[Mesa-dev] [PATCH] mesa: fix ARB_copy_image internal format check

Ilia Mirkin imirkin at alum.mit.edu
Sat Mar 7 10:18:48 PST 2015


[+jekstrand, who wrote the initial ARB_copy_image support and knows a
thing or two about texture formats]

On Sat, Mar 7, 2015 at 4:42 AM, Sean Burke <leftmostcat at gmail.com> wrote:
> The memory layout of compatible internal formats may differ in bytes per
> block, so TexFormat is not a reliable measure of compatibility. Additionally,
> the current check allows compressed textures of the same block size to be used,
> which is in violation of the spec.
>
> Additionally, add a fixme for another check against TexFormat.
> ---
>  src/mesa/main/copyimage.c | 127 ++++++++++++++++++++++++++++++++++++++--------
>  1 file changed, 106 insertions(+), 21 deletions(-)
>
> diff --git a/src/mesa/main/copyimage.c b/src/mesa/main/copyimage.c
> index 455929d..dd0b8ea 100644
> --- a/src/mesa/main/copyimage.c
> +++ b/src/mesa/main/copyimage.c
> @@ -33,6 +33,51 @@
>  #include "texobj.h"
>  #include "fbobject.h"
>  #include "textureview.h"
> +#include "glformats.h"
> +
> +enum mesa_block_class {
> +   BLOCK_CLASS_128_BITS,
> +   BLOCK_CLASS_64_BITS,
> +   BLOCK_CLASS_UNDEFINED
> +};
> +
> +struct internal_format_block_class_info {
> +   enum mesa_block_class block_class;
> +   GLenum internal_format;
> +};
> +
> +static const struct internal_format_block_class_info compressed_formats[] = {
> +   {BLOCK_CLASS_128_BITS, GL_COMPRESSED_RGBA_S3TC_DXT3_EXT},
> +   {BLOCK_CLASS_128_BITS, GL_COMPRESSED_SRGB_ALPHA_S3TC_DXT3_EXT},
> +   {BLOCK_CLASS_128_BITS, GL_COMPRESSED_RGBA_S3TC_DXT5_EXT},
> +   {BLOCK_CLASS_128_BITS, GL_COMPRESSED_SRGB_ALPHA_S3TC_DXT5_EXT},
> +   {BLOCK_CLASS_128_BITS, GL_COMPRESSED_RG_RGTC2},
> +   {BLOCK_CLASS_128_BITS, GL_COMPRESSED_SIGNED_RG_RGTC2},
> +   {BLOCK_CLASS_128_BITS, GL_COMPRESSED_RGBA_BPTC_UNORM},
> +   {BLOCK_CLASS_128_BITS, GL_COMPRESSED_SRGB_ALPHA_BPTC_UNORM},
> +   {BLOCK_CLASS_128_BITS, GL_COMPRESSED_RGB_BPTC_SIGNED_FLOAT},
> +   {BLOCK_CLASS_128_BITS, GL_COMPRESSED_RGB_BPTC_UNSIGNED_FLOAT},
> +   {BLOCK_CLASS_64_BITS, GL_COMPRESSED_RGB_S3TC_DXT1_EXT},
> +   {BLOCK_CLASS_64_BITS, GL_COMPRESSED_SRGB_S3TC_DXT1_EXT},
> +   {BLOCK_CLASS_64_BITS, GL_COMPRESSED_RGBA_S3TC_DXT1_EXT},
> +   {BLOCK_CLASS_64_BITS, GL_COMPRESSED_SRGB_ALPHA_S3TC_DXT1_EXT},
> +   {BLOCK_CLASS_64_BITS, GL_COMPRESSED_RED_RGTC1},
> +   {BLOCK_CLASS_64_BITS, GL_COMPRESSED_SIGNED_RED_RGTC1}
> +};
> +
> +static const struct internal_format_block_class_info uncompressed_formats[] = {
> +   {BLOCK_CLASS_128_BITS, GL_RGBA32UI},
> +   {BLOCK_CLASS_128_BITS, GL_RGBA32I},
> +   {BLOCK_CLASS_128_BITS, GL_RGBA32F},
> +   {BLOCK_CLASS_64_BITS, GL_RGBA16F},
> +   {BLOCK_CLASS_64_BITS, GL_RG32F},
> +   {BLOCK_CLASS_64_BITS, GL_RGBA16UI},
> +   {BLOCK_CLASS_64_BITS, GL_RG32UI},
> +   {BLOCK_CLASS_64_BITS, GL_RGBA16I},
> +   {BLOCK_CLASS_64_BITS, GL_RG32I},
> +   {BLOCK_CLASS_64_BITS, GL_RGBA16},
> +   {BLOCK_CLASS_64_BITS, GL_RGBA16_SNORM}
> +};
>
>  static bool
>  prepare_target(struct gl_context *ctx, GLuint name, GLenum *target, int level,
> @@ -253,6 +298,59 @@ check_region_bounds(struct gl_context *ctx,
> struct gl_texture_image *tex_image,
>     return true;
>  }
>
> +static bool
> +compressed_format_compatible(struct gl_context *ctx,
> +                             GLenum compressedFormat, GLenum otherFormat)
> +{
> +   GLuint i;
> +   enum mesa_block_class compressedClass = BLOCK_CLASS_UNDEFINED,
> +                         otherClass = BLOCK_CLASS_UNDEFINED;
> +
> +   if (_mesa_is_compressed_format(ctx, otherFormat)) {
> +      return false;
> +   }
> +
> +   for (i = 0; i < ARRAY_SIZE(compressed_formats); i++) {
> +      if (compressed_formats[i].internal_format == compressedFormat) {
> +         compressedClass = compressed_formats[i].block_class;
> +         break;
> +      }
> +   }

