r600g: Get rid of r600_blit_uncompress_depth_ptr.
[mesa.git] / src / gallium / drivers / r600 / r600_texture.c
index 30d79ebdd6ffb3af06ec11754cb38de9c3f040cd..e2745624575c1860cf9423f49df52bec1457a4f6 100644 (file)
  *      Jerome Glisse
  *      Corbin Simpson
  */
+#include <errno.h>
 #include <pipe/p_screen.h>
 #include <util/u_format.h>
 #include <util/u_math.h>
 #include <util/u_inlines.h>
 #include <util/u_memory.h>
 #include "state_tracker/drm_driver.h"
-#include "r600_screen.h"
-#include "r600_context.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;
 
-static unsigned long r600_texture_get_offset(struct r600_resource_texture *rtex,
-                                       unsigned level, unsigned zslice,
-                                       unsigned face)
+/* 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)
 {
-       unsigned long offset = rtex->offset[level];
+       struct pipe_transfer *transfer = (struct pipe_transfer*)rtransfer;
+       struct pipe_resource *texture = transfer->resource;
+
+       ctx->resource_copy_region(ctx, rtransfer->staging_texture,
+                               0, 0, 0, 0, texture, transfer->level,
+                               &transfer->box);
+}
+
+
+/* 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 pipe_transfer *transfer = (struct pipe_transfer*)rtransfer;
+       struct pipe_resource *texture = transfer->resource;
+       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,
+                                 0, &sbox);
+
+       ctx->flush(ctx, 0, NULL);
+}
+
+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) {
        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];
+               return offset + layer * rtex->layer_size[level];
        default:
-               assert(zslice == 0 && face == 0);
+               assert(layer == 0);
                return offset;
        }
 }
 
-static void r600_setup_miptree(struct r600_screen *rscreen, struct r600_resource_texture *rtex)
+static unsigned r600_get_pixel_alignment(struct pipe_screen *screen,
+                                        enum pipe_format format,
+                                        unsigned array_mode)
+{
+       struct r600_screen* rscreen = (struct r600_screen *)screen;
+       unsigned pixsize = util_format_get_blocksize(format);
+       int p_align;
+
+       switch(array_mode) {
+       case V_038000_ARRAY_1D_TILED_THIN1:
+               p_align = MAX2(8,
+                              ((rscreen->tiling_info->group_bytes / 8 / pixsize)));
+               break;
+       case V_038000_ARRAY_2D_TILED_THIN1:
+               p_align = MAX2(rscreen->tiling_info->num_banks,
+                              (((rscreen->tiling_info->group_bytes / 8 / pixsize)) *
+                               rscreen->tiling_info->num_banks)) * 8;
+               break;
+       case V_038000_ARRAY_LINEAR_GENERAL:
+       default:
+               p_align = rscreen->tiling_info->group_bytes / pixsize;
+               break;
+       }
+       return p_align;
+}
+
+static unsigned r600_get_height_alignment(struct pipe_screen *screen,
+                                         unsigned array_mode)
+{
+       struct r600_screen* rscreen = (struct r600_screen *)screen;
+       int h_align;
+
+       switch (array_mode) {
+       case V_038000_ARRAY_2D_TILED_THIN1:
+               h_align = rscreen->tiling_info->num_channels * 8;
+               break;
+       case V_038000_ARRAY_1D_TILED_THIN1:
+               h_align = 8;
+               break;
+       default:
+               h_align = 1;
+               break;
+       }
+       return h_align;
+}
+
+static unsigned r600_get_base_alignment(struct pipe_screen *screen,
+                                       enum pipe_format format,
+                                       unsigned array_mode)
+{
+       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 h_align = r600_get_height_alignment(screen, array_mode);
+       int b_align;
+
+       switch (array_mode) {
+       case V_038000_ARRAY_2D_TILED_THIN1:
+               b_align = MAX2(rscreen->tiling_info->num_banks * rscreen->tiling_info->num_channels * 8 * 8 * pixsize,
+                              p_align * pixsize * h_align);
+               break;
+       case V_038000_ARRAY_1D_TILED_THIN1:
+       default:
+               b_align = rscreen->tiling_info->group_bytes;
+               break;
+       }
+       return b_align;
+}
+
+static unsigned mip_minify(unsigned size, unsigned level)
+{
+       unsigned val;
+       val = u_minify(size, level);
+       if (level > 0)
+               val = util_next_power_of_two(val);
+       return val;
+}
+
+static unsigned r600_texture_get_stride(struct pipe_screen *screen,
+                                       struct r600_resource_texture *rtex,
+                                       unsigned level)
+{
+       struct pipe_resource *ptex = &rtex->resource.base.b;
+       unsigned width, stride, tile_width;
+
+       if (rtex->pitch_override)
+               return rtex->pitch_override;
+
+       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);
+
+       return stride;
+}
+
+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;
+       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);
+}
+
+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;
+
+       switch (array_mode) {
+       case V_0280A0_ARRAY_LINEAR_GENERAL:
+       case V_0280A0_ARRAY_LINEAR_ALIGNED:
+       case V_0280A0_ARRAY_1D_TILED_THIN1:
+       default:
+               rtex->array_mode[level] = array_mode;
+               break;
+       case V_0280A0_ARRAY_2D_TILED_THIN1:
+       {
+               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);
+
+               w = mip_minify(ptex->width0, level);
+               h = mip_minify(ptex->height0, level);
+               if (w < tile_width || h < tile_height)
+                       rtex->array_mode[level] = V_0280A0_ARRAY_1D_TILED_THIN1;
+               else
+                       rtex->array_mode[level] = array_mode;
+       }
+       break;
+       }
+}
+
+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;
-       unsigned long w, h, pitch, size, layer_size, i, offset;
+       struct radeon *radeon = (struct radeon *)screen->winsys;
+       enum chip_class chipc = r600_get_family_class(radeon);
+       unsigned pitch, size, layer_size, i, offset;
+       unsigned nblocksy;
 
