[Ffmpeg-devel] [PATCH] dead code elimination

[Roman Shaposhnick]

On Tue, 2007-02-27 at 00:54 +0100, Aurelien Jacobs wrote:
> On Mon, 26 Feb 2007 20:40:28 +0100
> Dominik 'Rathann' Mierzejewski <dominik at rangers.eu.org> wrote:
> 
> > On Monday, 26 February 2007 at 01:46, M?ns Rullg?rd wrote:
> > [...]
> > > Not OK.  This should be set for all 64-bit CPUs we currently have
> > > special settings for.  These include Alpha, PPC64, and Sparc64.  We
> > > don't seem to have any specific checks for MIPS64, a patch to rectify
> > > this would of course be welcome.
> > 
> > sparc64 is unlikely to be faster than sparc32. I can benchmark it if
> > you tell me how.
> 
> I don't know sparc very well, and indeed it seems that sparc32 is often
> faster than sparc64. But here it's not about sparc32 vs. sparc64.
> It's about sparc64 using some 32bits optimized code vs. sparc64 using
> some 64bits optimized code (ie. the code surrounded by HAVE_FAST_64BIT).
> 
> To benchmark it, I used START/STOP_TIMER at begining/end of
> idctRowCondDC(), but unfortunatly START/STOP_TIMER is not implemented
> for sparc :-(
> 
> If you find a good way to benchmark it, I'm interested in the results.
> Else I will probably commit this patch as is.

  What OS are we talking about here ? On Solaris one can use gethrtime()
instead of rdtsc. In fact, it might be even available on Linux.

  I think if Michael has nothing against it -- it might be a good idea
to make read_time() in libavutil/common.h default to gethrtime() if
the low level support is now provided by a particular architecture.

Thanks,
Roman.