r600g: put depth and stencil into one backing buffer
[mesa.git] / src / gallium / drivers / r600 / r600_texture.c
index b7600e90eb5a51ac2f00480fa055f7f1be1708b6..f0cf1f593d84289f43089202a3a3cfac6f9db00f 100644 (file)
 #include <errno.h>
 #include <pipe/p_screen.h>
 #include <util/u_format.h>
+#include <util/u_format_s3tc.h>
 #include <util/u_math.h>
 #include <util/u_inlines.h>
 #include <util/u_memory.h>
-#include "state_tracker/drm_driver.h"
 #include "pipebuffer/pb_buffer.h"
 #include "r600_pipe.h"
 #include "r600_resource.h"
-#include "r600_state_inlines.h"
 #include "r600d.h"
 #include "r600_formats.h"
 
-extern struct u_resource_vtbl r600_texture_vtbl;
-
 /* 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 pipe_transfer *transfer = (struct pipe_transfer*)rtransfer;
        struct pipe_resource *texture = transfer->resource;
-       struct pipe_subresource subdst;
 
-       subdst.face = 0;
-       subdst.level = 0;
        ctx->resource_copy_region(ctx, rtransfer->staging_texture,
-                               subdst, 0, 0, 0, texture, transfer->sr,
-                               transfer->box.x, transfer->box.y, transfer->box.z,
-                               transfer->box.width, transfer->box.height);
+                               0, 0, 0, 0, texture, transfer->level,
+                               &transfer->box);
 }
 
 
@@ -61,39 +54,35 @@ static void r600_copy_from_staging_texture(struct pipe_context *ctx, struct r600
 {
        struct pipe_transfer *transfer = (struct pipe_transfer*)rtransfer;
        struct pipe_resource *texture = transfer->resource;
-       struct pipe_subresource subsrc;
-
-       subsrc.face = 0;
-       subsrc.level = 0;
-       ctx->resource_copy_region(ctx, texture, transfer->sr,
+       struct pipe_box sbox;
+
+       sbox.x = sbox.y = sbox.z = 0;
+       sbox.width = transfer->box.width;
+       sbox.height = transfer->box.height;
+       /* XXX that might be wrong */
+       sbox.depth = 1;
+       ctx->resource_copy_region(ctx, texture, transfer->level,
                                  transfer->box.x, transfer->box.y, transfer->box.z,
-                                 rtransfer->staging_texture, subsrc,
-                                 0, 0, 0,
-                                 transfer->box.width, transfer->box.height);
+                                 rtransfer->staging_texture,
+                                 0, &sbox);
 
-       ctx->flush(ctx, 0, NULL);
+       r600_flush(ctx, NULL, RADEON_FLUSH_ASYNC);
 }
 
-static unsigned r600_texture_get_offset(struct r600_resource_texture *rtex,
-                                       unsigned level, unsigned zslice,
-                                       unsigned face)
+unsigned r600_texture_get_offset(struct r600_resource_texture *rtex,
+                                       unsigned level, unsigned layer)
 {
        unsigned offset = rtex->offset[level];
 
-       switch (rtex->resource.base.b.target) {
+       switch (rtex->resource.b.b.b.target) {
        case PIPE_TEXTURE_3D:
-               assert(face == 0);
-               return offset + zslice * rtex->layer_size[level];
        case PIPE_TEXTURE_CUBE:
-               assert(zslice == 0);
-               return offset + face * rtex->layer_size[level];
        default:
-               assert(zslice == 0 && face == 0);
-               return offset;
+               return offset + layer * rtex->layer_size[level];
        }
 }
 
-static unsigned r600_get_pixel_alignment(struct pipe_screen *screen,
+static unsigned r600_get_block_alignment(struct pipe_screen *screen,
                                         enum pipe_format format,
                                         unsigned array_mode)
 {
@@ -111,6 +100,9 @@ static unsigned r600_get_pixel_alignment(struct pipe_screen *screen,
                               (((rscreen->tiling_info->group_bytes / 8 / pixsize)) *
                                rscreen->tiling_info->num_banks)) * 8;
                break;
+       case V_038000_ARRAY_LINEAR_ALIGNED:
+               p_align = MAX2(64, rscreen->tiling_info->group_bytes / pixsize);
+               break;
        case V_038000_ARRAY_LINEAR_GENERAL:
        default:
                p_align = rscreen->tiling_info->group_bytes / pixsize;
@@ -130,8 +122,10 @@ static unsigned r600_get_height_alignment(struct pipe_screen *screen,
                h_align = rscreen->tiling_info->num_channels * 8;
                break;
        case V_038000_ARRAY_1D_TILED_THIN1:
+       case V_038000_ARRAY_LINEAR_ALIGNED:
                h_align = 8;
                break;
+       case V_038000_ARRAY_LINEAR_GENERAL:
        default:
                h_align = 1;
                break;
@@ -145,7 +139,7 @@ static unsigned r600_get_base_alignment(struct pipe_screen *screen,
 {
        struct r600_screen* rscreen = (struct r600_screen *)screen;
        unsigned pixsize = util_format_get_blocksize(format);
-       int p_align = r600_get_pixel_alignment(screen, format, array_mode);
+       int p_align = r600_get_block_alignment(screen, format, array_mode);
        int h_align = r600_get_height_alignment(screen, array_mode);
        int b_align;
 
@@ -155,6 +149,8 @@ static unsigned r600_get_base_alignment(struct pipe_screen *screen,
                               p_align * pixsize * h_align);
                break;
        case V_038000_ARRAY_1D_TILED_THIN1:
+       case V_038000_ARRAY_LINEAR_ALIGNED:
+       case V_038000_ARRAY_LINEAR_GENERAL:
        default:
                b_align = rscreen->tiling_info->group_bytes;
                break;
@@ -171,59 +167,46 @@ static unsigned mip_minify(unsigned size, unsigned level)
        return val;
 }
 
