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

aco/insert_exec_mask: optimize top-level transition to exact before demote 

fossil-db (Sienna Cichlid):
Totals from 5767 (3.55% of 162293) affected shaders:
Instrs: 3264949 -> 3257527 (-0.23%); split: -0.23%, +0.00%
CodeSize: 17835692 -> 17806004 (-0.17%); split: -0.17%, +0.00%
Latency: 45990060 -> 45987924 (-0.00%); split: -0.00%, +0.00%
InvThroughput: 7643850 -> 7643835 (-0.00%); split: -0.00%, +0.00%
Copies: 193641 -> 186219 (-3.83%); split: -3.84%, +0.01%

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/15244>
This commit is contained in:
Rhys Perry 2022-02-23 17:29:25 +00:00 committed by Marge Bot
parent 42a5be975a
commit d068eb53e8
1 changed files with 10 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

13
src/amd/compiler/aco_insert_exec_mask.cpp
View File

@ -585,8 +585,7 @@ process_instructions(exec_ctx& ctx, Block* block, std::vector<aco_ptr<Instructio
}
} else if (instr->opcode == aco_opcode::p_demote_to_helper) {
/* turn demote into discard_if with only exact masks */
assert((ctx.info[block->index].exec[0].second & (mask_type_exact | mask_type_global)) ==
(mask_type_exact | mask_type_global));
assert(ctx.info[block->index].exec[0].second == (mask_type_exact | mask_type_global));
int num;
Temp cond, exit_cond;
@ -605,7 +604,15 @@ process_instructions(exec_ctx& ctx, Block* block, std::vector<aco_ptr<Instructio
}
} else {
/* demote_if: transition to exact */
transition_to_Exact(ctx, bld, block->index);
if (block->kind & block_kind_top_level && ctx.info[block->index].exec.size() == 2 &&
ctx.info[block->index].exec.back().second & mask_type_global) {
/* We don't need to actually copy anything into exact, since the s_andn2
* instructions later will do that.
*/
ctx.info[block->index].exec.pop_back();
} else {
transition_to_Exact(ctx, bld, block->index);
}
assert(instr->operands[0].isTemp());
cond = instr->operands[0].getTemp();
num = ctx.info[block->index].exec.size() - 1;