Let's kill 80bit real at CTFE
tsbockman via Digitalmars-d
digitalmars-d at puremagic.com
Sat Sep 10 20:30:01 PDT 2016
On Friday, 9 September 2016 at 11:50:08 UTC, Stefan Koch wrote:
> Hi,
>
> In short 80bit real are a real pain to support cross-platform.
> emulating them in software is prohibitively slow,
Supposedly, well-optimized 128-bit software floating-point is
actually a bit faster than hardware 80-bit. (Intel keeps 80-bit
around primarily for backwards compatibility, and hasn't put much
effort or die space into improving its hardware speed in many
years.)
> and more importantly hard to get right.
> 64bit floating-point numbers are supported on more
> architectures and are much better supported.
> They are also trivial to use at ctfe.
> I vote for killing the 80bit handling at constant folding.
>
> Destroy!
This has been debated to death in the past. Reducing the maximum
precision is not an acceptable option, but 80-bit could
potentially be replaced by a software 128-bit implementation if
that helps.
More information about the Digitalmars-d
mailing list