Uses the __DRIimage loader interfaces.
v2: Fix _XIOErrors when DRI3 isn't present (change by anholt). Apparently
XCB just terminates your connection if you don't check for extensions
before using them, instead of returning an error like you'd expect.
Signed-off-by: Keith Packard <keithp@keithp.com>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
Reviewed-by: Eric Anholt <eric@anholt.net>
These provide an interface between the driver and the loader to allocate
color buffers through the DRIimage extension interface rather than through a
loader-specific extension (as is used by DRI2, for instance).
The driver uses the loader 'getBuffers' interface to allocate color buffers.
The loader uses the createNewScreen2, createNewDrawable, createNewContext,
getAPIMask and createContextAttribs APIS (mostly shared with DRI2).
This interface will work with the DRI3 loader, and should also work with GBM
and other loaders so that drivers need not be customized for each new loader
interface, as long as they provide this image interface.
v2: Fix build of i915 and i965 together (by anholt)
Signed-off-by: Keith Packard <keithp@keithp.com>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
Reviewed-by: Eric Anholt <eric@anholt.net>
The __DRI_IMAGE_FORMAT codes are used by the image extension, drivers need to
be able to translate between them. Instead of duplicating this translation in
each driver, create a shared version.
Signed-off-by: Keith Packard <keithp@keithp.com>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
Instead of assuming that the size will be height * pitch, have the caller pass
in the size explicitly.
Signed-off-by: Keith Packard <keithp@keithp.com>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
This just renames them so that they can be used with the DRI3 extension
without causing too much confusion.
Signed-off-by: Keith Packard <keithp@keithp.com>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
There's only one minor functional change, for immediates the pixel offsets
are no longer added since the values are all the same for all elements in
any case (it might be better if those weren't stored as soa vectors in the
first place maybe).
Reviewed-by: Zack Rusin <zackr@vmware.com>
After adding $(DEFINES) to AM_CPPFLAGS, the __glXGetCurrentContext
wrapper function is no longer needed and causes compile errors. Using
the correct defines causes it to be a macro!
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
These tests primarilly ensure that the functions added by this extension
don't abuse other interfaces (e.g., glx_screen::query_renderer_integer)
when provided bad data.
These tests helped me find a couple small bugs in the initial
implementation.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
The enumerated values are currently allocated from Intel's range.
v2: Fix a typo. Update the list of functions to which the new enums can
be passed. The "Current" versions were previously missing. Both things
noticed by Marek.
v3: Fix typo in return type of glXQueryRendererIntegerMESA in the spec
body (noticed by Ken). Fix typo in issue #14 referencing itself instead
of issue #13 (noticed by Dave).
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Acked-by: Marek Olšák <marek.olsak@amd.com>
Acked-by: Dave Airlie <airlied@redhat.com>
The new functions for this extension were added to a separate file
(dri2_query_renderer.c) to facilitate unit testing. I tried putting
them in dri2_glx.c, and it resulting in an unending chain of
dependencies. It was the proverbial threading hanging from a sweater.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
This structures will be accessed by internal functions that will be
added in a file separate from dri2_glx.c. The new code will be added to
a new file to facilitate unit testing.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
v2: Use sysconf instead of sysinfo for improved portability. Suggested
by Ken.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
v2: Use sysconf instead of sysinfo for improved portability. Suggested
by Ken.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
v2: Add assertions that the version string has the expected format.
This will catch build errors (or changes to the version string format)
in debug build without exposing release builds to buffer over-runs.
Suggested by Ken.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
This will soon be used in intel_screen.c from a function that doesn't
have a gl_context.
v2: Delete local variables that are now unused. This matches v1 of the
changes to the i915 driver.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
This will soon be used in intel_screen.c from a function that doesn't
have a gl_context.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
This will soon be used in intel_screen.c from a function that doesn't
have a gl_context.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
This will soon be used in intel_screen.c from a function that doesn't
have a gl_context.
v2: Remove spurious break after return.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
This will be used to let apps query hardware and driver limits before
creating a GL context.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
If the application requests reset notifiction, connect up the reset
status query method and set gl_context::ResetStrategy.
v2: Update based on kernel interface / libdrm changes.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Soon some drivers will support a different set of flags than other
drivers. If some flags have to be filtered in the driver, we might as
well filter all of them in the driver.
The changes in nouveau use tabs because nouveau seems to have it's own
indentation rules.
v2: Fix some rebase failures noticed by Ken (returning the wrong types,
etc.).
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
No drivers advertise the DRI2 extension yet, so no driver should ever
see a value other than false for notify_reset.
The changes in nouveau use tabs because nouveau seems to have it's own
indentation rules.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Drivers still have to implement dd_function_table::GetGraphicsResetStatus.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
These will be used to determine whether to signal a GPU reset after
another context in the share group has observed a reset.
v2: Change ShareGroupReset from GLboolean to bool. Suggested by Brian.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
This allows drivers to determine whether a GPU reset has occured. It
should return non-zero status if a reset was observed by the specified
context. Another mechanism will be used to observe resets occuring in
other contexts in the share group.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
This isn't going to be used in the actual implemenation of
glGetGraphicsResetStatus.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
v2: Add comments on the purpose of the auxiliary data structures.
Check for atomic counter overlaps. Use the contains_atomic()
convenience method. Add static assert with the number of expected
shader stages.
v3: Don't resize atomic arrays.
v4: Add comment on the reason why we don't resize atomic counter
arrays. Use 'strcmp(...) == 0' instead of '!strcmp(...)'.
v5 (idr): Don't use STL in the linker.
Signed-off-by: Francisco Jerez <currojerez@riseup.net>
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
v2: Mark atomic counters as read-only variables. Move offset overlap
code to the linker. Use the contains_atomic() convenience method.
v3: Use pointer to integer instead of non-const reference. Add
comment so we remember to add a spec quotation from the next GLSL
release once the issue of atomic counter aggregation within
structures is clarified.
v4 (idr): Don't use std::map because it's overkill. Add an assertion
that ctx->Const.MaxAtomicBufferBindings <= MAX_COMBINED_ATOMIC_BUFFERS.
Signed-off-by: Francisco Jerez <currojerez@riseup.net>
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
This reverts most of commit 0f2da77307.
(I chose to leave the additions to brw_defines.h.)
My previous Ironlake implementation was somewhat broken: counter data
was global, rather than per-context. This meant that performance
monitors captured data from your compositor, 2D driver, and other 3D
programs.
Originally, I believed that Sandybridge and later had an easy way to
avoid this problem (setting per-context flags in OACONTROL), while
Ironlake did not. So I'd intended to leave it as a known limitation of
performance monitoring support on Ironlake. However, this turned out
not to be true.
Unfortunately, our hardware only has one set of aggregating performance
counters shared between all 3D programs, and their values are not saved
or restored by hardware contexts. Also, at least on Sandybridge and
Ivybridge, the counters lose their values if the GPU goes to sleep.
To work around both of these problems, we have to snapshot the
performance counters at the beginning and end of each batch, similar to
how we handle query objects on platforms that don't support hardware
contexts.
For occlusion queries, this batch bookending approach is fairly simple:
only one occlusion query can be active at a time, and the result is a
single integer. Performance monitors are more complex: an arbitrary
number of monitors can be active at a time, each monitoring some subset
of our ~30 observability counters. Individual monitors can be started
and stopped at any point during the batch. Tracking where each monitor
started/ended relative to batch flushes ends up being a pain. And you
can run out of space in the buffer.
Properly supporting this required some serious rearchitecting of the
code. Rather than writing patches to try and morph a broken system into
a working one (which operates quite differently), I decided it would be
simplest to revert the old code and start fresh. Parts will look
familiar, but other parts are new.
I also decided it would be best to include Sandybridge and Ivybridge
support from the start, since the newer platforms have added complexity
that I wanted to make sure worked. They're also what most people care
about these days.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Previously, we only exposed them in desktop GL or with:
#extension GL_OES_standard_derivatives : enable
GLSL ES 3.00 includes these without an extension, so we need to expose
them by default.
Note that the above #extension line results in an error or desktop GL,
so we don't need to worry about this.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
This just tells the state tracker to turn on the GL_ARB_shader_texture_lod
extension. This simply allows the GLSL compiler to emit TXL and TXD
instructions for both vertex and fragment shaders. We already support
these opcodes in the svga driver. Though, the shadow2DGrad() Piglit
tests are failing.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
Improves performance of RoboHornet's 2D Canvas toDataURL benchmark
[http://www.robohornet.org/#e=canvastodataurl] by approximately 5x
on Baytrail on ChromiumOS.
Elapsed time drops by -81.4861% +/- 1.22619% (n=3 s=14.9105, confidence=95%).
Reviewed-by: Chad Versace <chad.versace@linux.intel.com>
Uses SSE 4.1's MOVNTDQA instruction (streaming load) to read from
uncached memory without polluting the cache.
Reviewed-by: Chad Versace <chad.versace@linux.intel.com>
This patch make changes to correctly set up the Dispatch GRF Start
Register in case of 'SIMD16 only' FS dispatch.
This fixes an issue of incorrect rendering on dolphin emulator with
GL_SAMPLE_SHADING enabled.
Signed-off-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
This theoretically works on earlier hardware as well, but the extension
requires at least GL3.0.
Signed-off-by: Chris Forbes <chrisf@ijw.co.nz>
Reviewed-by: Eric Anholt <eric@anholt.net>
V2: fix interaction with VertexAttribFormat, since that landed after
this was originally written
Signed-off-by: Chris Forbes <chrisf@ijw.co.nz>
Reviewed-by: Eric Anholt <eric@anholt.net>
With this patch, the llvmpipe and draw modules will calculate the depth bias
according to floating point depth buffer semantics described in the
arb_depth_buffer_float specification, when the driver has a z buffer bound
with a format type of UTIL_FORMAT_TYPE_FLOAT.
By default, the driver will use the existing UNORM calculation for depth bias.
A new function, draw_set_zs_format, was added to calculate the Minimum
Resolvable Depth value and floating point depth sense for the draw module.
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
This brings over the batch-wrap-prevention and aperture space checking
code from the normal brw_draw.c path, so that we don't need to flush the
batch every time.
There's a risk here if the intel_emit_post_sync_nonzero_flush() call isn't
high enough up in the state emit sequences -- before, we implicitly had
one at the batch flush before any state was emitted, so Mesa's workaround
emits didn't really matter. Since the SNB fixes by Ken, I didn't see any
regressions after 3 piglit runs.
Improves cairo-gl performance by 13.7733% +/- 1.74876% (n=30/32)
Improves minecraft apitrace performance by 1.03183% +/- 0.482297% (n=90).
Reduces low-resolution GLB 2.7 performance by 1.17553% +/- 0.432263% (n=88)
Reduces Lightsmark performance by 3.70246% +/- 0.322432% (n=126)
No statistically significant performance difference on unigine tropics
(n=10)
No statistically significant performance difference on openarena (n=755)
The two apps that are hurt happen to include stalls on busy buffer
objects, so I think this is an effect of missing out on an opportune
flush.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
update_array() and update_array_format() are changed to update the new
attrib and binding states, and the client arrays become derived state.
Reviewed-by: Eric Anholt <eric@anholt.net>
...and rename it to _mesa_bind_buffer_gen().
This is so the function can be called from _mesa_BindVertexBuffer().
This patch also adds a caller parameter so we can report the right
entry point in error messages.
Based on a patch by Eric Anholt.
Reviewed-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
This will become derived state as part of the ARB_vertex_attrib_binding
support.
Reviewed-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Split out the code for updating the array format into a new function
called update_array_format(). This function will be called by both
update_array() and the new glVertexAttrib*Format() entry points in
ARB_vertex_attrib_binding.
Reviewed-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Currently, has_surface_tile_offset is equivalent to gen == 4 && !is_g4x.
We already use it for related checks in brw_wm_surface_state.c, so it
makes sense to use it here too. It's simpler and more future-proof.
Broadwell also lacks surface tile offsets. With this patch, I won't
need to update any generation checking; I can simply not set the flag.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Eric Anholt <eric@anholt.net>
Hardware docs say we can only use SIMD8 dispatch in this condition.
Reviewed-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Chris Forbes <chrisf@ijw.co.nz>
We weren't adding the soa offsets when constructing the indices
for the gather functions. That meant that we were always returning
the data in the first element.
(Copied straight from the same fix for temps.)
While here fix up a couple of broken comments in the fetch functions,
plus don't name a straight float type float4 which is just confusing.
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
Reviewed-by: Zack Rusin <zackr@vmware.com>
Otherwise OutputSurface interop has funny results sometimes.
This fixes interop with the mpv media player.
v2 (chk): add proper locking
Signed-off-by: Christian König <christian.koenig@amd.com>
V2: Add comment explaining what emit_alpha_test() is for;
fix spurious temp and bogus whitespace.
Signed-off-by: Chris Forbes <chrisf@ijw.co.nz>
Reviewed-by: Eric Anholt <eric@anholt.net>
The same setup is required here as when the user-provided shader
explicitly uses KIL or discard.
Signed-off-by: Chris Forbes <chrisf@ijw.co.nz>
Reviewed-by: Eric Anholt <eric@anholt.net>
We have to do this in the shader instead, since these gens lack an
independent RT0 alpha value in their render target write messages.
Signed-off-by: Chris Forbes <chrisf@ijw.co.nz>
Reviewed-by: Eric Anholt <eric@anholt.net>
Now that brw_update_texture_buffer_surface() uses the virtual
emit_buffer_surface_state() function, it works for Gen7+ too.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
Now that brw_create_constant_surface uses a virtual function internally,
it doesn't need to be virtual itself. We can delete the Gen7+ variant
and simplify things.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
This will allow us to combine the Gen4-6 and Gen7 variants of these
functions.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
This entails adding "mocs" and "rw" parameters to the Gen4-5 version.
I made it actually pay attention to the rw flag (even though it is
always false), but mocs is always ignored.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
fix: intel_screen.c:1320:4: warning: initialization from
incompatible pointer type [enabled by default]
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Before the series with 3c9dc2d31b to
dynamically assign our binding table indices, we didn't really track our
binding table count per shader, so we never filled in these fields.
Affects cairo-gl trace runtime by -2.47953% +/- 1.07281% (n=20)
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
SSE can't handle true vector shifts (with variable shift count),
so llvm is turning them into a mess of extracts, scalar shifts and inserts.
It is however possible to emulate them in lp_build_minify with float muls,
which should be way faster (saves over 20 instructions per 8-wide
lp_build_minify). This wouldn't work for "generic" 32bit shifts though
since we've got only 24bits of mantissa (actually for left shifts it would
work by using sse41 int mul instead of float mul but not for right shifts).
Note that this has very limited scope for now, since this is only used with
per-pixel lod (otherwise we're avoiding the non-constant shift count by doing
per-quad shifts manually), and only 1d textures even then (though the latter
should change).
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
This will enable removing the dd_function_table::Scissor hook in the
near future.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Francisco Jerez <currojerez@riseup.net>
This will enable removing the dd_function_table::DepthRange hook in the
near future.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Francisco Jerez <currojerez@riseup.net>
The x, y, width, and height parameters aren't used by radeon_viewport,
so don't pass them. This should make future changes to the
dd_function_table::Viewport interface a little easier.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Jordan Justen <jljusten@gmail.com>
Acked-by: Alex Deucher <alexander.deucher@amd.com>
Cc: Courtney Goeltzenleuchter <courtney@lunarg.com>
The i830 and the i915 driver have the same dd_function_table::Viewport
function... it just has two names and lives in two places. Using a
single implementation allows cleaning up the saved_viewport nonsense
too.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Jordan Justen <jljusten@gmail.com>
Cc: Courtney Goeltzenleuchter <courtney@lunarg.com>
The i965 driver never installed a dd_function_table::Viewport function,
so this wrapper never actually did anything.
No piglit regressions on IVB on DRI2.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Jordan Justen <jljusten@gmail.com>
Cc: Courtney Goeltzenleuchter <courtney@lunarg.com>
* Goodbye BeOS, we hardly knew thee
* As BeOS was gcc2 only, there was little chance
of this being useful.
* Doesn't effect Haiku in any meaningful way
Reviewed-by: Brian Paul <brianp@vmware.com>
Previously, when packing geometry shader input varyings like this:
in float foo[3];
in float bar[3];
lower_packed_varyings would declare a packed varying like this:
(declare (shader_in flat) (array ivec4 3) packed:foo[0],bar[0])
That's confusing, since the packed varying acutally stores all three
values of foo and all three values of bar.
This patch causes it to generate the more sensible declaration:
(declare (shader_in flat) (array ivec4 3) packed:foo,bar)
Note that there should be no functional change for users of geometry
shaders, since the packed name is only used for generating debug
output. But this should reduce confusion when using INTEL_DEBUG=gs.
Reviewed-by: Eric Anholt <eric@anholt.net>
* Call mesa viewport call on winndow resize
* Add initial postprocessing code
* Pass hgl_context to private statetracker
as it is more useful than GalliumContext
* Use Lock and Unlock functions to standardize
GalliumContext locking
* Create texture resources in texture validation
Acked-by: Brian Paul <brianp@vmware.com>
The changes between Gen6-7 are minimal, and can easily be solved with
an extra generation check. This cuts a lot of duplicated code.
It also helps prevent even more duplication for Broadwell.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Eric Anholt <eric@anholt.net>
The latency information has been obtained empirically from
measurements taken on Haswell and Ivy Bridge.
Acked-by: Paul Berry <stereotype441@gmail.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
This can deal with all the 15 32-bit untyped atomic operations the
hardware supports, but only INC and PREDEC are going to be exposed
through the API for now.
v2: Represent atomics as GLSL intrinsics. Add support for variably
indexed atomic counter arrays.
v3: Add comment on why we don't need to assign uniform storage for
atomic counters.
Reviewed-by: Paul Berry <stereotype441@gmail.com>
This can deal with all the 15 32-bit untyped atomic operations the
hardware supports, but only INC and PREDEC are going to be exposed
through the API for now.
v2: Represent atomics as GLSL intrinsics. Add support for variably
indexed atomic counter arrays. Fix interaction with fragment
discard.
v3: Add comment on why we don't need to assign uniform storage for
atomic counters.
Reviewed-by: Paul Berry <stereotype441@gmail.com>
This patch fixes the three dead code elimination passes and the
VEC4/FS instruction scheduling passes so they leave instructions with
side effects alone.
At some point it might be interesting to have the instruction
scheduler calculate the exact memory dependencies between atomic ops,
but they're rare enough that it seems unlikely that it will make any
practical difference.
Reviewed-by: Paul Berry <stereotype441@gmail.com>
Inspired by a patch sent to the mailing list by Tom Stellard, but
using a different algorithm to calculate the optimal block size that
has been found to be considerably more effective.
Reviewed-by: Tom Stellard <thomas.stellard@amd.com>
dso_list was added as an argument for createInternalizePass in 3.4, and then
it was removed again in the same llvm version.
Tested-by: Mike Lothian <mike@fireburn.co.uk>
Reviewed-by: Tom Stellard <thomas.stellard@amd.com>
The new option clamps GL_MAX_SAMPLES to a hardware-supported MSAA mode.
If negative, then no clamping occurs.
v2: (for Paul)
- Add option to i965 only, not to all DRI drivers.
- Do not realy on int->uint cast to convert negative
values to large positive values. Explicitly check for
clamp_max_samples < 0.
v3: (for Ken)
- Don't allow clamp_max_samples to alter context version.
- Use clearer for-loop and correct comment.
- Rename variables.
v4: (for Ken)
- Merge identical if-branches.
Reviewed-and-tested-by: Kenneth Graunke <kenneth@whitecape.org>
Signed-off-by: Chad Versace <chad.versace@linux.intel.com>
Actually link VS out / FS in based on semantic info, keeping in mind
that position/pointsize can also be an input to the FS. This fixes a
few fragment shaders which were using gl_Position.
Signed-off-by: Rob Clark <robclark@freedesktop.org>
Fixes use of full-precision in fragment shader (ie. don't clobber r0.x
since that can be used by future bary instructions for varying fetch).
And makes use of full-precision the default in fragment shader (but can
be overriden via FD_MESA_DEBUG=fraghalf).
Seems like half precision is often not enough for texture coordinates.
The blob compiler is clever enough to keep texture coords in full
precision registers while using half precision for everything else. But
we aren't quite that clever yet, so better to default to full precision.
Signed-off-by: Rob Clark <robclark@freedesktop.org>
Handle some relative addressing constraints: cannot handle const or
relative in cat5 and src2 of cat3.
Signed-off-by: Rob Clark <robclark@freedesktop.org>
- Enable GEN7_WM_MSDISPMODE_PERSAMPLE, GEN7_WM_POSOFFSET_SAMPLE,
GEN7_WM_OMASK_TO_RENDER_TARGET as per extension's specification.
- Only enable one of GEN7_WM_8_DISPATCH_ENABLE or GEN7_WM_16_DISPATCH_ENABLE
when GEN7_WM_MSDISPMODE_PERSAMPLE is enabled. Refer IVB PRM Vol. 2, Part 1,
Page 288 for details.
V2:
- Use shared function _mesa_get_min_invocations_per_fragment().
- Use brw_wm_prog_data variables: uses_pos_offset, uses_omask.
V3:
- Enable simd16 dispatch with per sample shading.
- Make changes to give preference to 'simd16 only' mode over
'simd8 only' mode in case of non 1x per sample shading.
Signed-off-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
- Enable GEN6_WM_MSDISPMODE_PERSAMPLE, GEN6_WM_POSOFFSET_SAMPLE,
GEN6_WM_OMASK_TO_RENDER_TARGET as per extension's specification.
- Only enable one of GEN6_WM_8_DISPATCH_ENABLE or GEN6_WM_16_DISPATCH_ENABLE
when GEN6_WM_MSDISPMODE_PERSAMPLE is enabled.
Refer SNB PRM Vol. 2, Part 1, Page 279 for details.
V2:
- Use shared function _mesa_get_min_invocations_per_fragment().
- Use brw_wm_prog_data variables: uses_pos_offset, uses_omask.
V3:
- Enable simd16 dispatch with per sample shading.
- Make changes to give preference to 'simd16 only' mode over
'simd8 only' mode in case of non 1x per sample shading.
Signed-off-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
V2:
- Update comments
- Add a special backend instructions to compute sample_mask.
- Add a new variable uses_omask in brw_wm_prog_data.
V3:
- Make changes to support simd16 mode.
- Delete redundant AND instruction and handle the register
stride in FS backend instruction.
Signed-off-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
V2:
- Update comments
- Add compute_sample_id variables in brw_wm_prog_key
- Add a special backend instruction to compute sample_id.
V3:
- Make changes to support simd16 mode.
Signed-off-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
This is required while adding builtin system value vec{2, 3, 4}
variables. For example:
(declare (sys) vec2 gl_SamplePosition)
Without this patch above glsl ir splits in to:
(declare (temporary) float gl_SamplePosition_x)
(declare (temporary) float gl_SamplePosition_y)
Signed-off-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
This function is used to test if we need to do per sample shading or
per fragment shading.
V2: Use MAX2() to make sure the function returns a number >= 1.
Signed-off-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
New builtins added by GL_ARB_sample_shading:
in vec2 gl_SamplePosition
in int gl_SampleID
in int gl_NumSamples
out int gl_SampleMask[]
V2: - Use SWIZZLE_XXXX for STATE_NUM_SAMPLES.
- Use "result.samplemask" in arb_output_attrib_string.
- Add comment to explain the size of gl_SampleMask[] array.
- Make gl_SampleID and gl_SamplePosition system values.
Signed-off-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Number of samples will be required in fragment shader program by new
GLSL builtin uniform "gl_NumSamples".
V2: Use "state.numsamples" in place of "state.num.samples"
Use _NEW_BUFFERS flag in place of _NEW_MULTISAMPLE
Signed-off-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Ian Romanick <idr@freedesktop.org>
Reviewed-by: Ken Graunke <kenneth@whitecape.org>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
New functions added by GL_ARB_sample_shading:
glMinSampleShadingARB()
New enums:
GL_SAMPLE_SHADING_ARB
GL_MIN_SAMPLE_SHADING_VALUE_ARB
V2: Update comments.
Create new GL4x.xml.
Remove redundant code in get.c.
Update the API_XML list in Makefile.am.
Add extra_gl40_ARB_sample_shading predicate to get.c.
V3:
Fix make check failure.
Add checks for desktop GL.
Use GLfloat in place of GLclampf in glMinSampleShading().
Signed-off-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Ken Graunke <kenneth@whitecape.org>
This patch implements the common support code required for the
GL_ARB_sample_shading extension.
V2: Move GL_ARB_sample_shading to ARB extension list.
Signed-off-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Ian Romanick <idr@freedesktop.org>
Reviewed-by: Ken Graunke <kenneth@whitecape.org>
Only one program's instruction count is changed, but a shader in Tropics
is also affected.
instructions in affected programs: 326 -> 320 (-1.84%)
Reviewed-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
total instructions in shared programs: 1409124 -> 1406971 (-0.15%)
instructions in affected programs: 158376 -> 156223 (-1.36%)
Reviewed-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
Helps a lot of Steam games.
total instructions in shared programs: 1409360 -> 1409124 (-0.02%)
instructions in affected programs: 20842 -> 20606 (-1.13%)
Reviewed-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
This only operates on constant/uniform values for now, because otherwise I'd
have to deal with killing my available CSE entries when assignments happen,
and getting even this working in the tree ir was painful enough.
As is, it has the following effect in shader-db:
total instructions in shared programs: 1524077 -> 1521964 (-0.14%)
instructions in affected programs: 50629 -> 48516 (-4.17%)
GAINED: 0
LOST: 0
And, for tropics, that accounts for most of the effect, the FPS
improvement is 11.67% +/- 0.72% (n=3).
v2: Use read_only field of the variable, manually check the lod_info union
members, use get_num_operands(), rename cse_operands_visitor to
is_cse_candidate_visitor, move all is-a-candidate logic to that
function, and call it before checking for CSE on a given rvalue, more
comments, use private keyword.
Reviewed-by: Paul Berry <stereotype441@gmail.com>
Prior to the GLSL CSE pass, all of our testing happened to have a freshly
computed temporary in op[1], from the multiply by 16 to get a byte offset.
As of CSE you'll get var_refs of a reused value when you've got multiple
loads from the same offset.
Make a proper temporary for computing our temporary value, to avoid
shifting the value farther and farther down. Avoids a regression in
gs-float-array-variable-index
Reviewed-by: Paul Berry <stereotype441@gmail.com>
Previously, the write of each 32-bit half might land in separate batch
buffers, which is insane.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Chris Forbes <chrisf@ijw.co.nz>
This depends on ARB_transform_feedback2, so I've predicated it on the
ability to do register writes.
It also depends on ARB_transform_feedback3, which is the only reason we
couldn't expose it previously.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
This extension is written a bit strangely. Although it introduces the
concept of multiple transform feedback streams, it doesn't actually
provide more than a single stream.
The ARB_gpu_shader5 extension is what introduces the ability to write to
streams other than stream #0 and increases the required number of streams.
Since we don't yet support ARB_gpu_shader5, we can safely enable
ARB_transform_feedback3 even though we only support a single stream.
This does provide some useful functionality: applications can now use
more than one interleaved transform feedback buffer.
v2: Only expose the extension if ARB_transform_feedback2 is also
available, to avoid confusing applications (suggested by Ian).
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
ARB_transform_feedback3 allows applications to insert blank space
between interleaved varyings by adding fake 1, 2, 3, or 4-component
varyings named gl_SkipComponents[1234].
Mesa's core data structures don't explicitly track these, instead simply
tracking the buffer offset for each real varying. If there is padding
due to gl_SkipComponents, these will not be contiguous.
Our hardware takes the specification quite literally. Instead of
specifying offsets for each varying, it assumes they're all contiguous
and requires you to program fake varyings for each "hole".
This patch adds support for emitting SO_DECL structures for these holes.
Although we've lost the information about exactly how the application
specified their padding (i.e. gl_SkipComponents2, gl_SkipComponents2
vs. a single gl_SkipComponents4), it shouldn't matter. We just need to
emit the right amount of space. This patch emits the minimal number of
hole SO_DECL structures.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Currently, we emit one SO_DECL structure per output, so we use the index
in the Outputs[] array as the index into the so_decl[] array as well.
In order to support the fake "gl_SkipComponents[1234]" varyings from
ARB_transform_feedback3, we'll need to emit SO_DECLs to fill in the
holes between successive outputs. This means we'll likely emit more
SO_DECLs than there are outputs, so we need to count it explicitly.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
With Linux 3.12, register writes work on Ivybridge and Baytrail, but not
Haswell. That will be fixed in a future kernel revision, at which point
this extension will automatically be enabled.
v2: Use I915_GEM_DOMAIN_INSTRUCTION for the register read, and also
correctly set the writeable flag when mapping (caught by Eric).
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Eric Anholt <eric@anholt.net>
We only want to enable ARB_transform_feedback2 if we can write to
registers from batchbuffers. In order to test that, we need to be able
to submit batches. And for batches to work, we need to program the
initial pipeline state (like PIPELINE_SELECT), which is done from
brw_state_init().
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Eric Anholt <eric@anholt.net>
Implementing the GetTransformFeedbackVertexCount() driver hook allows
the VBO module to call us with the right number of vertices.
The hardware doesn't directly count the number of vertices written by
SOL, so we instead use the SO_NUM_PRIMS_WRITTEN(n) counters and multiply
by the number of vertices per primitive.
Unfortunately, counting the number of primitives generated is tricky:
a program might pause a transform feedback operation, start a second one
with a different object, then switch back and resume. Both transform
feedback operations share the SO_NUM_PRIMS_WRITTEN counters.
To work around this, we save the counter values at Begin, Pause, Resume,
and End. This "bookends" each section where transform feedback is
active for the current object. Adding up differences of pairs gives
us the number of primitives generated. (This is similar to what we
do for occlusion queries on platforms without hardware contexts.)
v2: Fix missing parenthesis in assertion (caught by Eric Anholt).
v3: Reuse prim_count_bo rather than freeing it and immediately
allocating a new one (suggested by Topi Pohjolainen).
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
Renaming it makes it obvious that it isn't used, and the assertion
verifies that the VBO module never passes us such an object.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
DrawTransformFeedback() needs to obtain the number of vertices written
to a particular stream during the last Begin/EndTransformFeedback block.
The new driver hook returns exactly that information.
Gallium drivers already implement this by passing the transform feedback
object to the drawing function, counting the number of vertices written
on the GPU, and using draw indirect. This is efficient, but doesn't
always work:
If vertex data comes from user arrays, then the VBO module needs to
know how many vertices to upload, so we need to synchronously count.
Gallium drivers are currently broken in this case.
It also doesn't work if primitive restart is done in software. For
normal drawing, vbo_draw_arrays() performs software primitive restart,
splitting the draw call in two. vbo_draw_transform_feedback() currently
doesn't because it has no idea how many vertices need to be drawn.
The new driver hook gives it that information, allowing us to reuse
the existing vbo_draw_arrays() code to do everything right.
On Intel hardware (at least Ivybridge), using the draw indirect approach
is difficult since the hardware counts primitives, rather than vertices,
which requires doing some simple math. So we always use this hook.
Gallium drivers will likely want to use this hook in some cases, but
want to use the existing draw indirect approach where possible. Hence,
I've added a flag to allow drivers to opt-in to this call.
v2: Make it possible to implement this hook but only use this path
when necessary (suggested by Marek).
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
The ARB_transform_feedback2 extension introduces the ability to pause
and resume transform feedback sessions. Although only one can be active
at a time, it's possible to switch between multiple transform feedback
objects while paused.
In order to facilitate this, we need to save/restore the SO_WRITE_OFFSET
registers so that after resuming, the GPU continues writing where it
left off.
This functionality also exists in ES 3.0, but somehow we completely
forgot to implement it.
v2: Reduce alignment from 4096 to 64 (it seemed excessive).
v3: Use I915_GEM_DOMAIN_INSTRUCTION instead of RENDER, for consistency
with other writes. It shouldn't matter on IVB+.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
This adds the basic driver hooks to allocate/free the brw variant.
It doesn't contain any additional information yet, but it will soon.
v2: Use the new _mesa_init_transform_feedback_object helper function
(requested by Eric and Ian).
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Eric Anholt <eric@anholt.net>
This picks up a missing obj->EverBound = GL_FALSE line, and will catch
any new fields that get added in the future.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
Both Gallium and i965 subclass gl_transform_feedback_object, which
requires implementing a custom NewTransformFeedback hook. Creating a
helper function to initialize the fields avoids code duplication and
divergence.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
We'd like to CSE some instructions, like CMP, that often have null
destinations. Instead of replacing them with MOVs to null, just don't
emit the MOV.
Reviewed-by: Paul Berry <stereotype441@gmail.com>
This avoids a lot of message setup we had to do otherwise. Improves
GLB2.7 performance with register spilling force enabled by 1.6442% +/-
0.553218% (n=4).
v2: Use BRW_PREDICATE_NONE, improve a comment (by Paul).
Reviewed-by: Paul Berry <stereotype441@gmail.com>
We were clearing the reg_offset before trying to use it. Oops. Fixes
glsl-fs-texture2drect with the reg spilling debug enabled.
Reviewed-by: Paul Berry <stereotype441@gmail.com>
Things blew up when I enabled the debug register spill code without
disabling 16-wide, so I decided to just fix 16-wide spilling.
We still don't generate 16-wide when register spilling happens as part of
allocation (since we expect it to be slower), but now we can experiment
with allowing it in some cases in the future.
Reviewed-by: Paul Berry <stereotype441@gmail.com>
I believe this will never happen in SIMD8 mode, but it could for SIMD16
when we fix it.
v2: Fix off-by-one in my register counting comment (caught by Paul).
Reviewed-by: Paul Berry <stereotype441@gmail.com> (v1)
Now that reg spilling generates new vgrfs, we were looping forever if you
ever turned it on.
Instead, move the debug code into the register allocator right near where
we'd be doing spilling anyway, which should more accurately reflect how
register spilling occurs in the wild.
Reviewed-by: Paul Berry <stereotype441@gmail.com>
I'm going to need to reuse this for fixing register spilling on SIMD16.
Note that BRW_MAX_MRF is 16, which is the same as BRW_MAX_GRF -
GEN7_MRF_HACK_START.
Reviewed-by: Paul Berry <stereotype441@gmail.com>
The ICD loader should be responsible for installing headers.
Reviewed and Tested-by: Aaron Watry <awatry@gmail.com>
Reviewed-by: Francisco Jerez <currojerez@riseup.net>
When faced with a million instructions that all became candidates at the
same time (none of which individually reduce register pressure), the ones
on the critical path are more likely to be the ones that will free up some
candidates soon.
shader-db:
total instructions in shared programs: 1681070 -> 1681070 (0.00%)
instructions in affected programs: 0 -> 0
GAINED: 40
LOST: 74
Fixes indistinguishable-from-hanging behavior in GLES3conform's
uniform_buffer_object_max_uniform_block_size test, regressed by
c3c9a8c857. Given that
93bd627d5a was unlocked by that commit, the
net effect on 16-wide program count is still quite positive, and I think
this should give us more stable scheduling (less dependency on original
instruction emit order).
v2: Comment suggestions by Paul
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=70943
Reviewed-by: Paul Berry <stereotype441@gmail.com>
This is a step in doing scheduling as described in Muchnick (p538). A
difference is that our latency function is only specific to one
instruction (it doesn't describe, for example, the different latency
between WAR of a send's arguments and RAW of a send's destination), but
that's changeable later. We also don't separately compute the postorder
traversal of the graph, since we can use the setting of the delay field as
the "visited" flag.
Reviewed-by: Paul Berry <stereotype441@gmail.com>
Commit aec20d66d9
(automake: properly handle non-default expat installation),
assumed that up-to date distributions use a recent version
of expat that handles security vunerabilities CVE-2012-1147
and CVE-2012-1148. Seems like this is not always the case
and they prefer to backport only the fix, rather than use
the updated library.
This commit adds a default case -lexpat whenever expat is
not found, while properly handling expat.pc if present.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=71022
Reported-By: Bryce Harrington <b.harrington@samsung.com>
Reported-By: Vinson Lee <vlee@freedesktop.org>
Tested-by: Bryce Harrington <b.harrington@samsung.com>
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
This will simplify the addition of layout(location) qualifiers for
separate shader objects. This was validated with new piglit tests
arb_explicit_attrib_location/1.30/compiler/not-enabled-01.vert and
arb_explicit_attrib_location/1.30/compiler/not-enabled-02.vert.
v2: Refactor error checking to check_explicit_attrib_location_allowed
and eliminate the gotos. Suggested by Paul.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
Use mode_string to get the name of the variable mode. Slightly change
the control flow. Both of these changes make it easier to support
separate shader object location layouts.
The format of the message changed because mode_string can return a
string like "shader output". This would result in an awkward message
like "vertex shader shader output..."
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
I made this a function (instead of a method of ir_variable) because it
made the change set smaller, and I expect that there will be an overload
that takes an ir_var_mode enum. Having both functions used the same way
seemed better.
v2: Add missing case for ir_var_system_value.
v3: Change the ir_var_mode_count case to just break. Move the assertion
and the return outside the switch-statment. In the unlikely event that
var->mode is an invalid value other than ir_var_mode_count, the
assertion will still fire, and in release builds we won't wind up
returning a garbage pointer. Suggested by Paul.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
Since the separation of ir_var_function_in and ir_var_shader_in (similar
for out), this check is no longer necessary. Previously, global_scope
was the only way to tell which was which.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
Future patches will add some extra code to this path, and some of that
code will want to exit from the explicit location code early.
v2: Change a geometry shader "break" to a "return" so that try to apply
a bogus geometry shader location qualifier (which could cause cascading
errors). Suggested by Paul.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
The return value has been unused since commit d348b0c. This was
originally included in another patch, but it was split out by Ian
Romanick.
v2: Drop unnecessary final return. Suggested by Paul.
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
Cc: Eric Anholt <eric@anholt.net>
Incremental builds were failing because not all generated source files
were missing dependencies to src/mapi/glapi/gen/*.xml.
Hopefully this change will be the end of these incremental build
failures.
This avoids a defect in lower_output_reads.
The problem is lower_output_reads treats the gl_FragData array as a single
variable. It first redirects all output writes to a temporary variable (array)
and then writes the whole temporary variable to the output, generating
assignments to all elements of gl_FragData.
BTW this pass can be modified to lower all arrays, not just inputs and outputs.
The question is whether it is worth it.
Reviewed-by: Paul Berry <stereotype441@gmail.com>
v2: addressed Paul Berry's comments
Use PKG_CHECK_MODULE over requesting the user to setup the
option at configure time. Drop unused EXPAT_INCLUDE and
update all targets.
NOTE: The this commit removes the --with-expat configure
option. One should ensure that the expat they wish to use
has expat.pc file accessible by pkg-config.
v2:
* Add note about the removal of --with-expat
(per Tom Stellard)
* Drop EXPAT_CFLAGS for targets that do not build DRI_COMMON
(spotted by Matt Turner)
v3:
* Rebase on top of megadrivers (drop EXPAT_CFLAGS from swrast)
Acked-by: Matt Turner <mattst88@gmail.com> (v2)
Reviewed-by: Tom Stellard <thomas.stellard@amd.com> (v2)
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Conflicts:
configure.ac
src/mesa/drivers/dri/common/Makefile.am
The function should have never used it in the first place as it was
a left over from the DRI1 days of the nouveau ddx. While we're around
check if KMS is supported before opening the nouveau device, and
add support for Fermi & Kepler cards.
Compile tested only due to the lack of a Fermi/Kepler card.
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
The function xf86GetEntityInfo() retrieves the entity rather than
doing any changes. Remove this no-op code.
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Reviewed-by: Tom Stellard <thomas.stellard@amd.com>
v2: Remove xf86PciInfo.h, all drivers provide their own PCI ID list
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
A convenient front end to indices generate/translate code, for emulating
primitives which are not supported natively by the driver.
This handles saving/restoring index buffer state, etc.
Signed-off-by: Rob Clark <robclark@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
The idea of the original order was that you'd dead code eliminate accesses
to push constants. But I've never seen a case of that (nor has
shader-db), while we frequently see sparse accesses of large constant
arrays that would overflow into pull constants.
Cuts pull constant use on csgo, serious sam, planeshift, and the cave:
total instructions in shared programs: 1695103 -> 1688795 (-0.37%)
instructions in affected programs: 92024 -> 85716 (-6.85%)
GAINED: 339
LOST: 0
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
* Use more consistant data sources
* Fix improper color space assignments
* Remove unnecessary comments and code
* Drop unnecessary round_up function (this was leftover
from moving winsys code out of renderer)
Acked-by: Brian Paul <brianp@vmware.com>
* Instead of assuming the displaytarget is the same
stride / colorspace as the destination, lets
actually check the source bitmap.
* Fixes random stride issues in rendering
Acked-by: Brian Paul <brianp@vmware.com>
The MRF variant is going to be used extensively by the atomic counter
intrinsics to assemble untyped atomic and surface read messages
easily.
Reviewed-by: Paul Berry <stereotype441@gmail.com>
The maximum number of atomic buffer objects is somewhat arbitrary, we
can change it in the future easily if it turns out it's not enough...
v2: Add comments with the relevant mesa dirty bits. Fix usage of
BRW_NEW_UNIFORM_BUFFER in the GS ABO state atom.
v3: Update binding table layout diagrams.
v4: Resolve conflicts with the recent dynamic surface index assignment changes.
Reviewed-by: Paul Berry <stereotype441@gmail.com>
Almost a trivial change, it boils down to renaming a few identifiers
so their names still make sense for opaque types other than sampler.
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Fix the linker to deal with intrinsic functions which are undefined
all the way down to the driver back-end, and introduce intrinsic
definition helpers in the built-in generator.
We still need to figure out what kind of interface we want for drivers
to communicate to the GLSL front-end which of the supported intrinsics
should use a default GLSL implementation and which should use a
hardware-specific override. As there's no default GLSL implementation
for atomic ops, this seems like something we can worry about later on.
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
v2: Define local helper function to generate ir_call nodes in the
builtin generator.
And use it to forbid comparisons of opaque operands. According to the
GL 4.2 specification:
> Except for array indexing, structure member selection, and
> parentheses, opaque variables are not allowed to be operands in
> expressions.
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
v2: Fix GLSL version in which the type became available. Add
contains_atomic() convenience method. Split off atomic counter
comparison error checking to a separate patch that will handle all
opaque types. Include new ir_variable fields for atomic types.
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
This patch implements the common support code required for the
ARB_shader_atomic_counters extension. It defines the necessary data
structures for tracking atomic counter buffer objects (from now on
"ABOs") associated with some specific context or shader program, it
implements support for binding buffers to an ABO binding point and
querying the existing atomic counters and buffers declared by GLSL
shaders.
v2: Fix extension checks. Drop unused MAX_ATOMIC_BUFFERS constant.
Acked-by: Paul Berry <stereotype441@gmail.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Add XML file for the dispatch code generator, update the
dispatch_sanity test and add stub definition for the new entry point.
Reviewed-by: Paul Berry <stereotype441@gmail.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
These ralloc contexts belong to a specific object and are being
deallocated manually from the class destructor. Now that we've hooked
up destructors to ralloc there's no reason for them to be children of
any other context, and doing so might to lead to double frees under
some circumstances. The class destructor has all the responsibility
of freeing class memory resources now.
This patch makes sure that class destructors are called as they should
be when a C++ object allocated by ralloc is released.
Based on a previous patch by Kenneth Graunke, but it doesn't exhibit
the ~0.8% performance regression in shader compilation times because
we now use the HAS_TRIVIAL_DESTRUCTOR() macro to detect the typical
case where the indirect function call can be avoided because the
object's destructor doesn't need to do anything.
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Only implemented on GCC and Clang for now. Other compilers use a
dummy implementation that always returns false, which should be a safe
[but slightly inefficient] assumption in all cases.
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
This will let us use strcasecmp() from anywhere inside Mesa without
having to worry about the fact that it doesn't exist in MSVC.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
The layer coming from GS needs to be clamped (not sure if that's actually
the correct error behavior but we need something) as the number can be higher
than the amount of layers in the fb. However, this code was using the layer
calculation from the scene, and this was actually calculated in
lp_scene_begin_rasterization() hence too late (so setup was using the value
from the _previous_ scene or just zero if it was the first scene).
Since the value is used in both rasterization and setup, move calculation up
to lp_scene_begin_binning() though it's a bit more inconvenient to calculate
there. (Theoretically could move _all_ code which was in
lp_scene_begin_rasterization() to there, because ever since we got rid of
swizzled render/depth buffers our "map" functions preparing the fb data for
render don't actually change the data in there at all, but it feels like
it would be a hack.)
v2: improve comments
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
Before we were only checking the st->vertex_array_out_of_memory flag
after updating array state. But if there's two consecutive glDrawArrays
calls and the first one is skipped because of OOM, the second one should
be skipped too.
Cc: 9.2 <mesa-stable@lists.freedesktop.org>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Orbital Explorer was generating a 4000 instruction geometry shader, which
was taking 275 trips through dead code elimination and register
coalescing, each of which updated live variables to get its work done, and
invalidated those live variables afterwards.
By using bitfields instead of bools (reducing the working set size by a
factor of 8) in live variables analysis, it drops from 88% of the profile
to 57%, and reduces overall runtime from I-got-bored-and-killed-it (Paul
says 3+ minutes) to 10.5 seconds.
Compare to f179f419d1 on the FS side.
Reviewed-by: Paul Berry <stereotype441@gmail.com>
This prevents unnecessary (and wrong) register allocation in the
scheduler for preloaded values in fixed registers.
Fixes interpolation-mixed.shader_test on rv770
(and probably on all other pre-evergreen chips).
Signed-off-by: Vadim Girlin <vadimgirlin@gmail.com>
Tested-by: Andreas Boll <andreas.boll.dev@gmail.com>
I noticed this in a shader in Unigine Heaven that was spilling. While it
doesn't really reduce register pressure, it shaves a few instructions
anyway (7955 -> 7882).
v2: Fix turning "0 >> x" into "x" instead of "0" (caught by Erik
Faye-Lund).
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
While ir_builder is slightly less efficient, we're only increasing the
work when there's actual optimization being done, and it's way more
readable code.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Matt and I had each screwed up these common required patterns recently, in
ways that wouldn't have been noticed for a long time if not for code
review. Just enforce it in the caller so that we don't rely on code
review catching these bugs.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
There is nothing in the OpenGL specification which prevents the user from
calling glGenQueries to generate a new query object while another object is
active. Neither is there anything in the Mesa implementation which prevents
this. So remove the INVALID_OPERATION errors in this case.
Similarly, it is explicitly allowed by the OpenGL specification to delete an
active query, so remove the assertion for that case, replacing it with the
necesssary state updates to end the query, (clear the bindpt pointer and call
into the driver's EndQuery hook).
