Re: [Simulavr-devel] using simulavr for profiling

[Theodore A. Roth]

On 17 Mar 2003, Richard Barrington wrote:

:)Hi,
:)
:)Has there been any development toward getting simulavr to produce gprof
:)output? I think this would be a useful thing... Any thoughts on what
:)would be involved to make this happen?

I've never looked into it. I could be useful though.

As for making it happen, I'm not sure what that would entail.

The way I see it happening on a Linux/Unix system, is that the compiler 
inserts extra calls into you code which hook into the gprof code. I'm not 
sure if this is the right approach for simulavr.

I'm not sure how to approach this though. My first impression seems to be to 
just hook into the call, rcall and return opcodes in simulavr. On each call 
or rcall, log the current clock count and PC, on each return, look up the 
log entry by where you are going to return to (i.e. the PC), then dump the 
entry to a gprof formatted file.

This doesn't seem too difficult, but you will take a run-time speed hit by 
implementing it. That's not too bad though, compared with the information 
you will gain.

I'll have to add this to my todo list...


Ted Roth