gallium: extend pipe_context::flush for it to accept an END_OF_FRAME flag
[mesa.git] / src / gallium / drivers / i915 / i915_resource_texture.c
index 92ba3ddb81e08e09324f3c86e72947f3192cce43..d9f58c01ac19e4daab958e56d439c7fbc1d184ff 100644 (file)
 #include "util/u_format.h"
 #include "util/u_math.h"
 #include "util/u_memory.h"
+#include "util/u_rect.h"
 
 #include "i915_context.h"
 #include "i915_resource.h"
 #include "i915_screen.h"
 #include "i915_winsys.h"
+#include "i915_debug.h"
 
 
+#define DEBUG_TEXTURES 0
+
 /*
  * Helper function and arrays
  */
  * Initial offset for Cube map.
  */
 static const int initial_offsets[6][2] = {
-   {0, 0},
-   {0, 2},
-   {1, 0},
-   {1, 2},
-   {1, 1},
-   {1, 3}
+   [PIPE_TEX_FACE_POS_X] = {0, 0},
+   [PIPE_TEX_FACE_POS_Y] = {1, 0},
+   [PIPE_TEX_FACE_POS_Z] = {1, 1},
+   [PIPE_TEX_FACE_NEG_X] = {0, 2},
+   [PIPE_TEX_FACE_NEG_Y] = {1, 2},
+   [PIPE_TEX_FACE_NEG_Z] = {1, 3},
 };
 
 /**
  * Step offsets for Cube map.
  */
 static const int step_offsets[6][2] = {
-   {0, 2},
-   {0, 2},
-   {-1, 2},
-   {-1, 2},
-   {-1, 1},
-   {-1, 1}
+   [PIPE_TEX_FACE_POS_X] = { 0, 2},
+   [PIPE_TEX_FACE_POS_Y] = {-1, 2},
+   [PIPE_TEX_FACE_POS_Z] = {-1, 1},
+   [PIPE_TEX_FACE_NEG_X] = { 0, 2},
+   [PIPE_TEX_FACE_NEG_Y] = {-1, 2},
+   [PIPE_TEX_FACE_NEG_Z] = {-1, 1},
 };
 
-/* XXX really need twice the size if x is already pot?
-   Otherwise just use util_next_power_of_two?
-*/
-static unsigned
-power_of_two(unsigned x)
+/**
+ * For compressed level 2
+ */
+static const int bottom_offsets[6] = {
+   [PIPE_TEX_FACE_POS_X] = 16 + 0 * 8,
+   [PIPE_TEX_FACE_POS_Y] = 16 + 1 * 8,
+   [PIPE_TEX_FACE_POS_Z] = 16 + 2 * 8,
+   [PIPE_TEX_FACE_NEG_X] = 16 + 3 * 8,
+   [PIPE_TEX_FACE_NEG_Y] = 16 + 4 * 8,
+   [PIPE_TEX_FACE_NEG_Z] = 16 + 5 * 8,
+};
+
+static INLINE unsigned
+align_nblocksx(enum pipe_format format, unsigned width, unsigned align_to)
+{
+   return align(util_format_get_nblocksx(format, width), align_to);
+}
+
+static INLINE unsigned
+align_nblocksy(enum pipe_format format, unsigned width, unsigned align_to)
+{
+   return align(util_format_get_nblocksy(format, width), align_to);
+}
+
+static INLINE unsigned
+get_pot_stride(enum pipe_format format, unsigned width)
 {
-   unsigned value = 1;
-   while (value < x)
-      value = value << 1;
-   return value;
+   return util_next_power_of_two(util_format_get_stride(format, width));
 }
 
+static INLINE const char*
+get_tiling_string(enum i915_winsys_buffer_tile tile)
+{
+   switch(tile) {
+   case I915_TILE_NONE:
+      return "none";
+   case I915_TILE_X:
+      return "x";
+   case I915_TILE_Y:
+      return "y";
+   default:
+      assert(FALSE);
+      return "?";
+   }
+}
+
+
 /*
  * More advanced helper funcs
  */
@@ -92,48 +131,58 @@ power_of_two(unsigned x)
 
 static void
 i915_texture_set_level_info(struct i915_texture *tex,
-                             unsigned level,
-                             unsigned nr_images,
-                             unsigned w, unsigned h, unsigned d)
+                            unsigned level, unsigned nr_images)
 {
    assert(level < Elements(tex->nr_images));
+   assert(nr_images);
+   assert(!tex->image_offset[level]);
 
    tex->nr_images[level] = nr_images;
+   tex->image_offset[level] = MALLOC(nr_images * sizeof(struct offset_pair));
+   tex->image_offset[level][0].nblocksx = 0;
+   tex->image_offset[level][0].nblocksy = 0;
+}
 
-   /*
-   DBG("%s level %d size: %d,%d,%d offset %d,%d (0x%x)\n", __FUNCTION__,
-       level, w, h, d, x, y, tex->level_offset[level]);
-   */
-
-   /* Not sure when this would happen, but anyway: 
-    */
-   if (tex->image_offset[level]) {
-      FREE(tex->image_offset[level]);
-      tex->image_offset[level] = NULL;
-   }
-
-   assert(nr_images);
-   assert(!tex->image_offset[level]);
+INLINE unsigned i915_texture_offset(struct i915_texture *tex,
+                                    unsigned level, unsigned layer)
+{
+   unsigned x, y;
+   x = tex->image_offset[level][layer].nblocksx
+      * util_format_get_blocksize(tex->b.b.format);
+   y = tex->image_offset[level][layer].nblocksy;
 