CC: <mesa-stable@lists.freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
Tested-by: Brian Paul <brianp@vmware.com>
The normal drawing path does this, and it's necessary on Ivybridge,
so let's try it on Sandybridge too. It's not explicitly documented
as necessary, but might help with hangs.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Tested-by: Xinkai Chen <yeled.nova@gmail.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
Cc: "9.2" <mesa-stable@lists.freedesktop.org>
From the documentation:
"[DevIVB] 3DSTATE_DEPTH_BUFFER must always be programmed along with the
other Depth/Stencil state commands(i.e. 3DSTATE_CLEAR_PARAMS,
3DSTATE_STENCIL_BUFFER, or 3DSTATE_HIER_DEPTH_BUFFER)."
We normally do this, but BLORP was failing to do so in the case where it
disables depth.
Not observed to fix anything yet.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Tested-by: Xinkai Chen <yeled.nova@gmail.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
Cc: "9.2" <mesa-stable@lists.freedesktop.org>
For some reason, we put the flush in the caller, rather than just before
emitting the packet. This is more than a cosmetic problem: BLORP calls
gen6_emit_3dstate_multisample() directly, and so it missed the flush.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Tested-by: Xinkai Chen <yeled.nova@gmail.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
Cc: "9.2" <mesa-stable@lists.freedesktop.org>
From the comments above intel_emit_post_sync_nonzero_flush:
"[DevSNB-C+{W/A}] Before any depth stall flush (including those
produced by non-pipelined state commands), software needs to first
send a PIPE_CONTROL with no bits set except Post-Sync Operation != 0."
This suggests that every non-pipelined (0x79xx) command needs a
post-sync non-zero flush before it.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Tested-by: Xinkai Chen <yeled.nova@gmail.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
Cc: "9.2" <mesa-stable@lists.freedesktop.org>
Otherwise the gen6 w/a in the kernel won't kick in and the write will
land nowhere.
Inspired by a patch Ken pointed me at which had the same issue (but
isn't yet merged and also for a gen7+ feature). An audit of the entire
driver didn't reveal any other case than the one in in the write_reg
helper used by the gen6 queryobj code.
Acked-by: Kenneth Graunke <kenneth@whitecape.org>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Tested-by: Xinkai Chen <yeled.nova@gmail.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
Cc: "9.2" <mesa-stable@lists.freedesktop.org>
The main purpose of this patch is to increase readability of
the array code by introducing is_unsized_array() to glsl_types.
Some redundent is_array() checks are also removed, and small number
of other related clean ups.
The introduction of is_unsized_array() should also make the
ARB_arrays_of_arrays code simpler and more readable when it arrives.
V2: Also replace code that checks for unsized arrays directly with the
length variable
Signed-off-by: Timothy Arceri <t_arceri@yahoo.com.au>
v3 (Paul Berry <stereotype441@gmail.com>): clean up formatting.
Separate whitespace cleanups to their own patch.
Reviewed-by: Paul Berry <stereotype441@gmail.com>
When a geometry shader is present, the fragment shader gl_PrimitiveID
input acts like an ordinary varying, receiving data from the gs
gl_PrimitiveID output. When there's no geometry shader, we have to
ask the fixed function SF hardware to provide the primitive ID to the
fragment shader instead.
Previously, the SF setup code would handle this situation by
recognizing that the FS gl_PrimitiveID input didn't match to any VS
output; since normally an FS input with no corresponding VS output
leads to undefined data, the SF setup code used to just arbitrarily
assign it to receive data from attribute 0.
This patch changes the SF setup code so that instead of arbitrarily
using attribute 0, it assigns the unmatched FS input to receive
gl_PrimitiveID. In the case where the FS input really is
gl_PrimitiveID, this produces the intended result. In all other
cases, no harm is done since GL specifies that the behaviour is
undefined.
Fixes piglit test primitive-id-no-gs.
v2: If an attribute is already being overridden with point
coordinates, don't try to also override it with gl_PrimitiveID. This
is necessary to avoid regressing piglit tests such as
shaders/glsl-fs-pointcoord.
Reviewed-by: Eric Anholt <eric@anholt.net>
v2: Actually implement interop between the gallium
state tracker and the VDPAU backend.
v3: Make it also available in non legacy contexts,
fix video buffer sharing.
v4: deny interop if we don't have the same screen object
v5: rebased on upstream changes
v6: implemented VDPAUGetSurfaceivNV, improved error handling,
unregister all surfaces in VDPAUFiniNV
v7: squash merge with Mareks changes
Signed-off-by: Christian König <christian.koenig@amd.com>
ir_txf expects an ivec* coordinate, and may be larger than ivec2;
shuffle things around so that this will work.
V2: Fix style nits, use ir_builder
Signed-off-by: Chris Forbes <chrisf@ijw.co.nz>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
It turns out that nonzero offsets with gsampler2DRect don't work -- they
just return garbage. Work around this by folding the offset into the
coord.
Done as an IR pass rather than yet another hack in the visitors because
it's clear what's going on this way. Can possibly reuse this to replace
the existing txf coord+offset hacks.
V2: Use ir_builder
Signed-off-by: Chris Forbes <chrisf@ijw.co.nz>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
We don't have a message that does 4 independent offsets; a lowering
pass needs to lower it to 4 normal gather4s before reaching this
point.
Signed-off-by: Chris Forbes <chrisf@ijw.co.nz>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Note that gather4_po_c's parameters are too long for SIMD16. It might be
worth emitting 2xSIMD8 messages in this case at some point.
Signed-off-by: Chris Forbes <chrisf@ijw.co.nz>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
gather4_c's argument layout is straightforward -- refz just goes on the
end.
gather4_po_c's layout however -- the array index is replaced with refz.
Signed-off-by: Chris Forbes <chrisf@ijw.co.nz>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
ARB_gpu_shader5's textureGather*() functions which take shadow samplers
have a separate `refz` parameter rather than adding it to the
coordinate.
Signed-off-by: Chris Forbes <chrisf@ijw.co.nz>
Reviewed-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
V3: fixup crazy check for whether we need to emit the coordinate after
custom handling.
Signed-off-by: Chris Forbes <chrisf@ijw.co.nz>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Some texturing ops are about to have nonconstant offset support; the
offset in the header in these cases should be zero.
Signed-off-by: Chris Forbes <chrisf@ijw.co.nz>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
The generator code ends up clearer this way than if we had to sniff
via the message length. Implemented via the gather4_po message in
hardware, which is present in Gen7 and later.
Signed-off-by: Chris Forbes <chrisf@ijw.co.nz>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Prior to ARB_gpu_shader5 / GLSL 4.0, the offset is required to be
a constant expression.
With that extension, it is relaxed to be an arbitrary expression.
Signed-off-by: Chris Forbes <chrisf@ijw.co.nz>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Since 062317d667 (i965: Go back to using the kernel SOL reset feature.)
we've been flushing the batch on BeginTransformFeedback(). So it's not
necessary to do it on EndTransformFeedback(). A PIPE_CONTROL will work.
This makes gen7_end_transform_feedback() exactly the same as the gen6
variant. However, they'll diverge again shortly.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Eric Anholt <eric@anholt.net>
This was a hack to avoid choosing to schedule all texturing before
consumption of any texture results due to the way dependency chains worked
out in the presence of MRFs. On gen7, we don't have MRFs, so the problem
doesn't apply, and this was just badly constraining our scheduling.
total instructions in shared programs: 1615306 -> 1612534 (-0.17%)
instructions in affected programs: 9958 -> 7186 (-27.84%)
GAINED: 259
LOST: 9
Reviewed-by: Matt Turner <mattst88@gmail.com>
The LIFO plan was simple: Take the most recently made available
instructions, and pick those first.
But because of the order we were pushing things onto our list of
available-to-schedule instructions, it meant that when a set of
instructions was made available at the same time (for example, everything
at the start of the program that didn't depend on other instructions) we'd
schedule them in reverse order.
If you had 10 texture calls in a row in your program, each with
independent argument setup, we'd set up the last texture call's args and
execute it first, even though we wouldn't be able to consume its results
until we'd finished the other 9 texture calls (assuming consumption of
texture results happens near each texture call, and combines it with
another texture result, which is normal for a convolution shader).
To fix this, walk the list for doing LIFO in the order that instructions
were originally generated in the program, but choose to push
newly-made-available instructions to the other end of the list instead.
total instructions in shared programs: 1587242 -> 1586290 (-0.06%)
instructions in affected programs: 7801 -> 6849 (-12.20%)
GAINED: 76
LOST: 67
Thanks to Chia-I Wu for pointing out the bug in my first version of the
patch that made it a huge loss.
Reviewed-by: Matt Turner <mattst88@gmail.com>
When PIPE_CAP_MIXED_FRAMEBUFFER_SIZES is not provided, parts of
ARB_framebuffer_object can't be supported, such as on NV30.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Signed-off-by: Marek Olšák <marek.olsak@amd.com>
This CAP will determine whether ARB_framebuffer_object can be enabled.
The nv30 driver does not allow mixing swizzled and linear zsbuf/cbuf
textures.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Signed-off-by: Marek Olšák <marek.olsak@amd.com>
In commit 1b4a737 (glsl: Support redeclaration of VS and GS
gl_PerVertex output), I added code to ensure that when an unnamed
gl_PerVertex interface block is redeclared, any ir_variables that
weren't included in the redeclaration are removed from the IR (and the
symbol table). This ensures that only those variables that were
explicitly redeclared may be used.
However, when I wrote this code, I neglected to match the variable
mode when finding variables to remove. This meant that redeclaring a
built-in output block might cause the built-in input gl_in to be
accidentally removed.
Fixes piglit test gs-redeclares-pervertex-out-only.
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
The GLSL 4.10 rules for redeclaration of built-in interface blocks
(which we've chosen to regard as clarifications of GLSL 1.50) only
require gl_PerVertex blocks to match in shaders that actually use
those blocks. The easiest way to implement this is to detect
situations where a compiled shader doesn't refer to any elements of
gl_PerVertex, and remove all the associated ir_variables from the
shader at the end of ast-to-ir conversion.
Fixes piglit tests
linker/interstage-{pervertex,pervertex-in,pervertex-out}-redeclaration-unneeded.
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Normally when a built-in array (such as gl_ClipDistance) is
redeclared, we call get_variable_being_redeclared() to do the
redeclaration, and it in turn calls check_builtin_array_max_size() to
make sure that the redeclared array size isn't too large.
However when a built-in array is redeclared as part of redeclaring
gl_in, we don't call get_variable_being_redeclared() (since the
individual built-ins aren't each represented by their own ir_variable
anymore). So we need to add an explicit call to
check_builtin_array_max_size() to make sure the new array size isn't
too large.
Note: at the moment this is redundant with a test that's done at link
time, so there's no change to piglit results. But the patch that
follows will prevent link errors from being reported if gl_PerVertex
isn't used, so in order to prevent that patch from causing
regressions, we need to add the compile check now. Besides, it's
nicer to report this error at compile time anyhow.
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
The queries GEOMETRY_VERTICES_OUT, GEOMETRY_INPUT_TYPE, and
GEOMETRY_OUTPUT_TYPE (defined by GL 3.2) differ from the corresponding
queries in ARB_geometry_shader4 in the following ways:
- They use different enum values
- They can only be queried; they cannot be set.
- Attempting to query them yields INVALID_OPERATION if the program is
not linked, or lacks a geometry shader.
This patch switches us over from the ARB_geometry_shader4 behaviour to
the GL 3.2 behaviour.
Fixes piglit test query-gs-prim-types.
v2: Improve comment above has_core_gs.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
When program_resource_visitor visits variables that were created by
lower_named_interface_blocks, it needs to do extra work to un-do the
effects of lower_named_interface_blocks and construct the proper API
names.
Fixes piglit test
spec/glsl-1.50/execution/interface-blocks-api-access-members.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
These variables will need to be treated specially by
program_resource_visitor, so that they can be addressed through the
API using their interface block name (and array index, for interface
block arrays).
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Later patches will use this information to do proper error checking of
interpolation qualifiers that appear inside of interface blocks.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
In future patches, we will need this in order to interpret
interpolation qualifiers that appear inside interface blocks.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Although in principle there is no hardware limitation that prevents
gl_MaxGeometryInputComponents from being set to 128 on Gen7, we have
the following limitations in the vec4 compiler back end:
- Registers assigned to geometry shader inputs can't be spilled or
later re-used for any other purpose.
- The last 16 registers are set aside for the "MRF hack", meaning they
can only be used to send messages, and not for general purpose
computation.
- Up to 32 registers may be reserved for push constants, even if there
is sufficient register pressure to make this impractical.
A shader using 128 geometry input components, and having an input type
of triangles_adjacency, would use up:
- 1 register for r0 (which holds URB handles and various pieces of
control information).
- 1 register for gl_PrimitiveID.
- 102 registers for geometry shader inputs (17 registers per input
vertex, assuming DUAL_INSTANCED dispatch mode and allowing for one
register of overhead for gl_Position and gl_PointSize, which are
present in the URB map even if they are not used).
- Up to 32 registers for push constants.
- 16 registers for the "MRF hack".
That's a total of 152 registers, which is well over the 128 registers
the hardware supports.
Fortunately, the GLSL 1.50 spec allows us to reduce
gl_MaxGeometryInputComponents to 64. Doing that frees up 48
registers, brining the total down to 104 registers, leaving 24
registers available to do computation.
Fixes piglit test
spec/glsl-1.50/execution/geometry/max-input-components.
Reviewed-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
This is similar to what we do for 16-wide vs 8-wide fragment shaders.
First we try compiling the geometry shader in DUAL_OBJECT mode. If we
can't do that without spilling, we fall back on DUAL_INSTANCED mode,
which should require less spilling (since it uses an interleaved
layout of payload registers).
In an ideal world we'd fall back to SINGLE mode, which would allow us
to interleave general-purpose registers too (resulting in even less
likelihood of spilling). But at the moment, the vec4 generator and
visitor classes don't have the infrastructure to interleave general
purpose registers, so DUAL_INSTANCED is the best we can do.
As a side benefit this paves the way for implementing instanced
geometry shaders (which are incompatible with DUAL_OBJECT mode).
Since most geometry shaders used in piglit testing are small,
DUAL_INSTANCED mode won't get exercised very much in a normal piglit
run. To force DUAL_INSTANCED mode to be used for all geometry
shaders, set INTEL_DEBUG=nodualobj.
Reviewed-by: Eric Anholt <eric@anholt.net>
Geometry shaders that run in "DUAL_INSTANCED" mode store their inputs
in vec4's. This means that when compiling gl_PointSize input
swizzling (a MOV instruction which uses a geometry shader input as
both source and destination), we need to do two things:
- Set force_writemask_all to ensure that the MOV happens regardless of
which channels are enabled.
- Set the source register region to <4;4,1> (instead of <0;4,1> to
satisfy register region restrictions.
v2: move the source register region fixup to the top of
vec4_generator::generate_vec4_instruction(), so that it applies to all
instructions rather than just MOV.
Reviewed-by: Eric Anholt <eric@anholt.net>
In future patches, this will allow us to first try compiling a
geometry shader in DUAL_OBJECT mode (which is more efficient but uses
more registers) and then if spilling is required, fall back on
DUAL_INSTANCED mode.
Reviewed-by: Eric Anholt <eric@anholt.net>
Otherwise the scheduler would be invoked with prog_data->total_grf ==
0, causing havoc.
In a future patch, this will allow us to try compiling a geometry
shader in DUAL_OBJECT mode with spilling disabled, and then fall back
to DUAL_INSTANCED mode if that failed.
Reviewed-by: Eric Anholt <eric@anholt.net>
When geometry shaders are operated in "single" or "dual instanced"
mode, a single set of geometry shader inputs is interleaved into the
thread payload (with each payload register containing a pair of
inputs) in order to save register space.
This patch modifies vec4_visitor::lower_attributes_to_hw_regs so that
it can handle the interleaved format.
Reviewed-by: Eric Anholt <eric@anholt.net>
All geometry shaders begin this instruction:
mov(1) g0.2<1>:ud 0x0:ud { align1 }
which sets up GRF0 properly for scratch reads and writes. Since this
instruction has a SIMD size of 1, it will only have an effect if the
first channel is enabled. In practice, the hardware seems to always
dispatch geometry shaders with the first channel enabled, but I can't
find anything in the docs to guarantee that.
So to be on the safe side, set force_writemask_all on the instruction,
which guarantees that it will have the desired effect regardless of
which channels are enabled.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
When lower_named_interface_blocks lowers a built-in interface block
member to an ir_variable, it needs to set explicit_location in the
ir_variable. Otherwise the linker gets confused and treats the
variable as a generic varying.
Fixes the following piglit tests, which were regressed by commit
63974c0 (glsl: Simplify the interface to
link_invalidate_variable_locations):
- clip-distance-bulk-copy
- clip-distance-in-bulk-read
- clip-distance-in-explicitly-sized
- clip-distance-in-param
- clip-distance-in-values
- core-inputs
- gs-redeclares-both-pervertex-blocks
- gs-redeclares-pervertex-in-only
- redeclare-pervertex-subset-vs-to-gs
- unsized-in-named-interface-block-gs
- unsized-in-named-interface-block-multiple
- unsized-in-unnamed-interface-block-gs
- unsized-in-unnamed-interface-block-multiple
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=70820
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
This should never have been in the program key in the first place,
since it's determined by the shader source, not by GL state. Change
the code to just refer to gl_program::UsesClipDistanceOut directly.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
This will make it easier for back-ends to share code between geometry
shader and vertex shader compilation. Also, it is renamed to
"UsesClipDistanceOut" to clarify that (a) in geometry shaders, it
refers to the gl_ClipDistance output rather than the gl_ClipDistance
input, and (b) it is irrelevant in fragment shaders.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Since gl_ClipDistance is lowered from an array of floats to an array
of vec4's during compilation, transform feedback has special logic to
keep track of the pre-lowered array size so that attempting to perform
transform feedback on gl_ClipDistance produces a result with the
correct size.
Previously, this special logic always consulted the vertex shader's
size for gl_ClipDistance. This patch fixes it so that it uses the
geometry shader's size for gl_ClipDistance when a geometry shader is
in use.
Fixes piglit test spec/glsl-1.50/transform-feedback-type-and-size.
v2: Change the type of LastClipDistanceArraySize to "unsigned", and
clarify the comment above it.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
We've always overriden
ctx->Const.{Vertex,Fragment}Program.MaxTextureImageUnits to reflect
the number of texture image units supported by the hardware (rather
than using the default values assigned by Mesa core) so it seems
sensible to do that for GeometryProgram.MaxTextureImageUnits too. We
set it to 0 if geometry shaders aren't supported.
Once that is done, we can just unconditionally add
GeometryProgram.MaxTextureImageUnits to MaxCombinedTextureImageUnits.
Fixes piglit test "spec/glsl-1.50/built-in
constants/gl_MaxCombinedTextureImageUnits".
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
The encoding of constant, relative, and relative-const src registers is
a bit more complex than originally thought, which gives an extra bit to
encode const reg # at expense of taking a bit from relative offset.
In most cases a3xx seems to actually use a scheme whereby it can encode
an extra bit for const register. You have three possible encodings in
thirteen bits:
register: (11 bits for N.c)
00........... rN.c
relative: (10 bits for N)
010.......... r<a0.x + N>
011.......... c<a0.x + N>
const: (12 bits for N.c)
1............ cN.c
Which means we can deal w/ more consts than previously thought.
Signed-off-by: Rob Clark <robclark@freedesktop.org>
d3d10 requires that cube corners are filtered with accurate weights (that
is, the weight of the non-existing corner texel should be evenly distributed
to the other 3 texels). OpenGL does not require this (but recommends it).
This requires us to use different filtering code, since we need per-texel
weights which our 2d lerp doesn't (and can't) do. And of course the (now
per element) weights need to be adjusted too for it to work.
Invoke the new filtering code whenever there's an edge to keep things simpler,
as it will work for edges too not just corners but of course it's only needed
with corners.
More ugly code for not much gain but at least a hacked up cubemap demo
shows very nice corners now... Not sure yet if and how this should be
configurable...
v2: incorporate feedback from Jose, only use special corner filtering code
when there's a corner not when there's only an edge (as corner filtering code
is slower, though a perf difference was only measureable when always
forcing edge code). Plus some minor style fixes.
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
No driver uses it any more, and it's been replaced by megadrivers.
v2: Remove always-on conditional for NEED_LIBPROGRAM (review by Emil)
Reviewed-by: Matt Turner <mattst88@gmail.com> (v1)
Reviewed-by: Emil Velikov <emil.l.velikov@gmail.com>
v2: drop dridir now that it's unused.
v3: Fix linking after rebase when building just swrast from classic but a
drm-using gallium driver.
v4: Consistently put spaces around += in the updated Makefile.am block.
v5: Set a global driverAPI variable so loaders don't have to update to
createNewScreen2() (though they may want to for thread safety).
Reviewed-by: Matt Turner <mattst88@gmail.com> (v3)
Reviewed-by: Emil Velikov <emil.l.velikov@gmail.com>
This required some reordering of headers to ensure that the symbol name
redefines happened before any prototypes.
v2: drop dridir now that it's unused.
v3: Consistently put spaces around += in the updated Makefile.am blocks.
v4: Set a global driverAPI variable so loaders don't have to update to
createNewScreen2() (though they may want to for thread safety).
Reviewed-by: Matt Turner <mattst88@gmail.com> (v2)
Reviewed-by: Emil Velikov <emil.l.velikov@gmail.com>
i915 has symbols for formerly-shared code that conflict with i965, so we
define them away using gen-symbol-redefs.py. Options considered:
- This option. Downsides: The symbols in profiling and debugging don't
match the source. The symbol list may change in the future and we won't
notice without manually running the tool again.
- Use objcopy --localize-hidden to automatically demote our symbols to
locals. This didn't work on i965 due to c++ weak symbols (which can't
be localized), but could work on i915. We could do it on i915 only, but
it does produce libtool warnings at link time due to libtool not knowing
if the resulting .o file is safe to link (stupid libtool). Plus you end
up with different symbols of the same name, which is confusing for
debugging too. On the other hand, no future symbol conflicts long term.
- Write our own libelf tool that handles c++ weak symbols like we want and
apply it to all drivers. All the downsides of above, but applies
uniformly across drivers.
- Edit the files to just rename all the i915 or i965 symbols that
conflict. There are on the order of 100 that have a prefix we used to
share, so it would take a bit of typing. Fewest downsides, but still
can have conflicts long term.
Ultimately, this is the least invasive change at the moment, and we can
see if the "more symbol conflicts appear later" thing is a real concern or
not.
Note that the ability to compile a version of i915 without INTEL_DEBUG env
support is dropped. It's too useful.
v2: drop dridir now that it's unused.
v3: Consistently put spaces around += in the updated Makefile.am block.
v4: Set a global driverAPI variable so loaders don't have to update to
createNewScreen2() (though they may want to for thread safety).
Reviewed-by: Matt Turner <mattst88@gmail.com> (v2)
Reviewed-by: Emil Velikov <emil.l.velikov@gmail.com>
v2: drop dridir now that it's unused.
v3: Consistently put spaces around += in the updated Makefile.am block.
v4: Set a global driverAPI variable so loaders don't have to update to
createNewScreen2() (though they may want to for thread safety).
v5: Fix missed public symbol in nouveau. (caught by Emil)
Reviewed-by: Matt Turner <mattst88@gmail.com> (v2)
Reviewed-by: Emil Velikov <emil.l.velikov@gmail.com>
Previously, we've split things such that mesa core is in libdricore,
exposing the whole Mesa core interface in the global namespace, and the
i965_dri.so code all links against that. Along with polluting application
namespace terribly, it requires extra PLT indirections and prevents LTO.
Instead, we can build all of the driver contents into the same .so with
just a few symbols exposed to be referenced from the actual driver .so
file, allowing LTO and reducing our exposed symbol count massively.
FPS improvement on GLB2.7 with INTEL_NO_HW=1: 2.61061% +/- 1.16957% (n=50)
(without LTO, just the PLT reductions from this commit)
Note that the X Server requires commit
7ecfab47eb221dbb996ea6c033348b8eceaeb893 to successfully load this driver!
v2: Set a global driverAPI variable so loaders don't have to update to
createNewScreen2() (though they may want to for thread safety).
v3: Drop AM_CPPFLAGS addition (Emil pointed out I'd missed some cflags
that would be necessary, though only if we actually relied on them).
v4: Fix install with DESTDIR set.
Reviewed-by: Matt Turner <mattst88@gmail.com> (v1)
Reviewed-by: Emil Velikov <emil.l.velikov@gmail.com> (v2)
As we move to megadrivers, we are unable to build multiple drivers with
the same public global symbol per driver (Think an X Server with an intel
and a nouveau driver, and the X Server implementing indirect for both --
we have to actually talk to the right driver). By slipping the
driDriverAPI vtable into the driver's extension list, we can replace the
usage of the global symbol with usage of the loader-dlsym()ed driver
information.
v2: Pull in the hunk to avoid crashing on null driver_extensions. Thanks,
Emil!
Reviewed-by: Matt Turner <mattst88@gmail.com> (v1)
Reviewed-by: Chad Versace <chad.versace@linux.intel.com>
Reviewed-by: Emil Velikov <emil.l.velikov@gmail.com>
This will allow a megadrivers build to reference the actual driver being
loaded from the shared dri_util screen creation code.
v2: Fix indentation, fallback case in EGL (review by Emil).
Reviewed-by: Matt Turner <mattst88@gmail.com> (v1)
Reviewed-by: Chad Versace <chad.versace@linux.intel.com> (v1)
Reviewed-by: Emil Velikov <emil.l.velikov@gmail.com>
The previous interface relied on a static struct, which meant that the
driver didn't get a chance to edit the struct before the struct got used.
For megadrivers, I want struct specific to the driver being loaded.
v2: Fix the prototype in the docs (caught by Marek). Since the driver
name was in the function, we didn't need to also pass it in.
v3: Fix asprintf error checking (caught by Matt's gcc).
Reviewed-by: Matt Turner <mattst88@gmail.com> (v1)
Reviewed-by: Chad Versace <chad.versace@linux.intel.com>
Reviewed-by: Emil Velikov <emil.l.velikov@gmail.com>
Turns out already we have this nice mechanism for providing optional
things from the driver to the loader, and I was going to have to rename
the public global symbol to avoid conflicts when doing megadrivers.
While the former __driConfigOptions is technically loader interface, this
is the only loader that made use of that symbol. Continue paying
attention to it if we can't find the new option, to retain compatibility
with old drivers.
Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Chad Versace <chad.versace@linux.intel.com>
Reviewed-by: Emil Velikov <emil.l.velikov@gmail.com>
Based on a similar fix from Aaron Watry. It seems unlikely that we
will ever need a kernel-specific setting for this, and the Gallium API
doesn't support it. Remove kernel::max_block_size() altogether.
The gallium vbuf module, which we've been using for some time now, takes
care of uploading user-space vertex/index data into real buffers. The
upload code in the svga driver was unused.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
Print info about packing, format, type, and tiling. This will help debug
future issues with this fastpath.
Reviewed-by: Frank Henigman <fjhenigman@google.com>
Signed-off-by: Chad Versace <chad.versace@linux.intel.com>
Fixes texture corruption of Weston clients on cairo-glesv2 backend.
Commit 49ed599 introduced the bug.
Corruption occured when glTexSubImage called
intel_texsubimage_tiled_memcpy() with:
x,y=10,9
w,h=7,7
format=GL_ALPHA(0x1906)
type=GL_UNSIGNED_BYTE(0x1401)
gl_format=MESA_FORMAT_A8(0x18)
packing.alignemnt=4
The function miscalculated the source image's stride as w*cpp=7 without
taking into account the packing alignment. The actual stride was 8.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=70435
Reported-by: U. Artie Eoff <ullysses.a.eoff@intel.com>
Tested-by: Kristian Høgsberg <krh@bitplanet.net>
Reviewed-by:Frank Henigman <fjhenigman@google.com>
Signed-off-by: Chad Versace <chad.versace@linux.intel.com>
In commit 800610f (i965/fs: Improve accuracy of dFdy() to match
dFdx()) I unrolled the high-accuracy dFdy() computation from a single
SIMD16 instruction to two SIMD8 instructions because of text I found
in the i965 (gen4) PRM saying that instruction compression could not
be used in align16 mode. I couldn't find similar text in later
hardware docs, and I observed problems trying to use instruction
compression on align16 mode on Ivy Bridge, so I assumed that the
restriction still applied and the associated documentation had simply
been lost.
After consultation with the hardware engineers, it turns out this is
not the case. In point of fact, the restriction was dropped in gen5,
re-introduced in Ivy Bridge, and dropped again in Haswell. The reason
I didn't notice this is that in the Ivy Bridge documentation, the
restriction was in a different section, and described using different
language.
Now that we know that the restriction only applies to Gen4 and Ivy
Bridge, we can limit the unrolling to those platforms.
Tested on gen5, gen6, and gen7 (both Ivy Bridge and Haswell).
Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
From the GLSL 1.50 spec, section 4.3.8.1 (Input Layout Qualifiers):
The layout qualifier identifiers for geometry shader inputs are
layout-qualifier-id
points
lines
lines_adjacency
triangles
triangles_adjacency
And from section 4.3.8.2 (Output Layout Qualifiers)
The layout qualifier identifiers for geometry shader outputs are
layout-qualifier-id
points
line_strip
triangle_strip
max_vertices = integer-constant
We were erroneously allowing line_strip and triangle_strip to be used
as input qualifiers, and we were allowing lines, lines_adjacency,
triangles, and triangles_adjacency to be used as output qualifiers.
Fixes piglit tests "glsl-1.50-gs-{input,output}-layout-qualifiers *".
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
On DOTA2, framerate on dota2-de1.dem in windowed mode on my laptop
improves by 7.69854% +/- 0.909163% (n=3). In a microbenchmark hitting
this code path (wall time of piglit vbo-subdata-many), runtime decreases
from 0.8 to 0.05 seconds.
v2: Use out of range start/end instead of separate bool for the active
flag (suggestion by Jordan), fix double-upload in the stalling path.
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
The brw_prepare_vertices that sets up buffers[] depends on these
parameters, so don't let brw_prepare_vertices() skip it.
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Supporting this extension turns out to simplify our code a bit over not
supporting this extension, once the glBufferSubData() synchronization code
lands.
v2: Use 16 byte alignment like we do for uniform buffers, due to unaligned
access penalties.
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com> (v1)
This was mostly for the i915 system-memory VBO code, which we don't have
any more, but since that existed we've ended up producing dependencies on
it being there.
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
If glBufferData(), glBufferSubData(0, obj->Size), or similar happens, we
get a new drm_intel_bo for the buffer object, and thus need to re-upload
texture buffer state so we point at the new data.
Fixes the new piglit GL_ARB_texture_buffer_object/data-sync
Cc: "9.2" <mesa-stable@lists.freedesktop.org>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
The new function replaces four old functions: set_fragment/vertex/
geometry/compute_sampler_views().
Note: at this time, it's expected that the 'start' parameter will
always be zero.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Tested-by: Emil Velikov <emil.l.velikov@gmail.com>
Writing a 64-bit register value to memory is sufficiently complicated
that it makes sense to reuse this function rather than duplicating it.
Exposing it outside of gen6_queryobj.c means it needs a more descriptive
function name. It could probably be moved to brw_util.c or somewhere
else, but this works too.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Eric Anholt <eric@anholt.net>
The current callers just want to write a single register, so combining
the register read with a pipeline flush made sense. However, in the
future we'll want to do multiple register reads back to back, and we'll
only want to flush once.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Eric Anholt <eric@anholt.net>
The unit tests added in the previous commits prove some things about the
state of some internal data structures. The most important of these is
that all built-in input and output variables have explicit_location
set. This means that link_invalidate_variable_locations doesn't need to
know the range of non-generic shader inputs or outputs. It can simply
reset location state depending on whether explicit_location is set.
There are two additional assumptions that were already implicit in the
code that comments now document.
- ir_variable::is_unmatched_generic_inout is only used by the linker
when connecting outputs from one shader stage to inputs of another
shader stage.
- Any varying that has explicit_location set must be a built-in. This
will be true until GL_ARB_separate_shader_objects is supported.
As a result, the input_base and output_base parameters to
link_invalidate_variable_locations are no longer necessary, and the code
for resetting locations and setting is_unmatched_generic_inout can be
simplified.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
Validates:
- ir_variable::explicit_location should not be modified.
- If ir_variable::explicit_location is not set, ir_variable::location,
ir_variable::location_frac, and
ir_variable::is_unmatched_generic_inout must be reset to 0.
- If ir_variable::explicit_location is set, ir_variable::location
should not be modified. ir_variable::location_frac, and
ir_variable::is_unmatched_generic_inout must be reset to 0.
Previous unit tests have shown that all non-generic inputs / outputs
have explicit_location set.
v2: Split the link_invalidate_variable_locations interface change out to
a separate patch. Remove the vertex_in_builtin_without_explicit and
vertex_out_builtin_without_explicit tests. There was a lot of good
discussion about this on the mailing list to which I refer the
interested reader. Both changes suggested by Paul.
http://lists.freedesktop.org/archives/mesa-dev/2013-October/046652.html
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
This will make it easier to unit test this function in successive
patches. Also, correct the prototype in linker.h. It was... wrong.
v2: Split the interface change from adding the unit tests. Suggested by
Paul.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
Checks that the variables generated meet certain criteria.
- Geometry shader inputs have an explicit location.
- Geometry shader outputs have an explicit location.
- Fragment shader-only varying locations are not used.
- Geometry shader uniforms and system values don't have an explicit
location.
- Geometry shader constants don't have an explicit location and are
read-only.
- No other kinds of geometry variables exist.
It does not verify that an specific variables exist.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
Checks that the variables generated meet certain criteria.
- Fragment shader inputs have an explicit location.
- Fragment shader outputs have an explicit location.
- Vertex / geometry shader-only varying locations are not used.
- Fragment shader uniforms and system values don't have an explicit
location.
- Fragment shader constants don't have an explicit location and are
read-only.
- No other kinds of fragment variables exist.
It does not verify that an specific variables exist.
v2: Use _mesa_varying_slot_in_fs in
fragment_builtin.inputs_have_explicit_location. Suggested by Paul.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
Checks that the variables generated meet certain criteria.
- Vertex shader inputs have an explicit location.
- Vertex shader outputs have an explicit location.
- Fragment shader-only varying locations are not used.
- Vertex shader uniforms and system values don't have an explicit
location.
- Vertex shader constants don't have an explicit location and are
read-only.
- No other kinds of vertex variables exist.
It does not verify that an specific variables exist.
v2: Fix memory management mistakes in
common_builtin::string_starts_with_prefix. Clean up error message
reporting in common_builtin::no_invalid_variable_modes. Both suggested
by Paul.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
Ever since the addition of interface blocks with instance names, we have
had an implicit invariant:
var->type->is_interface() ==
(var->type == var->interface_type)
The odd use of == here is intentional because !var->type->is_interface()
implies var->type != var->interface_type.
Further, if var->type->is_array() is true, we have a related implicit
invariant:
var->type->fields.array->is_interface() ==
(var->type->fields.array == var->interface_type)
However, the ir_variable constructor doesn't maintain either invariant.
That seems kind of silly... and I tripped over it while writing some
other code. This patch makes the constructor do the right thing, and it
introduces some tests to verify that behavior.
v2: Add general-ir-test to .gitignore. Update the description of the
ir_variable invariant for arrays in the commit message. Both suggested
by Paul.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
After some discussions about the correct way to update
_mesa_program_state_string, I decided to make a unit test for the
function. It turns out that the function didn't work quite the way I
thought. The unit test proves that the code was already correct.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Cc: Anuj Phogat <anuj.phogat@gmail.com>
Silence a bunch of MSVC type conversion warnings.
Changed return type of S_FIXED to int32_t (signed). The result
is the same. It just seems more intuitive that a signed conversion
function should return a signed value.
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
This extension never saw any real use so remove it.
v2: also update tests/num_strings.cpp for 'make check'
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Dead code elimination would get rid of the extra instructions, but
skipping this saves iterations through the optimization loop.
From shader-db:
N Min Max Median Avg Stddev
x 14672 3 16 3 3.1334515 0.59904168
+ 14672 1 16 3 2.8955153 0.77732963
Difference at 95.0% confidence
-0.237936 +/- 0.0158798
-7.59342% +/- 0.506783%
(Student's t, pooled s = 0.693935)
Embarassingly, the classic shadow mapping shader:
void main() { }
used to require three iterations through the optimization loop.
With this patch, it only requires one (which makes no progress).
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
Previously one side could be UD while the other was float.
V2: Prefer float; apparently IVB can dispatch float ops faster. (Thanks
Eric)
Signed-off-by: Chris Forbes <chrisf@ijw.co.nz>
Reviewed-by: Eric Anholt <eric@anholt.net>
Gather unconditionally uses a header, but in some cases the
texture_offset value will be zero.
V2: Don't introduce a bogus conversion.
Signed-off-by: Chris Forbes <chrisf@ijw.co.nz>
Reviewed-by: Eric Anholt <eric@anholt.net>
Previously, Mesa followed the linkage rules outlined in the GLSL
1.20-1.40 specs, which (collectively) said that GLSL versions 1.10 and
1.20 could be linked together, but no other versions could be linked.
In GLSL 4.30, the linkage rules were relaxed so that any two desktop
GLSL versions can be linked together. This change was made because it
reflected the behaviour of nearly all existing implementations (see
Khronos bug 8463). Mesa was one of the few (perhaps the only)
exceptions to prohibit cross-linking of some GLSL versions.
Since the GLSL linkage rules were deliberately relaxed in order to
match the behaviour of existing implementations, it seems appropriate
to relax the rules in Mesa too (even though Mesa doesn't support GLSL
4.30 yet).
Note that linking ES and desktop shaders is still prohibited, as is
linking ES shaders having different GLSL versions.
Fixes piglit tests "shaders/version-mixing {interstage,intrastage}".
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Eric Anholt <eric@anholt.net>
Identifiers with double underscores are reserved, and using them has
undefined behavior according to the C++ spec. It's unlikely to make
any difference, but...
Tested-by: Tom Stellard <thomas.stellard@amd.com>
For seamless cube filtering it is necessary to determine new faces and new
coords per sample. The logic for this is _seriously_ complex (what needs
to happen is very "asymmetric" wrt face, x/y under/overflow), further
complicated by the fact that if the 4 samples are in a corner (meaning we
only have actually 3 samples, and all 3 are on different faces) then
falling off the edge is happening _both_ on x and y axis simultaneously.
There was a noticeable performance hit in mesa's cubemap demo when seamless
filtering was forced on (just below 10 percent or so in a debug build, when
disabling all filtering hacks, otherwise it would probably be a bit more) and
when always doing the logic, hence use a branch which it only does it if any
of the pixels in a quad (or in two quads) actually hit this. With that there
was no measurable performance hit in the cubemap demo (neither in a debug nor
release buidl), but this will vary (cubemap demo very rarely hits edges).
Might also be different on other cpus, as this forces SoA sampling path which
potentially can be quite a bit slower.
Note that as for corners, this code gets all the 3 samples which actually
exist right, and the 4th texel will simply be the same as one of the others,
meaning that filter weights will be a bit wrong. This however should be
enough for full OpenGL (but not d3d10) compliance.
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Using atomic function for ncs is superfluous since it is
protected by a mutex anyway. Also lock the mutex only once
while retrieving the next CS for submission.
Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Fixes this build error.
CC clientattrib.lo
In file included from ../../include/GL/glx.h:333,
from glxclient.h:45,
from clientattrib.c:32:
../../include/GL/glxext.h:275: error: redefinition of typedef ‘GLXContextID’
../../include/GL/glx.h:171: note: previous declaration of ‘GLXContextID’ was here
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=70591
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Everything necessary for these appears to be implemented. We'll want to
add more tests to guard against bugs, but it should be functionally
complete.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
translate_sse.c contains code for msabi on x86_64, but it appears to be
untested.
Currently arguments 1 and 2 passed to the generated code are moved as 32-bit
quantities into the registers used by sysvabi, irrespective of the architecture.
Since these may be pointers, they must be moved as 64-bit quantities to avoid
truncation.
Commit f4dd099171 disabled tranlate_sse.c on MinGW
x86_64, I don't know if was due to this issue, or a different one...
Signed-off-by: Jon TURNEY <jon.turney@dronecode.org.uk>
Reviewed-by: Brian Paul <brianp@vmware.com>
Cygwin also uses the msabi calling convention on x86_64, not the sysvabi calling
convention
Signed-off-by: Jon TURNEY <jon.turney@dronecode.org.uk>
Reviewed-by: Brian Paul <brianp@vmware.com>
ignored, and an empty message aborts the commit.
The heap is NX on 64-bit Cygwin, so use the rtasm_exec_malloc() implementation
which uses mmap() to allocate an anonymous page with execute permission, rather
than the one which just uses malloc().
Signed-off-by: Jon TURNEY <jon.turney@dronecode.org.uk>
Reviewed-by: Brian Paul <brianp@vmware.com>
With most of the virtual functions gone, brwInitVtbl() is now tiny.
Merging it into the caller allows us to delete the entire file.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Eric Anholt <eric@anholt.net>
Now that i915 and i965 have been split, the separation between
intelDestroyContext and brw_destroy_context is kind of arbitrary.
This patch replaces the only brw->vtbl.destroy() call with the body
of brw_destroy_context (the only implementation of that virtual
function).
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Eric Anholt <eric@anholt.net>
dri_bo_release is a helper function that calls drm_intel_bo_unreference
but then also sets the pointer to NULL. This is unnecessary, since
brw_destroy_context is called from intelDestroyContext, which also frees
brw completely.
If you're still trying to access them, you've got bigger problems.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Eric Anholt <eric@anholt.net>
Having almost the entire body of the function indented one level for a
check that should never happen seems silly. Just early return.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Eric Anholt <eric@anholt.net>
Since the i915/i965 split, there's only one implementation of this
virtual function. We may as well just call it directly.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Eric Anholt <eric@anholt.net>
Since the i915/i965 split, there's only one implementation of this
virtual function. We may as well just call it directly.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Eric Anholt <eric@anholt.net>
In commit f878d20 (glsl: Update ir_variable::max_ifc_array_access
properly), I accidentally used the wrong kind of check to determine
whether the variable being accessed was an interface instance (I used
var->get_interface_type() != NULL when I should have used
var->is_interface_instance()). As a result, if an unnamed interface
block contained a struct which contained an array,
update_max_array_access() would mistakenly interpret the struct as a
named interface block and try to dereference a null
var->max_ifc_array_access.
This patch corrects the check, fixing the null dereference.
Fixes piglit test interface-block-struct-nesting.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=70368
Reviewed-by: Matt Turner <mattst88@gmail.com>
In desktop GLSL, location qualifiers are case-insensitive. In GLSL
ES, they are case-sensitive. This patch handles the difference by
using a new function to match layout qualifiers,
match_layout_qualifier(), which calls either strcmp() or strcasecmp()
as appropriate.
Fixes piglit tests:
- layout-not-case-sensitive-in.geom
- layout-not-case-sensitive-max-vert.geom
- layout-not-case-sensitive-out.geom
- layout-not-case-sensitive.frag
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
This fixes a number of piglit crashes when running on a hacked up llvmpipe.
Signed-off-by: Dave Airlie <airlied@redhat.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
This just adds the missing bits so the ubo tests don't crash.
Signed-off-by: Dave Airlie <airlied@redhat.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
From the ARB_geometry_shader4 spec (section Geometry Shader outputs):
"The built-in special variable gl_Position is intended to hold the
homogeneous vertex position. Writing gl_Position is optional."
Signed-off-by: Dave Airlie <airlied@redhat.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
v2 (Paul Berry <stereotype441@gmail.com>: Split out to separate patch
(previously this was part of "glsl: add builtins for geometry
shaders.")
Signed-off-by: Dave Airlie <airlied@redhat.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Exposing 10-bit color configs confuses too many applications that try to
use the chooser to pick an 8 bit config. The chooser consider an fbconfig
with more bits a better match and will thus give a 10 bit config when an
application asks for a config with GLX_RED_SIZE 1 or 8.
One key example is glxinfo, which does this, and then doesn't specify that
it needs a config where GLX_DRAWABLE_TYPE has the GLX_WINDOW_BIT set.
This way it ends up with a 10 bit config that it can't use to create a
GLX window and fails to log extensions.
This reverts commit f354bcc177.
https://bugs.freedesktop.org/show_bug.cgi?id=70557
We can't perform DCE using the liveness pass between GVN and GCM
because it relies on the correct schedule, but GVN doesn't care about
preserving correctness - it's rescheduled later by GCM.
This patch makes dce_cleanup pass perform simple DCE
between GVN and GCM instead of relying on liveness pass.
Fixes https://bugs.freedesktop.org/show_bug.cgi?id=70088
Signed-off-by: Vadim Girlin <vadimgirlin@gmail.com>
Two extra instructions in some heroesofnewerth shaders, but a win for
everything else.
total instructions in shared programs: 1531352 -> 1530815 (-0.04%)
instructions in affected programs: 121898 -> 121361 (-0.44%)
Reviewed-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
This reverts commit 94d05bf87a as it has a
few problems:
- it breaks windows builds becuase env[LLVM_CXXFLAGS] is never set there
- it is merging not only rtti, but the whole cxxflags (defines etc)
which has proven to be a source of troubles (breaks debugging etc.)
This is required in order for clang to correctly handle the OpenCL C
barrier() builtin which has the following restrictions acording to
the OpenCL 1.1 Specification:
If barrier is inside a conditional statement, then all work-items must
enter the conditional if any work-item enters the conditional statement
and executes the barrier.
If barrier is inside a loop, all work-items must execute the barrier for
each iteration of the loop before any are allowed to continue execution
beyond the barrier.
By linking before otimizations, we can replace calls to barrier() with
calls to a target specific intrinsic which has the noduplicate attribute
This attribute prevents clang from performing optimizations which could
violate the above rules.
This attribute must be applied to the call instruction that invokes
the function, so it is not enough to add this attribute the barrier()
declaration.
As a bonus this will probably speed up compile times since we will no
longer need to run link-time optimizations.
During the recent bind_sampler_states() interface change in gallium
we changed the CSO single_sampler_done() function so that if we were
decreasing the number of sampler states bound in the driver, we'd
null-out the "extra/old" sampler states to unbind them. See commit
1e2fbf265.
However, we didn't make the corresponding fix for sampler views.
This caused an assertion to fail in the svga driver which checked
that the number of sampler views matched the number of sampler states.
This patch fixes cso_restore_sampler_views() so that it nulls-out
the extra/old sampler views if the number of new views is less than
the number of current/old views.
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
Use GL_MAP_INVALIDATE_RANGE, UNSYNCHRONIZED and FLUSH_EXPLICIT flags
when mapping VBOs during display list compilation. This mirrors what
we do for immediate-mode VBO building in vbo_exec_vtx_map().
