Re: gprof 2.14 glitches w/ powerpc elf binaries

[Nick Clifton]

Hi Nju,

> We did not use gprof in the conventional manner. We captured an assembly code
> trace of the program  running in standalone mode on a ppc405 processor using a
> Agilent's/IBM trace port analyzer. I wrote a Perl script that converts the trace
> file into a gmon.out format. Someone else here at Xilinx wrote a similar
> application that converted a trace file (of a different format) to a gmon.out
> format. We both found out that gprof tacks on functions that were never 
> executed.
> Each address had its own histogram bin, so there can't be any issues of crossing
> function boundaries in the bins. I have attached the .elf binary file and the
> gmon.out file (to be used on a little endian machine like a linux x86). treck 
> elf
> never uses functions like memcpy or Letext, which appear in gprof's histogram 
> and
> c-g tables.

Have you tried running gprof with the -D command line switch ?  I think 
that this will get fix the spurious references to Letext for you.

As for the references to memcpy - I think that these really do happen. 
It appears to be called from the code inside xil_printf.c.  It may be 
that the code there uses the "=" operator to perform a structure 
assignment.  The compiler will sometimes translate this into a call to 
memcpy.  eg:

   typedef struct { int a; int b; int c; } s;
   s * ptr1, * ptr2;

   * ptr1 = * ptr2;

Cheers
  Nick