Commit graph

18 commits

Author SHA1 Message Date
Caio Oliveira
ff44f4d278 intel/brw: Update outdated comments
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/32536>
2025-02-11 09:13:28 +00:00
Caio Oliveira
cf3bb77224 intel/brw: Rename fs_visitor to brw_shader
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/32536>
2025-02-11 09:13:28 +00:00
Caio Oliveira
352a63122f intel/brw: Rename files brw_fs.cpp/h to brw_shader.cpp/h
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/32536>
2025-02-11 09:13:28 +00:00
Caio Oliveira
f82bcd56fc intel/brw: Add functions to allocate VGRF space
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/33334>
2025-02-06 08:33:03 -08:00
Caio Oliveira
ea87bab4ce intel/brw: Remove 'using namespace brw' directives
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/33418>
2025-02-06 07:58:55 -08:00
Caio Oliveira
1ade9a05d8 intel/brw: Use brw prefix instead of namespace for analysis implementations
Also drop the 'fs' prefix when applicable.

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/33048>
2025-02-05 21:47:07 +00:00
Caio Oliveira
2b92eb0b2c intel/brw: Use brw prefix instead of namespace for dep analysis enum
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/33048>
2025-02-05 21:47:07 +00:00
Caio Oliveira
d59bd421a2 intel/brw: Rename fs_inst to brw_inst
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/33114>
2025-01-31 00:57:21 +00:00
Caio Oliveira
f18dee3618 intel/brw: Fallback to SEND from SEND_GATHER if possible
After optimization happen, if the sources are still in one or two
contigous spans for some reason (e.g. some data read from memory
now being written), it is beneficial to just use regular SEND
and avoid having to set the ARF scalar instruction.

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Lionel Landwerlin <None>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/32410>
2025-01-30 04:43:58 +00:00
Caio Oliveira
b6b32933ad intel/brw: Use SHADER_OPCODE_SEND_GATHER in Xe3
Add an optimization pass to turn regular SENDs into SEND_GATHERs.
This allows the payload to be "broken" into smaller pieces that
can be further optimized, which _may_ result in

- less register pressure (no need to contiguous space), and
- less instructions (no need to MOV to such space).

For debugging, the INTEL_DEBUG=no-send-gather option skips this
optimization, and reporting how many opportunities were missed.

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Lionel Landwerlin <None>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/32410>
2025-01-30 04:43:58 +00:00
Caio Oliveira
5ac82efd35 intel/brw: Rename fs_builder to brw_builder
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/33076>
2025-01-18 16:12:55 +00:00
Caio Oliveira
f2d4c9db92 intel/brw: Rename brw_fs_builder.h to brw_builder.h
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/33076>
2025-01-18 16:12:54 +00:00
Caio Oliveira
634daf2827 intel/brw: Rename brw_fs_validate to brw_validate
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/32843>
2025-01-13 23:56:22 +00:00
Lionel Landwerlin
8ac7802ac8 brw: move final send lowering up into the IR
Because we do emit the final send message form in code generation, a
lot of emissions look like this :

  add(8)  vgrf0,    u0, 0x100
  mov(1)   a0.1, vgrf0          # emitted by the generator
  send(8)   ...,  a0.1

By moving address register manipulation in the IR, we can get this
down to :

  add(1)  a0.1,   u0, 0x100
  send(8)  ..., a0.1

This reduce register pressure around some send messages by 1 vgrf.

All lost shaders in the below results are fragment SIMD32, due to the
throughput estimator. If turned off, we loose no SIMD32 shaders with
this change.

