fdo-mirrors/mesa
1
0
Fork 0
mirror of https://gitlab.freedesktop.org/mesa/mesa.git synced 2026-05-06 13:48:06 +02:00
Code Issues Projects Releases Packages Wiki Activity Actions 1

jay/ra: don't reserve registers when not spilling

Browse source 
No changes at SIMD16. At SIMD32:

Totals:
Instrs: 4691895 -> 4691989 (+0.00%); split: -0.03%, +0.03%
CodeSize: 76010880 -> 76011248 (+0.00%); split: -0.03%, +0.03%
Number of spill instructions: 54369 -> 54223 (-0.27%)
Number of fill instructions: 56668 -> 56519 (-0.26%)

Totals from 71 (2.68% of 2647) affected shaders:
Instrs: 75963 -> 76057 (+0.12%); split: -1.67%, +1.79%
CodeSize: 1229792 -> 1230160 (+0.03%); split: -1.71%, +1.74%
Number of spill instructions: 146 -> 0 (-inf%)
Number of fill instructions: 149 -> 0 (-inf%)

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/41064>
This commit is contained in:
Alyssa Rosenzweig 2026-04-20 11:49:53 -04:00 • committed by Marge Bot
parent e5bf153d4f
commit 7d2a88a9e5
1 changed files with 3 additions and 7 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

10
src/intel/compiler/jay/jay_register_allocate.c
View file

@ -1548,15 +1548,11 @@ spill_file(jay_function *f, enum jay_file file, bool *spilled)
limit = 13;
}
/* Ensures we don't XOR swap, XXX: TODO: FIXME */
limit--;
if (f->demand[file] > limit) {
/* In the worst case, we
* require 2 temporary registers to lower a memory-memory swap produced by
* parallel copy lowering, so adjust the limit to be num_regs - 2.
/* In the worst case when spilling, we require 2 temporary registers to
* lower a memory-memory swap produced by parallel copy lowering.
*/
limit--;
limit -= 2;
/* If we spill, we need to reserve UGPRs for spilling */
if (!(*spilled)) {