AlexIndustrial/mesa
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

microsoft/compiler: Respect write masks when lowering unaligned loads and stores

Browse Source 
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/37778>
This commit is contained in:
Jesse Natalie
2025-10-10 11:13:53 -07:00
committed by Marge Bot
parent b3242516ad
commit c2d288bf97
1 changed files with 11 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/microsoft/compiler/dxil_nir.c
+11 -4
View File
@@ -1943,7 +1943,7 @@ split_unaligned_store(nir_builder *b, nir_intrinsic_instr *intrin, unsigned alig
nir_def *value = intrin->src[1].ssa;
unsigned comp_size = value->bit_size / 8;
unsigned num_comps = value->num_components;
b->cursor = nir_before_instr(&intrin->instr);
nir_deref_instr *ptr = nir_src_as_deref(intrin->src[0]);
@@ -1953,9 +1953,13 @@ split_unaligned_store(nir_builder *b, nir_intrinsic_instr *intrin, unsigned alig
unsigned num_stores = DIV_ROUND_UP(comp_size * num_comps, alignment);
for (unsigned i = 0; i < num_stores; ++i) {
nir_def *substore_val = nir_extract_bits(b, &value, 1, i * alignment * 8, 1, alignment * 8);
nir_deref_instr *elem = nir_build_deref_ptr_as_array(b, cast, nir_imm_intN_t(b, i, cast->def.bit_size));
nir_store_deref_with_access(b, elem, substore_val, ~0, access);
unsigned first_src_comp = i * alignment / comp_size;
/* If write mask has holes, alignment will be set to scalar size */
if (nir_intrinsic_write_mask(intrin) & (1u << first_src_comp)) {
nir_def *substore_val = nir_extract_bits(b, &value, 1, i * alignment * 8, 1, alignment * 8);
nir_deref_instr *elem = nir_build_deref_ptr_as_array(b, cast, nir_imm_intN_t(b, i, cast->def.bit_size));
nir_store_deref_with_access(b, elem, substore_val, ~0, access);
}
}
}
@@ -2015,6 +2019,9 @@ lower_unaligned_load_store(nir_builder *b, nir_instr *instr, void *state)
if (intrin->intrinsic == nir_intrinsic_load_deref)
return split_unaligned_load(b, intrin, alignment);
else {
/* If the write mask has holes, then force the split to be per-component */
if (nir_intrinsic_write_mask(intrin) != (1 << intrin->src[1].ssa->num_components) - 1)
alignment = MIN2(alignment, intrin->src[1].ssa->bit_size * 8);
split_unaligned_store(b, intrin, alignment);
return NIR_LOWER_INSTR_PROGRESS_REPLACE;
}