This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: PATCH for CSE
- To: law at cygnus dot com
- Subject: Re: PATCH for CSE
- From: Mark Mitchell <mark at markmitchell dot com>
- Date: Thu, 2 Jul 1998 12:40:30 -0700
- Cc: egcs-patches at cygnus dot com
- References: <24533.899364681@hurl.cygnus.com>
- Reply-to: mark at markmitchell dot com
>>>>> "Jeffrey" == Jeffrey A Law <law@hurl.cygnus.com> writes:
Jeffrey> I added it to the testsuite last night :-)
Jeffrey> execute/980701-1.c
Jeff --
Here's a patch for this bug. With this patch, I renew my petition
:-) for my earlier CSE_ADDRESS_COST patch to favor the use of
ADDRESSOFs.
Thanks,
--
Mark Mitchell mark@markmitchell.com
Mark Mitchell Consulting http://www.markmitchell.com
Thu Jul 2 12:35:50 1998 Mark Mitchell <mark@markmitchell.com>
* cse.c (cse_insn): When SETting (MEM (ADDRESSOF (X))) to Y,
don't claim that the former is equivalent to the latter.
Index: cse.c
===================================================================
RCS file: /egcs/carton/cvsfiles/egcs/gcc/cse.c,v
retrieving revision 1.38
diff -c -p -r1.38 cse.c
*** cse.c 1998/07/01 05:10:47 1.38
--- cse.c 1998/07/02 19:33:07
*************** cse_insn (insn, in_libcall_block)
*** 7405,7410 ****
--- 7417,7423 ----
if (sets[i].rtl)
{
register rtx dest = SET_DEST (sets[i].rtl);
+ rtx inner_dest = sets[i].inner_dest;
register struct table_elt *elt;
/* Don't record value if we are not supposed to risk allocating
*************** cse_insn (insn, in_libcall_block)
*** 7452,7460 ****
rehash_using_reg (dest);
sets[i].dest_hash = HASH (dest, GET_MODE (dest));
}
- elt = insert (dest, sets[i].src_elt,
- sets[i].dest_hash, GET_MODE (dest));
elt->in_memory = (GET_CODE (sets[i].inner_dest) == MEM
&& (! RTX_UNCHANGING_P (sets[i].inner_dest)
|| FIXED_BASE_PLUS_P (XEXP (sets[i].inner_dest,
--- 7465,7483 ----
rehash_using_reg (dest);
sets[i].dest_hash = HASH (dest, GET_MODE (dest));
}
+
+ if (GET_CODE (inner_dest) == MEM
+ && GET_CODE (XEXP (inner_dest, 0)) == ADDRESSOF)
+ /* Given (SET (MEM (ADDRESSOF (X))) Y) we don't want to say
+ that (MEM (ADDRESSOF (X))) is equivalent to Y.
+ Consider the case in which the address of the MEM is
+ passed to a function, which alters the MEM. Then, if we
+ later use Y instead of the MEM we'll miss the update. */
+ elt = insert (dest, 0, sets[i].dest_hash, GET_MODE (dest));
+ else
+ elt = insert (dest, sets[i].src_elt,
+ sets[i].dest_hash, GET_MODE (dest));
elt->in_memory = (GET_CODE (sets[i].inner_dest) == MEM
&& (! RTX_UNCHANGING_P (sets[i].inner_dest)
|| FIXED_BASE_PLUS_P (XEXP (sets[i].inner_dest,