|
From: | Lars Segerlund |
Subject: | Re: [Gomp-discuss] Architecture |
Date: | Wed, 22 Jan 2003 07:52:01 +0100 |
User-agent: | Mozilla/5.0 (X11; U; Linux i586; en-US; rv:1.2.1) Gecko/20021226 Debian/1.2.1-9 |
If you look at my suggestion, we would not need to wory about the threading model until code generation, the important thing would be to make gcc's GIMPLE tree's understand the directives, and then the threading model and handling of the primitives comes in in the code generation phase, not the parsing/optimizing part ( obviously these tree elements would have to make some sence to the optimizer :-) ).
If you have a look at treelang in the tree-ssa branch you can see that tree generation is quite simple, so we might even start off without a frontend, but only a stub which feeds 'specific' tree's to the backend. Thus the backend could be completed before the frontend work, and then the handling of the omp directives would be simplified to the insertion of a single tree node in the frontends.
This way, a certain 'independence' of the threading model would be reached ( actually it would be had for free :-) ).
So the question is, what needs to be specified and what is unknown ? / Lars Pop Se'bastian wrote:
On Tue, Jan 21, 2003 at 08:46:18AM -0500, Scott Robert Ladd wrote:
snip
I still have 3 exams this month, and then I'll be here for 5 months :-)Our first task, I think, is to decide on a threading model. I can look at what various OpenMP compilers do as well. I have several compilers (mostly Fortran) I can use to get an idea of different approaches. I think this will be our biggest challenge, given gcc's multi-platform focus. Are pthreads universal "enough", or do we need to take a generic view of threads?There is already some work in libgcc: have a look at gthr* files in gcc's directory: gthr-aix.h gthr-dce.h gthr-posix.h gthr-rtems.h gthr-single.h gthr-solaris.h gthr-vxworks.h gthr-win32.h gthr.h -> /* Threads compatibility routines for libgcc2. */ I don't know wether we could use this lib for a system independent thread model, but someone whispers that's in this direction we should scratch... Sebastian _______________________________________________ Gomp-discuss mailing list address@hidden http://mail.nongnu.org/mailman/listinfo/gomp-discuss
[Prev in Thread] | Current Thread | [Next in Thread] |