This is still very much a work in progress

This one takes a pointer and a size.  This enables compiling shaders
that are embedded in files with other data.

This regressed on i965 at some point, so I wanted to test it.

Ah my friend texture rectangles.  The spec for
GL_ARB_texture_rectangle is a little bit broken because it was
developed before the #extension mechanism existed.  Because of this,
shaders do not need to have the '#extension GL_ARB_texture_rectangle:
enable' directive.  Issue #15 in the spec says:

 15) How does this extension interact with GLSL based on the "OpenGL
     Shading Language Extension Conventions"?

     Unfortunately, this extension was specified and implemented
     contemporaneously with the GLSL Extension Conventions and because
     of this timing does not follow its guidance for #extension and
     adornment of new GLSL names.  Because this extension has both an
     API interaction (adding a new rectangle texture target) and a GLSL
     interaction (functions and sampler types for accessing texture
     rectangles), you can't practically use the GLSL texture rectangle
     functionality without the API functionality.  For this reason,
     detecting the GL_ARB_texture_rectangle string is sufficient for
     assuming the GLSL functionality is present.

     Conceptually, you can consider the declaration
     #extension GL_ARB_texture_rectangle : require, to allow support
     for texture rectangles, to be implicitly prepended to every
     GLSL shader when ARB_texture_rectangle is advertised.

     All future GLSL extensions should follow the "OpenGL Shading Language
     Extension Conventions" however.

This test may be obsoleted later if we stop relayouting due to
baselevel, but for now this shows off a failure in the relayout code
when doing a fallback blit.

This lets us pass additional args into piglit tests without having
to worry about their order.

Maciej Cencora authored 15 years ago and Brian Paul committed 15 years ago

It doesn't segfault now on r300.

Use a larger window so to avoid symetry in the Y positioning of the
test quads.  This helps to expose Y axis inversion bugs.

But these new tests are temporarily disabled until some bugs in
Mesa are fixed (so the test acts exactly as it did before).

Before we were using a 60x70 texture and not checking for
GL_ARB_texture_non_power_of_two.

While it was designed to catch failure on i965, it also catches
failure with Mesa software.

Shows off failure in the 965 driver in unaliasing registers for GLSL.

Probably should fix this to either use pot or rects.

Signed-off-by: Dave Airlie <airlied@redhat.com>