sna: Pass a hint to the backends when using copy_boxes for readback

If we expect to only emit this set of copy_boxes() and then submit the
batch, we would prefer to use the BLT for its lower latency.

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

14 files changed, 69 insertions, 51 deletions

diff --git a/src/sna/gen2_render.c b/src/sna/gen2_render.c
index 5fd995b3..b41c3865 100644
--- a/src/sna/gen2_render.c
+++ b/src/sna/gen2_render.c
@@ -2409,7 +2409,7 @@ static inline Bool prefer_blt_fill(struct sna *sna)
 #endif
 }
 
-static inline Bool prefer_blt_copy(struct sna *sna)
+static inline Bool prefer_blt_copy(struct sna *sna, unsigned flags)
 {
 #if PREFER_BLT_COPY
 	return true;
@@ -2864,7 +2864,7 @@ static Bool
 gen2_render_copy_boxes(struct sna *sna, uint8_t alu,
 		       PixmapPtr src, struct kgem_bo *src_bo, int16_t src_dx, int16_t src_dy,
 		       PixmapPtr dst, struct kgem_bo *dst_bo, int16_t dst_dx, int16_t dst_dy,
-		       const BoxRec *box, int n)
+		       const BoxRec *box, int n, unsigned flags)
 {
 	struct sna_composite_op tmp;
 
@@ -2882,7 +2882,7 @@ gen2_render_copy_boxes(struct sna *sna, uint8_t alu,
 	DBG(("%s (%d, %d)->(%d, %d) x %d\n",
 	     __FUNCTION__, src_dx, src_dy, dst_dx, dst_dy, n));
 
-	if (prefer_blt_copy(sna) &&
+	if (prefer_blt_copy(sna, flags) &&
 	    sna_blt_compare_depth(&src->drawable, &dst->drawable) &&
 	    sna_blt_copy_boxes(sna, alu,
 			       src_bo, src_dx, src_dy,
@@ -3048,7 +3048,7 @@ gen2_render_copy(struct sna *sna, uint8_t alu,
 #endif
 
 	/* Prefer to use the BLT */
-	if (prefer_blt_copy(sna) &&
+	if (prefer_blt_copy(sna, 0) &&
 	    sna_blt_compare_depth(&src->drawable, &dst->drawable) &&
 	    sna_blt_copy(sna, alu,
 			 src_bo, dst_bo,
diff --git a/src/sna/gen3_render.c b/src/sna/gen3_render.c
index 63bbd766..cfc8c63e 100644
--- a/src/sna/gen3_render.c
+++ b/src/sna/gen3_render.c
@@ -4040,7 +4040,7 @@ static Bool
 gen3_render_copy_boxes(struct sna *sna, uint8_t alu,
 		       PixmapPtr src, struct kgem_bo *src_bo, int16_t src_dx, int16_t src_dy,
 		       PixmapPtr dst, struct kgem_bo *dst_bo, int16_t dst_dx, int16_t dst_dy,
-		       const BoxRec *box, int n)
+		       const BoxRec *box, int n, unsigned flags)
 {
 	struct sna_composite_op tmp;
 
diff --git a/src/sna/gen4_render.c b/src/sna/gen4_render.c
index c3a82046..c5775360 100644
--- a/src/sna/gen4_render.c
+++ b/src/sna/gen4_render.c
@@ -2576,11 +2576,22 @@ gen4_render_copy_one(struct sna *sna,
 	FLUSH(op);
 }
 
+static inline bool prefer_blt_copy(struct sna *sna, unsigned flags)
+{
+#if PREFER_BLT
+	return true;
+	(void)sna;
+#else
+	return sna->kgem.mode != KGEM_RENDER;
+#endif
+	(void)flags;
+}
+
 static Bool
 gen4_render_copy_boxes(struct sna *sna, uint8_t alu,
 		       PixmapPtr src, struct kgem_bo *src_bo, int16_t src_dx, int16_t src_dy,
 		       PixmapPtr dst, struct kgem_bo *dst_bo, int16_t dst_dx, int16_t dst_dy,
-		       const BoxRec *box, int n)
+		       const BoxRec *box, int n, unsigned flags)
 {
 	struct sna_composite_op tmp;
 
