gold patch committed: Provide own version of ffsll

[Ian Lance Taylor]

Dave Korn <dave.korn.cygwin@googlemail.com> writes:

> Ian Lance Taylor wrote:
>> gold uses ffsll, particularly since gcc expands it inline when
>> possible.  However, some systems don't have ffsll.  I committed this
>> patch to add a version when the system does not have it.  I also tweaked
>> the other replacement functions, adding declarations so that they would
>> build without warnings.
>
>   Did you consider adding these to libiberty instead?  I don't know if they're
> all appropriate but some of them might well make handy additions, mightn't
> they?

I thought about it.  I added memmem to libiberty.  The version of mremap
in gold isn't general; it only does what gold needs.  ffsll is kind of
special purpose and I'm not convinced anything else will call it.


> +int
> +ffsll (long long arg)
> +{
> +  unsigned long long i;
> +  int ret;
> +
> +  ret = 0;
> +  for (i = (unsigned long long) arg; i != 0; i >>= 1)
> +    ++ret;
> +  return ret;
> +}
>
> ... does that really find the first set bit?  It looks to me like it counts
> all the one bits.  Possibly the terminating condition was meant to test "(i &
> 1) != 0"?

It doesn't matter too much, since gold will only call it with a power of
2.  But, yeah, I suppose it ought to work properly.  I committed the
appended patch.  Thanks.

Ian


2009-03-30  Ian Lance Taylor  <iant@google.com>

	* ffsll.c (ffsll): Correct implementation.


-------------- next part --------------
A non-text attachment was scrubbed...
Name: foo.patch
Type: text/x-patch
Size: 572 bytes
Desc: Fix ffsll
URL: <https://sourceware.org/pipermail/binutils/attachments/20090330/462d6562/attachment.bin>