4 files changed, 51 insertions, 32 deletions

diff --git a/src/gallium/drivers/radeonsi/si_compute.c b/src/gallium/drivers/radeonsi/si_compute.c
index 8c016ba65ab..d0e481a3f15 100644
--- a/src/gallium/drivers/radeonsi/si_compute.c
+++ b/src/gallium/drivers/radeonsi/si_compute.c
@@ -801,6 +801,9 @@ static void si_launch_grid(
 
 	si_need_cs_space(sctx);
 
+	if (sctx->b.log)
+		si_log_compute_state(sctx, sctx->b.log);
+
 	if (!sctx->cs_shader_state.initialized)
 		si_initialize_compute(sctx);
 
@@ -847,6 +850,9 @@ static void si_launch_grid(
 
 	si_ce_post_draw_synchronization(sctx);
 
+	if (unlikely(sctx->current_saved_cs))
+		si_trace_emit(sctx);
+
 	sctx->compute_is_busy = true;
 	sctx->b.num_compute_calls++;
 	if (sctx->cs_shader_state.uses_scratch)
diff --git a/src/gallium/drivers/radeonsi/si_debug.c b/src/gallium/drivers/radeonsi/si_debug.c
index bf63f9f5216..278887c7063 100644
--- a/src/gallium/drivers/radeonsi/si_debug.c
+++ b/src/gallium/drivers/radeonsi/si_debug.c
@@ -1067,34 +1067,38 @@ static void si_dump_debug_state(struct pipe_context *ctx, FILE *f,
 			si_dump_command("Wave information", "umr -O bits -wa", f);
 		}
 	}
+}
 
-	struct u_log_context log;
-	u_log_context_init(&log);
-
-	if (flags & PIPE_DUMP_CURRENT_STATES)
-		si_dump_framebuffer(sctx, &log);
-
-	if (flags & PIPE_DUMP_CURRENT_SHADERS) {
-		si_dump_gfx_shader(sctx, &sctx->vs_shader, &log);
-		si_dump_gfx_shader(sctx, &sctx->tcs_shader, &log);
-		si_dump_gfx_shader(sctx, &sctx->tes_shader, &log);
-		si_dump_gfx_shader(sctx, &sctx->gs_shader, &log);
-		si_dump_gfx_shader(sctx, &sctx->ps_shader, &log);
-		si_dump_compute_shader(&sctx->cs_shader_state, &log);
-
-		si_dump_descriptor_list(&sctx->descriptors[SI_DESCS_RW_BUFFERS],
-					"", "RW buffers", 4, SI_NUM_RW_BUFFERS,
-					si_identity, &log);
-		si_dump_gfx_descriptors(sctx, &sctx->vs_shader, &log);
-		si_dump_gfx_descriptors(sctx, &sctx->tcs_shader, &log);
-		si_dump_gfx_descriptors(sctx, &sctx->tes_shader, &log);
-		si_dump_gfx_descriptors(sctx, &sctx->gs_shader, &log);
-		si_dump_gfx_descriptors(sctx, &sctx->ps_shader, &log);
-		si_dump_compute_descriptors(sctx, &log);
-	}
+void si_log_draw_state(struct si_context *sctx, struct u_log_context *log)
+{
+	if (!log)
+		return;
 
-	u_log_new_page_print(&log, f);
-	u_log_context_destroy(&log);
+	si_dump_framebuffer(sctx, log);
+
+	si_dump_gfx_shader(sctx, &sctx->vs_shader, log);
+	si_dump_gfx_shader(sctx, &sctx->tcs_shader, log);
+	si_dump_gfx_shader(sctx, &sctx->tes_shader, log);
+	si_dump_gfx_shader(sctx, &sctx->gs_shader, log);
+	si_dump_gfx_shader(sctx, &sctx->ps_shader, log);
+
+	si_dump_descriptor_list(&sctx->descriptors[SI_DESCS_RW_BUFFERS],
+				"", "RW buffers", 4, SI_NUM_RW_BUFFERS,
+				si_identity, log);
+	si_dump_gfx_descriptors(sctx, &sctx->vs_shader, log);
+	si_dump_gfx_descriptors(sctx, &sctx->tcs_shader, log);
+	si_dump_gfx_descriptors(sctx, &sctx->tes_shader, log);
+	si_dump_gfx_descriptors(sctx, &sctx->gs_shader, log);
+	si_dump_gfx_descriptors(sctx, &sctx->ps_shader, log);
+}
+
+void si_log_compute_state(struct si_context *sctx, struct u_log_context *log)
+{
+	if (!log)
+		return;
+
+	si_dump_compute_shader(&sctx->cs_shader_state, log);
+	si_dump_compute_descriptors(sctx, log);
 }
 
 static void si_dump_dma(struct si_context *sctx,
@@ -1246,13 +1250,17 @@ void si_check_vm_faults(struct r600_common_context *ctx,
 			sctx->apitrace_call_number);
 
 	switch (ring) {
-	case RING_GFX:
-		si_dump_debug_state(&sctx->b.b, f,
-				    PIPE_DUMP_CURRENT_STATES |
-				    PIPE_DUMP_CURRENT_SHADERS |
-				    PIPE_DUMP_LAST_COMMAND_BUFFER);
-		break;
+	case RING_GFX: {
+		struct u_log_context log;
+		u_log_context_init(&log);
 
+		si_log_draw_state(sctx, &log);
+		si_log_compute_state(sctx, &log);
+
+		u_log_new_page_print(&log, f);
+		u_log_context_destroy(&log);
+		break;
+	}
 	case RING_DMA:
 		si_dump_dma(sctx, saved, f);
 		break;
diff --git a/src/gallium/drivers/radeonsi/si_pipe.h b/src/gallium/drivers/radeonsi/si_pipe.h
index 9ba1dfbab49..2dca72d7f3c 100644
--- a/src/gallium/drivers/radeonsi/si_pipe.h
+++ b/src/gallium/drivers/radeonsi/si_pipe.h
@@ -508,6 +508,8 @@ void si_init_cp_dma_functions(struct si_context *sctx);
 /* si_debug.c */
 void si_auto_log_cs(void *data, struct u_log_context *log);
 void si_log_hw_flush(struct si_context *sctx);
+void si_log_draw_state(struct si_context *sctx, struct u_log_context *log);
+void si_log_compute_state(struct si_context *sctx, struct u_log_context *log);
 void si_init_debug_functions(struct si_context *sctx);
 void si_check_vm_faults(struct r600_common_context *ctx,
 			struct radeon_saved_cs *saved, enum ring_type ring);
diff --git a/src/gallium/drivers/radeonsi/si_state_draw.c b/src/gallium/drivers/radeonsi/si_state_draw.c
index 4505bc6ed62..9631f18ff7f 100644
--- a/src/gallium/drivers/radeonsi/si_state_draw.c
+++ b/src/gallium/drivers/radeonsi/si_state_draw.c
@@ -1361,6 +1361,9 @@ void si_draw_vbo(struct pipe_context *ctx, const struct pipe_draw_info *info)
 
 	si_need_cs_space(sctx);
 
+	if (unlikely(sctx->b.log))
+		si_log_draw_state(sctx, sctx->b.log);
+
 	/* Since we've called r600_context_add_resource_size for vertex buffers,
 	 * this must be called after si_need_cs_space, because we must let
 	 * need_cs_space flush before we add buffers to the buffer list.