Thanks, you're right this might be a problem on platforms with 64-bit
pointers. (BTW, did you confirm what would actually happen in this case?)
I checked in a fix to CVS.
On Mon, Mar 24, 2003 at 07:28:30PM -0600, [EMAIL PROTECTED] wrote:
> I am definitely not familiar with the Crypto++ library but am
> trying to use it as a compiler benchmark.
>
> In the 5.1 codebase, I see the function below (first few lines).
> Since the index i is word32 which is unsigned, I would think
> that expressions like i-8 and i-5 and i-3 would wrap around
> to large positive numbers which is not what is wanted here.
>
> Are there any special procedures to follow for porting
> to a 64-bit target?
>
> -- Tom Deneau
>
>
>
> void Serpent::Base::UncheckedSetKey(CipherDir direction, const byte *userKey,
> unsigned int keylen)
> {
> AssertValidKeyLength(keylen);
>
> word32 *k = m_key;
> GetUserKey(LITTLE_ENDIAN_ORDER, k, 8, userKey, keylen);
>
> word32 i,a,b,c,d,e;
>
> if (keylen < 32)
> k[keylen/4] |= word32(1) << ((keylen%4)*8);
>
> k += 8;
> word32 t = k[-1];
> for (i = 0; i < 132; ++i)
> k[i] = t = rotlFixed(k[i-8] ^ k[i-5] ^ k[i-3] ^ t ^ 0x9e3779b9 ^ i,
> 11);
> k -= 20;
