sna: Implement extended fallback handling for src == dst copies

Only marginally better than falling all the way back to using the CPU,
is to perform a double copy to workaround the overlapping copy.

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

10 files changed, 157 insertions, 69 deletions

diff --git a/src/sna/gen2_render.c b/src/sna/gen2_render.c
index 373866b9..d10f03e6 100644
--- a/src/sna/gen2_render.c
+++ b/src/sna/gen2_render.c
@@ -2520,16 +2520,11 @@ gen2_render_copy_boxes(struct sna *sna, uint8_t alu,
 	    too_large(src->drawable.width, src->drawable.height) ||
 	    src_bo->pitch > 8192 ||
 	    too_large(dst->drawable.width, dst->drawable.height) ||
-	    dst_bo->pitch < 8 || dst_bo->pitch > 8192) {
-		if (!sna_blt_compare_depth(&src->drawable, &dst->drawable))
-			return FALSE;
-
-		return sna_blt_copy_boxes(sna, alu,
-					  src_bo, src_dx, src_dy,
-					  dst_bo, dst_dx, dst_dy,
-					  dst->drawable.bitsPerPixel,
-					  box, n);
-	}
+	    dst_bo->pitch < 8 || dst_bo->pitch > 8192)
+		return sna_blt_copy_boxes_fallback(sna, alu,
+						   src, src_bo, src_dx, src_dy,
+						   dst, dst_bo, dst_dx, dst_dy,
+						   box, n);
 
 	if (!kgem_check_bo(&sna->kgem, dst_bo, src_bo, NULL))
 		kgem_submit(&sna->kgem);
diff --git a/src/sna/gen3_render.c b/src/sna/gen3_render.c
index fe3a359c..4ada863d 100644
--- a/src/sna/gen3_render.c
+++ b/src/sna/gen3_render.c
@@ -3630,16 +3630,11 @@ gen3_render_copy_boxes(struct sna *sna, uint8_t alu,
 	    src_bo->pitch > 8192 ||
 	    too_large(src->drawable.width, src->drawable.height) ||
 	    dst_bo->pitch > 8192 ||
-	    too_large(dst->drawable.width, dst->drawable.height)) {
-		if (!sna_blt_compare_depth(&src->drawable, &dst->drawable))
-			return FALSE;
-
-		return sna_blt_copy_boxes(sna, alu,
-					  src_bo, src_dx, src_dy,
-					  dst_bo, dst_dx, dst_dy,
-					  dst->drawable.bitsPerPixel,
-					  box, n);
-	}
+	    too_large(dst->drawable.width, dst->drawable.height))
+		return sna_blt_copy_boxes_fallback(sna, alu,
+						   src, src_bo, src_dx, src_dy,
+						   dst, dst_bo, dst_dx, dst_dy,
+						   box, n);
 
 	if (!kgem_check_bo(&sna->kgem, dst_bo, src_bo, NULL))
 		kgem_submit(&sna->kgem);
diff --git a/src/sna/gen4_render.c b/src/sna/gen4_render.c
index 1c6c22ef..aec5ca04 100644
--- a/src/sna/gen4_render.c
+++ b/src/sna/gen4_render.c
@@ -2279,16 +2279,11 @@ gen4_render_copy_boxes(struct sna *sna, uint8_t alu,
 
 	if (!(alu == GXcopy || alu == GXclear) || src_bo == dst_bo ||
 	    src->drawable.width > 8192 || src->drawable.height > 8192 ||
-	    dst->drawable.width > 8192 || dst->drawable.height > 8192) {
-		if (!sna_blt_compare_depth(&src->drawable, &dst->drawable))
-			return FALSE;
-
-		return sna_blt_copy_boxes(sna, alu,
-					  src_bo, src_dx, src_dy,
-					  dst_bo, dst_dx, dst_dy,
-					  dst->drawable.bitsPerPixel,
-					  box, n);
-	}
+	    dst->drawable.width > 8192 || dst->drawable.height > 8192)
+		return sna_blt_copy_boxes_fallback(sna, alu,
+						   src, src_bo, src_dx, src_dy,
+						   dst, dst_bo, dst_dx, dst_dy,
+						   box, n);
 
