mirror of
https://gitlab.freedesktop.org/mesa/mesa.git
synced 2026-06-21 05:58:22 +02:00
cb8da4bb41
4K sparse allocations are super slow because they thrash TLB, so it's better
to disallow them.
radeonsi already reports 64K alignment for sparse buffers, but RADV
reported 4K.
VMEM latency with 4K pages:
Size , 1024, 1536, 2K, 3K, 4K, 6K, 8K, 12K, 16K, 24K, 32K, 48K, 64K, 96K, 128K, 192K, 256K, 384K, 512K, 768K, 1024K, 1536K, 2M, 3M, 4M,
Latency, 86, 86, 86, 86, 89, 86, 86, 87, 87, 126, 127, 126, 128, 387, 409, 448, 423, 450, 446, 424, 430, 429, 428, 433, 432
^ ^ ^ ^ ^ ^ ^ ^ ^ ^
VMEM latency with 64K pages (it's lower, up to 2x lower in some cases):
Size , 1024, 1536, 2K, 3K, 4K, 6K, 8K, 12K, 16K, 24K, 32K, 48K, 64K, 96K, 128K, 192K, 256K, 384K, 512K, 768K, 1024K, 1536K, 2M, 3M, 4M,
Latency, 87, 86, 86, 86, 86, 86, 86, 87, 87, 125, 127, 126, 128, 198, 196, 197, 198, 376, 392, 411, 381, 401, 392, 432, 451,
^ ^ ^ ^ ^ ^ ^ ^ ^ ^
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/42127>
|
||
|---|---|---|
| .. | ||
| amd | ||
| android_stub | ||
| asahi | ||
| broadcom | ||
| c11 | ||
| compiler | ||
| drm-shim | ||
| egl | ||
| etnaviv | ||
| freedreno | ||
| gallium | ||
| gbm | ||
| getopt | ||
| gfxstream | ||
| glx | ||
| gtest | ||
| imagination | ||
| imgui | ||
| intel | ||
| kosmickrisp | ||
| loader | ||
| mesa | ||
| microsoft | ||
| nouveau | ||
| panfrost | ||
| poly | ||
| tool | ||
| util | ||
| virtio | ||
| vulkan | ||
| x11 | ||
| .clang-format | ||
| meson.build | ||