Re: libm patch

[Stathis Kamperis <ekamperi%gmail.com@localhost>]

2010/8/27 Christos Zoulas <christos%astron.com@localhost>:
> Stathis Kamperis  <ekamperi%gmail.com@localhost> wrote:
>>2010/8/23 Stathis Kamperis <ekamperi%gmail.com@localhost>:
>>> Hi everyone,
>>>
>>> I have been continuing my work on libm (former gSoC project) and I
>>> have a patch[1] that I'd like to share with you.
>>> It's a bit mixed in terms of changes (sorry):
>>>
>>> * Added a few missing MLINKS
>>> * Rewritten some elementary complex functions to not rely on gcc
>>> extentions (creal, cimag, conj)
>>> * Implemented some new elementary real+complex functions (fabsl,
>>> copysignl, cproj*)
>>> * Probably the most important change, is the introduction of
>>> nextafterl() and friends, that will allow us to measure[2] the ULPs of
>>> the long double precision functions that will arrive soon
>>> * Some tidy up here and there
>
> good.
>
>>>
>>> The patch is WIP.
>>> I'm sharing early to get feedback, as my build cycles are quite long
>>> and I could batch changes :)
>>>
>>> Thoughts on:
>>>
>>> * Style (whether __ should be prepended to float_complex for instance
>>> (it isn't strictly needed, but gives a visual hint on the custom
>>> origin of the type))
>
> I don't know. What does FreeBSD do?
Just checked. They don't use __. I guess we could do the same, for
less noise while diffing against them.

>>> * File layout (should cproj.c be split into cproj.c, cprojf.c, cprojl.c?)
>>> * Everything else
>>> are very welcome.
>
> I think so to follow existing practice.
Ok, I'll break it up.

>>
>>Given the lack of general interest, I have filled a PR for this so it
>>doesn't get lost:
>>Re: lib/43807
>>
>>Until the experts can take a look at it :)
>
> Is this committable? There is interest, but not too many cycles to look at it.
It was a few days ago I checked (applied cleanly & built fine). Let me
make the changes (remove __ from custom types, break cproj.c to
type-specific files) and then give a new heads up.

Thanks!

Stathis