All DatabasesMacTech Vol 10-1994
Books
Volume Number: 10
Issue Number: 10
Column Tag: Book Reviews
Scanlin on Books
By Mike Scanlin, Mountain View, CA
POWER and PowerPC
By Weiss and Smith
Morgan Kaufmann Publishers, Inc. 1994.
ISBN 1-55860-279-8.
408 pages (hardback).
If you are working on learning PowerPC assembly language, and you want a good,
solid technical information that gives you both a base to start with and a reference to
return to once you’ve started going, POWER and PowerPC delivers both. It is a very
complete look at the POWER and PowerPC architectures and contains information that
would be of interest to anyone who really wants to delve into PowerPC 601
programming.
The book covers three main areas: (1) the POWER architecture, (2) the first
two POWER implementations, the POWER1 and POWER2, used by IBM in the
RS/6000, and (3) the PowerPC architecture and PowerPC 601 implementation.
There are other areas which are interesting but which probably aren’t as important to
most Macintosh programmers, including: A comparison of the POWER and PowerPC
architectures, a comparison of the PowerPC 601 and DEC Alpha 21064, and the IEEE
754 Floating-Point standard.
The best part of this book, in my Mac-centric opinion, are Chapters 7-9, which
describe the 601 in detail. It starts off with a discussion of the instruction formats and
goes on to show how some of the less-obvious instructions work (like rotate with
mask). It then goes on to explain the 601’s pipelines, branch processing and caches.
Within each of these discussions the examples are clearly illustrated with sample code
fragments. You’ll be able to see where and why pipeline stalls occur (and what you can
do to avoid them in some cases), how to optimize your branches and exactly how the
combined instruction and data cache works. Understanding these issues is a key part of
being able to optimize for the 601 when you need to (in addition to helping you identify
why certain code fragments run slower than you would expect).
This book is not a tutorial on how to program in PowerPC assembly language. It
is, however, one of those rare technical books that is a pleasure to read for all the
right reasons: the examples are clear, the examples are worth studying, the authors
know their stuff and, it’s presented in a neatly typeset and illustrated manner. I would
recommend it to self-motivated people who want to start learning PowerPC assembly
language programming or to anyone working in a high-level language who wants to
know more about their underlying processor.
Zen of Code Optimization
By Michael Abrash
The Coriolis Group, Inc. 1994.
ISBN 1-883577-03-9.
449 pages (soft cover, w/disk).
The guru who so many years ago brought us The Zen of Assembly Language has
returned. He has now released a new and improved version of the ideas and examples
contained in that sacred volume. And he has added new tricks for the latest Intel
processors, the 486 and the Pentium.
Now, you are probably asking yourself “Why would I care about a bunch of
optimization tricks on Intel processors when I’m a Mac fanatic?” The answer is
because there is something for everyone in this book. Even if you ignore all of the Intel
assembly code he presents (there is no 68K code at all) you can’t help but be
impressed with the methodology he used to determine the optimal instructions as well
as the clear (and often times humorous) explanations of the finer points of assembly
language programming. Mr. Abrash has a Zen-like understanding of the Intel
processors and the environments they live in. He is also a gifted writer who can make
an otherwise dry topic come to life. Even though I have personally vowed never to
write Intel assembly code, I thoroughly enjoyed both his former Zen book and this
latest one.
The author sums up the book’s essence rather well, “This book is the diary of a
personal passion, my quest for ways to write the fastest possible software for
IBM-compatible computers in C, C++, and assembly language. ... it is a journal of my
exploration of the flexible mind in action (with, to be sure, a generous leavening of
potent low-level optimization tricks).” This book is the summary of years of effort
studying the subtle behavior of Intel processors. And most of it is presented in
easy-to-read, story-like prose that is both fun to read and very educational.
The book starts off by giving us the Zen Timer; a little piece of code used
throughout that gives you the most precise timings possible of your Intel code
fragments. After all, you have to be able to measure your code accurately to know if
your latest change really improved things or not.
