r600g: Add $(DEFINES) to AM_CXXFLAGS
[mesa.git] / src / gallium / drivers / r600 / r600_texture.c
index 2df390d50510c9efaf8e6a95c390e138cc72a5f2..7f5752d81e6c401d719fd264d1f86f06fb319f57 100644 (file)
 #include "util/u_format_s3tc.h"
 #include "util/u_memory.h"
 
+
 /* Copy from a full GPU texture to a transfer's staging one. */
 static void r600_copy_to_staging_texture(struct pipe_context *ctx, struct r600_transfer *rtransfer)
 {
+       struct r600_context *rctx = (struct r600_context*)ctx;
        struct pipe_transfer *transfer = (struct pipe_transfer*)rtransfer;
-       struct pipe_resource *texture = transfer->resource;
-
-       ctx->resource_copy_region(ctx, &rtransfer->staging->b.b,
-                               0, 0, 0, 0, texture, transfer->level,
-                               &transfer->box);
+       struct pipe_resource *dst = &rtransfer->staging->b.b;
+       struct pipe_resource *src = transfer->resource;
+
+       if (src->nr_samples <= 1) {
+               if (!rctx->screen->dma_blit(ctx, dst, 0, 0, 0, 0,
+                                           src, transfer->level,
+                                           &transfer->box)) {
+                       /* async dma could not be use */
+                       ctx->resource_copy_region(ctx, dst, 0, 0, 0, 0,
+                                                 src, transfer->level, &transfer->box);
+               }
+       } else {
+               /* Resolve the resource. */
+               struct pipe_blit_info blit;
+
+               memset(&blit, 0, sizeof(blit));
+               blit.src.resource = src;
+               blit.src.format = src->format;
+               blit.src.level = transfer->level;
+               blit.src.box = transfer->box;
+               blit.dst.resource = dst;
+               blit.dst.format = dst->format;
+               blit.dst.box.width = transfer->box.width;
+               blit.dst.box.height = transfer->box.height;
+               blit.dst.box.depth = transfer->box.depth;
+               blit.mask = PIPE_MASK_RGBA;
+               blit.filter = PIPE_TEX_FILTER_NEAREST;
+
+               ctx->blit(ctx, &blit);
+       }
 }
 
-
 /* Copy from a transfer's staging texture to a full GPU one. */
 static void r600_copy_from_staging_texture(struct pipe_context *ctx, struct r600_transfer *rtransfer)
 {
+       struct r600_context *rctx = (struct r600_context*)ctx;
        struct pipe_transfer *transfer = (struct pipe_transfer*)rtransfer;
        struct pipe_resource *texture = transfer->resource;
        struct pipe_box sbox;
 
-       u_box_origin_2d(transfer->box.width, transfer->box.height, &sbox);
+       u_box_3d(0, 0, 0, transfer->box.width, transfer->box.height, transfer->box.depth, &sbox);
 
-       ctx->resource_copy_region(ctx, texture, transfer->level,
-                                 transfer->box.x, transfer->box.y, transfer->box.z,
-                                 &rtransfer->staging->b.b,
-                                 0, &sbox);
+       if (!rctx->screen->dma_blit(ctx, texture, transfer->level,
+                                   transfer->box.x, transfer->box.y, transfer->box.z,
+                                   &rtransfer->staging->b.b, 0, &sbox)) {
+               /* async dma could not be use */
+               ctx->resource_copy_region(ctx, texture, transfer->level,
+                                         transfer->box.x, transfer->box.y, transfer->box.z,
+                                         &rtransfer->staging->b.b,
+                                         0, &sbox);
+       }
 }
 
 unsigned r600_texture_get_offset(struct r600_texture *rtex,
@@ -69,7 +101,7 @@ static int r600_init_surface(struct r600_screen *rscreen,
                             struct radeon_surface *surface,
                             const struct pipe_resource *ptex,
                             unsigned array_mode,
-                            bool is_transfer, bool is_flushed_depth)
+                            bool is_flushed_depth)
 {
        const struct util_format_description *desc =
                util_format_description(ptex->format);
@@ -87,8 +119,7 @@ static int r600_init_surface(struct r600_screen *rscreen,
        surface->array_size = 1;
        surface->last_level = ptex->last_level;
 
