1. Jon Stanley EE382N-4 Spring 2011

2.  Objectives  Tasks  Keil MCB2300 ARM7 Evaluation Board ◦ System overview ◦ Measurement setup ◦ Instruction power usage ◦ Code power comparison  Beagleboard-XM ◦ System overview ◦ Android and Ubuntu OS experience ◦ Measurement setup ◦ Code power comparison  Summary  Other notes

3.  Explore several points in Lecture 16 on Power Aware Computing  Characterize power of an ARM processor  Characterize energy use of code ◦ Low level (bare metal) ◦ High level (Linux OS) ◦ Coding methods

4.  Power and energy measurements  Software ◦ Instruction level ◦ Small routines ◦ C-code comparisons:  Delays, timer interrupts, OS scheduling ◦ Linux OS impact ◦ Optimization impact ◦ Data type size impact  Hardware ◦ Keil MCP2300 and Beagleboard-XM

5.  NXP LPC2387 microcontroller ◦ ARM7 processor core ◦ GPIO to LEDs, ADC I/O, etc.  uVision IDE and Flash Magic for programming  Bare metal code ◦ More accurate measure of power ◦ Can measure power of each instruction type

6.  Measurement setup ◦ Ammeter with uA resolution due to tiny changes in power consumption of different instructions ◦ Use good shielded probe to VBUS jumper on board ◦ Careful of current drift if instrument isn’t high quality

7. 0loopADD R0, R0, #1 1ADD R1, R1, #1 2ADD R2, R2, #1 3ADD R3, R3, #1 4ADD R4, R4, #1 5ADD R0, R0, #1 … 999ADD R4, R4, #1 B loop  Same instruction 1000 times to minimize effect of Branch on power  Consider data hazards when writing test code

8. InstructionCurrent (mA) NOP41.94 MOV41.56 ADD41.59 ADD with stall41.51 ADD with LSL41.52 MOV with LSL41.57 MUL41.65 B41.05 PC Load to Jump40.86  NOP uses more power?  Stalled pipeline uses less power, but uses more energy due to longer execution time  ADD uses less power than MUL  MOV to PC uses less power than Branch

9. loopADD R0, R0, #1 CMP R0, #1 ADDLT R1, R1, #1 ADDEQ R1, R1, #1 ADDGT R1, R1, #1 B loop  No measurable difference at instruction level  Branch would be better when there is large amounts of code between each conditional statement to reduce execution time loopADD R0, R0, #1 CMP R0, #1 BLT jump BEQ jump jumpADD R1, R1, #1 B loop MethodCurrent (mA) Conditional42.33 Branch42.33

10.  C Code  LED PWM dimming  Timer interrupt PWM ◦ Interrupt handler ◦ LED control by interrupt  Delay PWM ◦ Main loop uses delays ◦ Main loop controls LED  Surprising result? ◦ Delay did not put processor in idle (like OS sleep). ◦ Tests were repeated later in Linux OS. MethodCurrent (mA) Interrupt92.31 Delay92.31

11.  Released in 2010  Features: ◦ TI DM3730 microcontroller  1GHz ARM Cortex-A8 core ◦ 512MB DDR ◦ Ethernet 10/100 ◦ 4-port USB ◦ RS232 ◦ Stereo audio ◦ S-video and DVI ◦ Boots from 4GB MicroSD card ◦ Comes with Angstrom Linux OS verification image ◦ Power usage is a realistic scenario of a system

12.  Installation files: http://software- dl.ti.com/dsps/dsps_public_sw/sdo_tii/TI_Android _DevKit/02_00_00/exports/AM37X.tar.gz  README is contained within the tarball  PC with Ubuntu 10 to build file system on SD card  Android boot-up and program execution was slow  Not friendly for developing code within OS

13.  Installation files and instructions: http://elinux.org/BeagleBoardUbuntu  Use instructions for “ Demo Image - Maverick 10.10 ” ◦ Installation will take at least 4 hours ◦ Need internet connectivity - use “dhclient usb1” to get IP ◦ Note: the NetInstall method resulted in broken OS for me  PC with Ubuntu 10 to build file system on SD card  Ubuntu GUI runs faster than Android ◦ Built-in terminal and friendly for developing within OS ◦ “sudo apt-get install gcc” to install GCC compiler

