MikeDvorskiy
MikeDvorskiy
[oneDPL] permutation_iterator and permutation_view improvements 1. Added support usage case like in Boost (https://www.boost.org/doc/libs/1_38_0/libs/iterator/doc/permutation_iterator.html) 2. Aded support C++ functor as an index map for permutation_iterator (https://oneapi-src.github.io/oneDPL/parallel_api/iterators.html). 3. Added support C++...
A fix for exclusive in-place scan. Fixed the second problem place (local serial scan within a workgroup) in oneDPL code. Added the caching of an input value for the next...
[oneDPL] + a fix a potential error (passing a temporary sycl::buffer to a sycl::accessor by left reference)
(2) - @ValentinaKats Could you please add details about alias "namespace dpl = oneapi::dpl" to the OneAPI spec or(and) oneAPI documentation? _Originally posted by @SergeyKopienko in https://github.com/oneapi-src/oneDPL/pull/558#discussion_r832914117_
1. Support permutation iterator based on a host iterator 2. The tests for permutation iterator usage.
Should we forward `__value`? Question is relevant to all similar places in that file _Originally posted by @rarutyun in https://github.com/oneapi-src/oneDPL/pull/458#discussion_r811077743_
It's still dangerous to forward policy keeping in mind that we use it later. But as we discussed there lots of such places. We need to either do global refactoring...
Actually, we changed just macros (with substring _OMP_ ) which enable OpenMP backend. (guards and key name for make in command line was kept as 'omp', till to libcxx integration...
1. class buffer 2. stuff from parallel scan pattern (upsweep, down sweep etc) 3. to try to write abstraction impl layer which contains the all sorting/ merging optimizations. _Originally posted...
I am a bit concerned how we name guards. Let's create a task to revise that. Currently INTERNAL means nothing at all but PSTL might be there and it brings...