gallium/radeon: add driver queries for compute/dma call stats and spills

also print the average count per frame

Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

6 files changed, 45 insertions, 1 deletions

diff --git a/src/gallium/drivers/radeon/r600_pipe_common.h b/src/gallium/drivers/radeon/r600_pipe_common.h
index 57fa9e367a..fb6d1a56d1 100644
--- a/src/gallium/drivers/radeon/r600_pipe_common.h
+++ b/src/gallium/drivers/radeon/r600_pipe_common.h
@@ -505,6 +505,9 @@ struct r600_common_context {
 	unsigned			max_db; /* for OQ */
 	/* Misc stats. */
 	unsigned			num_draw_calls;
+	unsigned			num_spill_draw_calls;
+	unsigned			num_compute_calls;
+	unsigned			num_spill_compute_calls;
 	unsigned			num_dma_calls;
 	uint64_t			num_alloc_tex_transfer_bytes;
 
diff --git a/src/gallium/drivers/radeon/r600_query.c b/src/gallium/drivers/radeon/r600_query.c
index e3626144dd..f090f1c011 100644
--- a/src/gallium/drivers/radeon/r600_query.c
+++ b/src/gallium/drivers/radeon/r600_query.c
@@ -75,6 +75,18 @@ static boolean r600_query_sw_begin(struct r600_common_context *rctx,
 	case R600_QUERY_DRAW_CALLS:
 		query->begin_result = rctx->num_draw_calls;
 		break;
+	case R600_QUERY_SPILL_DRAW_CALLS:
+		query->begin_result = rctx->num_spill_draw_calls;
+		break;
+	case R600_QUERY_COMPUTE_CALLS:
+		query->begin_result = rctx->num_compute_calls;
+		break;
+	case R600_QUERY_SPILL_COMPUTE_CALLS:
+		query->begin_result = rctx->num_spill_compute_calls;
+		break;
+	case R600_QUERY_DMA_CALLS:
+		query->begin_result = rctx->num_dma_calls;
+		break;
 	case R600_QUERY_REQUESTED_VRAM:
 	case R600_QUERY_REQUESTED_GTT:
 	case R600_QUERY_VRAM_USAGE:
@@ -127,6 +139,18 @@ static bool r600_query_sw_end(struct r600_common_context *rctx,
 	case R600_QUERY_DRAW_CALLS:
 		query->end_result = rctx->num_draw_calls;
 		break;
+	case R600_QUERY_SPILL_DRAW_CALLS:
+		query->end_result = rctx->num_spill_draw_calls;
+		break;
+	case R600_QUERY_COMPUTE_CALLS:
+		query->end_result = rctx->num_compute_calls;
+		break;
+	case R600_QUERY_SPILL_COMPUTE_CALLS:
+		query->end_result = rctx->num_spill_compute_calls;
+		break;
+	case R600_QUERY_DMA_CALLS:
+		query->end_result = rctx->num_dma_calls;
+		break;
 	case R600_QUERY_REQUESTED_VRAM:
 	case R600_QUERY_REQUESTED_GTT:
 	case R600_QUERY_VRAM_USAGE:
@@ -1139,7 +1163,11 @@ err:
 static struct pipe_driver_query_info r600_driver_query_list[] = {
 	X("num-compilations",		NUM_COMPILATIONS,	UINT64, CUMULATIVE),
 	X("num-shaders-created",	NUM_SHADERS_CREATED,	UINT64, CUMULATIVE),
-	X("draw-calls",			DRAW_CALLS,		UINT64, CUMULATIVE),
+	X("draw-calls",			DRAW_CALLS,		UINT64, AVERAGE),
+	X("spill-draw-calls",		SPILL_DRAW_CALLS,	UINT64, AVERAGE),
+	X("compute-calls",		COMPUTE_CALLS,		UINT64, AVERAGE),
+	X("spill-compute-calls",	SPILL_COMPUTE_CALLS,	UINT64, AVERAGE),
+	X("dma-calls",			DMA_CALLS,		UINT64, AVERAGE),
 	X("requested-VRAM",		REQUESTED_VRAM,		BYTES, AVERAGE),
 	X("requested-GTT",		REQUESTED_GTT,		BYTES, AVERAGE),
 	X("buffer-wait-time",		BUFFER_WAIT_TIME,	MICROSECONDS, CUMULATIVE),
diff --git a/src/gallium/drivers/radeon/r600_query.h b/src/gallium/drivers/radeon/r600_query.h
index a3c9aa29ab..b573a17bd0 100644
--- a/src/gallium/drivers/radeon/r600_query.h
+++ b/src/gallium/drivers/radeon/r600_query.h
@@ -42,6 +42,10 @@ struct r600_resource;
 
 enum {
 	R600_QUERY_DRAW_CALLS = PIPE_QUERY_DRIVER_SPECIFIC,
+	R600_QUERY_SPILL_DRAW_CALLS,
+	R600_QUERY_COMPUTE_CALLS,
+	R600_QUERY_SPILL_COMPUTE_CALLS,
+	R600_QUERY_DMA_CALLS,
 	R600_QUERY_REQUESTED_VRAM,
 	R600_QUERY_REQUESTED_GTT,
 	R600_QUERY_BUFFER_WAIT_TIME,
diff --git a/src/gallium/drivers/radeonsi/si_compute.c b/src/gallium/drivers/radeonsi/si_compute.c
index 09cbe1337e..2f7e1721c8 100644
--- a/src/gallium/drivers/radeonsi/si_compute.c
+++ b/src/gallium/drivers/radeonsi/si_compute.c
@@ -308,6 +308,8 @@ static bool si_switch_compute_shader(struct si_context *sctx,
 
 	sctx->cs_shader_state.emitted_program = program;
 	sctx->cs_shader_state.offset = offset;
+	sctx->cs_shader_state.uses_scratch =
+		config->scratch_bytes_per_wave != 0;
 
 	return true;
 }
@@ -487,6 +489,10 @@ static void si_launch_grid(
 	si_emit_dispatch_packets(sctx, info);
 
 	si_ce_post_draw_synchronization(sctx);
+
+	sctx->b.num_compute_calls++;
+	if (sctx->cs_shader_state.uses_scratch)
+		sctx->b.num_spill_compute_calls++;
 }
 
 
diff --git a/src/gallium/drivers/radeonsi/si_pipe.h b/src/gallium/drivers/radeonsi/si_pipe.h
index d4c55146ef..7e68bb707d 100644
--- a/src/gallium/drivers/radeonsi/si_pipe.h
+++ b/src/gallium/drivers/radeonsi/si_pipe.h
@@ -136,6 +136,7 @@ struct si_cs_shader_state {
 	struct si_compute		*emitted_program;
 	unsigned			offset;
 	bool				initialized;
+	bool				uses_scratch;
 };
 
 struct si_textures_info {
diff --git a/src/gallium/drivers/radeonsi/si_state_draw.c b/src/gallium/drivers/radeonsi/si_state_draw.c
index 0a85d5a0ca..641becf7bc 100644
--- a/src/gallium/drivers/radeonsi/si_state_draw.c
+++ b/src/gallium/drivers/radeonsi/si_state_draw.c
@@ -987,6 +987,8 @@ void si_draw_vbo(struct pipe_context *ctx, const struct pipe_draw_info *info)
 
 	pipe_resource_reference(&ib.buffer, NULL);
 	sctx->b.num_draw_calls++;
+	if (G_0286E8_WAVESIZE(sctx->spi_tmpring_size))
+		sctx->b.num_spill_draw_calls++;
 }
 
 void si_trace_emit(struct si_context *sctx)