Newsgroups: comp.sys.transputer
From: Michael Haardt <(michael)u31b3hs@pool.informatik.rwth-aachen.de>
Subject: [Summary] Linux and transputers
Keywords: linux, b004, device driver, assembler
Organization: An old and gray machine, somewhere in Moria.
Date: Tue,  2 Nov 93 23:45:01 MET

Quite some people seem to be interested in it, and so I decided to post
something about the current state of Linux and transputers:

There is a kernel space link device driver available.  It already works
fine for B004 and compatibles.  The next version will be able to support
multiple cards and have a better data transfer rate, which of course
depends a lot on the speed of your CPU (in the PC).  This driver was
written by Christoph Niemann.  The old userspace driver will disappear
soon, it is only used for some diagnostic things today.

All three, iserver, afserver and cserver are available, using that kernel
link device driver.  I have not tested cserver, but iserver and afserver
work fine.  I ported iserver, Christoph Niemann ported afserver and cserver.

There is an assembler and linker available, written by me from scratch.
They work fine under Linux.  Currently only 32 bit transputers are
supported, that may change though.  A paper about this project is in
preparation.

I have heard of several tries to port early versions of GCC to
transputers.  No success is known to me.  I think it is not only a
question of porting a compiler by writing a new backend, because most
people want extensions to use parallelism.  _RUMOURS_ say, RMS said
"forget it" about the possibility of GCC for transputers.  I have no
source for that statement.

In general, I think those compilers which have a tree or DAG structured
representation of intermediate code are good candidates to get ported,
because swapping subtrees and optimizing on a slightly attributed
tree is the thing to do in order to create good code.  I ported Small-C
once to get some experience in code generation for transputers after I
learned assembler.  The result is as expected: easy port, poor code.
The peephole optimizer can optimize simple things in detecting very
small trees to evaluate different using pattern matching, but of course
fails on the general case.  I hope being able to write more on this
subject soon. :)  Don't hold your breath ...

I don't know about link transfer rates under DOS, but file system
accesses are definitively much faster with Linux.  It got reported that
transputer hosted applications work faster because of this.  No wonder,
because the server is the bottleneck in I/O.

Last but not least: You can get the above mentioned software at the
anynonymous ftp server ftp.thp.uni-koeln.de, pub/linux/transputer/*.
This archive will be frequently updated, because development speeds up
more and more at the moment.

Michael
--
Twiggs and root are a wonderful tree (tm) Twiggs & root 1992 :-)