Locality Optimization in the Implementation of FORALL
Yang Xue-jun · Computer Engineering and Science · 2005
The FORALL construct is a Fortran 95 syntax, the effective implementation of the FORALL construct is full of challenge. Especially, the locality optimization is very important for the compiler to generate efficient code when implementing the FORALL construct. In this paper, we introduce two locality optimizing methods, Temporary Space Group and Nested Loop Sort, to implement the FORALL construct in the G 95 compiler. The experimental results show that locality optimization is very efficient to gain speed benefit. Compared with the Intel EFC compiler, our implementation gains a performance improvement up to 30% for some kind of applications.