gallium/radeon: print tiling index when printing texture info
[mesa.git] / src / gallium / drivers / radeon / r600_texture.c
index d3a498fb35d9c3841b5b9b473c57d35823e01aa2..2f2c17cd05e6d1020ea388ff04c39dde3b40a5c5 100644 (file)
@@ -37,8 +37,9 @@
 
 static void r600_texture_discard_cmask(struct r600_common_screen *rscreen,
                                       struct r600_texture *rtex);
-static unsigned r600_choose_tiling(struct r600_common_screen *rscreen,
-                                  const struct pipe_resource *templ);
+static enum radeon_surf_mode
+r600_choose_tiling(struct r600_common_screen *rscreen,
+                  const struct pipe_resource *templ);
 
 
 bool r600_prepare_for_dma_blit(struct r600_common_context *rctx,
@@ -191,126 +192,88 @@ static unsigned r600_texture_get_offset(struct r600_texture *rtex, unsigned leve
 static int r600_init_surface(struct r600_common_screen *rscreen,
                             struct radeon_surf *surface,
                             const struct pipe_resource *ptex,
-                            unsigned array_mode,
-                            bool is_flushed_depth)
+                            enum radeon_surf_mode array_mode,
+                            unsigned pitch_in_bytes_override,
+                            unsigned offset,
+                            bool is_imported,
+                            bool is_scanout,
+                            bool is_flushed_depth,
+                            bool tc_compatible_htile)
 {
        const struct util_format_description *desc =
                util_format_description(ptex->format);
        bool is_depth, is_stencil;
+       int r;
+       unsigned i, bpe, flags = 0;
 
        is_depth = util_format_has_depth(desc);
        is_stencil = util_format_has_stencil(desc);
 
-       surface->npix_x = ptex->width0;
-       surface->npix_y = ptex->height0;
-       surface->npix_z = ptex->depth0;
-       surface->blk_w = util_format_get_blockwidth(ptex->format);
-       surface->blk_h = util_format_get_blockheight(ptex->format);
-       surface->blk_d = 1;
-       surface->array_size = 1;
-       surface->last_level = ptex->last_level;
-
        if (rscreen->chip_class >= EVERGREEN && !is_flushed_depth &&
            ptex->format == PIPE_FORMAT_Z32_FLOAT_S8X24_UINT) {
-               surface->bpe = 4; /* stencil is allocated separately on evergreen */
+               bpe = 4; /* stencil is allocated separately on evergreen */
        } else {
-               surface->bpe = util_format_get_blocksize(ptex->format);
+               bpe = util_format_get_blocksize(ptex->format);
                /* align byte per element on dword */
-               if (surface->bpe == 3) {
-                       surface->bpe = 4;
+               if (bpe == 3) {
+                       bpe = 4;
                }
        }
 
-       surface->nsamples = ptex->nr_samples ? ptex->nr_samples : 1;
-       surface->flags = RADEON_SURF_SET(array_mode, MODE);
-
-       switch (ptex->target) {
-       case PIPE_TEXTURE_1D:
-               surface->flags |= RADEON_SURF_SET(RADEON_SURF_TYPE_1D, TYPE);
-               break;
-       case PIPE_TEXTURE_RECT:
-       case PIPE_TEXTURE_2D:
-               surface->flags |= RADEON_SURF_SET(RADEON_SURF_TYPE_2D, TYPE);
-               break;
-       case PIPE_TEXTURE_3D:
-               surface->flags |= RADEON_SURF_SET(RADEON_SURF_TYPE_3D, TYPE);
-               break;
-       case PIPE_TEXTURE_1D_ARRAY:
-               surface->flags |= RADEON_SURF_SET(RADEON_SURF_TYPE_1D_ARRAY, TYPE);
-               surface->array_size = ptex->array_size;
-               break;
-       case PIPE_TEXTURE_CUBE_ARRAY: /* cube array layout like 2d array */
-               assert(ptex->array_size % 6 == 0);
-       case PIPE_TEXTURE_2D_ARRAY:
-               surface->flags |= RADEON_SURF_SET(RADEON_SURF_TYPE_2D_ARRAY, TYPE);
-               surface->array_size = ptex->array_size;
-               break;
-       case PIPE_TEXTURE_CUBE:
-               surface->flags |= RADEON_SURF_SET(RADEON_SURF_TYPE_CUBEMAP, TYPE);
-               break;
-       case PIPE_BUFFER:
-       default:
-               return -EINVAL;
-       }
-
        if (!is_flushed_depth && is_depth) {
-               surface->flags |= RADEON_SURF_ZBUFFER;
+               flags |= RADEON_SURF_ZBUFFER;
 
