This is the mail archive of the gcc-patches@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]

[PATCH] Fix PR 24823/flow


Hello,

the problem described in the PR occurs because PARALLELs aren't
correctly handled as destinations of a SET in mark_used_regs.

The attached patch moves the code from the "case SET:" branch of
mark_used_regs into a separate function and calls it for every 
part of a PARALLEL if necessary.

This patch mimics the former behavior of falling through to the 
code after the switch statement although I think that this is 
another bug. If the SET case is not finished with a return, all
destinations not handled in the "case SET:" branch are passed to
other calls of mark_used_regs and therefore are marked used.
I've tried to simply place a return at the end of the "case SET:"
but got a few regressions which I'ven't tracked down yet.

I've bootstrapped that patch on s390, s390x and i686.
No testsuite regressions.

It fixes the PR testcase. The testcase is fortran code. I will add it
to the testsuite with a separated patch if someone thinks it is worth 
doing.

OK for gcc 4.1 and 4.2 branches?

Bye,

-Andreas-


2005-11-25  Andreas Krebbel  <krebbel1@de.ibm.com>

	PR 24823
	* flow.c (mark_used_dest_regs): New function.
	(mark_used_regs): Call mark_used_dest_regs.


Index: gcc/flow.c
===================================================================
*** gcc/flow.c.orig	2005-11-24 10:15:02.000000000 +0100
--- gcc/flow.c	2005-11-25 16:50:59.000000000 +0100
*************** mark_used_reg (struct propagate_block_in
*** 3857,3862 ****
--- 3857,3945 ----
      }
  }
  
+ /* Scan expression X for registers which have to be marked used in PBI.  
+    X is considered to be the SET_DEST rtx of SET.  TRUE is returned if
+    X could be handled by this function.  */
+ 
+ static bool
+ mark_used_dest_regs (struct propagate_block_info *pbi, rtx x, rtx cond, rtx insn)
+ {
+   int regno;
+   bool mark_dest = false;
+   rtx dest = x;
+   
+   /* On some platforms calls return values spread over several 
+      locations.  These locations are wrapped in a EXPR_LIST rtx
+      together with a CONST_INT offset.  */
+   if (GET_CODE (x) == EXPR_LIST
+       && GET_CODE (XEXP (x, 1)) == CONST_INT)
+     x = XEXP (x, 0);
+   
+   /* If storing into MEM, don't show it as being used.  But do
+      show the address as being used.  */
+   if (MEM_P (x))
+     {
+ #ifdef AUTO_INC_DEC
+       if (pbi->flags & PROP_AUTOINC)
+ 	find_auto_inc (pbi, x, insn);
+ #endif
+       mark_used_regs (pbi, XEXP (x, 0), cond, insn);
+       return true;
+     }
+ 	    
+   /* Storing in STRICT_LOW_PART is like storing in a reg
+      in that this SET might be dead, so ignore it in TESTREG.
+      but in some other ways it is like using the reg.
+      
+      Storing in a SUBREG or a bit field is like storing the entire
+      register in that if the register's value is not used
+ 	       then this SET is not needed.  */
+   while (GET_CODE (x) == STRICT_LOW_PART
+ 	 || GET_CODE (x) == ZERO_EXTRACT
+ 	 || GET_CODE (x) == SUBREG)
+     {
+ #ifdef CANNOT_CHANGE_MODE_CLASS
+       if ((pbi->flags & PROP_REG_INFO) && GET_CODE (x) == SUBREG)
+ 	record_subregs_of_mode (x);
+ #endif
+       
+       /* Modifying a single register in an alternate mode
+ 	 does not use any of the old value.  But these other
+ 	 ways of storing in a register do use the old value.  */
+       if (GET_CODE (x) == SUBREG
+ 	  && !((REG_BYTES (SUBREG_REG (x))
+ 		+ UNITS_PER_WORD - 1) / UNITS_PER_WORD
+ 	       > (REG_BYTES (x)
+ 		  + UNITS_PER_WORD - 1) / UNITS_PER_WORD))
+ 	;
+       else
+ 	mark_dest = true;
+       
+       x = XEXP (x, 0);
+     }
+   
+   /* If this is a store into a register or group of registers,
+      recursively scan the value being stored.  */
+   if (REG_P (x)	&& 
+       (regno = REGNO (x),
+        !(regno == FRAME_POINTER_REGNUM
+ 	  && (!reload_completed || frame_pointer_needed)))
+ #if FRAME_POINTER_REGNUM != HARD_FRAME_POINTER_REGNUM
+       && !(regno == HARD_FRAME_POINTER_REGNUM
+ 	    && (!reload_completed || frame_pointer_needed))
+ #endif
+ #if FRAME_POINTER_REGNUM != ARG_POINTER_REGNUM
+       && !(regno == ARG_POINTER_REGNUM && fixed_regs[regno])
+ #endif
+ 	  )
+     {
+       if (mark_dest)
+ 	mark_used_regs (pbi, dest, cond, insn);
+       return true;
+     }
+   return false;
+ }
+ 
  /* Scan expression X and store a 1-bit in NEW_LIVE for each reg it uses.
     This is done assuming the registers needed from X are those that
     have 1-bits in PBI->REG_LIVE.
*************** static void
*** 3868,3874 ****
  mark_used_regs (struct propagate_block_info *pbi, rtx x, rtx cond, rtx insn)
  {
    RTX_CODE code;
-   int regno;
    int flags = pbi->flags;
  
