[Mesa-dev] [PATCH 3/5] radeonsi: add support for displayable DCC for 1 RB chips

Marek Olšák maraeo at gmail.com
Thu Feb 28 21:20:23 UTC 2019


From: Marek Olšák <marek.olsak at amd.com>

This is the simpler codepath - just disable RB and pipe alignment for DCC.
---
 src/amd/common/ac_gpu_info.c               |  2 +
 src/amd/common/ac_gpu_info.h               |  3 +
 src/amd/common/ac_surface.c                | 25 +++++++-
 src/amd/common/ac_surface.h                |  2 +-
 src/gallium/drivers/radeon/radeon_winsys.h |  6 ++
 src/gallium/drivers/radeonsi/si_texture.c  | 74 ++++++++++++++++++++--
 src/gallium/winsys/amdgpu/drm/amdgpu_bo.c  |  8 +++
 7 files changed, 113 insertions(+), 7 deletions(-)

diff --git a/src/amd/common/ac_gpu_info.c b/src/amd/common/ac_gpu_info.c
index fc8c6a09d2f..a6d249a6d2f 100644
--- a/src/amd/common/ac_gpu_info.c
+++ b/src/amd/common/ac_gpu_info.c
@@ -496,20 +496,22 @@ void ac_print_gpu_info(struct radeon_info *info)
 	       info->pci_domain, info->pci_bus,
 	       info->pci_dev, info->pci_func);
 	printf("    pci_id = 0x%x\n", info->pci_id);
 	printf("    family = %i\n", info->family);
 	printf("    chip_class = %i\n", info->chip_class);
 	printf("    num_compute_rings = %u\n", info->num_compute_rings);
 	printf("    num_sdma_rings = %i\n", info->num_sdma_rings);
 	printf("    clock_crystal_freq = %i\n", info->clock_crystal_freq);
 	printf("    tcc_cache_line_size = %u\n", info->tcc_cache_line_size);
 
+	printf("    use_display_dcc_unaligned = %u\n", info->use_display_dcc_unaligned);
+
 	printf("Memory info:\n");
 	printf("    pte_fragment_size = %u\n", info->pte_fragment_size);
 	printf("    gart_page_size = %u\n", info->gart_page_size);
 	printf("    gart_size = %i MB\n", (int)DIV_ROUND_UP(info->gart_size, 1024*1024));
 	printf("    vram_size = %i MB\n", (int)DIV_ROUND_UP(info->vram_size, 1024*1024));
 	printf("    vram_vis_size = %i MB\n", (int)DIV_ROUND_UP(info->vram_vis_size, 1024*1024));
 	printf("    gds_size = %u kB\n", info->gds_size / 1024);
 	printf("    gds_gfx_partition_size = %u kB\n", info->gds_gfx_partition_size / 1024);
 	printf("    max_alloc_size = %i MB\n",
 	       (int)DIV_ROUND_UP(info->max_alloc_size, 1024*1024));
diff --git a/src/amd/common/ac_gpu_info.h b/src/amd/common/ac_gpu_info.h
index b1ef9c53734..99fed520618 100644
--- a/src/amd/common/ac_gpu_info.h
+++ b/src/amd/common/ac_gpu_info.h
@@ -49,20 +49,23 @@ struct radeon_info {
 	/* Device info. */
 	const char                  *name;
 	uint32_t                    pci_id;
 	enum radeon_family          family;
 	enum chip_class             chip_class;
 	uint32_t                    num_compute_rings;
 	uint32_t                    num_sdma_rings;
 	uint32_t                    clock_crystal_freq;
 	uint32_t                    tcc_cache_line_size;
 
