Re: basic-block profiling

[Saisanthosh Balakrishnan <sai at cs dot wisc dot edu>]

Following up on my previous email, I just now saw an entry in
Changelog:

<quote>
2002-05-21  Richard Henderson  <rth@redhat.com>

        * system.h (BLOCK_PROFILER, BLOCK_PROFILER_CODE,
        FUNCTION_BLOCK_PROFILER, FUNCTION_BLOCK_PROFILER_EXIT,
        MACHINE_STATE_SAVE, MACHINE_STATE_RESTORE): Poison.

        * toplev.c (display_help): Kill -a -ax help.
....
</quote>


Why was the code considered "poisonous"? Any other ways I could have
basic-block profiling?

-Sai


> I've been using gcc's (3.0.3, Linux, x86) -a (basic-block profiling)
> to get the execution frequency counts of basic blocks. "-ax", a
> documented flag, that is supposed to provide jump address and bb
> traces never worked for me. I get the following error with gcc -ax
> simple.c:
> 
> simple.c is the first example program in gcov's info page.
> simple.c: In function `main':
> simple.c:14: Unrecognizable insn:
> (call_insn 92 91 93 (call (mem:QI (symbol_ref ("__bb_trace_ret")) 0)
>         (const_int 0 [0x0])) -1 (nil)
> 		    (nil)
> 			    (nil))
> simple.c:14: Internal compiler error in insn_default_length, at
> insn-attrtab.c:223
> Please submit a full bug report,
> with preprocessed source if appropriate.
> See <URL:http://www.gnu.org/software/gcc/bugs.html> for instructions.
> 
> I assumed that the support for basic-block profiling was in initial
> stages in gcc-3.0.3, and downloaded gcc-3.2. I was surprised, when gcc
> reported -a and -ax as unsupported flags. I did not see any references
> to these flags in toplev.c (libgcc2.c etc.). However, the flags are
> documented in gcc-3.2 info page! Could the authors of profiling
> (and/or basic-block profiling), respond as to what's going on, what I
> could expect to see in future releases of gcc?
> 
> Thanks in advance.
> Sai
>