Deep searches for broadband extended gravitational-wave emission bursts by heterogeneous computing

We present a heterogeneous search algorithm for broadband extended gravitational-wave emission (BEGE), expected from gamma-ray bursts and energetic core-collapse supernovae. It searches the $(f,\dot{f})$-plane for long duration bursts by inner engines slowly exhausting their energy reservoir by matched filtering on a {\em Graphics Processor Unit} (GPU) over a template bank of millions of one-second duration chirps. Parseval's Theorem is used to predict the standard deviation $\sigma$ of filter output, taking advantage of near-Gaussian noise in LIGO S6 data over 350-2000 Hz. Tails exceeding a mulitple of $\sigma$ are communicated back to a {\em Central Processing Unit} (CPU). This algorithm attains about 65\% efficiency overall, normalized to the Fast Fourier Transform (FFT). At about one million correlations per second over data segments of 16 s duration $(N=2^{16}$ samples), better than real-time analysis is achieved on a cluster of about a dozen GPUs. We demonstrate its application to the capture of high frequency hardware LIGO injections. This algorithm serves as a starting point for deep all-sky searches in both archive data and real-time analysis in current observational runs.