From 0dbda153aae548a4087f7364c9013583a076e0e9 Mon Sep 17 00:00:00 2001 From: Jason Ekstrand Date: Thu, 17 Mar 2016 11:31:48 -0700 Subject: [PATCH] nir/algebraic: Flag inexact optimizations Many of our optimizations, while great for cutting shaders down to size, aren't really precision-safe. This commit tries to flag all of the inexact floating-point optimizations so they don't get run on values that are flagged "exact". It's a bit conservative and maybe flags some safe optimizations as unsafe but that's better than missing one. Reviewed-by: Francisco Jerez --- src/compiler/nir/nir_opt_algebraic.py | 121 +++++++++++++------------- 1 file changed, 62 insertions(+), 59 deletions(-) diff --git a/src/compiler/nir/nir_opt_algebraic.py b/src/compiler/nir/nir_opt_algebraic.py index 7e3aa5aa798..53633233f2b 100644 --- a/src/compiler/nir/nir_opt_algebraic.py +++ b/src/compiler/nir/nir_opt_algebraic.py @@ -61,19 +61,19 @@ optimizations = [ (('fabs', ('fneg', a)), ('fabs', a)), (('iabs', ('iabs', a)), ('iabs', a)), (('iabs', ('ineg', a)), ('iabs', a)), - (('fadd', a, 0.0), a), + (('~fadd', a, 0.0), a), (('iadd', a, 0), a), (('usadd_4x8', a, 0), a), (('usadd_4x8', a, ~0), ~0), - (('fadd', ('fmul', a, b), ('fmul', a, c)), ('fmul', a, ('fadd', b, c))), + (('~fadd', ('fmul', a, b), ('fmul', a, c)), ('fmul', a, ('fadd', b, c))), (('iadd', ('imul', a, b), ('imul', a, c)), ('imul', a, ('iadd', b, c))), - (('fadd', ('fneg', a), a), 0.0), + (('~fadd', ('fneg', a), a), 0.0), (('iadd', ('ineg', a), a), 0), (('iadd', ('ineg', a), ('iadd', a, b)), b), (('iadd', a, ('iadd', ('ineg', a), b)), b), - (('fadd', ('fneg', a), ('fadd', a, b)), b), - (('fadd', a, ('fadd', ('fneg', a), b)), b), - (('fmul', a, 0.0), 0.0), + (('~fadd', ('fneg', a), ('fadd', a, b)), b), + (('~fadd', a, ('fadd', ('fneg', a), b)), b), + (('~fmul', a, 0.0), 0.0), (('imul', a, 0), 0), (('umul_unorm_4x8', a, 0), 0), (('umul_unorm_4x8', a, ~0), a), @@ -81,29 +81,29 @@ optimizations = [ (('imul', a, 1), a), (('fmul', a, -1.0), ('fneg', a)), (('imul', a, -1), ('ineg', a)), - (('ffma', 0.0, a, b), b), - (('ffma', a, 0.0, b), b), - (('ffma', a, b, 0.0), ('fmul', a, b)), + (('~ffma', 0.0, a, b), b), + (('~ffma', a, 0.0, b), b), + (('~ffma', a, b, 0.0), ('fmul', a, b)), (('ffma', a, 1.0, b), ('fadd', a, b)), (('ffma', 1.0, a, b), ('fadd', a, b)), - (('flrp', a, b, 0.0), a), - (('flrp', a, b, 1.0), b), - (('flrp', a, a, b), a), - (('flrp', 0.0, a, b), ('fmul', a, b)), - (('flrp', a, b, ('b2f', c)), ('bcsel', c, b, a), 'options->lower_flrp'), + (('~flrp', a, b, 0.0), a), + (('~flrp', a, b, 1.0), b), + (('~flrp', a, a, b), a), + (('~flrp', 0.0, a, b), ('fmul', a, b)), + (('~flrp', a, b, ('b2f', c)), ('bcsel', c, b, a), 'options->lower_flrp'), (('flrp', a, b, c), ('fadd', ('fmul', c, ('fsub', b, a)), a), 