[Mesa-dev] [PATCH 2/3] radeonsi: add sampling of 4:2:2 subsampled textures

[Grigori Goronzy]

This makes 4:2:2 video surfaces work in VDPAU.
---
 src/gallium/drivers/radeon/r600_texture.c |  5 +-
 src/gallium/drivers/radeonsi/si_blit.c    | 91 ++++++++++++++++++-------------
 src/gallium/drivers/radeonsi/si_state.c   | 15 +++++
 3 files changed, 71 insertions(+), 40 deletions(-)

diff --git a/src/gallium/drivers/radeon/r600_texture.c b/src/gallium/drivers/radeon/r600_texture.c
index 3a37465..a20b0c8 100644
--- a/src/gallium/drivers/radeon/r600_texture.c
+++ b/src/gallium/drivers/radeon/r600_texture.c
@@ -737,9 +737,8 @@ static unsigned r600_choose_tiling(struct r600_common_screen *rscreen,
 	 * Compressed textures must always be tiled. */
 	if (!(templ->flags & R600_RESOURCE_FLAG_FORCE_TILING) &&
 	    !util_format_is_compressed(templ->format)) {
-		/* Tiling doesn't work with the 422 (SUBSAMPLED) formats on R600-Cayman. */
-		if (rscreen->chip_class <= CAYMAN &&
-		    desc->layout == UTIL_FORMAT_LAYOUT_SUBSAMPLED)
+		/* Tiling doesn't work with the 422 (SUBSAMPLED) formats on R600+. */
+		if (desc->layout == UTIL_FORMAT_LAYOUT_SUBSAMPLED)
 			return RADEON_SURF_MODE_LINEAR_ALIGNED;
 
 		/* Cursors are linear on SI.
diff --git a/src/gallium/drivers/radeonsi/si_blit.c b/src/gallium/drivers/radeonsi/si_blit.c
index e02615f..8c3e136 100644
--- a/src/gallium/drivers/radeonsi/si_blit.c
+++ b/src/gallium/drivers/radeonsi/si_blit.c
@@ -548,46 +548,63 @@ static void si_resource_copy_region(struct pipe_context *ctx,
 		dstx = util_format_get_nblocksx(orig_info[1].format, dstx);
 		dsty = util_format_get_nblocksy(orig_info[1].format, dsty);
 	} else if (!util_blitter_is_copy_supported(sctx->blitter, dst, src)) {
-		unsigned blocksize = util_format_get_blocksize(src->format);
-
-		switch (blocksize) {
-		case 1:
-			si_change_format(src, src_level, &orig_info[0],
-					 PIPE_FORMAT_R8_UNORM);
-			si_change_format(dst, dst_level, &orig_info[1],
-					 PIPE_FORMAT_R8_UNORM);
-			break;
-		case 2:
-			si_change_format(src, src_level, &orig_info[0],
-					 PIPE_FORMAT_R8G8_UNORM);
-			si_change_format(dst, dst_level, &orig_info[1],
-					 PIPE_FORMAT_R8G8_UNORM);
-			break;
-		case 4:
-			si_change_format(src, src_level, &orig_info[0],
-					 PIPE_FORMAT_R8G8B8A8_UNORM);
-			si_change_format(dst, dst_level, &orig_info[1],
-					 PIPE_FORMAT_R8G8B8A8_UNORM);
-			break;
-		case 8:
-			si_change_format(src, src_level, &orig_info[0],
-					 PIPE_FORMAT_R16G16B16A16_UINT);
-			si_change_format(dst, dst_level, &orig_info[1],
-					 PIPE_FORMAT_R16G16B16A16_UINT);
-			break;
-		case 16:
+		if (util_format_is_subsampled_422(src->format)) {
+			/* XXX untested */
 			si_change_format(src, src_level, &orig_info[0],
-					 PIPE_FORMAT_R32G32B32A32_UINT);
+					 PIPE_FORMAT_R8G8B8A8_UINT);
 			si_change_format(dst, dst_level, &orig_info[1],
-					 PIPE_FORMAT_R32G32B32A32_UINT);
-			break;
-		default:
-			fprintf(stderr, "Unhandled format %s with blocksize %u\n",
-				util_format_short_name(src->format), blocksize);
-			assert(0);
+					 PIPE_FORMAT_R8G8B8A8_UINT);
+
+			sbox = *src_box;
+			sbox.x = util_format_get_nblocksx(orig_info[0].format, src_box->x);
+			sbox.width = util_format_get_nblocksx(orig_info[0].format, src_box->width);
+			src_box = &sbox;
+			dstx = util_format_get_nblocksx(orig_info[1].format, dstx);
+
+			restore_orig[0] = TRUE;
+			restore_orig[1] = TRUE;
+		} else {
+			unsigned blocksize = util_format_get_blocksize(src->format);
+
+			switch (blocksize) {
+			case 1:
+				si_change_format(src, src_level, &orig_info[0],
+						PIPE_FORMAT_R8_UNORM);
+				si_change_format(dst, dst_level, &orig_info[1],
+						PIPE_FORMAT_R8_UNORM);
+				break;
+			case 2:
+				si_change_format(src, src_level, &orig_info[0],
+						PIPE_FORMAT_R8G8_UNORM);
+				si_change_format(dst, dst_level, &orig_info[1],
+						PIPE_FORMAT_R8G8_UNORM);
+				break;
+			case 4:
+				si_change_format(src, src_level, &orig_info[0],
+						PIPE_FORMAT_R8G8B8A8_UNORM);
+				si_change_format(dst, dst_level, &orig_info[1],
+						PIPE_FORMAT_R8G8B8A8_UNORM);
+				break;
+			case 8:
+				si_change_format(src, src_level, &orig_info[0],
+						PIPE_FORMAT_R16G16B16A16_UINT);
+				si_change_format(dst, dst_level, &orig_info[1],
+						PIPE_FORMAT_R16G16B16A16_UINT);
+				break;
+			case 16:
+				si_change_format(src, src_level, &orig_info[0],
+						PIPE_FORMAT_R32G32B32A32_UINT);
+				si_change_format(dst, dst_level, &orig_info[1],
+						PIPE_FORMAT_R32G32B32A32_UINT);
+				break;
+			default:
+				fprintf(stderr, "Unhandled format %s with blocksize %u\n",
+					util_format_short_name(src->format), blocksize);
+				assert(0);
+			}
+			restore_orig[0] = TRUE;
+			restore_orig[1] = TRUE;
 		}
-		restore_orig[0] = TRUE;
-		restore_orig[1] = TRUE;
 	}
 