-       rtex->bpt = util_format_get_blocksize(ptex->format);
        for (i = 0, offset = 0; i <= ptex->last_level; i++) {
-               w = u_minify(ptex->width0, i);
-               h = u_minify(ptex->height0, i);
-               pitch = util_format_get_stride(ptex->format, align(w, 64));
-               layer_size = pitch * h;
-               if (ptex->target == PIPE_TEXTURE_CUBE)
-                       size = layer_size * 6;
+               r600_texture_set_array_mode(screen, rtex, i, array_mode);
+
+               pitch = r600_texture_get_stride(screen, rtex, i);
+               nblocksy = r600_texture_get_nblocksy(screen, rtex, i);
+
+               layer_size = pitch * nblocksy;
+
+               if (ptex->target == PIPE_TEXTURE_CUBE) {
+                       if (chipc >= R700)
+                               size = layer_size * 8;
+                       else
+                               size = layer_size * 6;
+               }
                else
                        size = layer_size * u_minify(ptex->depth0, i);
+               /* align base image and start of miptree */
+               if ((i == 0) || (i == 1))
+                       offset = align(offset, r600_get_base_alignment(screen, ptex->format, array_mode));
                rtex->offset[i] = offset;
                rtex->layer_size[i] = layer_size;
-               rtex->pitch[i] = pitch;
+               rtex->pitch_in_bytes[i] = pitch;
+               rtex->pitch_in_pixels[i] = pitch_to_width(ptex->format, pitch);
                offset += size;
        }
        rtex->size = offset;
 }
 