DG2 results:

  Assassin's Creed Valhalla:
  Totals from 2044 (96.87% of 2110) affected shaders:
  Instrs: 852879 -> 832044 (-2.44%); split: -2.45%, +0.00%
  Subgroup size: 23832 -> 23824 (-0.03%)
  Cycle count: 53345742 -> 52144277 (-2.25%); split: -5.08%, +2.82%
  Spill count: 729 -> 554 (-24.01%); split: -28.40%, +4.39%
  Fill count: 2005 -> 1256 (-37.36%)
  Scratch Memory Size: 25600 -> 19456 (-24.00%); split: -32.00%, +8.00%
  Max live registers: 116765 -> 115058 (-1.46%)
  Max dispatch width: 19152 -> 18872 (-1.46%); split: +0.21%, -1.67%

  Cyberpunk 2077:
  Totals from 1181 (93.43% of 1264) affected shaders:
  Instrs: 667192 -> 663615 (-0.54%); split: -0.55%, +0.01%
  Subgroup size: 13016 -> 13032 (+0.12%)
  Cycle count: 17383539 -> 17986073 (+3.47%); split: -0.93%, +4.39%
  Spill count: 12 -> 8 (-33.33%)
  Fill count: 9 -> 6 (-33.33%)

  Dota2:
  Totals from 173 (11.59% of 1493) affected shaders:
  Cycle count: 274403 -> 280817 (+2.34%); split: -0.01%, +2.34%
  Max live registers: 5787 -> 5779 (-0.14%)
  Max dispatch width: 1344 -> 1152 (-14.29%)

  Hitman3:
  Totals from 5072 (95.39% of 5317) affected shaders:
  Instrs: 2879952 -> 2841804 (-1.32%); split: -1.32%, +0.00%
  Cycle count: 153208505 -> 165860401 (+8.26%); split: -2.22%, +10.48%
  Spill count: 3942 -> 3200 (-18.82%)
  Fill count: 10158 -> 8846 (-12.92%)
  Scratch Memory Size: 257024 -> 223232 (-13.15%)
  Max live registers: 328467 -> 324631 (-1.17%)
  Max dispatch width: 43928 -> 42768 (-2.64%); split: +0.09%, -2.73%

  Fortnite:
  Totals from 360 (4.82% of 7472) affected shaders:
  Instrs: 778068 -> 777925 (-0.02%)
  Subgroup size: 3128 -> 3136 (+0.26%)
  Cycle count: 38684183 -> 38734579 (+0.13%); split: -0.06%, +0.19%
  Max live registers: 50689 -> 50658 (-0.06%)

  Hogwarts Legacy:
  Totals from 1376 (84.00% of 1638) affected shaders:
  Instrs: 758810 -> 749727 (-1.20%); split: -1.23%, +0.03%
  Cycle count: 27778983 -> 28805469 (+3.70%); split: -1.42%, +5.12%
  Spill count: 2475 -> 2299 (-7.11%); split: -7.47%, +0.36%
  Fill count: 2677 -> 2445 (-8.67%); split: -9.90%, +1.23%
  Scratch Memory Size: 99328 -> 89088 (-10.31%)
  Max live registers: 84969 -> 84671 (-0.35%); split: -0.58%, +0.23%
  Max dispatch width: 11848 -> 11920 (+0.61%)

  Metro Exodus:
  Totals from 92 (0.21% of 43072) affected shaders:
  Instrs: 262995 -> 262968 (-0.01%)
  Cycle count: 13818007 -> 13851266 (+0.24%); split: -0.01%, +0.25%
  Max live registers: 11152 -> 11140 (-0.11%)

  Red Dead Redemption 2 :
  Totals from 451 (7.71% of 5847) affected shaders:
  Instrs: 754178 -> 753811 (-0.05%); split: -0.05%, +0.00%
  Cycle count: 3484078523 -> 3484111965 (+0.00%); split: -0.00%, +0.00%
  Max live registers: 42294 -> 42185 (-0.26%)

  Spiderman Remastered:
  Totals from 6820 (98.02% of 6958) affected shaders:
  Instrs: 6921500 -> 6747933 (-2.51%); split: -4.16%, +1.65%
  Cycle count: 234400692460 -> 236846720707 (+1.04%); split: -0.20%, +1.25%
  Spill count: 72971 -> 72622 (-0.48%); split: -8.08%, +7.61%
  Fill count: 212921 -> 198483 (-6.78%); split: -12.37%, +5.58%
  Scratch Memory Size: 3491840 -> 3410944 (-2.32%); split: -12.05%, +9.74%
  Max live registers: 493149 -> 487458 (-1.15%)
  Max dispatch width: 56936 -> 56856 (-0.14%); split: +0.06%, -0.20%

