fdo-mirrors/mesa
1
0
Fork 0
mirror of https://gitlab.freedesktop.org/mesa/mesa.git synced 2026-05-07 04:58:05 +02:00
Code Issues Projects Releases Packages Wiki Activity Actions 4

pan/bi: Handle arbitrary size constants

Browse source 
This isn't hard to do and it gives us a lot more flexibility in what NIR
we can consume.

Reviewed-by: Christoph Pillmayer <christoph.pillmayer@arm.com>
Reviewed-by: Lars-Ivar Hesselberg Simonsen <lars-ivar.simonsen@arm.com>
Reviewed-by: Lorenzo Rossi <lorenzo.rossi@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/40769>
This commit is contained in:
Faith Ekstrand 2026-04-02 10:48:52 -04:00
parent 0e5626d717
commit c8f61b6b0e
1 changed files with 25 additions and 16 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

41
src/panfrost/compiler/bifrost/bifrost_compile.c
View file

@ -2274,25 +2274,34 @@ bi_emit_intrinsic(bi_builder *b, nir_intrinsic_instr *instr)
static void
bi_emit_load_const(bi_builder *b, nir_load_const_instr *instr)
{
/* Accumulate all the channels of the constant, as if we did an
* implicit SEL over them */
uint64_t acc = 0;
const unsigned sz = instr->def.bit_size;
const unsigned comps = instr->def.num_components;
bi_index temp[BI_MAX_VEC];
for (unsigned i = 0; i < instr->def.num_components; ++i) {
uint64_t v =
nir_const_value_as_uint(instr->value[i], instr->def.bit_size);
acc |= (v << (i * instr->def.bit_size));
}
if (instr->def.bit_size <= 32) {
bi_mov_i32_to(b, bi_get_index(instr->def.index), bi_imm_u32(acc));
if (sz == 64) {
for (unsigned i = 0; i < comps; ++i) {
uint64_t v = nir_const_value_as_uint(instr->value[i], 64);
temp[i * 2 + 0] = bi_mov_i32(b, bi_imm_u32(v));
temp[i * 2 + 1] = bi_mov_i32(b, bi_imm_u32(v >> 32));
}
bi_emit_collect_to(b, bi_get_index(instr->def.index), temp, comps * 2);
} else {
uint32_t imm_2x32[2] = { acc & 0xffffffff, (acc >> 32) & 0xffffffff };
bi_index temp[2] = { bi_temp(b->shader), bi_temp(b->shader) };
bi_mov_i32_to(b, temp[0], bi_imm_u32(imm_2x32[0]));
bi_mov_i32_to(b, temp[1], bi_imm_u32(imm_2x32[1]));
assert(sz == 8 || sz == 16 || sz == 32);
const unsigned comps_per_word = 32 / sz;
const unsigned words = DIV_ROUND_UP(comps, comps_per_word);
for (unsigned w = 0; w < words; w++) {
uint32_t acc = 0;
for (unsigned i = 0; i < comps_per_word; i++) {
unsigned c = w * comps_per_word + i;
if (c >= comps)
break;
bi_emit_collect_to(b, bi_get_index(instr->def.index), temp, 2);
uint32_t v = nir_const_value_as_uint(instr->value[c], sz);
acc |= v << (i * sz);
}
temp[w] = bi_mov_i32(b, bi_imm_u32(acc));
}
bi_emit_collect_to(b, bi_get_index(instr->def.index), temp, words);
}
}