@@ -2597,7 +2608,7 @@ gen4_render_copy_boxes(struct sna *sna, uint8_t alu,
 				  box, n);
 #endif
 
-	if (prefer_blt(sna) &&
+	if (prefer_blt_copy(sna, flags) &&
 	    sna_blt_compare_depth(&src->drawable, &dst->drawable) &&
 	    sna_blt_copy_boxes(sna, alu,
 			       src_bo, src_dx, src_dy,
diff --git a/src/sna/gen5_render.c b/src/sna/gen5_render.c
index 46a37a0b..49cc17ed 100644
--- a/src/sna/gen5_render.c
+++ b/src/sna/gen5_render.c
@@ -2909,7 +2909,7 @@ static Bool
 gen5_render_copy_boxes(struct sna *sna, uint8_t alu,
 		       PixmapPtr src, struct kgem_bo *src_bo, int16_t src_dx, int16_t src_dy,
 		       PixmapPtr dst, struct kgem_bo *dst_bo, int16_t dst_dx, int16_t dst_dy,
-		       const BoxRec *box, int n)
+		       const BoxRec *box, int n, unsigned flags)
 {
 	struct sna_composite_op tmp;
 
diff --git a/src/sna/gen6_render.c b/src/sna/gen6_render.c
index 81fad355..778a0722 100644
--- a/src/sna/gen6_render.c
+++ b/src/sna/gen6_render.c
@@ -3238,9 +3238,11 @@ static bool prefer_blt_bo(struct sna *sna,
 
 static inline bool prefer_blt_copy(struct sna *sna,
 				   PixmapPtr src, struct kgem_bo *src_bo,
-				   PixmapPtr dst, struct kgem_bo *dst_bo)
+				   PixmapPtr dst, struct kgem_bo *dst_bo,
+				   unsigned flags)
 {
 	return (sna->kgem.ring == KGEM_BLT ||
+		(flags & COPY_LAST && sna->kgem.mode == KGEM_NONE) ||
 		prefer_blt_bo(sna, src, src_bo) ||
 		prefer_blt_bo(sna, dst, dst_bo));
 }
@@ -3280,7 +3282,7 @@ static Bool
 gen6_render_copy_boxes(struct sna *sna, uint8_t alu,
 		       PixmapPtr src, struct kgem_bo *src_bo, int16_t src_dx, int16_t src_dy,
 		       PixmapPtr dst, struct kgem_bo *dst_bo, int16_t dst_dx, int16_t dst_dy,
-		       const BoxRec *box, int n)
+		       const BoxRec *box, int n, unsigned flags)
 {
 	struct sna_composite_op tmp;
 
@@ -3302,7 +3304,7 @@ gen6_render_copy_boxes(struct sna *sna, uint8_t alu,
 		      dst_bo, dst_dx, dst_dy,
 		      box, n)));
 
-	if (prefer_blt_copy(sna, src, src_bo, dst, dst_bo) &&
+	if (prefer_blt_copy(sna, src, src_bo, dst, dst_bo, flags) &&
 	    sna_blt_compare_depth(&src->drawable, &dst->drawable) &&
 	    sna_blt_copy_boxes(sna, alu,
 			       src_bo, src_dx, src_dy,
@@ -3537,7 +3539,7 @@ gen6_render_copy(struct sna *sna, uint8_t alu,
 	     src->drawable.width, src->drawable.height,
 	     dst->drawable.width, dst->drawable.height));
 
-	if (prefer_blt_copy(sna, src, src_bo, dst, dst_bo) &&
+	if (prefer_blt_copy(sna, src, src_bo, dst, dst_bo, 0) &&
 	    sna_blt_compare_depth(&src->drawable, &dst->drawable) &&
 	    sna_blt_copy(sna, alu,
 			 src_bo, dst_bo,
diff --git a/src/sna/gen7_render.c b/src/sna/gen7_render.c
index d3032869..f9b2e9e5 100644
--- a/src/sna/gen7_render.c
+++ b/src/sna/gen7_render.c
@@ -3322,9 +3322,11 @@ static bool prefer_blt_bo(struct sna *sna,
 
