fdo-mirrors/mesa
1
0
Fork 0
mirror of https://gitlab.freedesktop.org/mesa/mesa.git synced 2026-04-25 15:40:46 +02:00
Code Issues Projects Releases Packages Wiki Activity Actions 2

aco: introduce helper to swap valu operands with modifiers

Browse source 
Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23059>
This commit is contained in:
Georg Lehmann 2023-05-18 10:54:45 +02:00 • committed by Marge Bot
parent 28e2031e3a
commit 6a53af3fc8
4 changed files with 23 additions and 17 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

3
src/amd/compiler/aco_assembler.cpp
View file

@ -160,8 +160,7 @@ emit_instruction(asm_context& ctx, std::vector<uint32_t>& out, Instruction* inst
instr->opcode = aco_opcode::v_fma_f16;
instr->format = (Format)((uint32_t)instr->format & ~(uint32_t)Format::VOP2);
} else if (instr->opcode == aco_opcode::v_fmamk_f16) {
std::swap(instr->operands[1], instr->operands[2]);
instr->valu().opsel[1].swap(instr->valu().opsel[2]);
instr->valu().swapOperands(1, 2);
instr->opcode = aco_opcode::v_fma_f16;
instr->format = (Format)((uint32_t)instr->format & ~(uint32_t)Format::VOP2);
}

18
src/amd/compiler/aco_ir.h
View file

@ -1402,6 +1402,8 @@ struct VALU_instruction : public Instruction {
bitfield_array8<uint32_t, 15, 3> opsel_hi; /* VOP3P */
bitfield_bool<uint32_t, 18> clamp; /* VOP3, VOP3P, SDWA, VINTERP_inreg */
};
void swapOperands(unsigned idx0, unsigned idx1);
};
static_assert(sizeof(VALU_instruction) == sizeof(Instruction) + 4, "Unexpected padding");
@ -1718,6 +1720,22 @@ struct Pseudo_reduction_instruction : public Instruction {
static_assert(sizeof(Pseudo_reduction_instruction) == sizeof(Instruction) + 4,
"Unexpected padding");
inline void
VALU_instruction::swapOperands(unsigned idx0, unsigned idx1)
{
if (this->isSDWA() && idx0 != idx1) {
assert(idx0 < 2 && idx1 < 2);
std::swap(this->sdwa().sel[0], this->sdwa().sel[1]);
}
assert(idx0 < 3 && idx1 < 3);
std::swap(this->operands[idx0], this->operands[idx1]);
this->neg[idx0].swap(this->neg[idx1]);
this->abs[idx0].swap(this->abs[idx1]);
this->opsel[idx0].swap(this->opsel[idx1]);
this->opsel_lo[idx0].swap(this->opsel_lo[idx1]);
this->opsel_hi[idx0].swap(this->opsel_hi[idx1]);
}
extern thread_local aco::monotonic_buffer_resource* instruction_buffer;
struct instr_deleter_functor {

12
src/amd/compiler/aco_optimizer.cpp
View file

@ -1429,9 +1429,8 @@ label_instruction(opt_ctx& ctx, aco_ptr<Instruction>& instr)
instr->operands[i] = op;
continue;
} else if (!instr->isVOP3() && can_swap_operands(instr, &instr->opcode)) {
instr->operands[i] = instr->operands[0];
instr->operands[0] = op;
instr->valu().opsel[0].swap(instr->valu().opsel[i]);
instr->operands[i] = op;
instr->valu().swapOperands(0, i);
continue;
} else if (can_use_VOP3(ctx, instr)) {
instr->format = asVOP3(instr->format);
@ -4829,12 +4828,7 @@ select_instruction(opt_ctx& ctx, aco_ptr<Instruction>& instr)
if (i != 0) {
if (!can_swap_operands(instr, &instr->opcode, 0, i))
continue;
std::swap(instr->operands[0], instr->operands[i]);
instr->valu().neg[0].swap(instr->valu().neg[i]);
instr->valu().abs[0].swap(instr->valu().abs[i]);
instr->valu().opsel[0].swap(instr->valu().opsel[i]);
instr->valu().opsel_lo[0].swap(instr->valu().opsel_lo[i]);
instr->valu().opsel_hi[0].swap(instr->valu().opsel_hi[i]);
instr->valu().swapOperands(0, i);
}
if (!can_use_DPP(ctx.program->gfx_level, instr, info.is_dpp8()))

7
src/amd/compiler/aco_optimizer_postRA.cpp
View file

@ -522,12 +522,7 @@ try_combine_dpp(pr_opt_ctx& ctx, aco_ptr<Instruction>& instr)
if (i != 0) {
if (!can_swap_operands(instr, &instr->opcode, 0, i))
continue;
std::swap(instr->operands[0], instr->operands[i]);
instr->valu().neg[0].swap(instr->valu().neg[i]);
instr->valu().abs[0].swap(instr->valu().abs[i]);
instr->valu().opsel[0].swap(instr->valu().opsel[i]);
instr->valu().opsel_lo[0].swap(instr->valu().opsel_lo[i]);
instr->valu().opsel_hi[0].swap(instr->valu().opsel_hi[i]);
instr->valu().swapOperands(0, i);
}
if (!can_use_DPP(ctx.program->gfx_level, instr, dpp8))