Shader stats from VERDE:
Default scheduler:
Totals:
SGPRS: 491272 -> 488672 (-0.53 %)
VGPRS: 289980 -> 311093 (7.28 %)
Code Size: 11091656 -> 11219948 (1.16 %) bytes
LDS: 97 -> 97 (0.00 %) blocks
Scratch: 1732608 -> 2246656 (29.67 %) bytes per wave
Max Waves: 78063 -> 77352 (-0.91 %)
Wait states: 0 -> 0 (0.00 %)
Looking at some of the worst regressions, I get:
- The VGPR increase seems to be caused by the fact that if PS has used less
than 16 VGPRs, now it will always use 16 VGPRs and sometimes even 20.
However, the wave count remains at 10 if VGPRs <= 24, so no harm there.
- The scratch increase seems to be caused by SGPR spilling.
The unnecessary SGPR spilling has been an ongoing issue with the compiler
and it's completely fixable by rematerializing s_loads or reordering
instructions.
SI scheduler:
Totals:
SGPRS: 374848 -> 374576 (-0.07 %)
VGPRS: 284456 -> 307515 (8.11 %)
Code Size: 11433068 -> 11535452 (0.90 %) bytes
LDS: 97 -> 97 (0.00 %) blocks
Scratch: 509952 -> 522240 (2.41 %) bytes per wave
Max Waves: 79456 -> 78217 (-1.56 %)
Wait states: 0 -> 0 (0.00 %)
VGPRs - same story as before. The SI scheduler doesn't spill SGPRs so much
and generally spills way less than the default scheduler.
(522240 spills vs 2246656 spills)
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Still disabled.
Only prologs & epilogs are compiled in draw calls, but each variant of those
is compiled only once per process.
VS is always compiled as hw VS.
TES is always compiled as hw VS.
LS and ES stages are always compiled on demand.
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
It only exports the primitive ID.
Also used by TES when it's compiled as VS.
The VS input location of the primitive ID input is v2.
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Prologs (shader binaries inserted before the API shader binary) need to
know this, so that they won't change the input registers unintentionally.
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Kepler compute support is really different than Fermi and it's not
ready yet.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
Changes from v3:
- move the previous OP_SELP change to the previous commit
Changes from v2:
- make sure the op is OP_SELP when emitting the predicate and add one
assert
- use bld.getSSA() for mkOp2()
- add cross edge between tryLockAndSetBB and joinBB
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Acked-by: Ilia Mirkin <imirkin@alum.mit.edu>
This OP_SELP insn will be used to handle compare and swap subops.
Changes from v2:
- fix logic for GK110+
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
Shared memory address space (FILE_MEMORY_SHARED) must be used instead
of global memory when a shared memory area is declared.
Changes from v2:
- oops, do not remove TGSI_FILE_BUFFER in a switch in
nv50_ir_from_tgsi.cpp
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reduce likelihood of collision with real buffers by placing the
hole at the top of the 4G area. This fixes some indirect draw+compute
tests with large buffers.
Suggested by Ilia Mirkin.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
When indirect compute is used, the size of the grid (in blocks) is
stored as three integers inside a buffer. This requires a macro to
set up GRIDDIM_YX and GRIDDIM_Z.
Changes from v2:
- do not launch the grid if the number of groups for a dimension is 0
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
Textures and samplers don't seem to be aliased between COMPUTE and 3D.
Changes from v2:
- refactor the code to share (almost) the same logic between 3d and
compute
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
This is loosely based on 3D. Shader buffers are bound on c15 (the
driver constbuf) at offset 0x200.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
Changes from v3:
- add new validation state for COMPUTE driver constbuf
Changes from v2:
- always bind the driver consts even if user params come in via clover
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
This will be used to invalidate 3D driver constbuf when using COMPUTE
and vice-versa. This is needed because this CB contains a bunch of
useful information like the addrs of shader buffers.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
Loosely based on 3D.
Changs from v3:
- invalidate COMPUTE CBs after validating 3D CBs because they are
aliased
Changes from v2:
- get rid of the 's' param to nvc0_cb_bo_push() because it doesn't
matter to upload constbufs for compute using the 3d chan
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
Compute shaders are totally unsupported. This avoids Clover to
report that OpenCL is supported on Tesla because it's a lie.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Pierre Moreau <pierre.morrow@free.fr>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
A return value of '-1' means that there was error during swap with a
window drawable, in this case we set error as EGL_BAD_NATIVE_WINDOW.
v2: coding style cleanup, better commit message
Signed-off-by: Matt Roper <matthew.d.roper@intel.com>
Cc: "11.0 11.1" <mesa-stable@lists.freedesktop.org
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
Null-check on "*value" is currently done in _eglGetSyncAttrib, which is
after eglGetSyncAttribKHR dereferences it.
Move the check a layer up (in the beginning of eglGetSyncAttribKHR) to
avoid segfaults.
Cc: "11.0 11.1" <mesa-stable@lists.freedesktop.org
Signed-off-by: Dongwon Kim <dongwon.kim@intel.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
[Emil Velikov: tweak commit message, add stable tag]
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
It's basically the same thing as GL_ARB_texture_stencil8 except that
glCopyTexImage isn't supported, so add STENCIL_INDEX to the list of
invalid GLES formats for glCopyTexImage.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Eduardo Lima Mitev <elima@igalia.com>
- LOD must be provided in .w for TXF (even for buffer textures)
- User buffer must be valid at draw time
- Must have a sampler associated with the sampler view
This makes PBO uploads work again on nouveau.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
The base format is a function of the user-requested format, while the
driver format is not. So we should use the base format instead.
The driver format can be anything. Specifically in the stencil-only
case, it might be a depth/stencil format. However we still want to
refuse such an attachment when bound to GL_DEPTH.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Brian Paul <brianp@vmware.com>
This reduces a glTexImage(GL_RGBA, GL_UNSIGNED_BYTE) hot spot in when
storing the texture as BGRA.
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Instead of discarding the texture we created, keep it around in case
the next glDrawPixels draws the same image again. This is intended
to help application which draw the same image several times in a row,
either within a frame or subsequent frames.
Reviewed-by: Charmaine Lee <charmainel@vmware.com>
Specifically, for the case where we initialize a dmat with a source
matrix that has fewer columns/rows.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Need to set some non-zero limits for MaxCombinedUniformComponents,
otherwise we hit an "Too many <type> shader uniform components" error
in the linker.
Signed-off-by: Rob Clark <robclark@freedesktop.org>
Reviewed-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
This fixes a glDrawPixels regression since b63fe0552b. The new
quad-drawing utility code uses 3 vertex attributes (xyz, rgba, st).
For glDrawPixels path we don't use the rgba attribute so there's a
gap in the TGSI VS input declarations (INPUT[0] = pos, INPUT[2] =
texcoord). The TGSI->VGPU10 translations code did not handle this
correctly. I missed this because my VM was configured for HWv11
while testing.
Another way to fix this would be to change the tgsi_scan.c code so
that the tgsi_shader_info::num_inputs (and num_outputs) included
the unused inputs/outputs. These counts would then actually be
"max input register index + 1" rather than "number of used inputs".
But that change could impact all drivers so put it off for now.
No regressions found with piglit or typical GL apps.
v2: also update alloc_system_value_index() to use info.file_max[]
Reviewed-by: Charmaine Lee <charmainel@vmware.com>
Because the if statement that checks whether we have a return
statement is valid only on x86, surround it with X86 or X86-64
arch defines
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Currently, disassemble() directly prints to stdout. This has broke the
profiling support for llvmpipe JIT code.
This patch redirects the output to an sstream object, which is then
either gets printed to stdout (for assembly debugging) or gets written
to a file in /tmp/ (for profiling support).
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
src/gallium/drivers/trace/tr_context.c:1713:39: warning: ‘rbug_blocker_flags’ defined but not used [-Wunused-const-variable]
static const struct debug_named_value rbug_blocker_flags[] = {
^~~~~~~~~~~~~~~~~~
Note that use of rbug_blocker_flags was removed in:
commit 5494332128
Author: Jakob Bornecrantz <jakob@vmware.com>
Date: Wed May 12 19:26:19 2010 +0100
trace: Remove rbug from trace
Signed-off-by: Rob Clark <robdclark@gmail.com>
src/gallium/auxiliary/pipebuffer/pb_bufmgr_mm.c: In function ‘mm_bufmgr_create_from_buffer’:
src/gallium/auxiliary/pipebuffer/pb_bufmgr_mm.c:288:4:
warning: statement is indented as if it were guarded by... [-Wmisleading-indentation]
if(mm->map)
^~
src/gallium/auxiliary/pipebuffer/pb_bufmgr_mm.c:286:1: note:
...this ‘if’ clause, but it is not
if(mm->heap)
^~
Signed-off-by: Rob Clark <robdclark@gmail.com>
src/gallium/auxiliary/hud/font.c:234:22: warning: ‘Fixed8x13_Character_159’ defined but not used [-Wunused-const-variable]
static const GLubyte Fixed8x13_Character_159[] = { 9, 0, 0, 0, 0, 0, 0,170, 0, 0, 0,130, 0, 0, 0,130, 0, 0, 0,130, 0, 0, 0,170, 0, 0, 0, 0, 0};
^~~~~~~~~~~~~~~~~~~~~~~
.... many more..
These are simply unused, just #if 0 them out for now, in case someone
wants to use them in the future.
Signed-off-by: Rob Clark <robdclark@gmail.com>
src/mesa/main/texstore.c:92:22: warning: ‘map_1032’ defined but not used [-Wunused-const-variable]
static const GLubyte map_1032[6] = { 1, 0, 3, 2, ZERO, ONE };
^~~~~~~~
src/mesa/main/texstore.c:91:22: warning: ‘map_3210’ defined but not used [-Wunused-const-variable]
static const GLubyte map_3210[6] = { 3, 2, 1, 0, ZERO, ONE };
^~~~~~~~
src/mesa/main/texstore.c:90:22: warning: ‘map_identity’ defined but not used [-Wunused-const-variable]
static const GLubyte map_identity[6] = { 0, 1, 2, 3, ZERO, ONE };
^~~~~~~~~~~~
These appear to be unused since:
commit 8ec6534b26
Author: Iago Toral Quiroga <itoral@igalia.com>
AuthorDate: Wed Oct 15 13:42:11 2014 +0200
mesa: Use _mesa_format_convert to implement texstore_rgba.
Signed-off-by: Rob Clark <robdclark@gmail.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
src/compiler/glsl/lower_discard_flow.cpp:79:1: warning: ‘ir_visitor_status {anonymous}::lower_discard_flow_visitor::visit_enter(ir_loop_jump*)’ defined but not used [-Wunused-function]
lower_discard_flow_visitor::visit_enter(ir_loop_jump *ir)
^~~~~~~~~~~~~~~~~~~~~~~~~~
The base class method that was intended to be overridden was
'visit(ir_loop_jump *ir)', not visit_enter().
Signed-off-by: Rob Clark <robdclark@gmail.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
src/compiler/glsl/ast_to_hir.cpp: In function ‘unsigned int ast_process_struct_or_iface_block_members(exec_list*, _mesa_glsl_parse_state*, exec_list*, glsl_struct_field**, bool, glsl_matrix_layout, bool, ir_variable_mode, ast_type_qualifier*,
unsigned int, unsigned int)’:
src/compiler/glsl/ast_to_hir.cpp:6339:52: warning: ‘first_member_has_explicit_location’ may be used uninitialized in this function [-Wmaybe-uninitialized]
if (!layout->flags.q.explicit_location &&
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~^~
((first_member_has_explicit_location &&
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
!qual->flags.q.explicit_location) ||
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
(!first_member_has_explicit_location &&
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
qual->flags.q.explicit_location))) {
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Signed-off-by: Rob Clark <robdclark@gmail.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
src/mesa/drivers/dri/i965/brw_fs_copy_propagation.cpp:244:1: warning:
‘void {anonymous}::fs_copy_prop_dataflow::dump_block_data() const’ defined but not used [-Wunused-function]
fs_copy_prop_dataflow::dump_block_data() const
^~~~~~~~~~~~~~~~~~~~~
From looking at git history, it looks like this is intended to be unused
(ie. just for adding on-demand debug prints)
Signed-off-by: Rob Clark <robdclark@gmail.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
src/util/hash_table.h:111:23: warning: ‘_mesa_fnv32_1a_offset_bias’ defined but not used [-Wunused-const-variable]
static const uint32_t _mesa_fnv32_1a_offset_bias = 2166136261u;
^~~~~~~~~~~~~~~~~~~~~~~~~~
Signed-off-by: Rob Clark <robdclark@gmail.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Android builds with -Wunused-parameter enabled which results in spewing
lots of warnings. Disable it so more meaningful warnings are more visible.
Signed-off-by: Rob Herring <robh@kernel.org>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
Use the LOCAL_CFLAGS_{32/64} instead of arch specific variants to define
the DEFAULT_DRIVER_DIR. This enables building for arm64.
Cc: Chih-Wei Huang <cwhuang@android-x86.org>
Signed-off-by: Rob Herring <robh@kernel.org>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
TARGET_CC is not defined for the secondary arch on combined 32/64-bit
builds. The build system uses 2ND_TARGET_CC instead and it is not meant
to be used in module makefiles. LOCAL_CC was used to provide C only
flags as -std=c99 is not valid for C++ files. Since Android 4.4,
LOCAL_CONLYFLAGS was added to set compiler flags on C files only, so it
can be used now instead of LOCAL_CC.
This will break on pre-4.4 versions of Android, but it unlikely anyone
is using current Mesa with such an old version of Android.
Cc: Chih-Wei Huang <cwhuang@android-x86.org>
Signed-off-by: Rob Herring <robh@kernel.org>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
Pass the additional config attributes to dri2_add_config to set them
instead of open coding them. This is in preparation to add more attributes.
Signed-off-by: Rob Herring <robh@kernel.org>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
From ARB_sample_shading:
"gl_NumSamples is the total number of samples in the framebuffer,
or one if rendering to a non-multisample framebuffer"
So make sure to always pass in at least 1.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Edward O`Callaghan <eocallaghan@alterapraxis.com>
Reviewed-by: Neil Roberts <neil@linux.intel.com>
This patch moves the calculation of current uniforms to
link_uniforms, which makes use of UniformRemapTable which
stores all the reserved uniform locations.
Location assignment for implicit uniforms now tries to use
any gaps left in the table after the location assignment
for explicit uniforms. This gives us more space to store more
uniforms.
Patch is based on earlier patch with following changes/additions:
1: Move the counting of explicit locations to
check_explicit_uniform_locations and then pass
the number to link_assign_uniform_locations.
2: Count the number of empty slots in UniformRemapTable
and store them in a list_head.
3: Try to find an empty slot for implicit locations from
the list, if that fails resize UniformRemapTable.
Fixes following CTS tests:
ES31-CTS.explicit_uniform_location.uniform-loc-mix-with-implicit-max
ES31-CTS.explicit_uniform_location.uniform-loc-mix-with-implicit-max-array
Signed-off-by: Tapani Pälli <tapani.palli@intel.com>
Signed-off-by: Plamena Manolova <plamena.manolova@intel.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=93696
Just like the rest of the msaa "implementation" it's just fake for now...
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
This basically saves the current pipeline state, sets up state for
rendering, constructs a set of textured quads, renders, then restores
the previous pipeline state.
It shouldn't be hard to implement a similar function for non-gallium
drives. With some code refactoring, the vertex definition code could
probably be shared.
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
This improves the performance of applications which use glXUseXFont()
or wglUseFontBitmaps() and glCallLists() to draw bitmap text.
Basically, we collect all the glBitmap images from the display lists
and put them into a texture atlas. To render the bitmaps for a
glCallLists() command, we render a set of textured quads where each
quad is textured with one bitmap image. Actually, the rendering part
has to be done by the Mesa driver or Mesa/gallium state tracker.
Note that GLUT demos that use glutBitmapCharacter() don't benefit
from this.
v2, per Nicolai Hähnle:
- check the max tex rect size is at least 1024.
- add comment in dd.h that texture_rectangle is required.
- in _mesa_DeleteLists(), try to delete the atlas before the list(s)
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Gallium doesn't present these as GL_RED-style. A swizzle is necessary to
present the proper data in the unused components.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Also adds some of the Iris/Pro parts which we previously didn't have named.
v2: 0x192d is gt3, not gt4
Adding some 'e' tags for eDRAM parts
Signed-off-by: Ben Widawsky <benjamin.widawsky@intel.com>
Acked-by: Michał Winiarski <michal.winiarski@intel.com>
The restriction on multisampled integer texture formats only applies to
GLES 3.0, so don't apply it to GLES 3.1 contexts. This fixes a slew of
dEQP-GLES31.functional.state_query.internal_format.*
tests, which now all pass.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Timothy Arceri <timothy.arceri@collabora.com>
Every stage has a corresponding 3DSTATE_CONSTANT_XS packet, so having
the code to create and emit push constant buffers in genX_vs_state.c
is a little strange. Moving it to a separate file seems more logical.
v2 [Ken]: Rebase on master, explain motivation in the commit message.
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Gen4/5's SEL instruction cannot use conditional modifiers, so min/max
are implemented as CMP + SEL. Handling that after optimization lets us
CSE more.
On Ironlake:
total instructions in shared programs: 6426035 -> 6422753 (-0.05%)
instructions in affected programs: 326604 -> 323322 (-1.00%)
helped: 1411
total cycles in shared programs: 129184700 -> 129101586 (-0.06%)
cycles in affected programs: 18950290 -> 18867176 (-0.44%)
helped: 2419
HURT: 328
Reviewed-by: Francisco Jerez <currojerez@riseup.net>
This will prevent optimization passes from introducing unsupported
library calls.
Tested-by: Michel Dänzer <michel.daenzer@amd.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Even though it's a no-op, it's important to keep track of the type so
that we can pick the properly-signed op later on.
This fixes dEQP-GLES3.functional.shaders.precision.uint.highp_div_fragment,
which ended up using IDIV instead of UDIV.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Cc: mesa-stable@lists.freedesktop.org
Note that this results in a different transformation for the viewport's
Z axis (depth range), but that doesn't matter for this case.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
On gen7 (Ivy Bridge, Haswell), we will get a GPU hang if an indirect
dispatch is used, but one of the dimensions is 0.
Therefore we use predicated rendering on the GPGPU_WALKER command to
handle this case.
Fixes piglit test: spec/arb_compute_shader/zero-dispatch-size
From the ARB_compute_shader spec, under DispatchCompute:
"If the work group count in any dimension is zero, no work groups are
dispatched."
And then for DispatchComputeIndirect:
... "is equivalent (assuming no errors are generated) to calling
DispatchCompute with <num_groups_x>, <num_groups_y> and
<num_groups_z>" ...
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=94100
Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Ben Widawsky <benjamin.widawsky@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Tested-by: Ilia Mirkin <imirkin@alum.mit.edu>
This seems to give more reliable results. More similar to what we do on
a3xx, although I think it breaks the a3xx theory that the four sets of
results map to each MRT (since we appear to still only have four sets on
a4xx). The divide-by-two is a bit odd, but seems to be needed for some
reason.
Signed-off-by: Rob Clark <robclark@freedesktop.org>
Some hw queries need their sample memory locations to have certain
alignment. At the moment that isn't an issue, since the only hw query
is occlusion, so all samples have the same size. But when others are
added with different sample sizes, this starts to be a problem.
All current and immediately upcoming hw queries simply need their
sample address aligned to their size, so let's use that for now.
Signed-off-by: Rob Clark <robclark@freedesktop.org>
Add enable hook for hw query providers. Some will need to configure
perfctr selector registers, which we want to do at the start of the
submit.
Signed-off-by: Rob Clark <robclark@freedesktop.org>
This will be needed to support converting from cycle counts to time for
performance related queries (initially time-elapsed, but there are some
additional performance counters that could be wired up).
Signed-off-by: Rob Clark <robclark@freedesktop.org>
src/gallium/drivers/freedreno/ir3/ir3_compiler_nir.c: In function ‘emit_tex’:
src/gallium/drivers/freedreno/ir3/ir3_compiler_nir.c:1368:26: warning: unused variable ‘const_off’ [-Wunused-variable]
struct ir3_instruction *const_off[4];
^~~~~~~~~
unused since:
commit 8750299a42
Author: Jason Ekstrand <jason.ekstrand@intel.com>
Date: Tue Feb 9 14:51:28 2016 -0800
nir: Remove the const_offset from nir_tex_instr
Signed-off-by: Rob Clark <robdclark@gmail.com>
helps shaders in saints row IV, bioshock infinite and shadow warrior
total instructions in shared programs : 1914931 -> 1903900 (-0.58%)
total gprs used in shared programs : 247920 -> 247785 (-0.05%)
total local used in shared programs : 5673 -> 5673 (0.00%)
total bytes used in shared programs : 17558272 -> 17457320 (-0.57%)
local gpr inst bytes
helped 0 137 719 719
hurt 0 12 0 0
v2: remove this opt for OP_SLCT and check against float for OP_SET
v3: simplified the code
Signed-off-by: Karol Herbst <nouveau@karolherbst.de>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
When there's a predicate, it just goes onto the sources list. If the
quadop only has a single regular source, we will end up thinking that
the predicate is the second source. Check explicitly for the predSrc so
that we don't accidentally emit the wrong thing.
This fixes a bunch of dEQP-GLES3.functional.shaders.derivate.* tests.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Cc: mesa-stable@lists.freedesktop.org
In a situation where the seamless setting isn't available on a
per-texture basis (G200+ Teslas, and all Fermis), assume that all
samplers will have it identically set, and enable accordingly.
This fixes arb_seamless_cubemap piglit test on Fermi and Tesla.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Noticed by Ilia when I was trying to figure out why some app was failing
to use ETC2.
Signed-off-by: Rob Clark <robclark@freedesktop.org>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
Without this NVF0_COMPUTE environment variable, compute support is
initialized by default and this is not what we want for now because
it might break 3D. It will be enabled by default once we are sure it
won't break anything.
Please note that compute support on GM200+ is not enabled yet because
it needs to be double-checked.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
Fortunately, compute support on GM107 is very close to GK110, except
the GK110_COMPUTE.UNK02C4 which is invalid and should not be used.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
Because our firmware doesn't support the GK110_COMPUTE.FIRMWARE[0x6]
method the GPU hangs when it is used. Removing it fix the issue and
allow to launch compute shaders on GK110+.
Tested on GK208 and GM107.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
We already have one in the IR code that can be used everywhere its
needed in the AST code so remove the one from the AST.
Reviewed-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
This is usually handled by the backends in order to handle the
various interactions with the gl_*Color built-ins.
The problem is this means linking will fail if one side on the
interface adds the smooth qualifier to the varying and the other
side just uses the default even though they match.
This fixes various deqp tests. The spec is not clear what to for
desktop GL so leave it as is for now.
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=92743
This fixes the following dEQP test and the other compswap variants.
dEQP-GLES31.functional.ssbo.atomic.compswap.highp_int
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
When the number of uniform blocks is less than 12,
ARB_uniform_buffer_object can't be enabled and the maximum GL version
is not even 3.1...
This fixes a regression introduced in 7c79c1e (st/mesa: add compute
shader state) if the maximum number of uniform blocks allowed for
compute shaders is less than 12. This happens on Kepler but this might
also affect other Gallium drivers.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reported-by: Tobias Klausmann <tobias.johannes.klausmann@mni.thm.de>
Tested-by: Tobias Klausmann <tobias.johannes.klausmann@mni.thm.de>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Tobias Klausmann <tobias.johannes.klausmann@mni.thm.de>
The ARB_compute_shader spec says:
"If the work group count in any dimension is zero, no work groups
are dispatched."
Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
See Ivy Bridge PRM, Volume 2, Part 2, 1.8.4 INTERFACE_DESCRIPTOR_DATA:
DWORD 5, bits 20:16: "This field indicates how much shared local
memory the thread group requires. The amount is specified in 4k
blocks, but only powers of 2 are allowed: 0, 4k, 8k, 16k, 32k and 64k
per half-slice."
For Haswell, see Volume 2d, INTERFACE_DESCRIPTOR_DATA:
DWORD 5, bits 20:16: With text identical to the Ivy Bridge PRM.
For Broadwell, see Volume 2d, INTERFACE_DESCRIPTOR_DATA:
DWORD 6, bits 20:16: With text identical to the Ivy Bridge PRM.
Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Ben Widawsky <benjamin.widawsky@intel.com>
cso_save_state() takes a bitmask of state items to save. Calling
cso_restore_state() restores those states.
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
Define a new st_util_vertex structure which is a bit smaller (9 floats
versus the previous 12 floats per vertex). Clean up the glClear,
glDrawPixels and glBitmap code that sets up the vertex data and does the
drawing so it's all very similar. This can lead to more consolidation.
v2: add assertion that vertex buffer slot == 0 to catch possible future
change in cso_get_aux_vertex_buffer_slot() behavior.
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
LLVM removed LLVMAddTargetData for the 3.9 release in r260919. For the two
places in mesa where this is called, only enable the lines when compiling
for less then 3.9.
For the radeon driver, I'm not sure how to check if any other LLVM calls need
to be adjusted. I think since the target data used is extracted from the
LLVMModule, it isn't necessary to pass it back to LLVM again.
The code does compile, and at least for radeonsi does run OpenGL games.
[ Michel Dänzer: Move #if closer to LLVMAddTargetData in lp_bld_init.c,
and add HAVE_LLVM < 0x0309 guards around now unused occurrences of TD
and data_layout ]
Signed-off-by: Matthew Dawson <matthew@mjdsystems.ca>
Reviewed-and-Tested-by: Michel Dänzer <michel.daenzer@amd.com>
Alos use the opportunity to mark inputs constant. (Context has to be
given as read-write to intel_miptree_supports_non_msrt_fast_clear()
to support debug output).
Signed-off-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
Reviewed-by: Ben Widawsky <benjamin.widawsky@intel.com>
v2 (Ben): Use combination of msaa_layout and number of samples
instead of introducing explicit type for lossless
compression (intel_miptree_is_lossless_compressed()).
Signed-off-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
Reviewed-by: Ben Widawsky <benjamin.widawsky@intel.com>
v2 (Ben): Use combination of msaa_layout and number of samples
instead of introducing explicit type for lossless
compression (intel_miptree_is_lossless_compressed()).
Signed-off-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
Reviewed-by: Ben Widawsky <benjamin.widawsky@intel.com>
v2 (Ben): Use combination of msaa_layout and number of samples
instead of introducing explicit type for lossless
compression (intel_miptree_is_lossless_compressed()).
Signed-off-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
Reviewed-by: Ben Widawsky <benjamin.widawsky@intel.com>
v2 (Ben): Use combination of msaa_layout and number of samples
instead of introducing explicit type for lossless
compression.
v3 (Ben): Squash with "i965: Resolve color buffer also in
lossless compression case" and clarify simple
non-compressed fast clear case.
Signed-off-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
Reviewed-by: Ben Widawsky <benjamin.widawsky@intel.com>
This got pushed accidentally in the first place but wasn't reverted
as it didn't regress piglit but instead fixed one newly introduced
test exercising a corner in case in i965 driver. However, saving and
restoring vertex buffer context is complicated and requires more
thought.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=94150
Signed-off-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
Reviewed-by: Tapani Palli <tapani.palli@intel.com>
Adds support for the new TIC layout that's present on Maxwell GPUs,
heavily based on the code for the existing layout.
This code is required for GM20x support. While GM10x supports the older
layout still, this commit switches it to use the updated version instead.
Piglit testing shows zero regressions on GM107.
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Acked-by: Ilia Mirkin <imirkin@alum.mit.edu>
We previously stored texture format information as it would appear in
the TIC.
We're about to support the new TIC layout that appeared with Maxwell,
so it makes more sense to store the data in a split-out format.
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Acked-by: Ilia Mirkin <imirkin@alum.mit.edu>
We've previously had identical naming between vertex and texture
formats, so it mostly made sense to define these together.
However, upcoming patches are going to transition the driver over to
using updated texture header definitions using NVIDIA's naming, and this
will no longer be the case.
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Acked-by: Ilia Mirkin <imirkin@alum.mit.edu>
This enables ARB_shader_image_load_store and ARB_shader_image_size when
the backend claims support for these. It will also implicitly enable the
image component of ARB_shader_texture_image_samples.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Dave Airlie <airlied@redhat.com>
Make them akin to shader buffers, with no refcounting/etc. Just used to
pass data about the bound image in ->set_shader_images.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Dave Airlie <airlied@redhat.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Silences the following GCC warning:
mesa/src/gallium/drivers/vc4/vc4_qir_schedule.c: In function 'qir_schedule_instructions':
mesa/src/gallium/drivers/vc4/vc4_qir_schedule.c:578:16: warning: missing braces around initializer [-Wmissing-braces]
struct schedule_state state = { 0 };
^
Signed-off-by: Rhys Kidd <rhyskidd@gmail.com>
Signed-off-by: Eric Anholt <eric@anholt.net>
Variable was previously always set to true. Accordingly, the later
assert() served no active purpose.
Found with GCC warning and code inspection:
mesa/src/gallium/drivers/vc4/vc4_qpu_emit.c: In function'vc4_generate_code':
mesa/src/gallium/drivers/vc4/vc4_qpu_emit.c:315:22: warning: variable 'handled_qinst_cond' set but not used [-Wunused-but-set-variable]
bool handled_qinst_cond = true;
^
Signed-off-by: Rhys Kidd <rhyskidd@gmail.com>
Signed-off-by: Eric Anholt <eric@anholt.net>
The two consumers want to know that the destination will be exactly the
source, which is not true if we might not set the destination.
Signed-off-by: Eric Anholt <eric@anholt.net>
This fixes a number of dEQP tests, such as:
dEQP-GLES31.functional.program_interface_query.buffer_limited_query.resource_query
It was expecting the length to be set even in the bufSize == 0 case.
Also _mesa_get_program_resourceiv does some error checking on the
resource which should probably happen even in the bufSize == 0 case as
well although there's no dEQP test for that.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
We already have this logic in the gallium/util functions so
lets reduce some entropy while here.
V.2:
Apply change to nv50 also as suggested by Samuel Pitoiset.
Signed-off-by: Edward O'Callaghan <eocallaghan@alterapraxis.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
brw_draw_upload.c uploads VertexID/InstanceID first, then DrawID.
So we need to assign the attribute mapping in that order as well.
Fixes the following Pigit tests with the vec4 backend:
- arb_shader_draw_parameters-drawid vertexid
- arb_shader_draw_parameters-drawid-indirect basevertex
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
The ImageAccess array is statically sized to MAX_IMAGE_UNIFORMS:
GLenum ImageAccess[MAX_IMAGE_UNIFORMS];
There was no bounds checking ensuring we don't overflow. Passing in a
shader with too many uniforms would cause writes to extend into other
fields, such as sh->NumImages.
Later linker checks already handle reporting an error when there are too
many images, so just avoid corrupting structures here.
This rearranges the logic a bit to look more like the sampler case.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Timothy Arceri <timothy.arceri@collabora.com>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Tested-by: Jordan Justen <jordan.l.justen@intel.com>
GL_ARB_texture_multisample and GLES 3.1 expect the initial value to be
GL_TRUE. This fixes
dEQP-GLES31.functional.state_query.texture_level.texture_2d_multisample_array.fixed_sample_locations_integer
and a few related tests.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Chris Forbes <chrisf@ijw.co.nz>
SPIR-V has a concept of a function type that's used fairly heavily. We
could special-case function types in SPIR-V -> NIR but it's easier if we
just add support to glsl_types.
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
This is to be used by SPIR-V for representing a sampler that isn't attached
to any particular image. In SPIR-V, all of the interesting bits such as
dimensionality, sampled type, etc. come from the image, the bare "sampler"
type simply uses a sampled type of VOID and 0 values for the rest.
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
It's a bit more descriptive since it is the base type that you get when you
sample from it. Also, the next commit adds a bare "sampler" type and we
need glsl_type::sampler_type available for a public static member.
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Fixes several of the
"dEQP-GLES31.functional.image_load_store*load_store*single_layer" dEQP
tests that use image formats we implement using untyped surface
messages.
Cc: mesa-stable@lists.freedesktop.org
Tested-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Changes from v3:
- dump the TGSI compute program
Changes from v2:
- remove use of MALLOC()
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
According to the spec, this also increases the following minimum values:
- MAX_COMBINED_TEXTURE_IMAGE_UNITS 96 (6*16), was 80
- MAX_UNIFORM_BUFFER_BINDINGS 72 (6*12), was 60
ARB_compute_shader is not enabled by default because images support is
still not implemented yet. If you want to use it you need to set
MESA_EXTENSION_OVERRIDE=GL_ARB_compute_shader.
Changes from v2:
- make use of the new PIPE_CAP_SHADER_SUPPORTED_IRS cap instead of
enabling the extension when PIPE_CAP_COMPUTE is enabled.
- query for PIPE_CAP_COMPUTE first
- s/shader_supported_irs/compute_supported_irs/
- disable ARB_compute_shader and add a comment which explains why
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
LOCAL_INVOCATION_ID, WORK_GROUP_ID and NUM_WORK_GROUPS are respectively
mapped to THREAD_ID, BLOCK_ID and GRID_SIZE.
Changes from v2:
- add assertions in st_translate_program()
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
This adds GLSL intrinsics for load/store and atomic operations.
Changes from v2:
- use PROGRAM_MEMORY instead of PROGRAM_BUFFER
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
Compute needs a new and different validation path.
Changes from v2:
- make use of unreachable() instead of assert() when the pipeline is
invalid
- move the st_pipeline enumeration to st_context.h instead of st_api.h
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
This introduces TGSI_FILE_MEMORY for shared, global and local memory.
Only shared memory is currently supported.
Changes from v2:
- introduce TGSI_FILE_MEMORY
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
This cap indicates the supported representations of programs. It should
be a mask of pipe_shader_ir bits. It will allow to enable
ARB_compute_shader if the underlying driver supports TGSI.
Changes from v2:
- improve description of PIPE_SHADER_CAP_SUPPORTED_IRS
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Like indirect draw, we need to store a resource and an offset that
needs to be 4 byte aligned. When indirect is used, the size of the
grid (in blocks) is stored with three 32-bit integers.
Changes from v2:
- s/most values/block sizes/
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
This introduces pipe_grid_info which contains all information to
describe a launch_grid call. This will be used to implement indirect
compute in the same fashion as indirect draw.
Changes from v2:
- correctly initialize pipe_grid_info for nv50/nvc0
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
Changes from v2:
- removed cso_{save,restore}_compute_shader() functions and the
compute_shader_saved variable because disabling compute shaders for
meta ops is not currently needed
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
The size of shared variables needs to be stored in gl_compute_program
in order to set up pipe_compute_state::req_local_mem.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
This will allow to query the underlying drivers for the maximum
total storage size of all variables declared as <shared> with
PIPE_COMPUTE_CAP_MAX_LOCAL_SIZE.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
Until now there has been only one type of color buffer that needs
to resolved - namely single sampled fast clear. As even the
sampler engine in GPU doesn't understand the associated meta data,
the color values need to be always resolved prior to reading them.
From SKL onwards there is new scheme supported called the lossless
compression of single sampled color buffers. This is something that
is understood by the sampling engine and therefore resolving of
these types of buffers is not necessary before sampling.
This patch adds means to make the distinction when considering if
resolve is needed.
Signed-off-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
Reviewed-by: Ben Widawsky <benjamin.widawsky@intel.com>
In addition to simply calling miptree_create() the higher level
call intel_miptree_create() also considers if the buffer should
be associated with an auxiliary buffer based on the given format.
Here we are allocating an auxiliary buffer which in turn has such
format that would mislead intel_miptree_create_layout() later on
to try to associate the auxiliary buffer with an auxiliary buffer.
To prevent this the actual buffer creation logic was split out
into its own function. Lets invoke that instead.
v2 (Ben): Do not signal msaa layout with explicit argument but
using layout_flags instead.
Signed-off-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
Reviewed-by: Ben Widawsky <benjamin.widawsky@intel.com>
This allows ls, and scripts to get the file names in the correct order of
optimization.
Signed-off-by: Ben Widawsky <benjamin.widawsky@intel.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
The L3 partitioning code tries to look at all programs - both render
programs (VS/TCS/TES/GS/FS) and compute (CS).
After calling brw_clear_cache, all prog_data pointers are invalid and
point to freed data. The intention was that flagging the dirty bits for
all programs would cause the next draw call to re-run the atoms for each
program stage, uploading new programs and installing new, valid pointers.
However, this doesn't quite work in our new multi-pipeline world. When
drawing or dispatching a compute workload, we only consider the programs
for the appropriate pipeline: drawing sets up VS/TCS/TES/GS/FS, but not
CS, and vice versa. This leaves pointers dangling a bit longer than
intended.
The L3 configuration code tries to inspect the prog_data for all shader
stages, so that we avoid having to reconfigure it when swapping back and
forth between render and compute workloads. So we can't have dangling
pointers.
The fix is simple: have brw_clear_cache NULL out stale prog_data
pointers, making it safe to inspect. The next L3 configuration pass
will see either the render shaders or compute shader as missing for
one go around, but will pick them up when both pipelines have run.
In other words, we'll simply reconfigure L3 twice, which is safe,
if a tiny bit wasteful - but then again, we just threw every compiled
shader we had on the floor and started recompiling the from scratch,
which is massively more wasteful, so it's not much of a concern.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=93790
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Francisco Jerez <currojerez@riseup.net>
Reviewed-by: Jordan Justen <jljusten@gmail.com>
If there is no pipe info log, we would unconditionally deref length,
which was only optionally there. _mesa_copy_string handles the source
being null, as well as the length, so may as well just always call it.
Fixes a segfault in
dEQP-GLES31.functional.state_query.program_pipeline.info_log
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Brian Paul <brianp@vmware.com>
Similar to commit dd9d2963d6 (mesa: AtomicBufferBindings should be
initialized to zero.), we should reset these to zero when unbinding.
This fixes a number of dEQP failures due to cross-test pollution. The
tests properly unbound everything, but when querying the values again,
the expectation was that they would be 0.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Francisco Jerez <currojerez@riseup.net>
Similar as for AUX1-3, these enums aren't invalid (i.e. -1) but also not
supported by mesa. Returning BUFFER_COUNT causes the proper error to be
returned by ReadBuffer and other functions. This resolves some failures
in
dEQP-GLES31.functional.debug.negative_coverage.get_error.buffer.read_buffer
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Brian Paul <brianp@vmware.com>
This fixes
dEQP-GLES31.functional.debug.negative_coverage.get_error.buffer.clear_bufferfv
and brings the logic up to spec with GL 4.5
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Brian Paul <brianp@vmware.com>
This fixes
dEQP-GLES31.functional.debug.negative_coverage.get_error.buffer.clear_bufferuiv
and brings the logic up to spec with GL 4.5
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Brian Paul <brianp@vmware.com>
There's a hunk above which sets INVALID_ENUM for GL_DEPTH
unconditionally.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Brian Paul <brianp@vmware.com>
This fixes
dEQP-GLES31.functional.state_query.texture.texture_2d_multisample.depth_stencil_mode_integer
and a few related tests.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Nanley Chery <nanley.g.chery@intel.com>
We were implementing those the same way than
the default pool, which is sub-optimal.
The buffer is supposed to return pointer to
a ram copy when user locks, and automatically
update the vram copy when needed.
v2: Rename NineBuffer9_Validate to NineBuffer9_Upload
Rename validate_buffers to update_managed_buffers
Initialize NineBuffer9 managed fields after the resource
is allocated. In case of allocation failure, when the dtor
is executed, This->base.pool is then rightfully set.
Signed-off-by: Axel Davy <axel.davy@ens.fr>
For 32 bits, incoming stack is 4-byte aligned.
We need to realign the stack to 16-byte at some point,
or there are issues later (crash with SSE, llvm, etc).
This patch chooses to align the stack at API entry points.
Signed-off-by: Axel Davy <axel.davy@ens.fr>
using MIN/MAX is fine instead of CLAMP.
NRM doesn't exist anymore.
Signed-off-by: Axel Davy <axel.davy@ens.fr>
Reviewed-by: Patrick Rudolph <siro@das-labor.org>
We had several issues of crashes with it.
This should fix it.
Signed-off-by: Axel Davy <axel.davy@ens.fr>
Reviewed-by: Patrick Rudolph <siro@das-labor.org>
Add new argument to d3d9_to_pipe_format_checked to
be able to bypass format support checks. This argument
is set to TRUE when the requested Pool is SCRATCH.
Signed-off-by: Axel Davy <axel.davy@ens.fr>
Reviewed-by: Patrick Rudolph <siro@das-labor.org>
Returns INVALIDCALL when trying to create a surface
of unsupported format.
In practice, apps are supposed to check for format
support before trying to create a render target
of that format. However some bad behaving apps
could just try to create the surface and deduce if
it failed that it wasn't supported.
Signed-off-by: Axel Davy <axel.davy@ens.fr>
Reviewed-by: Patrick Rudolph <siro@das-labor.org>
Texture and CubeTexture use common code,
and thus ATI1/ATI2 is already implemented
for CubeTexture.
Signed-off-by: Axel Davy <axel.davy@ens.fr>
Reviewed-by: Patrick Rudolph <siro@das-labor.org>
We were having checks at both Create*Texture functions
and in ctors.
Move all Create*Texture checks to ctors.
Signed-off-by: Axel Davy <axel.davy@ens.fr>
Reviewed-by: Patrick Rudolph <siro@das-labor.org>
This->base.base.resource is worth NULL
for SYSTEMMEM textures.
Signed-off-by: Axel Davy <axel.davy@ens.fr>
Reviewed-by: Patrick Rudolph <siro@das-labor.org>
We do not support shared textures, thus no need to set
the shared flag.
Signed-off-by: Axel Davy <axel.davy@ens.fr>
Reviewed-by: Patrick Rudolph <siro@das-labor.org>
We do not create a resource for SYSTEMMEM textures,
thus we do not need to set resource usage.
The only exception is vertexbuffer SYSTEMMEM, since
we do use a pipe resource for them.
Signed-off-by: Axel Davy <axel.davy@ens.fr>
Reviewed-by: Patrick Rudolph <siro@das-labor.org>
We already use these for gallium in
src/gallium/auxiliary/os/os_memory_stdc.h and it's always better to
minimize divergences between MinGW and MSVC.
Reviewed-by: Brian Paul <brianp@vmware.com>
Auxiliary buffers are always created with sample number of zero
which effectively prevents intel_miptree_create_layout() from trying
to associate auxiliary buffers with auxiliary buffers.
Now that there is more direct path available lets start using it
instead and stop even checking for such (im)possibility.
v2 (Ben): Do not signal msaa layout with explicit argument but
using layout_flags instead.
Signed-off-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
Reviewed-by: Ben Widawsky <benjamin.widawsky@intel.com>
Currently the logic allocating and setting up miptrees is closely
combined with decision making when to re-allocate buffers in
X-tiled layout and when to associate colors with auxiliary buffers.
These auxiliary buffers are in turn also represented as miptrees
and are created by the same miptree creation logic calling itself
recursively. This means considering in vain if the auxiliary buffers
should be represented in X-tiled layout or if they should be
associated with auxiliary buffers again.
While this is somewhat unnecessary, this doesn't impose any problems
currently. Miptrees for auxiliary buffers are created as simgle-sampled
fusing the consideration for multi-sampled compression auxiliary
buffers. The format in turn is such that is not applicable for
single-sampled fast clears (that would require accompaning auxiliary
buffer).
But once the driver starts to support lossless compression of color
buffers the auxiliary buffer will have a format that would itself
be applicable for lossless compression. This would be rather
difficult and ugly to detect in the current miptree creation logic,
and therefore this patch seeks to separate the association logic
from the general allocation and setup steps.
v2 (Ben):
- Do not reconsider for X-tiling in intel_miptree_create()
as it was just forced to Y-tiling in miptree_create().
- Do not drop checks for allocation failures.
Signed-off-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
This makes the logic a little more explicit and helps to keep
subsequent patches easier to read.
Suggested-by: Ben Widawsky <benjamin.widawsky@intel.com>
Signed-off-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
Part of brw_try_draw_prims() is a check to validate textures
(brw_validate_textures()). In case of textures that currently have
only level zero but are marked for mipmap generation, i965 driver
will decide to replace the underlying buffer with a larger one
capable of holding also the additional levels. This results into
blit from the original buffer to the newly allocated (see
intel_miptree_copy_teximage()). This blit is currently handled with
blitter engine and hence it won't effect the ongoing draw operation.
However, this blit in turn may trigger color resolve on the source
buffer. In principle, this should be possible with fast cleared
buffers but I only started hitting it when I enabled lossless
compression (that reguires similar resolve to fast cleared buffers).
Now, the color resolve is a meta operation and uses the same drawing
path we are already in middle of. After quite a bit of debugging I
realized that the resolve will modify the current vbo setup but it
won't restore it afterwards resulting in the original draw call
using wrong vertex data.
When brw_try_draw_prims() gets called, the vbo logic in the Mesa
core (see vbo_draw_arrays()) has just bound the vbo (see
vbo_bind_arrays() and recalculate_input_bindings()). Color resolve
operation will overwrite the vbo setup by calling vbo_bind_arrays()
against the resolve rectangle (see brw_draw_rectlist()). Once the
color resolve is done the vbo setup is left to the resolve rectangle
state and the original drawing call yields bogus results.
This patch aims to restore the original state after the color
resolve by calling vbo_bind_arrays() yet again after the vertex
array state in the core context have been restored.
Now having said all this, I'd also like to state that I'm quite
uncomfortable with the nested meta operations. Ths original draw
call in this case is in fact a meta operation itself. It is a blit
from level zero to level one when generating the additional mipmap
levels (see _mesa_meta_GenerateMipmap()). Imagine the complexity
if the blit in the middle from buffer to another would go to meta
path also instead of blitter.
I would very tempted to try to move all the resolves to happen
before a meta operation is started.
Additionally I still feel that work I did earlier in the spring/
summer time moving meta operations to use direct state upload
bypassing the core context would make sense.
v2: Force input recalculation by setting the flag explicitly
v3: Do not attempt to restore vbo for opengles1 which doesn't
support vertex buffer objects.
Signed-off-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
Validation may kick off copies and subsequently color resolves.
Color resolves (and the copies themselves if ending up in meta path)
will overwrite the internal driver state but are not prepared to
restore it. Instead of adding that capability the validation can be
simply performed before the state is updated.
Signed-off-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
Setting brw->ctx.NewDriverState and brw->ctx.NewGLState affects
the dirty bits for the current pipeline. But, we need to flag
everything dirty on *both* pipelines, so that when we switch
back, we'll realize our programs are stale and re-upload them.
To accomplish this, flag the saved state for both pipelines.
Only one of them should matter, but this way we don't have to
check which we need to set. It's harmless to set the other.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=93790
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Tested-by: Ilia Mirkin <imirkin@alum.mit.edu>
Feedback from Khronos is that 'invariant' should be allowed on block
members for desktop OpenGL. Fix piglit regression added by fe1e89a0:
invariant-qualifier-in-out-block-01.vert
v2:
- Allow it for in/out blocks in OpenGL ES too, so when OES_shader_io_blocks
is supported we don't need to do any change (Timothy)
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=89330
Signed-off-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
Reviewed-by: Timothy Arceri <timothy.arceri@collabora.com>
I think this was just missed; Curro and I were probably writing
code simultaneously and forgot to combine them at the end.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
We really need to stop pulling information directly out of shaders for
state setup. For one thing, if we want any sort of an on-disk shader
cache, having all of this metadata in one place is going to be crucial.
Also, passing it all through prog_data cleans up the compiler <-> state
setup API substantially.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
It's extremely FS specific so the fact that we have a stage check in the
middle of it is rather bogus. While were here, we rename
setup_payload_gen4 and setup_payload_gen6 to make it obvious that they are
both FS specific.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
This 'words' parameter is there since 2011 but it has never been used.
While we are at it, get rid of the extern declaration.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
Commit c98deb18d5 in 2010 disallowed embedded struct definitions
in ES. Then in 2013 d9bb8b7b56 disallowed it for everything but
GLSL 1.10.
Commit c98deb18d5 seemed the cleanest way to do the check so its
been extended to cover GL and the other version has been removed.
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
We no longer need to build any part of Mesa with Windows SDK 7.0.7600 or
MSVC 2008. MSVC 2013 will be the oldest we support.
In practice this means people are now free to declare variables in the
middle of blocks, on the whole Mesa tree.
Care should still be taken with variable length arrays and void pointer
arithmetic.
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Hella-acked-by: Ian Romanick <ian.d.romanick@intel.com>
The indirect dispatch registers were whitelisted in command parser
version 5. (Version 5 is available as of Linux 4.4)
Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Broken by one of my cleanups. Spotted by luck.
Radeonsi doesn't care, because all shader create callbacks go to the same
function.
Reviewed-by: Brian Paul <brianp@vmware.com>
When NIR was originally drafted, there was no easy way to determine if
something was constant or not. The result was that we had lots of
special-casing for constant values such as this. Now that load_const
instructions are SSA-only, it's really easy to find constants and this
isn't really needed anymore.
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Reviewed-by: Rob Clark <robclark@gmail.com>
This fixes two issues. First, we had a use-after-free in the case where
the instruction got deleted and we tried to return mov->dest.write_mask.
Second, in the case where we are doing a self-mov of a register, we delete
those channels that are moved to themselves from the write-mask. This
means that those channels aren't reported as being handled even though they
are. We now stash off the write-mask before remove unneeded channels so
that they still get reported as handled.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=94073
Reviewed-by: Matt Turner <mattst88@gmail.com>
Cc: "11.0 11.1" <mesa-stable@lists.freedesktop.org>
This fixes an assertion failure in [at least] one of the Unreal Engine Linux
demo/games that uses DXT1 compression. Specifically, the "Vehicle Game".
At some point, the game ends up trying to blit mip level whose size is 2x2,
which is smaller than a DXT1 block. As a result, the assertion in the blit path
is triggered. It should be safe to simply make sure we align the width and
height, which is sadly an example of compression being less efficient.
NOTE: The demo seems to work fine without the assert, and therefore release
builds of mesa wouldn't stumble over this. Perhaps there is some unnoticeable
corruption, but I had trouble spotting it.
Thanks to Jason for looking at my backtrace and figuring out what was going on.
v2: Use NPOT alignment to make sure ASTC is handled properly (Ilia)
Remove comment about how this doesn't fix other bugs, because it does.
Cc: "11.0 11.1" <mesa-stable@lists.freedesktop.org
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=93358
Signed-off-by: Ben Widawsky <benjamin.widawsky@intel.com>
Tested-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
tl;dr: For many types of GL object, we can *NEVER* use the Gen function.
In OpenGL ES (all versions!) and OpenGL compatibility profile,
applications don't have to call Gen functions. The GL spec is very
clear about how you can mix-and-match generated names and non-generated
names: you can use any name you want for a particular object type until
you call the Gen function for that object type.
Here's the problem scenario:
- Application calls a meta function that generates a name. The first
Gen will probably return 1.
- Application decides to use the same name for an object of the same
type without calling Gen. Many demo programs use names 1, 2, 3,
etc. without calling Gen.
- Application calls the meta function again, and the meta function
replaces the data. The application's data is lost, and the app
fails. Have fun debugging that.
Fixes piglit 'object-namespace-pollution glGetTexImage-compressed
renderbuffer' test.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=92363
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
tl;dr: For many types of GL object, we can *NEVER* use the Gen function.
In OpenGL ES (all versions!) and OpenGL compatibility profile,
applications don't have to call Gen functions. The GL spec is very
clear about how you can mix-and-match generated names and non-generated
names: you can use any name you want for a particular object type until
you call the Gen function for that object type.
Here's the problem scenario:
- Application calls a meta function that generates a name. The first
Gen will probably return 1.
- Application decides to use the same name for an object of the same
type without calling Gen. Many demo programs use names 1, 2, 3,
etc. without calling Gen.
- Application calls the meta function again, and the meta function
replaces the data. The application's data is lost, and the app
fails. Have fun debugging that.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=92363
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
Nothing left in meta does anything with the RBO binding, so we don't
need to save or restore it. The FBO binding is still modified.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
This has the advantage that it does not pollute the global binding
state. It also enables later patches that will stop calling
_mesa_GenRenderbuffers / _mesa_CreateRenderbuffers which pollute the
renderbuffer namespace.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
This has the advantage that it does not pollute the global binding
state. It also enables later patches that will stop calling
_mesa_GenRenderbuffers / _mesa_CreateRenderbuffers which pollute the
renderbuffer namespace.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
Pulls the parts of renderbuffer_storage that aren't just parameter
validation out into a function that can be called from other parts of
Mesa (e.g., meta).
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
This function previously was only used in fbobject.c and contained a
bunch of API validation. Split the function into
framebuffer_renderbuffer that is static and contains the validation, and
_mesa_framebuffer_renderbuffer that is suitable for calling from
elsewhere in Mesa (e.g., meta).
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
The texture slot is expanded to 16 dwords containing 2 descriptors.
Those can be:
- Image and fmask, or
- Image and sampler state
By carefully choosing the locations, we can put all three into one slot,
with the fmask and sampler state being mutually exclusive.
This improves shaders in 2 ways:
- 2 user SGPRs are unused, shaders can use them as temporary registers now
- each pair of descriptors is always on the same cache line
v2: cosmetic changes: add back v8i32, don't load a sampler state & fmask
at the same time
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
v2: Clarify the relation between num_tiles_pipes and GB_TILE_MODE and the fix
needed for Tahiti as suggested by Marek.
Signed-off-by: Alexandre Demers <alexandre.f.demers@gmail.com>
Signed-off-by: Marek Olšák <marek.olsak@amd.com>
This avoids a possible NULL dereference because ureg_create() might
return a NULL pointer.
Spotted by coverity.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Two things were broken here:
- The depth/stencil surface dimensions were broken for MSAA.
- Sample count was programmed incorrectly.
Result was the depth resolve didn't work correctly on MSAA surfaces, and
so sampling the surface later produced garbage.
Fixes the new piglit test arb_texture_multisample-sample-depth, and
various artifacts in 'tesseract' with msaa=4 glineardepth=0.
Fixes freedesktop bug #76396.
Not observed any piglit regressions on Haswell.
v2: Just set brw_hiz_op_params::dst.num_samples rather than adding a
helper function (Ken).
Signed-off-by: Chris Forbes <chrisf@ijw.co.nz>
v3: moved the alignment needed for hiz+msaa to brw_blorp.cpp, as
suggested by Chad Versace (Alejandro Piñeiro on behalf of Chris
Forbes)
Signed-off-by: Alejandro Piñeiro <apinheiro@igalia.com>
Reviewed-by: Ben Widawsky <benjamin.widawsky@intel.com>
Tested-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
The assertion is inside a condition mandating num_samples > 1 and
therefore the first half of the constraint is always met. The
second half in turn would only be applicable for single sampled
case and moreover it is trying to falsely check against surface
type instead of format.
Subsequent patches will introduce proper support for the lossless
compression and dropping this here makes the patches a little
simpler.
Signed-off-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
Reviewed-by: Ben Widawsky <benjamin.widawsky@intel.com>
This is no longer necessary...and it doesn't make much sense to
have inputs as destinations.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Chris Forbes <chrisf@ijw.co.nz>
gl_PointSize is delivered in the .w component of the VUE header, while
the language expects it to be a float (and thus in the .x component).
Previously, we emitted MOVs to copy it over to the .x component.
But this is silly - we can just use a .wwww swizzle and access it
without copying anything or clobbering the value stored at .x
(which admittedly is useless).
Removes the last use of ATTR destinations.
v2: Use BRW_SWIZZLE_WWWW, not SWIZZLE_WWWW (caught by GCC).
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Chris Forbes <chrisf@ijw.co.nz>
This patch re-implements the pre-Haswell VS attribute workarounds.
Instead of emitting shader code in the vec4 backend, we now simply
call a NIR pass to emit the necessary code.
This simplifies the vec4 backend. Beyond deleting code, it removes
the primary use of ATTR as a destination. It also eliminates the
requirement that the vec4 VS backend express the ATTR file in terms
of VERT_ATTRIB_* locations, giving us a bit more flexibility.
This approach is a little different: rather than munging the attributes
at the top, we emit code to fix them up when they're accessed. However,
we run the optimizer afterwards, so CSE should eliminate the redundant
math. It may even be able to fuse it with other calculations based on
the input value.
shader-db does not handle non-default NOS settings, so I have no
statistics about this patch.
Note that the scalar backend does not implement VS attribute
workarounds, as they are unnecessary on hardware which allows SIMD8 VS.
v2: Do one multiply for FIXED rescaling and select components from
either the original or scaled copy, rather than multiplying each
component separately (suggested by Matt Turner).
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Chris Forbes <chrisf@ijw.co.nz>
Use st->internal_target instead of PIPE_TEXTURE_2D when choosing the
texture format. Probably no real difference, but let's be consistent.
Simplify a test when determining whether we need normalized texcoords.
Add a new assertion.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Bitmaps may be drawn with a PIPE_TEXTURE_2D or PIPE_TEXTURE_RECT resource
as determined at context creation by checking if PIPE_CAP_NPOT_TEXTURES is
supported. But many places in the bitmap code were hard-coded to use
PIPE_TEXTURE_2D. Use st->internal_target instead.
I think an older NV chip is the only case where a gallium driver does not
support NPOT textures. Bitmap drawing was probably broken for that GPU.
Also, we only need one sampler state with texcoord normalization set up
according to st->internal_target.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
The fast path for Intel's ReadPixels() unintentionally omits clipping
the specified area to a valid one. Rather than clip in various
corner-cases, perform this operation in the API validation stage.
The bug in intel_readpixels_tiled_memcpy() showed itself when the winsys
ReadBuffer's height was smaller than the one specified by ReadPixels().
yoffset became negative, which was an invalid input for tiled_to_linear().
v2: Move clipping to validation stage (Jason)
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=92193
Reported-by: Marta Löfstedt <marta.lofstedt@intel.com>
Cc: "11.0 11.1" <mesa-stable@lists.freedesktop.org>
Signed-off-by: Nanley Chery <nanley.g.chery@intel.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
These logical texture instructions can have a *lot* of sources. It's much
safer if we have symbolic names for them.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
This commit adds the capability to NIR to support separate textures and
samplers. As it currently stands, glsl_to_nir only sets the texture deref
and leaves the sampler deref alone as it did before and nir_lower_samplers
assumes this. Backends can still assume that they are combined and only
look at only at the texture index. Or, if they wish, they can assume that
they are separate because nir_lower_samplers, tgsi_to_nir, and prog_to_nir
all set both texture and sampler index whenever a sampler is required (the
two indices are the same in this case).
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
We're about to separate the two concepts. When we do, the sampler will
become optional. Doing a rename first makes the separation a bit more
safe because drivers that depend on GLSL or TGSI behaviour will be fine to
just use the texture index all the time.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Bit 0 of the Patch Header is "TR DS Cache Disable". Setting that bit
disables the DS Cache for tessellator-output topologies resulting in
stitch-transition regions (but leaves it enabled for other cases).
We probably shouldn't leave this to chance - the URB could contain
garbage - which could result in the cache randomly being turned on
or off.
This patch makes the final EOT write 0 to the first DWord (which
only contains this one bit). This ensures the cache is always on.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Direct access to intr->const_index[n], where different slots have
different meanings, is somewhat confusing.
Instead, let's put some extra info in nir_intrinsic_infos[] about which
slots map to what, and add some get/set helpers. The helpers validate
that the field being accessed (base/writemask/etc) is applicable for the
intrinsic opc, for some extra safety. And nir_print can use this to
dump out decoded const_index fields.
Signed-off-by: Rob Clark <robclark@freedesktop.org>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
If the only stage is MESA_SHADER_COMPUTE, we should complain that
there's nothing coming out of the geometry shader stage just as
we would if the first stage were MESA_SHADER_FRAGMENT.
Also, it's valid for tessellation shaders to be the stage producing
transform feedback varyings, so mention those in the compiler error.
Found by inspection.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Timothy Arceri <timothy.arceri@collabora.com>
This fixes FP16 conversion instructions for VI, which has 16-bit floats,
but not SI & CI, which can't disable denorms for those instructions.
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
It was partly a state and partly emulated by shader code, but since we want
to do this in a fragment shader prolog, we need to put it into the shader
key, which will be used to generate the prolog.
This also removes the spi_ps_input states and moves the registers
to the PS state.
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
BCOLOR inputs were immediately after COLOR inputs. Thus, all following inputs
were offset by 1 if color_two_side was enabled, and not offset if it was not
enabled, which is a variation that's problematic if we want to have 1 variant
per shader and the variant doesn't care about color_two_side (that should be
handled by other bytecode attached at the beginning).
Instead, move BCOLOR inputs after all other inputs, so BCOLOR0 is at location
"num_inputs" if it's present. BCOLOR1 is next.
This also allows removing si_shader::nparam and
si_shader::ps_input_param_offset, which are useless now.
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
When glCallLists() is compiled into a display list, preserve the call
as a single glCallLists rather than 'n' glCallList calls. This will
matter for an upcoming display list optimization project.
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Generate GL_INVALID_VALUE if n < 0. Return early if n==0 or lists==NULL.
v2: fix formatting, also check for lists==NULL.
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Most apps don't use glBitmap so don't allocate the bitmap cache or
gallium state objects/shaders/etc until the first call to st_Bitmap().
v2: simplify a conditional, per Gustaw Smolarczyk.
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Move setup/restoration of rendering state into helper functions.
This makes the draw_bitmap_quad() function much more concise.
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Both st/mesa and i965 should return a true/false result now, and the
only other driver implementing queries (radeon) doesn't support
ARB_occlusion_query2 which added that pname.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
This reduces code duplication. It also adds support for drivers where the
fragment position is a system value.
Suggested-by: Jose Fonseca <jfonseca@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
These are used in GLSL IR to removed unused varyings and match
transform feedback variables. There is no need to use these in NIR.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
The existing code was very hard to follow and has been the source
of at least 3 bugs in the past year.
The existing code also has a bug for SSO where if we have a
multi-stage SSO for example a tes -> gs program, if we try to use
transform feedback with gs the existing code would look for the
transform feedback varyings in the tes stage and fail as it can't
find them.
V2: Add more code comments, always try to remove unused inputs
to the first stage.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
We really just needed to skip the existing ES < 3.1 check if we have
a compute shader, all other scenarios are already covered.
* No shaders is a link error.
* Geom or Tess without Vertex is a link error which means we always
require a Vertex shader and hence a Fragment shader.
* Finally a Compute shader linked with any other stage is a link error.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Previously an empty program would go through the entire
link_shaders() function and we would have to be careful
not to cause a segfault.
In core profile also now set link_status to false by
generating an error, it was previously set to true.
From Section 7.3 (PROGRAM OBJECTS) of the OpenGL 4.5 spec:
"Linking can fail for a variety of reasons as specified in the
OpenGL Shading Language Specification, as well as any of the
following reasons:
- No shader objects are attached to program."
V2: Only generate an error in core profile and add spec quote (Ian)
V3: generate error in ES too, remove previous check which was only
applying the rule to GL 4.5/ES 3.1 and above. My understand is that
this spec change is clarifying previously undefined behaviour and
therefore should be applied retrospectively. The ES CTS tests for
this are in ES 2 I suspect it was passing because it would have
generated an error for not having both a vertex and fragment shader.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Helps 11 shaders in UnrealEngine4 demos.
I seriously hope they would have given us bitfieldReverse() if we
exposed GL 4.0 (but we do expose ARB_gpu_shader5, so why not use that
anyway?).
instructions in affected programs: 4875 -> 4633 (-4.96%)
cycles in affected programs: 270516 -> 244516 (-9.61%)
I suspect there's a *lot* of room to improve nir_search/opt_algebraic's
handling of this. We'd actually like to match, e.g., step2 by matching
step1 once and then doing a pointer comparison for the second instance
of step1, but unfortunately we generate an enormous tuple for instead.
The .text size increases by 6.5% and the .data by 17.5%.
text data bss dec hex filename
22957 45224 0 68181 10a55 nir_libnir_la-nir_opt_algebraic.o
24461 53160 0 77621 12f35 nir_libnir_la-nir_opt_algebraic.o
I'd be happy to remove this if Unreal4 uses bitfieldReverse() if it is
in a GL 4.0 context once we expose GL 4.0.
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
The next patch adds an algebraic rule that uses the constant 0xff00ff00.
Without this change, the build fails with
return hex(struct.unpack('I', struct.pack('i', self.value))[0])
struct.error: 'i' format requires -2147483648 <= number <= 2147483647
The hex() function handles integers of any size, and assigning a
negative value to an unsigned does what we want in C. The pack/unpack is
unnecessary (and as we see, buggy).
Reviewed-by: Dylan Baker <baker.dylan.c@gmail.com>
Walking the SSA definitions in order means that we consider the smallest
algebraic optimizations before larger optimizations. So if a smaller
rule is part of a larger rule, the smaller one will happen first,
preventing the larger one from happening.
instructions in affected programs: 32721 -> 32611 (-0.34%)
helped: 106
In programs whose nir_optimize loop count changes (129 of them):
before: 1164 optimization loops
after: 1071 optimization loops
Of the 129 affected, 16 programs' optimization loop counts increased.
Prevents regressions and annoyances in the next commits.
Reviewed-by: Eduardo Lima Mitev <elima@igalia.com>
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
I don't know why, but we never hooked up this pass Eric wrote.
Otherwise, you can end up with stupid scalarized code such as:
vec4 ssa_7 = load_const (0.0, 0.0, 0.0, 0.0)
vec4 ssa_8 = ...
vec1 ssa_9 = feq ssa_8, ssa_7
vec1 ssa_10 = feq ssa_8.y, ssa_7.y
vec1 ssa_11 = feq ssa_8, ssa_7.z
vec1 ssa_12 = feq ssa_8.y, ssa_7.w
ssa_8.xyxy == <0, 0, 0, 0> should only take two feq instructions.
shader-db on Skylake:
total instructions in shared programs: 9121153 -> 9120749 (-0.00%)
instructions in affected programs: 32421 -> 32017 (-1.25%)
helped: 277
HURT: 69
total cycles in shared programs: 69003364 -> 69000912 (-0.00%)
cycles in affected programs: 899186 -> 896734 (-0.27%)
helped: 313
HURT: 403
This also prevents regressions when disabling channel expressions.
v2: Don't call opt_cse afterwards (requested by Matt). It should
happen in the optimization loop below anyway.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Eduardo Lima Mitev <elima@igalia.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
The aim of this is to work towards removing UniformHash from the program
struct so that we don't need to hold onto it in memory and pass it around
outside the linker.
Reviewed-by: Dave Airlie <airlied@redhat.com>
There are never render target reads, so there are no scheduling hazards.
Giving the extra flexibility to the scheduler makes it possible to do
FB writes as soon as their sources are available, reducing register
pressure. It also makes it possible to do the payload setup for more
than one FB write message at a time, which could better hide latency.
shader-db results on Skylake:
total instructions in shared programs: 9110254 -> 9110211 (-0.00%)
instructions in affected programs: 2898 -> 2855 (-1.48%)
helped: 3
HURT: 0
LOST: 0
GAINED: 1
A reduction in instruction counts is surprising, but legitimate:
the three shaders helped were spilling, and reducing register
pressure allowed us to issue fewer spills/fills.
total cycles in shared programs: 69035108 -> 68928820 (-0.15%)
cycles in affected programs: 4412402 -> 4306114 (-2.41%)
helped: 4457
HURT: 213
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
reuse the sampler deref handling code to do the same
thing for atomics.
Acked-by: Ilia Mirkin <imirkin@alum.mit.edu>
Signed-off-by: Dave Airlie <airlied@redhat.com>
The state tracker never handled this properly, and it finally
annoyed me for the second time so I decided to fix it properly.
This is inspired by the NIR sampler lowering code and I only realised
NIR seems to do its deref ordering different to GLSL at the last
minute, once I got that things got much easier.
it fixes a bunch of tests in
tests/spec/arb_gpu_shader5/execution/sampler_array_indexing/
v2: fix AoA tests when forced on.
I was right I didn't need all that code, fixing the AoA code
meant cleaning up a chunk of code I didn't like in the array
handling.
v3: start generalising the code a bit more for atomics.
v3.1: use UniformRemapTable
v4: handle uniforms differently using the param_index,
and go back to UniformStorage
fix issues identified by Timothy with deref handling.
v4.1: squash const fix and move handling 1D const out
of recursive function.
Reviewed-by: Timothy Arceri <timothy.arceri@collabora.com>
Acked-by: Ilia Mirkin <imirkin@alum.mit.edu>
Signed-off-by: Dave Airlie <airlied@redhat.com>
We have a requirement to store the index into the mesa parameterlist
for uniforms. Up until now we've overwritten var->data.location with
this info. However this then stops us accessing UniformStorage,
which is needed to do proper dereferencing.
Add a new variable to ir_variable to store this value in, and change
the two uses to use it correctly.
Reviewed-by: Timothy Arceri <timothy.arceri@collabora.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
Its previous name was somewhat misleading, this really behaves like a
RW cache flush rather than an invalidation.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
The state cache is also L3-backed so it seems sensible to make sure
it's clean as we do for other RO caches before repartitioning the L3.
This wasn't part of my original L3 partitioning code because I was
able to reproduce hangs on Gen7 hardware when the state cache
invalidation happened asynchronously with previous 3D rendering, which
should no longer be possible after the previous change.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
We need to split the stalling flush from the RO cache invalidation
into a different PIPE_CONTROL command to make sure that the top of the
pipe invalidation happens after any previous rendering is complete.
Otherwise it's possible for previous rendering to pollute the L3 cache
in the short window of time between RO invalidation and the completion
of the stalling flush. Fixes rendering artifacts on Unigine Heaven,
Metro Last Light Redux and Metro 2033 Redux.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=93540
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=93599
Tested-by: Darius Spitznagel <d.spitznagel@goodbytez.de>
Tested-by: Martin Peres <martin.peres@linux.intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
The SEL instruction with predication mode NONE emitted when the atomic
operation doesn't need to be predicated is a no-op and might rely on
undocumented hardware behaviour. Noticed by chance while looking at
the assembly output.
Reviewed-by: Matt Turner <mattst88@gmail.com>
The errors.c file had grown quite large so split off this extension
code into its own file. This involved making a handful of functions
non-static.
Acked-by: Timothy Arceri <timothy.arceri@collabora.com>
This fixes a crash with bin/arb_clear_texture-base-formats and
probably some other tests which use clear_texture().
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
The builtin data can get released with a glReleaseShaderCompiler call.
We're careful everywhere to clone everything that comes out of builtins
except here, where we accidentally return the signature belonging to the
builtin version, rather than the locally-cloned one.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Timothy Arceri <timothy.arceri@collabora.com>
Tested-by: Rob Herring <robh@kernel.org>
Cc: mesa-stable@lists.freedesktop.org
The builtin function shader is part of the builtin state, released
when glReleaseShaderCompiler is called. We must ensure that the
builtins have been (re)initialized before attempting to link with the
builtin shader.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Timothy Arceri <timothy.arceri@collabora.com>
Tested-by: Rob Herring <robh@kernel.org>
Cc: mesa-stable@lists.freedesktop.org
All interface blocks will have been lowered by this point so just
use an assert. Returning false would have caused all sorts of
problems if they were not lowered yet and there is an assert to
catch this later anyway.
We also update the tests to reflect this change.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
The vec4 backend, at the end, does this:
if (inst->is_3src()) {
for (int i = 0; i < 3; i++) {
if (inst->src[i].vstride == BRW_VERTICAL_STRIDE_0)
assert(brw_is_single_value_swizzle(inst->src[i].swizzle));
So make sure that we use the same conditions when trying to
copy-propagate. UNIFORMs will be converted to vstride 0 in
convert_to_hw_regs, but so will ATTRs when interleaved (as will happen
in a GS with multiple attributes). Since the vstride is not set at
copy-prop time, infer it by inspecting dispatch_mode and reject ATTRs if
they have non-scalar swizzles and are interleaved.
Fixes assertion errors in dolphin-generated geometry shaders (or
misrendering on opt builds) on Sandybridge or on IVB/HSW with
INTEL_DEBUG=nodualobj.
Co-authored-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=93418
Cc: "11.0 11.1" <mesa-stable@lists.freedesktop.org>
If you're worried about the duplication of some CAPs, we can remove them
later.
v2: add fields for memory eviction stats
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
GLsync objects had a race condition when used from multiple threads
(which is the main point of the extension, really); it could be
validated as a sync object at the beginning of the function, and then
deleted by another thread before use, causing crashes. Fix this by
changing all casts from GLsync to struct gl_sync_object to a new
function _mesa_get_and_ref_sync() that validates and increases
the refcount.
In a similar vein, validation itself uses _mesa_set_search(), which
requires synchronization -- it was called without a mutex held, causing
spurious error returns and other issues. Since _mesa_get_and_ref_sync()
now takes the shared context mutex, this problem is also resolved.
Fixes bug #92757, found while developing Nageru, my live video mixer
(due for release at FOSDEM 2016).
v2: Marek: silence warnings, fix declaration after code
Signed-off-by: Steinar H. Gunderson <sesse@google.com>
Cc: "11.0 11.1" <mesa-stable@lists.freedesktop.org>
Signed-off-by: Marek Olšák <marek.olsak@amd.com>
Yet another change motivated by AMD GPUPerfStudio compatibility. These groups
are not directly accessible from userspace, and AMD GPUPerfStudio does not
actually query them - it just requires them to be there. Hence, adding
a placeholder for now.
Reviewed-by: Edward O'Callaghan <eocallaghan@alterapraxis.com>
Acked-by: Marek Olšák <marek.olsak@amd.com>
This is yet another change motivated by appeasing AMD GPUPerfStudio's
hardcoding of performance counter group numbers.
Reviewed-by: Edward O'Callaghan <eocallaghan@alterapraxis.com>
Acked-by: Marek Olšák <marek.olsak@amd.com>
As documented in the comment, AMD GPUPerfStudio unfortunately hardcodes the
order of performance counter groups. Let's do the pragmatic thing and present
the same order as Catalyst/Crimson.
Reviewed-by: Edward O'Callaghan <eocallaghan@alterapraxis.com>
Acked-by: Marek Olšák <marek.olsak@amd.com>
This group was used by older versions of AMD GPUPerfStudio (via
AMD_performance_monitor) to identify the GPU family, and GPUPerfStudio
still complains when it isn't available.
Reviewed-by: Edward O'Callaghan <eocallaghan@alterapraxis.com>
Acked-by: Marek Olšák <marek.olsak@amd.com>
Set R600_DEBUG=preoptir to dump the LLVM IR before optimization passes,
to allow diagnosing problems caused by optimization passes.
Note that in order to compile the resulting IR with llc, you will first
have to run at least the mem2reg pass, e.g.
opt -mem2reg -S < shader.ll | llc -march=amdgcn -mcpu=bonaire
Signed-off-by: Michel Dänzer <michel.daenzer@amd.com> (original patch)
Signed-off-by: Nicolai Hähnle <nicolai.haehnle@amd.com> (w/ debug flag)
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
We can get rid of our reference immediately, since the driver will hold
onto it for us.
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
While rather unlikely, uploads _can_ fail. Doing them earlier means
we'll have to restore less state when they do fail, and it's slightly
easier to check the restore code.
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Previously the framebuffer default sample count was taken directly
from the value given by the application. On the i965 driver on HSW if
the value wasn't one that is supported by the hardware it would hit an
assert when it tried to program the state for it. This patch fixes it
by adding a derived sample count to the state for the default
framebuffer. The driver can then quantize this to one of the valid
values in its UpdateState handler when the _NEW_BUFFERS state changes.
_mesa_geometric_samples is changed to use the new derived value.
Fixes the piglit test arb_framebuffer_no_attachments-query
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=93957
Cc: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
Fixup to commit 03b3eb90d - the number of buffers could be larger than
the number of elements, in which case we'd pass a negative argument to
PUSH_SPACE, which would be bad. While we're at it, merge it with the
other PUSH_SPACE at the top of the function.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Cc: mesa-stable@lists.freedesktop.org
The spill logic will insert convert ops when moving between files. It
seems like the emission logic wasn't quite ready for these converts.
Tested on fermi, and visually looked at nvdisasm output for maxwell.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Cc: mesa-stable@lists.freedesktop.org
Use align_free to free memory allocated
with align_malloc.
Signed-off-by: Axel Davy <axel.davy@ens.fr>
Reviewed-by: Patrick Rudolph <siro@das-labor.org>
The color inputs must automatically use centroid whether
multisampling is used or not.
Signed-off-by: Axel Davy <axel.davy@ens.fr>
Reviewed-by: Patrick Rudolph <siro@das-labor.org>
sem.reg.mod & NINED3DSPDM_CENTROID is worth 4 when
centroid is requested, whereas
TGSI_INTERPOLATE_LOC_CENTROID is worth 1.
Signed-off-by: Axel Davy <axel.davy@ens.fr>
This enables to use fast clears in the following
case:
pixel shader renders to 1 RT
4 RT bound
clear
new pixel shader bound that renders to 4 RTs
Previously the fast clear path wouldn't be hit,
because when trying the fast clear path,
the framebuffer state would be configured for 1 RT,
instead of 4.
Signed-off-by: Axel Davy <axel.davy@ens.fr>
Reviewed-by: Patrick Rudolph <siro@das-labor.org>
Some docs say linear filtering is always used when
app does shadow mapping.
Signed-off-by: Axel Davy <axel.davy@ens.fr>
Reviewed-by: Patrick Rudolph <siro@das-labor.org>
Respect block alignment for ATI1/ATI2 format when trying to lock a
surface using LockRect().
Fixes failing WINE tests device.c test_surface_blocks() tests.
Signed-off-by: Patrick Rudolph <siro@das-labor.org>
Reviewed-by: Axel Davy <axel.davy@ens.fr>
Testing Win behaviour seems to show wrong states
are accepted, but then depending on the states
some specific 'good' behaviours happen.
This adds some validation to catch invalid
states and have these 'good' behaviours
when it happens.
Also reorders SetRenderState to match the expected
optimisation:
(Value == previous Value) => return immediately,
which affects D3D9 hacks too.
Signed-off-by: Axel Davy <axel.davy@ens.fr>
Signed-off-by: Patrick Rudolph <siro@das-labor.org>
Add config option override_vendorid to report a fake card in d3dadapter9 drm.
Signed-off-by: Patrick Rudolph <siro@das-labor.org>
Reviewed-by: Axel Davy <axel.davy@ens.fr>
Implement a device private memory counter similar to Win 7.
Only textures and surfaces increment vidmem and may return
ERR_OUTOFVIDEOMEMORY. Vertexbuffers and indexbuffers creation always
succeedes, even when out of video memory.
Fixes "Vampire: The Masquerade - Bloodlines" allocating resources until crash.
Fixes "Age of Conan" allocating resources until crash.
Fixes failing WINE test device.c test_vidmem_accounting().
Signed-off-by: Patrick Rudolph <siro@das-labor.org>
Reviewed-by: Axel Davy <axel.davy@ens.fr>
Apps can know if the window is occluded by checking for
specific error messages. The behaviour is different
for Device9 and Device9Ex.
This allow games to release the mouse and stop rendering
until the focus is restored.
In case of multiple swapchain we do care only of the device one.
Signed-off-by: Patrick Rudolph <siro@das-labor.org>
Reviewed-by: Axel Davy <axel.davy@ens.fr>
To keep compatible with older ID3DPresent interfaces (used to talk
with Wine), store the minor version num accessible to all
statetracker functions (in the NineDevice9 structure).
Signed-off-by: Patrick Rudolph <siro@das-labor.org>
Reviewed-by: Axel Davy <axel.davy@ens.fr>
flush_resource needs to be called before flush (for
fast clear resolve, etc).
Removes useless computation of resource (it is
already set correctly).
Signed-off-by: Axel Davy <axel.davy@ens.fr>
Reviewed-by: Patrick Rudolph <siro@das-labor.org>
Return D3DERR_INVALIDCALL instead of E_POINTER.
On error set ppBackBuffer to NULL.
Multiple swapchains can only be created in windowed mode as
windowed swapchain.
Set backbuffer to NULL in NineDevice9_GetBackBuffer, but not
in NineSwapChain9_GetBackBuffer.
This fixes all WINE's device.c test_swapchain() tests.
Signed-off-by: Patrick Rudolph <siro@das-labor.org>
Reviewed-by: Axel Davy <axel.davy@ens.fr>
When no window is specified, we should revert to the focus window.
This deserves more tests however (what if the device swapchain is
already using the focus window ?)
Fixes crash for FFXIV
Signed-off-by: Axel Davy <axel.davy@ens.fr>
Reviewed-by: Patrick Rudolph <siro@das-labor.org>
In case swapchain creation fails This->swapchains[i] might be NULL and
causes a crash.
Signed-off-by: Patrick Rudolph <siro@das-labor.org>
Reviewed-by: Axel Davy <axel.davy@ens.fr>
Return errors in case of invalid presentation parameters.
Fixes failing WINE tests device.c test_swapchain_parameters().
Signed-off-by: Patrick Rudolph <siro@das-labor.org>
Reviewed-by: Axel Davy <axel.davy@ens.fr>
Store a copy of GUID in the header that is under our control and use it
as key for the hashtable instead of using the application provided pointer.
The application might change the memory after leaving the function.
Fixes a crash for issue https://github.com/iXit/Mesa-3D/issues/130
Signed-off-by: Patrick Rudolph <siro@das-labor.org>
Reviewed-by: Axel Davy <axel.davy@ens.fr>
To ease debugging print the GUID instead of the pointer to it.
Signed-off-by: Patrick Rudolph <siro@das-labor.org>
Reviewed-by: Axel Davy <axel.davy@ens.fr>
The values of box.z and box.depth weren't set and lead to a crash.
Signed-off-by: Patrick Rudolph <siro@das-labor.org>
Reviewed-by: Axel Davy <axel.davy@ens.fr>
Tests show in case of multisample mismatch between the depth-stencil
buffer and the render target, then it is not cleared.
Fixes failing WINE test visual.c test_multisample_mismatch().
Signed-off-by: Patrick Rudolph <siro@das-labor.org>
Reviewed-by: Axel Davy <axel.davy@ens.fr>
Fixes crash for non-square textures.
We were using the height instead of the
width for some calculations.
Signed-off-by: Axel Davy <axel.davy@ens.fr>
Reviewed-by: Patrick Rudolph <siro@das-labor.org>
Add support for D3DFMT_R8G8B8. It allows format conversion for
surfaces of pool scratch.
Usually gallium formats equivalents for d3d9 formats
have their names reversed.
The gallium format PIPE_FORMAT_R8G8B8_UNORM is the right
equivalent here, and its name is likely wrong (reversed).
Fixes a crash in TmNationsForever.
Signed-off-by: Patrick Rudolph <siro@das-labor.org>
Reviewed-by: Axel Davy <axel.davy@ens.fr>
Shade mode flat is only working if pixelshaders have interpolate
set to TGSI_INTERPOLATE_COLOR on color inputs.
Fixes failing WINE tests visual.c test_shademode().
Signed-off-by: Patrick Rudolph <siro@das-labor.org>
Reviewed-by: Axel Davy <axel.davy@ens.fr>
Report success instead of failing as there's no resource for those surfaces.
Fixes a crash in Crysis: Warhead.
Signed-off-by: Patrick Rudolph <siro@das-labor.org>
Reviewed-by: Axel Davy <axel.davy@ens.fr>
Previous vertex elements code update
was protected by
'if ((group & (NINE_STATE_VDECL | NINE_STATE_VS)) ||
state->changed.stream_freq & ~1)'
itself protected by
'if (group & (NINE_STATE_COMMON | NINE_STATE_VS))'
If no state is changed except the stream frequency,
no update would happen.
This patch solves the problem by adding a new
NINE_STATE_STREAMFREQ state.
Another way would be to add state->changed.stream_freq & ~1
check to the main test.
Signed-off-by: Axel Davy <axel.davy@ens.fr>
Reviewed-by: Patrick Rudolph <siro@das-labor.org>
Some apps do redundant SetStreamSourceFreq calls.
Catch them to improve performance.
Signed-off-by: Axel Davy <axel.davy@ens.fr>
Reviewed-by: Patrick Rudolph <siro@das-labor.org>
The indexbuffer9 codebase was lagging behind the one of vertexbuffer9.
Add buffer9 as common code base for indexbuffer9 and vertexbuffer9.
Signed-off-by: Patrick Rudolph <siro@das-labor.org>
Reviewed-by: Axel Davy <axel.davy@ens.fr>
This seems cleaner to actually reference the resources for vtxbuf,
rather than relying on the fact the bound d3d streams do.
Signed-off-by: Axel Davy <axel.davy@ens.fr>
When an application sets a vertex shader, we are supposed
to use it, and when no vertex shader are set, we are supposed
to revert to fixed function vertex shader.
It seems there is an exception: when the vertex declaration
has a position_t index, we should revert to fixed function
vertex shader.
Up to know we were checking if device->state.vs is set
to know whether to use programmable shader or not.
With this commit we determine whether we use programmable shader
or not when vertex shader/declaration are set, but
stateblocks do complicate things a bit.
Signed-off-by: Axel Davy <axel.davy@ens.fr>
Reviewed-by: Patrick Rudolph <siro@das-labor.org>
NineUnknown_ctor increments the refcount even in case of an error.
Restructure the code to prevent refcount increments.
Fixes a couple of wine tests.
Signed-off-by: Patrick Rudolph <siro@das-labor.org>
Reviewed-by: Axel Davy <axel.davy@ens.fr>
Textures in SYSTEMMEM don't have resources attached.
Instead of returning an error for them, StretchRect
was crashing.
This changes the check order to fix that case.
Signed-off-by: Axel Davy <axel.davy@ens.fr>
Reviewed-by: Patrick Rudolph <siro@das-labor.org>
The last weighted element is one minus the sum of all previous weights.
Fixes WINE test visual.c test_vertex_blending.
Signed-off-by: Patrick Rudolph <siro@das-labor.org>
Reviewed-by: Axel Davy <axel.davy@ens.fr>
In case of non local viewer the value has to be subtracted.
Fixes failing WINE tests in test_specular_lighting() (visual.c)
Signed-off-by: Patrick Rudolph <siro@das-labor.org>
Reviewed-by: Axel Davy <axel.davy@ens.fr>
Implement fixed function D3DRS_SPECULARENABLE.
Fixes failing WINE tests in test_specular_lighting() (visual.c)
Signed-off-by: Patrick Rudolph <siro@das-labor.org>
Reviewed-by: Axel Davy <axel.davy@ens.fr>
vs1.1 rounds a0 to lowest integer, while
other versions do round to closest.
To use the same path as the other versions (with ARR),
we were substracting 0.5 for vs1.1 to get round to lowest.
This gives wrong result if a0 is set to 0:
round(0 - 0.5) = -1
Instead just use ARL for vs1.1
Signed-off-by: Axel Davy <axel.davy@ens.fr>
Reviewed-by: Patrick Rudolph <siro@das-labor.org>
The documentation of the flag doesn't make sense.
To sum up the doc, if not set, specular alpha contains fog,
and if set specular alpha contains 0 (except for ff).
However in practice when the flag is there, apps do use specular alpha
as if it could be used normally, which makes much more sense than the doc.
Signed-off-by: Axel Davy <axel.davy@ens.fr>
Reviewed-by: Patrick Rudolph <siro@das-labor.org>
Based on a gallivm patch by Ilia Mirkin.
+8 piglit regressions due to precision issues (I blame the tests)
The benefit is that we'll get v_cvt_f32_f16 and v_cvt_f16_f32 instead
of emulation with integer instructions. They are GLSL 4.00 intrinsics.
Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>
No instruction counts changed, but:
total cycles in shared programs: 64834502 -> 64781530 (-0.08%)
cycles in affected programs: 16331544 -> 16278572 (-0.32%)
helped: 4757
HURT: 4288
GAINED: 66
LOST: 20
I remember trying this when I first wrote the pass, but it wasn't
helpful at the time.
Reviewed-by: Francisco Jerez <currojerez@riseup.net>
We must fetch all sources into the instruction stream before generating
the instruction that uses them. Otherwise we'll define values after
using them, which won't work so well.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Tested-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
The EXT spec has been updated to:
- logically combine the es2_profile and es_profile exts
- allow any legal version to be requested
dEQP tests request a specific ES version when using GLX, so this allows
dEQP upstream to run against GLX with the appropriate X server patch
(which had similar disabling logic).
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Matt Turner <mattst88@gmail.com> (v1)
Reviewed-by: Adam Jackson <ajax@redhat.com> (v3)
v1 -> v2:
- distinguish between DRI_API_GLES{,2,3}
- add GLX_EXT_create_context_es_profile client-side support
v2 -> v3:
- fix error in computing mask
While this is the default, private .emacs files might have it set to
something else. No harm in forcing it to 0.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
To prevent build failures when a large patch series is committed, like
happened in https://ci.appveyor.com/project/jrfonseca-fdo/mesa/build/322
due to 10 commits between dac2964f3e and
6f428328d3 where submitted before the
build slave started the git clone.
100 commits should be bigger than any patch series seen in practice, and
it takes practically the same time to download as 5 commits.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
We seem to end up w/ duplication between compiler/Makefile.sources and
compiler/glsl/Makefile.sources. The latter appears unused. Delete it.
Signed-off-by: Rob Clark <robclark@freedesktop.org>
This parameter is equivalent to the corresponding OpenGL implementation
limit which is in texels, not bytes.
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
This is already used internally in si_resource_copy_region for compressed
textures, so the only real change here is the adjusted surface size
computation.
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Edward O'Callaghan <eocallaghan@alterapraxis.com>
We will write our own version of texsubimage for PBO uploads, and we will
want to call that here as well.
Reviewed-by: Edward O'Callaghan <eocallaghan@alterapraxis.com>
Use instancing to generate two triangles for each destination layer and use
a geometry shader to route the layer index.
v2:
- directly write layer in VS if supported by the driver (Marek Olšák)
Reviewed-by: Edward O'Callaghan <eocallaghan@alterapraxis.com>
Create a PIPE_BUFFER sampler view on the pixel-unpack buffer, and draw
the image on the texture with a fragment shader that maps fragment
coordinates to buffer coordinates.
Modifications by Nicolai Hähnle:
- various cleanups and fixes (e.g. error handling, corner cases)
- split try_pbo_upload into two functions, which will allow code to be
shared with compressed texture uploads
- modify the source format selection to only test for support against
the PIPE_BUFFER target
v2:
- update handling of TGSI_SEMANTIC_POSITION for recent changes in master
- MaxTextureBufferSize is number of texels, not bytes (Ilia Mirkin)
- only enable when integers are supported (Marek Olšák)
- try harder to hit the TextureBufferOffsetAlignment
- remove unnecessary MOV from the fragment shader
Reviewed-by: Edward O'Callaghan <eocallaghan@alterapraxis.com>
We need to tell the address generation functions about the dimensionality of
the texture to correctly implement the part of Section 3.8.1 (Texture Image
Specification) of the OpenGL 2.1 specification which says:
"For the purposes of decoding the texture image, TexImage2D is
equivalent to calling TexImage3D with corresponding arguments
and depth of 1, except that
...
* UNPACK SKIP IMAGES is ignored."
Fixes a low impact bug that was found by chance while browsing the spec and
extending piglit tests.
Cc: "11.0 11.1" <mesa-stable@lists.freedesktop.org>
Reviewed-by: Edward O'Callaghan <eocallaghan@alterapraxis.com>
This cap indicates whether pipe->create_surface can reinterpret a texture
as a surface with a format of different block width/height (but equal
block size).
v2: fix whitespace
Reviewed-by: Edward O'Callaghan <eocallaghan@alterapraxis.com>
This cap indicates that the driver only supports R, RG, RGB and RGBA
formats for PIPE_BUFFER sampler views.
v2: move into "unsupported features" section for nouveau (Ilia Mirkin)
Reviewed-by: Edward O'Callaghan <eocallaghan@alterapraxis.com>
When set to a truish value, this globally disables the minmax cache for all
buffer objects.
No #ifdef DEBUG guards because this option can be interesting for
benchmarking.
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
When applications stream their index buffers, the caches for those BOs become
useless and add overhead, so we want to disable them. The tricky part is
coming up with the right heuristic for *when* to disable them.
The first question is which hit rate to aim for. Since I'm not aware of any
interesting borderline applications that do something like "draw two or three
times for each upload", I just kept it simple.
The second question is how soon we should give up on the caching. Applications
might have a warm-up phase where they fill a buffer gradually but then keep
reusing it. For this reason, I count the number of indices that hit and miss
(instead of the number of calls that hit or miss), since comparing that to
the size of the buffer makes sense.
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Some games developers are unaware that an index buffer in a VBO still needs
to be read by the CPU if some varying data comes from a user pointer (unless
glDrawRangeElements and friends are used). This is particularly bad when
they tell us that the index buffer should live in VRAM.
This cache helps, e.g. lifting This War Of Mine (a particularly bad
offender) from under 10fps to slightly over 20fps on a Carrizo.
Note that there is nothing prohibiting a user from rendering from multiple
threads simultaneously with the same index buffer, hence the locking. (The
internal buffer map taken for the buffer still leads to a race, but at least
the locks are a move in the right direction.)
v2: disable the cache on USAGE_TEXTURE_BUFFER as well (Chris Forbes)
v3:
- use bool instead of GLboolean for MinMaxCacheDirty (Ian Romanick)
- replace the sticky USAGE_PERSISTENT_WRITE_MAP bit by a direct
AccessFlags check
Reviewed-by: Chris Forbes <chrisf@ijw.co.nz> (v2)
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
We will add more code for caching/memoization. Moving the existing code
into its own file helps keep things modular.
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Note that the conversion of the clear data (when data != NULL) can fail due
to an out of memory condition, but it does not check any error conditions
mandated by the spec. Therefore, it is safe to skip when size == 0.
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
We will want to disable minmax index caching for buffers that are used in this
way.
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
We will want to disable minmax index caching for buffers that are used in this
way.
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
The scaling list should be filled out with zig zag scan
v2: integrate zig zag scan for list 4x4 to vl(Christian)
v3: move list determination out from the loop(Ilia)
Cc: "11.0 11.1" <mesa-stable@lists.freedesktop.org>
Signed-off-by: Leo Liu <leo.liu@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
GEN8_SURFACE_AUX_MODE_NONE is 0, so this is a no-op.
Yet, this also makes it clear that we can compare aux_mode to the
other GEN8_SURFACE_AUX_MODE_ values. We will want to compare to
GEN8_SURFACE_AUX_MODE_HIZ.
v2: Some very minor cherry-pick conflicts due to moving it around in the series.
Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
Reviewed-by: Ben Widawsky <benjamin.widawsky@intel.com>
Signed-off-by: Ben Widawsky <benjamin.widawsky@intel.com>
Whether multisampling is turned on depends, in part, on whether
attachments are themselves multisample surfaces. However when there are
no attachments, we should rely on the default geometry for this.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
This fixes dEQP-GLES31.functional.fbo.completeness.no_attachments
When the width or height are 0, the framebuffer is incomplete. We may
also not have been passing the new state down to the driver when the
widths/heights/etc changed. Make sure to dirty the state so that the
framebuffer state is revalidated at draw time.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
This fixes it.
States which also need to be taken into account:
- SPI color formats - each down-conversion format supports only a limited set
of SPI formats
- whether MSAA resolving and logic op are enabled
These need special handling:
- blending
- disabled channels
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
The motivation is to simplify the Stoney RB+ code.
Intensity is already treated as red except here.
No piglit regressions.
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
The equivalent of the last patch for the hash table. I'm not aware of
any issues this fixes.
v2:
- use entry_is_deleted (Timothy)
Reviewed-by: Timothy Arceri <timothy.arceri@collabora.com>
Signed-off-by: Connor Abbott <cwabbott0@gmail.com>
When we delete entries in the hash set, we mark them "deleted" by
setting their key to the deleted_key, which points to a dummy
deleted_key_value. When searching for an entry, we normally skip over
those, but set_add() had some code for searching for duplicate entries
which forgot to skip over deleted entries. This led to a segfault inside
the NIR vectorization pass, since its key comparison function
interpreted the memory where deleted_key_value resides as a pointer and
tried to dereference it.
v2:
- add better commit message (Timothy)
- use entry_is_deleted (Timothy)
Reviewed-by: Timothy Arceri <timothy.arceri@collabora.com>
Signed-off-by: Connor Abbott <cwabbott0@gmail.com>
This reverts commit ab30426e33.
Apparently the memory isn't quite as aligned when this gets called
as it should be, causing crashes. (Albeit this looks independent
from this code, should crash just as well if ssse3 is enabled when
compiling without this patch.)
https://bugs.freedesktop.org/show_bug.cgi?id=93962
Add support for these opcodes, the conversion functions were already
there albeit need some new packing stuff.
Just like the tgsi version, piglit won't like it for all the same
reasons, so it's disabled (UP2H passes piglit arb_shader_language_packing
tests, albeit since PK2H won't due to those rounding differences I don't
know if that one works or not as the piglit test is rather difficult to
deal with).
Reviewed-by: Brian Paul <brianp@vmware.com>
Add support for these opcodes, the conversion functions were already
there albeit need some new packing stuff.