-static unsigned r600_texture_get_stride(struct pipe_screen *screen,
-                                       struct r600_resource_texture *rtex,
-                                       unsigned level)
+static unsigned r600_texture_get_nblocksx(struct pipe_screen *screen,
+                                         struct r600_resource_texture *rtex,
+                                         unsigned level)
 {
-       struct r600_screen* rscreen = (struct r600_screen *)screen;
-       struct pipe_resource *ptex = &rtex->resource.base.b;
-       struct radeon *radeon = (struct radeon *)screen->winsys;
-       enum chip_class chipc = r600_get_family_class(radeon);
-       unsigned width, stride, tile_width;
+       struct pipe_resource *ptex = &rtex->resource.b.b.b;
+       unsigned nblocksx, block_align, width;
+       unsigned blocksize = util_format_get_blocksize(rtex->real_format);
 
        if (rtex->pitch_override)
-               return rtex->pitch_override;
+               return rtex->pitch_override / blocksize;
 
        width = mip_minify(ptex->width0, level);
-       if (util_format_is_plain(ptex->format)) {
-               tile_width = r600_get_pixel_alignment(screen, ptex->format,
-                                                     rtex->array_mode[level]);
-               width = align(width, tile_width);
-       }
-       stride = util_format_get_stride(ptex->format, width);
+       nblocksx = util_format_get_nblocksx(rtex->real_format, width);
 
-       return stride;
+       block_align = r600_get_block_alignment(screen, rtex->real_format,
+                                             rtex->array_mode[level]);
+       nblocksx = align(nblocksx, block_align);
+       return nblocksx;
 }
 
 static unsigned r600_texture_get_nblocksy(struct pipe_screen *screen,
                                          struct r600_resource_texture *rtex,
                                          unsigned level)
 {
-       struct pipe_resource *ptex = &rtex->resource.base.b;
+       struct pipe_resource *ptex = &rtex->resource.b.b.b;
        unsigned height, tile_height;
 
        height = mip_minify(ptex->height0, level);
-       if (util_format_is_plain(ptex->format)) {
-               tile_height = r600_get_height_alignment(screen,
-                                                       rtex->array_mode[level]);
-               height = align(height, tile_height);
-       }
-       return util_format_get_nblocksy(ptex->format, height);
-}
-
-/* Get a width in pixels from a stride in bytes. */
-static unsigned pitch_to_width(enum pipe_format format,
-                                unsigned pitch_in_bytes)
-{
-    return (pitch_in_bytes / util_format_get_blocksize(format)) *
-            util_format_get_blockwidth(format);
+       height = util_format_get_nblocksy(rtex->real_format, height);
+       tile_height = r600_get_height_alignment(screen,
+                                               rtex->array_mode[level]);
+       height = align(height, tile_height);
+       return height;
 }
 
 static void r600_texture_set_array_mode(struct pipe_screen *screen,
                                        struct r600_resource_texture *rtex,
                                        unsigned level, unsigned array_mode)
 {
-       struct pipe_resource *ptex = &rtex->resource.base.b;
+       struct pipe_resource *ptex = &rtex->resource.b.b.b;
 
        switch (array_mode) {
        case V_0280A0_ARRAY_LINEAR_GENERAL:
@@ -237,11 +220,11 @@ static void r600_texture_set_array_mode(struct pipe_screen *screen,
                unsigned w, h, tile_height, tile_width;
 
                tile_height = r600_get_height_alignment(screen, array_mode);
-               tile_width = r600_get_pixel_alignment(screen, ptex->format, array_mode);
+               tile_width = r600_get_block_alignment(screen, rtex->real_format, array_mode);
 
                w = mip_minify(ptex->width0, level);
                h = mip_minify(ptex->height0, level);
-               if (w < tile_width || h < tile_height)
+               if (w <= tile_width || h <= tile_height)
                        rtex->array_mode[level] = V_0280A0_ARRAY_1D_TILED_THIN1;
                else
                        rtex->array_mode[level] = array_mode;
@@ -254,80 +237,230 @@ static void r600_setup_miptree(struct pipe_screen *screen,
                               struct r600_resource_texture *rtex,
                               unsigned array_mode)
 {
-       struct pipe_resource *ptex = &rtex->resource.base.b;
-       struct radeon *radeon = (struct radeon *)screen->winsys;
+       struct pipe_resource *ptex = &rtex->resource.b.b.b;
+       struct radeon *radeon = ((struct r600_screen*)screen)->radeon;
        enum chip_class chipc = r600_get_family_class(radeon);
-       unsigned pitch, size, layer_size, i, offset;
-       unsigned nblocksy;
+       unsigned size, layer_size, i, offset;
+       unsigned nblocksx, nblocksy;
 
        for (i = 0, offset = 0; i <= ptex->last_level; i++) {
+               unsigned blocksize = util_format_get_blocksize(rtex->real_format);
+               unsigned base_align = r600_get_base_alignment(screen, rtex->real_format, array_mode);
+
                r600_texture_set_array_mode(screen, rtex, i, array_mode);
 
-               pitch = r600_texture_get_stride(screen, rtex, i);
+               nblocksx = r600_texture_get_nblocksx(screen, rtex, i);
                nblocksy = r600_texture_get_nblocksy(screen, rtex, i);
 
-               layer_size = pitch * nblocksy;
-
+               layer_size = nblocksx * nblocksy * blocksize;
                if (ptex->target == PIPE_TEXTURE_CUBE) {
                        if (chipc >= R700)
                                size = layer_size * 8;
                        else
                                size = layer_size * 6;
                }
-               else
+               else if (ptex->target == PIPE_TEXTURE_3D)
                        size = layer_size * u_minify(ptex->depth0, i);
+               else
+                       size = layer_size * ptex->array_size;
+
                /* align base image and start of miptree */
                if ((i == 0) || (i == 1))
-                       offset = align(offset, r600_get_base_alignment(screen, ptex->format, array_mode));
+                       offset = align(offset, base_align);
                rtex->offset[i] = offset;
                rtex->layer_size[i] = layer_size;
-               rtex->pitch_in_bytes[i] = pitch;
-               rtex->pitch_in_pixels[i] = pitch_to_width(ptex->format, pitch);
+               rtex->pitch_in_blocks[i] = nblocksx; /* CB talks in elements */
+               rtex->pitch_in_bytes[i] = nblocksx * blocksize;
+
                offset += size;
        }
        rtex->size = offset;
 }
 