 	DBG(("%s (%d, %d)->(%d, %d) x %d\n",
 	     __FUNCTION__, src_dx, src_dy, dst_dx, dst_dy, n));
diff --git a/src/sna/gen5_render.c b/src/sna/gen5_render.c
index 4f5b57ca..6d89ca3a 100644
--- a/src/sna/gen5_render.c
+++ b/src/sna/gen5_render.c
@@ -2581,16 +2581,11 @@ gen5_render_copy_boxes(struct sna *sna, uint8_t alu,
 
 	if (!(alu == GXcopy || alu == GXclear) || src_bo == dst_bo ||
 	    too_large(src->drawable.width, src->drawable.height) ||
-	    too_large(dst->drawable.width, dst->drawable.height)) {
-		if (!sna_blt_compare_depth(&src->drawable, &dst->drawable))
-			return FALSE;
-
-		return sna_blt_copy_boxes(sna, alu,
-					  src_bo, src_dx, src_dy,
-					  dst_bo, dst_dx, dst_dy,
-					  dst->drawable.bitsPerPixel,
-					  box, n);
-	}
+	    too_large(dst->drawable.width, dst->drawable.height))
+		return sna_blt_copy_boxes_fallback(sna, alu,
+						   src, src_bo, src_dx, src_dy,
+						   dst, dst_bo, dst_dx, dst_dy,
+						   box, n);
 
 	DBG(("%s (%d, %d)->(%d, %d) x %d\n",
 	     __FUNCTION__, src_dx, src_dy, dst_dx, dst_dy, n));
diff --git a/src/sna/gen6_render.c b/src/sna/gen6_render.c
index 8808300a..c276f7e8 100644
--- a/src/sna/gen6_render.c
+++ b/src/sna/gen6_render.c
@@ -2791,16 +2791,12 @@ gen6_render_copy_boxes(struct sna *sna, uint8_t alu,
 
 	if (!(alu == GXcopy || alu == GXclear) || src_bo == dst_bo ||
 	    too_large(src->drawable.width, src->drawable.height) ||
-	    too_large(dst->drawable.width, dst->drawable.height)) {
-		if (!sna_blt_compare_depth(&src->drawable, &dst->drawable))
-			return FALSE;
+	    too_large(dst->drawable.width, dst->drawable.height))
+		return sna_blt_copy_boxes_fallback(sna, alu,
+						   src, src_bo, src_dx, src_dy,
+						   dst, dst_bo, dst_dx, dst_dy,
+						   box, n);
 
-		return sna_blt_copy_boxes(sna, alu,
-					  src_bo, src_dx, src_dy,
-					  dst_bo, dst_dx, dst_dy,
-					  dst->drawable.bitsPerPixel,
-					  box, n);
-	}
 	tmp.op = alu == GXcopy ? PictOpSrc : PictOpClear;
 
 	tmp.dst.pixmap = dst;
diff --git a/src/sna/gen7_render.c b/src/sna/gen7_render.c
index 0e18badc..dc97d181 100644
--- a/src/sna/gen7_render.c
+++ b/src/sna/gen7_render.c
@@ -2904,16 +2904,12 @@ gen7_render_copy_boxes(struct sna *sna, uint8_t alu,
 
 	if (!(alu == GXcopy || alu == GXclear) || src_bo == dst_bo ||
 	    too_large(src->drawable.width, src->drawable.height) ||
-	    too_large(dst->drawable.width, dst->drawable.height)) {
-		if (!sna_blt_compare_depth(&src->drawable, &dst->drawable))
-			return FALSE;
+	    too_large(dst->drawable.width, dst->drawable.height))
+		return sna_blt_copy_boxes_fallback(sna, alu,
+						   src, src_bo, src_dx, src_dy,
+						   dst, dst_bo, dst_dx, dst_dy,
+						   box, n);
 
-		return sna_blt_copy_boxes(sna, alu,
-					  src_bo, src_dx, src_dy,
-					  dst_bo, dst_dx, dst_dy,
-					  dst->drawable.bitsPerPixel,
-					  box, n);
-	}
 	tmp.op = alu == GXcopy ? PictOpSrc : PictOpClear;
 
 	tmp.dst.pixmap = dst;
diff --git a/src/sna/sna.h b/src/sna/sna.h
index 8deccc19..1fa31222 100644
--- a/src/sna/sna.h
+++ b/src/sna/sna.h
@@ -428,6 +428,7 @@ PixmapPtr sna_pixmap_create_upload(ScreenPtr screen,
 
 struct sna_pixmap *sna_pixmap_move_to_gpu(PixmapPtr pixmap);
 struct sna_pixmap *sna_pixmap_force_to_gpu(PixmapPtr pixmap);
+struct kgem_bo *sna_pixmap_change_tiling(PixmapPtr pixmap, uint32_t tiling);
 
