Optimized use of temporary arrays in polynomial multiplication

This commit is contained in:
Andreas Steffen 2014-02-27 15:22:48 +01:00
parent bf24960cbe
commit f87f28ec68
2 changed files with 24 additions and 24 deletions

View File

@ -92,14 +92,13 @@ METHOD(ntru_poly_t, get_indices, uint16_t*,
* the indices of its +1 and -1 coefficients results in polynomial c.
* This is a convolution operation
*/
static void ring_mult_indices(uint16_t *a, indices_len_t len, uint16_t *indices,
uint16_t N, uint16_t mod_q_mask, uint16_t *c)
static void ring_mult_i(uint16_t *a, indices_len_t len, uint16_t *indices,
uint16_t N, uint16_t mod_q_mask, uint16_t *t,
uint16_t *c)
{
uint16_t *t;
int i, j, k;
/* allocate and initialize temporary array t */
t = malloc(N * sizeof(uint16_t));
/* initialize temporary array t */
for (k = 0; k < N; k++)
{
t[k] = 0;
@ -144,56 +143,53 @@ static void ring_mult_indices(uint16_t *a, indices_len_t len, uint16_t *indices,
{
c[k] = t[k] & mod_q_mask;
}
/* cleanup */
free(t);
}
METHOD(ntru_poly_t, ring_mult, void,
private_ntru_poly_t *this, uint16_t *a, uint16_t *c)
{
uint16_t *bi = this->indices, mod_q_mask = this->q - 1;
uint16_t *t1, *t2;
uint16_t *bi = this->indices;
uint16_t mod_q_mask = this->q - 1;
int i;
/* allocate temporary array t1 */
t1 = malloc(this->N * sizeof(uint16_t));
if (this->num_polynomials == 1)
{
ring_mult_indices(a, this->indices_len[0], bi, this->N, mod_q_mask, c);
ring_mult_i(a, this->indices_len[0], bi, this->N, mod_q_mask, t1, c);
}
else
{
uint16_t *t1, *t2;
int i;
/* allocate temporary arrays */
t1 = malloc(this->N * sizeof(uint16_t));
/* allocate temporary array t2 */
t2 = malloc(this->N * sizeof(uint16_t));
/* t1 = a * b1 */
ring_mult_indices(a, this->indices_len[0], bi, this->N, mod_q_mask, t1);
ring_mult_i(a, this->indices_len[0], bi, this->N, mod_q_mask, t1, t1);
/* t1 = (a * b1) * b2 */
bi += this->indices_len[0].p + this->indices_len[0].m;
ring_mult_indices(t1, this->indices_len[1], bi, this->N, mod_q_mask, t1);
ring_mult_i(t1, this->indices_len[1], bi, this->N, mod_q_mask, t2, t1);
/* t2 = a * b3 */
bi += this->indices_len[1].p + this->indices_len[1].m;
ring_mult_indices(a, this->indices_len[2], bi, this->N, mod_q_mask, t2);
ring_mult_i(a, this->indices_len[2], bi, this->N, mod_q_mask, t2, t2);
/* c = (a * b1 * b2) + (a * b3) */
for (i = 0; i < this->N; i++)
{
c[i] = (t1[i] + t2[i]) & mod_q_mask;
}
/* cleanup */
free(t1);
free(t2);
}
free(t1);
}
METHOD(ntru_poly_t, destroy, void,
private_ntru_poly_t *this)
{
memwipe(this->indices, get_size(this));
memwipe(this->indices, sizeof(uint16_t) * get_size(this));
free(this->indices);
free(this);
}

View File

@ -38,12 +38,16 @@ struct ntru_poly_t {
size_t (*get_size)(ntru_poly_t *this);
/**
* @return array containing the indices of the non-zero coefficients
* @return array containing the indices of the non-zero coefficients
*/
uint16_t* (*get_indices)(ntru_poly_t *this);
/**
* @return array containing the indices of the non-zero coefficients
* Multiply polynomial a with ntru_poly_t object b having sparse coeffients
* to form result polynomial c = a * b
*
* @param a input polynomial a
* @param b output polynomial c
*/
void (*ring_mult)(ntru_poly_t *this, uint16_t *a, uint16_t *c);