sna/video: Correct scaling of source offsets

When applying pan and zoom to a mismatched video, it would inevitably
miscompute the origin and scale factors.

Reported-by: Matti Hamalainen <ccr@tnsp.org>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=61610
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>

8 files changed, 141 insertions, 120 deletions

diff --git a/src/sna/gen3_render.c b/src/sna/gen3_render.c
index 0bac8d31..33116124 100644
--- a/src/sna/gen3_render.c
+++ b/src/sna/gen3_render.c
@@ -3024,8 +3024,10 @@ gen3_composite_picture(struct sna *sna,
 
 		switch (source->type) {
 		case SourcePictTypeSolidFill:
-			DBG(("%s: solid fill [%08x], format %x\n",
-			     __FUNCTION__, source->solidFill.color, picture->format));
+			DBG(("%s: solid fill [%08x], format %08x\n",
+			     __FUNCTION__,
+			     (unsigned)source->solidFill.color,
+			     (unsigned)picture->format));
 			ret = gen3_init_solid(channel, source->solidFill.color);
 			break;
 
@@ -5103,10 +5105,8 @@ gen3_emit_video_state(struct sna *sna,
 		OUT_BATCH(SS2_COLORSPACE_CONVERSION |
 			  (FILTER_LINEAR << SS2_MAG_FILTER_SHIFT) |
 			  (FILTER_LINEAR << SS2_MIN_FILTER_SHIFT));
-		OUT_BATCH((TEXCOORDMODE_CLAMP_EDGE <<
-			   SS3_TCX_ADDR_MODE_SHIFT) |
-			  (TEXCOORDMODE_CLAMP_EDGE <<
-			   SS3_TCY_ADDR_MODE_SHIFT) |
+		OUT_BATCH((TEXCOORDMODE_CLAMP_EDGE << SS3_TCX_ADDR_MODE_SHIFT) |
+			  (TEXCOORDMODE_CLAMP_EDGE << SS3_TCY_ADDR_MODE_SHIFT) |
 			  (0 << SS3_TEXTUREMAP_INDEX_SHIFT) |
 			  SS3_NORMALIZED_COORDS);
 		OUT_BATCH(0x00000000);
@@ -5220,30 +5220,24 @@ gen3_emit_video_state(struct sna *sna,
 		/* sampler 0 */
 		OUT_BATCH((FILTER_LINEAR << SS2_MAG_FILTER_SHIFT) |
 			  (FILTER_LINEAR << SS2_MIN_FILTER_SHIFT));
-		OUT_BATCH((TEXCOORDMODE_CLAMP_EDGE <<
-			   SS3_TCX_ADDR_MODE_SHIFT) |
-			  (TEXCOORDMODE_CLAMP_EDGE <<
-			   SS3_TCY_ADDR_MODE_SHIFT) |
+		OUT_BATCH((TEXCOORDMODE_CLAMP_EDGE << SS3_TCX_ADDR_MODE_SHIFT) |
+			  (TEXCOORDMODE_CLAMP_EDGE << SS3_TCY_ADDR_MODE_SHIFT) |
 			  (0 << SS3_TEXTUREMAP_INDEX_SHIFT) |
 			  SS3_NORMALIZED_COORDS);
 		OUT_BATCH(0x00000000);
 		/* sampler 1 */
 		OUT_BATCH((FILTER_LINEAR << SS2_MAG_FILTER_SHIFT) |
 			  (FILTER_LINEAR << SS2_MIN_FILTER_SHIFT));
-		OUT_BATCH((TEXCOORDMODE_CLAMP_EDGE <<
-			   SS3_TCX_ADDR_MODE_SHIFT) |
-			  (TEXCOORDMODE_CLAMP_EDGE <<
-			   SS3_TCY_ADDR_MODE_SHIFT) |
+		OUT_BATCH((TEXCOORDMODE_CLAMP_EDGE << SS3_TCX_ADDR_MODE_SHIFT) |
+			  (TEXCOORDMODE_CLAMP_EDGE << SS3_TCY_ADDR_MODE_SHIFT) |
 			  (1 << SS3_TEXTUREMAP_INDEX_SHIFT) |
 			  SS3_NORMALIZED_COORDS);
 		OUT_BATCH(0x00000000);
 		/* sampler 2 */
 		OUT_BATCH((FILTER_LINEAR << SS2_MAG_FILTER_SHIFT) |
 			  (FILTER_LINEAR << SS2_MIN_FILTER_SHIFT));
-		OUT_BATCH((TEXCOORDMODE_CLAMP_EDGE <<
-			   SS3_TCX_ADDR_MODE_SHIFT) |
-			  (TEXCOORDMODE_CLAMP_EDGE <<
-			   SS3_TCY_ADDR_MODE_SHIFT) |
+		OUT_BATCH((TEXCOORDMODE_CLAMP_EDGE << SS3_TCX_ADDR_MODE_SHIFT) |
+			  (TEXCOORDMODE_CLAMP_EDGE << SS3_TCY_ADDR_MODE_SHIFT) |
 			  (2 << SS3_TEXTUREMAP_INDEX_SHIFT) |
 			  SS3_NORMALIZED_COORDS);
 		OUT_BATCH(0x00000000);
@@ -5359,7 +5353,6 @@ gen3_emit_video_state(struct sna *sna,
 				(sna->kgem.nbatch - id - 2);
 		}
 	}
-
 }
 
 static void
@@ -5395,37 +5388,51 @@ gen3_render_video(struct sna *sna,
 		  struct sna_video *video,
 		  struct sna_video_frame *frame,
 		  RegionPtr dstRegion,
-		  short src_w, short src_h,
-		  short drw_w, short drw_h,
-		  short dx, short dy,
 		  PixmapPtr pixmap)
 {
 	struct sna_pixmap *priv = sna_pixmap(pixmap);
 	BoxPtr pbox = REGION_RECTS(dstRegion);
 	int nbox = REGION_NUM_RECTS(dstRegion);
-	int width = dstRegion->extents.x2 - dstRegion->extents.x1;
-	int height = dstRegion->extents.y2 - dstRegion->extents.y1;
+	int dst_width = dstRegion->extents.x2 - dstRegion->extents.x1;
+	int dst_height = dstRegion->extents.y2 - dstRegion->extents.y1;
+	int src_width = frame->src.x2 - frame->src.x1;
+	int src_height = frame->src.y2 - frame->src.y1;
+	float src_offset_x, src_offset_y;
 	float src_scale_x, src_scale_y;
 	int pix_xoff, pix_yoff;
 	struct kgem_bo *dst_bo;
 	bool bilinear;
 	int copy = 0;
 
-	DBG(("%s: %dx%d -> %dx%d\n", __FUNCTION__, src_w, src_h, drw_w, drw_h));
+	DBG(("%s: src:%dx%d (frame:%dx%d) -> dst:%dx%d\n", __FUNCTION__,
+	     src_width, src_height, frame->width, frame->height, dst_width, dst_height));
 
 	dst_bo = priv->gpu_bo;
 	if (dst_bo == NULL)
 		return false;
 
+	bilinear = src_width != dst_width || src_height != dst_height;
+
+	src_scale_x = (float)src_width / dst_width / frame->width;
+	src_offset_x = (float)frame->src.x1 / frame->width - dstRegion->extents.x1 * src_scale_x;
+
+	src_scale_y = (float)src_height / dst_height / frame->height;
+	src_offset_y = (float)frame->src.y1 / frame->height - dstRegion->extents.y1 * src_scale_y;
+	DBG(("%s: src offset (%f, %f), scale (%f, %f)\n",
+	     __FUNCTION__, src_offset_x, src_offset_y, src_scale_x, src_scale_y));
+
 	if (too_large(pixmap->drawable.width, pixmap->drawable.height) ||
 	    !gen3_check_pitch_3d(dst_bo)) {
 		int bpp = pixmap->drawable.bitsPerPixel;
 
+		if (too_large(dst_width, dst_height))
+			return false;
+
 		dst_bo = kgem_create_2d(&sna->kgem,
-					width, height, bpp,
+					dst_width, dst_height, bpp,
 					kgem_choose_tiling(&sna->kgem,
 							   I915_TILING_X,
-							   width, height, bpp),
+							   dst_width, dst_height, bpp),
 					0);
 		if (!dst_bo)
 			return false;
@@ -5434,9 +5441,6 @@ gen3_render_video(struct sna *sna,
 		pix_yoff = -dstRegion->extents.y1;
 		copy = 1;
 	} else {
-		width = pixmap->drawable.width;
-		height = pixmap->drawable.height;
-
 		/* Set up the offset for translating from the given region
 		 * (in screen coordinates) to the backing pixmap.
 		 */
@@ -5447,32 +5451,27 @@ gen3_render_video(struct sna *sna,
 		pix_xoff = 0;
 		pix_yoff = 0;
 #endif
-	}
 
