Optimize your Fortran Application for SIMD and Multi-Core Parallelism using the Intel Compiler

Moore’s Law no longer gives us steadily increasing clock speeds, but instead gives us more cores and wider SIMD units. Applications need to make effective use of these in order to continue extracting performance improvements from current and future processor generations.

Find out to how obtain good SIMD performance through explicit vector programming in Fortran or through automatic vectorization, and learn how they compare. Get advice and avoid common pitfalls when creating, running, debugging and optimizing an OpenMP program using Intel Fortran. We will briefly summarize other optimizations in Intel Fortran, and there will be an opportunity to ask questions about any of these. Replies to questions not answered online will be posted afterwards.

For more complete information about compiler optimizations, see our Optimization Notice.

Comments

Ron,

Ron,

Thanks for this talk, that I have just found. It is very informative.
Where can I find the second part about parallelization ?

Also, should: a(j:n) = a(j+offset:n) +b(j:n) + c(j:n)
be written as: a(j:n) = a(j+offset:n+offset) +b(j:n) + c(j:n)
or is this not necessary?

John