r600g: don't suspend timer queries for u_blitter

Timer queries should be able to measure the time spent in u_blitter as well.

Queries are split into two groups: the timer ones and the others (streamout,
occlusion), because we should only suspend non-timer queries for u_blitter,
and later if the non-timer queries are suspended, the context flush should
only suspend and resume the timer queries.

Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>

5 files changed, 88 insertions, 23 deletions

diff --git a/src/gallium/drivers/r600/r600_blit.c b/src/gallium/drivers/r600/r600_blit.c
index b9471b81af..d58abea487 100644
--- a/src/gallium/drivers/r600/r600_blit.c
+++ b/src/gallium/drivers/r600/r600_blit.c
@@ -45,7 +45,7 @@ static void r600_blitter_begin(struct pipe_context *ctx, enum r600_blitter_op op
 {
 	struct r600_context *rctx = (struct r600_context *)ctx;
 
-	r600_suspend_queries(rctx);
+	r600_suspend_nontimer_queries(rctx);
 
 	util_blitter_save_blend(rctx->blitter, rctx->states[R600_PIPE_STATE_BLEND]);
 	util_blitter_save_depth_stencil_alpha(rctx->blitter, rctx->states[R600_PIPE_STATE_DSA]);
@@ -95,7 +95,7 @@ static void r600_blitter_end(struct pipe_context *ctx)
 					       rctx->saved_render_cond_mode);
 		rctx->saved_render_cond = NULL;
 	}
-	r600_resume_queries(rctx);
+	r600_resume_nontimer_queries(rctx);
 }
 
 static unsigned u_num_layers(struct pipe_resource *r, unsigned level)
diff --git a/src/gallium/drivers/r600/r600_hw_context.c b/src/gallium/drivers/r600/r600_hw_context.c
index 9679458c98..5a48b3d963 100644
--- a/src/gallium/drivers/r600/r600_hw_context.c
+++ b/src/gallium/drivers/r600/r600_hw_context.c
@@ -814,7 +814,8 @@ void r600_need_cs_space(struct r600_context *ctx, unsigned num_dw,
 	}
 
 	/* Count in queries_suspend. */
-	num_dw += ctx->num_cs_dw_queries_suspend;
+	num_dw += ctx->num_cs_dw_nontimer_queries_suspend;
+	num_dw += ctx->num_cs_dw_timer_queries_suspend;
 
 	/* Count in streamout_end at the end of CS. */
 	num_dw += ctx->num_cs_dw_streamout_end;
@@ -1243,16 +1244,21 @@ void r600_context_flush(struct r600_context *ctx, unsigned flags)
 {
 	struct radeon_winsys_cs *cs = ctx->cs;
 	struct r600_block *enable_block = NULL;
-	bool queries_suspended = false;
+	bool timer_queries_suspended = false;
+	bool nontimer_queries_suspended = false;
 	bool streamout_suspended = false;
 
 	if (cs->cdw == ctx->atom_start_cs.atom.num_dw)
 		return;
 
 	/* suspend queries */
-	if (ctx->num_cs_dw_queries_suspend) {
-		r600_suspend_queries(ctx);
-		queries_suspended = true;
+	if (ctx->num_cs_dw_timer_queries_suspend) {
+		r600_suspend_timer_queries(ctx);
+		timer_queries_suspended = true;
+	}
+	if (ctx->num_cs_dw_nontimer_queries_suspend) {
+		r600_suspend_nontimer_queries(ctx);
+		nontimer_queries_suspended = true;
 	}
 
 	if (ctx->num_cs_dw_streamout_end) {
@@ -1284,8 +1290,11 @@ void r600_context_flush(struct r600_context *ctx, unsigned flags)
 	}
 
 	/* resume queries */
-	if (queries_suspended) {
-		r600_resume_queries(ctx);
+	if (timer_queries_suspended) {
+		r600_resume_timer_queries(ctx);
+	}
+	if (nontimer_queries_suspended) {
+		r600_resume_nontimer_queries(ctx);
 	}
 
