vpowf parameters backwards on Intel?

Subject: vpowf parameters backwards on Intel?

From: glenn andreas <email@hidden>

Date: Mon, 24 Apr 2006 14:23:04 -0500

Delivered-to: email@hidden

I've got an app that is vectorized, and I'm getting different results between Altivec/PPC and SSE/Intel builds. I've narrowed it down to a call to vpowf (as found in vecLib/vfp.h).

So if you take the following code:

#include <vecLib/vecLib.h>

#include <stdio.h>

int main(int argc, const char *argv[])

{

#ifdef __VEC__

vFloat two = (vFloat)(2.0,2.0,2.0,2.0);

vFloat >

#else

vFloat two = (vFloat){2.0,2.0,2.0,2.0};

vFloat >

#endif

vFloat sqrtTwo = vpowf(two, oneHalf);

printf("%vf ^ %vf = %vf\n",two, oneHalf, sqrtTwo);

return 0;

}

compile and run on PPC:

gandreas% gcc -faltivec -framework vecLib -o vpowfppc -isysroot /Developer/SDKs/MacOSX10.4u.sdk vpowf.c

gandreas% ./vpowfppc

2.000000 2.000000 2.000000 2.000000 ^ 0.500000 0.500000 0.500000 0.500000 = 1.414214 1.414214 1.414214 1.414214

Doing the same for i386:

gandreas% gcc -arch i386 -framework vecLib -o vpowf386 -isysroot /Developer/SDKs/MacOSX10.4u.sdk vpowf.c

gandreas% ./vpowfi386

2.000000 2.000000 2.000000 2.000000 ^ 0.500000 0.500000 0.500000 0.500000 = 0.250000 0.250000 0.250000 0.250000

Looks like the parameters are backwards to me...

Glenn Andreas email@hidden

<http://www.gandreas.com/> wicked fun!

quadrium | build, mutate, evolve | images, textures, backgrounds, art

_______________________________________________ Do not post admin requests to the list. They will be ignored. PerfOptimization-dev mailing list (email@hidden) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/perfoptimization-dev/email@hidden This email sent to email@hidden