Raul P. Pelaez 2018. FastCorrelation
Computes the correlation of two signals (autocorrelation if the signals are the same).
You can compute using a GPU with cuFFT or using a CPU with FFTW

BUILD:

Run:

mkdir build; cd build; cmake ..; make;

CMake will autodetect CUDA and FFTW3 and generate a valid Makefile for the current system.
You can also change the DONT_USE_CUDA option in cmake to force CPU only mode.
The option USE_MKL will force the usage of the MKL FFTW implementation (CMake will automatically try to find it if fftw3 is not present).

USAGE:

$ cat two_column_file  | correlation -N [N]  > corr.dat  

The output will have three columns:

	lag corr(lag) std(lag)

std = sqrt(<corr^2> - ^2)

If you have several files with realizations of the same signal, say signal1, signal2,...signalN you can interleave them to use with correlation as:

$ paste -d'\n' signal* | correlation ... > corr.dat  

INPUT FORMAT:

The input must have two columns with the following order:

       A_t0_signal1 B_t0_signal1  
       A_t0_signal2 B_t0_signal2  
       ...  
       A_t0_signal_nsignals   B_t0_signal_nsignals  
       A_t1_signal_nsignals   B_t1_signal_nsignals  
       ...  
       A_t_N_signal_nsignals  B_t_N_signal_nsignals  

OPTIONS:

-N: signal length  
-nsignals: number of signals in the file (they all must have length N) (default is 1).  
-windowSize: If present the signal will be cut in pieces of windowSize size for averaging (default is N).  
-maxLag: Compute up to lag maxLag (default is N). A lower value will increase performance and the results will be numerically identical up to maxLag.  
-noPad: If not present the signall will be padded with zeros up to maxLag in FFT mode. (default is false)  
-scale: Scale mode (default is biased), can be:  
      none:   Return the unscaled correlation, R.  
      biased: Return the biased average, R/N  
  unbiased: Return the unbiased average, R(t)/(N-t)  
-prec: float or double, specify the precision mode (default is float).  
-device: GPU or CPU, specify the device (default is CPU) to run FFT.  
-h or -help: Print this info.