fdo-mirrors/mesa
1
0
Fork 0
mirror of https://gitlab.freedesktop.org/mesa/mesa.git synced 2026-05-24 08:28:16 +02:00
Code Issues Projects Releases Packages Wiki Activity Actions 1
219537 commits 66 branches 936 tags 986 MiB
Commit graph

744 commits

Author SHA1 Message Date
Georg Lehmann
ab773fc5d4 nir/opt_algebraic: fix frsq clamp pattern 
This is not NaN correct.
And also make the pattern 32bit only because the constant is hard coded
FLT_MAX.

Fixes: 780b5c1037 ("nir/algebraic: Simplify some Inf and NaN avoidance code")
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/39987>
 2026-03-07 05:01:42 +00:00
Georg Lehmann
ba30de1f97 nir/opt_algebraic: remove pattern that skips iabs with range analysis 
Fixes: f2a59fdea6 ("nir: remove non float nir_analyse_range support")
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/39987>
 2026-03-07 05:01:41 +00:00
Caio Oliveira
da57fbfb07 nir: Fix constant folding for iadd_sat 
Use INT_MIN instead of INT_MAX for underflow.

Fixes: cc4b50b023 ("nir/opcodes: use u_overflow to fix incorrect checks")
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Pierre-Eric Pelloux-Prayer <pelloux@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/40252>
 2026-03-06 22:26:07 +00:00
Georg Lehmann
6a218e346d nir: remove lower_vector_cmp 
Use nir_lower_alu_width or nir_lower_alu_to_scalar instead.

Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/40197>
 2026-03-04 19:50:28 +00:00
Georg Lehmann
3e6e1e213c nir: remove fall_equal/fany_nequal opcodes 
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/40197>
 2026-03-04 19:50:27 +00:00
Georg Lehmann
7194dfcc2c nir/opt_algebraic: optimize b2i(a) * b to bcsel 
Foz-DB Navi48:
Totals from 3180 (2.77% of 114655) affected shaders:
MaxWaves: 85526 -> 85446 (-0.09%)
Instrs: 2681446 -> 2678641 (-0.10%); split: -0.17%, +0.07%
CodeSize: 14295536 -> 14284628 (-0.08%); split: -0.13%, +0.05%
VGPRs: 174792 -> 174636 (-0.09%); split: -0.16%, +0.07%
SpillSGPRs: 306 -> 308 (+0.65%)
Latency: 14078973 -> 14070122 (-0.06%); split: -0.07%, +0.01%
InvThroughput: 2774242 -> 2764051 (-0.37%); split: -0.37%, +0.00%
VClause: 41744 -> 41734 (-0.02%); split: -0.10%, +0.07%
SClause: 58176 -> 58154 (-0.04%); split: -0.05%, +0.01%
Copies: 222967 -> 223108 (+0.06%); split: -0.14%, +0.20%
Branches: 57317 -> 57322 (+0.01%)
PreSGPRs: 140454 -> 140451 (-0.00%); split: -0.01%, +0.00%
PreVGPRs: 131649 -> 131540 (-0.08%); split: -0.09%, +0.01%
VALU: 1509318 -> 1505443 (-0.26%); split: -0.26%, +0.00%
SALU: 384419 -> 385838 (+0.37%); split: -0.01%, +0.38%
VOPD: 13272 -> 13286 (+0.11%); split: +0.14%, -0.03%

Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/40160>
 2026-03-02 15:58:30 +00:00
Georg Lehmann
3d304d5647 nir/opt_algebraic: remove is_used_once on outer instruction 
This just prevents useful optimizations.
is_used_once only makes sense on inner instructions, to prevent
creating more new instructions than will be removed.

