diff options
| -rw-r--r-- | src/intel/compiler/brw_nir_lower_mem_access_bit_sizes.c | 10 |
1 files changed, 5 insertions, 5 deletions
diff --git a/src/intel/compiler/brw_nir_lower_mem_access_bit_sizes.c b/src/intel/compiler/brw_nir_lower_mem_access_bit_sizes.c index ef9aa206b44..19abc16a9c5 100644 --- a/src/intel/compiler/brw_nir_lower_mem_access_bit_sizes.c +++ b/src/intel/compiler/brw_nir_lower_mem_access_bit_sizes.c @@ -81,15 +81,15 @@ lower_mem_load_bit_size(nir_builder *b, nir_intrinsic_instr *intrin, intrin->intrinsic == nir_intrinsic_load_scratch; assert(intrin->dest.is_ssa); - if (intrin->dest.ssa.bit_size == 32 && - (!needs_scalar || intrin->num_components == 1)) - return false; - const unsigned bit_size = intrin->dest.ssa.bit_size; const unsigned num_components = intrin->dest.ssa.num_components; const unsigned bytes_read = num_components * (bit_size / 8); const unsigned align = nir_intrinsic_align(intrin); + if (bit_size == 32 && align >= 32 && + (!needs_scalar || intrin->num_components == 1)) + return false; + nir_ssa_def *result; nir_src *offset_src = nir_get_io_offset_src(intrin); if (bit_size < 32 && nir_src_is_const(*offset_src)) { @@ -167,7 +167,7 @@ lower_mem_store_bit_size(nir_builder *b, nir_intrinsic_instr *intrin, assert(writemask < (1 << num_components)); if ((value->bit_size <= 32 && num_components == 1) || - (value->bit_size == 32 && + (value->bit_size == 32 && align >= 32 && writemask == (1 << num_components) - 1 && !needs_scalar)) return false; |