CSA Transputer Education Kit - User Guide
1990
Computer System Architects
47 Pages
© Computer System Architects 1990.
WARNING: This equipment generates, uses and can radiate radio frequency energy, and if not installed and used in accordance with the instruction manual, may cause interference to radio communications. Operation of this equipment in a residential area is likely to cause interference. If the user desires to correct this interference, he must do so at his own expense. Modification of this product will require compliance with FCC Part 15 rules.
Computer System Architects reserves the right to make changes in specifications at any time and without notice. Information contained in this manual is derived with permission from published Inmos documents and CSA engineering. The information furnished by Computer System Architects is believed to be accurate; however, no responsibility is assumed for its use, nor for any infringements of patents or other rights of third parties resulting from its use. No license is granted under any patents, trademarks, or other rights of Computer System Architects or of the Inmos group of companies.
Computer System Architects and the CSA logo are trademarks of Computer System Architects, Inc. Inmos and IMS are trademarks of the Inmos Group of Companies. IBM is a registered trademark of International Business Machines Corporation. Apple II is a registered trademark of Apple Computer, Inc. Macintosh is a trademark licensed to Apple Computer, Inc.
Software written by CSA and included with this package is provided for assurance testing and function demonstration. This software is provided by CSA under license. Use of this software on non-CSA products will not be supported. This software is © Copyright 1988, 1989, 1990 Computer System Architects, and is marked as such.
Other software contained in the demonstration programs supplied with this package is provided by Inmos Limited free of charge and without any support. This software is marked as such, and is Copyright Inmos Limited. Except for the liability arising from the due course of law, Inmos accepts no liability whatsoever with respect to these programs.
Logical Systems C - Notice of License: A paid single machine software license is included for the C software provided in this package by Logical Systems, Corvallis, Oregon, USA. This license is printed on the software package.
Occam Toolset - Notice of License: A paid single machine software license is included for the Occam software provided in this package by the Inmos division of SGS-Thomson. This license is printed on the software package.
Inmos is a member of the SGS-Thomson Microelectronics group.
Contents
1 Introduction 1.1 Forward 1.2 Brief Theory of Operation 2 Installation Guide 2.1 Getting Started 2.2 Hardware Installation 2.2.1 Anti-static Precautions 2.2.2 Board Installation 2.2.3 Running Hardware Checks 2.2.4 Mandelbrot Demonstration 2.2.5 Additional Memory 2.3 Software Options and Memory 2.4 Software Installation 2.4.1 Installing the C Toolset 2.4.2 Installing the Occam Toolset 2.5 Your First Transputer Program 2.5.1 A C Program in On-Chip Memory 2.5.2 A C Program 2.5.3 An Assembly Program 2.5.4 An Occam Program 2.6 The Next Step 3 Installation Options 3.1 Board Layout 3.1.1 List of Components 3.1.2 Optional Components 3.1.3 Jumper Locations 3.2 Adding Memory 3.2.1 Memory Speed 3.2.2 Memory Size 3.2.3 Memory Tests 3.3 PC/Link Interface Settings 3.3.1 PC/Link Interface 3.3.2 PC/Link Speed 3.4 Transputer Settings 3.4.1 Transputer Speed 3.4.2 Link Speeds 3.5 PC-Xptr Jumpers 3.6 Setting Up Multi-Transputer Networks 3.6.1 Chain Jumper Settings 3.6.2 Tree Jumper Settings 3.6.3 Mesh Jumper Settings 4 External Interface 4.1 Description 4.1.1 LED Port 4.1.2 Programmed 1/0 Port 4.1.3 Power Transistors 4.2 Examples 4.2.1 Printer Interface 4.2.2 Eight-Channel A to D Converter 4.2.3 Two-Channel D to A Converter 4.2.4 Two-digit LED Display 5 Troubleshooting 5.1 Simple Solutions 5.1.1 Installing the Board 5.1.2 Installing the Memory 5.1.3 Software 5.1.4 Miscellaneous 5.2 Step by Step Diagnostics 5.2.1 PC/Link Interface 5.2.2 Transputer 5.2.3 Memory 5.2.4 Software 5.2.5 Add-On-Processor Boards 5.2.6 Mufti-Board Networks 6 C In On-Chip Memory 6.1 The Basics 6.2 The Lt I/O Library 6.2.1 Output Using the Lt I/O Library 6.2.2 Input Using the Lt I/O Library 6.2.3 Lt I/O Heap Management 6.2.4 Lt I/O Library Reference 6.3 Optimizing Memory Use 6.4 Multi-processing In On-Chip Memory
Appendix
A.1 Memory Map A.2 Memory Mapped 1/O interface Addresses A.2.1 PC Link A.2.2 PC Link SS (SubSystem) System Services A.2.3 Transputer SS (SubSystem) System Services A.2.4 External Interface LED Port and Power Transistors., A.2.5 External Interface Programmed I/O Port A.3 Board Schematics A.4 External Cable Schematics A.4.1 Cables between Transputer Education Kit Boards... A.4.2 Education Kit to CSA PART Series Internal A.4.3 Education Kit to CSA PART Series External A.4.4 Education Kit to Inmos Boards Index
Forward
"What we are witnessing today is the last hurrah of serial processing."