Foz-DB Navi48:
Totals from 16989 (14.82% of 114655) affected shaders:
MaxWaves: 434379 -> 434353 (-0.01%); split: +0.01%, -0.01%
Instrs: 29030794 -> 29022514 (-0.03%); split: -0.07%, +0.04%
CodeSize: 155293092 -> 155262816 (-0.02%); split: -0.05%, +0.03%
VGPRs: 1093980 -> 1094088 (+0.01%); split: -0.01%, +0.02%
SpillSGPRs: 9801 -> 9803 (+0.02%); split: -0.03%, +0.05%
Latency: 356327270 -> 356283384 (-0.01%); split: -0.03%, +0.02%
InvThroughput: 58239439 -> 58229374 (-0.02%); split: -0.03%, +0.01%
VClause: 451716 -> 451815 (+0.02%); split: -0.07%, +0.09%
SClause: 654614 -> 654556 (-0.01%); split: -0.03%, +0.03%
Copies: 1809805 -> 1809297 (-0.03%); split: -0.20%, +0.17%
Branches: 552382 -> 552384 (+0.00%); split: -0.00%, +0.00%
PreSGPRs: 947188 -> 947224 (+0.00%); split: -0.01%, +0.02%
PreVGPRs: 879583 -> 880173 (+0.07%); split: -0.01%, +0.08%
VALU: 16317859 -> 16309975 (-0.05%); split: -0.07%, +0.02%
SALU: 4256121 -> 4259315 (+0.08%); split: -0.05%, +0.12%
SMEM: 1067069 -> 1067070 (+0.00%)
VOPD: 440855 -> 440792 (-0.01%); split: +0.05%, -0.07%

Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/40138>
 2026-03-02 15:24:36 +00:00
Georg Lehmann
41878e5714 nir_opt_algebraic: remove unneeded is_not_const 
These were needed when we didn't constant fold inside nir_search,
to prevent infinite loops.
But now all they do is slow down pattern matching.

Foz-DB Navi48:
Totals from 107 (0.09% of 114655) affected shaders:
Instrs: 162439 -> 162481 (+0.03%); split: -0.01%, +0.03%
CodeSize: 943056 -> 942988 (-0.01%); split: -0.03%, +0.02%
Latency: 971667 -> 970865 (-0.08%); split: -0.09%, +0.00%
InvThroughput: 164452 -> 164521 (+0.04%); split: -0.02%, +0.07%
Copies: 7980 -> 7982 (+0.03%)
VALU: 103572 -> 103566 (-0.01%); split: -0.05%, +0.04%
SALU: 12825 -> 12878 (+0.41%)
VOPD: 5235 -> 5190 (-0.86%)

Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/40138>
 2026-03-02 15:24:36 +00:00
Georg Lehmann
374cbc17a4 nir_opt_algebraic: reassociate fadd into ffma where one factor is a constant 
This restriction doesn't really make sense, probably an accident.

Foz-DB Navi48:
Totals from 2290 (2.00% of 114655) affected shaders:
MaxWaves: 57496 -> 57510 (+0.02%); split: +0.06%, -0.03%
Instrs: 2817419 -> 2816209 (-0.04%); split: -0.12%, +0.08%
CodeSize: 15218816 -> 15220576 (+0.01%); split: -0.09%, +0.10%
VGPRs: 147456 -> 147384 (-0.05%); split: -0.07%, +0.02%
Latency: 13757114 -> 13751833 (-0.04%); split: -0.13%, +0.09%
InvThroughput: 2463343 -> 2462482 (-0.03%); split: -0.07%, +0.04%
VClause: 40137 -> 40153 (+0.04%); split: -0.07%, +0.11%
SClause: 57351 -> 57385 (+0.06%); split: -0.12%, +0.18%
Copies: 135482 -> 136258 (+0.57%); split: -0.22%, +0.79%
Branches: 30886 -> 30894 (+0.03%)
PreSGPRs: 113470 -> 113462 (-0.01%); split: -0.03%, +0.02%
PreVGPRs: 117554 -> 117591 (+0.03%); split: -0.01%, +0.04%
VALU: 1682734 -> 1681557 (-0.07%); split: -0.10%, +0.03%
SALU: 390685 -> 391301 (+0.16%); split: -0.07%, +0.22%
VOPD: 6159 -> 6254 (+1.54%); split: +1.72%, -0.18%

Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/40138>
 2026-03-02 15:24:36 +00:00
Georg Lehmann
b949122908 nir/opt_algebraic: remove loops for b2f/b2i equality handling 
The feq/fneu patterns already existed, and there is no reason to use bit size based
loops here.

Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/40138>
 2026-03-02 15:24:36 +00:00
Georg Lehmann
83091276f8 nir_opt_algebraic: remove more specific cmp+bcsel opts 
Only some minimal difference from pattern ordering:

Foz-DB Navi48:
Totals from 3 (0.00% of 114655) affected shaders:
Instrs: 4556 -> 4533 (-0.50%)
CodeSize: 23716 -> 23608 (-0.46%)
Latency: 27424 -> 26336 (-3.97%)
InvThroughput: 4674 -> 4672 (-0.04%)
SClause: 107 -> 105 (-1.87%)
Copies: 351 -> 346 (-1.42%)
Branches: 130 -> 126 (-3.08%)
VALU: 2598 -> 2595 (-0.12%)
SALU: 561 -> 555 (-1.07%)
SMEM: 169 -> 167 (-1.18%)

Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/40138>
 2026-03-02 15:24:36 +00:00
Georg Lehmann
4190241795 nir/opt_algebraic: optimize all comparisons of b2f/b2i with constants 
Foz-DB Navi48:
Totals from 857 (0.75% of 114655) affected shaders:
Instrs: 1136993 -> 1132422 (-0.40%); split: -0.48%, +0.08%
CodeSize: 6096636 -> 6070832 (-0.42%); split: -0.48%, +0.06%
VGPRs: 49668 -> 49620 (-0.10%)
Latency: 24014661 -> 24044601 (+0.12%); split: -0.04%, +0.16%
InvThroughput: 4182482 -> 4183708 (+0.03%); split: -0.12%, +0.15%
VClause: 17698 -> 17695 (-0.02%)
SClause: 25214 -> 25213 (-0.00%)
Copies: 81474 -> 81396 (-0.10%); split: -0.79%, +0.69%
Branches: 24722 -> 24650 (-0.29%); split: -0.36%, +0.07%
PreSGPRs: 43338 -> 43291 (-0.11%); split: -0.22%, +0.11%
VALU: 652975 -> 649760 (-0.49%); split: -0.50%, +0.00%
SALU: 153961 -> 153797 (-0.11%); split: -0.72%, +0.61%
VOPD: 10650 -> 10684 (+0.32%); split: +0.38%, -0.07%

Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/40138>
 2026-03-02 15:24:36 +00:00
Georg Lehmann
ef6f5377da nir/opt_algebraic: remove fcmp+fneg patterns that are cleaned up earlier 
No Foz-DB changes, as expected.

Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/40138>
 2026-03-02 15:24:36 +00:00
Georg Lehmann
a5334ec239 nir/opt_algebraic: generalize late fcmp(fneg(a), const) patterns 
No reason just to do this for 1.0.

Foz-DB Navi48:
Totals from 44 (0.04% of 114655) affected shaders:
CodeSize: 111620 -> 111476 (-0.13%)

Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/40138>
 2026-03-02 15:24:35 +00:00
Georg Lehmann
6b464785b9 nir/opt_algebraic: optimize d3d9 iand(a, inot(b)) 
Foz-DB GFX1201:
Totals from 24 (0.02% of 112525) affected shaders:
Instrs: 15598 -> 15426 (-1.10%); split: -1.17%, +0.06%
CodeSize: 88716 -> 88260 (-0.51%); split: -0.98%, +0.46%
Latency: 54419 -> 53965 (-0.83%); split: -0.91%, +0.08%
InvThroughput: 10294 -> 10166 (-1.24%); split: -1.28%, +0.04%
VClause: 302 -> 300 (-0.66%)
SClause: 367 -> 363 (-1.09%); split: -1.63%, +0.54%
Copies: 712 -> 705 (-0.98%); split: -3.09%, +2.11%
PreSGPRs: 1402 -> 1424 (+1.57%); split: -0.14%, +1.71%
PreVGPRs: 850 -> 848 (-0.24%)
VALU: 9730 -> 9591 (-1.43%)
SALU: 1579 -> 1649 (+4.43%)

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/40104>
 2026-02-26 14:44:01 +00:00
Georg Lehmann
a3f9c347bf nir/opt_algebraic: optimize b2f(a) - 1.0 to -b2f(a) 
Foz-DB GFX1201:
Totals from 81 (0.07% of 112525) affected shaders:
Instrs: 95048 -> 94965 (-0.09%); split: -0.13%, +0.05%
CodeSize: 532148 -> 531864 (-0.05%); split: -0.09%, +0.04%
SpillSGPRs: 122 -> 125 (+2.46%)
Latency: 440372 -> 440402 (+0.01%); split: -0.02%, +0.03%
InvThroughput: 296078 -> 296173 (+0.03%); split: -0.03%, +0.06%
VClause: 1449 -> 1456 (+0.48%); split: -0.21%, +0.69%
SClause: 2249 -> 2256 (+0.31%); split: -0.09%, +0.40%
Copies: 3956 -> 3965 (+0.23%); split: -0.10%, +0.33%
PreVGPRs: 2900 -> 2899 (-0.03%)
VALU: 61212 -> 61098 (-0.19%); split: -0.19%, +0.01%
SALU: 6970 -> 6981 (+0.16%)

