[Mono-dev] Cilk for Mono
Liang, Qichang (梁启昌)
kcleung.sourceforge at gmail.com
Fri Nov 5 23:18:43 EDT 2010
Before we continue, we should have a good read on I-Ting Lee's Master
thesis (MIT) on her JCilk implementation:
http://people.csail.mit.edu/angelee/papers/angelee-msthesis.pdf
http://supertech.csail.mit.edu/papers/danaher-thesis.pdf
These two papers also discuss the language syntax:
http://people.csail.mit.edu/angelee/papers/jcilk-scp.pdf
http://people.csail.mit.edu/angelee/papers/jcilk-scool.pdf
2010/11/6 Liang, Qichang (梁启昌) <kcleung.sourceforge at gmail.com>:
> Glad to hear that we have System.Threading.Tasks in place,
>
>> For Task and work-stealing schedulers we have System.Threading.Tasks already
>> (albeit with no language support).
>
> so to implement the Cilk task scheduling system, does the runtime
> library, which contains the Cilk work-stealing scheduler just need to
> overload some interfaces / classes from System.Threading.Tasks? If
> so, could you please point me to the right one?
>
> Then the next step is obviously work on the dmcs compiler.............
>
>> Plus Cilk is released under GPL (afaict) which make it incompatible for us.
>
> This is irrelevant to us. It is only the Cilk *code*, which is
> written in C/C++ is licensed under GPL, not the algorithm itself, and
> as I said, the Cilk algorithm itself is not patented. Since the
> runtime library, which implements the stealing algorithm, will be
> written in C#, so we have to re-implement the algorithm anyway, so the
> license is *irrelevant* to us.
>
> When I look at the System.Threading.Tasks API, it looks tedious for me
> as a programmer to use, and it resembles a bit like the good old
> Pthread. To improve programmability and performance, should we go
> ahead and set up the Cilk subproject under mono?
>
> Cheers,
>
> Qichang
>
More information about the Mono-devel-list
mailing list