Just like the tgsi version, piglit won't like it for all the same
reasons, so it's disabled (UP2H passes piglit arb_shader_language_packing
tests, albeit since PK2H won't due those rounding differences I don't
know if that one works or not as the piglit test is rather difficult to
deal with).
The util functions handle the half-float conversion.
Note that piglit won't like it much due to:
a) The util functions use magic float mul conversion but when run inside
softpipe/llvmpipe, denorms are flushed to zero, therefore when the conversion
is from/to f16 denorm the result will be zero. This is a bug which should be
fixed in these functions (should not rely on denorms being available), but
will happen elsewhere just the same (e.g. conversion to f16 render targets).
b) The util functions use trunc round mode rather than round-to-nearest. This
is NOT a bug (as it is a d3d10 requirement). This will result of rounding not
representable finite values to MAX_F16 rather than INFINITY. My belief is the
piglit tests are wrong here but it's difficult to tell (generally glsl
rounding mode is undefined, however I'm not sure if rounding mode might need
to be consistent for different operations). Nevertheless, for gl it would be
better to use round-to-nearest, but using different rounding for GL and d3d10
is an unsolved problem (as it affects things like conversion to f16 render
targets, clear colors, this shader opcode).
Hence for now don't enable the cap bit (so the code is unused).
(Code is from imirkin, comment from sroland)
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Roland Scheidegger <sroland@vmvware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
If the tri is fully inside a scissor edge (or rather, we just use the
bounding box of the tri for the comparison), then we can drop these
additional scissor "planes" early. We do not even need to allocate
space for them in the tri.
The math actually appears to be slightly iffy due to bounding boxes
being rounded, but it doesn't matter in the end.
Those scissor rects are costly - the 4 planes from the scissor are
already more expensive to calculate than the 3 planes from the tri itself,
and it also prevents us from using the specialized raster code for small
tris.
This helps openarena performance by about 8% or so. Of course, it helps
there that while openarena often enables scissoring (and even moves the
scissor rect around) I have not seen a single tri actually hit the
scissor rect, ever.
v2: drop individual scissor edges, and do it earlier, not even allocating
space for them.
v3: help the compiler a bit with simpler code, suggested by Brian.
Reviewed-by: Brian Paul <brianp@vmware.com>
When we switched to 64bit rasterization, we could no longer use straight
aligned loads for loading the plane data. However, what the code actually
does for loading 3 planes, is 12 scalar loads + 9 unpacks, and then there's
another 8 unpacks for the transpose we need (!).
It would be possible to do the (scalar) loads of course already transposed
(at least saving the additional unpacks), however instead just use
(un)aligned vector loads, and recalculate the eo values, which is much less
instructions (note in case of the triangle_32_3_4 case, the eo values are
not even used, making the scalar loads + unpacks for them all the more
pointless).
This drops execution time of the triangle_32_3_4 function considerably,
albeit it doesn't really make a measurable difference (for small tris we're
essentially limited by vertex throughput in any case), for triangle_32_3_16
it's essentially noise (the loop is more costly than the initial code there).
(I'm thinking about just ditching storing the eo values in the plane data,
so could switch back to using aligned planes, however right now they are
still used in the other raster functions dealing with planes with scalar
code. Also not touching the ppc code, might not be that bad there in any
case.)
Reviewed-by: Brian Paul <brianp@vmware.com>
The existing code used ssse3, and because it isn't compiled in a separate
file compiled with that, it is usually not used (that, of course, could
be fixed...), whereas sse2 is always present at least with 64bit builds.
This should be pretty much as fast as the pshufb version, albeit those
code paths aren't really used on chips without llc in any case.
v2: fix andnot argument order, add comments
v3: use pshuflw/hw instead of shifts (suggested by Matt Turner), cut comments
Reviewed-by: Matt Turner <mattst88@gmail.com>
Enabling swrast on Android causes a link error because vtest is missing.
Signed-off-by: Rob Herring <robh@kernel.org>
Signed-off-by: Dave Airlie <airlied@redhat.com>
The virgl reference counting of buffers is broken for prime fd buffers.
Each prime fd passed into virgl_drm_winsys_resource_create_handle creates
a new resource. The solution requires creating a separate hash table to
track flink names separately from prime handles.
Signed-off-by: Rob Herring <robh@kernel.org>
Signed-off-by: Dave Airlie <airlied@redhat.com>
It is necessary to share the screen between mesa and gralloc to
properly ref count resources. This implements a hash lookup on
the file description to re-use an already created screen. This is
a similar implementation as freedreno and radeon.
Signed-off-by: Rob Herring <robh@kernel.org>
Signed-off-by: Dave Airlie <airlied@redhat.com>
The change is necessary to avoid the following building error in android:
external/mesa/src/gallium/drivers/nouveau/nouveau_vp3_video_bsp.c: In function 'nouveau_vp3_bsp_next':
external/mesa/src/gallium/drivers/nouveau/nouveau_vp3_video_bsp.c:269:14: error: 'bsp_bo' undeclared (first use in this function)
assert(bsp_bo->size >= str_bsp->w0[0] + num_bytes[i]);
^
This matches the declaration of the variables in question.
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
We use this logic to detect live ranges and then do plain renaming
across the whole codebase. As such, to prevent WaW hazards, we have to
treat a write as if it were also a read.
For example, the following sequence was observed before this patch:
13: UIF TEMP[6].xxxx :0
14: ADD TEMP[6].x, CONST[6].xxxx, -IN[3].yyyy
15: RCP TEMP[7].x, TEMP[3].xxxx
16: MUL TEMP[3].x, TEMP[6].xxxx, TEMP[7].xxxx
17: ADD TEMP[6].x, CONST[7].xxxx, -IN[3].yyyy
18: RCP TEMP[7].x, TEMP[3].xxxx
19: MUL TEMP[4].x, TEMP[6].xxxx, TEMP[7].xxxx
While after this patch it becomes:
13: UIF TEMP[7].xxxx :0
14: ADD TEMP[7].x, CONST[6].xxxx, -IN[3].yyyy
15: RCP TEMP[8].x, TEMP[3].xxxx
16: MUL TEMP[4].x, TEMP[7].xxxx, TEMP[8].xxxx
17: ADD TEMP[7].x, CONST[7].xxxx, -IN[3].yyyy
18: RCP TEMP[8].x, TEMP[3].xxxx
19: MUL TEMP[5].x, TEMP[7].xxxx, TEMP[8].xxxx
Most importantly note that in the first example, the second RCP is done
on the result of the MUL while in the second, the second RCP should have
the same value as the first. Looking at the GLSL source, it is apparent
that both of the RCP's should have had the same source.
Looking at what's going on, the GLSL looks something like
float tmin_8;
float tmin_10;
tmin_10 = tmin_8;
... lots of code ...
tmin_8 = tmpvar_17;
... more code that never looks at tmin_8 ...
And so we end up with a last_read somewhere at the beginning, and a
first_write somewhere at the bottom. For some reason DCE doesn't remove
it, but even if that were fixed, DCE doesn't handle 100% of cases, esp
including loops.
With the last_read somewhere high up, we overwrite the previously
correct (and large) last_read with a low one, and then proceed to decide
to merge all kinds of junk onto this temp. Even if that weren't the
case, and there were just some writes after the last read, then we might
still overwrite a merged value with one of those.
As a result, we should treat a write as a last_read for the purpose of
determining the live range.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Dave Airlie <airlied@redhat.com>
Cc: mesa-stable@lists.freedesktop.org
And mark nir_op_pack_uvec4_to_uint unreachable, since it's only produced
by lowering pack[SU]norm4x8 which the vec4 backend does not need.
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
A future patch will want to use designated initalizers, which aren't
available in C++, but this is C.
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Enable GL_OES_geometry_shader enums for OpenGL ES 3.1.
V4: EXTRA tokens updated according to comments from Ilia Mirkin.
V5: Account for check_extra does not evaluate "or" lazy. Fix issues
with EXTRA_EXT_FB_NO_ATTACH_CS.
Signed-off-by: Marta Lofstedt <marta.lofstedt@linux.intel.com>
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
This happens especially with exports and varying packing, where the last
bits aren't always filled in. We end up trying to do quad-wide stores,
which ends up being a lot of register moves that carefully preserve the
nop value. Instead don't do the stores.
total instructions in shared programs : 6131375 -> 6125267 (-0.10%)
total gprs used in shared programs : 910139 -> 895501 (-1.61%)
total local used in shared programs : 15328 -> 15328 (0.00%)
local gpr inst
helped 0 7442 4693
hurt 0 90 2687
Most of the helped/hurt instruction changes are by one or two ops
because can no longer do quad-wide stores in all cases.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
If an instruction has multiple defs, we have to do a lot more checks to
make sure that we can move it forward. Among other things, various code
likes to do
a, b = tex()
if () c = a
else c = b
which means that a single phi node will have results pointing at the
same instruction. We obviously can't propagate the tex in this case, but
properly accounting for this situation is tricky. Just don't try for
instructions with multiple defs.
This fixes about 20 shaders in shader-db, including the dolphin efb2ram
shader.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Cc: mesa-stable@lists.freedesktop.org
It appears that the nvidia render engine is quite picky when it comes to
linear surfaces. It doesn't like non-256-byte aligned offsets, and
apparently doesn't even do non-256-byte strides.
This makes arb_clear_buffer_object-unaligned pass on both nv50 and nvc0.
As a side-effect this also allows RGB32 clears to work via GPU data
upload instead of synchronizing the buffer to the CPU (nvc0 only).
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu> # tested on GF108, GT215
Tested-by: Nick Sarnie <commendsarnex@gmail.com> # GK208
Cc: mesa-stable@lists.freedesktop.org
Since we emulate clip-planes, the clip-vertex is used within the VS
itself (thanks to nir_lower_clip). So just ignore it as a VS output.
Fixes a boatload of piglit tests that were asserting on unknown
varying slot.
(Also unrelated spelling/typo fix.)
Signed-off-by: Rob Clark <robclark@freedesktop.org>
With glsl_to_nir we end up with local variables, instead of global, for
arrays.
Note that we'll eventually have to do something more clever, I think,
when we support multiple functions, but that will probably take some
work in a few places.
Signed-off-by: Rob Clark <robclark@freedesktop.org>
With tgsi_to_nir we get this as a normal input with VARYING_SLOT_FACE.
But glsl_to_nir plus nir_lower_system_values this becomes an intrinsic.
Signed-off-by: Rob Clark <robclark@freedesktop.org>
When using the "shared" vertex array configuration strategy, we bind
each of the buffers as a separate array. However there can be holes in
such vertex buffer lists, so just emit a disable for those.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Cc: mesa-stable@lists.freedesktop.org
Teach the emitter that the two registers are sequential, and drop the
second arg entirely, in favor of a double-wide first argument.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
This largely leaves the existing image logic alone. When image support
is added this will have to be harmonized somehow.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Issue a MEM_BARRIER. No idea if this is sufficient. As there are no
tests for this, it'll have to do for now.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
(address, length) pairs are uploaded to the driver constbuf as well to
make these values available to the shaders.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
This makes PROGRAM_IMMEDIATE a first-class gl_register_file type, and
adds PROGRAM_BUFFER to the list. These are used purely inside
glsl_to_tgsi conversion.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Currently any access params (coherent/volatile/restrict) are being lost
when lowering to the ssbo load/store intrinsics. Keep track of the
variable being used, and bake its access params in as the last arg of
the load/store intrinsics.
If the variable is accessed via an instance block, then 'variable'
points to the instance block variable and not the field inside the
instance block that we are accessing. In order to check access
parameters for the field itself we need to detect this case and keep
track of the corresponding field struct so we can extract the specific
field access information from there instead.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Marek Olšák <marek.olsak@amd.com> (v1)
v1 -> v2: add tracking of struct field
v2 -> v3: minor adjustments based on Iago's feedback
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Interfaces can have image properties set in case they are buffer
interfaces. Make sure not to lose this information.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
In particular, AMDGPU_GEM_CREATE_CPU_GTT_USWC can affect even BOs created
in VRAM if they get evicted to GTT. In general there's no need to
restrict any of the flags to any particular domains.
Reviewed-by: Edward O'Callaghan <eocallaghan@alterapraxis.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Very modest effect, but it's clearly the right thing to do.
total instructions in shared programs : 6131491 -> 6131398 (-0.00%)
total gprs used in shared programs : 910157 -> 910131 (-0.00%)
total local used in shared programs : 15328 -> 15328 (0.00%)
local gpr inst bytes
helped 0 55 85 85
hurt 0 26 20 20
Signed-off-by: Karol Herbst <nouveau@karolherbst.de>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
Following shader-db results on GK110:
total instructions in shared programs : 6141510 -> 6131491 (-0.16%)
total gprs used in shared programs : 910187 -> 910157 (-0.00%)
total local used in shared programs : 15328 -> 15328 (0.00%)
local gpr inst bytes
helped 0 18 821 821
hurt 0 0 0 0
Signed-off-by: Karol Herbst <nouveau@karolherbst.de>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
ARB_gpu_shader_fp64 spec says:
"This extension does not support interpolation of double-precision
values; doubles used as fragment shader inputs must be qualified as
"flat"."
Fixes the regressions added by commit 781d278:
arb_gpu_shader_fp64-double-gettransformfeedbackvarying
arb_gpu_shader_fp64-tf-interleaved
arb_gpu_shader_fp64-tf-interleaved-aligned
arb_gpu_shader_fp64-tf-separate
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=93878
Signed-off-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
Reviewed-by: Timothy Arceri <timothy.arceri@collabora.com>
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
Just make sure that after we've submitted, we get to at least 5
(global) submits ago before we go on to do more. Prevents up to
seconds of lag with window movement in X with xcompmgr -c. There may
be useful tuning to do in the future, but for now this gets us
usability.
Cc: "11.0 11.1" <mesa-stable@lists.freedesktop.org>
Signed-off-by: Eric Anholt <eric@anholt.net>
On an error return, the returned seqno will probably be unset, so we'd
lose track of what we've submitted so far for waiting on in the
future.
Cc: "11.0 11.1" <mesa-stable@lists.freedesktop.org>
Signed-off-by: Eric Anholt <eric@anholt.net>
When RA fails, and we spill, we have to clean everything up before doing
RA again. We were forgetting to reset the hi/lo linked lists - at
least the hi list is guaranteed to still have pointers to now-deleted
RIG nodes.
Signed-off-by: Karol Herbst <nouveau@karolherbst.de>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
Cc: mesa-stable@lists.freedesktop.org
v2: drop inline keyword
drop radeon_llvm_dispose_kernel_module wrapper
v3: move definitions to .c file
use in radeonsi
Signed-off-by: Jan Vesely <jan.vesely@rutgers.edu>
Signed-off-by: Marek Olšák <marek.olsak@amd.com>
The addition of spi_shader_col_format killed all color outputs
in precompiled shaders.
Reviewed-by: Michel Dänzer <michel.daenzer@amd.com> (v1)
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com> (v1)
v2: also set the alpha func (trivial)
For now this will be enabled in tandem with GL_OES_geometry_shader.
Should a driver come along that wants to separate them out, another
enable can be added.
Also adds the missed GL_OES_geometry_shader define in glcpp.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Marta Lofstedt <marta.lofstedt@intel.com>
At a later stage we might want to split out the NIR specific [XXX:
which one was it], as to make things move obvious and rename the files
appropriately. This patch aims to split it out of nir.
Signed-off-by: Emil Velikov <emil.velikov@collabora.com>
Acked-by: Matt Turner <mattst88@gmail.com>
Acked-by: Jose Fonseca <jfonseca@vmware.com>
Currently it's an empty library, although it'll be used to store common
code between GLSL and NIR that is compiler specific (rather than generic
as the one in src/util).
XXX: strictly speaking we could add a python/mako parser to generate the
relevant files instead including builtin_type_macros.h in such a manner.
Signed-off-by: Emil Velikov <emil.velikov@collabora.com>
Acked-by: Matt Turner <mattst88@gmail.com>
Acked-by: Jose Fonseca <jfonseca@vmware.com>
This currently just writes out the name of dump files, which can be useful
to easily correlate those files with other log outputs (driver debug output,
apitrace calls, etc.)
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
This changes the default behavior of 'always' mode to be consistent with
hang detection mode.
I have used this to more easily compare dumped command streams using diff.
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
The new name for the intrinsic was introduced in LLVM r258558.
v2: use ternary operator instead of preprocessor
Reviewed-by: Michel Dänzer <michel.daenzer@amd.com> (v1)
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
When setting the conservative thread counts, I halved everything. That isn't
correct for the wm, which has nothing to do with actual thread counts. I suck.
BXT only has 1 slice, and there is some ambiguity about subslices, so just
reserve the max possible for now. It looks like this might fix:
piglit.spec.glsl-1_50.execution.variable-indexing.gs-output-array-vec4-index-wr.bxtm64.
I kind of question why that is, but it is what Jenkins says.
Mark is current running some of the other blacklisted tests on this patch. (it
effects anything requiring scratch space).
Cc: mesa-stable <mesa-stable@lists.freedesktop.org>
Cc: Neil Roberts <neil@linux.intel.com>
Signed-off-by: Ben Widawsky <benjamin.widawsky@intel.com>
Acked-by: Kenneth Graunke <kenneth@whitecape.org>
Tested-by: Mark Janes <mark.a.janes@intel.com>
_mesa_texture_parameteriv is used because (the more obvious)
_mesa_texture_parameteri just stuffs the parameter in an array and calls
_mesa_texture_parameteriv. This just cuts out the middleman.
As a side bonus we no longer need check that ARB_stencil_texturing is
supported. The test doesn't allow non-supporting implementations to
avoid any work, and it's redundant with the value-changed test.
Fix bug #93717 because the state restore commands at the bottom of
_mesa_meta_GenerateMipmap no longer depend on the bound state.
Fixes piglit arb_direct_state_access-generatetexturemipmap with the
changes recently sent to the piglit mailing list. See the bugzilla
entry for more info.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=93717
Cc: "11.0 11.1" <mesa-stable@lists.freedesktop.org>
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
Commit c246828c added the code to save and restore the stencil
texturing mode. The restore, however, was erroneously inside the
'target != GL_TEXTURE_RECTANGLE' block.
Fixes piglit test 'arb_stencil_texturing-blit_corrupts_state
GL_TEXTURE_RECTANGLE'.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Cc: "11.0 11.1" <mesa-stable@lists.freedesktop.org>
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
Commit 055093e removed the call to _mesa_meta_in_progress, and meta.h
has not been necessary in src/mesa/main/enable.c since.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
The key for a geometry shader would be interpreted as the key for a vertex
shader further down the line, which really doesn't make sense.
This does not affect the contents of shader->key because geometry shaders
don't have any key entries anyway.
Reviewed-by: Edward O'Callaghan <eocallaghan@alterapraxis.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
It is only used during shader creation now, so no need to keep it around
afterwards.
Reviewed-by: Edward O'Callaghan <eocallaghan@alterapraxis.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
We now have an explicit parameter that contains the same information, and
this will allow us to get rid of is_gs_copy_shader in the si_shader struct.
Reviewed-by: Edward O'Callaghan <eocallaghan@alterapraxis.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Specifically, when the API switches from using a GS to not using a GS and then
back to using the same GS again, we do not have to re-send all the GS state,
but we do have to send VGT_GS_MODE. So make VGT_GS_MODE consistently be a part
of the VS state.
This fixes a rendering bug in Dolphin, but surely other applications are
affected as well.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=93648
Cc: "11.0 11.1" <mesa-stable@lists.freedesktop.org>
Reviewed-by: Edward O'Callaghan <eocallaghan@alterapraxis.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
This reverts commit 67e3098703.
It breaks a bunch of geometry shader tests, such as "spec@!opengl 3.2@minmax"
and others depending on the glGet queries.
In the old hand-writen implementation of atan2, the calculation of
atan(y/x) was performed conditionally in the "then" block of the
outermost if statement. I believe I accidentally lifted this out
into unconditional code when converting to IR builder.
For reference, the original hand-written IR is visible in commit
722eff674b.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Cc: Erik Faye-Lund <kusmabite@gmail.com>
OpenGL's dual color blending feature was specified so that an
implementation could support both multiple render targets (MRT) and
dual source blending. Fragment shader outputs specify both "location"
(the render target number) and "index" (either color 0 or 1).
I believe DirectX only has the notion of "location" - if using dual
color blending, location 0 or 1 will specify the operands. If not,
then location means the render target index. The two features can't
be used together.
As such, some applications mistakenly try to use <loc = 0, index = 0>
and <loc = 1, index = 0> in a shader used for dual color blending with
a single render target, rather than the correct <loc = 0, index = 0>
and <loc = 0, index = 1>.
In particular, Unigine Heaven 4.0 and Valley 1.0 suffer from this bug.
Unigine is aware of the problem, and quickly developed a fix, but has
not bothered to change the download link on their website to a working
copy in over a year. People were still using the broken version and
complaining. We tried working around this by disabling dual color
blending, but that apparently hurts performance, and people were once
again unhappy.
On i965, dual source blending is achieved by using different framebuffer
write messages than normal rendering. So, we have to compile different
code for the two cases. We're not being pedantic: we actually have to
know in order to function.
Normally, dual source blending is detectable in the shader: if a shader
has an output with index = 1, then it's meant for blending, not MRT.
With the broken inputs, they're indistinguishable, so we can only tell
by looking at the current GL state.
This patch implements a new drirc workaround:
export dual_color_blend_by_location=true
which makes the i965 driver detect when OpenGL state is configured for
dual source blending, and recompile the fragment shader to use the right
messages. In that case, we allow either location = 1 or index = 1 to
specify the second source for the blending equations.
It also re-enables GL_ARB_blend_func_extended for Unigine.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=92233
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Acked-by: Ilia Mirkin <imirkin@alum.mit.edu>
v2: After more discussion with hw teams, the kernel already contains the
optimal settings allowing us to use all CUs.
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
This was originally removed here:
commit 031d350132
Author: Kenneth Graunke <kenneth@whitecape.org>
Date: Tue Aug 25 16:59:12 2015 -0700
i965/vs: Unify URB entry size/read length calculations between backends.
Then added back:
commit bd198b9f0a
Author: Kenneth Graunke <kenneth@whitecape.org>
Date: Fri Aug 14 16:01:33 2015 -0700
i965/vs: Simplify fs_visitor's ATTR file.
Note that the authorship dates are out of order, but the above reflects the
order of the commit dates.
Signed-off-by: Ben Widawsky <benjamin.widawsky@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
When a fragment shader is used that has no outputs but does conditional
discard (KILL_IF), all fragments are killed without this patch.
By comparing various register settings, my conclusion is that the exec mask
is either not properly forwarded to the DB by NULL exports or ends up being
unused, at least when there is _only_ a NULL export (the ISA documentation
claims that NULL exports can be used to override a previously exported exec
mask).
Of the various approaches I have tried to work around the problem, this one
seems to be the least invasive one.
v2: take discard by alpha test into account as well
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=93761
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Updates the _mesa_has_geometry_shaders function to also look
for OpenGL ES 3.1 contexts that has OES_geometry_shader enabled.
Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
because not using SPI_SHADER_32_ABGR doubles fill rate.
We should also get optimal performance if alpha isn't needed or blending
isn't enabled.
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
This does change the behavior slightly:
If a shader writes COLOR[i] and that color buffer isn't bound,
the shader will export MRT_NULL instead and discard the IR tree that
calculates the output. The only exception is alpha-to-coverage, which
requires an alpha export.
v2: - update a comment about 16BPC
- account for MRTZ when when fixing alpha-test/kill
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Normally there's a producer and consumer, and the producer var gets
picked. In both the vertex->gs and tes->gs cases, that's the un-arrayed
version.
In the SSO case, however, there is no producer. So we picked the arrayed
GS variable, and as a result, used more slots than we should. More
critically, these slots would also no longer line up with the producer's
calculation. To fix this, we need to fix up the type of the variable
based on stage no matter what.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=93650
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Timothy Arceri <timothy.arceri@collabora.com>
Cc: "11.0 11.1" <mesa-stable@lists.freedesktop.org>
By using whole static libraries the android buildsystem provides
whole-archive (alike) solution. This means that we don't need to worry
about the order of the static libraries and any reverse, recursive or
circular dependencies that they have between one another.
Without this the linker will discard any unused hunks of one library
and we'll end up with unresolved symbols as those are required by
another static library. This issue has become more prominent with the
introduction of pipe-loader.
Whole static libraries has been used in i915/i965 for a very long
time, so we might do the same.
v2:
- Better commit message (Ilia)
- Keep external dependencies as [normal] static libs (Mauro)
Cc: mesa-stable@lists.freedesktop.org
Cc: Mauro Rossi <issor.oruam@gmail.com>
Reported-by: Mauro Rossi <issor.oruam@gmail.com>
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
The commit b4e198f47f changed the offset and bits parameters of the
bitfield insert operation from scalars to vectors. However, the lowering
of ldexp on doubles operates on each vector component and emits scalar
code (since it has to deal with the lower and upper 32-bit chunks of
each double component), so it needs its bits and offset parameters to
be scalars.
Fixes fp64 regression (crash) in:
spec/arb_gpu_shader_fp64/execution/built-in-functions/fs-ldexp-dvec4.shader_test
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
Since the GREMEDY extensions are normally only exposed by the gremedy
debugger (and could possibly trigger debug paths in the app), we don't
expose the extension by default, but instead only with
ST_DEBUG=gremedy.
Signed-off-by: Rob Clark <robclark@freedesktop.org>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
The texture mipmap completeness checking code was checking whether all
of the faces have the same size. However this is pointless because the
code just above it checks whether the face has the expected size
calculated for the mipmap level anyway so the error condition could
never be reached. This patch just removes it.
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
According to the GL 1.4 spec section 3.8.10, a cubemap texture is only
complete if:
• The level base arrays of each of the six texture images making up
the cube map have identical, positive, and square dimensions.
• The level base arrays were each specified with the same internal
format.
• The level base arrays each have the same border width.
Previously the texture completeness code was only checking the first
point. This patch makes it additionally check the other two.
This fixes the following two dEQP tests:
deqp-gles2.functional.texture.completeness.cube.format_mismatch_rgba_rgb_level_0_neg_z
deqp-gles2.functional.texture.completeness.cube.format_mismatch_rgb_rgba_level_0_pos_z
And also this Piglit test:
spec/!opengl 2.0/incomplete-cubemap-format
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=93792
Cc: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
The buffers are referenced from r600_update_driver_const_buffers()
-> r600_set_constant_buffer() -> u_upload_data(), but nothing
ever releases the reference. Similar case with driver_consts.
Found using valgrind.
Signed-off-by: Grazvydas Ignotas <notasas@gmail.com>
Cc: <mesa-stable@lists.freedesktop.org>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
main/shaderapi.c:1318:51: warning: format specifies type 'unsigned int' but the argument has type 'GLhandleARB' (aka 'unsigned long') [-Wformat]
_mesa_debug(ctx, "glDeleteObjectARB(%u)\n", obj);
~~ ^~~
%lu
Signed-off-by: Jeremy Huddleston Sequoia <jeremyhu@apple.com>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Removes the public symbol _glapi_create_table_from_handle from
libGL.so.1.2.0 on all platforms except Darwin.
Since the symbol is not used on other platforms it makes sense to
build glapi_gentable.c only on Darwin.
As a side effect it accelerates the build a bit and reduces the size
of libGL.so.1.2.0 as follows:
size lib/libGL.so.1.2.0 on my system shows
text data bss dec hex filename
469211 21848 2720 493779 788d3 lib/libGL.so.1.2.0 before
420988 11240 2720 434948 6a304 lib/libGL.so.1.2.0 after
A little bit of history:
_glapi_create_table_from_handle was introduced in
commit 85937f4c0d
Author: Jeremy Huddleston <jeremyhu@apple.com>
Date: Thu Jun 9 16:59:49 2011 -0700
glapi: Add API that can create a _glapi_table from a dlfcn handle
Example usage:
void *handle = dlopen(opengl_library_path, RTLD_LOCAL);
struct _glapi_table *disp = _glapi_create_table_from_handle(handle,
"gl");
Signed-off-by: Jeremy Huddleston <jeremyhu@apple.com>
and the only user in mesa was added in
commit f35913b96e
Author: Jeremy Huddleston <jeremyhu@apple.com>
Date: Thu Jun 9 17:29:51 2011 -0700
apple: Use _glapi_create_table_from_handle to initialize our
dispatch table
Signed-off-by: Jeremy Huddleston <jeremyhu@apple.com>
gl_gentable.py was also used for XQuartz in xserver 1.11 - 1.14.
v2: Fix typos in commit message
Add missing XORG_GLAPI_OUTPUTS += \ into src/mapi/glapi/gen/Makefile.am
Add glapi_gentable.c to EXTRA_DIST for inclusion in the release
tarball
v3: Fix commit message: s/gl_gentable.c/glapi_gentable.c/
Reported-by: Arlie Davis <arlied@google.com>
Cc: Jeremy Huddleston <jeremyhu@apple.com>
Signed-off-by: Andreas Boll <andreas.boll.dev@gmail.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
This patch significantly reduces the size of the libGL.so binary. It does
not change the (externally visible) behavior of libGL.so at all.
gl_gentable.py generates a function, _glapi_create_table_from_handle.
This function allocates a large dispatch table, consisting of 1300 or so
function pointers, and fills this dispatch table by doing symbol lookups
on a given shared library. Previously, gl_gentable.py would generate a
single, very large _glapi_create_table_from_handle function, with a short
cluster of lines for each entry point (function). The idiom it generates
was a NULL check, a call to snprintf, a call to dlsym / GetProcAddress,
and then a store into the dispatch table. Since this function processes
a large number of entry points, this code is duplicated many times over.
We can encode the same information much more compactly, by using a lookup
table. The previous total size of _glapi_create_table_from_handle on x64
was 125848 bytes. By using a lookup table, the size of
_glapi_create_table_from_handle (and the related lookup tables) is reduced
to 10840 bytes. In other words, this enormous function is reduced by 91%.
The size of the entire libGL.so binary (measured when stripped) itself drops
by 15%.
So the purpose of this change is to reduce the binary size, which frees up
disk space, memory, etc.
size lib/libGL.so.1.2.0 on my system shows (Andreas)
text data bss dec hex filename
565947 11256 2720 579923 8d953 lib/libGL.so.1.2.0 before
469211 21848 2720 493779 788d3 lib/libGL.so.1.2.0 after
v2: Incorporate Matt's feedback.
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Jeremy Huddleston Sequoia <jeremyhu@apple.com>
Tested-by: Jeremy Huddleston Sequoia <jeremyhu@apple.com>
Signed-off-by: Andreas Boll <andreas.boll.dev@gmail.com>
Take reading shader outputs into account, and use setFlagsDef for the
carry since we rely on having i->flagsDef being set.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Doing that is clearly a bug. We can't quite assert as st/mesa may hit this,
but increase at least visibility of it a bit.
(For the non-refcounted objects it would be illegal too, but we can't detect
that unless we'd store the context ourselves. Plus, those don't tend to cause
random crashes at context or object destruction time... So just sampler views,
surfaces and so targets for now.)
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
I removed this mistakenly in 2dbc20e456. I
actually thought it should not be necessary and a piglit run didn't show
any differences, but this shouldn't have been in there.
draw_prepare_shader_outputs() is in fact dependent on NEW_RASTERIZER.
The new polygon-mode-facing test indeed shows why this is necessary, there's
lots of invalid reads and writes with valgrind (also crashes without
valgrind), because the pre-pipeline vertex size doesn't match the
post-pipeline vertex size (note this won't help much with stages which don't
have the prepare hook which can grow the vertex size, in particular the wide
point stage, but this isn't used by llvmpipe). The test still won't pass, of
course, but it is only usage of uninitialized values now, which is much
less dangerous...
(Albeit I'm pretty sure for i915 it really is not needed anymore as it
doesn't care about the extra outputs and doesn't call
draw_prepare_shader_outputs().)
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
Patch moves uniform calculation to happen during link_uniforms, this
is possible with help of UniformRemapTable that has all the reserved
locations.
Location assignment for implicit locations is changed so that we
utilize also the 'holes' that explicit uniform location assignment
might have left in UniformRemapTable, this makes it possible to fit
more uniforms as previously we were lazy here and wasting space.
Fixes following CTS tests:
ES31-CTS.explicit_uniform_location.uniform-loc-mix-with-implicit-max
ES31-CTS.explicit_uniform_location.uniform-loc-mix-with-implicit-max-array
v2: code cleanups, increment NumUniformRemapTable correctly, fix
find_empty_block to work properly and add some more comments.
Signed-off-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Marta Lofstedt <marta.lofstedt@intel.com>
Fixes piglit regression after fixes to duplicate layout rules.
Previously catching multiple layouts was relying on the code
meant to catch duplicates within a single layout(...), this
change triggers the rules for multiple layouts.
Cc: Mark Janes <mark.a.janes@intel.com>
Reviewed-by: Francisco Jerez <currojerez@riseup.net>
If we have a d24x8 format, there is no stencil. Therefore, we can always
clear these bits too, which means this will be some kind of memset rather
than read-modify-write.
This is good for some 7% increase or so in gears with huge window size -
seems to have a bigger effect if things aren't in caches. Of course, any
real app won't spend nearly as much time comparatively in clearing
depth buffer in the first place, so the speedup will be much lower.
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
Fixes a number of GLES31 CTS failures and hangs on various hardware:
ES31-CTS.texture_gather.plain-gather-depth-2d
ES31-CTS.texture_gather.plain-gather-depth-2darray
ES31-CTS.texture_gather.plain-gather-depth-cube
ES31-CTS.texture_gather.offset-gather-depth-2d
ES31-CTS.texture_gather.offset-gather-depth-2darray
ES31-CTS.layout_binding.sampler2D_layout_binding_texture_ComputeShader
ES31-CTS.layout_binding.sampler2DArray_layout_binding_texture_ComputeShader
ES31-CTS.explicit_uniform_location.uniform-loc-types-samplers
ES31-CTS.compute_shader.resources-texture
Some of them were actually passing by luck on some generations even
though we weren't uploading sampler state tables explicitly for the
compute stage, most likely because they relied on the cached sampler
state left from previous rendering to be close enough.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=92589
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=93312
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=93325
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=93407
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=93725
Reported-by: Marta Lofstedt <marta.lofstedt@intel.com>
Reviewed-by: Marta Lofstedt <marta.lofstedt@intel.com>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
This reuses the NEW_SAMPLER_STATE_TABLE state bit (currently only used
on pre-Gen7 hardware) to signal that the sampler state tables have
changed in order to make sure that the GPGPU interface descriptor is
updated.
Reviewed-by: Marta Lofstedt <marta.lofstedt@intel.com>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
I have a patch that writes shaders as .shader_test files, and it uses
this function to create the headers (i.e. [vertex shader]).
[tess ctrl shader] isn't a valid shader_runner header - it's spelled
out as [tessellation control shader].
There's no real reason to abbreviate it, so spell it out.
v2: Rebase on Rob's patches to move the code.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Even if re-linking fails rendering shouldn't fail as the previous
succesfully linked program will still be available. It also shouldn't
be possible to have an unlinked program as part of the current rendering
state.
This fixes a subtest in:
ES31-CTS.sepshaderobjs.StateInteraction
This change should improve performance on CPU limited benchmarks as noted
in commit d6c6b186cf.
>From Section 7.3 (Program Objects) of the OpenGL 4.5 spec:
"If a program object that is active for any shader stage is re-linked
unsuccessfully, the link status will be set to FALSE, but any existing
executables and associated state will remain part of the current rendering
state until a subsequent call to UseProgram, UseProgramStages, or
BindProgramPipeline removes them from use. If such a program is attached to
any program pipeline object, the existing executables and associated state
will remain part of the program pipeline object until a subsequent call to
UseProgramStages removes them from use. An unsuccessfully linked program may
not be made part of the current rendering state by UseProgram or added to
program pipeline objects by UseProgramStages until it is successfully
re-linked."
"void UseProgram(uint program);
...
An INVALID_OPERATION error is generated if program has not been linked, or
was last linked unsuccessfully. The current rendering state is not modified."
V2: apply the rule to both core and compat.
Cc: Tapani Pälli <tapani.palli@intel.com>
Cc: Brian Paul <brianp@vmware.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
From the ARB_shading_language_420pack spec:
"More than one layout qualifier may appear in a single
declaration. If the same layout-qualifier-name occurs in
multiple layout qualifiers for the same declaration, the
last one overrides the former ones."
The parser was already failing correctly when the extension is
not available but testing for duplicates within a single layout
qualifier was still causing this to fail when available as both
cases share the same function for merging.
Here we add a parameter to differentiate between the two uses
and apply it to the duplicate test.
Acked-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Chris Forbes <chrisf@ijw.co.nz>
In order to only create a single node for each default declaration
we add a new boolean parameter to the in/out merge function to
only create one once we reach the rightmost layout qualifier.
From the ARB_shading_language_420pack spec:
"More than one layout qualifier may appear in a single
declaration. If the same layout-qualifier-name occurs in
multiple layout qualifiers for the same declaration, the
last one overrides the former ones."
Acked-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Chris Forbes <chrisf@ijw.co.nz>
This will allow merging of duplicate layout qualifiers as allowed
by ARB_shading_language_420pack
Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Chris Forbes <chrisf@ijw.co.nz>
This is added by ARB_enhanced_layouts although it doesn't fit
into any of the six main changes so we enable this independently.
From the ARB_enhanced_layouts spec:
"More than one layout qualifier may appear in a single
declaration. Additionally, the same layout-qualifier-name
can occur multiple times within a layout qualifier or across
multiple layout qualifiers in the same declaration"
Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Chris Forbes <chrisf@ijw.co.nz>
Apparently, nobody has combined stippling with a fragment shader
containing immediates in almost five years...
Fixes a bug in Kodi with radeonsi reported by Christian König.
Cc: "11.0 11.1" <mesa-stable@lists.freedesktop.org>
Tested-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
The whole point of inlining sources is to reduce loads. We can end up in
a situation where one value is used a lot of times, and one value is
used only once per instruction. The once-per-instruction one is the one
that should get inlined, but with the previous algorithm, it was given
no preference.
This flips things around to preferring putting less-referenced values
into src1 which increases the likelihood of them being inlined.
While we're at it, adjust the heuristic to not treat 0 as an immediate,
as well as (effectively) check for situations where LIMMs can't be
loaded. All this yields improvements on nvc0:
total instructions in shared programs : 6261157 -> 6255985 (-0.08%)
total gprs used in shared programs : 945082 -> 943417 (-0.18%)
total local used in shared programs : 30372 -> 30288 (-0.28%)
total bytes used in shared programs : 50089256 -> 50047880 (-0.08%)
local gpr inst bytes
helped 21 822 3332 3332
hurt 0 278 565 565
And more importantly avoids generating really bad code with SSBOs, where
we end up checking a lot of different values (usually immediates) against
the length.
On nv50 we get comparable results, and even improve packing (bytes went
down more than instructions):
total instructions in shared programs : 6346564 -> 6341277 (-0.08%)
total gprs used in shared programs : 728719 -> 725131 (-0.49%)
total local used in shared programs : 3552 -> 3552 (0.00%)
total bytes used in shared programs : 43995688 -> 43932928 (-0.14%)
local gpr inst bytes
helped 0 1380 3252 3774
hurt 0 287 1710 1365
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
An issue could still occur if the base level is set, but fixing that
would require a lot more logic.
This fixes the recently-failing texelFetch 3D tests because the mipmaps
were no longer being generated, which in turn caused the copying logic
to be hit, which in turn didn't work because of the broken
width/height/depth.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Once we go past half of the "GPR" register file, it seems like we need
to run frag shader with smaller threadsize. (The vertex shader already
runs at TWO_QUADS, which is the minimum.)
Signed-off-by: Rob Clark <robclark@freedesktop.org>
Some a4xx firmware doesn't implement the "PFD" (prefetch-disabled)
version of the CP_INDIRECT_BUFFER packet. So allow for PFD vs PFE per
generation. Switch a3xx and a4xx over to using prefetch-enabled version
(which is also what blob does.. it seems only on a2xx we cannot use
PFE).
Signed-off-by: Rob Clark <robclark@freedesktop.org>
And only call it from r600_invalidate_resource for buffer resources.
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
This patch fixes a bug when building a pack instruction.
For POWER (altivec), in case the destination is signed and the
src width is 32, we need to use vpkswss. The original code used vpkuwus,
which emits an unsigned result.
This fixes the following piglit tests on ppc64le:
- spec@arb_color_buffer_float@gl_rgba8-drawpixels
- shaders@glsl-fs-fogscale
I've also corrected some coding style issues in the function.
v2: Returned else statements to vmware style
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
One of the oglconform tests was crashing here, and it was
due to not cloning the actual parameters before creating the
new call. This makes a call clone function that does the right
things to make sure we clone all the needed info, and points
the callee at it. (It differs from ->clone due to this).
this may fix https://bugs.freedesktop.org/show_bug.cgi?id=93722, I had this
patch in my cts fixes tree, but hadn't had time to make sure I liked it.
Cc: "11.0 11.1" <mesa-stable@lists.freedesktop.org>
Signed-off-by: Dave Airlie <airlied@redhat.com>
Reviewed-by: Timothy Arceri <timothy.arceri@collabora.com>
Any duplicates in a single declaration will already fail the
generic duplicates test due to the explicit_stream flag being set.
Reviewed-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
This will allow the ARB_shading_language_420pack rules in
glsl_parser.yy for catching duplicate layout qualifiers to be
triggered for the stream identifier rather than relying on the
code meant to catch duplicates within a single layout(...)
Reviewed-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
From Section 7.9 (SUBROUTINE UNIFORM VARIABLES) of the OpenGL
4.5 Core spec:
"The command
void UniformSubroutinesuiv(enum shadertype, sizei count,
const uint *indices);
will load all active subroutine uniforms for shader stage
shadertype with subroutine indices from indices, storing
indices[i] into the uniform at location i. The indices for
any locations between zero and the value of
ACTIVE_SUBROUTINE_UNIFORM_LOCATIONS minus one which are not
used will be ignored."
V2: simplify NULL check suggested by Jason.
Acked-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Dave Airlie <airlied@redhat.com>
Cc: "11.0 11.1" mesa-stable@lists.freedesktop.orghttps://bugs.freedesktop.org/show_bug.cgi?id=93731
Apparently the IPA op decided to stop working with offsets. Need to
figure out if we need to do an AL2P situation or something similar. For
now just turn it back off.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
This patch fixes a classic "confuse the enemy" bug.
_mm_andnot_si128 (SSE) and vec_andc (VMX) do the same operation, but the
arguments are opposite.
_mm_andnot_si128 performs "r = (~a) & b" while
vec_andc performs "r = a & (~b)"
To make sure this error won't return in another place, I added a wrapper
function, vec_andnot_si128, in u_pwr8.h, which makes the swap inside.
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
In fad158a0 ("freedreno/ir3: array rework") the src # (n) shifted by
one, but missed updating delay-slot calc.
Signed-off-by: Rob Clark <robclark@freedesktop.org>
It at least happens with some piglit tests, like
$piglit/bin/vp-address-01
VERT
DCL IN[0]
DCL IN[1]
DCL OUT[0], POSITION
DCL OUT[1], COLOR
DCL CONST[0..7]
DCL ADDR[0]
0: ARL ADDR[0].x, IN[1].xxxx
1: MOV_SAT OUT[1], CONST[ADDR[0].x-1]
2: DP4 OUT[0].x, CONST[4], IN[0]
3: DP4 OUT[0].y, CONST[5], IN[0]
4: DP4 OUT[0].z, CONST[6], IN[0]
5: DP4 OUT[0].w, CONST[7], IN[0]
6: END
Signed-off-by: Rob Clark <robclark@freedesktop.org>
Seems like in certain cases, we cannot use c<a0.x+0> as the third src to
cat3 instructions. This may be slightly conservative, we may only have
this restriction when the first src is also const.
This fixes, for example, +24/-0 of the variable-indexing piglit tests.
Signed-off-by: Rob Clark <robclark@freedesktop.org>
If we handle separately the special case of eliminating output mov
(which includes keeps and various other cases where we don't have a
consuming instruction's src register to collapse things into), we
can simplify the logic.
Signed-off-by: Rob Clark <robclark@freedesktop.org>
Shuffle things slightly, passing instr-data to ra_name() to reduce the
number of places where we need to add support for array names.
Signed-off-by: Rob Clark <robclark@freedesktop.org>
nir.h is a bit inconsistent about 'typedef struct {} nir_foo' vs
'typedef struct nir_foo {} nir_foo'. But missing struct name tags is
inconvenient when you need a fwd declaration without pulling in all
of nir.
So add missing struct name tag for nir_variable, and a couple other
spots where it would likely be useful.
Signed-off-by: Rob Clark <robclark@freedesktop.org>
Reviewed-by: Edward O'Callaghan <eocallaghan@alterapraxis.com>
This adds code that is basically the same as the code in umod, udiv and idiv.
However, unlike idiv we return -1.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
The ARB has decided that implicit conversions should be performed for
bitwise operators in future language revisions. Implementations of
current language revisions may or may not perform them.
This patch makes Mesa apply implicti conversions even on current
language versions. Applications appear to expect this behavior,
and there's really no downside to doing so.
Fixes shader compilation in Shadow of Mordor.
Bugzilla: https://www.khronos.org/bugzilla/show_bug.cgi?id=1405
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Timothy Arceri <timothy.arceri@collabora.com>
Cc: mesa-stable@lists.freedesktop.org
In the vertex and fragment stages, the hardware is nice to us and leaves
g0.2 zerod out for us so we can use it for headers. However, in compute,
geometry, and tessellation stages, the hardware is not so nice. In
particular, for compute shaders on BDW, the hardware places some debug bits
in 23:15. As it happens, bit 15 is interpreted by the sampler as the alpha
channel mask. This means that if you use a texturing instruction with a
header in a compute shader, you may randomly get the alpha channel
disabled. Since channel masks affect the return length of the sampler
message, this can lead the GPU to expect a different mlen to the one you
specified in the shader and this, in turn, hangs your GPU.
Cc: "11.1" <mesa-stable@lists.freedesktop.org>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
The cleaning up was quite a performance hog (making pipe_resource_reference
the number two in profilers on the vertex path, and 3rd overall, with its
cousin pipe_reference_described not far behind) if there were lots
of tiny draw calls (ipers). Now the reason was really that it was blindly
calling this for all potential shader views (so 32 each for vs and gs) even
though the app never touched a single one which could have been fixed,
however I can't come up with a good reason why we refcount these. We've got
references, of course, in the sampler views, which should be quite sufficient
as we do all vertex and geometry shader execution fully synchronous.
(Calling prepare_shader_sampling for all draw calls even if there were no
changes looks quite suboptimal too, but generally we don't really expect vs/gs
shader sampling to be used much with llvmpipe, and there's even an early exit
if there aren't any views to avoid the "null loop" albeit it's now no longer
always trying to loop through all 32 slots. Maybe improve another time...).
Of course, if we manage to make vertex loads run asynchronously some day,
we need references again, but adding that back would be the least of the
problems...
Also only set LP_NEW_SAMPLER_VIEW for fragment sampler views. Nothing on the
vertex side depends on it (I suppose we'd really wanted a separate flag in
any case).
(Good for a 3% improvement or so in ipers under the right conditions.)
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
This was not really a leak per se, but we were referencing the textures for
longer than intended. If textures were set via llvmpipe_set_sampler_views()
(for fs) and then picked up by lp_setup_set_fragment_sampler_views(), they
were referenced in the setup state. However, the only way to unreference them
was by replacing them with another texture, and not when the texture slot
was replaced with a NULL sampler view. (They were then further also referenced
by the scene too which might have additional minor side effects as we limit
the memory size which is allowed to be referenced by a scene in a rather crude
way.) Only setup destruction (at context destruction time) then finally would
get rid of the references.
Fix this by noting the number of textures the last time, and unreference
things if the new view is NULL (avoiding having to unreference things
always up to PIPE_MAX_SHADER_SAMPLER_VIEWS which would also have worked).
Found by code inspection, no test...
v2: rename var
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
Only modify interpolation type for integer-based varyings or when the
consumer is known and different than fragment shader.
If we are linking separate shader programs and the consumer is unknown,
the consumer could be added later and be a fragment shader. If we
modify the interpolation type in this case, we could read wrong
values in the fragment shader inputs, as shown in bug 93320.
Fixes the following CTS test:
ES31-CTS.vertex_attrib_binding.advanced-bindingUpdate
Fixes the following dEQP tests:
dEQP-GLES31.functional.separate_shader.random.102
dEQP-GLES31.functional.separate_shader.random.111
dEQP-GLES31.functional.separate_shader.random.115
dEQP-GLES31.functional.separate_shader.random.17
dEQP-GLES31.functional.separate_shader.random.22
dEQP-GLES31.functional.separate_shader.random.23
dEQP-GLES31.functional.separate_shader.random.3
dEQP-GLES31.functional.separate_shader.random.32
dEQP-GLES31.functional.separate_shader.random.39
dEQP-GLES31.functional.separate_shader.random.64
dEQP-GLES31.functional.separate_shader.random.73
dEQP-GLES31.functional.separate_shader.random.91
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=93320
Signed-off-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
This shouldn't hurt anything, and I'm about to introduce a pass that
will want it.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
This makes it a pass, hiding the parameter structs and block callbacks
so it's simpler to work with.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Shorter than compiler->scalar_stage[MESA_SHADER_VERTEX], which can
help with line-wrapping.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
nir_build_ivec4 is more readable and succinct than using nir_build_imm
directly, even if you have C99.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
If shader declares uniform explicit location in one stage but
implicit in another, explicit location should be used. Patch marks
implicit uniforms as explicit if they were explicit in previous stage.
This makes sure that we don't treat them implicit later when assigning
locations.
Fixes following CTS test:
ES31-CTS.explicit_uniform_location.uniform-loc-implicit-in-some-stages3
v2: move check to cross_validate_globals (Timothy)
Signed-off-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Timothy Arceri <timothy.arceri@collabora.com>
The RS and hardware binding tables are only supported on the 3D
pipeline and can lead to corruption if left enabled during a GPGPU
workload. Disable it when switching to the GPGPU (or media) pipeline
and re-enable it when switching back to the 3D pipeline.
Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Abdiel Janulgue <abdiel.janulgue@linux.intel.com>
This hardware bug can supposedly lead to a hang on IVB and VLV.
Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
AFAIK brw_emit_select_pipeline() is only called once during context
init on Gen4-5, at which point the pipeline is likely to be already
idle so it may just happen to work by luck regardless of the MI_FLUSH.
Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Switching the current pipeline while it's not completely idle or the
read and write caches aren't flushed can lead to corruption. Fixes
misrendering of at least the following Khronos CTS test:
ES31-CTS.shader_image_load_store.basic-allTargets-store-fs
The stall and flushes are no longer required on Gen8+.
v2: Emit PIPE_CONTROL with non-zero post-sync op before the write
cache flush on SNB due to hardware bug. (Ken)
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=93323
Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
This hardware bug can cause a hang on context restore while the
current pipeline is set to GPGPU (BDWGFX HSD 1909593). In addition to
clearing the valid bit, mark the CC state as dirty to make sure that
the CC indirect state pointer is re-emitted when we switch back to the
3D pipeline.
Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
This will be used on Gen8+ to make sure that the color calculator
state pointers are re-emitted when switching back to the 3D pipeline
after some GPGPU workload due to a hardware workaround. There are
other state bits already defined that could be used to achieve the
same effect but they all cause a ton of unrelated state to be
re-emitted (e.g. BRW_NEW_STATE_BASE_ADDRESS), so just define a new
one, state bits are cheap.
Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reduces local memory usage in a lot of Metro 2033 Redux and a few KSP
shaders:
total local used in shared programs : 54116 -> 30372 (-43.88%)
Probably modest advantage to execution, but it's an imporant
prerequisite to dropping some of the TGSI optimizations done by the
state tracker.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Previously we were treating any indirect temp array usage to mean that
everything should end up in lmem. The MemoryOpt pass would clean a lot
of that up later, but in the meanwhile we would lose a lot of
opportunity for optimization.
This helps a lot of Metro 2033 Redux and a handful of KSP shaders:
total instructions in shared programs : 6288373 -> 6261517 (-0.43%)
total gprs used in shared programs : 944051 -> 945131 (0.11%)
total local used in shared programs : 54116 -> 54116 (0.00%)
A typical case is for register usage to double and for instructions to
halve. A future commit can also optimize local memory usage size to be
reduced with better packing.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Indirect constbuf indexing works by using very large offsets. However if
an indirect constbuf index load is const-propagated, it becomes a very
large const offset. Take that into account when legalizing the SSA by
moving the high parts of that offset into the file index. Also disallow
very large (or small) indices on most other instructions.
This fixes regressions in ubo_array_indexing/*-two-arrays piglit tests.
Fixes: abd326e81b (nv50/ir: propagate indirect loads into instructions)
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
For those formats that support hw mipmap generation, use the
DXGenMips command. Otherwise fallback to the mipmap generation utility.
Tested with piglit, OpenGL apps (Heaven, Turbine, Cinebench)
v2: make sure the texture surface was created with the render target bind flag
set relocation flag to SVGA_RELOC_WRITE for the texture surface
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
The actual increment of the num-generate-mipmap counter will be done
in a subsequent patch when hw generate mipmap is supported.
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
This patch adds a new interface to support hardware mipmap generation.
PIPE_CAP_GENERATE_MIPMAP is added to allow a driver to specify
if this new interface is supported; if not supported, the state tracker will
fallback to mipmap generation by rendering/texturing.
v2: add PIPE_CAP_GENERATE_MIPMAP to the disabled section for all drivers
v3: add format to the generate_mipmap interface to allow mipmap generation
using a format other than the resource format
v4: fix return type of trace_context_generate_mipmap()
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
The OpenGL specifications for bitfieldExtract() says:
The result will be undefined if <offset> or <bits> is negative, or if
the sum of <offset> and <bits> is greater than the number of bits
used to store the operand.
Therefore passing bits=32, offset=0 is legal and defined in GLSL.
But the earlier SM5 ubfe/ibfe opcodes are specified to accept a bitfield width
ranging from 0-31. As such, Intel and AMD instructions read only the low 5 bits
of the width operand, making them not able to implement the GLSL-specified
behavior directly.
This commit adds ubfe/ibfe operations from SM5 and a lowering pass for
bitfield_extract to to handle the trivial case of <bits> = 32 as
bitfieldExtract:
bits > 31 ? value : bfe(value, offset, bits)
Fixes:
ES31-CTS.shader_bitfield_operation.bitfieldExtract.uvec3_0
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=92595
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Tested-by: Marta Lofstedt <marta.lofstedt@intel.com>
The OpenGL specifications for bitfieldInsert() says:
The result will be undefined if <offset> or <bits> is negative, or if
the sum of <offset> and <bits> is greater than the number of bits
used to store the operand.
Therefore passing bits=32, offset=0 is legal and defined in GLSL.
But the earlier SM5 bfi opcode is specified to accept a bitfield width
ranging from 0-31. As such, Intel and AMD instructions read only the low
5 bits of the width operand, making them not able to implement the
GLSL-specified behavior directly.
This commit fixes the lowering of bitfield_insert to handle the trivial
case of <bits> = 32 as
bitfieldInsert:
bits > 31 ? insert : bfi(bfm(bits, offset), insert, base)
Fixes:
ES31-CTS.shader_bitfield_operation.bitfieldInsert.uint_2
ES31-CTS.shader_bitfield_operation.bitfieldInsert.uvec4_3
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=92595
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Tested-by: Marta Lofstedt <marta.lofstedt@intel.com>
We check that a bunch of raster operations are disabled in
blit_copy_pixels(). We also need to check that color logicop is
disabled.
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
The whole point of AMD_pinned_memory is that applications don't have to map
buffers via OpenGL - but they're still allowed to, so make sure we don't break
the link between buffer object and user memory unless explicitly instructed
to.
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
This accomodates a streaming pattern where the discard flag is set when the
application wraps back to the beginning of the buffer.
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
It makes sense to re-use pipe->invalidate_resource for the purpose of
glInvalidateBufferData, but this function is already implemented in vc4
where it doesn't have the expected behavior. So add a capability flag
to indicate that the driver supports the expected behavior.
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Change the check to be in line with what the quoted spec fragment says.
I have sent out a piglit test for this as well.
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
The internal Mesa format used for a texture might not match the one
requested in the internalFormat when the texture was created, for
example if the driver is internally remapping RGB textures to RGBA.
Otherwise it can cause false positives for completeness if one mipmap
image is created as RGBA and the other as RGB because they would both
have an RGBA Mesa format. If we check the InternalFormat instead then
we are directly checking the API usage which I think better matches
the intention of the check.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=93700
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
I spotted this while looking for what needs updating in future platforms.
I'm too lazy to go through the git logs, but it was probably missed by Jason
when all the brw refactoring happened.
Signed-off-by: Ben Widawsky <benjamin.widawsky@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Intel/AMD's hardware instructions do not handle arguments of 32.
Constant evaluation should not produce a result different from the
hardware instruction.
The s/1ull/1u/ change is intentional: previously we wanted defined
behavior for the "1 << 32" case, but we're making this case undefined so
we can make it 1u and save ourselves a 64-bit operation.
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
If a Python codegen script failed, it would write a zero-byte file,
which on subsequent invocations of make would trick it into thinking the
file was appropriately generated.
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
We would like to be able to combine
result.x = bitfieldExtract(src0.x, src1.x, src2.x);
result.y = bitfieldExtract(src0.y, src1.y, src2.y);
result.z = bitfieldExtract(src0.z, src1.z, src2.z);
result.w = bitfieldExtract(src0.w, src1.w, src2.w);
into a single ivec4 bitfieldInsert operation. This should be possible
with most drivers.
This patch changes the offset and bits parameters from scalar ints
to ivecN or uvecN. The type of all three operands will be the same,
for simplicity.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
We would like to be able to combine
result.x = bitfieldInsert(src0.x, src1.x, src2.x, src3.x);
result.y = bitfieldInsert(src0.y, src1.y, src2.y, src3.y);
result.z = bitfieldInsert(src0.z, src1.z, src2.z, src3.z);
result.w = bitfieldInsert(src0.w, src1.w, src2.w, src3.w);
into a single ivec4 bitfieldInsert operation. This should be possible
with most drivers.
This patch changes the offset and bits parameters from scalar ints
to ivecN or uvecN. The type of all four operands will be the same,
for simplicity.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
TGSI doesn't use these - it just translates ir_quadop_bitfield_insert
directly. NIR can handle ir_quadop_bitfield_insert as well.
These opcodes were only used for i965, and with Jason's recent patches,
we can do this lowering in NIR (which also gains us SPIR-V handling).
So there's not much point to retaining this GLSL IR lowering code.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
NIR's bfm, like Intel/AMD's hardware instructions and GLSL IR's
ir_binop_bfm takes <bits> as src0 and <offset> as src1.
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
I added this code right at the end, and got it wrong.
Only used by the WGL_ARB_render_texture code.
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
Reviewed-by: Charmaine Lee <charmainel@vmware.com>
The RGBX surface formats aren't renderable so we internally remap them
to RGBA when rendering. They are retained as RGBX when used as
textures. However since the previous patch fast clears are disabled
for surfaces that use a different format for rendering than for
texturing. To avoid this situation we can just pretend not to support
RGBX formats at all. This will cause the upper layers of mesa to pick
an RGBA format internally instead. This should be safe because we
always override the alpha component to 1.0 for RGBX in the texture
swizzle anyway. We could also do this for all gens except that it's a
bit more difficult when the hardware doesn't support texture
swizzling. Gens using the blorp have further problems because that
doesn't implement this swizzle override.
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
Commit 8926dc8 added a check where we add packed varyings of output
stage only when we have multiple stages, however duplicates are already
handled by changes in commit 0508d950 and we want to add outputs also in
case where we have only one stage.
Fixes regression caused by 8926dc8 for following test:
ES31-CTS.program_interface_query.separate-programs-vertex
Signed-off-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Marta Lofstedt <marta.lofstedt@intel.com>
Reviewed-by: Timothy Arceri <timothy.arceri@collabora.com>
The trick here is to recognize that in the c + n * dcdx calculations,
not only can the lower FIXED_ORDER bits not change (as the dcdx values
have those all zero) but that this means the sign bit of the calculations
cannot be different as well, that is
sign(c + n*dcdx) == sign((c >> FIXED_ORDER) + n*(dcdx >> FIXED_ORDER)).
That shaves off more than enough bits to never require 64bit masks.
A shifted plane c value could still easily exceed 32 bits, however since we
throw out planes which are trivial accept even before binning (and similarly
don't even get to see tris for which there was a trivial reject plane)) this
is never a problem.
The idea isnt't all that revolutionary, in fact something similar was tried
ages ago (9773722c2b) back when the values were
only 32 bit anyway. I believe now it didn't quite work then because the
adjustment needed for testing trivial reject / partial masks wasn't handled
correctly.
This still keeps the separate 32/64 bit paths for now, as the 32 bit one still
looks minimally simpler (and also because if we'd pass in dcdx/dcdy/eo unscaled
from setup which would be a good reason to ditch the 32 bit path, we'd need to
change the special-purpose rasterization functions for small tris).
This passes piglit triangle-rasterization (-fbo -auto -max_size
-subpixelbits 8) and triangle-rasterization-overdraw (with some hacks
to make it work correctly with large sizes) easily (full piglit as
well of course, but most tests wouldn't use triangles large enough to
be affected, that is tris with a bounding box over 128x128).
The profiler says indeed time spent in rast_tri functions is reduced
substantially, BUT of course only if the tris are large. I measured a 3%
improvement in mesa gloss demo when supersized to twice the screen size...
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
Otherwise some planes we get in rasterization have subpixel precision, others
not. Doesn't matter so far, but will soon. (OpenGL actually supports viewports
with subpixel accuracy, so could even do bounding box calcs with that).
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
This is quite a few less instructions, albeit still do the 2 64bit muls
with scalar c code (they'd need way more shuffles, plus fixup for the signed
mul so it totally doesn't seem worth it - x86 can do 32x32->64bit signed
scalar muls natively just fine after all (even on 32bit).
(This still doesn't have a very measurable performance impact in reality,
although profiler seems to say time spent in setup indeed has gone down by
10% or so overall. Maybe good for a 3% or so improvement in openarena.)
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
Discovered by accident, valgrind was complaining (could have possibly caused
us to create redundant geometry shader variants).
v2: convinced by Brian and Jose, just use memset for both gs and vs keys,
just as easy and less error prone.
.length() on an unsized SSBO variable doesn't actually read any data
from the SSBO, and is allowed on variables marked 'writeonly'.
Fixes compute shader compilation in Shadow of Mordor.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
This adds barrier dependencies around TCS_OPCODE_URB_WRITE, preventing
reads and writes from being incorrectly scheduled.
Fixes rendering in GFXBench 4.0's tessellation demo.
For some reason, we haven't ever listed URB writes as having
side-effects. This hasn't been a problem because in most stages, we
never read from the URB, and only write to each location once.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=93526
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Chris Forbes <chrisf@ijw.co.nz>
If the constructor fails before the LIST_INIT calls the pointers
will be null and the deconstructor will segfault.
Signed-off-by: Tom St Denis <tom.stdenis@amd.com>
Reviewed-by: Leo Liu <leo.liu@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Patch changes linker to allocate gl_shader_variable instead of using
ir_variable. This makes it possible to get rid of ir_variables and ir
in memory after linking.
v2: check that we do not create duplicate entries with
packed varyings
v3: document 'patch' bit (Ilia Mirkin)
Signed-off-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Linker missed a check for situation where we exceed max amount of
uniform locations with explicit + implicit locations. Patch adds this
check to already existing iteration over uniforms in linker.
Fixes following CTS test:
ES31-CTS.explicit_uniform_location.uniform-loc-negative-link-max-num-of-locations
v2: use var->type->uniform_locations() (Timothy)
Signed-off-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Timothy Arceri <timothy.arceri@collabora.com>
We already check if the driver changed the completeness, we don't
need to duplicate that check. Let's just early out there instead.
Signed-off-by: Erik Faye-Lund <kusmabite@gmail.com>
Reviewed-by: Timothy Arceri <timothy.arceri@collabora.com>
This hasn't been in use since c476305 ("gallium/util: pregenerate
half float tables"), where the last bit of run-time init using this
was killed. So let's just get rid of the pointless header.
Signed-off-by: Erik Faye-Lund <kusmabite@gmail.com>
Reviewed-by: Timothy Arceri <timothy.arceri@collabora.com>
Re-binding compute constant buffers after launching a grid have no effects
because they are not currently validated and because dirty_cp is not updated
accordingly. This might also prevent weird future behaviours when UBOs will
be bound for compute.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
The path that depends on this will be avoided (by fallback_required) if
the extension is not supported. _mesa_set_sampler_srgb_decode does not
generate GL errors (by design), so there are no problems there.
I kept this change separate and last because it is one of the few in the
series that is not a candidate for the stable branch.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
All of the calls after the first _mesa_bind_sampler call are DSA style
calls that don't depend on the current binding.
I kept this change separate and last because it is one of the few in the
series that is not a candidate for the stable branch.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
tl;dr: For many types of GL object, we can *NEVER* use the Gen function.
In OpenGL ES (all versions!) and OpenGL compatibility profile,
applications don't have to call Gen functions. The GL spec is very
clear about how you can mix-and-match generated names and non-generated
names: you can use any name you want for a particular object type until
you call the Gen function for that object type.
Here's the problem scenario:
- Application calls a meta function that generates a name. The first
Gen will probably return 1.
- Application decides to use the same name for an object of the same
type without calling Gen. Many demo programs use names 1, 2, 3,
etc. without calling Gen.
- Application calls the meta function again, and the meta function
replaces the data. The application's data is lost, and the app
fails. Have fun debugging that.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=92363
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
Some meta operations can be called recursively. Future changes (the
"Don't pollute the ... namespace" changes) will cause objects with
invalid names to be used. If a nested meta operation tries to restore
an object named 0xDEADBEEF, it will fail.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
tl;dr: For many types of GL object, we can *NEVER* use the Gen function.
In OpenGL ES (all versions!) and OpenGL compatibility profile,
applications don't have to call Gen functions. The GL spec is very
clear about how you can mix-and-match generated names and non-generated
names: you can use any name you want for a particular object type until
you call the Gen function for that object type.
Here's the problem scenario:
- Application calls a meta function that generates a name. The first
Gen will probably return 1.
- Application decides to use the same name for an object of the same
type without calling Gen. Many demo programs use names 1, 2, 3,
etc. without calling Gen.
- Application calls the meta function again, and the meta function
replaces the data. The application's data is lost, and the app
fails. Have fun debugging that.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=92363
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
Some meta operations can be called recursively. Future changes (the
"Don't pollute the ... namespace" changes) will cause objects with
invalid names to be used. If a nested meta operation tries to restore
an object named 0xDEADBEEF, it will fail.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
tl;dr: For many types of GL object, we can *NEVER* use the Gen function.
In OpenGL ES (all versions!) and OpenGL compatibility profile,
applications don't have to call Gen functions. The GL spec is very
clear about how you can mix-and-match generated names and non-generated
names: you can use any name you want for a particular object type until
you call the Gen function for that object type.
Here's the problem scenario:
- Application calls a meta function that generates a name. The first
Gen will probably return 1.
- Application decides to use the same name for an object of the same
type without calling Gen. Many demo programs use names 1, 2, 3,
etc. without calling Gen.
- Application calls the meta function again, and the meta function
replaces the data. The application's data is lost, and the app
fails. Have fun debugging that.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=92363
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
Some meta operations can be called recursively. Future changes (the
"Don't pollute the ... namespace" changes) will cause objects with
invalid names to be used. If a nested meta operation tries to restore
an object named 0xDEADBEEF, it will fail.
v2: Add a comment explaining why samp_obj_save is set to NULL in
_mesa_meta_fb_tex_blit_begin. This came out of review feedback from
Jason.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
This requires tracking the sampler object using the gl_sampler_object*
instead of the object name.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
Pulls the parts of _mesa_BindSampler that aren't just parameter
validation out into a function that can be called from other parts of
Mesa (e.g., meta).
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
I was going to send this as review for dce1e1a8, but I missed that
window. This saves 64 bytes of unshared data and prelaces it with 96
bytes shared text. My guess is that some of the calls to memcpy get
optimized to something else.
text data bss dec hex filename
7847613 220208 27432 8095253 7b8615 i965_dri.so before
7847709 220144 27432 8095285 7b8635 i965_dri.so after
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
Cc: Brian Paul <brianp@vmware.com>
The ISO C99 standard (7.18.4) specifies that C++
implementations should define UINT64_C only when
__STDC_CONSTANT_MACROS is defined.
Because we now use UINT64_C in our cpp files (since commit
208bfc493d), we need to add this define.
This also solves compilation errors with GCC 4.8.x on ppc64le machines.
v2: add this define to SCons build system
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
Gen9+ requires us to emit 3DSTATE_BINDING_TABLE_POINTERS_HS for the
hull shader push constants to take effect. The passthrough TCS uses
push constants for the default tessellation levels. So, when those
change, we need to re-upload the binding table as well.
Fixes five Piglit tests on Skylake:
- spec/arb_tessellation_shader/vs-tes-vertex
- spec/arb_tessellation_shader/vs-tes-tessinner-tessouter-inputs-quads
- spec/arb_tessellation_shader/vs-tes-tessinner-tessouter-inputs-tris
- spec/arb_tessellation_shader/tes-read-texture
- spec/arb_tessellation_shader/tess_with_geometry
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
In testing KBL, I found:
- urb size was not set for slices gt1.5, gt2, and gt3. The value I
used for these slices (384) was taken from an earlier patch authored
by Ben Widawsky.
- slice count was missing. This field was added by
a403ad4f5a
With this commit, KBL passes piglit at parity with SKL.
Note: As requested by Kristian, Sarah modified this patch to drop
setting urb size for gt1.5, gt2, and gt3, since the correct default is
set in the GEN9 macro by commit c1e38ad370
"i965/skl: Use larger URB size where available."
Signed-off-by: Mark Janes <mark.a.janes@intel.com>
Signed-off-by: Sarah Sharp <sarah.a.sharp@linux.intel.com>
Reviewed-by: Kristian Høgsberg Kristensen <kristian.h.kristensen@intel.com>
Cc: "11.1" <mesa-stable@lists.freedesktop.org>
For the case where we convert a double to an int, we should
round the same as we do for floats.
This fixes GL41-CTS.gpu_shader_fp64.state_query
v2: add IROUNDD (Ilia)
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
Signed-off-by: Dave Airlie <airlied@redhat.com>
The lower_named_interface_blocks() pass is called before we try
assign locations to varyings so this shouldn't be reachable.
Reviewed-by: Edward O'Callaghan <eocallaghan@alterapraxis.com>
The lower_named_interface_blocks() pass is called before we try
assign locations to varyings so this shouldn't be reachable.
Reviewed-by: Edward O'Callaghan <eocallaghan@alterapraxis.com>
OpenGL 2.0 function StencilOp() is in part internally implemented via
StencilOpSeparate(). This change happened some time ago, however the
accompanying doxygen todo comment was not accordingly updated.
Replace the outdated portion of this doxygen todo comment, leaving the
remainder unchanged.
Also better respect the 80 character suggested line length in this file.
v2: Fully remove comment, following code review by t_arceri@yahoo.com.au
Signed-off-by: Rhys Kidd <rhyskidd@gmail.com>
Reviewed-by: Thomas Helland <thomashelland90@gmail.com>
Reviewed-by: Timothy Arceri <timothy.arceri@collabora.com>
Currently, opt_vectorize() tries to combine:
result.x = bitfieldInsert(src0.x, src1.x, src2.x, src3.x);
result.y = bitfieldInsert(src0.y, src1.y, src2.y, src3.y);
result.z = bitfieldInsert(src0.z, src1.z, src2.z, src3.z);
result.w = bitfieldInsert(src0.w, src1.w, src2.w, src3.w);
into a single ir_quadop_bitfield_insert opcode, which operates on
ivec4s. However, GLSL IR's opcodes currently require the bits and
offset parameters to be scalar integers. So, this breaks.
We want to be able to vectorize this eventually, but for now, just
chicken out and make opt_vectorize() bail by marking all the bitfield
insert/extract related opcodes as horizontal. This is a relatively
uncommon case today, so we'll do the simple fix for stable branches,
and fix it properly on master.
Fixes assertion failures when compiling Shadow of Mordor vertex shaders
on i965 in vec4 mode (where OptimizeForAOS enables opt_vectorize()).
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Cc: mesa-stable@lists.freedesktop.org
SCons doesn't understand nir yet and doesn't want to compile the glsl to
nir pass. Move the files to their own variable so we can add it only for
automake.
Tested-by: Brian Paul <brianp@vmware.com>
These are used by code that doesn't necessarily link to libglsl.la. Move
them to shader_enums.[ch] where we keep similar helpers.
Reviewed-by: Matt Turner <mattst88@gmail.com>
The scope of libi965_compiler.la is to be able to take nir shaders and
generate i965 EU code. As such, we don't want the GLSL IR lowering
passes in the library. With this change, libi965_compiler.la no longer
needs to link to libglsl.la.
Reviewed-by: Matt Turner <mattst88@gmail.com>
libglsl_la_SOURCES includes both NIR_FILES and LIBGLSL_FILES, so for
libglsl.la consumers, this is a no-op. libnir.la however no longer uses
any GLSL IR infrastructure and can be used without also linking to
libglsl.la.
Acked-by: Matt Turner <mattst88@gmail.com>
The freedesktop.org blog feeds aren't mentioned on either mesa3d.org or
any of the graphics project wikis (including the DRI wiki) on
freedeskop.org. Fix that by linking to it from the sidebar.
Signed-off-by: Sarah Sharp <sarah.a.sharp@linux.intel.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Specify that the operation only applies to the x component, not
per-component as previously specified. This is unnecessary for GL and
creates additional complications for images which need to support these
operations as well.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Each load/store on most hardware can specify what caching to do. Since
SSBO allows individual variables to also have separate caching modes,
allow loads/stores to have the qualifiers instead of attempting to
encode them in declarations.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
I believe that `1u << x`, where x >= 32 yields undefined results
according to the C standard.
Particularly MSVC says `warning C4334: '<<' : result of 32-bit shift
implicitly converted to 64 bits (was 64-bit shift intended?)`.
Reviewed-by: Brian Paul <brianp@vmware.com>
For profiling mesa's code, especially llvmpipe, PROFILE should be
defined. Currently, this define can only be generated if mesa is
built using scons.
This patch makes it possible to generate this define also when building
mesa through automake tools.
v2:
- Change --enable-llvmpipe-profile to --enable-profile
- Add -fno-omit-frame-pointer to CFLAGS and CXXFLAGS when enabling profile
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
It can be trivially derived from the number of already declared system
values. This allows ureg users not to worry about which index to choose.
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Edward O'Callaghan <eocallaghan@alterapraxis.com>
The piglit copyteximage check has recently been augmented to test this, but
apparently it hasn't been fixed in Mesa so far.
This language also already appears in the OpenGL 2.1 spec (Ian).
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
We don't need these for GLSL or ARB, but we need them for SPIR-V
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Both were defined as returning bool but the gpu_shader5 functions are
defined to return int. Also, we had the parameters for usub borrwo
backwards in the folding expression.
Reviewed-by: Matt Turner <mattst88@gmail.com>
The draw groups are now split up into groups of 32 if there's a
non-packed stride, or in groups of 400-500 if the draw data is packed.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
This shifts all indirect draws to go through the new function. If the
driver doesn't have support for multi draws, we break those up and
perform N draws. Otherwise, we pass everything through for just a single
draw call.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
This makes it possible to support indirect multidraws as well as having
the number of such draws to come from a separate GPU resource.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
All indirect draws are passed to the new draw function. By default
there's a fallback implementation which pipes it right back to
draw_prims, but eventually both the fallback and draw_prim's support for
indirect drawing should be removed.
This should allow a backend to properly support ARB_multi_draw_indirect
and ARB_indirect_parameters.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Acked-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
The sse path was pretty much disabled for practical purposes because the
largest allowed fb size was 128x128. So, adapt it for 64bit plane calculations.
This is actually not that difficult, though a problem is that we can't do
a signed 32x32->64bit mul, only unsigned, so need to fix that up. Overall,
the code still looks reasonable, though it's not like changes there in
setup really make much of a difference in the end...
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
eo, just like dcdx and dcdy, cannot overflow 32bit.
Store it as unsigned though just in case (it cannot be negative, but
in theory twice as big as dcdx or dcdy so this gives it one more bit).
This doesn't really change anything, albeit it might help minimally on
32bit archs.
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Back in the day (before 24678700ed) the values
were not actually in a struct but even then I can't see why we didn't simply
align the values. Especially since it's trivial to do so.
(Not that it actually matters since the code is pretty much unused for now.)
Reviewed-by: Oded Gabbay <oded.gabbay@gmail.com>
Otherwise, clipped lines would have undefined stippling reset bit if line
stippling is enabled.
(Untested, and I just assume copying over the bits from the original line
is actually the right thing to do.)
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
The unfilled stage was not filling in the prim header, and the line stage
then decided to reset the stipple counter or not based on the uninitialized
data. This causes some failures in conform linestipple test (albeit quite
randomly happening depending on environment).
So fill in the prim header in the unfilled stage - I am not entirely sure
if anybody really needs determinant after that stage, but there's at least
later stages (wide line for instance) which copy over the determinant as well.
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
This was added in 54f583a20 since then error handling has improved.
The test this was added to fix now fails earlier since 01822706ec
Reviewed-by: Matt Turner <mattst88@gmail.com>
In lp_build_conv() and lp_build_conv_auto(), there is a special case of
conversion when sse2 is present. That code path is suitable without any
changes to altivec, because all the functions that are called in that
code path already support altivec.
This patch increase the FPS in POWER arch across the board
between 10%-25%
I checked ipers, glxgears, glxspheres64, openarena, xonotic and glmark2.
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
The function will be extended to dump all binaries shaders will consist of,
so si_shader* makes sense here.
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Eventually, I'd like to dump stats for several combined binaries, which is
why you don't see a binary parameter in si_shader_dump_stats
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
We won't compile shaders in draw calls, but we will concatenate shader
binaries according to states in draw calls, so keep the binaries.
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
There will be 1 config per variant, which will be a union of configs
from {prolog, main, epilog}. For now, just add the structure.
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
This doesn't fix a known bug, but better safe than sorry.
Also, simplify the expression in si_shader.c.
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
because the API pixel shader binary will not emulate alpha test one day,
so the KILL_ENABLE bit must be determined elsewhere.
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
vector_insert takes a vector, a scalar location, and a scalar value,
and produces a new vector with that component updated. As such, it
can't be vectorized properly.
vector_extract takes a vector and a scalar location, and returns
that scalar component of the vector. Vectorization doesn't really
make any sense.
Treating both as horizontal operations makes sure the vectorizer
won't try to touch these.
Found by inspection.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
This makes it more similar to llvmpipe. It also allows us to let draw emit
code handle things like getting zeros for non-existing vs outputs
automatically. There probably isn't really any overhead either way, there isn't
really any "simply copy everything" code in the emit path it would copy each
attrib individually just the same. Likewise, we still do another mapping step
in softpipe as the layout may still not match exactly (same as in llvmpipe,
should probably nuke the pointless mapping in both drivers).
This fixes the piglit arb_fragment_layer_viewport no_gs/no_write tests.
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Edward O'Callaghan <eocallaghan@alterapraxis.com>
They can't actually be 0 (as position is there) but should avoid confusion.
This was supposed to have been done by af7ba989fb
but I accidentally pushed an older version of the patch in the end...
Also prettify slightly. And make some notes about the confusing and useless
fs input "map".
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Edward O'Callaghan <eocallaghan@alterapraxis.com>
draw emit couldn't care less what the interpolation mode is...
This somehow looked like it would matter, all drivers more or less
dutifully filled that in correctly. But this is only used for emit,
if draw needs to know about interpolation mode (for clipping for instance)
it will get that information from the vs anyway.
softpipe actually used to depend on that interpolation parameter, as it
abused that structure quite a bit but no longer.
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Edward O'Callaghan <eocallaghan@alterapraxis.com>
softpipe would calculate two "vertex layouts". The second one was however
just used for internal purposes, draw would know nothing about it even though
it looked exactly the same as the other one we tell draw about.
So, store that information separately as this was just confusing.
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Edward O'Callaghan <eocallaghan@alterapraxis.com>
Unlike llvmpipe, softpipe always tells draw to emit the vertices as-is.
The two vertex layouts it calculates are a bit confusing, one which is just
used to tell draw to emit vertices as-is, and the other which has draw written
all over it but draw is completely unaware of and is used only to look up the
correct interpolation info later in setup.
Thus, the slots used are different to what llvmpipe does (I'm going to clean
up the confusing two layout stuff).
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Edward O'Callaghan <eocallaghan@alterapraxis.com>
It was actually slightly buggy (missing initialization / setup not dependent
on new vs albeit I didn't see issues), but the case of non-existing attributes
is now handled by draw emit code so don't need that anymore.
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Edward O'Callaghan <eocallaghan@alterapraxis.com>
Previously the code would just redirect requests for attributes which
don't exist to use output 0. Rework this to output all zeros instead which
seems more useful - in particular some extensions like
ARB_fragment_layer_viewport require 0 in the fs even if it wasn't output by
previous stages. That way, drivers don't have to special case this depending
if the vs/gs outputs some attribute or not.
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
Reviewed-by: Edward O'Callaghan <eocallaghan@alterapraxis.com>
Rename SVGA_HINT_FLAG_DRAW_EMITTED to SVGA_HINT_FLAG_CAN_PRE_FLUSH
because preemptive flush can be unblocked by more commands than
draw.
Reviewed-by: Brian Paul <brianp@vmware.com>
The existing code effectively turns off preemptive flushing for all
but the regions used for draws. This turns out to be overly
restrictive as some memory regions, e.g. GMR, may never get a draw
when used as a DMA upload staging area, causing problems for apps
that upload a large amount of textures, e.g. Unigine Heaven.
This patch fixes the Unigine Heaven memory allocation error and
has been verified to not cause a regression in the previous extended
retina display issue.
Reviewed-by: Thomas Hellstrom <thellstrom@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
In emit_input_declarations(), we are skipping declarations for those
registers that are not being used. But in emit_vertex_attrib_instructions(),
we are still emitting instructions to tweak the vertex attributes even if
they are not being used. This causes an assert in the backend because an
input register is not declared in the shader. This patch fixes the problem
by skipping the instruction if the vertex attribute is not being used.
Changes in this patch is originated from the code snippet from Jose as
suggested in bug 1530161.
Tested with piglit, Heaven, Turbine, glretrace.
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
If the only dirty state is mesa's _NEW_PROGRAM_CONSTANTS flag, we can
skip state validation before drawing a bitmap since that state doesn't
effect bitmap rendering.
This further increases the performance of the ipers demo on llvmpipe
to about what it was before commit 36c93a6fae.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
We were checking the dirty->st flags but not the dirty->mesa flags.
When we took the early return, we didn't clear the dirty->mesa flags
so the next time we called st_validate_state() we'd often flush the
glBitmap cache. And since st_validate_state() is called from
st_Bitmap(), it meant we flushed the bitmap cache for every glBitmap()
call.
This change seems to recover most of the performance loss observed
with the ipers demo on llvmpipe since commit commit 36c93a6fae.
Cc: mesa-stable@lists.freedesktop.org
Reviewed-by: José Fonseca <jfonseca@vmware.com>
Previously each member was being counted as using a single slot,
count_attribute_slots() fixes the count for array and struct members.
Also don't assign a negitive to the unsigned expl_location variable.
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
Previously we were only reserving a single location for arrays and
structs.
We also didn't take into account implicit locations clashing with
explicit locations when assigning locations for their arrays or
structs.
This patch fixes both issues.
V5: fix regression for patch inputs/outputs in tessellation shaders
V4: just use count_attribute_slots() to get the number of slots,
also calculate the correct number of slots to reserve for gs and
tess stages by making use of the new get_varying_type() helper.
V3: handle arrays of structs
V2: also fix for arrays of arrays and structs.
Acked-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Edward O'Callaghan <eocallaghan@alterapraxis.com>
This will be used in the following patch for calculating array sizes correctly
when reserving explicit varying locations.
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Edward O'Callaghan <eocallaghan@alterapraxis.com>
Previously we would pack varyings before trying to remove them, this
relied on the packing pass not packing varyings with a location of -1
to avoid packing varyings that should be removed.
However this meant unused varyings with an explicit location would be
packed before they could be removed when we enable packing of them in a
later patch.
V2: fix regression in V1 removing unused varyings in multi-stage SSO,
fix regression with single stage programs.
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Edward O'Callaghan <eocallaghan@alterapraxis.com>
ALIGN_DIVUP is a driver specific(r600g) macro that duplicates DIV_ROUND_UP functionality.
Replacing it with DIV_ROUND_UP eliminates this problems.
Signed-off-by: Krzysztof A. Sobiecki <sobkas@gmail.com>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
I had the driver all tested for the last series, and in my last build I
noticed that get_swizzled_channel was unused now, and removed
it... apparently without testing to find that I removed the wrong channel
swizzle function.
We routinely have code like:
vec1 ssa_220 = fge ssa_104, ssa_61
vec1 ssa_199 = bcsel ssa_220, ssa_106, ssa_105
and we would compare fge's args and choose between ~0 and 0 to generate
ssa_220, then compare ssa_220 to 0 and choose between bcsel's args.
Instead, try to notice the pattern and compare between fge's args to
select between bcsel's args.
total instructions in shared programs: 88019 -> 87574 (-0.51%)
instructions in affected programs: 9985 -> 9540 (-4.46%)
total estimated cycles in shared programs: 245752 -> 245237 (-0.21%)
estimated cycles in affected programs: 17232 -> 16717 (-2.99%)
We can't use its other features currently (mostly because we don't want
Newton-Raphson on rcps for texture coordinates), but it gets us started.
This eliminates some comparisons with constants in GLB2.7 and ETQW traces
at the QIR level by moving the comparisons into NIR, where they get
constant-folded out.
instructions in affected programs: 165 -> 156 (-5.45%)
total uniforms in shared programs: 32087 -> 32085 (-0.01%)
total estimated cycles in shared programs: 245762 -> 245752 (-0.00%)
estimated cycles in affected programs: 461 -> 451 (-2.17%)
I'm moving away from QIR being SSA (since NIR is doing lots of SSA
optimization for us now) and instead having QIR just be QPU operations
with virtual registers. By making our SELs be composed of two MOVs, we
could potentially coalesce the registers for the MOV's src and dst and
eliminate the MOV.
total instructions in shared programs: 88448 -> 88028 (-0.47%)
instructions in affected programs: 39845 -> 39425 (-1.05%)
total estimated cycles in shared programs: 246306 -> 245762 (-0.22%)
estimated cycles in affected programs: 162887 -> 162343 (-0.33%)
If you want the SF of the value of a register produced from a series of
packing MOVs or conditional MOVs, we can't just SF on the last MOV into
the register.
Fix a 's/unsigned int/unsigned/' consistency case while here.
Found-by: Coccinelle
Signed-off-by: Edward O'Callaghan <eocallaghan@alterapraxis.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Fix silly issue with MSVC case fall-though support to need
a extra 'break;'
Found-by: Coccinelle
Signed-off-by: Edward O'Callaghan <eocallaghan@alterapraxis.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
This patch converts the SSE-optimized lp_rast_triangle_32_3_16()
to VMX/VSX.
I measured the results on POWER8 machine with 32 cores at 3.4GHz and
16GB of RAM.
FPS/Score
Name Before After Delta
------------------------------------------------
openarena 16.35 16.7 2.14%
xonotic 4.707 4.97 5.57%
glmark2 didn't show a significant (more than 1%) difference.
v2: Make sure code is build only on POWER8 LE machine
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
This patch converts the SSE-optimized build_mask_32() and
build_mask_linear_32() to VMX/VSX.
I measured the results on POWER8 machine with 32 cores at 3.4GHz and
16GB of RAM.
FPS/Score
Name Before After Delta
------------------------------------------------
glmark2 (score) 139.8 142.7 2.07%
openarena and xonotic didn't show a significant (more than 1%)
difference.
v2: Make sure code is build only on POWER8 LE machine
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
This patch converts the SSE optimization done in do_triangle_ccw to
VMX/VSX.
I measured the results on POWER8 machine with 32 cores at 3.4GHz and
16GB of RAM.
FPS/Score
Name Before After Delta
------------------------------------------------
glmark2 (score) 136.6 139.8 2.34%
openarena 16.14 16.35 1.30%
xonotic 4.655 4.707 1.11%
v2:
- Convert loads to use aligned loads
- Make sure code is build only on POWER8 LE machine
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
This file provides a portability layer that will make it easier to convert
SSE-based functions to VMX/VSX-based functions.
All the functions implemented in this file are prefixed using "vec_".
Therefore, when converting from SSE-based function, one needs to simply
replace the "_mm_" prefix of the SSE function being called to "vec_".
Having said that, not all functions could be converted as such, due to the
differences between the architectures. So, when doing such
conversion hurt the performance, I preferred to implement a more ad-hoc
solution. For example, converting the _mm_shuffle_epi32 needed to be done
using ad-hoc masks instead of a generic function.
All the functions in this file support both little-endian and big-endian
but currently the file is build only on POWER8 LE machine.
All of the functions are implemented using the Altivec/VMX intrinsics,
except one where I needed to use inline assembly (due to missing
intrinsic).
v2:
- Use vec_vgbbd instead of __builtin_vec_vgbbd
- Add an aligned load function
- Don't use typeof()
- Make file build only on POWER8 LE machine
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
To determine if we could use special POWER8 assembly directives, we first
need to detect whether we are running on POWER8 architecture. This patch
adds this detection to configure.ac and adds the necessary compilation
flags accordingly.
v2:
- Add option to disable POWER8 instructions generation
- Detect whether building on BE or LE machine and build with
-mpower8-vector only on LE machine
- Make the printed messages more standard
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
The nir_opt_algebraic rule
(('fadd', ('flog2', a), ('fneg', ('flog2', b))), ('flog2', ('fdiv', a, b))),
can produce new fdiv operations, which need to be lowered on i965,
as we don't actually implement fdiv. (Normally, we handle this in
GLSL IR's lower_instructions pass, but in the above case we introduce
an fdiv after that point. So, make NIR do it for us.)
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Cc: mesa-stable@lists.freedesktop.org
Compute shaders require reconfiguring the L3 for shared local memory
support. We have to be able to write the L3 registers to do that.
This effectively turns off compute shaders prior to Kernel 4.2.
(Previously, the extension enable was in an API_OPENGL_CORE conditional.
However, that isn't necessary - core Mesa extension handling already
restricts it properly. I've moved it out in this patch.)
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Francisco Jerez <currojerez@riseup.net>
This fixes some piglit subtests for ARB_program_interface_query.
V3: remove some of the unnecessary parentheses
V2: fix alignment
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
There is a function dedicated to demoting unused varyings lets
trust it to do its job.
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Edward O'Callaghan <eocallaghan@alterapraxis.com>
After lowering the matching flag is_unmatched_generic_inout is lost so
we need to move this validation before lowering.
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Edward O'Callaghan <eocallaghan@alterapraxis.com>
An SSO program can have multiple stages and we only want to add the externally
facing varyings. The current code was adding both the packed inputs and outputs
for the first and last stage of each program.
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Edward O'Callaghan <eocallaghan@alterapraxis.com>
Conditions modified allow skl+ to use blitter:
- for all tiling formats
- to write data to YF/YS tiled surfaces
Signed-off-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
Overlapping blits are anyway undefined in OpenGL. So no need
of overlap check here.
Signed-off-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
It's very rare that a GL app calls glVertex3dv(), but one in particular
calls it lot, always with Z = 0. Check for that condition and convert
the call into glVertex2f. This reduces VBO memory used and reduces
the number of times we have to switch between float[2] and float[3]
vertex formats in the svga driver. This results in a small but
measurable performance improvement.
Reviewed-by: Charmaine Lee <charmainel@vmware.com>
We only want to set the SVGA_NEW_STIPPLE dirty flag when the polygon
stipple state changes. Before, we only set the flag when we were
enabling stipple, but not disabling.
We don't really have to add SVGA_NEW_STIPPLE to the dirty FS state
set since it's a subset of SVGA_NEW_RAST, but let's be explicit.
This doesn't fix any known bugs.
Reviewed-by: Charmaine Lee <charmainel@vmware.com>
and svga_set_sampler_views(). If there's no change, return early
and don't set a SVGA_NEW_x dirty state flag.
Reviewed-by: Charmaine Lee <charmainel@vmware.com>
gcc 4.9.3 shows the following error:
brw_vue_map.c:260:20: warning: array subscript is above array bounds
[-Warray-bounds]
return brw_names[slot - VARYING_SLOT_MAX];
This is because BRW_VARYING_SLOT_COUNT is a valid value for the enum
type. Adding an assert will generate no additional code but will teach
the compiler to not complain.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Timothy Arceri <timothy.arceri@collabora.com>
The counter was not set but used by the nouveau driver.
It is required otherwise visual output is garbage.
Signed-off-by: Julien Isorce <j.isorce@samsung.com>
Reviewed-by: Christian Koenig <christian.koenig@amd.com>
This fixes the same tests that commit 8cf2e892f was attempting to fix:
ES31-CTS.shader_storage_buffer_object.advanced-unsizedArrayLength-cs-std430-vec-bindrangeOffset
ES31-CTS.shader_storage_buffer_object.advanced-unsizedArrayLength-cs-std430-vec-bindrangeSize
as confirmed by Samuel.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Cc: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
Cc: Marta Lofstedt <marta.lofstedt@intel.com>
Reviewed-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
This reverts commit 8cf2e892fc. It's
entirely bogus to attempt to store anything about the binding in the
buffer object itself, which might be bound any number of times.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Cc: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
Cc: Marta Lofstedt <marta.lofstedt@intel.com>
Reviewed-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
Instead, keep track of GL_DEBUG_OUTPUT and (un)install the pipe_debug_callback
accordingly. Hardware drivers can still use the absence of the callback to
skip more expensive operations in the normal case, and users can no longer be
surprised by the need to set the debug flag at context creation time.
v2:
- re-add the proper initialization of debug contexts (Ilia Mirkin)
- silence a potential warning (Ilia Mirkin)
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
Commit 5bb5eeea fixes a bug indicating that the surfaces should have the
API buffer size. Hovewer it picked the wrong value.
This patch adds a new variable, which takes into account
glBindBufferRange() values. This patch fixes the following CTS
regressions:
ES31-CTS.shader_storage_buffer_object.advanced-unsizedArrayLength-cs-std430-vec-bindrangeOffset
ES31-CTS.shader_storage_buffer_object.advanced-unsizedArrayLength-cs-std430-vec-bindrangeSize
Signed-off-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
Reviewed-by: Marta Lofstedt <marta.lofstedt@intel.com>
Reported by Tom^ on IRC. The original intent was to mark the pointer
constant as well as the data being pointed to, so move the *.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Note these are a bit uglier, due to avoidance of GNU C extensions. But
drivers which do not need to be built with compilers that don't support
the extension can wrap these macros with their own.
Signed-off-by: Rob Clark <robclark@freedesktop.org>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Found during NIR_TEST_CLONE=1 piglit run. We were using block->index
but forgetting to require it. Causing things to not work with a cloned
shader which didn't preserve block_index.
Signed-off-by: Rob Clark <robclark@freedesktop.org>
Immediately convert into NIR and do an initial key-agnostic lowering/
optimization pass. This should let us share most of the per-variant
transformations between each variant, and hopefully minimize the draw-
time variant creation part of the compilation process.
Signed-off-by: Rob Clark <robclark@freedesktop.org>
It will still hit a compile_assert() in emit_tex, which has the
advantage of dumping out the offending shader.
Signed-off-by: Rob Clark <robclark@freedesktop.org>
Instead of iterating over all the buffer resources looking for coherent
buffers, we keep track of a context-wide count. This will save some
iterations (and CPU cycles) in 99.99% case because usually coherent
buffers are not so used.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
If there's a linked TES program, we should just use the actual
primitive mode. If not, just guess triangles (as we did before).
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Using the push model for inputs is much more efficient than pulling
inputs - the hardware can simply copy a large chunk into URB registers
at thread creation time, rather than having the thread send messages to
request data from the L3 cache. Unfortunately, it's possible to have
more TES inputs than fit in registers, so we have to fall back to the
pull model in some cases.
However, it turns out that most tessellation evaluation shaders are
fairly simple, and don't use many inputs. An arbitrary cut-off of
32 vec4 slots (16 registers) is more than sufficient to ensure that
100% of TES inputs are pushed for Shadow of Mordor, Unigine Heaven,
GPUTest/TessMark, and SynMark.
Note that unlike most SIMD8 stages, this actually reads packed vec4
data, since that is what our vec4 TCS programs write.
Improves performance in GPUTest's tessmark_x64 microbenchmark
by 93.4426% +/- 5.35541% (n = 25) on my Lenovo X250 at 1024x768.
Improves performance in Synmark's Gl40TerrainFlyTess microbenchmark
by 22.74% +/- 0.309394% (n = 5).
