Path: utzoo!utgpu!watmath!clyde!att!rutgers!mailrus!cornell!rochester!pt.cs.cmu.edu!*!postman+
From: pgc+@andrew.cmu.edu (Paul G. Crumley)
Newsgroups: comp.sys.ibm.pc.rt
Subject: Re: Why the original RT seemed/was slow (was ...)
Message-ID: <0XZPROy00jaRQ1D1Fy@andrew.cmu.edu>
Date: 1 Dec 88 21:19:22 GMT
References: <5046@polya.Stanford.EDU> <1287@auschs.UUCP> <1309@auschs.UUCP>,
	<3736@pt.cs.cmu.edu>
Organization: Carnegie Mellon
Lines: 100
In-Reply-To: <3736@pt.cs.cmu.edu>

Hello,

After reading so many interesting statements about the RT/PC family of
processors I feel I should add my few cents worth too!

I have been a user of RT/PC family processors from the beginning of the product.
 We had some RT/PC prototypes and have had one of everything here at the ITC
since.  I have used both the 4.2 and 4.3 derived UNIX(tm) and AIX.  I also have
experience with SUN and uVax equipment.

A couple of observations from previous complaints:

1) keyboard click:

This is controlled by the MACH device drivers.  If you don't like the way the
click works tell the MACH people.

2) bad code generation under HC:

I have used a number of versions of HC.  Even at this time, after many releases,
HC still does a miserable job of generating code.  I have watched the HC
compiler (with the -O switch set!)

- move arguments passed in registers to other other registers for no reason.
- store locals on the stack even when plenty of registers are available.
- fail to remove parts of the prolog and epilog not needed when a routine never
calls a function.
- recalulate values in loops.
- save more registers than needed upon entering a routine.
- you get the idea.....

None of these mistakes even takes into account the need to have a compiler for a
delayed fetch/store machine that understands the importance of ording operations
to keep the memory subsytem from causing waits.

In general, HC seems primitive compared to the compilers available under AIX.
My experience has been that the AIX compilers do a substantiallly better job of
code generation.  These compilers understand optimizations like careful register
allocation, eliminated parts of the prolog/epilog that are not needed for a
particular routine, ordering fetches and stores, inlining certain functions, etc.

I don't doubt that much of the performance of RT/PCs running AOS can be gobbled
up by HC's code generation.  I have found that I can routinely do better than HC
in optimizing the generated code.  This is not true for the PL/.8 based
compilers and the AIX compiler.

I'm sure IBM realizes how important good compiler technology is to RISC
processors and will continue to provide high-perforance compilers under the AIX
operating system.

3) lack of useful support and service people from IBM:

Perhaps our IBM representatives are superior to most IBM employees.  I have
gotten reliably good results from the sales and service people.

4) "state of the art"ness of the RT/PC:

In general, I have found the RT/PC to be a very good machine from all aspects
including cost/performance, reliability, repairability, keyboard layout, display
(the 6155 is too small but the Moniterm(tm) is fine), availability of many I/O
functions because of the use of an AT compatible bus, power, noise, ...

It could always be better but as a person that has to help support RT/PCs at the
hardware level I have found them very easy to deal with.

5) AIX:

Having used some of the early versions of AIX on old RT/PCs I must agree that
the original system was a real dog.  (wuff, wuff)  Newer versions of AIX on APC
hardware are quite respectable.  The AIX operating system also has a number of
facilities and features not available in the AOS software.

I have been able to port fairly complex software from AOS to AIX and back by
only changing the Make files and a couple #ifdefs.  (complex is software that
uses the IP network, accesses hardware directly, uses X, etc.)

Though some may consider the installation procedures for AIX to be a real pain,
I find the ability to install and reconfigure device drivers on a running
system, hardware error logging, support for vendor hardware, etc. in AIX to be
very helpful in installing and maintaining a system.

The VRM is a source of much debate.  I have not found it to get in the way very
often but know it can hurt system performance in some situations.


I personally believe that many of the performance problems experienced by users
of the AOS system are caused by inadequate compilers (though HC 2.0 is due any
day now...).

All things considered, I have found the RT/PC to be a very well designed
hardware base.  The AOS system is a very nice version of BSD flavor UNIX and AIX
is a more robust system that is getting more compabible with the BSD world with
each new release.



Best Regards,

Paul G. Crumley
ITC/CMU