Status of coarray implementation and performance

[OP1]

Is there any documentation regarding the respecting performance of coarray with respect to MPI for the latest Intel Fortran compilers (Linux/Cluster)? A couple years ago the focus was on implementation of coarray - with an explicit disclosure that performance and/or optimization were taking a very distant back seat to that primary goal.

Are there today published benchmarks for IVF coarrays that would help us make architectural decisions for a new, large code? Coarray performance was allegedly (and understandably) dismal with IVF a few years back. What is the current status today?

Thanks,
Olivier

[Ron_Green]

In Intel Visual Fortran Composer XE 2013 Update 2 we introduced some data blocking/aggregation improvements.  So be sure to use that compiler and nothing older. 

There is still a large body of work to be done for performance in the coming years.  It will be an ongoing process, much as getting array syntax optimized was back in the early days after Fortran 90 was introduced.  Do not expect a magical point in time when all performance issues suddenly disappear.  It will vary from application to application, usage by usage and will be an ongoing process for many years.

We have had customers tell us of dramatic improvements in performance with Update 2.  All I can say is try it for yourself with your application.

ron