-   tex->image_offset[level] = (unsigned *) MALLOC(nr_images * sizeof(unsigned));
-   tex->image_offset[level][0] = 0;
+   return y * tex->stride + x;
 }
 
 static void
 i915_texture_set_image_offset(struct i915_texture *tex,
-                              unsigned level, unsigned img, unsigned x, unsigned y)
+                              unsigned level, unsigned img,
+                              unsigned nblocksx, unsigned nblocksy)
 {
-   if (img == 0 && level == 0)
-      assert(x == 0 && y == 0);
-
+   /* for the first image and level make sure offset is zero */
+   assert(!(img == 0 && level == 0) || (nblocksx == 0 && nblocksy == 0));
    assert(img < tex->nr_images[level]);
 
-   tex->image_offset[level][img] = y * tex->stride + x * util_format_get_blocksize(tex->b.b.format);
+   tex->image_offset[level][img].nblocksx = nblocksx;
+   tex->image_offset[level][img].nblocksy = nblocksy;
 
-   /*
-   printf("%s level %d img %d pos %d,%d image_offset %x\n",
-       __FUNCTION__, level, img, x, y, tex->image_offset[level][img]);
-   */
+#if DEBUG_TEXTURES
+   debug_printf("%s: %p level %u, img %u (%u, %u)\n", __FUNCTION__,
+                tex, level, img, x, y);
+#endif
+}
+
+static enum i915_winsys_buffer_tile
+i915_texture_tiling(struct i915_screen *is, struct i915_texture *tex)
+{
+   if (!is->debug.tiling)
+      return I915_TILE_NONE;
+
+   if (tex->b.b.target == PIPE_TEXTURE_1D)
+      return I915_TILE_NONE;
+
+   /* Use X tiling for 2D, 3D and compressed textures */
+   return I915_TILE_X;
 }
 
 
@@ -153,26 +202,26 @@ i9x5_scanout_layout(struct i915_texture *tex)
    if (pt->last_level > 0 || util_format_get_blocksize(pt->format) != 4)
       return FALSE;
 
-   i915_texture_set_level_info(tex, 0, 1,
-                               pt->width0,
-                               pt->height0,
-                               1);
+   i915_texture_set_level_info(tex, 0, 1);
    i915_texture_set_image_offset(tex, 0, 0, 0, 0);
 
    if (pt->width0 >= 240) {
-      tex->stride = power_of_two(util_format_get_stride(pt->format, pt->width0));
-      tex->total_nblocksy = align(util_format_get_nblocksy(pt->format, pt->height0), 8);
-      tex->hw_tiled = I915_TILE_X;
+      tex->stride = align(util_format_get_stride(pt->format, pt->width0), 64);
+      tex->total_nblocksy = align_nblocksy(pt->format, pt->height0, 8);
+      tex->tiling = I915_TILE_X;
+   /* special case for cursors */
    } else if (pt->width0 == 64 && pt->height0 == 64) {
-      tex->stride = power_of_two(util_format_get_stride(pt->format, pt->width0));
-      tex->total_nblocksy = align(util_format_get_nblocksy(pt->format, pt->height0), 8);
+      tex->stride = get_pot_stride(pt->format, pt->width0);
+      tex->total_nblocksy = align_nblocksy(pt->format, pt->height0, 8);
    } else {
       return FALSE;
    }
 
+#if DEBUG_TEXTURE
    debug_printf("%s size: %d,%d,%d offset %d,%d (0x%x)\n", __FUNCTION__,
       pt->width0, pt->height0, util_format_get_blocksize(pt->format),
       tex->stride, tex->total_nblocksy, tex->stride * tex->total_nblocksy);
+#endif
 
    return TRUE;
 }
@@ -192,23 +241,82 @@ i9x5_display_target_layout(struct i915_texture *tex)
    if (pt->width0 < 240)
       return FALSE;
 
-   i915_texture_set_level_info(tex, 0, 1,
-                               pt->width0,
-                               pt->height0,
-                               1);
+   i915_texture_set_level_info(tex, 0, 1);
    i915_texture_set_image_offset(tex, 0, 0, 0, 0);
 
-   tex->stride = power_of_two(util_format_get_stride(pt->format, pt->width0));
-   tex->total_nblocksy = align(util_format_get_nblocksy(pt->format, pt->height0), 8);
-   tex->hw_tiled = I915_TILE_X;
+   tex->stride = align(util_format_get_stride(pt->format, pt->width0), 64);
+   tex->total_nblocksy = align_nblocksy(pt->format, pt->height0, 8);
+   tex->tiling = I915_TILE_X;
 
