fdo-mirrors/mesa
1
0
Fork 0
mirror of https://gitlab.freedesktop.org/mesa/mesa.git synced 2025-12-24 21:50:12 +01:00
Code Issues Projects Releases Packages Wiki Activity Actions 1

amd/lower_mem_access_bit_sizes: be more careful with 8/16-bit scratch load

Browse source 
Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Backport-to: 25.3
Reviewed-by: Georg Lehmann <dadschoorse@gmail.com>
(cherry picked from commit e89b22280f)

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/38010>
This commit is contained in:
Rhys Perry 2025-10-17 15:39:20 +01:00 • committed by Dylan Baker
parent 2b8675fd86
commit c1cf6e75ae
2 changed files with 7 additions and 6 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

2
.pick_status.json
View file

@ -224,7 +224,7 @@
"description": "amd/lower_mem_access_bit_sizes: be more careful with 8/16-bit scratch load",
"nominated": true,
"nomination_type": 4,
"resolution": 0,
"resolution": 1,
"main_sha": null,
"because_sha": null,
"notes": null

11
src/amd/common/nir/ac_nir_lower_mem_access_bit_sizes.c
View file

@ -121,10 +121,11 @@ lower_mem_access_cb(nir_intrinsic_op intrin, uint8_t bytes, uint8_t bit_size, ui
return res;
}
const bool is_buffer_load = intrin == nir_intrinsic_load_ubo ||
intrin == nir_intrinsic_load_ssbo ||
intrin == nir_intrinsic_load_constant;
if (is_smem) {
const bool is_buffer_load = intrin == nir_intrinsic_load_ubo ||
intrin == nir_intrinsic_load_ssbo ||
intrin == nir_intrinsic_load_constant;
const bool supported_subdword = cb_data->gfx_level >= GFX12 &&
intrin != nir_intrinsic_load_push_constant &&
(!cb_data->use_llvm || intrin != nir_intrinsic_load_ubo);
@ -194,8 +195,8 @@ lower_mem_access_cb(nir_intrinsic_op intrin, uint8_t bytes, uint8_t bit_size, ui
const uint32_t max_pad = 4 - MIN2(combined_align, 4);
/* Global loads don't have bounds checking, so increasing the size might not be safe. */
if (intrin == nir_intrinsic_load_global || intrin == nir_intrinsic_load_global_constant) {
/* Global/scratch loads don't have bounds checking, so increasing the size might not be safe. */
if (!is_buffer_load) {
if (align_mul < 4) {
/* If we split the load, only lower it to 32-bit if this is a SMEM load. */
const unsigned chunk_bytes = align(bytes, 4) - max_pad;