  Strange Brigade:
  Totals from 3769 (91.21% of 4132) affected shaders:
  Instrs: 1354476 -> 1321474 (-2.44%)
  Cycle count: 25351530 -> 25339190 (-0.05%); split: -1.64%, +1.59%
  Max live registers: 199057 -> 193656 (-2.71%)
  Max dispatch width: 30272 -> 30240 (-0.11%)

  Witcher 3:
  Totals from 25 (2.40% of 1041) affected shaders:
  Instrs: 24621 -> 24606 (-0.06%)
  Cycle count: 2218793 -> 2217503 (-0.06%); split: -0.11%, +0.05%
  Max live registers: 1963 -> 1955 (-0.41%)

LNL results:

  Assassin's Creed Valhalla:
  Totals from 1928 (98.02% of 1967) affected shaders:
  Instrs: 856107 -> 835756 (-2.38%); split: -2.48%, +0.11%
  Subgroup size: 41264 -> 41280 (+0.04%)
  Cycle count: 64606590 -> 62371700 (-3.46%); split: -5.57%, +2.11%
  Spill count: 915 -> 669 (-26.89%); split: -32.79%, +5.90%
  Fill count: 2414 -> 1617 (-33.02%); split: -36.62%, +3.60%
  Scratch Memory Size: 62464 -> 44032 (-29.51%); split: -36.07%, +6.56%
  Max live registers: 205483 -> 202192 (-1.60%)

  Cyberpunk 2077:
  Totals from 1177 (96.40% of 1221) affected shaders:
  Instrs: 682237 -> 678931 (-0.48%); split: -0.51%, +0.03%
  Subgroup size: 24912 -> 24944 (+0.13%)
  Cycle count: 24355928 -> 25089292 (+3.01%); split: -0.80%, +3.81%
  Spill count: 8 -> 3 (-62.50%)
  Fill count: 6 -> 3 (-50.00%)
  Max live registers: 126922 -> 125472 (-1.14%)

  Dota2:
  Totals from 428 (32.47% of 1318) affected shaders:
  Instrs: 89355 -> 89740 (+0.43%)
  Cycle count: 1152412 -> 1152706 (+0.03%); split: -0.52%, +0.55%
  Max live registers: 32863 -> 32847 (-0.05%)

  Fortnite:
  Totals from 5354 (81.72% of 6552) affected shaders:
  Instrs: 4135059 -> 4239015 (+2.51%); split: -0.01%, +2.53%
  Cycle count: 132557506 -> 132427302 (-0.10%); split: -0.75%, +0.65%
  Spill count: 7144 -> 7234 (+1.26%); split: -0.46%, +1.72%
  Fill count: 12086 -> 12403 (+2.62%); split: -0.73%, +3.35%
  Scratch Memory Size: 600064 -> 604160 (+0.68%); split: -1.02%, +1.71%

  Hitman3:
  Totals from 4912 (97.09% of 5059) affected shaders:
  Instrs: 2952124 -> 2916824 (-1.20%); split: -1.20%, +0.00%
  Cycle count: 179985656 -> 189175250 (+5.11%); split: -2.44%, +7.55%
  Spill count: 3739 -> 3136 (-16.13%)
  Fill count: 10657 -> 9564 (-10.26%)
  Scratch Memory Size: 373760 -> 318464 (-14.79%)
  Max live registers: 597566 -> 589460 (-1.36%)

