Re: [PATCH v7 2/5] x86/asm: add volatile, clobbers and zero-length check in inline memcmp

On Fri, 24 Jul 2026 17:25:21 -0700
"H. Peter Anvin" <[email protected]> wrote:

> On 2026-07-24 16:57, H. Peter Anvin wrote:
> > 
> > And of course I got the comparison backwards (negative means a < b, thus we
> > want a - b not b - a). Here is a fixed version.
> > 
> > int memcmp(const void *a, const void *b, size_t size)
> > {
> >         int diff;
> > 
> >         asm volatile("xor %0,%0 ;"      /* Sets ZF for the size = 0 case */
> >                      "repe cmpsb ;"
> >                      "jz 1f ;"          /* When size = 0 loading is unsafe */
> >                      "movb -1(%1),%b0 ;"
> >                      "movzbl -1(%2),%k2 ;"
> >                      "sub %k2,%0 ;"
> >                      "1:"
> >                      : "=&q" (diff), "+D" (a), "+S" (b), "+c" (size)
> >                      : : "memory");
> >         return diff;
> > }
> >   
> For extra credit, this version is even smaller in 16- and 32-bit mode, but
> larger in 64-bit mode (because it depends on the order of the CMPSB operands,

The function is inlined, so it will make no difference.

	David

> which is the inverse of what the x86-64 ABI expects; swapping the order of "a"
> and "b" and adding a cmc instruction improves the x86-64 size, but 64 bits is
> not where the really tight code is...)
> 
> int memcmp(const void *a, const void *b, size_t size)
> {
>         int diff;
> 
>         asm volatile("xor %0,%0 ;"
>                      "repe cmpsb ;"
>                      "jz 1f ;"
>                      "sbb %0,%0 ;"
>                      "or $1,%0 ;"
>                      "1:"
>                      : "=&r" (diff), "+S" (a), "+D" (b), "+c" (size)
>                      : : "memory");
>         return diff;
> }
> 
>