Thrust 1.15.0 (NVIDIA HPC SDK 21.11)

[alliepiper]

Summary

Thrust 1.15.0 accompanies the NVIDIA HPC SDK 21.11 release. It provides numerous bugfixes, including non-numeric thrust::sequence support, several MSVC-related compilation fixes, fewer conversion warnings, counting_iterator initialization, and documentation updates.

Deprecation Notices

A future version of Thrust will remove support for CUDA Dynamic Parallelism (CDP).

This will only affect calls to Thrust algorithms made from CUDA device-side code that currently launches a kernel; such calls will instead execute sequentially on the calling GPU thread instead of launching a device-wide kernel.

Bug Fixes

• Sequence: Specialise compute_sequence_values for builtin arithmetic types #1507: Allow thrust::sequence to work with non-numeric types. Thanks to Ben Jude (@bjude) for this contribution.
• Fixed call of max() on MSVC compiler #1509: Avoid macro collision when calling max() on MSVC. Thanks to Thomas (@TomInTheShell) for this contribution.
• Fix default constructor of counting iterator #1514: Initialize all members in counting_iterator's default constructor.
• Fix std::allocator traits #1518: Fix std::allocator_traits on MSVC + C++17.
• First checkpoint for Wconversion warning fixes #1530: Fix several -Wconversion warnings. Thanks to Matt Stack (@matt-stack) for this contribution.
• changed "std::device_vector" to "thrust::device_vector" #1539: Fixed typo in thrust::for_each documentation. Thanks to Salman (@untamedimpala) for this contribution.

Other Enhancements

• Use CUB version of merge sort #1511: Replace the internal merge sort implementation with cub::DeviceMergeSort.

---

This discussion was created from the release Thrust 1.15.0 (NVIDIA HPC SDK 21.11).