RE: [RFC PATCH v1 3/5] tools/nolibc: x86-64: Use `rep cmpsb` for `memcmp()`

[David Laight]

From: Willy Tarreau
> Sent: 30 August 2023 22:27
> 
> On Wed, Aug 30, 2023 at 08:57:24PM +0700, Ammar Faizi wrote:
> > Simplify memcmp() on the x86-64 arch.
> >
> > The x86-64 arch has a 'rep cmpsb' instruction, which can be used to
> > implement the memcmp() function.
> >
> >     %rdi = source 1
> >     %rsi = source 2
> >     %rcx = length
> >
> > Signed-off-by: Ammar Faizi <ammarfaizi2@xxxxxxxxxxx>
> > ---
> >  tools/include/nolibc/arch-x86_64.h | 19 +++++++++++++++++++
> >  tools/include/nolibc/string.h      |  2 ++
> >  2 files changed, 21 insertions(+)
> >
> > diff --git a/tools/include/nolibc/arch-x86_64.h b/tools/include/nolibc/arch-x86_64.h
> > index 42f2674ad1ecdd64..6c1b54ba9f774e7b 100644
> > --- a/tools/include/nolibc/arch-x86_64.h
> > +++ b/tools/include/nolibc/arch-x86_64.h
> > @@ -214,4 +214,23 @@ __asm__ (
> >  	"retq\n"
> >  );
> >
> > +#define NOLIBC_ARCH_HAS_MEMCMP
> > +static int memcmp(const void *s1, const void *s2, size_t n)
> > +{
> > +	const unsigned char *p1 = s1;
> > +	const unsigned char *p2 = s2;
> > +
> > +	if (!n)
> > +		return 0;
> > +
> > +	__asm__ volatile (
> > +		"rep cmpsb"
> > +		: "+D"(p2), "+S"(p1), "+c"(n)
> > +		: "m"(*(const unsigned char (*)[n])s1),
> > +		  "m"(*(const unsigned char (*)[n])s2)
> > +	);
> > +
> > +	return p1[-1] - p2[-1];
> > +}
> 
> Out of curiosity, given that you implemented the 3 other ones directly
> in an asm statement, is there a particular reason this one mixes a bit
> of C and asm ? It would probably be something around this, in the same
> vein:
> 
>   memcmp:
>     xchg  %esi,%eax   // source1

Aren't the arguments in %rdi, %rsi and %rdx so you only
need to move the count (below) ?
(Looks like you copied memchr())

	David

>     mov   %rdx,%rcx   // count
>     rep   cmpsb       // source2 in rdi; sets ZF on equal, CF if src1<src2
>     seta  %al         // 0 if src2 <= src1, 1 if src2 > src1
>     sbb   $0, %al     // 0 if src2 == src1, -1 if src2 < src1, 1 if src2 > src1
>     movsx %al, %eax   // sign extend to %eax
>     ret
> 
> Note that the output logic could have to be revisited, I'm not certain but
> at first glance it looks valid.
> 
> Regards,
> Willy

-
Registered Address Lakeside, Bramley Road, Mount Farm, Milton Keynes, MK1 1PT, UK
Registration No: 1397386 (Wales)