mutter/cogl/cogl-texture.c

/*
 * Cogl
 *
 * An object oriented GL/GLES Abstraction/Utility Layer
 *
 * Copyright (C) 2007,2008,2009 Intel Corporation.
 * Copyright (C) 2010 Red Hat, Inc.
 *
 * This library is free software; you can redistribute it and/or
 * modify it under the terms of the GNU Lesser General Public
 * License as published by the Free Software Foundation; either
 * version 2 of the License, or (at your option) any later version.
 *
 * This library is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 * Lesser General Public License for more details.
 *
 * You should have received a copy of the GNU Lesser General Public
 * License along with this library. If not, see <http://www.gnu.org/licenses/>.
 *
 *
 *
 * Authors:
 *  Matthew Allum  <mallum@openedhand.com>
 *  Neil Roberts   <neil@linux.intel.com>
 *  Robert Bragg   <robert@linux.intel.com>
 */

#ifdef HAVE_CONFIG_H
#include "config.h"
#endif

#include "cogl-internal.h"
#include "cogl-util.h"
#include "cogl-bitmap.h"
#include "cogl-bitmap-private.h"
#include "cogl-buffer-private.h"
#include "cogl-pixel-buffer-private.h"
#include "cogl-private.h"
#include "cogl-texture-private.h"
#include "cogl-texture-driver.h"
#include "cogl-texture-2d-sliced-private.h"
#include "cogl-texture-2d-private.h"
#include "cogl-texture-2d-gl.h"
#include "cogl-texture-rectangle-private.h"
#include "cogl-sub-texture-private.h"
#include "cogl-atlas-texture-private.h"
#include "cogl-pipeline.h"
#include "cogl-context-private.h"
#include "cogl-object-private.h"
#include "cogl-object-private.h"
#include "cogl-primitives.h"
#include "cogl-framebuffer-private.h"
#include "cogl1-context.h"
#include "cogl-sub-texture.h"
#include "cogl-primitive-texture.h"

#include <string.h>
#include <stdlib.h>
#include <math.h>

uint32_t
cogl_texture_error_quark (void)
{
  return g_quark_from_static_string ("cogl-texture-error-quark");
}

/* XXX:
 * The CoglObject macros don't support any form of inheritance, so for
 * now we implement the CoglObject support for the CoglTexture
 * abstract class manually.
 */

void
_cogl_texture_register_texture_type (const CoglObjectClass *klass)
{
  _COGL_GET_CONTEXT (ctxt, NO_RETVAL);

  ctxt->texture_types = g_slist_prepend (ctxt->texture_types, (void *) klass);
}

CoglBool
cogl_is_texture (void *object)
{
  CoglObject *obj = (CoglObject *)object;
  GSList *l;

  _COGL_GET_CONTEXT (ctxt, FALSE);

  if (object == NULL)
    return FALSE;

  for (l = ctxt->texture_types; l; l = l->next)
    if (l->data == obj->klass)
      return TRUE;

  return FALSE;
}

void *
cogl_texture_ref (void *object)
{
  if (!cogl_is_texture (object))
    return NULL;

  _COGL_OBJECT_DEBUG_REF (CoglTexture, object);

  cogl_object_ref (object);

  return object;
}

void
cogl_texture_unref (void *object)
{
  if (!cogl_is_texture (object))
    {
      g_warning (G_STRINGIFY (cogl_texture_unref)
                 ": Ignoring unref of CoglObject "
                 "due to type mismatch");
      return;
    }

  _COGL_OBJECT_DEBUG_UNREF (CoglTexture, object);

  cogl_object_unref (object);
}

void
_cogl_texture_init (CoglTexture *texture,
                    CoglContext *context,
                    const CoglTextureVtable *vtable)
{
  texture->context = context;
  texture->vtable = vtable;
  texture->framebuffers = NULL;
}

void
_cogl_texture_free (CoglTexture *texture)
{
  g_free (texture);
}

static CoglBool
_cogl_texture_needs_premult_conversion (CoglPixelFormat src_format,
                                        CoglPixelFormat dst_format)
{
  return ((src_format & dst_format & COGL_A_BIT) &&
          src_format != COGL_PIXEL_FORMAT_A_8 &&
          dst_format != COGL_PIXEL_FORMAT_A_8 &&
          (src_format & COGL_PREMULT_BIT) !=
          (dst_format & COGL_PREMULT_BIT));
}

CoglPixelFormat
_cogl_texture_determine_internal_format (CoglPixelFormat src_format,
                                         CoglPixelFormat dst_format)
{
  /* If the application hasn't specified a specific format then we'll
   * pick the most appropriate. By default Cogl will use a
   * premultiplied internal format. Later we will add control over
   * this. */
  if (dst_format == COGL_PIXEL_FORMAT_ANY)
    {
      if (COGL_PIXEL_FORMAT_CAN_HAVE_PREMULT (src_format))
        return src_format | COGL_PREMULT_BIT;
      else
        return src_format;
    }
  else
    /* XXX: It might be nice to make this match the component ordering
       of the source format when the formats are otherwise the same
       because on GL there is no way to specify the ordering of the
       internal format. However when using GLES with the
       GL_EXT_texture_format_BGRA8888 the order of the internal format
       becomes important because it must exactly match the format of
       the uploaded data. That means that if someone creates a texture
       with some RGBA data and then later tries to upload BGRA data we
       do actually have to swizzle the components */
    return dst_format;
}

CoglBitmap *
_cogl_texture_prepare_for_upload (CoglBitmap      *src_bmp,
                                  CoglPixelFormat  dst_format,
                                  CoglPixelFormat *dst_format_out,
                                  GLenum          *out_glintformat,
                                  GLenum          *out_glformat,
                                  GLenum          *out_gltype)
{
  CoglPixelFormat src_format = cogl_bitmap_get_format (src_bmp);
  CoglBitmap *dst_bmp;

  _COGL_GET_CONTEXT (ctx, NULL);

  dst_format = _cogl_texture_determine_internal_format (src_format,
                                                        dst_format);