+/* Figure out whether u_blitter will fallback to a transfer operation.
+ * If so, don't use a staging resource.
+ */
+static boolean permit_hardware_blit(struct pipe_screen *screen,
+                                       const struct pipe_resource *res)
+{
+       unsigned bind;
+
+       if (util_format_is_depth_or_stencil(res->format))
+               bind = PIPE_BIND_DEPTH_STENCIL;
+       else
+               bind = PIPE_BIND_RENDER_TARGET;
+
+       /* hackaround for S3TC */
+       if (util_format_is_compressed(res->format))
+               return TRUE;
+           
+       if (!screen->is_format_supported(screen,
+                               res->format,
+                               res->target,
+                               res->nr_samples,
+                                bind))
+               return FALSE;
+
+       if (!screen->is_format_supported(screen,
+                               res->format,
+                               res->target,
+                               res->nr_samples,
+                                PIPE_BIND_SAMPLER_VIEW))
+               return FALSE;
+
+       switch (res->usage) {
+       case PIPE_USAGE_STREAM:
+       case PIPE_USAGE_STAGING:
+               return FALSE;
+
+       default:
+               return TRUE;
+       }
+}
+
+static boolean r600_texture_get_handle(struct pipe_screen* screen,
+                                       struct pipe_resource *ptex,
+                                       struct winsys_handle *whandle)
+{
+       struct r600_resource_texture *rtex = (struct r600_resource_texture*)ptex;
+       struct r600_resource *resource = &rtex->resource;
+       struct radeon *radeon = ((struct r600_screen*)screen)->radeon;
+
+       return r600_bo_get_winsys_handle(radeon, resource->bo,
+                       rtex->pitch_in_bytes[0], whandle);
+}
+
+static void r600_texture_destroy(struct pipe_screen *screen,
+                                struct pipe_resource *ptex)
+{
+       struct r600_resource_texture *rtex = (struct r600_resource_texture*)ptex;
+       struct r600_resource *resource = &rtex->resource;
+
+       if (rtex->flushed_depth_texture)
+               pipe_resource_reference((struct pipe_resource **)&rtex->flushed_depth_texture, NULL);
+
+       if (resource->bo) {
+               r600_bo_reference(&resource->bo, NULL);
+       }
+       FREE(rtex);
+}
+
+static const struct u_resource_vtbl r600_texture_vtbl =
+{
+       r600_texture_get_handle,        /* get_handle */
+       r600_texture_destroy,           /* resource_destroy */
+       r600_texture_get_transfer,      /* get_transfer */
+       r600_texture_transfer_destroy,  /* transfer_destroy */
+       r600_texture_transfer_map,      /* transfer_map */
+       u_default_transfer_flush_region,/* transfer_flush_region */
+       r600_texture_transfer_unmap,    /* transfer_unmap */
+       u_default_transfer_inline_write /* transfer_inline_write */
+};
+
 static struct r600_resource_texture *
 r600_texture_create_object(struct pipe_screen *screen,
                           const struct pipe_resource *base,
                           unsigned array_mode,
                           unsigned pitch_in_bytes_override,
                           unsigned max_buffer_size,
-                          struct r600_bo *bo)
+                          struct r600_bo *bo,
+                          boolean alloc_bo)
 {
        struct r600_resource_texture *rtex;
        struct r600_resource *resource;
-       struct radeon *radeon = (struct radeon *)screen->winsys;
+       struct radeon *radeon = ((struct r600_screen*)screen)->radeon;
 
        rtex = CALLOC_STRUCT(r600_resource_texture);
        if (rtex == NULL)
                return NULL;
 
        resource = &rtex->resource;
-       resource->base.b = *base;
-       resource->base.vtbl = &r600_texture_vtbl;
-       pipe_reference_init(&resource->base.b.reference, 1);
-       resource->base.b.screen = screen;
+       resource->b.b.b = *base;
+       resource->b.b.vtbl = &r600_texture_vtbl;
+       pipe_reference_init(&resource->b.b.b.reference, 1);
+       resource->b.b.b.screen = screen;
        resource->bo = bo;
        rtex->pitch_override = pitch_in_bytes_override;
+       rtex->real_format = base->format;
+
+       /* We must split depth and stencil into two separate buffers on Evergreen. */
+       if (r600_get_family_class(((struct r600_screen*)screen)->radeon) >= EVERGREEN &&
+           util_format_is_depth_and_stencil(base->format)) {
+               struct pipe_resource stencil;
+               unsigned stencil_pitch_override = 0;
+
+               switch (base->format) {
+               case PIPE_FORMAT_Z24_UNORM_S8_USCALED:
+                       rtex->real_format = PIPE_FORMAT_Z24X8_UNORM;
+                       break;
+               case PIPE_FORMAT_S8_USCALED_Z24_UNORM:
+                       rtex->real_format = PIPE_FORMAT_X8Z24_UNORM;
+                       break;
+               case PIPE_FORMAT_Z32_FLOAT_S8X24_USCALED:
+                       rtex->real_format = PIPE_FORMAT_Z32_FLOAT;
+                       break;
+               default:
+                       assert(0);
+                       FREE(rtex);
+                       return NULL;
+               }
+
+               /* Divide the pitch in bytes by 4 for stencil, because it has a smaller pixel size. */
+               if (pitch_in_bytes_override) {
+                       assert(base->format == PIPE_FORMAT_Z24_UNORM_S8_USCALED ||
+                              base->format == PIPE_FORMAT_S8_USCALED_Z24_UNORM);
+                       stencil_pitch_override = pitch_in_bytes_override / 4;
+               }
+
+               /* Allocate the stencil buffer. */
+               stencil = *base;
+               stencil.format = PIPE_FORMAT_S8_USCALED;
+               rtex->stencil = r600_texture_create_object(screen, &stencil, array_mode,
+                                                          stencil_pitch_override,
+                                                          max_buffer_size, NULL, FALSE);
+               if (!rtex->stencil) {
+                       FREE(rtex);
+                       return NULL;
+               }
+               /* Proceed in creating the depth buffer. */
+       }
+
+       /* only mark depth textures the HW can hit as depth textures */
+       if (util_format_is_depth_or_stencil(rtex->real_format) && permit_hardware_blit(screen, base))
+               rtex->depth = 1;
 
