GCC compiler specific question

What about the address of the function? It should be unique, and it is perfect for building function tables!

Regards,

Zara

Zara schrieb:

Good idea, but the problem is I have to implement a search algorithm to find out if the currently processed function is already present in the table or not. I wanted to avoid to do this way because of timing issues.

Here is a small code snippet to give a rough idea of what I intend:

void FunctionName(void) { # ifdef __PROF__ Prof_g[FunctionIndex].sFuncName = __func__; Prof_g[FunctionIndex].nNCalls++; Prof_g[FunctionIndex].nStartTime = CTM4_g.nMCSM11Counter; # endif /* __PROF__ */ ... FunctionBody ... # ifdef __PROF__ Prof_g[FunctionIndex].nStopTime = CTM4_g.nMCSM11Counter; Prof_g[FunctionIndex].nCurExecTime = Prof_g.nStopTime - Prof_g.nStartTime; # endif /* __PROF__ */ }

Binary search? Just 2log(N) iterations for N functions. 10000 functions take

14 iterations.

Meindert

I suppose you could do something like:

int global_function_number; ...

void FunctionName(void) { #ifdef __PROF__ static int FunctionIndex; extern int global_function_number; if(!FunctionIndex) { FunctionIndex = global_function_number++; } Prof_g[FunctionIndex].sFuncName = __func__; Prof_g[FunctionIndex].nNCalls++; Prof_g[FunctionIndex].nStartTime = CTM4_g.nMCSM11Counter; ...

--

John Devereux

Instead of adding explicit profiling code to the source, have you considered Monte Carlo profiling ? For counting function calls, you could use the gcc -p or -pg flag.

Its address.

Why? Do you want to re-invent the -pg option and gprof? And/or

-fprofile-arcs and gcov?

--
Hans-Bernhard Broeker (broeker@physik.rwth-aachen.de)
Even if all the snow were burnt, ashes would remain.

Hans-Bernhard Broeker schrieb:

No, I don't want to reinvent any GNU tools. But I cannot imagine that it is possible to get a real picture of an embedded software with asynchronous interrupts from the outside world with these tools. So I decided to follow the way I described.

Look for -finstrument-functions option. GCC will generate calls to the entry and exit profiling functions which you can implement as you wish.

Vadim

You may not have set out wanting to --- by you definitely appear to be doing exactly that.

The timer used by -pg is inside the target system, i.e. it'd be a timer of your embedded system. The builtin featur you're looking at

*is* -pg, and more to the point, the mcount() function used by it. Your best by far is going to be to change mcount() and/or the

-fprofile-arcs support machinery for your purpose. The tricky part will be to get the generated data off the embedded system and presentable for gprof, of course.

--
Hans-Bernhard Broeker (broeker@physik.rwth-aachen.de)
Even if all the snow were burnt, ashes would remain.