Improves performance in Shadow of Mordor at low settings with
tessellation enabled at 1280x720 by 2.12197% +/- 0.478553% (n = 4).
shader-db statistics for files containing tessellation shaders:
total instructions in shared programs: 184358 -> 181181 (-1.72%)
instructions in affected programs: 27971 -> 24794 (-11.36%)
helped: 226
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
We need a MOV to replicate g0.0<0,1,0> to all 8 channels. Since the
message payload is a single register, MOV seemed more sensible than
LOAD_PAYLOAD. However, MOV cannot be CSE'd, while LOAD_PAYLOAD can.
All input loads can use the same header - we don't need to re-expand
g0 every time. CSE accomplishes this, saving instructions.
shader-db statistics for files containing tessellation shaders:
total instructions in shared programs: 186923 -> 184358 (-1.37%)
instructions in affected programs: 30536 -> 27971 (-8.40%)
helped: 226
HURT: 0
total cycles in shared programs: 1009850 -> 1005356 (-0.45%)
cycles in affected programs: 168206 -> 163712 (-2.67%)
helped: 226
HURT: 0
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
While most align16 instructions only support a SubRegNum of 0 or 4
(using swizzling to control the other channels), 3-src instructions
actually support arbitrary SubRegNums. When the RepCtrl bit is set,
we believe it ignores the swizzle and uses the equivalent of a <0,1,0>
region from the subnr.
In the past, we adopted a vec4-centric approach of specifying subnr of
0 or 4 and a swizzle, then having brw_eu_emit.c convert that to a proper
SubRegNum. This isn't a great fit for the scalar backend, where we
don't set swizzles at all, and happily set subnrs in the range [0, 7].
This patch changes brw_eu_emit.c to use subnr and swizzle directly,
relying on the higher levels to set them sensibly.
This should fix problems where scalar sources get copy propagated into
3-src instructions in the FS backend. I've only observed this with
TES push model inputs, but I suppose it could happen in other cases.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Diagnostics sent during code generation and the every error message reported
by LLVMTargetMachineEmitToMemoryBuffer are disjoint reporting mechanisms. We
take care of both and also send an explicit message indicating failure at the
end, so that log parsers can more easily tell the boundary between shader
compiles.
Removed an fprintf that could never be triggered.
Reviewed-by: Edward O'Callaghan <eocallaghan@alterapraxis.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
This will allow us to send shader debug info via the context's debug callback.
Reviewed-by: Edward O'Callaghan <eocallaghan@alterapraxis.com> (v1)
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
The output via stderr is very helpful for ad-hoc debugging tasks, so that remains
unchanged, but having the information available via debug messages as well
will allow the use of parallel shader-db runs.
Shader stats are always provided (if the context is a debug context, that is),
but you still have to enable the appropriate R600_DEBUG flags to get
disassembly (since it is rather spammy and is only generated by LLVM when we
explicitly ask for it).
Reviewed-by: Edward O'Callaghan <eocallaghan@alterapraxis.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
The fixed alignment of u_upload_mgr will go away.
This is the first step.
The motivation is that one u_upload_mgr can have multiple users,
each allocating from the same buffer, but requiring a different alignment.
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
The function only aligned the size, but not the offset.
The offset was aligned only when the previous suballocation was aligned.
That yielded the correct offset alignment if the alignment was constant
for all suballocations.
Instead, directly align the offset, but allow an unaligned size.
There is no change in behavior, because the alignment is constant
at the moment.
This a prerequisite for allowing a variable alignment for suballocations.
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Spotted by luck. The GLSL uniform storage is only associated once
in LinkShader and can't be reallocated afterwards, because that would
break the association.
v2: don't remove st_upload_constants calls, clarify why they're needed
Cc: 11.0 11.1 <mesa-stable@lists.freedesktop.org>
First off, we can't flush in the middle of a command. Secondly
requesting the extra push space might cause a flush to happen. If that
flush happens, we'd have to do the PUSH_REFN again. So instead do
PUSH_REFN after the push space request. This helps avoid rare crashes
with supertuxkart in libdrm due to assertion failures.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Cc: "11.0 11.1" <mesa-stable@lists.freedesktop.org>
While playing with fp64, I disable varying packing to debug
something else, and noticed we never emitted half the output
movs for double matrix arrays.
We should be moving the left index two slots for dual
source doubles, and the right index two slots for non-vs
input doubles.
Signed-off-by: Dave Airlie <airlied@redhat.com>
vertex inputs are counted differently in some cases, with
vertex inputs we need to make sure we don't double count them.
Signed-off-by: Dave Airlie <airlied@redhat.com>
Otherwise we end up emitting the wrong index for the second
double.
This fixes dmat-vs-gs-tcs-tes.shader_test and dvec3-vs-gs-tcs-tes.shader_test
Signed-off-by: Dave Airlie <airlied@redhat.com>
It's important for the double instruction emission code that
the writemasks are correct going in for double so it know
which channels to replicate.
This fixes it for the array and matrix cases.
Signed-off-by: Dave Airlie <airlied@redhat.com>
This code takes into account double inputs in the array
shrinking code. This fixes some issues with doubles
and geom/tess inputs.
Signed-off-by: Dave Airlie <airlied@redhat.com>
This handles the case where a double output is stored
in an array, and tracks it for use in the double
instruction emit code.
Signed-off-by: Dave Airlie <airlied@redhat.com>
This is just a precursor patch to a fix for doubles with
tessellation that I've written.
We need to descend into output arrays in that case and
mark dst's as double.
Signed-off-by: Dave Airlie <airlied@redhat.com>
varying_matches::record tries to compute the number of components in
each varying, which varying_matches::assign_locations uses to assign
locations. With varying packing, it uses glsl_type::component_slots()
to come up with a reasonable value.
Without varying packing, it fell back to an open-coded computation
that didn't bother to handle structs at all. I believe we can simply
use 4 * glsl_type::count_attribute_slots(false), which already handles
these cases correctly.
Partially fixes rendering in GFXBench 4.0's tessellation benchmark.
(NVE0 is almost right after this, but i965 is still mostly garbage.)
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Timothy Arceri <timothy.arceri@collabora.com>
Cc: "11.0 11.1" <mesa-stable@lists.freedesktop.org>
Unigine Heaven 4.0 and Valley 1.0 use dual color blending but don't
specify which fragment shader output is which, so there's at best a
50/50 chance of us guessing it correctly. This is invalid.
Unigine fixed this in 4.1 and 1.1 versions over a year and a half ago,
but hasn't actually released them for whatever reason. So, add the
workaround back so that it works for most people.
Fixes Heaven 4.0/Valley 1.0 rendering on Ivybridge. For whatever
reason, Broadwell worked. 4.1 and 1.1 have always worked.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=92233
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Cc: mesa-stable@lists.freedesktop.org
Hooks up the new system values, passes the drawid in.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
This will allow the state tracker to inform the driver where in a
broken-up multidraw we currently are. This can then be passed into the
vertex shader.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
This allows the state tracker to know that the various draw parameters
are available in vertex shaders.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
The add might actually have a 0 as an argument, which would convert it
into a mov. Make sure to detect that. Also avoid the hack of putting the
immediate directly into the instruction, instead use a mov to put it
into place and let the later LoadPropagation pass place it if possible.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
The imulExtended tests of the shader bitfield tests of the
OpenGL ES 3.1 CTS, fail on gen8+, when BRW_REGISTER_TYPE_W
is used for SHADER_OPECODE_MULH.
Also, remove unused helper function:
static inline bool type_is_signed(unsigned type)
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=92595
Signed-off-by: Marta Lofstedt <marta.lofstedt@linux.intel.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
There used to be more members but they now share other fields
in order to keep memory use low.
Also making the naming more generic will allow us to reuse the
field for explicit byte offsets within blocks for
ARB_enhanced_layouts.
Reviewed-by: Edward O'Callaghan <eocallaghan@alterapraxis.com>
A hugely common case when using nir_builder is to have a shader with a
single function called main. This adds a helper that gives you just that.
This commit also makes us use it in the NIR control-flow unit tests as well
as tgsi_to_nir and prog_to_nir.
Reviewed-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Acked-by: Kenneth Graunke <kenneth@whitecape.org>
GL_ARB_shader_draw_parameters added two new system values. This gets us
back to mapping mesa system values to the right TGSI semantics.
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
If we're doing an indirect draw, prims[i].basevertex is always 0 and the
real base vertex value is in the indirect parameter buffer. We try to
avoid flagging BRW_NEW_VERTICES if prims[i].basevertex doesn't change,
which then breaks down for indirect draws. Thus, if a program uses base
vertex or base instance, and the draw call is indirect, always flag
BRW_NEW_VERTICES. A new piglit test,
spec/ARB_shader_draw_parameters/drawid-indirect-vertexid tests this.
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
This optimizes a + b - b to just a. Modest shader-db results (BDW):
total instructions in shared programs: 7842452 -> 7841862 (-0.01%)
instructions in affected programs: 61938 -> 61348 (-0.95%)
total loops in shared programs: 2131 -> 2131 (0.00%)
helped: 263
HURT: 0
GAINED: 0
LOST: 0
but the optimization turns
gl_VertexID - gl_BaseVertexARB
into just a reference to SYSTEM_VALUE_VERTEX_ID_ZERO_BASE, which the
i965 hardware supports natively. That means we can avoid using the
internal vertex buffer for gl_BaseVertexARB in this case.
Reviewed-by: Eduardo Lima Mitev <elima@igalia.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
We have to break open a new vec4 for gl_DrawIDARB. We've used up all
space in the vec4 we use for SGVS and gl_DrawIDARB has to come from its
own separate vertex buffer anyway. This is because we point the vb for
base vertex and base instance into the draw parameter BO for indirect
draw calls, but the draw id is generated by mesa in a different buffer.
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
We already have gl_BaseVertexARB in the .x component of the SGVS vec4
and plug gl_BaseInstanceARB into the last free component (.y).
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
fs_visitor::emit_vs_system_value() looks like it's trying to handle
SYSTEM_VALUE_VERTEX_ID, but we should never see that value in the
backend.
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
The drivers will need this for passing in gl_DrawIDARB. For indirect
multidraw calls, we get the prim array and prim[i].draw_id == i and is
redundant. But for non-indirect calls, we get one primitive at a time
and need the draw_id field.
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
This option allows replacing a single shader by a pre-compiled ELF object
as generated by LLVM's llc, for example. This can be useful for debugging a
deterministically occuring error in shaders (and has in fact helped find
the causes of https://bugs.freedesktop.org/show_bug.cgi?id=93264).
v2: drop the debug flag, use DEBUG_GET_ONCE_OPTION instead
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
This changes the count slightly (because of si_generate_gs_copy_shader), but
this is only relevant for the driver-specific num-compilations query. It sets
the stage for the next commit.
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Setting interleave on the TCS EOT message causes Ivybridge hardware to
GPU hang like crazy. Individual tests would pass, but running even a
simple test like nop.shader_test in a loop would hang within 1-3 runs.
Adding sleep delays worked around the problem, somehow.
Interleave doesn't make much sense given that we only have one patch
URB handle, not two. Complete doesn't seem useful either.
There's no reason to actually set those bits. We were just being lazy.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Edward O'Callaghan <eocallaghan@alterapraxis.com>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Pre-Broadwell hardware requires us to manually release the ICP Handles
by issuing URB read messages with the "Complete" bit set. We can do
this in pairs to use fewer URB read messages.
Based heavily on work from Chris Forbes.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Edward O'Callaghan <eocallaghan@alterapraxis.com>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
When Connor originally drafted NIR, he copied the same function+overload
system that GLSL IR had with a few names changed. However, this
double-indirection is not really needed and has only served to confuse
people. Instead, let's just have functions which may not have unique names
and may or may not have an implementation. If someone wants to do overload
resolving, they can hav a hash table based function+overload system in the
overload resolving pass. There's no good reason to keep it in core NIR.
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Acked-by: Kenneth Graunke <kenneth@whitecape.org>
ir3 bits are
Reviewed-by: Rob Clark <robclark@gmail.com>
NIR has never been built with MSVC2008, so we shouldn't add
MSVC2008_COMPAT_CFLAGS to anything that uses it. This allows us to get
rid of the pragma in tgsi_to_nir.c.
Build tested with freedreno.
v2: Use MSVC2013_COMPAT_CLFAGS instead.
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
Signed-off-by: Connor Abbott <cwabbott0@gmail.com>
The BDW PRM Vol2a: Command Reference: Instructions, section MEDIA_CURBE_LOAD,
says that 'CURBE Total Data Length' and 'CURBE Data Start Address' are
64-byte aligned. This is different from previous gens, that were 32-byte
aligned.
v2 (Jordan):
- CURBE Data Start Address is also 64-byte aligned.
- The call to brw_state_batch should also use 64-byte alignment.
- Improve PRM reference.
v3:
* New patch from Jordan. Always align base and size to 64 bytes.
Fixes the following SSBO CTS tests on BDW:
ES31-CTS.shader_storage_buffer_object.basic-atomic-case1-cs
ES31-CTS.shader_storage_buffer_object.basic-operations-case1-cs
ES31-CTS.shader_storage_buffer_object.basic-operations-case2-cs
ES31-CTS.shader_storage_buffer_object.basic-stdLayout_UBO_SSBO-case2-cs
ES31-CTS.shader_storage_buffer_object.advanced-write-fragment-cs
ES31-CTS.shader_storage_buffer_object.advanced-indirectAddressing-case2-cs
ES31-CTS.shader_storage_buffer_object.advanced-matrix-cs
And many other CS CTS tests as reported by Marta Lofstedt.
(Commit message is from Iago, but in v3, code is from Jordan.)
Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Tested-by: Iago Toral Quiroga <itoral@igalia.com>
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Everything is in place and I'm not aware of any further issues.
Tested with:
- Piglit
- Tessmark
- Unigine Heaven
- Shadow of Mordor
- GRID Autosport
I have patches to backport this to Haswell, Ivybridge, and Baytrail as
well (the first Intel hardware to support tessellation), but there are
still a lot of GPU hangs left to debug. So that will come later.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
GL_ARB_separate_shader_objects allows the application to mix-and-match
TCS and TES programs separately. This means that the interface between
the two stages isn't known until the final SSO pipeline is in place.
This isn't a great match for our hardware: the TCS and TES have to agree
on the Patch URB entry layout. Since we store data as per-patch slots
followed by per-vertex slots, changing the number of per-patch slots can
significantly alter the layout. This can easily happen with SSO.
To handle this, we store the [Patch]OutputsWritten and [Patch]InputsRead
bitfields in the TCS/TES program keys, introducing program recompiles.
brw_upload_programs() decides the layout for both TCS and TES, and
passes it to brw_upload_tcs/tes(), which store it in the key.
When creating the NIR for a shader specialization, we override
nir->info.inputs_read (and friends) to the program key's values.
Since everything uses those, no further compiler changes are needed.
This also replaces the hack in brw_create_nir().
To avoid recompiles, brw_precompile_tes() looks to see if there's a
TCS in the linked shader. If so, it accounts for the TCS outputs,
just as brw_upload_programs() would. This eliminates all recompiles
in the non-SSO case. In the SSO case, there should only be recompiles
when using a TCS and TES that have different input/output interfaces.
Fixes Piglit's mix-and-match-tcs-tes test.
v2: Pull the brw_upload_programs code into a brw_upload_tess_programs()
helper function (requested by Jordan Justen).
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
With tessellation shaders and SSO, we won't be able to always decide on
VUE map layouts at LinkProgram time. Unfortunately, we have to delay it
until shader specialization time.
However, uniform lowering cannot be deferred - brw_codegen_*_prog()
reads nir->num_uniforms. Fortunately, we don't need to defer it -
uniform, system value, atomic, and sampler lowering can safely stay
where it is. This patch moves those to brw_lower_nir()'s only caller,
renames brw_lower_nir() to brw_nir_lower_io(), and introduces calls
to that.
For non-tessellation stages, I chose to call brw_nir_lower_io() from
brw_create_nir(), so it's still done at the same time. There's no
need to defer it, and doing it at LinkProgram time is nice.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
This way, I can safely use brw_tcs_prog_key::program_string_id == 0
to mean "not filled out because no program exists", which avoids the
need for adding an extra boolean to that struct.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
When the application hasn't supplied a TCS, and we have to create one,
we need to know what VS outputs to copy to TES inputs.
To do this, we create a new program key field, and set it to the TES
InputsRead bitfield.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
When using tessellation on OpenGL without a TCS, default values for
gl_TessLevelOuter/gl_TessLevelInner are provided via the API.
Core Mesa will flag ctx->DriverFlags.NewDefaultTessLevels whenever those
values change. We add a corresponding BRW_NEW_DEFAULT_TESS_LEVELS flag
and hook it up to HS push constants (which will be used to upload these
default values to the autogenerated TCS).
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
With the automatic-TCS creation, we won't have a prog, but still need to
upload push constants.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Tessellation control shaders are optional, but evaluation shaders will
always be present when using tessellation. However, we'll always enable
the TCS (HS) hardware stage when using tessellation - we'll just create
a program on the fly.
That program, however, won't have a gl_program or gl_shader_program.
So we shouldn't check brw->tess_ctrl_program or
shader_prog->_LinkedShaders[MESA_SHADER_TESS_CTRL] - if we want to know
whether tessellation is enabled, we should look for a TES.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
This is trying to enforce the fact that the hardware requires HS, TE,
and DS to be enabled or disabled together. But it's kind of an ad-hoc
attempt, and not too useful.
More importantly, we aren't going to have a gl_shader_program for the
TCS which is automatically generated when none is present. (We'll just
handle it in the driver backend.) So, these will trip for no reason.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
For several reasons, I don't think it's particularly useful to have
separate flags:
1. Most of the time, tessellation shaders are paired, so both will be
replaced at the same time.
2. The data layout is tightly coupled. Both need to agree on the number
of per-patch slots in the VUE map. Even adding extra TCS outputs
that aren't read by the TES will trigger the need for recompiles.
3. The TCS is optional from an API perspective, but required by the
hardware whenever tessellation is enabled. So, atoms that deal with
the TCS must check brw->tess_eval_program (BRW_NEW_TESS_EVAL_PROGRAM?)
rather than brw->tess_ctrl_program to tell whether tessellation is
enabled.
So, not only is it unlikely to be useful, it's a bit confusing to get
right. Simply using one flag for both simplifies this.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
If there's no evaluation shader, tessellation is disabled. The upload
functions would just bail. Instead, don't bother calling them.
This will simplify the optional-TCS case a bit, as brw_upload_tcs can
assume that we're doing tessellation.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
I need access to glsl_type::vec2_type from C. Wrapping vec() also gives
us access to vec3 if we need it.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Instead of performing the read-modify-write cycle in glsl->nir, we can
simply emit a partial writemask. For locals, nir_lower_vars_to_ssa will
do the equivalent read-modify-write cycle for us, so we continue to get
the same SSA values we had before.
Because glsl_to_nir calls nir_lower_outputs_to_temporaries, all outputs
are shadowed with temporary values, and written out as whole vectors at
the end of the shader. So, most consumers will still not see partial
writemasks.
However, nir_lower_outputs_to_temporaries bails for tessellation control
shader outputs. So those remain actual variables, and stores to those
variables now get a writemask. nir_lower_io passes that through. This
means that TCS outputs should actually work now.
This is a functional change for tessellation control shaders.
v2: Relax the nir_validate assert to allow partial writemasks.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
Tessellation control shaders need to be careful when writing outputs.
Because multiple threads can concurrently write the same output
variables, we need to only write the exact components we were told.
Traditionally, for sub-vector writes, we've read the whole vector,
updated the temporary, and written the whole vector back. This breaks
down with concurrent access.
This patch prepares the way for a solution by adding a writemask field
to store_var intrinsics, as well as the other store intrinsics. It then
updates all produces to emit a writemask of "all channels enabled". It
updates nir_lower_io to copy the writemask to output store intrinsics.
Finally, it updates nir_lower_vars_to_ssa to handle partial writemasks
by doing a read-modify-write cycle (which is safe, because local
variables are specific to a single thread).
This should have no functional change, since no one actually emits
partial writemasks yet.
v2: Make nir_validate momentarily assert that writemasks cover the
complete value - we shouldn't have partial writemasks yet
(requested by Jason Ekstrand).
v3: Fix accidental SSBO change that arose from merge conflicts.
v4: Don't try to handle writemasks in ir3_compiler_nir - my code
for indirects was likely wrong, and TTN doesn't generate partial
writemasks today anyway. Change them to asserts as requested by
Rob Clark.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com> [v3]
Patch makes following changes for interface matching:
- do not try to match builtin variables
- handle swizzle in input name, as example 'a.z' should
match with 'a'
- add matching by location
- check that amount of inputs and outputs matches
These changes make interface matching tests to work in:
ES31-CTS.sepshaderobjs.StateInteraction
The test still does not pass completely due to errors in rendering
output. IMO this is unrelated to interface matching.
Note that type matching is not done due to varying packing which
changes type of variable, this can be added later on. Preferably
when we have quicker way to iterate resources and have a complete
list of all existed varyings (before packing) available.
v2: add spec reference, return true on desktop since we do not
have failing cases for it, inputs and outputs amount do not
need to match on desktop.
v3: add some more spec reference, remove desktop specifics since
not used for now on desktop, add match by location qualifier,
rename input_stage and output_stage as producer and consumer
as suggested by Timothy.
Signed-off-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Timothy Arceri <timothy.arceri@collabora.com>
The i965 driver uses this function to decide if it can disable the
FS unit in the absence of color/depth writes. We don't want to disable
the unit in the presence of SSBOs, since the fragment shader could
be writing to it.
We could go a step further and check not just for the presence of SSBOs
but also if the shader code writes to them. Does not look worth the trouble
though and we are not doing this for atomic buffers either anyway.
v2: put this into a generic _mesa_active_fragment_shader_has_side_effects
function instead of having one specific for SSBOs (Jason).
Fixes the following CTS test:
ES31-CTS.shader_storage_buffer_object.advanced-usage-sync-vsfs
Reviewed-by: Francisco Jerez <currojerez@riseup.net>
On Haswell we need to set the UAV_ONLY WM state bit when there are no colour
or depth buffer writes and on all hardware we should set the early
depth/stencil control field to PSEXEC unless early fragment tests are enabled
to make sure that the fragment shader is executed regardless of whether
per-fragment tests pass or not as the spec requires.
So far we have been doing this for images only, but we should apply the same
treatment to all side effectful scenarios. Suggested by Curro.
This is not strictly required for compliance with the original
ARB_shader_atomic_counters extension, it's only necessary to get the execution
semantics specified in GL4.2+ right.
v2:
- Mark active_fs_has_side_effects as constant. (Curro)
- Mention that this is only only necessary to get the execution semantics
specified in GL4.2+ right. (Curro)
Reviewed-by: Francisco Jerez <currojerez@riseup.net>
Some drivers can disable the FS unit if there is nothing in the shader code
that writes to an output (i.e. color, depth, etc). Right now, mesa has
a function to check for atomic buffers and the i965 driver also checks for
images. Refactor this logic into a generic function that we can use for
any source of side effects in a fragment shader. Suggested by Jason.
v2:
- Use '_Shader', as suggested by Tapani, to fix the following CTS test:
ES31-CTS.shader_atomic_counters.advanced-usage-many-draw-calls2
Reviewed-by: Francisco Jerez <currojerez@riseup.net>
The hardware provides us no decent way of getting at the number of input
vertices in the patch topology from the tessellation control shader.
It's actually very surprising - normally this sort of information would
be available in the thread payload.
For the precompile, we guess that the number of vertices will be the
same for both the input and output patches. This usually seems to be
the case.
On Gen8+, we could pass in an extra push constant containing this value.
We may be able to do that on Haswell too. It's quite a bit trickier on
Ivybridge, however.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
The TCS is the first tessellation shader stage, and the most
complicated. It has access to each of the control points in the input
patch, and computes a new output patch. There is one logical invocation
per output control point; all invocations run in parallel, and can
communicate by reading and writing output variables.
One of the main responsibilities of the TCS is to write the special
gl_TessLevelOuter[] and gl_TessLevelInner[] output variables which
control how much new geometry the hardware tessellation engine will
produce. Otherwise, it simply writes outputs that are passed along
to the TES.
We run in SIMD4x2 mode, handling two logical invocations per EU thread.
The hardware doesn't properly manage the dispatch mask for us; it always
initializes it to 0xFF. We wrap the whole program in an IF..ENDIF block
to handle an odd number of invocations, essentially falling back to
SIMD4x1 on the last thread.
v2: Update comments (requested by Jordan Justen).
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
The TES is essentially a post-tessellator VS, which has access to the
entire TCS output patch, and a special gl_TessCoord input. Otherwise,
they're very straightforward.
This patch implements SIMD8 tessellation evaluation shaders for Gen8+.
The tessellator can generate a lot of geometry, so operating in SIMD8
mode (8 vertices per thread) is more efficient than SIMD4x2 mode (only
2 vertices per thread). I have another patch which implements SIMD4x2
mode for older hardware (or via an environment variable override).
We currently handle all inputs via the pull model.
v2: Improve comments (suggested by Jordan Justen).
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
This field is used as a flag to optimise out any varyings that don't have
a matching varying on the other side of the interface.
The value should be the same for all varyings (except for SSO but we can't
optimise those) by the time they reach nir and are no longer be needed.
Acked-by: Jason Ekstrand <jason.ekstrand@intel.com>
Also emits a method to properly bind the class to a subchannel, which
was missing previously. The kernel currently doesn't care, but this
will break if it ever decides to (ie. to support multiple sw classes).
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Reviewed-by: Emil Velikov <emil.l.velikov@gmail.com>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Tested-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
The kernel previously exposed incorrect classes for some of the chipsets
that this code supports. It no longer does, but the older object ioctls
have compatibility to avoid breaking userspace.
This needs to be fixed before switching over to the newer interfaces.
Rather than hardcoding chipset->class like the rest of the driver does,
this makes use of (new) sclass queries to determine what's available.
v2.
- update to use symbolic class identifier from <nvif/class.h>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Reviewed-by: Emil Velikov <emil.l.velikov@gmail.com>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Tested-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Switching to the newer libdrm entry-points tells libdrm that it's OK to
make use of newer kernel interfaces.
We want to be able to isolate any bugs to either the interfaces changes,
or the use of NVIF itself. As such, this commit has a slight hack which
forces libdrm to continue using the older kernel interfaces.
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Reviewed-by: Emil Velikov <emil.l.velikov@gmail.com>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Tested-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
The winsys layer would attempt to cleanup the nouveau_device if screen
init failed, however, in most paths the pipe driver would have already
destroyed it, resulting in accesses to freed memory etc.
This commit fixes the problem by allowing the winsys to detect whether
the pipe driver's destroy function needs to be called or not.
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Reviewed-by: Emil Velikov <emil.l.velikov@gmail.com>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Tested-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
If oViewport is written, vertex reuse need to be turned off.
If oViewport is constant, vertex reuse is fine, and VPORT_PROVOKE_DISABLE
need to be set. (we don't have enough info to program VPORT_PROVOKE).
Fixes: arb_viewport_array-render-viewport-2 and some CTS tests.
v2: drop vport provoke write, drop initial state writing this
on evergreen, only program it on evergreen.
Signed-off-by: Dave Airlie <airlied@redhat.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
If oViewport is written, vertex reuse need to be turned off.
If oViewport is constant, vertex reuse is fine, and VPORT_PROVOKE_DISABLE
need to be set. (We don't know if oViewport is constant so we
skip this.)
Fixes: arb_viewport_array-render-viewport-2 and some CTS tests.
v2: drop writing to provoke disable, drop write in initial
state.
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
This function deals with vertex inputs and fragment
outputs, so we should count the attribute locations
correctly for the vertex inputs.
Reviewed-by: Timothy Arceri <timothy.arceri@collabora.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
I was cleverly using one iteration to obtain a pointer to the last item
in ralloc's singly list child list, while also setting parents.
Unfortunately, I forgot to set the parent on that last item.
Cc: "11.1 11.0 10.6" <mesa-stable@lists.freedesktop.org>
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
This doubles the element width for the types that are greater
than 2 elements wide.
Reviewed-by: Timothy Arceri <timothy.arceri@collabora.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
This doesn't apply to other stages. This is only
used in the mesa/st code, which needs further fixes.
Reviewed-by: Timothy Arceri <timothy.arceri@collabora.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
This is a rewrite of vc4_opt_qpu_schedule.c to operate on QIR. Texture
fetch can probably take as much as the rest of the cycles of the program,
so it's important to hide our other cycles during it (which is hard to do
after register allocation). Also, we can queue up multiple texture
requests before collecting the resulting samples, so that we keep the
texture unit busy more of the time.
High-settings openarena performance +2.35849% +/- 0.221154% (n=7). Also
about 2-3% on the multiarb demo. 8 piglit tests
(ext_framebuffer_multisample accuracy depthstencil) go from failing in
rendering to failing in register allocation, but hopefully I can fix that
up with some better register pressure handling here.
total instructions in shared programs: 87723 -> 88448 (0.83%)
instructions in affected programs: 78411 -> 79136 (0.92%)
total estimated cycles in shared programs: 276583 -> 246306 (-10.95%)
estimated cycles in affected programs: 265691 -> 235414 (-11.40%)
There's only high latency between a complete texture fetch setup and
collecting its result, not between each step of setting up the texture
fetch request.
So vertex shader input attributes are handled different than internal
varyings between shader stages, dvec3 and dvec4 only count as
one slot for vertex attributes, but for internal varyings, they
count as 2.
This patch comments all the uses of this API to clarify what we
pass in, except one which needs further investigation
Signed-off-by: Dave Airlie <airlied@redhat.com>
Reviewed-by: Timothy Arceri <timothy.arceri@collabora.com>
The old function didn't work for matrices, and we need this
in other places to fix some other problems, so move to a helper
in glsl type and fix the one user so far.
A dual slot double is one that has 3 or 4 components in it's
base type.
Signed-off-by: Dave Airlie <airlied@redhat.com>
Reviewed-by: Oded Gabbay <oded.gabbay@gmail.com>
Reviewed-by: Timothy Arceri <timothy.arceri@collabora.com>
As in the previous patches, these can be implemented as
any(v) -> any_nequal(v, false)
all(v) -> all_equal(v, true)
and their removal simplifies the code in the next patch.
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
The GLSL IR to TGSI/Mesa IR paths for any_nequal have the same
optimizations the ir_unop_any paths had.
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
NaNs mean it should be clipped, otherwise the NaNs might get passed to the
next stages (if clipping didn't happen for another reason already), which
might cause all kind of problems.
The llvm path got this right already (possibly by luck), but this isn't used
when there's a gs active.
Found by code inspection, verified with some hacked piglit test and some more
hacked debug output.
(Note the clipper can still itself incorrectly generate NaN and INF position
values in its output prims (at least after w divide / viewport transform) even
if the inputs weren't NaNs, if the position data of the vertices is
"sufficiently bad".)
Reviewed-by: Brian Paul <brianp@vmware.com>
Those stages only really work for OGL-style texturing (so number of samplers
and views mostly the same, certainly for the max values).
These get often set up all at once, thus there might be max number of both
even if all of them are just NULL. We must not set the max number of samplers
and views to the same value since that will lead to terrible things if a driver
supports more views than samplers (and the state tracker set up all the views).
(This will not make these stages magically work if a shader uses dx10-style
texturing, they might still replace an actually used sview in that case.)
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
This patch modifies the SSE4.1 test in configure.ac to use a global
variable to initialize vector variables. In addition, we now return the
value of the computation instead of 0.
This is done so gcc 4.9 (and lower) won't optimize the SSE4.1 assembly
instructions (when using -O1 and higher), because then the configure test
might incorrectly pass even though the assembler doesn't support the
SSE4.1 instructions (the test will pass because the compiler does support the intrinsics).
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=91806
Cc: "11.0 11.1" <mesa-stable@lists.freedesktop.org>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
OpenGLES 3.1 cannot be enabled for gen 7 (Ivy Bridge, Haswell) since
they are still missing ARB_stencil_texturing.
Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Marta Lofstedt <marta.lofstedt@intel.com>
Previously we were checking the desktop OpenGL ARB_compute_shader
requirements, but for OpenGLES 3.1, the requirements are lower.
Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Marta Lofstedt <marta.lofstedt@intel.com>
The OpenGL ARB_compute_shader extension specfication requires at least
1024 for GL_MAX_COMPUTE_WORK_GROUP_INVOCATIONS, whereas OpenGLES 3.1
only required 128.
Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
The maximum number of resident warps per multiprocessor is 64 on
Kepler instead of 48 on Fermi.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
It's nonsense to drain the pipeline like this.
v2: keep the drain for DMA-buf exports.
v3: flush before the export and after compositing and add TODO comment.
Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Julien Isorce <j.isorce@samsung.com>
Tested-by: Julien Isorce <j.isorce@samsung.com>
This reverts commit 839793680f.
The patch was breaking DRI3 because driGLFormatToImageFormat does not
handle MESA_FORMAT_B8G8R8X8_SRGB which ended up making it fail to
create the renderbuffer and it would later crash. It's not trivial to
add this format because there is no __DRI_IMAGE_FORMAT nor
__DRI_IMAGE_FOURCC define for the format either. I'm not sure how
difficult adding this would be and whether adding a new format would
require some sort of new version for DRI. Seeing as this might take a
while to fix I think it makes sense to just revert the patch in the
meantime in order to avoid regressing master.
It is also not handled in intel_gles3_srgb_workaround and there may be
other cases where it breaks.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=93388
Acked-by: Jason Ekstrand <jason.ekstrand@intel.com>
Previously the GL spec required that whenever glBlitFramebuffer is
used with either buffer being multisampled, the internal formats must
match. However the GL 4.4 spec was later changed to remove this
restriction. In the section entitled “Changes in the released
Specification of July 22, 2013” it says:
“Relax BlitFramebuffer in section 18.3.1 so that format conversion can
take place during multisample blits, since drivers already allow this
and some apps depend on it.”
If most drivers already allowed this in earlier versions I think it's
safe to assume that this is a spec bug and it should also be allowed
in all versions.
This patch just removes the restriction on desktop GL. For GLES there
are conformance tests that assert the previous behaviour so it is
probably safer to leave it in.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=92706
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
"image" is not ready yet since it will be set at
the end of the function by: *image = *img;
Signed-off-by: Julien Isorce <j.isorce@samsung.com>
Reviewed-by: Christian K<C3><B6>nig <christian.koenig@amd.com>
We just ignored them altogether. While this feature is rather old-fashioned
supporting it is actually rather trivial.
This fixes the associated piglit tests (2 gl-1.0-edgeflag, 2 gl-2.0-edgeflag
and all (7) of point-vertex-id).
v2: comment fixes, and make the use of the edgeflag in clipmask consistent
with when it's actually there (should be impossible to hit a case where the
difference would actually matter but still...)
Reviewed-by: Brian Paul <brianp@vmware.com>
This just adds confusion, these parameters are used when fetching vertices
by translate, but certainly not when emitting hw vertices for drivers, they
make no sense there (setting them has no consequences otherwise since there
won't be any elements with instance_divisor set). So just set them to 0 (the
draw_pipe_vbuf code for emitting vertices when the draw pipeline is run
already does exactly that).
Also while here do some whitespace cleanup.
Reviewed-by: Brian Paul <brianp@vmware.com>
Now that we have a helper in the builder for system values and a helper in
core NIR to get the intrinsic opcode, there's really no point in having
things split out into a helper function. This commit "modernizes" this
pass to use helpers better and look more like newer passes.
Reviewed-by: Eric Anholt <eric@anholt.net>
The VC4_DEBUG=cl,qpu is nice and all, but I want to be able to get more
detailed dumps, and to replay the same exact commands in simulation. For
that I need a dump with all of the VBOs, shaders, shader recs, etc. This
dump can be parsed by vc4-gpu-tools.
For now this is only doable from simulator mode, because otherwise we
don't have access to the RCL contents generated by the kernel.
Any update here should have been the same as in
vc4_set_framebuffer_state(), except for the point where vc4_blit.c
temporarily sets different state for its different buffers.
This is apparently a weirdness of gallium -- nr_samples==1 is occasionally
used and means the same thing as nr_samples==0. Fixes a bunch of
ARB_framebuffer_srgb blit cases in piglit.
It's really harsh to abort() the X Server because of a momentary failure
(particularly -ENOMEM). I don't see a way to pass an -ENOMEM up the stack
from here, but we can at least log to stderr before proceeding on.
Cc: "11.1" <mesa-stable@lists.freedesktop.org>
Fixes the following Lintian (Debian package checker) error:
privacy-breach-logo
usr/share/doc/mesa-common-dev/contents.html
(http://sourceforge.net/sflogo.php?group_id=3&type=1)
usr/share/doc/mesa-common-dev/thanks.html
(http://sourceforge.net/sflogo.php?group_id=3&type=1)
The extended description of this tag is:
This package creates a potential privacy breach by fetching a logo
at runtime.
Before using a local copy you should check that the logo is suitable
for main. You can get help with determining this by posting a link to
the logo and a copy of, or a link to, the logo copyright and license
information to the debian-legal mailing list.
Please replace any scripts, images, or other remote resources with
non-remote resources. It is preferable to replace them with text and
links but local copies of the remote resources are also acceptable as
long as they don't also make calls to remote services. Please ensure
that the remote resources are suitable for Debian main before making
local copies of them.
Severity: serious, Certainty: possible
Check: files, Type: binary, udeb
Signed-off-by: Andreas Boll <andreas.boll.dev@gmail.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
This test is a left-over of the initial development. It is unneeded and
misleading, so let's get rid of it.
Reviewed-by: Edward O'Callaghan <eocallaghan@alterapraxis.com>
It seems like disabling earlyz on a4xx also, by defaults, disables
fragcoord.z to the FS. For frag shaders that both read fragcoord(.z)
and write fragdepth, we need to set some extra bits to prevent a
lockup.
This lets us get rid of the hack of disabling fragcoord.z (which
prevented 0ad from lockups, but resulted in rendering corruption). Also
fixes fbo-depth-sample-compare.
Signed-off-by: Rob Clark <robclark@freedesktop.org>
The picture id in this case is a VA-API surface handle, checking
for a certain value can't be correct.
Signed-off-by: Christian König <christian.koenig@amd.com>
v2: (by Ken, incorporating feedback from Matt Turner):
- Rewrite the push constant allocation code to be clearer.
- Only apply the minimum VS entries workaround on Gen 8.
v3: (by Ken)
- Fix a bug in v2 where we failed to allocate the full push constant
space when the number of enabled stages didn't divide the available
push constant space evenly. (Any left over space is now allocated
to the PS, as it was in v1.)
- Fix an off-by-one error in v2's number of enabled stages calculation.
- Use DIV_ROUND_UP for nicer formatting.
- Line wrapping fixes.
Signed-off-by: Chris Forbes <chrisf@ijw.co.nz>
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
This change also adds explicit location support for structs and interfaces which
is currently missing in Mesa but is allowed with SSO and GLSL 1.50+.
Reviewed-by: Edward O'Callaghan <eocallaghan@alterapraxis.com>
This makes the code easier to follow, should be more efficient
and will makes it easier to add matching via explicit locations
in the following patch.
This patch also replaces the hash table with the newer
resizable hash table this should be more suitable as the table
is likely to only contain a small number of entries.
Reviewed-by: Edward O'Callaghan <eocallaghan@alterapraxis.com>
vertex header had both clip_pos and clip_vertex.
We only really need one (clip_pos) because the draw llvm shader would
overwrite the position output from the vs with the viewport transformed.
However, we don't really need the second one, which was only really used
for gl_ClipVertex - if the shader didn't have that the values were just
duplicated to both clip_pos and clip_vertex. So, just use this from the vs
output instead when we actually need it.
Also change clip debug to output both the data from clip_pos and the
clipVertex output (if available).
Makes some things more complex, some things less complex, but seems more
easy to understand what clipping actually does (and what values it uses
to do its magic).
Reviewed-by: Brian Paul <brianp@vmware.com
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
Seems obvious now this should use the data from position and not clip_vertex
(albeit might not really make a difference).
Reviewed-by: Brian Paul <brianp@vmware.com
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
clip -> clip_vertex and pre_clip_pos -> clip_pos.
Looks more obvious to me what these values actually represent (so use
something resembling the vs output names).
Reviewed-by: Brian Paul <brianp@vmware.com
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
This is just for code cleanup, conceptually the have_clipdist really
isn't per-vertex state, so don't put it there (just dependent on the
shader). Even though there wasn't really any overhead associated with
this, we shouldn't store random shader information in the vertex header.
Reviewed-by: Brian Paul <brianp@vmware.com
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
I'm pretty sure this should use position (i.e. pre_clip_pos) and not
the output from clipVertex. Albeit piglit doesn't care. It is what we
use in the clip test, and it is what every other driver does (as they
don't even have clipVertex output and lower the additional planes to
clip distances).
Reviewed-by: Brian Paul <brianp@vmware.com
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
This is a newer convention, which we prefer over ALIGN(x, n) / n.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
The compact VUE map only works when varying packing is in use.
Unfortunately, varying packing is disabled for TCS inputs.
This is needed to fix Piglit's tcs-input-read-array-interface test.
v2: Make lines fit in 80 columns (caught by Jordan Justen).
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
TCS outputs and TES inputs both refer to a common "patch URB entry"
shared across all invocations. First, there are some number of
per-patch entries. Then, there are per-vertex entries accessed via
an offset for the variable and a stride times the vertex index.
Because these calculations need to be done in both the vec4 and scalar
backends, it's simpler to just compute the offset calculations in NIR.
It doesn't necessarily make much sense to use per-vertex intrinsics
afterwards, but that at least means we don't lose the per-patch vs.
per-vertex information.
v2: Use is_input/is_output helpers (suggested by Jordan Justen).
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
TES outputs work exactly like VS outputs, so we can simply add a case
statement for those.
TCS inputs are very similar to geometry shaders - they're arrays of
per-vertex data. We use the same method I used for the scalar GS
backend.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Without varying packing, if a VS writes a compound variable, and the GS
only reads part of it, the base location of the variable may not
actually be in the VUE map.
To cope with this, we do lowering in terms of varying slots, add any
constant offsets to the base, and then do the VUE map remapping. This
ensures we only look up VUE map entries for slots which actually exist.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
My tessellation branch has two additional remap functions. I don't want
to replicate this logic there.
v2: Handle inputs/outputs separately (suggested by Jason Ekstrand).
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
Shared variables and input reworks landed around the same time.
Presumably, this was some sort of mistake in rebase conflict resolution.
This really only affects the num_indices field in nir_intrinsic_infos,
which is rarely used. However, it's used by the printer.
Found by inspection.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
GL_DRAW_FRAMEBUFFER does not exist in OpenGL ES 1.x, and since
_mesa_meta_begin hasn't been called yet, we have to work-around API
difficulties. The whole reason that GL_DRAW_FRAMEBUFFER is used instead
of GL_FRAMEBUFFER is that the read framebuffer may be different. This
is moot in OpenGL ES 1.x.
I have another patch series that would also fix this (by removing the
calls to _mesa_BindFramebuffer and friends), but it's not quite ready
yet... and I think it may be a bit heavy for some stable branches.
Consider this a stop-gap fix.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=93215
Cc: "11.0 11.1" <mesa-stable@lists.freedesktop.org>
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
Remove unused variables from clear_state and use a hardcoded location
for color uniform to get rid of 2 more variables. Modify shaders to use
explicit location for vertex attribute too as extension is enabled.
Signed-off-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
GRID Autosport uses SSO shaders. When a tessellation evaluation shader
is passed through this, it triggers assertion failures down the line
with unassigned varying locations. Make sure to do this when the first
shader in the pipeline is not a vertex shader.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Timothy Arceri <timothy.arceri@collabora.com>
Cc: "11.0 11.1" <mesa-stable@lists.freedesktop.org>
Previously if the visual didn't have an alpha channel then it would
pick a format that is not sRGB-capable. I don't think there's any
reason not to always have an sRGB-capable visual. Since 28090b30 there
are now visuals advertised without an alpha channel which means that
games that don't request alpha bits in the config would end up without
an sRGB-capable visual. This was breaking supertuxkart which assumes
the winsys buffer is always sRGB-capable.
The previous code always used an RGBA format if the visual config
itself was marked as sRGB-capable regardless of whether the visual has
alpha bits. I think we don't actually advertise any sRGB-capable
visuals (but we just use sRGB formats anyway) so it shouldn't make any
difference. However this patch also changes it to use RGBX if an
sRGB-capable visual is requested without alpha bits for consistency.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=92759
Cc: "11.0 11.1" <mesa-stable@lists.freedesktop.org>
Cc: Ilia Mirkin <imirkin@alum.mit.edu>
Suggested-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
brw_init_surface_formats overrides the render format for RGBX formats
which aren't supported for rendering so that they internally use RGBA
instead. However, B8G8R8X8_SRGB was missing so it wasn't marked as a
renderable format. This patch just adds it.
Cc: "11.0 11.1" <mesa-stable@lists.freedesktop.org>
Cc: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
With the current algorithm, we only look at tex uses. However there's a
write-after-write hazard where we might decide to, on some path, not use
a texture's output at all, but instead to write a different value to
that register. However without the barrier, the texture might complete
later and overwrite that value.
This fixes Unreal Elemental demo on GK110/GK208, flightgear on GK10x,
and likely other random-looking failures.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Cc: "11.1" <mesa-stable@lists.freedesktop.org>
In some cases shaders want non-default rounding when converting float to
integer. This can be done in one go, so merge the two ops. This comes up
in the packUnorm4x8 & co functions, as well as a few random shaders.
Overall shader-db impact is minimal, helping a handful of witcher2 and
other misc shaders.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
The conversion of 32-bit integer multiplies into 16-bit ones happens
after the regular optimization loop. However it's fairly common to
multiply by a small integer, rendering some of the expansion pointless.
Firstly, propagate immediates when possible into mul ops, secondly just
remove the ops when they are unnecessary.
Including the change to generate imad immediates, the effect is:
total instructions in shared programs : 6365463 -> 6351898 (-0.21%)
total gprs used in shared programs : 728684 -> 728684 (0.00%)
total local used in shared programs : 9904 -> 9904 (0.00%)
total bytes used in shared programs : 44001576 -> 44036120 (0.08%)
local gpr inst bytes
helped 0 0 3288 4
hurt 0 0 0 842
It's easy for this to hurt bytes since we end up always generating the
8-byte form, while we can't always get rid of the immediate in question.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
There will usually be a split before the mad op, peer through that and
pick out the right word of the immediate.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Support emission of the short imad, but also include it in the various
logic that tries to make it possible to emit.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
On NV50, we use 16-bit reg units (to make it all work with half-regs). A
few places assumed that it was always in 32-bit units.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
This helps in the use of GALLIUM_DDEBUG_SKIP: first run a target application
with skip set to a very large number and note how many draw calls happen
before the bug. Then re-run, skipping the corresponding number of calls.
Despite the additional run, this can still be much faster than not skipping
anything.
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
When we know that hangs occur only very late in a reproducible run (e.g.
apitrace), we can save a lot of debugging time by skipping the flush and hang
detection for earlier draw calls.
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
ARB_fragment_layer_viewport requires that if a fs reads layer or viewport
index but it wasn't output by gs (or vs with other extensions), then it reads
0. This never worked for llvmpipe, and is surprisingly non-trivial to fix.
The problem is the mechanism to handle non-existing outputs in draw is rather
crude, it will simply redirect them to whatever is at output 0, thus later
stages will just get garbage. So, rather than trying to fix this up (which
looks non-trivial), fix this up in llvmpipe setup by detecting this case there
and output a fixed zero directly.
While here, also optimize the hw vertex layout a bit - previously if the gs
outputted layer (or vp) and the fs read those inputs, we'd add them twice
to the vertex layout, which is unnecessary.
And do some minor cleanup, slots don't require that many bits, there was some
bogus (but harmless) float/int mixup for psize slot too, make the slots all
unsigned (we always put pos at pos zero thus everything else has to be positive
if it exists), and make sure they are properly initialized (layer and vp index
slot were not which looked fishy as they might not have got set back to zero
when changing from a gs which outputs them to one which does not).
This fixes the failures in piglit's arb_fragment_layer_viewport group
(3 each for layer and vp).
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
This greatly reduces the number of SetSamplers() commands for some
applications.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
Reviewed-by: Charmaine Lee <charmainel@vmware.com>
The offset for loads is in src[0]. This was a copy+paste error in the
nir_intrinsic_load/store refactoring. This commit fixes a segfault in
ES31-CTS.compute_shader.work-group-size. I have no idea how piglit failed
to catch this...
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=93348
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
This is brw_gs_surface_state.c copy and pasted twice with search and
replace.
brw_binding_table.c code is similarly copy and pasted.
v2: Drop dword_pitch related fields.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Acked-by: Jason Ekstrand <jason.ekstrand@intel.com>
Tessellation evaluation shaders work almost identically to vertex
shaders - we have a set of URB writes at the end of the program, and the
last one should terminate it.
Geometry shaders really are the special case, where multiple
EmitVertex() calls trigger URB writes in the middle of the program.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
Tessellation evaluation shaders will use g4 instead. For now, make an
fs_reg called urb_handle and use that in place of hardcoding g1.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
This is a helper function for setting up the local invocation ID
payload according to the cs_prog_data generated by the compiler. It's
intended to be available to users of libi965_compiler so move it there.
GL likes to saturate your incoming color, but if that color's coming from
unpacking from unorms, there's no point. Ideally we'd have a range
propagation pass that cleans these up in NIR, but that doesn't seem to be
going to land soon. It seems like we could do a one-off optimization in
nir_opt_algebraic, except that doesn't want to operate on expressions
involving unpack_unorm_4x8, since it's sized.
total instructions in shared programs: 87879 -> 87761 (-0.13%)
instructions in affected programs: 6044 -> 5926 (-1.95%)
total estimated cycles in shared programs: 349457 -> 349252 (-0.06%)
estimated cycles in affected programs: 6172 -> 5967 (-3.32%)
No SSPD on openarena (which had the biggest gains, in its VS/CSes), n=15.
The caller isn't going to expect it from a return, so it would probably
get misinterpreted. If the caller had an unpack in its reg, that's fine,
but don't lose track of it.
I apparently broke this in a late refactor, in such a way that I decided
its tests were some of those interminable ones that I should just
blacklist from my testing. As a result, the refactors related to it were
totally wrong.
Mostly related to making sure the rasterizer can correctly
pick out the correct scissor box for the current viewport.
Signed-off-by: Edward O'Callaghan <eocallaghan@alterapraxis.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
When GL_FRAMEBUFFER_SRGB is enabled any single-sampled renderbuffers
are resolved in intel_update_state because the hardware can't cope
with fast clears on SRGB buffers. In that case it's pointless to do a
fast clear because it will just be immediately resolved.
Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
SRGB buffers are not marked as losslessly compressible so previously
they would not be used for fast clears. However in practice the
hardware will never actually see that we are using SRGB buffers for
fast clears if we use the linear equivalent format when clearing and
make sure to resolve the buffer as a linear format before sampling
from it.
This is an important use case because by default the window system
framebuffers are created as SRGB so without this fast clears won't be
used there.
Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
SKL can't cope with the CCS buffer for SRGB buffers. Normally the
hardware won't see the SRGB formats because when GL_FRAMEBUFFER_SRGB
is disabled these get mapped to their linear equivalents. In order to
avoid relying on the CCS buffer when it is enabled this patch now
makes it flush the renderbuffers.
Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
For single-sampled textures the MCS buffer is only used to implement
fast clears. However the surface always needs to be resolved before
being used as a texture anyway so the the MCS buffer doesn't actually
achieve anything. This is important for Gen9 because in that case SRGB
surfaces are not supported for fast clears and we don't want the
hardware to see the MCS buffer in that case.
Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
Adds MESA_META_FRAMEBUFFER_SRGB to the meta save state so that
GL_FRAMEBUFFER_SRGB will be disabled when performing the fast clear.
That way the render surface state will be programmed with the linear
equivalent format during the clear. This is important for Gen9 because
the SRGB formats are not marked as losslessly compressible so in
theory they aren't support for fast clears. It shouldn't make any
difference whether GL_FRAMEBUFFER_SRGB is enabled for the fast clear
operation because the color is not actually written to the framebuffer
so there is no chance for the hardware to apply the SRGB conversion on
it anyway.
Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
This simplified (basically duplicated) version of pb_cache_manager will
allow removing some ugly hacks from radeon and amdgpu winsyses and
flatten simplify their design.
The difference is that winsyses must manually add buffers to the cache
in "destroy" functions and the cache doesn't know about the buffers before
that. The integration is therefore trivial and the impact on the winsys
design is negligible.
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Reviewed-by: Edward O'Callaghan <eocallaghan@alterapraxis.com>
Acked-by: Michel Dänzer <michel.daenzer@amd.com>
This is the recommended setting according to hw people and it makes Hyper-Z
stable. Just the two magic states.
This fixes Evergreen, Cayman, SI, CI, VI (using the Cayman code).
Cc: 11.0 11.1 <mesa-stable@lists.freedesktop.org>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
This reverts commit 32f05fadbb.
It turned out the problem with Stoney was caused by incorrect handling of
a non-power-two VRAM size in the kernel driver.
This is an optional BIOS setting and can be worked around by choosing
a different VRAM size in the BIOS.
Cc: 11.1 <mesa-stable@lists.freedesktop.org>
If we have a dmat2[4], then dmat2[0] is at 17, dmat2[1] at 19,
dmat2[2] at 21 etc. The old code was returning 17,18,19.
I think this code is also wrong for float matricies as well.
There is now a piglit for the float case.
This partly fixes:
GL41-CTS.vertex_attrib_64bit.limits_test
[airlied: update with Tapani suggestion to clean it up].
Cc: "11.0 11.1" <mesa-stable@lists.freedesktop.org>
Reviewed-by: Timothy Arceri <timothy.arceri@collabora.com>
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
Discovered this when working on other clip code, apparently didn't work
correctly - the combination of linear interpolated values and using
gl_ClipVertex produced wrong values (failing all such combinations
in piglits glsl-1.30 interpolation tests, named
interpolation-noperspective-XXX-vertex).
Use the pre-clip-pos values when determining the interpolation factor to
fix this.
Noone really understands this code well, but everybody agrees this looks
sane... This fixes all those failing tests (10 in total) both with
the llvm and non-llvm draw paths, with no piglit regressions.
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Dave Airlie <airlied@redhat.com>
There is some special-casing needed in a competent back-end. However, they
can do their special-casing easily enough based on whether or not the
offset is a constant. In the mean time, having the *_indirect variants
adds special cases a number of places where they don't need to be and, in
general, only complicates things. To complicate matters, NIR had no way to
convdert an indirect load/store to a direct one in the case that the
indirect was a constant so we would still not really get what the back-ends
wanted. The best solution seems to be to get rid of the *_indirect
variants entirely.
This commit is a bunch of different changes squashed together:
- nir: Get rid of *_indirect variants of input/output load/store intrinsics
- nir/glsl: Stop handling UBO/SSBO load/stores differently depending on indirect
- nir/lower_io: Get rid of load/store_foo_indirect
- i965/fs: Get rid of load/store_foo_indirect
- i965/vec4: Get rid of load/store_foo_indirect
- tgsi_to_nir: Get rid of load/store_foo_indirect
- ir3/nir: Use the new unified io intrinsics
- vc4: Do all uniform loads with byte offsets
- vc4/nir: Use the new unified io intrinsics
- vc4: Fix load_user_clip_plane crash
- vc4: add missing src for store outputs
- vc4: Fix state uniforms
- nir/lower_clip: Update to the new load/store intrinsics
- nir/lower_two_sided_color: Update to the new load intrinsic
NIR and i965 changes are
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
NIR indirect declarations and vc4 changes are
Reviewed-by: Eric Anholt <eric@anholt.net>
ir3 changes are
Reviewed-by: Rob Clark <robdclark@gmail.com>
NIR changes are
Acked-by: Rob Clark <robdclark@gmail.com>
There was way too much incrementing of things going on. Instead, let's
just start everything off at the right base location, and then increment in
the loop.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
In case a state tracker unbinds every slot by a seperate
pipe->set_vertex_buffers() call, starting from slot zero, the number
of bound buffers would not reach zero at all.
The current algorithm does not account for pre-existing holes in the
buffer list.
Unbinding all buffers at once or starting at the top-most slot results
in correct behaviour.
Calculating the correct number of bound buffers fixes a NULL pointer
dereference in nvc0_validate_vertex_buffers_shared().
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=93004
Signed-off-by: Patrick Rudolph <siro@das-labor.org>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
Cc: "11.0 11.1" <mesa-stable@lists.freedesktop.org>
According to the GLES3 spec, blitting between multisample FBOs with
different internal formats should not be allowed. The
compatible_resolve_formats function implements this check. Previously
it had a shortcut where if the Mesa formats of the two renderbuffers
were the same then it would assume the blit is ok. However some
drivers map different internal formats to the same Mesa format, for
example it might implement both GL_RGB and GL_RGBA textures with
MESA_FORMAT_R8G8B8A_UNORM. The function is used to generate a GL error
according to what the GL spec requires so the blit should not be
allowed in that case. This patch just removes the shortcut so that it
only ever looks at the internal format.
Note that I posted a related patch to disable this check altogether
for desktop GL. However this function is still used on GLES3 because
there are conformance tests that require this behaviour so this patch
is still useful.
Cc: Marek Olšák <maraeo@gmail.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
The tiled memcpy doesn't work for copying from RGBX to RGBA because it
doesn't override the alpha component to 1.0. Commit 2cebaac479 added
a check to disable it for RGBX formats by looking at the TexFormat.
However a lot of the rest of the code base is written with the
assumption that an RGBA texture can be used internally to implement a
GL_RGB texture. If that is done then this check breaks. This patch
makes it instead check the base format of the texture which I think
more directly matches the intention.
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
Since Gen8 this is allowed as a rendering target so we don't need to
override it to B8G8R8A8. This is helpful on Gen9+ where using this
override causes fast clears not to work.
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Ben Widawsky <benjamin.widawsky@intel.com>
Previously fast clear was disallowed on Gen9 for MSRTs with the claim
that some formats don't work but we didn't understand why. On further
investigation it seems the formats that don't work are the ones where
the render surface format is being overriden to a different format
than the one used for texturing. The one used for texturing is not
actually a renderable format. It arguably makes sense that the sampler
hardware doesn't handle the fast color correctly in these cases
because it shouldn't be possible to end up with a fast cleared surface
that is non-renderable.
This patch changes the limitation to prevent fast clear for surfaces
where the format for rendering is overriden.
Reviewed-by: Ben Widawsky <benjamin.widawsky@intel.com>
If GL_FRAMEBUFFER_SRGB is enabled when writing to an SRGB-capable
framebuffer then the color will be converted from linear to SRGB
before being written. There is no chance for the hardware to do this
itself because it can't modify the clear color that is programmed in
the surface state so it seems pretty clear that the driver should be
handling this itself.
Note that this wasn't a problem before Gen9 because previously we were
only able to do fast clears to 0 or 1 and those values are the same in
linear and SRGB space.
Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
Enable ARB_compute_shader on gen7+, on hardware that supports the
OpenGL 4.3 requirements of a local group size of 1024.
With SIMD16 support, this is limited to Ivy Bridge and Haswell.
Broadwell will work with a local group size up to 896 on SIMD16
meaning programs that use this size or lower should run when setting
MESA_EXTENSION_OVERRIDE=GL_ARB_compute_shader.
Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
Shared variables can be accessed by other threads within the same
local workgroup. This prevents us from performing certain
optimizations with shared variables.
Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
When an intrinsic atomic operation is used on a shared variable, we
translate it to a new 'shared variable' specific intrinsic function
call.
For example, a call to __intrinsic_atomic_add when used on a shared
variable will be translated to a call to
__intrinsic_atomic_add_shared.
v3:
* Fix stale comments copied from SSBOs (Iago)
Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
The compiler probably already blocks this earlier on, but we should be
checking for an SSBO here.
Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
When an atomic function is called, we need to check to see if it is
for an SSBO variable before lowering it to the SSBO specific intrinsic
function.
v2:
* is_in_buffer_block => is_in_shader_storage_block (Iago)
Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
The atomic functions can also be used with shared variables in compute
shaders.
When lowering the intrinsic in lower_ubo_reference, we still create an
SSBO specific intrinsic since SSBO accesses can be indirectly
addressed, whereas all compute shader shared variable live in a single
shared variable area.
v2:
* Also remove the _internal suffix from ssbo atomic intrinsic names (Iago)
Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
In this lowering pass, shared variables are decomposed into intrinsic
calls.
v2:
* Send mem_ctx as a parameter (Iago)
v3:
* Shared variables don't have an associated interface block (Iago)
* Always use 430 packing (Iago)
* Comment / whitespace cleanup (Iago)
Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
This code will also be usable by the pass to lower shared variables.
Note, that *const_offset is adjusted by setup_buffer_access so it must
be initialized before calling setup_buffer_access.
v2:
* Add comment for lower_buffer_access::setup_buffer_access
Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
This class has code that will be shared by lower_ubo_reference and
lower_shared_reference. (lower_shared_reference will be used to
support compute shader shared variables.)
v2:
* Add lower_buffer_access.h to makefile (Emil)
* Remove static is_dereferenced_thing_row_major from
lower_buffer_access.cpp. This will become a lower_buffer_access
method in the next commit.
* Pass mem_ctx as parameter rather than using a member variable (Iago)
Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
This allows the code in emit_access to be generic enough to also be
for lowering shared variables.
Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
v2:
* Rename ssbo_get_array_length to ssbo_unsized_array_length_access (Iago)
* Use always use this-> when referencing buffer_access_type (Iago)
Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
Otherwise packed and inactive varyings get optimized away. This needs
to be prevented when using separate shader objects where interface
needs to be preserved.
Signed-off-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Timothy Arceri <timothy.arceri@collabora.com>
This will cause validation to run during next draw, this is done
because possible changes in used stages and programs can cause
invalid pipeline state.
This fixes a subtest in following CTS test:
ES31-CTS.sepshaderobjs.StateInteraction
Signed-off-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Timothy Arceri <timothy.arceri@collabora.com>
Doesn't have any effect in practice I don't think, but
CTS reads back using GetVertexAttrib.
This fixes: GL41-CTS.vertex_attrib_64bit.get_vertex_attrib
Cc: "11.0 11.1" <mesa-stable@lists.freedesktop.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
src/gallium/tests/trivial/compute.c expects samplers to be cleaned
when the samplers list is NULL.
Like in radeon, the function behave like when the number of samplers
parameter is set to 0.
[small s/hwsco/hwcso/ typo fix]
Signed-off-by: Rob Clark <robclark@freedesktop.org>
Let us avoid trapping in hardware from a SIGFPE and instead
assert on a zero divisor.
Hint: This can occur if a PIPE_PRIM_? is not handled in
u_prim_vertex_count() that results in ' info ' not
being initialized in the expected manner.
Further, we also fix a possibly NULL pointer dereference
from ' info ' being NULL from a u_prim_vertex_count() call.
Signed-off-by: Edward O'Callaghan <eocallaghan@alterapraxis.com>
Signed-off-by: Marek Olšák <marek.olsak@amd.com>
When a buffer is created with GL_STATIC_DRAW, its contents should not
be changed frequently. But that's exactly what one application I'm
debugging does. This patch adds code to try to detect inefficient
buffer use in a couple places. The GL_ARB_debug_output mechanism is
used to report the issue.
NVIDIA's driver detects these sort of things too.
Other types of inefficient buffer use could also be detected in the
future.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
Every other gen the representation of the URB size was changed and
previous ones weren't updated. I'd be willing to write a series
normalizing this to be KB on all generations if anybody else cares.
This is going to require some rather intrusive kernel changes to fix
properly, in the meantime (and forever on at least pre-v4.1 kernels)
we'll have to restore the hardware defaults at the end of every batch
in which the L3 configuration was changed to avoid interfering with
the DDX and GL clients that use an older non-L3-aware version of Mesa.
Reviewed-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
v2: Optimize look-up of the default configuration by assuming it's the
first entry of the L3 config array in order to avoid an FPS
regression in GpuTest Triangle and SynMark OglBatch2-7 on most
affected platforms.
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
The L3 state atom calculates the target L3 partition weights when the
program bound to some shader stage is modified, and in case they are
far enough from the current partitioning it makes sure that the L3
state is re-emitted.
v2: Fix for inconsistent units the context URB size is expressed in.
Clamp URB size to 1008 KB on SKL due to FF hardware limitation.
Reviewed-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
Acked-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
This calculates a rather conservative partitioning of the L3 cache
based on the shaders currently bound to the pipeline and whether they
use SLM, atomics, images or scratch space. The result is intended to
be fine-tuned later on based on other pipeline state.
Note that the L3 partitioning calculated for VLV in the non-SLM non-DC
case differs from the hardware defaults in that it doesn't include a
DC partition and has twice as much RO cache space -- This is an
intentional functional change that improves performance in several
bandwidth-bound benchmarks on VLV (5% significance): SynMark
OglTexFilterAniso by 14.18%, SynMark OglTexFilterTri by 7.15%, Unigine
Heaven by 4.91%, SynMark OglShMapPcf by 2.15%, GpuTest Fur by 1.83%,
SynMark OglDrvRes by 1.80%, SynMark OglVsTangent by 1.71%, and a few
other benchmarks from the Finnish system by less than 1%.
Reviewed-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
The input of the L3 set-up code is a vector giving the approximate
desired relative size of each partition. This implements logic to
compare the input vector against the table of validated configurations
for the device and pick the closest compatible one.
Reviewed-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
Improves performance of the arb_shader_image_load_store-atomicity
piglit test by over 25x (which isn't a real benchmark it's just heavy
on atomics -- the improvement in a microbenchmark I wrote a while ago
seemed to be even greater). The drawback is one needs to be
extra-careful not to hang the GPU (in fact the whole system). A DC
partition must have been allocated on L3, the "convert L3 cycle for DC
to UC" bit may not be set, the MOCS L3 cacheability bit must be set
for all surfaces accessed using DC atomics, and the SCRATCH1 and
ROW_CHICKEN3 bits must be kept in sync.
A fairly recent kernel is required for the command parser to allow
writes to these registers.
Reviewed-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
Acked-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
It should be possible to use additional L3 configurations other than
the ones listed in the tables of validated allocations ("BSpec »
3D-Media-GPGPU Engine » L3 Cache and URB [IVB+] » L3 Cache and URB [*]
» L3 Allocation and Programming"), but it seems sensible for now to
hard-code the tables in order to stick to the hardware docs. Instead
of setting up the arbitrary L3 partitioning given as input, the
closest validated L3 configuration will be looked up in these tables
and used to program the hardware.
The included tables should work for Gen7-9. Note that the quantities
are specified in ways rather than in KB, this is because the L3
control registers expect the value in ways, and because by doing that
we can re-use a single table for all GT variants of the same
generation (and in the case of IVB/HSW and CHV/SKL across different
generations) which generally have different L3 way sizes but allow the
same combinations of way allocations.
v2: Use slice count from the devinfo structure instead of the gt
number to implement get_l3_way_size().
Reviewed-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
Acked-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
According to the hardware docs a DC flush is sufficient to make
CS_STALL happy, there's no need to add STALL_AT_SCOREBOARD whenever
it's present.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
This will make sure that we recalculate the URB layout anytime the URB
size is modified by the L3 partitioning code.
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
This stores the result of can_do_pipelined_register_writes() in the
context struct so we can find out later whether LRI can be used to
program the L3 configuration.
v2:
* Split change of gen check in can_do_pipelined_register_writes (jljusten)
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
Allow for pipelined register writes for gen < 7.
v2:
* Split from another patch and adjust comment (jljusten)
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
This fixes:
glsl-1.50/execution/geometry/dynamic_input_array_index.shader_test
my profanity.
We need to load the AR register with the value from the index reg
Cc: "11.0 11.1" <mesa-stable@lists.freedesktop.org>
Signed-off-by: Dave Airlie <airlied@redhat.com>
This fixes:
arb_transform_feedback3-ext_interleaved_two_bufs_gs
arb_transform_feedback3-ext_interleaved_two_bufs_gs_max
transform-feedback-builtins
If we are only emitting one ring, then emit all output
buffers on it.
Cc: "11.0 11.1" <mesa-stable@lists.freedesktop.org>
Signed-off-by: Dave Airlie <airlied@redhat.com>
See: `commit e82c527f1fc2f8ddc64954ecd06b0de3cea92e93`
which is where a block in src maps to a pixel in dst and vice versa.
e.g. DXT1 <-> R32G32_UINT
DXT5 <-> R32G32B32A32_UINT
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Signed-off-by: Edward O'Callaghan <eocallaghan@alterapraxis.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
The only effect here is a space savings - 822 programs in shader-db
affected with the following overall change:
total bytes used in shared programs : 44154976 -> 44139880 (-0.03%)
Fixes: 641eda0c (nv50/ir: r63 is only 0 if we are using less than 63 registers)
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Cc: "11.0 11.1" <mesa-stable@lists.freedesktop.org>
This allows us to use the short encoding, and potentially fold
immediates in later on.
total instructions in shared programs : 6379731 -> 6367861 (-0.19%)
total gprs used in shared programs : 728502 -> 728683 (0.02%)
total local used in shared programs : 9904 -> 9904 (0.00%)
total bytes used in shared programs : 44661008 -> 44154976 (-1.13%)
local gpr inst bytes
helped 0 51 7267 20306
hurt 0 232 125 274
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Operations that take immediates can only encode registers up to 64. This
fixes a shader in a "Powered by Unity" intro.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
We already semi-did this but the list of uses as unsorted, so it was
unreliable. Sort the uses by bb and serial, and don't unspill for each
instruction in a sequence. (And also don't unspill multiple times for a
single instruction that uses the value in question multiple times.)
This causes a minor reduction in generated instructions for shader-db
(as few programs spill) but more importantly it brings determinism to
each run's output.
On SM10:
total instructions in shared programs : 6387945 -> 6379359 (-0.13%)
total gprs used in shared programs : 728544 -> 728544 (0.00%)
total local used in shared programs : 9904 -> 9904 (0.00%)
local gpr inst bytes
helped 0 0 322 322
hurt 0 0 0 0
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
This fixes the fetching of fp64 inputs to the geometry shader,
this fixes the recently posted piglit's
arb_gpu_shader_fp64/execution/gs-fs-vs-double-array.shader_test
arb_vertex_attrib_64bit/execution/gs-fs-vs-attrib-double-array.shader_test
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com> [v1]
Reviewed-by: Eric Anholt <eric@anholt.net> [v1]
v2: Move new rule to Boolean simplification section
Add a a@bool != true simplification
Suggested-by: Neil Roberts <neil@linux.intel.com>
... and allow the "binding" qualifier in ES 3.1 as well.
GLSL ES 3.1 incorporates only a few features from the extension
ARB_shading_language_420pack: the relaxed qualifier ordering
requirements and the binding qualifier.
Cc: "11.1" <mesa-stable@lists.freedesktop.org>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Unforunately the Appveyor -> SourceForge connection seems a bit
unreliable, causing frequent build failures while downloading
winflexbison (approx once every 2 days).
Fetching winflexbison archive into Appveyor's cache should eliminate
these.
Fetching Python modules from PyPI doesn't seem to be a problem, so they
are left alone for now, though they could eventually get the same
treatment.
We still have several failures in the newly enabled tests in simulation:
sRGB downsampling is done as if it was just linear, stencil blits are not
supported on MSAA either, and derivatives are still not supported
(breaking some MSAA simulation shaders). So, other than sRGB downsampling
quality, things seem to be in good shape.
This is the core of ARB_texture_multisample. Most of the piglit tests for
GL_ARB_texture_multisample require GL 3.0, but exposing support for this
lets us use the gallium blitter for multisample resolves. We can
sometimes multisample resolve using just the RCL, but that requires that
the blit is 1:1, unflipped, and aligned to tile boundaries.
This includes GL_SAMPLE_COVERAGE, GL_SAMPLE_ALPHA_TO_ONE, and
GL_SAMPLE_ALPHA_TO_COVAGE.
I haven't implemented a dithering function yet, and gallium doesn't give
me a good chance to do so for GL_SAMPLE_COVERAGE.
I only stumbled on this while experimenting due to reading about HW-2905.
I don't know if the EZ disable in the Z-clear is actually necessary, but
go with it for now.
The recent unaligned fix successfully prevented RCL blits that weren't
aligned inside of the surface, but we also want to be able to do RCL blits
for the whole surface when the width or height of the surface aren't
aligned (we don't care what renders inside of the padding).
This commit pushes makes uniform offsets be terms of bytes starting with
nir_lower_io. They get converted to be in terms of vec4s or floats when we
cram them in the UNIFORM register file but reladdr remains in terms of
bytes all the way down to the point where we lower it to a pull constant
load.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
The one and only place where the FS backend allows reladdr is on uniforms.
For locals, inputs, and outputs, we lower it away before the backend ever
sees it. This commit gets rid of the dead indirect handling code.
Cc: "11.0" <mesa-stable@lists.freedesktop.org>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Previously, the VS_OPCODE_PULL_CONSTANT_LOAD opcode operated on
vec4-aligned byte offsets on Iron Lake and below and worked in terms of
vec4 offsets on Sandy Bridge. On Ivy Bridge, we add a new *LOAD_GEN7
variant which works in terms of vec4s. We're about to change the GEN7
version to work in terms of bytes, so this is a nice unification.
Cc: "11.0" <mesa-stable@lists.freedesktop.org>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
It is legal to have a texture view of a single layer from a 2D array texture;
you can sample from it, or render to it. Intel hardware needs to be made aware
when it is using a 2d array surface in the surface state. The texture view is
just a 2d surface with the backing miptree actually being a 2d array surface.
This caused the previous code would not set the right bit in the surface state
since it wasn't considered an array texture.
I spotted this early on in debug but brushed it off because it is clearly not
needed on other platforms (since they all pass). I have no idea how this works
properly on other platforms (I think gen7 introduced the bit in the state, but I
am too lazy to check). As such, I have opted not to modify gen7, though I
believe the current code is wrong there as well.
Thanks to Chris for helping me debug this.
v2: Just use the underlying mt's target type to make the array determination.
This replaces a bug in the first patch which was incorrectly relying only
on non-zero depth (not sure how that had no failures). (Ilia)
Cc: Chris Forbes <chrisf@ijw.co.nz>
Reported-by: Mark Janes <mark.a.janes@intel.com> (Jenkins)
References: https://www.opengl.org/registry/specs/ARB/texture_view.txt
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=92609
Signed-off-by: Ben Widawsky <benjamin.widawsky@intel.com>
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
For some reason this has been disabled for integers ever since codegen
was merged, despite there being emission code for IMAD. Seems to work.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
According to nvdisasm both the immediate and non-imm cases use the same
bits. Both of these flags are quite rarely set though.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Cc: "11.0 11.1" <mesa-stable@lists.freedesktop.org>
From the 3DSTATE_URB_DS documentation:
"Project: IVB, HSW
If Domain Shader Thread Dispatch is Enabled then the minimum number of
handles that must be allocated is 10 URB entries."
"Project: BDW+
If Domain Shader Thread Dispatch is Enabled then the minimum number of
handles that must be allocated is 34 URB entries."
When the HS is run in SINGLE_PATCH mode (the only mode we support
today), there is no minimum for HS - it's just zero.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
For now, this just splits the existing code to disable these stages into
separate atoms/files. We can then replace it with real code.
v2: Bump the render atoms in this patch so it compiles (in my branch,
I'd bumped it in an earlier patch). 61 seems to be the minimum
that works, which doesn't match the old value + the number of atoms
I added in this patch, so apparently we had some slop before.
v3: Actually disable the DS unit on Gen8+.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net> [v1]
Reviewed-by: Matt Turner <mattst88@gmail.com>
We actually leave the sampler unset for OP_TXF, which caused the GK104+
logic to treat some texel fetches as indirect. While this works, it's
incredibly wasteful. This only happened when the texture was > 0 (since
sampler remained == 0).
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Cc: "11.0 11.1" <mesa-stable@lists.freedesktop.org>
_mesa_is_array_texture provides the same functionality and:
1. it returns bool instead of GLboolean
2. it's not related to the texture format (texformat.c)
3. the name's a little shorter
v2: remove _mesa_tex_target_is_array instead (Brian Paul)
Reviewed-by: Brian Paul <brianp@vmware.com>
Both methods provide the same functionality, so one would be
removed.
v2: use _mesa_is_array_texture and not the other way (Brian Paul)
Reviewed-by: Brian Paul <brianp@vmware.com>
Use the new debug callback hook to report conformance, performance
and fallbacks to the state tracker. The state tracker, in turn can
report this issues to the user via the GL_ARB_debug_output extension.
More issues can be reported in the future; this is just a start.
v2: remove conditionals around pipe_debug_message() calls since the
check is now done in the macro itself.
v3: remove unneeded dummy %s substitutions
Acked-by: Ilia Mirkin <imirkin@alum.mit.edu>,
Reviewed-by: José Fonseca <jfonseca@vmware.com>
So the callers don't have to do it.
v2: also check cb!=NULL in the macro
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: José Fonseca <jfonseca@vmware.com>
This function is unfinished there is a bunch more validation rules
that need to be applied here. We will still want to call it for desktop
GL we just don't want to validate precision so move the ES check to
reflect this.
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
The validation api doesn't trigger this error so just move it to the
code called during rendering.
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
Cc: Kenneth Graunke <kenneth@whitecape.org>
At last on ARUBA this is required to stop tessellation hanging
in heaven.
This removes one of the SIMDs from use by the HS/LS.
Reviewed-by: Edward O'Callaghan <eocallaghan@alterapraxis.com>
Tested-by: Edward O'Callaghan <eocallaghan@alterapraxis.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
This enables tessellation for evergreen/cayman,
This will need changes before committing depending
on what hw works etc.
working are CAYMAN/REDWOOD/BARTS/TURKS/SUMO/CAICOS
v2: only enable on evergreen and above.
Reads from the queue shouldn't be merged for now read operations.
Reads from the queue shouldn't be merged for now, or put in
T slots.
Signed-off-by: Dave Airlie <airlied@redhat.com>
At least one SIMD must be kept away from the HS/LS
stages in order to avoid a hw issue on evergreen/cayman.
This patch implements this workaround.
Signed-off-by: Dave Airlie <airlied@redhat.com>
This adds handling for TESSINNER/TESSOUTER in the TES
where they need to be fetched from LDS,
and TESSCOORD which comes in via r0.
It also handle primitive ID and invocation ID.
Signed-off-by: Dave Airlie <airlied@redhat.com>
when tessellation is enabled the TES shader is responsible
for handling streamout and exports.
This adds the streamout and export workarounds to TES,
and also makes sure TES sets up spi_sid.
Signed-off-by: Dave Airlie <airlied@redhat.com>
When we are finished the shader, we read back all the tess factors
from LDS and write them to special global memory storage using
GDS instructions.
This also handles adding NOP when GDS or ENDLOOP end the TCS.
Signed-off-by: Dave Airlie <airlied@redhat.com>
TCS outputs whenever they are written in the shader,
need to be written to LDS not temporaries, this handles
this case. It also fixes up the case where the output
is a relative addressed output, so we don't try to apply
the relative address at the wrong time.
Signed-off-by: Dave Airlie <airlied@redhat.com>
This handles the logic for doing fetches from LDS for
TCS and TES. For TCS we need to fetch both inputs and outputs,
for TES only inputs need to be fetched.
v2: use 24-bit ops.
Signed-off-by: Dave Airlie <airlied@redhat.com>
This retrievs the offset into the LDS for a patch or
non-patch variable, it takes the RelPatch channel
and a temporary register.
Signed-off-by: Dave Airlie <airlied@redhat.com>
This function retrieves the tess input/output info
from the tess constant buffer that is bound to the shader.
This uses a vfetch to get the values into the shader.
Signed-off-by: Dave Airlie <airlied@redhat.com>
These utilities are to be used to do things like integer adds and
multiplies to be used in calculating the LDS offsets etc.
It handles CAYMAN MULLO differences as well.
Signed-off-by: Dave Airlie <airlied@redhat.com>
When using tessellation on eg/ni chipsets, we must disable
dynamic GPRs to workaround a hw bug where the GPU hangs
when too many things get queued.
This implements something like the r600 code to emit
the transition between static and dynamic GPRs, and to
statically allocate GPRs when tessellation is enabled.
Signed-off-by: Dave Airlie <airlied@redhat.com>
If we have no tess control shader, then we have to use a fallback
one that just writes the tessellation factors.
Signed-off-by: Dave Airlie <airlied@redhat.com>
This creates a constant buffer with the information about
the layout of the LDS memory that is given to the vertex, tess
control and tess evaluation shaders.
This also programs the LDS size and the LS_HS_CONFIG registers,
on evergreen only.
v2: calculate lds hs num waves properly (Marek)
Emit the state only when something has changed (airlied).
Signed-off-by: Dave Airlie <airlied@redhat.com>
This update the setting of the shader stages register
when tess is enabled and add the setting of the VGT_TF_PARAM
register from the tess shader properties.
Signed-off-by: Dave Airlie <airlied@redhat.com>
This initialises the tess min/max using fglrx values,
and also initialises a number of other registers related
to tessellation.
v1.1: caicos doesn't have some registers.
Signed-off-by: Dave Airlie <airlied@redhat.com>
This just adds printing for the hw shader types, and hooks it up.
Reviewed-by: Glenn Kennard <glenn.kennard@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
This just adds the LDS ops to the SB bytecode reader/writers.
Signed-off-by: Glenn Kennard <glenn.kennard@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
These are used in tessellation shaders to read/write values
between VS/TCS/TES.
This splits the eg alu assembler out to handle these
instructions.
Signed-off-by: Dave Airlie <airlied@redhat.com>
This just adds support to the decoder, not actual SB support.
v1.1: fixup GDS relative mode. (Glenn).
Reviewed-by: Glenn Kennard <glenn.kennard@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
This function is going to get a lot messier with tessellation
so I'm going to use some macros to try and clean some bits
of common code up.
Signed-off-by: Dave Airlie <airlied@redhat.com>
This moves to using an array of hw stages for the atoms.
Note this drops the 23 from the vertex shader, this value
is calculated internally when shaders are bound, so not
required here.
Signed-off-by: Dave Airlie <airlied@redhat.com>
This changes the r600 specific GPR adjustment code
to use the stage defines, and arrays.
This is prep work for the tess changes later.
Reviewed-by: Oded Gabbay <oded.gabbay@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
Add a list of defines for the HW stages.
We will use this for GPR calculations amongst other things.
Reviewed-by: Oded Gabbay <oded.gabbay@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
Use NULL tests of the form `if (ptr)' or `if (!ptr)'.
They do not depend on the definition of the symbol NULL.
Further, they provide the opportunity for the accidental
assignment, are clear and succinct.
Signed-off-by: Edward O'Callaghan <eocallaghan@alterapraxis.com>
Signed-off-by: Marek Olšák <marek.olsak@amd.com>
Use NULL tests of the form `if (ptr)' or `if (!ptr)'.
They do not depend on the definition of the symbol NULL.
Further, they provide the opportunity for the accidental
assignment, are clear and succinct.
Signed-off-by: Edward O'Callaghan <eocallaghan@alterapraxis.com>
Signed-off-by: Marek Olšák <marek.olsak@amd.com>
On SM20 this gives:
total instructions in shared programs : 6299222 -> 6294240 (-0.08%)
total gprs used in shared programs : 944139 -> 944068 (-0.01%)
total local used in shared programs : 54116 -> 54116 (0.00%)
local gpr inst bytes
helped 0 126 2781 2781
hurt 0 55 11 11
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
This way $r1 = $r0 + 4; c1[$r1] becomes c1[$r0+4].
On SM35:
total instructions in shared programs : 6206257 -> 6185058 (-0.34%)
total gprs used in shared programs : 911045 -> 910722 (-0.04%)
total local used in shared programs : 39072 -> 39072 (0.00%)
local gpr inst bytes
helped 0 417 4195 4195
hurt 0 280 0 0
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
This works when the add also has an immediate. This often happens in
address calculations. These addresses can then be inlined as well.
On code targeted to SM35:
total instructions in shared programs : 6223346 -> 6206257 (-0.27%)
total gprs used in shared programs : 911075 -> 911045 (-0.00%)
total local used in shared programs : 39072 -> 39072 (0.00%)
local gpr inst bytes
helped 0 119 3664 3664
hurt 0 74 15 15
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Even if the rasterizer has scissor disabled, we'll have whatever
vc4->scissor bounds were last set when someone set up a scissor, so we
shouldn't clip to them in that case.
Fixes piglit fbo-blit-rect, and a lot of MSAA tests once they're enabled.
We could potentially handle scissored blits when they're tile aligned, but
it doesn't seem worth it. If you're doing a scissored blit, you're
probably a testcase.
Fixes piglit's fbo-scissor-blit fbo
This implements more performance metrics than the previous support,
but some other metrics still need to be figured out.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
These performance metrics will be re-introduced in an upcoming
patch that will follow the same design as Fermi.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
inst_issued is performance metric not a hardware event on Kepler (SM30).
It will be re-introduced in an upcoming patch.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
SM30 is the compute capability version for GK104/GK106/GK107.
This also introduces a new signal group selection called UNK0F.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Move these to 'disasm' instead of the more verbose 'optmsgs' since, like
the tgsi dumps, it is useful without the more verbose compiler logging
enabled.
Signed-off-by: Rob Clark <robclark@freedesktop.org>
For MSAA, we store full resolution tile buffer contents, which have their
own tiling format. Since they're full resolution buffers, we have to
align their size to full tiles.
We were checking that the blit started at 0 and was 1:1, but not that it
went to the full width of the surface, or that the width was aligned to a
tile. We then told it to blit to the full width/height of the surface,
causing contents to be stomped in a bunch of MSAA tests that happen to
include half-screen-width blits to 0,0.
I've played with a few different approaches to tweak instruction
priority according to how much they increase/decrease register pressure,
etc. But nothing seems to change the fact that compared to original
(pre-multiple-block-support) scheduler, in some edge cases we are
generating shaders w/ 5-6x higher register usage.
The problem is that the priority queue approach completely looses the
dependency between instructions, and ends up scheduling all paths at the
same time.
Original reason for switching was that recursive approach relied on
starting from the shader outputs array. But we can achieve more or less
the same thing by starting from the depth-sorted list.
shader-db results:
total instructions in shared programs: 113350 -> 105183 (-7.21%)
total dwords in shared programs: 219328 -> 211168 (-3.72%)
total full registers used in shared programs: 7911 -> 7383 (-6.67%)
total half registers used in shader programs: 109 -> 109 (0.00%)
total const registers used in shared programs: 21294 -> 21294 (0.00%)
half full const instr dwords
helped 0 322 0 711 215
hurt 0 163 0 38 4
The shaders hurt tend to gain a register or two. While there are also a
lot of helped shaders that only loose a register or two, the more
complex ones tend to loose significanly more registers used. In some
more extreme cases, like glsl-fs-convolution-1.shader_test it is more
like 7 vs 34 registers!
Signed-off-by: Rob Clark <robclark@freedesktop.org>
It causes confusion in sched if we need to split_addr() since otherwise
we wouldn't easily know which block the new addr instr will be scheduled
in. So just side-step the whole situation.
Signed-off-by: Rob Clark <robclark@freedesktop.org>
We'll need to add similar for ir3_instruction, but following the pattern
to use 'id' seems confusing. Let's just go w/ generic 'data' as the
name.
Signed-off-by: Rob Clark <robclark@freedesktop.org>
Undefining the NDEBUG is relevant for release build, as they are the
ones that set it.
[Emil Velikov: split from previous patch]
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
This follows the src/util/u_atomic_test.c model of undefining NDEBUG
unconditionally throughouth the XvMC tests, to force asserts regardless
of debug mode.
The comment on u_atomic_test.c is also fixed (read 'debug' where it
should have been 'release').
v2: s/debug/release/ in relevant comments
Signed-off-by: Giuseppe Bilotta <giuseppe.bilotta@gmail.com>
[Emil Velikov: keep the src/util/ hunk as separate patch]
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Since we're using nir_lower_outputs_to_temporaries to shadow all our
outputs, it's impossible to actually get an indirect store. The code we
had to "handle" this was pretty bogus as it created a register with a
reladdr and then stuffed it in a fixed varying slot without so much as a
MOV. Not only does this not do the MOV, it also puts the indirect on the
wrong side of the transaction. Let's just delete the broken dead code.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
It's not really buying us anything at this point. It's just a way of
remapping one offset namespace onto another. We can just use the location
namespace the whole way through.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
The original change to put zeroes directly into instructions created
conditional mov's with the zero immediate. However that can't be
emitted, so make sure to replace the zero with r63.
Fixes: 52a800a68 (nv50/ir: allow immediate 0 to be loaded anywhere)
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
A situation where there's a 128-bit load where the last component gets
DCE'd causes a 96-bit load to be generated, which no GPU can actually
emit. Avoid generating such instructions by scaling back to 64-bit on
the first load when splitting.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Cc: "11.0 11.1" <mesa-stable@lists.freedesktop.org>
This was just plain broken. It used always the value from v0 (for vp_index)
but would pass the value from the provoking vertex to later stages - but only
if there was a corresponding fs input, otherwise the layer/vp index would get
lost completely (as it would try to interpolate the (unsigned) values as
floats).
So, make it obey provoking vertex rules (drivers relying on draw will need to
do the same). And make sure that the default interpolation mode (when no
corresponding fs input is found) for them is constant.
Also, change the code a bit so constant inputs aren't interpolated then
copied over later.
Fixes the new piglit test gl-layer-render-clipped.
v2: more consistent whitespaces fixes for function defs, and more tab killing
(overall still not quite right however).
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
Same as for llvmpipe, albeit softpipe only really handles multiple layers,
not multiple viewports/scissors.
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
d3d10 actually requires using provoking (first) vertex. GL is happy with
any vertex (as long as we say it's undefined in the corresponding queries).
Up to now we actually used vertex 0 for viewport index, and vertex 1 for
layer (for tris), which really didn't make sense (probably a typo). Also,$
since we reorder vertices of clockwise triangle, that actually meant we used
a different vertex depending if the traingle was cw or ccw (still ok by gl).
However, it should be consistent with what draw (clip) does, and using
provoking vertex seems like the sensible choice (draw clip will be fixed
next as it is totally broken there).
While here, also use the correct viewport always even when not needed
in setup (we pass it down to jit fragment shader it might be needed there
for getting correct near/far depth values).
No piglit changes.
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
We can't dump it in the real driver, since the kernel doesn't give us a
handle to it (except after a GPU hang, using a root ioctl). In the
simulator we can.
It should be MSVC2008_COMPAT_CFLAGS and not MSVC2008_COMPAT_CXXFLAGS.
This is why the recent util_blitter breakage went unnoticed on autotools
builds.
Trivial.
nir is the exception among gallium/auxiliary -- we don't need to compile
it with MSVC2008 yet. And this enables us to use
-Werror=declaration-after-statement in the next commit as we should,
without complicated fixes to tgsi_to_nir module.
Trvial. Tested with GCC and Clang.
Currently it stores strlen(buf) whenever the user originally provided a
negative value for length.
Although I've not seen any explicit text in the spec, CTS requires that
the very same length (be that negative value or not) is returned back on
Pop.
So let's push down the length < 0 checks, tweak the meaning of
gl_debug_message::length and fix GetDebugMessageLog to add and count the
null terminators, as required by the spec.
v2: return correct total length in GetDebugMessageLog
v3: rebase (drop _mesa_shader_debug hunk).
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Reviewed-by: Timothy Arceri <timothy.arceri@collabora.com>
We're about to rework the meaning of gl_debug_message::length to only
store the user provided data. Thus we should add an explicit validation
for null terminated strings.
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Reviewed-by: Timothy Arceri <timothy.arceri@collabora.com>
These new (relative to ARB_debug_output) tokens, have been explicitly
separated from the existing ones in the spec text. With the reference
to glDebugMessageInsert was dropped.
At the same time, further down the spec says:
"The value of <type> must be one of the values from Table 5.4"
... and these two are listed in Table 5.4.
The GL 4.3 and GLES 3.2 do not give any hints on the former
'definition', plus CTS requires that the tokens are valid values for
glDebugMessageInsert.
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Reviewed-by: Timothy Arceri <timothy.arceri@collabora.com>
As per the spec quote:
"All messages are initially enabled unless their assigned severity
is DEBUG_SEVERITY_LOW"
We already had MEDIUM and HIGH set, let's toggle NOTIFICATION as well.
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Reviewed-by: Timothy Arceri <timothy.arceri@collabora.com>
We already have one group (the default) as specified in the spec. So
lets return its size, rather than the index of the current group.
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Reviewed-by: Timothy Arceri <timothy.arceri@collabora.com>
The extension requires (cough implements) GetPointervKHR (alias of
GetPointerv) which in itself is available for ES 1.1 enabled mesa.
Anyone willing to fish around and implement it for ES 1.0 is more than
welcome to revert this commit. Until then lets restrict things.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=93048
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Reviewed-by: Timothy Arceri <timothy.arceri@collabora.com>
There are a few legacy OpenGL apps on Windows which need this extension.
We basically use glCopyTex[Sub]Image to implement wglBindTexImageARB (see
the implementation notes for details).
v2: refactor code to use st_copy_framebuffer_to_texture() helper function.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
Reviewed-by: Charmaine Lee <charmainel@vmware.com>
This helper is used by the WGL state tracker to implement the
wglBindTexImageARB() function.
This is basically a new "meta" function. However, we're not putting
it in the src/mesa/drivers/common/ directory because that code is not
linked with gallium-based drivers.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
Reviewed-by: Charmaine Lee <charmainel@vmware.com>
Atomic counters and Images were using ctx::Shader that does not take in
to account program pipeline changes, ctx::_Shader must be used for SSO to
work. Commit c0347705 already changed ubo's to use this.
Fixes failures seen with following Piglit test:
arb_separate_shader_object-atomic-counter
Signed-off-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Francisco Jerez <currojerez@riseup.net>
Cc: "11.0 11.1" <mesa-stable@lists.freedesktop.org>
Noticed this when looking at a trace that caused flags to spill to/from
registers. The flags source/destination wasn't encoded correctly
according to both envydis and nvdisasm.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
The algorithm expects the entire CFG to be reachable, so make sure that
we hit every node. Otherwise we will end up with uninitialized data,
memory corruption, etc.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
For example if there are only returns, the break bb will not end up part
of the CFG. However there will have been a prebreak already emitted for
it, and when hitting the RET that comes after, we will try to insert the
current (i.e. break) BB into the graph even though it will be
unreachable. This makes the SSA code sad.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Cc: "11.0 11.1" <mesa-stable@lists.freedesktop.org>
There's a post-RA fixup to replace 0's with $r63 (or $r127 if too many
regs are used), so just as nvc0, let an immediate 0 be loaded anywhere.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
This appeared in brw_vs.c and brw_wm.c, should have appeared in
brw_gs.c, and was soon going to have to be in brw_tcs.c and brw_tes.c as
well.
So, instead, move it to a central location (which has to know about both
struct brw_context and perf_debug()).
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
That texture mask thing doesn't seem to be needed for surface ops, so
just as nve4+, let do that only for texture ops.
This fixes a segfault with 'test_surface_st' from
gallium/tests/trivial/compute.c on Fermi because this test uses sustp.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
AppVeyor doesn't require an appveyor.yml in the repos (in fact it has
some limitations as noted in comments below), but doing so has two great
advantages over the web UI:
- appveyor.yml can be revisioned together with the code, so instructions
should always be in synch with the code
- appveyor.yml can be reused for people's private repositories (be on
fdo or GitHub, etc.)
Acked-by: Roland Scheidegger <sroland@vmware.com>
Previously util_blitter_clear_depth_stencil() could not clear more
than the first layer. We need to generalise this as we did for
util_blitter_clear_render_target().
Signed-off-by: Edward O'Callaghan <eocallaghan@alterapraxis.com>
Signed-off-by: Marek Olšák <marek.olsak@amd.com>
Previously util_blitter_clear_render_target() could not clear more
than the first layer. We need to generalise this so that
ARB_clear_texture can pass the 3d piglit test.
Signed-off-by: Edward O'Callaghan <eocallaghan@alterapraxis.com>
Signed-off-by: Marek Olšák <marek.olsak@amd.com>
These are implementation-dependent queries, but so far we just returned the
value of whatever the current provoking vertex convention was set to, which
was clearly wrong.
Just make this a variable in the context constants like for other things
which are implementation dependent (I assume all drivers will want to set
this to the same value for both queries), and set it to GL_UNDEFINED_VERTEX
which is correct for everybody (and drivers can override it).
Reviewed-by: Brian Paul <brianp@vmware.com>
CC: <mesa-stable@lists.freedesktop.org>
Since I just broke the scons build, I figured I'd make Travis test that I
don't break it again in the future. The script runs the builds in
parallel across VMs, so it still takes just 5 minutes to turn around
results.
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
The NVIDIA binary driver and Intel's closed source driver both expose
14 here, rather than the GL minimum of 12. Let's follow suit.
