[Bug c/33498] New: Optimizer (-O2) may convert a normal loop to infinite

[dada1 at cosmosbay dot com]

gcc-4.2.0 and gcc-4.2.1 cannot compile properly this function if -O2 is
selected

It generates an infinite loop :(

No problem for previous version (gcc-4.1.2 is OK)

$ cat bug.c
void table_init(int *value)
{
int i;
int val = 0x03020100;

for (i = 0; i < 256/4; i++) {
value[i] = val;
val += 0x04040404;
}
}

$ gcc -O2 -S bug.c
$ cat bug.s
.file   "bug.c"
.text
.p2align 4,,15
.globl table_init
.type   table_init, @function
table_init:
pushl   %ebp
movl$50462976, %edx
movl%esp, %ebp
movl$1, %eax
movl8(%ebp), %ecx
.p2align 4,,7
.L2:
movl%edx, -4(%ecx,%eax,4)
addl$67372036, %edx
addl$1, %eax
jmp .L2
.size   table_init, .-table_init
.ident  "GCC: (GNU) 4.2.1"
.section.note.GNU-stack,"",@progbits


-- 
   Summary: Optimizer (-O2) may convert a normal loop to infinite
   Product: gcc
   Version: 4.2.1
Status: UNCONFIRMED
  Severity: critical
  Priority: P3
 Component: c
AssignedTo: unassigned at gcc dot gnu dot org
    ReportedBy: dada1 at cosmosbay dot com
 GCC build triplet: i686-pc-linux-gnu
  GCC host triplet: i686-pc-linux-gnu
GCC target triplet: i686-pc-linux-gnu


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=33498

[Bug tree-optimization/33498] [4.2/4.3 Regression] Optimizer (-O2) may convert a normal loop to infinite

[dada1 at cosmosbay dot com]

--- Comment #10 from dada1 at cosmosbay dot com  2007-09-20 08:17 ---
> 
> What happens is that ivopts decide to use val as the variable to use in the
> exit compare; they compute what its final value will be (67305984), and 
> replace
> the exit test by val != 67305984.
> 
> There is not much I can do with that in ivopts.  I could make ivopts avoid
> preserving signed variables appearing in the source code that provably
> overflow; but I do not think we want to introduce this kind of hacks to handle
> code with undefined behavior.
> 

This code is valid. Integer overflows of a counter may happen in any program.

i = 0x7fff,
i += 1; /* IS VALID */
/* Here, gcc-4.1.2 can emit some infinite loop because programmer is lazy ! */

At very least, gcc should emit a BIG WARNING or ERROR

The integer overflow is not a excuse for a compiler to generate an infinite
loop.

int i;
int some_int = 0;
for (i = 0 ; i < 100 ; i++) {
some_func(some_int);
some_int += 0x4000; /* yes, it can 'overflow'... big deal */
}

Are you telling me that *any* integer overflow allows a compiler to generate a
buggy code without any notice ? Interesting.


-- 

dada1 at cosmosbay dot com changed:

   What|Removed |Added

 Status|RESOLVED|REOPENED
 Resolution|INVALID |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=33498