'options->lower_flrp'), (('ffract', a), ('fsub', a, ('ffloor', a)), 'options->lower_ffract'), - (('fadd', ('fmul', a, ('fadd', 1.0, ('fneg', ('b2f', c)))), ('fmul', b, ('b2f', c))), ('bcsel', c, b, a), 'options->lower_flrp'), - (('fadd', ('fmul', a, ('fadd', 1.0, ('fneg', c ))), ('fmul', b, c )), ('flrp', a, b, c), '!options->lower_flrp'), - (('fadd', a, ('fmul', ('b2f', c), ('fadd', b, ('fneg', a)))), ('bcsel', c, b, a), 'options->lower_flrp'), - (('fadd', a, ('fmul', c , ('fadd', b, ('fneg', a)))), ('flrp', a, b, c), '!options->lower_flrp'), + (('~fadd', ('fmul', a, ('fadd', 1.0, ('fneg', ('b2f', c)))), ('fmul', b, ('b2f', c))), ('bcsel', c, b, a), 'options->lower_flrp'), + (('~fadd', ('fmul', a, ('fadd', 1.0, ('fneg', c ))), ('fmul', b, c )), ('flrp', a, b, c), '!options->lower_flrp'), + (('~fadd', a, ('fmul', ('b2f', c), ('fadd', b, ('fneg', a)))), ('bcsel', c, b, a), 'options->lower_flrp'), + (('~fadd', a, ('fmul', c , ('fadd', b, ('fneg', a)))), ('flrp', a, b, c), '!options->lower_flrp'), (('ffma', a, b, c), ('fadd', ('fmul', a, b), c), 'options->lower_ffma'), - (('fadd', ('fmul', a, b), c), ('ffma', a, b, c), '!options->lower_ffma'), + (('~fadd', ('fmul', a, b), c), ('ffma', a, b, c), '!options->lower_ffma'), # Comparison simplifications - (('inot', ('flt', a, b)), ('fge', a, b)), - (('inot', ('fge', a, b)), ('flt', a, b)), - (('inot', ('feq', a, b)), ('fne', a, b)), - (('inot', ('fne', a, b)), ('feq', a, b)), + (('~inot', ('flt', a, b)), ('fge', a, b)), + (('~inot', ('fge', a, b)), ('flt', a, b)), + (('~inot', ('feq', a, b)), ('fne', a, b)), + (('~inot', ('fne', a, b)), ('feq', a, b)), (('inot', ('ilt', a, b)), ('ige', a, b)), (('inot', ('ige', a, b)), ('ilt', a, b)), (('inot', ('ieq', a, b)), ('ine', a, b)), @@ -132,15 +132,15 @@ optimizations = [ (('imax', a, a), a), (('umin', a, a), a), (('umax', a, a), a), - (('fmin', ('fmax', a, 0.0), 1.0), ('fsat', a), '!options->lower_fsat'), - (('fmax', ('fmin', a, 1.0), 0.0), ('fsat', a), '!options->lower_fsat'), + (('~fmin', ('fmax', a, 0.0), 1.0), ('fsat', a), '!options->lower_fsat'), + (('~fmax', ('fmin', a, 1.0), 0.0), ('fsat', a), '!options->lower_fsat'), (('fsat', a), ('fmin', ('fmax', a, 0.0), 1.0), 'options->lower_fsat'), (('fsat', ('fsat', a)), ('fsat', a)), (('fmin', ('fmax', ('fmin', ('fmax', a, 0.0), 1.0), 0.0), 1.0), ('fmin', ('fmax', a, 0.0), 1.0)), - (('ior', ('flt', a, b), ('flt', a, c)), ('flt', a, ('fmax', b, c))), - (('ior', ('flt', a, c), ('flt', b, c)), ('flt', ('fmin', a, b), c)), - (('ior', ('fge', a, b), ('fge', a, c)), ('fge', a, ('fmin', b, c))), - (('ior', ('fge', a, c), ('fge', b, c)), ('fge', ('fmax', a, b), c)), + (('~ior', ('flt', a, b), ('flt', a, c)), ('flt', a, ('fmax', b, c))), + (('~ior', ('flt', a, c), ('flt', b, c)), ('flt', ('fmin', a, b), c)), + (('~ior', ('fge', a, b), ('fge', a, c)), ('fge', a, ('fmin', b, c))), + (('~ior', ('fge', a, c), ('fge', b, c)), ('fge', ('fmax', a, b), c)), (('fabs', ('slt', a, b)), ('slt', a, b)), (('fabs', ('sge', a, b)), ('sge', a, b)), (('fabs', ('seq', a, b)), ('seq', a, b)), @@ -191,35 +191,35 @@ optimizations = [ (('iand', 0xff, ('ushr', a, 24)), ('ushr', a, 24)), (('iand', 0xffff, ('ushr', a, 16)), ('ushr', a, 16)), # Exponential/logarithmic identities - (('fexp2', ('flog2', a)), a), # 2^lg2(a) = a - (('flog2', ('fexp2', a)), a), # lg2(2^a) = a + (('~fexp2', ('flog2', a)), a), # 2^lg2(a) = a + (('~flog2', ('fexp2', a)), a), # lg2(2^a) = a (('fpow', a, b), ('fexp2', ('fmul', ('flog2', a), b)), 'options->lower_fpow'), # a^b = 2^(lg2(a)*b) - (('fexp2', ('fmul', ('flog2', a), b)), ('fpow', a, b), '!options->lower_fpow'), # 2^(lg2(a)*b) = a^b - (('fexp2', ('fadd', ('fmul', ('flog2', a), b), ('fmul', ('flog2', c), d))), - ('fmul', ('fpow', a, b), ('fpow', c, d)), '!options->lower_fpow'), # 2^(lg2(a) * b + lg2(c) + d) = a^b * c^d - (('fpow', a, 1.0), a), - (('fpow', a, 2.0), ('fmul', a, a)), - (('fpow', a, 4.0), ('fmul', ('fmul', a, a), ('fmul', a, a))), - (('fpow', 2.0, a), ('fexp2', a)), - (('fpow', ('fpow', a, 2.2), 0.454545), a), - (('fpow', ('fabs', ('fpow', a, 2.2)), 0.454545), ('fabs', a)), - (('fsqrt', ('fexp2', a)), ('fexp2', ('fmul', 0.5, a))), - (('frcp', ('fexp2', a)), ('fexp2', ('fneg', a))), - (('frsq', ('fexp2', a)), ('fexp2', ('fmul', -0.5, a))), - (('flog2', ('fsqrt', a)), ('fmul', 0.5, ('flog2', a))), - (('flog2', ('frcp', a)), ('fneg', ('flog2', a))), - (('flog2', ('frsq', a)), ('fmul', -0.5, ('flog2', a))), - (('flog2', ('fpow', a, b)), ('fmul', b, ('flog2', a))), - (('fadd', ('flog2', a), ('flog2', b)), ('flog2', ('fmul', a, b))), - (('fadd', ('flog2', a), ('fneg', ('flog2', b))), ('flog2', ('fdiv', a, b))), - (('fmul', ('fexp2', a), ('fexp2', b)), ('fexp2', ('fadd', a, b))), + (('~fexp2', ('fmul', ('flog2', a), b)), ('fpow', a, b), '!options->lower_fpow'), # 2^(lg2(a)*b) = a^b + (('~fexp2', ('fadd', ('fmul', ('flog2', a), b), ('fmul', ('flog2', c), d))), + ('~fmul', ('fpow', a, b), ('fpow', c, d)), '!options->lower_fpow'), # 2^(lg2(a) * b + lg2(c) + d) = a^b * c^d + (('~fpow', a, 1.0), a), + (('~fpow', a, 2.0), ('fmul', a, a)), + (('~fpow', a, 4.0), ('fmul', ('fmul', a, a), ('fmul', a, a))), + (('~fpow', 2.0, a), ('fexp2', a)), + (('~fpow', ('fpow', a, 2.2), 0.454545), a), + (('~fpow', ('fabs', ('fpow', a, 2.2)), 0.454545), ('fabs', a)), + (('~fsqrt', ('fexp2', a)), ('fexp2', ('fmul', 0.5, a))), + (('~frcp', ('fexp2', a)), ('fexp2', ('fneg', a))), + (('~frsq', ('fexp2', a)), ('fexp2', ('fmul', -0.5, a))), + (('~flog2', ('fsqrt', a)), ('fmul', 