int test_heuristic_scalar_product()
{
int result = 1;
ulong count1, count2, i;
double ** mat;
F_mpz_mat_t B;
int expo[50];
for (count1 = 0; count1 < 1000; count1++)
{
ulong rows = z_randint(50) + 1;
ulong cols = z_randint(50) + 2;
F_mpz_mat_init(B, rows, cols);
F_mpz_randmat(B, rows, cols, 900);
mat = d_mat_init(rows, cols);
for (i = 0; i < rows; i++)
expo[i] = _F_mpz_vec_to_d_vec_2exp(mat[i], B->rows[i], cols);
for (count2 = 0; (count2 < 100); count2++)
{
ulong r1 = z_randint(rows);
ulong r2 = z_randint(rows);
double d1 = heuristic_scalar_product(mat[r1], mat[r2], cols,
B, r1, r2, expo[r1] + expo[r2]);
double d2 = heuristic_scalar_product(mat[r1], mat[r1], cols,
B, r1, r1, expo[r1] + expo[r1]);
double d3 = heuristic_scalar_product(mat[r2], mat[r2], cols,
B, r2, r2, expo[r2] + expo[r2]);
_d_vec_add(mat[r2], mat[r1], mat[r2], cols);
_F_mpz_vec_add(B->rows[r2], B->rows[r1], B->rows[r2], cols);
double d4 = heuristic_scalar_product(mat[r2], mat[r2], cols,
B, r2, r2, expo[r2] + expo[r2]);
result = (fabs(d4 - d3 - d2 - 2*d1) < 1.0E-12);
if (!result)
{
printf("count2 = %ld Failed expo[r1] = %d expo[r2] = %d, d1 = %lf, d2 = %lf, d3 = %lf, d4 = %f\n", count2, expo[r1], expo[r2], d1, d2, d3, d4);
}
expo[r2] = _F_mpz_vec_to_d_vec_2exp(mat[r2], B->rows[r2], cols);
}
F_mpz_mat_clear(B);
d_mat_clear(mat);
}
return result;
}
void LLL(F_mpz_mat_t B)
{
int kappa, kappa2, d, n, i, j, zeros, kappamax;
double ** mu, ** r, ** appB, ** appSP;
double * s, * mutmp, * appBtmp, * appSPtmp;
double tmp = 0.0;
int * expo, * alpha;
mp_limb_t * Btmp;
n = B->c;
d = B->r;
ulong shift = getShift(B);
alpha = (int *) malloc(d * sizeof(int));
expo = (int *) malloc(d * sizeof(int));
mu = d_mat_init(d, d);
r = d_mat_init(d, d);
appB = d_mat_init(d, n);
appSP = d_mat_init(d, d);
s = (double *) malloc (d * sizeof(double));
appSPtmp = (double *) malloc (d * sizeof(double));
for (i = 0; i < d; i++)
for (j = 0; j < d; j++)
appSP[i][j] = NAN;//0.0/0.0;
/* ************************** */
/* Step1: Initialization Step */
/* ************************** */
for (i = 0; i < d; i++)
expo[i] = F_mpz_mat_set_line_d(appB[i], B, i, n);
/* ********************************* */
/* Step2: Initializing the main loop */
/* ********************************* */
kappamax = 0;
i = 0;
do
appSP[i][i] = d_vec_norm(appB[i], n);
while ((appSP[i][i] <= 0.0) && (++i < d)); // Fixme : should this be EPS not 0.0
zeros = i - 1; /* all vectors B[i] with i <= zeros are zero vectors */
kappa = i + 1;
if (zeros < d - 1) r[i][i] = appSP[i][i];
for (i = zeros + 1; i < d; i++)
alpha[i] = 0;
while (kappa < d)
{
if (kappa > kappamax) kappamax++; // Fixme : should this be kappamax = kappa instead of kappamax++
/* ********************************** */
/* Step3: Call to the Babai algorithm */
/* ********************************** */