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/40104>
 2026-02-26 14:44:01 +00:00
Alyssa Rosenzweig
42c4f7935a nir: optimize u2u32(unpack_32_2x16_split_*) 
Noticed while playing with pixel coord things.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@intel.com>
Reviewed-by: Georg Lehmann <dadschoorse@gmail.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/40056>
 2026-02-24 19:16:56 +00:00
Georg Lehmann
5d5f99bfe8 nir/opt_algebraic: create more b2f if sign of zero doesn't matter 
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/39966>
 2026-02-19 15:21:27 +00:00
Georg Lehmann
d87943ad3d nir/opt_algebraic: preserve signed zero when creating new b2f 
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/39966>
 2026-02-19 15:21:27 +00:00
Rob Clark
8cc99edb7b nir: Fill in missing conversion opts
Some checks are pending
macOS-CI / macOS-CI (dri) (push) Waiting to run
Details
macOS-CI / macOS-CI (xlib) (push) Waiting to run
Details 
I noticed we were missing:

    (('u2f16', ('u2u64', 'a@32')), ('u2f16', a))

This was do to coupling the u2f/i2f opts with i2i/u2u in the same loop
(with different positionals).  The `if B <= S\ncontinue` doesn't apply
to the second part.  So just split these into two loops.

Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/14848
Signed-off-by: Rob Clark <rob.clark@oss.qualcomm.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/39899>
 2026-02-18 15:13:21 +00:00
Rhys Perry
fd22c48b2a nir/algebraic: remove ignore_exact 
This was used because the exact bit meant something different for
comparisons than it did for the replacement expression, but that isn't the
case anymore.

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Georg Lehmann <dadschoorse@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/39809>
 2026-02-18 14:04:22 +00:00
Georg Lehmann
6a662a59b7 nir/opt_algebraic: optimize 1.0 - b2f(a) to b2f(inot(a)) 
Which can then be cleaned up further.

Foz-DB Navi48:
Totals from 4156 (3.62% of 114655) affected shaders:
MaxWaves: 102580 -> 102620 (+0.04%)
Instrs: 11696222 -> 11679986 (-0.14%); split: -0.16%, +0.02%
CodeSize: 64452544 -> 64379204 (-0.11%); split: -0.13%, +0.02%
VGPRs: 288256 -> 288172 (-0.03%)
SpillSGPRs: 7290 -> 7297 (+0.10%)
Latency: 160690992 -> 160643825 (-0.03%); split: -0.05%, +0.02%
InvThroughput: 26869332 -> 26849963 (-0.07%); split: -0.09%, +0.02%
VClause: 237078 -> 237003 (-0.03%); split: -0.04%, +0.01%
SClause: 270560 -> 270564 (+0.00%); split: -0.01%, +0.01%
Copies: 936165 -> 937970 (+0.19%); split: -0.07%, +0.26%
Branches: 302981 -> 302992 (+0.00%); split: -0.00%, +0.00%
PreSGPRs: 244967 -> 245303 (+0.14%)
PreVGPRs: 232930 -> 232886 (-0.02%); split: -0.02%, +0.00%
VALU: 6200283 -> 6187264 (-0.21%); split: -0.23%, +0.02%
SALU: 1759176 -> 1760275 (+0.06%); split: -0.10%, +0.16%
VOPD: 447502 -> 446194 (-0.29%); split: +0.14%, -0.43%

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/39917>
 2026-02-17 10:01:21 +00:00
Georg Lehmann
f7222d6939 nir/opt_algebraic: remove few uses of integer nir_analyze_range 
Surprisingly, this has an effect on GFX1201:

Totals from 66 (0.08% of 82405) affected shaders:
Instrs: 200725 -> 201517 (+0.39%)
CodeSize: 978676 -> 981488 (+0.29%)
Latency: 291736 -> 291760 (+0.01%)
InvThroughput: 31556 -> 31604 (+0.15%)
Copies: 11928 -> 12588 (+5.53%)
Branches: 14850 -> 15048 (+1.33%)
SALU: 68981 -> 69509 (+0.77%)

I say surprisingly, because nir_analyze_range handles nothing but
constants and bcsel for integers. Maybe rdr2 is actually
hitting some weird bcsel(a, #b, #c) == 0 case where b and c are not 0?
No, I looked at a few of those shaders, and it's just noise from changed
instruction order.

Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/39756>
 2026-02-16 18:08:53 +00:00
Georg Lehmann
4e2f1345d8 nir/opt_algebraic: make fcmp(a+b, 0.0) -> fcmp(a, -b) exact using ninf 
And remove some cases that never happen because we remove fneg on compare with constants.

Foz-DB Navi48:
Totals from 1305 (1.58% of 82405) affected shaders:
MaxWaves: 32872 -> 32854 (-0.05%)
Instrs: 4554013 -> 4551638 (-0.05%); split: -0.06%, +0.01%
CodeSize: 25269108 -> 25255428 (-0.05%); split: -0.06%, +0.00%
VGPRs: 87660 -> 87732 (+0.08%)
Latency: 33291152 -> 33285023 (-0.02%); split: -0.03%, +0.01%
InvThroughput: 8965288 -> 8963071 (-0.02%); split: -0.03%, +0.00%
VClause: 104008 -> 103947 (-0.06%); split: -0.09%, +0.03%
SClause: 97577 -> 97574 (-0.00%); split: -0.01%, +0.00%
Copies: 372741 -> 372628 (-0.03%); split: -0.05%, +0.02%
Branches: 134076 -> 134072 (-0.00%)
PreSGPRs: 65109 -> 65110 (+0.00%); split: -0.00%, +0.00%
PreVGPRs: 68911 -> 68968 (+0.08%); split: -0.01%, +0.10%
VALU: 2247091 -> 2245815 (-0.06%); split: -0.07%, +0.01%
SALU: 810190 -> 810001 (-0.02%); split: -0.02%, +0.00%
VOPD: 205075 -> 205016 (-0.03%); split: +0.04%, -0.07%

Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/39641>
 2026-02-10 18:42:03 +00:00
Georg Lehmann
ef7dd040d9 nir/opt_algebraic: make a < 0.0 ? -a : a exact using search helpers 
Foz-DB Navi21:
Totals from 104 (0.13% of 82405) affected shaders:
Instrs: 175964 -> 175514 (-0.26%); split: -0.26%, +0.00%
CodeSize: 909008 -> 908744 (-0.03%); split: -0.05%, +0.02%
Latency: 1515203 -> 1514560 (-0.04%); split: -0.05%, +0.01%
InvThroughput: 308751 -> 308573 (-0.06%); split: -0.06%, +0.00%
Copies: 10318 -> 10315 (-0.03%); split: -0.06%, +0.03%
PreVGPRs: 5767 -> 5755 (-0.21%)
VALU: 108151 -> 107745 (-0.38%)
VOPD: 738 -> 737 (-0.14%)

Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/39641>
 2026-02-10 18:42:03 +00:00
Georg Lehmann
0474ad1504 nir/opt_algebraic: make ffract(is_integral) exact using nnan 
Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/39641>
 2026-02-10 18:42:03 +00:00
Georg Lehmann
b8d1763e0a nir/opt_algebraic: make some more fcmp patterns exact using nnan 
No Foz-DB changes.

Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/39641>
 2026-02-10 18:42:03 +00:00
Georg Lehmann
8d52c59505 nir/opt_algebraic: make some fmin/fmax/fsat patterns exact using nsz/nnan 
Foz-DB Navi48:
Totals from 90 (0.11% of 82405) affected shaders:
Instrs: 52109 -> 52032 (-0.15%); split: -0.16%, +0.01%
CodeSize: 263916 -> 263900 (-0.01%); split: -0.05%, +0.05%
Latency: 504693 -> 504775 (+0.02%); split: -0.01%, +0.03%
InvThroughput: 81444 -> 81157 (-0.35%)
Copies: 2894 -> 2895 (+0.03%)
VALU: 30097 -> 29991 (-0.35%)

Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/39641>
 2026-02-10 18:42:03 +00:00
Georg Lehmann
486ea54184 nir/opt_algebraic: make bcsel(fcmp(b, a), b, a) -> fmin/fmax patterns exact 
These patterns need is_only_used_as_float because fmin/fmax might change NaN
patterns, while bcsel is bit exact. For the same reason, the replacement
must not add undefined results, so make the replacement NaN/inf preserving.

It's impossible to make them signed zero correct (-0.0 == +0.0),
so it's also important that the user alu doesn't care.

Otherwise, the only thing that matters is is whether a is NaN.