Without this, Shadow of Mordor fails to render correctly and triggers
OpenGL errors:
Mesa: User error: GL_INVALID_VALUE in glBindBufferBase(index=68)
Mesa: User error: GL_INVALID_VALUE in glUniformBlockBinding(block binding 68 >= 60)
There are 5 stages (VS, TCS, TES, GS, FS), and 12 * 5 = 60 is too small.
14 * 5 = 70 will work just fine.
Tapani believes this will also help Alien Isolation.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Acked-by: Tapani Pälli <tapani.palli@intel.com>
Acked-by: Matt Turner <mattst88@gmail.com>
The first pass marked dead instructions as opcode = NOP, and a second
pass deleted those instructions so that the live ranges used in the
first pass wouldn't change.
But since we're walking the instructions in reverse order, we can just
do everything in one pass. The only thing we have to do is walk the
blocks in reverse as well.
Reviewed-by: Francisco Jerez <currojerez@riseup.net>
Removes dead code from glsl-mat-from-int-ctor-03.shader_test.
Reported-by: Juan A. Suarez Romero <jasuarez@igalia.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
The docs say we should send the emit after the ring writes,
so lets do that and not have an ALU in between.
Signed-off-by: Dave Airlie <airlied@redhat.com>
For the compute support, we might stick buffers as surfaces. This fixes
an assertion when executing src/gallium/tests/trivial/compute.
To avoid using these "restricted" surfaces as render targets, these
assertions have been moved. Note that it's already handled for the
framebuffer thing on nvc0.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
String literals cannot exceed 65535 characters for MSVC. Instead of
emiting a string, emit an array of characters.
v2: fix indentation and add comment in the gl_enums.py file about this
ugliness.
Reviewed-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
This just builds/installs our dependencies, and runs "make check". I'm
interested in integrating more tests into it, but this seems like a pretty
easy first start.
If your personal branches of Mesa are on github, you can enable it on your
account and the repository (see
https://docs.travis-ci.com/user/for-beginners), then any pushes you do
will get their HEAD commit tested, and any pull requests to your tree will
get their merge commits tested.
Now when people need new extensions, they can skip the entire
enum-definition process, and we can stop reviewing new extension XML for
its enum content.
This also brings in a new enum that I wanted to use in enum_strings.cpp
for testing the code generator.
v2: Drop comment about disabled GL_1PASS_EXT test.
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
With GLES 3.1, GL 4.5, and many new vendor extensions about to get their
enums added, we jump up to 85k of table.
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Sometimes GL likes to rename an old enum when it grows a more general
purpose, and we should prefer the new name. Changes from this:
GL_POINT/LINE_SIZE_* (1.1) -> GL_SMOOTH_POINT/LINE_SIZE_* (1.2)
GL_FOG_COORDINATE_* (1.4) -> GL_FOG_COORD_* (1.5)
GL_SOURCE[012]_RGB/ALPHA (1.3) -> GL_SRC0_RGB (1.5)
GL_COPY_READ/WRITE_BUFFER (3.1) -> GL_COPY_READ_BUFFER_BINDING (4.2)
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Asking the table for bitfield names doesn't make any sense. For 0x10, do
you want GL_GLYPH_HORIZONTAL_BEARING_ADVANCE_BIT_NV or
GL_COLOR_BUFFER_BIT4_QCOM or GL_POLYGON_STIPPLE_BIT or
GL_SHADER_GLOBAL_ACCESS_BARRIER_BIT_NV? Giving a useful answer would
depend on a whole lot of context.
This also fixes a bad enum table entry, where we chose GL_HINT_BIT instead
of GL_ABGR_EXT for 0x8000, so we can now fix its entry in the enum_strings
test.
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
I've used a bunch of python code to cut out new enums so that the two
generated files can be diffed. I'll remove all that hardcoding in the
following commits. All remaining differences between the generated code:
- GL_TEXTURE_BUFFER_FORMAT didn't appear in GL3 when TBOs got merged to
core, so it now gets an _ARB suffix instead.
- Blacklisting can't keep EXT_sso's GL_ACTIVE_PROGRAM_EXT from becoming
GL_ACTIVE_PROGRAM -- in our hash table, GL_ACTIVE_PROGRAM_EXT points at
the GLES2 enum's value (aka GL_CURRENT_PROGRAM). By not blacklisting
the core name, we get both enums translated.
- GL_DRAW_FRAMEBUFFER_BINDING and GL_FRAMEBUFFER_BINDING both appeared in
GL3 as synonyms, and the new code happens to choose
GL_FRAMEBUFFER_BINDING instead.
- GL_TEXTURE_COMPONENTS and GL_TEXTURE_INTERNAL_FORMAT both appear in 1.1,
and the new code chooses GL_TEXTURE_INTERNAL_FORMAT instead (which seems
better, to me)
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
emacs whines at me every time I open the file about these unsafe
variables, and the file was reformatted from 8 space to 4 space long ago.
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
GL_ALL_ATTRIB_BITS is a thing, and GL_CLIENT_ALL_ATTRIB_BITS, but I don't
see GL_ALL_CLIENT_ATTRIB_BITS in my grepping of khronos XML, GL extension
specs, GL 1.1, GL 2.2, and GL 4.4.
Reviewed-by: Brian Paul <brianp@vmware.com>
Mesa hasn't been using these enums and the finalized specs don't reference
them, so losing them from our generated enum-to-string code should be
fine. Reduces diffs to generating from Khronos XML, which has these enums
noted defined but commented out from any consumers.
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
In converting to using the Khronos XML, I found that our XML had these two
swapped, and the text spec agreed.
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
The intention here is to keep a pristine copy of the upstream gl.xml that
can be updated at any time with a new version, and use that to generate
Mesa code from instead of our private XML.
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
The previous contents appeared to be the output of some form of code
generation for all enums, with a few entries hand-edited to deal with
oddness. The downside to this was that when an enum gets promoted from
vendor to _EXT or _EXT to _ARB or _ARB to core, make check starts failing
even when the commiter has done nothing wrong. Instead of black-box
testing the code generation, pick a few enums that intentionally poke the
interesting cases of code generation.
People editing the code generator should be diffing the generated code
anyway. This should catch when they fail to do so, without throwing false
negatives when people update the GL XML.
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
When probing for devices, clover will call pipe_loader_probe() twice.
The first time to retrieve the number of devices, and then second time
to retrieve the device structures.
We currently assume that the return value of both calls will be the
same, but this will not be the case if a device happens to disappear
between the two calls.
When a device disappears, the pipe_loader_probe() will add a NULL
device to the device list, so we need to handle this.
v2:
- Keep range for loop
Reviewed-by: Francisco Jerez <currojerez@riseup.net>
Acked-by: Emil Velikov <emil.l.velikov@gmail.com>
CC: <mesa-stable@lists.freedesktop.org>
Improves register pressure, since otherwise we end up emitting
loads for all the elements in the RHS and them emitting
stores for all elements in the LHS.
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Improves register pressure, since otherwise we end up emitting
loads for all the elements in the RHS and them emitting
stores for all elements in the LHS.
v2:
- Mark progress properly. This also fixes some instances where the added
nodes with individual element copies where not being lowered, which is
expected behavior as explained in the documentation for
visit_list_elements.
- Only need to do this if the RHS is a buffer-backed variable.
- We can also have arrays inside structs. A later patch will make it so
we also split struct copies and end up with multiple
ir_dereference_record assignments, so make sure that if any of these
is an array copy, we also split it.
Fixes the following piglit tests:
tests/spec/arb_shader_storage_buffer_object/execution/large-field-copy.shader_test
tests/spec/arb_shader_storage_buffer_object/linker/copy-large-array.shader_test
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Other hardwares than AMD require to parse:
VAPictureParameterBufferH264.ReferenceFrames[16]
Signed-off-by: Julien Isorce <j.isorce@samsung.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
In general max_references cannot be based on num_render_targets.
This patch allows to allocate buffers with an accurate size.
I.e. no more than necessary. For other codecs it is a fixed
value 2.
This is similar behaviour as vaapi/vdpau-driver.
For now HEVC case defaults to num_render_targets as before.
But it could also benefits this change by setting a more
accurate max_references number in handlePictureParameterBuffer.
Signed-off-by: Julien Isorce <j.isorce@samsung.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Emil Velikov <emil.l.velikov@gmail.com>
The current implementation looks for array dereferences on gl_FragData and
immediately proceeds to lower them, however this is not enough because we
can have array access on vector variables too, like in this code:
out vec4 color;
void main()
{
int i;
for (i = 0; i < 4; i++)
color[i] = 1.0;
}
Fix it by making sure that the actual variable being dereferenced is an array.
Fixes a crash in:
spec/arb_gpu_shader_fp64/execution/built-in-functions/fs-ldexp-dvec4.shader_test
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
We need to emit at least one cut/emit in every
geometry shader, the easiest workaround it to
stick a single CUT at the top of each geom shader.
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Cc: "10.6 11.0 11.1" <mesa-stable@lists.freedesktop.org>
Signed-off-by: Dave Airlie <airlied@redhat.com>
This fixes an issue where the addition of the FLAT qualifier in
varying_matches::record() can break the expected varying order.
It also avoids a future issue with the relaxing of interpolation
qualifier matching constraints in GLSL 4.50.
V2: (by Timothy Arceri)
* reworked comment slightly
Signed-off-by: Gregory Hainaut <gregory.hainaut@gmail.com>
Reviewed-by: Timothy Arceri <timothy.arceri@collabora.com>
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
GL_ARB_separate_shader_objects allow matching by name variable or block
interface. Input varyings can't be removed because it is will impact the
location assignment.
This fixes the bug 79783 and likely any application that uses
GL_ARB_separate_shader_objects extension.
V2 (by Timothy Arceri):
* simplify now that builtins are not set as always active
Signed-off-by: Gregory Hainaut <gregory.hainaut@gmail.com>
Reviewed-by: Timothy Arceri <timothy.arceri@collabora.com>
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
https://bugs.freedesktop.org/show_bug.cgi?id=79783
The value will be set in separate-shader program when an input/output
must remains active. e.g. when deadcode removal isn't allowed because
it will create interface location/name-matching mismatch.
v3:
* Rename the attribute
* Use ir_variable directly instead of ir_variable_refcount_visitor
* Move the foreach IR code in the linker file
v4:
* Fix variable name in assert
v5 (by Timothy Arceri):
* Rename functions and reword comments
* Don't set always active on builtins
Signed-off-by: Gregory Hainaut <gregory.hainaut@gmail.com>
Reviewed-by: Timothy Arceri <timothy.arceri@collabora.com>
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
This change allows used defined inputs/outputs with explicit locations
to be removed if they are detected to not be used between shaders
at link time.
To enable this we change the is_unmatched_generic_inout field to be
flagged when we have a user defined varying. Previously
explicit_location was assumed to be set only in builtins however SSO
allows the user to set an explicit location.
We then add a function to match explicit locations between shaders.
V2: call match_explicit_outputs_to_inputs() after
is_unmatched_generic_inout has been initialised.
Cc: Gregory Hainaut <gregory.hainaut@gmail.com>
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
When working on tessellation shaders, I created some vec4 virtual
opcodes for creating message headers through a sequence like:
mov(8) g7<1>UD 0x00000000UD { align1 WE_all 1Q compacted };
mov(1) g7.5<1>UD 0x00000100UD { align1 WE_all };
mov(1) g7<1>UD g0<0,1,0>UD { align1 WE_all compacted };
mov(1) g7.3<1>UD g8<0,1,0>UD { align1 WE_all };
This is done in the generator since the vec4 backend can't handle align1
regioning. From the visitor's point of view, this is a single opcode:
hs_set_output_urb_offsets vgrf7.0:UD, 1U, vgrf8.xxxx:UD
Normally, there's no hazard between sources and destinations - an
instruction (naturally) reads its sources, then writes the result to the
destination. However, when the virtual instruction generates multiple
hardware instructions, we can get into trouble.
In the above example, if the register allocator assigned vgrf7 and vgrf8
to the same hardware register, then we'd clobber the source with 0 in
the first instruction, and read back the wrong value in the last one.
It occured to me that this is exactly the same problem we have with
SIMD16 instructions that use W/UW or B/UB types with 0 stride. The
hardware implicitly decodes them as two SIMD8 instructions, and with
the overlapping regions, the first would clobber the second.
Previously, we handled that by incrementing the live range end IP by 1,
which works, but is excessive: the next instruction doesn't actually
care about that. It might also be the end of control flow. This might
keep values alive too long. What we really want is to say "my source
and destinations interfere".
This patch creates new infrastructure for doing just that, and teaches
the register allocator to add interference when there's a hazard. For
my vec4 case, we can determine this by switching on opcodes. For the
SIMD16 case, we just move the existing code there.
I audited our existing virtual opcodes that generate multiple
instructions; I believe FS_OPCODE_PACK_HALF_2x16_SPLIT needs this
treatment as well, but no others.
v2: Rebased by mattst88.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
We've apparently always been botching JIP for sequences such as:
do
cmp.f0.0 ...
(+f0.0) break
...
if
...
else
...
endif
...
while
Normally, UIP is supposed to point to the final destination of the jump,
while in nested control flow, JIP is supposed to point to the end of the
current nesting level. It essentially bounces out of the current nested
control flow, to an instruction that has a JIP which bounces out another
level, and so on.
In the above example, when setting JIP for the BREAK, we call
brw_find_next_block_end(), which begins a search after the BREAK for the
next ENDIF, ELSE, WHILE, or HALT. It ignores the IF and finds the ELSE,
setting JIP there.
This makes no sense at all. The break is supposed to skip over the
whole if/else/endif block entirely. They have a sibling relationship,
not a nesting relationship.
This patch fixes brw_find_next_block_end() to track depth as it does
its search, and ignore anything not at depth 0. So when it sees the
IF, it ignores everything until after the ENDIF. That way, it finds
the end of the right block.
I noticed this while reading some assembly code. We believe jumping
earlier is harmless, but makes the EU walk through a bunch of disabled
instructions for no reason. I noticed that GLBenchmark Manhattan had
a shader that contained a BREAK with a bogus JIP, but didn't measure
any performance improvement (it's likely miniscule, if there is any).
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Francisco Jerez <currojerez@riseup.net>
This just splits out a common pattern into an inline function
to make things cleaner to read.
Reviewed-by: Glenn Kennard <glenn.kennard@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
We really should initialise HS/LS_2 and SQ_LDS_ALLOC exists
on all evergreen not just cayman, so we should initialise
it as well.
Reviewed-by: Glenn Kennard <glenn.kennard@gmail.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
This adds the defines for a bunch of registers and shader
values that are required to implement tessellation.
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Glenn Kennard <glenn.kennard@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
Move some common code into one place, tess will also need
to use this function.
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
There was only a single user which was using strlen(buf).
As this function is not user facing (i.e. we don't need to feed back
original length via a callback), we can simplify things.
Suggested-by: Timothy Arceri <timothy.arceri@collabora.com>
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Reviewed-by: Timothy Arceri <timothy.arceri@collabora.com>
Add some checks if the original/dup'd fd is valid and ensure that we
don't leak it on error. The former is implicitly handled within the
pipe_loader, although let's make things explicit and check beforehand.
Spotted by Coverity (CID 1339865)
Cc: mesa-stable@lists.freedesktop.org
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
In theory this wouldn't be an issue, as we'll find the correct name and
break out of the loop before we hit the sentinel.
Let's fix this and avoid issues in the future.
Spotted by Coverity (CID 1339869, 1339870, 1339871)
Cc: mesa-stable@lists.freedesktop.org
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Earlier commit factored out the mpeg4 IQ matrix handling into separate
function, although it forgot to add a break in its case statement.
Thus the data ended up partially overwritten as the mpeg4 and h265
structs are members of the desc union.
Spotted by Coverity (CID 1341052)
Fixes: 64761a841d "st/va: move MPEG4 functions into separate file"
Cc: Julien Isorce <j.isorce@samsung.com>
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Non-timer queries are suspended during blits. When the blits end, the queries
are resumed, but this resume operation itself might run out of CS space and
trigger a flush. When this happens, we must prevent a duplicate suspend during
preflush suspend, and we must also prevent a duplicate resume when the CS flush
returns back to the original resume operation.
This fixes a regression that was introduced by:
commit 8a125afa6e
Author: Nicolai Hähnle <nhaehnle@gmail.com>
Date: Wed Nov 18 18:40:22 2015 +0100
radeon: ensure that timing/profiling queries are suspended on flush
The queries_suspended_for_flush flag is redundant because suspended queries
are not removed from their respective linked list.
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reported-by: Axel Davy <axel.davy@ens.fr>
Cc: "11.1" <mesa-stable@lists.freedesktop.org>
Tested-by: Axel Davy <axel.davy@ens.fr>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Zero length arrays are non standard:
warning C4200: nonstandard extension used : zero-sized array in struct/union
Cannot generate copy-ctor or copy-assignment operator when UDT contains a zero-sized array
And all code does `N * sizeof query_result->batch[0]`, so it should work
exactly the same.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Added to OpenGL 4.3 section, tagged as 'in progress (elima)'. See
https://bugs.freedesktop.org/show_bug.cgi?id=92687.
Thanks to Thomas H.P. Andersen for remainding me about this.
v1: - Update the already existing entry in section 4.3
instead (Ilia Mirkin).
- Added my BZ nickname as contact person (Felix Schwarz).
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
From Section 11.1.3.11 (Validation) of the GLES 3.1 spec:
"An INVALID_OPERATION error is generated by any command that trans-
fers vertices to the GL or launches compute work if the current set
of active program objects cannot be executed, for reasons including:"
It then goes on to list the rules we validate in the
_mesa_validate_program_pipeline() function.
For ValidateProgramPipeline the only mention of generating an error is:
"An INVALID_OPERATION error is generated if pipeline is not a name re-
turned from a previous call to GenProgramPipelines or if such a name has
since been deleted by DeleteProgramPipelines,"
Which we handle separately.
This fixes:
ES31-CTS.sepshaderobjs.PipelineApi
No regressions on the eEQP 3.1 tests.
Cc: Gregory Hainaut <gregory.hainaut@gmail.com>
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
Rather than assigning inloc up front, when we don't yet know if it will
be unused, assign it last thing before the legalize pass.
Also, realize when inputs are unused (since for frag shader's we can't
rely on them being removed from ir->inputs[]). This doesn't make sense
if we don't also dynamically assign the inloc's, since we could end up
telling the hw the wrong # of varyings (since we currently assume that
the # of varyings and max-inloc are related..)
Signed-off-by: Rob Clark <robclark@freedesktop.org>
Make the interpolation / point-sprite replacement mode setup deal with
varying packing.
In a later commit, we switch to packing just the varying components that
are actually used by the frag shader, so we won't be able to assume
everything is vec4's aligned to vec4. Which would highly confuse the
previous vinterp/vpsrepl logic.
Signed-off-by: Rob Clark <robclark@freedesktop.org>
The thread scratch space is thread-local so using the full IA-coherent
stateless surface index (255 since Gen8) is unnecessary and
potentially expensive. On Gen8 and early steppings of Gen9 this is
not a functional change because the kernel already sets bit 4 of
HDC_CHICKEN0 which overrides all HDC memory access to be non-coherent
in order to workaround a hardware bug.
This happens to fix a full system hang when running any spilling code
on a pre-production SKL GT4e machine I have on my desk (forcing all
HDC access to non-coherent from the kernel up to stepping F0 might be
a good idea though regardless of this patch), and improves performance
of the OglPSBump2 SynMark benchmark run with INTEL_DEBUG=spill_fs by
33% (11 runs, 5% significance) on a production SKL GT2 (on which HDC
IA-coherency is apparently functional so it wouldn't make sense to
disable globally).
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
Unfortunately Gen7 scratch block reads and writes seem to be hardwired
to BTI 255 even on Gen9+ where that index causes the dataport to do an
IA-coherent read or write. This change is required for the next patch
to be correct, since otherwise we would be writing to the scratch
space using non-coherent access and then reading it back using
IA-coherent reads, which wouldn't be guaranteed to return the value
previously written to the same location without introducing an
additional HDC flush in between.
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
Since the query names are not very enlightening, and there are thousands
of them, GALLIUM_HUD=help should only show the first and last query name
for each hardware block.
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
No drivers currently implement ARB_geometry_shader4, nor are there
any plans to implement it. We only support the version of geometry
shaders that was incorporated into OpenGL 3.2 / GLSL 1.50.
Signed-off-by: Marta Lofstedt <marta.lofstedt@linux.intel.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Patch adds additional mask for tracking which vertex arrays have
associated vertex buffer binding set. This array can be directly
compared to which vertex arrays are enabled and should match when
drawing.
Fixes following CTS tests:
ES31-CTS.draw_indirect.negative-noVBO-arrays
ES31-CTS.draw_indirect.negative-noVBO-elements
v2: update mask in vertex_array_attrib_binding
v3: rename mask and make it track _BoundArrays which matches what
was actually originally wanted (Fredrik Höglund)
v4: code cleanup, check for GLES 3.1 (Fredrik Höglund)
Signed-off-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Fredrik Höglund <fredrik@kde.org>
This was missed in commit 59cfb21d ("targets: use the non-inline sw
helpers").
Fixes build failure:
CXXLD libXvMCgallium.la
../../../../src/gallium/auxiliary/pipe-loader/.libs/libpipe_loader_static.a(libpipe_loader_static_la-pipe_loader_sw.o):(.data.rel.ro+0x0): undefined reference to `sw_screen_create'
collect2: error: ld returned 1 exit status
Makefile:756: recipe for target 'libXvMCgallium.la' failed
make[3]: *** [libXvMCgallium.la] Error 1
Trivial.
Analogous to previous commit. As we no longer have anyone who uses NIR
we can drop the link.
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Acked-by: Rob Clark <robdclark@gmail.com>
Previously (with the inline ones) things were embedded into the
pipe-loader, which means that we cannot control/select what we want in
each target.
That also meant that at runtime we ended up with the empty
sw_screen_create() as the GALLIUM_SOFTPIPE/LLVMPIPE were not set.
v2: Cover all the targets, not just dri.
Cc: "11.1" <mesa-stable@lists.freedesktop.org>
Cc: Ilia Mirkin <imirkin@alum.mit.edu>
Cc: Edward O'Callaghan <edward.ocallaghan@koparo.com>
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Tested-by: Oded Gabbay <oded.gabbay@gmail.com>
Tested-by: Nick Sarnie <commendsarnex@gmail.com>
While we correctly set output[] for composite varyings, we set completely
bogus values for output_components[], making emit_urb_writes() output
zeros instead of the actual values.
Unfortunately, our simple approach goes out the window, and we need to
recurse into structs to get the proper value of vector_elements for each
field.
Together with the previous patch, this fixes rendering in an upcoming
game from Feral Interactive.
v2: Use pointers instead of pass-by-mutable-reference (Jason, Matt).
Cc: "11.1 11.0" <mesa-stable@lists.freedesktop.org>
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Apparently we have literally no support for FS varying struct inputs.
This is somewhat surprising, given that we've had tests for that very
feature that have been passing for a long time.
Normally, varying packing splits up structures for us, so we don't see
them in the backend. However, with SSO, varying packing isn't around
to save us, and we get actual structs that we have to handle.
This patch changes fs_visitor::emit_general_interpolation() to work
recursively, properly handling nested structs/arrays/and so on.
(It's easier to read with diff -b, as indentation changes.)
When using the vec4 VS backend, this fixes rendering in an upcoming
game from Feral Interactive. (The scalar VS backend requires additional
bug fixes in the next patch.)
v2: Use pointers instead of pass-by-mutable-reference (Jason, Matt).
Cc: "11.1 11.0" <mesa-stable@lists.freedesktop.org>
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Expose most of the performance counter groups that are exposed by Catalyst.
Ideally, the driver will work with GPUPerfStudio at some point, but we are not
quite there yet. In any case, this is the reason for grouping multiple
instances of hardware blocks in the way it is implemented.
The counters can also be shown using the Gallium HUD. If one is interested to
see how work is distributed across multiple shader engines, one can set the
environment variable RADEON_PC_SEPARATE_SE=1 to obtain finer-grained performance
counter groups.
Part of the implementation is in radeon because an implementation for
older hardware would largely follow along the same lines, but exposing
a different set of blocks which are programmed slightly differently.
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Performance monitor queries can become very big, especially considering that
instances of a block in different shader engines are queried separately.
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
The enable of AMD_performance_monitor is no longer related to whether
queries are run by the GPU since the commit mentioned below.
Suggested-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
commit ddf27a3dd0
Author: Nicolai Hähnle <nhaehnle@gmail.com>
Date: Tue Nov 10 13:35:01 2015 +0100
gallium: remove pipe_driver_query_group_info field type
Most applications never use performance counters, so allow drivers to
skip potentially expensive initialization steps.
A driver that wants to use this must enable the appropriate extension(s)
at context initialization and set the InitPerfMonitorGroups driver function
which will be called the first time information about the performance monitor
groups is actually used.
The init_groups helper is called for API functions that can be called before
a monitor object exists. Functions that require an existing monitor object
can rely on init_groups having been called before.
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Previously pass did not traverse to those array dereferences which were
used as indices to arrays. This fixes Synmark2 Gl42CSCloth application
issues.
Signed-off-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Juha-Pekka Heikkila <juhapekka.heikkila@gmail.com>
templat->interlaced is 0 if not NV12 which is the case currently
when using VPP.
Signed-off-by: Julien Isorce <j.isorce@samsung.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
This setting is only used by glTexCoordPointer and related glEnable
calls. Since the preceeding commits removed all of those, it is not
necessary to save, reset to default, or restore this state.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
Nothing left in meta does anything with the VBO binding, so we don't
need to save or restore it. The VAO binding is still modified.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
tl;dr: For many types of GL object, we can *NEVER* use the Gen function.
In OpenGL ES (all versions!) and OpenGL compatibility profile,
applications don't have to call Gen functions. The GL spec is very
clear about how you can mix-and-match generated names and non-generated
names: you can use any name you want for a particular object type until
you call the Gen function for that object type.
Here's the problem scenario:
- Application calls a meta function that generates a name. The first
Gen will probably return 1.
- Application decides to use the same name for an object of the same
type without calling Gen. Many demo programs use names 1, 2, 3,
etc. without calling Gen.
- Application calls the meta function again, and the meta function
replaces the data. The application's data is lost, and the app
fails. Have fun debugging that.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=92363
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
tl;dr: For many types of GL object, we can *NEVER* use the Gen function.
In OpenGL ES (all versions!) and OpenGL compatibility profile,
applications don't have to call Gen functions. The GL spec is very
clear about how you can mix-and-match generated names and non-generated
names: you can use any name you want for a particular object type until
you call the Gen function for that object type.
Here's the problem scenario:
- Application calls a meta function that generates a name. The first
Gen will probably return 1.
- Application decides to use the same name for an object of the same
type without calling Gen. Many demo programs use names 1, 2, 3,
etc. without calling Gen.
- Application calls the meta function again, and the meta function
replaces the data. The application's data is lost, and the app
fails. Have fun debugging that.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=92363
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
tl;dr: For many types of GL object, we can *NEVER* use the Gen function.
In OpenGL ES (all versions!) and OpenGL compatibility profile,
applications don't have to call Gen functions. The GL spec is very
clear about how you can mix-and-match generated names and non-generated
names: you can use any name you want for a particular object type until
you call the Gen function for that object type.
Here's the problem scenario:
- Application calls a meta function that generates a name. The first
Gen will probably return 1.
- Application decides to use the same name for an object of the same
type without calling Gen. Many demo programs use names 1, 2, 3,
etc. without calling Gen.
- Application calls the meta function again, and the meta function
replaces the data. The application's data is lost, and the app
fails. Have fun debugging that.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=92363
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
The fixed-function attribute paths don't get the DSA treatment because
there are no DSA entry-points for fixed-function attributes. These
could have been added, but this is a temporary patch intended to make
later patches easier to review.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
Meta currently does this, but future changes will make this impossible.
Explicitly do it as a step in the patch series now to catch any possible
kinks.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
tl;dr: For many types of GL object, we can *NEVER* use the Gen function.
In OpenGL ES (all versions!) and OpenGL compatibility profile,
applications don't have to call Gen functions. The GL spec is very
clear about how you can mix-and-match generated names and non-generated
names: you can use any name you want for a particular object type until
you call the Gen function for that object type.
Here's the problem scenario:
- Application calls a meta function that generates a name. The first
Gen will probably return 1.
- Application decides to use the same name for an object of the same
type without calling Gen. Many demo programs use names 1, 2, 3,
etc. without calling Gen.
- Application calls the meta function again, and the meta function
replaces the data. The application's data is lost, and the app
fails. Have fun debugging that.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=92363
Reviewed-by: Abdiel Janulgue <abdiel.janulgue@linux.intel.com>
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
Instead of going through the GL API implementation functions, use the
lower-level functions. This means that we have to keep track of a
pointer to the gl_buffer_object and the gl_vertex_array_object.
This has two advantages. First, it avoids a bunch of CPU overhead in
looking up objects and validing API parameters. Second, and much more
importantly, it will allow us to stop calling _mesa_GenBuffers /
_mesa_CreateBuffers and pollute the buffer namespace (next patch).
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Abdiel Janulgue <abdiel.janulgue@linux.intel.com>
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
Future patches will use the brw_context instead. Keeping this
non-functional change separate should make the function changes easier
to review.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Abdiel Janulgue <abdiel.janulgue@linux.intel.com>
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
Pulls the parts of enable_vertex_array_attrib that aren't just parameter
validation out into a function that can be called from other parts of
Mesa (e.g., meta).
_mesa_enable_vertex_array_attrib can also be used to enable
fixed-function arrays.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
Pulls the parts of update_array_format that aren't just parameter
validation out into a function that can be called from other parts of
Mesa (e.g., meta).
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
This reverts commit a280e83d71.
It breaks INTEL_DEBUG=fs output. For example,
glsl-fs-discard-01.shader_test has 11 instructions but only prints 5.
Acked-by: Matt Turner <mattst88@gmail.com>
Coverity noticed that we were passing this by value, and it's 152 bytes.
Reviewed-by: Timothy Arceri <timothy.arceri@collabora.com>
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
It's only called from C, it compiles as C, so just compile it as C.
Notice the missing extern "C" on the definition of the function, which
would screw things up if the prototype wasn't parsed before the
definition.
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
We were including it in headers, which then caused it to be included in
tons of places it wasn't needed.
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
These functions' prototypes are marked with extern "C", which apparently
overrides a lack of extern "C" at the definition site if the prototype
has been seen first.
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Now that backend_reg inherits from brw_reg, we have to be careful to
avoid the object slicing problem.
Reviewed-by: Francisco Jerez <currojerez@riseup.net>
In the next patch, I make backend_reg's inheritance from brw_reg
private, which confuses clang when it sees the type "struct brw_reg" in
the derived class constructors, thinking it is referring to the
privately inherited brw_reg:
brw_fs.cpp:366:23: error: 'brw_reg' is a private member of 'brw_reg'
fs_reg::fs_reg(struct brw_reg reg) :
^
brw_shader.h:39:22: note: constrained by private inheritance here
struct backend_reg : private brw_reg
^~~~~~~~~~~~~~~
brw_reg.h:232:8: note: member is declared here
struct brw_reg {
^
Avoid this by marking brw_reg with the scope resolution operator.
In order to do this, we have to change the signature of the
backend_reg(brw_reg) constructor to take a reference to a brw_reg in
order to avoid unresolvable ambiguity about which constructor is
actually being called in the other modifications in this patch.
As far as I understand it, the rule in C++ is that if multiple
constructors are available for parent classes, the one closest to you in
the class heirarchy is closen, but if one of them didn't take a
reference, that screws things up.
Reviewed-by: Francisco Jerez <currojerez@riseup.net>
3333977556 added support for ASTC textures to
gallium. They don't have any helpers hooked up for software decoding, however,
so cannot support them in drivers relying on util code for decoding.
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
Removing the fake format helpers (1c7d0a6aa4)
caused this to fail. These formats were never supported, but previously
they would have asserted in the generated jit functions (which, due to lack
of test cases for these formats, were never called) whereas we now assert when
trying to build the jit function. So, skip them completely.
This fixes https://bugs.freedesktop.org/show_bug.cgi?id=93092
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
Kind of a handy function. And I'll want it available outside of i965
for common nir-pass helpers.
Signed-off-by: Rob Clark <robclark@freedesktop.org>
Reviewed-by: Nicolai Hähnle <nhaehnle@gmail.com>
The dirty area in this call isn't related to the screen at all.
v2: set clear dirty area to false as well
Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Julien Isorce <j.isorce@samsung.com>
Reviewed-by: Leo Liu <leo.liu@amd.com>
Enables 200+ dEQP SSO tests to proceed past validation,
and fixes a ES31-CTS.sepshaderobjs.PipelineApi subtest.
V2: split out change that reverts a previous patch into its own commit,
move variable declaration to top of function, and fix some formatting
all suggested by Ian.
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
Cc: "11.1" <mesa-stable@lists.freedesktop.org>
This reverts commit ba02f7a3b6.
The commit checked whether the pipeline was currently bound instead
of checking whether it had ever been bound. The previous setting
of Validated during object creation makes this unnecessary. The
real problem was that Validated was not properly set to false
elsewhere in the code. This is fixed by a later patch.
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
Cc: "11.1" <mesa-stable@lists.freedesktop.org>
This should fix the getteximage-depth test that currently asserts.
I was hitting problem with virgl as well in this area.
This moves the 1D array handling code to a single place.
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Ben Skeggs <bskeggs@redhat.com>
Cc: "10.6 11.0 11.1" <mesa-stable@lists.freedesktop.org>
Signed-off-by: Dave Airlie <airlied@redhat.com>
Now that nir_lower_tex can do texture swizzle lowering, we can use that
instead of repeating more-or-less the same code in both backends. This
both allows us to share code and means that things like the tg4
work-arounds are somewhat simpler because they don't have to take the
swizzle into account.
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
nir_ssa_def_rewrite_uses is one of the older helpers in NIR and predated
both of those. Now it can be substantially simplified.
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Previously, if someone accidentally made an instruction that refers to its
own SSA destination, the validator wouldn't catch it. The reason for this
is that it validated the destination too early and, by the time it got to
the source, the destination SSA value was already added to the set of seen
SSA values so it would assume that it came from some previous instruction.
By moving destination validation to be after source validation, the SSA
value is not in the list of seen values and the validator will catch
self-referential instructions.
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Previously, we had a rescale_texcoords helper in the FS backend for
handling rescaling of texture coordinates. Now that we can do variants in
NIR, we can use nir_lower_tex to do the rescaling for us. This allows us
to delete the i965-specific code and gives us proper TEXTURE_RECTANGLE and
GL_CLAMP handling in vertex and geometry shaders.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
This allows us to insert NIR passes between initial NIR compilation and
optimization (link time) and actual backend code-gen. In particular, it
will allow us to do shader variants in NIR and share some of that shader
variant code between backends.
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
At the moment, brw_create_nir just calls the three stages in sequence so
there's not much difference. Soon, however, we will want to start doing
variants in NIR at which point the postprocessing step will have to move
from shader create time to codegen time.
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
This fixes a regression introduced in b1a83b5d1 that caused basically all
shaders to fail to compile on 32-bit platforms.
Reported-by: Mark Janes <mark.a.janes@intel.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
It appears that the hardware wants the integer to be scaled the same way
that the hardware representation is. snorm16 uses one of the float
factors, so this is only relevant for snorm8.
This fixes a number of subcases of
bin/fbo-blending-formats GL_EXT_texture_snorm
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Cc: mesa-stable@lists.freedesktop.org
This doesn't account for the ldr/hdr distinction... that will probably
have to be exposed via a separate cap. When relevant hardware appears,
this can be worked out.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
This was a silly hack that kept growing and growing. Instead, just write
NULLs for those functions. No need to have helpers that just assert(0)
when you call them.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Not too long ago, the dri3 code was living in src/glx, which in itself
was guarded by HAVE_DRI_GLX. As the name suggests we didn't dive into
the folder when dri was disabled, thus we missed that dri3 does not
consider/honour --enable-dri.
Cc: mesa-stable@lists.freedesktop.org
Fixes: 6bd9ba7d07 "loader: Add dri3 helper"
Cc: Pali Rohár <pali.rohar@gmail.com>
Reported-by: Pali Rohár <pali.rohar@gmail.com>
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
It looks like the sampler hardware doesn't take into account the
surface format when sampling a cleared color after a fast clear has
been done. So for example if you clear a GL_RED surface to 1,1,1,1
then the sampling instructions will return 1,1,1,1 instead of 1,0,0,1.
This patch makes it override the color that is programmed in the
surface state in order to swizzle for luminance and intensity as well
as overriding the missing components.
Fixes the ext_framebuffer_multisample-fast-clear Piglit test.
v2: Handle luminance and intensity formats
Reviewed-by: Ben Widawsky <benjamin.widawsky@intel.com>
v2: do the same in tgsi_to_nir (Samuel)
v3: added missing cases after rebase (Iago)
v4: Add a blank space after '#' in one of the comments (Matt)
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
This way the caller doesn't have to initialize all 4 channels when they
aren't using them.
v2: Fix signed/unsigned comparison warning (Iago)
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
There are various restrictions on what the hstride can be that depend on
the Gen, and now that we're using hstride == 2 for packing/unpacking
doubles, we're going to run into these restrictions a lot more often.
Pull them out into a separate function, and move the one restriction we
checked previously into it.
Reviewed-by: Matt Turner <mattst88@gmail.com>
This can happen when the source of the compare was split by the SIMD
lowering pass. Potentially, we could allow the case where the exec size
of scan_inst is larger, and scan_inst has the right quarter selected,
but doing that seems a little more risky.
v2: Merge the bail condition into the the previous if/break block (Matt)
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
If we tried to get/set something that was exactly 64 bits, we would
try to do (1 << 64) - 1 to calculate the mask which doesn't give us all
1's like we want.
v2 (Iago)
- Replace ~0 by ~0ull
- Removed unnecessary parenthesis
v3 (Kristian)
- Avoid the conditional
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
Regression as of 64710db664
We can't use the type returned by get_interface_type() as
the interface type has arrays removed.
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
In case that the buffer has no bind at all, assume it can be a regular
buffer. This can happen on buffers created through the ARB_dsa
interfaces.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Cc: "11.0 11.1" <mesa-stable@lists.freedesktop.org>
With ARB_direct_state_access, buffers can be created without any binding
hints at all. We still need to allocate these buffers to VRAM or GART,
as we don't have logic down the line to place them into GPU-mappable
space. Ideally we'd be able to shift these things around based on usage.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=92438
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Cc: "11.0 11.1" <mesa-stable@lists.freedesktop.org>
They're exclusive at build time, but the ilo entry is always present, so
we'd try to use it and fail out.
v2: Add comment in the code, from Emil.
Cc: mesa-stable@lists.freedesktop.org
Reviewed-by: Emil Velikov <emil.l.velikov@gmail.com>
A prior, literal reading of the ASTC spec led to the prohibition
of some compressed formats being used against the targets:
TEXTURE_CUBE_MAP_ARRAY and TEXTURE_3D. Since the spec does not specify
interactions with other extensions for specific compressed textures,
remove such interactions.
Fixes the following Piglit tests on Gen9:
piglit.spec.arb_direct_state_access.getcompressedtextureimage
piglit.spec.arb_get_texture_sub_image.arb_get_texture_sub_image-getcompressed
piglit.spec.arb_texture_cube_map_array.fbo-generatemipmap-cubemap array s3tc_dxt1
piglit.spec.ext_texture_compression_s3tc.getteximage-targets cube_array s3tc
v2. Don't interact with other specific compressed formats (Ian).
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=91927
Suggested-by: Neil Roberts <neil@linux.intel.com>
Signed-off-by: Nanley Chery <nanley.g.chery@intel.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
Provide the ability to prevent any permanently enabled extension
from appearing in the string returned by glGetString[i]().
Signed-off-by: Nanley Chery <nanley.g.chery@intel.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Tested-by: Brian Paul <brianp@vmware.com>
I noticed that brw_vs.c does this.
I believe the point is that nir->num_uniforms is either counted in
scalar components (in scalar mode), or vec4 slots (in vector mode).
But we want param_count to be in scalar components regardless, so
we have to scale up in vector mode.
We don't have to scale up in scalar mode, though.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
I think I may have regressed this in the NIR conversion. TGSI-to-NIR is
putting the PSIZ in the .x channel, not .w, so we were grabbing some
garbage for point size, which ended up meaning just not drawing points.
Fixes glean pointAtten and pointsprite.
Same fix as on a3xx - set the second (tiny) layer size bitfield to the
smallest level's size so that the hw knows not to minify beyond that.
This fixes texelFetch sampler3D piglits.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Cc: mesa-stable@lists.freedesktop.org
When layer is the container, slices are tightly packed inside of each
layer. We don't need any additional alignment. On a3xx, each slice
contains all the layers, so having alignment makes sense.
This fixes a whole slew of array-related piglits, including texelFetch
and tex-miplevel-selection varieties.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Cc: mesa-stable@lists.freedesktop.org
We delay the null check only to jump through hoops to work around that.
Check early to make our lives easier.
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Acked-by: Rob Clark <robclark@freedesktop.org>
As of earlier all the targets use the non inline version. Don't forget
to remove the function prototypes/declarations.
v2: rebase on top of virgl support.
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Acked-by: Rob Clark <robclark@freedesktop.org>
Covert DRI to use only the pipe-loader interface.
With drisw_create_screen and kms_swrast_create_screen replaced by their
pipe-loader equivalent, we can now drop them.
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Acked-by: Rob Clark <robclark@freedesktop.org>
We cannot use this C99 feature here quite yet, as the code needs to be
build with MSVC prior to 2013.
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Acked-by: Rob Clark <robclark@freedesktop.org>
Unlike the inline ones, here we'd want to have an extern definition of
the functions. This is required as with follow-up commits, we'll
gradually start using the static pipe-loader, with the latter needing
the symbols.
These are direct copy from the inline version.
v2:
- rebase on top of virgl support
- add "driver missing" printfs (Nicolai)
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Acked-by: Rob Clark <robclark@freedesktop.org>
Rather than having all targets include the file, with only some defining
the relevant guard macro, just move things where they are used.
v2: rebase on top of virgl support.
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Acked-by: Rob Clark <robclark@freedesktop.org>
As of last few commits we have a static and dynamic pipe-loader. Either
of which will be used with (almost) all targets..
We can look into allowing the user to select which way the targets are
built, be that 'static for all' or 'per target' in follow up commits.
After which we can look into building only the static or dynamic
version, although building both shouldn't cause any issues.
Hack/workaround alert:
Control the standalone pipe-drivers via HAVE_CLOVER. Will need to be
fixed as the targets are converted/configure knobs are in.
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Acked-by: Rob Clark <robclark@freedesktop.org>
Analogous to previous commit with a small catch.
As the sw inline helpers are mere wrappers, and the screen <> winsys
split is more prominent (with the latter not being part of the final
pipe-driver), things will just work.
v2: rebase on top of earlier 'consolitate teardown' changes
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Acked-by: Rob Clark <robclark@freedesktop.org>
Add a list of driver descriptors and select one from the list, during
probe time.
As we'll need to have all the driver pipe_foo_screen_create() functions
provided externally (i.e. from another static lib) we need a separate
(non-inline) drm_helper, which contains the function declarations.
v2: rebase on top of virgl support.
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Acked-by: Rob Clark <robclark@freedesktop.org>
It is to be used in contrast of the dynamic one. The state-tracker does
not need to know if the pipe-driver is built into the final blob or
a separate object. This will allow us to move the logic to the final
step (in target) where the appropriate pipe-loader will be chosen.
Cc: Tom Stellard <thomas.stellard@amd.com>
Cc: Francisco Jerez <currojerez@riseup.net>
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Acked-by: Rob Clark <robclark@freedesktop.org>
With the next commits we'll introduce a 'static' version, which will
essentially load the statically linked-in pipe-drivers, rather than the
standalone pipe-$foo.so ones.
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Acked-by: Rob Clark <robclark@freedesktop.org>
Rather than giving false hopes that things might work, just check at
probe time. This allows us to remove the duplication and consolidate
the code wrt the upcomming static pipe-loader.
Cc: Tom Stellard <thomas.stellard@amd.com>
Cc: Francisco Jerez <currojerez@riseup.net>
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Acked-by: Rob Clark <robclark@freedesktop.org>
i.e. plug some (hard to hit) memory leaks.
v2: fix rebase fallout - really teardown the winsys (Brian)
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Acked-by: Rob Clark <robclark@freedesktop.org>
Move the winsys into the pipe-target, similar to the hardware
pipe-driver.
v2:
- move int declaration outside of loop (Brian)
- fold the teardown into a goto + separate function.
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Acked-by: Rob Clark <robclark@freedesktop.org>
Currently the location is determined at configure/build time and
consistently copied across gallium. Just remove the extra argument, and
use PIPE_SEARCH_DIR where appropriate.
This will allow us to remove the duplication in the *configuration and
*screen_create APIs by moving util_dl_get_proc_address() and friends to
probe time.
v2: rebase on top of vl_winsys_drm.c addition
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Acked-by: Rob Clark <robclark@freedesktop.org>
Since the up-streaming of nine, the static target was used by default.
The dynamic pipe-drivers being available only via manual tweak of
configure.ac.
As we'll be removing the library_path argument from the pipe-loader with
follow-up commits, we can remove D3D9_DRIVERS_PATH/D3D9_DRIVERS_DIR.
Everyone doing local hacking on nine, or wishing to have a env override
can bring them back within the pipe-loader.
Cc: Axel Davy <axel.davy@ens.fr>
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Acked-by: Rob Clark <robclark@freedesktop.org>
... in favour of HAVE_LIBDRM. After all we solely want to build the code
when the latter is available.
In the not too distant future we will remove the libudev/sysfs
dependency and simplify configure.ac even further.
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Acked-by: Rob Clark <robclark@freedesktop.org>
They serve little to no purpose, as we don't need any additional
dependencies (headers and/or symbols). On the other hand dropping them
will allow us to use GALLIUM_PIPE_LOADER_DEFINES in only one single
place - the pipe-loader.
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Acked-by: Rob Clark <robclark@freedesktop.org>
Currently every target makes sure that the screen is non-null prior to
using the debug (trace including) wrappers. If that no longer holds true
we want to know and fix this ASAP rather than silently bailing out.
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Acked-by: Rob Clark <robclark@freedesktop.org>
nir_locals, nir_ssa_values, and nir_system_values are all dst_reg (not
that that makes a whole lot of sense to me), and only nir_inputs is a
src_reg.
Reviewed-by: Francisco Jerez <currojerez@riseup.net>
In most cases (when the negate is copy propagated and the MOV removed),
this is two instructions on Gen >= 8 and only two instructions on
earlier platforms -- and it doesn't use the flag register.
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
a4xx hardware has real support for RGTC so there's no need to fake it
like we do on a3xx. Undo the hacks, and keep track of an "internal
format" of a resource, which on a3xx will be different, triggering the
transfer-time conversions to take place.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Looks like a4xx hw does this in a more standard way and we don't need to
hack around it like we do on a3xx. Fixes GL_ALPHA formats in
fbo-blending-formats, fbo-colormask-formats, and fbo-alphatest-formats.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Cc: mesa-stable@lists.freedesktop.org
Instead of playing the guessing game as to which texture format reads
from which border color encoding type, just write both of them always.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
There are not native RGBX render formats, so we must manually force
dst_alpha to be one, same as for a3xx.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
The point of prepare_buffer is to ensure that the query buffer contains valid
initial data for conditional rendering: as long as the buffer is initialized
correctly, the GPU is able to tell whether query results have been written
already (and wait or fall back to unconditional rendering if desired).
This means prepare_buffer needs to be called again when a buffer is reused.
Conversely, for queries that cannot be used for conditional rendering
(notably pipeline statistics), we can re-use buffers immediately, and they
do not need to be initialized.
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Tested-by: Andy Furniss <adf.lists@gmail.com>
To graph the number of bytes uploaded to GPU per frame (vertex buffer data,
constant buffer data, texture data, etc).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
Reviewed-by: Charmaine Lee <charmainel@vmware.com>
ARB_explicit_uniform_location allows the index for subroutine functions
to be explicitly set in the shader.
This patch reduces the restriction on the index qualifier in
validate_layout_qualifiers() to allow it to be applied to subroutines
and adds the new subroutine qualifier validation to ast_function::hir().
ast_fully_specified_type::has_qualifiers() is updated to allow the
index qualifier on subroutine functions when explicit uniform locations
is available.
A new check is added to ast_type_qualifier::merge_qualifier() to stop
multiple function qualifiers from being defied, before this patch this
would cause a segfault.
Finally a new variable is added to ir_function_signature to store the
index. This value is validated and the non explicit values assigned in
link_assign_subroutine_types().
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
This patch replaces the old interger constant qualifiers with either
the new ast_layout_expression type if the qualifier requires merging
or ast_expression if the qualifier can't have mulitple declarations
or if all but the newest qualifier is simply ignored.
We also update the process_qualifier_constant() helper to be
similar to the one in the ast_layout_expression class, but in
this case it will be used to process the ast_expression qualifiers.
Global shader layout qualifier validation is moved out of the parser
in this change as we now need to evaluate any constant expression
before doing the validation.
V2: Fix minimum value check for vertices (Emil)
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
In this patch we introduce a new ast type for holding the new
compile-time constant expressions. The main reason for this is that
we can no longer do merging of layout qualifiers before they have been
converted into GLSL IR so we need to store them to be proccessed later.
The new type has two helper functions:
- process_qualifier_constant()
Used to merge and then evaluate qualifier expressions
- merge_qualifier()
Simply appends a qualifier to a list to be merged later by
process_qualifier_constant()
In order to avoid cascading error messages the process_qualifier_constant()
helpers return a bool
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
This will allow us to add error checking to this function
in a later patch, if we don't move it the error messages
will go missing.
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
This change moves the binding layout handing code into an apply
function to be consistent with other helper functions in the ast
code, and to encapsulate the code so that when we introduce
compile time constants the code will be much cleaner.
One small downside is for unnamed interface blocks we will now
be revalidating the binding for each member its applied to.
However this seems a small sacrifice in order to have code which
is readable.
We also remove the incorrect comment in the named interface code
about propagating bindings to members which seems to have been
copied from the unnamed interface code.
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
This validation is moved later so we can validate the
max value when compile time constant support is added in a
later patch.
Reviewed-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
We are moving this out of the parser in preparation for compile
time constant support.
The reason a validation function is used rather than an apply
function like what is used with bindings is because glsl allows
streams to be defined on members of blocks even though they must
match the stream thats associated with the current block, this
means we need access to the value after validation to do this
comparision.
V2: Fix typo in comment (Emil)
Reviewed-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
Use new helper that will in a later patch allow for
compile time constants.
Reviewed-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
The minimum value for index is validated in apply_explicit_location()
and we want to remove validation from the parser so we can add
compile time constant support.
Reviewed-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
We are moving this out of the parser in preparation for compile
time constant support.
Reviewed-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
For now this just validates that a qualifier is inside its
minimum boundary, in a later patch we will expand it to
evaluate compile time constants.
Reviewed-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
SKL supports the ability to do fast clears and resolves of 32b RGBA as both
integer and floats. This patch only enables float color clears because we
haven't yet enabled integer color clears, (HW support for that was added in
BDW).
v2: Remove LUMINANCE16F and INTENSITY16F special cases since they are now
handled by Neil's patch to disable MSAA fast clears.
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Reviewed-by: Neil Roberts <neil@linux.intel.com>
Reviewed-by: Chad Versace <chad.versace@intel.com>
This reverts commit 8a0c85b258.
It's not a strict revert because I don't want to bring back the gen < 9 check at
this point in time.
Reviewed-by: Neil Roberts <neil@linux.intel.com>
The impetus for this patch comes from a seemingly benign statement within the
spec (quoted within the patch).
It is very important for clearing multiple color buffer attachments and can be
observed in the following piglit tests:
spec/arb_framebuffer_object/fbo-drawbuffers-none glclear
spec/ext_framebuffer_multisample/blit-multiple-render-targets 0
v2: Doing the framebuffer binding only once (Chad)
Directly use the renderbuffers from the mt (Chad)
v3: Patch from Neil whose feedback I originally missed.
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Reviewed-by: Chad Versace <chad.versace@intel.com>
Reviewed-by: Neil Roberts <neil@linux.intel.com>
Some of the information originally in this commit message is now in the patch
before this.
SKL adds compressible render targets and as a result mutates some of the
programming for fast clears and resolves. There is a new internal surface type
called the CCS. The old AUX_MCS bit becomes AUX_CCS_D. "Auxiliary Surfaces For
Sampled Tiled Resource".
The formats which are supported are defined in the table titled "Render Target
Surface Types [SKL+]". There is no PRM yet to reference. The previously
implemented helper function already does the right thing provided the table is
correct.
v2: Use better English in commit message (Matt)
s/compressable/compressible/ (Matt)
Don't compare bools to true (Matt)
Use the helper function and don't increase the context size - this is mostly
implemented in the patch just before this (Chad, Neil)
Remove an "invalid" assert (Chad)
Fix assertion to check num_samples > 1, instead of num_samples (Chad)
v3:
Use Matt's code as Requested-by: Chad. I didn't even look at it since Chad said
he was fine with that, and presumably Matt is fine with it.
v4: Use better quote from spec (Topi)
Cc: Chad Versace <chad.versace@intel.com>
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
Background: Prior to Skylake and since Ivybridge Intel hardware has had the
ability to use a MCS (Multisample Control Surface) as auxiliary data in
"compression" operations on the surface. This reduces memory bandwidth. This
hardware was either used for MSAA compression, or fast clear operations. On
Gen8, a similar mechanism exists to allow the hiz buffer to be sampled from, and
therefore this feature is sometimes referred to more generally as "AUX buffers".
Skylake adds the ability to have the display engine directly source compressed
surfaces on top of the ability to sample from them. Inference dictates that
enabling this display features adds a restriction to the formats which could
actually be compressed. This is backed up by a blurb in the AUX_CCS_D section
from the RENDER_SURFACE_STATE: "In addition, if the surface is bound to the
sampling engine, Surface Format must be supported for Render Target Compression
for surfaces bound to the sampling engine." The current set of surfaces seems
to be a subset as compared to previous gens (see the next patch). Also, if I had
to guess I would guess that future gens add support for more surface formats. To
make handling this a bit easier to read, and more future proof, the support for
this is moved into the surface formats table.
Along with the modifications to the table, a helper function is also provided to
determine if a surface is CCS_E compatible. Because fast clears are currently
disabled on SKL, we can plumb the helper all the way through here, and not
actually have anything break.
v2:
- rename ccs to ccs_e; Requested-by: Chad
- rename lossless_compression to lossless_compression Requested-by: Chad
- change meaning of brw_losslessly_compressible_format Requested-by: Chad
- related changes to the code to reflect this.
- remove excess ccs (Chad)
v3:
- Commit message changes (Topi)
- Const some things which could be const (Topi)
Requested-by: Chad Versace <chad.versace@intel.com>
Requested-by: Neil Roberts <neil@linux.intel.com>
Signed-off-by: Ben Widawsky <benjamin.widawsky@intel.com>
Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
Reviewed-by: Chad Versace <chad.versace@intel.com>
Patch was originally called:
i965/skl: Enable fast color clears on SKL
Skylake introduces some differences in the way that fast clears are programmed
and in the restrictions for using fast clears. Since some of these are
non-obvious, and fast clears are currently disabled globally, we can enable the
simple stuff here and leave the weirder stuff and separately reviewable work.
Based on a patch originally from Kristian.
Note that within this patch the change in scaling factors could be achieved with
this hunk instead. I've opted to keep things more like how the docs describe it
however.
--- a/src/mesa/drivers/dri/i965/intel_mipmap_tree.c
+++ b/src/mesa/drivers/dri/i965/intel_mipmap_tree.c
@@ -150,9 +150,13 @@ intel_get_non_msrt_mcs_alignment(struct brw_context *brw,
/* In release builds, fall through */
case I915_TILING_Y:
*width_px = 32 / mt->cpp;
- *height = 4;
+ if (brw->gen >= 9)
+ *height = 2;
+ else
+ *height = 4;
v2: Add braces for the multiline (Matt + Chad)
Comment updates (requested by Chad)
Modified commit message
Commit message from Chad explaining the MCS height change (Chad)
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Reviewed-by: Neil Roberts <neil@linux.intel.com>
Reviewed-by: Chad Versace <chad.versace@intel.com>
Commit b9b40ef9b7 moved the file, but forgot to update the reference in
the makefile. Thus the out of tree build was busted :\
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
With llvm 3.7 semi-dropping the autoconf build, we rely on their cmake
build. With the latter of which annoyingly using another (busted?)
SONAME.
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
The queries_suspended_for_flush flag is redundant because suspended queries
are not removed from their respective linked list.
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Some drivers (in particular radeon[si], but also freedreno judging from
a quick grep) may want to expose performance counters that cannot be
individually enabled or disabled.
Allow such drivers to mark driver-specific queries as requiring a new
type of batch query object that is used to start and stop a list of queries
simultaneously.
v3: adjust recently added nv50 queries
v2: documentation for create_batch_query
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Tested-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
It is easy enough to pre-determine the required size, and arrays are
generally better behaved especially when they get large.
v2: make sure init_perf_monitor returns true when no counters are active
(spotted by Samuel Pitoiset)
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Tested-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Previously, when a performance monitor was initialized, an inner loop through
all driver queries with string comparisons for each enabled performance
monitor counter was used. This hurts when a driver exposes lots of queries.
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Tested-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
This was only used to implement an unnecessarily restrictive interpretation
of the spec of AMD_performance_monitor. The spec says
A performance monitor consists of a number of hardware and software
counters that can be sampled by the GPU and reported back to the
application.
I guess one could take this as a requirement that counters _must_ be sampled
by the GPU, but then why are they called _software_ counters? Besides,
there's not much reason _not_ to expose all counters that are available,
and this simplifies the code.
v3: add a missing change in the nouveau driver (thanks Samuel Pitoiset)
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Tested-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
texel fetches don't use any samplers. Previously we just set the same
number for both texture and sampler unit (as per "ordinary" gl style
sampling where the numbers are always the same) however this would trigger
some assertions checking that the sampler index isn't over PIPE_MAX_SAMPLERS
limit elsewhere with d3d10, so just set to 0.
(Fixing the assertion instead isn't really an option, the sampler isn't
really used but might still pass an out-of-bound pointer around and even
copy some things from it.)
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
The change is necessary to avoid building errors in glsl and i965
modules due to missing glsl_types.h header
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
In preparation for supporting GL_KHR_debug in OpenGL ES
v2: add a missing hunk in _mesa_IsEnabled (Emil)
Signed-off-by: Boyan Ding <boyan.j.ding@gmail.com>
Reviewed-by: Emil Velikov <emil.l.velikov@gmail.com>
As defined in the spec
when implemented in an OpenGL ES context, all entry points defined
by this extension must have a "KHR" suffix.
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Current behavior on the interface matching:
layout (location = 0) out0; // Assigned to VARYING_SLOT_VAR0 by user
out1; // Assigned to VARYING_SLOT_VAR0 by the linker
New behavior on the interface matching:
layout (location = 0) out0; // Assigned to VARYING_SLOT_VAR0 by user
out1; // Assigned to VARYING_SLOT_VAR1 by the linker
v4:
* Fix variable name in assert
Signed-off-by: Gregory Hainaut <gregory.hainaut@gmail.com>
Reviewed-by: Timothy Arceri <timothy.arceri@collabora.com>
Analogous to previous commit. While we're here prefix all functions
identically -> vl_dri2_foo
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Acked-by: Alex Deucher <alexander.deucher@amd.com>
As of last commit everyone is using the vl_screen dispatch, thus we can
hide this function from the headers and make it static.
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Acked-by: Alex Deucher <alexander.deucher@amd.com>
As mentioned previously, it will allow us to use different vl backend in
a generic way from either video state-tracker.
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Acked-by: Alex Deucher <alexander.deucher@amd.com>
In a preparation of having proper multi-platform/backend handling in VL.
With follow up commits we'll introduce a dispatch within vl_screen
similar to the one in pipe_screen. This way any VL state-tracker can
operate seamlessly, considering the backend/platform is properly setup.
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Acked-by: Alex Deucher <alexander.deucher@amd.com>
The current code is busted in a number of ways.
- initially checks for omx_display (rather than omx_screen), which may
or may not be around.
- blindly feeds the empty env variable string to loader_open_device()
- reads the env variable every time get_screen is called
- the latter manifests into memory leaks, and other issues as one sets
the variable between two get_screen calls.
Additionally it cleans up a couple of extra bits
- drops unneeded set/check of omx_display.
- make the teardown (put_screen) order was not symmetrical to the setup
(get_screen)
v2: Drop the "is empty string" check (Leo)
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Reviewed-by: Leo Liu <leo.liu@amd.com>
Rather than duplicating things, just use the generic AM_CPPFLAGS. This
has the fortunate side-effect of adding VISIBILITY_CFLAGS for the dri3
helper. The latter of which was erroneously exposing some internal
symbols.
Signed-off-by: Emil Velikov <emil.velikov@collabora.com>
Reported-by: Kai Wasserbäch <kai@dev.carbon-project.org>
Tested-by: Kai Wasserbäch <kai@dev.carbon-project.org>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
Reviewed-by: Matt Turner <mattst88@gmail.com>
On the vec4 backend, textureSamplesIdentical() will always return
false. There are currently no test cases for the vec4 backend, so we
don't have much confidence in any implementation. We also don't think
anyone is likely to miss it.
v2: Handle immediate value for MCS smarter. Rebase on changes to
nir_texop_sampels_identical (missing second parameter). Suggested by
Jason.
v3: Add Neil's code to handle 16x MSAA in the FS. Also rebase on top of
f9a9ba5e. Stub out the vec4 implementation.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Signed-off-by: Neil Roberts <neil@linux.intel.com>
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com> [v2]
Reviewed-by: Chris Forbes <chrisf@ijw.co.nz> [v2]
This is the NIR analog to GLSL IR ir_samples_identical.
v2: Don't add the second nir_tex_src_ms_index parameter. Suggested by
Ken and Jason.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
Reviewed-by: Chris Forbes <chrisf@ijw.co.nz>
v2: Add Neil to the list of contributors. I meant to do that before,
but Matt reminded me.
v3: Fix typos noticed by Nicolai.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Chris Forbes <chrisf@ijw.co.nz>
Note: not quite perfect, we should use type_size vfunc (in
compiler_options or nir_shader?) to determine how much we
increment num_inputs/outputs/uniforms. But we don't have
that yet, so let's at least fix things for the existing
users of these passes.
Signed-off-by: Rob Clark <robclark@freedesktop.org>
Acked-by: Jason Ekstrand <jason.ekstrand@intel.com>
Otherwise, passing -1 gets you:
error: invalid conversion from 'int' to 'nir_variable_mode' [-fpermissive]
Signed-off-by: Rob Clark <robdclark@gmail.com>
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
This fixes teximage-colors, fbo-generatemipmap-formats, and probably
others (in relation to the RGB5 formats, others still fail).
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Cc: mesa-stable@lists.freedesktop.org
The lower layers assume that we support this, and it's been core since
GL 1.4. This fixes a slew of piglit tests, especially around
tex-miplevel-selection.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Cc: mesa-stable@lists.freedesktop.org
The hardware can actually generates vertexid when vertices come from
a client-side buffer like when glDrawElements is used.
This doesn't fix (or break) any piglit tests but it improves the
previous attempt of Ilia (c830d19 "nv50: avoid using inline vertex
data submit when gl_VertexID is used")
The only disadvantage is that only works on G84+, but we don't really
care of that weird and old NV50 chipset.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
W/UW immediates are 16-bits, but those 16-bits must be replicated
in the high 16-bits of the 32-bit field.
Remove the useless W/UW immediate saturating code, since we'll now be
using the appropriate immediate (and W/UW immediates in the IR can now
no longer be larger than 16-bits).
Reviewed-by: Emil Velikov <emil.velikov@collabora.co.uk>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Cuts 10k of .text, of which only 776 bytes are the fs_reg constructor
implementations themselves.
text data bss dec hex filename
5204535 214112 27784 5446431 531b1f i965_dri.so before
5193977 214112 27784 5435873 52f1e1 i965_dri.so after
Reviewed-by: Emil Velikov <emil.velikov@collabora.co.uk>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
This partially reverts commit bbf8239f92.
I didn't like that commit to begin with -- computing things at compile
time is fine -- but for purposes of verifying that the resulting values
are correct, looking up 0x00 and 0x30 in a table is a lot better than
evaluating a recursive function.
Anyway, by making brw_imm_vf4() take the actual 8-bit restricted floats
directly (instead of only integral values that would be converted to
restricted float), we can use this function as a replacement for the
vector float src_reg/fs_reg constructors.
brw_float_to_vf() is not currently an inline function, so it will not be
evaluated at compile time. I'll address that in a follow-up patch.
Reviewed-by: Emil Velikov <emil.velikov@collabora.co.uk>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Enable developers to know if the table's alphabetical sorting
is maintained or lost.
v2: Move "*" next to pointer name (Matt)
Include extensions_table.h instead of extensions.h (Ian)
Remove extra " *" in comment (Ian)
Signed-off-by: Nanley Chery <nanley.g.chery@intel.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Make it easier to determine where to add new extensions.
Performed with the vim sort command.
v2: Insert newline after last #define (Matt)
Signed-off-by: Nanley Chery <nanley.g.chery@intel.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
gl_MaxDualSourceDrawBuffersEXT - Maximum dual-source draw buffers supported
For ESSL 1.0, it provides two builtins since you can't have user-defined
color output variables:
gl_SecondaryFragColorEXT
gl_SecondaryFragDataEXT[MaxDSDrawBuffers]
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
This adds a state for the maximum dual source draw variables available
and the variable for determining if the extension has been enabled
in the program shaders.
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
I think the intention was to mark the "this" parameter as const, but
const goes on the other end to do that.
In file included from glsl_symbol_table.cpp:26:0:
ast.h:339:35: warning: type qualifiers ignored on function return type [-Wignored-qualifiers]
const bool is_single_dimension()
^
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Timothy Arceri <timothy.arceri@collabora.com>
This allows arbitrary non-constant indices on GS input arrays,
both for the vertex index, and any array offsets beyond that.
All indirects are handled via the pull model. We could potentially
handle indirect addressing of pushed data as well, but it would add
additional code complexity, and we usually have to pull inputs anyway
due to the sheer volume of input data. Plus, marking pushed inputs
as live due to indirect addressing could exacerbate register pressure
problems pretty badly. We'd need to be careful.
v2: Use updated MOV_INDIRECT opcode.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Abdiel Janulgue <abdiel.janulgue@linux.intel.com>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
- env GALLIUM_HUD_VISIBLE: control default visibility
- env GALLIUM_HUD_SIGNAL_TOGGLE: toggle visibility via signal
Signed-off-by: Marek Olšák <marek.olsak@amd.com>
This commit adds code for testing nir_shader_clone by running it after each
and every optimization pass and throwing away the old shader. Testing
nir_shader_clone is hidden behind a new INTEL_CLONE_NIR environment
variable.
Reviewed-by: Rob Clark <robclark@freedesktop.org>
Acked-by: Kenneth Graunke <kenneth@whitecape.org>
Failing to call nir_metadata_preserve() can have nasty consequences:
some pass breaks dominance information, but leaves it marked as valid,
causing some subsequent pass to go haywire and probably crash.
This pass adds a simple validation mechanism to ensure passes handle
this properly. We add a new bogus metadata flag that isn't used for
anything in particular, set it before each pass, and ensure it *isn't*
still set after the pass. nir_metadata_preserve will reset the flag,
so correct passes will work, and bad passes will assert fail.
(I would have made these functions static inline, but nir.h is included
in C++, so we can't bit-or enums without lots of casting...)
Thanks to Dylan Baker for the idea.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
OPT() is the normal macro for passes that return booleans, while OPT_V()
is a variant that works for passes that don't properly report progress.
(Such passes should be fixed to return a boolean, eventually.)
These macros take care of calling nir_validate_shader() and setting
progress appropriately. In the future, it would be easy to add shader
dumping similar to INTEL_DEBUG=optimizer by extending the macro.
v2 (Jason Ekstrand):
- Fix an unused variable warning
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
The a4xx bits corresponding to 'freedreno/a3xx: add fake RGTC support
(required for GL3)'
TODO some more r/e.. maybe we get lucky and hw supports some of this
directly? For now this will help us enable gl3.
Signed-off-by: Rob Clark <robclark@freedesktop.org>
The main issue is that the current logic looked into cso->u.tex, which
is the wrong side of the union to look into for texture buffers. While I
was at it, it was easy enough to add the logic to handle offsets
(first_element).
- reduce texture buffer size limit (determined experimentally)
- don't look at first/last levels, instead look at first/last element
- include the first element offset
- set offset alignment to 16 (determined experimentally)
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Signed-off-by: Rob Clark <robclark@freedesktop.org>
A smarter implementation would make it possible to attach this to emit
state for the BY_REGION versions to avoid breaking the tiling. But this
is a start.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Signed-off-by: Rob Clark <robclark@freedesktop.org>
Also throw in LATC while we're at it (same exact format). This could be
made more efficient by keeping a shadow compressed texture to use for
returning at map time. However... it's not worth it for now...
presumably compressed textures are not updated often.
Lastly fix up Z32S8 transfers to non-0 layers.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Signed-off-by: Rob Clark <robclark@freedesktop.org>
The previously RE'd formats were from an ES driver implementing
OES_vertex_type_10_10_10_2 and thus backwards. A future change could add
the 2_10_10_10 support.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Signed-off-by: Rob Clark <robclark@freedesktop.org>
We'd end up in a state where shader uses no inputs, yet num_elements is
greater than zero. Triggered by a TF vertex shader which did:
gl_Position = vec4(0.0, 0.0, 0.0, 0.0);
resulting in a binning pass variant with no inputs.
Includes equiv fix in a4xx, even though we don't have binning-pass
enabled yet on a4xx.
Signed-off-by: Rob Clark <robclark@freedesktop.org>
point_size_per_vertex is always TRUE for GLES, causing us to configure
the hw as if gl_PointSize was written, even if it was not. Which makes
for grumpy hw.
Signed-off-by: Rob Clark <robclark@freedesktop.org>
In nv50, and in the python script that Rob circulated, we do:
bld.mkCmp(OP_SET, CC_GE, TYPE_U32, (s = bld.getSSA()), TYPE_U32, m, b);
Do the same in the nir div lowering pass. This fixes the large-udiv-udiv
piglit tests on freedreno.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Cc: mesa-stable@lists.freedesktop.org
Signed-off-by: Rob Clark <robclark@freedesktop.org>
This patch disables the use of VSX instructions, as they cause some
piglit tests to fail
For more details, see: https://llvm.org/bugs/show_bug.cgi?id=25503#c7
With this patch, ppc64le reaches parity with x86-64 as far as piglit test
suite is concerned.
v2:
- Added check that we have at least LLVM 3.4
- Added the LLVM bug URL as a comment in the code
v3:
- Only disable VSX if Altivec is supported, because if Altivec support
is missing, then VSX support doesn't exist anyway.
- Change original patch description.
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
Cc: "11.0" <mesa-stable@lists.freedesktop.org>
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
3DSTATE_TE has partitioning, output topology, and domain fields,
each of which has several enumerated values. We'll also need to
switch on the domain, so enums (rather than #defines) seem like a
natural fit.
I chose to put these in brw_compiler.h because they'll be stored
in struct brw_tes_prog_data, which will live there.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
We always want to prefer the VGPU10 formats over the VGPU9 ones when
we have VGPU10 support.
Original patch by Jose and updated by Brian.
Reviewed-by: Charmaine Lee <charmainel@vmware.com>
Reviewed-by: José Fonseca <jfonseca@vmware.com>
This is important for the case of sampling from a depth texture. In
that case, we need to sample the texture as if it were a single-channel
color texture. For other/color formats, we can use the format as-is.
Reviewed-by: Charmaine Lee <charmainel@vmware.com>
Reviewed-by: José Fonseca <jfonseca@vmware.com>
The idea here is that driver queries implemented outside of common code
will use the same query buffer handling with different logic for starting
and stopping the corresponding counters.
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
[Fixed a rebase conflict and re-tested before pushing.]
Move r600_query and r600_query_hw into the header because we will want to
reuse the buffer handling and suspend/resume logic outside of the common
radeon code.
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
[Fixed a rebase conflict and re-tested before pushing.]
Software queries are all queries that do not require suspend/resume
and explicit handling of result buffers.
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
[Fixed a rebase conflict and re-tested before pushing.]
The goal here is to be able to move the implementation details of hardware-
specific queries (in particular, performance counters) out of the common code.
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
[Fixed a rebase conflict and re-tested before pushing.]
More query-related structures will have to be moved into their own
header file to support hardware-specific performance counters.
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
There are currently a bunch of formats that behave strangely when
sampling the cleared color from the MCS buffer on SKL. They seem to
mostly be formats that don't have an alpha component, although it's
not all of them, and we haven't yet found anything in the specs which
would explain this. For now to be on the safe side this patch just
prevents fast clears for MSRTs on SKL altogether so that when fast
clears are eventually enabled it will only be for single-sampled
surfaces. The assumption is that clears are probably more likely to be
used in single-sampled applications anyway so we can at least get them
working and we can enable MSRTs later once we understand the problem
better.
This patch should have no functional effect other than perhaps
receiving fewer perf_debug messages on SKL+.
v2: Improve the commit message to avoid saying the patch disables fast
clears because it will be merged before fast clears are enabled
for any surfaces so it doesn't actually disable anything.
Reviewed-by: Ben Widawsky <benjamin.widawsky@intel.com>
Reviewed-by: Chad Versace <chad.versace@intel.com>
DEQP likes to do math on uniforms, and the "fmaxabs dst, uni, uni" to get
the absolute value would get lowered. The lowering doesn't bother to try
to restrict the lifetime of the lowered uniforms, so we'd end up register
allocation failng due to this on 5 of the tests (More tests still fail in
RA, which look like we'll need to reduce lowered uniform lifetimes to
fix).
No changes on shader-db, though fewer extra MOVs are generated on even
glxgears (MOVs pair well enough that it ends up being the same instruction
count).
It looks like nir_lower_idiv is going to use it soon, so add support.
With Ilia's change, this fixes one case in fs-op-div-large-uint-uint (with
GL 3.0 forced on).
Cc: "11.0" <mesa-stable@lists.freedesktop.org>
This helps address a coverity warning and prevents future questions about this
code.
Reported-by: Coverity (via Ilia)
Cc: Ilia Mirkin <imirkin@alum.mit.edu>
Signed-off-by: Ben Widawsky <benjamin.widawsky@intel.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Checking that the flag has been set is all the validation thats
needed here.
Also not calling the binding validation function will make things
much simpler when adding compile time constant support as we
won't need to resolve the binding value.
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
Previously if the member was an array of matrices then a
warning message would be incorrectly given.
Also the struct case could never be met so it has been removed.
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
Previously we only gave the location for some members and never
gave the variable location. In those cases we were just giving
the location of the struct/block.
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
For struct and block members previously we were doing it for
every variable declaration.
So for example
struct S {
atomic_uint x, y, z;
};
Would previously generate three error messages when one is sufficient.
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
We now also only apply these rules to variables rather than also
trying to apply them to function params.
V2: move code for handling stream layout qualifier
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
Even though both tessellation shader stages must be used together, I
still think it makes sense to add separate debug flags for each stage.
It makes it possible to read the TCS/HS, rule out problems, then read
the TES/DS separately, without sifting through as much printed text.
I decided to add both the GL names (tcs/tes) and hardware names (hs/ds)
so they can be used interchangeably.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
This is needed for the FILE * type in brw_print_vue_map().
Apparently, all files that include brw_compiler.h already pick this up
via some include chain, so this isn't actually a build fix. However,
I have patches which introduce new consumers of brw_compiler.h that
fail to build because of the missing #include.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
v2: From Martin Peres
- Tell we are compiling the dri3 backend in configure.ac
- Update the Makefile.am
- get rid of the LIBDRM_HAS_RENDERNODE_SUPPORT macro
- fix some warnings related to EGLuint64KHR to int64_t conversions
- use dri2_get_dri_config to get the __DRIconfig instead of open-coding it
- replace the occasional tabs with spaces
v3: From Martin Peres
- fix and indent problem (Matt Turner)
- drop the authenticate function, use NULL in the vtable instead (Emil)
- drop some useless includes (Emil Velikov)
- mandate libdrm (Emil Velikov)
- link to xcb-dri3 (Kristian Høgsberg)
- convert to the new loader interface for drwable (Kristian)
- remove some dead code after the dropping of some vfuncs (Kristian)
- add a comment on the topic of rendering to the frontbuffer
v4: From Martin Peres
- do not expose the preserved swap behavior (Acked by Eric Anholt)
Signed-off-by: Boyan Ding <boyan.j.ding@gmail.com>
Signed-off-by: Martin Peres <martin.peres@linux.intel.com>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
Reviewed-by: Emil Velikov <emil.velikov@collabora.co.uk>
dri3 for EGL will use different struct other than dri2_egl_surface for
an EGL surface, the common code only uses __DRIdrawable from that
struct, so instead of converting _EGLSurface to dri2_egl_surface, let
the platform code return the __DRIdrawable by its own (although the
current platforms use the same function).
v2: From Martin Peres
- convert to the new drawable interface (Kristian)
Signed-off-by: Boyan Ding <boyan.j.ding@gmail.com>
Signed-off-by: Martin Peres <martin.peres@linux.intel.com>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
Reviewed-by: Emil Velikov <emil.velikov@collabora.co.uk>
v2: From Martin Peres
- convert to the new drawable interface
- delete dead code after the dropping of some vfuncs
- delete the width and height attributes since they are found in the helper
Signed-off-by: Boyan Ding <boyan.j.ding@gmail.com>
Signed-off-by: Martin Peres <martin.peres@linux.intel.com>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
Reviewed-by: Emil Velikov <emil.velikov@collabora.co.uk>
v2: From Martin Peres
- Try to fit in the 80-col limit as much as possible
v3: From Martin Peres
- introduce loader_dri3_helper.la to avoid dragging the xcb dep everywhere (Kristian & Emil)
- get rid of the width, height, dri_screen and is_different_gpu vfuncs (Kristian)
- replace the create/destroy functions with init/fini for dri3 drawables
- prefix static functions with dri3_ and exported ones with loader_dri3 (Emil)
- keep the function definition consistent (Emil)
Signed-off-by: Boyan Ding <boyan.j.ding@gmail.com>
Signed-off-by: Martin Peres <martin.peres@linux.intel.com>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
Reviewed-by: Emil Velikov <emil.velikov@collabora.co.uk>
brw_compile_gs() should return a pointer to unsigned, but it is returning the
bool 'false' at some point, hence annoying us with a compiler warning:
In function 'const unsigned int* brw::brw_compile_gs(const brw_compiler*,
void*, void*, const brw_gs_prog_key*, brw_gs_prog_data*, const nir_shader*,
gl_shader_program*, int, unsigned int*, char**)':
brw_vec4_gs_visitor.cpp:776:14: warning: converting 'false' to pointer type
'const unsigned int*' [-Wconversion-null]
return false;
^
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Precision qualifier should be ignored on desktop OpenGL.
v2: include spec quote (Samuel)
Signed-off-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
Up until now, we've been letting core Mesa initialize it to 36 for us
(which is presumably BRW_MAX_UBO (12) * (VS+GS+FS stages -> 3)).
With compute and tessellation, we need to increase this.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
This was getting pretty out of hand, and with compute partially in place
and tessellation on the way, it was only going to get worse.
This patch makes a "stage exists?" predicate and a "number of stages"
count and uses them to clean up a lot of calculations. We can just
loop over shader stages and set things for the ones that exist. For
combined counts, we can just multiply by the number of stages.
It also tries to organize a little bit.
We should probably use _mesa_has_geometry_shaders/tessellation/compute
here, but we can't because ctx->Version isn't initialized yet. Perhaps
that could be fixed in the future.
No change in "glxinfo -l" on Broadwell.
v2: Drop stray compute shader hunk. Mark stage_exists as const.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
I was going to add scalar_tcs and scalar_tes flags, and then thought
better of it and decided to convert this to an array. Simpler.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Since 779cabfc7d the same txformat table entries
are used for "normal" texturing as well as for blits. However, I forgot to put
in an entry for the bgrx8 (le) and xrgb8 (be) formats - the normal texturing
path can't hit them because the radeon tex format chooser will never chose
them, but we get that format from the dri buffers (at least I assume we got
it from there).
This is untested but essentially addressing the same bug as for radeon.
(I don't think that the second entry per le/be table is actually necessary,
but shouldn't hurt...)
Tested-by: Ian Romanick <ian.d.romanick@intel.com>
Acked-by: Alex Deucher <alexander.deucher@amd.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Cc: "11.0" <mesa-stable@lists.freedesktop.org>
Since d21320f625 the same txformat table entries
are used for "normal" texturing as well as for blits. However, I forgot to put
in an entry for the bgrx8 (le) and xrgb8 (be) formats - the normal texturing
path can't hit them because the radeon tex format chooser will never chose
them, but we get that format from the dri buffers (at least I assume we got
it from there). This caused lots of piglit regressions (and probably lots of
trouble outside piglit too).
This fixes bug https://bugs.freedesktop.org/show_bug.cgi?id=92900.
Tested-by: Ian Romanick <ian.d.romanick@intel.com>
Acked-by: Alex Deucher <alexander.deucher@amd.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Cc: "11.0" <mesa-stable@lists.freedesktop.org>
Previously GL_FRAMEBUFFER was used. However, if GL_EXT_framebuffer_blit
is supported (note: it is supported by every Mesa driver), this is
*sometimes* an alias for GL_DRAW_FRAMEBUFFER (getters) and *sometimes*
an alias for *both* GL_DRAW_FRAMEBUFFER and GL_READ_FRAMEBUFFER
(setters). As a result, the code saved one binding but modified both.
If the bindings were different, the GL_READ_FRAMEBUFFER would be
incorrect on exit.
Fixes the piglit fbo-generatemipmap-versus-READ_FRAMEBUFFER test.
Ideally this function would use DSA functions and not modify the binding
at all. However, that would be a much more intrusive change because
_mesa_meta_bind_fbo_image would also need to be modified.
_mesa_meta_bind_fbo_image has a lot of callers. Much of this code is
about to get a major rework due to bug #92363, so I don't think it
matters too much. In fact, I discovered this bug while working on the
other bug. Le bon temps!
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
Cc: "10.6 11.0" <mesa-stable@lists.freedesktop.org>
Replace the current loop by a direct call to _mesa_fls() function.
It also fixes an implicit bug in the current code where num_textures
seems to be one value less than it should be when sh->Program->SamplersUsed > 0.
For instance, num_textures is 0 instead of 1 when
sh->Program->SamplersUsed is 1.
Signed-off-by: Juan A. Suarez Romero <jasuarez@igalia.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
If a source operand in a MOV has source modifiers, then we cannot
copy-propagate it from the parent instruction and remove the MOV.
v2: remove the check for source modifiers from is_move() (Jason)
v3: Put the check for source modifiers back into is_move() since
this function is called from copy_prop_alu_src(). Add source
modifiers checks to is_vec() instead.
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
Only the regular "clear" call is supposed to respect the render
condition. The rest should ignore it.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
The geometry and tessellation control shader stages both read from
multiple URB entries (one per vertex). The thread payload contains
several URB handles which reference these separate memory segments.
In GLSL, these inputs are represented as per-vertex arrays; the
outermost array index selects which vertex's inputs to read. This
array index does not necessarily need to be constant.
To handle that, we need to use indirect addressing on GRFs to select
which of the thread payload registers has the appropriate URB handle.
(This is before we can even think about applying the pull model!)
This patch introduces a new opcode which performs a MOV from a
source using VxH indirect addressing (which allows each of the 8
SIMD channels to select distinct data.)
Based on a patch by Jason Ekstrand.
v2: Rename from INDIRECT_THREAD_PAYLOAD_MOV to MOV_INDIRECT; make it
a bit more generic. Use regs_read() instead of hacking up the
register allocator. (Suggested by Jason Ekstrand.)
v3: Fix regs_read() to be more accurate for small unaligned regions.
Also rebase on Matt's work.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com> [v3]
Reviewed-by: Abdiel Janulgue <abdiel.janulgue@linux.intel.com> [v1]
These compute-related MP performance counters have been reverse
engineered using CUPTI which is part of NVIDIA CUDA.
As for nvc0, we use a compute kernel to read out those performance
counters, and the command stream to configure them. Note that Tesla
only exposes 4 MP performance counters, while Fermi has 8.
Only G84+ is supported because G80 is an old and weird card.
Tested on G84, G96, G200, MCP79 and GT218 with glxgears, glxspheres64,
xonotic-glx, heaven and valley.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Tested-by: Pierre Moreau <pierre.morrow@free.fr>
Acked-by: Ilia Mirkin <imirkin@alum.mit.edu>
This adds the ability to launch simple compute kernels like the one I
will use to read out MP performance counters in the upcoming patch.
This compute support is based on the work of Francisco Jerez (aka curro)
that he did as part of his EVoC project in 2011/2012 to get OpenCL
working on Tesla. His original work can be found here:
https://github.com/curro/mesa/commits/nv50-compute
I did some improvements on the original code, like fixing using both 3D
and COMPUTE simultaneously, improving global buffers binding, and making
the code closer to what nvc0 already does. This compute support has been
tested by Pierre Moreau and myself with some compute kernels. This is a
step towards OpenCL.
Speaking about this, it seems like compute programs overlap fragment
programs when they are used both. To fix this, we need to re-validate
fragment programs when binding compute programs and vice versa.
Note that, textures, samplers and surfaces still need to be implemented.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Tested-by: Pierre Moreau <pierre.morrow@free.fr>
Acked-by: Ilia Mirkin <imirkin@alum.mit.edu>
As for nvc0, we need to free memory allocated by interpolation
parameters. This fixes a memory leak spotted by valgrind.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
No need to allocate more GPR than used in the compute kernel which
reads MP performance counters on Fermi.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
nir/nir_control_flow.c: In function ‘split_block_cursor.isra.11’:
nir/nir_control_flow.c:460:15: warning: ‘after’ may be used uninitialized in this function [-Wmaybe-uninitialized]
*_after = after;
^
nir/nir_control_flow.c:458:16: warning: ‘before’ may be used uninitialized in this function [-Wmaybe-uninitialized]
*_before = before;
^
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
We need to use per-slot offsets when there's non-uniform indexing,
as each SIMD channel could have a different index. We want to use
them for any non-constant index (even if uniform), as it lives in
the message header instead of the descriptor, allowing us to set
offsets in GRFs rather than immediates.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Abdiel Janulgue <abdiel.janulgue@linux.intel.com>
GLSL 4.00 and GL_ARB_gpu_shader5 introduced a new int -> uint implicit
conversion rule and updated the rules for modulus to use them. (In
earlier languages, none of the implicit conversion rules did anything
relevant, so there was no point in applying them.)
This allows expressions such as:
int foo;
uint bar;
uint mod = foo % bar;
Cc: mesa-stable@lists.freedesktop.org
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
I've been carrying around a patch to do this for the last few months,
and it's been exceedingly useful for debugging GS and tessellation
problems. I've caught lots of bugs by inspecting the interface
expectations of two adjacent stages.
It's not that much spam, so I figure we may as well just print it.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Acked-by: Matt Turner <mattst88@gmail.com>
This makes expressions like component(fs_reg(ATTR, n), 7) get a proper
<0,1,0> region instead of the invalid <0,8,0>.
Nobody uses this today, but I plan to.
v2: Rebase on Matt's changes; simplify.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com> [v1]
With the many variants of IO intrinsics, particular sources are often in
different locations. It's convenient to say "give me the indirect
offset" or "give me the vertex index" and have it just work, without
having to think about exactly which kind of intrinsic you have.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
We'd like to shadow these when possible, but the current code doesn't
work properly for TCS outputs. For now, disable it.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
Normally, we rely on nir_lower_outputs_to_temporaries to create shadow
variables for outputs, buffering the results and writing them all out
at the end of the program. However, this is infeasible for tessellation
control shader outputs.
Tessellation control shaders can generate multiple output vertices, and
write per-vertex outputs. These are arrays indexed by the vertex
number; each thread only writes one element, but can read any other
element - including those being concurrently written by other threads.
The barrier() intrinsic synchronizes between threads.
Even if we tried to shadow every output element (which is of dubious
value), we'd have to read updated values in at barrier() time, which
means we need to allow output reads.
Most stages should continue using nir_lower_outputs_to_temporaries(),
but in theory drivers could choose not to if they really wanted.
v2: Rebase to accomodate Jason's review feedback.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
Similar to nir_load_per_vertex_input, but for outputs. This is not
useful in geometry shaders, but will be useful in tessellation shaders.
v2: Change stage_uses_per_vertex_outputs() to is_per_vertex_output(),
taking a nir_variable (requested by Jason Ekstrand).
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
Tessellation control shader inputs are an array indexed by the vertex
number, like geometry shader inputs. There aren't per-patch TCS inputs.
Tessellation evaluation shaders have both per-vertex and per-patch
inputs. Per-vertex inputs get the new intrinsics; per-patch inputs
continue to use the ordinary load_input intrinsics, as they already
work like we want them to.
v2: Change stage_uses_per_vertex_inputs into is_per_vertex_input(),
which takes a variable (requested by Jason Ekstrand).
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
intel_asm_annotation.c: In function ‘annotation_insert_error’:
intel_asm_annotation.c:214:18:
warning: ‘ann’ may be used uninitialized in this function
[-Wmaybe-uninitialized]
ann->error = ralloc_strdup(annotation->mem_ctx, error);
^
I initially tried changing the type of ann_count to unsigned (is
currently int), since that in addition to the check that it's non-zero
at the beginning of the function seems sufficient to prove that it must
be greater than zero. Unfortunately that wasn't sufficient.
The first four values (2-bits) are hardware values, and VGRF, ATTR, and
UNIFORM remain values used in the IR.
Reviewed-by: Emil Velikov <emil.velikov@collabora.co.uk>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
HW_REGs are (were!) kind of awful. If the file was HW_REG, you had to
look at different fields for type, abs, negate, writemask, swizzle, and
a second file. They also caused annoying problems like immediate sources
being considered scheduling barriers (commit 6148e94e2) and other such
nonsense.
Instead use ARF/FIXED_GRF/MRF for fixed registers in those files.
After a sufficient amount of time has passed since "GRF" was used, we
can rename FIXED_GRF -> GRF, but doing so now would make rebasing awful.
Reviewed-by: Emil Velikov <emil.velikov@collabora.co.uk>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
The fs_reg() constructors for immediates set stride to 0, except for
vector-immediates, which set stride to 1. This patch makes the fs_reg
constructor that takes a brw_reg do likewise, so that stride is set
correctly for cases such as fs_reg(brw_imm_v(...)).
The generator asserts that this is true (and presumably it's useful in
some optimization passes?) and the VF fs_reg constructors did this (by
virtue of the fact that it doesn't override what init() does).
In the next commit, calling this constructor with brw_imm_* will generate
an IMM file register rather than a HW_REG, making this change necessary
to avoid breakage with existing uses of brw_imm_v().
Reviewed-by: Emil Velikov <emil.velikov@collabora.co.uk>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
We use brw_imm_v() to produce type-V immediates, which generates a
brw_reg with fs_reg's .file set to HW_REG. The next commit will rid us
of HW_REGs, so we need to handle BRW_REGISTER_TYPE_V in the IMM case.
Reviewed-by: Emil Velikov <emil.velikov@collabora.co.uk>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
The 2-bit hardware register file field is ARF, GRF, MRF, IMM.
Rename GRF to VGRF (virtual GRF) so that we can reuse the GRF name to
mean an assigned general purpose register.
Reviewed-by: Emil Velikov <emil.velikov@collabora.co.uk>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
I'm going to begin using brw_reg's file field in backend_reg and its
derivatives, and in order to keep the hardware value for ARF as 0, we
have to do something different.
Reviewed-by: Emil Velikov <emil.velikov@collabora.co.uk>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
The test (file == BAD_FILE) works on registers for which the constructor
has not run because BAD_FILE is zero. The next commit will move
BAD_FILE in the enum so that it's no longer zero.
In the case of this->outputs, the constructor was being run implicitly,
and we were unnecessarily memsetting is to zero.
Reviewed-by: Emil Velikov <emil.velikov@collabora.co.uk>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
In addition to combining another field, we get replace silliness like
"reg.reg" with something that actually makes sense, "reg.nr"; and no one
will ever wonder again why dst.reg isn't a dst_reg.
Moving the now 16-bit nr field to a 16-bit boundary decreases code size
by about 3k.
Reviewed-by: Emil Velikov <emil.velikov@collabora.co.uk>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Since backend_reg now inherits brw_reg, we can use it in place of the
fixed_hw_reg field.
Reviewed-by: Emil Velikov <emil.velikov@collabora.co.uk>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Put fields that are meaningless with an immediate in the same storage
with the immediate. This leaves fields type, file, nr, subnr in the
first dword where there's now extra room for expansion.
Reviewed-by: Emil Velikov <emil.velikov@collabora.co.uk>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Generated by
sed -i -e 's/\.bits\././g' *.c *.h *.cpp
sed -i -e 's/dw1\.//g' *.c *.h *.cpp
and then reverting changes to comments in gen7_blorp.cpp and
brw_fs_generator.cpp.
There wasn't any utility offered by forcing the programmer to list these
to access their fields. Removing them will reduce churn in future
commits.
This is C11 (and gcc has apparently supported it for sometime
"compatibility with other compilers")
See https://gcc.gnu.org/onlinedocs/gcc/Unnamed-Fields.html
Reviewed-by: Emil Velikov <emil.velikov@collabora.co.uk>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Switching from an implicitly-sized type field to field with an explicit
bit width is safe because we have fewer than 2^4 types, and gcc will
warn if you attempt to set a value that will not fit.
Reviewed-by: Emil Velikov <emil.velikov@collabora.co.uk>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Instead use the ones provided by brw_reg. Also allows us to handle
HW_REGs in the negate() functions.
Reviewed-by: Emil Velikov <emil.velikov@collabora.co.uk>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Since the types of the expression were
bool ? src_reg : (bool ? brw_reg : brw_reg)
the result of the second (nested) ternary would be implicitly
converted to a src_reg by the src_reg(struct brw_reg) constructor. I.e.,
bool ? src_reg : src_reg(bool ? brw_reg : brw_reg)
In the next patch, I make backend_reg (the parent of src_reg) inherit
from brw_reg, which changes this expression to return brw_reg, which
throws away any fields that exist in the classes derived from brw_reg.
I.e.,
src_reg(bool ? brw_reg(src_reg) : bool ? brw_reg : brw_reg)
Generally this code was gross, and wasn't actually shorter or easier to
read than an if ladder.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Chris Forbes <chrisf@ijw.co.nz>
This reduces the shader key for ES.
Use a fixed attrib location based on (semantic name, index).
The ESGS item size is determined by the physical index of the highest ES
output, so it's almost always larger than before, but I think that
shouldn't matter as long as the ESGS ring buffer is large enough.
Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>
I discovered that increasing the ESGS ring size fixes GS hangs on Tonga,
so let's do it properly.
There is now a separate init_config_gs_rings state that is not immutable,
because GS rings are resized when needed.
This also saves some memory. Most apps won't need more than 1MB
per ring per shader engine.
Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Not needed anymore. A similar flag will be introduced in the next commit,
which will be private in radeonsi.
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
need_cs_space isn't invoked so often and is called before all commands too.
This is a lot cleaner. The code in radeon_add_to_buffer_list always seemed
dodgy to me.
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
KCACHE, TC L1 and TC L2 are renamed to:
- SMEM L1
- VMEM L1
- GLOBAL L2
You can easily tell what they are used for now.
Shaders must deal with coherency issues between both L1s manually,
e.g. by setting GLC=1 or by using s_dcache_*.
BOTH_ICACHE_KCACHE was an unused definition.
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
This can't crash currently, but it would crash if clear_buffer
from u_blitter were used with a clean context.
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Discovered by luck. This code path hasn't been exercised since transform
feedback was implemented.
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Use a LIBDIR variable, set per-platform.
Update the Mesa configuration flags.
Run update-initramfs or dracut, update /etc/modules
Signed-off-by: Brian Paul <brianp@vmware.com>
eglSwapBuffersWithDamage accepts damage-region rectangles to hint the
compositor that it only needs to redraw certain areas, which was passed
through the wl_surface_damage request, as designed.
Wayland also offers a buffer transformation interface, e.g. to allow
users to render pre-rotated buffers. Unfortunately, there is no way to
query buffer transforms, and the damage region was provided in surface,
rather than buffer, co-ordinate space.
Users could in theory account for this themselves, but EGL also requires
co-ordinates to be passed in GL/mathematical co-ordinate space, with an
inversion to Wayland's natural/scanout co-ordinate space, so
transformations other than a 180-degree rotation will fail as EGL
attempts to subtract the region from (its view of the) surface height.
Pending creation and acceptance of a wl_surface.buffer_damage request,
which will accept co-ordinates in buffer co-ordinate space, pessimise to
always sending full-surface damage.
bce64c6c provides the explanation for why we send maximum-range damage,
rather than the full size of the surface: in the presence of buffer
transformations, full-surface damage may not actually cover the entire
surface.
