fdo-mirrors/mesa
1
0
Fork 0
mirror of https://gitlab.freedesktop.org/mesa/mesa.git synced 2025-12-20 22:30:12 +01:00
Code Issues Projects Releases Packages Wiki Activity Actions 1

nir/algebraic: Invert comparisons less often

Browse source 
This fixes the piglit test range_analysis_fsat_of_nan.shader_test.  That
test contains some code like

    o = saturate(X) > 0 ? vec4(1.0, 0.0, 0.0, 1.0)
                        : vec4(0.0, 1.0, 0.0, 1.0);

A clever optimizer will convert this to

    o = vec4(float(saturate(X) > 0),
             float(!(saturate(X) > 0)),
             0, 1);

Due to the ordering of optimizations in the compiler, the `saturate`
operations are removed.  This is safe even in the presense of NaN.

    o = vec4(float(X > 0), float(!(X > 0)), 0, 1);

Since the calculations are not marked precise, an overzealous
optimizer may reduce this to

    o = vec4(float(X > 0), float(X <= 0), 0, 1);

This will result in black being output.  The GLSL spec gives quite a bit
of leeway with respect to NaN, but that seems too far.  The shader
author asked for a result of red or green.  A result of black is still
"undefined behavior," but it's also a little mean.

This also enables CSE to do its job better.

v2: Update A530 expected image checksum for minetest.trace.

Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/4531
Fixes: 0dbda153aa ("nir/algebraic: Flag inexact optimizations")
Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>

Tiger Lake
total instructions in shared programs: 21041563 -> 21041789 (<.01%)
instructions in affected programs: 992066 -> 992292 (0.02%)
helped: 526
HURT: 548
helped stats (abs) min: 1 max: 16 x̄: 2.48 x̃: 2
helped stats (rel) min: 0.04% max: 5.56% x̄: 0.74% x̃: 0.49%
HURT stats (abs)   min: 1 max: 27 x̄: 2.80 x̃: 2
HURT stats (rel)   min: 0.04% max: 4.55% x̄: 0.59% x̃: 0.38%
95% mean confidence interval for instructions value: -0.00 0.42
95% mean confidence interval for instructions %-change: -0.12% <.01%
Inconclusive result (value mean confidence interval includes 0).

total cycles in shared programs: 855885569 -> 856118189 (0.03%)
cycles in affected programs: 343637248 -> 343869868 (0.07%)
helped: 907
HURT: 541
helped stats (abs) min: 1 max: 7724 x̄: 206.45 x̃: 36
helped stats (rel) min: <.01% max: 29.97% x̄: 1.01% x̃: 0.37%
HURT stats (abs)   min: 1 max: 14177 x̄: 776.09 x̃: 31
HURT stats (rel)   min: <.01% max: 29.94% x̄: 1.24% x̃: 0.35%
95% mean confidence interval for cycles value: 84.30 237.00
95% mean confidence interval for cycles %-change: -0.32% -0.01%
Inconclusive result (value mean confidence interval and %-change mean confidence interval disagree).

LOST:   3
GAINED: 5

Ice Lake
total instructions in shared programs: 20027107 -> 20025352 (<.01%)
instructions in affected programs: 1068856 -> 1067101 (-0.16%)
helped: 1153
HURT: 273
helped stats (abs) min: 1 max: 14 x̄: 1.83 x̃: 1
helped stats (rel) min: 0.03% max: 5.66% x̄: 0.61% x̃: 0.35%
HURT stats (abs)   min: 1 max: 15 x̄: 1.29 x̃: 1
HURT stats (rel)   min: 0.16% max: 1.30% x̄: 0.58% x̃: 0.60%
95% mean confidence interval for instructions value: -1.33 -1.13
95% mean confidence interval for instructions %-change: -0.43% -0.34%
Instructions are helped.

total cycles in shared programs: 979499227 -> 979448725 (<.01%)
cycles in affected programs: 344261539 -> 344211037 (-0.01%)
helped: 1079
HURT: 441
helped stats (abs) min: 1 max: 9384 x̄: 147.78 x̃: 48
helped stats (rel) min: <.01% max: 31.83% x̄: 0.90% x̃: 0.33%
HURT stats (abs)   min: 1 max: 7220 x̄: 247.07 x̃: 32
HURT stats (rel)   min: <.01% max: 31.30% x̄: 1.52% x̃: 0.53%
95% mean confidence interval for cycles value: -70.01 3.56
95% mean confidence interval for cycles %-change: -0.35% -0.05%
Inconclusive result (value mean confidence interval includes 0).