+#if DEBUG_TEXTURE
    debug_printf("%s size: %d,%d,%d offset %d,%d (0x%x)\n", __FUNCTION__,
       pt->width0, pt->height0, util_format_get_blocksize(pt->format),
       tex->stride, tex->total_nblocksy, tex->stride * tex->total_nblocksy);
+#endif
 
    return TRUE;
 }
 
+/**
+ * Helper function for special layouts
+ */
+static boolean
+i9x5_special_layout(struct i915_texture *tex)
+{
+   struct pipe_resource *pt = &tex->b.b;
+
+   /* Scanouts needs special care */
+   if (pt->bind & PIPE_BIND_SCANOUT)
+      if (i9x5_scanout_layout(tex))
+         return TRUE;
+
+   /* Shared buffers needs to be compatible with X servers
+    *
+    * XXX: need a better name than shared for this if it is to be part
+    * of core gallium, and probably move the flag to resource.flags,
+    * rather than bindings.
+    */
+   if (pt->bind & (PIPE_BIND_SHARED | PIPE_BIND_DISPLAY_TARGET))
+      if (i9x5_display_target_layout(tex))
+         return TRUE;
+
+   return FALSE;
+}
+
+/**
+ * Cube layout used on i915 and for non-compressed textures on i945.
+ */
+static void
+i9x5_texture_layout_cube(struct i915_texture *tex)
+{
+   struct pipe_resource *pt = &tex->b.b;
+   const unsigned nblocks = util_format_get_nblocksx(pt->format, pt->width0);
+   unsigned level;
+   unsigned face;
+
+   assert(pt->width0 == pt->height0); /* cubemap images are square */
+
+   /* double pitch for cube layouts */
+   tex->stride = align(nblocks * util_format_get_blocksize(pt->format) * 2, 4);
+   tex->total_nblocksy = nblocks * 4;
+
+   for (level = 0; level <= pt->last_level; level++)
+      i915_texture_set_level_info(tex, level, 6);
+
+   for (face = 0; face < 6; face++) {
+      unsigned x = initial_offsets[face][0] * nblocks;
+      unsigned y = initial_offsets[face][1] * nblocks;
+      unsigned d = nblocks;
+
+      for (level = 0; level <= pt->last_level; level++) {
+         i915_texture_set_image_offset(tex, level, face, x, y);
+         d >>= 1;
+         x += step_offsets[face][0] * d;
+         y += step_offsets[face][1] * d;
+      }
+   }
+}
+
 
 /*
  * i915 layout functions
@@ -223,36 +331,23 @@ i915_texture_layout_2d(struct i915_texture *tex)
    unsigned width = pt->width0;
    unsigned height = pt->height0;
    unsigned nblocksy = util_format_get_nblocksy(pt->format, pt->width0);
+   unsigned align_y = 2;
 
-   /* used for scanouts that need special layouts */
-   if (pt->bind & PIPE_BIND_SCANOUT)
-      if (i9x5_scanout_layout(tex))
-         return;
-
-   /* shared buffers needs to be compatible with X servers 
-    * 
-    * XXX: need a better name than shared for this if it is to be part
-    * of core gallium, and probably move the flag to resource.flags,
-    * rather than bindings.
-    */
-   if (pt->bind & (PIPE_BIND_SHARED | PIPE_BIND_DISPLAY_TARGET))
-      if (i9x5_display_target_layout(tex))
-         return;
+   if (util_format_is_s3tc(pt->format))
+      align_y = 1;
 
    tex->stride = align(util_format_get_stride(pt->format, pt->width0), 4);
    tex->total_nblocksy = 0;
 
    for (level = 0; level <= pt->last_level; level++) {
-      i915_texture_set_level_info(tex, level, 1, width, height, 1);
+      i915_texture_set_level_info(tex, level, 1);
       i915_texture_set_image_offset(tex, level, 0, 0, tex->total_nblocksy);
 
-      nblocksy = align(MAX2(2, nblocksy), 2);
-
       tex->total_nblocksy += nblocksy;
 
       width = u_minify(width, 1);
       height = u_minify(height, 1);
-      nblocksy = util_format_get_nblocksy(pt->format, height);
+      nblocksy = align_nblocksy(pt->format, height, align_y);
    }
 }
 
@@ -275,7 +370,7 @@ i915_texture_layout_3d(struct i915_texture *tex)
    /* XXX: hardware expects/requires 9 levels at minimum.
     */
    for (level = 0; level <= MAX2(8, pt->last_level); level++) {
-      i915_texture_set_level_info(tex, level, depth, width, height, depth);
+      i915_texture_set_level_info(tex, level, depth);
 
       stack_nblocksy += MAX2(2, nblocksy);
 
@@ -301,56 +396,21 @@ i915_texture_layout_3d(struct i915_texture *tex)
    tex->total_nblocksy = stack_nblocksy * pt->depth0;
 }
 