+	/* Disable RB and pipe alignment to skip the retile blit. (1 RB chips only) */
+	bool                        use_display_dcc_unaligned;
+
 	/* Memory info. */
 	uint32_t                    pte_fragment_size;
 	uint32_t                    gart_page_size;
 	uint64_t                    gart_size;
 	uint64_t                    vram_size;
 	uint64_t                    vram_vis_size;
 	unsigned                    gds_size;
 	unsigned                    gds_gfx_partition_size;
 	uint64_t                    max_alloc_size;
 	uint32_t                    min_alloc_size;
diff --git a/src/amd/common/ac_surface.c b/src/amd/common/ac_surface.c
index 91004e032a3..6802ab2badb 100644
--- a/src/amd/common/ac_surface.c
+++ b/src/amd/common/ac_surface.c
@@ -471,21 +471,22 @@ static unsigned cik_get_macro_tile_index(struct radeon_surf *surf)
 	assert(index < 16);
 	return index;
 }
 
 static bool get_display_flag(const struct ac_surf_config *config,
 			     const struct radeon_surf *surf)
 {
 	unsigned num_channels = config->info.num_channels;
 	unsigned bpe = surf->bpe;
 
-	if (surf->flags & RADEON_SURF_SCANOUT &&
+	if (!(surf->flags & RADEON_SURF_Z_OR_SBUFFER) &&
+	    surf->flags & RADEON_SURF_SCANOUT &&
 	    config->info.samples <= 1 &&
 	    surf->blk_w <= 2 && surf->blk_h == 1) {
 		/* subsampled */
 		if (surf->blk_w == 2 && surf->blk_h == 1)
 			return true;
 
 		if  (/* RGBA8 or RGBA16F */
 		     (bpe >= 4 && bpe <= 8 && num_channels == 4) ||
 		     /* R5G6B5 or R5G5B5A1 */
 		     (bpe == 2 && num_channels >= 3) ||
@@ -1208,21 +1209,21 @@ static int gfx9_compute_miptree(ADDR_HANDLE addrlib,
 			din.numFrags = in->numFrags;
 			din.numMipLevels = in->numMipLevels;
 			din.dataSurfaceSize = out.surfSize;
 
 			ret = Addr2ComputeDccInfo(addrlib, &din, &dout);
 			if (ret != ADDR_OK)
 				return ret;
 
 			surf->u.gfx9.dcc.rb_aligned = din.dccKeyFlags.rbAligned;
 			surf->u.gfx9.dcc.pipe_aligned = din.dccKeyFlags.pipeAligned;
-			surf->u.gfx9.dcc_pitch_max = dout.pitch - 1;
+			surf->u.gfx9.display_dcc_pitch_max = dout.pitch - 1;
 			surf->dcc_size = dout.dccRamSize;
 			surf->dcc_alignment = dout.dccRamBaseAlign;
 			surf->num_dcc_levels = in->numMipLevels;
 
 			/* Disable DCC for levels that are in the mip tail.
 			 *
 			 * There are two issues that this is intended to
 			 * address:
 			 *
 			 * 1. Multiple mip levels may share a cache line. This
@@ -1444,20 +1445,33 @@ static int gfx9_compute_surface(ADDR_HANDLE addrlib,
 		AddrSurfInfoIn.numSlices = config->info.depth;
 	else if (config->is_cube)
 		AddrSurfInfoIn.numSlices = 6;
 	else
 		AddrSurfInfoIn.numSlices = config->info.array_size;
 
 	/* This is propagated to HTILE/DCC/CMASK. */
 	AddrSurfInfoIn.flags.metaPipeUnaligned = 0;
 	AddrSurfInfoIn.flags.metaRbUnaligned = 0;
 
+	/* The display hardware can only read DCC with RB_ALIGNED=0 and
+	 * PIPE_ALIGNED=0. PIPE_ALIGNED really means L2CACHE_ALIGNED.
+	 *
+	 * The CB block requires RB_ALIGNED=1 except 1 RB chips.
+	 * PIPE_ALIGNED is optional, but PIPE_ALIGNED=0 requires L2 flushes
+	 * after rendering, so PIPE_ALIGNED=1 is recommended.
+	 */
+	if (info->use_display_dcc_unaligned && is_color_surface &&
+	    AddrSurfInfoIn.flags.display) {
+		AddrSurfInfoIn.flags.metaPipeUnaligned = 1;
+		AddrSurfInfoIn.flags.metaRbUnaligned = 1;
+	}
+
 	switch (mode) {
 	case RADEON_SURF_MODE_LINEAR_ALIGNED:
 		assert(config->info.samples <= 1);
 		assert(!(surf->flags & RADEON_SURF_Z_OR_SBUFFER));
 		AddrSurfInfoIn.swizzleMode = ADDR_SW_LINEAR;
 		break;
 
 	case RADEON_SURF_MODE_1D:
 	case RADEON_SURF_MODE_2D:
 		if (surf->flags & RADEON_SURF_IMPORTED) {
@@ -1516,20 +1530,27 @@ static int gfx9_compute_surface(ADDR_HANDLE addrlib,
 
 	surf->is_linear = surf->u.gfx9.surf.swizzle_mode == ADDR_SW_LINEAR;
 