 static inline bool prefer_blt_copy(struct sna *sna,
 				   PixmapPtr src, struct kgem_bo *src_bo,
-				   PixmapPtr dst, struct kgem_bo *dst_bo)
+				   PixmapPtr dst, struct kgem_bo *dst_bo,
+				   unsigned flags)
 {
 	return (sna->kgem.ring == KGEM_BLT ||
+		(flags & COPY_LAST && sna->kgem.mode == KGEM_NONE) ||
 		prefer_blt_bo(sna, src, src_bo) ||
 		prefer_blt_bo(sna, dst, dst_bo));
 }
@@ -3364,7 +3366,7 @@ static Bool
 gen7_render_copy_boxes(struct sna *sna, uint8_t alu,
 		       PixmapPtr src, struct kgem_bo *src_bo, int16_t src_dx, int16_t src_dy,
 		       PixmapPtr dst, struct kgem_bo *dst_bo, int16_t dst_dx, int16_t dst_dy,
-		       const BoxRec *box, int n)
+		       const BoxRec *box, int n, unsigned flags)
 {
 	struct sna_composite_op tmp;
 
@@ -3386,7 +3388,7 @@ gen7_render_copy_boxes(struct sna *sna, uint8_t alu,
 		      dst_bo, dst_dx, dst_dy,
 		      box, n)));
 
-	if (prefer_blt_copy(sna, src, src_bo, dst, dst_bo) &&
+	if (prefer_blt_copy(sna, src, src_bo, dst, dst_bo, flags) &&
 	    sna_blt_compare_depth(&src->drawable, &dst->drawable) &&
 	    sna_blt_copy_boxes(sna, alu,
 			       src_bo, src_dx, src_dy,
@@ -3644,7 +3646,7 @@ gen7_render_copy(struct sna *sna, uint8_t alu,
 	     src->drawable.width, src->drawable.height,
 	     dst->drawable.width, dst->drawable.height));
 
-	if (prefer_blt_copy(sna, src, src_bo, dst, dst_bo) &&
+	if (prefer_blt_copy(sna, src, src_bo, dst, dst_bo, 0) &&
 	    sna_blt_compare_depth(&src->drawable, &dst->drawable) &&
 	    sna_blt_copy(sna, alu,
 			 src_bo, dst_bo,
diff --git a/src/sna/sna_accel.c b/src/sna/sna_accel.c
index 537b2871..1919b2ef 100644
--- a/src/sna/sna_accel.c
+++ b/src/sna/sna_accel.c
@@ -594,7 +594,7 @@ struct kgem_bo *sna_pixmap_change_tiling(PixmapPtr pixmap, uint32_t tiling)
 	if (!sna->render.copy_boxes(sna, GXcopy,
 				    pixmap, priv->gpu_bo, 0, 0,
 				    pixmap, bo, 0, 0,
-				    &box, 1)) {
+				    &box, 1, 0)) {
 		DBG(("%s: copy failed\n", __FUNCTION__));
 		kgem_bo_destroy(&sna->kgem, bo);
 		return NULL;
@@ -1232,7 +1232,7 @@ skip_inplace_map:
 				ok = sna->render.copy_boxes(sna, GXcopy,
 							    pixmap, priv->gpu_bo, 0, 0,
 							    pixmap, priv->cpu_bo, 0, 0,
-							    box, n);
+							    box, n, COPY_LAST);
 			if (!ok)
 				sna_read_boxes(sna,
 					       priv->gpu_bo, 0, 0,
@@ -1690,7 +1690,7 @@ sna_drawable_move_region_to_cpu(DrawablePtr drawable,
 				ok = sna->render.copy_boxes(sna, GXcopy,
 							    pixmap, priv->gpu_bo, 0, 0,
 							    pixmap, priv->cpu_bo, 0, 0,
-							    box, n);
+							    box, n, COPY_LAST);
 			if (!ok)
 				sna_read_boxes(sna,
 					       priv->gpu_bo, 0, 0,
@@ -1798,7 +1798,7 @@ sna_drawable_move_region_to_cpu(DrawablePtr drawable,
 						ok = sna->render.copy_boxes(sna, GXcopy,
 									    pixmap, priv->gpu_bo, 0, 0,
 									    pixmap, priv->cpu_bo, 0, 0,
-									    box, n);
+									    box, n, COPY_LAST);
 
