[RFC] Addition of range-based parallel algorithms

[akukanov]

Overview

For several years oneDPL provides an experimental implementation of parallel algorithms that work with ranges, similar to the C++20 standard Ranges library but with limited functionality. We think it's time to move forward and make these a solid feature of the library.

Details

The major differences comparing to the experimental implementation will be:

• Requiring C++20 (on top of the general library requirements)
• Support for all execution policies, not only the device_policy
• Support for a subset of standard views in namespace std::ranges, instead of our own custom implementation
• Better alignment with C++20 range algorithms, including constraints and support for projections.

The API design will generally follow the "C++ parallel range algorithms" proposal P3179, with necessary adjustments such as namespace modifications.

The specification and the implementation will gradually expand over time. The existing experimental implementation will remain intact until all its functionality is covered.

[akukanov]

Pull requests that add the first set of parallel range algorithms:
#1457
#1471

Update: the implementation has been merged.

[akukanov]

The specification change PR to add the first set of parallel range algorithms; uxlfoundation/oneAPI-spec#569

Update: the specification patch has been merged.

[akukanov]

The RFC to add more parallel range algorithms: #2037
The corresponding oneDPL specification update: uxlfoundation/oneAPI-spec#614

Update: both have been merged.

[akukanov]

By now, the oneDPL specification v1.5 draft has analogues for all the new C++26 parallel algorithms that take ranges, except for generate.
Meanwhile. oneDPL 2022.10 implements another 16 of these algorithms (RFC #2269).