[llvm-bugs] [Bug 43841] narrow casted FMA op?

[via llvm-bugs]

https://bugs.llvm.org/show_bug.cgi?id=43841

Sanjay Patel <spatel+llvm at rotateright.com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
         Resolution|---                         |INVALID
             Status|NEW                         |RESOLVED

--- Comment #4 from Sanjay Patel <spatel+llvm at rotateright.com> ---
(In reply to Steve Canon from comment #3)
> Figueroa doesn't cover it, so don't feel bad. The necessary width is so
> staggeringly huge as to be basically useless for every case except half (you
> basically need emax - emin + p bits to prevent double rounding, so more than
> 256 bits for float). Half only needs 39 bits though, so yes, double suffices.

Thanks - I figured there had to be some limit, but I didn't know how to
formulate it. We could leave this report open to handle the extreme cases with
half, but I don't think that's a common case. 

I'd still like to be able to do this transform generally when given enough
compiler flag rope, so I've opened bug 43847 for that (it requires a different
set of fixes within clang/LLVM).

-- 
You are receiving this mail because:
You are on the CC list for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-bugs/attachments/20191030/de56eb8f/attachment.html>