A profiling tool that I've written as a hobby project. This is a simple profiling tool that can be uesd to profile C programs. I haven't really tested it's correctness throughly, yet.


HOWTO:
-------------------------------------------------------------------------------------
 - compile the program along with the "aprof.c" file, using the "-finstrument-functions" compile time option.
              eg: gcc -g proc1.c proc2.c aprof.c -finstrument-functions
              
 - How to read the output:
   The output currently is very clumsy. The output consists of the function address, the self ticks taken and the total ticks taken.
   To map addr --> function_name, 
   you can use "nm"
           eg:  nm <bin_file> | grep "<addr>"


Limitations:
-------------------------------------------------------------------------------------
 - Does not work with recursive functions.
 - Profiles ALL the functions. You cannot selectively choose which functions to profile
   and which to leave alone.
 - Cannot use with multiple threads.

Advantages:
-------------------------------------------------------------------------------------
 - The profiler code itself is not accounted for anywhere. OK that was a bit bold. Some neglible stuff from the profiler does get accounted :-)
   But I've tried to keep it to a minimum.
 - The user does not need to modify the existing code for profiling (except the Makefile).


TODO:
-------------------------------------------------------------------------------------
 - Support recursive functions.
 - Give users the facility to define "profile points" wherein the user will be able to start the profiling from that point on upto a certain point in future.
 - Better output !!
 - Misc optimizations / cleanup.
 - Improve README


How it works:
-------------------------------------------------------------------------------------
GCC provides a compile-time option "-finstrument-functions" which basically adds some piece of code at the start and exit of each function that essentially calls "__cyg_profile_func_enter()" just after entry into the function and "__cyg_profile_exit()" just before unwinding the function.

So in __cyg_profile_enter, we start tracking the time for a function, that can be identified via the arguments passed. We also "stop" accounting for the "self_time" of the parent function and instead start tracking the "child_time" of the parent function.

In __cyg_profile_exit, we account for how many ticks the current function has taken, i.e. the "self_time" of the current function. We also add this time to the "child_time" of the parent function. And we start tracking the parent function's "self_time" again.

This gives us 2 metrics:
     - self ticks: The ticks taken by the function in question only. It does not include any ticks that are taken by any of it's children.
     - total ticks: total ticks = self_ticks + child_ticks. This gives the total time for which the frame for this function was on the stack.
      

Changelog:
-------------------------------------------------------------------------------------
2012.02.26:
  -- Version 0.1