 					if (!ok)
 						sna_read_boxes(sna,
@@ -1823,7 +1823,7 @@ sna_drawable_move_region_to_cpu(DrawablePtr drawable,
 					ok = sna->render.copy_boxes(sna, GXcopy,
 								    pixmap, priv->gpu_bo, 0, 0,
 								    pixmap, priv->cpu_bo, 0, 0,
-								    box, n);
+								    box, n, COPY_LAST);
 				if (!ok)
 					sna_read_boxes(sna,
 						       priv->gpu_bo, 0, 0,
@@ -1848,7 +1848,7 @@ sna_drawable_move_region_to_cpu(DrawablePtr drawable,
 						ok = sna->render.copy_boxes(sna, GXcopy,
 									    pixmap, priv->gpu_bo, 0, 0,
 									    pixmap, priv->cpu_bo, 0, 0,
-									    box, n);
+									    box, n, COPY_LAST);
 					if (!ok)
 						sna_read_boxes(sna,
 							       priv->gpu_bo, 0, 0,
@@ -2096,7 +2096,7 @@ sna_pixmap_move_area_to_gpu(PixmapPtr pixmap, const BoxRec *box, unsigned int fl
 				ok = sna->render.copy_boxes(sna, GXcopy,
 							    pixmap, priv->cpu_bo, 0, 0,
 							    pixmap, priv->gpu_bo, 0, 0,
-							    box, n);
+							    box, n, 0);
 			if (!ok) {
 				if (pixmap->devPrivate.ptr == NULL) {
 					assert(priv->stride && priv->ptr);
@@ -2134,7 +2134,7 @@ sna_pixmap_move_area_to_gpu(PixmapPtr pixmap, const BoxRec *box, unsigned int fl
 			ok = sna->render.copy_boxes(sna, GXcopy,
 						    pixmap, priv->cpu_bo, 0, 0,
 						    pixmap, priv->gpu_bo, 0, 0,
-						    box, 1);
+						    box, 1, 0);
 		if (!ok) {
 			if (pixmap->devPrivate.ptr == NULL) {
 				assert(priv->stride && priv->ptr);
@@ -2163,7 +2163,7 @@ sna_pixmap_move_area_to_gpu(PixmapPtr pixmap, const BoxRec *box, unsigned int fl
 			ok = sna->render.copy_boxes(sna, GXcopy,
 						    pixmap, priv->cpu_bo, 0, 0,
 						    pixmap, priv->gpu_bo, 0, 0,
-						    box, n);
+						    box, n,0);
 		if (!ok) {
 			if (pixmap->devPrivate.ptr == NULL) {
 				assert(priv->stride && priv->ptr);
@@ -2641,7 +2641,7 @@ sna_pixmap_move_to_gpu(PixmapPtr pixmap, unsigned flags)
 			ok = sna->render.copy_boxes(sna, GXcopy,
 						    pixmap, priv->cpu_bo, 0, 0,
 						    pixmap, priv->gpu_bo, 0, 0,
-						    box, n);
+						    box, n, 0);
 		if (!ok) {
 			if (pixmap->devPrivate.ptr == NULL) {
 				assert(priv->stride && priv->ptr);
@@ -3657,7 +3657,7 @@ sna_self_copy_boxes(DrawablePtr src, DrawablePtr dst, GCPtr gc,
 		if (!sna->render.copy_boxes(sna, alu,
 					    pixmap, priv->gpu_bo, dx, dy,
 					    pixmap, priv->gpu_bo, tx, ty,
-					    box, n)) {
+					    box, n, 0)) {
 			DBG(("%s: fallback - accelerated copy boxes failed\n",
 			     __FUNCTION__));
 			goto fallback;
@@ -3874,7 +3874,7 @@ sna_copy_boxes(DrawablePtr src, DrawablePtr dst, GCPtr gc,
 			if (!sna->render.copy_boxes(sna, alu,
 						    src_pixmap, src_priv->gpu_bo, src_dx, src_dy,
 						    dst_pixmap, bo, dst_dx, dst_dy,
-						    box, n)) {
+						    box, n, 0)) {
 				DBG(("%s: fallback - accelerated copy boxes failed\n",
 				     __FUNCTION__));
 				goto fallback;
@@ -3909,7 +3909,7 @@ sna_copy_boxes(DrawablePtr src, DrawablePtr dst, GCPtr gc,
 			if (!sna->render.copy_boxes(sna, alu,
 						    src_pixmap, src_priv->gpu_bo, src_dx, src_dy,
 						    dst_pixmap, bo, dst_dx, dst_dy,
-						    box, n)) {
+						    box, n, 0)) {
 				DBG(("%s: fallback - accelerated copy boxes failed\n",
 				     __FUNCTION__));
 				goto fallback;
@@ -3945,7 +3945,7 @@ sna_copy_boxes(DrawablePtr src, DrawablePtr dst, GCPtr gc,
 			if (!sna->render.copy_boxes(sna, alu,
 						    src_pixmap, src_priv->cpu_bo, src_dx, src_dy,
 						    dst_pixmap, bo, dst_dx, dst_dy,
-						    box, n)) {
+						    box, n, 0)) {
 				DBG(("%s: fallback - accelerated copy boxes failed\n",
 				     __FUNCTION__));
 				goto fallback;
@@ -4005,7 +4005,7 @@ sna_copy_boxes(DrawablePtr src, DrawablePtr dst, GCPtr gc,
 			if (!sna->render.copy_boxes(sna, alu,
 						    tmp, sna_pixmap_get_bo(tmp), dx, dy,
 						    dst_pixmap, bo, dst_dx, dst_dy,
-						    box, n)) {
+						    box, n, 0)) {
 				DBG(("%s: fallback - accelerated copy boxes failed\n",
 				     __FUNCTION__));
 				tmp->drawable.pScreen->DestroyPixmap(tmp);
diff --git a/src/sna/sna_display.c b/src/sna/sna_display.c
index f1a0b841..f59eed52 100644
--- a/src/sna/sna_display.c
+++ b/src/sna/sna_display.c
@@ -814,7 +814,7 @@ void sna_copy_fbcon(struct sna *sna)
 	ok = sna->render.copy_boxes(sna, GXcopy,
 				    scratch, bo, sx, sy,
 				    sna->front, priv->gpu_bo, dx, dy,
-				    &box, 1);
+				    &box, 1, 0);
 	if (!DAMAGE_IS_ALL(priv->gpu_damage))
 		sna_damage_add_box(&priv->gpu_damage, &box);
 