 	/* set all valid group as dirty so they get reemited on
diff --git a/src/gallium/drivers/r600/r600_pipe.c b/src/gallium/drivers/r600/r600_pipe.c
index 9929baa226..b422d75307 100644
--- a/src/gallium/drivers/r600/r600_pipe.c
+++ b/src/gallium/drivers/r600/r600_pipe.c
@@ -248,7 +248,8 @@ static struct pipe_context *r600_create_context(struct pipe_screen *screen, void
 	rctx->chip_class = rscreen->chip_class;
 
 	LIST_INITHEAD(&rctx->dirty_states);
-	LIST_INITHEAD(&rctx->active_query_list);
+	LIST_INITHEAD(&rctx->active_timer_queries);
+	LIST_INITHEAD(&rctx->active_nontimer_queries);
 	LIST_INITHEAD(&rctx->dirty);
 	LIST_INITHEAD(&rctx->resource_dirty);
 	LIST_INITHEAD(&rctx->enable_list);
diff --git a/src/gallium/drivers/r600/r600_pipe.h b/src/gallium/drivers/r600/r600_pipe.h
index bc0ffc0385..62d831086f 100644
--- a/src/gallium/drivers/r600/r600_pipe.h
+++ b/src/gallium/drivers/r600/r600_pipe.h
@@ -309,8 +309,18 @@ struct r600_context {
 
 	/* The list of active queries. Only one query of each type can be active. */
 	int			num_occlusion_queries;
-	struct list_head	active_query_list;
-	unsigned		num_cs_dw_queries_suspend;
+
+	/* Manage queries in two separate groups:
+	 * The timer ones and the others (streamout, occlusion).
+	 *
+	 * We do this because we should only suspend non-timer queries for u_blitter,
+	 * and later if the non-timer queries are suspended, the context flush should
+	 * only suspend and resume the timer queries. */
+	struct list_head	active_timer_queries;
+	unsigned		num_cs_dw_timer_queries_suspend;
+	struct list_head	active_nontimer_queries;
+	unsigned		num_cs_dw_nontimer_queries_suspend;
+
 	unsigned		num_cs_dw_streamout_end;
 
 	unsigned		backend_mask;
@@ -395,8 +405,10 @@ void r600_flush(struct pipe_context *ctx, struct pipe_fence_handle **fence,
 
 /* r600_query.c */
 void r600_init_query_functions(struct r600_context *rctx);
-void r600_suspend_queries(struct r600_context *ctx);
-void r600_resume_queries(struct r600_context *ctx);
+void r600_suspend_nontimer_queries(struct r600_context *ctx);
+void r600_resume_nontimer_queries(struct r600_context *ctx);
+void r600_suspend_timer_queries(struct r600_context *ctx);
+void r600_resume_timer_queries(struct r600_context *ctx);
 
 /* r600_resource.c */
 void r600_init_context_resource_functions(struct r600_context *r600);
diff --git a/src/gallium/drivers/r600/r600_query.c b/src/gallium/drivers/r600/r600_query.c
index 25731c290f..53440ae734 100644
--- a/src/gallium/drivers/r600/r600_query.c
+++ b/src/gallium/drivers/r600/r600_query.c
@@ -25,6 +25,13 @@
 #include "util/u_memory.h"
 #include "r600_hw_context_priv.h"
 
+static bool r600_is_timer_query(unsigned type)
+{
+	return type == PIPE_QUERY_TIME_ELAPSED ||
+	       type == PIPE_QUERY_TIMESTAMP ||
+	       type == PIPE_QUERY_TIMESTAMP_DISJOINT;
+}
+
 static struct r600_resource *r600_new_query_buffer(struct r600_context *ctx, unsigned type)
 {
 	unsigned j, i, num_results, buf_size = 4096;
@@ -123,7 +130,11 @@ static void r600_emit_query_begin(struct r600_context *ctx, struct r600_query *q
 	cs->buf[cs->cdw++] = PKT3(PKT3_NOP, 0, 0);
 	cs->buf[cs->cdw++] = r600_context_bo_reloc(ctx, query->buffer.buf, RADEON_USAGE_WRITE);
 
