freedreno/a6xx: Set a level's pitch based on minified level0 pitch, not width0.

Found from piglit fbo-generatemipmaps failures, then tracked down with the
texturator test.  The piece that really revealed things was finding that
1024x1 linear RGBA8 on the older blob drivers would have a pitch of 5120
instead of 4096, and the following levels minified that pitch.

Fixes ~124 piglit tests (~8.5% of piglit failures) on cheza.

Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/3987>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/3987>

2 files changed, 274 insertions, 13 deletions

diff --git a/src/freedreno/fdl/fd6_layout.c b/src/freedreno/fdl/fd6_layout.c
index 82a04f688e7..3d4efb1d0a8 100644
--- a/src/freedreno/fdl/fd6_layout.c
+++ b/src/freedreno/fdl/fd6_layout.c
@@ -63,6 +63,20 @@ static const struct {
 #define RGB_TILE_HEIGHT_ALIGNMENT 16
 #define UBWC_PLANE_SIZE_ALIGNMENT 4096
 
+static int
+fdl6_pitchalign(struct fdl_layout *layout, int ta, int level)
+{
+	const struct util_format_description *format_desc =
+		util_format_description(layout->format);
+
+	uint32_t pitchalign = 64;
+	if (fdl_tile_mode(layout, level))
+		pitchalign = tile_alignment[ta].pitchalign;
+	if (format_desc->layout == UTIL_FORMAT_LAYOUT_ASTC)
+		pitchalign *= util_format_get_blockwidth(layout->format);
+	return pitchalign;
+}
+
 /* NOTE: good way to test this is:  (for example)
  *  piglit/bin/texelFetch fs sampler3D 100x100x8
  */
@@ -88,8 +102,6 @@ fdl6_layout(struct fdl_layout *layout,
 	if (tile_alignment[layout->cpp].ubwc_blockwidth == 0)
 		layout->ubwc = false;
 
-	const struct util_format_description *format_desc =
-		util_format_description(format);
 	int ta = layout->cpp;
 
 	/* The z16/r16 formats seem to not play by the normal tiling rules: */
@@ -110,6 +122,8 @@ fdl6_layout(struct fdl_layout *layout,
 		layout->base_align = 64;
 	}
 
+	uint32_t pitch0 = util_align_npot(width0, fdl6_pitchalign(layout, ta, 0));
+
 	for (uint32_t level = 0; level < mip_levels; level++) {
 		uint32_t depth = u_minify(depth0, level);
 		struct fdl_slice *slice = &layout->slices[level];
@@ -125,14 +139,9 @@ fdl6_layout(struct fdl_layout *layout,
 			width = u_minify(width0, level);
 			height = u_minify(height0, level);
 		}
-		uint32_t pitchalign;
 
-		if (tile_mode) {
-			pitchalign = tile_alignment[ta].pitchalign;
+		if (tile_mode)
 			height = align(height, tile_alignment[ta].heightalign);
-		} else {
-			pitchalign = 64;
-		}
 
 		/* The blits used for mem<->gmem work at a granularity of
 		 * 32x32, which can cause faults due to over-fetch on the
@@ -144,11 +153,8 @@ fdl6_layout(struct fdl_layout *layout,
 		if (level == mip_levels - 1)
 			height = align(height, 32);
 
-		if (format_desc->layout == UTIL_FORMAT_LAYOUT_ASTC)
-			slice->pitch =
-				util_align_npot(width, pitchalign * util_format_get_blockwidth(format));
-		else
-			slice->pitch = align(width, pitchalign);
+		slice->pitch = util_align_npot(u_minify(pitch0, level),
+				fdl6_pitchalign(layout, ta, level));
 
