GitLab

Without them, the build will fail on a system without the X headers with
a recent Mesa from the master branch. The EGL_CFLAGS_OTHER define the
MESA_EGL_NO_X11_HEADERS macro that guards the #includes of the X
headers.
Signed-off-by: Mircea Gherzan <mircea.gherzan@intel.com>
Signed-off-by: Ben Widawsky <benjamin.widawsky@intel.com>

No GLSL version implies GLSL 1.10, but that isn't supported by
fglrx in core profiles.

This lets the tests run and fail in more interesting ways on fglrx.
Signed-off-by: Dave Airlie <airlied@redhat.com>

BITMAP is a wingdi.h typedef.

Also, remove another bit of leftover devel/debug code.

Reviewed-by: Jose Fonseca <jfoneca@vmware.com>

  skip the test if there is no linker available
  fix CL_LINKER_NOT_AVAILABLE detection by testing device one by one
  print the build log in case of error

Fixes the arb_shader_image_load_store-max-images compute stage
subtest.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=94140Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>

Reported-by: Mark Janes <mark.a.janes@intel.com>
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Brian Paul <brianp@vmware.com>

Demonstrates a bug in Intel color resolve logic. If rendering to
texture is preceded by fast clear subsequent reads from the
texture require the underlying color buffer to be resolved. In
case of mipmap generation Intel driver needs to reallocate the
underlying buffer to fit the subsequent levels. However, copy of
the original level zero to newly allocated buffer is delayed until
the driver is about to generate level one using level zero as
source. This causes a color resolve of the original level zero.
Both the generation of level one and the resolve are meta
operations, and the latter will interfere with the sampler
settings of the former.
Signed-off-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
Reviewed-by: Brian Paul <brianp@vmware.com>

The test was passing a pointer to an uninitialized integer value as the
glShaderSource length parameter.  This is meant to be an array
containing the length of each shader string.

Instead, just pass NULL, which makes the GL assume the strings are
null-terminated (which they are).

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=93335Reviewed-by: Mark Janes <mark.a.janes@intel.com>

CMake actually marks that we require six 1.4.0, however, I can't find
any packages anywhere for 1.4.0, and the lowest version I've seen
requested is 1.5.2.

This fixes requirements for working with six 1.5.2, and sets tox to use
1.5.2 (and a suitable version of mock). Primarily there are a few things
we're using that are not available: six.moves.getcwd, six.viewvalues,
six.python_2_unicode_compatible.
Signed-off-by: Dylan Baker <dylanx.c.baker@intel.com>
Tested-by: Brian Paul <brianp@vmware.com>

From ARB_shader_image_load_store spec:

"    (0) How does this extension differ from the similar
        EXT_shader_image_load_store?

      RESOLVED:  The functionality provided by this extension is very similar
      to that provided by EXT_shader_image_load_stores.  There are some
      functional differences.

        * "size" layout qualifiers replaced with "format" qualifiers.

        * Image loads aren't restricted to "1x8", "1x16", "1x32", "2x32", and
          "4x32" formats.  Instead, each supported image format has a layout
          qualifier, and values loaded from images are converted to an
          vec4/ivec4/uvec4 representation appropriate for the image format."

size4x32 was valid for EXT_shader_image_load_store. That explains
why this test works properly on proprietary NVIDIA drivers, as
both extensions are supported.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=93336Reviewed-by: Tapani Pälli <tapani.palli@intel.com>

program_interface_query: no need to check link status when using piglit_build_simple_program_multiple_shaders

From piglit_build_simple_program_multiple_shaders documentation:
/**
 * Builds and links a program from optional sources, throwing
 * PIGLIT_FAIL on error. The last target must be 0.
 */

So internally is already calling piglit_link_check_status, deleting
the program and reporting PIGLIT_FAIL if fails.

So it is not needed to check the link status. That would be needed
when using piglit_build_simple_program_unlinked_multiple_shaders.
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>

This fixes a run-time error with python 3.4.
Reviewed-by: Dylan Baker <baker.dylan.c@gmail.com>

During the hybridization this little bit was missed, causing the to
subprocess to return bytes, while the rest of the code expected unicode.
Signed-off-by: Dylan Baker <dylanx.c.baker@intel.com>
Tested-by: Tom Stellard <thomas.stellard@amd.com>

V3: rebased, fix variable copy & paste error, update the
relink_program_created_by_glCreateShaderProgram() test to check
for linking error when in core profile and no error in compat
(Timothy).

v2: Add missing extension check for GL_ARB_transform_feedback2.  Use
configure_transform_feedback_object utility function.  Use
CreateShaderProgram_with_xfb utility function.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Timothy Arceri <timothy.arceri@collabora.com>

I filled a spec bug for this as the GL Core spec seems to conflict
with the GLSL spec on this and its been confirmed that they are
not intended to be allowed.

>From Section 7.3 (PROGRAM OBJECTS) of the OpenGL 4.5 spec:

    "Linking can fail for a variety of reasons as specified in the
    OpenGL Shading Language Specification, as well as any of the
    following reasons:

     - No shader objects are attached to program."

Cc: Ian Romanick <idr@freedesktop.org>

There are two tests, the second test covers what the first
test is doing plus more, so remove the first test.

Add a check that glUseProgram doesn't generate an error when
linked program is empty.

Finally move the remaining test out of the shaders folder.

Cc: Ilia Mirkin <imirkin@alum.mit.edu>
Cc: Ian Romanick <idr@freedesktop.org>

Verifies the screen position specified by glViewport is not effected by
the GL_CLIP_ORIGIN state.