-               if (is_stencil) {
-                       surface->flags |= RADEON_SURF_SBUFFER |
-                                         RADEON_SURF_HAS_SBUFFER_MIPTREE;
+               if (tc_compatible_htile &&
+                   array_mode == RADEON_SURF_MODE_2D) {
+                       /* TC-compatible HTILE only supports Z32_FLOAT.
+                        * Promote Z16 to Z32. DB->CB copies will convert
+                        * the format for transfers.
+                        */
+                       bpe = 4;
+                       flags |= RADEON_SURF_TC_COMPATIBLE_HTILE;
                }
-       }
-       if (rscreen->chip_class >= SI) {
-               surface->flags |= RADEON_SURF_HAS_TILE_MODE_INDEX;
+
+               if (is_stencil)
+                       flags |= RADEON_SURF_SBUFFER;
        }
 
        if (rscreen->chip_class >= VI &&
            (ptex->flags & R600_RESOURCE_FLAG_DISABLE_DCC ||
             ptex->format == PIPE_FORMAT_R9G9B9E5_FLOAT))
-               surface->flags |= RADEON_SURF_DISABLE_DCC;
+               flags |= RADEON_SURF_DISABLE_DCC;
 
-       if (ptex->bind & PIPE_BIND_SCANOUT) {
+       if (ptex->bind & PIPE_BIND_SCANOUT || is_scanout) {
                /* This should catch bugs in gallium users setting incorrect flags. */
-               assert(surface->nsamples == 1 &&
-                      surface->array_size == 1 &&
-                      surface->npix_z == 1 &&
-                      surface->last_level == 0 &&
-                      !(surface->flags & RADEON_SURF_Z_OR_SBUFFER));
+               assert(ptex->nr_samples <= 1 &&
+                      ptex->array_size == 1 &&
+                      ptex->depth0 == 1 &&
+                      ptex->last_level == 0 &&
+                      !(flags & RADEON_SURF_Z_OR_SBUFFER));
 
-               surface->flags |= RADEON_SURF_SCANOUT;
+               flags |= RADEON_SURF_SCANOUT;
        }
-       return 0;
-}
 
-static int r600_setup_surface(struct pipe_screen *screen,
-                             struct r600_texture *rtex,
-                             unsigned pitch_in_bytes_override,
-                             unsigned offset)
-{
-       struct r600_common_screen *rscreen = (struct r600_common_screen*)screen;
-       unsigned i;
-       int r;
+       if (is_imported)
+               flags |= RADEON_SURF_IMPORTED;
 
-       r = rscreen->ws->surface_init(rscreen->ws, &rtex->surface);
+       r = rscreen->ws->surface_init(rscreen->ws, ptex, flags, bpe,
+                                     array_mode, surface);
        if (r) {
                return r;
        }
 
-       rtex->size = rtex->surface.bo_size;
-
-       if (pitch_in_bytes_override && pitch_in_bytes_override != rtex->surface.level[0].pitch_bytes) {
+       if (pitch_in_bytes_override && pitch_in_bytes_override != surface->level[0].pitch_bytes) {
                /* old ddx on evergreen over estimate alignment for 1d, only 1 level
                 * for those
                 */
-               rtex->surface.level[0].nblk_x = pitch_in_bytes_override / rtex->surface.bpe;
-               rtex->surface.level[0].pitch_bytes = pitch_in_bytes_override;
-               rtex->surface.level[0].slice_size = pitch_in_bytes_override * rtex->surface.level[0].nblk_y;
+               surface->level[0].nblk_x = pitch_in_bytes_override / bpe;
+               surface->level[0].pitch_bytes = pitch_in_bytes_override;
+               surface->level[0].slice_size = pitch_in_bytes_override * surface->level[0].nblk_y;
        }
 
