Image processing in airborne applications using multicore embedded computers

• Multi-core speedup — Speedup highly depends on parallelizability of algorithms (serial dependencies vs. independent computation) — Results indicate performance of the benchmark • SIMD speedup — Achieved speedup of 1.3 considerably low — Speedup highly depends on efficient utilization of the AltiVecunit, e.g. constant availability of input data • Compiler optimization — Speedup between 1.5 and 1.7 — The higher the per-compiler optimization the lower the effect of compiler optimization (cf. hogC which has the highest degree of pre-compiler optimization) — Compiler optimization can partly compensate for pre-compiler optimizations of code • Overall Conclusions — Modern multi-core processors with SIMD features provide huge performance boost compared to single-core variants (for average cases) — Overall performance of CPUs still outperformed by GPUs