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

aco: ensure to not extract more components than have been fetched

Browse Source 
Fixes: 7015d2c249 ('aco: fix scratch loads which cross element_size boundaries')
Cc: 20.1 <mesa-stable@lists.freedesktop.org>
Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6024>
This commit is contained in:
Daniel Schürmann
2020-07-22 16:41:41 +02:00
committed by Marge Bot
parent 6bc5ce7a91
commit bd75e99233
1 changed files with 5 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/amd/compiler/aco_instruction_selection.cpp
+5 -4
View File
@@ -472,7 +472,7 @@ void byte_align_vector(isel_context *ctx, Temp vec, Operand offset, Temp dst, un
offset = Operand(0u);
}
unsigned num_components = dst.bytes() / component_size;
unsigned num_components = vec.bytes() / component_size;
if (vec.regClass() == dst.regClass()) {
assert(offset.constantValue() == 0);
bld.copy(Definition(dst), vec);
@@ -480,17 +480,18 @@ void byte_align_vector(isel_context *ctx, Temp vec, Operand offset, Temp dst, un
return;
}
emit_split_vector(ctx, vec, vec.bytes() / component_size);
emit_split_vector(ctx, vec, num_components);
std::array<Temp, NIR_MAX_VEC_COMPONENTS> elems;
RegClass rc = RegClass(RegType::vgpr, component_size).as_subdword();
assert(offset.constantValue() % component_size == 0);
unsigned skip = offset.constantValue() / component_size;
for (unsigned i = 0; i < num_components; i++)
elems[i] = emit_extract_vector(ctx, vec, i + skip, rc);
for (unsigned i = skip; i < num_components; i++)
elems[i - skip] = emit_extract_vector(ctx, vec, i, rc);
/* if dst is vgpr - split the src and create a shrunk version according to the mask. */
if (dst.type() == RegType::vgpr) {
num_components = dst.bytes() / component_size;
aco_ptr<Pseudo_instruction> create_vec{create_instruction<Pseudo_instruction>(aco_opcode::p_create_vector, Format::PSEUDO, num_components, 1)};
for (unsigned i = 0; i < num_components; i++)
create_vec->operands[i] = Operand(elems[i]);