Alessio Matt�ffffe8 wrote:
Hi all,
I implemented an FFT algorithm on mica2/mica2dot platform.
Now, I want to profile the NeSC code that I wrote. I used two timer to calculate the elapsed time for computing the FFT algorithm and, in particular, I used the SysTimeC component that should provide a timer with 1.2 microsec accuracy (http://mail.millennium.berkeley.edu/pipermail/tinyos/2003-December/000150.html)
In your opinion, is it a right procedure to evaluate the performances of mica2/mica2dot motes ?
Is there a way to evaluate the processing cycles (Atmel ATmega128L CPU cycles) that my algorithm spends ?
Is there a profiling tool for NeSC code or for AVR Assembly (today, I read about AVRORA simulator on the mailing list, but it is a simulator and I want to quantify the real performances of mica2/mica2dot) ?

you can try PowerTOSSIM (http://www.eecs.harvard.edu/~shnayder/ptossim/), but it's not extremely accurate as far as measuring CPU cycles goes. For something like FFT profiling, I'd recommend trying atemu (http://www.isr.umd.edu/CSHCN/research/atemu/)-it does a cycle level simulation of the atmega128, so should be quite accurate.


To get timings:

* Put Leds.redOn(), Leds.redOff() around the code you want to time
* make mica2
* atemu main.exe
Type g

It will print messages like

000                  970 event led Green off
000                13688 event led Red on

The second column is the cycle count at that point.  Subtract.

Hope this helps,
-Victor
_______________________________________________
Tinyos-users mailing list
[EMAIL PROTECTED]
http://mail.Millennium.Berkeley.EDU/mailman/listinfo/tinyos-users

Reply via email to