sna/gen[23]: Fixup render targets with pitches below hw minimum

gen2/3 have a restriction that the 3D pipeline cannot render to a pixmap
with a pitch less than 8/16 respectively. Rather than mandating all
pixmaps to be created with a stride greater than 16, fixup the bo for
the rare occasions when it is necessary.

Reported-by: Paul Neumann <paul104x@yahoo.de>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=43688
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>

4 files changed, 131 insertions, 6 deletions

diff --git a/src/sna/gen2_render.c b/src/sna/gen2_render.c
index d35f2da5..cd2dcf42 100644
--- a/src/sna/gen2_render.c
+++ b/src/sna/gen2_render.c
@@ -1186,7 +1186,8 @@ gen2_composite_picture(struct sna *sna,
 }
 
 static Bool
-gen2_composite_set_target(struct sna_composite_op *op,
+gen2_composite_set_target(struct sna *sna,
+			  struct sna_composite_op *op,
 			  PicturePtr dst)
 {
 	struct sna_pixmap *priv;
@@ -1200,6 +1201,22 @@ gen2_composite_set_target(struct sna_composite_op *op,
 	if (priv == NULL)
 		return FALSE;
 
+	if (priv->gpu_bo->pitch < 8) {
+		struct kgem_bo *bo;
+
+		if (priv->pinned)
+			return FALSE;
+
+		bo = kgem_replace_bo(&sna->kgem, priv->gpu_bo,
+				     op->dst.width, op->dst.height, 16,
+				     op->dst.pixmap->drawable.bitsPerPixel);
+		if (bo == NULL)
+			return FALSE;
+
+		kgem_bo_destroy(&sna->kgem, priv->gpu_bo);
+		priv->gpu_bo = bo;
+	}
+
 	op->dst.bo = priv->gpu_bo;
 	op->damage = &priv->gpu_damage;
 	if (sna_damage_is_all(&priv->gpu_damage, op->dst.width, op->dst.height))
@@ -1396,7 +1413,7 @@ gen2_render_composite(struct sna *sna,
 					    width,  height,
 					    tmp);
 
-	if (!gen2_composite_set_target(tmp, dst)) {
+	if (!gen2_composite_set_target(sna, tmp, dst)) {
 		DBG(("%s: unable to set render target\n",
 		     __FUNCTION__));
 		return FALSE;
@@ -1822,7 +1839,7 @@ gen2_render_composite_spans(struct sna *sna,
 	if (need_tiling(sna, width, height))
 		return FALSE;
 
-	if (!gen2_composite_set_target(&tmp->base, dst)) {
+	if (!gen2_composite_set_target(sna, &tmp->base, dst)) {
 		DBG(("%s: unable to set render target\n",
 		     __FUNCTION__));
 		return FALSE;
diff --git a/src/sna/gen3_render.c b/src/sna/gen3_render.c
index ea9d40ab..33bdfd24 100644
--- a/src/sna/gen3_render.c
+++ b/src/sna/gen3_render.c
@@ -2175,7 +2175,9 @@ gen3_align_vertex(struct sna *sna,
 }
 
 static Bool
-gen3_composite_set_target(struct sna_composite_op *op, PicturePtr dst)
+gen3_composite_set_target(struct sna *sna,
+			  struct sna_composite_op *op,
+			  PicturePtr dst)
 {
 	struct sna_pixmap *priv;
 
@@ -2189,6 +2191,22 @@ gen3_composite_set_target(struct sna_composite_op *op, PicturePtr dst)
 	if (priv == NULL)
 		return FALSE;
 
+	if (priv->gpu_bo->pitch < 16) {
+		struct kgem_bo *bo;
+
+		if (priv->pinned)
+			return FALSE;
+
+		bo = kgem_replace_bo(&sna->kgem, priv->gpu_bo,
+				     op->dst.width, op->dst.height, 16,
+				     op->dst.pixmap->drawable.bitsPerPixel);
+		if (bo == NULL)
+			return FALSE;
+
+		kgem_bo_destroy(&sna->kgem, priv->gpu_bo);
+		priv->gpu_bo = bo;
+	}
+
 	op->dst.bo = priv->gpu_bo;
 	op->damage = &priv->gpu_damage;
 	if (sna_damage_is_all(op->damage, op->dst.width, op->dst.height))
@@ -2380,7 +2398,7 @@ gen3_render_composite(struct sna *sna,
 
 	memset(&tmp->u.gen3, 0, sizeof(tmp->u.gen3));
 
