This is the mail archive of the gcc@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: Dead code at gcc/tree-ssa-loop.c:772?


On 6/6/19 6:20 AM, Martin Liška wrote:
Hi.

The code is dead:

    757	char *
    758	get_lsm_tmp_name (tree ref, unsigned n, const char *suffix)
    759	{
    760	  char ns[2];
    761	
    762	  lsm_tmp_name_length = 0;
    763	  gen_lsm_tmp_name (ref);
    764	  lsm_tmp_name_add ("_lsm");
    765	  if (n < 10)
    766	    {
    767	      ns[0] = '0' + n;
    768	      ns[1] = 0;
    769	      lsm_tmp_name_add (ns);
    770	    }
    771	  return lsm_tmp_name;
    772	  if (suffix != NULL)
    773	    lsm_tmp_name_add (suffix);
    774	}

Andrew is it a typo or an issue?
Thanks,
Martin

Dunno. It was written in 2005.
2005-08-16  Zdenek Dvorak  <dvorakz@suse.cz>

        * tree-ssa-loop-im.c (MAX_LSM_NAME_LENGTH, lsm_tmp_name,
        lsm_tmp_name_length): New.
        (lsm_tmp_name_add, gen_lsm_tmp_name, get_lsm_tmp_name): New functions.
        (schedule_sm): Use get_lsm_tmp_name instead of "lsm_tmp".

The whole thing is a little odd since you cant get more than 10 tmp names without suddenly all being the same name.

I dont know anything about the code, my guess is the return should be after the 'if'.  the only callers appears to pass ~0 as the value for N. execute_sm_if_changed_flag_set()  adds '_flag' as a suffix and execute_sm()  calls it without the suffix.

My guess is the return should be moved to the bottom so that those 2 get different names, so it could be a problem as it is.   Someone who know the loop code better could comment..

Andrew


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]