llvmpipe: get rid of tiled/linear layout remains
[mesa.git] / src / gallium / drivers / llvmpipe / lp_texture.c
index ff4773fd7c502c27b853c98bc5d1b2a51a733fb9..0ac3528f7accd7ca5add50498785290a05dcd30d 100644 (file)
@@ -46,9 +46,9 @@
 #include "lp_context.h"
 #include "lp_flush.h"
 #include "lp_screen.h"
-#include "lp_tile_image.h"
 #include "lp_texture.h"
 #include "lp_setup.h"
+#include "lp_state.h"
 
 #include "state_tracker/sw_winsys.h"
 
@@ -59,45 +59,6 @@ static struct llvmpipe_resource resource_list;
 static unsigned id_counter = 0;
 
 
-static INLINE boolean
-resource_is_texture(const struct pipe_resource *resource)
-{
-   switch (resource->target) {
-   case PIPE_BUFFER:
-      return FALSE;
-   case PIPE_TEXTURE_1D:
-   case PIPE_TEXTURE_2D:
-   case PIPE_TEXTURE_RECT:
-   case PIPE_TEXTURE_3D:
-   case PIPE_TEXTURE_CUBE:
-      return TRUE;
-   default:
-      assert(0);
-      return FALSE;
-   }
-}
-
-
-
-/**
- * Allocate storage for llvmpipe_texture::layout array.
- * The number of elements is width_in_tiles * height_in_tiles.
- */
-static enum lp_texture_layout *
-alloc_layout_array(unsigned num_slices, unsigned width, unsigned height)
-{
-   const unsigned tx = align(width, TILE_SIZE) / TILE_SIZE;
-   const unsigned ty = align(height, TILE_SIZE) / TILE_SIZE;
-
-   assert(num_slices * tx * ty > 0);
-   assert(LP_TEX_LAYOUT_NONE == 0); /* calloc'ing LP_TEX_LAYOUT_NONE here */
-
-   return (enum lp_texture_layout *)
-      CALLOC(num_slices * tx * ty, sizeof(enum lp_texture_layout));
-}
-
-
-
 /**
  * Conventional allocation path for non-display textures:
  * Just compute row strides here.  Storage is allocated on demand later.
@@ -111,18 +72,22 @@ llvmpipe_texture_layout(struct llvmpipe_screen *screen,
    unsigned width = pt->width0;
    unsigned height = pt->height0;
    unsigned depth = pt->depth0;
+   uint64_t total_size = 0;
+   unsigned layers = pt->array_size;
 
    assert(LP_MAX_TEXTURE_2D_LEVELS <= LP_MAX_TEXTURE_LEVELS);
    assert(LP_MAX_TEXTURE_3D_LEVELS <= LP_MAX_TEXTURE_LEVELS);
 
    for (level = 0; level <= pt->last_level; level++) {
 
-      /* Row stride and image stride (for linear layout) */
+      /* Row stride and image stride */
       {
          unsigned alignment, nblocksx, nblocksy, block_size;
 
          /* For non-compressed formats we need to align the texture size
           * to the tile size to facilitate render-to-texture.
+          * XXX this blows up 1d/1d array textures by unreasonable
+          * amount (factor 64), probably should do something about it.
           */
          if (util_format_is_compressed(pt->format))
             alignment = 1;
@@ -137,18 +102,16 @@ llvmpipe_texture_layout(struct llvmpipe_screen *screen,
 
          lpr->row_stride[level] = align(nblocksx * block_size, 16);
 
-         lpr->img_stride[level] = lpr->row_stride[level] * nblocksy;
-      }
+         /* if row_stride * height > LP_MAX_TEXTURE_SIZE */
+         if (lpr->row_stride[level] > LP_MAX_TEXTURE_SIZE / nblocksy) {
+            /* image too large */
+            goto fail;
+         }
 
-      /* Size of the image in tiles (for tiled layout) */
-      {
-         const unsigned width_t = align(width, TILE_SIZE) / TILE_SIZE;
-         const unsigned height_t = align(height, TILE_SIZE) / TILE_SIZE;
-         lpr->tiles_per_row[level] = width_t;
-         lpr->tiles_per_image[level] = width_t * height_t;
+         lpr->img_stride[level] = lpr->row_stride[level] * nblocksy;
       }
 
-      /* Number of 3D image slices or cube faces */
+      /* Number of 3D image slices, cube faces or texture array layers */
       {
          unsigned num_slices;
 
@@ -156,12 +119,26 @@ llvmpipe_texture_layout(struct llvmpipe_screen *screen,
             num_slices = 6;
          else if (lpr->base.target == PIPE_TEXTURE_3D)
             num_slices = depth;
+         else if (lpr->base.target == PIPE_TEXTURE_1D_ARRAY ||
+                  lpr->base.target == PIPE_TEXTURE_2D_ARRAY)
+            num_slices = layers;
          else
             num_slices = 1;
 
          lpr->num_slices_faces[level] = num_slices;
+      }
+
+      /* if img_stride * num_slices_faces > LP_MAX_TEXTURE_SIZE */
+      if (lpr->img_stride[level] >
+          LP_MAX_TEXTURE_SIZE / lpr->num_slices_faces[level]) {
+         /* volume too large */
+         goto fail;
+      }
 
-         lpr->layout[level] = alloc_layout_array(num_slices, width, height);
+      total_size += (uint64_t) lpr->num_slices_faces[level]
+                  * (uint64_t) lpr->img_stride[level];
+      if (total_size > LP_MAX_TEXTURE_SIZE) {
+         goto fail;
       }
 
       /* Compute size of next mipmap level */
@@ -171,9 +148,26 @@ llvmpipe_texture_layout(struct llvmpipe_screen *screen,
    }
 
    return TRUE;
+
+fail:
+   return FALSE;
 }
 
 
+/**
+ * Check the size of the texture specified by 'res'.
+ * \return TRUE if OK, FALSE if too large.
+ */
+static boolean
+llvmpipe_can_create_resource(struct pipe_screen *screen,
+                             const struct pipe_resource *res)
+{
+   struct llvmpipe_resource lpr;
+   memset(&lpr, 0, sizeof(lpr));
+   lpr.base = *res;
+   return llvmpipe_texture_layout(llvmpipe_screen(screen), &lpr);
+}
+
 
 static boolean
 llvmpipe_displaytarget_layout(struct llvmpipe_screen *screen,
@@ -184,19 +178,12 @@ llvmpipe_displaytarget_layout(struct llvmpipe_screen *screen,
    /* Round up the surface size to a multiple of the tile size to
     * avoid tile clipping.
     */
-   const unsigned width = align(lpr->base.width0, TILE_SIZE);
-   const unsigned height = align(lpr->base.height0, TILE_SIZE);
-   const unsigned width_t = width / TILE_SIZE;
-   const unsigned height_t = height / TILE_SIZE;
+   const unsigned width = MAX2(1, align(lpr->base.width0, TILE_SIZE));
+   const unsigned height = MAX2(1, align(lpr->base.height0, TILE_SIZE));
 
-   lpr->tiles_per_row[0] = width_t;
-   lpr->tiles_per_image[0] = width_t * height_t;
    lpr->num_slices_faces[0] = 1;
    lpr->img_stride[0] = 0;
 
