Using CUDA to help with the convertion

Developer
Apr 22, 2013 at 8:52 PM
I am doing some research but I think we could use nvidia cuda cores in gpu to help with converting the code. If we can do so it would take a lot of load off the cpu. I'm not completely possitive yet but it's something I am going to look into.
Coordinator
Apr 22, 2013 at 8:54 PM
Sounds interesting. Keep me posted.
Developer
Apr 22, 2013 at 8:55 PM
Will do.
Developer
Apr 22, 2013 at 11:23 PM
If I am unable to do something with cuda definitely use multi-threads to convert the code to make it fastest.
Coordinator
Apr 22, 2013 at 11:50 PM
I was a skeptical when you said that we could use CUDA. CUDA depends apon an ARM based compiler that uses a c like language.

We could use multi-processing and not multi-threading. Multi-threading is not designed for processing. It's designed to be able to do background IO tasks like downloading a file, or in situations where there is latency. However, you make a very good point about designing an algorithm that be run over symmetric multiprocessing. We could make the compiler support some form of inter-process communication. Where by we use multiple instances of the compiler. Which makes it is easier to force/control where the process resides in. For example, in windows you can specify which logical core or physical processor a process is executed on. I know that the .NET framework permits you to access the task management features like this.

At the moment, i feel that i might be too close to the compilation paradigm. I'v been reading the EMCA-334 specification and documenting the lexicon. I have also been noting the syntax rules written in the document. I haven't considered the system process to that level. In my mind, I've been saying "...just focus on the lexicon, we will get to the interpreter and assembler writer later...". I am afraid to get too ahead of myself.

Still, it is good to look a head while keeping a firm grip on the now.
Developer
Apr 23, 2013 at 12:04 AM
Edited Apr 23, 2013 at 12:07 AM
Ok I will keep you info if I think of away to use multi processes or cuda.
Developer
Apr 23, 2013 at 9:01 PM
Basically with cuda there is a warpper to c# but in order to utilize the cuda cores to make the lexer fastest we would need to convert many c# files at one time. So as this may serve as an option in future when we are using multi-class conversions it will not benefit at this point in time.
Coordinator
Apr 23, 2013 at 10:22 PM
Wow that sounds really cool.
Developer
Apr 24, 2013 at 1:14 AM
Here's a cuda lib for c# https://managedcuda.codeplex.com/