-static void
-i915_texture_layout_cube(struct i915_texture *tex)
-{
-   struct pipe_resource *pt = &tex->b.b;
-   unsigned width = pt->width0, height = pt->height0;
-   const unsigned nblocks = util_format_get_nblocksx(pt->format, pt->width0);
-   unsigned level;
-   unsigned face;
-
-   assert(width == height); /* cubemap images are square */
-
-   /* double pitch for cube layouts */
-   tex->stride = align(nblocks * util_format_get_blocksize(pt->format) * 2, 4);
-   tex->total_nblocksy = nblocks * 4;
-
-   for (level = 0; level <= pt->last_level; level++) {
-      i915_texture_set_level_info(tex, level, 6, width, height, 1);
-      width /= 2;
-      height /= 2;
-   }
-
-   for (face = 0; face < 6; face++) {
-      unsigned x = initial_offsets[face][0] * nblocks;
-      unsigned y = initial_offsets[face][1] * nblocks;
-      unsigned d = nblocks;
-
-      for (level = 0; level <= pt->last_level; level++) {
-         i915_texture_set_image_offset(tex, level, face, x, y);
-         d >>= 1;
-         x += step_offsets[face][0] * d;
-         y += step_offsets[face][1] * d;
-      }
-   }
-}
-
 static boolean
 i915_texture_layout(struct i915_texture * tex)
 {
-   struct pipe_resource *pt = &tex->b.b;
-
-   switch (pt->target) {
+   switch (tex->b.b.target) {
    case PIPE_TEXTURE_1D:
    case PIPE_TEXTURE_2D:
-      i915_texture_layout_2d(tex);
+   case PIPE_TEXTURE_RECT:
+      if (!i9x5_special_layout(tex))
+         i915_texture_layout_2d(tex);
       break;
    case PIPE_TEXTURE_3D:
       i915_texture_layout_3d(tex);
       break;
    case PIPE_TEXTURE_CUBE:
-      i915_texture_layout_cube(tex);
+      i9x5_texture_layout_cube(tex);
       break;
    default:
       assert(0);
@@ -370,7 +430,7 @@ static void
 i945_texture_layout_2d(struct i915_texture *tex)
 {
    struct pipe_resource *pt = &tex->b.b;
-   const int align_x = 2, align_y = 4;
+   int align_x = 4, align_y = 2;
    unsigned level;
    unsigned x = 0;
    unsigned y = 0;
@@ -379,15 +439,10 @@ i945_texture_layout_2d(struct i915_texture *tex)
    unsigned nblocksx = util_format_get_nblocksx(pt->format, pt->width0);
    unsigned nblocksy = util_format_get_nblocksy(pt->format, pt->height0);
 
-   /* used for scanouts that need special layouts */
-   if (tex->b.b.bind & PIPE_BIND_SCANOUT)
-      if (i9x5_scanout_layout(tex))
-         return;
-
-   /* shared buffers needs to be compatible with X servers */
-   if (tex->b.b.bind & (PIPE_BIND_SHARED | PIPE_BIND_DISPLAY_TARGET))
-      if (i9x5_display_target_layout(tex))
-         return;
+   if (util_format_is_s3tc(pt->format)) {
+      align_x = 1;
+      align_y = 1;
+   }
 
    tex->stride = align(util_format_get_stride(pt->format, pt->width0), 4);
 
@@ -397,9 +452,9 @@ i945_texture_layout_2d(struct i915_texture *tex)
     * 2nd mipmap level out past the width of its parent.
     */
    if (pt->last_level > 0) {
-      unsigned mip1_nblocksx 
-         = align(util_format_get_nblocksx(pt->format, u_minify(width, 1)), align_x)
-         + util_format_get_nblocksx(pt->format, u_minify(width, 2));
+      unsigned mip1_nblocksx =
+         align_nblocksx(pt->format, u_minify(pt->width0, 1), align_x) +
+         util_format_get_nblocksx(pt->format, u_minify(pt->width0, 2));
 
       if (mip1_nblocksx > nblocksx)
          tex->stride = mip1_nblocksx * util_format_get_blocksize(pt->format);
@@ -411,11 +466,9 @@ i945_texture_layout_2d(struct i915_texture *tex)
    tex->total_nblocksy = 0;
 
    for (level = 0; level <= pt->last_level; level++) {
-      i915_texture_set_level_info(tex, level, 1, width, height, 1);
+      i915_texture_set_level_info(tex, level, 1);
       i915_texture_set_image_offset(tex, level, 0, x, y);
 
-      nblocksy = align(nblocksy, align_y);
-
       /* Because the images are packed better, the final offset
        * might not be the maximal one:
        */
@@ -424,16 +477,15 @@ i945_texture_layout_2d(struct i915_texture *tex)
       /* Layout_below: step right after second mipmap level.
        */
       if (level == 1) {
-         x += align(nblocksx, align_x);
-      }
-      else {
+         x += nblocksx;
+      } else {
          y += nblocksy;
       }
 
       width  = u_minify(width, 1);
       height = u_minify(height, 1);
-      nblocksx = util_format_get_nblocksx(pt->format, width);
-      nblocksy = util_format_get_nblocksy(pt->format, height);
+      nblocksx = align_nblocksx(pt->format, width, align_x);
+      nblocksy = align_nblocksy(pt->format, height, align_y);
    }
 }
 
@@ -461,7 +513,7 @@ i945_texture_layout_3d(struct i915_texture *tex)
       int y = 0;
       unsigned q, j;
 
-      i915_texture_set_level_info(tex, level, depth, width, height, depth);
+      i915_texture_set_level_info(tex, level, depth);
 
       for (q = 0; q < depth;) {
          for (j = 0; j < pack_x_nr && q < depth; j++, q++) {
@@ -496,99 +548,94 @@ static void
 i945_texture_layout_cube(struct i915_texture *tex)
 {
    struct pipe_resource *pt = &tex->b.b;
-   unsigned level;
-
    const unsigned nblocks = util_format_get_nblocksx(pt->format, pt->width0);
+   const unsigned dim = pt->width0;
+   unsigned level;
    unsigned face;
-   unsigned width = pt->width0;
-   unsigned height = pt->height0;
 
