Enlarge the buffer hashlist to prevent large numbers of misses
due to adding more buffers than can be cached in the hashlist.
The game I tested had CS's with up to 1500 buffers and the overhead
of amdgpu_lookup_buffer for various sizes was:
4096 1.97% (new value)
2048 4.37%
1024 6.92%
512 9.47% (old value)
(percentage of CPU usage in render thread as determined by perf)
The time spent in amdgpu_add_buffer self is ~4.2% in all cases and
for 4096 the time needed to clear the hashlist is still < 0.10%,
so I am not expecting significant regressions.
Signed-off-by: Bas Nieuwenhuizen <b...@basnieuwenhuizen.nl>
---
src/gallium/winsys/amdgpu/drm/amdgpu_cs.h | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/src/gallium/winsys/amdgpu/drm/amdgpu_cs.h
b/src/gallium/winsys/amdgpu/drm/amdgpu_cs.h
index 6ad3cdd..a2fb44a 100644
--- a/src/gallium/winsys/amdgpu/drm/amdgpu_cs.h
+++ b/src/gallium/winsys/amdgpu/drm/amdgpu_cs.h
@@ -76,7 +76,7 @@ struct amdgpu_cs {
uint8_t *flags;
struct amdgpu_cs_buffer *buffers;
- int buffer_indices_hashlist[512];
+ int buffer_indices_hashlist[4096];
uint64_t used_vram;
uint64_t used_gart;
--
2.7.2
_______________________________________________
mesa-dev mailing list
mesa-dev@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/mesa-dev
