Automatic Tuning of Whole Applications:

For many years, retargeting of applications for new architectures has been a major headache for high performance computation. As new architectures have emerged at dizzying speed, we have moved from uniprocessors, to vector machines, symmetric multiprocessors, synchronous parallel arrays, distributed-memory parallel computers, and scalable clusters. Over the past year, clusters based on multicore chips (with partially shared cache hierarchies) have further added to the complexity of computing platforms. Each new architecture, and even each new model of a given architecture, has required retargeting and retuning every application, often at the cost of many person-months or years of effort.