[Mono-dev] Cilk for Mono

[Liang, Qichang (梁启昌）]

What do you mean for the "runtime"?

In the cilk-5.4.6 for C

http://supertech.csail.mit.edu/cilk/

.cilk and .cilkh files is first compiled to vanilla .c and .h files by
the cilk2c compiler.  The generated .c and .h files includes headers
of the runtime library (in C).  These generated .c and .h files are in
turn compiled by an ordinary C compiler such as gcc and linked to the
runtime library.

As discussed in my previous posts, Cilk can have similar arrangement in C#:

The extended language (C# plus spawn and sync) is parsed by dmcs, then
the AST is expanded in-place within dmcs to conform to vanilla C#, and
finally dmcs emits CIL that calls the runtime library that is written
in vanilla C# and compiled into CIL.

Therefore compatibility with .NET should not be affected, because:
- the virtual machine and the CIL language is not changed.
- any valid C# programs are also valid in the extended language

So any program that can run on vanilla .NET should be run on mono.

Finally, we can make the Cilk support to be off by default, and can be
turned on by supplying a flag to dmcs, such as "-fcilk".

In the default mode, cilk keywords are not allowed and code generated
will *not* link to cilk runtime library
if "-fcilk" is addedm then dmcs will enter the cilk mode and produce
CIL that may call the cilk runtime library, and the resultant code
would run natively in mono, or if run on native .NET, the cilk runtime
library path must be made available to .NET

Would this work?

Qichang


On Fri, Nov 5, 2010 at 9:56 PM, Stifu <stifu at free.fr> wrote:
>
> Would this require a change in the runtime, or only in the compiler?
> If the runtime needs to be updated, that would break compatibility with
> .NET.
> I doubt this could work as an external library, right?
>
>
> kcleung wrote:
>>
>> I am surprised that the Cilk task-stealing mechanism, which is
>> algorithmically proven to be the most efficient, is not included in C#
>> 5.0 specification.
>>
>> However would it be feasible for us to implement Cilk in dmcs for Mono
>> 3.0?
>>
>> The resultant language will have the new keywords "spawn" and "sync"
>> (with support for atomic "+=", "-=" operators on spawn results) and
>> inlets.  The compiler should also forbid using results from spawn
>> before calling "sync"
>>
>> Beside performance advantages, the cilk model is really much simpler
>> than the async interface in C# 5.0
>>
>> Since Cilk (developed by MIT) is *not* patented, there should be no
>> legal obstacles preventing us from implementing the Cilk flow model
>> onto C#, or other mono / .NET languages.
>>
>> What do you think?
>>
>> Cheers,
>>
>> Qichang
>> _______________________________________________
>> Mono-devel-list mailing list
>> Mono-devel-list at lists.ximian.com
>> http://lists.ximian.com/mailman/listinfo/mono-devel-list
>>
>>
>
> --
> View this message in context: http://mono.1490590.n4.nabble.com/Cilk-for-Mono-tp3028260p3028313.html
> Sent from the Mono - Dev mailing list archive at Nabble.com.
> _______________________________________________
> Mono-devel-list mailing list
> Mono-devel-list at lists.ximian.com
> http://lists.ximian.com/mailman/listinfo/mono-devel-list
>