-   lpr->layout[0] = alloc_layout_array(1, width, height);
-   //lpr->layout[0][0] = LP_TEX_LAYOUT_LINEAR;
-
    lpr->dt = winsys->displaytarget_create(winsys,
                                           lpr->base.bind,
                                           lpr->base.format,
@@ -204,7 +191,20 @@ llvmpipe_displaytarget_layout(struct llvmpipe_screen *screen,
                                           16,
                                           &lpr->row_stride[0] );
 
-   return lpr->dt != NULL;
+   if (lpr->dt == NULL)
+      return FALSE;
+
+   {
+      void *map = winsys->displaytarget_map(winsys, lpr->dt,
+                                            PIPE_TRANSFER_WRITE);
+
+      if (map)
+         memset(map, 0, height * lpr->row_stride[0]);
+
+      winsys->displaytarget_unmap(winsys, lpr->dt);
+   }
+
+   return TRUE;
 }
 
 
@@ -223,32 +223,37 @@ llvmpipe_resource_create(struct pipe_screen *_screen,
 
    /* assert(lpr->base.bind); */
 
-   if (resource_is_texture(&lpr->base)) {
-      if (lpr->base.bind & PIPE_BIND_DISPLAY_TARGET) {
+   if (llvmpipe_resource_is_texture(&lpr->base)) {
+      if (lpr->base.bind & (PIPE_BIND_DISPLAY_TARGET |
+                            PIPE_BIND_SCANOUT |
+                            PIPE_BIND_SHARED)) {
          /* displayable surface */
          if (!llvmpipe_displaytarget_layout(screen, lpr))
             goto fail;
-         assert(lpr->layout[0][0] == LP_TEX_LAYOUT_NONE);
       }
       else {
          /* texture map */
          if (!llvmpipe_texture_layout(screen, lpr))
             goto fail;
-         assert(lpr->layout[0][0] == LP_TEX_LAYOUT_NONE);
       }
-      assert(lpr->layout[0]);
    }
    else {
       /* other data (vertex buffer, const buffer, etc) */
-      const enum pipe_format format = templat->format;
-      const uint w = templat->width0 / util_format_get_blockheight(format);
-      const uint h = templat->height0 / util_format_get_blockwidth(format);
-      const uint d = templat->depth0;
-      const uint bpp = util_format_get_blocksize(format);
-      const uint bytes = w * h * d * bpp;
+      const uint bytes = templat->width0;
+      assert(util_format_get_blocksize(templat->format) == 1);
+      assert(templat->height0 == 1);
+      assert(templat->depth0 == 1);
+      assert(templat->last_level == 0);
       lpr->data = align_malloc(bytes, 16);
+      /*
+       * buffers don't really have stride but it's probably safer
+       * (for code doing same calculations for buffers and textures)
+       * to put something sane in there.
+       */
+      lpr->row_stride[0] = bytes;
       if (!lpr->data)
          goto fail;
+      memset(lpr->data, 0, bytes);
    }
 
    lpr->id = id_counter++;
@@ -267,7 +272,7 @@ llvmpipe_resource_create(struct pipe_screen *_screen,
 
 static void
 llvmpipe_resource_destroy(struct pipe_screen *pscreen,
-                         struct pipe_resource *pt)
+                          struct pipe_resource *pt)
 {
    struct llvmpipe_screen *screen = llvmpipe_screen(pscreen);
    struct llvmpipe_resource *lpr = llvmpipe_resource(pt);
@@ -276,38 +281,12 @@ llvmpipe_resource_destroy(struct pipe_screen *pscreen,
       /* display target */
       struct sw_winsys *winsys = screen->winsys;
       winsys->displaytarget_destroy(winsys, lpr->dt);
-
-      if (lpr->tiled[0].data) {
-         align_free(lpr->tiled[0].data);
-         lpr->tiled[0].data = NULL;
-      }
-
-      FREE(lpr->layout[0]);
    }
-   else if (resource_is_texture(pt)) {
-      /* regular texture */
-      uint level;
-
+   else if (llvmpipe_resource_is_texture(pt)) {
       /* free linear image data */
-      for (level = 0; level < Elements(lpr->linear); level++) {
-         if (lpr->linear[level].data) {
-            align_free(lpr->linear[level].data);
-            lpr->linear[level].data = NULL;
-         }
-      }
-
-      /* free tiled image data */
-      for (level = 0; level < Elements(lpr->tiled); level++) {
-         if (lpr->tiled[level].data) {
-            align_free(lpr->tiled[level].data);
-            lpr->tiled[level].data = NULL;
-         }
-      }
-
-      /* free layout flag arrays */
-      for (level = 0; level < Elements(lpr->tiled); level++) {
-         FREE(lpr->layout[level]);
-         lpr->layout[level] = NULL;
+      if (lpr->linear_img.data) {
+         align_free(lpr->linear_img.data);
+         lpr->linear_img.data = NULL;
       }
    }
    else if (!lpr->userBuffer) {
@@ -329,26 +308,20 @@ llvmpipe_resource_destroy(struct pipe_screen *pscreen,
  */
 void *
 llvmpipe_resource_map(struct pipe_resource *resource,
-                     unsigned face,
-                     unsigned level,
-                     unsigned zslice,
-                      enum lp_texture_usage tex_usage,
-                      enum lp_texture_layout layout)
+                      unsigned level,
+                      unsigned layer,
+                      enum lp_texture_usage tex_usage)
 {
    struct llvmpipe_resource *lpr = llvmpipe_resource(resource);
    uint8_t *map;
 
-   assert(face < 6);
    assert(level < LP_MAX_TEXTURE_LEVELS);
+   assert(layer < (u_minify(resource->depth0, level) + resource->array_size - 1));
 
    assert(tex_usage == LP_TEX_USAGE_READ ||
           tex_usage == LP_TEX_USAGE_READ_WRITE ||
           tex_usage == LP_TEX_USAGE_WRITE_ALL);
 
-   assert(layout == LP_TEX_LAYOUT_NONE ||
-          layout == LP_TEX_LAYOUT_TILED ||
-          layout == LP_TEX_LAYOUT_LINEAR);
-
    if (lpr->dt) {
       /* display target */
       struct llvmpipe_screen *screen = llvmpipe_screen(resource->screen);
@@ -363,34 +336,23 @@ llvmpipe_resource_map(struct pipe_resource *resource,
          dt_usage = PIPE_TRANSFER_READ_WRITE;
       }
 
-      assert(face == 0);
       assert(level == 0);
-      assert(zslice == 0);
+      assert(layer == 0);
 
       /* FIXME: keep map count? */
       map = winsys->displaytarget_map(winsys, lpr->dt, dt_usage);
 
       /* install this linear image in texture data structure */
-      lpr->linear[level].data = map;
+      lpr->linear_img.data = map;
 
       /* make sure tiled data gets converted to linear data */
-      map2 = llvmpipe_get_texture_image(lpr, 0, 0, tex_usage, layout);
-      if (layout == LP_TEX_LAYOUT_LINEAR)
-         assert(map == map2);
-
+      map2 = llvmpipe_get_texture_image(lpr, 0, 0, tex_usage);
       return map2;
    }
-   else if (resource_is_texture(resource)) {
-      /* regular texture */
-      if (resource->target != PIPE_TEXTURE_CUBE) {
-         assert(face == 0);
-      }
-      if (resource->target != PIPE_TEXTURE_3D) {
-         assert(zslice == 0);
-      }
+   else if (llvmpipe_resource_is_texture(resource)) {
 
