LIGO Document P1400213-v1
- Detecting gravitational waves emitted by compact binary coalescences can provide important astrophysical information about the progenitor through a process called parameter estimation. Because parameter estimation, which involves finding theoretical waveforms and posterior probability distributions, is computationally expensive, we aim to expedite the process by optimizing on Intel hardware on the Stampede computing cluster, by adapting Intel compatible compilers and functions to improve performance. Our results indicate that the Intel hardware compatible MKL FFT implementation runs an order of magnitude faster than the generic FFTW implementation. Thus, given this improvement, we investigated optimizing the LSC Algorithms Library parameter estimation code as it utilizes FFTs. By reducing the computational cost of parameter estimation via optimization on hardware, longer gravitational wave signals can be analyzed, with the potential of extracting more information about the progenitor. We also compare the performance of the reduced order quadrature method for parameter estimation, which reduced the total runtime by an order of magnitude compared to the regular quadrature method.
DCC Version 3.4.1, contact