-       if (rscreen->chip_class >= EVERGREEN &&
-           !is_transfer && !is_flushed_depth &&
+       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 */
        } else {
@@ -133,6 +164,7 @@ static int r600_init_surface(struct r600_screen *rscreen,
                surface->array_size = ptex->array_size;
                break;
        case PIPE_TEXTURE_2D_ARRAY:
+       case PIPE_TEXTURE_CUBE_ARRAY: /* cube array layout like 2d layout for now */
                surface->flags |= RADEON_SURF_SET(RADEON_SURF_TYPE_2D_ARRAY, TYPE);
                surface->array_size = ptex->array_size;
                break;
@@ -147,11 +179,12 @@ static int r600_init_surface(struct r600_screen *rscreen,
                surface->flags |= RADEON_SURF_SCANOUT;
        }
 
-       if (!is_transfer && !is_flushed_depth && is_depth) {
+       if (!is_flushed_depth && is_depth) {
                surface->flags |= RADEON_SURF_ZBUFFER;
 
                if (is_stencil) {
-                       surface->flags |= RADEON_SURF_SBUFFER;
+                       surface->flags |= RADEON_SURF_SBUFFER |
+                                          RADEON_SURF_HAS_SBUFFER_MIPTREE;
                }
        }
        return 0;
@@ -179,7 +212,8 @@ static int r600_setup_surface(struct pipe_screen *screen,
                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;
                if (rtex->surface.flags & RADEON_SURF_SBUFFER) {
-                       rtex->surface.stencil_offset = rtex->surface.level[0].slice_size;
+                       rtex->surface.stencil_offset =
+                       rtex->surface.stencil_level[0].offset = rtex->surface.level[0].slice_size;
                }
        }
        for (i = 0; i <= ptex->last_level; i++) {
@@ -236,6 +270,7 @@ static void r600_texture_destroy(struct pipe_screen *screen,
        if (rtex->flushed_depth_texture)
                pipe_resource_reference((struct pipe_resource **)&rtex->flushed_depth_texture, NULL);
 
+        pipe_resource_reference((struct pipe_resource**)&rtex->htile, NULL);
        pb_reference(&resource->buf, NULL);
        FREE(rtex);
 }
@@ -302,7 +337,6 @@ static void r600_texture_allocate_fmask(struct r600_screen *rscreen,
        r600_texture_get_fmask_info(rscreen, rtex,
                                    rtex->resource.b.b.nr_samples, &fmask);
 
-       /* Reserve space for FMASK while converting bits back to bytes. */
        rtex->fmask_bank_height = fmask.bank_height;
        rtex->fmask_offset = align(rtex->size, fmask.alignment);
        rtex->fmask_size = fmask.size;
@@ -365,12 +399,13 @@ static void r600_texture_allocate_cmask(struct r600_screen *rscreen,
 #endif
 }
 
+DEBUG_GET_ONCE_BOOL_OPTION(print_texdepth, "R600_PRINT_TEXDEPTH", FALSE);
+
 static struct r600_texture *
 r600_texture_create_object(struct pipe_screen *screen,
                           const struct pipe_resource *base,
                           unsigned pitch_in_bytes_override,
                           struct pb_buffer *buf,
-                          boolean alloc_bo,
                           struct radeon_surface *surface)
 {
        struct r600_texture *rtex;
@@ -400,7 +435,7 @@ r600_texture_create_object(struct pipe_screen *screen,
                return NULL;
        }
 
-       if (base->nr_samples > 1 && !rtex->is_depth && alloc_bo) {
+       if (base->nr_samples > 1 && !rtex->is_depth && !buf) {
                r600_texture_allocate_cmask(rscreen, rtex);
                r600_texture_allocate_fmask(rscreen, rtex);
        }
@@ -411,16 +446,57 @@ r600_texture_create_object(struct pipe_screen *screen,
                return NULL;
        }
 