-      map = llvmpipe_get_texture_image(lpr, face + zslice, level,
-                                       tex_usage, layout);
+      map = llvmpipe_get_texture_image(lpr, layer, level,
+                                       tex_usage);
       return map;
    }
    else {
@@ -404,9 +366,8 @@ llvmpipe_resource_map(struct pipe_resource *resource,
  */
 void
 llvmpipe_resource_unmap(struct pipe_resource *resource,
-                       unsigned face,
                        unsigned level,
-                       unsigned zslice)
+                       unsigned layer)
 {
    struct llvmpipe_resource *lpr = llvmpipe_resource(resource);
 
@@ -415,14 +376,8 @@ llvmpipe_resource_unmap(struct pipe_resource *resource,
       struct llvmpipe_screen *lp_screen = llvmpipe_screen(resource->screen);
       struct sw_winsys *winsys = lp_screen->winsys;
 
-      assert(face == 0);
       assert(level == 0);
-      assert(zslice == 0);
-
-      /* make sure linear image is up to date */
-      (void) llvmpipe_get_texture_image(lpr, face + zslice, level,
-                                        LP_TEX_USAGE_READ,
-                                        LP_TEX_LAYOUT_LINEAR);
+      assert(layer == 0);
 
       winsys->displaytarget_unmap(winsys, lpr->dt);
    }
@@ -434,7 +389,7 @@ llvmpipe_resource_data(struct pipe_resource *resource)
 {
    struct llvmpipe_resource *lpr = llvmpipe_resource(resource);
 
-   assert(!resource_is_texture(resource));
+   assert(!llvmpipe_resource_is_texture(resource));
 
    return lpr->data;
 }
@@ -442,27 +397,23 @@ llvmpipe_resource_data(struct pipe_resource *resource)
 
 static struct pipe_resource *
 llvmpipe_resource_from_handle(struct pipe_screen *screen,
-                             const struct pipe_resource *template,
-                             struct winsys_handle *whandle)
+                              const struct pipe_resource *template,
+                              struct winsys_handle *whandle)
 {
    struct sw_winsys *winsys = llvmpipe_screen(screen)->winsys;
-   struct llvmpipe_resource *lpr = CALLOC_STRUCT(llvmpipe_resource);
-   unsigned width, height, width_t, height_t;
+   struct llvmpipe_resource *lpr;
 
    /* XXX Seems like from_handled depth textures doesn't work that well */
 
-   if (!lpr)
-      return NULL;
+   lpr = CALLOC_STRUCT(llvmpipe_resource);
+   if (!lpr) {
+      goto no_lpr;
+   }
 
    lpr->base = *template;
    pipe_reference_init(&lpr->base.reference, 1);
    lpr->base.screen = screen;
 
-   width = align(lpr->base.width0, TILE_SIZE);
-   height = align(lpr->base.height0, TILE_SIZE);
-   width_t = width / TILE_SIZE;
-   height_t = height / TILE_SIZE;
-
    /*
     * Looks like unaligned displaytargets work just fine,
     * at least sampler/render ones.
@@ -472,8 +423,6 @@ llvmpipe_resource_from_handle(struct pipe_screen *screen,
    assert(lpr->base.height0 == height);
 #endif
 
-   lpr->tiles_per_row[0] = width_t;
-   lpr->tiles_per_image[0] = width_t * height_t;
    lpr->num_slices_faces[0] = 1;
    lpr->img_stride[0] = 0;
 
@@ -481,13 +430,9 @@ llvmpipe_resource_from_handle(struct pipe_screen *screen,
                                                template,
                                                whandle,
                                                &lpr->row_stride[0]);
-   if (!lpr->dt)
-      goto fail;
-
-   lpr->layout[0] = alloc_layout_array(1, lpr->base.width0, lpr->base.height0);
-
-   assert(lpr->layout[0]);
-   assert(lpr->layout[0][0] == LP_TEX_LAYOUT_NONE);
+   if (!lpr->dt) {
+      goto no_dt;
+   }
 
    lpr->id = id_counter++;
 
@@ -497,8 +442,9 @@ llvmpipe_resource_from_handle(struct pipe_screen *screen,
 
    return &lpr->base;
 
- fail:
+no_dt:
    FREE(lpr);
+no_lpr:
    return NULL;
 }
 
@@ -520,34 +466,50 @@ llvmpipe_resource_get_handle(struct pipe_screen *screen,
 
 
 static struct pipe_surface *
-llvmpipe_get_tex_surface(struct pipe_screen *screen,
-                         struct pipe_resource *pt,
-                         unsigned face, unsigned level, unsigned zslice,
-                         unsigned usage)
+llvmpipe_create_surface(struct pipe_context *pipe,
+                        struct pipe_resource *pt,
+                        const struct pipe_surface *surf_tmpl)
 {
    struct pipe_surface *ps;
 
-   assert(level <= pt->last_level);
+   assert(surf_tmpl->u.tex.level <= pt->last_level);
+   if (!(pt->bind & (PIPE_BIND_DEPTH_STENCIL | PIPE_BIND_RENDER_TARGET)))
+      debug_printf("Illegal surface creation without bind flag\n");
 
    ps = CALLOC_STRUCT(pipe_surface);
    if (ps) {
       pipe_reference_init(&ps->reference, 1);
       pipe_resource_reference(&ps->texture, pt);
-      ps->format = pt->format;
-      ps->width = u_minify(pt->width0, level);
-      ps->height = u_minify(pt->height0, level);
-      ps->usage = usage;
-
-      ps->face = face;
-      ps->level = level;
-      ps->zslice = zslice;
+      ps->context = pipe;
+      ps->format = surf_tmpl->format;
+      if (llvmpipe_resource_is_texture(pt)) {
+         assert(surf_tmpl->u.tex.level <= pt->last_level);
+         ps->width = u_minify(pt->width0, surf_tmpl->u.tex.level);
+         ps->height = u_minify(pt->height0, surf_tmpl->u.tex.level);
+         ps->u.tex.level = surf_tmpl->u.tex.level;
+         ps->u.tex.first_layer = surf_tmpl->u.tex.first_layer;
+         ps->u.tex.last_layer = surf_tmpl->u.tex.last_layer;
+         if (ps->u.tex.first_layer != ps->u.tex.last_layer) {
+            debug_printf("creating surface with multiple layers, rendering to first layer only\n");
+         }
+      }
+      else {
+         /* setting width as number of elements should get us correct renderbuffer width */
+         ps->width = surf_tmpl->u.buf.last_element - surf_tmpl->u.buf.first_element + 1;
+         ps->height = pt->height0;
+         ps->u.buf.first_element = surf_tmpl->u.buf.first_element;
+         ps->u.buf.last_element = surf_tmpl->u.buf.last_element;
+         assert(ps->u.buf.first_element <= ps->u.buf.last_element);
+         assert(ps->u.buf.last_element < ps->width);
+      }
    }
    return ps;
 }
 
 
 static void 
-llvmpipe_tex_surface_destroy(struct pipe_surface *surf)
+llvmpipe_surface_destroy(struct pipe_context *pipe,
+                         struct pipe_surface *surf)
 {
    /* Effectively do the texture_update work here - if texture images
     * needed post-processing to put them into hardware layout, this is
@@ -559,18 +521,26 @@ llvmpipe_tex_surface_destroy(struct pipe_surface *surf)
 }
 