-	bilinear = src_w != drw_w || src_h != drw_h;
-
-	src_scale_x = ((float)src_w / frame->width) / drw_w;
-	src_scale_y = ((float)src_h / frame->height) / drw_h;
-
-	DBG(("%s: src offset=(%d, %d), scale=(%f, %f), dst offset=(%d, %d)\n",
-	     __FUNCTION__,
-	     dx, dy, src_scale_x, src_scale_y, pix_xoff, pix_yoff));
+		dst_width  = pixmap->drawable.width;
+		dst_height = pixmap->drawable.height;
+	}
 
 	gen3_video_get_batch(sna, dst_bo);
 	gen3_emit_video_state(sna, video, frame, pixmap,
-			      dst_bo, width, height, bilinear);
+			      dst_bo, dst_width, dst_height, bilinear);
 	do {
 		int nbox_this_time = gen3_get_inline_rectangles(sna, nbox, 4);
 		if (nbox_this_time == 0) {
 			gen3_video_get_batch(sna, dst_bo);
 			gen3_emit_video_state(sna, video, frame, pixmap,
-					      dst_bo, width, height, bilinear);
+					      dst_bo, dst_width, dst_height, bilinear);
 			nbox_this_time = gen3_get_inline_rectangles(sna, nbox, 4);
+			assert(nbox_this_time);
 		}
 		nbox -= nbox_this_time;
 
 		OUT_BATCH(PRIM3D_RECTLIST | (12 * nbox_this_time - 1));
-		while (nbox_this_time--) {
+		do {
 			int box_x1 = pbox->x1;
 			int box_y1 = pbox->y1;
 			int box_x2 = pbox->x2;
@@ -5480,27 +5479,31 @@ gen3_render_video(struct sna *sna,
 
 			pbox++;
 
-			DBG(("%s: box (%d, %d), (%d, %d)\n",
-			     __FUNCTION__, box_x1, box_y1, box_x2, box_y2));
+			DBG(("%s: dst (%d, %d), (%d, %d) + (%d, %d); src (%f, %f), (%f, %f)\n",
+			     __FUNCTION__, box_x1, box_y1, box_x2, box_y2, pix_xoff, pix_yoff,
+			     box_x1 * src_scale_x + src_offset_x,
+			     box_y1 * src_scale_y + src_offset_y,
+			     box_x2 * src_scale_x + src_offset_x,
+			     box_y2 * src_scale_y + src_offset_y));
 
 			/* bottom right */
 			OUT_BATCH_F(box_x2 + pix_xoff);
 			OUT_BATCH_F(box_y2 + pix_yoff);
-			OUT_BATCH_F((box_x2 - dx) * src_scale_x);
-			OUT_BATCH_F((box_y2 - dy) * src_scale_y);
+			OUT_BATCH_F(box_x2 * src_scale_x + src_offset_x);
+			OUT_BATCH_F(box_y2 * src_scale_y + src_offset_y);
 
 			/* bottom left */
 			OUT_BATCH_F(box_x1 + pix_xoff);
 			OUT_BATCH_F(box_y2 + pix_yoff);
-			OUT_BATCH_F((box_x1 - dx) * src_scale_x);
-			OUT_BATCH_F((box_y2 - dy) * src_scale_y);
+			OUT_BATCH_F(box_x1 * src_scale_x + src_offset_x);
+			OUT_BATCH_F(box_y2 * src_scale_y + src_offset_y);
 
