Sampler Lit Tests

[cperkinsintel]

Between the coordinate_normalization_mode , addressing_mode and filtering_mode there are 16 valid combinations that a sampler should support for RGBA images. Here we are adding each combination as its own test.
Some of the devices do not support particular combinations. For example, the CPU device is not correctly supporting linear interpolation at this time. In those cases, I have simply skipped the //RUN: directive for the not-yet-supporting device and left a comment in its place. Let me know if this is not the correct approach.

Signed-off-by: Chris Perkins chris.perkins@intel.com

[vladimirlaz]

The failures are dependency issue. Will restart tests once it is fixed.

[vladimirlaz]

The dependency issues have been resolved. Please fix RUN line to have tests run properly.

[cperkinsintel]

@vladimirlaz - How does one find out how/where these tests are supposedly failing? When I click the Details link it just results in a Network error.

I pull this branch, merged it against the latest of intel/llvm-test-suite, and tested it locally. I have the latest pull of intel/llvm sycl branch built. For me the ESIMD tests fail, but they are the only ones, and they seem to always fail.

[vladimirlaz]

If the test caused timeout on some configuration we can only mark them UNSUPPORTED.
SYCL :: Plugin/interop-level-zero-keep-ownership.cpp failyure can be ignored.

[pvchupin]

@cperkinsintel, Sampler/unnormalized-none-linear.cpp is failing for NFC change: intel/llvm#3319

[pvchupin]

@bader, does it make sense to add such sort of tests to CTS instead?

[bader]

@bader, does it make sense to add such sort of tests to CTS instead?

It does if CTS doesn't cover this already.

[cperkinsintel]

@bader @pvchupin - The CTS tests don't cover the samplers at all. All they do is make sure the interface is fully defined and available. There is no actual testing of functionality.
I was hoping to re-do these tests as CTS tests at a later date and submit them to the Khronos repo. But my guess is that the process might be longer? We wanted these tests available immediately, because we have tickets that teams are actively working, and having tests here and now is valuable.

[pvchupin]

@cperkinsintel, it sounds reasonable for this case. Thanks. For the future (and less urgent) cases we may want to consider to contribute directly to CTS. Let's keep that in mind.