Re: [2.6.6-BK] x86_64 has buggy ffs() implementation

[Gabriel Paubert]

On Wed, May 12, 2004 at 08:31:56PM +0000, H. Peter Anvin wrote:
> Followup to:  <1084369416.16624.53.camel@xxxxxxxxxxxxxxxxx>
> By author:    Anton Altaparmakov <aia21@xxxxxxxxx>
> In newsgroup: linux.dev.kernel
> >
> > Hi Andi, Andrew, Linus,
> > 
> > x86_64 has incorrect include/asm-x86_64/bitops.h::ffs() implementation. 
> > It uses "g" instead of "rm" in the insline assembled bsfl instruction. 
> > (This was spotted by Yuri Per.)
> > 
> > bsfl does not accept constant values but only memory ones.  On i386 the
> > correct "rm" is used.
> > 
> > This causes NTFS build to fail as gcc optimizes a variable into a
> > constant and ffs() then fails to assemble.
> > 
> 
> Of course, this is a good reason to do a __builtin_constant_p()
> wrapper that gcc can optimize:
> 
> static __inline__ __attribute_const__ int ffs(int x)
> {  
> 	if ( __builtin_constant_p(x) ) {
> 		unsigned int y = (unsigned int)x;
> 		if ( y >= 0x80000000 )
> 			return 32;
> 		else if ( y >= 0x40000000 )
> 			return 31;
> 		else if /* ... you get the idea ... */

Either I'm asleep or you are emulating bsrl, not bsfl. It
should rather be:

	if ( y & 0x00000001) return 1;
	if ( y & 0x00000002) return 2;
	if ( y & 0x00000004) return 3;
	...
	if ( y & 0x80000000) return 32;
	return 0;

No need for the else clauses either because of the return.
But maybe even __builtin_ffs(y) would work in this case.

	Gabriel
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/