total spills in shared programs: 10564 -> 10568 (0.04%)
spills in affected programs: 143 -> 147 (2.80%)
helped: 0
HURT: 1

total fills in shared programs: 11343 -> 11347 (0.04%)
fills in affected programs: 287 -> 291 (1.39%)
helped: 0
HURT: 1

LOST:   3
GAINED: 2

Skylake
total instructions in shared programs: 18192274 -> 18190128 (-0.01%)
instructions in affected programs: 1000188 -> 998042 (-0.21%)
helped: 1149
HURT: 55
helped stats (abs) min: 1 max: 14 x̄: 1.92 x̃: 1
helped stats (rel) min: 0.04% max: 6.67% x̄: 0.67% x̃: 0.42%
HURT stats (abs)   min: 1 max: 2 x̄: 1.05 x̃: 1
HURT stats (rel)   min: 0.16% max: 0.55% x̄: 0.27% x̃: 0.26%
95% mean confidence interval for instructions value: -1.87 -1.69
95% mean confidence interval for instructions %-change: -0.67% -0.58%
Instructions are helped.

total cycles in shared programs: 960856054 -> 960728040 (-0.01%)
cycles in affected programs: 340840968 -> 340712954 (-0.04%)
helped: 1079
HURT: 233
helped stats (abs) min: 1 max: 7640 x̄: 170.95 x̃: 46
helped stats (rel) min: <.01% max: 30.20% x̄: 0.96% x̃: 0.28%
HURT stats (abs)   min: 1 max: 6864 x̄: 242.23 x̃: 26
HURT stats (rel)   min: <.01% max: 34.64% x̄: 2.10% x̃: 0.22%
95% mean confidence interval for cycles value: -135.62 -59.53
95% mean confidence interval for cycles %-change: -0.59% -0.25%
Cycles are helped.

LOST:   15
GAINED: 1

Broadwell
total instructions in shared programs: 17855624 -> 17853580 (-0.01%)
instructions in affected programs: 1012209 -> 1010165 (-0.20%)
helped: 1105
HURT: 52
helped stats (abs) min: 1 max: 13 x̄: 1.90 x̃: 1
helped stats (rel) min: 0.03% max: 6.67% x̄: 0.67% x̃: 0.36%
HURT stats (abs)   min: 1 max: 1 x̄: 1.00 x̃: 1
HURT stats (rel)   min: 0.13% max: 0.52% x̄: 0.26% x̃: 0.25%
95% mean confidence interval for instructions value: -1.86 -1.67
95% mean confidence interval for instructions %-change: -0.68% -0.58%
Instructions are helped.

total cycles in shared programs: 1029905447 -> 1029840699 (<.01%)
cycles in affected programs: 347102680 -> 347037932 (-0.02%)
helped: 1007
HURT: 211
helped stats (abs) min: 1 max: 1360 x̄: 89.76 x̃: 48
helped stats (rel) min: <.01% max: 16.26% x̄: 0.69% x̃: 0.25%
HURT stats (abs)   min: 1 max: 1297 x̄: 121.51 x̃: 20
HURT stats (rel)   min: <.01% max: 31.31% x̄: 1.21% x̃: 0.20%
95% mean confidence interval for cycles value: -62.39 -43.92
95% mean confidence interval for cycles %-change: -0.47% -0.25%
Cycles are helped.

total spills in shared programs: 20335 -> 20333 (<.01%)
spills in affected programs: 19 -> 17 (-10.53%)
helped: 2
HURT: 0

total fills in shared programs: 25905 -> 25899 (-0.02%)
fills in affected programs: 23 -> 17 (-26.09%)
helped: 2
HURT: 0

LOST:   9
GAINED: 0

Haswell
total instructions in shared programs: 16418516 -> 16417293 (<.01%)
instructions in affected programs: 223785 -> 222562 (-0.55%)
helped: 590
HURT: 67
helped stats (abs) min: 1 max: 15 x̄: 2.19 x̃: 1
helped stats (rel) min: 0.03% max: 6.52% x̄: 0.87% x̃: 0.60%
HURT stats (abs)   min: 1 max: 2 x̄: 1.04 x̃: 1
HURT stats (rel)   min: 0.04% max: 1.85% x̄: 0.44% x̃: 0.25%
95% mean confidence interval for instructions value: -2.01 -1.71
95% mean confidence interval for instructions %-change: -0.80% -0.67%
Instructions are helped.

total cycles in shared programs: 1037179754 -> 1037084874 (<.01%)
cycles in affected programs: 352541071 -> 352446191 (-0.03%)
helped: 1093
HURT: 182
helped stats (abs) min: 1 max: 888 x̄: 111.03 x̃: 64
helped stats (rel) min: <.01% max: 27.30% x̄: 0.84% x̃: 0.20%
HURT stats (abs)   min: 1 max: 6777 x̄: 145.49 x̃: 21
HURT stats (rel)   min: <.01% max: 24.10% x̄: 1.99% x̃: 0.29%
95% mean confidence interval for cycles value: -88.10 -60.73
95% mean confidence interval for cycles %-change: -0.58% -0.29%
Cycles are helped.

