AlexIndustrial/mesa
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

aco: use v_add_f{16,32} with clamp for fsat

Browse Source 
v_add can be dual issued on gfx11, v_med3 cannot.
Don't use v_add directly to still optimize omod(fsat(x)).

Foz-DB GFX1100:
Totals from 32702 (24.24% of 134913) affected shaders:
Latency: 475008203 -> 474928037 (-0.02%); split: -0.02%, +0.00%
InvThroughput: 59226198 -> 59140787 (-0.14%); split: -0.14%, +0.00%

Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21402>
This commit is contained in:
Georg Lehmann
2023-02-22 17:31:06 +01:00
committed by Marge Bot
parent 3a0bc8f007
commit 177dba62a1
3 changed files with 47 additions and 21 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/amd/compiler/tests/test_optimizer.cpp
+4 -4
View File
@@ -1174,12 +1174,12 @@ BEGIN_TEST(optimize.casts)
writeout(2, fmul(u2u16(bld.vop2_e64(aco_opcode::v_mul_f32, bld.def(v1), Operand::c32(0xbf800000u), bld.as_uniform(a16))), a16));
//! v1: %res3_tmp = v_mul_f32 %a, %a
//! v2b: %res3 = v_med3_f16 0, 1.0, %res3_tmp
//! v2b: %res3 = v_add_f16 %res3_tmp, 0 clamp
//! p_unit_test 3, %res3
writeout(3, fsat(u2u16(fmul(a, a))));
//! v2b: %res4_tmp = v_mul_f16 %a16, %a16
//! v1: %res4 = v_med3_f32 0, 1.0, %res4_tmp
//! v1: %res4 = v_add_f32 %res4_tmp, 0 clamp
//! p_unit_test 4, %res4
writeout(4, fsat(bld.as_uniform(fmul(a16, a16))));
@@ -1701,12 +1701,12 @@ BEGIN_TEST(optimize.mad_mix.cast)
writeout(3, f2f32(u2u16(fmul(a, a))));
//! v1: %res4_mul = v_fma_mix_f32 lo(%a16), %a, -0
//! v2b: %res4 = v_med3_f16 0, 1.0, %res4_mul
//! v2b: %res4 = v_add_f16 %res4_mul, 0 clamp
//! p_unit_test 4, %res4
writeout(4, fsat(u2u16(fmul(f2f32(a16), a))));
//! v2b: %res5_mul = v_fma_mixlo_f16 %a, %a, -0
//! v1: %res5 = v_med3_f32 0, 1.0, %res5_mul
//! v1: %res5 = v_add_f32 %res5_mul, 0 clamp
//! p_unit_test 5, %res5
writeout(5, fsat(bld.as_uniform(f2f16(fmul(a, a)))));