-struct pipe_resource *r600_texture_create(struct pipe_screen *screen,
-                                               const struct pipe_resource *templ)
+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_resource_texture *rtex;
        struct r600_resource *resource;
-       struct r600_screen *rscreen = r600_screen(screen);
+       struct radeon *radeon = (struct radeon *)screen->winsys;
 
        rtex = CALLOC_STRUCT(r600_resource_texture);
-       if (!rtex) {
+       if (rtex == NULL)
                return NULL;
-       }
+
        resource = &rtex->resource;
-       resource->base.b = *templ;
+       resource->base.b = *base;
        resource->base.vtbl = &r600_texture_vtbl;
        pipe_reference_init(&resource->base.b.reference, 1);
        resource->base.b.screen = screen;
-       r600_setup_miptree(rscreen, rtex);
+       resource->bo = bo;
+       rtex->pitch_override = pitch_in_bytes_override;
 
-       /* FIXME alignment 4096 enought ? too much ? */
-       resource->domain = r600_domain_from_usage(resource->base.b.bind);
-       resource->bo = radeon_bo(rscreen->rw, 0, rtex->size, 4096, NULL);
-       if (resource->bo == NULL) {
-               FREE(rtex);
-               return NULL;
+       if (array_mode)
+               rtex->tiled = 1;
+       r600_setup_miptree(screen, rtex, array_mode);
+
+       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);
+
+               resource->bo = r600_bo(radeon, rtex->size, base_align, base->bind, base->usage);
+               if (!resource->bo) {
+                       FREE(rtex);
+                       return NULL;
+               }
+       }
+       return rtex;
+}
+
+struct pipe_resource *r600_texture_create(struct pipe_screen *screen,
+                                               const struct pipe_resource *templ)
+{
+       unsigned array_mode = 0;
+       static int force_tiling = -1;
+
+       /* 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) {
+               if (!(templ->flags & R600_RESOURCE_FLAG_TRANSFER) &&
+                   !(templ->bind & PIPE_BIND_SCANOUT)) {
+                       array_mode = V_038000_ARRAY_2D_TILED_THIN1;
+               }
        }
 
-       return &resource->base.b;
+       return (struct pipe_resource *)r600_texture_create_object(screen, templ, array_mode,
+                                                                 0, 0, NULL);
+
 }
 
 static void r600_texture_destroy(struct pipe_screen *screen,
@@ -113,156 +349,385 @@ static void r600_texture_destroy(struct pipe_screen *screen,
 {
        struct r600_resource_texture *rtex = (struct r600_resource_texture*)ptex;
        struct r600_resource *resource = &rtex->resource;
-       struct r600_screen *rscreen = r600_screen(screen);
+       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) {
-               radeon_bo_decref(rscreen->rw, resource->bo);
+               r600_bo_reference(radeon, &resource->bo, NULL);
        }
        FREE(rtex);
 }
 
-static struct pipe_surface *r600_get_tex_surface(struct pipe_screen *screen,
+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);
+}
+
+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 pipe_surface *surface = CALLOC_STRUCT(pipe_surface);
-       unsigned long offset;
+       struct r600_surface *surface = CALLOC_STRUCT(r600_surface);
+       unsigned 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);
-       pipe_reference_init(&surface->reference, 1);
-       pipe_resource_reference(&surface->texture, texture);
-       surface->format = texture->format;
-       surface->width = u_minify(texture->width0, level);
-       surface->height = u_minify(texture->height0, level);
-       surface->offset = offset;
-       surface->usage = flags;
-       surface->zslice = zslice;
-       surface->texture = texture;
-       surface->face = face;
-       surface->level = level;
-       return surface;
+       /* 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.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.usage = surf_tmpl->usage;
+       surface->base.texture = texture;
+       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(pipe->screen, rtex->array_mode[level]);
+       surface->aligned_height = align(surface->base.height, tile_height);
+       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);
 }
 
+
 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 r600_resource_texture *rtex;
-       struct r600_resource *resource;
-       struct radeon_bo *bo = NULL;
-
-       bo = radeon_bo(rw, whandle->handle, 0, 0, NULL);
-       if (bo == NULL) {
-               return NULL;
-       }
+       struct r600_bo *bo = NULL;
+       unsigned array_mode = 0;
 