total spills in shared programs: 17457 -> 17456 (<.01%)
spills in affected programs: 12 -> 11 (-8.33%)
helped: 1
HURT: 0

total fills in shared programs: 20387 -> 20385 (<.01%)
fills in affected programs: 15 -> 13 (-13.33%)
helped: 1
HURT: 0

LOST:   6
GAINED: 1

Ivy Bridge and earlier platforms had similar results. (Ivy Bridge shown)
total instructions in shared programs: 15515482 -> 15513998 (<.01%)
instructions in affected programs: 239739 -> 238255 (-0.62%)
helped: 573
HURT: 57
helped stats (abs) min: 1 max: 20 x̄: 2.73 x̃: 2
helped stats (rel) min: 0.03% max: 9.84% x̄: 0.94% x̃: 0.55%
HURT stats (abs)   min: 1 max: 2 x̄: 1.39 x̃: 1
HURT stats (rel)   min: 0.09% max: 1.85% x̄: 0.52% x̃: 0.35%
95% mean confidence interval for instructions value: -2.57 -2.14
95% mean confidence interval for instructions %-change: -0.89% -0.73%
Instructions are helped.

total cycles in shared programs: 584509880 -> 584463152 (<.01%)
cycles in affected programs: 11765280 -> 11718552 (-0.40%)
helped: 661
HURT: 152
helped stats (abs) min: 1 max: 3073 x̄: 101.99 x̃: 32
helped stats (rel) min: <.01% max: 34.38% x̄: 1.46% x̃: 0.50%
HURT stats (abs)   min: 1 max: 6637 x̄: 136.10 x̃: 15
HURT stats (rel)   min: <.01% max: 24.19% x̄: 1.75% x̃: 0.25%
95% mean confidence interval for cycles value: -82.79 -32.16
95% mean confidence interval for cycles %-change: -1.11% -0.61%
Cycles are helped.

LOST:   9
GAINED: 0

Tiger Lake
Instructions in all programs: 160905127 -> 160900949 (-0.0%)
SENDs in all programs: 6812418 -> 6812085 (-0.0%)
Loops in all programs: 38225 -> 38225 (+0.0%)
Cycles in all programs: 7431911114 -> 7433914697 (+0.0%)
Spills in all programs: 192582 -> 192582 (+0.0%)
Fills in all programs: 304539 -> 304537 (-0.0%)

Ice Lake
Instructions in all programs: 145296733 -> 145292370 (-0.0%)
SENDs in all programs: 6863818 -> 6863485 (-0.0%)
Loops in all programs: 38219 -> 38219 (+0.0%)
Cycles in all programs: 8798257570 -> 8800204360 (+0.0%)
Spills in all programs: 216880 -> 216880 (+0.0%)
Fills in all programs: 334250 -> 334248 (-0.0%)

Skylake
Instructions in all programs: 135891485 -> 135887357 (-0.0%)
SENDs in all programs: 6803031 -> 6802698 (-0.0%)
Loops in all programs: 38216 -> 38216 (+0.0%)
Cycles in all programs: 8442221881 -> 8444201959 (+0.0%)
Spills in all programs: 194839 -> 194839 (+0.0%)
Fills in all programs: 301116 -> 301114 (-0.0%)

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10012>
This commit is contained in:
Ian Romanick 2020-02-25 10:36:23 -08:00 • committed by Marge Bot
parent 49177b9e2f
commit 4246c2869c
6 changed files with 11 additions and 15 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

