fdo-mirrors/mesa
1
0
Fork 0
mirror of https://gitlab.freedesktop.org/mesa/mesa.git synced 2025-12-20 05:10:11 +01:00
Code Issues Projects Releases Packages Wiki Activity Actions 2

pan/bi: Keep vectorized phis
Some checks failed
macOS-CI / macOS-CI (dri) (push) Has been cancelled
Details
macOS-CI / macOS-CI (xlib) (push) Has been cancelled
Details

Browse source 
When the source contains vectorized phis, keep them to avoid
generating unnecessary moves.

Reviewed-by: Faith Ekstrand <faith.ekstrand@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/38772>
This commit is contained in:
Romaric Jodin 2025-11-17 09:41:33 +01:00 • committed by Marge Bot
parent d3a890a58e
commit 6b67ca2e8f
1 changed files with 12 additions and 3 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

15
src/panfrost/compiler/bifrost/bifrost_compile.c
View file

@ -5021,7 +5021,7 @@ bi_emit_phi_deferred(bi_context *ctx, bi_block *block, bi_instr *I)
nir_phi_instr *phi = I->phi;
/* Guaranteed by lower_phis_to_scalar */
assert(phi->def.num_components == 1);
assert(phi->def.bit_size * phi->def.num_components <= 32);
nir_foreach_phi_src(src, phi) {
bi_block *pred = bi_from_nir_block(ctx, src->pred);
@ -5502,7 +5502,16 @@ bi_vectorize_filter(const nir_instr *instr, const void *data)
{
unsigned gpu_id = *((unsigned *)data);
/* Defaults work for everything else */
if (instr->type == nir_instr_type_phi) {
unsigned bit_size = nir_instr_as_phi(instr)->def.bit_size;
if (bit_size == 8)
return 4;
if (bit_size == 16)
return 2;
return 1;
}
/* Do not vectorize all non-ALU instruction */
if (instr->type != nir_instr_type_alu)
return 0;
@ -6355,7 +6364,7 @@ bifrost_postprocess_nir(nir_shader *nir, unsigned gpu_id)
NIR_PASS(_, nir, nir_lower_alu_width, bi_vectorize_filter, &gpu_id);
NIR_PASS(_, nir, nir_lower_load_const_to_scalar);
NIR_PASS(_, nir, nir_lower_all_phis_to_scalar);
NIR_PASS(_, nir, nir_lower_phis_to_scalar, bi_vectorize_filter, &gpu_id);
NIR_PASS(_, nir, nir_lower_flrp, 16 | 32 | 64, false /* always_precise */);
NIR_PASS(_, nir, nir_lower_var_copies);
NIR_PASS(_, nir, nir_lower_alu);