sna: Reuse the same upload buffer for the duration of the batch

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>

4 files changed, 67 insertions, 8 deletions

diff --git a/src/sna/kgem.c b/src/sna/kgem.c
index db579d09..cee45139 100644
--- a/src/sna/kgem.c
+++ b/src/sna/kgem.c
@@ -1381,7 +1381,9 @@ static void kgem_commit(struct kgem *kgem)
 {
 	struct kgem_request *rq = kgem->next_request;
 	struct kgem_bo *bo, *next;
+	struct list release;
 
+	list_init(&release);
 	list_for_each_entry_safe(bo, next, &rq->buffers, request) {
 		assert(next->request.prev == &bo->request);
 
@@ -1408,8 +1410,21 @@ static void kgem_commit(struct kgem *kgem)
 			/* proxies are not used for domain tracking */
 			list_del(&bo->request);
 			bo->rq = NULL;
+			bo->exec = &_kgem_dummy_exec;
+			if (bo->map)
+				list_add_tail(&bo->request, &release);
 		}
 	}
+	while (!list_is_empty(&release)) {
+		bo = list_first_entry(&release, struct kgem_bo, request);
+		DBG(("%s: releasing upload cache, handle=%d\n",
+		     __FUNCTION__, bo->handle));
+		list_del(&bo->request);
+		assert(*(struct kgem_bo **)bo->map == bo);
+		*(struct kgem_bo **)bo->map = NULL;
+		bo->map = NULL;
+		kgem_bo_destroy(kgem, bo);
+	}
 
 	if (rq == &_kgem_static_request) {
 		struct drm_i915_gem_set_domain set_domain;
@@ -2842,7 +2857,7 @@ static void _kgem_bo_delete_partial(struct kgem *kgem, struct kgem_bo *bo)
 	DBG(("%s: size=%d, offset=%d, parent used=%d\n",
 	     __FUNCTION__, bo->size.bytes, bo->delta, io->used));
 
-	if (bo->delta + bo->size.bytes == io->used) {
+	if (ALIGN(bo->delta + bo->size.bytes, 64) == io->used) {
 		io->used = bo->delta;
 		bubble_sort_partial(&kgem->active_partials, io);
 	}
@@ -2850,9 +2865,11 @@ static void _kgem_bo_delete_partial(struct kgem *kgem, struct kgem_bo *bo)
 
 void _kgem_bo_destroy(struct kgem *kgem, struct kgem_bo *bo)
 {
+	DBG(("%s: handle=%d, proxy? %d\n",
+	     __FUNCTION__, bo->handle, bo->proxy != NULL));
+
 	if (bo->proxy) {
-		assert(bo->map == NULL);
-		if (bo->io && bo->exec == NULL)
+		if (bo->io && (bo->exec == NULL || bo->proxy->rq == NULL))
 			_kgem_bo_delete_partial(kgem, bo);
 		kgem_bo_unref(kgem, bo->proxy);
 		kgem_bo_binding_free(kgem, bo);
@@ -3873,6 +3890,15 @@ struct kgem_bo *kgem_upload_source_image(struct kgem *kgem,
 	return bo;
 }
 
+void kgem_proxy_bo_attach(struct kgem_bo *bo,
+			  struct kgem_bo **ptr)
+{
+	DBG(("%s: handle=%d\n", __FUNCTION__, bo->handle));
+	assert(bo->map == NULL);
+	bo->map = ptr;
+	*ptr = kgem_bo_reference(bo);
+}
+
 void kgem_buffer_read_sync(struct kgem *kgem, struct kgem_bo *_bo)
 {
 	struct kgem_partial_bo *bo;
@@ -3880,7 +3906,8 @@ void kgem_buffer_read_sync(struct kgem *kgem, struct kgem_bo *_bo)
 	int domain;
 
 	assert(_bo->io);
-	assert(_bo->exec == NULL);
+	assert(_bo->exec == &_kgem_dummy_exec);
+	assert(_bo->rq == NULL);
 	if (_bo->proxy)
 		_bo = _bo->proxy;
 	assert(_bo->exec == NULL);
diff --git a/src/sna/kgem.h b/src/sna/kgem.h
index dff8bb28..52e5e9c0 100644
--- a/src/sna/kgem.h
+++ b/src/sna/kgem.h
@@ -200,6 +200,7 @@ struct kgem_bo *kgem_upload_source_image(struct kgem *kgem,
 					 const void *data,
 					 BoxPtr box,
 					 int stride, int bpp);
+void kgem_proxy_bo_attach(struct kgem_bo *bo, struct kgem_bo **ptr);
 
 int kgem_choose_tiling(struct kgem *kgem,
 		       int tiling, int width, int height, int bpp);
diff --git a/src/sna/sna_accel.c b/src/sna/sna_accel.c
index 1b0ac3c4..dc330ca8 100644
--- a/src/sna/sna_accel.c
+++ b/src/sna/sna_accel.c
@@ -1064,8 +1064,10 @@ skip_inplace_map:
 	}
 
 done:
-	if (flags & MOVE_WRITE)
+	if (flags & MOVE_WRITE) {
 		priv->source_count = SOURCE_BIAS;
+		assert(priv->gpu_bo == NULL || priv->gpu_bo->proxy == NULL);
+	}
 
 	if ((flags & MOVE_ASYNC_HINT) == 0 && priv->cpu_bo) {
 		DBG(("%s: syncing CPU bo\n", __FUNCTION__));
@@ -1260,6 +1262,7 @@ sna_drawable_move_region_to_cpu(DrawablePtr drawable,
 
 		if (priv->stride && priv->gpu_bo &&
 		    region_inplace(sna, pixmap, region, priv)) {
+			assert(priv->gpu_bo->proxy == NULL);
 			if (sync_will_stall(priv->gpu_bo) &&
 			    priv->gpu_bo->exec == NULL)
 				kgem_retire(&sna->kgem);
@@ -1371,6 +1374,7 @@ sna_drawable_move_region_to_cpu(DrawablePtr drawable,
 	if (priv->gpu_bo == NULL)
 		goto done;
 
