These reworks were combined into this patch:
* Matt Turner: i965: Disable NoDDChk/NoDDClr test on Gen12+
* Francisco Jerez: intel/eu/validate/gen12: Disable
qword_low_power_no_depctrl eu_validate test.
Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Francisco Jerez <currojerez@riseup.net>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Calculated the number for allocation and did not
reserve space ....
Fixes: 2117c53b72 "radv: Add temporary datastructure for submissions."
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Variables spilled on both branch legs need to be assigned to the same spilling slot.
These affinities can be transitive through multiple merge blocks.
Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
This patch makes the live variable analysis more precise
w.r.t. killed phi operands and the block's register pressure.
Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Converting to 'Conventional SSA Form' ensures correctness w.r.t. spilling of phi nodes.
Previously, it was possible that phi operands have intersecting live-ranges, and thus,
couldn't get spilled to the same spilling slot. For this reason, ACO tried to avoid to
spill phis, even if it was beneficial.
This patch implements a conversion pass which is currently only called if spilling is necessary.
Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Fixes these deqp tests (and more):
dEQP-GLES2.functional.draw.draw_arrays.points.single_attribute
dEQP-GLES2.functional.draw.draw_arrays.points.multiple_attributes
dEQP-GLES2.functional.draw.draw_arrays.points.default_attribute
dEQP-GLES2.functional.draw.draw_elements.points.single_attribute
dEQP-GLES2.functional.draw.draw_elements.points.multiple_attributes
dEQP-GLES2.functional.draw.draw_elements.points.default_attribute
Signed-off-by: Jonathan Marek <jonathan@marek.ca>
Reviewed-by: Christian Gmeiner <christian.gmeiner@gmail.com>
The final version of previous stencil fix patch ended up breaking one-sided
stencil.
Fixes remaining failures in these deqp tests (tested on GC3000/GC7000L):
dEQP-GLES2.functional.fragment_ops.depth_stencil.*
Note: deqp tests require --deqp-gl-config-name=rgba8888d24s8ms0
Fixes: 05da025f ("etnaviv: fix two-sided stencil")
Signed-off-by: Jonathan Marek <jonathan@marek.ca>
Reviewed-by: Christian Gmeiner <christian.gmeiner@gmail.com>
Fixes remaining failures in these deqp tests (tested on GC3000/GC7000L):
dEQP-GLES2.functional.polygon_offset.*
Fixes: 6c3c05dc ("etnaviv: fix polygon offset")
Signed-off-by: Jonathan Marek <jonathan@marek.ca>
Reviewed-by: Christian Gmeiner <christian.gmeiner@gmail.com>
On gen11 and older, compressed images are tiled and aligned to 4K. On
gen12 this 4K alignment restriction was removed. However, only aligning
the fast clear color buffer to 64B (a cacheline, as it's on the
documentation) is causing some bugs where the fast clear color is not
converted during the fast clear operation. Aligning things to 4K seems
to fix it.
v2: Fix typo case in the comment (Nanley)
v3: Rebase and fix conflicts.
v4: Fix rebase mistake (Nanley).
Reviewed-by: Nanley Chery <nanley.g.chery@intel.com>
On gen11 and older, compressed images are tiled and aligned to 4K. On
gen12 this 4K alignment restriction was removed. However, only aligning
the fast clear color buffer to 64B (a cacheline, as it's on the
documentation) is causing some bugs where the fast clear color is not
converted during the fast clear operation. Aligning things to 4K seems
to fix it.
v2: Assert that image->planes[plane].offset is 4K aligned (Nanley)
Reviewed-by: Nanley Chery <nanley.g.chery@intel.com>
While we're at it, make sure we error out if it's not supported when
required.
This brings us a bit closer to being able to test on SwiftShader, which
doesn't currently support KHR_external_memory_fd.
TGL will have separate tables for src0 and src1, so the shared function
will no longer make sense.
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
The EU compaction unit test fuzzes the compaction code by flipping bits.
We use a simple skip_bits() function with a list of reserved bits to
ignore, but for more complex cases like invalid combinations of register
file:type, we need either machinery to check validity or for these
functions to simply inform us whether a combination was valid.
enum brw_reg_type a 4-bit field in brw_reg, so rather than expanding it
with an "INVALID" value, just return -1 and let the caller check for
that.
Scott suggested redefining unreachable() within the unit test to
longjmp() which would allow driver code like this to still use it and
allow the test to handle expected failures like this. If that plan works
out, I plan to revert this.
Mostly for vertex formats, but they are supported as texture formats too
(untested however).
Signed-off-by: Jonathan Marek <jonathan@marek.ca>
Reviewed-by: Rob Clark <robdclark@gmail.com>
If src/dst addresses are dw aligned and size is > 4 then we align
byte count to dw as well.
PAL implementation works like this.
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Previously, the scheduler tried to move up instructions from below depending
VMEM instructions only to move them down again when scheduling the VMEM
instruction.
Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
These got lost due to some refactoring.
Due to the way our scheduler works currently, for now
we add back the reorder flag for divergent loads only.
Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
This patch changes VMEM scheduling in a way that they can only
be moved upwards by previous VMEM instructions but not downwards.
This way, it improves the order of VMEM instructions in relation
to their users.
Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Previously, we allowed all shaders to reduce the number of max_waves to as low as 5.
Restricting this on shaders with low register demand, increases the total number of waves
while the VMEM def-use distances hardly change.
This patch also changes the max number of move operations per MEM instruction.
Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
This shaves around 4-5% off of a CPU-limited example running with the
Dawn WebGPU implementation.
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
In 0e4a75f917, Ken added a flag brw_stage_prog_data which indicates
whether any UBO pulls ever occur. Unfortunately, he neglected to set
the bit in the vec4 back-end. This was fine at the time because the
optimization was intended for iris which does not support gen7 and using
the vec4 back-end on Gen8+ requires an environment variable. We want to
use this in Vulkan which does support Gen7 so we want the information
from the vec4 back-end as well as scalar.
Fixes: 0e4a75f917 "intel/compiler: Record whether any pull constant..."
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
RADV_PERFTEST=outooforder has been removed a while ago. This fixes
dumping the options into hang reports.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
This is actually a non-threaded implementation. I'd summarize this
as event-based submission.
When submit happens we walk a tree of submissions that depend on
the syncobj signal operations to be submitted and if those submission
we no other dependencies we start to execute them immediately.
Or, well I still use a list to avoid issues with long chains and
the stacksize when using recursion.
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
This does not fully do wait-before-submit, to be done in a follow
up patch.
For kernels without support for timeline syncobjs, this adds an
implementation of non-shareable timelines using legacy syncobjs.
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
This will lead to fewer pipelines in the cache, which is assumed to
become our most unavoidable performance bottle-neck down the line.
Reviewed-by: Dave Airlie <airlied@redhat.com>
This is a function with timeout support for reading from the pipe
between processes used for secure compile.
Initially we hardcode the timeout to 5 seconds. We can adjust the
timeout limit in future if needed.
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
This will be used in the following patch to support timeouts for
reading the pipe between processes.
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
This skips touching %ebx most times and it shows that glGetString performance
increased from 114M/s to 120M/s on my desktop.
Reviewed-by: Matt Turner <mattst88@gmail.com>
Signed-off-by: Lepton Wu <lepton@chromium.org>
Remove hard coded 16 and use entry_generate_or_patch to patch
public stubs. The generated code actually is sightly tighter
than before since the "nop" instructions before the final "jmp"
get removed.
Reviewed-by: Matt Turner <mattst88@gmail.com>
Signed-off-by: Lepton Wu <lepton@chromium.org>
The code works exactly the same with before. Just split this function
out so we can reuse it.
Reviewed-by: Matt Turner <mattst88@gmail.com>
Signed-off-by: Lepton Wu <lepton@chromium.org>
The x86 assembly language stub in src/mapi/entry_x86_tsd.h does not
generate PIC (position-independent code). This causes text relocations
which bring troubles on recent versions of FreeBSD, OpenBSD, Android.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=108541
Reviewed-by: Matt Turner <mattst88@gmail.com>
Signed-off-by: Lepton Wu <lepton@chromium.org>
Vulkan requires that only one bit for the ordering is set, but old
versions of GLSLang just set all the bits. This was fixed as part of
c51287d744
but we can still find older versions (or shaders compiled with it)
around.
So instead of failing, emit a warning and fallback to the effective
result of any combination of multiple bits: AcquireRelease.
Closes: https://gitlab.freedesktop.org/mesa/mesa/issues/2018
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
We have to resolve destination surfaces if we are bliting to and from
the same surface.
v2: Revert unrelated change (Nanley Chery)
Signed-off-by: Sagar Ghuge <sagar.ghuge@intel.com>
Reviewed-by: Nanley Chery <nanley.g.chery@intel.com>
Let aux surface state tracker track the stencil buffer's aux state while
clearing depth stencil buffer.
v2: Fix condition check (Nanley Chery)
Signed-off-by: Sagar Ghuge <sagar.ghuge@intel.com>
Reviewed-by: Nanley Chery <nanley.g.chery@intel.com>
Even though stencil buffer compression looks like regular lossless color
compression w/o fast clear support, we have to resolve stencil buffer
with WM_HZ_OP packet.
v2: Check if resource is stencil with helper function (Nanley Chery)
v3: Remove unnecessary included file (Nanley Chery)
v4: (Nanley Chery)
- Avoid stencil buffer aux state transition by improving condition check
Signed-off-by: Sagar Ghuge <sagar.ghuge@intel.com>
Reviewed-by: Nanley Chery <nanley.g.chery@intel.com>
On Gen12+, Stencil buffer's lossless compression should be resolved
with WM_HZ_OP packet.
Signed-off-by: Sagar Ghuge <sagar.ghuge@intel.com>
Reviewed-by: Nanley Chery <nanley.g.chery@intel.com>
We never saw any failures regarding this typo but it's good to assign
correct stencil view while constructing blorp_params.
Fixes: 0cabf93b80 "intel/blorp: Add an entrypoint for clearing depth and stencil"
Signed-off-by: Sagar Ghuge <sagar.ghuge@intel.com>
Reviewed-by: Nanley Chery <nanley.g.chery@intel.com>
On Gen12, the CCS buffer address doesn't have to be referenced in state
packets. In the case of a stencil buffer with CCS, the kernel won't know
the location of the CCS unless an extra call is made to pin its address.
To avoid this extra call, make the CCS part of the main surface.
v2. Update comment above bo_size. (Jordan)
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
The functions used during aux buffer configuration and creation only
return false for exceptional errors. Don't proceed with surface creation
in those cases.
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
The original value of 256 was under the assumption that you're a batch
buffer which is likely going to have a large number of relocations.
However, pipeline objects on Gen7 will have at most 6 relocations (one
per shader stage and one for the workaround BO) so this is a lot of
per-pipeline wasted space.
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
The old relocation list code always allocated 256 relocations and a hash
set up-front without knowing whether or not we really need them. In
particular, in the softpin case, this is two fairly large allocations
that we don't need to be making. Also, for pipeline objects on haswell
where we don't have softpin, we don't need relocations unless scratch is
used so this is extra data per-pipeline. Instead, we should do it
on-demand. This shaves 3.5% off of a cpu-limited example running with
the Dawn WebGPU implementation.
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
For gen12 we set the streamout buffers using 4 separate
commands instead of 3DSTATE_SO_BUFFER.
Signed-off-by: Plamena Manolova <plamena.manolova@intel.com>
Acked-by: Jason Ekstrand <jason@jlekstrand.net>
Extract out values for the handful of unknown registers which have
different values across different a6xx models, to simplify adding
support for new a6xx's.
Signed-off-by: Rob Clark <robdclark@chromium.org>
E.g. documentation-only changes cannot affect the outcome of the
pipeline, so don't waste resources on running it.
The thing we need to be careful about here is that the container stage
jobs must always run if any later stage jobs using the corresponding
docker images run. We're currently using the same .ci-run-policy
template for all jobs, so this is trivially true.
v2:
* Add bin/ and common.py (Eric Engestrom)
Acked-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com> # v1
Reviewed-by: Eric Engestrom <eric.engestrom@intel.com>
A few equations/programming changes for ICL.
v2: Fix a couple of issues in naming and floating/integer operations (Ken)
Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Acked-by: Kenneth Graunke <kenneth@whitecape.org>
Commit 9edcce2a32 bumped the required libdrm-amdgpu version to
2.4.100. Update the version we use in our CI scripts to avoid CI
build failures.
Also bump the debian image name for this change to take effect.
Note that amdgpu is only built with the debian-buster image,
so only this image requires an update.
Fixes: 9edcce2a ("ac: get tcc_harvested from the kernel")
Signed-off-by: Alexandros Frantzis <alexandros.frantzis@collabora.com>
Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
The anv_batch_bo contents are linked one to another, and when printing
we have to start with the first of those. Since in `u_vector` new
elements are added to the head, to get the first element we need the
vector's tail.
Fixes: 32ffd90002 ("anv: add support for INTEL_DEBUG=bat")
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Previously subgroup shuffle was implemented using the bpermute
instruction, which only works accross half-waves, so by itself it's
not suitable for implementing subgroup shuffle when the shader is
running in wave64 mode.
This commit adds a trick using shared VGPRs that allows to implement
subgroup shuffle still relatively effectively in this mode.
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Fixes p_reduce (all cluster sizes), p_inclusive_scan and p_exclusive_scan
with all reduction operations.
Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
The existing "fallback" code didn't actually do anything, so this
removes it, and instead we just always fallback to `iris` for future
PCI IDs.
Suggested-by: Kenneth Graunke <kenneth@whitecape.org>
Signed-off-by: Eric Engestrom <eric.engestrom@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
When this code was merged, this wasn't necessary because the
state-tracker would do it later anyway. But this recently got changed,
without changing the code that depended on this.
Arguably, this was a mistake in the lowering pass to begin with. Either
way, let's fix it by not assuming that the lowering code gets called
later when it's not needed.
This fixed user-defined clip-planes in Zink.
Signed-off-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Fixes: eaffdad108 ("st/mesa: don't lower_global_vars_to_local for VS if there are no dead inputs")
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Add case for MCS_CCS so that we get the correct aux usage while copy
operation.
v2: Fix commit subject (Nanley Chery)
Signed-off-by: Sagar Ghuge <sagar.ghuge@intel.com>
Reviewed-by: Nanley Chery <nanley.g.chery@intel.com>
Depending on MCS_CSS or MCS we can emit blorp blit shaders.
As we support MCS_CSS and MCS, it makes sense to use
isl_aux_usage_has_mcs function.
v2: Fix commit message (Nanley Chery)
Signed-off-by: Sagar Ghuge <sagar.ghuge@intel.com>
Reviewed-by: Nanley Chery <nanley.g.chery@intel.com>
If aux for MCS is already configured, don't configure again.
v2: Fix missing period in commit message (Nanley Chery)
Signed-off-by: Sagar Ghuge <sagar.ghuge@intel.com>
Reviewed-by: Nanley Chery <nanley.g.chery@intel.com>
The Vulkan spec says that an implementation has to support one of
VK_FORMAT_X8_D24_UNORM_PACK32 and VK_FORMAT_D32_SFLOAT, as well of
one of VK_FORMAT_D24_UNORM_S8_UINT and VK_FORMAT_D32_SFLOAT_S8_UINT.
So let's keep track which one is supported of earch pair, and emulate
one on top of the other one.
This won't give the exact result for comparisons, or when mapping and
unmapping the resources. But it's better than flat out failing to create
the resource, and we can fix the map/unmap issue later if needed.
Tested-by: Duncan Hopkins <duncan@thefoundry.co.uk>
If a modifier specifies an aux, it must be created.
Fixes: 75a3947af4 ("iris/resource: Fall back to no aux if creation fails")
Reviewed-by: Sagar Ghuge <sagar.ghuge@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Make sure the res struct is free'd before returning.
Fixes: 2dce0e94a3 ("iris: Initial commit of a new 'iris' driver for Intel Gen8+ GPUs.")
Reviewed-by: Sagar Ghuge <sagar.ghuge@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Store the converted depth value into two dwords. Avoids regressing the
piglit test "fbo-depth-array depth-clear", when HIZ_CCS sampling is
enabled in a later commit.
Reviewed-by: Sagar Ghuge <sagar.ghuge@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Write through to the CCS if the surface is used as a texture and can be
sampled by the HW with CCS.
Reviewed-by: Sagar Ghuge <sagar.ghuge@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Check that the alignment requirements for HIZ_CCS are satisfied by using
this function.
Reviewed-by: Sagar Ghuge <sagar.ghuge@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Prevent the piglit test,
amd_vertex_shader_layer-layered-depth-texture-render, from regressing in
in a future commit.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Prepare this function to be used in iris and to handle new Gen12 behavior.
Reviewed-by: Sagar Ghuge <sagar.ghuge@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Add a helper to determine if an ISL surface supports the write-through
mode of HIZ_CCS.
Reviewed-by: Sagar Ghuge <sagar.ghuge@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
The HIZ_CCS and MCS_CCS auxiliary surface modes require that drivers
store information about two aux buffers. We choose to represent this as
HiZ/MCS being the primary aux surface and the CCS as an secondary/extra
aux surface. This representation has the effect of placing most of the
code that will have to choose between the two aux surfaces around the
aux-map entry points.
Reviewed-by: Sagar Ghuge <sagar.ghuge@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Instead of guessing an aux_usage, then confirming it if the
isl_surf_get_*_surf functions are successful, just call the ISL
functions up-front. This will help us to more easily determine if a
depth buffer supports HIZ_CCS.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Add an extra aux parameter which will be filled out with CCS if the
first two isl_surf parameters fit the requirements for HiZ_CCS.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
While this format isn't listed in BSpec: 53911, other documentation and
empirical evidence suggest that it's fine to remap it to R32_FLOAT. I've
filed a bug for the BSpec page.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
We'll start doing slow depth clears more often on HIZ_CCS buffers in a
future commit. Reduce the performance impact by making them use less
bandwidth.
From the Depth Test section of the BSpec:
This function is enabled by the Depth Test Enable state variable. If
enabled, the pixel's ("source") depth value is first computed. After
computation the pixel's depth value is clamped to the range defined
by Minimum Depth and Maximum Depth in the selected CC_VIEWPORT state.
Then the current ("destination") depth buffer value for this pixel is
read.
and from the Depth Buffer Updates section of the BSpec:
If depth testing is disabled or the depth test passed, the incoming
pixel's depth value is written to the Depth Buffer.
Taken together, it's clear that depth testing isn't necessary to perform
a depth buffer clear. Mark Janes and I analyzed this patch with
frameretrace and a depthrange piglit test. I disabled HiZ to ensure we'd
get slow depth clears. We've observed the bandwidth consumption by the
depth buffer access to be cut ~50% on BDW and SKL during depth clears.
On a more graphically intensive workload, the Shadowmapping Sascha
benchmark, I took the average of 3 runs on a BDW with a display
resolution of about 1920x1200 (minus some desktop environment
decorations). I measured a 22.61% FPS improvement when HiZ is disabled.
v2. The BSpec doesn't mandate this behavior, update comment accordingly.
(Ken)
Fixes: bc4bb5a7e3 ("intel/blorp: Emit more complete DEPTH_STENCIL state")
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
In ISL:
Update the format table to add CCS_E support for some 8BPP formats,
some 16BPP formats, and R10G10B10A2_UNORM_SRGB.
In the helper for determining CCS_E support, we return false for some
16BPP formats because they aren't properly handled in blorp_copy().
In BLORP:
Allow the new and non-problematic formats for CCS_E-enabled copies.
v2. Update other fields for A1B5G5R5_UNORM and A4B4G4R4_UNORM in table.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com> (v1)
The CCS could be described in a number of ways, but this format was
chosen to minimize churn in the drivers. We may decide on an different
direction in the future.
v2. Increase alignment for display surfaces. (Nanley)
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com> (v1)
Acked-by: Kenneth Graunke <kenneth@whitecape.org>
Use a helper that will automatically handle Gen12's CCS tiling when
creating a CCS isl_surf.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
In the function which translates ISL tilings to i915 tilings, map ISL's
HiZ and CCS tilings to Y instead of NONE (linear). The HW docs describe
HiZ and pre-Gen12 CCS surfaces as being Y-tiled in memory.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Avoid the compiler warnings for the new enums that will be introduced in
a future commit.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
The isl_surf structs for Gen12's CCS won't describe how many slices in
the main surface can be compressed. All slices will be compressable if
CCS is enabled, so lookup the main surface's logical dimension.
v2. Add a space before a `?`. (Jordan)
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
This isn't accurate enough for HiZ which can have a discontiguous range
of supported aux slices. This also won't work with the plan to represent
Gen12 CCS as a single slice surface.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
From "Render Target Fast Clear" description for Gen12:
"SW must store clear color using MI_STORE_DATA_IMM with
ForceWriteCompletionCheck bit set."
From Instruction_MI_STORE_DATA_IMM, bitfield 10 (when set to 1):
"Following the last write from this command, Command Streamer
will wait for all previous writes are completed and in global
observable domain before moving to next command."
We use 4 SDIs to store the clear color (one per channel). From the
description, it looks to me that setting that flag only on the last SDI
should be enough.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Gen12's CCS requires that the main surface have a pitch aligned to 512B.
v2. Provide a BSpec citation. (Ken)
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
add_aux_state_tracking_buffer() actually checks the aux usage when
determining how many dwords to allocate for state tracking. Move the
function call to the point after the CCS_E aux usage is assigned.
Fixes: de3be61801 ("anv/cmd_buffer: Rework aux tracking")
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Avoid failing the `info->use_clear_address` assertion in ISL on Gen12+.
Fixes: 6c9f9a82d7 ("intel/genxml,isl: Add gen12 render surface state changes")
Reported-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
In gen12 we use the 3DSTATE_DEPTH_BOUNDS instruction
to enable depth bounds testing.
Signed-off-by: Plamena Manolova <plamena.manolova@intel.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
In gen12 we use the 3DSTATE_DEPTH_BOUNDS instruction
to enable depth bounds testing.
Signed-off-by: Plamena Manolova <plamena.manolova@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
In gen12 we add the 3DSTATE_DEPTH_BOUNDS instruction
which enables support for depth bounds testing.
Signed-off-by: Plamena Manolova <plamena.manolova@gmail.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
When resolving a merge-conflict, I accidentally only updated the
ARM64-tag tag. Let's correct this.
Fixes: 3d529c1739 ("gitlab-ci: also build Zink on CI")
Some implementations don't support the lineWidth-feature, so let's
avoid setting invalid state to them. But since we don't have a fallback
for this, inform the user.
Acked-by: Jordan Justen <jordan.l.justen@intel.com>
The driver can report a minimum alignment for UBOs, and that can be
larger than 64, which we've currently been using. Let's play ball, and
use the reported value instead.
Acked-by: Jordan Justen <jordan.l.justen@intel.com>
There's two things that goes wrong in this code on some drivers:
1. Rounding off the line-width to granularity can push it outside the
legal range.
2. A granularity of 0.0 results in NaN, because we divide by zero.
So let's make this code a bit more robust.
Acked-by: Jordan Justen <jordan.l.justen@intel.com>
We're now adding interface-types during code-emitting, so we need to
defer emitting the entry-point. No biggie, spirv_builder is prepares for
this.
Acked-by: Jordan Justen <jordan.l.justen@intel.com>
This is the only call-site that wants to specify unique values per
component for any of the get_*_constant functions. So let's give this
its own implementation instead, so we can ease the burden for the rest.
Acked-by: Jordan Justen <jordan.l.justen@intel.com>
While we're at it, let's move emit_float_const to the same location as
this needs to be defined at.
Acked-by: Jordan Justen <jordan.l.justen@intel.com>
This is going to make it easier to verify that 1-bit float sizes don't
leak into the rest of the code.
Acked-by: Jordan Justen <jordan.l.justen@intel.com>
We don't implement the get_timestamp context-method, so this is just
going to crash if anyone tries to use it. Let's implement it later.
Acked-by: Jordan Justen <jordan.l.justen@intel.com>
This isn't as inaccurate as the comment says, the Vulkan documentation
even seems to suggest this is the same. Let's drop the comment.
Acked-by: Jordan Justen <jordan.l.justen@intel.com>
Because si.waitSemaphoreCount is 0, this won't even be looked at by the
driver, so let's just drop it.
Acked-by: Jordan Justen <jordan.l.justen@intel.com>
This inlines submit_cmdbuf into zink_end_batch, the only place it's
used. This makes the code a bit more straight-forward to read.
Acked-by: Jordan Justen <jordan.l.justen@intel.com>
These aren't guaranteed to be vectors, they can also be scalars. The
var-part is the significant part here, not the vector-ness. So let's
rename these.
Acked-by: Jordan Justen <jordan.l.justen@intel.com>
These track nir-registers, so it's clearer if we refer to them by that
name instead. There's potentially more vars than these.
Acked-by: Jordan Justen <jordan.l.justen@intel.com>
we don't need to track the resources for the samplers any longer, as
the sampler view holds a reference instead.
Acked-by: Jordan Justen <jordan.l.justen@intel.com>
The primitive topology is a bit of an odd-ball, as it's the only
truly draw-call specific state that needs to be passed to the program to
get a pipeline.
So let's make this a bit more explict, by passing it separately. This
makes the flow of data a bit easier to wrap your head around.
Acked-by: Jordan Justen <jordan.l.justen@intel.com>
This adds bitcasting to uint everywhere for now,
and stores all spir-v ssa values as uints.
It also casts bool to 0/0xffffffff for now
(nir 1-bit bools may be coming in the future).
This fixes a lot of piglit tests to pass now
Acked-by: Jordan Justen <jordan.l.justen@intel.com>
In vulkan, the Z-range of clip-space goes from 0..W instead of -W..+W
as is the case in OpenGL. So we need to transform the Z-range to
account for this.
Acked-by: Jordan Justen <jordan.l.justen@intel.com>
Here's zink, a so far pretty simple vulkan-gallium driver that is able
to translate some applications from OpenGL to Vulkan.
The compiler is quite limited for now, this will be improved on later.
Signed-off-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Acked-by: Jordan Justen <jordan.l.justen@intel.com>
Do not flush NaN to 0.
Fixes
dEQP-VK.spirv_assembly.instruction.compute.opquantize.propagated_nans
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
It's similar to GFX9+. Shadow of Mordor (Vulkan beta) hits that
path and it works fine.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
This interface allows the aux-map code in the intel/common library to
allocate and free buffers.
Reworks:
* free gen_buffer in gen_aux_map_buffer_free. (Rafael)
Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Commit d2b60e433e introduced restrictions (as per GLES spec) on the
internal format. We need to setup a sized format for the texture image
so framebuffers created with that are considered complete.
This change fixes following Android CTS test in AHardwareBufferNativeTests
category:
SingleLayer_ColorTest_GpuColorOutputAndSampledImage_R10G10B10A2_UNORM
Signed-off-by: Tapani Pälli <tapani.palli@intel.com>
Fixes: d2b60e433e ("mesa/main: R10G10B10_(A2) formats are not color renderable in ES")
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
I thought there was hardware support for this, but it seems to broken,
or at least more complex than I believed.
Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
This memory needs to still be available after all the drawing is done
and forgotten about, so cannot be transient.
Also clear the result so that no rendering returns a zero.
Signed-off-by: Urja Rannikko <urjaman@gmail.com>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Serialized NIR is required for clover with the SPIR-V pipeline. With
this change and PAN_MESA_DEBUG=deqp, clinfo is able to successfully
probe panfrost.
Code from Nouveau (commit 7955fabcf8 by
Karol Herbst).
Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
A device supported by kmsro will not automatically probe kmsro since the
driver name will be panfrost/lima/v3d/..., not "kmsro". Since kmsro is a
bit of a catch-all for generic (mostly embedded) GPUs, add a fallback on
kmsro for the dynamic loader.
Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Reviewed-by: Christian Gmeiner <christian.gmeiner@gmail.com>
Acked-by: Karol Herbst <kherbst@redhat.com>
kmsro is used by numerous embedded GPUs for a common winsys abstraction.
Let's add support for it for the dynamic pipe loader, so clover can
probe on these drivers.
We build the target with Panfrost. When other drivers need kmsro+clover,
we can revisit the build system part; my mesonfu is wanting.
Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Reviewed-by: Christian Gmeiner <christian.gmeiner@gmail.com>
Acked-by: Karol Herbst <kherbst@redhat.com>
Can be enabled via the environment variable which tells the
driver how many compilation threads are expected to be called,
and therefore how many forked processes the driver should
create.
For example we would expect to call fossilize replay with
something like this:
RADV_SECURE_COMPILE_THREADS=8 ./fossilize-replay --num-threads 8 \
--shader-cache-size 0 --ignore-derived-pipelines pipeline_cache.foz
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
This added support for the fork, the installation of the seccomp
filter, and the main loop for the actual compilation to be called
from i.e. run_secure_compile_device().
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
This function will be called by the parent process when doing a
secure compile. It first selects a free process to work with then
passes it all the information it needs to compile the pipeline.
Once the pipeline information has been passed to the secure
process, it then waits around to read/write any disk cache entries
required before exiting.
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
This allows the secure process to read and write to the disk cache
via the parent process. This commit just adds the functionality
needed for the secure process, the following commit will add the
functionality for the parent process.
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
These will be used by the following commits to hold information about
the forked secure compile processes.
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
This will be used to identify information being passed between the
parent and secure process during a secure compile.
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
This can be usefull for debugging the on disk cache, but is also
useful in the following patch for secure compiles which will be
used to compile huge pipeline collections. These pipeline
collections can be multiple GBs and the in memory cache grows to
multiple GBs very quickly when they are compiled so we want to
be able to turn off the in memory cache.
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
This is cleaner and avoids having to read/write an additional copy of
topology for use with secure compile.
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
This adds a new CI job that runs on windows with MSVC. It currently
builds softpipe and osmesa, and runs the related unit tests. It does
rely on meson's wraps for zlib, but I've set up caching of the wrap
dependencies so hopefully that wont be a problem.
I really wanted to user powershell for this, but there just isn't an
easy way to do that, it's much easier to use batch scripts, so thats
what I used.
The leading `/` for .gitlab-ci/lava... must be removed because windows
doesn't understand it, and when it reads the file the job ends in error.
Reviewed-by: Eric Engestrom <eric@engestrom.ch>
All of these (bug titles, patch titles, features, and people's names)
can contain characters that are not valid html. Just escape everything
for safety.
Fixes: 86079447da
("scripts: Add a gen_release_notes.py script")
Reviewed-by: Eric Engestrom <eric@engestrom.ch>
Reviewed-by: Juan A. Suarez <jasuarez@igalia.com>
I made a bad assumption; I assumed this would be run in the release
branch. But we don't do that, we run in the master branch. As a result
we need to pass the version as an argument.
Fixes: 3226b12a09
("release: Add an update_release_calendar.py script")
Reviewed-by: Eric Engestrom <eric@engestrom.ch>
Reviewed-by: Juan A. Suarez <jasuarez@igalia.com>
Which is very likely .Z > 0 releases.
Fixes: 86079447da
("scripts: Add a gen_release_notes.py script")
Reviewed-by: Eric Engestrom <eric@engestrom.ch>
Reviewed-by: Juan A. Suarez <jasuarez@igalia.com>
If they use the `Fixes: #1` form.
Fixes: 86079447da
("scripts: Add a gen_release_notes.py script")
Reviewed-by: Eric Engestrom <eric@engestrom.ch>
Reviewed-by: Juan A. Suarez <jasuarez@igalia.com>
Previously this would result in the .0 warning be generated for .z > 0
and the .z == 0 would get the other message.
Fixes: 86079447da
("scripts: Add a gen_release_notes.py script")
Reviewed-by: Eric Engestrom <eric@engestrom.ch>
Reviewed-by: Juan A. Suarez <jasuarez@igalia.com>
After the discussion in
https://github.com/KhronosGroup/OpenGL-API/issues/45
the section 8.17 (texture completeness) of the OpenGL 4.6 core profile
was changed to explicitly say that multisample texture completeness
ignores filter state of the texture.
"Using the preceding definitions, a texture is complete unless any of the
following conditions hold true:
...
- The minification filter requires a mipmap (is neither NEAREST nor LINEAR),
the texture is not multisample, and the texture is not mipmap complete.
- The texture is not multisample; either the magnification filter is not
NEAREST, or the minification filter is neither NEAREST nor NEAREST_-
MIPMAP_NEAREST; and any of
– The internal format of the texture is integer (see table 8.12).
– The internal format is STENCIL_INDEX.
– The internal format is DEPTH_STENCIL, and the value of DEPTH_-
STENCIL_TEXTURE_MODE for the texture is STENCIL_INDEX."
Signed-off-by: Danylo Piliaiev <danylo.piliaiev@globallogic.com>
Signed-off-by: Illia Iorin <illia.iorin@globallogic.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
MAX_VARYINGS_INCL_PATCH subtracts VARYING_SLOT_VAR0 giving us a size
that's too small, so BITSET_SET writes words out of bounds, corrupting
the stack and causing all kinds of chaos. VARYING_SLOT_TESS_MAX is
the right value to use here, as it's the largest location.
Closes: 2002
Fixes: ee2050b111 ("nir: Use BITSET for tracking varyings in lower_io_arrays")
Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
[12/60] Compiling C object 'src/gallium/auxiliary/eb820e8@@gallium@sta/rbug_rbug_texture.c.o'.
FAILED: src/gallium/auxiliary/eb820e8@@gallium@sta/rbug_rbug_texture.c.o
[...]
../src/gallium/auxiliary/rbug/rbug_texture.c: In function 'rbug_send_texture_info_reply':
../src/gallium/auxiliary/rbug/rbug_texture.c:302:21: error: implicit declaration of function 'alloca'; did you mean 'malloc'? [-Werror=implicit-function-declaration]
uint32_t *height = alloca(sizeof(uint32_t) * height_len);
^~~~~~
malloc
../src/gallium/auxiliary/rbug/rbug_texture.c:302:21: warning: initialization makes pointer from integer without a cast [-Wint-conversion]
../src/gallium/auxiliary/rbug/rbug_texture.c:303:20: warning: initialization makes pointer from integer without a cast [-Wint-conversion]
uint32_t *depth = alloca(sizeof(uint32_t) * height_len);
^~~~~~
cc1: some warnings being treated as errors
Include c99_alloca.h to portably make the alloca() prototype available.
See also: 498d9d0f, adfb9c5c, fc8139b1
Fixes: 6174cba7 ("rbug: fix transmitted texture sizes")
Reviewed-by: Eric Engestrom <eric@engestrom.ch>
Reviewed-by: Christian Gmeiner <christian.gmeiner@gmail.com>
Rather than supplying a mask/swizzle to compose with the original, just
supply the offset of the allocated register so we can directly offset
the mask/swizzle, without resorting to composition.
This is simpler, cleaner, and will generalize to non-32-bit.
Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
GFX10 hazards require a different approach compared to previous
generations, for example it doesn't need s_nop, and most hazards
can't be solved by adding NOPs at all. Also, they are not
resolved by branch instructions.
This commit reorganizes aco_insert_NOPs so that there is now a
separate pass for GFX10. The new GFX10 pass also respects the
control flow of the shader.
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
This commit refines the VMEMtoScalarWriteHazard mitigation, based
upon a closer look at what LLVM does. Also changes the code to
match the structure of the other hazard mitigations.
* The hazard is not only triggered by VMEM, FLAT and GLOBAL
but also SCRATCH and DS instructions.
* The SMEM/SALU instructions only cause a hazard when they
write a register that the VMEM/etc. are reading.
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
There is a hazard caused by there is a branch between a
VMEM/GLOBAL/SCRATCH instruction and a DS instruction.
This commit adds a workaround that avoids the problem.
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
There is a hazard that happens when an SMEM instruction
reads an SGPR and then a VALU instruction writes that same SGPR.
This commit adds a workaround that avoids the problem.
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
There is a hazard when a non-VALU instruction reads the EXEC mask
and then a VALU instruction writes the EXEC mask.
This commit adds a workaround that avoids the problem.
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Any permlane instruction that follows any VOPC instruction can cause a hazard,
this commit implements a workaround that avoids this causing a problem.
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
ACO currently mitigates VMEMtoScalarWriteHazard and Offset3fBug
(names from LLVM). There are some bugs that ACO needn't care about.
Just to be on the safe side, add an assertion that makes sure
that we aren't hit by FlatSegmentOffsetBug.
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
From the Vulkan spec 1.1.126 :
"VK_SHADER_FLOAT_CONTROLS_INDEPENDENCE_32_BIT_ONLY_KHR specifies
that shader float controls for 32-bit floating point can be set
independently; other bit widths must be set identically to each
other."
Forgot to update this when I enabled that extension recently.
Fixes dEQP-VK.spirv_assembly.instruction.compute.float_controls.independence_settings.independence_setting
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
This saves one return and a simple benchmark which calls glGetString
repeatedly on my desktop shows it improves calls per second from 118M
to 128M.
Signed-off-by: Lepton Wu <lepton@chromium.org>
Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
Most GPU require the sample count is power of 2. Just remove those
formats with unusual sample count. This decreases dEQP EGL tests run
time a lot.
Signed-off-by: Lepton Wu <lepton@chromium.org>
Reviewed-by: Eric Anholt <eric@anholt.net>
MAX_VARYINGS_INCL_PATCH is greater than 64, so we'll need more that 64
bits (per component) to track which vars have indirects. This pass was
trying to track patch varyings (which start at bit 63) in a separate
64 bit word, but failed to subtract VARYING_SLOT_PATCH0 and accessed
out of bounds.
Do away with the ad-hoc bit mask tracking and just use a BITSET.
Fixes: dEQP-GLES31.functional.tessellation.user_defined_io.per_patch_block.vertex_io_array_size_implicit.triangles
Reviewed-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
Signed-off-by: Kristian H. Kristensen <hoegsberg@google.com>
In some cases, in particular when you have things that can be src
modifiers ((abs)/(neg)), once eliminating one mov, there is a
possibility to remove another. Handle this by re-visiting an
instruction after eliminating a copy on one of it's srcs.
Signed-off-by: Rob Clark <robdclark@chromium.org>
Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
These date back to relatively early days of ir3, when a lot was still
not well understood. But according to CI (and what I've seen blob
driver do), these are not actually real restrictions.
Signed-off-by: Rob Clark <robdclark@chromium.org>
Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
Now that we fixed the sharp edges that this was papering over, we can
relax the restriction about eliminating a mov coming out of a fanout
(for example from result of texture fetch).
Signed-off-by: Rob Clark <robdclark@chromium.org>
Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
This avoids copy-propagating a high register into an instruction which
cannot consume it.
Signed-off-by: Rob Clark <robdclark@chromium.org>
Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
We did this properly already for split/fanout. But collect was missed.
Extract out a helper to share.
This way we avoid copy propagating a mov from high or half reg into an
instruction which cannot consume a high/half reg.
Signed-off-by: Rob Clark <robdclark@chromium.org>
Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
1) deduplicate IR3_SHADER_DEBUG=disasm versus fs/vs/etc handling
2) standardize shader stage name prints, in particular VERT vs BVERT
3) don't mix stderr and stdout
Signed-off-by: Rob Clark <robdclark@chromium.org>
Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
Avoid keeping track of the idx and all possible image operands for
each operation. Note for convenience we split up the handling of
ImageOperandsOffsetMask and ImageOperandsConstOffsetMask.
Suggested by Jason Ekstrand.
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Change the information to also include the category, so that the
particulars of BitEnum enumeration can be handled in the template.
Acked-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Emit barriers with semantics matching the access operand and the
storage class of the pointer.
v2: Fix order of visible / available emission relative to the
operations. (Bas)
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Set the memory semantics and scope for later emitting the barrier.
Note the barrier emission code already exist in vtn_handle_image for
the Image atomics.
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Add a helper to split the memory semantics into before and after the
operation, and use that result to emit memory barriers.
v2: Be more explicit about which bits we are keeping around when
splitting memory semantics into a before and after. For now
we are ignoring Volatile. (Jason)
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Including the right storage memory semantic based on the storage class
of the operation. These will be used later to emit memory barriers.
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Three groups of tests, effectively defining what cases the
optimization is allowed or prevented
- Redudant loads (a load generated the value)
- Propagate SSA values (a store generated the value)
- Propagate a var (a copy generated the value)
Change the shader type of the tests to be COMPUTE so
nir_var_mem_shared can also be used. Doesn't affect the semantic of
the copy propagation.
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Add a NIR instrinsic that represent a memory barrier in SPIR-V /
Vulkan Memory Model, with extra attributes that describe the barrier:
- Ordering: whether is an Acquire or Release;
- "Cache control": availability ("ensure this gets written in the memory")
and visibility ("ensure my cache is up to date when I'm reading");
- Variable modes: which memory types this barrier applies to;
- Scope: how far this barrier applies.
Note that unlike in SPIR-V, the "Storage Semantics" and the "Memory
Semantics" are split into two different attributes so we can use
variable modes for the former.
NIR passes that took barriers in consideration were also changed
- nir_opt_copy_prop_vars: clean up the values for the mode of an
ACQUIRE barrier. Copy propagation effect is to "pull up a load" (by
not performing it), which is what ACQUIRE restricts.
- nir_opt_dead_write_vars and nir_opt_combine_writes: clean up the
pending writes for the modes of an RELEASE barrier. Dead writes
effect is to "push down a store", which is what RELEASE restricts.
- nir_opt_access: treat the ACQUIRE and RELEASE as a full barrier for
the modes. This is conservative, but since this is a GL-specific
pass, doesn't make a difference for now.
v2: Fix the scoped barrier handling in copy propagation. (Jason)
Add scoped barrier handling to nir_opt_access and
nir_opt_combine_writes. (Rhys)
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
This warning is different. Meson support for windows is less mature than
for other platforms, and the goal here is to alert people that
eventually we plan to drop scons and move to meson, and that they should
try out meson and report issues.
Reviewed-by: Eric Anholt <eric@anholt.net>
At this point meson should be able to handle all of the non-windows
platforms just fine; we'd like to be able to stop maintaining scons for
those platforms sooner than later.
Reviewed-by: Eric Anholt <eric@anholt.net>
This ensures that we get python3's print() function behavior even in
python2, instead of python2's print statement behavior. We'll be using
this in the next patch.
Reviewed-by: Eric Anholt <eric@anholt.net>
On GFX8 the number of records is in bytes while on other chips
it's in units of "stride".
Fixes dEQP-VK.robustness.vertex_access.*.draw.vertex_* on RAVEN.