Foz-DB Navi48:
Totals from 453 (0.55% of 82405) affected shaders:
MaxWaves: 8242 -> 8270 (+0.34%)
Instrs: 2382059 -> 2380094 (-0.08%); split: -0.09%, +0.00%
CodeSize: 13197208 -> 13179488 (-0.13%); split: -0.14%, +0.00%
VGPRs: 44688 -> 44604 (-0.19%)
Latency: 22839894 -> 22838985 (-0.00%); split: -0.01%, +0.00%
InvThroughput: 4873352 -> 4872924 (-0.01%)
VClause: 50862 -> 50883 (+0.04%); split: -0.02%, +0.06%
SClause: 54000 -> 53993 (-0.01%)
Copies: 250215 -> 250233 (+0.01%); split: -0.00%, +0.01%
PreVGPRs: 39694 -> 39620 (-0.19%)
VALU: 1116881 -> 1116073 (-0.07%); split: -0.07%, +0.00%
SALU: 492799 -> 492139 (-0.13%); split: -0.14%, +0.00%
VOPD: 85457 -> 85461 (+0.00%)

Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/39641>
 2026-02-10 18:42:03 +00:00
Georg Lehmann
f55668bb50 nir/opt_algebraic: update flt -> fneu patterns 
And remove the ones that are redundant because we already move the fneg to
the constant source.

No Foz-DB changes.

Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/39641>
 2026-02-10 18:42:03 +00:00
Georg Lehmann
15b13d5fd4 nir/opt_algebraic: optimize flt/fge(#c, fadd(a, #b)) 
I guess these were missing because the author forgot flt/fge aren't commutative.

Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/39641>
 2026-02-10 18:42:03 +00:00
Georg Lehmann
2355b63cb5 nir/opt_algebraic: use better float control for some fcmp patterns 
Foz-DB Navi48:
Totals from 1084 (1.32% of 82405) affected shaders:
Instrs: 1969973 -> 1968947 (-0.05%); split: -0.08%, +0.02%
CodeSize: 11349704 -> 11344884 (-0.04%); split: -0.06%, +0.02%
VGPRs: 59076 -> 59064 (-0.02%); split: -0.06%, +0.04%
Latency: 20766031 -> 20755032 (-0.05%); split: -0.07%, +0.01%
InvThroughput: 2849402 -> 2846733 (-0.09%); split: -0.10%, +0.01%
VClause: 40736 -> 40740 (+0.01%)
SClause: 91835 -> 91832 (-0.00%)
Copies: 217961 -> 217868 (-0.04%); split: -0.07%, +0.02%
Branches: 60045 -> 60031 (-0.02%)
PreSGPRs: 50639 -> 50618 (-0.04%); split: -0.06%, +0.02%
PreVGPRs: 39593 -> 39590 (-0.01%); split: -0.01%, +0.01%
VALU: 960270 -> 959524 (-0.08%); split: -0.10%, +0.02%
SALU: 326638 -> 326680 (+0.01%); split: -0.04%, +0.06%
VOPD: 23963 -> 23929 (-0.14%); split: +0.04%, -0.18%

Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/39641>
 2026-02-10 18:42:03 +00:00
Georg Lehmann
7238888d93 nir/opt_algebraic: remove redundant patterns with fcmp(fneg(...), #c) 
We already have patterns to move the negation to the constant.

No Foz-DB changes.

Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@intel.com>
Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/39641>
 2026-02-10 18:42:03 +00:00
Georg Lehmann
03c497f236 nir/opt_algebraic: make 1.0 - fsat(a) -> fsat(1.0 - a) pattern exact using nnan 
Foz-DB Navi48:
Totals from 50 (0.06% of 82405) affected shaders:
CodeSize: 137072 -> 137456 (+0.28%)

Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/39641>
 2026-02-10 18:42:02 +00:00
Georg Lehmann
79e4530a9b nir/opt_algebraic: make pattern pushing fmul into bcsel exact 
The only special case here is d == -0.0.

Foz-DB Navi48:
Totals from 3 (0.00% of 82405) affected shaders:
CodeSize: 29140 -> 29188 (+0.16%)
InvThroughput: 2945 -> 2951 (+0.20%)
VALU: 3217 -> 3223 (+0.19%)

Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@intel.com>
Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/39641>
 2026-02-10 18:42:02 +00:00
Georg Lehmann
a3bc94a3d0 nir/opt_algebraic: remove inexact from floor->trunc pattern 
This was marked inexact because of me in !21475, but I don't see why now,
even after checking all the special values.

