Development of High-Speed Processing of Gravitational Wave Radiometry using GPGPU

全球引力波观测网络预计将于2017年或2018年左右启动。引力波辐射测量是一种利用多探测器和长时间积分的方法，可以从天空的某个方向提取引力信号，即使是不相干的和微弱的波。然而，辐射测量计算需要冗长的计算，如天空位置的迭代和长时间观测的短块。因此，我们的研究计划的目标是使用GPGPU（通用计算图形处理单元）的辐射计算的高速化。我们实现了约100倍的速度与以前的CPU相比，GPGPU的计算。在同等条件下比较GPU和CPU并不简单，但结果是在相同的性价比下得出的。利用这种加速，我们研究了辐射测量应用的变化，例如使用不同波段的源分离滤波器。

The global gravitational wave observation network is expected to start around year 2017 or 2018. Gravitational wave radiometry is a method which will make possible to extract gravitational signal from certain direction of the sky using multi-detectors and long duration integration, even if waves are incoherent and weak. However, the radiometry calculation requires verbose calculations such as iteration for sky positions and of short chunks for long observation. Thus, the aim of our research program is high-speed up of radiometry calculation using GPGPU (General-purpose computing on graphics processing units).We achieved about 100 times faster calculation with GPGPU comparing with previous CPU. It is not straightforward to compare GPU and CPU in equal condition, but the result is derived on same cost-performance. Using this speed-up, we studied the variation of application of radiometry, e.g. source separate filter using different waveband.