This improves performance for applications which interleave display
list compilation with execution. For example:
glNewList(A);
glBegin/End prims;
glEndList();
glCallList(A);
glNewList(B);
glBegin/End prims;
glEndList();
glCallList(B);
Mesa's vbo module tries to combine the vertex data from lists A and B
into the same VBO when there's room. Before, when we mapped the VBO for
building list B, we did so with GL_MAP_WRITE_BIT only. Even though we
were writing to an unused part of the buffer, the map would stall until
the preceeding drawing call finished.
Use the extra map flags and FlushMappedBufferRange() to avoid the stall.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
Instead of checking width==height in four places, just do it in
_mesa_legal_texture_dimensions() where we do the other width, height,
depth checks. Similarly, move the check that cube map array depth is
a multiple of 6.
This change also fixes some missing cube dimension checks for the
glTexStorage[23]D() functions.
Remove width==height assertion in _mesa_get_tex_max_num_levels() since
that's called before the other size checks for glTexStorage.
Cc: "9.2" <mesa-stable@lists.freedesktop.org>
We can now add GBM support for the 10 bit/channel formats which lets us
create a gbm surface that we can use with KMS for display hardware that
support the format.
Signed-off-by: Kristian Høgsberg <krh@bitplanet.net>
We add support for the ARGB2101010 color format to the DRI image extension,
which allows DRI loaders to create a __DRIimage with this color format.
Signed-off-by: Kristian Høgsberg <krh@bitplanet.net>
This commit enables ARGB2101010 system framebuffers (that is, DRI drawables)
for the i965 drivers. This is done by generating DRI configs that advertise
this color format as well as teaching intelCreateBuffer to pick the right
color format when it sees such a DRI config.
Signed-off-by: Kristian Høgsberg <krh@bitplanet.net>
This extends the common dri driver infrastructure with the ability to create
__DRIconfigs for 10 bits/channel + 2 bit alphs formats. This still has
to be supported and requested by a driver, so this doesn't enable anthing yet.
Signed-off-by: Kristian Høgsberg <krh@bitplanet.net>
The EGLConfig doesn't have the rgba masks, only the rgba sizes. To
make sure a config is usable with a given GBM/KMS format, we need a way
to make sure the formats really match.
All callers now use the more correct rgba mask mechanism for filtering
out mathcing DRI configs. Even if depth and buffer size match, the
color component layout can be different, or in case or ARGB8888 and
ARGB2101010 the color components can even be different sizes.
Since anything that the depth check would reject is also rejected by
the rgba mask comparison, the depth parameter is redundant and not
specific enough. We should probably have removed it when the rgba
masks argument was introduced, but better late than never.
Signed-off-by: Kristian Høgsberg <krh@bitplanet.net>
Matching on visual depth to buffer size makes 8 bpc RGBA look similar to
10 bit RGB with 2 bit alphs - both have buffer size 32. Instead, build
the rgba masks from the visual data and use that for finding matching
DRI configs.
We need to keep the special case that allows us to match 24 bit visuals
to DRI configs with buffer size 32. We do that by creating an alpha
mask of "all the non-rgb bits" for 24 bit visuals and matching a second
time with that.
Signed-off-by: Kristian Høgsberg <krh@bitplanet.net>
Add information for RGB565 to the table of image formats so that we can
create a __DRIimage for that format. This in turn enables RGB565
wayland clients.
Signed-off-by: Kristian Høgsberg <krh@bitplanet.net>
With this patch Wayland clients can now ask EGL for RGB 565 format buffers
and attach them to a Wayland compositor.
Signed-off-by: Kristian Høgsberg <krh@bitplanet.net>
* The rtti fix actually dug up a bug in the scons build scripts.
* Autotools took the LLVM cpp and cxx flags, while scons only took
the cpp flags.
* This grabs the cxx flags and applies them where needed. We may
want to make the same change for the llvm cpp flags in scons.
* The only linux platform I can find with LLVM no-rtti is Ubuntu.
* Fixes bug #70471
Tested-by: Vinson Lee <vlee@freedesktop.org>
The original intent of the variable was to prevent adding
libdrm dependency for non drm drivers (swrast). This is
already handled with __NOT_HAVE_DRM_H, and with the recent
merge of the dri_util and drisw_util code this variable has
started causing build issues.
Eg. the following will fail
$ ./autogen.sh --with-dri-drivers=swrast --with-gallium-drivers=
$ make
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Reviewed-by: Andreas Boll <andreas.boll.dev@gmail.com>
When a geometry shader is active, the transform feedback primitive
type ("mode") needs to be validated against the geometry shader output
primitive type, not the primitive type passed to the glDraw*()
function.
Fixes the following piglit tests:
- glsl-1.50-geometry-primitive-types GL_LINES
- glsl-1.50-geometry-primitive-types GL_LINES_ADJACENCY
- glsl-1.50-geometry-primitive-types GL_LINE_STRIP
- glsl-1.50-geometry-primitive-types GL_LINE_STRIP_ADJACENCY
- glsl-1.50-geometry-primitive-types GL_TRIANGLES
- glsl-1.50-geometry-primitive-types GL_TRIANGLES_ADJACENCY
- glsl-1.50-geometry-primitive-types GL_TRIANGLE_FAN
Exposes previously hidden failures in the following piglit tests:
- glsl-1.50-geometry-primitive-id-restart GL_LINES other
- glsl-1.50-geometry-primitive-id-restart GL_LINES_ADJACENCY other
- glsl-1.50-geometry-primitive-id-restart GL_LINE_LOOP ffs
- glsl-1.50-geometry-primitive-id-restart GL_LINE_LOOP other
- glsl-1.50-geometry-primitive-id-restart GL_LINE_STRIP other
- glsl-1.50-geometry-primitive-id-restart GL_LINE_STRIP_ADJACENCY other
- glsl-1.50-geometry-primitive-id-restart GL_TRIANGLES other
- glsl-1.50-geometry-primitive-id-restart GL_TRIANGLES_ADJACENCY other
- glsl-1.50-geometry-primitive-id-restart GL_TRIANGLE_FAN ffs
- glsl-1.50-geometry-primitive-id-restart GL_TRIANGLE_FAN other
- glsl-1.50-geometry-primitive-id-restart GL_TRIANGLE_STRIP other
- glsl-1.50-geometry-primitive-id-restart GL_TRIANGLE_STRIP_ADJACENCY other
(These failures were previously hidden due to a flaw in the test: it
doesn't check for GL errors. I'll fix the test shortly).
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Ivy Bridge's "reorder enable" bit gives us a binary choice for the
order in which vertices from triangle strips are delivered to the
geometry shader. Neither choice follows the OpenGL spec, but setting
the bit is better, because it gets triangle orientation correct.
Haswell replaces the "reorder enable" bit with a new "reorder mode"
bit (which occupies the same location in the command packet). This
bit gives us a different binary choice, which affects both triangle
strips and triangle strips with adjacency. Setting the bit ("reorder
trailing") gives the proper order according to the OpenGL spec.
So in either case we want to set the bit.
On Ivy Bridge, fixes piglit test "triangle-strip-orientation".
On Haswell, fixes piglit tests "glsl-1.50-geometry-primitive-types
{GL_TRIANGLE_STRIP,GL_TRIANGLE_STRIP_ADJACENCY}" and
"glsl-1.50-geometry-tri-strip-ordering-with-prim-restart *".
v2: Rename the bit to "REORDER_TRAILING" for consistency with Haswell
docs.
Reviewed-by: Matt Turner <mattst88@gmail.com>
Despite the name, this field wasn't being set to the dispatch width at
all; it was always 8. The only place it was used was that the
constant buffer read length was aligned to it, and as far as I can
tell from the docs, there is no need to align this value to the
dispatch width; aligning it to a multiple of 8 is sufficient. So I've
just replaced it with a hardcoded 8.
v2: In gen6_wm_state, use brw->wm.base.push_const_size for consistency
with VS and GS state upload.
Reviewed-by: Eric Anholt <eric@anholt.net>
This patch populates the following built-in GLSL 1.50 variables based
on constants stored in ctx->Const:
- gl_MaxVertexOutputComponents
- gl_MaxGeometryInputComponents
- gl_MaxGeometryOutputComponents
- gl_MaxFragmentInputComponents
- gl_MaxGeometryTextureImageUnits
- gl_MaxGeometryOutputVertices
- gl_MaxGeometryTotalOutputComponents
- gl_MaxGeometryUniformComponents
- gl_MaxGeometryVaryingComponents
On i965/gen7, fixes all Piglit tests in "spec/glsl-1.50/built-in
constants/*" except for gl_MaxCombinedTextureImageUnits and
gl_MaxGeometryUniformComponents.
Reviewed-by: Matt Turner <mattst88@gmail.com>
Now that both vec4 and fs are dynamically assigning offsets, a lot of the
code is the same.
v2: Avoid passing around the next offset through the class. (Review by
Paul)
Reviewed-by: Paul Berry <stereotype441@gmail.com>
Note that the dropped comment in brw_context.h is mostly (better written)
in brw_binding_table.c as well.
Reviewed-by: Paul Berry <stereotype441@gmail.com>
It would be nice to be able to pack our binding table so that programs
that use 1 render target don't upload an extra BRW_MAX_DRAW_BUFFERS - 1
binding table entries. To do that, we need the compiled program to have
information on where its surfaces go.
v2: Rename size to size_bytes to be more explicit.
Reviewed-by: Paul Berry <stereotype441@gmail.com>
* As discussed on the mailing list,
forced no-rtti breaks C++ public
API's such as the Haiku C++ libGL.so
* -fno-rtti *can* be still set however
instead of blindly forcing -fno-rtti,
we can rely on the llvm-config
--cppflags output.
If the system llvm is built without
rtti (default), the no-rtti flag will be
present in llvm-config --cppflags
(which we pick up on)
If llvm is built with rtti
(REQUIRES_RTTI=1), then -fno-rtti is
removed from llvm-config --cppflags.
* We could selectively add / remove rtti
from various components, however mixing
rtti and non-rtti code is tricky and
could introduce missing symbols.
* This needs impact tested.
Reviewed-by: Francisco Jerez <currojerez@riseup.net>
Add simple plain C routines for NV12<->YV12 and YUYV<->UYVY
conversions. The NV12->YV12 conversion is commonly used, for instance
by VLC.
Reviewed-by: Christian König <christian.koenig@amd.com>
Textures that likely reside in VRAM, are mapped for reading and
don't require direct mapping should be staged into GTT, to avoid bad
performance. This fixes readback performance of VDPAU surfaces.
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
This new bind flag forces linear storage, but does not have other
side effects like R600_RESOURCE_FLAG_TRANSFER.
Reviewed-by: Christian König <christian.koenig@amd.com>
Currently it's hardcoded in the shader, so every change requires
compilation of the shader variant, killing the performance
in Serious Sam 3 and probably other apps.
This patch passes alpha_ref in the user sgpr and removes it from
the shader key.
Signed-off-by: Vadim Girlin <vadimgirlin@gmail.com>
Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>
This fixes the issue when dst and src is the same reg and operation on one
channel overwrites the source for other channels, e.g.:
UMUL TEMP[2].xyz, TEMP[0].xyzz, TEMP[2].xxxx
In this example the result of the operation on channel x is written in
TEMP[2].x and then used as a second source operand for channels y and z
instead of original value in TEMP[2].x.
This patch stores the results in temp reg and moves them to
dst after performing operation on all channels.
Fixes https://bugs.freedesktop.org/show_bug.cgi?id=70327
Signed-off-by: Vadim Girlin <vadimgirlin@gmail.com>
v2: Keep the random 32-bit only version of memcpy, since Ian says I
can't delete it without data proving it isn't useful.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
brw_context.h includes imports.h which includes compiler.h which already
defines these.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
These make it easy to convert a floating point value to a fixed point
numbers. The second parameter is the number of bits used for the
fractional part of the number.
It looks like core Mesa has similar functions already, but none that
allows an arbitrary number of fractional bits. The more generic version
is probably useful to everyone.
r600g apparently has an identical copy of the S_FIXED macro, but doesn't
include this file. I'm not sure what to do about that, so I'm just
going to leave it for now.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
This seems generally useful, so it may as well live in core Mesa.
In fact, the comment for ALIGN() in macros.h actually says to "see also"
ROUND_DOWN_TO, which...was in a driver somewhere.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
intel_batchbuffer_init() sets up initial batchbuffer state; it seems
like a reasonable place to initialize this flag.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Configuring which dirty flags we want sounds like a job for
brw_init_state().
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
It actually just wants generation checking, and brw->gen is the usual
way of doing that. In the future, we'll also want to check brw->hw_ctx,
which isn't available from the screen.
While we're changing the function signature, convert from camel case to
our usual naming conventions.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
There's no point in having two files for context functions. This patch
moves the code from intel_context.c into brw_context.c unmodified
(other than whitespace fixes).
Right now, this looks silly; future patches will merge functions and
tidy things up.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
brw_init_surface_formats already sets entries in TextureFormatsSupported
to true; it may as well take care of initializing it to false too.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
This flag is only used in one place, and is only set on one platform.
Just check for original Gen4 in the relevant function.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
This seems like a better place for it, and helps clean up
brwCreateContext (which is full of a lot of random stuff).
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
This was always set to false, and is only used for debugging.
To enable it, simply change the if (0) block and recompile.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Since each kind of device has its own brw_device_info structure, we can
simply store the URB and thread limits there. This eliminates all the
large if-ladders, and simplifies the context initialization code quite a
bit.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
This option was useful during initial development, but it's been ages
since I've heard of anyone using it. Plus, Gen7+ mandates separate
stencil, so it was really only useful on Sandybridge anyway.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
The idea is that struct brw_device_info should store statically-known
information about hardware features. Using the new family name in the
PCI ID table, we can easily grab the right structure.
This is basically the equivalent of intel_device_info in the kernel.
This patch also makes the new structure available from intel_screen, but
nothing uses it. Right now, it looks very redundant with existing
fields, but that will change.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
I removed this a while ago, since we never used it, but I'm finally
resurrecting the idea in the next commits.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Nothing uses the #define name, and it's not terribly useful - the
numerical ID serves the same purpose. The only thing we could really do
with it is generate slightly prettier preprocessed code. But who looks
at that?
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Using a helper function clarifies the context initialization code.
I would've liked to completely centralize it, but moving the optionCache
code from intelInitExtensions into here would've required setting flags
in the context, which seems like a waste.
v2: Rebase for the introduction of disable_derivative_optimization.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Now that intelInitContext isn't shared between i915 and i965, the split
is fairly arbitrary. This patch moves a bunch of the basic context
creation and generation checking code up to the top-level function
(and slightly earlier).
More will follow.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Now that there isn't an intel_context structure, the split between
brw_context.[ch] and intel_context.[ch] is rather awkward and arbitrary.
Removing intel_context.[ch] seems desirable, but not everything really
belongs in brw_context.[ch], either.
Moving INTEL_DEBUG handling into separate intel_debug.[ch] files should
make them relatively easy to find.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
"error" is a very generic name. dri_ctx_error is the name used in
intelInitContext(), which is more specific.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Geometry shader support is now working well, and adequately piglit
tested. There are just a few piglit failures left to fix. So there's
no need for an "experimental" warning anymore.
Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Chris Forbes <chrisf@ijw.co.nz>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Geometry shaders were the last thing we needed to finish before
turning on GLSL 1.50 and GL 3.2 support. They are now working well,
with just a few piglit failures left to fix.
Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Chris Forbes <chrisf@ijw.co.nz>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
With code dump enabled LLVM may generate disassembly during compilation.
Show this disassembly when available and prefer it to SI bytecode dump.
Reviewed-by: Tom Stellard <thomas.stellard@amd.com>
Signed-off-by: Jay Cornwall <jay@jcornwall.me>
Fix coord wrapping (and face selection too) in case of edges.
Unfortunately, the coord wrapping is way more complicated than what
the code did, as it depends on the face and the direction where the
texel falls off the face (the logic needed to get this right in fact
seems utterly ridiculous).
Also fix a bug in (y direction under/overflow) face selection.
And get rid of complicated cube corner handling. Just like edge case,
the coord wrapping was wrong and it seems very difficult to fix.
I'm near certain it can't always work anyway (though ordinary seamless
filtering on edge has actually a similar problem but not as severe)
because we don't have per-pixel face, hence could have multiple corner
texels which would make it very difficult to average the remaining texels
correctly. Hence simply pick a texel which would only have fallen off one
edge but not both instead, which is not quite accurate but actually I think
should be enough to meet OpenGL (but not d3d10) requirements.
v2: small fixes suggested by Brian, add some comments.
Reviewed-by: Brian Paul <brianp@vmware.com>
The previous limit of of 128*1024 was reported to cause frequent recompiles
in some apps due to shader variant thrashing on IRC in some apps leading
to noticeable lags.
Note that the LP_MAX_SHADER_VARIANTS limit (1024) was more or less impossible
to reach, since even simple fragment shaders without texturing (glxgears) used
more than twice than 128 instructions, hence the instruction limit would have
always been reached first (excluding things like trivial shaders not writing
color). Even with the new limit it is VERY likely the instruction limit is hit
first.
Should help with such lags due to recompiles (though other shader types have
their own limits, LP_MAX_SETUP_VARIANTS and DRAW_MAX_SHADER_VARIANTS, in
particular the latter seems a bit small (128)).
Reviewed-by: Brian Paul <brianp@vmware.com>
Fixes this build error.
CC imports.lo
../../src/mesa/main/imports.c: In function '_mesa_strtof':
../../src/mesa/main/imports.c:570:20: error: expected '=', ',', ';', 'asm' or '__attribute__' before 'loc'
../../src/mesa/main/imports.c:570:20: error: 'loc' undeclared (first use in this function)
../../src/mesa/main/imports.c:570:20: note: each undeclared identifier is reported only once for each function it appears in
../../src/mesa/main/imports.c:572:7: error: implicit declaration of function 'newlocale'
../../src/mesa/main/imports.c:572:23: error: 'LC_CTYPE_MASK' undeclared (first use in this function)
../../src/mesa/main/imports.c:574:4: error: implicit declaration of function 'strtof_l'
../../src/mesa/main/imports.c:580:1: warning: control reaches end of non-void function
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Now that libEGL has been fixed to not leak all kinds of symbols, gbm
links to its own copy of the libwayland-drm.a helper library. That means
we can't rely on comparing the addresses of a static vtable symbol in that
library to determine if a wl_buffer is a wl_drm_buffer. Instead, we
move the vtable into the wl_drm struct and use that for comparing.
https://bugs.freedesktop.org/show_bug.cgi?id=69437
Cc: 9.2 <mesa-stable@lists.freedesktop.org>
execinfo.h is not available on NetBSD.
Fixes this bulid error.
CC glapi_gentable.lo
glapi_gentable.c:44:22: fatal error: execinfo.h: No such file or directory
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
This was overriding the top-level .dir-locals.el causing some settings
(like forcing spaces instead of tabs!) to be lost.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
We should be able to safely set the framebuffer state without a
fragment shader bound. bind_ps_state will take care of updating the
necessary state bits later.
v2: check in update_db_shader_control
Fixes GL2ExtensionTests/egl_image_external/TestSimpleUnassociated.test
which is part of gles2/3 conformance suite. Here image external
textures are switched to be treated the same as 2D textures. These
can be associated with the fallback texture providing fixed sample
values of (0, 0, 0, 1).
The OES_EGL_image_external spec says:
"Sampling an external texture which is not associated with any EGLImage
sibling will return a sample value of (0,0,0,1)."
"External textures cannot be used with TexImage2D, TexSubImage2D,
CompressedTexImage2D, CompressedTexSubImage2D, CopyTexImage2D, or
CopyTexSubImage2D, and an INVALID_ENUM error will be generated if
this is attempted."
And quoting Chad:
"That's enforced in _mesa_TexImage*() by calling
legal_teximage_target(), and enforced in _mesa_TexSubImage*() by
calling legal_texsubmimage_target(). Each of the
legal_tex*image_target() functions reject external textures.
Therefore, allowing GL_TEXTURE_EXTERNAL_OES in store_texsubimage()
won't violate the above spec quote.
I think it's safe to allow GL_TEXTURE_EXTERNAL_OES in
store_texsubimage(), as long as the texture has only a single
plane. Luckily, that's the only type of external textures that
Mesa currently supports."
CC: Chad Versace <chad.versace@linux.intel.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
Signed-off-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
Extend the fast texture upload from BGRA X-tiled to include RGBA,
Alpha/Luminance, and Y-tiled. Speed improvements, measured with
mesa demos teximage program, on 256 x 256 texture, in MB/s, on a
Sandy Bridge (Ivy is comparable):
before after increase
BGRA/X-tiled 3266 4524 1.39x
BGRA/Y-tiled 1739 3971 2.28x
RGBA/X-tiled 474 4694 9.90x
RGBA/Y-tiled 477 3368 7.06x
L/X-tiled 1268 1516 1.20x
L/Y-tiled 1439 1581 1.10x
v2: Cosmetic changes only: reformat and reword comments, make doxygen-friendly,
rename variables, use existing macros, add an assert.
Signed-off-by: Frank Henigman <fjhenigman@google.com>
Reviewed-and-tested-by: Chad Versace <chad.versace@linux.intel.com>
This is part of the prep for megadrivers, which won't allow using a single
global symbol due to the fact that there will be multiple drivers built
into the same dri.so file. For that, we'll need screen init to take a
reference to the driver to set up this vtable.
v2: Fix two missed references to driDriverAPI.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org> (v1)
The intel_screen.c used to be a dispatch to one of 3 driver functions, but
was down to 1, so it was kind of a waste. In addition, it was trying to
free all of the data that might have been partially freed in the kernel
3.6 check (which comes after intelInitContext, and thus might have had
driverPrivate set and result in intelDestroyContext() doing work on the
freed data). By moving the driverPrivate setup earlier, we can use
intelDestroyContext() consistently and avoid such problems in the future.
v2: Adjust the prototype of brwCreateContext to use the proper enum
(fixing a compiler warning in some builds)
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org> (v1)
If bufmgr didn't get created, then screen creation failed, and we never
should have got here in the first place. This was added by Chris Wilson
in 2010 with no explanation for why it would be needed.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
i965, i915, radeon, r200, swrast, and nouveau were mostly trying to do the
same logic, except where they failed to. Notably, swrast had code that
appeared to try to enable GLES1/2 but forgot to set api_mask (thus
preventing any gles context from being created), and the non-intel drivers
didn't support MESA_GL_VERSION_OVERRIDE.
nouveau still relies on _mesa_compute_version(), because I don't know what
its limits actually are, and gallium drivers don't declare limits up front
at all. I think I've heard talk about doing so, though.
v2: Compat max version should be 30 (noted by Ken)
Drop r100's custom max version check, too (noted by Emil Velikov)
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
The only important difference was not calling drmGetVersion, and making
the swrast extension vtable. That doesn't justify duplicating the other
330 lines of code.
v2: fix the scons build (code by Emil Velikov)
v3: fix scons build with swrast-only (code by Emil Velikov)
v4: Drop the new define I added, when we already have __NOT_HAVE_DRM_H.
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Looking at Lightsmark's shaders, the way we used MRFs (or in gen7's
case, GRFs) was bad in a couple of ways. One was that it prevented
compute-to-MRF for the common case of a texcoord that gets used
exactly once, but where the texcoord setup all gets emitted before the
texture calls (such as when it's a bare fragment shader input, which
gets interpolated before processing main()). Another was that it
introduced a bunch of dependencies that constrained scheduling, and
forced waits for texture operations to be done before they are
required. For example, we can now move the compute-to-MRF
interpolation for the second texture send down after the first send.
The downside is that this generally prevents
remove_duplicate_mrf_writes() from doing anything, whereas previously
it avoided work for the case of sampling from the same texcoord twice.
However, I suspect that most of the win that originally justified that
code was in avoiding the WAR stall on the first send, which this patch
also avoids, rather than the small cost of the extra instruction. We
see instruction count regressions in shaders in unigine, yofrankie,
savage2, hon, and gstreamer.
Improves GLB2.7 performance by 0.633628% +/- 0.491809% (n=121/125, avg of
~66fps, outliers below 61 dropped).
Improves openarena performance by 1.01092% +/- 0.66897% (n=425).
No significant difference on Lightsmark (n=44).
v2: Squash in the fix for register unspilling for send-from-GRF, fixing a
segfault in lightsmark.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Acked-by: Matt Turner <mattst88@gmail.com>
For texturing from GRFs, we now have payloads of arbitrary sizes up to the
message length limit.
v2 (Kenneth Graunke): Rebase on intel_context -> brw_context change.
v3: Add some comment text.
v4: Change some magic 16s to BRW_MAX_MRF (noted by Ken). Leave the 11,
which is the magic "max sampler message length". BRW_MAX_MRF sizing
on the little int arrays is retained because I could see us needing to
extend in the future if we move to GRFs for FB writes (those go to at
least 12 long in a quick scan of the specs)
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org> (v2)
Acked-by: Matt Turner <mattst88@gmail.com>
This will let us coalesce into texture-from-GRF arguments, which would
otherwise be prevented due to the live interval for the whole vgrf
extending across all the MOVs setting up the channels of the message
v2 (Kenneth Graunke): Rebase for renames.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Now optimization passes will be able to look at the per-channel ranges.
v2: Rebase on various optimization pass changes.
v3 (Kenneth Graunke): Rename live_variables to live_intervals; split
introduction of invalidate_live_intervals() into a separate patch.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
When compacting the list of VGRFs, we patch up the live interval ranges
(which are indexed by VGRF number). Unfortunately, once we make
per-component data available, this will become too complicated to
maintain. Instead, simply invalidate them.
This was pulled out of a patch by Eric Anholt.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Eric Anholt <eric@anholt.net>
In compute_live_intervals(), start and end are shorter names for
the virtual_grf_start and virtual_grf_end class members.
Now that the fs_live_intervals class has arrays named start and end
which are indexed by var, rather than VGRF, reusing the name is
confusing. Plus, most of the code has been factored out, so using the
long names isn't as inconvenient.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Eric Anholt <eric@anholt.net>
This is the information we'll actually use to replace the
virtual_grf_start[]/end[] arrays.
No change in shader-db.
v2 (Kenneth Graunke): Rebase; minor comment updates.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
These blocks are about to grow some more code, and the indentation was
getting out of hand.
v2 (Kenneth Graunke): Rebase, minor typo fixes and style changes.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
For now, this simply sets live_intervals_valid = false, but in the
future it will do something more sophisticated.
Based on a patch by Eric Anholt.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Eric Anholt <eric@anholt.net>
This significantly improves our handling of VGRFs of size > 1.
Previously, we only marked VGRFs as def'd if the whole register was
written by a single instruction. Large VGRFs which were written
piecemeal would not be considered def'd at all, even if they were
ultimately completely written.
Without being def'd, these were then marked "live in" to the basic
block, often extending the range to preceding blocks and sometimes
even the start of the program.
The new per-component tracking gives more accurate live intervals,
which makes register coalescing more effective.
In the future, this should help with texturing from GRFs on Gen7+.
A sampler message might be represented by a 2-register VGRF which
holds the texture coordinates. If those are incoming varyings,
they'll be produced by two PLN instructions, which are piecemeal writes.
No reduction in shader-db instruction counts. However, code which
prints the live interval ranges does show that some VGRFs now have
smaller (and more correct) live intervals.
v2: Rebase on current send-from-GRF code requiring adding extra use[]s.
v3: Rebase on live intervals fix to include defs in the end of the
interval.
v4 (Kenneth Graunke): Rebase; split off a few preparatory patches;
add lots of comments; minor style changes; rewrite commit message.
v5 (Eric Anholt): whitespace nit.
Written-by: Eric Anholt <eric@anholt.net> [v1-3]
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org> [v4]
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Eric Anholt <eric@anholt.net> (v4)
num_vars was shorthand for the number of virtual GRFs. num_vgrfs is a
bit clearer. Plus, the next patch will introduce "vars" which are
distinct from vgrfs.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Eric Anholt <eric@anholt.net>
This has no functional effect, but should make subsequent changes a
little simpler.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Eric Anholt <eric@anholt.net>
From section 4.1.9 (Arrays) of the GLSL 4.40 spec (as of revision 7):
However, unless noted otherwise, blocks cannot be redeclared;
an unsized array in a user-declared block cannot be sized
through redeclaration.
The only place where the spec notes that interface blocks can be
redeclared is to allow for redeclaration of built-in interface blocks
such as gl_PerVertex. Therefore, user-defined interface blocks can
never be redeclared. This is a clarification of previous intent (see
Khronos bug 10659).
We were already preventing interface block redeclaration using the
same block name at compile time, but we weren't preventing interface
block redeclaration using the same instance name (and different block
names) at compile time. And we weren't preventing an instance name
from conflicting with a previously-declared ordinary variable.
In practice the problem would be caught at link time, but only because
of a coincidence: since ast_interface_block::hir() wasn't doing any
checking to see if the instance name already existed in the shader, it
was creating a second ir_variable in the shader having the same name
but a different type. Coincidentally, when the linker checked for
intrastage consistency of global variable declarations, it treated the
two declarations from the same shader as a conflict, so it reported a
link error.
But it seems dangerous to rely on that linker behaviour to catch
illegal redeclarations that really ought to be detected at compile
time.
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
This patch verifies that:
- The gl_PerVertex input interface block may only be redeclared in a
geometry shader, and that it may only be redeclared as gl_in[].
- The gl_PerVertex output interface block may only be redeclared in a
vertex or geometry shader, and that it may only be redeclared as a
non-array without an interface name.
- gl_PerVertex may not be redeclared as any other type of interface
block (i.e. as a uniform interface block).
As a side-effect, the code now keeps track of what the previous
declaration of gl_PerVertex was--this will be needed in future
patches.
Fixes piglit tests:
- spec/glsl-1.50/compiler/gs-redeclares-pervertex-in-with-incorrect-name.geom
- spec/glsl-1.50/compiler/gs-redeclares-pervertex-out-as-array.geom
- spec/glsl-1.50/compiler/gs-redeclares-pervertex-out-with-instance-name.geom
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
In later patches, we'll use this in order to implement the required
behaviour that after the gl_PerVertex interface block has been
redeclared, only members of the redeclared interface block may be
used.
v2: Update the function name and comment to clarify that we aren't
actually removing the variable from the symbol table, just disabling
it.
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
This will be used by future patches to change an ir_variable's
interface type when the gl_PerVertex built-in interface block is
redeclared.
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
This patch modifies the get_variable_being_redeclared() function so
that it no longer relies on the ast_declaration for the variable being
redeclared. In future patches, this will allow
get_variable_being_redeclared() to be used for processing
redeclarations of the built-in gl_PerVertex interface block.
v2: Also make get_variable_being_redeclared() static.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Note: we need to make an exception for the gl_PerVertex interface
block, since in geometry shaders it is allowed to be redeclared with
the instance name gl_in. Future patches will make redeclaration of
gl_PerVertex work properly.
Fixes piglit test
spec/glsl-1.50/compiler/interface-block-instance-name-uses-gl-prefix.vert.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Note: we need to make an exception for the gl_PerVertex interface
block, since built-in variables are allowed to be redeclared inside
it. Future patches will make redeclaration of gl_PerVertex work
properly.
Fixes piglit tests:
- spec/glsl-1.50/compiler/interface-block-array-elem-uses-gl-prefix.vert
- spec/glsl-1.50/compiler/named-interface-block-elem-uses-gl-prefix.vert
- spec/glsl-1.50/compiler/unnamed-interface-block-elem-uses-gl-prefix.vert
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Note: we need to make an exception for the gl_PerVertex interface
block, since this is allowed to be redeclared. Future patches will
make redeclaration of gl_PerVertex work properly.
Fixes piglit test
spec/glsl-1.50/compiler/interface-block-name-uses-gl-prefix.vert.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Note: some limited amount of redeclaration is actually allowed,
provided the shader is redeclaring the built-in gl_PerVertex interface
block. Support for this will be added in future patches.
Fixes piglit tests
spec/glsl-1.50/compiler/unnamed-interface-block-elem-conflicts-with-prev-{block-elem,global}.vert.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
GLSL reserves identifiers beginning with "gl_" or containing "__", but
we haven't been consistent about enforcing this rule. This patch
makes a new function to check whether identifier names are valid. In
the process it closes a loophole where we would previously allow
function argument names to contain "__".
v2: Rename check_valid_identifier() -> validate_identifier(). Add
curly braces in validate_identifier().
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
In commit e2660770731b018411fbe1620cacddaf8dff5287 (glsl: Keep track
of location for interface block fields), I neglected to update
glsl_type::record_key_compare to account for the fact that interface
types now contain location information. As a result, interface types
that differ only by their location information would not be properly
distinguished.
At the moment this is not a problem, because the only interface block
in which location information != -1 is gl_PerVertex, and gl_PerVertex
is always created in the same way. However, in the patches that
follow, we'll be adding new ways to create gl_PerVertex (by
redeclaring it), so we'll need location information to be handled
properly.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Although these interfaces can't be accessed directly by GLSL (since
they don't have an instance name), they will be necessary in order to
allow redeclarations of gl_PerVertex.
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Currently, we create just a single gl_PerVertex interface block for
geometry shader inputs. In later patches, we'll also need to create
an interface block for geometry and vertex shader outputs. Moving the
code into its own class will make reuse easier.
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
We use a location of -1 for variables which don't have their own
assigned locations--this includes ir_variables which represent named
interface blocks. Technically the location assigned to gl_in doesn't
matter, since gl_in is only accessed via its members (which have their
own locations). But it's nice to be consistent.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Calling radeon_drm_cs_flush from multiple threads might cause deadlocks,
fix this by immediately signaling the semaphore after waiting for it.
This is a candidate for the stable branch(es).
Partially fixes: https://bugs.freedesktop.org/show_bug.cgi?id=70123
v2: some fixes on commit message
Signed-off-by: Christian König <christian.koenig@amd.com>
Unless the polygon fill mode is different from PIPE_POLYGON_MODE_FILL,
so checking the the polygon mode is sufficient.
Testing done: no regression in polygon-mode-offset
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Not used since ages, and it wouldn't work at all with explicit derivatives now
(not that it did before as it ignored them but now the code would just use
the derivs pre-projected which would be quite random numbers).
v2: also get rid of 3 helper functions no longer used.
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
They need some special handling. Quite complicated.
Additionally, use the same code for implicit derivatives too if no_rho_approx
and no_quad_lod is set, because it seems while generally it should be ok
to use per quad lod for implicit derivatives there's at least some test which
insists that in case of cubemaps the shared lod value MUST come from a pixel
inside the primitive (due to the derivatives becoming different if a different
larger major axis is chosen).
v2: based on Brian's feedback, clean up code a bit.
And use sign bit of major axis instead of pre-select s/t/r sign for coord
mirroring (which should be the same in the end, saves 2 ands).
Also fix two bugs with select/mirror of derivatives, the minor axes need to
use major axis sign as well (instead of major derivative axis sign), and
don't mistakenly use absolute values of major derivative and inverse major
values.
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
There's two reasons for this:
1) even when ignoring rho approximation for cube maps, the result is still
not correct, but it's better as the max error at edges is now sqrt(2) instead
of 2 (which was a full mip level), same as it is for ordinary 2d maps when
doing rho approximations (so the error actually goes from factor 2 at edges and
sqrt(2) completely inside a face to sqrt(2) at edges and 0 inside a face).
2) I want to repurpose rho_no_approx for cubemaps for fully correct cubemap
derivatives (so don't need yet another debug var).
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
We were already setting the array size of unsized arrays that appeared
inside unnamed interface blocks, but we weren't updating
ir_variable::interface_type to reflect the new array size, causing
bogus link errors.
This patch causes array_sizing_visitor to keep track of all the
unnamed interface types it sees, and the ir_variables corresponding to
each one. After the visitor runs, a new function,
fixup_unnamed_interface_types(), adjusts each unnamed interface type
to correctly correspond with the array sizes in the ir_variables.
Fixes piglit tests:
- spec/glsl-1.50/execution/unsized-in-unnamed-interface-block-gs
- spec/glsl-1.50/execution/unsized-in-unnamed-interface-block-multiple
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
When multiple shaders of the same type access an interface block
containing an unsized array, we need to set the array size based on
the maximum array element accessed across all the shaders. This is
similar to what we already do with unsized arrays occurring outside of
interface blocks.
Note: one corner case is not yet addressed by these patches: the case
where one compilation unit defines an interface block containing
unsized arrays and another compilation unit defines the same interface
block containing sized arrays.
Fixes piglit test:
- spec/glsl-1.50/execution/unsized-in-named-interface-block-multiple
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Unsized arrays appearing inside named interface blocks now get a
proper size assigned by the array_sizing_visitor.
Fixes piglit tests:
- spec/glsl-1.50/execution/unsized-in-named-interface-block
- spec/glsl-1.50/execution/unsized-in-named-interface-block-gs
- spec/glsl-1.50/linker/unsized-in-named-interface-block
- spec/glsl-1.50/linker/unsized-in-named-interface-block-gs
- spec/glsl-1.50/linker/unsized-in-unnamed-interface-block-gs (*)
(*) is fixed by dumb luck--support for unsized arrays in unnamed
interface blocks will come in a later patch.
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
This patch modifies update_max_array_access() so that it updates
ir_variable::max_ifc_array_access to reflect the shader's use of
arrays appearing within interface blocks.
v2: Use an ordinary function in ast_array_index.cpp rather than a
virtual function in ir_rvalue. Avoid dereferencing NULL when handling
accesses to ordinary structs.
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
For interface blocks that contain arrays, this field will contain the
maximum element of each contained array that is accessed by the
shader. This is a first step toward supporting unsized arrays in
interface blocks.
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Currently, when converting an access to an array element from ast to
IR, we need to see if the array is an ir_dereference_variable, and if
so update the variable's max_array_access.
When we add support for unsized arrays in interface blocks, we'll also
need to account for cases where the array is an ir_dereference_record
and the record is an interface block.
To make this easier, move the update into its own function.
v2: Use an ordinary function in ast_array_index.cpp rather than a
virtual function in ir_rvalue.
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Although it's not explicitly stated in the GLSL 1.50 spec, unsized
arrays are allowed in interface blocks.
section 1.2.3 (Changes from revision 5 of version 1.5) of the GLSL
1.50 spec says:
* Completed full update to grammar section. Tested spec examples
against it:
...
* add unsized arrays for block members
And section 7.1 (Vertex and Geometry Shader Special Variables)
includes an unsized array in the built-in gl_PerVertex interface
block:
out gl_PerVertex {
vec4 gl_Position;
float gl_PointSize;
float gl_ClipDistance[];
};
Furthermore, GLSL 4.30 contains an example of an unsized array
occurring inside an interface block. From section 4.3.9 (Interface
Blocks):
uniform Transform { // API uses "Transform[2]" to refer to instance 2
mat4 ModelViewMatrix;
mat4 ModelViewProjectionMatrix;
vec4 a[]; // array will get implicitly sized
float Deformation;
} transforms[4];
This patch adds the parser rule to support unsized arrays inside
interface blocks. Later patches in the series will add the
appropriate semantics to handle them.
Fixes piglit tests:
- spec/glsl-1.50/execution/unsized-in-unnamed-interface-block
- spec/glsl-1.50/linker/unsized-in-unnamed-interface-block
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Interface declarations have two names associated with them: the block
name and the instance name. It's the block name that needs to be
passed to get_interface_instance(). This patch renames the argument
so that there's no confusion.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
BLORP performs blits by drawing a rectangle with a shader that samples
from the source texture, and writes color data to the destination.
The sampler always returns 32-bit RGBA float data, regardless of the
source format's component ordering or data type. Likewise, the render
target write message takes 32-bit RGBA float data, and converts it
appropriately. So the bulk of the work is already taken care of for us.
This greatly accelerates a lot of CopyTexSubImage calls, and makes
Legends of Aethereus playable on Ivybridge. At the default settings,
LOA continually blits between SRGBA8888 (the window format) and
RGBA16_FLOAT. Since neither BLORP nor our BLT paths supported this,
it fell back to meta, spending 33% of the CPU in floorf() converting
between floats and half-floats.
v2: Use != instead of ^ (suggested by Ian). Note that only
CopyTexSubImage is affected by this patch (caught by Eric).
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Chad Versace <chad.versace@linux.intel.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
The previous code for sRGB overrides assumes that the source and
destination formats are equal, other than the color space. This won't
be feasible when we add support for format conversions.
Here are a few cases, and how the old code handled them:
1. RGB8 -> SRGB8, MSAA ==> SRGB8 -> SRGB8
2. RGB8 -> SRGB8, single ==> RGB8 -> RGB8
3. SRGB8 -> RGB8, MSAA ==> RGB8 -> RGB8
4. SRGB8 -> RGB8, single ==> SRGB8 -> SRGB8
Apparently, preserving the behavior of #1 is important. When doing a
multisample to single-sample resolve, blending the samples together in
an sRGB correct fashion results in a noticably higher quality image.
It also is necessary to pass Piglit's EXT_framebuffer_multisample
accuracy color tests.
Paul, Eric, Anuj, and I talked about this, and aren't sure that it
matters in the other cases.
This patch preserves the behavior of #1, but otherwise reverts to
doing everything in linear space, changing the behavior of case #4.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Chad Versace <chad.versace@linux.intel.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
We could conceivably use BRW_SURFACEFORMAT_R24_UNORM_X8_TYPELESS for
Z24 source images, allowing conversions from Z24 to either Z16 or Z32F.
Unfortunately, we can't use it for destination images since it isn't
supported as a render target.
Using different formats for sources or destinations would be painful,
so for now, punt.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Chad Versace <chad.versace@linux.intel.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Currently, all that matters is that we copy the correct number of bits,
so any format that has 32-bits of data will work fine.
Once BLORP begins handling format conversions, the sampler will need to
correctly interpret the data. We don't need a depth format, but we do
need the right number of components and data type (FLOAT).
For Z32F, this means using R32_FLOAT.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Chad Versace <chad.versace@linux.intel.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Currently, all that matters is that we copy the correct number of bits,
so any format that has 16-bits of data will work fine.
Once BLORP begins handling format conversions, the sampler will need to
correctly interpret the data. We don't need a depth format, but we do
need the right number of components and data type (UNORM).
For Z16, this means using R16_UNORM.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Chad Versace <chad.versace@linux.intel.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Once blorp gains the ability to do format conversions, it's conceivable
that the source format may be texturable but not supported as a render
target. This would break Paul's code, which assumes that it can use the
render_target_format array even for the source format.
There are three ways to convert MESA_FORMAT enums to BRW_SURFACEFORMAT
enums:
1. brw_format_for_mesa_format()
This translates the Mesa format to the most equivalent BRW format.
2. brw->render_target_format[]
This is used for renderbuffers, and handles the subset of formats
that are renderable. However, it's not always equivalent, since
it overrides a few non-renderable formats. For example, it
converts B8G8R8X8_UNORM to B8G8R8A8_UNORM so it can be rendered to.
3. translate_tex_format()
This is used for textures. It wraps brw_format_for_mesa_format(),
but overrides depth textures, and one sRGB case on Gen4.
BLORP has a fourth function, which uses brw->render_target_format[]
and overrides depth formats (differently than translate_tex_format).
This patch makes the BLORP function to use brw_format_for_mesa_format()
for textures/source data, since not everything will be a render target.
It continues using brw->render_target_format[] for render targets, since
it needs the format overrides that provides.
We don't use translate_tex_format() since the additional overrides are
not useful or simply redundant.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Chad Versace <chad.versace@linux.intel.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
As we're moving towards expanding the number of subpixel
bits and the width of the variables used in the computations
we need to make this code a bit more centralized.
Signed-off-by: Zack Rusin <zackr@vmware.com>
Reviewed-by: José Fonseca <jfonseca@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
The code introduces two new 32bit integer multiplication opcodes which
can be used to produce correct 64 bit results. GLSL, OpenCL and D3D10+
require them. We use two seperate opcodes, because they match the
behavior of GLSL and OpenCL, are a lot easier to add than a single
opcode with multiple destinations and because there's not much (any)
difference wrt code-generation.
Signed-off-by: Zack Rusin <zackr@vmware.com>
Reviewed-by: José Fonseca <jfonseca@vmware.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
The UD values were getting set up as floats. This happened to work out
because they were used as the second argument where the first was a dword,
and gen6+ doesn't do source conversions. But it did trigger fulsim
warnings, and it meant if you used the push constant as the first operand
you would have been disappointed.
Reviewed-by: Paul Berry <stereotype441@gmail.com>
The EGL library has some references to x11 but it gets the link flags
from the XCB_DRI2_LIBS if and only if HAVE_EGL_PLATFORM_X11 is true.
The X11_LIBS variable was probably coming from a PKG_CHECK_MODULES (x11)
earlier in history.
If it is possible to have HAVE_EGL_DRIVER_GLX without HAVE_EGL_PLATFORM_X11
then the link flags for libX11 should be passed. However, it won't come
from X11_LIBS which is undefined.
Reported-by: Emil Velikov <emil.l.velikov@gmail.com>
Acked-by: Emil Velikov <emil.l.velikov@gmail.com>
Signed-off-by: Gaetan Nadon <memsize@videotron.ca>
The compiler cannot find the Xlib.h in the installed system headers.
All supplied include directives point to inside the mesa module.
The X11_CFLAGS variable is undefined (not defined in config.status).
It appears the intent was to use X11_INCLUDES defined in configure.ac.
The Xlib.h file is not installed on my workstation. It is supplied in
the libx11-dev package. This allows an X developer control over which
version of this file is used for X development.
Use to test: --enable-gallium-egl --enable-xlib-glx --disable-dri
Acked-by: Brian Paul <brianp@vmware.com>
Signed-off-by: Gaetan Nadon <memsize@videotron.ca>
The compiler cannot find the Xlib.h in the installed system headers.
All supplied include directives point to inside the mesa module.
The X11_CFLAGS variable is undefined (not defined in config.status).
It appears the intent was to use X11_INCLUDES defined in configure.ac.
The Xlib.h file is not installed on my workstation. It is supplied in
the libx11-dev package. This allows an X developer control over which
version of this file is used for X development.
Acked-by: Brian Paul <brianp@vmware.com>
Signed-off-by: Gaetan Nadon <memsize@videotron.ca>
The X11_CFLAGS variable is undefined (not defined in config.status).
It appears the intent was to use X11_INCLUDES defined in configure.ac.
It is used for building the code in the x11 subdir.
The build does not fail on this one as LIBDRM_CFLAGS happens to have
the inludedir value as the one for X11. It will not always be the case.
The option --enable-gallium-egl is required durimg configuration.
Acked-by: Brian Paul <brianp@vmware.com>
Signed-off-by: Gaetan Nadon <memsize@videotron.ca>
OutputSurfaces have simple YCbCr rendering functionality built in,
but so far only 4:2:0 subsampling worked correctly. This fixes 4:2:2
and 4:4:4 formats.
Reviewed-by: Christian König <christian.koenig@amd.com>
It doesn't work (decodes to garbage) with most videos on UVD 3.0. Worse
yet, it often results in random memory corruption or GPU hangs. Rumor
has it only the newest UVD hardware could do it anyway.
