In several places in comm_tod_noise_mod.f90 there is code analogous to
dt(1:ntod) = d_prime(:)
dt(2*ntod:ntod+1:-1) = dt(1:ntod)
call sfftw_execute_dft_r2c(plan_fwd, dt, dv)The array dt represents an even function, i.e. the FFT transforms a function symmetric around the origin. This is in fact a discrete cosine transform of type 2, which can in principle be computed twice as quickly than the real-valued FFT used by the code.
If a significant fraction of overall run time is spent on these FFTs, it's probably worth switching to DCTs. They are supported by FFTW (see http://www.fftw.org/fftw3_doc/Real-even_002fodd-DFTs-_0028cosine_002fsine-transforms_0029.html).
In several places in
comm_tod_noise_mod.f90there is code analogous toThe array
dtrepresents an even function, i.e. the FFT transforms a function symmetric around the origin. This is in fact a discrete cosine transform of type 2, which can in principle be computed twice as quickly than the real-valued FFT used by the code.If a significant fraction of overall run time is spent on these FFTs, it's probably worth switching to DCTs. They are supported by FFTW (see http://www.fftw.org/fftw3_doc/Real-even_002fodd-DFTs-_0028cosine_002fsine-transforms_0029.html).