Related articles |
---|
Info needed (Sparc C compiler w/o register window) clim@ICSI.Berkeley.EDU (1992-04-14) |
Re: Info needed (Sparc C compiler w/o register window) pardo@cs.washington.edu (1992-04-16) |
No threads on SPARC? (was Re: Info needed (Sparc C compiler w/o regist mueller@mount.fsu.edu (1992-04-17) |
Re: No threads on SPARC? pardo@cs.washington.edu (1992-04-20) |
Newsgroups: | comp.compilers,comp.lang.c |
From: | mueller@mount.fsu.edu (Frank Mueller) |
Followup-To: | comp.compilers |
Keywords: | registers, sparc |
Organization: | Florida State University Computer Science |
References: | 92-04-058 92-04-068 |
Date: | Fri, 17 Apr 1992 17:44:47 GMT |
pardo@cs.washington.edu (David Keppel) writes:
> why you can't write a truly user-space threads package on the SPARC and
I disagree, I implemented a stripped-down version of preemtive Pthreads
using library implementation techniques. Your paper essentially points out
the same problems we encountered (e.g. having to flush the register
windows upon context switches, or, to be more precise, upon changing the
*stack pointer*, that's really what matters.) but non of the problems are
unsolvable. (At a later time, we might distribute Pthreads, we still have
to implement a few functions and to change from POSIX draft 5 to draft 6.)
Of course, we don't allocate a register window per thread since it's too
expensive. (David Keppel also pointed that out.) The exclusive use of
register windows for a thread might be of limited use for special-purpose
real-time application with a constant number n of threads where
n <= #register windows / 2.
(I believe you need two windows per thread to give traps some place to run.)
Frank
mueller@cs.fsu.edu
--
Return to the
comp.compilers page.
Search the
comp.compilers archives again.