Re: linux + usemorebits: gcc error?

[Hugo van der Sanden <hv at crypt dot compulink dot co dot uk>]

Earlier I told p5p:
:Build and make test worked fine with normal options, but adding
:-Dusemorebits gave me:
:          CCCMD =  cc -DPERL_CORE -c -DDEBUGGING -fno-strict-aliasing -D_LARGEFILE_SOURCE -D_FILE_OFFSET_BITS=64 -I/usr/include/gdbm -g -O6 -Wall
:perlio.c: In function `PerlIORaw_pushed':
:perlio.c:1050: Unrecognizable insn:
:(insn/i 247 246 248 (set (reg:SI 85)
:        (mem/s:SI (plus:SI (mult:SI (subreg:SI (reg:DI 78) 0)
:                    (const_int 8 [0x8]))
:                (reg:SI 84)) 0)) -1 (insn_list 246 (nil))
:    (expr_list:REG_DEAD (reg:SI 84)
:        (nil)))
:perlio.c:1050: confused by earlier errors, bailing out
:make: *** [perlio.o] Error 1
:
:Version is perl patchlevel @17185; command was:
:   ( ./Configure -des -Dprefix=/opt/ldblead -Doptimize='-g -O6' -Dusedevel -Uversiononly -Dusemorebits && make test ) 2>&1 | tee zz

The problem seems to be caused by the O3 optimiser and specific to
long longs. Here's the smallest gcc test case I could come up with:

  crypt% cat subperlio.c
  typedef struct { int kind; } t_1;
  typedef struct { t_1 *funcs; } t_2;
  typedef struct { t_2 *array; } t_3;
  typedef struct s_4 {
      struct s_4 *next;
      int flags;
  } t_4;
  
  extern t_1 *v_1;
  extern t_3 *v_3;
  
  void f_1(t_4 *f);
  void f_2(int kind);
  
  t_1 *f_3(long long n) {
      if (n < (long long)v_1->kind)
          f_2(v_3->array[(int)n].funcs->kind);
      return (t_1*)0;
  }
  
  t_1 *f_4(int n) {
      if (n < 0)
          n += v_1->kind;
      return f_3((long long)n);
  }
  
  void f_5(t_4 *f) {
      while (((long long)f->flags) & 1) {
          if (&(f->next)) {
              f_1(f);
          } else {
              (void)f_4(0);
          }
      }
  }
  crypt% cc -o subperlio.o subperlio.c -O2 -c
  crypt% cc -o subperlio.o subperlio.c -O3 -c
  subperlio.c: In function `f_5':
  subperlio.c:36: Unrecognizable insn:
  (insn/i 74 73 75 (set (reg:SI 58)
          (mem/s:SI (plus:SI (mult:SI (subreg:SI (reg:DI 49) 0)
                      (const_int 4 [0x4]))
                  (reg:SI 57)) 4)) -1 (insn_list 73 (nil))
      (expr_list:REG_DEAD (reg:SI 57)
          (nil)))
  subperlio.c:36: confused by earlier errors, bailing out
  crypt% 

Here's some system details:
  crypt% gcc -v
  Reading specs from /usr/lib/gcc-lib/i386-redhat-linux/2.96/specs
  gcc version 2.96 20000731 (Red Hat Linux 7.1 2.96-81)
  crypt% cat /etc/redhat-release 
  Red Hat Linux release 7.1 (Seawolf)
  crypt% cat /proc/cpuinfo 
  processor       : 0
  vendor_id       : AuthenticAMD
  cpu family      : 6
  model           : 4
  model name      : AMD Athlon(tm) Processor
  stepping        : 2
  cpu MHz         : 1247.989
  cache size      : 256 KB
  fdiv_bug        : no
  hlt_bug         : no
  f00f_bug        : no
  coma_bug        : no
  fpu             : yes
  fpu_exception   : yes
  cpuid level     : 1
  wp              : yes
  flags           : fpu vme de pse tsc msr pae mce cx8 sep mtrr pge mca cmov pat pse36 mmx fxsr syscall mmxext 3dnowext 3dnow
  bogomips        : 2490.36
  
  crypt% 

For p5p: I'll see if I can come up with a workaround to avoid the failure,
but failing that I guess we'll need to drop to -O2 or less for perlio.c
under these conditions (ie, I guess, +perlio, +use64bitint, gcc v2.96).

Hope this helps,

Hugo van der Sanden