        /* Support only 2D textures without mipmaps */
-       if (templ->target != PIPE_TEXTURE_2D || templ->depth0 != 1 || templ->last_level != 0)
+       if ((templ->target != PIPE_TEXTURE_2D && templ->target != PIPE_TEXTURE_RECT) ||
+             templ->depth0 != 1 || templ->last_level != 0)
                return NULL;
 
-       rtex = CALLOC_STRUCT(r600_resource_texture);
-       if (rtex == NULL)
+       bo = r600_bo_handle(rw, whandle->handle, &array_mode);
+       if (bo == NULL) {
                return NULL;
+       }
 
-       resource = &rtex->resource;
-       resource->base.b = *templ;
-       resource->base.vtbl = &r600_texture_vtbl;
-       pipe_reference_init(&resource->base.b.reference, 1);
-       resource->base.b.screen = screen;
-       resource->bo = bo;
-       rtex->pitch_override = whandle->stride;
-       rtex->bpt = util_format_get_blocksize(templ->format);
-       rtex->pitch[0] = whandle->stride;
-       rtex->offset[0] = 0;
-       rtex->size = align(rtex->pitch[0] * templ->height0, 64);
-
-       return &resource->base.b;
+       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)
+                                               unsigned level, int layer)
 {
        /* FIXME */
        return PIPE_REFERENCED_FOR_READ | PIPE_REFERENCED_FOR_WRITE;
 }
 
+int r600_texture_depth_flush(struct pipe_context *ctx,
+                            struct pipe_resource *texture)
+{
+       struct r600_resource_texture *rtex = (struct r600_resource_texture*)texture;
+       struct pipe_resource resource;
+
+       if (rtex->flushed_depth_texture)
+               goto out;
+
+       resource.target = PIPE_TEXTURE_2D;
+       resource.format = texture->format;
+       resource.width0 = texture->width0;
+       resource.height0 = texture->height0;
+       resource.depth0 = 1;
+       resource.last_level = 0;
+       resource.nr_samples = 0;
+       resource.usage = PIPE_USAGE_DYNAMIC;
+       resource.bind = 0;
+       resource.flags = R600_RESOURCE_FLAG_TRANSFER;
+
+       resource.bind |= PIPE_BIND_DEPTH_STENCIL;
+
+       rtex->flushed_depth_texture = (struct r600_resource_texture *)ctx->screen->resource_create(ctx->screen, &resource);
+       if (rtex->flushed_depth_texture == NULL) {
+               R600_ERR("failed to create temporary texture to hold untiled copy\n");
+               return -ENOMEM;
+       }
+
+out:
+       /* XXX: only do this if the depth texture has actually changed:
+        */
+       r600_blit_uncompress_depth(ctx, rtex);
+       return 0;
+}
+
+/* Needs adjustment for pixelformat:
+ */
+static INLINE unsigned u_box_volume( const struct pipe_box *box )
+{
+       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)
 {
        struct r600_resource_texture *rtex = (struct r600_resource_texture*)texture;
+       struct pipe_resource resource;
        struct r600_transfer *trans;
+       int r;
+       boolean use_staging_texture = FALSE;
+
+       /* We cannot map a tiled texture directly because the data is
+        * in a different order, therefore we do detiling using a blit.
+        *
+        * Also, use a temporary in GTT memory for read transfers, as
+        * the CPU is much happier reading out of cached system memory
+        * than uncached VRAM.
+        */
+       if (rtex->tiled)
+               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;
 
        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;
-       trans->transfer.stride = rtex->pitch[sr.level];
-       trans->offset = r600_texture_get_offset(rtex, sr.level, box->z, sr.face);
+       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);
+               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;
+               }
+       } 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;
+               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;
+               }
+               /* Create the temporary texture. */
+               trans->staging_texture = ctx->screen->resource_create(ctx->screen, &resource);
+               if (trans->staging_texture == NULL) {
+                       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 =
+                       ((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);
+               }
+               return &trans->transfer;
+       }
+       trans->transfer.stride = rtex->pitch_in_bytes[level];
+       trans->offset = r600_texture_get_offset(rtex, level, box->z);
        return &trans->transfer;
 }
 
