Re: [PATCH] Fix PR61306: improve handling of sign and cast in bswap

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

On Tue, Jun 10, 2014 at 10:49 AM, Thomas Preud'homme
<thomas.preudhomme@arm.com> wrote:
>> From: Richard Biener [mailto:richard.guenther@gmail.com]
>> Sent: Wednesday, June 04, 2014 5:39 PM
>> To: Thomas Preud'homme
>>
>
>>
>> I'm failing to get why you possibly need two casts ... you should
>> only need one, from the bswap/load result to the final type
>> (zero-extended as you say - so the load type should simply be
>> unsigned which it is already).
>
> You are right indeed. I failed to realize that the problems I
> encountered were caused by an initially wrong understanding of
> the reason behind PR61306. All this code is not necessary.
>
>>
>> So I think that the testcase in the patch is fixed already by
>> doing the n->type change (and a proper sign-extension detection).
>>
>> Can you please split that part out?
>
> Doing so I realize the patch was incomplete. Sign extension can
> be triggered in two distinct place in the code (right shift and cast)
> that can both lead to incorrect code being generated. With some
> efforts I managed to create two testcases that work both on
> GCC trunk but also GCC 4.9 and 4.8.
>
> ChangeLog entries are:
>
> *** gcc/ChangeLog ***
>
> 2014-06-05  Thomas Preud'homme  <thomas.preudhomme@arm.com>
>
>         PR tree-optimization/61306
>         * tree-ssa-math-opts.c (struct symbolic_number): Store type of
>         expression instead of its size.
>         (do_shift_rotate): Adapt to change in struct symbolic_number. Return
>         false to prevent optimization when the result is unpredictable due to
>         arithmetic right shift of signed type with highest byte is set.
>         (verify_symbolic_number_p): Adapt to change in struct symbolic_number.
>         (init_symbolic_number): Likewise.
>         (find_bswap_or_nop_1): Likewise. Return NULL to prevent optimization
>         when the result is unpredictable due to sign extension.
>
> *** gcc/testsuite/ChangeLog ***
>
> 2014-06-05  Thomas Preud'homme  <thomas.preudhomme@arm.com>
>
>         * gcc.c-torture/execute/pr61306-1.c: New test.
>         * gcc.c-torture/execute/pr61306-2.c: Likewise.
>         * gcc.c-torture/execute/pr61306-3.c: Likewise.
>
>
> diff --git a/gcc/testsuite/gcc.c-torture/execute/pr61306-1.c b/gcc/testsuite/gcc.c-torture/execute/pr61306-1.c
> new file mode 100644
> index 0000000..f6e8ff3
> --- /dev/null
> +++ b/gcc/testsuite/gcc.c-torture/execute/pr61306-1.c
> @@ -0,0 +1,39 @@
> +#ifdef __INT32_TYPE__
> +typedef __INT32_TYPE__ int32_t;
> +#else
> +typedef int int32_t;
> +#endif
> +
> +#ifdef __UINT32_TYPE__
> +typedef __UINT32_TYPE__ uint32_t;
> +#else
> +typedef unsigned uint32_t;
> +#endif
> +
> +#define __fake_const_swab32(x) ((uint32_t)(                  \
> +        (((uint32_t)(x) & (uint32_t)0x000000ffUL) << 24) |    \
> +        (((uint32_t)(x) & (uint32_t)0x0000ff00UL) <<  8) |    \
> +        (((uint32_t)(x) & (uint32_t)0x00ff0000UL) >>  8) |    \
> +        (( (int32_t)(x) &  (int32_t)0xff000000UL) >> 24)))
> +
> +/* Previous version of bswap optimization failed to consider sign extension
> +   and as a result would replace an expression *not* doing a bswap by a
> +   bswap.  */
> +
> +__attribute__ ((noinline, noclone)) uint32_t
> +fake_bswap32 (uint32_t in)
> +{
> +  return __fake_const_swab32 (in);
> +}
> +
> +int
> +main(void)
> +{
> +  if (sizeof (int32_t) * __CHAR_BIT__ != 32)
> +    return 0;
> +  if (sizeof (uint32_t) * __CHAR_BIT__ != 32)
> +    return 0;
> +  if (fake_bswap32 (0x87654321) != 0xffffff87)
> +    __builtin_abort ();
> +  return 0;
> +}
> diff --git a/gcc/testsuite/gcc.c-torture/execute/pr61306-2.c b/gcc/testsuite/gcc.c-torture/execute/pr61306-2.c
> new file mode 100644
> index 0000000..6cbbd19
> --- /dev/null
> +++ b/gcc/testsuite/gcc.c-torture/execute/pr61306-2.c
> @@ -0,0 +1,40 @@
> +#ifdef __INT16_TYPE__
> +typedef __INT16_TYPE__ int16_t;
> +#else
> +typedef short int16_t;
> +#endif
> +
> +#ifdef __UINT32_TYPE__
> +typedef __UINT32_TYPE__ uint32_t;