Tested on GFX6, GFX8, GFX10 and RAVEN.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Flagged by UBSan:
../src/util/tests/fast_idiv_by_const/fast_idiv_by_const_test.cpp:233:14: runtime error: negation of -2147483648 cannot be represented in type 'int'; cast to an unsigned type to negate this value to itself
#0 0x55b4c1a2a428 in rand_sint ../src/util/tests/fast_idiv_by_const/fast_idiv_by_const_test.cpp:233
#1 0x55b4c1a2ad3a in random_sdiv_test ../src/util/tests/fast_idiv_by_const/fast_idiv_by_const_test.cpp:308
#2 0x55b4c1a2b837 in fast_idiv_by_const_int32_Test::TestBody() ../src/util/tests/fast_idiv_by_const/fast_idiv_by_const_test.cpp:410
#3 0x55b4c1abc13f in void testing::internal::HandleSehExceptionsInMethodIfSupported<testing::Test, void>(testing::Test*, void (testing::Test::*)(), char const*) ../src/gtest/src/gtest.cc:2402
#4 0x55b4c1aa7a4d in void testing::internal::HandleExceptionsInMethodIfSupported<testing::Test, void>(testing::Test*, void (testing::Test::*)(), char const*) ../src/gtest/src/gtest.cc:2438
#5 0x55b4c1a4ce57 in testing::Test::Run() ../src/gtest/src/gtest.cc:2474
#6 0x55b4c1a4f530 in testing::TestInfo::Run() ../src/gtest/src/gtest.cc:2656
#7 0x55b4c1a51cbe in testing::TestCase::Run() ../src/gtest/src/gtest.cc:2774
#8 0x55b4c1a6d698 in testing::internal::UnitTestImpl::RunAllTests() ../src/gtest/src/gtest.cc:4649
#9 0x55b4c1abfd58 in bool testing::internal::HandleSehExceptionsInMethodIfSupported<testing::internal::UnitTestImpl, bool>(testing::internal::UnitTestImpl*, bool (testing::internal::UnitTestImpl::*)(), char const*) ../src/gtest/src/gtest.cc:2402
#10 0x55b4c1aab425 in bool testing::internal::HandleExceptionsInMethodIfSupported<testing::internal::UnitTestImpl, bool>(testing::internal::UnitTestImpl*, bool (testing::internal::UnitTestImpl::*)(), char const*) ../src/gtest/src/gtest.cc:2438
#11 0x55b4c1a64cba in testing::UnitTest::Run() ../src/gtest/src/gtest.cc:4257
#12 0x55b4c1ae4b73 in RUN_ALL_TESTS() ../src/gtest/include/gtest/gtest.h:2233
#13 0x55b4c1ae4a33 in main ../src/gtest/src/gtest_main.cc:37
#14 0x7ff172d1dbba in __libc_start_main ../csu/libc-start.c:308
#15 0x55b4c1a28dc9 in _start (/home/daenzer/src/mesa-git/mesa/build-amd64-sanitize/src/util/tests/fast_idiv_by_const/fast_idiv_by_const_test+0x96dc9)
../src/util/tests/fast_idiv_by_const/fast_idiv_by_const_test.cpp:309:52: runtime error: negation of -9223372036854775808 cannot be represented in type 'long int'; cast to an unsigned type to negate this value to itself
#0 0x563b24dafd2d in random_sdiv_test ../src/util/tests/fast_idiv_by_const/fast_idiv_by_const_test.cpp:309
#1 0x563b24db0f0f in fast_idiv_by_const_int64_Test::TestBody() ../src/util/tests/fast_idiv_by_const/fast_idiv_by_const_test.cpp:473
#2 0x563b24e41111 in void testing::internal::HandleSehExceptionsInMethodIfSupported<testing::Test, void>(testing::Test*, void (testing::Test::*)(), char const*) ../src/gtest/src/gtest.cc:2402
#3 0x563b24e2ca1f in void testing::internal::HandleExceptionsInMethodIfSupported<testing::Test, void>(testing::Test*, void (testing::Test::*)(), char const*) ../src/gtest/src/gtest.cc:2438
#4 0x563b24dd1e29 in testing::Test::Run() ../src/gtest/src/gtest.cc:2474
#5 0x563b24dd4502 in testing::TestInfo::Run() ../src/gtest/src/gtest.cc:2656
#6 0x563b24dd6c90 in testing::TestCase::Run() ../src/gtest/src/gtest.cc:2774
#7 0x563b24df266a in testing::internal::UnitTestImpl::RunAllTests() ../src/gtest/src/gtest.cc:4649
#8 0x563b24e44d2a in bool testing::internal::HandleSehExceptionsInMethodIfSupported<testing::internal::UnitTestImpl, bool>(testing::internal::UnitTestImpl*, bool (testing::internal::UnitTestImpl::*)(), char const*) ../src/gtest/src/gtest.cc:2402
#9 0x563b24e303f7 in bool testing::internal::HandleExceptionsInMethodIfSupported<testing::internal::UnitTestImpl, bool>(testing::internal::UnitTestImpl*, bool (testing::internal::UnitTestImpl::*)(), char const*) ../src/gtest/src/gtest.cc:2438
#10 0x563b24de9c8c in testing::UnitTest::Run() ../src/gtest/src/gtest.cc:4257
#11 0x563b24e69b45 in RUN_ALL_TESTS() ../src/gtest/include/gtest/gtest.h:2233
#12 0x563b24e69a05 in main ../src/gtest/src/gtest_main.cc:37
#13 0x7f9a90330bba in __libc_start_main ../csu/libc-start.c:308
#14 0x563b24daddc9 in _start (/home/daenzer/src/mesa-git/mesa/build-amd64-sanitize/src/util/tests/fast_idiv_by_const/fast_idiv_by_const_test+0x96dc9)
v2:
* Use INT64_MIN instead of LLONG_MIN (Jason Ekstrand)
* Simpler test for INT64_MIN result from rand_sint (Jason Ekstrand)
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Adam Jackson <ajax@redhat.com>
Shifting int64_t values left into the sign bit has undefined behaviour:
../src/util/fast_idiv_by_const.c:175:14: runtime error: left shift of 131 by 56 places cannot be represented in type 'long int'
#0 0x561337ed10c1 in sign_extend ../src/util/fast_idiv_by_const.c:175
#1 0x561337ed1335 in util_compute_fast_sdiv_info ../src/util/fast_idiv_by_const.c:239
#2 0x561337e17519 in fast_idiv_by_const_int8_Test::TestBody() ../src/util/tests/fast_idiv_by_const/fast_idiv_by_const_test.cpp:357
#3 0x561337ea815d in void testing::internal::HandleSehExceptionsInMethodIfSupported<testing::Test, void>(testing::Test*, void (testing::Test::*)(), char const*) ../src/gtest/src/gtest.cc:2402
#4 0x561337e93a6b in void testing::internal::HandleExceptionsInMethodIfSupported<testing::Test, void>(testing::Test*, void (testing::Test::*)(), char const*) ../src/gtest/src/gtest.cc:2438
#5 0x561337e38e75 in testing::Test::Run() ../src/gtest/src/gtest.cc:2474
#6 0x561337e3b54e in testing::TestInfo::Run() ../src/gtest/src/gtest.cc:2656
#7 0x561337e3dcdc in testing::TestCase::Run() ../src/gtest/src/gtest.cc:2774
#8 0x561337e596b6 in testing::internal::UnitTestImpl::RunAllTests() ../src/gtest/src/gtest.cc:4649
#9 0x561337eabd76 in bool testing::internal::HandleSehExceptionsInMethodIfSupported<testing::internal::UnitTestImpl, bool>(testing::internal::UnitTestImpl*, bool (testing::internal::UnitTestImpl::*)(), char const*) ../src/gtest/src/gtest.cc:2402
#10 0x561337e97443 in bool testing::internal::HandleExceptionsInMethodIfSupported<testing::internal::UnitTestImpl, bool>(testing::internal::UnitTestImpl*, bool (testing::internal::UnitTestImpl::*)(), char const*) ../src/gtest/src/gtest.cc:2438
#11 0x561337e50cd8 in testing::UnitTest::Run() ../src/gtest/src/gtest.cc:4257
#12 0x561337ed0b91 in RUN_ALL_TESTS() ../src/gtest/include/gtest/gtest.h:2233
#13 0x561337ed0a51 in main ../src/gtest/src/gtest_main.cc:37
#14 0x7f85ba483bba in __libc_start_main ../csu/libc-start.c:308
#15 0x561337e14dc9 in _start (/home/daenzer/src/mesa-git/mesa/build-amd64-sanitize/src/util/tests/fast_idiv_by_const/fast_idiv_by_const_test+0x96dc9)
../src/util/tests/fast_idiv_by_const/fast_idiv_by_const_test.cpp:51:14: runtime error: left shift of negative value -63
#0 0x55fc3c0e67cc in strunc ../src/util/tests/fast_idiv_by_const/fast_idiv_by_const_test.cpp:51
#1 0x55fc3c0e6d93 in smul_high ../src/util/tests/fast_idiv_by_const/fast_idiv_by_const_test.cpp:140
#2 0x55fc3c0e7067 in fast_sdiv ../src/util/tests/fast_idiv_by_const/fast_idiv_by_const_test.cpp:181
#3 0x55fc3c0e858b in fast_idiv_by_const_int8_Test::TestBody() ../src/util/tests/fast_idiv_by_const/fast_idiv_by_const_test.cpp:358
#4 0x55fc3c17915d in void testing::internal::HandleSehExceptionsInMethodIfSupported<testing::Test, void>(testing::Test*, void (testing::Test::*)(), char const*) ../src/gtest/src/gtest.cc:2402
#5 0x55fc3c164a6b in void testing::internal::HandleExceptionsInMethodIfSupported<testing::Test, void>(testing::Test*, void (testing::Test::*)(), char const*) ../src/gtest/src/gtest.cc:2438
#6 0x55fc3c109e75 in testing::Test::Run() ../src/gtest/src/gtest.cc:2474
#7 0x55fc3c10c54e in testing::TestInfo::Run() ../src/gtest/src/gtest.cc:2656
#8 0x55fc3c10ecdc in testing::TestCase::Run() ../src/gtest/src/gtest.cc:2774
#9 0x55fc3c12a6b6 in testing::internal::UnitTestImpl::RunAllTests() ../src/gtest/src/gtest.cc:4649
#10 0x55fc3c17cd76 in bool testing::internal::HandleSehExceptionsInMethodIfSupported<testing::internal::UnitTestImpl, bool>(testing::internal::UnitTestImpl*, bool (testing::internal::UnitTestImpl::*)(), char const*) ../src/gtest/src/gtest.cc:2402
#11 0x55fc3c168443 in bool testing::internal::HandleExceptionsInMethodIfSupported<testing::internal::UnitTestImpl, bool>(testing::internal::UnitTestImpl*, bool (testing::internal::UnitTestImpl::*)(), char const*) ../src/gtest/src/gtest.cc:2438
#12 0x55fc3c121cd8 in testing::UnitTest::Run() ../src/gtest/src/gtest.cc:4257
#13 0x55fc3c1a1b91 in RUN_ALL_TESTS() ../src/gtest/include/gtest/gtest.h:2233
#14 0x55fc3c1a1a51 in main ../src/gtest/src/gtest_main.cc:37
#15 0x7fd224759bba in __libc_start_main ../csu/libc-start.c:308
#16 0x55fc3c0e5dc9 in _start (/home/daenzer/src/mesa-git/mesa/build-amd64-sanitize/src/util/tests/fast_idiv_by_const/fast_idiv_by_const_test+0x96dc9)
v2:
* Use two casts instead of changing the argument type (Jason Ekstrand)
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Adam Jackson <ajax@redhat.com>
Otherwise a smaller type may be promoted to int, which can hit undefined
behaviour:
../src/gallium/auxiliary/util/u_half.h:126:29: runtime error: left shift of 32768 by 16 places cannot be represented in type 'int'
#0 0x5646ff63d488 in util_half_to_float ../src/gallium/auxiliary/util/u_half.h:126
#1 0x5646ff63d749 in _mesa_half_to_float ../src/util/half_float.c:145
#2 0x5646ff54d557 in nir_const_value_negative_equal ../src/compiler/nir/nir_instr_set.c:372
#3 0x5646ff44d29a in const_value_negative_equal_test_nir_type_float16_trivially_true_Test::TestBody() ../src/compiler/nir/tests/negative_equal_tests.cpp:121
#4 0x5646ff505c05 in void testing::internal::HandleSehExceptionsInMethodIfSupported<testing::Test, void>(testing::Test*, void (testing::Test::*)(), char const*) ../src/gtest/src/gtest.cc:2402
#5 0x5646ff4f1513 in void testing::internal::HandleExceptionsInMethodIfSupported<testing::Test, void>(testing::Test*, void (testing::Test::*)(), char const*) ../src/gtest/src/gtest.cc:2438
#6 0x5646ff4979b5 in testing::Test::Run() ../src/gtest/src/gtest.cc:2474
#7 0x5646ff49a08e in testing::TestInfo::Run() ../src/gtest/src/gtest.cc:2656
#8 0x5646ff49c81c in testing::TestCase::Run() ../src/gtest/src/gtest.cc:2774
#9 0x5646ff4b81f6 in testing::internal::UnitTestImpl::RunAllTests() ../src/gtest/src/gtest.cc:4649
#10 0x5646ff50981e in bool testing::internal::HandleSehExceptionsInMethodIfSupported<testing::internal::UnitTestImpl, bool>(testing::internal::UnitTestImpl*, bool (testing::internal::UnitTestImpl::*)(), char const*) ../src/gtest/src/gtest.cc:2402
#11 0x5646ff4f4eeb in bool testing::internal::HandleExceptionsInMethodIfSupported<testing::internal::UnitTestImpl, bool>(testing::internal::UnitTestImpl*, bool (testing::internal::UnitTestImpl::*)(), char const*) ../src/gtest/src/gtest.cc:2438
#12 0x5646ff4af818 in testing::UnitTest::Run() ../src/gtest/src/gtest.cc:4257
#13 0x5646ff52e639 in RUN_ALL_TESTS() ../src/gtest/include/gtest/gtest.h:2233
#14 0x5646ff52e4f9 in main ../src/gtest/src/gtest_main.cc:37
#15 0x7f6bacb78bba in __libc_start_main ../csu/libc-start.c:308
#16 0x5646ff448019 in _start (/home/daenzer/src/mesa-git/mesa/build-amd64-sanitize/src/compiler/nir/negative_equal+0x17c019)
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Adam Jackson <ajax@redhat.com>
Otherwise a smaller type may be promoted to int, which can hit undefined
behaviour:
../src/intel/compiler/brw_packed_float.c:66:17: runtime error: left shift of 128 by 24 places cannot be represented in type 'int'
#0 0x5604a03969aa in brw_vf_to_float ../src/intel/compiler/brw_packed_float.c:66
#1 0x5604a0391305 in vf_float_conversion_test_test_vf_to_float_Test::TestBody() ../src/intel/compiler/test_vf_float_conversions.cpp:70
#2 0x5604a041a323 in void testing::internal::HandleSehExceptionsInMethodIfSupported<testing::Test, void>(testing::Test*, void (testing::Test::*)(), char const*) ../src/gtest/src/gtest.cc:2402
#3 0x5604a0405c31 in void testing::internal::HandleExceptionsInMethodIfSupported<testing::Test, void>(testing::Test*, void (testing::Test::*)(), char const*) ../src/gtest/src/gtest.cc:2438
#4 0x5604a03ab03b in testing::Test::Run() ../src/gtest/src/gtest.cc:2474
#5 0x5604a03ad714 in testing::TestInfo::Run() ../src/gtest/src/gtest.cc:2656
#6 0x5604a03afea2 in testing::TestCase::Run() ../src/gtest/src/gtest.cc:2774
#7 0x5604a03cb87c in testing::internal::UnitTestImpl::RunAllTests() ../src/gtest/src/gtest.cc:4649
#8 0x5604a041df3c in bool testing::internal::HandleSehExceptionsInMethodIfSupported<testing::internal::UnitTestImpl, bool>(testing::internal::UnitTestImpl*, bool (testing::internal::UnitTestImpl::*)(), char const*) ../src/gtest/src/gtest.cc:2402
#9 0x5604a0409609 in bool testing::internal::HandleExceptionsInMethodIfSupported<testing::internal::UnitTestImpl, bool>(testing::internal::UnitTestImpl*, bool (testing::internal::UnitTestImpl::*)(), char const*) ../src/gtest/src/gtest.cc:2438
#10 0x5604a03c2e9e in testing::UnitTest::Run() ../src/gtest/src/gtest.cc:4257
#11 0x5604a0442d57 in RUN_ALL_TESTS() ../src/gtest/include/gtest/gtest.h:2233
#12 0x5604a0442c17 in main ../src/gtest/src/gtest_main.cc:37
#13 0x7f9a1983dbba in __libc_start_main ../csu/libc-start.c:308
#14 0x5604a0390d89 in _start (/home/daenzer/src/mesa-git/mesa/build-amd64-sanitize/src/intel/compiler/vf_float_conversions+0x8dd89)
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Adam Jackson <ajax@redhat.com>
To avoid it, use the modulo of the number of bits in the value being
shifted, which is presumably what ended up happening on x86.
Flagged by UBSan:
../src/intel/compiler/brw_eu_validate.c:974:33: runtime error: shift exponent 64 is too large for 64-bit type 'long unsigned int'
#0 0x561abb612ab3 in general_restrictions_on_region_parameters ../src/intel/compiler/brw_eu_validate.c:974
#1 0x561abb617574 in brw_validate_instructions ../src/intel/compiler/brw_eu_validate.c:1851
#2 0x561abb53bd31 in validate ../src/intel/compiler/test_eu_validate.cpp:106
#3 0x561abb555369 in validation_test_source_cannot_span_more_than_2_registers_Test::TestBody() ../src/intel/compiler/test_eu_validate.cpp:486
#4 0x561abb742651 in void testing::internal::HandleSehExceptionsInMethodIfSupported<testing::Test, void>(testing::Test*, void (testing::Test::*)(), char const*) ../src/gtest/src/gtest.cc:2402
#5 0x561abb72e64d in void testing::internal::HandleExceptionsInMethodIfSupported<testing::Test, void>(testing::Test*, void (testing::Test::*)(), char const*) ../src/gtest/src/gtest.cc:2438
#6 0x561abb6d5451 in testing::Test::Run() ../src/gtest/src/gtest.cc:2474
#7 0x561abb6d7b2a in testing::TestInfo::Run() ../src/gtest/src/gtest.cc:2656
#8 0x561abb6da2b8 in testing::TestCase::Run() ../src/gtest/src/gtest.cc:2774
#9 0x561abb6f5c92 in testing::internal::UnitTestImpl::RunAllTests() ../src/gtest/src/gtest.cc:4649
#10 0x561abb74626a in bool testing::internal::HandleSehExceptionsInMethodIfSupported<testing::internal::UnitTestImpl, bool>(testing::internal::UnitTestImpl*, bool (testing::internal::UnitTestImpl::*)(), char const*) ../src/gtest/src/gtest.cc:2402
#11 0x561abb732025 in bool testing::internal::HandleExceptionsInMethodIfSupported<testing::internal::UnitTestImpl, bool>(testing::internal::UnitTestImpl*, bool (testing::internal::UnitTestImpl::*)(), char const*) ../src/gtest/src/gtest.cc:2438
#12 0x561abb6ed2b4 in testing::UnitTest::Run() ../src/gtest/src/gtest.cc:4257
#13 0x561abb768b3b in RUN_ALL_TESTS() ../src/gtest/include/gtest/gtest.h:2233
#14 0x561abb7689fb in main ../src/gtest/src/gtest_main.cc:37
#15 0x7f525e5a9bba in __libc_start_main ../csu/libc-start.c:308
#16 0x561abb538ed9 in _start (/home/daenzer/src/mesa-git/mesa/build-amd64-sanitize/src/intel/compiler/eu_validate+0x1b8ed9)
Reviewed-by: Adam Jackson <ajax@redhat.com>
`strerror()` takes an `errno`, not the negative value returned by the
`ioctl()`.
Instead of fixing this as `"%s", strerror(errno)`, let's just use the
`"%m"` shortcut for it.
Fixes: 2b5f30b1d9 ("anv: implement VK_INTEL_performance_query")
Signed-off-by: Eric Engestrom <eric.engestrom@intel.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
The kernel total GMR/DMA size is limited, but it's definitely possible for the
kernel to allow a larger buffer allocation to succeed, but command
submission using that buffer as a GMR would fail typically causing an
application crash.
So have the winsys limit the size of GMR/DMA buffers. The pipe driver will
then resort to allocating smaller buffers and perform the DMA transfer in
multiple bands, also allowing for the pre-flush mechanism to kick in.
This avoids the related application crashes.
Fixes: e7843273fa ("winsys/svga: Update to vmwgfx kernel module 2.1")
Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Charmaine Lee <charmainel@vmware.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Even with banded DMA uploads, st->hwbuf is always non-NULL, but when we've
allocated a software buffer to hold the full upload, unmapping of the
hardware buffer has already been done before
svga_texture_transfer_unmap_dma(), and the code was performing an unmap of
an already mapped buffer.
Fix this by testing for software buffer not present.
Fixes: a9c4a861d5 ("svga: refactor svga_texture_transfer_map/unmap functions")
Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Charmaine Lee <charmainel@vmware.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Update to 5.4-rc4 so we can test Panfrost on devices with Mali T720 and
T820.
A bug was found that prevented things working at all on RK3288 devices,
so we carry a patch for now in my personal fork.
Signed-off-by: Tomeu Vizoso <tomeu.vizoso@collabora.com>
Acked-by: Daniel Stone <daniels@collabora.com>
If there are queued shaders to be written to disk, wait for that.
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
We were relying on specific pass ordering in st to avoid setting
inputs_read/outputs_written for edge flags. Instead, just assume
that it happens and throw out the results we don't want.
We should probably revisit this and try and add a vertex element
property like I originally wanted so we can avoid having it be
associated with the VS altogether.
I recently changed the slow depth/stencil clear path to make sure
depth values are explicitly exported by the fragment shader. This
is actually only useful when VK_EXT_depth_range_unrestricted is
enabled.
While this path is correct, it introduced a performance regression
with Heroes of the Storm, Shadow of Mordor (Vulkan beta) and
probably more titles. This is because it prevents the hardware
to do some optimizations like discarding fragments.
This commit re-introduces the previous (a bit faster) slow
depth/stencil clear path and it selects the unrestricted path
only if VK_EXT_depth_range_unrestricted is enabled.
Closes: https://gitlab.freedesktop.org/mesa/mesa/issues/863
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
descriptorCount is the number of bytes into the descriptor, so
it shouldn't be used as an index. srcArrayElement/dstArrayElement
specify the starting byte offset within the binding to copy from/to.
This fixes new CTS tests:
dEQP-VK.binding_model.descriptor_copy.*.inline_uniform_block_*
dEQP-VK.binding_model.descriptor_copy.*.mix_3
dEQP-VK.binding_model.descriptor_copy.*.mix_array1
Fixes: 8d2654a419 ("radv: Support VK_EXT_inline_uniform_block.")
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
It used to cause weird issues on GFX10 in the past with vkmark and
Wreckfest, and they can't be reproduced now. Shadow Of Mordor
(Vulkan beta) hits that path and it works fine.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Fixes some crashes with dEQP-VK.geometry.layered.*.secondary_cmd_buffer
on Raven and other chips that allow rbplus.
This just prevents a crash and rbplus probaby needs more work.
Cc: 19.2 <mesa-stable@lists.freedesktop.org>
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
The driver only supports up to 8 samples, so it's useless to
create more pipelines than needed.
This fixes a conditional jump reported by Valgrind on GFX10:
==194282== Conditional jump or move depends on uninitialised value(s)
==194282== at 0xDBF925A: radv_gfx10_compute_bin_size (radv_pipeline.c:3242)
==194282== by 0xDBF95A6: radv_pipeline_generate_binning_state (radv_pipeline.c:3334)
==194282== by 0xDBFC1A0: radv_pipeline_generate_pm4 (radv_pipeline.c:4440)
==194282== by 0xDBFD15E: radv_pipeline_init (radv_pipeline.c:4764)
==194282== by 0xDBFD23E: radv_graphics_pipeline_create (radv_pipeline.c:4788)
==194282== by 0xDBB95A3: create_pipeline (radv_meta_clear.c:114)
==194282== by 0xDBB9AC5: create_color_pipeline (radv_meta_clear.c:297)
==194282== by 0xDBBCF05: radv_device_init_meta_clear_state (radv_meta_clear.c:1277)
==194282== by 0xDB9ACD9: radv_device_init_meta (radv_meta.c:363)
==194282== by 0xDB7FE3A: radv_CreateDevice (radv_device.c:2080
This is caused by an out of bound access of 'fmask_array' (ie. index
is 4 as for 16 samples).
Cc: <mesa-stable@lists.freedesktop.org>
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
v2: Introduce the appropriate pipe controls
Properly deal with changes in metric sets (using execbuf parameter)
Record marker at query end
v3: Fill out PerfCntr1&2
v4: Introduce vkUninitializePerformanceApiINTEL
v5: Use new execbuf extension mechanism
v6: Fix comments in genX_query.c (Rafael)
Use PIPE_CONTROL workarounds (Rafael)
Refactor on the last kernel series update (Lionel)
v7: Only I915_PERF_IOCTL_CONFIG when perf stream is already opened (Lionel)
Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Rafael Antognolli <rafael.antognolli@intel.com>
We have 2 of those we can configure to source programmable events.
Those are not part of the OA reports. Configuration happens in i915
through the metric set selected by the application. On the Mesa side
we'll just sample those and do a diff.
Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Rafael Antognolli <rafael.antognolli@intel.com>
Pull new updates from drm-next as of the following commit:
commit f1b4a9217efd61d0b84c6dc404596c8519ff6f59
Merge: 400e91347e1d f3a36d469621
Author: Dave Airlie <airlied@redhat.com>
Date: Tue Oct 22 15:04:00 2019 +1000
Merge tag 'du-next-20191016' of git://linuxtv.org/pinchartl/media into drm-next
Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
We want to query the content of register configurations from the
kernel. Let's pull this out of the query.
Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Rafael Antognolli <rafael.antognolli@intel.com>
The Vulkan performance query extension is a bit lower level than the
GL one. Expose some of the functions to do the result accumulation
directly in the Anv driver.
Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Rafael Antognolli <rafael.antognolli@intel.com>
This is useful for PBO texture upload with GL_RGB and GL_UNSIGNED_BYTE.
v2: Vasily Khoruzhick provided an update for the Lima CI expectations.
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
p_extract_vector's second operand is in units of the definition size, not
dwords.
v2: move extract_subvector() to right before ds_write_helper
Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Small typo resulted in not converting footprint to vec4, meaning that we
could potentially ask for quite a few more registers than required
Signed-off-by: Rob Clark <robdclark@chromium.org>
Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
If the load_interpolated_input is scalarized, we would be too
conservative about deciding the tex instruction wasn't a candidate to
pre-fetch:
vec1 32 ssa_0 = load_const (0x00000000 /* 0.000000 */)
vec2 32 ssa_1 = intrinsic load_barycentric_pixel () (0) /* interp_mode=0 */
vec1 32 ssa_2 = intrinsic load_interpolated_input (ssa_1, ssa_0) (0, 0) /* base=0 */ /* component=0 */ /* packed:v_uv,v_uv1 */
vec1 32 ssa_3 = intrinsic load_interpolated_input (ssa_1, ssa_0) (0, 1) /* base=0 */ /* component=1 */ /* packed:v_uv,v_uv1 */
vec2 32 ssa_8 = vec2 ssa_2, ssa_3
vec4 32 ssa_9 = tex ssa_8 (coord), 0 (texture), 0 (sampler)
Really we don't care that the texcoord components come from different
load_interpolated_input instructions, just that they have consecutive
varying offsets.
Reported-by: Eduardo Lima Mitev <elima@igalia.com>
Reviewed-by: Eduardo Lima Mitev <elima@igalia.com>
Signed-off-by: Rob Clark <robdclark@chromium.org>
Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
Previously, we used one hashset per BB, so that we could
always initialize the current hashset from the immediate
dominator. This patch changes the behavior to a single
hashmap using the block index per instruction to resolve
dominance.
Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Some of these lowerings aren't supported for drivers that supports
tesselation and geometry shaders. Let's add a couple of asserts to make
it obvious if these have been enabled when it's not possible.
Signed-off-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
v2:
* Use LLVM 8 from buster-backports
v3:
* Use LLVM 7 again for armhf, llvmpipe is still broken there with LLVM 8
Acked-by: Eric Engestrom <eric.engestrom@intel.com>
This allows running the regression tests.
One downside is that we can't easily build the Vulkan overlay layer,
because only x86 binaries of the glslang validator are available. If
that's important, we could either use those binaries via qemu, or build
it from source.
v2:
* Add :amd64 suffix to existing debian-9/10 job names (Eric Engestrom)
Acked-by: Eric Engestrom <eric.engestrom@intel.com> # v1
Apparently needs: in a definition overwrites inherited ones. So
.deqp-test effectively didn't declare needs: for debian-10, which means
any jobs based on .deqp-test could spuriously run after the debian-10
job failed or was cancelled.
Use https:// URLs in the APT configuration.
Drop --no-install-recommends, the image generation template disables
installation of recommended packages in /etc/apt/apt.conf.
Run apt-get autoremove at the end, cleaning up packages which were
installed to satisfy dependencies but are no longer needed.
Reviewed-by: Eric Engestrom <eric.engestrom@intel.com>
On GFX9, the driver is able to do an optimized fast depth/stencil
clear with only one aspect (ie. clear the stencil part of a
depth/stencil image). When this happens, the driver should only
update the clear values of the given aspect.
Note that it's currently only supported on GFX9 but I have some
local patches that extend this optimized path for other gens.
Closes: https://gitlab.freedesktop.org/mesa/mesa/issues/1967
Cc: 19.2 <mesa-stable@lists.freedesktop.org>
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
On Gen12, we support mixed mode HF/F operands, and also 3 source
instruction supports immediate value support, so keep immediate as it
is, if it fits properly in 16 bit field.
Signed-off-by: Sagar Ghuge <sagar.ghuge@intel.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
On Gen >= 12, if src0 or src2 holds immediate value, we need set
src[0/2]_is_imm bits instead of register file.
Signed-off-by: Sagar Ghuge <sagar.ghuge@intel.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
On Gen >= 10, Either src0 or src2 can use 16-bit immediate value, but
not both.
Signed-off-by: Sagar Ghuge <sagar.ghuge@intel.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
It's been throwing the following error today:
"<Fault -32603: 'Internal Server Error (contact server administrator
for details): could not extend file "base/17952/18226": No space left
on device\nHINT: Check free disk space.\n'>"
Reviewed-by: Daniel Stone <daniels@collabora.com>
If you set LP_NUM_THREADS=0 compute shaders would hang,
just execute the workloads in sequence if we have no threads
in the pool.
Fixes: 1b24e3ba75 ("llvmpipe: add compute threadpool + mutex")
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
This file is created in 2a0d45ae6c but
addition to android makefiles was omitted. It breaks the build with
missing references which are defined in this file.
List the file in ir3_SOURCES to make the build succeed.
Signed-off-by: Marijn Suijten <marijns95@gmail.com>
This fixes some crashes with dEQP-VK.descriptor_indexing.* when
read_first_invocation has its source from a descriptor.
Most of these tests still fail because of an LLVM bug (they work
with ACO).
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
v2: make variable names snake_case
v2: minor cleanups in emit_udiv()
v2: fix Panfrost build failure
v3: use an enum instead of a boolean flag in nir_lower_idiv()'s signature
v4: remove nir_op_urcp
v5: drop nv50 path
v5: rebase
v6: add back nv50 path
v6: add comment for nir_lower_idiv_path enum
v7: rename _nv50/_llvm to _fast/_precise
v8: fix etnaviv build failure
Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Remove emit_alpha_to_coverage workaround from backend compiler and start
using ported workaround from NIR.
v2: Copy comment from brw_fs_visitor (Caio Marcelo de Oliveira Filho)
Fixes piglit test on HSW:
- arb_sample_shading-builtin-gl-sample-mask-mrt-alpha-to-coverage-combinations
Signed-off-by: Sagar Ghuge <sagar.ghuge@intel.com>
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Importing this pass from fs_visitor::emit_alpha_to_coverage_workaround()
in intel/compiler.
v2 (Caio Marcelo de Oliveira Filho):
- Track store output and sample mask instruction
- Nest math insturction for more readability
- Bail out early if no gl_SampleMask
v3: (Caio Marcelo de Oliveira Filho):
- Do math instructions after instruction block
- Restructure code
- Move pass under src/intel/compiler
v4: (Caio Marcelo de Oliveira Filho):
- Organize dither mask calculation
Signed-off-by: Sagar Ghuge <sagar.ghuge@intel.com>
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
0.49.0 can compile most of mesa with ICC or ICL, but not SWR without
additional workarounds in our meson.build files. Bumping patch version
is easier and shouldn't be a big burden anyway, especially to cover a
niche compiler. The check originally only covered ICC, but now covers
ICL as well.
Fixes: 3740ffb59c
("meson: add switches for SWR with MSVC")
Closes: https://gitlab.freedesktop.org/mesa/mesa/issues/1937
Acked-by: Eric Engestrom <eric.engestrom@intel.com>
Due to a bug in GFX10 hardware, s_nop instructions must be added
if a branch is at 0x3f. We already do this, but forgot to also update
the constant addresses that come after this instruction.
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Currently if you have an SMEM store followed by an SMEM load that
loads the same location as was written, it won't work because the
store isn't finished before the load is executed. This is NOT
mitigated by an s_nop instruction on GFX10.
Since we currently don't have proper alias analysis, this commit adds
a workaround which will insert an s_waitcnt lgkmcnt(0) before each
SSBO load if they follow a store. We should further refine this in
the future when we can make sure to only add the wait when we load the
same thing as has been stored.
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
The current implementation does not synchronize on BO readiness when
DISCARD_WHOLE_RES flag is set, which can lead to misbehaviours when the
resource being updated is being used by one of the pending or already
flushed batches.
Adding unconditional BO synchronization would do the trick, but we can
sometimes optimize this path by re-allocating a new BO instead of
waiting for the existing one to be ready.
Reported-by: Daniel Stone <daniels@collabora.com>
Reported-by: Heinrich Fink <heinrich.fink@daqri.com>
Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Daniel Stone <daniels@collabora.com>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
According to the OES_geometry_shader spec, section Dependencies:
"OpenGL ES 3.1 and OpenGL ES Shading Language 3.10
are required."
Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
We currently doesn't maintain it correctly and the buffer gets leaked if
surface is destroyed before calling swapping buffers.
From Android frameworks/native/libs/nativewindow/include/system/window.h:
The window holds a reference to the buffer between dequeueBuffer and
either queueBuffer or cancelBuffer, so clients only need their own
reference if they might use the buffer after queueing or canceling it.
v2: Remove our own reference.
Fixes: 0212db3504 ("egl/android: Cancel any outstanding ANativeBuffer in surface destructor")
Reviewed-by: Chia-I Wu <olvaffe@gmail.com> (v1)
Reviewed-By: Tapani Pälli <tapani.palli@intel.com>
Signed-off-by: Lepton Wu <lepton@chromium.org>
If a pipeline has both graphics and compute, descriptors are same.
While we are at it, use queue->device for simplicity.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
To make sure a trace file is generated in case the driver crashes
during the hang report generation (which happens sometimes).
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
This information has never been useful. All descriptors are
already dumped with colors etc, and it's more useful.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
A bunch of blend tests fixed on T760. A single blend test regressed on
both T760/T860 but I am unable to reproduce locally so am just
documenting the regression and moving on.
Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
We would like to eliminate not just entire dead instructions, but also
dead components, which increases scheduler flexibility (since some
vector instructions can become scalar after eliminating dead
components). This also will allow better RA in the future.
Results are meh.
total instructions in shared programs: 3453 -> 3451 (-0.06%)
instructions in affected programs: 60 -> 58 (-3.33%)
helped: 2
HURT: 0
total bundles in shared programs: 1826 -> 1824 (-0.11%)
bundles in affected programs: 33 -> 31 (-6.06%)
helped: 2
HURT: 0
total quadwords in shared programs: 3144 -> 3144 (0.00%)
quadwords in affected programs: 0 -> 0
helped: 0
HURT: 0
total registers in shared programs: 321 -> 321 (0.00%)
registers in affected programs: 45 -> 45 (0.00%)
helped: 11
HURT: 11
helped stats (abs) min: 1 max: 1 x̄: 1.00 x̃: 1
helped stats (rel) min: 16.67% max: 50.00% x̄: 39.70% x̃: 50.00%
HURT stats (abs) min: 1 max: 1 x̄: 1.00 x̃: 1
HURT stats (rel) min: 100.00% max: 100.00% x̄: 100.00% x̃: 100.00%
95% mean confidence interval for registers value: -0.45 0.45
95% mean confidence interval for registers %-change: -1.87% 62.18%
Inconclusive result (value mean confidence interval includes 0).
total threads in shared programs: 445 -> 447 (0.45%)
threads in affected programs: 2 -> 4 (100.00%)
helped: 1
HURT: 0
Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
This allows for vec16 dependencies in the scheduler, not that we have
any yet (thankfully).
Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Now that we have notion of byte masks, liveness tracking can be updated
to reflect this extra granularity without loss of correctness.
Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Read component masks don't have a particular type associated, since the
type of the ALU operation may not match the type of the operands in
question. So let's generate byte masks instead, and update the rest of
the compiler to use byte masks when analyzing reads.
Preparation for mixed types.
Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
There are essentially two formats of masks in play beginning with this
commit: masks per-channel and masks per-byte. The former make sense
within a given fixed-size instruction; the latter are
typesize-independent. It turns out you need the latter to meaningfully
manipulate instructions containing multiple sizes (which is quite
possible with ALU operations).
Similarly, we have mir_srcsize. We calculate the size of the source by
analyzing the size of the instruction itself and stepping down if there
is a half-modifier.
Finally, we have mir_round_bytemask_down, for when we want to take a
byte mask and "round it down" to a given component size, so that we can
use it as a component mask.
Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
This will allow us to encode properties about the load/store ops like we
do for ALU ops. We include now properties about whether we have a store,
and if there are special cases on the load/store op. We also tag each
instruction by its natural size... this is probably not totally right,
but it's a start.
Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
The trick is realizing even with a destination override, the masks are encoded in the same mode as the
instruction itself, rather than stepping down. The override means that
the smaller type is used, but the mask is parsed as if it were the
higher type. Overriding down is down by printed by blinding doing this. Overriding up can be thought of as printing in the upper size, but shifting the alphabet to use the upper half, i.e. shifting xyzw to become abcd.
Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Add some comments explaining what's going on in a more natural flow in
order to solve the actual bug.
Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Fixes: 2d914ebe81 ("pan/midgard: Fix memory corruption in register spilling")
This allows a write to proceed to an uninitialized part of a buffer
even when the GPU is using the previously-initialized portions.
Such a situation can be triggered with the following API usage example:
glBufferSubData(..., offset, size, data1);
glDrawArrays(...);
// append new vertex data
glBufferSubData(..., offset+size, size, data2);
glDrawArrays(...);
Same is done for freedreno, nouveau and radeon.
Signed-off-by: Christian Gmeiner <christian.gmeiner@gmail.com>
Reviewed-by: Lucas Stach <l.stach@pengutronix.de>
This is the layout used in the GL API, and maps directly to PIPE
formats with no endianness trickery. As with the LA change, this
fixes big-endian fetching from texbos. Also cleans up some endian
shenanigans in shader images.
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Now that Mesa is also using an array format for LA, nothing was using
these. (And, clearly, no HW driver had exposed them).
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
The array format is what the GL API wants (fixing texbos on
big-endian), and matches directly to gallium's corresponding array
format. The only driver exposing A8L8 was radeon/r200 in big-endian,
where the HW's underlying format was trying to read as array and we
needed to flip things around to make our packed format come out right
(note that while the radeon format tables had both AL and LA,
ChooseTextureFormat would only pick one of them based on endianness).
v2: Don't make r200/radeon use endian swaps.
v3: Rebase on dropping the r200 _be/_le format table removal patch
v4: reword commit message to explain why we can drop both formats
from radeon.
Reviewed-by: Marek Olšák <marek.olsak@amd.com> (v1)
The array format is what the GL API wants (and we made a mistake in
the format returned for texbos on big-endian!), and it's exactly what
the gallium-side PIPE_FORMAT_L16A16 is. The only downside is that
dri_util tries to fall back to sampling RG16 using LA16, which doesn't
have a match for big-endian any more. No HW drivers supported A16L16
anyway.
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
The first arg to OUT_BATCH_RELOC is ignored, we actually wanted these
in the third arg. They're always 0 so far, so it didn't matter.
v2: Reword commit message that I don't end up using the tile bits, but
keep the commit as a cleanup anyway.
Reviewed-by: Marek Olšák <marek.olsak@amd.com> (v1)
No matter what, we deref the texFormat from the table, except for a
mistake in cpp=4 where we pulled a 0 out of the table either way.
v2: Rebase on dropping r200 table deduplication patch.
Reviewed-by: Marek Olšák <marek.olsak@amd.com> (v1)
PP stack size should be set to maximum PP stack size, not to stack size of
last shader.
Fixes: 27e7603c34 ("lima: fix ppir spill stack allocation")
Tested-by: Icenowy Zheng <icenowy@aosc.io>
Reviewed-by: Erico Nunes <nunes.erico@gmail.com>
Signed-off-by: Vasily Khoruzhick <anarsoul@gmail.com>
Similiar to iadd, we can fold an added constant value from an imad24_ir3
into the load_uniform's constant offset. This avoids some cases where
the addition of imad24_ir3 could otherwise be a regression in instr
count.
Signed-off-by: Rob Clark <robdclark@chromium.org>
Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
Reviewed-by: Eduardo Lima Mitev <elima@igalia.com>
We can't encode immed sources for cat3 (mad) instructions, but we can
use const in first or third src. We handled this case already, but we
weren't considering that we could lower immed to const.
For manhattan:
total instructions in shared programs: 35202 -> 34718 (-1.37%)
instructions in affected programs: 14931 -> 14447 (-3.24%)
helped: 90
HURT: 0
total full in shared programs: 2451 -> 2359 (-3.75%)
full in affected programs: 653 -> 561 (-14.09%)
helped: 69
HURT: 2
Signed-off-by: Rob Clark <robdclark@chromium.org>
Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
Lower amul to either imul or imul24, depending on whether 24b is enough
bits to calculate an offset within the thing being dereferenced.
Signed-off-by: Rob Clark <robdclark@chromium.org>
Used for address/offset calculation (ie. array derefs), where we can
potentially use less than 32b for the multiply of array idx by element
size. For backends that support `imul24`, this gives a lowering pass
an easy way to find multiplies that potentially can be converted to
`imul24`.
Signed-off-by: Rob Clark <robdclark@chromium.org>
Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
Reviewed-by: Eduardo Lima Mitev <elima@igalia.com>
Some hardware can do 24b multiply in a single instruction, but not 32b.
However in most cases 24b is sufficient for address/offset calculation.
Signed-off-by: Rob Clark <robdclark@chromium.org>
Acked-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
Reviewed-by: Eduardo Lima Mitev <elima@igalia.com>
ir3 compiler has a signed integer multiply-add instruction (MAD_S24)
that is used for different offset calculations in the backend.
Since we intend to move some of these calculations to NIR, we need
a new ALU op that can directly represent it.
Signed-off-by: Rob Clark <robdclark@chromium.org>
Acked-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
Reviewed-by: Eduardo Lima Mitev <elima@igalia.com>
Otherwise, if the base type is (for example) uint32, we would
incorrectly think that PoT optimizations could not apply.
Signed-off-by: Rob Clark <robdclark@chromium.org>
Reviewed-by: Jason Ekstsrand <jason@jleksrand.net>
Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
Reviewed-by: Eduardo Lima Mitev <elima@igalia.com>
The pass should run once at the end of shader compilation, for a4xx
onwards. It iterates texture sampling instructions and mark those
eligibile for pre-dispatch by changing the tex op from 'tex' to
'tex_prefetch'. An instruction is eligibile if:
* The coordinate is a vector where all its components come from a
shader input.
* The order of the components match exactly that of the input (no
swizzles).
* The instruction is in the 'main' function, and in the outer
most-block.
The first two restrictions were arrived to empirically, so more
testing could tighten or loosen it.
The 3rd restriction is there to allow moving the instructions
eligible for pre-dispatch to the beginning of the shader, so
that we don't block the registers holding the result for too
long.
Signed-off-by: Rob Clark <robdclark@chromium.org>
Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
It seems that pre-fs texture fetch only works if ij_pix ends up in r0.x.
I've tried unknown zero bits, to no avail, and blob also seems to force
r0.x when this feature is used.
Signed-off-by: Rob Clark <robdclark@chromium.org>
Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
Useful to see in disassembly listing texture fetches that were moved to
pre-dispatch.
Signed-off-by: Rob Clark <robdclark@chromium.org>
Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
If the only use of varyings is a pre-shader texture-fetch, we still need
to issue a bary.f with the end-input flag, otherwise we'll block further
VS invocations, as the hw will think varying storage is still busy.
Signed-off-by: Rob Clark <robdclark@chromium.org>
Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
It is possible that the result of a pre-fs texture fetch is an output
(or partially an output) of the FS. Sine the meta:tex_prefetch
instructions are dropped before the assembler, we need to account for
this when we fixup the register footprint.
Signed-off-by: Rob Clark <robdclark@chromium.org>
Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
Add a placeholder instruction to track texture fetches made prior to FS
shader dispatch. These, like meta:input instructions are scheduled
before any real instructions, so that RA realizes their result values
are live before the first real instruction. And to give legalize a way
to track usage of fetched sample requiring (sy) sync flags.
There is some related special handling for varying texcoord inputs used
for pre-fs-fetch, so that they are not DCE'd and remain in linkage
between FS and previous stage. Note that we could almost avoid this
special handling by giving meta:tex_prefetch real src arguments, except
that in the FS stage, inputs are actual bary.f/ldlv instructions.
Signed-off-by: Rob Clark <robdclark@chromium.org>
Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
When we enable pre-dispatch texture fetch, we could have a scenario
where the barycentric i/j coord sysval is not used in the shader, but
only used for the varying fetch for the pre-dispatch texture fetch.
In this case we need to take care not to DCE this sysval.
Signed-off-by: Rob Clark <robdclark@chromium.org>
Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
Will be needed for special handling of SYSTEM_VALUE_BARYCENTRIC_PIXEL
(ij_pix) when pre-fs texture fetch is enabled.
Signed-off-by: Rob Clark <robdclark@chromium.org>
Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
Not sure I remember how long this has been unused for. But it's unused
now.
Signed-off-by: Rob Clark <robdclark@chromium.org>
Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
This is like nir_texop_tex, but signals that the sampling coordinates
are immutable during the shader stage, in a way that allows the HW
that supports pre-dispatching sampling operations to pre-fetch
the result prior to scheduling the shader stage.
This is introduced to support the feature in Freedreno. Adreno HW
from a4xx supports it.
A NIR pass introduced later in this series will detect sampling
operations that are eligible for pre-dispatch, and replace
nir_texop_tex by this new op, to tell the backend to enable
pre-fetch.
Signed-off-by: Rob Clark <robdclark@chromium.org>
Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
We don't use cmake normally because it always results in static linking.
This is very problematic for *nix OSes which expect shared linking by
default, but for windows this isn't a problem as LLVM doesn't support
shared linking on windows anyway.
Reviewed-by: Adam Jackson <ajax@redhat.com>
For building on Windows (when not using cygwin), users may want to use a
binary wrap of LLVM, this provides a fallback to the LLVM dependency
which may be used in this case
Reviewed-by: Adam Jackson <ajax@redhat.com>
MIN filter is only used when LOD MAX is at least 4 (I guess the 2 LSB don't
actually exist).
Signed-off-by: Jonathan Marek <jonathan@marek.ca>
Reviewed-by: Lucas Stach <l.stach@pengutronix.de>
The etnaviv kernel driver will only ever flush write caches. As both
the TX descriptor and instruction cache are read caches they must be
flushed from the user cmdstream at an appropriate time.
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
Reviewed-by: Jonathan Marek <jonathan@marek.ca>
It's just a matter of writing the addressing mode into the
texture descriptor.
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
Reviewed-by: Jonathan Marek <jonathan@marek.ca>
Create a separate implementation file with texture-descriptor-based
sampler views and sampler states. Initialize the one or the other
based on the GPU. There is so little in common that this seemed more
appropriate that keeping them as one type of state object would
only be confusing.
This commit is actually a combiation of the original commit by
Wladimir, fixes and TS implementation from Jonathan and changed to
use softpin by Lucas.
Signed-off-by: Wladimir J. van der Laan <laanwj@gmail.com>
Signed-off-by: Jonathan Marek <jonathan@marek.ca>
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
Reviewed-by: Guido Günther <agx@sigxcpu.org>
Reviewed-by: Jonathan Marek <jonathan@marek.ca>
Halti5 uses texture descriptors to control the samplers, and thus needs to
know the GPU virtual address for the texture buffers to fill into the
descriptor buffer. Without softpin userspace has no control over the GPU
VM and also no way to fix up the texture descriptor buffer, so there is
no point in creating a screen on a Halti5 device without softpin being
available.
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
Reviewed-by: Jonathan Marek <jonathan@marek.ca>
If softpin is available on the kernel side, we transparently replace the
relocs with self-managed GPU virtual addresses. This allows to skip some
work at the kernel side, as it doesn't need to touch the command stream
anymore before submitting it to the hardware.
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
Reviewed-by: Jonathan Marek <jonathan@marek.ca>
Replace the per-screen locking of flushing with per-context one and
add per-context lock around command stream buffer accesses, to prevent
cross-context flushing from corrupting these command stream buffers.
Signed-off-by: Marek Vasut <marex@denx.de>
Reallocate the command stream buffer in case it is too small.
The older kernel versions are limited to 64 kiB buffer, so
limit the size to avoid oversized buffers.
Signed-off-by: Marek Vasut <marex@denx.de>
Have each context track which resources it marked as pending read and
pending write. Have each resource track in which context it is pending.
This way, it is possible to identify when a resource is both pending
read and pending write at the same time. Moreover, the status field
can be correctly calculated and updated when necessary.
Signed-off-by: Marek Vasut <marex@denx.de>
This way we can ensure that the pipe driver tracking of pending resources
stays in sync with the actual command buffer state, even if a space
reservation triggers a forced flush.
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
Reviewed-by: Jonathan Marek <jonathan@marek.ca>
As long as a resource is pending in any context we must not destroy
it, otherwise we'll hit a classical use-after-free with fireworks.
To avoid this take a reference when the resource is first added to
the pending set and put the reference when no longer pending.
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
Reviewed-by: Jonathan Marek <jonathan@marek.ca>
Currently, the screen tracks all resources for all contexts, but this
is not correct. Each context should track the resources it uses. This
also allows a context to detect whether a resource is used by another
context and to notify another context using a resource that the current
context is done using the resource.
Signed-off-by: Marek Vasut <marex@denx.de>
Cc: Christian Gmeiner <christian.gmeiner@gmail.com>
Cc: Guido Günther <guido.gunther@puri.sm>
Cc: Lucas Stach <l.stach@pengutronix.de>
This exposes what's required for DX and this is what we already
configure. The driver flushes denorms for FP32 and preserves them
for FP16/FP64. Note that we can't allow both preserving and
flushing denorms because this won't work for merged shaders. This
will require LLVM to update the float mode register to make it work.
Only enabled on GFX8+ with the LLVM path because it's untested on
previous chips and ACO doesn't support it.
This extension is required for SPIRV 1.4.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Because some instructions will be optimized by the backend compiler,
the driver has to manually flush to zero to keep the result exact.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
The new Mac OS X images apparently already have python2 and python3,
and `brew` considers asking to install something already installed
as a fatal error...
Signed-off-by: Eric Engestrom <eric.engestrom@intel.com>
This will expose GL_EXT_primitive_bounding_box and
GL_OES_primitive_bounding_box after previous commits
expose OpenGL ES 3.1 once Compute Shaders are available.
Reviewed-by: Eric Anholt <eric@anholt.net>
This adapts the v3d driver to the new CL submit ioctl interface that
allows the driver to request a flush of the caches after the render
job has completed. This seems to eliminate the kernel write violation
errors reported during CTS and Piglit excutions, fixing some CTS tests
and GPU resets along the way.
v2:
- Adapt to changes in the kernel side.
- Disable shader storage and shader images if the kernel doesn't
implement cache flushing.
