mirror of
https://gitlab.freedesktop.org/mesa/mesa.git
synced 2026-05-21 17:38:08 +02:00
3d7d2115f8
Replace the vk_meta_fill_buffer call with direct panlib precomp dispatches: a KERNEL(32) uint4 bulk path for 16-byte-aligned fills and a KERNEL(32) uint32 path otherwise, each with a KERNEL(1) scalar tail for sub-workgroup remainders. gpu-ratemeter vk.bufbw on Mali-G610 MC4 shows a 1.15-1.18x median speedup across alignment classes and roughly 5x on fills <= 512 B, thanks to the removed pipeline bind / descriptor-set setup that vk_meta_fill_buffer pays per call. Signed-off-by: Christian Gmeiner <cgmeiner@igalia.com> Reviewed-by: Lars-Ivar Hesselberg Simonsen <lars-ivar.simonsen@arm.com> Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/41079> |
||
|---|---|---|
| .. | ||
| ci | ||
| clc | ||
| compiler | ||
| drm-shim | ||
| ds | ||
| genxml | ||
| lib | ||
| libpan | ||
| model | ||
| perf | ||
| shared | ||
| tools | ||
| vulkan | ||
| .clang-format | ||
| meson.build | ||