        if (offset) {
-               for (i = 0; i < ARRAY_SIZE(rtex->surface.level); ++i)
-                       rtex->surface.level[i].offset += offset;
+               for (i = 0; i < ARRAY_SIZE(surface->level); ++i)
+                       surface->level[i].offset += offset;
        }
        return 0;
 }
@@ -497,7 +460,12 @@ static void r600_degrade_tile_mode_to_linear(struct r600_common_context *rctx,
        rtex->resource.b.b.bind = templ.bind;
        pb_reference(&rtex->resource.buf, new_tex->resource.buf);
        rtex->resource.gpu_address = new_tex->resource.gpu_address;
+       rtex->resource.vram_usage = new_tex->resource.vram_usage;
+       rtex->resource.gart_usage = new_tex->resource.gart_usage;
+       rtex->resource.bo_size = new_tex->resource.bo_size;
+       rtex->resource.bo_alignment = new_tex->resource.bo_alignment;
        rtex->resource.domains = new_tex->resource.domains;
+       rtex->resource.flags = new_tex->resource.flags;
        rtex->size = new_tex->size;
        rtex->surface = new_tex->surface;
        rtex->non_disp_tiling = new_tex->non_disp_tiling;
@@ -616,35 +584,31 @@ void r600_texture_get_fmask_info(struct r600_common_screen *rscreen,
                                 struct r600_fmask_info *out)
 {
        /* FMASK is allocated like an ordinary texture. */
-       struct radeon_surf fmask = rtex->surface;
+       struct pipe_resource templ = rtex->resource.b.b;
+       struct radeon_surf fmask = {};
+       unsigned flags, bpe;
 
        memset(out, 0, sizeof(*out));
 
-       fmask.bo_alignment = 0;
-       fmask.bo_size = 0;
-       fmask.nsamples = 1;
-       fmask.flags |= RADEON_SURF_FMASK;
-
-       /* Force 2D tiling if it wasn't set. This may occur when creating
-        * FMASK for MSAA resolve on R6xx. On R6xx, the single-sample
-        * destination buffer must have an FMASK too. */
-       fmask.flags = RADEON_SURF_CLR(fmask.flags, MODE);
-       fmask.flags |= RADEON_SURF_SET(RADEON_SURF_MODE_2D, MODE);
+       templ.nr_samples = 1;
+       flags = rtex->surface.flags | RADEON_SURF_FMASK;
 
-       if (rscreen->chip_class >= SI) {
-               fmask.flags |= RADEON_SURF_HAS_TILE_MODE_INDEX;
-       }
+       /* Use the same parameters and tile mode. */
+       fmask.bankw = rtex->surface.bankw;
+       fmask.bankh = rtex->surface.bankh;
+       fmask.mtilea = rtex->surface.mtilea;
+       fmask.tile_split = rtex->surface.tile_split;
 
        switch (nr_samples) {
        case 2:
        case 4:
-               fmask.bpe = 1;
+               bpe = 1;
                if (rscreen->chip_class <= CAYMAN) {
                        fmask.bankh = 4;
                }
                break;
        case 8:
-               fmask.bpe = 4;
+               bpe = 4;
                break;
        default:
                R600_ERR("Invalid sample count for FMASK allocation.\n");
@@ -655,10 +619,11 @@ void r600_texture_get_fmask_info(struct r600_common_screen *rscreen,
         * This can be fixed by writing a separate FMASK allocator specifically
         * for R600-R700 asics. */
        if (rscreen->chip_class <= R700) {
-               fmask.bpe *= 2;
+               bpe *= 2;
        }
 
-       if (rscreen->ws->surface_init(rscreen->ws, &fmask)) {
+       if (rscreen->ws->surface_init(rscreen->ws, &templ, flags, bpe,
+                                     RADEON_SURF_MODE_2D, &fmask)) {
                R600_ERR("Got error in surface_init while allocating FMASK.\n");
                return;
        }
@@ -672,8 +637,8 @@ void r600_texture_get_fmask_info(struct r600_common_screen *rscreen,
        out->tile_mode_index = fmask.tiling_index[0];
        out->pitch_in_pixels = fmask.level[0].nblk_x;
        out->bank_height = fmask.bankh;
-       out->alignment = MAX2(256, fmask.bo_alignment);
-       out->size = fmask.bo_size;
+       out->alignment = MAX2(256, fmask.surf_alignment);
+       out->size = fmask.surf_size;
 }
 
 static void r600_texture_allocate_fmask(struct r600_common_screen *rscreen,
@@ -704,8 +669,8 @@ void r600_texture_get_cmask_info(struct r600_common_screen *rscreen,
        unsigned macro_tile_width = util_next_power_of_two(sqrt_pixels_per_macro_tile);
        unsigned macro_tile_height = pixels_per_macro_tile / macro_tile_width;
 