Joel Birnbaum
Hewlett-Packard
Congratulations. With your purchase of a Transputer Educational Kit from Computer System Architects, you have invested in the most promising computer technology available. This exciting technology is multiprocessing. Transputers are revolutionary parallel microprocessors that make multiprocessing viable and affordable today.
I frequently interface with scientists and engineers throughout North America as a marketing manager for SGS-Thompson-Inmos, the company that manufactures the transputer and teamed with CSA to build your kit. Many of these technical people are achieving scientific breakthroughs and are building futuristic applications using the same development tools you now own.
The importance of what this kit will teach you cannot be overstated. The powerful concepts and tools are unique, oftentimes contrary to mainstream methodologies promoted by the technical and academic communities.
When transputer technology was being born in the early 1980s, it was championed by visionary thinkers and imaginative programmers looking to extend the boundaries of computer science. But conventionalists regarded the transputer as a whimsical idea that would amount to nothing more than an interesting toy. One eminent scientist even postulated a law of diminishing performance to disprove benchmarks obtained by large transputer systems.
Today the transputer is broadly regarded as an innovative machine at the forefront of computer technology. It delivers unmatched performance for many applications and ranked fourth in world sales of 32-bit microprocessors last year. Major Fortune 500 companies and large military contractors are using transputers to build next generation systems, and transputer applications are playing a role in America's space program. Let me briefly examine this dramatic turn-around and relate it to what you will be learning.
We live in an age of tremendous discovery. If you are in your forties, half of the world's knowledge has been produced since you left school. This explosive growth of science is being made possible by computers that extend our ability to manage information and solve problems which stagger the human mind.
Applications for computers have increased far beyond the basic "number crunching" they were invented for. Besides calculating mathematical expressions, computers also process, analyze, control, synchronize, and even characterize data.
Some computer systems convert signal and image information into speech and vision. Graphics computers transform mass numbers, humanly indigestible in their raw form, into three-dimensional maps of the human body, subsurface geological features, protein molecules, and objects in outer space.
Other computers are used as smart, responsive, "real-time controllers" for focusing cameras, robot motion, anti-lock breaking, nuclear reactors, and weapon guidance. Perhaps most amazing are neurocomputers which implement neural network paradigms to produce machine learning or "artificial intelligence."
Scientific investigation and computer technology have become almost inseparable. Even though the performance of today's computers is impressive, the quest for knowledge continues to demand more computational power. Major problems on all frontiers of science are so complex they overwhelm even the most powerful supercomputers, capable of executing 1.5 billion calculations per second.
Such computers achieve their speed with advanced process technologies that reduce the size of chip components and enhance their conductive properties. Semiconductor devices holding two million components are now in production, but designers are running into limitations imposed by the laws of physics. These limitations are dictated by the wavelength of light used to etch circuit patterns on wafers, heat dissipated from current resistance in conductors, and leakage of electrons as insulation layers become increasingly thin.
Some exotic solutions are being explored to overcome these speed barriers. One is the creation of photonic circuits that use flashes of light, instead of electric current, to transmit digital information through. strands of glass called optical fibers. These fibers dissipate no heat and require no electrical insulation. The most radical concept for a photonic circuit is the "biochip," a three-dimensional device made of organic (carbon based) molecules. These tiny molecular switches could be packed together in far greater densities than semiconductor components.
These prospects are exciting but they will not become commercial realities for some time, and the price of process technology in terms of unit and system costs is already soaring. Supercomputers utilizing Gallium Arsenide or supercooled CMOS circuits require very fast memories and cost millions of dollars to own and operate.