 	/* Initialize the surface. */
diff --git a/src/gallium/drivers/radeonsi/si_state.c b/src/gallium/drivers/radeonsi/si_state.c
index 19d2b55..cbd51ad 100644
--- a/src/gallium/drivers/radeonsi/si_state.c
+++ b/src/gallium/drivers/radeonsi/si_state.c
@@ -1102,6 +1102,19 @@ static uint32_t si_translate_texformat(struct pipe_screen *screen,
 		}
 	}
 
+	if (desc->layout == UTIL_FORMAT_LAYOUT_SUBSAMPLED) {
+		switch (format) {
+		case PIPE_FORMAT_R8G8_B8G8_UNORM:
+		case PIPE_FORMAT_G8R8_B8R8_UNORM:
+			return V_008F14_IMG_DATA_FORMAT_GB_GR;
+		case PIPE_FORMAT_G8R8_G8B8_UNORM:
+		case PIPE_FORMAT_R8G8_R8B8_UNORM:
+			return V_008F14_IMG_DATA_FORMAT_BG_RG;
+		default:
+			goto out_unknown;
+		}
+	}
+
 	if (desc->layout == UTIL_FORMAT_LAYOUT_S3TC) {
 
 		if (!enable_s3tc)
@@ -2481,6 +2494,8 @@ static struct pipe_sampler_view *si_create_sampler_view(struct pipe_context *ctx
 					num_format = V_008F14_IMG_NUM_FORMAT_UNORM;
 					break;
 				}
+			} else if (desc->layout == UTIL_FORMAT_LAYOUT_SUBSAMPLED) {
+				num_format = V_008F14_IMG_NUM_FORMAT_UNORM;
 			} else {
 				num_format = V_008F14_IMG_NUM_FORMAT_FLOAT;
 			}
-- 
1.8.3.2