  /* OpenGL supports specifying a different format for the internal
     format when uploading texture data. We should use this to convert
     formats because it is likely to be faster and support more types
     than the Cogl bitmap code. However under GLES the internal format
     must be the same as the bitmap format and it only supports a
     limited number of formats so we must convert using the Cogl
     bitmap code instead */

  /* If the driver doesn't natively support alpha textures then it
   * won't work correctly to convert to/from component-alpha
   * textures */

  if ((ctx->private_feature_flags & COGL_PRIVATE_FEATURE_FORMAT_CONVERSION) &&
      ((ctx->private_feature_flags & COGL_PRIVATE_FEATURE_ALPHA_TEXTURES) ||
       (src_format != COGL_PIXEL_FORMAT_A_8 &&
        dst_format != COGL_PIXEL_FORMAT_A_8) ||
       src_format == dst_format))
    {
      /* If the source format does not have the same premult flag as the
         dst format then we need to copy and convert it */
      if (_cogl_texture_needs_premult_conversion (src_format,
                                                  dst_format))
        {
          dst_bmp = _cogl_bitmap_convert (src_bmp,
                                          src_format ^ COGL_PREMULT_BIT);

          if (dst_bmp == NULL)
            return NULL;
        }
      else
        dst_bmp = cogl_object_ref (src_bmp);

      /* Use the source format from the src bitmap type and the internal
         format from the dst format type so that GL can do the
         conversion */
      ctx->driver_vtable->pixel_format_to_gl (ctx,
                                              src_format,
                                              NULL, /* internal format */
                                              out_glformat,
                                              out_gltype);
      ctx->driver_vtable->pixel_format_to_gl (ctx,
                                              dst_format,
                                              out_glintformat,
                                              NULL,
                                              NULL);

    }
  else
    {
      CoglPixelFormat closest_format;

      closest_format = ctx->driver_vtable->pixel_format_to_gl (ctx,
                                                               dst_format,
                                                               out_glintformat,
                                                               out_glformat,
                                                               out_gltype);

      if (closest_format != src_format)
        dst_bmp = _cogl_bitmap_convert (src_bmp, closest_format);
      else
        dst_bmp = cogl_object_ref (src_bmp);
    }

  if (dst_format_out)
    *dst_format_out = dst_format;

  return dst_bmp;
}

static inline int
calculate_alignment (int rowstride)
{
  int alignment = 1 << (_cogl_util_ffs (rowstride) - 1);

  return MIN (alignment, 8);
}

void
_cogl_texture_prep_gl_alignment_for_pixels_upload (int pixels_rowstride)
{
  _COGL_GET_CONTEXT (ctx, NO_RETVAL);

  GE( ctx, glPixelStorei (GL_UNPACK_ALIGNMENT,
                          calculate_alignment (pixels_rowstride)) );
}

void
_cogl_texture_prep_gl_alignment_for_pixels_download (int bpp,
                                                     int width,
                                                     int rowstride)
{
  int alignment;

  _COGL_GET_CONTEXT (ctx, NO_RETVAL);

  /* If no padding is needed then we can always use an alignment of 1.
   * We want to do this even though it is equivalent to the alignment
   * of the rowstride because the Intel driver in Mesa currently has
   * an optimisation when reading data into a PBO that only works if
   * the alignment is exactly 1.
   *
   * https://bugs.freedesktop.org/show_bug.cgi?id=46632
   */

  if (rowstride == bpp * width)
    alignment = 1;
  else
    alignment = calculate_alignment (rowstride);

  GE( ctx, glPixelStorei (GL_PACK_ALIGNMENT, alignment) );
}

CoglTexture *
cogl_texture_new_with_size (unsigned int     width,
			    unsigned int     height,
                            CoglTextureFlags flags,
			    CoglPixelFormat  internal_format)
{
  CoglTexture *tex;

  _COGL_GET_CONTEXT (ctx, NULL);

  if ((_cogl_util_is_pot (width) && _cogl_util_is_pot (height)) ||
      (cogl_has_feature (ctx, COGL_FEATURE_ID_TEXTURE_NPOT_BASIC) &&
       cogl_has_feature (ctx, COGL_FEATURE_ID_TEXTURE_NPOT_MIPMAP)))
    {
      /* First try creating a fast-path non-sliced texture */
      tex = COGL_TEXTURE (cogl_texture_2d_new_with_size (ctx,
                                                         width, height,
                                                         internal_format,
                                                         NULL));
    }
  else
    tex = NULL;

  if (tex)
    {
      CoglBool auto_mipmap = !(flags & COGL_TEXTURE_NO_AUTO_MIPMAP);
      cogl_primitive_texture_set_auto_mipmap (COGL_PRIMITIVE_TEXTURE (tex),
                                              auto_mipmap);
    }
  else
    {
      /* If it fails resort to sliced textures */
      int max_waste = flags & COGL_TEXTURE_NO_SLICING ? -1 : COGL_TEXTURE_MAX_WASTE;
      tex = COGL_TEXTURE (cogl_texture_2d_sliced_new_with_size (ctx,
                                                                width,
                                                                height,
                                                                max_waste,
                                                                internal_format,
                                                                NULL));
    }

  return tex;
}

CoglTexture *
cogl_texture_new_from_data (unsigned int width,
			    unsigned int height,
                            CoglTextureFlags flags,
			    CoglPixelFormat format,
			    CoglPixelFormat internal_format,
			    unsigned int rowstride,
			    const uint8_t *data)
{
  CoglBitmap *bmp;
  CoglTexture *tex;

  _COGL_GET_CONTEXT (ctx, NULL);

  if (format == COGL_PIXEL_FORMAT_ANY)
    return NULL;

  if (data == NULL)
    return NULL;

  /* Rowstride from width if not given */
  if (rowstride == 0)
    rowstride = width * _cogl_pixel_format_get_bytes_per_pixel (format);

  /* Wrap the data into a bitmap */
  bmp = cogl_bitmap_new_for_data (ctx,
                                  width, height,
                                  format,
                                  rowstride,
                                  (uint8_t *) data);

  tex = cogl_texture_new_from_bitmap (bmp, flags, internal_format);

  cogl_object_unref (bmp);

  return tex;
}

CoglTexture *
cogl_texture_new_from_bitmap (CoglBitmap *bitmap,
                              CoglTextureFlags flags,
                              CoglPixelFormat  internal_format)
{
  CoglAtlasTexture *atlas_tex;
  CoglTexture *tex;

