[Coco] 6809/6309 Power Analysis

Thu Jan 21 18:16:14 EST 2010

I'm not so sure this presents that big of a problem. If the intent to is to 
measure pure processor current, you're probably right; however, I think 
think current sourced to read/write memory and used by the circuits in 
support of the processor is meaningful if the intent to support greener 
development methodology as a new branch of science. It could certainly lend 
credence to register-based optimizations to reduce current drain accessing 
memory if that's the scope.

--------------------------------------------------
From: "Christian Lesage" <hyperfrog at gmail.com>
Sent: Thursday, January 21, 2010 5:55 PM
To: "CoCoList for Color Computer Enthusiasts" <coco at maltedmedia.com>
Subject: Re: [Coco] 6809/6309 Power Analysis

> I believe it's going to be tricky if you intend to measure the power 
> consumption related to the execution unit alone. You have to remember that 
> part of the total power is delivered to the external busses and varies 
> with the input or output levels as well as on the load. For instance, no 
> matter what instruction will be be executed, the power consumption could 
> differ whether you output $FFFF or $0000 on the address bus in order to 
> fetch it. In one case, you are sourcing current, while in the other case, 
> you are sinking current. The same applies to the data bus, but in addition 
> to that, whether you read or write a value will also affect power 
> consumption. As regards the address bus, your program will be outputting a 
> value there whenever it needs to read a new program or data byte. You only 
> have a limited control over the addresses being output. As regards the 
> data bus, your program will always be at least fetching a new program byte 
> every few cycles, and again, you have a limited control over what's being 
> read. I believe the particular opcode of a given instruction as well as 
> the operand's value could affect power consumption (e.g. reading $FF could 
> require more/less power than reading $00). It also depends on whether the 
> busses are buffered from the memory and I/O circuits, and how "efficient" 
> the buffers are (e.g. driving a HCT CMOS buffer requires less power than 
> driving an LS TTL one).
>
> Christian
>
> Boisy G. Pitre wrote:
>> Calling all hardware/electrical engineering gurus on the list...
>>
>> As some of you know, I am working on my thesis this semester where I am 
>> exploring power-aware compiler optimizations.  I've selected (surprise) 
>> the 6809/6309 as my evaluation processor and will be retooling RELIC to 
>> use the power information that I am attempting to gather to make 
>> decisions about instruction use based on the current that they draw.
>>
>> I'm hoping you hardware gurus can give me some input on the measurement 
>> part, in which I am attempting to determine the current draw of the 6809 
>> when executing specific instructions.  The absolute value is not 
>> important; it is the relative difference of current draw for different 
>> instructions that I need to move forward.
>>
>> Just to give you an idea of the type of equipment that I have access to 
>> at the UL campus:
>>
>> - Agilent Technologies MS06104A Mixed Signal Oscilloscope with analog 
>> probes as well as two 54620-61601 8-wire Logic Analyzer probe cables
>> - Agilent Portable Logic Analyzer, don't have the exact model number but 
>> looks like a microwave oven and I am told costs around $55K.
>>
>> The ideas brought up so far for measuring current at the instruction 
>> level:
>> - use a PIC with an A/D convertor to measure the voltage.
>> - use a shunt resistor (james recommended this method to me as well in 
>> private email) to measure current and outlined here: 
>> http://jet-server.commtest.co.nz/kb2/11624.htm
>>
>> On the software side, I have constructed several test executables that 
>> run under NitrOS-9 and execute a particular instruction over and over 
>> (with interrupts masked of course). The baseline measurement instruction 
>> is CWAI, which puts the processor in sleep mode.  The code looks 
>> something like this:
>>
>>     cwai #^IntMasks
>>
>> With interrupts masked there should be no activity and the cwai should 
>> put the processor to a wait mode.  I would then take the measuring device 
>> of choice and measure the current for a fixed amount of time (say 1 
>> second).
>>
>> Afterwards, I would run the following program which has a series of 100 
>> MUL instructions (MUL is chosen for presumably higher use of CPU):
>>
>>      ldd       #$0000
>>      orcc     #IntMasks
>> toploop:
>>      mul
>>      mul
>>      mul
>>      ....
>>      mul          * 100th instruction
>>      bra       toploop
>>
>> And again, use the same measuring device to measure the current draw for 
>> 1 second.
>>
>> With a low and high mark, I could then begin measuring other instructions 
>> and see where they fall with the high-low current range.
>>
>> My concern is the accuracy of the measurement, since I expect that the 
>> difference in current draw between instructions is small.  This is just a 
>> guess, but the level of accuracy that I would need to measure the 
>> difference in current draw between instructions would have to be AT LEAST 
>> 10e-3, or .001 amp (1 milliamp).
>>
>> Suggestions?  How would you guys do this? Do I really need the high 
>> dollar equipment that I have access to, or do I need a high accuracy, 
>> sensitive multimeter that can measure current in very very small amounts?
>>
>> Any input from the hardware experts here would be much appreciated.
>> --
>> Boisy G. Pitre
>> http://www.tee-boy.com/
>>
>>
>>
>> --
>> Coco mailing list
>> Coco at maltedmedia.com
>> http://five.pairlist.net/mailman/listinfo/coco
>>
>>
>
>
> --
> Coco mailing list
> Coco at maltedmedia.com
> http://five.pairlist.net/mailman/listinfo/coco
> 