Passes with nvidia, fails with Mesa until _mesa_get_viewport_xform()
is fixed.

v2: also enable back-face culling to make sure front/back-face polygon
orientation isn't upset by the clip origin.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=93813Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>

The modern piglit convention is "ext_foo_bar-testname".  This patch
renames the executables to:
  arb_clip_control-clip-control
  arb_clip_control-depth-precision

Arguably, the first test should be split into two separate tests that
exercise state setting/getting and rendering.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>

Acked-by: Ilia Mirkin <imirkin@alum.mit.edu>

This refactors the tox.ini a little bit since some of the requirements
of the accel platform on 2.7 (backports.lzma and subprocess32) are not
required in 3.3+, and mock is not required on 3.3+ for any profile.
Signed-off-by: Dylan Baker <dylanx.c.baker@intel.com>
Acked-by: Jose Fonseca <jfonseca@vmware.com>

This makes the piglit executable run with python 3 instead of python 2,
the legacy scripts (piglit-run, piglit-summary-html, etc) still run with
python 2.

The goal of this series has been to make python 2.7 a fallback rather
than a main option. If one wants to use python 2, it's a trivial change
to the python script to get that.
Signed-off-by: Dylan Baker <dylanx.c.baker@intel.com>
Acked-by: Jose Fonseca <jfonseca@vmware.com>

The generators have supported 3.x for some time, this now makes 3.x the
default, and falls back to 2.7

This has been tested with 3.5 and 2.7
Signed-off-by: Dylan Baker <dylanx.c.baker@intel.com>
Acked-by: Jose Fonseca <jfonseca@vmware.com>

In python 3 the current implementation provides bytes instead of
unicode, but the write mode is set for unicode.
Signed-off-by: Dylan Baker <dylanx.c.baker@intel.com>
Acked-by: Jose Fonseca <jfonseca@vmware.com>

Python 3 doesn't allow mixing of tabs and spaces for indent, only a
choice between one or the other. Piglit uses spaces.
Signed-off-by: Dylan Baker <dylanx.c.baker@intel.com>
Acked-by: Jose Fonseca <jfonseca@vmware.com>

On python 2 re.compile sets no flags, so asking if the flags bitfield ==
re.IGNORECASE works. For python 3 this isn't the case (the default it
returns for me is 100010), so the test needs to correctly check that
re.IGNORECASE is in the bitfield.
Signed-off-by: Dylan Baker <dylanx.c.baker@intel.com>
Acked-by: Jose Fonseca <jfonseca@vmware.com>

For some reason in python 3 (all versions) some of the mock patchers in
this opengl test leak out into other modules. This doesn't happen in
python 2. I'm not sure why this fixes the issue, but it does.
Signed-off-by: Dylan Baker <dylanx.c.baker@intel.com>
Acked-by: Jose Fonseca <jfonseca@vmware.com>

By default the builtins don't appear in the python 3 namespace, and
create=True must be set to make this work.
Signed-off-by: Dylan Baker <dylanx.c.baker@intel.com>
Acked-by: Jose Fonseca <jfonseca@vmware.com>

This makes the GLSLParserTest class only support str (unicode in python
2), and not bytes (str in python 2).
Signed-off-by: Dylan Baker <dylanx.c.baker@intel.com>
Acked-by: Jose Fonseca <jfonseca@vmware.com>

This changes collect_system_info to convert the raw output of each
command to unicode before returning it.
Signed-off-by: Dylan Baker <dylanx.c.baker@intel.com>
Acked-by: Jose Fonseca <jfonseca@vmware.com>

Signed-off-by: Dylan Baker <dylanx.c.baker@intel.com>
Acked-by: Jose Fonseca <jfonseca@vmware.com>

This changes the junit.py script to always convert bytes into unicode
before writing, since we do declare that our xml is encoded in unicode.
Signed-off-by: Dylan Baker <dylanx.c.baker@intel.com>
Acked-by: Jose Fonseca <jfonseca@vmware.com>

In python2 one can seek to negatives (-1 is the end of the file, for
example). In python 2.6+ (and exclusively in python 3) one instead uses
the second argument of seek to say whether to start at the start or end
of the file, and then an exact point. So file.seek(-1) -> file.seek(0,
2)
Signed-off-by: Dylan Baker <dylanx.c.baker@intel.com>
Acked-by: Jose Fonseca <jfonseca@vmware.com>

In python 2 < 2.6, generators define a next() method. In python 2.6+
(including 2.7) they should define a __next__ method and are called with
the next() function.

In python 2.6 and 2.7 both are supported, but in 3.x only the __next__
method is supported
Signed-off-by: Dylan Baker <dylanx.c.baker@intel.com>
Acked-by: Jose Fonseca <jfonseca@vmware.com>

Also fix unittest that relied on compare to None having python 2
behavior.
Signed-off-by: Dylan Baker <dylanx.c.baker@intel.com>
Acked-by: Jose Fonseca <jfonseca@vmware.com>

This actually needs to handle bytes, since it gets assigned from
subprocess directly.
Signed-off-by: Dylan Baker <dylanx.c.baker@intel.com>
Acked-by: Jose Fonseca <jfonseca@vmware.com>

Signed-off-by: Dylan Baker <dylanx.c.baker@intel.com>
Acked-by: Jose Fonseca <jfonseca@vmware.com>

Mostly this involves marking str (unicode) and bytes explicitly and not
mixing the two.
Signed-off-by: Dylan Baker <dylanx.c.baker@intel.com>
Acked-by: Jose Fonseca <jfonseca@vmware.com>