  _COGL_GET_CONTEXT (ctx, FALSE);

  /* First try putting the texture in the atlas */
  if ((atlas_tex = _cogl_atlas_texture_new_from_bitmap (bitmap,
                                                        flags,
                                                        internal_format)))
    return COGL_TEXTURE (atlas_tex);

  /* If that doesn't work try a fast path 2D texture */
  if ((_cogl_util_is_pot (bitmap->width) &&
       _cogl_util_is_pot (bitmap->height)) ||
      (cogl_has_feature (ctx, COGL_FEATURE_ID_TEXTURE_NPOT_BASIC) &&
       cogl_has_feature (ctx, COGL_FEATURE_ID_TEXTURE_NPOT_MIPMAP)))
    {
      tex = COGL_TEXTURE (cogl_texture_2d_new_from_bitmap (bitmap,
                                                           internal_format,
                                                           NULL));
    }
  else
    tex = NULL;

  if (tex)
    {
      CoglBool auto_mipmap = !(flags & COGL_TEXTURE_NO_AUTO_MIPMAP);
      cogl_primitive_texture_set_auto_mipmap (COGL_PRIMITIVE_TEXTURE (tex),
                                              auto_mipmap);
    }
  else
    {
      /* Otherwise create a sliced texture */
      tex = COGL_TEXTURE (_cogl_texture_2d_sliced_new_from_bitmap (bitmap,
                                                             flags,
                                                             internal_format));
    }

  return tex;
}

CoglTexture *
cogl_texture_new_from_file (const char        *filename,
                            CoglTextureFlags   flags,
                            CoglPixelFormat    internal_format,
                            CoglError           **error)
{
  CoglBitmap *bmp;
  CoglTexture *texture = NULL;
  CoglPixelFormat src_format;

  _COGL_GET_CONTEXT (ctx, NULL);

  _COGL_RETURN_VAL_IF_FAIL (error == NULL || *error == NULL, NULL);

  bmp = cogl_bitmap_new_from_file (filename, error);
  if (bmp == NULL)
    return NULL;

  src_format = cogl_bitmap_get_format (bmp);

  /* We know that the bitmap data is solely owned by this function so
     we can do the premult conversion in place. This avoids having to
     copy the bitmap which will otherwise happen in
     _cogl_texture_prepare_for_upload */
  internal_format =
    _cogl_texture_determine_internal_format (src_format, internal_format);
  if (!_cogl_texture_needs_premult_conversion (src_format, internal_format) ||
      _cogl_bitmap_convert_premult_status (bmp, src_format ^ COGL_PREMULT_BIT))
    texture = cogl_texture_new_from_bitmap (bmp, flags, internal_format);

  cogl_object_unref (bmp);

  return texture;
}

CoglTexture *
cogl_texture_new_from_foreign (GLuint           gl_handle,
			       GLenum           gl_target,
			       GLuint           width,
			       GLuint           height,
			       GLuint           x_pot_waste,
			       GLuint           y_pot_waste,
			       CoglPixelFormat  format)
{
#if HAVE_COGL_GL
  if (gl_target == GL_TEXTURE_RECTANGLE_ARB)
    {
      CoglTextureRectangle *texture_rectangle;
      CoglSubTexture *sub_texture;

      _COGL_GET_CONTEXT (ctx, NULL);

      if (x_pot_waste != 0 || y_pot_waste != 0)
        {
          /* It shouldn't be necessary to have waste in this case since
           * the texture isn't limited to power of two sizes. */
          g_warning ("You can't create a foreign GL_TEXTURE_RECTANGLE cogl "
                     "texture with waste\n");
          return NULL;
        }

      texture_rectangle = cogl_texture_rectangle_new_from_foreign (ctx,
                                                                   gl_handle,
                                                                   width,
                                                                   height,
                                                                   format,
                                                                   NULL);

      /* CoglTextureRectangle textures work with non-normalized
       * coordinates, but the semantics for this function that people
       * depend on are that all returned texture works with normalized
       * coordinates so we wrap with a CoglSubTexture... */
      sub_texture = cogl_sub_texture_new (ctx,
                                          COGL_TEXTURE (texture_rectangle),
                                          0, 0, width, height);
      return COGL_TEXTURE (sub_texture);
    }
#endif

  if (x_pot_waste != 0 || y_pot_waste != 0)
    return COGL_TEXTURE (_cogl_texture_2d_sliced_new_from_foreign (gl_handle,
                                                                   gl_target,
                                                                   width,
                                                                   height,
                                                                   x_pot_waste,
                                                                   y_pot_waste,
                                                                   format));
  else
    {
      _COGL_GET_CONTEXT (ctx, NULL);
      return COGL_TEXTURE (cogl_texture_2d_new_from_foreign (ctx,
                                                             gl_handle,
                                                             width,
                                                             height,
                                                             format,
                                                             NULL));
    }
}

CoglBool
_cogl_texture_is_foreign (CoglTexture *texture)
{
  if (texture->vtable->is_foreign)
    return texture->vtable->is_foreign (texture);
  else
    return FALSE;
}

CoglTexture *
cogl_texture_new_from_sub_texture (CoglTexture *full_texture,
                                   int sub_x,
                                   int sub_y,
                                   int sub_width,
                                   int sub_height)
{
  _COGL_GET_CONTEXT (ctx, NULL);
  return COGL_TEXTURE (cogl_sub_texture_new (ctx,
                                             full_texture, sub_x, sub_y,
                                             sub_width, sub_height));
}

unsigned int
cogl_texture_get_width (CoglTexture *texture)
{
  return texture->vtable->get_width (texture);
}

unsigned int
cogl_texture_get_height (CoglTexture *texture)
{
  return texture->vtable->get_height (texture);
}

CoglPixelFormat
cogl_texture_get_format (CoglTexture *texture)
{
  return texture->vtable->get_format (texture);
}

unsigned int
cogl_texture_get_rowstride (CoglTexture *texture)
{
  CoglPixelFormat format = cogl_texture_get_format (texture);
  /* FIXME: This function should go away. It previously just returned
     the rowstride that was used to upload the data as far as I can
     tell. This is not helpful */