-       if (array_mode)
-               rtex->tiled = 1;
        r600_setup_miptree(screen, rtex, array_mode);
 
+       /* If we initialized separate stencil for Evergreen. place it after depth. */
+       if (rtex->stencil) {
+               unsigned stencil_align, stencil_offset;
+
+               stencil_align = r600_get_base_alignment(screen, rtex->stencil->real_format, array_mode);
+               stencil_offset = align(rtex->size, stencil_align);
+
+               for (unsigned i = 0; i <= rtex->stencil->resource.b.b.b.last_level; i++)
+                       rtex->stencil->offset[i] += stencil_offset;
+
+               rtex->size = stencil_offset + rtex->stencil->size;
+       }
+
        resource->size = rtex->size;
 
-       if (!resource->bo) {
-               struct pipe_resource *ptex = &rtex->resource.base.b;
-               int base_align = r600_get_base_alignment(screen, ptex->format, array_mode);
+       /* Now create the backing buffer. */
+       if (!resource->bo && alloc_bo) {
+               struct pipe_resource *ptex = &rtex->resource.b.b.b;
+               unsigned base_align = r600_get_base_alignment(screen, ptex->format, array_mode);
 
                resource->bo = r600_bo(radeon, rtex->size, base_align, base->bind, base->usage);
                if (!resource->bo) {
+                       pipe_resource_reference((struct pipe_resource**)&rtex->stencil, NULL);
                        FREE(rtex);
                        return NULL;
                }
        }
+
+       if (rtex->stencil)
+               rtex->stencil->resource.bo = rtex->resource.bo;
        return rtex;
 }
 
@@ -335,83 +468,67 @@ struct pipe_resource *r600_texture_create(struct pipe_screen *screen,
                                                const struct pipe_resource *templ)
 {
        unsigned array_mode = 0;
-        static int force_tiling = -1;
+       static int force_tiling = -1;
 
-        /* Would like some magic "get_bool_option_once" routine.
+       /* Would like some magic "get_bool_option_once" routine.
         */
-       if (force_tiling == -1)
-                force_tiling = debug_get_bool_option("R600_FORCE_TILING", FALSE);
+       if (force_tiling == -1) {
+#if 0
+               /* reenable when 2D tiling is fixed better */
+               struct r600_screen *rscreen = (struct r600_screen *)screen;
+               if (r600_get_minor_version(rscreen->radeon) >= 9)
+                       force_tiling = debug_get_bool_option("R600_TILING", TRUE);
+#endif
+               force_tiling = debug_get_bool_option("R600_TILING", FALSE);
+       }
 
-       if (force_tiling) {
+       if (force_tiling && permit_hardware_blit(screen, templ)) {
                if (!(templ->flags & R600_RESOURCE_FLAG_TRANSFER) &&
                    !(templ->bind & PIPE_BIND_SCANOUT)) {
                        array_mode = V_038000_ARRAY_2D_TILED_THIN1;
                }
        }
 
-       return (struct pipe_resource *)r600_texture_create_object(screen, templ, array_mode,
-                                                                 0, 0, NULL);
-
-}
+       if (!(templ->flags & R600_RESOURCE_FLAG_TRANSFER) &&
+           util_format_is_compressed(templ->format))
+               array_mode = V_038000_ARRAY_1D_TILED_THIN1;
 
-static void r600_texture_destroy(struct pipe_screen *screen,
-                                struct pipe_resource *ptex)
-{
-       struct r600_resource_texture *rtex = (struct r600_resource_texture*)ptex;
-       struct r600_resource *resource = &rtex->resource;
-       struct radeon *radeon = (struct radeon *)screen->winsys;
-
-       if (rtex->flushed_depth_texture)
-               pipe_resource_reference((struct pipe_resource **)&rtex->flushed_depth_texture, NULL);
-
-       if (resource->bo) {
-               r600_bo_reference(radeon, &resource->bo, NULL);
-       }
-       FREE(rtex);
-}
-
-static boolean r600_texture_get_handle(struct pipe_screen* screen,
-                                       struct pipe_resource *ptex,
-                                       struct winsys_handle *whandle)
-{
-       struct r600_resource_texture *rtex = (struct r600_resource_texture*)ptex;
-       struct r600_resource *resource = &rtex->resource;
-       struct radeon *radeon = (struct radeon *)screen->winsys;
-
-       return r600_bo_get_winsys_handle(radeon, resource->bo,
-                       rtex->pitch_in_bytes[0], whandle);
+       return (struct pipe_resource *)r600_texture_create_object(screen, templ, array_mode,
+                                                                 0, 0, NULL, TRUE);
 }
 
-static struct pipe_surface *r600_get_tex_surface(struct pipe_screen *screen,
+static struct pipe_surface *r600_create_surface(struct pipe_context *pipe,
                                                struct pipe_resource *texture,
-                                               unsigned face, unsigned level,
-                                               unsigned zslice, unsigned flags)
+                                               const struct pipe_surface *surf_tmpl)
 {
        struct r600_resource_texture *rtex = (struct r600_resource_texture*)texture;
        struct r600_surface *surface = CALLOC_STRUCT(r600_surface);
-       unsigned offset, tile_height;
+       unsigned level = surf_tmpl->u.tex.level;
 