Reviewed-by: Christian König <christian.koenig@amd.com>
The DPB size calculations seem to be off; there is various random
corruption happening, even with advanced profile. Always assuming
a minimum number of references appears to fix it, similarly to
H.264. This might overallocate the DPB. Also clean up the SPS/PPS
field setup so that it matches VC-1 specifications better.
With these changes, all advanced profile VC-1 files I could get my
hand on work fine.
Reviewed-by: Christian König <christian.koenig@amd.com>
UVD can only support NV12 in the case of hardware decoding, but we
can still use all other formats for software decoding. Use the UNKNOWN
profile to signal that we're not interesting in hardware decoding.
v2: use profile instead of entrypoint
Reviewed-by: Christian König <christian.koenig@amd.com>
which contains -Wl,-Bsymbolic. If I understand it correctly, it prevents
symbols from clashing if multiple drivers are loaded at the same time.
Tested-by: Emil Velikov <emil.l.velikov@gmail.com>
Copy sechalf to the new register, otherwise we would read wrong HW registers.
Signed-off-by: Chia-I Wu <olv@lunarg.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
When the instruction to send the sampler message is forced uncompressed or
sechalf, send SIMD8 one even in SIMD16 mode.
Signed-off-by: Chia-I Wu <olv@lunarg.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
SIMD8 sampler messages are allowed in SIMD16 mode, and they could not work
without BRW_COMPRESSION_2NDHALF. Later PRMs (gen5 and later) do not
explicitly state whether BRW_COMPRESSION_2NDHALF is allowed, but they do have
examples using send with SecHalf. It should be safe to assume SecHalf is
valid.
Signed-off-by: Chia-I Wu <olv@lunarg.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
gl_PointSize is stored in the w component of VARYING_SLOT_PSIZ, but
the geometry shader infrastructure assumes that it should look for all
geometry shader inputs of type float in the x component. So when
compiling a geomtery shader that uses a gl_PointSize input, fix it up
during the shader prolog by moving the w component to the x component.
This is similar to how we emit fixups and workarounds for vertex
shader attributes.
Fixes piglit test spec/glsl-1.50/execution/geometry/core-inputs.
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
This corresponds to the lowering of gl_ClipDistance to
gl_ClipDistanceMESA for vertex and geometry shader outputs. Since
this lowering pass occurs after lower_named_interface blocks, it deals
with 2D arrays (gl_ClipDistance[vertex][clip_plane]) rather than 1D
arrays in an interface block
(gl_in[vertex].gl_ClipDistance[clip_plane]).
v2 (Paul Berry <stereotype441@gmail.com>): Fix indexing order for
gl_ClipDistance input lowering. Properly lower bulk assignment of
gl_ClipDistance inputs. Rework for GLSL 1.50 style geometry shaders.
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
v3 (Paul Berry <stereotype441@gmail.com>): Add comments and assertions
to clarify that the 2D version of clip distance is only used for
geometry shader inputs.
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Previously, builtin_variables.cpp was written assuming that we
supported ARB_geometry_shader4 style geometry shader inputs, meaning
that each built-in varying input to a geometry was supplied via an
array variable whose name ended in "In", e.g. gl_PositionIn or
gl_PointSizeIn.
However, in GLSL 1.50 style geometry shaders, things work
differently--built-in inputs are supplied to geometry shaders via a
built-in interface block called gl_in, which contains all the built-in
inputs using their usual names (e.g. the gl_Position input is supplied
to the geometry shader as gl_in[i].gl_Position).
This patch adds the necessary logic to builtin_variables.cpp to create
the gl_in interface block and populate it accordingly for geometry
shader inputs. The old ARB_geometry_shader4 style varyings are
removed, though they can easily be added back in the future if we
decide to support ARB_geometry_shader4.
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
This patch adds a "location" element to struct glsl_struct_field, so
that we can keep track of the gl_varying_slot associated with each
built-in geometry shader input.
In lower_named_interface_blocks, we use this value to populate the
"location" field in the ir_variable that stores each geometry shader
input.
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
For a few reasons.
1: In the (current) common case, these conditionals are never true. All
we're doing by checking them is slowing down MakeCurrent. The server
does these checks already anyway.
2: GLX >= 3.0 contexts may legally be made current without a bound
framebuffer.
This does not fix piglit/glx-create-context-current-no-framebuffer, but
is a prerequisite for fixing it.
Cc: "9.1 9.2" <mesa-stable@lists.freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
Signed-off-by: Adam Jackson <ajax@redhat.com>
The new sampler bind sends us NULL samplers, so we need to count
the number of valid samplers ourselves. This fixes ~500 piglit
regressions from the sampler rework.
While we're at it, let's also support start.
In 3DSTATE_DEPTH_BUFFER, we set Width and Height to the miptree slice's
physical dimensions. (Logical and physical dimensions may differ for
multisample surfaces).
However, in SURFACE_STATE, we always set Width and Height to the slice's
logical dimensions. We should do the same for 3DSTATE_DEPTH_BUFFER,
because the hw docs say so.
No Piglit regressions (-x glx -x glean) on Ivybridge with Wayland.
v2: No Piglit regressions, for real this time.
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Signed-off-by: Chad Versace <chad.versace@linux.intel.com>
Now, one can do the following to generate and read the i965 Doxygen:
cd $MESA_TOP/doxygen
make
firefox i965/index.html
Reviewed-by: Frank Henigman <fjhenigman@google.com>
Signed-off-by: Chad Versace <chad.versace@linux.intel.com>
The registers in the architecture register file don't share much in
common, so there's no point in grouping them together. Use the HW_REG
class instead. The vec4 backend already does this.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Accidentally pushed an old version of the patch.
v2: Set destination register using brw_null_reg().
Reviewed-by: Paul Berry <stereotype441@gmail.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
CSE would otherwise combine the two mul(8) emitted by [iu]mulExtended:
mul(8) acc0 x y
mach(8) null x y
mov(8) lsb acc0
...
mul(8) acc0 x y
mach(8) msb x y
Into:
mul(8) temp x y
mov(8) acc0 temp
mach(8) null x y
mov(8) lsb acc0
...
mov(8) acc0 temp
mach(8) msb x y
But mul(8) into the accumulator produces more than 32-bits of precision,
which is required and lost if multiplying into a general register and
moving to the accumulator.
Reviewed-by: Eric Anholt <eric@anholt.net>
These built-ins have two "out" parameters, which makes implementing them
efficiently with our current compiler infrastructure difficult. Instead,
implement them in terms of the existing ir_binop_mul IR (to return the
low 32-bits) and a new ir_binop_mul64 which returns the high 32-bits.
v2: Rename mul64 -> imul_high as suggested by Ken.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
i965 implements this with a single (multiple destination) instruction,
SUBB. Emitting SUBB directly from usubBorrow() would be ideal, but our
optimization passes don't know how to copy with expressions with
side-effects.
Radeon has an SUBB_UINT instruction that only generates the borrow
bit. I've chosen to go this route and implement usubBorrow() by doing the
subtraction and the borrow operations separately.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
i965 implements this with a single (multiple destination) instruction,
ADDC. Emitting ADDC directly from uaddCarry() would be ideal, but our
optimization passes don't know how to copy with expressions with
side-effects.
Radeon has an ADDC_UINT instruction that only generates the carry
bit. I've chosen to go this route and implement uaddCarry() by doing the
addition and the carry operations separately.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Calculates the carry out of the addition of two values and the
borrow from subtraction respectively. Will be used in uaddCarry() and
usubBorrow() built-in implementations.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
The only GLSL extension that is not enabled is AMD_vertex_shader_layer.
I think the standalone-compiler could enable this (as shading language
support is complete), but no driver enables it.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Infer whether or not to use ES based on the GLSL version (100 or 300 are
for ES). This replaces the --glsl-es command line option. Set various
compiler limits based on the minimums required for the specified GLSL
version.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Acked-by: Kenneth Graunke <kenneth@whitecape.org>
The choices aren't just 0 and 1, so using the enum names is much more
clear.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
This allows application developers to use Mesa's compiler as a
standalone validator for their shaders.
This is mostly a revert of commit 569f0e4.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Pull the data directly from the context like the other varying related
limits. The parser state shadow copies were added back when the parser
state didn't have a pointer to the context. There's no reason to do it
now days.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
gl_MaxVertexOutputVectors => ctx->Const.VertexProgram.MaxOutputComponents
gl_MaxFragmentInputVectors => ctx->Const.FragmentProgram.MaxInputComponents
v2: Add types so that the code compiles. Pointed out by Brian.
v3: Leave gl_MaxVaryingFloats et al. as-is. Suggested by Paul.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Brian Paul <brianp@vmware.com> [v2]
Reviewed-by: Marek Olšák <marek.olsak@amd.com> [v2]
Reviewed-by: Paul Berry <stereotype441@gmail.com> [v2]
Starting with OpenGL 3.2 input limits and output limits for stages may
not match. This means they need to be accounted separately.
No piglit regressions.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
No need to keep a copy of the message in system memory anymore,
since it should now be in GART memory on newer chips.
Signed-off-by: Christian König <christian.koenig@amd.com>
This fixes issues where get_rt_format would see a 0 format because the
nouveau_surface had not been properly initialized. Fixes crash on
supertuxkart startup (which still fails due to out-of-vram issues).
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Signed-off-by: Francisco Jerez <currojerez@riseup.net>
As of ARB_gpu_shader5, textureGather doesn't always read the
post-swizzle RED channel -- so we can't just look at the red swizzle
state.
Theoretically we could only flag the quirk if *some* green swizzle is in
use, but that's probably more trouble than it's worth.
Signed-off-by: Chris Forbes <chrisf@ijw.co.nz>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
- For HSW: Select the channel based on the component selected (swizzle
is done in HW)
- For IVB: Select the channel based on the swizzle state for the
component selected. Only apply the RG32F w/a if we actually want
green -- we're about to flag it regardless of swizzle state.
Signed-off-by: Chris Forbes <chrisf@ijw.co.nz>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
- For HSW: Select the channel based on the component selected (swizzle
is done in HW)
- For IVB: Select the channel based on the swizzle state for the
component selected. Only apply the RG32F w/a if we actually want
green -- we're about to flag it regardless of swizzle state.
Signed-off-by: Chris Forbes <chrisf@ijw.co.nz>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
- gsampler2DRect support
- optional `comp` parameter
Future patches will add shadow sampler support and
textureGatherOffsets().
Signed-off-by: Chris Forbes <chrisf@ijw.co.nz>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
ARB_gpu_shader5 introduces new variants of textureGather* which have an
explicit component selector, rather than relying purely on the sampler's
swizzle state.
This patch adds the GLSL plumbing for the extra parameter.
Signed-off-by: Chris Forbes <chrisf@ijw.co.nz>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Theoretically would work on Gen5 as well but requires GLSL 1.30, which
is not (yet) enabled by default there.
V2: Enable for Gen5 conditionally on GLSL version.
Signed-off-by: Chris Forbes <chrisf@ijw.co.nz>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Previously we special-cased textureSize() but this is the more correct
condition.
Signed-off-by: Chris Forbes <chrisf@ijw.co.nz>
Reviewed-by: Matt Turner <mattst88@gmail.com>
* This in essence means that Mesa would be
taking control of Haiku's OpenGL kit.
* This works by dispatching renderers from the
OpenGL add-ons directory
This patch fixes this Oracle Solaris Studio build error.
"../../src/glsl/ir_constant_expression.cpp", line 1398: Error: The function "isnormal" must have a prototype.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
All texture instructions can use offsets, not just TXF. Offsets into
the literals array were wrong, too.
Signed-off-by: Marek Olšák <marek.olsak@amd.com>
We return 0 for GL_NUM_SHADER_BINARY_FORMATS, so
GL_SHADER_BINARY_FORMATS should not write any data to the application
buffer.
Fixes piglit test 'arb_get_program_binary-overrun shader'.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
As we march over the source buffer we're uploading in pieces, we
need to memcpy from the current offset, not the start of the buffer.
Fixes graphical corruption when drawing very large vertex buffers.
Cc: "9.2" <mesa-stable@lists.freedesktop.org>
Reviewed-by: Matthew McClure <mcclurem@vmware.com>
This patch adds the lrint, lrintf, llrint, and llrintf rounding utility
functions. When packing unorm depth values, we will round to nearest.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
With commit cb1febb07, I have incorrectly removed HAVE_COMMON_DRI
assuming that swrast does not need to build the translations for
driconf options, as effectively swrast/drisw does not use them.
With the incoming unification work of dri and drisw, it makes
sense just to revert the offending hunk.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=70057
Reported-by: Vinson Lee <vlee@freedesktop.org>
Tested-by: Vinson Lee <vlee@freedesktop.org>
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Previously, we computed dFdy() using the following instruction:
add(8) dst<1>F src<4,4,0)F -src.2<4,4,0>F { align1 1Q }
That had the disadvantage that it computed the same value for all 4
pixels of a 2x2 subspan, which meant that it was less accurate than
dFdx(). This patch changes it to the following instruction when
c->key.high_quality_derivatives is set:
add(8) dst<1>F src<4,4,1>.xyxyF -src<4,4,1>.zwzwF { align16 1Q }
This gives it comparable accuracy to dFdx().
Unfortunately, align16 instructions can't be compressed, so in SIMD16
shaders, instead of emitting this instruction:
add(16) dst<1>F src<4,4,1>.xyxyF -src<4,4,1>.zwzwF { align16 1H }
We need to unroll to two instructions:
add(8) dst<1>F src<4,4,1>.xyxyF -src<4,4,1>.zwzwF { align16 1Q }
add(8) (dst+1)<1>F (src+1)<4,4,1>.xyxyF -(src+1)<4,4,1>.zwzwF { align16 2Q }
Fixes piglit test spec/glsl-1.10/execution/fs-dfdy-accuracy.
Acked-by: Chris Forbes <chrisf@ijw.co.nz>
Reviewed-by: Eric Anholt <eric@anholt.net>
The inferface/prototype in native_wayland_bufmgr.h uses boolean/int, as
well as the rest of the file. Convert to improve consistency and to
prevent gcc compiler warnings due to type miss-match.
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Clean up inconsistency in enum decoration:
- Use the undecorated enums where possible.
- MAX_PROGRAM_TEXTURE_GATHER_COMPONENTS_ARB remains decorated, since it
has no undecorated equivalent in GL4.
Signed-off-by: Chris Forbes <chrisf@ijw.co.nz>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=70054
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
The new surface channel select bits allow us to avoid having to
recompile the shader for this workaround.
Signed-off-by: Chris Forbes <chrisf@ijw.co.nz>
Reviewed-and-tested-by: Kenneth Graunke <kenneth@whitecape.org>
This allows us to use a different surface format for gather4, which is
required for R32G32_FLOAT to work on Gen7.
V4: - Only emit alternate surface state for shaders which will actually
use it.
- Pass a simple 'for_gather' flag rather than a function pointer.
The callee can decide what w/a to apply.
Signed-off-by: Chris Forbes <chrisf@ijw.co.nz>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Worst-case is that *every* texunit uses a format that needs overriding.
V4: Place the gather slots last, so shaders which don't use gather don't
get penalized by having a huge binding table.
Signed-off-by: Chris Forbes <chrisf@ijw.co.nz>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
gather4 GREEN channel against a surface with format R32G32_FLOAT doesn't work
correctly on IVB. w/a from bspec:
- use R32G32_FLOAT_LD = 0x97 instead, for gather4 only.
- select BLUE channel to read GREEN
Signed-off-by: Chris Forbes <chrisf@ijw.co.nz>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
V4: Only flag quirks if there are any uses of gather in the shader,
to avoid spurious recompiles just because someone happened to use
RG32F.
Signed-off-by: Chris Forbes <chrisf@ijw.co.nz>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Pretty much the same as the FS case. Channel select goes in the header,
V2: Less mangling.
V3: Avoid sampling at all, for degenerate swizzles.
Signed-off-by: Chris Forbes <chrisf@ijw.co.nz>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Lowers ir_tg4 (from textureGather and textureGatherOffset builtins) to
SHADER_OPCODE_TG4.
The usual post-sampling swizzle workaround can't work for ir_tg4,
so avoid doing that:
* For R/G/B/A swizzles use the hardware channel select (lives in the
same dword in the header as the texel offset), and then don't do
anything afterward in the shader.
* For 0/1 swizzles blast the appropriate constant over all the output
channels instead of sampling.
V2: Avoid duplicating header enabling block
V3: Avoid sampling at all, for degenerate swizzles.
Signed-off-by: Chris Forbes <chrisf@ijw.co.nz>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Adds the Gen7 message IDs, a new SHADER_OPCODE_TG4 pseudo-op, and
low-level support for emitting it via generate_tex().
V3: Updated for changes in master.
Signed-off-by: Chris Forbes <chrisf@ijw.co.nz>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
When used with a cube array in VS, failed assertion in ir_validate:
Assignment count of LHS write mask channels enabled not
matching RHS vector size (3 LHS, 4 RHS).
To fix this, swizzle the RHS correctly for the writemask.
This showed up in the ARB_texture_gather tests, which exercise cube
arrays in the VS.
Signed-off-by: Chris Forbes <chrisf@ijw.co.nz>
Cc: "9.2" <mesa-stable@lists.freedesktop.org>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Consider only the top-left and top-right pixels to approximate DDX in a 2x2
subspan, unless the application requests a more accurate approximation via
GL_FRAGMENT_SHADER_DERIVATIVE_HINT or this optimization is disabled from the
new driconf option disable_derivative_optimization.
This results in a less accurate approximation. However, it improves the
performance of Xonotic with Ultra settings by 24.3879% +/- 0.832202% (at 95.0%
confidence) on Haswell. No noticeable image quality difference observed.
The improvement comes from faster sample_d. It seems, on Haswell, some
optimizations are introduced to allow faster sample_d when all pixels in a
subspan have the same derivative. I considered SAMPLE_STATE too, which allows
one to control the quality of sample_d on Haswell. But it gave much worse
image quality without giving better performance comparing to this change.
No piglit quick.tests regression on Haswell (tested with v1).
v2: better guess for precompile program key
Signed-off-by: Chia-I Wu <olv@lunarg.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Chris Forbes <chrisf@ijw.co.nz>
We have the destination framebuffer object passed in; there's no need to
go digging around in the context.
Signed-off-by: Chris Forbes <chrisf@ijw.co.nz>
Reviewed-by: Chad Versace <chad.versace@linux.intel.com>
Using it encourages the (IMHO worrying) practice of leaving member
variables uninitialized in constructor definitions. This macro
shouldn't be necessary anymore after the last patch series fixing all
its users to initialize all member variables from the class
constructor. Remove it.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
All member variables of glsl_to_tgsi_instruction are already being
initialized from its implicitly defined constructor, it's not
necessary to use rzalloc to allocate its memory.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
All member variables of ir_to_mesa_instruction are already being
initialized from its implicitly defined constructor, it's not
necessary to use rzalloc to allocate its memory.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
All member variables of vec4_live_variables are already being
initialized from its constructor, it's not necessary to use rzalloc to
allocate its memory, and doing so makes it more likely that we will
start relying on the allocator to zero out all memory if the class is
ever extended with new member variables.
That's bad because it ties objects to some specific allocation scheme,
and gives unpredictable results when an object is created with a
different allocator -- Stack allocation, array allocation, or
aggregation inside a different object are some of the useful
possibilities that come to my mind.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
All member variables of fs_live_variables are already being
initialized from its constructor, it's not necessary to use rzalloc to
allocate its memory, and doing so makes it more likely that we will
start relying on the allocator to zero out all memory if the class is
ever extended with new member variables.
That's bad because it ties objects to some specific allocation scheme,
and gives unpredictable results when an object is created with a
different allocator -- Stack allocation, array allocation, or
aggregation inside a different object are some of the useful
possibilities that come to my mind.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
All member variables of fs_inst are already being initialized from its
constructor, it's not necessary to use rzalloc to allocate its memory,
and doing so makes it more likely that we will start relying on the
allocator to zero out all memory if the class is ever extended with
new member variables.
That's bad because it ties objects to some specific allocation scheme,
and gives unpredictable results when an object is created with a
different allocator -- Stack allocation, array allocation, or
aggregation inside a different object are some of the useful
possibilities that come to my mind.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
All member variables of ip_record are already being initialized from
its constructor, it's not necessary to use rzalloc to allocate its
memory, and doing so makes it more likely that we will start relying
on the allocator to zero out all memory if the class is ever extended
with new member variables.
That's bad because it ties objects to some specific allocation scheme,
and gives unpredictable results when an object is created with a
different allocator -- Stack allocation, array allocation, or
aggregation inside a different object are some of the useful
possibilities that come to my mind.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
The cfg_t object relies on the memory allocator zeroing out its
contents before it's initialized, which is quite an unusual practice
in the C++ world because it ties objects to some specific allocation
scheme, and gives unpredictable results when an object is created with
a different allocator -- Stack allocation, array allocation, or
aggregation inside a different object are some of the useful
possibilities that come to my mind. Initialize all fields from the
constructor and stop using the zeroing allocator.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
The bblock_t object relies on the memory allocator zeroing out its
contents before it's initialized, which is quite an unusual practice
in the C++ world because it ties objects to some specific allocation
scheme, and gives unpredictable results when an object is created with
a different allocator -- Stack allocation, array allocation, or
aggregation inside a different object are some of the useful
possibilities that come to my mind. Initialize all fields from the
constructor and stop using the zeroing allocator.
v2: Use zero initialization for numeric types instead of default construction.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
All member variables of ast_type_qualifier are already being
initialized from its implicitly defined constructor, it's not
necessary to use rzalloc to allocate its memory.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
All member variables of ast_node are already being initialized from
its constructor, but some of its derived classes were leaving members
uninitialized -- Fix them.
Using rzalloc makes it more likely that we will start relying on the
allocator to zero out all memory if the class is ever extended with
new member variables. That's bad because it ties objects to some
specific allocation scheme, and gives unpredictable results when an
object is created with a different allocator -- Stack allocation,
array allocation, or aggregation inside a different object are some of
the useful possibilities that come to my mind.
v2: Use NULL initialization instead of default construction for pointers.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
The vec4_instruction object relies on the memory allocator zeroing out
its contents before it's initialized, which is quite an unusual
practice in the C++ world because it ties objects to some specific
allocation scheme, and gives unpredictable results when an object is
created with a different allocator -- Stack allocation, array
allocation, or aggregation inside a different object are some of the
useful possibilities that come to my mind. Initialize all fields from
the constructor and stop using the zeroing allocator.
Reviewed-by: Paul Berry <stereotype441@gmail.com>
Reviewed-by: Chad Versace <chad.versace@linux.intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
The _mesa_glsl_parse_state object relies on the memory allocator
zeroing out its contents before it's initialized, which is quite an
unusual practice in the C++ world because it ties objects to some
specific allocation scheme, and gives unpredictable results when an
object is created with a different allocator -- Stack allocation,
array allocation, or aggregation inside a different object are some of
the useful possibilities that come to my mind. Initialize all fields
from the constructor and stop using the zeroing allocator.
Reviewed-by: Paul Berry <stereotype441@gmail.com>
Reviewed-by: Chad Versace <chad.versace@linux.intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Several C++ source files include "main/uniforms.h" from an extern "C"
block, which is both unnecessary, because "uniforms.h" already checks
for a C++ compiler and sets the right linkage, and incorrect, because
the header file includes other C++ headers ("glsl_types.h" and
"ir_uniform.h") that are supposed to get C++ linkage.
Reviewed-by: Paul Berry <stereotype441@gmail.com>
Reviewed-by: Chad Versace <chad.versace@linux.intel.com>
In commit 247f90c77e (i965/gs: Set
control data header size/format appropriately for EndPrimitive()), I
incorrectly numbered the DWORDs in the 3DSTATE_GS command starting
from 1 instead of starting from 0. This caused the control data
format to be programmed into the wrong DWORD, resulting in corruption
in some geometry shaders that used an output type of points.
This patch numbers the DWORDs starting from 0, as we do for all other
commands, which causes the control data format to be programmed into
the correct DWORD.
Reviewed-by: Chad Versace <chad.versace@linux.intel.com>
The spec doesn't say GL_INVALID_VALUE should be raised for bufSize <= 0.
In any case, memcpy(len < 0) will lead to a crash, so don't allow it.
CC: "9.2" <mesa-stable@lists.freedesktop.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Error checking bufSize isn't mentioned in the spec, but it is in the
man pages. However, I believe the man page is incorrect. Typically,
GL functions that take GLsizei parameters check that they're positive
or non-negative. Negative values don't make sense here.
A spec bug has been filed with Khronos/ARB.
v2: check for negative values, not <= 0.
This incorporates Vinson's change to check for a null src pointer as
detected by coverity.
Also, rename the function params to be src/dst, const-qualify src,
and use GL types to match the calling functions. And add some more
comments.
Reviewed-by: Timothy Arceri <t_arceri@yahoo.com.au>
libllvmradeon.la is available whenever NEED_RADEON_LLVM is set, using
R600_NEED_RADEON_GALLIUM is rather ambiguous and unnecessary. Drop it
in favour of NEED_RADEON_LLVM.
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Reviewed-by: Tom Stellard <thomas.stellard@amd.com>
The mesa/drivers/dri/Makefile.am already guards the individual
targets/subdirs with HAVE_*_DRI before including them. Thus making
the additional check within each Makefile.am unnecessary.
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
libdricommon.la is available whenever a non swrast driver is built.
All the classic dri drivers make use of the prebuild library but all
of the gallium ones rebuild it explicitly.
While we're here gallium/{llvm,soft}pipe does not require HAVE_COMMON_DRI
thus do not set in during configure.
v2: [Emil] Add commit message and drop HAVE_COMMON_DRI from configure.ac
v3: [Emil] Rebase and resolve targets/r*/dri conflicts
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Reviewed-by: Tom Stellard <thomas.stellard@amd.com>
shader has already been dereferenced earlier so cannot be null here.
Fixes "Dereference before null check" defect reported by Coverity.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
The block size for all formats is currently at least 1 byte. Add an
assertion for this.
This should silence several Coverity "Division or modulo by zero"
defects.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
There is an earlier null check for draw so draw could be null here as
well.
Fixes "Dereference after null check" defect reported by Coverity.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
Fix build errors.
CC surface.lo
surface.c: In function 'vlVdpVideoSurfaceGetBitsYCbCr':
surface.c:247:10: error: implicit declaration of function 'util_copy_rect' [-Werror=implicit-function-declaration]
CC output.lo
output.c: In function 'vlVdpOutputSurfaceGetBitsNative':
output.c:216:4: error: implicit declaration of function 'util_copy_rect' [-Werror=implicit-function-declaration]
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Fix build error.
CC device.lo
device.c: In function 'vlVdpDefaultSamplerViewTemplate':
device.c:251:4: error: implicit declaration of function 'util_format_description' [-Werror=implicit-function-declaration]
device.c:251:9: warning: assignment makes pointer from integer without a cast [enabled by default]
device.c:252:12: error: dereferencing pointer to incomplete type
device.c:252:28: error: 'UTIL_FORMAT_SWIZZLE_0' undeclared (first use in this function)
device.c:252:28: note: each undeclared identifier is reported only once for each function it appears in
device.c:254:12: error: dereferencing pointer to incomplete type
device.c:256:12: error: dereferencing pointer to incomplete type
device.c:258:12: error: dereferencing pointer to incomplete type
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
This patch fixes the build error introduced with commit
81bb98e928.
CC subpicture.lo
subpicture.c: In function 'upload_sampler':
subpicture.c:181:4: error: implicit declaration of function 'util_copy_rect' [-Werror=implicit-function-declaration]
subpicture.c: In function 'XvMCClearSubpicture':
subpicture.c:304:21: error: storage size of 'uc' isn't known
subpicture.c:328:4: error: implicit declaration of function 'util_fill_rect' [-Werror=implicit-function-declaration]
subpicture.c:304:21: warning: unused variable 'uc' [-Wunused-variable]
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
The svga/d3d9 convention is that pixel centers are at integer coordinates.
Fixes piglit glsl-arb-fragment-coord-conventions test.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
Using the map/unmap path for glTexImage is a little bit faster
than blitting. Also, this fixes about 50 assorted piglit failures
that seem to be related to the blit version of glReadPixels.
Reviewed-by: Thomas Hellstrom <thellstrom@vmware.com>
Reviewed-by: José Fonseca <jfonseca@vmware.com>
u_rect.h was including u_surface.h just to avoid touching a bunch
of other source files after some functions were moved from u_rect.h
to u_surface.h. This patch cleans up that hack.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
The format of the window system framebuffer changed from ARGB8888 to
SARGB8, but we're still supposed to render to it the same as ARGB8888
unless the user flipped the GL_FRAMEBUFFER_SRGB switch.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
NOTE: This is a candidate for stable branches.
I believe this extension was enabled by accident. As far as I can tell,
there has never been any code in Mesa to actually support it. Not only
that, this extension is only useful in the common-lite profile, and Mesa
does the common profile.
This "fixes" the piglit test oes_matrix_get-api.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Cc: "9.1 9.2" <mesa-stable@lists.freedesktop.org>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
For some reason that I don't yet fully understand, Glaze does not work with
libEGL unless libEGL is linked with -Bsymbolic.[*]
Beyond that specific reason, all of the reasons for which libGL.so is linked
with -Bsymbolic, (see the commit history), should also apply here.
[*] The specific behavior I am seeing is that when Glaze calls dlopen for
libEGL.so, ifunc resolvers within Glaze for EGL functions are called before
the dlopen returns. These resolvers cannot succeed, as they need the return
value from dlopen in order to find the functions to resolve to. I don't know
what's causing these resolvers to be called, but I have verified that linking
libEGL with -Bsymbolic causes this problematic behavior to stop.
CC: "9.1 and 9.2" <mesa-stable@lists.freedesktop.org>
Reviewed-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Chad Versace <chad.versace@linux.intel.com>
From the bspec documentation of the SEND instruction:
"destination region cannot cross the 256-bit register boundary."
To avoid violating this restriction when executing SIMD16 texturing
operations (such as those used by blorp), we need to ensure that the
destination of the SEND instruction doesn't exceed 256 bits in size.
An easy way to do this is to set the type of the destination register
to UW (unsigned word), since 16 unsigned words can fit inside a
256-bit register. Fortunately, this has no effect on the sampling
operation, since the sampler always infers the destination data type
from the sampler message rather than from the type of the instruction
operand.
Previously, we did this for texturing operations issued by the vec4
and fs back-ends, but not for blorp. This patch makes blorp use the
same trick.
I haven't observed any behavioural difference on actual hardware due
to this patch, but it avoids a warning from the simulator so it seems
like the right thing to do.
Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
Acked-by: Chad Versace <chad.versace@linux.intel.com>
We originally had a path just did the loop and called
ctx->Driver.AllocTextureImageBuffer(), which I moved into Mesa core. But
we can do better, avoiding incorrect miptree size guesses and later
texture validations by just directly allocating the miptree and setting it
to all the images.
v2: drop debug printf.
Reviewed-by: Chad Versace <chad.versace@linux.intel.com>
As long as the baselevel, maxlevel still sit inside the range we had
previously validated, there's no need to reallocate the texture.
I also hope this makes our texture validation logic much more obvious.
It's taken me enough tries to write this change, that's for sure. Reduces
miptree copy count on a piglit run by 1.3%, though the change in amount of
data moved is much smaller.
Reviewed-by: Chad Versace <chad.versace@linux.intel.com>
Given that a teximage that calls us with this flag set will immediately
proceed to allocate the other levels, we can probably just go ahead and
allocate those levels now.
Reduces miptree copies in piglit by about .05%.
Reviewed-by: Chad Versace <chad.versace@linux.intel.com>
If the caller shows up with GL_BASE_LEVEL != 0, it doesn't mean that the
texture will over the course of its lifetime have that nonzero baselevel,
it means that the caller is filling the texture from the bottom up for
some reason (one could imagine demand-loading detailed texture layers at
runtime, for example). If we allocate from just the current baselevel, it
means when they come along with the next level up, we'll have to allocate
a new miptree and copy all of our bits out of the first miptree.
Reviewed-by: Chad Versace <chad.versace@linux.intel.com>
Let's say you started allocating your 2D texture with level 2 of a tree as
a 1x1 image. The driver doesn't know if this means that level 0 is 4x4 or
4x1 or 1x4, so we would just allocate a single 1x1 and let it get copied
in to the real location at texture validate time later.
Since this is just a temporary allocation that *will* get copied, the
extra space allocation of just taking the normal path which will happen to
producing a 4x1 level 0, 2x1 level 1, and 1x1 level 2 is the right way to
go, to reduce complexity in the normal case.
No change in miptree copies over the course of a piglit run.
Reviewed-by: Chad Versace <chad.versace@linux.intel.com>
This has no effect currently, because intel_finalize_mipmap_tree() always
makes mt->first_level == tObj->BaseLevel.
The change I made before to handle it
(b1080cfbdb) got very close to working, but
after fixing some unrelated bugs in the series, it still left
tex-miplevel-selection producing errors when testing textureLod(). The
problem is that for explicit LODs, the sampler's LOD clamping is ignored,
and only the surface's MIP clamping is respected. So we need to use
surface mip clamping, which applies on top of the sampler's mip clamping,
so the sampler change gets backed out.
Now actually tested with a non-regressing series producing a non-zero
computed baselevel.
Reviewed-by: Chad Versace <chad.versace@linux.intel.com>
We know that the object's mt is equal to the firstimage's mt because it's
gone through intel_finalize_mipmap_tree(). Saves a lookup of firstimage
on pre-gen7.
v2: Merge in the warning fix that appeared later in the series (noted by
Chad)
Reviewed-by: Chad Versace <chad.versace@linux.intel.com>
The function r600_choose_tiling is new and needs a review.
The only change in functionality is that it enables 2D tiling for compressed
textures on SI. It was probably accidentally turned off.
v2: don't make scanout buffers linear
More work needs to be done for this to be entirely shared with r600g.
I'm just trying to share r600_texture.c now.
The reason I put the implementation to si_descriptors.c is that the emit
function had already been there.
This doesn't fix any known issue (I haven't run piglit with this yet),
but the code was obviously completely wrong. It looks like copy-pasted from CMP.
Reviewed-by: Tom Stellard <thomas.stellard@amd.com>
If the argument to emit_bool_to_cond_code() is an ir_expression, we
loop over the operands, calling accept() on each of them, which
generates assembly code to compute that subexpression. We then emit
one or two final instruction that perform the top-level operation on
those operands.
If it's not an expression (say, a boolean-valued variable), we simply
call accept() on the whole value.
In commit 80ecb8f1 (i965/fs: Avoid generating extra AND instructions on
bool logic ops), Eric made logic operations jump out of the expression
path to the non-expression path.
Unfortunately, this meant that we would first accept() the two operands,
skip generating any code that used them, then accept() the whole
expression, generating code for the operands a second time.
Dead code elimination would always remove the first set of redundant
operand assembly, since nothing actually used them. But we shouldn't
generate it in the first place.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Ironlake's counters are always enabled; userspace can simply send a
MI_REPORT_PERF_COUNT packet to take a snapshot of them. This makes it
easy to implement.
The counters are documented in the source code for the intel-gpu-tools
intel_perf_counters utility.
v2: Adjust for core data structure changes. Add a table mapping buffer
object offsets to exposed counters (which changes each generation).
Finally, add report ID assertions to sanity check the BO layout
(thanks to Carl Worth).
v3: Update for core BeginPerfMonitor hook changes (requested by Brian).
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
This provides an interface for applications (and OpenGL-based tools) to
access GPU performance counters. Since the exact performance counters
available vary between vendors and hardware generations, the extension
provides an API the application can use to get the names, types, and
minimum/maximum values of all available counters. Counters are also
organized into groups.
Applications create "performance monitor" objects, select the counters
they want to track, and Begin/End monitoring, much like OpenGL's query
API. Multiple monitors can be in flight simultaneously.
v2: Pass ctx to all driver hooks (suggested by Christoph), and attempt
to fix overallocation of bitsets (caught by Christoph). Incomplete.
v3: Significantly rework core data structures. Store counters in groups
rather than in a global list. Use their array index in the group's
counter list as the ID rather than trying to store a globally unique
counter ID. Use bitsets for active counters within a group, and
also track which groups are active so that's easy to query.
v4: Remove _mesa_ prefix on static functions; detect out of memory
conditions in new_performance_monitor(); make BeginPerfMonitor hook
return a boolean rather than setting m->Active or raising an error.
Switch to GLuint/unsigned for NumGroups, NumCounters, and
MaxActiveCounters (which also means switching a bunch of temporary
variable types). All suggested by Brian Paul. Also, remove
commented out code at the bottom of the block. Finally, fix the
dispatch sanity test (noticed by Ian Romanick).
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Brian Paul <brianp@vmware.com> [v3]
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
This is better than overriding the extension enable based on the
language version; it's robust against shaders that do:
#version 140
#extension GL_ARB_uniform_buffer_object : disable
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Explicit attribute locations are supported with GLSL 3.30, GLSL ES 3.00,
or "#extension GL_ARB_explicit_attrib_location: enable". Using a helper
function makes it easy to check for this.
This enables support in GLSL 3.30, which was previously missing.
Previously, we overrode the extension enable flag for ES 3.00. This is
not robust against a shader such as:
#version 330
#extension GL_ARB_explicit_attrib_location : disable
Disabling extensions should not remove core language functionality.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Hardware requires the magnitude of the largest component to not exceed
1; brw_cubemap_normalize ensures that this is the case.
Unfortunately, we would previously multiply the array index for cube
arrays by the normalization factor. The incorrect array index would then
cause the sampler to attempt to access either the wrong cube, or memory
outside the cube surface entirely, resulting in garbage rendering or in
the worst case, hangs.
Alter the normalization pass to only multiply the .xyz components.
Fixes broken rendering in the arb_texture_cube_map_array-cubemap piglit,
which was recently adjusted to provoke this behavior.
V2: Fix indent.
Signed-off-by: Chris Forbes <chrisf@ijw.co.nz>
Cc: "9.2" mesa-stable@lists.freedesktop.org
Reviewed-by: Eric Anholt <eric@anholt.net>
Compress empty triangles (don't emit more than one in a row) and
never emit empty triangles if we already generated a triangle
covering a non-null area. We can't skip all null-triangles
because c_primitives expects ones that were generated from vertices
exactly at the clipping-plane, to be emitted.
Signed-off-by: Zack Rusin <zackr@vmware.com>
Reviewed-by: José Fonseca <jfonseca@vmware.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
We need to count the clipper primitives before the rasterizer
discards one it considers to be null.
Signed-off-by: Zack Rusin <zackr@vmware.com>
Reviewed-by: José Fonseca <jfonseca@vmware.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
We need to subdivide triangles if either of the dimensions is
larger than the max edge length, not when both of them are larger.
Signed-off-by: Zack Rusin <zackr@vmware.com>
Reviewed-by: José Fonseca <jfonseca@vmware.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
The fix is at the end (TGSI_TEXTURE_SHADOWCUBE handling), but I also
restructured the code for it to be more readable.
Fixes spec/!OpenGL 3.0/sampler-cube-shadow.
Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>
This fixes some piglits, e.g:
spec/!OpenGL 3.0/required-renderbuffer-attachment-formats.
This can be ported to r600g.
Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>
Only create one screen for each winsys instance.
This helps with buffer sharing and interop handling.
v2: rebased and some minor cleanup
Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
This reverts commit 755c11dc5e.
We agreed that this is band-aid that's not very useful and
the proper solution is to rewrite the rasterization algo
so that it operates on 64 bit values.
Signed-off-by: Zack Rusin <zackr@vmware.com>
No such argument exists since this commit:
commit 92f3fca0ea
Author: Ian Romanick <ian.d.romanick@intel.com>
AuthorDate: Sun Aug 21 17:23:58 2011 -0700
Commit: Ian Romanick <ian.d.romanick@intel.com>
CommitDate: Tue Aug 23 14:52:09 2011 -0700
mesa: Remove target parameter from dd_function_table::BufferSubData
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
When subdiving a triangle we're using a temporary array to store
the new coordinates for the subdivided triangles. Unfortunately
the array used for that was not aligned properly causing
random crashes in the llvm jit code which was trying to load
vectors from it.
Signed-off-by: Zack Rusin <zackr@vmware.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
This patch fixes the MSVC build error introduced by commit
673129e0b9.
enums.c
mesa\main\enums.c(3776) : error C2143: syntax error : missing ';' before 'type'
mesa\main\enums.c(3781) : error C2065: 'elt' : undeclared identifier
mesa\main\enums.c(3781) : warning C4047: '!=' : 'int' differs in levels of indirection from 'void *'
mesa\main\enums.c(3782) : error C2065: 'elt' : undeclared identifier
mesa\main\enums.c(3782) : error C2223: left of '->offset' must point to struct/union
mesa\main\enums.c(3782) : warning C4033: '_mesa_lookup_enum_by_nr' must return a value
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Normally, LD_PRELOAD will take precedence over your own symbols, which you
want for things like malloc() in libc. But we don't have any local
symbols we would want overridden (like hash_table_insert(), for example!),
so tell the linker to resolve them internally. This also avoids calls
through the PLT.
Saves almost 100k on libdricore's size, and gets us a bunch of the
performance back that we had with non-dricore.
Reviewed-by: Ian Romanick <ian.d.romanick@.intel.com>
This gives the compiler the chance to inline and not export class symbols
even in the absence of LTO. Saves about 60kb on disk.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ian Romanick <ian.d.romanick@.intel.com>
Noticed while grepping through the code for something else.
v2: Don't convert really-runtime asserts to static asserts.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ian Romanick <ian.d.romanick@.intel.com>
Since it's only used for debug information, we can misalign the struct and
save the disk space. Another 19k on a 64-bit build.
v2: Make a compiler.h macro to only use the attribute if we know we can.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ian Romanick <ian.d.romanick@.intel.com>
Now that there's no name -> enum direction, we can drop the extra strings,
and merge the offsets table and the reduced_enums table.
Between the previous commit and this one, Mesa core drops by 30k.
Acked-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ian Romanick <ian.d.romanick@.intel.com>
It's been unused for a long time. I stopped digging through git history
as of 2009.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ian Romanick <ian.d.romanick@.intel.com>
Unfortunately d3d10 requires a lot higher precision (e.g.
wgf11clipping tests for it). The smallest number of precision
bits with which it passes is 8. That means that we need to
decrease the maximum length of an edge that we can handle without
subdivision by 4 bits. Abstracted the code a bit to make it easier
to change once to switch to 64bit rasterization.
Signed-off-by: Zack Rusin <zackr@vmware.com>
Reviewed-by: José Fonseca <jfonseca@vmware.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
This patch fixes these MSVC build errors.
ir_constant_expression.cpp
src\glsl\ir_constant_expression.cpp(564) : warning C4244: '=' : conversion from 'int' to 'float', possible loss of data
src\glsl\ir_constant_expression.cpp(1384) : error C3861: 'isnormal': identifier not found
src\glsl\ir_constant_expression.cpp(1385) : error C3861: 'copysign': identifier not found
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=69541
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Acked-by: Matt Turner <mattst88@gmail.com>
Share the winsys between different fd's if they point to the same device.
Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Waiting for an empty queue is nonsense and can lead to deadlocks if we have
multiple waiters or another thread that continuously sends down new commands.
Just post the cs to the queue and immediately wait for it to finish.
This is a candidate for the stable branch.
Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Kill the thread only after we checked that it's not used any more, not before.
Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
The rescale_texcoord(), if it does something, will return just the
GLSL-sized coordinate, leaving out the 3rd and 4th components where we
were storing our projected shadow compare and the texture projector.
Deref the shadow compare before using the shared rescale-the-coordinate
code to fix the problem.
Fixes piglit tex-shadow2drect.shader_test and txp-shadow2drect.shader_test
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=69525
NOTE: This is a candidate for stable branches.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Probably non-intentional, but the SURFACE_STATE setup refactoring
for buffer surfaces had missed the scs bits when creating constant
surface states.
Fixes broken GLB 2.5 on Haswell where the knight's textures are missing
Signed-off-by: Abdiel Janulgue <abdiel.janulgue@linux.intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
These classes declared a placement new operator, but didn't declare a
delete operator. Switching to the macro gives them a delete operator,
which probably is a good idea anyway.
This also eliminates a lot of boilerplate.
v2: Properly use RZALLOC in Mesa IR/TGSI translators. Caught by Eric
and Chad.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Chad Versace <chad.versace@linux.intel.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Most of our C++ classes define placement new and delete operators so we
can do convenient allocation via:
thing *foo = new(mem_ctx) thing(...)
Currently, this is done via a lot of boilerplate. By adding simple
macros to ralloc, we can condense this to a single line, making it
trivial to add this feature to a new class.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Chad Versace <chad.versace@linux.intel.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Allocate a CMASK on demand and use it to fast clear single-sample
colorbuffers. Both FBOs and window system colorbuffers are fast
cleared. Expand as needed when colorbuffers are mapped or displayed
on screen.
v2: cosmetics, move transfer expansion into dma_blit
Signed-off-by: Marek Olšák <marek.olsak@amd.com>
r600g needs explicit flushing before DRI2 buffers are presented on the screen.
v2: add (stub) implementations for all drivers, fix frontbuffer flushing
v3: fix galahad
Signed-off-by: Marek Olšák <marek.olsak@amd.com>
We must take rounding in consideration when re-scaling to narrow
normalized channels, such as 2-bit normalized alpha.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
GL_MAX_GEOMETRY_TEXTURE_IMAGE_UNITS, GL_MAX_GEOMETRY_OUTPUT_VERTICES,
GL_MAX_GEOMETRY_TOTAL_OUTPUT_COMPONENTS, and
GL_MAX_GEOMETRY_UNIFORM_COMPONENTS all have the same enum value and
meaning as their _ARB counterparts.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
The comment '# GL 3.0 / GLES3' was incorrect. The
MAX_VERTEX_OUTPUT_COMPONENTS and MAX_FRAGMENT_INPUT_COMPONENTS queries
were added in OpenGL 3.2 (with geometry shaders) and OpenGL ES 3.0.
This just fixes that comment.
v2: Add the GEOMETRY queries in the existing '# GL 3.2' section since
they have nothing to do with GLES3. Suggested by Paul.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
In OpenGL ES 3.0 the minimum-maximum for GL_MAX_VERTEX_OUTPUT_VECTORS is 16,
but the minimum-maximum for GL_MAX_FRAGMENT_INTPUT_VECTORS is 15.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
In OpenGL ES 3.0 the minimum-maximum for GL_MAX_VERTEX_OUTPUT_VECTORS is 16,
but the minimum-maximum for GL_MAX_FRAGMENT_INTPUT_VECTORS is 15.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
Now that MaxVaryings is > 16, VertexProgram.MaxOutputComponents,
GeometryProgram.MaxInputComponents, GeometryProgram.MaxOutputComponents,
and FragmentProgram.MaxInputComponents also need to be set.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Cc: Paul Berry <stereotype441@gmail.com>
Previously gl_constants::MaxVaryingComponents was used. Now
gl_constants::VertexProgram::MaxOutputs and
gl_constants::GeometryProgram::MaxOutputs are used.
This means that st_extensions.c had to be updated to set these fields
instead of MaxVaryingComponents. It was previously the only place that
set MaxVaryingComponents.
