4 files changed, 65 insertions, 5 deletions

diff --git a/src/gallium/drivers/r600/r600_pipe.c b/src/gallium/drivers/r600/r600_pipe.c
index 81448d49ee9..0e113a1b827 100644
--- a/src/gallium/drivers/r600/r600_pipe.c
+++ b/src/gallium/drivers/r600/r600_pipe.c
@@ -1077,6 +1077,24 @@ static uint64_t r600_get_timestamp(struct pipe_screen *screen)
 			rscreen->info.r600_clock_crystal_freq;
 }
 
+static int r600_get_driver_query_info(struct pipe_screen *screen,
+				      unsigned index,
+				      struct pipe_driver_query_info *info)
+{
+	struct pipe_driver_query_info list[] = {
+		{"draw-calls", R600_QUERY_DRAW_CALLS, 0},
+	};
+
+	if (!info)
+		return Elements(list);
+
+	if (index >= Elements(list))
+		return 0;
+
+	*info = list[index];
+	return 1;
+}
+
 struct pipe_screen *r600_screen_create(struct radeon_winsys *ws)
 {
 	struct r600_screen *rscreen = CALLOC_STRUCT(r600_screen);
@@ -1183,6 +1201,7 @@ struct pipe_screen *r600_screen_create(struct radeon_winsys *ws)
 	rscreen->screen.fence_reference = r600_fence_reference;
 	rscreen->screen.fence_signalled = r600_fence_signalled;
 	rscreen->screen.fence_finish = r600_fence_finish;
+	rscreen->screen.get_driver_query_info = r600_get_driver_query_info;
 	r600_init_screen_resource_functions(&rscreen->screen);
 
 	util_format_s3tc_init();
diff --git a/src/gallium/drivers/r600/r600_pipe.h b/src/gallium/drivers/r600/r600_pipe.h
index 285d45fd371..9c3a4c2b229 100644
--- a/src/gallium/drivers/r600/r600_pipe.h
+++ b/src/gallium/drivers/r600/r600_pipe.h
@@ -75,6 +75,8 @@
 #define R600_CONTEXT_PS_PARTIAL_FLUSH		(1 << 6)
 #define R600_CONTEXT_FLUSH_AND_INV_DB_META      (1 << 7)
 
+#define R600_QUERY_DRAW_CALLS		(PIPE_QUERY_DRIVER_SPECIFIC + 0)
+
 struct r600_context;
 struct r600_bytecode;
 struct r600_shader_key;
@@ -476,6 +478,9 @@ struct r600_query {
 	unsigned				num_cs_dw;
 	/* linked list of queries */
 	struct list_head			list;
+	/* for custom non-GPU queries */
+	uint64_t begin_result;
+	uint64_t end_result;
 };
 
 struct r600_so_target {
@@ -621,6 +626,7 @@ struct r600_context {
 	unsigned			num_cs_dw_nontimer_queries_suspend;
 	/* If queries have been suspended. */
 	bool				nontimer_queries_suspended;
+	unsigned			num_draw_calls;
 
 	/* Render condition. */
 	struct pipe_query		*current_render_cond;
diff --git a/src/gallium/drivers/r600/r600_query.c b/src/gallium/drivers/r600/r600_query.c
index 782ad26e3b2..6ee4bbd3a00 100644
--- a/src/gallium/drivers/r600/r600_query.c
+++ b/src/gallium/drivers/r600/r600_query.c
@@ -41,6 +41,13 @@ static struct r600_resource *r600_new_query_buffer(struct r600_context *ctx, uns
 {
 	unsigned j, i, num_results, buf_size = 4096;
 	uint32_t *results;
+
+	/* Non-GPU queries. */
+	switch (type) {
+	case R600_QUERY_DRAW_CALLS:
+		return NULL;
+	}
+
 	/* Queries are normally read by the CPU after
 	 * being written by the gpu, hence staging is probably a good
 	 * usage pattern.
@@ -270,8 +277,8 @@ static void r600_emit_query_predication(struct r600_context *ctx, struct r600_qu
 static struct pipe_query *r600_create_query(struct pipe_context *ctx, unsigned query_type)
 {
 	struct r600_context *rctx = (struct r600_context *)ctx;
-
 	struct r600_query *query;
+	bool skip_allocation = false;
 
 	query = CALLOC_STRUCT(r600_query);
 	if (query == NULL)
@@ -301,16 +308,22 @@ static struct pipe_query *r600_create_query(struct pipe_context *ctx, unsigned q
 		query->result_size = 32;
 		query->num_cs_dw = 6;
 		break;
+	/* Non-GPU queries. */
+	case R600_QUERY_DRAW_CALLS:
+		skip_allocation = true;
+		break;
 	default:
 		assert(0);
 		FREE(query);
 		return NULL;
 	}
 
-	query->buffer.buf = r600_new_query_buffer(rctx, query_type);
-	if (!query->buffer.buf) {
-		FREE(query);
-		return NULL;
+	if (!skip_allocation) {
+		query->buffer.buf = r600_new_query_buffer(rctx, query_type);
+		if (!query->buffer.buf) {
+			FREE(query);
+			return NULL;
+		}
 	}
 	return (struct pipe_query*)query;
 }
@@ -343,6 +356,13 @@ static void r600_begin_query(struct pipe_context *ctx, struct pipe_query *query)
 		return;
 	}
 
+	/* Non-GPU queries. */
+	switch (rquery->type) {
+	case R600_QUERY_DRAW_CALLS:
+		rquery->begin_result = rctx->num_draw_calls;
+		return;
+	}
+
 	/* Discard the old query buffers. */
 	while (prev) {
 		struct r600_query_buffer *qbuf = prev;
@@ -373,6 +393,13 @@ static void r600_end_query(struct pipe_context *ctx, struct pipe_query *query)
 	struct r600_context *rctx = (struct r600_context *)ctx;
 	struct r600_query *rquery = (struct r600_query *)query;
 
+	/* Non-GPU queries. */
+	switch (rquery->type) {
+	case R600_QUERY_DRAW_CALLS:
+		rquery->end_result = rctx->num_draw_calls;
+		return;
+	}
+
 	r600_emit_query_end(rctx, rquery);
 
 	if (r600_query_needs_begin(rquery->type) && !r600_is_timer_query(rquery->type)) {
@@ -407,6 +434,13 @@ static boolean r600_get_query_buffer_result(struct r600_context *ctx,
 	unsigned results_base = 0;
 	char *map;
 
+	/* Non-GPU queries. */
+	switch (query->type) {
+	case R600_QUERY_DRAW_CALLS:
+		result->u64 = query->end_result - query->begin_result;
+		return TRUE;
+	}
+
 	map = r600_buffer_mmap_sync_with_rings(ctx, qbuf->buf,
 						PIPE_TRANSFER_READ |
 						(wait ? 0 : PIPE_TRANSFER_DONTBLOCK));
diff --git a/src/gallium/drivers/r600/r600_state_common.c b/src/gallium/drivers/r600/r600_state_common.c
index 34c70edf575..8190873ce7b 100644
--- a/src/gallium/drivers/r600/r600_state_common.c
+++ b/src/gallium/drivers/r600/r600_state_common.c
@@ -1517,6 +1517,7 @@ static void r600_draw_vbo(struct pipe_context *ctx, const struct pipe_draw_info
 	}
 
 	pipe_resource_reference(&ib.buffer, NULL);
+	rctx->num_draw_calls++;
 }
 
 void r600_draw_rectangle(struct blitter_context *blitter,