+       assert(surf_tmpl->u.tex.first_layer == surf_tmpl->u.tex.last_layer);
        if (surface == NULL)
                return NULL;
-       offset = r600_texture_get_offset(rtex, level, zslice, face);
+       /* XXX no offset */
+/*     offset = r600_texture_get_offset(rtex, level, surf_tmpl->u.tex.first_layer);*/
        pipe_reference_init(&surface->base.reference, 1);
        pipe_resource_reference(&surface->base.texture, texture);
-       surface->base.format = texture->format;
+       surface->base.context = pipe;
+       surface->base.format = surf_tmpl->format;
        surface->base.width = mip_minify(texture->width0, level);
        surface->base.height = mip_minify(texture->height0, level);
-       surface->base.offset = offset;
-       surface->base.usage = flags;
-       surface->base.zslice = zslice;
+       surface->base.usage = surf_tmpl->usage;
        surface->base.texture = texture;
-       surface->base.face = face;
-       surface->base.level = level;
+       surface->base.u.tex.first_layer = surf_tmpl->u.tex.first_layer;
+       surface->base.u.tex.last_layer = surf_tmpl->u.tex.last_layer;
+       surface->base.u.tex.level = level;
 
-       tile_height = r600_get_height_alignment(screen, rtex->array_mode[level]);
-       surface->aligned_height = align(surface->base.height, tile_height);
+       surface->aligned_height = r600_texture_get_nblocksy(pipe->screen,
+                                                           rtex, level);
        return &surface->base;
 }
 
-static void r600_tex_surface_destroy(struct pipe_surface *surface)
+static void r600_surface_destroy(struct pipe_context *pipe,
+                                struct pipe_surface *surface)
 {
        pipe_resource_reference(&surface->texture, NULL);
        FREE(surface);
@@ -422,8 +539,9 @@ struct pipe_resource *r600_texture_from_handle(struct pipe_screen *screen,
                                               const struct pipe_resource *templ,
                                               struct winsys_handle *whandle)
 {
-       struct radeon *rw = (struct radeon*)screen->winsys;
+       struct radeon *rw = ((struct r600_screen*)screen)->radeon;
        struct r600_bo *bo = NULL;
+       unsigned stride = 0;
        unsigned array_mode = 0;
 
        /* Support only 2D textures without mipmaps */
@@ -431,29 +549,17 @@ struct pipe_resource *r600_texture_from_handle(struct pipe_screen *screen,
              templ->depth0 != 1 || templ->last_level != 0)
                return NULL;
 
-       bo = r600_bo_handle(rw, whandle->handle, &array_mode);
+       bo = r600_bo_handle(rw, whandle, &stride, &array_mode);
        if (bo == NULL) {
                return NULL;
        }
 
        return (struct pipe_resource *)r600_texture_create_object(screen, templ, array_mode,
-                                                                 whandle->stride,
-                                                                 0,
-                                                                 bo);
-}
-
-static unsigned int r600_texture_is_referenced(struct pipe_context *context,
-                                               struct pipe_resource *texture,
-                                               unsigned face, unsigned level)
-{
-       /* FIXME */
-       return PIPE_REFERENCED_FOR_READ | PIPE_REFERENCED_FOR_WRITE;
+                                                                 stride, 0, bo, FALSE);
 }
 
-int (*r600_blit_uncompress_depth_ptr)(struct pipe_context *ctx, struct r600_resource_texture *texture);
-
 int r600_texture_depth_flush(struct pipe_context *ctx,
-                            struct pipe_resource *texture)
+                            struct pipe_resource *texture, boolean just_create)
 {
        struct r600_resource_texture *rtex = (struct r600_resource_texture*)texture;
        struct pipe_resource resource;
@@ -466,7 +572,8 @@ int r600_texture_depth_flush(struct pipe_context *ctx,
        resource.width0 = texture->width0;
        resource.height0 = texture->height0;
        resource.depth0 = 1;
-       resource.last_level = 0;
+       resource.array_size = 1;
+       resource.last_level = texture->last_level;
        resource.nr_samples = 0;
        resource.usage = PIPE_USAGE_DYNAMIC;
        resource.bind = 0;
@@ -480,10 +587,14 @@ int r600_texture_depth_flush(struct pipe_context *ctx,
                return -ENOMEM;
        }
 
+       ((struct r600_resource_texture *)rtex->flushed_depth_texture)->is_flushing_texture = TRUE;
 out:
+       if (just_create)
+               return 0;
+
        /* XXX: only do this if the depth texture has actually changed:
         */
-       r600_blit_uncompress_depth_ptr(ctx, rtex);
+       r600_blit_uncompress_depth(ctx, rtex);
        return 0;
 }
 
@@ -491,52 +602,12 @@ out:
  */
 static INLINE unsigned u_box_volume( const struct pipe_box *box )
 {
-        return box->width * box->depth * box->height;
+       return box->width * box->depth * box->height;
 };
 
-
-/* Figure out whether u_blitter will fallback to a transfer operation.
- * If so, don't use a staging resource.
- */
-static boolean permit_hardware_blit(struct pipe_screen *screen,
-                                    struct pipe_resource *res)
-{
-        unsigned bind;
-
-        if (util_format_is_depth_or_stencil(res->format))
-                bind = PIPE_BIND_DEPTH_STENCIL;
-        else
-                bind = PIPE_BIND_RENDER_TARGET;
-
-       /* See r600_resource_copy_region: there is something wrong
-         * with depth resource copies at the moment so avoid them for
-         * now.
-         */
-       if (util_format_get_component_bits(res->format,
-                                           UTIL_FORMAT_COLORSPACE_ZS,
-                                           0) != 0)
-                return FALSE;
-
-        if (!screen->is_format_supported(screen,
-                                         res->format,
-                                         res->target,
-                                         res->nr_samples,
-                                         bind, 0))
-                return FALSE;
-
-        if (!screen->is_format_supported(screen,
-                                         res->format,
-                                         res->target,
-                                         res->nr_samples,
-                                         PIPE_BIND_SAMPLER_VIEW, 0))
-                return FALSE;
-
-        return TRUE;
-}
-
 struct pipe_transfer* r600_texture_get_transfer(struct pipe_context *ctx,
                                                struct pipe_resource *texture,
-                                               struct pipe_subresource sr,
+                                               unsigned level,
                                                unsigned usage,
                                                const struct pipe_box *box)
 {
@@ -553,53 +624,56 @@ struct pipe_transfer* r600_texture_get_transfer(struct pipe_context *ctx,
         * the CPU is much happier reading out of cached system memory
         * than uncached VRAM.
         */
-       if (rtex->tiled)
+       if (R600_TEX_IS_TILED(rtex, level))
                use_staging_texture = TRUE;
 
