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

aco: support v_fma_f32_dpp as fma_mix

Browse Source 
Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25220>
This commit is contained in:
Georg Lehmann
2023-09-14 13:15:39 +02:00
committed by Marge Bot
parent 5e9fad48bf
commit 7d7657ef74
1 changed files with 17 additions and 10 deletions
Download Patch File Download Diff File Expand all files Collapse all files

27
src/amd/compiler/aco_optimizer.cpp
View File

@@ -3929,28 +3929,34 @@ can_use_mad_mix(opt_ctx& ctx, aco_ptr<Instruction>& instr)
if (ctx.program->gfx_level == GFX9 && ctx.fp_mode.denorm16_64) if (ctx.program->gfx_level == GFX9 && ctx.fp_mode.denorm16_64)
return false; return false;
if (instr->valu().omod)
return false;
switch (instr->opcode) { switch (instr->opcode) {
case aco_opcode::v_add_f32: case aco_opcode::v_add_f32:
case aco_opcode::v_sub_f32: case aco_opcode::v_sub_f32:
case aco_opcode::v_subrev_f32: case aco_opcode::v_subrev_f32:
case aco_opcode::v_mul_f32: case aco_opcode::v_mul_f32: return !instr->isSDWA() && !instr->isDPP();
case aco_opcode::v_fma_f32: break; case aco_opcode::v_fma_f32:
return ctx.program->dev.fused_mad_mix || !instr->definitions[0].isPrecise();
case aco_opcode::v_fma_mix_f32: case aco_opcode::v_fma_mix_f32:
case aco_opcode::v_fma_mixlo_f16: return true; case aco_opcode::v_fma_mixlo_f16: return true;
default: return false; default: return false;
} }
if (instr->opcode == aco_opcode::v_fma_f32 && !ctx.program->dev.fused_mad_mix &&
instr->definitions[0].isPrecise())
return false;
return !instr->valu().omod && !instr->isSDWA() && !instr->isDPP();
} }
void void
to_mad_mix(opt_ctx& ctx, aco_ptr<Instruction>& instr) to_mad_mix(opt_ctx& ctx, aco_ptr<Instruction>& instr)
{ {
bool is_add = instr->opcode != aco_opcode::v_mul_f32 && instr->opcode != aco_opcode::v_fma_f32; ctx.info[instr->definitions[0].tempId()].label &= label_f2f16 | label_clamp | label_mul;
if (instr->opcode == aco_opcode::v_fma_f32) {
instr->format = (Format)((uint32_t)withoutVOP3(instr->format) | (uint32_t)(Format::VOP3P));
instr->opcode = aco_opcode::v_fma_mix_f32;
return;
}
bool is_add = instr->opcode != aco_opcode::v_mul_f32;
aco_ptr<VALU_instruction> vop3p{ aco_ptr<VALU_instruction> vop3p{
create_instruction<VALU_instruction>(aco_opcode::v_fma_mix_f32, Format::VOP3P, 3, 1)}; create_instruction<VALU_instruction>(aco_opcode::v_fma_mix_f32, Format::VOP3P, 3, 1)};
@@ -3975,7 +3981,6 @@ to_mad_mix(opt_ctx& ctx, aco_ptr<Instruction>& instr)
vop3p->pass_flags = instr->pass_flags; vop3p->pass_flags = instr->pass_flags;
instr = std::move(vop3p); instr = std::move(vop3p);
ctx.info[instr->definitions[0].tempId()].label &= label_f2f16 | label_clamp | label_mul;
if (ctx.info[instr->definitions[0].tempId()].label & label_mul) if (ctx.info[instr->definitions[0].tempId()].label & label_mul)
ctx.info[instr->definitions[0].tempId()].instr = instr.get(); ctx.info[instr->definitions[0].tempId()].instr = instr.get();
} }
@@ -4044,6 +4049,8 @@ combine_mad_mix(opt_ctx& ctx, aco_ptr<Instruction>& instr)
op[i] = conv->operands[0]; op[i] = conv->operands[0];
if (!check_vop3_operands(ctx, instr->operands.size(), op)) if (!check_vop3_operands(ctx, instr->operands.size(), op))
continue; continue;
if (!conv->operands[0].isOfType(RegType::vgpr) && instr->isDPP())
continue;
if (!instr->isVOP3P()) { if (!instr->isVOP3P()) {
bool is_add = bool is_add =