 	/* Query whether the surface is displayable. */
 	bool displayable = false;
 	if (!config->is_3d && !config->is_cube) {
 		r = Addr2IsValidDisplaySwizzleMode(addrlib, surf->u.gfx9.surf.swizzle_mode,
 					   surf->bpe * 8, &displayable);
 		if (r)
 			return r;
+
+		/* Display needs unaligned DCC. */
+		if (info->use_display_dcc_unaligned &&
+		    surf->num_dcc_levels &&
+		    (surf->u.gfx9.dcc.pipe_aligned ||
+		     surf->u.gfx9.dcc.rb_aligned))
+			displayable = false;
 	}
 	surf->is_displayable = displayable;
 
 	switch (surf->u.gfx9.surf.swizzle_mode) {
 		/* S = standard. */
 		case ADDR_SW_256B_S:
 		case ADDR_SW_4KB_S:
 		case ADDR_SW_64KB_S:
 		case ADDR_SW_VAR_S:
 		case ADDR_SW_64KB_S_T:
diff --git a/src/amd/common/ac_surface.h b/src/amd/common/ac_surface.h
index 7ae166c70a3..eb50c37c3c2 100644
--- a/src/amd/common/ac_surface.h
+++ b/src/amd/common/ac_surface.h
@@ -142,21 +142,21 @@ struct gfx9_surf_layout {
     enum gfx9_resource_type     resource_type; /* 1D, 2D or 3D */
     uint16_t                    surf_pitch; /* in blocks */
     uint16_t                    surf_height;
 
     uint64_t                    surf_offset; /* 0 unless imported with an offset */
     /* The size of the 2D plane containing all mipmap levels. */
     uint64_t                    surf_slice_size;
     /* Mipmap level offset within the slice in bytes. Only valid for LINEAR. */
     uint32_t                    offset[RADEON_SURF_MAX_LEVELS];
 
-    uint16_t                    dcc_pitch_max;  /* (mip chain pitch - 1) */
+    uint16_t                    display_dcc_pitch_max;  /* (mip chain pitch - 1) */
 
     uint64_t                    stencil_offset; /* separate stencil */
 };
 
 struct radeon_surf {
     /* Format properties. */
     unsigned                    blk_w:4;
     unsigned                    blk_h:4;
     unsigned                    bpe:5;
     /* Number of mipmap levels where DCC is enabled starting from level 0.
diff --git a/src/gallium/drivers/radeon/radeon_winsys.h b/src/gallium/drivers/radeon/radeon_winsys.h
index 82feef39487..4e53c992fdf 100644
--- a/src/gallium/drivers/radeon/radeon_winsys.h
+++ b/src/gallium/drivers/radeon/radeon_winsys.h
@@ -210,20 +210,26 @@ struct radeon_bo_metadata {
             unsigned                tile_split;
             unsigned                mtilea;
             unsigned                num_banks;
             unsigned                stride;
             bool                    scanout;
         } legacy;
 
         struct {
             /* surface flags */
             unsigned swizzle_mode:5;
+
+            /* DCC flags */
+            /* [31:8]: max offset = 4GB - 256; 0 = DCC disabled */
+            unsigned dcc_offset_256B:24;
+            unsigned dcc_pitch_max:14;   /* (mip chain pitch - 1) for DCN */
+            unsigned dcc_independent_64B:1;
         } gfx9;
     } u;
 