Signed-off-by: Daniel Stone <daniels@collabora.com>
Reviewed-by: Pekka Paalanen <pekka.paalanen@collabora.co.uk>
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
The change proposed in the review leads to piglit regressions because
is_move() is used in other places and relies on the checks for source
modifiers to be there.
Revert this until we agree on a better solution.
Commit 8b28b35 added 'shared' as a keyword for compute shaders
but it broke the existing 'shared' layout qualifier support for
uniform and shader storage blocks.
This patch fixes 578 dEQP-GLES31.functional.ssbo.* tests.
v2:
- Move SHARED to interface_block_layout_qualifier (Timothy)
- Don't remove "shared" case insensitive check (Timothy)
- Remove the clearing of shared_storage flag (Timothy)
Signed-off-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
Reviewed-by: Timothy Arceri <timothy.arceri@collabora.com>
If a source operand in a MOV has source modifiers, then we cannot
copy-propagate it from the parent instruction and remove the MOV.
v2: remove the check for source source modifiers from is_move() (Jason)
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
This was a remnant of an early attempt to handle output reads in
vars_to_ssa. That attempt was abandon a long time ago but these few lines
were aparently left in the pass and managed to evade review.
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Previously, we walked through a given deref_node's copies and, after
lowering the copy away, removed it from both the source and destination
copy sets. This commit changes this to only remove it from the other
node's copy set (not the one we're lowering). At the end of the loop, we
just throw away the copy set for the node we're lowering since that node no
longer has any copies. This has two advantages:
1) It's more efficient because we're doing potentially half as many set
search operations.
2) It now properly handles copies from a node to itself. Perviously, it
would delete the copy from the set when processing the destinatioon and
then assert-fail when we couldn't find it for the source.
Cc: "11.0" <mesa-stable@lists.freedesktop.org>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=92588
Reviewed-by: Timothy Arceri <timothy.arceri@collabora.com>
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
The shader-subroutine code creates uniforms of type SUBROUTINE for
subroutines that are then read as integers in the backends. If we ever
want to do any optimizations on these, we'll need to come up with a better
plan where they are actual scalars or something, but this works for now.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=92859
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Mesa unconditionally sets this driver flag to true in
_mesa_init_extensions(). There is therefore no need for
the driver to communicate support for this extension.
Replace the driver capability flag with ::dummy_true.
Signed-off-by: Nanley Chery <nanley.g.chery@intel.com>
Reviewed-by: Chad Versace <chad.versace@intel.com>
This commit accidentally used a '==' when '=' was intended.
commit 96b22fb080
Author: Kristian Høgsberg Kristensen <krh@bitplanet.net>
Date: Wed Nov 4 14:58:54 2015 -0800
glsl: Use array deref for access to vector components
Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
Make API context and version checks done by the helper functions pass
unconditionally while meta is in progress. This transparently makes
extension checks solely dependent on struct gl_extensions while in meta.
v2: Use an 8-bit data type instead of a GLuint
Signed-off-by: Nanley Chery <nanley.g.chery@intel.com>
Reviewed-by: Chad Versace <chad.versace@intel.com>
Rename the following types and variables:
* struct extension -> struct mesa_extension,
like the mesa_format type.
* extension_table -> _mesa_extension_table,
like the _mesa_extension_override_{enables,disables} structs.
Suggested-by: Marek Olšák <marek.olsak@amd.com>
Suggested-by: Chad Versace <chad.versace@intel.com>
Signed-off-by: Nanley Chery <nanley.g.chery@intel.com>
Reviewed-by: Chad Versace <chad.versace@intel.com>
Generate functions which determine if an extension is supported in the
current context. Initially, enums were going to be explicitly used with
_mesa_extension_supported(). The idea to embed the function and enums
into generated helper functions was suggested by Kristian Høgsberg.
For performance, the function body no longer uses
_mesa_extension_supported() and, as suggested by Chad Versace, the
functions are also declared static inline.
v2: Place function qualifiers on separate line (Chad)
v3: Move function curly brace to new line (Chad)
Signed-off-by: Nanley Chery <nanley.g.chery@intel.com>
Reviewed-by: Chad Versace <chad.versace@intel.com>
The api_set field has no users outside of _mesa_extension_supported().
Remove it and allow the version field to take its place.
The brunt of the transformation was performed with the following vim commands:
s/\(GL [^,]\+\),\s*\d*,\s*\d*\(,\s*\d*\)\(,\s*\d*\)/\1, GLL, GLC\2\3/g
s/\(GLL [^,]\+\)\,\s*\d*/\1, GLL/g
s/\(GLC [^,]\+\)\(,\s*\d*\),\s*\d*\(,\s*\d*\)\(,\s*\d*\)/\1\2, GLC\3\4/g
s/\( ES1[^,]*\)\(,\s*\(\w\|\d\)\+\)\(,\s*\(\w\|\d\)\+\),\s*\d*/\1\2\4, ES1/g
s/\( ES2[^,]*\)\(,\s*\(\w\|\d\)\+\)\(,\s*\(\w\|\d\)\+\)\(,\s*\(\w\|\d\)\+\),\s*\d*/\1\2\4\6, ES2/g
Signed-off-by: Nanley Chery <nanley.g.chery@intel.com>
Reviewed-by: Chad Versace <chad.versace@intel.com>
Replace open-coded checks for extension support with
_mesa_extension_supported().
Signed-off-by: Nanley Chery <nanley.g.chery@intel.com>
Reviewed-by: Chad Versace <chad.versace@intel.com>
Create a function which determines if an extension is supported in the
current context.
v2: Use common variable names (Emil)
Insert new line between variables and return statement (Chad)
Rename api_set variable to api_bit (Chad)
Signed-off-by: Nanley Chery <nanley.g.chery@intel.com>
Reviewed-by: Chad Versace <chad.versace@intel.com>
Enable limiting advertised extension support by context version with
finer granularity. This new field is currently unused and is set to
0 everywhere. When it is used, a value of 0 will indicate that the
extension is supported for any version of a context.
v2: Use uint*t type for version and note the expected values (Emil)
Use an 8-bit data type
Reformat macro for better readability (Chad)
v3: Note preparatory nature of commit (Chad)
Signed-off-by: Nanley Chery <nanley.g.chery@intel.com>
Reviewed-by: Chad Versace <chad.versace@intel.com>
With this infrastructure set in place, we can now reuse the entries to
generate useful code.
v2: Add the new file into Makefile.sources (Emil)
Signed-off-by: Nanley Chery <nanley.g.chery@intel.com>
Reviewed-by: Chad Versace <chad.versace@intel.com>
Now that we're using macros, remove the redundant text from each entry.
Remove comments between the entries to make editing easier and separate
the sections with blank lines. Structure the EXT macros in a way that
helps reviewers verify that no meaning has been altered.
v2: Indent the entries (Chad)
Signed-off-by: Nanley Chery <nanley.g.chery@intel.com>
Reviewed-by: Chad Versace <chad.versace@intel.com>
Simplify future updates to the extension struct array by removing
the sentinel.
Signed-off-by: Nanley Chery <nanley.g.chery@intel.com>
Reviewed-by: Chad Versace <chad.versace@intel.com>
Initially just checks that sources are non-NULL, which would have
alerted us to the problem fixed by commit 6c846dc5.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Will allow annotations to contain error messages (indicating an
instruction violates a rule for instance) that are printed after the
disassembly of the block.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Often annotations are identical between sets of consecutive
instructions. We can perhaps avoid some memory allocations by reusing
the previous annotation.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
And why did IFF have a destination?
I suspect that once upon a time the disassembler used this information
to know which fields to find the jump targets in. The jump targets have
moved, so the disassembler has to know how to handle these
per-generation anyway.
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Add some instructions: illegal, movi, sends, sendsc.
Remove some instructions with reused opcodes: msave, mrestore, push,
pop, goto. I did have some gross code for disassembling opcodes
per-generation, but there's very little meaningful overlap so it's
probably not needed.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
This would have caught the locking bug that was fixed in the earlier
"st/wgl: fix locking issue in stw_st_framebuffer_present_locked()"
patch.
v2: minor coding style changes by Brian.
Reviewed-by: Sinclair Yeh <syeh@vmware.com>
Reviewed-by: Charmaine Lee <charmainel@vmware.com>
v2: update comments on the stw_framebuffer::mutex field regarding locking
order.
Reviewed-by: Sinclair Yeh <syeh@vmware.com>
Reviewed-by: Charmaine Lee <charmainel@vmware.com>
To get declaration for debug_printf() directly instead of getting it
indirectly through os_thread.h
Reviewed-by: Sinclair Yeh <syeh@vmware.com>
Reviewed-by: Charmaine Lee <charmainel@vmware.com>
This is Windows-only code so we can use the native Win32 functions for
critical sections. This will also allow us to (cleanly) add some mutex
check/debug code in subsequent patches.
Reviewed-by: Sinclair Yeh <syeh@vmware.com>
Reviewed-by: Charmaine Lee <charmainel@vmware.com>
We support "cpu" but not "cpu#" because there's no good way of querying
per-cpu usage. Also, the cpu usage is for the process, not the whole
system.
Original code cobbled together by Brian and then fixed/polished by Jose.
Signed-off-by: Brian Paul <brianp@vmware.com>
Patch sets matrix_stride as 0 for non matrix uniforms that are in a
atomic counter buffer. Matrix stride calculation for actual matrix
uniforms is done during link_assign_uniform_locations.
From ARB_program_interface_query specification:
GL_MATRIX_STRIDE:
"For active variables not declared as a matrix or array of matrices,
zero is written to <params>. For active variables not backed by a
buffer object, -1 is written to <params>, regardless of the variable
type."
Signed-off-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Marta Lofstedt <marta.lofstedt@intel.com>
This information will be used by cross stage validation of varyings
for pipeline objects.
Signed-off-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
We will need this later on when we implement proper support for
precision qualifiers in the drivers and also to do link time checks for
uniforms as indicated by the spec.
This patch also adds compile-time checks for variables without precision
information (currently, Mesa only checks that a default precision is set
for floats in fragment shaders).
As indicated by Ian, the addition of the precision information to
ir_variable has been done using a bitfield and pahole to identify an
available hole so that memory requirements for ir_variable stay the
same.
v2 (Ian):
- Avoid if-ladders by defining arrays of supported sampler names and
indexing
into them with type->sampler_array + 2 * type->sampler_shadow
- Make the code that selects the precision qualifier to use an utility
function
- Fix a typo
v3 (Tapani):
- rebased
- squashed in "Precision qualifiers are not allowed on structs"
- fixed select_gles_precision for sampler arrays
- fixed precision_qualifier_allowed for arrays of structs
v4 (Tapani):
- add atomic_uint handling
- do not allow precision qualifier on images
(issues reported by Marta)
v5 (Tapani):
- support precision qualifier on image types
v6 (Tapani):
- set precision qualifier on interface block members
Reviewed-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
Notice that the spec requires that a default precision has been set for every
type used by a shader that can use a precision qualifier and does not have a
predefined precision, however, at the moment, Mesa only checks this for floats
in the fragment shader. This is probably because the GLSL ES 1.0 specs mentions
this case specifically, but GLSL ES 3.0 clarifies that the same applies to
other types:
"The fragment language has no default precision qualifier for floating point
types. Hence for float, floating point vector and matrix variable
declarations, either the declaration must include a precision qualifier or
the default float precision must have been previously declared. Similarly,
there is no default precision qualifier for the following sampler types in
either the vertex or fragment language:
sampler3D;
samplerCubeShadow;
sampler2DShadow;
sampler2DArray;
sampler2DArrayShadow;
isampler2D;
isampler3D;
isamplerCube;
isampler2DArray;
usampler2D;
usampler3D;
usamplerCube;
usampler2DArray;"
we will fix this in a later patch.
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
The GLSL ES spec specifies default precision qualifiers for certain types,
so populate the symbol table with these.
Notice that the desktop GLSL spec also indicates defaults for some types
but this is not really useful since precision qualifiers are completely
ignored in desktop GLSL.
v2: simplify and add samplerExternalOES, specified by
OES_EGL_image_external (Tapani)
v3: add atomic_uint (reported missing by Marta)
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
These have scoping rules that match the ones defined for other things such
as variables, so we want them in the symbol table.
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
FS_OPCODE_GET_BUFFER_SIZE is calculated with a resinfo's sampler message.
This patch adjusts the number of registers written by the opcode
following what the PRM spec says about the number of registers written
by the SIMD8 and SIMD16's writeback messages for sampler messages.
Signed-off-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
The comment in the code details the restriction. Thanks to Ken for having a very
helpful conversation with me, and spotting the blurb in the link I sent him :P.
There are still stability problems for me on GT4, but this definitely helps with
some of the failures.
v2: Comment fixes
Cc: mesa-stable@lists.freedesktop.org
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Many intrinsics only apply to a particular stage (such as discard).
In other cases, we may want to interpret them differently based on
the stage (such as load_primitive_id or load_input).
The current method isn't that pretty - we handle all intrinsics in
one giant function. Sometimes we assert on stage, sometimes we forget.
Different behaviors are handled via if-ladders based on stage.
This commit introduces new nir_emit_<stage>_intrinsic() functions,
and makes nir_emit_instr() call those. In turn, those fall back to
the generic nir_emit_intrinsic() function for cases they don't want
to handle specially.
This makes it clear which intrinsics only exist in one stage, and makes
it easy to handle inputs/outputs differently for various stages.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
For compressed textures, the image size is not necessarily a multiple of
the block size (e.g. the last mip levels). Section 18.3.2 (Copying
Between Images) of the OpenGL 4.5 Core Profile spec says:
An INVALID_VALUE error is generated if the dimensions of either
subregion exceeds the boundaries of the corresponding image
object, or if the image format is compressed and the dimensions of
the subregion fail to meet the alignment constraints of the
format.
and Section 8.7 (Compressed Texture Images) says:
An INVALID_OPERATION error is generated if any of the following
conditions occurs:
* width is not a multiple of four, and width + xoffset is not
equal to the value of TEXTURE_WIDTH.
* height is not a multiple of four, and height + yoffset is not
equal to the value of TEXTURE_HEIGHT.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=92860
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Acked-by: Alex Deucher <alexander.deucher@amd.com>
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Cc: mesa-stable@lists.freedesktop.org
Just a minor code change to make it obvious that NULL is returned when
we don't find the given HWND.
Reviewed-by: Sinclair Yeh <syeh@vmware.com>
Reviewed-by: Charmaine Lee <charmainel@vmware.com>
Reviewed-by: José Fonseca <jfonseca@vmware.com>
When stw_st_framebuffer_present_locked() is called, the
stw_framebuffer's mutex will already be locked. Normally, the
stw_framebuffer_present_locked() function calls
stw_framebuffer_release() to unlock the mutex when it's done. But if
for some reason the 'resource' pointer in
stw_st_framebuffer_present_locked() is null, we'd return without
unlocking the stw_framebuffer. This fixes that to avoid potential
deadlocks.
Reviewed-by: Charmaine Lee <charmainel@vmware.com>
A while back, we moved to directly emitting the Gen7+ state when
constructing the binding tables. These flags are only used on
Gen4-6, which emit all the binding table pointers at once.
We gain nothing by having separate flags, so combine them.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
Inspired by a patch by Fabian Bieler.
Fabian defined a _3DPRIM_PATCHLIST_0 macro (which isn't actually a valid
topology type); I instead chose to make a macro that takes an argument.
He also took the number of patch vertices from _mesa_prim (which was set
to ctx->TessCtrlProgram.patch_vertices) - I chose to use it directly to
avoid the need for the VBO patch.
v2: Change macro to 0x20 + (n - 1) instead of 0x1F + n to better match
the documentation (suggested by Ian).
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
When both fadd and fmul instructions have at least one operand that is a
constant and it is only used once, the total number of instructions can
be reduced from 3 (1 ffma + 2 load_const) to 2 (1 fmul + 1 fadd); because
the constants will be progagated as immediate operands of fmul and fadd.
This patch detects these situations and prevents fusing fmul+fadd into ffma.
Shader-db results on i965 Haswell:
total instructions in shared programs: 6235835 -> 6225895 (-0.16%)
instructions in affected programs: 1124094 -> 1114154 (-0.88%)
total loops in shared programs: 1979 -> 1979 (0.00%)
helped: 7612
HURT: 843
GAINED: 4
LOST: 0
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
Because the next patch will add an optimization that is specific to i965,
we want to move this loweing pass to that driver altogether.
This is safe because i965 is the only consumer.
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
We've assumed that we could lower per-component vector access from
vec[i] = scalar
to
vec = ir_triop_vector_insert(vec, scalar, i)
but with SSBOs (and compute shader SLM and tesselation outputs) this is
no longer valid. If a vector is "externally visible", multiple threads
can write independent components simultaneously. With lowering to
ir_triop_vector_insert, each thread read the entire vector, changes one
component, then writes out the entire vector. This is racy.
Instead of generating a ir_binop_vector_extract when we see v[i], we
generate ir_dereference_array. We then add a lowering pass to lower the
ir_dereference_array to ir_binop_vector_extract for rvalues and for to
vector_insert for lvalues in a separate lowering pass.
The resulting IR is the same as before, but we now have a window between
ast->ir conversion and the lowering pass where v[i] appears in the IR as
an array deref. This lets us run lowering passes that lower the vector
access to I/O (eg for SSBO load/store) before we lower the per-component
access to full vector writes.
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Signed-off-by: Kristian Høgsberg Kristensen <krh@bitplanet.net>
All GLSL IR consumers run this lowering pass so we can move it to the
linker. This moves the pass up quite a bit, but that's the point: it
needs to run before we throw away information about per-component vector
access.
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Signed-off-by: Kristian Høgsberg Kristensen <krh@bitplanet.net>
We always pass in shader->ir and we already pass in the shader, so just
drop the exec_list. Most passes either take just a exec_list or a
shader, so this seems more consistent.
Reviewed-by: Timothy Arceri <timothy.arceri@collabora.com>
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Signed-off-by: Kristian Høgsberg Kristensen <krh@bitplanet.net>
Its only user now returns a nir_ssa_def *, and we'll need this since the
builder returns a nir_ssa_def *.
Signed-off-by: Connor Abbott <cwabbott0@gmail.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
A long time ago, before NIR was even merged to master, glsl_to_nir used
registers and these sources were actually register sources. But nowadays
everything in glsl_to_nir is an SSA value, so stop pretending that by
evaluating an rvalue we can get an arbitrary nir_src. Most importantly,
we need this since the builder takes nir_ssa_def * sources directly.
Signed-off-by: Connor Abbott <cwabbott0@gmail.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Ideally we should have a _mesa_cleanup_buffer_object function in
src/mesa/bufferobj.c so that the destruction logic resided in a single
place.
Reviewed-by: Brian Paul <brianp@vmware.com>
These tessellation shader related fields need plumbing through NIR.
v2: Use uint32_t instead of uint64_t to match the source type of
GLbitfield (caught by Iago Toral).
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Since X has undefined contents in new pixmaps, it will allocate new
textures for an FBO and draw to them without an explicit clear. For
VC4, it's much faster to emit a clear than the load of the actual
undefined memory contents, so just do that instead.
I'm not sure what the caller does is appropriate (just have a NULL sampler
at this slot), but it fixes the immediate crash.
Cc: "11.0" <mesa-stable@lists.freedesktop.org>
I was afraid our callers weren't prepared for this, but it looks like
at least for resource creation, mesa/st throws an error appropriately.
Cc: "11.0" <mesa-stable@lists.freedesktop.org>
Shared variables are stored in a common pool accessible by all threads
in a compute shader local work group.
These variables are similar to OpenCL's local/__local variables.
Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Timothy Arceri <timothy.arceri@collabora.com>
v2:
* Move shared parsing under storage qualifiers (tarceri)
* Fail to compile if shared is used in non-compute shader (tarceri)
* Use separate shared_storage bit for shared variables (tarceri)
Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Timothy Arceri <timothy.arceri@collabora.com>
Qualifiers on member variables are redundent all we need to do
if check if it matches the stream associated with the block and
throw an error if its not.
Reviewed-by: Samuel Iglesias Gonsalvez <siglesias@igalia.com>
Cc: Emil Velikov <emil.l.velikov@gmail.com>
The stw_st_framebuffer_present_locked() function was getting called
twice per SwapBuffers. First, when st_context_iface::flush() was
called from DrvSwapBuffers() because the ST_FLUSH_FRONT flag was
given. Second, by stw_st_swap_framebuffer_locked() which does the
actual SwapBuffers.
Two code changes:
1. Pass ST_FLUSH_END_OF_FRAME, instead of ST_FLUSH_FRONT.
2. Move the implementation of stw_flush_current_locked() into
DrvSwapBuffers() since it's not called anywhere else.
Not much change in perf for benchmarks like Lightsmark, but some simple
Mesa demos are measurably faster.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
And put 8-bit/channel formats before 5/6/5 formats.
The ChoosePixelFormat() function seems to be finicky about format
selection. Putting the MSAA formats after the non-MSAA formats
means most apps get a low-numbered format. Now we generally get
the same pixel format regardless of whether using vgpu9 or 10.
VMware bug 1455030
Reviewed-by: José Fonseca <jfonseca@vmware.com>
This allows to use apitrace's retracediff script on Windows to retrace and
compare two builds of a Mesa based opengl32.dll/ICD side-by-side.
See also e4a4f15f5b
This will allow dec/enc/transcode without X
v2: use env override even with X,
use loader_open_device instead of open
v3: clean up
Signed-off-by: Leo Liu <leo.liu@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
This will allow the state trackers to use render nodes
with screen creation
v2: dup fd for pipe loader
Signed-off-by: Leo Liu <leo.liu@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
There is no dev in drv, and dev should be from vl_screen here
Signed-off-by: Leo Liu <leo.liu@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Altough the compute support is still not complete because textures and
surfaces need to be implemented, it allows to launch very simple compute
kernel like one which reads reading MP performance counters.
This turns on PIPE_CAP_COMPUTE and PIPE_SHADER_COMPUTE.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
There might only be a single arg (e.g. cvt), so use mode rather than
looking at the source directly. Also we don't want to rely on the type
of the value, which can be unreliable, but instead use the
instruction's. This works out well since mkSplit doesn't adjust the
type.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Not reachable from TGSI since it only has UMUL, no IMUL. However it's
surprising that setting argument types to s32 will cause sign to get
lost.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Force the fence to get kicked off, which won't actually wait for its
completion, but any additional work will be put onto a fresh list.
This fixes crashes in teximage-colors --benchmark with too many active
maps.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
As pointed out by Emil, this sometimes hangs, appears to be due to threading
need to rethink how this stuff works for llvmpipe.
Signed-off-by: Dave Airlie <airlied@redhat.com>
Coverity reported that ret could only be 0 or 1, since it
was setting ret = fn() > 0, instead of doing (ret = fn()) > 0.
Signed-off-by: Dave Airlie <airlied@redhat.com>
Previously, we were assuming that everything read/wrote exactly 1 logical
GRF (1 in SIMD8 and 2 in SIMD16). This isn't actually true. In
particular, the PLN instruction reads 2 logical registers in one of the
components. This commit changes post-RA scheduling to use regs_read and
regs_written instead so that we add enough dependencies.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=92770
Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Now that we support 64 bit immediates in insnCanLoad, we need to swap
64 bit immediate sources too for optimal effect.
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
Teach insnCanLoad about double immediates, together with the
"Add support for merge-s to the ConstantFolding pass"
This turns the following (nvc0) code:
1: mov u32 $r2 0x00000000 (8)
2: mov u32 $r3 0x3fe00000 (8)
3: add f64 $r0d $r0d $r2d (8)
Into:
1: add f64 $r0d $r0d 0.500000 (8)
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
This allows later passes like LoadPropagation to properly deal with 64
bit immediates.
If the new 64 bit load this introduces does not get optimized away then
split64BitOpPostRA() will split this into 2 instructions again.
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
Add support for encoding double immediates (up to 20 bits of precision)
into the generated gm107 machine-code.
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
Add support for encoding double immediates (up to 20 bits of precision)
into the generated nvc0 machine-code.
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
For these nir intrinsics, we emit the same code as
nir_intrinsic_memory_barrier:
* nir_intrinsic_memory_barrier_atomic_counter
* nir_intrinsic_memory_barrier_buffer
* nir_intrinsic_memory_barrier_image
We treat these nir intrinsics as no-ops:
* nir_intrinsic_group_memory_barrier
* nir_intrinsic_memory_barrier_shared
v3:
* Add comment for no-op cases (curro)
v4:
* Moving comment to a separate patch authored by curro
Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Francisco Jerez <currojerez@riseup.net>
When these functions are called in glsl-ir, we create a corresponding
nir intrinsic function call.
Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Francisco Jerez <currojerez@riseup.net>
When these functions are called in GLSL code, we create an intrinsic
function call:
* groupMemoryBarrier => __intrinsic_group_memory_barrier
* memoryBarrierAtomicCounter => __intrinsic_memory_barrier_atomic_counter
* memoryBarrierBuffer => __intrinsic_memory_barrier_buffer
* memoryBarrierImage => __intrinsic_memory_barrier_image
* memoryBarrierShared => __intrinsic_memory_barrier_shared
v2:
* Consolidate with memoryBarrier function/intrinsic creation (curro)
v3:
* Instead of add_memory_barrier_function, add an intrinsic_name
parameter to _memory_barrier (curro)
Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Francisco Jerez <currojerez@riseup.net>
Before it was only possible to convert a NV12 surface to
RGBA or BGRA. This patch uses the same post processing
function, "handleVAProcPipelineParameterBufferType", but
add definitions for RGBX and BGRX.
This patch also makes vlVaQuerySurfaceAttributes more generic
to avoid copy and pasting the same lines.
Signed-off-by: Julien Isorce <j.isorce@samsung.com>
Reviewed-by: Christian K<C3><B6>nig <christian.koenig@amd.com>
Reviewed-by: Emil Velikov <emil.l.velikov@gmail.com>
Useful is one wants to create RGBX or BGRX surfaces.
The infrastructure is such that it required just a
few definitions to support these formats.
Signed-off-by: Julien Isorce <j.isorce@samsung.com>
Reviewed-by: Christian K<C3><B6>nig <christian.koenig@amd.com>
Reviewed-by: Emil Velikov <emil.l.velikov@gmail.com>
The various cf nodes all get allocated w/ shader as their ralloc_parent,
so lets make this more explicit. Plus couple other corrections/
clarifications.
Signed-off-by: Rob Clark <robclark@freedesktop.org>
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
Commit 342e68dc60 (nvc0: remove BGRA4 format support) removed the
support to fix a WoW trace. However after further experimentation, I was
able to get the blit to work by using a different "fake" format in the
2d engine.
The reason why this worked on nv50 is that nv50 falls back to the 3d
blit path in case either the src or the dst aren't "faithfully"
supported, while nvc0 only does it for the dst format. RG8 is better
supported by the nvc0 2d engine than R16.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
The scalar VS backend has never handled float[] and vec2[] outputs
correctly (my original code was broken). Outputs need to be padded
out to vec4 slots.
In fs_visitor::nir_setup_outputs(), we tried to process each vec4 slot
by looping from 0 to ALIGN(type_size_scalar(type), 4) / 4. However,
this is wrong: type_size_scalar() for a float[2] would return 2, or
for vec2[2] it would return 4. This looked like a single slot, even
though in reality each array element would be stored in separate vec4
slots.
Because of this bug, outputs[] and output_components[] would not get
initialized for the second element's VARYING_SLOT, which meant
emit_urb_writes() would skip writing them. Nothing used those values,
and dead code elimination threw a party.
To fix this, we introduce a new type_size_vec4_times_4() function which
pads array elements correctly, but still counts in scalar components,
generating correct indices in store_output intrinsics.
Normally, varying packing avoids this problem by turning varyings into
vec4s. So this doesn't actually fix any Piglit or dEQP tests today.
However, if varying packing is disabled, things would be broken.
Tessellation shaders can't use varying packing, so this fixes various
tcs-input Piglit tests on a branch of mine.
v2: Shorten the implementation of type_size_4x to a single line (caught
by Connor Abbott), and rename it to type_size_vec4_times_4()
(renaming suggested by Jason Ekstrand). Use type_size_vec4
rather than using type_size_vec4_times_4 and then dividing by 4.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
This will allow gallium drivers to send messages to KHR_debug endpoints
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Do it in the visitor, like we do for other opcodes.
v2: use const, get rid of useless surf_index temporary (Curro)
Reviewed-by: Francisco Jerez <currojerez@riseup.net>
Do it in the visitor, like we do for other opcodes.
v2: use const, get rid of useless surf_index temporary (Curro)
Reviewed-by: Francisco Jerez <currojerez@riseup.net>
Right now the generator marks direct surfaces as used but leaves marking of
indirect surfaces to the caller. Just make the callers handle marking in both
cases for consistency.
v2: Use const, do not add unnecessary temporary (Curro)
Reviewed-by: Francisco Jerez <currojerez@riseup.net>
Right now the generator marks direct surfaces as used but leaves marking of
indirect surfaces to the caller. Just make the callers handle marking in both
cases for consistency.
Reviewed-by: Francisco Jerez <currojerez@riseup.net>
Right now the generator marks direct surfaces as used but leaves marking of
indirect surfaces to the caller. Just make the callers handle marking in both
cases for consistency.
v2: Use const and remove useless surf_index temporary (Curro)
Reviewed-by: Francisco Jerez <currojerez@riseup.net>
Previously there was a problem in i965 where if 16x MSAA is used then
some of the sample positions are exactly on the 0 x or y axis. When
the MSAA copy blit shader interpolates the texture coordinates at
these sample positions it was possible that it would jump to a
neighboring texel due to rounding errors. It is likely that these
positions would be used on 16x MSAA because that is where they are
defined to be in D3D.
To fix that this patch makes it use interpolateAtOffset in the blit
shader whenever 16x MSAA is used and the GL_ARB_gpu_shader5 extension
is available. This forces it to interpolate the texture coordinates at
the pixel center to avoid these problematic positions.
This fixes ext_framebuffer_multisample-unaligned-blit and
ext_framebuffer_multisample-clip-and-scissor-blit with 16x MSAA on
SKL+.
v2: Use interpolateAtOffset instead of interpolateAtSample
v3: Always try to enable GL_ARB_gpu_shader5 in the shader
[Ian Romanick]
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
Previously this extension was only enabled when blitting between two
multisampled buffers. However I don't think it does any harm to just
enable it all the time. The ‘enable’ option is used instead of
‘require’ so that the shader will still compile if the extension isn't
available in the cases where it isn't used. This will make the next
patch simpler because it wants to add another optional extension.
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
The destination rectangle is now drawn at 4x4 the size and the shader
code to calculate the sample number is adjusted accordingly.
Acked-by: Ben Widawsky <ben@bwidawsk.net>
In order to accomodate 16x MSAA, the starting sample pair index is now
3 bits rather than 2 on SKL+.
Reviewed-by: Ben Widawsky <ben@bwidawsk.net>
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
The gen7_surface_msaa_bits function already returns the right values
for 16 samples but it just needs its assert to be relaxed.
Reviewed-by: Ben Widawsky <ben@bwidawsk.net>
When 16x MSAA is used for sampling with texelFetch the compiler needs
to use a different instruction which passes more arguments for the MCS
data. Previously on skl+ it was unconditionally using this new
instruction. However since 16x MSAA is probably going to be pretty
rare, it is probably worthwhile to avoid using this instruction for
the other sample counts. In order to do that this patch adds a new
member to brw_sampler_prog_key_data to track when a sampler refers to
a buffer with 16 samples.
Note that this isn't done for the vec4 backend because it wouldn't
change how many registers it uses.
Acked-by: Ben Widawsky <ben@bwidawsk.net>
In order to support 16x MSAA, skl+ has a wider version of ld2dms that
takes two parameters for the MCS data. The MCS data in the response
still fits in a single register so we just need to ensure we copy both
values rather than just the lower one.
Acked-by: Ben Widawsky <ben@bwidawsk.net>
In order to support 16x MSAA, skl+ has a wider version of ld2dms that
takes two parameters for the MCS data. The MCS data retrieved from the
ld_mcs instruction already returns 4 or 8 registers and is documented
to return zeroes for the mcsh value when the sample count is less than
16.
v2: Use get_lowered_simd_width to fall back to SIMD8 instructions when
the message length would be too long in SIMD16.
Reviewed-by: Ben Widawsky <ben@bwidawsk.net>
This is the standard pattern used by the other 3D graphics API.
BDW has slots for these values, but they aren't actually used until
SKL. Even though the documentation for BDW says they must be zero, it
doesn't seem to cause any harm to program them anyway.
The comment above for the 8x sample positions says that the hardware
implements centroid interpolation by picking the centre-most sample
that is inside the primitive. That implies that it might be worthwhile
to pick a pattern that includes 0.5,0.5. However by experimentation
this doesn't seem to actually be the case. With the sample positions
in this patch, if I modify the piglit test below so that it instead
reports the centroid position, it reports 0.492188,0.421875 which
doesn't match any of the positions. If I modify the sample positions
so that they include one at exactly 0.5,0.5 it doesn't help and it
reports another position which is even further from the center for
some reason.
arb_gpu_shader5-interpolateAtSample-different
Kenneth Graunke experimented with some other patterns that have a
higher standard deviation but I think after some discussion it was
decided that it would be better to pick the same pattern as the other
graphics API in case there are games that rely on this pattern.
(Based on a patch by Kenneth Graunke)
Cc: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ben Widawsky <ben at bwidawsk.net>
Equivalent to commit 8ac3b525c but with sel operations. In this case
we select the PredCtrl based on the writemask.
This patch helps on cases like this:
1: cmp.l.f0.0 vgrf40.0.x:F, vgrf0.zzzz:F, vgrf7.xxxx:F
2: cmp.nz.f0.0 null:D, vgrf40.xxxx:D, 0D
3: (+f0.0) sel vgrf41.0.x:UD, vgrf6.xxxx:UD, vgrf5.xxxx:UD
In this case, cmod propagation can't optimize instruction #2, because
instructions #1 and #2 have different writemasks, and we can't update
directly instruction #2 writemask because our code thinks that sel at
instruction #3 reads all four channels of the flag, when it actually
only reads .x.
So, with this patch, the previous case becames this:
1: cmp.l.f0.0 vgrf40.0.x:F, vgrf0.zzzz:F, vgrf7.xxxx:F
2: cmp.nz.f0.0 null:D, vgrf40.xxxx:D, 0D
3: (+f0.0.x) sel vgrf41.0.x:UD, vgrf6.xxxx:UD, vgrf5.xxxx:UD
Now only the x channel of the flag is used, allowing dead code
eliminate to update the writemask at the second instruction:
1: cmp.l.f0.0 vgrf40.0.x:F, vgrf0.zzzz:F, vgrf7.xxxx:F
2: cmp.nz.f0.0 null.x:D, vgrf40.xxxx:D, 0D
3: (+f0.0.x) sel vgrf41.0.x:UD, vgrf6.xxxx:UD, vgrf5.xxxx:UD
So now cmod propagation can simplify out #2:
1: cmp.l.f0.0 vgrf40.0.x:F, attr18.wwww:F, vgrf7.xxxx:F
2: (+f0.0.x) sel vgrf41.0.x:UD, vgrf6.xxxx:UD, vgrf5.xxxx:UD
Shader-db numbers:
total instructions in shared programs: 6235835 -> 6228008 (-0.13%)
instructions in affected programs: 219850 -> 212023 (-3.56%)
total loops in shared programs: 1979 -> 1979 (0.00%)
helped: 1192
HURT: 0
We also have the "reserved for kick" space available. Some of my earlier
changes can probably be removed, but this is a quick fix for some of the
rarer fallout.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Cc: <mesa-stable@lists.freedesktop.org>
This greatly increases the pressure you can put on the driver before
create fails. Ultimately we need to let the kernel take control of
our cached BOs and just take them from us (and other clients)
directly, but this is a very easy patch for the moment.
Cc: "11.0" <mesa-stable@lists.freedesktop.org>
I thought that aliased functions didn't need to be added, but that might
only be if the function aliases something in the same {desktop,ES}
space. Resolves the dispatch sanity test failure.
Fixes: 13b19aa81 (mesa: expose support for GL_EXT_buffer_storage)
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=92824
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Very long line loops which spanned 3 or more vertex buffers were not
handled correctly and could result in stray lines.
The piglit lineloop test draws 10000 vertices by default, and is not
long enough to trigger this. Even 'lineloop -count 100000' doesn't
trigger the bug.
For future reference, the issue can be reproduced by changing Mesa's
VBO_VERT_BUFFER_SIZE to 4096 and changing the piglit lineloop test to
use glVertex2f(), draw 3 loops instead of 1, and specifying -count
1023.
Acked-by: Sinclair Yeh <syeh@vmware.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
When we emulate XOR logicop mode with blend-subtract, we need to ensure
that the fragment shader always emits white. We had this implemented
for VGPU9, but not VGPU10.
VMware bug 1545492.
Reviewed-by: Charmaine Lee <charmainel@vmware.com>
This actually stored the values as 8bit linear values in the cache,
then did another srgb->linear conversion...
We don't want to do the former (decoding 8bit srgb values to 8bit linear
completely defeats the purpose of srgb in the first place), so just decode
to 8bit srgb.
Fixes piglit.spec.ext_texture_srgb.texwrap formats-s3tc tests.
There is nothing wrong with the code today, but as one modifies the code it
turns out to be not too difficult to mess up the code, and this easy assertion
should catch such driver implementation failures quickly.
Cc: Kristian Høgsberg <krh@bitplanet.net>
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Reviewed-by: Chad Versace <chad.versace@intel.com>
Reviewed-by: Neil Roberts <neil@linux.intel.com>
This extension requires ES 3.1 since it relies on glMemoryBarrier.
For testing purposes I temporarily moved glMemoryBarrier to be an ES 3.0
function.
This has been tested with the piglit in the ML and the Dolphin emulator.
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
V3: clamp array index to the correct size (the size of the current array
rather than the inner array) Francisco Jerez.
V2: avoid useless zero-initialization and addition for the first AoA level,
avoid redundant temporary, make use of type_size_scalar(), rename aoa_size
to element_size, assign the indirect indexing temporary directly to
image.reladdr, and replace while loop with a for loop. All suggested
by Francisco Jerez.
Reviewed-by: Francisco Jerez <currojerez@riseup.net>
compressed textures are very slow because decoding is rather complex
(and because there's no jit code code to decode them too for non-technical
reasons).
Thus, add some texture cache which holds a couple of decoded blocks.
Right now this handles only s3tc format albeit it could be extended to work
with other formats rather trivially as long as the result of decode fits into
32bit per texel (ideally, rgtc actually would decode to more than 8 bits
per channel, but even then making it work for it shouldn't be too difficult).
This can improve performance noticeably but don't expect wonders (uncompressed
is unsurprisingly still faster). It's also possible it might be slower in
some cases (using nearest filtering for example or if there's otherwise not
many cache hits, the cache is only direct mapped which isn't great).
Also, actual decode of a block relies on util code, thus even though always
full blocks are decoded it is done texel by texel - this could obviously
benefit greatly from simd-optimized code decoding full blocks at once...
Note the cache is per (raster) thread, and currently only used for fragment
shaders.
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
There are currently two methods in llvmpipe code to calculate coeffs to
be used as inputs for the fragment shader. The two methods use slightly
different ways to do the floating point calculations and thus produce
slightly different results.
The decision which method to use is determined by the size of the vector
that is used by the platform.
For vectors with size of more than 128bit, a single-step method is used,
in which coeffs_init_simple() + attribs_update_simple() are called.
For vectors with size of 128bit or less, a two-step method is used, in
which coeffs_init() + attribs_update() are called.
This causes some piglit tests (clip-distance-bulk-copy,
interface-vs-unnamed-to-fs-unnamed) to fail when using platforms with
128bit vectors (such as ppc64le or x86-64 without AVX).
This patch makes platforms with 128bit vectors use the single-step
method (aka "simple" method) instead of the two-step method.
This would make the resulting coeffs identical between more platforms,
make sure the piglit tests passes, and make debugging and maintainability
a bit easier as the generated LLVM IR will be the same for more platforms.
The performance impact is negligible for x86-64 without AVX, and
basically non-existent for ppc64le, as it can be seen from the following
benchmarking results:
- glxspheres, on ppc64le:
- original code: 4.892745317 frames/sec 5.460303857 Mpixels/sec
- with the patch: 4.932083873 frames/sec 5.504205571 Mpixels/sec
- Additional 0.8% performance boost
- glxspheres, on x86-64 without AVX:
- original code: 20.16418809 frames/sec 22.50323395 Mpixels/sec
- with the patch: 20.31328989 frames/sec 22.66963152 Mpixels/sec
- Additional 0.74% performance boost
- glmark2, on ppc64le:
- original code: score of 58
- with my change: score of 57
- glmark2, on x86-64 without AVX:
- original code: score of 175
- with the patch: score of 167
- Impact of of -4.5% on performance
- OpenArena, on ppc64le:
- original code: 3398 frames 1719.0 seconds 2.0 fps
255.0/505.9/2773.0/0.0 ms
- with the patch: 3398 frames 1690.4 seconds 2.0 fps
241.0/497.5/2563.0/0.2 ms
- 29 seconds faster with the patch, which is about 2%
- OpenArena, on x86-64 without AVX:
- original code: 3398 frames 239.6 seconds 14.2 fps
38.0/70.5/719.0/14.6 ms
- with the patch: 3398 frames 244.4 seconds 13.9 fps
38.0/71.9/697.0/14.3 ms
- 0.3 fps slower with the patch (about 2%)
Additional details can be found at:
http://lists.freedesktop.org/archives/mesa-dev/2015-October/098635.html
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
pipe->flush never returned SDMA fences. This fixes it.
This is only an issue on amdgpu where fences can signal out of order.
Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>
This is hidden behind INTEL_SCALAR_GS=1 for now, as we don't yet support
instanced geometry shaders, and Orbital Explorer's shader spills like
crazy. But the infrastructure is in place, and it's largely working.
v2: Lots of rebasing.
v3: (feedback from Kristian Høgsberg)
- Handle stride and subreg_offset correctly for ATTRs; use a helper.
- Fix missing emit_shader_time_end() call.
- Delete dead code after early EOT in static vertex case to avoid
tripping asserts in emit_shader_time_end().
- Use proper D/UD type in intexp2().
- Fix "EndPrimitve" and "to that" typos.
- Assert that invocations == 1 so we know this is missing.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
We really ought to compute the VUE map at link time and stash it, rather
than recomputing it here, but with the mess of program structures I
wasn't sure where to put it. We can improve that later.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
Jason reworked this so it isn't simply ST_GS anymore...it's either -1
(not enabled) or an actual offset.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
On future generation platforms the color clear value is stored elsewhere in the
surface state. By extracting this logic, we can cleanly implement the difference
in an upcoming patch.
Should have no functional impact.
v2: Move hunk from the next patch into this patch (Matt)
Whitespace fix (Ben)
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Reviewed-by: Neil Roberts <neil@linux.intel.com>
The allocate_surface_state already zeroes out the surface state, and doing it
later in the function is destructive for what we want to accomplish when we
split out support for gen9 fast clears (next patch).
NOTE: Only dword 12 actually needed to be fixed, but it seemed more consistent
to remove the other instances as well. I can make an argument both ways (open
coding it, vs. not). I can rework the next patch if requires.
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Reviewed-by: Chad Versace <chad.versace@intel.com>
Reviewed-by: Neil Roberts <neil@linux.intel.com>
To get the size (in bytes) of a compute parameter, clover first calls
get_compute_param() with a NULL data pointer. The RET() macro is based
on nv50.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
A few new PCI ids are added here, and one is removed (0x190B) because it no
longer seems to exist anywhere.
v2-4:
Only use ascii characters (Ilia)
0x1921 is no longer marked as f
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Signed-off-by: Ben Widawsky <benjamin.widawsky@intel.com>
Like other gen8+ hardware, the hardware automatically scales up thread counts.
We must be careful about the URB sizes since GT4 adds another slice.
One of the existing PCI IDs is actually mislabeled as GT3. Arguably this is a
real bug since the URB size will be wrong. Because this patch is simply meant to
add the missing IDs, that will be fixed in a later patch.
v2: No longer relevant.
v3: Update the wm thread count to support GT4. The WM thread count is used to
determine the maximum scratch space required. Currently the code always
allocates the maximum amount even though lower GT SKUs require less. The formula
is threads_per_psd * subslices_per_slice * slices
Cc: mesa-stable@lists.freedesktop.org
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Signed-off-by: Ben Widawsky <benjamin.widawsky@intel.com>
Note: The OpenGL 4.3 - 4.5 specification language for DispatchCompute
appears to have an error regarding the max allowed values. When adding
the specification citation, we note why the code does not match the
specification language.
v2:
* Updates based on review from Iago
Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Cc: Iago Toral Quiroga <itoral@igalia.com>
Cc: Marta Lofstedt <marta.lofstedt@intel.com>
Reviewed-by: Marta Lofstedt <marta.lofstedt@intel.com>
There is some discrepancy between the return values for some error
cases for the DispatchComputeIndirect call in the ARB_compute_shader
specification. Regarding the indirect parameter, in one place the
extension spec lists that the error returned for invalid values should
be INVALID_OPERATION, while later it specifies INVALID_VALUE.
The OpenGL 4.3 and OpenGLES 3.1 specifications appear to be consistent
in requiring the INVALID_VALUE error return in this case.
Here we update the code to match the main specifications, and update
the citations use the main specification rather than the extension
specification.
v2:
* Updates based on review from Iago
Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Cc: Iago Toral Quiroga <itoral@igalia.com>
Cc: Marta Lofstedt <marta.lofstedt@intel.com>
Reviewed-by: Marta Lofstedt <marta.lofstedt@intel.com>
We've made a mistake in calling the Channel Enable bits "writemask",
because they do more than control which channels of the destination are
written -- they actually control which channels are enabled (surprise!
surprise!)
So, if we emit
cmp.z.f0(8) null.xy<1>D g10<4,4,1>.xyzzD g2<0,4,1>.xyzzD
mov(8) g12<1>.xUD 0x00000000UD
(+f0.all4h) mov(8) g12<1>.xUD 0xffffffffUD
where the CMP instruction has only .xy channel enables, it won't write
the .zw channels of the flag register, which are of course read by the
+f0.all4 predicate.
We need to always emit CMP instructions whose flag result might be read
by such a predicate with all channels enabled.
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
Since introduction of SSBO, UniformStorage contains not just uniforms
but also buffer variables, this needs to be taken in to account when
calculating active uniforms with GL_ACTIVE_UNIFORMS and
GL_ACTIVE_UNIFORM_MAX_LENGTH.
No Piglit regressions.
Signed-off-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Eduardo Lima Mitev <elima@igalia.com>
Reviewed-by: Timothy Arceri <timothy.arceri@collabora.com>
gl_active_atomic_buffer contains index to UniformStorage, we need to
calculate resource index for that gl_uniform_storage.
Fixes following CTS tests:
ES31-CTS.program_interface_query.atomic-counters
ES31-CTS.program_interface_query.atomic-counters-one-buffer
No Piglit regressions.
Signed-off-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Marta Lofstedt <marta.lofstedt@intel.com>
These helpers are ran for same case the same loop. Here joined
their operation so the loop is ran just once. Also fixed
out-of-memory condition here.
v2: Make the loop simpler to read as per Tapani's suggestion
Signed-off-by: Juha-Pekka Heikkila <juhapekka.heikkila@gmail.com>
Reviewed-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
Tested-by: Tapani Pälli <tapani.palli@intel.com>
Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Marta Lofstedt <marta.lofstedt@intel.com>
[itoral@igalia.com: Reviewed-by for all except the ctx->_Shader change]
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
So I've known this was broken before, cogl has a workaround
for it from what I know, but with the gallium based swrast
drivers BlitFramebuffer from back to front or vice-versa
was pretty broken.
The legacy swrast driver tracks when a front buffer is used
and does the get/put images when it is mapped/unmapped,
so this patch attempts to add the same functionality to the
gallium drivers.
It creates a new context interface to denote when a front
buffer is being created, and passes a private pointer to it,
this pointer is then used to decide on map/unmap if the
contents should be updated from the real frontbuffer using
get/put image.
This is primarily to make gtk's gl code work, the only
thing I've tested so far is the glarea test from
https://github.com/ebassi/glarea-example.git
v2: bump extension version,
check extension version before calling get image. (Ian)
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=91930
Cc: <mesa-stable@lists.freedesktop.org>
Signed-off-by: Dave Airlie <airlied@redhat.com>
Wrap some of the 'omg it's getting out of hand' long lines, and
re-indent where things feel off.
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Reviewed-by: Dave Airlie <airlied@redhat.com>
Include what you want, rather than relying on a header foo.h N levels
down the include chain, to provide something that you need.
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Reviewed-by: Dave Airlie <airlied@redhat.com>
The only two remaining cases of (struct virgl_resource *) require a
closer look. Either the error checking is missing or the arguments
provided feel wrong.
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Reviewed-by: Dave Airlie <airlied@redhat.com>
The screen already has a pointer to the (base) winsys object.
With the latter of which implemented/sub-classed as either drm or sw
based one, depending on the target.
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Reviewed-by: Dave Airlie <airlied@redhat.com>
Strictly speaking virgl_hw.h should reside in the driver folder, as
it describes the hardware. Moving it allows us to nuke the following
strange dependency
winsys/vtest > driver > winsys/drm
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Reviewed-by: Dave Airlie <airlied@redhat.com>
Use the relevant GALLIUM_foo_CFLAGS which has all the requirements
(not to mention VISIBITY_CFLAGS) and keep ../ out of the include
directives.
Signed-off-by: Emil Velikov <emil.velikov@collabora.com>
Reviewed-by: Dave Airlie <airlied@redhat.com>
The drm/ prefix is required, if using the kernel provided headers. As
most distros don't ship them it and we already depend on libdrm (which
adds the relevant -I flag) just drop the drm/ from the include.
Once a libdrm release with the virtgpu_drm.h header is released, we can
drop our local copy of the file.
Signed-off-by: Emil Velikov <emil.velikov@collabora.com>
Reviewed-by: Dave Airlie <airlied@redhat.com>
v2:
- Add a few const qualifiers for good measure.
- Drop unneeded retype()s (Matt)
- Convert timestamp to SIMD8/16, as fs_visitor::get_timestamp() returns
SIMD4 (Connor)
v3:
- Remove unneeded temporary + MOV (Connor)
Signed-off-by: Emil Velikov <emil.velikov@collabora.com>
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
We're about to reuse get_timestamp() for the nir_intrinsic_shader_clock.
In the latter the generalisation does not apply, so move the smear()
where needed. This also makes the function analogous to the vec4 one.
v2: Tweak the comment - The caller -> We (Matt, Connor).
v3: More comment tweaks (Connor)
Signed-off-by: Emil Velikov <emil.velikov@collabora.com>
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
v2: Add flags and inline comment/description.
v3: None of the input/outputs are variables
v4: Drop clockARB reference, relate code motion barrier comment wrt
intrinsic flag.
v5: Drop the "thus we can eliminate..." comment (Connor)
Signed-off-by: Emil Velikov <emil.velikov@collabora.com>
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
While we are at it, store the rotate offset for occlusion queries to
nv50_hw_query like on nvc0.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Pierre Moreau <pierre.morrow@free.fr>
Like for nvc0, this will allow to split different types of queries and
to prepare the way for both global performance counters and MP counters.
While we are at it, make use of nv50_query struct instead of pipe_query.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
I.e. implements:
VaAcquireBufferHandle
VaReleaseBufferHandle
for memory of type VA_SURFACE_ATTRIB_MEM_TYPE_DRM_PRIME
And apply relatives change to:
vlVaMapBuffer
vlVaUnMapBuffer
vlVaDestroyBuffer
Implementation inspired from cgit.freedesktop.org/vaapi/intel-driver
Tested with gstreamer-vaapi with nouveau driver.
Signed-off-by: Julien Isorce <j.isorce@samsung.com>
Reviewed-by: Emil Velikov <emil.l.velikov@gmail.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
And apply relatives change to:
vlVaBufferSetNumElements
vlVaCreateBuffer
vlVaMapBuffer
vlVaUnmapBuffer
vlVaDestroyBuffer
vlVaPutImage
It is unfortunate that there is no proper va buffer type and struct
for this. Only possible to use VAImageBufferType which is normally
used for normal user data array.
On of the consequences is that it is only possible VaDeriveImage
is only useful on surfaces backed with contiguous planes.
Implementation inspired from cgit.freedesktop.org/vaapi/intel-driver
Signed-off-by: Julien Isorce <j.isorce@samsung.com>
Reviewed-by: Emil Velikov <emil.l.velikov@gmail.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Add support for VA_PROFILE_NONE and VAEntrypointVideoProc
in the 4 following functions:
vlVaQueryConfigProfiles
vlVaQueryConfigEntrypoints
vlVaCreateConfig
vlVaQueryConfigAttributes
Signed-off-by: Julien Isorce <j.isorce@samsung.com>
Reviewed-by: Emil Velikov <emil.l.velikov@gmail.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Add support for VPP in the following functions:
vlVaCreateContext
vlVaDestroyContext
vlVaBeginPicture
vlVaRenderPicture
vlVaEndPicture
Add support for VAProcFilterNone in:
vlVaQueryVideoProcFilters
vlVaQueryVideoProcFilterCaps
vlVaQueryVideoProcPipelineCaps
Add handleVAProcPipelineParameterBufferType helper.
One application is:
VASurfaceNV12 -> gstvaapipostproc -> VASurfaceRGBA
Signed-off-by: Julien Isorce <j.isorce@samsung.com>
Reviewed-by: Emil Velikov <emil.l.velikov@gmail.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Inspired from http://cgit.freedesktop.org/vaapi/intel-driver/
especially src/i965_drv_video.c::i965_CreateSurfaces2.
This patch is mainly to support gstreamer-vaapi and tools that uses
this newer libva API. The first advantage of using VaCreateSurfaces2
over existing VaCreateSurfaces, is that it is possible to select which
the pixel format for the surface. Indeed with the simple VaCreateSurfaces
function it is only possible to create a NV12 surface. It can be useful
to create a RGBA surface to use with video post processing.
The avaible pixel formats can be query with VaQuerySurfaceAttributes.
Signed-off-by: Julien Isorce <j.isorce@samsung.com>
Reviewed-by: Emil Velikov <emil.l.velikov@gmail.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
If formats are not the same vlVaPutImage re-creates the video
buffer with the right format. But if the creation of this new
video buffer fails then the surface looses its current buffer.
Let's just destroy the previous buffer on success.
Signed-off-by: Julien Isorce <j.isorce@samsung.com>
Reviewed-by: Emil Velikov <emil.l.velikov@gmail.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Added PIPE_VIDEO_CHROMA_FORMAT_NONE in p_format.h
and return it by default in ChromaToPipe.
Renamed YCbCrToPipe to VaFourccToPipeFormat because it now
contains RGB.
Implemented PipeFormatToVaFourcc which will be used later in
VlVaDeriveImage.
Note that gstreamer-vaapi check all the VAImageFormat fields.
Signed-off-by: Julien Isorce <j.isorce@samsung.com>
Reviewed-by: Emil Velikov <emil.l.velikov@gmail.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Commit 4565b6f did not update the basename match's check for
the case that string would exactly match the name of the
variable if the suffix "[0]" were appended to it.
Fixes two dEQP-GLES31 tests:
dEQP-GLES31.functional.program_interface_query.shader_storage_block.resource_list.block_array
dEQP-GLES31.functional.program_interface_query.shader_storage_block.resource_list.block_array_single_element
v2:
- Change the position of rname_has_array_index_zero to avoid an out-of-bounds
read. Reported by Tapani Pälli.
Signed-off-by: Samuel Iglesias Gonsalvez <siglesias@igalia.com>
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
Previously, we were using some heuristics to try and detect when a write
was about to begin a live range, or when a read was about to end a live
range. We never used the liveness analysis information used by the
register allocator, though, which meant that the scheduler's and the
allocator's ideas of when a live range began and ended were different.
Not only did this make our estimate of the register pressure benefit of
scheduling an instruction wrong in some cases, but it was preventing us
from knowing the actual register pressure when scheduling each
instruction, which we want to have in order to switch to register
pressure scheduling only when the register pressure is too high.
This commit rewrites the register pressure tracking code to use the same
model as our register allocator currently uses. We use the results of
liveness analysis, as well as the compute_payload_ranges() function that
we split out in the last commit. This means that we compute live ranges
twice on each round through the register allocator, although we could
speed it up by only recomputing the ranges and not the live in/live out
sets after scheduling, since we only shuffle around instructions within
a single basic block when we schedule.
Shader-db results on bdw:
total instructions in shared programs: 7130187 -> 7129880 (-0.00%)
instructions in affected programs: 1744 -> 1437 (-17.60%)
helped: 1
HURT: 1
total cycles in shared programs: 172535126 -> 172473226 (-0.04%)
cycles in affected programs: 11338636 -> 11276736 (-0.55%)
helped: 876
HURT: 873
LOST: 8
GAINED: 0
v2: use regs_read() in more places.
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
We'll need this for the scheduler too, since it wants to know when the
live ranges of payload registers end in order to model them in our
register pressure calculations.
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
The heuristic we're using is rather lame, since it assumes everything is
non-uniform and loops execute 10 times, but it should be enough for
measuring improvements in the scheduler that don't result in a change in
the number of instructions.
v2:
- Switch loops and cycle counts to be compatible with older shader-db.
- Make loop heuristic 10x to match with spilling code.
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
Before, we would only do scheduling after register allocation if we
spilled, despite the fact that the pre-RA scheduler was only supposed to
be for register pressure and set the latencies of every instruction to
1. This meant that unless we spilled, which we rarely do, then we never
considered instruction latencies at all, and we usually never bothered
to try and hide texture fetch latency. Although a later commit removes
the setting the latency to 1 part, we still want to always run the
post-RA scheduler since it's able to take the false dependencies that
the register allocator creates into account, and it can be more
aggressive than the pre-RA scheduler since it doesn't have to worry
about register pressure at all.
Test master post-ra-sched diff %diff
bench_OglPSBump2 396.730 402.386 5.656 +1.400%
bench_OglPSBump8 244.370 247.591 3.221 +1.300%
bench_OglPSPhong 241.117 242.002 0.885 +0.300%
bench_OglPSPom 59.555 59.725 0.170 +0.200%
bench_OglShMapPcf 86.149 102.346 16.197 +18.800%
bench_OglVSTangent 388.849 395.489 6.640 +1.700%
bench_trex 65.471 65.862 0.390 +0.500%
bench_trexoff 69.562 70.150 0.588 +0.800%
bench_heaven 25.179 25.254 0.074 +0.200%
Reviewed-by: Jason Ekstrand <jasoan.ekstrand@intel.com>
Although write-after-write dependencies have the same latency as
read-after-write dependencies due to how the register scoreboard works,
write-after-read dependencies aren't checked by the EU at all, so
they're purely a constraint on how the scheduler can order the
instructions.
v2: fix accumulator dependencies too.
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
The issue time for an instruction is how many cycles it takes to
actually put it into the pipeline. If there's a pipeline stall that
causes the instruction to be delayed, we should first take that into
account to figure out when the instruction would start executing and
*then* add the issue time. The old code had it backwards, and so we
would underestimate the total time whenever we thought there would be a
pipeline stall by up to the issue time of the instruction.
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
Uses the same technique as for nvc0 of fixups before upload, and
evicting in case of state change. Removes one source of variants kept by
st/mesa.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
These are often useful in debugging, and the writemask (actually
"Channel Enables") determines more than just what goes into the
destination.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
No functional change, since they were both 3, but BRW_IMMEDIATE_VALUE is
the hardware value and IMM was the IR value -- and you can see that
BRW_IMMEDIATE_VALUE was correctly used in the context of this patch.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
No functional change, since they were both 3, but BRW_IMMEDIATE_VALUE is
the hardware value and IMM was the IR value -- and you can see that
BRW_IMMEDIATE_VALUE was correctly used in the context of this patch.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
I've checked with piglit and one tests fails, but it fails
on evergreen as well, so will get fixed later.
Otherwise SB seems to be working fine for geom shaders on my
rv635.
Signed-off-by: Dave Airlie <airlied@redhat.com>
We really weren't taking advantage of vec4_generator being a class.
By adding a "p" parameter to the helper methods, and "prog_data" to
ones which need binding table information, we can convert everything
to static functions.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
The public API for the generator is brw_vec4_generate_code(); nobody
actually needs to use the class. This means we can extend it without
triggering the recompiles associated with altering brw_vec4.h.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
vec4_generator is a class for convenience, but only exports a single
method as its public API. It makes much more sense to just export a
single function.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
This patch makes the visitor convert registers to the HW_REG file at the
very end, after register allocation, post-RA scheduling, and dependency
control flagging. After that, everything is in fixed brw_regs.
This simplifies the code generator, as it can just use the hardware
registers rather than having to interpret our abstract files. In
particular, interpreting the UNIFORM file meant reading prog_data
to figure out where push constants are supposed to start.
Having the part of the code that performs register allocation also
translate everything to hardware registers seems sensible.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Some constants (like 1.0 and 0.5) could be inlined as immediate inputs
without using their literal value. The r600_bytecode_special_constants()
function emulates the negative of these constants by using NEG modifier.
However some shaders define -1.0 constant and want to use it as 1.0.
They do so by using ABS modifier. But r600_bytecode_special_constants()
set NEG in addition to ABS. Since NEG modifier have priority over ABS one,
we get -|1.0| as result, instead of |1.0|.
The patch simply prevents the additional switching of NEG when ABS is set.
[According to Ivan Kalvachev, this bug was fond via
https://github.com/iXit/Mesa-3D/issues/126 and
https://github.com/iXit/Mesa-3D/issues/127]
Signed-off-by: Ivan Kalvachev <ikalvachev@gmail.com>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
CC: <mesa-stable@lists.freedesktop.org>
Without the clamping by NumLevels, the state tracker would reallocate the
texture storage (incorrect) and even fail to copy the base level image
after reallocation, leading to the graphical glitch of
https://bugs.freedesktop.org/show_bug.cgi?id=91993 .
A piglit test has been submitted for review as well (subtest of
arb_texture_storage-texture-storage).
v2: also bypass all calls to st_finalize_texture (suggested by Marek Olšák)
Cc: mesa-stable@lists.freedesktop.org
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
In OpenGL ES, the COMPRESSED_TEXTURE_FORMATS query returns the set of
supported specific compressed formats. Since ASTC formats fit within
that category, include them in the set and update the
NUM_COMPRESSED_TEXTURE_FORMATS query as well.
This enables GLES2-based ASTC dEQP tests to run. See the Bugzilla for
more info.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=92193
Reported-by: Tapani Pälli <tapani.palli@intel.com>
Suggested-by: Ian Romanick <ian.d.romanick@intel.com>
Signed-off-by: Nanley Chery <nanley.g.chery@intel.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
In agreement with the extension spec and commit
dd0eb00487, filter FXT1 formats to the
desktop GL profiles. Now we no longer advertise such formats as supported
in an ES context and then throw an INVALID_ENUM error when the client
tries to use such formats with CompressedTexImage2D.
Fixes the following 26 dEQP tests:
* dEQP-GLES2.functional.negative_api.texture.compressedteximage2d_invalid_border
* dEQP-GLES2.functional.negative_api.texture.compressedteximage2d_invalid_border_cube_neg_x
* dEQP-GLES2.functional.negative_api.texture.compressedteximage2d_invalid_border_cube_neg_y
* dEQP-GLES2.functional.negative_api.texture.compressedteximage2d_invalid_border_cube_neg_z
* dEQP-GLES2.functional.negative_api.texture.compressedteximage2d_invalid_border_cube_pos_x
* dEQP-GLES2.functional.negative_api.texture.compressedteximage2d_invalid_border_cube_pos_y
* dEQP-GLES2.functional.negative_api.texture.compressedteximage2d_invalid_border_cube_pos_z
* dEQP-GLES2.functional.negative_api.texture.compressedteximage2d_invalid_size
* dEQP-GLES2.functional.negative_api.texture.compressedteximage2d_level_max_cube_pos
* dEQP-GLES2.functional.negative_api.texture.compressedteximage2d_level_max_tex2d
* dEQP-GLES2.functional.negative_api.texture.compressedteximage2d_neg_level_cube
* dEQP-GLES2.functional.negative_api.texture.compressedteximage2d_neg_level_tex2d
* dEQP-GLES2.functional.negative_api.texture.compressedteximage2d_neg_width_height_cube_neg_x
* dEQP-GLES2.functional.negative_api.texture.compressedteximage2d_neg_width_height_cube_neg_y
* dEQP-GLES2.functional.negative_api.texture.compressedteximage2d_neg_width_height_cube_neg_z
* dEQP-GLES2.functional.negative_api.texture.compressedteximage2d_neg_width_height_cube_pos_x
* dEQP-GLES2.functional.negative_api.texture.compressedteximage2d_neg_width_height_cube_pos_y
* dEQP-GLES2.functional.negative_api.texture.compressedteximage2d_neg_width_height_cube_pos_z
* dEQP-GLES2.functional.negative_api.texture.compressedteximage2d_neg_width_height_tex2d
* dEQP-GLES2.functional.negative_api.texture.compressedteximage2d_width_height_max_cube_neg_x
* dEQP-GLES2.functional.negative_api.texture.compressedteximage2d_width_height_max_cube_neg_y
* dEQP-GLES2.functional.negative_api.texture.compressedteximage2d_width_height_max_cube_neg_z
* dEQP-GLES2.functional.negative_api.texture.compressedteximage2d_width_height_max_cube_pos_x
* dEQP-GLES2.functional.negative_api.texture.compressedteximage2d_width_height_max_cube_pos_y
* dEQP-GLES2.functional.negative_api.texture.compressedteximage2d_width_height_max_cube_pos_z
* dEQP-GLES2.functional.negative_api.texture.compressedteximage2d_width_height_max_tex2d
v2. Use _mesa_is_desktop_gl() (Ilia, Ian)
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Signed-off-by: Nanley Chery <nanley.g.chery@intel.com>
Not sure if this is actually reachable in practice (to have a complex
copy with MS textures).
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
This makes sure that user is still able to query properties about
variables that have gotten removed by opt_dead_builtin_varyings pass.
Fixes following OpenGL ES 3.1 test:
ES31-CTS.program_interface_query.output-layout
No Piglit regressions.
v2: cleanup, drop extra parenthesis (Topi)
Signed-off-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Marta Lofstedt <marta.lofstedt@intel.com>
This is required to store information about fragdata arrays, currently
these variables get lost and cannot be retrieved later in sensible way
for program interface queries. List will be utilized by next patch.
Patch also modifies opt_dead_builtin_varyings pass to build list when
lowering fragdata arrays. This is identical approach as taken with
packed varyings pass.
Signed-off-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Marta Lofstedt <marta.lofstedt@intel.com>
From ARB_program_interface_query:
"For the property of BUFFER_DATA_SIZE, then the implementation-dependent
minimum total buffer object size, in basic machine units, required to hold
all active variables associated with an active uniform block, shader
storage block, or atomic counter buffer is written to <params>. If the
final member of an active shader storage block is array with no declared
size, the minimum buffer size is computed assuming the array was declared
as an array with one element."
Fixes the following dEQP-GLES31 tests:
dEQP-GLES31.functional.program_interface_query.shader_storage_block.buffer_data_size.named_block
dEQP-GLES31.functional.program_interface_query.shader_storage_block.buffer_data_size.unnamed_block
dEQP-GLES31.functional.program_interface_query.shader_storage_block.buffer_data_size.block_array
v2:
- Fix comment's indentation and explain that the parser already
checked that unsized array is in last element of a shader
storage block (Iago).
- Add assert (Iago).
Signed-off-by: Samuel Iglesias Gonsalvez <siglesias@igalia.com>
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Normally, we could read gl_Layer from bits 26:16 of R0.0. However, the
specification requires that bogus out-of-range 32-bit values written by
previous stages need to appear in the fragment shader as-written.
Instead, we pass in the full 32-bit value from the VUE header as an
extra flat-shaded varying. We have the SF override the value to 0
when the previous stage didn't actually write a value (it's actually
defined to return 0).
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Chris Forbes <chrisf@ijw.co.nz>
Traditionally, we've hardcoded "URB Entry Read Offset" to 1 (which
represents 2 vec4 varying slots) to skip over the 8 DWord VUE header.
In order to support ARB_fragment_layer_viewport, we'll need to read
from that header. This patch adds the basic plumbing necessary to
calculate a value dynamically and hook it up in the SBE packets.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Chris Forbes <chrisf@ijw.co.nz>
Integer varyings need to be flat qualified - all others were already.
I think we just missed this. Presumably some hardware passes this via
sideband and ignores attribute interpolation, so no one has noticed.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Chris Forbes <chrisf@ijw.co.nz>
Commit 268008f98c changed unused VUE map
slots to be initialized with BRW_VARYING_SLOT_PAD, not COUNT. I missed
updating this. It also means that commit message was wrong, as some
code *did* rely slots being initialized to COUNT.
This may fix a bug with SSO programs with > 16 FS input varyings.
I think we probably just emitted extra pointless code, but probably
didn't break anything. We might also just have no tests for that.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Chris Forbes <chrisf@ijw.co.nz>
Avoid deferring building shaders until draw time, should hopefully
reduce any stuttering, as well as enable shader-db style analysis.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Unfortunately flatshading is an all-or-nothing proposition on nvc0,
while GL 3.0 calls for the ability to selectively specify explicit
interpolation parameters on gl_Color/gl_SecondaryColor which would
override the flatshading setting. This allows us to fix up the
interpolation settings after shader generation based on rasterizer
settings.
While we're at it, we can add support for dynamically forcing all
(non-flat) shader inputs to be interpolated per-sample, which allows
st/mesa to not generate variants for these.
Fixes the remaining failing glsl-1.30/execution/interpolation piglits.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
This was introduced in GLSL IR after NIR development had branched.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
nir_intrinsic_load_patch_vertices_in corresponds to gl_PatchVerticesIn,
a special input in both the TCS and TES stages.
nir_intrinsic_load_tess_coord corresponds to gl_TessCoord, a special
tessellation evaluation shader input.
nir_intrinsic_load_tess_level_outer/inner correspond to the
gl_TessLevelOuter[] and gl_TessLevelInner[] evaluation shader inputs,
which we treat as system values because they're stored specially.
(These intrinsics are only for the TES - the TCS uses output variables.)
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Consider the case of two nearly identical GLSL fragment shaders:
out vec4 color;
void main() { color = vec4(1); }
and
layout(early_fragment_tests) in;
out vec4 color;
void main() { color = vec4(1); }
These shaders compile to the exact same assembly, but have distinct
values for brw_wm_prog_data::early_fragment_tests.
Since these are two independent GLSL shaders, they have different
program keys - notably, brw_wm_prog_key::program_string_id differs.
When uploading the second, brw_upload_cache will find an existing copy
of the assembly in the cache BO, which means matching_data will be
non-NULL. Although we create a second cache item (with the new key
and prog_data), we set item->offset to the existing copy and avoid
re-uploading duplicate assembly.
However, brw_search_cache() would only flag BRW_NEW_*_PROG_DATA if
item->offset differed from the supplied offset. With reuse, both
programs have the same offset, but prog_data changed. We have to
flag it, but failed to.
To fix this, we simply need to check if the aux (prog_data) pointer
changed. If either the assembly or the prog_data differs, flag it.
This fixes a regression since 1bba29ed40,
where Topi fixed brw_upload_cache() to actually reuse identical
assembly. Prior to that, reuse basically never happened due to bugs.
Unfortunately, this code apparently wasn't prepared to handle reuse!
Fixes GPU hangs in Dolphin on Broadwell.
Huge thanks to Pierre Bourdon and Ilia Mirkin for debugging this
and helping track down the real issue.
Cc: "11.0" <mesa-stable@lists.freedesktop.org>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=92623
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
Tested-by: Pierre Bourdon <delroth@gmail.com>
With just the right sequence of per-vertex commands and state changes,
it's possible for this assertion to fail (such as with viewperf11's
lightwave-06-1 test). Instead of asserting, return 0 so that the
caller knows the VBO is full and needs to be flushed.
Reviewed-by: Charmaine Lee <charmainel@vmware.com>
which is where a block in src maps to a pixel in dst and vice versa.
e.g. DXT1 <-> R32G32_UINT
DXT5 <-> R32G32B32A32_UINT
Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>
The variable is already of type src_reg. creating a new instance only to
destroy it seems unnecessary.
Signed-off-by: Emil Velikov <emil.velikov@collabora.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
There is only one function that can be called, which is well known at
compilation time.
The abstraction used here seems unnecessary, so let's use a direct call
to brw_stage_prog_data_free() when appropriate, cut down the size of
struct brw_cache.
Signed-off-by: Emil Velikov <emil.velikov@collabora.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
The maximum number of active variables for shader storage blocks should
take into account the specific rules for shader storage blocks, i.e. for
an active shader storage block member declared as an array, an entry
will be generated only for the first array element, regardless of its type.
Fixes 3 dEQP-GLES31.functional.* tests:
dEQP-GLES31.functional.program_interface_query.shader_storage_block.active_variables.named_block
dEQP-GLES31.functional.program_interface_query.shader_storage_block.active_variables.unnamed_block
dEQP-GLES31.functional.program_interface_query.shader_storage_block.active_variables.block_array
Signed-off-by: Samuel Iglesias Gonsalvez <siglesias@igalia.com>
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
From OpenGL ES 3.1 specification, section 10.5:
"DrawArraysIndirect requires that all data sourced for the
command, including the DrawArraysIndirectCommand
structure, be in buffer objects, and may not be called when
the default vertex array object is bound."
Signed-off-by: Marta Lofstedt <marta.lofstedt@linux.intel.com>
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
For debugging, bug reports, etc.
This is not in the radeonsi directory, but it is about radeonsi.
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
From OpenGL 4.4 specification, section 10.4 and
Open GL Es 3.1 section 10.5:
"An INVALID_VALUE error is generated if indirect is not a multiple
of the size, in basic machine units, of uint."
However, the current code follow the ARB_draw_indirect:
https://www.opengl.org/registry/specs/ARB/draw_indirect.txt
"INVALID_OPERATION is generated by DrawArraysIndirect and
DrawElementsIndirect if commands source data beyond the end
of a buffer object or if <indirect> is not word aligned."
V2: After discussions on the list, it was suggested to
only keep the INVALID_VALUE error.
Signed-off-by: Marta Lofstedt <marta.lofstedt@linux.intel.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
From ARB_program_query_interface spec:
"uint GetProgramResourceIndex(uint program, enum programInterface,
const char *name);
[...]
If <name> exactly matches the name string of one of the active resources
for <programInterface>, the index of the matched resource is returned.
Additionally, if <name> would exactly match the name string of an active
resource if "[0]" were appended to <name>, the index of the matched
resource is returned. [...]"
"A string provided to GetProgramResourceLocation or
GetProgramResourceLocationIndex is considered to match an active variable
if:
[...]
* if the string identifies the base name of an active array, where the
string would exactly match the name of the variable if the suffix
"[0]" were appended to the string;
[...]
"
Fixes the following two dEQP-GLES31 tests:
dEQP-GLES31.functional.program_interface_query.shader_storage_block.resource_list.block_array
dEQP-GLES31.functional.program_interface_query.shader_storage_block.resource_list.block_array_single_element
v2:
- Add AoA support (Timothy)
- Apply it too for GetUniformLocation(), GetUniformName() and others
because ARB_program_interface_query says that they are equivalent
to GetProgramResourceLocation() and GetProgramResourceName() (Tapani)
Signed-off-by: Samuel Iglesias Gonsalvez <siglesias@igalia.com>
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
This paves the way for copy propagating our unpacks. We end up with a
small change on shader-db:
total instructions in shared programs: 89390 -> 89251 (-0.16%)
instructions in affected programs: 19041 -> 18902 (-0.73%)
which appears to be because we no longer convert MOVs for an FMAX dst,
r4.unpack, r4.unpack (instead of the previous MOV dst, r4.unpack), and
this ends up with a slightly better schedule.
At one point I thought packs and unpacks were in the same field of the
instruction. They aren't. These instructions therefore never cause a
pack.
total instructions in shared programs: 89472 -> 89390 (-0.09%)
instructions in affected programs: 15261 -> 15179 (-0.54%)
When a TCS is present, the TES input gl_PatchVerticesIn is actually a
constant - it's simply the # of output vertices specified by the TCS
layout qualifiers. So, we can replace the system value with a constant,
which may allow further optimization, and will likely be more efficient.
If the TCS is absent, we can't do this optimization.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Previously we could create a renderbuffer with format
MESA_FORMAT_R8G8B8A8_UNORM, convert that renderbuffer to an EGLImage,
then FAIL to convert the EGLImage back to a renderbuffer because
reasons. Just use the same check in
intel_image_target_renderbuffer_storage that brw_render_target_supported
uses.
There are more checks in brw_render_target_supported, but I don't think
they are necessary here. A different approach would be to refactor
brw_render_target_supported to take rb->Format and rb->NumSamples as
parameters (instead of a gl_renderbuffer) and use the new function here.
Fixes:
ES2-CTS.gtf.GL2ExtensionTests.egl_image.egl_image
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
Tested-by: Tapani Pälli <tapani.palli@intel.com>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=92476
Cc: "10.3 10.4 10.5 10.6 11.0" <mesa-stable@lists.freedesktop.org>
This is more optimal as it means we no longer have to upload the same set
of ABO surfaces to all stages in the program.
This also fixes a bug where since commit c0cd5b var->data.binding was
being used as a replacement for atomic buffer index, but they don't have
to be the same value they just happened to end up the same when binding is 0.
Reviewed-by: Francisco Jerez <currojerez@riseup.net>
Reviewed-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
Cc: Ilia Mirkin <imirkin@alum.mit.edu>
Cc: Alejandro Piñeiro <apinheiro@igalia.com>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=90175
f16c intrinsic can only be emitted when AVX is used. So when we disable AVX
due to forcing 128bit vectors we must not use this intrinsic (depending on
llvm version, this worked previously because llvm used AVX even when we didn't
tell it to, however I've seen this fail with llvm 3.3 since
718249843b which seems to have the side effect
of disabling avx in llvm albeit it only touches sse flags really, but
with ea421e919a it's now really disabled).
Albeit being able to use AVX with 128bit vectors also would have its uses, the
code as is really was meant to emulate jit code creation for less capable cpus.
v2: add some (ifdefed out) missing de-featuring options for simulating
less capable cpus.
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
Patch adds additional check to make sure we don't return locations for
structures or arrays of structures.
From page 79 of the OpenGL 4.2 spec:
"A valid name cannot be a structure, an array of structures, or any
portion of a single vector or a matrix."
v2: use without-array() to simplify code (Timothy)
No Piglit or CTS regressions observed.
Signed-off-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
One instruction instead of four, and it turns out you do this a lot for
the Over operator.
total uniforms in shared programs: 32168 -> 32087 (-0.25%)
uniforms in affected programs: 318 -> 237 (-25.47%)
total instructions in shared programs: 89830 -> 89472 (-0.40%)
instructions in affected programs: 6434 -> 6076 (-5.56%)
I left the function to obtain the revision because it is, and will continue to
be useful in the future. I'd rather not have to dig it up every time we need it.
Comments left at the implementation to say as much.
This was accidentally left here when I moved the early platform support:
commit 28ed1e08e8
Author: Ben Widawsky <benjamin.widawsky@intel.com>
Date: Fri Aug 7 13:58:37 2015 -0700
i965/skl: Remove early platform support
Signed-off-by: Ben Widawsky <benjamin.widawsky@intel.com>
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
According to piglit/xonotic/neverball/stc, blend/rasterize/zsa state
will always be bound (never null). And the null checks were in-
consistent anyways, so remove them.
Signed-off-by: Rob Clark <robclark@freedesktop.org>
Uses the DCC buffer instead of the CMASK buffer. The ELIMINATE_FAST_CLEAR
still works. Furthermore, with DCC compression we can directly clear
to a limited set of colors such that we do not need a postprocessing step.
v2 Marek: check dcc_buffer && dirty_level_mask in set_sampler_view
Signed-off-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Signed-off-by: Marek Olšák <marek.olsak@amd.com>
Can't see why anyone would ever want to use this, but it was clearly broken.
This fixes the piglit texwrap offset test using this combination.
Reviewed-by: Dave Airlie <airlied@redhat.com>
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
When using nearest filtering and clamp / clamp to edge wrapping results could
be wrong for negative offsets. Fix this by adding the offset before doing
the conversion to int coords (could also use floor instead of trunc int
conversion but probably more complex on "typical" cpu).
This fixes the piglit texwrap offset failures with this filter/wrap combo
(which only leaves the linear/mirror repeat combination broken).
Reviewed-by: Dave Airlie <airlied@redhat.com>
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
For vertex/geometry shader sampling, this is the same as for llvmpipe - just
use the original resource target.
For fragment shader sampling though (which does not use first-layer based mip
offsets) adjust the sampling code to use first_layer in the non-array cases.
While here also fix up some code which looked wrong wrt buffer texel fetch
(no piglit change).
Reviewed-by: Dave Airlie <airlied@redhat.com>
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
Just need to use resource target not view target when calculating
first-layer based mip offsets. (This is a gl specific problem since
d3d10 does not distinguish between non-array and array resources neither
at the resource nor view level, only at the shader level.)
Fixes new piglit arb_texture_view sampling-2d-array-as-2d-layer test.
Reviewed-by: Dave Airlie <airlied@redhat.com>
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
This patch was originally written before stoney support
was merged. Add stoney.
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
As the alignment requirements can be 32 KiB or more, also adding
an aligned buffer creation function.
DCC is disabled for textures that can be shared as sharing the
DCC buffers has not been implemented yet.
Signed-off-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Signed-off-by: Marek Olšák <marek.olsak@amd.com>
Out of 7063 shaders from my shader-db:
- 6564 (93%) shaders don't have any state parameters.
- 347 (5%) shaders have 1 state parameter for WPOS lowering.
- The remaining 2% have more state parameters, usually matrices.
Reviewed-by: Brian Paul <brianp@vmware.com>
The edgeflag comes in as ubyte with glEdgeFlagPointer but as float with
plain immediate glEdgeFlag. Avoid reading bytes that weren't meant for
the edgeflag in the pointer case.
Fixes intermittent failures with gl-2.0-edgeflag piglit (and valgrind
complaints about reading uninitialized memory).
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Cc: mesa-stable@lists.freedesktop.org
We can't do this all the time, because you want blending to be done in
linear space, and sRGB would lose too much precision being done in 4x8.
The win on instructions is pretty huge when you can, though.
total uniforms in shared programs: 32065 -> 32168 (0.32%)
uniforms in affected programs: 327 -> 430 (31.50%)
total instructions in shared programs: 92644 -> 89830 (-3.04%)
instructions in affected programs: 15580 -> 12766 (-18.06%)
Improves openarena performance at 1920x1080 from 10.7fps to 11.2fps.
This corresponds to instructions used on vc4 for its blending inside of
shaders. I've seen these opcodes on other architectures before, but I
think it's the first time these are needed in Mesa.
v2: Rename to 'u' instead of 'i', since they're all 'u'norm (from review
by jekstrand)
Reviewed-by: Francisco Jerez <currojerez@riseup.net>
Write groups of enabled components together.
Signed-off-by: Kristian Høgsberg Kristensen <krh@bitplanet.net>
Now that we don't read each component one-by-one, we don't need the
temoprary vgrf for the offset. More importantly, this register was type
UD while the nir source was type D. This broke copy propagation and left
a redundant MOV in the generated code.
Reviewed-by: Francisco Jerez <currojerez@riseup.net>
Signed-off-by: Kristian Høgsberg Kristensen <krh@bitplanet.net>
The scalar destination registers break copy propagation. Instead compute
the results to a regular register and then reference a component when we
later use the result as a source.
Reviewed-by: Francisco Jerez <currojerez@riseup.net>
Signed-off-by: Kristian Høgsberg Kristensen <krh@bitplanet.net>
The emit_untyped_read and emit_untyped_write helpers already uniformize
the surface index argument. No need to do it before calling them.
Reviewed-by: Matt Turner <mattst88@gmail.com>
Signed-off-by: Kristian Høgsberg Kristensen <krh@bitplanet.net>
The destination for SHADER_OPCODE_FIND_LIVE_CHANNEL is always a UD
register. When we replace the opcode with a MOV, make sure we use a UD
immediate 0 so copy propagation doesn't bail because of non-matching
types.
Reviewed-by: Matt Turner <mattst88@gmail.com>
Signed-off-by: Kristian Høgsberg Kristensen <krh@bitplanet.net>
Instead of looping through single-component reads, read all components
in one go.
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Signed-off-by: Kristian Høgsberg Kristensen <krh@bitplanet.net>
We always set the mask to 0xffff, which is what it defaults to when no
header is present. Let's drop the header instead.
v2: Only remove header for untyped reads. Typed reads always need the
header.
Reviewed-by: Francisco Jerez <currojerez@riseup.net>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Signed-off-by: Kristian Høgsberg Kristensen <krh@bitplanet.net>
Commit f17b78 added an alternative reads_flag(channel) that returned
if the instruction was reading a specific channel flag. By mistake it
only took into account the predicate, but when the opcode is
VS_OPCODE_UNPACK_FLAGS_SIMD4X2 there isn't any predicate, but the flag
are used.
That mistake caused some regressions on old hw. More information on
this bug:
https://bugs.freedesktop.org/show_bug.cgi?id=92621
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
This is a little bit like the mprotect-based fencing I've experimented
with, but it's simple and low overhead. The downside is that only catches
writes, not reads.
It didn't catch any bad writes on a current piglit run, but may be useful
in the future.
Move scratch_size out of ilo_state_shader_kernel_info and
ilo_state_compute_interface_info. A scratch space is shared by all
kernels/interfaces. Update builder to emit relocs for scratch bos.
Location has never been able to be a negative value because it has
always been validated in the parser.
Also the linker doesn't check for negatives like the comment claims.
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
virgl/vtest is a swrast driver that allows the
virgl acceleration to be tested without having
a virtual machine.
The backend has a unix socket server that
this connects to.
This is run by setting
LIBGL_ALWAYS_SOFTWARE=y
GALLIUM_DRIVER=virpipe
In this mode all renderering is sent over
a socket to the remote renderer, and the
results are readback and copies to the screen
using drisw. This works well enough to develop
new features and to help debug.
Signed-off-by: Dave Airlie <airlied@redhat.com>
virgl is the 3D acceleration backend for the
virtio-gpu shipping with qemu.
The 3D acceleration is designed around gallium
and TGSI as the virtualisation layer. The backend
renderer translates the virgl interface into
OpenGL currently.
This is the initial import of the driver to mesa.
The kernel driver portions are lined up for drm-next.
Currently this driver supports up to GL3.3 and some
misc extensions if the host driver exposes it. It is
planned to iterate the virgl API to new GL levels
as mesa host drivers gain features.
v2: fix resource tracking across flushes to avoid
->bind hack in mapping.
consolidate mapping and waiting code for transfers.
use u_range for dirt tracking.
handle larger shaders in protocol.
include virtgpu_drm.h in mesa for now.
add translation layer for gallium tgsi to virgl tgsi.
Signed-off-by: Dave Airlie <airlied@redhat.com>
This is used to detect error in virgl if we overflow the shader
dumping buffers.
v2: return a bool.
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
This adds support to the parser to accept hex values as floats,
and then adds support to the dumper to allow the user to select
to dump float as 32-bit hex numbers.
This is required to get accurate values for virgl use of TGSI.
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
On ultra high resolution modes, the preemptive flush flag can be
set midway through command submission, a condition that cannot be
recovered from a flush-retry, causing rendering artifacts.
This patch prevents a preemtive_flush until a draw has been
emitted.
Signed-off-by: Sinclair Yeh <syeh@vmware.com>
Reviewed-by: Thomas Hellstrom <thellstrom@vmware.com>
Reviewed-by: Charmaine Lee <charmainel@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
The svga device doesn't directly support quads, quad strips or polygons
so we have to convert those types to indexed triangle lists. But we
can sometimes avoid that if we're drawing flat/constant-colored prims
and we don't have to worry about provoking vertex.
Reviewed-by: Charmaine Lee <charmainel@vmware.com>
Reviewed-by: José Fonseca <jfonseca@vmware.com>
Provoking vertex comes into play when doing flat shading. But if we know
that all fragments in a primitive are the same color, the provoking vertex
doesn't matter. Check for that case and use whichever provoking vertex
convention is supported by the device.
This avoids generating an index buffer to do the PV conversion.
Reviewed-by: Charmaine Lee <charmainel@vmware.com>
Reviewed-by: José Fonseca <jfonseca@vmware.com>
Examine the fragment shader to try to detect TGSI shaders which use
"MOV OUT[0], CONST[i]" to write a constant value for the fragment color.
In this case, all fragments will have the same color (unless blending is
enabled).
This is a common case for OpenGL code such as: glColor(), glBegin(),
glVertex(), ..., glEnd() when lighting/fog/etc are disabled. In this
case, the Mesa/gallium state tracker actually generates a simple
"MOV OUT[0], CONST[i]" fragment shader.
This will be used by the next commit to avoid provoking vertex conversion
(creating/rewriting an index buffer) when drawing flat-shaded primitives.
Reviewed-by: Charmaine Lee <charmainel@vmware.com>
Reviewed-by: José Fonseca <jfonseca@vmware.com>
The st_RasterPos() function goes to great pains to implement the
rasterpos transformation. It basically uses gallium's draw module to
execute the vertex shader to draw a point, then capture that point's
attributes.
But glRasterPos isn't typically used with a vertex shader so we can
usually use the old/fixed-function implementation which is a lot simpler
and faster.
This can add up for legacy apps that make a lot of calls to glRasterPos.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
We'll remove it from the tnl module next. By lifting this code into core
Mesa we can use it from the gallium state tracker.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Instead of calling memcpy() 'n' times, we can do it all at once since
the source and dest regions are all contiguous.
Reviewed-by: Matt Turner <mattst88@gmail.com>
The section for UVD 2 and older was not updated
when HEVC support was added. Reported by Kano
on irc.
v2: integrate the UVD2 and older checks into the
main switch statement.
v3: handle encode checking as well. Encode is
already checked in the top case statement, so
drop encode checks in the lower case statement.
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Cc: mesa-stable@lists.freedesktop.org
v2: externalize pred_ctrl_align16 from brw_disasm.c instead of adding
a copy on brw_vec4.c, as suggested by Matt Turner
Reviewed-by: Matt Turner <mattst88@gmail.com>
The complete way to do this would be parse INTEL_DEBUG and
print the output if DEBUG_VS (or a new one) is present
(see intel_debug.c).
But that seems like an overkill for the unit tests, that
after all, the most common use case is being run when
calling make check.
v2: use the same idea for the fs counterpart too, as suggested by
Matt Turner
Reviewed-by: Matt Turner <mattst88@gmail.com>
This include the same tests coming from test_fs_cmod_propagation, (non
vector glsl types included) plus some new with vec4 types, inspired on
the regressions found while the optimization was a work in progress.
Additionally, the check of number of instructions after the
optimization was changed from EXPECT_EQ to ASSERT_EQ. This was done to
avoid a crash on failing tests that expected no optimization, as after
checking the number of instructions, there were some checks related to
this last instruction opcode/conditional mod.
v2: update tests after Matt Turner's review of the optimization pass
v3: tweaks on the tests (mostly on the comments), after Matt Turner's
review
Reviewed-by: Matt Turner <mattst88@gmail.com>
vec4 port of fs_cmod_propagation.
Shader-db results (no vec4 grepping):
total instructions in shared programs: 6240413 -> 6235841 (-0.07%)
instructions in affected programs: 401933 -> 397361 (-1.14%)
total loops in shared programs: 1979 -> 1979 (0.00%)
helped: 2265
HURT: 0
v2: remove extra space and combine two if blocks, as suggested by
Matt Turner
v3: add condition check to bail out if current inst and inst being
scanned has different writemask, as pointed by Matt Turner
v3: updated shader-db numbers
v4: remove block from foreach_inst_in_block_*_starting_from after
commit 801f151917
Reviewed-by: Matt Turner <mattst88@gmail.com>
vec4_live_variables tracks now each flag channel independently, so
vec4_dead_code_eliminate can update the writemask of null registers,
based on which component are alive at the moment. This would allow
vec4_cmod_propagation to optimize out several movs involving null
registers.
v2: added support to track each flag channel independently at vec4
live_variables, as v1 assumed that it was already doing it, as
pointed by Francisco Jerez
v3: general cleaningn after Matt Turner's review
Reviewed-by: Matt Turner <mattst88@gmail.com>
Gen8+ lifted the register region restriction that an instruction whose
destination spans two registers must have sources that also span two
registers.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
The AND and SHR produce a scalar value that we had been replicating
across $dispatch_width channels. The immediate MOV produces only four
useful channels of data.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
Not a functional difference, but register is loaded with a signed
immediate (V) and added to a signed type (D) producing a signed result
(D).
Also change the type of g0 to allow for compaction.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
We implement textureQueryLevels (which takes no arguments, save the
sampler) using the resinfo message (which takes an argument of LOD).
Without initializing it, we'd generate a MOV from the null register to
load the LOD argument.
Essentially the same logic applies to texture. A vertex shader cannot
compute derivatives and so cannot produce an LOD, so TXL with an LOD of
0.0 is used.
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
An untyped surface read is volatile because it might be affected by a
write.
In the ES31-CTS.compute_shader.resources-max test, two back to back
read/modify/writes of an SSBO variable looked something like this:
r1 = untyped_surface_read(ssbo_float)
r2 = r1 + 1
untyped_surface_write(ssbo_float, r2)
r3 = untyped_surface_read(ssbo_float)
r4 = r3 + 1
untyped_surface_write(ssbo_float, r4)
And after CSE, we had:
r1 = untyped_surface_read(ssbo_float)
r2 = r1 + 1
untyped_surface_write(ssbo_float, r2)
r4 = r1 + 1
untyped_surface_write(ssbo_float, r4)
Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
v2: remove useless source_stencil_to_render_target (Ken)
Squash in the actual packing function, which also got to
v2:
Move the definition of the OPCODE outside of FB_WRITE opcodes (Matt)
Reorder the regioning to be in VWH order (Matt)
Don't retype src in the backend, just assert instead (Matt)
Rename the debug prints to something better (Matt)
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Gen9 adds the ability to write out a stencil value, so we need to expand the
virtual payload by one. Abstracting this now makes that change easier to read.
I was admittedly confused early on about some of the hardcoding. If people
believe the resulting code is inferior, I am not super attached to the patch.
v2:
Remove explicit numbering from the enumeration (Matt).
Use a real naming scheme, and reference it in the opcode definition (Curro)
Add a missed hardcoded logical position in get_lowered_simd_width (Ben)
Add an assertion to make sure the component numbering is correct (Ben)
Cc: Matt Turner <mattst88@gmail.com>
Cc: Francisco Jerez <currojerez@riseup.net>
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Before the change "tgsi/scan: use properties for clip/cull distance
writemasks", the tgsi_shader_info::num_written_clipdistance field
was a multiple of four, now it's an accurate count. In the svga
driver, we need a minor change to the loop test.
Reviewed-by: Charmaine Lee <charmainel@vmware.com>
It's stored in bits 31:27 of g1 (along with the URB handles).
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
Unlike the vs/wm structs, brw_gs_compile is actually useful: it contains
the input VUE map and information about the control data headers.
Passing this in allows us to share that code in brw_gs.c, and calculate
them before deciding on vec4 vs. scalar mode, as it's independent of
that choice.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
This patch introduces a brw->scalar_gs flag, similar to brw->scalar_vs,
which controls whether or not to use SIMD8 geometry shaders.
For now, we control it via a new environment variable, INTEL_SCALAR_GS.
This provides a convenient way to try it out.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
Geometry shaders have additional header data at the beginning of their
output URB entries. Shaders that use EndPrimitive() or multiple streams
have a control data header; shaders with a dynamic vertex count have an
additional vec4 slot to hold the 32-bit vertex count (and 96 bits of
padding).
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
The GS will emit a bunch of vertices, and we don't want to do an EOT
prematurely. We'll emit GS_OPCODE_THREAD_END when we want to terminate
the thread.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
GS doesn't have ClampVertexColor, and we don't want to go through VS
structures.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
Tessellation shaders and SIMD8 geometry shaders may need to resort to
the pull model for inputs at times. When set, the state upload code
will tell the hardware to provide URB handles for input data.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
In scalar mode, geometry shader inputs can easily take up hundreds of
registers. This makes pushing VUE entries impractical; we'll need to
resort to the pull model in some cases.
To support this, we introduce a new opcode corresponding to the "URB
Read SIMD8" message.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
In the vec4 backend, we have a vec4_instruction::urb_write_flags field.
There are many kinds of flags for SIMD4x2 messages.
