fdo-mirrors/mesa
1
0
Fork 0
mirror of https://gitlab.freedesktop.org/mesa/mesa.git synced 2025-12-24 21:50:12 +01:00
Code Issues Projects Releases Packages Wiki Activity Actions 1

aco: change order in combine_minmax()

Browse source 
Prepare for future optimizations.

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19933>
This commit is contained in:
Rhys Perry 2022-11-16 18:10:38 +00:00 • committed by Marge Bot
parent ce5838599d
commit dfbc8e0192
2 changed files with 7 additions and 7 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

10
src/amd/compiler/aco_optimizer.cpp
View file

@ -2625,22 +2625,22 @@ bool
combine_minmax(opt_ctx& ctx, aco_ptr<Instruction>& instr, aco_opcode opposite, aco_opcode minmax3)
{
/* TODO: this can handle SDWA min/max instructions by using opsel */
if (combine_three_valu_op(ctx, instr, instr->opcode, minmax3, "012", 1 | 2))
if (combine_three_valu_op(ctx, instr, instr->opcode, minmax3, "120", 1 | 2))
return true;
/* min(-max(a, b), c) -> min3(c, -a, -b) *
* max(-min(a, b), c) -> max3(c, -a, -b) */
/* min(-max(a, b), c) -> min3(-a, -b, c) *
* max(-min(a, b), c) -> max3(-a, -b, c) */
for (unsigned swap = 0; swap < 2; swap++) {
Operand operands[3];
bool neg[3], abs[3], clamp, precise;
uint8_t opsel = 0, omod = 0;
bool inbetween_neg;
if (match_op3_for_vop3(ctx, instr->opcode, opposite, instr.get(), swap, "012", operands, neg,
if (match_op3_for_vop3(ctx, instr->opcode, opposite, instr.get(), swap, "120", operands, neg,
abs, &opsel, &clamp, &omod, &inbetween_neg, NULL, NULL, &precise) &&
inbetween_neg) {
ctx.uses[instr->operands[swap].tempId()]--;
neg[0] = !neg[0];
neg[1] = !neg[1];
neg[2] = !neg[2];
create_vop3_for_op3(ctx, minmax3, instr, operands, neg, abs, opsel, clamp, omod);
return true;
}

4
src/amd/compiler/tests/test_optimizer.cpp
View file

@ -718,14 +718,14 @@ BEGIN_TEST(optimize.minmax)
if (!setup_cs("v1", (amd_gfx_level)i))
continue;
//! v1: %res0 = v_max3_f32 0, -0, %a
//! v1: %res0 = v_max3_f32 -0, %a, 0
//! p_unit_test 0, %res0
Temp xor0 = fneg(inputs[0]);
Temp min = bld.vop2(aco_opcode::v_min_f32, bld.def(v1), Operand::zero(), xor0);
Temp xor1 = fneg(min);
writeout(0, bld.vop2(aco_opcode::v_max_f32, bld.def(v1), Operand::zero(), xor1));
//! v1: %res1 = v_max3_f32 0, -0, -%a
//! v1: %res1 = v_max3_f32 -0, -%a, 0
//! p_unit_test 1, %res1
min = bld.vop2(aco_opcode::v_min_f32, bld.def(v1), Operand::zero(), Operand(inputs[0]));
xor1 = fneg(min);