 		slice->offset = layout->size;
 		uint32_t blocks = util_format_get_nblocks(format,
diff --git a/src/freedreno/fdl/fd6_layout_test.c b/src/freedreno/fdl/fd6_layout_test.c
index c687029f5ee..0f31ebb3ca8 100644
--- a/src/freedreno/fdl/fd6_layout_test.c
+++ b/src/freedreno/fdl/fd6_layout_test.c
@@ -118,6 +118,261 @@ static const struct testcase testcases[] = {
 		},
 	},
 #endif
+
+	/* A layout that we failed on (129 wide has a surprise level 1 pitch
+	 * increase), and the sizes bracketing it.
+	 */
+	{
+		.format = PIPE_FORMAT_R8G8B8A8_UNORM,
+		.layout = {
+			.width0 = 128, .height0 = 1,
+			.slices = {
+				{ .offset = 0, .pitch = 512 },
+				{ .offset = 512, .pitch = 256 },
+				{ .offset = 768, .pitch = 256 },
+				{ .offset = 1024, .pitch = 256 },
+				{ .offset = 1280, .pitch = 256 },
+				{ .offset = 1536, .pitch = 256 },
+				{ .offset = 1792, .pitch = 256 },
+				{ .offset = 2048, .pitch = 256 },
+			},
+		},
+	},
+	{
+		.format = PIPE_FORMAT_R8G8B8A8_UNORM,
+		.layout = {
+			.width0 = 129, .height0 = 1,
+			.slices = {
+				{ .offset = 0, .pitch = 768 },
+				{ .offset = 768, .pitch = 512 },
+				{ .offset = 1280, .pitch = 256 },
+				{ .offset = 1536, .pitch = 256 },
+				{ .offset = 1792, .pitch = 256 },
+				{ .offset = 2048, .pitch = 256 },
+				{ .offset = 2304, .pitch = 256 },
+				{ .offset = 2560, .pitch = 256 },
+			},
+		},
+	},
+	{
+		.format = PIPE_FORMAT_R8G8B8A8_UNORM,
+		.layout = {
+			.width0 = 130, .height0 = 1,
+			.slices = {
+				{ .offset = 0, .pitch = 768 },
+				{ .offset = 768, .pitch = 512 },
+				{ .offset = 1280, .pitch = 256 },
+				{ .offset = 1536, .pitch = 256 },
+				{ .offset = 1792, .pitch = 256 },
+				{ .offset = 2048, .pitch = 256 },
+				{ .offset = 2304, .pitch = 256 },
+				{ .offset = 2560, .pitch = 256 },
+			},
+		},
+	},
+
+	/* The 129 failure seems to be across formats, let's test some cpps */
+	{
+		.format = PIPE_FORMAT_R8_UNORM,
+		.layout = {
+			.width0 = 129, .height0 = 1,
+			.slices = {
+				{ .offset = 0, .pitch = 192 },
+				{ .offset = 192, .pitch = 128 },
+				{ .offset = 320, .pitch = 64 },
+				{ .offset = 384, .pitch = 64 },
+				{ .offset = 448, .pitch = 64 },
+				{ .offset = 512, .pitch = 64 },
+				{ .offset = 576, .pitch = 64 },
+				{ .offset = 640, .pitch = 64 },
+			},
+		},
+	},
+	{
+		.format = PIPE_FORMAT_R16_UINT,
+		.layout = {
+			.width0 = 129, .height0 = 1,
+			.slices = {
+				{ .offset = 0, .pitch = 384 },
+				{ .offset = 384, .pitch = 256 },
+				{ .offset = 640, .pitch = 128 },
+				{ .offset = 768, .pitch = 128 },
+				{ .offset = 896, .pitch = 128 },
+				{ .offset = 1024, .pitch = 128 },
+				{ .offset = 1152, .pitch = 128 },
+				{ .offset = 1280, .pitch = 128 },
+			},
+		},
+	},
+	{
+		.format = PIPE_FORMAT_R32G32B32A32_FLOAT,
+		.layout = {
+			.width0 = 129, .height0 = 1,
+			.slices = {
+				{ .offset = 0, .pitch = 3072 },
+				{ .offset = 3072, .pitch = 2048 },
+				{ .offset = 5120, .pitch = 1024 },
+				{ .offset = 6144, .pitch = 1024 },
+				{ .offset = 7168, .pitch = 1024 },
+				{ .offset = 8192, .pitch = 1024 },
+				{ .offset = 9216, .pitch = 1024 },
+				{ .offset = 10240, .pitch = 1024 },
+			},
+		},
+	},
+
+	/* The 129 failure replicated at every +256 pixels wide.  Pick one of
+	 * them, and this time increase the height as a new variable as well.
+	 */
+	{
+		.format = PIPE_FORMAT_R8G8B8A8_UNORM,
+		.layout = {
+			.width0 = 385, .height0 = 128,
+			.slices = {
+				{ .offset = 0, .pitch = 1792 },
+				{ .offset = 229376, .pitch = 1024 },
+				{ .offset = 294912, .pitch = 512 },
+				{ .offset = 311296, .pitch = 256 },
+				{ .offset = 315392, .pitch = 256 },
+				{ .offset = 317440, .pitch = 256 },
+				{ .offset = 318464, .pitch = 256 },
+				{ .offset = 318976, .pitch = 256 },
+				{ .offset = 319232, .pitch = 256 },
+			},
