[Openmp-commits] [PATCH] D108534: [OpenMP][Docs] add clang to LLVM_ENABLE_PROJECTS in build instructions

[Jon Chesterfield via Phabricator via Openmp-commits]

JonChesterfield added a comment.

The challenge is mostly that clang and deviceRTL (and host runtime to some extent) need to agree on the internal api. E.g. whether a function returns i32 or i64, as something I changed a few days ago for nvptx. That internal API is not stable, documented or versioned. That lets code move across the boundary and generally evolve.

However, that only runs reliably if clang uses the deviceRTL from the same monorepo hash. Some other clang will have different assumptions about the library. It's not sufficient to build the deviceRTL with the clang the user will run on host code. It really has to be the deviceRTL that came from the same checkout as the clang the user will use.

This is difficult to specify reliably without requiring an exact match (e.g. burn the git sha into all the pieces and refuse to run if some differ) or bootstrapping. Enable_runtimes makes the bootstrap easy to get right.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D108534/new/

https://reviews.llvm.org/D108534