-       unsigned pitch_elements = align(rtex->surface.npix_x, macro_tile_width);
-       unsigned height = align(rtex->surface.npix_y, macro_tile_height);
+       unsigned pitch_elements = align(rtex->resource.b.b.width0, macro_tile_width);
+       unsigned height = align(rtex->resource.b.b.height0, macro_tile_height);
 
        unsigned base_align = num_pipes * pipe_interleave_bytes;
        unsigned slice_bytes =
@@ -756,8 +721,8 @@ static void si_texture_get_cmask_info(struct r600_common_screen *rscreen,
 
        unsigned base_align = num_pipes * pipe_interleave_bytes;
 
-       unsigned width = align(rtex->surface.npix_x, cl_width*8);
-       unsigned height = align(rtex->surface.npix_y, cl_height*8);
+       unsigned width = align(rtex->resource.b.b.width0, cl_width*8);
+       unsigned height = align(rtex->resource.b.b.height0, cl_height*8);
        unsigned slice_elements = (width * height) / (8*8);
 
        /* Each element of CMASK is a nibble. */
@@ -809,8 +774,9 @@ static void r600_texture_alloc_cmask_separate(struct r600_common_screen *rscreen
        }
 
        rtex->cmask_buffer = (struct r600_resource *)
-               pipe_buffer_create(&rscreen->b, PIPE_BIND_CUSTOM,
-                                  PIPE_USAGE_DEFAULT, rtex->cmask.size);
+               r600_aligned_buffer_create(&rscreen->b, 0, PIPE_USAGE_DEFAULT,
+                                          rtex->cmask.size,
+                                          rtex->cmask.alignment);
        if (rtex->cmask_buffer == NULL) {
                rtex->cmask.size = 0;
                return;
@@ -827,28 +793,30 @@ static void r600_texture_alloc_cmask_separate(struct r600_common_screen *rscreen
        p_atomic_inc(&rscreen->compressed_colortex_counter);
 }
 
-static unsigned r600_texture_get_htile_size(struct r600_common_screen *rscreen,
-                                           struct r600_texture *rtex)
+static void r600_texture_get_htile_size(struct r600_common_screen *rscreen,
+                                       struct r600_texture *rtex)
 {
        unsigned cl_width, cl_height, width, height;
        unsigned slice_elements, slice_bytes, pipe_interleave_bytes, base_align;
        unsigned num_pipes = rscreen->info.num_tile_pipes;
 
+       rtex->surface.htile_size = 0;
+
        if (rscreen->chip_class <= EVERGREEN &&
            rscreen->info.drm_major == 2 && rscreen->info.drm_minor < 26)
-               return 0;
+               return;
 
        /* HW bug on R6xx. */
        if (rscreen->chip_class == R600 &&
-           (rtex->surface.level[0].npix_x > 7680 ||
-            rtex->surface.level[0].npix_y > 7680))
-               return 0;
+           (rtex->resource.b.b.width0 > 7680 ||
+            rtex->resource.b.b.height0 > 7680))
+               return;
 
        /* HTILE is broken with 1D tiling on old kernels and CIK. */
        if (rscreen->chip_class >= CIK &&
            rtex->surface.level[0].mode == RADEON_SURF_MODE_1D &&
            rscreen->info.drm_major == 2 && rscreen->info.drm_minor < 38)
-               return 0;
+               return;
 
        /* Overalign HTILE on P2 configs to work around GPU hangs in
         * piglit/depthstencil-render-miplevels 585.
@@ -883,11 +851,11 @@ static unsigned r600_texture_get_htile_size(struct r600_common_screen *rscreen,
                break;
        default:
                assert(0);
-               return 0;
+               return;
        }
 
-       width = align(rtex->surface.npix_x, cl_width * 8);
-       height = align(rtex->surface.npix_y, cl_height * 8);
+       width = align(rtex->resource.b.b.width0, cl_width * 8);
+       height = align(rtex->resource.b.b.height0, cl_height * 8);
 
        slice_elements = (width * height) / (8 * 8);
        slice_bytes = slice_elements * 4;
@@ -895,33 +863,40 @@ static unsigned r600_texture_get_htile_size(struct r600_common_screen *rscreen,
        pipe_interleave_bytes = rscreen->info.pipe_interleave_bytes;
        base_align = num_pipes * pipe_interleave_bytes;
 