I believe that the structure is allocated by calloc, so the value should
be initialized to zero in non-Gallium drivers before and after my
change. Right now nobody enables GL_ARB_geometry_shader4, so it's
pretty much dead code anyway.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
Cc: Zack Rusin <zackr@vmware.com>
In OpenGL 3.2 these are independently queryable. In addition, the spec
has different minimum-maximums for various values.
GL_MAX_VERTEX_OUTPUT_COMPONENTS is 64, but
GL_MAX_GEOMETRY_OUTPUT_COMPONENTS (and GL_MAX_FRAGMENT_INPUT_COMPONENTS)
is 128.
In OpenGL ES 3.0 these are also independently queryable. The spec has
different minimum-maximums for various values.
GL_MAX_VERTEX_OUTPUT_VECTORS is 16, but GL_MAX_FRAGMENT_INTPUT_VECTORS
is 15.
None of these values are used yet. I have just added space to the
structures. Future patches will add users and eventually remove some
old fields.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
Cc: Zack Rusin <zackr@vmware.com>
This was an embarassingly large amount of copy and pasted code,
and it wasn't particularly simple code either. By factoring it out
into a helper function, we consolidate the complexity.
v2: Properly NULL-check bo. Caught by Eric Anholt.
v3: Do the subtraction by 1 in gen7_emit_buffer_surface_state, rather
than making callers do it. This makes the buffer_size parameter
the actual size of the buffer. Suggested by Paul Berry.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Eric Anholt <eric@anholt.net>
This was an embarassingly large amount of copy and pasted code,
and it wasn't particularly simple code either. By factoring it out
into a helper function, we consolidate the complexity.
v2: Properly NULL-check bo. Caught by Eric Anholt.
v3: Do the subtraction by 1 in gen7_emit_buffer_surface_state, rather
than making callers do it. This makes the buffer_size parameter
the actual size of the buffer. Suggested by Paul Berry.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Eric Anholt <eric@anholt.net>
The value that's split into width/height/depth needs to be the size of
the buffer minus one. This makes it consistent with the constant buffer
and shader time SURFACE_STATE setup code.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Eric Anholt <eric@anholt.net>
This fixes myriads of regressions since commit 169f9c030c
("i965: Add an assertion that writemask != NULL for non-ARFs.").
On Sandybridge, our control flow handling (such as brw_IF) does:
brw_set_dest(p, insn, brw_imm_w(0));
insn->bits1.branch_gen6.jump_count = 0;
This results in a IMM destination with zero for the writemask. IMM
destinations are rather bizarre, but the code has been working for ages,
so I'm loathe to change it.
Fixes glxgears on Sandybridge.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Eric Anholt <eric@anholt.net>
I would use _mesa_delete_shader, but it's declared static, and we don't
really need any of the stuff in it anyway.
This fixes a memory leak caught by Valgrind.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Eric Anholt <eric@anholt.net>
&a and &b are the address of the local stack variables, not the actual
structures. Instead of comparing the fields of a and b, we compared
...some stack memory.
Not a candidate for stable since GS code doesn't exist in 9.2.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Chad Versace <chad.versace@linux.intel.com>
The code to upload the binding tables for each stage was scattered
across brw_{vs,gs,wm}_surface_state.c and brw_misc_state.c, which also
contain a lot of code to populate individual SURFACE_STATE structures.
This patch brings all the binding table upload code together, and splits
it out from the code which fills in SURFACE_STATE entries.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Eric Anholt <eric@anholt.net>
This is not quite the same: brw_upload_binding_table() also has code to
early-return if there are no entries, while the existing code did not.
The PS binding table is unlikely to be empty since it will have at least
one color buffer. If it ever is empty, early returning seems wise.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Eric Anholt <eric@anholt.net>
Instead of passing in a brw_vec4_prog_data structure, we can simply
pass the one field it needs: the number of entries in the binding table.
We also need to pass in the shader time surface index rather than
hardcoding SURF_INDEX_VEC4_SHADER_TIME.
Since the resulting function is stage-agnostic, this patch removes
"vec4_" from the name.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Eric Anholt <eric@anholt.net>
This is probably more efficient. At any rate, it's less code.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Eric Anholt <eric@anholt.net>
The first comment was a bit stale; there are more kinds of surfaces than
textures and pull constants.
The second was a leftover "to do" comment for something I already did.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Eric Anholt <eric@anholt.net>
xlib_sw_winsys.h:5:22: fatal error: X11/Xlib.h: No such file or directory
The compiler cannot find the Xlib.h in the installed system headers.
All supplied include directives point to inside the mesa module.
The X11_CFLAGS variable is undefined (not defined in config.status).
It appears the intent was to use X11_INCLUDES defined in configure.ac.
The Xlib.h file is not installed on my workstation. It is supplied in
the libx11-dev package. This allows an X developer control over which
version of this file is used for X development.
Signed-off-by: Gaetan Nadon <memsize@videotron.ca>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
egl_glx.c:40:22: fatal error: X11/Xlib.h: No such file or directory
The compiler cannot find the Xlib.h in the installed system headers.
All supplied include directives point to inside the mesa module.
The X11_CFLAGS variable is undefined (not defined in config.status).
It appears the intent was to use X11_INCLUDES defined in configure.ac.
The Xlib.h file is not installed on my workstation. It is supplied in
the libx11-dev package. This allows an X developer control over which
version of this file is used for X development.
Signed-off-by: Gaetan Nadon <memsize@videotron.ca>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Technically without seamless filtering enabled GL allows any wrap mode, which
made sense when supporting true borders (can get seamless effect with border
and CLAMP_TO_BORDER), but gallium doesn't support borders and d3d9 requires
wrap modes to be ignored and it's a pain to fix up the sampler state (as it
makes it texture dependent). It is difficult to imagine a situation where an
app really wants another behavior so just cheat here. (It looks like some
graphics hw (intel) actually requires this too hence it should be safe.)
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
This removes a lot of code, but not everything, as util_blit_pixels_tex
is still useful when one needs to override pipe_sampler_view::swizzle_?.
Reviewed-by: Zack Rusin <zackr@vmware.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
By calling util_map_texcoords2d_onto_cubemap.
A new parameter for util_blit_pixels_tex is necessary, as
pipe_sampler_view::first_layer is always supposed to point to the first
face when sampling from cubemaps.
Reviewed-by: Zack Rusin <zackr@vmware.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
The previous names were really confusing to talk about:
- brw_fs_visitor() contained methods named emit_whatever().
- brw_fs_generator() contained methods named generate_whatever(), but
lived in brw_fs_emit.cpp.
So when someone said "the emit layer", or "emit code", we weren't sure
whether they meant the visitor's emit() functions or the generator in
brw_fs_emit.cpp.
By renaming these files, the method names, class names, and file names
all match, which is much less confusing.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Acked-by: Paul Berry <stereotype441@gmail.com>
Acked-by: Eric Anholt <eric@anholt.net>
I initially implemented frexp() as an IR opcode with a lowering pass,
but since it returns a value and has an out-parameter, it would break
assumptions our optimization passes make about ir_expressions being pure
(i.e., having no side effects).
For example, if opt_tree_grafting encounters this code:
uniform float u;
void main()
{
int exp;
float f = frexp(u, out exp);
float g = float(exp)/256.0;
float h = float(exp) + 1.0;
gl_FragColor = vec4(f, g, h, g + h);
}
it may try to optimize it to this:
uniform float u;
void main()
{
int exp;
float g = float(exp)/256.0;
float h = float(exp) + 1.0;
gl_FragColor = vec4(frexp(u, out exp), g, h, g + h);
}
Some hardware has an instruction which performs frexp(), but we would
need some other compiler infrastructure to be able to generate it, such
as an intrinsics system that would allow backends to emit specific code
for particular bits of IR.
Reviewed-by: Paul Berry <stereotype441@gmail.com>
Note the parameter name change in the int version of ir_constant, to
avoid the conflict with the loop iterator.
v2: Make analogous change to builtin_builder::imm().
Reviewed-by: Paul Berry <stereotype441@gmail.com>
These data structures are used for debug output, so it wasn't hurting
anything that there were missing bits. But it's good to keep things
up to date.
This patch also adds static asserts so that the {brw,cache}_bits[]
arrays are the proper size, so that we don't forget to add to them in
the future. Unfortunately there's no convenient way to assert that
mesa_bits[] is the proper size.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
If the geometry shader refers to the built-in variable
gl_PrimitiveIDIn, we need to set a bit in 3DSTATE_GS to tell the
hardware to dispatch primitive ID to r1, and we need to leave room for
it when allocating registers.
Note: this feature doesn't yet work properly when software primitive
restart is in use (the primitive ID counter will incorrectly reset
with each primitive restart, since software primitive restart works by
performing multiple draw calls). I plan to address that in a future
patch series.
Fixes piglit test "spec/glsl-1.50/execution/geometry/primitive-id-in".
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
When we previously implemented primitive restart, we didn't add cases
to brw_primitive_restart.c's can_cut_index_handle_prims() for the
primitive types that are introduced with geometry shaders. It turns
out that all of the new primitive types are supported by hardware
primitive restart.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
As part of its support for geometry shaders, GL 3.2 introduces four
new primitive types: GL_LINES_ADJACENCY, GL_LINE_STRIP_ADJACENCY,
GL_TRIANGLES_ADJACENCY, and GL_TRIANGLE_STRIP_ADJACENCY.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Simply adjust wrap mode to clamp_to_edge. This is all that's needed for a
correct implementation for nearest filtering, and it's way better than
using repeat wrap for instance for linear filtering (though obviously this
doesn't actually do seamless filtering).
v2: fix s/t wrap not r/s...
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
Specifying a miptree layout makes no sense for constant buffers.
This has no functional change since BRW_SURFACE_MIPMAPLAYOUT_BELOW is
just a #define for 0.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
This paves the way for using gen7_upload_constant_state for PS data.
The formula is copied from gen7_wm_state.c.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
GL 3.2 requires us to support 128 varying components for geometry
shader outputs and fragment shader inputs, and 64 varying components
otherwise. But there's no hardware limitation that restricts us to 64
varying components, and core Mesa doesn't currently allow different
stages to have different maximum values, so just go ahead and enable
128 varying components for all stages. This gets us better test
coverage anyway.
Even though we are only working on GL 3.2 support for gen7 right now,
gen6 also supports 128 varying components, so go ahead and switch it
on there too.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Previously we only ever did 1 URB write, since the maximum number of
varyings we support is small enough to fit in 1 URB write (when using
BRW_URB_SWIZZLE_NONE, which is what the pre-Gen7 GS always uses). But
we're about to increase the number of varying components we support
from 64 to 128.
With 128 varyings, the most URB writes we'll have to do is 2, but it's
just as easy to write a general-purpose loop.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
The "{VS,GS} URB Entry Allocation Size" fields of 3DSTATE_URB allow
values in the range 0-4, but they are U8-1 fields, so the range of
possible allocation sizes is 1-5. We were erroneously prohibiting a
size of 5.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Previously we only ever did 1 or 2 URB writes, since the maximum
number of varyings we support is small enough to fit in 2 URB writes.
But GL 3.2 requires the geometry shader to support 128 output varying
components, and this could require up to 3 URB writes.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Since the SF/SBE stage is only capable of performing arbitrary
reorderings of 16 varying slots, we can't arrange the fragment shader
inputs in an arbitrary order if there are more than 16 input varying
slots in use. We need to make sure that slots 16-31 match the
corresponding outputs of the previous pipeline stage.
The easiest way to accomplish this is to just make all varying slots
match up with the previous pipeline stage.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
The for loop was rather silly. In addition to checking brw->gen < 6
on each loop iteration, it took pains to exclude bits from
fp->Base.InputsRead that don't correspond to fragment shader inputs.
But those bits would never have been set in the first place, since the
only bits that are ever set in fp->Base.InputsRead are fragment shader
inputs.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Now that the vertex shader output VUE map is determined solely by a
64-bit bitfield, we don't have to store it in its entirety in the
geometry shader program key; instead, we can just store the bitfield,
and let the geometry shader infer the VUE map at compile time.
This dramatically reduces the size of the geometry shader program key,
which we want to keep small since it gets recomputed whenever the
active program changes.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Previously, on Gen6+, we laid out the vertex (or geometry) shader VUE
map differently depending whether user clipping was active. If it was
active, we put the clip distances in slots 2 and 3 (where the clipper
expects them); if it was inactive, we assigned them in the order of
the gl_varying_slot enum.
This made for unnecessary recompiles, since turning clipping on/off
for a shader that used gl_ClipDistance might rearrange the varyings.
It also required extra bookkeeping, since it required the user
clipping flag to be provided to brw_compute_vue_map() as a parameter.
With this patch, we always put clip distances at in slots 2 and 3 if
they are written to. do_vs_prog() and do_gs_prog() are responsible
for ensuring that clip distances are written to when user clipping is
enabled (as do_vs_prog() previously did for gen4-5).
This makes the only input to brw_compute_vue_map() a bitfield of which
varyings the shader writes to, a fact that we'll take advantage of in
forthcoming patches.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Previously, if a fragment shader accessed gl_FragCoord or
gl_FrontFacing, we would assign them their own slots in the fragment
shader input attribute array, using up space that could be made
available to real varyings. This was not strictly necessary (since
these values are not true varyings, and are instead computed from
other data available in the FS payload). But we had to do it anyway
because the SF/SBE setup code assumed that every 1 bit in the
gl_program::InputsRead bitfield corresponded to a genuine varying
variable.
Now that the SF/SBE code consults brw_wm_prog_data and only sets up
the attributes that the fragment shader actually needs, we don't have
to do this anymore.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Previously, the SF/SBE setup code delivered varying inputs to the FS
in the order in which they appear in the gl_program::InputsRead
bitfield, since that's what the FS expects.
When we add support for more than 64 varying components, this will no
longer always be the case, because the Gen6+ SF/SBE stage is only
capable of performing arbitrary reorderings of 16 varying slots. So,
when there are more than 16 vec4's worth of varying inputs, the FS
will have to adjust the order its input varyings in order to partially
match the order of outputs from the geometry or vertex shader.
To allow extra flexibility in the ordering of FS varyings, this patch
causes the SF/SBE to deliver varying inputs to the FS in exactly the
order that the FS requests, by consulting brw_wm_prog_data::urb_setup
and brw_wm_prog_data::num_varying_inputs.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
We always program the SF unit to start reading the vertex URB entry at
offset 1. In upcoming patches, we'll be adding FS code that relies on
this. So consistently use the constant BRW_SF_URB_ENTRY_READ_OFFSET
rather than hardcoding a 1.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Previously, we assumed that the number of varying inputs consumed by
the fragment shader was equal to the number of bits set in
gl_program::InputsRead. However, we'll soon be making two changes
that will cause that not to be true:
- We'll stop wasting varying input space for gl_FragCoord and
gl_FrontFacing, which aren't varyings.
- For fragment shaders that have more than 16 varying inputs, we'll
adjust the layout of the inputs to account for the fact that the
SF/SBE pipeline stage can't reorder inputs beyond the first 16; if
there are GS outputs that the FS doens't use (or vice versa) this
may cause the number of FS varying inputs to change.
So, instead of trying to guess the number of FS inputs from
gl_program::InputsRead, simply read it from
brw_wm_prog_data:num_varying_inputs, which is guaranteed to be correct
since it's populated by fs_visitor::calculate_urb_setup().
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
On gen4-5, the FS stage reads varying inputs from URB entries that
were output by the SF thread, where each register stores the
interpolation setup for two components of a vec4, therefore the FS
urb_read_length is twice the number of FS input varyings. On gen6+,
varying inputs are directly deposited in the FS payload by the SF/SBE
fixed function logic, so urb_read_length is irrelevant.
However, in future patches, it will be nice to be able to consult
brw_wm_prog_data to determine how many varying inputs the FS expects
(rather than inferring it from gl_program::InputsRead). So instead of
storing urb_read_length, we simply store num_varying_inputs in
brw_wm_prog_data. On gen4-5, we multiply this by 2 to recover the URB
read length.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
At the moment, for Gen6+, the FS assumes that all varying inputs are
delivered to it in the order in which they appear in the
gl_program::InputsRead bitfield, and the SF/SBE setup code ensures
that they are delivered in this order.
When we add support for more than 64 varying components, this will no
longer always be possible, because the Gen6+ SF/SBE stage is only
capable of performing arbitrary reorderings of 16 varying slots.
To allow extra flexibility in the ordering of FS varyings, this patch
causes the FS to advertise exactly what ordering it expects.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
For each sampler type, this tests that:
- The base type is GLSL_TYPE_SAMPLER.
- The dimensionality is set correctly.
- The returned data type is correct.
- The sampler_array and sampler_shadow flags are set correctly.
- sampler_coordinate_components() returns the correct value.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ian Romanick <idr@freedesktop.org>
It seems a user app can get us into this state, I trigger the fail
running fbo-maxsize inside virgl, it fails to create the backing
storage for the texture object, but then segfaults here when it
should fail the completeness test.
Cc: "9.2" <mesa-stable@lists.freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
Fix the return type and allow src and dst types for comparison
to be separate, this at least fixes the two test cases I've written.
v2: drop the u32->s32 change
Acked-by: Christoph Bumiller <christoph.bumiller@speed.at>
Signed-off-by: Dave Airlie <airlied@redhat.com>
When the old contents do not need to be preserved, it is faster to
create a new backing bo rather than stall.
Signed-off-by: Rob Clark <robclark@freedesktop.org>
max_index may be 0xffffffff. The hardware does not need 1 + max_index
(although it does not hurt unless max_index wraps around to zero).
Signed-off-by: Rob Clark <robclark@freedesktop.org>
For mem->gmem we don't sample depth/stencil as it's native type. So we
need to setup the swizzle state for the sampler based on the format used
for sampling.
Signed-off-by: Rob Clark <robclark@freedesktop.org>
Needed by some games, like etuxracer and supertuxkart which use alpha
test rather than blending, to handle texture transparency.
Signed-off-by: Rob Clark <robclark@freedesktop.org>
With a debug option to force DIRECT (mainly to make it easier for
capturing cmdstream dumps). Using INDIRECT for large shaders at least
makes a noticable reduction in CPU load, which helps for CPU limited
games.
Signed-off-by: Rob Clark <robclark@freedesktop.org>
Because of how the tiling works, we can't really flush at arbitrary
points very easily. So wraparound is handled by resetting to top of
ringbuffer. Previously this would stall until current rendering is
complete. Instead cycle through multiple ringbuffers to avoid a stall.
Signed-off-by: Rob Clark <robclark@freedesktop.org>
Emit markers by writing to scratch registers in order to "triangulate"
gpu lockup position from post-mortem register dump. By comparing
register values in post-mortem dump to command-stream, it is possible to
narrow down which DRAW_INDX caused the lockup.
Signed-off-by: Rob Clark <robclark@freedesktop.org>
Have a single helper that all draws come through.. mainly for a
convenient debug and instrumentation point.
Signed-off-by: Rob Clark <robclark@freedesktop.org>
The varying-out config comes from the inputs of the frag shader (so that
we aren't exporting unneeded varyinges). The varyings-count should come
from the frag shader as well, to avoid a discrepency in configuration
and resulting gpu lockup.
Signed-off-by: Rob Clark <robclark@freedesktop.org>
These functions are defined in EXT_texture_array, which makes no
mention of what shader types they should be allowed in. At the time
EXT_texture_array was introduced, functions ending in "Lod" were
available only in vertex shaders, however this restriction was lifted
in later spec versions and extensions.
We already have the function lod_exists_in_stage() for figuring out
whether functions ending in "Lod" should be available, so just re-use
that.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Since BRW_MAX_WM_SURFACES is greater than BRW_MAX_VEC4_SURFACES, the
existing array isn't large enough to be used by the WM. Increasing it
will make it possible to share them.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
These are largely based on the similar fields in brw->wm.
v2: Add a better comment than "Scratch buffer".
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
Everyone at the Khronos meeting was as surprised that GLSL didn't
already support this as we were. Several vendors said they'd ship it,
but there didn't seem to be enough interest to put in the effort to make
it ARB or KHR.
v2: Fix a couple typos and rename the spec file to
EXT_shader_integer_mix.spec. Suggested by Roland.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
start_instance doesn't affect gl_InstanceID.
There's no piglit test, but it's kinda obvious the code was wrong.
Reviewed-by: Christian König <christian.koenig@amd.com>
The shader is responsible for writing to streamout buffers using
the TBUFFER_STORE_FORMAT_* instructions.
The locations of some input SGPRs and VGPRs are assigned dynamically, because
the input SGPRs controlling streamout are not declared if they are not needed,
decreasing the indices of all following inputs.
Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>
So that the "init" state is always emitted first and not later in draw_vbo.
This fixes streamout where the "init" state, which disables streamout,
was emitted in draw_vbo after streamout was enabled.
Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>
This could happen if set_stream_output_targets is called twice
in a row without a draw call in between.
Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>
Paused transform feedback objects may refer to a program other than the
current program. If any active objects refer to a program, LinkProgram
must reject the request to relink.
The code to detect this is ugly since _mesa_HashWalk is awkward to use,
but unfortunately we can't use hash_table_foreach since there's no way
to get at the underlying struct hash_table (and even then, we'd need to
handle locking somehow).
Fixes the last subcase of Piglit's new ARB_transform_feedback2
api-errors test.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
This is actually a pretty important error condition: otherwise, you
could set up transform feedback with one program, and resume it with
a program that generates a completely different set of outputs.
Fixes a subcase of Piglit's new ARB_transform_feedback2 api-errors test.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
The next few patches will use this for API error checking.
All of the drivers appear to CALLOC_STRUCT transform feedback objects,
so this should be properly NULL initialized on creation.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Fixes a subcase of Piglit's new ARB_transform_feedback2 api-errors test.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Move the code back into the common UVD files since we now
have base structures for R600 and radeonsi.
Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Fixes FTBFS on kfreebsd-*
Debian GNU/kFreeBSD doesn't provide getprogname() since it uses stdlib.h
from glibc. Instead it provides program_invocation_short_name from glibc.
You can find the same order in src/mesa/drivers/dri/common/xmlconfig.c
Cc: "9.2" <mesa-stable@lists.freedesktop.org>
Tested-by: Julien Cristau <jcristau@debian.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
It was wrong for EXP.y, as we clamped the source before computing the
fractional part, and this opcode should be rarely used, so it's not
worth the hassle.
We used to pass the number of components actually used for the
coordinate (rather than padding, shadow comparitors, and projectors) by
hand, specifying it on every _texture() call.
The new helper function can just compute this, eliminating a lot of
potential mistakes.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
This computes the number of components necessary to address a sampler
based on its dimensionality. It will be useful for texturing built-ins.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
It is planned to ship openSUSE 13.1 with -shared libs.
nouveau.la, nv30.la, nv50.la and nvc0.la are currently LIBADDs in all nouveau
related targets.
This change makes it possible to easily build one shared libnouveau.so which is
then LIBADDed.
Also dlopen will be faster for one library instead of three and build time on
-jX will be reduced.
Whitespace fixes were requested by 'git am'.
Signed-off-by: Johannes Obermayr <johannesobermayr@gmx.de>
Acked-by: Christoph Bumiller <christoph.bumiller@speed.at>
Acked-by: Ian Romanick <ian.d.romanick@intel.com>
According to GLSL, the shader may call EndPrimitive() at any point
during its execution, causing the line or triangle strip currently
being output to be terminated and a new strip to be begun.
This is implemented in gen7 hardware by using one control data bit per
vertex, to indicate whether EndPrimitive() was called after that
vertex was emitted.
In order to make this work without sacrificing too much efficiency, we
accumulate 32 control data bits at a time in a GRF. When we have
accumulated 32 bits (or when the shader terminates), we output them to
the appropriate DWORD in the control data header and reset the
accumulator to 0.
We have to take special care to make sure that EndPrimitive() calls
that occur prior to the first vertex have no effect.
Since geometry shaders that output a large number of vertices are
likely to be rare, an optimization kicks in if max_vertices <= 32. In
this case, we know that we can wait until the end of shader execution
before any control data bits need to be output.
I've tried to write the code in such a way that in the future, we can
easily adapt it to output stream ID bits (which are two bits/vertex
instead of one).
Fixes piglit tests "spec/glsl-1.50/glsl-1.50-geometry-end-primitive *".
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Previously, brw_urb_WRITE() would always generate a URB_WRITE_HWORD
message, we always wanted to write data to the URB in pairs of varying
slots or larger (an HWORD is 32 bytes, which is 2 varying slots).
In order to support geometry shader EndPrimitive functionality, we'll
need the ability to write to just a single OWORD (16 byte) slot, since
we'll only be outputting 32 of the control data bits at a time. So
this patch adds a flag that will cause brw_urb_WRITE to generate a
URB_WRITE_OWORD message.
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Previously, brw_urb_WRITE() would unconditionally override the channel
masks in the URB_WRITE message to 0xff (indicating that all channels
should be written to the URB).
In order to support geometry shader EndPrimitive functionality, we'll
need the ability to set the channel masks programatically, so that we
can output just 32 of the control data bits at a time. So this patch
adds a flag that will prevent brw_urb_WRITE() from overriding them.
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
The gen7 geometry shader uses a "control data header" at the beginning
of the output URB entry to store either
(a) flag bits (1 bit/vertex) indicating whether EndPrimitive() was
called after each vertex, or
(b) stream ID bits (2 bits/vertex) indicating which stream each vertex
should be sent to (when multiple transform feedback streams are in
use).
Fortunately, OpenGL only requires separate streams to be supported
when the output type is points, and EndPrimitive() only has an effect
when the output type is line_strip or triangle_strip, so it's not a
problem that these two uses of the control data header are mutually
exclusive.
This patch modifies do_vec4_gs_prog() to determine the correct
hardware settings for configuring the control data header, and
modifies upload_gs_state() to propagate these settings to the
hardware.
In addition, it modifies do_vec4_gs_prog() to ensure that the output
URB entry is large enough to contain both the output vertices *and*
the control data header.
Finally, it modifies vec4_gs_visitor so that it accounts for the size
of the control data header when computing the offset within the URB
where output vertex data should be stored.
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
v2: Fixed incorrect handling of IVB/HSW differences.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
This information will be useful in the i965 back end, since we can
save some compilation effort if we know from the outset that the
shader never calls EndPrimitive().
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
v2: Do not attempt to share the code that uploads
3DSTATE_BINDING_TABLE_POINTERS_GS, 3DSTATE_SAMPLER_STATE_POINTERS_GS,
or 3DSTATE_GS with VS.
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
v3: Add _NEW_TRANSFORM to gen7_gs_state.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
This will allow us to reuse some code when setting up the geometry
shader stage.
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Mesa provides the wayland-egl libs and the pkgconfig file, but the headers
originate from the wayland package. Ensure everything matches, by requiring
application builds to look at the wayland headers as well.
Signed-off-by: Torsten Duwe <duwe@suse.de>
Signed-off-by: Johannes Obermayr <johannesobermayr@gmx.de>
Commit b77316ad75
st/dri: always copy new DRI front and back buffers to corresponding MSAA buffers
introduced creating a pipe_context for every call to validate, which is not required
because the callers have a context anyway.
Only exception is egl_g3d_create_pbuffer_from_client_buffer, can someone test if it
still works with NULL passed as context for validate? From examining the code I
believe it does, but I didn't thoroughly test it.
Signed-off-by: Maarten Lankhorst <maarten.lankhorst@canonical.com>
Cc: 9.2 <mesa-stable@lists.freedesktop.org>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
We've observed GPU hangs on Ivybridge from the following instruction:
mov(8) g115<1>.F 0D { align16 WE_normal NoDDChk 1Q };
There should be no reason to ever set the writemask on a destination
register to zero, except for perhaps the ARF NULL register.
This patch adds an assertion to enforce this for non-ARF registers.
Excluding ARFs is conservative yet should still catch the majority
of mistakes.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Chris Forbes <chrisf@ijw.co.nz>
Otherwise, coordinates with four components would result in a MOV
with a destination writemask that has no channels enabled:
mov(8) g115<1>.F 0D { align16 WE_normal NoDDChk 1Q };
At best, this is stupid: we emit code that shouldn't do anything.
Worse, it apparently causes GPU hangs (observable with Chris's
textureGather test on CubeArrays.)
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Chris Forbes <chrisf@ijw.co.nz>
Reviewed-by: Ian Romanick <idr@freedesktop.org>
Cc: mesa-stable@lists.freedesktop.org
This was originally introduced by commit
ba47aabc98, but unfortunately the commit message
doesn't go into much detail about why +INF would be a problem here.
A similar issue exists for STATE_FOG_PARAMS_OPTIMIZED, but allowing infinity
there would potentially introduce NaNs where they shouldn't exist, depending
on the values of fog end and the fog coord. Since STATE_FOG_PARAMS_OPTIMIZED
is only used for fixed function (including ARB_fragment_program with fog
option), and the calculation there probably isn't very stable to begin with
when fog start and end are close together, it seems best to just leave it
alone.
This fixes piglit glsl-fs-fogscale, and a couple of Wine D3D tests. No piglit
regressions on Cayman.
Signed-off-by: Henri Verbeet <hverbeet@gmail.com>
Tested-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
278372b47e added the uninitialized pointer
field gl_sync_object:Label. A free of this pointer, added in commit
6d8dd59cf5, resulted in a crash.
This patch fixes piglit ARB_sync regressions with swrast introduced by
6d8dd59cf5.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
These instructions will be used with immediate arguments in the upcoming
ldexp lowering pass and frexp implementation.
v2: Add vec4 support as well.
Reviewed-by: Paul Berry <stereotype441@gmail.com>
It's a ?: that operates per-component on vectors. Will be used in
upcoming lowering pass for ldexp and the implementation of frexp.
csel(selector, a, b):
per-component result = selector ? a : b
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
builtin_info was originally going to be a structure containing a bunch
of information, but after various rewrites, it turned into a boolean
availability predicate.
builtin_avail is a better name than builtin_info, since it doesn't
store any information other than availability.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
This creates a new replacement for the existing built-in function code.
The new module lives in builtin_functions.cpp (not builtin_function.cpp)
and exists in parallel with the existing system. It isn't used yet.
The new built-in function code takes a significantly different approach:
Instead of implementing built-ins via printed IR, build time scripts,
and run time parsing, we now implement them directly in C++, using
ir_builder. This translates to faster load times, and a much less
complex build system.
It also takes a different approach to built-in availability: each
signature now stores a boolean predicate, which makes it easy to
construct arbitrary expressions based on _mesa_glsl_parse_state's
fields. This is much more flexible than the old system, and also
easier to use.
Built-ins are also now stored in a single gl_shader object, rather
than being spread out across a number of shaders that need to be linked.
When searching for a matching prototype, we simply consult the
availability predicate. This also simplifies the code.
v2: Incorporate Matt Turner's feedback: use the new fma() function rather
than expr(). Don't expose textureQueryLOD() in GLSL 4.00 (since it
was renamed to textureQueryLod()). Also correct some #undefs.
v3: Incorporate Paul Berry's feedback: rename legacy to compatibility;
add comments to explain a few things; fix uvec availability; include
shaderobj.h instead of repeating the _mesa_new_shader prototype.
v4: Fix lack of TEX_PROJECT on textureProjGrad[Offset] (caught by oglc).
Add an out_var convenience function (more feedback by Matt Turner).
v5: Rework availability predicates for Lod functions. They were broken.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Enthusiastically-acked-by: Paul Berry <stereotype441@gmail.com>
Each ir_factory needs an instruction list and memory context in order to
be useful. Rather than creating an object and manually assigning these,
we can just use optional parameters in the constructor.
This makes it possible to create a ready-to-use factory in one line:
ir_factory body(&sig->body, mem_ctx);
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
Adding new convenience emitters makes it easier to generate IR involving
these opcodes.
bitfield_insert is particularly useful, since there is no expr() for
quadops.
v2: Add fma() and rename lrp() operands to x/y/a to match the GLSL
specification (suggested by Matt Turner). Fix whitespace issues.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
IR builder already offers a lot of swizzling functions, such as
swizzle_xxxx, swizzle_z, or swizzle_for_size.
The swizzle_xxxx style is convenient if you statically know which
components you want. swizzle_for_size is great if you want to select
the first few components. However, if you want to select components
based on, say, a loop counter, none of those are sufficient.
IR builder actually already had support for arbitrary swizzling, but
didn't expose it. This patch exposes that API.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
dotlike() uses ir_binop_mul for scalars, and ir_binop_dot for vectors.
When generating built-in functions, we often want to use regular
multiply for scalar signatures, and dot() for vector signatures.
ir_binop_dot only works on vectors, so we have to switch opcodes,
even if the code is otherwise identical. dotlike() makes this easy.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
We use "ret" as the function name since "return" is a C++ keyword, and
"ir_return" is already a class name.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
This adds two new signatures:
assign(lhs, rhs, condition, writemask);
assign(lhs, rhs, condition);
All the other existing APIs still exist.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
We already have ir_expression constructors for unary and binary
operations, which automatically infer the type based on the opcode and
operand types.
These are convenient and also required for ir_builder support.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
This isn't strictly necessary, since creators of ir_texture objects
should set LOD when relevant. However, it's nice to have a NULL pointer
in case they forget.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
During compilation, we'll use this to determine built-in availability.
The plan is to have a single shader containing every built-in in every
version of the language, but filter out the ones that aren't actually
available to the shader being compiled.
At link time, we don't actually need this filtering capability: we've
already imported prototypes for every built-in that the shader actually
calls, and they're flagged as is_builtin(). The linker doesn't import
any additional prototypes, so it won't pull in any unavailable
built-ins. When resolving prototypes to function definitions, the
linker ensures the values of is_builtin() match, which means that a
shader can't trick the linker into importing the body of an unavailable
built-in by defining a suspiciously similar prototype.
In other words, during linking, we can just pass in NULL. It will work
out fine.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
We can simply call the stored predicate function. If state is NULL,
just report that the function is available.
v2: Add a comment (requested by Paul Berry).
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
This promises the method won't modify the contents of the object.
This allows us to call it even with a const pointer to the state.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
A signature is a built-in if and only if builtin_info != NULL, so we
don't actually need a separate flag bit. Making a boolean-valued
method allows existing code to ask the same question while not worrying
about the internal representation.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
For the upcoming built-in function rewrite, we'll need to be able to
answer "Is this built-in function signature available?".
This is actually a somewhat complex question, since it depends on the
language version, GLSL vs. GLSL ES, enabled extensions, and the current
shader stage.
Storing such a set of constraints in a structure would be painful, so
instead we store a function pointer. When creating a signature, we
simply point to a predicate that inspects _mesa_glsl_parse_state and
answers whether the signature is available in the current shader.
Unfortunately, IR reader doesn't actually know when built-in functions
are available, so this patch makes it lie and say that they're always
present. This allows us to hook up the new functionality; it just won't
be useful until real data is populated. In the meantime, the existing
profile mechanism ensures built-ins are available in the right places.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
Otherwise, coordinates with four components would result in a MOV
with a destination writemask that has no channels enabled:
mov(8) g115<1>.F 0D { align16 WE_normal NoDDChk 1Q };
At best, this is stupid: we emit code that shouldn't do anything.
Worse, it apparently causes GPU hangs (observable with Chris's
textureGather test on CubeArrays.)
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Cc: Chris Forbes <chrisf@ijw.co.nz>
Cc: mesa-stable@lists.freedesktop.org
The change is very small. Do seamless filtering if either the context
enable is set or the sampler enable is set.
The AMD_seamless_cubemap_per_texture says:
"If TEXTURE_CUBE_MAP_SEAMLESS_ARB is emabled (sic) globally or the
value of the texture's TEXTURE_CUBE_MAP_SEAMLESS_ARB parameter is
TRUE, seamless cube map sampling is enabled..."
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Chris Forbes <chrisf@ijw.co.nz>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Appendix F.2 of the OpenGL ES 3.0.0 spec says:
"OpenGL ES 3.0 requires that all cube map filtering be
seamless. OpenGL ES 2.0 specified that a single cube map face be
selected and used for filtering."
Setting the field only in the context will work fine with sampler
objects (and drivers that support AMD_seamless_cubemap_per_texture)
because seamless filtering is used if *either* the context or the
sampler enable it:
"If TEXTURE_CUBE_MAP_SEAMLESS_ARB is emabled (sic) globally or the
value of the texture's TEXTURE_CUBE_MAP_SEAMLESS_ARB parameter is
TRUE, seamless cube map sampling is enabled..."
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Chris Forbes <chrisf@ijw.co.nz>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
Reported-by: Maxence Le Dore <maxence.ledore@gmail.com>
Thanked-by: Maxence Le Dore <maxence.ledore@gmail.com>
There is no GL_TEXTURE_CUBE_MAP_SEAMLESS in any version of OpenGL ES or
in any extension that applies to OpenGL ES. The same error check
already occurs for glTexParameteri.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Chris Forbes <chrisf@ijw.co.nz>
Cc: Maxence Le Dore <maxence.ledore@gmail.com>
This aligns the gfx, compute, and dma IBs to 8 DW boundries.
This aligns the the IB to the fetch size of the CP for optimal
performance. Additionally, r6xx hardware requires at least 4
DW alignment to avoid a hw bug. This also aligns the DMA
IBs to 8 DW which is required for the DMA engine. This
alignment is already handled in the gallium driver, but that
patch can be removed now that it's done in the winsys.
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
CC: "9.2" <mesa-stable@lists.freedesktop.org>
CC: "9.1" <mesa-stable@lists.freedesktop.org>
We support indirect addressing only on the vertex index, but some
shaders also use indirect addressing on attributes. This patch
adds support for indirect addressing on both dimensions inside
gs arrays.
Signed-off-by: Zack Rusin <zackr@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: José Fonseca <jfonseca@vmware.com>
When adding a new buffer to the beginning of the memory pool, we were
accidentally deleting the buffer that was first in the buffer list.
This was caused by a bug in the memory pool's linked list
implementation.
DPA2 is listed in the "Defeatured Instructions" section of the
965 PRM, Volume 4:
"The following instructions are removed from Gen4 implementation mainly
due to implementation cost/schedule reasons. They are candidates for
future generations."
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
RSR and RSL are listed in the "Defeatured Instructions" section of the
965 PRM, Volume 4:
"The following instructions are removed from Gen4 implementation mainly
due to implementation cost/schedule reasons. They are candidates for
future generations."
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Fixes a bug where if an uniform array is passed to a function the accesses
to the array are not propagated so later all but the first vector of the
uniform array are removed in parcel_out_uniform_storage resulting in
broken shaders and out of bounds access to arrays in
brw::vec4_visitor::pack_uniform_registers.
Cc: mesa-stable@lists.freedesktop.org
Reviewed-and-Tested-by: Matt Turner <mattst88@gmail.com>
Signed-off-by: Dominik Behr <dbehr@chromium.org>
Haswell GT2 and GT3 require the number of vertex shader URB entries to
be at least 64, not 32.
At the moment, we always meet this requirement automatically, because
in the absence of a geometry shader, we assign all available URB space
to the vertex shader. But when we turn on support for geometry
shaders, this lower limit will become important.
Reviewed-by: Chad Versace <chad.versace@linux.intel.com>
This patch creates a new file brw_vec4_vs_visitor.cpp, to contain code
that is specific to the vertex shader. Now the organization of vertex
shader and geometry shader visitor code is symmetric: vs-specific code
is in brw_vec4_vs_visitor.cpp, gs-specific code is in
brw_vec4_gs_visitor.cpp, and code shared between vs and gs is in
brw_vec4_visitor.cpp.
Acked-by: Kenneth Graunke <kenneth@whitecape.org>
This will allow it to be shared between brw_vec4_visitor.cpp and
brw_vec4_vs_visitor.cpp (which will be created in the next patch).
Acked-by: Kenneth Graunke <kenneth@whitecape.org>
The fixup code emulates non-BGRA render targets by adding an
extra instruction at the end of fragment shaders to swizzle the
output. To do this, we also swizzle the blend function. However
an oversight until now was that the writemask wasn't getting
swizzled. This patch fixes that which fixes a bunch of piglit
tests.
The old code in dri2_glx suffered from a typographical error that caused
the default version to be 2.1 instead of 1.2 (minimum required by the
Linux OpenGL ABI). drisw_glx had a similar error resulting in a default
version of 0.1.
Some driver/card combinations (r200/RV280, i915/915G) don't support
OpenGL 2.1. These create in some corner cases an indirect context
instead of a direct context when calling glXCreateContextAttribsARB().
This happens because of a bad default value. To avoid this, just used
the default value specified by the GLX_ARB_create_context specification:
"The default values for GLX_CONTEXT_MAJOR_VERSION_ARB and
GLX_CONTEXT_MINOR_VERSION_ARB are 1 and 0 respectively. In this
case, implementations will typically return the most recent version
of OpenGL they support which is backwards compatible with OpenGL 1.0
(e.g. 3.0, 3.1 + GL_ARB_compatibility, or 3.2 compatibility
profile)"
Refactor all the default value setting to dri2_convert_glx_attribs, and
make sure the correct defaults are set in that one place.
Signed-off-by: Rico Schüller <kgbricola@web.de>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Bugzilla http://bugs.winehq.org/show_bug.cgi?id=34238
Cc: "9.1 9.2" <mesa-stable@lists.freedesktop.org>
This patch adds liveness analysis to i915g and a couple
optimizations which benefit from it. One interesting
optimization turns (fake) indirect texture accesses into direct
texture accesses (the i915 supports a maximum of 4 indirect
texture accesses). Among other things this fixes a bunch of
piglit tests.
The vertex shader color outputs (gl_FrontColor, gl_BackColor,
gl_FrontSecondaryColor, and gl_BackSecondaryColor) don't have the same
names as the matching fragment shader color inputs (gl_Color and
gl_SecondaryColor). As a result, the qualifiers on them were not being
properly cross validated.
Full spec compliance required ir_variable::used and
ir_variable::assigned be set properly. Without the preceeding patch,
which fixes the ::clone method to copy them, this will not be the case.
Fixes all of the previously failing piglit
spec/glsl-1.30/linker/interpolation-qualifiers tests.
v2: Update callers of cross_validate_types_and_qualifiers and
cross_validate_front_and_back_color. The function signature changed in
v2 of a previous patch. Suggested by Paul.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=47755
The new function, cross_validate_types_and_qualifiers, will have
multiple callers from this file in future commits.
v2: Don't pass the names of the producer / consumer stages to
cross_validate_types_and_qualifiers. Instead, pass the types and get
the names only in the error paths. Suggested by Paul.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
Changes to the grammar for GL_ARB_shading_language_420pack (commit
6eec502) moved precision qualifiers out of the type_specifier production
chain. This caused declarations such as:
struct S {
lowp float f;
};
to generate parse errors. Section 4.1.8 (Structures) of both the GLSL
ES 1.00 spec and GLSL 1.30 specs says:
"Member declarators may contain precision qualifiers, but may not
contain any other qualifiers."
So, it sure seems like we shouldn't generate a parse error. :)
Instead of type_specifier, use fully_specified_type in struct members.
However, fully_specified_type allows a lot of other qualifiers that are
not allowed on structure members, so expeclitly disallow them.
Note, this makes struct_declaration look an awful lot like
member_declaration (used for interface blocks). We may want to
(somehow) unify these rules to reduce code duplication at some point.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=68753
Reported-by: Aras Pranckevicius <aras@unity3d.com>
Cc: Aras Pranckevicius <aras@unity3d.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Cc: "9.2" <mesa-stable@lists.freedesktop.org>
Remap any type or severity exclusive to KHR_debug to
something suitable for ARB_debug_output
Signed-off-by: Timothy Arceri <t_arceri@yahoo.com.au>
Reviewed-by: Brian Paul <brianp@vmware.com>
V3: make sure to add null terminator when setting label,
generate error when the client specifies an explicit
length that exceeds MAX_LABEL_LENGTH, set label pointer
to NULL when freed, and output correct length in
MAX_LABEL_LENGTH error message.
V2: fixed indentation of comment
Signed-off-by: Timothy Arceri <t_arceri@yahoo.com.au>
Reviewed-by: Brian Paul <brianp@vmware.com>
The main GL context's swtnl_im field is the VBO module's vbo_context
structure. Using the name "swtnl" in the name is confusing since
some drivers use hardware texturing and lighting, but still rely on the
VBO module for drawing.
v2: Forward declare the type and use that instead of void *
(suggested by Eric Anholt).
v3: Remove unnecessary cast (pointed out by by Topi Pohjolainen).
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
Some drawing functions take a single _mesa_prim object, while others
take an array of primitives. Both kinds of functions used a parameter
called "prim" (the singular form), which was confusing.
Using the plural form, "prims," clearly communicates that the parameter
is an array of primitives.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
can_cut_index_handle_prims() was passed an array of _mesa_prim objects
and a count, and ran a loop for that many iterations. However, it
treated the array like a pointer, repeatedly checking the first element.
This patch makes it actually check every primitive.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
Fixes assertion failues in 24 piglit tests with
MESA_GL_VERSION_OVERRIDE=3.0, 12 of which are now passing.
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
LIBGL_SHOW_FPS=1 makes GLX print FPS every second while other values do
nothing. Extend it so that LIBGL_SHOW_FPS=N will print the FPS every N
seconds.
Signed-off-by: Chia-I Wu <olvaffe@gmail.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
The VBO module actually calls us with an array of _mesa_prim objects.
For example, it may break up a DrawArrays() call into multiple
primitives when primitive restart is enabled.
Previously, we treated prim like a pointer, always accessing element 0.
This worked because all of the primitive objects in a single draw call
have the same value for num_instances and basevertex.
However, accessing an array as a pointer and using the wrong object's
fields is misleading. For stylistic reasons alone, we should use the
right object.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
These functions have almost identical code; the only difference is that
a few of the bits moved around. Adding a few trivial conditionals
allows the same function to work on all generations, and the resulting
code is still quite readable.
v2: Comment that the workaround flush is only necessary on SNB
(requested by Paul Berry).
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
Fixes broken rendering if these MRFs contained anything other than zero.
NOTE: This is a candidate for stable branches.
Signed-off-by: Chris Forbes <chrisf@ijw.co.nz>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
There is a slight functionality change. Previously we would compute a
common value for num_samplers for all stages, and populate that many
entries in each stage's surf_offset table regardless of how many
samplers each stage used. Now we only populate the number of entries
in the surf_offset table corresponding to the number of samplers
actually used by the stage.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Previously these functions would accept a pointer to the binding table
and an index indicating which entry in the binding table should be
updated. Now they merely take a pointer to the binding table entry to
be updated.
This will make it easier to generalize brw_texture_surfaces to support
geometry shaders.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
This patch implements pull constant upload, binding table upload, and
surface setup for geometry shaders, by re-using vertex shader code
that was generalized in previous patches.
Based on work by Eric Anholt <eric@anholt.net>.
v2: Update ditry bits for brw_gs_ubo_surfaces to account for commit
77d8fbc (mesa: add & use a new driver flag for UBO updates instead of
_NEW_BUFFER_OBJECT).