-   /*
-   printf("%s %i, %i\n", __FUNCTION__, pt->width0, pt->height0);
-   */
-
-   assert(width == height); /* cubemap images are square */
+   assert(pt->width0 == pt->height0); /* cubemap images are square */
+   assert(util_next_power_of_two(pt->width0) == pt->width0); /* npot only */
+   assert(util_format_is_s3tc(pt->format)); /* compressed only */
 
    /*
-    * XXX Should only be used for compressed formats. But lets
-    * keep this code active just in case.
-    *
     * Depending on the size of the largest images, pitch can be
     * determined either by the old-style packing of cubemap faces,
     * or the final row of 4x4, 2x2 and 1x1 faces below this.
+    *
+    * 64  * 2 / 4 = 32
+    * 14 * 2 = 28
     */
-   if (nblocks > 32)
-      tex->stride = align(nblocks * util_format_get_blocksize(pt->format) * 2, 4);
+   if (pt->width0 >= 64)
+      tex->stride = nblocks * 2 * util_format_get_blocksize(pt->format);
    else
-      tex->stride = 14 * 8 * util_format_get_blocksize(pt->format);
+      tex->stride = 14 * 2 * util_format_get_blocksize(pt->format);
 
-   tex->total_nblocksy = nblocks * 4;
+   /*
+    * Something similary apply for height as well.
+    */
+   if (pt->width0 >= 4)
+      tex->total_nblocksy = nblocks * 4 + 1;
+   else
+      tex->total_nblocksy = 1;
 
-   /* Set all the levels to effectively occupy the whole rectangular region.
-   */
-   for (level = 0; level <= pt->last_level; level++) {
-      i915_texture_set_level_info(tex, level, 6, width, height, 1);
-      width /= 2;
-      height /= 2;
-   }
+   /* Set all the levels to effectively occupy the whole rectangular region */
+   for (level = 0; level <= pt->last_level; level++)
+      i915_texture_set_level_info(tex, level, 6);
 
    for (face = 0; face < 6; face++) {
-      unsigned x = initial_offsets[face][0] * nblocks;
-      unsigned y = initial_offsets[face][1] * nblocks;
-      unsigned d = nblocks;
+      /* all calculations in pixels */
+      unsigned total_height = tex->total_nblocksy * 4;
+      unsigned x = initial_offsets[face][0] * dim;
+      unsigned y = initial_offsets[face][1] * dim;
+      unsigned d = dim;
 
-#if 0 /* Fix and enable this code for compressed formats */
-      if (nblocks == 4 && face >= 4) {
-         y = tex->total_height - 4;
+      if (dim == 4 && face >= 4) {
          x = (face - 4) * 8;
-      }
-      else if (nblocks < 4 && (face > 0)) {
-         y = tex->total_height - 4;
+         y = tex->total_nblocksy * 4 - 4; /* 4 = 1 block */
+      } else if (dim < 4 && (face > 0)) {
          x = face * 8;
+         y = total_height - 4;
       }
-#endif
 
       for (level = 0; level <= pt->last_level; level++) {
-         i915_texture_set_image_offset(tex, level, face, x, y);
+         i915_texture_set_image_offset(tex, level, face,
+                                       util_format_get_nblocksx(pt->format, x),
+                                       util_format_get_nblocksy(pt->format, y));
 
          d >>= 1;
 
-#if 0 /* Fix and enable this code for compressed formats */
          switch (d) {
-            case 4:
-               switch (face) {
-                  case PIPE_TEX_FACE_POS_X:
-                  case PIPE_TEX_FACE_NEG_X:
-                     x += step_offsets[face][0] * d;
-                     y += step_offsets[face][1] * d;
-                     break;
-                  case PIPE_TEX_FACE_POS_Y:
-                  case PIPE_TEX_FACE_NEG_Y:
-                     y += 12;
-                     x -= 8;
-                     break;
-                  case PIPE_TEX_FACE_POS_Z:
-                  case PIPE_TEX_FACE_NEG_Z:
-                     y = tex->total_height - 4;
-                     x = (face - 4) * 8;
-                     break;
-               }
-            case 2:
-               y = tex->total_height - 4;
-               x = 16 + face * 8;
-               break;
-
-            case 1:
-               x += 48;
-               break;
-            default:
-#endif
+         case 4:
+            switch (face) {
+            case PIPE_TEX_FACE_POS_X:
+            case PIPE_TEX_FACE_NEG_X:
                x += step_offsets[face][0] * d;
                y += step_offsets[face][1] * d;
-#if 0
                break;
+            case PIPE_TEX_FACE_POS_Y:
+            case PIPE_TEX_FACE_NEG_Y:
+               y += 12;
+               x -= 8;
+               break;
+            case PIPE_TEX_FACE_POS_Z:
+            case PIPE_TEX_FACE_NEG_Z:
+               y = total_height - 4;
+               x = (face - 4) * 8;
+               break;
+            }
+            break;
+         case 2:
+            y = total_height - 4;
+            x = bottom_offsets[face];
+            break;
+         case 1:
+            x += 48;
+            break;
+         default:
+            x += step_offsets[face][0] * d;
+            y += step_offsets[face][1] * d;
+            break;
          }
-#endif
       }
    }
 }
@@ -596,18 +643,21 @@ i945_texture_layout_cube(struct i915_texture *tex)
 static boolean
 i945_texture_layout(struct i915_texture * tex)
 {
-   struct pipe_resource *pt = &tex->b.b;
-
-   switch (pt->target) {
+   switch (tex->b.b.target) {
    case PIPE_TEXTURE_1D:
    case PIPE_TEXTURE_2D:
-      i945_texture_layout_2d(tex);
+   case PIPE_TEXTURE_RECT:
+      if (!i9x5_special_layout(tex))
+         i945_texture_layout_2d(tex);
       break;
    case PIPE_TEXTURE_3D:
       i945_texture_layout_3d(tex);
       break;
    case PIPE_TEXTURE_CUBE:
-      i945_texture_layout_cube(tex);
+      if (!util_format_is_s3tc(tex->b.b.format))
+         i9x5_texture_layout_cube(tex);
+      else
+         i945_texture_layout_cube(tex);
       break;
    default:
       assert(0);
@@ -640,80 +690,95 @@ i915_texture_get_handle(struct pipe_screen * screen,
 
