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]

Re: ICE, OSF/1, egcs-19990927


On Thu, Sep 30, 1999 at 01:31:48PM -0700, kargl@pag.apl.washington.edu wrote:
> DragBS.c:581: Internal compiler error in `fixup_var_refs_1',
> at function.c:1749

For the curious, this happened because 

(insn 258 256 260 (set (reg:DI 126)
        (and:DI (plus:DI (addressof:DI (mem/s:SI (plus:DI (reg:DI 63 FP)
                            (const_int 144 [0x90])) 0) 113 0x200000aed00)
                (const_int 1 [0x1]))
            (const_int -8 [0xfffffffffffffff8]))) -1 (nil)
    (nil))

was broken out from a MEM because the address was deemed invalid. 
I didn't look to see how force_operand (or whoever) could be 
changed to accomodate such weird addressing modes, because we
actually don't want the address broken out.

So a small tweek to one of GO_IF_LEGITIMATE_ADDRESS's subroutines
allows ADDRESSOF to be a base of an ANDed address.  With the change,
purge_addressof successfully substitutes a reg for the ADDRESSOF
and compilation proceeds apace.


r~


	* alpha.h (GO_IF_LEGITIMATE_SIMPLE_ADDRESS): Accept addressof
	as the base of a PLUS.

Index: alpha.h
===================================================================
RCS file: /cvs/gcc/egcs/gcc/config/alpha/alpha.h,v
retrieving revision 1.79
diff -c -p -d -r1.79 alpha.h
*** alpha.h	1999/10/14 04:20:55	1.79
--- alpha.h	1999/10/15 05:39:09
*************** do {						\
*** 1452,1457 ****
--- 1452,1460 ----
  	      && CONSTANT_ADDRESS_P (XEXP (X, 1)))			\
  	    goto ADDR;							\
  	}								\
+       else if (GET_CODE (tmp) == ADDRESSOF				\
+ 	       && CONSTANT_ADDRESS_P (XEXP (X, 1)))			\
+ 	goto ADDR;							\
      }									\
  }
  


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