15.  Hardware: BeagleBoard with Ubuntu ◦ Fluctuating current consumption so a DMM is not practical for an average current measurement ◦ NI cDAQ-9174 with 9227 analog current module ◦ Sample current consumption at 2kS/sec ◦ Measure average current with ~1.5 minute window ◦ Turn off user-defined LEDs to reduce noise  echo none >/sys/class/leds/beagleboard\:\:usr0/trigger  echo none >/sys/class/leds/beagleboard\:\:usr1/trigger  Software: 2 C-codes for each test ◦ First code has simple for loop for current measure ◦ Second code uses gettimeofday() and runs the same for loop 1000 times to determine average runtime

16.  Simple code that accumulates iteration count main() { int i,c; while(1) { for(i=0;i<10000;i++) c += i; }

17. Idle OS Running Test Program

18.  HW2 exercise – gcc creates poor asm code that performs unnecessary memory operations  Direct correlation to energy consumption  Energy = V * (I code – I IdleOS ) * T  3x decrease in energy consumption by eliminating unnecessary memory operations TestCurrent (mA) Time (us) Coulomb (mA*us) Idle OS602 For loop sum in C6731359585 Optimized for loop in asm677453375

19.  Coulomb usage comparison of for loop sum code using different data types for the variable “c”  Coulomb usage appears correlated to both the complexity of the library function that performs the add and size of the data type TestBytesCurrent (mA) Time (us) Coulomb (mA*us) Idle OS602 Short265718410120 Integer46731359585 Float463989132967 Double866048728246

20.  Test application: ◦ Loop that writes to variable. 50% duty. ~2ms period.  Test 1: Delay ◦ Cannot depend on CPU clock period for delay timing due to OS, so poll using gettimeofday  Test 2: OS Scheduler ◦ Use usleep for timing with us resolution TestCurrent (mA) Time (us) Coulomb (mA*us) Jitter (us) MinAvgMax Idle OS598 Delay6212020464607107209 Scheduler59822380371809162

21.  Smallest data type generally lower energy  Use better compilers or check assembly code  Direct correlation between code performance and energy usage for ARM processors ◦ Beware of stalling the pipeline ◦ Look at Coulomb usage of different code approaches  Fast algorithms may use more power than slow ones but use less energy overall ◦ Memory transactions are energy expensive ◦ Use interrupts or OS scheduler instead of polling in OS  Results may not entirely hold for x86 or other processors, but these can be characterized using the methods outlined here

22.  Issues: ◦ Difficult to find documentation ◦ Community-oriented development ◦ Fairly new board  Processor voltage and clock is adjustable but driver library not fully merged in Ubuntu yet ◦ cpufrequtils package  Processor current consumption is accessible to software on the board via device on I2C

23.  Processor current consumption is accessible to software on the board ◦ Not recommended for precision gauging because 10-bit ADC resolution yields ~53mA per LSB with series resistor  Hardware details in BeagleBoard user manual  Software: ◦ http://groups.google.com/group/beagleboard/browse_t hread/thread/7810fb7a93e44a4e ◦ Software is currently broken ◦ /dev/I2C-1 file access issue in Ubuntu  “sudo i2cdump 1 0x4a” in terminal returns “Device or resource busy” error  Permission conflict with OS over i2c file access?  “sudo fuser –km /dev/i2c-1” will freeze the OS if attempting to kill any processes that use i2c-1

24.  Current consumption comparison of various system energy saving or shutdown options accessible from the Ubuntu GUI TestCurrent (mA) Idle OS602 Blank screen598 Suspend566 Processor halted (shutdown)380

25.  Embedded System Power ◦ Lecture 16 – Power Aware Programming ◦ http://www.newelectronics.co.uk/electronics-technology/optimising-the- power-consumption-of-embedded-systems/30528/ ◦ http://www.wimserc.org/members/Papers/1215608754-PID632073.pdf ◦ http : //www.netrino.com/node/178 ◦ http://low-powerdesign.com/article_cypress_093010.html  Keil MCB2300 ◦ http://www.keil.com/support/man/docs/mcb2300/mcb2300_intro.htm  Beagleboard-XM ◦ Beagleboard.org ◦ Google Groups on Beagleboard topics