AlexIndustrial/mesa
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

ac/nir/lower_ngg: rename clip_cull_dist_mask and use it correctly

Browse Source 
We incorrectly used it to determine whether the shader should cull, which
luckily had no effect because it wasn't used everywhere.

cull_clipdist_mask should be used instead, which also reflects whether
clip planes are enabled in GL.

clip_cull_dist_mask is renamed to export_clipdist_mask to make it clear.

Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/35352>
This commit is contained in:
Marek Olšák
2025-05-30 06:17:45 -04:00
committed by Marge Bot
parent f6af3c0e17
commit ed0f393607
5 changed files with 15 additions and 15 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/amd/common/nir/ac_nir.h
+4 -3
View File
@@ -165,16 +165,17 @@ typedef struct {
unsigned max_workgroup_size;
unsigned wave_size;
uint8_t clip_cull_dist_mask;
/* The mask of clip and cull distances that the shader should export. */
uint8_t export_clipdist_mask;
/* The mask of clip and cull distances that the shader should cull against.
* If no clip and cull distance outputs are present, it will load clip planes and cull
* either against CLIP_VERTEX or POS.
*/
uint8_t cull_clipdist_mask;
bool write_pos_to_clipvertex;
/* Remove clip/cull distance components that are missing in clip_cull_dist_mask, improving
/* Remove clip/cull distance components that are missing in export_clipdist_mask, improving
* throughput by up to 50% (3 pos exports -> 2 pos exports). The caller shouldn't set no-op
* components (>= 0) in clip_cull_dist_mask to remove those completely. No-op components
* components (>= 0) in export_clipdist_mask to remove those completely. No-op components
* should be determined by nir_opt_clip_cull_const before this.
*/
bool pack_clip_cull_distances;
src/amd/common/nir/ac_nir_lower_ngg.c
+5 -6
View File
@@ -359,8 +359,7 @@ remove_culling_shader_output(nir_builder *b, nir_intrinsic_instr *intrin, void *
unsigned base = io_sem.location == VARYING_SLOT_CLIP_DIST1 ? 4 : 0;
base += component;
/* valid clipdist component mask */
unsigned mask = (s->options->clip_cull_dist_mask >> base) & writemask;
unsigned mask = (s->options->cull_clipdist_mask >> base) & writemask;
u_foreach_bit(i, mask) {
add_clipdist_bit(b, nir_channel(b, store_val, i), base + i,
s->clipdist_neg_mask_var);
@@ -369,7 +368,8 @@ remove_culling_shader_output(nir_builder *b, nir_intrinsic_instr *intrin, void *
break;
}
case VARYING_SLOT_CLIP_VERTEX:
ac_nir_store_var_components(b, s->clip_vertex_var, store_val, component, writemask);
if (s->options->cull_clipdist_mask)
ac_nir_store_var_components(b, s->clip_vertex_var, store_val, component, writemask);
break;
default:
break;
@@ -1040,8 +1040,7 @@ add_deferred_attribute_culling(nir_builder *b, nir_cf_list *original_extracted_c
/* Relative patch ID is a special case because it doesn't need an extra dword, repack separately. */
s->repacked_rel_patch_id = nir_local_variable_create(impl, glsl_uint_type(), "repacked_rel_patch_id");
if (s->options->clip_cull_dist_mask ||
s->options->cull_clipdist_mask) {
if (s->options->cull_clipdist_mask) {
s->clip_vertex_var =
nir_local_variable_create(impl, glsl_vec4_type(), "clip_vertex");
s->clipdist_neg_mask_var =
@@ -1746,7 +1745,7 @@ ac_nir_lower_ngg_nogs(nir_shader *shader, const ac_nir_lower_ngg_options *option
}
ac_nir_export_position(b, options->hw_info->gfx_level,
options->clip_cull_dist_mask,
options->export_clipdist_mask,
options->write_pos_to_clipvertex,
options->pack_clip_cull_distances,
!options->has_param_exports,
src/amd/common/nir/ac_nir_lower_ngg_gs.c
+1 -1
View File
@@ -468,7 +468,7 @@ ngg_gs_emit_output(nir_builder *b, nir_def *max_num_out_vtx, nir_def *max_num_ou
export_outputs &= ~VARYING_BIT_LAYER;
ac_nir_export_position(b, s->options->hw_info->gfx_level,
s->options->clip_cull_dist_mask,
s->options->export_clipdist_mask,
s->options->write_pos_to_clipvertex,
s->options->pack_clip_cull_distances,
!s->options->has_param_exports,
src/amd/vulkan/radv_shader.c
+4 -4
View File
@@ -791,7 +791,7 @@ radv_lower_ngg(struct radv_device *device, struct radv_shader_stage *ngg_stage,
options.hw_info = &pdev->info;
options.max_workgroup_size = info->workgroup_size;
options.wave_size = info->wave_size;
options.clip_cull_dist_mask = info->outinfo.clip_dist_mask | info->outinfo.cull_dist_mask;
options.export_clipdist_mask = info->outinfo.clip_dist_mask | info->outinfo.cull_dist_mask;
options.vs_output_param_offset = info->outinfo.vs_output_param_offset;
options.has_param_exports = info->outinfo.param_exports || info->outinfo.prim_param_exports;
options.can_cull = nir->info.stage != MESA_SHADER_GEOMETRY && info->has_ngg_culling;
@@ -825,9 +825,9 @@ radv_lower_ngg(struct radv_device *device, struct radv_shader_stage *ngg_stage,
unsigned hw_workgroup_size = ALIGN(info->workgroup_size, info->wave_size);
bool scratch_ring = false;
NIR_PASS(_, nir, ac_nir_lower_ngg_mesh, &pdev->info, options.clip_cull_dist_mask,
options.vs_output_param_offset, options.has_param_exports, &scratch_ring, info->wave_size,
hw_workgroup_size, gfx_state->has_multiview_view_index, info->ms.has_query, pdev->mesh_fast_launch_2);
NIR_PASS(_, nir, ac_nir_lower_ngg_mesh, &pdev->info, options.export_clipdist_mask, options.vs_output_param_offset,
options.has_param_exports, &scratch_ring, info->wave_size, hw_workgroup_size,
gfx_state->has_multiview_view_index, info->ms.has_query, pdev->mesh_fast_launch_2);
ngg_stage->info.ms.needs_ms_scratch_ring = scratch_ring;
} else {
unreachable("invalid SW stage passed to radv_lower_ngg");
src/gallium/drivers/radeonsi/si_shader.c
+1 -1
View File
@@ -1131,7 +1131,7 @@ static void si_lower_ngg(struct si_shader *shader, nir_shader *nir,
.disable_streamout = !shader->info.num_streamout_vec4s,
.vs_output_param_offset = temp_info->vs_output_param_offset,
.has_param_exports = shader->info.nr_param_exports,
.clip_cull_dist_mask = clip_cull_dist_mask,
.export_clipdist_mask = clip_cull_dist_mask,
.kill_pointsize = key->ge.opt.kill_pointsize,
.kill_layer = key->ge.opt.kill_layer,
.force_vrs = sel->screen->options.vrs2x2,