Babai(kappa, B, mu, r, s, appB, expo, appSP, alpha[kappa], zeros,
kappamax, FLINT_MIN(kappamax + 1 + shift, n));
/* ************************************ */
/* Step4: Success of Lovasz's condition */
/* ************************************ */
/* ctt * r.coeff[kappa-1][kappa-1] <= s[kappa-2] ?? */
tmp = r[kappa-1][kappa-1] * ctt;
tmp = ldexp (tmp, 2*(expo[kappa-1] - expo[kappa]));
if (tmp <= s[kappa-1])
{
alpha[kappa] = kappa;
tmp = mu[kappa][kappa-1] * r[kappa][kappa-1];
r[kappa][kappa] = s[kappa-1] - tmp;
kappa++;
} else
{
/* ******************************************* */
/* Step5: Find the right insertion index kappa */
/* kappa2 remains the initial kappa */
/* ******************************************* */
kappa2 = kappa;
do
{
kappa--;
if (kappa > zeros + 1)
{
tmp = r[kappa-1][kappa-1] * ctt;
tmp = ldexp(tmp, 2*(expo[kappa-1] - expo[kappa2]));
}
} while ((kappa >= zeros + 2) && (s[kappa-1] <= tmp));
for (i = kappa; i < kappa2; i++)
if (kappa <= alpha[i]) alpha[i] = kappa;
for (i = kappa2; i > kappa; i--) alpha[i] = alpha[i-1];
for (i = kappa2 + 1; i <= kappamax; i++)
if (kappa < alpha[i]) alpha[i] = kappa;
alpha[kappa] = kappa;
/* ****************************** */
/* Step6: Update the mu's and r's */
/* ****************************** */
mutmp = mu[kappa2];
for (i = kappa2; i > kappa; i--) mu[i] = mu[i-1];
mu[kappa] = mutmp;
mutmp = r[kappa2];
for (i = kappa2; i > kappa; i--) r[i] = r[i-1];
r[kappa] = mutmp;
r[kappa][kappa] = s[kappa];
/* ************************ */
/* Step7: Update B and appB */
/* ************************ */
Btmp = B->rows[kappa2];
for (i = kappa2; i > kappa; i--) B->rows[i] = B->rows[i-1];
B->rows[kappa] = Btmp;
appBtmp = appB[kappa2];
for (i = kappa2; i > kappa; i--) appB[i] = appB[i-1];
appB[kappa] = appBtmp;
j = expo[kappa2];
for (i = kappa2; i > kappa; i--) expo[i] = expo[i-1];
expo[kappa] = j;
/* *************************** */
/* Step8: Update appSP: tricky */
/* *************************** */
for (i = 0; i <= kappa2; i++) appSPtmp[i] = appSP[kappa2][i];
for (i = kappa2 + 1; i <= kappamax; i++) appSPtmp[i] = appSP[i][kappa2];
for (i = kappa2; i > kappa; i--)
{
for (j = 0; j < kappa; j++) appSP[i][j] = appSP[i-1][j];
appSP[i][kappa] = appSPtmp[i-1];
for (j = kappa + 1; j <= i; j++) appSP[i][j] = appSP[i-1][j-1];
for (j = kappa2 + 1; j <= kappamax; j++) appSP[j][i] = appSP[j][i-1];
}
for (i = 0; i < kappa; i++) appSP[kappa][i] = appSPtmp[i];
appSP[kappa][kappa] = appSPtmp[kappa2];
for (i = kappa2 + 1; i <= kappamax; i++) appSP[i][kappa] = appSPtmp[i];
if (r[kappa][kappa] <= 0.0)
{
zeros++;
kappa++;
appSP[kappa][kappa] = d_vec_norm(appB[kappa], n);
r[kappa][kappa] = appSP[kappa][kappa];
}
kappa++;
}
}
free(alpha);
free(expo);
d_mat_clear(mu);
d_mat_clear(r);
d_mat_clear(appB);
d_mat_clear(appSP);
free(s);
free(appSPtmp);
}