Automatic relaxation: towards elimination of “relaxed”.Use of non-blocking communication (future work).Performance portability: make application performance less sensitive to the architecture.Usability of UPC: fine-grained programs with coarse-grained performance.Automatic optimizations of communication are key to.GAS languages can support both programming styles.shared memory programming style has more communication events.Fine-grained (OpenMP style) version still slower.GAS language outperforms MPI+Fortran (flat is good!).Source-to-source translation a good strategy for portable PGAS language implementations.Translator generated C code can be as efficient as original C code.Generating Optimizable (Vectorizable) Code Both compiler have improved since these were measured.Competitive with HP compiler, which generates native code.General blocked cyclic pointer 7x slower for addition.Indefinite pointers almost as fast as regular C pointers.Phaseless pointers are an important optimization.64 bits on small machines, 128 on large, word or struct.Pointer size/representation easily reconfigured.Some machine-specific special cases for some memory layouts.Phaseless (cyclic and indefinite) – skip phase update.Source of overhead for updating and de-referencing.A pointer needs a “phase” to keep track of where it is in a block.Block-cyclic, cyclic, and indefinite (always local). ![]()
0 Comments
Leave a Reply. |