void fft_mfa_truncate_sqrt2_inner(mp_limb_t ** ii, mp_limb_t ** jj, mp_size_t n,
mp_bitcnt_t w, mp_limb_t ** t1, mp_limb_t ** t2,
mp_limb_t ** temp, mp_size_t n1, mp_size_t trunc, mp_limb_t * tt)
{
mp_size_t i, j, s;
mp_size_t n2 = (2*n)/n1;
mp_size_t trunc2 = (trunc - 2*n)/n1;
mp_size_t limbs = (n*w)/FLINT_BITS;
mp_bitcnt_t depth = 0;
mp_bitcnt_t depth2 = 0;
while ((UWORD(1)<<depth) < n2) depth++;
while ((UWORD(1)<<depth2) < n1) depth2++;
ii += 2*n;
jj += 2*n;
/* convolutions on relevant rows */
for (s = 0; s < trunc2; s++)
{
i = n_revbin(s, depth);
fft_radix2(ii + i*n1, n1/2, w*n2, t1, t2);
if (ii != jj) fft_radix2(jj + i*n1, n1/2, w*n2, t1, t2);
for (j = 0; j < n1; j++)
{
mp_size_t t = i*n1 + j;
mpn_normmod_2expp1(ii[t], limbs);
if (ii != jj) mpn_normmod_2expp1(jj[t], limbs);
fft_mulmod_2expp1(ii[t], ii[t], jj[t], n, w, tt);
}
ifft_radix2(ii + i*n1, n1/2, w*n2, t1, t2);
}
ii -= 2*n;
jj -= 2*n;
/* convolutions on rows */
for (i = 0; i < n2; i++)
{
fft_radix2(ii + i*n1, n1/2, w*n2, t1, t2);
if (ii != jj) fft_radix2(jj + i*n1, n1/2, w*n2, t1, t2);
for (j = 0; j < n1; j++)
{
mp_size_t t = i*n1 + j;
mpn_normmod_2expp1(ii[t], limbs);
if (ii != jj) mpn_normmod_2expp1(jj[t], limbs);
fft_mulmod_2expp1(ii[t], ii[t], jj[t], n, w, tt);
}
ifft_radix2(ii + i*n1, n1/2, w*n2, t1, t2);
}
}
void fft_convolution(mp_limb_t ** ii, mp_limb_t ** jj, long depth,
long limbs, long trunc, mp_limb_t ** t1,
mp_limb_t ** t2, mp_limb_t ** s1, mp_limb_t * tt)
{
long n = (1L<<depth), j;
long w = (limbs*FLINT_BITS)/n;
long sqrt = (1L<<(depth/2));
if (depth <= 6)
{
trunc = 2*((trunc + 1)/2);
fft_truncate_sqrt2(ii, n, w, t1, t2, s1, trunc);
if (ii != jj)
fft_truncate_sqrt2(jj, n, w, t1, t2, s1, trunc);
for (j = 0; j < trunc; j++)
{
mpn_normmod_2expp1(ii[j], limbs);
if (ii != jj) mpn_normmod_2expp1(jj[j], limbs);
fft_mulmod_2expp1(ii[j], ii[j], jj[j], n, w, tt);
}
ifft_truncate_sqrt2(ii, n, w, t1, t2, s1, trunc);
for (j = 0; j < trunc; j++)
{
mpn_div_2expmod_2expp1(ii[j], ii[j], limbs, depth + 2);
mpn_normmod_2expp1(ii[j], limbs);
}
} else
{
trunc = 2*sqrt*((trunc + 2*sqrt - 1)/(2*sqrt));
fft_mfa_truncate_sqrt2_outer(ii, n, w, t1, t2, s1, sqrt, trunc);
if (ii != jj)
fft_mfa_truncate_sqrt2_outer(jj, n, w, t1, t2, s1, sqrt, trunc);
fft_mfa_truncate_sqrt2_inner(ii, jj, n, w, t1, t2, s1, sqrt, trunc, tt);
ifft_mfa_truncate_sqrt2_outer(ii, n, w, t1, t2, s1, sqrt, trunc);
}
}
int mpn_mulmod_Bexpp1(mp_ptr r, mp_srcptr i1, mp_srcptr i2, mp_size_t limbs, mp_ptr tt)
{
mp_size_t bits = limbs * GMP_LIMB_BITS;
mp_bitcnt_t depth1, depth = 1;
mp_size_t w1, off;
mp_limb_t c = 2 * i1[limbs] + i2[limbs];
if (c & 1)
{
mpn_neg_n(r, i1, limbs + 1);
mpn_normmod_2expp1(r, limbs);
return 0;
} else if (c & 2)
{
mpn_neg_n(r, i2, limbs + 1);
mpn_normmod_2expp1(r, limbs);
return 0;
}
if (limbs <= FFT_MULMOD_2EXPP1_CUTOFF)
{
if(bits)
r[limbs] = mpn_mulmod_2expp1_basecase(r, i1, i2, c, bits, tt);
else
r[limbs] = 0;
return r[limbs];
}
while ((((mp_limb_t)1)<<depth) < bits) depth++;
if (depth < 12) off = mulmod_2expp1_table_n[0];
else off = mulmod_2expp1_table_n[MIN(depth, FFT_N_NUM + 11) - 12];
depth1 = depth/2 - off;
w1 = bits/(((mp_limb_t)1)<<(2*depth1));
mpir_fft_mulmod_2expp1(r, i1, i2, limbs, depth1, w1);
return r[limbs];
}
int
main(void)
{
mp_bitcnt_t depth, w;
flint_rand_t state;
printf("fft/ifft_mfa_truncate_sqrt2....");
fflush(stdout);
flint_randinit(state);