 void r600_texture_transfer_destroy(struct pipe_context *ctx,
-                                  struct pipe_transfer *trans)
+                                  struct pipe_transfer *transfer)
 {
-       pipe_resource_reference(&trans->resource, NULL);
-       FREE(trans);
+       struct r600_transfer *rtransfer = (struct r600_transfer*)transfer;
+       struct r600_resource_texture *rtex = (struct r600_resource_texture*)transfer->resource;
+
+       if (rtransfer->staging_texture) {
+               if (transfer->usage & PIPE_TRANSFER_WRITE) {
+                       r600_copy_from_staging_texture(ctx, rtransfer);
+               }
+               pipe_resource_reference(&rtransfer->staging_texture, NULL);
+       }
+       if (rtex->flushed_depth_texture) {
+               pipe_resource_reference((struct pipe_resource **)&rtex->flushed_depth_texture, NULL);
+       }
+       pipe_resource_reference(&transfer->resource, NULL);
+       FREE(transfer);
 }
 
 void* r600_texture_transfer_map(struct pipe_context *ctx,
                                struct pipe_transfer* transfer)
 {
        struct r600_transfer *rtransfer = (struct r600_transfer*)transfer;
-       struct r600_resource *resource;
+       struct r600_bo *bo;
        enum pipe_format format = transfer->resource->format;
-       struct r600_screen *rscreen = r600_screen(ctx->screen);
+       struct radeon *radeon = (struct radeon *)ctx->screen->winsys;
+       unsigned offset = 0;
+       unsigned usage = 0;
        char *map;
 
-       r600_flush(ctx, 0, NULL);
+       if (rtransfer->staging_texture) {
+               bo = ((struct r600_resource *)rtransfer->staging_texture)->bo;
+       } else {
+               struct r600_resource_texture *rtex = (struct r600_resource_texture*)transfer->resource;
 
-       resource = (struct r600_resource *)transfer->resource;
-       if (radeon_bo_map(rscreen->rw, resource->bo)) {
-               return NULL;
+               if (rtex->flushed_depth_texture)
+                       bo = ((struct r600_resource *)rtex->flushed_depth_texture)->bo;
+               else
+                       bo = ((struct r600_resource *)transfer->resource)->bo;
+
+               offset = rtransfer->offset +
+                       transfer->box.y / util_format_get_blockheight(format) * transfer->stride +
+                       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;
        }
-       radeon_bo_wait(rscreen->rw, resource->bo);
 
-       map = resource->bo->data;
+       if (transfer->usage & PIPE_TRANSFER_UNSYNCHRONIZED) {
+               usage |= PB_USAGE_UNSYNCHRONIZED;
+       }
+
+       map = r600_bo_map(radeon, bo, usage, ctx);
+       if (!map) {
+               return NULL;
+       }
 
-       return map + rtransfer->offset +
-               transfer->box.y / util_format_get_blockheight(format) * transfer->stride +
-               transfer->box.x / util_format_get_blockwidth(format) * util_format_get_blocksize(format);
+       return map + offset;
 }
 
 void r600_texture_transfer_unmap(struct pipe_context *ctx,
                                 struct pipe_transfer* transfer)
 {
-       struct r600_screen *rscreen = r600_screen(ctx->screen);
-       struct r600_resource *resource;
+       struct r600_transfer *rtransfer = (struct r600_transfer*)transfer;
+       struct radeon *radeon = (struct radeon *)ctx->screen->winsys;
+       struct r600_bo *bo;
+
+       if (rtransfer->staging_texture) {
+               bo = ((struct r600_resource *)rtransfer->staging_texture)->bo;
+       } else {
+               struct r600_resource_texture *rtex = (struct r600_resource_texture*)transfer->resource;
 
