Compositional Approach Applied to Loop Specialization

An optimizing compiler has a hard time to generate a code which will perform at top speed for an arbitrary data set size. In general, the low level optimization process must take into account parameters such as loop trip count for generating efficient code. The code can be specialized depending upon data set size ranges, at the expense of code expansion and decision tree overhead. We propose for loop structures a new method to specialize code at the assembly level, cutting drastically the overhead cost with a new folding approach. Our technique can generate and combine sequentially at the assembly level several versions, tuned for small, medium and large iteration number. We first show on the SPEC benchmarks the need for specialization on small loops. Then we demonstrate the benefit of our method on kernels with detailed results.

• Type: Conference
• Year: 2007
• Where: EUROPAR
• Authors: Lamia Djoudi, Jean-Thomas Acquaviva, Denis Barthou
• Keywords: Assembly Level, Code, Data Set Size, EUROPAR 2007