@@ -2780,7 +2780,7 @@ sna_crtc_redisplay(xf86CrtcPtr crtc, RegionPtr region)
 		if (sna->render.copy_boxes(sna, GXcopy,
 					   sna->front, sna_pixmap_get_bo(sna->front), 0, 0,
 					   &tmp, sna_crtc->bo, -tx, -ty,
-					   REGION_RECTS(region), REGION_NUM_RECTS(region)))
+					   REGION_RECTS(region), REGION_NUM_RECTS(region), 0))
 			return;
 	}
 
@@ -2909,7 +2909,8 @@ disable:
 					     sna->front, new, 0, 0,
 					     sna->front, old, 0, 0,
 					     REGION_RECTS(region),
-					     REGION_NUM_RECTS(region));
+					     REGION_NUM_RECTS(region),
+					     COPY_LAST);
 		kgem_submit(&sna->kgem);
 
 		sna_pixmap(sna->front)->gpu_bo = old;
diff --git a/src/sna/sna_dri.c b/src/sna/sna_dri.c
index 4519f170..a10efc7b 100644
--- a/src/sna/sna_dri.c
+++ b/src/sna/sna_dri.c
@@ -568,7 +568,7 @@ sna_dri_copy_to_front(struct sna *sna, DrawablePtr draw, RegionPtr region,
 	sna->render.copy_boxes(sna, GXcopy,
 			       (PixmapPtr)draw, src_bo, -draw->x, -draw->y,
 			       pixmap, dst_bo, dx, dy,
-			       boxes, n);
+			       boxes, n, COPY_LAST);
 
