fdo-mirrors/mesa
1
0
Fork 0
mirror of https://gitlab.freedesktop.org/mesa/mesa.git synced 2026-05-21 06:48:09 +02:00
Code Issues Projects Releases Packages Wiki Activity Actions 1
172671 commits 66 branches 936 tags 998 MiB
Commit graph

2571 commits

Author SHA1 Message Date
Daniel Schürmann
a56ddca4e8 aco: make all exec accesses non-temporaries 
So that they are not counted into the register demand.

Totals from 107336 (77.00% of 139391) affected shaders (Navi10):
VGPRs: 4023452 -> 4023248 (-0.01%); split: -0.01%, +0.01%
SpillSGPRs: 14088 -> 12571 (-10.77%); split: -11.03%, +0.26%
CodeSize: 266816164 -> 266765528 (-0.02%); split: -0.04%, +0.02%
MaxWaves: 1553339 -> 1553374 (+0.00%); split: +0.00%, -0.00%
Instrs: 50977701 -> 50973093 (-0.01%); split: -0.02%, +0.01%
Cycles: 1733911128 -> 1733605320 (-0.02%); split: -0.05%, +0.03%
VMEM: 40867650 -> 40900204 (+0.08%); split: +0.13%, -0.05%
SMEM: 6835980 -> 6829073 (-0.10%); split: +0.10%, -0.20%
VClause: 1032783 -> 1032788 (+0.00%); split: -0.01%, +0.01%
SClause: 2103705 -> 2104115 (+0.02%); split: -0.09%, +0.11%
Copies: 3195658 -> 3193656 (-0.06%); split: -0.30%, +0.24%
Branches: 1140213 -> 1140120 (-0.01%); split: -0.05%, +0.04%
PreSGPRs: 3603785 -> 3437064 (-4.63%); split: -5.13%, +0.50%
PreVGPRs: 3321996 -> 3321850 (-0.00%)

Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/8870>
 2021-02-12 22:41:31 +00:00
Daniel Schürmann
5d7b3bf1a7 aco: handle non-temp phi definitions and operands 
This will be necessary as we make exec non-temp.

No fossil-db changes.

Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/8870>
 2021-02-12 22:41:31 +00:00
Daniel Schürmann
e663a15098 aco: don't create unnecessary exec phi on merge blocks 
No fossil-db changes.

Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/8870>
 2021-02-12 22:41:31 +00:00
Daniel Schürmann
44a76ba16d aco: use VCC as regular SGPR pair on GFX10 
There is no need to reserve it for special purposes, only.

Totals from 139391 (100.00% of 139391) affected shaders (Navi10):
VGPRs: 4738296 -> 4738156 (-0.00%); split: -0.01%, +0.00%
SpillSGPRs: 16188 -> 14968 (-7.54%); split: -7.60%, +0.06%
CodeSize: 294204472 -> 294118048 (-0.03%); split: -0.04%, +0.01%
MaxWaves: 2119584 -> 2119619 (+0.00%); split: +0.00%, -0.00%
Instrs: 56075079 -> 56056235 (-0.03%); split: -0.05%, +0.01%
Cycles: 1757781564 -> 1755354032 (-0.14%); split: -0.16%, +0.02%
VMEM: 52995887 -> 52996319 (+0.00%); split: +0.07%, -0.07%
SMEM: 9005338 -> 9004858 (-0.01%); split: +0.16%, -0.17%
VClause: 1178436 -> 1178331 (-0.01%); split: -0.02%, +0.01%
SClause: 2403649 -> 2404542 (+0.04%); split: -0.14%, +0.18%
Copies: 3447073 -> 3432417 (-0.43%); split: -0.66%, +0.23%
Branches: 1166542 -> 1166422 (-0.01%); split: -0.11%, +0.10%
PreSGPRs: 4229322 -> 4235538 (+0.15%)
PreVGPRs: 3817111 -> 3817040 (-0.00%)

Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/8921>
 2021-02-12 19:00:18 +00:00
Daniel Schürmann
112f389261 aco: don't abort() if disassembly fails 
We used that to catch assembly errors in the past,
but now, there are too many hardware features we
use in ACO that are not supported by the LLVM disassembler,
that it is not really suited anymore as a debugging tool.

Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/8921>
 2021-02-12 19:00:18 +00:00
Daniel Schürmann
171fbe3ae1 aco: check get_reg_specified() on register hints 
This ensures that max_used_sgpr is adjusted accordingly.

Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/8921>
 2021-02-12 19:00:18 +00:00
Daniel Schürmann
dd16e21e97 aco: also consider VCC in get_reg_specified() 
This allows split_vector and others to keep their VCC position.

Totals from 4573 (3.28% of 139391) affected shaders (Navi10):
CodeSize: 54292268 -> 54289324 (-0.01%); split: -0.03%, +0.03%
Instrs: 10327645 -> 10326941 (-0.01%); split: -0.04%, +0.04%
Cycles: 744410748 -> 744034732 (-0.05%); split: -0.07%, +0.02%
VMEM: 749093 -> 749092 (-0.00%); split: +0.00%, -0.00%
SMEM: 269306 -> 269322 (+0.01%)
SClause: 358746 -> 358744 (-0.00%)
Copies: 826051 -> 823910 (-0.26%); split: -0.55%, +0.29%
Branches: 355074 -> 356493 (+0.40%); split: -0.01%, +0.41%

Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/8921>
 2021-02-12 19:00:18 +00:00
Daniel Schürmann
947bf0bd67 aco: don't decrease the vgpr_limit when encountering bpermute 
Instead we recalculate vgpr_limit on demand, depending on
the number of needed shared VGPRs.

Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/8921>
 2021-02-12 19:00:18 +00:00
Daniel Schürmann
b98a4d4dd7 aco: refactor GPR limit calculation 
This patch delays the calculation of GPR limits in order to
precisely incorporate extra registers (VCC etc.) and shared VGPRs.

Additionally, the allocation granularity is used to set the config.
This has some effect on the reported SGPR stats.

Totals (Navi10):
SGPRs: 6971787 -> 17753642 (+154.65%)

Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/8921>
 2021-02-12 19:00:18 +00:00
Daniel Schürmann
eaf681724e aco: change gpr_alloc_granule to full alignment 
This also switches the alloc_granule of Tonga and Iceland
to 96, so that the calculation is consistent.
Also changes the granularity for RDNA to 16 to keep
better stats with the upcoming patch.

Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/8921>
 2021-02-12 19:00:18 +00:00
Daniel Schürmann
bacc3b36f5 aco: fix shared VGPR allocation on RDNA2 
VGPRs are now allocated in blocks of 8 normal
or 16 shared VGPRs, respectively.

Fixes: 14a5021aff ('aco/gfx10: Refactor of GFX10 wave64 bpermute.')

Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/8921>
 2021-02-12 19:00:18 +00:00
Rhys Perry
f7575fa71f aco: fix adjust_vertex_fetch_alpha 
These offsets were wrong and didn't match the old behaviour.

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Fixes: e8220e106b ("aco: optimize AC_FETCH_FORMAT_SNORM alpha adjust")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/8935>
 2021-02-10 09:37:17 +00:00
Rhys Perry
e2608312d3 aco: remove loop to flag loop blocks as WQM 
This is no longer necessary.

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/8446>
 2021-02-09 17:52:17 +00:00
Rhys Perry
ed020008b5 aco: rewrite setting of Exact_Branch 
Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/8446>
 2021-02-09 17:52:17 +00:00
Rhys Perry
f0074a6f05 aco: do not flag all blocks WQM to ensure we enter all nested loops in WQM 
This should no longer be necessary since the mark_block_wqm() we use to
flag break conditions as WQM now adds block to the worklist. With them
added to the worklist, get_block_needs() will add WQM to block_needs.

Adding WQM to block_needs here without adding the block to the worklist
(like we do here) can cause issues because it does not ensure that the
predecessors' branches are in WQM (needed for it to be possible to
transition to WQM in the block). This happened in an Overwatch shader.

No fossil-db changes.

