Hi,

In my experience, there are very few instances where parallelism can be usefully concealed behind a library interface. OpenMP has very high overhead and will help only for very long-running functions -- at least under Microsoft's compiler on x86 and x64 platforms.

Right, but this is something that depends of the algorithm. For example, let's take the regular expression library implementation in Boost (Xpressive). Pattern matching is an expensive task and it's a kind of thing that comes in a library set like Boost. Researchers have tried parallelize regular expression matching ( http://ieeexplore.ieee.org/iel5/10901/34300/01636074.pdf). So, I guess there would be researches that are based on parallel libraries construction.

Algorithms that are likely to be applied to very large datasets could have the OMP pragmas inserted optionally but they would need to be protected by #ifdef logic (or given distinct names) because otherwise the overhead will destroy programs that make more frequent calls on smaller datasets.

Ok, but the feasibility study of the parallel version of an algorithm should be a part of the parallelization task. Not all of the parallelizable tasks would be really interesting running in parallel.

Also consider that, in an application that is already parallelized, there are no extra cores for the library to use internally.

Well, but what if the library is already parallelized? Maybe the extra task generated by parallelize the real application would be dispensable.

On a smaller scale, adding "vectorized" and/or "streaming producer-consumer" interfaces for selected libraries may help a lot by encouraging use of vector instruction / execution units, unrolling of loops and improving instruction and data locality.

This is a pretty good idea. Best, -- Cristianno Martins Mastering in Computer Science State University of Campinas cristianno.martins@students.ic.unicamp.br skype: cristiannomartins gTalk: cristiannomartins msn: cristiannomartins@hotmail.com