1. 2016-6-91 The Stack An Hong han@ustc.edu.cn 2015 Fall han@ustc.edu.cn School of Computer Science and Technology Lecture on Introduction to Computing Systems

2. 2016-6-92 Review: Using Memory Memory Just a big “array” “Indexed” by address Accessed with loads and stores LD/LDR/LDI Read a word out of memory Use different addressing mode ST/STR/STI Place a word in memory Use different addressing mode

3. 2016-6-93 Review: Using Memory Problem What if the memory you want to access is far away? LD/ST won’t work (PC-relative) LDR/STR won’t work alone (need to get address in register) Solution:LDI/STI Place address of far away value nearby Load address, then load/store from that

4. 2016-6-94 Recursion Need Per-subroutine-invocation data space (activation record) Approach Allocate new activation record on a stack whenever a subroutine is called Subroutine uses its own activation record to hold invocationspecific data (e.g., local variables, saved registers) Note Given that Breakout is recursive, we will need activation records

5. 2016-6-95 Recursion What’s the problem with... recursion? Each subroutine invocation gets its own activation record... but how? First call to Foo (SaveR7 contains address of Next) Second call to Foo (SaveR7 contains address of After) First return from Foo (returns to After) Second return from Foo (returns to After again!!!)

6. 2016-6-96 Stack Data Structure A LIFO (last-in first-out) storage structure The first thing you put in is the last thing you take out The last thing you put in is the first thing you take out Two main operations PUSH: add an item to the stack POP: remove an item from the stack Error conditions: Underflow (try to pop from empty stack) Overflow (try to push onto full stack) A register (eg. R6) holds address of top of stack (TOS)

7. 2016-6-97 A Physical Stack Coin holder Last quarter in is the first quarter out (LIFO)

8. 2016-6-98 A Software Stack Data items don't move in memory, just our idea about where TOP of the stack is By convention, R6 holds the Top of Stack (TOS) Pointer (SP) x3FFF x4000 x4001 x4002 x4003 x4004

9. 2016-6-99 Basic Push and Pop Code Note: Stacks can grow in either direction (toward higher address or toward lower addresses) x3FFF x4000 x4001 x4002 x4003 x4004 PUSH ADD R6, R6, #1; increment stack ptr STR R0, R6, #0; store data(R0) to TOS POP LDR R0, R6, #0; load data(R0) from TOS ADD R6, R6, #-1; decrement stack ptr

10. 2016-6-910 Pop with Underflow Detection If we try to pop too many items off the stack, an underflow condition occurs. Check for underflow by checking TOS before removing data. Return status code in R5 (0 for success, 1 for underflow) POP LD R1, EMPTY ADD R2, R6, R1 ; Compare stack pointer BRz UNDER ; with x3FFF LDR R0, R6, #0 ; The actual ‘pop’ ADD R6, R6, #-1 ; Adjust stack pointer AND R5, R5, #0 ; Success: return R5 = 0 RET UNDER AND R5, R5, #0 ; Underflow: return R5 = 1 ADD R5, R5, #1 RET EMPTY.FILL xC001; EMPTY = -x3FFF

11. 2016-6-911 Push with Overflow Detection If we try to push too many items onto the stack, an overflow condition occurs. Check for underflow by checking TOS before adding data. Return status code in R5 (0 for success, 1 for overflow) PUSH LD R1, FULL ADD R2, R6, R1 ; Compare stack pointer BRz OVER ; with x4004 ADD R6, R6, #1; Adjust stack pointer STR R0, R6, #0 ; The actual ‘push’ AND R5, R5, #0 ; Success: return R5 = 0 RET OVER AND R5, R5, #0 ADD R5, R5, #1; Overflow: return R5 = 1 RET FULL.FILL xBFFC ; FULL = -x4004

12. 2016-6-912 PUSH & POP in LC-3 - 1 POPST R2,Sv2; save, needed by POP ST R1,Sv1; save, needed by POP LD R1, EMPTY ; EMPTY contains –x3FFF ADD R2, R6, R1 ; Compare stack pointer with x3FFF BRz Fail_exit; Branch if stack empty LDR R0, R6, #0 ; The actual ‘pop’ ADD R6, R6, #-1 ; Adjust stack pointer RET EMPTY.FILL xC001; EMPTY = -x3FFF PUSHST R2,Sv2; save, needed by PUSH ST R1,Sv1; save, needed by PUSH LD R1, FULL ADD R2, R6, R1 ; Compare stack pointer BRz Fail_exit ; with x4004 ADD R6, R6, #1; Adjust stack pointer STR R0, R6, #0 ; The actual ‘push’ RET FULL.FILL xBFFC ; FULL = -x4004

