Re: [Qemu-devel] [Patch] Swapping functions for Mac OS X

[Laurent Amon]

Since I just looked, I can confirm it.

cpp includes this when including bswap.h :

static __inline
uint16_t
OSReadSwapInt16(
    __volatile void               * base,
    uintptr_t                          offset
)
{
    uint16_t result;
    __asm__ __volatile("lhbrx %0, %1, %2"
                     : "=r" (result)
                     : "b%"  (base), "r" (offset)
                     : "memory");
    return result;
}

And all wrappers are static inline so everything *should* get ASM 
inline.

However, I ran a few benchmarks and it makes no difference. So either 
gcc 3.3 is intelligent enough to do it anyway, or the code does not 
rely much on them, or they do not perform on my PPC7410. This bears 
more investigation.

Lga.

On 26 juil. 04, at 19:48, address@hidden wrote:

> Benjamin,
>
> I think that the NXSwap functions are defined as inlined asm in the 
> header.
>
> Pierre
>
> Quoting Benjamin Herrenschmidt <address@hidden>:
>
> > On Sat, 2004-07-17 at 09:50, Pierre d'Herbemont wrote:
> > > Hi Fabrice,
> > >
> > > I think it should be quite useful to use the swapping functions on 
> > > Mac
> > > OS X. However I don't see a huge performance gain with this patch.
> >
> > Ugh ? How cna it be useful to call into library functions why you can
> > use inline assembly ? Actually, if the swap is done on data 
> > loaded/stored
> > from/to memory, then you can use the ppc endian swapped load/store
> > instructions
> >
> > Ben.
> >
> >
> >
> > _______________________________________________
> > Qemu-devel mailing list
> > address@hidden
> > http://lists.nongnu.org/mailman/listinfo/qemu-devel
>
>
>
>
> _______________________________________________
> Qemu-devel mailing list
> address@hidden
> http://lists.nongnu.org/mailman/listinfo/qemu-devel