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
The hardware requires that after constant buffers for a stage are
allocated using a 3DSTATE_PUSH_CONSTANT_ALLOC_{VS,HS,DS,GS,PS}
command, and prior to execution of a 3DPRIMITIVE, the corresponding
stage's constant buffers must be reprogrammed using a
3DSTATE_CONSTANT_{VS,HS,DS,GS,PS} command.
Previously we didn't need to worry about this, because we only
programmed 3DSTATE_PUSH_CONSTANT_ALLOC_{VS,HS,DS,GS,PS} once on
startup (or, previous to that, whenever BRW_NEW_CONTEXT was flagged).
But now that we reallocate the constant buffers whenever geometry
shaders are switched on and off, we need to make sure the constant
buffers are reprogrammed.
We do this by adding a new bit, BRW_NEW_PUSH_CONSTANT_ALLOCATION, to
brw->state.dirty.brw.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Previously, we would always use the same push constant allocation
regardless of what shader programs were being run: the available push
constant space was split into 2 equal size partitions, one for the
vertex shader, and one for the fragment shader.
Now that we are adding geometry shader support, we need to do
something smarter. This patch adjusts things so that when a geometry
shader is in use, we split the available push constant space into 3
nearly-equal size partitions instead of 2.
Since the push constant allocation is now affected by GL state, it can
no longer be set up by brw_upload_initial_gpu_state(); instead it must
be set up by a state atom.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
This is required by the internal hardware docs and the PRM. Probably
the reason we were getting away with not doing it was because we only
emitted 3DSTATE_PUSH_CONSTANT_ALLOC_PS during startup. However that's
going to change with the introduction of geometry shaders.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Previously, we gave all of the URB space (other than the small amount
that is used for push constants) to the vertex shader. However, when
a geometry shader is active, we need to divide it up between the
vertex and geometry shaders.
The size of the URB entries for the vertex and geometry shaders can
vary dramatically from one shader to the next. So it doesn't make
sense to simply split the available space in two. In particular:
- On Ivy Bridge GT1, this would not leave enough space for the worst
case geometry shader, which requires 64k of URB space.
- Due to hardware-imposed limits on the maximum number of URB entries,
sometimes a given shader stage will only be capable of using a small
amount of URB space. When this happens, it may make sense to
allocate substantially less than half of the available space to that
stage.
Our algorithm for dividing space between the two stages is to first
compute (a) the minimum amount of URB space that each stage needs in
order to function properly, and (b) the amount of additional URB space
that each stage "wants" (i.e. that it would be capable of making use
of). If the total amount of space available is not enough to satisfy
needs + wants, then each stage's "wants" amount is scaled back by the
same factor in order to fit.
When only a vertex shader is active, this algorithm produces
equivalent results to the old algorithm (if the vertex shader stage
can make use of all the available URB space, we assign all the space
to it; if it can't, we let it use as much as it can).
In the future, when we need to support tessellation control and
tessellation evaluation pipeline stages, it should be straightforward
to expand this algorithm to cover them.
v2: Use "unsigned" rather than "GLuint".
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
This paves the way for sharing the code that will set up the vertex
and geometry shader pipeline state.
v2: Rename the base class to brw_stage_state.
Reviewed-by: Chad Versace <chad.versace@linux.intel.com>
Defines that previously referred to VS now refer to VEC4, since they
will be shared by the user-programmable vertex shader and geometry
shader stages.
Defines that previously referred to the Gen6 geometry shader stage
(which is only used for transform feedback) are now renamed to
explicitly refer to Gen6, to avoid confusion with the Gen7
user-programmable geometry shader stage.
Based on work by Eric Anholt <eric@anholt.net>.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Chad Versace <chad.versace@linux.intel.com>
This will avoid confusion when we add geometry shaders, since these
data structures will be shared by vertex and geometry shaders.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Chad Versace <chad.versace@linux.intel.com>
Now that the name "gs" is no longer used to refer to the legacy fixed
function geometry shaders, we can use it to refer to user-defined
geometry shaders.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Chad Versace <chad.versace@linux.intel.com>
"ff" is for "fixed function". This frees up the name "gs" to refer to
user-defined geometry shaders.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Chad Versace <chad.versace@linux.intel.com>
This mimics r600g. The R600_CONTEXT_xxx flags are added to rctx->b.flags
and si_emit_cache_flush emits the packets. That's it. The shared radeon code
tells us when the streamout cache should be flushed, so we have to check
the flags anyway.
There is a new atom "cache_flush", because caches must be flushed *after*
resource descriptors are changed in memory.
Functional changes:
* Write caches are flushed at the end of CS and read caches are flushed
at its beginning.
* Sampler view states are removed from si_state, they only held the flush
flags.
* Everytime a shader is changed, the I cache is flushed. Is this needed?
Due to a hw bug, this also flushes the K cache.
* The WRITE_DATA packet is changed to use TC, which fixes a rendering issue
in openarena. I'm not sure how TC interacts with CP DMA, but for now it
seems to work better than any other solution I tried. (BTW CIK allows us
to use TC for CP DMA.)
* Flush the K cache instead of the texture cache when updating resource
descriptors (due to a hw bug, this also flushes the I cache).
I think the K cache flush is correct here, but I'm not sure if the texture
cache should be flushed too (probably not considering we use TC
for WRITE_DATA, but we don't use TC for CP DMA).
* The number of resource contexts is decreased to 16. With all of these cache
changes, 4 doesn't work, but 8 works, which suggests I'm actually doing
the right thing here and the pipeline isn't drained during flushes.
Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Tested-by: Tom Stellard <thomas.stellard@amd.com>
There is a new "class" si_buffer_resources, which should be good enough for
implementing any kind of buffer bindings (constant buffers, vertex buffers,
streamout buffers, shader storage buffers, etc.)
I don't even keep a copy of pipe_constant_buffer - we don't need it.
The main motivation behind this is to have a well-tested infrastrusture
for setting up streamout buffers.
Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Tested-by: Tom Stellard <thomas.stellard@amd.com>
This streamout state code will be used by radeonsi.
There are new structures r600_common_context and r600_common_screen.
What is inherited by what is shown here:
pipe_context -> r600_common_context -> r600_context
pipe_screen -> r600_common_screen -> r600_screen
The common structures reside in drivers/radeon. Currently they only contain
enough functionality to be able to handle streamout. Eventually I'd like
the whole pipe_screen implementation to be shared and some of the context
stuff too.
This is quite big, but most changes are because of the new structures and
the fact r600_write_value is replaced by radeon_emit.
Thanks to Tom Stellard for fixing the build for r600g/compute.
Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Tested-by: Tom Stellard <thomas.stellard@amd.com>
It is incorrect to assume that src[0] of a SEND-from-GRF opcode is the
GRF. For example, FS_OPCODE_UNIFORM_PULL_CONSTANT_LOAD uses src[1] for
the GRF.
To be safe, loop over all the source registers and mark any GRFs. We
probably won't ever have more than one, but it's simpler to just check
all three rather than attempting to bail early.
Not observed to fix anything yet, but likely to. Parallels the bug fix
in the previous commit, which actually does fix known failures.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Eric Anholt <eric@anholt.net>
Cc: mesa-stable@lists.freedesktop.org
It is incorrect to assume that src[0] of a SEND-from-GRF opcode is the GRF.
VS_OPCODE_PULL_CONSTANT_LOAD_GEN7 uses an IMM as src[0], and stores the
GRF as src[1].
To be safe, loop over all the source registers and mark any GRFs. We
probably won't ever have more than one, but it's simpler to just check
all three rather than attempting to bail early.
Fixes assertion failures in Unigine Sanctuary since we started making
register allocation rely on split_virtual_grfs working. (The register
classes were actually sufficient, we were just interpreting an IMM as
a virtual GRF number.)
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=68637
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Eric Anholt <eric@anholt.net>
Cc: mesa-stable@lists.freedesktop.org
pstipple/aaline stages used PIPE_MAX_SAMPLER instead of
PIPE_MAX_SHADER_SAMPLER_VIEWS when dealing with sampler views.
Now these stages can't actually handle sampler_unit != texture_unit anyway
(they cannot work with d3d10 shaders at all due to using tex not sample
opcodes as "mixed mode" shaders are impossible) but this leads to crashes if
a driver just installs these stages and then more than PIPE_MAX_SAMPLER views
are set even if the stages aren't even used.
Reviewed-by: Zack Rusin <zackr@vmware.com>
Turns out we don't need to do much extra work for detecting this case,
since we are guaranteed to get a empty static texture state in this case,
hence just rely on format being 0 and return all zero then.
Previously needed dummy textures (would just have crashed on format being 0
otherwise) which cannot return the correct result for size queries and when
sampling textures with wrap modes using border.
As a bonus should hugely increase performance when sampling unbound textures -
too bad it isn't a useful feature :-).
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
Reviewed-by: Zack Rusin <zackr@vmware.com>
No idea if this is working right but copied straight from llvmpipe.
(Not only does this check the so_target but also use buffer->data instead
of buffer for the mapping.)
Just trying to get rid of a segfault testing something else...
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
Reviewed-by: Zack Rusin <zackr@vmware.com>
The only reason this was needed was because the fetch texel function had to
get the (dynamic) border color, but this is now done much earlier.
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
Instead of enhancing the AoS path so it can deal with it, just use SoA. Fixing
AoS path wouldn't be all that difficult (use all the same logic as SoA) but
considered not worth it for now.
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
If the app is asking us to do GL_COMPRESSED_RGBA, then the app obviously
doesn't have pre-compressed data to hand us. So don't choose a storage
format that we won't actually be able to compress and store.
Fixes black screen in warzone2100 when libtxc_dxtn is not present. Also
66 piglit tests.
NOTE: This is a candidate for the 9.2 branch.
Reported-by: Paul Wise <pabs@debian.org>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
You should only be flagging the formats as supported if you support them
anyway.
NOTE: This is a candidate for the 9.2 branch. (required for next commit)
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Thanks to Ken for trawling through my neglected public branches and
finding the bug in this change (inside a megacommit) that made me abandon
this work.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
This avoids the need to get the inter- and intra-tile offset and adjust
our miptree info based on them.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
These are things that happen to be occurring because of the batch flush at
the start of the blorp op (which exists to prevent batch space or aperture
space overflow), but the intention was for this sequence of state resets at
the end of blorp to be everything necessary for the next draw call.
Found when debugging the next commit, by comparing brw_new_batch() and
intel_batchbuffer_reset() to brw_blorp_exec().
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
The code that got replaced with map_raw didn't do the flush, but now
map_raw() is responsible for it and we don't have to worry about it.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
This gives us more information about why we're flushing that we can
use for handling our throttling.
v2 (Kenneth Graunke): Rebase on latest master, add missing
FLUSH_VERTICES and FLUSH_CURRENT, which fixes a regression in Glean's
polygonOffset test.
v3 (anholt): Drop FLUSH_CURRENT -- FLUSH_VERTICES is what we need, which
is "get any queued prims out of VBO and into the driver", not "update
ctx->Current so we can read it with the CPU." Also drop batch->used
check, which intel_batchbuffer_flush() does anyway.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
This was already happening because blorp happens to flush at the end of
every call, but we have been talking about removing that at some point,
and this would surely get overlooked.
v2 (Kenneth Graunke): Rebase on latest master. Note that we did remove
the other flush, and this change actually did get overlooked!
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
intel_flush() now did nothing except call through (and
intel_batchbuffer_flush() does the no-op check, too!)
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
df06745c5a made it so that we didn't
allocate extra uniform space for unused clip planes, which also
incidentally made us not allocate any space at all, which we were relying
on for this no-uniforms case. Instead of putting the knowledge of this
special HW exception into the thing that normally preallocates prog_data
for us, just allocate it here.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=68766
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Since we can have per-pixel lod we should also honor the filter per-pixel
(in fact we didn't honor it per quad neither in the multiple quad case).
Do this by running the linear path and simply beating the weights into shape
(the sample with the higher weight is the one which should have been chosen
with nearest filtering hence adjust filter weight to 1.0/0.0 based on that).
If all pixels use nearest filter (either min and mag) then still run just a
nearest filter as this is way cheaper (probably around 4 times faster for 2d,
more for 3d case) and it should be relatively rare that pixels really need
different filtering. OTOH if all pixels would require linear don't do anything
special since the linear path with filter adjustments shouldn't really be all
that much more expensive than ordinary linear, and we think it's rare that
min/mag filters are configured differently so there doesn't seem much value
in trying to optimize this further.
This does not yet fix the AoS path (though currently AoS is only used for
single quads hence it could be considered less broken, just never honoring
per-pixel filter decision but doing it per quad).
v2: simplify code a bit (unify min linear and min nearest cases)
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
While a sqrt here and there shouldn't hurt much (depending on the cpu) it is
possible to completely omit it since rho is only used for calculating lod and
there log2(x) == 0.5*log2(x^2). Depending on the exact path taken for
calculating lod this means we get a simple mul instead of sqrt (in case of
nearest mip filter in fact we don't need to replace the sqrt with something
else at all), only in some not very useful path this doesn't work (combined
brilinear calculation of int level and fractional lod, accurate rho calc but
brilinear filtering seems odd).
Apart from being faster as an added bonus this should increase our crappy
fractional accuracy of lod, since fast_log2 is only good for ~3bits and this
should increase accuracy by one bit (though not used if dimension is just one
as we'd need an extra mul there as we never had the squared rho in the first
place).
v2: use separate ilog2_sqrt function if we have squared rho.
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
This is just preparation for per-pixel (or per-quad in case of multiple quads)
min/mag filter since some assumptions about number of miplevels being equal
to number of lods no longer holds true.
This change does not change behavior yet (though theoretically when forcing
per-element path it might be slower with different min/mag filter since the
code will respect this setting even when there's no mip maps now in this case,
so some lod calcs will be done per-element just ultimately still the same
filter used for all pixels).
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
Fixes "Missing break in switch" defect reported by Coverity.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
The downstream android kernel driver is "kgsl", the upstream drm/kms
driver is called "msm". Since libdrm_freedreno handles the differences
between the two, we need to load the same thing for either device.
Signed-off-by: Rob Clark <robclark@freedesktop.org>
We need to set the flag on all the .xyzw components that are written by
the instruction, not just on .x. Otherwise a later use of rN.y (for
example) will not trigger the appropriate sync bit to be set.
Signed-off-by: Rob Clark <robclark@freedesktop.org>
Seems like most/all instructions have some restrictions about const src
registers. In seems like the 2 src (cat2) instructions can take at most
one const, and the 3 src (cat3) instructions can take at most one const
in the first 2 arguments. And so on. Handle this properly now.
Signed-off-by: Rob Clark <robclark@freedesktop.org>
GLSL 1.30 doesn't allow precision qualifiers on sampler types,
but in GLSL ES, sampler types are also allowed. This seems like
an oversight (since the intention of including these in GLSL 1.30
is to allow compatibility with ES shaders).
Currently, Mesa allows "default" precision qualifiers to be set for
sampler types in GLSL (commit d5948f2). This patch makes it follow
GLSL ES rules and also allow declaring sampler variables with a
precision qualifier in GLSL 1.30 (and later). e.g.
uniform lowp sampler2D sampler;
This fixes a shader compilation error in Khronos OpenGL conformance
test "depth_texture_mipmap".
V2: Update comments.
Signed-off-by: Ian Romanick <idr@lists.freedesktop.org>
Signed-off-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Ian Romanick <idr@lists.freedesktop.org>
Cc: <mesa-stable@lists.freedesktop.org>
Previously, we allocated space in brw_vs_prog_data's params and
pull_params arrays for MAX_CLIP_PLANES vec4s---even when it wasn't
necessary.
On a 64-bit architecture, this used 0.5 kB of space (8 clip planes *
4 floats per plane * 8 bytes per float pointer * 2 arrays of pointers =
512 bytes). Since this cost was per-vertex shader, it added up.
Conveniently, we already store the number of clip plane constants in the
program key. By using that, we can allocate the exact amount of space
needed. For the common case where user clipping is disabled, this means
0 bytes.
While we're here, mention exactly what code requires this extra space,
since it wasn't obvious.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
When validating draw parameters move check for 0 draw count last
(drawing with count 0 is not an error), so that other parameters (e.g.: the
primitive type) are validated and the correct errors (if applicable) are
generated.
>From the OpenGL 3.3 spec page 33 (page 48 of the PDF):
"[Regarding DrawArraysOneInstance, in terms of which other draw operations
are defined:]
If count is negative, an INVALID_VALUE error is generated."
This patch also changes the bahavior of MultiDrawElements to perform the draw
operation if some primitive's index counts are zero.
Signed-off-by: Fabian Bieler <fabianbieler@fastmail.fm>
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
MAD will be generated directly from ir_triop_fma, so this assertion
checks that all ir_expressions are usable.
Reviewed-by: Paul Berry <stereotype441@gmail.com>
MSAA was tested by one user on RS690 and it works for him with color
compression (CMASK) disabled. Our theory is that his chipset lacks CMASK RAM.
Since we don't have hardware documentation about which chipsets actually have
CMASK RAM, I had to take a guess based on the presence of HiZ.
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
In particular noone is interested in the vertex count, so drop that,
and also drop the duplicated num_primitives_generated /
so.primitives_storage_needed variables in drivers. I am unable for now to figure
out if primitives_storage_needed in SO stats (used for d3d10) should
increase if SO is disabled, though the equivalent num_primitives_generated
used for OpenGL definitely should increase. In any case we were only counting
when SO is active both in softpipe and llvmpipe anyway so don't pretend there's
an independent num_primitives_generated counter which would count always.
(This means the PIPE_QUERY_PRIMITIVES_GENERATED count will still be wrong just
as before, should eventually fix this by doing either separate counting for this
query or adjust the code so it always counts this even if SO is inactive depending
on what's correct for d3d10.)
Reviewed-by: Brian Paul <brianp@vmware.com>
SEQ and SNE are not native i915 instructions, so they each generate at
least 3 instructions. If both operands are uniforms or constants, we
get 5 instructions like:
U[1] = MOV CONST[1]
U[0].xyz = SGE CONST[0].xxxx, U[1]
U[1] = MOV CONST[1].-x-y-z-w
R[0].xyz = SGE CONST[0].-x-x-x-x, U[1]
R[0].xyz = MUL R[0], U[0]
This code is stupid. Instead of having the individual calls to
i915_emit_arith generate the moves to utemps, do it in the caller. This
results in code like:
U[1] = MOV CONST[1]
U[0].xyz = SGE CONST[0].xxxx, U[1]
R[0].xyz = SGE CONST[0].-x-x-x-x, U[1].-x-y-z-w
R[0].xyz = MUL R[0], U[0]
This allows fs-temp-array-mat2-index-col-wr and
fs-temp-array-mat2-index-row-wr to fit in hardware limits (instead of
falling back to software rasterization).
NOTE: Without pending patches to the piglit tests, these tests will now
fail. This is an unrelated, pre-existing issue.
v2: Copy most of the body of the commit message into comments in the
code. Suggested by Eric.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
The command is submitted once the event has been triggered, but it might not
have completed yet. Therefore, we have to add it to deps in order to wait on it.
Signed-off-by: Niels Ole Salscheider <niels_ole@salscheider-online.de>
Reviewed-by: Francisco Jerez <currojerez@riseup.net>
I previously fixed this partly in 9e8400f4c9,
however I didn't go far enough in testing it, now when I parse a TGSI shader
with arrays in it my iterator can see the ArrayID set to the proper value.
Reviewed-by: Brian Paul <brianp@vmware.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
Now that we use a fixed set of register classes, we can set up the
register set and conflict graphs once, at context creation, rather than
on every VS compile. This is obviously less expensive, and also what
we already do in the FS backend.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
We're soon going to be calling brw_alloc_reg_set() from outside of the
visitor, where we don't have the precomputed "max_grf" variable handy.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
For now, nothing else can get allocated over them. That may change at
some point in the future.
This also means that base_reg_count can be computed without knowing the
number of registers used for the payload, which is required if we want
to allocate the register set once at context creation time.
See commit 551e1cd44f, which implemented
virtually identical code in the FS backend.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
Arrays, structures, and matrices use large VGRFs of arbitrary sizes.
However, split_virtual_grfs() breaks those down into VGRFs of size 1.
For reference, commit 5d90b98879 is the
analogous change to the FS backend.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
(From a suggestion by Francisco Jerez)
If an enum represents a bitfield of flags, e.g.:
enum E {
A = 1,
B = 2,
C = 4,
D = 8,
};
then C++ normally prohibits statements like this:
enum E x = A | B;
because A and B are implicitly converted to ints before OR-ing them,
and an int can't be stored in an enum without a type cast. C, on the
other hand, allows an int to be implicitly converted to an enum
without casting.
In the past we've dealt with this situation by storing flag bitfields
as ints. This avoids ugly casting at the expense of some type safety
that C++ would normally have offered (e.g. we get no warning if we
accidentally use the wrong enum type).
However, we can get the best of both worlds if we override the |
operator. The ugly casting is confined to the operator overload, and
we still get the benefit of C++ making sure we don't use the wrong
enum type.
v2: Remove unnecessary comment and unnecessary use of "enum" keyword.
Use static_cast.
Reviewed-by: Chad Versace <chad.versace@linux.intel.com>
Reviewed-by: Francisco Jerez <currojerez@riseup.net>
We never noticed that this field was uninitialized because it is only
used in an error path that reports internal Mesa errors.
But it's silly to have it around anyway because &brw->ctx is
equivalent.
Should fix Coverity defect CID 1063351: Uninitialized pointer field
(UNINIT_CTOR) /src/mesa/drivers/dri/i965/brw_vec4_emit.cpp: 148
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
If brwNewProgram is asked to create a program for an unrecognized
target, don't bother falling back on _mesa_new_program(). That just
hides bugs.
Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
v2: Use assert() rather than _mesa_problem().
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Chad Versace <chad.versace@linux.intel.com>
330.frag is a direct copy of 150.frag.
330.glsl is 150.glsl combined with ARB_shader_bit_encoding.glsl.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
These are necessary in order to compile the built-in functions.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
glIsQuery is supposed to return false for names returned by glGenQueries
until their first use. BeginQuery is a use, but QueryCounter is also a
use.
From the ARB_timer_query spec:
"A timer query object is created with the command
void QueryCounter(uint id, enum target);
[...] If <id> is an unused query object name, the
name is marked as used [...]"
Fixes Piglit's spec/ARB_timer_query/query-lifetime.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Chad Versace <chad.versace@linux.intel.com>
Cc: mesa-stable@lists.freedesktop.org
I assume this should have been part of commit
7727fbb7c5. This (obviously) fixes a lot tests.
Signed-off-by: Henri Verbeet <hverbeet@gmail.com>
Reviewed-by: Marek Olšák <maraeo@gmail.com>
Commit 53e20b8b introduced the use of a template to initialize some
common fields. Move this copying of fields to before the common vp3
fields are initialized.
Reported-by: Martin Peres <martin.peres@labri.fr>
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Christian König <christian.koenig@amd.com>
The cmps.f.* instruction doesn't actually seem to give a float 1.0 or
0.0 output. It either needs a cov.u16f16 or add.s + sel.f16. This
makes SGT/SLT/etc more similar to CMP, so handle them in trans_cmp().
This fixes a bunch of piglit tests.
Signed-off-by: Rob Clark <robclark@freedesktop.org>
It seems there are a number of cases where instructions have limitations
about taking reading src's from const register file, so make
get_unconst() a bit easier to use.
Signed-off-by: Rob Clark <robclark@freedesktop.org>
We probably should get rid of assert() entirely, but at this stage it is
more useful for things to crash where we can catch it in a debugger.
With compile_error() we have a single place to set an error flag (to
bail out and return an error on the next instruction) so that will be a
small change later when enough of the compiler bugs are sorted.
But re-arrange/cleanup the error/assert stuff so we at least get a dump
of the TGSI that triggered it. So we see some useful output in piglit
logs.
Signed-off-by: Rob Clark <robclark@freedesktop.org>
Don't crash when no color buffer bound. Something caught when starting
to run piglit, fixes a hanful of piglit tests.
Signed-off-by: Rob Clark <robclark@freedesktop.org>
Category 4 instructions (rsq, rcp, sqrt, etc) seem to be unable to take
a const register as src. In these cases we need to move the src to a
temporary gpr first.
This is the second case of such a restriction, where the instruction
encoding appears to support a const src, but in fact the hw appears to
ignore that bit. So split things out into a helper that can be re-used
for any instructions which have this limitation.
Signed-off-by: Rob Clark <robclark@freedesktop.org>
Our current (rather naive) register assignment is based on mapping
different register files (INPUT, OUTPUT, TEMP, CONST, etc) based on the
max register index of the preceding file. But in some cases, the lowest
used register in a file might not be zero. In which case
file_count[file] != file_max[file] + 1.
Signed-off-by: Rob Clark <robclark@freedesktop.org>
Sometimes things other than color dst need saturating, like if there is
a 'clamp(foo, 0.0, 1.0)'. So for saturated dst add the extra
instructions to fix up dst.
Signed-off-by: Rob Clark <robclark@freedesktop.org>
The 1st src to add.s needs (r) flag (repeat), otherwise it will end up:
add.s dst.xyzw, tmp.xxxx -1
instead of:
add.s dst.xyzw, tmp.xyzw, -1
Also, if we are using a temporary dst to avoid clobbering one of the src
registers, we actually need to use that as the dst for the sel
instruction.
Signed-off-by: Rob Clark <robclark@freedesktop.org>
This patch adds support for:
PIPE_COMPUTE_CAP_MAX_INPUT_SIZE
PIPE_COMPUTE_CAP_MAX_LOCAL_SIZE
Return the values reported by the closed source driver for now.
Signed-off-by: Niels Ole Salscheider <niels_ole@salscheider-online.de>
Reviewed-by: Tom Stellard <thomas.stellard@amd.com>
Previously, the min/mag switchover point when using nearest/none mip
filter was effectively -0.5 which can't be right. Looks like new OpenGL
thinks it's ok if it's always 0.0 (older versions required 0.5 in some
cases), let's hope everybody else thinks that's fine too.
Refactor this slightly and get the per-quad/per-pixel min/mag decision
values further down to sampling, though still only the first component
is used yet.
While here also fix code trying to skip lod bias application etc. when
mipfilter is none, as this is still needed for determining min/mag filter.
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
As of "2f142d59 build: Add --enable-gallium-osmesa flag." the pkgconfig
file from classic osmesa is no longer installed when building gallium
osmesa, so copy it to gallium osmesa and install the copy instead.
CC: "9.2" <mesa-stable@lists.freedesktop.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
This patch introduces the vec4_gs_visitor class, which translates
geometry shaders from GLSL IR to back-end opcodes.
This class is derived from vec4_visitor (which is also the base class
for vec4_vs_visitor), so as a result most of the back end code is
shared. The only parts that differ are:
- Geometry shaders use a different input payload organization, since
the inputs need to match up with the outputs of the previous
pipeline stage (vec4_gs_visitor::setup_payload() and
vec4_gs_visitor::setup_varying_inputs()).
- Geometry shader input array dereferences need a special stride
computation, since all geometry shader inputs are interleaved into
one giant array (vec4_gs_visitor::compute_array_stride()).
- There are no geometry shader system values
(vec4_gs_visitor::make_reg_for_system_value()).
- At the beginning of a geometry shader, extra data in R0 needs to be
zeroed out, and a vertex counter needs to be initialized
(vec4_gs_visitor::emit_prolog()).
- When EmitVertex() appears in the shader, the current contents of
output variables need to be emitted to the URB, and the vertex
counter needs to be incremented
(vec4_gs_visitor::visit(ir_emit_vertex *)).
- When generating a URB_WRITE message to output vertex data, the
current state of the vertex counter needs to be used to store a
write offset in the message header
(vec4_gs_visitor::emit_urb_write_header()).
- The URB_WRITE message that outputs vertex data needs to be sent
using GS_OPCODE_URB_WRITE, since VS_OPCODE_URB_WRITE would overwrite
the offsets in the message header
(vec4_gs_visitor::emit_urb_write_opcode()).
- At the end of a geometry shader, the final vertex count needs to be
delivered using a URB WRITE message
(vec4_gs_visitor::emit_thread_end()).
- EndPrimitive() functionality is not implemented yet
(vec4_gs_visitor::visit(ir_end_primitive *)).
- There is no support for assembly shaders
(vec4_gs_visitor::emit_program_code()).
v2: Make num_input_vertices const. Refer to registers as rN rather
than gN, for consistency with the PRM. Fix misspelling. Improve
comment in the ir_emit_vertex visitor explaining why we emit vertices
inside a conditional. Enclose the conditional code in the
ir_emit_vertex visitor between curly braces.
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
This will be used by geometry shaders to implement the EmitVertex()
function, since it requires writing data to a dynamically-determined
offset within the geometry shader's URB entry.
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
The arguments to brw_urb_WRITE() were getting pretty unwieldy, and we
have to add more flags to support geometry shaders anyhow.
Also plumb these flags through brw_clip_emit_vue(),
brw_set_urb_message(), and the vec4_instruction class.
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
When I initially generalized the vec4_visitor class in preparation for
geometry shaders, I assumed that the setup_attributes() function would
need to be different between vertex and geometry shaders, but its
caller, setup_payload(), could be shared. So I made
setup_attributes() a virtual function.
It turns out this isn't true; setup_payload() needs to be different
too, since the geometry shader payload sometimes includes an extra
register (primitive ID) that has to come before uniforms.
So setup_payload() needs to be the virtual function instead of
setup_attributes().
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Both 3DSTATE_VS and 3DSTATE_GS have a dispatch_grf_start_reg control,
which determines the register where the hardware delivers data sourced
from the URB (push constants followed by per-vertex input data).
For vertex shaders, we always set dispatch_grf_start_reg to 1, since
R1 is always the first register available for push constants in vertex
shaders.
For geometry shaders, we'll need the flexibility to set
dispatch_grf_start_reg to different values depending on the behvaiour
of the geometry shader; if it accesses gl_PrimitiveIDIn, we'll need to
set it to 2 to allow the primitive ID to be delivered to the thread in
R1.
This patch eliminates the assumption that dispatch_grf_start_reg is
always 1. In vec4_visitor, we record the regnum that was passed to
vec4_visitor::setup_uniforms() in prog_data for later use. In
vec4_generator, we consult this value when converting an abstract
UNIFORM register to a concrete hardware register. And in the code
that emits 3DSTATE_VS, we set dispatch_grf_start_reg based on the
value recorded in prog_data.
This will allow us to set dispatch_grf_start_reg to the appropriate
value when compiling geometry shaders. Vertex shaders will continue
to always use a dispatch_grf_start_reg of 1.
v2: Make dispatch_grf_start_reg "unsigned" rather than "GLuint".
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
This patch moves the following things into brw_vec4.{cpp,h}:
- struct brw_vec4_compile
- struct brw_vec4_prog_key
- brw_vec4_prog_data_compare()
- brw_vec4_prog_data_free()
This will allow us to avoid having to include brw_vs.h in
geometry-shader-specific files.
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
The patch that follows will move the definition of struct
brw_vec4_prog_key from brw_vs.h to brw_vec4.h, making it necessary for
brw_vs.h to include brw_vec4.h (because brw_vs.h defines struct
brw_vs_prog_key, which contains brw_vec4_prog_key as a member). Since
brw_vs.h is included from C source files, that means that brw_vec4.h
will need to be safe to include from C. Same for brw_shader.h, since
it is included by brw_vec4.h.
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
This is backwards from what we are going to want in the long term, which is:
- brw_vec4.h declares general-purpose vec4 infrastructure needed by
both VS and GS
- brw_vs.h includes brw_vec4.h and adds VS-specific parts.
- brw_gs.h includes brw_vec4.h and adds GS-specific parts.
Note that at the moment brw_vec.h contains a fair amount of
VS-specific declarations--I plan to address that in a later patch.
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Otherwise any GS that requires lowering (e.g. one that uses
gl_ClipDistance as an input or output) will fail to work.
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
This patch extracts the following logic from
validate_vertex_shader_executable():
(a) Generate an error if the shader writes to both gl_ClipDistance and
gl_ClipVertex.
(b) Record whether the shader writes to gl_ClipDistance in
gl_shader_program for use by the back-end.
(c) Record the size of gl_ClipDistance in gl_shader_program for use by
transform feedback logic.
And moves it into a function that is shared between vertex and
geometry shaders.
Strictly speaking we only need to have shared logic for (b) and (c)
right now (since (a) only matters in compatibility contexts, and we're
only implementing geometry shaders in core contexts right now). But
the three are closely related enough that it seems sensible to keep
them together.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
enums were being converted twice resulting in incorrect values.
The extra conversion has been removed and the redundant assert is
removed also.
Cc: 9.2 <mesa-stable@lists.freedesktop.org>
Signed-off-by: Timothy Arceri <t_arceri@yahoo.com.au>
Reviewed-by: Brian Paul <brianp@vmware.com>
The previous value of (GLuint64) ~0 has some problems:
GL_MAX_SERVER_WAIT_TIMEOUT is supposed to be a GLuint64 value, but has
to be queried via GetInteger64v(), which returns a GLint64. This means
that some applications are likely to treat it as a signed integer, where
~0 means -1. Negative values are nonsensical and problematic.
When interpreted correctly, ~0 translates to about 0.58 million years,
which seems rather excessive.
This patch changes it to 0x1fff7fffffff, which is about 1.11 years.
This is still plenty long, and is the same as both an int64 and uint64.
Applications that accidentally store it in a 32-bit int/unsigned also
get a non-negative value, which is again the same as both int and
unsigned. This value was suggested by Ian Romanick.
v2: Add the ULL prefix on the constant (suggested by Ian).
Fixes Piglit's spec/!OpenGL 3.2/get-integer-64v.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Cc: mesa-stable@lists.freedesktop.org
_mesa_meta_begin() sets up an orthographic project and initializes the
viewport based on the current drawbuffer's width and height. This is
likely the window size, since it occurs before the meta operation binds
any temporary buffers.
decompress_texture_image needs the viewport to be the size of the image
it's trying to draw. Otherwise, it may only draw part of the image.
v2: Actually set the projection properly too.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=68250
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Cc: Mak Nazecic-Andrlon <owlberteinstein@gmail.com>
Fixes inconsistent failure of gles2conform/GL2Tests/glUniform/glUniform.test
under gnome-shell. What follows is a description of the bug and its fix.
When intel_update_renderbuffers() allocates a miptree for a winsys
renderbuffer, it propagates the renderbuffer's format to become also the
miptree's format.
If the winsys color buffer format is SARGB, then, in the first call to
eglMakeCurrent, intel_gles3_srgb_workaround() changes the renderbuffer's
format to ARGB. That is, it changes the format from sRGB to non-sRGB.
However, it changes the renderbuffer's format *after*
intel_update_renderbuffers() has allocated the renderbuffer's miptree.
Therefore, when eglMakeCurrent returns, the miptree format (SARGB)
differs from the renderbuffer format (ARGB).
If the X server reallocates the color buffer,
intel_update_renderbuffers() will create a new miptree for the
renderbuffer. The new miptree's format (ARGB) will differ from old
miptree's format (SARGB). This mismatch between old and new miptrees
causes bugs.
Fix the bug by moving intel_gles3_srgb_workaround() to occur *before*
intel_update_renderbuffers().
CC: "9.2" <mesa-stable@lists.freedesktop.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=67934
Signed-off-by: Chad Versace <chad.versace@linux.intel.com>
Except for explicit derivs with cube maps which are very bogus anyway.
Just like explicit lod this is only used if no_quad_lod is set in
GALLIVM_DEBUG env var.
Minification is terrible on cpus which don't support true vector shifts
(but should work correctly). Cannot do the min/mag filter decision (if
they are different) per pixel though, only selecting different mip levels
works.
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
block size depth is always 1 even for compressed formats (unless someone
invents true 3d compressed formats at least which we can't represent).
Nearest (and soa) path had it right.
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
We have set up 3DSTATE_SBE (or 3DSTATE_SF on GEN6) in
ilo_shader_select_kernel_routing(). There is no need to pass the last shader
stage to the GPE function.
The Gallium implementation is apparently not ready for regular
consumption, so as much as I hate adding more build-time options, here's
another.
Acked-by: Brian Paul <brianp@vmware.com>
The variable means that UBO qualifiers are allowed in a particular
context (e.g., not allowed in a struct field declaration), rather than a
particular set of UBO qualifiers are valid.
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
This was invaluable when debugging the global copy propagation
algorithm. We may as well commit it in case someone needs to print
out the sets in the future.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
The (complicated!) math is all identical, there's just minimal differences how
sign bit is calculated plus there's an additional subtraction for the argument
going into the polynomial for cos.
The logic stays 100% the same (with a small exception, sign bit calculation for
sin is minimally simplified, applying sign mask after xoring the arguments
instead of applying it to each argument).
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
IVB/BYT also has the same L3 cacheability control in MOCS as HSW,
so let's make use of it.
pts/xonotic and pts/reaction @ 1920x1080 gain ~4% on my IVB GT2. Most
other things show less gains/no regressions, except furmark which
loses some 10 points.
I didn't have a BYT at hand for testing.
v2: Don't check (brw->gen == 7) in gen7 functions. (chadv)
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Chad Versace <chad.versace@linux.intel.com>
Just spotted these unpopulated MOCS fields when comparing the code
against BSpec. Set the MOCS to the same as everywhere else in Haswell:
L3-cacheable.
v2: Annotate state packet fields (chadv).
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Chad Versace <chad.versace@linux.intel.com>
The NVIDIA driver doesn't expose them, and piglit's
arb_texture_compression-invalid-formats expects them to not be there.
This, with the previous commit, fixes piglit
arb_texture_compression-invalid-formats.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
Cc: "9.2" <mesa-stable@lists.freedesktop.org>
This is required by the spec, and it's a bit tricky because the default
precision is scoped. As a result, I'm slightly abusing the symbol
table.
Fixes piglit no-default-float-precision.frag tests and the piglit
default-precision-nested-scope-0[1234].frag tests that are currently on
the piglit mailing list for review.
On IRC I got confirmation from cwabbot that ARM (Mali T6xx and T400)
enforces this requirement and from kusma that NVIDIA (Tegra2) enforces
this requirement. We should be safe from regressing shipping
applications.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Cc: "9.2" <mesa-stable@lists.freedesktop.org>
We never noticed this before because we previously didn't enfoce GLSL ES
fragement shader requirements that precision be defined. There may also
have been some interaction here with the addition of
GL_ARB_shading_language_420pack, but it doesn't appear to me that it
added any new bugs (just perhaps uncovered some old ones).
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Cc: "9.2" <mesa-stable@lists.freedesktop.org>
Going to need this soon (not going to bother with avx2 intrinsics at this time
but don't want to do workarounds for true vector shifts if llvm itself can use
them just fine and won't need the gazillion instruction emulation).
Not really tested other than my cpu returns 0 for these features...
(I have no idea if llvm actually would emit avx2/xop instructions neither...)
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
Need to check the wrap mode of the actually used coords not a fixed 2.
While checking more than necessary would only potentially disable aos and
not cause any harm I'm pretty sure for 3d textures it could have caused
assertion failures (if s,t coords have simple filter and r not).
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
Turns out it is actually very complicated to figure out what a format really
is wrt range, as using channel information for determining unorm/snorm etc.
doesn't work for a bunch of cases - namely compressed, subsampled, other.
Also while here add clamping for uint/sint as well - d3d10 doesn't actually
need this (can only use ld with these formats hence no border) and we could
do this outside the shader for GL easily (due to the fixed texture/sampler
relation) do it here too just so I can forget about it.
v2: move border color clamping out of fetch texel. Also change it to clamp
the whole border vector at once (and use vectorized load of border color),
which saves a couple of instructions - needs some different handling of
mixed signed/unsigned formats so skip the per channel stuff and just derive
this from first channel except for special formats.
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
There's a new debug value used to disable per-quad lod optimizations
in fragment shader (ignored for vs/gs as the results are just too wrong
typically). Also trying to detect if a supplied lod value is really a
scalar (if it's coming from immediate or constant file) in which case
sampler code can use this to stay on per-quad-lod path (in fact for
explicit lod could simplify even further and use same lod for both
quads in the avx case but this is not implemented yet).
Still need to actually implement per-element lod bias (and derivatives),
and need to handle per-element lod in size queries.
v2: fix comments, prettify.
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
The rules were writing files to e.g. util/u_indices_gen.py, but in an
out-of-tree build this directory doesn't exist in the build directory. So,
create the directories just in case.
Cc: mesa-stable@lists.freedesktop.org
Reviewed-by: Matt Turner <mattst88@gmail.com>
Signed-off-by: Ross Burton <ross.burton@intel.com>
The LLVM R600 backend currently always uses separate VGPRs for these.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=68162
(Centroid interpolation is identical to center interpolation without
multisampling, so the shader hardware was only pre-loading one set of
interpolation coefficients, and the pixel shader code was using
uninitialized values as the centroid interpolation coefficients)
Cc: mesa-stable@lists.freedesktop.org
Tested-by: Laurent Carlier <lordheavym@gmail.com>
Now that we have the number of samplers available, we don't need to
iterate over all 16. This should be particularly helpful for vertex
shaders.
v2: Use the correct shader program (caught by Paul Berry).
This needs to initialize the exact same set of sampler swizzles as
the actual key setup, or else we end up doing recompiles due to some
being XYZW and others being 0.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
The first field of a record in a UBO has the aligment of the record
itself.
Fixes piglit vs-struct-pad, fs-struct-pad, and (with the patch posted to
the piglit list that extends the test) layout-std140.
NOTE: The bit of strangeness with the version of visit_field without the
record_type poitner is because that method is pure virtual in the base
class. The original implementation of the class did this to ensure
derived classes remembered to implement that flavor. Now they can
implement either flavor but not both. I don't know a C++ way to enforce
that.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=68195
Cc: "9.2 9.1" mesa-stable@lists.freedesktop.org
The outer-most record is passed into the visit_field method for
the first field. In other words, in the following structure:
struct S1 {
vec4 v;
float f;
};
struct S {
S1 s1;
S1 s2;
};
uniform Ubo {
S s;
};
s.s1.v would get record_type = S (because s1.v is the first non-record
field in S), and s.s2.v would get record_type = S1. s.s1.f and s.s2.f
would get record_type = NULL becuase they aren't the first field of
anything.
This new overload isn't used yet, but the next patch will add several
uses.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
Cc: "9.2 9.1" mesa-stable@lists.freedesktop.org
For some reason, we didn't use this information even though the VS
backend has computed it (albeit poorly) for ages.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
Unlike the FS, the VS backend already computed the binding table size.
However, it did so poorly: after compilation, it looked to see if any
pull constants/textures/UBOs were in use, and set num_surfaces to the
maximum surface index for that category. If the VS only used a single
texture or UBO, this overcounted by quite a bit.
The shader time surface was also noted at state upload time (during
drawing), not at compile time, which is inefficient. I believe it also
had an off by one error.
This patch computes it accurately, while also simplifying the code.
It also renames num_surfaces to binding_table_size, since num_surfaces
wasn't actually the number of surfaces used. For example, a VS that
used one UBO and no other surfaces would have set num_surfaces to
SURF_INDEX_VS_UBO(1) == 18, rather than 1. A bit of a misnomer there.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
Computing the minimum size was easy, and done at compile-time for no
extra overhead here. Making the binding table smaller wastes less batch
space.
Adding the CACHE_NEW_WM_PROG dirty bit isn't strictly necessary, since
other atoms depend on it and flag BRW_NEW_SURFACES. However, it's best
to add it for clarity and safety. It shouldn't add any new overhead.
v2: Use binding_table_size, rather than max_surface_index.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
By tracking the maximum surface index used by the shader, we know just
how small we can make the binding table.
Since it depends entirely on the shader program, we can just compute
it once at compile time, rather than at binding table emit time (which
happens during drawing).
v2: Store binding_table_size, rather than max_surface_index, for
consistency with the VS (which needs to be able to represent 0
surfaces).
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
SURF_INDEX_DRAW() has been the identity function since the dawn of time,
and both the shader code and binding table upload code relied on that,
simply using X rather than SURF_INDEX_DRAW(X).
Even if that continues to be true, using the macro clarifies the code.
The comment about draw buffers needing to be first in order for
headerless render target writes to work turned out to be wrong; with
this change, SURF_INDEX_DRAW can be changed to arbitrary indices and
everything continues working.
The confusion was over the "Render Target Index" field in the FB write
message header. If it were a binding table index, then RT 0 would have
to be at index 0 for headerless FB writes to work. However, it's
actually an index into the blend state table, so there's no problem.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Cc: Paul Berry <stereotype441@gmail.com>
Now that we have the number of samplers available, we don't need to
iterate over all 16. This should be particularly helpful for vertex
shaders.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
Previously, we computed sampler counts when generating the SAMPLER_STATE
table. By computing it earlier, we should be able to shorten a bunch of
loops.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
This allows us to avoid uploading the VS sampler state table if only the
fragment program changes.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
Now, each shader stage has a sampler state table that only refers to the
samplers actually used by that problem. This should make the VS table
non-existant or very small.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
Also upload separate sampler default/texture border color entries.
At the moment, this is completely idiotic: both tables contain exactly
the same contents, so we're simply wasting batch space and CPU time.
However, soon we'll only upload data for textures actually /used/ in
a particular stage, which will usually make the VS table empty and
very likely eliminate all redundancy. This is just a stepping stone.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
Like the previous patch, this simply pushes direct access to brw->wm up
one level in the call chain. Rather than passing the whole array, we
just pass a pointer to the correct spot in the array, similar to what we
do for the actual sampler state structure.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
When we begin uploading separate sampler state tables for VS and FS,
we won't be able to use &brw->wm.sdc_offset[ss_index]. By passing it in
as a parameter, we push the problem up to the caller.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
Currently, we only have a single sampler state table shared among all
stages, so we just copy wm.sampler_count into vs.sampler_count.
In the future, each shader stage will have its own SAMPLER_STATE table,
at which point we'll need these separate sampler counts.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
I believe the data flow analysis actually works now, and it should be
safe to re-enable global copy propagation. It even does things now.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
Since the initial value for livein is an overestimation (0xffffffff),
it's extremely likely that it will shrink, which means we can't simply
OR in new bits - we need to fully recompute it based on the current
liveout values.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
Since we start with an overestimation of livein (0xffffffff), successive
steps can actually take away values. This means we can't simply OR in
new liveout values; we need to recompute it from scratch at each
iteration of the fixed-point algorithm.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
The starting block always has livein = 0 and liveout = copy. Since we
start with real data, not estimates, there's no need to refine it with
the fixed point algorithm.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
The previous commit properly initialized liveout. This previous
(and incorrect) initialization is no longer necessary.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
Previously, livein was initialized to 0 for all blocks. According to
the textbook, it should be the universal set (~0) for all blocks except
the one representing the start of the program (which should be 0).
liveout also needs to be initialized to COPY for the initial block.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
According to page 360 of the textbook, the proper formula for liveout
is:
CPout(n) = COPY(i) union (CPin(i) - KILL(i))
Previously, we omitted COPY.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
Excluding the existing liveout bits is a deviation from the textbook
algorithm. The reason for doing so was to determine if the value
changed, which means the fixed-point algorithm needs to run for another
iteration.
