another Altivec question


    Jan 06 2007 | 5:24 pm

    • Jan 06 2007 | 5:37 pm
      After further investigations, I have replaced the (short) in my object struct by a (long) and it seems it has solved the problem... If anyone could explain it to me or send me to a proper documentation (or tell me it does not solve the problem for real ;-) that would be very much appreciated
      thanks
      pa
    • Jan 06 2007 | 7:58 pm
      > I have replaced the (short) in my object struct by a (long) and it
      > seems it has solved the problem...
      Sorry for the noise, but it does not solve my problem at all
      finally... I was just lucky. So if someone can give me any hint, I
      would be very grateful...
      pa
    • Jan 06 2007 | 8:42 pm
      On Jan 6, 2007, at 11:58 AM, Pierre Alexandre Tremblay wrote:
      >> I have replaced the (short) in my object struct by a (long) and it
      >> seems it has solved the problem...
      >
      > Sorry for the noise, but it does not solve my problem at all
      > finally... I was just lucky. So if someone can give me any hint, I
      > would be very grateful...
      Just an off the cuff comment, but I would suggest you not use vectors
      allocated in structs which might not have 16byte alignment (unless
      you use some non-standard struct packing strategy), and instead
      retain pointers to vectors allocated by getbytes16() or similar.
      Hope this helps.
      -Joshua
    • Jan 07 2007 | 2:29 pm
      Dear Joshua
      Thank you for your swift help. I have not tried this solution yet,
      but I am quite surprised by the proposal since the plus~ code
      provided in the UB SKD is doing exactly the same thing that I do. I
      have looked again at my code, and did not find any new hypothesis.
      I will try the pointer thing later today, but I will have to do my
      own destruction routine to get rid of the allocated memory... except
      if I am wrong again!
      Thanks again
    • Jan 07 2007 | 5:26 pm