fdo-mirrors/mesa
1
0
Fork 0
mirror of https://gitlab.freedesktop.org/mesa/mesa.git synced 2026-05-04 20:38:06 +02:00
Code Issues Projects Releases Packages Wiki Activity Actions 2

aco: always optimize v_mad to v_madak in presence of literals

Browse source 
v_mad and v_madak are both 64-bit instructions, so it doesn't
increase code size to always apply a 32-bit literal instead of
using v_mad and a sgpr which contains that literal.

Found with some Youngblood shaders but help some other games.

vkpipeline-db (VEGA10):
Totals from affected shaders:
SGPRS: 46168 -> 46016 (-0.33 %)
VGPRS: 45576 -> 45564 (-0.03 %)
Code Size: 5187208 -> 5179584 (-0.15 %) bytes
Max Waves: 3297 -> 3297 (0.00 %)

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4410>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4410>
This commit is contained in:
Samuel Pitoiset 2020-04-01 18:09:43 +02:00 • committed by Marge Bot
parent 63b4fcba33
commit c953292630
1 changed files with 11 additions and 2 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

13
src/amd/compiler/aco_optimizer.cpp
View file

@ -2633,7 +2633,15 @@ void select_instruction(opt_ctx &ctx, aco_ptr<Instruction>& instr)
literal_idx = i;
}
}
if (literal_uses < threshold) {
/* Limit the number of literals to apply to not increase the code
* size too much, but always apply literals for v_mad->v_madak
* because both instructions are 64-bit and this doesn't increase
* code size.
* TODO: try to apply the literals earlier to lower the number of
* uses below threshold
*/
if (literal_uses < threshold || literal_idx == 2) {
ctx.uses[instr->operands[literal_idx].tempId()]--;
mad_info->check_literal = true;
mad_info->literal_idx = literal_idx;
@ -2760,7 +2768,8 @@ void apply_literals(opt_ctx &ctx, aco_ptr<Instruction>& instr)
/* apply literals on MAD */
if (instr->opcode == aco_opcode::v_mad_f32 && ctx.info[instr->definitions[0].tempId()].is_mad()) {
mad_info* info = &ctx.mad_infos[ctx.info[instr->definitions[0].tempId()].val];
if (info->check_literal && ctx.uses[instr->operands[info->literal_idx].tempId()] == 0) {
if (info->check_literal &&
(ctx.uses[instr->operands[info->literal_idx].tempId()] == 0 || info->literal_idx == 2)) {
aco_ptr<Instruction> new_mad;
if (info->literal_idx == 2) { /* add literal -> madak */
new_mad.reset(create_instruction<VOP2_instruction>(aco_opcode::v_madak_f32, Format::VOP2, 3, 1));