chapter 2 sections 2.1 – 2.10 based on slides from dr. iyad f. jafar instructions: language of the...

Click here to load reader

Upload: ariel-crosley

Post on 14-Dec-2015

217 views

Category:

Documents


0 download

TRANSCRIPT

  • Slide 1

Chapter 2 Sections 2.1 2.10 Based on slides from Dr. Iyad F. Jafar Instructions: Language of the Computer Slide 2 Outline The Von Neumann Architecture Busses & Memory Program Execution The Computer Language Instruction Set Architecture The MIPS ISA MIPS Design Principles Fallacies and Pitfalls Further Reading More Examples 2 Slide 3 The Von Neumann Architecture 3 John Von Neumann,1945 A computer consists of Processor Memory Input and Output Stored-program concept as opposed to program- controlled computers ! Programs and data are loaded into memory prior to execution Different components are connected via set of shared wires called Busses for information exchange Compare to Harvard Architecture! Processor Memory Devices Datapath Control Input Output Slide 4 Busses In any computer, there is usually three types of busses Data Used to exchange data between components Bidirectional Typical size 8, 16, 32, and 64 bits Address Used to specify the source/destination Unidirectional The size of the bus specifies the number of entities (memory locations,I/O) that can be addressed Control Used to specify the operation requested by the CPU (READ, WRITE, and other handshaking operations) 4 Slide 5 0 Memory 5 B n-1 B1B1 0 B0B0 B1B1 1 B0B0 B1B1 2 B0B0 B1B1 M-1 B0B0 n bits (Memory Width) Address (Log 2 M bits) Control Data (n bits) Slide 6 Program Execution Program = Instructions + Data Programs are loaded into memory prior to execution In order to execute the instructions, the CPU Fetches the instruction Read the instruction from memory (Control Unit) Decodes the instruction Understand what should be done (Control Unit) Executes the instruction Perform the required task (Datapath) This is done repeatedly until the end of program ! 6 Fetch Decode Execute Slide 7 The Computer Language Commanding computers requires speaking their language; Binary electric signals at the hardware level (Machine Language) Tedious, time consuming, error-prone, knowledge about processor architecture Make it easier! Represent binary instruction by words (Assembly Language). Convert words to binary manually or using Assemblers Tedious, time consuming, error-prone, knowledge about processor architecture One line corresponds to one instruction!! Make it even easier! Make expressing instructions closer to the way humans express operations (High- level Languages ) Conversion to machine language is done automatically using Compilers 7 Suppose we want to perform two operations 1)Add two numbers X and Y and store the result in Z 2)Assign element 5 from array ARR to W 00001 0011 0001 0010 00101 0010 0000 0111 ADD Z, X, Y READW, ARR[5] Z = X + Y W = ARR[5] Slide 8 Instruction Set Architecture Designing processors requires specifying the type and number of instructions the processor is supposed to support and how a programmer can use it This is specified by the instruction set architecture (ISA) of the processor which includes Instructions (type, encoding, operation ) Memory and I/O access Registers (Why do we need registers !) Different processors have different ISA IA-32, MIPS, SPARC, ARM, DEC, HP, IBM, The same ISA can be implemented in different ways to achieve different goals Single-cycle, multi-cycle, pipelining, 8 Slide 9 Instruction Set Architecture Generally, the design of ISA could follow one of two schools CISC - Complex Instruction Set Computers RISC - Reduced Instruction Set Computers (~1980s ) 9 CISCRISC Many instructions and addressing modes Few instructions and addressing modes Instructions have different levels of complexity (different size and execution time) Simple instructions of fixed size Shorter programsLonger programs Relatively slowRelatively fast Expensive !!! (not really)Cheaper !! (not really) Intel, AMD, Cyrix MIPS, Sun SPARC, HP PA-RISC, IBM PowerPC Slide 10 The MIPS ISA Historical Facts Microprocessor without Interlocked Pipeline Stages First MIPS processor by John Hennessey in 1981 at Stanford University MIPS Technologies, Inc. 1984 First commercial model R2000 1985 General Features RISC 32-bit and 64-bit Register-Register Architecture ( Load-Store ) ISA revisions MIPS I, MIPS II, MIPS III, MIPS IV, MIPS V, MIPS32, and MIPS64 Different processor models R2000, R3000, R4000, R10000 10 Slide 11 The MIPS ISA Register File 11 0 0 1 2 B 31 B1B1 31 B0B0 32 bits B 31 B1B1 B0B0 B1B1 B0B0 B1B1 B0B0 Read Addr 1 Read Addr 2 Write Addr Write Data Read/Write Read Data 1 Read Data 2 Inside the CPU SRAM !! Hold data temporarily Some sort of caching of data Why two read ports ? 