[libc-commits] [libc] [llvm] [libc][math] Implement asinhbf16 (PR #235448)

[via libc-commits]

================

----------------
Tejaswi2517 wrote:

Thanks for the feedback. Just to clarify, do you mean an implementation that doesn't depend on asinhf, similar to the standalone asinbf16 and acosbf16 implementations?

Currently, I reuse asinhf and round the result once to bfloat16. I've tested all 65,536 input bit patterns in all four rounding modes against a 300-bit reference, and also checked for double-rounding mismatches through float. I've shared the details and verification scripts in my earlier comment.

Would you prefer a standalone implementation, or would the current approach be acceptable? I'm happy to work on a standalone implementation if that's preferred. Thanks again!

https://github.com/llvm/llvm-project/pull/235448