fdo-mirrors/mesa
1
0
Fork 0
mirror of https://gitlab.freedesktop.org/mesa/mesa.git synced 2026-05-09 04:38:03 +02:00
Code Issues Projects Releases Packages Wiki Activity Actions 1

ac/nir: optimize barycentric_at_sample(sample_id) in ac_lower_ps_early

Browse source 
Replace it with barycentric_sample.

Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/33024>
This commit is contained in:
Marek Olšák 2025-01-02 22:08:16 -05:00
parent 43f6b2655e
commit 33134f9503
3 changed files with 25 additions and 4 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

16
src/amd/common/nir/ac_nir.h
View file

@ -275,7 +275,21 @@ typedef struct {
bool load_sample_positions_always_loads_current_ones;
bool dynamic_rasterization_samples;
int force_front_face; /* 0 -> keep, 1 -> set to true, -1 -> set to false */
unsigned ps_iter_samples; /* >= 2 forces sample interpolation, affects sample_mask_in lowering */
/* barycentrics:
* ps_iter_samples >= 2:
* * All barycentrics are changed to per-sample interpolation except at_offset/at_sample.
* * barycentric_at_sample(sample_id) is replaced by barycentric_sample.
*
* sample_mask_in:
* ps_iter_samples == 2, 4:
* * sample_mask_in is changed to (sample_mask_in & (ps_iter_mask << sample_id))
* ps_iter_samples == 8:
* * sample_mask_in is replaced by 1 << sample_id.
*
* When ps_iter_samples is equal to rasterization samples, set ps_iter_samples = 8 for this pass.
*/
unsigned ps_iter_samples;
/* fbfetch_output */
bool fbfetch_is_1D;

9
src/amd/common/nir/ac_nir_lower_ps_early.c
View file

@ -383,19 +383,26 @@ lower_ps_intrinsic(nir_builder *b, nir_intrinsic_instr *intrin, void *state)
return true;
case nir_intrinsic_load_barycentric_at_sample: {
unsigned mode = nir_intrinsic_interp_mode(intrin);
nir_def *sample_id = intrin->src[0].ssa;
if (s->options->force_center_interp_no_msaa) {
nir_def_replace(&intrin->def, nir_load_barycentric_pixel(b, 32, .interp_mode = mode));
return true;
}
if (s->options->ps_iter_samples >= 2 &&
sample_id->parent_instr->type == nir_instr_type_intrinsic &&
nir_instr_as_intrinsic(sample_id->parent_instr)->intrinsic == nir_intrinsic_load_sample_id) {
nir_def_replace(&intrin->def, nir_load_barycentric_sample(b, 32, .interp_mode = mode));
return true;
}
/* If load_sample_positions_always_loads_current_ones is true, load_sample_positions_amd
* always loads the sample positions that are currently set in the rasterizer state
* even if MSAA is disabled.
*/
nir_def *num_samples = s->options->load_sample_positions_always_loads_current_ones ?
nir_undef(b, 1, 32) : nir_load_rasterization_samples_amd(b);
nir_def *sample_id = intrin->src[0].ssa;
nir_def *sample_pos = nir_load_sample_positions_amd(b, 32, sample_id, num_samples);
sample_pos = nir_fadd_imm(b, sample_pos, -0.5f);

4
src/gallium/drivers/radeonsi/si_shader.c
View file

@ -2487,11 +2487,11 @@ static struct nir_shader *si_get_nir_shader(struct si_shader *shader, struct si_
key->ps.mono.interpolate_at_sample_force_center,
.load_sample_positions_always_loads_current_ones = true,
.force_front_face = key->ps.opt.force_front_face_input,
/* This forces per-sample interpolation (if at least 2) and lowers sample_mask_in. */
/* This does a lot of things. See the description in ac_nir_lower_ps_early_options. */
.ps_iter_samples = key->ps.part.prolog.samplemask_log_ps_iter ?
(1 << key->ps.part.prolog.samplemask_log_ps_iter) :
(key->ps.part.prolog.force_persp_sample_interp ||
key->ps.part.prolog.force_linear_sample_interp ? 2 : 1),
key->ps.part.prolog.force_linear_sample_interp ? 2 : 0),
.fbfetch_is_1D = key->ps.mono.fbfetch_is_1D,
.fbfetch_layered = key->ps.mono.fbfetch_layered,