Re: [OT] Altivec
Re: [OT] Altivec
- Subject: Re: [OT] Altivec
- From: David Duncan <email@hidden>
- Date: Fri, 4 Jul 2003 10:58:06 -0400
On Friday, July 4, 2003, at 06:43 AM, Mark's Studio wrote:
First is NSLog not supposed to print out vectors with the %vd format?
No idea. If it just calls into the BSD printf(), then I would think not
(at least according to the man page) as it is not supported there.
Here is the problem
peakBuffer is a const void * [NSData mutableBytes]
vSInt8 lmin,lmax is from the vec_min(),vec_max()
*(vSInt8 *)peakBuffer = lmin; NSLog(@"%d",*(SInt8
*)peakBuffer); peakBuffer+= 1;
*(vSInt8 *)peakBuffer = rmin; NSLog(@"%d",*(SInt8
*)peakBuffer); peakBuffer+= 1;
*(vSInt8 *)peakBuffer = lmax; NSLog(@"%d",*(SInt8
*)peakBuffer); peakBuffer+= 1;
*(vSInt8 *)peakBuffer = rmax; NSLog(@"%d",*(SInt8
*)peakBuffer); peakBuffer+= 1;
Altivec loads & stores always happen on 16-byte boundries, so your
efforts to store the vector into peakBuffer will always put them in an
aligned location - which is probably not where you want it.
// this is just to check what was stored
peakBuffer-= 4;
SInt8 t1,t2,t3,t4;
t1 = *(SInt8 *)peakBuffer; peakBuffer+= 1;
t2 = *(SInt8 *)peakBuffer; peakBuffer+= 1;
t3 = *(SInt8 *)peakBuffer; peakBuffer+= 1;
t4 = *(SInt8 *)peakBuffer; peakBuffer+= 1;
NSLog(@"%d %d %d %d",t1,t2,t3,t4);
Here is a bit of the output, the original value is right, but why is
the check not the same ?
And if peakBuffer is not 16-byte aligned, then the check will print
something completely different from the store above, as the scalar
loads here can happen in places where you could have not stored
anything in the prior section.
Is SInt8 and vSInt8 not the same size?
No they are not. SInt8 is 1 byte, while vSInt8 is 16 bytes (as are all
vector types). Altivec also only loads on a 16-byte boundry - there is
no way to load an entire vector off this alignment in a single
instruction. So given what your seeing, I'd make sure that peakBuffer
is 16-byte aligned.
--
Reality is what, when you stop believing in it, doesn't go away.
Failure is not an option. It is a privilege reserved for those who try.
David Duncan
_______________________________________________
coreaudio-api mailing list | email@hidden
Help/Unsubscribe/Archives:
http://www.lists.apple.com/mailman/listinfo/coreaudio-api
Do not post admin requests to the list. They will be ignored.
| References: | |
| | >[OT] Altivec (From: "Mark's Studio" <email@hidden>) |