+		},
+	},
+
+	/* At 257-259 (and replicated every +256 pixels) we had another failure. */
+	{
+		.format = PIPE_FORMAT_R8G8B8A8_UNORM,
+		.layout = {
+			.width0 = 257, .height0 = 1,
+			.slices = {
+				{ .offset = 0, .pitch = 1280 },
+				{ .offset = 1280, .pitch = 768 },
+				{ .offset = 2048, .pitch = 512 },
+				{ .offset = 2560, .pitch = 256 },
+				{ .offset = 2816, .pitch = 256 },
+				{ .offset = 3072, .pitch = 256 },
+				{ .offset = 3328, .pitch = 256 },
+				{ .offset = 3584, .pitch = 256 },
+				{ .offset = 3840, .pitch = 256 },
+			},
+		},
+	},
+	{
+		.format = PIPE_FORMAT_R8G8B8A8_UNORM,
+		.layout = {
+			.width0 = 258, .height0 = 1,
+			.slices = {
+				{ .offset = 0, .pitch = 1280 },
+				{ .offset = 1280, .pitch = 768 },
+				{ .offset = 2048, .pitch = 512 },
+				{ .offset = 2560, .pitch = 256 },
+				{ .offset = 2816, .pitch = 256 },
+				{ .offset = 3072, .pitch = 256 },
+				{ .offset = 3328, .pitch = 256 },
+				{ .offset = 3584, .pitch = 256 },
+				{ .offset = 3840, .pitch = 256 },
+			},
+		},
+	},
+	{
+		.format = PIPE_FORMAT_R8G8B8A8_UNORM,
+		.layout = {
+			.width0 = 259, .height0 = 1,
+			.slices = {
+				{ .offset = 0, .pitch = 1280 },
+				{ .offset = 1280, .pitch = 768 },
+				{ .offset = 2048, .pitch = 512 },
+				{ .offset = 2560, .pitch = 256 },
+				{ .offset = 2816, .pitch = 256 },
+				{ .offset = 3072, .pitch = 256 },
+				{ .offset = 3328, .pitch = 256 },
+				{ .offset = 3584, .pitch = 256 },
+				{ .offset = 3840, .pitch = 256 },
+			},
+		},
+	},
+	{
+		.format = PIPE_FORMAT_R8G8B8A8_UNORM,
+		.layout = {
+			.width0 = 260, .height0 = 1,
+			.slices = {
+				{ .offset = 0, .pitch = 1280 },
+				{ .offset = 1280, .pitch = 768 },
+				{ .offset = 2048, .pitch = 512 },
+				{ .offset = 2560, .pitch = 256 },
+				{ .offset = 2816, .pitch = 256 },
+				{ .offset = 3072, .pitch = 256 },
+				{ .offset = 3328, .pitch = 256 },
+				{ .offset = 3584, .pitch = 256 },
+				{ .offset = 3840, .pitch = 256 },
+			},
+		},
+	},
+
+	/* And, again for the 257-9 failure, test a replica with a larger size*/
+	{
+		.format = PIPE_FORMAT_R8G8B8A8_UNORM,
+		.layout = {
+			.width0 = 513, .height0 = 32,
+			.slices = {
+				{ .offset = 0, .pitch = 2304 },
+				{ .offset = 73728, .pitch = 1280 },
+				{ .offset = 94208, .pitch = 768 },
+				{ .offset = 100352, .pitch = 512 },
+				{ .offset = 102400, .pitch = 256 },
+				{ .offset = 102912, .pitch = 256 },
+				{ .offset = 103168, .pitch = 256 },
+				{ .offset = 103424, .pitch = 256 },
+				{ .offset = 103680, .pitch = 256 },
+				{ .offset = 103936, .pitch = 256 },
+			},
+		},
+	},
+
+	/* Oh, look.  The 513-517 failure extends up to 518 at the next texture
+	 * level!
+	 */
+	{
+		.format = PIPE_FORMAT_R8G8B8A8_UNORM,
+		.layout = {
+			.width0 = 518, .height0 = 1,
+			.slices = {
+				{ .offset = 0, .pitch = 2304 },
+				{ .offset = 2304, .pitch = 1280 },
+				{ .offset = 3584, .pitch = 768 },
+				{ .offset = 4352, .pitch = 512 },
+				{ .offset = 4864, .pitch = 256 },
+				{ .offset = 5120, .pitch = 256 },
+				{ .offset = 5376, .pitch = 256 },
+				{ .offset = 5632, .pitch = 256 },
+				{ .offset = 5888, .pitch = 256 },
+				{ .offset = 6144, .pitch = 256 },
+			},
+		},
+	},
+
+	/* Tiled mode testing of the unusual 1/2-bytes-per-pixel pitch alignment */
+	{
+		.format = PIPE_FORMAT_R8_UNORM,
+		.layout = {
+			.tile_mode = TILE6_3,
+			.width0 = 129, .height0 = 1,
+			.slices = {
+				{ .offset = 0, .pitch = 256 },
+				{ .offset = 8192, .pitch = 128 },
+				{ .offset = 12288, .pitch = 128 },
+				{ .offset = 16384, .pitch = 128 },
+				{ .offset = 20480, .pitch = 64 },
+				{ .offset = 20544, .pitch = 64 },
+				{ .offset = 20608, .pitch = 64 },
+				{ .offset = 20672, .pitch = 64 },
+			},
+		},
+	},
+
 };
 
 static bool test_layout(const struct testcase *testcase)