fdo-mirrors/mesa
1
0
Fork 0
mirror of https://gitlab.freedesktop.org/mesa/mesa.git synced 2025-12-21 07:10:09 +01:00
Code Issues Projects Releases Packages Wiki Activity Actions 1

aco: rewrite apply_sgprs()

Browse source 
This will make it easier to apply two different sgprs (for GFX10) or apply
the same sgpr twice (just remove the break).

No pipeline-db changes.

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/2883>
This commit is contained in:
Rhys Perry 2019-11-22 14:55:25 +00:00 • committed by Marge Bot
parent 0be7409069
commit cef7879719
1 changed files with 51 additions and 36 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

87
src/amd/compiler/aco_optimizer.cpp
View file

@ -1868,61 +1868,76 @@ bool combine_clamp(opt_ctx& ctx, aco_ptr<Instruction>& instr,
void apply_sgprs(opt_ctx &ctx, aco_ptr<Instruction>& instr) void apply_sgprs(opt_ctx &ctx, aco_ptr<Instruction>& instr)
{ {
/* apply sgprs */ /* find candidates and create the set of sgprs already read */
uint32_t sgpr_idx = 0; unsigned sgpr_ids[2] = {0, 0};
uint32_t sgpr_info_id = 0; uint32_t operand_mask = 0;
bool has_sgpr = false; bool has_literal = false;
uint32_t sgpr_ssa_id = 0; for (unsigned i = 0; i < instr->operands.size(); i++) {
/* find 'best' possible sgpr */ if (instr->operands[i].isLiteral())
for (unsigned i = 0; i < instr->operands.size(); i++) has_literal = true;
{
if (instr->operands[i].isLiteral()) {
has_sgpr = true;
break;
}
if (!instr->operands[i].isTemp()) if (!instr->operands[i].isTemp())
continue; continue;
if (instr->operands[i].getTemp().type() == RegType::sgpr) { if (instr->operands[i].getTemp().type() == RegType::sgpr) {
has_sgpr = true; if (instr->operands[i].tempId() != sgpr_ids[0])
sgpr_ssa_id = instr->operands[i].tempId(); sgpr_ids[!!sgpr_ids[0]] = instr->operands[i].tempId();
continue;
} }
ssa_info& info = ctx.info[instr->operands[i].tempId()]; ssa_info& info = ctx.info[instr->operands[i].tempId()];
if (info.is_temp() && info.temp.type() == RegType::sgpr) { if (info.is_temp() && info.temp.type() == RegType::sgpr)
operand_mask |= 1u << i;
}
unsigned max_sgprs = 1;
if (has_literal)
max_sgprs--;
unsigned num_sgprs = !!sgpr_ids[0] + !!sgpr_ids[1];
/* keep on applying sgprs until there is nothing left to be done */
while (operand_mask) {
uint32_t sgpr_idx = 0;
uint32_t sgpr_info_id = 0;
uint32_t mask = operand_mask;
/* choose a sgpr */
while (mask) {
unsigned i = u_bit_scan(&mask);
uint16_t uses = ctx.uses[instr->operands[i].tempId()]; uint16_t uses = ctx.uses[instr->operands[i].tempId()];
if (sgpr_info_id == 0 || uses < ctx.uses[sgpr_info_id]) { if (sgpr_info_id == 0 || uses < ctx.uses[sgpr_info_id]) {
sgpr_idx = i; sgpr_idx = i;
sgpr_info_id = instr->operands[i].tempId(); sgpr_info_id = instr->operands[i].tempId();
} }
} }
} operand_mask &= ~(1u << sgpr_idx);
if (!has_sgpr && sgpr_info_id != 0) {
ssa_info& info = ctx.info[sgpr_info_id]; /* Applying two sgprs require making it VOP3, so don't do it unless it's
* definitively beneficial.
* TODO: this is too conservative because later the use count could be reduced to 1 */
if (num_sgprs && ctx.uses[sgpr_info_id] > 1)
break;
Temp sgpr = ctx.info[sgpr_info_id].temp;
bool new_sgpr = sgpr.id() != sgpr_ids[0] && sgpr.id() != sgpr_ids[1];
if (new_sgpr && num_sgprs >= max_sgprs)
continue;
if (sgpr_idx == 0 || instr->isVOP3()) { if (sgpr_idx == 0 || instr->isVOP3()) {
instr->operands[sgpr_idx] = Operand(info.temp); instr->operands[sgpr_idx] = Operand(sgpr);
ctx.uses[sgpr_info_id]--;
ctx.uses[info.temp.id()]++;
} else if (can_swap_operands(instr)) { } else if (can_swap_operands(instr)) {
instr->operands[sgpr_idx] = instr->operands[0]; instr->operands[sgpr_idx] = instr->operands[0];
instr->operands[0] = Operand(info.temp); instr->operands[0] = Operand(sgpr);
ctx.uses[sgpr_info_id]--; /* swap bits using a 4-entry LUT */
ctx.uses[info.temp.id()]++; uint32_t swapped = (0x3120 >> (operand_mask & 0x3)) & 0xf;
operand_mask = (operand_mask & ~0x3) | swapped;
} else if (can_use_VOP3(instr)) { } else if (can_use_VOP3(instr)) {
to_VOP3(ctx, instr); to_VOP3(ctx, instr);
instr->operands[sgpr_idx] = Operand(info.temp); instr->operands[sgpr_idx] = Operand(sgpr);
ctx.uses[sgpr_info_id]--; } else {
ctx.uses[info.temp.id()]++; continue;
} }
/* we can have two sgprs on one instruction if it is the same sgpr! */ sgpr_ids[num_sgprs++] = sgpr.id();
} else if (sgpr_info_id != 0 &&
sgpr_ssa_id == ctx.info[sgpr_info_id].temp.id() &&
ctx.uses[sgpr_info_id] == 1 &&
can_use_VOP3(instr)) {
to_VOP3(ctx, instr);
instr->operands[sgpr_idx] = Operand(ctx.info[sgpr_info_id].temp);
ctx.uses[sgpr_info_id]--; ctx.uses[sgpr_info_id]--;
ctx.uses[ctx.info[sgpr_info_id].temp.id()]++; ctx.uses[sgpr.id()]++;
break; /* for testing purposes, only apply 1 new sgpr */
} }
} }