[PATCH] D150291: [Clang] Rename internal type identifier(s) for __bf16 to BF16Ty

M. Zeeshan Siddiqui via Phabricator via cfe-commits cfe-commits at lists.llvm.org
Fri May 12 11:42:09 PDT 2023


codemzs added a comment.

In D150291#4338118 <https://reviews.llvm.org/D150291#4338118>, @tahonermann wrote:

>> I do wonder if we need two bfloat implementations, but for that I'll leave a comment on D149573 <https://reviews.llvm.org/D149573>.
>
> Given the discussions occurring in D149573 <https://reviews.llvm.org/D149573>, let's hold off on landing this for now. It is sounding like we might have direction for repurposing `__bf16` for `std::bfloat16_t` (and a future `_BFloat16` C type; with `__bf16` retained as an alternate spelling). If we go in that direction, then we would presumably want a change that goes in the opposite direction of this patch; a change that migrates "bf16" names towards "bfloat16". Let's focus on confirming that direction first. I'll mark this as requesting changes for now while we figure this out.

Thank you for your guidance, @tahonermann. I agree it's prudent to establish a confirmed direction for reusing `__bf16` to implement `std::bfloat16_t` as an arithmetic type before proceeding. If we decide to take this route, I understand that we'll be looking towards a change that aligns "bf16" names more closely with "bfloat16".

In the meantime, I will revert to my initial change in D149573 <https://reviews.llvm.org/D149573> which repurposed `__bf16` type, and present it for further discussion. This, along with a summary of the RFC discussion, should help us reach a consensus. I welcome any further insights or considerations you might think pertinent to this process.


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D150291/new/

https://reviews.llvm.org/D150291



More information about the cfe-commits mailing list