Fixes CTS tests:
KHR-GLES31.core.shader_image_size.basic-nonMS-fs-float
KHR-GLES31.core.shader_image_size.basic-nonMS-fs-int
KHR-GLES31.core.shader_image_size.basic-nonMS-fs-uint
KHR-GLES31.core.shader_image_size.advanced-nonMS-fs-float
KHR-GLES31.core.shader_image_size.advanced-nonMS-fs-int
KHR-GLES31.core.shader_image_size.advanced-nonMS-fs-uint
KHR-GLES31.core.shader_atomic_counters.advanced-usage-many-draw-calls2
KHR-GLES31.core.shader_atomic_counters.advanced-usage-draw-update-draw
KHR-GLES31.core.shader_storage_buffer_object.advanced-unsizedArrayLength-fs-int
KHR-GLES31.core.shader_storage_buffer_object.advanced-unsizedArrayLength-fs-std140-matR
KHR-GLES31.core.shader_storage_buffer_object.advanced-unsizedArrayLength-fs-std140-struct
KHR-GLES31.core.shader_storage_buffer_object.advanced-unsizedArrayLength-fs-std430-matC-pad
KHR-GLES31.core.shader_storage_buffer_object.advanced-unsizedArrayLength-fs-std430-vec
Reviewed-by: Eric Anholt <eric@anholt.net>
Now that the UAPI has landed, add the pipe_context function for
dispatching compute shaders. This is the last major feature for GLES 3.1,
though it's not enabled quite yet.
That we set for any TMU write on spills and general tmu. It is then
used as part of v3d_emit_gl_shader_state later.
v2: add a new flag instead at v3d_compiler instead of dirty the flag
at v3dx if there is any spill (change suggested by Eric, added by
Alejandro)
v3: set this for anything that is not a load and do it also in
v3d40_vir_emit_image_load_store (Eric)
Reviewed-by: Eric Anholt <eric@anholt.net>
The SCR_INIT macro used to install the rbug resource_changed method
will only do so when the driver below rbug exposes this method, so
the check will always evaluate to true.
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
All the other context method initialzation follow the order of the pipe_context
structure definition making it easy to find unimplemented methods in rbug.
Move the flush_resource init to follow the same order.
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
All resources passed to the drivers below rbug need to be unwrapped before
being passed down. We missed to do this for the index buffer resource when
this was made part of the draw_info structure.
Fixes: 330d0607ed (gallium: remove pipe_index_buffer and set_index_buffer)
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
The rbug wire format defines the texture size parameters to be uint32_t sized
and uses memcpy to move the function parameters to the message structure.
This caused totally wrong transmitted texture sizes since the height and depth
paramterds have been changed to uint16_t in the gallium API. Fix this by doing
an explicit conversion to the correct representation before packing into the
wire message.
Fixes: e6428092f5 (gallium: decrease the size of pipe_resource - 64 -> 48 bytes)
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
Using 0 as the backlog argument to listen() is exploiting implementation
defined behavior and will lead to no connections being accepted on some
libc implementations.
Quote of the listen manpage: "A backlog argument of 0 may allow the socket to
accept connections, in which case the length of the listen queue may be set to
an implementation-defined minimum value."
Fix this by using a more sensible backlog value.
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
It seems that for desktop GL this was included with ARB_gpu_shader5, but
for OpenGL ES this is already included with the base extension and there is
a CTS test that checks this.
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Implement the 3 functions using the texturestorage_error() helper.
_mesa_lookup_or_create_texture is always called to make sure that 'texture'
is initialized (even if the texturestorage_error() generates an error afterwards).
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
When we import a resource through Gallium, we need to take account of
the offset parameter passed.
Fixes a failure seen with the VIVID V4L2 driver, which would create NV12
resources within the same BO, with an offset. Sample pipeline to
reproduce (replace videoN with your actual VIVID device node):
gst-launch-1.0 v4l2src device=/dev/videoN ! video/x-raw,format=NV12 ! glimagesink
Signed-off-by: Daniel Stone <daniels@collabora.com>
Reviewed-by: Tomeu Vizoso <tomeu.vizoso@collabora.com>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Reported-by: Nicolas Dufresne <nicolas.dufresne@collabora.com>
Tested-by: Nicolas Dufresne <nicolas.dufresne@collabora.com>
Reworks:
* Change subject from "iris: Align main surface allocation to 64k on gen12+"
* Make use of isl surf alignment. (Nanley)
Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reworks:
* Fill out the format's entry in the ISL format table. (Nanley)
* Support CCS_E-enabled BLORP copies with the format. (Nanley)
Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Sagar Ghuge <sagar.ghuge@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
This can be useful to measure whether memory access optimizations are
having the desired effect. For example, we might see a reduction in
image loads/stores, or constant buffer loads. We can already see this
in cycle estimates to some extent, but this is a more direct approach,
minus a lot of the noise of random scheduler shuffling.
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
The removed st_nir_opts calls are mostly redundant.
There is an improvement with shader-db on radeonsi:
Before:
real 1m54.047s
user 28m37.857s
sys 0m7.573s
After:
real 1m52.012s
user 28m3.412s
sys 0m7.808s
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
DPH isn't actually commutative, so this doesn't work. If the immediate
in src0 would be a VF candidate, we could do better. *shrug*
No shader-db changes on any Intel platform.
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Fixes: b04beaf41d ("intel/vec4: Try both sources as candidates for being immediates")
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Fixes: 09705747d7 ("nir/algebraic: Reassociate fadd into fmul in DPH-like pattern")
This function was difficult to implement for new formats due to the
combination of endianness and swapbytes support. Since it's mostly
used for fast paths, bugs in it were often missed during testing.
Just reimplement it on top of the recent
_mesa_format_from_format_and_type() which can give us a canonical
MESA_FORMAT for a format and type enum (while respecting endianness).
Fixes:
- R4G4B4A4_UNORM, B4G4R4_UINT, R4G4B4A4_UINT incorrectly matched with
swapBytes (you can't just reverse the channels if the channels
aren't bytes)
- A4R4G4B4_UNORM and A4R4G4B4_UINT missing BGRA/4444_REV matches
- failing to match RGB/BGR unorm8 array formats on BE
- 2101010 formats incorrectly matching with swapBytes set.
- UINT/SINT byte formats failed to match with swapBytes set.
This deletes the part of tests/mesa_formats.cpp that called
_mesa_format_matches_format_and_type() to make sure it didn't
assertion fail, as it now would assertion fail due to the fact that we
were passing an invalid format (GL_RG) for most types.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
In desktop GL, you can specify things like GL_DEPTH_COMPONENT/GL_BYTE as a
ReadPixels format, and we need to be able to represent that to see if we
have proper MESA_FORMATs for them. That's exactly what the
mesa_array_format enum is for.
v2: Drop _mesa from static fn.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
We had missed this case where GLES3 allows glReadPixels(DEPTH, UINT_24_8),
and just got lucky by the readpixels path never asking for the matching
format from this function.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
The GL spec says the 24-bit component is in the high bits, and
format_unpack.c looks at the high 24 bits in the S8Z24 case, not
Z24SS8.
Avoids a regression in the next commit.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
The unreachable() that follows isn't very useful for debug, and by adding
this here we get a nice description of the failure in debug builds.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
When we don't have streamout enabled, we have to read this register to
get the number of primitives emitted.
Signed-off-by: Kristian H. Kristensen <hoegsberg@google.com>
When used in a GS pipeline, the VS doesn't end with the END
instruction. Instead it chains to the GS, which continues running with
the same register allocation. The intended use cases seems to be that
you can compile a regular VS (ie outputs in registers and ending with
END) but then tack on link-time generated code past the END to write
the outputs using STLW, in case the VS is used with GS.
Signed-off-by: Kristian H. Kristensen <hoegsberg@google.com>
We don't know what kind of loads we might have to wait on when coming
in from chsh in the VS so set both sync flags.
Signed-off-by: Kristian H. Kristensen <hoegsberg@google.com>
These sysvals have to be unclobbered by VS and in the same registers
in both VS and GS, since the chsh from VS to GS doesn't reload the
values. We use the pre-color argument to ir3_ra() to always place
these values in r0.x and r0.y.
Signed-off-by: Kristian H. Kristensen <hoegsberg@google.com>
Inputs are the GS header, which contains vertex ID, local primitive ID
and thread ID as well as primitive ID. The setup is a little different
from other sysvals, since we always have to receive them in the VS so
that it can pass them on into the GS.
The vertex flag outputs from GS is set up as a proper nir output in
the lowering pass and doesn't need special handling here.
Signed-off-by: Kristian H. Kristensen <hoegsberg@google.com>
This implements the load_vs_primitive_stride_ir3,
load_vs_vertex_stride_ir3 and load_primitive_location_ir3 intrinsics,
used for getting the primitive layout strides and locations.
Signed-off-by: Kristian H. Kristensen <hoegsberg@google.com>
This introduces two new lowering passes. One to lower VS to explicit
outputs using STLW and one to lower GS to load input using LDLW and
implement the GS specific functionality.
Signed-off-by: Kristian H. Kristensen <hoegsberg@google.com>
Since the presence of GS changes how the VS operates we need to track
that in the shader key.
Signed-off-by: Kristian H. Kristensen <hoegsberg@google.com>
These intrinsics will let us do all the offset calculations in nir,
which is nicer to work with and lets nir_opt_algebraic eat it all up.
Signed-off-by: Kristian H. Kristensen <hoegsberg@google.com>
Before, offset held the offset, which can be either immediate or a
register. Use a third register to hold the offset so that we can use
a register.
Signed-off-by: Kristian H. Kristensen <hoegsberg@google.com>
Just add the constructors for now and special case similar to END so
we don't remove them.
Signed-off-by: Kristian H. Kristensen <hoegsberg@google.com>
We know what these do an either write them in the program stateobj or
don't need to write them.
Signed-off-by: Kristian H. Kristensen <hoegsberg@google.com>
Tests the combinations of cases of RAW, WAW and WAR hazards involving
both inorder and outoforder instructions. Also tests that
dependencies combine and propagate correctly through control
flow (loops and conditionals).
v2: Add an extra test illustrating that the non-logical CFG edge
between then-block and else-block is being taking into
account. (Curro)
Reviewed-by: Francisco Jerez <currojerez@riseup.net>
LLVM 8 did remove both the signed and unsigned sse2/avx intrinsics in
the end, and provide arch-independent llvm intrinsics instead.
Fixes a crash when using snorm framebuffers (tested with piglit
arb_color_buffer_float-render GL_RGBA8_SNORM -auto).
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
Reviewed-by: Dave Airlie <airlied@redhat.com>
CC: <mesa-stable@lists.freedesktop.org>
If two jobs use the same GEM object at the same time, the job that
finishes first will (previous to this commit) close the GEM object, even
if there's a job still referencing it.
To prevent this, have all jobs use the same panfrost_bo for a given GEM
object, so it's only closed once the last job is done with it.
Signed-off-by: Tomeu Vizoso <tomeu.vizoso@collabora.com>
Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Daniel Stone <daniels@collabora.com>
Reviewed-by: Rohan Garg <rohan.garg@collabora.com>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
A new throttle fence was initialized to 1, and increased by 1
again when it's put in drawable->throttle_fence; the ref was
decreased by 1 when it's removed from drawable->throttle_fence,
and never reached to 0, caused leak.
Fixes: ff77bf5cbf7 ("gallium: simplify throttle implementation")
Closes: https://gitlab.freedesktop.org/mesa/mesa/issues/1949
Signed-off-by: James Xiong <james.xiong@intel.com>
Reported-by: Florian Wesch <fw@info-beamer.com>
Reviewed-by: Jose Maria Casanova Crespo <jmcasanova@igalia.com>
This allows us to make sure clipdist is emitted as a scalar array rather
than two vec4s. This matches SPIR-V semantics, and will be useful for
Zink.
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
This actually corresponds to legal GL depth-ranges, because depth-clear
values are always in the 0..1 range in OpenGL.
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
This will prevent us from accidentally falling back to the wrap-db
instead of using locally installed versions.
Reviewed-by: Eric Engestrom <eric.engestrom@intel.com>
The one debian provides is broken in buster+, so I've just written my
own. This allows meson to find the installed zlib and prevents it from
falling back to wraps.
Reviewed-by: Eric Engestrom <eric.engestrom@intel.com>
It's not really needed, and there's no debian package for it so we're
forced to fall back to wraps in mesa's CI. This can be problematic in
itself.
Reviewed-by: Eric Engestrom <eric.engestrom@intel.com>
nvc0 and I assume radeonsi as well hit an assert inside glsl_to_tgsi as atan
instructions get inserted into the shader.
Fixes: cece947a8d ("glsl/builtin: Add alternate versions of atan using new ops")
Cc: Neil Roberts <nroberts@igalia.com>
Signed-off-by: Karol Herbst <kherbst@redhat.com>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Rob Clark <robdclark@gmail.com>
We're trying to cast the return type to the type of the var, but instead
we were casting `sizeof(*v)`.
Fixes: 6df72e970c ("util: Make u_atomic.h typeless.")
Fixes: 0a7f17cf5b ("util/u_atomic: add p_atomic_xchg")
Signed-off-by: Eric Engestrom <eric.engestrom@intel.com>
Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
It's already defined in `m_debug_util.h`, along with an explanation of
what it is and how to use it.
Signed-off-by: Eric Engestrom <eric.engestrom@intel.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
'struct lima_context' has to be declared before usage in lima_program.h
Signed-off-by: Andreas Baierl <ichgeh@imkreisrum.de>
Reviewed-by: Vasily Khoruzhick <anarsoul@gmail.com>
Fixes build errors of:
In file included from ../src/intel/vulkan/anv_private.h:48,
from ../src/intel/vulkan/genX_blorp_exec.c:26:
../src/intel/common/gen_gem.h: In function ‘gen_ioctl’:
../src/intel/common/gen_gem.h:68:15: error: implicit declaration of function ‘ioctl’ [-Werror=implicit-function-declaration]
68 | ret = ioctl(fd, request, arg);
| ^~~~~
In file included from ../include/c11/threads_posix.h:35,
from ../include/c11/threads.h:66,
from ../src/mesa/main/mtypes.h:39,
from ../src/intel/compiler/brw_compiler.h:30,
from ../src/intel/vulkan/anv_private.h:51,
from ../src/intel/vulkan/genX_blorp_exec.c:26:
/usr/include/unistd.h: At top level:
/usr/include/unistd.h:471:12: error: conflicting types for ‘ioctl’
471 | extern int ioctl(int, int, ...);
| ^~~~~
/usr/include/unistd.h:471:1: note: a parameter list with an ellipsis can’t match an empty parameter name list declaration
471 | extern int ioctl(int, int, ...);
| ^~~~~~
In file included from ../src/intel/vulkan/anv_private.h:48,
from ../src/intel/vulkan/genX_blorp_exec.c:26:
../src/intel/common/gen_gem.h:68:15: note: previous implicit declaration of ‘ioctl’ was here
68 | ret = ioctl(fd, request, arg);
| ^~~~~
Signed-off-by: Alan Coopersmith <alan.coopersmith@oracle.com>
Acked-by: Eric Engestrom <eric.engestrom@intel.com>
gcc is very particular about where you place the (void) cast
The previous placement made it error out with:
In file included from disk_cache.c:40:0:
../../src/util/u_atomic.h:203:29: error: void value not ignored as it ought to be
#define p_atomic_add(v, i) ((void) \
^
disk_cache.c:658:4: note: in expansion of macro ‘p_atomic_add’
p_atomic_add(cache->size, size);
^
Signed-off-by: Alan Coopersmith <alan.coopersmith@oracle.com>
Reviewed-by: Eric Engestrom <eric.engestrom@intel.com>
Fixes build failures on Solaris in C++ files using gcc:
../src/util/u_math.h:628:41: error: expected ‘,’ or ‘...’ before ‘dest’
628 | util_memcpy_cpu_to_le32(void * restrict dest, const void * restrict src, size_t n)
| ^~~~
../src/util/u_math.h: In function ‘void* util_memcpy_cpu_to_le32(void*)’:
../src/util/u_math.h:641:18: error: ‘dest’ was not declared in this scope
641 | return memcpy(dest, src, n);
| ^~~~
../src/util/u_math.h:641:24: error: ‘src’ was not declared in this scope
641 | return memcpy(dest, src, n);
| ^~~
../src/util/u_math.h:641:29: error: ‘n’ was not declared in this scope; did you mean ‘yn’?
641 | return memcpy(dest, src, n);
| ^
| yn
Signed-off-by: Alan Coopersmith <alan.coopersmith@oracle.com>
Acked-by: Eric Engestrom <eric.engestrom@intel.com>
It doesn't make sense. You already spilled it once, and it didn't help.
Don't try again, or you'll end up in a loop.
Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Essentially an off-by-one error ... bit of an edge case, but seems to
occur in some glamor shaders.
Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
The new frame throttling implemention interacts unfortunately with
pipelining, leading to fence fds leaking like crazy and ultimately apps
crashing quickly.
With this patch, apps still crash but not as quickly. We need to either
figure out the real cause or revert the core changes.
Nevertheless, we don't want frame throttling in the first place, so.
Fixes: a65e29ccb2 ("gallium: simplify throttle implementation")
Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
This commit moves the target check before using _mesa_get_current_tex_object
to fix a "Mesa implementation error: bad target in _mesa_get_current_tex_object()"
error.
Fixes: 9dd1f7cec0 ("mesa: pass gl_texture_object as arg to not depend on state")
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
We don't really need to impose this condition, but we do need to cope
with the slightly more general case.
Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
A buffer and its aux are imported separately, if the aux import is
not completed yet when resource_get_param is called, merge the
separate aux a.k.a the 2nd image into the main image.
Fixes: 246eebba4a ("iris: Export and import surfaces with modifiers that have aux data")
Signed-off-by: James Xiong <james.xiong@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Once again, we were handling back-to-front in the GLES3 case, but not
the desktop GL case.
Fixes GTF-GL46.gtf30.GL3Tests.framebuffer_srgb.framebuffer_srgb_default_encoding when run with --deqp-surface-type=pbuffer --deqp-gl-context-type=egl.
Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
We were looking at ctx->DrawBuffer when asking about the read buffer,
which was good enough for CTS purposes, but definitely not right.
Reviewed-by: Eric Engestrom <eric.engestrom@intel.com>
Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
The first time you call glXMakeCurrent, current != ctx. As a result we
would never look up whether the drawable already had an XMesaDrawable,
and would instead always create one. Then XMesaBufferList would have two
different buffers for the same XID, and you'd be reading and drawing to
different places, and that's not what you want at all.
Instead just always look up the drawable.
Fixes: db8be355 (gallium/xlib: Remove drawable caching from the MakeCurrent path)
Closes: https://gitlab.freedesktop.org/mesa/mesa/issues/1196
Reviewed-by: Michel Dänzer <mdaenzer@redhat.com>
This reverts commit 2ca8629fa9.
This was initially ported from RadeonSI, but in the meantime it has
been reverted because it might hang. Be conservative and re-introduce
this packet emission.
Unfortunately this doesn't fix anything known.
Cc: 19.2 <mesa-stable@lists.freedesktop.org>
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Variables with same location should use the same driver_location.
Signed-off-by: Jonathan Marek <jonathan@marek.ca>
Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
Taken from nir_lower_samplers. Sampler arrays don't work though, this is
just to avoid an assert fail in ir3.
Signed-off-by: Jonathan Marek <jonathan@marek.ca>
Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
Notably includes centroid varying bits that were missing.
Signed-off-by: Jonathan Marek <jonathan@marek.ca>
Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
Noticed while debugging a tiling-looking issue by comparing our gmem
blit setup to freedreno's.
Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
Fixes ir3 compiler failure failure in
dEQP-VK.renderpass.dedicated_allocation.formats.r8g8b8a8_unorm.clear.clear_draw
(now just a rendering failure where the subpass clear isn't happening)
Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
Fixes assertion failures in
dEQP-VK.api.image_clearing.core.clear_color_image.2d.* for these
formats, though the test set as a whole is stil failing.
Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
Deal with tiled r8g8 having different alignment and other updates taken
from fd6_resource. Additionally track image samples/cpp.
Signed-off-by: Jonathan Marek <jonathan@marek.ca>
Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
Avoids hangs and some texture tests are happy with just this.
Signed-off-by: Jonathan Marek <jonathan@marek.ca>
Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
Changes to make compressed, tiled, 3d, etc textures work
Signed-off-by: Jonathan Marek <jonathan@marek.ca>
Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
* Fix R16G16 SCALED and R16G16B16A16 SCALED having texture format
* Fix B5G6R5 swap value
* Use R8_UINT instead of R8_UNORM for S8_UINT rb format
* Disable 96-bit texture formats instead having a check for NPOT formats
* Don't fail assert on D24X8 format
Signed-off-by: Jonathan Marek <jonathan@marek.ca>
Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
Not supported, so always set pointer to NULL
Signed-off-by: Jonathan Marek <jonathan@marek.ca>
Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
Note: for output type U32, negative LOD is not sign extended from 16 bits
Signed-off-by: Jonathan Marek <jonathan@marek.ca>
Reviewed-by: Rob Clark <robclark@gmail.com>
GPUs with a single supported vertex stream must use the single state
address to program the stream.
Fixes: 3d09bb390a (etnaviv: GC7000: State changes for HALTI3..5)
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
Reviewed-by: Jonathan Marek <jonathan@marek.ca>
This gs_iface doesn't seem to require a dependence on the tgsi
context, except for the swr end prim code.
This refactors the API to include all the info that the swr
code needs in the interface rather than having to dig it out of
the struct inheritance.
This is a precursor to adding NIR support to llvmpipe.
Reviewed-by: Jan Zielinski <jan.zielinski@intel.com>
Fixes the following building error:
external/mesa/src/gallium/winsys/amdgpu/drm/amdgpu_winsys.c:42:10:
fatal error: 'ac_llvm_util.h' file not found
^~~~~~~~~~~~~~~~
1 error generated.
Fixes: 3a08110 ("amd: Move all amd/common code that depends on LLVM to amd/llvm.")
Signed-off-by: Mauro Rossi <issor.oruam@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
All gallium drivers currently set MAX_FRAME_IN_FLIGHT to either 1
or 0, which means that the drivers either throttle on the previous
render or don't throttle, the current implementation is more
complicated than necessary and can be simplified.
Signed-off-by: James Xiong <james.xiong@intel.com>
Reviewed-by: Michel Dänzer <mdaenzer@redhat.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Adds alternate versions of the atan builtin functions that use
ir_unop_atan and ir_binop_atan2 instead of inlining to the IR
implementation of the function. These alternatives are selected if the
IR is going to be consumed by NIR. In that case the IR ops will be
translated to the appropriate NIR op.
Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
Adds ir_binop_atan2 and ir_unop_atan. When converting to NIR these are
expanded out using the appropriate builtin generator. If they are used
with anything else then it will just hit an assert.
Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
Moves build_atan and build_atan2 into nir_builtin_builder. The goal is
to be able to use this from the GLSL translator too.
Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
When users pass a config to `eglCreateWindowSurface` it requests double
buffering, but if the config doesn't have the appropriate `__DRIconfig`,
`eglCreateWindowSurface` fails with a `EGL_BAD_MATCH`.
Given that such behaviour is completely unacceptable, we drop the
`EGL_WINDOW_BIT` if we don't have at least one `__DRIconfig` supporting double
buffering, otherwise dropping the `EGL_PIXMAP_BIT`.
Fixes: 049f343e8a "egl: Allow 24-bit visuals for 32-bit RGBA8888 configs"
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=67676
Cc: mesa-stable@lists.freedesktop.org
Reviewed-by: Adam Jackson <ajax@redhat.com>
Signed-off-by: Hal Gentz <zegentzy@protonmail.com>
This commit does this by allowing both RGB and RGBA visuals to match with
EGL configs. We also expose the `EGL_MESA_config_select_group` egl
extension, which is similar to GLX's visual select group extension, to
allow the RGBA visuals to get less priority.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=67676
Fixes: 049f343e8a "egl: Allow 24-bit visuals for 32-bit RGBA8888 configs"
Cc: mesa-stable@lists.freedesktop.org
Reviewed-by: Adam Jackson <ajax@redhat.com>
Signed-off-by: Hal Gentz <zegentzy@protonmail.com>
The bit moved on gen12 in order to prepare for dual-SIMD8 dispatch.
This implementation isn't an entirely complete as it only works on SIMD8
and SIMD16 and not dual-SIMD8.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Apparently the ts_request_type and ts_resource_select thread spawner
message descriptor bits were removed from the hardware at least since
ICL. Drop them in order to avoid assertion failures on Gen12+
platforms which don't have any encoding for this. On Gen9+ these are
probably just ignored by the hardware, so this is unlikely to have had
any functional implications prior to Gen12.
v2: Mark TS message fields as non-existing in brw_inst.h on ICL. (Caio)
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
The WAIT instruction has been removed, but SYNC.bar can be used
instead to wait for a notification on n0.0.
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Apparently this field was removed on SKL, and according to the
hardware docs for previous platforms "This field is only valid for a
ForwardMsg message. It is ignored for other messages. The BarrierMsg
message always increments the N0 notification counter".
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Confirmed no regressions after a full Piglit run on TGL with the
brw_fs_test_dispatch_packing() test enabled.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
They look like a NULL source if you don't look at the address mode.
Reviewed-by: Francisco Jerez <currojerez@riseup.net>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
The following fix-up by Jordan Justen is squashed in:
intel/eu/validate: gen12 send instruction doesn't have a dst type field
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Due to hardware bug filed as HSDES#1604601757.
v2: Only return if result of fs_inst::can_do_source_mods() is known to
be false for the case new orthogonal restrictions are implemented
below in the future. (Caio)
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Kept as a separate commit in order to avoid distracting reviewers of
the software scoreboard pass with memory management boilerplate.
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Gen12+ hardware lacks the register scoreboard logic that used to
guarantee data coherency between register reads and writes in previous
generations. This lowering pass runs after register allocation in
order to make up for it.
It works by performing global dataflow analysis in order to determine
the set of potential dependencies of every instruction in the shader,
and then inserts any required SWSB annotations and additional SYNC
instructions in order to guarantee data coherency.
v2: Drop unnecessary _safe list iteration (Caio).
v3: Temporarily workaround potential WaR hazard between FPU
instruction and subsequent out-of-order write, pending
clarification from the hardware team. Drop redundant tracking of
implicit access of acc0-1, since the hardware guarantees coherency
of these (but not the other accumulators...).
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Reviewers are encouraged to audit the code generation pass
independently for the case I missed some potential data hazard or new
code has been added in the meantime.
v2: Add SYNC instruction to cr0 workaround in brw_float_controls_mode().
v3: Drop likely redundant (and potentially harmful) RegDist SWSB
annotation from ce0 read in brw_find_live_channel() (Caio).
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
An effect similar to the one formerly provided by setting thread
control to "switch" can be achieved now by setting a RegDist of 1 on
the SWSB field.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
A future lowering pass will simulate the same behavior originally
provided by NoDDChk/NoDDClr at the IR level by using appropriate SWSB
annotations.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
The new SEND instruction behaves like the former SENDS instruction.
The original single-payload SEND instruction is gone.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
The SEND instruction is now four-source. The descriptor is no longer
part of source 1, so avoid touching it to avoid corruption while
initializing the descriptor.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Quite a lot of churn because the encoding of most hardware opcodes has
changed unfortunately.
v2: Split dot-product description fixes to separate patch (Caio).
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
The encoding of almost every instruction field has changed in Gen12,
so this involves adding a Gen12+ bitfield spec to every brw_inst
macro. In addition some new macros are required to handle certain
discontiguous and variable-length fields.
This commit doesn't actually include the Gen12 updated bitfield specs,
only the macros are extended here for reviewability.
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
v2: Rename FDC() to FFDC() and FDC1() to FDC() for consistency with
the existing F() and FF() macros.
This edge doesn't exist in the original scalar program, but it
represents a potential control flow path the EU will take in cases
where control flow isn't uniform across channels of the same SIMD
thread.
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
This edge doesn't exist in the original scalar program, but it
represents a potential control flow path the EU will take in cases
where the condition isn't uniform across channels of the same SIMD
thread.
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Currently only the physical back-edge is represented, which
incidentally also leads to the exit block of the loop, but we need the
direct logical edge in addition for our logical CFG representation to
be complete.
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
This represents two control flow graphs in the same cfg_t data
structure: The physical CFG that will include all possible control
flow paths the EU can physically take, and the logical CFG restricted
to the control flow paths that exist in the original scalar program.
The latter is a subset of the former because in case of divergence the
SIMD vectorized program will take control flow paths that aren't part
of the original scalar program.
The bblock_link constructor and bblock_t::add_successor() now take a
"kind" parameter that specifies whether the edge is purely physical or
whether it's part of both the logical and physical CFGs (a logical
edge is of course always guaranteed to be in the physical CFG as
well). bblock_t::is_predecessor_of() and ::is_successor_of() also
take a kind parameter specifying which CFG is being queried. The '~>'
notation will be used now in order to represent purely physical edges
in IR dumps.
This commit doesn't actually add nor remove any edges from the CFG
(the only edges marked as purely physical here are the two WHILE loop
ones that already existed). Optimization passes should continue using
the same (incomplete) physical CFG they were using before until
they're fixed to do something smarter in a later commit, so this
shouldn't lead to any functional changes.
v2: Remove tabs from lines changed in this file (Caio).
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Having the IR opcodes locked to their hardware representation is risky
because it causes opcodes as different as BRC and IFF to compare equal
at the IR level (luckily the back-end only ever uses one opcode from
each group, right now), and it prevents us from supporting
instructions that change their hardware representation across
generations, which will become a problem on Gen12+ platforms.
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Change brw_inst_set_opcode() and brw_inst_opcode() to call
brw_opcode_encode/decode() transparently in order to translate between
hardware and IR opcodes, and update the EU compaction code in order to
do the same as needed, so we can eventually drop the one-to-one
correspondence between hardware and IR opcodes.
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
This rewrites the current opcode description tables as a more compact
flat data structure. The purpose is to allow efficient constant-time
look-up by either HW or IR opcode, which will allow us to drop the
hard-coded correspondence between HW and IR opcodes -- See the next
commits for the rationale.
brw_eu.c is now built as C++ source so we can take advantage of
pointers to member in order to make the look-up function work
regardless of the opcode_desc member used as look-up key.
v2: Optimize devinfo struct comparison (Caio)
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
The brw_inst opcode accessors are going away in one of the following
commits. We could potentially replace them with the new helpers that
do opcode remapping, but that would lead to a circular dependency
between brw_inst.h and brw_eu.h. This way we also avoid ordering
issues that can cause the semantics of the ex_desc accessors to change
depending on whether the ex_desc field is set after or before the
opcode instruction field.
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
This is required because SEND message payload sources are fetched
asynchronously by the hardware, which can lead to WaR data corruption
on Gen12+ platforms if not handled specially by the compiler to
guarantee proper synchronization.
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
And after discard-only loops. Otherwise we end up with dead code
which confuses nir_repair_ssa into adding a whole bunch of uses
of undefined. However, for derefs, we sometimes always expect to
get a variable instead of undefined.
Fixes dEQP-VK.graphicsfuzz.write-red-in-loop-nest on radv.
Fixes: c832820ce9 "nir/dead_cf: Repair SSA if the pass makes progress"
Closes: https://gitlab.freedesktop.org/mesa/mesa/issues/1928
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
p_as_uniform can get CSE'd, which can be incorrect and break some
dEQP-VK.descriptor_indexing.* tests.
Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Fixes the UBO/SSBO dEQP-VK.descriptor_indexing.* tests
v2: remove bld.copy() usage
Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
This can happen when bcsel is used between the results of two
vulkan_resource_index. It's also probably needed for non-uniform
descriptor indexing
Fixes dEQP-VK.spirv_assembly.instruction.compute.variable_pointers.compute.reads_opselect_two_buffers
Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
v2: always assert on the texture/sampler handle's num_components
v3: replicate the deref inside the loop
v4: remove a case of useless line wrapping
Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Now that the base resource is allowed to be incompatible with PE, we can
make a smarter choice of tiling mode to avoid allocating a PE compatible
base that is never used for regular textures. This affects GPUs like GC2000
where there is no tiling compatible with both PE and TE.
Signed-off-by: Jonathan Marek <jonathan@marek.ca>
Reviewed-by: Christian Gmeiner <christian.gmeiner@gmail.com>
For PE-incompatible layouts, use a mechanism similar to what texture does
to create a compatible base resource.
Signed-off-by: Jonathan Marek <jonathan@marek.ca>
Reviewed-by: Christian Gmeiner <christian.gmeiner@gmail.com>
Remove the "addressing_mode" state, which is currently set incorrectly, and
instead deduce the addressing mode from the tiling layout.
Signed-off-by: Jonathan Marek <jonathan@marek.ca>
Reviewed-by: Christian Gmeiner <christian.gmeiner@gmail.com>
It simplifies the definitions of jobs using the Debian 10 image.
The needs: was previously missing from the llvmpipe/softpipe test jobs,
so they could spuriously run if the debian-10 job failed or was
cancelled.
Reviewed-by: Eric Engestrom <eric.engestrom@intel.com>
The coroutine split pass is missing a dependency before LLVM 9.0,
and fails to initialise properly if the CallGraphWrapperPass hasn't
be initialised earlier (x86 does it due to some of it's passes
requiring it).
This is a workaround for llvm 8 (coroutines are only supported in 8
and higher). It adds another pass that has a dependency on the pass
the coroutines split requires. This pass shouldn't have any raal
effects.
Fixes: d32690b43c (gallivm: add coroutine pass manager support)
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
This is needed as part of GLES3.1 and helps for ARB_gpu_shader5.
Fixes: KHR-GLES31.core.texture_gather.* cases
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
This will encode the component selection value (0, 1, 2, 3) into
the X swizzle of the sampler, if the driver requests it.
Signed-off-by: Dave Airlie <airlied@redhat.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Accessing the TG4 component via immediates in the llvmpipe backend is quite
messy (like really messy). Roland suggested we change the instruction encoding,
so introduce a cap to allow the component to be selected to be store in the
sampler swizzle, which should be otherwise unused.
I could probably switch all drivers over, but virgl would need some work that
I'd prefer not to rush it.
Signed-off-by: Dave Airlie <airlied@redhat.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
This job uses the vs2017 backend of meson (msbuild) as opposed to the
ninja backend used on MacOS and Linux.
v7: - rebase on master
- remove llvm (we'll add that back later)
- remove cygwin (we'll add that back later too)
v6: - rebase on master, including the addition of cygwin
- consolidate 3 appveyor patches into this one patch
v5 - use the new b_vscrt option instead of manually specifying the crt
v4: - rebase on python3 generators
- cache meson wraps
- Build x86 instead of x86_64, since that's what the pre-built LLVM
is
- update to vs2017 from vs2015
- set the default-library to static
- use the new vscrt override
- add the /m switch to msbuild to make the build somewhat faster
Acked-by: Kristian H. Kristensen <hoegsberg@google.com>
Currently meson doesn't correctly handle passing compiled binaries to
scripts in tests. This patch looks to the future (0.53) when meson will
have this functionality, but also immediately it fixes these tests in
cross compiles by causing them to return 77, which meson interprets as
skip.
Acked-by: Kristian H. Kristensen <hoegsberg@google.com>
MSVC is generally happy, but mingw errors. I've spent as much time
(several days) trying to squash all of these warnings and I'm done with
it, just leave them as warnings with MinGW.
Acked-by: Kristian H. Kristensen <hoegsberg@google.com>
This has always been present in the scons build, so it should be in
the meson build as well.
Acked-by: Eric Engestrom <eric.engestrom@intel.com>
Acked-by: Kristian H. Kristensen <hoegsberg@google.com>
Mesa uses the lib prefix, and doesn't use a version for it's dynamic
libraries, which meson defaults to.
v2: - this patch
Acked-by: Eric Engestrom <eric.engestrom@intel.com>
Acked-by: Kristian H. Kristensen <hoegsberg@google.com>
It crashes hard (pop-up window and all).
v2: - Change comment to FIXME
Reviewed-by: Eric Engestrom <eric.engestrom@intel.com>
Acked-by: Kristian H. Kristensen <hoegsberg@google.com>
I can't figure out why symbols are being exposed that shouldn't.
v2: - change comment to FIXME
Reviewed-by: Eric Engestrom <eric.engestrom@intel.com>
Acked-by: Kristian H. Kristensen <hoegsberg@google.com>
They require the pipe-loaders, which require xmlconfig, which doesn't
build with msvc.
Acked-by: Eric Engestrom <eric.engestrom@intel.com>
Acked-by: Kristian H. Kristensen <hoegsberg@google.com>
There are quite a few tests that require getopt, when using MSVC we need
to use the bundled version of getopt since there isn't a system version.
Acked-by: Kristian H. Kristensen <hoegsberg@google.com>
Because the macros for exporting dll symbols and using TLS are mutually
exclusive.
Reviewed-by: Eric Engestrom <eric.engestrom@intel.com>
Acked-by: Kristian H. Kristensen <hoegsberg@google.com>
This makes two changes for SWR,
The first is that it reorders the arguments to try to put the ICL ones
first. This is required to support older versions of meson that don't
add enough "error in this case" switches to ICL, which causes it to
happy accept -mavx (for example) even though it doesn't support them,
resulting in compilation failures.
The second is to fix the names of the libraries, setting the soversion
to '' will result in <lib>.dll, instead of <lib>-0.dll. Since these are
not versioned dll's, but implement an internal API we should communicate
that. It's also what scons does.
Acked-by: Kristian H. Kristensen <hoegsberg@google.com>
There isn't an obvious command line switch here, /arch:AVX *might* be
the right thing, but meson doesn't know what to do here either and
leaves the -msse4.1 and -mstackrealign.
Reviewed-by: Eric Engestrom <eric.engestrom@intel.com>
Acked-by: Kristian H. Kristensen <hoegsberg@google.com>
v2: - Add missing D to pound define
- Simply define the variable rather than set it to 1 (mirrors
android.mk not scons)
Acked-by: Kristian H. Kristensen <hoegsberg@google.com>
There's a mingw bug for this, it exports __builtin_posix_memalign but
not posix_memalign, so the check will succeed, but compiling will fail.
Reviewed-by: Eric Engestrom <eric.engestrom@intel.com>
Acked-by: Kristian H. Kristensen <hoegsberg@google.com>
v2: - set so_version to '' (only affects windows)
- always set lib prefix to 'lib', even on msvc
v5: - key NO_EXPORTS on shared glapi instead of gles.
Acked-by: Kristian H. Kristensen <hoegsberg@google.com>
v4: - Fix check for broken mingw (should be for x86 not x86_64)
- Add comment about why check is needed
Acked-by: Kristian H. Kristensen <hoegsberg@google.com>
v4: - Handle enable gles properly
- Add comments about what various #defines do
v5: - key NO_EXPORTS on shared glapi instead of gles.
Acked-by: Kristian H. Kristensen <hoegsberg@google.com>
v4: - Retain scons comments for windows specific defines
v5: - key GLAPI_NO_EXPORTS off of shared-glapi instead of gles
Acked-by: Kristian H. Kristensen <hoegsberg@google.com>
These are needed to control the export or symbols due to differences
between the way windows and *nix handle symbol exports.
Reviewed-by: Eric Anholt <eric@anholt.net> (v2)
Acked-by: Kristian H. Kristensen <hoegsberg@google.com>
v5: - key NO_EXPORT off of shared-glapi instead of gles
v4: - Fix typo in warning code (4246 -> 4267)
- Copy comments from scons for what MSVC warnings codes do
- Merge linker argument changes into this commit
v5: - Add /GR- on windows if LLVM is build without rtti (equivalent to
GCc's -fno-rtti')
- Add /wd4291, which is catching the same hting that
-Wno-non-virtual-dtor is on GCC/Clang
Acked-by: Kristian H. Kristensen <hoegsberg@google.com>
MinGW defines only _WIN32, but doesn't have fcntl, so we need to use the
windows path.
Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Acked-by: Kristian H. Kristensen <hoegsberg@google.com>
Not sure if this is a bug in the user or not, but some CTS
tests fail due to using an 8 byte constant buffer.
Fixes: KHR-GLES31.core.layout_binding.block_layout_binding_block_VertexShader
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
since images are a single level, minify before passing the w/h
to draw.
Fixes: KHR-GLES31.core.shader_image_size.basic-nonMS-vs-*
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Due to use vmovdqa instructions in the asm, which require 16-byte
aligned buffers.
This fixes a crash in
KHR-GLES31.core.texture_buffer.texture_buffer_texture_buffer_range
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Preparation for a later commit.
Fixes: 93df862b6a ("meson: re-add incorrect pkg-config files with GLVND for backward compatibility")
Signed-off-by: Eric Engestrom <eric.engestrom@intel.com>
Reviewed-by: Dylan Baker <dylan@pnwbakers.com>
This reflects better what is provided by glvnd or not.
Fixes: 93df862b6a ("meson: re-add incorrect pkg-config files with GLVND for backward compatibility")
Signed-off-by: Eric Engestrom <eric.engestrom@intel.com>
Reviewed-by: Dylan Baker <dylan@pnwbakers.com>
SCons and Meson have never supported that feature, and Autotools was
deleted over 6 months ago and no-one complained yet, so it's pretty
obvious nobody cares about it.
Fixes: 95aefc94a9 ("Delete autotools")
Signed-off-by: Eric Engestrom <eric.engestrom@intel.com>
Acked-by: Dylan Baker <dylan@pnwbakers.com>
This is a security feature to disallow malicious apps from passing
a buffer that is too small.
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
To abstract things a bit, this adds a helper function in radv_android.c.
However, this means we have to link in radv_android.c on non-android as
well, which means some scaffolding changes.
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Since we really cannot share them ever.
Also remove an unused switch.
Fixes: b70829708a "radv: Implement VK_KHR_external_memory"
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Derived from the Intel code.
For the internal format we just use the internal Vulkan format,
as we have Vulkan formats for all android formats we care about.
For the ycbcr properties we just do something. I do not have a real
clue what would be recommended.
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
The minigbm comment really says it all. We should
fix minigbm as well, but for now this is the more
robust solution.
Note that this only changes width and height for
the surface creation, not for the image and hence
also not for the sampler, where it would wreak
havoc due to the normalized coords.
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
We want this flexibility because in GFX10 we lose any stride fields,
so we have to make sure our width/height are in alignment with
the external image we import.
Furthermore, we need the ability to inject tiling modifiers on import
time which is strictly after create time for Android. So, with the
layout & patch functions being fully independent of pCreateInfo, we
can delay it until import/bind time.
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Run dEQP on boards with Mali 400 and 450 in Baylibre's lab.
There's lots of skipped tests because of crashes and undetermined
behavior. May be a good idea to run the tests with valgrind and fix any
issues found.
Signed-off-by: Tomeu Vizoso <tomeu.vizoso@collabora.com>
Reviewed-by: Vasily Khoruzhick <anarsoul@gmail.com>
Reviewed-by: Neil Armstrong <narmstrong@baylibre.com>
As the non-LAVA runner script does, have per-GPU version files listing
the tests that are to be skipped, due to being very slow, unstable, etc.
Signed-off-by: Tomeu Vizoso <tomeu.vizoso@collabora.com>
Reviewed-by: Vasily Khoruzhick <anarsoul@gmail.com>
Reviewed-by: Neil Armstrong <narmstrong@baylibre.com>
Create basic aub_context on GEM_CONTEXT_CREATE.
Set it up and submit a context + ring + pphwsp during execbuf
submission, if it has not been initialized yet.
v2: Write the HWSP only once per engine (Lionel).
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
v2:
- Only dump context if there were no erros (Lionel).
- Store counter for context handles in aub_file (Lionel).
v3:
- Add a comment about aub_context -> GEM context (Lionel).
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
We want to be able to create contexts on demand, and increase the GGTT
as needed for that. Use the aub_map_ggtt() function for that.
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
We want to reuse it in execlists_setup().
v2: Rename it to write_ggtt_ptes() (Lionel).
v3: Rename it to aub_map_ggtt() (Lionel).
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
When the timestamp is not ready (ie. UINT64_MAX), the availabily bit
should be zero. The previous code used to copy the timestamp value
as the availabily bit and that's completely wrong.
Because it's not that simple to emit a conditional with the CP, the
driver now uses a compute shader for copying timestamp query results.
Fixes dEQP-VK.pipeline.timestamp.misc_tests.reset_query_before_copy.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Otherwise, the GPU might write timestamp queries after the reset
operation. This is similar to other query operations.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
These are not needed anymore, since PhyReg has an implicit
conversion operator that can convert it to unsigned int,
which is equivalent to accessing this field.
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
According to LLVM, branches with an offset of 0x3f are buggy.
v2: (by Timur Kristóf)
- extract the GFX10 specific part to its own function
Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
The DLC bit is now set to 1 for all loads when GLC is also set,
but cleared to 0 for all stores (otherwise it causes issues),
and also cleared to 0 for atomics.
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Also remove img_format from aco_ir, since it can be calculated
from dfmt and nfmt. So only the assember needs to deal with it.
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
We'd like to use some functions, for example some
ac_shader_util functions in ACO, so we need to link
ACO to AC.
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
We'd like to include some of these in C++ code later.
Specifically, ACO is written in C++ and we would like to use
some of this code in ACO in order to avoid code duplication.
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Specifically when reading the primitive counters.
This fixed ~700 CTS tests using this pattern:
dEQP-GLES3.functional.transform_feedback.*
when run after tests like
dEQP-GLES3.functional.prerequisite.read_pixels on the same
caselist. When run individually those tests were passing because
prim_counts_offset was zero.
Fixes: 0f2d1dfe65 ("v3d: use the GPU to
record primitives written to transform feedback")
Reviewed-by: Jose Maria Casanova Crespo <jmcasanova@igalia.com>
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
The initial patch only fixed up the NIR path, but forgot
the TGSI path needed fixing as well.
Fixes: f92226931b ("st/mesa: Prefer R8 for bitmap textures")
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
With libc++ (LLVM's STL implementation), the original code does not compile because an
appropriate vector constructor cannot be found (for the _ForwardIterator one, requirement
is_constructible is not satisfied).
<sys/param.h> is required for NetBSD version detection,
and __NetBSD__ must be used to detect even on older releases.
Reviewed-by: Eric Engestrom <eric.engestrom@intel.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
eglGetDisplay is awful because you have to inspect the pointer you're
given and guess what type of native display it corresponds to. We make
it worse by caching the type of the first such display we detect, so if
the second call to eglGetDisplay is to a different display type, kaboom.
Fortunately this is a problem that can be solved with the delete key.
Reviewed-by: Eric Engestrom <eric.engestrom@intel.com>
Closes: https://gitlab.freedesktop.org/mesa/mesa/issues/156
Previously, this could have made the resource divergent in code like
that which is genereated by nir_lower_non_uniform_access.
Fixes: da8ed68a ('nir: replace nir_move_load_const() with nir_opt_sink()')
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Previously, for code like:
loop {
loop {
a = load_ubo()
}
use(a)
}
adjust_block_for_loops() would return the block before the first loop.
Now we compute the range of allowed blocks and then walk the dominance
tree directly, guaranteeing directly that we always choose a block that
dominates all the uses and is dominated by the definition.
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
And use a new p_discard_early_exit instruction. This fixes some cases
where a definition having the same register as an operand causes issues.
v2: rename instruction to p_exit_early_if
v2: modify the existing instruction instead of creating a new one
v3: merge the "i == num - 1" IFs
Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Cloning texture loads isn't a good idea since we may move it into
a block that is not shared between all the invocations of the shader.
We'd like to avoid that since it may result in undefined behavior.