+       /* Tiled depth textures utilize the non-displayable tile order. */
+       rtex->non_disp_tiling = rtex->is_depth && rtex->surface.level[0].mode >= RADEON_SURF_MODE_1D;
+
+       /* only enable hyperz for PIPE_TEXTURE_2D not for PIPE_TEXTURE_2D_ARRAY
+        * Thought it might still be interessting to use hyperz for texture
+        * array without using fast clear features
+        */
+       rtex->htile = NULL;
+       if (!(base->flags & (R600_RESOURCE_FLAG_TRANSFER | R600_RESOURCE_FLAG_FLUSHED_DEPTH)) &&
+           util_format_is_depth_or_stencil(base->format) &&
+           rscreen->use_hyperz &&
+           base->target == PIPE_TEXTURE_2D &&
+           rtex->surface.level[0].nblk_x >= 32 &&
+           rtex->surface.level[0].nblk_y >= 32) {
+               unsigned sw = rtex->surface.level[0].nblk_x * rtex->surface.blk_w;
+               unsigned sh = rtex->surface.level[0].nblk_y * rtex->surface.blk_h;
+               unsigned htile_size;
+               unsigned npipes = rscreen->info.r600_num_tile_pipes;
+
+               /* this alignment and htile size only apply to linear htile buffer */
+               sw = align(sw, 16 << 3);
+               sh = align(sh, npipes << 3);
+               htile_size = (sw >> 3) * (sh >> 3) * 4;
+               /* must be aligned with 2K * npipes */
+               htile_size = align(htile_size, (2 << 10) * npipes);
+
+               rtex->htile = (struct r600_resource*)pipe_buffer_create(&rscreen->screen, PIPE_BIND_CUSTOM,
+                                                                       PIPE_USAGE_STATIC, htile_size);
+               if (rtex->htile == NULL) {
+                       /* this is not a fatal error as we can still keep rendering
+                        * without htile buffer
+                        */
+                       R600_ERR("r600: failed to create bo for htile buffers\n");
+               } else {
+                       void *ptr;
+                       ptr = rscreen->ws->buffer_map(rtex->htile->cs_buf, NULL, PIPE_TRANSFER_WRITE);
+                       memset(ptr, 0x0, htile_size);
+                       rscreen->ws->buffer_unmap(rtex->htile->cs_buf);
+               }
+       }
+
        /* Now create the backing buffer. */
-       if (!buf && alloc_bo) {
+       if (!buf) {
                unsigned base_align = rtex->surface.bo_alignment;
                unsigned usage = R600_TEX_IS_TILED(rtex, 0) ? PIPE_USAGE_STATIC : base->usage;
 
-               if (!r600_init_resource(rscreen, resource, rtex->size, base_align, base->bind, usage)) {
+               if (!r600_init_resource(rscreen, resource, rtex->size, base_align, FALSE, usage)) {
                        FREE(rtex);
                        return NULL;
                }
-       } else if (buf) {
+       } else {
                /* This is usually the window framebuffer. We want it in VRAM, always. */
                resource->buf = buf;
                resource->cs_buf = rscreen->ws->buffer_get_cs_handle(buf);
@@ -433,6 +509,52 @@ r600_texture_create_object(struct pipe_screen *screen,
                memset(ptr + rtex->cmask_offset, 0xCC, rtex->cmask_size);
                rscreen->ws->buffer_unmap(resource->cs_buf);
        }
+
+       if (debug_get_option_print_texdepth() && rtex->is_depth && rtex->non_disp_tiling) {
+               printf("Texture: npix_x=%u, npix_y=%u, npix_z=%u, blk_w=%u, "
+                      "blk_h=%u, blk_d=%u, array_size=%u, last_level=%u, "
+                      "bpe=%u, nsamples=%u, flags=%u\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->surface.flags);
+               if (rtex->surface.flags & RADEON_SURF_ZBUFFER) {
+                       for (int i = 0; i <= rtex->surface.last_level; i++) {
+                               printf("  Z %i: offset=%llu, slice_size=%llu, npix_x=%u, "
+                                      "npix_y=%u, npix_z=%u, nblk_x=%u, nblk_y=%u, "
+                                      "nblk_z=%u, pitch_bytes=%u, mode=%u\n",
+                                      i, (unsigned long long)rtex->surface.level[i].offset,
+                                      (unsigned long long)rtex->surface.level[i].slice_size,
+                                      u_minify(rtex->resource.b.b.width0, i),
+                                      u_minify(rtex->resource.b.b.height0, i),
+                                      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);
+                       }
+               }
+               if (rtex->surface.flags & RADEON_SURF_SBUFFER) {
+                       for (int i = 0; i <= rtex->surface.last_level; i++) {
+                               printf("  S %i: offset=%llu, slice_size=%llu, npix_x=%u, "
+                                      "npix_y=%u, npix_z=%u, nblk_x=%u, nblk_y=%u, "
+                                      "nblk_z=%u, pitch_bytes=%u, mode=%u\n",
+                                      i, (unsigned long long)rtex->surface.stencil_level[i].offset,
+                                      (unsigned long long)rtex->surface.stencil_level[i].slice_size,
+                                       u_minify(rtex->resource.b.b.width0, i),
+                                      u_minify(rtex->resource.b.b.height0, i),
+                                      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);
+                       }
+               }
+       }
        return rtex;
 }
 
