Let me get back to you when I get a chance to put together more detailed information.
But essentially it would be a code performance profiler directly hooked in to GCC for C code and I beleave there is now a gnu compiler for opencl which is a open code standard for writing vendor neutral code for GPGPU's and I know the GPL Radeon driver claims to support it on some Radeon.
The goal would be to advise the developer of thing that would prevent a program from spanning multiple cores like use of shared memory, improper flags when creating posix threads, and or insufficient parallel execution of tasks which could be run in parallel, not only in their own code but in any libraries they are using, which would prevent them from fully utilizing the hardware.
-- Sent from my HP Pre3
On Sep 8, 2013 13:50, Nima Sahraneshin-Samani <[log in to unmask]> wrote:
Dear Paul,
I want to know more a bout it . what should i do for this?
regards,
Nima
On 9/8/13, Paul Robert Marino <[log in to unmask]> wrote:
> I don't know if would work because its more general but extremely
> useful for Scientific computing. Cray use to have a tool for their
> Vector Processors which would evaluate code and advise on if it
> "Vectorized" correctly and what prevented it from doing so. The Vector
> processors were similar in design to a modern GPGPU. What they meant
> by Vectorizing was that the code once compiled would utilize all of
> the cores effectively in parallel. With the proliferation of
> relatively inexpensive GPGPU's and the increasing number cores in main
> CPU's this is a tool that is sadly missing and becoming more needed
> especially in the realm of scientific computing which in many cases
> needs to be able to get high performance out of off the shelf and in
> many cases older computers.
>
>
> On Sat, Sep 7, 2013 at 2:17 PM, Nima Sahraneshin-Samani
> <[log in to unmask]> wrote:
>> Dear All,
>>
>> I'm a master student of computer science-scientific computing and
>> interested to do a project related to operation systems.
>> I talked about OS with my supervisor and he accepted that if i
>> found a good idea for my master thesis around OS i can do
>> that.Could you please give me a project that be near to scientific
>> computation ?
>>
>> regards,
>> Nima
>