 static void
 i915_texture_destroy(struct pipe_screen *screen,
-                    struct pipe_resource *pt)
+                     struct pipe_resource *pt)
 {
    struct i915_texture *tex = i915_texture(pt);
    struct i915_winsys *iws = i915_screen(screen)->iws;
    uint i;
 
-   /*
-     DBG("%s deleting %p\n", __FUNCTION__, (void *) tex);
-   */
-
-   iws->buffer_destroy(iws, tex->buffer);
+   if (tex->buffer)
+      iws->buffer_destroy(iws, tex->buffer);
 
    for (i = 0; i < Elements(tex->image_offset); i++)
-      if (tex->image_offset[i])
-         FREE(tex->image_offset[i]);
+      FREE(tex->image_offset[i]);
 
    FREE(tex);
 }
 
-static struct pipe_transfer * 
-i915_texture_get_transfer(struct pipe_context *context,
-                         struct pipe_resource *resource,
-                         struct pipe_subresource sr,
-                         unsigned usage,
-                         const struct pipe_box *box)
-{
-   struct i915_texture *tex = i915_texture(resource);
-   struct pipe_transfer *transfer = CALLOC_STRUCT(pipe_transfer);
-   if (transfer == NULL)
-      return NULL;
-
-   transfer->resource = resource;
-   transfer->sr = sr;
-   transfer->usage = usage;
-   transfer->box = *box;
-   transfer->stride = tex->stride;
-
-   return transfer;
-}
-
-
 static void *
 i915_texture_transfer_map(struct pipe_context *pipe,
-                         struct pipe_transfer *transfer)
+                          struct pipe_resource *resource,
+                          unsigned level,
+                          unsigned usage,
+                          const struct pipe_box *box,
+                          struct pipe_transfer **ptransfer)
 {
-   struct pipe_resource *resource = transfer->resource;
+   struct i915_context *i915 = i915_context(pipe);
    struct i915_texture *tex = i915_texture(resource);
+   struct i915_transfer *transfer = util_slab_alloc(&i915->texture_transfer_pool);
+   boolean use_staging_texture = FALSE;
    struct i915_winsys *iws = i915_screen(pipe->screen)->iws;
-   struct pipe_subresource sr = transfer->sr;
-   struct pipe_box *box = &transfer->box;
    enum pipe_format format = resource->format;
    unsigned offset;
    char *map;
 
-   if (resource->target == PIPE_TEXTURE_CUBE) {
-      offset = tex->image_offset[sr.level][sr.face];
-   }
-   else if (resource->target == PIPE_TEXTURE_3D) {
-      offset = tex->image_offset[sr.level][box->z];
+   if (transfer == NULL)
+      return NULL;
+
+   transfer->b.resource = resource;
+   transfer->b.level = level;
+   transfer->b.usage = usage;
+   transfer->b.box = *box;
+   transfer->b.stride = tex->stride;
+   transfer->staging_texture = NULL;
+   /* XXX: handle depth textures everyhwere*/
+   transfer->b.layer_stride = 0;
+
+   /* if we use staging transfers, only support textures we can render to,
+    * because we need that for u_blitter */
+   if (i915->blitter &&
+       util_blitter_is_copy_supported(i915->blitter, resource, resource,
+                                     PIPE_MASK_RGBAZS) &&
+       (usage & PIPE_TRANSFER_WRITE) &&
+       !(usage & (PIPE_TRANSFER_READ | PIPE_TRANSFER_DONTBLOCK | PIPE_TRANSFER_UNSYNCHRONIZED)))
+      use_staging_texture = TRUE;
+
+   use_staging_texture = FALSE;
+
+   if (use_staging_texture) {
+      /* 
+       * Allocate the untiled staging texture.
+       * If the alloc fails, transfer->staging_texture is NULL and we fallback to a map() 
+       */
+      transfer->staging_texture = i915_texture_create(pipe->screen, resource, TRUE);
    }
-   else {
-      offset = tex->image_offset[sr.level][0];
-      assert(sr.face == 0);
+
+   if (resource->target != PIPE_TEXTURE_3D &&
+       resource->target != PIPE_TEXTURE_CUBE)
       assert(box->z == 0);
+
+   if (transfer->staging_texture) {
+      tex = i915_texture(transfer->staging_texture);
+   } else {
+      /* TODO this is a sledgehammer */
+      tex = i915_texture(resource);
+      pipe->flush(pipe, NULL, 0);
    }
 
