A Comparison of Compiler Tiling Algorithms

摘要: Linear algebra codes contain data locality which can be exploited by tiling multiple loop nests. Several approaches to have been suggested for avoiding conflict misses in low associativity caches. We propose a new technique based on intra-variable padding and compare its performance with existing techniques. Results show improves of matrix multiply over 100% some cases range sizes. Comparing the efficacy different algorithms, we discover rectangular tiles are slightly more efficient than square tiles. Overall, from 0-250%. Copying at run time proves quite effective.