fdo-mirrors/mesa
1
0
Fork 0
mirror of https://gitlab.freedesktop.org/mesa/mesa.git synced 2025-12-25 00:00:11 +01:00
Code Issues Projects Releases Packages Wiki Activity Actions 1

winsys/amdgpu: overallocate buffers for faster address translation on Gfx9

Browse source 
Sadly, the 3 games I tested (DeusEx:MD, DiRT Rally, DOTA 2) are unaffected
by the overallocation, because I guess their buffers don't fall into
the small range below a power-of-two size.

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
This commit is contained in:
Marek Olšák 2018-11-21 18:06:54 -05:00
parent 8c00f778fc
commit f737fe00a0
1 changed files with 24 additions and 0 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

24
src/gallium/winsys/amdgpu/drm/amdgpu_bo.c
View file

@ -442,6 +442,30 @@ static struct amdgpu_winsys_bo *amdgpu_create_bo(struct amdgpu_winsys *ws,
/* VRAM or GTT must be specified, but not both at the same time. */
assert(util_bitcount(initial_domain & RADEON_DOMAIN_VRAM_GTT) == 1);
/* Gfx9: Overallocate the size to the next power of two for faster address
* translation if we don't waste too much memory.
*/
if (ws->info.chip_class >= GFX9) {
uint64_t next_pot_size = util_next_power_of_two64(size);
/* For slightly lower than 4 GB allocations, at most 32 MB are wasted.
* For slightly lower than 256 MB allocations, at most 2 MB are wasted.
* For slightly lower than 64 MB allocations, at most 512 KB are wasted.
*
* Waste at most 0.79% (1/127) of the size if we decide to overallocate.
*/
uint64_t max_overalloc = next_pot_size >> 7;
/* If the next power-of-two size is <= the page size, waste up to
* 6.25% (1/16) of the size if we decide to overallocate.
*/
if (next_pot_size <= ws->info.pte_fragment_size)
max_overalloc = next_pot_size >> 4;
if (size + max_overalloc >= next_pot_size)
size = next_pot_size;
}
bo = CALLOC_STRUCT(amdgpu_winsys_bo);
if (!bo) {
return NULL;