Each one of these is just a handful of options, rather than going
through the list each time, it'd be a lot more performant to have each
one just be a

switch (glformat) {
case ...
case ...
  return BLOCK_CLASS_64;
case ..
case ..
  return 128
default:
  return undefined;
}

> +
> +   for (i = 0; i < ARRAY_SIZE(uncompressed_formats); i++) {
> +      if (uncompressed_formats[i].internal_format == otherFormat) {
> +         otherClass = uncompressed_formats[i].block_class;
> +         break;
> +      }
> +   }
> +
> +   if (compressedClass == BLOCK_CLASS_UNDEFINED
> +       || otherClass == BLOCK_CLASS_UNDEFINED)
> +      return false;
> +
> +   return compressedClass == otherClass;
> +}
> +
> +static bool
> +copy_format_compatible(struct gl_context *ctx,
> +                                GLenum srcFormat, GLenum dstFormat)
> +{
> +   if (srcFormat == dstFormat) {
> +      return true;
> +   }
> +   else if (_mesa_is_compressed_format(ctx, srcFormat)) {
> +      return compressed_format_compatible(ctx, srcFormat, dstFormat);
> +   }
> +   else if (_mesa_is_compressed_format(ctx, dstFormat)) {
> +      return compressed_format_compatible(ctx, dstFormat, srcFormat);
> +   }
> +   else {
> +      return _mesa_texture_view_compatible_format(ctx,
> +                                                  srcFormat,
> +                                                  dstFormat);

>From my read of the spec, this should be

else if (ctx->Extensions.ARB_texture_view)

Or alternatively the check should be moved into
_mesa_texture_view_compatible_format. Actually that may be preferable.

The spec says:

Dependencies on OpenGL 4.3 and ARB_texture_view:

    If OpenGL 4.3 or later and ARB_texture_view are not supported,
    any references to view-compatible image formats should be removed.



