fdo-mirrors/mesa
1
0
Fork 0
mirror of https://gitlab.freedesktop.org/mesa/mesa.git synced 2026-03-12 13:20:33 +01:00
Code Issues Projects Releases Packages Wiki Activity Actions 4

nir: don't assume indicies are always 32bit when accessing them as raw data

Browse source 
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@intel.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/40299>
This commit is contained in:
Georg Lehmann 2026-03-09 12:57:17 +01:00 • committed by Marge Bot
parent aa831b6690
commit abfd6a4df9
5 changed files with 24 additions and 20 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

3
src/compiler/nir/nir.h
View file

@ -2116,6 +2116,9 @@ typedef struct nir_intrinsic_info {
/** the number of constant indices used by the intrinsic */
uint8_t num_indices;
/** the number of 32bit slots used for storing constant indices. */
uint8_t num_index_slots;
/** list of indices */
uint8_t indices[NIR_INTRINSIC_MAX_CONST_INDEX];

4
src/compiler/nir/nir_instr_set.c
View file

@ -242,7 +242,7 @@ hash_intrinsic(uint32_t hash, const nir_intrinsic_instr *instr)
hash = XXH32(v, sizeof(v), hash);
}
hash = XXH32(instr->const_index, info->num_indices * sizeof(instr->const_index[0]), hash);
hash = XXH32(instr->const_index, info->num_index_slots * sizeof(instr->const_index[0]), hash);
for (unsigned i = 0; i < nir_intrinsic_infos[instr->intrinsic].num_srcs; i++)
hash = hash_src(hash, &instr->src[i]);
@ -747,7 +747,7 @@ nir_instrs_equal(const nir_instr *instr1, const nir_instr *instr2)
return false;
}
for (unsigned i = 0; i < info->num_indices; i++) {
for (unsigned i = 0; i < info->num_index_slots; i++) {
if (intrinsic1->const_index[i] != intrinsic2->const_index[i])
return false;
}

1
src/compiler/nir/nir_intrinsics_c.py
View file

@ -41,6 +41,7 @@ const nir_intrinsic_info nir_intrinsic_infos[nir_num_intrinsics] = {
.dest_bit_sizes = ${hex(reduce(operator.or_, opcode.bit_sizes, 0))},
.bit_size_src = ${opcode.bit_size_src},
.num_indices = ${opcode.num_indices},
.num_index_slots = ${opcode.num_indices},
% if opcode.indices:
.indices = {
% for i in range(len(opcode.indices)):

2
src/compiler/nir/nir_lower_mem_access_bit_sizes.c
View file

@ -52,7 +52,7 @@ dup_mem_intrinsic(nir_builder *b, nir_intrinsic_instr *intrin,
}
dup->num_components = num_components;
for (unsigned i = 0; i < info->num_indices; i++)
for (unsigned i = 0; i < info->num_index_slots; i++)
dup->const_index[i] = intrin->const_index[i];
nir_set_io_offset(dup, offset);

