ippsMean 2

[carlo-turri]

Hi.
What I meant in the prev thread ippsMean 1 , ippsMeanColumn has the solution in how to make the averages for example in the FFT , let's suppose to have 1024 of FFT's beans magnetudes and we want to average them for 20 times , with ippsMeanColumn simple giving the 1024x20 array's pointer you have the result for each column in one 1024 1D array . no loop just one call , what Isee with ippsMean instead you have just a number for the average of one row, seem to be another job . Is this right ? or what else now can do the mentioned FFT job ?
Thanks
Cartu

[Vladimir_Dudnik]

Cartu,

could you please compare performance of loop of 20 iterations of calls ippsMean from 7.0 with performance of ippsMeanColumn from 6.1? I actually do not think there is big difference in performance.

Regards,
Vladimir

[jeffc111]

I think the two are not analogous. When we want to average across a set of vectors, the data is not aligned such that ippsMean can be used. If it were, it might mean thousands of calls to ippsMean.

The preferred alternative is, I think, is to use ippsAdd_*_I repeatedly, followed by an ippsDivC_*_I.