[PATCH 3/7] gpu: ipu-v3: image-convert: limit input seam position to hardware requirements

[Philipp Zabel]

Limit the input seam position to an interval that guarantees the tile
size does not exceed 1024 pixels after the IC downsizing section and
that space is left for the next tile.

Signed-off-by: Philipp Zabel <p.zabel at pengutronix.de>
---
 drivers/gpu/ipu-v3/ipu-image-convert.c | 30 ++++++++++++++++++++------
 1 file changed, 24 insertions(+), 6 deletions(-)

diff --git a/drivers/gpu/ipu-v3/ipu-image-convert.c b/drivers/gpu/ipu-v3/ipu-image-convert.c
index 4ac7377cfe5c..9f631a0c39ea 100644
--- a/drivers/gpu/ipu-v3/ipu-image-convert.c
+++ b/drivers/gpu/ipu-v3/ipu-image-convert.c
@@ -481,12 +481,23 @@ static void find_best_seam(struct ipu_image_convert_ctx *ctx,
 	unsigned int min_diff = UINT_MAX;
 	unsigned int out_start;
 	unsigned int out_end;
+	unsigned int in_start;
+	unsigned int in_end;
 
 	/* Start within 1024 pixels of the right / bottom edge */
 	out_start = max_t(int, index * out_align, out_edge - 1024);
 	/* End before having to add more columns to the left / rows above */
 	out_end = min_t(unsigned int, out_edge, index * 1024 + 1);
 
+	/*
+	 * Limit input seam position to make sure that the downsized input tile
+	 * to the right or bottom does not exceed 1024 pixels.
+	 */
+	in_start = max_t(int, index * in_align,
+			 in_edge - (1024 << downsize_coeff));
+	in_end = min_t(unsigned int, in_edge,
+		       index * (1024 << downsize_coeff) + 1);
+
 	/*
 	 * Output tiles must start at a multiple of 8 bytes horizontally and
 	 * possibly at an even line horizontally depending on the pixel format.
@@ -496,6 +507,7 @@ static void find_best_seam(struct ipu_image_convert_ctx *ctx,
 	for (out_pos = out_start; out_pos < out_end; out_pos += out_align) {
 		unsigned int in_pos;
 		unsigned int in_pos_aligned;
+		unsigned int in_pos_rounded;
 		unsigned int abs_diff;
 
 		/*
@@ -516,9 +528,16 @@ static void find_best_seam(struct ipu_image_convert_ctx *ctx,
 		 * start the input tile at, 19.13 fixed point.
 		 */
 		in_pos_aligned = round_closest(in_pos, 8192U * in_align);
+		/* Convert 19.13 fixed point to integer */
+		in_pos_rounded = in_pos_aligned / 8192U;
+
+		if (in_pos_rounded < in_start)
+			continue;
+		if (in_pos_rounded >= in_end)
+			break;
 
 		if ((in_burst > 1) &&
-		    (in_edge - in_pos_aligned / 8192U) % in_burst)
+		    (in_edge - in_pos_rounded) % in_burst)
 			continue;
 
 		if (in_pos < in_pos_aligned)
@@ -527,19 +546,18 @@ static void find_best_seam(struct ipu_image_convert_ctx *ctx,
 			abs_diff = in_pos - in_pos_aligned;
 
 		if (abs_diff < min_diff) {
-			in_seam = in_pos_aligned;
+			in_seam = in_pos_rounded;
 			out_seam = out_pos;
 			min_diff = abs_diff;
 		}
 	}
 
 	*_out_seam = out_seam;
-	/* Convert 19.13 fixed point to integer seam position */
-	*_in_seam = DIV_ROUND_CLOSEST(in_seam, 8192U);
+	*_in_seam = in_seam;
 
-	dev_dbg(dev, "%s: out_seam %u(%u) in [%u, %u], in_seam %u(%u) diff %u.%03u\n",
+	dev_dbg(dev, "%s: out_seam %u(%u) in [%u, %u], in_seam %u(%u) in [%u, %u] diff %u.%03u\n",
 		__func__, out_seam, out_align, out_start, out_end,
-		*_in_seam, in_align, min_diff / 8192,
+		in_seam, in_align, in_start, in_end, min_diff / 8192,
 		DIV_ROUND_CLOSEST(min_diff % 8192 * 1000, 8192));
 }
 
-- 
2.20.1