 
-static struct pipe_transfer *
-llvmpipe_get_transfer(struct pipe_context *pipe,
-                     struct pipe_resource *resource,
-                     struct pipe_subresource sr,
-                     unsigned usage,
-                     const struct pipe_box *box)
+static void *
+llvmpipe_transfer_map( struct pipe_context *pipe,
+                       struct pipe_resource *resource,
+                       unsigned level,
+                       unsigned usage,
+                       const struct pipe_box *box,
+                       struct pipe_transfer **transfer )
 {
-   struct llvmpipe_resource *lprex = llvmpipe_resource(resource);
-   struct llvmpipe_transfer *lpr;
+   struct llvmpipe_context *llvmpipe = llvmpipe_context(pipe);
+   struct llvmpipe_screen *screen = llvmpipe_screen(pipe->screen);
+   struct llvmpipe_resource *lpr = llvmpipe_resource(resource);
+   struct llvmpipe_transfer *lpt;
+   struct pipe_transfer *pt;
+   ubyte *map;
+   enum pipe_format format;
+   enum lp_texture_usage tex_usage;
+   const char *mode;
 
    assert(resource);
-   assert(sr.level <= resource->last_level);
+   assert(level <= resource->last_level);
 
    /*
     * Transfers, like other pipe operations, must happen in order, so flush the
@@ -580,11 +550,11 @@ llvmpipe_get_transfer(struct pipe_context *pipe,
       boolean read_only = !(usage & PIPE_TRANSFER_WRITE);
       boolean do_not_block = !!(usage & PIPE_TRANSFER_DONTBLOCK);
       if (!llvmpipe_flush_resource(pipe, resource,
-                                   sr.face, sr.level,
-                                   0, /* flush_flags */
+                                   level,
                                    read_only,
                                    TRUE, /* cpu_access */
-                                   do_not_block)) {
+                                   do_not_block,
+                                   __FUNCTION__)) {
          /*
           * It would have blocked, but state tracker requested no to.
           */
@@ -593,49 +563,32 @@ llvmpipe_get_transfer(struct pipe_context *pipe,
       }
    }
 
-   lpr = CALLOC_STRUCT(llvmpipe_transfer);
-   if (lpr) {
-      struct pipe_transfer *pt = &lpr->base;
-      pipe_resource_reference(&pt->resource, resource);
-      pt->box = *box;
-      pt->sr = sr;
-      pt->stride = lprex->row_stride[sr.level];
-      pt->slice_stride = lprex->img_stride[sr.level];
-      pt->usage = usage;
-
-      return pt;
+   /* Check if we're mapping the current constant buffer */
+   if ((usage & PIPE_TRANSFER_WRITE) &&
+       (resource->bind & PIPE_BIND_CONSTANT_BUFFER)) {
+      unsigned i;
+      for (i = 0; i < Elements(llvmpipe->constants[PIPE_SHADER_FRAGMENT]); ++i) {
+         if (resource == llvmpipe->constants[PIPE_SHADER_FRAGMENT][i].buffer) {
+            /* constants may have changed */
+            llvmpipe->dirty |= LP_NEW_CONSTANTS;
+            break;
+         }
+      }
    }
-   return NULL;
-}
-
 
-static void 
-llvmpipe_transfer_destroy(struct pipe_context *pipe,
-                              struct pipe_transfer *transfer)
-{
-   /* Effectively do the texture_update work here - if texture images
-    * needed post-processing to put them into hardware layout, this is
-    * where it would happen.  For llvmpipe, nothing to do.
-    */
-   assert (transfer->resource);
-   pipe_resource_reference(&transfer->resource, NULL);
-   FREE(transfer);
-}
-
-
-static void *
-llvmpipe_transfer_map( struct pipe_context *pipe,
-                       struct pipe_transfer *transfer )
-{
-   struct llvmpipe_screen *screen = llvmpipe_screen(pipe->screen);
-   ubyte *map;
-   struct llvmpipe_resource *lpr;
-   enum pipe_format format;
-   enum lp_texture_usage tex_usage;
-   const char *mode;
+   lpt = CALLOC_STRUCT(llvmpipe_transfer);
+   if (!lpt)
+      return NULL;
+   pt = &lpt->base;
+   pipe_resource_reference(&pt->resource, resource);
+   pt->box = *box;
+   pt->level = level;
+   pt->stride = lpr->row_stride[level];
+   pt->layer_stride = lpr->img_stride[level];
+   pt->usage = usage;
+   *transfer = pt;
 
-   assert(transfer->sr.face < 6);
-   assert(transfer->sr.level < LP_MAX_TEXTURE_LEVELS);
+   assert(level < LP_MAX_TEXTURE_LEVELS);
 
    /*
    printf("tex_transfer_map(%d, %d  %d x %d of %d x %d,  usage %d )\n",
@@ -645,7 +598,7 @@ llvmpipe_transfer_map( struct pipe_context *pipe,
           transfer->usage);
    */
 
-   if (transfer->usage == PIPE_TRANSFER_READ) {
+   if (usage == PIPE_TRANSFER_READ) {
       tex_usage = LP_TEX_USAGE_READ;
       mode = "read";
    }
@@ -655,34 +608,29 @@ llvmpipe_transfer_map( struct pipe_context *pipe,
    }
 
    if (0) {
-      struct llvmpipe_resource *lpr = llvmpipe_resource(transfer->resource);
       printf("transfer map tex %u  mode %s\n", lpr->id, mode);
    }
 
-
-   assert(transfer->resource);
-   lpr = llvmpipe_resource(transfer->resource);
    format = lpr->base.format;
 
-   map = llvmpipe_resource_map(transfer->resource,
-                              transfer->sr.face,
-                              transfer->sr.level,
-                              transfer->box.z,
-                               tex_usage, LP_TEX_LAYOUT_LINEAR);
+   map = llvmpipe_resource_map(resource,
+                               level,
+                               box->z,
+                               tex_usage);
 
 
    /* May want to do different things here depending on read/write nature
     * of the map:
     */
-   if (transfer->usage & PIPE_TRANSFER_WRITE) {
+   if (usage & PIPE_TRANSFER_WRITE) {
       /* Do something to notify sharing contexts of a texture change.
        */
       screen->timestamp++;
    }
-   
+
    map +=
-      transfer->box.y / util_format_get_blockheight(format) * transfer->stride +
-      transfer->box.x / util_format_get_blockwidth(format) * util_format_get_blocksize(format);
+      box->y / util_format_get_blockheight(format) * pt->stride +
+      box->x / util_format_get_blockwidth(format) * util_format_get_blocksize(format);
 
    return map;
 }
@@ -695,30 +643,73 @@ llvmpipe_transfer_unmap(struct pipe_context *pipe,
    assert(transfer->resource);
 
    llvmpipe_resource_unmap(transfer->resource,
-                          transfer->sr.face,
-                          transfer->sr.level,
-                          transfer->box.z);
+                           transfer->level,
+                           transfer->box.z);
+
+   /* Effectively do the texture_update work here - if texture images
+    * needed post-processing to put them into hardware layout, this is
+    * where it would happen.  For llvmpipe, nothing to do.
+    */
+   assert (transfer->resource);
+   pipe_resource_reference(&transfer->resource, NULL);
+   FREE(transfer);
 }
 
-static unsigned int
+unsigned int
 llvmpipe_is_resource_referenced( struct pipe_context *pipe,
-                               struct pipe_resource *presource,
-                               unsigned face, unsigned level)
+                                 struct pipe_resource *presource,
+                                 unsigned level)
 {
    struct llvmpipe_context *llvmpipe = llvmpipe_context( pipe );
 
