[Piglit] [PATCH v3] glsl-4.50: add a test for helper invocations

[Ilia Mirkin]

On Thu, Nov 12, 2015 at 5:28 PM, Ilia Mirkin <imirkin at alum.mit.edu> wrote:
> On Thu, Nov 12, 2015 at 5:22 PM, Glenn Kennard <glenn.kennard at gmail.com> wrote:
>> Tested on r600g implementation
>>
>> Signed-off-by: Glenn Kennard <glenn.kennard at gmail.com>
>> ---
>> Ported to desktop GL 4.5 from Ilia's GLES test case, needed since r600g uses
>> coarse derivatives by default, and shader control over that requires
>> GL_ARB_derivative_control.
>>
>>  .../execution/helper-invocation.shader_test        | 30 ++++++++++++++++++++++
>>  1 file changed, 30 insertions(+)
>>  create mode 100644 tests/spec/glsl-4.50/execution/helper-invocation.shader_test
>>
>> diff --git a/tests/spec/glsl-4.50/execution/helper-invocation.shader_test b/tests/spec/glsl-4.50/execution/helper-invocation.shader_test
>> new file mode 100644
>> index 0000000..d72d490
>> --- /dev/null
>> +++ b/tests/spec/glsl-4.50/execution/helper-invocation.shader_test
>> @@ -0,0 +1,30 @@
>> +[require]
>> +GL >= 4.5
>> +GLSL >= 4.50
>> +GL_ARB_derivative_control
>> +
>> +[vertex shader passthrough]
>> +
>> +[fragment shader]
>> +#version 450
>> +#extension GL_ARB_derivative_control : require
>
> This shouldn't be necessary -- mesa should be fixed rather than adding
> this here. [Also remove from requirements above.]
>
>> +precision highp float;
>
> This was just for ES, not needed here.
>
> With the above 2 fixed,
>
> Reviewed-by: Ilia Mirkin <imirkin at alum.mit.edu>
>
>> +out vec4 color;
>> +void main() {
>> +  float helper = float(gl_HelperInvocation);
>> +  color = vec4(abs(dFdxFine(helper)), abs(dFdyFine(helper)), 0, 1);

Actually since we're doing abs and derivatives... maybe stick helper
into blue? Otherwise an inverted implementation would also pass (i.e.
having frag helper = 0 for the lit pixels)

>> +}
>> +
>> +[test]
>> +clear color 0.2 0.2 0.2 0.2
>> +clear
>> +
>> +# A single pixel being drawn, guaranteed that both the X and Y
>> +# derivatives will be 1.
>> +draw rect ortho 0 0 1 1
>> +probe rect rgba (0, 0, 1, 1) (1, 1, 0, 1)
>> +
>> +# A larger rect being drawn, check that the first 2x2 quad, all of
>> +# which is definitely going to be covered, actually ended up as all 0's
>> +draw rect ortho 2 2 10 10
>> +probe rect rgba (2, 2, 2, 2) (0, 0, 0, 1)
>> --
>> 1.9.1
>>
>> _______________________________________________
>> Piglit mailing list
>> Piglit at lists.freedesktop.org
>> http://lists.freedesktop.org/mailman/listinfo/piglit