-       if ((usage & PIPE_TRANSFER_READ) &&
-            u_box_volume(box) > 1024)
-                use_staging_texture = TRUE;
-
-        /* XXX: Use a staging texture for uploads if the underlying BO
-         * is busy.  No interface for checking that currently? so do
-         * it eagerly whenever the transfer doesn't require a readback
-         * and might block.
-         */
-        if ((usage & PIPE_TRANSFER_WRITE) &&
-            !(usage & (PIPE_TRANSFER_READ |
-                       PIPE_TRANSFER_DONTBLOCK |
-                       PIPE_TRANSFER_UNSYNCHRONIZED)))
-                use_staging_texture = TRUE;
-
-        /*if (!permit_hardware_blit(ctx->screen, texture) ||
-            (texture->flags & R600_RESOURCE_FLAG_TRANSFER))*/
-                use_staging_texture = FALSE;
+       if ((usage & PIPE_TRANSFER_READ) && u_box_volume(box) > 1024)
+               use_staging_texture = TRUE;
+
+       /* XXX: Use a staging texture for uploads if the underlying BO
+        * is busy.  No interface for checking that currently? so do
+        * it eagerly whenever the transfer doesn't require a readback
+        * and might block.
+        */
+       if ((usage & PIPE_TRANSFER_WRITE) &&
+                       !(usage & (PIPE_TRANSFER_READ |
+                                       PIPE_TRANSFER_DONTBLOCK |
+                                       PIPE_TRANSFER_UNSYNCHRONIZED)))
+               use_staging_texture = TRUE;
+
+       if (!permit_hardware_blit(ctx->screen, texture) ||
+               (texture->flags & R600_RESOURCE_FLAG_TRANSFER))
+               use_staging_texture = FALSE;
 
        trans = CALLOC_STRUCT(r600_transfer);
        if (trans == NULL)
                return NULL;
        pipe_resource_reference(&trans->transfer.resource, texture);
-       trans->transfer.sr = sr;
+       trans->transfer.level = level;
        trans->transfer.usage = usage;
        trans->transfer.box = *box;
        if (rtex->depth) {
-                /* XXX: only readback the rectangle which is being mapped?
-                 */
-                /* XXX: when discard is true, no need to read back from depth texture
-                 */
-               r = r600_texture_depth_flush(ctx, texture);
+               /* XXX: only readback the rectangle which is being mapped?
+               */
+               /* XXX: when discard is true, no need to read back from depth texture
+               */
+               r = r600_texture_depth_flush(ctx, texture, FALSE);
                if (r < 0) {
                        R600_ERR("failed to create temporary texture to hold untiled copy\n");
                        pipe_resource_reference(&trans->transfer.resource, NULL);
                        FREE(trans);
                        return NULL;
                }
+               trans->transfer.stride = rtex->flushed_depth_texture->pitch_in_bytes[level];
+               trans->offset = r600_texture_get_offset(rtex->flushed_depth_texture, level, box->z);
+               return &trans->transfer;
        } else if (use_staging_texture) {
                resource.target = PIPE_TEXTURE_2D;
                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;
@@ -625,16 +699,17 @@ struct pipe_transfer* r600_texture_get_transfer(struct pipe_context *ctx,
                }
 
                trans->transfer.stride =
-                        ((struct r600_resource_texture *)trans->staging_texture)->pitch_in_bytes[0];
+                       ((struct r600_resource_texture *)trans->staging_texture)->pitch_in_bytes[0];
                if (usage & PIPE_TRANSFER_READ) {
                        r600_copy_to_staging_texture(ctx, trans);
                        /* Always referenced in the blit. */
-                       ctx->flush(ctx, 0, NULL);
+                       r600_flush(ctx, NULL, 0);
                }
                return &trans->transfer;
        }
-       trans->transfer.stride = rtex->pitch_in_bytes[sr.level];
-       trans->offset = r600_texture_get_offset(rtex, sr.level, box->z, sr.face);
+       trans->transfer.stride = rtex->pitch_in_bytes[level];
+       trans->transfer.layer_stride = rtex->layer_size[level];
+       trans->offset = r600_texture_get_offset(rtex, level, box->z);
        return &trans->transfer;
 }
 
@@ -642,7 +717,8 @@ void r600_texture_transfer_destroy(struct pipe_context *ctx,
                                   struct pipe_transfer *transfer)
 {
        struct r600_transfer *rtransfer = (struct r600_transfer*)transfer;
-       struct r600_resource_texture *rtex = (struct r600_resource_texture*)transfer->resource;
+       struct pipe_resource *texture = transfer->resource;
+       struct r600_resource_texture *rtex = (struct r600_resource_texture*)texture;
 
        if (rtransfer->staging_texture) {
                if (transfer->usage & PIPE_TRANSFER_WRITE) {
@@ -650,9 +726,12 @@ void r600_texture_transfer_destroy(struct pipe_context *ctx,
                }
                pipe_resource_reference(&rtransfer->staging_texture, NULL);
        }
-       if (rtex->flushed_depth_texture) {
-               pipe_resource_reference((struct pipe_resource **)&rtex->flushed_depth_texture, NULL);
+
+       if (rtex->depth && !rtex->is_flushing_texture) {
+               if ((transfer->usage & PIPE_TRANSFER_WRITE) && rtex->flushed_depth_texture)
+                       r600_blit_push_depth(ctx, rtex);
        }
+
        pipe_resource_reference(&transfer->resource, NULL);
        FREE(transfer);
 }
@@ -660,12 +739,12 @@ void r600_texture_transfer_destroy(struct pipe_context *ctx,
 void* r600_texture_transfer_map(struct pipe_context *ctx,
                                struct pipe_transfer* transfer)
 {
+       struct r600_pipe_context *rctx = (struct r600_pipe_context *)ctx;
        struct r600_transfer *rtransfer = (struct r600_transfer*)transfer;
        struct r600_bo *bo;
        enum pipe_format format = transfer->resource->format;
-       struct radeon *radeon = (struct radeon *)ctx->screen->winsys;
+       struct radeon *radeon = rctx->screen->radeon;
        unsigned offset = 0;
-       unsigned usage = 0;
        char *map;
 
        if (rtransfer->staging_texture) {
@@ -683,30 +762,7 @@ void* r600_texture_transfer_map(struct pipe_context *ctx,
                        transfer->box.x / util_format_get_blockwidth(format) * util_format_get_blocksize(format);
        }
 
