topic Hi Alon, in Intel® Integrated Performance Primitives

Matrix Subtraction

alon_b_ — Tue, 20 Jan 2015 15:48:51 GMT

Hey,

This is probably an obvious question..

I have an operation I'd like to optimize with intel.

currently the input and output are 8 bit matrices, and I'd like the output to be the subtraction of those 2 matrices.

It seems obvious to me that the output would have to be at least 16 bit..

what is the ipp function to do it? I'm looking around the ippmSub but can't seem to find it.

Thanks,

Alon

pls look at the image

Gennady_F_Intel — Wed, 21 Jan 2015 07:34:11 GMT

pls look at the image processing functions: Subtracts pixel values of two images. - ippiSub_*.*

But as I said, Input bits and

alon_b_ — Wed, 21 Jan 2015 15:10:52 GMT

But as I said, Input bits and output bits are same count, how is overflow handled?

0-256 = -256

256-0 = 256.

you need 16 bits for 8 bit subtraction.

Hi Alon,

Igor_A_Intel — Wed, 21 Jan 2015 15:19:59 GMT

Hi Alon,

IPP for integer data works with saturation - therefore for 0-256 you'll get 0. If you need precise result - yes, you are right, you should convert both sources to 16s (for example with ippiConvert_8u16s_CxR) and then use ippiSub for 16s data type.

regards, Igor