-   if (presource->target == PIPE_BUFFER)
-      return PIPE_UNREFERENCED;
-   
+   /*
+    * XXX checking only resources with the right bind flags
+    * is unsafe since with opengl state tracker we can end up
+    * with resources bound to places they weren't supposed to be
+    * (buffers bound as sampler views is one possibility here).
+    */
+   if (!(presource->bind & (PIPE_BIND_DEPTH_STENCIL |
+                            PIPE_BIND_RENDER_TARGET |
+                            PIPE_BIND_SAMPLER_VIEW)))
+      return LP_UNREFERENCED;
+
    return lp_setup_is_resource_referenced(llvmpipe->setup, presource);
 }
 
 
+/**
+ * Returns the largest possible alignment for a format in llvmpipe
+ */
+unsigned
+llvmpipe_get_format_alignment( enum pipe_format format )
+{
+   const struct util_format_description *desc = util_format_description(format);
+   unsigned size = 0;
+   unsigned bytes;
+   unsigned i;
+
+   for (i = 0; i < desc->nr_channels; ++i) {
+      size += desc->channel[i].size;
+   }
+
+   bytes = size / 8;
+
+   if (!util_is_power_of_two(bytes)) {
+      bytes /= desc->nr_channels;
+   }
+
+   if (bytes % 2 || bytes < 1) {
+      return 1;
+   } else {
+      return bytes;
+   }
+}
+
 
 /**
  * Create buffer which wraps user-space data.
  */
-static struct pipe_resource *
+struct pipe_resource *
 llvmpipe_user_buffer_create(struct pipe_screen *screen,
                             void *ptr,
                             unsigned bytes,
@@ -739,6 +730,7 @@ llvmpipe_user_buffer_create(struct pipe_screen *screen,
    buffer->base.width0 = bytes;
    buffer->base.height0 = 1;
    buffer->base.depth0 = 1;
+   buffer->base.array_size = 1;
    buffer->userBuffer = TRUE;
    buffer->data = ptr;
 
@@ -748,33 +740,12 @@ llvmpipe_user_buffer_create(struct pipe_screen *screen,
 
 /**
  * Compute size (in bytes) need to store a texture image / mipmap level,
- * for just one cube face or one 3D texture slice
+ * for just one cube face, one array layer or one 3D texture slice
  */
 static unsigned
-tex_image_face_size(const struct llvmpipe_resource *lpr, unsigned level,
-                    enum lp_texture_layout layout)
+tex_image_face_size(const struct llvmpipe_resource *lpr, unsigned level)
 {
-   const unsigned width = u_minify(lpr->base.width0, level);
-   const unsigned height = u_minify(lpr->base.height0, level);
-
-   assert(layout == LP_TEX_LAYOUT_TILED ||
-          layout == LP_TEX_LAYOUT_LINEAR);
-
-   if (layout == LP_TEX_LAYOUT_TILED) {
-      /* for tiled layout, force a 32bpp format */
-      const enum pipe_format format = PIPE_FORMAT_B8G8R8A8_UNORM;
-      const unsigned block_size = util_format_get_blocksize(format);
-      const unsigned nblocksy =
-         util_format_get_nblocksy(format, align(height, TILE_SIZE));
-      const unsigned nblocksx =
-         util_format_get_nblocksx(format, align(width, TILE_SIZE));
-      const unsigned buffer_size = block_size * nblocksy * nblocksx;
-      return buffer_size;
-   }
-   else {
-      /* we already computed this */
-      return lpr->img_stride[level];
-   }
+   return lpr->img_stride[level];
 }
 
 
@@ -783,182 +754,72 @@ tex_image_face_size(const struct llvmpipe_resource *lpr, unsigned level,
  * including all cube faces or 3D image slices
  */
 static unsigned
-tex_image_size(const struct llvmpipe_resource *lpr, unsigned level,
-               enum lp_texture_layout layout)
+tex_image_size(const struct llvmpipe_resource *lpr, unsigned level)
 {
-   const unsigned buf_size = tex_image_face_size(lpr, level, layout);
+   const unsigned buf_size = tex_image_face_size(lpr, level);
    return buf_size * lpr->num_slices_faces[level];
 }
 
 
-/**
- * This function encapsulates some complicated logic for determining
- * how to convert a tile of image data from linear layout to tiled
- * layout, or vice versa.
- * \param cur_layout  the current tile layout
- * \param target_layout  the desired tile layout
- * \param usage  how the tile will be accessed (R/W vs. read-only, etc)
- * \param new_layout_return  returns the new layout mode
- * \param convert_return  returns TRUE if image conversion is needed
- */
-static void
-layout_logic(enum lp_texture_layout cur_layout,
-             enum lp_texture_layout target_layout,
-             enum lp_texture_usage usage,
-             enum lp_texture_layout *new_layout_return,
-             boolean *convert)
-{
-   enum lp_texture_layout other_layout, new_layout;
-
-   *convert = FALSE;
-
-   new_layout = 99; /* debug check */
-
-   if (target_layout == LP_TEX_LAYOUT_LINEAR) {
-      other_layout = LP_TEX_LAYOUT_TILED;
-   }
-   else {
-      assert(target_layout == LP_TEX_LAYOUT_TILED);
-      other_layout = LP_TEX_LAYOUT_LINEAR;
-   }
-
-   new_layout = target_layout;  /* may get changed below */
-
-   if (cur_layout == LP_TEX_LAYOUT_BOTH) {
-      if (usage == LP_TEX_USAGE_READ) {
-         new_layout = LP_TEX_LAYOUT_BOTH;
-      }
-   }
-   else if (cur_layout == other_layout) {
-      if (usage != LP_TEX_USAGE_WRITE_ALL) {
-         /* need to convert tiled data to linear or vice versa */
-         *convert = TRUE;
-
-         if (usage == LP_TEX_USAGE_READ)
-            new_layout = LP_TEX_LAYOUT_BOTH;
-      }
-   }
-   else {
-      assert(cur_layout == LP_TEX_LAYOUT_NONE ||
-             cur_layout == target_layout);
-   }
-
-   assert(new_layout == LP_TEX_LAYOUT_BOTH ||
-          new_layout == target_layout);
-
-   *new_layout_return = new_layout;
-}
-
-
 /**
  * Return pointer to a 2D texture image/face/slice.
  * No tiled/linear conversion is done.
  */
 ubyte *
 llvmpipe_get_texture_image_address(struct llvmpipe_resource *lpr,
-                                   unsigned face_slice, unsigned level,
-                                   enum lp_texture_layout layout)
+                                   unsigned face_slice, unsigned level)
 {
    struct llvmpipe_texture_image *img;
    unsigned offset;
 
