patches/gcc/3.4.6/180-pr15068-fix.patch
author "Yann E. MORIN" <yann.morin.1998@anciens.enib.fr>
Mon Mar 28 01:05:18 2011 +0200 (2011-03-28)
changeset 2362 0888065f8c4d
parent 339 bd5e0a849352
permissions -rw-r--r--
cc/gcc: cleanup the _or_later logic

So far, we've had a version always select appropriate _or_later option,
which in turn would select all previous _or_later options.

Because the dependencies on companion libs were cumulative, that was
working OK. But the upcoming 4.6 will no longer depend on libelf, so
we can't keep the cumulative scheme we've been using so far.

Have each release family select the corresponding dependencies, instead
of relying on selecting previous _or_later.

Signed-off-by: "Yann E. MORIN" <yann.morin.1998@anciens.enib.fr>
     1 diff -durN gcc-3.4.6.orig/gcc/flow.c gcc-3.4.6/gcc/flow.c
     2 --- gcc-3.4.6.orig/gcc/flow.c	2005-09-01 22:51:09.000000000 +0200
     3 +++ gcc-3.4.6/gcc/flow.c	2007-08-15 23:00:30.000000000 +0200
     4 @@ -1884,6 +1884,7 @@
     5  	  rtx set_src = SET_SRC (pc_set (BB_END (bb)));
     6  	  rtx cond_true = XEXP (set_src, 0);
     7  	  rtx reg = XEXP (cond_true, 0);
     8 + 	  enum rtx_code inv_cond;
     9  
    10  	  if (GET_CODE (reg) == SUBREG)
    11  	    reg = SUBREG_REG (reg);
    12 @@ -1892,11 +1893,13 @@
    13  	     in the form of a comparison of a register against zero.  
    14  	     If the condition is more complex than that, then it is safe
    15  	     not to record any information.  */
    16 -	  if (GET_CODE (reg) == REG
    17 + 	  inv_cond = reversed_comparison_code (cond_true, BB_END (bb));
    18 + 	  if (inv_cond != UNKNOWN
    19 +	      && GET_CODE (reg) == REG
    20  	      && XEXP (cond_true, 1) == const0_rtx)
    21  	    {
    22  	      rtx cond_false
    23 -		= gen_rtx_fmt_ee (reverse_condition (GET_CODE (cond_true)),
    24 +		= gen_rtx_fmt_ee (inv_cond,
    25  				  GET_MODE (cond_true), XEXP (cond_true, 0),
    26  				  XEXP (cond_true, 1));
    27  	      if (GET_CODE (XEXP (set_src, 1)) == PC)