  /* Just guess at a suitable rowstride */
  return (_cogl_pixel_format_get_bytes_per_pixel (format)
          * cogl_texture_get_width (texture));
}

int
cogl_texture_get_max_waste (CoglTexture *texture)
{
  return texture->vtable->get_max_waste (texture);
}

CoglBool
cogl_texture_is_sliced (CoglTexture *texture)
{
  return texture->vtable->is_sliced (texture);
}

/* If this returns FALSE, that implies _foreach_sub_texture_in_region
 * will be needed to iterate over multiple sub textures for regions whos
 * texture coordinates extend out of the range [0,1]
 */
CoglBool
_cogl_texture_can_hardware_repeat (CoglTexture *texture)
{
  return texture->vtable->can_hardware_repeat (texture);
}

/* NB: You can't use this with textures comprised of multiple sub textures (use
 * cogl_texture_is_sliced() to check) since coordinate transformation for such
 * textures will be different for each slice. */
void
_cogl_texture_transform_coords_to_gl (CoglTexture *texture,
                                      float *s,
                                      float *t)
{
  texture->vtable->transform_coords_to_gl (texture, s, t);
}

CoglTransformResult
_cogl_texture_transform_quad_coords_to_gl (CoglTexture *texture,
                                           float *coords)
{
  return texture->vtable->transform_quad_coords_to_gl (texture, coords);
}

GLenum
_cogl_texture_get_gl_format (CoglTexture *texture)
{
  return texture->vtable->get_gl_format (texture);
}

CoglBool
cogl_texture_get_gl_texture (CoglTexture *texture,
			     GLuint *out_gl_handle,
			     GLenum *out_gl_target)
{
  return texture->vtable->get_gl_texture (texture,
                                          out_gl_handle, out_gl_target);
}

CoglTextureType
_cogl_texture_get_type (CoglTexture *texture)
{
  return texture->vtable->get_type (texture);
}

void
_cogl_texture_pre_paint (CoglTexture *texture, CoglTexturePrePaintFlags flags)
{
  texture->vtable->pre_paint (texture, flags);
}

void
_cogl_texture_ensure_non_quad_rendering (CoglTexture *texture)
{
  texture->vtable->ensure_non_quad_rendering (texture);
}

CoglBool
cogl_texture_set_region_from_bitmap (CoglTexture *texture,
                                     int src_x,
                                     int src_y,
                                     int dst_x,
                                     int dst_y,
                                     unsigned int dst_width,
                                     unsigned int dst_height,
                                     CoglBitmap *bmp)
{
  CoglBool ret;

  _COGL_RETURN_VAL_IF_FAIL ((cogl_bitmap_get_width (bmp) - src_x)
                            >= dst_width, FALSE);
  _COGL_RETURN_VAL_IF_FAIL ((cogl_bitmap_get_height (bmp) - src_y)
                            >= dst_height, FALSE);

  /* Shortcut out early if the image is empty */
  if (dst_width == 0 || dst_height == 0)
    return TRUE;

  /* Note that we don't prepare the bitmap for upload here because
     some backends may be internally using a different format for the
     actual GL texture than that reported by
     cogl_texture_get_format. For example the atlas textures are
     always stored in an RGBA texture even if the texture format is
     advertised as RGB. */

  ret = texture->vtable->set_region (texture,
                                     src_x, src_y,
                                     dst_x, dst_y,
                                     dst_width, dst_height,
                                     bmp);

  return ret;
}

CoglBool
cogl_texture_set_region (CoglTexture *texture,
			 int src_x,
			 int src_y,
			 int dst_x,
			 int dst_y,
			 unsigned int dst_width,
			 unsigned int dst_height,
			 int width,
			 int height,
			 CoglPixelFormat format,
			 unsigned int rowstride,
			 const uint8_t *data)
{
  CoglContext *ctx = texture->context;
  CoglBitmap *source_bmp;
  CoglBool ret;

  _COGL_RETURN_VAL_IF_FAIL ((width - src_x) >= dst_width, FALSE);
  _COGL_RETURN_VAL_IF_FAIL ((height - src_y) >= dst_height, FALSE);

  /* Check for valid format */
  if (format == COGL_PIXEL_FORMAT_ANY)
    return FALSE;

  /* Rowstride from width if none specified */
  if (rowstride == 0)
    rowstride = _cogl_pixel_format_get_bytes_per_pixel (format) * width;

  /* Init source bitmap */
  source_bmp = cogl_bitmap_new_for_data (ctx,
                                         width, height,
                                         format,
                                         rowstride,
                                         (uint8_t *) data);

  ret = cogl_texture_set_region_from_bitmap (texture,
                                             src_x, src_y,
                                             dst_x, dst_y,
                                             dst_width, dst_height,
                                             source_bmp);

  cogl_object_unref (source_bmp);

  return ret;
}

/* Reads back the contents of a texture by rendering it to the framebuffer
 * and reading back the resulting pixels.
 *
 * It will perform multiple renders if the texture is larger than the
 * current glViewport.
 *
 * It assumes the projection and modelview have already been setup so
 * that rendering to 0,0 with the same width and height of the viewport
 * will exactly cover the viewport.
 *
 * NB: Normally this approach isn't normally used since we can just use
 * glGetTexImage, but may be used as a fallback in some circumstances.
 */
static void
do_texture_draw_and_read (CoglFramebuffer *fb,
                          CoglPipeline *pipeline,
                          CoglTexture *texture,
                          CoglBitmap *target_bmp,
                          float *viewport)
{
  float rx1, ry1;
  float rx2, ry2;
  float tx1, ty1;
  float tx2, ty2;
  int bw,  bh;
  CoglBitmap *rect_bmp;
  unsigned int tex_width, tex_height;
  CoglContext *ctx = fb->context;

  tex_width = cogl_texture_get_width (texture);
  tex_height = cogl_texture_get_height (texture);

  ry2 = 0;
  ty2 = 0;