13. 2016-6-913 PUSH & POP in LC-3 - 3 Sv1.FILL x0000 Sv2.FILL x0000 Success_exit LD R1, Sv1 ;Restore reg values LD R2, Sv2 ; AND R5, R5, #0 ; Success: return R5 = 0 RET ; Fail_exit LD R1, Sv1;Restore reg values LD R2, Sv2 AND R5, R5, #0 ADD R5, R5, #1; Overflow: return R5 = 1 RET

14. 2016-6-914 Arithmetic Using a Stack Instead of registers, some ISA's use a stack for source and destination operations: a zero-address machine. Example: ADD instruction pops two numbers from the stack, adds them, and pushes the result to the stack. ADD vs. ADD R0,R1,R2 Evaluating (A+B)·(C+D) using a stack: push A push B ADD push C push D ADD MULTIPLY pop result x3FFA x3FFB x3FFC x3FFD x3FFE x3FFF x3FFA SP

15. 2016-6-915 (25+17) x (3+2) 25 x3FFB x3FFC x3FFD x3FFE x3FFF x3FFA SP 25 17 x3FFC x3FFB x3FFC x3FFD x3FFE x3FFF x3FFA SP 42 17 x3FFB x3FFC x3FFD x3FFE x3FFF x3FFA SP 42 3 x3FFC x3FFB x3FFC x3FFD x3FFE x3FFF x3FFA SP 42 3 2 x3FFD x3FFB x3FFC x3FFD x3FFE x3FFF x3FFA SP 42 5 2 x3FFC x3FFB x3FFC x3FFD x3FFE x3FFF x3FFA SP 210 5 2 x3FFB x3FFC x3FFD x3FFE x3FFF x3FFA SP 210 5 2 x3FFA x3FFB x3FFC x3FFD x3FFE x3FFF x3FFA SP

16. 2016-6-916 Data Type Conversion I/O Keyboard input routines read ASCII characters (not binary values) Console output routines write ASCII (‘s’ not “x73”) Consider this program: TRAP x23 ; input from keyboard ADD R1, R0, #0 ; move to R1 TRAP x23 ; input from keyboard ADD R0, R1, R0 ; add two inputs TRAP x21 ; display result TRAP x25 ; HALT User inputs ‘2’ and ‘3’ -- what happens? Result displayed: ‘e’ Why? ASCII '2' (x32) + ASCII '3' (x33) = ASCII 'e' (x65)

17. 2016-6-917 ASCII to Binary Single digit numbers are trivial (subtract x30) E.g., ‘7’ is ASCII x37, x37 - x30 = x7 Input Assume we've read three ASCII digits (e.g., "259") into a memory buffer How do we convert this to a number we can use? Convert first character to digit (subtract x30) and multiply by 100 Convert second character to digit and multiply by 10 Convert third character to digit Add the three digits together

18. 2016-6-918 ASCII to Binary Conversion Algorithm

19. 2016-6-919 Multiplication How can we multiply a number by 100? Approach 0 -Use the MUL instruction Approach 1 -Add to itself 10 times Approach 2 -Add 10 to itself times (better if number < 10) Approach 3 -Look it up! Only practical if number of multiplicands is small

20. 2016-6-920 Code for Lookup Table ; multiply R0 by 100, using lookup table ; LEA R1, Lookup100 ; R1 = table base ADD R1, R1, R0 ; add index (R0) LDR R0, R1, #0 ; load from M[R1]... Lookup100.FILL #0 ; entry 0.FILL #100 ; entry 1.FILL #200 ; entry 2.FILL #300 ; entry 3.FILL #400 ; entry 4.FILL #500 ; entry 5.FILL #600 ; entry 6.FILL #700 ; entry 7.FILL #800 ; entry 8.FILL #900 ; entry 9

21. Complete ASCII to Binary Conversion Code ASCII "259” to value 259 2016-6-921 ASCIIBUF 3 R1

22. 2016-6-922 Complete ASCII to Binary Conversion Code (1 of 3) ; Three-digit buffer at ASCIIBUF. ; R1 tells how many digits to convert. ; Put resulting decimal number in R0. ASCIItoBinary AND R0, R0, #0 ;clear result ADD R1, R1, #0 ;test # digits BRz DoneAtoB;done if no digits ; LD R3, NegZero ;R3 = -x30 LEA R2, ASCIIBUF;ptr to the first digit ADD R2, R2, R1;R2 =(R2)+(R1) ADD R2, R2, #-1 ;points to ones digit ; LDR R4, R2, #0 ;load digit ADD R4, R4, R3 ;convert to number ADD R0, R0, R4 ;add ones contribution ;

