Re: [PATCH, x86_64]: Use soft-fp for TFmode (128bit) FP.

["Uros Bizjak" <ubizjak at gmail dot com>]

On 4/13/07, François-Xavier Coudert <fxcoudert@gmail.com> wrote:

(I didn't notice I have pressed reply-to-sender, not reply-to-all in
my previous response).

> I experimented a bit with your patch and the Fortran front-end and
> library support and I have a few questions. There's very naive, so the
> answers might be short :)
>
> * For some of the tests I did, I get undefined references to
> __trunctfxf2 and __extendxftf2. I assume that these are the
> "conversions to/from XFmode" that you have been discussing with jsm,
> is that right? Otherwise, if it's something unexpected, I'll send you
> my testcase.

Attached to this message, please find an untested patch that
implements missing functions. You have to change t-fprules-softfp64 to
account for new functions. I plan to test tem in the evening (CET),
but in case they don't compile, the changes should be minimal. These
small changes can sometime create brain shutdown ;)

> * is there any library outside GCC that brings support for basic math
> functions? is there any plan to integrate such support into GCC?

I don't know of any. However, these functions are a foundation that
allows building these libraries. OTOH, I can't imagine an application
that requires dynamic range offered by 128bit FP. Perhaps to count
every fundamental particle in space? Or in high-high-high-end audio
processing ;)

> * is there any library outside GCC that brings support for I/O
> functions? is there any plan for such support in GCC?

There was a patch proposed for inclusion in glibc. I suggest that you
refrain from reading the response from the maintainer.

Uros.

Attachment: conv-xf.tar.bz2
Description: BZip2 compressed data