-       if (transfer->usage & PIPE_TRANSFER_WRITE) {
-               usage |= PB_USAGE_CPU_WRITE;
-
-               if (transfer->usage & PIPE_TRANSFER_DISCARD) {
-               }
-
-               if (transfer->usage & PIPE_TRANSFER_FLUSH_EXPLICIT) {
-               }
-       }
-
-       if (transfer->usage & PIPE_TRANSFER_READ) {
-               usage |= PB_USAGE_CPU_READ;
-       }
-
-       if (transfer->usage & PIPE_TRANSFER_DONTBLOCK) {
-               usage |= PB_USAGE_DONTBLOCK;
-       }
-
-       if (transfer->usage & PIPE_TRANSFER_UNSYNCHRONIZED) {
-               usage |= PB_USAGE_UNSYNCHRONIZED;
-       }
-
-       map = r600_bo_map(radeon, bo, usage, ctx);
-       if (!map) {
+       if (!(map = r600_bo_map(radeon, bo, rctx->ctx.cs, transfer->usage))) {
                return NULL;
        }
 
@@ -717,7 +773,7 @@ void r600_texture_transfer_unmap(struct pipe_context *ctx,
                                 struct pipe_transfer* transfer)
 {
        struct r600_transfer *rtransfer = (struct r600_transfer*)transfer;
-       struct radeon *radeon = (struct radeon *)ctx->screen->winsys;
+       struct radeon *radeon = ((struct r600_screen*)ctx->screen)->radeon;
        struct r600_bo *bo;
 
        if (rtransfer->staging_texture) {
@@ -734,23 +790,10 @@ void r600_texture_transfer_unmap(struct pipe_context *ctx,
        r600_bo_unmap(radeon, bo);
 }
 
-struct u_resource_vtbl r600_texture_vtbl =
+void r600_init_surface_functions(struct r600_pipe_context *r600)
 {
-       r600_texture_get_handle,        /* get_handle */
-       r600_texture_destroy,           /* resource_destroy */
-       r600_texture_is_referenced,     /* is_resource_referenced */
-       r600_texture_get_transfer,      /* get_transfer */
-       r600_texture_transfer_destroy,  /* transfer_destroy */
-       r600_texture_transfer_map,      /* transfer_map */
-       u_default_transfer_flush_region,/* transfer_flush_region */
-       r600_texture_transfer_unmap,    /* transfer_unmap */
-       u_default_transfer_inline_write /* transfer_inline_write */
-};
-
-void r600_init_screen_texture_functions(struct pipe_screen *screen)
-{
-       screen->get_tex_surface = r600_get_tex_surface;
-       screen->tex_surface_destroy = r600_tex_surface_destroy;
+       r600->context.create_surface = r600_create_surface;
+       r600->context.surface_destroy = r600_surface_destroy;
 }
 
 static unsigned r600_get_swizzle_combined(const unsigned char *swizzle_format,
@@ -767,11 +810,7 @@ static unsigned r600_get_swizzle_combined(const unsigned char *swizzle_format,
        };
 
        if (swizzle_view) {
-               /* Combine two sets of swizzles. */
-               for (i = 0; i < 4; i++) {
-                       swizzle[i] = swizzle_view[i] <= UTIL_FORMAT_SWIZZLE_W ?
-                               swizzle_format[swizzle_view[i]] : swizzle_view[i];
-               }
+               util_format_compose_swizzles(swizzle_format, swizzle_view, swizzle);
        } else {
                memcpy(swizzle, swizzle_format, 4);
        }
@@ -802,13 +841,16 @@ static unsigned r600_get_swizzle_combined(const unsigned char *swizzle_format,
 }
 
 /* texture format translate */
-uint32_t r600_translate_texformat(enum pipe_format format,
-                                 const unsigned char *swizzle_view, 
+uint32_t r600_translate_texformat(struct pipe_screen *screen,
+                                 enum pipe_format format,
+                                 const unsigned char *swizzle_view,
                                  uint32_t *word4_p, uint32_t *yuv_format_p)
 {
        uint32_t result = 0, word4 = 0, yuv_format = 0;
        const struct util_format_description *desc;
        boolean uniform = TRUE;
+       static int r600_enable_s3tc = -1;
+
        int i;
        const uint32_t sign_bit[4] = {
                S_038010_FORMAT_COMP_X(V_038010_SQ_FORMAT_COMP_SIGNED),
@@ -844,6 +886,12 @@ uint32_t r600_translate_texformat(enum pipe_format format,
                        result = FMT_8;
                        word4 |= S_038010_NUM_FORMAT_ALL(V_038010_SQ_NUM_FORMAT_INT);
                        goto out_word4;
+               case PIPE_FORMAT_Z32_FLOAT:
+                       result = FMT_32_FLOAT;
+                       goto out_word4;
+               case PIPE_FORMAT_Z32_FLOAT_S8X24_USCALED:
+                       result = FMT_X24_8_32_FLOAT;
+                       goto out_word4;
                default:
                        goto out_unknown;
                }
@@ -851,52 +899,92 @@ uint32_t r600_translate_texformat(enum pipe_format format,
        case UTIL_FORMAT_COLORSPACE_YUV:
                yuv_format |= (1 << 30);
                switch (format) {
-                case PIPE_FORMAT_UYVY:
-                case PIPE_FORMAT_YUYV:
+               case PIPE_FORMAT_UYVY:
+               case PIPE_FORMAT_YUYV:
                default:
                        break;
                }
                goto out_unknown; /* TODO */
-               
+
        case UTIL_FORMAT_COLORSPACE_SRGB:
                word4 |= S_038010_FORCE_DEGAMMA(1);
-               if (format == PIPE_FORMAT_L8A8_SRGB || format == PIPE_FORMAT_L8_SRGB)
-                       goto out_unknown; /* fails for some reason - TODO */
                break;
 