23. 2016-6-923 Complete ASCII to Binary Conversion Code(2 of 3) ADD R1, R1, #-1 ;one less digit BRz DoneAtoB ;done if zero ADD R2, R2, #-1 ;points to tens digit ; LDR R4, R2, #0 ;load ‘tens’ digit ADD R4, R4, R3 ;convert to number LEA R5, Lookup10 ;multiply by 10 ADD R5, R5, R4 LDR R4, R5, #0 ADD R0, R0, R4 ;adds tens contribution to total ; ADD R1, R1, #-1 ;one less digit BRz DoneAtoB ;done if zero ADD R2, R2, #-1 ;points to hundreds digit ; LDR R4, R2, #0 ;load digit ADD R4, R4, R3 ;convert to number LEA R5, Lookup100 ;multiply by 100 ADD R5, R5, R4 LDR R4, R5, #0 ADD R0, R0, R4 ;adds 100's contrib

24. 2016-6-924 Complete ASCII to Binary Conversion Code(3 of 3) DoneAtoB RET NegZero.FILL xFFD0 ;-x30 ASCIIBUF.BLKW 4 Lookup10.FILL #0.FILL #10.FILL #20... Lookup100.FILL #0.FILL #100.FILL #200...

25. 2016-6-925 Binary to ASCII Conversion Converting a 2's complement binary value to a three-digit decimal number Resulting characters can be output using OUT Instead of multiplying, we need to divide by 100 to get hundreds digit. Why wouldn't we use a lookup table for this problem? Subtract 100 repeatedly from number to divide. First, check whether number is negative. Write sign character (+ or -) to buffer and make positive.

26. 2016-6-926 Binary to ASCII Conversion Code (1 of 3) ; R0 is between -999 and +999. ; Put sign character in ASCIIBUF, followed by three ; ASCII digit characters. BinaryToASCII LEA R1, ASCIIBUF ;ptr to result string ADD R0, R0, #0 ;test sign of value BRn NegSign LD R2, ASCIIplus ;store '+' STR R2, R1, #0 BR Begin100 NegSign LD R2, ASCIIneg ;store '-' STR R2, R1, #0 NOT R0, R0 ;convert value to pos ADD R0, R0, #1 Begin100 LD R2, ASCIIoffset LD R3, Neg100 Loop100 ADD R0, R0, R3 BRn End100 ADD R2, R2, #1 ;add one to digit BR Loop100

27. 2016-6-927 Binary to ASCII Conversion Code(2 of 3) End100STR R2, R1, #1 ;store ASCII 100's digit LD R3, Pos100 ADD R0, R0, R3 ;restore last subtract ; LD R2, ASCIIoffset LD R3, Neg10 Loop100 ADD R0, R0, R3 BRn End10 ADD R2, R2, #1 ;add one to digit BR Loop10 End10STR R2, R1, #2 ;store ASCII 10's digit ADD R0, R0, #10 ;restore last subtract ; LD R2, ASCIIoffset ADD R2, R2, R0 ;convert one's digit STR R2, R1, #3 ;store one's digit RET ;

28. 2016-6-928 Binary to ASCII Conversion Code(3 of 3) ASCIIplus.FILL x002B ;plus sign ASCII code ASCIIneg.FILL x002D ;neg sign ASCII code ASCIIoffset.FILL x0030 ;zero’s ASCII code Neg100.FILL xFF9C ;-100 Pos100.FILL x0064;100 Neg10.FILL xFFF6 ;-10

29. 2016-6-929 Interrupt-Driven I/O Timing of I/O controlled by device 1. Report: Tells processor when something interesting happens -Example: when character is entered on keyboard -Example: when monitor is ready for next character -Example: when block has been transferred from disk to memory 2. Processing: Processor interrupts its normal instruction processing and executes a Interrupt Service Routine(ISR) (like a TRAP) -1. Figure out what device is causing the interrupt -2. Execute routine to deal with event -3. Resume execution No need for processor to poll device Can perform other useful work Interrupt is an unscripted subroutine call, triggered by an external event

30. 2016-6-930 Interrupt-Driven I/O Program A is executing instruction n Program A is executing instruction n+1 Program A is executing instruction n+2 Program A is executing instruction n+3 Program A is executing instruction n+4 ……

31. 2016-6-931 Interrupt-Driven I/O Program A is executing instruction n Program A is executing instruction n+1 Program A is executing instruction n+2 Interrupt!!! Program A is executing instruction n+3 Program A is executing instruction n+4 ……