-   map = iws->buffer_map(iws,
-                        tex->buffer,
-                        (transfer->usage & PIPE_TRANSFER_WRITE) ? TRUE : FALSE);
-   if (map == NULL)
+   offset = i915_texture_offset(tex, transfer->b.level, box->z);
+
+   map = iws->buffer_map(iws, tex->buffer,
+                         (transfer->b.usage & PIPE_TRANSFER_WRITE) ? TRUE : FALSE);
+   if (map == NULL) {
+      pipe_resource_reference(&transfer->staging_texture, NULL);
+      FREE(transfer);
       return NULL;
+   }
+
+   *ptransfer = &transfer->b;
 
    return map + offset +
-      box->y / util_format_get_blockheight(format) * transfer->stride +
+      box->y / util_format_get_blockheight(format) * transfer->b.stride +
       box->x / util_format_get_blockwidth(format) * util_format_get_blocksize(format);
 }
 
@@ -721,20 +786,125 @@ static void
 i915_texture_transfer_unmap(struct pipe_context *pipe,
                            struct pipe_transfer *transfer)
 {
-   struct i915_texture *tex = i915_texture(transfer->resource);
+   struct i915_context *i915 = i915_context(pipe);
+   struct i915_transfer *itransfer = (struct i915_transfer*)transfer;
+   struct i915_texture *tex = i915_texture(itransfer->b.resource);
    struct i915_winsys *iws = i915_screen(tex->b.b.screen)->iws;
+
+   if (itransfer->staging_texture)
+      tex = i915_texture(itransfer->staging_texture);
+
    iws->buffer_unmap(iws, tex->buffer);
+
+   if ((itransfer->staging_texture) &&
+       (transfer->usage & PIPE_TRANSFER_WRITE)) {
+      struct pipe_box sbox;
+
+      u_box_origin_2d(itransfer->b.box.width, itransfer->b.box.height, &sbox);
+      pipe->resource_copy_region(pipe, itransfer->b.resource, itransfer->b.level,
+                                   itransfer->b.box.x, itransfer->b.box.y, itransfer->b.box.z,
+                                   itransfer->staging_texture,
+                                   0, &sbox);
+      pipe->flush(pipe, NULL, 0);
+      pipe_resource_reference(&itransfer->staging_texture, NULL);
+   }
+
+   util_slab_free(&i915->texture_transfer_pool, itransfer);
 }
 
+#if 0
+static void i915_transfer_inline_write( struct pipe_context *pipe,
+                                 struct pipe_resource *resource,
+                                 unsigned level,
+                                 unsigned usage,
+                                 const struct pipe_box *box,
+                                 const void *data,
+                                 unsigned stride,
+                                 unsigned layer_stride)
+{
+   struct pipe_transfer *transfer = NULL;
+   struct i915_transfer *itransfer = NULL;
+   const uint8_t *src_data = data;
+   unsigned i;
+
+   transfer = pipe->transfer_get(pipe,
+                                 resource,
+                                 level,
+                                 usage,
+                                 box );
+   if (transfer == NULL)
+      goto out;
+
+   itransfer = (struct i915_transfer*)transfer;
+
+   if (itransfer->staging_texture) {
+      struct i915_texture *tex = i915_texture(itransfer->staging_texture);
+      enum pipe_format format = tex->b.b.format;
+      struct i915_winsys *iws = i915_screen(tex->b.b.screen)->iws;
+      size_t offset;
+      size_t size;
+
+      offset = i915_texture_offset(tex, transfer->level, transfer->box.z);
+
+      for (i = 0; i < box->depth; i++) {
+         if (!tex->b.b.last_level &&
+                     tex->b.b.width0 == transfer->box.width) {
+             unsigned nby = util_format_get_nblocksy(format, transfer->box.y);
+             assert(!offset);
+             assert(!transfer->box.x);
+             assert(tex->stride == transfer->stride);
+
+             offset += tex->stride * nby;
+             size = util_format_get_2d_size(format, transfer->stride,
+                             transfer->box.height);
+             iws->buffer_write(iws, tex->buffer, offset, size, transfer->data);
+
+         } else {
+             unsigned nby = util_format_get_nblocksy(format, transfer->box.y);
+             int i;
+             offset += util_format_get_stride(format, transfer->box.x);
+             size = transfer->stride;
+
+             for (i = 0; i < nby; i++) {
+                     iws->buffer_write(iws, tex->buffer, offset, size, transfer->data);
+                     offset += tex->stride;
+             }
+         }
+         offset += layer_stride;
+      }
+   } else {
+      uint8_t *map = pipe_transfer_map(pipe, &itransfer->b);
+      if (map == NULL)
+         goto nomap;
+
+      for (i = 0; i < box->depth; i++) {
+         util_copy_rect(map,
+                        resource->format,
+                        itransfer->b.stride, /* bytes */
+                        0, 0,
+                        box->width,
+                        box->height,
+                        src_data,
+                        stride,       /* bytes */
+                        0, 0);
+         map += itransfer->b.layer_stride;
+         src_data += layer_stride;
+      }
+nomap:
+      if (map)
+         pipe_transfer_unmap(pipe, &itransfer->b);
+   }
 