-       resource = (struct r600_resource *)transfer->resource;
-       radeon_bo_unmap(rscreen->rw, resource->bo);
+               if (rtex->flushed_depth_texture) {
+                       bo = ((struct r600_resource *)rtex->flushed_depth_texture)->bo;
+               } else {
+                       bo = ((struct r600_resource *)transfer->resource)->bo;
+               }
+       }
+       r600_bo_unmap(radeon, bo);
 }
 
 struct u_resource_vtbl r600_texture_vtbl =
 {
-       u_default_resource_get_handle,  /* get_handle */
+       r600_texture_get_handle,        /* get_handle */
        r600_texture_destroy,           /* resource_destroy */
        r600_texture_is_referenced,     /* is_resource_referenced */
        r600_texture_get_transfer,      /* get_transfer */
@@ -273,58 +738,58 @@ struct u_resource_vtbl r600_texture_vtbl =
        u_default_transfer_inline_write /* transfer_inline_write */
 };
 
-void r600_init_screen_texture_functions(struct pipe_screen *screen)
+void r600_init_surface_functions(struct r600_pipe_context *r600)
 {
-       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,
-                                         const unsigned char *swizzle_view)
+               const unsigned char *swizzle_view)
 {
-    unsigned i;
-    unsigned char swizzle[4];
-    unsigned result = 0;
-    const uint32_t swizzle_shift[4] = {
-           16, 19, 22, 25,
-    };
-    const uint32_t swizzle_bit[4] = {
-           0, 1, 2, 3,
-    };
-
-    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];
-        }
-    } else {
-        memcpy(swizzle, swizzle_format, 4);
-    }
-
-    /* Get swizzle. */
-    for (i = 0; i < 4; i++) {
-        switch (swizzle[i]) {
-            case UTIL_FORMAT_SWIZZLE_Y:
-                result |= swizzle_bit[1] << swizzle_shift[i];
-                break;
-            case UTIL_FORMAT_SWIZZLE_Z:
-                result |= swizzle_bit[2] << swizzle_shift[i];
-                break;
-            case UTIL_FORMAT_SWIZZLE_W:
-                result |= swizzle_bit[3] << swizzle_shift[i];
-                break;
-            case UTIL_FORMAT_SWIZZLE_0:
-                result |= V_038010_SQ_SEL_0 << swizzle_shift[i];
-                break;
-            case UTIL_FORMAT_SWIZZLE_1:
-                result |= V_038010_SQ_SEL_1 << swizzle_shift[i];
-                break;
-            default: /* UTIL_FORMAT_SWIZZLE_X */
-                result |= swizzle_bit[0] << swizzle_shift[i];
-        }
-    }
-    return result;
+       unsigned i;
+       unsigned char swizzle[4];
+       unsigned result = 0;
+       const uint32_t swizzle_shift[4] = {
+               16, 19, 22, 25,
+       };
+       const uint32_t swizzle_bit[4] = {
+               0, 1, 2, 3,
+       };
+
+       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];
+               }
+       } else {
+               memcpy(swizzle, swizzle_format, 4);
+       }
+
+       /* Get swizzle. */
+       for (i = 0; i < 4; i++) {
+               switch (swizzle[i]) {
+               case UTIL_FORMAT_SWIZZLE_Y:
+                       result |= swizzle_bit[1] << swizzle_shift[i];
+                       break;
+               case UTIL_FORMAT_SWIZZLE_Z:
+                       result |= swizzle_bit[2] << swizzle_shift[i];
+                       break;
+               case UTIL_FORMAT_SWIZZLE_W:
+                       result |= swizzle_bit[3] << swizzle_shift[i];
+                       break;
+               case UTIL_FORMAT_SWIZZLE_0:
+                       result |= V_038010_SQ_SEL_0 << swizzle_shift[i];
+                       break;
+               case UTIL_FORMAT_SWIZZLE_1:
+                       result |= V_038010_SQ_SEL_1 << swizzle_shift[i];
+                       break;
+               default: /* UTIL_FORMAT_SWIZZLE_X */
+                       result |= swizzle_bit[0] << swizzle_shift[i];
+               }
+       }
+       return result;
 }
 
 /* texture format translate */
@@ -344,19 +809,31 @@ uint32_t r600_translate_texformat(enum pipe_format format,
        };
        desc = util_format_description(format);
 