5 5 32 5 Slide 12 #NamePurpose $0$zeroConstant zero $1$atReserved for assembler $2$v0Function return value $3$v1 $4$a0Function parameter $5$a1 $6$a2 $7$a3 $8$t0Temporary Caller-saved $9$t1 $10$t2 $11$t3 $12$t4 $13$t5 $14$t6 $15$t7 #NamePurpose $16$s0Temporary Callee-saved $17$s1 $18$s2 $19$s3 $20$s4 $21$s5 $22$s6 $23$s7 $24$t8Temporary Caller-saved $25$t9 $26$k0Reserved for OS $27$k1 $28$gpGlobal pointer $29$spStack pointer $30$fpFrame pointer $31$raFunction return address The MIPS ISA Register File When writing assembly, these registers can be referenced by their address (number) or name General purpose and special purpose registers 12 Slide 13 The MIPS ISA Register File There are other registers ! Not directly accessible (no address) PC: Program counter Instruction sequencing LO and HI Used with multiplication and division instructions 13 PC LO HI 32 bits Slide 14 The MIPS ISA - Instructions Different categories Arithmetic, logical, memory, flow control All instructions are encoded in binary using 32 bits (Fixed Size!!!) Three different formats depending on the number and type of operands, and operation R-type I-type J-type Instruction encoding Opcode operation code that specifies the operation in binary Operands the ins and outs of the instruction 14 Slide 15 Arithmetic instructions The MIPS ISA - Instructions 15 A = B + C F = C - A ADDA, B, C SUBF, C, A ADD$s0, $s1, $s2 #$s0 = $s1 + $s2 SUB$t2, $s6, $s4 #$t2 = $s6 - $s4 OperationDestination, Source1, Source2 destination source1 op source2 Each arithmetic instruction performs one operation Each instruction has three operands The operands are in the file registers of the datapath The order of operands is fixed Slide 16 Arithmetic instructions The MIPS ISA - Instructions 16 Example 1. Given the following piece of C code, find the equivalent assembly code using the basic MIPS ISA given so far. a = b c d = 3 * a Solution: assume that the variables a, b, c, and d are associated with registers $s0, $s1, $s2, and $s3, respectively, by the compiler. sub $s0, $s1, $s2 # $s0 contains b c add $s3, $s0,$s0 # $s3 contains 2*a add $s3, $s3, $s0 # $s3 contains 3*a Slide 17 Arithmetic instructions The MIPS ISA - Instructions 17 Example 2. Given the following piece of C code, find the equivalent assembly code using the basic MIPS ISA given so far. f = (g + h) (i + j) Solution: assume that the variables f, g, h, i, and j are associated with registers $s0, $s1, $s2, $s3, and $s4, respectively, by the compiler. add $t0, $s1, $s2 # $t0 contains g + h add $t1, $s3,$s4 # $t1 contains i + j sub $s0, $t0, $t1 # $s0 contains (g+h) (i+j) More efficient translation!??? Slide 18 The MIPS ISA - Instructions Arithmetic Instructions Machine Language Any instruction has to be encoded using 32 bits including the operation and its operands 18 ADD $t0, $s1, $s4 oprsrtrd shamt funct 65555 6 op 6-bitsopcode that specifies the operation rs 5-bitsregister file address of the first source operand rt 5-bitsregister file address of the second source operand rd 5-bitsregister file address of the results destination shamt 5-bitsshift amount (for shift instructions) funct 6-bitsfunction code augmenting the opcode R-type Slide 19 The MIPS ISA - Instructions Arithmetic Instructions Machine Language Example 3. What is the machine code for ADD $s0, $t0, $s4 19 ADD $s0, $t0, $s4 00000001000101001000000000100000 s 65555 6 0000 0001 0001 0100 1000 0000 0010 0000 B 0 1 1 4 8 0 2 0 H Slide 20 The MIPS ISA - Instructions Arithmetic Instructions Machine Language Example 4. What is the machine code for SUB $s1, $s2, $s3 20 SUB $s1, $s2, $s3 00000010010100111000100000100010 s 65555 6 0000 0010 0101 0011 1000 1000 0010 