     /* Additional metadata associated with the buffer, in bytes.
      * The maximum size is 64 * 4. This is opaque for the winsys & kernel.
      * Supported by amdgpu only.
      */
     uint32_t                size_metadata;
     uint32_t                metadata[64];
 };
diff --git a/src/gallium/drivers/radeonsi/si_texture.c b/src/gallium/drivers/radeonsi/si_texture.c
index 581f90a7b2f..cb62f153e59 100644
--- a/src/gallium/drivers/radeonsi/si_texture.c
+++ b/src/gallium/drivers/radeonsi/si_texture.c
@@ -30,20 +30,21 @@
 #include "util/u_memory.h"
 #include "util/u_pack_color.h"
 #include "util/u_resource.h"
 #include "util/u_surface.h"
 #include "util/u_transfer.h"
 #include "util/os_time.h"
 #include <errno.h>
 #include <inttypes.h>
 #include "state_tracker/drm_driver.h"
 #include "amd/common/sid.h"
+#include "amd/common/gfx9d.h"
 
 static enum radeon_surf_mode
 si_choose_tiling(struct si_screen *sscreen,
 		 const struct pipe_resource *templ, bool tc_compatible_htile);
 
 
 bool si_prepare_for_dma_blit(struct si_context *sctx,
 			     struct si_texture *dst,
 			     unsigned dst_level, unsigned dstx,
 			     unsigned dsty, unsigned dstz,
@@ -344,20 +345,25 @@ static void si_get_display_metadata(struct si_screen *sscreen,
 	if (sscreen->info.chip_class >= GFX9) {
 		if (metadata->u.gfx9.swizzle_mode > 0)
 			*array_mode = RADEON_SURF_MODE_2D;
 		else
 			*array_mode = RADEON_SURF_MODE_LINEAR_ALIGNED;
 
 		*is_scanout = metadata->u.gfx9.swizzle_mode == 0 ||
 			      metadata->u.gfx9.swizzle_mode % 4 == 2;
 
 		surf->u.gfx9.surf.swizzle_mode = metadata->u.gfx9.swizzle_mode;
+
+		if (metadata->u.gfx9.dcc_offset_256B) {
+			surf->u.gfx9.display_dcc_pitch_max = metadata->u.gfx9.dcc_pitch_max;
+			assert(metadata->u.gfx9.dcc_independent_64B == 1);
+		}
 	} else {
 		surf->u.legacy.pipe_config = metadata->u.legacy.pipe_config;
 		surf->u.legacy.bankw = metadata->u.legacy.bankw;
 		surf->u.legacy.bankh = metadata->u.legacy.bankh;
 		surf->u.legacy.tile_split = metadata->u.legacy.tile_split;
 		surf->u.legacy.mtilea = metadata->u.legacy.mtilea;
 		surf->u.legacy.num_banks = metadata->u.legacy.num_banks;
 
 		if (metadata->u.legacy.macrotile == RADEON_LAYOUT_TILED)
 			*array_mode = RADEON_SURF_MODE_2D;
@@ -610,20 +616,29 @@ static void si_set_tex_bo_metadata(struct si_screen *sscreen,
 				   struct si_texture *tex)
 {
 	struct radeon_surf *surface = &tex->surface;
 	struct pipe_resource *res = &tex->buffer.b.b;
 	struct radeon_bo_metadata md;
 
 	memset(&md, 0, sizeof(md));
 
 	if (sscreen->info.chip_class >= GFX9) {
 		md.u.gfx9.swizzle_mode = surface->u.gfx9.surf.swizzle_mode;
+
+		if (tex->dcc_offset && !tex->dcc_separate_buffer) {
+			uint64_t dcc_offset = tex->dcc_offset;
+
+			assert((dcc_offset >> 8) != 0 && (dcc_offset >> 8) < (1 << 24));
+			md.u.gfx9.dcc_offset_256B = dcc_offset >> 8;
+			md.u.gfx9.dcc_pitch_max = tex->surface.u.gfx9.display_dcc_pitch_max;
+			md.u.gfx9.dcc_independent_64B = 1;
+		}
 	} else {
 		md.u.legacy.microtile = surface->u.legacy.level[0].mode >= RADEON_SURF_MODE_1D ?
 					   RADEON_LAYOUT_TILED : RADEON_LAYOUT_LINEAR;
 		md.u.legacy.macrotile = surface->u.legacy.level[0].mode >= RADEON_SURF_MODE_2D ?
 					   RADEON_LAYOUT_TILED : RADEON_LAYOUT_LINEAR;
 		md.u.legacy.pipe_config = surface->u.legacy.pipe_config;
 		md.u.legacy.bankw = surface->u.legacy.bankw;
 		md.u.legacy.bankh = surface->u.legacy.bankh;
 		md.u.legacy.tile_split = surface->u.legacy.tile_split;
 		md.u.legacy.mtilea = surface->u.legacy.mtilea;
@@ -699,29 +714,65 @@ static void si_get_opaque_metadata(struct si_screen *sscreen,
 		return;
 
