[Bug middle-end/42794] New: PRE produces illegal PHI node

["sergei_lus at yahoo dot com" <gcc-bugzilla at gcc dot gnu dot org>]

The following (reduced) code:

typedef enum {
    TYPE_NON_IDR,
    TYPE_IDR,
} NAL_UNIT_TYPE;
typedef struct recordTag
{
} Record;
typedef struct {
    unsigned int ActualSize;
    unsigned short *Slice;
}Info;

typedef struct {
} Params;

typedef struct {
  NAL_UNIT_TYPE unit_type;
} NAL_UNIT;

unsigned int foo( Info *Decode, unsigned int nal_len)
{
    NAL_UNIT nal_unit;
    unsigned short *Backend;
    unsigned char complete;
    int *BufLen = (int *)&Decode->ActualSize;

    do{
        *BufLen = *BufLen - nal_len;
        if (((nal_unit.unit_type) == TYPE_NON_IDR))
        {
            Decode->Slice = Backend;
        }
    }while (*BufLen >0);

    Finish( &complete);
}

Produces infinite loop with this options: -O2 
>gcc -v
Using built-in specs.
Target: x86_64-unknown-linux-gnu
Configured with: ../configure --enable-threads=posix
--prefix=/x/x86_gcc_4.4/bin --enable-languages=c,c++ --disable-checking
Thread model: posix
gcc version 4.4.0 (GCC)

The issue first appears in tree PRE (from reduced2.c.084t.pre):
...
<bb 2>:
  D.1896_2 = &Decode_1(D)->ActualSize;
  BufLen_3 = (int *) D.1896_2;
  pretmp.16_34 = Decode_1(D)->ActualSize;

<bb 3>:
  # prephitmp.17_35 = PHI <pretmp.16_34(2), D.1905_12(8)>
  D.1907_14 = prephitmp.17_35;
  D.1899_7 = D.1907_14 - nal_len_6(D);
  D.1900_8 = (int) D.1899_7;
  *BufLen_3 = D.1900_8;
  if (nal_unit$unit_type_16(D) == 0)
    goto <bb 4>;
  else
    goto <bb 7>;

<bb 7>:
  goto <bb 5>;

<bb 4>:
  Decode_1(D)->Slice = Backend_10(D);
  pretmp.16_36 = Decode_1(D)->ActualSize;

<bb 5>:
  # prephitmp.17_37 = PHI <prephitmp.17_35(7), pretmp.16_36(4)>
  D.1905_12 = prephitmp.17_37;
  D.1906_13 = (int) D.1905_12;
  if (D.1906_13 > 0)
    goto <bb 8>;
  else
    goto <bb 6>;


The first "argument" of this PHI: # prephitmp.17_37 = PHI <prephitmp.17_35(7),
pretmp.16_36(4)> is wrong - instead of decremented value it uses original one.

PRE only makes an earlier DF analysis issue evident. The problem is elsewhere.
Any suggestions are highly appreciated.


-- 
           Summary: PRE produces illegal PHI node
           Product: gcc
           Version: 4.4.0
            Status: UNCONFIRMED
          Severity: major
          Priority: P3
         Component: middle-end
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: sergei_lus at yahoo dot com


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42794