0010 B 0 2 5 3 8 8 2 2 H Slide 21 The MIPS ISA - Instructions Logical Instructions AND, OR, NOR Operation is performed between corresponding bits (bitwise) R-type instruction format Example 5. What is the machine code for AND $s1, $t2, $s2 op = 0x0 rs = $t2 = 0x0A rt = $s2= 0x12 rd = $s1 = 0x11 shamt = 0x00 funct = 0x24 21 AND $s4, $s3, $t4 # $s4 = $s3 & $t4 OR $s0, $t7, $s3 # $s0 = $t7 | $s3 NOR $s2, $t6, $s3 # $s2 = ~($t6 | $s3) oprsrtrdshamtfunc 00000001010100101000100000100100 Slide 22 The MIPS ISA - Instructions 22 Arithmetic and Logic Instructions with Immediate In many occasions, we encounter high-level statements such as X = Y + 5 Z = W 4 F = 514 x R If (C>-3) This implies that the operation is between a register and some constant ! Can we do this with ADD and SUB instructions? Where do these constants come from? MIPS ISA specifies a new set of instructions that deal with immediate data ADDI, ORI, ANDI, XORI, Slide 23 The MIPS ISA - Instructions 23 Arithmetic and Logic Instructions with Immediate ADDI $t5, $s3, 130 ADDIU$s1,$s2, 15 oprsrtImmediate 65516 I-type Instruction Register File Sign Extension + 16 32 Reg[rs] Immediate Addressing! Slide 24 The MIPS ISA - Instructions 24 Arithmetic and Logic Instructions with Immediate ANDI $t1, $t3, 12 ORI$s3,$a0, 123 oprsrtImmediate 65516 I-type Instruction Register File Zero Extension & 16 32 Reg[rs] Immediate Addressing! Slide 25 The MIPS ISA - Instructions 25 Arithmetic and Logic Instructions with Immediate What if the immediate is greater than 16 bits? In MIPS, this is achieved in two steps using the Load Upper Immediate (LUI) and ORI instructions Suppose we want to add the constant (0000 0010 0100 0010 0001 0011 1100 1100) 2 to $s0 LUI$t0, 1000 1111 0001 0010 # loads the upper 16 bits of $t0 and sets the lower 16 bits to 0 LUI$t0, (0000 0010 0100 0010) 2 ORI$at, $t0, (0001 0011 1100 1100) 2 Slide 26 Memory instructions MIPS ISA address bus is 32 bits, i.e. it can address up to 2 32 (4 G) locations Memory width is usually 8 bits (byte) The MIPS ISA - Instructions 26 8 bits 0x00 00 00 00 0x00 00 00 01 0x00 00 00 02 0xFF FF FF FF MIPS CPU 32-bit Address Bus (32 bits) Data Bus (32 bits) Read Write Slide 27 Memory instructions Most memories are byte addressable ! Every four consecutive locations (8 bits each) form a word (32 bits) ! In other words, the addresses of two consecutive locations differ by 4 The MIPS ISA - Instructions 27 8 bits 0x00 00 00 00 0x00 00 00 01 0x00 00 00 02 0x00 00 00 03 0x00 00 00 04 0x00 00 00 05 0x00 00 00 06 0x00 00 00 07 0x00 00 00 08 0x00 00 00 09 0x00 00 00 0A 0x00 00 00 0B 0x00 00 00 0C 0x00 00 00 0D 0x00 00 00 0E 32 bits 0x00 00 00 00 0x00 00 00 04 0x00 00 00 08 0x00 00 00 0C 0x00 00 00 10 0x00 00 00 14 0x00 00 00 18 0x00 00 00 1C 0x00 00 00 20 Slide 28 Memory instructions Reading from memory requires specifying the address to read from and a register to store the read data Writing to memory requires specifying an address to write to and a register whose content will be written to memory Simple ! However, addressing a memory location requires 32 bits Where do these bits come from ?! Are they stored in memory?! Are they part of the instruction ?! Displacement Addressing In MIPS the address is formed by adding the content of some register (the base register) to 16-bit signed constant (offset or displacement) that is part of the instruction itself. The offset is sign-extended to 32 bits to match the size of the register. The MIPS ISA - Instructions 28 Slide 29 Memory instructions The Load instruction reads 32 bit value (A word) into a register The MIPS ISA - Instructions 29 lw $t0, 4 ($s4) OperationDestinationBase RegisterOffset Reg[$t0] = MEM[Reg[$s4]+sign_extend(4)] Register File + Instruction Memory Sign Extension 16 32 offset Memory Data Register Content Address Slide 30 Memory instructions The Store instruction writes the 32 bits (A word) found in a register into a memory location The MIPS ISA - Instructions 30 sw $t5, -12 ($t1) NemunicDestinationBase RegisterOffset MEM[Reg[$t1]+sign_extend(-12)] = Reg[$t5] Register File + Instruction Memory Sign Extension 16 32 offset Register Content Address Slide 31 Memory instructions The memory is byte-addressable. However, the