Reviewed-by: Andreas Baierl <ichgeh@imkreisrum.de>
Signed-off-by: Vasily Khoruzhick <anarsoul@gmail.com>
Without this, the test jobs could spuriously run after the container
job failed or was cancelled, even if the build job didn't run at all.
Reviewed-by: Tomeu Vizoso <tomeu.vizoso@collabora.com>
The spec has probably been misinterpreted during RADV bringup.
This fixes GPU hangs with dEQP-VK.binding_model.*offset_nonzero*.
Fixes: f4e499ec79 ("radv: add initial non-conformant radv vulkan driver")
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
If it's not available, we fall back to A8. This should work on all drivers,
because we depend on it in the display-list code already.
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
This should help avoid stalls in the pixel mask array in certain
non-promoted depth cases. It especially helps for Z16, as each bit
in the PMA corresponds to two pixels when using Z16, as opposed to
the usual one pixel.
Improves performance in GFXBench5 TRex by 22% (n=1).
Fixes Piglit's gl-2.1-polygon-stipple-fs on iris.
Fixes: 63f24c3c01 ("gallium: Enable MESA_framebuffer_flip_y")
Reviewed-by: Fritz Koenig <frkoenig@google.com>
Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
Implement glFramebufferParameteriMESA on GLES 3 so
that the extension is not dependant on GLES 3.1
Reviewed-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
bound_vertex_buffers doesn't include extra draw parameters buffers.
Tracking this correctly is kind of complicated, and iris_destroy_state
isn't exactly in a hot path, so just loop over all VBO bindings.
Fixes: 4122665dd9 (iris: Enable ARB_shader_draw_parameters support)
Reported-by: Sergii Romantsov <sergii.romantsov@globallogic.com>
On Solaris, sys/sysmacros.h has long-deprecated copies of major() & minor()
but not makedev().
sys/mkdev.h has all three and is the preferred choice.
Let's make sure we check for all 3 major(), minor() and makedev().
Reported-by: Alan Coopersmith <alan.coopersmith@oracle.com>
Signed-off-by: Eric Engestrom <eric.engestrom@intel.com>
Reviewed-by: Alan Coopersmith <alan.coopersmith@oracle.com>
Tested-by: Alan Coopersmith <alan.coopersmith@oracle.com>
The list of AMD/ATI devices supported by radeon/r200/r300/r600 is
complete, so anything else must use radeonsi.
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
When only the depth/stencil bufs are cleared, we should make sure the
color content is reloaded into the tile buffers if we want to preserve
their content.
Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
glClear()s are expected to be the first thing GL apps do before drawing
new things. If there's already an existing batch targetting the same
FBO that has draws attached to it, we should make sure the new clear
gets a new batch assigned to guaranteed that the FB content is actually
cleared with the requested color/depth/stencil values.
We create a panfrost_get_fresh_batch_for_fbo() helper for that and
call it from panfrost_clear().
Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Hitting any fallback path on Broxton as we require clflushing the whole
buffer even for an upload of a subtexture. However, since gallium
provides a pbo upload path, allow it to sample packed RGB if supported.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
It gets used by the gallium auxiliary draw module, which gets used
pretty much always when LLVM is used as JIT.
At the same time most builds don't hit the issue here because the
shared library of LLVM contains all modules.
Fixes: d32690b43c ("gallivm: add coroutine pass manager support")
Closes: https://gitlab.freedesktop.org/mesa/mesa/issues/951
Reviewed-by: Gert Wollny <gert.wollny@collabora.com>
This commit is a step towards the goal of being able to build RADV
without LLVM. In the future we would like to offer the option to
use RADV solely with ACO. There is still a need for the common AMD
code located in amd/common but the LLVM specific parts need to be
separated.
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Acked-by: Marek Olšák <marek.olsak@amd.com>
Acked-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
This mirrors the intrinsics in the GLSL IR. One could imagine an
alternate definition where reading the semantic would account for the
READ_HELPER functionality, but that feels potentially dodgy and could be
subject to CSE unpleasantness.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
u_upload_mgr sets it, so that util_range_add can skip the lock.
The time spent in tc_transfer_flush_region decreases from 0.8% to 0.2%
in torcs on radeonsi.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Meson automatically tracks any file included by a file it already tracks,
and `pci_id_driver_map.h` & `loader.h` are included by `loader.c`, while
`loader_dri3_helper.h` is included by `loader_dri3_helper.c`.
Signed-off-by: Eric Engestrom <eric.engestrom@intel.com>
Acked-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
These can appear after loop unrolling.
v2: stylistic changes
v2: replace state->mem_ctx with state->shader
v2: add bounds checking
v3: use nir_intrinsic_range() for bounds checking
v3: fix issue where partially out-of-bounds reads are replaced with undefs
v4: fix merge conflicts during rebase
v5: split into two commits
v6: set constant_data to NULL after freeing (fixes nir_sweep()/Iris)
v7: don't remove the constant data if there are no constant loads
Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Connor Abbott <cwabbott0@gmail.com> (v6)
Acked-by: Ian Romanick <ian.d.romanick@intel.com>
We only have the subgroup variant in NIR (equivalent to clockARB), so
only support that for now.
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
In the test stage, we can use any of the two container images as we
arent going to do anything architecture-dependent when submitting the
jobs to LAVA.
But if we are in a pipeline in which the images need to be rebuilt and
one finishes much earlier than the other, it could happen that the test
job that executes first fails to find the container image.
To avoid that, have each job in the test stage to use the image that has
been already implicitly built by depending on the build job for the
given arch.
Signed-off-by: Tomeu Vizoso <tomeu.vizoso@collabora.com>
Reviewed-by: Michel Dänzer <mdaenzer@redhat.com>
This reverts commit 19546108d3.
This commit breaks the build because lima implements
->set_damage_region(). I guess we'll need more discussion before
removing the ->set_damage_region() hook.
This reverts commit 492ffbed63.
BACK_LEFT attachment can be outdated when the user calls
KHR_partial_update(), leading to a damage region update on the
wrong pipe_resource object.
Let's not expose the ->set_damage_region() method until the core is
fixed to handle that properly.
Cc: mesa-stable@lists.freedesktop.org
Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Acked-by: Daniel Stone <daniels@collabora.com>
In preparation for testing drivers other than Panfrost in LAVA labs.
Signed-off-by: Tomeu Vizoso <tomeu.vizoso@collabora.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
Include Panfrost's gitlab.ci.yml file from Mesa's main .gitlab-ci.yml so
we test on devices with Panfrost.
This uses LAVA to schedule jobs in the devices and will be the base for
testing Etnaviv, Lima, etc.
Signed-off-by: Tomeu Vizoso <tomeu.vizoso@collabora.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
This is a port of Nanley's 904c2a617d
from i965 to iris.
One concern is that iris uses larger batches, and also emits far fewer
commands, so we may come closer to the 500 limit within a batch, and
could need to supplement this with actual counting. Manhattan 3.0 had
239 3DSTATE_CONSTANT_PS packets in a batch, Unigine Valley had 155.
So it seems like we're still in the realm of safety.
We're missing the offset of the slice in the subslice mask...
This worked for most platforms that don't have first slice fused off
because we would reread the same mask from slice0 again and again...
Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Fixes: c1900f5b0f ("intel: devinfo: add helper functions to fill fusing masks values")
Gitlab: https://gitlab.freedesktop.org/mesa/mesa/issues/1869
Reviewed-by: Mark Janes <mark.a.janes@intel.com>
We were supplying __DRI2_THROTTLE_SWAPBUFFER, rather than the obvious
choice of __DRI2_THROTTLE_COPYSUBBUFFER. This meant that we hit the
swap-based frame throttling. glXCopySubBuffer doesn't seem like it's
intended to be a frame boundary, so we'd like to avoid this throttling.
Tested-by: Michel Dänzer <mdaenzer@redhat.com> # DRI3 only
Reviewed-by: Michel Dänzer <mdaenzer@redhat.com>
glXCopySubBufferMESA copies data from the back buffer to the front,
so it needs to perform a MSAA downsampling operation just like
glXSwapBuffers would.
Currently, the CopySubBuffer implementations supply a throttle reason
of __DRI2_THROTTLE_SWAPBUFFERS, so they hit this path and work today.
But we'd like to avoid swapbuffer throttling in this case, so the next
patch will change that reason.
Tested-by: Michel Dänzer <mdaenzer@redhat.com> # DRI3 only
Reviewed-by: Michel Dänzer <mdaenzer@redhat.com>
I thought I fixed this, but I guess I must have broken it again.
Fixes various dEQP-VK.draw.* tests
Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Working on the algebraic implementation, I was being driven nuts by my
editor not highlighting and handling indentation for the C code. It turns
out that it's basically not pass-specific code, and we can move it over to
the relevant .c file. Replaces 30KB of code with 34KB of data on my i965
build. No perf diff on shader-db (n=3)
Reviewed-by: Ian Romanick <ian.d.romainck@intel.com>
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
This lets us memoize range analysis work across instructions. Reduces
runtime of shader-db on Intel by -30.0288% +/- 2.1693% (n=3).
Fixes: 405de7ccb6 ("nir/range-analysis: Rudimentary value range analysis pass")
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Having passes generate these is just making more work for copy
propagation (and thus probably calling more optimization passes)
later. Noticed while trying to debug nir_opt_algebraic()
top-to-bottom having O(n^2) behavior due to not finding new matches in
replacement code.
Reviewed-by: Ian Romanick <ian.d.romainck@intel.com>
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
This matches what we do for uses_sample_qualifier, and what we
do in ir_set_program_inouts.cpp as well.
Reviewed-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
This simplifies ACO and allows the lowered code to be optimized (in
particular, constant folded).
Totals from affected shaders:
SGPRS: 1776 -> 1776 (0.00 %)
VGPRS: 1436 -> 1436 (0.00 %)
Spilled SGPRs: 0 -> 0 (0.00 %)
Spilled VGPRs: 0 -> 0 (0.00 %)
Private memory VGPRs: 0 -> 0 (0.00 %)
Scratch size: 0 -> 0 (0.00 %) dwords per thread
Code Size: 203452 -> 203564 (0.06 %) bytes
LDS: 0 -> 0 (0.00 %) blocks
Max Waves: 103 -> 103 (0.00 %)
At least some of the code size increase seems to be from literals being
applied to instructions as a result of constant folding.
v2: remove fmod/frem handling in init_context()
Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
If the instruction interpolateAtCentroid is used the extra interpolator
must also be enabled in the state.
Fixes: fs-interpolateatcentroid-block
Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Now that we have live_out calculated per block as metadata, calculating
liveness of an instruction at a given point in the program becomes O(n)
to the size of the block worst-case, rather than O(n) the program.
Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Callers should have liveness info ready. Ideally we'd have a nice
metadata tracking framework like NIR to handle this automatically, but
for now this will allow us to make forward progress... when we're about
to do something with liveness, invalidate everything ahead to force a
clean calculation.
Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
This will allow us to explicitly invalidate liveness analysis results so
we can cache liveness results.
Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
By definition, once liveness analysis has occurred:
live_out = OR {succ} succ->live_in
Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
There are unfortunately two distinct liveness analysis passes in the
compiler right now -- one good (but complex) pass used by RA based on
solving data flow equations, and one awful (but simple) pass used for
dead code elimination and bundling based on an abstract walk of the AST.
Let's move RA's pass into shared code so we can work on unifying.
Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
This allows us to fill in ctx->temp_count explicitly, even if we haven't
squished down the MIR.
Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
We already enforce this with the SSA/register distinction in the
backend. There is no need to duplicate this logic merely for an assert.
Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Now that we have track inter-batch dependencies, the flush done in
panfrost_set_framebuffer_state() is no longer needed. Let's get rid of
it.
Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Now that we have all the pieces in place to support pipelining batches
we can get rid of the drmSyncobjWait() at the end of
panfrost_batch_submit().
Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
We don't have to flush all batches when we're only interested in
reading/writing a specific BO. Thanks to the
panfrost_flush_batches_accessing_bo() and panfrost_bo_wait() helpers
we can now flush only the batches touching the BO we want to access
from the CPU.
This fixes the dEQP-GLES2.functional.fbo.render.texsubimage.* tests.
Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
This is needed if we want to free the panfrost_batch object at submit
time in order to not have to GC the batch on the next job submission.
Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Will be useful to make the ioctl(WAIT_BO) call conditional on BOs that
are not exported/imported (meaning that all GPU accesses are known
by the context).
Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
This will allow us to only flush batches touching a specific resource,
which is particularly useful when the CPU needs to access a BO.
Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
And use it in panfrost_flush() to flush all batches, and not only the
one currently bound to the context.
We also replace all internal calls to panfrost_flush() by
panfrost_flush_all_batches() ones.
Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
The panfrost_fence logic currently waits on the last submitted batch,
but the batch serialization that was enforced in
panfrost_batch_submit() is about to go away, allowing for several
batches to be pipelined, and the last submitted one is not necessarily
the one that will finish last.
We need to make sure the fence logic waits on all flushed batches, not
only the last one.
Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
The idea is to track which BO are being accessed and the type of access
to determine when a dependency exists. Thanks to that we can build a
dependency graph that will allow us to flush batches in the correct
order.
Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
We'll soon need to freeze a batch not only when it's flushed, but also
when another batch depends on us, so let's add a helper to avoid
duplicating the logic.
Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
We just replace the per-context out_sync object by a pointer to the
the fence of the last last submitted batch. Pipelining of batches will
come later.
Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
So we can store the flags as data and keep the BO as a key. This way
we keep track of the type of access done on BOs.
Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
The type of access being done on a BO has impacts on job scheduling
(shared resources being written enforce serialization while those
being read only allow for job parallelization) and BO lifetime (the
fragment job might last longer than the vertex/tiler ones, if we can,
it's good to release BOs earlier so that others can re-use them
through the BO re-use cache).
Let's pass extra access flags to panfrost_batch_add_bo() and
panfrost_batch_create_bo() so the batch submission logic can take the
appropriate when submitting batches. Note that this information is not
used yet, we're just patching callers to pass the correct flags here.
Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
The CTS finally has agreed to drop the requirement for a
565-no-depth-no-stencil config for ES 3.0. Hence we can now remove the
code to satisfy this requirement using a pbuffer-only visual with
whatever other buffers the driver happens to have given us.
This reverts commit 82607f8a90,
commit 6ad31c4ff3 and
commit dacb11a585.
v2:
- Reference the VK-GL-CTS issue (Eric E.).
v3:
- Don't revert
fc21394bc4 ("egl: Quiet warning about front buffer rendering for pixmaps/pbuffers")
(Kenneth).
References: VK-GL-CTS issue 1601.
Cc: mesa-stable@lists.freedesktop.org
Signed-off-by: Andres Gomez <agomez@igalia.com>
Acked-by: Eric Engestrom <eric.engestrom@intel.com>
Acked-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
This script is responsible for generating an entire page in the
docs/relnotes/ directory. It includes a template for the page, and uses
mako to fill in the necessary bits. It is designed to be purely fire and
forget, calculating previous versions, shortlogs, bug fixes, and dates.
Acked-by: Eric Engestrom <eric.engestrom@intel.com>
Acked-by: Juan A. Suarez <jasuarez@igalia.com>
The next patch is going to introduce a tool that creates the entire
release html page for us, without any user intervention. As such we
can't be editing it. To that end the script will read the
new_features.txt file to get a list of new features.
This is a flat text file, one entry per line.
Acked-by: Eric Engestrom <eric.engestrom@intel.com>
Acked-by: Juan A. Suarez <jasuarez@igalia.com>
On 64 bits platforms, some atomic operations like __sync_fetch_and_add()
have constant time, but on 32 bits platforms they are implemented with a
loop and might take much longer.
Additionally, it seems like if their operands are not aligned to 64
bits, they also require extra memory accesses. From the Intel
Architecture's Developer Manual Vol. 1, 4.1.1:
"A word or doubleword operand that crosses a 4-byte boundary or a
quadword operand that crosses an 8-byte boundary is considered
unaligned and requires two separate memory bus cycles for access."
Forcing the u64 field to be aligned to 64 bits seems to make the unit
tests that are stressing this finish much faster.
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
This replaces to old Bugzilla: tag, which no longer makes sense because
we don't use bugzilla anymore.
Reviewed-by: Eric Anholt <eric@anholt.net> (v1)
Reviewed-by: Eric Engestrom <eric.engestrom@intel.com>
v1 by Topi Pohjolainen
v2,v3 by Anuj Phogat:
- Apply for gen >= 11
- Remove wa_bug_xxx function
- Use helper functions
Signed-off-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
This lowers fmod and frem at NIR level like RadeonSI. fmod is
already lowered directly in NIR->LLVM, and frem will be lowered by
LLVM anyways.
This fixes a LLVM crash with:
dEQP-VK.glsl.builtin.precision_fp16_storage32b.frem.compute.scalar.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
... because it's wrong to do so. The error path out of
dri2_initialize_drm ends with dri2_display_destroy, which calls
functions in the vtable we're trying to set up, so if we dlclose the
driver then those function pointers will point off into space and things
crash.
Noticed this because after !1923 eglinfo would crash when setting up the
GBM platform. This was something of a cascade failure, because my kernel
is too old for DRM_IOCTL_I915_GETPARAM to work without DRM_AUTH, so i965
wouldn't load. platform_drm.c then got very confused when it tries to
load swrast as a dri2 driver.
Reviewed-by: Eric Anholt <eric@anholt.net>
uintptr_t is 32 bits in a 32-bits build, resulting in shifting out
of bounds.
Reviewed-by: Eric Engestrom <eric@engestrom.ch>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
We need the continue CS for referencing the tess/GDS/sample position BOs.
Fixes: 46e52df34d "radv: add tessellation ring allocation support. (v2)"
Fixes: e1dc3ab753 "radv/gfx10: allocate GDS/OA buffer objects for NGG streamout"
Fixes: 1171b304f3 "radv: overhaul fragment shader sample positions."
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Instead of a single cache shared between all jobs, but reduce the
maximum cache size to 1.5G (from 5G).
Rationale for smaller cache:
Pulling & pushing a 5G cache could take a long time. Consider
https://gitlab.freedesktop.org/mesa/mesa/-/jobs/684010 (click the "Show
complete raw" button to see timestamps): Pulling the cache took
1569927241-1569927194 = 47 seconds, pushing it 1569927671-1569927519
= 152, for a total of 199 seconds. The actual build took comparable
1569927518-1569927243 = 275 seconds, despite no cache hits from ccache.
In other words, the cache transfers almost doubled the job duration,
and they would have negated any build time benefits from ccache even
with a high cache hit rate.
Also, the smaller caches avoid blowing up storage requirements for them
too much.
Rationale for per-job caches:
Making a single cache significantly smaller might result in cached
build products from one job getting evicted by another job, reducing
the likelihood of cache hits from previous pipelines.
v2:
* Move up "ccache --max-size=1500M" call (Eric Engestrom)
Reviewed-by: Eric Engestrom <eric.engestrom@intel.com>
To truly to do this correctly, we'll have to fix the discrepancy between
drm_virtgpu_3d_transfer_to_host and virtio_gpu_transfer_host_3d. However,
this is a good starting point.
Since virtio-gpu only supports self-import and export, this should be fine.
Let's only do WINSYS_HANDLE_TYPE_FD for this currently.
Reviewed by: Robert Tarasov <tutankhamen@chromium.org>
The winsys might supply dimensions that are different than
those we calculate. In additional, it may supply virtualized
modifiers.
In practice, a stride != bpp * width and virtualized modifiers don't
happen yet, but the plan is to move in that direction.
Also make virgl_resource_layout static.
Reviewed by: Robert Tarasov <tutankhamen@chromium.org>
i915 will report ENODEV on generations prior to Haswell because there
is no point in reporting values on those. This is prior any fusing
could happen on parts with identical PCI ids.
This query call was previously only triggered on generations that
support performance queries, which happens to match generation for
which i915 reports topology, but the commit pointed below started
using it on all generations.
Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Gitlab: https://gitlab.freedesktop.org/mesa/mesa/issues/1860
Cc: <mesa-stable@lists.freedesktop.org>
Fixes: 96e1c945f2 ("i965: Move device info initialization to common code")
Reviewed-by: Mark Janes <mark.a.janes@intel.com>
Random hangs no longer happen, I'm actually not sure if they were
related to this.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Make sure to export the expected clear values to the depth
stencil attachment.
This fixes dEQP-VK.pipeline.depth_range_unrestricted.* on GFX10.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
The number of vertices has to be adjusted with the output primitive
type.
This fixes dEQP-VK.transform_feedback.simple.triangle_strip_*.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
The GS outputs are stored differently in the LDS storage, they
are indexed by out_idx which is incremented for each stored DWORD.
Thus, we need a different path for exporting the stream outputs.
This fixes a bunch of CTS failures when NGG GS is force enabled.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
The LDS storage allocated for stream outputs is 4 * N, where N
is the number of outputs. So, we have to store/load with N as index
and not with the output location as index.
This doesn't fix anything known but it should fix out-of-bounds
access and it also reduces the number of outputs written to the
LDS storage.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Some hardware has a bug with triangle strips and it is signalled by the
flag BUG_FIXED8 whether this bug has been fixed. So only enable triangle
strips when this flag is set.
Thanks: Jonathan Marek and Christian Gmeiner for the pointers
v2: Add TODO to indicate that the handling should be refined
(Jonathan & Christian)
Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Reviewed-by: Christian Gmeiner <christian.gmeiner@gmail.com>
It's unused here, and undefined in scons. It is used in targets/osmesa,
but it's properly defined there already.
Reviewed-by: Eric Engestrom <eric.engestrom@intel.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
On the Android Antutu benchmark we ran into an assert in ISL where the
(base layer + num layers) > total layers. It turns out the core of
mesa forgot to clear the _Layer variable, potentially leaving an
inconsistent value.
v2: Pull setting u->_Layer out of the conditional blocks (Jason)
Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Cc: <mesa-stable@lists.freedesktop.org>
Reviewed-by: Eric Engestrom <eric.engestrom@intel.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
In converting to shift/size-based validation, we lost a condition from
the ARGB/XRGB equivalence check, which left it working one way round
but not the other, and broke applications like glmark2-es2-drm on some
platforms. Restore the equivalent check that *both* configs actually
have an alpha channel before considering a mismatch.
Fixes: 7b4ed2b513 ("egl: Convert configs to use shifts and sizes instead of masks")
Signed-off-by: Robin Murphy <robin.murphy@arm.com>
Reviewed-by: Adam Jackson <ajax@redhat.com>
Reviewed-by: Eric Engestrom <eric.engestrom@intel.com>
1. The hgl.c file is a read-only file versus read-write.
Ref: src/gallium/state_trackers/hgl/hgl.c
2. I've included the Haiku-specific patches I used to get a successful
build of Mesa 19.1.7 on Haiku using the meson/ninja build procedure.
Shows "[764/764] linking target ... libswpipe.so" at build completion.
v2:
Remove autotools files (Eric)
v3:
Update the patch
Reported-by: Ken Mays <kmays2000@gmail.com>
Tested-by: Ken Mays <kmays2000@gmail.com>
CC: mesa-stable@lists.freedesktop.org
Reviewed-by: Alexander von Gluck IV <kallisti5@unixzen.com>
After 41549a18e6 ("i965: Enable OpenGL 4.6 for Gen8+"), i965
implements GL_ARB_gl_spirv, GL_ARB_spirv_extensions and OpenGL 4.6.
After 15e439071d ("iris: Enable ARB_gl_spirv and ARB_spirv_extensions"),
iris implements GL_ARB_gl_spirv, GL_ARB_spirv_extensions and OpenGL
4.6.
v2:
- Explicit the support is for i965 and iris.
v3:
- Add also GL_ARB_spirv_extensions to the release notes (Alejandro).
Signed-off-by: Andres Gomez <agomez@igalia.com>
Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Found when building for Android in C99 mode. Include bitscan.h to ensure ffs is
available.
Fixes: 7b4ed2b5 ("egl: Convert configs to use shifts and sizes instead of masks")
Signed-off-by: Kevin Strasser <kevin.strasser@intel.com>
Reviewed-by: Eric Engestrom <eric.engestrom@intel.com>
The order of comparison has changed, so we need to invert the logic of
"insert_left" when using rb_tree_insert_at().
Fixes: dae33052db (util/rb_tree: Reverse the order of comparison
functions).
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
To enable EXT_demote_to_helper_invocation:
This extension adds a "demote" keyword that is similar to "discard" but
only suppresses subsequent writes and outputs to the framebuffer, and
does not terminate the execution of the invocation. For the remainder
of the execution, the invocation is "demoted" to act like a helper
invocation.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
From EXT_demote_to_helper_invocation, implemented with the existing
nir_intrinsic_is_helper_invocation.
Such builtin is necessary when using `demote` because we can't
redefine the value of gl_HelperInvocation (since it is an input
variable).
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
When the EXT_demote_to_helper_invocation extension is enabled,
`demote` is treated as a keyword, and produces an ir_demote.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
To represent the new `demote` keyword when using
EXT_demote_to_helper_invocation extension. Most of the changes are to
include it in the visitors.
Demote is not considered a control flow, so also include an empty
visit member function in ir_control_flow_visitor.
Only NIR actually supports `demote`, so assert the translations for
TGSI and Mesa's gl_program -- since the demote is not expected to
appear for those.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
We can't just check for the BO base address, we need to check for the
full address including any offset we may have applied. When updating
the address, we need to include the offset again.
Fixes: 5ad0c88dbe ("iris: Replace buffer backing storage and rebind to update addresses.")
A while back, Michael Larabel noticed that Paraview's Wavelet Volume
case runs significantly slower on iris than i965. It turns out this
is because we enable CCS_E for 32-bit floating point formats, while
i965 disables it, with an oblique comment saying that we benchmarked
it (on what exactly?) and determined that it was a loss.
Paraview uses both R32_FLOAT and R32G32B32A32_FLOAT, and I observed
large framerate drops when enabling CCS_E for either format. However,
several other benchmarks (Aztec Ruins, many Synmark cases) use 16-bit
floating point formats, with no apparent ill effects.
So, disable compression for 32-bit float formats for now, but leave it
enabled for 16-bit float formats as they seem to be working fine.
Improves performance in Paraview's Wavelet Volume test by 62% on a
Skylake GT4e.
Fixes: 3cfc6a207b ("iris: Fill out res->aux.possible_usages")
Tests done with llvm-config indicate that there are only 2 libraries in
irreader and not in engine, LLVMAsmParser and LLVMIRReader and both of them
are part of coroutines so I replaced irreader with coroutines and added
libraries unique to coroutines.
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
Now that we have constant adjustment logic abstracted, we can do this
safely. Along with the csel inversion patch, this allows many more
common csel ops to inline their condition in the bundle.
Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
If we can reuse constant slots from other instructions, we would like to
do so to include more instructions per bundle.
Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
If an instruction could be scheduled to vmul to satisfy the writeout
conditions, let's do that and save an instruction+cycle per fragment
shader.
Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
We still emit in-order but we switch to using the bundles created from
the new scheduler, which will allow greater flexibility and room for
out-of-order optimization.
Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
We require chosen instructions to be "close", to avoid ballooning
register pressure. This is a kludge that will go away once we have
proper liveness tracking in the scheduler, but for now it prevents a lot
of needless spilling.
v2: Lower threshold to 6 (from 8). Schedule is hurt, but a few shaders
that spilled excessively are fixed.
Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Derp
We can bundle two load/store together. This eliminates the need for
explicit load/store pairing in a prepass, as well.
Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Conditions for branches don't have a swizzle explicitly in the emitted
binary, but they do implicitly get swizzled in whatever instruction
wrote r31, so we need to handle that.
Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Conditional instructions (csel and conditional branches) require their
condition to be written to a special condition pipeline register (r31.w
for scalar, r31.xyzw for vector). However, pipeline registers are live
only for the duration of a single bundle. As such, the logic to schedule
conditionals correct is surprisingly complex. Essentially, we see if we
could stuff the conditional within the same bundle as the csel/branch
without breaking anything; if we can, we do that. If we can't, we add a
dummy move to make room.
Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
A bit of a kludge but allows setting an implicit dependency of synthetic
conditional moves on the actual condition, fixing code generated like:
vmul.feq r0, ..
sadd.imov r31, .., r0
vadd.fcsel [...]
The imov runs simultaneous with feq so it gets garbage results, but it's
too late to add an actual dependency practically speaking, since the new
synthetic imov doesn't have a node associated.
Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
In the future, we will want to keep track of which components of
constants of various sizes correspond to which parts of the bundle
constants, like in the old scheduler. For now, let's just stub it out
for a simple rule of one instruction with embedded constants per bundle.
We can eventually do better, of course.
Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
We don't actually do any scheduling here yet, but add per-tag helpers to
consume an instruction, print it, pop it off the worklist.
Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
It's not always obvious what the optimal bundle type should be. Let's
break out the logic to decide.
Currently set for purely in-order operation.
Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
After we've chosen an instruction, popped it off, and processed it, it's
time to update the worklist, removing that instruction from the
dependency graph to allow its dependents to be put onto the worklist.
Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
In the future, this routine will implement the core scheduling logic to
decide which instruction out of the worklist will be scheduled next, in
a way that minimizes cycle count and register pressure.
In the present, we are more interested in replicating in-order
scheduling with the much-more-powerful out-of-order model. So rather
than discriminating by a register pressure estimate, we simply choose
the latest possible instruction in the worklist.
Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
We would like to flatten a linked list of midgard_instructions into an
array of midgard_instruction pointers on the heap.
Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
It's not based on the writemask and it can't be inferred; it's just
intrinsic to the op itself.
Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
This worked better than my original v3d-local pass for just subs, and is a
huge win over not producing subs.
total instructions in shared programs: 6408469 -> 6167932 (-3.75%)
total threads in shared programs: 153784 -> 154104 (0.21%)
total uniforms in shared programs: 2157078 -> 1905823 (-11.65%)
total max-temps in shared programs: 904546 -> 895796 (-0.97%)
total spills in shared programs: 4959 -> 4993 (0.69%)
total fills in shared programs: 6558 -> 6670 (1.71%)
total sfu-stalls in shared programs: 25845 -> 25175 (-2.59%)
total inst-and-stalls in shared programs: 6434314 -> 6193107 (-3.75%)
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
There are some optimizations which are only implemented for additions
and some optimizations which assume that subtractions have been lowered.
By lowering all subtractions first and later recombine for backends
which prefer this option, we don't have to implement them twice.
This patch also moves lower_negate to nir_opt_algebraic_late() to enable
these optimizations for backends which make use of it.
Reviewed-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Upcoming changes to sub optimization will make this pass required. Over
the course of that series, we see uniforms +.46%, instructions -.24%
(seems like a fine tradeoff -- uniforms are 1/2 the size of instructions
as far as cache occupancy)
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Without this, it was theoretically possible for the jobs to run before
the docker image was ready.
v2:
* Use - list syntax instead of [] (Eric Engestrom)
Reviewed-by: Eric Engestrom <eric.engestrom@intel.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
This allows most build jobs to run before the stretch or arm64 docker
images are ready.
v2:
* Use - list syntax instead of [] (Eric Engestrom)
Reviewed-by: Eric Engestrom <eric.engestrom@intel.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
This allows the *-old-llvm jobs to run before the buster docker images
are ready.
v2:
* Use - list syntax instead of [] (Eric Engestrom)
Reviewed-by: Eric Engestrom <eric.engestrom@intel.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
This patch is based on 28e3f85e09/mingw-w64-mesa/link-ole32.patch but with tweaks to avoid MSVC build break when applied.
v2: Create Mingw platform alias pointing to windows host platform define to avoid spurious crosscompilation;
v3: Fix obviously wrong compiler flags for swr driver;
v4: Update original patch URL because it has been relocated;
v5: Don't bother patching autools stuff as it's not used by MSYS2 Mingw-w64 build and it's days are numbered anyway;
v6: After Mingw posix flag fix in 295851eb things are far simpler as we don't need more linking of uuid, ole32, version and shell32 than what is already in place.
As X86AsmPrinter component is gone, LLVMX86AsmPrinter got replaced
with LLVMRemarks, LLVMBitstreamReader and LLVMDebugInfoDWARF.
Tests done with llvm-config on both LLVM 8 and 9 indicate that
mcjit, bitwriter and x86asmprinter fully fit inside engine component.
On other platforms and with meson build mcdisassembler was used to replace
X86AsmPrinter but mcdisassembler also fully fits inside engine component
for LLVM>=8 according to same tests.
v2: Avoid duplicating code related to Mingw pthreads.
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
Cc: 19.1 19.2 <mesa-stable@lists.freedesktop.org>
On 19.1 this patch does not apply cleanly without 88eb2a1f
Looks like blob uses following values for uniforms buffer:
0 for 8 bytes
1 for 16 bytes
2 for 24 bytes
2 for 32 bytes
3 for 40 bytes
3 for 48 bytes
3 for 56 bytes
3 for 64 bytes
4 for 72 bytes
It all looks like log2(size / 8) rounded up, so let's do the same.
Fixes: 931fc2a7b3f9("lima: do not set the PP uniforms address lowest bits")
Reviewed-by: Icenowy Zheng <icenowy@aosc.io>
Signed-off-by: Vasily Khoruzhick <anarsoul@gmail.com>
Android building rules are added in src/amd/Android.compiler.mk
libmesa_aco static library is built conditionally to radeonsi
as done for vulkan.radv module
This will prevent Android build errors for non x86 systems
filter-out compiler/aco_instruction_selection_setup.cpp source,
as already included by compiler/aco_instruction_selection.cpp
and would cause several multiple definition linker errors
NOTE: libLLVM requires AMDGPU Disassembler to build radv with aco
Fixes: 93c8ebf ("aco: Initial commit of independent AMD compiler")
Fixes: a70a998 ("radv/aco: Setup alternate path in RADV to support the experimental ACO compiler")
Signed-off-by: Mauro Rossi <issor.oruam@gmail.com>
Fixes a few building errors similar to the following:
In file included from external/mesa/src/amd/compiler/aco_instruction_selection.cpp:26:
In file included from external/libcxx/include/algorithm:639:
external/libcxx/include/utility:321:9:
error: implicit instantiation of undefined template 'std::__1::array<aco::Temp, 4>'
_T2 second;
^
Fixes: 93c8ebf ("aco: Initial commit of independent AMD compiler")
Signed-off-by: Mauro Rossi <issor.oruam@gmail.com>
Fixes the following piglit test: fragdepth_gles2 (for ETNA_MESA_DEBUG=nir)
Signed-off-by: Jonathan Marek <jonathan@marek.ca>
Reviewed-by: Christian Gmeiner <christian.gmeiner@gmail.com>
Note it can still be improved a bit:
* Use alu swizzle to determine if src is scalar
* Take into account new immediates in the multiple uniform src lowering
Signed-off-by: Jonathan Marek <jonathan@marek.ca>
Reviewed-by: Christian Gmeiner <christian.gmeiner@gmail.com>
This can improve performance by allowing the LAST_VARYING_2X bit to be
set when possible (and possibility more benefits on HALTI5 where the
number of components is set for each varying).
Signed-off-by: Jonathan Marek <jonathan@marek.ca>
Reviewed-by: Christian Gmeiner <christian.gmeiner@gmail.com>
LOAD starts reading into the first enabled destination component, and
doesn't skip disabled components, so we need to allocate a destination with
contiguous components.
Signed-off-by: Jonathan Marek <jonathan@marek.ca>
Reviewed-by: Christian Gmeiner <christian.gmeiner@gmail.com>
Only invert front facing when glFrontFace is GL_CW.
Fixes following deqp test:
dEQP-GLES2.functional.shaders.builtin_variable.frontfacing
Signed-off-by: Jonathan Marek <jonathan@marek.ca>
Reviewed-by: Christian Gmeiner <christian.gmeiner@gmail.com>
The PP uniforms address register in render state is not a direct pointer
to the uniforms storage -- instead, it points to an one-item array, and
the array item is the real pointer to the uniforms storage.
This register reuses some of its LSBs as a size field. Currently the
size is set according to the length of the real uniforms storage.
However, as the register itself contains only a pointer to the one-item
array, the size field should be set to the length of the one-item array
and subtract it by 1, which means a fixed value of 0. That means we can
just omit it now.
Test shows this should be the correct approach to set this register.
Signed-off-by: Icenowy Zheng <icenowy@aosc.io>
Reviewed-by: Vasily Khoruzhick <anarsoul@gmail.com>
glsl 4.4 spec section '5.9 expressions':
"The operator is multiply (*), where both operands are matrices or one operand is a vector and the
other a matrix. A right vector operand is treated as a column vector and a left vector operand as a
row vector. In all these cases, it is required that the number of columns of the left operand is equal
to the number of rows of the right operand. Then, the multiply (*) operation does a linear
algebraic multiply, yielding an object that has the same number of rows as the left operand and the
same number of columns as the right operand. Section 5.10 “Vector and Matrix Operations”
explains in more detail how vectors and matrices are operated on."
This fix disallows a multiplication of incompatible matrices like:
mat4x3(..) * mat4x3(..)
mat4x2(..) * mat4x2(..)
mat3x2(..) * mat3x2(..)
....
CC: <mesa-stable@lists.freedesktop.org>
Reviewed-by: Eric Anholt <eric@anholt.net>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=111664
Signed-off-by: Andrii Simiklit <andrii.simiklit@globallogic.com>
According to the 1.1.123 spec:
"The implementation will attempt to create all pipelines, and only
return VK_NULL_HANDLE values for those that actually failed."
Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
I was inheriting the one from src/freedreno with funny tabs, while
this driver is written with normal Mesa 3-space indents.
Unfortunately I have to add both files, because I use emacs and emacs
prefers .dir-locals to .editorconfig :(
Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
We include shader_enums.h from freedreno's compiler for both GL and
Vulkan, and the main/config.h include resulted in polluting the
namespace with things like MAX_VIEWPORTS that other Vulkan drivers use
as their driver-specific maximums.
Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
Without this, we were DCEing flag writes because we didn't think their
results were used because we didn't understand that an ANY32 predicate
actually read all the flags.
Fixes: df1aec763e "i965/fs: Define methods to calculate the flag..."
Reviewed-by: Matt Turner <mattst88@gmail.com>
Passes most of piglit's tests regarding arb_framebuffer_object
and unlocks some more piglit tests.
Signed-off-by: Christian Gmeiner <christian.gmeiner@gmail.com>
Reviewed-by: Jonathan Marek <jonathan@marek.ca>
We are using util_resource_copy_region(..) as fallback which supports
different formats for src and dst. Improves the experience when running
deqp or piglit with a debug build.
Signed-off-by: Christian Gmeiner <christian.gmeiner@gmail.com>
Reviewed-by: Jonathan Marek <jonathan@marek.ca>
Prior to xvmc 1.0.12 libxvmc incorrectly required libxv, but that was
fixed. This results in compilation failures for the gallium xvmc tracker
and tools. This patch fixes that by explicitly linking to libxv.
Fixes: 22a817af8a
("meson: build gallium xvmc state tracker")
Closes: https://gitlab.freedesktop.org/mesa/mesa/issues/1844
Reviewed-by: Adam Jackson <ajax@redhat.com>
We don't want to require a visual for the drawable, because there exist
fbconfigs that don't correspond to any visual (say a 565 pixmap|pbuffer
config on a depth-24 display). Fortunately, we don't need one either.
Passing the visual to XCreateImage serves only to fill in the XImage's
{red,green,blue}_mask fields, which libX11 itself never uses, they exist
only for the client's convenience, and we don't care. And we already
have the drawable depth in glx_config::rgbBits. So replace the
XVisualInfo field in the drawable private with a pointer to the
glx_config.
Having done that driswCreateGCs becomes trivial, so inline it into its
caller.
Gitlab: https://gitlab.freedesktop.org/mesa/mesa/issues/1194
Reviewed-by: Eric Anholt <eric@anholt.net>
There's no reason to have two GCs here. The only difference between
them is that swapgc would generate graphics exposures, except we only
ever use this GC for PutImage, and PutImage doesn't generate graphics
exposures. We also don't need to explicitly ChangeGC to GXCopy, because
that's the default.
Reviewed-by: Eric Anholt <eric@anholt.net>
Looks like r16_unorm might have precision issues.
dEQP-VK.api.copy_and_blit.core.image_to_image.all_formats.color.r16_unorm.r16_unorm.general_general
fails, but the dumped images in the xml are the same so
I'd guess the low bits are the issue.
r8_unorm and r16_uint work.
Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
3D blits & format reinterpretation are still TBD.
Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
We need the loop header phis for the outer exec masks. Needed for
dEQP-VK.glsl.demote.dynamic_loop_texture
Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
NIR may emit a single instrinsic to load several packed varyings,
but that's suboptimal for Utgard PP for several reasons:
- varyings that are used as sampler inputs can be passed using
pipeline register with increased precision
- we have small number of regs, so using a vec4 regs for storing
two vec2 varyings increases reg pressure.
Add NIR pass to split a single load into several loads and utilize
it in lima.
Reviewed-by: Qiang Yu <yuq825@gmail.com>
Signed-off-by: Vasily Khoruzhick <anarsoul@gmail.com>
According to radeonsi, GLM doesn't support WB alone, so
we have to set INV too when WB is set.
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Avoids getting a "load_output" in a case like this:
gl_Position = ubuf.MVP * ubuf.position[gl_VertexIndex];
frag_pos = gl_Position.xyz;
Signed-off-by: Jonathan Marek <jonathan@marek.ca>
Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
Acked-by: Eric Anholt <eric@anholt.net>
Lower these to something compatible with ir3, and save the descriptor set
and binding information.
Signed-off-by: Jonathan Marek <jonathan@marek.ca>
Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
Acked-by: Eric Anholt <eric@anholt.net>
The old version of the iterators relies on a &iter->field != NULL check
which works fine on older GCC but newer GCC versions and clang have
optimizations that break if you do pointer math on a null pointer. The
correct solution to this is to do the null comparisons before we do any
sort of &iter->field or use rb_node_data to do the reverse operation.
Acked-by: Michel Dänzer <mdaenzer@redhat.com>
Tested-by: Michel Dänzer <mdaenzer@redhat.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
This is based on the fix I used for the same problem on V3D. In this
case, it fixes all but the the
dEQP-GLES2.functional.texture.filtering.2d.*_npot cases of
dEQP-GLES2.functional.texture.filtering.2d.*'s failures.
Acked-by: Rob Clark <robdclark@chromium.org>
As Vasily discovered, the bit 7 of the word 1 of the texture descriptor
is set when reloading the framebuffer, to use framebuffer-based offset
rather than normalized one. This bit also works for regular textures to
enable accessing with non-normalized offset.
Add support for rectangle texture by setting this bit for
PIPE_TEXTURE_RECT.
Suggested-by: Vasily Khoruzhick <anarsoul@gmail.com>
Signed-off-by: Icenowy Zheng <icenowy@aosc.io>
Reviewed-by: Vasily Khoruzhick <anarsoul@gmail.com>
Per the valgrind output below, we were returning the pointer to freed
memory if none of the later conditional pointer assignments were
executed. This caused dEQP CI jobs to crash on certain runners,
presumably due to a double-free down the line.
Also, we were skipping to the out: label before the vendor_id & chip_id
variables used by it were initialized, resulting in broken
LIBGL_DEBUG=verbose output such as
libGL: pci id for fd 4: 51108f00:51108f00, driver radeonsi
Fixes: 5a545e355b "loader: always map the "amdgpu" kernel driver name to radeonsi (v2)"
==403== Invalid read of size 1
==403== at 0x4AFD576: surfaceless_probe_device (platform_surfaceless.c:316)
==403== by 0x4AFD915: dri2_initialize_surfaceless (platform_surfaceless.c:391)
==403== by 0x4AF5EEA: dri2_initialize (egl_dri2.c:984)
==403== by 0x4AF5EEA: dri2_initialize (egl_dri2.c:958)
==403== by 0x4AF1EEC: _eglMatchAndInitialize (egldriver.c:75)
==403== by 0x4AF1F3B: _eglMatchDriver (egldriver.c:96)
==403== by 0x4AE9367: eglInitialize (eglapi.c:617)
==403== by 0x1D99C9: tcu::surfaceless::EglRenderContext::EglRenderContext(glu::RenderConfig const&, tcu::CommandLine const&) [clone .constprop.57] (in /deqp/modules/gles2/deqp-gles2)
==403== by 0x1DABB0: tcu::surfaceless::ContextFactory::createContext(glu::RenderConfig const&, tcu::CommandLine const&, glu::RenderContext const*) const (in /deqp/modules/gles2/deqp-gles2)
==403== by 0x53EBD1: glu::createRenderContext(tcu::Platform&, tcu::CommandLine const&, glu::RenderConfig const&, glu::RenderContext const*) (in /deqp/modules/gles2/deqp-gles2)
==403== by 0x53EFE9: glu::createDefaultRenderContext(tcu::Platform&, tcu::CommandLine const&, glu::ApiType) (in /deqp/modules/gles2/deqp-gles2)
==403== by 0x1DE07A: deqp::gles2::Context::Context(tcu::TestContext&) (in /deqp/modules/gles2/deqp-gles2)
==403== by 0x1DB5EF: deqp::gles2::TestPackage::init() (in /deqp/modules/gles2/deqp-gles2)
==403== Address 0x56bd340 is 0 bytes inside a block of size 4 free'd
==403== at 0x48369AB: free (in /usr/lib/x86_64-linux-gnu/valgrind/vgpreload_memcheck-amd64-linux.so)
==403== by 0x4B01767: loader_get_driver_for_fd (loader.c:464)
==403== by 0x4AFD553: surfaceless_probe_device (platform_surfaceless.c:308)
==403== by 0x4AFD915: dri2_initialize_surfaceless (platform_surfaceless.c:391)
==403== by 0x4AF5EEA: dri2_initialize (egl_dri2.c:984)
==403== by 0x4AF5EEA: dri2_initialize (egl_dri2.c:958)
==403== by 0x4AF1EEC: _eglMatchAndInitialize (egldriver.c:75)
==403== by 0x4AF1F3B: _eglMatchDriver (egldriver.c:96)
==403== by 0x4AE9367: eglInitialize (eglapi.c:617)
==403== by 0x1D99C9: tcu::surfaceless::EglRenderContext::EglRenderContext(glu::RenderConfig const&, tcu::CommandLine const&) [clone .constprop.57] (in /deqp/modules/gles2/deqp-gles2)
==403== by 0x1DABB0: tcu::surfaceless::ContextFactory::createContext(glu::RenderConfig const&, tcu::CommandLine const&, glu::RenderContext const*) const (in /deqp/modules/gles2/deqp-gles2)
==403== by 0x53EBD1: glu::createRenderContext(tcu::Platform&, tcu::CommandLine const&, glu::RenderConfig const&, glu::RenderContext const*) (in /deqp/modules/gles2/deqp-gles2)
==403== by 0x53EFE9: glu::createDefaultRenderContext(tcu::Platform&, tcu::CommandLine const&, glu::ApiType) (in /deqp/modules/gles2/deqp-gles2)
==403== Block was alloc'd at
==403== at 0x483577F: malloc (in /usr/lib/x86_64-linux-gnu/valgrind/vgpreload_memcheck-amd64-linux.so)
==403== by 0x4EE5E09: strndup (strndup.c:43)
==403== by 0x4B010B1: loader_get_kernel_driver_name (loader.c:101)
==403== by 0x4B016AF: loader_get_driver_for_fd (loader.c:462)
==403== by 0x4AFD553: surfaceless_probe_device (platform_surfaceless.c:308)
==403== by 0x4AFD915: dri2_initialize_surfaceless (platform_surfaceless.c:391)
==403== by 0x4AF5EEA: dri2_initialize (egl_dri2.c:984)
==403== by 0x4AF5EEA: dri2_initialize (egl_dri2.c:958)
==403== by 0x4AF1EEC: _eglMatchAndInitialize (egldriver.c:75)
==403== by 0x4AF1F3B: _eglMatchDriver (egldriver.c:96)
==403== by 0x4AE9367: eglInitialize (eglapi.c:617)
==403== by 0x1D99C9: tcu::surfaceless::EglRenderContext::EglRenderContext(glu::RenderConfig const&, tcu::CommandLine const&) [clone .constprop.57] (in /deqp/modules/gles2/deqp-gles2)
==403== by 0x1DABB0: tcu::surfaceless::ContextFactory::createContext(glu::RenderConfig const&, tcu::CommandLine const&, glu::RenderContext const*) const (in /deqp/modules/gles2/deqp-gles2)
Acked-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
This new option can help debug shader compiler problems when
there are issues with the meta shaders.