+out:
+   if (itransfer)
+      pipe_transfer_destroy(pipe, &itransfer->b);
+}
+#endif
 
-struct u_resource_vtbl i915_texture_vtbl = 
+struct u_resource_vtbl i915_texture_vtbl =
 {
    i915_texture_get_handle,          /* get_handle */
    i915_texture_destroy,             /* resource_destroy */
-   NULL,                             /* is_resource_referenced */
-   i915_texture_get_transfer,        /* get_transfer */
-   u_default_transfer_destroy,       /* transfer_destroy */
    i915_texture_transfer_map,        /* transfer_map */
    u_default_transfer_flush_region,   /* transfer_flush_region */
    i915_texture_transfer_unmap,              /* transfer_unmap */
@@ -742,16 +912,14 @@ struct u_resource_vtbl i915_texture_vtbl =
 };
 
 
-
-
 struct pipe_resource *
 i915_texture_create(struct pipe_screen *screen,
-                    const struct pipe_resource *template)
+                    const struct pipe_resource *template,
+                    boolean force_untiled)
 {
    struct i915_screen *is = i915_screen(screen);
    struct i915_winsys *iws = is->iws;
    struct i915_texture *tex = CALLOC_STRUCT(i915_texture);
-   size_t tex_size;
    unsigned buf_usage = 0;
 
    if (!tex)
@@ -762,6 +930,11 @@ i915_texture_create(struct pipe_screen *screen,
    pipe_reference_init(&tex->b.b.reference, 1);
    tex->b.b.screen = screen;
 
+   if ( (force_untiled) || (template->usage == PIPE_USAGE_STREAM) )
+      tex->tiling = I915_TILE_NONE;
+   else
+      tex->tiling = i915_texture_tiling(is, tex);
+
    if (is->is_i945) {
       if (!i945_texture_layout(tex))
          goto fail;
@@ -770,8 +943,6 @@ i915_texture_create(struct pipe_screen *screen,
          goto fail;
    }
 
-   tex_size = tex->stride * tex->total_nblocksy;
-
    /* for scanouts and cursors, cursors arn't scanouts */
 
    /* XXX: use a custom flag for cursors, don't rely on magically
@@ -782,23 +953,15 @@ i915_texture_create(struct pipe_screen *screen,
    else
       buf_usage = I915_NEW_TEXTURE;
 
-   tex->buffer = iws->buffer_create(iws, tex_size, 64, buf_usage);
+   tex->buffer = iws->buffer_create_tiled(iws, &tex->stride, tex->total_nblocksy,
+                                             &tex->tiling, buf_usage);
    if (!tex->buffer)
       goto fail;
 
-   /* setup any hw fences */
-   if (tex->hw_tiled) {
-      assert(tex->sw_tiled == I915_TILE_NONE);
-      iws->buffer_set_fence_reg(iws, tex->buffer, tex->stride, tex->hw_tiled);
-   }
-
-   
-#if 0
-   void *ptr = ws->buffer_map(ws, tex->buffer,
-      PIPE_BUFFER_USAGE_CPU_WRITE);
-   memset(ptr, 0x80, tex_size);
-   ws->buffer_unmap(ws, tex->buffer);
-#endif
+   I915_DBG(DBG_TEXTURE, "%s: %p stride %u, blocks (%u, %u) tiling %s\n", __func__,
+            tex, tex->stride,
+            tex->stride / util_format_get_blocksize(tex->b.b.format),
+            tex->total_nblocksy, get_tiling_string(tex->tiling));
 
    return &tex->b.b;
 
@@ -817,13 +980,15 @@ i915_texture_from_handle(struct pipe_screen * screen,
    struct i915_winsys *iws = is->iws;
    struct i915_winsys_buffer *buffer;
    unsigned stride;
+   enum i915_winsys_buffer_tile tiling;
 
    assert(screen);
 
-   buffer = iws->buffer_from_handle(iws, whandle, &stride);
+   buffer = iws->buffer_from_handle(iws, whandle, &tiling, &stride);
 
    /* Only supports one type */
-   if (template->target != PIPE_TEXTURE_2D ||
+   if ((template->target != PIPE_TEXTURE_2D &&
+       template->target != PIPE_TEXTURE_RECT) ||
        template->last_level != 0 ||
        template->depth0 != 1) {
       return NULL;
@@ -839,12 +1004,19 @@ i915_texture_from_handle(struct pipe_screen * screen,
    tex->b.b.screen = screen;
 
    tex->stride = stride;
+   tex->tiling = tiling;
+   tex->total_nblocksy = align_nblocksy(tex->b.b.format, tex->b.b.height0, 8);
 
-   i915_texture_set_level_info(tex, 0, 1, template->width0, template->height0, 1);
+   i915_texture_set_level_info(tex, 0, 1);
    i915_texture_set_image_offset(tex, 0, 0, 0, 0);
 
    tex->buffer = buffer;
 
+   I915_DBG(DBG_TEXTURE, "%s: %p stride %u, blocks (%u, %u) tiling %s\n", __func__,
+            tex, tex->stride,
+            tex->stride / util_format_get_blocksize(tex->b.b.format),
+            tex->total_nblocksy, get_tiling_string(tex->tiling));
+
    return &tex->b.b;
 }