fdo-mirrors/mesa
1
0
Fork 0
mirror of https://gitlab.freedesktop.org/mesa/mesa.git synced 2026-01-04 02:40:11 +01:00
Code Issues Projects Releases Packages Wiki Activity Actions 1

aco: combine extracts into s_pack_ll_b32_b16

Browse source 
fossil-db (navi21):
Totals from 3 (0.00% of 79395) affected shaders:
Instrs: 45941 -> 45924 (-0.04%)
CodeSize: 241768 -> 241756 (-0.00%)
Latency: 176501 -> 176491 (-0.01%)
Copies: 6884 -> 6882 (-0.03%)
SALU: 6101 -> 6088 (-0.21%)

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Georg Lehmann <dadschoorse@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/29912>
This commit is contained in:
Rhys Perry 2024-06-06 15:25:13 +01:00 • committed by Marge Bot
parent 98cb50297b
commit ca161a96d1
2 changed files with 67 additions and 1 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

18
src/amd/compiler/aco_optimizer.cpp
View file

@ -1050,6 +1050,13 @@ can_apply_extract(opt_ctx& ctx, aco_ptr<Instruction>& instr, unsigned idx, ssa_i
} else if (instr->isVALU() && sel.size() == 2 && !instr->valu().opsel[idx] &&
can_use_opsel(ctx.program->gfx_level, instr->opcode, idx)) {
return true;
} else if (instr->opcode == aco_opcode::s_pack_ll_b32_b16 && sel.size() == 2 &&
(idx == 1 || ctx.program->gfx_level >= GFX11 || !sel.offset())) {
return true;
} else if (sel.size() == 2 &&
((instr->opcode == aco_opcode::s_pack_lh_b32_b16 && idx == 0) ||
(instr->opcode == aco_opcode::s_pack_hl_b32_b16 && idx == 1))) {
return true;
} else if (instr->opcode == aco_opcode::p_extract) {
SubdwordSel instrSel = parse_extract(instr.get());
@ -1124,6 +1131,13 @@ apply_extract(opt_ctx& ctx, aco_ptr<Instruction>& instr, unsigned idx, ssa_info&
!info.instr->operands[0].isOfType(RegType::vgpr))
instr->format = asVOP3(instr->format);
}
} else if (instr->opcode == aco_opcode::s_pack_ll_b32_b16) {
if (sel.offset())
instr->opcode = idx ? aco_opcode::s_pack_lh_b32_b16 : aco_opcode::s_pack_hl_b32_b16;
} else if (instr->opcode == aco_opcode::s_pack_lh_b32_b16 ||
instr->opcode == aco_opcode::s_pack_hl_b32_b16) {
if (sel.offset())
instr->opcode = aco_opcode::s_pack_hh_b32_b16;
} else if (instr->opcode == aco_opcode::p_extract) {
SubdwordSel instrSel = parse_extract(instr.get());
@ -3784,7 +3798,7 @@ combine_instruction(opt_ctx& ctx, aco_ptr<Instruction>& instr)
if (instr->definitions.empty() || is_dead(ctx.uses, instr.get()))
return;
if (instr->isVALU()) {
if (instr->isVALU() || instr->isSALU()) {
/* Apply SDWA. Do this after label_instruction() so it can remove
* label_extract if not all instructions can take SDWA. */
for (unsigned i = 0; i < instr->operands.size(); i++) {
@ -3811,7 +3825,9 @@ combine_instruction(opt_ctx& ctx, aco_ptr<Instruction>& instr)
instr->operands[i].setTemp(info.instr->operands[0].getTemp());
}
}
}
if (instr->isVALU()) {
if (can_apply_sgprs(ctx, instr))
apply_sgprs(ctx, instr);
combine_mad_mix(ctx, instr);

50
src/amd/compiler/tests/test_optimizer.cpp
View file

@ -1977,3 +1977,53 @@ BEGIN_TEST(optimize.vinterp_inreg_output_modifiers)
finish_opt_test();
END_TEST
BEGIN_TEST(optimize.s_pack)
//>> s1: %a, s1: %b, s1: %c = p_startpgm
if (!setup_cs("s1 s1 s1", GFX11))
return;
Temp lo = bld.pseudo(aco_opcode::p_extract, bld.def(s1), bld.def(s1, scc), inputs[1],
Operand::c32(0), Operand::c32(16u), Operand::c32(false));
Temp hi = bld.pseudo(aco_opcode::p_extract, bld.def(s1), bld.def(s1, scc), inputs[2],
Operand::c32(1), Operand::c32(16u), Operand::c32(false));
//! s1: %res0 = s_pack_lh_b32_b16 %b, %c
//! p_unit_test 0, %res0
writeout(0, bld.sop2(aco_opcode::s_pack_ll_b32_b16, bld.def(s1), lo, hi));
//! s1: %res1 = s_pack_ll_b32_b16 %b, %b
//! p_unit_test 1, %res1
writeout(1, bld.sop2(aco_opcode::s_pack_ll_b32_b16, bld.def(s1), lo, lo));
//! s1: %res2 = s_pack_hl_b32_b16 %c, %b
//! p_unit_test 2, %res2
writeout(2, bld.sop2(aco_opcode::s_pack_ll_b32_b16, bld.def(s1), hi, lo));
//! s1: %res3 = s_pack_hh_b32_b16 %c, %c
//! p_unit_test 3, %res3
writeout(3, bld.sop2(aco_opcode::s_pack_ll_b32_b16, bld.def(s1), hi, hi));
lo = bld.pseudo(aco_opcode::p_extract, bld.def(s1), bld.def(s1, scc), inputs[1], Operand::c32(0),
Operand::c32(16u), Operand::c32(false));
hi = bld.pseudo(aco_opcode::p_extract, bld.def(s1), bld.def(s1, scc), inputs[2], Operand::c32(1),
Operand::c32(16u), Operand::c32(false));
//! s1: %res4 = s_pack_ll_b32_b16 %a, %b
//! p_unit_test 4, %res4
writeout(4, bld.sop2(aco_opcode::s_pack_ll_b32_b16, bld.def(s1), inputs[0], lo));
//! s1: %res5 = s_pack_lh_b32_b16 %a, %c
//! p_unit_test 5, %res5
writeout(5, bld.sop2(aco_opcode::s_pack_ll_b32_b16, bld.def(s1), inputs[0], hi));
//! s1: %res6 = s_pack_ll_b32_b16 %b, %a
//! p_unit_test 6, %res6
writeout(6, bld.sop2(aco_opcode::s_pack_ll_b32_b16, bld.def(s1), lo, inputs[0]));
//! s1: %res7 = s_pack_hl_b32_b16 %c, %a
//! p_unit_test 7, %res7
writeout(7, bld.sop2(aco_opcode::s_pack_ll_b32_b16, bld.def(s1), hi, inputs[0]));
finish_opt_test();
END_TEST