Re: [PATCH] Fix PR61375: cancel bswap optimization when value doesn't fit in a HOST_WIDE_INT

[Richard Biener <richard dot guenther at gmail dot com>]

On Tue, Jun 10, 2014 at 4:30 AM, Thomas Preud'homme
<thomas.preudhomme@arm.com> wrote:
> When analyzing a bitwise AND with a constant as part of a bitwise OR,
> the bswap pass stores the constant in a int64_t variable without checking
> if it fits. As a result, we get ICE when the constant is an __int128 value.
> This affects GCC trunk but also GCC 4.9 and 4.8 (and possibly earlier
> version as well).
>
>
> ChangeLog are changed as follows:
>
> *** gcc/ChangeLog ***
>
> 2014-06-05  Thomas Preud'homme  <thomas.preudhomme@arm.com>
>
>         PR tree-optimization/61375
>         * tree-ssa-math-opts.c (init_symbolic_number): Cancel optimization if
>         symbolic number cannot be represented in an unsigned HOST_WIDE_INT.
>         (find_bswap_or_nop_1): Likewise.
>
> *** gcc/testsuite/ChangeLog ***
>
> 2014-06-05  Thomas Preud'homme  <thomas.preudhomme@arm.com>
>
>         PR tree-optimization/61375
>         * gcc.c-torture/execute/pr61375-1.c: New test.
>
>
> diff --git a/gcc/testsuite/gcc.c-torture/execute/pr61375.c b/gcc/testsuite/gcc.c-torture/execute/pr61375.c
> new file mode 100644
> index 0000000..58df57a
> --- /dev/null
> +++ b/gcc/testsuite/gcc.c-torture/execute/pr61375.c
> @@ -0,0 +1,34 @@
> +#ifdef __UINT64_TYPE__
> +typedef __UINT64_TYPE__ uint64_t;
> +#else
> +typedef unsigned long long uint64_t;
> +#endif
> +
> +#ifndef __SIZEOF_INT128__
> +#define __int128 long long
> +#endif
> +
> +/* Some version of bswap optimization would ICE when analyzing a mask constant
> +   too big for an HOST_WIDE_INT (PR210931).  */
> +
> +__attribute__ ((noinline, noclone)) uint64_t
> +uint128_central_bitsi_ior (unsigned __int128 in1, uint64_t in2)
> +{
> +  __int128 mask = (__int128)0xffff << 56;
> +  return ((in1 & mask) >> 56) | in2;
> +}
> +
> +int main(int argc)
> +{
> +  __int128 in = 1;
> +#ifdef __SIZEOF_INT128__
> +  in <<= 64;
> +#endif
> +  if (sizeof (uint64_t) * __CHAR_BIT__ != 64)
> +    return 0;
> +  if (sizeof (unsigned __int128) * __CHAR_BIT__ != 128)
> +    return 0;
> +  if (uint128_central_bitsi_ior (in, 2) != 0x102)
> +    __builtin_abort ();
> +  return 0;
> +}
> diff --git a/gcc/tree-ssa-math-opts.c b/gcc/tree-ssa-math-opts.c
> index 658b341..95b3f25 100644
> --- a/gcc/tree-ssa-math-opts.c
> +++ b/gcc/tree-ssa-math-opts.c
> @@ -1717,6 +1717,8 @@ init_symbolic_number (struct symbolic_number *n, tree src)
>    if (n->size % BITS_PER_UNIT != 0)
>      return false;
>    n->size /= BITS_PER_UNIT;
> +  if (n->size > (int)sizeof (unsigned HOST_WIDE_INT))

this should be sizeof (uint64_t) on the trunk and sizeof (unsigned
HOST_WIDEST_INT) on branches.

> +    return false;
>    n->range = n->size;
>    n->n = CMPNOP;
>
> @@ -1883,6 +1885,8 @@ find_bswap_or_nop_1 (gimple stmt, struct symbolic_number *n, int limit)
>             type_size = TYPE_PRECISION (gimple_expr_type (stmt));
>             if (type_size % BITS_PER_UNIT != 0)
>               return NULL_TREE;
> +           if (type_size > (int)sizeof (unsigned HOST_WIDE_INT) * 8)
> +             return NULL_TREE;

Likewise.

>             if (type_size / BITS_PER_UNIT < (int)(sizeof (int64_t)))
>               {
>
> Is this OK for trunk?

Ok for trunk with using uint64_t.

 What about backports for 4.8 and 4.9? Would a
> reworked patch for these versions be accepted? The change would
> be trivial: the code in init_symbolic_number now was moved from
> some other place.

Backports are welcome - please post a patch.

Thanks,
Richard.

> Best regards,
>
> Thomas
>
>