Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Path: utzoo!mnetor!uunet!seismo!sundc!pitstop!sun!amdcad!tim
From: tim@amdcad.AMD.COM (Tim Olson)
Newsgroups: comp.arch
Subject: Re: register windows
Message-ID: <18855@amdcad.AMD.COM>
Date: Sun, 25-Oct-87 20:27:02 EST
Article-I.D.: amdcad.18855
Posted: Sun Oct 25 20:27:02 1987
Date-Received: Tue, 27-Oct-87 05:53:13 EST
References: <201@PT.CS.CMU.EDU> <933@cpocd2.UUCP> <821@mips.UUCP> <18843@amdcad.AMD.COM> <833@mips.UUCP>
Reply-To: tim@amdcad.UUCP (Tim Olson)
Organization: Advanced Micro Devices
Lines: 52
Keywords: register windows, interrupt latency

In article <833@mips.UUCP> hansen@mips.UUCP (Craig Hansen) writes:

| Loads/stores to variables through global pointer register: 15.2% (Because of
| use of the global pointer register, these references are single instructions,
| all of which have non-zero offset values.) I would suspect that the Am29000
| statistics do not count these as non-zero offset values, though they are
| more than half of all the references.


You are correct; these were not counted.  The statistics were gathered
by looking (at "runtime") for explicit address calculations of load
addresses.  I don't really consider the use of the global pointer
register to be an "non-zero offset" addressing mode, however.  Because
the lower 16 bits of the global pointer register are zero, the offset is
really just a concatenation -- no "add" is performed.

| Tim should also be careful, though, to realize that the tools and
| compilers used to measure the architecture also influences the results.

Certainly.

| Unless you set out to design both the architecture and the compiler
| concurrently, you won't be able to make good trade-offs between them.

THE compiler?  Which one?  C? Pascal? FORTRAN? Ada? Common LISP? Smalltalk?
Obviously you must design the architecture with a great deal of thought
put into how the resources are to be used by software and what can be
easily/cheaply performed in software vs hardware, but I disagree with
your statement.  MIPS certainly didn't write an entire Ada compiler
before starting architectural design, but that in no way decreases the
viability of Ada running on a MIPS machine.

| It should also be noted, again, that the selection of programs used as
| benchmarks will influence the results too. Making your architectural
| trade-offs on the basis of Dhrystone, or even nroff, which is not very
| representative of more modern C code, isn't very smart.

Making architectural decisions based on *any* single program isn't very
smart.  You should examine a large body of code, looking at older,
heavily-used programs as well as more "modern" code (output of C++
compilers, object-oriented programming).

Discounting nroff is kind of silly, however; it happens to be a large,
heavily-used utility.  MIPS used it, along with other significant
UNIX utilities, in early papers on the R2000 [Rowen, et al, VLSI Systems
Design, March 1986], and continues to reference it in John Mashey's
performance briefs.

	-- Tim Olson
	Advanced Micro Devices
	(tim@amdcad.amd.com)