However, there are really only two (per-slot offset, use channel masks)
for SIMD8 messages. Rather than adding a boolean flag for per-slot
offsets (polluting all instructions), I decided to just make three new
opcodes.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
This commit moves the large pile of setup calculations we have to do for
geometry shaders out of brw_gs_emit and into brw_compile_gs. This has a
couple of nice implications. First, it's less work that the caller of
brw_compile_gs has to do. Second, it's consistent with the vertex and
fragment stages. Finally, it allows us to put brw_gs_compile back behind
the API boundary where it belongs.
v2 (Jason Ekstrand):
- Pull the changes to use nir info into a separate patch
- Put brw_gs_compile into brw_shader.h rather than brw_vec4_gs_visitor.h
so that we can use it for scalar GS.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Previously, we were pulling bits from GL data structures in order to set up
the prog_data. However, in this brave new world of NIR, we want to be
pulling it out of the NIR shader whenever possible. This way, we can move
all this setup code into brw_compile_gs without depending on the old GL
stuff.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
All the documentation I can find says that this bit (and functionality) only
exists on SKL+. Since the bit isn't yet used, there is no real impact here.
The original code was added by Ken here (a surprisingly long time ago):
commit f3c6d6f1e1
Author: Kenneth Graunke <kenneth@whitecape.org>
Date: Thu Nov 29 21:00:27 2012 -0800
i965: Update 3DSTATE_PS, 3DSTATE_WM, and add 3DSTATE_PS_EXTRA.
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Documentation is sparse, but it appears to have existed on G45 and ILK
as a second bit extension of the mask_control field. Setting the pair of
bits to 0b11 enables "NoCMask".
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
It only exists on Gen6+, and the next patches will add compaction
support for the (unused) field in the same location on earlier
platforms.
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
The next commit will add assertions dependent on devinfo->gen.
Use compact()/uncompact() macros where possible, like the 3-src code
does.
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Otherwise we'd emit a MOV from the null register (which isn't allowed).
Helps 24 programs in shader-db (the geometry shaders in GSCloth):
instructions in affected programs: 302 -> 262 (-13.25%)
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
GNU make predefines RM to rm -f but this is not required by POSIX
so ensure that RM is set. This fixes "make clean" on OpenBSD.
v2: use AC_CHECK_PROG
Signed-off-by: Jonathan Gray <jsg@jsg.id.au>
CC: "10.6 11.0" <mesa-stable@lists.freedesktop.org>
Reviewed-by: Emil Velikov <emil.l.velikov@gmail.com>
In bfdae9149e I disabled the opt_sampler_eot optimisation for TG4
message types because I found by experimentation that it doesn't work.
I wrote in the comment that I couldn't find any documentation for this
problem. However I've now found the documentation and it has
additional restrictions on further message types so this patch updates
the comment and adds the others.
Reviewed-by: Matt Turner <mattst88@gmail.com>
Since 49374fab5d these macros no longer actually use the block
argument. I think this is worth doing to make the macros easier to use
because they already have really long names and a confusing set of
arguments.
Reviewed-by: Matt Turner <mattst88@gmail.com>
This fixes assigning explicit locations in the CTS test:
ES31-CTS.explicit_uniform_location.uniform-loc-arrays-of-arrays
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
process_parameters() will now be called earlier because we need
actual_parameters processed earlier so we can use it with
match_subroutine_by_name() to get the subroutine variable, we need
to do this inside the recursive function generate_array_index() because
we can't create the ir_dereference_array() until we have gotten to the
outermost array.
For the remainder of the array dimensions the type doesn't matter so we
can just use the existing _mesa_ast_array_index_to_hir() function to
process the ast.
Reviewed-by: Dave Airlie <airlied@redhat.com>
Check current_var directly instead of using the passed in record_type.
This fixes following failing CTS test:
ES31-CTS.explicit_uniform_location.uniform-loc-types-structs
No Piglit regressions.
Signed-off-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Marta Lofstedt <marta.lofstedt@intel.com>
Reviewed-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
UniformRemapTable is used only for remapping user specified uniform
locations to driver internally used ones, shader storage buffer
variables should not utilize uniform locations.
Signed-off-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Marta Lofstedt <marta.lofstedt@intel.com>
Reviewed-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
A third instance of this was needed but missed in the previous commit.
Return 32 as for the two other cases.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Charmaine Lee <charmainel@vmware.com>
When the draw module splits long line loops, the sections are emitted
as line strips. But the primitive type wasn't set correctly so each
section was being drawn as a loop, introducing extra line segments.
To fix this, we pass a new DRAW_LINE_LOOP_AS_STRIP flag to the run()
function. The linear/elt_run() functions have to check for this flag
and set their primitive type accordingly.
No piglit regressions. Fixes piglit's lineloop with -count 4097 or
higher.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=81174
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Patch just does some refactoring to make the code look better. No
functional changes in here.
Signed-off-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
When a long GL_LINE_LOOP prim was split across primitives we drew
stray lines. See previous commit for details.
This patch converts GL_LINE_LOOP prims into GL_LINE_STRIP prims so
that drivers don't have to worry about the _mesa_prim::begin/end flags.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=81174
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
Acked-by: Sinclair Yeh <syeh@vmware.com>
When long GL_LINE_LOOP primitives don't fit in one vertex buffer they
have to be split across buffers. The code to do this was basically correct
but drivers had to pay special attention to the _mesa_prim::begin,end flags
in order to draw the sections of the line loop properly. Apparently, the
only drivers to do this were those using the old 'tnl' module for software
vertex processing.
Now we convert the split pieces of GL_LINE_LOOP prims into GL_LINE_STRIP
primitives so that drivers don't have to worry about the special begin/end
flags. The only time a driver will get a GL_LINE_LOOP prim is when the
whole thing fits in one vertex buffer.
Mostly fixes bug 81174, but not completely. There's another bug somewhere
in the src/gallium/auxiliary/draw/ code. If the piglit lineloop test is
run with -count 4096, rendering is correct, but with -count 4097 there are
stray lines. 4096 is a magic number in the draw code (search for "4096").
Also note that this does not fix long line loops in display lists. The
next patch fixes that.
v2: fix incorrect -1 in vbo_compute_max_verts(), per Charmaine. Remove
incorrect assertion which was added in vbo_copy_vertices().
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=81174
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=49779
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=28130
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
Reviewed-by: Sinclair Yeh <syeh@vmware.com>
As before, use a new 'last_prim' pointer to simplify things. Plus, add
some const qualifiers.
v2: use 'sz' in another place, per Sinclair. And update subject line.
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
Reviewed-by: Sinclair Yeh <syeh@vmware.com>
Use a new 'last_prim' pointer to simplify things.
v2: remove unneeded assert(exec->vtx.prim_count > 0)
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
Reviewed-by: Sinclair Yeh <syeh@vmware.com>
Whenever we got a glColor, glNormal, glTexCoord, etc. call outside a
glBegin/End pair, we'd immediately map a vertex buffer to begin
accumulating vertex data. In some cases, such as with display lists,
this led to excessive vertex buffer mapping. For example, if we have
a display list such as:
glNewList(42, GL_COMPILE);
glBegin(prim);
glVertex2f();
...
glVertex2f();
glEnd();
glEndList();
Then did:
glColor3f();
glCallList(42);
We'd map a vertex buffer as soon as we saw glColor3f but we'd never
actually write anything to it. Note that the vertex position data
was put into a vertex buffer during display list compilation.
With this change, we delay mapping the vertex buffer until we actually
have a vertex to write to it (triggered by a glVertex() call). In the
above case, we no longer map a vertex buffer when setting the color and
calling the list.
For drivers such as VMware's, reducing buffer mappings gives improved
performance.
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
If we didn't find a gallium surface format that exactly matched the
glDrawPixels format/type combination, we used some other 32-bit packed
RGBA format and swizzled the whole image in the mesa texstore/format code.
That slow path can be avoided in some common cases by using the
pipe_samper_view's swizzle terms to do the swizzling at texture sampling
time instead.
For now, only GL_RGBA/ubyte and GL_BGRA/ubyte combinations are supported.
In the future other formats and types like GL_UNSIGNED_INT_8_8_8_8 could
be added.
v2: fix incorrect swizzle setup (need to invert the tex format's swizzle)
Reviewed by: Jose Fonseca <jfonseca@vmware.com>
So that we can use it directly from the mesa/gallium state tracker.
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
Reviewed-by: Sinclair Yeh <syeh@vmware.com>
Before, if make_texture() or st_create_texture_sampler_view() failed
we silently no-op'd the glDrawPixels. Now, set GL_OUT_OF_MEMORY.
This also allows us to un-nest a bunch of code.
v2: also check if allocation of sv[1] fails, per Jose.
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
src/mesa/drivers/dri/i965/brw_program.c:94:39:
warning: passing argument 1 of ‘_mesa_init_gl_program’ from incompatible
pointer type [-Wincompatible-pointer-types]
return _mesa_init_gl_program(&prog->program, target, id);
^
Runtime was unaffected as brw_geometry_program is subclassed from
gl_geometry_program, thus the address passed was the same.
Fixes: bcb56c2c69 (program: convert _mesa_init_gl_program() to take
struct gl_program *)
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
This avoids a serious r600g bug leading to a GPU hang.
The chances this bug will get fixed are pretty low now.
I deeply regret listening to others and not pushing this patch, leaving
other users with a GPU-crashing driver. Yes, it should be fixed
in the compiler and it's ugly, but users couldn't care less about that.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=86720
Cc: 11.0 10.6 <mesa-stable@lists.freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
This exposes more information to NIR's optimization, and should be
particularly useful when we do range-based optimization.
total uniforms in shared programs: 32066 -> 32065 (-0.00%)
uniforms in affected programs: 21 -> 20 (-4.76%)
total instructions in shared programs: 93104 -> 92630 (-0.51%)
instructions in affected programs: 31901 -> 31427 (-1.49%)
The TGSI usage mask can't be used, because these are declared as an output
array of 2 elements.
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Brian Paul <brianp@vmware.com>
The "current" shader pointer is moved from the CSO to the context, so that
the CSO is mostly immutable.
The only drawback is that the "current" pointer isn't saved when unbinding
a shader and it must be looked up when the shader is bound again.
This is also a prerequisite for multithreaded shader compilation.
Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>
All tests pass. We don't need to do much - just set CUBE if the view
target is CUBE or CUBE_ARRAY, otherwise set the resource target.
The reason this can be so simple is that texture instructions
have a greater effect on the target than the sampler view.
Thanks Glenn for the piglit test.
Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>
This also removes the validation from the parser as it is not required
and once arb_enhanced_layouts comes along we wont be able to do validation
on the stream qualifier in the parser anyway as it adds constant expression
support to the stream qualifier.
Reviewed-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
Cc: 11.0 <mesa-stable@lists.freedesktop.org>
Fixes regression cased by bb5aeb8549
We don't care about the swizzle when building the name so just skip over it.
Tested-by: Markus Wick <markus@selfnet.de>
Reviewed-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
NIR considers bcsel to produce and consume unsigned types, leading to
SEL instructions operating on unsigned types when the data is really
floating-point. Previous to this patch, saturate propagation would
happily transform
(+f0) sel g20:UD, g30:UD, g40:UD
mov.sat g50:F, g20:F
into
(+f0) sel.sat g20:UD, g30:UD, g40:UD
mov g50:F, g20:F
But since the meaning of .sat is dependent on the type of the
destination register, this is not valid.
Instead, allow saturate propagation to change the types of dest/source
on instructions that are simply copying data in order to propagate the
saturate modifier.
Fixes bad code gen in 158 programs.
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
Make them members of fs_inst/vec4_instruction for use elsewhere.
Also fix the fs version to check that dst.type == src[1].type and for
!saturate.
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
At this point, the compiler API has been substantially simplified. In the
spirit of Kristian's making a compiler library, this commit makes a single
header file that contains, more-or-less, the entire compiler API.
There's still a bit of cleanup to do particularly in the area of geometry
shaders. However, this gets us much closer to having a separate compiler.
Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
This commit moves the common/modern stuff. Some legacy stuff such as
setting use_alt_mode was left because it needs to know whether or not we're
an ARB program.
Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
This commit removes all dependence on GL state by getting rid of the
brw_context parameter and the GL data structures.
Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
This commit removes all dependence on GL state by getting rid of the
brw_context parameter and the GL data structures. Unfortunately, we still
have to pass in the gl_shader_program for gen6 because it's needed for
transform feedback.
Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
This commit removes all dependence on GL state by getting rid of the
brw_context parameter and the GL data structures.
v2 (Jason Ekstrand):
- Patch use_legacy_snorm_formula through as a function argument rather
than trying to go through the shader key.
Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
This commit removes all dependence on GL state by getting rid of the
brw_context parameter and the GL data structures.
Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
This way we can have other stage-specific info without consuming too much
extra space. While we're at it, we make sure that the geometry info is
only set if we're actually a goemetry shader.
Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
Undoes early hacks, and ensures nir/glsl_types.cpp is built once, and
only once.
The root problem is that SCons doesn't know about NIR nor any source
file in the NIR_FILES source list.
Tested with libgl-gdi and libgl-xlib scons targets.
Reviewed-by: Brian Paul <brianp@vmware.com>
Spotted by Roland. Luckily, this code should never really be hit
since the const buffer size and offset should already be multiples
of 16. I could probably add more assertions to that effect, but
let's just fix the arithmetic for now.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
The latter holds both UBOs and SSBOs, but here we only want UBOs.
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
The latter holds both UBOs and SSBOs, but here we only want UBOs.
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
This is the only place in the driver where we use this. Since we now work
with separate index spaces, always use NumUniformBlocks and
NumShaderStorageBlocks instead of NumBufferInterfaceBlocks to be more
consistent with the rest of the code.
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Now that we have separate index spaces for UBOs and SSBOs we do not need
to iterate through BufferInterfaceBlocks any more, we can just take the
UBO count directly from NumUniformBlocks.
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
It looks like binding a constant buffer on compute overwrites the 3D
state. To avoid that, we already re-bind all the 3D constant buffers
after launching a compute grid but this is not enough.
Binding the constant buffer of input parameters for the compute state at
initialization corrupts the 3D constant buffers, and it's just useless
to bind it because this is not needed until we really launch a grid.
This fixes some piglit regressions related to interpolation tests
introduced in "nvc0: enable compute support by default on Fermi".
Fixes: 00d6186 (nvc0: enable compute support by default on Fermi)
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
Marek made core Mesa call ProgramStringNotify(), which solves this
properly. The hack is no longer needed.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
VS, GS, and FS continue doing the same thing they did before. We can
simplify the FS code a bit because it is always scalar.
Compute shaders now assert that there are no outputs instead of doing
a loop over 0 outputs.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
The previous version has precision issues. This can be a problem
with tessellation. Sadly, I can't find the article where I read it
anymore. I'm not sure if the unsafe-fp-math flag would be enough to revert
this.
v2: added the comment
Changing the matrix mode alone has no effect on rendering and does
not need to trigger a flush or state validation.
Reviewed-by: Eric Anholt <eric@anholt.net>
Signed-off-by: Marek Olšák <marek.olsak@amd.com>
commit a6a6a71092
Author: Rob Clark <robclark@freedesktop.org>
AuthorDate: Sat Oct 10 14:13:50 2015 -0400
glsl: (mostly) remove libglsl_util
Was a bit too ambitious on removal of libglsl_util.. it is still needed
by some of the tests.
Signed-off-by: Rob Clark <robclark@freedesktop.org>
As explained in the CUDA toolkit documentation, "a metric is a
characteristic of an application that is calculated from one or more
event values."
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
Now that NIR does not depend on glsl, we can (mostly[*]) get rid of the
libglsl_util hack.
[*] glsl_compiler is the one remaining user of libglsl_util
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
Signed-off-by: Rob Clark <robclark@freedesktop.org>
Move glsl_types into NIR, now that the dependency on glsl_symbol_table
has been split out.
Possibly makes sense to rename things at this point, but if we do that
I'd like to keep it split out into a separate patch to make git history
easier to follow (IMHO).
v2: fix android build
v3: I f***ing hate scons.. but at least it builds
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Signed-off-by: Rob Clark <robclark@freedesktop.org>
First step at untangling NIR's dependency on glsl_types without bringing
in the dependency on glsl_symbol_table. The builtin types are now in
glsl_types (which will end up in NIR), but adding them to the symbol-
table stays in builtin_types.cpp (which will not be part of NIR).
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
Signed-off-by: Rob Clark <robclark@freedesktop.org>
This was only being done in one of the two process methods.
Fixes an issue with samplers using the array size of a previous record.
Tested-by: Marek Olšák <marek.olsak@amd.com>
Cc: Jason Ekstrand <jason@jlekstrand.net>
MP counters on GF100/GF110 (compute capability 2.0) are buggy
because there is a context-switch problem that we need to fix.
Results might be wrong sometimes, be careful!
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
GF100 and GF110 chipsets are compute capability 2.0, while the other
Fermi chipsets are compute capability 2.1. That's why, some MP counters
are different between these chipsets and we need to handle variants.
Signed-off-by: Samuel Pitoiet <samuel.pitoiset@gmail.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
Compute support was not enabled by default because weird effects
on 3D state happened, but I can't reproduce them anymore.
This also enables MP performance counters by default on Fermi.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
Because we can't expose the number of hardware counters needed for each
different query, we don't want to allow more than one active query
simultaneously to avoid failure when the maximum number of counters
is reached. Note that these groups of GPU counters are currently only
used by AMD_performance_monitor.
Like for Kepler, this limits the maximum number of active queries
to 1 on Fermi.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
When a card has more than one GPC, the grid used by the compute
kernel which reads MP performance counters seems to be too small.
The consequence is that the kernel is not launched on all TPCs.
Increasing the grid size using the number of GPCs now launches
enough blocks and we can read MP performance counters of all TPCs.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
NOUVEAU_GETPARAM_GRAPH_UNITS param returns the number of GPCs, the total
number of TPCs and the number of ROP units. Note that when the DRM
version is too old the default number of GPCs is fixed to 4.
This will be used to launch the compute kernel which is used to read MP
performance counters over all GPCs.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
Memory access have to be aligned to 128-bits. Note that this
doesn't happen when the card only has TPC.
This patch fixes the following dmesg fail:
gr: GPC0/TPC1/MP trap: global 00000004 [MULTIPLE_WARP_ERRORS] warp 000f
[UNALIGNED_MEM_ACCESS]
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
For strange reasons, the signal id depends on the slot selected on Fermi
but not on Kepler. Fortunately, the signal ids are just offseted by the
slot id!
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
Queries which use more than one MP counters was misconfigured and
computing the final result was also wrong because sources need to
be configured on different hardware counters instead.
According to the blob, computing the result is now as follows:
FOR i..n
val += ctr[i] * pow(2, i)
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
On Fermi, we have one domain of 8 MP counters while we have
two domains of 4 MP counters on Kepler.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
Sequence fields are located at MP[i] + 0x20 in the buffer object.
This is used to check if result is available for MP[i].
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
Writing 0x408000 to 0x419e00 (like on Kepler) has no effect on Fermi
because we only have one domain of 8 counters. Instead, we have to
write 0x80000000.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
Writing 0x1fcb to 0x419eac is definitely not related to MP counters and
has no effect on Fermi (although this enables MP counters on Kepler).
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
The way we configure MP performance counters is going to pretty
different between Fermi and Kepler. Having two separate functions
is much better.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
Add new GALLIUM_HUD queries for:
num-shaders
num-resources
num-state-objects
num-validations
map-buffer-time
num-surface-views
num-resources-mapped
num-flushes
Most of this patch was originally written by Neha. Additional clean-ups
and num-flushes counter added by Brian Paul.
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Charmaine Lee <charmainel@vmware.com>
Silences 5 warnings of the type:
state_tracker/st_cb_program.c: In function 'st_new_program':
state_tracker/st_cb_program.c:108:7: warning: passing argument 1 of
'_mesa_init_gl_program' from incompatible pointer type [enabled by default]
return _mesa_init_gl_program(&prog->Base, target, id);
^
Reviewed-by: Emil Velikov <emil.l.velikov@gmail.com>
This reverts commit 0de5e0f3fb.
Michel Dänzer spotted two piglit regressions from the change. I suspect
that removing the FLUSH_VERTICES() actually exposed a bug elsewhere but
I don't have time to hunt down the root issue at this time.
has_shader_storage_buffer_objects() returns true also if the OpenGL
context is 4.30 or ES 3.1.
Previously, we were saying that all atomic*() GLSL builtin functions
for SSBOs were not available when OpenGL ES 3.1 context was in use.
Fixes 48 dEQP-GLES31 tests:
dEQP-GLES31.functional.ssbo.atomic.*
Signed-off-by: Samuel Iglesias Gonsalvez <siglesias@igalia.com>
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Otherwise there are problems when user overrides version and application
such as Piglit wants to detect used api with glGetString(GL_VERSION).
This makes it currently impossible to run glslparsertest tests for
OpenGL ES when using version override.
Below is example when using MESA_GLES_VERSION_OVERRIDE=3.1.
Before:
"3.1 Mesa 11.1.0-devel (git-24a1a15)"
After:
"OpenGL ES 3.1 Mesa 11.1.0-devel (git-78042ff)"
v2: only include api prefix for OpenGL ES (Boyan Ding)
Signed-off-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Cc: "11.0" <mesa-stable@lists.freedesktop.org>
Before d31f98a272 and 56e2bdbca3 we had a sigle index space for UBOs
and SSBOs, so NumBufferInterfaceBlocks would contain the combined number of
blocks, not just one kind. This means that for shader programs using both
UBOs and SSBOs, we were setting num_ssbos and num_ubos to a larger number than
we should. Since the above commits we have separate index spaces for each
so we can just get the right numbers.
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
(yes, we want PRI?64, but we want the x version rather than the u
version)
Signed-off-by: Rob Clark <robclark@freedesktop.org>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
nir_variable_create already inserts it in the right list for us so
inserting it again causes a linked list corruption.
Reviewed-by: Matt Turner <mattst88@gmail.com>
This has the better name to use. Aparently, sh->Name is usually 0.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Neil Roberts <neil@linux.intel.com>
For glBlendFunc and glBlendFuncSeparate(), the _UsesDualSrc flag
will be the same for all buffers, so no need to compute it N times.
Reviewed-by: Eric Anholt <eric@anholt.net>
A redundant call to glBlendFuncSeparateiARB() is more likely than getting
invalid values, so do the no-op check first.
Reviewed-by: Eric Anholt <eric@anholt.net>
Streamline the checking for no state change in _mesa_BlendFuncSeparate()
(and _mesa_BlendFunc()). If _BlendFuncPerBuffer is false, we only need
to check the 0th buffer state. Move argument validation after the no-op
check.
I'm looking at an app that issues about 1000 redundant glBlendFunc()
calls per frame!
Reviewed-by: Eric Anholt <eric@anholt.net>
We can skip to the end of _mesa_update_state_locked() if only the
_NEW_LINE flag is set since none of the derived state depends on it
(just like _NEW_CURRENT_ATTRIB). Note that we still call the
ctx->Driver.UpdateState() function, of course.
v2: use bitmask-based test, per Eric.
Reviewed-by: Eric Anholt <eric@anholt.net>
Changing the matrix mode alone has no effect on rendering and does
not need to trigger a flush or state validation.
Reviewed-by: Eric Anholt <eric@anholt.net>
Rather than accepting a void pointer, only to down and up cast around
it, convert the function to take the base (struct gl_program) pointer.
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
V3: use a check_*_allowed style function for requirements checking
rather than has_* which doesn't encapsulate the error message
V2: add missing 's' to the extension name in error messages
and add decimal place in version string
Reviewed-by: Marta Lofstedt <marta.lofstedt@intel.com>
This adds support for setting up the UniformBlock structures for AoA
and also adds support for resizing AoA blocks with a packed layout.
Reviewed-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
Add support for setting the max access of an unsized member
of an interface array of arrays.
For example ifc[j][k].foo[i] where foo is unsized.
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
This marks all counters in an AoA as active.
For AoA all but the innermost array are treated as separate
counters/uniforms. The Nvidia binary also goes further and
finds inactive counters in the AoA, in future we should do
this too, however this gets things working for the time being.
This change also removes the use of UniformHash for atomic counters,
this avoids having to generate name strings used as hash keys.
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
This allows the correct offset to be calculated for use in indirect
indexing of samplers.
Reviewed-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
Currently only one ir assignment is removed for each var in a single
dead code optimisation pass. This means if a var has more than one
assignment, then it requires all the glsl optimisations to be run again
for each additional assignment to be removed.
Another pass is also required to remove the variable itself.
With this change all assignments and the variable are removed in a single
pass.
Some of the arrays of arrays conformance tests that were looping
through 8 dimensions ended up with a var with hundreds of assignments.
This change helps ES31-CTS.arrays_of_arrays.InteractionFunctionCalls1
go from around 3 min 20 sec -> 2 min
ES31-CTS.arrays_of_arrays.InteractionFunctionCalls2 went from
around 9 min 20 sec to 7 min 30 sec
I had difficulty getting the public shader-db to give a consistent result
with or without this change but the results seemed unchanged at between
15-20 seconds.
Thomas Helland measured change with shader-db on his machine from
approx 117 secs to 112 secs.
V3: Simplify freeing of list as suggested by Ian, and spelling fixes.
V2: Add assert to be sure references are counted before assignments.
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Tested-By: Thomas Helland <thomashelland90@gmail.com>
Tested-by: Ian Romanick <ian.d.romanick@intel.com>
V3: move patch after fixes to ast for AoA and add const to helper
as suggested by Ian
V2: move single dimensional array detection into a helper
Signed-off-by: Timothy Arceri <t_arceri@yahoo.com.au>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
V3: Fix setting of data.location for struct AoA UBO members
V2: Handle arrays of arrays in the same way structures are handled
The ARB_arrays_of_arrays spec doesn't give very many details on how
AoA uniforms are intended to be implemented. However in the
ARB_program_interface_query spec there are details that show AoA are
intended to be handled in a similar way to structs.
Issues 7 from the ARB_program_interface_query spec:
We define rules consistent with our enumeration rules for
other complex types. For existing one-dimensional arrays, we enumerate
a single entry if the array is an array of basic types, or separate
entries for each array element if the array is an array of structures.
We follow similar rules here. For a uniform array such as:
uniform vec4 a[5][4][3];
we enumerate twenty different entries ("a[0][0][0]" through
"a[4][3][0]"), each of which is treated as an array with three elements.
This is morally equivalent to what you'd get if you worked around the
limitation in current GLSL via:
struct ArrayBottom { vec4 c[3]; };
struct ArrayMid { ArrayBottom b[3]; };
uniform ArrayMid a[5];
which would enumerate "a[0].b[0].c[0]" through "a[4].b[3].c[0]".
Reviewed-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
The ES31-CTS.compute_shader.pipeline-compute-chain test case generates
an unsigned index by using gl_LocalInvocationID.x and
gl_LocalInvocationID.y as array indices.
Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
The ES31-CTS.compute_shader.pipeline-compute-chain test case generates
an unsigned index by using gl_LocalInvocationID.x and
gl_LocalInvocationID.y as array indices.
Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
The commit shown below caused compute shaders to hit the unreachable
in the default of the switch block. Since compute shaders don't have
any inputs, we can make brw_nir_lower_inputs a no-op for CS.
commit 2953c3d761
Author: Kenneth Graunke <kenneth@whitecape.org>
Date: Fri Aug 14 15:15:11 2015 -0700
i965/vs: Map scalar VS input locations properly; avoid tons of MOVs.
Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
pipe_surface_reference have problems with deleted contexts,
so use of pipe_surface_release might be more appropriate.
Fixes Wasteland 2 Director's Cut crash on start.
Cc: mesa-stable@lists.freedesktop.org
Reviewed-by: Brian Paul <brianp@vmware.com>
The split of Uniform blocks and shader storage block only loops
up to MESA_SHADER_FRAGMENT and igonres compute shaders.
This cause segfault when running the OpenGL ES 3.1 CTS tests
with GL_ARB_compute_shader enabled.
V2: Changed to use MESA_SHADER_STAGES instead of
MESA_SHADER_COMPUTE
Reviewed-by: Francisco Jerez <currojerez@riseup.net>
Signed-off-by: Marta Lofstedt <marta.lofstedt@linux.intel.com>
Patch moves existing calculation code from shader_query.cpp to happen
during program resource list creation.
No Piglit or CTS regressions were observed during testing.
Signed-off-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
Patch adds 2 new fields to gl_uniform_storage so that we don't need to
calculate these values during runtime shader queries. This is required by
upcoming changes to free GLSL IR after linking.
Patch moves 3 booleans inside structure so that structure size stays the
same after this change.
Signed-off-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
Currently, these arrays in gl_shader and gl_shader_program hold both
UBOs and SSBOs, so this looks like a better name. We were already
using NumBufferInterfaceBlocks in gl_shader_program, so this makes
things more consistent as well.
In a later patch we will add {Num}UniformBlocks and
{Num}ShaderStorageBlocks which will contain only references to
UBOs and SSBOs respectively that will provide backends with
a separate index space for both types of objects.
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
We get these when we operate on vector variables with array accessors
(i.e. things like a[0] where 'a' is a vec4). When we call variable_referenced()
on these expressions we want to return a reference to 'a' instead of NULL.
This fixes a problem where we pass a[0] as the first argument to an atomic
SSBO function that expects a buffer variable. In order to check this, we use
variable_referenced(), but that is currently returning NULL in this case, since
the underlying rvalue is a vector_extract expression.
Tested-by: Markus Wick <markus@selfnet.de>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
NIR is typeless so this is the only way to keep track of the
type to select the proper atomic to use.
v2:
- Use imin,imax,umin,umax for the intrinsic names (Connor Abbott)
- Change message for unreachable paths (Michael Schellenberger)
Tested-by: Markus Wick <markus@selfnet.de>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
The variable 'i' is a value in [0, MAT_ATTRIB_MAX-1] so subtracting
VERT_ATTRIB_GENERIC0 gave a bogus value and we executed the default
switch clause for all loop iterations.
This doesn't fix any known issues but was clearly incorrect.
Cc: mesa-stable@lists.freedesktop.org
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
This is the result of applying several rules:
From OpenGL 4.3 spec, section 7.6.2.2 "Standard Uniform Block Layout":
"2. If the member is a two- or four-component vector with components
consuming N basic machine units, the base alignment is 2N or 4N,
respectively."
[...]
"4. If the member is an array of scalars or vectors, the base alignment
and array stride are set to match the base alignment of a single array
element, according to rules (1), (2), and (3), and rounded up to the
base alignment of a vec4."
[...]
"7. If the member is a row-major matrix with C columns and R rows, the
matrix is stored identically to an array of R row vectors with C
components each, according to rule (4)."
[...]
"When using the std430 storage layout, shader storage blocks will be
laid out in buffer storage identically to uniform and shader storage
blocks using the std140 layout, except that the base alignment and
stride of arrays of scalars and vectors in rule 4 and of structures in
rule 9 are not rounded up a multiple of the base alignment of a vec4."
In summary: vec2 has a base alignment of 2*N, a row-major mat2xY is
stored like an array of Y row vectors with 2 components each. Because
of std430 storage layout, the base alignment of the array of vectors
is not rounded up to vec4, so it is still 2*N.
Fixes 15 dEQP tests:
dEQP-GLES31.functional.ssbo.layout.single_basic_type.std430.row_major_lowp_mat2
dEQP-GLES31.functional.ssbo.layout.single_basic_type.std430.row_major_mediump_mat2
dEQP-GLES31.functional.ssbo.layout.single_basic_type.std430.row_major_highp_mat2
dEQP-GLES31.functional.ssbo.layout.single_basic_type.std430.row_major_lowp_mat2x3
dEQP-GLES31.functional.ssbo.layout.single_basic_type.std430.row_major_mediump_mat2x3
dEQP-GLES31.functional.ssbo.layout.single_basic_type.std430.row_major_highp_mat2x3
dEQP-GLES31.functional.ssbo.layout.single_basic_type.std430.row_major_lowp_mat2x4
dEQP-GLES31.functional.ssbo.layout.single_basic_type.std430.row_major_mediump_mat2x4
dEQP-GLES31.functional.ssbo.layout.single_basic_type.std430.row_major_highp_mat2x4
dEQP-GLES31.functional.ssbo.layout.single_basic_array.std430.row_major_mat2
dEQP-GLES31.functional.ssbo.layout.single_basic_array.std430.row_major_mat2x3
dEQP-GLES31.functional.ssbo.layout.single_basic_array.std430.row_major_mat2x4
dEQP-GLES31.functional.ssbo.layout.instance_array_basic_type.std430.row_major_mat2
dEQP-GLES31.functional.ssbo.layout.instance_array_basic_type.std430.row_major_mat2x3
dEQP-GLES31.functional.ssbo.layout.instance_array_basic_type.std430.row_major_mat2x4
v2:
- Add spec quote in both commit log and code (Timothy)
Signed-off-by: Samuel Iglesias Gonsalvez <siglesias@igalia.com>
Reviewed-by: Marta Lofstedt <marta.lofstedt@intel.com>
Previously, ATTR was indexed by VERT_ATTRIB_* slots; at the end of
compilation, assign_vs_urb_setup() translated those into GRF units,
and converted ATTR to HW_REGs.
This patch moves the transslation earlier, making ATTR work in terms of
GRF units from the beginning. assign_vs_urb_setup() simply has to add
the number of payload registers and push constants to obtain the final
hardware GRF number. (We can't do this earlier as those values aren't
known.)
ATTR still supports reg_offset; however, it's simply added to reg.
It's not clear whether this is valuable or not.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
The act of ensuring that there is space can cause a flush to happen,
which will emit the current screen fence. If that is the fence we're
trying to wait on, then it will have been emitted as a result of doing
the PUSH_SPACE. Don't attempt to emit it a second time.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Fixes: 8053c9208f (nouveau: avoid emitting new fences unnecessarily)
Cc: mesa-stable@lists.freedesktop.org
Fixes:
ES3-CTS.shaders.negative.constant_sequence
spec/glsl-es-3.00/compiler/global-initializer/from-sequence.vert
spec/glsl-es-3.00/compiler/global-initializer/from-sequence.frag
v2: Fix a couple copy-and-paste mistake in the spec quotations.
Suggested by Matt.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Cc: "10.6 11.0" <mesa-stable@lists.freedesktop.org>
This will be used in the next patch to enforce some language sematics.
v2: Fix inverted logic in
ast_function_expression::has_sequence_subexpression. The method
originally had a different name and a different meaning. I fixed the
logic in ast_to_hir.cpp, but I only changed the names in
ast_function.cpp.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Marta Lofstedt <marta.lofstedt@intel.com> [v1]
Reviewed-by: Matt Turner <mattst88@gmail.com>
Cc: "10.6 11.0" <mesa-stable@lists.freedesktop.org>
v2: Combine this check with the existing const and uniform checks. This
change depends on the previous patch (glsl: Only set
ir_variable::constant_value for const-decorated variables).
Fixes:
ES2-CTS.shaders.negative.initialize
ES3-CTS.shaders.negative.initialize
spec/glsl-es-1.00/compiler/global-initializer/from-attribute.vert
spec/glsl-es-1.00/compiler/global-initializer/from-uniform.vert
spec/glsl-es-1.00/compiler/global-initializer/from-uniform.frag
spec/glsl-es-1.00/compiler/global-initializer/from-global.vert
spec/glsl-es-1.00/compiler/global-initializer/from-global.frag
spec/glsl-es-1.00/compiler/global-initializer/from-varying.frag
spec/glsl-es-3.00/compiler/global-initializer/from-uniform.vert
spec/glsl-es-3.00/compiler/global-initializer/from-uniform.frag
spec/glsl-es-3.00/compiler/global-initializer/from-in.vert
spec/glsl-es-3.00/compiler/global-initializer/from-in.frag
spec/glsl-es-3.00/compiler/global-initializer/from-global.vert
spec/glsl-es-3.00/compiler/global-initializer/from-global.frag
Note: spec/glsl-es-3.00/compiler/global-initializer/from-sequence.*
still fail because the result of a sequence operator is still considered
to be a constant expression.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=92304
Reviewed-by: Tapani Pälli <tapani.palli@intel.com> [v1]
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com> [v1]
Reviewed-by: Matt Turner <mattst88@gmail.com>
Cc: "10.6 11.0" <mesa-stable@lists.freedesktop.org>
Right now we're also setting for uniforms, and that doesn't seem to hurt
things. The next patch will make general global variables in GLSL ES,
and those definitely should not have constant_value set!
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Cc: "10.6 11.0" <mesa-stable@lists.freedesktop.org>
This is the way layout(binding=xxx) works from GLSL. The old method
just happened to work (and significantly predated support for
layout(binding=xxx)), but future changes will break this.
v2: Remove some stale comments. Suggested by Matt and Chris Forbes.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Cc: "10.6 11.0" <mesa-stable@lists.freedesktop.org>
In d4a24745 (August 2012), Paul made functions calls not be constant
expressions in GLSL ES 1.00. Since this feature was added in desktop
GLSL 1.20, we believed that it was added in GLSL ES 3.00. That turns
out to be completely wrong. Built-in functions have always been allowed
as constant expressions in GLSL ES, and the patch adds the (many) spec
quotations to prove it.
While we never previously encountered this, a later patch enforces a GLSL
ES 1.00 rule that global variable initializers must be constant
expressions. Without this fix, several dEQP tests fail.
Fixes:
tests/spec/glsl-es-1.00/compiler/const-initializer/from-function.frag
tests/spec/glsl-es-1.00/compiler/const-initializer/from-function.vert
tests/spec/glsl-es-1.00/compiler/const-initializer/from-sequence-in-function.frag
tests/spec/glsl-es-1.00/compiler/const-initializer/from-sequence-in-function.vert
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Cc: "10.0 10.1 10.2 10.3 10.4 10.5 10.6 11.0" <mesa-stable@lists.freedesktop.org>
Yes, I know we don't maintain stable branches that far back, but that
*is* how far back this bug goes!
Vertex attributes of different categories (constant/per-instance/
per-vertex) go into different buffers for translation, and this is now
properly reflected in the vertex buffers passed to the driver.
Fixes e.g. piglit's point-vertex-id divisor test.
Cc: mesa-stable@lists.freedesktop.org
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Also fix style / wrong indentation along the way and make the messages
more uniform.
Reviewed-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
GLSL Spec 4.20.8, 4.3 Storage Qualifiers:
"Initializers in global declarations may only be used in declarations of
global variables with no storage qualifier, with a const qualifier or
with a uniform qualifier."
We do this for input variables, but not for output variables. AMD and NVIDIA
proprietary drivers don't allow this either.
Reviewed-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
For the VS and FS stages that use ARB_vertex_program or
ARB_fragment_program we don't have a shader program, however,
when debuging is enabled, we call brw_dump_ir like this:
brw_dump_ir("vertex", prog, &vs->base, &vp->program.Base);
where vs will be NULL (since prog is NULL).
As pointed out by Chris, this &vs->base is not really a dereference,
it simply computes a new address that just happens to be 0x0 because
the offset of base in brw_shader is 0. Then brw_dump_ir will see a
NULL pointer and not do anything. This is why this does not crash at
the moment. However, this does not look very safe (it would crash
for any location of base that is not the first in brw_shader), so
patch it to prevent a potential (even if unlikely) problem in the
future.
Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
The initial glGetUniformdv support didn't cover all the
casting cases that are apparantly legal, and cts seems to
test for them.
I've updated the piglit test to cover these cases now.
v2: fix indentation - it's all broken in this file (Ilia)
fix src/dst index tracking in light of fp64 support (Ilia)
cc: "11.0" <mesa-stable@lists.freedesktop.org>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
Signed-off-by: Dave Airlie <airlied@redhat.com>
Curro added this in commit 3ee2daf23d (before the vec4/NIR backend was
added) but it was missed in the new NIR backend. Add it there as well.
instructions in affected programs: 1857 -> 1810 (-2.53%)
helped: 15
Reviewed-by: Francisco Jerez <currojerez@riseup.net>
We still have to push everything out, might as well kick earlier and
flip pushbufs when we know we'll need it. This resolves some issues with
the new policy of making sure that we always leave a bit of room at the
end for fences.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Fixes: 47d11990b (nouveau: make sure there's always room to emit a fence)
Cc: mesa-stable@lists.freedesktop.org
Right now we emit on every kick, but this is only necessary if something
will ever be able to observe that the fence completed. If there are no
refs, leave the fence alone and emit it another day.
This also happens to work around an issue for the kick handler -- a kick
can be a result of e.g. nouveau_bo_wait or explicit kick, or it can be
due to lack of space in the pushbuf. We want the emit to happen in the
current batch, so we want there to always be enough space. However an
explicit kick could take the reserved space for the implicitly-triggered
kick's fence emission if it happened right after. With the new mechanism,
hopefully there's no way to cause two fences to be emitted into the same
reserved space.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Fixes: 47d11990b (nouveau: make sure there's always room to emit a fence)
Cc: mesa-stable@lists.freedesktop.org
In theory, GF110+ should also support NVC8_COMPUTE_CLASS but, in practice,
a ILLEGAL_CLASS dmesg fail appears when using it.
This fixes compute support and MP performance counters on GF110.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
For scalar VS, I'll need this in brw_fs.cpp as well. It seems silly to
redeclare it in three places.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Previously, we used nir_lower_io with the scalar type_size function,
which mapped VERT_ATTRIB_* locations to...some numbers. Then, in
fs_visitor::nir_setup_inputs(), we created temporaries indexed by
those numbers, and emitted MOVs from the actual ATTR registers to
those temporaries. Virtually all of these were copy propagated away,
but it's still ugly.
This patch reworks our input lowering to produce NIR lower_input
intrinsics that properly index into the ATTR file, so we can access
it directly.
No changes in shader-db.
v2: Fix unreachable() message (Ken), update commit message (Matt).
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
nr_attributes is used to compute first_non_payload_grf, which is the
first register we're allowed to use for ordinary register allocation.
The hardware requires us to read at least one pair of values, but we're
completely free to overwrite that garbage register with whatever we like.
Instead of altering nr_attributes, we should alter urb_read_length, which
only affects the amount we ask the VF to read. This should save us a
register in trivial cases (which admittedly isn't very useful).
While we're at it, improve the explanation in the comments.
v2: Actually do what I said (caught by Ilia).
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Both the vec4 and scalar VS backends had virtually identical URB entry
size and read length calculations. We can move those up a level to
backend-agnostic code and reuse it for both.
Unfortunately, the backends need to know nr_attributes to compute
first_non_payload_grf, so I had to store that in prog_data. We could
use urb_read_length, but that's nr_attributes rounded up to a multiple
of two, so doing so would waste a register in some cases.
There's more code to be removed in the vec4 backend, but that will
come in a follow-on patch.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Switch statements introduce a bogus loop with an unconditional break at
the end of the loop, just before the while...so the while is unreachable
and has no immediate dominator.
v2: With less exuberance
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
The (gen < 9) check in brw_clear() was too broad. It disabled all types
of fast color clears:
a. singlesample rep clears
b. singlesample MCS fast clears
c. multisample MCS fast clears
The MCS clears are still buggy, but the rep clear works well. So let's
enable it.
Reviewed-by: Neil Roberts <neil@linux.intel.com>
Fast color clears are disabled for gen9 (see the checks in
brw_meta_fast_clear), so there is no reason to allocate the MCS and
track its clear/resolve state.
Reviewed-by: Neil Roberts <neil@linux.intel.com>
The translate functions is split into two:
- translation to TGSI
- creating the variant (TGSI transformations only)
Reviewed-by: Dave Airlie <airlied@redhat.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Tested-by: Brian Paul <brianp@vmware.com>
The samplers for DrawPixels data and the pixel map are assigned to slots
which don't overlap with the existing sampler slots.
The texture coordinates for the user texture are uploaded as a constant.
Reviewed-by: Dave Airlie <airlied@redhat.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Tested-by: Brian Paul <brianp@vmware.com>
- there is no connection to user fragment shaders, so having these as
shader variants makes no sense
- don't use Mesa IR, use TGSI
- don't create gl_fragment_program, just create the shader CSO
v2: generate exactly the same shader as before to fix llvmpipe
Reviewed-by: Dave Airlie <airlied@redhat.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Tested-by: Brian Paul <brianp@vmware.com>
No other shader stage has a "prepare" function.
This will allow removing some variables from st_vertex_program.
Also, prepare_fragment_program was a dead prototype.
Reviewed-by: Dave Airlie <airlied@redhat.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Tested-by: Brian Paul <brianp@vmware.com>
Drivers weren't notified about this at all.
This allows disabling on-demand compilation in drivers.
Reviewed-by: Dave Airlie <airlied@redhat.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Tested-by: Brian Paul <brianp@vmware.com>
First step towards inverting the dependency between glsl and nir (so nir
can be used without glsl). Also solves this issue with 'make distclean'
Making distclean in mesa
make[2]: Entering directory '/mnt/sdb1/Src64/Mesa-git/mesa/src/mesa'
Makefile:2486: ../glsl/.deps/shader_enums.Plo: No such file or directory
make[2]: *** No rule to make target '../glsl/.deps/shader_enums.Plo'. Stop.
make[2]: Leaving directory '/mnt/sdb1/Src64/Mesa-git/mesa/src/mesa'
Makefile:684: recipe for target 'distclean-recursive' failed
make[1]: *** [distclean-recursive] Error 1
make[1]: Leaving directory '/mnt/sdb1/Src64/Mesa-git/mesa/src'
Makefile:615: recipe for target 'distclean-recursive' failed
make: *** [distclean-recursive] Error 1
Reported-by: Andy Furniss <adf.lists@gmail.com>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
Signed-off-by: Rob Clark <robclark@freedesktop.org>
The call to _mesa_test_texobj_completeness() is unnecessary if the
texture is already known to be complete. If the texture object is
dirtied in the meantime _BaseComplete and _MipmapComplete will be
reset to false. _mesa_is_image_unit_valid() will start to be called
more frequently in a future commit, so it seems desirable to avoid the
unnecessary work.
Tested-by: Ye Tian <yex.tian@intel.com>
CC: "11.0" <mesa-stable@lists.freedesktop.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
A future commit will remove all texture object-dependent derived state
from the image unit struct to make validation unnecessary on texture
state changes. Instead of checking gl_image_unit::_Valid drivers will
be required to call this function when needed to find out whether an
image unit is in a valid state and whether access from the shader is
allowed.
Tested-by: Ye Tian <yex.tian@intel.com>
CC: "11.0" <mesa-stable@lists.freedesktop.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
The hardware documentation relating to the UAV HW-assisted coherency
mechanism and UAV access enable bits is scarce and sometimes
contradictory, and there's quite some guesswork behind this commit, so
let me summarize the background first: HSW and later hardware have
infrastructure to support a stricter form of data coherency between
shader invocations from separate primitives. The mechanism is
controlled by the "Accesses UAV" bits on 3DSTATE_VS, _HS, _DS, _GS and
_PS (or _PS_EXTRA on BDW+), and the "UAV Coherency Required" bit on
the 3DPRIMITIVE command.
Regardless of whether "UAV Coherency Required" is set, the hardware
fixed-function units will increment a per-stage semaphore for each
request received if "Accesses UAV" is set for the same or any lower
stage. An implicit DC flush is emitted by the lowermost stage with
"Accesses UAV" set once it's done processing the request, this also
happens regardless of the value of "UAV Coherency Required". The
completion of the DC flush will cause the same stage and all previous
ones to decrement the semaphore, marking the UAV accesses for the
primitive as coherent with L3.
The "UAV Coherency Required" 3DPRIMITIVE bit will cause a pipeline
stall before any threads are dispatched for the first FF stage with
"Accesses UAV" set until the semaphore is cleared for the same stage.
Effectively this guarantees that UAV memory accesses performed by
previous primitives from any stage will be strictly ordered (and
thanks to the implicit DC flush visible in memory) with UAV accesses
from the following primitives.
None of this is required by the usual image, atomic counter and SSBO
GL APIs which have very relaxed cross-primitive coherency and ordering
requirements, so we don't actually ever set the "UAV Coherency
Required" bit -- Ordering with respect to shader invocations from
previous stages on the same primitive where there is a data dependency
is of course already guaranteed as the spec requires, regardless of
this mechanism being enabled. We do set the "Accesses UAV" bits
though since my commit ac7664e493 (which
this patch partially reverts), mainly because of comments like the
following from the BDW PRM:
> 3DSTATE_GS
>[...]
> 12 Accesses UAV
> Format: Enable
> This field must be set when GS has a UAV access.
There are similar comments in the documentation for the other
3DSTATE_*S commands. The "must" part is misleading and unjustified
AFAIK. Most of the "Accesses UAV" bits don't seem to have any side
effects other than the implicit DC flushes and the related
book-keeping in anticipation for a subsequent primitive with "UAV
Coherency Required" set, so in most cases they are unnecessary and may
incur a performance penalty. There is an exception though. On Gen8+
the PS_EXTRA UAV access bit influences the calculation of the PS
UAV-only and ThreadDispatchEnable signals which on previous
generations were set explicitly by the driver, so we cannot always
avoid enabling it on the PS stage.
The primary motivation for this change is that in fact the hardware
coherency mechanism is buggy and will cause a rather non-deterministic
hang on Gen8 when VS is the only stage with "Accesses UAV" set and the
processing of a request terminates immediately after the implicit DC
flush is sent for a previous primitive with no additional vertices
being emitted for the second primitive, what will cause the hardware
to skip sending a second DC flush and cause the VS to stall
indefinitely waiting for a response from the DC (BDWGFX HSD 1912017).
This hardware bug can be reproduced on current master with the
spec@arb_shader_image_load_store@host-mem-barrier@Indirect/RaW piglit
subtest (if you have the patience to run it a few dozen times).
The proposed workaround is to insert CS STALLs speculatively between
3DPRIMITIVE commands when "Accesses UAV" is enabled for the VS stage
only. Because this would affect one of the hottest paths in the
driver and likely decrease performance even further due to the
unnecessary serialization, and because we don't actually need the
implicit DC flushes, it seems better to just disable them.
Cc: 11.0 <mesa-stable@lists.freedesktop.org>
This was originally added to nir_instrs_equal() instead of
nir_instr_can_cse() incorrectly, but this was fixed when moving to the
instruction set API (as it had to be, otherwise hashing wouldn't work).
Now, this is dead code since instr_can_rewrite() will only return true
for texture instructions that use an index, so we can turn the check into
an assert. This also means that now nir_instrs_equal(instr, instr) will
always return true unless it assert-fails.
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
Signed-off-by: Connor Abbott <cwabbott0@gmail.com>
This was previously tied to CSE, since it would only work for
instructions where nir_can_cse() (now instr_can_rewrite()) returned
true. Now that CSE uses the instruction set abstraction which only uses
this internally, we can make it local to nir_instr_set.c.
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
Signed-off-by: Connor Abbott <cwabbott0@gmail.com>
This replaces an O(n^2) algorithm with an O(n) one, while allowing us to
import most of the infrastructure required for GVN. The idea is to walk
the dominance tree depth-first, similar when converting to SSA, and
remove the instructions from the set when we're done visiting the
sub-tree of the dominance tree so that the only instructions in the set
are the instructions that dominate the current block.
No piglit regressions. No shader-db changes.
Compilation time for full shader-db:
Difference at 95.0% confidence
-35.826 +/- 2.16018
-6.2852% +/- 0.378975%
(Student's t, pooled s = 3.37504)
v2:
- rebase on start_block removal
- remove useless state struct
- change commit message
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
Signed-off-by: Connor Abbott <cwabbott0@gmail.com>
This will replace direct usage of nir_instrs_equal() in the CSE pass,
which reduces an O(n^2) algorithm with an effectively O(n) one. It'll
also be useful for implementing GVN on top of GCM.
v2:
- Add texture support.
- Add more comments.
- Rename instr_can_hash() to instr_can_rewrite() since it's really more
about whether its uses can be rewritten, and it's implicitly used by
nir_instrs_equal() as well.
- Rename nir_instr_set_add() to nir_instr_set_add_or_rewrite() (Jason).
- Make the HASH() macro less magical (Topi).
- Rewrite the commit message.
v3:
- For sorting phi sources, use a VLA, store pointers to the sources, and
compare the predecessor pointer directly (Jason).
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
Signed-off-by: Connor Abbott <cwabbott0@gmail.com>
Right now nir_instrs_equal() is tied pretty tightly to CSE, but we're
going to introduce the idea of an instruction set and tie it to that
instead. In anticipation of that, move this into its own file where
we'll add the rest of the instruction set implementation later.
v2: Rebase on texture support.
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
Signed-off-by: Connor Abbott <cwabbott0@gmail.com>
If a non-const sample number is given to interpolateAtSample it will
now generate an indirect send message with the sample ID similar to
how non-const sampler array indexing works. Previously non-const
values were ignored and instead it ended up using a constant 0 value.
The generator will try to determine if the sample ID is dynamically
uniform via nir_src_is_dynamically_uniform. If not it will query the
pixel interpolator in a loop, once for each different live sample
number. The next live sample number is found using emit_uniformize. If
multiple live channels have the same sample number then they will be
handled in a single iteration of the loop. The loop is necessary
because the indirect send message doesn't seem to have a way to
specify a different value for each fragment.
This fixes the following two Piglit tests:
arb_gpu_shader5-interpolateAtSample-nonconst
arb_gpu_shader5-interpolateAtSample-dynamically-nonuniform
v2: Handle dynamically non-uniform sample ids.
v3: Remove the BREAK instruction and predicate the WHILE directly.
Make the tokens arrays const. (Matt Turner)
v4: Iterate over the live channels instead of each possible sample
number.
v5: Don't special case immediate values in
brw_pixel_interpolator_query. Make a better wrapper for the
function to set up the PI send instruction. Ensure that the SHL
instructions are scalar. (Francisco Jerez).
Reviewed-by: Francisco Jerez <currojerez@riseup.net>
It is possible to directly predicate the WHILE instruction. In this
case there will be a second successor block because the execution can
resume from the instruction after the loop. This will be used in a
subsequent patch.
Reviewed-by: Matt Turner <mattst88@gmail.com>
Adds nir_src_is_dynamically_uniform which returns true if the source
is known to be dynamically uniform. This will be used in a later patch
to add a workaround for cases that only work with dynamically uniform
sources. Note that the function is not definitive, it can return false
negatives (but not false positives). Currently it only detects
constants and uniform accesses. It could easily be extended to include
more cases.
Reviewed-by: Matt Turner <mattst88@gmail.com>
This will allow to split SW and HW queries in an upcoming patch.
While we are at it, make use of nvc0_query struct instead of pipe_query.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
From ARB_program_query_interface:
For the property ARRAY_SIZE, a single integer identifying the number of
active array elements of an active variable is written to <params>. The
array size returned is in units of the type associated with the property
TYPE. For active variables not corresponding to an array of basic types,
the value one is written to <params>. If the variable is a shader
storage block member in an array with no declared size, the value zero
is written to <params>.
v2:
- Unsized arrays of arrays have an array size different than zero
v3:
- Arrays and unsized arrays will have an array_stride > 0. Use it
instead of is_unsized_array flag (Timothy).
Signed-off-by: Samuel Iglesias Gonsalvez <siglesias@igalia.com>
Reviewed-by: Timothy Arceri <t_arceri@yahoo.com.au>
From ARB_shader_storage_buffer_object:
"When using the ARB_program_interface_query extension to enumerate the
set of active buffer variables, only the first element of arrays (sized
or unsized) will be enumerated"
_mesa_program_resource_array_size() is used when getting the name (and
name length) of the active variables. When it is an unsized array,
we want to indicate it has one active element so the returned name
would have "[0]" at the end.
v2:
- Use array_stride > 0 and array_elements == 0 to detect unsized
arrays. Because of that, we don't need is_unsized_array flag
(Timothy)
Signed-off-by: Samuel Iglesias Gonsalvez <siglesias@igalia.com>
Reviewed-by: Timothy Arceri <t_arceri@yahoo.com.au>
When the active variable is an array which is already a top-level
shader storage block member, don't return its array size and stride
when querying TOP_LEVEL_ARRAY_SIZE and TOP_LEVEL_ARRAY_STRIDE
respectively.
Fixes the following 12 dEQP-GLES31 tests:
dEQP-GLES31.functional.ssbo.layout.single_basic_array.shared.mat3x4
dEQP-GLES31.functional.ssbo.layout.single_basic_array.shared.row_major_mat3x4
dEQP-GLES31.functional.ssbo.layout.single_basic_array.shared.column_major_mat3x4
dEQP-GLES31.functional.ssbo.layout.single_basic_array.packed.mat3x4
dEQP-GLES31.functional.ssbo.layout.single_basic_array.packed.row_major_mat3x4
dEQP-GLES31.functional.ssbo.layout.single_basic_array.packed.column_major_mat3x4
dEQP-GLES31.functional.ssbo.layout.single_basic_array.std140.mat3x4
dEQP-GLES31.functional.ssbo.layout.single_basic_array.std140.row_major_mat3x4
dEQP-GLES31.functional.ssbo.layout.single_basic_array.std140.column_major_mat3x4
dEQP-GLES31.functional.ssbo.layout.single_basic_array.std430.mat3x4
dEQP-GLES31.functional.ssbo.layout.single_basic_array.std430.row_major_mat3x4
dEQP-GLES31.functional.ssbo.layout.single_basic_array.std430.column_major_mat3x4
v2:
- Fix check when the shader storage block is instanced
- Write auxiliary function to do the check.
v3:
- Check if full_instanced_name is NULL just after allocation (Ilia)
- Remove () from one strcmp() in the if statement (Ilia)
Signed-off-by: Samuel Iglesias Gonsalvez <siglesias@igalia.com>
Tested-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
Patch adds missing type (used with NV_read_depth) so that it gets
handled correctly. This fixes errors seen with following CTS test:
ES3-CTS.gtf.GL3Tests.packed_pixels.packed_pixels
Signed-off-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Cc: "11.0" <mesa-stable@lists.freedesktop.org>
Before, we were unconditionally assigning the TargetIndex field in
_mesa_BindTexture(), even if it was already set properly. Now we
initialize TargetIndex wherever we initialize the Target field, in
_mesa_initialize_texture_object(), finish_texture_init(), etc.
v2: also update the meta_copy_image code. In make_view() the
view_tex_obj->Target field was set, but not the TargetIndex field.
Also, remove a second, redundant assignment to view_tex_obj->Target.
Add sanity check assertions too.
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
Tested-by: Mark Janes <mark.a.janes@intel.com>
Callers of create_texture() will either pass target=0 or a validated
GL texture target enum so no need to do another error check inside
the loop.
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
Tested-by: Mark Janes <mark.a.janes@intel.com>
We can now link the unit tests against just libi965_compiler.la. This
lets us drop a lot of DRI driver dependencies, but we still pull in all
of libmesa and more.
This also provides a few standalone users of libi965_compiler.la, which
will help us accidentally using i965_dri.so functions from the compiler.
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
Signed-off-by: Kristian Høgsberg Kristensen <krh@bitplanet.net>
This introduces a new libtool helper library, libi965_compiler.la. This
library is moderately self-contained, but still needs to link to all of
libmesa.la among other things.
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
Signed-off-by: Kristian Høgsberg Kristensen <krh@bitplanet.net>
brw_get_shader_time_index() is all tangled up in brw_context state and
we can't call it from the compiler. Thanks the Jasons recent
refactoring, we can just get the index and pass to the emit functions
instead.
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
Signed-off-by: Kristian Høgsberg Kristensen <krh@bitplanet.net>
We call this from the compiler so move it to brw_shader.cpp.
Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
Signed-off-by: Kristian Høgsberg Kristensen <krh@bitplanet.net>
This function computes the next power of two, but at least 1024. We can
do that by bitwise or'ing in 1023 and calling util_next_power_of_two().
We use brw_get_scratch_size() from the compiler so we need it out of
brw_program.c. We could move it to brw_shader.cpp, but let's make it a
small inline function instead.
Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
Signed-off-by: Kristian Høgsberg Kristensen <krh@bitplanet.net>
brw_program.c won't be part of the compiler library, but we need
brw_mark_surface_used() in the compiler. Move to brw_shader.cpp.
Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
Signed-off-by: Kristian Høgsberg Kristensen <krh@bitplanet.net>
The initial motivation for this patch was to avoid calling
brw_cs_prog_local_id_payload_dwords() in gen7_cs_state.c from the
compiler. This commit ends up refactoring things a bit more so as to
split out the logic to build the local id payload to brw_fs.cpp. This
moves the payload building closer to the compiler code that uses the
payload layout and makes it available to other users of the compiler.
Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
Signed-off-by: Kristian Høgsberg Kristensen <krh@bitplanet.net>
We want to use the rest of brw_shader.cpp with the rest of the compiler
without pulling in the GLSL linking code.
Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
Signed-off-by: Kristian Høgsberg Kristensen <krh@bitplanet.net>
We need the debug flag parsing and INTEL_DEBUG in the compiler, but we
don't want the dependency on bufmgr (libdrm_intel) in there. Move to
intel_screen.c.
There are now only two lines left in brw_process_intel_debug_variable(),
but we keep it in intel_debug.h to avoid having to expose
'debug_control' as a global variable.
Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Signed-off-by: Kristian Høgsberg Kristensen <krh@bitplanet.net>
We want to use intel_debug.c in code that doesn't link to dri common.
v2: Remove unnecessary stddef.h include (Topi), use util/debug.h
in all DRI driver and remove driParseDebugString() (Iago).
Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
Signed-off-by: Kristian Høgsberg Kristensen <krh@bitplanet.net>
We move these calls one level up into the codegen functions.
Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
Signed-off-by: Kristian Høgsberg Kristensen <krh@bitplanet.net>
Previously the name of the nir shader was being freed prematurely during
nir_sweep. Since 756613ed35 the name was later being used to generate
filenames for the optimiser debug output and these would end up with
garbage from the dangling pointer.
Co-authored-by: Neil Roberts <neil@linux.intel.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Comit d48ac93066 addressed this for VS, but we forgot to do the same for
URB writes generated by the gen6 GS.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Varyings can be considered inputs or outputs of a program only when
SSO is in use. With multi-stage programs, inputs contain only inputs
for first stage and outputs contains outputs of the final shader stage.
I've tested that fix works for Assault Android Cactus (demo version)
and does not cause Piglit or CTS regressions in glGetProgramiv tests.
Following ES 3.1 CTS separate shader tests that do query properties
of varyings in SSO shader programs pass:
ES31-CTS.program_interface_query.separate-programs-vertex
ES31-CTS.program_interface_query.separate-programs-fragment
Signed-off-by: Tapani Pälli <tapani.palli@intel.com>
Tested-by: Dieter Nützel <Dieter@nuetzel-hh.de>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=92122
The EXT_texture_format_BGRA8888 extension (which mesa supports
unconditionally) adds a new format and internal format called GL_BGRA_EXT.
Previously, this was not really handled at all in
_mesa_ex3_error_check_format_and_type. When the checks were tightened in
commit f15a7f3c, we accidentally tightened things too far and GL_BGRA_EXT
would always cause an error to be thrown.
There were two primary issues here. First, is that
_mesa_es3_effective_internal_format_for_format_and_type didn't handle the
GL_BGRA_EXT format. Second is that it blindly uses _mesa_base_tex_format
which returns GL_RGBA for GL_BGRA_EXT. This commit fixes both of these
issues as well as adds explicit checks that GL_BGRA_EXT is only ever used
with GL_BGRA_EXT and GL_UNSIGNED_BYTE.
Signed-off-by: Jason Ekstrand <jason.ekstrand@intel.com>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=92265
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Cc: "11.0" <mesa-stable@lists.freedesktop.org>
Fixes a regression that broke EGL since
commit 858f2f2ae6
Author: Emil Velikov <emil.l.velikov@gmail.com>
Date: Sun Sep 13 12:25:27 2015 +0100
egl/dri2: ease srgb __DRIconfig conditionals
Version 3: Simplify the code comment, word wrap commit description.
Version 2: Return GL_FALSE if ARB_shadow is unsupported instead of
pretending to store the value as suggested by Brian Paul.
This fixes a GL error warning on r200 in Wine.
The GL_ARB_sampler_objects extension does not specify a dependency on
GL_ARB_shadow or GL_ARB_depth_texture for setting the depth texture
compare mode and function. Silently ignore attempts to change these
settings. They won't matter without a depth texture being assigned
anyway.
Reviewed-by: Brian Paul <brianp@vmware.com>
The svga3d device requires constant buffers to be a multiple of 16 bytes
in size. OpenGL UBOs may not fit that restriction. As a work-around,
round the size up if possible, else round down.
Note that this patch only effects UBO constant buffers (index 1 or higher),
not the 0th/default constant buffer.
Fixes the game Grim Fandango Remastered. VMware bug 1510130.
Reviewed-by: Charmaine Lee <charmainel@vmware.com>
Reviewed-by: José Fonseca <jfonseca@vmware.com>
One can simplify the if-else chain, by declaring the driconfigs as a
two sized array, whist using srgb as a index to the correct entry.
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Acked-by: Alex Deucher <alexander.deucher@amd.com>
Move all the enums but CONTEXT_FLAGS. The spec seems quite explicit
about the latter (wrt OpenGL ES)
"In OpenGL ES versions prior to and including ES 3.1 there is no
CONTEXT_FLAGS state and therefore the CONTEXT_FLAG_DEBUG_BIT cannot
be queried."
v2 [Emil Velikov] Rebase.
v3 [Emil Veliokv] Drop the CONTEXT_FLAGS hunk - not applicable for GLES
Signed-off-by: Matthew Waters <ystreet00@gmail.com>
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
v2 [Emil Velikov]
- Rebase.
- Correct version in gles11 dispatch_sanity.
- Move the extension enable to a separate patch.
Signed-off-by: Matthew Waters <ystreet00@gmail.com>
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
I started seeing a lot of situations on nv30 where fence emission
wouldn't fit into the previous buffer (causing assertions). This ensures
that whenever checking for space, we always leave a bit of extra room
for the fence emission commands. Adjusts the nv30 and nvc0 fence
emission logic to bypass the space checking as well.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Cc: mesa-stable@lists.freedesktop.org
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Similar to 9ffc1049ca (freedreno/ir3: use nir two-sided-color lowering).
No piglit regression.
Signed-off-by: Boyan Ding <boyan.j.ding@gmail.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
Bring the following commit over to i915:
commit ec542d7457
Author: Eric Anholt <eric@anholt.net>
Date: Mon Mar 3 10:43:10 2014 -0800
i965: Drop broken front_buffer_reading/drawing optimization.
Not sure if it might fix anything, but since the i965 and i915 used to
share a bunch of that code, it would seem reasonable the same problems
could be present in the i915 code still, and the i965 approach is well
tested by now so bringing it over seems fairly safe.
No piglit regressions on 855.
v2: Rebase on _mesa_is_front_buffer_* refactor.
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
There are multiple similar implementations of these functions, and a
later patch was going to add another.
v2: Move removing intel_framebuffer to a different patch.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
v2: Since state_tracker does not call _mesa_init_driver_functions, we
need to initialize the dd::NewFramebuffer pointer to
_mesa_new_framebuffer here. Suggested by Brian.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Flip the cull bits when rendering to a user fbo on gen2. This
was already done on gen3 (since before git history starts)
but was missing from the gen2 code.
Fixes rendering of the driver+kart model in supertuxkart kart
selection screen.
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
The hardware can draw lines 0.5 to 7.5 pixels wide. Adjust the limits
to 1.0-7.0. The old limits seems to be from the era when i915 and i965
were sharing this code.
Not really sure if 1.0-7.0 is correct. Maybe it could be 0.5.7.5 as
those are the hw limits, or maybe some combination of the two?
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
The sub-pixel adjustment for points was killed off in
commit 60d762aa62
Author: Xiang, Haihao <haihao.xiang@intel.com>
Date: Wed Jan 2 11:38:51 2008 +0800
i915: Needn't adjust pixel centers. fix#12944
so if we don't need it in intel_tris.c we don't need it in
intel_render.c either, which means we can allow intel_render.c to render
points.
No apparent regressions on PNV in ES1 or ES2 conformance.
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
intelFastRenderClippedPoly() renders the polygon using triangles. For
polygons the provoking vertex is always the first one, and currently
this function assumes that the provoking vertex for triangles is the
last one. In case the user changed the provoking vertex convention,
the hardware may be configured to treat the first vertex of triangles
as the provoking vertex. So check the convention and emit the triangles
in the appropriate order to avoid having to change the hardware
provoking vertex convention for rendering polygons.
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
When drawing quads using triangles we need to be careful to make
the provoking vertices match when flat shading.
v2: Major rebase on top of Ian's other t_dd_dmatmp.h work.
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
When rendering quad strips via tri strips we can't get the provoking
vertex right, so disallow flat shading.
v2: Major rebase on top of Ian's other t_dd_dmatmp.h work.
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
We can allow rendering flat shaded polygons using tri fans if we check
the provoking vertex convention.
v2 (idr): Remove _EXT suffixes from GL_FIRST_VERTEX_CONVENTION.
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
From http://lists.freedesktop.org/archives/mesa-dev/2015-May/084883.html:
"There are no real error cases here, just dead code.
validate_render() is supposed to make sure we never call these
functions if the code can't actually render the primitives. The
fprintf()+return branches should really just contain assert(0) or
equivalent."
I also rearranged the if-else-block in render_quad_strip_verts to look
more like the other functions. A future patch is going to change a
bunch of that code anyway.
v2: Make "unreachable" message more descriptive. Suggested by Iago.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Suggested-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
With NIR, it actually hurts things.
total instructions in shared programs: 6529329 -> 6528888 (-0.01%)
instructions in affected programs: 14833 -> 14392 (-2.97%)
helped: 299
HURT: 1
In all affected programs I inspected (including the single hurt one) the
pass CSE'd some multiplies and caused some reassociation (e.g., caused
(A * B) * C to be A * (B * C)) when the original intermediate result was
reused elsewhere.
Acked-by: Kenneth Graunke <kenneth@whitecape.org>
We're not using any fs_inst fields, and the next commit will make the
peephole used by the vec4 backend.
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
We never emit IF instructions with an embedded comparison (lost in the
switch to NIR), so this code is not used. If we want to readd support,
we should have a pass that merges a CMP instruction with an IF or a
WHILE instruction after other optimizations have run.
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
According to the Intel Software Development Manual (Volume 1: Basic
Architecture, 12.10.3 Streaming Load Hint Instruction):
Streaming loads may be weakly ordered and may appear to software to
execute out of order with respect to other memory operations.
Software must explicitly use fences (e.g. MFENCE) if it needs to
preserve order among streaming loads or between streaming loads and
other memory operations.
That is, a memory fence is needed to preserve the order between the GPU
writing the buffer and the streaming loads reading it back.
Reported-by: Joseph Nuzman <joseph.nuzman@intel.com>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
There are three types of fast clears:
a. fast depth clears
b. fast singlesample color clears
c. fast multisample color clears
Function intel_miptree_is_fast_clear_capable() checks if a miptree
supports fast clears of type (b).
Rename the function to disambiguate what it does:
old: intel_miptree_is_fast_clear_capable
new: intel_miptree_supports_non_msrt_fast_clear
The functionally accidentally rejected multisampled color surfaces
because it thought they were singlesample array surfaces. Fix that by
explicitly rejecting surfaces with samples > 1.
This fix would have been needed before we enabled layered fast
singlesample color clears (introduced in gen8), which we want to do
eventually. For now, though, this patch changes no behavior; it just
fixes how the driver chooses its behavior.
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
intel_tiling_supports_non_msrt_mcs() and
intel_miptree_is_fast_clear_capable() are not used outside of
intel_mipmap_tree.c.
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
We need a virtual destructor when at least one of the class' methods is virtual.
Failure to do so might lead to undefined behavior when destructing derived classes.
Fixes the following warning:
brw_vec4_gs_visitor.cpp: In function 'const unsigned int* brw::brw_gs_emit(brw_context*, gl_shader_program*, brw_gs_compile*, void*, unsigned int*)':
brw_vec4_gs_visitor.cpp:703:11: warning: deleting object of polymorphic class type 'brw::vec4_gs_visitor' which has non-virtual destructor might cause undefined behaviour [-Wdelete-non-virtual-dtor]
delete gs;
Curro: This shouldn't be causing any actual bugs at the moment because
gen6_gs_visitor is the only subclass of vec4_visitor destroyed through
a pointer of a base class (vec4_gs_visitor *) and its destructor is
basically the same as its parent's. Anyway it seems sensible to change
this so it doesn't bite us in the future.
Reviewed-by: Francisco Jerez <currojerez@riseup.net>
Fixes following Piglit test:
global-scope-binding-qualifier.frag
Signed-off-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
In theory we can't break this assertion since the compiler frontend checks
that we don't exceed any of the individual limits, but it does not hurt to
be extra safe.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
These share the space with UBO surfaces but we need to make sure we
allocate enough space for both sets (12 of each)
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
The thing you want to do with the output files is diff them, which is
made more difficult by line numbers changing.
Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
It seems like things are either coming in slighly wrong, or perhaps
uploaded incorrectly, but either way passing them through the translate
module seems to fix everything. Eventually we should figure out what's
going wrong and fix it "for real", but this should do for now.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Cc: mesa-stable@lists.freedesktop.org
This puts us in line with what the DDX/DRI2 st are expecting. It also
happens to work... no idea why, but seems better to have it work than to
ask lots of questions.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Cc: mesa-stable@lists.freedesktop.org
The uniform will only be of a single type so store the data for
opaque types in a single array.
Cc: Francisco Jerez <currojerez@riseup.net>
Cc: Ilia Mirkin <imirkin@alum.mit.edu>
Geometry and tessellation shaders process multiple vertices; their
inputs are arrays indexed by the vertex number. While GLSL makes
this look like a normal array, it can be very different behind the
scenes.
On Intel hardware, all inputs for a particular vertex are stored
together - as if they were grouped into a single struct. This means
that consecutive elements of these top-level arrays are not contiguous.
In fact, they may sometimes be in completely disjoint memory segments.
NIR's existing load_input intrinsics are awkward for this case, as they
distill everything down to a single offset. We'd much rather keep the
vertex ID separate, but build up an offset as normal beyond that.
This patch introduces new nir_intrinsic_load_per_vertex_input
intrinsics to handle this case. They work like ordinary load_input
intrinsics, but have an extra source (src[0]) which represents the
outermost array index.
v2: Rebase on earlier refactors.
v3: Use ssa defs instead of nir_srcs, rebase on earlier refactors.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
get_io_offset() already walks the dereference chain and discovers
whether or not we have an indirect; we can just return that rather than
computing it a second time via deref_has_indirect(). This means moving
the call a bit earlier.
By returning a nir_ssa_def *, we can pass back both an existence flag
(via NULL checking the pointer) and the value in one parameter. It
also simplifies the code somewhat. nir_lower_samplers works in a
similar fashion.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
This is only a half of the work. The next patch will handle
gl_SampleID/SamplePos, which is the other half of ARB_sample_shading.
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
Required by ARB_sample_shading for drivers that don't want a shader variant
in st/mesa.
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
Acked-by: Roland Scheidegger <sroland@vmware.com>
We will only do depth-only or stencil-only decompress blits, whichever is
needed by textures, instead of always doing both.
Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>
Now that everything comes in through NIR, we can pick this directly out of
the shader source and don't need to reference the gl_fragment_program.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
This is a common enough operation that it's nice to not have to think about
the arguments to foreach_list_typed every time.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Drivers and state trackers that use LLVM for generating code, must
register the targets they use with LLVM's global TargetRegistry.
The TargetRegistry is not thread-safe, so all targets must be added
to the registry before it can be queried for target information.
When drivers and state trackers initialize their own targets, they need
a way to force gallivm to initialize its targets at the same time.
Otherwise, there can be a race condition in some multi-threaded
applications (e.g. glx-multihreaded-shader-compile in piglit),
when one thread creates a context for a driver that uses LLVM (e.g.
radeonsi) and another thread creates a gallivm context (glxContextCreate
does this).
The race happens when the driver thread initializes its LLVM targets and
then starts using the registry before the gallivm thread has a chance to
register its targets.
This patch allows users to force gallivm to register its targets by
calling the gallivm_init_llvm_targets() function.
v2:
- Use call_once and remove mutexes and static initializations.
- Replace gallivm_init_llvm_{begin,end}() with
gallivm_init_llvm_targets().
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Mathias Fröhlich <Mathias.Froehlich@web.de>
Reviewed-by: Emil Velikov <emil.l.velikov@gmail.com>
CC: "10.6 11.0" <mesa-stable@lists.freedesktop.org>
Unfortunately, we can't get rid of them entirely. The FS backend still
needs gl_program for handling TEXTURE_RECTANGLE. The GS vec4 backend still
needs gl_shader_program for handling transfom feedback. However, the VS
needs neither and we can substantially reduce the amount they are used.
One day we will be free from their tyranny.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
It doesn't exist for anything other than an assert that, as far as I can
tell, isn't possible to trip. Soon, we will remove prog from the visitor
entirely and this will become even more impossible to hit.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
The texunit variable we create and assign in nir_emit_texture gets passed
through two more layers of function calls before it gets to its sole use in
rescale_texcoord. The best part is that we already pass the sampler into
rescale_texcoord so we can just look it up there.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
As of now, uniform setup is more-or-less unified between vec4 and fs and no
longer requires the fs_visitor. This makes uniform setup more of a
language/API thing than a backend compiler thing. This commit moves
setting up the stage_prog_data.params arrays to the same place as we set up
the rest of stage_prog_data.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Setting up binding tables really has little to do with the actual process
of turning shaders into instructions; it's more part of setting up
prog_data. This commit moves it out of the visitors and with the rest of
the prog_data setup stuff.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
This really has nothing to do with the backend compiler and we'd like to
eventually be able to set this up earlier in the compile process.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
The way we deal with GLSL uniforms and builtins is basically the same in
both the vec4 and the fs backend. This commit takes the best parts of both
implementations and pulls the common code into a shared helper function.
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
The way we deal with ARB program uniforms is basically the same in both the
vec4 and the fs backend. This commit takes the best parts of both
implementations and pulls the common code into a shared helper function.
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Previously, we were counting up uniforms as we set them up. However, this
count should be exactly identical to shader->num_uniforms provided by
nir_assign_var_locations. (If it's not, we're in trouble anyway because
that means that locations don't match up.) This matches what the fs
backend is already doing.
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
I tried to do this once before but Curro pointed out that having it in
backend_shader meant it could use the setup_vec4_uniform_values helper
which did different things in vec4 and fs. Now the setup_uniform_values
function differs only by an assert in the two backends so there's no real
good reason to be using it anymore.
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
The uniform_vector_size array was only ever used by pack_uniform_registers
which no longer needs it.
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Previously, pack_uniform_registers worked based on the size of the uniform
as given to us when we initially set up the uniforms. However, we have to
walk through the uniforms and figure out liveness anyway, so we migh as
well record the number of channels used as we go. This may also allow us
to pack things tighter in a few cases.
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
That way, if we do the usual thing of multiplying vector_elements by
matrix_columns we get the actual number of components in the type as per
component_slots().
While we're at it, we also switch to using the actual C++ field
initializers for vector_elements and matrix_columns.
Reviewed-by: Dave Airlie <airlied@redhat.com>
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Previously, we had a bunch of code in each stage to figure out how many
slots we needed in stage_prog_data.param. This code was mostly identical
across the stages and had been copied and pasted around. Unfortunately,
this meant that any time you did something special, you had to add code for
it to each of these places. In particular, none of the stages took
subroutines into account; they were working entirely by accident. By
taking this data from the NIR shader, we know the exact number of entries
we need and everything goes a bit smoother.
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
The next commit will add code to codegen_vs_prog that requires the NIR
shader to be there in all cases. It doesn't hurt anything to just move it
from brw_vs_emit to its only caller.
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
GLSL IR vs. NIR shader-db results for vec4 programs on i965:
total instructions in shared programs: 1499328 -> 1388354 (-7.40%)
instructions in affected programs: 1245199 -> 1134225 (-8.91%)
helped: 7469
HURT: 2440
GLSL IR vs. NIR shader-db results for vec4 programs on G4x:
total instructions in shared programs: 1436799 -> 1325825 (-7.72%)
instructions in affected programs: 1205599 -> 1094625 (-9.20%)
helped: 7469
HURT: 2440
GLSL IR vs. NIR shader-db results for vec4 programs on Iron Lake:
total instructions in shared programs: 1436654 -> 1325682 (-7.72%)
instructions in affected programs: 1205503 -> 1094531 (-9.21%)
helped: 7468
HURT: 2440
GLSL IR vs. NIR shader-db results for vec4 programs on Sandy Bridge:
total instructions in shared programs: 2016249 -> 1787033 (-11.37%)
instructions in affected programs: 1850547 -> 1621331 (-12.39%)
helped: 14856
HURT: 1481
GLSL IR vs. NIR shader-db results for vec4 programs on Ivy Bridge:
total instructions in shared programs: 1848027 -> 1648216 (-10.81%)
instructions in affected programs: 1660279 -> 1460468 (-12.03%)
helped: 14668
HURT: 1369
GLSL IR vs. NIR shader-db results for vec4 programs on Bay Trail:
total instructions in shared programs: 1848027 -> 1648216 (-10.81%)
instructions in affected programs: 1660279 -> 1460468 (-12.03%)
helped: 14668
HURT: 1369
GLSL IR vs. NIR shader-db results for vec4 programs on Haswell:
total instructions in shared programs: 1848027 -> 1648216 (-10.81%)
instructions in affected programs: 1660279 -> 1460468 (-12.03%)
helped: 14668
HURT: 1369
I also ran our full suite of benchmarks on a Haswell and had the following
statistically significant (according to ministat) changes:
Test master-glsl master-nir diff
bench_OglGeomPoint 461.556 463.006 1.450
bench_OglTerrainFlyInst 184.484 187.574 3.090
bench_OglTerrainPanInst 132.412 136.307 3.895
bench_OglTexFilterAniso 19.653 19.645 -0.008
bench_OglTexFilterTri 58.333 58.009 -0.324
bench_OglVSInstancing 65.049 65.327 0.278
bench_trexoff 69.474 69.694 0.220
bench_valley 40.708 41.125 0.417
v2 (Jason Ekstrand):
- Remove more uses of NirOptions as a switch
- New shader-db numbers
- Added benchmark numbers
Reviewed-by: Matt Turner <mattst88@gmail.com>
Commit 0a1adaf11d (nir: Report progress
from nir_lower_system_values().) introduced a bug caught by Valgrind:
==823== Conditional jump or move depends on uninitialised value(s)
==823== at 0xB09020C: convert_block (nir_lower_system_values.c:68)
==823== by 0xB079FB8: foreach_cf_node (nir.c:1310)
==823== by 0xB07A0AF: nir_foreach_block (nir.c:1336)
==823== by 0xB09026B: convert_impl (nir_lower_system_values.c:79)
...
==823== Uninitialised value was created by a stack allocation
==823== at 0xB090249: convert_impl (nir_lower_system_values.c:76)
which is trivially fixed by initializing progress.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Add a macro GL_LIB_NAME to hold the filename that configure comes up with
based on the --with-gl-lib-name and --enable-mangling options.
In driOpenDriver, use the GL_LIB_NAME macro instead of hard-coding
"libGL.so.1".
v2: Add an #ifndef/#define for GL_LIB_NAME so that non-autoconf builds will
work.
v3: Fix the library filename in the Makefile.
Signed-off-by: Kyle Brenneman <kbrenneman@nvidia.com>
Reviewed-by: Emil Velikov <emil.l.velikov@gmail.com>
Cc: "10.6 11.0" <mesa-stable@lists.freedesktop.org>
When USE_MGL_NAMESPACE is defined, _glapi_get_stub will check for the "m"
prefix before trying to skip it, so that "glFoo" and "mglFoo" are
equivalent.
This should let it work with all the places where something calls
_glapi_get_proc_offset with a hard-coded name that starts with the normal
"gl" prefix.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=55552
Signed-off-by: Kyle Brenneman <kbrenneman@nvidia.com>
Reviewed-by: Emil Velikov <emil.l.velikov@gmail.com>
Cc: "10.6 11.0" <mesa-stable@lists.freedesktop.org>
Fixes following Piglit test:
member-invalid-binding-qualifier.frag
Signed-off-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
When writing to a column of a row-major matrix, each component of the
vector is stored to non-consecutive memory addresses, so we generate
one instruction per component.
This patch skips the disabled components in the writemask, saving some
store instructions plus avoid storing wrong data on each disabled
component.
Signed-off-by: Samuel Iglesias Gonsalvez <siglesias@igalia.com>
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Fixes a failing subtest in:
ES31-CTS.shader_storage_buffer_object.negative-glsl-compileTime
Signed-off-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
Reading this output was really confusing. reg represents attribute
slots; reg_offset is the x/y/z/w component (0..3) within a vec4 slot.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
The code for input lowering is going to get significantly more
complicated shortly, so I wanted to pull it out. Vertex shader inputs
are handled nearly identically regardless of vec4/scalar mode, so I
opted to not split that.
I thought about having each function actually do the lowering, but one
pass through nir_lower_io that handles all types (which weren't handled
earlier) is probably more efficient.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
We may want to use different type_size functions for (e.g.) inputs
vs. uniforms. Passing in -1 for mode ignores this, handling all
modes as before.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
If the texture object exists, but the Name field is zero, it means
the object was created but never bound to a target. Trying to bind it
in _mesa_BindTextureUnit() should generate GL_INVALID_OPERATION.
Fixes piglit's arb_direct_state_access-bind-texture-unit test.
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
This helper was only called from _mesa_BindTextureUnit(). It's simpler
to just inline it.
The error check / code / message in the helper was incorrect. It was
written for glBindTextures(), not glBindTextureUnit(). The correct
error for a bad texture unit number is GL_INVALID_VALUE. The error
message now reports the unit number rather than a GL_TEXTUREi enum.
Fixes a failure in piglit's arb_direct_state_access-bind-texture-unit test.
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
Before, we were doing the actual _mesa_reference_texobj() call and
ctx->Driver.BindTexture() and misc housekeeping in three different
places. This consolidates the common code in a new bind_texture()
function.
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
Since we store both in UniformBlocks, we can't just compute the index by
subtracting the array address start, we need to count the number of
buffers of the approriate type.
v2:
- Just fall back to calc_resource_index (Tapani)
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
The old code had some significant problems with respect to
sampler2DArray textures. The biggest problem was that some of the code
would use vec3 for the texture coordinate type, and other parts of the
code would use vec2. The resulting shader would not even compile.
Since there were not tests for this path, nobody noticed.
The input to the fragment shader is always treated as a vec3. If the
source data is only vec2, the vertex puller will supply 0 for the .z
component. The texture coordinate passed to the fragment shader is
always a vec2 that comes from the .xy part of the vertex shader input.
The layer, taken from the .z of the vertex shader input is passed
separately as a flat integer. If the generated fragment shader does not
use the layer integer, the GLSL linker will eliminate all the dead code
in the vertex shader.
Fixes the new piglit tests "blit-scaled samples=2 with
gl_texture_2d_multisample_array", etc. on i965.
Note for stable maintainer: This patch may depend on 46037237, and that
patch should be safe for stable.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
Cc: Topi Pohjolainen <topi.pohjolainen@intel.com>
Cc: Jordan Justen <jordan.l.justen@intel.com>
Cc: "10.6 11.0" <mesa-stable@lists.freedesktop.org>
Add comments that link the driver's miptree structures to the hardware
structures documented in the PRM. This provides sorely needed
orientation to developers new to the miptree code. And for miptree
veterans, this clarifies some of the more obscure miptree data.
For each driver struct field that closely corresponds to a
hardware struct field, add a PRM reference to that hardware field's
name. For example,
struct intel_mipmap_tree {
...
/**
* @brief One of GL_TEXTURE_2D, GL_TEXTURE_2D_ARRAY, etc.
*
* @see RENDER_SURFACE_STATE.SurfaceType
* @see RENDER_SURFACE_STATE.SurfaceArray
* @see 3DSTATE_DEPTH_BUFFER.SurfaceType
*/
GLenum target;
...
};
Also annotate the INTEL_MSAA_LAYOUT_* enums with the name of the PRM
sections that documents the layout.
v2: Replace "2D subimage" with "slice", and define what a "slice" is.
For Ben.
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com> (v1)
Reviewed-by: Ben Widawsky <benjamin.widawsky@intel.com> (v1)
The values of intel_mipmap_tree::align_w and ::align_h correspond to the
hardware enums HALIGN_* and VALIGN_*.
See the confusion?
align_h != HALIGN
align_h == VALIGN
Reduce the confusion by renaming the variables to match the hardware
enum names:
git ls-files |
xargs sed -i -e 's/align_w/halign/g' \
-e 's/align_h/valign/g'
Suggested-by: Kenneth Graunke <kenneth@whitecape.org>
Acked-by: Ben Widawsky <benjamin.widawsky@intel.com>
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
Because that's what it is. It's an untiled, *linear* miptree.
v2:
- Add space after /*.
- Use one comment per function argument.
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
Acked-by: Ben Widawsky <benjamin.widawsky@intel.com>
The comment for intel_miptree_map::mode claimed that it was a bitmask of
GL_MAP_{READ,WRITE,INVALIDATE}_BIT. In reality, the bitmask may include
any of {GL,BRW}_MAP_*_BIT.
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
Acked-by: Ben Widawsky <benjamin.widawsky@intel.com>
Clarify that this bit extends the set of GL_MAP_*_BIT enums.
Also fix typo of "temporary".
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
Acked-by: Ben Widawsky <benjamin.widawsky@intel.com>
Jason open coded this in 60befc63 when cleaning up some ugly code;
using our existing helper tidies it up a bit more.
v2: Drop inline (suggested by Matt).
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
i915 fragment programs utilize the texture coordinate registers
for both texture coordinates and varyings. Unfortunately the
code doesn't check if the same index might be in use for both.
It just naively uses the index to pick a texture unit, which
could lead to collisions.
Add an extra mapping step to allocate non conflicting texture
units for both uses.
The issue can be reproduced with a pair of simple shaders like
these:
attribute vec4 in_mod;
varying vec4 mod;
void main() {
mod = in_mod;
gl_TexCoord[0] = gl_MultiTexCoord0;
gl_Position = gl_ModelViewProjectionMatrix * gl_Vertex;
}
varying vec4 mod;
uniform sampler2D tex;
void main() {
gl_FragColor = texture2D(tex, vec2(gl_TexCoord[0])) * mod;
}
Fixes many piglit tests on i915:
glsl-link-varyings-2
glsl-orangebook-ch06-bump
interpolation-none-gl_frontcolor-smooth-fixed
interpolation-none-gl_frontcolor-smooth-none
interpolation-none-gl_frontcolor-smooth-vertex
interpolation-none-gl_frontsecondarycolor-smooth-fixed
interpolation-none-gl_frontsecondarycolor-smooth-vertex
interpolation-none-gl_frontsecondarycolor-smooth-none
interpolation-none-other-flat-fixed
interpolation-none-other-flat-none
interpolation-none-other-flat-vertex
interpolation-none-other-smooth-fixed
interpolation-none-other-smooth-none
interpolation-none-other-smooth-vertex
v2 [idr]: Minor formatting tweaks.
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Cc: "11.0" <mesa-stable@lists.freedesktop.org>
I830_UPLOAD_RASTER_RULES and I830_UPLOAD_TEX(0) are trying to occupy
the same bit. Move the texture bits upwards a bit to make room for
I830_UPLOAD_RASTER_RULES.
Now the driver will actually upload the raster rules which is rather
important to get the provoking vertex right. Fixes the appearance
of glxgears teeth on gen2.
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Cc: "10.6 11.0" <mesa-stable@lists.freedesktop.org>
Commit 1665d29ee3 introduced an incorrect
format specifier that operates on GLintptr indirect within the function
_mesa_DispatchComputeIndirect().
This patch mitigates the introduced GCC warning:
src/mesa/main/compute.c: In function '_mesa_DispatchComputeIndirect':
src/mesa/main/compute.c:53:7: warning: format '%d' expects argument of type 'int', but argument 3 has type 'GLintptr' [-Wformat=]
_mesa_debug(ctx, "glDispatchComputeIndirect(%d)\n", indirect);
^
v2: Amend for Boyan Ding <boyan.j.ding@gmail.com> feedback.
Signed-off-by: Rhys Kidd <rhyskidd@gmail.com>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
intel_update_winsys_renderbuffer_miptree() will release the existing
miptree when wrapping a new DRI2 buffer, so we can remove the early
release and so prevent a NULL mt dereference should importing the new
DRI2 name fail for any reason. (Reusing the old DRI2 name will result
in the rendering going astray, to a stale buffer, and not shown on the
screen, but it allows us to issue a warning and not crash much later in
innocent code.)
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=86281
Reviewed-by: Martin Peres <martin.peres@linux.intel.com>
Reviewed-by: Chad Versace <chad.versace@intel.com>
From GLSL 1.50 spec, section 4.1.8 "Structures":
"Structures must have at least one member declaration."
So the base_alignment should be higher than zero.
Signed-off-by: Samuel Iglesias Gonsalvez <siglesias@igalia.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
If the string being copied is not NULL-terminated the result of
strlen() is undefined.
Signed-off-by: Samuel Iglesias Gonsalvez <siglesias@igalia.com>
Reviewed-by: Neil Roberts <neil@linux.intel.com>
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
Fix OpenGL ES 3.1 conformance tests: advanced-readWrite-case1-vsfs
and advanced-matrix-vsfs.
v2:
- Fix SHADER_OPCODE_MEMORY_FENCE emission and the allocation of 'tmp'
(Francisco).
Signed-off-by: Samuel Iglesias Gonsalvez <siglesias@igalia.com>
Tested-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Francisco Jerez <currojerez@riseup.net>
From ARB_program_interface_query:
"For an active shader storage block member declared as an array, an
entry will be generated only for the first array element, regardless
of its type. For arrays of aggregate types, the enumeration rules are
applied recursively for the single enumerated array element."
v2:
- Simplify 'if' conditions and return true if it is not a buffer
variable, because these rules only apply to buffer variables (Timothy).
Signed-off-by: Samuel Iglesias Gonsalvez <siglesias@igalia.com>
Reviewed-by: Timothy Arceri <t_arceri@yahoo.com.au>
This matches the function signature created in
lower_ubo_reference_visitor::ssbo_store which has a void return.
Suggested-by: Jason Ekstrand <jason@jlekstrand.net>
Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
At least on Intel hardware, gl_PrimitiveIDIn comes in as a special part
of the payload rather than a normal input. This is typically what we
use system values for. Dave and Ilia also agree that a system value
would be nicer.
At some point, we should change it at the GLSL IR level as well. But
that requires changing most of the drivers. For now, let's at least
make NIR do the right thing, which is easy.
v2: Add a comment about not creating a temporary (suggested by Iago).
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
For 8-bit RGB(A) texture formats we set the PIPE_BIND_RENDER_TARGET flag
to try to get a hardware format which also supports rendering (for FBO
textures). Do the same thing for floating point formats.
This allows the Redway3D Flat demo to run.
Cc: 10.6 11.0 <mesa-stable@lists.freedesktop.org>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
I've temporarily added code like this many times. Wrap it in a
conditional that can be enabled when needed.
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
This code also sets cs_prog_data->uses_num_work_groups which is later
used by state setup to indicate that the gl_NumWorkGroups surface
needs to be setup.
Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
This will only be setup when the prog_data uses_num_work_groups
boolean is set.
At this point nothing will set uses_num_work_groups, but soon code
will set it when emitting code for the intrinsic that loads
gl_NumWorkGroups.
We can't emit this surface information earlier at the start of the
DispatchCompute* call because we may not have generated the program
yet. Until we generate the program, we don't know if the
gl_NumWorkGroups variable is accessed.
We also can't emit the surface as part of the brw_cs_state atom,
because we might not need the surface if gl_NumWorkGroups is not used
by the program.
Lastly, we cannot emit the surface later (after state upload) in the
DispatchCompute* call, because it needs to be run before the
brw_cs_state atom is emitted, since it changes the surface state.
Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
If glDispatchComputeIndirect is used, then the value for this variable
must be read from the indirect BO.
To allow the same generated code to support indirect and
glDispatchCompute, we will also setup a BO for the number of work
groups using the intel_upload_data mechanism. This will only be
required if the gl_NumWorkGroups variable is accessed.
Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
We will need this in an atom to setup a surface to read the
gl_NumWorkGroups values from.
Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
Unlike rendering (BINDING_TABLE_POINTERS_*S), compute doesn't have a
binding table pointers command. Instead it is part of the
MEDIA_INTERFACE_DESCRIPTOR structure loaded by the brw_cs_state atom.
Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
We need to re-emit push constansts when a new batch is started since
the push constants are stored in the batch. We also need to re-emit
the MEDIA_INTERFACE_DESCRIPTOR (in brw_cs_state) since it is stored in
the batch.
Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
Patch also refactors name length queries which were using array size
in computation, this has to be done in same time to avoid regression in
arb_program_interface_query-resource-query Piglit test.
Fixes rest of the failures with
ES31-CTS.program_interface_query.no-locations
v2: make additional check only for GS inputs
v3: create helper function for resource name length
so that it gets calculated only in one place
Signed-off-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Martin Peres <martin.peres@linux.intel.com>
The comment says that it should be impossible for decl_type to be NULL
here, so don't try to handle the case where it is, simply add an assert.
>>> CID 1324977: Null pointer dereferences (FORWARD_NULL)
>>> Comparing "decl_type" to null implies that "decl_type" might be null.