No Foz-DB changes.

Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@intel.com>
Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/39641>
 2026-02-10 18:42:02 +00:00
Georg Lehmann
da7abb1337 nir/opt_algebraic: mark fmulz(finite, finite) -> fmul pattern as nsz 
No Foz-DB chagnes.

Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@intel.com>
Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/39641>
 2026-02-10 18:42:02 +00:00
Georg Lehmann
ea87f1f9bc nir/opt_algebraic: add a - a with nnan 
Foz-DB Navi48:
Totals from 576 (0.70% of 82405) affected shaders:
MaxWaves: 16706 -> 16726 (+0.12%)
Instrs: 618677 -> 580965 (-6.10%); split: -6.10%, +0.00%
CodeSize: 3022552 -> 2861612 (-5.32%); split: -5.33%, +0.00%
VGPRs: 28008 -> 28860 (+3.04%); split: -0.51%, +3.56%
Latency: 2689318 -> 2655887 (-1.24%); split: -1.25%, +0.01%
InvThroughput: 403512 -> 393404 (-2.51%); split: -2.51%, +0.00%
VClause: 7584 -> 7577 (-0.09%); split: -0.17%, +0.08%
SClause: 19974 -> 19086 (-4.45%); split: -4.48%, +0.03%
Copies: 43862 -> 40888 (-6.78%); split: -6.87%, +0.09%
Branches: 12457 -> 11407 (-8.43%)
PreSGPRs: 28315 -> 27046 (-4.48%); split: -4.53%, +0.05%
PreVGPRs: 20751 -> 19397 (-6.52%)
VALU: 317224 -> 290151 (-8.53%); split: -8.53%, +0.00%
SALU: 124297 -> 121347 (-2.37%); split: -2.39%, +0.02%
VMEM: 11918 -> 11907 (-0.09%)
SMEM: 27582 -> 26241 (-4.86%)

Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@intel.com>
Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/39641>
 2026-02-10 18:42:02 +00:00
Georg Lehmann
16db9f79d1 nir/opt_algebraic: remove inexact a * 0.0 patterns 
We already have some with nnan,nsz.

Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@intel.com>
Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/39641>
 2026-02-10 18:42:02 +00:00
Georg Lehmann
e443229644 nir/opt_algebraic: mark newly created fmulz nan/inf preserving 
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@intel.com>
Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/39641>
 2026-02-10 18:42:02 +00:00
Georg Lehmann
b678899ef8 nir/opt_algebraic: use nan/inf/sz preserve flags instead of exact for cmp/min/max replacement 
And remove some, because they should be covered by the search pattern anyway.

Foz-DB Navi48:
Totals from 560 (0.68% of 82405) affected shaders:
MaxWaves: 11279 -> 11291 (+0.11%)
Instrs: 5214229 -> 5214386 (+0.00%); split: -0.02%, +0.02%
CodeSize: 29613884 -> 29616740 (+0.01%); split: -0.01%, +0.02%
VGPRs: 50400 -> 50328 (-0.14%)
Latency: 36481700 -> 36481157 (-0.00%); split: -0.01%, +0.01%
InvThroughput: 7309905 -> 7307905 (-0.03%); split: -0.05%, +0.02%
VClause: 131423 -> 131424 (+0.00%); split: -0.00%, +0.00%
SClause: 111485 -> 111499 (+0.01%); split: -0.00%, +0.01%
Copies: 441899 -> 442029 (+0.03%); split: -0.02%, +0.05%
Branches: 165599 -> 165597 (-0.00%)
PreVGPRs: 43558 -> 43525 (-0.08%)
VALU: 2573609 -> 2573324 (-0.01%); split: -0.03%, +0.02%
SALU: 851172 -> 851271 (+0.01%); split: -0.01%, +0.02%
VOPD: 366409 -> 366934 (+0.14%); split: +0.23%, -0.08%

Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/39641>
 2026-02-10 18:42:02 +00:00
Georg Lehmann
45cb1d3b6f nir/opt_algebraic: remove unpack_half_2x16_split 
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/39511>
 2026-02-06 06:12:36 +00:00
Georg Lehmann
b18d9c1b33 nir/opt_algebraic: optimize unpack_32_2x16 of extract 
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/39511>
 2026-02-06 06:12:36 +00:00
Marek Olšák
a3f022d0a2 nir: reassociate a $op (b ? #c : #d) for div, mod, rem
Some checks are pending
macOS-CI / macOS-CI (dri) (push) Waiting to run
Details
macOS-CI / macOS-CI (xlib) (push) Waiting to run
Details 
This eliminates expensive div, mod, rem opcodes with non-constant src1 being
constant src1 hiding behind bcsel.

gcc and LLVM are missing this.

