topic Re: buffersize in function ippmInvert_ma_32f_S2 in Intel® Integrated Performance Primitives

buffersize in function ippmInvert_ma_32f_S2

mweissbach — Wed, 10 May 2006 18:42:49 GMT

Hi

I tried to use the ippmInvert_ma_32f_S2-Function for small matrices. I recognized, that the size of the buffer has to be larger than heightWidth^2. In my case the function works fine, if bufferSize = 1.5 * heightWidth^2 is used. I don't know if its a bug or the error on my side.

I'm using the IPP 4.0.

Thanks a lot

M.W.

Message Edited by MWeissbach on 05-10-200605:57 AM

Re: buffersize in function ippmInvert_ma_32f_S2

Vladimir_Dudnik — Wed, 17 May 2006 21:22:36 GMT

Hello,

there is answer from our expert:

The IPP 4.0 documentation contained the bug which has been fixed since the version IPP 4.1.

The actual buffer size is (heightWidth^2+heightWidth).

I would like to advise using the last IPP versions for small matrices which provide better performance and reliability.

Note, the latest IPP version is v5.1

Regards,
Vladimir