 			/* top left */
 			OUT_BATCH_F(box_x1 + pix_xoff);
 			OUT_BATCH_F(box_y1 + pix_yoff);
-			OUT_BATCH_F((box_x1 - dx) * src_scale_x);
-			OUT_BATCH_F((box_y1 - dy) * src_scale_y);
-		}
+			OUT_BATCH_F(box_x1 * src_scale_x + src_offset_x);
+			OUT_BATCH_F(box_y1 * src_scale_y + src_offset_y);
+		} while (--nbox_this_time);
 	} while (nbox);
 
 	if (copy) {
diff --git a/src/sna/gen4_render.c b/src/sna/gen4_render.c
index 4abcb97b..f78c5da1 100644
--- a/src/sna/gen4_render.c
+++ b/src/sna/gen4_render.c
@@ -1315,18 +1315,21 @@ gen4_render_video(struct sna *sna,
 		  struct sna_video *video,
 		  struct sna_video_frame *frame,
 		  RegionPtr dstRegion,
-		  short src_w, short src_h,
-		  short drw_w, short drw_h,
-		  short dx, short dy,
 		  PixmapPtr pixmap)
 {
 	struct sna_composite_op tmp;
-	int nbox, pix_xoff, pix_yoff;
+	int dst_width = dstRegion->extents.x2 - dstRegion->extents.x1;
+	int dst_height = dstRegion->extents.y2 - dstRegion->extents.y1;
+	int src_width = frame->src.x2 - frame->src.x1;
+	int src_height = frame->src.y2 - frame->src.y1;
+	float src_offset_x, src_offset_y;
 	float src_scale_x, src_scale_y;
+	int nbox, pix_xoff, pix_yoff;
 	struct sna_pixmap *priv;
 	BoxPtr box;
 
-	DBG(("%s: %dx%d -> %dx%d\n", __FUNCTION__, src_w, src_h, drw_w, drw_h));
+	DBG(("%s: %dx%d -> %dx%d\n", __FUNCTION__,
+	     src_width, src_height, dst_width, dst_height));
 
 	priv = sna_pixmap_force_to_gpu(pixmap, MOVE_READ | MOVE_WRITE);
 	if (priv == NULL)
@@ -1341,7 +1344,7 @@ gen4_render_video(struct sna *sna,
 	tmp.dst.format = sna_format_for_depth(pixmap->drawable.depth);
 	tmp.dst.bo = priv->gpu_bo;
 
-	if (src_w == drw_w && src_h == drw_h)
+	if (src_width == dst_width && src_height == dst_height)
 		tmp.src.filter = SAMPLER_FILTER_NEAREST;
 	else
 		tmp.src.filter = SAMPLER_FILTER_BILINEAR;
@@ -1375,9 +1378,11 @@ gen4_render_video(struct sna *sna,
 	pix_yoff = 0;
 #endif
 
-	/* Use normalized texture coordinates */
-	src_scale_x = ((float)src_w / frame->width) / (float)drw_w;
-	src_scale_y = ((float)src_h / frame->height) / (float)drw_h;
+	src_scale_x = (float)src_width / dst_width / frame->width;
+	src_offset_x = frame->src.x1 / frame->width - dstRegion->extents.x1 * src_scale_x;
+
+	src_scale_y = (float)src_height / dst_height / frame->height;
+	src_offset_y = frame->src.y1 / frame->height - dstRegion->extents.y1 * src_scale_y;
 
 	box = REGION_RECTS(dstRegion);
 	nbox = REGION_NUM_RECTS(dstRegion);
@@ -1392,16 +1397,16 @@ gen4_render_video(struct sna *sna,
 		gen4_get_rectangles(sna, &tmp, 1, gen4_video_bind_surfaces);
 
 		OUT_VERTEX(r.x2, r.y2);
-		OUT_VERTEX_F((box->x2 - dx) * src_scale_x);
-		OUT_VERTEX_F((box->y2 - dy) * src_scale_y);
+		OUT_VERTEX_F(box->x2 * src_scale_x + src_offset_x);
+		OUT_VERTEX_F(box->y2 * src_scale_y + src_offset_y);
 
 		OUT_VERTEX(r.x1, r.y2);
-		OUT_VERTEX_F((box->x1 - dx) * src_scale_x);
-		OUT_VERTEX_F((box->y2 - dy) * src_scale_y);
+		OUT_VERTEX_F(box->x1 * src_scale_x + src_offset_x);
+		OUT_VERTEX_F(box->y2 * src_scale_y + src_offset_y);
 
