Mesa (main): ac/nir: Use gs_accepted variable after culling.

[GitLab Mirror]

Module: Mesa
Branch: main
Commit: a2d02c0c11009a7018e9760ed44d38131ccfd195
URL:    http://cgit.freedesktop.org/mesa/mesa/commit/?id=a2d02c0c11009a7018e9760ed44d38131ccfd195

Author: Timur Kristóf <timur.kristof at gmail.com>
Date:   Thu Jul 15 14:32:34 2021 +0200

ac/nir: Use gs_accepted variable after culling.

This prevents us from recalculating the EXEC mask later
in the shader, and removes the requirement for
counting the number of primitives.

The stats are better than expected because they also
show that some code that is still there is now DCE'd by ACO.

Fossil DB results on Sienna Cichlid (with NGGC on):

Totals from 58239 (45.27% of 128647) affected shaders:
SpillSGPRs: 330 -> 340 (+3.03%)
CodeSize: 166356072 -> 162805724 (-2.13%)
Instrs: 31920041 -> 31089256 (-2.60%)
Latency: 138815742 -> 138113669 (-0.51%); split: -0.54%, +0.03%
InvThroughput: 22459553 -> 22404840 (-0.24%); split: -0.26%, +0.02%
SClause: 753746 -> 753765 (+0.00%); split: -0.00%, +0.01%
Copies: 3226647 -> 3268973 (+1.31%); split: -0.45%, +1.76%
Branches: 1223441 -> 1223440 (-0.00%); split: -0.00%, +0.00%
PreSGPRs: 2025339 -> 2091013 (+3.24%)

No Fossil DB changes with NGGC off.

Signed-off-by: Timur Kristóf <timur.kristof at gmail.com>
Reviewed-by: Daniel Schürmann <daniel at schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11908>

---

 src/amd/common/ac_nir_lower_ngg.c | 11 +++++++++--
 1 file changed, 9 insertions(+), 2 deletions(-)

diff --git a/src/amd/common/ac_nir_lower_ngg.c b/src/amd/common/ac_nir_lower_ngg.c
index 626313f6436..92b0a63e231 100644
--- a/src/amd/common/ac_nir_lower_ngg.c
+++ b/src/amd/common/ac_nir_lower_ngg.c
@@ -318,7 +318,11 @@ emit_ngg_nogs_prim_exp_arg(nir_builder *b, lower_ngg_nogs_state *st)
 static void
 emit_ngg_nogs_prim_export(nir_builder *b, lower_ngg_nogs_state *st, nir_ssa_def *arg)
 {
-   nir_if *if_gs_thread = nir_push_if(b, nir_build_has_input_primitive_amd(b));
+   nir_ssa_def *gs_thread = st->gs_accepted_var
+                            ? nir_load_var(b, st->gs_accepted_var)
+                            : nir_build_has_input_primitive_amd(b);
+
+   nir_if *if_gs_thread = nir_push_if(b, gs_thread);
    {
       if (!arg)
          arg = emit_ngg_nogs_prim_exp_arg(b, st);
@@ -584,7 +588,8 @@ compact_vertices_after_culling(nir_builder *b,
 
    /* If all vertices are culled, set primitive count to 0 as well. */
    nir_ssa_def *num_exported_prims = nir_build_load_workgroup_num_input_primitives_amd(b);
-   num_exported_prims = nir_bcsel(b, nir_ieq_imm(b, num_live_vertices_in_workgroup, 0u), nir_imm_int(b, 0u), num_exported_prims);
+   nir_ssa_def *fully_culled = nir_ieq_imm(b, num_live_vertices_in_workgroup, 0u);
+   num_exported_prims = nir_bcsel(b, fully_culled, nir_imm_int(b, 0u), num_exported_prims);
 
    nir_if *if_wave_0 = nir_push_if(b, nir_ieq(b, nir_build_load_subgroup_id(b), nir_imm_int(b, 0)));
    {
@@ -642,6 +647,7 @@ compact_vertices_after_culling(nir_builder *b,
    nir_pop_if(b, if_gs_accepted);
 
    nir_store_var(b, es_accepted_var, es_survived, 0x1u);
+   nir_store_var(b, gs_accepted_var, nir_bcsel(b, fully_culled, nir_imm_false(b), nir_build_has_input_primitive_amd(b)), 0x1u);
 }
 
 static void
@@ -958,6 +964,7 @@ add_deferred_attribute_culling(nir_builder *b, nir_cf_list *original_extracted_c
    nir_pop_if(b, if_es_thread);
 
    nir_store_var(b, es_accepted_var, es_thread, 0x1u);
+   nir_store_var(b, gs_accepted_var, nir_build_has_input_primitive_amd(b), 0x1u);
 
    /* Remove all non-position outputs, and put the position output into the variable. */
    nir_metadata_preserve(impl, nir_metadata_none);