[lsb-discuss] OpenMP

[Wichmann, Mats D]

This landed in my mailbox... I know Robert isn't exactly
"in the office" right now but wondering if there are any
reactions.  No, I didn't get left with hints as to what
these deficiencies are...

====

Abstract: The well-known OpenMP programming model is based on compiler hints (so-called pragmas) that inform the compiler about how to transform the sequential code into a parallel version.  Programmers identify code fragments for parallelization and add parallelization hints to it.  OpenMP 3.0 increased OpenMP parallelization capabilities by introducing tasks.  But it turns out that for modern applications, OpenMP lacks essential features.  In this talk, I will present three issues that make writing optimal code with OpenMP 3.0 cumbersome and error prone.  The talk will shed some light on the proposals that might make into the next OpenMP specification to solve these issues.  The discussed topics include: user-defined reductions for non-standard reductions, better support for control of OpenMP task constructs, and an outlook to OpenMP error handling features.

====


Does this mean that OpenMP is still immature enough
that it ought not to be considered?  I know we won't
get to it for 4.1 anyway given the current constraints,
but it would be good to understand more of these issues.