From: Paolo Bonzini <bonzini@gnu.org>
Date: Mon, 17 Aug 2009 17:20:18 GMT
Subject: Re: Linus' sha1 is much faster!
Message-ID: <4A899152.8070607@gnu.org>
In-Reply-To: <alpine.LFD.2.00.0908171228570.6044@xanadu.home>

> my first version of
> get_be32() was a macro that did this:
>
> #define SHA_SRC(t) \
>    ({ unsigned char *__d = (unsigned char *)&data[t]; \
>       (__d[0]<<  24) | (__d[1]<<  16) | (__d[2]<<  8) | (__d[3]<<  0); })
>
> With such a construct, gcc would always allocate a register to hold __d
> and then dereference that with an offset from 0 to 3.  Whereas:
>
> #define SHA_SRC(t) \
>     ({   unsigned char *__d = (unsigned char *)data; \
>          (__d[(t)*4 + 0]<<  24) | (__d[(t)*4 + 1]<<  16) | \
>          (__d[(t)*4 + 2]<<   8) | (__d[(t)*4 + 3]<<   0); })
>
> does produce optimal assembly as only the register holding the data
> pointer is dereferenced with the absolute byte offset.  I suspect your
> usage of inline functions has the same effect as the first SHA_SRC
> definition above.

Yes, that's what happens.

Paolo