[Mesa-dev] [PATCH 7/7] radeonsi: implement ARB/KHR_parallel_shader_compile callbacks

[Marek Olšák]

From: Marek Olšák <marek.olsak at amd.com>

---
 src/gallium/drivers/radeonsi/si_pipe.c | 31 ++++++++++++++++++++++++++
 1 file changed, 31 insertions(+)

diff --git a/src/gallium/drivers/radeonsi/si_pipe.c b/src/gallium/drivers/radeonsi/si_pipe.c
index b965d9d64d4..7dbd4cb2c40 100644
--- a/src/gallium/drivers/radeonsi/si_pipe.c
+++ b/src/gallium/drivers/radeonsi/si_pipe.c
@@ -19,20 +19,21 @@
  * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
  * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
  * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
  * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
  * USE OR OTHER DEALINGS IN THE SOFTWARE.
  */
 
 #include "si_pipe.h"
 #include "si_public.h"
 #include "si_shader_internal.h"
+#include "si_compute.h"
 #include "sid.h"
 
 #include "ac_llvm_util.h"
 #include "radeon/radeon_uvd.h"
 #include "gallivm/lp_bld_misc.h"
 #include "util/disk_cache.h"
 #include "util/u_log.h"
 #include "util/u_memory.h"
 #include "util/u_suballoc.h"
 #include "util/u_tests.h"
@@ -826,20 +827,46 @@ static void si_disk_cache_create(struct si_screen *sscreen)
 	 */
 	STATIC_ASSERT(ALL_FLAGS <= UINT_MAX);
 	shader_debug_flags |= (uint64_t)sscreen->info.address32_hi << 32;
 
 	sscreen->disk_shader_cache =
 		disk_cache_create(sscreen->info.name,
 				  cache_id,
 				  shader_debug_flags);
 }
 
+static void si_set_max_shader_compiler_threads(struct pipe_screen *screen,
+					       unsigned max_threads)
+{
+	struct si_screen *sscreen = (struct si_screen *)screen;
+
+	/* This function doesn't allow a greater number of threads than
+	 * the queue had at its creation. */
+	util_queue_adjust_num_threads(&sscreen->shader_compiler_queue,
+				      max_threads);
+	/* Don't change the number of threads on the low priority queue. */
+}
+
+static bool si_is_parallel_shader_compilation_finished(struct pipe_screen *screen,
+						       void *shader,
+						       unsigned shader_type)
+{
+	if (shader_type == PIPE_SHADER_COMPUTE) {
+		struct si_compute *cs = (struct si_compute*)shader;
+
+		return util_queue_fence_is_signalled(&cs->ready);
+	}
+	struct si_shader_selector *sel = (struct si_shader_selector *)shader;
+
+	return util_queue_fence_is_signalled(&sel->ready);
+}
+
 struct pipe_screen *radeonsi_screen_create(struct radeon_winsys *ws,
 					   const struct pipe_screen_config *config)
 {
 	struct si_screen *sscreen = CALLOC_STRUCT(si_screen);
 	unsigned hw_threads, num_comp_hi_threads, num_comp_lo_threads, i;
 
 	if (!sscreen) {
 		return NULL;
 	}
 
@@ -856,20 +883,24 @@ struct pipe_screen *radeonsi_screen_create(struct radeon_winsys *ws,
 	}
 
 	sscreen->debug_flags = debug_get_flags_option("R600_DEBUG",
 						      debug_options, 0);
 	sscreen->debug_flags |= debug_get_flags_option("AMD_DEBUG",
 						       debug_options, 0);
 
 	/* Set functions first. */
 	sscreen->b.context_create = si_pipe_create_context;
 	sscreen->b.destroy = si_destroy_screen;
+	sscreen->b.set_max_shader_compiler_threads =
+		si_set_max_shader_compiler_threads;
+	sscreen->b.is_parallel_shader_compilation_finished =
+		si_is_parallel_shader_compilation_finished;
 
 	si_init_screen_get_functions(sscreen);
 	si_init_screen_buffer_functions(sscreen);
 	si_init_screen_fence_functions(sscreen);
 	si_init_screen_state_functions(sscreen);
 	si_init_screen_texture_functions(sscreen);
 	si_init_screen_query_functions(sscreen);
 
 	/* Set these flags in debug_flags early, so that the shader cache takes
 	 * them into account.
-- 
2.17.1