 		OUT_VERTEX(r.x1, r.y1);
-		OUT_VERTEX_F((box->x1 - dx) * src_scale_x);
-		OUT_VERTEX_F((box->y1 - dy) * src_scale_y);
+		OUT_VERTEX_F(box->x1 * src_scale_x + src_offset_x);
+		OUT_VERTEX_F(box->y1 * src_scale_y + src_offset_y);
 
 		if (!DAMAGE_IS_ALL(priv->gpu_damage)) {
 			sna_damage_add_box(&priv->gpu_damage, &r);
diff --git a/src/sna/gen5_render.c b/src/sna/gen5_render.c
index 9a9c223e..8c8a996f 100644
--- a/src/sna/gen5_render.c
+++ b/src/sna/gen5_render.c
@@ -1299,18 +1299,21 @@ gen5_render_video(struct sna *sna,
 		  struct sna_video *video,
 		  struct sna_video_frame *frame,
 		  RegionPtr dstRegion,
-		  short src_w, short src_h,
-		  short drw_w, short drw_h,
-		  short dx, short dy,
 		  PixmapPtr pixmap)
 {
 	struct sna_composite_op tmp;
-	int nbox, pix_xoff, pix_yoff;
+	int dst_width = dstRegion->extents.x2 - dstRegion->extents.x1;
+	int dst_height = dstRegion->extents.y2 - dstRegion->extents.y1;
+	int src_width = frame->src.x2 - frame->src.x1;
+	int src_height = frame->src.y2 - frame->src.y1;
+	float src_offset_x, src_offset_y;
 	float src_scale_x, src_scale_y;
+	int nbox, pix_xoff, pix_yoff;
 	struct sna_pixmap *priv;
 	BoxPtr box;
 
-	DBG(("%s: %dx%d -> %dx%d\n", __FUNCTION__, src_w, src_h, drw_w, drw_h));
+	DBG(("%s: %dx%d -> %dx%d\n", __FUNCTION__,
+	     src_width, src_height, dst_width, dst_height));
 
 	priv = sna_pixmap_force_to_gpu(pixmap, MOVE_READ | MOVE_WRITE);
 	if (priv == NULL)
@@ -1325,7 +1328,7 @@ gen5_render_video(struct sna *sna,
 	tmp.dst.format = sna_format_for_depth(pixmap->drawable.depth);
 	tmp.dst.bo = priv->gpu_bo;
 
-	if (src_w == drw_w && src_h == drw_h)
+	if (src_width == dst_width && src_height == dst_height)
 		tmp.src.filter = SAMPLER_FILTER_NEAREST;
 	else
 		tmp.src.filter = SAMPLER_FILTER_BILINEAR;
@@ -1359,9 +1362,11 @@ gen5_render_video(struct sna *sna,
 	pix_yoff = 0;
 #endif
 
-	/* Use normalized texture coordinates */
-	src_scale_x = ((float)src_w / frame->width) / (float)drw_w;
-	src_scale_y = ((float)src_h / frame->height) / (float)drw_h;
+	src_scale_x = (float)src_width / dst_width / frame->width;
+	src_offset_x = frame->src.x1 / frame->width - dstRegion->extents.x1 * src_scale_x;
+
+	src_scale_y = (float)src_height / dst_height / frame->height;
+	src_offset_y = frame->src.y1 / frame->height - dstRegion->extents.y1 * src_scale_y;
 
 	box = REGION_RECTS(dstRegion);
 	nbox = REGION_NUM_RECTS(dstRegion);
@@ -1376,16 +1381,16 @@ gen5_render_video(struct sna *sna,
 		gen5_get_rectangles(sna, &tmp, 1, gen5_video_bind_surfaces);
 
 		OUT_VERTEX(r.x2, r.y2);
-		OUT_VERTEX_F((box->x2 - dx) * src_scale_x);
-		OUT_VERTEX_F((box->y2 - dy) * src_scale_y);
+		OUT_VERTEX_F(box->x2 * src_scale_x + src_offset_x);
+		OUT_VERTEX_F(box->y2 * src_scale_y + src_offset_y);
 
 		OUT_VERTEX(r.x1, r.y2);
-		OUT_VERTEX_F((box->x1 - dx) * src_scale_x);
-		OUT_VERTEX_F((box->y2 - dy) * src_scale_y);
+		OUT_VERTEX_F(box->x1 * src_scale_x + src_offset_x);
+		OUT_VERTEX_F(box->y2 * src_scale_y + src_offset_y);
 