load and store instructions deal with words (4 bytes)! Reading from memory is effectively reading four bytes! How these bytes are ordered in the 32 bit register ? Storing to memory is storing 32 bits in four consecutive locations. What is the order by which these bits are stored in 4 different locations? Two approaches Little endian: the least significant byte is associated with the location of lowest address (Intel) Big endian: the most significant byte is associated with the lowest address (MIPS) The MIPS ISA - Instructions 31 Slide 32 Reading a word from memory The MIPS ISA - Instructions 32 0x10 0x33 0x43 0xFD 0x00 00 00 F2 0x00 00 00 F3 0x00 00 00 F4 0x00 00 00 F5 8 bits Memory 0x100x330x430xFD Register Little Endian 32 bits 0x100x330x430xFD Register Big Endian 32 bits Byte 0Byte 1Byte 2Byte 3 Byte 0Byte 1Byte 2Byte 3 Slide 33 Memory Instructions Example 7. Convert the following high-level statements to MIPS assembly language G = 2 * F A[17] = A[16] + G Solution assume that the variables F and G are associated with registers $s4 and $s5, respectively, and the base address of the array A is in $s0. All values are 32-bit integers. add $s5, $s4, $s4 # $s5 contains 2*F lw $t0, 64($s0) # $t0 contains A[16] (note the offset is 64) add $t0, $t0, $s5 # $t0 contains A[16] + G sw $t0, 68($s0) # store $t0 in A[17] (note the offset is 68) The MIPS ISA - Instructions 33 Slide 34 The MIPS ISA - Instructions Memory Instructions - Machine Language 34 lw $t0, 4 ($s4) sw $t2, -6 ($at) oprsrtOffset (displacement) 65516 op 6-bitsopcode that specifies the operation rs 5-bitsregister file address containing the base address rt 5-bitsregister file address of destination register (LW) or the source register (SW) Offset 16-bitsdisplacement or offset (number of bytes to move, positive or negative) I-type How far can we move in memory when using LW and SW? Slide 35 Memory Instructions Example 6 The MIPS ISA - Instructions 35 Machine code for lw $s5, 4($s1) in hexadecimal op = 0x23 rs = $s1 = 0x11 rt = $s5 = 0x15 offset = 0x0004 Machine code for sw $s0, 16($s4) in hexadecimal op = 0x2b rs = $s4 = 0x14 rt = $s0 = 0x10 offset = 0x0010 6 bits5 bits 16 bits OpcodersrtOffset 10001110001101010000 0000 0000 0100 10101110100100000000 0000 0001 0000 Slide 36 Memory Instructions MIPS ISA defines memory instructions that can load/store bytes (8 bits) and half words (16 bits) lb $t0, 1($s3) #load byte from memory (sign extention) lbu $t0, 1($s3) # load byte from memory (zeros extension) sb $t0, 6($s3) #store byte to memory lh $t0, 1($s3) #load half word from memory (sign extension) lhu $t0, 1($s3) #load half word from memory (sign extension) sh $t0, 6($s3) #store half word to memory When loading a byte into 32-bit register, where is it stored? Lower 8 bits of the register ! What about remaining bits? When loading a half word into 32-bit register, where is it stored? Lower 16 bits of the register ! What about remaining bits? How about sb and sh instructions? The MIPS ISA - Instructions 36 Slide 37 Memory Instructions The MIPS ISA - Instructions 37 0xD0 0xC3 0x43 0xFD 0x00 00 00 F2 0x00 00 00 F3 0x00 00 00 F4 0x00 00 00 F5 8 bits Memory 0x00 0xD0 $s0 32 bits Byte 0Byte 1Byte 2Byte 3 LBU$s0, 2 ($s4) # assume Reg[$s4] = 0x00 00 00 F0 0xFF 0xD0 $s0 32 bits Byte 0Byte 1Byte 2Byte 3 LB$s0, 2 ($s4) # assume Reg[$s4] = 0x00 00 00 F0 0x30 0x04 0x00 00 00 F0 0x00 00 00 F1 What if the instructions are LH and LHU? Slide 38 Flow Control Instructions Instructions are loaded in memory! The normal execution of programs is sequential; one instruction after another! Keeping track of execution is done through a special register called the Program Counter (PC) 32 bits (Why???) It is incremented automatically by 4 (why!) after an instruction is fetched Thus, its contents always holds the address of the next instruction! It is not directly accessible !!! (Why) However, in our programs, we often write statements that skip some parts of the program, conditionally or unconditionally !! (IF, SWITCH, WHILE, GOTO ) How is this implemented in the hardware !? The MIPS ISA - Instructions 38 Slide 39 Flow Control Instructions How to skip instructions in memory to execute others? Basically, this requires using