Using system calls to create/context switch threads in assembly code

Hi all,
I am writing a compiler for a simple concurrent programming language. The compiler has to produce assembly code for Intel-32 architectures.
One of the major issue for code generation is how to create multithreaded assembly code. Linux provides system calls for creating threads, clone(,,,,). However, it seems there is no system call for doing context swiching for threads (is that so?). Also, our compiler has to deal with synchronizations among threads, so there should be a ready-queue that holds all ready-to-run threads and several waiting-queue. In order to do context swtiching, the compiler must be able to access each thread's CPU registers values (more specificly, the running thread's CPU register values need to be saved with the thread during context switching, so an array representing the CPU registers is needed to contruct a thread). I am not sure all of those are supported by the threads created with system calls.
Is it possible to construct a multi-threaded assembly program without making Linux system calls? Odd question, isn't? I do hope there is a solution.
Anyone has some suggestions?
Thanks

tony

In the end, something will have to make a system call. As far as I know, glibc and libpthread don't have access to any secret API, so they have to manage what they do using system calls. Have you looked through the pthread sources? I haven't, but they are just a small add-on to glibc. Looks like about 7200 lines of code (using cat *.{c,h} | egrep -cv "(^$|^/\*)") for version 2.5, which isn't a whole lot to search through. In case you aren't aware, look for the linuxthreads package at most glibc ftp sites. The glibc functions with system call names are generally just wrapped system calls, so you can probably infer the correct system calls based on what libpthread does.
ta0kira