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

aco: make p_wqm a marker instruction without Operands/Definitions

Browse Source 
Totals from 28277 (36.93% of 76572) affected shaders: (GFX11)

MaxWaves: 833930 -> 833898 (-0.00%); split: +0.01%, -0.01%
Instrs: 21366950 -> 21353346 (-0.06%); split: -0.11%, +0.05%
CodeSize: 112855368 -> 112610508 (-0.22%); split: -0.24%, +0.03%
VGPRs: 1157748 -> 1158540 (+0.07%); split: -0.10%, +0.17%
SpillSGPRs: 2465 -> 2463 (-0.08%); split: -0.16%, +0.08%
Latency: 168339886 -> 168383646 (+0.03%); split: -0.10%, +0.12%
InvThroughput: 25164895 -> 25158376 (-0.03%); split: -0.08%, +0.06%
VClause: 347660 -> 346256 (-0.40%); split: -0.55%, +0.15%
SClause: 794460 -> 799521 (+0.64%); split: -0.33%, +0.97%
Copies: 1151908 -> 1148370 (-0.31%); split: -0.54%, +0.23%
Branches: 359447 -> 359437 (-0.00%); split: -0.01%, +0.00%
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25038>
This commit is contained in:
Daniel Schürmann
2023-09-02 10:30:02 +02:00
committed by Marge Bot
parent 1275981df8
commit 040142684c
5 changed files with 11 additions and 30 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/amd/compiler/aco_optimizer.cpp
-13
View File
@@ -662,7 +662,6 @@ alu_can_accept_constant(const aco_ptr<Instruction>& instr, unsigned operand)
case aco_opcode::v_cndmask_b32: return operand != 2;
case aco_opcode::s_addk_i32:
case aco_opcode::s_mulk_i32:
case aco_opcode::p_wqm:
case aco_opcode::p_extract_vector:
case aco_opcode::p_split_vector:
case aco_opcode::v_readlane_b32:
@@ -2071,11 +2070,6 @@ label_instruction(opt_ctx& ctx, aco_ptr<Instruction>& instr)
instr->operands[2].setTemp(ctx.info[instr->operands[2].tempId()].temp);
}
break;
case aco_opcode::p_wqm:
if (instr->operands[0].isTemp() && ctx.info[instr->operands[0].tempId()].is_scc_invert()) {
ctx.info[instr->definitions[0].tempId()].set_temp(instr->operands[0].getTemp());
}
break;
case aco_opcode::s_mul_i32:
/* Testing every uint32_t shows that 0x3f800000*n is never a denormal.
* This pattern is created from a uniform nir_op_b2f. */
@@ -4787,13 +4781,6 @@ select_instruction(opt_ctx& ctx, aco_ptr<Instruction>& instr)
instr->opcode == aco_opcode::s_cselect_b32) &&
instr->operands[2].isTemp()) {
ctx.info[instr->operands[2].tempId()].set_scc_needed();
} else if (instr->opcode == aco_opcode::p_wqm && instr->operands[0].isTemp() &&
ctx.info[instr->definitions[0].tempId()].is_scc_needed()) {
/* Propagate label so it is correctly detected by the uniform bool transform */
ctx.info[instr->operands[0].tempId()].set_scc_needed();
/* Fix definition to SCC, this will prevent RA from adding superfluous moves */
instr->definitions[0].setFixed(scc);
}
/* check for literals */