Re: GCC build of HEAD failed for native with your patch on 2004-08-12T11:01:41Z.

[Geoffrey Keating <geoffk at apple dot com>]

This appears to be due to Zack's patch to the gen* code.  It reproduces  
on a cross to powerpc-eabisim, and the problem is that recog() starts  
with:


int
recog (rtx x0 ATTRIBUTE_UNUSED,
        rtx insn ATTRIBUTE_UNUSED,
        int *pnum_clobbers ATTRIBUTE_UNUSED)
{
  rtx * const operands ATTRIBUTE_UNUSED = &recog_data.operand[0];
  rtx x1 ATTRIBUTE_UNUSED;
  rtx x2 ATTRIBUTE_UNUSED;
  rtx x3 ATTRIBUTE_UNUSED;
  rtx x4 ATTRIBUTE_UNUSED;
  rtx x5 ATTRIBUTE_UNUSED;
  rtx x6 ATTRIBUTE_UNUSED;
  rtx x7 ATTRIBUTE_UNUSED;
  int tem ATTRIBUTE_UNUSED;
  recog_data.insn = NULL_RTX;

  switch (GET_CODE (x0))
    {
    case SET:
      goto L1;
    case PARALLEL:
      goto L14308;
    case UNSPEC_VOLATILE:
      goto L14333;
    case RETURN:
      goto L14334;
    case CONST_INT:
      goto L14335;
    case TRAP_IF:
      goto L7594;
    case PREFETCH:
      goto L7710;
    case UNSPEC:
      goto L14343;
    default:
     break;
   }
 L7622: ATTRIBUTE_UNUSED_LABEL
  if (XVECLEN (x0, 0) >= 3
      && any_operand (x0, VOIDmode))
    {
      operands[0] = x0;
      goto L7623;
    }

but after the default case in the switch is hit, there's no guarantee  
that x0 is a vector of any kind, and in fact it usually isn't; so this  
was crashing trying to recognize

(insn:HI 44 41 49 0  
/Network/Servers/hills/Volumes/capanna/gkeating/co/egcs-mainline/gcc/ 
gcc/libgcc2.c:466 (use (reg/i:DI 3 3 [ <result> ])) -1  
(insn_list:REG_DEP_ANTI 47 (insn_list:REG_DEP_ANTI 6  
(insn_list:REG_DEP_ANTI 7 (insn_list:REG_DEP_ANTI 20  
(insn_list:REG_DEP_ANTI 22 (insn_list:REG_DEP_ANTI 24  
(insn_list:REG_DEP_ANTI 25 (insn_list:REG_DEP_ANTI 27  
(insn_list:REG_DEP_ANTI 28 (insn_list:REG_DEP_ANTI 30  
(insn_list:REG_DEP_ANTI 31 (insn_list:REG_DEP_ANTI 32  
(insn_list:REG_DEP_ANTI 34 (insn_list 41 (nil)))))))))))))))

It appears to have lost the code that implemented this bit from  
gccint.info:

`(match_parallel N PREDICATE [SUBPAT...])'
...
     When matching an insn, it matches if the body of the
     insn is a `parallel' expression with at least as many elements as
     the vector of SUBPAT expressions in the `match_parallel', if each
     SUBPAT matches the corresponding element of the `parallel', _and_
     the function PREDICATE returns nonzero on the `parallel' that is
     the body of the insn.

note the part about requiring the body to be a PARALLEL.

--
Geoff Keating <geoffk@apple.com>

Attachment: smime.p7s
Description: S/MIME cryptographic signature