-   if (layout == LP_TEX_LAYOUT_LINEAR) {
-      img = &lpr->linear[level];
-   }
-   else {
-      assert (layout == LP_TEX_LAYOUT_TILED);
-      img = &lpr->tiled[level];
-   }
+   img = &lpr->linear_img;
+   offset = lpr->linear_mip_offsets[level];
 
    if (face_slice > 0)
-      offset = face_slice * tex_image_face_size(lpr, level, layout);
-   else
-      offset = 0;
+      offset += face_slice * tex_image_face_size(lpr, level);
 
    return (ubyte *) img->data + offset;
 }
 
 
-static INLINE enum lp_texture_layout
-llvmpipe_get_texture_tile_layout(const struct llvmpipe_resource *lpr,
-                                 unsigned face_slice, unsigned level,
-                                 unsigned x, unsigned y)
-{
-   uint i;
-   assert(resource_is_texture(&lpr->base));
-   assert(x < lpr->tiles_per_row[level]);
-   i = face_slice * lpr->tiles_per_image[level]
-      + y * lpr->tiles_per_row[level] + x;
-   return lpr->layout[level][i];
-}
-
-
-static INLINE void
-llvmpipe_set_texture_tile_layout(struct llvmpipe_resource *lpr,
-                                 unsigned face_slice, unsigned level,
-                                 unsigned x, unsigned y,
-                                 enum lp_texture_layout layout)
-{
-   uint i;
-   assert(resource_is_texture(&lpr->base));
-   assert(x < lpr->tiles_per_row[level]);
-   i = face_slice * lpr->tiles_per_image[level]
-      + y * lpr->tiles_per_row[level] + x;
-   lpr->layout[level][i] = layout;
-}
-
-
-/**
- * Set the layout mode for all tiles in a particular image.
- */
-static INLINE void
-llvmpipe_set_texture_image_layout(struct llvmpipe_resource *lpr,
-                                  unsigned face_slice, unsigned level,
-                                  unsigned width_t, unsigned height_t,
-                                  enum lp_texture_layout layout)
-{
-   const unsigned start = face_slice * lpr->tiles_per_image[level];
-   unsigned i;
-
-   for (i = 0; i < width_t * height_t; i++) {
-      lpr->layout[level][start + i] = layout;
-   }
-}
-
-
 /**
- * Allocate storage for a linear or tile texture image (all cube
- * faces and all 3D slices.
+ * Allocate storage for a linear image
+ * (all cube faces and all 3D slices, all levels).
  */
 static void
-alloc_image_data(struct llvmpipe_resource *lpr, unsigned level,
-                 enum lp_texture_layout layout)
+alloc_image_data(struct llvmpipe_resource *lpr)
 {
    uint alignment = MAX2(16, util_cpu_caps.cacheline);
+   uint level;
+   uint offset = 0;
 
-   if (lpr->dt)
-      assert(level == 0);
+   if (lpr->dt) {
+      /* we get the linear memory from the winsys, and it has
+       * already been zeroed
+       */
+      struct llvmpipe_screen *screen = llvmpipe_screen(lpr->base.screen);
+      struct sw_winsys *winsys = screen->winsys;
 
-   if (layout == LP_TEX_LAYOUT_TILED) {
-      /* tiled data is stored in regular memory */
-      uint buffer_size = tex_image_size(lpr, level, layout);
-      lpr->tiled[level].data = align_malloc(buffer_size, alignment);
+      assert(lpr->base.last_level == 0);
+
+      lpr->linear_img.data =
+         winsys->displaytarget_map(winsys, lpr->dt,
+                                   PIPE_TRANSFER_READ_WRITE);
    }
    else {
-      assert(layout == LP_TEX_LAYOUT_LINEAR);
-      if (lpr->dt) {
-         /* we get the linear memory from the winsys */
-         struct llvmpipe_screen *screen = llvmpipe_screen(lpr->base.screen);
-         struct sw_winsys *winsys = screen->winsys;
-
-         lpr->linear[0].data =
-            winsys->displaytarget_map(winsys, lpr->dt,
-                                      PIPE_TRANSFER_READ_WRITE);
+      /* not a display target - allocate regular memory */
+      /*
+       * Offset calculation for start of a specific mip/layer is always
+       * offset = lpr->linear_mip_offsets[level] + lpr->img_stride[level] * layer
+       */
+      for (level = 0; level <= lpr->base.last_level; level++) {
+         uint buffer_size = tex_image_size(lpr, level);
+         lpr->linear_mip_offsets[level] = offset;
+         offset += align(buffer_size, alignment);
       }
-      else {
-         /* not a display target - allocate regular memory */
-         uint buffer_size = tex_image_size(lpr, level, LP_TEX_LAYOUT_LINEAR);
-         lpr->linear[level].data = align_malloc(buffer_size, alignment);
+      lpr->linear_img.data = align_malloc(offset, alignment);
+      if (lpr->linear_img.data) {
+         memset(lpr->linear_img.data, 0, offset);
       }
    }
 }
@@ -966,142 +827,48 @@ alloc_image_data(struct llvmpipe_resource *lpr, unsigned level,
 
 
 /**
- * Return pointer to texture image data (either linear or tiled layout)
+ * Return pointer to texture image data
  * for a particular cube face or 3D texture slice.
  *
  * \param face_slice  the cube face or 3D slice of interest
  * \param usage  one of LP_TEX_USAGE_READ/WRITE_ALL/READ_WRITE
- * \param layout  either LP_TEX_LAYOUT_LINEAR or _TILED or _NONE
  */
 void *
 llvmpipe_get_texture_image(struct llvmpipe_resource *lpr,
                            unsigned face_slice, unsigned level,
-                           enum lp_texture_usage usage,
-                           enum lp_texture_layout layout)
+                           enum lp_texture_usage usage)
 {
-   /*
-    * 'target' refers to the image which we're retrieving (either in
-    * tiled or linear layout).
-    * 'other' refers to the same image but in the other layout. (it may
-    *  or may not exist.
-    */
    struct llvmpipe_texture_image *target_img;
-   struct llvmpipe_texture_image *other_img;
    void *target_data;
-   void *other_data;
-   const unsigned width = u_minify(lpr->base.width0, level);
-   const unsigned height = u_minify(lpr->base.height0, level);
-   const unsigned width_t = align(width, TILE_SIZE) / TILE_SIZE;
-   const unsigned height_t = align(height, TILE_SIZE) / TILE_SIZE;
-   enum lp_texture_layout other_layout;
-   boolean only_allocate;
-
-   assert(layout == LP_TEX_LAYOUT_NONE ||
-          layout == LP_TEX_LAYOUT_TILED ||
-          layout == LP_TEX_LAYOUT_LINEAR);
+   unsigned target_offset;
+   unsigned *target_off_ptr;
 
    assert(usage == LP_TEX_USAGE_READ ||
           usage == LP_TEX_USAGE_READ_WRITE ||
           usage == LP_TEX_USAGE_WRITE_ALL);
 
-   /* check for the special case of layout == LP_TEX_LAYOUT_NONE */
-   if (layout == LP_TEX_LAYOUT_NONE) {
-      only_allocate = TRUE;
-      layout = LP_TEX_LAYOUT_TILED;
-   }
-   else {
-      only_allocate = FALSE;
-   }
-
    if (lpr->dt) {
-      assert(lpr->linear[level].data);
-   }
-
-   /* which is target?  which is other? */
-   if (layout == LP_TEX_LAYOUT_LINEAR) {
-      target_img = &lpr->linear[level];
-      other_img = &lpr->tiled[level];
-      other_layout = LP_TEX_LAYOUT_TILED;
-   }
-   else {
-      target_img = &lpr->tiled[level];
-      other_img = &lpr->linear[level];
-      other_layout = LP_TEX_LAYOUT_LINEAR;
+      assert(lpr->linear_img.data);
    }
 
+   target_img = &lpr->linear_img;
+   target_off_ptr = lpr->linear_mip_offsets;
    target_data = target_img->data;
-   other_data = other_img->data;
 
    if (!target_data) {
       /* allocate memory for the target image now */
-      alloc_image_data(lpr, level, layout);
+      alloc_image_data(lpr);
       target_data = target_img->data;
    }
 
