From f71fc26393e5dab0d49c2cd9a6be55b00bb87790 Mon Sep 17 00:00:00 2001 From: Georg Lehmann Date: Sat, 22 Mar 2025 06:53:20 +0100 Subject: [PATCH] nir/opt_algebraic: generalize fmax(fadd(a, b), 0.0) to fsat by not requiring fneg Not a large effect, but it's positive and makes the pattern simpler. Foz-DB Navi21: Totals from 1 (0.00% of 79789) affected shaders: Instrs: 145 -> 138 (-4.83%) CodeSize: 784 -> 756 (-3.57%) Latency: 1495 -> 1487 (-0.54%) InvThroughput: 210 -> 196 (-6.67%) VALU: 103 -> 96 (-6.80%) Reviewed-by: Alyssa Rosenzweig Part-of: --- src/compiler/nir/nir_opt_algebraic.py | 10 +++++----- 1 file changed, 5 insertions(+), 5 deletions(-) diff --git a/src/compiler/nir/nir_opt_algebraic.py b/src/compiler/nir/nir_opt_algebraic.py index 4a350bee129..a8c2fba3b07 100644 --- a/src/compiler/nir/nir_opt_algebraic.py +++ b/src/compiler/nir/nir_opt_algebraic.py @@ -1022,14 +1022,14 @@ optimizations.extend([ (('fmax', ('fneg', ('fmin', b, a)), b), ('fmax', ('fabs', b), ('fneg', a))), (('fmin', ('fneg', ('fmax', b, a)), b), ('fmin', ('fneg', ('fabs', b)), ('fneg', a))), - # If a in [0,b] then b-a is also in [0,b]. Since b in [0,1], max(b-a, 0) = - # fsat(b-a). + # If a in [-b,0] then a+b is in [0,b]. Since b in [0,1], max(a+b, 0) = + # fsat(a+b). # - # If a > b, then b-a < 0 and max(b-a, 0) = fsat(b-a) = 0 + # If a < -b, then a+b < 0 and max(a+b, 0) = fsat(a+b) = 0 # # This should be NaN safe since max(NaN, 0) = fsat(NaN) = 0. - (('fmax', ('fadd(is_used_once)', ('fneg', 'a(is_not_negative)'), '#b(is_zero_to_one)'), 0.0), - ('fsat', ('fadd', ('fneg', a), b)), '!options->lower_fsat'), + (('fmax', ('fadd(is_used_once)', 'a(is_not_positive)', '#b(is_zero_to_one)'), 0.0), + ('fsat', ('fadd', a, b)), '!options->lower_fsat'), (('extract_u8', ('imin', ('imax', a, 0), 0xff), 0), ('imin', ('imax', a, 0), 0xff)),