-	ctx->num_cs_dw_queries_suspend += query->num_cs_dw;
+	if (r600_is_timer_query(query->type)) {
+		ctx->num_cs_dw_timer_queries_suspend += query->num_cs_dw;
+	} else {
+		ctx->num_cs_dw_nontimer_queries_suspend += query->num_cs_dw;
+	}
 }
 
 static void r600_emit_query_end(struct r600_context *ctx, struct r600_query *query)
@@ -167,7 +178,12 @@ static void r600_emit_query_end(struct r600_context *ctx, struct r600_query *que
 	cs->buf[cs->cdw++] = r600_context_bo_reloc(ctx, query->buffer.buf, RADEON_USAGE_WRITE);
 
 	query->buffer.results_end += query->result_size;
-	ctx->num_cs_dw_queries_suspend -= query->num_cs_dw;
+
+	if (r600_is_timer_query(query->type)) {
+		ctx->num_cs_dw_timer_queries_suspend -= query->num_cs_dw;
+	} else {
+		ctx->num_cs_dw_nontimer_queries_suspend -= query->num_cs_dw;
+	}
 }
 
 static void r600_emit_query_predication(struct r600_context *ctx, struct r600_query *query,
@@ -324,7 +340,12 @@ static void r600_begin_query(struct pipe_context *ctx, struct pipe_query *query)
 	r600_update_occlusion_query_state(rctx, rquery->type, 1);
 
 	r600_emit_query_begin(rctx, rquery);
-	LIST_ADDTAIL(&rquery->list, &rctx->active_query_list);
+
+	if (r600_is_timer_query(rquery->type)) {
+		LIST_ADDTAIL(&rquery->list, &rctx->active_timer_queries);
+	} else {
+		LIST_ADDTAIL(&rquery->list, &rctx->active_nontimer_queries);
+	}
 }
 
 static void r600_end_query(struct pipe_context *ctx, struct pipe_query *query)
@@ -525,23 +546,45 @@ static void r600_render_condition(struct pipe_context *ctx,
 	}
 }
 
-void r600_suspend_queries(struct r600_context *ctx)
+void r600_suspend_nontimer_queries(struct r600_context *ctx)
 {
 	struct r600_query *query;
 
-	LIST_FOR_EACH_ENTRY(query, &ctx->active_query_list, list) {
+	LIST_FOR_EACH_ENTRY(query, &ctx->active_nontimer_queries, list) {
 		r600_emit_query_end(ctx, query);
 	}
-	assert(ctx->num_cs_dw_queries_suspend == 0);
+	assert(ctx->num_cs_dw_nontimer_queries_suspend == 0);
+}
+
+void r600_resume_nontimer_queries(struct r600_context *ctx)
+{
+	struct r600_query *query;
+
+	assert(ctx->num_cs_dw_nontimer_queries_suspend == 0);
+
+	LIST_FOR_EACH_ENTRY(query, &ctx->active_nontimer_queries, list) {
+		r600_emit_query_begin(ctx, query);
+	}
+}
+
+void r600_suspend_timer_queries(struct r600_context *ctx)
+{
+	struct r600_query *query;
+
+	LIST_FOR_EACH_ENTRY(query, &ctx->active_timer_queries, list) {
+		r600_emit_query_end(ctx, query);
+	}
+
+	assert(ctx->num_cs_dw_timer_queries_suspend == 0);
 }
 
-void r600_resume_queries(struct r600_context *ctx)
+void r600_resume_timer_queries(struct r600_context *ctx)
 {
 	struct r600_query *query;
 
-	assert(ctx->num_cs_dw_queries_suspend == 0);
+	assert(ctx->num_cs_dw_timer_queries_suspend == 0);
 
-	LIST_FOR_EACH_ENTRY(query, &ctx->active_query_list, list) {
+	LIST_FOR_EACH_ENTRY(query, &ctx->active_timer_queries, list) {
 		r600_emit_query_begin(ctx, query);
 	}
 }