The simpler way to do that is to save the value from step (N-1) and
compare it to the new value at step N.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
This is the "COPY" set from Muchnick's textbook, which is necessary
to do the dataflow algorithm correctly.
v2: Simplify initialization based on Paul Berry's observation that
out_acp contains exactly what needs to be in the COPY set.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
Although this function currently only initializes the KILL set, it will
soon initialize other data flow sets as well.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
To compute the actual liveout/livein data flow values, we start with
some initial values and apply a fixed-point algorithm until they settle.
Previously, we iterated through all blocks, updating both liveout and
livein together in one pass. This is awkward, since computing livein
for a block requires knowing liveout for all parent blocks. Not all
of those parent blocks may have been processed yet.
This patch separates the two. First, we update liveout for all blocks.
At iteration N of the fixed-point algorithm, this uses livein values
from iteration N-1. Secondly, we update livein for all blocks. At
step N, this uses the liveout information we just computed (in step N).
This ensures each computation has a consistent picture of the data,
rather than seeing an random mix of data from steps N-1 and N depending
on the order of the blocks in the CFG data structure.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
This variable indicates that the fixed-point algorithm made changes to
the data at this step, so it needs to run for another iteration.
"progress" seems a nicer name for that.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
The fixed-point algorithm needs to run at least once, so a do-while loop
is more natural.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
The dataflow analysis used for global copy propagation is severely
broken, and I believe it doesn't actually do anything. Fixing it will
require a lot of changes, each of which might break things.
Once all the fixes land, we can re-enable this.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
Any decent compiler will do this for us, although doing this
will make grepping through the code alot easier.
v2: In both mixer and query interface
v3: rebase
Reviewed-by: Christian König <christian.koenig@amd.com> [v1]
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Code should loop through and cleanup the three (VL_NUM_COMPONENTS) idct
buffers, rather than doing the first one three times.
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Free any allocated memory and return BadAlloc if create_video_buffer()
has failed to create a buffer.
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Not seen in the wild yet, but seems like a reasonable thing to do.
[suggested by Christian]
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
I was looking into some minor 422 issues/discrepencies I noticed long
ago using vdpau on my rv790.
I noticed that there is code that is halving height rather than width -
422 is full height AFAIK.
Making the changes below doesn't actually make any noticable difference
to what I was looking into.
Maybe there are more but here's three I've found so far
Reviewed-by: Christian König <christian.koenig@amd.com>
We are getting close to the maximum number of BRW_NEW_* bits that can
be stored in brw->state.dirty.brw without overflowing 32 bits, and
geometry shaders are going to add more. Add a STATIC_ASSERT so that
we will be alerted when we need to switch to 64 bits.
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Previously, we were asserting that each driver specified an NConfigOptions
exactly equal to the number of options they supplied, leading to frequent
bugs when people would forget to adjust the value when adjusting driver
options. Instead, just overallocate the table by a bit and leave sanity
checking to the assert in findOption().
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Consistently using a "The ___ driver hook." line at the the top of each
function's comment block makes it easy to see at a glance what function
is being implemented.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
This code upload performs batched uploads via a BO. By moving it out to
a separate file, intel_buffer_objects.c only provides the core buffer
object functionality.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
GL_APPLE_object_purgeable creates a mechanism for marking OpenGL objects
as "purgeable" so they can be thrown away when system resources become
scarce. It specifically applies to buffer objects, textures, and
renderbuffers.
The intel_buffer_objects.c file provides core functionality for GL
buffer objects, such as MapBufferRange and CopyBufferSubData. Having
texture and renderbuffer functionality in that file is a bit strange.
The 2010 copyright on the new file is because Chris Wilson first added
this code in January 2010 (commit 755915fa).
v2: Actually remember to call the new dd table setup function.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
The surface allocator understands the scanout flag just fine.
This seems to improve performance for Ubuntu Unity on top of st/xorg
and it fixes the cursor.
Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>
This started as an attempt to add support for MSAA texture transfers and
MSAA depth-stencil decompression for the DB->CB copy path.
It has gotten a bit out of control, but it's for the greater good.
Some changes do not make much sense, they are there just to make it look
like the other driver.
With a few cosmetic modifications, r600_texture.c can be shared with
a symlink.
Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>
This is glBlitFramebuffer support for MSAA surfaces as required by GL 3.0
and texturing as required by GL 3.2 and GL_ARB_texture_multisample.
Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>
This is basic MSAA support which should work with most apps.
Some features are missing, those will be implemented by other commits.
Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>
It moves all sampler view descriptors to a buffer.
It supports partial resource updates and it can also unbind resources
(required for FMASK texturing).
The buffer contains all sampler view descriptors for one shader stage,
represented as an array. On top of that, there are N arrays in the buffer,
which are used to emulate context registers as implemented by the previous
ASICs (each array is a context).
This uses the RCU synchronization approach to avoid read-after-write hazards
as discussed in the thread:
"radeonsi: add FMASK texture binding slots and resource setup"
CP DMA is used to clear the descriptors at context initialization and to copy
the descriptors from one context to the next.
v2: - use PKT3_DMA_DATA on CIK (I'll test CIK later)
- turn the bool CP DMA parameters into self-explanatory flags
- add a nice simple API for packet emission to radeon_winsys.h
- use 256 contexts, 128 causes texture corruption in openarena
It shouldn't be necessary to call radeon_winsys::cs_flush() from
radeonsi_launch_grid(), because the state tracker is responsible for
flushing the pipeline at the appropriate time. The current behavior is
also wrong, because radeonsi_launch_grid() submits packets to the
compute ring, but when the state tracker calls pipe->flush() everything
is submitted to the graphics ring. This has the potential to create a
race condition.
The downside of removing this flush is that the compute dispatch packets
will be sent to the graphics ring rather than the compute ring.
In the future we will need to come up with a way to detect 'compute'
command streams and submit them to the appropriate ring.
Signed-off-by: Marek Olšák <marek.olsak@amd.com>
Previously, INTEL_DEBUG=bat would dump messages like:
intel_mipmap_tree.c:1643: Batchbuffer flush with 456b used
This only reported the space used for command packets, and didn't
report any information on the space used for indirect state.
Now it dumps:
intel_context.c:366: Batchbuffer flush with 6128b (pkt) + 4288b (state)
= 10416b (31.8%)
This conveniently shows the breakdown of space used for packets vs.
state, as well as the percentage of batchbuffer space.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
We emit these before configuring depth in the normal path, or actually
using the depth buffer in BLORP - we just failed to emit them when
disabling depth altogether.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Chad Versace <chad.versace@linux.intel.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
We emit these before configuring depth in the normal path, or actually
using the depth buffer in BLORP - we just failed to emit them when
disabling depth altogether.
On Sandybridge, this also requires the post_sync_nonzero flush.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Chad Versace <chad.versace@linux.intel.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Previously, copy propagation would cause bitcast_f2u(abs(float)) to
be performed in a single step, but the application of source modifiers
(abs, neg) happens after type conversion, leading to incorrect results.
That is, for bitcast_f2u(abs(float)) we would in fact generate code to
do abs(bitcast_f2u(float)).
For example, whereas bitcast_f2u(abs(float)) might result in a register
argument such as
(abs)g2.2<0,1,0>UD
v2: Set interfered = true and break in register_coalesce instead of
returning false.
Reviewed-by: Paul Berry <stereoytpe441@gmail.com>
Necessary to avoid combining a bitcast and a modifier into a single
operation. Otherwise if safe, the MOV should be removed by
copy-propagation or register coalescing.
With this and the next patch, there are only four changes in shader-db:
all a single extra instruction. The code does something like
mov a.w, -b.x
and copy propagation doesn't work because it only handles no-op
swizzles. Seems acceptable, given the known limitation of our copy
propagation.
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Paul Berry <stereoytpe441@gmail.com>
Currently single sample scaled blits with GL_LINEAR filter falls
back to meta path. Patch removes this limitation in BLORP engine
and implements single sample scaled blit with bilinear filter.
No piglit, gles3 regressions are observed with this patch on Ivybridge.
V2: Use "sample" message to utilize the linear filtering functionality
built in to hardware.
V3: Define a bool variable (bilinear_filter) to handle the conditions
for GL_LINEAR blits.
Signed-off-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
New function clamp_tex_coords() clamps the texture coordinates
to texture boundaries. This function will also be utilized later
for the BLORP implementation of single-sample scaled blit with
bilinear filter.
Signed-off-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
When we talk about both multi-sample and single-sample scaled blits,
rect_grid_{x1, y1} are more appropriate variable names as compared
to sample_grid_{x1, y1}. There are no functional changes in this patch.
It just prepares for the BLORP implementation of single-sample scaled
blit with bilinear filter.
Signed-off-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
This patch fixes a case of framebuffer blitting with renderbuffer
as color attachment and GL_LINEAR filter. Meta implementation of
glBlitFrambuffer() converts source color buffer to a texture and
uses it to do the scaled blitting in to destination buffer. Using
the exact source rectangle to create the texture does incorrect
linear filtering along the edges. This patch makes the changes to
extend the texture edges by one pixel in x, y directions. This
ensures correct linear filtering.
It fixes failing piglit fbo-attachments-blit-scaled-linear test.
Signed-off-by: Anuj Phogat <anuj.phogat@gmail.com>
CC: "9.2" <mesa-stable@lists.freedesktop.org>
CC: "9.1" <mesa-stable@lists.freedesktop.org>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
h264/mpeg4 remain disabled for pre-nvc0, there's some minor
bug/difference which causes the decoding to hang after some frames.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
The error code was changed from INVALID_VALUE to INVALID_OPERATION
in OpenGL 3.3. We should also generate an error when size is BGRA
and normalized is FALSE.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Sandybridge is the only platform that supports an IF instruction
with an embedded comparison. In this case, we need to emit a CMP
to go along with the SEL.
Fixes regressions in Piglit's glsl-fs-atan-3, fs-unpackHalf2x16,
fs-faceforward-float-float-float, isinf-and-isnan fs_basic, and
isinf-and-isnan fs_fbo.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=68086
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
Tested-by: lu hua <huax.lu@intel.com>
If clipdistance for one of the vertices is nan (or inf) then the
entire primitive should be discarded.
Signed-off-by: Zack Rusin <zackr@vmware.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Doing the comparisons pre-filter is highly recommended by OpenGL (and d3d9)
and definitely required by d3d10.
This actually doesn't do it pre-filter but more "in-filter" as otherwise
need to push the comparisons even further down into fetch code and this
also trivially allows using a somewhat cheaper lerp.
Doing it pre-filter would actually have some performance advantage for UNORM
formats (because the comparisons should be done in texture format, we'd only
need to convert the shadow ref coord to texture format once, but in turn would
save converting the per-sample texture values to floats) but this gets a bit
messy as this has implications for border color handling as well (which needs
to be done prior to depth comparisons, hence would also need to convert border
color to texture format too or use some other tricks like doing separate border
color / shadow ref comparison and simply using that result directly when doing
border replacement).
Should make no difference for nearest filtering, and performance for linear
filtering should be mostly the same too (essentially have one more comparison
instruction per sample, and replace the sub/mul/add lerp with a sub/and/and/add
special "lerp" which all in all shouldn't be much of a difference).
v2: get rid of old code completely
Reviewed-by: Zack Rusin <zackr@vmware.com>
E.g. the Source engine seems to always write to gl_ClipVertex, but normally
doesn't enable any GL_CLIP_DISTANCEn states. This change removes some
irrelevant parts from the generated vertex shader code in such cases.
Reviewed-by: Tom Stellard <thomas.stellard@amd.com>
This is a very well hidden bug found by accident (only the fixed glean
tstencil2 test so far seems to hit it).
We must use new mask with combined s_pass values and orig_mask values
for zpass/zfail stencil ops, otherwise both the sfail op and one of
zpass/zfail op are applied (probably not hit in most tests because
some of the ops tend to be KEEP usually).
Note: this is a candidate for the 9.2 branch.
Reviewed-by: Zack Rusin <zackr@vmware.com>
Should get rid of some float-to-int conversions (with negation).
No piglit regressions (with llvmpipe).
v2: fix bogus formatting spotted by Brian.
Reviewed-by: Brian Paul <brianp@vmware.com>
This also allows people who don't want to install the binary blobs
required for VP2 to still get MPEG decoding.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
There's no need to use a clip flag for NEGW on these gens, so
no reason we can't just enable 8 planes.
V2: - Bump (and document!) MAX_VERTS in the clip code.
- Fix clip flag masks in the clip unit state and in the shader
prolog
- Move this to the end of the series for less breakage.
Signed-off-by: Chris Forbes <chrisf@ijw.co.nz>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
This does the same thing as we do for triangle clipping -- select the
appropriate source (either dot(hpos,fixed plane) or a clipdistance
slot).
Signed-off-by: Chris Forbes <chrisf@ijw.co.nz>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
Nothing in the clipper uses gl_ClipVertex any more, so we don't care
where it is.
V2: Don't bother fishing out the clipvertex offset either.
Signed-off-by: Chris Forbes <chrisf@ijw.co.nz>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
Soon the dp4 is only going to be used for fixed clip planes.
V2: Remove old inaccurate comment about the behavior of this function;
add a better explanation above.
Signed-off-by: Chris Forbes <chrisf@ijw.co.nz>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
V2: - Use the new VS_OPCODE_UNPACK_FLAGS_SIMD4X2 to correctly split the
flags for the two vertices being processed together.
- Don't apply bogus masking of clip flags. The set of plane enables
aren't included in the shader key, and we wouldn't want the
recompiles anyway.
V3: - Tidy up spurious instructions, name temps properly.
Signed-off-by: Chris Forbes <chrisf@ijw.co.nz>
[V2] Reviewed-by: Paul Berry <stereotype441@gmail.com>
Splits the bottom 8 bits of f0.0 for further wrangling
in a SIMD4x2 program. The 4 bits corresponding to the channels in each
program flow are copied to the LSBs of dst.x visible to each flow.
This is useful for working with clipping flags in the VS.
V3: - Fixup immediate types
- Teach scheduler about the hidden dep on flags
Signed-off-by: Chris Forbes <chrisf@ijw.co.nz>
V2: Reviewed-by: Paul Berry <stereotype441@gmail.com>
We're about to have an instruction that depends on the flags but isn't
predicated. This lays the groundwork.
Signed-off-by: Chris Forbes <chrisf@ijw.co.nz>
Previously we had disabled interpolation of the clip distances as a
special case, since they were unused.
Signed-off-by: Chris Forbes <chrisf@ijw.co.nz>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
We need to produce clip flags for the vertex header on Gen4/5, so
clip plane lowering has to be done before we try to emit the flags/psiz
attribute.
Signed-off-by: Chris Forbes <chrisf@ijw.co.nz>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
V2: We don't particularly care where they fall in the VUE map, as long
as they are allocated somewhere, and occupy two contiguous slots. Don't
fiddle with the SF layout at all -- there's no need.
Signed-off-by: Chris Forbes <chrisf@ijw.co.nz>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
Fixes build error introduced with commit
d1ba1055d9.
CC nouveau_video.lo
nouveau_video.c: In function 'nouveau_screen_get_video_param':
nouveau_video.c:866:33: error: 'screen' undeclared (first use in this function)
nouveau_video.c:866:33: note: each undeclared identifier is reported only once for each function it appear
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
This makes things a bit nicer, and more importantly it fixes an issue
where a "downgraded" array texture (due to view reduced to 1 layer and
addressed with (non-array) samplec instruction) would use the wrong
coord as shadow reference value. (This could also be fixed by passing
target through the sampler interface much the same way as is done for
size queries, might do this eventually anyway.)
And if we'd ever want to support (shadow) cube map arrays, we'd need
5 coords in any case.
v2: fix bugs (texel fetch using wrong layer coord for 1d, shadow tex
using wrong shadow coord for 2d...). Plus need to project the shadow
coord, and just for fun keep projecting the layer coord too.
Reviewed-by: Zack Rusin <zackr@vmware.com>
Instead of passing s,t,r coordinates pass a coord array - the reason is that
I need to pass more coords (in particular for shadow "coord", future will also
need another one for cube map arrays) so just pass them as an array.
Also, to simplify things, use fixed location for the shadow reference value I
want to get rid of the silly "where is the right coord value" game.
Keep old-style however for aos sampling (which is not going to need shadow
coord, though for cube map arrays it still would need fixing).
(Next patch will pass those through using the new arrangement directly from
sampler interface.)
v2: fix up soa split path (unreachable currently but still...)
Reviewed-by: Zack Rusin <zackr@vmware.com>
We need to put border color into texture format color space which
essentially means clamping for non-float, normalized formats (not entirely
sure if we're also meant to quantize the float but it's probably ok not to
do it thankfully).
For OpenGL we could do this easily outside generated code due to the
1:1 sampler/texture correspondence but not for d3d10 which is terrible
(as we recalculate a constant over and over again per shader invocation).
Fortunately border color should be rare enough that we don't care THAT much.
Reviewed-by: Zack Rusin <zackr@vmware.com>
If the fragment shader is null then pixel shader invocations have
to be equal to zero. And if we're running a null ps then clipper
invocations and primitives should be equal to zero but only
if both stancil and depth testing are disabled.
Signed-off-by: Zack Rusin <zackr@vmware.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Calling the prepare outputs cleans up the slot assignments
for outputs, unfortunately aapoint and aaline didn't have
code to reset their slots after the initial setup, this
was messing up our slot assignments. The unfilled stage
was just missing the initial assignment of the face slot.
This fixes all of the reported piglit failures.
Signed-off-by: Zack Rusin <zackr@vmware.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
In commit 8fc41df (glsl: Modify ir_set_program_inouts to handle
geometry shaders), when attempting to pattern match the "foo" part of
expressions such as:
foo[i][j]
foo[i]
I incorrectly called as_dereference_variable() on the subexpression
foo[i] instead of foo. As a result, the pattern never matched, so
ir_set_program_inouts would fall back on marking the entire variable
as used, rather than just the portion indexed by the array.
This didn't result in incorrect behaviour, but it could have resulted
in inefficiency by causing the back-end to allocate resources for
unused parts of an input or output array.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
This patch adds the level query support to the video decoders
and uses some more reasonable defaults.
v2: (ck) add commit message
Reviewed-by: Christian König <christian.koenig@amd.com>
Previously we would emit a warning for empty declarations like
float;
We would also emit the same warning for things like
highp float;
However, this second case is most likely the application trying to set
the default precision. This makes the compiler generate a stronger
warning with some suggestion of a fix.
It really seems like this should be an error. I'll bet that 100% of the
time someone writes 'highp float;' the actually meant 'precision highp
float;'. Alas, both AMD and NVIDIA accept this syntax, and the spec
doesn't explicitly forbid it.
This makes piglit's precision-05.vert generate the following warnings:
0:12(11): warning: empty declaration with precision qualifier, to set the default precision, use `precision lowp float;'
0:13(12): warning: empty declaration with precision qualifier, to set the default precision, use `precision mediump int;'
v2: Add { } around a one-line if body and fix a comment. Suggested by
Ken.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Cc: "9.2" <mesa-stable@lists.freedesktop.org>
Previously, we were accidentally calling
handle_geometry_shader_input_decl() on non-input interface block
declarations, resulting in bogus error checking.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Previously, if a geometry shader input was declared as a non-array, we
would flag the proper compiler error, but then before we got a chance
to report it to the client, handle_geometry_shader_input_decl() would
assertion fail.
With this patch, handle_geometry_shader_input_decl() ignores
non-arrays.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Move the arrays to the new header brw_multisample_state.h, which will be
shared with Broadwell code.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Signed-off-by: Chad Versace <chad.versace@linux.intel.com>
Place each array in the brw namespace by renaming it:
sample_positions_4x -> brw_multisample_positions_4x
sample_positions_8x -> brw_multisample_positions_8x
This prepares for moving the arrays to a header shared by gen6 and gen8.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Signed-off-by: Chad Versace <chad.versace@linux.intel.com>
GLSL ES does not allow unsized arrays, and GLSL ES 1.00 does not allow
array initializers. However, GLSL ES 3.00 allows array initializers,
and the initializer can explicitly size the array. The specification
even includes some examples of this:
float x[] = float[2] (1.0, 2.0); // declares an array of size 2
float y[] = float[] (1.0, 2.0, 3.0); // declares an array of size 3
float a[5];
float b[] = a;
Move the unsized array check to after the initializer has been
processed. If the array is still unsized, generate the error. This
should have no effect in GLSL ES 1.00 because, as previously mentioned,
array initializers are not allowed.
Fixes piglit "glsl-es-3.00 compiler array-sized-by-initializer.vert".
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Cc: "9.1 9.2" <mesa-stable@lists.freedesktop.org>
The functional change is that now invalidate_framebuffer is called if
the texture is actually detached from one of the currently bound FBOs.
Previously this was only done for renderbuffers.
The remaining changes make the texture delete path look more similar to
the renderbuffer delete path. This includes adding relevant spec
quotations to justify the behavior.
Fixes piglit fbo-incomplete "delete texture of bound FBO" test.
v2: Move 'fb->Attachment[i].Texture == att' check from previous patch to
this patch... where it was intended to be in the first place. Noticed
by Chad.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Chad Versace <chad.versace@linux.intel.com>
Cc: "9.2" <mesa-stable@lists.freedesktop.org>
Also add a return value indicating whether any work was done.
This will be used by the next patch.
v2: Move 'fb->Attachment[i].Texture == att' check to the next
patch... where it was intended to be in the first place. Noticed by
Chad.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Chad Versace <chad.versace@linux.intel.com>
Cc: "9.2" <mesa-stable@lists.freedesktop.org>
Looks like the same issue that was seen with MULADD in trans slot on
R7xx also affects MULADD_IEEE (maybe all OP3 instructions and MULADD is
just a most frequently used?). So the workaround is to not allow affected
instructions to be placed into the trans slot.
Fixes https://bugs.freedesktop.org/show_bug.cgi?id=67927
Signed-off-by: Vadim Girlin <vadimgirlin@gmail.com>
Cc: "9.2" <mesa-stable@lists.freedesktop.org>
FSEQ/FSGE/FSLT/FSNE work just the same as SEQ/SGE/SLT/SNE except skip the
select.
And just for consistency use the same appropriate ordered/unordered comparisons
for the old opcodes as well.
Reviewed-by: Zack Rusin <zackr@vmware.com>
Also while here add a bunch of other forgotten (integer) instructions to
tgsi_util_get_inst_usage_mask() (which isn't used for much except optimizing
away unused input components), though it may still be incomplete.
Reviewed-by: Zack Rusin <zackr@vmware.com>
Newer graphic languages don't want messy float mask results but instead true
"boolean" mask results for float comparisons. Otherwise just need to convert
the floats back to integers. Need to keep the old opcodes however due to both
legacy (gl and d3d9) needing them and because older hw can't really deal with
integers. These new FSEQ/FSGE/FSLT/FSNE opcodes are part of integer API and
hence must be supported if a driver claims to support glsl 1.30 (or
PIPE_SHADER_CAP_INTEGERS).
Reviewed-by: Zack Rusin <zackr@vmware.com>
Because we must maintain an exec_mask even if there's currently nothing
on the mask stack, we can still have an exec_mask at the end of the program.
Effectively, this mask should be set back to default when returning from main.
Without relying on END/RET opcode (I think it's valid to have neither) it is
actually difficult to do this, as there doesn't seem any reasonable place to
do it, so instead let's just say the exec_mask is invalid outside main (which
it really is effectively).
The problem is that geometry shader called end_primitive outside the shader
(in the epilogue), and as a result used a bogus mask, leading to bugs if we
had to set the (somewhat misnamed) ret_in_main bit anywhere. So just avoid
the mask combining function when called from outside the shader.
Reviewed-by: Zack Rusin <zackr@vmware.com>
The code was quite weird, the second comparison was in fact a complete no-op
and we can also do the comparison with the vector directly instead of scalar,
which should not also be faster but it is way more obvious how that mask
is actually going to look like.
Reviewed-by: Zack Rusin <zackr@vmware.com>
Instead of reducing masks to 0/1 simply use the mask directly as -1.
Also use some signed comparison instead of unsigned (as far as I understand
these values have to be (very) small and signed means llvm doesn't have to
apply additional logic to do the unsigned comparisons the cpu can't do).
Saves a couple of instructions in some test geometry shader here.
v2: that was a bit to much optimization, don't skip combining the masks...
Reviewed-by: Zack Rusin <zackr@vmware.com>
CMP instructions use BRW_ARF_NULL as a destination. Prior to this
patch, dump_instruction() decoded the destination as "???".
Now it decodes BRW_ARF_NULL as "(null)" and other ARFs numerically.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
This resulted in printouts like:
246: cmp.cmod.f0.0
???, vgrf152, 0.000000f, (null),
With this patch, CMP is properly printed on one line.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Many GLSL shaders contain code of the form:
x = condition ? foo : bar
The compiler emits an ir_if tree for this, since each subexpression
might be a complex tree that could have side-effects and short-circuit
logic operations.
However, the common case is to simply pick one of two constants or
variable's values---which is exactly what SEL is for. Replacing IF/ELSE
with SEL also simplifies the control flow graph, making optimization
passes which work on basic blocks more effective.
The shader-db statistics:
total instructions in shared programs: 1655247 -> 1503234 (-9.18%)
instructions in affected programs: 949188 -> 797175 (-16.02%)
2,970 shaders were helped, none hurt. Gained 181 SIMD16 programs.
This helps Valve's Source Engine games (max -41.33%), The Cave
(max -33.33%), Serious Sam 3 (max -18.64%), Yo Frankie! (max -30.19%),
Zen Bound (max -22.22%), GStreamer (max -6.12%), and GLBenchmark 2.7
(max -1.94%).
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
The instruction
(+f0.0) SEL dst, src0, src1
will write either src0 or src1 to dst, depending on the predicate.
Unlike most predicated instructions, it always writes to dst.
fs_inst::is_partial_write() is supposed to return true if the whole
register is guaranteed to be written. The !inst->predicated check makes
sense for most instructions, which might not write the whole register,
but SEL is a special case.
This caused live interval analysis to ignore the destination of
predicated SEL instructions when computing "def" information.
Requires the previous commit to avoid regressions.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
The existing inst->is_partial_write() already disallows predicated
instructions, so this has no functional change. However, it's worth
doing explicitly since the CSE pass does not consider the flag register.
This means it could blindly factor out operations that use the same
sources, but which have different condition codes set.
This prevents a regression in the next commit.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Usually, the driver creates both 8-wide and 16-wide variants of every
fragment shader. When 16-wide compilation fails, it logs a performance
warning explaining why only an 8-wide program exists.
However, when there are pull parameters, the driver won't even bother
trying the 16-wide compile (since it would fail). In this case, it
failed to emit a performance warning, leaving no explanation for the
missing 16-wide program.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Fix ilo_gpe_init_view_surface_for_buffer to allow buffer to be NULL, and add
ilo_gpe_set_view_surface_bo to set it later. This allows us to set up
SURFACE_STATE early for constant buffers backed by user buffers.
In finalize_index_buffer(), when the current index buffer was destroyed due to
u_upload_data(), it may happen that the new index buffer is at the same
address as the old one. Comparing the pointers to the two buffers could fail
to work, and 3DSTATE_INDEX_BUFFER would be incorrectly skipped.
Holding a reference to the current index buffer before calling u_upload_data()
should fix the problem.
Makes this flag appear in the output for INTEL_DEBUG=state
Signed-off-by: Chris Forbes <chrisf@ijw.co.nz>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
INTEL_DEBUG=vue now emits a listing of each slot in the VUE map,
and the corresponding interpolation mode.
V2: Fix whitespace issues.
Signed-off-by: Chris Forbes <chrisf@ijw.co.nz>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
My previous attempt at doing so double-failed miserably (minification of
zero still gives one, and even if it would not the value was never written
anyway).
While here also rename the confusingly named int_vec bld as we have int vecs
of different sizes, and rename need_nr_mips (as this also changes out-of-bounds
behavior) to is_sviewinfo too.
Reviewed-by: Zack Rusin <zackr@vmware.com>
d3d10 has no notion of distinct array resources neither at the resource nor
sampler view level. However, shader dcl of resources certainly has, and
d3d10 expects resinfo to return the values according to that - in particular
a resource might have been a 1d texture with some array layers, then the
sampler view might have only used 1 layer so it can be accessed both as 1d
or 1d array texture (I think - the former definitely works). resinfo of a
resource decleared as array needs to return number of array layers but
non-array resource needs to return 0 (and not 1). Hence fix this by passing
the target from the shader decl to emit_size_query and use that (in case of
OpenGL the target will come from the instruction itself).
Could probably do the same for actual sampling, though it may not matter there
(as the bogus components will essentially get clamped away), possibly could
wreak havoc though if it REALLY doesn't match (which is of course an error
but still).
Reviewed-by: Zack Rusin <zackr@vmware.com>
Specifically, must return 0 for non-existent mip levels (and non-existent
textures which is an unsolved problem) for everything but total mip count.
Reviewed-by: Zack Rusin <zackr@vmware.com>
GLSL 1.50 incorporates the functionality of the
ARB_fragment_coord_conventions extension, so we need to make this
functionality available even if the extension isn't enabled.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
From section E.1 (Profiles and Deprecated Features of OpenGL 3.0)
of the OpenGL 3.0 spec:
"LineWidth is not deprecated, but values greater than 1.0
will generate an INVALID VALUE error"
From context it is clear that values greater than 1.0 should only
generate an INVALID VALUE error in a forward-compatible context.
The code was correctly quoting this spec text, but it was disallowing
all line widths in forward-compatible contexts, instead of just widths
greater than 1.0.
This patch introduces the correct check, so that setting a line width
of 1.0 or less is permitted.
Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
We can't be injecting the primitive id's in the pipeline because
by that time the primitives have already been decomposed. To
properly number the primitives we need to handle the adjacency
primitives by hand. This patch moves the prim id injection into
the original primitive assembler and completely removes the
useless pipeline stage.
Signed-off-by: Zack Rusin <zackr@vmware.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Without reseting the vertex id, with primitives where the same
vertex is used with different primitives (e.g. tri/lines strips)
our vbuf module won't re-emit those vertices with the changed
primitive id. So lets reset the vertex id whenever injecting
new primitive id to make sure that the vertex data is correctly
emitted.
Signed-off-by: Zack Rusin <zackr@vmware.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Before inserting new front face and prim id outputs cleanup
the old extra outputs, otherwise our cache will use previous
output slots which will break as soon as outputs of the current
shader don't match the last.
Signed-off-by: Zack Rusin <zackr@vmware.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
libEGL was incorrectly exporting *all* symbols, public and private.
This patch adds -fvisibility=hidden to libEGL's linker flags to ensure
that only symbols annotated with __attribute__((visibility("default")))
get exported.
Sanity-checked with libEGL's builtin DRI2 driver and the i965 DRI driver
by running Piglit on X/EGL and by running weston-gears on Weston as an
X client.
Sanity-checked with libEGL's Gallium driver (which is not built-in) and
the swrast Gallium driver by running es2gears_x11.
Kristian reviewed the symbol diff in `nm libEGL.so`.
CC: "9.2" <mesa-stable@lists.freedesktop.org>
CC: Ian Romanick <idr@freedesktop.org>
Acked-by: Kristian Høgsberg <krh@bitplanet.net>
Reviewed-by: Jakob Bornecrantz <jakob@vmware.com>
Signed-off-by: Chad Versace <chad.versace@linux.intel.com>
This is wrong both for OpenGL and d3d. (In fact clamping is a side effect
of converting to depth format, so this should really do quantization too
at least in d3d10 for the comparisons to be truly correct.)
Reviewed-by: Zack Rusin <zackr@vmware.com>
Clearly the returned values need to be per-element if the lod is per element.
Does not actually change behavior yet.
Reviewed-by: Zack Rusin <zackr@vmware.com>
This opcode is quite problematic in tgsi, while it tries to mirror
d3d10 resinfo it can't really do what's stated there due to missing
the crazy return type modifiers. Hence specify this is ignored along
with the swizzle.
(Other options would be to have multiple opcodes or specify the ret
type modifier maybe in dst_reg as there's padding bits left there but
it is the only instruction allowing this.)
Reviewed-by: Zack Rusin <zackr@vmware.com>
For d3d10 and ARB_robust_buffer_access_behavior, we are required to return
0 for out-of-bounds coordinates (for which we can just enable the code already
there was just disabled). Additionally, also need to return 0 for
out-of-bounds mip level and out-of-bounds layer. This changes the logic
so instead of clamping the level/layer, an out-of-bound mask is computed
instead in this case (actual clamping then can be omitted just like with
coordinates, since we set the fetch offset to zero if that happens anyway).
Reviewed-by: Zack Rusin <zackr@vmware.com>
While so far this only causes some harmless test failures, there's lots more
cpus with DAZ. All 64bit capable ones can do it (particularly relevant for
AMD cpus as they supported sse3 very very late) but if really necessary we
can check support for that for real with some more magic.
(In fact just about ANY cpu with sse2 can support DAZ, I believe the only
exception are first gen P4 (Willamette) and from those only early steppings
which can't do it it's almost like intel forgot to add it... - a real pity
though docs say you can't just try to set it as they will throw a GPF.)
While this was meant to address https://bugs.freedesktop.org/show_bug.cgi?id=67672
it does not fix it. Most likely the tests need fixing as I don't think
there's any guarantee about denorm handling in the reference math library
functions if the flags aren't set to standard values. Nevertheless enabling
DAZ on all cpus which can do it should be the right thing to do.
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
Should be much faster, seems to work in softpipe.
While here (also it's now disabled) fix up the pow factor - the former value
is what is in GL core it is however not actually accurate to fp32 standard
(as it is 1.0/2.4), and if someone would do all the accurate math there's no
reason to waste 8 mantissa bits or so...
v2: use real table generating function instead of just printing the values
(might take a bit longer as it does calculations on some 3+ million floats
but much more descriptive obviously).
Also fix up another inaccurate pow factor (this time in the python code) -
wondering where the couple one bit errors came from :-(.
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
Reviewed-by: Zack Rusin <zackr@vmware.com>
Since Wayland 1.2, struct wl_buffer and a few functions are deprecated.
References to wl_buffer are replaced with wl_resource and some getter
functions and calls to deprecated functions are replaced with the proper
new API. The latter changes are related to resource versioning.
Signed-off-by: Ander Conselvan de Oliveira <ander.conselvan.de.oliveira@intel.com>
A listener is added just after the interface is bound, in
registry_handle_global().
Signed-off-by: Ander Conselvan de Oliveira <ander.conselvan.de.oliveira@intel.com>
The helper provides a series of functions to easy the implementation
of the WL_bind_wayland_display extension on different platforms. But
even with the helpers there was still a bit of duplicated code between
platforms, with the drm authentication being the only part that
differs.
This patch changes the bufmgr interface to provide a self contained
object with a create function that takes a drm authentication callback
as an argument. That way all the helper functions are made static and
the "_helper" suffix was removed from the sources file name.
This change also removes the mix of Wayland client and server code in
the wayland drm platform source file. All the uses of libwayland-server
are now contained in native_wayland_drm_bufmgr.c.
Changes to the drm platform are only compile tested.
Signed-off-by: Ander Conselvan de Oliveira <ander.conselvan.de.oliveira@intel.com>
Since llvm -3.4svn r187618, TargetOptions doesn't provide
RealignStack, so only enable it with llvm<3.4
This option must now be specified using function attributes, see LLVM
commit r187618
Reviewed-by: Tom Stellard <thomas.stellard@amd.com>
This command reads a value from memory and writes it to a register (the
opposite of MI_STORE_REGISTER_MEM). It's only available on Gen7+.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
This prevents a crash in a future patch.
_mesa_initialize_context() creates a default transform feedback object
by calling the NewTransformFeedbackObject() driver hook. Eventually,
we'll want to subclass that and allocate a buffer object. This means
passing brw->bufmgr to drm_intel_alloc_bo(), and crashing if it isn't
initialized yet.
The buffer manager is actually already initialized; we just hadn't
copied the pointer from intel_screen to intel_context quite early
enough.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
Gen7+ supports four transform feedback streams. Using a function-like
macro makes it easy to access them by stream number or loop over them.
"GEN7_" prefixes are more common than "_IVB" suffixes, so use that.
Gen6 only supports a single stream, so the single #define should be
fine. However, SO_NUM_PRIM_STORAGE_NEEDED was a poor name. For one,
the word "NUM" doesn't appear in the actual name of the register.
It's also confusingly generic, as it doesn't exist on Gen7+. Add a
"GEN6_" prefix for clarity.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
Gen7+ supports four transform feedback streams. Using a function-like
macro makes it easy to access them by stream number or loop over them.
"GEN7_" prefixes are more common than "_IVB" suffixes, so we use that.
Gen6 only supports a single stream, so the single #define should be
fine. However, SO_NUM_PRIMS_WRITTEN was confusingly generic, as it
doesn't exist on Gen7+. Add a "GEN6_" prefix for clarity.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
Previously only the slice of a 3D texture was validated in the FBO
completeness check. This fixes the failure in the 'invalid layer of an
array texture' subtest of piglit's fbo-incomplete test.
v2: 1D_ARRAY textures have Depth == 1. Instead, compare against Height.
v3: Handle CUBE_MAP_ARRAY textures too. Noticed by Marek.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Cc: "9.1 9.2" mesa-stable@lists.freedesktop.org
This fixes the segfault in the 'invalid slice of 3D texture' and
'invalid layer of an array texture' subtests of piglit's fbo-incomplete
test.
The 'invalid layer of an array texture' subtest still fails.
v2: Fix off-by-one comparison error noticed by Chris Forbes. Also,
1D_ARRAY textures have Depth == 1. Instead, compare against Height.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org> [v1]
Cc: "9.1 9.2" mesa-stable@lists.freedesktop.org
Allow user-generated names for glBindFramebufferEXT on desktop GL.
Disallow its use altogether for core profiles.
Names bound with glBindFramebuffer in desktop OpenGL are still
(incorrectly) shared across the share group instead of being
per-context. This gets us a bit closer to being strictly conformant.
v2: Disallow glBindFramebufferEXT in 3.1 by not installing it in the
dispatch table. Suggested by Jordan.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org> [v1]
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com> [v1]
Cc: mesa-stable@lists.freedesktop.org
Allow user-generated names for glBindRenderbufferEXT on desktop GL.
Disallow its use altogether for core profiles.
v2: Disallow glBindRenderbufferEXT in 3.1 by not installing it in the
dispatch table. Suggested by Jordan.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org> [v1]
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com> [v1]
Cc: mesa-stable@lists.freedesktop.org
These are only used on Gen4-5. Why waste the 8kB of space?
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Eric Anholt <eric@anholt.net>
If gs is null, then freeing state->shader.tokens would result in a null
dereference.
Fixes "Dereference after null check" defect reported by Coverity.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
Fixes "Uninitialized pointer read" defect reported by Coverity.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
unfilled_stage::face_slot is of type int.
Fixes "Unsigned compared against 0" defect reported by Coverity.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
pp_free_bos dereferences ppq without a null check.
Fixes "Dereference before null check" defect reported by Coverity.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
the issue is that stream output is run before the pipeline, which
means that unless we decompose the primitives before the so
then things crash. we could convert the entire stream output
code into a pipeline stage but it will take a bit, so for now
fix the crashes by simply re-adding the old input assembler
which is run before the SO.
Signed-off-by: Zack Rusin <zackr@vmware.com>
we used to have a face primitive assembler that we ran after if
the gs was missing but we had adjacency primitives in the pipeline,
lets convert it to a pipeline stage, which allows us to use it
to inject outputs (primitive id) into the vertices. it's also
a lot cleaner because the decomposition is already handled for us.
Signed-off-by: Zack Rusin <zackr@vmware.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Inject front face only if the fragment shader uses it and
propagate through all channels because otherwise we'll
need to figure out the exact swizzle that the fs expects and
it's just simpler to make sure all the components within
the front face register are correctly set.
Signed-off-by: Zack Rusin <zackr@vmware.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Replace "fulldecl->Semantic.Name/Index" with semName/semIndex.
Simplify if/else logic for TGSI_FILE_OUTPUT code.
Remove old comment.
Fix indentation.
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
Previously we would mark a renderbuffer as needing a depth resolve.
But, to support layered rendering, we need to look at the attachment
instead, since the attachment knows if layered rendering is being
used.
Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
Reviewed-by: Chad Versace <chad.versace@linux.intel.com>
This function is needed to support layered rendering. With
layered rendering, the attachment stores the state of whether
layered rendering is being used.
Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
Reviewed-by: Chad Versace <chad.versace@linux.intel.com>
When layered rendering is being used, we should not set
FORCE_ZERO_RTAINDEX in the clip state to allow render target
array values other than zero to be used.
Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
This restriction was related to programming the offset fields
of the depth buffer packet. We are now setting these offsets
to 0 now, so this restriction should no longer be required.
Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
Previously we would always find the 2D sub-surface of interest,
and then program the surface to this location. Now we always
program the 3DSTATE_DEPTH_BUFFER at the start of the surface.
To select the lod/slice, we utilize the lod & minimum array
element fields.
As part of this change, we must revert 1f112ccf:
Revert "i965/gen7: Align all depth miplevels to 8 in the X direction."
We also must disable brw_workaround_depthstencil_alignment for
gen >= 7. Now the hardware will handle alignment when rendering
to additional slices/LODs.
v2:
* Merge with recent MOCS changes
Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
For gen >= 7, we will use the lod/minimum-array-element fields to
support layered rendering. This means that we must restrict
the depth & stencil attachments to match in various more retrictive
ways. (Now the width, height, depth, LOD and layer must match)
The reason width, height, and depth must match is that the hardware
has a single set of width, height, and depth settings (in
3DSTATE_DEPTH_BUFFER) that affect both the depth and stencil buffers.
Since these controls determine the miptree layout, they need to be
set correctly in order for lod and minimum-array-element to work
properly. So the only way rendering can work is if the width,
height, and depth match.
In the future, if this restriction proves to be a problem (say
because some crucial client application relies on rendering to
different levels/layers of stencil and depth buffers), then we can
always work around the restriction by copying depth and/or stencil
data to a temporary buffer prior to rendering (much in the same way
that brw_workaround_depthstencil_alignment() does today for
gen < 7), but hopefully that won't be necessary.
Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
When performing hiz ops, we must ensure that the region sizes
have an 8 aligned width and 4 aligned height. We can tweak the
size for blorp hiz operations at LOD 0, but for the others we
can't. Therefore, we disable hiz for these miplevels if they
don't meet the size alignment requirements.
Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
This will be used in 3DSTATE_DEPTH_BUFFER in a later patch.
Note: Cube maps are treated as 2D arrays with 6 times as
many array elements as the cube map array would have.
Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
This will be used in 3DSTATE_DEPTH_BUFFER in a later patch.
Note: Cube maps are treated as 2D arrays with 6 times as
many array elements as the cube map array would have.
Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
In a future pass this will allow us to exit-early from this
routine to disable it for gen >= 7.
Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
Some videos specify mb_adaptive_frame_field_flag instead of
field_pic_flag. This implies that the pic height needs to be halved, and
this field needs to be passed to the VP engine.
Cc: "9.2" mesa-stable@lists.freedesktop.org
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
The loop was iterating over all the fs inputs and setting them
to perspective interpolation, then after the loop we were
creating extra output slots with the correct interpolation. Instead
of injecting bogus extra outputs, just set the interpolation
on front face and prim id correctly when doing the initial scan
of fs inputs.
Signed-off-by: Zack Rusin <zackr@vmware.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
clipping would drop the extra outputs because it always
used the number of standard vertex shader outputs, without
geometry shader or extra outputs. The commit makes sure
that clipping with geometry shaders which have more outputs
than the current vertex shader and with extra outputs correctly
propagates the entire vertex.
Signed-off-by: Zack Rusin <zackr@vmware.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
Draw module can decompose primitives into wireframe models, which
is a fancy word for 'lines', unfortunately that decomposition means
that we weren't able to preserve the original front-face info which
could be derived from the original primitives (lines don't have a
'face'). To fix it allow draw module to inject a fake face semantic
into outputs from which the backends can figure out the original
frontfacing info of the primitives.
Signed-off-by: Zack Rusin <zackr@vmware.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
Draw sometimes injects extra shader outputs (aa points, lines or
front face), unfortunately most of the pipeline and llvm code
didn't handle them at all. It only worked if number of inputs
happened to be bigger or equal to the number of shader outputs
plus the extra injected outputs. In particular when running
the pipeline which depends on the vertex_id in the vertex_header
things were completely broken. The patch adjust the code to
correctly use the total number of shader outputs (the standard
ones plus the injected ones) to make it all stop crashing and
work.
Signed-off-by: Zack Rusin <zackr@vmware.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Instead of using the magical 4 use the above computed
vertex size. Doesn't change the behavior, just makes the code
a bit cleaner.
Signed-off-by: Zack Rusin <zackr@vmware.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
when dumping shader outputs it's nice to have the integer
values of the outputs, in particular because some values
are integers.
Signed-off-by: Zack Rusin <zackr@vmware.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
Adding code to detect the usage of prim id and front face
semantics in fragment shaders.
Signed-off-by: Zack Rusin <zackr@vmware.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
we forgot to add ucmp to the list of opcodes, so it was never
generated for ureg.
Signed-off-by: Zack Rusin <zackr@vmware.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
The spec says that front-face is true if the value is >0 and false
if it's <0. To make sure that we follow the spec, lets just
subtract 0.5 from our value (llvmpipe did 1 for frontface and 0
otherwise), which will get us a positive num for frontface and
negative for backface.
Signed-off-by: Zack Rusin <zackr@vmware.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
We'll need proper values for max_gs_threads when we eventually support
geometry shaders. Also, we initialize it for every other platform.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
Commit 2548092ad8 switched the sense of interpolation qualifier
checks in order to permit them on geometry shader in/out variables.
In doing so, it accidentally allowed interpolation qualifiers to be
applied to ordinary variables and function parameters.
Fixes a regression in Piglit's local-smooth-01.frag.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
Commit 7cfefe6965 introduced a check for whether linked->Type equals
GL_GEOMETRY_SHADER. However, linked may be NULL due to an earlier error
condition.
Since the entire function after the error path is (or should be) guarded
by linked != NULL checks, we may as well just return early and remove
the checks.
