fdo-mirrors/mesa
1
0
Fork 0
mirror of https://gitlab.freedesktop.org/mesa/mesa.git synced 2026-05-08 15:38:09 +02:00
Code Issues Projects Releases Packages Wiki Activity Actions 1

aco: fix combining max(-min(a, b), c) if a or b uses the neg modifier

Browse source 
No fossils-db changes.

Cc: 20.2, 20.3
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7657>
(cherry picked from commit 0fcd379184)
This commit is contained in:
Samuel Pitoiset 2020-11-17 17:14:49 +01:00 • committed by Dylan Baker
parent 60e001e95f
commit 4c5d644bdb
2 changed files with 27 additions and 4 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

8
src/amd/compiler/aco_optimizer.cpp
View file

@ -2023,8 +2023,8 @@ bool combine_minmax(opt_ctx& ctx, aco_ptr<Instruction>& instr, aco_opcode opposi
uint64_t omod_clamp = ctx.info[instr->definitions[0].tempId()].label &
(label_omod_success | label_clamp_success);
/* min(-max(a, b), c) -> min3(-a, -b, c) *
* max(-min(a, b), c) -> max3(-a, -b, c) */
/* min(-max(a, b), c) -> min3(c, -a, -b) *
* max(-min(a, b), c) -> max3(c, -a, -b) */
for (unsigned swap = 0; swap < 2; swap++) {
Operand operands[3];
bool neg[3], abs[3], clamp, precise;
@ -2036,8 +2036,8 @@ bool combine_minmax(opt_ctx& ctx, aco_ptr<Instruction>& instr, aco_opcode opposi
&clamp, &omod, &inbetween_neg, NULL, NULL, &precise) &&
inbetween_neg) {
ctx.uses[instr->operands[swap].tempId()]--;
neg[1] = true;
neg[2] = true;
neg[1] = !neg[1];
neg[2] = !neg[2];
create_vop3_for_op3(ctx, minmax3, instr, operands, neg, abs, opsel, clamp, omod);
if (omod_clamp & label_omod_success)
ctx.info[instr->definitions[0].tempId()].set_omod_success(instr.get());

23
src/amd/compiler/tests/test_optimizer.cpp
View file

@ -209,3 +209,26 @@ BEGIN_TEST(optimize.add3)
finish_opt_test();
END_TEST
BEGIN_TEST(optimize.minmax)
for (unsigned i = GFX8; i <= GFX10; i++) {
//>> v1: %a, s2: %_:exec = p_startpgm
if (!setup_cs("v1", (chip_class)i))
continue;
//! v1: %res0 = v_max3_f32 0, -0, %a
//! p_unit_test 0, %res0
Temp xor0 = bld.vop2(aco_opcode::v_xor_b32, bld.def(v1), Operand(0x80000000u), Operand(inputs[0]));
Temp min = bld.vop2(aco_opcode::v_min_f32, bld.def(v1), Operand(0u), xor0);
Temp xor1 = bld.vop2(aco_opcode::v_xor_b32, bld.def(v1), Operand(0x80000000u), min);
writeout(0, bld.vop2(aco_opcode::v_max_f32, bld.def(v1), Operand(0u), xor1));
//! v1: %res1 = v_max3_f32 0, -0, -%a
//! p_unit_test 1, %res1
min = bld.vop2(aco_opcode::v_min_f32, bld.def(v1), Operand(0u), Operand(inputs[0]));
xor1 = bld.vop2(aco_opcode::v_xor_b32, bld.def(v1), Operand(0x80000000u), min);
writeout(1, bld.vop2(aco_opcode::v_max_f32, bld.def(v1), Operand(0u), xor1));
finish_opt_test();
}
END_TEST