Re: [RFH] Folding of (int)round to lround (and friends)

[Richard Guenther <rguenther at suse dot de>]

On Tue, 6 Dec 2005, Kaveh R. Ghazi wrote:

>  > patch below implements folding of integer conversions of
>  > the rounding functions to the specific C99 routines.  And there
>  > is the problem - given suitable expanders, code generated for
>  > lround can be a lot better than for (int)round, but lround is
>  > only available from C99 (and maybe doesn't have a library fallback).
>  > So with the patch below I get
>  > 
>  > #include <math.h>
>  > 
>  > int foo(double x)
>  > {
>  >   return round(x);
>  > }
>  > 
>  > > gcc -c -O2 t.c
>  > t.c: In function 'foo':
>  > t.c:5: warning: incompatible implicit declaration of built-in function 
>  > 'llround'
>  > 
>  > which I can only avoid with -std=c99.  How do we generally deal with
>  > this?  Not do the transformation if !c99?
> 
> I think you want to check the macro TARGET_C99_FUNCTIONS in your code.
> 
> However I thought this transformation, (int)round to lround, was
> already done by GCC.  See: gcc.dg/torture/builtin-convert-2.c
> 
> Why do we need your patch?

Doh, we indeed do.  But it's buried in convert.c:convert_to_integer.
A more appropriate place is fold().  Also with that code we, too,
get the spurious warning, just this time for 'round', not llround:

> gcc -O2 -c t.c
t.c: In function 'foo':
t.c:5: warning: incompatible implicit declaration of built-in function 
'round'

I think warning here is a bug.

Richard.