topic Vector find optimize routine in Intel® oneAPI Math Kernel Library

Vector find optimize routine

lgmendes — Wed, 24 Feb 2010 15:32:43 GMT

Dear all,

How can I write the following code using the mkl (or other Intel library) optimized vector routines? The main idea is to remove the for in the "main" code.

for (v=vsize;v--;){

if (estimate > 0.05) {
estimate = 200.0;
} else {
estimate = 1.0;
}

}

Thanks in advance,

Luis

Vector find optimize routine

Gennady_F_Intel — Wed, 24 Feb 2010 15:57:51 GMT

Luis,

there are no such routines into MKL, but you can find similar functionality in Intel IPP.

Please see into signal processing functionality - threshold functions, which performs the threshold operation on the elements of a vector by limiting the element values by specified value.

as an example:

IppStatus ippsThreshold_64f(const Ipp64f* pSrc, Ipp64f* pDst, int len, Ipp64f level, IppCmpOp relOp);

where Ipp64f == double precison

Please let us know if any furher guiestions or example .

--Gennady

Vector find optimize routine

TimP — Wed, 24 Feb 2010 16:08:39 GMT

Quoting lgmendes

The main idea is to remove the for in the "main" code.
for (v=vsize;v--;){

if (estimate > 0.05) {
estimate = 200.0;
} else {
estimate = 1.0;
}

}

You should be able to vectorize this with icc, something along the line of

#pragma vector always

for (v=0; vestimate = (estimate > 0.05) ? 200. : 1.

If you don't like for() you should be using Fortran:

estimate(0:vsize-1) = merge(200d0,1d0,estimate(0:vsize-1) > .05d0)