20
src/compiler/nir/nir_opt_algebraic.py
View file

@ -350,16 +350,16 @@ optimizations.extend([
('iadd', ('ishl', ('iadd', a, c), d), ('ishl', b, d))),
# Comparison simplifications
(('~inot', ('flt', a, b)), ('fge', a, b)),
(('~inot', ('fge', a, b)), ('flt', a, b)),
(('inot', ('feq', a, b)), ('fneu', a, b)),
(('inot', ('fneu', a, b)), ('feq', a, b)),
(('inot', ('ilt', a, b)), ('ige', a, b)),
(('inot', ('ult', a, b)), ('uge', a, b)),
(('inot', ('ige', a, b)), ('ilt', a, b)),
(('inot', ('uge', a, b)), ('ult', a, b)),
(('inot', ('ieq', a, b)), ('ine', a, b)),
(('inot', ('ine', a, b)), ('ieq', a, b)),
(('~inot', ('flt(is_used_once)', a, b)), ('fge', a, b)),
(('~inot', ('fge(is_used_once)', a, b)), ('flt', a, b)),
(('inot', ('feq(is_used_once)', a, b)), ('fneu', a, b)),
(('inot', ('fneu(is_used_once)', a, b)), ('feq', a, b)),
(('inot', ('ilt(is_used_once)', a, b)), ('ige', a, b)),
(('inot', ('ult(is_used_once)', a, b)), ('uge', a, b)),
(('inot', ('ige(is_used_once)', a, b)), ('ilt', a, b)),
(('inot', ('uge(is_used_once)', a, b)), ('ult', a, b)),
(('inot', ('ieq(is_used_once)', a, b)), ('ine', a, b)),
(('inot', ('ine(is_used_once)', a, b)), ('ieq', a, b)),
(('iand', ('feq', a, b), ('fneu', a, b)), False),
(('iand', ('flt', a, b), ('flt', b, a)), False),

2
src/freedreno/ci/traces-freedreno.yml
View file

@ -437,7 +437,7 @@ traces:
- device: freedreno-a306
checksum: 9227cc8d4e6445f2323438340f2a5d9b
- device: freedreno-a530
checksum: 76fbd196f69594c927dc994b4492dd50
checksum: d0d655d81fabeb4087bf7c4837301f2a
- device: freedreno-a630
checksum: d70eb1f72d5d50e6cb617520234f20fb
- path: neverball/neverball.trace

1
src/gallium/drivers/freedreno/ci/piglit-freedreno-a530-fails.txt
View file

@ -334,4 +334,3 @@ spec@glsl-1.30@execution@clipping@fs-clip-distance-interpolated,Crash
spec@glsl-1.30@execution@fs-large-local-array-vec2,Crash
spec@glsl-1.30@execution@fs-large-local-array-vec3,Crash
spec@glsl-1.30@execution@fs-large-local-array-vec4,Crash
spec@glsl-1.30@execution@range_analysis_fsat_of_nan,Fail

1
src/gallium/drivers/freedreno/ci/piglit-freedreno-a630-fails.txt
View file

@ -309,7 +309,6 @@ spec@glsl-1.20@execution@tex-miplevel-selection gl2:textureproj 2d,Crash
spec@glsl-1.20@execution@tex-miplevel-selection gl2:textureproj 2d_projvec4,Crash
spec@glsl-1.20@execution@tex-miplevel-selection gl2:textureproj 2dshadow,Crash
spec@glsl-1.20@execution@tex-miplevel-selection gl2:textureproj 3d,Crash
spec@glsl-1.30@execution@range_analysis_fsat_of_nan,Fail
spec@glsl-1.30@execution@tex-miplevel-selection texture() 1d,Crash
spec@glsl-1.30@execution@tex-miplevel-selection texture() 1darray,Crash
spec@glsl-1.30@execution@tex-miplevel-selection texture() 1darrayshadow,Crash

1
src/gallium/drivers/llvmpipe/ci/llvmpipe-quick_shader.txt
View file

@ -158,7 +158,6 @@ spec/ext_shader_framebuffer_fetch_non_coherent/execution/gles3/simple-ms8: skip
spec/ext_shader_image_load_formatted/execution/image_checkerboard: skip
spec/glsl-1.10/preprocessor/extension-defined-test: skip
spec/glsl-1.10/preprocessor/extension-if-1: skip
spec/glsl-1.30/execution/range_analysis_fsat_of_nan: fail
spec/glsl-1.50/execution/compatibility/clipping/gs-clip-vertex-const-accept: skip
spec/glsl-1.50/execution/compatibility/clipping/gs-clip-vertex-const-reject: skip
spec/glsl-1.50/execution/compatibility/clipping/gs-clip-vertex-different-from-position: skip

1
src/gallium/drivers/softpipe/ci/softpipe-quick.txt
View file

@ -2702,7 +2702,6 @@ spec/glsl-1.10/execution/fs-dfdx-accuracy: warn
spec/glsl-1.10/execution/fs-dfdy-accuracy: warn
spec/glsl-1.10/preprocessor/extension-defined-test: skip
spec/glsl-1.10/preprocessor/extension-if-1: skip
spec/glsl-1.30/execution/range_analysis_fsat_of_nan: fail
spec/glsl-1.30/execution/tex-miplevel-selection texturegrad 1d: fail
spec/glsl-1.30/execution/tex-miplevel-selection texturegrad 1darray: fail
spec/glsl-1.30/execution/tex-miplevel-selection texturegrad 1darrayshadow: fail