Porting to pika

[albestro]

This is the PR in preparation for the next HPX release.

Changelog:

• Update minimum C++ standard requirement to C++17 (changed in Initial sender/receiver adaptations for tile algorithms #371 and Set C++17 flags in DLA-F #427)
• update HPX minimum version both in CMake and spack package

TODO:

• Adapt CI to use the actual release branch
• Remove tag_invoke/dispatch compatibility macros
  Rename tag_dispatch back to tag_invoke STEllAR-GROUP/hpx#5600
• Use execution::then instead of transform, and execution::start_detached instead of detach (renamed)
  Rename execution::transform to execution::then and execution::detach to execution::start_detached STEllAR-GROUP/hpx#5644
• Use execution::transfer instead of on (renamed)
  Rename on execution::schedule_from, rename just_on to just_transfer, and add transfer STEllAR-GROUP/hpx#5656
• CUDA and C++17 Porting to pika #422 (comment)

Word to @msimberg for any major change (in case there are).

[msimberg]

Word to @msimberg for any major change (in case there are).

C++17 is the major one. The rest are additions to the sender/receiver implementation, but that doesn't affect master at all at the moment.

[albestro]

About C++17 migration of DLAF codebase, there is https://en.cppreference.com/w/cpp/error/uncaught_exception that slightly changes its API.

[teonnik]

Note that the move to C++17 will require that we use rather recent versions of CUDA and nvcc (>11.0.167): https://gist.github.com/ax3l/9489132

[msimberg]

bors try

[msimberg]

A few notes about the move to pika:

• Most of it is renaming hpx to pika
• hpx/local headers and hpx::local namespaces have dropped local since there's no corresponding distributed
• I've required an exact version (0.1.0) in CMake since until 1.0.0 we will not guarantee API stability (we'll follow https://semver.org/, which allows breaking changes on pre-1.0.0 versions)
• The spack package currently points to my fork of spack. I'll replace hpx-local with pika in the upstream spack repository once I see this is working.
• I've changed some hpx/include headers to pika/modules headers. The hpx/include headers came from distributed HPX and were meant to be replaced eventually. pika/modules are also not official public API headers, but work for now. I've opened issues to replace them with official public headers: https://github.com/pika-org/pika/issues?q=is%3Aissue+is%3Aopen+API+header+.

[msimberg]

bors try

[bors]

try

Already running a review

[msimberg]

bors try-

[msimberg]

bors try

[msimberg]

bors try-

[msimberg]

bors try

[bors]

try

Build failed:

• ci/gitlab/full-pipeline

[msimberg]

bors try

[bors]

try

Build succeeded:

• ci/gitlab/full-pipeline

[msimberg]

bors try

[bors]

try

Build succeeded:

• ci/gitlab/full-pipeline

[msimberg]

bors try

[bors]

try

Build succeeded:

• ci/gitlab/full-pipeline

[msimberg]

bors try

[bors]

try

Build succeeded:

• ci/gitlab/full-pipeline

[msimberg]

try

Build succeeded:

* [ci/gitlab/full-pipeline](https://gitlab.com/cscs-ci/eth-cscs/DLA-Future/-/pipelines/458562157)

Note to self: this build was meaningless since it used the cached images. I wanted to try a new commit on pika main so I'll push a commit with an updated spack hash.

[msimberg]

bors try

[bors]

try

Build succeeded:

• ci/gitlab/full-pipeline

[msimberg]

Pika still needs a release, and the release added to the spack package, but this is ready for reviews otherwise.

[msimberg]

bors try

[bors]

try

Build succeeded:

• ci/gitlab/full-pipeline