 	DBG(("%s: flushing? %d\n", __FUNCTION__, flush));
 	if (flush) { /* STAT! */
@@ -654,7 +654,7 @@ sna_dri_copy_from_front(struct sna *sna, DrawablePtr draw, RegionPtr region,
 	sna->render.copy_boxes(sna, GXcopy,
 			       pixmap, src_bo, dx, dy,
 			       (PixmapPtr)draw, dst_bo, -draw->x, -draw->y,
-			       boxes, n);
+			       boxes, n, COPY_LAST);
 
 	if (region == &clip)
 		pixman_region_fini(&clip);
@@ -697,7 +697,7 @@ sna_dri_copy(struct sna *sna, DrawablePtr draw, RegionPtr region,
 	sna->render.copy_boxes(sna, GXcopy,
 			       (PixmapPtr)draw, src_bo, 0, 0,
 			       (PixmapPtr)draw, dst_bo, 0, 0,
-			       boxes, n);
+			       boxes, n, COPY_LAST);
 
 	if (region == &clip)
 		pixman_region_fini(&clip);
diff --git a/src/sna/sna_io.c b/src/sna/sna_io.c
index 8d045484..ffa86b10 100644
--- a/src/sna/sna_io.c
+++ b/src/sna/sna_io.c
@@ -276,7 +276,7 @@ fallback:
 					if (!sna->render.copy_boxes(sna, GXcopy,
 								    dst, src_bo, src_dx, src_dy,
 								    &tmp, dst_bo, -tile.x1, -tile.y1,
-								    clipped, c-clipped)) {
+								    clipped, c-clipped, COPY_LAST)) {
 						kgem_bo_destroy(&sna->kgem, dst_bo);
 						if (clipped != stack)
 							free(clipped);
@@ -316,7 +316,7 @@ fallback:
 			if (!sna->render.copy_boxes(sna, GXcopy,
 						    dst, src_bo, src_dx, src_dy,
 						    &tmp, dst_bo, -extents.x1, -extents.y1,
-						    box, nbox)) {
+						    box, nbox, COPY_LAST)) {
 				kgem_bo_destroy(&sna->kgem, dst_bo);
 				goto fallback;
 			}
@@ -689,7 +689,7 @@ tile:
 						n = sna->render.copy_boxes(sna, GXcopy,
 									   &tmp, src_bo, -tile.x1, -tile.y1,
 									   dst, dst_bo, dst_dx, dst_dy,
-									   clipped, c - clipped);
+									   clipped, c - clipped, 0);
 					else
 						n = 1;
 
@@ -736,7 +736,7 @@ tile:
 			n = sna->render.copy_boxes(sna, GXcopy,
 						   &tmp, src_bo, -extents.x1, -extents.y1,
 						   dst, dst_bo, dst_dx, dst_dy,
-						   box, nbox);
+						   box, nbox, 0);
 
 			kgem_bo_destroy(&sna->kgem, src_bo);
 
@@ -973,7 +973,7 @@ fallback:
 		n = sna->render.copy_boxes(sna, GXcopy,
 					   &tmp, src_bo, -extents.x1, -extents.y1,
 					   dst, dst_bo, dst_dx, dst_dy,
-					   box, nbox);
+					   box, nbox, 0);
 
 		kgem_bo_destroy(&sna->kgem, src_bo);
 
@@ -1144,7 +1144,7 @@ indirect_replace(struct sna *sna,
 	ret = sna->render.copy_boxes(sna, GXcopy,
 				     pixmap, src_bo, 0, 0,
 				     pixmap, bo, 0, 0,
-				     &box, 1);
+				     &box, 1, 0);
 
 	kgem_bo_destroy(kgem, src_bo);
 