@@ -441,25 +563,33 @@ struct pipe_resource *r600_texture_create(struct pipe_screen *screen,
 {
        struct r600_screen *rscreen = (struct r600_screen*)screen;
        struct radeon_surface surface;
-       unsigned array_mode = 0;
+       const struct util_format_description *desc = util_format_description(templ->format);
+       unsigned array_mode;
        int r;
 
-       if (!(templ->flags & R600_RESOURCE_FLAG_TRANSFER)) {
-               if (!(templ->bind & PIPE_BIND_SCANOUT) &&
-                   templ->usage != PIPE_USAGE_STAGING &&
-                   templ->usage != PIPE_USAGE_STREAM) {
+       /* Default tiling mode for staging textures. */
+       array_mode = V_038000_ARRAY_LINEAR_ALIGNED;
+
+       /* Tiling doesn't work with the 422 (SUBSAMPLED) formats. That's not an issue,
+        * because 422 formats are used for videos, which prefer linear buffers
+        * for fast uploads anyway. */
+       if (!(templ->flags & R600_RESOURCE_FLAG_TRANSFER) &&
+           desc->layout != UTIL_FORMAT_LAYOUT_SUBSAMPLED) {
+               if (templ->flags & R600_RESOURCE_FLAG_FORCE_TILING) {
+                       array_mode = V_038000_ARRAY_2D_TILED_THIN1;
+               } else if (!(templ->bind & PIPE_BIND_SCANOUT) &&
+                          templ->usage != PIPE_USAGE_STAGING &&
+                          templ->usage != PIPE_USAGE_STREAM &&
+                          templ->target != PIPE_TEXTURE_1D &&
+                          templ->target != PIPE_TEXTURE_1D_ARRAY &&
+                          templ->height0 > 3) {
                        array_mode = V_038000_ARRAY_2D_TILED_THIN1;
                } else if (util_format_is_compressed(templ->format)) {
                        array_mode = V_038000_ARRAY_1D_TILED_THIN1;
                }
        }
 
-       /* XXX tiling is broken for the 422 formats */
-       if (util_format_description(templ->format)->layout == UTIL_FORMAT_LAYOUT_SUBSAMPLED)
-               array_mode = V_038000_ARRAY_LINEAR_ALIGNED;
-
        r = r600_init_surface(rscreen, &surface, templ, array_mode,
-                             templ->flags & R600_RESOURCE_FLAG_TRANSFER,
                              templ->flags & R600_RESOURCE_FLAG_FLUSHED_DEPTH);
        if (r) {
                return NULL;
@@ -469,7 +599,7 @@ struct pipe_resource *r600_texture_create(struct pipe_screen *screen,
                return NULL;
        }
        return (struct pipe_resource *)r600_texture_create_object(screen, templ,
-                                                                 0, NULL, TRUE, &surface);
+                                                                 0, NULL, &surface);
 }
 
 struct pipe_surface *r600_create_surface_custom(struct pipe_context *pipe,
@@ -479,6 +609,8 @@ struct pipe_surface *r600_create_surface_custom(struct pipe_context *pipe,
 {
        struct r600_surface *surface = CALLOC_STRUCT(r600_surface);
 
