⭐ 欢迎来到虫虫下载站! | 📦 资源下载 📁 资源专辑 ℹ️ 关于我们
⭐ 虫虫下载站

📄 http:^^www.cs.wisc.edu^~cs354-2^cs354^lec.notes^case.studies.html

📁 This data set contains WWW-pages collected from computer science departments of various universities
💻 HTML
📖 第 1 页 / 共 2 页
字号:
Date: Tue, 05 Nov 1996 00:32:49 GMTServer: NCSA/1.5Content-type: text/htmlLast-modified: Wed, 30 Aug 1995 21:21:35 GMTContent-length: 20185<html><head><title> Lecture notes - Chapter 14 - Architectures</title></head><BODY><h1> Chapter 14 -- Architectures</h1><pre>PERSPECTIVE ON ARCHITECTURE DESIGN----------------------------------factors in computer design:   speed       as fast as possible, of course       dependent on technology and cost   cost/price       profit, non-profit, mass market, single use   useablility       shared/single user, size of machine, OS/software issues,       power requirements       depends on intended use!   intended market       mass market, scientific research, home use, multiple users,       instructional, application specific   technologyprice/performance curve       ^  | perf. |  |         x          |           x          Want to be to the "left"  of these,          |      x               to have higher performance for the          |                      price.  "More bang for the buck."          |    x   x          _________________		    		    price ->technology -- a perspective----------    electromechanical (1930) -- used mechanical relays    vacuum tubes (1945)      space requirement:  room          Manchester Mark 1 (late 1940s)    transistors      discrete (late 1950s)         space requirement:  a large cabinet to a room	 Examples:	    CDC 6600	    B 5000	    Atlas (?)	    PDP 11/10      SSI,    MSI (mid-late 1960s)      1-10   10-100 transistors         space requirement:  a cabinet	 Examples:	   Cray 1	   VAX 11/780      LSI   (early 1970s)      100-10,000 transistors         space requirement:  a board	 Examples:      VLSI  (late 1970s - today)      >10,000 transistors         space requirement:  a chip, or chip set, board	 Examples:	    MIPS R2000	    Intel 386 (~275,000 transistors)	    SparcRISC vs. CISC------------- RISC - Reduced Instruction Set Computer   The term was first used to name a research architecture at   Berkeley:  the RISC microprocessor.  It has come to (loosely) mean a   single chip processor that has the following qualities:     1. load/store architecture     2. very few addressing modes     3. simple instructions     4. pipelined implementation     5. small instruction set -- easily decoded instructions     6. fixed-size instructions CISC - Complex Instruction Set Computer   This term was coined to distinguish computers that were not RISC.   It generally is applied to computers that have the following   qualities:     1. complex instructions     2. large instruction set     3. many addressing modesdifficulties with these terms - not precisely defined - term introduced/applied to earlier machines - "RISC" has become a marketing toolsingle chip constraint----------------------As technologies advanced, it became possible to put a processor on a single VLSI chip.  Designs became driven by how much (how many transistors) could go on the 1 chip. Why?  1.  The time it takes for an electrical signal to cross a  chip are significantly less than the time for the signal to  get driven off the chip to somewhere else.       2.  The number of pins available was limited.  So, the desire is to have as little interaction of the chip   with the outside world as possible.  It cannot be eliminated,   but it can be minimized.The earliest of single processors on a chip had to carefullypick and choose what went on the chip.  Cutting-edge designstoday can fit everything but main memory on the chip.how the world has changed-------------------------  earliest computers had their greatest difficulties in getting  the hardware to work --    technology difficulties        space requirements       cooling requirements  given a working computer, scientists would jump through whatever    hoops necessary to use it.  as hardware has gotten (much) faster and cheaper, attention has been    diverted to software.      OS      compilers      optimizers      IPC (inter-process communication)1 instruction at a time isn't enough.  The technologyisn't "keeping up."  