fdo-mirrors/mesa
1
0
Fork 0
mirror of https://gitlab.freedesktop.org/mesa/mesa.git synced 2025-12-23 22:00:13 +01:00
Code Issues Projects Releases Packages Wiki Activity Actions 1

nir/load_store_vectorize: also parse offsets through u2u64 if additions don't wrap around

Browse source 
Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/37163>
This commit is contained in:
Daniel Schürmann 2025-07-17 11:33:34 +02:00 • committed by Marge Bot
parent 084add9959
commit acb47d2c78
1 changed files with 16 additions and 9 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

25
src/compiler/nir/nir_opt_load_store_vectorize.c
View file

@ -305,11 +305,14 @@ get_effective_alu_op(nir_scalar scalar)
* sources is a constant, update "def" to be the non-constant source, fill "c"
* with the constant and return true. */
static bool
parse_alu(nir_scalar *def, nir_op op, uint64_t *c)
parse_alu(nir_scalar *def, nir_op op, uint64_t *c, bool require_nuw)
{
if (!nir_scalar_is_alu(*def) || get_effective_alu_op(*def) != op)
return false;
if (require_nuw && !nir_def_as_alu(def->def)->no_unsigned_wrap)
return false;
nir_scalar src0 = nir_scalar_chase_alu_src(*def, 0);
nir_scalar src1 = nir_scalar_chase_alu_src(*def, 1);
if (op != nir_op_ishl && nir_scalar_is_const(src0)) {
@ -337,20 +340,23 @@ parse_offset(nir_scalar *base, uint64_t *base_mul, uint64_t *offset)
uint64_t mul = 1;
uint64_t add = 0;
bool progress = false;
bool require_nuw = false;
do {
uint64_t mul2 = 1, add2 = 0;
progress = parse_alu(base, nir_op_imul, &mul2);
progress = parse_alu(base, nir_op_imul, &mul2, require_nuw);
mul *= mul2;
mul2 = 0;
progress |= parse_alu(base, nir_op_ishl, &mul2);
progress |= parse_alu(base, nir_op_ishl, &mul2, require_nuw);
mul <<= mul2;
progress |= parse_alu(base, nir_op_iadd, &add2);
progress |= parse_alu(base, nir_op_iadd, &add2, require_nuw);
add += add2 * mul;
if (nir_scalar_is_alu(*base) && nir_scalar_alu_op(*base) == nir_op_mov) {
if (nir_scalar_is_alu(*base) && (nir_scalar_alu_op(*base) == nir_op_mov ||
nir_scalar_alu_op(*base) == nir_op_u2u64)) {
require_nuw |= nir_scalar_alu_op(*base) == nir_op_u2u64;
*base = nir_scalar_chase_alu_src(*base, 0);
progress = true;
}
@ -384,8 +390,6 @@ static unsigned
add_to_entry_key(nir_scalar *offset_defs, uint64_t *offset_defs_mul,
unsigned offset_def_count, nir_scalar def, uint64_t mul)
{
mul = util_mask_sign_extend(mul, def.def->bit_size);
for (unsigned i = 0; i <= offset_def_count; i++) {
if (i == offset_def_count || cmp_scalar(def, offset_defs[i])) {
/* insert before i */
@ -453,9 +457,10 @@ create_entry_key_from_deref(void *mem_ctx,
*offset_base += offset * stride;
if (base.def) {
uint64_t mul = util_mask_sign_extend(base_mul * stride, index->bit_size);
offset_def_count += add_to_entry_key(offset_defs, offset_defs_mul,
offset_def_count,
base, base_mul * stride);
base, mul);
}
break;
}
@ -493,6 +498,7 @@ static unsigned
parse_entry_key_from_offset(struct entry_key *key, unsigned size, unsigned left,
nir_scalar base, uint64_t base_mul, uint64_t *offset)
{
unsigned original_bit_size = base.def->bit_size;
uint64_t new_mul;
uint64_t new_offset;
parse_offset(&base, &new_mul, &new_offset);
@ -505,7 +511,7 @@ parse_entry_key_from_offset(struct entry_key *key, unsigned size, unsigned left,
assert(left >= 1);
if (left >= 2) {
if (left >= 2 && base.def->bit_size == original_bit_size) {
if (nir_scalar_is_alu(base) && nir_scalar_alu_op(base) == nir_op_iadd) {
nir_scalar src0 = nir_scalar_chase_alu_src(base, 0);
nir_scalar src1 = nir_scalar_chase_alu_src(base, 1);
@ -515,6 +521,7 @@ parse_entry_key_from_offset(struct entry_key *key, unsigned size, unsigned left,
}
}
base_mul = util_mask_sign_extend(base_mul, original_bit_size);
return add_to_entry_key(key->offset_defs, key->offset_defs_mul, size, base, base_mul);
}