-       rtex->htile.pitch = width;
-       rtex->htile.height = height;
-       rtex->htile.xalign = cl_width * 8;
-       rtex->htile.yalign = cl_height * 8;
-
-       return (util_max_layer(&rtex->resource.b.b, 0) + 1) *
+       rtex->surface.htile_alignment = base_align;
+       rtex->surface.htile_size =
+               (util_max_layer(&rtex->resource.b.b, 0) + 1) *
                align(slice_bytes, base_align);
 }
 
 static void r600_texture_allocate_htile(struct r600_common_screen *rscreen,
                                        struct r600_texture *rtex)
 {
-       unsigned htile_size = r600_texture_get_htile_size(rscreen, rtex);
+       uint32_t clear_value;
+
+       if (rtex->tc_compatible_htile) {
+               clear_value = 0x0000030F;
+       } else {
+               r600_texture_get_htile_size(rscreen, rtex);
+               clear_value = 0;
+       }
 
-       if (!htile_size)
+       if (!rtex->surface.htile_size)
                return;
 
        rtex->htile_buffer = (struct r600_resource*)
-                            pipe_buffer_create(&rscreen->b, PIPE_BIND_CUSTOM,
-                                               PIPE_USAGE_DEFAULT, htile_size);
+               r600_aligned_buffer_create(&rscreen->b, PIPE_BIND_CUSTOM,
+                                          PIPE_USAGE_DEFAULT,
+                                          rtex->surface.htile_size,
+                                          rtex->surface.htile_alignment);
        if (rtex->htile_buffer == NULL) {
                /* this is not a fatal error as we can still keep rendering
                 * without htile buffer */
                R600_ERR("Failed to create buffer object for htile buffer.\n");
        } else {
-               r600_screen_clear_buffer(rscreen, &rtex->htile_buffer->b.b, 0,
-                                        htile_size, 0, R600_COHERENCY_NONE);
+               r600_screen_clear_buffer(rscreen, &rtex->htile_buffer->b.b,
+                                        0, rtex->surface.htile_size,
+                                        clear_value, R600_COHERENCY_NONE);
        }
 }
 
@@ -930,18 +905,18 @@ void r600_print_texture_info(struct r600_texture *rtex, FILE *f)
        int i;
 
        fprintf(f, "  Info: npix_x=%u, npix_y=%u, npix_z=%u, blk_w=%u, "
-               "blk_h=%u, blk_d=%u, array_size=%u, last_level=%u, "
+               "blk_h=%u, array_size=%u, last_level=%u, "
                "bpe=%u, nsamples=%u, flags=0x%x, %s\n",
-               rtex->surface.npix_x, rtex->surface.npix_y,
-               rtex->surface.npix_z, rtex->surface.blk_w,
-               rtex->surface.blk_h, rtex->surface.blk_d,
-               rtex->surface.array_size, rtex->surface.last_level,
-               rtex->surface.bpe, rtex->surface.nsamples,
+               rtex->resource.b.b.width0, rtex->resource.b.b.height0,
+               rtex->resource.b.b.depth0, rtex->surface.blk_w,
+               rtex->surface.blk_h,
+               rtex->resource.b.b.array_size, rtex->resource.b.b.last_level,
+               rtex->surface.bpe, rtex->resource.b.b.nr_samples,
                rtex->surface.flags, util_format_short_name(rtex->resource.b.b.format));
 
-       fprintf(f, "  Layout: size=%"PRIu64", alignment=%"PRIu64", bankw=%u, "
+       fprintf(f, "  Layout: size=%"PRIu64", alignment=%u, bankw=%u, "
                "bankh=%u, nbanks=%u, mtilea=%u, tilesplit=%u, pipeconfig=%u, scanout=%u\n",
-               rtex->surface.bo_size, rtex->surface.bo_alignment, rtex->surface.bankw,
+               rtex->surface.surf_size, rtex->surface.surf_alignment, rtex->surface.bankw,
                rtex->surface.bankh, rtex->surface.num_banks, rtex->surface.mtilea,
                rtex->surface.tile_split, rtex->surface.pipe_config,
                (rtex->surface.flags & RADEON_SURF_SCANOUT) != 0);
@@ -961,17 +936,16 @@ void r600_print_texture_info(struct r600_texture *rtex, FILE *f)
                        rtex->cmask.yalign, rtex->cmask.slice_tile_max);
 