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Add a function called ac_get_fs_input_vgpr_cnt which will return
the number of input VGPRs used by an AMD shader. Previously,
radv and radeonsi had the same code duplicated, but this commit also
allows them to share this code.
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
This commit allows RADV to set the shared VGPR count according to
the shader config.
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
This commit moves ac_get_tbuffer_format, ac_get_sampler_dim and
ac_get_image_dim into ac_shader_util, thus enabling them to be used
by compilers other than LLVM.
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
The aim of this commit is to keep ac_shader_util LLVM-free,
since we would like to use it in ACO later.
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
v2: rename pass_temp to pass_flags
v2: also CSE reductions
v3: add ds_swizzle_b32 support
v3: check gds/offset0/offset1 fields
Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
We want to generate PC files for non-glvnd builds and for builds with
old glvnd, but the current logic doesn't do that, it builds them
unconditionally, and for GLES it builds the shared libraries, which is
also not what we want. This does not generate .pc files for gles1 or
gles2. Which it we weren't doing before either, making this not a
regression but a return to status-quo.o
Closes: https://gitlab.freedesktop.org/mesa/mesa/issues/1838
Fixes: 93df862b6a
("meson: re-add incorrect pkg-config files with GLVND for backward compatibility")
Reviewed-by: Matt Turner <mattst88@gmail.com>
This allows the reslut of mov and bcsel to be separately interpreted as
float or int depending on the use.
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Some shaders are hurt by this change because now a
load_const(0x00000000) is not recognized as eq_zero when loaded as a
float. This behavior is restored in a later patch (nir/range-analysis:
Use types to provide better ranges from bcsel and mov).
v2: Add a comment about reinterpretation of int/uint/bool. Suggested by
Caio. Rewrite condition the check for types being float versus checking
for types not being all the things that aren't float.
Fixes: 405de7ccb6 ("nir/range-analysis: Rudimentary value range analysis pass")
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
All Gen7+ platforms had similar results. (Ice Lake shown)
total instructions in shared programs: 16327543 -> 16328255 (<.01%)
instructions in affected programs: 55928 -> 56640 (1.27%)
helped: 0
HURT: 208
HURT stats (abs) min: 1 max: 16 x̄: 3.42 x̃: 3
HURT stats (rel) min: 0.33% max: 6.74% x̄: 1.31% x̃: 1.12%
95% mean confidence interval for instructions value: 3.06 3.79
95% mean confidence interval for instructions %-change: 1.17% 1.46%
Instructions are HURT.
total cycles in shared programs: 363682759 -> 363683977 (<.01%)
cycles in affected programs: 325758 -> 326976 (0.37%)
helped: 44
HURT: 133
helped stats (abs) min: 1 max: 179 x̄: 33.61 x̃: 5
helped stats (rel) min: 0.06% max: 14.21% x̄: 2.47% x̃: 0.29%
HURT stats (abs) min: 1 max: 157 x̄: 20.28 x̃: 14
HURT stats (rel) min: 0.07% max: 14.44% x̄: 1.42% x̃: 0.73%
95% mean confidence interval for cycles value: 0.38 13.39
95% mean confidence interval for cycles %-change: -0.06% 0.96%
Inconclusive result (%-change mean confidence interval includes 0).
Sandy Bridge
total instructions in shared programs: 10787433 -> 10787443 (<.01%)
instructions in affected programs: 1842 -> 1852 (0.54%)
helped: 0
HURT: 10
HURT stats (abs) min: 1 max: 1 x̄: 1.00 x̃: 1
HURT stats (rel) min: 0.33% max: 1.85% x̄: 0.73% x̃: 0.49%
95% mean confidence interval for instructions value: 1.00 1.00
95% mean confidence interval for instructions %-change: 0.36% 1.10%
Instructions are HURT.
total cycles in shared programs: 153724543 -> 153724563 (<.01%)
cycles in affected programs: 8407 -> 8427 (0.24%)
helped: 1
HURT: 3
helped stats (abs) min: 18 max: 18 x̄: 18.00 x̃: 18
helped stats (rel) min: 0.98% max: 0.98% x̄: 0.98% x̃: 0.98%
HURT stats (abs) min: 4 max: 18 x̄: 12.67 x̃: 16
HURT stats (rel) min: 0.21% max: 0.75% x̄: 0.56% x̃: 0.72%
95% mean confidence interval for cycles value: -21.31 31.31
95% mean confidence interval for cycles %-change: -1.11% 1.46%
Inconclusive result (value mean confidence interval includes 0).
No shader-db changes on Iron Lake or GM45.
We're using vs and fs now, and adding hs, ds and gs soon. It's
confusing enough that we have both DS/TCS and HS/TES. At least for VS
and FS there doesn't have to be multiple names.
Signed-off-by: Kristian H. Kristensen <hoegsberg@google.com>
RET as a last instruction could be safely ignored.
Remove it to prevent crashes/warnings in case underlying driver
doesn't implement arbitrary returns.
A better way would be to remove the RET after the whole shader
is parsed which will handle a possible case when the last RET is
followed by a comment.
CC: <mesa-stable@lists.freedesktop.org>
Signed-off-by: Danylo Piliaiev <danylo.piliaiev@globallogic.com>
Reviewed-by: Axel Davy <davyaxel0@gmail.com>
--oneline shortens hashes, while --oneline=pretty doesn't, otherwise
they are the same. Having full hashes is convenient as that is the
format that the bin/.cherry-ignore script requires to work correctly.
Reviewed-by: Eric Engestrom <eric.engestrom@intel.com>
Reviewed-by: Juan A. Suarez <jasuarez@igalia.com>
The main reason to do this is that 19.2 has slipped by two weeks, and
such the 19.3 branch is due to happen extremely close to the release of
19.2.0. I think it would be better to have a little more time between
releases for developers and for packagers.
This would still have the 19.3 release out before December, even if it
slips by 1 week.
Acked-By: Karol Herbst <kherbst@redhat.com>
Acked-by: Juan A. Suarez <jasuarez@igalia.com>
This is a bit counter-intuitive, but the issue is that GLVND is broken
in versions <= 1.1.1, so we need to keep wrongly providing these files
to cover up their mistake, otherwise the rest of the world ends up
broken.
Suggested-by: Dylan Baker <dylan@pnwbakers.com>
Cc: mesa-stable@lists.freedesktop.org
Signed-off-by: Eric Engestrom <eric.engestrom@intel.com>
Reviewed-by: Dylan Baker <dylan@pnwbakers.com>
We currently lower them, but nir_opt_algebraic() can add new ones because
lower_sub=true.
Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
When handling two variables with overlapping locations, we process the
one with lower location first, and then extend the location ->
driver_location map to guarantee that it's contiguous for the second
variable too. But the loop had the wrong bound, so we weren't extending
the map 100%, which could lead to problems later such as an incorrect
num_inputs. The loop index i is an index into the slots of the variable,
so we need to stop at the final slot of the variable (var_size) instead
of the number of unassigned slots.
This fixes
spec@arb_enhanced_layouts@execution@component-layout@vs-fs-array-interleave-range
on radeonsi NIR.
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
both clang and gcc warn with:
"moving a local object in a return statement prevents copy elision"
Signed-off-by: Karol Herbst <kherbst@redhat.com>
Reviewed-by: Pierre Moreau <dev@pmoreau.org>
This moves the fix from commit 361f3d19f1 to happen in get_param
(used now instead of get_handle by st/dri). This fixes artifacts
seen with Xorg and CCS_E.
Fixes: fc12fd05f5 "iris: Implement pipe_screen::resource_get_param"
Signed-off-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Without this, we'll incorrectly round off huge values to the nearest
representable double instead of keeping it at the exact value as
we're supposed to.
Found by inspecting compiler-warnings.
Signed-off-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Fixes: 85faf5082f ("glsl: Add 64-bit integer support for constant expressions")
Reviewed-by: Eric Engestrom <eric.engestrom@intel.com>
Currently we add dependecies in 3 cases:
1) One node consumes value produced by another node
2) Sequency dependencies
3) Write after read dependencies
2) and 3) only affect scheduler decisions since we still can use pipeline
register if we have only 1 dependency of type 1).
Add 3 dependency types and mark dependencies as we add them.
Reviewed-by: Qiang Yu <yuq825@gmail.com>
Signed-off-by: Vasily Khoruzhick <anarsoul@gmail.com>
It makes no sense to clone texture coords if it's not varying, moreover
we don't support cloning ALU nodes.
Fixes: 1c1890fa70 ("lima/ppir: clone uniforms and load_coords into each successor")
Reviewed-by: Andreas Baierl <ichgeh@imkreisrum.de>
Signed-off-by: Vasily Khoruzhick <anarsoul@gmail.com>
because vl doesn't call flush_resource and I wasn't able to find
all places where flush_resource needs to be called.
This fixes corrupted / unflushed surfaces with fullscreen videos on Raven.
Cc: 19.1 19.2 <mesa-stable@lists.freedesktop.org>
This fixes some piglit tests on radeonsi NIR where a varying is
initialized to a constant array in the vertex shader. Varying packing
after nir_lower_io_to_temporaries creates writemasked stores which
persist after pulling the constant initialization down into the fragment
shader.
While we're here, rewrite handle_constant_store() to do the loop over
components outside the switch, so that we don't have to duplicate the
writemask checking for every bitsize.
Fixes: 1235850522 ("nir: Add a large constants optimization pass")
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Fixes a compilation error when building libnouveau:
In file included from ../src/gallium/drivers/nouveau/nv50/nv50_program.c:25:
../src/compiler/nir/nir.h:1115:10: fatal error: nir_intrinsics.h: No such file or directory
#include "nir_intrinsics.h"
^~~~~~~~~~~~~~~~~~
compilation terminated.
Fixes: f014ae3c7c ("nouveau: add support for nir")
Signed-off-by: Stephen Barber <smbarber@chromium.org>
Reviewed-by: Eric Engestrom <eric.engestrom@intel.com>
Reviewed-by: Karol Herbst <kherbst@redhat.com>
Released today and hangs on RADV. We don't have the root cause yet,
but this should unblock people playing the game.
No drirc because the radv debugflags are not usable from drirc and
I want this backported.
CC: <mesa-stable@lists.freedesktop.org>
Reviewed-by: Dave Airlie <airlied@redhat.com>
flrp was forgotten when already adding the rounding mode for other
instructions.
Fixes: ba1e25e1aa ("i965/fs: set rounding mode when emitting fadd, fmul and ffma instructions")
Suggested-by: Ian Romanick <ian.d.romanick@intel.com>
Signed-off-by: Andres Gomez <agomez@igalia.com>
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Reviewed-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
After
1711bf6cf2 ("intel/fs: Generate better code for fsign multiplied by a value"),
the conflicts resolution for setting the rounding mode after the
fused fmul and fsign optimization is non obvious.
Basically, the optimization doesn't really result in a MUL, or any
other operation which would need to have the rounding mode set. Hence,
we set it just before the actual MUL in the treatment of fmul.
Fixes: ba1e25e1aa ("i965/fs: set rounding mode when emitting fadd, fmul and ffma instructions")
Suggested-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Signed-off-by: Andres Gomez <agomez@igalia.com>
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Reviewed-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
The script only handles commits with "Fixes: <sha1>" where <sha1> is
equal or great than 8 chars. But <sha1> can be smaller, like 7 chars.
This commit relax the restriction to handle <sha1> 4 or more chars.
Fixes: 533fead423 ("bin/get-pick-list.sh: tweak the commit sha matching pattern")
Acked-by: Eric Engestrom <eric.engestrom@intel.com>
Signed-off-by: Juan A. Suarez Romero <jasuarez@igalia.com>
The scheduler doesn't expect them. To do this, I had to refactor the
registration part of gpir_node_create_dest() to be separate from
creating and inserting the node, since the last two now aren't done when
handling moves. This adds more code but creates the possibility of
automatically inserting input dependencies when inserting nodes, similar
to what's done in NIR with the use-def lists (this isn't done yet).
Reviewed-by: Vasily Khoruzhick <anarsoul@gmail.com>
We guarantee that a complex1 op is always used by postlog2 directly by
rewriting the postlog2 op to be a move when there would be a move
inserted between them. But we weren't doing this in all circumstances
where there might be a move. Move the logic to place_move() so that it
always happens. Fixes a few log tests that happened to start failing due
to changes in the register allocator leading to a different scheduling
order.
Reviewed-by: Vasily Khoruzhick <anarsoul@gmail.com>
This commit adds the framework for cross-basic-block register
allocation. Like ARM's compiler, we assume that the value registers
aren't usable across branches, which means we have to use physical
registers to store any value that crosses a basic block. There are three
parts to this:
1. When translating from NIR, we rely on the NIR out-of-ssa pass to
coalesce values into registers. We insert store_reg instructions for
values used in more than one basic block, and load_reg instructions for
values not defined in the same basic block (or defined after their use,
for loops). So by the time we've translated out of NIR we've already
split things into values (which are only used in the same basic block)
and registers (which are only used in different basic blocks than where
they're defined).
2. We allocate the registers at the same time that we allocate the
values, before the final scheduler. Unlike the values, where the
assigned color is fake, we assign the actual physical index & component
to physregs at this stage. load_reg and store_reg are treated as moves
in the allocator and when creating write-after-read dependencies.
3. Finally, in the main scheduler we have to avoid overwriting existing
live physregs when spilling. First, we have to tell the scheduler which
physical registers are live at the end of each block, to avoid
overwriting those. If a register is only live at the beginning, we can
reuse it for spilling after the last original use in the final program
happens, i.e. before any original use is scheduled, but we have to be
careful to add the proper dependencies so that the spill write is
scheduled before the original reads. To handle this we repurpose
reg_link for uses to be used by the scheduler.
A few register-related things copied over from NIR or from other
drivers can be dropped.
Reviewed-by: Vasily Khoruzhick <anarsoul@gmail.com>
Because branch conditions have to be in the pass slot, there is no
unconditional branch, and realistically the pass slot has to contain a
move when branching (there's nothing it does that would be useful for
operating on booleans, so we can't use it for anything when computing
the branch condition), we put the branch instruction in the pass slot
and at codegen time turn it into a move of the branch condition. This
means that it doesn't have to be special-cased like store instructions
are in the scheduler. Because of this decision we can remove the
half-implemented BRANCH codegen slot. Finally, we (ab)use the existing
schedule_first mechanism to make sure that branches are always last in
the basic block.
Reviewed-by: Vasily Khoruzhick <anarsoul@gmail.com>
When picking a node to be scheduled, we try to schedule its children as
well. But we shouldn't try to schedule nodes which only have a fake
dependency on the original node, since this isn't the point of
scheduling children at the same time and can break some expectations of
the rest of the code.
Reviewed-by: Vasily Khoruzhick <anarsoul@gmail.com>
Somewhat terrifyingly, we never sent this for direct contexts, which
means the server never knew the context/drawable bindings. To handle
this sanely, pull the request code up out of the indirect backend, and
rewrite the context switch path to call it as appropriate. This
attempts to preserve the existing behavior of not calling unbind() on
the context if its refcount would not drop to zero.
Of course, you can't just do this indiscriminately, because this is GLX
and extant X servers have bugs and everything is terrible. To wit:
- For 1.20.x prior to 1.20.6, you can bind a direct context once, but
the second time you try to modify the context's binding you will get
GLXBadContextTag. This includes unbinding the context. And "deleting"
the context will leak memory, because it will still appear to be
current.
- For 1.19 and earlier, glXMakeCurrent(dpy, None, ctx) should be legal
for GL 3.0+ contexts, but the server will throw BadMatch.
To guard against this, we only send the request for indirect contexts
unless the server is known good, and only mention one context at a time
in such a request; if switching between contexts, we first unbind the
old, and then bind the new. Note that the second VendorRelease() version
is to catch XFree86 4.x and Xorg [67].x, which almost certainly have the
above bugs. Other servers might report different version numbers here,
but we can't do direct rendering against them, so this should be safe.
Fixes glx-make-context, glx-multi-window-single-context and
glx-query-drawable-glx_fbconfig_id-window. Sufficiently old piglit will
regress on glx-make-glxdrawable-current (throwing BadMatch), which is
fixed by mesa/piglit!116.
From the MEDIA_VFE_STATE docs:
"Starting with this configuration, the Maximum Number of Threads must
be set to (#EU * 8) for GPGPU dispatches.
Although there are only 7 threads per EU in the configuration, the
FFTID is calculated as if there are 8 threads per EU, which in turn
requires a larger amount of Scratch Space to be allocated by the
driver."
It's pretty clear that we need to increase this for scratch address
calculations, because the FFTID has a certain bit-pattern. The quote
above seems to indicate that we should increase the actual thread count
programmed in MEDIA_VFE_STATE as well, but we think the intention is to
only bump the scratch space.
Fixes GPU hangs in Bioshock Infinite and Synmark's CSDof on Icelake 8x8.
Fixes: 5ac804bd9a ("intel: Add a preliminary device for Ice Lake")
Reviewed-by: Matt Turner <mattst88@gmail.com>
This reverts commit 729de1488f.
It turns out that, although the register is in the logical context,
it isn't whitelisted, so we can't actually write it from userspace
batch buffers. The write just becomes a noop, which is why we saw
no performance changes.
I manually whitelisted it, and still observed no performance gains, but
it did regress KHR-GL46.texture_cube_map_array.color_depth_attachments
on the iris driver. So we might need to fix something before enabling
this. To prevent it randomly getting turned on should the kernel ever
whitelist this register, we revert the patch for now.
'α' has never appeared in any genxml files, so there's no need to
replace it with the word "alpha".
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Use VPC_SO_OVERRIDE to control whether we do streamout in binning or
draw pass. Normally we want to do streamout in binning pass, except
when there is a single tile and binning passed is skipped.
Signed-off-by: Rob Clark <robdclark@chromium.org>
Reviewed-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
We could bit doing streamout from binning pass. In this case we want to
use the full VS which doesn't have (potentially streamed out) varyings
stripped out.
Signed-off-by: Rob Clark <robdclark@chromium.org>
Reviewed-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
In a3268599f3, I attempted to fix nir_repair_ssa for unreachable
blocks. However, that commit missed the possibility that the use is in
a block which, itself, is unreachable. In this case, we can end up in
an infinite loop trying to replace a def with itself. Even though a
no-op replacement is a fine operation, it keeps extending the end of the
uses list as we're walking it. Instead of explicitly checking for the
group of conditions, just check if the phi builder gives us a different
def. That's guaranteed to be 100% reliable and, while it lacks symmetry
with the is_valid checks, should be more reliable.
Fixes: a3268599 "nir/repair_ssa: Repair dominance for unreachable..."
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
pipe->clear() is not called for partial clears, which mesa emulates by
drawing a quad.
Furthermore, drivers should not use rasterizer state information for
scissor information (which was being used to handle the partial clears).
So, remove the partial clear support since it was not supposed to be
handled by pipe->clear() anyway.
This fixes issues with clearing after switching to different sized
framebuffers.
Signed-off-by: Erico Nunes <nunes.erico@gmail.com>
Reviewed-by: Vasily Khoruzhick <anarsoul@gmail.com>
Reviewed-by: Qiang Yu <yuq825@gmail.com>
->padded_count should be large enough to cover all vertices pointed by
the index array. Use the local vertex_count variable that contains the
updated vertex_count value for the indexed draw case.
Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
fixes "sorry, unimplemented: non-trivial designated initializers not supported"
Fixes: deb04adf2a ("clover: add support for passing kernels as nir to the driver")
Signed-off-by: Karol Herbst <kherbst@redhat.com>
Reviewed-by: Eric Engestrom <eric@engestrom.ch>
Allocating BOs is expensive, so we should avoid doing that by caching
freed BOs.
BO cache is modelled after one in v3d driver and works as follows:
- in lima_bo_create() check if we have matching BO in cache and return
it if there's one, allocate new BO otherwise.
- in lima_bo_unreference() (renamed from lima_bo_free()): put BO in
cache instead of freeing it and remove all stale BOs from cache
Reviewed-by: Qiang Yu <yuq825@gmail.com>
Signed-off-by: Vasily Khoruzhick <anarsoul@gmail.com>
os_time_get_absolute_timeout(0) returns current time, while kernel
driver expects 0 as value to poll BO status and return immediately.
Fix it by setting abs_timeout to 0 if timeout_ns is 0
Reviewed-by: Qiang Yu <yuq825@gmail.com>
Signed-off-by: Vasily Khoruzhick <anarsoul@gmail.com>
Some time weston set full damage region. It is
more effient to use the cached pp stream instead
of dynamically create one.
Reviewed-and-Tested-by: Vasily Khoruzhick <anarsoul@gmail.com>
Signed-off-by: Qiang Yu <yuq825@gmail.com>
This extension set a damage region for each
buffer swap which can be used to reduce buffer
reload cost by only feed damage region's tile
buffer address for PP.
Reviewed-and-Tested-by: Vasily Khoruzhick <anarsoul@gmail.com>
Signed-off-by: Qiang Yu <yuq825@gmail.com>
The PLBU expects the viewport's 4 borders' coordinates, however
currently we're feeding the coordinate of the left-bottom point and the
size to it, which leads to misrendering when the left-bottom point is
not (0,0).
Change the macros for the viewport PLBU command, and the data feed to
it. The code to calculate the 4 borders is ported from Panfrost.
Signed-off-by: Icenowy Zheng <icenowy@aosc.io>
Reviewed-by: Qiang Yu <yuq825@gmail.com>
ACO depends on C++14, but radeonsi/radv with LLVM 8,9 do not. Let us
only require it for RADV, since that is the only user.
Fixes: a70a998718 "radv/aco: Setup alternate path in RADV to support the experimental ACO compiler"
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
required for OpenCL
v2: adjust to changes in previous commits
v3: properly convert to NIR in nvc0_cp_state_create
Signed-off-by: Karol Herbst <kherbst@redhat.com>
Reviewed-by: Pierre Moreau <pierre.morrow@free.fr> (v1)
v2: minor formatting fixes
v3: call glsl_type_singleton_init_or_ref and glsl_type_singleton_decref
v4: capitalize and punctuate comments
fix text_executable -> text_intermediate in TODO
make glsl_type_singleton wrapper static
v5: rewrite how we run the nir passes
v6: fix unhandled case switch warning in st/mesa
Signed-off-by: Karol Herbst <kherbst@redhat.com>
Reviewed-by: Francisco Jerez <currojerez@riseup.net> (v4)
v2: rework arguments to compiler::compile_program
add assert to device::ir_format
v3: remove PIPE_SHADER_IR_SPIRV
change title
Signed-off-by: Karol Herbst <kherbst@redhat.com>
Reviewed-by: Francisco Jerez <currojerez@riseup.net> (v2)
Reviewed-by: Pierre Moreau <pierre.morrow@free.fr>
Most drivers have actually no binary format and just store the IR directly
as a single entry point blob.
v2: add a cap to switch between single or multi entry point binaries
v3: remove the entry_point field
v4: remove PIPE_CAP_MULTI_ENTRY_POINT_BINARIES
v5: remove supports_multiple_entry_points
Signed-off-by: Karol Herbst <kherbst@redhat.com>
Reviewed-by: Pierre Moreau <pierre.morrow@free.fr>
Changes since:
* v12:
- remove autotools (Karol Herbst)
- Remove the callback in format_validation_msg. (Francisco Jerez)
- Removed is_binary_spirv. (Francisco Jerez)
- Pass a string reference to is_valid_spirv instead of the
notification callback. (Francisco Jerez)
* v11: Fix compilation error introduced in v11.
* v10:
- Reuse format_validation_msg in is_valid_spirv.
- Remove LVL2STR macro in format_validation_msg.
* v9: Add `clover_cpp_std` to the overrides of the `libclspirv` target
in Meson.
* v7: Add DEFINES to libclspirv and libclover, in autotools, as they
would otherwise never know whether CLOVER_ALLOW_SPIRV has been
defined (Dave Airlie)
* v6: Update the dependency name (meson) and the libs variable
(Makefile) due to the replacement of llvm-spirv to the new
official SPIRV-LLVM-Translator.
* v5: Changed to match the updated “clover/llvm: Allow translating from
SPIR-V to LLVM IR” in the v6.
Reviewed-by: Karol Herbst <kherbst@redhat.com>
Reviewed-by: Francisco Jerez <currojerez@riseup.net>
Changes since:
* v12 (Karol Herbst):
- rename CLOVER_ALLOW_SPIRV to HAVE_CLOVER_SPIRV
* v11 (Karol Herbst):
- only set new defines for clover to speed up recompilation
- remove autotools
* v10:
- Add a new flag (`--enable-opencl-spirv` for autotools, and
`-Dopencl-spirv=true` for meson) for enabling SPIR-V support in
clover, and never automagically enable it without that flag. (Dylan Baker)
- When enabling the SPIR-V support, the SPIRV-Tools and
SPIRV-LLVM-Translator libraries are now required dependencies.
* v7:
- Properly align LLVMSPIRVLib comment (Dylan Baker)
- Only define CLOVER_ALLOW_SPIRV when **both** dependencies are found:
autotools was only requiring one or the other.
* v6: Replace the llvm-spirv repository by the new official
SPIRV-LLVM-Translator.
* v4: Add a comment saying where to find llvm-spirv (Karol Herbst).
* v3:
- make SPIRV-Tools and llvm-spirv optional (Francisco Jerez);
- bump requirement for llvm-spirv to version 0.2
* v2:
- Bump the required version of SPIRV-Tools to the latest release;
- Add a dependency on llvm-spirv.
Reviewed-by: Dylan Baker <dylan@pnwbakers.com> (v10)
Reviewed-by: Karol Herbst <kherbst@redhat.com>
Reviewed-by: Francisco Jerez <currojerez@riseup.net>
Gen11 doesn't require us to bypass the L2 cache for BC* images anymore.
The documentation is a bit hard to follow on this point, but the Windows
driver clearly only applies this workaround on Gen9, and their commit
history indicates that this was an intentional change to drop the
workaround for Gen11+.
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Currently there is no way to make no context current w/gallium + osmesa.
The non-gallium version of osmesa does this if the context and buffer
passed to `OSMesaMakeCurrent` are both null. This small change makes it
so that this is also the case with the gallium version.
Cc: mesa-stable@lists.freedesktop.org
Signed-off-by: Hal Gentz <zegentzy@protonmail.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
We can't really handle it in the little-core 64-bit case but it's not
really needed there. Where we really want this is for when we need to
do 16 -> 8-bit conversions.
Reviewed-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Because byte immediates aren't a thing on GEN hardware, we return a
signed or unsigned word immediate in the byte case.
Reviewed-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
During generate_shuffle(), when we use byte sized registers we end up
with a destination stride of 2. We don't take the stride into
consideration when selecting the group offset for the last MOV
operation, which means we end up moving things to the wrong place,
leaving the last few channels untouched. Take the destination stride
in consideration so we don't miss the last channels.
v2: Assert this is not necessary for the IVB special case (Jason).
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
The new order matches that of the comparison functions accepted by the C
standard library qsort() functions. Being consistent with qsort will
hopefully help avoid developer confusion.
The only current user of the red-black tree is aub_mem.c which is pretty
easy to fix up.
Reviewed-by: Lionel Landwerlin <lionel.g.lndwerlin@intel.com>
When I wrote the red-black tree implementation, I wrote tests for it but
they never got imported into mesa.
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
This effectively breaks the instance dispatch table in 2 with entry
points using a physical device as first argument getting their own
dispatch table.
As a result we now have to check instance & physical device dispatch
table instead of just the instance dispatch table before.
Signed-off-by: Eric Engestrom <eric.engestrom@intel.com>
Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
We were using the current drawable of the context to name the
appropriate screen for creating the bitmaps. But one, the current
drawable can be None, and two, it can be a GLXDrawable. Passing either
one as the second argument to XCreatePixmap will throw BadDrawable. Use
the root window of the context's screen instead.
Gitlab: https://gitlab.freedesktop.org/mesa/mesa/issues/89
LOLed-by: Kristian H. Kristensen <hoegsberg@google.com>
Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
atof() is locale-dependent (sigh), which means 1.3 becomes 1.0 if the
locale's decimal separator isn't a full-stop. Just use the protocol
major/minor instead. This would be slightly broken if the server
generically implements 1.3+ but a particular screen is only capable of
less, but in practice no such servers exist.
Gitlab: https://gitlab.freedesktop.org/mesa/mesa/issues/74
Reviewed-by: Eric Engestrom <eric.engestrom@intel.com>
Some shaders do not use 'invariant' in vertex and (possibly) geometry
shader stages on some outputs that are intended to be invariant. For
various reasons, this optimization may not be fully applied in all
shaders used for different rendering passes of the same geometry. This
can result in Z-fighting artifacts (at best). For now, disable this
optimization in these stages.
In tessellation stages applications seem to use 'precise' when
necessary, so allow the optimization in those stages.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=111490
Fixes: 09705747d7 ("nir/algebraic: Reassociate fadd into fmul in DPH-like pattern")
All Gen8+ platforms had similar results. (Ice Lake shown)
total instructions in shared programs: 16194726 -> 16344745 (0.93%)
instructions in affected programs: 2855172 -> 3005191 (5.25%)
helped: 6
HURT: 20279
helped stats (abs) min: 1 max: 3 x̄: 1.33 x̃: 1
helped stats (rel) min: 0.44% max: 1.00% x̄: 0.54% x̃: 0.44%
HURT stats (abs) min: 1 max: 32 x̄: 7.40 x̃: 7
HURT stats (rel) min: 0.14% max: 42.86% x̄: 8.58% x̃: 6.56%
95% mean confidence interval for instructions value: 7.34 7.45
95% mean confidence interval for instructions %-change: 8.48% 8.67%
Instructions are HURT.
total cycles in shared programs: 364471296 -> 365014683 (0.15%)
cycles in affected programs: 32421530 -> 32964917 (1.68%)
helped: 2925
HURT: 16144
helped stats (abs) min: 1 max: 403 x̄: 18.39 x̃: 5
helped stats (rel) min: <.01% max: 22.61% x̄: 1.97% x̃: 1.15%
HURT stats (abs) min: 1 max: 18471 x̄: 36.99 x̃: 15
HURT stats (rel) min: 0.02% max: 52.58% x̄: 5.60% x̃: 3.87%
95% mean confidence interval for cycles value: 21.58 35.41
95% mean confidence interval for cycles %-change: 4.36% 4.52%
Cycles are HURT.
There's nothing whatsoever compiler-specific about it other than that's
currently where it's used.
Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
There's a missing prev_ldst = NULL; assignment in the new logic,
but even with this fixed it seems to regress some applications,
so let's revert the change until we find the real problem.
This reverts commit c9bebae287.
BLORP always turns off TCS/TES/GS. If regular drawing also has them
disabled (the overwhelmingly common case), then leaving them disabled
is just fine by us and we can skip dirtying them, as that would just
re-disable them a second time on the next draw.
If they are actually enabled, however, we do need to flag them.
Cuts 52% of the 3DSTATE_HS packets in an Aztec Ruins trace.
Later generations support bindless for samplers, images, and buffers and
thus per-stage descriptors are not limited by the binding table size.
However, gen8 doesn't support bindless images and thus needs to report a
lower per-stage limit so that all combinations of descriptors that fit
within the advertised limits are reported as supported by
vkGetDescriptorSetLayoutSupport.
Fixes test dEQP-VK.api.maintenance3_check.descriptor_set
Fixes: 79fb0d27f3 ("anv: Implement SSBOs bindings with GPU addresses in the descriptor BO")
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Commit d1e1563bb6 added a NULL check for eglGetSyncAttribKHR
but eglGetSyncAttrib does not do this. Patch adds same check to
happen with eglGetSyncAttrib.
Fixes crashes in (when exposing EGL 1.5):
dEQP-EGL.functional.fence_sync.invalid.get_invalid_value
Signed-off-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Eric Engestrom <eric.engestrom@intel.com>
Cc: mesa-stable@lists.freedesktop.org
This improves a couple of things:
1. We now only update anything if the shader actually cares.
Previously, is_indexed_draw was causing us to flag dirty vertex
buffers, elements, and SGVs every time the shader switched between
indexed and non-indexed draws. This is a very common situation,
but we only need that information if the shader uses gl_BaseVertex.
We were also flagging things when switching between indirect/direct
draws as well, and now we only bother if it matters.
2. We upload new draw parameters only when necessary.
When we detect that the draw parameters have changed, we upload a
new copy, and use that. Previously we were uploading it every time
the vertex buffers were dirty (for possibly unrelated reasons) and
the shader needed that info. Tying these together also makes the
code a bit easier to follow.
In Civilization VI's benchmark, this code was flagging dirty state
many times per frame (49 average, 16 median, 614 maximum). Now it
occurs exactly once for the entire run.
This makes use of the total job size limiting feature added in the
previous patch.
The idea is to avoid an excessive build up in memory use due to the
use of both the UTIL_QUEUE_INIT_RESIZE_IF_FULL and
UTIL_QUEUE_INIT_USE_MINIMUM_PRIORITY flags.
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
When both UTIL_QUEUE_INIT_RESIZE_IF_FULL and
UTIL_QUEUE_INIT_USE_MINIMUM_PRIORITY are set, we can get into a
situation where the queue never executes and grows to a huge size
due to all other threads being busy.
This is the case with the shader cache when attempting to compile a
huge number of shaders up front. If all threads are busy compiling
shaders the cache queues memory use can climb into the many GBs
very fast.
The use of these two flags with the shader cache is intended to
allow shaders compiled at runtime to be compiled as fast as possible.
To avoid huge memory use but still allow the queue to perform
optimally in the run time compilation case, we now add the ability
to track memory consumed by the jobs in the queue and limit it to
a hardcoded 256MB which should be more than enough.
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Since we set the UTIL_QUEUE_INIT_USE_MINIMUM_PRIORITY flag this should
have little impact on low core systems. However just about all modern
CPUs currently available that run Mesa have *at least* 4 cores. For
these CPUs allowing more threads can result in the queue being
processed faster and avoid excessive memory use due to a backlog of
cache entrys building up in the queue.
This change helps avoid a huge build up of cache entrys in the queue
due to using both the UTIL_QUEUE_INIT_USE_MINIMUM_PRIORITY and
UTIL_QUEUE_INIT_RESIZE_IF_FULL flags.
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
The current code can create functions with a width of 32, which is not
supported by our hardware. Add some code to simplify how we express
what we want and prevent such cases.
For some unknown reason, all the tests I could run seem to work even
with these unsupported MOVs.
Fixes: b0858c1cc6 "intel/fs: Add a couple of simple helper opcodes"
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
There are cases where we try to generate registers with a stride of
32, while the hardware maximum is just 16. This happens, for example,
when using 8 bit integers on SIMD32. This results in a crash because
the variable 'width' has a value of 32:
../../src/intel/compiler/brw_reg.h:550: brw_reg brw_vecn_reg(unsigned
int, brw_reg_file, unsigned int, unsigned int): Assertion `!"Invalid
register width"' failed.
This change prevents the crash and makes the tests pass.
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
IMHO the code is easier to understand this way, being explicit that
we're doing exactly the same thing every time.
No functional changes.
v2: Adjust the loop breaking condition (Jason).
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
When dealing with uint16_t and uint8_t on SIMD32 we can do all the
operations using just 2 registers, so we don't hit the recursion at
the beginning of emit_scan(). Because of that, we need to actually
compute scan/reduce for channels 31:16.
v2: Still missed instructions (Jason).
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
We want this for tessellation eventually, but we can turn it on now.
Shader-db results:
total instructions in shared programs: 8612905 -> 8611387 (-0.02%)
instructions in affected programs: 164952 -> 163434 (-0.92%)
total dwords in shared programs: 11952000 -> 11950560 (-0.01%)
dwords in affected programs: 68096 -> 66656 (-2.11%)
total full in shared programs: 315019 -> 315009 (<.01%)
full in affected programs: 1642 -> 1632 (-0.61%)
total constlen in shared programs: 2463654 -> 2463654 (0.00%)
constlen in affected programs: 0 -> 0
total (ss) in shared programs: 152379 -> 152409 (0.02%)
(ss) in affected programs: 1503 -> 1533 (2.00%)
total (sy) in shared programs: 96473 -> 96525 (0.05%)
(sy) in affected programs: 654 -> 706 (7.95%)
total max_sun in shared programs: 1172454 -> 1172472 (<.01%)
max_sun in affected programs: 104 -> 122 (17.31%)
Signed-off-by: Kristian H. Kristensen <hoegsberg@google.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
Also, swap vs and fs constructor or so fs comes first.
Signed-off-by: Kristian H. Kristensen <hoegsberg@google.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
When using xfb and rasterizing, the fragment shader may have fewer
inputs than the vertex shader outputs. We can't rely on gl_Position to
be placed at fs->total_in, but have to instead remember where we add
it in the link map and use that location.
Fixes 100+ tesselation dEQPs under
dEQP-GLES31.functional.tessellation.primitive_discard.*
dEQP-GLES31.functional.tessellation.user_defined_io.*
Reviewed-by: Eric Anholt <eric@anholt.net>
New added cases "stole" the previous break.
Fixes: 420ad0a1a3 ("spirv: check support for SPV_KHR_float_controls capabilities")
Reviewed-by: Eric Engestrom <eric.engestrom@intel.com>
If we can entirely push uniform data, we don't need a SURFACE_STATE
descriptor for pulling data. Since constant uploads are a very common
operation, and being able to push all data is also very common, we would
like to avoid the overhead in this case.
This patch defers uploading new descriptors. Instead of handling that
at iris_set_constant_buffer, we do it at iris_update_compiled_shaders,
where we can see the currently bound shader variants. If any need pull
descriptors, and descriptors are missing, we update them and flag that
the binding table also needs to be refreshed.
Improves performance in GFXBench5 gl_driver2 on an i7-6770HQ by
31.9774% +/- 1.12947% (n=15).
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
I would like for iris to be able to avoid setting up SURFACE_STATE
for UBOs in the common case where all constants are pushed.
Unfortunately, we don't know up front whether everything will be
pushed: the backend is allowed to demote pushed UBOs to pull loads
fairly late in the process. This is probably desirable though, as
we'd like the backend to be able to re-pull pushed data to break up
long live ranges in response to register pressure.
Here we simply add a "are there any pull loads at all" boolean to
prog_data, which is a bit crude but at least allows us to skip work
in the common "everything pushed" case. We could skip more work by
tracking exactly which UBO surfaces are pulled in a bitmask, but I
wanted to avoid bringing back the old mark_surface_used() mechanism.
Finer-grained tracking could allow us to skip a bit more work when
multiple UBOs are in use and /some/ are 100% pushed, but others are
accessed via pulls. However, I'm not sure how common this is and
it would save at most 4 pull descriptors, so we defer that for now.
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
We now track per-stage bind history for constant and shader buffers,
shader images, and sampler views by adding an extra res->bind_stages
field to go with res->bind_history.
This lets us flag IRIS_DIRTY_CONSTANTS for only the specific stages
involved, and also skip some CPU overhead in iris_rebind_buffer.
Cuts 4% of 3DSTATE_CONSTANT_XS packets in a Shadow of Mordor trace
on Icelake.
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
The underlying buffer isn't changing - so we don't need to update any
SURFACE_STATE descriptors - we just might have new constants, meaning
we need to re-emit 3DSTATE_CONSTANT_XS. On Gen9, this means we need
to update 3DSTATE_BINDING_TABLE_POINTERS_XS too, but that's now handled
by the explicit check in the previous patch.
On Gen9, this should cause us to re-emit the binding table /pointer/ on
writing to a buffer with PIPE_BIND_CONSTANT_BUFFER, rather than emitting
a whole new /table/.
On Gen8 and Gen11, this avoids binding table churn altogether.
Cuts 61% of 3DSTATE_BINDING_TABLE_POINTERS_XS packets in a Shadow of
Mordor trace on Icelake.
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Right now, we usually flag both IRIS_DIRTY_{CONSTANTS,BINDINGS}_XS,
because we have SURFACE_STATE for constant buffers in case the shaders
access them via pull mode.
But this flagging is overkill in many cases. Gen8 and Gen11 don't need
it at all. Gen9 doesn't need that large of a hammer in all cases.
Just handle it explicitly so the right thing happens.
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
We upload a new SURFACE_STATE for the UBO/SSBO in question, which
means that we need new binding tables as well.
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Apparently we already enabled it without having support ...
Not sure if we also need to set disable_start_of_prim when the PS
has memory writes, but this mirrors radeonsi.
Doubles fillrate in my dual_quad_bench from ~16 pixels/cycles to
~32 pixels/cycle on a Raven.
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
The pass assumed that "Most ALU ops produce an undefined result if any
source is undef" which is completely untrue. Due to how we lower if
statements to selects and then optimize on those selects later, we
simply cannot make that assumption. In particular this pass tried to
replace an ior of undef and true, which had been generated by
optimizing a select which itself came from flattening an if statement,
to undef causing a miscompilation for a CTS test with radeonsi NIR.
We fix this by always doing what the non-undef path did, i.e. duplicate
the instruction twice. If there are cases where the instruction before
the loop can be folded away due to having an undef source, we should add
these to opt_undef instead.
The comment above the pass says that if the phi source from before the
loop is undef, and we can fold the instruction before the loop to undef,
then we can ignore sources of the original instruction that don't
dominate the block before the loop because we don't need them to create
the instruction before the loop. This is incorrect, because the
instruction at the bottom of the loop would get those sources from the
wrong loop iteration. The code never actually did what the comment said,
so we only have to update the comment to match what the pass actually
does. We also update the example to more closely match what most actual
loops look like after vtn and peephole_select.
There are no shader-db changes with i965, radeonsi NIR, or radv. With
anv and my vkpipeline-db there's only one change:
total instructions in shared programs: 14125290 -> 14125300 (<.01%)
instructions in affected programs: 2598 -> 2608 (0.38%)
helped: 0
HURT: 1
total cycles in shared programs: 2051473437 -> 2051473397 (<.01%)
cycles in affected programs: 36697 -> 36657 (-0.11%)
helped: 1
HURT: 0
Fixes
KHR-GL45.shader_subroutine.control_flow_and_returned_subroutine_values_used_as_subroutine_input
with radeonsi NIR.
Akin to 1a25980c46 ("egl: drop incorrect pkg-config file for
glvnd") and b01524fff0 ("meson: don't build libGLES*.so with
GLVND") , removes a pkg-config file that shouldn't have been there in
the first place, but was needed because of that GLVND bug.
Now that the glvnd bug has been fixed, it was apparent that this gl.pc
pkg-config file was forgotten to be removed, so let's do just that :)
Suggested-by: Matt Turner <mattst88@gmail.com>
Cc: mesa-stable@lists.freedesktop.org
Signed-off-by: Eric Engestrom <eric.engestrom@intel.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Having Python and C variables sharing name in the same block of code
makes its understanding a bit confusing. Make it explicit that the
Python bit_size variable refers to the destination bit size.
Suggested-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Signed-off-by: Andres Gomez <agomez@igalia.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
No GS copy shader if a pipeline enables NGG GS.
This fixes
dEQP-VK.pipeline.executable_properties.graphics.*geometry_stage*.
Fixes: 86864eedd2 ("radv: Implement radv_GetPipelineExecutablePropertiesKHR.")
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
* Set missing STENCIL_CONFIG_EXT2 bits
* Swap stencil sides when rendering CCW
Fixes following deqp tests (which were 99% failing):
dEQP-GLES2.functional.fragment_ops.depth_stencil.*
Note: deqp tests require --deqp-gl-config-name=rgba8888d24s8ms0
Signed-off-by: Jonathan Marek <jonathan@marek.ca>
Reviewed-by: Christian Gmeiner <christian.gmeiner@gmail.com>
This is needed in particular to get a recent enough version of meson in
the stretch image, but should be generally beneficial.
Acked-by: Eric Engestrom <eric.engestrom@intel.com>
Pros:
* Less fragile due to not mixing packages from stretch and buster
* No longer need to use third-party LLVM packages
* The buster image now uses GCC 8 for C++ as well (previously 6 for C++,
8 for C), allowing to drop some hacks
Con:
* The stretch image now only uses GCC 6 for C as well as C++
* Need separate jobs for testing old LLVM versions
Reviewed-by: Eric Engestrom <eric.engestrom@intel.com>
If installing new packages would require removing previously installed
ones, this flag causes apt-get to abort with an error instead,
preventing later obscure failures due to the missing packages.
Reviewed-by: Eric Engestrom <eric.engestrom@intel.com>
If we want to execute several batches in parallel they need to have
their own tiler and scratchpad BOs. Let move those objects to
panfrost_batch and allocate them on a per-batch basis.
Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
If we want the batch dependency tracking to work correctly we must
make sure all BOs are added to the batch->bos set early enough. Adding
FBO BOs when generating the fragment job is clearly to late. Add a
panfrost_batch_add_fbo_bos helper and call it in the clear/draw path.
Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
This helper automates the panfrost_bo_create()+panfrost_batch_add_bo()+
panfrost_bo_unreference() sequence that's done for all per-batch BOs.
Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Thanks to that we avoid the recursive call into panfrost_bo_create()
and we can get rid of panfrost_bo_release() by inlining the code in
panfrost_bo_unreference().
Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
panfrost_bo_unreference() should be used instead.
The only difference caused by this change is that the scratchpad,
tiler_heap and tiler_dummy BOs are now returned to the cache instead
of being freed when a context is destroyed. This is only a problem if
we care about context isolation, which apparently is not the case since
transient BOs are already returned to the per-FD cache (and all contexts
share the same address space anyway, so enforcing context isolation
is almost impossible).
Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Store a screen pointer in panfrost_bo so we don't have to pass a screen
object to all functions manipulating the BO.
Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Right now, the BO API is spread over pan_{allocate,resource,screen}.h.
Let's move all BO related definitions to a separate header file.
Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
pan_drm.c was only meaningful when we were supporting 2 kernel drivers
(mali_kbase, and the drm one). Now that there's now kernel-driver
abstraction we're better off moving those functions were they belong:
* BO related functions in pan_bo.c
* fence related functions + query_gpu_version() in pan_screen.c
* submit related functions in pan_job.c
While at it, we rename the functions according to the place they're
being moved to.
Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
ctx is allocated with rzalloc() which takes care of zero-ing the memory
region. No need to call memset(0) on top.
Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
That's what we do for other per-batch BOs, and we'll soon add an helper
to automate this create_bo()+add_bo()+bo_unreference() sequence, so
let's prepare the code to ease this transition.
Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Some BOs are used by batches but never explicitly added to the BO set.
This is currently not a problem because we wait for the execution of
a batch to be finished before releasing a BO, but we will soon relax
this rule.
Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
The DRM driver expects an array of u32, let's use the correct type, even
if using an int works in practice because it's still a 32-bit integer.
Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Looks like only HALT2 GPUs have support for it but that is not yet
implemented so disable ARB_shadow for now.
