Re: giv combination patch

[Toon Moene <toon at moene dot indiv dot nluug dot nl>]

I wrote:

> >     1. Combining related General Induction Variables.

RTH:

>  The following code is a first step along this path.  It
>  works correctly for givs related by a constant as below,
>  or a simple constant as with higher order ranks of
>  staticly dimensioned arrays.

Unfortunately, trying this patch against egcs-971207, it doesn't  
work, over here:

stage1/xgcc -Bstage1/  -DIN_GCC    -O2 -g  -DHAVE_CONFIG_H   -o  
genrecog \
 genrecog.o rtl.o ` case "obstack.o" in ?*) echo obstack.o ;; esac  
` ` case "" in ?*) echo  ;; esac ` ` case "" in ?*) echo  ;; esac `
./genrecog ./config/m68k/m68k.md > tmp-recog.c
/bin/sh: 25584 Memory fault
make[1]: *** [stamp-recog] Error 139
make[1]: Leaving directory `/Users/toon/Unix/compilers/egcs-971207/gcc'
make: *** [bootstrap] Error 2

I've tried whatever-you-can-think-of to get some information out of  
this, but gdb reports:

Program generated(1): Memory access exception on address 0xfed3d594  
(invalid address).
Reading in symbols for ./genrecog.c...done.
add_to_sequence (pattern=0x20222, last=0x3fffad0,  
position=0x1098<Address 0x1098 out of bounds>) at ./genrecog.c:391
391                     if (preds[i].codes[1] == 0 && new->code ==  
UNKNOWN)
(gdb) p i
$1 = 9
(gdb) p new
$2 = (struct decision *) 0x2526c
(gdb) p new->code
$3 = UNKNOWN
(gdb) p preds[i].codes[1]
$5 = UNKNOWN
(gdb) list
386                     int j;
387                     int allows_const_int = 0;
388
389                     new->pred = i;
390
391                     if (preds[i].codes[1] == 0 && new->code ==  
UNKNOWN)
392                       {
393                         new->code = preds[i].codes[0];
394                         if (! strcmp ("const_int_operand",  
new->tests))
395                           new->tests = 0, new->pred = -1;

... etc ...

and I can't get any subexpression of line 391 generating a  
segmentation fault, i.e. nothing revealing.

HTH,
Toon.