[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date: Mon, 5 Mar 2012 10:25:25 -0500
From: Rich Felker <dalias@...ifal.cx>
To: musl@...ts.openwall.com
Subject: Re: libm
On Mon, Mar 05, 2012 at 04:17:11PM +0100, Szabolcs Nagy wrote:
> * Rich Felker <dalias@...ifal.cx> [2012-03-05 09:04:59 -0500]:
> > On Mon, Mar 05, 2012 at 09:51:35AM +0100, Szabolcs Nagy wrote:
> > Hm? If you need a macro to test whether an argument is an integer or
> > floating point, try this:
> >
> > #define __IS_FP(x) !!((1?1:(x))/2)
>
> oh nice
> (and subtle)
Note that it avoids evaluating x too, so it's safe in macros that need
to avoid evaluating the argument more than once.
> > adopted it. In any case I see no good way to define them without
> > compound literals except the function calls..
> >
>
> btw while testing these macros i noticed that when
> multiple classification macros are used fpclassify
> gets called many times
> (with the previous solution)
>
> the extra calls could be optimized by adding
> __attribute__((const)) to fpclassify
> (resulted less calls, smaller code size etc)
Indeed. I was wrongly thinking they'd need __attribute__((__pure__))
rather than const, which is trickier because it's only available in
newish gcc. But const has been around since before 2.95 even, so just
#ifdef __GNUC__ is a sufficient check for using it.
Alternatively I might just move the definitions of fpclassify (at
least for non-LD types) back into math.h as static inline. This would
let gcc optimize it down a lot more.
Rich
Powered by blists - more mailing lists
Confused about mailing lists and their use?
Read about mailing lists on Wikipedia
and check out these
guidelines on proper formatting of your messages.