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Fixes: 661922f6ac ("aco: add block to worklist in mark_block_wqm()")
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/4066
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/8446>
 2021-02-09 17:52:17 +00:00
Rhys Perry
e8220e106b aco: optimize AC_FETCH_FORMAT_SNORM alpha adjust 
This looks like it was copied from LLVM, which didn't have a fmax
intrinsic.

fossil-db (GFX8):
Totals from 43 (0.03% of 140385) affected shaders:
CodeSize: 49660 -> 49488 (-0.35%)
Instrs: 10434 -> 10348 (-0.82%)
Cycles: 41736 -> 41392 (-0.82%)
VMEM: 13793 -> 13719 (-0.54%)

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/8918>
 2021-02-09 12:58:22 +00:00
Rhys Perry
d1f93261b1 aco: always set exec_live=false 
Register demand calculation for exec masks doesn't always match
get_live_changes() and get_temp_registers(). For now, just set
exec_live=false.

fossil-db (GFX10.3):
Totals from 108230 (77.64% of 139391) affected shaders:
SGPRs: 5759658 -> 5756818 (-0.05%); split: -0.08%, +0.03%
VGPRs: 4061104 -> 4061248 (+0.00%); split: -0.00%, +0.01%
SpillSGPRs: 14114 -> 15198 (+7.68%); split: -0.10%, +7.78%
CodeSize: 266548396 -> 266603288 (+0.02%); split: -0.01%, +0.03%
MaxWaves: 1390885 -> 1390855 (-0.00%); split: +0.00%, -0.00%
Instrs: 50983353 -> 50992972 (+0.02%); split: -0.02%, +0.04%
Cycles: 1733042048 -> 1735443264 (+0.14%); split: -0.02%, +0.16%
VMEM: 41933625 -> 41914722 (-0.05%); split: +0.04%, -0.09%
SMEM: 7197675 -> 7197789 (+0.00%); split: +0.16%, -0.16%
VClause: 1050885 -> 1050978 (+0.01%); split: -0.02%, +0.03%
SClause: 2074913 -> 2071844 (-0.15%); split: -0.23%, +0.08%
Copies: 3181464 -> 3188125 (+0.21%); split: -0.38%, +0.59%
Branches: 1127526 -> 1127716 (+0.02%); split: -0.10%, +0.12%
PreSGPRs: 3376687 -> 3586076 (+6.20%); split: -0.00%, +6.20%
PreVGPRs: 3339740 -> 3339811 (+0.00%)

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Cc: mesa-stable
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/8807>
 2021-02-05 17:03:17 +00:00
Rhys Perry
cc8613c8d5 aco: add fallback algorithm in get_reg() 
The generated code is often terrible, but the situations where this is
needed are rare.

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Tony Wasserka <tony.wasserka@gmx.de>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/8713>
 2021-02-05 14:41:24 +00:00
Tony Wasserka
5df057a3e6 aco/ra: Add helper to get a PhysRegInterval for the register demand 
Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/8737>
 2021-02-03 08:17:49 +00:00
Jason Ekstrand
d360a996f9 vulkan: Add common instance and physical device structs 
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/8676>
 2021-02-01 18:54:24 +00:00
Rhys Perry
b2dbe2b87b aco: implement non-uniform get_ssbo_size 
Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/3711
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7969>
 2021-01-27 13:00:33 +00:00
Tony Wasserka
5fdc298e4b aco/ra: Remove preprocessor guards for print_regs 
This debug helper isn't used anywhere by default, but compiling it anyway
ensures future refactors don't break it.

Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/8719>
 2021-01-27 12:08:47 +01:00
Tony Wasserka
3675aefa84 aco/ra: Fix build with print_regs enabled 
Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/8719>
 2021-01-27 12:08:47 +01:00
Tony Wasserka
b241a050da aco/ra: Fix print_regs using the wrong constant to check for blocked slots 
Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/8719>
 2021-01-27 12:08:47 +01:00
Rhys Perry
0f178290ca aco: don't affect isPrecise() after applying output modifiers 
fossil-db (GFX10.3):
Totals from 26679 (19.14% of 139391) affected shaders:
SGPRs: 1757155 -> 1757059 (-0.01%); split: -0.05%, +0.04%
VGPRs: 1175932 -> 1173556 (-0.20%); split: -0.21%, +0.01%
CodeSize: 86203592 -> 85572480 (-0.73%); split: -0.73%, +0.00%
MaxWaves: 315513 -> 315805 (+0.09%); split: +0.10%, -0.00%
Instrs: 16297785 -> 16143745 (-0.95%); split: -0.95%, +0.00%

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/8718>
 2021-01-26 22:22:58 +00:00
Daniel Schürmann
b06609e903 aco: fix nir_intrinsic_ballot with wave32 
Found by inspection.

Fixes: 21db083504 ('aco/wave32: Allow setting the subgroup ballot size to 64-bit.')

Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/8703>
 2021-01-26 21:06:48 +00:00
Daniel Schürmann
9a49760e82 aco: fix VCC hint on boolean subgroup operations 
Found by inspection.

Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/8712>
 2021-01-26 14:59:30 +00:00
Rhys Perry
2f0d480c73 aco: optimize out a*1.0 if it's used as a float 
fossil-db (GFX10):
Totals from 370 (0.27% of 139391) affected shaders:
CodeSize: 641436 -> 634156 (-1.13%); split: -1.14%, +0.00%
Instrs: 117668 -> 115739 (-1.64%); split: -1.64%, +0.00%

fossil-db (GFX10.3):
Totals from 370 (0.27% of 139391) affected shaders:
SGPRs: 26888 -> 26912 (+0.09%)
VGPRs: 13964 -> 13916 (-0.34%)
CodeSize: 692008 -> 679008 (-1.88%)
MaxWaves: 4779 -> 4783 (+0.08%)
Instrs: 134265 -> 132026 (-1.67%); split: -1.67%, +0.00%

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5523>
 2021-01-26 11:36:13 +00:00
Rhys Perry
54a09545ec aco: optimize a*0.0 
fossil-db (GFX10):
Totals from 1943 (1.39% of 139391) affected shaders:
SGPRs: 99952 -> 99544 (-0.41%); split: -0.44%, +0.03%
VGPRs: 60880 -> 60272 (-1.00%); split: -1.02%, +0.02%
CodeSize: 5138488 -> 5107500 (-0.60%); split: -0.61%, +0.01%
MaxWaves: 32193 -> 32380 (+0.58%)
Instrs: 983178 -> 975684 (-0.76%); split: -0.77%, +0.01%

fossil-db (GFX10.3):
Totals from 1943 (1.39% of 139391) affected shaders:
SGPRs: 99832 -> 99648 (-0.18%); split: -0.25%, +0.06%
VGPRs: 64708 -> 63944 (-1.18%); split: -1.27%, +0.09%
CodeSize: 5196732 -> 5157632 (-0.75%); split: -0.76%, +0.00%
MaxWaves: 27478 -> 27486 (+0.03%); split: +0.06%, -0.03%
Instrs: 1007222 -> 998737 (-0.84%); split: -0.84%, +0.00%

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5523>
 2021-01-26 11:36:13 +00:00
Rhys Perry
0c3d8e8e2e aco: disable a*1.0 optimization if the instruction is precise 
fossil-db (GFX10):
Totals from 10370 (7.44% of 139391) affected shaders:
SGPRs: 564072 -> 564016 (-0.01%); split: -0.01%, +0.00%
VGPRs: 248312 -> 248532 (+0.09%); split: -0.02%, +0.11%
CodeSize: 12866732 -> 13208904 (+2.66%); split: -0.00%, +2.66%
MaxWaves: 190198 -> 190170 (-0.01%)
Instrs: 2460818 -> 2545351 (+3.44%)

fossil-db (GFX10.3):
Totals from 10370 (7.44% of 139391) affected shaders:
SGPRs: 563904 -> 564272 (+0.07%); split: -0.16%, +0.22%
VGPRs: 289344 -> 295016 (+1.96%); split: -0.88%, +2.84%
CodeSize: 13519204 -> 14197020 (+5.01%); split: -0.00%, +5.01%
MaxWaves: 155946 -> 154566 (-0.88%)
Instrs: 2719177 -> 2806919 (+3.23%); split: -0.00%, +3.23%

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5523>
 2021-01-26 11:36:13 +00:00
Samuel Pitoiset
bb8f87088c radv,aco: fix shifting input VGPRs for the LS VGPR init bug on GFX9 
We were incorrectly shifting the input VGPRs for the instance ID
for chips affected by the LS VGPR init bug (ie. Vega10 and Raven).

When there is no HS threads, the hardware loads the LS VGPR
starting from VGPR 0, so they should be shifted by two.

This fixes some sort of vertex explosion with Squad, Visage, Barn
Finders and probably more titles that use tessellation. Note that
only Vega10 and Raven were affected by this bug.