 #define MOVE_WRITE 0x1
 #define MOVE_READ 0x2
diff --git a/src/sna/sna_accel.c b/src/sna/sna_accel.c
index 47665dd3..60bd39fc 100644
--- a/src/sna/sna_accel.c
+++ b/src/sna/sna_accel.c
@@ -322,7 +322,7 @@ static uint32_t sna_pixmap_choose_tiling(PixmapPtr pixmap, uint32_t tiling)
 				  pixmap->drawable.bitsPerPixel);
 }
 
-static bool sna_pixmap_change_tiling(PixmapPtr pixmap, uint32_t tiling)
+struct kgem_bo *sna_pixmap_change_tiling(PixmapPtr pixmap, uint32_t tiling)
 {
 	struct sna_pixmap *priv = sna_pixmap(pixmap);
 	struct sna *sna = to_sna_from_pixmap(pixmap);
@@ -335,7 +335,7 @@ static bool sna_pixmap_change_tiling(PixmapPtr pixmap, uint32_t tiling)
 
 	if (priv->pinned) {
 		DBG(("%s: can't convert pinned bo\n", __FUNCTION__));
-		return false;
+		return NULL;
 	}
 
 	bo = kgem_create_2d(&sna->kgem,
@@ -345,7 +345,7 @@ static bool sna_pixmap_change_tiling(PixmapPtr pixmap, uint32_t tiling)
 			    tiling, 0);
 	if (bo == NULL) {
 		DBG(("%s: allocation failed\n", __FUNCTION__));
-		return false;
+		return NULL;
 	}
 
 	box.x1 = box.y1 = 0;
@@ -362,8 +362,7 @@ static bool sna_pixmap_change_tiling(PixmapPtr pixmap, uint32_t tiling)
 	}
 
 	kgem_bo_destroy(&sna->kgem, priv->gpu_bo);
-	priv->gpu_bo = bo;
-	return true;
+	return priv->gpu_bo = bo;
 }
 
 static inline void sna_set_pixmap(PixmapPtr pixmap, struct sna_pixmap *sna)
@@ -7273,11 +7272,9 @@ sna_poly_fill_rect_stippled_blt(DrawablePtr drawable,
 		DBG(("%s: converting bo from Y-tiling\n", __FUNCTION__));
 		/* This is cheating, but only the gpu_bo can be tiled */
 		assert(bo == sna_pixmap(pixmap)->gpu_bo);
-
-		if (!sna_pixmap_change_tiling(pixmap, I915_TILING_X))
+		bo = sna_pixmap_change_tiling(pixmap, I915_TILING_X);
+		if (bo == NULL)
 			return false;
-
-		bo = sna_pixmap(pixmap)->gpu_bo;
 	}
 
 	if (!sna_drawable_move_to_cpu(&stipple->drawable, MOVE_READ))
@@ -7589,11 +7586,11 @@ sna_glyph_blt(DrawablePtr drawable, GCPtr gc,
 	bo = priv->gpu_bo;
 	if (bo->tiling == I915_TILING_Y) {
 		DBG(("%s: converting bo from Y-tiling\n", __FUNCTION__));
-		if (!sna_pixmap_change_tiling(pixmap, I915_TILING_X)) {
+		bo = sna_pixmap_change_tiling(pixmap, I915_TILING_X);
+		if (bo == NULL) {
 			DBG(("%s -- fallback, dst uses Y-tiling\n", __FUNCTION__));
 			return false;
 		}
-		bo = priv->gpu_bo;
 	}
 
 	get_drawable_deltas(drawable, pixmap, &dx, &dy);
diff --git a/src/sna/sna_blt.c b/src/sna/sna_blt.c
index a488102c..02ddf3ff 100644
--- a/src/sna/sna_blt.c
+++ b/src/sna/sna_blt.c
@@ -1843,3 +1843,117 @@ Bool sna_blt_copy_boxes(struct sna *sna, uint8_t alu,
 
