[flang-commits] [flang] [flang] WIP: attempting to add an atomicadd intrinsinc (PR #123490)

Valentin Clement バレンタイン クレメン via flang-commits flang-commits at lists.llvm.org
Mon Jan 20 07:37:39 PST 2025


clementval wrote:

> Actually, intrinsic _module_ procedure must have a Fortran interface defined an intrinsic module file because they are only recognized if those modules are used. The table can also be used for those procedures since [this commit for ieee](https://github.com/llvm/llvm-project/commit/124338dd8016dc7589685d9fc731b178f2b5d379#diff-6bb3b4929e13ba83349cd5d5e97eabaf0d3b88a87b83710c3842c69ca8801a6f)

My bad ... I didn't remember about this. 

> @Renaud-K, in your test in the PR description, I do not see a use cudadevice, I think it is needed so that atomicaddd is resolved in semantics as being declared in the intrinsic module, and not being some user defined procedure.

`cudadevice` is implicitly "used" in cuda fortran device/global procedures.

https://github.com/llvm/llvm-project/pull/123490


More information about the flang-commits mailing list