Data fitting Doppler data processor with OpenMP and CUDA GPU backends

kernel algrithm : http://www1.icsi.berkeley.edu/~storn/code.html#csou

Author

Nianchuan Jian and Dmitry Mikushin

Prerequisites

$ sudo apt-get install pgplot5

Install and run

• $ make -j24
• $ cd bin
• $ ./phase_track

Release notes

This version uses the new makefile system, which is not bound to Intel compiler. The package could be compiled with any modern set of GNU C/C++11 and Fortran compilers.

This version uses Fortran version of libspice for data I/O: https://sourceforge.net/projects/ngspice/files/ng-spice-rework/27/ The source code of necessary function is shipped together with this package, no external libraries are needed.

Function evaluation can use either CPU (OpenMP) implementation or GPU (CUDA/Thrust). On systems with NVIDIA GPU available, evaulation will use GPU version by default. In order to enforce the use of CPU (OpenMP) version on GPU-enabled machine, disable GPUs visibility:

$ CUDA_VISIBLE_DEVICES= ./phase_check

GPU version is set to build for SM 3.5 (Kepler) and 6.1 (Pascal) GPUs. In order to add any other compute capability, locate the following lines in the Makefile and adjust them accordingly:

GPUARCH += -gencode=arch=compute_35,code=sm_35 # Kepler GK110 CC 3.5
GPUARCH += -gencode=arch=compute_61,code=sm_61 # Pascal GP106 CC 6.1