        if (rtex->htile_buffer)
-               fprintf(f, "  HTile: size=%u, alignment=%u, pitch=%u, height=%u, "
-                       "xalign=%u, yalign=%u\n",
+               fprintf(f, "  HTile: size=%u, alignment=%u, TC_compatible = %u\n",
                        rtex->htile_buffer->b.b.width0,
-                       rtex->htile_buffer->buf->alignment, rtex->htile.pitch,
-                       rtex->htile.height, rtex->htile.xalign, rtex->htile.yalign);
+                       rtex->htile_buffer->buf->alignment,
+                       rtex->tc_compatible_htile);
 
        if (rtex->dcc_offset) {
-               fprintf(f, "  DCC: offset=%"PRIu64", size=%"PRIu64", alignment=%"PRIu64"\n",
+               fprintf(f, "  DCC: offset=%"PRIu64", size=%"PRIu64", alignment=%u\n",
                        rtex->dcc_offset, rtex->surface.dcc_size,
                        rtex->surface.dcc_alignment);
-               for (i = 0; i <= rtex->surface.last_level; i++)
+               for (i = 0; i <= rtex->resource.b.b.last_level; i++)
                        fprintf(f, "  DCCLevel[%i]: enabled=%u, offset=%"PRIu64", "
                                "fast_clear_size=%"PRIu64"\n",
                                i, rtex->surface.level[i].dcc_enabled,
@@ -979,10 +953,10 @@ void r600_print_texture_info(struct r600_texture *rtex, FILE *f)
                                rtex->surface.level[i].dcc_fast_clear_size);
        }
 
-       for (i = 0; i <= rtex->surface.last_level; i++)
+       for (i = 0; i <= rtex->resource.b.b.last_level; i++)
                fprintf(f, "  Level[%i]: offset=%"PRIu64", slice_size=%"PRIu64", "
                        "npix_x=%u, npix_y=%u, npix_z=%u, nblk_x=%u, nblk_y=%u, "
-                       "nblk_z=%u, pitch_bytes=%u, mode=%u\n",
+                       "pitch_bytes=%u, mode=%u, tiling_index = %u\n",
                        i, rtex->surface.level[i].offset,
                        rtex->surface.level[i].slice_size,
                        u_minify(rtex->resource.b.b.width0, i),
@@ -990,18 +964,18 @@ void r600_print_texture_info(struct r600_texture *rtex, FILE *f)
                        u_minify(rtex->resource.b.b.depth0, i),
                        rtex->surface.level[i].nblk_x,
                        rtex->surface.level[i].nblk_y,
-                       rtex->surface.level[i].nblk_z,
                        rtex->surface.level[i].pitch_bytes,
-                       rtex->surface.level[i].mode);
+                       rtex->surface.level[i].mode,
+                       rtex->surface.tiling_index[i]);
 
        if (rtex->surface.flags & RADEON_SURF_SBUFFER) {
                fprintf(f, "  StencilLayout: tilesplit=%u\n",
                        rtex->surface.stencil_tile_split);
-               for (i = 0; i <= rtex->surface.last_level; i++) {
+               for (i = 0; i <= rtex->resource.b.b.last_level; i++) {
                        fprintf(f, "  StencilLevel[%i]: offset=%"PRIu64", "
                                "slice_size=%"PRIu64", npix_x=%u, "
                                "npix_y=%u, npix_z=%u, nblk_x=%u, nblk_y=%u, "
-                               "nblk_z=%u, pitch_bytes=%u, mode=%u\n",
+                               "pitch_bytes=%u, mode=%u, tiling_index = %u\n",
                                i, rtex->surface.stencil_level[i].offset,
                                rtex->surface.stencil_level[i].slice_size,
                                u_minify(rtex->resource.b.b.width0, i),
@@ -1009,9 +983,9 @@ void r600_print_texture_info(struct r600_texture *rtex, FILE *f)
                                u_minify(rtex->resource.b.b.depth0, i),
                                rtex->surface.stencil_level[i].nblk_x,
                                rtex->surface.stencil_level[i].nblk_y,
-                               rtex->surface.stencil_level[i].nblk_z,
                                rtex->surface.stencil_level[i].pitch_bytes,
-                               rtex->surface.stencil_level[i].mode);
+                               rtex->surface.stencil_level[i].mode,
+                               rtex->surface.stencil_tiling_index[i]);
                }
        }
 }
@@ -1020,8 +994,6 @@ void r600_print_texture_info(struct r600_texture *rtex, FILE *f)
 static struct r600_texture *
 r600_texture_create_object(struct pipe_screen *screen,
                           const struct pipe_resource *base,
-                          unsigned pitch_in_bytes_override,
-                          unsigned offset,
                           struct pb_buffer *buf,
                           struct radeon_surf *surface)
 {
@@ -1035,6 +1007,7 @@ r600_texture_create_object(struct pipe_screen *screen,
 