> +#else
> +typedef unsigned uint32_t;
> +#endif
> +
> +#define __fake_const_swab32(x) ((uint32_t)(                          \
> +        (((uint32_t)         (x) & (uint32_t)0x000000ffUL) << 24) |   \
> +        (((uint32_t)(int16_t)(x) & (uint32_t)0x00ffff00UL) <<  8) |   \
> +        (((uint32_t)         (x) & (uint32_t)0x00ff0000UL) >>  8) |   \
> +        (((uint32_t)         (x) & (uint32_t)0xff000000UL) >> 24)))
> +
> +
> +/* Previous version of bswap optimization failed to consider sign extension
> +   and as a result would replace an expression *not* doing a bswap by a
> +   bswap.  */
> +
> +__attribute__ ((noinline, noclone)) uint32_t
> +fake_bswap32 (uint32_t in)
> +{
> +  return __fake_const_swab32 (in);
> +}
> +
> +int
> +main(void)
> +{
> +  if (sizeof (uint32_t) * __CHAR_BIT__ != 32)
> +    return 0;
> +  if (sizeof (int16_t) * __CHAR_BIT__ != 16)
> +    return 0;
> +  if (fake_bswap32 (0x81828384) != 0xff838281)
> +    __builtin_abort ();
> +  return 0;
> +}
> diff --git a/gcc/testsuite/gcc.c-torture/execute/pr61306-3.c b/gcc/testsuite/gcc.c-torture/execute/pr61306-3.c
> new file mode 100644
> index 0000000..6086e27
> --- /dev/null
> +++ b/gcc/testsuite/gcc.c-torture/execute/pr61306-3.c
> @@ -0,0 +1,13 @@
> +short a = -1;
> +int b;
> +char c;
> +
> +int
> +main ()
> +{
> +  c = a;
> +  b = a | c;
> +  if (b != -1)
> +    __builtin_abort ();
> +  return 0;
> +}
> diff --git a/gcc/tree-ssa-math-opts.c b/gcc/tree-ssa-math-opts.c
> index 658b341..d92c433 100644
> --- a/gcc/tree-ssa-math-opts.c
> +++ b/gcc/tree-ssa-math-opts.c
> @@ -1622,7 +1622,7 @@ make_pass_cse_sincos (gcc::context *ctxt)
>
>  struct symbolic_number {
>    uint64_t n;
> -  int size;
> +  tree type;
>    tree base_addr;
>    tree offset;
>    HOST_WIDE_INT bytepos;
> @@ -1652,13 +1652,15 @@ do_shift_rotate (enum tree_code code,
>                  struct symbolic_number *n,
>                  int count)
>  {
> +  int bitsize = TYPE_PRECISION (n->type);
> +
>    if (count % 8 != 0)
>      return false;
>
>    /* Zero out the extra bits of N in order to avoid them being shifted
>       into the significant bits.  */
> -  if (n->size < (int)sizeof (int64_t))
> -    n->n &= ((uint64_t)1 << (n->size * BITS_PER_UNIT)) - 1;
> +  if (bitsize < 8 * (int)sizeof (int64_t))
> +    n->n &= ((uint64_t)1 << bitsize) - 1;
>
>    switch (code)
>      {
> @@ -1666,20 +1668,23 @@ do_shift_rotate (enum tree_code code,
>        n->n <<= count;
>        break;
>      case RSHIFT_EXPR:
> +      /* Arithmetic shift of signed type: result is dependent on the value.  */
> +      if (!TYPE_UNSIGNED (n->type) && (n->n & (0xff << (bitsize - 8))))
> +       return false;
>        n->n >>= count;
>        break;
>      case LROTATE_EXPR:
> -      n->n = (n->n << count) | (n->n >> ((n->size * BITS_PER_UNIT) - count));
> +      n->n = (n->n << count) | (n->n >> (bitsize - count));
>        break;
>      case RROTATE_EXPR:
> -      n->n = (n->n >> count) | (n->n << ((n->size * BITS_PER_UNIT) - count));
> +      n->n = (n->n >> count) | (n->n << (bitsize - count));
>        break;
>      default:
>        return false;
>      }
>    /* Zero unused bits for size.  */
> -  if (n->size < (int)sizeof (int64_t))
> -    n->n &= ((uint64_t)1 << (n->size * BITS_PER_UNIT)) - 1;
> +  if (bitsize < 8 * (int)sizeof (int64_t))
> +    n->n &= ((uint64_t)1 << bitsize) - 1;
>    return true;
>  }
>
> @@ -1696,7 +1701,7 @@ verify_symbolic_number_p (struct symbolic_number *n, gimple stmt)
>    if (TREE_CODE (lhs_type) != INTEGER_TYPE)
>      return false;
>
> -  if (TYPE_PRECISION (lhs_type) != n->size * BITS_PER_UNIT)
> +  if (TYPE_PRECISION (lhs_type) != TYPE_PRECISION (n->type))
>      return false;
>
>    return true;
> @@ -1708,20 +1713,23 @@ verify_symbolic_number_p (struct symbolic_number *n, gimple stmt)
>  static bool
>  init_symbolic_number (struct symbolic_number *n, tree src)
>  {
> +  int size;
> +
>    n->base_addr = n->offset = n->alias_set = n->vuse = NULL_TREE;
>
>    /* Set up the symbolic number N by setting each byte to a value between 1 and