+        assert(templ->u.tex.first_layer <= u_max_layer(texture, templ->u.tex.level));
+        assert(templ->u.tex.last_layer <= u_max_layer(texture, templ->u.tex.level));
        assert(templ->u.tex.first_layer == templ->u.tex.last_layer);
        if (surface == NULL)
                return NULL;
@@ -488,21 +620,19 @@ struct pipe_surface *r600_create_surface_custom(struct pipe_context *pipe,
        surface->base.format = templ->format;
        surface->base.width = width;
        surface->base.height = height;
-       surface->base.usage = templ->usage;
        surface->base.u = templ->u;
        return &surface->base;
 }
 
 static struct pipe_surface *r600_create_surface(struct pipe_context *pipe,
-                                               struct pipe_resource *texture,
+                                               struct pipe_resource *tex,
                                                const struct pipe_surface *templ)
 {
-       struct r600_texture *rtex = (struct r600_texture*)texture;
        unsigned level = templ->u.tex.level;
 
-       return r600_create_surface_custom(pipe, texture, templ,
-                                         rtex->surface.level[level].npix_x,
-                                         rtex->surface.level[level].npix_y);
+       return r600_create_surface_custom(pipe, tex, templ,
+                                          u_minify(tex->width0, level),
+                                         u_minify(tex->height0, level));
 }
 
 static void r600_surface_destroy(struct pipe_context *pipe,
@@ -547,14 +677,14 @@ struct pipe_resource *r600_texture_from_handle(struct pipe_screen *screen,
        else if (micro == RADEON_LAYOUT_TILED)
                array_mode = V_0280A0_ARRAY_1D_TILED_THIN1;
        else
-               array_mode = 0;
+               array_mode = V_038000_ARRAY_LINEAR_ALIGNED;
 
-       r = r600_init_surface(rscreen, &surface, templ, array_mode, false, false);
+       r = r600_init_surface(rscreen, &surface, templ, array_mode, false);
        if (r) {
                return NULL;
        }
        return (struct pipe_resource *)r600_texture_create_object(screen, templ,
-                                                                 stride, buf, FALSE, &surface);
+                                                                 stride, buf, &surface);
 }
 
 bool r600_init_flushed_depth_texture(struct pipe_context *ctx,
@@ -591,6 +721,7 @@ bool r600_init_flushed_depth_texture(struct pipe_context *ctx,
        }
 
        (*flushed_depth_texture)->is_flushing_texture = TRUE;
+       (*flushed_depth_texture)->non_disp_tiling = false;
        return true;
 }
 
@@ -603,11 +734,10 @@ static void *r600_texture_transfer_map(struct pipe_context *ctx,
 {
        struct r600_context *rctx = (struct r600_context*)ctx;
        struct r600_texture *rtex = (struct r600_texture*)texture;
-       struct pipe_resource resource;
        struct r600_transfer *trans;
        boolean use_staging_texture = FALSE;
        enum pipe_format format = texture->format;
-       struct radeon_winsys_cs_handle *buf;
+       struct r600_resource *buf;
        unsigned offset = 0;
        char *map;
 
@@ -628,7 +758,7 @@ static void *r600_texture_transfer_map(struct pipe_context *ctx,
 
        /* Use a staging texture for uploads if the underlying BO is busy. */
        if (!(usage & PIPE_TRANSFER_READ) &&
-           (rctx->ws->cs_is_buffer_referenced(rctx->cs, rtex->resource.cs_buf, RADEON_USAGE_READWRITE) ||
+           (r600_rings_is_buffer_referenced(rctx, rtex->resource.cs_buf, RADEON_USAGE_READWRITE) ||
             rctx->ws->buffer_is_busy(rtex->resource.buf, RADEON_USAGE_READWRITE))) {
                use_staging_texture = TRUE;
        }
@@ -655,6 +785,13 @@ static void *r600_texture_transfer_map(struct pipe_context *ctx,
                */
                struct r600_texture *staging_depth;
 