> +   }
> +}
> +
>  void GLAPIENTRY
>  _mesa_CopyImageSubData(GLuint srcName, GLenum srcTarget, GLint srcLevel,
>                         GLint srcX, GLint srcY, GLint srcZ,
> @@ -265,7 +363,7 @@ _mesa_CopyImageSubData(GLuint srcName, GLenum
> srcTarget, GLint srcLevel,
>     struct gl_texture_object *srcTexObj, *dstTexObj;
>     struct gl_texture_image *srcTexImage, *dstTexImage;
>     GLuint src_bw, src_bh, dst_bw, dst_bh;
> -   int i, srcNewZ, dstNewZ, Bpt;
> +   int i, srcNewZ, dstNewZ;
>
>     if (MESA_VERBOSE & VERBOSE_API)
>        _mesa_debug(ctx, "glCopyImageSubData(%u, %s, %d, %d, %d, %d, "
> @@ -291,6 +389,8 @@ _mesa_CopyImageSubData(GLuint srcName, GLenum
> srcTarget, GLint srcLevel,
>                         &dstTexObj, &dstTexImage, &tmpTexNames[1], "dst"))
>        goto cleanup;
>
> +   /* XXX: We can't rely on TexFormat to give us information about the texture
> +    * format specified by the user. */

In theory, yes... but in practice this check only matters for
compressed formats, and those map fairly between gl and mesa formats.
(i.e. the driver would never pick a *different* compressed format for
a particular GL one, unlike with color textures where that's done all
the time).

>     _mesa_get_format_block_size(srcTexImage->TexFormat, &src_bw, &src_bh);
>     if ((srcX % src_bw != 0) || (srcY % src_bh != 0) ||
>         (srcWidth % src_bw != 0) || (srcHeight % src_bh != 0)) {
> @@ -306,15 +406,6 @@ _mesa_CopyImageSubData(GLuint srcName, GLenum
> srcTarget, GLint srcLevel,
>        goto cleanup;
>     }
>
> -   /* Very simple sanity check.  This is sufficient if one of the textures
> -    * is compressed. */
> -   Bpt = _mesa_get_format_bytes(srcTexImage->TexFormat);
> -   if (_mesa_get_format_bytes(dstTexImage->TexFormat) != Bpt) {
> -      _mesa_error(ctx, GL_INVALID_VALUE,
> -                  "glCopyImageSubData(internalFormat mismatch)");
> -      goto cleanup;
> -   }
> -
>     if (!check_region_bounds(ctx, srcTexImage, srcX, srcY, srcZ,
>                              srcWidth, srcHeight, srcDepth, "src"))
>        goto cleanup;
> @@ -324,17 +415,11 @@ _mesa_CopyImageSubData(GLuint srcName, GLenum
> srcTarget, GLint srcLevel,
>                              (srcHeight / src_bh) * dst_bh, srcDepth, "dst"))
>        goto cleanup;
>
> -   if (_mesa_is_format_compressed(srcTexImage->TexFormat)) {
> -      /* XXX: Technically, we should probaby do some more specific checking
> -       * here.  However, this should be sufficient for all compressed
> -       * formats that mesa supports since it is a direct memory copy.
> -       */
> -   } else if (_mesa_is_format_compressed(dstTexImage->TexFormat)) {
> -   } else if (_mesa_texture_view_compatible_format(ctx,
> -                                                   srcTexImage->InternalFormat,
> -
> dstTexImage->InternalFormat)) {
> -   } else {
> -      return; /* Error logged by _mesa_texture_view_compatible_format */
> +   if (!copy_format_compatible(ctx, srcTexImage->InternalFormat,
> +                               dstTexImage->InternalFormat)) {
> +      _mesa_error(ctx, GL_INVALID_VALUE,
> +                  "glCopyImageSubData(internalFormat mismatch)");


    INVALID_OPERATION is generated
     * if either object is a texture and the texture is not complete,
     * if the source and destination formats are not compatible,
     * if the source and destination number of samples do not match,
     * if one image is compressed and the other is uncompressed and the
       block size of compressed image is not equal to the texel size
       of the compressed image.

I think this should be GL_INVALID_OPERATION, not GL_INVALID_VALUE.

> +      goto cleanup;
>     }
>
>     for (i = 0; i < srcDepth; ++i) {
> --
> 2.1.0
> _______________________________________________
> mesa-dev mailing list
> mesa-dev at lists.freedesktop.org
> http://lists.freedesktop.org/mailman/listinfo/mesa-dev


More information about the mesa-dev mailing list