1. 25 Sep, 2011 6 commits
  2. 22 Sep, 2011 2 commits
  3. 20 Sep, 2011 5 commits
    • Paul Berry's avatar
      Update clip distance tests to use 8 clip distances. · 732f8af2
      Paul Berry authored
      The clipping tests previously in Piglit only tested the first 6
      clipping planes, since prior to GLSL 1.30, a conforming implementation
      was allowed to only support 6 clipping planes.  However, in GLSL 1.30,
      the minimum value of gl_MaxClipDistances and gl_MaxClipPlanes was
      increased to 8.
      This patch modifies vs-clip-distance-all-planes-enabled.shader_test to
      exercise all 8 clipping planes.
      The previous implementation of
      vs-clip-distance-all-planes-enabled.shader_test used to also verify,
      as a side effect, that gl_ClipDistance could be implicitly sized.  The
      new version does not (it requires gl_ClipDistance to be explicitly
      sized, since it indexes into it using a loop variable).  So I've added
      a new test, vs-clip-distance-implicitly-sized.shader_test, to verify
      Tested using the nVidia proprietary Linux driver.
      Reviewed-by: default avatarIan Romanick <ian.d.romanick@intel.com>
    • Paul Berry's avatar
      Test "double negative" corner case in copy propagation. · 08d16f9e
      Paul Berry authored
      This patch adds two tests that exercise a subtle corner case bug found
      in Mesa with i965: If a value is negated, and then negated again, then
      copy propagation would fail to un-do the negation.
      The bug only exists on i965 vertex shaders, but I went ahead and
      added the equivalent fragment shader test for completeness.
      Reviewed-by: default avatarIan Romanick <ian.d.romanick@intel.com>
    • Paul Berry's avatar
      Add tests of whole-array assignment and out/inout params. · 4642d5a6
      Paul Berry authored
      In GLSL 1.10, whole arrays are not lvalues (even though their elements
      are).  This means it is illegal to use an array on the left hand side
      of an assigment, or to use it as an out or inout parameter of a
      In GLSL 1.20 and later versions, an array is a proper lvalue, so all
      of these uses are allowed.
      In GLSL 1.00 ES, it is illegal to use an array on the left hand side
      of an assignment, but using it as an out or inout parameter of a
      function is ok.
      This patch adds tests to verify correct behavior in GLSL 1.10, GLSL
      1.20, and GLSL ES 1.00.  In addition it verifies that in GLSL 1.20, a
      built-in array may be used as an lvalue (since this currently
      exercises a different code path in Mesa).
      Reviewed-by: default avatarIan Romanick <ian.d.romanick@intel.com>
    • Paul Berry's avatar
      glslparsertest: use dummy shaders when linking for GLSL 1.00 ES. · 881d0274
      Paul Berry authored
      GLES requires both vertex and fragment shaders to be present in order
      to link.  From section 2.10.3 (Program Objects) of the GLES 2.0 spec:
          "Linking will also fail ... if program does not contain both a
          vertex shader and a fragment shader ..."
      This patch adds logic to glslparsertest so that when attempting to
      verify that a vertex program links, it attaches a dummy fragment
      program, and vice versa.  This logic is only used when the requested
      GLSL version is 1.00 (indicating that GLES should be used).
      Reviewed-by: default avatarIan Romanick <ian.d.romanick@intel.com>
    • Jose Fonseca's avatar
      vertex-program-two-side: Use sprintf instead of asprintf. · c6fece4b
      Jose Fonseca authored
      Fixes Windows build, as Windows doesn't support asprintf.  We could
      implement our own asprintf, but doesn't seem necessary here, as static
      buffers fit equally well.
  4. 19 Sep, 2011 7 commits
  5. 16 Sep, 2011 2 commits
  6. 15 Sep, 2011 1 commit
  7. 12 Sep, 2011 1 commit
  8. 11 Sep, 2011 1 commit
  9. 09 Sep, 2011 8 commits
  10. 08 Sep, 2011 7 commits