I am interested in being able to reproduce the results from Intel Performance Primitives to floating point round off error. I’ve found that I’m able to reproduce the results in ippiWarpAffine and ippiWarpAffineLinear for 32f_C1R using nppiWarpAffineBack_32f_C1R for the bilinear interpolation case.
I would like to request some enhancements to nppiWarpAffineBack since it’s the algorithm I’m looking for.
Support for 64f_C1R as is found in nppiWarpAffine.
Support for a different cubic interpolation algorithm. I’d like the same behavior as:
status = ippiWarpAffineCubicInit(sizeSrc, sizeDst, dataType, T, direction, 1, // numChannels set to 1 Ipp64f(0.0), Ipp64f(0.5), borderType, fillVal+k, 0, pSpec, initBuf); ippiWarpAffineCubic_64f_C1R(src, srcStep, dst, dstStep, dstOffset, sizeDst, pSpec, pBuffer);
In other words, I’d like the CATMUL_ROM cubic spline interpolant discussed in the IPP doc:
Not sure if this is the right venue for asking for enhancement requests, I don’t have a question to ask. Happy to submit this somewhere else if that’s the right way to go.