+       word4 |= r600_get_swizzle_combined(desc->swizzle, swizzle_view);
+
        /* Colorspace (return non-RGB formats directly). */
        switch (desc->colorspace) {
                /* Depth stencil formats */
        case UTIL_FORMAT_COLORSPACE_ZS:
                switch (format) {
                case PIPE_FORMAT_Z16_UNORM:
-                       result = V_028010_DEPTH_16;
+                       result = FMT_16;
                        goto out_word4;
+               case PIPE_FORMAT_X24S8_USCALED:
+                       word4 |= S_038010_NUM_FORMAT_ALL(V_038010_SQ_NUM_FORMAT_INT);
                case PIPE_FORMAT_Z24X8_UNORM:
-                       result = V_028010_DEPTH_X8_24;
-                       goto out_word4;
                case PIPE_FORMAT_Z24_UNORM_S8_USCALED:
-                       result = V_028010_DEPTH_8_24;
+                       result = FMT_8_24;
+                       goto out_word4;
+               case PIPE_FORMAT_S8X24_USCALED:
+                       word4 |= S_038010_NUM_FORMAT_ALL(V_038010_SQ_NUM_FORMAT_INT);
+               case PIPE_FORMAT_X8Z24_UNORM:
+               case PIPE_FORMAT_S8_USCALED_Z24_UNORM:
+                       result = FMT_24_8;
+                       goto out_word4;
+               case PIPE_FORMAT_S8_USCALED:
+                       result = FMT_8;
+                       word4 |= S_038010_NUM_FORMAT_ALL(V_038010_SQ_NUM_FORMAT_INT);
                        goto out_word4;
                default:
                        goto out_unknown;
@@ -365,8 +842,8 @@ 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;
                }
@@ -381,12 +858,32 @@ uint32_t r600_translate_texformat(enum pipe_format format,
        default:
                break;
        }
-       
-       word4 |= r600_get_swizzle_combined(desc->swizzle, swizzle_view);
 
        /* S3TC formats. TODO */
        if (desc->layout == UTIL_FORMAT_LAYOUT_S3TC) {
-               goto out_unknown;
+               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;
+
+               switch (format) {
+               case PIPE_FORMAT_DXT1_RGB:
+               case PIPE_FORMAT_DXT1_RGBA:
+                       result = FMT_BC1;
+                       goto out_word4;
+               case PIPE_FORMAT_DXT3_RGBA:
+                       result = FMT_BC2;
+                       goto out_word4;
+               case PIPE_FORMAT_DXT5_RGBA:
+                       result = FMT_BC3;
+                       goto out_word4;
+               default:
+                       goto out_unknown;
+               }
        }
 
 
@@ -412,7 +909,7 @@ uint32_t r600_translate_texformat(enum pipe_format format,
                        if (desc->channel[0].size == 5 &&
                            desc->channel[1].size == 6 &&
                            desc->channel[2].size == 5) {
-                               result |= V_0280A0_COLOR_5_6_5;
+                               result = FMT_5_6_5;
                                goto out_word4;
                        }
                        goto out_unknown;
@@ -421,14 +918,14 @@ uint32_t r600_translate_texformat(enum pipe_format format,
                            desc->channel[1].size == 5 &&
                            desc->channel[2].size == 5 &&
                            desc->channel[3].size == 1) {
-                               result |= V_0280A0_COLOR_1_5_5_5;
+                               result = FMT_1_5_5_5;
                                goto out_word4;
                        }
                        if (desc->channel[0].size == 10 &&
                            desc->channel[1].size == 10 &&
                            desc->channel[2].size == 10 &&
                            desc->channel[3].size == 2) {
-                               result |= V_0280A0_COLOR_10_10_10_2;
+                               result = FMT_10_10_10_2;
                                goto out_word4;
                        }
                        goto out_unknown;
@@ -436,79 +933,89 @@ uint32_t r600_translate_texformat(enum pipe_format format,
                goto out_unknown;
        }
 