 	/* Return if DCC is enabled. The texture should be set up with it
 	 * already.
 	 */
 	if (md->size_metadata >= 10 * 4 && /* at least 2(header) + 8(desc) dwords */
 	    md->metadata[0] != 0 &&
 	    md->metadata[1] == si_get_bo_metadata_word1(sscreen) &&
 	    G_008F28_COMPRESSION_EN(desc[6])) {
 		tex->dcc_offset = (uint64_t)desc[7] << 8;
+
+		if (sscreen->info.chip_class >= GFX9) {
+			/* Fix up parameters for displayable DCC. Some state
+			 * trackers don't set the SCANOUT flag when importing
+			 * displayable images, so we have to recover the correct
+			 * parameters here.
+			 */
+			tex->surface.u.gfx9.dcc.pipe_aligned =
+				G_008F24_META_PIPE_ALIGNED(desc[5]);
+			tex->surface.u.gfx9.dcc.rb_aligned =
+				G_008F24_META_RB_ALIGNED(desc[5]);
+
+			/* If DCC is unaligned, this can only be a displayable image. */
+			if (!tex->surface.u.gfx9.dcc.pipe_aligned &&
+			    !tex->surface.u.gfx9.dcc.rb_aligned)
+				tex->surface.is_displayable = true;
+		}
 		return;
 	}
 
 	/* Disable DCC. These are always set by texture_from_handle and must
 	 * be cleared here.
 	 */
 	tex->dcc_offset = 0;
 }
 
+static bool si_has_displayable_dcc(struct si_texture *tex)
+{
+	struct si_screen *sscreen = (struct si_screen*)tex->buffer.b.b.screen;
+
+	if (sscreen->info.chip_class <= VI)
+		return false;
+
+	/* This needs a cache flush before scanout.
+	 * (it can't be scanned out and rendered to simultaneously)
+	 */
+	if (sscreen->info.use_display_dcc_unaligned &&
+	    tex->dcc_offset &&
+	    !tex->surface.u.gfx9.dcc.pipe_aligned &&
+	    !tex->surface.u.gfx9.dcc.rb_aligned)
+		return true;
+
+	return false;
+}
+
 static boolean si_texture_get_handle(struct pipe_screen* screen,
 				     struct pipe_context *ctx,
 				     struct pipe_resource *resource,
 				     struct winsys_handle *whandle,
 				     unsigned usage)
 {
 	struct si_screen *sscreen = (struct si_screen*)screen;
 	struct si_context *sctx;
 	struct si_resource *res = si_resource(resource);
 	struct si_texture *tex = (struct si_texture*)resource;
@@ -752,21 +803,24 @@ static boolean si_texture_get_handle(struct pipe_screen* screen,
 			assert(res->b.b.bind & PIPE_BIND_SHARED);
 			assert(res->flags & RADEON_FLAG_NO_SUBALLOC);
 			assert(!(res->flags & RADEON_FLAG_NO_INTERPROCESS_SHARING));
 			assert(tex->surface.tile_swizzle == 0);
 		}
 
