xf86-video-intel: 4 commits - src/sna/blt.c src/sna/gen6_common.h src/sna/gen9_render.c src/sna/sna_display.c

[Chris Wilson]

 src/sna/blt.c         |   22 ++++++++++++----------
 src/sna/gen6_common.h |    8 ++++----
 src/sna/gen9_render.c |    2 +-
 src/sna/sna_display.c |   10 ++++++++--
 4 files changed, 25 insertions(+), 17 deletions(-)

New commits:
commit c3dc8310575acdbbef4565f53644384c7ce4a6f5
Author: Chris Wilson <chris at chris-wilson.co.uk>
Date:   Fri Apr 15 19:34:15 2016 +0100

    sna: Mark the transformed cursor image as dirty
    
    So that when we size from transformed cursors to non-transformed, we
    remember to clear the entire area.
    
    Signed-off-by: Chris Wilson <chris at chris-wilson.co.uk>

diff --git a/src/sna/sna_display.c b/src/sna/sna_display.c
index ff80a4e..3b5ae87 100644
--- a/src/sna/sna_display.c
+++ b/src/sna/sna_display.c
@@ -5684,8 +5684,14 @@ static struct sna_cursor *__sna_get_cursor(struct sna *sna, xf86CrtcPtr crtc)
 	cursor->rotation = rotation;
 	cursor->transformed = transformed;
 	cursor->serial = sna->cursor.serial;
-	cursor->last_width = width;
-	cursor->last_height = height;
+	if (transformed) {
+		/* mark the transformed rectangle as dirty, not input */
+		cursor->last_width = size;
+		cursor->last_height = size;
+	} else {
+		cursor->last_width = width;
+		cursor->last_height = height;
+	}
 	return cursor;
 }
 
commit d30d276aeebe666e2c71a275bb2c71781c75448b
Author: Chris Wilson <chris at chris-wilson.co.uk>
Date:   Fri Apr 15 19:26:34 2016 +0100

    sna/blt: Reuse computed partial tile offset in copy_from_tiled
    
    Signed-off-by: Chris Wilson <chris at chris-wilson.co.uk>

diff --git a/src/sna/blt.c b/src/sna/blt.c
index e82647e..c1ac166 100644
--- a/src/sna/blt.c
+++ b/src/sna/blt.c
@@ -721,13 +721,16 @@ memcpy_from_tiled_x__swizzle_0__sse2(const void *src, void *dst, int bpp,
 	width *= cpp;
 	assert(dst_stride >= width);
 	if (src_x & tile_mask) {
-		dst_stride -= width;
 		offset_x = (src_x & tile_mask) * cpp;
 		length_x = min(tile_width - offset_x, width);
+		dst_stride -= width;
 		dst_stride += (width - length_x) & 15;
-	} else
+	} else {
+		offset_x = 0;
 		dst_stride -= width & ~15;
+	}
 	assert(dst_stride >= 0);
+	src_x = (src_x >> tile_shift) * tile_size;
 
 	while (height--) {
 		unsigned w = width;
@@ -735,17 +738,16 @@ memcpy_from_tiled_x__swizzle_0__sse2(const void *src, void *dst, int bpp,
 
 		tile_row += src_y / tile_height * src_stride * tile_height;
 		tile_row += (src_y & (tile_height-1)) * tile_width;
+		tile_row += src_x;
 		src_y++;
 
-		if (src_x) {
-			tile_row += (src_x >> tile_shift) * tile_size;
-			if (src_x & tile_mask) {
-				memcpy(dst, tile_row + offset_x, length_x);
-				tile_row += tile_size;
-				dst = (uint8_t *)dst + length_x;
-				w -= length_x;
-			}
+		if (offset_x) {
+			memcpy(dst, tile_row + offset_x, length_x);
+			tile_row += tile_size;
+			dst = (uint8_t *)dst + length_x;
+			w -= length_x;
 		}
+
 		if ((uintptr_t)dst & 15) {
 			while (w >= tile_width) {
 				from_sse128xNu(dst,
commit bb69256b5270b6ff64a87db4eb308f1cc856679c
Author: Chris Wilson <chris at chris-wilson.co.uk>
Date:   Fri Apr 15 15:16:00 2016 +0100

    sna/gen6: Encourage migration of small BLT operations
    
    Signed-off-by: Chris Wilson <chris at chris-wilson.co.uk>

diff --git a/src/sna/gen6_common.h b/src/sna/gen6_common.h
index 898cda8..83adc7d 100644
--- a/src/sna/gen6_common.h
+++ b/src/sna/gen6_common.h
@@ -46,9 +46,6 @@ inline static bool can_switch_to_blt(struct sna *sna,
 	if (sna->kgem.ring != KGEM_RENDER)
 		return true;
 
-	if (NO_RING_SWITCH(sna))
-		return false;
-
 	if (bo && RQ_IS_BLT(bo->rq))
 		return true;
 
@@ -61,6 +58,9 @@ inline static bool can_switch_to_blt(struct sna *sna,
 	if (bo && RQ_IS_RENDER(bo->rq))
 		return false;
 
+	if (NO_RING_SWITCH(sna))
+		return false;
+
 	if (flags & COPY_LAST)
 		return true;
 
@@ -207,7 +207,7 @@ prefer_blt_fill(struct sna *sna, struct kgem_bo *bo, unsigned flags)
 		if (!prefer_blt_ring(sna, bo, 0))
 			return false;
 	} else {
-	    if (can_switch_to_blt(sna, bo, 0))
+	    if (can_switch_to_blt(sna, bo, COPY_LAST))
 		    return true;
 	}
 
commit 0d38419cbeaae5f8b5260e4d4dd9cfa0d2694eae
Author: Chris Wilson <chris at chris-wilson.co.uk>
Date:   Fri Apr 15 14:58:57 2016 +0100

    sna/gen9: Update mocs selection
    
    Since the choice is now boolean (use PTE or use WB), remove the third
    uncached condition.
    
    Signed-off-by: Chris Wilson <chris at chris-wilson.co.uk>

diff --git a/src/sna/gen9_render.c b/src/sna/gen9_render.c
index 889cded..3c1f07a 100644
--- a/src/sna/gen9_render.c
+++ b/src/sna/gen9_render.c
@@ -1427,7 +1427,7 @@ gen9_bind_bo(struct sna *sna,
 		domains = I915_GEM_DOMAIN_RENDER << 16 |I915_GEM_DOMAIN_RENDER;
 	} else
 		domains = I915_GEM_DOMAIN_SAMPLER << 16;
-	ss[1] = (is_dst && is_uncached(sna, bo)) ? 0 : is_scanout ? MOCS_PTE << 24 : MOCS_WB << 24;
+	ss[1] = (is_scanout || (is_dst && is_uncached(sna, bo))) ? MOCS_PTE << 24 : MOCS_WB << 24;
 	ss[2] = ((width - 1)  << SURFACE_WIDTH_SHIFT |
 		 (height - 1) << SURFACE_HEIGHT_SHIFT);
 	ss[3] = (bo->pitch - 1) << SURFACE_PITCH_SHIFT;