  Hogwarts Legacy:
  Totals from 1471 (96.33% of 1527) affected shaders:
  Instrs: 748749 -> 766214 (+2.33%); split: -0.71%, +3.05%
  Cycle count: 33301528 -> 34426308 (+3.38%); split: -1.30%, +4.68%
  Spill count: 3278 -> 3070 (-6.35%); split: -8.30%, +1.95%
  Fill count: 4553 -> 4097 (-10.02%); split: -10.85%, +0.83%
  Scratch Memory Size: 251904 -> 217088 (-13.82%)
  Max live registers: 168911 -> 168106 (-0.48%); split: -0.59%, +0.12%

  Metro Exodus:
  Totals from 18356 (49.81% of 36854) affected shaders:
  Instrs: 7559386 -> 7621591 (+0.82%); split: -0.01%, +0.83%
  Cycle count: 195240612 -> 196455186 (+0.62%); split: -1.22%, +1.84%
  Spill count: 595 -> 546 (-8.24%)
  Fill count: 1604 -> 1408 (-12.22%)
  Max live registers: 2086937 -> 2086933 (-0.00%)

  Red Dead Redemption 2:
  Totals from 4171 (79.31% of 5259) affected shaders:
  Instrs: 2619392 -> 2719587 (+3.83%); split: -0.00%, +3.83%
  Subgroup size: 86416 -> 86432 (+0.02%)
  Cycle count: 8542836160 -> 8531976886 (-0.13%); split: -0.65%, +0.53%
  Fill count: 12949 -> 12970 (+0.16%); split: -0.43%, +0.59%
  Scratch Memory Size: 401408 -> 385024 (-4.08%)

  Spiderman Remastered:
  Totals from 6639 (98.94% of 6710) affected shaders:
  Instrs: 6877980 -> 6800592 (-1.13%); split: -3.11%, +1.98%
  Cycle count: 282183352210 -> 282100051824 (-0.03%); split: -0.62%, +0.59%
  Spill count: 63147 -> 64218 (+1.70%); split: -7.12%, +8.82%
  Fill count: 184931 -> 175591 (-5.05%); split: -10.81%, +5.76%
  Scratch Memory Size: 5318656 -> 5970944 (+12.26%); split: -5.91%, +18.17%
  Max live registers: 918240 -> 906604 (-1.27%)

  Strange Brigade:
  Totals from 3675 (92.24% of 3984) affected shaders:
  Instrs: 1462231 -> 1429345 (-2.25%); split: -2.25%, +0.00%
  Cycle count: 37404050 -> 37345292 (-0.16%); split: -1.25%, +1.09%
  Max live registers: 361849 -> 351265 (-2.92%)

  Witcher 3:
  Totals from 13 (46.43% of 28) affected shaders:
  Instrs: 593 -> 660 (+11.30%)
  Cycle count: 28302 -> 28714 (+1.46%)

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Reviewed-by: Caio Oliveira <caio.oliveira@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/28199>
2025-01-11 08:41:42 +00:00
Kenneth Graunke
4ab04799ee brw: Delete assign_constant_locations and push_constant_loc[]
The push_constant_loc[] array is always an identity mapping these days,
so it's kind of pointless.  Just use the original uniform number and
skip the unnecessary "remap" step.  With that gone, and shrinking UBO
ranges gone, assign_constant_locations() is now empty and can be removed
as well.

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/32841>
2025-01-06 12:45:47 +00:00
Kenneth Graunke
93e186e1a4 brw: Delete pull constant lowering
Now that we never shrink ranges in the backend, we never lower push
constants to pull constants late in the backend either.  get_pull_loc
will never return true, and so all of brw_lower_constant_loads becomes
a noop.

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/32841>
2025-01-06 12:45:47 +00:00
Caio Oliveira
e1aebf8a0c intel/brw: Remove 'fs' prefix from passes and related functions
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/32813>
2025-01-02 18:11:05 +00:00
Caio Oliveira
25384dccc0 intel/brw: Remove 'fs' prefix from passes filenames
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/32813>
2025-01-02 18:11:05 +00:00
Renamed from src/intel/compiler/brw_fs_opt.cpp (Browse further)