 		OUT_VERTEX(r.x1, r.y1);
-		OUT_VERTEX_F((box->x1 - dx) * src_scale_x);
-		OUT_VERTEX_F((box->y1 - dy) * src_scale_y);
+		OUT_VERTEX_F(box->x1 * src_scale_x + src_offset_x);
+		OUT_VERTEX_F(box->y1 * src_scale_y + src_offset_y);
 
 		if (!DAMAGE_IS_ALL(priv->gpu_damage)) {
 			sna_damage_add_box(&priv->gpu_damage, &r);
diff --git a/src/sna/gen6_render.c b/src/sna/gen6_render.c
index 31953881..8d15bd80 100644
--- a/src/sna/gen6_render.c
+++ b/src/sna/gen6_render.c
@@ -1573,20 +1573,23 @@ gen6_render_video(struct sna *sna,
 		  struct sna_video *video,
 		  struct sna_video_frame *frame,
 		  RegionPtr dstRegion,
-		  short src_w, short src_h,
-		  short drw_w, short drw_h,
-		  short dx, short dy,
 		  PixmapPtr pixmap)
 {
 	struct sna_composite_op tmp;
-	int nbox, pix_xoff, pix_yoff;
+	int dst_width = dstRegion->extents.x2 - dstRegion->extents.x1;
+	int dst_height = dstRegion->extents.y2 - dstRegion->extents.y1;
+	int src_width = frame->src.x2 - frame->src.x1;
+	int src_height = frame->src.y2 - frame->src.y1;
+	float src_offset_x, src_offset_y;
 	float src_scale_x, src_scale_y;
+	int nbox, pix_xoff, pix_yoff;
 	struct sna_pixmap *priv;
 	unsigned filter;
 	BoxPtr box;
 
 	DBG(("%s: src=(%d, %d), dst=(%d, %d), %dx[(%d, %d), (%d, %d)...]\n",
-	     __FUNCTION__, src_w, src_h, drw_w, drw_h,
+	     __FUNCTION__,
+	     src_width, src_height, dst_width, dst_height,
 	     REGION_NUM_RECTS(dstRegion),
 	     REGION_EXTENTS(NULL, dstRegion)->x1,
 	     REGION_EXTENTS(NULL, dstRegion)->y1,
@@ -1611,7 +1614,7 @@ gen6_render_video(struct sna *sna,
 	tmp.floats_per_vertex = 3;
 	tmp.floats_per_rect = 9;
 
-	if (src_w == drw_w && src_h == drw_h)
+	if (src_width == dst_width && src_height == dst_height)
 		filter = SAMPLER_FILTER_NEAREST;
 	else
 		filter = SAMPLER_FILTER_BILINEAR;
@@ -1647,9 +1650,11 @@ gen6_render_video(struct sna *sna,
 	pix_yoff = 0;
 #endif
 
-	/* Use normalized texture coordinates */
-	src_scale_x = ((float)src_w / frame->width) / (float)drw_w;
-	src_scale_y = ((float)src_h / frame->height) / (float)drw_h;
+	src_scale_x = (float)src_width / dst_width / frame->width;
+	src_offset_x = frame->src.x1 / frame->width - dstRegion->extents.x1 * src_scale_x;
+
+	src_scale_y = (float)src_height / dst_height / frame->height;
+	src_offset_y = frame->src.y1 / frame->height - dstRegion->extents.y1 * src_scale_y;
 
 	box = REGION_RECTS(dstRegion);
 	nbox = REGION_NUM_RECTS(dstRegion);
@@ -1664,16 +1669,16 @@ gen6_render_video(struct sna *sna,
 		gen6_get_rectangles(sna, &tmp, 1, gen6_emit_video_state);
 
 		OUT_VERTEX(r.x2, r.y2);
-		OUT_VERTEX_F((box->x2 - dx) * src_scale_x);
-		OUT_VERTEX_F((box->y2 - dy) * src_scale_y);
+		OUT_VERTEX_F(box->x2 * src_scale_x + src_offset_x);
+		OUT_VERTEX_F(box->y2 * src_scale_y + src_offset_y);
 