+               assert(rtex->resource.b.b.nr_samples <= 1);
+               if (rtex->resource.b.b.nr_samples > 1) {
+                       R600_ERR("mapping MSAA zbuffer unimplemented\n");
+                       FREE(trans);
+                       return NULL;
+               }
+
                if (!r600_init_flushed_depth_texture(ctx, texture, &staging_depth)) {
                        R600_ERR("failed to create temporary texture to hold untiled copy\n");
                        FREE(trans);
@@ -667,44 +804,58 @@ static void *r600_texture_transfer_map(struct pipe_context *ctx,
                                           0, 0);
 
                trans->transfer.stride = staging_depth->surface.level[level].pitch_bytes;
+                trans->transfer.layer_stride = staging_depth->surface.level[level].slice_size;
                trans->offset = r600_texture_get_offset(staging_depth, level, box->z);
                trans->staging = (struct r600_resource*)staging_depth;
        } else if (use_staging_texture) {
-               resource.target = PIPE_TEXTURE_2D;
+               struct pipe_resource resource;
+               struct r600_texture *staging;
+
+               memset(&resource, 0, sizeof(resource));
                resource.format = texture->format;
                resource.width0 = box->width;
                resource.height0 = box->height;
                resource.depth0 = 1;
                resource.array_size = 1;
-               resource.last_level = 0;
-               resource.nr_samples = 0;
                resource.usage = PIPE_USAGE_STAGING;
-               resource.bind = 0;
                resource.flags = R600_RESOURCE_FLAG_TRANSFER;
-               /* For texture reading, the temporary (detiled) texture is used as
-                * a render target when blitting from a tiled texture. */
-               if (usage & PIPE_TRANSFER_READ) {
-                       resource.bind |= PIPE_BIND_RENDER_TARGET;
-               }
-               /* For texture writing, the temporary texture is used as a sampler
-                * when blitting into a tiled texture. */
-               if (usage & PIPE_TRANSFER_WRITE) {
-                       resource.bind |= PIPE_BIND_SAMPLER_VIEW;
+
+               /* We must set the correct texture target and dimensions if needed for a 3D transfer. */
+               if (box->depth > 1 && u_max_layer(texture, level) > 0)
+                       resource.target = texture->target;
+               else
+                       resource.target = PIPE_TEXTURE_2D;
+
+               switch (resource.target) {
+               case PIPE_TEXTURE_1D_ARRAY:
+               case PIPE_TEXTURE_2D_ARRAY:
+               case PIPE_TEXTURE_CUBE_ARRAY:
+                       resource.array_size = box->depth;
+                       break;
+               case PIPE_TEXTURE_3D:
+                       resource.depth0 = box->depth;
+                       break;
+               default:;
                }
+
+
                /* Create the temporary texture. */
-               trans->staging = (struct r600_resource*)ctx->screen->resource_create(ctx->screen, &resource);
-               if (trans->staging == NULL) {
+               staging = (struct r600_texture*)ctx->screen->resource_create(ctx->screen, &resource);
+               if (staging == NULL) {
                        R600_ERR("failed to create temporary texture to hold untiled copy\n");
                        FREE(trans);
                        return NULL;
                }
-
-               trans->transfer.stride =
-                       ((struct r600_texture *)trans->staging)->surface.level[0].pitch_bytes;
+               trans->staging = &staging->resource;
+               trans->transfer.stride = staging->surface.level[0].pitch_bytes;
+               trans->transfer.layer_stride = staging->surface.level[0].slice_size;
                if (usage & PIPE_TRANSFER_READ) {
                        r600_copy_to_staging_texture(ctx, trans);
-                       /* Always referenced in the blit. */
-                       r600_flush(ctx, NULL, 0);
+                       /* flush gfx & dma ring, order does not matter as only one can be live */
+                       if (rctx->rings.dma.cs) {
+                               rctx->rings.dma.flush(rctx, 0);
+                       }
+                       rctx->rings.gfx.flush(rctx, 0);
                }
        } else {
                trans->transfer.stride = rtex->surface.level[level].pitch_bytes;
@@ -713,9 +864,9 @@ static void *r600_texture_transfer_map(struct pipe_context *ctx,
        }
 