 		/* Since shader image stores don't support DCC on VI,
 		 * disable it for external clients that want write
 		 * access.
 		 */
-		if (usage & PIPE_HANDLE_USAGE_SHADER_WRITE && tex->dcc_offset) {
+		if ((usage & PIPE_HANDLE_USAGE_SHADER_WRITE && tex->dcc_offset) ||
+		    /* Displayable DCC requires an explicit flush. */
+		    (!(usage & PIPE_HANDLE_USAGE_EXPLICIT_FLUSH) &&
+		     si_has_displayable_dcc(tex))) {
 			if (si_texture_disable_dcc(sctx, tex)) {
 				update_metadata = true;
 				/* si_texture_disable_dcc flushes the context */
 				flush = false;
 			}
 		}
 
 		if (!(usage & PIPE_HANDLE_USAGE_EXPLICIT_FLUSH) &&
 		    (tex->cmask_buffer || tex->dcc_offset)) {
 			/* Eliminate fast clear (both CMASK and DCC) */
@@ -1005,21 +1059,21 @@ void si_print_texture_info(struct si_screen *sscreen,
 				tex->surface.htile_alignment,
 				tex->surface.u.gfx9.htile.rb_aligned,
 				tex->surface.u.gfx9.htile.pipe_aligned);
 		}
 
 		if (tex->dcc_offset) {
 			u_log_printf(log, "  DCC: offset=%"PRIu64", size=%u, "
 				"alignment=%u, pitch_max=%u, num_dcc_levels=%u\n",
 				tex->dcc_offset, tex->surface.dcc_size,
 				tex->surface.dcc_alignment,
-				tex->surface.u.gfx9.dcc_pitch_max,
+				tex->surface.u.gfx9.display_dcc_pitch_max,
 				tex->surface.num_dcc_levels);
 		}
 
 		if (tex->surface.u.gfx9.stencil_offset) {
 			u_log_printf(log, "  Stencil: offset=%"PRIu64", swmode=%u, epitch=%u\n",
 				tex->surface.u.gfx9.stencil_offset,
 				tex->surface.u.gfx9.stencil.swizzle_mode,
 				tex->surface.u.gfx9.stencil.epitch);
 		}
 		return;
@@ -1192,22 +1246,23 @@ si_texture_create_object(struct pipe_screen *screen,
 			if (!tex->surface.fmask_size || !tex->surface.cmask_size)
 				goto error;
 		}
 
 		/* Shared textures must always set up DCC here.
 		 * If it's not present, it will be disabled by
 		 * apply_opaque_metadata later.
 		 */
 		if (tex->surface.dcc_size &&
 		    (buf || !(sscreen->debug_flags & DBG(NO_DCC))) &&
-		    !(tex->surface.flags & RADEON_SURF_SCANOUT)) {
-			/* Reserve space for the DCC buffer. */
+		    (sscreen->info.use_display_dcc_unaligned ||
+		     !(tex->surface.flags & RADEON_SURF_SCANOUT))) {
+			/* Add space for the DCC buffer. */
 			tex->dcc_offset = align64(tex->size, tex->surface.dcc_alignment);
 			tex->size = tex->dcc_offset + tex->surface.dcc_size;
 		}
 	}
 
 	/* Now create the backing buffer. */
 	if (!buf) {
 		si_init_resource_fields(sscreen, resource, tex->size,
 					  tex->surface.surf_alignment);
 
@@ -1503,20 +1558,31 @@ static struct pipe_resource *si_texture_from_winsys_buffer(struct si_screen *ssc
 
 	tex = si_texture_create_object(&sscreen->b, templ, buf, &surface);
 	if (!tex)
 		return NULL;
 
 	tex->buffer.b.is_shared = true;
 	tex->buffer.external_usage = usage;
 
 	si_get_opaque_metadata(sscreen, tex, &metadata);
 
+	/* Displayable DCC requires an explicit flush. */
+	if (dedicated &&
+	    !(usage & PIPE_HANDLE_USAGE_EXPLICIT_FLUSH) &&
+	    si_has_displayable_dcc(tex)) {
+		/* TODO: do we need to decompress DCC? */
+		if (si_texture_discard_dcc(sscreen, tex)) {
+			/* Update BO metadata after disabling DCC. */
+			si_set_tex_bo_metadata(sscreen, tex);
+		}
+	}
+
 	assert(tex->surface.tile_swizzle == 0);
 	return &tex->buffer.b.b;
 }
 