 		OUT_VERTEX(r.x1, r.y2);
-		OUT_VERTEX_F((box->x1 - dx) * src_scale_x);
-		OUT_VERTEX_F((box->y2 - dy) * src_scale_y);
+		OUT_VERTEX_F(box->x1 * src_scale_x + src_offset_x);
+		OUT_VERTEX_F(box->y2 * src_scale_y + src_offset_y);
 
 		OUT_VERTEX(r.x1, r.y1);
-		OUT_VERTEX_F((box->x1 - dx) * src_scale_x);
-		OUT_VERTEX_F((box->y1 - dy) * src_scale_y);
+		OUT_VERTEX_F(box->x1 * src_scale_x + src_offset_x);
+		OUT_VERTEX_F(box->y1 * src_scale_y + src_offset_y);
 
 		if (!DAMAGE_IS_ALL(priv->gpu_damage)) {
 			sna_damage_add_box(&priv->gpu_damage, &r);
diff --git a/src/sna/gen7_render.c b/src/sna/gen7_render.c
index 0dae8c43..4453797f 100644
--- a/src/sna/gen7_render.c
+++ b/src/sna/gen7_render.c
@@ -1715,20 +1715,23 @@ gen7_render_video(struct sna *sna,
 		  struct sna_video *video,
 		  struct sna_video_frame *frame,
 		  RegionPtr dstRegion,
-		  short src_w, short src_h,
-		  short drw_w, short drw_h,
-		  short dx, short dy,
 		  PixmapPtr pixmap)
 {
 	struct sna_composite_op tmp;
-	int nbox, pix_xoff, pix_yoff;
+	int dst_width = dstRegion->extents.x2 - dstRegion->extents.x1;
+	int dst_height = dstRegion->extents.y2 - dstRegion->extents.y1;
+	int src_width = frame->src.x2 - frame->src.x1;
+	int src_height = frame->src.y2 - frame->src.y1;
+	float src_offset_x, src_offset_y;
 	float src_scale_x, src_scale_y;
+	int nbox, pix_xoff, pix_yoff;
 	struct sna_pixmap *priv;
 	unsigned filter;
 	BoxPtr box;
 
 	DBG(("%s: src=(%d, %d), dst=(%d, %d), %dx[(%d, %d), (%d, %d)...]\n",
-	     __FUNCTION__, src_w, src_h, drw_w, drw_h,
+	     __FUNCTION__,
+	     src_width, src_height, dst_width, dst_height,
 	     REGION_NUM_RECTS(dstRegion),
 	     REGION_EXTENTS(NULL, dstRegion)->x1,
 	     REGION_EXTENTS(NULL, dstRegion)->y1,
@@ -1753,7 +1756,7 @@ gen7_render_video(struct sna *sna,
 	tmp.floats_per_vertex = 3;
 	tmp.floats_per_rect = 9;
 
-	if (src_w == drw_w && src_h == drw_h)
+	if (src_width == dst_width && src_height == dst_height)
 		filter = SAMPLER_FILTER_NEAREST;
 	else
 		filter = SAMPLER_FILTER_BILINEAR;
@@ -1789,9 +1792,11 @@ gen7_render_video(struct sna *sna,
 	pix_yoff = 0;
 #endif
 
-	/* Use normalized texture coordinates */
-	src_scale_x = ((float)src_w / frame->width) / (float)drw_w;
-	src_scale_y = ((float)src_h / frame->height) / (float)drw_h;
+	src_scale_x = (float)src_width / dst_width / frame->width;
+	src_offset_x = frame->src.x1 / frame->width - dstRegion->extents.x1 * src_scale_x;
+
+	src_scale_y = (float)src_height / dst_height / frame->height;
+	src_offset_y = frame->src.y1 / frame->height - dstRegion->extents.y1 * src_scale_y;
 
 	box = REGION_RECTS(dstRegion);
 	nbox = REGION_NUM_RECTS(dstRegion);
@@ -1806,16 +1811,16 @@ gen7_render_video(struct sna *sna,
 		gen7_get_rectangles(sna, &tmp, 1, gen7_emit_video_state);
 
 		OUT_VERTEX(r.x2, r.y2);
-		OUT_VERTEX_F((box->x2 - dx) * src_scale_x);
-		OUT_VERTEX_F((box->y2 - dy) * src_scale_y);
+		OUT_VERTEX_F(box->x2 * src_scale_x + src_offset_x);
+		OUT_VERTEX_F(box->y2 * src_scale_y + src_offset_y);
 
