[FFmpeg-devel] libavutil simd

[Nigel Pearson]

> Not useful. A utility means forking. Lookup how this utility gets its
> information

Its copyright is the Apple Public Source License,
so I don't think it is appropriate to dissect it.


I found a page that lists another method:
http://www.cocoadev.com/index.pl?HowToGetHardwareAndNetworkInfo
which shows Mach calls for detecting CPU.



It also mentions a sysctl parameter, hw.optional.altivec.
I am not sure that this will work on *BSD,
but it does the trick on OS X:

#include <sys/sysctl.h>

int main()
{
     size_t len = 0;
     char   *name = "hw.optional.altivec";
     long   result;

     if (sysctlbyname(name,  NULL,  &len, NULL, 0) == -1)
         return 0;   // altivec is not even an option on an i386

     if (len > sizeof(result))
         return 0;   // this is not a size or frequency, so should be  
32bits

     if (sysctlbyname(name, &result, &len, NULL, 0) == -1)
         return 0;


     printf("sysctl(%s)  returned %d\n", name, result);
}


No guarantee that this doesn't cause a kernel context switch.
Same thing for the Mach calls.


...
> your test is backwards. You should assume altivec
> is available unless one of a fixed set of pre-altivec cpu types is
> detected. The number of cpu models without altivec is already
> fixed/finite. The number of cpu models with altivec is unlimited since
> all future ones will have it.

Moot, because I suspect that Apple
will only ship Intels from now on?

--
Nigel Pearson, nigel at ind.tansu.com.au|"Gentlemen,
Telstra Net. Eng., Sydney, Australia | you can't fight in here-
Office: 9202 3900    Fax:  9261 3912 |   this is the war room!"
Mobile: 0408 664435  Home: 9792 6998 |     Dr Strangelove