🚀 go-pugleaf

A bug report from James Clark <jjc@jclark.com> reports that GCC 2.1
misoptimizes this program:
	main()
	{
	  signed char c = -1;
	  printf("%d\n", c < 0);
	  exit(0);
	}
It prints zero instead of one.

I've found this bug.  The optimizer is noting that "c" equals 0xFF,
but it is forgetting that 0xFF is a _byte_ value (QImode).  So when
the optimizer SIGN_EXTENDs that constant, the constant looks like it
has already been extended.  This bug results in a wrong value of 255
(which is, of course, not less than zero) instead of the correct -1.

The below patch makes GCC generate and retain the correct machine
modes for constants.

----------------------------------------------------------------------------

Thu Apr  9 17:22:19 1992  Chip Salzenberg  (chip at tct.com)

	* emit-rtl.c (gen_rtx): Generate CONST_INT with requested MODE.
	  (gen_lowpart_common): Ditto.

Index: emit-rtl.c
*************** gen_rtx (va_alist)
*** 257,260 ****
--- 257,261 ----
  
        rt_val = rtx_alloc (code);
+       rt_val->mode = mode;
        INTVAL (rt_val) = arg;
      }
*************** gen_lowpart_common (mode, x)
*** 533,537 ****
  	    : (GET_MODE_BITSIZE (mode) == HOST_BITS_PER_INT
  	       && GET_CODE (x) == CONST_INT) ? x
! 	    : gen_rtx (CONST_INT, VOIDmode,
  		       (GET_MODE_MASK (mode)
  			& (GET_CODE (x) == CONST_INT
--- 534,538 ----
  	    : (GET_MODE_BITSIZE (mode) == HOST_BITS_PER_INT
  	       && GET_CODE (x) == CONST_INT) ? x
! 	    : gen_rtx (CONST_INT, mode,
  		       (GET_MODE_MASK (mode)
  			& (GET_CODE (x) == CONST_INT
-- 
Chip Salzenberg at Teltronics/TCT  <chip@tct.com>, <73717.366@compuserve.com>
      "Informix 4GL is not a 4G, and it's barely an L."  -- John Tombs