sna/gen3+: Use nearest for unscaled videos

If the output is unscaled, then we do not require pixel interpolation
(and planar formats are exactly subsampled).

References: https://bugs.freedesktop.org/show_bug.cgi?id=58185
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>

5 files changed, 33 insertions, 15 deletions

diff --git a/src/sna/gen3_render.c b/src/sna/gen3_render.c
index b092976f..f203e7ec 100644
--- a/src/sna/gen3_render.c
+++ b/src/sna/gen3_render.c
@@ -3528,7 +3528,8 @@ gen3_emit_video_state(struct sna *sna,
 		      struct sna_video_frame *frame,
 		      PixmapPtr pixmap,
 		      struct kgem_bo *dst_bo,
-		      int width, int height)
+		      int width, int height,
+		      bool bilinear)
 {
 	struct gen3_render_state *state = &sna->render_state.gen3;
 	uint32_t id, ms3, rewind;
@@ -3887,6 +3888,7 @@ gen3_render_video(struct sna *sna,
 	float src_scale_x, src_scale_y;
 	int pix_xoff, pix_yoff;
 	struct kgem_bo *dst_bo;
+	bool bilinear;
 	int copy = 0;
 
 	DBG(("%s: %dx%d -> %dx%d\n", __FUNCTION__, src_w, src_h, drw_w, drw_h));
@@ -3927,6 +3929,8 @@ gen3_render_video(struct sna *sna,
 #endif
 	}
 
+	bilinear = src_w != drw_w || src_h != drw_h;
+
 	src_scale_x = ((float)src_w / frame->width) / drw_w;
 	src_scale_y = ((float)src_h / frame->height) / drw_h;
 
@@ -3936,13 +3940,13 @@ gen3_render_video(struct sna *sna,
 
 	gen3_video_get_batch(sna, dst_bo);
 	gen3_emit_video_state(sna, video, frame, pixmap,
-			      dst_bo, width, height);
+			      dst_bo, width, height, bilinear);
 	do {
 		int nbox_this_time = gen3_get_inline_rectangles(sna, nbox, 4);
 		if (nbox_this_time == 0) {
 			gen3_video_get_batch(sna, dst_bo);
 			gen3_emit_video_state(sna, video, frame, pixmap,
-					      dst_bo, width, height);
+					      dst_bo, width, height, bilinear);
 			nbox_this_time = gen3_get_inline_rectangles(sna, nbox, 4);
 		}
 		nbox -= nbox_this_time;
diff --git a/src/sna/gen4_render.c b/src/sna/gen4_render.c
index 75b07a3c..f80fa835 100644
--- a/src/sna/gen4_render.c
+++ b/src/sna/gen4_render.c
@@ -1690,7 +1690,10 @@ gen4_render_video(struct sna *sna,
 	tmp.dst.format = sna_format_for_depth(pixmap->drawable.depth);
 	tmp.dst.bo = priv->gpu_bo;
 
-	tmp.src.filter = SAMPLER_FILTER_BILINEAR;
+	if (src_w == drw_w && src_h == drw_h)
+		tmp.src.filter = SAMPLER_FILTER_NEAREST;
+	else
+		tmp.src.filter = SAMPLER_FILTER_BILINEAR;
 	tmp.src.repeat = SAMPLER_EXTEND_PAD;
 	tmp.src.bo = frame->bo;
 	tmp.mask.bo = NULL;
diff --git a/src/sna/gen5_render.c b/src/sna/gen5_render.c
index bff5066b..6e6389a1 100644
--- a/src/sna/gen5_render.c
+++ b/src/sna/gen5_render.c
@@ -1688,7 +1688,10 @@ gen5_render_video(struct sna *sna,
 	tmp.dst.format = sna_format_for_depth(pixmap->drawable.depth);
 	tmp.dst.bo = priv->gpu_bo;
 
-	tmp.src.filter = SAMPLER_FILTER_BILINEAR;
+	if (src_w == drw_w && src_h == drw_h)
+		tmp.src.filter = SAMPLER_FILTER_NEAREST;
+	else
+		tmp.src.filter = SAMPLER_FILTER_BILINEAR;
 	tmp.src.repeat = SAMPLER_EXTEND_PAD;
 	tmp.src.bo = frame->bo;
 	tmp.mask.bo = NULL;
diff --git a/src/sna/gen6_render.c b/src/sna/gen6_render.c
index 64830925..801236f0 100644
--- a/src/sna/gen6_render.c
+++ b/src/sna/gen6_render.c
@@ -186,10 +186,6 @@ static const struct blendinfo {
 #define FILL_FLAGS(op, format) GEN6_SET_FLAGS(FILL_SAMPLER, gen6_get_blend((op), false, (format)), GEN6_WM_KERNEL_NOMASK, FILL_VERTEX)
 #define FILL_FLAGS_NOBLEND GEN6_SET_FLAGS(FILL_SAMPLER, NO_BLEND, GEN6_WM_KERNEL_NOMASK, FILL_VERTEX)
 
