Fafalada said:
I'm curious, what kind of new language development is MS working on?
Don't know what MS research is doing, but what you need to do to a language to make it parallel friendly isn't exactly rocket science.
First thing you need to do is make mutable variables extremly RARE, ideally none existent.
After that you can help the compiler out by making all scope explicit give it enough information that it can establish all possible codepaths.
Loose anything in the language semantics that dictates order of potentially parallel operations (make sequential operations explicit).
Add some sort of native concept for streams with sliding windows.
Add Vectors as a native type (The compiler can then do all the nice stuff it does with other built in types).
Add Hints.
Most functional languages meet the first two criteria. The third is just a language design decision, and the last three aren't terribly hard to come up with solutions for. With all that the compiler has enough information to have a shot at generating parallel code, and it's very likely to make extremly poor decisions doing it.
I think a better model is to implement the above minus point 3, and use it to generate jobs that an explicitly parallel model can run. Building a language like the above makes it much harder to screw yourself with unintentional side effects.
Of course even with all that (and what's described is a language with very little in common with C or C++) you still have to solve the scheduling problem and given the instruction latencies of these processors and the in order designs that's far from a given.