Signed-off-by: Christian Gmeiner <christian.gmeiner@gmail.com>
Reviewed-by: Jonathan Marek <jonathan@marek.ca>
Reviewed-by: Eric Anholt <eric@anholt.net>
drmIoctl handles EAGAIN itself and actually it always return -1 on errors.
Remove the wrong handling of its return value. Also, print a warning when
it fails.
v2: - use _debug_printf instead of fprintf (Gurchetan Singh)
Signed-off-by: Lepton Wu <lepton@chromium.org>
Reviewed-by: Eric Anholt <eric@anholt.net> (v1)
The compiler now sets the "Null Render Target" bit in the RT write
extended message descriptor, causing it to write to an implicit null
surface without us needing to set one up in the binding table.
Together with the last patch, this improves performance in Car Chase on
an Icelake 8x8 (locked to 700Mhz) by 0.0445526% +/- 0.0132736% (n=832).
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
When there are no color regions (i.e. a depth only pass), we can set
the "Null Render Target" bit in the Gen11 RT write extended message
descriptor to indicate that it should behave as if it's writing to a
null render target, without the need for a binding table entry.
This lets drivers avoid setting up that null RT binding table entry,
but more importantly means the HW doesn't actually have to bother
looking up the surface state.
Together with the next patch, this improves performance in Car Chase on
an Icelake 8x8 (locked to 700Mhz) by 0.0445526% +/- 0.0132736% (n=832).
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
This adds support for
VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_FLOAT_CONTROLS_PROPERTIES_KHR and
enables de Vulkan and SPIR-V extensions.
Also, notice that this includes the updates applied to the
VkPhysicalDeviceFloatControlsPropertiesKHR structure in the extension
VK_KHR_shader_float_controls v4 and Vulkan 1.1.116.
Signed-off-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
Signed-off-by: Andres Gomez <agomez@igalia.com>
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
The remove_extra_rounding_modes() optimization will remove duplicated
rounding mode changes.
v2:
- Fix bug in the rounding mode change (Alejandro).
v3:
- Fix rounding modes.
v4:
- Updated to renamed shader info member and enum values (Andres).
v5:
- Simplify flags logic operations (Caio).
Signed-off-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
Signed-off-by: Andres Gomez <agomez@igalia.com>
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
We need this function to emit code that setups the control register
later with the defined execution mode for the shader. Therefore, we
emit it as the first instruction.
v2:
- Fix bug in setting the default mode mask in brw_rnd_mode_from_nir().
- Fix support for rounding modes in brw_rnd_mode_from_nir().
v3:
- Updated to renamed shader info member and enum values (Andres).
v4:
- Add actual emission as first instruction of emit_nir_code (Caio).
Signed-off-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
Signed-off-by: Andres Gomez <agomez@igalia.com>
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Before this commit, we had only FPRoundingMode decoration (the per
instruction one) that is applied during the SPIR-V handling. In
vtn_alu we find out the rounding mode, and generate the code
accordingly that later will be used to look for the respective
nir_op_f2f16_{rtz,rtne}.
Per-instruction gets prioritized because we make them explicit
conversions (with RTZ or RTNE nir opcodes) and they will override the
default execution mode defined with float controls. However, we need
to come back to the mode defined by float controls after the execution
of the FP Rounding instruction.
Therefore, the new SHADER_OPCODE_FLOAT_CONTROL_MODE opcode will be
used to set the default rounding mode and denorms treatment in the
whole shader while the pre-existent SHADER_OPCODE_RND_MODE, will be
used as prioritized rounding mode in a per-instruction basis.
v2:
- Fix bug in defining BRW_CR0_FP_MODE_MASK.
v3:
- Update comment (Caio).
v4:
- Split the patch into the helper and the new opcode (this
one) (Caio).
v5:
- Add an explanation on the actual purpose and priority of the newly
introduced opcode in the commit log (Caio).
Signed-off-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
v2:
- Fix bug in defining BRW_CR0_FP_MODE_MASK.
v3:
- Update comment (Caio).
v4:
- Split the patch into the helper (this one) and the new
opcode (Caio).
Signed-off-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
The denorm mode is set in the control register, no need to do
something else.
v2:
- Add an assert to make sure that we realize if this assumption is
broken in the future (Caio).
Signed-off-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
Signed-off-by: Andres Gomez <agomez@igalia.com>
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
If we have fsin or fcos trigonometric operations with constant values
as inputs, we will multiply the result by 0.99997 in
brw_nir_apply_trig_workarounds, making the result wrong.
Adjusting the rules so they do not apply to const values we let a
later constant fold to deal with it.
v2:
- Do not early constant fold but only apply the trig workaround for
non constants (Caio).
- Add fixes tag to commit log (Caio).
Fixes: bfd17c76c1 "i965: Port INTEL_PRECISE_TRIG=1 to NIR."
Signed-off-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
Signed-off-by: Andres Gomez <agomez@igalia.com>
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Until now, it was using the floating point version of fmin/fmax,
instead of the double version.
Signed-off-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
v2:
- Replace hard coded value with DBL_MIN (Connor).
v3:
- Have into account the FLOAT_CONTROLS_DENORM_PRESERVE_FP64
flag (Caio).
Signed-off-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
Signed-off-by: Andres Gomez <agomez@igalia.com>
Reviewed-by: Connor Abbott <cwabbott0@gmail.com> [v2]
According to VK_KHR_shader_float_controls:
"Denormalized values obtained via unpacking an integer into a vector
of values with smaller bit width and interpreting those values as
floating-point numbers must: be flushed to zero, unless the entry
point is declared with the code:DenormPreserve execution mode."
v2:
- Add nir_op_unpack_half_2x16_flush_to_zero opcode (Connor).
v3:
- Adapt to use the new NIR lowering framework (Andres).
v4:
- Updated to renamed shader info member and enum values (Andres).
v5:
- Simplify flags logic operations (Caio).
Signed-off-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
Signed-off-by: Andres Gomez <agomez@igalia.com>
Reviewed-by: Connor Abbott <cwabbott0@gmail.com> [v2]
If FLOAT_CONTROLS_SIGNED_ZERO_INF_NAN_PRESERVE or
FLOAT_CONTROLS_DENORM_FLUSH_TO_ZERO are enabled, do not apply the
inexact optimizations so the VK_KHR_shader_float_controls execution
mode is respected.
v2:
- Do not apply inexact optimizations if SHADER_DENORM_FLUSH_TO_ZERO is
enabled (Andres).
v3:
- Updated to renamed shader info member (Andres).
v4:
- Directly access execution mode instead of dragging it by parameter (Caio).
Signed-off-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
Signed-off-by: Andres Gomez <agomez@igalia.com>
Reviewed-by: Connor Abbott <cwabbott0@gmail.com> [v1]
With the arrival of VK_KHR_shader_float_controls algebraic
optimizations for float types of the form (('fop', a, b), a) become
inexact depending on the execution mode.
For example, if we have activated SHADER_DENORM_FLUSH_TO_ZERO, in case
of a denorm value for the "a" parameter, we cannot return it still as
a denorm, it needs to be flushed to zero. Therefore, we mark now all
those operations as inexact.
Suggested-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
Signed-off-by: Andres Gomez <agomez@igalia.com>
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
v2:
- Move the op-code specific knowledge to nir_opcodes.py even if it
means a rount trip conversion (Connor).
Signed-off-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
Signed-off-by: Andres Gomez <agomez@igalia.com>
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
According to Vulkan spec, the new execution modes affect only
correctly rounded SPIR-V instructions, which includes fadd, fsub and
fmul.
v2:
- Fix fmul, fsub and fadd round-to-zero definitions, they should use
auxiliary functions to calculate the proper value because Mesa uses
round-to-nearest-even rounding mode by default (Connor).
v3:
- Do an actual fused multiply-add at ffma (Connor).
v4:
- Simplify fadd and fmul for bit sizes < 64 (Connor).
- Do not use double ffma for 32 bits float (Connor).
Signed-off-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
Signed-off-by: Andres Gomez <agomez@igalia.com>
Reviewed-by: Connor Abbott <cwabbott0@gmail.com> [v3]
f2f16's rounding modes are already handled and f2f64 don't need it
as there is not a floating point type with higher bit size than 64 for
now.
Signed-off-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
In order to be coherent with the pre-existent API for half floats,
this new API for double is the one meant to be used when doing double
to float conversions. It is no more than a wrapper for the softfloat.h
API but we meant to keep that one private.
v2:
- Fix bug in _mesa_double_to_float_rtz() in the inf/nan detection
using the exponent value.
v3:
- Replace custom f64 -> f32 implementations with the softfloat
one (Andres).
v4:
- Added API usage clarifying comments (Caio).
Signed-off-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
Signed-off-by: Andres Gomez <agomez@igalia.com>
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
In order to be coherent with the pre-existent functions, this new API
is the one meant to be used when doing half float to float
conversions. It is no more than a wrapper for the softfloat.h API but
we meant to keep that one private.
v2:
- Replace custom f32 -> f16 RTZ implementation with the softfloat
one (Andres).
v3:
- Added API usage clarifying comments (Caio).
Signed-off-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
Signed-off-by: Andres Gomez <agomez@igalia.com>
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Implemented fadd, fsub, fmul and ffma for doubles and ffma for floats,
rounding to zero, using a modified implementation from Berkely
Softfloat 3e Library.
Their implementation correctness has been checked with the Berkeley
TestFloat Release 3e tool for x86_64.
v2:
- Reuse util_last_bit64() in _mesa_count_leading_zeros64()
implementation (Connor).
v3:
- Add a specific ffma for floats version (Connor).
- Implement the ffma for doubles version (Andres).
- Lots of fixes in fadd, fsub and fmul (Andres).
- Improved documentation (Andres).
v4:
- Added f64 -> f32 conversion function (Andres).
- Added f32 -> f16 RTZ conversion function (Andres).
Signed-off-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
Signed-off-by: Andres Gomez <agomez@igalia.com>
Tested-by: Andres Gomez <agomez@igalia.com>
Acked-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
v2:
- Refactor conditions and shared function (Connor).
- Move code to nir_eval_const_opcode() (Connor).
- Don't flush to zero on fquantize2f16
From Vulkan spec, VK_KHR_shader_float_controls section:
"3) Do denorm and rounding mode controls apply to OpSpecConstantOp?
RESOLVED: Yes, except when the opcode is OpQuantizeToF16."
v3:
- Fix bit size (Connor).
- Fix execution mode on nir_loop_analize (Connor).
v4:
- Adapt after API changes to nir_eval_const_opcode (Andres).
v5:
- Simplify constant_denorm_flush_to_zero (Caio).
v6:
- Adapt after API changes and to use the new constant
constructors (Andres).
- Replace MAYBE_UNUSED with UNUSED as the first is going
away (Andres).
v7:
- Adapt to newly added calls (Andres).
- Simplified the auxiliary to flush denorms to zero (Caio).
- Updated to renamed supported capabilities member (Andres).
Signed-off-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
Signed-off-by: Andres Gomez <agomez@igalia.com>
Reviewed-by: Connor Abbott <cwabbott0@gmail.com> [v4]
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
v2:
- Added more functions.
v3:
- Simplify most of the functions (Caio).
v4:
- Updated to renamed enum values (Andres).
Signed-off-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
Signed-off-by: Andres Gomez <agomez@igalia.com>
Reviewed-by: Connor Abbott <cwabbott0@gmail.com> [v2]
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com> [v3]
v2:
- Add support for rounding modes for each floating point bit size.
v3:
- Commit e68871f6a4 ("spirv: Handle constants and types before
execution modes") changed when the execution modes are handled,
which affects the result of the floating point constants when the
rounding mode is set in the execution mode. Moved the handling of
the rounding modes before we handle the constants.
v4:
- Rename vtn_decoration "literals" to "operands" (Andres).
- Simplify execution mode parsing util function (Caio).
- Extend the comment about the timing of the handling of the rounding
modes (Caio).
v5:
- Correct extension name (Caio).
- Rename shader info member (Andres).
- Rename float controls enum (Andres).
Signed-off-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
Signed-off-by: Andres Gomez <agomez@igalia.com>
Reviewed-by: Connor Abbott <cwabbott0@gmail.com> [v3]
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Yes, some tests fail, but we can turn those into XFAILs at meson time.
Better to keep the things that work working than not cover them at all.
Unfortunately XPASS results will not cause the build to fail until we
update CI to meson 0.51 or newer.
Reviewed-by: Daniel Stone <daniels@collabora.com>
Since struct timespec's tv_sec member is of type time_t, adjust the
expected value to allow for the truncation which will occur with 32-bit
time_t.
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Adam Jackson <ajax@redhat.com>
Otherwise it never gets closed, this fixes errors seen with deqp-egl
where we end up opening 1024 files.
Fixes: 2dce0e94 ("iris: Initial commit of a new 'iris' driver for Intel Gen8+ GPUs.")
Signed-off-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
In case the GLSL version is 130 or higher, we've already enabled
ARB_shader_bit_encoding a bit earlier in this same function. So this
condition will always be true.
Signed-off-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
The PLBU seems to preserve scissor state between draws, and since lima doesn't
emit PLBU_CMD_SCISSORS() if scissor test is disabled, it uses state from previous draw.
Fix it by emitting PLBU_CMD_SCISSORS() for full fb if scissor test is disabled.
Signed-off-by: Icenowy Zheng <icenowy@aosc.io>
Reviewed-by: Vasily Khoruzhick <anarsoul@gmail.com>
Reviewed-by: Qiang Yu <yuq825@gmail.com>
SPIR-V 1.5 incorported the SPV_EXT_shader_viewport_index_layer but
splitting into the two capabilities above. Just handle them as we
support the extension already.
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
We can't deref list_(first/last)_entries unless we know we have at least
one. Instead, just use our IP we've been tracking as we go to set up the
start ip, and fill in the end IP as we walk instructions.
Fixes a complaint in valgrind on
dEQP-GLES3.functional.transform_feedback.* which sometimes has an
empty main (non-END) block when the VS inputs are just directly mapped
to outputs without any ALU ops.
Reviewed-by: Rob Clark <robdclark@chromium.org>
Table 23.54 of the OpenGL 4.5 spec lists the minimum values for
GL_POINT_SIZE_RANGE as [1, 1]. So zero is not allowed (even though
arguably this could be useful for MSAA rendering, where a sub-1px
point might cover only some samples...)
This fixes the WebGL 2.0 conformance suite's state.gl-get-calls test
on Chromium on Linux, which uses desktop OpenGL. The test checks that
the minimum value of GL_ALIASED_POINT_SIZE_RANGE is 1. Unfortunately,
that query doesn't exist in desktop GL, so it checks POINT_SIZE_RANGE,
which is the anti-aliased value. There's not really anything better
for Chromium to do here, unfortunately. When running Chromium with
--api=es3, it maps it to the correct query and the test already works.
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
Previously, internalformat GL_RGBA and type GL_UNSIGNED_SHORT_5_5_5_1
was promoted to RGBA8888 as the table entry with the 5551 formats
is listed below the 8888 entry, and it also doesn't have GL_RGBA as
a possible internalformat.
Using actual 5551 fixes the following dEQP-EGL test:
- dEQP-EGL.functional.image.modify.tex_rgb5_a1_tex_subimage_rgba8
Reviewed-by: Eric Anholt <eric@anholt.net>
Currently scons puts them in src/mapi/glapi, meosn puts them in
src/mapi/glapi/gen. This results in some things being compilable only by
one or the other, put them in the same places so that everyone is happy.
Reviewed-by: Eric Engestrom <eric.engestrom@intel.com>
It's useful for analyzing shader binaries produced by ARM mali offline
compiler which outputs files in MBS format. MBS is mali binary shader,
currently parser just extracts shader binary and ignores everything else.
Reviewed-and-tested-by: Connor Abbott<cwabbott0@gmail.com>
Signed-off-by: Vasily Khoruzhick <anarsoul@gmail.com>
Update GL headers and xml API from upstream Khronos registry (commit
3d0c3eb). Keep `BUILDING_MESA` quirk in glext.h.
mesa/extensions: Expose EXT_EGL_sync instead of MESA_EGL_sync to reflect
Khronos request of changing this extension's scope from MESA to EXT.
EGL_EGL_sync is also the name of the extension that has been merged into
the upstream Khronos GL registry.
Remove MESA_EGL_sync spec txt from Mesa tree as it is now published as
EXT by Khronos.
v1: Remove MESA_EGL_sync spec and squash commits (Eric E)
Reviewed-by: Eric Engestrom <eric.engestrom@intel.com>
Reviewed-by: Daniel Stone <daniels@collabora.com>
This might allow the arm64 tests to start running earlier.
Reviewed-by: Eric Engestrom <eric.engestrom@intel.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
v2:
* Preserve setting NIR_VALIDATE=0 for all arm64_* jobs
* Preserve setting DEQP_SKIPS=deqp-default-skips.txt for
arm64_a306_gles2 jobs
Reviewed-by: Eric Engestrom <eric.engestrom@intel.com> # v1
Reviewed-by: Eric Anholt <eric@anholt.net>
Support for multiple inheritance was added to GitLab recently.
Reviewed-by: Eric Engestrom <eric.engestrom@intel.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
This allows the arm64_a306_gles2 jobs to run as soon as the meson-arm64
job has finished.
Fixes: 6f0dc087b7 "freedreno: Introduce gitlab-based CI."
Reviewed-by: Eric Engestrom <eric.engestrom@intel.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
Commit f3e978db incorrectly assumed the maximum number of
samplers was equal to the max number of defined samplers
e.g. where bindings skip slots.
This fixes an assert in si_nir_load_sampler_desc() for an
enemy territory quake wars shader. And fixes potential bugs with
incorrect bounds limiting in the same code for production builds
of mesa.
Fixes: f3e978db ("radeonsi/nir: Remove uniform variable scanning")
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
It's still disabled by default because transform feedback randomly
hangs and it seems like it's related to GDS (cf. RadeonSI).
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Otherwise the next streamout operation will overwrite GDS. This
can be improved by tracking if there is a streamout operation in
flight. Currently the driver unconditionally flushes but that
doesn't matter much as NGG streamout is disabled by default.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
NGG streamout uses GDS and we have to make sure that another
process isn't going to overwrite GDS while our shaders are busy.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Otherwise the wave IDs are probably 0 and it hangs. NGG_WAVE_ID_EN
generates wave IDs for GDS OA.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
This internal option is turned off by default because NGG streamout
still hangs. It seems like it's related to GDS as RadeonSI.
That option will be turned on once all issues are resolved.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Documentation for pipe_context::flush states:
"NOTE: use screen->fence_reference() (or equivalent) to transfer
new fence ref to **fence, to ensure that previous fence is unref'd"
Hence we need to unref previous out_fence.
Signed-off-by: Danylo Piliaiev <danylo.piliaiev@globallogic.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
A filed of nir_variable.location may be equel to -1.
That may cause copying to invalid address of list-node,
making some internal fields corrupted.
Patch fixes segfault during freeing context due to
corrupted address of ralloc_header.destructor.
v2: copy data if var is constant (Connor Abbott)
CC: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Fixes: b6d4753568 (nir/large_constants: De-duplicate constants)
Signed-off-by: Sergii Romantsov <sergii.romantsov@globallogic.com>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=111676
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Vulkan applications can register with the following structure :
typedef struct VkApplicationInfo {
VkStructureType sType;
const void* pNext;
const char* pApplicationName;
uint32_t applicationVersion;
const char* pEngineName;
uint32_t engineVersion;
uint32_t apiVersion;
} VkApplicationInfo;
This enables the Vulkan implementations to apply workarounds based off
matching this description.
Here we add a new parameter for matching the driconfig options with
the following :
<device driver="anv">
<application engine_name_match="MyOwnEngine.*" engine_versions="10:12,40:42">
<option name="blaaah" value="true" />
</application>
</device>
v2: switch engine name match to use regexps
v3: Verify that the regexec returns REG_NOMATCH for match failure (Eric)
v4: Add missing bit that went to the following commit (Eric)
Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Eric Engestrom <eric.engestrom@intel.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Cc: 19.2 <mesa-stable@lists.freedesktop.org>
Seen a couple flakes on this one so far. Not sure if it is a real
driver problem or not, but skip it to unblock things.
Signed-off-by: Rob Clark <robdclark@chromium.org>
It was calloc'd to 0 which is PIPE_PRIM_POINTS, which means that we
fail to notice an initial primitive of points being new, and fail at
updating the "primitive is points or lines" field.
We do not need to reset this on device loss because we're tracking
the last primitive mode sent to us on the CPU via draw_vbo, not the
last primitive mode sent to the GPU.
Fixes several tests:
- dEQP-GLES3.functional.clipping.point.wide_point_clip
- dEQP-GLES3.functional.clipping.point.wide_point_clip_viewport_center
- dEQP-GLES3.functional.clipping.point.wide_point_clip_viewport_corner
Fixes: dcfca0af7c ("iris: Set XY Clipping correctly.")
If people fix bugs without updating the expected-fails list, then we
end up with a lack of coverage of those failures in the future. Also,
some day down the line another developer ends up trying to figure out
if the bug was actually fixed or their environment is just failing to
reproduce it.
Suggested-by: Kristian H. Kristensen <hoegsberg@google.com>
Reviewed-by: Adam Jackson <ajax@redhat.com>
Acked-by: Rob Clark <robdclark@chromium.org>
Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
This hasn't failed for me in ~5 minutes of looping over
dEQP-GLES3.functional.fbo.msaa.*
Reviewed-by: Adam Jackson <ajax@redhat.com>
Acked-by: Rob Clark <robdclark@chromium.org>
Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
Add a ppir dummy node for nir_ssa_undef_instr, create a reg for it and mark
it as undefined, so that regalloc can set it non-interfering to avoid
register pressure.
Signed-off-by: Andreas Baierl <ichgeh@imkreisrum.de>
Reviewed-by: Vasily Khozuzhick <anarsoul@gmail.com>
Reviewed-by: Erico Nunes <nunes.erico@gmail.com>
Building w/ AOSP, I was hitting the following error:
external/mesa3d/src/amd/Android.common.mk:95: error: missing separator.
Which was due to the changes to mesa-build-with-llvm missing
a line continuation.
Fixes: 96b592696f
Signed-off-by: John Stultz <john.stultz@linaro.org>
We are about to patch panfrost_flush() to flush all pending batches,
not only the current one. In order to do that, we need to move the
'flush single batch' code to panfrost_batch_submit().
While at it, we get rid of the existing pipelining logic, which is
currently unused and replace it by an unconditional wait at the end of
panfrost_batch_submit(). A new pipeline logic will be introduced later
on.
Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
panfrost_flush() is about to be reworked to flush all pending batches,
but we want the fence to block on the last one. Let's move the fence
creation logic in panfrost_flush() to prepare for this situation.
Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
panfrost_draw_vbo() Might call the primeconvert/without_prim_restart
helpers which will enter the ->draw_vbo() again. Let's delay
payloads[].offset_start initialization so we don't initialize them
twice.
Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
panfrost_attach_vt_xxx() functions are now passed a batch, and the
generated FB desc is kept in panfrost_batch so we can switch FBs
without forcing a flush. The postfix->framebuffer field is restored
on the next attach_vt_framebuffer() call if the batch already has an
FB desc.
Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
So we can emit SET_VALUE jobs for a batch that's not currently bound
to the context.
Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
We'll soon be able to flush a batch that's not currently bound to the
context, which means ctx->pipe_framebuffer will not necessarily be the
FBO targeted by the wallpaper draw. Let's prepare for this case and
use ctx->wallpaper_batch in panfrost_blit_wallpaper().
Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
We need that if we want to upload transient buffers to a batch that's
not currently bound to the context, which in turn will be needed if we
want to relax the batch serialization we have right now (only flush
batches when we need to: on a flush request, or when one batch depends
on the result of other batches).
Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Rename panfrost_is_scanout() into panfrost_batch_is_scanout(), pass it
a batch instead of a context and move the code to pan_job.c.
With this in place, we can now test if a batch is targeting a scanout
FB even if this batch is not bound to the context.
Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Will be replaced by something similar but using a BOs as keys instead
of resources.
Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
This way we have all the fb_state information directly attached to a
batch and can pass only the batch to functions emitting CMDs, which is
needed if we want to be able to queue CMDs to a batch that's not
currently bound to the context.
Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
mir_foreach_instr_in_block_safe() is based on list_for_each_entry_safe()
which is designed to protect against removal of the current entry, but
removing the entry placed just after the current one will lead to a
use-after-free situation.
Luckily, the midgard_pair_load_store() logic guarantees that the
instruction being removed (if any) is never placed just after ins which
in turn guarantees that the hidden __next variable always points to a
valid object.
Took me a bit of time to realize that this code was safe, so I'm
suggesting to get rid of the inner mir_foreach_instr_in_block_from()
loop and rework the code so that the removed instruction is always the
current one (which is what the list_for_each_entry_safe() API was
initially designed for).
While at it, we also get rid of the unecessary insert(ins)/remove(ins)
dance by simply moving the instruction around.
Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
list_for_each_entry() does not allow modifying the current item pointer.
Let's rework the skip-instructions logic in schedule_block() to not
break this rule.
Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
The V3D documentation states that primitive counters are reset when
we emit Tile Binning Mode Configuration items, which we do at the start
of each draw call, however, in the actual hardware this doesn't seem to
take effect when transform feedback is not active (this doesn't happen in
the simulator). This causes a problem in the following scenario:
glBeginTransformFeedback()
glDrawArrays()
glPauseTransformFeedback()
glDrawArrays()
glResumeTransformFeedback()
glEndTransformFeedback()
The TF pause will trigger a flush of the primitive counters, which results
in a correct number of primitives up to that point. In theory, the counter
should then be reset when we execute the draw after pausing TF, but that
doesn't happen, and since TF is enabled again by the resume command before
we end recording, by the time we end the transform feedback recording we
again check the counters, but instead of reading 0, we read again the same
value we read at the time we paused, incorrectly accumulating that value
again.
In theory, we should be able to avoid this by using the other method to
reset the primitive counters: using operation 1 instead of 0 when we
flush the counts to the buffer at the time we pause, but again, this
doesn't seem to be work and we still see obsolete counts by the time we
end transform feedback.
This patch fixes the problem by not accumulating TF primitive counts
unless we know we have actually queued draw calls during transform
feedback, since that seems to effectively reset the counters. This should
also be more performant, since it saves unnecessary stalls for the
primitive counters to be updated when we know there haven't been any
new primitives drawn.
Fixes CTS tests:
dEQP-GLES3.functional.transform_feedback.*
Reviewed-by: Eric Anholt <eric@anholt.net>
This was updating the counter for the indexed draw path only, but we are
already updating the counter for all paths a bit later, so this is only
duplicating counts for indexed paths.
Reviewed-by: Eric Anholt <eric@anholt.net>
Instead of running it with the Wayland platform, which introduces
unwanted dependencies and complexity.
Makes tests run 30% faster, as well.
Signed-off-by: Tomeu Vizoso <tomeu.vizoso@collabora.com>
streamout_buffers is assigned after that function, so the previous
fix was completely wrong. This probably fix something when streamout
buffers and push constants are used/inlined in the same shader.
Fixes: 378e2d2414 ("radv: fix computing number of user SGPRs for streamout buffers")
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
_mesa_texstore_z32f_x24s8 calculates source rowStride at a
pace of 64-bit, this will make inaccuracy offset if the width
of src image is an odd number. Modify src pointer to int_32* as
source image format is gl_float which is 32-bit per pixel.
Reviewed by Ilia Mirkin
Signed-off-by: Jiadong Zhu <Jiadong.Zhu@amd.com>
Signed-off-by: Marek Olšák <marek.olsak@amd.com>
The AnTuTu "garden" benchmark overflows the fixed size constbuffer
stateobject, so lets be more clever and calculate (a potentially
slightly pessimistic) actual size.
Signed-off-by: Rob Clark <robdclark@chromium.org>
fd6_blitter.c:724:31: warning: passing argument 1 of ‘fd_resource_level_linear’ discards ‘const’ qualifier from pointer target type [-Wdiscarded-qualifiers]
Signed-off-by: Rob Clark <robdclark@chromium.org>
Reviewed-by: Eric Anholt <eric@anholt.net>
Since freedreno's kernel and GPU reset seem to be totally solid, we
don't need to have the complexity of the LAVA setup that panfrost has.
Instead, we can register some boards as shared gitlab runners and have
the jobs run out of a docker container just like we do for llvmpipe.
Just make sure that the DRI device node is passed through to the
containers in the gitlab config ('devices = ["/dev/dri"]' under
runners.docker).
If a runner fails (networking dies, kernel panic, etc.) it'll take out
one build but the rest can keep going since gitlab-runner is what
pulls jobs. Since the runner pulls jobs, it also means that they can
live behind firewalls instead of needing some public address to be
accessed by gitlab.fd.o.
For now, enable it just on db410c (A307) and cheza (A630) as those are
the hardware that I have plenty of. A307 is only testing GLES2 since
running all of GLES3 takes too long for the number of boards I've
brought up.
Acked-by: Rob Clark <robdclark@chromium.org>
Acked-by: Kenneth Graunke <kenneth@whitecape.org>
Sometimes you just want confirmation that dEQP really picked up the
driver we built you thought. This is not as good as one might like,
because git isn't present in the cross-build image.
Acked-by: Rob Clark <robdclark@chromium.org>
Acked-by: Kenneth Graunke <kenneth@whitecape.org>
A handful of tests on freedreno have been close to the watchdog
timeout, and now sporadically fail since range analysis has slowed
down the compiler for them.
Acked-by: Rob Clark <robdclark@chromium.org>
Acked-by: Kenneth Graunke <kenneth@whitecape.org>
This brings back the fallback previously present in
st_nir_lookup_parameter_index(): if there's no parameter associated
with the variable, use a parameter from a variable with the same
prefix.
We'll have to sort out something for SPIR-V, but in the meantime let's
fix GLSL.
Fixes: b6384e57f5 ("mesa/st: Lookup parameters without using names")
Reviewed-by: Eric Anholt <eric@anholt.net>
Tested-by: Eric Anholt <eric@anholt.net>
As introduced in "v3d: flag dirty state when binding new sampler states"
we need to add support for compute states. New flag VC5_DIRTY_COMPTEX and
VC5_DIRTY_UNCOMPILED_CS are introduced.
Reaching 33 flags at the dirty field forces us to change the type to
uint_64. Flags are reordered and empty continuous bits are available
for future pipeline stages.
v2: Update flag conditions to compile cs shader. (Eric Antholt)
Now dirty flags use uint_64t and flags are reordered.
Added VC5_DIRTY_UNCOMPILED_CS flag.
Reviewed-by: Eric Anholt <eric@anholt.net>
Translating TGSI_INTERPOLATE_COLOR as INTERP_MODE_SMOOTH made
it for drivers impossible to have flatshaded color inputs.
Translate it to INTERP_MODE_NONE which drivers interpret as
smooth or flat depending on flatshading state.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=111467
Fixes: 770faf54 ("tgsi_to_nir: Improve interpolation modes.")
Signed-off-by: Danylo Piliaiev <danylo.piliaiev@globallogic.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
The patch adds support for HAL_PIXEL_FORMAT_RGBA_1010102 on
Android platform.
Fixes android.media.cts.DecoderTest#testVp9HdrStaticMetadata
which failed in egl due to "Unsupported native buffer format 0x2b"
on Android.
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
Signed-off-by: Chenglei Ren <chenglei.ren@intel.com>
We have only two defines that aren't from DRM_FORMAT_*: SARGB and
SABGR. Keep only those as __DRI_IMAGE_FOURCC and garbage collect the
rest.
While this header is also used from the X server, the X server doesn't
use any __DRI_IMAGE enums.
Reviewed-by: Eric Engestrom <eric.engestrom@intel.com>
Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
Taken from drm-misc-next 268de6530aa1 ("drm: mst: Fix query_payload
ack reply struct")
Reviewed-by: Eric Engestrom <eric.engestrom@intel.com>
Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
Previously, ReadPixels, PBO upload/download, and clears would call
cso_save_state with CSO_PAUSE_QUERIES, causing cso_context to call
pipe->set_active_query_state() twice for each operation. This can
potentially cause driver work to enable/disable statistics counters.
But often, there are no queries happening which need to be paused.
By keeping a simple tally of active queries, we can skip this work.
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Initial benchmarking didn't show any performance benefits. But it might eventually.
Signed-off-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
For example, the surfaceless platform only supports pbuffers. If the
driver supports MSAA, we would still create a config, but it would have
no supported surface types. That's meaningless, so don't do it.
Reviewed-by: Eric Engestrom <eric.engestrom@intel.com>
To go any further than this would be to break the current version of
Android.
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
[ Michel Dänzer: Dropped jessie line from debian-install.sh again ]
Upgrading to a newer g++ causes older LLVM/clang packages to be
removed.
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
Reviewed-by: Adam Jackson <ajax@redhat.com>
Something seems to have changed in Debian buster causing installation
of the other foreign packages to fail without this.
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
Reviewed-by: Adam Jackson <ajax@redhat.com>
IIRC, designated initializers are not legal C++.
Fixes the MSVC build.
Fixes: 83fd1e58 ("glsl/nir: Add and use a gl_nir_link() function")
Reviewed-by: Neha Bhende <bhenden@vmware.com>
This is unsupported by meson and may become a hard error in the future.
Fixes: 5adfc8602c
("lima/ppir: move sin/cos input scaling into NIR")
Reviewed-by: Vasily Khoruzhick <anarsoul@gmail.com>
Looks like .out_sync wasn't set in lima_submit_start(), as result
submit completion fence was never signalled.
Reviewed-by: Qiang Yu <yuq825@gmail.com>
Signed-off-by: Vasily Khoruzhick <anarsoul@gmail.com>
This is a pipe format, not a boolean.
Fixes: 5849e0612c ("gallium/auxiliary: Add util_format_get_depth_only() helper.")
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
Fixes dEQP-GLES3.functional.texture.specification.texstorage3d.size.3d_2x2x2_2_levels
Signed-off-by: Rob Clark <robdclark@chromium.org>
Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
If the lowest (largest) mipmap level is too small to tile, then don't
bother pretending.
Note that this requires initializing pipe->screen before
fd_resource_level_linear() is called.
Signed-off-by: Rob Clark <robdclark@chromium.org>
Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
This will also "unlock" OpenGL 4.6 for Iris!
v2: Also enable PIPE_CAP_GL_SPIRV_VARIABLE_POINTERS.
Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com> [v1]
Perform all the NIR linking steps in order. Change iris and i965 to
use it. Suggested by Alejandro.
v2: Add gl_nir_linker_options struct.
Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com> [v1]
The PIPE_CAP_GL_SPIRV capability enables ARB_gl_spirv and
ARB_spirv_extensions, and will make sure the corresponding SPIR-V
capabilities and extensions lists are initialized.
The additional PIPE_CAP_GL_SPIRV_VARIABLE_POINTERS capability enables
the support for Variable Pointers in SPIR-V shaders. This depends on
the driver and is not mandatory for ARB_gl_spirv support.
v2: Add a PIPE_CAP for Variable Pointers. (Marek)
Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com> [v1]
There's no such case, if we load prog->nir from the shader cache, we
shouldn't hit this path.
Suggested-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
The SPIR-V codepath uses NIR linking, so we have to preprocess after
the linking steps, which makes things slightly different than GLSL.
To make more clear when the preprocess is happening, I've ended up
inlining st_nir_get_mesa_program() into its caller.
The goal was to make both GLSL and SPIR-V to use the same preprocess
function, the exceptions are:
- SPIR-V codepath don't support NIR state slots yet;
- GLSL lowers shared memory early, so we don't do the deref lowering
for those.
For now I didn't bother to rename other functions and files (now that
many of them apply to both GLSL and SPIR-V), but we should do this in
further patches.
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
Refactor to split the glsl_to_nir conversion from the preprocessing
NIR passes into separate functions, so we can use them in SPIR-V.
Unlike in GLSL, there we'll need to perform a few passes with the NIR
linker before doing the individual preprocess calls.
No behavior should change with this patch.
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
Use the new MainUniformStorageIndex field in Parameter instead. It
was added so we could match those in the SPIR-V case, where names are
optional.
v2: Use MainUniformStorageIndex for all cases.
Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com> [v1]
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
Use the new UniformStorageIndex field in Parameter instead. This
mechanism was added so we could match those in the SPIR-V case, where
names are optional.
v2: Use UniformStorageIndex for all cases. (Timothy)
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
When creating Parameters, fill in the associated uniform storage
indices, like it is done with the NIR linker used for SPIR-V. This
will allow later code to not rely on names (which would never work for
SPIR-V where names are optional).
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
The parameter lists were not being created nor filled since i965
doesn't use them. In Gallium they are used for uniform handling, so
add a way to fill them.
The gl_uniform_storage struct got two new fields that let us go
- from a Parameter to the matching UniformStorage and,
- from the variable to the *first* UniformStorage
without relying on names -- since they are optional for ARB_gl_spirv.
Later patches will make use of them.
v2: Do not fill parameters for i965. (Timothy)
Use uint32_t for the new attributes. (Marek)
v3: Serialize the new fields. (Timothy)
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
The gl_register_file doesn't need 16 bits, so shorten it and use the
extra room for 'Padded' (also mark it as a single bit). This shrinks
the struct size from 32 bytes to 24 bytes.
See also 4794fbc86e ("mesa: reduce the size of gl_program_parameter")
that shrinked from 40 to 24 and later 7536af670b ("glsl: fix shader
cache for packed param list") that added `Padded`.
v2: Use just 5 bits for gl_register_file. (Timothy)
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
Every uniform that have the "gl_" name also have some state slots. So
use the state_slots like we did in 57b6184931 ("i965: account for NIR
uniforms without name").
This removes the dependency on names, which are optional when using
ARB_gl_spirv.
Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Don't use the UNMAPPED_UNIFORM_LOC (-1) to set the unsigned
max_uniform_location. Those unmapped uniforms don't have to be
accounted at this point.
Fixes: 7a9e5cdfbb ("nir/linker: Add gl_nir_link_uniforms()")
Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
This mirrors the haiku build which uses a platform.
v2: - Fix some rebase problems
Reviewed-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Eric Engestrom <eric.engestrom@intel.com>
v4: - Don't wrap a single file in a list to match mesa style
- Use null_dep instead of empty list
Reviewed-by: Eric Anholt <eric@anholt.net> (v3)
Reviewed-by: Eric Engestrom <eric.engestrom@intel.com>
v4: - Don't run checks on Windows that will always fail
Reviewed-by: Eric Anholt <eric@anholt.net> (v3)
Reviewed-by: Eric Engestrom <eric.engestrom@intel.com>
Which will allow meson to build a shared glapi build with mingw.
v2: - Add symbol to symbol check test
Reviewed-by: Eric Anholt <eric@anholt.net> (v1)
Reviewed-by: Eric Engestrom <eric.engestrom@intel.com>
It doesn't compile due to undefined symbols, which are in
libglapi_static, so I don't understand the problem.
Reviewed-by: Eric Engestrom <eric.engestrom@intel.com>
Currently the praser for s expressions assumes that newlines will be \n,
resulting in incorrect parsing on windows, where the newline is \r\n.
This patch just adds \r? to the regular expression used to parse the s
expressions, which fixes at 1 test on windows.
Reviewed-by: Eric Engestrom <eric.engestrom@intel.com>
Since the system value refactor, we've accidentally only been setting
cbuf->buffer_size in the UBO case, and not in the uploaded-constants
case. We use cbuf->buffer_size to fill out the SURFACE_STATE entry,
so it needs to be initialized in both cases.
Fixes: 3b6d787e40 ("iris: move sysvals to their own constant buffer")
This fixes some interactions when NGG GS is enabled. It fixes:
- dEQP-VK.clipping.user_defined.clip_cull_distance_dynamic_index.*geom*
- dEQP-VK.tessellation.geometry_interaction.passthrough.*
For some reasons, using the computed ESGS ring size randomly hangs
with CTS. For now, just use the maximum LDS size for ESGS.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
This shouldn't be in NIR->LLVM because ACO also needs the shader
info. This will also help for computing some NGG values that are
necessary for declaring LDS symbols.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
ac_surface computes it for amdgpu.
radeon_drm_surface computes it for radeon.
Acked-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
The VBO module maps a buffer with GL_MAP_FLUSH_EXPLICIT, and keeps
appending data, and calling glFlushMappedBufferRange(). We were
invalidating the VF cache each time it flushed a new range, which
results in a ton of VF flushes.
If the contents of the destination in the target range are undefined
(never even possibly written), this patch makes us assume that it's
likely not in the cache and so cache invalidations are required. If
the destination range is defined, we continue cache flushing as we may
need to expunge stale data.
This eliminates 88% of the VF cache invalidates on Manhattan 3.0.
Improves performance in Manhattan 3.0 on my Icelake 8x8 with the GPU
frequency locked to 700Mhz by 0.376724% +/- 0.0989183% (n=10).
This cuts roughly 85% of the 3DSTATE_SAMPLER_STATE_POINTERS_PS calls in
the J2DBench images test. For some reason, the state tracker is calling
bind_sampler_state with the same sampler state in a bunch of cases.
The line stipple pattern and factor only matter if line stippling is
actually enabled. Otherwise, we can safely ignore it.
PBO upload may give us zero for line stipple information, while normal
drawing tends to give us an actual stipple pattern such as 0xffff. This
was causing us to flag IRIS_DIRTY_LINE_STIPPLE way too often, leading to
useless 3DSTATE_LINE_STIPPLE commands, which are non-pipelined and thus
very expensive.
Improves performance in Manhattan 3.0 on Skylake GT4e by
0.149261% +/- 0.0380796% (n=210). On an Icelake 8x8 with the GPU
frequency locked at 700Mhz, improves by 0.423756% +/- 0.222843% (n=3).
The entire point of schedule_first is that the node has to be scheduled
as soon as possible without any moves because it doesn't produce a
proper floating-point value, or its value changes depending on where you
read it. We were still introducing a move for preexp2 in some cases
though, even if it got scheduled as soon as possible, which broke some
exp() tests. Fix that.
Reviewed-by: Vasily Khoruzhick <anarsoul@gmail.com>
Tested-by: Vasily Khoruzhick <anarsoul@gmail.com>
The whole point of schedule_first nodes is that they need to be
scheduled as soon as possible, so if a schedule_first node is the
successor in a fake dependency that prevents it from being scheduled
after its parent, that can cause problems. We need to add these fake
dependencies to the parent as well, and we need to guarantee that the
pre-RA scheduler puts schedule_first nodes right before their parents in
order to prevent this from adding cycles to the dependency graph.
Reviewed-by: Vasily Khoruzhick <anarsoul@gmail.com>
Tested-by: Vasily Khoruzhick <anarsoul@gmail.com>
The idea was to make sure schedule_first nodes were always first in the
ready list. I made sure they were inserted first, but not that other
nodes wouldn't later be scheduled ahead of them. Fixes
spec@glsl-1.10@execution@built-in-functions@vs-exp-float and probably
others.
Reviewed-by: Vasily Khoruzhick <anarsoul@gmail.com>
Tested-by: Vasily Khoruzhick <anarsoul@gmail.com>
The point of the function is to avoid creating a complex move which is
used by certain slots in the next instruction, but unscheduled
successors will never be in the next instruction. Found while debugging
a crash that the previous commit fixed.
Reviewed-by: Vasily Khoruzhick <anarsoul@gmail.com>
Tested-by: Vasily Khoruzhick <anarsoul@gmail.com>
The scheduler assumes that load nodes are always duplicated so that they
can always be scheduled eventually and therefore they never need to be
spilled. But some lowerings were running after the pre-RA scheduler,
whereas duplication has to happen before then since it's needed for the
scheduler to do a better job reducing register pressure. This meant
that lowerings were introducing multiple uses of a load instruction,
which broke the scheduler's expectation and resulted in infinite loops
in situations where the only nodes available to spill were load nodes.
Spilling load nodes would be silly, so we want to fix the lowerings
rather than the scheduler. Just do all lowerings before the pre-RA
scheduler, which also helps with reducing pressure since the scheduler
can more accurately compute the pressure.
Fixeslima/mesa#104.
Reviewed-by: Vasily Khoruzhick <anarsoul@gmail.com>
Tested-by: Vasily Khoruzhick <anarsoul@gmail.com>
Change needed to fix the following building error:
In file included from external/mesa/src/intel/vulkan/anv_device.c:43:
external/mesa/src/util/xmlpool.h:115:10: fatal error: 'xmlpool/options.h' file not found
^~~~~~~~~~~~~~~~~~~
1 error generated.
Fixes: 4dcb1ff ("anv: add support for driconf")
Signed-off-by: Mauro Rossi <issor.oruam@gmail.com>
Reviewed-by: Eric Engestrom <eric@engestrom.ch>
So we can move all the BO logic into this file instead of having it
spread over pan_resource.c, pan_drm.c and pan_bo_cache.c.
Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
The last users have been converted to use plain BOs. Let's get rid of
this abstraction. We can always consider adding it back if we need it
at some point.
Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Some fields in panfrost_context are unused (probably leftovers from
previous refactor). Let's get rid of them.
Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
ctx->{scratchpad,tiler_heap,tiler_dummy} are allocated using
panfrost_drm_allocate_slab() but they never any of the SLAB-based
allocation logic. Let's convert those fields to plain BOs.
Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Right now, the transient memory allocator implements its own BO caching
mechanism, which is not really needed since we already have a generic
BO cache. Let's simplify things a bit.
Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
What we currently call a job is actually a batch containing several jobs
all attached to a rendering operation targeting a specific FBO.
Let's rename structs, functions, variables and fields to reflect this
fact.
Suggested-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
This commit follow OES_EGL_sync to universially enable use of EGL sync
objects with desktop OpenGL contexts.
Reviewed-by: Daniel Stone <daniels@collabora.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
The dead_cf pass calls into the CF manipulation helpers which attempt to
keep NIR's SSA form sane. However, when the only break is removed from
a loop, dominance gets messed up anyway because the CF SSA clean-up code
only looks at phis and doesn't consider the case of code becoming
unreachable. One solution to this would be to put the loop into LCSSA
form before we modify any of its contents. Another (and the approach
taken by this pass) is to just run the repair_ssa pass afterwards
because the CF manipulation helpers are smart enough to keep all the
use/def stuff sane; they just don't always preserve dominance
properties.
While we're here, we clean up some bogus indentation.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=111405
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=111069
Cc: mesa-stable@lists.freedesktop.org
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
NIR currently assumes that unreachable blocks are trivially dominated by
everything. However, when considering well-formed SSA, there is no path
from any block to an unreachable block. Therefore, we can break any
use-def chains where the use is in an unreachable block. This removes
any dependencies on code created by uses in unreachable blocks and lets
DCE do a better job of cleaning it up.
Cc: mesa-stable@lists.freedesktop.org
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
We already bail and don't split the vars but we were passing a NULL to
_mesa_hash_table_search which is not allowed.
Fixes: f1cb3348f1 "nir/split_vars: Properly bail in the presence of ..."
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
In the case where the stencil clear is nicely aligned, we can clear
stencil much more efficiently by mapping it as a wide format (say
RGBA32_UINT) and blasting out the stencil clear value with a repclear.
On Unigine Heaven, this makes one stencil clear go from non-trivial to
unnoticeable when looking at per-draw timings.
In order for this change to work properly, ANV needs to do a bit more
flushing around depth and stencil clears. i965 and iris already have
the cache tracking logic to handle this so no changes are required
there.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
This isn't known to fix any current bugs but it does prevent a
regression in a subsequent commit.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Fixes: 02bc4aabb48 ('nir/lower_io_to_vector: allow FS outputs to be vectorized')
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
u_endian.h needs to be included, otherwise PIPE_ARCH_BIG_ENDIAN might not
be defined on big-endian architectures and the endian conversion macros
will be incorrect.
I don't think anything is broken because of this, I just noticed this when
looking at the file.
Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
This bit redirects the state cache from the unified/RO sections of the
L3 cache to the "CS command buffer" section of the cache, which would
be set up via TCCNTLREG. The documentation says:
"Additionaly, this redirection should be enabled only if there is a
non-zero allocation for the CS command buffer section."
We don't allocate any cache to the CS command buffer section, so
enabling this redirection effectively disabled the state cache.
The Windows driver only sets up that section when using POSH, which
we do not currently use. So, leave it unallocated and disable the
redirection to get a functional state cache again.
Improves performance in Civilization VI by 18%, Manhattan 3.0 by 6%,
and Car Chase by 2%.
Jason pointed out that the caches likely refer to offsets from dynamic
and surface state base addresses, so when we change those, we need to
invalidate the caches.
Comment borrowed from src/intel/vulkan/genX_cmd_buffer.c.
The driver can't determine PIPE_QUERY_PRIMITIVES_GENERATED or
PIPE_QUERY_PRIMITIVES_EMITTED once we support geometry or
tessellation, since these stages add primitives at runtime. Use the
WRITE_PRIMITIVE_COUNTS event to write back the primitive counts and
implement a hw query for this.
Reviewed-by: Rob Clark <robdclark@gmail.com>
The GPU writes out streamout offsets as it goes to the FLUSH_BASE
pointer. We use that value with CP_MEM_TO_REG when appending to the
stream so that we don't have to track the offsets with the CPU in the
driver. This ensures that streamout continues to work once we enable
geometry and tessellation shader stages that add geometry.
Reviewed-by: Rob Clark <robdclark@gmail.com>
Should fix some issues we're seeing. And use REALLOC instead of realloc.
Reviewed-by: Dave Airlie <airlied@redhat.com>
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
This has lower_io_to_vector try to turn variables into arrays of 4-sized
vectors when possible and fall back to the old approach when that isn't
possible.
This is so that lower_io_to_vector can guarantee that only one variable is
used for each fragment shader output.
v2: handle dual-source blending
v3: don't try to merge structs and non-32-bit types in get_flat_type()
v3: fix per-vertex inputs
v3: fix and cleanup location advancement in get_flat_type() and it's
calling code
v4: prioritize the original mode over the flat mode
v4: don't create flat variables to merge only one variable
v5: don't skip an entire slot when encountering structs in the old mode
Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
It shouldn't matter much because output varyings should have been
compacted during NIR shader linking but it mirrors what the driver
does when emitting NGG GS vertex parameters.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
If the fragment shader needs the layer index, we have to allocate
one more dword in the NGG GS storage. Found by inspection. This
doesn't fix anything known.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Sometimes LAVA jobs will timeout due to transient issues, and the Gitlab
job will fail in that case. Increase the timeouts to reduce the
likeliness of that happening and reduce false positives.
Signed-off-by: Tomeu Vizoso <tomeu.vizoso@collabora.com>
So repositories don't need to be specially configured with a token to
access LAVA, store this token in a bind volume for a special runner.
Signed-off-by: Tomeu Vizoso <tomeu.vizoso@collabora.com>
For loops which condition is false on the first iteration
iteration count was falsely calculated under the assumption
that loop's condition is true until it becomes false, meaning
it's true at least one time.
Now such loops are reported as having 0 iteration.
Similar to the fix e71fc7f2 done in NIR.
Fixes tests/shaders/glsl-fs-loop-while-false-02.shader_test
Signed-off-by: Danylo Piliaiev <danylo.piliaiev@globallogic.com>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
This commit removes the GLSL dependency in TTN by manually recording
the textures used and calling nir_lower_samplers
instead of its GL counterpart.
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Lowering samplers is needed to produce NIR that can actually be
consumed by some gallium drivers, so it doesn't make sense to
to keep it only in the GLSL code.
This commit introduces nir_lower_samplers to compiler/nir,
while maintains the GL-specific function too.
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
This fixes a memory leak in the flush code:
Direct leak of 128 byte(s) in 1 object(s) allocated from:
#0 in __interceptor_realloc .../gcc-8.3.0/libsanitizer/asan/asan_malloc_linux.cc:105
#1 in si_buffer_do_flush_region src/gallium/drivers/radeonsi/si_buffer.c:573
#2 in si_buffer_flush_region src/gallium/drivers/radeonsi/si_buffer.c:608
#3 in si_buffer_flush_region src/gallium/drivers/radeonsi/si_buffer.c:597
Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
This patch partially reverts 20294dc ("mesa: Enable asm unconditionally, ...")
Android makefile build logic needs to disable assembler optimization
in 32bit builds to avoid text relocations for libglapi.so shared
Fixes the following build error with Android x86 32bit target:
[ 0% 4/477] target SharedLib: libglapi (out/target/product/x86/obj/SHARED_LIBRARIES/libglapi_intermediates/LINKED/libglapi.so)
FAILED: out/target/product/x86/obj/SHARED_LIBRARIES/libglapi_intermediates/LINKED/libglapi.so
...
prebuilts/gcc/linux-x86/x86/x86_64-linux-android-4.9/x86_64-linux-android/bin/ld: warning: shared library text segment is not shareable
prebuilts/gcc/linux-x86/x86/x86_64-linux-android-4.9/x86_64-linux-android/bin/ld: error: treating warnings as errors
clang-6.0: error: linker command failed with exit code 1 (use -v to see invocation)
Fixes: 20294dc ("mesa: Enable asm unconditionally, now that gen_matypes is gone.")
Signed-off-by: Mauro Rossi <issor.oruam@gmail.com>
Acked-by: Eric Engestrom <eric@engestrom.ch>
The codegen handles it and it adds the correct casts. This fixes
a bunch of LLVM validation errors when enabling Wave32 for compute.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Load a 32-bit value then convert to 1-bit. Convert 1-bit to 32-bit
value, then Store it.
These cases started to appear when we changed Anvil to use derefs for
shared memory.
v2: Use `bit_size` in a couple of places we were missing. (Jason)
Reassign `value` instead of `src[0]`. (Jason)
Fixes: 024a46a407 ("anv: use derefs for shared memory access")
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
This reverts commit c0504569ea. Now that
we're doing interpolation lowering in NIR, we can continue to stride the
FS input registers directly in the brw_fs_nir code like we did before.
This fixes SIMD32 fragment shaders which broke because lower_simd_width
depended on the 0 stride to split PLN instructions correctly.
Reviewed-by: Francisco Jerez <currojerez@riseup.net>
This commit does two things. First, it simplifies the way we compute
the FB write group bit. There's no reason to use a ternary because
inst->group / 16 can only be 0 or 1. Second, it fixes an order-of-
operations bug where the ternary wasn't selecting between (1 << 11) and
0 but between (1 << 11) and 0 | brw_dp_write_desc(...).
Fixes: 0d9648416 "intel/compiler: Use generic SEND for Gen7+ FB writes"
Reviewed-by: Francisco Jerez <currojerez@riseup.net>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Utgard PP is vec4 architecture, so lowering phis to scalars
increases instruction count and potentially interferes with
spilling.
Tested-by: Andreas Baierl <ichgeh@imkreisrum.de>
Reviewed-by: Eric Anholt <eric@anholt.net>
Signed-off-by: Vasily Khoruzhick <anarsoul@gmail.com>
For render formats, update fd2_pipe2color to only work with HW supported
render formats, and remove the format whitelist is_format_supported. This
patch enables float render formats (which work).
For vertex/texture formats, use a generic function which translates using
the bitsize of the channels. Since we fake support for some vertex formats,
check for these in is_format_supported to avoid enabling them as sampler
formats.
Signed-off-by: Jonathan Marek <jonathan@marek.ca>
Reviewed-by: Rob Clark <robdclark@chromium.org>
Use fd_gmem_restore_format() to avoid trying to use unsupported Z24S8/Z16
render formats for gmem restore.
Also apply this change to gmem2mem so it doesn't depend on fd2_pipe2color
working with depth formats.
gmem2mem/mem2gmem also doesn't need to use the swap/swizzle, since dst/src
formats are the same.
Signed-off-by: Jonathan Marek <jonathan@marek.ca>
Reviewed-by: Rob Clark <robdclark@chromium.org>
Fixes failures in the following deqp tests:
dEQP-GLES2.functional.polygon_offset.*
Signed-off-by: Jonathan Marek <jonathan@marek.ca>
Reviewed-by: Rob Clark <robdclark@chromium.org>
Reviewed-by: Eric Anholt <eric@anholt.net>
Fixes failures in the following deqp tests:
dEQP-GLES2.functional.fragment_ops.*src_alpha_saturate*
Signed-off-by: Jonathan Marek <jonathan@marek.ca>
Reviewed-by: Rob Clark <robdclark@chromium.org>
Reviewed-by: Eric Anholt <eric@anholt.net>
Fixes the following deqp test:
dEQP-GLES2.functional.shaders.builtin_variable.pointcoord
Signed-off-by: Jonathan Marek <jonathan@marek.ca>
Reviewed-by: Rob Clark <robdclark@chromium.org>
Reviewed-by: Eric Anholt <eric@anholt.net>
Some instructions generated by int/bool float lowering need to be lowered
by opt_algebraic.
Fixes: 43dbd7d6
Signed-off-by: Jonathan Marek <jonathan@marek.ca>
Reviewed-by: Rob Clark <robdclark@chromium.org>
Reviewed-by: Eric Anholt <eric@anholt.net>
Utgard PP has vector fcsel operation, but its condition is scalar. Add
filtering callback that checks whether {b,f}csel condition is not scalar
to lower {b,f}csel to scalar only in this case.
Reviewed-by: Qiang Yu <yuq825@gmail.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
Signed-off-by: Vasily Khoruzhick <anarsoul@gmail.com>
Set of opcodes doesn't have enough flexibility in certain cases. E.g.
Utgard PP has vector conditional select operation, but condition is always
scalar. Lowering all the vector selects to scalar increases instruction
number, so we need a way to filter only those ops that can't be handled
in hardware.
Reviewed-by: Qiang Yu <yuq825@gmail.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Signed-off-by: Vasily Khoruzhick <anarsoul@gmail.com>
This appears to work fine (with the additional constraint of keeping the
indirect load in the same block that a0.x was loaded).
We can probably lift this restriction on earlier gens after testing.
Signed-off-by: Rob Clark <robdclark@chromium.org>
Reviewed-by: Eric Anholt <eric@anholt.net>
Need to use ir3_instr_set_address(), otherwise the instruction might not
get added to the indirects table. This becomes a problem when we turn
on copy propagation for relative accesses, as check_instr() in the sched
pass won't realize there is an indirect consumer of address register
load that is ready to be scheduled.
Signed-off-by: Rob Clark <robdclark@chromium.org>
Reviewed-by: Eric Anholt <eric@anholt.net>
An instruction can reference only a single address register value.
Add an assert to catch bugs.
Also, address value should also be local to the same block as the
instruction.
(The one spot where changing the instruction address is actually legit
needs to clear the address first.)
Signed-off-by: Rob Clark <robdclark@chromium.org>
Reviewed-by: Eric Anholt <eric@anholt.net>
After the next patch enabling copy propagation for relative sources,
we'll need to dereference the n'th src in valid_flags(), so we actually
need to swap the sources before calling valid_flags().
But the logic was already a bit cumbersome, so move it into a helper
function.
Signed-off-by: Rob Clark <robdclark@chromium.org>
Reviewed-by: Eric Anholt <eric@anholt.net>
The live_values and use_count was not being properly updated. This
starts triggering problems with the next patch, where we allow copy
propagation for RELATIV access.
Signed-off-by: Rob Clark <robdclark@chromium.org>
Reviewed-by: Eric Anholt <eric@anholt.net>
Move the constant part of the indirect offset into nir intrinsic base.
When we have multiple indirect accesses with different constant offsets,
this lets other opt passes clean up things to use a single address
register value.
Signed-off-by: Rob Clark <robdclark@chromium.org>
Reviewed-by: Eric Anholt <eric@anholt.net>
Now that spilling ops can be inserted into existing instructions, it
makes sense to increase cost to spill registers that would cause the
creation of a new instruction.
Experimental results showed that penalizing too much due to this caused
worse results, however it is beneficial as a tie resolver between
registers with the same number of components.
Signed-off-by: Erico Nunes <nunes.erico@gmail.com>
Reviewed-by: Vasily Khoruzhick <anarsoul@gmail.com>
Avoid creating unnecessary instructions for the load/store temp nodes
when not required, to further reduce register pressure.
The store_temp operation seems to be unable to do any spilling.
At least the offline shader seems to never output instructions accessing
swizzled components, and attempting to output that in ppir results in
errors. So, force spilled registers to allocate a full vec4 register.
This seems to be the optimal way as it is possible to always keep stores
and temps in a single instruction that can be pipelined.
Signed-off-by: Erico Nunes <nunes.erico@gmail.com>
Reviewed-by: Vasily Khoruzhick <anarsoul@gmail.com>
One ssa created in the spillinc code in ppir_update_spilled_src was not
properly being marked 'spilled', which made it a candidate for future
spilling attempts.
Since it was being inserted by the spilling code itself, let's mark it
unspillable to avoid an infinite spilling loop.
Signed-off-by: Erico Nunes <nunes.erico@gmail.com>
Reviewed-by: Vasily Khoruzhick <anarsoul@gmail.com>
Shaders must not attempt to write to the register files in the last
three instructions, but that doesn't include the magic registers:
nop ; nop ; thrsw; ldtmu.- *** ERROR ***
nop ; nop
nop ; nop
v2: Simplify validation rules. (Eric Anholt)
v3: Adjust validation even more. (Eric Anholt)
Reviewed-by: Eric Anholt <eric@anholt.net>
For radeonsi, we will prefer the NIR pass as it'll generate better code
(some index calculation and a single load vs. a load, then index
calculation, then another load) and oftentimes NIR optimization can kick
in and make all the access indices constant.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
This prevents regressions when disabling indirect lowering. Sometimes
the only use of an input array was copying it to the array created by
nir_lower_io_to_temporaries, and without lowering indirects we wouldn't
have eliminated the temporary array until after linking, which was too
late to remove unused code in the producer.
No shader-db changes with radeonsi NIR.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
Setup a constant global variable that LLVM will stick in a .rodata
section and generate PC-relative loads for.
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
We usually use these counts as a simple way to figure out if a change
reduces the number of instructions or shrinks an instruction. However,
since .rodata sections aren't executed, we shouldn't be counting their
size for this analysis. Make the linker return the total executable
size, and use it to report the more useful size in both drivers.
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Removing GL_FRAMEBUFFER_FLIP_Y_MESA token from glheader.h as it is now
provided by glext.h
Reviewed-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Sync extension spec of MESA_framebuffer_flip_y to what has been merged
upstream in the GL registry. Update now carries the accepted GL
extension no.
v2: split GL headers update off to separate commit
Reviewed-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Integrating headers from upstream registry [0] master branch. Effective
GL registry commit integrated:
9d534f9312e56c72df763207e449c6719576fd54
Keeping the following quirks local to Mesa:
- glext.h: BUILDING_MESA guard (see !1492)
- glxext.h: glXQueryGLXPbufferSGIX: 'int' return type (Mesa) vs while
'void' (GL registry)
- glxext.h: GLX_RENDERER_ID_MESA is still expected by some mesa tests,
even though its token has been removed from the spec (see
docs/specs/MESA_query_renderer.spec)
- glxext.h: glXGetTransparentIndexSUN / PFNGLXGETTRANSPARENTINDEXSUNPROC
argument pTransparentIndex has type 'unsigned long *' (Mesa) vs. 'long
*' (GL registry)
[0] https://github.com/KhronosGroup/OpenGL-Registry
Reviewed-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Given that we occasionally touch this code and probably nobody really
wants to think about it, introduce a minimal test so that we know we
haven't completely broken OSMesa.
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
When run in optirun, applications that linked to `libGLX.so` and then
proceeded to querying Mesa for extension strings caused a SEGV in Mesa.
`glXQueryExtensionsString` was calling a chain of functions that
eventually led to `__glXQueryServerString`. This function would call
`xcb_glx_query_server_string` then `xcb_glx_query_server_string_reply`.
The latter for some unknown reason returned `NULL`. Passing this `NULL`
to `xcb_glx_query_server_string_string_length` would cause a SEGV as the
function tried to dereference it.
The reason behind the function returning `NULL` is yet to be determined,
however, simply checking that the ptr is not `NULL` resolves this. A
similar check has been added to `__glXGetString` for completeness sake,
although not immediately necessary.
In addition to that, we stumbled into a similar problem in
`AllocAndFetchScreenConfigs` which tries to access the configs to free
them if `__glXQueryServerString` fails. This, of course, SEGVs, because the
configs are yet to have been allocated. Simply continuing past the configs
if their config ptrs are `NULL` resolves this. We also switch to `calloc`
to make sure that the config ptrs are `NULL` by default, and not some
uninitialized value.
Cc: mesa-stable@lists.freedesktop.org
Fixes: 24b8a8cfe8 "glx: implement __glXGetString, hide __glXGetStringFromServer"
Fixes: cb3610e37c "Import the GLX client side library, formerly from xc/lib/GL/glx. Build it "
Reviewed-by: Adam Jackson <ajax@redhat.com>
Signed-off-by: Hal Gentz <zegentzy@protonmail.com>
The precision for a function return type is now stored in
ir_function_signature. This will later be useful to implement mediump
to float16 lowering. In the meantime it is also useful to catch errors
where a function is redeclared with a different precision.
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
This adds the dispatch code. It creates a job for the number
of blocks in the grid, and dispatches them to the threadpool
implementation. The threadpool then calls the JIT code to
execute the coroutines.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
This creates the coroutine execution environment and the
main compute shaders that get executed inside it.
Each compute shader block is executed in it's own coroutine
execution shader, which each "thread" being a coroutine executed
inside it in sequence.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
This doesn't actually build any of the shaders yet, but just
builds up the framework necessary to start building the shaders
and variants.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
The compute shader will need it's own context like the frag shader
has, this just introduces the framework struct and allocates/frees
for it in the right places.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
When the code is executing an hits a barrier, it will suspend
the coroutine and return control to the coroutine dispatcher.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
In order to efficiently run a number of compute blocks, use
a threadpool that just allows for jobs with unique sequential
ids to be dispatched.
In order to share the texture/image/sampler code with compute
shaders we need to reorg them to be at the front of context
same as draw does for vs/gs sharing.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
These wrap the coroutine intrinsics and also add some higher
level wrappers around coroutine begin, end and suspend procedures
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
We were previously not doing at least some of the checks. This uses the
same logic that is used in glTexImage*.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Dave Airlie <airlied@redhat.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Each BSD has slightly different sysctl for retrieving per-CPU times.
FreeBSD returns long while NetBSD returns uint64_t. On OpenBSD return
type differs between summation and per-CPU times. DragonFly is
compatible with FreeBSD.
Signed-off-by: Jan Beich <jbeich@FreeBSD.org>
Based on the vc4 implementation.
Fixes Android RenderEngine::flush() routine:
android.googlesource.com/platform/frameworks/native/+/refs/tags/android-o-mr1-iot-release-smart-clock-fcs/services/surfaceflinger/RenderEngine/RenderEngine.cpp#225
Signed-off-by: Roman Stratiienko <roman.stratiienko@globallogic.com>
Reviewed-by: Qiang Yu <yuq825@gmail.com>
Try more aggressive approach with cloning uniform and coord loads.
Uniform load can be inserted into any instruction, so let's do that. ARM site
claim that penalty for cache miss is one clock, so we don't lose anything if
we merge it into instruction that uses the result. As side effect we can also
pipeline it and thus decrease reg pressure.
Do the same for varyings that hold texture coords, but for different reason:
looks like there's a special path for coords that increases precision if
varying that holds it is pipelined. If we don't pipeline it and load coords
from a register its precision is fp16 and thus only 10 bits which is not enough
to accurately sample textures of size 1024 or larger.
Since instruction can hold only one uniform load and one varying load,
node_to_instr now creates a move using helper introduced in previous commit if
slot is already taken. As side effect of this change we can also try to
pipeline texture loads and create a move if attempt fails.
Reviewed-by: Erico Nunes <nunes.erico@gmail.com>
Signed-off-by: Vasily Khoruzhick <anarsoul@gmail.com>
It can load value from varying directly as well. Also load_regs is the
only op that has a source, so add src_num field to load node and set it
accordingly.
Reviewed-by: Erico Nunes <nunes.erico@gmail.com>
Signed-off-by: Vasily Khoruzhick <anarsoul@gmail.com>
When lowering from ubo, use the constant base field in the load_uniform
instruction for the constant part of the offset. Doesn't change much
for constant indexing, but this will help for indirect indexing because
constant-folding can't completely clean up the result.
Signed-off-by: Rob Clark <robdclark@chromium.org>
Reviewed-by: Eric Anholt <eric@anholt.net>
This looks like clear copy-and-pasteos, and fixes:
dEQP-GLES2.functional.draw.random.40
(on A307 and A630, both tested in the new CI farm)
Reviewed-by: Rob Clark <robdclark@chromium.org>
We can get all the information we need from NIR. It's slightly less
accurate, but radeonsi doesn't use the extra information. The old code
also overcounted atomic counters, which led to problems when everything
was used at once.
Fixes KHR-GL45.compute_shader.resources-max.
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Otherwise it's impossible to know the maximum SSBO index for both
internal TGSI shaders from TTN (which don't have any notion of atomic
counters and no offset) as well as shaders from GLSL.
I fixed everything I could find while grepping for num_ssbos and
num_abos, which hopefully is everything (iris was the only user I could
find that uses it in a meaningful way).
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
This adds a bit of unneccesary code on radeonsi, since whether
unnormalized coordinates are used is known at compile time with GL, but
I wasn't sure if it was worth the few instructions to plumb everything
through, especially for something so rare -- my shader-db doesn't have
any instances where this changes anything.
Fixes CTS tests I created at
https://github.com/cwabbott0/VK-GL-CTS/tree/unnorm-gather-tests
Acked-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
The workaround was originally written based on amdgpu-pro traces, but
since then radeonsi has got its own slightly different version. Use the
radeonsi version instead, to be consistent and because it'll be slightly
more convenient for handling unnormalized coordinates.
Acked-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Technically, the user might have set EGL_DISPLAY instead of
EGL_PLATFORM, but since the former is deprecated let's just mention the
latter in the warning message.
Signed-off-by: Eric Engestrom <eric.engestrom@intel.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
This routine was made obsolete over a series of reworks of memory
allocation; Tomeu's changes to shader memory allocation finally made
this unused as cppcheck noted.
Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Reviewed-by: Tomeu Vizoso <tomeu.vizoso@collabora.com>
swr_shader.cpp: In function ‘void (* swr_compile_gs(swr_context*, swr_jit_gs_key&))(HANDLE, HANDLE, SWR_GS_CONTEXT*)’:
swr_shader.cpp:732:44: error: ‘make_unique’ was not declared in this scope
ctx->gs->map.insert(std::make_pair(key, make_unique<VariantGS>(builder.gallivm, func)));
^~~~~~~~~~~
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Jan Zielinski <jan.zielinski@intel.com>
While the documentation for _BitScanReverse64 on MSDN says that it's
available on ARM, this isn't true. It's only available on ARM64. So
let's match reality.
Signed-off-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Acked-by: Matt Turner <mattst88@gmail.com>
This code generates CVTSD2SI, which requires SSE2. So let's fix the
required SSE-version.
Signed-off-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Fixes: 5de29ae (util: try to use SSE instructions with MSVC and 32-bit gcc)
Reviewed-by: Matt Turner <mattst88@gmail.com>
This has been unused since 183db3a645 ("glsl: move half<->float
convertion to util"), Oct 10 2015. Let's drop needlessly including it.
Signed-off-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Lionel found actual documentation for this at long last. Apparently
it actually is a sampler cache limitation that was mostly fixed on
Icelake. Unfortunately, it seems there are still issues with ASTC
and non-ASTC sampler views. Still, we can lessen the flush condition
from "format mismatch" to "ASTC mismatch", which eliminates most of
the flushing here.
We also update the documentation to refer to the workaround name.
strchrnul is not available on macOS.
pipe_loader.c:141:14: error: implicit declaration of function 'strchrnul' is invalid in C99 [-Werror,-Wimplicit-function-declaration]
next = strchrnul(library_paths, ':');
^
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Acked-by: Eric Engestrom <eric.engestrom@intel.com>
The majority of these only apply the start argument to the input, but a
few of them also does for the output-array. util_primconvert, the only
user of this argument expects this pass a non-zero start-argument does
not expect this to be applied to the output; if it is, it will write
outside of allocated memory, leading to VRAM corruption.
The reason this doesn't seem to have been noticed before, is that no
driver currently use util_primconvert to convert a primitive-type to
itself, which is the cases where this was broken. But for Zink, this
will no longer be true, because we need to eliminate the use of 8-bit
index-buffers.
Signed-off-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Fixes: 28f3f8d413 ("gallium/auxiliary/indices: add start param")
Reviewed-by: Rob Clark <robdclark@chromium.org>
Commit 6f7306c029 ("swr/rast: Refactor memory API between rasterizer
core and swr") unintentionally removed changes for llvm-9.0.
Fixes: 6f7306c029 ("swr/rast: Refactor memory API between rasterizer core and swr")
Fixes: 5dd9ad1570 ("swr/rasterizer: Better implementation of scatter")
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Jan Zielinski <jan.zielinski@intel.com>
We already had a perfectly cromulent pass for this, but one landed in
common NIR code so let's switch and lighten our tree.
Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
This optimization depended on RA running before scheduling. It therefore
no longer applies and is now unused.
Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
This is a tradeoff.
Scheduling before RA means we don't do RA on what-will-become pipeline
registers. Importantly, it means the scheduler is able to reorder
instructions, as registers have not been decided yet.
Unfortunately, it also complicates register spilling, since the spills
themselves won't get bundled optimally and we can only spill twice per
ALU bundle (only one spill per bundle allowed here). It also prevents us
from eliminating dead moves introduced by register allocation, as they
are not dead before RA. The shader-db regressions are from poor spilling
choices introduced by the new bundling requirements. These could be
solved by the combination of a post-scheduler (to combine adjacent
spills into bundles) with a VLIW-aware spill cost calculation.
Nevertheless, the change is small enough that I feel it's worth it to
eat a tiny shader-db regression for the sake of flexibility.
Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Rather than using a pile of hacks and awkward constructs in MIR to
ensure the writeout parameter gets written into r0, let's add a
dedicated shadow register class for writeout (interfering with work
register r0) so we can express the writeout condition succintly and
directly.
Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
There's no slot for it; you'll end up writing into the void and
clobbering stuff. Don't. do it.
Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
When running the register allocator after scheduling, the MIR looks a
little different, so we need to extend the RA to handle a few of these
extra cases correctly.
Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
After scheduling, we still have valid MIR, but we have additional
bundling annotations which we would like to keep debug, so print these.
Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Rather than a vague "br.??" line, annotate the branch with its target
type (useful for disambiguating discards) and whether it was inverted.
Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
I'm not sure if this is strictly necessary but it makes debugging easier
and minimizes the diff with the experimental scheduler.
Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Scheduling occurs on a per-block basis, strongly assuming that a given
block contains at most a single branch. This does not always map to the
source NIR control flow, particularly when discard intrinsics are
involved. The solution is to allow scheduling barriers, which will
terminate a block early in code generation and open a new block.
To facilitate this, we need to move some post-block processing to a new
pass, rather than relying hackily on the current_block pointer.
This allows us to cleanup some logic analyzing branches in other parts
of the driver us well, now that the MIR is much more well-formed.
Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
It's sometimes convenient to call this with no instruction specified. By
definition, a missing instruction cannot reference any argument, so
let's check for NULL and shortciruit to false.
Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
The branch has the writeout specified in its source list, making this
special even if it's not explicitly part of r0.
Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
In order to run register allocation after scheduling, it is sometimes
necessary to be able to insert instructions into an already-scheduled
program. This is suboptimal, since it forces us to do a worst-case
scheduling, but it is nevertheless required for correct handling of
spills/fills. Let's add helpers to insert instructions as standalone
bundles for use in spilling code.
These helpers are minimal -- they *only* work on load/store ops or
moves. They should not be used for anything but register spilling; any
other instructions should be added prior to the schedule.
Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
While it doesn't matter with an unconditional move to the conditional
register (r31), when we try to elide that move we'll need to track the
swizzle explicitly, and there is no slot for that yet since ALU ops are
normally binary.
Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Oh boy. Midgard scheduling is crazy... These are all just the
requirements, not even the algorithm yet.
Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
This does not affect shaders in any way. Rather, it makes the shader-db
instruction count recorded in the compiler accurate with the in-order
scheduler, matching up with what we calculate from pandecode.
Though shaders are the same, instruction counts cannot be compared
across this commit for this reason.
Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Allow a direct link to the PDF itself from the authors themselves,
rather than a paywall splash page.
Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Acked-by: Rob Clark <robdclark@chromium.org>
The GLX extension strings are independent of any context, so abusing the
direct_support bit to control this extension's visibility is wrong.
This reverts commit 079d0717fc896bc8086b037d0ed22642274986c7.
Reported-by: Michel Dänzer <michel@daenzer.net>
Reviewed-by: Michel Dänzer <michel@daenzer.net>
Memory allocated through panfrost_allocate_transient() is likely to
come from the transient pool. Let's add the BO backing the allocated
memory region to the job batch so the kernel can retain this BO while
jobs are executed.
In practice that has never been a problem because the transient pool
is never shrinked, and even if it was, we still control the lifetime of
the job, so there's no reason for this BO to be freed before the GPU is
done executing the batch. But it still make sense to add the BO for
debugging purpose.
Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
This fix two dEQP tests for virgl:
dEQP-EGL.functional.image.create.gles2_cubemap_positive_x_rgba_texture
dEQP-EGL.functional.image.render_multiple_contexts.gles2_cubemap_positive_x_rgba8_texture
Signed-off-by: Lepton Wu <lepton@chromium.org>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
* Fix 2D/2DArray/3D tiling parameters:
There is a bottom threshold for width and height.
* Renable tiling for Cubemap, after setting the right parameters.
Reviewed-by: Rob Clark <robdclark@gmail.com>
Equivalent of 0c1dd9dee "broadcom/vc4: Allow importing linear BOs with
arbitrary offset/stride." for v3d.
Allows YUV buffers with a single buffer and plane offsets to be
passed in.
Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.org>
Reviewed-by: Eric Anholt <eric@anholt.net>
Input to GS is just a set of attributes, so remove explicit setup of
'position' which is meaningless for GS input processing.
Reviewed-by: Alok Hota <alok.hota@intel.com>
RADV no longer uses specific LLVM options compared to the common code.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Dave Airlie <airlied@redhat.com>
The patch adds support for 64 bit HAL_PIXEL_FORMAT_RGBA_FP16
for android platform.
Fixes android.graphics.cts.BitmapColorSpaceTest#test16bitHardware
which failed in egl due to "Unsupported native buffer format 0x16"
on chromebooks.
Signed-off-by: Nataraj Deshpande <nataraj.deshpande@intel.com>
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
v2: Update several of the comments. Drop some redundant uses of
ASSERT_UNION_OF_OTHERS_MATCHES_UNKNOWN_*_SOURCE source. Suggested by
Caio.
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Suggested-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
One shader from Metro Last Light and the rest from Rochard. In the
Rochard cases, something like:
min(1.0, max(pow(saturate(x), y), z))
was transformed to
saturate(max(pow(saturate(x), y), z))
because the result of the pow must be >= 0.
The Metro Last Light case was similar. An instance of
min(pow(abs(x), y), 1.0)
became
saturate(pow(abs(x), y))
v2: Fix some comments. Suggested by Caio.
v3: Fix setting is_intgral when the exponent might be negative. See
also Mesa MR !1778.
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
All Intel platforms had similar results. (Ice Lake shown)
total instructions in shared programs: 16280670 -> 16280659 (<.01%)
instructions in affected programs: 1130 -> 1119 (-0.97%)
helped: 11
HURT: 0
helped stats (abs) min: 1 max: 1 x̄: 1.00 x̃: 1
helped stats (rel) min: 0.72% max: 1.43% x̄: 1.03% x̃: 0.97%
95% mean confidence interval for instructions value: -1.00 -1.00
95% mean confidence interval for instructions %-change: -1.19% -0.86%
Instructions are helped.
total cycles in shared programs: 367168430 -> 367168270 (<.01%)
cycles in affected programs: 10281 -> 10121 (-1.56%)
helped: 10
HURT: 1
helped stats (abs) min: 16 max: 18 x̄: 17.00 x̃: 17
helped stats (rel) min: 1.31% max: 2.43% x̄: 1.79% x̃: 1.70%
HURT stats (abs) min: 10 max: 10 x̄: 10.00 x̃: 10
HURT stats (rel) min: 3.10% max: 3.10% x̄: 3.10% x̃: 3.10%
95% mean confidence interval for cycles value: -20.06 -9.04
95% mean confidence interval for cycles %-change: -2.36% -0.32%
Cycles are helped.
I discovered this while looking at a shader that was hurt by some other
work I'm doing. When I examined the changes, I was confused that one
instance of a comparison that was used in a discard_if was (incorrectly)
eliminated, while another instance used by a bcsel was (correctly) not
eliminated. I had to use NIR_PRINT=true to see exactly where things
when wrong.
A bunch of shaders in Goat Simulator, Dungeon Defenders, Sanctum 2, and
Strike Suit Zero were impacted.
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Fixes: 405de7ccb6 ("nir/range-analysis: Rudimentary value range analysis pass")
All Intel platforms had similar results. (Ice Lake shown)
total instructions in shared programs: 16280659 -> 16281075 (<.01%)
instructions in affected programs: 21042 -> 21458 (1.98%)
helped: 0
HURT: 136
HURT stats (abs) min: 1 max: 9 x̄: 3.06 x̃: 3
HURT stats (rel) min: 1.16% max: 6.12% x̄: 2.23% x̃: 2.03%
95% mean confidence interval for instructions value: 2.93 3.19
95% mean confidence interval for instructions %-change: 2.08% 2.37%
Instructions are HURT.
total cycles in shared programs: 367168270 -> 367170313 (<.01%)
cycles in affected programs: 172020 -> 174063 (1.19%)
helped: 14
HURT: 111
helped stats (abs) min: 2 max: 80 x̄: 21.21 x̃: 9
helped stats (rel) min: 0.10% max: 4.47% x̄: 1.35% x̃: 0.79%
HURT stats (abs) min: 2 max: 584 x̄: 21.08 x̃: 5
HURT stats (rel) min: 0.12% max: 17.28% x̄: 1.55% x̃: 0.40%
95% mean confidence interval for cycles value: 5.41 27.28
95% mean confidence interval for cycles %-change: 0.64% 1.81%
Cycles are HURT.
Found by inspection. I tried really, really hard to make a test case
that would trigger this problem, but I was unsuccesful. It's very hard
to get an instruction to produce a ne_zero result without ne_zero
sources. The most plausible way is using bcsel. That proves
problematic because bcsel interprets its sources as integers, so it
cannot currently be used to "clean" values for floating point
instructions.
No shader-db changes on any Intel platform.
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Fixes: 405de7ccb6 ("nir/range-analysis: Rudimentary value range analysis pass")
Fixes piglit tests (new in piglit!110):
- fs-underflow-fma-compare-zero.shader_test
- fs-underflow-mul-compare-zero.shader_test
v2: Add back part of comment accidentally deleted. Noticed by
Caio. Remove is_not_zero function as it is no longer used.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=111308
Fixes: fa116ce357 ("nir/range-analysis: Range tracking for ffma and flrp")
Fixes: 405de7ccb6 ("nir/range-analysis: Rudimentary value range analysis pass")
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
All Gen7+ platforms** had similar results. (Ice Lake shown)
total instructions in shared programs: 16278465 -> 16279492 (<.01%)
instructions in affected programs: 16765 -> 17792 (6.13%)
helped: 0
HURT: 23
HURT stats (abs) min: 7 max: 275 x̄: 44.65 x̃: 8
HURT stats (rel) min: 1.15% max: 17.51% x̄: 4.23% x̃: 1.62%
95% mean confidence interval for instructions value: 9.57 79.74
95% mean confidence interval for instructions %-change: 1.85% 6.61%
Instructions are HURT.
total cycles in shared programs: 367135159 -> 367154270 (<.01%)
cycles in affected programs: 279306 -> 298417 (6.84%)
helped: 0
HURT: 23
HURT stats (abs) min: 13 max: 6029 x̄: 830.91 x̃: 54
HURT stats (rel) min: 0.17% max: 45.67% x̄: 7.33% x̃: 0.49%
95% mean confidence interval for cycles value: 100.89 1560.94
95% mean confidence interval for cycles %-change: 0.94% 13.71%
Cycles are HURT.
total spills in shared programs: 8870 -> 8869 (-0.01%)
spills in affected programs: 19 -> 18 (-5.26%)
helped: 1
HURT: 0
total fills in shared programs: 21904 -> 21901 (-0.01%)
fills in affected programs: 81 -> 78 (-3.70%)
helped: 1
HURT: 0
LOST: 0
GAINED: 1
** On Broadwell, a shader was hurt for spills / fills instead of
helped.
No changes on any earlier platforms.
Fix the a / b ordering in some compares. Delete duplicate patterns.
Add a table explaining things. While I was cleaning this up, I managed
to confuse myself. The table helped sort that out.
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
This didn't fix bug #111308, but it was found will trying to find the
actual cause of that bug.
Fixes piglit tests (new in piglit!110):
- fs-fract-of-NaN.shader_test
- fs-lt-nan-tautology.shader_test
- fs-ge-nan-tautology.shader_test
No shader-db changes on any Intel platform.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=111308
Fixes: b77070e293 ("nir/algebraic: Use value range analysis to eliminate tautological compares")
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
We enabled fast clears at level > 0, but didn't minify the dimensions
when comparing the box size, so we always thought it was a partial
clear and as a result never actually enabled any.
This eliminates some slow clears in Civilization VI, but they are mostly
during initialization and not the main rendering.
Thanks to Dan Walsh for noticing we had too many slow clears.
Fixes: 393f659ed8 ("iris: Enable fast clears on other miplevels and layers than 0.")
Reviewed-by: Rafael Antognolli <rafael.antognolli@intel.com>
Otherwise it doesn't exist and can't be parsed, so everything dies at
screen init time.
Fixes: 6dc4ddc5f8 ("iris: use driconf for 'bo_reuse' parameter")
Some functionality has been added to deqp-volt to only print
regressions, so update our version of it and use the new options.
Signed-off-by: Tomeu Vizoso <tomeu.vizoso@collabora.com>
Acked-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
LLVM 7.0 ditched the pmulu intrinsics.
This is only a trivial patch to use the fallback code instead.
It'll likely produce atrocious code since the pattern doesn't match what
llvm itself uses in its autoupgrade paths, hence the pattern won't be
recognized.
Should fix https://bugs.freedesktop.org/show_bug.cgi?id=111496
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
Reviewed-by: Dave Airlie <airlied@redhat.com>
Fixes a possible data race spotted while debugging on other EGL
related failures where glFinish and eglCreateContext are going on at
the same time:
==11558== Possible data race during read of size 1 at 0x5E78CD0 by thread #23
==11558== Locks held: 1, at address 0x5E77CA8
==11558== at 0x61B71D4: bo_alloc_internal (brw_bufmgr.c:639)
==11558== by 0x61B7328: brw_bo_alloc (brw_bufmgr.c:669)
==11558== by 0x61EF975: recreate_growing_buffer (intel_batchbuffer.c:231)
==11558== by 0x61EFAAE: intel_batchbuffer_reset (intel_batchbuffer.c:255)
==11558== by 0x61EFB85: intel_batchbuffer_reset_and_clear_render_cache (intel_batchbuffer.c:280)
==11558== by 0x61F0507: brw_new_batch (intel_batchbuffer.c:551)
==11558== by 0x61F12C1: _intel_batchbuffer_flush_fence (intel_batchbuffer.c:888)
==11558== by 0x61BDD6B: intel_glFlush (brw_context.c:296)
==11558== by 0x61BDDB9: intel_finish (brw_context.c:307)
==11558== by 0x623831B: _mesa_Finish (context.c:1906)
==11558== by 0x46D556: deqp::egl::GLES2ThreadTest::Operation::execute(tcu::ThreadUtil::Thread&)
==11558== by 0x721502: tcu::ThreadUtil::Thread::run()
==11558==
==11558== This conflicts with a previous write of size 1 by thread #26
==11558== Locks held: 1, at address 0x5D09878
==11558== at 0x61B98A9: brw_bufmgr_enable_reuse (brw_bufmgr.c:1541)
==11558== by 0x61BF09D: brw_process_driconf_options (brw_context.c:854)
==11558== by 0x61BF6CA: brwCreateContext (brw_context.c:993)
==11558== by 0x621181F: driCreateContextAttribs (dri_util.c:473)
==11558== by 0x53FE87B: dri2_create_context (egl_dri2.c:1388)
==11558== by 0x53EE7BE: eglCreateContext (eglapi.c:807)
==11558== by 0x5C8AB9: eglw::FuncPtrLibrary::createContext(void*, void*, void*, int const*) const
==11558== by 0x46E027: deqp::egl::GLES2ThreadTest::CreateContext::exec(tcu::ThreadUtil::Thread&)
Signed-off-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
When u_upload_mgr fills up a buffer, it unmaps and destroys it. Our
unmap function was automatically performing the equivalent of a
FlushMappedBufferRange call in this case. Because the buffer mapping
is persistent and coherent, we don't actually do any flushing when we
do the rest of the writes to the buffer - we were just doing one final
one at the end. But we would be using the uploaded contents on the
GPU the whole time.
This certainly shouldn't be necessary for streaming buffers, and if
such flushing and dirtying is necessary for coherent buffers, this is
wildly insufficient.
Drops a small number of constant packets and PIPE_CONTROL flushes from
most benchmarks that I've looked at. Doesn't seem to make much of an
impact on performance, however.
Thanks to Felix Degrood for noticing that we were emitting more
3DSTATE_CONSTANT_* packets than we needed to.
NIR shaders use GLSL types (note: these live outside libglsl), and
nine needs to properly initialize these just like the other state
trackers. This fixes an assertion failure when TTN is used.
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Fixes:
dEQP-GLES3.functional.shaders.switch.switch_in_do_while_loop_dynamic_vertex
dEQP-GLES3.functional.shaders.switch.switch_in_do_while_loop_dynamic_fragment
Signed-off-by: Rob Clark <robdclark@chromium.org>
Reviewed-by: Eric Anholt <eric@anholt.net>
While resolving jumps to skip intermediate jumps from the structured
CFG, maintain the successors and predecessors correctly.
Signed-off-by: Rob Clark <robdclark@chromium.org>
Reviewed-by: Eric Anholt <eric@anholt.net>
This adds the ability for intel devices that:
* Only load on i965
* Only load on iris
* First attempt i965, and try iris next
* First attempt iris, and try i965 next
Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
If a field name differs slightly between two generations then this
change will still add the fields into the same group.
For example, these will be treated as equal:
* "Software Exception" and "Software Exception"
* "Per Thread" and "Per-Thread"
Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
We were clamping the LOD to force non-mipmap filtering, but that means
that the HW doesn't get to select between the min and mag filters.
Setting MIPFILTER_LINEAR_FAR appears to force non-mipmap filtering.
Fixes all failures in dEQP-GLES2.functional.texture.filtering.2d.*
Reviewed-by: Rob Clark <robdclark@chromium.org>
See the previous commit for the explanation of the Fixes tag.
Hurts 21 shaders in shader-db. All of the hurt shaders are in Unreal
Engine 4 tech demos.
Reviewed-by: Matt Turner <mattst88@gmail.com>
Fixes: 7afa26d4e3 ("nir: Add lowering for nir_op_bitfield_reverse.")
This caused a problem on Sandybridge where an open-coded
bitfieldReverse() function could be optimized to a
nir_op_bitfield_reverse that would generate an unsupported BFREV
instruction in the backend. This was encountered in some Unreal4 tech
demos in shader-db. The bug was not previously noticed because we don't
actually try to run those demos on Sandybridge.
The fixes tag is a bit a lie. The actual bug was introduced about
26,000 commits earlier in 371c4b3c48 ("nir: Recognize open-coded
bitfield_reverse."). Without the NIR lowering pass, the flag needed to
avoid the optimization does not exist. Hopefully nobody will care to
fix this on an earlier Mesa release.
Reviewed-by: Matt Turner <mattst88@gmail.com>
Fixes: 7afa26d4e3 ("nir: Add lowering for nir_op_bitfield_reverse.")
Reduces the size of the u_format_table.c file by 140k (out of 1.64M)
and makes me less confused about endianness in gallium.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Acked-by: Adam Jackson <ajax@redhat.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
The formats affected are:
- LA x (16_FLOAT, 32_FLOAT, 32_UINT, 32_SINT)
- R8G8B8 x (UNORM, SNORM, SRGB, USCALED, SSCALED, UINT, SINT)
- RG/RGB/RGBA x (64_FLOAT, 32_FLOAT, 16_FLOAT, 32_UNORM, 32_SNORM,
32_USCALED, 32_SSCALED, 32_FIXED, 32_UINT, 32_SINT)
- RGB/RGBA x (16_UNORM, 16_SNORM, 16_USCALED, 16_SSCALED,
16_UINT, 16_SINT)
- RGBx16 x (UNORM, SNORM, FLOAT, UINT, SINT)
- RGBx32 x (FLOAT, UINT, SINT)
- RA x (16_FLOAT, 32_FLOAT, 32_UINT, 32_SINT)
The updated st_formats.c unit test checks that the formats affected by
this change are all array formats in the equivalent Mesa format (if
any). Mesa's array format definition is clear: the value stored is an
array (increasing memory address) of values of the channel's type.
It's also the only thing that makes sense for the RGB types, or very
large types like RGBA64_FLOAT (A should not move to the low address
because the cpu is BE).
Acked-by: Roland Scheidegger <sroland@vmware.com>
Acked-by: Adam Jackson <ajax@redhat.com>
Tested-by: Matt Turner <mattst88@gmail.com> (unit tests on BE)
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Nothing accessed the .value field, just the .chan. Unwrap all the
code from the union, for clarity (and 13k less generated code).
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Acked-by: Adam Jackson <ajax@redhat.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Shaves 30k off of the 1.6M .c file, and makes for less noise for me
trying to understand how gallium formats actually work.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Acked-by: Adam Jackson <ajax@redhat.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Instructions attached to blocks are never explicitly freed. Let's
use ralloc() to attach those objects to the compiler context so that
they are automatically freed when the ctx object is freed.
Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
I don't know how Meson didn't hit this issue, when it too already uses
-Werror=incompatible-pointer-types
Fixes: 3dd299c3d5 ("glx: Sync <GL/glxext.h> with Khronos")
Reviewed-by: Eric Engestrom <eric@engestrom.ch>
Reviewed-by: Adam Jackson <ajax@redhat.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Looks like initial RE was wrong and some fields have different purpose.
I.e. there's no "disable_mipmap" field, it's actually part of another field
that selects mipmap filtering.
Also fix layout position.
