COS10004 Computer Systems

Lecture 8.3 LED Flash (part 3) – a better busy wait

(the Timer Register)

Dr Chris McCarthy
 INSERT DELAY WITH BUSY WAIT TIMER
•  Pseudocode:
–  Program GPIO18 LED for output
–  Loop1:
•  Turn LED on (pull GPIO18 high)
•  Busy wait
•  Turn LED off (pull GPIO18 low)
•  Busy wait
–  branch to loop1

25/9/20 COS10004 Computer Systems 2

 A DUMB TIMER
•  Variables:
–  r0 = GPIO base address
–  r1 = working memory (for setting bits, registers)
–  use r2 for timing
•  mov r2,$3F0000 Does a 'busy
wait' - uses 100%
loop1: of CPU
sub r2,#1
cmp r2,#0
bne loop1

25/9/20 COS10004 Computer Systems 3

OK2 RUNNING ON PI 2B AND PI 4B
 A BETTER TIMER
•  RPi has a dedicated timer register:
–  independent of clock speed.
–  housed inside the same chip as the ARM CPU, the
GPIO, GPU, RAM and most other things.
–  This chip is called the SoC (System on a Chip).
•  The Timer registers start at
BASE address + 0x3000
–  Timer counts 1 microsecond intervals (106 per
second)

25/9/20 COS10004 Computer Systems 5

 A BETTER TIMER
•  The RPi timer registers:
Byte offset Size / Read or
Name Description
(from BASE) Bytes Write

Register used to control and clear

0x3000 4 Control / Status timer channel comparator RW
matches.

A counter that increments at

0x3004 8 Counter R
1MHz.

0x300C 4 Compare 0 0th Comparison register. RW

0x3010 4 Compare 1 1st Comparison register. RW

0x3014 4 Compare 2 2nd Comparison register. RW

0x3018 4 Compare 3 3rd Comparison register. RW

25/9/20 COS10004 Computer Systems 6

 PSEUDOCODE
Still does a 'busy
•  pseudocode: wait' – but
measures real
store base address of timer time
store delay
mov start_time (=now)
loop:
read now
remaining_time = (now – starttime)
compare remaining_time, delay
loop if remaining_time <= delay

25/9/20 COS10004 Computer Systems 7

 READING THE TIME (TIMESTAMP)
•  Variables:
•  r3 - base address of timer
•  r4 - desired delay
•  r6, r7 - where the current time will be stored
•  r5 - start time Timer returns
64-bit time, but
•  r8 - elapsed time (r6-r5) we only need
the lower
register (r6) -
must be an even
register
25/9/20 COS10004 Computer Systems 8
 WITH REGISTERS

store base address of timer (r3)

store delay (r4)
mov start_time (r5)(=current_time (r6))
timerloop:
read current_time (r6)
remaining_time (r8)= current_time (r6) –
start_time (r5)
compare remaining_time (r8), delay (r4)
loop if LE (remaining_time <= delay)

25/9/20 COS10004 Computer Systems 9

 delimiter
RECALL LOAD REGISTER ?
ldr r0,[r1]

register 0

r1 must contain a
pointer to a value in
memory

•  load register 0 with the value pointed to by r1

25/9/20 COS10004 Computer Systems 10

 A BETTER TIMER
•  BASE = $FE000000 d suffix
TIMER_OFFSET = $3000 means
mov r3,BASE double-word
orr r3,TIMER_OFFSET ;base address of timer (64 bits)
•  ldrd r6,r7,[r3,#4] ;copy timestamp [4 bytes after r3] to r6 (low
byte), r7(high byte).
–  d suffix means double-word (64 bit)
–  http://infocenter.arm.com/help/index.jsp?topic=
/com.arm.doc.ddi0338g/Chddeedh.html

•  mov r4,$80000 ;0.524sec

25/9/20 COS10004 Computer Systems 11

 WITH CODE (4B)
BASE = $FE000000 •  r3=BASE + TIMER_OFFSET + 4
TIMER_OFFSET = $3000 (0x3F003004)
mov r3,BASE •  [r3,#4] means
orr r3,TIMER_OFFSET ;store base address of timer (r3)
value at ((address in r3) + 4)
mov r4,$80000 ;store delay (r4)
ldrd r6,r7,[r3,#4] can't re-use loop label - each
must be unique: loop1,
mov r5,r6 ;mov starttime (r5)(=currenttime (r6))
timerloop: loop2, loop3...
ldrd r6,r7,[r3,#4] ;read currenttime (r6)
sub r8,r6,r5 ;remainingtime (8)= currenttime (r6) - starttime (r5)
cmp r8,r4 ;compare remainingtime (r8), delay (r4)
bls timerloop ;loop if LE (reaminingtime <= delay)

