fdo-mirrors/mesa
1
0
Fork 0
mirror of https://gitlab.freedesktop.org/mesa/mesa.git synced 2026-04-27 11:00:37 +02:00
Code Issues Projects Releases Packages Wiki Activity Actions 1

pan/bi: Revamp bi_optimize_nir

Browse source 
This reorder things a bit, ensure we attempt more agressive vectorisation,
attempt to optimize cf and more.

Inspiration from NAK's optimize_nir function.

Signed-off-by: Mary Guillemard <mary.guillemard@collabora.com>
Acked-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Acked-by: Eric R. Smith <eric.smith@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/36629>
This commit is contained in:
Mary Guillemard 2025-07-29 14:52:43 +00:00 • committed by Marge Bot
parent ef7095c85b
commit c5c196e6d5
1 changed files with 39 additions and 6 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

45
src/panfrost/compiler/bifrost_compile.c
View file

@ -5424,6 +5424,8 @@ mem_vectorize_cb(unsigned align_mul, unsigned align_offset, unsigned bit_size,
static void
bi_optimize_nir(nir_shader *nir, unsigned gpu_id, nir_variable_mode robust2_modes)
{
unsigned arch = pan_arch(gpu_id);
NIR_PASS(_, nir, nir_opt_shrink_stores, true);
bool progress;
@ -5434,27 +5436,58 @@ bi_optimize_nir(nir_shader *nir, unsigned gpu_id, nir_variable_mode robust2_mode
NIR_PASS(progress, nir, nir_lower_vars_to_ssa);
NIR_PASS(progress, nir, nir_lower_wrmasks, should_split_wrmask, NULL);
NIR_PASS(_, nir, nir_opt_copy_prop_vars);
NIR_PASS(_, nir, nir_opt_dead_write_vars);
NIR_PASS(_, nir, nir_opt_combine_stores, nir_var_all);
NIR_PASS(_, nir, nir_lower_alu_width, bi_vectorize_filter, &gpu_id);
NIR_PASS(_, nir, nir_opt_vectorize, bi_vectorize_filter, &gpu_id);
NIR_PASS(progress, nir, nir_copy_prop);
NIR_PASS(progress, nir, nir_opt_remove_phis);
NIR_PASS(progress, nir, nir_opt_dce);
NIR_PASS(progress, nir, nir_opt_dead_cf);
NIR_PASS(progress, nir, nir_opt_cse);
nir_opt_peephole_select_options peephole_select_options = {
.limit = 64,
.expensive_alu_ok = true,
};
NIR_PASS(progress, nir, nir_opt_peephole_select, &peephole_select_options);
NIR_PASS(progress, nir, nir_opt_peephole_select,
&peephole_select_options);
NIR_PASS(progress, nir, nir_opt_idiv_const, 8);
NIR_PASS(progress, nir, nir_opt_algebraic);
NIR_PASS(progress, nir, nir_opt_constant_folding);
NIR_PASS(progress, nir, nir_opt_undef);
NIR_PASS(progress, nir, nir_lower_undef_to_zero);
NIR_PASS(progress, nir, nir_opt_dead_cf);
NIR_PASS(progress, nir, nir_opt_shrink_vectors, false);
bool loop_progress = false;
NIR_PASS(loop_progress, nir, nir_opt_loop);
progress |= loop_progress;
if (loop_progress) {
/* If nir_opt_loop makes progress, then we need to clean things up
* if we want any hope of nir_opt_if or nir_opt_loop_unroll to make
* progress.
*/
NIR_PASS(progress, nir, nir_copy_prop);
NIR_PASS(progress, nir, nir_opt_dce);
}
/* XXX: On Bifrost (G52), this cause a failure on
* "dEQP-VK.graphicsfuzz.spv-composite-phi" and is related to an unknown
* scheduling issue */
if (arch >= 9)
NIR_PASS(
progress, nir, nir_opt_if,
nir_opt_if_optimize_phi_true_false | nir_opt_if_avoid_64bit_phis);
NIR_PASS(progress, nir, nir_opt_phi_to_bool);
NIR_PASS(progress, nir, nir_opt_loop_unroll);
NIR_PASS(progress, nir, nir_opt_remove_phis);
NIR_PASS(progress, nir, nir_opt_undef);
} while (progress);
NIR_PASS(_, nir, nir_lower_undef_to_zero);
NIR_PASS(_, nir, nir_opt_shrink_vectors, false);
nir_load_store_vectorize_options vectorize_opts = {
.modes = nir_var_mem_global |
nir_var_mem_shared |