Reviewed-by: Qiang Yu <yuq825@gmail.com>
Signed-off-by: Vasily Khoruzhick <anarsoul@gmail.com>
This fixes the following CTS test on 32-bit systems:
GTF-GL46.gtf30.GL3Tests.packed_depth_stencil.packed_depth_stencil_init
It does glGetTexImage of a 16-bit SNORM image, requesting 32-bit UNORM
data. In get_tex_rgba_uncompressed, we round trip through float to
handle image transfer ops for clamping. _mesa_format_convert does:
_mesa_float_to_unorm(0.571428597f, 32)
which translated to:
_mesa_lroundevenf(0.571428597f * 0xffffffffu)
which produced different results on 64-bit and 32-bit systems:
64-bit: result = 0x92492500
32-bit: result = 0x80000000
This is because the size of "long" varies between the two systems, and
0x92492500 is too large to fit in a signed 32-bit integer. To fix this,
we switch to the new _mesa_i64roundevenf function which always does the
64-bit operation.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=104395
Fixes: 594fc0f859 ("mesa: Replace F_TO_I() with _mesa_lroundevenf().")
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
This always returns a int64_t, translating to _mesa_lroundevenf on
systems where long is 64-bit, and llrintf where "long long" is needed.
Fixes: 594fc0f859 ("mesa: Replace F_TO_I() with _mesa_lroundevenf().")
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
GLX_EXT_import_context operates only on indirect contexts, a direct
context cannot possibly support it. Without this change the extension
will appear in the combined GLX extension string even if it is missing
from the server string, indicating a lack of required server support.
At least on Linux, we can use the ELF auxiliary vector to
detect the presence of AltiVec, VSX and other CPU features
without having to go through handling SIGILL, which has
various problems of its own.
A similar thing is already being done for ARM to detect NEON.
Reviewed-by: Matt Turner <mattst88@gmail.com>
Signed-off-by: Daniel Kolesa <daniel@octaforge.org>
Gen11 adds support for specifying the render target index and src0
alpha present bits in the extended message descriptor. Previously,
we had to use a message header for this, requiring extra instructions
to write the fields, and two registers of extra payload.
Improves performance on my ICL 8x8 frequency locked to 700Mhz, on iris:
GfxBench5 Manhattan 3.0: 2.13635% +/- 0.159859% (n=5)
GfxBench5 Aztec Ruins: 1.57173% +/- 0.128749% (n=5)
Synmark2 OglDeferred: 2.86914% +/- 0.191211% (n=10)
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
This takes care of generate_fb_write/fire_fb_write/brw_fb_WRITE's stuff
earlier in the visitor. It will also make it easier to generate SENDSC
messages with indirect extended descriptors in a few patches.
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Annoyingly, these bits exist in some extended message descriptors
(in particular render target writes), but they don't have any
corresponding bits in the ISA encoding. So we can't use an immediate
and have to fall back to an indirect extended descriptor.
Thanks to Jason Ekstrand for reminding me that you can still set these
bits via an indirect descriptor, even if they don't exist in the ISA.
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
src0 vstride and type overlap with bits of the extended descriptor.
brw_set_desc() also sets the extended descriptor to 0. So by setting
the descriptor, then setting src0, we were accidentally setting a bunch
of extended descriptor bits unintentionally.
When using this infrastructure for framebuffer writes (in a future
patch), this ended up setting the extended descriptor bit 20, which is
"Null Render Target" on Icelake, causing nothing to be written to the
framebuffer.
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
We need two different values of the register, one for NGG and one for
legacy, in order to fix edge flags for the legacy pipeline.
Passing the ngg flag to emit_clip_regs would be too complicated,
so CONTEXT_REG_RMW is used for partial register updates.
Acked-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Right now we're leaking all block and instruction objects allocated by
the compiler. Let's clean things up before leaving
midgard_compile_shader_nir().
Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
src/gallium/winsys/svga/drm/vmw_screen.c: In function ‘vmw_dev_compare’:
src/gallium/winsys/svga/drm/vmw_screen.c:48:12: warning: implicit declaration of function ‘major’ [-Wimplicit-function-declaration]
48 | return (major(*(dev_t *)key1) == major(*(dev_t *)key2) &&
| ^~~~~
src/gallium/winsys/svga/drm/vmw_screen.c:49:12: warning: implicit declaration of function ‘minor’ [-Wimplicit-function-declaration]
49 | minor(*(dev_t *)key1) == minor(*(dev_t *)key2)) ? 0 : 1;
| ^~~~~
That file (and many others) already has the proper #include with their
respective guards, but scons wasn't defining them, resulting in implicit
functions being used instead (and an always-true check that's probably
breaking something down the line).
Note that I'm cheating a bit here because Scons doesn't seem to have
a clean way to detect the existence of major() et al. as functions or
macros, so I'm taking the shortcut of just detecting the presence of the
header and assuming its contents is what we expect.
Signed-off-by: Eric Engestrom <eric.engestrom@intel.com>
Reviewed-By: Jose Fonseca <jfonseca@vmware.com>
This mirrors the vs/gs keys, and will be needed when adding images
support.
The const changes also mirror how the draw code work (as is needed
when we add images)
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Not sure how I missed this before, but compswap was hitting an
assert here as it is it's own special case.
Fixes: b5ac381d8f ("gallivm: add buffer operations to the tgsi->llvm conversion.")
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Looks like a copy/paste error. This patch prevents a segfault when
running the following on BDW:
INTEL_DEBUG=no8,no16,do32 ./deqp-vk -n \
dEQP-VK.subgroups.arithmetic.compute.subgroupmin_dvec4
For the curious, the message we're getting is:
CS compile failed: Failure to register allocate. Reduce number
of live scalar values to avoid this.
Fixes: 864737ce6c ("i965/fs: Build 32-wide compute shader when needed.")
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
No driver implements them yet, but this is a long way toward gallium
having matching format enums for Mesa formats.
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
To add ASTC 3D compression formats, we need to be able to express the
block depth. While I'm touching every line, line up the columns of
the CSV again as they've drifted over time.
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
When moving constants, if switching to a floating-point representation
doesn't break anything, we'd rather have an fmov than an imov,
permitting inlining the constant in many circumstances.
total quadwords in shared programs: 3408 -> 3366 (-1.23%)
quadwords in affected programs: 1188 -> 1146 (-3.54%)
helped: 41
HURT: 0
helped stats (abs) min: 1 max: 2 x̄: 1.02 x̃: 1
helped stats (rel) min: 0.19% max: 25.00% x̄: 9.65% x̃: 11.11%
95% mean confidence interval for quadwords value: -1.07 -0.98
95% mean confidence interval for quadwords %-change: -11.38% -7.93%
Quadwords are helped.
Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Storing constants as float doesn't make sense when we have integer
instructions; better to switch to be integer natively and coerce to/from
float rather than the opposite.
Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
This fixes dEQP-GLES3.functional.texture.specification subtests on iris:
- texsubimage3d_depth.depth24_stencil8_2d_array
- texsubimage3d_depth.depth32f_stencil8_2d_array
- texsubimage3d_depth.depth_component32f_2d_array
- texsubimage3d_depth.depth_component24_2d_array
- texstorage2d.format.depth24_stencil8_2d
- texstorage2d.format.depth32f_stencil8_2d
- texstorage2d.format.depth_component24_2d
- texstorage2d.format.depth_component32f_2d
- texstorage3d.format.depth24_stencil8_2d_array
- texstorage3d.format.depth32f_stencil8_2d_array
- texstorage3d.format.depth_component24_2d_array
- texstorage3d.format.depth_component32f_2d_array
Here, something appears to be going wrong with having this bit set
during blorp_copy operations for texture upload, which override the
format to R8G8B8A8_UINT.
AFAICT this bit should have no effect for integer surfaces, as it has
to do with blending, and integer blending is not a thing. So it should
be harmless to disable it.
The Windows driver appears to be setting this bit universally, so
I am unclear why we would need to. Perhaps they simply haven't run
into this issue.
Fixes: f741de236b ("isl: Enable Unorm Path in Color Pipe")
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Jason suggested I remove this in review, and he's right. AFAICT this
affects blending, and that just isn't going to happen on buffers.
Fixes: f741de236b ("isl: Enable Unorm Path in Color Pipe")
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
For each mipmaps, the driver will store the clear values (8-bytes)
and the TC-compat zrange value (4-bytes).
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
The workaround was entirely in common code, and it's needed in radeonsi
too so just always do it when necessary. Fixes
KHR-GL45.shader_image_load_store.advanced-allStages-oneImage on gfx9
with LLVM 8.
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
GL and Vulkan allow you to bind a single layer of a 3D texture to a 2D
image, and we weren't implementing a workaround for that on gfx9 that
TGSI was. Copy it over.
Fixes KHR-GL45.shader_image_load_store.non-layered_binding with radeonsi
NIR.
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
16-bit and 32-bit values match hardware values but 8-bit doesn't.
This fixes dEQP-VK.pipeline.input_assembly.* with 8-bit index.
Fixes: 372c3dcfdb ("radv: implement VK_EXT_index_type_uint8")
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl
The virgl formats are fixed in time snapshots of the gallium ones,
we just need to provide a translation table between them when
we enter the hardware.
This fixes a regression since Eric renumbered the gallium table.
Fixes: c45c33a5a2 (gallium: Remove manual defining of PIPE_FORMAT enum values.)
Bugzilla: https://bugs.freedesktop.org/111454
v1 by Dave Airlie <airlied@redhat.com>
v2: virgl: Add a number of formats to the table that are used, e.g. for vertex
attributes
v3: cover some more missing formats from a piglit run
Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
pp has vector units and some operations can be optimized when bundled
together.
Benchmarking this with piglit shaders shows that the instruction count
can be greatly reduced on many examples with vectorize.
Signed-off-by: Erico Nunes <nunes.erico@gmail.com>
Reviewed-by: Qiang Yu <yuq825@gmail.com>
nir vec4 fcsel assumes that each component of the condition will be used
to select the same component from the options, but pp can't implement
that since it only has 1 component for the condition.
Signed-off-by: Erico Nunes <nunes.erico@gmail.com>
Reviewed-by: Qiang Yu <yuq825@gmail.com>
The previous spill stack was fixed and too small, and caused instability
in programs requiring spilling for roughly more than one value.
This patch adds a dynamic calculation of the buffer size based on stack
utilization and switches it to a separate allocation at flush time that
will fit the shader that requires the largest buffer.
Signed-off-by: Erico Nunes <nunes.erico@gmail.com>
Reviewed-by: Vasily Khoruzhick <anarsoul@gmail.com>
Reviewed-by: Qiang Yu <yuq825@gmail.com>
It's not used by anything anymore now that so much lowering has been
moved into NIR. Sadly, we still need on in brw_compile_gs() for
geometry shaders on Sandy Bridge. Short of a lot of pointless work,
that one's probably not going away.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Create a unified table to handle pipe format to texture
and render target format lookup.
Reviewed-by: Vasily Khoruzhick <anarsoul@gmail.com>
Reviewed-by: Erico Nunes <nunes.erico@gmail.com>
Signed-off-by: Qiang Yu <yuq825@gmail.com>
Put the uncached GTT type at a higher index than the visible VRAM type,
rather than having GTT first.
When we don't have dedicated VRAM, we don't have a non-visible VRAM
type, and the property flags for GTT and visible VRAM are identical.
According to the spec, for types with identical flags, we should give
the one with better performance a lower index.
Previously, apps which follow the spec guidance for choosing a memory
type would have picked the GTT type in preference to visible VRAM (all
Feral games will do this), and end up with lower performance.
On a Ryzen 5 2500U laptop (Raven Ridge), this improves average FPS in
the Rise of the Tomb Raider benchmark by up to ~30%. Tested a couple of
other (Feral) games and saw similar improvement on those as well.
Signed-off-by: Alex Smith <asmith@feralinteractive.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Cc: 19.2 <mesa-stable@lists.freedesktop.org>
(Bas: CCing this to 19.2-rc due to high impact and limited complexity)
Add better liveness analysis that was modelled after one in vc4.
It uses live ranges and is aware of multiple blocks which is prerequisite
for adding CF support
Tested-by: Andreas Baierl <ichgeh@imkreisrum.de>
Reviewed-by: Qiang Yu <yuq825@gmail.com>
Reviewed-by: Erico Nunes <nunes.erico@gmail.com>
Signed-off-by: Vasily Khoruzhick <anarsoul@gmail.com>
Mali4x0 supports only gl_FragColor. gl_FragDepth is not supported.
Check that we don't get anything but gl_FragColor in shader outputs.
Reviewed-by: Qiang Yu <yuq825@gmail.com>
Signed-off-by: Vasily Khoruzhick <anarsoul@gmail.com>
We don't have a special OP to store color in PP, all we need to do is to
store gl_FragColor into reg0, thus it's just a mov and therefore ALU node.
Yet we still need to indicate that it's store_color op so regalloc ignores
its destination.
Tested-by: Andreas Baierl <ichgeh@imkreisrum.de>
Reviewed-by: Qiang Yu <yuq825@gmail.com>
Reviewed-by: Erico Nunes <nunes.erico@gmail.com>
Signed-off-by: Vasily Khoruzhick <anarsoul@gmail.com>
Create ppir block for each corresponding NIR block and populate
its successors. It will be used later in liveness analysis and
in CF support
Tested-by: Andreas Baierl <ichgeh@imkreisrum.de>
Reviewed-by: Qiang Yu <yuq825@gmail.com>
Reviewed-by: Erico Nunes <nunes.erico@gmail.com>
Signed-off-by: Vasily Khoruzhick <anarsoul@gmail.com>
We can get following from NIR:
(1) r1 = r2
(2) r2 = ssa1
Note that r2 is read before it's assigned, so there's no node for
it in comp->var_nodes. We need to create a dummy node in this case
which sole purpose is to hold ppir_dest with reg in it.
Tested-by: Andreas Baierl <ichgeh@imkreisrum.de>
Reviewed-by: Qiang Yu <yuq825@gmail.com>
Reviewed-by: Erico Nunes <nunes.erico@gmail.com>
Signed-off-by: Vasily Khoruzhick <anarsoul@gmail.com>
There can be several root nodes, i.e.:
(1) r0 = r1
(2) r2 = r3
(3) branch if (ssa1)
We need to make (3) depend on (1) and (2), old code added
dependency only for (2), and (1) was kept as root node since there
is no branch/discard or store color between two movs.
Tested-by: Andreas Baierl <ichgeh@imkreisrum.de>
Reviewed-by: Qiang Yu <yuq825@gmail.com>
Reviewed-by: Erico Nunes <nunes.erico@gmail.com>
Signed-off-by: Vasily Khoruzhick <anarsoul@gmail.com>
ppir_lower_load() and ppir_lower_load_texture() assume that node
is in the same block as its successors, fix it by cloning each
ld_uni and ld_tex to every block.
It also reduces register pressure since values never cross block
boundaries and thus never appear in live_in or live_out of any block,
so do it for varyings as well.
Tested-by: Andreas Baierl <ichgeh@imkreisrum.de>
Reviewed-by: Qiang Yu <yuq825@gmail.com>
Reviewed-by: Erico Nunes <nunes.erico@gmail.com>
Signed-off-by: Vasily Khoruzhick <anarsoul@gmail.com>
Const nodes are now cloned for each user, i.e. const is guaranteed to have
exactly one successor, so we can use ppir_do_one_node_to_instr() and
drop insert_to_each_succ_instr()
Tested-by: Andreas Baierl <ichgeh@imkreisrum.de>
Reviewed-by: Qiang Yu <yuq825@gmail.com>
Reviewed-by: Erico Nunes <nunes.erico@gmail.com>
Signed-off-by: Vasily Khoruzhick <anarsoul@gmail.com>
On commit f6e7de41d7, we started emitting 3DSTATE_LINE_STIPPLE as part
of the non-dynamic state. That gets re-emitted every time we bind a new
VkPipeline. But that instruction is non-pipelined, and it caused a perf
regression of about 9-10% on Dota2.
This commit makes anv_dynamic_state_copy() return a mask with only the
state that has changed when copying it. 3DSTATE_LINE_STIPPLE won't be
emitted anymore unless it has changed, fixing the problem above.
v2: Improve commit message and add documentation about skipped checks
(Jason)
Fixes: f6e7de41d7 ("anv: Implement VK_EXT_line_rasterization")
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
We can statically determine from the disassembly if helper invocations
will be needed, so we can validate the corresponding bit in the
cmdstream and thus avoid printing the bit itself in the decode.
Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
We check for texture ops which calculate derivatives (either explicitly
via dFd* or implicitly) and mark the shader as requiring helper
invocations.
Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
The variables level and start_layer are not initialized, then
initialized if we have a BUFFER_BIT_DEPTH set. We assert on them
later using the same check. This should be enough but GCC 9.1.1 is
not convinced, so let's initialize the variables.
Acked-by: Eric Engestrom <eric.engestrom@intel.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Code that used it was removed in 4ebe6b2e72 ("tgsi: Drop the SSE2
constants setup that's been dead code since 2011.")
Acked-by: Eric Engestrom <eric.engestrom@intel.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Initialize `next_batch_addr` and `second_level`. If the batch is well
formed, those values will be overriden, if not, they are as good as
uninitialized garbage.
Acked-by: Eric Engestrom <eric.engestrom@intel.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
The helper check_node_type() is only used when DEBUG is set (in the
function below), but ASSERTED macro uses NDEBUG. So just guard the
helper with #ifdef. If we see more such cases we might consider a
ASSERTED-like macro for the DEBUG case.
Reviewed-by: Eric Engestrom <eric.engestrom@intel.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Compiler can't see that d is initialized.
../src/intel/compiler/brw_vec4_nir.cpp: In function ‘int brw::try_immediate_source(const nir_alu_instr*, brw::src_reg*, bool, const gen_device_info*)’:
../src/intel/compiler/brw_vec4_nir.cpp:984:12: warning: ‘d’ may be used uninitialized in this function [-Wmaybe-uninitialized]
984 | d = MAX2(-d, d);
Assert that we expect at least one component -- hence d going to be
set. That by itself is not enough, so also zero initialize the
variable.
Acked-by: Eric Engestrom <eric.engestrom@intel.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
...by copying the implementation of anv_get_absolute_timeout().
Appears to fix a CTS test with 32-bit builds:
GTF-GL46.gtf32.GL3Tests.sync.sync_functionality_clientwaitsync_flush
Fixes: f459c56be6 ("iris: Add fence support using drm_syncobj")
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Eric Engestrom <eric@engestrom.ch>
Rafael Antognolli tracked down a performance gap between i965 and iris
in Synmark2's OglCSDof microbenchmark, noting that iris was performing
substantially more memory reads and writes, with substantially fewer
L3 hits. He suggested that something might be wrong with MOCS, or L3
configs, at which point I came up with a theory...
It would appear that the STATE_BASE_ADDRESS command updates the MOCS
settings for various base addresses even if you don't specify the
"Modify Enable" bit for that address. Until now, we had been setting
only the MOCS for bases we intended to change, leaving the others
"blank" which is MOCS table entry 0, which is uncached.
Most data access has a more specific MOCS (e.g. in SURFACE_STATE),
but scratch access uses the Stateless Data Port Access MOCS from
STATE_BASE_ADDRESS. So this meant all scratch access was uncached.
Improves performance in Synmark2's OglCSDof by 2x, bringing iris
on par with the existing i965 driver.
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Fix this build error on macOS.
../src/glx/apple/glx_empty.c:158:4: error: void function 'glXQueryGLXPbufferSGIX' should not return a value [-Wreturn-type]
return 0;
^ ~
Fixes: 3dd299c3d5 ("glx: Sync <GL/glxext.h> with Khronos")
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Eric Engestrom <eric@engestrom.ch>
Reviewed-by: Adam Jackson <ajax@redhat.com>
Similarly to before, this didn't properly handle varying structs with
doubles in them.
This doesn't fix any tests, but was noticed while looking at the code.
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
The old version wasn't as accurate as it could be, and didn't handle
double variables inside structs correctly. Walk the path to compute the
actual components affected.
In combination with the previous commit fixes
KHR-GL45.enhanced_layouts.varying_structure_locations.
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
This is already done in get_deref_offset() in the common code. We were
adding it twice accidentally.
Fixes KHR-GL45.enhanced_layouts.varying_array_locations.
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Some users of this function (e.g. GS inputs) currently only work with
constant offsets. We got lucky since all the tests used an array index
of 0, so the non-constant part was always 0. But we still need to handle
this.
This doesn't fix any CTS test, but was noticed while debugging one.
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Now that LLVM 9 will be released soon, we will only support
LLVM 8, 9 and master (10).
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Fixes errors seen with eglSetBlobCacheFuncsANDROID on Android when
running dEQP that terminates and reinitializes a display.
Fixes: 6f5b57093b "egl: add support for EGL_ANDROID_blob_cache"
Signed-off-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Eric Engestrom <eric.engestrom@intel.com>
We were always resolving the buffer as if we were accessing it via
CPU maps, which don't understand any auxiliary surfaces. But we often
copy to a temporary using BLORP, which understands compression just
fine. So we can avoid the resolve, and accelerate the copy as well.
Fixes: 9d1334d2a0 ("iris: Use copy_region and staging resources to avoid transfer stalls")
Reviewed-by: Rafael Antognolli <rafael.antognolli@intel.com>
This doesn't work for compressed formats, as the source texture and
temporary texture would have different block sizes. (Forcing the driver
to always take the GPU path would expose the bug.) Instead, just use
the source format for the temporary, and let blorp_copy deal with
overrides.
The one case where we can't do this is ASTC, because isl won't let us
create a linear ASTC surface. Fall back to the CPU paths there for now.
Fixes: 9d1334d2a0 ("iris: Use copy_region and staging resources to avoid transfer stalls")
Reviewed-by: Rafael Antognolli <rafael.antognolli@intel.com>
Gen11 stores the fast clear color in an "indirect clear buffer", as
a packed pixel value. Gen9 hardware stores it as a float or integer
value, which is interpreted via the format. We were trying to store
that in a buffer, for similarity with Icelake, and MI_COPY_MEM_MEM
it from there to the actual SURFACE_STATE bytes where it's stored.
This unfortunately doesn't work for blorp_copy(), which does bit-for-bit
copies, and overrides the format to a CCS-compatible UINT format. This
causes the clear color to be interpreted in the overridden format.
Normally, we provide the clear color on the CPU, and blorp_blit.c:2611
converts it to a packed pixel value in the original format, then unpacks
it in the overridden format, so the clear color we use expands to the
bits we originally desired.
However, BLORP doesn't support this pack/unpack with an indirect clear
buffer, as it would need to do the math on the GPU. On Gen11+, it isn't
necessary, as the hardware does the right thing.
This patch changes Gen9 to stop using an indirect clear buffer and
simply do PIPE_CONTROLs with post-sync write immediate operations
to store the new color over the surface states for regular drawing.
BLORP continues streaming out surface states, and handles fast clear
colors on the CPU.
Fixes: 53c484ba8a ("iris: blorp using resolve hooks")
Reviewed-by: Rafael Antognolli <rafael.antognolli@intel.com>
For renderable surfaces, we allocate SURFACE_STATEs for each bit in
res->aux.possible_usages. Sampler views use res->aux.sampler_usages.
When pinning buffers, we call surf_state_offset_for_aux() to calculate
the offset to the desired surface state. surf_state_offset_for_aux()
took an aux_modes parameter, which should be one of those two fields.
However...it was not using that parameter. It always used the broader
res->aux.possible_usages field directly.
One of the callers, update_clear_value(), was passing incorrect masks
for this parameter. It iterated through the bits in order, using
u_bit_scan(), which destructively modifies the mask. So each time we
called it, the count of bits before our selected mode was 0, which would
cause us to always update the SURFACE_STATE for ISL_AUX_USAGE_NONE,
rather than updating each in turn. This was hidden by the earlier bug
where surf_state_offset_for_aux() ignored the parameter.
Fixes: 7339660e80 ("iris: Add aux.sampler_usages.")
Reviewed-by: Rafael Antognolli <rafael.antognolli@intel.com>
This is genxml, we can compile out this code.
Fixes: 2660667284 ("iris/gen8: Re-emit the SURFACE_STATE if the clear color changed.")
Reviewed-by: Rafael Antognolli <rafael.antognolli@intel.com>
Rather than passing through the transformed gl_Position, we can use the
hardware-level varying for this, which will correctly handle
gl_FragCoord.w
Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
The offset is added to the base address, so we need to subtract it from
the size to maintain the same end address and thus prevent a buffer
overflow:
end_address = start_address + size
start_address' = start_address + offset
size' = size - offset
end_address' = start_address' + size'
= (start_address + offset) + (size - offset)
= (start_address + size) + (offset - offset)
= start_address + size
= end_address
QED.
Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
We need a special path for special varyings so we parse them correctly
instead of throwing an error when they inevitably point to bad memory.
Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
The hardware doesn't care, and a lot of Panfrost code relies on an
oversized buffer. The important part is that (stride *
padded_num_vertices) is no greater than size, which we'll need to check
once we validate instancing.
Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
We don't need to dump the contents necessary, but having the stub with
the address is useful.
Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
I don't know who thought this mask was a good idea but unfortunately it
must have been me.
Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
If we permit more $whatever through than the shader needs, that's a bit
of a waste, but it isn't an error.
Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
We don't actually care about the *contents* of the index buffer, but we
would rather like to ensure it is present and of the correct size.
Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
We can infer these stats in many cases from the disassembly, so we
should try to sanity check where we can. We may need to be fuzzy about
analysis, since analysis gives us a bound but we don't mind if it's not
used fully by the shader.
Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
We could do better by forcing the checks to *equal* zero (right now, an
indeterminate answer will pass the checks), but this is a start to guard
against some egregious cases.
Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
There are a number of conditions we need to test for to statically check
for TILE_RANGE_FAULTs, but once these checks are in order, we can print
as-is.
Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
These tags need to match up with what's actually described by the MFBD,
so check this. Once this is checked, since the type and contents of the
FBD are obvious from printing above, there's no need to explicitly mark
off the framebuffer line.
Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
For shaders using exclusively direct attribute/varyings, we can work
this out statically. For shaders with indirect access, we just set an
upper bound of 16 (the max attributes/varyings we support) and the
actual count will be reported regardless.
We proceed similarly for textures/samplers, as well as for UBOs. While
UBOs can be *indexed* indirectly, the *UBO itself* -- which is what we
count in the shader descriptor (rather than the UBO descriptors) -- is
statically determinable.
Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
This one is a little tricky, but the idea is that:
r16-r23 are always uniforms
r8-r15 are sometimes work, sometimes uniforms...
...but as work, they are always written before use
...and as uniforms, they are never written before use
So we use that heuristic to determine the count to feed the machine.
We'll record work register use in the next commit.
Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Panfrost is the only user of the macro; we are better off expanding than
having random stuff in nir.h.
Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Right now it always returns zero, but as of:
commit a48a6b8a40
Author: Adam Jackson <ajax@redhat.com>
Date: Tue Nov 14 15:13:05 2017 -0500
glx: Prepare driFetchDrawable for no-config contexts
We were hoping it would return true if the drawable could actually be
looked up. It wasn't, so that didn't go very well. With the most recent
update to <GL/glxext.h> glXQueryGLXPbufferSGIX (correctly) returns void,
so there's no longer anything else besides driFetchDrawable that depends
on the return value from __glXGetDrawableAttribute.
Acked-by: Eric Engestrom <eric.engestrom@intel.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
Minor fixups required to keep the prototypes matching and to remove
mention of retired enums.
Acked-by: Eric Engestrom <eric.engestrom@intel.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
We added this utility for vulkan where all timeouts are given as
uint64_t values. We can switch from signed to unsigned as this is the
only user and if we ever deal with signed integers somewhere else
we'll have to be careful to use the corresponding
timespec_(add|sub)_msec and always pass absolute values.
v2: Forgot to drop the test calling add_nsec() with a negative number
Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reported-by: Juan A. Suarez Romero <jasuarez@igalia.com>
Fixes: d2d70c3bb5 ("util: add a timespec helper")
Acked-by: Daniel Stone <daniels@collabora.com>
This fixes a regression introduced with scan&reduce operations
on GFX10. Note that some subgroups CTS still fail on GFX10 but
I assume it's a different issue.
This fixes dEQP-VK.subgroups.arithmetic.*.subgroupexclusive*.
Fixes: 227c29a80d "amd/common/gfx10: implement scan & reduce operations"
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Commit fixes current crashes with Vulkan applications on Android.
Fixes: c0376a1234 "util: add anon_file.h for all memfd/temp file usage"
Signed-off-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Eric Engestrom <eric@engestrom.ch>
v2: Pass through to oscreen rather than faking it (review from Marek).
Fixes: 0346b70083 ("gallium/screen: Add pipe_screen::resource_get_param")
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
At compressed_tex_sub_image we only can obtain the tex_object after
compressed_subtexture_target_check is validated for TEX_MODE_CURRENT.
So if the target is wrong the error is raised to the user.
This completes the fix for the regression introduced on "mesa: refactor
compressed_tex_sub_image function" of the pending failing tests:
dEQP-GLES3.functional.negative_api.texture.compressedtexsubimage3d
dEQP-GLES31.functional.debug.negative_coverage.get_error.texture.compressedtexsubimage3d
v2: Fix warning that texObj might be used uninitialized (Gert Wollny)
Fixes: 7df233d68d ("mesa: refactor compressed_tex_sub_image function")
Reviewed-By: Gert Wollny <gert.wollny@collabora.com>
Expose configs when allow_fp16_configs has been enabled and
DRI_LOADER_CAP_FP16 is set in the loader.
Also, make kms_swrast_dri respect format bpp, to allow for allocating
buffers wider than 32 bpp.
Make fp16 opt-in for gallium.
Signed-off-by: Kevin Strasser <kevin.strasser@intel.com>
Reviewed-by: Adam Jackson <ajax@redhat.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Emil Velikov <emil.l.velikov@gmail.com>
Expose configs when allow_fp16_configs has been enabled and
DRI_LOADER_CAP_FP16 is set in the loader.
Also, define a new dri configuration option so users can disable exposure of
fp16 formats. Make fp16 opt-in for i965.
Signed-off-by: Kevin Strasser <kevin.strasser@intel.com>
Reviewed-by: Adam Jackson <ajax@redhat.com>
Reviewed-by: Emil Velikov <emil.l.velikov@gmail.com>
Add dri formats for RGBA ordered 64 bpp IEEE 754 half precision floating
point. Leverage existing offscreen render support for
MESA_FORMAT_RGBA_FLOAT16 and MESA_FORMAT_RGBX_FLOAT16.
Signed-off-by: Kevin Strasser <kevin.strasser@intel.com>
Reviewed-by: Adam Jackson <ajax@redhat.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
In the case that __DRI_ATTRIB_FLOAT_BIT is set in the dri config, set
EGL_COLOR_COMPONENT_TYPE_FLOAT_EXT in the egl config. Add a field to the
platform driver visual to indicate if it has components that are in floating
point form.
Signed-off-by: Kevin Strasser <kevin.strasser@intel.com>
Reviewed-by: Adam Jackson <ajax@redhat.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
In order to handle pixel formats that consist of floating point data, enable
floatMode field in the dri config, and set __DRI_ATTRIB_FLOAT_BIT in the
render type attribute.
Signed-off-by: Kevin Strasser <kevin.strasser@intel.com>
Reviewed-by: Adam Jackson <ajax@redhat.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
Change dri2_add_config to take arrays of shifts and sizes, and compare with
those set in the dri config. Convert all platform driver masks
to shifts and sizes.
In order to handle older drivers, where shift attributes aren't available,
we fall back to the mask attributes and compute the shifts with ffs.
Signed-off-by: Kevin Strasser <kevin.strasser@intel.com>
Reviewed-by: Emil Velikov <emil.l.velikov@gmail.com>
bitcount is free from the pipe header dependencies that make u_math.h hard
to include by non-gallium specific code, so move it to bitscan.h. bitscan.h
is included by u_math.h so existing references will continue working.
Signed-off-by: Kevin Strasser <kevin.strasser@intel.com>
Reviewed-by: Emil Velikov <emil.l.velikov@gmail.com>
The existing mask attributes can only support up to 32 bpp. Introduce
per-channel SHIFT attributes that indicate how many bits, from lsb towards
msb, the bit field is offset. A shift of -1 will indicate that there is no
bit field set for the channel.
As old loaders will still be looking for masks, we set the masks to 0 for
any formats wider than 32 bpp.
Signed-off-by: Kevin Strasser <kevin.strasser@intel.com>
Reviewed-by: Adam Jackson <ajax@redhat.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Emil Velikov <emil.l.velikov@gmail.com>
The driver checks dri config options and loader caps to filter out certain
formats during config creation. Fold 4 call sites under a single helper
function.
Signed-off-by: Kevin Strasser <kevin.strasser@intel.com>
Reviewed-by: Emil Velikov <emil.l.velikov@gmail.com>
This debug option allows vkGet[Instance/Device]ProcAddr() to succeed
even if the extension associated with the requested entrypoint was not
enabled.
This has come in handy in a few instances when debugging VR
applications, so I thought it would be good to have a cleaned up version
upstreamed.
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
This should fix glDepthRangef issues. Eventually, something similar
should allow implementing the depth bounds test.
Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
A pair of special flags can turn the texture/sampler handle fields into
register selects. This means code like:
texture(uTextures[hr28.w], ...)
can be compiled to something like:
texture ..., fsampler[hr28.w], texture[hr28.w]
Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
This data structure is shared in other parts of the texture word, so
let's streamline printing.
Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
This allows nodes to be unsigned and prevents a class of weird
signedness bugs identified by Coverity.
Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
This path shouldn't be possible for in-spec shaders, but let's be
defensive. (Because security, right? Mostly because Coverity.)
Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
This better matches all the other atomic intrinsics such as those for
SSBOs and shared variables where the sign is part of the intrinsic
opcode. Both generators (GLSL and SPIR-V) know the sign from the type
of the image variable or handle. In SPIR-V, signed min/max are separate
opcodes from unsigned.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Eric Anholt <eric@anholt.net>
We can smush this into one-line per record as per usual. We still need
more validation and cleaning this up, especially around instancing. But
for LINEAR records, it works okay already.
Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
This consolidates texture format and dimensionality into something simple:
tiled rgba8_unorm.rgb1: 512x512
Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Textures of a smaller dimension don't need higher dimensions printed.
This allows us to be more compact, while enforcing verification that
higher dimensions must be zero.
Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
unknown3A I think I've actually seen on T6xx but.. we'll see what
happens in traces going forward. We don't want the zero noise normally,
and if they show up in the wild, we want to draw attention to them.
Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
This dramatically reduces visual clutter: now an entire
attribute/varying record looks something like:
rgba32f attribute_0[16].bgra;
which is equivalent to the raw structure:
{
.index = 0,
.format = MALI_FORMAT_RGBA32F,
.swizzle = (MALI_CHANNEL_BLUE << 9) | ....,
.src_offset = 16,
}
Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
We want to make sure we don't access a component in the swizzle that
doesn't exist in the format, since that is (as far as I know) undefined.
Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
We've never seen them, so if they come up in trace, we want to draw
attention to that.
Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Varying discard is not used by Panfrost, but the blob uses it sometimes
to have some padding in the varyings table, probably to minimize
per-draw overhead. (...We should maybe consider this ourselves!)
Let's check for this and ensure the rest of the record is consistent
with a discarded varying.
Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
It's a legacy GL thing... we don't really need to handle it *right* now,
but we shouldn't crash..
Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
This CAP controls a desktop-only extension. If the corresponding support
exists in the hardware, we don't know how to use it.
Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Subtle issue masked by how we emitted SET_VALUE jobs, but this case can
and does occur, so let's fix it.
Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
This normalizes the printed format. It also makes it easier for the
future when we may introduce semantic _warn and _error handlers.
A tripped zero is essentially a hazard to check for.
Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
If this bit is clear, MFBD preload will be enabled, and you.. don't want
that. (At least, when the bit is clear, the old contents of the
framebuffer will be preserved. I'm assuming this is what "MFBD preload"
refers to in kbase.)
Validate that this bit is always set.
Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
There is no "chunknown" structure; that part of the union is an artefact
from falsely believing vertex/tiler MFBDs could have render targets
attached (they can't). These are just plain old AFBC fields, and if
there is no AFBC, it's error to set these field.
Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
For our purposes of driver debugging, the contents of uniform buffers
are rarely interesting; we're more concerned about the metadata setting
them up.
We do need to be careful to validate the sizes of both uniforms and
uniform buffers.
Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Many structures in the command stream have a GPU address and size
determined statically. We should check that the pointers we are passed
are valid and the buffers they point to are big enough for the given
size. If they're not, an MMU fault would be raised.
Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Verify sizes / masks / etc against something logical to cull down the
trace space and automatically guard against a number of potential
hazards.
Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
While the algorithm for computing the header size has been correct for a
while, we used a major hack to conservatively guess the body size. Let's
scrap that and figure out the algorithm we actually need to use to be
bit-identical with what the hardware expects.
We do have to be careful to add the header size to total comptued BO
size.
It's not clear how big the polygon list needs to be in practice -- but
it has to be somewhat bigger than the polygon list itself. This needs
more investigation. If we size the polygon list exactly based on the
polygon_list_size field, we get faults like:
[ 1224.219886] panfrost ff9a0000.gpu: Unhandled Page fault in AS0 at VA 0x000000001BDE8000
Reason: TODO
raw fault status: 0x660003C3
decoded fault status: SLAVE FAULT
exception type 0xC3: TRANSLATION_FAULT_LEVEL3
access type 0x3: WRITE
source id 0x6600
Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
The other commented lines just add noise/entropy we don't want, and can
in fact crash the trace due to asserts failing.
Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
The polygon sizes are computed from the width/height/flags, so we can
reverse the computation and use our computation to verify the two
computation algorithms are bit-identical. If they are, we can omit the
computed fields.
Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
We have the BOs available; ensure that the bounds specified in the
command stream are actually the correct bounds.
Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
This allows the caller to call track_mmap multiple times for the same
gpu_va for the purpose of updating the mmap. This is used to trace
invisible BOs with kbase and doesn't apply to native traces.
Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
This allows us to catch a class of errors (for negative offsets, etc)
automatically.
Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
The on-the-wire representation of workgroups is not 1:1 to the decoded
Gallium-level workgroups (there are multiple valid encodings; see the
previous commit). Nevertheless, since we're now bit-identical in packing
vs the blob, we can check for a canonical form and only print the
verbose trace if we fail the canonical form.
Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
This is a blob quirk; in so much as I know, the hardware doesn't care.
But we're trying to be bit-identical to take as much entropy out of
traces as possible, so let's introduce the quirk.
Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
The routines in this file have no dependency on Gallium. Let's share
them so they can be used for a theoretical future Vulkan driver or, more
immediately, consulted when tracing.
Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
It's obvious that it's linked by virtue of us printing the struct it
links against. No need to repeat ourselves.
Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
The last remaining stuff was ARB_gl_spirv and ARB_spirv_extensions.
Note that it is really likely that we can enable it for some Gen7 (as
4.5 was), but it was not tested yet.
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
To help make sure we are running tests in the ideal number of threads,
print load stats to make obvious when there's a problem with
utilization.
This will be specially useful when we run tests on a wider variety of
devices.
Signed-off-by: Tomeu Vizoso <tomeu.vizoso@collabora.com>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Some runners may be configured such that the qemu binary might not be
available by the time we need to start running commands within the
chroot.
So make sure that it's there to avoid suprising problems in that case.
Signed-off-by: Tomeu Vizoso <tomeu.vizoso@collabora.com>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
A number of things can go wrong when building the rootfs from within a
non-native chroot, so make sure to print the bootstrap.log so we can
tell what's going on.
Signed-off-by: Tomeu Vizoso <tomeu.vizoso@collabora.com>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
It's able to run tests in parallel, fully utilizing the HW and
shortening considerable the time it takes.
Needed to disable tests in RK3288 for now because Volt doesn't support
armhf yet, though this should be fixed soon.
Tests are now run with --deqp-gl-config-name=rgba8888d24s8ms0, so we are
hitting a few more failures in tests that previously were being skipped.
The time to run the tests decreases from around 8 minutes to 1:45
minutes, allowing for extending coverage without increasing CI times too
much.
Signed-off-by: Tomeu Vizoso <tomeu.vizoso@collabora.com>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
This gives a nice boost, +20% at this time on my Vega 56. Shader
ballot should be enabled by default at some point but it reduces
performance a bit (-6%) with Wolfeinstein II. Enable it only for
Youngblood at the moment, like what we did for Talos in the past.
As a bonus point, it gets rid of some minor artifacts that only
happens when ballot is disabled for some reasons.
Cc: 19.2 <mesa-stable@lists.freedesktop.org
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Loops like:
block block_0:
vec1 32 ssa_2 = load_const (0x00000020)
vec1 32 ssa_3 = load_const (0x00000001)
loop {
vec1 32 ssa_7 = phi block_0: ssa_3, block_4: ssa_9
vec1 1 ssa_8 = ige ssa_2, ssa_7
if ssa_8 {
break
} else {
}
vec1 32 ssa_9 = iadd ssa_7, ssa_1
}
Were treated as having more than 1 iteration and after unrolling
produced wrong results, however such loop will exit during
the first iteration if not unrolled.
So we check if loop will actually loop.
Fixes tests/shaders/glsl-fs-loop-while-false-02.shader_test
Signed-off-by: Danylo Piliaiev <danylo.piliaiev@globallogic.com>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
The comments say that we should remove continue if it is the last
intruction in a loop however we remove any kind of jump.
Signed-off-by: Danylo Piliaiev <danylo.piliaiev@globallogic.com>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
Otherwise hangs are possible. This register was already set for
GS and NGG.
Fixes: 5eaed7ecfc "radv/gfx10: enable support for NAVI10, NAVI12 and NAVI14"
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Should take the max of the 2.
Fixes: ea337c8b7e "radv/gfx10: fix VS input VGPRs with the legacy path"
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
An application quitting before the destroying its GL context and
binding a NULL context might still have a radeonsi compiler thread
running and potentially still accessing the types.
Therefore take a reference for the duration of the threads' lifetime.
v2: Only ref the glsl types, the builtins should be used by the time
shader data gets to a gallium driver.
Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
The issue we're running into when running CTS is that glsl types are
deleted while builtins depending on them are not.
This happens because on one hand we have glsl types ref counted, but
builtins are not. Instead builtins are destroyed when unloading libGL
or explicitly calling glReleaseShaderCompiler().
This change removes almost entirely any dealing with glsl types
ref/unref by letting the builtins deal with it instead. In turn we
introduce a builtin ref count mechanism. Each GL context takes a
reference on the builtins when compiling a shader for the first time.
It releases the reference when the context is destroyed. It can also
explicitly release those when glReleaseShaderCompiler() is called.
Finally we also take a reference on the glsl types when loading libGL
to avoid recreating glsl types too often.
v2: Ensure we take a reference if we don't have one in link step (Lionel)
Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=110796
Reviewed-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
The compute paths in vl are a bit AMD-specific. For example, they (on
nouveau), try to use a BGRX8 image format, which is not supported.
Fixing all this is probably possible, but since the compute paths aren't
in any way better, it's difficult to care.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=111213
Fixes: 9364d66cb7 (gallium/auxiliary/vl: Add video compositor compute shader render)
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
# Install python wheels, necessary to install SCons via pip
- python -m pip install wheel
# Install SCons
- python -m pip install scons==3.0.1
- scons --version
# Install flex/bison
- set WINFLEXBISON_ARCHIVE=win_flex_bison-%WINFLEXBISON_VERSION%.zip
- if not exist "%WINFLEXBISON_ARCHIVE%" appveyor DownloadFile "https://github.com/lexxmark/winflexbison/releases/download/v%WINFLEXBISON_VERSION%/%WINFLEXBISON_ARCHIVE%"
- 7z x -y -owinflexbison\ "%WINFLEXBISON_ARCHIVE%" > nul
- set Path=%CD%\winflexbison;%Path%
- win_flex --version
- win_bison --version
# Download and extract LLVM
- if not exist "%LLVM_ARCHIVE%" appveyor DownloadFile "https://people.freedesktop.org/~jrfonseca/llvm/%LLVM_ARCHIVE%"
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=109630">Bug 109630</a> - vkQuake flickering geometry under Intel</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=110395">Bug 110395</a> - Shadows are flickering in SuperTuxKart</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=111113">Bug 111113</a> - ANGLE BlitFramebufferTest.MultisampleDepthClear/ES3_OpenGL fails on Intel Ubuntu19.04</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=111267">Bug 111267</a> - [CM246] Flickering with multiple draw calls within the same graphics pipeline if a compute pipeline is present</li>
</ul>
<h2>Changes</h2>
<p>Bas Nieuwenhuizen (4):</p>
<ul>
<li>radv: Do non-uniform lowering before bool lowering.</li>
<li>ac/nir: Use correct cast for readfirstlane and ptrs.</li>
<li>radv: Avoid binning RAVEN hangs.</li>
<li>radv: Avoid VEGA/RAVEN scissor bug in binning.</li>
</ul>
<p>Danylo Piliaiev (1):</p>
<ul>
<li>i965: Emit a dummy MEDIA_VFE_STATE before switching from GPGPU to 3D</li>
</ul>
<p>Eric Engestrom (1):</p>
<ul>
<li>util: fix mem leak of program path</li>
</ul>
<p>Erik Faye-Lund (2):</p>
<ul>
<li>gallium/dump: add missing query-type to short-list</li>
<li>gallium/dump: add missing query-type to short-list</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=111411">Bug 111411</a> - SPIR-V shader leads to GPU hang, sometimes making machine unstable</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=110814">Bug 110814</a> - KWin compositor crashes on launch</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=111069">Bug 111069</a> - Assertion fails in nir_opt_remove_phis.c during compilation of SPIR-V shader</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=111271">Bug 111271</a> - Crash in eglMakeCurrent</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=111401">Bug 111401</a> - Vulkan overlay layer - async compute not supported, making overlay disappear in Doom</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=111405">Bug 111405</a> - Some infinite 'do{}while' loops lead mesa to an infinite compilation</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=111467">Bug 111467</a> - WOLF RPG Editor + Gallium Nine Standalone: Rendering issue when using Iris driver</li>
<li><ahref="https://gitlab.freedesktop.org/mesa/mesa/issues/1878">Issue #1878</a> - meson.build:1447:6: ERROR: Problem encountered: libdrm required for gallium video statetrackers when using x11</li>
</ul>
<h2>Changes</h2>
<p>Adam Jackson (1):</p>
<ul>
<li>docs: Update bug report URLs for the gitlab migration</li>
</ul>
<p>Alan Coopersmith (5):</p>
<ul>
<li>c99_compat.h: Don't try to use 'restrict' in C++ code</li>
<li>util: Make Solaris implemention of p_atomic_add work with gcc</li>
<li>util: Workaround lack of flock on Solaris</li>
<li>meson: recognize "sunos" as the system name for Solaris</li>
<li>intel/common: include unistd.h for ioctl() prototype on Solaris</li>
</ul>
<p>Andreas Gottschling (1):</p>
<ul>
<li>drisw: Fix shared memory leak on drawable resize</li>
</ul>
<p>Andres Gomez (3):</p>
<ul>
<li>docs: Add the maximum implemented Vulkan API version in 19.1 rel notes</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=103674">Bug 103674</a> - u_queue.c:173:7: error: implicit declaration of function 'timespec_get' is invalid in C99</li>
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=104395">Bug 104395</a> - [CTS] GTF-GL46.gtf32.GL3Tests.packed_pixels.packed_pixels tests fail on 32bit Mesa</li>
@@ -118,13 +116,10 @@ TBD.
<li><ahref="https://bugs.freedesktop.org/show_bug.cgi?id=111734">Bug 111734</a> - Geometry shader with double interpolators fails in LLVM</li>
</ul>
</ul>
<h2>Changes</h2>
<ul>
<p>Adam Jackson (1):</p>
<ul>
<li>docs: Update bug report URLs for the gitlab migration</li>
@@ -444,8 +439,6 @@ TBD.
<li>travis: Fail build if any command in if statement fails.</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.
