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

aco: force linear for event_vmem_sample and event_vmem_bvh

Browse Source 
I don't know if this issue affects GFX12, but workaround it anyway to be
safe.

fossil-db (gfx1200):
Totals from 3463 (4.36% of 79395) affected shaders:
Instrs: 9794280 -> 9833253 (+0.40%); split: -0.00%, +0.40%
CodeSize: 52306040 -> 52457988 (+0.29%); split: -0.01%, +0.30%
Latency: 90549385 -> 93617517 (+3.39%); split: -0.00%, +3.39%
InvThroughput: 13189030 -> 13602942 (+3.14%); split: -0.00%, +3.14%

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Georg Lehmann <dadschoorse@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/32373>
This commit is contained in:
Rhys Perry
2024-11-27 14:51:32 +00:00
committed by Marge Bot
parent 86c63b29bc
commit fd19ff0b9e
1 changed files with 2 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/amd/compiler/aco_insert_waitcnt.cpp
+2 -1
View File
@@ -562,7 +562,8 @@ insert_wait_entry(wait_ctx& ctx, Definition def, wait_event event, uint8_t vmem_
/* We can't safely write to unwritten destination VGPR lanes with DS/VMEM on GFX11 without
* waiting for the load to finish.
*/
uint32_t ds_vmem_events = event_lds | event_gds | event_vmem | event_flat;
uint32_t ds_vmem_events =
event_lds | event_gds | event_vmem | event_vmem_sample | event_vmem_bvh | event_flat;
bool force_linear = ctx.gfx_level >= GFX11 && (event & ds_vmem_events);
insert_wait_entry(ctx, def.physReg(), def.regClass(), event, true, vmem_types, force_linear);