Increasingly, multiprocessing is being adopted around the world as the sensible alternative. Instead of using process technology to achieve raw speed, problems are solved more quickly by using dozens or even thousands of parallel computers networked together. This strategy is not unlike brain cells, called neurons, which are tied together in massive networks that allow them to quickly recognize patterns like smells or process information for split second reasoning.
Nature has elected to process information concurrently because the universe is enormously parallel. Any system that can be viewed as a collection of simultaneous (parallel) parts or events is said to possess concurrency. Virtually all systems in astronomy, genetics, physics, geology, biology, and chemistry exhibit concurrency because natural phenomena rarely occur in a nice serial fashion - the way most programmers develop their code.
Computers used for speech regulation, neural networking, jet engine control, medical diagnosis, and airport scheduling all describe pieces of the real world. To accurately model nuclear reactions, global weather, biosystems, experimental drugs, formation of new stars, and air bag expansion in a car crash, scientists and engineers visually simulate the net effect of many simultaneous forces, events, and physical laws. Although some computers may be faster at performing rote calculations, the transputer is specifically designed to handle this complexity.
Conventional computers and programs are "sequential," meaning one instruction or task must be completed before the next can begin. Parts of a problem are always executed in serial order, even if they actually take place at the same time. When sequential programs are used to describe real phenomena, the impact of real-time interactions is lost and only simplistic models are possible.
Conversely, transputers are "parallel" machines. They achieve their speed by executing the composite parts of a problem together as they occur. A single transputer handles different parts by "juggling" rapidly between them, or a group of transputers simultaneously on parts distributed among them.
The method of programming is essentially the same for either approach. In the second case, transputers are combined as building blocks in multiprocessor networks to obtain nearly linear performance enhancement. In other words, a network of N transputers will typically execute code N times faster than a single transputer.
Multiprocessing is as much a conceptual advancement as it is a technological advancement. Although the transputer's architecture is being studied by computer manufacturers everywhere, the real trick to multiprocessing is learning how to write code that exploits the parallelism of problems.
The most far-reaching breakthrough in this area is the Occam programming language. Parallel programming is not trivial to master, but Occam provides a framework for developing programs which express parallelism explicitly, in an understandable way. Occam can also be proven mathematically, making it attractive for theoretical research and secure applications.
Because its architecture is designed to execute the Occam model of concurrency, the transputer is sometimes regarded as a hardware implementation of the language. This unprecedented level of integration between software and hardware is part of the reason behind the transputer's impressive performance. Other parallel programming languages exist, but most are written and supported only by PhDs and none have been transformed into an innovative computer architecture.
Occam is being successfully taught to undergraduate students around the world, and some universities have made it part of their required curriculum. National user groups have formed a worldwide community of students, faculty, and industry professionals who use Occam because it makes the description of concurrent systems straightforward and comprehensible.
Transputers, like ordinary microprocessors, can also be programmed in standard high-level languages. Your kit includes comprehensive tools and documentation for both Occam and C.
A company called Logical Systems has built a special "Parallel C" compiler that embeds Occam's powerful constructs within C code to ease the transition from serial to parallel programming. This useful compiler also makes the porting of existing C applications onto transputer systems possible. An increasing number of industry applications are combining Occam with more familiar languages like C in pragmatic ways.
CSA recommends using your single transputer board to become familiar with this kit's extensive software development tools. Look at the exercises in your workbook and try to start writing and debugging code. It is important to approach Occam with an open mind until its use feels natural. First time programmers will sometimes learn quicker than experienced programmers who have an ingrained sequential approach to writing code.
Additional PC add-in boards can be purchased from CSA to construct inexpensive multiprocessor systems for accelerating your code. These boards are easy to interconnect in arbitrary network configurations with plug-in link cables. These networks may be located inside one PC chassis or distributed among separate PCs in a parallel processing laboratory.
With four additional 1 Mbyte boards, you can build a formidable PC multiprocessor for under $1,500 which delivers 50 MILLION INSTRUCTIONS PER SECOND (MIPS). No other technology in the world places this kind of performance within reach of typical consumers. Most 25 MIPS workstations sell for $30,000 and up. A Hypercube machine is several orders of magnitude more expensive than a comparable transputer network, and is an engineering nightmare by comparison. Sizable transputer systems can be created in classrooms or teaching labs if each student purchases their own kit for the course. Some commercial systems in use today have several thousand transputers.
It is our hope that your state-of-the-art transputer kit becomes a source of valuable training and fun. You are now one of many pioneers exploring today's newest, most powerful, and fastest growing computer technology.
Mark Hopkins
Strategic Projects Manager
SGS-Thomson Microelectronics