4 files changed, 44 insertions, 4 deletions

diff --git a/src/gallium/drivers/radeonsi/si_pipe.h b/src/gallium/drivers/radeonsi/si_pipe.h
index 0d00a9b17b4..0fd3ec46755 100644
--- a/src/gallium/drivers/radeonsi/si_pipe.h
+++ b/src/gallium/drivers/radeonsi/si_pipe.h
@@ -1105,6 +1105,9 @@ struct si_context {
 	unsigned			num_resident_handles;
 	uint64_t			num_alloc_tex_transfer_bytes;
 	unsigned			last_tex_ps_draw_ratio; /* for query */
+	unsigned			compute_num_verts_accepted;
+	unsigned			compute_num_verts_rejected;
+	unsigned			compute_num_verts_ineligible; /* due to low vertex count */
 	unsigned			context_roll;
 
 	/* Queries. */
diff --git a/src/gallium/drivers/radeonsi/si_query.c b/src/gallium/drivers/radeonsi/si_query.c
index 1dd9249d57c..1d541f194f3 100644
--- a/src/gallium/drivers/radeonsi/si_query.c
+++ b/src/gallium/drivers/radeonsi/si_query.c
@@ -255,6 +255,15 @@ static bool si_query_sw_begin(struct si_context *sctx,
 		query->begin_result =
 			p_atomic_read(&sctx->screen->num_shader_cache_hits);
 		break;
+	case SI_QUERY_PD_NUM_PRIMS_ACCEPTED:
+		query->begin_result = sctx->compute_num_verts_accepted;
+		break;
+	case SI_QUERY_PD_NUM_PRIMS_REJECTED:
+		query->begin_result = sctx->compute_num_verts_rejected;
+		break;
+	case SI_QUERY_PD_NUM_PRIMS_INELIGIBLE:
+		query->begin_result = sctx->compute_num_verts_ineligible;
+		break;
 	case SI_QUERY_GPIN_ASIC_ID:
 	case SI_QUERY_GPIN_NUM_SIMD:
 	case SI_QUERY_GPIN_NUM_RB:
@@ -420,6 +429,15 @@ static bool si_query_sw_end(struct si_context *sctx,
 		query->end_result =
 			p_atomic_read(&sctx->screen->num_shader_cache_hits);
 		break;
+	case SI_QUERY_PD_NUM_PRIMS_ACCEPTED:
+		query->end_result = sctx->compute_num_verts_accepted;
+		break;
+	case SI_QUERY_PD_NUM_PRIMS_REJECTED:
+		query->end_result = sctx->compute_num_verts_rejected;
+		break;
+	case SI_QUERY_PD_NUM_PRIMS_INELIGIBLE:
+		query->end_result = sctx->compute_num_verts_ineligible;
+		break;
 	case SI_QUERY_GPIN_ASIC_ID:
 	case SI_QUERY_GPIN_NUM_SIMD:
 	case SI_QUERY_GPIN_NUM_RB:
@@ -465,6 +483,12 @@ static bool si_query_sw_get_result(struct si_context *sctx,
 		result->u64 = (query->end_result - query->begin_result) * 100 /
 			      (query->end_time - query->begin_time);
 		return true;
+	case SI_QUERY_PD_NUM_PRIMS_ACCEPTED:
+	case SI_QUERY_PD_NUM_PRIMS_REJECTED:
+	case SI_QUERY_PD_NUM_PRIMS_INELIGIBLE:
+		result->u64 = ((unsigned)query->end_result -
+			       (unsigned)query->begin_result) / 3;
+		return true;
 	case SI_QUERY_GPIN_ASIC_ID:
 		result->u32 = 0;
 		return true;
@@ -1782,6 +1806,10 @@ static struct pipe_driver_query_info si_driver_query_list[] = {
 	X("GPU-surf-sync-busy",		GPU_SURF_SYNC_BUSY,	UINT64, AVERAGE),
 	X("GPU-cp-dma-busy",		GPU_CP_DMA_BUSY,	UINT64, AVERAGE),
 	X("GPU-scratch-ram-busy",	GPU_SCRATCH_RAM_BUSY,	UINT64, AVERAGE),
+
+	X("pd-num-prims-accepted",	PD_NUM_PRIMS_ACCEPTED,	UINT64, AVERAGE),
+	X("pd-num-prims-rejected",	PD_NUM_PRIMS_REJECTED,	UINT64, AVERAGE),
+	X("pd-num-prims-ineligible",	PD_NUM_PRIMS_INELIGIBLE,UINT64, AVERAGE),
 };
 
