Dan Hoeflinger
Dan Hoeflinger
Our testing for `permutation_iterator` needs additional coverage for a variety of base iterator types. We missed some bugs related to this until they were caught by others, and this is...
Bug Description --- As implemented before this PR, `set_intersection` using `oneapi::dpl::execution::par` has a bug which results in incorrect semantics for some situations. For performance reasons, in some situations, we would...
In a number of algorithms within oneDPL, we rely upon compiling kernels and then querying exact workgroup size of compiled kernels to work around an issue with incorrect returns from...
Align `__get_sycl_range` with SYCL runtime in it's treatment of `write` access mode and `no_init{}`
Currently `__get_sycl_range` with a `write` access mode and a host_iterator input does not include a "copy-in" for data. The copy-in is skipped, as the `sycl::buffer` is not initialized with the...
As described in the comment below, it should be possible to implement vectorized in-place exclusive scan, rather than turning off vectorization for those cases. --- I agree that in-place exclusive...
Pre-including standard library implementation headers for tests to make sure we have knowledge of the STL implementation when creating macros. Also, turning off `PSTL_USE_PARALLEL_POLICIES` when a valid tbb implementation is...
Adds infrastructure to be able to shorten some commonly used test infrastructure by 1) allowing scaling of the `max_n` (100000) and `small_n` (16) 2) allowing scaling of the step (3.14.15)...
Following the lead of https://github.com/oneapi-src/oneDPL/pull/1543, we have a similar issue in another place. This PR adjusts the standard library implementation specific macros withing `test_config.h` to be lazily evaluated later when...
Summary --- This PR enables comparison ``,`=` of our internal tuple `oneapi::dpl::__internal::tuple` with another `oneapi::dpl::__internal::tuple` or `std::tuple` with equal number of elements where corresponding elements from rhs and lhs are...
@MikeDvorskiy and I discussed some issues with device copyable. He is working on a PR to resolve them. The 2 issues we found which need addressing are: 1) Types with...