CS61c: Introduction to Synchronous Digital SystemsJ. WawrzynekOctober 8, 2007Reading: P&H, Appendix B1 Instruction Set ArchitectureAmong the topics we studied thus far this semester, was the following big idea. Our high-level languageprograms (C, or C++ for instance) are converted to a list of assembly language instructions which interm are converted to a set of machine instructions.C Language constructs/data structures↓assembly language list of CPU instructions (with memory references)↓machine language list of binary coded machine instructions↓?How are these instructions carried out? What happens to the machine instructions? One thing wecan do is to read them into a simulator (such as MARS or SPIM). However, most of the time, we areinterested in actually running the instruction on real hardware. How does the hardware execute ourinstructions? That is the topic of the remainder of the semester.The set of opcodes that a processor can carry out, along with a description of the processor “state”(all the user visible registers), is called the Instruction Set Architecture (ISA).SoftwareInstruction Set ArchitectureHardwareIt is an abstraction of the hardware for the benefit of the software. More precisely it is a 2-waycontract between the hardware and the software.12 Wawrzynek 2007cUCBThe hardware promises that it will faithfully implement the ISA and the software promisesto only try to do things supported by the ISA.Software designers can write operating systems, compilers, and applications without fear that theirsoftware will not work on new versions of the hardware. Also, the software really doesn’t know (norshould care) about how the hardware implements the ISA. Therefore the hardware designer is free todesign and optimize the implementation of the processor as he/she sees fit. It can be tailored to the latesttechnology, or to some particular market need (a particular price or performance point for instance), aslong as the design meets the contract by faithfully implementing the ISA. As new, more advanced,implementations can come along, all our old programs and operating systems will still function.The ISA is an abstraction and doesn’t specify in detail how the instruction execution should beimplemented. Consequently their are many different ways to design the processor. We will study afew of the basic concepts behind processor design, leaving many of the performance enhancements forCS152.2 The Physical HardwareWe will start by looking at a photograph of a microprocessor. This one is from a few years ago, but stillis representative of today’s processors.CS61c Lecture Notes 3This microprocessor happens to be a powerPC (similar to what user to be inside Apple computers).However, most processors, even ones with different instructions set, for instance the MIPS, look verysimilar to this picture.Notice the various blocks labeled on the photograph. We will be discussing these later this semester.Here are the specifications for this particular processor:• “Superscalar (3 instructions/cycle)”• 6 execution units (2 integer and 1 double precision IEEE floating point)• 32 KByte Instructions and 32 KBytes Data L1 caches• Dual Memory Management Units (MMU) with Translation Lookaside Buffers (TLB)• External L2 Cache interface with integrated controller and cache tags, supports up to 1 MByteexternal L2 cacheThe picture gives you a good idea of the relative complexity of the various sub-blocks.In the world of integrated circuit (chip) design area ≡ cost. Bigger chips are much more expensiveto make than smaller chips. Chip real estate is a precious commodity. Designers work hard to makeeach function as physically small as possible without hurting overall performance.If we zoomed in with a microscope, each subpiece would look very similar to the others, becausethey are all made of the same basic stuff—wires and transistors. (In fact, we call these devices “inte-grated circuits” because they put together multiple transistors and the necessary wires to connect themtogether. Early chips (non-ICs) had a single transistor or two without the necessary wires. These daysstate-of-the-art ICs have transistor counts in the 10’s to 100’s of Millions). We are not going to go intothe details of how transistors operate. Suffice it to say that small collections of transistors are joined tomake simple functional blocks which in turn are joined to make bigger blocks, etc. We will stop ourinvestigations at the level of the simplest function units, logic gates and flip-flops. Below is an exampleof a simple unit. We will only consider the representation on the right, not the transistor-level details.EECS40 covers transistor-level implementations in detail.Surrounding the inner part of the chip—the core—is a set of connections to the outside world.Usually these connect through some wires in the plastic or ceramic package to the printed circuit board(PCB). In the case of most computers this PCB would be the motherboard. Some of these connectionsgo to the main memory and the system bus. A fair number of the pins are used to connect to the powersupply. The power supply takes the 110 Volt AC from the wall socket (provided by PG&E) and convertsit to low voltage DC (usually in the range of around 1 to 5 volts, depending on the particular chip used).4 Wawrzynek 2007cUCBThe DC voltage is measured relative to ground potential (GND). Power connections to the chip fromthe power supply are of two types; GND, and DC Voltage (labeled “Vdd”).The energy provided by the power supply drives the operation of the processor. The energy is usedto move electric charge from place to place in the circuits and in the process is dissipated as heat. Thesedays a processor my use on the order of 10 watts—like a not-so-bright light bulb.3 Signals and WaveformsAnother special connection to the chip is the clock input signal. The clock signal is generated on themotherboard and sent to the chip where it is distributed throughout the processor on internal wires. Theclock signal is the heartbeat of the system. It controls the timing of the flow of electric charge (and thusinformation) through out the processor and off-chip to the memory and system bus. If we had a verysmall probe we could examine the signal on the clock wires, by looking at the voltage level on the wire.We would see a blur, because the clock signal is oscillating at a very high frequency (around 1 billioncycles/second or 1GHz,
View Full Document
Unlocking...