KonstantinSeurer
/
mesa
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

aco/ra: improve split_vector register assignment if the operand is not killed 

This allows for more coalescing when lowering the copies.

Totals from 44801 (33.21% of 134913) affected shaders: (GFX10.3)
VGPRs: 1513264 -> 1513248 (-0.00%)
CodeSize: 113354240 -> 113172872 (-0.16%); split: -0.16%, +0.00%
Instrs: 21648793 -> 21603397 (-0.21%); split: -0.21%, +0.00%
Latency: 95762290 -> 95757403 (-0.01%); split: -0.01%, +0.00%
InvThroughput: 15427354 -> 15427341 (-0.00%); split: -0.00%, +0.00%
Copies: 2065330 -> 2019933 (-2.20%); split: -2.20%, +0.00%

Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15414>
This commit is contained in:
Daniel Schürmann 2022-03-16 17:25:35 +01:00 committed by Marge Bot
parent 499dc20e6a
commit 5e6e47ecea
1 changed files with 15 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

16
src/amd/compiler/aco_register_allocation.cpp
View File

@ -2793,10 +2793,24 @@ register_allocation(Program* program, std::vector<IDSet>& live_out_per_block, ra
/* find free reg */
if (instr->opcode == aco_opcode::p_split_vector) {
PhysReg reg = instr->operands[0].physReg();
RegClass rc = definition->regClass();
for (unsigned j = 0; j < i; j++)
reg.reg_b += instr->definitions[j].bytes();
if (get_reg_specified(ctx, register_file, definition->regClass(), instr, reg))
if (get_reg_specified(ctx, register_file, rc, instr, reg)) {
definition->setFixed(reg);
} else if (i == 0) {
RegClass vec_rc = RegClass::get(rc.type(), instr->operands[0].bytes());
DefInfo info(ctx, ctx.pseudo_dummy, vec_rc, -1);
std::pair<PhysReg, bool> res = get_reg_simple(ctx, register_file, info);
reg = res.first;
if (res.second && get_reg_specified(ctx, register_file, rc, instr, reg))
definition->setFixed(reg);
} else if (instr->definitions[i - 1].isFixed()) {
reg = instr->definitions[i - 1].physReg();
reg.reg_b += instr->definitions[i - 1].bytes();
if (get_reg_specified(ctx, register_file, rc, instr, reg))
definition->setFixed(reg);
}
} else if (instr->opcode == aco_opcode::p_wqm ||
instr->opcode == aco_opcode::p_parallelcopy) {
PhysReg reg = instr->operands[i].physReg();