sna/gen7: Reduce dst readbacks for unsupported sources

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

1 files changed, 109 insertions, 0 deletions

diff --git a/src/sna/gen7_render.c b/src/sna/gen7_render.c
index 3159da32..68ec2219 100644
--- a/src/sna/gen7_render.c
+++ b/src/sna/gen7_render.c
@@ -2202,6 +2202,109 @@ try_blt(struct sna *sna, int width, int height)
 	return FALSE;
 }
 
+static bool
+is_solid(PicturePtr picture)
+{
+	return  picture->pDrawable->width == 1 &&
+		picture->pDrawable->height == 1 &&
+		picture->repeat;
+}
+
+static bool
+is_gradient(PicturePtr picture)
+{
+	if (picture->pDrawable)
+		return FALSE;
+
+	return picture->pSourcePict->type != SourcePictTypeSolidFill;
+}
+
+static bool
+source_fallback(PicturePtr p)
+{
+	return is_gradient(p) || !gen7_check_filter(p) || !gen7_check_repeat(p);
+}
+
+static bool
+gen7_composite_fallback(struct sna *sna,
+			PicturePtr src,
+			PicturePtr mask,
+			PicturePtr dst)
+{
+	struct sna_pixmap *priv;
+	PixmapPtr src_pixmap;
+	PixmapPtr mask_pixmap;
+	PixmapPtr dst_pixmap;
+
+	if (!gen7_check_dst_format(dst->format)) {
+		DBG(("%s: unknown destination format: %d\n",
+		     __FUNCTION__, dst->format));
+		return TRUE;
+	}
+
+	dst_pixmap = get_drawable_pixmap(dst->pDrawable);
+	src_pixmap = src->pDrawable ? get_drawable_pixmap(src->pDrawable) : NULL;
+	mask_pixmap = (mask && mask->pDrawable) ? get_drawable_pixmap(mask->pDrawable) : NULL;
+
+	/* If we are using the destination as a source and need to
+	 * readback in order to upload the source, do it all
+	 * on the cpu.
+	 */
+	if (src_pixmap == dst_pixmap && source_fallback(src)) {
+		DBG(("%s: src is dst and will fallback\n",__FUNCTION__));
+		return TRUE;
+	}
+	if (mask_pixmap == dst_pixmap && source_fallback(mask)) {
+		DBG(("%s: mask is dst and will fallback\n",__FUNCTION__));
+		return TRUE;
+	}
+
+	/* If anything is on the GPU, push everything out to the GPU */
+	priv = sna_pixmap(dst_pixmap);
+	if (priv && priv->gpu_damage) {
+		DBG(("%s: dst is already on the GPU, try to use GPU\n",
+		     __FUNCTION__));
+		return FALSE;
+	}
+
+	if (src_pixmap && !is_solid(src) && !source_fallback(src)) {
+		priv = sna_pixmap(src_pixmap);
+		if (priv && priv->gpu_damage) {
+			DBG(("%s: src is already on the GPU, try to use GPU\n",
+			     __FUNCTION__));
+			return FALSE;
+		}
+	}
+	if (mask_pixmap && !is_solid(mask) && !source_fallback(mask)) {
+		priv = sna_pixmap(mask_pixmap);
+		if (priv && priv->gpu_damage) {
+			DBG(("%s: mask is already on the GPU, try to use GPU\n",
+			     __FUNCTION__));
+			return FALSE;
+		}
+	}
+
+	/* However if the dst is not on the GPU and we need to
+	 * render one of the sources using the CPU, we may
+	 * as well do the entire operation in place onthe CPU.
+	 */
+	if (source_fallback(src)) {
+		DBG(("%s: dst is on the CPU and src will fallback\n",
+		     __FUNCTION__));
+		return TRUE;
+	}
+
+	if (mask && source_fallback(mask)) {
+		DBG(("%s: dst is on the CPU and mask will fallback\n",
+		     __FUNCTION__));
+		return TRUE;
+	}
+
+	DBG(("%s: dst is not on the GPU and the operation should not fallback\n",
+	     __FUNCTION__));
+	return FALSE;
+}
+
 static Bool
 gen7_render_composite(struct sna *sna,
 		      uint8_t op,
@@ -2240,6 +2343,9 @@ gen7_render_composite(struct sna *sna,
 			      width, height, tmp))
 		return TRUE;
 
+	if (gen7_composite_fallback(sna, src, mask, dst))
+		return FALSE;
+
 	if (need_tiling(sna, width, height))
 		return sna_tiling_composite(op, src, mask, dst,
 					    src_x, src_y,
@@ -2603,6 +2709,9 @@ gen7_render_composite_spans(struct sna *sna,
 	if (need_tiling(sna, width, height))
 		return FALSE;
 
+	if (gen7_composite_fallback(sna, src, NULL, dst))
+		return FALSE;
+
 	tmp->base.op = op;
 	if (!gen7_composite_set_target(&tmp->base, dst))
 		return FALSE;