 static struct pipe_resource *si_texture_from_handle(struct pipe_screen *screen,
 						    const struct pipe_resource *templ,
 						    struct winsys_handle *whandle,
 						    unsigned usage)
 {
 	struct si_screen *sscreen = (struct si_screen*)screen;
diff --git a/src/gallium/winsys/amdgpu/drm/amdgpu_bo.c b/src/gallium/winsys/amdgpu/drm/amdgpu_bo.c
index 58979bd4ea7..c1863057370 100644
--- a/src/gallium/winsys/amdgpu/drm/amdgpu_bo.c
+++ b/src/gallium/winsys/amdgpu/drm/amdgpu_bo.c
@@ -1219,20 +1219,24 @@ static void amdgpu_buffer_get_metadata(struct pb_buffer *_buf,
    assert(bo->bo && "must not be called for slab entries");
 
    r = amdgpu_bo_query_info(bo->bo, &info);
    if (r)
       return;
 
    tiling_flags = info.metadata.tiling_info;
 
    if (bo->ws->info.chip_class >= GFX9) {
       md->u.gfx9.swizzle_mode = AMDGPU_TILING_GET(tiling_flags, SWIZZLE_MODE);
+
+      md->u.gfx9.dcc_offset_256B = AMDGPU_TILING_GET(tiling_flags, DCC_OFFSET_256B);
+      md->u.gfx9.dcc_pitch_max = AMDGPU_TILING_GET(tiling_flags, DCC_PITCH_MAX);
+      md->u.gfx9.dcc_independent_64B = AMDGPU_TILING_GET(tiling_flags, DCC_INDEPENDENT_64B);
    } else {
       md->u.legacy.microtile = RADEON_LAYOUT_LINEAR;
       md->u.legacy.macrotile = RADEON_LAYOUT_LINEAR;
 
       if (AMDGPU_TILING_GET(tiling_flags, ARRAY_MODE) == 4)  /* 2D_TILED_THIN1 */
          md->u.legacy.macrotile = RADEON_LAYOUT_TILED;
       else if (AMDGPU_TILING_GET(tiling_flags, ARRAY_MODE) == 2) /* 1D_TILED_THIN1 */
          md->u.legacy.microtile = RADEON_LAYOUT_TILED;
 
       md->u.legacy.pipe_config = AMDGPU_TILING_GET(tiling_flags, PIPE_CONFIG);
@@ -1252,20 +1256,24 @@ static void amdgpu_buffer_set_metadata(struct pb_buffer *_buf,
                                        struct radeon_bo_metadata *md)
 {
    struct amdgpu_winsys_bo *bo = amdgpu_winsys_bo(_buf);
    struct amdgpu_bo_metadata metadata = {0};
    uint64_t tiling_flags = 0;
 
    assert(bo->bo && "must not be called for slab entries");
 
    if (bo->ws->info.chip_class >= GFX9) {
       tiling_flags |= AMDGPU_TILING_SET(SWIZZLE_MODE, md->u.gfx9.swizzle_mode);
+
+      tiling_flags |= AMDGPU_TILING_SET(DCC_OFFSET_256B, md->u.gfx9.dcc_offset_256B);
+      tiling_flags |= AMDGPU_TILING_SET(DCC_PITCH_MAX, md->u.gfx9.dcc_pitch_max);
+      tiling_flags |= AMDGPU_TILING_SET(DCC_INDEPENDENT_64B, md->u.gfx9.dcc_independent_64B);
    } else {
       if (md->u.legacy.macrotile == RADEON_LAYOUT_TILED)
          tiling_flags |= AMDGPU_TILING_SET(ARRAY_MODE, 4); /* 2D_TILED_THIN1 */
       else if (md->u.legacy.microtile == RADEON_LAYOUT_TILED)
          tiling_flags |= AMDGPU_TILING_SET(ARRAY_MODE, 2); /* 1D_TILED_THIN1 */
       else
          tiling_flags |= AMDGPU_TILING_SET(ARRAY_MODE, 1); /* LINEAR_ALIGNED */
 
       tiling_flags |= AMDGPU_TILING_SET(PIPE_CONFIG, md->u.legacy.pipe_config);
       tiling_flags |= AMDGPU_TILING_SET(BANK_WIDTH, util_logbase2(md->u.legacy.bankw));
-- 
2.17.1



More information about the mesa-dev mailing list