Create a "processing" module ?

Base classes and backends

An increasing number of classes have several backends.

In principle the nabu modules breakdown tries to follow the tomography steps (estimation, pre-processing, phase, reconstruction, stitching). But these classes can be used in any step (ex. fft), so it doesn't fit this scheme.

For now many of these are in nabu.misc, but the name is unfortunate. The idea is always the same:

• One base class with common code between backend, and possibly the reference (numpy-based) implementation
• Cuda and OpenCL backend based on this class

In the current state (see below), the different modules/classes are scattered in a not-so-consistent way.

There are at least 2 options:

1. Put everything in a common "processing" module, and have feature.py, feature_cuda.py and feature_opencl.py
2. Put only the base classes in "misc", and the backends in other folders: misc.feature, cuda.feature, opencl.feature. This is roughly the current state. The drawback is that the tests (for all backends) are in misc, which sounds weird.

Current state

Padding:

• PaddingBase in misc
• CudaPadding in cuda
• OpenCLPadding in opencl

Histogram

• PartialHistogram in misc
• CudaPartialHistogram in misc
• OpenCLPartialHistogram TBD

Unsharp

• UnsharpMask in misc
• CudaUnsharpMask in misc
• OpenCLUnsharpMask in misc

Rotation

• Rotation in misc
• CudaRotation in misc
• OpenCLRotation TBD

Processing

• ProcessingBase in misc
• CudaProcessing in cuda
• OpenCLProcessing in opencl

Kernel

• KernelBase in misc
• CudaKernel in cuda
• OpenCLKernel in opencl

Convolution

• CudaConvolution in cuda
• OpenclConvolution in silx

MedFilt

• MedianFilter in cuda
• no other backend