 	return TRUE;
 }
+
+static void box_extents(const BoxRec *box, int n, BoxRec *extents)
+{
+	*extents = *box;
+	while (--n) {
+		box++;
+		if (box->x1 < extents->x1)
+			extents->x1 = box->x1;
+		if (box->y1 < extents->y1)
+			extents->y1 = box->y1;
+
+		if (box->x2 > extents->x2)
+			extents->x2 = box->x2;
+		if (box->y2 > extents->y2)
+			extents->y2 = box->y2;
+	}
+}
+
+Bool sna_blt_copy_boxes_fallback(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 nbox)
+{
+	struct kgem_bo *free_bo = NULL;
+	Bool ret;
+
+	DBG(("%s: alu=%d, n=%d\n", __FUNCTION__, alu, nbox));
+
+	if (!sna_blt_compare_depth(&src->drawable, &dst->drawable)) {
+		DBG(("%s: mismatching depths %d -> %d\n",
+		     __FUNCTION__, src->drawable.depth, dst->drawable.depth));
+		return FALSE;
+	}
+
+	if (src_bo == dst_bo) {
+		DBG(("%s: dst == src\n", __FUNCTION__));
+
+		if (src_bo->tiling == I915_TILING_Y) {
+			struct kgem_bo *bo;
+
+			DBG(("%s: src is Y-tiled\n", __FUNCTION__));
+
+			assert(src_bo == sna_pixmap(src)->gpu_bo);
+			bo = sna_pixmap_change_tiling(src, I915_TILING_X);
+			if (bo == NULL) {
+				BoxRec extents;
+
+				DBG(("%s: y-tiling conversion failed\n",
+				     __FUNCTION__));
+
+				box_extents(box, nbox, &extents);
+				free_bo = kgem_create_2d(&sna->kgem,
+							 extents.x2 - extents.x1,
+							 extents.y2 - extents.y1,
+							 src->drawable.bitsPerPixel,
+							 I915_TILING_X, 0);
+				if (free_bo == NULL) {
+					DBG(("%s: fallback -- temp allocation failed\n",
+					     __FUNCTION__));
+					return FALSE;
+				}
+
+				if (!sna_blt_copy_boxes(sna, GXcopy,
+							src_bo, src_dx, src_dy,
+							free_bo, -extents.x1, -extents.y1,
+							src->drawable.bitsPerPixel,
+							box, nbox)) {
+					DBG(("%s: fallback -- temp copy failed\n",
+					     __FUNCTION__));
+					kgem_bo_destroy(&sna->kgem, free_bo);
+					return FALSE;
+				}
+
+				src_dx = -extents.x1;
+				src_dy = -extents.y1;
+				src_bo = free_bo;
+			} else
+				dst_bo = src_bo = bo;
+		}
+	} else {
+		if (src_bo->tiling == I915_TILING_Y) {
+			DBG(("%s: src is y-tiled\n", __FUNCTION__));
+			assert(src_bo == sna_pixmap(src)->gpu_bo);
+			src_bo = sna_pixmap_change_tiling(src, I915_TILING_X);
+			if (src_bo == NULL) {
+				DBG(("%s: fallback -- src y-tiling conversion failed\n",
+				     __FUNCTION__));
+				return FALSE;
+			}
+		}
+
+		if (dst_bo->tiling == I915_TILING_Y) {
+			DBG(("%s: dst is y-tiled\n", __FUNCTION__));
+			assert(dst_bo == sna_pixmap(dst)->gpu_bo);
+			dst_bo = sna_pixmap_change_tiling(dst, I915_TILING_X);
+			if (dst_bo == NULL) {
+				DBG(("%s: fallback -- dst y-tiling conversion failed\n",
+				     __FUNCTION__));
+				return FALSE;
+			}
+		}
+	}
+
+	ret =  sna_blt_copy_boxes(sna, alu,
+				  src_bo, src_dx, src_dy,
+				  dst_bo, dst_dx, dst_dy,
+				  dst->drawable.bitsPerPixel,
+				  box, nbox);
+
+	if (free_bo)
+		kgem_bo_destroy(&sna->kgem, free_bo);
+
+	return ret;
+}
diff --git a/src/sna/sna_render.h b/src/sna/sna_render.h
index 95720fa2..bebbed26 100644
--- a/src/sna/sna_render.h
+++ b/src/sna/sna_render.h
@@ -522,6 +522,10 @@ Bool sna_blt_copy_boxes(struct sna *sna, uint8_t alu,
 			struct kgem_bo *dst_bo, int16_t dst_dx, int16_t dst_dy,
 			int bpp,
 			const BoxRec *box, int n);
+Bool sna_blt_copy_boxes_fallback(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 nbox);
 
 Bool sna_get_pixel_from_rgba(uint32_t *pixel,
 			     uint16_t red,