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

aco/optimizer: do can_eliminate_and_exec() optimization later

Browse Source 
This will allow to optimize s_not(v_cmp()) safely.

Totals from 1024 (0.76% of 134913) affected shaders: (GFX10.3)
CodeSize: 5695424 -> 5701860 (+0.11%); split: -0.00%, +0.11%
Scratch: 242688 -> 241664 (-0.42%)
Instrs: 1040656 -> 1041635 (+0.09%); split: -0.00%, +0.09%
Latency: 16842282 -> 16922790 (+0.48%); split: -0.06%, +0.54%
InvThroughput: 4772728 -> 4810868 (+0.80%); split: -0.10%, +0.90%
VClause: 20013 -> 20000 (-0.06%); split: -0.12%, +0.05%
Copies: 115057 -> 114384 (-0.58%); split: -1.22%, +0.63%
Branches: 34531 -> 34532 (+0.00%); split: -0.00%, +0.01%
PreSGPRs: 46263 -> 46267 (+0.01%)

Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18253>
This commit is contained in:
Daniel Schürmann
2022-08-24 12:25:23 +02:00
committed by Marge Bot
parent 14b15b4038
commit 79a8e8b5b2
1 changed files with 14 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/amd/compiler/aco_optimizer.cpp
+14 -3
View File
@@ -1939,9 +1939,6 @@ label_instruction(opt_ctx& ctx, aco_ptr<Instruction>& instr)
* s_and is unnecessary. */
ctx.info[instr->definitions[0].tempId()].set_temp(instr->operands[0].getTemp());
break;
} else if (can_eliminate_and_exec(ctx, instr->operands[0].getTemp(), instr->pass_flags)) {
ctx.info[instr->definitions[0].tempId()].set_temp(instr->operands[0].getTemp());
break;
}
}
FALLTHROUGH;
@@ -4489,6 +4486,20 @@ select_instruction(opt_ctx& ctx, aco_ptr<Instruction>& instr)
return;
}
/* This optimization is done late in order to be able to apply otherwise
* unsafe optimizations such as the inverse comparison optimization.
*/
if (instr->opcode == aco_opcode::s_and_b32 || instr->opcode == aco_opcode::s_and_b64) {
if (instr->operands[0].isTemp() && fixed_to_exec(instr->operands[1]) &&
ctx.uses[instr->operands[0].tempId()] == 1 && ctx.uses[instr->definitions[1].tempId()] == 0 &&
can_eliminate_and_exec(ctx, instr->operands[0].getTemp(), instr->pass_flags)) {
ctx.uses[instr->operands[0].tempId()]--;
ctx.info[instr->operands[0].tempId()].instr->definitions[0].setTemp(instr->definitions[0].getTemp());
instr.reset();
return;
}
}
/* Combine DPP copies into VALU. This should be done after creating MAD/FMA. */
if (instr->isVALU()) {
for (unsigned i = 0; i < instr->operands.size(); i++) {