   retry:
--- 3951,3956 ----
*************** mark_used_regs (struct propagate_block_i
*** 3969,4041 ****
  
      case SET:
        {
! 	rtx testreg = SET_DEST (x);
! 	int mark_dest = 0;
! 
! 	/* If storing into MEM, don't show it as being used.  But do
! 	   show the address as being used.  */
! 	if (MEM_P (testreg))
! 	  {
! #ifdef AUTO_INC_DEC
! 	    if (flags & PROP_AUTOINC)
! 	      find_auto_inc (pbi, testreg, insn);
! #endif
! 	    mark_used_regs (pbi, XEXP (testreg, 0), cond, insn);
! 	    mark_used_regs (pbi, SET_SRC (x), cond, insn);
! 	    return;
! 	  }
! 
! 	/* Storing in STRICT_LOW_PART is like storing in a reg
! 	   in that this SET might be dead, so ignore it in TESTREG.
! 	   but in some other ways it is like using the reg.
! 
! 	   Storing in a SUBREG or a bit field is like storing the entire
! 	   register in that if the register's value is not used
! 	   then this SET is not needed.  */
! 	while (GET_CODE (testreg) == STRICT_LOW_PART
! 	       || GET_CODE (testreg) == ZERO_EXTRACT
! 	       || GET_CODE (testreg) == SUBREG)
! 	  {
! #ifdef CANNOT_CHANGE_MODE_CLASS
! 	    if ((flags & PROP_REG_INFO) && GET_CODE (testreg) == SUBREG)
! 	      record_subregs_of_mode (testreg);
! #endif
! 
! 	    /* Modifying a single register in an alternate mode
! 	       does not use any of the old value.  But these other
! 	       ways of storing in a register do use the old value.  */
! 	    if (GET_CODE (testreg) == SUBREG
! 		&& !((REG_BYTES (SUBREG_REG (testreg))
! 		      + UNITS_PER_WORD - 1) / UNITS_PER_WORD
! 		     > (REG_BYTES (testreg)
! 			+ UNITS_PER_WORD - 1) / UNITS_PER_WORD))
! 	      ;
! 	    else
! 	      mark_dest = 1;
! 
! 	    testreg = XEXP (testreg, 0);
! 	  }
! 
! 	/* If this is a store into a register or group of registers,
! 	   recursively scan the value being stored.  */
! 
! 	if ((GET_CODE (testreg) == PARALLEL
! 	     && GET_MODE (testreg) == BLKmode)
! 	    || (REG_P (testreg)
! 		&& (regno = REGNO (testreg),
! 		    ! (regno == FRAME_POINTER_REGNUM
! 		       && (! reload_completed || frame_pointer_needed)))
! #if FRAME_POINTER_REGNUM != HARD_FRAME_POINTER_REGNUM
! 		&& ! (regno == HARD_FRAME_POINTER_REGNUM
! 		      && (! reload_completed || frame_pointer_needed))
! #endif
! #if FRAME_POINTER_REGNUM != ARG_POINTER_REGNUM
! 		&& ! (regno == ARG_POINTER_REGNUM && fixed_regs[regno])
! #endif
! 		))
  	  {
- 	    if (mark_dest)
- 	      mark_used_regs (pbi, SET_DEST (x), cond, insn);
  	    mark_used_regs (pbi, SET_SRC (x), cond, insn);
  	    return;
  	  }
--- 4051,4068 ----
  
      case SET:
        {
! 	rtx dest = SET_DEST (x);
! 	int i;
! 	bool ret = false;
! 
! 	if (GET_CODE (dest) == PARALLEL)
! 	  for (i = 0; i < XVECLEN (dest, 0); i++)
! 	    ret |= mark_used_dest_regs (pbi, XVECEXP (dest, 0 ,i), cond, insn);
! 	else
! 	  ret = mark_used_dest_regs (pbi, dest, cond, insn);
! 	
! 	if (ret)
  	  {
  	    mark_used_regs (pbi, SET_SRC (x), cond, insn);
  	    return;
  	  }


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