0.5, ('flog2', a))), + (('~flog2', ('frcp', a)), ('fneg', ('flog2', a))), + (('~flog2', ('frsq', a)), ('fmul', -0.5, ('flog2', a))), + (('~flog2', ('fpow', a, b)), ('fmul', b, ('flog2', a))), + (('~fadd', ('flog2', a), ('flog2', b)), ('flog2', ('fmul', a, b))), + (('~fadd', ('flog2', a), ('fneg', ('flog2', b))), ('flog2', ('fdiv', a, b))), + (('~fmul', ('fexp2', a), ('fexp2', b)), ('fexp2', ('fadd', a, b))), # Division and reciprocal - (('fdiv', 1.0, a), ('frcp', a)), + (('~fdiv', 1.0, a), ('frcp', a)), (('fdiv', a, b), ('fmul', a, ('frcp', b)), 'options->lower_fdiv'), - (('frcp', ('frcp', a)), a), - (('frcp', ('fsqrt', a)), ('frsq', a)), + (('~frcp', ('frcp', a)), a), + (('~frcp', ('fsqrt', a)), ('frsq', a)), (('fsqrt', a), ('frcp', ('frsq', a)), 'options->lower_fsqrt'), - (('frcp', ('frsq', a)), ('fsqrt', a), '!options->lower_fsqrt'), + (('~frcp', ('frsq', a)), ('fsqrt', a), '!options->lower_fsqrt'), # Boolean simplifications (('ieq', 'a@bool', True), a), (('ine', 'a@bool', True), ('inot', a)), @@ -256,7 +256,7 @@ optimizations = [ (('iand', 0xffff, a), ('extract_u16', a, 0), '!options->lower_extract_word'), # Subtracts - (('fsub', a, ('fsub', 0.0, b)), ('fadd', a, b)), + (('~fsub', a, ('fsub', 0.0, b)), ('fadd', a, b)), (('isub', a, ('isub', 0, b)), ('iadd', a, b)), (('ussub_4x8', a, 0), a), (('ussub_4x8', a, ~0), 0), @@ -264,7 +264,7 @@ optimizations = [ (('isub', a, b), ('iadd', a, ('ineg', b)), 'options->lower_sub'), (('fneg', a), ('fsub', 0.0, a), 'options->lower_negate'), (('ineg', a), ('isub', 0, a), 'options->lower_negate'), - (('fadd', a, ('fsub', 0.0, b)), ('fsub', a, b)), + (('~fadd', a, ('fsub', 0.0, b)), ('fsub', a, b)), (('iadd', a, ('isub', 0, b)), ('isub', a, b)), (('fabs', ('fsub', 0.0, a)), ('fabs', a)), (('iabs', ('isub', 0, a)), ('iabs', a)), @@ -393,10 +393,13 @@ for op in ['flt', 'fge', 'feq', 'fne', # they help code generation but do not necessarily produce code that is # more easily optimizable. late_optimizations = [ + # Most of these optimizations aren't quite safe when you get infinity or + # Nan involved but the first one should be fine. (('flt', ('fadd', a, b), 0.0), ('flt', a, ('fneg', b))), - (('fge', ('fadd', a, b), 0.0), ('fge', a, ('fneg', b))), - (('feq', ('fadd', a, b), 0.0), ('feq', a, ('fneg', b))), - (('fne', ('fadd', a, b), 0.0), ('fne', a, ('fneg', b))), + (('~fge', ('fadd', a, b), 0.0), ('fge', a, ('fneg', b))), + (('~feq', ('fadd', a, b), 0.0), ('feq', a, ('fneg', b))), + (('~fne', ('fadd', a, b), 0.0), ('fne', a, ('fneg', b))), + (('fdot2', a, b), ('fdot_replicated2', a, b), 'options->fdot_replicates'), (('fdot3', a, b), ('fdot_replicated3', a, b), 'options->fdot_replicates'), (('fdot4', a, b), ('fdot_replicated4', a, b), 'options->fdot_replicates'), -- 2.30.2