+       /* Find the first non-VOID channel. */
+       for (i = 0; i < 4; i++) {
+               if (desc->channel[i].type != UTIL_FORMAT_TYPE_VOID) {
+                       break;
+               }
+       }
+
+       if (i == 4)
+               goto out_unknown;
+
        /* uniform formats */
-       switch (desc->channel[0].type) {
+       switch (desc->channel[i].type) {
        case UTIL_FORMAT_TYPE_UNSIGNED:
        case UTIL_FORMAT_TYPE_SIGNED:
-               if (!desc->channel[0].normalized &&
+               if (!desc->channel[i].normalized &&
                    desc->colorspace != UTIL_FORMAT_COLORSPACE_SRGB) {
                        goto out_unknown;
                }
 
-               switch (desc->channel[0].size) {
+               switch (desc->channel[i].size) {
                case 4:
                        switch (desc->nr_channels) {
                        case 2:
-                               result |= V_0280A0_COLOR_4_4;
+                               result = FMT_4_4;
                                goto out_word4;
                        case 4:
-                               result |= V_0280A0_COLOR_4_4_4_4;
+                               result = FMT_4_4_4_4;
                                goto out_word4;
                        }
                        goto out_unknown;
                case 8:
                        switch (desc->nr_channels) {
                        case 1:
-                               result |= V_0280A0_COLOR_8;
+                               result = FMT_8;
                                goto out_word4;
                        case 2:
-                               result |= V_0280A0_COLOR_8_8;
+                               result = FMT_8_8;
                                goto out_word4;
                        case 4:
-                               result |= V_0280A0_COLOR_8_8_8_8;
+                               result = FMT_8_8_8_8;
                                goto out_word4;
                        }
                        goto out_unknown;
                case 16:
                        switch (desc->nr_channels) {
                        case 1:
-                               result |= V_0280A0_COLOR_16;
+                               result = FMT_16;
                                goto out_word4;
                        case 2:
-                               result |= V_0280A0_COLOR_16_16;
+                               result = FMT_16_16;
                                goto out_word4;
                        case 4:
-                               result |= V_0280A0_COLOR_16_16_16_16;
+                               result = FMT_16_16_16_16;
                                goto out_word4;
                        }
                }
                goto out_unknown;
 
        case UTIL_FORMAT_TYPE_FLOAT:
-               switch (desc->channel[0].size) {
+               switch (desc->channel[i].size) {
                case 16:
                        switch (desc->nr_channels) {
                        case 1:
-                               result |= V_0280A0_COLOR_16_FLOAT;
+                               result = FMT_16_FLOAT;
                                goto out_word4;
                        case 2:
-                               result |= V_0280A0_COLOR_16_16_FLOAT;
+                               result = FMT_16_16_FLOAT;
                                goto out_word4;
                        case 4:
-                               result |= V_0280A0_COLOR_16_16_16_16_FLOAT;
+                               result = FMT_16_16_16_16_FLOAT;
                                goto out_word4;
                        }
                        goto out_unknown;
                case 32:
                        switch (desc->nr_channels) {
                        case 1:
-                               result |= V_0280A0_COLOR_32_FLOAT;
+                               result = FMT_32_FLOAT;
                                goto out_word4;
                        case 2:
-                               result |= V_0280A0_COLOR_32_32_FLOAT;
+                               result = FMT_32_32_FLOAT;
                                goto out_word4;
                        case 4:
-                               result |= V_0280A0_COLOR_32_32_32_32_FLOAT;
+                               result = FMT_32_32_32_32_FLOAT;
                                goto out_word4;
                        }
                }
@@ -519,7 +1026,6 @@ out_word4:
                *word4_p = word4;
        if (yuv_format_p)
                *yuv_format_p = yuv_format;
-//     fprintf(stderr,"returning %08x %08x %08x\n", result, word4, yuv_format);
        return result;
 out_unknown:
 //     R600_ERR("Unable to handle texformat %d %s\n", format, util_format_name(format));