Those macros have nothing to do with this function. Instead, replace A1
with A[1], etc. But I don't understand how the alternative implementation
can help significantly with your stack space problem. The branchless
version only uses up 32 bytes more stack space than the alternative.
On Tue, Mar 11, 2003 at 01:01:27PM -0600, Dave Sheely wrote:
> Sirs,
>
> Referring to the Crypto++ 5.0 implementation of
> "Portable::Multiply2" below (c.f., file
> "integer.cpp", I would prefer to use the implementation which is
> commented-out as I am a little
> light on stack space. However, this alternate implementation appears to
> be missing the definition
> or declaration of symbol "N2". Where might I find this?
>
> On a completely different note, is there or are there plans
> for a C implementation of
> Crypto++?
>
>
> Regards,
> David
>
>
> #define A0 A
> #define A1 (A+N2)
> #define B0 B
> #define B1 (B+N2)
>
> #define T0 T
> #define T1 (T+N2)
> #define T2 (T+N)
> #define T3 (T+N+N2)
>
> #define R0 R
> #define R1 (R+N2)
> #define R2 (R+N)
> #define R3 (R+N+N2)
>
> void Portable::Multiply2(word *C, const word *A, const word *B)
> {
> /*
> word s;
> dword d;
>
> if (A1 >= A0)
> if (B0 >= B1)
> {
> s = 0;
> d = (dword)(A1-A0)*(B0-B1);
> }
> else
> {
> s = (A1-A0);
> d = (dword)s*(word)(B0-B1);
> }
> else
> if (B0 > B1)
> {
> s = (B0-B1);
> d = (word)(A1-A0)*(dword)s;
> }
> else
> {
> s = 0;
> d = (dword)(A0-A1)*(B1-B0);
> }
> */
> // this segment is the branchless equivalent of above
> word D[4] = {A[1]-A[0], A[0]-A[1], B[0]-B[1], B[1]-B[0]};
> unsigned int ai = A[1] < A[0];
> unsigned int bi = B[0] < B[1];
> unsigned int di = ai & bi;
> dword d = (dword)D[di]*D[di+2];
> D[1] = D[3] = 0;
> unsigned int si = ai + !bi;
> word s = D[si];
>
> dword A0B0 = (dword)A[0]*B[0];
> C[0] = LOW_WORD(A0B0);
>
> dword A1B1 = (dword)A[1]*B[1];
> dword t = (dword) HIGH_WORD(A0B0) + LOW_WORD(A0B0) + LOW_WORD(d) +
> LOW_WORD(A1B1);
> C[1] = LOW_WORD(t);
>
> t = A1B1 + HIGH_WORD(t) + HIGH_WORD(A0B0) + HIGH_WORD(d) +
> HIGH_WORD(A1B1) - s;
> C[2] = LOW_WORD(t);
> C[3] = HIGH_WORD(t);
> }