+	assert(priv->gpu_bo->proxy == NULL);
 	if (priv->clear) {
 		int n = REGION_NUM_RECTS(region);
 		BoxPtr box = REGION_RECTS(region);
@@ -1581,8 +1585,10 @@ done:
 		RegionTranslate(region, -dx, -dy);
 
 out:
-	if (flags & MOVE_WRITE)
+	if (flags & MOVE_WRITE) {
 		priv->source_count = SOURCE_BIAS;
+		assert(priv->gpu_bo == NULL || priv->gpu_bo->proxy == NULL);
+	}
 	if (priv->cpu_bo) {
 		DBG(("%s: syncing cpu bo\n", __FUNCTION__));
 		kgem_bo_sync__cpu(&sna->kgem, priv->cpu_bo);
@@ -1942,6 +1948,7 @@ done:
 
 	DBG(("%s: using GPU bo with damage? %d\n",
 	     __FUNCTION__, *damage != NULL));
+	assert(priv->gpu_bo->proxy == NULL);
 	return priv->gpu_bo;
 
 use_gpu_bo:
@@ -1951,6 +1958,7 @@ use_gpu_bo:
 			  &to_sna_from_pixmap(pixmap)->active_pixmaps);
 	*damage = NULL;
 	DBG(("%s: using whole GPU bo\n", __FUNCTION__));
+	assert(priv->gpu_bo->proxy == NULL);
 	return priv->gpu_bo;
 
 use_cpu_bo:
@@ -2083,6 +2091,7 @@ sna_pixmap_force_to_gpu(PixmapPtr pixmap, unsigned flags)
 
 	if (DAMAGE_IS_ALL(priv->gpu_damage)) {
 		DBG(("%s: GPU all-damaged\n", __FUNCTION__));
+		assert(!priv->gpu_bo->proxy || (flags & MOVE_WRITE) == 0);
 		return sna_pixmap_mark_active(to_sna_from_pixmap(pixmap), priv);
 	}
 
@@ -2285,6 +2294,7 @@ done:
 		}
 	}
 active:
+	assert(!priv->gpu_bo->proxy || (flags & MOVE_WRITE) == 0);
 	return sna_pixmap_mark_active(sna, priv);
 }
 
@@ -2532,6 +2542,8 @@ static bool upload_inplace(struct sna *sna,
 		return false;
 
 	if (priv->gpu_bo) {
+		assert(priv->gpu_bo->proxy == NULL);
+
 		if (!kgem_bo_map_will_stall(&sna->kgem, priv->gpu_bo))
 			return true;
 
@@ -3389,6 +3401,8 @@ sna_copy_boxes(DrawablePtr src, DrawablePtr dst, GCPtr gc,
 		dst_priv->clear = false;
 	}
 
+	assert(dst_priv->gpu_bo == NULL || dst_priv->gpu_bo->proxy == NULL);
+
 	/* Try to maintain the data on the GPU */
 	if (dst_priv->gpu_bo == NULL &&
 	    ((dst_priv->cpu_damage == NULL && copy_use_gpu_bo(sna, dst_priv, &region)) ||
diff --git a/src/sna/sna_render.c b/src/sna/sna_render.c
index 09a33c16..b5d314b8 100644
--- a/src/sna/sna_render.c
+++ b/src/sna/sna_render.c
@@ -496,6 +496,14 @@ static struct kgem_bo *upload(struct sna *sna,
 		channel->offset[1] -= box->y1;
 		channel->scale[0] = 1.f/channel->width;
 		channel->scale[1] = 1.f/channel->height;
+
+		if (pixmap->usage_hint == 0 &&
+		    channel->width  == pixmap->drawable.width &&
+		    channel->height == pixmap->drawable.height) {
+			struct sna_pixmap *priv = sna_pixmap(pixmap);
+			if (priv)
+				kgem_proxy_bo_attach(bo, &priv->gpu_bo);
+		}
 	}
 
 	return bo;
@@ -526,7 +534,8 @@ sna_render_pixmap_bo(struct sna *sna,
 	priv = sna_pixmap(pixmap);
 	if (priv) {
 		if (priv->gpu_bo &&
-		    (DAMAGE_IS_ALL(priv->gpu_damage) || !priv->cpu_damage)) {
+		    (DAMAGE_IS_ALL(priv->gpu_damage) || !priv->cpu_damage ||
+		     priv->gpu_bo->proxy)) {
 			channel->bo = kgem_bo_reference(priv->gpu_bo);
 			return 1;
 		}
@@ -1148,12 +1157,20 @@ sna_render_picture_extract(struct sna *sna,
 				upload = false;
 			}
 		}
-		if (upload)
+		if (upload) {
 			bo = kgem_upload_source_image(&sna->kgem,
 						      pixmap->devPrivate.ptr,
 						      &box,
 						      pixmap->devKind,
 						      pixmap->drawable.bitsPerPixel);
+			if (pixmap->usage_hint == 0 &&
+			    box.x2 - box.x1 == pixmap->drawable.width &&
+			    box.y2 - box.y1 == pixmap->drawable.height) {
+				struct sna_pixmap *priv = sna_pixmap(pixmap);
+				if (priv)
+					kgem_proxy_bo_attach(bo, &priv->gpu_bo);
+			}
+		}
 	}
 	if (src_bo) {
 		bo = kgem_create_2d(&sna->kgem, w, h,