Accelerated DSP functions on the CBE for the X-Midas Toolkit

We characterize the impact of the Cell Broadband Engine architecture, on commonly used radar DSP algorithms. We use the capabilities of the CBE to accelerate several key computational kernels including Matrix Multiplication, Matrix Inversion, and the Finite Impulse Response (FIR) filter. These algorithms are implemented and benchmarked as library routines within the X-Midas Toolkit. We observe speedups as large as 1200x for complex matrix multiplication, but speedups of 40x to 60x are more typical. We find that system I/O overhead within the X-Midas toolkit severely limited the performance of the applications.