Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Posting-Version: version B 2.10.2 9/18/84; site watdcsu.UUCP
Path: utzoo!watmath!watnot!watdcsu!herbie
From: herbie@watdcsu.UUCP (Herb Chong [DCS])
Newsgroups: net.works
Subject: Re: Not again!?  Assembler vs High-Level languages
Message-ID: <1228@watdcsu.UUCP>
Date: Thu, 11-Apr-85 11:29:06 EST
Article-I.D.: watdcsu.1228
Posted: Thu Apr 11 11:29:06 1985
Date-Received: Fri, 12-Apr-85 00:01:55 EST
References: <483@terak.UUCP> <2251@wateng.UUCP> <492@terak.UUCP> <5461@utzoo.UUCP>
Reply-To: herbie@watdcsu.UUCP (Herb Chong [DCS])
Organization: U of Waterloo
Lines: 54
Summary: 

In article <5461@utzoo.UUCP> laura@utzoo.UUCP (Laura Creighton) writes:
>I have mostly used assembler, not when I needed something *fast* as
>when I needed something *real small*. But there it is easy -- write the
>code in C, produce the assembler and then tighten that. (but not on the
>pdp-8, though. I don't think there *is* a C compiler for the pdp-8)
>
>Laura Creighton
>utzoo!laura

on some systems other than unix, assembler is used because of the
special restrictions of the design of the system.  in VM/CMS, many
system services are awkward to get at directly from a high-level
language (poor design) but are very useful.  other system services are
designed to not be accessible from a high level language.  CMS itself
occupies a part of the user's address space but is protected against
write by ordinary programs.  a special area (called the transient area)
is reserved for commands that are not part of the kernel but are often
used.  it's only 8k in size, but there are many compelling reasons to
write commands that use that area.

i wrote a version of the Boyer-Moore string matching algorithm to fit
in 8K, and it contains a shell sort, error handler, and interrupt
handler.  the most important reasons for making it transient was to
make it available to be called by any high-level language program and
to use system services not available otherwise.  because CMS is in the
same address space as a user program, it is possible to use kernel code
by branching directly to it instead of doing a system call (SVC)
provided certain conditons are met.  i do i/o by branching directly to
the i/o routines in the kernel.  execution time is reduced by 1/3
because i bypass the system call interface.  this is a well documented,
but seldom used, feature of CMS because of the restrictions on write
access to the kernel.  it is possible (using our Waterloo C compiler)
to write a C program that fits in the transient area, but the minimal
library occupies about 4K, leaving 4K for code and variables.

using assembler partly depends also upon the coding style used.  everyone
writes it their own way unless compelled to do otherwise.  writing code
that is meant to be modified by others requires discipline and consistency.
at waterloo, we modify and maintain our own version of VM/SP3 that is
sufficiently different from others that it takes getting used to. it's
all written in /370 assembler.  maintaining it is no more trouble, in
my opinion, than 4.2bsd code, not because /370 assembler is so good, but
because badly written C code (as most of 4.2bsd is) is worse than well
written assembler.

miscellaneous ramblings from the keyboard of
Herb Chong...

I'm user-friendly -- I don't byte, I nybble....

UUCP:  {decvax|utzoo|ihnp4|allegra|clyde}!watmath!water!watdcsu!herbie
CSNET: herbie%watdcsu@waterloo.csnet
ARPA:  herbie%watdcsu%waterloo.csnet@csnet-relay.arpa
NETNORTH, BITNET, EARN: herbie@watdcs, herbie@watdcsu