No piglit regressions observed.
Reviewed-by: Timothy Arceri <t_arceri@yahoo.com.au>
Add an assert on the result of as_dereference() not being NULL:
>>> CID 1324978: Null pointer dereferences (NULL_RETURNS)
>>> Dereferencing a null pointer "deref_record->record->as_dereference()".
Since we are introducing a new variable to hold the result of
as_dereference(), take the opportunity to rename deref_record_type to
interface_type and just name the new variable interface_deref, which is
less confusing.
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
We don't use param in this part of the code, so no point in advancing
the pointer forward:
>>> CID 1324983: Code maintainability issues (UNUSED_VALUE)
>>> Assigning value from "param->get_next()" to "param" here, but that stored value is overwritten before it can be used.
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
v2:
- Add strndup.h to Makefile.sources (Emil)
- Use calloc instead of malloc (Emil).
- Check if allocation fails (Emil, Jose)
- Add '#pragma once' and include stdlib.h to strndup.h (Jose)
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=92124
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
Because it counts shader storage blocks too.
v2:
- Use NumBufferInterfaceBlocks instead (Jordan).
Signed-off-by: Samuel Iglesias Gonsalvez <siglesias@igalia.com>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
NumUniformBlocks also counts shader storage blocks.
NumUniformBlocks variable will be renamed in a later patch to avoid
misunderstandings.
v2:
- Modify the condition to use !IsShaderStorage and the list of
uniform blocks (Timothy)
Signed-off-by: Samuel Iglesias Gonsalvez <siglesias@igalia.com>
Reviewed-by: Timothy Arceri <t_arceri@yahoo.com.au>
This condition restricts the use of fast copy blit to cases
where starting pixel of src and dst is oword (16 byte) aligned.
Many piglit tests (if using fast copy blit in Mesa) failed earlier
because I missed adding this condition.Fast copy blit is currently
enabled for use only with Yf/Ys tiling.
Signed-off-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Chad Versace <chad.versace@intel.com>
The bo will often come from a slab in which case it doesn't matter. But
for larger allocations this will be in its own bo, and we have to make
sure to wait until it's no longer used in order for it to be freed.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Cc: mesa-stable@lists.freedesktop.org
Tested-by: Marcin Ślusarz <marcin.slusarz@gmail.com>
If there is an unflushed fence on the bo, then the resource may still be
used in commands built up in the local pushbuf. Flushing can cause all
sorts of unwanted effects, so just free the bo when the relevant fence
is hit.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Cc: mesa-stable@lists.freedesktop.org
Tested-by: Marcin Ślusarz <marcin.slusarz@gmail.com>
This function isn't specific to miptrees. So, drop the "miptree"
from function name.
V3: Add a comment explaining how the 1D Array texture height and
depth is interpreted by Intel hardware.
Signed-off-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Chad Versace <chad.versace@intel.com>
I misinterpreted the alignmnet restriction in XY_FAST_COPY_BLT earlier.
Instead of checking pitch for 64KB alignmnet we need to check it for
tile widh alignment.
Signed-off-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Chad Versace <chad.versace@intel.com>
Current code checks the alignment restrictions only for Y tiling.
From Broadwell PRM vol 10:
"pitch is of 512Byte granularity for Tile-X: This means the tiled-x
surface pitch can be (512, 1024, 1536, 2048...)/4 (in Dwords)."
This patch adds the restriction for X tiling as well.
Signed-off-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Ben Widawsky <ben@bwidawsk.net>
Reviewed-by: Chad Versace <chad.versace@intel.com>
It takes care of using the correct tile width if we later use other
tiling patterns for aux miptree.
V2: Remove the comment about using Yf for aux miptree.
Signed-off-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Chad Versace <chad.versace@intel.com>
This will require change in the parameters passed to
intel_miptree_get_tile_masks().
V2: Rearrange the order of parameters. (Ben)
Change the name to intel_get_tile_masks(). (Topi)
V3: Use temporary variables in intel_get_tile_masks()
for clarity. Fix mask_y computation.
Signed-off-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
Reviewed-by: Chad Versace <chad.versace@intel.com>
V2:
- Do the tile width/height computations in the new helper
function and use it later in intel_miptree_get_tile_masks().
- Change the name to intel_get_tile_dims().
V3: Return the tile_h in number of rows in place of bytes.
Document the units of tile_w, tile_h parameters.
Signed-off-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
Reviewed-by: Chad Versace <chad.versace@intel.com>
When validating format+type+internalFormat for texture pixel operations
on GLES3, the effective internal format should be used if the one
specified is an unsized internal format. Page 127, section "3.8 Texturing"
of the GLES 3.0.4 spec says:
"if internalformat is a base internal format, the effective internal
format is a sized internal format that is derived from the format and
type for internal use by the GL. Table 3.12 specifies the mapping of
format and type to effective internal formats. The effective internal
format is used by the GL for purposes such as texture completeness or
type checks for CopyTex* commands. In these cases, the GL is required
to operate as if the effective internal format was used as the
internalformat when specifying the texture data."
v2: Per the spec, Luminance8Alpha8, Luminance8 and Alpha8 should not be
considered sized internal formats. Return the corresponding unsize format
instead.
v4: * Improved comments in
_mesa_es3_effective_internal_format_for_format_and_type().
* Splitted patch to separate chunk about reordering of
error_check_subtexture_dimensions() error check, which is not directly
related with this patch.
v5: Dropped the splitted patch because it was actually a work around 3
dEQP tests that are buggy:
dEQP-GLES2.functional.negative_api.texture.texsubimage2d_neg_offset
dEQP-GLES2.functional.negative_api.texture.texsubimage2d_offset_allowed
dEQP-GLES2.functional.negative_api.texture.texsubimage2d_neg_wdt_hgt
Cc: "11.0" <mesa-stable@lists.freedesktop.org>
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
Tested-by: Mark Janes <mark.a.janes@intel.com>
This function will be needed as part of validating the combination of format,
type and internal format of texture pixel operations, which happens in
glformats files. Specifically, we want to be able to obtain the base format
of a resolved effective internal format, to compare it with the original
internal format passed.
Also, since this function deals solely with GL formats, it fits better in
glformats where the rest of similar format functionality rests.
The function is moved as-is, without any modification.
Cc: "11.0" <mesa-stable@lists.freedesktop.org>
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
Tested-by: Mark Janes <mark.a.janes@intel.com>
The more specific GLES constrains should be checked after the general
validation performed by _mesa_error_check_format_and_type(). This is also
for consistency with the error checks order of glTexSubImage ops.
v3: The change of order uncovered a bug that regresses a couple of piglit
tests written against OpenGL-ES 1.1 spec, which expects an INVALID_VALUE
instead of the INVALID_ENUM returned by _mesa_error_check_format_and_type()
when an invalid format is passed to glTexImage2D. This version of the patch
accounts for those cases.
Fixes 1 dEQP test:
* dEQP-GLES3.functional.negative_api.texture.teximage2d
Cc: "11.0" <mesa-stable@lists.freedesktop.org>
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
Tested-by: Mark Janes <mark.a.janes@intel.com>
IVB and VLV hang sporadically when an untyped surface read or write
message is used to access a surface of format other than RAW, as may
happen when there is a mismatch between the format qualifier of the
image uniform and the format of the actual image bound to the
pipeline. According to the spec this condition gives undefined
results but may not lead to program termination (which is one of the
possible outcomes of the hang). Fix it by checking at runtime whether
the surface is of the right type.
Fixes the "arb_shader_image_load_store.invalid/format mismatch" piglit
subtest.
Reported-by: Mark Janes <mark.a.janes@intel.com>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=91718
CC: mesa-stable@lists.freedesktop.org
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
This will be used to share the same logic between buffer and image
creation.
v2: Make memory flag set constants local to validate_flags. (Serge
Martin)
This reverts commit 586142658e.
The specs are not explicit about any restrictions related to the types allowed
on buffer variables, however, the description of opaque types (like atomic
counters) is in conclict with the purpose of buffer variables:
"The opaque types declare variables that are effectively opaque
handles to other objects. These objects are
accessed through built-in functions, not through direct reading or
writing of the declared variable.
(...)
Opaque variables cannot be treated as l-values;(...)"
Also, Mesa is already disallowing opaque types in interface blocks anyway, so
that commit was not really achieving anything.
Reviewed-by: Francisco Jerez <currojerez@riseup.net>
With static vertex counts, the final EOT write doesn't actually write
any data - it's just there to end the thread. Typically, the last
thing before ending the thread will be an EmitVertex() call, resulting
in a URB write. We can just set EOT on that.
Note that this isn't always possible - there might be an intervening
SSBO write/image store, or the URB write may have been in a loop.
shader-db statistics for geometry shaders only:
total instructions in shared programs: 3173 -> 3149 (-0.76%)
instructions in affected programs: 176 -> 152 (-13.64%)
helped: 8
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
GS_OPCODE_SET_WRITE_OFFSET is a MUL with a constant src[1] and special
strides. We can easily make the generator handle constant src[0]
arguments by instead generating a MOV with the product of both operands.
This isn't necessarily a win in and of itself - instead of a MUL, we
generate a MOV, which should be basically the same cost. However, we
can probably avoid the earlier MOV to put src[0] into a register.
shader-db statistics for geometry shaders only:
total instructions in shared programs: 3207 -> 3173 (-1.06%)
instructions in affected programs: 3207 -> 3173 (-1.06%)
helped: 11
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Broadwell's 3DSTATE_GS contains new "Static Output" and "Static Vertex
Count" fields, which control a new optimization. Normally, geometry
shaders can output arbitrary numbers of vertices, which means that
resource allocation has to be done on the fly. However, if the number
of vertices is statically known, the hardware can pre-allocate resources
up front, which is more efficient.
Thanks to the new NIR GS intrinsics, this is easy. We just call the
function introduced in the previous commit to get the vertex count.
If it obtains a count, we stop emitting the extra 32-bit "Vertex Count"
field in the VUE, and instead fill out the 3DSTATE_GS fields.
Improves performance of Gl32GSCloth by 5.16347% +/- 0.12611% (n=91)
on my Lenovo X250 laptop (Broadwell GT2) at 1024x768.
shader-db statistics for geometry shaders only:
total instructions in shared programs: 3227 -> 3207 (-0.62%)
instructions in affected programs: 242 -> 222 (-8.26%)
helped: 10
v2: Don't break non-NIR paths (just skip this optimization).
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
The visitor was setting a mlen that was wrong for Broadwell, but the
generator was ignoring it and doing the right thing regardless. We may
as well move the logic fully into the visitor. This will be useful in
the next commit as well.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Some hardware (such as Broadwell) can run geometry shaders more
efficiently when the number of vertices emitted is statically known.
This pass provides a way to obtain the constant vertex count, or
-1 indicating that the vertex count is unknown/non-constant.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
The old code was disasterously complex - spread across multiple atoms
which may not even run, inspecting the dirty bits to try and decide
whether it was necessary to do checks...storing VS information in
brw_context...extra flagging...
This code tripped me and Carl up very badly when working on the
shader cache code. It's very fragile and hard to maintain.
Now that geometry shaders only depend on their inputs and don't have
to worry about the VS VUE map, we can dramatically simplify this:
just compute the VUE map coming out of the geometry shader stage
in brw_upload_programs. If it changes, flag it. Done.
v2: Also check vue_map.separable.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Chris Forbes <chrisf@ijw.co.nz>
Because we only support geometry shaders in core profile, we can safely
ignore any driver-extending of VS outputs.
Those are:
- Legacy userclipping (doesn't exist in core profile)
- Edgeflag copying (Gen4-5 only, no GS support)
- Point coord replacement (Gen4-5 only, no GS support)
- front/back color hacks (Gen4-5 only, no GS support)
v2: Rebase; leave a comment about why SSO works.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Chris Forbes <chrisf@ijw.co.nz>
Previously, our VUE map code always assigned slots to varyings
sequentially, in one contiguous block.
This was a bad fit for separate shaders - the GS input layout depended
or the VS output layout, so if we swapped out vertex shaders, we might
have to recompile the GS on the fly - which rather defeats the point of
using separate shader objects. (Tessellation would suffer from this
as well - we could have to recompile the HS, DS, and GS.)
Instead, this patch makes the VUE map for separate shaders use a fixed
layout, based on the input/output variable's location field. (This is
either specified by layout(location = ...) or assigned by the linker.)
Corresponding inputs/outputs will match up by location; if there's a
mismatch, we're allowed to have undefined behavior.
This may be less efficient - depending what locations were chosen, we
may have empty padding slots in the VUE. But applications presumably
use small consecutive integers for locations, so it hopefully won't be
much worse in practice.
3% of Dota 2 Reborn shaders are hurt, but only by 2 instructions.
This seems like a small price to pay for avoiding recompiles.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Chris Forbes <chrisf@ijw.co.nz>
Our plan of assigning consecutive slots doesn't work properly for
separate shader objects - at least, if we want to avoid recompiling them
whenever the interface changes.
As a first step, make assign_vue_map take an explicit slot parameter,
rather than implicitly incrementing it.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Chris Forbes <chrisf@ijw.co.nz>
Nothing actually relies on unused slots being initialized to
BRW_VARYING_SLOT_COUNT. Soon, we're going to have VUE maps with holes
in them, at which point pre-filling with BRW_VARYING_SLOT_PAD make a lot
more sense.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Chris Forbes <chrisf@ijw.co.nz>
We can't just break for padding slots. Instead, treat them like
unwritten output variables, so we handle flushing and incrementing
urb_offset correctly.
Paul introduced the concept of padding slots back in 2011, but we've
never actually used them for anything. So it's unsurprising that the
scalar VS backend didn't handle them quite right.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Chris Forbes <chrisf@ijw.co.nz>
Passing NULL to C11 threads functions isn't safe, so there's no need for
our implementation to handle it. Cuts about 1k of .text.
text data bss dec hex filename
5009514 198440 26328 5234282 4fde6a i965_dri.so before
5008346 198440 26328 5233114 4fd9da i965_dri.so after
Reviewed-by: Timothy Arceri <t_arceri@yahoo.com.au>
Reviewed-by: Brian Paul <brianp@vmware.com>
Commit 1ca25ab (glsl: Do not eliminate 'shared' or 'std140' blocks
or block members) considered as active 'shared' and 'std140' uniform
blocks and uniform block arrays, but did not include the block array
elements. Because of that, it was possible to have an active uniform
block array without any elements marked as used, making the assertion
((b->num_array_elements > 0) == b->type->is_array())
in link_uniform_blocks() fail.
Fixes the following 5 dEQP tests:
* dEQP-GLES3.functional.ubo.random.nested_structs_instance_arrays.18
* dEQP-GLES3.functional.ubo.random.nested_structs_instance_arrays.24
* dEQP-GLES3.functional.ubo.random.nested_structs_arrays_instance_arrays.19
* dEQP-GLES3.functional.ubo.random.all_per_block_buffers.49
* dEQP-GLES3.functional.ubo.random.all_shared_buffer.36
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=83508
Tested-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Timothy Arceri <t_arceri@yahoo.com.au>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
v2:
- Add tessellation shader constants support
v3:
- Add GLES 3.1 support.
v4:
- Move the getters to the proper place
Signed-off-by: Samuel Iglesias Gonsalvez <siglesias@igalia.com>
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
Including TOP_LEVEL_ARRAY_SIZE and TOP_LEVEL_ARRAY_STRIDE queries.
v2:
- Use std430_array_stride() to get top level array stride following std430's rules.
Signed-off-by: Samuel Iglesias Gonsalvez <siglesias@igalia.com>
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
The error location won't be right, but fixing that would require to check
for this as we process each type of AST node that can involve a variable
read.
v2:
- Limit the check to buffer variables, image variables have different
semantics involved.
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
v2:
- Merge the error check for the readonly qualifier with the already
existing check for variables flagged as readonly (Timothy).
- Limit the check to buffer variables, image variables have different
semantics involved (Curro).
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
v2:
- Memory qualifiers on shader storage buffer objects do not come in the form
of layout qualifiers, they are block-level qualifiers.
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
v2:
- Save memory qualifier info in the top level members of a shader
storage block.
- Add a checks to record_compare() which is used when comparing
shader storage buffer declarations in different shaders.
- Always report an error for incompatible readonly/writeonly
definitions, whether they are present at block or field level.
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
According to ARB_uniform_buffer_object spec:
"If the parameter (starting offset or size) was not specified when the
buffer object was bound (e.g. if bound with BindBufferBase), or if no
buffer object is bound to <index>, zero is returned."
Signed-off-by: Samuel Iglesias Gonsalvez <siglesias@igalia.com>
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
These handle querying the buffer name attached to a giving binding point
as well as the start offset and size of that buffer.
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
Defined in ARB_shader_storage_buffer_object extension.
Signed-off-by: Samuel Iglesias Gonsalvez <siglesias@igalia.com>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
v2:
- Add ssbo_in the names of the static functions so it is clear that this
is specific to SSBO atomics.
v3:
- Move the check after the loop (Kristian Høgsberg)
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
The original GLSL IR intrinsics have been lowered to an internal
version that accepts a block index and an offset instead of a
SSBO reference.
v2 (Connor):
- Document the sources used by the atomic intrinsics.
Reviewed-by: Connor Abbott <connor.w.abbott@intel.com>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
The first argument to SSBO atomics is a reference to a SSBO buffer variable
so we want to compute its block index and offset and provide these values
to an internal version of the intrinsic that takes them instead of the
buffer variable reference.
v2:
- Support single components of integer vectors to be passed in as arguments.
- Get interface packing information from interface's type.
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
In a later commit we will need to handle ir_swizzle nodes too, which are
not an ir_dereference. That can happen, for example, when we pass a
component of an integer vector as argument to any of the SSBO atomic
functions.
Signed-off-by: Samuel Iglesias Gonsalvez <siglesias@igalia.com>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
Shader Storage Buffer Object will add new atomic functions that are not
associated with counters, so better have atomic counter-specific functions
explicitly include the word "counter" in their names.
Reviewed-by: Timothy Arceri <t_arceri@yahoo.com.au>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
This patch moves nir_instr_insert_after_cf_list call into each case
in the intrinsics switch at nir_visitor::visit(ir_call *ir) and
define a nir_dest variable which will be used when handling
ir->return_deref after the switch.
This patch simplifies the code for nir_intrinsic_load_ssbo
implementation changes we are going to do next.
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
v2 (Connor):
- Make the STORE() macro take arguments for the extra sources (and their
size) and any extra indices required.
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
Implement helper functions that can be used to construct and send
untyped and typed surface read, write and atomic messages to the
shared dataport unit.
v2: Split from the FS implementation.
v3: Rewrite to avoid evil array_reg, emit_collect and emit_zip.
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
These functions handle the conversion of a vec4 into the form expected
by the dataport unit in message and message return payloads. The
conversion is not always trivial because some messages don't support
SIMD4x2 for some generations, in which case a strided copy may be
necessary.
v2: Split from the FS implementation.
v3: Rewrite to avoid evil array_reg, emit_collect and emit_zip.
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
See "i965/fs: Introduce FS IR builder." for the rationale.
v2: Drop scalarizing VEC4 builder.
v3: Take a backend_shader as constructor argument. Improve handling
of debug annotations and execution control flags. Rename "instr"
variable. Initialize cursor to NULL by default and add method to
explicitly point the builder at the end of the program.
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
Notice that we should differentiate between shader storage blocks and
uniform blocks, since they have different limits.
Signed-off-by: Samuel Iglesias Gonsalvez <siglesias@igalia.com>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
Otherwise, generate a link time error as per the
ARB_shader_storage_buffer_object spec.
v2:
- Fix error message (Jordan)
v3:
- Move std140_size() changes to its own patch (Kristian)
Signed-off-by: Samuel Iglesias Gonsalvez <siglesias@igalia.com>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
v2:
- Get interface packing information from interface's type, not the
variable type.
- Simplify is_std430 condition in emit_access() for readability (Jordan)
- Add a commment explaing why array of three-component vector case is
different in std430 than the rest of cases.
- Add calls to std430_array_stride().
v3:
- Simplify size_mul change for std430's case (Jordan)
- Fix commit log lines length (Jordan)
- Pass 'packing' instead of 'is_std430' to emit_access() (Kristian)
Signed-off-by: Samuel Iglesias Gonsalvez <siglesias@igalia.com>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
They are used to calculate the offset, array stride of uniform/shader
storage buffer variables. Take into account this info to get the right
value for std430.
v2:
- Fix commit log line length and indention. (Jordan)
Signed-off-by: Samuel Iglesias Gonsalvez <siglesias@igalia.com>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
v2:
- Fix a missing check in has_layout()
v3:
- Mention shader storage block in error message for layout qualifiers
(Kristian).
Signed-off-by: Samuel Iglesias Gonsalvez <siglesias@igalia.com>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
They are used to calculate size, base alignment and array stride values
for a glsl_type following std430 rules.
v2:
- Paste OpenGL 4.3 spec wording as it mentions stride of array. (Jordan)
Signed-off-by: Samuel Iglesias Gonsalvez <siglesias@igalia.com>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
This kind of definitions:
layout(xxx) buffer;
was not supported by commit 84fc5fece0.
Signed-off-by: Samuel Iglesias Gonsalvez <siglesias@igalia.com>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
Also if GL_ARB_shading_language_420pack extension is enabled.
Signed-off-by: Samuel Iglesias Gonsalvez <siglesias@igalia.com>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
The returned drm buffer object has a size multiple of 4096 but that should not
be exposed to the API user, which is working with a different size.
As far as I can see this problem is only visible in the calculation of the
length of unsized arrays used in SSBOs, as the implementation of this needs
to query the underlying buffer size via a message.
Signed-off-by: Samuel Iglesias Gonsalvez <siglesias@igalia.com>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
Otherwise we can expect odd things to happen if, for example, we ask
for the size of the attached buffer from shader code, since that
might query this value from the surface we uploaded and get random
results.
Signed-off-by: Samuel Iglesias Gonsalvez <siglesias@igalia.com>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
v2:
- Remove inst->regs_written assignment as the instruction only
writes to one register.
Signed-off-by: Samuel Iglesias Gonsalvez <siglesias@igalia.com>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
Notice that Skylake needs to include a header in the sampler message
so it will need some tweaks to work there.
Signed-off-by: Samuel Iglesias Gonsalvez <siglesias@igalia.com>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
This is how backends provide the buffer size required to compute
the size of unsized arrays in the previous patch
Signed-off-by: Samuel Iglesias Gonsalvez <siglesias@igalia.com>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
v2:
- Reduce the number of lines over 80 character line width
limit. (Thomas Hellan)
v3:
- Inject the formula to compute the array length in the IR, backends
only need to provide the buffer size (Curro)
- Create an auxiliary function to simplify code (Jordan Justen)
- Rename variables (Jordan Justen)
Signed-off-by: Samuel Iglesias Gonsalvez <siglesias@igalia.com>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
The unsized array length is computed with the following formula:
array.length() =
max((buffer_object_size - offset_of_array) / stride_of_array, 0)
Of these, only the buffer size needs to be provided by the backends, the
frontend already knows the values of the two other variables.
This patch identifies the cases where we need to get the length of an
unsized array, injecting ir_unop_ssbo_unsized_array_length expressions
that will be lowered (in a later patch) to inject the formula mentioned
above.
It also adds the ir_unop_get_buffer_size expression that drivers will
implement to provide the buffer length.
v2:
- Do not define a triop that will force backends to implement the
entire formula, they should only need to provide the buffer size
since the other values are known by the frontend (Curro).
v3:
- Call state->has_shader_storage_buffer_objects() in ast_function.cpp instead
of using state->ARB_shader_storage_buffer_object_enable (Tapani).
Signed-off-by: Samuel Iglesias Gonsalvez <siglesias@igalia.com>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
They only can be defined in the last position of the shader
storage blocks.
When an unsized array is used in different shaders, it might be
converted in different sized arrays, avoid get a linker error
in that case.
v2:
- Rework error condition and error messages (Timothy Arceri)
v3:
- Move OpenGL ES check to its own patch.
Signed-off-by: Samuel Iglesias Gonsalvez <siglesias@igalia.com>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
Buffer variables are the same as uniforms, only that read/write, so we want
the same treatment.
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
Since these are a special kind of UBOs we emit them together reusing the
same infrastructure, however, we use a RAW surface so we can reuse
existing untyped read/write/atomic messages which include a pixel mask
header that we need to set to obtain correct behavior with helper
invocations of the fragment shader.
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
v2:
- Set it after the driver's MaxShaderStorageBuffers value assignment.
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
We use the same dirty state for SSBOs and UBOs because they share the
same infrastructure.
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
This should be a cacheline (64 bytes) so that we can safely have the
CPU and GPU writing the same SSBO on non-cachecoherent systems (our
Atom CPUs). With UBOs, the GPU never writes, so there's no
problem. For an SSBO, the GPU and the CPU can be updating disjoint
regions of the buffer simultaneously and that will break if the
regions overlap the same cacheline.
v2:
- Use cacheline size (64 bytes) instead of 16 bytes (Kristian).
- Update commit log and add a comment in the code explaining
why we use cacheline size (Ben).
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
This makes sure that user is still able to query properties about
variables that have gotten packed by lower_packed_varyings pass.
Fixes following OpenGL ES 3.1 test:
ES31-CTS.program_interface_query.separate-programs-vertex
v2: fix 'name included in packed list' check (Ilia Mirkin)
v3: iterate over instances of name using strtok_r (Ilia Mirkin)
Signed-off-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Marta Lofstedt <marta.lofstedt@intel.com>
This is required to store information about packed varyings, currently
these variables get lost and cannot be retrieved later in sensible way
for program interface queries. List will be utilized by next patch.
Signed-off-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Marta Lofstedt <marta.lofstedt@intel.com>
If the immutable compressed texture didn't have the full mip pyramid,
this didn't work, because it tried to generate mip levels for non-existing
levels. _mesa_prepare_mipmap_level() would correctly handle this by returning
FALSE if the mip level didn't exist, however we actually created the
non-existing mip level right before that because we used _mesa_get_tex_image()
before calling _mesa_prepare_mipmap_level(). It would then proceed to crash
(we allocated the mip level, which is a bad idea on an immutable texture,
but didn't initialize the values, leading to assertion failures or segfaults).
Fix this by using _mesa_select_tex_image() instead and call it after
_mesa_prepare_mipmap_level(), as that function will allocate missing mip levels
for non-immutable textures already.
This fixes a (2 year old) crash with astromenace which was hack-fixed in ubuntu
packages instead: http://bugs.debian.org/718680 (I guess most apps do full mip
chains - I believe this app not doing it is actually unintentional, always one
level less than full mip chain...).
Cc: "10.6 11.0" <mesa-stable@lists.freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
... with only ARB_shader_atomic_counters.
I expected to see interactions with ARB_tessellation_shader in the
ARB_shader_atomic_counters spec, but they do not exist. It seems that we
should unconditionally expose these variables in the presence of
ARB_shader_atomic_counters:
gl_MaxTessControlAtomicCounters
gl_MaxTessEvaluationAtomicCounters
This partially reverts commit da7adb99e8. The commit also affected
gl_MaxTessControlImageUniforms and gl_MaxTessEvaluationImageUniforms
similarly but the ARB_shader_image_load_store spec does list an
interaction with ARB_tessellation_shader.
Cc: "11.0" <mesa-stable@lists.freedesktop.org>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=92095
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Without this commit, copy propagation is discarded if it involves
a uniform with an instruction that has 3 sources. But 3 sourced
instructions can access scalar values.
For example, this is what vec4_visitor::fix_3src_operand() is already
doing:
if (src.file == UNIFORM && brw_is_single_value_swizzle(src.swizzle))
return src;
Shader-db results (unfiltered) on NIR:
total instructions in shared programs: 6259650 -> 6241985 (-0.28%)
instructions in affected programs: 812755 -> 795090 (-2.17%)
helped: 7930
HURT: 0
Shader-db results (unfiltered) on IR:
total instructions in shared programs: 6445822 -> 6441788 (-0.06%)
instructions in affected programs: 296630 -> 292596 (-1.36%)
helped: 2533
HURT: 0
v2:
- Updated commit message, using Matt Turner suggestions
- Move the check after we've created the final value, as Jason
Ekstrand suggested
- Clean up the condition
v3:
- Move the check back to the original place, to keep things
tidy, as suggested by Jason Ekstrand
v4:
- Fixed missing is_single_value_swizzle() as pointed by Jason Ekstrand
Reviewed-by: Matt Turner <mattst88@gmail.com>
When we assign hw regs to attributes, we don't incorporate the stride
and subreg_offset from the fs_reg. It's rarely used, but the integer
multiplication lowering uses unusual stride and subreg_offset
combination breaks when one source is an attribute.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=91970
Cc: "10.6 11.0" <mesa-stable@lists.freedesktop.org>
Signed-off-by: Kristian Høgsberg Kristensen <krh@bitplanet.net>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Previously, core Mesa's _mesa_CopyImageSubData() created temporary textures
to wrap renderbuffer sources/destinations. This caused a bit of a mess in
the Mesa/gallium state tracker because we had to basically undo that
wrapping.
Instead, change ctx->Driver.CopyImageSubData() to take both gl_renderbuffer
and gl_texture_image src/dst pointers (one being null, the other non-null)
so the driver can handle renderbuffer vs. texture as needed.
For the i965 driver, we basically moved the code that wrapped textures
around renderbuffers from copyimage.c down into the met and driver code.
The old code in copyimage.c also made some questionable calls to
_mesa_BindTexture(), etc. which weren't undone at the end.
v2 (Jason Ekstrand): Rework the intel bits
v3 (Brian Paul): Update the temporary st_CopyImageSubData() function.
Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
Tested-by: Kai Wasserbäch <kai@dev.carbon-project.org>
Tested-by: Nick Sarnie <commendsarnex@gmail.com>
Check for PIPE_FORMAT_R8_UNORM when setting up the copy shader.
Also re-enable the dest alpha blending with A8 destination that
actually turned out to be correct.
Verified using rendercheck that the composite operators
overreverse, in, out, atop, atopreverse and xor seem to work fine
with a8 destiation.
v2: Fix a copy-paste error.
Reported-by: Jose Fonseca <jfonseca@vmware.com>
Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
It doesn't matter whether a write is saturated or not, in another
implementation it might even have been a separate opcode. This code was
most likely copied from the copy-propagation pass (where one does have
to distinguish saturation).
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
I left a bunch of code indented a level in the previous patch to make
the diff easier to read. But now we should fix that.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
By performing the vertex counting in NIR, we're able to elide a ton of
useless safety checks around every EmitVertex() call:
total instructions in shared programs: 3952 -> 3720 (-5.87%)
instructions in affected programs: 3491 -> 3259 (-6.65%)
helped: 11
HURT: 0
Improves performance in Gl32GSCloth by 0.671742% +/- 0.142202% (n=621)
on Haswell GT3e at 1024x768.
This should also make it easier to implement Broadwell's "Static Vertex
Count" feature someday.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
This patch also introduces a lowering pass to convert the simple GS
intrinsics to the new ones. See the comments above that for the
rationale behind the new intrinsics.
This should be useful for i965; it's a generic enough mechanism that I
could see other drivers potentially using it as well, so I don't feel
too bad about putting it in the generic code.
v2:
- Use nir_after_block_before_jump for the cursor (caught by Jason
Ekstrand - I'd mistakenly used nir_after_block when rebasing this
code onto the new NIR control flow API).
- Remove the old emit_vertex intrinsic at the end, rather than in
the middle (requested by Jason).
- Use state->... directly rather than locals (requested by Jason).
- Report progress from nir_lower_gs_intrinsics() (requested by me).
- Remove "Authors:" section from file comment (requested by
Michael Schellenberger Costa).
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
The NIR control flow modification API churns the block structure,
splitting blocks, stitching them back together, and so on. Preserving
information about block dominance is hard (and probably not worthwhile).
This patch makes nir_cf_extract() throw away all metadata, like we do
when adding/removing jumps.
We then make the dead control flow pass compute dominance information
right before it uses it. This is necessary because earlier work by the
pass may have invalidated it.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
Calling unlink_blocks(block, block->successors[0]) will successfully
unlink the first successor, but then will shift block->successors[1]
down to block->successor[0]. So the successors[1] != NULL check will
always fail.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
Consider the case of "while (...) { break }". Or in NIR:
block block_0 (0x7ab640):
...
/* succs: block_1 */
loop {
block block_1:
/* preds: block_0 */
break
/* succs: block_2 */
}
block block_2:
Calling nir_handle_remove_jump(block_1, nir_jump_break) will remove the break.
Unfortunately, it would mangle the predecessors and successors.
Here, block_2->predecessors->entries == 1, so we would create a fake
link, setting block_1->successors[1] = block_2, and adding block_1 to
block_2's predecessor set. This is illegal: a block cannot specify the
same successor twice. In particular, adding the predecessor would have
no effect, as it was already present in the set.
We'd then call unlink_block_successors(), which would delete the fake
link and remove block_1 from block_2's predecessor set. It would then
delete successors[0], and attempt to remove block_1 from block_2's
predecessor set a second time...except that it wouldn't be present,
triggering an assertion failure.
The fix appears to be simple: simply unlink the block's successors and
recreate them to point at the correct blocks first. Then, add the fake
link. In the above example, removing the break would cause block_1 to
have itself as a successor (as it becomes an infinite loop), so adding
the fake link won't cause a duplicate successor.
v2: Add comments (requested by Connor Abbott) and fix commit message.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
There is a bug where we mess up predecessors/successors due to the
ordering of unlinking/recreating edges/adding fake edges. In order to
fix that, I need everything in one routine.
However, calling block_add_normal_succs() isn't safe from
cleanup_cf_node() - it would crash trying to insert phi undefs.
So unfortunately I need to add a parameter.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
Consider the following NIR:
block block_0;
/* succs: block_1 block_2 */
if (...) {
block block_1;
...
} else {
block block_2;
}
Calling split_block_beginning() on block_1 would break block_0's
successors: link_block() sets both successors of a block, so calling
link_block(block_0, new_block, NULL) would throw away the second
successor, leaving only /* succ: new_block */. This is invalid: the
block before an if statement must have two successors.
Changing the call to link_block(pred, new_block, pred->successors[0])
would correctly leave both successors in place, but because unlink_block
may shift successor[1] to successor[0], it may not preserve the original
order. NIR maintains a convention that successor[0] must point to the
"then" block, while successor[1] points to the "else" block, so we need
to take care to preserve this ordering.
This patch creates a new function that swaps out one successor for
another, preserving the ordering. It then uses this to fix the issue.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
I need to do this in a second place, and I'd rather make a helper
function than cut and paste the code.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
This is invalid, and causes disasters if we try to unlink successors:
removing the first will work, but removing the second copy will fail
because the block isn't in the successor's predecessor set any longer.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
It's possible that, if a vecN operation is involved in a phi node, that we
could end up moving from a register to itself. If swizzling is involved,
we need to emit the move but. However, if there is no swizzling, then the
mov is a no-op and we might as well not bother emitting it.
Shader-db results on Haswell:
total instructions in shared programs: 6262536 -> 6259558 (-0.05%)
instructions in affected programs: 184780 -> 181802 (-1.61%)
helped: 838
HURT: 0
Reviewed-by: Eduardo Lima Mitev <elima@igalia.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
I don't know of any piglit tests that are currently broken. However, there
is nothing stopping a vecN instruction from getting source modifiers and
lower_vec_to_movs is run after we lower to source modifiers.
Reviewed-by: Eduardo Lima Mitev <elima@igalia.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
../../../../../src/mesa/tnl_dd/t_dd_dmatmp.h:83:28: warning: comparison between signed and unsigned integer expressions [-Wsign-compare]
nr = MIN2(currentsz, count - j);
^
../../../../../src/mesa/tnl_dd/t_dd_dmatmp.h:83:55: warning: signed and unsigned type in conditional expression [-Wsign-compare]
nr = MIN2(currentsz, count - j);
^
../../../../../src/mesa/tnl_dd/t_dd_dmatmp.h:116:25: warning: comparison between signed and unsigned integer expressions [-Wsign-compare]
nr = MIN2(currentsz, count - j);
^
../../../../../src/mesa/tnl_dd/t_dd_dmatmp.h:116:52: warning: signed and unsigned type in conditional expression [-Wsign-compare]
nr = MIN2(currentsz, count - j);
^
../../../../../src/mesa/tnl_dd/t_dd_dmatmp.h:140:25: warning: comparison between signed and unsigned integer expressions [-Wsign-compare]
nr = MIN2(currentsz, count - j);
^
../../../../../src/mesa/tnl_dd/t_dd_dmatmp.h:140:52: warning: signed and unsigned type in conditional expression [-Wsign-compare]
nr = MIN2(currentsz, count - j);
^
../../../../../src/mesa/tnl_dd/t_dd_dmatmp.h: In function 'intel_render_line_loop_verts':
../../../../../src/mesa/tnl_dd/t_dd_dmatmp.h:174:28: warning: comparison between signed and unsigned integer expressions [-Wsign-compare]
nr = MIN2(currentsz, count - j);
^
../../../../../src/mesa/tnl_dd/t_dd_dmatmp.h:174:55: warning: signed and unsigned type in conditional expression [-Wsign-compare]
nr = MIN2(currentsz, count - j);
^
../../../../../src/mesa/tnl_dd/t_dd_dmatmp.h:224:25: warning: comparison between signed and unsigned integer expressions [-Wsign-compare]
nr = MIN2(currentsz, count - j);
^
../../../../../src/mesa/tnl_dd/t_dd_dmatmp.h:224:52: warning: signed and unsigned type in conditional expression [-Wsign-compare]
nr = MIN2(currentsz, count - j);
^
../../../../../src/mesa/tnl_dd/t_dd_dmatmp.h:255:25: warning: comparison between signed and unsigned integer expressions [-Wsign-compare]
nr = MIN2(currentsz, count - j);
^
../../../../../src/mesa/tnl_dd/t_dd_dmatmp.h:255:52: warning: signed and unsigned type in conditional expression [-Wsign-compare]
nr = MIN2(currentsz, count - j);
^
../../../../../src/mesa/tnl_dd/t_dd_dmatmp.h:281:25: warning: comparison between signed and unsigned integer expressions [-Wsign-compare]
nr = MIN2(currentsz, count - j + 1);
^
../../../../../src/mesa/tnl_dd/t_dd_dmatmp.h:281:56: warning: signed and unsigned type in conditional expression [-Wsign-compare]
nr = MIN2(currentsz, count - j + 1);
^
../../../../../src/mesa/tnl_dd/t_dd_dmatmp.h: In function 'intel_render_poly_verts':
../../../../../src/mesa/tnl_dd/t_dd_dmatmp.h:313:28: warning: comparison between signed and unsigned integer expressions [-Wsign-compare]
nr = MIN2(currentsz, count - j + 1);
^
../../../../../src/mesa/tnl_dd/t_dd_dmatmp.h:313:59: warning: signed and unsigned type in conditional expression [-Wsign-compare]
nr = MIN2(currentsz, count - j + 1);
^
../../../../../src/mesa/tnl_dd/t_dd_dmatmp.h:365:28: warning: comparison between signed and unsigned integer expressions [-Wsign-compare]
nr = MIN2(currentsz, count - nr);
^
../../../../../src/mesa/tnl_dd/t_dd_dmatmp.h:365:56: warning: signed and unsigned type in conditional expression [-Wsign-compare]
nr = MIN2(currentsz, count - nr);
^
../../../../../src/mesa/tnl_dd/t_dd_dmatmp.h:83:28: warning: comparison between signed and unsigned integer expressions [-Wsign-compare]
nr = MIN2(currentsz, count - j);
^
../../../../../src/mesa/tnl_dd/t_dd_dmatmp.h:83:55: warning: signed and unsigned type in conditional expression [-Wsign-compare]
nr = MIN2(currentsz, count - j);
^
../../../../../src/mesa/tnl_dd/t_dd_dmatmp.h:116:25: warning: comparison between signed and unsigned integer expressions [-Wsign-compare]
nr = MIN2(currentsz, count - j);
^
../../../../../src/mesa/tnl_dd/t_dd_dmatmp.h:116:52: warning: signed and unsigned type in conditional expression [-Wsign-compare]
nr = MIN2(currentsz, count - j);
^
../../../../../src/mesa/tnl_dd/t_dd_dmatmp.h:140:25: warning: comparison between signed and unsigned integer expressions [-Wsign-compare]
nr = MIN2(currentsz, count - j);
^
../../../../../src/mesa/tnl_dd/t_dd_dmatmp.h:140:52: warning: signed and unsigned type in conditional expression [-Wsign-compare]
nr = MIN2(currentsz, count - j);
^
../../../../../src/mesa/tnl_dd/t_dd_dmatmp.h: In function 'radeon_dma_render_line_loop_verts':
../../../../../src/mesa/tnl_dd/t_dd_dmatmp.h:174:28: warning: comparison between signed and unsigned integer expressions [-Wsign-compare]
nr = MIN2(currentsz, count - j);
^
../../../../../src/mesa/tnl_dd/t_dd_dmatmp.h:174:55: warning: signed and unsigned type in conditional expression [-Wsign-compare]
nr = MIN2(currentsz, count - j);
^
../../../../../src/mesa/tnl_dd/t_dd_dmatmp.h:224:25: warning: comparison between signed and unsigned integer expressions [-Wsign-compare]
nr = MIN2(currentsz, count - j);
^
../../../../../src/mesa/tnl_dd/t_dd_dmatmp.h:224:52: warning: signed and unsigned type in conditional expression [-Wsign-compare]
nr = MIN2(currentsz, count - j);
^
../../../../../src/mesa/tnl_dd/t_dd_dmatmp.h:255:25: warning: comparison between signed and unsigned integer expressions [-Wsign-compare]
nr = MIN2(currentsz, count - j);
^
../../../../../src/mesa/tnl_dd/t_dd_dmatmp.h:255:52: warning: signed and unsigned type in conditional expression [-Wsign-compare]
nr = MIN2(currentsz, count - j);
^
../../../../../src/mesa/tnl_dd/t_dd_dmatmp.h:281:25: warning: comparison between signed and unsigned integer expressions [-Wsign-compare]
nr = MIN2(currentsz, count - j + 1);
^
../../../../../src/mesa/tnl_dd/t_dd_dmatmp.h:281:56: warning: signed and unsigned type in conditional expression [-Wsign-compare]
nr = MIN2(currentsz, count - j + 1);
^
../../../../../src/mesa/tnl_dd/t_dd_dmatmp.h: In function 'radeon_dma_render_poly_verts':
../../../../../src/mesa/tnl_dd/t_dd_dmatmp.h:313:28: warning: comparison between signed and unsigned integer expressions [-Wsign-compare]
nr = MIN2(currentsz, count - j + 1);
^
../../../../../src/mesa/tnl_dd/t_dd_dmatmp.h:313:59: warning: signed and unsigned type in conditional expression [-Wsign-compare]
nr = MIN2(currentsz, count - j + 1);
^
../../../../../src/mesa/tnl_dd/t_dd_dmatmp.h:365:28: warning: comparison between signed and unsigned integer expressions [-Wsign-compare]
nr = MIN2(currentsz, count - nr);
^
../../../../../src/mesa/tnl_dd/t_dd_dmatmp.h:365:56: warning: signed and unsigned type in conditional expression [-Wsign-compare]
nr = MIN2(currentsz, count - nr);
^
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Two drivers use this file, and neither supports ELTs.
No piglit regressions on i915 (G33) or radeon (Radeon 7500).
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Two drivers use this file, and both support triangle fans.
No piglit regressions on i915 (G33) or radeon (Radeon 7500).
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
v2: Fix '- nr' typo noticed by Marius.
No piglit regressions on i915 (G33) or radeon (Radeon 7500).
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Brian Paul <brianp@vmware.com> [v1]
Two drivers use this file, and both support triangle strips.
No piglit regressions on i915 (G33) or radeon (Radeon 7500).
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Two drivers use this file, and both support triangles.
No piglit regressions on i915 (G33) or radeon (Radeon 7500).
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Two drivers use this file, and both support line strips.
No piglit regressions on i915 (G33) or radeon (Radeon 7500).
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Two drivers use this file, and both support lines.
No piglit regressions on i915 (G33) or radeon (Radeon 7500).
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Two drivers use this file, and neither supports quads.
No piglit regressions on i915 (G33) or radeon (Radeon 7500).
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Two drivers use this file, and neither supports quad strips.
No piglit regressions on i915 (G33) or radeon (Radeon 7500).
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
The value passed in count previously was "vertex after the last vertex
to be processed." Calling that "count" was misleading and kind of mean.
Looking at the code, many functions immediately do "count-start" to get
back the true count. That's just silly.
If it is better for the loops to be 'for (j = start; j < (start +
count); j++)', GCC will do that transformation.
NOTE: There is some strange formatting left by this patch. That was
done to make it more obvious that the before and after code is
equivalent. These will be fixed in the next patch.
No piglit regressions on i915 (G33) or radeon (Radeon 7500).
v2: Fix a remaining (count-start) in render_quad_strip_verts.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Brian Paul <brianp@vmware.com> [v1]
Cc: "10.6 11.0" <mesa-stable@lists.freedesktop.org>
From section 9.2. Binding and Managing Framebuffer Objects:
"Upon successful return from Get*FramebufferAttachmentParameteriv, if
pname is FRAMEBUFFER_ATTACHMENT_OBJECT_TYPE, then params will contain
one of NONE, FRAMEBUFFER_DEFAULT, TEXTURE, or RENDERBUFFER, identifying
the type of object which contains the attached image."
And then it clarifies further:
"If the value of FRAMEBUFFER_ATTACHMENT_OBJECT_TYPE is NONE, then
either no framebuffer is bound to target; or the default framebuffer is
bound, attachment is DEPTH or STENCIL, and the number of depth or stencil
bits, respectively, is zero"
Currently, if the default framebuffer is bound, we always return
GL_FRAMEBUFFER_DEFAULT for FRAMEBUFFER_ATTACHMENT_OBJECT_TYPE, but
according to the spec, when GL_DEPTH or GL_STENCIL attachments are
the ones being queried, we should return GL_NONE if they don't exist.
Fixes the following dEQP test:
dEQP-GLES3.functional.state_query.fbo.framebuffer_attachment_x_size_initial
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Cc: "10.6" <mesa-stable@lists.freedesktop.org>
Patch fixes a crash in conformance test that tries out different
invalid arguments for glShaderSource and glGetShaderSource:
ES2-CTS.gtf.GL.glGetShaderSource.getshadersource_programhandle
This is a regression from commit:
04e201d0c0
Additions in v2 also fix following failing deqp test:
dEQP-GLES[2|3].functional.negative_api.shader.shader_source
v2: cleanup function, do check earlier (Iago Toral)
Signed-off-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
We don't use any of the code after the switch anyway. Since we check for
num_components == 1 and early-return, it doesn't get executed so
everything's ok. However, it makes it much clearer what's going on if we
simply do an early return.
Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
v2 (Ken):
- Squash together commits for HS, DS, and TE, as well as fixes.
- Add INTEL_MASK variants so we can use SET_FIELD if we want.
- Rename GEN7_HS_INSTANCE_CONTROL to GEN7_HS_INSTANCE_COUNT to match
the documentation.
- Add some more fields from the PRMs.
- Add Broadwell variants.
Signed-off-by: Chris Forbes <chrisf@ijw.co.nz>
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
"r600g: apply disable workaround on all scissors" forgot to update
num_dw, fix it.
Fixes: fbb423b433 "r600g: apply disable workaround on all scissors"
Reported-and-tested-by: Markus Trippelsdorf <markus@trippelsdorf.de>
Signed-off-by: Grazvydas Ignotas <notasas@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
Now it is more similar to brw_fs_copy_propagation, with three
clear stages:
1) Build up the value we are propagating as if it were the source of a
single MOV:
2) Check that we can propagate that value
3) Build the final value
Previously everything was somewhat messed up, making the
implementation on some specific cases, like knowing if you can
propagate from a previous instruction even with type mismatches, even
messier (for example, with the need of maintaining more of one
has_source_modifiers). The refactoring clears stuff, and gives
support to this mentioned use case without doing anything extra
(for example, only one has_source_modifiers is used).
Shader-db results for vec4 programs on Haswell:
total instructions in shared programs: 1683842 -> 1669037 (-0.88%)
instructions in affected programs: 739837 -> 725032 (-2.00%)
helped: 6237
HURT: 0
v2: using 'arg' index to get the from inst was wrong
v3: rebased against last change on the previous patch of the series
v4: don't need to track instructions on struct copy_entry, as we
only set the source on a direct copy
v5: change the approach for a refactoring
v6: tweaked comments
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
The function was a no-op and if the ctx->Driver.BindFramebuffer pointer
is null, Mesa won't try to use it.
Reviewed-by: Matt Turner <mattst88@gmail.com>
According to OpenGL ES 3.1 specification 10.3.1:
"An INVALID_OPERATION error is generated if buffer is not zero
or a name returned from a previous call to GenBuffers,
or if such a name has since been deleted with DeleteBuffers."
This error check was previously limited to OpenGL Core.
Signed-off-by: Marta Lofstedt <marta.lofstedt@intel.com>
The only functional change here is that we now set EmitNoIndirectOutput and
EmitNoIndirectTemp for compute shaders. Compute shaders don't have outputs
per-se and we should have been setting EmitNoIndirectTemp all along.
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
All SKL SKUs except the lowest one which has half the L3 size actually have 384K
of URB per slice.
For once, I can explain how this mistake was made and how it was missed in
review... Historically when we enable a platform and put the production sizes,
you can simply look at the "smallest" SKU and see what its URB size is (and we
assumed it was the 1 slice variant). Since on newer platforms the URB sizes are
scaled automatically by HW, this was sufficient. On SKL, this is a bit different
as the lowest SKU actually has half of the L3 fused off. GT2 is the 1 slice (not
GT1) variant and it has 384K.
There are no Jenkins tests fixed (or regressions) and we don't expect any fixes
here because you can always run with less URB size.
Thanks to Sarah for bringing this to my attention.
Cc: Sarah Sharp <sarah.a.sharp@intel.com>
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Designated initializers are not allowed in C++ (not even C++11). Since
nir_lower_samplers is now using nir_builder, and nir_lower_samplers is in
C++, this breaks the build on some compilers. Aparently, GCC 5 allows it
in some limited extent because mesa still builds on my system without this
patch.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=92052
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
With the only C++ function having its own wrapper we can 'demote' this
file to a normal C one. This allows us to get rid of extern C { #include
<foo.h> } 'hacks'. Plus some of the headers may use C99 initializers,
which are not supported by the ISO standard.
This may cause build issue on incremental builds. If so run the
following:
sed -i -e 's|samplers\.cpp|samplers.c|' src/glsl/nir/.deps/nir_lower_samplers.Plo
Fixes: ef8eebc6ad5(nir: support indirect indexing samplers in struct arrays)
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Reported-by: Gottfried Haider <gottfried.haider@gmail.com>
Tested-by: Gottfried Haider <gottfried.haider@gmail.com>
Reviewed-by: Timothy Arceri <t_arceri@yahoo.com.au>
compr4 is represented by setting the high bit on the MRF number.
We need to mask it out before sanity checking the register number.
Fixes ~8000 assert fails on Ironlake and G45.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=92066
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
res_ptr already contains the resource values. fmask_ptr needs to be
looked up relative to the start of the resource params.
Note that this only affects indirect loads of MS sampler arrays.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Cc: "11.0" <mesa-stable@lists.freedesktop.org>
There are some bug reports about shaders failing to compile in gen6
because MRF 14 is used when we need to spill. For example:
https://bugs.freedesktop.org/show_bug.cgi?id=86469https://bugs.freedesktop.org/show_bug.cgi?id=90631
Discussion in bugzilla pointed to the fact that gen6 might actually have
24 MRF registers available instead of 16, so we could use other MRF
registers and avoid these conflicts (we still need to investigate why
some shaders need up to MRF 14 anyway, since this is not expected).
Notice that the hardware docs are not clear about this fact:
SNB PRM Vol4 Part2's "Table 5-4. MRF Registers Available in Device
Hardware" says "Number per Thread" - "24 registers"
However, SNB PRM Vol4 Part1, 1.6.1 Message Register File (MRF) says:
"Normal threads should construct their messages in m1..m15. (...)
Regardless of actual hardware implementation, the thread should
not assume th at MRF addresses above m15 wrap to legal MRF registers."
Therefore experimentation was necessary to evaluate if we had these extra
MRF registers available or not. This was tested in gen6 using MRF
registers 21..23 for spilling and doing a full piglit run (all.py) forcing
spilling of everything on the FS backend. It was also tested by doing
spilling of everything on both the FS and the VS backends with a piglit run
of shader.py. In both cases no regressions were observed. In fact, many of
these tests where helped in the cases where we forced spilling, since that
triggered the same underlying problem described in the bug reports. Here are
some results using INTEL_DEBUG=spill_fs,spill_vec4 for a shader.py run on
gen6 hardware:
Using MRFs 13..15 for spilling:
crash: 2, fail: 113, pass: 6621, skip: 5461
Using MRFs 21..23 for spilling:
crash: 2, fail: 12, pass: 6722, skip: 5461
This patch sets the ground for later patches to implement spilling
using MRF registers 21..23 in gen6.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
In a later patch we will make BRW_MAX_MRF return a different value depending
on the hardware generation, but it is inconvenient to add a gen parameter
to the brw_reg functions only for the assertions, so move these to places where
we have the hardware generation available.
Ken suggested to add the asserts to brw_set_src0 and brw_set_dest since that
would make sure that we catch all uses of MRF registers, even those coming
from modules that generate native code directly, like blorp. Unfortunately,
this is very late in the process which can make things harder to debug, so add
asserts to the generator as well.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Until now we only used MRFs 1..15 for regular SEND messages, so the
message length could not possibly exceed the maximum size. Soon we'll
allow to use MRF registers 1..23 in gen6, so we need to be careful
not to build messages that can go beyond the limit. That could occur,
specifically, when building URB write messages, which we may need to
split in chunks due to their size. Previously we would simply go and
create a new message when we reached MRF 13 (since 13..15 were
reserved for spilling), now we also want to check the size of the
message explicitly.
Besides adding that condition to split URB write messages properly,
this patch also adds asserts in the generator. Notice that
brw_inst_set_mlen already asserts for this, but asserting in the
generators is easy and can make debugging easier in some cases.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Not something actually hit in real life (now state is never non-null,
but only case state->syms is null is if nir_print_instr() path). But it
was something I overlooked the first time, so might as well fix it.
*** CID 1324642: Null pointer dereferences (REVERSE_INULL)
/src/glsl/nir/nir_print.c: 299 in print_var_decl()
293
294 fprintf(fp, " (%s, %u)", loc, var->data.driver_location);
295 }
296
297 fprintf(fp, "\n");
298
>>> CID 1324642: Null pointer dereferences (REVERSE_INULL)
>>> Null-checking "state" suggests that it may be null, but it has already been dereferenced on all paths leading to the check.
299 if (state) {
300 _mesa_set_add(state->syms, name);
301 _mesa_hash_table_insert(state->ht, var, name);
302 }
303 }
304
Signed-off-by: Rob Clark <robclark@freedesktop.org>
For consistency, either we have all class members dereferenced, or none.
In this case, very few are so lets get rid of them all.
Reviewed-by: Timothy Arceri <t_arceri@yahoo.com.au>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reuse utility functions instead of reimplementing the same logic.
* _mesa_is_compressed_format() performs the required checking to
determine format support in the current context.
* _mesa_gl_compressed_format_base_format() returns the base format.
As a side effect, we now check that we're in a desktop context when
determining support for the FXT1 and RGTC formats. This is in agreement
with our extension table and the glext headers.
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
Signed-off-by: Nanley Chery <nanley.g.chery@intel.com>
Instead of case statements, use _mesa_get_format_layout() to
determine if a GL format is part of a family of compressed formats.
v2. restrict LATC formats to API_OPENGL_COMPAT (Ilia).
rename the variable mFormat to m_format.
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
Signed-off-by: Nanley Chery <nanley.g.chery@intel.com>
This enables us to predicate statments on a compressed format being
a type of LATC format. Also, remove the comment that lists the enum
(it was getting a tad long).
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
Signed-off-by: Nanley Chery <nanley.g.chery@intel.com>
With this, we completely switch over to nir lowering passes instead of
tgsi_lowering. So one step closer to supporting direct glsl or spirv to
nir support for freedreno a3xx/a4xx.
Signed-off-by: Rob Clark <robclark@freedesktop.org>
Some hardware needs to clamp texture coordinates to [0.0, 1.0] in the
shader to emulate GL_CLAMP. This is added to lower_tex_proj since, in
the case of projected coords, the clamping needs to happen *after*
projection.
v2: comments/suggestions from Ilia and Eric, use txs to get texture size
and clamp RECT textures to their dimensions rather than [0.0, 1.0] to
avoid having to lower RECT textures to 2D.
Signed-off-by: Rob Clark <robclark@freedesktop.org>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
v2: comments/suggestions from Ilia and Eric, split out get_texture_size()
helper so we can use it in the next commit for clamping RECT textures.
Signed-off-by: Rob Clark <robclark@freedesktop.org>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Some hardware, such as adreno a3xx, supports txp on some but not all
sampler types. In this case we want more fine grained control over
which texture projectors get lowered.
v2: split out nir_lower_tex_options struct to make it easier to
add the additional parameters coming in the following patches
Signed-off-by: Rob Clark <robclark@freedesktop.org>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Since the following patches will add additional tex-lowering related
functionality, which doesn't make sense to split out into a separate
pass (as they would require duplication of the projector lowering
logic), let's give this pass a more generic name.
Signed-off-by: Rob Clark <robclark@freedesktop.org>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
SEL and MOV instructions, as long as they don't have source modifiers, are
just copying bits around. So those kind of instruction could be propagated
even if there are type mismatches. This is needed because NIR generates
integer SEL and MOV instructions whenever it doesn't know what else to
generate.
This commit adds support for copy propagation using current instruction
as reference.
Equivalent to commit 472ef9 but for vec4.
v2: include check for saturate, as Jason Ekstrand suggested
v3: check that the dst.type and the src type are the same, in order to
solve (among others) the following deqp regression with v2:
dEQP-GLES3.functional.shaders.operator.unary_operator.minus.lowp_uint_vertex
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
brw_fs_visitor.cpp: In member function 'void fs_visitor::emit_urb_writes()':
brw_fs_visitor.cpp:977:58: warning: comparison between signed and unsigned integer expressions [-Wsign-compare]
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
OpenGL ES 3.0 spec 3.7.2 "Transfer of Pixel Rectangles" specifies
DEPTH_COMPONENT, UNSIGNED_INT as a valid couple, validation for
internal format is checked by is_float_depth().
Fix regression caused by 81d2fd91a9 in:
ES3-CTS.gtf.GL3Tests.packed_pixels.packed_pixels
Test uses GL_DEPTH_COMPONENT, UNSIGNED_INT only when GL_NV_read_depth
extension is present.
v2: change check in _mesa_error_check_format_and_type to be explicit
for ES 2.0+, desktop OpenGL does not allow this behaviour + uses
this function for both glReadPixels and glDrawPixels validation.
(No Piglit regressions seen with v2.)
Signed-off-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com> [v1]
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=92009
Cc: "10.6 11.0" <mesa-stable@lists.freedesktop.org>
Fixes:
In file included from nir/nir_lower_samplers.cpp:27:0:
nir/nir_builder.h: In function 'nir_ssa_def* nir_channel(nir_builder*, nir_ssa_def*, int)':
nir/nir_builder.h:222:37: warning: narrowing conversion of 'c' from 'int' to 'unsigned int' inside { } is ill-formed in C++11 [-Wnarrowing]
unsigned swizzle[4] = {c, c, c, c};
Signed-off-by: Rob Clark <robclark@freedesktop.org>
Use nir_lower_clip pass for adding the VS/FS instructions to handle
user-clip-planes and CLIPDIST. Wire up support for load_user_clip_plane
intrinsic to fetch ucp[plane] values as driver-params (passed as const's
to the shader).
Signed-off-by: Rob Clark <robclark@freedesktop.org>
The vertex shader lowering adds calculation for CLIPDIST, if needed
(ie. user-clip-planes), and the frag shader lowering adds conditional
kills based on CLIPDIST value (which should be treated as a normal
interpolated varying by the driver).
Note that this won't quite do the right thing in the face of MSAA plus
user-clip-planes, since all the samples would be killed or not (rather
than potentially only a portion of them). But it's better than no UCP
support at all for drivers that don't have this in hw.
Signed-off-by: Rob Clark <robclark@freedesktop.org>
Reviewed-by: Eric Anholt <eric@anholt.net>
For lowering user-clip-planes, we need a way to pass the enabled/used
user-clip-planes in to shader.
Signed-off-by: Rob Clark <robclark@freedesktop.org>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Used internally in freedreno/ir3 to calc stream-out position. Seems
like a generic enough way to implement stream-out (using str instrs),
plus it avoids compiler warnings by sneaking in a non-enum value in
switch statements.
Signed-off-by: Rob Clark <robclark@freedesktop.org>
When updating texture buffers, we might end up replacing the whole
buffer. Check that the tic address matches the resource address, and if
not, update the tic and reupload it.
This fixes:
arb_direct_state_access-texture-buffer
arb_texture_buffer_object-data-sync
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Cc: "11.0" <mesa-stable@lists.freedesktop.org>
Instructions with difference in PM field can actually be paired up if
the one without PM doesn't do packing/unpacking and non-NOP
packing/unpacking operations from PM instruction aren't added to the
other without PM.
total instructions in shared programs: 48209 -> 47460 (-1.55%)
instructions in affected programs: 11688 -> 10939 (-6.41%)
Signed-off-by: Boyan Ding <boyan.j.ding@gmail.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
The idea here is not that it gives register coalescing a little bit of a
helping hand. It doesn't actually fix the coalescing problems, but it
seems to help a good bit.
Shader-db results for vec4 programs on Haswell:
total instructions in shared programs: 1746280 -> 1683959 (-3.57%)
instructions in affected programs: 1259166 -> 1196845 (-4.95%)
helped: 11363
HURT: 148
v2 (Jason Ekstrand):
- Run nir_move_vec_src_uses_to_dest after going out of SSA
- New shader-db numbers
Reviewed-by: Eduardo Lima Mitev <elima@igalia.com>
The provided indices have the very nice property that if A dominates B then
A->index <= B->index. We should document that somewhere.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Various pieces of code to create compressed textures will first
generate an uncompressed RGBA texture into a temporary buffer,
and then read from that buffer while creating the final compressed
texture in the requested format.
The code reading from the temporary buffer assumes the buffer is
formatted as an array of bytes in RGBA order. However, the buffer
is filled using a _mesa_texstore call with MESA_FORMAT_R8G8B8A8_UNORM
format -- this is defined as an array of *integers* holding the
RGBA values in packed format (least-significant to most-significant).
This means incorrect bytes are accessed on big-endian systems.
This patch fixes this by using the MESA_FORMAT_A8B8G8R8_UNORM format
instead on big-endian systems when filling the buffer. This fixes
about 100 piglit test case failures on s390x for me.
Signed-off-by: Ulrich Weigand <ulrich.weigand@de.ibm.com>
Tested-by: Oded Gabbay <oded.gabbay@gmail.com>
Cc: "10.6" "11.0" <mesa-stable@lists.freedesktop.org>
Signed-off-by: Dave Airlie <airlied@gmail.com>
XA has been using L8_UNORM for a8 and yuv component surfaces.
This commit instead makes XA prefer R8_UNORM since it's assumed to have a
higher availability.
Also neither of these formats are suitable as destination formats using
destination alpha blending, so reject those operations.
Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
From OpenGL 4.5 Core spec (7.13):
"If pipeline is a name that has been generated (without subsequent
deletion) by GenProgramPipelines, but refers to a program pipeline
object that has not been previously bound, the GL first creates a
new state vector in the same manner as when BindProgramPipeline
creates a new program pipeline object."
I interpret this as "If GetProgramPipelineiv gets called without a
bound (but valid) pipeline object, the state should reflect initial
state of a new pipeline object." This is also expected behaviour by
ES31-CTS.sepshaderobjs.PipelineApi conformance test.
Signed-off-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Marta Lofstedt <marta.lofstedt@intel.com>
From OpenGL ES 3.1 spec (7.12):
"Most properties set within program objects are specified not to
take effect until the next call to LinkProgram or ProgramBinary.
Some properties further require a successful call to either of
these commands before taking effect. GetProgramiv returns the
properties currently in effect for program, which may differ from
the properties set within program since the most recent call to
LinkProgram or ProgramBinary, which have not yet taken effect. If
there has been no such call putting changes to pname into effect,
initial values are returned."
Signed-off-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Marta Lofstedt <marta.lofstedt@intel.com>
As a bonus we get indirect support for arrays of arrays for free.
V5: couple of small clean-ups suggested by Jason.
V4: fix struct member location caclulation, use nir_ssa_def rather than
nir_src for the indirect as suggested by Jason
V3: Use nir_instr_rewrite_src() with empty src rather then clearing
the use_link list directly for the old indirects as suggested by Jason
V2: Fixed validation error in debug build
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
This will allow us to access the uniform later on without resorting to
building a name string and looking it up in UniformHash.
V3: remove line wrap change from this patch
V2: store slot number for all non-UBO uniforms to make code more
consitent, renamed explicit_binding to explicit_location and added
comment about what it does. Store the location at every shader stage.
Updated data.location comments in ir/nir.h.
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
This is required so that the next patch can safely assign the slot id
to the var.
The ids are now assigned in the order we want before allocating storage
so there is no need to sort the storage array and move things around.
V2: rename variable to make code easier to follow as suggested by Jason
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
This allows the correct offset to be easily calculated for indirect
indexing when a struct array contains multiple samplers, or any crazy
nesting.
The indices for the folling struct will now look like this:
Sampler index: 0 Name: s[0].tex
Sampler index: 1 Name: s[1].tex
Sampler index: 2 Name: s[0].si.tex
Sampler index: 3 Name: s[1].si.tex
Sampler index: 4 Name: s[0].si.tex2
Sampler index: 5 Name: s[1].si.tex2
Before this change it looked like this:
Sampler index: 0 Name: s[0].tex
Sampler index: 3 Name: s[1].tex
Sampler index: 1 Name: s[0].si.tex
Sampler index: 4 Name: s[1].si.tex
Sampler index: 2 Name: s[0].si.tex2
Sampler index: 5 Name: s[1].si.tex2
struct S_inner {
sampler2D tex;
sampler2D tex2;
};
struct S {
sampler2D tex;
S_inner si;
};
uniform S s[2];
V3: Update comments with suggestions from Jason
V2: rename struct array counter to have better name
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
This reverts commit 48961fa3ba.
glamor/Xwayland use this, the spec saying something when it
was written, and the fact that the comment says Mesa relies on it
hasn't changed.
I also don't have a copy of this patch in my mail archive, which
seems wierd, did it get posted to mesa-dev?
Signed-off-by: Dave Airlie <airlied@redhat.com>
Even though luminance formats don't have alpha, we still want the alpha
output to go to the blender. This fixes the luminance blending tests.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Cc: "11.0" <mesa-stable@lists.freedesktop.org>
(originally part of previous patch, split out to separate patch by Rob)
v2: squash in some fixes from Eric
v3: Another fix from Eric for point coords.
Signed-off-by: Rob Clark <robclark@freedesktop.org>
This avoids exceeding the size of the .index bitfield since it got
truncated, and should make our NIR look more like the NIR that the rest of
the NIR developers are working on.
v2: split out vc4 updates, first patch uses varying_slot_to_tgsi_semantic()
helper, and second patch does the actual conversion.
v3: add frag_result_to_tgsi_semantic() helper and don't try to map
frag_results to semantic name/index as if they were varying_slot's
v4: use VERT_ATTRIB_ for VS inputs
v5: Fix vc4 build.
Signed-off-by: Rob Clark <robclark@freedesktop.org>
v2: split out moving of FILE *fp into state structure into it's own
(more complete patch) to reduce the noise in this one
Signed-off-by: Rob Clark <robclark@freedesktop.org>
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Rename print_var_state to print_state, and stuff FILE ptr into the state
object. This avoids passing around an extra parameter everywhere.
v2: even more extensive conversion.. use state *everywhere* instead of
FILE ptr, and convert nir_print_instr() to use state as well
Signed-off-by: Rob Clark <robclark@freedesktop.org>
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Similar to fee0686c21, but in this case to
ensure that drm_gralloc and libGLES_mesa are sharing a single screen.
Bumps libdrm_freedreno version dependency, as it requires the new
fd_device_fd() API.
Signed-off-by: Rob Clark <robclark@freedesktop.org>
When preparing the barrier payload, the instructions should operate in
simd8 mode since we only use 1 payload register.
fs_inst::regs_read is also updated to indicate that it only reads one
register for SHADER_OPCODE_BARRIER.
These issues were flagged by:
commit cadd7dd384
Author: Jason Ekstrand <jason.ekstrand@intel.com>
Date: Thu Jul 2 15:41:02 2015 -0700
i965/fs: Add a very basic validation pass
Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
C++ gets cranky if we take references of temporaries. This isn't a problem
yet in master because nir_builder is never used from C++. However, it will
be in the future so we should fix it now.
Reviewed-by: Rob Clark <robclark@freedesktop.org>
Both use the same layout for the buffer containing border-color values,
so rather than duplicating the logic in a4xx, split it out into a
helper.
Signed-off-by: Rob Clark <robclark@freedesktop.org>
Now that we have a replicating fdot instruction, we can actually coalesce
into the destinations of vec4 instructions. We couldn't really do this
before because, if the destination had to end up in .z, we couldn't
reswizzle the instruction. With a replicated destination, the result ends
up in all channels so we can just set the writemask and we're done.
Shader-db results for vec4 programs on Haswell:
total instructions in shared programs: 1747753 -> 1746280 (-0.08%)
instructions in affected programs: 143274 -> 141801 (-1.03%)
helped: 667
HURT: 0
It turns out that dot-products matter...
Reviewed-by: Eduardo Lima Mitev <elima@igalia.com>
Fortunately, nir_constant_expr already auto-splats if "dst" never shows up
in the constant expression field so we don't need to do anything there.
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Reviewed-by: Eduardo Lima Mitev <elima@igalia.com>
The old pass blindly inserted a bunch of moves into the shader with no
concern for whether or not it was really needed. This adds code to try and
coalesce into the destination of the instruction providing the value.
Shader-db results for vec4 shaders on Haswell:
total instructions in shared programs: 1754420 -> 1747753 (-0.38%)
instructions in affected programs: 231230 -> 224563 (-2.88%)
helped: 1017
HURT: 2
This approach is heavily based on a different patch by Eduardo Lima Mitev
<elima@igalia.com>. Eduardo's patch did this in a separate pass as opposed
to integrating it into nir_lower_vec_to_movs.
Reviewed-by: Eduardo Lima Mitev <elima@igalia.com>
Previously, we did this thing with keeping track of a separate start_idx
which was different from the iteration variable. I think this was a relic
of the way that GLSL IR implements writemasks. In NIR, if a given bit in
the writemask is unset then that channel is just "unused", not missing. In
particular, a vec4 operation with a writemask of 0xd will use sources 0, 2,
and 3 and leave source 1 alone. We can simplify things a good deal (and
make them correct) by removing this "compacting" step.
Reviewed-by: Eduardo Lima Mitev <elima@igalia.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
We made this switch in the FS backend some time ago and it seems to make a
number of things a bit easier. In particular, supporting SSA values takes
very little work in the backend and allows us to take advantage of the
majority of the SSA information even after we've gotten rid of Phi nodes.
Reviewed-by: Eduardo Lima Mitev <elima@igalia.com>
v2 (Jason Ekstrand):
- Use nir_instr_rewrite_dest
- Pass the impl directly into lower_vec_to_movs_block
Reviewed-by: Eduardo Lima Mitev <elima@igalia.com>
Previously, we were passing the shader around, we were just calling it
"mem_ctx". However, the nir_shader is (and must be for the purposes of
mark-and-sweep) the mem_ctx so we might as well pass it around explicitly.
Reviewed-by: Eduardo Lima Mitev <elima@igalia.com>
Currently the validation pass only validates that regs_read and
regs_written are consistent with the sizes of VGRF's. We can add more as
we find it to be useful.
Reviewed-by: Matt Turner <mattst88@gmail.com>
In certain conditions, we have to do bounds-checking in the shader for
image_load_store. The way this works for image loads is that we do a
predicated load and then emit a series of selects, one per component,
that gives us 0 or the loaded value depending on whether or not you're
in bounds. However, we were hard-coding 4 components which may not be
correct. Instead, we should be using size which is the number of
components read.
Reviewed-by: Francisco Jerez <currojerez@riseup.net>
According to the extensions table and our glext headers,
OES_compressed_ETC1_RGB8_texture is only supported in
GLES1 and GLES2. Since we may give users a GLES3 context
when a GLES2 context is requested, we also allow this
extension for GLES3 as well.
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
Signed-off-by: Nanley Chery <nanley.g.chery@intel.com>
Driver vendors do this as well. The extension specification
lists GLES 1.1 or 2.0 as requirements.
Reviewed-by: Chad Versace <chad.versace@intel.com>
Signed-off-by: Nanley Chery <nanley.g.chery@intel.com>
It's extensively used by XA for a8- and planar yuv component surfaces.
This fixes broken XA yuv blits using vgpu10 contexts.
Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Currently the check was incorrect as it did not consider the (unlikely)
case of fd == 0. In order to fix this we should first correctly
initialize it to -1, as the swrast implementations leave it set to zero
(props to calloc()).
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Reviewed-by: Boyan Ding <boyan.j.ding@gmail.com>
Move the fcntl(dupfd_cloexec) to the else branch where it belongs.
Otherwise it's not immediately obvious that the code is hit, only when
an existing device is used.
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Reviewed-by: Boyan Ding <boyan.j.ding@gmail.com>
v2: [Emil Velikov]
Rework the error path to a common goto, close only if we own the fd.
v3; [Emil Velikov]
Always close the fd (we either opened the device or dup'd) (Boyan, Ian)
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Reviewed-by: Boyan Ding <boyan.j.ding@gmail.com>
At the moment if a gbm buffer is imported and the gbm buffer
has an old-style GBM_BO_FORMAT format, the import will crash,
since it's passed directly to DRI functions that expect
a fourcc format (as provided by the newer GBM_FORMAT
definitions)
This commit addresses the problem in two ways:
1) it prevents invalid formats from leading to a crash by
returning EINVAL if the image couldn't be created
2) it translates GBM_BO_FORMAT formats into the comparable
GBM_FORMAT formats.
Reference: https://bugzilla.gnome.org/show_bug.cgi?id=753531
CC: "10.6 11.0" <mesa-stable@lists.freedesktop.org>
Reviewed-by: Emil Velikov <emil.l.velikov@gmail.com>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
We're trying to avoid a libdrm dependency in the core compiler, so let's
move the perf_debug code one level up from the brw_*_emit() helpers to
the brw_codegen_*_prog() helpers.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Signed-off-by: Kristian Høgsberg Kristensen <krh@bitplanet.net>
All other precompile functions live in the brw_<stage>.c files, make fs
follow the convention.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Signed-off-by: Kristian Høgsberg Kristensen <krh@bitplanet.net>
This moves the compute shader code around in order to make the way the
code is split up more consistent. There should be no functional changes.
Typically we have a few files per stage:
brw_vs.c, brw_wm.c brw_gs.c:
code to drive code generation and implement precompiling and
cache search.
genX_<stage>_state.c
gen specific implementation of the state emission for the shader
stage.
The brw_*_emit() functions are all in the same files as the visitor
classes they use (with the exception of VS, which may use either vec4 or
fs).
To make compute follow this convention, we move the brw_cs_emit()
function into brw_fs.cpp. We can then rename brw_cs.cpp to brw_cs.c and
do this in C like the other similar files. Finally, move state setup
and atoms to gen7_cs_state.c.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Signed-off-by: Kristian Høgsberg Kristensen <krh@bitplanet.net>
Curiously this has no actual effect. I think it's because the first 8
textures are bound in multiple slots for some reason. However seems
prudent to use these the same way as regular texturing, esp in the case
where there are more than 8 textures bound.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
If the register types do not match and the instruction
that contains the final destination is saturated, register
coalescing generated non-equivalent code.
This did not happen when using IR because types usually
matched, but it is visible in nir-vec4.
For example,
mov vgrf7:D vgrf2:D
mov.sat m4:F vgrf7:F
is coalesced to:
mov.sat m4:D vgrf2:D
The patch prevents coalescing in such scenario, unless the
instruction we want to coalesce into is a MOV (without type
conversion implied). In that case, the patch sets the register
types to the type of the final destination.
Shader-db results in HSW (only vec4 instructions shown):
total instructions in shared programs: 1754415 -> 1754416 (0.00%)
instructions in affected programs: 74 -> 75 (1.35%)
helped: 0
HURT: 1
GAINED: 0
LOST: 0
Only one extra instruction in one of the shaders, that comes from
eliminating a saturation error by preventing register coalesce.
Cc: "10.6 11.0" <mesa-stable@lists.freedesktop.org>
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
We lower gl_LocalInvocationIndex based on the extension spec formula:
gl_LocalInvocationIndex =
gl_LocalInvocationID.z * gl_WorkGroupSize.x * gl_WorkGroupSize.y +
gl_LocalInvocationID.y * gl_WorkGroupSize.x +
gl_LocalInvocationID.x;
https://www.opengl.org/registry/specs/ARB/compute_shader.txt
We need to set this variable in main(), even if gl_LocalInvocationIndex
is not referenced by the shader. (It may be used by a linked shader.)
Therefore, we can't eliminate it as a dead variable.
Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
Also rename to _mesa_get_main_function_signature.
We will call it near the end of compilation to insert some code into
main for initializing some compute shader global variables.
Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Eduardo Lima Mitev <elima@igalia.com>
We lower gl_GlobalInvocationID based on the extension spec formula:
gl_GlobalInvocationID =
gl_WorkGroupID * gl_WorkGroupSize + gl_LocalInvocationID
https://www.opengl.org/registry/specs/ARB/compute_shader.txt
We need to set this variable in main(), even if gl_GlobalInvocationID
is not referenced by the shader. (It may be used by a linked shader.)
Therefore, we can't eliminate these as dead variables.
Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
This is to avoid needless float<->int conversions, since all
face-related computations are made on integers. Spotted by Emil
Velikov.
Reviewed-by: Brian Paul <brianp@vmware.com>
New enum to add to switch so compiler doesn't complain.
commit 1807a08e4f
Author: Ilia Mirkin <imirkin@alum.mit.edu>
AuthorDate: Thu Aug 27 23:05:03 2015 -0400
Commit: Ilia Mirkin <imirkin@alum.mit.edu>
CommitDate: Thu Sep 10 17:38:33 2015 -0400
nir: add nir_texop_texture_samples and convert from glsl
Signed-off-by: Rob Clark <robclark@freedesktop.org>
Sometimes a useful thing for compilers (or, for example, tgsi_to_nir) to
know. And pretty trivial for scan to figure this out for us.
Signed-off-by: Rob Clark <robclark@freedesktop.org>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
Cypress/Cayman/Aruba, earlier r6xx/r7xx chips only support a subset
of the needed fp64 ops, and don't do GL4 anyway.
Signed-off-by: Glenn Kennard <glenn.kennard@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
Only for Cypress/Cayman/Aruba, older chips have only partial fp64 support.
Uses float intermediate values so only accurate for int24 range, which
matches what the blob does.
Signed-off-by: Glenn Kennard <glenn.kennard@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
I'm going to want a driver constant buffer for tess to coordinate
LDS storage, so before I go tackling that I decided to merge the
clip/samplepos and texture info buffers into one. So I can steal
the spare one.
This creates a single constant buffer between the two, with
clip/samplepos taking up a reserved 128 bytes at the start.
Reviewed-by: Edward O'Callaghan <eocallaghan@alterapraxis.com>
Reviewed-by: Glenn Kennard <glenn.kennard@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
V2: -Change to "not started" for most entries
-Add status for multisample_2d_array
-Change shader_multisample_interpolation to "not_stared"
V3 (idr): Move the GLES 3.2 section after the "Additional functions"
section from GLES 3.1. Note that GL_KHR_texture_compression_astc_hdr is
done for i965 on gen9+ hardware. Note that GL_OES_shader_io_blocks is
based on some features from GLSL 1.50.
Signed-off-by: Thomas Helland <thomashelland90@gmail.com>
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Tapani Pälli <tapani.palli@intel.com> [v2]
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
This commit makes a lot of variables constant - this is basically done
by moving the computation to variable definition. Some of them are
moved into lower scopes (like in img_filter_2d_ewa).
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Add a small inline function doing the casting - this is to make sure
we don't do a cast from some completely unrelated type. This commit
does not make tgsi_sampler parameters const in vfuncs themselves for
now - probably llvmpipe would need looking at before making such a
change.
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Those functions actually could always take them as constants.
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Those functions actually could always take them as constants.
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
A followup from previous commit - since all functions called by
query_lod take pointers to const sp_sampler_view and const sp_sampler,
which are taken from tgsi_sampler subclass, we can the tgsi_sampler as
const itself now.
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
This is to prepare for making tgsi_sampler parameter in query_lod a
const too. These functions do not modify anything in either sampler or
view anymore.
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
With that, sp_sampler_view instances are not abused anymore as a local
storage, so we can later make them constant.
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
As of a10d4937, we would really like things associated with an instruction
to be allocated out of that instruction and not out of the shader. In
particular, you should be passing the instruction that will ultimately be
holding the source into nir_src_copy rather than an arbitrary memory
context.
We also change the prototypes of nir_dest_copy and nir_alu_src/dest_copy to
explicitly take an instruction so we catch this earlier in the future.
Cc: "11.0" <mesa-stable@lists.freedesktop.org>
Reviewed-by: Thomas Helland <thomashelland90@gmail.com>
We copy the output, make the old output the temporary, and give the
temporary a new name. The copy keeps the pointer to the old name. This
works just fine up until the point where we lower things to SSA and delete
the old variable and, with it, the name. Instead, we should re-parent to
the copy.
Reviewed-by: Eduardo Lima Mitev <elima@igalia.com>
opt_register_coalesce stopped to check previous instructions to
coalesce with if somebody else was writing on the same
destination. This can be optimized to check if somebody else was
writing to the same channels of the same destination using the
writemask.
Shader DB results (taking into account only vec4):
total instructions in shared programs: 1781593 -> 1734957 (-2.62%)
instructions in affected programs: 1238390 -> 1191754 (-3.77%)
helped: 12782
HURT: 0
GAINED: 0
LOST: 0
v2: removed some parenthesis, fixed indentation, as suggested by
Matt Turner
v3: added brackets, for consistency, as suggested by Eduardo Lima
Reviewed-by: Matt Turner <mattst88@gmail.com>
This makes it possible for NIR shaders to know the number of output
vertices and the number of invocations. Drivers could also access
these directly without going through gl_program.
We should probably add InputType and OutputType here too, but currently
those are stored as GL_* enums, and I wanted to avoid using those in
NIR, as I suspect Vulkan/SPIR-V will use different enums. (We should
probably make our own.)
We could add VerticesIn, but it's easily computable from the input
topology, so I'm not sure whether it's worth it. It's also currently
not stored in gl_shader (only gl_shader_program), which would require
changes to the glsl_to_nir interface or require us to store it there.
This is a bit of duplication of data...ideally, we would factor these
substructs out of gl_program, gl_shader_program, and nir_shader, creating
a gl_geometry_info class...but it would need to go in a new place (in
src/glsl?) that isn't mtypes.h nor nir.h.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
These provide a convenient way to do simple variable loads and stores.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
Previously the result of the complicated clamp() expression just dropped
on the floor: clamp does not modify any of its parameters. Looking at
the surrounding code, I believe this is supposed to modify the value of
tex_coord.
This change (along with a change to avoid the use of
brw_blorp_framebuffer) does not affect any existing piglit tests. I'm
not sure what this clamp is trying to accomplish, so I'm not sure how to
write a test to exercise this path.
I also noticed another bug in this code. There is no way the array
texture case could possibly work. This will generate code for the
TEXEL_FETCH macro like:
#define TEXEL_FETCH(coord) texelFetch(texSampler, ivec3(coord), sample_map[int(2 * fract(coord.x))]);
Since the coord parameter of this macro is a vec2 at all invocations, no
expansion of this macro will even compile.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
Cc: Topi Pohjolainen <topi.pohjolainen@intel.com>
Cc: Jordan Justen <jordan.l.justen@intel.com>
These only occurred in release builds, but they occurred in every file
that included intel_batchbuffer.h. Lots of spam. :(
intel_batchbuffer.h: In function 'intel_batchbuffer_advance':
intel_batchbuffer.h:153:47: warning: unused parameter 'brw' [-Wunused-parameter]
intel_batchbuffer_advance(struct brw_context *brw)
^
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
The for_bo parameter of intel_miptree_create_layout appears to be unused
since 27eedca when Eric removed some Gen5 code (after the i915 and i965
drivers parted ways).
intel_mipmap_tree.c: In function 'old_intel_miptree_create_layout':
intel_mipmap_tree.c:77:35: warning: unused parameter 'for_bo' [-Wunused-parameter]
bool for_bo)
^
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
Hasn't existed in the i915 source since the i915 and i965 drivers parted
ways.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
This hasn't been used outside intel_mipmap_tree.c since d5d4ba9 started
using meta instead of the blitter for PBO TexSubImage. While we're
here, remove the unused brw parameter from the function formerly known
as intel_miptree_unmap_raw.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
These only occurred in release builds, but they occurred in every file
that included intel_mipmap_tree.h. Lots of spam. :(
intel_mipmap_tree.h: In function 'intel_miptree_check_level_layer':
intel_mipmap_tree.h:595:59: warning: unused parameter 'mt' [-Wunused-parameter]
intel_miptree_check_level_layer(struct intel_mipmap_tree *mt,
^
intel_mipmap_tree.h:596:42: warning: unused parameter 'level' [-Wunused-parameter]
uint32_t level,
^
intel_mipmap_tree.h:597:42: warning: unused parameter 'layer' [-Wunused-parameter]
uint32_t layer)
^
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
The target parameter of compute_msaa_layout appears to be unused since
83b83fb when support for CMS textures was added for Gen7.
The brw parameter of intel_get_non_msrt_mcs_alignment appears to be
unused since e92fbdc when the GEN check (along with the "can we fast
clear" decision) was moved to a different function.
intel_mipmap_tree.c: In function 'compute_msaa_layout':
intel_mipmap_tree.c:62:73: warning: unused parameter 'target' [-Wunused-parameter]
compute_msaa_layout(struct brw_context *brw, mesa_format format, GLenum target,
^
intel_mipmap_tree.c: In function 'intel_get_non_msrt_mcs_alignment':
intel_mipmap_tree.c:143:54: warning: unused parameter 'brw' [-Wunused-parameter]
intel_get_non_msrt_mcs_alignment(struct brw_context *brw,
^
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
Cc: Ben Widawsky <benjamin.widawsky@intel.com>
If we skip a vbuffer we need to make sure we NULL out
the contents, otherwise when it gets passed to the driver
it will get confused.
This was hit by:
GL41-CTS.gpu_shader_fp64.varyings
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
Enable barrier in MEDIA_INTERFACE_DESCRIPTOR if the program uses the
barrier() GLSL function.
On Ivy Bridge and Haswell, this allows the piglit test
tests/spec/arb_compute_shader/execution/simple-barrier-atomics.shader_test
to pass. On gen8, this enables a similar test with a local group size
of 896 to pass.
Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
first_non_payload_grf may be updated in assign_urb_setup for FS or
assign_vs_urb_setup for VS.
We need to set this in assign_curb_setup for compute shaders since cs
does not have an assign_cs_urb_setup like assign_urb_setup (fs) or
assign_vs_urb_setup (vs).
Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
[v2: kayden-supplied code in fs_nir replacing need for logical opcode]
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
grep -lr 'sub license' | while read f; do \
sed --in-place -e 's/sub license/sublicense/' $f ;\
done
grep -lr 'NON-INFRINGEMENT' | while read f; do \
sed --in-place -e 's/NON-INFRINGEMENT/NONINFRINGEMENT/' $f ;\
done
As noted by Matt, both of these changes match the MIT license text found
at http://opensource.org/licenses/MIT.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Acked-by: Matt Turner <mattst88@gmail.com>
Passes the shader piglit tests and introduces no regressions.
This commit finally makes use of the refactoring in previous
commits.
v2:
- adapted the code to changes in previous commits (renames,
need_cube_convert stuff)
- splitted too long lines
Reviewed-by: Brian Paul <brianp@vmware.com>
This introduces new vfunc in tgsi_sampler just for this opcode. I
decided against extending get_samples vfunc to return the mipmap level
and LOD - the function's prototype is already too scary and doing the
sampling for textureQueryLod would be a waste of time.
v2:
- splitted too long lines
Reviewed-by: Brian Paul <brianp@vmware.com>
These functions will be used by textureQueryLod.
v2:
- renamed mip_level_* funcs to mip_rel_level_* to indicate that
these functions return mip level relative to base level and
documented them
- renamed a level member in sp_filter_funcs struct to relative_level
- changed mip_rel_level_none and mip_rel_level_nearest to return mip
level relative to base level, mip_rel_level_linear already did
that
- documented clamp_lod function
Reviewed-by: Brian Paul <brianp@vmware.com>
This is to avoid tying the conversion to the sampling -
textureQueryLod will need to do the conversion too, but it does not do
any sampling.
So instead of a "get_samples" vfunc, there is just a bool saying
whether the conversion is needed or not. This solution keeps a nice
property of not adding any overhead for the common case (2D textures).
v2:
- replaced the "convert_coords" vfunc with a "need_cube_convert"
boolean to avoid overhead of copying arrays in common case
- removed an unused typedef
- splitted too long lines in convert_cube
- const fixes in convert_cube
Reviewed-by: Brian Paul <brianp@vmware.com>
This function will be later used by textureQueryLod. The
img_filter_func are optional, because textureQueryLod will not need
them.
v2:
- adapted to changes in previous commit (renames)
- simplified conditions a bit
- updated docs
- splitted too long lines
Reviewed-by: Brian Paul <brianp@vmware.com>
Putting this function pointer into a struct enables grouping of
several related functions in a single place. For now it is just a
single function, but the struct will be later extended with a
mip_level_func for returning relative mip level.
v2:
- renamed sp_mip struct to sp_filter_funcs
- renamed sp_filter_funcs instances from mip_foo to funcs_foo
- splitted too long lines
- sp_sampler now holds a pointer to sp_filter_funcs instead of an
instance of it
- some const fixes
Reviewed-by: Brian Paul <brianp@vmware.com>
textureQueryLod returns a vec2 with a mipmap information and a
LOD. The latter needs to be not clamped.
v2:
- changed the "not_clamped" part to "unclamped"
- corrected "clamp into" to "clamp to"
- splitted too long lines
Reviewed-by: Brian Paul <brianp@vmware.com>
The level-of-detail bias wasn't simply added in the explicit LOD case.
This case seems to be tested only in piglit's
fs-texturequerylod-nearest-biased test, which is currently skipped, as
softpipe does not support textureQueryLod at the moment.
Reviewed-by: Brian Paul <brianp@vmware.com>
Basically, do the same thing as for buffer_unmap, but use the explicit range
instead. It's for apps which want to map a whole buffer and mark touched
ranges explicitly.
Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>
When checking for LLVM shared libraries, use IMP_LIB_EXT for the extension for
shared libraries appropriate to the target, rather than hardcoding '.so'
Also add some comments to explain why we have this circus of pain.
Signed-off-by: Jon TURNEY <jon.turney@dronecode.org.uk>
Reviewed-by: Tom Stellard <thomas.stellard@amd.com>
Reviewed-by: Emil Velikov <emil.l.velikov@gmail.com>
mesa/src/mesa/drivers/dri/i965/brw_eu_compact.c: In function 'set_3src_control_index':
mesa/src/mesa/drivers/dri/i965/brw_eu_compact.c:805:22: warning: comparison between signed and unsigned integer expressions [-Wsign-compare]
for (int i = 0; i < ARRAY_SIZE(gen8_3src_control_index_table); i++) {
^
mesa/src/mesa/drivers/dri/i965/brw_eu_compact.c: In function 'set_3src_source_index':
mesa/src/mesa/drivers/dri/i965/brw_eu_compact.c:839:22: warning: comparison between signed and unsigned integer expressions [-Wsign-compare]
for (int i = 0; i < ARRAY_SIZE(gen8_3src_source_index_table); i++) {
^
mesa/src/mesa/drivers/dri/i965/brw_state_dump.c: In function 'dump_sampler_state':
mesa/src/mesa/drivers/dri/i965/brw_state_dump.c:382:18: warning: comparison between signed and unsigned integer expressions [-Wsign-compare]
for (i = 0; i < size / 16; i++) {
^
mesa/src/mesa/drivers/dri/i965/brw_state_upload.c: In function 'brw_pipeline_state_finished':
mesa/src/mesa/drivers/dri/i965/brw_state_upload.c:801:13: warning: comparison between signed and unsigned integer expressions [-Wsign-compare]
if (i != pipeline) {
^
mesa/src/mesa/drivers/dri/i965/intel_mipmap_tree.c: In function 'intel_gen7_hiz_buf_create':
mesa/src/mesa/drivers/dri/i965/intel_mipmap_tree.c:1544:47: warning: comparison between signed and unsigned integer expressions [-Wsign-compare]
for (int level = mt->first_level; level <= mt->last_level; ++level) {
^
mesa/src/mesa/drivers/dri/i965/intel_mipmap_tree.c: In function 'intel_gen8_hiz_buf_create':
mesa/src/mesa/drivers/dri/i965/intel_mipmap_tree.c:1638:44: warning: comparison between signed and unsigned integer expressions [-Wsign-compare]
for (int level = mt->first_level; level <= mt->last_level; ++level) {
^
mesa/src/mesa/drivers/dri/i965/intel_mipmap_tree.c: In function 'intel_miptree_alloc_hiz':
mesa/src/mesa/drivers/dri/i965/intel_mipmap_tree.c:1771:44: warning: comparison between signed and unsigned integer expressions [-Wsign-compare]
for (int level = mt->first_level; level <= mt->last_level; ++level) {
^
mesa/src/mesa/drivers/dri/i965/intel_mipmap_tree.c:1775:33: warning: comparison between signed and unsigned integer expressions [-Wsign-compare]
for (int layer = 0; layer < mt->level[level].depth; ++layer) {
^
Signed-off-by: Rhys Kidd <rhyskidd@gmail.com>
Reviewed-by: Thomas Helland <thomashelland90@gmail.com>
Reviewed-by: Emil Velikov <emil.l.velikov@gmail.com>
mesa/src/mesa/program/prog_to_nir.c: In function 'setup_registers_and_variables':
/mesa/src/mesa/program/prog_to_nir.c:1059:22: warning: comparison between signed and unsigned integer expressions [-Wsign-compare]
for (int i = 0; i < c->prog->NumTemporaries; i++) {
^
Signed-off-by: Rhys Kidd <rhyskidd@gmail.com>
Reviewed-by: Thomas Helland <thomashelland90@gmail.com>
Reviewed-by: Jan Vesely <jan.vesely@rutgers.edu>
Reviewed-by: Emil Velikov <emil.l.velikov@gmail.com>
mesa/src/glsl/nir/nir_lower_tex_projector.c: In function 'nir_lower_tex_projector_block':
mesa/src/glsl/nir/nir_lower_tex_projector.c:63:25: warning: comparison between signed and unsigned integer expressions [-Wsign-compare]
for (int i = 0; i < tex->num_srcs; i++) {
^
mesa/src/glsl/nir/nir_lower_tex_projector.c: In function 'nir_lower_tex_projector_block':
mesa/src/glsl/nir/nir_lower_tex_projector.c:114:38: warning: comparison between signed and unsigned integer expressions [-Wsign-compare]
for (int i = proj_index + 1; i < tex->num_srcs; i++) {
^
mesa/src/glsl/nir/nir_lower_tex_projector.c: In function 'nir_lower_tex_projector_block':
mesa/src/glsl/nir/nir_lower_tex_projector.c:53:39: warning: comparison between signed and unsigned integer expressions [-Wsign-compare]
for (proj_index = 0; proj_index < tex->num_srcs; proj_index++) {
^
mesa/src/glsl/nir/nir_lower_tex_projector.c:57:22: warning: comparison between signed and unsigned integer expressions [-Wsign-compare]
if (proj_index == tex->num_srcs)
^
mesa/src/glsl/nir/nir_search.c: In function 'match_value':
mesa/src/glsl/nir/nir_search.c:84:22: warning: comparison between signed and unsigned integer expressions [-Wsign-compare]
for (int i = 0; i < num_components; ++i)
^
mesa/src/glsl/nir/nir_search.c: In function 'match_value':
mesa/src/glsl/nir/nir_search.c:110:28: warning: comparison between signed and unsigned integer expressions [-Wsign-compare]
for (int i = 0; i < num_components; ++i) {
^
mesa/src/glsl/nir/nir_search.c: In function 'match_value':
mesa/src/glsl/nir/nir_search.c:139:19: warning: comparison between signed and unsigned integer expressions [-Wsign-compare]
if (i < num_components)
^
mesa/src/glsl/nir/nir_opt_peephole_ffma.c: In function 'get_mul_for_src':
mesa/src/glsl/nir/nir_opt_peephole_ffma.c:130:27: warning: comparison between signed and unsigned integer expressions [-Wsign-compare]
for (unsigned i = 0; i < num_components; i++)
^
Signed-off-by: Rhys Kidd <rhyskidd@gmail.com>
Reviewed-by: Thomas Helland <thomashelland90@gmail.com>
Reviewed-by: Jan Vesely <jan.vesely@rutgers.edu>
Reviewed-by: Emil Velikov <emil.l.velikov@gmail.com>
Resolve a series of missing field initializer warnings within get_hash_params.py
Of the form:
In file included from mesa/src/mesa/main/get.c:495:0:
mesa/src/mesa/main/get_hash.h:180:5: warning: missing initializer for field
'extra' of 'const struct value_desc' [-Wmissing-field-initializers]
{ GL_POINT_SIZE_ARRAY_BUFFER_BINDING_OES, LOC_CUSTOM, TYPE_INT, 0 },
^
mesa/src/mesa/main/get.c:165:15: note: 'extra' declared here
const int *extra;
^
This patch addresses some likely code rot around the *extra field, where the
initialization is via C code generated indirectly from a Python script.
