sna/gen5+: Prefer using the BLT when either src or dst is untiled

The cost of the TLB miss on every sample far outweighs the impact of the
context (and ring) switch.

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

3 files changed, 41 insertions, 7 deletions

diff --git a/src/sna/gen5_render.c b/src/sna/gen5_render.c
index 507a947c..d4870b5f 100644
--- a/src/sna/gen5_render.c
+++ b/src/sna/gen5_render.c
@@ -2401,6 +2401,15 @@ gen5_copy_bind_surfaces(struct sna *sna,
 	gen5_emit_state(sna, op, offset);
 }
 
+static inline bool prefer_blt_copy(struct sna *sna,
+				   struct kgem_bo *src_bo,
+				   struct kgem_bo *dst_bo)
+{
+	return (src_bo->tiling == I915_TILING_NONE ||
+		dst_bo->tiling == I915_TILING_NONE ||
+		sna->kgem.mode == KGEM_BLT);
+}
+
 static Bool
 gen5_render_copy_boxes(struct sna *sna, uint8_t alu,
 		       PixmapPtr src, struct kgem_bo *src_bo, int16_t src_dx, int16_t src_dy,
@@ -2409,7 +2418,7 @@ gen5_render_copy_boxes(struct sna *sna, uint8_t alu,
 {
 	struct sna_composite_op tmp;
 
-	if (sna->kgem.mode == KGEM_BLT &&
+	if (prefer_blt_copy(sna, src_bo, dst_bo) &&
 	    sna_blt_compare_depth(&src->drawable, &dst->drawable) &&
 	    sna_blt_copy_boxes(sna, alu,
 			       src_bo, src_dx, src_dy,
@@ -2559,7 +2568,7 @@ gen5_render_copy(struct sna *sna, uint8_t alu,
 {
 	DBG(("%s (alu=%d)\n", __FUNCTION__, alu));
 
-	if (sna->kgem.mode == KGEM_BLT &&
+	if (prefer_blt_copy(sna, src_bo, dst_bo) &&
 	    sna_blt_compare_depth(&src->drawable, &dst->drawable) &&
 	    sna_blt_copy(sna, alu,
 			 src_bo, dst_bo,
@@ -2604,8 +2613,15 @@ gen5_render_copy(struct sna *sna, uint8_t alu,
 	if (!kgem_check_bo(&sna->kgem, dst_bo, src_bo, NULL))
 		kgem_submit(&sna->kgem);
 
-	if (kgem_bo_is_dirty(src_bo))
+	if (kgem_bo_is_dirty(src_bo)) {
+		if (sna_blt_compare_depth(&src->drawable, &dst->drawable) &&
+		    sna_blt_copy(sna, alu,
+				 src_bo, dst_bo,
+				 dst->drawable.bitsPerPixel,
+				 op))
+			return TRUE;
 		kgem_emit_flush(&sna->kgem);
+	}
 
 	gen5_copy_bind_surfaces(sna, &op->base);
 	gen5_align_vertex(sna, &op->base);
diff --git a/src/sna/gen6_render.c b/src/sna/gen6_render.c
index 164f69d5..6fbf6365 100644
--- a/src/sna/gen6_render.c
+++ b/src/sna/gen6_render.c
@@ -2581,6 +2581,15 @@ gen6_emit_copy_state(struct sna *sna,
 	gen6_emit_state(sna, op, offset);
 }
 
+static inline bool prefer_blt_copy(struct sna *sna,
+				   struct kgem_bo *src_bo,
+				   struct kgem_bo *dst_bo)
+{
+	return (src_bo->tiling == I915_TILING_NONE ||
+		dst_bo->tiling == I915_TILING_NONE ||
+		sna->kgem.ring == KGEM_BLT);
+}
+
 static Bool
 gen6_render_copy_boxes(struct sna *sna, uint8_t alu,
 		       PixmapPtr src, struct kgem_bo *src_bo, int16_t src_dx, int16_t src_dy,
@@ -2604,7 +2613,7 @@ gen6_render_copy_boxes(struct sna *sna, uint8_t alu,
 	     __FUNCTION__, src_dx, src_dy, dst_dx, dst_dy, n, alu,
 	     src_bo == dst_bo));
 
-	if (sna->kgem.ring == KGEM_BLT &&
+	if (prefer_blt_copy(sna, src_bo, dst_bo) &&
 	    sna_blt_compare_depth(&src->drawable, &dst->drawable) &&
 	    sna_blt_copy_boxes(sna, alu,
 			       src_bo, src_dx, src_dy,
@@ -2760,7 +2769,7 @@ gen6_render_copy(struct sna *sna, uint8_t alu,
 	     src->drawable.width, src->drawable.height,
 	     dst->drawable.width, dst->drawable.height));
 
-	if (sna->kgem.ring == KGEM_BLT &&
+	if (prefer_blt_copy(sna, src_bo, dst_bo) &&
 	    sna_blt_compare_depth(&src->drawable, &dst->drawable) &&
 	    sna_blt_copy(sna, alu,
 			 src_bo, dst_bo,
diff --git a/src/sna/gen7_render.c b/src/sna/gen7_render.c
index efe8bb53..1e614b85 100644
--- a/src/sna/gen7_render.c
+++ b/src/sna/gen7_render.c
@@ -2701,6 +2701,15 @@ gen7_emit_copy_state(struct sna *sna,
 	gen7_emit_state(sna, op, offset);
 }
 
+static inline bool prefer_blt_copy(struct sna *sna,
+				   struct kgem_bo *src_bo,
+				   struct kgem_bo *dst_bo)
+{
+	return (src_bo->tiling == I915_TILING_NONE ||
+		dst_bo->tiling == I915_TILING_NONE ||
+		sna->kgem.ring == KGEM_BLT);
+}
+
 static Bool
 gen7_render_copy_boxes(struct sna *sna, uint8_t alu,
 		       PixmapPtr src, struct kgem_bo *src_bo, int16_t src_dx, int16_t src_dy,
@@ -2724,7 +2733,7 @@ gen7_render_copy_boxes(struct sna *sna, uint8_t alu,
 	     __FUNCTION__, src_dx, src_dy, dst_dx, dst_dy, n, alu,
 	     src_bo == dst_bo));
 
-	if (sna->kgem.ring == KGEM_BLT &&
+	if (prefer_blt_copy(sna, src_bo, dst_bo) &&
 	    sna_blt_compare_depth(&src->drawable, &dst->drawable) &&
 	    sna_blt_copy_boxes(sna, alu,
 			       src_bo, src_dx, src_dy,
@@ -2880,7 +2889,7 @@ gen7_render_copy(struct sna *sna, uint8_t alu,
 	     src->drawable.width, src->drawable.height,
 	     dst->drawable.width, dst->drawable.height));
 
-	if (sna->kgem.ring == KGEM_BLT &&
+	if (prefer_blt_copy(sna, src_bo, dst_bo) &&
 	    sna_blt_compare_depth(&src->drawable, &dst->drawable) &&
 	    sna_blt_copy(sna, alu,
 			 src_bo, dst_bo,