nouveau: potentially delay buffer destruction

Rather than spinning waiting for the GPU to finish with the buffer,
add a callback on the buffer's fence object instead.  Gives a sizable
performance increase in OA.

diff --git a/src/gallium/winsys/dri/nouveau/nouveau_bo.c b/src/gallium/winsys/dri/nouveau/nouveau_bo.c
index 40d4667c4778fb2d296d10006bb7dbe860c12b2c..5dbb7d037444c7d9967acbeee69e665c508b500b 100644 (file)
--- a/src/gallium/winsys/dri/nouveau/nouveau_bo.c
+++ b/src/gallium/winsys/dri/nouveau/nouveau_bo.c
@@ -206,6 +206,18 @@ nouveau_bo_ref(struct nouveau_device *dev, uint64_t handle,
        return 0;
 }
 
+static void
+nouveau_bo_del_cb(void *priv)
+{
+       struct nouveau_bo_priv *nvbo = priv;
+
+       nouveau_fence_ref(NULL, &nvbo->fence);
+       nouveau_mem_free(nvbo->base.device, &nvbo->drm, &nvbo->map);
+       if (nvbo->sysmem && !nvbo->user)
+               free(nvbo->sysmem);
+       free(nvbo);
+}
+
 void
 nouveau_bo_del(struct nouveau_bo **bo)
 {
@@ -220,11 +232,9 @@ nouveau_bo_del(struct nouveau_bo **bo)
                return;
 
        if (nvbo->fence)
-               nouveau_fence_wait(&nvbo->fence);
-       nouveau_mem_free(nvbo->base.device, &nvbo->drm, &nvbo->map);
-       if (nvbo->sysmem && !nvbo->user)
-               free(nvbo->sysmem);
-       free(nvbo);
+               nouveau_fence_signal_cb(nvbo->fence, nouveau_bo_del_cb, nvbo);
+       else
+               nouveau_bo_del_cb(nvbo);
 }
 
 int