 #undef X
diff --git a/src/gallium/drivers/radeonsi/si_query.h b/src/gallium/drivers/radeonsi/si_query.h
index 6ff0a239cb3..c1918f4df2d 100644
--- a/src/gallium/drivers/radeonsi/si_query.h
+++ b/src/gallium/drivers/radeonsi/si_query.h
@@ -112,6 +112,9 @@ enum {
 	SI_QUERY_GPIN_NUM_SE,
 	SI_QUERY_TIME_ELAPSED_SDMA,
 	SI_QUERY_TIME_ELAPSED_SDMA_SI, /* emulated, measured on the CPU */
+	SI_QUERY_PD_NUM_PRIMS_ACCEPTED,
+	SI_QUERY_PD_NUM_PRIMS_REJECTED,
+	SI_QUERY_PD_NUM_PRIMS_INELIGIBLE,
 
 	SI_QUERY_FIRST_PERFCOUNTER = PIPE_QUERY_DRIVER_SPECIFIC + 100,
 };
diff --git a/src/gallium/drivers/radeonsi/si_state_draw.c b/src/gallium/drivers/radeonsi/si_state_draw.c
index b4b3fe323ee..485efcb0dff 100644
--- a/src/gallium/drivers/radeonsi/si_state_draw.c
+++ b/src/gallium/drivers/radeonsi/si_state_draw.c
@@ -1591,14 +1591,17 @@ static void si_draw_vbo(struct pipe_context *ctx, const struct pipe_draw_info *i
 			}
 		}
 	} else {
-		direct_count = info->count * instance_count;
+		/* Multiply by 3 for strips and fans to get an approximate vertex
+		 * count as triangles. */
+		direct_count = info->count * instance_count *
+			       (prim == PIPE_PRIM_TRIANGLES ? 1 : 3);
 	}
 
 	/* Determine if we can use the primitive discard compute shader. */
 	if (si_compute_prim_discard_enabled(sctx) &&
-	    /* Multiply by 3 for strips and fans to get the vertex count as triangles. */
-	    direct_count * (prim == PIPE_PRIM_TRIANGLES ? 1 : 3) >
-	    sctx->prim_discard_vertex_count_threshold &&
+	    (direct_count > sctx->prim_discard_vertex_count_threshold ?
+	     (sctx->compute_num_verts_rejected += direct_count, true) : /* Add, then return true. */
+	     (sctx->compute_num_verts_ineligible += direct_count, false)) && /* Add, then return false. */
 	    (!info->count_from_stream_output || pd_msg("draw_opaque")) &&
 	    (primitive_restart ?
 	     /* Supported prim types with primitive restart: */
@@ -1648,10 +1651,13 @@ static void si_draw_vbo(struct pipe_context *ctx, const struct pipe_draw_info *i
 			index_size = 4;
 			instance_count = 1;
 			primitive_restart = false;
+			sctx->compute_num_verts_rejected -= direct_count;
+			sctx->compute_num_verts_accepted += direct_count;
 			break;
 		case SI_PRIM_DISCARD_DISABLED:
 			break;
 		case SI_PRIM_DISCARD_DRAW_SPLIT:
+			sctx->compute_num_verts_rejected -= direct_count;
 			goto return_cleanup;
 		}
 	}