>       the byte size of rhs1.  The highest order byte is set to n->size and the
>       lowest order byte to 1.  */
> -  n->size = TYPE_PRECISION (TREE_TYPE (src));
> -  if (n->size % BITS_PER_UNIT != 0)
> +  n->type = TREE_TYPE (src);
> +  size = TYPE_PRECISION (n->type);
> +  if (size % BITS_PER_UNIT != 0)
>      return false;
> -  n->size /= BITS_PER_UNIT;
> -  n->range = n->size;
> +  size /= BITS_PER_UNIT;
> +  n->range = size;
>    n->n = CMPNOP;
>
> -  if (n->size < (int)sizeof (int64_t))
> -    n->n &= ((uint64_t)1 << (n->size * BITS_PER_UNIT)) - 1;
> +  if (size < (int)sizeof (int64_t))
> +    n->n &= ((uint64_t)1 << (size * BITS_PER_UNIT)) - 1;
>
>    return true;
>  }
> @@ -1857,12 +1865,12 @@ find_bswap_or_nop_1 (gimple stmt, struct symbolic_number *n, int limit)
>         {
>         case BIT_AND_EXPR:
>           {
> -           int i;
> +           int i, size = TYPE_PRECISION (n->type) / BITS_PER_UNIT;
>             uint64_t val = int_cst_value (rhs2);
>             uint64_t tmp = val;
>
>             /* Only constants masking full bytes are allowed.  */
> -           for (i = 0; i < n->size; i++, tmp >>= BITS_PER_UNIT)
> +           for (i = 0; i < size; i++, tmp >>= BITS_PER_UNIT)
>               if ((tmp & 0xff) != 0 && (tmp & 0xff) != 0xff)
>                 return NULL_TREE;
>
> @@ -1878,21 +1886,30 @@ find_bswap_or_nop_1 (gimple stmt, struct symbolic_number *n, int limit)
>           break;
>         CASE_CONVERT:
>           {
> -           int type_size;
> +           int type_size, old_type_size;
> +           tree type;
>
> -           type_size = TYPE_PRECISION (gimple_expr_type (stmt));
> +           type = gimple_expr_type (stmt);
> +           type_size = TYPE_PRECISION (type);
>             if (type_size % BITS_PER_UNIT != 0)
>               return NULL_TREE;
>
> +           /* Sign extension: result is dependent on the value.  */
> +           old_type_size = TYPE_PRECISION (n->type);
> +           if (!TYPE_UNSIGNED (n->type)
> +               && type_size > old_type_size
> +               && n->n & (0xff << (old_type_size - 8)))
> +             return NULL_TREE;
> +
>             if (type_size / BITS_PER_UNIT < (int)(sizeof (int64_t)))
>               {
>                 /* If STMT casts to a smaller type mask out the bits not
>                    belonging to the target type.  */
>                 n->n &= ((uint64_t)1 << type_size) - 1;
>               }
> -           n->size = type_size / BITS_PER_UNIT;
> +           n->type = type;
>             if (!n->base_addr)
> -             n->range = n->size;
> +             n->range = type_size / BITS_PER_UNIT;
>           }
>           break;
>         default:
> @@ -1905,7 +1922,7 @@ find_bswap_or_nop_1 (gimple stmt, struct symbolic_number *n, int limit)
>
>    if (rhs_class == GIMPLE_BINARY_RHS)
>      {
> -      int i;
> +      int i, size;
>        struct symbolic_number n1, n2;
>        uint64_t mask;
>        tree source_expr2;
> @@ -1931,7 +1948,7 @@ find_bswap_or_nop_1 (gimple stmt, struct symbolic_number *n, int limit)
>           if (!source_expr2)
>             return NULL_TREE;
>
> -         if (n1.size != n2.size)
> +         if (TYPE_PRECISION (n1.type) != TYPE_PRECISION (n2.type))
>             return NULL_TREE;
>
>           if (!n1.vuse != !n2.vuse ||
> @@ -1997,8 +2014,9 @@ find_bswap_or_nop_1 (gimple stmt, struct symbolic_number *n, int limit)
>           n->base_addr = n1.base_addr;
>           n->offset = n1.offset;
>           n->bytepos = n1.bytepos;
> -         n->size = n1.size;
> -         for (i = 0, mask = 0xff; i < n->size; i++, mask <<= BITS_PER_UNIT)
> +         n->type = n1.type;
> +         size = TYPE_PRECISION (n->type) / BITS_PER_UNIT;
> +         for (i = 0, mask = 0xff; i < size; i++, mask <<= BITS_PER_UNIT)
>             {
>               uint64_t masked1, masked2;
>
>
> Is this OK for trunk? Does this bug qualify for a backport patch to
> 4.8 and 4.9 branches?

This is ok for trunk and also for backporting (after a short while to
see if there is any fallout).

Thanks,
RIchard.

> Best regards,
>
> Thomas
>
>