-   if (face_slice > 0) {
-      unsigned target_offset, other_offset;
+   target_offset = target_off_ptr[level];
 
-      target_offset = face_slice * tex_image_face_size(lpr, level, layout);
-      other_offset = face_slice * tex_image_face_size(lpr, level, other_layout);
-      if (target_data) {
-         target_data = (uint8_t *) target_data + target_offset;
-      }
-      if (other_data) {
-         other_data = (uint8_t *) other_data + other_offset;
-      }
+   if (face_slice > 0) {
+      target_offset += face_slice * tex_image_face_size(lpr, level);
    }
 
-   if (only_allocate) {
-      /* Just allocating tiled memory.  Don't initialize it from the
-       * linear data if it exists.
-       */
-      return target_data;
-   }
-
-   if (other_data) {
-      /* may need to convert other data to the requested layout */
-      enum lp_texture_layout new_layout;
-      unsigned x, y;
-
-      /* loop over all image tiles, doing layout conversion where needed */
-      for (y = 0; y < height_t; y++) {
-         for (x = 0; x < width_t; x++) {
-            enum lp_texture_layout cur_layout =
-               llvmpipe_get_texture_tile_layout(lpr, face_slice, level, x, y);
-            boolean convert;
-
-            layout_logic(cur_layout, layout, usage, &new_layout, &convert);
-
-            if (convert && other_data && target_data) {
-               if (layout == LP_TEX_LAYOUT_TILED) {
-                  lp_linear_to_tiled(other_data, target_data,
-                                     x * TILE_SIZE, y * TILE_SIZE,
-                                     TILE_SIZE, TILE_SIZE,
-                                     lpr->base.format,
-                                     lpr->row_stride[level],
-                                     lpr->tiles_per_row[level]);
-               }
-               else {
-                  assert(layout == LP_TEX_LAYOUT_LINEAR);
-                  lp_tiled_to_linear(other_data, target_data,
-                                     x * TILE_SIZE, y * TILE_SIZE,
-                                     TILE_SIZE, TILE_SIZE,
-                                     lpr->base.format,
-                                     lpr->row_stride[level],
-                                     lpr->tiles_per_row[level]);
-               }
-            }
-
-            if (new_layout != cur_layout)
-               llvmpipe_set_texture_tile_layout(lpr, face_slice, level, x, y,
-                                                new_layout);
-         }
-      }
-   }
-   else {
-      /* no other data */
-      llvmpipe_set_texture_image_layout(lpr, face_slice, level,
-                                        width_t, height_t, layout);
+   if (target_data) {
+      target_data = (uint8_t *) target_data + target_offset;
    }
 
    return target_data;
@@ -1110,15 +877,12 @@ llvmpipe_get_texture_image(struct llvmpipe_resource *lpr,
 
 /**
  * Return pointer to start of a texture image (1D, 2D, 3D, CUBE).
- * All cube faces and 3D slices will be converted to the requested
- * layout if needed.
  * This is typically used when we're about to sample from a texture.
  */
 void *
 llvmpipe_get_texture_image_all(struct llvmpipe_resource *lpr,
                                unsigned level,
-                               enum lp_texture_usage usage,
-                               enum lp_texture_layout layout)
+                               enum lp_texture_usage usage)
 {
    const int slices = lpr->num_slices_faces[level];
    int slice;
@@ -1127,7 +891,7 @@ llvmpipe_get_texture_image_all(struct llvmpipe_resource *lpr,
    assert(slices > 0);
 
    for (slice = slices - 1; slice >= 0; slice--) {
-      map = llvmpipe_get_texture_image(lpr, slice, level, usage, layout);
+      map = llvmpipe_get_texture_image(lpr, slice, level, usage);
    }
 
    return map;
@@ -1135,9 +899,7 @@ llvmpipe_get_texture_image_all(struct llvmpipe_resource *lpr,
 
 
 /**
- * Get pointer to a linear image (not the tile!) where the tile at (x,y)
- * is known to be in linear layout.
- * Conversion from tiled to linear will be done if necessary.
+ * Get pointer to a linear image (not the tile!) at tile (x,y).
  * \return pointer to start of image/face (not the tile)
  */
 ubyte *
@@ -1146,188 +908,24 @@ llvmpipe_get_texture_tile_linear(struct llvmpipe_resource *lpr,
                                  enum lp_texture_usage usage,
                                  unsigned x, unsigned y)
 {
-   struct llvmpipe_texture_image *linear_img = &lpr->linear[level];
-   enum lp_texture_layout cur_layout, new_layout;
-   const unsigned tx = x / TILE_SIZE, ty = y / TILE_SIZE;
-   boolean convert;
-   uint8_t *tiled_image, *linear_image;
-
-   assert(resource_is_texture(&lpr->base));
-   assert(x % TILE_SIZE == 0);
-   assert(y % TILE_SIZE == 0);
-
-   if (!linear_img->data) {
-      /* allocate memory for the linear image now */
-      alloc_image_data(lpr, level, LP_TEX_LAYOUT_LINEAR);
-   }
-
-   /* compute address of the slice/face of the image that contains the tile */
-   tiled_image = llvmpipe_get_texture_image_address(lpr, face_slice, level,
-                                                    LP_TEX_LAYOUT_TILED);
-   linear_image = llvmpipe_get_texture_image_address(lpr, face_slice, level,
-                                                     LP_TEX_LAYOUT_LINEAR);
-
-   /* get current tile layout and determine if data conversion is needed */
-   cur_layout = llvmpipe_get_texture_tile_layout(lpr, face_slice, level, tx, ty);
-
-   layout_logic(cur_layout, LP_TEX_LAYOUT_LINEAR, usage,
-                &new_layout, &convert);
-
-   if (convert && tiled_image && linear_image) {
-      lp_tiled_to_linear(tiled_image, linear_image,
-                         x, y, TILE_SIZE, TILE_SIZE, lpr->base.format,
-                         lpr->row_stride[level],
-                         lpr->tiles_per_row[level]);
-   }
-
-   if (new_layout != cur_layout)
-      llvmpipe_set_texture_tile_layout(lpr, face_slice, level, tx, ty, new_layout);
-
-   return linear_image;
-}
-
-
-/**
- * Get pointer to tiled data for rendering.
- * \return pointer to the tiled data at the given tile position
- */
-ubyte *
-llvmpipe_get_texture_tile(struct llvmpipe_resource *lpr,
-                          unsigned face_slice, unsigned level,
-                          enum lp_texture_usage usage,
-                          unsigned x, unsigned y)
-{
-   struct llvmpipe_texture_image *tiled_img = &lpr->tiled[level];
-   enum lp_texture_layout cur_layout, new_layout;
-   const unsigned tx = x / TILE_SIZE, ty = y / TILE_SIZE;
-   boolean convert;
-   uint8_t *tiled_image, *linear_image;
-   unsigned tile_offset;
-
-   assert(x % TILE_SIZE == 0);
-   assert(y % TILE_SIZE == 0);
-
-   if (!tiled_img->data) {
-      /* allocate memory for the tiled image now */
-      alloc_image_data(lpr, level, LP_TEX_LAYOUT_TILED);
-   }
-
-   /* compute address of the slice/face of the image that contains the tile */
-   tiled_image = llvmpipe_get_texture_image_address(lpr, face_slice, level,
-                                                    LP_TEX_LAYOUT_TILED);
-   linear_image = llvmpipe_get_texture_image_address(lpr, face_slice, level,
-                                                     LP_TEX_LAYOUT_LINEAR);
-
-   /* get current tile layout and see if we need to convert the data */
-   cur_layout = llvmpipe_get_texture_tile_layout(lpr, face_slice, level, tx, ty);
-
-   layout_logic(cur_layout, LP_TEX_LAYOUT_TILED, usage, &new_layout, &convert);
-   if (convert && linear_image && tiled_image) {
-      lp_linear_to_tiled(linear_image, tiled_image,
-                         x, y, TILE_SIZE, TILE_SIZE, lpr->base.format,
-                         lpr->row_stride[level],
-                         lpr->tiles_per_row[level]);
-   }
-
-   if (!tiled_image)
-      return NULL;
-
-   if (new_layout != cur_layout)
-      llvmpipe_set_texture_tile_layout(lpr, face_slice, level, tx, ty, new_layout);
-
-   /* compute, return address of the 64x64 tile */
-   tile_offset = (ty * lpr->tiles_per_row[level] + tx)
-         * TILE_SIZE * TILE_SIZE * 4;
-
-   return (ubyte *) tiled_image + tile_offset;
-}
-
-
-/**
- * Get pointer to tiled data for rendering.
- * \return pointer to the tiled data at the given tile position
- */
-void
-llvmpipe_unswizzle_cbuf_tile(struct llvmpipe_resource *lpr,
-                             unsigned face_slice, unsigned level,
-                             unsigned x, unsigned y,
-                             uint8_t *tile)
-{
-   struct llvmpipe_texture_image *linear_img = &lpr->linear[level];
-   const unsigned tx = x / TILE_SIZE, ty = y / TILE_SIZE;
+   struct llvmpipe_texture_image *linear_img = &lpr->linear_img;
    uint8_t *linear_image;
 
