sna/gen3: Prefer to use the BLT to clear the scratch glyph pixmaps

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

3 files changed, 50 insertions, 40 deletions

diff --git a/src/sna/gen3_render.c b/src/sna/gen3_render.c
index bae65051..4dd027fb 100644
--- a/src/sna/gen3_render.c
+++ b/src/sna/gen3_render.c
@@ -49,8 +49,11 @@
 #define NO_COPY 0
 #define NO_COPY_BOXES 0
 #define NO_FILL 0
+#define NO_FILL_ONE 0
 #define NO_FILL_BOXES 0
 
+#define PREFER_BLT_FILL 1
+
 enum {
 	SHADER_NONE = 0,
 	SHADER_ZERO,
@@ -2966,10 +2969,9 @@ gen3_emit_video_state(struct sna *sna,
 		  S2_TEXCOORD_FMT(5, TEXCOORDFMT_NOT_PRESENT) |
 		  S2_TEXCOORD_FMT(6, TEXCOORDFMT_NOT_PRESENT) |
 		  S2_TEXCOORD_FMT(7, TEXCOORDFMT_NOT_PRESENT));
-	OUT_BATCH((2 << S6_DEPTH_TEST_FUNC_SHIFT) |
-		  (2 << S6_CBUF_SRC_BLEND_FACT_SHIFT) |
+	OUT_BATCH((2 << S6_CBUF_SRC_BLEND_FACT_SHIFT) |
 		  (1 << S6_CBUF_DST_BLEND_FACT_SHIFT) |
-		  S6_COLOR_WRITE_ENABLE | (2 << S6_TRISTRIP_PV_SHIFT));
+		  S6_COLOR_WRITE_ENABLE);
 
 	sna->render_state.gen3.last_blend = 0;
 	sna->render_state.gen3.last_sampler = 0;
@@ -3663,6 +3665,15 @@ gen3_render_fill_boxes_try_blt(struct sna *sna,
 				  pixel, box, n);
 }
 
+static inline Bool prefer_fill_blt(struct sna *sna)
+{
+#if PREFER_BLT_FILL
+	return true;
+#else
+	return sna->kgem.mode != KGEM_RENDER;
+#endif
+}
+
 static Bool
 gen3_render_fill_boxes(struct sna *sna,
 		       CARD8 op,
@@ -3698,7 +3709,7 @@ gen3_render_fill_boxes(struct sna *sna,
 						      dst, dst_bo,
 						      box, n);
 
-	if (sna->kgem.mode != KGEM_RENDER &&
+	if (prefer_fill_blt(sna) &&
 	    gen3_render_fill_boxes_try_blt(sna, op, format, color,
 					   dst, dst_bo,
 					   box, n))
@@ -3803,7 +3814,7 @@ gen3_render_fill(struct sna *sna, uint8_t alu,
 #endif
 
 	/* Prefer to use the BLT if already engaged */
-	if (sna->kgem.mode != KGEM_RENDER &&
+	if (prefer_fill_blt(sna) &&
 	    sna_blt_fill(sna, alu,
 			 dst_bo, dst->drawable.bitsPerPixel,
 			 color,
@@ -3875,13 +3886,13 @@ gen3_render_fill_one(struct sna *sna, PixmapPtr dst, struct kgem_bo *bo,
 {
 	struct sna_composite_op tmp;
 
-#if NO_FILL_BOXES
+#if NO_FILL_ONE
 	return gen3_render_fill_one_try_blt(sna, dst, bo, color,
 					    x1, y1, x2, y2, alu);
 #endif
 
 	/* Prefer to use the BLT if already engaged */
-	if (sna->kgem.mode != KGEM_RENDER &&
+	if (prefer_fill_blt(sna) &&
 	    gen3_render_fill_one_try_blt(sna, dst, bo, color,
 					 x1, y1, x2, y2, alu))
 		return TRUE;
@@ -3905,10 +3916,17 @@ gen3_render_fill_one(struct sna *sna, PixmapPtr dst, struct kgem_bo *bo,
 	tmp.floats_per_vertex = 2;
 	tmp.floats_per_rect = 6;
 