32. 2016-6-932 Interrupt-Driven I/O Program A is executing instruction n Program A is executing instruction n+1 Program A is executing instruction n+2 1.Interrupt signal is detected 1.Program A is put into suspended animation 2.The needs os the I/O device start being carried out 2.The needs os the I/O device being carried out 3.Program A is brought back to life Program A is executing instruction n+3 Program A is executing instruction n+4 ……

33. 2016-6-933 When to Use Interrupts When timing of external event is uncertain Example: incoming packet from network TRAP vs. Interrupt, 12.5s vs. 0.00001s @ 100 char read from KB When device operation takes a long time Example: start a disk transfer, disk interrupts when transfer is finished processor can do something else in the meantime When event is rare but critical Example: building on fire -- save and shut down!

34. 2016-6-934 How is Interrupt Signaled? External interrupt signal by Devices: INT Device sets INT=1 when it wants to cause an interrupt Interrupt vector: INTV 8-bit signal for device to identify itself Also used as entry into Interrupt Vector Table, which gives starting address of Interrupt Service Routine (ISR) -Just like Trap Vector Table and Trap Service Routine -TVT: x0000 to x00FF -IVT: x0100 to x01FF

35. Interrupt Signal to CPU Several things must be true for an I/O device to actually interrupt the processor (ALL the three elements are present) The I/O device must want service(ready bit = 1) The device must have the right to request the service(interrupt enable bit = 1). The device request must be more urgent than what the processor is currently doing. 2016-6-935

36. 2016-6-936 Interrupt Signal to CPU At the device side(I/O device set) Control register has “Interrupt Enable” bit Must be set for interrupt to be generated At the processor(CPU set) Sometimes have “Interrupt Mask” register (LC-3 does not) When set, processor ignores INT signal Why? - Example: may not want to be interrupted while in ISR KBSR 1514 0 I/O device set: ready bit 13 CPU set: interrupt enable bit interrupt signal to processor(INT)

37. The State of a Program 2016-6-937 1514131211109876543210 PrPLNZPPSR Cond Codes Priority Level The Processor Status Register PSR[15] 0, privileged (supervisor) mode 1, unprivileged(user)mode The Priority Level of a program being executed PL[10:8] 0( 最低 )~7( 最高 ) Priv Mode

38. Interrupt Signal to CPU 2016-6-938 PL1 Device 14 0 13 PL2 Device 14 0 13 PL7 Device 14 0 13 Priority encoder INT B A ? A>B PL of executing program …... What if more than one device wants to interrupt? External logic controls which one gets to drive signals

39. 2016-6-939 Testing for Interrupt Signal CPU looks at signal between STORE and FETCH phases If not set, continues with next instruction If set, transfers control to interrupt service routine EA OP EX S S F F D D interrupt signal? Transfer to ISR Transfer to ISR NO YES

40. 2016-6-940 How Does Processor Handle It? Examines INT signal just before starting FETCH phase If INT=1, don’t fetch next instruction Instead -Save program state (PC, PSR (privilege and CCs)) on stack -Update PSR (set privilege bit) -Index INTV into IVT to get start address of ISR (put in PC) After service routine RTI instruction restores PSR and PC from stack Need a different return instruction, because  RET gets PC from R7 and doesn’t update PSR -RTI gets PC from R7 and always clears privilege bit in PSR Processor only checks between STORE and FETCH phases -- Why?

41. 2016-6-941 Supervisor Mode and the Stack Problem PC and PSR shouldn’t be saved on user stack What if R6 is uninitialized? What if user has set R6 to refer to OS memory? User could see OS data (when trap returns) Solution Create two versions of R6 (stack pointer) in register file  One is user stack pointer (what we’ve been using all along) -The other is supervisor stack pointer Extra register file logic selects the appropriate register based on privilege bit in current PSR Bottom line: OS code always uses its own stack

42. 2016-6-942 Example (1)

43. 2016-6-943 Example (2)

44. 2016-6-944 Example (3)

45. 2016-6-945 Example (4)

46. 2016-6-946 Example (5)

47. 2016-6-947 Example (6)

48. 进一步学习 第 11-19 章 32 位 CPU 设计,~50 条指令 RDMA 竞赛 全国并行应用优化大赛 ISC-SCC 国际大学生超算竞赛 SC-SCC 国际大学生超算竞赛 2016-6-948

49. 感谢大家对这门课教学工作的配合与支持 ! 2016-6-949

50. 祝大家期末考试取得好成绩 ! 2016-6-950

51. 祝大家新年快乐 ! 2016-6-951