ac/nir,radeonsi: replace llvm ngg gs invocation query with nir

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Signed-off-by: Qiang Yu <yuq825@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20156>

diff --git a/src/amd/common/ac_nir_lower_ngg.c b/src/amd/common/ac_nir_lower_ngg.c
index a3e2658..bf075a3 100644 (file)
--- a/src/amd/common/ac_nir_lower_ngg.c
+++ b/src/amd/common/ac_nir_lower_ngg.c
@@ -2469,6 +2469,8 @@ ngg_gs_shader_query(nir_builder *b, lower_ngg_gs_state *s)
 
             if (count)
                nir_atomic_add_gs_emit_prim_count_amd(b, count);
+
+            nir_atomic_add_gs_invocation_count_amd(b, num_active_threads);
          }
          nir_pop_if(b, if_pipeline_query);
       }

diff --git a/src/gallium/drivers/radeonsi/gfx10_shader_ngg.c b/src/gallium/drivers/radeonsi/gfx10_shader_ngg.c
index 17d0bea..2392e73 100644 (file)
--- a/src/gallium/drivers/radeonsi/gfx10_shader_ngg.c
+++ b/src/gallium/drivers/radeonsi/gfx10_shader_ngg.c
@@ -124,35 +124,6 @@ void gfx10_ngg_export_vertex(struct ac_shader_abi *abi)
    si_llvm_build_vs_exports(ctx, outputs, num_outputs);
 }
 
-void gfx10_ngg_gs_emit_begin(struct si_shader_context *ctx)
-{
-   LLVMBuilderRef builder = ctx->ac.builder;
-   LLVMValueRef tmp;
-
-   if (ctx->screen->info.gfx_level < GFX11) {
-      tmp = si_is_gs_thread(ctx);
-      ac_build_ifcc(&ctx->ac, tmp, 15090);
-         {
-            tmp = GET_FIELD(ctx, GS_STATE_PIPELINE_STATS_EMU);
-            tmp = LLVMBuildTrunc(builder, tmp, ctx->ac.i1, "");
-            ac_build_ifcc(&ctx->ac, tmp, 5109); /* if (GS_PIPELINE_STATS_EMU) */
-            LLVMValueRef args[] = {
-               ctx->ac.i32_1,
-               ngg_get_emulated_counters_buf(ctx),
-               LLVMConstInt(ctx->ac.i32,
-                            si_query_pipestat_end_dw_offset(ctx->screen, PIPE_STAT_QUERY_GS_INVOCATIONS) * 4,
-                            false),
-               ctx->ac.i32_0,                            /* soffset */
-               ctx->ac.i32_0,                            /* cachepolicy */
-            };
-
-            ac_build_intrinsic(&ctx->ac, "llvm.amdgcn.raw.buffer.atomic.add.i32", ctx->ac.i32, args, 5, 0);
-            ac_build_endif(&ctx->ac, 5109);
-         }
-      ac_build_endif(&ctx->ac, 15090);
-   }
-}
-
 static void clamp_gsprims_to_esverts(unsigned *max_gsprims, unsigned max_esverts,
                                      unsigned min_verts_per_prim, bool use_adjacency)
 {

diff --git a/src/gallium/drivers/radeonsi/si_shader_internal.h b/src/gallium/drivers/radeonsi/si_shader_internal.h
index 93e3925..e258371 100644 (file)
--- a/src/gallium/drivers/radeonsi/si_shader_internal.h
+++ b/src/gallium/drivers/radeonsi/si_shader_internal.h
@@ -183,7 +183,6 @@ LLVMValueRef gfx10_get_thread_id_in_tg(struct si_shader_context *ctx);
 unsigned gfx10_ngg_get_vertices_per_prim(struct si_shader *shader);
 bool gfx10_ngg_export_prim_early(struct si_shader *shader);
 void gfx10_ngg_export_vertex(struct ac_shader_abi *abi);
-void gfx10_ngg_gs_emit_begin(struct si_shader_context *ctx);
 unsigned gfx10_ngg_get_scratch_dw_size(struct si_shader *shader);
 bool gfx10_ngg_calculate_subgroup_info(struct si_shader *shader);
 

diff --git a/src/gallium/drivers/radeonsi/si_shader_llvm.c b/src/gallium/drivers/radeonsi/si_shader_llvm.c
index 5a241c8..6605495 100644 (file)
--- a/src/gallium/drivers/radeonsi/si_shader_llvm.c
+++ b/src/gallium/drivers/radeonsi/si_shader_llvm.c
@@ -902,10 +902,6 @@ bool si_llvm_translate_nir(struct si_shader_context *ctx, struct si_shader *shad
           shader->key.ge.as_ngg && !shader->key.ge.as_es && !shader->key.ge.opt.ngg_culling)
          ac_build_s_barrier(&ctx->ac, ctx->stage);
 
-      /* NGG GS: handle GS_STATE_PIPELINE_STATS_EMU */
-      if (ctx->stage == MESA_SHADER_GEOMETRY && shader->key.ge.as_ngg)
-         gfx10_ngg_gs_emit_begin(ctx);
-
       LLVMValueRef thread_enabled = NULL;
 
       if ((ctx->stage == MESA_SHADER_GEOMETRY && !shader->key.ge.as_ngg) ||