Skip to content

nccl, aws-ofi-nccl: add new versions and add rocm support to aws-ofi-nccl#3885

Draft
msimberg wants to merge 11 commits intospack:developfrom
msimberg:aws-ofi-nccl-with-rocm
Draft

nccl, aws-ofi-nccl: add new versions and add rocm support to aws-ofi-nccl#3885
msimberg wants to merge 11 commits intospack:developfrom
msimberg:aws-ofi-nccl-with-rocm

Conversation

@msimberg
Copy link
Contributor

Since 1.18.0 aws-ofi-nccl has support for rocm (aws/aws-ofi-nccl#1014). This effectively replaces aws-ofi-rccl. Note that --with-rccl is not needed. --with-nccl could possibly also be removed, but I have not tested that change so leaving it out from here.

version("1.7.1", sha256="d50a160c7aba76445e5c895fba0f3dbfdec51f702d218168a5e5017806cf0fb0")
version("1.6.0", sha256="19a6fc91afe9a317fd3154c897fa219eab48fcdddefa66d881f1843c1165f7ee")

variant("cuda", default=True, description="Enable CUDA support")
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

is there any reason for not making this a CudaPackage/ROCmPackage?

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

No, I'm glad you asked. It just didn't cross my mind... Let me try to make that change.

depends_on("cuda", when="+cuda")
depends_on("nccl fabrics=auto", when="+cuda")
depends_on("hip", when="+rocm")
depends_on("rccl", when="+rocm")
Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I should actually check if this is needed since --with-rccl isn't needed...

@msimberg msimberg marked this pull request as draft March 18, 2026 14:50
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants