fdo-mirrors/mesa
1
0
Fork 0
mirror of https://gitlab.freedesktop.org/mesa/mesa.git synced 2026-05-08 09:08:10 +02:00
Code Issues Projects Releases Packages Wiki Activity Actions 1

aco: disallow various v_add_u32 opts if modifiers are used

Browse source 
Check for clamp, SDWA or DPP. The optimization isn't possible with SDWA
and DPP, so it would have been skipped anyway. Doing any of these with a
clamp modifier present would be incorrect.

No fossil-db changes.

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Cc: mesa-stable
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7045>
This commit is contained in:
Rhys Perry 2020-10-07 11:45:30 +01:00 • committed by Dylan Baker
parent 7ea1f6ff78
commit 944bdf259c
3 changed files with 30 additions and 2 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

2
.pick_status.json
View file

@ -8167,7 +8167,7 @@
"description": "aco: disallow various v_add_u32 opts if modifiers are used",
"nominated": true,
"nomination_type": 0,
"resolution": 0,
"resolution": 3,
"master_sha": null,
"because_sha": null
},

2
src/amd/compiler/aco_optimizer.cpp
View file

@ -2783,7 +2783,7 @@ void combine_instruction(opt_ctx &ctx, Block& block, aco_ptr<Instruction>& instr
else combine_three_valu_op(ctx, instr, aco_opcode::s_xor_b32, aco_opcode::v_xor3_b32, "012", 1 | 2);
} else if (instr->opcode == aco_opcode::v_add_u32) {
if (combine_add_sub_b2i(ctx, instr, aco_opcode::v_addc_co_u32, 1 | 2)) ;
else if (ctx.program->chip_class >= GFX9) {
else if (ctx.program->chip_class >= GFX9 && !instr->usesModifiers()) {
if (combine_three_valu_op(ctx, instr, aco_opcode::s_xor_b32, aco_opcode::v_xad_u32, "120", 1 | 2)) ;
else if (combine_three_valu_op(ctx, instr, aco_opcode::v_xor_b32, aco_opcode::v_xad_u32, "120", 1 | 2)) ;
else if (combine_three_valu_op(ctx, instr, aco_opcode::s_add_i32, aco_opcode::v_add3_u32, "012", 1 | 2)) ;

28
src/amd/compiler/tests/test_optimizer.cpp
View file

@ -181,3 +181,31 @@ BEGIN_TEST(optimize.const_comparison_ordering)
finish_opt_test();
END_TEST
BEGIN_TEST(optimize.add3)
//>> v1: %a, v1: %b, v1: %c, s2: %_:exec = p_startpgm
if (!setup_cs("v1 v1 v1", GFX9))
return;
//! v1: %res0 = v_add3_u32 %a, %b, %c
//! p_unit_test 0, %res0
Builder::Result tmp = bld.vop2(aco_opcode::v_add_u32, bld.def(v1), inputs[1], inputs[2]);
writeout(0, bld.vop2(aco_opcode::v_add_u32, bld.def(v1), inputs[0], tmp));
//! v1: %tmp1 = v_add_u32 %b, %c clamp
//! v1: %res1 = v_add_u32 %a, %tmp1
//! p_unit_test 1, %res1
tmp = bld.vop2_e64(aco_opcode::v_add_u32, bld.def(v1), inputs[1], inputs[2]);
static_cast<VOP3A_instruction *>(tmp.instr)->clamp = true;
writeout(1, bld.vop2(aco_opcode::v_add_u32, bld.def(v1), inputs[0], tmp));
//! v1: %tmp2 = v_add_u32 %b, %c
//! v1: %res2 = v_add_u32 %a, %tmp2 clamp
//! p_unit_test 2, %res2
tmp = bld.vop2(aco_opcode::v_add_u32, bld.def(v1), inputs[1], inputs[2]);
tmp = bld.vop2_e64(aco_opcode::v_add_u32, bld.def(v1), inputs[0], tmp);
static_cast<VOP3A_instruction *>(tmp.instr)->clamp = true;
writeout(2, tmp);
finish_opt_test();
END_TEST