Re: OPENMP
- Posted by mattlewis (admin) Oct 15, 2013
- 2406 views
Matt said: "What sinks in to me is that you are talking about things you don't really understand. It's fine to point this out and wonder about its applicability to euphoria. But when you start talking about how trivial it would be to make this great thing, you just make yourself look foolish and ignorant. "
Look at your own wxEuphoria and its complexity.
Then look at the functions etc that exist in OPENMP.
Purely from the point of view of wrapping it, I would say it is only one quarter of a challenge or even less.
You really have no clue about this.
As to "applicability". Intel, AMD and all those big companies are supporting OPENMP. I understand that Euphoria runs mainly on Intel/AMD processors (We won't talk about Mac and Androids here). It is almost axiomatic that a C language library specifically developed for the multicore processors of today should be compatible with a language developed mainly in C and somewhat in x86 assembler.
The only factor I have to look at is the current C compiler you all are using. OPENMP is compatible with a lot of C compilers, and I think I mentioned earlier that GCC compiler is the one I have in mend for this specific marriage. If GCC compiler will not compile Euphoria then it is a different story. I mentioned Clang in passing, knowing that you people have not used or worked with Clang.
I know you are probably immune to reason and facts, but what part of, "Dropping this into parts of euphoria would break many things. We would need to do a lot of work with euphoria to make it work with this." don't you understand?
If I was foolish or ignorant, cpmpanies would not pay me $20,000 just to supervise computer language related projects or hire me for $200/hour for a minimum contract of 6 hours/day for 3 days.
Yes that is what I charge. I mention it because you people do not seem to appreciate the value of good advice given with no financial motive just because I appreciate the hard work you all have done for free for a long time.
And the best part of it is that their work is completed and they are eager to pay and hire me again.
I can't imagine why they would pay you that much if this is the sort of things you do for them. Sounds like an excellent racket. In your gigs, do you actually do anything? Or do you suggest random frameworks that look appealing and walk away from the havoc you create?
I will therefore, reemphasize that Euphoria needs OPENMP, programmers can and will use it and derive great benefit from it, and the coding itself (without testing) can be done in 2 days. I might have been off in my estimate of 2 weeks of testing by author(s). In most projects I advice farming out of the field testing after a brief in-house testing and the way Euphoria is beta tested is similar to farming out in commercial projects.
We've known for some time that some form of multi-threading is needed and wanted. This is no revelation. Your post is interesting in that looking at openmp might suggest some interesting implementation possibilities, but your estimates or ideas about how it would work are complete fantasy. That you talk about "wrapping" it is enough to know that you haven't a clue how any of it works.
Take a look at this tutorial to get an idea. Some telling quotes:
- Most OpenMP parallelism is specified through the use of compiler directives which are imbedded in C/C++ or Fortran source code.
- The API provides for the placement of parallel regions inside other parallel regions.
- The OpenMP API is comprised of three distinct components. As of version 3.1:
- Compiler Directives (20)
- Runtime Library Routines (32)
- Environment Variables (9)
Matt
