OpenHarmony/third_party_mesa3d
2
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

aco: handle opsel in combine_constant_comparison_ordering

Browse Source 
Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22069>
This commit is contained in:
Georg Lehmann
2023-03-28 21:48:04 +02:00
committed by Marge Bot
parent d8f07a0ddc
commit 0896ecec9a
1 changed files with 14 additions and 25 deletions
Download Patch File Download Diff File Expand all files Collapse all files

35
src/amd/compiler/aco_optimizer.cpp
View File

@@ -2472,9 +2472,8 @@ combine_constant_comparison_ordering(opt_ctx& ctx, aco_ptr<Instruction>& instr)
Instruction* nan_test = follow_operand(ctx, instr->operands[0], true); Instruction* nan_test = follow_operand(ctx, instr->operands[0], true);
Instruction* cmp = follow_operand(ctx, instr->operands[1], true); Instruction* cmp = follow_operand(ctx, instr->operands[1], true);
if (!nan_test || !cmp || nan_test->isSDWA() || cmp->isSDWA()) if (!nan_test || !cmp || nan_test->isSDWA() || cmp->isSDWA() || nan_test->isDPP() ||
return false; cmp->isDPP())
if (nan_test->isSDWA() || cmp->isSDWA())
return false; return false;
aco_opcode expected_nan_test = is_or ? aco_opcode::v_cmp_neq_f32 : aco_opcode::v_cmp_eq_f32; aco_opcode expected_nan_test = is_or ? aco_opcode::v_cmp_neq_f32 : aco_opcode::v_cmp_eq_f32;
@@ -2497,17 +2496,15 @@ combine_constant_comparison_ordering(opt_ctx& ctx, aco_ptr<Instruction>& instr)
if (prop_nan0 != prop_nan1) if (prop_nan0 != prop_nan1)
return false; return false;
if (nan_test->isVOP3()) {
VALU_instruction& vop3 = nan_test->valu(); VALU_instruction& vop3 = nan_test->valu();
if (vop3.neg[0] != vop3.neg[1] || vop3.abs[0] != vop3.abs[1] || if (vop3.neg[0] != vop3.neg[1] || vop3.abs[0] != vop3.abs[1] || vop3.opsel[0] != vop3.opsel[1])
vop3.opsel[0] != vop3.opsel[1])
return false; return false;
}
int constant_operand = -1; int constant_operand = -1;
for (unsigned i = 0; i < 2; i++) { for (unsigned i = 0; i < 2; i++) {
if (cmp->operands[i].isTemp() && if (cmp->operands[i].isTemp() &&
original_temp_id(ctx, cmp->operands[i].getTemp()) == prop_nan0) { original_temp_id(ctx, cmp->operands[i].getTemp()) == prop_nan0 &&
cmp->valu().opsel[i] == nan_test->valu().opsel[0]) {
constant_operand = !i; constant_operand = !i;
break; break;
} }
@@ -2518,24 +2515,16 @@ combine_constant_comparison_ordering(opt_ctx& ctx, aco_ptr<Instruction>& instr)
uint64_t constant_value; uint64_t constant_value;
if (!is_operand_constant(ctx, cmp->operands[constant_operand], bit_size, &constant_value)) if (!is_operand_constant(ctx, cmp->operands[constant_operand], bit_size, &constant_value))
return false; return false;
if (is_constant_nan(constant_value, bit_size)) if (is_constant_nan(constant_value >> (cmp->valu().opsel[constant_operand] * 16), bit_size))
return false; return false;
aco_opcode new_op = is_or ? get_unordered(cmp->opcode) : get_ordered(cmp->opcode); aco_opcode new_op = is_or ? get_unordered(cmp->opcode) : get_ordered(cmp->opcode);
Instruction* new_instr; Instruction* new_instr = create_instruction<VALU_instruction>(new_op, cmp->format, 2, 1);
if (cmp->isVOP3()) { new_instr->valu().neg = cmp->valu().neg;
VALU_instruction* new_vop3 = new_instr->valu().abs = cmp->valu().abs;
create_instruction<VALU_instruction>(new_op, asVOP3(Format::VOPC), 2, 1); new_instr->valu().clamp = cmp->valu().clamp;
VALU_instruction& cmp_vop3 = cmp->valu(); new_instr->valu().omod = cmp->valu().omod;
new_vop3->neg = cmp_vop3.neg; new_instr->valu().opsel = cmp->valu().opsel;
new_vop3->abs = cmp_vop3.abs;
new_vop3->clamp = cmp_vop3.clamp;
new_vop3->omod = cmp_vop3.omod;
new_vop3->opsel = cmp_vop3.opsel;
new_instr = new_vop3;
} else {
new_instr = create_instruction<VALU_instruction>(new_op, Format::VOPC, 2, 1);
}
new_instr->operands[0] = copy_operand(ctx, cmp->operands[0]); new_instr->operands[0] = copy_operand(ctx, cmp->operands[0]);
new_instr->operands[1] = copy_operand(ctx, cmp->operands[1]); new_instr->operands[1] = copy_operand(ctx, cmp->operands[1]);
new_instr->definitions[0] = instr->definitions[0]; new_instr->definitions[0] = instr->definitions[0];