-	if (!gen3_composite_set_target(tmp, dst)) {
+	if (!gen3_composite_set_target(sna, tmp, dst)) {
 		DBG(("%s: unable to set render target\n",
 		     __FUNCTION__));
 		return FALSE;
@@ -2913,7 +2931,7 @@ gen3_render_composite_spans(struct sna *sna,
 	if (need_tiling(sna, width, height))
 		return FALSE;
 
-	if (!gen3_composite_set_target(&tmp->base, dst)) {
+	if (!gen3_composite_set_target(sna, &tmp->base, dst)) {
 		DBG(("%s: unable to set render target\n",
 		     __FUNCTION__));
 		return FALSE;
diff --git a/src/sna/kgem.c b/src/sna/kgem.c
index 1c9da8ec..34ce7459 100644
--- a/src/sna/kgem.c
+++ b/src/sna/kgem.c
@@ -2356,3 +2356,85 @@ void kgem_bo_set_binding(struct kgem_bo *bo, uint32_t format, uint16_t offset)
 		bo->binding.next = b;
 	}
 }
+
+struct kgem_bo *
+kgem_replace_bo(struct kgem *kgem,
+		struct kgem_bo *src,
+		uint32_t width,
+		uint32_t height,
+		uint32_t pitch,
+		uint32_t bpp)
+{
+	struct kgem_bo *dst;
+	uint32_t br00, br13;
+	uint32_t handle;
+	uint32_t size;
+	uint32_t *b;
+
+	DBG(("%s: replacing bo handle=%d, size=%dx%d pitch=%d, with pitch=%d\n",
+	     __FUNCTION__, src->handle,  width, height, src->pitch, pitch));
+
+	/* We only expect to be called to fixup small buffers, hence why
+	 * we only attempt to allocate a linear bo.
+	 */
+	assert(src->tiling == I915_TILING_NONE);
+
+	size = height * pitch;
+
+	dst = search_linear_cache(kgem, size, true);
+	if (dst == NULL)
+		dst = search_linear_cache(kgem, size, false);
+	if (dst == NULL) {
+		handle = gem_create(kgem->fd, size);
+		if (handle == 0)
+			return NULL;
+
+		dst = __kgem_bo_alloc(handle, size);
+	}
+	dst->pitch = pitch;
+
+	kgem_set_mode(kgem, KGEM_BLT);
+	if (!kgem_check_batch(kgem, 8) ||
+	    !kgem_check_reloc(kgem, 2) ||
+	    !kgem_check_bo_fenced(kgem, src, dst, NULL)) {
+		_kgem_submit(kgem);
+		_kgem_set_mode(kgem, KGEM_BLT);
+	}
+
+	br00 = XY_SRC_COPY_BLT_CMD;
+	br13 = pitch;
+	pitch = src->pitch;
+	if (kgem->gen >= 40 && src->tiling) {
+		br00 |= BLT_SRC_TILED;
+		pitch >>= 2;
+	}
+
+	br13 |= 0xcc << 16;
+	switch (bpp) {
+	default:
+	case 32: br00 |= BLT_WRITE_ALPHA | BLT_WRITE_RGB;
+		 br13 |= 1 << 25; /* RGB8888 */
+	case 16: br13 |= 1 << 24; /* RGB565 */
+	case 8: break;
+	}
+
+	b = kgem->batch + kgem->nbatch;
+	b[0] = br00;
+	b[1] = br13;
+	b[2] = 0;
+	b[3] = height << 16 | width;
+	b[4] = kgem_add_reloc(kgem, kgem->nbatch + 4, dst,
+			      I915_GEM_DOMAIN_RENDER << 16 |
+			      I915_GEM_DOMAIN_RENDER |
+			      KGEM_RELOC_FENCED,
+			      0);
+	b[5] = 0;
+	b[6] = pitch;
+	b[7] = kgem_add_reloc(kgem, kgem->nbatch + 7, src,
+			      I915_GEM_DOMAIN_RENDER << 16 |
+			      KGEM_RELOC_FENCED,
+			      0);
+	kgem->nbatch += 8;
+
+	return dst;
+}
diff --git a/src/sna/kgem.h b/src/sna/kgem.h
index 14faa709..8649512e 100644
--- a/src/sna/kgem.h
+++ b/src/sna/kgem.h
@@ -170,6 +170,14 @@ int kgem_choose_tiling(struct kgem *kgem,
 		       int tiling, int width, int height, int bpp);
 bool kgem_can_create_2d(struct kgem *kgem,
 			int width, int height, int bpp, int tiling);
+
+struct kgem_bo *
+kgem_replace_bo(struct kgem *kgem,
+		struct kgem_bo *src,
+		uint32_t width,
+		uint32_t height,
+		uint32_t pitch,
+		uint32_t bpp);
 enum {
 	CREATE_EXACT = 0x1,
 	CREATE_INACTIVE = 0x2,