instructions that change the contents of the PC to point to the address where the target instructions are loaded (branch/jump address)! In MIPS, there are no instructions that can modify the PC directly! Alternatively, we have two conditional and three unconditional flow control instructions! These instructions can change the contents of the PC indirectly! The MIPS ISA - Instructions 39 Slide 40 Conditional Flow Control Instructions Branch If Equal Instruction (BEQ) Checks if two registers are equal! If true, then the PC (containing the address of next instruction) is incremented or decremented by adding a signed 16-bit offset (that is part of the instruction) after it is multiplied by 4 (WHY?). If false, program execution continues normally from the address in PC. This is called PC-relative addressing! The MIPS ISA - Instructions 40 BEQ $t0, $t1, 15 Operation Tested RegistersOffset Slide 41 Conditional Flow Control Instructions Branch If Equal Instruction (BEQ) The MIPS ISA - Instructions 41 PC Reg1Reg2 Instruction =? x4 MUXMUX Sign Extension + Address of instruction following BEQ Branch Address BEQ . MEM . 16 bit offset 32 bit instruction from branch address (two registers are equal) 32 bit instruction next to BEQ (Two registers are not equal) Slide 42 Conditional Flow Control Instructions Branch If Not Equal Instruction (BNE) Checks if two registers are not equal! If true, then the PC (containing the address of next instruction) is incremented or decremented by adding a signed 16-bit offset (that is part of the instruction) after it is multiplied by 4 (WHY?)). If false, program execution continues normally from the address in PC. This is called PC-relative addressing! The MIPS ISA - Instructions 42 BNE $t0, $t1, -23 Operation Tested RegistersOffset Slide 43 Flow Control Instructions (Conditional) Branch If Not Equal Instruction (BNE) The MIPS ISA - Instructions 43 PC Reg1Reg2 Instruction =? x4 MUXMUX Sign Extension + Address of instruction following BNE Branch Address BNE . MEM . 16 bit offset32 bit instruction from branch address (Two registers are not equal) 32 bit instruction next to BNE (Two registers are equal) Slide 44 Flow Control Instructions Example 8. Convert the following high-level statements to MIPS assembly language if (x == y) x = x + y end The MIPS ISA - Instructions 44 Solution assume that the variables x and y are associated with registers $s4 and $s5, respectively. BNE$s4, $s5, 1 # if $s5 ~= $s4, skip one instruction ADD$s4, $s4, $s5 # add x and y if they are equal OR, we can use labels instead of counting how many instructions to skip! BNE$s4, $s5, SKIP # if $s5 ~= $s4, skip one instruction ADD$s4, $s4, $s5 # add x and y if they are equal SKIP: . Slide 45 The MIPS ISA - Instructions Flow Control Instructions -Machine Language 45 BEQ $t0, $t1, 4 BNE $t2, $v1, -214 oprsrtOffset (displacement) 65516 op 6-bitsopcode that specifies the operation rs 5-bitsfirst register file address to be compared rt 5-bitssecond register file address to be compared Offset 16-bitsdisplacement or offset (number of instructions to skip, positive or negative!) I-type How many instructions can we skip when using BNE and BEQ? Slide 46 Conditional Flow Control Instructions Example 9. Machine code for BEQ $s5, $s1, 20 in hexadecimal. op = 0x04 rs = $s5 = 0x15 rt = $s1 = 0x11 offset = 0x0014 Machine code for BNE $s0, $s1, -13 in hexadecimal op = 0x05 rs = $s0 = 0x10 rt = $s1 = 0x11 offset = 0xFFFD The MIPS ISA - Instructions 46 00010010101100010000 0000 0001 0100 00010110000100011111 1111 1111 0011 Slide 47 Conditional Flow Control Instructions BEQ and BNE instructions can check for equality only? How about >, = ??? There are no instructions such bgt, bls, bge, ble!! In MIPS ISA, this can be done with the aid of The Set On Less than Instructions (SLT and SLTU) Compares if one register is less than another one If so, a third register is loaded with 1 Otherwise, the third register is loaded with 0 The $Zero register Register number zero Hardwired to zero Can be read only! The MIPS ISA - Instructions 47 Slide 48 Conditional Flow Control Instructions SLT Instruction The MIPS ISA - Instructions 48 SLT$s1, $s2, $s3#set $s1 if $s2 < $s3, else clear $s1 $S2 $S3 - $s2-$s3