        if (trans->staging) {
-               buf = ((struct r600_resource *)trans->staging)->cs_buf;
+               buf = trans->staging;
        } else {
-               buf = ((struct r600_resource *)texture)->cs_buf;
+               buf = &rtex->resource;
        }
 
        if (rtex->is_depth || !trans->staging)
@@ -723,7 +874,7 @@ static void *r600_texture_transfer_map(struct pipe_context *ctx,
                        box->y / util_format_get_blockheight(format) * trans->transfer.stride +
                        box->x / util_format_get_blockwidth(format) * util_format_get_blocksize(format);
 
-       if (!(map = rctx->ws->buffer_map(buf, rctx->cs, usage))) {
+       if (!(map = r600_buffer_mmap_sync_with_rings(rctx, buf, usage))) {
                pipe_resource_reference((struct pipe_resource**)&trans->staging, NULL);
                FREE(trans);
                return NULL;
@@ -776,18 +927,26 @@ void r600_init_surface_functions(struct r600_context *r600)
        r600->context.surface_destroy = r600_surface_destroy;
 }
 
-static unsigned r600_get_swizzle_combined(const unsigned char *swizzle_format,
-               const unsigned char *swizzle_view)
+unsigned r600_get_swizzle_combined(const unsigned char *swizzle_format,
+                                  const unsigned char *swizzle_view,
+                                  boolean vtx)
 {
        unsigned i;
        unsigned char swizzle[4];
        unsigned result = 0;
-       const uint32_t swizzle_shift[4] = {
+       const uint32_t tex_swizzle_shift[4] = {
                16, 19, 22, 25,
        };
+       const uint32_t vtx_swizzle_shift[4] = {
+               3, 6, 9, 12,
+       };
        const uint32_t swizzle_bit[4] = {
                0, 1, 2, 3,
        };
+       const uint32_t *swizzle_shift = tex_swizzle_shift;
+
+       if (vtx)
+               swizzle_shift = vtx_swizzle_shift;
 
        if (swizzle_view) {
                util_format_compose_swizzles(swizzle_format, swizzle_view, swizzle);
@@ -826,11 +985,14 @@ uint32_t r600_translate_texformat(struct pipe_screen *screen,
                                  const unsigned char *swizzle_view,
                                  uint32_t *word4_p, uint32_t *yuv_format_p)
 {
+       struct r600_screen *rscreen = (struct r600_screen *)screen;
        uint32_t result = 0, word4 = 0, yuv_format = 0;
        const struct util_format_description *desc;
        boolean uniform = TRUE;
        static int r600_enable_s3tc = -1;
        bool is_srgb_valid = FALSE;
+       const unsigned char swizzle_xxxx[4] = {0, 0, 0, 0};
+       const unsigned char swizzle_yyyy[4] = {1, 1, 1, 1};
 
        int i;
        const uint32_t sign_bit[4] = {
@@ -841,38 +1003,62 @@ uint32_t r600_translate_texformat(struct pipe_screen *screen,
        };
        desc = util_format_description(format);
 
-       word4 |= r600_get_swizzle_combined(desc->swizzle, swizzle_view);
+       /* Depth and stencil swizzling is handled separately. */
+       if (desc->colorspace != UTIL_FORMAT_COLORSPACE_ZS) {
+               word4 |= r600_get_swizzle_combined(desc->swizzle, swizzle_view, FALSE);
+       }
 
