[Openmp-commits] [PATCH] D62393: [OPENMP][NVPTX]Mark parallel level counter as volatile.

[Jon Chesterfield via Phabricator via Openmp-commits]

JonChesterfield added a comment.

In D62393#1699629 <https://reviews.llvm.org/D62393#1699629>, @jlebar wrote:

> Based on this and my comment above, I strongly oppose this patch.
>
> ptxas from CUDA 8 has many, many known bugs.  I oppose working around them in LLVM; it will eat up a huge amount of maintainers' time, and you will never cover all or even many of the bugs.


Ideologically I'm right there with you. Killing off cuda 8 support works for me - there's presumably an official deprecation process we should follow. Announcements and so forth.

Pragmatically, you're overlooking a couple of things. We don't need to cover many of the bugs to keep the deviceRTL working - it's 6k lines of relatively straightforward cuda that will exercise a small subset of the nvptx toolchain.

More fundamentally, it may be necessary to keep cuda 8 working despite the broken toolchain. People write code against buggy compilers, and when the compiler is changed, their code behaves differently. The use case is wanting to use a recent llvm, e.g. for x86 related security patches, while also continuing to use a buggy and outdated cuda toolchain so their numerical codes keep getting the same answers.

So if we can drop cuda 8, great. If some of the maintainers have a customer that requires cuda 8 to continue working with a recent llvm - which is sad but plausible - we're better off supporting that use case in tree than forcing an out of tree fork.


Repository:
  rOMP OpenMP

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D62393/new/

https://reviews.llvm.org/D62393