 		OUT_VERTEX(r.x1, r.y2);
-		OUT_VERTEX_F((box->x1 - dx) * src_scale_x);
-		OUT_VERTEX_F((box->y2 - dy) * src_scale_y);
+		OUT_VERTEX_F(box->x1 * src_scale_x + src_offset_x);
+		OUT_VERTEX_F(box->y2 * src_scale_y + src_offset_y);
 
 		OUT_VERTEX(r.x1, r.y1);
-		OUT_VERTEX_F((box->x1 - dx) * src_scale_x);
-		OUT_VERTEX_F((box->y1 - dy) * src_scale_y);
+		OUT_VERTEX_F(box->x1 * src_scale_x + src_offset_x);
+		OUT_VERTEX_F(box->y1 * src_scale_y + src_offset_y);
 
 		if (!DAMAGE_IS_ALL(priv->gpu_damage)) {
 			sna_damage_add_box(&priv->gpu_damage, &r);
diff --git a/src/sna/sna_render.h b/src/sna/sna_render.h
index 01176c6a..1a15b70e 100644
--- a/src/sna/sna_render.h
+++ b/src/sna/sna_render.h
@@ -245,9 +245,6 @@ struct sna_render {
 		      struct sna_video *video,
 		      struct sna_video_frame *frame,
 		      RegionPtr dstRegion,
-		      short src_w, short src_h,
-		      short drw_w, short drw_h,
-		      short dx, short dy,
 		      PixmapPtr pixmap);
 
 	bool (*fill_boxes)(struct sna *sna,
diff --git a/src/sna/sna_video.c b/src/sna/sna_video.c
index ecf52d21..a0eb0f4e 100644
--- a/src/sna/sna_video.c
+++ b/src/sna/sna_video.c
@@ -449,9 +449,9 @@ sna_video_copy_data(struct sna *sna,
 {
 	uint8_t *dst;
 
-	DBG(("%s: handle=%d, size=%dx%d [%d], rotation=%d, is-texture=%d\n",
+	DBG(("%s: handle=%d, size=%dx%d [%d], pitch=[%d,%d] rotation=%d, is-texture=%d\n",
 	     __FUNCTION__, frame->bo ? frame->bo->handle : 0,
-	     frame->width, frame->height, frame->size,
+	     frame->width, frame->height, frame->size, frame->pitch[0], frame->pitch[1],
 	     video->rotation, video->textured));
 	DBG(("%s: image=(%d, %d), (%d, %d), source=(%d, %d), (%d, %d)\n",
 	     __FUNCTION__,
@@ -462,6 +462,8 @@ sna_video_copy_data(struct sna *sna,
 
 	/* In the common case, we can simply the upload in a single pwrite */
 	if (video->rotation == RR_Rotate_0 && !video->tiled) {
+		DBG(("%s: unrotated, untiled fast paths: is-planar?=%d\n",
+		     __FUNCTION__, is_planar_fourcc(frame->id)));
 		if (is_planar_fourcc(frame->id)) {
 			int w = frame->image.x2 - frame->image.x1;
 			int h = frame->image.y2 - frame->image.y1;
@@ -508,6 +510,8 @@ sna_video_copy_data(struct sna *sna,
 				return true;
 			}
 		}
+
+		DBG(("%s: source cropped, fallback\n", __FUNCTION__));
 	}
 
 	/* copy data, must use GTT so that we keep the overlay uncached */
diff --git a/src/sna/sna_video_textured.c b/src/sna/sna_video_textured.c
index e5cae859..bd203259 100644
--- a/src/sna/sna_video_textured.c
+++ b/src/sna/sna_video_textured.c
@@ -287,10 +287,7 @@ sna_video_textured_put_image(ScrnInfoPtr scrn,
 	}
 
 	ret = Success;
-	if (!sna->render.video(sna, video, &frame, clip,
-			       src_w, src_h, drw_w, drw_h,
-			       drw_x - src_x, drw_y - src_y,
-			       pixmap)) {
+	if (!sna->render.video(sna, video, &frame, clip, pixmap)) {
 		DBG(("%s: failed to render video\n", __FUNCTION__));
 		ret = BadAlloc;
 	} else