Opinions on coding of Parallelism in C++

[craigmanson]

Hey All

I'm working on some exciting new Parallelism tooling for C++ with the University of St Andrews in Scotland.

We gave a presentation on this at the Edinburgh C++ Meetup group late last year (http://cppedinburgh.uk/ under ParaFormance) and had some good conversations.

I'm interested in feedback and opinions of professional developers on what we're trying to do. This is critical to help us successfully spinout from the University and finish creating a great tool for everyone to use.

We've created a survey to find out more about peoples needs and challenges with Parallelism in C++:
https://www.surveymonkey.co.uk/r/newsoftwaresurvey

It'd be great if you could help us be filling this out.

There'll be a draw for $100 of Amazon vouchers and the possibility of receiving a summary of the results for participants.

Also really happy to answer questions here or in private message about the project. The survey is a good way to prevent having to ask 20 different people what IDE they use....

Many Thanks
Craig

[Cubbi]

Hows does it mesh with the parallelism approach taken by ISO C++? (generic algorithms that use parallel and vectorized executors)
How does it mesh with the approach taken by Intel in Cilk+? (task parallelism with work-stealing)

[craigmanson]

In the tool we're creating we are building upon and extending these mechanisms. Currently we're inserting parallel code using the TBB or OpenMP libraries. In the future this will be extended to cover other approaches.

Essentially the tool is taking serial code and then rewriting that to a version which will execute in parallel, plumbing into 3rd party libraries to wrap the threading. It builds upon the capabilities of these libraries to offer a selection of parallel patterns which the code can be structured into. Exposing the possible options libraries have and extending them.

This part of the tool is not intended to produce results you could not achieve manually. It is intended to massively increase the speed and accuracy with which you can write parallel code.

Further to this there's aspects of safety checking (proactively identifying concurrency issues) and insight into where in the codebase parallelism should be applied. All helping you to develop products quicker and more robustly.

Sorry for the delay responding - been away at Hipeac conf all last week!