diff --git a/src/sna/sna_render.c b/src/sna/sna_render.c
index a072994a..c22965c7 100644
--- a/src/sna/sna_render.c
+++ b/src/sna/sna_render.c
@@ -1167,7 +1167,7 @@ sna_render_picture_extract(struct sna *sna,
 			if (!sna->render.copy_boxes(sna, GXcopy,
 						    pixmap, src_bo, 0, 0,
 						    &tmp, bo, -box.x1, -box.y1,
-						    &box, 1)) {
+						    &box, 1, 0)) {
 				kgem_bo_destroy(&sna->kgem, bo);
 				bo = NULL;
 			}
diff --git a/src/sna/sna_render.h b/src/sna/sna_render.h
index 1cc3af19..0eb7e90a 100644
--- a/src/sna/sna_render.h
+++ b/src/sna/sna_render.h
@@ -237,7 +237,9 @@ struct sna_render {
 	Bool (*copy_boxes)(struct sna *sna, uint8_t alu,
 			   PixmapPtr src, struct kgem_bo *src_bo, int16_t src_dx, int16_t src_dy,
 			   PixmapPtr dst, struct kgem_bo *dst_bo, int16_t dst_dx, int16_t dst_dy,
-			   const BoxRec *box, int n);
+			   const BoxRec *box, int n, unsigned flags);
+#define COPY_LAST 0x1
+
 	Bool (*copy)(struct sna *sna, uint8_t alu,
 		     PixmapPtr src, struct kgem_bo *src_bo,
 		     PixmapPtr dst, struct kgem_bo *dst_bo,
diff --git a/src/sna/sna_tiling.c b/src/sna/sna_tiling.c
index 27a9dc81..b20eceb8 100644
--- a/src/sna/sna_tiling.c
+++ b/src/sna/sna_tiling.c
@@ -659,7 +659,7 @@ sna_tiling_fill_boxes(struct sna *sna,
 				if (!sna->render.copy_boxes(sna, GXcopy,
 							     dst, dst_bo, 0, 0,
 							     &tmp, bo, -dx, -dy,
-							     REGION_RECTS(&this), REGION_NUM_RECTS(&this)))
+							     REGION_RECTS(&this), REGION_NUM_RECTS(&this), 0))
 					goto err;
 
 				RegionTranslate(&this, -dx, -dy);
@@ -671,7 +671,7 @@ sna_tiling_fill_boxes(struct sna *sna,
 				if (!sna->render.copy_boxes(sna, GXcopy,
 							     &tmp, bo, 0, 0,
 							     dst, dst_bo, dx, dy,
-							     REGION_RECTS(&this), REGION_NUM_RECTS(&this)))
+							     REGION_RECTS(&this), REGION_NUM_RECTS(&this), 0))
 					goto err;
 
 				kgem_bo_destroy(&sna->kgem, bo);
@@ -898,11 +898,11 @@ sna_tiling_copy_boxes(struct sna *sna, uint8_t alu,
 			i = (sna->render.copy_boxes(sna, GXcopy,
 						    src, src_bo, src_dx, src_dy,
 						    &p, tmp_bo, -tile.x1, -tile.y1,
-						    clipped, c - clipped) &&
+						    clipped, c - clipped, 0) &&
 			     sna->render.copy_boxes(sna, alu,
 						    &p, tmp_bo, -tile.x1, -tile.y1,
 						    dst, dst_bo, dst_dx, dst_dy,
-						    clipped, c - clipped));
+						    clipped, c - clipped, 0));
 
 			kgem_bo_destroy(&sna->kgem, tmp_bo);
 
diff --git a/src/sna/sna_trapezoids.c b/src/sna/sna_trapezoids.c
index 36defa10..d83c083c 100644
--- a/src/sna/sna_trapezoids.c
+++ b/src/sna/sna_trapezoids.c
@@ -5779,7 +5779,7 @@ trap_mask_converter(PicturePtr picture,
 	sna->render.copy_boxes(sna, GXcopy,
 			       scratch, sna_pixmap_get_bo(scratch), -extents.x1, -extents.x1,
 			       pixmap, priv->gpu_bo, x, y,
-			       &extents, 1);
+			       &extents, 1, 0);
 	mark_damaged(pixmap, priv, &extents ,x, y);
 
 	screen->DestroyPixmap(scratch);
@@ -5857,7 +5857,7 @@ trap_upload(PicturePtr picture,
 	sna->render.copy_boxes(sna, GXcopy,
 			       scratch, sna_pixmap_get_bo(scratch), -extents.x1, -extents.x1,
 			       pixmap, priv->gpu_bo, x, y,
-			       &extents, 1);
+			       &extents, 1, 0);
 	mark_damaged(pixmap, priv, &extents, x, y);
 
 	screen->DestroyPixmap(scratch);