The next couple of chapters teach you various low-level things you need to know
to really optimize for the Intel processors, such as: the prefetch queue cycle-eater,
dynamic RAM refresh cycle-eater, and the display adapter cycle-eater. The interaction
of these cycle-eaters leads to some surprising results (like you can’t trust the
instruction times in the Intel manuals).
Once the basics are understood (and the reader has accepted “assume nothing;
time everything”) the book proceeds to apply that knowledge to some real-world
problems. In particular, the Boyer-Moore string searching algorithm is studied and
optimized. There are many examples of peephole optimizations (like fast
multiplication by 5 or 9 with the LEA instruction). There are examples given on how
to manipulate common data structures efficiently, such as linked lists.
The last few chapters are devoted to the Pentium. In addition to showing you how
many of the 386 and 486 tricks (taught in earlier chapters) will break on the
Pentium, there is an in-depth discussion of the Pentium and it’s U-pipe and V-pipe
and how to keep them both full most of the time. Sadly, like the 68K family, it is not
possible to simultaneously optimize for all members of the Intel family.
As an optimizing assembly language programmer, I found it refreshing to find
someone who is both a true master at assembly language programming and at the same
time capable of making all the right trade-offs when coding in mixed C and assembly. I
would recommend Zen of Code Optimization as a 10 on a scale of 10 if you are working
on any Intel processor and I would give it an 8 out of 10 for anyone working on the
Macintosh who is interested in writing high performance code.
JPEG: Still Image Data Compression Standard
By Pennebaker and Mitchell
Van Nostrand Reinhold. 1993.
ISBN 0-442-01272-1.
638 pages (hardback).
Ever wonder how those color painting programs manage to store 10MB of pixels
in a 1MB file? Well, most of them use JPEG (Joint Photographic Experts Group)
image compression. If you’ve ever wanted to know how it works, or implement it
yourself, then JPEG: Still Image Data Compression Standard is the book for you.
Written by two members of the JPEG standard committee, this book gives many
of the hows and whys of JPEG that are not in the official JPEG specification (which is
given in an appendix). There is a good chance that this book will tell you more than you
really want to know about JPEG. It contains a LOT of information.
The book is written for everyone, from non-technical people to programmers to
mathematicians. But don’t let the inclusion of some non-technical info dissuade you -
there is more technical and mathematical info here than you probably care to read.
Each section is marked with one of three “technical difficulty” symbols: one for
non-technical readers, one for people with intermediate technical skills and one for
people with advanced technical skills who are either going to implement a JPEG engine
or else just like hard math problems.
The beginning of the book goes over some basic imaging concepts for the
uninitiated (such as low-pass filters and the difference between luminance and
chrominance). It then introduces you to the Discrete Cosine Transform (DCT) that is
the heart of JPEG. This discussion is very complete and certainly makes it clear how
both one dimensional and two dimensional DCTs work (with good illustrations and
examples). It also explains some of the ‘blocking’ effects you sometimes see with JPEG
images.
After the DCT, the book goes on to explain the various JPEG modes of operation
(sequential, progressive, lossless and hierarchical) and the syntax of the JPEG data
stream. If you’ve read the spec and not been clear on any of those things then this
book’s discussion of them will clear them up for you (it certainly did for me).
Once you’ve run your image data through the DCT and quantized it, the last step of
JPEG is to entropy encode the quantized values. JPEG allows two methods of entropy
encoding: Huffman or arithmetic. The book spends ample time on both methods (several
chapters, in fact, including one on probability estimation) and, depending on how much
you like math, you’ll come away either really confused or really understanding how it
works. (The explanations are clear, but it’s difficult material.)
The last part of the book gives comparisons of performance for the different
kinds of JPEG compression, a list of JPEG applications and vendors, a history of JPEG,
possible future directions of JPEG and a discussion of non-JPEG compression standards
(JBIG, MPEG, fax).
This book is as complete as you could possibly want on the subject of JPEG image
compression. It is a must-read for JPEG implementors and recommended reading for
those people who like to understand how common algorithms work or who want to know
more about imaging algorithms in general.
Referenced by (60):