-#define VIDEO_SAMPLER \
-	SAMPLER_OFFSET(SAMPLER_FILTER_BILINEAR, SAMPLER_EXTEND_PAD, \
-		       SAMPLER_FILTER_NEAREST, SAMPLER_EXTEND_NONE)
-
 #define GEN6_SAMPLER(f) (((f) >> 16) & 0xfff0)
 #define GEN6_BLEND(f) (((f) >> 0) & 0xfff0)
 #define GEN6_KERNEL(f) (((f) >> 16) & 0xf)
@@ -1986,6 +1982,7 @@ gen6_render_video(struct sna *sna,
 	int nbox, dxo, dyo, pix_xoff, pix_yoff;
 	float src_scale_x, src_scale_y;
 	struct sna_pixmap *priv;
+	unsigned filter;
 	BoxPtr box;
 
 	DBG(("%s: src=(%d, %d), dst=(%d, %d), %dx[(%d, %d), (%d, %d)...]\n",
@@ -2014,8 +2011,15 @@ gen6_render_video(struct sna *sna,
 	tmp.floats_per_vertex = 3;
 	tmp.floats_per_rect = 9;
 
+	if (src_w == drw_w && src_h == drw_h)
+		filter = SAMPLER_FILTER_NEAREST;
+	else
+		filter = SAMPLER_FILTER_BILINEAR;
+
 	tmp.u.gen6.flags =
-		GEN6_SET_FLAGS(VIDEO_SAMPLER, NO_BLEND,
+		GEN6_SET_FLAGS(SAMPLER_OFFSET(filter, SAMPLER_EXTEND_PAD,
+					       SAMPLER_FILTER_NEAREST, SAMPLER_EXTEND_NONE),
+			       NO_BLEND,
 			       is_planar_fourcc(frame->id) ?
 			       GEN6_WM_KERNEL_VIDEO_PLANAR :
 			       GEN6_WM_KERNEL_VIDEO_PACKED,
diff --git a/src/sna/gen7_render.c b/src/sna/gen7_render.c
index fea5a101..270488cb 100644
--- a/src/sna/gen7_render.c
+++ b/src/sna/gen7_render.c
@@ -209,10 +209,6 @@ static const struct blendinfo {
 #define FILL_FLAGS(op, format) GEN7_SET_FLAGS(FILL_SAMPLER, gen7_get_blend((op), false, (format)), GEN7_WM_KERNEL_NOMASK, FILL_VERTEX)
 #define FILL_FLAGS_NOBLEND GEN7_SET_FLAGS(FILL_SAMPLER, NO_BLEND, GEN7_WM_KERNEL_NOMASK, FILL_VERTEX)
 
-#define VIDEO_SAMPLER \
-	SAMPLER_OFFSET(SAMPLER_FILTER_BILINEAR, SAMPLER_EXTEND_PAD, \
-		       SAMPLER_FILTER_NEAREST, SAMPLER_EXTEND_NONE)
-
 #define GEN7_SAMPLER(f) (((f) >> 16) & 0xfff0)
 #define GEN7_BLEND(f) (((f) >> 0) & 0x7ff0)
 #define GEN7_READS_DST(f) (((f) >> 15) & 1)
@@ -2098,6 +2094,7 @@ gen7_render_video(struct sna *sna,
 	int nbox, dxo, dyo, pix_xoff, pix_yoff;
 	float src_scale_x, src_scale_y;
 	struct sna_pixmap *priv;
+	unsigned filter;
 	BoxPtr box;
 
 	DBG(("%s: src=(%d, %d), dst=(%d, %d), %dx[(%d, %d), (%d, %d)...]\n",
@@ -2126,8 +2123,15 @@ gen7_render_video(struct sna *sna,
 	tmp.floats_per_vertex = 3;
 	tmp.floats_per_rect = 9;
 
+	if (src_w == drw_w && src_h == drw_h)
+		filter = SAMPLER_FILTER_NEAREST;
+	else
+		filter = SAMPLER_FILTER_BILINEAR;
+
 	tmp.u.gen7.flags =
-		GEN7_SET_FLAGS(VIDEO_SAMPLER, NO_BLEND,
+		GEN7_SET_FLAGS(SAMPLER_OFFSET(filter, SAMPLER_EXTEND_PAD,
+					      SAMPLER_FILTER_NEAREST, SAMPLER_EXTEND_NONE),
+			       NO_BLEND,
 			       is_planar_fourcc(frame->id) ?
 			       GEN7_WM_KERNEL_VIDEO_PLANAR :
 			       GEN7_WM_KERNEL_VIDEO_PACKED,