OpenHarmony/third_party_mesa3d
2
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

radeonsi/gfx11: fix compute scratch buffer - WAVES is always per SE

Browse Source 
Fixes: ba02ed91a6 - ac/gfx11: fix the scratch buffer

Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19477>
This commit is contained in:
Marek Olšák
2022-11-02 14:34:58 -04:00
parent 803802c29f
commit bdfacd0a24
4 changed files with 6 additions and 6 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
src/amd/common/ac_shader_util.c
View File

@@ -916,7 +916,7 @@ void ac_set_reg_cu_en(void *cs, unsigned reg_offset, uint32_t value, uint32_t cl
} }
/* Return the register value and tune bytes_per_wave to increase scratch performance. */ /* Return the register value and tune bytes_per_wave to increase scratch performance. */
void ac_get_scratch_tmpring_size(const struct radeon_info *info, bool compute, void ac_get_scratch_tmpring_size(const struct radeon_info *info,
unsigned bytes_per_wave, unsigned *max_seen_bytes_per_wave, unsigned bytes_per_wave, unsigned *max_seen_bytes_per_wave,
uint32_t *tmpring_size) uint32_t *tmpring_size)
{ {
@@ -949,8 +949,8 @@ void ac_get_scratch_tmpring_size(const struct radeon_info *info, bool compute,
*max_seen_bytes_per_wave = MAX2(*max_seen_bytes_per_wave, bytes_per_wave); *max_seen_bytes_per_wave = MAX2(*max_seen_bytes_per_wave, bytes_per_wave);
unsigned max_scratch_waves = info->max_scratch_waves; unsigned max_scratch_waves = info->max_scratch_waves;
if (info->gfx_level >= GFX11 && !compute) if (info->gfx_level >= GFX11)
max_scratch_waves /= info->num_se; /* WAVES is per SE for SPI_TMPRING_SIZE. */ max_scratch_waves /= info->num_se; /* WAVES is per SE */
/* TODO: We could decrease WAVES to make the whole buffer fit into the infinity cache. */ /* TODO: We could decrease WAVES to make the whole buffer fit into the infinity cache. */
*tmpring_size = S_0286E8_WAVES(max_scratch_waves) | *tmpring_size = S_0286E8_WAVES(max_scratch_waves) |

2
src/amd/common/ac_shader_util.h
View File

@@ -166,7 +166,7 @@ void ac_set_reg_cu_en(void *cs, unsigned reg_offset, uint32_t value, uint32_t cl
unsigned value_shift, const struct radeon_info *info, unsigned value_shift, const struct radeon_info *info,
void set_sh_reg(void*, unsigned, uint32_t)); void set_sh_reg(void*, unsigned, uint32_t));
void ac_get_scratch_tmpring_size(const struct radeon_info *info, bool compute, void ac_get_scratch_tmpring_size(const struct radeon_info *info,
unsigned bytes_per_wave, unsigned *max_seen_bytes_per_wave, unsigned bytes_per_wave, unsigned *max_seen_bytes_per_wave,
uint32_t *tmpring_size); uint32_t *tmpring_size);

2
src/gallium/drivers/radeonsi/si_compute.c
View File

@@ -566,7 +566,7 @@ static bool si_switch_compute_shader(struct si_context *sctx, struct si_compute
} }
unsigned tmpring_size; unsigned tmpring_size;
ac_get_scratch_tmpring_size(&sctx->screen->info, true, ac_get_scratch_tmpring_size(&sctx->screen->info,
config->scratch_bytes_per_wave, config->scratch_bytes_per_wave,
&sctx->max_seen_compute_scratch_bytes_per_wave, &tmpring_size); &sctx->max_seen_compute_scratch_bytes_per_wave, &tmpring_size);

2
src/gallium/drivers/radeonsi/si_state_shaders.cpp
View File

@@ -4054,7 +4054,7 @@ static bool si_update_scratch_relocs(struct si_context *sctx)
bool si_update_spi_tmpring_size(struct si_context *sctx, unsigned bytes) bool si_update_spi_tmpring_size(struct si_context *sctx, unsigned bytes)
{ {
unsigned spi_tmpring_size; unsigned spi_tmpring_size;
ac_get_scratch_tmpring_size(&sctx->screen->info, false, bytes, ac_get_scratch_tmpring_size(&sctx->screen->info, bytes,
&sctx->max_seen_scratch_bytes_per_wave, &spi_tmpring_size); &sctx->max_seen_scratch_bytes_per_wave, &spi_tmpring_size);
unsigned scratch_needed_size = sctx->max_seen_scratch_bytes_per_wave * unsigned scratch_needed_size = sctx->max_seen_scratch_bytes_per_wave *