TEmplate Massively PaRAllel Library for Efficient N-Dimensional Signal Processing

We present TEMPeRA: a Cpp library for efficient parallel signal processing, with a focus on image deconvolution. TEMPeRA makes porting new algorithms from MATLAB to Cpp easier than with the conventional methods. The library provides a class to describe a signal, with main point-wise algebraic operations, that is compatible with standard library generic algorithms. Interface for linear operators is also provided, suitable for modeling the blurring effect introduced by acquisition devices, such as telescopes or microscopes. Both classes defined in the library can exploit either CPU or GPU, using CUDA: this allows the end user to write device independent code. Moreover, thanks to policy-based template design, support for different architectures is introduced. The library is then exploited for the implementation of Richardson-Lucy deconvolution algorithm. Benchmark results shows remarkable speedup when comparing serial (CPU) code and parallel (CUDA) implementation.