[Mono-dev] Jit and C# Compiler

Khaled A. K. Mohammed khaled.mohammed at gmail.com
Thu Oct 5 11:33:45 EDT 2006


On Thu, 2006-10-05 at 14:18 +0200, Cedric Vivier wrote:
> Hi!
> Sounds very interesting!
> If I am correct you want to detect (thus extract) parallelism from a
> CIL image right?
Yes, the extra information will help JIT to understand the nature of the

> Do you have any research paper describing the algorithm you plan to
> use ?

Well, one of the student of Prof. Tarek (University of Toronto) have
modified Jikes (the IBM java compiler) to do something similar. His
changes in compiler created two files: a java byte code and a registry
for runtime environment. And his changes in JIT will thread any method
it finds and will use the registry information to handle dependency
Prof. Tarek has done a lot of research on inserting extra information in
binary image. So I am guessing he has what I would need. 

The first step is to research on .NET to find out what kind of
optimization for multicore processor (parellelism) is currently

> Please keep us informed of your advancement :)
Sure, thanks.
> --
> cv
> 2006/10/5, Khaled A. K. Mohammed <khaled.mohammed at gmail.com>:
>         Hi,
>         I am doing a course on compiler. I am planning to choose a
>         project based
>         on .NET Framework/Mono.
>         I am briefly describing the project below.
>         The project will aim to improve the execution of CIL image on
>         multicore 
>         processors. First step is to change the C# compiler to insert
>         custom
>         instructions into the managed binary (just like how the line
>         #s are
>         stored). These instructions will be parsed by JIT to pick
>         valuable
>         information about instruction dependency and will be able to
>         make smart 
>         decisions. Yes, the size of binaries will grow. We will work
>         for a
>         optimal solutions - may be by having a percentage size
>         threshold.
>         This is a research/school project to prove a concept.
>         Lots of comments/suggestions are welcomed. 
>         Regards,
>         Khaled
>         _______________________________________________
>         Mono-devel-list mailing list
>         Mono-devel-list at lists.ximian.com
>         http://lists.ximian.com/mailman/listinfo/mono-devel-list

More information about the Mono-devel-list mailing list