fdo-mirrors/mesa
1
0
Fork 0
mirror of https://gitlab.freedesktop.org/mesa/mesa.git synced 2026-01-03 20:10:17 +01:00
Code Issues Projects Releases Packages Wiki Activity Actions 1

rusticl/kernel: Do not run kernels with a workgroup size beyond work_dim

Browse source 
When no workgroup size is specified we try to run with the most optimal one
possible. However we didn't take into account that we shouldn't run a
workgroup of higher dimensionality than requested by the application.

Fixes: 376d1e6667 ("rusticl: implement cl_khr_suggested_local_work_size")
(cherry picked from commit d46be8fbf2)

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/38803>
This commit is contained in:
Karol Herbst 2025-11-11 17:51:26 +01:00 • committed by Dylan Baker
parent 57ea82d834
commit eb51b0b39a
2 changed files with 11 additions and 5 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

2
.pick_status.json
View file

@ -684,7 +684,7 @@
"description": "rusticl/kernel: Do not run kernels with a workgroup size beyond work_dim",
"nominated": true,
"nomination_type": 2,
"resolution": 0,
"resolution": 1,
"main_sha": null,
"because_sha": "376d1e6667a80e1811d4c25115633817f16666a7",
"notes": null

14
src/gallium/frontends/rusticl/core/kernel.rs
View file

@ -1478,7 +1478,13 @@ impl Kernel {
}
}
fn optimize_local_size(&self, d: &Device, grid: &mut [usize; 3], block: &mut [u32; 3]) {
fn optimize_local_size(
&self,
d: &Device,
work_dim: u32,
grid: &mut [usize; 3],
block: &mut [u32; 3],
) {
if !block.contains(&0) {
for i in 0..3 {
// we already made sure everything is fine
@ -1492,10 +1498,10 @@ impl Kernel {
usize_block[i] = block[i] as usize;
}
self.suggest_local_size(d, 3, grid, &mut usize_block);
self.suggest_local_size(d, work_dim as usize, grid, &mut usize_block);
for i in 0..3 {
block[i] = usize_block[i] as u32;
block[i] = 1.max(usize_block[i] as u32);
}
}
@ -1549,7 +1555,7 @@ impl Kernel {
let api_grid = grid;
self.optimize_local_size(q.device, &mut grid, &mut block);
self.optimize_local_size(q.device, work_dim, &mut grid, &mut block);
Ok(Box::new(move |cl_ctx, ctx| {
let hw_max_grid = ctx.dev.max_grid_size();