Module: Mesa Branch: main Commit: 0f5c8c3dc35561f23a5383cb05075898942ce0c5 URL: http://cgit.freedesktop.org/mesa/mesa/commit/?id=0f5c8c3dc35561f23a5383cb05075898942ce0c5
Author: Pierre-Eric Pelloux-Prayer <[email protected]> Date: Tue Jan 17 10:14:52 2023 +0100 vbo: lower VBO_SAVE_BUFFER_SIZE to avoid large VRAM usage The ideal case for performance is to have a single buffer for all display list. The caveat is that large buffers are less likely to be freed because they're refcounted: it only takes 1 user (diplay list) to keep it in VRAM. This lowers VRAM usage when replaying the trace attached of the trace attached to !6140 from 5.5 GB to about 1.8 GB. Viewperf snx performance isn't affected. Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/6140 Cc: mesa-stable Reviewed-by: Marek Olšák <[email protected]> Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20748> --- src/mesa/vbo/vbo_save.h | 6 +++++- 1 file changed, 5 insertions(+), 1 deletion(-) diff --git a/src/mesa/vbo/vbo_save.h b/src/mesa/vbo/vbo_save.h index 7782129a7ec..54ac188dc8c 100644 --- a/src/mesa/vbo/vbo_save.h +++ b/src/mesa/vbo/vbo_save.h @@ -111,8 +111,12 @@ _vbo_save_get_stride(const struct vbo_save_vertex_list *node) /* Default size for the buffer holding the vertices and the indices. * A bigger buffer helps reducing the number of draw calls but may * waste memory. + * 1MB was picked because a lower value reduces viewperf snx tests + * performance but larger values cause high VRAM usage (because + * larger buffers will be shared by more display lists which reduces + * the likelyhood of freeing the buffer). */ -#define VBO_SAVE_BUFFER_SIZE (20*1024*1024) +#define VBO_SAVE_BUFFER_SIZE (1024 * 1024) #define VBO_SAVE_PRIM_MODE_MASK 0x3f struct vbo_save_vertex_store {