        resource = &rtex->resource;
        resource->b.b = *base;
+       resource->b.b.next = NULL;
        resource->b.vtbl = &r600_texture_vtbl;
        pipe_reference_init(&resource->b.b.reference, 1);
        resource->b.b.screen = screen;
@@ -1043,10 +1016,17 @@ r600_texture_create_object(struct pipe_screen *screen,
        rtex->is_depth = util_format_has_depth(util_format_description(rtex->resource.b.b.format));
 
        rtex->surface = *surface;
-       if (r600_setup_surface(screen, rtex, pitch_in_bytes_override, offset)) {
-               FREE(rtex);
-               return NULL;
-       }
+       rtex->size = rtex->surface.surf_size;
+
+       rtex->tc_compatible_htile = rtex->surface.htile_size != 0;
+       assert(!!(rtex->surface.flags & RADEON_SURF_TC_COMPATIBLE_HTILE) ==
+              rtex->tc_compatible_htile);
+
+       /* TC-compatible HTILE only supports Z32_FLOAT. */
+       if (rtex->tc_compatible_htile)
+               rtex->db_render_format = PIPE_FORMAT_Z32_FLOAT;
+       else
+               rtex->db_render_format = base->format;
 
        /* Tiled depth textures utilize the non-displayable tile order.
         * This must be done after r600_setup_surface.
@@ -1110,7 +1090,9 @@ r600_texture_create_object(struct pipe_screen *screen,
        /* Now create the backing buffer. */
        if (!buf) {
                r600_init_resource_fields(rscreen, resource, rtex->size,
-                                         rtex->surface.bo_alignment);
+                                         rtex->surface.surf_alignment);
+
+               resource->flags |= RADEON_FLAG_HANDLE;
 
                if (!r600_alloc_resource(rscreen, resource)) {
                        FREE(rtex);
@@ -1119,7 +1101,13 @@ r600_texture_create_object(struct pipe_screen *screen,
        } else {
                resource->buf = buf;
                resource->gpu_address = rscreen->ws->buffer_get_virtual_address(resource->buf);
+               resource->bo_size = buf->size;
+               resource->bo_alignment = buf->alignment;
                resource->domains = rscreen->ws->buffer_get_initial_domain(resource->buf);
+               if (resource->domains & RADEON_DOMAIN_VRAM)
+                       resource->vram_usage = buf->size;
+               else if (resource->domains & RADEON_DOMAIN_GTT)
+                       resource->gart_usage = buf->size;
        }
 
        if (rtex->cmask.size) {
@@ -1158,8 +1146,9 @@ r600_texture_create_object(struct pipe_screen *screen,
        return rtex;
 }
 
-static unsigned r600_choose_tiling(struct r600_common_screen *rscreen,
-                                  const struct pipe_resource *templ)
+static enum radeon_surf_mode
+r600_choose_tiling(struct r600_common_screen *rscreen,
+                  const struct pipe_resource *templ)
 {
        const struct util_format_description *desc = util_format_description(templ->format);
        bool force_tiling = templ->flags & R600_RESOURCE_FLAG_FORCE_TILING;
@@ -1227,20 +1216,27 @@ struct pipe_resource *r600_texture_create(struct pipe_screen *screen,
 {
        struct r600_common_screen *rscreen = (struct r600_common_screen*)screen;
        struct radeon_surf surface = {0};
+       bool is_flushed_depth = templ->flags & R600_RESOURCE_FLAG_FLUSHED_DEPTH;
+       bool tc_compatible_htile =
+               rscreen->chip_class >= VI &&
+               (templ->flags & PIPE_RESOURCE_FLAG_TEXTURING_MORE_LIKELY) &&
+               !(rscreen->debug_flags & DBG_NO_HYPERZ) &&
+               !is_flushed_depth &&
+               templ->nr_samples <= 1 && /* TC-compat HTILE is less efficient with MSAA */
+               util_format_is_depth_or_stencil(templ->format);
+
        int r;
 