Reviewed-by: Georg Lehmann <dadschoorse@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/39560>
 2026-02-02 21:34:48 +00:00
Georg Lehmann
ad6f8291bf nir/opt_algebraic: rework ignore_exact to work like other internal conditions 
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/39616>
 2026-01-31 15:30:25 +00:00
Georg Lehmann
70f0e75262 nir/opt_algebraic: optimize pack_half_2x16_rtz of float converted from 16bit 
Foz-DB Navi48:
Totals from 177 (0.21% of 82405) affected shaders:
Instrs: 326628 -> 325955 (-0.21%); split: -0.21%, +0.00%
CodeSize: 1726720 -> 1722500 (-0.24%); split: -0.24%, +0.00%
Latency: 5076631 -> 5075700 (-0.02%); split: -0.02%, +0.00%
InvThroughput: 596010 -> 595598 (-0.07%); split: -0.07%, +0.00%
VClause: 3613 -> 3616 (+0.08%)
Copies: 24427 -> 24501 (+0.30%); split: -0.06%, +0.36%
VALU: 182468 -> 182029 (-0.24%); split: -0.24%, +0.00%
SALU: 55449 -> 55452 (+0.01%); split: -0.01%, +0.01%

Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/39531>
 2026-01-29 14:44:37 +00:00
Georg Lehmann
c3e12429c5 nir/opt_algebaric: improve a < 0.0 ? 0.0 : sqrt(a) pattern 
Fix the NaN correctness of the original pattern, and add more variants.

Foz-DB Navi48:
Totals from 372 (0.45% of 82405) affected shaders:
Instrs: 208946 -> 207522 (-0.68%); split: -0.71%, +0.03%
CodeSize: 1116436 -> 1109804 (-0.59%); split: -0.61%, +0.02%
VGPRs: 19452 -> 19104 (-1.79%)
Latency: 1121222 -> 1120423 (-0.07%); split: -0.13%, +0.05%
InvThroughput: 158228 -> 157567 (-0.42%); split: -0.61%, +0.19%
VClause: 3695 -> 3704 (+0.24%)
Copies: 9516 -> 9606 (+0.95%); split: -0.24%, +1.19%
VALU: 118696 -> 118031 (-0.56%); split: -0.61%, +0.05%
VOPD: 380 -> 372 (-2.11%)

Reviewed-by: Emma Anholt <emma@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/39507>
 2026-01-29 11:29:48 +00:00
Georg Lehmann
f872c13707 nir/opt_algebraic: use contract instead of inexact for more patterns 
These use more precise operations, so contract is enough.

Foz-DB Navi48:
Totals from 248 (0.30% of 82405) affected shaders:
Instrs: 284686 -> 284318 (-0.13%); split: -0.14%, +0.01%
CodeSize: 1528856 -> 1527520 (-0.09%); split: -0.10%, +0.01%
Latency: 2368390 -> 2367345 (-0.04%); split: -0.06%, +0.01%
InvThroughput: 346623 -> 346335 (-0.08%); split: -0.09%, +0.01%
SClause: 6752 -> 6756 (+0.06%); split: -0.12%, +0.18%
Copies: 14685 -> 14694 (+0.06%); split: -0.01%, +0.07%
VALU: 179922 -> 179727 (-0.11%); split: -0.11%, +0.01%
SALU: 28706 -> 28707 (+0.00%)
VOPD: 1196 -> 1198 (+0.17%)

Reviewed-by: Emma Anholt <emma@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/39507>
 2026-01-29 11:29:48 +00:00
Georg Lehmann
d8ef28671d nir/opt_algebraic: use correct syntax to create exact fsat 
Fixes: 3b06824e4c ("nir/opt_algebraic: optimize some post peephole select patterns")

Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@intel.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/39586>
 2026-01-28 18:46:22 +00:00
Georg Lehmann
b2d9615000 nir/opt_algebraic: optimize bcsel to hi 16bits with undef lo 
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Acked-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/39412>
 2026-01-26 10:54:20 +00:00