So, do more than one instructionat a time:parallelism-----------  instruction level (ILP) -- pipelining  superscalar -- more than one instruction at a time  multis  VLIW  supercomputer  WHICH OF THESE IS "BEST" and "FASTEST" DEPENDS ON WHAT PROGRAM  IS BEING RUN -- THE INTENDED USEAGE.on the 68000 Family--------------------- released in the late 1970's- an early "processor on a chip"- a lot of its limitations have to do with what could fit on a VLSI  chip in the late 1970'sINSTRUCTIONS - a relatively simple set (like the MIPS) but NOT a load/store arch. - a two-address architecture - most instructions are specified in 16 bits -- fixed size. - tight encoding, it is difficult to distinguish opcode from operands,   but the m.s. 4 bits are always part of the opcode. integer arithmetic   different opcode for varying size data   (add.b    add.w      add.l) logical   different opcode for varying size data control instructions   conditional branches, jumps   (condition code mechanism used --  where most instructions    had the effect of setting the condition codes) procedure mechanisms   call and return instructions floating point ??   (I guess not!) decimal string   arithmetic presuming representation of binary coded decimalREGISTERS  16 32-bit general purpose registers,  only one is not general purpose (it is a stack pointer)  the PC is not part of the general purpose registers  the registers are divided up into two register files of 8,    one is called the D (data) registers, and the other    is called the A (address) registers.  This is a distinction    similar to the CRAY 1.  A7 is the stack pointer.DATA TYPES byte word (16 bits) longword (32 bits) addresses are really their own data type.   arithmetic on A registers is 32 bit arithmetic.  However,   pin limitations on the VLSI chip required a reduced   size of address.  Addresses that travel on/off chip are   24 bits -- and the memory is byte-addressable.  So   a 24-bit address specifies one of 16Mbyte memory locations.each instruction operates on a fixed data typeOPERAND ACCESS the number of operands for each individual instructions is fixed like the VAX, the addressing mode of an operand does not depend on the instruction.  To simplify things, one of the operands (of a 2 operand instruction) must usually come from the registers. the number/type of addressing modes is much larger than the MIPS, but fewer than the VAX. the text has a detailed discussion of the 68000 addressing modes. READ IT!PERFORMANCE  ?, they got faster as new technologies got faster.SIZE  1 64-pin VLSI chip (a huge number of pins at that time)the Intel iAPX 86-----------------the 8086 - late 1970's release - a one- or two- address architecture (depending on how you look   at it) - part of a chip set   pin limitations on the chips made for some unusual architectural   design decisions. - compatible with earlier 8085 chip set a memory-to-memory architecture, sort of 3 16-bit registers, plus an accumulator, a stack pointer, and a PC   (plus 4 more that deal with segments) condition codes are set by most instructions,and are used   in branching an unusual addressing scheme: (due to a 16-bit limitation for pins used for addresses)   divide all of memory into fixed-size, 64K byte pieces   call each piece a segment.   addresses are 16 bits, and specify an offset within one of   the segments   4 extra registers hold segment base addresses, and an   effective address is computed (sort of) by    1. specifying which segment register    2. adding the 16-bit address to the contents of the       segment register  this addressing scheme "cramps a programmer's style".  code has to fit into a segment -- so does data, and  so does a stack.all about the Cray 1--------------------  There has always been a drive to design the best, fastest computer in  the world.  Whatever computer is the fastest has generally been called  a supercomputer.  The Cray 1 earned this honor, and was the fastest for a relatively long  period of time.  The man who designed the machine,  Semour Cray, is a bit of an eccentric,  but he can get away with it because he's so good.  The Cray 1 has an  exceptionally "clean" design, and that makes it fast.  (This is probably  a bit exaggerated due to my bias -- the Cray 1 is probably my favorite  computer.)  Mostly my opinion:  To make the circuitry as fast as possible, the Cray 1 took 2 paths    1.  Physical -- a relatively "time-tested" technology was used, but	much attention was paid to making circuits physically close	(Semour was aware of the limits imposed by the speed of light.)

⌨️ 快捷键说明

复制代码 Ctrl + C
搜索代码 Ctrl + F
全屏模式 F11
切换主题 Ctrl + Shift + D
显示快捷键 ?
增大字号 Ctrl + =
减小字号 Ctrl + -