        r = r600_init_surface(rscreen, &surface, templ,
-                             r600_choose_tiling(rscreen, templ),
-                             templ->flags & R600_RESOURCE_FLAG_FLUSHED_DEPTH);
-       if (r) {
-               return NULL;
-       }
-       r = rscreen->ws->surface_best(rscreen->ws, &surface);
+                             r600_choose_tiling(rscreen, templ), 0, 0,
+                             false, false, is_flushed_depth,
+                             tc_compatible_htile);
        if (r) {
                return NULL;
        }
-       return (struct pipe_resource *)r600_texture_create_object(screen, templ, 0,
-                                                                 0, NULL, &surface);
+
+       return (struct pipe_resource *)
+              r600_texture_create_object(screen, templ, NULL, &surface);
 }
 
 static struct pipe_resource *r600_texture_from_handle(struct pipe_screen *screen,
@@ -1282,16 +1278,13 @@ static struct pipe_resource *r600_texture_from_handle(struct pipe_screen *screen
        else
                array_mode = RADEON_SURF_MODE_LINEAR_ALIGNED;
 
-       r = r600_init_surface(rscreen, &surface, templ, array_mode, false);
+       r = r600_init_surface(rscreen, &surface, templ, array_mode, stride,
+                             offset, true, metadata.scanout, false, false);
        if (r) {
                return NULL;
        }
 
-       if (metadata.scanout)
-               surface.flags |= RADEON_SURF_SCANOUT;
-
-       rtex = r600_texture_create_object(screen, templ, stride,
-                                         offset, buf, &surface);
+       rtex = r600_texture_create_object(screen, templ, buf, &surface);
        if (!rtex)
                return NULL;
 
@@ -1347,6 +1340,7 @@ bool r600_init_flushed_depth_texture(struct pipe_context *ctx,
                }
        }
 
+       memset(&resource, 0, sizeof(resource));
        resource.target = texture->target;
        resource.format = pipe_format;
        resource.width0 = texture->width0;
@@ -2094,7 +2088,7 @@ static void vi_separate_dcc_try_enable(struct r600_common_context *rctx,
        if (!tex->resource.is_shared ||
            !(tex->resource.external_usage & PIPE_HANDLE_USAGE_EXPLICIT_FLUSH) ||
            tex->resource.b.b.target != PIPE_TEXTURE_2D ||
-           tex->surface.last_level > 0 ||
+           tex->resource.b.b.last_level > 0 ||
            !tex->surface.dcc_size)
                return;
 
@@ -2350,12 +2344,12 @@ static void si_set_optimal_micro_tile_mode(struct r600_common_screen *rscreen,
                                           struct r600_texture *rtex)
 {
        if (rtex->resource.is_shared ||
-           rtex->surface.nsamples <= 1 ||
+           rtex->resource.b.b.nr_samples <= 1 ||
            rtex->surface.micro_tile_mode == rtex->last_msaa_resolve_target_micro_mode)
                return;
 
        assert(rtex->surface.level[0].mode == RADEON_SURF_MODE_2D);
-       assert(rtex->surface.last_level == 0);
+       assert(rtex->resource.b.b.last_level == 0);
 
        /* These magic numbers were copied from addrlib. It doesn't use any
         * definitions for them either. They are all 2D_TILED_THIN1 modes with
@@ -2380,29 +2374,29 @@ static void si_set_optimal_micro_tile_mode(struct r600_common_screen *rscreen,
                switch (rtex->last_msaa_resolve_target_micro_mode) {
                case 0: /* displayable */
                        switch (rtex->surface.bpe) {
-                       case 8:
+                       case 1:
                             rtex->surface.tiling_index[0] = 10;
                             break;
-                       case 16:
+                       case 2:
                             rtex->surface.tiling_index[0] = 11;
                             break;
-                       default: /* 32, 64 */
+                       default: /* 4, 8 */
                             rtex->surface.tiling_index[0] = 12;
                             break;
                        }
                        break;
                case 1: /* thin */
                        switch (rtex->surface.bpe) {
-                       case 8:
+                       case 1:
                                 rtex->surface.tiling_index[0] = 14;
                                 break;
-                       case 16:
+                       case 2:
                                 rtex->surface.tiling_index[0] = 15;
                                 break;
-                       case 32:
+                       case 4:
                                 rtex->surface.tiling_index[0] = 16;
                                 break;
-                       default: /* 64, 128 */
+                       default: /* 8, 16 */
                                 rtex->surface.tiling_index[0] = 17;
                                 break;
                        }