topic Optimal 3D representation??? in Intel® Integrated Performance Primitives

Optimal 3D representation???

apolo74 — Wed, 20 Jul 2011 08:42:54 GMT

What would be the most efficient way of representing a 3D object?

In my application I'll need to apply convolution to a 3D object; this is not a problem for 1D or 2D:

1D. Ipp32f* myArray = ippsMalloc_32f( sizeX ); ...; ippsConv_32f( ... );

2D. Ipp32f* myMatrix = ippsMalloc_32f( sizeX * sizeY ); ...; ippiConvValid_32f_C1R( ... );

3D. ??? Maybe a vector of sizeZ, but then how to apply the convolution in the 3rd dimension

or a Ipp32f**, but again the same problem as before. And a really big Ipp32f* [sizeX * sizeY * sizeZ] gives me the same problem, how to apply the convolution on the 3rd dimension.

Thanks for any help and suggestion,

Boris

Optimal 3D representation???

Joseph_S_Intel — Fri, 22 Jul 2011 23:10:07 GMT

You might want to try using the Vector Statistical part of the Intel Math Kernel Library http://software.intel.com/en-us/articles/intel-mkl/ it handles multidimensional convolution which would include 3D convolution

Optimal 3D representation???

apolo74 — Mon, 25 Jul 2011 15:13:07 GMT

Thanks for the tip Walter, it seems that MKL has what I need. However I was wondering if you know about future plans for IPP to include a multi-dimensional approach (Convolution, Fourier). I like these libraries very much and I would like to stay with them if possible. Thanks again for the suggestion,

Boris