        default:
                break;
        }
 
-       /* S3TC formats. TODO */
+       if (r600_enable_s3tc == -1) {
+               struct r600_screen *rscreen = (struct r600_screen *)screen;
+               if (r600_get_minor_version(rscreen->radeon) >= 9)
+                       r600_enable_s3tc = 1;
+               else
+                       r600_enable_s3tc = debug_get_bool_option("R600_ENABLE_S3TC", FALSE);
+       }
+
+       if (desc->layout == UTIL_FORMAT_LAYOUT_RGTC) {
+               if (!r600_enable_s3tc)
+                       goto out_unknown;
+
+               switch (format) {
+               case PIPE_FORMAT_RGTC1_SNORM:
+               case PIPE_FORMAT_LATC1_SNORM:
+                       word4 |= sign_bit[0];
+               case PIPE_FORMAT_RGTC1_UNORM:
+               case PIPE_FORMAT_LATC1_UNORM:
+                       result = FMT_BC4;
+                       goto out_word4;
+               case PIPE_FORMAT_RGTC2_SNORM:
+               case PIPE_FORMAT_LATC2_SNORM:
+                       word4 |= sign_bit[0] | sign_bit[1];
+               case PIPE_FORMAT_RGTC2_UNORM:
+               case PIPE_FORMAT_LATC2_UNORM:
+                       result = FMT_BC5;
+                       goto out_word4;
+               default:
+                       goto out_unknown;
+               }
+       }
+
        if (desc->layout == UTIL_FORMAT_LAYOUT_S3TC) {
-                static int r600_enable_s3tc = -1;
 
-                if (r600_enable_s3tc == -1)
-                        r600_enable_s3tc = 
-                                debug_get_bool_option("R600_ENABLE_S3TC", FALSE);
+               if (!r600_enable_s3tc)
+                       goto out_unknown;
 
-                if (!r600_enable_s3tc)
-                        goto out_unknown;
+               if (!util_format_s3tc_enabled) {
+                       goto out_unknown;
+               }
 
                switch (format) {
                case PIPE_FORMAT_DXT1_RGB:
                case PIPE_FORMAT_DXT1_RGBA:
-                        result = FMT_BC1;
-                        goto out_word4;
+               case PIPE_FORMAT_DXT1_SRGB:
+               case PIPE_FORMAT_DXT1_SRGBA:
+                       result = FMT_BC1;
+                       goto out_word4;
                case PIPE_FORMAT_DXT3_RGBA:
-                        result = FMT_BC2;
-                        goto out_word4;
+               case PIPE_FORMAT_DXT3_SRGBA:
+                       result = FMT_BC2;
+                       goto out_word4;
                case PIPE_FORMAT_DXT5_RGBA:
-                        result = FMT_BC3;
-                        goto out_word4;
-                default:
-                        goto out_unknown;
-                }
+               case PIPE_FORMAT_DXT5_SRGBA:
+                       result = FMT_BC3;
+                       goto out_word4;
+               default:
+                       goto out_unknown;
+               }
+       }
+
+       if (format == PIPE_FORMAT_R9G9B9E5_FLOAT) {
+               result = FMT_5_9_9_9_SHAREDEXP;
+               goto out_word4;
+       } else if (format == PIPE_FORMAT_R11G11B10_FLOAT) {
+               result = FMT_10_11_11_FLOAT;
+               goto out_word4;
        }
 
 
-       for (i = 0; i < desc->nr_channels; i++) {       
+       for (i = 0; i < desc->nr_channels; i++) {
                if (desc->channel[i].type == UTIL_FORMAT_TYPE_SIGNED) {
                        word4 |= sign_bit[i];
                }
@@ -904,13 +992,11 @@ uint32_t r600_translate_texformat(enum pipe_format format,
 
        /* R8G8Bx_SNORM - TODO CxV8U8 */
 
-       /* RGTC - TODO */
-
        /* See whether the components are of the same size. */
        for (i = 1; i < desc->nr_channels; i++) {
                uniform = uniform && desc->channel[0].size == desc->channel[i].size;
        }
-       
+
        /* Non-uniform formats. */
        if (!uniform) {
                switch(desc->nr_channels) {
@@ -934,7 +1020,7 @@ uint32_t r600_translate_texformat(enum pipe_format format,
                            desc->channel[1].size == 10 &&
                            desc->channel[2].size == 10 &&
                            desc->channel[3].size == 2) {
-                               result = FMT_10_10_10_2;
+                               result = FMT_2_10_10_10;
                                goto out_word4;
                        }
                        goto out_unknown;
@@ -997,6 +1083,19 @@ uint32_t r600_translate_texformat(enum pipe_format format,
                                result = FMT_16_16_16_16;
                                goto out_word4;
                        }
+                       goto out_unknown;
+               case 32:
+                       switch (desc->nr_channels) {
+                       case 1:
+                               result = FMT_32;
+                               goto out_word4;
+                       case 2:
+                               result = FMT_32_32;
+                               goto out_word4;
+                       case 4:
+                               result = FMT_32_32_32_32;
+                               goto out_word4;
+                       }
                }
                goto out_unknown;
 
@@ -1028,8 +1127,9 @@ uint32_t r600_translate_texformat(enum pipe_format format,
                                goto out_word4;
                        }
                }
-               
+               goto out_unknown;
        }
+
 out_word4:
        if (word4_p)
                *word4_p = word4;
@@ -1037,6 +1137,6 @@ out_word4:
                *yuv_format_p = yuv_format;
        return result;
 out_unknown:
-//     R600_ERR("Unable to handle texformat %d %s\n", format, util_format_name(format));
+       /* R600_ERR("Unable to handle texformat %d %s\n", format, util_format_name(format)); */
        return ~0;
 }