Fixes crashes in 9 Piglit tests.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
v2:
- upon success close the given file descriptors
v3:
- use specific entry for dma buffers instead of the basic for
primes, and enable the extension based on the availability
of the hook
v4 (Chad):
- use ARRAY_SIZE
- improve the comment about the number of file descriptors
- in case of invalid format report EGL_BAD_ATTRIBUTE instead
of EGL_BAD_MATCH
- take into account specific error set by the driver.
v5:
- fix error handling
v6 (Chad):
- fix invalid plane count checking
v7 (Chad):
- fix indentation and reset loop counter before checking
for excess attributes
Signed-off-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
Reviewed-by: Chad Versace <chad.versace@linux.intel.com>
Memory originating outside mesa stack is meant to be for reading
only. In addition, the restrictions imposed by the image external
extension should apply. For example, users shouldn't be allowed
to generare mip-trees based on these images.
v2 (Chad): document using full extension names, fix the comment
style itself and emit description of error
Signed-off-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
Reviewed-by: Chad Versace <chad.versace@linux.intel.com>
As specified in:
http://www.khronos.org/registry/egl/extensions/EXT/EGL_EXT_image_dma_buf_import.txt
Checking for the valid fourcc values is left for drivers avoiding
dependency to drm header files here.
v2: enforce EGL_NO_CONTEXT
v3: declare the extension as EGL (not GLES)
v4: do not update eglext.h manually but rely on update from
Khronos instead
v5: (Eric) report invalid context as EGL_BAD_PARAMETER instead of as
EGL_BAD_CONTEXT
v6: (Chad) fix the checking for valid hints. Before all values were
rejected.
v7: (Chad) comment style change from
/**
* Multi-
* line
into
/* Multi-
* line
Signed-off-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
Reviewed-by: Chad Versace <chad.versace@linux.intel.com>
v2: do not break ABI, but instead introduce new entry point for
dma buffers and bump up the dri-interface version to eight
v3 (Chad): allow the hook to specify an error originating from the
driver. For now only unsupported format is considered.
I thought about rejecting the hints also as they are
addressing only YUV sampling which is not supported at
the moment but then thought against it as the spec is
not saying one way or the other.
v4 (Eric, Chad): restrict to rgb formatted only
v5: rebased on top of i915/i965 split
v6 (Chad): document using full extension name
Signed-off-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
Reviewed-by: Chad Versace <chad.versace@linux.intel.com>
Geometry shader support in the Mesa front end is still fairly
preliminary. Many features are untested, and the following things are
known not to work:
- The gl_in interface block
- The gl_ClipDistance input
- Transform feedback of geometry shader outputs
- Constants that are new in GLSL 1.50 (e.g. gl_MaxGeometryInputComponents)
This isn't a problem, since no back-end drivers currently enable
geometry shaders. However, to make sure no one gets the wrong
impression, emit a nasty warning to let the user know that geometry
shader support isn't complete.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Section 4.3.8.1 (Input Layout Qualifiers) of the GLSL 1.50 spec
contains some tricky rules for how the sizes of geometry shader input
arrays are related to the input layout specification. In essence,
those rules boil down to the following:
- If an input array declaration does not specify a size, and it
follows an input layout declaration, it is sized according to the
input layout.
- If an input layout declaration follows an input array declaration
that didn't specify a size, the input array declaration is given a
size at the time the input layout declaration appears.
- All input layout declarations and input array sizes must ultimately
match. Inconsistencies are reported as soon as they are detected,
at compile time if the inconsistency is within one compilation unit,
otherwise at link time.
- At least one compilation unit must contain an input layout
declaration.
(Note: the geom_array_resize_visitor class was contributed by Bryan
Cain <bryancain3@gmail.com>.)
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
From the GLSL ES 3.00 spec:
"All indexes used to index a uniform block array must be constant
integral expressions."
Similar text exists in GLSL specs since 1.50.
When we implemented this, the only type of interface block supported
by Mesa was uniform blocks, so we required all indexes used to index
any interface block to be constant integral expressions.
Now that we are adding interface block support for GLSL 1.50, we need
a more specific check.
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
This gets piglit's geometry-basic test running.
TODO: Still need to validate that the GS layout qualifiers don't get used
in places they shouldn't (like an interface block, or a particular shader
input or output)
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Next step is to validate them at link time.
v2 (Paul Berry <stereotype441@gmail.com>): Don't attempt to export the
layout qualifiers in the event of a compile error, since some of them
are set up by ast_to_hir(), and ast_to_hir() isn't guaranteed to have
run in the event of a compile error.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
v3 (Paul Berry <stereotype441@gmail.com>): Use PRIM_UNKNOWN to
represent "not set in this shader".
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Limited semantic checking (compatibility between declarations, checking
that they're in the right shader target, etc.) is done.
v2: Remove stray debug printfs.
v3 (Paul Berry <stereotype441@gmail.com>): Process input layout
qualifiers at ast_to_hir time rather than at parse time, since certain
error conditions depend on the relative ordering between input layout
qualifiers, declarations, and calls to .length().
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
We do some tests of qualifiers using a union containing an int and the
struct full of bitfields, so make sure the bitfields don't spill
outside the int.
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
From section 2.15 (Geometry Shaders) the OpenGL 3.2 spec:
A program object that includes a geometry shader must also include
a vertex shader; otherwise a link error will occur.
Reviewed-by: Paul Berry <stereotype441@gmail.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
ARB_geometry_shader4 spec Errors:
"The error INVALID_VALUE is generated by ProgramParameteriARB if <pname>
is GEOMETRY_VERTICES_OUT_ARB and <value> is negative."
Reviewed-by: Paul Berry <stereotype441@gmail.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
In geometry shaders, outputs are consumed at the time of a call to
EmitVertex() (as opposed to all other shader types, where outputs are
consumed when the shader exits). Therefore, when packing geometry
shader output varyings using lower_packed_varyings, we need to do the
packing at the time of the EmitVertex() call.
This patch accomplishes that by adding a new visitor class,
lower_packed_varyings_gs_splicer, which is responsible for splicing
the varying packing code into place wherever EmitVertex() is found.
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
This patch modifies lower_packed_varyings to store the packing code it
generates in a temporary exec_list, and then splice that list into the
shader's main() function when it's done. This paves the way for
supporting geometry shader outputs, where we'll have to splice a clone
of the packing code before every call to EmitVertex().
As a side benefit, varying packing code is now emitted in the same
order for inputs and outputs; this should make debug output a little
easier to read.
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Since geometry shader inputs are arrays (where the array index
indicates which vertex is being examined), varying packing needs to
treat them differently.
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
From section 4.3.4 (Inputs) of the GLSL 1.50 spec:
Geometry shader input variables get the per-vertex values written
out by vertex shader output variables of the same names. Since a
geometry shader operates on a set of vertices, each input varying
variable (or input block, see interface blocks below) needs to be
declared as an array.
Therefore, the element type of each geometry shader input array should
match the type of the corresponding vertex shader output.
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
The documentation for gl_shader_program.Geom and gl_geometry_program
says that the former is copied to the latter at link time, but this
wasn't happening. This patch causes _mesa_ir_link_shader() to perform
the copy, and updates comment accordingly.
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
This patch creates a single function to copy the the UsesClipDistance
flag from gl_shader_program.Vert to gl_vertex_program. Previously
this logic was duplicated in the i965-specific function
brw_link_shader() and the core mesa function _mesa_ir_link_shader().
This logic will have to be expanded to support geometry shaders, and I
don't want to have to update it in two separate places.
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
This commit adds all of the parsing and semantics for GLSL 150 style
geometry shaders.
v2 (Paul Berry <stereotype441@gmail.com>): Add a few missing calls to
get_pipeline_stage(). Fix some signed/unsigned comparison warnings.
Fix handling of NULL consumer in assign_varying_locations().
v3 (Bryan Cain <bryancain3@gmail.com>): fix indexing order of 2D
arrays. Also, allow interpolation qualifiers in geometry shaders.
v4 (Paul Berry <stereotype441@gmail.com>): Eliminate
get_pipeline_stage()--it is no longer needed thanks to 030ca23 (mesa:
renumber shader indices according to their placement in pipeline).
Remove 2D stuff. Move vertices_per_prim() to ir.h, so that it will be
accessible from outside the linker. Remove
inject_num_vertices_visitor. Rework for GLSL 1.50.
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
v5 (Paul Berry <stereotype441@gmail.com>): Split out
do_set_program_inouts() argument refactoring to a separate patch.
Move geom_array_resizing_visitor to later in the series.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
There's no reason to be clever about this. By making separate
allocations for vertex and fragment shaders, we'll allow geometry
shaders to be added without introducing any complication.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
v2 (Paul Berry <stereotype441@gmail.com>): Account for rework of
builtin_variables.cpp. Use INTERP_QUALIFIER_FLAT for gl_PrimitiveID
so that it will obey provoking vertex conventions. Convert to GLSL
1.50 style geometry shaders.
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
v3 (Paul Berry <stereotype441@gmail.com>): Be less obscure about
setting interpolation field of gl_Primitive variables.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
These correspond to the EmitVertex and EndPrimitive functions in GLSL.
v2 (Paul Berry <stereotype441@gmail.com>): Add stub implementations of
new pure visitor functions to i965's vec4_visitor and fs_visitor
classes.
v3 (Paul Berry <stereotype441@gmail.com>): Rename classes to be more
consistent with the names used in the GL spec.
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Previously, we assumed that the only way Mesa would expose geometry
shader support was via the ARB_geometry_shader4 extension. But this
extension has some extra complications over GL 3.2 (interactions with
compatibility-only features, and link-time initialization of the
constant gl_VerticesIn). So we want to allow for the possibility of
supporting GL 3.2 (with GLSL 1.50 style geometry shaders) even if
ctx->Extensions.ARB_geometry_shader4 is false.
This patch adds a new function, _mesa_has_geometry_shaders(), which
returns true if either ARB_geometry_shader4 is supported or the GL
version is at least 3.2 desktop. Since compute_version() only enables
GL 3.2 functionality when GLSL 1.50 support is present, a sufficient
way for a back-end to advertise geometry shader support is to set
ctx->Const.GLSLVersion >= 150.
v2: Remove unnecessary ctx->Const.GeometryShaders150 constant.
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
We can't just use a ".glsl" file since the Lod variants are only
available in vertex and geometry shaders, while the bias variants are
only available in the fragment shader.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Commit 586b4b5 (glsl: Also update implicit sizes of varyings at link
time) extended update_array_sizes() to apply to both uniforms and
shader ins/outs. However, doing creates problems for geometry
shaders, because update_array_sizes() assumes that variables with
matching names in different parts of the pipeline should have the same
sizes. With the addition of geometry shaders, this is no longer true
(e.g. both vertex and geometry shaders have a gl_ClipDistance output
variable, but there's no reason these variables should have the same
sizes).
The original reason for commit 586b4b5 (avoid problems with
gl_TexCoord being 0 length) has since been addressed by commit 6f53921
(linker: Ensure that unsized arrays have a size after linking). So go
ahead and switch update_array_sizes() back to only acting on uniforms.
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
According to GLSL, indexing into an array or matrix with an
out-of-range constant results in a compile error. However, indexing
with an out-of-range value that isn't constant merely results in
undefined results.
Since optimization passes (e.g. loop unrolling) can convert
non-constant array indices into constant array indices, it's possible
that ir_set_program_inouts will encounter a constant array index that
is out of range; if this happens, just mark the whole array as used.
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
The code in ir_set_program_inouts that marks just a portion of a
variable as used (rather than the whole variable) only works on a few
kinds of indexing operations:
- Indexing into matrices
- Indexing into arrays of matrices, vectors, or scalars.
Fortunately these are the only kinds of indexing operations that we
expect to see; everything else is either handled by a
previously-executed lowering pass or prohibited by GLSL.
However, that could conceivably change in the future (the GLSL rules
might change, or we might modify the lowering passes). To avoid
mysterious bugs in the future, let's have ir_set_program_inouts report
an assertion failure if it ever encounters an unexpected kind of
indexing operation (and in release builds, fall back to just marking
the whole variable as used).
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
This patch extracts the functions mark_whole_variable() and
try_mark_partial_variable() from the ir_set_program_inouts visitor
functions. This will make the code easier to follow when we add
geometry shader support.
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Our previous justification for leaving this function out of glsl_type
was that it implemented counting rules that were specific to GLSL
1.50. However, these counting rules also describe the number of
varying slots that Mesa will assign to a varying in the absence of
varying packing. That's useful to be able to compute from outside of
the linker code (a future patch will use it from
ir_set_program_inouts.cpp). So go ahead and move it to glsl_type.
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
This will allow us to add geometry shader support without having to
add another boolean argument.
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
llvm shifts are undefined for shift counts exceeding (or matching) bit width,
so need to apply a mask for the tgsi shift instructions.
v2: only use mask for the tgsi shift instructions, not for the build shift
helpers. None of the internal callers need this behavior, and while llvm can
optimize away the masking for constants there are legitimate cases where it
might not be able to do so even if we know that shift count must be smaller
than type width (currently all such callers do not use the build shift
helpers).
Reviewed-by: Zack Rusin <zackr@vmware.com>
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
c shifts are undefined for shift counts exceeding (or matching) bit width,
so need to apply a mask (on x86 it actually would usually probably work as
shifts do masking on int domain shifts - unless some auto-vectorizer would
come along at last as simd domain does not mask the shift count).
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
Previously, nothing was said what happens with shift counts exceeding
bit width of the values to shift. In theory 3 behaviors are possible:
1) undefined (classic c definition)
2) just shift out all bits (so result is zero, or -1 potentially for ashr)
3) mask the shift count to bit width - 1
API's either require 3) or are ok with 1). In particular, GLSL (as well as a
couple uninteresting legacy GL extensions) is happy with undefined, whereas
both OpenCL and d3d10 require 3). Consequently, most hw also implements 3).
So, for simplicity we just specify that 3) is required rather than saying
undefined and then needing state trackers to work around it.
Also while here specify shift count as a vector, not scalar. As far as I
can tell this was a doc bug, neither state trackers nor drivers used scalar
shift count.
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
This hasn't done anything in a long time, and it's only used in a couple
places...which means we couldn't use it without doing a bunch of work
anyway.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Previously, if TEXTURE_IMMUTABLE_FORMAT was TRUE, the levels were allowed to
be set like usual, but ARB_texture_storage states:
> if TEXTURE_IMMUTABLE_FORMAT is TRUE, then level_base is clamped to the range
> [0, <levels> - 1] and level_max is then clamped to the range [level_base,
> <levels> - 1], where <levels> is the parameter passed the call to
> TexStorage* for the texture object
Reviewed-by: Matt Turner <mattst88@gmail.com>
Signed-off-by: Corey Richardson <corey@octayn.net>
This patch ensures that integers will pass through unscathed. Doing
(useless) computations on them is risky, especially when their bit
patterns correspond to values like inf or nan.
[V1-2]: Signed-off-by: Olivier Galibert <galibert at pobox.com>
Signed-off-by: Chris Forbes <chrisf@ijw.co.nz>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Adds support for interpolating noperspective varyings linearly in screen
space when clipping.
Based on Olivier Galibert's patch from last year:
http://lists.freedesktop.org/archives/mesa-dev/2012-July/024341.html
At this point all -fixed and -vertex interpolation tests work.
V5: Add brw_clip_compile.has_noperspective_shading rather than another
key flag.
V6: Real bools.
[V1-2]: Signed-off-by: Olivier Galibert <galibert at pobox.com>
Signed-off-by: Chris Forbes <chrisf@ijw.co.nz>
Acked-by: Paul Berry <stereotype441@gmail.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Previously we only gave special treatment to the builtin color varyings.
This patch adds support for arbitrary flat-shaded varyings, which is
required for GLSL 1.30.
Based on Olivier Galibert's patch from last year:
http://lists.freedesktop.org/archives/mesa-dev/2012-July/024340.html
V5: Move key.do_flat_shading to brw_clip_compile.has_flat_shading
V6: Real bools.
[V1-2]: Signed-off-by: Olivier Galibert <galibert at pobox.com>
Signed-off-by: Chris Forbes <chrisf@ijw.co.nz>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Previously the SF only handled the builtin color varying specially.
This patch generalizes that support to cover user-defined varyings,
driven by the interpolation mode array set up alongside the VUE map.
Based on the following patches from Olivier Galibert:
- http://lists.freedesktop.org/archives/mesa-dev/2012-July/024335.html
- http://lists.freedesktop.org/archives/mesa-dev/2012-July/024339.html
With this patch, all the GLSL 1.3 interpolation tests that do not clip
(spec/glsl-1.30/execution/interpolation/*-none.shader_test) pass.
V5: Move key.do_flat_shading to brw_sf_compile.has_flat_shading; drop
vestigial hunks.
V6: Real bools.
Signed-off-by: Chris Forbes <chrisf@ijw.co.nz>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
The interpolation map (in brw->interpolation_mode) is a new auxiliary
structure alongside the post-GS VUE map, which describes the
interpolation modes for each VUE slot, for use by the clip and SF
stages.
This patch introduces a new state atom to compute the interpolation map,
and adjusts the program keys for the clip and SF stages, but it is not
actually used yet.
[V1-2]: Signed-off-by: Olivier Galibert <galibert at pobox.com>
V3: Updated for vue_map changes, intel -> brw merge, etc. (Chris Forbes)
V4: Compute interpolation map as a new state atom rather than tacking it
on the front of the clip setup
V5: Rework commit message, make interpolation_mode_map a struct.
Signed-off-by: Chris Forbes <chrisf@ijw.co.nz>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
We recently proposed a new syntax for stable-patch nominations such as:
CC: "9.2 and 9.1" <mesa-stable@lists.freedesktop.org>
and this has already appeared in the wild.
So we extend the regular expression to pick this up as well.
MP_TEMP_SIZE must be aligned to 0x8000, while TEMP_SIZE on NVE4_3D
must be aligned to 0x20000, so perform both alignments to be sure
we allocate enough space (actually the bo will most likely use 128
KiB pages and not aligning to that would be a waste anyway).
Cc: "9.2" mesa-stable@lists.freedesktop.org
YYLEX_PARAM is no longer supported as of Bison 3.0. Instead, the Bison
developers recommend using %lex-param.
%lex-param takes a type and variable name, similar to %parse-param,
so you can't pass an arbitrary expression like state->scanner. But Flex
insists on passing the actual scanner object, not an arbitrary object
like state.
To solve this, the parser defines a wrapper lex() function which accepts
"state," and calls Flex's lex() function with state->scanner.
Fixes the build with Bison 3.0. Also works with Bison 2.7.1.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=67354
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Tested-by: Laurent Carlier <lordheavym@gmail.com>
Cc: "9.2" mesa-stable@lists.freedesktop.org
YYLEX_PARAM is no longer supported as of Bison 3.0. Instead, the Bison
developers recommend using %lex-param.
%lex-param takes a type and variable name, similar to %parse-param,
so you can't pass an arbitrary expression like state->scanner. But Flex
insists on passing the actual scanner object, not an arbitrary object
like state.
To solve this, the parser defines a wrapper lex() function which accepts
"state," and calls Flex's lex() function with state->scanner.
Fixes the build with Bison 3.0. Also works with Bison 2.7.1.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=67354
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Tested-by: Laurent Carlier <lordheavym@gmail.com>
Cc: "9.2" mesa-stable@lists.freedesktop.org
I had removed it in commit 1e7776ca2b
because it was obviously wrong -- why do we care whether the server is a
version that emits events, if we're not watching for the server's events,
anyway? And why would you only invalidate on a server that emits
invalidate events, when the comment said to emit invalidates if the server
*doesn't*? Only, I missed that we otherwise don't flag that our buffers
might have changed at swap time at all, so the driver was only checking
for new buffers when triggered by the Viewport hack. Of course you don't
expect Viewport to be called after a swap.
So, this is effectively a revert of the previous commit, except that I
dropped the check for only emitting invalidates on a new server -- we
*always* need to invalidate if we're doing a SwapBuffers.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=63435
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Cc: "9.1 and 9.2" <mesa-stable@lists.freedesktop.org>
Previously we were using truncation, which gives the correct result
only for numbers in [0.5-1.0] range (because there's no mantissa bits
to do any rounding there).
This is frequently hit (and probably only used there) when converting
fragment depth to depth format (d24s8 etc.) or otherwise dealing with
depth format.
v2: as spotted by Jose, get rid of extra type (src_type is already unsigned).
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
The code that checks if some texture target is valid for
glGetTexLevelParameter*() was not programmed to check for multisampling
proxy textures. This made it impossible(?) to use the proxy textures
for their intended purpose as glGetTexLevelParameter*() would just fail
on you.
Reviewed-by: Brian Paul <brianp@vmware.com>
Cc: mesa-stable@lists.freedesktop.org
glTexStorage*() functions make textures immutable. This carries on to
proxy textures. Error checking in texture storage functions prevents
proxy textures from working after first time because internally, they
became immutable.
This commit makes the error checking ignore the immutability flag when
working with proxy textures.
Reviewed-by: Brian Paul <brianp@vmware.com>
Cc: mesa-stable@lists.freedesktop.org
When working with the glTexStorage*() functions, the error checking
checks that a non-default (i.e., non-zero) texture is currently bound.
However, this check made glTexStorage*() functions fail with proxy
textures when the default texture is bound. Proxy textures do not care
about the current texture bindings so for them this check should not
be done.
Reviewed-by: Brian Paul <brianp@vmware.com>
Cc: mesa-stable@lists.freedesktop.org
The function _mesa_get_tex_max_num_levels() is supposed to calculate
the number of mipmap levels but it was not written to handle proxy
textures, at best returning a maximum of 1 mipmap level. Because of
this, at least glTexStorage*() calls would incorrectly fail when used
with proxy textures with more than one mipmap level.
Reviewed-by: Brian Paul <brianp@vmware.com>
Cc: mesa-stable@lists.freedesktop.org
Free all our temporary buffers in one place at the end of the
function. Fixes memory leak detected by Coverity.
Note: This is a candidate for the 9.x branches
Cc: mesa-stable@lists.freedesktop.org
Reviewed-by: José Fonseca <jfonseca@vmware.com>
The variable 'usage' was being used for two different things.
Sometimes for PIPE_USAGE_x and other times for PIPE_TRANSFER_x.
This renames usage to access when we're talking about PIPE_TRANSFER_x
flags. Plus, add a bunch of comments to remind us what's going on.
Also, use unsigned for PIPE_TRANSFER_x bitmask to be consistent with
other places. And add a missing const qualifier.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
We should probably be using map()/unmap() when accessing resource
data, but this is a little better.
v2: assert that the resource is not a display target, per Jose.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
This was never a problem since the Mesa state tracker always gives
us a user-space constant buffer with buffer_offset=0. But if another
state tracker ever gave us a "HW" constant buffer with non-zero
buffer_offset we'd mis-render.
Also, use the correct buffer size. And move an assertion to the
top of the function.
Reviewed-by: Marek Olšák <maraeo@gmail.com>
Reviewed-by: José Fonseca <jfonseca@vmware.com>
The cap means _can_ accept user-space constant buffers; it doesn't
mean _only_ accepts user-space constant buffers.
v2: also update the PIPE_CAP_USER_VERTEX_BUFFERS and
PIPE_CAP_USER_INDEX_BUFFERS descriptions as well. Per Jose.
Reviewed-by: Marek Olšák <maraeo@gmail.com>
Reviewed-by: José Fonseca <jfonseca@vmware.com>
vblank_mode is read by dri_util.c and falls under the "dri2" driver name,
which is not connected to the actual Mesa/Gallium driver in any way.
Reviewed-by: Brian Paul <brianp@vmware.com>
PP saves current states to cso_context and then util_blit_pixels does
the same. cso_context doesn't like that and the original state is not
correctly restored.
Cc: mesa-stable@lists.freedesktop.org
Reviewed-by: Brian Paul <brianp@vmware.com>
Surprisingly all drivers supporting MSAA can already do this (r300g and r600g
for sure) and I think Christoph wanted to have this feature for his Nouveau
drivers anyway.
We recently adopted a new convention that patches can be nominated for the
stable branch by including a line in the commit message as follows:
CC: mesa-stable@lists.freedesktop.org
This is a convenient syntax as "git send-email" will notice this line and
automatically copy the resulting patch email to the mesa-stable mailing list.
Here we extend the regular expression in the get-pick-list.sh script to also
notice this pattern, (as well as the traditional "NOTE: This patch is a
candidate..." form.
The linker_error() function sets prog->LinkStatus to false. There's
no reason for the caller of linker_error() to also do so.
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
We're now emitting this error from a point where we have easy access
to the name of the block that failed to match, so go ahead and include
that in the error message, as we do for intrastage interface block
mismatches.
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
This patch changes link_shaders() so that it sets prog->LinkStatus to
true when it starts, and then relies on linker_error() to set it to
false if a link failure occurs.
Previously, link_shaders() would set prog->LinkStatus to true halfway
through its execution; as a result, linker functions that executed
during the first half of link_shaders() would have to do their own
success/failure tracking; if they didn't, then calling linker_error()
would add an error message to the log, but not cause the link to fail.
Since it wasn't always obvious from looking at a linker function
whether it was called before or after link_shaders() set
prog->LinkStatus to true, this carried a high risk of bugs.
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Previously we failed to link (which is correct), but we did not output
an error message, which could have been confusing for users.
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
A comment in link_intrastage_shaders(), and an if-test that followed
it, seemed to indicate that link_uniform_blocks() would return a
negative value in the event of an error. But this is not the
case--all error checking has already been performed by
validate_intrastage_interface_blocks(), and link_uniform_blocks() can
only return unsigned values.
So get rid of the if-test and change the return type of
link_intrastage_shaders() to clarify that it can only return unsigned
values.
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
To have non-static buffers in local memory, it is necessary to pass them
as arguments to the kernel.
For r600, the correct lds size must be set to the SQ_LDS_ALLOC register.
The correct size is the clover size plus the size reported by the
compiler.
Reviewed-by: Tom Stellard <thomas.stellard@amd.com>
Here is an updated patch with no line wrapping and respecting 80-column limit (for my changes).
v2: Tom Stellard
- Create global arguments for constant buffers so we don't break
r600g.
Reviewed-by: Tom Stellard <thomas.stellard@amd.com>
For GLSL programs, enabledTargets can have more than one bit set. For
example, a shader that uses sampler2D and samplerCube uniforms will have
both TEXTURE_2D_BIT and TEXTURE_CUBE_BIT set.
The code that sets _ReallyEnabled already handles this, selecting the
"highest priority" texture target. We should simply use that.
Fixes new Piglit test incomplete-textures-of-multiple-types.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=62698
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
Rather than having to keep track of all the build systems and their respecitve
definition of the mesa version, use a single top file VERSION. Every build
system is responsible for reading/parsing the file and using it
v2:
* remove useless bulletpoint from the documentation, suggested by Matt
* "Androing is Linux. Use '/' in stead of '\'", spotted by Chad V
* use cleaner code to get the version in scons, suggested by Chad V
v3:
* ensure leading and trailing whitespace characters are stripped while parsing
* android: handle GNU shell commands approapriately
Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Chad Versace <chad.versace@linux.intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
We already skip this for API_OPENGL_CORE; ES2+ is very similar.
The primary user of the swrast context is GL_SELECT and GL_FEEDBACK,
which have never existed in ES.
This saves approximately 18MB of memory in GLBenchmark 2.7 Egypt (ES2).
No regressions in es3conform on Ivybridge.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Chad Versace <chad.versace@linux.intel.com>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
Certain extensions only add functionality to particular shader stages.
(For example, ARB_draw_instanced only adds variables to the vertex
shader stage.)
Previously, we only allowed such extensions to be enabled in the shader
stages where they're useful. However, I've never found any text which
mandates that behavior; in my opinion, you should be able to turn on
extensions in any shader stage, even if they have no effect.
Fixes Piglit tests glslparsertest/glsl2/draw_buffers-05.vert and
ARB_draw_instanced/preprocessor/feature-macro-enabled.frag.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=29185
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
EGL_KHR_surfaceless_context extension allows contexts to be made current
without a default winsys fbo. This extension specifies what ES 1.1 and
2.0 should do (the ES 3.0 spec already does).
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Commit 8c3d3622d9 introduced a new assertion,
but since it causes lp_test_conv failures remove it again and let's hope
we don't really hit bugs caused by the potentially bogus code (it is possible
the assert() caught some cases which work correctly too).
The second 'const' says that the pointer itself is constant. This in
unenforcible in C++, so GCC emits a warning (see) below for each of
these functions in every file that includes glsl_types.h. It's a lot of
warning spam.
../../../src/glsl/glsl_types.h:176:58: warning: type qualifiers ignored on function return type [-Wignored-qualifiers]
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Cc: mesa-stable@lists.freedesktop.org
The majority of calls to _mesa_glsl_error(), _mesa_glsl_warning(), and
_mesa_glsl_parse_state::check_version() use a message that begins with
a lower case letter and ends without a period. This patch makes all
messages follow that convention.
Also, error/warning messages shouldn't end in '\n', since
_mesa_glsl_msg() automatically adds '\n' at the end of the message.
Reviewed-by: Matt Turner <mattst88@gmail.com>
Just like the UNORM case we need to use round to nearest, not trunc.
(There's also another problem, we're using the formula for SNORM->float
which will produce a value below -1.0 for the most negative value which
according to both OpenGL and d3d10 would need clamping. However, no actual
failures have been observed due to that hence keep cheating on that.)
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
I am not able to find _any_ rounding behavior specified for OpenGL for
float to half-float conversions. However, it is specified for fp11/fp10
which suggests round to next finite value but round-to-zero would also
be allowed, but finite values must not be flushed to infinity in either
case.
Hence I believe it makes sense to do the same for half-floats too.
We could probably also use round-to-zero consistently, which is in fact
required by d3d10 (but it doesn't seem to matter much).
Does not match the mesa core function doing the same though (which is
saying it was built to match intel gpus which I don't believe for a
second as it would cause failures in d3d10, moreover the PRM (for
ivy bridge, not listed in older manuals) while not specifying rounding
behavior clearly states finite numbers are never flushed to infinity).
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
unlike OpenGL, the texel swizzle is embedded in the instruction, so honor
that.
(Technically we now execute both the sampler_view swizzle and the
per-instruction swizzle but this should be quite ok.)
v2: add documentation note as it's not obvious.
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
GL_EXT_framebuffer_object differs from GL_ARB_framebuffer_object in ways
that we can't and don't implement in core profiles. Exposing it is a
lie, so we shouldn't do that.
It's possible the some other GL_EXT_framebuffer_* extensions should be
disabled, but it's not quite so clear cut.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
If any component used the ZERO or ONE swizzle, its corresponding member
in the `swizzle` array would never be initialized. We *mostly* got away
with this, except when that memory happened to contain a value that
clobbered another channel when combined using BRW_SWIZZLE4().
NOTE: This is a candidate for stable branches.
Signed-off-by: Chris Forbes <chrisf@ijw.co.nz>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
This fixes the dri2 opening to check if DRI_PRIME is set,
and picks the correct drm device path to open, this along
with a change to libvdpau allows vdpauinfo to work at least,
Martin Peres tested with nouveau, and there seems to be a
further issue with final displaying, it only works sometimes,
but this patch is at least necessary to help debug further.
Signed-off-by: Dave Airlie <airlied@redhat.com>
Cc: mesa-stable@lists.freedesktop.org
Reviewed-by: Christian König <christian.koenig@amd.com>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=67283
Tested-by: Armin K. <krejzi@email.com>
This reverts commit c9db037dc9.
Eric believes that the viewport hacks are still necessary for EGL;
invalidate events aren't hooked up properly.
This commit caused a regression where EFL applications wouldn't show
anything other than window decorations; GLBenchmark also showed issues.
The revert had conflicts due to the intel_context/brw_context merge.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=66606
Cc: mesa-stable@lists.freedesktop.org
Wrapping every character of an email address in <em> looks bizarre, and
makes it impossible to read the text. Apparently Brian did this in 2003
to try and obfuscate email addresses and avoid spam.
Of course, mesa-*@lists.freedesktop.org are public mailing lists and
trivial to find on the internet. So obfuscation buys us nothing
(assuming the <em> technique even works at all, which I doubt).
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
LOLed-at-by: Matt Turner :)
Bump major version, as the change to require explicit
xa_context_flush(), the addition of the handle-type parameter to
xa_surface_handle(), and change of surface to ref/unref will require a
minor change in DDX.
For freedreno DDX, we have to create the scanout GEM bo in a special way
(until we have our own KMS/DRM kernel driver.. and even then for
phones/tablets you probably need to use the android drivers if you don't
want to port the lcd panel driver support). The easiest way to handle
this is let the DDX create the scanout bo, and then create the xa
surface from that.
Signed-off-by: Rob Clark <robclark@freedesktop.org>
TargetOptions::NoFramePointerElimNonLeaf was removed in LLVM 3.4
r187093.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: José Fonseca <jfonseca@vmware.com>
The is_loop_terminator() function was asserting that the following
kind of if statement could never occur:
if (...) { } else { }
(presumably based on the assumption that such an if statement would be
eliminated by previous optimization stages). But that isn't the
case--it's possible that previous optimization stages might simplify
more complex code down to this empty if statement, in which case it
won't be eliminated until the next time through the optimization loop.
So is_loop_terminator() needs to handle it. Fortunately it's easy to
handle--it's not a loop terminator because it does nothing.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=64330
CC: mesa-stable@lists.freedesktop.org
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Two callers of brw_search_cache() weren't initializing that function's
inout_offset parameter: brw_blorp_const_color_params::get_wm_prog()
and brw_blorp_const_color_params::get_wm_prog().
That's a benign problem, since the only effect of not initializing
inout_offset prior to calling brw_search_cache() is that the bit
corresponding to cache_id in brw->state.dirty.cache may not be set
reliably. This is ok, since the cache_id's used by
brw_blorp_const_color_params::get_wm_prog() and
brw_blorp_blit_params::get_wm_prog() (BRW_BLORP_CONST_COLOR_PROG and
BRW_BLORP_BLIT_PROG, respectively) correspond to dirty bits that are
not used.
However, failing to initialize this parameter causes valgrind to
complain. So let's go ahead and fix it to reduce valgrind noise.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=66779
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
The linker matches up variables in interface blocks according to their
block name and variable name. When support for interface block arrays
was added in commit d6863acb, we renamed variables appearing in
interface blocks so that their name included the array size. For
example, in a block like this:
out foo {
float bar
} baz[3];
The variable "bar" would get renamed to "bar[3]".
This is unnecessary, and leads to problems in supporting geometry
shaders, since geometry shaders require vertex shader outputs which
are non-arrays to be linked up to geometry shader inputs which are
arrays.
This patch makes the behaviour of interface block arrays the same as
simple non-array interface blocks; in both cases, the variables
contained within them are not renamed.
Reviewed-by: Matt Turner <mattst88@gmail.com>
vertex id has to be unaffected by the start index (i.e. when calling
draw arrays with start_index = 5, the first vertex_id has to still
be 0, not 5) and it has to be equal to the index when performing
indexed rendering (in which case it has to be unaffected by the
index bias). This fixes our behavior.
Signed-off-by: Zack Rusin <zackr@vmware.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
The instance id system value always starts at 0, even if the
specified start instance is larger than 0. Instead of implicitly
setting instance id to instance id plus start instance and then
having to subtract instance id when computing the buffer offsets
lets just set instance id to the proper instance id. This fixes
instance id computation and cleansup buffer offset computation.
Signed-off-by: Zack Rusin <zackr@vmware.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
There are earlier returns for PIPE_FUNC_NEVER and PIPE_FUNC_ALWAYS. The
switch value of 'func' cannot be either of those values.
Fixes "Logically dead code" defects reported by Coverity.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: José Fonseca <jfonseca@vmware.com>
Looks like a thinko, "Hey, constant buffers can be at most 64 KiB
in size, offset can't be larger." But it can, of course.
I think piglit lacks a test for UBO and BindBufferRange that
tests if it actually works.
Since disabling denorms in draw_vbo() we require the util_cpu_caps to be
initialized there. Hence add another util_cpu_detect() call in
draw_create_context() which should ensure this.
(There is another call in draw_get_option_use_llvm() which only gets called
with x86 (not x86_64) but calling it always there wouldn't help since it most
likely wouldn't get called when compiling without llvm, so leave it alone
there.)
This fixes https://bugs.freedesktop.org/show_bug.cgi?id=66806.
(Because util_cpu_caps wasn't initialized when first calling util_fpstate_get()
hence it returning zero, but it would later get initialized by rtasm translate
code hence when draw call returned it unmasked all exceptions by calling
util_fpstate_set(). This was happening only with DRAW_USE_LLVM=0 or not
compiling with llvm, otherwise the llvm init code was calling it on time too.)
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
Reviewed-by: Zack Rusin <zackr@vmware.com>
Tested-by: Vinson Lee <vlee@freedesktop.org>
The codeword must be unsigned (otherwise will shift in 1's from above when
merging low/high parts so some texels decode wrong).
This also affects gallium's util/u_format_rgtc.
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
Reviewed-by: Zack Rusin <zackr@vmware.com>
For AVX it's not sufficient to only rely on the cpuid flags. If the CPU
supports these extensions, but the OS doesn't, issuing these insns will
trigger an undefined opcode exception.
In addition to the AVX cpuid bit we also need to:
* test cpuid for OSXSAVE support
* XGETBV to check if the OS saves/restores AVX regs on context switches
See "Detecting Availability and Support" at
http://software.intel.com/en-us/articles/introduction-to-intel-advanced-vector-extensions
Signed-off-by: Andre Heider <a.heider@gmail.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: José Fonseca <jfonseca@vmware.com>
Test infs, zeros and nans with our arith functions to assure
correct/defined behavior with those values.
Signed-off-by: Zack Rusin <zackr@vmware.com>
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Same as log2_safe, which means that it can handle infs, 0s and
nans.
Signed-off-by: Zack Rusin <zackr@vmware.com>
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Only the floating point operarators change everything else
is the same so it makes sense to share the code.
Signed-off-by: Zack Rusin <zackr@vmware.com>
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
sin/cos for anything not finite is nan and everything else has
to be between [-1, 1].
Signed-off-by: Zack Rusin <zackr@vmware.com>
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
That means that if input is:
* - less than zero (to and including -inf) then NaN will be returned
* - equal to zero (-denorm, -0, +0 or +denorm), then -inf will be returned
* - +infinity, then +infinity will be returned
* - NaN, then NaN will be returned
It's a separate function because the checks are a little bit costly
and in most cases are likely unnecessary.
Signed-off-by: Zack Rusin <zackr@vmware.com>
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
exp(0) has to be exactly 1, exp(-inf) has to be 0, exp(inf) has
to be inf and exp(nan) has to be nan, this fixes all of those
cases.
Signed-off-by: Zack Rusin <zackr@vmware.com>
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Both D3D10 and OpenCL say that if one the inputs is nan then
the other should be returned. To preserve that behavior
the patch fixes both the sse and the non-sse paths in both
functions and adds helper code for handling nans.
Signed-off-by: Zack Rusin <zackr@vmware.com>
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
It's not the first time that, due to missing build dependencies or
incomplete commits, we end up with a broken libGL.so that's missing
symbols, causing all tests to fail catastrophically.
Instead try to catch this sort of issues earlier.
2013-07-19 13:08:07 +01:00
1530 changed files with 78599 additions and 54859 deletions
People who are concerned with stability and reliability should stick
with a previous release or wait for Mesa 10.0.1.
</p>
<p>
Mesa 10.0 implements the OpenGL 3.3 API, but the version reported by
glGetString(GL_VERSION) or glGetIntegerv(GL_MAJOR_VERSION) /
glGetIntegerv(GL_MINOR_VERSION) depends on the particular driver being used.
Some drivers don't support all the features required in OpenGL 3.3. OpenGL
3.3 is <strong>only</strong> available if requested at context creation
because compatibility contexts are not supported.
</p>
<h2>MD5 checksums</h2>
<pre>
TBD.
</pre>
<h2>New features</h2>
<p>
Note: some of the new features are only available with certain drivers.
</p>
<ul>
<li>GL_AMD_seamless_cubemap_per_texture on i965.</li>
<li>GL_ARB_conservative_depth on i965.</li>
<li>GL_ARB_texture_gather on i965.</li>
<li>GL_ARB_texture_query_levels on i965.</li>
<li>GL_ARB_texture_mirror_clamp_to_edge.</li>
<li>GL_ARB_transform_feedback2, GL_ARB_transform_feedback3, and GL_ARB_transform_feedback_instanced on i965/Gen7 (with appropriate kernel support).</li>
<li>GL_ARB_sample_shading on i965.</li>
<li>GL_ARB_shader_atomic_counters on i965.</li>
<li>GL_ARB_vertex_attrib_binding</li>
<li>GL_ARB_vertex_type_10f_11f_11f_rev on i965 and r600g</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=47824">Bug 47824</a> - osmesa using --enable-shared-glapi depends on libgl</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=62362">Bug 62362</a> - Crash when using Wayland EGL platform</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=63435">Bug 63435</a> - [Regression since 9.0] Flickering in EGL OpenGL full-screen window with swap interval 1</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=64087">Bug 64087</a> - Webgl conformance shader-with-non-reserved-words crash when mesa is compiled without --enable-debug</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=65236">Bug 65236</a> - [i965] Rendering artifacts in VDrift/GL2</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=66558">Bug 66558</a> - RS690: 3D artifacts when playing SuperTuxKart</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=66847">Bug 66847</a> - compilation broken with llvm 3.3</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=66850">Bug 66850</a> - glGenerateMipmap crashes when using GL_TEXTURE_2D_ARRAY with compressed internal format</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=66921">Bug 66921</a> - [r300g] Heroes of Newerth: HiZ related corruption</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=67283">Bug 67283</a> - VDPAU doesn't work on hybrid laptop through DRI_PRIME</li>
</ul>
<h2>Changes</h2>
<p>The full set of changes can be viewed by using the following GIT command:</p>
<pre>
git log mesa-9.1.5..mesa-9.1.6
</pre>
<p>Andreas Boll (1):</p>
<ul>
<li>configure.ac: Require llvm-3.2 for r600g/radeonsi llvm backends</li>
</ul>
<p>Brian Paul (4):</p>
<ul>
<li>mesa: handle 2D texture arrays in get_tex_rgba_compressed()</li>
<li>meta: handle 2D texture arrays in decompress_texture_image()</li>
<li>mesa: implement mipmap generation for compressed 2D array textures</li>
<li>mesa: improve free() cleanup in generate_mipmap_compressed()</li>
</ul>
<p>Carl Worth (7):</p>
<ul>
<li>docs: Add 9.1.5 release md5sums</li>
<li>Merge 'origin/9.1' into stable</li>
<li>cherry-ignore: Drop 13 patches from the pick list</li>
<li>get-pick-list.sh: Include commits mentionining "CC: mesa-stable..." in pick list</li>
<li>get-pick-list: Allow for non-whitespace between "CC:" and "mesa-stable"</li>
<li>get-pick-list: Ignore commits which CC mesa-stable unless they say "9.1"</li>
<li>Bump version to 9.1.6</li>
</ul>
<p>Chris Forbes (5):</p>
<ul>
<li>i965/Gen4: Zero extra coordinates for ir_tex</li>
<li>i965/vs: Fix flaky texture swizzling</li>
<li>i965/vs: set up sampler state pointer for Gen4/5.</li>
<li>i965/vs: Put lod parameter in the correct place for Gen4</li>
<li>i965/vs: Gen4/5: enable front colors if back colors are written</li>
</ul>
<p>Christoph Bumiller (1):</p>
<ul>
<li>nv50,nvc0: s/uint16/uint32 for constant buffer offset</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=61635">Bug 61635</a> - glVertexAttribPointer(id, GL_UNSIGNED_BYTE, GL_FALSE,...) does not work</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=65958">Bug 65958</a> - GPU Lockup on Trinity 7500G</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=67548">Bug 67548</a> - glGetAttribLocation seems to be broken</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=68195">Bug 68195</a> - piglit tests vs-struct-pad and fs-struct-pad both fail</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=68250">Bug 68250</a> - Automatic mipmap generation with texture compression produces borders that fade to black</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=66779">Bug 66779</a> - Use of uninitialized stack variable with brw_search_cache()</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=68233">Bug 68233</a> - Valgrind errors in mesa</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=68250">Bug 68250</a> - Automatic mipmap generation with texture compression produces borders that fade to black</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=44618">Bug 44618</a> - Cross-compilation broken by glsl builtin_compiler</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=46632">Bug 46632</a> - Make the alignment checks for the readpixel blit fastpath a bit more lenient</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=47116">Bug 47116</a> - Enemy territory freezes with rs880 and commit fbebd431ec4e2e461a0cbcd5f3a04a000b8f6bbf</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=47248">Bug 47248</a> - autogen missing dependency on flex and bison, causes infinite loop in glsl build</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=50655">Bug 50655</a> - [r600g][RV670 HD3870] Ioquake games causes GPU lockup (waiting for 0x00003039 last fence id 0x00003030)</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=51471">Bug 51471</a> - [965gm] Corrupted graphics in corners of screen with pixel shaders enabled</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=51782">Bug 51782</a> - mesa-8.0.3: fails to compile against uclibc</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=58680">Bug 58680</a> - [IVB] Graphical glitches in 0 A.D</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=58872">Bug 58872</a> - Mac OS X configure: error: Couldn't find clock_gettime</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=59322">Bug 59322</a> - r300g MSAA breaks Half-Life 2 in Wine</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=59364">Bug 59364</a> - [bisected] Mesa build fails: clientattrib.c:33:22: fatal error: indirect.h: No such file or directory</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=59439">Bug 59439</a> - glCopyPixels generates no fragments (occlusion_query_meta_fragments test fails)</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=59440">Bug 59440</a> - glBitmap generates no fragments (occlusion_query_meta_fragments test fails)</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=60086">Bug 60086</a> - Wayland platform backend crashes if there's no back buffer during dri2_swap_buffers</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=60633">Bug 60633</a> - EXT_texture_sRGB does not work in game The Cave on IvyBridge</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=60737">Bug 60737</a> - In GLSL ES, a missing FS precision qualifier does not generate an error</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=60866">Bug 60866</a> - GLSL performance issues for uniform buffer objects</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=61036">Bug 61036</a> - Shader fails to build in LLVMpipe, aborts program</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=61200">Bug 61200</a> - insufficient linking of libxatracker.so</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=61635">Bug 61635</a> - glVertexAttribPointer(id, GL_UNSIGNED_BYTE, GL_FALSE,...) does not work</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=62466">Bug 62466</a> - r600g hyperz lockups with KSP 0.19</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=62669">Bug 62669</a> - HyperZ freeze when playing PrBoom-Plus demo with lots of monsters</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=62721">Bug 62721</a> - GPU lockup in Minecraft 1.5.1 with HyperZ</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=64959">Bug 64959</a> - Cannot build against EGL without X11</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=65112">Bug 65112</a> - glcpp hangs parsing line continuations</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=65958">Bug 65958</a> - GPU Lockup on Trinity 7500G</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=66450">Bug 66450</a> - JUNIPER UVD accelerated playback of MPEG 1/2 streams does not work</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=66606">Bug 66606</a> - [i965 bisected]GLBenchmark 2.5.1/2.7.0 sometimes render error with gnome-session enabling SNA</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=66713">Bug 66713</a> - Team Fortress 2 crashes with r600-sb on HD4850</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=67354">Bug 67354</a> - glsl_parser.cpp is broken with bison 3.0</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=67548">Bug 67548</a> - glGetAttribLocation seems to be broken</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=67927">Bug 67927</a> - R600_DEBUG=sb: Celestia show 2 earths, one wrongly rendered</li>
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.