+   assert(llvmpipe_resource_is_texture(&lpr->base));
    assert(x % TILE_SIZE == 0);
    assert(y % TILE_SIZE == 0);
 
    if (!linear_img->data) {
       /* allocate memory for the linear image now */
-      alloc_image_data(lpr, level, LP_TEX_LAYOUT_LINEAR);
+      /* XXX should probably not do that here? */
+      alloc_image_data(lpr);
    }
+   assert(linear_img->data);
 
    /* compute address of the slice/face of the image that contains the tile */
-   linear_image = llvmpipe_get_texture_image_address(lpr, face_slice, level,
-                                                     LP_TEX_LAYOUT_LINEAR);
-
-   {
-      uint ii = x, jj = y;
-      uint tile_offset = jj / TILE_SIZE + ii / TILE_SIZE;
-      uint byte_offset = tile_offset * TILE_SIZE * TILE_SIZE * 4;
-      
-      /* Note that lp_tiled_to_linear expects the tile parameter to
-       * point at the first tile in a whole-image sized array.  In
-       * this code, we have only a single tile and have to do some
-       * pointer arithmetic to figure out where the "image" would have
-       * started.
-       */
-      lp_tiled_to_linear(tile - byte_offset, linear_image,
-                         x, y, TILE_SIZE, TILE_SIZE,
-                         lpr->base.format,
-                         lpr->row_stride[level],
-                         1);       /* tiles per row */
-   }
-
-   llvmpipe_set_texture_tile_layout(lpr, face_slice, level, tx, ty,
-                                    LP_TEX_LAYOUT_LINEAR);
-}
+   linear_image = llvmpipe_get_texture_image_address(lpr, face_slice, level);
 
-
-/**
- * Get pointer to tiled data for rendering.
- * \return pointer to the tiled data at the given tile position
- */
-void
-llvmpipe_swizzle_cbuf_tile(struct llvmpipe_resource *lpr,
-                           unsigned face_slice, unsigned level,
-                           unsigned x, unsigned y,
-                           uint8_t *tile)
-{
-   uint8_t *linear_image;
-
-   assert(x % TILE_SIZE == 0);
-   assert(y % TILE_SIZE == 0);
-
-   /* compute address of the slice/face of the image that contains the tile */
-   linear_image = llvmpipe_get_texture_image_address(lpr, face_slice, level,
-                                                     LP_TEX_LAYOUT_LINEAR);
-
-   if (linear_image) {
-      uint ii = x, jj = y;
-      uint tile_offset = jj / TILE_SIZE + ii / TILE_SIZE;
-      uint byte_offset = tile_offset * TILE_SIZE * TILE_SIZE * 4;
-
-      /* Note that lp_linear_to_tiled expects the tile parameter to
-       * point at the first tile in a whole-image sized array.  In
-       * this code, we have only a single tile and have to do some
-       * pointer arithmetic to figure out where the "image" would have
-       * started.
-       */
-      lp_linear_to_tiled(linear_image, tile - byte_offset,
-                         x, y, TILE_SIZE, TILE_SIZE,
-                         lpr->base.format,
-                         lpr->row_stride[level],
-                         1);       /* tiles per row */
-   }
+   return linear_image;
 }
 
 
@@ -1340,12 +938,14 @@ llvmpipe_resource_size(const struct pipe_resource *resource)
    const struct llvmpipe_resource *lpr = llvmpipe_resource_const(resource);
    unsigned lvl, size = 0;
 
-   for (lvl = 0; lvl <= lpr->base.last_level; lvl++) {
-      if (lpr->linear[lvl].data)
-         size += tex_image_size(lpr, lvl, LP_TEX_LAYOUT_LINEAR);
-
-      if (lpr->tiled[lvl].data)
-         size += tex_image_size(lpr, lvl, LP_TEX_LAYOUT_TILED);
+   if (llvmpipe_resource_is_texture(resource)) {
+      for (lvl = 0; lvl <= lpr->base.last_level; lvl++) {
+         if (lpr->linear_img.data)
+            size += tex_image_size(lpr, lvl);
+      }
+   }
+   else {
+      size = resource->width0;
    }
 
    return size;
@@ -1393,22 +993,19 @@ llvmpipe_init_screen_resource_funcs(struct pipe_screen *screen)
    screen->resource_destroy = llvmpipe_resource_destroy;
    screen->resource_from_handle = llvmpipe_resource_from_handle;
    screen->resource_get_handle = llvmpipe_resource_get_handle;
-   screen->user_buffer_create = llvmpipe_user_buffer_create;
-
-   screen->get_tex_surface = llvmpipe_get_tex_surface;
-   screen->tex_surface_destroy = llvmpipe_tex_surface_destroy;
+   screen->can_create_resource = llvmpipe_can_create_resource;
 }
 
 
 void
 llvmpipe_init_context_resource_funcs(struct pipe_context *pipe)
 {
-   pipe->get_transfer = llvmpipe_get_transfer;
-   pipe->transfer_destroy = llvmpipe_transfer_destroy;
    pipe->transfer_map = llvmpipe_transfer_map;
    pipe->transfer_unmap = llvmpipe_transfer_unmap;
-   pipe->is_resource_referenced = llvmpipe_is_resource_referenced;
  
    pipe->transfer_flush_region = u_default_transfer_flush_region;
    pipe->transfer_inline_write = u_default_transfer_inline_write;
+
+   pipe->create_surface = llvmpipe_create_surface;
+   pipe->surface_destroy = llvmpipe_surface_destroy;
 }