Optimization of vertical and horizontal beamforming kernels on the PowerPC G4 processor with AltiVec technology
Y.H. Cho, D. Brunke, Gregory E. Allen, Brian L. Evans · 2002
Three-dimensional real-time digital sonar beamforming requires 4 to 12 GFLOPS, 1 to 2 GB of memory, and about 100 MB/s of I/O bandwidth. G.E. Allen and B.L. Evans have implemented a 4-GFLOP sonar beamformer in real-time on a Sun UltraSPARC II server with 16 333-MHz processors by utilizing the Visual Instruction Set (VIS) single-instruction multiple-data (SIMD) extensions. In this paper, we rewrite the horizontal and vertical beamforming kernels to use AltiVec SIMD extension for the PowerPC. AltiVec can execute up to four 32-bit floating-point multiply and accumulate (MAC) operations per instruction. In the PowerPC implementation, we prefetch and realign data for the I28-bit SIMD registers of AltiVec. We evaluate the performance of these beamforming kernels on the PowerPC and the UltraSPARC-II to evaluate the impact of the compiler, SIMD word alignment, and cache block alignment on performance.