  /* Walk Y axis until whole bitmap height consumed */
  for (bh = tex_height; bh > 0; bh -= viewport[3])
    {
      /* Rectangle Y coords */
      ry1 = ry2;
      ry2 += (bh < viewport[3]) ? bh : viewport[3];

      /* Normalized texture Y coords */
      ty1 = ty2;
      ty2 = (ry2 / (float) tex_height);

      rx2 = 0;
      tx2 = 0;

      /* Walk X axis until whole bitmap width consumed */
      for (bw = tex_width; bw > 0; bw-=viewport[2])
        {
          int width;
          int height;

          /* Rectangle X coords */
          rx1 = rx2;
          rx2 += (bw < viewport[2]) ? bw : viewport[2];

          width = rx2 - rx1;
          height = ry2 - ry1;

          /* Normalized texture X coords */
          tx1 = tx2;
          tx2 = (rx2 / (float) tex_width);

          /* Draw a portion of texture */
          cogl_framebuffer_draw_textured_rectangle (fb,
                                                    pipeline,
                                                    0, 0,
                                                    rx2 - rx1,
                                                    ry2 - ry1,
                                                    tx1, ty1,
                                                    tx2, ty2);

          /* Read into a temporary bitmap */
          rect_bmp = _cogl_bitmap_new_with_malloc_buffer
                                              (ctx,
                                               width, height,
                                               COGL_PIXEL_FORMAT_RGBA_8888_PRE);

          cogl_framebuffer_read_pixels_into_bitmap
                                   (fb,
                                    viewport[0], viewport[1],
                                    COGL_READ_PIXELS_COLOR_BUFFER,
                                    rect_bmp);

          /* Copy to target bitmap */
          _cogl_bitmap_copy_subregion (rect_bmp,
                                       target_bmp,
                                       0,0,
                                       rx1,ry1,
                                       width,
                                       height);

          /* Free temp bitmap */
          cogl_object_unref (rect_bmp);
        }
    }
}

/* Reads back the contents of a texture by rendering it to the framebuffer
 * and reading back the resulting pixels.
 *
 * NB: Normally this approach isn't normally used since we can just use
 * glGetTexImage, but may be used as a fallback in some circumstances.
 */
CoglBool
_cogl_texture_draw_and_read (CoglTexture *texture,
                             CoglBitmap *target_bmp,
                             GLuint target_gl_format,
                             GLuint target_gl_type)
{
  int bpp;
  CoglFramebuffer *framebuffer = cogl_get_draw_framebuffer ();
  float viewport[4];
  CoglBitmap *alpha_bmp;
  int target_width = cogl_bitmap_get_width (target_bmp);
  int target_height = cogl_bitmap_get_height (target_bmp);
  int target_rowstride = cogl_bitmap_get_rowstride (target_bmp);
  CoglContext *ctx = framebuffer->context;

  bpp = _cogl_pixel_format_get_bytes_per_pixel (COGL_PIXEL_FORMAT_RGBA_8888);

  /* Viewport needs to have some size and be inside the window for this */
  cogl_framebuffer_get_viewport4fv (framebuffer, viewport);
  if (viewport[0] <  0 || viewport[1] <  0 ||
      viewport[2] <= 0 || viewport[3] <= 0)
    return FALSE;

  /* Setup orthographic projection into current viewport (0,0 in top-left
   * corner to draw the texture upside-down so we match the way cogl_read_pixels
   * works)
   */

  _cogl_framebuffer_push_projection (framebuffer);
  cogl_framebuffer_orthographic (framebuffer,
                                 0, 0,
                                 viewport[2], viewport[3],
                                 0, 100);

  cogl_framebuffer_push_matrix (framebuffer);
  cogl_framebuffer_identity_matrix (framebuffer);

  /* Direct copy operation */

  if (ctx->texture_download_pipeline == NULL)
    {
      ctx->texture_download_pipeline = cogl_pipeline_new (ctx);
      cogl_pipeline_set_blend (ctx->texture_download_pipeline,
                               "RGBA = ADD (SRC_COLOR, 0)",
                               NULL);
    }

  cogl_pipeline_set_layer_texture (ctx->texture_download_pipeline, 0, texture);

  cogl_pipeline_set_layer_combine (ctx->texture_download_pipeline,
                                   0, /* layer */
                                   "RGBA = REPLACE (TEXTURE)",
                                   NULL);

  cogl_pipeline_set_layer_filters (ctx->texture_download_pipeline, 0,
                                   COGL_PIPELINE_FILTER_NEAREST,
                                   COGL_PIPELINE_FILTER_NEAREST);

  do_texture_draw_and_read (framebuffer,
                            ctx->texture_download_pipeline,
                            texture, target_bmp, viewport);

  /* Check whether texture has alpha and framebuffer not */
  /* FIXME: For some reason even if ALPHA_BITS is 8, the framebuffer
     still doesn't seem to have an alpha buffer. This might be just
     a PowerVR issue.
  GLint r_bits, g_bits, b_bits, a_bits;
  GE( ctx, glGetIntegerv (GL_ALPHA_BITS, &a_bits) );
  GE( ctx, glGetIntegerv (GL_RED_BITS, &r_bits) );
  GE( ctx, glGetIntegerv (GL_GREEN_BITS, &g_bits) );
  GE( ctx, glGetIntegerv (GL_BLUE_BITS, &b_bits) );
  printf ("R bits: %d\n", r_bits);
  printf ("G bits: %d\n", g_bits);
  printf ("B bits: %d\n", b_bits);
  printf ("A bits: %d\n", a_bits); */
  if ((cogl_texture_get_format (texture) & COGL_A_BIT)/* && a_bits == 0*/)
    {
      uint8_t *srcdata;
      uint8_t *dstdata;
      uint8_t *srcpixel;
      uint8_t *dstpixel;
      int x,y;
      int alpha_rowstride = bpp * target_width;

      if ((dstdata = _cogl_bitmap_map (target_bmp,
                                       COGL_BUFFER_ACCESS_WRITE,
                                       COGL_BUFFER_MAP_HINT_DISCARD)) == NULL)
        return FALSE;