Cc: mesa-stable
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/4129
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/3311
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Tested-by: Diego Viola <diego.viola@gmail.com>
Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/8694>
 2021-01-25 17:10:44 +00:00
Rhys Perry
f03c20ffae aco: fix WQM for texture instructions with args before the coordinates 
Previously, we might not have required all coordinates to be in WQM if
there were other args before them. We should probably also require that
the offset is in WQM.

fossil-db (GFX10.3):
Totals from 10053 (7.21% of 139391) affected shaders:
SGPRs: 911032 -> 911048 (+0.00%); split: -0.00%, +0.00%
VGPRs: 689856 -> 688412 (-0.21%); split: -0.26%, +0.05%
CodeSize: 84151460 -> 84140396 (-0.01%); split: -0.02%, +0.01%
MaxWaves: 77526 -> 77527 (+0.00%)
Instrs: 15972106 -> 15971521 (-0.00%); split: -0.01%, +0.01%

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/4153
Fixes: 4015b3651a ("aco: only require texture coordinates to be in WQM if NSA is used")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/8693>
 2021-01-25 16:31:39 +00:00
Rhys Perry
e115b01948 aco: return references in instruction cast methods 
Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/8595>
 2021-01-22 14:12:33 +00:00
Rhys Perry
1d245cd18b aco: use format-check methods 
Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/8595>
 2021-01-22 14:12:32 +00:00
Rhys Perry
70dbcfa1c9 aco: use instruction cast methods 
Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/8595>
 2021-01-22 14:12:32 +00:00
Rhys Perry
fb12302b8e aco: add instruction cast and format-check methods 
Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/8595>
 2021-01-22 14:12:32 +00:00
Rhys Perry
441ead5fb3 aco: remove Format::{VOP3A,VOP3B} 
These are really the same as Format::VOP3.

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/8595>
 2021-01-22 14:12:32 +00:00
Rhys Perry
824eba2148 aco: don't consider a phi trivial if same's register doesn't match the def 
For example:
 s2: %688:s[32-33] = p_linear_phi %3:s[10-11], %688:s[32-33]
would have been considered trivial.

This might happen due to parallelcopies when assigning phi registers.

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Fixes: 69b6069dd2 ("aco: refactor try_remove_trivial_phi() in RA")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/8645>
 2021-01-22 12:42:47 +00:00
Rhys Perry
dc19fe0e9f radv,aco: use deref_buffer_array_length 
Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Gitlab: https://gitlab.freedesktop.org/mesa/mesa/-/issues/3993
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/8163>
 2021-01-21 11:53:12 +00:00
Daniel Schürmann
7dcb9a0d8c aco/optimizer: convert extract_vector with index 0 into parallelcopies if possible 
Totals from 273 (0.20% of 139391) affected shaders (Navi10):
VGPRs: 11600 -> 11792 (+1.66%)
CodeSize: 1389304 -> 1383152 (-0.44%); split: -0.53%, +0.08%
MaxWaves: 3848 -> 3752 (-2.49%)
Instrs: 240228 -> 239478 (-0.31%); split: -0.37%, +0.06%
Cycles: 20637708 -> 20580024 (-0.28%); split: -0.46%, +0.18%
VMEM: 39164 -> 38831 (-0.85%); split: +0.06%, -0.91%
SMEM: 21743 -> 22204 (+2.12%)
VClause: 4787 -> 4783 (-0.08%)
Copies: 39057 -> 38308 (-1.92%); split: -2.28%, +0.37%
Branches: 6556 -> 6557 (+0.02%)

Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/8260>
 2021-01-21 11:05:36 +00:00
Daniel Schürmann
ebbf5fe716 aco/optimizer: expand subdword vectors with SGPRs on all generations 
No fossildb changes.

Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/8260>
 2021-01-21 11:05:36 +00:00
Daniel Schürmann
96fafcca63 aco: propagate temporaries into PSEUDO instructions if it can take it 
This patch relaxes copy-propagation for PSEUDO instructions with
subdword Operands / Definitions:
general:
- only propagate VGPR temps if the Definition is VGPR (or on p_as_uniform)

parallelcopy/create_vector/phis:
- size has to be the same

extract_vector/split_vector:
- propagate SGPR temps on GFX9+ or if the Definitions are not subdword
- split_vector: size must not increase