It resolves a number of warnings reported by GCC when configured to be pedantic.
$ gcc --version
gcc (Ubuntu 4.9.2-10ubuntu13) 4.9.2
No piglit regressions on Ironlake.
v2:
- Squash series into a single patch.
Signed-off-by: Rhys Kidd <rhyskidd@gmail.com>
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Reviewed-by: Jan Vesely <jan.vesely@rutgers.edu>
Reviewed-by: Thomas Helland <thomashelland90@gmail.com>
Reviewed-by: Emil Velikov <emil.l.velikov@gmail.com>
When parsing an variable declaration qualified with the typename
keyword, clang attempted to declare a variable with the type of non
type member "enum type type" of module::argument (within the header
file clover/core/module.hpp) instead of the typed member of
module::argument "enum type".
Replaced "typename" with "enum" to force clang to declare the variable
marg_type with type "enum type" of module::argument.
CC: "11.0" <mesa-stable@lists.freedesktop.org>
Reviewed-by: Francisco Jerez <currojerez@riseup.net>
Reviewed-by: Emil Velikov <emil.l.velikov@gmail.com>
Signed-off-by: Albert Freeman <albertwdfreeman@gmail.com>
Our old value of 16384 is the minimum value. DirectX apparently
requires 65536 at a minimum; that's also what nVidia and the Intel
Windows driver advertise. AMD advertises MAX_INT.
Ilia Mirkin noticed that "Shadow Warrior" uses UBOs larger than 16k
on Nouveau, which advertises 65536 bytes for this limit. Traces
captured on Nouveau don't work on i965 because our lower limit causes
the GLSL linker to reject the captured shaders. While this isn't
important in and of itself, it does suggest that raising the limit
would be beneficial.
We can read linear buffers up to 2^27 bytes in size, so raising this
should be safe; we could probably even go larger. For now, matching
nVidia and Intel/Windows seems like a good plan.
We have to reinitialize MaxCombinedUniformComponents as core Mesa will
have set it based on a stale value for MaxUniformBlockSize.
According to Tapani, there's an unreleased game that asserts on this.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
Cc: "11.0" <mesa-stable@lists.freedesktop.org>
It is advantageous to use r63 instead of r127 since r63 can fit into the
shorter encoding. However if we've RA'd over 63 registers, we must use
r127 as the replacement instead.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Cc: "11.0" <mesa-stable@lists.freedesktop.org>
Unfortunately nv50_ir phi nodes aren't directly connected to the CFG, so
the mapping between source and the actual BB is by inbound edge order.
So when manipulating edges one has to be extremely careful. We were
insufficiently careful when splitting critical edges which resulted in
the phi nodes being confused as to where their sources were coming from.
This primarily manifests itself with the TXL-lowering logic on nv50,
when it is inside of a conditional. I've been unable to trigger the
issue anywhere else so far. This resolves rendering failures
in a number of games like Two Worlds 2, Trine: Enchanted Edition, Trine 2,
XCOM:Enemy Unknown, Stacking. It also improves the situation in
Hearthstone, Sonic Generations, and The Raven: Legacy of a Master Thief.
However more work needs to be done there (splitting a lot more edges
solves it, so it's some other sort of RA-related issue).
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=90887
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Cc: "11.0" <mesa-stable@lists.freedesktop.org>
The purpose of the macro was to create the name_as_gs_input from name.
The previous commit removed the name_as_gs_input from add_varying, so
the macro is unnecessary.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
After inserting instructions the cursor.option becomes _after_instr
(even if it started life as an _after_block). So we cannot simply stash
the current cursor on the if/loop_stack. Otherwise we end up inserting
instructions after the endif/endloop in the block preceeding the if/
loop.
Signed-off-by: Rob Clark <robclark@freedesktop.org>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
CB updates to bound buffers need to go through the CB_DATA endpoints,
otherwise the shader may not notice that the updates happened.
Furthermore, these updates have to go in to the same address as the
bound buffer, otherwise, again, the shader may not notice updates.
So we keep track of all the places where a constbuf is bound, and
iterate over all of them when updating data. If a binding is found that
encompasses the region to be updated, then we use the settings of that
binding for the upload. Otherwise we upload as a regular data update.
This fixes piglit 'arb_uniform_buffer_object-rendering offset' as well
as blurriness in Witcher2.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=91890
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Cc: "11.0" <mesa-stable@lists.freedesktop.org>
This pass can be used as a helper for NIR producers so they don't have to
worry about creating the temporaries themselves.
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Some modern apps try to use msaa without keeping in mind the
restrictions on videomem of older cards. Resulting in dmesg saying:
[ 1197.850642] nouveau E[soffice.bin[3785]] fail ttm_validate
[ 1197.850648] nouveau E[soffice.bin[3785]] validating bo list
[ 1197.850654] nouveau E[soffice.bin[3785]] validate: -12
Because we are running out of video memory, after which the program
using the msaa visual freezes, and eventually the entire system freezes.
To work around this we do not allow msaa visauls by default and allow
the user to override this via NV30_MAX_MSAA.
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
[imirkin: move env var lookup to screen so that it's only done once]
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Cc: "10.6 11.0" <mesa-stable@lists.freedesktop.org>
We do not have a generic blitter on nv3x cards, so we must use the
sifm object for color resolving.
This commit divides the sources and dest surfaces in to tiles which
match the constraints of the sifm object, so that color resolving
will work properly on nv3x cards.
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
Cc: "11.0" <mesa-stable@lists.freedesktop.org>
Since debugging issues w/ fd's close()d at the wrong time can be quite
fun, this should probably be made more explicit in the docs.
Signed-off-by: Rob Clark <robclark@freedesktop.org>
This patch is necessary to avoid building error on android,
due to missing sid_tables.h generated sources
v2:[Emil Velikov] Correctly split the lists.
Fixes: fbbebeae10f(radeonsi: inline si_cmd_context_control)
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Otherwise the android build fails with
error : unable to find string literal operator ‘operator"" PRIx64’
There are several resources referring to the problem, which is related
to c++11, in our case used when building mesa for lollipop.
http://comments.gmane.org/gmane.comp.graphics.opensg.user/5883
I've not investigated all the semantics, some people even suggested a
bug in the gcc compiler,
I just saw the building error was solved with one little space for
lollipop and no side effect when c+11 not used.
v2: [Emil Velikov] add an alternative commit message from Mauro.
Cc: 11.0 <mesa-stable@lists.freedesktop.org>
Reviewed-by: Emil Velikov <emil.l.velikov@gmail.com>
There are a few bits this commit aims to resolve:
One can generalise the mkdir rule to a simple MKDIR_P $(@D) which will
expand appropriately for even if we change the subdir name, and/or add
new rules. We can also drop the explicit $(srcdir) prefix for the
dependency rules, they they are not strictly required, nor used
elsewhere in mesa.
Finally replace $< with explicit filename to be consistent through the
file, and honour PYTHON_FLAGS.
v2: Add comprehensive commit summary/message (Ian, Matt)
Cc: 11.0 <mesa-stable@lists.freedesktop.org>
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Rather than folding one variable within the other only to unwrap them,
just use the ones we need.
v2: bring back LOCAL_PATH prefix for nir_constant_expressions,h
Cc: 11.0 <mesa-stable@lists.freedesktop.org>
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Reviewed-by: Matt Turner <mattst88@gmail.com> (v1)
The glsl equivalent of "mesa: automake: rework the source generation
rules". Plus let's make things consistent and always explicitly provide
the header name.
v2: Rebase on top of reverted "remove custom AM_V_LEX/YACC" (Matt)
Cc: 11.0 <mesa-stable@lists.freedesktop.org>
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Same logic as previous commit applies.
Additionally remove the odd (set -e/mv/INDENT) from the rules.
The last one is the only one we remotely care about, if reading the
generated sources.
Upcoming work from DylanB which will replace the existing python
scripts with ones that produce more readable output anyway.
Cc: 11.0 <mesa-stable@lists.freedesktop.org>
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
A handful of changes/cleanups paving the way to bmake support:
- Remove optional $(srcdir)/ prefix for files in the prereq list.
- Drop the space after the AM_V_GEN variable.
- Using $< in a non-suffix rule is a GNU make idiom.
- Use $(@D) over $(dir $@). The latter is a POSIX standard.
v2: Cosmetic tweaks in the commit summary.
Cc: 11.0 <mesa-stable@lists.freedesktop.org>
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Reviewed-by: Matt Turner <mattst88@gmail.com> (v1)
This is the only place in mesa that uses this constuct which seems
to be GNUmake-ism. Attempting to build with POSIX make implementations
(bmake) would fail as below.
--- options.h ---
LOCALEDIR := .
sh: line 2: LOCALEDIR: command not found
*** [options.h] Error code 127
So let's keep things consistent and compatible by making the variable
non target specific.
v2:
- Bring back LOCALEDIR.
- Reword the commit message
- Change mesa-stable tag 10.6 > 11.0
Cc: 11.0 <mesa-stable@lists.freedesktop.org>
Cc: Jonathan Gray <jsg@jsg.id.au>
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
According to OpenGL ES 3.1 specification table : 20.2 and
OpenGL specification 4.4 table 23.4. The glGetIntegeri_v
functions should report the name of the buffer bound
when called with GL_VERTEX_BINDING_BUFFER.
Signed-off-by: Marta Lofstedt <marta.lofstedt@intel.com>
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
This code is all pretty much identical. We just needed the translation
from one enum value to the other.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
TRIFAN_NOSTIPPLE has always been 0x16 - 0x15 is marked "Reserved" on all
platforms. See the 965 PRM, Volume 2, Table 3-1, "3D Primitive Topology
Type Encoding" for a list.
We don't currently use this, and I don't expect we will, but we may as
well not leave the bogus value around.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Chris Forbes <chrisf@ijw.co.nz>
Docs suggest this is no longer required starting with Gen8.
Perf (no regressions in n=20)
OglMultithread 0.67%
OglTerrainPanInst 0.12%
trex 0.45%
warsow 0.64%
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Since 7a32652231
r600: Turn 'r600_shader_key' struct into union
we were accessing key fields that might be aliased in the union
with other fields, so we should check what shader type we are
compiling for before using key values from it.
v1.1: make it compile
v2: have caffeine, make it work - we don't set type
until later, so don't reference it until we've set it.
Reviewed-by: Edward O'Callaghan <eocallaghan@alterapraxis.com>
Cc: "11.0" <mesa-stable@lists.freedesktop.org>
Signed-off-by: Dave Airlie <airlied@redhat.com>
I meant to do this here, but it was in the wrong place:
commit c1151b18f2
Author: Ben Widawsky <benjamin.widawsky@intel.com>
Date: Wed Jun 24 20:07:54 2015 -0700
i965/skl: Use more compact hiz dimensions
NOTE: Jordan did go back and look at the original mailing list post. I mailed
the right thing, and pushed the wrong one.
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Neil Roberts <neil@linux.intel.com>
Indications are that if the colormask indicates a single bit set on
fermi, that value will always be read from $r0 instead of a potentially
higher register (if e.g. green is set). Not to upset the counting logic,
always set the header up with a full color mask for each RT. Such a
situation can basically only ever happen with generated blit shaders.
Fixes the following piglit on Fermi (Kepler is unaffected):
fbo-stencil blit GL_DEPTH32F_STENCIL8
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Cc: "10.6 11.0" <mesa-stable@lists.freedesktop.org>
Commit 2126c68e5c killed the array elements parameter on load/store
intrinsics that was stored in const_index[1]. It looks like that
patch missed to remove this assignment in the UBO path.
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
The sifm object has a limit of 1024x1024 for its input size and 2048x2048
for its output. The code checking this was trying to be clever resulting
in it seeing a surface of e.g 1024x256 being outside of the input size
limit.
This commit fixes this.
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
Cc: "10.6 11.0" <mesa-stable@lists.freedesktop.org>
X11_CFLAGS is never defined. Path to X11 headers is not needed here, so
just remove.
Future work: Using AM_CFLAGS here looks wrong, as this Makefile only builds
C++ files
Signed-off-by: Jon TURNEY <jon.turney@dronecode.org.uk>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Emil Velikov <emil.l.velikov@gmail.com>
Make sure errors are correcly propagated.
Also don't flush during state emission if emission fails.
Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Reviewed-by: Charmaine Lee <charmainel@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Like xa_surface_from_handle(), but takes a handle type, rather than
hard-coding 'shared' handle. This is needed to fix bugs seen with
xf86-video-freedreno with xrandr rotation, for example. The root issue
is that doing a GEM_OPEN ioctl on a bo that already has a GEM handle
associated with the drm_file will result in two unique handles for the
same bo. Which causes all sorts of follow-on fail.
v2:
- Add support for for fd handles.
- Avoid duplicating code.
- Bump xa version minor.
Signed-off-by: Rob Clark <robclark@freedesktop.org>
Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
My earlier attempt to fix this missed the fact that there was a #else
clause that assumes that you have openssh. This moves the whole thing
under #ifdef HAVE_SHA1 which should avoid this issue.
Fixes: 13bfa5201 (util: always include sha1 into the build)
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=91898
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Dave Airlie <airlied@gmail.com>
SHA1 is now used in all builds when HAVE_SHA1 is defined. Adjust src to
do the same thing, rather than predicating on shader cache.
Fixes: 04e201d0c0 ("mesa: change 'SHADER_SUBST' facility to work with env variables")
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Dave Airlie <airlied@gmail.com>
Nothing in the spec allows for the reduced precision, and this also
fixes st_QuerySamplesForFormat for nv50, which does not allow MS8 on
RGBA32F. Now this will be respected instead of reporting MS8 as
supported with an assumption that the format used will be RGBA16F.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Cc: "10.6 11.0" <mesa-stable@lists.freedesktop.org>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
vbuf is never null. We want to make sure that a resource was allocated
for the vbuf, which is *vbuf.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
The hardware only generates vertexid when vertices come from a VBO. This
fixes:
vertexid-drawelements
vertexid-drawarrays
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Cc: "11.0" <mesa-stable@lists.freedesktop.org>
The stride was being set to 0, which is illegal (and also non-sensical).
Also we must wait for the buffer to become available for reading as
otherwise a wrong value may be prefetched. Since we must wait for the
buffer anyways, and it's mapped and in GART, we may as well avoid the
annoyance of the indirect pushbuf submit.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Cc: mesa-stable@lists.freedesktop.org
Gen9 changes the meaning of this to coarse LOD quality mode. Although that's a
desirable thing to be setting, it doesn't match the gen8 behavior and this was
unintentional. More importantly, we don't ever use this field. So instead of
getting it "wrong" drop it entirely.
This is a respin of a patch which only [incorrectly] tried to address gen9.
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
round(val*dscale) produces a double result, as val and dscale are double.
However, LLVMConstInt receives unsigned long long, so there is an
implicit conversion from double to unsigned long long.
This is an undefined behavior. Therefore, we need to first explicitly
convert the round result to long long, and then let the compiler handle
conversion from that to unsigned long long.
This bug manifests itself in POWER, where all IMM values of -1 are being
converted to 0 implicitly, causing a wrong LLVM IR output.
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
CC: "10.6 11.0" <mesa-stable@lists.freedesktop.org>
Reviewed-by: Tom Stellard <thomas.stellard@amd.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Note this is not ideal. Since the sifm can only do source sizes upto
1024x1024 we end up using the blitter on nv4x, which is not that fast.
And on nv3x we end up using the cpu which is really slow.
Cc: "10.6 11.0" <mesa-stable@lists.freedesktop.org>
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
Scanout buffers on nv30 must always be non-swizzled and have special
width alignment constraints.
These constrains have been taken from the xf86-video-nouveau
src/nv_accel_common.c: nouveau_allocate_surface() function.
nouveau_allocate_surface() applies these width constraints only when a
tiled attribute is set, which it sets for all surfaces allocated via
dri, and this "tiling" is not the same as swizzling, scanout surfaces
must be linear / have a uniform_pitch or only complete garbage is shown.
This commit fixes dri3 on nv30 showing a garbled display, with dri3 the
scanout buffers are allocated by mesa, rather then by the ddx, and the
wrong stride of these buffers was causing the garbled display.
Cc: "10.6 11.0" <mesa-stable@lists.freedesktop.org>
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
The tiled memcpy fast paths perform a simple blit (with only a couple of
trivial pixel conversion routines) and do not accommodate PixelTransfer
operations. Therefore if any are set, fallback to the regular routines.
Note that PixelTransfer only applies to TexImage and ReadPixels, not to
GetTexImage.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Jason Ekstrand <jason.ekstrand@intel.com>
Cc: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Cc: mesa-stable@lists.freedesktop.org
If we have spilled/unspilled a register in the current instruction, avoid
emitting unspills for the same register in the same instruction or consecutive
instructions following the current one as long as they keep reading the spilled
register. This should allow us to avoid emitting costy unspills that come with
little benefit to register allocation.
v2:
- Apply the same logic when evaluating spilling costs (Curro).
v3:
- Abstract the logic that decides if a register can be reused in a function.
that can be used from both spill_reg and evaluate_spill_costs (Curro).
v4:
- Do not disallow reusing scratch_reg in predicated reads (Curro).
- Track if previous sources in the same instruction read scratch_reg (Curro).
- Return prev_inst_read_scratch_reg at the end (Curro).
- No need to explicitily skip scratch read/write opcodes in spill_reg (Curro).
- Fix the comments explaining what happens when we hit an instruction that
does not read or write scratch_reg (Curro)
- Return true early when the current or previous instructions read
scratch_reg with a compatible mask.
v5:
- Do not return true early, the loop should not be expensive anyway
and this adds more complexity (Curro).
Reviewed-by: Francisco Jerez <currojerez@riseup.net>
We now print out the name of the message instead of its numerical
value, and label the message control and surface numbers.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Chris Forbes <chrisf@ijw.co.nz>
The entire VUE map is computed based on the slots_valid bitfield;
calling brw_compute_vue_map on the same bitfield will return the
same result. So we can simply compare those.
struct brw_vue_map is 136 bytes; doing a single 8-byte comparison is
much cheaper and should work just as well.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Chris Forbes <chrisf@ijw.co.nz>
These were only for legacy userclipping, which we no longer support
in geometry shaders.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Chris Forbes <chrisf@ijw.co.nz>
According to the GLSL 1.50 specification, page 76:
"The shader must also set all values in gl_ClipDistance that have been
enabled via the OpenGL API, or results are undefined."
With this patch, we only enable clip distance writes when the shader
actually writes them. We no longer force a value to be written when
clip planes are enabled in the API. This could mean the first varying
slot would be used as clip distances - I believe it should be the safe
kind of undefined behavior.
Empirically, it doesn't seem to cause a problem.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Chris Forbes <chrisf@ijw.co.nz>
The legacy userclip fields are only used for the vertex shader, and at
that point there's only program_string_id and the tex struct, which are
common to all keys. So there's no need for a "VUE" key base class.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Chris Forbes <chrisf@ijw.co.nz>
This avoids a downcast of key, which won't exist in the base class soon.
I'm not a huge fan of this patch, but given that we're currently using
inheritance, this seems like the "right" way to do it. The alternative
is to make key a void pointer in the parent class and continue
downcasting.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Chris Forbes <chrisf@ijw.co.nz>
I'm about to remove the base class for VS/GS/HS/DS program keys, at
which point we won't be able to use key->tex anymore. Instead, we'll
need to store a direct pointer (like we do in the FS backend).
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Chris Forbes <chrisf@ijw.co.nz>
This is now only used for the vertex shader, so it makes sense to get it
out of any paths run by the geometry shader.
Instead of passing the gl_clip_plane array into the run() method (which
is shared among all subclasses), we add it as a vec4_vs_visitor
constructor parameter. This eliminates the bogus NULL parameter in the
GS case.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Chris Forbes <chrisf@ijw.co.nz>
There are two uses of this flag.
The primary use is checking whether we need to emit code to convert
legacy gl_ClipVertex/gl_Position clipping to clip distances. In this
case, we also have to upload the clip planes as uniforms, which means
setting nr_userclip_plane_consts to a positive value. Checking if it's
> 0 works for detecting this case.
Gen4-5 also wants to know whether we're doing clipping at all, so it can
emit user clip flags. Checking if output_reg[VARYING_SLOT_CLIP_DIST0]
is set to a real register suffices for this.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Chris Forbes <chrisf@ijw.co.nz>
We only support geometry shaders in core profiles, where gl_ClipVertex
doesn't exist. Presumably the even older behavior of clipping to
gl_Position isn't supported either. In fact, GLSL 1.50 page 76 claims:
"The shader must also set all values in gl_ClipDistance that have been
enabled via the OpenGL API, or results are undefined."
So we don't need to handle legacy clipping in geometry shaders. I think
Paul added this back when we were considering supporting the old
GL_ARB_geometry_shader4 extension.
This removes a non-orthagonal state dependency on GS compilation.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Chris Forbes <chrisf@ijw.co.nz>
This living in brw_fs.{h,cpp} is a historical artifact of us supporting
texturing for fragment shaders before any other stages. It's kind of
awkward given that we use it for all stages.
This avoids having to include brw_fs.h in geometry shader code in order
to access this function.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Chris Forbes <chrisf@ijw.co.nz>
Patch modifies existing shader source and replace functionality to work
with environment variables rather than enable dumping on compile time.
Also instead of _mesa_str_checksum, _mesa_sha1_compute is used to avoid
collisions.
Functionality is controlled via two environment variables:
MESA_SHADER_DUMP_PATH - path where shader sources are dumped
MESA_SHADER_READ_PATH - path where replacement shaders are read
v2: cleanups, add strerror if fopen fails, put all functionality
inside HAVE_SHA1 since sha1 is required
Signed-off-by: Tapani Pälli <tapani.palli@intel.com>
Suggested-by: Eero Tamminen <eero.t.tamminen@intel.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 472ef9a02f introduced code to
change the types of SEL and MOV instructions for moves that simply
"copy bits around". It didn't account for type conversion moves,
however. So it would happily turn this:
mov(8) vgrf6:D, -vgrf5:D
mov(8) vgrf7:F, vgrf6:UD
into this:
mov(8) vgrf6:D, -vgrf5:D
mov(8) vgrf7:D, -vgrf5:D
which erroneously drops the conversion to float.
Cc: "11.0 10.6" <mesa-stable@lists.freedesktop.org>
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
As far as I can tell, the behavior is preserved from the previous generations.
Before we set a single bit to tell the FS whether or not we'll be using an input
coverage mask. Now we have some options which are implementing various
extensions. These bits are used for the various conservative rasterization
mechanisms (for collision detection, binning, and whatever else).
I believe that the behavior is preserved because the problem which conservative
rasterization is attempting to fix would go away with the "NORMAL" mode (at the
cost of performance, I believe).
This patch serves as documentation of the change by creating the enums, as well
as giving some of the history with the links here so that the next person who
comes along and looks at it doesn't spend as long as I had to in order to
determine if there is an issue or not.
Previously, this algorithm had been done in software, and this can still be used
as long as we don't export an extension stating otherwise.
References: https://www.opengl.org/registry/specs/NV/conservative_raster.txt
References: https://http.developer.nvidia.com/GPUGems2/gpugems2_chapter42.html
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
This must be done before exporting a buffer as dmabuf fds, because
we lose track of who is using it and can't trust the reference counter.
Cc: 11.0 <mesa-stable@lists.freedesktop.org>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
This is probably the most called util function. It does almost nothing,
yet it can consume 10% of the CPU on the profile. This drops it down to 5%.
Reviewed-by: Brian Paul <brianp@vmware.com>
There doesn't seem any reason to start from 4.
Start from 1 instead (0 is left reserved to catch uninitialized atoms).
Signed-off-by: Marek Olšák <marek.olsak@amd.com>
Similarly to scissor states, we can use single atom to track all viewport
states. This will allow to simplify dirty atom handling later.
Signed-off-by: Marek Olšák <marek.olsak@amd.com>
During review of the "r600g: make all scissor states use single atom" patch
Marek Olšák noticed that scissor disable workaround should be applied on
all scissor states and not just first one, so let's do so.
Signed-off-by: Marek Olšák <marek.olsak@amd.com>
As suggested by Marek Olšák, we can use single atom to track all scissor
states. This will allow to simplify dirty atom handling later.
Signed-off-by: Marek Olšák <marek.olsak@amd.com>
It's legal to call glTexSubImage with zero values for the width,
height or depth. Previously this was breaking the PBO access
validation because it tries to work out the last pixel accessed by
getting the pixel at height-1 and depth-1 which would end up with
bogus values.
This was causing GL errors to be generated during the Piglit
texsubimage test, although the test was passing anyway.
v2: Also check for width == 0. Don't validate the start pointer if any
of the dimensions are zero.
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
In various versions of OpenGL and GLSL, it's possible to declare
multiple VS input variables with aliasing attribute locations.
So, when computing the storage requirements for vertex attributes,
we can't simply add up the sizes. Instead, we need to look at the
enabled slots.
This patch begins tracking which attributes are double types that
are larger than 128-bits (i.e. take up two vec4 slots). We then
count normal attributes once, and count the double-size attributes
a second time.
Fixes deQP functional.attribute_location.bind_aliasing.max_cond_* tests
on i965, which regressed with commit ad208d975a.
No Piglit changes on llvmpipe (which actually supports dvecs).
Cc: "10.6 11.0" <mesa-stable@lists.freedesktop.org>
Tested-by: Mark Janes <mark.a.janes@intel.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Dave Airlie <airlied@redhat.com>
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Previously we would allow glUniformMatrix4fv on a dmat4 and
glUniformMatrix4dv on a mat4. Both are illegal. That later also
overwrites the storage for the mat4 and causes bad things to happen.
Should fix the (new) arb_gpu_shader_fp64-wrong-type-setter piglit test.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Timothy Arceri <t_arceri@yahoo.com.au>
Cc: Dave Airlie <airlied@redhat.com>
Cc: "10.6 11.0" <mesa-stable@lists.freedesktop.org>
All of the other state upload functions are static because the only use
is in the brw_tracked_state structure.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Abdiel Janulgue <abdiel.janulgue@linux.intel.com>
All of the other state upload functions are static because the only use
is in the brw_tracked_state structure.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
Because the compiler already has enough things to complain about.
grep -rl 'const static' src/ | while read f
do
sed --in-place -e 's/const static/static const/g' $f
done
brw_eu_emit.c: In function 'brw_reg_type_to_hw_type':
brw_eu_emit.c:98:7: warning: 'static' is not at beginning of declaration [-Wold-style-declaration]
const static int imm_hw_types[] = {
^
brw_eu_emit.c:120:7: warning: 'static' is not at beginning of declaration [-Wold-style-declaration]
const static int hw_types[] = {
^
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
brw_upload_cs_push_constants was based on gen6_upload_push_constants.
v2:
* Add FINISHME comments about more efficient ways to push uniforms
Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Ben Widawsky <ben@bwidawsk.net>
Check for a valid framebuffer cbuf pointer before accessing its
associated surface.
Fix piglit test fbo-drawbuffers-none.
Reviewed-by: Brian Paul <brianp@vmware.com>
Commit b9ba8492 removes an unneeded pipe_surface_release() from
st_render_texture(). This implies a surface can now be reused for a
render buffer. Currently, when we render to a texture, we mark the
surface as dirty. But in svga_mark_surface_dirty(), if the surface
is already marked as dirty, it does not increment the texture age.
Any view to this texture might not be updated properly then.
With this patch, the texture age is incremented regardless of whether
the surface is already marked as dirty or not.
Fix bug 1499181.
Reviewed-by: Sinclair Yeh <syeh@vmware.com>
Commit b9ba8492 removes an unneeded pipe_surface_release() from
st_render_texture() and exposes a bug in the backed surface view
creation. Currently a backed surface view for a conflicted surface view
is created at framebuffer emit time. But if shader sampler views are changed
but framebuffer surface views remain unchanged, emit_framebuffer() will not
be called and conflicted surface views will not be detected.
To fix this, also check for conflicted surface views when setting sampler
views. If there is any conflicted surface views, enable the
framebuffer dirty bit so that the framebuffer emit code has a chance to
create a backed surface view for the conflicted surface view.
Fix cinebench-r11-test regression.
Reviewed-by: Brian Paul <brianp@vmware.com>
The lowered code reads from the destination, which isn't possible from
message registers.
Fixes the following dEQP tests on SNB:
dEQP-GLES3.functional.shaders.precision.int.highp_mul_fragment
dEQP-GLES3.functional.shaders.precision.int.mediump_mul_fragment
dEQP-GLES3.functional.shaders.precision.int.lowp_mul_fragment
Cc: "10.6 11.0" <mesa-stable@lists.freedesktop.org>
Tested-by: Mark Janes <mark.a.janes@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
This is a squash commit of roughly two years of development work.
Authors include:
Brian Paul
Charmaine Lee
Thomas Hellstrom
Jakob Bornecrantz
Sinclair Yeh
Mingcheng Chen
Kai Ninomiya
MengLin Wu
The driver supports OpenGL 3.3.
Signed-off-by: Brian Paul <brianp@vmware.com>
The VMware svga driver doesn't directly support pipe_screen::get_timestamp()
but we can do a work-around. However, we need a gallium context to do so.
This patch adds a new pipe_context::get_timestamp() function that will only
be called if the pipe_screen::get_timestamp() function is NULL.
Signed-off-by: Brian Paul <brianp@vmware.com>
This involves a few driver modifications to keep things building.
The driver may not actually run properly at this point.
Signed-off-by: Brian Paul <brianp@vmware.com>
If the user is specifying a subregion of a buffer using SKIP_ROWS and
SKIP_PIXELS, we must compute the buffer size carefully as the end of the
last row may be much shorter than stride*image_height*depth. The current
code tries to memcpy from beyond the end of the user data, for example
causing:
==28136== Invalid read of size 8
==28136== at 0x4C2D94E: memcpy@@GLIBC_2.14 (vg_replace_strmem.c:915)
==28136== by 0xB4ADFE3: brw_bo_write (brw_batch.c:1856)
==28136== by 0xB5B3531: brw_buffer_data (intel_buffer_objects.c:208)
==28136== by 0xB0F6275: _mesa_buffer_data (bufferobj.c:1600)
==28136== by 0xB0F6346: _mesa_BufferData (bufferobj.c:1631)
==28136== by 0xB37A1EE: create_texture_for_pbo (meta_tex_subimage.c:103)
==28136== by 0xB37A467: _mesa_meta_pbo_TexSubImage (meta_tex_subimage.c:176)
==28136== by 0xB5C8D61: intelTexSubImage (intel_tex_subimage.c:195)
==28136== by 0xB254AB4: _mesa_texture_sub_image (teximage.c:3654)
==28136== by 0xB254C9F: texsubimage (teximage.c:3712)
==28136== by 0xB2550E9: _mesa_TexSubImage2D (teximage.c:3853)
==28136== by 0x401CA0: UploadTexSubImage2D (teximage.c:171)
==28136== Address 0xd8bfbe0 is 0 bytes after a block of size 1,024 alloc'd
==28136== at 0x4C28C20: malloc (vg_replace_malloc.c:296)
==28136== by 0x402014: PerfDraw (teximage.c:270)
==28136== by 0x402648: Draw (glmain.c:182)
==28136== by 0x8385E63: ??? (in /usr/lib/x86_64-linux-gnu/libglut.so.3.9.0)
==28136== by 0x83896C8: fgEnumWindows (in /usr/lib/x86_64-linux-gnu/libglut.so.3.9.0)
==28136== by 0x838641C: glutMainLoopEvent (in /usr/lib/x86_64-linux-gnu/libglut.so.3.9.0)
==28136== by 0x8386C1C: glutMainLoop (in /usr/lib/x86_64-linux-gnu/libglut.so.3.9.0)
==28136== by 0x4019C1: main (glmain.c:262)
==28136==
==28136== Invalid read of size 8
==28136== at 0x4C2D940: memcpy@@GLIBC_2.14 (vg_replace_strmem.c:915)
==28136== by 0xB4ADFE3: brw_bo_write (brw_batch.c:1856)
==28136== by 0xB5B3531: brw_buffer_data (intel_buffer_objects.c:208)
==28136== by 0xB0F6275: _mesa_buffer_data (bufferobj.c:1600)
==28136== by 0xB0F6346: _mesa_BufferData (bufferobj.c:1631)
==28136== by 0xB37A1EE: create_texture_for_pbo (meta_tex_subimage.c:103)
==28136== by 0xB37A467: _mesa_meta_pbo_TexSubImage (meta_tex_subimage.c:176)
==28136== by 0xB5C8D61: intelTexSubImage (intel_tex_subimage.c:195)
==28136== by 0xB254AB4: _mesa_texture_sub_image (teximage.c:3654)
==28136== by 0xB254C9F: texsubimage (teximage.c:3712)
==28136== by 0xB2550E9: _mesa_TexSubImage2D (teximage.c:3853)
==28136== by 0x401CA0: UploadTexSubImage2D (teximage.c:171)
==28136== Address 0xd8bfbe8 is 8 bytes after a block of size 1,024 alloc'd
==28136== at 0x4C28C20: malloc (vg_replace_malloc.c:296)
==28136== by 0x402014: PerfDraw (teximage.c:270)
==28136== by 0x402648: Draw (glmain.c:182)
==28136== by 0x8385E63: ??? (in /usr/lib/x86_64-linux-gnu/libglut.so.3.9.0)
==28136== by 0x83896C8: fgEnumWindows (in /usr/lib/x86_64-linux-gnu/libglut.so.3.9.0)
==28136== by 0x838641C: glutMainLoopEvent (in /usr/lib/x86_64-linux-gnu/libglut.so.3.9.0)
==28136== by 0x8386C1C: glutMainLoop (in /usr/lib/x86_64-linux-gnu/libglut.so.3.9.0)
==28136== by 0x4019C1: main (glmain.c:262)
==28136==
Fixes regression from commit 7f396189f0
Author: Jason Ekstrand <jason.ekstrand@intel.com>
Date: Mon Jan 5 18:17:04 2015 -0800
meta: Add a BlitFramebuffers-based implementation of TexSubImage
v2: However, the teximage we create does need to be width x full_height x 1
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Jason Ekstrand <jason.ekstrand@intel.com>
Cc: Neil Roberts <neil@linux.intel.com>
Reviewed-by Neil Roberts <neil@linux.intel.com>
The src_reg constructor that received the glsl_type was using it
only to build the swizzle, but not to fill this->type as dst_reg
is doing.
This caused some type mismatch between movs and alu operations
on the NIR path, so copy propagation optimization was not applied
to remove unneeded movs if negate modifier was involved. This was
first detected on minus (negate+add) operations.
Shader DB results (taking into account only vec4):
total instructions in shared programs: 20019 -> 19934 (-0.42%)
instructions in affected programs: 2918 -> 2833 (-2.91%)
helped: 79
HURT: 0
GAINED: 0
LOST: 0
Reviewed-by: Matt Turner <mattst88@gmail.com>
Only a subset of AMD GPUs supported by r600g support doubles,
CAYMAN and CYPRESS are probably all we'll try and support, however
I don't have a CYPRESS so ignore that for now.
This disables SB support for doubles, as we think we need to
make the scheduler smarter to introduce delay slots.
[airlied: pushing this to avoid pain of rebasing, it mostly
works on cayman only so far, Glenn has some ideas about
delay slot issues we need to look into. turned off by
default for now]
Signed-off-by: Dave Airlie <airlied@redhat.com>
This patch is taken from work by Glenn and myself,
and I've spent some time making it all work here.
This adds support for the multiple streams part of
ARB_gpu_shader5 to r600g.
It doesn't enable ARB_gpu_shader5 yet.
Signed-off-by: Dave Airlie <airlied@redhat.com>
This adds a peephole and removes an assert that isn't
actually valid with some of the stream emit instructions.
Signed-off-by: Dave Airlie <airlied@redhat.com>
This just adds support to the assembler dumper and allows
stream instructions to be generated. Also fix up the stream
debugging to add stream info.
Signed-off-by: Dave Airlie <airlied@redhat.com>
The CTS packed_pixels test checks that readpixels doesn't write
into the space between rows, however we fail that here unless
we check the format and stride match.
This fixes all the core mesa problems with CTS packed_pixels
tests.
Cc: "11.0" <mesa-stable@lists.freedesktop.org>
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
The fastpath currently checks the RowLength != width, but
if you have a RowLength of 7, and Alignment of 4, then
that shouldn't match.
align the rowlength to the pack alignment before comparing.
This fixes compressed cases in CTS packed_pixels_pixelstore
test when SKIP_PIXELS is enabled, which causes row length
to get set.
v1.1: add fxt1 fix (Iago)
Cc: "11.0" <mesa-stable@lists.freedesktop.org>
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
We don't need to use the 3d image address here as that will
include SKIP_IMAGES, and we are only blitting a single
2D anyways, so just use the 2D path.
This fixes some memory overruns under CTS
packed_pixels.packed_pixels_pixelstore when PACK_SKIP_IMAGES
is used.
Cc: "11.0" <mesa-stable@lists.freedesktop.org>
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
GL3.3 added GL_ARB_texture_rgb10_a2ui, which specifies
a lot more things than just rgb10/a2ui.
While playing with ogl conform one of the tests must
attempted all valid formats for GL3.3 and hits the
unreachable here.
This adds the first chunk of formats that hit the
assert.
Cc: "11.0" <mesa-stable@lists.freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
In a number of places the SwapBytes handling didn't handle cases with
GL_(UN)PACK_ALIGNMENT set and 7 byte width cases aligned to 8 bytes.
This adds a common routine to swap bytes a 2D image and uses this
code in:
texture storage
texture get
readpixels
swrast drawpixels.
[airlied: updated with Brian's nitpicks].
Cc: "11.0" <mesa-stable@lists.freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
Let it be defined externally instead, allowing setting mechanisms other
than environment variables.
Reviewed-by: Zack Rusin <zackr@vmware.com>
Reviewed-by: Matthew McClure <mcclurem@vmware.com>
The first function translates prim restart indexes to be 0xffff or
0xffffffff.
The second splits indexed primitives with restart indexes into sub-
primitives without restart indexes.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
This adds a tgsi utility tgsi_add_aa_point to transform a fragment shader
to support anti-aliased wide point by computing the fragment distance from
the point center. This utility assumes the geometry shader is emitting
an extra generic output with point coord data. The semantic index of
this generic output is passed to the tgsi_add_aa_point utility.
Reviewed-by: Brian Paul <brianp@vmware.com>
This adds a tgsi utility tgsi_add_point_sprite to transform a geometry
shader to emulate wide points by drawing quads. This utility adds an
extra output for the original point position if the point position is
to be written to a stream output buffer. It also assumes the driver will
add a constant for inverse viewport scale after the user defined constants.
Reviewed-by: Brian Paul <brianp@vmware.com>
This could be used by any driver where the device doesn't directly
support two-sided lighting. This code modifies a fragment shader
to accecpt back-face colors and choose between the front/back colors
depending on the triangle's front-face sign.
These functions deal with inclusive coordinates, hence a 0/0/0/0 rect
returned when there's no intersection doesn't actually represent an empty
rectangle. Hence return 0/-1/0/-1 instead.
This fixes some problems in llvmpipe with empty scissor rects (which up
to now didn't really matter because while the intersect test returned the
wrong result all pixels were scissored away later anyway).
It isn't really obvious if intersection test should take into account empty
rectangles or if the caller should do it. But it looks like most callers
actually verified one of the rects but not the other, but since correctly
returning an empty rect that other rect could actually be empty leading to
more bugs. Hence just verify both rects for emptyness in the intersection
test itself which makes the code easier in the caller (though it will be
slower if the caller knows the rectangles are non-empty).
Reviewed-by: Zack Rusin <zackr@vmware.com>
This patch adds some more helper functions such as
. tgsi_transform_temps_decl
. tgsi_transform_output_decl
. tgsi_transform_dst_reg
. tgsi_transform_src_reg
Reviewed-by: Brian Paul <brianp@vmware.com>
The border colors are uploaded only once when the state is created.
This brings truly immutable sampler descriptors, because they don't have
to be updated every time a sampler state is re-bound.
It also moves the TA_BC_BASE_ADDR registers to init_config, removing one
more state. The catch is there is now a limit: only 4096 border colors can
be used by one context. I don't think that will be a problem.
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Acked-by: Christian König <christian.koenig@amd.com>
Since we don't put any resource descriptors in IBs, the space used by draw
calls is quite small.
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Acked-by: Christian König <christian.koenig@amd.com>
The main idea is to avoid setting CB_COLORi_INFO = 0 for i>0 repeatedly
when those colorbuffers aren't used. This is mainly for glamor.
Same for DB. Z_INFO and STENCIL_INFO need to be cleared only once.
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Acked-by: Christian König <christian.koenig@amd.com>
This mainly removes the cache misses when checking the dirty flags.
Not much else though.
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Acked-by: Christian König <christian.koenig@amd.com>
I need to initialize more atom IDs.
This adds 4 more si_init_atom calls, which simplifies the code.
(si_init_atom needs a different context type of the emit functions though)
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Acked-by: Christian König <christian.koenig@amd.com>
- convert 16 states to 1 atom
- only emit 1 scissor if VIEWPORT_INDEX isn't written
- use only one packet when emitting consecutive scissors
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Acked-by: Christian König <christian.koenig@amd.com>
Fixes regression from
commit 8c17d53823
Author: Kenneth Graunke <kenneth@whitecape.org>
Date: Wed Apr 15 03:04:33 2015 -0700
i965: Make intel_emit_linear_blit handle Gen8+ alignment restrictions.
which adjusted the coordinates to be relative to the nearest cacheline.
However, this then offsets the coordinates by up to 63 and this may then
cause them to overflow the BLT limits. For the well aligned large
transfer case, we can use 32bpp pixels and so reduce the coordinates by
4 (versus the current 8bpp pixels). We also have to be more careful
doing the last line just in case it may exceed the coordinate limit.
Reported-and-tested-by: kaillasse91@hotmail.fr
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=90734
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Kenneth Graunke <kenneth@whitecape.org>
Cc: Ian Romanick <ian.d.romanick@intel.com>
Cc: Anuj Phogat <anuj.phogat@gmail.com>
Cc: mesa-stable@lists.freedesktop.org
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
v2: fix detecting if the loop has any phi nodes after it.
v2: use nir_foreach_ssa_def() instead of nir_foreach_dest() when
checking for values live after the loop to catch const_load
instructions.
v2: fix handling return instructions
v2: add some documentation to loop_is_dead()
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
We were already doing this internally for iterating over a function
implementation, so just expose it directly.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
v2: use nir_cf_node_remove_after().
v2: use foreach_list_typed() instead of hardcoding a list walk.
v3: update to new control flow modification helpers.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
v2: use nir_cf_node_remove_after() instead of our own broken thing.
v3: use the new control flow modification helpers.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
I've been chasing a geom shader hang on rv635 since I wrote
r600 geom code, and finally I hacked some values from fglrx
in and I could run texelfetch without failures.
This is totally my fault as well, maths fail 101.
This makes geom shaders on r600 not fail heavily.
Cc: "10.6" "11.0" <mesa-stable@lists.freedesktop.org>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Reviewed-by: Edward O'Callaghan <eocallaghan@alterapraxis.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
According to OpenGL ES 3.1 specification, section 9.2.1 for
glFramebufferParameter and section 9.2.3 for glGetFramebufferParameteriv:
"An INVALID_ENUM error is generated if pname is not FRAMEBUFFER_DEFAULT_WIDTH,
FRAMEBUFFER_DEFAULT_HEIGHT, FRAMEBUFFER_DEFAULT_SAMPLES, or
FRAMEBUFFER_DEFAULT_FIXED_SAMPLE_LOCATIONS."
Therefore exclude OpenGL ES 3.1 from using the GL_FRAMEBUFFER_DEFAULT_LAYERS
parameter.
Signed-off-by: Marta Lofstedt <marta.lofstedt@intel.com>
Reviewed-by: Kevin Rogovin <kevin.rogovin at intel.com>
This *should* ensure that the cursor gets properly advanced in all cases.
We had a problem before where, if the cursor was created using
nir_after_cf_node on a non-block cf_node, that would call nir_before_block
on the block following the cf node. Instructions would then get inserted
in backwards order at the top of the block which is not at all what you
would expect from nir_after_cf_node. By just resetting to after_instr, we
avoid all these problems.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
This function's cases for non-generic compressed formats duplicate
the GL to MESA translation in _mesa_glenum_to_compressed_format().
This patch replaces the switch cases with a call to the translation
function. This change teaches this function about ASTC, thus enabling
ASTC for glTex*Storage*() calls.
Reviewed-by: Chad Versace <chad.versace@intel.com>
Signed-off-by: Nanley Chery <nanley.g.chery@intel.com>
MESA_FORMAT_RGBA_DXT5 should actually be reserved for GL_RGBA[4]_DXT5_S3TC.
Also, Gallium and other dri drivers (radeon and nouveau) follow this mapping
scheme.
Reviewed-by: Chad Versace <chad.versace@intel.com>
Signed-off-by: Nanley Chery <nanley.g.chery@intel.com>
As Glenn did for finalize_loop we need to update_cf when we
add a POP at the end of a shader.
I think this fixes one of the earlier shader going off end
of memory problems we've stopped.
Reviewed-by: Glenn Kennard <glenn.kennard@gmail.com>
Cc: "10.6" "11.0" <mesa-stable@lists.freedesktop.org>
Signed-off-by: Dave Airlie <airlied@redhat.com>
The docs specifically call out SEL with .l and .ge as the
implementations of MIN and MAX respectively. Among other things,
SEL with these conditional mods are commutative.
See commit 3b7f683f.
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Certain compressed formats require this setting. The docs don't go into much
detail as to why it's needed exactly.
This patch introduces no piglit regressions on gen9 (bsw is untested). Note that
the SKL "regressions" are fixed tests, and the egl_khr_gl_colorspace tests are
WTF. The patch also fixes nothing I can find.
http://otc-mesa-ci.jf.intel.com/job/Leeroy/127820/
v2:
Reworded commit message (Matt); Added piglit results link.
Restructured condition (Matt)
Moved check out to function (Nanley). I left the setting of the bit in the
surface state open coded because it seems to go better with the existing code.
v3:
Use and inline function only in gen8_emit_texture_surface_state() (Matt).
Cc: Matt Turner <mattst88@gmail.com>
Cc: Nanley Chery <nanleychery@gmail.com>
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Reviewed-by: Matt Turner <mattst88@gmail.com>
This can be done with a single pass for the instruction base,
and takes renumber_registers out of its spot on the profile.
Acked-by: Marek Olšák <marek.olsak@amd.com
Signed-off-by: Dave Airlie <airlied@redhat.com>
The glsl->tgsi convertor does some temporary register reduction
however in profiling shader-db this shows up quite highly,
so optimise things to reduce the number of loops through
all the instructions we do. This drops merge_registers
from 4-5% on the profile to 1%. I think this can be reduced
further by possibly optimising the renumber pass.
Acked-by: Marek Olšák <marek.olsak@amd.com
Signed-off-by: Dave Airlie <airlied@redhat.com>
Instead of looking this up lots, lets just cache it in the instruction
translation up front. I just noticed this function what high in a profile
of shader-db on radeonsi.
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
The selector is shared by all shader variants, so the
individual shaders shouldn't change it. Use tgsi_shader_scan()
results to set geometry properties within a
r600_create_shader_state() call and treat said propertices in
the selector as read-only within r600_shader_from_tgsi().
Signed-off-by: Edward O'Callaghan <eocallaghan@alterapraxis.com>
Signed-off-by: Marek Olšák <marek.olsak@amd.com>
Note that 'geometry shader properties' should be carried in the
selector state over the shader state in any case.
Signed-off-by: Edward O'Callaghan <eocallaghan@alterapraxis.com>
Signed-off-by: Marek Olšák <marek.olsak@amd.com>
The hardware is capable of dealing with GL1-style user clip planes.
No clip vertex, no clip distances. Fixes a number of ucp tests, as well
as neverball.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Cc: "11.0" <mesa-stable@lists.freedesktop.org>
According to NVIDIA, local performance counters (MP) are prefixed
with SM, while global performance counters (PCOUNTER) are called PM.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Otherwise this will crash on 32-bit, and it gets rid of
warnings building on 32-bit.
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
This greatly improves generated code, especially for the snorm variants,
since it is able to get rid of the lshift/rshift for sext, as well as
replacing each shift + mask with a single op.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Matt Turner <mattst88@gmail.com>
It is fairly tricky to detect the proper conditions for using bitfield
insert, but easy to just use it up front. This removes a lot of
instructions on nvc0 when invoking the packing builtins.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Matt Turner <mattst88@gmail.com>
No need to walk through instructions in blocks we know don't contain our
registers' live ranges.
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
I always thought that the is_control_flow() -> return false check was a
bad hack, and some previous attempts to remove it have failed and have
been reverted.
The previous two patches fix some problems that caused register
coalescing to not notice some interference between registers, which the
is_control_flow() check apparently works around.
With that fixed, we can calculate interference more accurately.
total instructions in shared programs: 6261319 -> 6257917 (-0.05%)
instructions in affected programs: 346282 -> 342880 (-0.98%)
helped: 1552
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
equals() returns false for registers with different types, using it
isn't appropriate to determine whether an is overwriting a register.
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Noticed when debugging things that lead to the next patch.
On G45 (and presumably ILK) this helps register coalescing:
total instructions in shared programs: 4077373 -> 4077340 (-0.00%)
instructions in affected programs: 43751 -> 43718 (-0.08%)
helped: 52
HURT: 2
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Zero sized uniforms can exist in the list, but they don't get get any space
allocated in prog_data->params or in the param_size array, so the size
should not be set for them. This was previously fixed in:
commit: 781dc7c0e1.
However,
commit: 259f7291de
removed the fix.
Signed-off-by: Marta Lofstedt <marta.lofstedt@intel.com>
Reviewed-by: Francisco Jerez <currojerez@riseup.net>
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
If the sampler object has been deleted in the same context the binding
will have been cleared. If it has been deleted in another context, the
spec does not say what should returned. None of the other binding point
queries check for deletion in another context.
Also, as names of deleted objects are free for reuse, the current code
didn't even work reliably.
Reviewed-by: Fredrik Höglund <fredrik@kde.org>
Signed-off-by: Fredrik Höglund <fredrik@kde.org>
This adds index queries (glGet*i_v) for GL_TEXTURE_BINDING_* and
GL_SAMPLER_BINDING, as well as textue queries
(glGetTex{,ture}Parameter*) for GL_TEXTURE_TARGET.
CC: "10.6 11.0" <mesa-stable@lists.freedesktop.org>
Reviewed-by: Fredrik Höglund <fredrik@kde.org>
Signed-off-by: Fredrik Höglund <fredrik@kde.org>
Shaders that contain instruction data after an instruction with EOP could end
up parsing that as an instruction, leading to various crashes and asserts in
SB as it gets very confused if it sees for instance a loop start instruction
jumping off to some random point.
Add a couple of asserts, and print EOP bit if set in old asm printer.
Signed-off-by: Glenn Kennard <glenn.kennard@gmail.com>
Cc: <mesa-stable@lists.freedesktop.org>
Signed-off-by: Dave Airlie <airlied@redhat.com>
Cube maps are special in that they have separate teximages for each
face. We handled that by copying the data to them separately, but in
case zoffset != 0 or depth != 6 we would read off the end of the client
array or modify the wrong images.
zoffset/depth have already been verified by the time the code gets to
this stage, so no need to double-check.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Brian Paul <brianp@vmware.com>
Cc: "10.6 11.0" <mesa-stable@lists.freedesktop.org>
The NIR cursor API is exactly what we want for the builder's insertion
point. This simplifies the API, the implementation, and is actually
more flexible as well.
This required a bit of reworking of TGSI->NIR's if/loop stack handling;
we now store cursors instead of cf_node_lists, for better or worse.
v2: Actually move the cursor in the after_instr case.
v3: Take advantage of nir_instr_insert (suggested by Connor).
v4: vc4 build fixes (thanks to Eric).
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Eric Anholt <eric@anholt.net> [v1]
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com> [v4]
Acked-by: Connor Abbott <cwabbott0@gmail.com> [v4]
This patch implements a general nir_instr_insert() function that takes a
nir_cursor for the insertion point. It then reworks the existing API to
simply be a wrapper around that for compatibility.
This largely involves moving the existing code into a new function.
Suggested by Connor Abbott.
v2: Make the legacy functions static inline in nir.h (requested by
Connor Abbott).
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
Acked-by: Connor Abbott <cwabbott0@gmail.com>
We want to use this for normal instruction insertion too, not just
control flow. Generally these functions are going to be extremely
useful when working with NIR, so I want them to be widely available
without having to include a separate file.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
Acked-by: Connor Abbott <cwabbott0@gmail.com>
Jumps must be the last instruction in a block, so inserting another
instruction after a jump is illegal.
Previously, we only checked this when the new instruction being inserted
was a jump. This is a red herring - inserting *any* kind of instruction
after a jump is illegal.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
Acked-by: Connor Abbott <cwabbott0@gmail.com>
Previously, we used PROGRAM_ARRAY only for variables which were
arrays or matrices. But if the variable is a structure containing
an array or matrix, we need to use PROGRAM_ARRAY for that too.
Before, we failed an assertion:
state_tracker/st_glsl_to_tgsi.cpp:4900:
Assertion `src_reg->file != PROGRAM_TEMPORARY' failed.
when running the piglit test
glsl-1.20/execution/fs-const-array-of-struct-of-array.shader_test
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
The split_virtual_grfs code doesn't properly rewrite reladdr so we need to
make sure that any uniform indirects are lowered away first.
This fixes the glsl-fs-uniform-indexed-by-swizzled-vec4.shader_test in piglit
Cc: "10.6" <mesa-stable@lists.freedesktop.org>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Now that all constant locations are assigned in a single function, we can
refactor it a bit to unify things. In particular, we now handle
pull_constant_loc and push_constant_loc more similarly and we only modify
stage_prog_data->params[] in one place at the end of the function.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
driParseDebugString() doesn't have actual code to parse comma separated
lists (or any other supported options?); instead it dumbly uses strstr().
This means that INTEL_DEBUG="vec4vs" will trigger both DEBUG_VEC4VS and
DEBUG_VS, as "vs" is also a substring.
We should probably improve the driconf parsing, but for now, just rename
the option so it's usable in the meantime.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Acked-by: Jason Ekstrand <jason.ekstrand@intel.com>
Acked-by: Kristian Høgsberg <krh@bitplanet.net>
v2: use ARB_texture_multisample enable bit
Patch adds extension enable bit and enables required keywords
and builtin functions for the extension.
Signed-off-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Marta Lofstedt <marta.lofstedt@intel.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
This patch creates a new macro, FETCH_COMPRESSED - similar in nature
to the other FETCH_* macros. This reduces repetition in the code that
deals with compressed textures.
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
Signed-off-by: Nanley Chery <nanley.g.chery@intel.com>
In agreement with the coding style, functions that aren't directly visible
to the GL API should prefer the use of bool over GLboolean.
Suggested-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
Signed-off-by: Nanley Chery <nanley.g.chery@intel.com>
Remove redundant checks and comments by grouping our calculations for
align_w and align_h wherever possible.
v2: reintroduce brw.
don't include functional changes.
don't adjust function parameters or create a new function.
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
Signed-off-by: Nanley Chery <nanley.g.chery@intel.com>
An ASTC block takes up 16 bytes for all block width and height configurations.
This size is not integrally divisible by all ASTC block widths. Therefore cpp
is changed to mean bytes per block if the texture is compressed.
Because the original definition was bytes per block divided by block width, all
references to the mipmap width must be divided the block width. This keeps the
address calculation formulas consistent. For example, the units for miptree_level
x_offset and miptree total_width has changed from pixels to blocks.
v2: reuse preexisting ALIGN_NPOT macro located in an i965 driver file.
v3: move ALIGN_NPOT into seperate commit.
simplify cpp assignment in copy_image_with_blitter().
update miptree width and offset variables in: intel_miptree_copy_slice(),
intel_miptree_map_gtt(), and brw_miptree_layout_texture_3d().
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
Signed-off-by: Nanley Chery <nanley.g.chery@intel.com>
In agreement with commit 4ab8d59a23, vertical alignment values are equal to
four times the block height on Gen9+.
v2: add newlines to separate declarations, statments, and comments.
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Neil Roberts <neil@linux.intel.com>
Signed-off-by: Nanley Chery <nanley.g.chery@intel.com>
ALIGN is changed to ALIGN_NPOT because alignment values are sometimes not
powers of two when working with ASTC.
v2: handle texture arrays and LDR-only systems.
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
Signed-off-by: Nanley Chery <nanley.g.chery@intel.com>
Aligning with a non-power-of-two number is a general task that can be used in
various places. This commit is required for the next one.
v2: add greater than 0 assertion (Anuj).
convert the macro to a static inline function.
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
Signed-off-by: Nanley Chery <nanley.g.chery@intel.com>
ALIGN and ROUND_DOWN_TO both require that the alignment value passed
into the macro be a power of two in the comments. Using software assertions
verifies this to be the case.
v2: use static inline functions instead of gcc-specific statement expressions (Brian).
v3: fix indendation (Brian).
v4: add greater than zero requirement (Anuj).
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
Signed-off-by: Nanley Chery <nanley.g.chery@intel.com>
Define two-thirds of the 2D Intel ASTC surface formats (LDR-only). This allows
a 1-to-1 mapping from the mesa format to the Intel format.
ASTC textures will default to being processed in LDR mode. If there is
hardware support for HDR/Full mode and the texture is not sRGB, add the
format bit necessary to process it in HDR/Full mode.
v2: remove extra newlines.
v3: follow existing coding style in translate_tex_format().
v4: expound on the GEN9_SURFACE_ASTC_HDR_FORMAT_BIT comment.
update SF table - ASTC is actually supported in Gen8.
v5: conform the ASTC MESA_FORMAT enums to the existing naming convention.
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
Signed-off-by: Nanley Chery <nanley.g.chery@intel.com>
This is necesary to initialize the gl_texture_image struct.
From the KHR_texture_compression_astc_ldr spec:
"Added to Section 3.8.6, Compressed Texture Images
Add the tokens specified above to Table 3.16, Compressed Internal Formats.
In all cases, the base internal format will be RGBA. The encoding allows
images to be encoded with fewer channels, but this is always presented as
RGBA to the sampler."
v2. use _mesa_is_astc_format().
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
Signed-off-by: Nanley Chery <nanley.g.chery@intel.com>
The ASTC spec was revised as follows:
Revision 2, April 28, 2015 - added CompressedTex{Sub,}Image3D to
commands accepting ASTC format tokens in the New Tokens section [...].
Support only exists in the HDR submode:
Add a second new column "3D Tex." which is empty for all non-ASTC
formats. If only the LDR profile is supported by the implementation,
this column is also empty for all ASTC formats. If both the LDR and HDR
profiles are supported only, this column is checked for all ASTC
formats.
LDR-only systems should generate an INVALID_OPERATION error when
attempting to call CompressedTexImage3D with the TEXTURE_3D target.
v2. return the proper error for LDR-only systems.
v3. update is_astc_format().
v4. use _mesa_is_astc_format().
v5. place logic in _mesa_target_can_be_compressed.
v6. fix issues handling ASTC formats.
Reviewed-by: Chad Versace <chad.versace@intel.com>
Signed-off-by: Nanley Chery <nanley.g.chery@intel.com>
In agreement with the ASTC spec, this makes calls to TexImage*D unsuccessful.
Implied by the spec, Generate[Texture]Mipmap and [Copy]Tex[Sub]Image*D calls
must be unsuccessful as well.
v2. actually force attempts to compress online to fail.
v3. indentation (Matt).
v4. update copytexture_error_check to account for CopyTexImage*D (Chad).
Reviewed-by: Chad Versace <chad.versace@intel.com>
Signed-off-by: Nanley Chery <nanley.g.chery@intel.com>
v2: correct the spelling of the sRGB variants.
remove spaces around "=" when setting the enum value.
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
Signed-off-by: Nanley Chery <nanley.g.chery@intel.com>
Define the mesa formats and make changes necessary for compilation
without errors. Also add support for _mesa_get_srgb_format_linear().
v2. conform the ASTC MESA_FORMAT enums to the existing naming convention.
v3. remove ASTC cases for _mesa_get_uncompressed_format(). This function is
only used for generating mipmaps - something ASTC formats do not support
due to lack of online compression.
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
Signed-off-by: Nanley Chery <nanley.g.chery@intel.com>
If the packet encoding is defined in the same format as register definitions,
the python script can process them automatically and the parser support
becomes trivial.
Acked-by: Christian König <christian.koenig@amd.com>
Acked-by: Alex Deucher <alexander.deucher@amd.com>
This adds trace points to all IBs and the parser prints them and also
prints which trace points were reached (executed) by the CP.
This can help pinpoint a problematic packet, draw call, etc.
Acked-by: Christian König <christian.koenig@amd.com>
Acked-by: Alex Deucher <alexander.deucher@amd.com>
GPU hang detection must be enabled by setting: GALLIUM_DDEBUG=[timeout in ms]
This may print too much information that we might not understand yet,
but some of the bits are very useful.
Acked-by: Christian König <christian.koenig@amd.com>
Acked-by: Alex Deucher <alexander.deucher@amd.com>
This makes writing a good IB parser a lot easier.
It generates 2 tables:
- packet3 table
- register table with all registers, fields, and named values
Acked-by: Christian König <christian.koenig@amd.com>
Acked-by: Alex Deucher <alexander.deucher@amd.com>
Instruction encoding isn't needed in Mesa.
The border color address registers were duplicated.
Acked-by: Christian König <christian.koenig@amd.com>
Acked-by: Alex Deucher <alexander.deucher@amd.com>
v2: lots of improvements
This is like identity or trace, but simpler. It doesn't wrap most states.
Run with:
GALLIUM_DDEBUG=1000 [executable]
where "executable" is the app and "1000" is in miliseconds, meaning that
the context will be considered hung if a fence fails to signal in 1000 ms.
If that happens, all shaders, context states, bound resources, draw
parameters, and driver debug information (if any) will be dumped into:
/home/$username/dd_dumps/$processname_$pid_$index.
Note that the context is flushed after every draw/clear/copy/blit operation
and then waited for to find the exact call that hangs.
You can also do:
GALLIUM_DDEBUG=always
to do the dumping after every draw/clear/copy/blit operation without
flushing and waiting.
Examples of driver states that can be dumped are:
- Hardware status registers saying which hw block is busy (hung).
- Disassembled shaders in a human-readable form.
- The last submitted command buffer in a human-readable form.
v2: drop pipe-loader changes, drop SConscript
rename dd.h -> dd_pipe.h
Acked-by: Christian König <christian.koenig@amd.com>
Acked-by: Alex Deucher <alexander.deucher@amd.com>
This works if drivers upsample on upload (like all radeon ones do).
The alternative is an unexpected GL error from anything calling
_mesa_update_state and possibly other issues.
Cc: 10.6 11.0 <mesa-stable@lists.freedesktop.org>
Reviewed-by: Dave Airlie <airlied@redhat.com>
Otherwise we get:
warning: 'num_user_sgprs' may be used uninitialized in this function
...
Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>
Signed-off-by: Marek Olšák <marek.olsak@amd.com>
Patch refactors existing parameters check to first check common enums
between desktop GL and GLES 3.1 and modifies get_tex_level_parameter_image
to be compatible with enums specified in 3.1.
v2: remove extra is_gles31() checks (suggested by Ilia)
Signed-off-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com> (v1)
Reviewed-by: Marta Lofstedt <marta.lofstedt@intel.com> (v1)
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
According to OpenGL ES specification section 7.12,
GL_COMPUTE_WORK_GROUP_SIZE, is supported by the
glGetProgramiv function.
Signed-off-by: Marta Lofstedt <marta.lofstedt@intel.com>
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
According to the OpenGL ES 3.1 specification chapter 17, the
MAX_COMPUTE_WORK_GROUP_COUNT and MAX_COMPUTE_WORK_GROUP_SIZE
is available for glGetIntegeri_v.
Signed-off-by: Marta Lofstedt <marta.lofstedt@linux.intel.com>
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
commit 26c549e69d
Author: Nanley Chery <nanley.g.chery@intel.com>
Date: Fri Jul 31 10:26:36 2015 -0700
mesa/formats: remove compressed formats from matching function
caused a regression in my CTS testing, this looks like a clear
thinko.
Reviewed-by: Nanley Chery <nanley.g.chery@intel.com>
sSigned-off-by: Dave Airlie <airlied@redhat.com>
I used this as some testing ground for investigating some compiler
bits initially (e.g. lrint calls etc.), figured I could do much better
in the end just for fun...
This is mathematically equivalent, but uses some tricks to avoid
doubles and also replaces some float math with ints. Good for another
performance doubling or so. As a side note, some quick tests show that
llvm's loop vectorizer would be able to properly vectorize this version
(which it failed to do earlier due to doubles, producing a mess), giving
another 3 times performance increase with sse2 (more with sse4.1), but this
may not apply to mesa.
No piglit change.
Acked-by: Marek Olšák <marek.olsak@amd.com>
This code (lifted straight from the extension) was doing things the most
inefficient way you could think of.
This drops some of the more expensive float operations, in particular
- int-cast floors (pointless, values always positive)
- 2 raised to (signed) integers (replace with simple exponent manipulation),
getting rid of a misguided comment in the process (implement with table...)
- float division (replace with mul of reverse of those exponents)
This is like 3 times faster (measured for float3_to_rgb9e5), though it depends
(e.g. llvm is clever enough to replace exp2 with ldexp whereas gcc is not,
division is not too bad on cpus with early-exit divs).
Note that keeping the double math for now (float x + 0.5), as the results may
otherwise differ.
Acked-by: Marek Olšák <marek.olsak@amd.com>
GetTexImage can read to stencil8 but only from
a stencil or depthstencil textures.
This fixes a bunch of failures in CTS
GL33-CTS.gtf32.GL3Tests.packed_pixels
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Cc: "11.0" <mesa-stable@lists.freedesktop.org>
Signed-off-by: Dave Airlie <airlied@redhat.com>
Enables _mesa_target_can_be_compressed to return the appropriate GL error
depending on it's inputs. Use the parameter to return the appropriate GL error
for ETC2 formats on GLES3.
Suggested-by: Chad Versace <chad.versace@intel.com>
Reviewed-by: Chad Versace <chad.versace@intel.com>
Signed-off-by: Nanley Chery <nanley.g.chery@intel.com>
All compressed formats return GL_FALSE and there isn't any evidence to
support that this behaviour would change. Remove all switch cases for
compressed formats.
v2. Since the exhaustive switch is removed, add a gtest to ensure
all formats are handled.
v3. Ensure that GL_NO_ERROR is set before returning.
v4. Fix an arg to _mesa_uncompressed_format_to_type_and_comps();
fix formatting and misc improvements (Chad).
Reviewed-by: Chad Versace <chad.versace@intel.com>
Signed-off-by: Nanley Chery <nanley.g.chery@intel.com>
We currently check that our format info table is sane during context
initialization in debug builds. Perform this check during
`make check` instead. This enables format testing in release builds
and removes the requirement of an exhuastive switch for
_mesa_uncompressed_format_to_type_and_comps().
v2. indentation and conditional inclusion fixes (Chad).
allow tests to continue running if any format fails
and display the failing format name.
Reviewed-by: Chad Versace <chad.versace@intel.com>
Signed-off-by: Nanley Chery <nanley.g.chery@intel.com>
I intend to remove nir_builder::cf_node_list, so I can't have this code
poking at it directly. The proper way is to set the insertion point and
then simply insert things there.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Eric Anholt <eric@anholt.net>
I intend to remove nir_builder::cf_node_list, so I can't have this code
poking at it directly. The proper way is to set the insertion point and
then simply insert things there.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Eric Anholt <eric@anholt.net>
This makes it easy for NIR passes to inspect what kind of shader they're
operating on.
Thanks to Michel Dänzer for helping me figure out where TGSI stores the
shader stage information.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
The comment above move_uniform_array_access_to_pull_constants was
completely bogus because it has nothing to do with lowering instructions.
Instead, it's assiging locations of pull constants.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Previously, we treated the entire UNIFORM file as if it had two elements:
One for direct things and one for indirect. This is substantially
different from how the old visitor code handled it where each element was
effectively its own uniform. This commit makes the NIR path more like the
old ir_visitor path where each uniform is separate. This should allow us
to more easily make decisions about what to push.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
In the i965 backend, we want to be able to "pull apart" the uniforms and
push some of them into the shader through a different path. In order to do
this effectively, we need to know which variable is actually being referred
to by a given uniform load. Previously, it was completely flattened by
nir_lower_io which made things difficult. This adds more information to
the intrinsic to make this easier for us.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Previously, there were four type_size() functions in play - the i965
compiler backend defined scalar and vec4 type_size() functions, and
nir_lower_io contained its own similar functions.
In fact, the i965 driver used nir_lower_io() and then looped over the
components using its own type_size - meaning both were in play. The
two are /basically/ the same, but not exactly in obscure cases like
subroutines and images.
This patch removes nir_lower_io's functions, and instead makes the
driver supply a function pointer. This gives the driver ultimate
flexibility in deciding how it wants to count things, reduces code
duplication, and improves consistency.
v2 (Jason Ekstrand):
- One side-effect of passing in a function pointer is that nir_lower_io is
now aware of and properly allocates space for image uniforms, allowing
us to drop hacks in the backend
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
v2 Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
If there are no parameters, we don't need to create a nir_variable to
hold them...and allocating an array of length 0 is pretty bogus.
Should avoid i965 backend assertions in future patches Jason and I are
working on.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
I want to use C function pointers to these, and they don't use anything
in the visitor classes anyway.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
This way they don't implicitly increment the uniforms variable and don't
have to be called in-sequence during uniform setup.
Reviewed-by: Francisco Jerez <currojerez@riseup.net>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
The commit:
commit b49371b8ed
Author: Connor Abbott <cwabbott0@gmail.com>
AuthorDate: Tue Jul 21 19:54:18 2015 -0700
nir: move control flow modification to its own file
split out some control flow related APIs into a separate header, but did
not update drivers.
Signed-off-by: Rob Clark <robclark@freedesktop.org>
The commit:
commit 8e0d4ef341
Author: Kenneth Graunke <kenneth@whitecape.org>
AuthorDate: Thu Aug 6 18:18:40 2015 -0700
nir: Delete the nir_function_impl::start_block field.
removed the start_block field without fixing up drivers..
Signed-off-by: Rob Clark <robclark@freedesktop.org>
Connor introduced this helper recently; we should use it here too.
I had to move the function earlier in the file for it to be available.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
This gives us some testing of it. Also, the old nir_cf_node_remove()
wasn't handling phi nodes correctly and was calling cleanup_cf_node()
too late.
Signed-off-by: Connor Abbott <connor.w.abbott@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
These will help us do a number of things, including:
- Early return elimination.
- Dead control flow elimination.
- Various optimizations, such as replacing:
if (foo) {
...
}
if (!foo) {
...
}
with:
if (foo) {
...
} else {
...
}
Signed-off-by: Connor Abbott <connor.w.abbott@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
This is a helper that will be shared between the new control flow
insertion and modification code.
Signed-off-by: Connor Abbott <connor.w.abbott@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
For now, it allows us to refactor the control flow insertion API's so
that there's a single entrypoint (with some wrappers). More importantly,
it will allow us to reduce the combinatorial explosion in the extract
function. There, we need to specify two points to extract, which may be
at the beginning of a block, the end of a block, or in the middle of a
block. And then there are various wrappers based off of that (before a
control flow node, before a control flow list, etc.). Rather than having
9 different functions, we can have one function and push the actual
logic of determining which variant to use down to the split function,
which will be shared with nir_cf_node_insert().
In the future, we may want to make the instruction insertion API's as
well as the builder use this, but that's a future cleanup.
Signed-off-by: Connor Abbott <connor.w.abbott@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
When we insert a single basic block A into another basic block B, we
will split B into C and D, insert A in the middle, and then splice
together C, A, and D. When we splice together C and A, we need to move
the successors of A into C -- except A has no successors, since it
hasn't been inserted yet. So in move_successors(), we need to handle the
case where the block whose successors are to be moved doesn't have any
successors. Fixing link_blocks() here prevents a segfault and makes it
work correctly.
Signed-off-by: Connor Abbott <connor.w.abbott@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
We may delete a control flow node which contains structured jumps to
other parts of the program. We need to remove the jump as a predecessor,
as well as remove any phi node sources which reference it. Right now,
the same problem exists for blocks that don't end in a jump instruction,
but with the new API it shouldn't be an issue, since blocks that don't
end in a jump must either point to another block in the same extracted
CF list or not point to anything at all.
Signed-off-by: Connor Abbott <connor.w.abbott@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Unlike calling nir_instr_remove(), calling nir_cf_node_remove() (and
later in the series, the nir_cf_list_delete()) implies that you're
removing instructions that may still have uses, except those
instructions are never executed so any uses will be undefined. When
cleaning up a CF node for deletion, we must clean up any uses of the
deleted instructions by making them point to undef instructions instead.
Signed-off-by: Connor Abbott <connor.w.abbott@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
In particular, handle the case where the earlier block ends in a jump
and the later block is empty. In that case, we want to preserve the jump
and remove any traces of the later block. Before, we would only hit this
case when removing a control flow node after a jump, which wasn't a
common occurance, but we'll need it to handle inserting a control flow
list which ends in a jump, which should be more common/useful.
Signed-off-by: Connor Abbott <connor.w.abbott@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Before, we would only split a block with a jump at the end if we were
inserting something after a block with a jump, which never happened in
practice. But now, we want to use this to extract control flow lists
which may end in a jump, in which case we really need to do the correct
patching up. As a side effect, when removing jumps we now correctly
insert undef phi sources in some corner cases, which can't hurt.
Signed-off-by: Connor Abbott <connor.w.abbott@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Before, the process of removing a jump and wiring up the remaining block
correctly was atomic, but with the new control flow modification it's
split into two parts: first, we extract the jump, which creates a new
block with re-wired successors as well as a free-floating jump, and then
we delete the control flow containing the jump, which removes the entry
in the predecessors and any phi node sources. Split up
nir_handle_remove_jumps() to accomodate this, and add the missing
support for removing phi node sources.
Signed-off-by: Connor Abbott <connor.w.abbott@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
We want to start reworking and expanding this code, but it'll be a lot
easier to do once we disentangle it from the rest of the stuff in nir.c.
Unfortunately, there are a few unavoidable dependencies in nir.c on
methods we'd rather not expose publicly, since if not used in very
specific situations they can cause Bad Things (tm) to happen. Namely, we
need to do some magical control flow munging when adding/removing jumps.
In the future, we may disallow adding/removing jumps in
nir_instr_insert_*() and nir_instr_remove(), and use separate functions
that are part of the control flow modification code, but for now we
expose them and put them in a separate, private header.
Signed-off-by: Connor Abbott <connor.w.abbott@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
cleanup_cf_node() is part of the control flow modification code, which
we're going to split into its own file, but remove_defs_uses() is an
internal function used by nir_instr_remove(). Break the dependency by
making cleanup_cf_node() use nir_instr_remove() instead, which simply
calls remove_defs_uses() and then removes the instruction from the list.
nir_instr_remove() does do extra things for jumps, though, so we avoid
calling it on jumps which matches the previous behavior (this will be
fixed later in the series).
Signed-off-by: Connor Abbott <connor.w.abbott@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
It was being used to initialize function impls and loops, even though
it's really a control flow modification helper. It's pretty trivial, so
just inline it to avoid the dependency.
Signed-off-by: Connor Abbott <connor.w.abbott@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
It's simply the first nir_cf_node in the nir_function_impl::body list,
which is easy enough to access - we don't to store a pointer to it
explicitly. Removing it means we don't need to maintain the pointer
when, say, splitting the start block when modifying control flow.
Thanks to Connor Abbott for suggesting this.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Only uncompressed formats have a non-void type and actual
components per pixel. Rename _mesa_format_to_type_and_comps
to _mesa_uncompressed_format_to_type_and_comps and require
callers to check if the format is not compressed.
v2. include compressed format cases to avoid gcc warnings (Chad).
Reviewed-by: Chad Versace <chad.versace@intel.com>
Signed-off-by: Nanley Chery <nanley.g.chery@intel.com>
On the older platforms where we don't have logical contexts preserving
state across batches, we emit the invariant state setup on every batch
using the brw_invariant_state atom. This includes the pipeline selection
which is cached with the introduction of
commit 0e0e23ef53
Author: Jordan Justen <jordan.l.justen@intel.com>
Date: Wed Apr 22 11:43:50 2015 -0700
i965/state: Emit pipeline select when changing pipelines
However, we do not reset the cache between batches on context-less
platforms resulting in us not setting the pipeline selection and can
cause GPU hangs if a media pipelined was loaded in the meantime (e.g.
mixing mplayer/gstreamer using libva and gnome-shell). A simple solution
is to just forcibly re-emit the pipeline select along with the invariant
state and reset the cache at that point.
Reported-and-tested-by: Tomasz C. <tomaszc@o2.pl>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=91254
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Jordan Justen <jordan.l.justen@intel.com>
Cc: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Cc: "10.6 11.0" <mesa-stable@lists.freedesktop.org>
This reverts commit 567394112d.
It regressed performance. It looks like smaller IBs are better, because
the GPU goes idle quicker and there is less waiting for buffers and fences.
Cc: 11.0 <mesa-stable@lists.freedesktop.org>
When the edge flag element is enabled then the elements are slightly
reordered so that the edge flag is always the last one. This was
confusing the code to upload the 3DSTATE_VF_INSTANCING state because
that is uploaded with a separate loop which has an instruction for
each element. The indices used in these instructions weren't taking
into account the reordering so the state would be incorrect.
v2: Use nr_elements instead of brw->vb.nr_enabled so that it will cope
when gl_VertexID is used.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=91292
Cc: <mesa-stable@lists.freedesktop.org>
Reviewed-by: Ben Widawsky <ben@bwidawsk.net>
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Tested-by: Mark Janes <mark.a.janes@intel.com>
The edge flag data on Gen6+ is passed through the fixed function hardware as
an extra attribute. According to the PRM it must be the last valid
VERTEX_ELEMENT structure. However if the vertex ID is also used then another
extra element is added to source the VID. This made it so the vertex ID is in
the wrong register in the vertex shader and the edge attribute is no longer in
the last element.
v2: Also implement for BDW+
v3 [by Ben]: Remove 10.5 tag. Too late.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=84677
Cc: <mesa-stable@lists.freedesktop.org>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Tested-by: Ben Widawsky <ben@bwidawsk.net>
Tested-by: Mark Janes <mark.a.janes@intel.com>
GL_OES_gpu_shader5 not started (based on parts of GL_ARB_gpu_shader5, which is done for some drivers)
GL_OES_primitive_bounding box not started
GL_OES_sample_shading not started (based on parts of GL_ARB_sample_shading, which is done for some drivers)
GL_OES_sample_variables not started (based on parts of GL_ARB_sample_shading, which is done for some drivers)
GL_OES_shader_image_atomic not started (based on parts of GL_ARB_shader_image_load_store, which is done for some drivers)
GL_OES_shader_io_blocks not started (based on parts of GLSL 1.50, which is done)
GL_OES_shader_multisample_interpolation not started (based on parts of GL_ARB_gpu_shader5, which is done)
GL_OES_tessellation_shader not started (based on GL_ARB_tessellation_shader, which is done for some drivers)
GL_OES_texture_border_clamp not started (based on GL_ARB_texture_border_clamp, which is done)
GL_OES_texture_buffer not started (based on GL_ARB_texture_buffer_object, GL_ARB_texture_buffer_range, and GL_ARB_texture_buffer_object_rgb32 that are all done)
GL_OES_texture_cube_map_array not started (based on GL_ARB_texture_cube_map_array, which is done for all drivers)
GL_OES_texture_stencil8 DONE (all drivers that support GL_ARB_texture_stencil8)
GL_OES_texture_storage_multisample_2d_array DONE (all drivers that support GL_ARB_texture_multisample)
More info about these features and the work involved can be found at
<ahref="relnotes/11.1.2.html">Mesa 11.1.2</a> is released.
This is a bug-fix release.
</p>
<h2>January 22, 2016</h2>
<p>
<ahref="relnotes/11.0.9.html">Mesa 11.0.9</a> is released.
This is a bug-fix release.
<br>
NOTE: It is anticipated that 11.0.9 will be the final release in the 11.0
series. Users of 11.0 are encouraged to migrate to the 11.1 series in order
to obtain future fixes.
</p>
<h2>January 13, 2016</h2>
<p>
<ahref="relnotes/11.1.1.html">Mesa 11.1.1</a> is released.
This is a bug-fix release.
</p>
<h2>December 21, 2015</h2>
<p>
<ahref="relnotes/11.0.8.html">Mesa 11.0.8</a> is released.
This is a bug-fix release.
</p>
<h2>December 15, 2015</h2>
<p>
<ahref="relnotes/11.1.0.html">Mesa 11.1.0</a> is released. This is a new
development release. See the release notes for more information about
the release.
</p>
<h2>December 9, 2015</h2>
<p>
<ahref="relnotes/11.0.7.html">Mesa 11.0.7</a> is released.
This is a bug-fix release.
</p>
<p>
Mesa demos 8.3.0 is also released.
See the <ahref="http://lists.freedesktop.org/archives/mesa-announce/2015-December/000191.html">announcement</a> for more information about the release.
You can download it from <ahref="ftp://ftp.freedesktop.org/pub/mesa/demos/8.3.0/">ftp.freedesktop.org/pub/mesa/demos/8.3.0/</a>.
</p>
<h2>November 21, 2015</h2>
<p>
<ahref="relnotes/11.0.6.html">Mesa 11.0.6</a> is released.
This is a bug-fix release.
</p>
<h2>November 11, 2015</h2>
<p>
<ahref="relnotes/11.0.5.html">Mesa 11.0.5</a> is released.
This is a bug-fix release.
</p>
<h2>October 24, 2015</h2>
<p>
<ahref="relnotes/11.0.4.html">Mesa 11.0.4</a> is released.
This is a bug-fix release.
</p>
<h2>October 10, 2015</h2>
<p>
<ahref="relnotes/11.0.3.html">Mesa 11.0.3</a> is released.
This is a bug-fix release.
</p>
<h2>October 3, 2015</h2>
<p>
<ahref="relnotes/10.6.9.html">Mesa 10.6.9</a> is released.
This is a bug-fix release.
<br>
NOTE: It is anticipated that 10.6.9 will be the final release in the 10.6
series. Users of 10.6 are encouraged to migrate to the 11.0 series in order
to obtain future fixes.
</p>
<h2>September 28, 2015</h2>
<p>
<ahref="relnotes/11.0.2.html">Mesa 11.0.2</a> is released.
This is a bug-fix release.
</p>
<h2>September 26, 2015</h2>
<p>
<ahref="relnotes/11.0.1.html">Mesa 11.0.1</a> is released.
This is a bug-fix release.
</p>
<h2>September 20, 2015</h2>
<p>
<ahref="relnotes/10.6.8.html">Mesa 10.6.8</a> is released.
This is a bug-fix release.
</p>
<h2>September 12, 2015</h2>
<p>
<ahref="relnotes/11.0.0.html">Mesa 11.0.0</a> is released. This is a new
development release. See the release notes for more information about
the release.
</p>
<h2>September 10, 2015</h2>
<p>
<ahref="relnotes/10.6.7.html">Mesa 10.6.7</a> is released.
This is a bug-fix release.
</p>
<h2>September 4, 2015</h2>
<p>
<ahref="relnotes/10.6.6.html">Mesa 10.6.6</a> is released.
This is a bug-fix release.
</p>
<h2>August 22, 2015</h2>
<p>
<ahref="relnotes/10.6.5.html">Mesa 10.6.5</a> is released.
This is a bug-fix release.
</p>
<h2>August 11 2015</h2>
<h2>August 11, 2015</h2>
<p>
<ahref="relnotes/10.6.4.html">Mesa 10.6.4</a> is released.
This is a bug-fix release.
</p>
<h2>July 26 2015</h2>
<h2>July 26, 2015</h2>
<p>
<ahref="relnotes/10.6.3.html">Mesa 10.6.3</a> is released.
This is a bug-fix release.
</p>
<h2>July 11 2015</h2>
<h2>July 11, 2015</h2>
<p>
<ahref="relnotes/10.6.2.html">Mesa 10.6.2</a> is released.
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=90925">Bug 90925</a> - "high fidelity": Segfault in _mesa_program_resource_find_name</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=91254">Bug 91254</a> - (regresion) video using VA-API on Intel slow and freeze system with mesa 10.6 or 10.6.1</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=91292">Bug 91292</a> - [BDW+] glVertexAttribDivisor not working in combination with glPolygonMode</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=91673">Bug 91673</a> - Segfault when calling glTexSubImage2D on storage texture to bound FBO</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=91726">Bug 91726</a> - R600 asserts in tgsi_cmp/make_src_for_op3</li>
</ul>
<h2>Changes</h2>
<p>Chris Wilson (2):</p>
<ul>
<li>i965: Prevent coordinate overflow in intel_emit_linear_blit</li>
<li>i965: Always re-emit the pipeline select during invariant state emission</li>
</ul>
<p>Daniel Scharrer (1):</p>
<ul>
<li>mesa: add missing queries for ARB_direct_state_access</li>
</ul>
<p>Dave Airlie (8):</p>
<ul>
<li>mesa/arb_gpu_shader_fp64: add support for glGetUniformdv</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=90621">Bug 90621</a> - Mesa fail to build from git</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=91526">Bug 91526</a> - World of Warcraft (on Wine) has UI corruption with nouveau</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=91719">Bug 91719</a> - [SNB,HSW,BYT] dEQP regressions associated with using NIR for vertex shaders</li>
</ul>
<h2>Changes</h2>
<p>Alejandro Piñeiro (1):</p>
<ul>
<li>i965/vec4: fill src_reg type using the constructor type parameter</li>
</ul>
<p>Antia Puentes (1):</p>
<ul>
<li>i965/vec4: Fix saturation errors when coalescing registers</li>
</ul>
<p>Emil Velikov (2):</p>
<ul>
<li>docs: add sha256 checksums for 10.6.7</li>
<li>cherry-ignore: add commit non applicable for 10.6</li>
</ul>
<p>Hans de Goede (4):</p>
<ul>
<li>nv30: Fix creation of scanout buffers</li>
<li>nv30: Implement color resolve for msaa</li>
<li>nv30: Fix max width / height checks in nv30 sifm code</li>
<li>nv30: Disable msaa unless requested from the env by NV30_MAX_MSAA</li>
</ul>
<p>Ian Romanick (2):</p>
<ul>
<li>mesa: Pass the type to _mesa_uniform_matrix as a glsl_base_type</li>
<li>mesa: Don't allow wrong type setters for matrix uniforms</li>
</ul>
<p>Ilia Mirkin (5):</p>
<ul>
<li>st/mesa: don't fall back to 16F when 32F is requested</li>
<li>nvc0: always emit a full shader colormask</li>
<li>nvc0: remove BGRA4 format support</li>
<li>st/mesa: avoid integer overflows with buffers >= 512MB</li>
<li>nv50, nvc0: fix max texture buffer size to 128M elements</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=38109">Bug 38109</a> - i915 driver crashes if too few vertices are submitted (Mesa 7.10.2)</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=55552">Bug 55552</a> - Compile errors with --enable-mangling</li>
@@ -45,8 +45,6 @@ because compatibility contexts are not supported.
<ul>
<ul>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=91993">Bug 91993</a> - Graphical glitch in Astromenace (open-source game).</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=92214">Bug 92214</a> - Flightgear crashes during splashboot with R600 driver, LLVM 3.7.0 and mesa 11.0.2</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=38109">Bug 38109</a> - i915 driver crashes if too few vertices are submitted (Mesa 7.10.2)</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=49779">Bug 49779</a> - Extra line segments in GL_LINE_LOOP</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=55552">Bug 55552</a> - Compile errors with --enable-mangling</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=71789">Bug 71789</a> - [r300g] Visuals not found in (default) depth = 24</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=79783">Bug 79783</a> - Distorted output in obs-studio where other vendors "work"</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=80821">Bug 80821</a> - When LIBGL_ALWAYS_SOFTWARE is set, KHR_create_context is not supported</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=81174">Bug 81174</a> - Gallium: GL_LINE_LOOP broken with more than 512 points</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=83508">Bug 83508</a> - [UBO] Assertion for array of blocks</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=84677">Bug 84677</a> - Triangle disappears with glPolygonMode GL_LINE</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=86720">Bug 86720</a> - [radeon] Europa Universalis 4 freezing during game start (10.3.3+, still broken on 11.0.2)</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=89014">Bug 89014</a> - PIPE_QUERY_GPU_FINISHED is not acting as expected on SI</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=90175">Bug 90175</a> - [hsw bisected][PATCH] atomic counters doesn't work for a binding point different to zero</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=90348">Bug 90348</a> - Spilling failure of b96 merged value</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=90631">Bug 90631</a> - Compilation failure for fragment shader with many branches on Sandy Bridge</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=90734">Bug 90734</a> - glBufferSubData is corrupting data when buffer is > 32k</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=90887">Bug 90887</a> - PhiMovesPass in register allocator broken</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=91044">Bug 91044</a> - piglit spec/egl_khr_create_context/valid debug flag gles* fail</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=91254">Bug 91254</a> - (regresion) video using VA-API on Intel slow and freeze system with mesa 10.6 or 10.6.1</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=91292">Bug 91292</a> - [BDW+] glVertexAttribDivisor not working in combination with glPolygonMode</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=91342">Bug 91342</a> - Very dark textures on some objects in indoors environments in Postal 2</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=91526">Bug 91526</a> - World of Warcraft (on Wine) has UI corruption with nouveau</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=91551">Bug 91551</a> - DXTn compressed normal maps produce severe artifacts on all NV5x and NVDx chipsets</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=91596">Bug 91596</a> - EGL_KHR_gl_colorspace (v2) causes problem with Android-x86 GUI</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=91716">Bug 91716</a> - [bisected] piglit.shaders.glsl-vs-int-attrib regresses on 32 bit BYT, HSW, IVB, SNB</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=91719">Bug 91719</a> - [SNB,HSW,BYT] dEQP regressions associated with using NIR for vertex shaders</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=91726">Bug 91726</a> - R600 asserts in tgsi_cmp/make_src_for_op3</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=91780">Bug 91780</a> - Rendering issues with geometry shader</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=91785">Bug 91785</a> - make check DispatchSanity_test.GLES31 regression</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=91788">Bug 91788</a> - [HSW Regression] Synmark2_v6 Multithread performance case FPS reduced by 36%</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=91847">Bug 91847</a> - glGenerateTextureMipmap not working (no errors) unless glActiveTexture(GL_TEXTURE1) is called before</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=91857">Bug 91857</a> - Mesa 10.6.3 linker is slow</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=91881">Bug 91881</a> - regression: GPU lockups since mesa-11.0.0_rc1 on RV620 (r600) driver</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=91898">Bug 91898</a> - src/util/mesa-sha1.c:250:25: fatal error: openssl/sha.h: No such file or directory</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=92052">Bug 92052</a> - nir/nir_builder.h:79: error: expected primary-expression before ‘.’ token</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=92054">Bug 92054</a> - make check gbm-symbols-check regression</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=92066">Bug 92066</a> - [ILK,G45,regression] New assertion on BRW_MAX_MRF breaks ilk and g45</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=92072">Bug 92072</a> - Wine breakage since d082c5324 (st/mesa: don't call st_validate_state in BlitFramebuffer)</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=92122">Bug 92122</a> - [bisected, cts] Regression with Assault Android Cactus</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=92124">Bug 92124</a> - shader_query.cpp:841:34: error: ‘strndup’ was not declared in this scope</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=92183">Bug 92183</a> - linker.cpp:3187:46: error: ‘strtok_r’ was not declared in this scope</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=92214">Bug 92214</a> - Flightgear crashes during splashboot with R600 driver, LLVM 3.7.0 and mesa 11.0.2</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=92221">Bug 92221</a> - Unintended code changes in _mesa_base_tex_format commit</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=92265">Bug 92265</a> - Black windows in weston after update mesa to 11.0.2-1</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=92363">Bug 92363</a> - [BSW/BDW] ogles1conform Gets test fails</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=92437">Bug 92437</a> - osmesa: Expose GL entry points for Windows build, via .def file</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=92438">Bug 92438</a> - Segfault in pushbuf_kref when running the android emulator (qemu) on nv50</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=92621">Bug 92621</a> - [G965 ILK G45] Regression: 24 piglit regressions in glsl-1.10</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=92623">Bug 92623</a> - Differences in prog_data ignored when caching fragment programs (causes hangs)</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=92634">Bug 92634</a> - gallium's vl_mpeg12_decoder does not work with st/va</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=92705">Bug 92705</a> - [clover] fail to build with llvm-svn/clang-svn 3.8</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=92709">Bug 92709</a> - "LLVM triggered Diagnostic Handler: unsupported call to function ldexpf in main" when starting race in stuntrally</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=92738">Bug 92738</a> - Randon R7 240 doesn't work on 16KiB page size platform</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=92744">Bug 92744</a> - [g965 Regression bisected] Performance regression and piglit assertions due to liveness analysis</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=92900">Bug 92900</a> - [regression bisected] About 700 piglit regressions is what could go wrong</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=92909">Bug 92909</a> - Offset/alignment issue with layout std140 and vec3</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=92985">Bug 92985</a> - Mac OS X build error "ar: no archive members specified"</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=93015">Bug 93015</a> - Tonga Elemental segfault + VM faults since radeon: implement r600_query_hw_get_result via function pointers</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=93235">Bug 93235</a> - [regression] dispatch sanity broken by GetPointerv</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=93266">Bug 93266</a> - gl_arb_shading_language_420pack does not allow binding of image variables</li>
</ul>
<h2>Changes</h2>
<li>MPEG4 decoding has been disabled by default in the VAAPI driver</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=91596">Bug 91596</a> - EGL_KHR_gl_colorspace (v2) causes problem with Android-x86 GUI</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=93628">Bug 93628</a> - Exception: attempt to use unavailable module DRM when building MesaGL 11.1.0 on windows</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=93648">Bug 93648</a> - Random lines being rendered when playing Dolphin (geometry shaders related, w/ apitrace)</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=93650">Bug 93650</a> - GL_ARB_separate_shader_objects is buggy (PCSX2)</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=93717">Bug 93717</a> - Meta mipmap generation can corrupt texture state</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=93722">Bug 93722</a> - Segfault when compiling shader with a subroutine that takes a parameter</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=93731">Bug 93731</a> - glUniformSubroutinesuiv segfaults when subroutine uniform is bound to a specific location</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=93761">Bug 93761</a> - A conditional discard in a fragment shader causes no depth writing at all</li>
@@ -152,25 +197,39 @@ if they're not installed in your system. You should be told what's missing.
<br>
<br>
<li>xf86-video-vmware: Now, once libxatracker is installed, we proceed with building and replacing the current Xorg driver. First check if your system is 32- or 64-bit. If you're building for a 32-bit system, you will not be needing the --libdir=/usr/lib64 option to autogen.
<li>xf86-video-vmware: Now, once libxatracker is installed, we proceed with
building and replacing the current Xorg driver.
First check if your system is 32- or 64-bit.
<pre>
cd $TOP/xf86-video-vmware
./autogen.sh --prefix=/usr --libdir=/usr/lib64
./autogen.sh --prefix=/usr --libdir=${LIBDIR}
make
sudo make install
</pre>
<li>vmwgfx kernel module. First make sure that any old version of this kernel module is removed from the system by issuing
print'scons: warning: Visual Studio versions prior to 2012 are known to produce incorrect code when optimizations are enabled ( https://bugs.freedesktop.org/show_bug.cgi?id=58718 )'
Some files were not shown because too many files have changed in this diff
Show More
Reference in New Issue
Block a user
Blocking a user prevents them from interacting with repositories, such as opening or commenting on pull requests or issues. Learn more about blocking a user.