      /* Create temp bitmap for alpha values */
      alpha_bmp =
        _cogl_bitmap_new_with_malloc_buffer (ctx,
                                             target_width,
                                             target_height,
                                             COGL_PIXEL_FORMAT_RGBA_8888);

      /* Draw alpha values into RGB channels */
      cogl_pipeline_set_layer_combine (ctx->texture_download_pipeline,
                                       0, /* layer */
                                       "RGBA = REPLACE (TEXTURE[A])",
                                       NULL);

      do_texture_draw_and_read (framebuffer,
                                ctx->texture_download_pipeline,
                                texture, alpha_bmp, viewport);

      /* Copy temp R to target A */

      srcdata = _cogl_bitmap_map (alpha_bmp,
                                  COGL_BUFFER_ACCESS_READ,
                                  0 /* hints */);

      for (y=0; y<target_height; ++y)
        {
          for (x=0; x<target_width; ++x)
            {
              srcpixel = srcdata + x*bpp;
              dstpixel = dstdata + x*bpp;
              dstpixel[3] = srcpixel[0];
            }
          srcdata += alpha_rowstride;
          dstdata += target_rowstride;
        }

      _cogl_bitmap_unmap (alpha_bmp);

      _cogl_bitmap_unmap (target_bmp);

      cogl_object_unref (alpha_bmp);
    }

  /* Restore old state */
  cogl_framebuffer_pop_matrix (framebuffer);
  _cogl_framebuffer_pop_projection (framebuffer);

  return TRUE;
}

static CoglBool
get_texture_bits_via_offscreen (CoglTexture    *texture,
                                int             x,
                                int             y,
                                int             width,
                                int             height,
                                uint8_t         *dst_bits,
                                unsigned int    dst_rowstride,
                                CoglPixelFormat dst_format)
{
  CoglContext *ctx = texture->context;
  CoglOffscreen *offscreen;
  CoglFramebuffer *framebuffer;
  CoglBitmap *bitmap;
  CoglBool ret;

  if (!cogl_has_feature (ctx, COGL_FEATURE_ID_OFFSCREEN))
    return FALSE;

  offscreen = _cogl_offscreen_new_to_texture_full
                                      (texture,
                                       COGL_OFFSCREEN_DISABLE_DEPTH_AND_STENCIL,
                                       0);

  if (offscreen == NULL)
    return FALSE;

  framebuffer = COGL_FRAMEBUFFER (offscreen);

  bitmap = cogl_bitmap_new_for_data (ctx,
                                     width, height,
                                     dst_format,
                                     dst_rowstride,
                                     dst_bits);
  ret = cogl_framebuffer_read_pixels_into_bitmap (framebuffer,
                                                  x, y,
                                                  COGL_READ_PIXELS_COLOR_BUFFER,
                                                  bitmap);
  cogl_object_unref (bitmap);

  cogl_object_unref (framebuffer);

  return ret;
}

static CoglBool
get_texture_bits_via_copy (CoglTexture *texture,
                           int x,
                           int y,
                           int width,
                           int height,
                           uint8_t *dst_bits,
                           unsigned int dst_rowstride,
                           CoglPixelFormat dst_format)
{
  unsigned int full_rowstride;
  uint8_t *full_bits;
  CoglBool ret = TRUE;
  int bpp;
  int full_tex_width, full_tex_height;

  full_tex_width = cogl_texture_get_width (texture);
  full_tex_height = cogl_texture_get_height (texture);

  bpp = _cogl_pixel_format_get_bytes_per_pixel (dst_format);

  full_rowstride = bpp * full_tex_width;
  full_bits = g_malloc (full_rowstride * full_tex_height);

  if (texture->vtable->get_data (texture,
                                 dst_format,
                                 full_rowstride,
                                 full_bits))
    {
      uint8_t *dst = dst_bits;
      uint8_t *src = full_bits + x * bpp + y * full_rowstride;
      int i;

      for (i = 0; i < height; i++)
        {
          memcpy (dst, src, bpp * width);
          dst += dst_rowstride;
          src += full_rowstride;
        }
    }
  else
    ret = FALSE;

  g_free (full_bits);

  return ret;
}

typedef struct
{
  int orig_width;
  int orig_height;
  CoglBitmap *target_bmp;
  uint8_t *target_bits;
  CoglBool success;
} CoglTextureGetData;

static void
texture_get_cb (CoglTexture *texture,
                const float *subtexture_coords,
                const float *virtual_coords,
                void        *user_data)
{
  CoglTextureGetData *tg_data = user_data;
  CoglPixelFormat format = cogl_bitmap_get_format (tg_data->target_bmp);
  int bpp = _cogl_pixel_format_get_bytes_per_pixel (format);
  unsigned int rowstride = cogl_bitmap_get_rowstride (tg_data->target_bmp);
  int subtexture_width = cogl_texture_get_width (texture);
  int subtexture_height = cogl_texture_get_height (texture);

  int x_in_subtexture = (int) (0.5 + subtexture_width * subtexture_coords[0]);
  int y_in_subtexture = (int) (0.5 + subtexture_height * subtexture_coords[1]);
  int width = ((int) (0.5 + subtexture_width * subtexture_coords[2])
               - x_in_subtexture);
  int height = ((int) (0.5 + subtexture_height * subtexture_coords[3])
                - y_in_subtexture);
  int x_in_bitmap = (int) (0.5 + tg_data->orig_width * virtual_coords[0]);
  int y_in_bitmap = (int) (0.5 + tg_data->orig_height * virtual_coords[1]);

  uint8_t *dst_bits;

  if (!tg_data->success)
    return;

  dst_bits = tg_data->target_bits + x_in_bitmap * bpp + y_in_bitmap * rowstride;

  /* If we can read everything as a single slice, then go ahead and do that
   * to avoid allocating an FBO. We'll leave it up to the GL implementation to
   * do glGetTexImage as efficiently as possible. (GLES doesn't have that,
   * so we'll fall through) */
  if (x_in_subtexture == 0 && y_in_subtexture == 0 &&
      width == subtexture_width && height == subtexture_height)
    {
      if (texture->vtable->get_data (texture,
                                     format,
                                     rowstride,
                                     dst_bits))
        return;
    }

