[cairo] [patch] Accelerate CAIRO_BITSWAP8 on 64 bit machines

[Chris Wilson]

On Tue, 29 May 2012 07:51:09 +0000, "Xing, Dongsheng" <dongsheng.xing at intel.com> wrote:
> -----Original Message-----
> From: Chris Wilson
> Sent: Tuesday, May 29, 2012 3:25 PM
> To: Xing, Dongsheng; cairo at cairographics.org
> Subject: Re: [cairo] [patch] Accelerate CAIRO_BITSWAP8 on 64 bit machines
> 
> > On Tue, 29 May 2012 02:03:58 +0000, "Xing, Dongsheng" wrote:
> > > Hi, CAIRO_BITSWAP8 costs seven operations on a 32 bit machine, but it 
> > > can be done with only four operations on 64 bit machines.
> > > This patch accelerates CAIRO_BITSWAP8 on 64 bit machines.
> >
> > This doesn't apply to master. Care to resend?
> 
> OK.
> 
> diff --git a/src/cairoint.h b/src/cairoint.h
> index 2c9f2dd..d6201f3 100644
> --- a/src/cairoint.h
> +++ b/src/cairoint.h
> @@ -169,7 +169,11 @@ do {                                       \
>   * Devised by Sean Anderson, July 13, 2001.
>   * Source: http://graphics.stanford.edu/~seander/bithacks.html#ReverseByteWith32Bits
>   */
> +#if (SIZEOF_LONG >= 8)
> +#define CAIRO_BITSWAP8(c) ((((c) * 0x80200802LLU & 0x884422110LLU) * 0x101010101LLU) >> 32)
> +#else
>  #define CAIRO_BITSWAP8(c) ((((c) * 0x0802LU & 0x22110LU) | ((c) * 0x8020LU & 0x88440LU)) * 0x10101LU >> 16)
> +#endif
> 
> I don't know why the patch doesn't apply to "master". 
> I guess the reason is my code is not the "master". Because I downloaded the code by 
>      git clone git://anongit.freedesktop.org/git/cairo
> I am not sure.

It justs means that the parent of this patch is not known upstream and
by my repository so git can not reconstruct the right patch for me.

> > > diff --git a/src/cairoint.h b/src/cairoint.h index 2c9f2dd..7e62d86 
> > > 100644
> > > --- a/src/cairoint.h
> > > +++ b/src/cairoint.h
> > > @@ -169,7 +169,11 @@ do {                                       \
> > >   * Devised by Sean Anderson, July 13, 2001.
> > >   * Source:
> > > http://graphics.stanford.edu/~seander/bithacks.html#ReverseByteWith32Bits
> > >   */
> > > +#if (SIZEOF_VOID_P >= 8)
> >
> > A more appropriate test here would be that "SIZEOF_LONG_LONG >= 8" as that is the condition used in the macro.
> 
> I think SIZEOF_LONG >= 8 is better. Because sizeof(long long)==8 on a 32 bit machine. And SIZEOF_LONG is the length of a native processor word.

Ok, if we can use SIZEOF_LONG here, wouuld 0x80200802LU suffice?
-Chris

-- 
Chris Wilson, Intel Open Source Technology Centre