Performance Analysis Tool

Hi all,

I just finished an application for a msp430. Unfortunately I need 2
redesigns to finally reach the required performance. The RMA tools I
know do not support things like "one task triggers the execution of
another task" or "bursts of execution in a cyclic manner (e.g. serial
interrupts).

Does anyone knows a good tool to model a real time system to get
performance figures early during design?

Thanks a lot,
Peter

"Peter Mueller" <peter.o.mueller@gmx.de> wrote in message
news:1g5yrw8.um280q129uhk0N%peter.o.mueller@gmx.de...
> Hi all,
>
> I just finished an application for a msp430. Unfortunately I need 2
> redesigns to finally reach the required performance. The RMA tools I
> know do not support things like "one task triggers the execution of
> another task" or "bursts of execution in a cyclic manner (e.g. serial
> interrupts).
>
> Does anyone knows a good tool to model a real time system to get
> performance figures early during design?

How much money do you want to spend?

What kind of answers do you need?

"John R. Strohm" <strohm@airmail.net> wrote in message 
> 
> How much money do you want to spend?
> 
> What kind of answers do you need?

Hi John,

I want to understand 
- if my design meets the requirements: speed, response time.
- find potential for tuning
- better understand where the system spends its time

Peter

"Peter Mueller" <peter.o.mueller@gmx.de> wrote in message
news:ae3e5034.0312171123.1fc583e5@posting.google.com...
> "John R. Strohm" <strohm@airmail.net> wrote in message
> >
> > How much money do you want to spend?
> >
> > What kind of answers do you need?
>
> Hi John,
>
> I want to understand
(A) > - if my design meets the requirements: speed, response time.
(B) > - find potential for tuning
(C) > - better understand where the system spends its time

OK, going on the assumption that this is a simple application on a single
microcontroller...

Debugging is usually simpler if you start with known working code in an easy
debug environment.  If your code doesn't work in a benign workstation
environment, it won't work on the embedded platform, either.  So build a sim
environment on a Linux box, and run gprof to profile it.  You WON'T see time
spent in library routines, just time spent in your code.  That gets you a good
start on item (C), and knowledge of (C) is the first thing that feeds (B).
Tuning something that in untuned form takes 1% of your total runtime isn't going
to buy you very much.

Unfortunately, for (A), beyond the usual analyses about how many events per
second you have to process, and how much computation is involved per event, and
how much data has to move through which pieces of the system per event and per
second, there isn't a good uniform methodology for tackling this kind of
problem.