Re: [PATCH] Add prototype for ssa_fast_dce, add dump output for ssa-ccp

[Daniel Berlin <dan at cgsoftware dot com>]

law@redhat.com writes:

>   In message <8766d1pw5l.fsf@cgsoftware.com>you write:
>  > ssa-ccp was missing a prototype for ssa_fast_dce.
>  > And it never said what it was doing when it removed edges/set
>  > registers constant, in the dump file.
>  > 
>  > (It's also missing a prototype in ssa.h, but i'll send that in a sec).
>  > 
>  > 2001-07-09  Daniel Berlin  <dan@cgsoftware.com>
>  > 
>  > 	* ssa-ccp.c: Add prototype for ssa_fast_dce.
>  > 	(optimize_unexecutable_edges): Add note about removal
>  > 	of edge, and removal of phi alternative,  to dump file. 
>  > 	(ssa_ccp_substitute_constants): Add note about register now being
>  > 	constant, and which uses were replaced in what insns, to dump file.
> You really need to be bootstrapping and testing your changes like we
> require everyone to do.  I'm not going to make a habit of bootstrapping
> changes from contributors who should know better.
Errr, I did.
I submitted the patch *before* someone fixed it (check the archives),
so it bootstrapped fine.

> 
> You also need to pay more attention to our coding standards.  Similarly,
> I won't make a habit of fixing coding standards issues for regular
> contributors.

My apologies, i'll try to be more careful in the future.


> 
>  > --- 756,770 ----
>  >   	      while (PHI_NODE_P (insn))
>  >   		{
>  >   		  remove_phi_alternative (PATTERN (insn), edge->src);
>  > + 		  if (rtl_dump_file)
>  > + 		    fprintf (rtl_dump_file, "Removing alternative for bb %d of 
>  > phi %d\n", 
> Bad line wrap here.
> 
>  > + 			     edge->src->index, SSA_NAME (PATTERN (insn)));
>  >   		  insn = NEXT_INSN (insn);
>  >   		}
>  >   	    }
>  > ! 	  if (rtl_dump_file)
>  > ! 	    fprintf (rtl_dump_file, "Removing edge from %d to %d because it is 
>  > not executable\n",
> And here.
> 
> 
>  > *************** ssa_ccp_substitute_constants ()
>  > *** 859,864 ****
>  > --- 864,872 ----
>  >   	     are consecutive at the start of the basic block.  */
>  >   	  if (! PHI_NODE_P (def))
>  >   	    {
>  > + 	      if (rtl_dump_file)
>  > + 		fprintf (rtl_dump_file, "Register %d is now set to a constant\n
>  > ", SSA_NAME (PATTERN (def)));
> And here.
> 
> 
>  > *************** ssa_ccp_substitute_constants ()
>  > *** 882,894 ****
>  >   		  && (GET_CODE (useinsn) == INSN
>  >   		      || GET_CODE (useinsn) == JUMP_INSN))
>  >   		{
>  > ! 		  validate_replace_src (regno_reg_rtx [i],
>  >   					values[i].const_value,
>  > ! 					useinsn);
>  >   		  INSN_CODE (useinsn) = -1;
>  >   		  df_insn_modify (df_analyzer,
>  >   				  BLOCK_FOR_INSN (useinsn),
>  >   				  useinsn);
>  >   		}
>  >   
>  >   	    }
>  > --- 890,910 ----
>  >   		  && (GET_CODE (useinsn) == INSN
>  >   		      || GET_CODE (useinsn) == JUMP_INSN))
>  >   		{
>  > ! 		  
>  > ! 		  if (validate_replace_src (regno_reg_rtx [i],
>  >   					    values[i].const_value,
>  > ! 					    useinsn))
>  > ! 		    {
>  > ! 		      if (rtl_dump_file)
>  > ! 			fprintf(rtl_dump_file, 
>  > ! 				"Use of register %d in insn %d replace with con
>  > stant value\n",
>  > ! 				i, INSN_UID (useinsn));
>  >   		      INSN_CODE (useinsn) = -1;
>  >   		      df_insn_modify (df_analyzer,
>  >   				      BLOCK_FOR_INSN (useinsn),
>  >   				      useinsn);
>  > + 		    }
> Line wrapping problems, you forgot to reindent code after inserting a new
> level of braces, you're missing whitespace between the fprintf and the
> paren to start its parameter list.
> 
> 
> I've fixed things this time, but don't expect me to do so on a regular
> basis.
> 
> Bootstrapped x86-linux and installed, except for the change which
> added a prototype for ssa_fast_dce which was fixed by someone else.
> 
> jeff
> 

-- 
"Today I was arrested for scalping low numbers at the deli.  I
sold a #3 for 28 bucks.
"-Steven Wright