        /* Colorspace (return non-RGB formats directly). */
        switch (desc->colorspace) {
-               /* Depth stencil formats */
+       /* Depth stencil formats */
        case UTIL_FORMAT_COLORSPACE_ZS:
                switch (format) {
+               /* Depth sampler formats. */
                case PIPE_FORMAT_Z16_UNORM:
+                       word4 |= r600_get_swizzle_combined(swizzle_xxxx, swizzle_view, FALSE);
                        result = FMT_16;
                        goto out_word4;
-               case PIPE_FORMAT_X24S8_UINT:
-                       word4 |= S_038010_NUM_FORMAT_ALL(V_038010_SQ_NUM_FORMAT_INT);
                case PIPE_FORMAT_Z24X8_UNORM:
                case PIPE_FORMAT_Z24_UNORM_S8_UINT:
+                       word4 |= r600_get_swizzle_combined(swizzle_xxxx, swizzle_view, FALSE);
                        result = FMT_8_24;
                        goto out_word4;
-               case PIPE_FORMAT_S8X24_UINT:
-                       word4 |= S_038010_NUM_FORMAT_ALL(V_038010_SQ_NUM_FORMAT_INT);
                case PIPE_FORMAT_X8Z24_UNORM:
                case PIPE_FORMAT_S8_UINT_Z24_UNORM:
+                       if (rscreen->chip_class < EVERGREEN)
+                               goto out_unknown;
+                       word4 |= r600_get_swizzle_combined(swizzle_yyyy, swizzle_view, FALSE);
                        result = FMT_24_8;
                        goto out_word4;
+               case PIPE_FORMAT_Z32_FLOAT:
+                       word4 |= r600_get_swizzle_combined(swizzle_xxxx, swizzle_view, FALSE);
+                       result = FMT_32_FLOAT;
+                       goto out_word4;
+               case PIPE_FORMAT_Z32_FLOAT_S8X24_UINT:
+                       word4 |= r600_get_swizzle_combined(swizzle_xxxx, swizzle_view, FALSE);
+                       result = FMT_X24_8_32_FLOAT;
+                       goto out_word4;
+               /* Stencil sampler formats. */
                case PIPE_FORMAT_S8_UINT:
+                       word4 |= S_038010_NUM_FORMAT_ALL(V_038010_SQ_NUM_FORMAT_INT);
+                       word4 |= r600_get_swizzle_combined(swizzle_xxxx, swizzle_view, FALSE);
                        result = FMT_8;
+                       goto out_word4;
+               case PIPE_FORMAT_X24S8_UINT:
                        word4 |= S_038010_NUM_FORMAT_ALL(V_038010_SQ_NUM_FORMAT_INT);
+                       word4 |= r600_get_swizzle_combined(swizzle_yyyy, swizzle_view, FALSE);
+                       result = FMT_8_24;
                        goto out_word4;
-               case PIPE_FORMAT_Z32_FLOAT:
-                       result = FMT_32_FLOAT;
+               case PIPE_FORMAT_S8X24_UINT:
+                       if (rscreen->chip_class < EVERGREEN)
+                               goto out_unknown;
+                       word4 |= S_038010_NUM_FORMAT_ALL(V_038010_SQ_NUM_FORMAT_INT);
+                       word4 |= r600_get_swizzle_combined(swizzle_xxxx, swizzle_view, FALSE);
+                       result = FMT_24_8;
                        goto out_word4;
                case PIPE_FORMAT_X32_S8X24_UINT:
                        word4 |= S_038010_NUM_FORMAT_ALL(V_038010_SQ_NUM_FORMAT_INT);
-               case PIPE_FORMAT_Z32_FLOAT_S8X24_UINT:
+                       word4 |= r600_get_swizzle_combined(swizzle_yyyy, swizzle_view, FALSE);
                        result = FMT_X24_8_32_FLOAT;
                        goto out_word4;
                default:
@@ -898,7 +1084,6 @@ uint32_t r600_translate_texformat(struct pipe_screen *screen,
        }
 
        if (r600_enable_s3tc == -1) {
-               struct r600_screen *rscreen = (struct r600_screen *)screen;
                if (rscreen->info.drm_minor >= 9)
                        r600_enable_s3tc = 1;
                else