Research has shown time and again that the most efficient method to improve parallelism is to let the compiler find the instances explicitly. The compiler contribution to total performance is steadily increasing. In part this is the result of the increased pipelines, wider issue bandwidth, and longer latencies, especially in regard to memory.
Since the compiler views the total code stream, it can make global optimizations as it runs through the code. This also takes advantage of the new architectural features available in the compiler, such as speculation and predication. Given the new centrality of the compiler to the performance issue, HP at the time of the Itanium-based product introduction supports five native compilers, as shown in Figure 15-1.
18.118.26.90