[libcxx-commits] [PATCH] D141779: [libc++][PSTL] Copy the headers into libc++

Louis Dionne via Phabricator via libcxx-commits libcxx-commits at lists.llvm.org
Thu Mar 30 09:03:55 PDT 2023


ldionne added a comment.

In D141779#4210839 <https://reviews.llvm.org/D141779#4210839>, @Mordante wrote:

> I'm not too familiar with pstl so I don't have a strong opinion. What I read sounds sensible, thanks @philnik &
> @ldionne for the detailed plan. I think it would be great to put some of these design ideas in a developer page for our documentation.

@philnik Can you put this in a DesignDoc?

In D141779#4213395 <https://reviews.llvm.org/D141779#4213395>, @rarutyun wrote:

> Since many different algorithms could map on single pattern it's common layer [...]

When we looked with @philnik, I think we did not find many patterns that were reused, which is why we came to the conclusion that having a systematic layer of patterns for all algorithms was not pulling its weigh. You know the library better, so if we misunderstood something, please let us know. If we didn't miss anything, then I do believe that keeping reusable helpers (patterns) as implementation-detail functions to be reused would be sufficient, without having to have a full layer of customization.

> The only reason why forwarding declarations were introduced is reducing the compilation time when parallel backends (i.e.., both TBB and OpenMP) are not available [...]

That's interesting, I had not thought of that. However, as @philnik said, I think we'll want to go for the serial backend when we don't have anything better available, so that shouldn't be an issue.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D141779/new/

https://reviews.llvm.org/D141779



More information about the libcxx-commits mailing list