  /* Next best option is a FBO and glReadPixels */
  if (get_texture_bits_via_offscreen (texture,
                                      x_in_subtexture, y_in_subtexture,
                                      width, height,
                                      dst_bits,
                                      rowstride,
                                      format))
    return;

  /* Getting ugly: read the entire texture, copy out the part we want */
  if (get_texture_bits_via_copy (texture,
                                 x_in_subtexture, y_in_subtexture,
                                 width, height,
                                 dst_bits,
                                 rowstride,
                                 format))
    return;

  /* No luck, the caller will fall back to the draw-to-backbuffer and
   * read implementation */
  tg_data->success = FALSE;
}

int
cogl_texture_get_data (CoglTexture *texture,
		       CoglPixelFormat format,
		       unsigned int rowstride,
		       uint8_t *data)
{
  CoglContext *ctx = texture->context;
  int bpp;
  int byte_size;
  CoglPixelFormat closest_format;
  GLenum closest_gl_format;
  GLenum closest_gl_type;
  CoglBitmap *target_bmp;
  int tex_width;
  int tex_height;
  CoglPixelFormat texture_format;

  CoglTextureGetData tg_data;

  texture_format = cogl_texture_get_format (texture);

  /* Default to internal format if none specified */
  if (format == COGL_PIXEL_FORMAT_ANY)
    format = texture_format;

  tex_width = cogl_texture_get_width (texture);
  tex_height = cogl_texture_get_height (texture);

  /* Rowstride from texture width if none specified */
  bpp = _cogl_pixel_format_get_bytes_per_pixel (format);
  if (rowstride == 0)
    rowstride = tex_width * bpp;

  /* Return byte size if only that requested */
  byte_size = tex_height * rowstride;
  if (data == NULL)
    return byte_size;

  closest_format =
    ctx->texture_driver->find_best_gl_get_data_format (ctx,
                                                       format,
                                                       &closest_gl_format,
                                                       &closest_gl_type);

  /* We can assume that whatever data GL gives us will have the
     premult status of the original texture */
  if (COGL_PIXEL_FORMAT_CAN_HAVE_PREMULT (closest_format))
    closest_format = ((closest_format & ~COGL_PREMULT_BIT) |
                      (texture_format & COGL_PREMULT_BIT));

  /* If the application is requesting a conversion from a
   * component-alpha texture and the driver doesn't support them
   * natively then we can only read into an alpha-format buffer. In
   * this case the driver will be faking the alpha textures with a
   * red-component texture and it won't swizzle to the correct format
   * while reading */
  if ((ctx->private_feature_flags & COGL_PRIVATE_FEATURE_ALPHA_TEXTURES) == 0)
    {
      if (texture_format == COGL_PIXEL_FORMAT_A_8)
        {
          closest_format = COGL_PIXEL_FORMAT_A_8;
          closest_gl_format = GL_RED;
          closest_gl_type = GL_UNSIGNED_BYTE;
        }
      else if (format == COGL_PIXEL_FORMAT_A_8)
        {
          /* If we are converting to a component-alpha texture then we
           * need to read all of the components to a temporary buffer
           * because there is no way to get just the 4th component.
           * Note: it doesn't matter whether the texture is
           * pre-multiplied here because we're only going to look at
           * the alpha component */
          closest_format = COGL_PIXEL_FORMAT_RGBA_8888;
          closest_gl_format = GL_RGBA;
          closest_gl_type = GL_UNSIGNED_BYTE;
        }
    }

  /* Is the requested format supported? */
  if (closest_format == format)
    /* Target user data directly */
    target_bmp = cogl_bitmap_new_for_data (ctx,
                                           tex_width,
                                           tex_height,
                                           format,
                                           rowstride,
                                           data);
  else
    target_bmp = _cogl_bitmap_new_with_malloc_buffer (ctx,
                                                      tex_width, tex_height,
                                                      closest_format);

  tg_data.orig_width = tex_width;
  tg_data.orig_height = tex_height;
  tg_data.target_bmp = target_bmp;
  tg_data.target_bits = _cogl_bitmap_map (target_bmp, COGL_BUFFER_ACCESS_WRITE,
                                          COGL_BUFFER_MAP_HINT_DISCARD);
  if (tg_data.target_bits == NULL)
    {
      cogl_object_unref (target_bmp);
      return 0;
    }
  tg_data.success = TRUE;

  /* If there are any dependent framebuffers on the texture then we
     need to flush their journals so the texture contents will be
     up-to-date */
  _cogl_texture_flush_journal_rendering (texture);

  /* Iterating through the subtextures allows piecing together
   * the data for a sliced texture, and allows us to do the
   * read-from-framebuffer logic here in a simple fashion rather than
   * passing offsets down through the code. */
  cogl_meta_texture_foreach_in_region (COGL_META_TEXTURE (texture),
                                       0, 0, 1, 1,
                                       COGL_PIPELINE_WRAP_MODE_REPEAT,
                                       COGL_PIPELINE_WRAP_MODE_REPEAT,
                                       texture_get_cb,
                                       &tg_data);

  _cogl_bitmap_unmap (target_bmp);

  /* XXX: In some cases _cogl_texture_2d_download_from_gl may fail
   * to read back the texture data; such as for GLES which doesn't
   * support glGetTexImage, so here we fallback to drawing the
   * texture and reading the pixels from the framebuffer. */
  if (!tg_data.success)
    _cogl_texture_draw_and_read (texture, target_bmp,
                                 closest_gl_format,
                                 closest_gl_type);

  /* Was intermediate used? */
  if (closest_format != format)
    {
      CoglBitmap *new_bmp;
      CoglBool result;

      /* Convert to requested format directly into the user's buffer */
      new_bmp = cogl_bitmap_new_for_data (ctx,
                                          tex_width, tex_height,
                                          format,
                                          rowstride,
                                          data);
      result = _cogl_bitmap_convert_into_bitmap (target_bmp, new_bmp);

      if (!result)
        /* Return failure after cleaning up */
        byte_size = 0;

      cogl_object_unref (new_bmp);
    }

  cogl_object_unref (target_bmp);

  return byte_size;
}

static void
_cogl_texture_framebuffer_destroy_cb (void *user_data,
                                      void *instance)
{
  CoglTexture *tex = user_data;
  CoglFramebuffer *framebuffer = instance;

  tex->framebuffers = g_list_remove (tex->framebuffers, framebuffer);
}

void
_cogl_texture_associate_framebuffer (CoglTexture *texture,
                                     CoglFramebuffer *framebuffer)
{
  static CoglUserDataKey framebuffer_destroy_notify_key;