Totals from 282 (0.20% of 140985) affected shaders (Polaris10):
VGPRs: 14520 -> 14408 (-0.77%)
CodeSize: 2693956 -> 2694316 (+0.01%); split: -0.20%, +0.21%
Instrs: 512874 -> 512864 (-0.00%); split: -0.16%, +0.16%
Cycles: 26338860 -> 26320652 (-0.07%); split: -0.36%, +0.29%
VMEM: 49460 -> 49634 (+0.35%); split: +0.47%, -0.12%
SMEM: 10035 -> 10036 (+0.01%)
VClause: 7675 -> 7674 (-0.01%)
Copies: 66012 -> 65943 (-0.10%); split: -1.31%, +1.20%
Branches: 17265 -> 17281 (+0.09%); split: -0.10%, +0.19%
PreVGPRs: 12211 -> 12124 (-0.71%)

Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/8260>
 2021-01-21 11:05:36 +00:00
Daniel Schürmann
21a7bea342 aco/validate: relax subdword restrictions 
This affects constants/SGPRs on GFX6-8 and
the operand regClass of SDWA instructions.

Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/8260>
 2021-01-21 11:05:36 +00:00
Daniel Schürmann
77c9629046 aco/validate: ensure that Operand and Definition size matches for parallelcopies 
Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/8260>
 2021-01-21 11:05:36 +00:00
Daniel Schürmann
8fb66187ec aco/validate: validate that p_create_vector operands are aligned unless they are subdword operands 
Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/8260>
 2021-01-21 11:05:36 +00:00
Daniel Schürmann
c0cec3a29b aco: generalize subdword constant copy lowering 
This will allow to propagate and emit sub-register constants
on all hardware generations.

Also fixes GFX8 constant emission to not use SDWA.

Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/8260>
 2021-01-21 11:05:36 +00:00
Daniel Schürmann
856fd4750d aco/optimizer: don't propagate subdword temps of different size 
It could happen that due to inconsistent copy-propagation

  v1 = p_parallelcopy v2b

instructions were left after optimization on GFX8.

Cc: 20.3
Cc: 21.0

Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/8260>
 2021-01-21 11:05:36 +00:00
Daniel Schürmann
cd870d1b6a aco/optimizer: don't copy-prop logical phis 
This is dangerous w.r.t. LCSSA-phis.

Totals from 746 (0.54% of 139391) affected shaders (Navi10):
CodeSize: 8592160 -> 8568156 (-0.28%); split: -0.30%, +0.02%
MaxWaves: 5172 -> 5171 (-0.02%); split: +0.02%, -0.04%
Instrs: 1653949 -> 1648489 (-0.33%); split: -0.36%, +0.03%
Cycles: 49474892 -> 49329224 (-0.29%); split: -0.33%, +0.03%
VMEM: 137574 -> 137421 (-0.11%); split: +0.18%, -0.29%
SMEM: 42391 -> 42439 (+0.11%); split: +0.12%, -0.01%
VClause: 26946 -> 26943 (-0.01%)
Copies: 130902 -> 126176 (-3.61%); split: -4.05%, +0.43%
Branches: 54891 -> 54556 (-0.61%); split: -0.64%, +0.03%
PreVGPRs: 53941 -> 53939 (-0.00%)

This has a slight effect on RA due to affinity changes.

Cc: 20.3
Cc: 21.0

Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/8260>
 2021-01-21 11:05:36 +00:00
Rhys Perry
914c61d6c0 radv,aco: don't use MUBUF for multi-channel loads on GFX8 with robustness2 
Fixes several dEQP-VK.robustness.robustness2.* tests on GFX8. Generations
other than GFX8 don't fail the tests because bounds-checking is done using
the index (making it per-vertex).

fossil-db (Polaris):
Totals from 1387 (0.99% of 140385) affected shaders:
(no statistics affected)

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Fixes: 03a0d39366 ("aco: use MUBUF in some situations instead of splitting vertex fetches")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7834>
 2021-01-20 17:57:56 +00:00
Rhys Perry
af9977a3d5 aco: add affinity for non-sequential MIMG operands 
fossil-db (GFX10.3):
Totals from 42008 (30.14% of 139391) affected shaders:
VGPRs: 2139116 -> 2147696 (+0.40%); split: -0.06%, +0.46%
CodeSize: 199109120 -> 198637852 (-0.24%); split: -0.24%, +0.01%
Instrs: 37713901 -> 37714574 (+0.00%); split: -0.02%, +0.03%
Cycles: 1621911328 -> 1621634168 (-0.02%); split: -0.02%, +0.01%

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/8523>
 2021-01-20 16:46:54 +00:00