Add test plan for oneapi_non_uniform_groups extension #866
Conversation
This commit adds a test plan for the [oneapi_non_uniform_groups](https://github.com/intel/llvm/blob/sycl/sycl/doc/extensions/experimental/sycl_ext_oneapi_non_uniform_groups.asciidoc) oneAPI extension. Signed-off-by: Larsen, Steffen <steffen.larsen@intel.com>
|
@Pennycook could I delegate my review role to you for this test plan? I think you'd have better insight since you were the original author of this extension spec. (I tried adding you in the "Reviewers" section, but for some reason GitHub doesn't find you there.) |
I think I can't be officially added as a reviewer because I'm not associated with the SYCL-CTS repository. Regardless, I'll take a look! |
Signed-off-by: Larsen, Steffen <steffen.larsen@intel.com>
Signed-off-by: Larsen, Steffen <steffen.larsen@intel.com>
Signed-off-by: Larsen, Steffen <steffen.larsen@intel.com>
This commit adds tests for the [sycl_ext_oneapi_non_uniform_groups](https://github.com/intel/llvm/blob/sycl/sycl/doc/extensions/experimental/sycl_ext_oneapi_non_uniform_groups.asciidoc) extension in accordance with the test plan currently in review here: KhronosGroup#866
|
Draft PR with the tests: #875 |
Signed-off-by: Larsen, Steffen <steffen.larsen@intel.com>
Co-authored-by: John Pennycook <john.pennycook@intel.com>
| ==== get_group_id | ||
|
|
||
| Check that `get_group_id()` return type is `id_type` and return value is | ||
| `1` if the predicate was `true` or `0` if the predicate was `false`. |
There was a problem hiding this comment.
It confused me, but this is what the specification says.
There was a problem hiding this comment.
Ironically, the implementation and extension specification slightly disagree on this. I have a patch to make them agree: intel/llvm#12905
|
|
||
| === The `fixed_size_group` class API | ||
|
|
||
| Let `N` be some power-of-two value greater than 1 that is expected to be a |
There was a problem hiding this comment.
Are you excluding the case of 1. That seems interesting to test corner cases, otherwise.
There was a problem hiding this comment.
For the class API tests we only test for one group configuration. The intention here was to pick one that wasn't 1 to avoid some trivial cases and to test a more normal group size. We could potentially run the tests for multiple configurations, if that would be more interesting? We do it for the group algorithms.
There was a problem hiding this comment.
Oh I thought you were running with several N and in that case testing also the boundaries would have made sense.
|
@keryell, could you please approve the PR if all your concerns are addressed? |
This commit adds a test plan for the oneapi_non_uniform_groups oneAPI extension.