  /* Note: we don't take a reference on the framebuffer here because
   * that would introduce a circular reference. */
  texture->framebuffers = g_list_prepend (texture->framebuffers, framebuffer);

  /* Since we haven't taken a reference on the framebuffer we setup
    * some private data so we will be notified if it is destroyed... */
  _cogl_object_set_user_data (COGL_OBJECT (framebuffer),
                              &framebuffer_destroy_notify_key,
                              texture,
                              _cogl_texture_framebuffer_destroy_cb);
}

const GList *
_cogl_texture_get_associated_framebuffers (CoglTexture *texture)
{
  return texture->framebuffers;
}

void
_cogl_texture_flush_journal_rendering (CoglTexture *texture)
{
  GList *l;

  /* It could be that a referenced texture is part of a framebuffer
   * which has an associated journal that must be flushed before it
   * can be sampled from by the current primitive... */
  for (l = texture->framebuffers; l; l = l->next)
    _cogl_framebuffer_flush_journal (l->data);
}

/* This function lets you define a meta texture as a grid of textures
 * whereby the x and y grid-lines are defined by an array of
 * CoglSpans. With that grid based description this function can then
 * iterate all the cells of the grid that lye within a region
 * specified as virtual, meta-texture, coordinates.  This function can
 * also cope with regions that extend beyond the original meta-texture
 * grid by iterating cells repeatedly according to the wrap_x/y
 * arguments.
 *
 * To differentiate between texture coordinates of a specific, real,
 * slice texture and the texture coordinates of a composite, meta
 * texture, the coordinates of the meta texture are called "virtual"
 * coordinates and the coordinates of spans are called "slice"
 * coordinates.
 *
 * Note: no guarantee is given about the order in which the slices
 * will be visited.
 *
 * Note: The slice coordinates passed to @callback are always
 * normalized coordinates even if the span coordinates aren't
 * normalized.
 */
void
_cogl_texture_spans_foreach_in_region (CoglSpan *x_spans,
                                       int n_x_spans,
                                       CoglSpan *y_spans,
                                       int n_y_spans,
                                       CoglTexture **textures,
                                       float *virtual_coords,
                                       float x_normalize_factor,
                                       float y_normalize_factor,
                                       CoglPipelineWrapMode wrap_x,
                                       CoglPipelineWrapMode wrap_y,
                                       CoglMetaTextureCallback callback,
                                       void *user_data)
{
  CoglSpanIter iter_x;
  CoglSpanIter iter_y;
  float slice_coords[4];

  /* Iterate the y axis of the virtual rectangle */
  for (_cogl_span_iter_begin (&iter_y,
                              y_spans,
                              n_y_spans,
                              y_normalize_factor,
                              virtual_coords[1],
                              virtual_coords[3],
                              wrap_y);
       !_cogl_span_iter_end (&iter_y);
       _cogl_span_iter_next (&iter_y))
    {
      if (iter_y.flipped)
        {
          slice_coords[1] = iter_y.intersect_end;
          slice_coords[3] = iter_y.intersect_start;
        }
      else
        {
          slice_coords[1] = iter_y.intersect_start;
          slice_coords[3] = iter_y.intersect_end;
        }

      /* Map the current intersection to normalized slice coordinates */
      slice_coords[1] = (slice_coords[1] - iter_y.pos) / iter_y.span->size;
      slice_coords[3] = (slice_coords[3] - iter_y.pos) / iter_y.span->size;

      /* Iterate the x axis of the virtual rectangle */
      for (_cogl_span_iter_begin (&iter_x,
                                  x_spans,
                                  n_x_spans,
                                  x_normalize_factor,
                                  virtual_coords[0],
                                  virtual_coords[2],
                                  wrap_x);
	   !_cogl_span_iter_end (&iter_x);
	   _cogl_span_iter_next (&iter_x))
        {
          CoglTexture *span_tex;
          float span_virtual_coords[4];

          if (iter_x.flipped)
            {
              slice_coords[0] = iter_x.intersect_end;
              slice_coords[2] = iter_x.intersect_start;
            }
          else
            {
              slice_coords[0] = iter_x.intersect_start;
              slice_coords[2] = iter_x.intersect_end;
            }

          /* Map the current intersection to normalized slice coordinates */
          slice_coords[0] = (slice_coords[0] - iter_x.pos) / iter_x.span->size;
          slice_coords[2] = (slice_coords[2] - iter_x.pos) / iter_x.span->size;

	  /* Pluck out the cogl texture for this span */
          span_tex = textures[iter_y.index * n_x_spans + iter_x.index];

          span_virtual_coords[0] = iter_x.intersect_start;
          span_virtual_coords[1] = iter_y.intersect_start;
          span_virtual_coords[2] = iter_x.intersect_end;
          span_virtual_coords[3] = iter_y.intersect_end;

          callback (COGL_TEXTURE (span_tex),
                    slice_coords,
                    span_virtual_coords,
                    user_data);
	}
    }
}

void
_cogl_texture_gl_flush_legacy_texobj_wrap_modes (CoglTexture *texture,
                                                 GLenum wrap_mode_s,
                                                 GLenum wrap_mode_t,
                                                 GLenum wrap_mode_p)
{
  texture->vtable->gl_flush_legacy_texobj_wrap_modes (texture,
                                                      wrap_mode_s,
                                                      wrap_mode_t,
                                                      wrap_mode_p);
}

void
_cogl_texture_gl_flush_legacy_texobj_filters (CoglTexture *texture,
                                              GLenum min_filter,
                                              GLenum mag_filter)
{
  texture->vtable->gl_flush_legacy_texobj_filters (texture,
                                                   min_filter, mag_filter);
}