-	tmp.src.u.gen3.type = SHADER_CONSTANT;
-	tmp.src.u.gen3.mode =
-		sna_rgba_for_color(color, dst->drawable.depth);
-	tmp.mask.u.gen3.mode = SHADER_NONE;
+	color = sna_rgba_for_color(color, dst->drawable.depth);
+	if (color == 0)
+		tmp.src.u.gen3.type = SHADER_ZERO;
+	else if (color == 0xff000000)
+		tmp.src.u.gen3.type = SHADER_BLACK;
+	else if (color == 0xffffffff)
+		tmp.src.u.gen3.type = SHADER_WHITE;
+	else
+		tmp.src.u.gen3.type = SHADER_CONSTANT;
+	tmp.src.u.gen3.mode = color;
+	tmp.mask.u.gen3.type = SHADER_NONE;
 	tmp.u.gen3.num_constants = 0;
 
 	if (!kgem_check_bo(&sna->kgem, bo, NULL)) {
diff --git a/src/sna/kgem.c b/src/sna/kgem.c
index a11e9553..4d876dbf 100644
--- a/src/sna/kgem.c
+++ b/src/sna/kgem.c
@@ -1313,30 +1313,29 @@ int kgem_choose_tiling(struct kgem *kgem, int tiling, int width, int height, int
 	if (tiling < 0)
 		return tiling;
 
-	if (tiling && width * bpp > 8 * 4096) {
+	if (tiling == I915_TILING_Y && height <= 16) {
+		DBG(("%s: too short [%d] for TILING_Y\n",
+		     __FUNCTION__,height));
+		tiling = I915_TILING_X;
+	}
+	if (tiling && width * bpp >= 8 * 4096) {
 		DBG(("%s: TLB miss between lines %dx%d (pitch=%d), forcing tiling %d\n",
 		     __FUNCTION__,
 		     width, height, width*bpp/8,
 		     tiling));
 		return -tiling;
 	}
+	if (tiling == I915_TILING_X && height < 4) {
+		DBG(("%s: too short [%d] for TILING_X\n",
+		     __FUNCTION__, height));
+		tiling = I915_TILING_NONE;
+	}
 
 	/* Before the G33, we only have a small GTT to play with and tiled
 	 * surfaces always require full fence regions and so cause excessive
 	 * aperture thrashing.
 	 */
 	if (kgem->gen < 33) {
-		if (tiling == I915_TILING_Y && height < 16) {
-			DBG(("%s: too short [%d] for TILING_Y\n",
-			     __FUNCTION__,height));
-			tiling = I915_TILING_X;
-		}
-		if (tiling == I915_TILING_X && height < 4) {
-			DBG(("%s: too short [%d] for TILING_X\n",
-			     __FUNCTION__, height));
-			tiling = I915_TILING_NONE;
-		}
-
 		if (tiling == I915_TILING_X && width * bpp < 8*512/2) {
 			DBG(("%s: too thin [%d] for TILING_X\n",
 			     __FUNCTION__, width));
diff --git a/src/sna/sna_glyphs.c b/src/sna/sna_glyphs.c
index 4db5bafb..13b0ccaf 100644
--- a/src/sna/sna_glyphs.c
+++ b/src/sna/sna_glyphs.c
@@ -651,19 +651,16 @@ next_glyph:
 	return TRUE;
 }
 
-static Bool
-clear_pixmap(struct sna *sna, PixmapPtr pixmap, PictFormat format)
+static void
+clear_pixmap(struct sna *sna, PixmapPtr pixmap)
 {
-	BoxRec box;
-	xRenderColor color = { 0 };
-
-	box.x1 = box.y1 = 0;
-	box.x2 = pixmap->drawable.width;
-	box.y2 = pixmap->drawable.height;
-
-	return sna->render.fill_boxes(sna, PictOpClear, format, &color,
-				      pixmap, sna_pixmap_get_bo(pixmap),
-				      &box, 1);
+	struct sna_pixmap *priv = sna_pixmap(pixmap);
+	assert(priv->gpu_only);
+	sna->render.fill_one(sna, pixmap, priv->gpu_bo, 0,
+			     0, 0,
+			     pixmap->drawable.width,
+			     pixmap->drawable.height,
+			     GXclear);
 }
 
 static Bool
@@ -742,11 +739,7 @@ glyphs_via_mask(struct sna *sna,
 		return FALSE;
 
 	ValidatePicture(mask);
-
-	if (!clear_pixmap(sna, pixmap, mask->format)) {
-		FreePicture(mask, 0);
-		return FALSE;
-	}
+	clear_pixmap(sna, pixmap);
 
 	memset(&tmp, 0, sizeof(tmp));
 	glyph_atlas = NULL;