25/9/20 COS10004 Computer Systems 12

 ALL THE CODE (B)
BASE = $20000000 ;B B+. Use $3F000000 for 2
GPIO_OFFSET = $200000
mov r0,BASE
orr r0,GPIO_OFFSET ;Base address of GPIO
mov r1,#1
lsl r1,#18 ;B
str r1,[r0,#4] ;enable output
mov r1,#1
lsl r1,#16
loop$:
str r1,[r0,#40] ;Turn on LED
;new timer
TIMER_OFFSET = $3000
mov r3,BASE
orr r3,TIMER_OFFSET ;store base address of timer (r3)
mov r4,524288;store delay (r4)
ldrd r6,r7,[r3,#4]
mov r5,r6 ;store starttime (r5)(=currenttime (r6))
loopt1:
ldrd r6,r7,[r3,#4] ;read currenttime (r6)
sub r8,r6,r5 ;remainingtime (8)= currenttime (r6) - starttime (r5)
cmp r8,r4 ;compare remainingtime (r8), delay (r4)
bls loopt1 ;loop if LE (reaminingtime <= delay)
str r1,[r0,#28] ;turn off LED
;re-use timer
ldrd r6,r7,[r3,#4]
mov r5,r6 ;store starttime (r5)(=currenttime (r6))
loopt2:
ldrd r6,r7,[r3,#4] ;read currenttime (r6)
sub r8,r6,r5 ;remainingtime (8)= currenttime (r6) - starttime (r5)
cmp r8,r4 ;compare remainingtime (r8), delay (r4)
bls loopt2 ;loop if LE (reaminingtime <= delay)
b loop$

25/9/20 COS10004 Computer Systems 13

 ALL THE CODE (2)
BASE = $FE000000 ;4B
GPIO_OFFSET = $200000
mov r0,BASE
orr r0,GPIO_OFFSET ;Base address of GPIO
mov r1,#1
lsl r1,#24
str r1,[r0,#4] ;enable output
mov r1,#1
lsl r1,#18
loop$:
str r1,[r0,#28] ;Turn on LED
;new timer
TIMER_OFFSET = $3000
mov r3,BASE
orr r3,TIMER_OFFSET ;store base address of timer (r3)
mov r4, 524288 ;store delay (r4)
ldrd r6,r7,[r3,#4]
mov r5,r6 ;store starttime (r5)(=currenttime (r6))
loopt1:
ldrd r6,r7,[r3,#4] ;read currenttime (r6)
sub r8,r6,r5 ;remainingtime (8)= currenttime (r6) - starttime (r5)
cmp r8,r4 ;compare remainingtime (r8), delay (r4)
bls loopt1 ;loop if LE (reaminingtime <= delay)
str r1,[r0,#40] ;turn off LED
;re-use timer
ldrd r6,r7,[r3,#4]
mov r5,r6 ;store starttime (r5)(=currenttime (r6))
loopt2:
ldrd r6,r7,[r3,#4] ;read currenttime (r6)
sub r8,r6,r5 ;remainingtime (8)= currenttime (r6) - starttime (r5)
cmp r8,r4 ;compare remainingtime (r8), delay (r4)
bls loopt2 ;loop if LE (reaminingtime <= delay)
b loop$

25/9/20 COS10004 Computer Systems 14

 USING THE TIMER REGISTER
•  An absolute timer (no longer dependent on
processor speed!
•  Our approach is still a form of busy wait though !
–  CPU still occupied for duration of delay
–  Essentially this is polling the timer
•  Is there a potentially even better approach ?
–  Well yeah! We could implement an interrupt based
timer
–  More complicated though (and beyond)
 SUMMARY
•  The timer register provides an accurate timer
that can easily be accessed
•  We read it using ldrd
–  a version of load register that reads 64 bit words
into two specified 32 bit registers
•  Our timer is better than before, but is still a
“busy wait” timer
–  Interrupt based timing would solve this