[Mesa-dev] [PATCH 3/6] u_queue: add util_queue_finish for waiting for previously added jobs

[Nicolai Hähnle]

From: Nicolai Hähnle <nicolai.haehnle at amd.com>

Schedule one job for every thread, and wait on a barrier inside the job
execution function.
---
 src/util/u_queue.c | 35 +++++++++++++++++++++++++++++++++++
 src/util/u_queue.h |  2 ++
 2 files changed, 37 insertions(+)

diff --git a/src/util/u_queue.c b/src/util/u_queue.c
index 8293ec661b0..7008391d210 100644
--- a/src/util/u_queue.c
+++ b/src/util/u_queue.c
@@ -18,21 +18,25 @@
  * 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.
  *
  * The above copyright notice and this permission notice (including the
  * next paragraph) shall be included in all copies or substantial portions
  * of the Software.
  */
 
 #include "u_queue.h"
+
+#include <alloca.h>
+
 #include "util/u_string.h"
+#include "util/u_thread.h"
 
 static void util_queue_killall_and_wait(struct util_queue *queue);
 
 /****************************************************************************
  * Wait for all queues to assert idle when exit() is called.
  *
  * Otherwise, C++ static variable destructors can be called while threads
  * are using the static variables.
  */
 
@@ -422,19 +426,50 @@ util_queue_drop_job(struct util_queue *queue, struct util_queue_fence *fence)
       }
    }
    mtx_unlock(&queue->lock);
 
    if (removed)
       util_queue_fence_signal(fence);
    else
       util_queue_fence_wait(fence);
 }
 
+static void
+util_queue_finish_execute(void *data, int num_thread)
+{
+   util_barrier *barrier = data;
+   util_barrier_wait(barrier);
+}
+
+/**
+ * Wait until all previously added jobs have completed.
+ */
+void
+util_queue_finish(struct util_queue *queue)
+{
+   util_barrier barrier;
+   struct util_queue_fence *fences = alloca(queue->num_threads * sizeof(*fences));
+
+   util_barrier_init(&barrier, queue->num_threads);
+
+   for (unsigned i = 0; i < queue->num_threads; ++i) {
+      util_queue_fence_init(&fences[i]);
+      util_queue_add_job(queue, &barrier, &fences[i], util_queue_finish_execute, NULL);
+   }
+
+   for (unsigned i = 0; i < queue->num_threads; ++i) {
+      util_queue_fence_wait(&fences[i]);
+      util_queue_fence_destroy(&fences[i]);
+   }
+
+   util_barrier_destroy(&barrier);
+}
+
 int64_t
 util_queue_get_thread_time_nano(struct util_queue *queue, unsigned thread_index)
 {
    /* Allow some flexibility by not raising an error. */
    if (thread_index >= queue->num_threads)
       return 0;
 
    return u_thread_get_time_nano(queue->threads[thread_index]);
 }
diff --git a/src/util/u_queue.h b/src/util/u_queue.h
index 3d9f19f4e6c..f92797ba5aa 100644
--- a/src/util/u_queue.h
+++ b/src/util/u_queue.h
@@ -203,20 +203,22 @@ void util_queue_destroy(struct util_queue *queue);
 
 /* optional cleanup callback is called after fence is signaled: */
 void util_queue_add_job(struct util_queue *queue,
                         void *job,
                         struct util_queue_fence *fence,
                         util_queue_execute_func execute,
                         util_queue_execute_func cleanup);
 void util_queue_drop_job(struct util_queue *queue,
                          struct util_queue_fence *fence);
 
+void util_queue_finish(struct util_queue *queue);
+
 int64_t util_queue_get_thread_time_nano(struct util_queue *queue,
                                         unsigned thread_index);
 
 /* util_queue needs to be cleared to zeroes for this to work */
 static inline bool
 util_queue_is_initialized(struct util_queue *queue)
 {
    return queue->threads != NULL;
 }
 
-- 
2.11.0