34
src/compiler/nir/nir_serialize.c
View file

@ -1053,7 +1053,7 @@ write_intrinsic(write_ctx *ctx, const nir_intrinsic_instr *intrin)
/* 10 bits for nir_intrinsic_op */
STATIC_ASSERT(nir_num_intrinsics <= 1024);
unsigned num_srcs = nir_intrinsic_infos[intrin->intrinsic].num_srcs;
unsigned num_indices = nir_intrinsic_infos[intrin->intrinsic].num_indices;
unsigned num_index_slots = nir_intrinsic_infos[intrin->intrinsic].num_index_slots;
assert(intrin->intrinsic < 1024);
union packed_instr header;
@ -1063,19 +1063,19 @@ write_intrinsic(write_ctx *ctx, const nir_intrinsic_instr *intrin)
header.intrinsic.intrinsic = intrin->intrinsic;
/* Analyze constant indices to decide how to encode them. */
if (num_indices) {
if (num_index_slots) {
unsigned max_bits = 0;
for (unsigned i = 0; i < num_indices; i++) {
for (unsigned i = 0; i < num_index_slots; i++) {
unsigned max = util_last_bit(intrin->const_index[i]);
max_bits = MAX2(max_bits, max);
}
if (max_bits * num_indices <= 8) {
if (max_bits * num_index_slots <= 8) {
header.intrinsic.const_indices_encoding = const_indices_all_combined;
/* Pack all const indices into 8 bits. */
unsigned bit_size = 8 / num_indices;
for (unsigned i = 0; i < num_indices; i++) {
unsigned bit_size = 8 / num_index_slots;
for (unsigned i = 0; i < num_index_slots; i++) {
header.intrinsic.packed_const_indices |=
intrin->const_index[i] << (i * bit_size);
}
@ -1095,18 +1095,18 @@ write_intrinsic(write_ctx *ctx, const nir_intrinsic_instr *intrin)
for (unsigned i = 0; i < num_srcs; i++)
write_src(ctx, &intrin->src[i]);
if (num_indices) {
if (num_index_slots) {
switch (header.intrinsic.const_indices_encoding) {
case const_indices_8bit:
for (unsigned i = 0; i < num_indices; i++)
for (unsigned i = 0; i < num_index_slots; i++)
blob_write_uint8(ctx->blob, intrin->const_index[i]);
break;
case const_indices_16bit:
for (unsigned i = 0; i < num_indices; i++)
for (unsigned i = 0; i < num_index_slots; i++)
blob_write_uint16(ctx->blob, intrin->const_index[i]);
break;
case const_indices_32bit:
for (unsigned i = 0; i < num_indices; i++)
for (unsigned i = 0; i < num_index_slots; i++)
blob_write_uint32(ctx->blob, intrin->const_index[i]);
break;
}
@ -1120,7 +1120,7 @@ read_intrinsic(read_ctx *ctx, union packed_instr header)
nir_intrinsic_instr *intrin = nir_intrinsic_instr_create(ctx->nir, op);
unsigned num_srcs = nir_intrinsic_infos[op].num_srcs;
unsigned num_indices = nir_intrinsic_infos[op].num_indices;
unsigned num_index_slots = nir_intrinsic_infos[op].num_index_slots;
if (nir_intrinsic_infos[op].has_dest)
read_def(ctx, &intrin->def, &intrin->instr, header);
@ -1143,12 +1143,12 @@ read_intrinsic(read_ctx *ctx, union packed_instr header)
}
}
if (num_indices) {
if (num_index_slots) {
switch (header.intrinsic.const_indices_encoding) {
case const_indices_all_combined: {
unsigned bit_size = 8 / num_indices;
unsigned bit_size = 8 / num_index_slots;
unsigned bit_mask = u_bit_consecutive(0, bit_size);
for (unsigned i = 0; i < num_indices; i++) {
for (unsigned i = 0; i < num_index_slots; i++) {
intrin->const_index[i] =
(header.intrinsic.packed_const_indices >> (i * bit_size)) &
bit_mask;
@ -1156,15 +1156,15 @@ read_intrinsic(read_ctx *ctx, union packed_instr header)
break;
}
case const_indices_8bit:
for (unsigned i = 0; i < num_indices; i++)
for (unsigned i = 0; i < num_index_slots; i++)
intrin->const_index[i] = blob_read_uint8(ctx->blob);
break;
case const_indices_16bit:
for (unsigned i = 0; i < num_indices; i++)
for (unsigned i = 0; i < num_index_slots; i++)
intrin->const_index[i] = blob_read_uint16(ctx->blob);
break;
case const_indices_32bit:
for (unsigned i = 0; i < num_indices; i++)
for (unsigned i = 0; i < num_index_slots; i++)
intrin->const_index[i] = blob_read_uint32(ctx->blob);
break;
}