Integer class hack due to Sun CC

[Jeffrey Walton]

Hi Everyone,

Integer class took a bug report under Sun Studio 12.4. Also see Issue 184: Error: The operand "___LKDB" cannot be assigned to (http://github.com/weidai11/cryptopp/issues/188).

Sun CC 12.4 is an important Sun compiler because its their C++11 compiler. I'm fairly certain the bug report is bogus, and its a compiler bug. Never the less, we had to work around it.

The hack is available at http://github.com/weidai11/cryptopp/commit/7e06c1dce4b0ebc89708ba7afbcc23ff9093f4ea. The essence of the hack is:

  #if (__SUNPRO_CC == 0x5130)
  # define MAYBE_CONST
  #else
  # define MAYBE_CONST const
  #endif

    ...
  static word LinearMultiply(word *C, const word *AA, word B, size_t N)
  {
    MAYBE_CONST word* A = const_cast<word*>(AA);

    word carry=0;
    for(unsigned i=0; i<N; i++)
    {
        Declare2Words(p);
        MultiplyWords(p, A[i], B);
        Acc2WordsBy1(p, carry);
        C[i] = LowWord(p);
        carry = HighWord(p);
    }
    return carry;
  }

Under Sun CC, we cast away the const-ness. Under other compilers, we cast it away but restore it through MAYBE_CONST. For the other compilers, we are just creating a shadow.

I want to ensure there are __**NO**__ performance hits for this one.

If possible, could we get some testing under Visual Studio 2015 and 2016?

Jeff

[Jeffrey Walton]

On Tuesday, June 14, 2016 at 7:46:53 AM UTC-4, Jeffrey Walton wrote:

Hi Everyone,

Integer class took a bug report under Sun Studio 12.4. Also see Issue 184: Error: The operand "___LKDB" cannot be assigned to (http://github.com/weidai11/cryptopp/issues/188).

Sun CC 12.4 is an important Sun compiler because its their C++11 compiler. I'm fairly certain the bug report is bogus, and its a compiler bug. Never the less, we had to work around it.

The hack is available at http://github.com/weidai11/cryptopp/commit/7e06c1dce4b0ebc89708ba7afbcc23ff9093f4ea. The essence of the hack is:

  #if (__SUNPRO_CC == 0x5130)
  # define MAYBE_CONST
  #else
  # define MAYBE_CONST const
  #endif

This was changed a small bit:

   #if (__SUNPRO_CC == 0x5130)
   # define MAYBE_CONST

   # define MAYBE_UNCONST_CAST const_cast<word*>

   #else
   # define MAYBE_CONST const

   # define MAYBE_UNCONST_CAST
   #endif

It allows us to avoid the un-const cast completely if its not needed. It gets used as expected:

    void Baseline_Square2(word *R, const word *AA)
    {
        // http://github.com/weidai11/cryptopp/issues/188
        MAYBE_CONST word* A = MAYBE_UNCONST_CAST(AA);

        Squ_2
    }

Also see http://github.com/weidai11/cryptopp/commit/c042616ba5951c843a9b6e2d1c7f7b5a3893ccb9.

To test these changes, jump on the Solaris branch with 'git checkout solaris'. Or, visit GitHub, select Solaris from the drop down, then select Download ZIP.

Jeff

[Jeffrey Walton]

This was committed at http://github.com/weidai11/cryptopp/commit/b1df5736a7191eb14adb178cfff2c8ea3daf638e.

Jeff