_flint_rand_init_gmp(state);
for (depth = 6; depth <= 13; depth++)
{
for (w = 1; w <= 5; w++)
{
mp_size_t n = (1UL<<depth);
mp_size_t trunc = 2*n + n_randint(state, 2*n) + 1;
mp_size_t n1 = (1UL<<(depth/2));
mp_size_t limbs = (n*w)/GMP_LIMB_BITS;
mp_size_t size = limbs + 1;
mp_size_t i;
mp_limb_t * ptr;
mp_limb_t ** ii, ** jj, * t1, * t2, * s1;
trunc = 2*n1*((trunc + 2*n1 - 1)/(2*n1));
ii = flint_malloc((4*(n + n*size) + 3*size)*sizeof(mp_limb_t));
for (i = 0, ptr = (mp_limb_t *) ii + 4*n; i < 4*n; i++, ptr += size)
{
ii[i] = ptr;
random_fermat(ii[i], state, limbs);
}
t1 = ptr;
t2 = t1 + size;
s1 = t2 + size;
for (i = 0; i < 4*n; i++)
mpn_normmod_2expp1(ii[i], limbs);
jj = flint_malloc(4*(n + n*size)*sizeof(mp_limb_t));
for (i = 0, ptr = (mp_limb_t *) jj + 4*n; i < 4*n; i++, ptr += size)
{
jj[i] = ptr;
mpn_copyi(jj[i], ii[i], size);
}
fft_mfa_truncate_sqrt2(ii, n, w, &t1, &t2, &s1, n1, trunc);
ifft_mfa_truncate_sqrt2(ii, n, w, &t1, &t2, &s1, n1, trunc);
for (i = 0; i < trunc; i++)
{
mpn_div_2expmod_2expp1(ii[i], ii[i], limbs, depth + 2);
mpn_normmod_2expp1(ii[i], limbs);
}
for (i = 0; i < trunc; i++)
{
if (mpn_cmp(ii[i], jj[i], size) != 0)
{
printf("FAIL:\n");
printf("n = %ld, trunc = %ld\n", n, trunc);
printf("Error in entry %ld\n", i);
abort();
}
}
flint_free(ii);
flint_free(jj);
}
}
flint_randclear(state);
printf("PASS\n");
return 0;
}
int
main(void)
{
int i, result;
flint_rand_t state;
printf("get/set_fft....");
fflush(stdout);
flint_randinit(state);
/* convert back and forth and compare */
for (i = 0; i < 10000; i++)
{
fmpz * a, * b;
mp_bitcnt_t bits;
long len, limbs;
mp_limb_t ** ii, * ptr;
long i, bt;
bits = n_randint(state, 300) + 1;
len = n_randint(state, 300) + 1;
limbs = 2*((bits - 1)/FLINT_BITS + 1);
ii = flint_malloc((len + len*(limbs + 1))*sizeof(mp_limb_t));
ptr = (mp_limb_t *) ii + len;
for (i = 0; i < len; i++, ptr += (limbs + 1))
ii[i] = ptr;
a = _fmpz_vec_init(len);
b = _fmpz_vec_init(len);
_fmpz_vec_randtest(a, state, len, bits);
bt = _fmpz_vec_get_fft(ii, a, limbs, len);
for (i = 0; i < len; i++)
mpn_normmod_2expp1(ii[i], limbs);
_fmpz_vec_set_fft(b, len, ii, limbs, bt < 0);
result = (_fmpz_vec_equal(a, b, len));
if (!result)
{
printf("FAIL:\n");
_fmpz_vec_print(a, len), printf("\n\n");
_fmpz_vec_print(b, len), printf("\n\n");
abort();
}
_fmpz_vec_clear(a, len);
_fmpz_vec_clear(b, len);
}
/* convert back and forth unsigned and compare */
for (i = 0; i < 10000; i++)
{
fmpz * a, * b;
mp_bitcnt_t bits;
long len, limbs;
mp_limb_t ** ii, * ptr;
long i, bt;
bits = n_randint(state, 300) + 1;
len = n_randint(state, 300) + 1;
limbs = 2*((bits - 1)/FLINT_BITS + 1);
ii = flint_malloc((len + len*(limbs + 1))*sizeof(mp_limb_t));
ptr = (mp_limb_t *) ii + len;
for (i = 0; i < len; i++, ptr += (limbs + 1))
ii[i] = ptr;
a = _fmpz_vec_init(len);
b = _fmpz_vec_init(len);
_fmpz_vec_randtest_unsigned(a, state, len, bits);
bt = _fmpz_vec_get_fft(ii, a, limbs, len);
_fmpz_vec_set_fft(b, len, ii, limbs, bt < 0);
result = (_fmpz_vec_equal(a, b, len));
if (!result)
{
printf("FAIL:\n");
_fmpz_vec_print(a, len), printf("\n\n");
_fmpz_vec_print(b, len), printf("\n\n");
abort();
}
_fmpz_vec_clear(a, len);
_fmpz_vec_clear(b, len);
}
flint_randclear(state);
_fmpz_cleanup();
printf("PASS\n");
return 0;
}
int
main(void)
{
mp_bitcnt_t depth, w;
flint_rand_t state;
printf("fft/ifft_radix2....");
fflush(stdout);
flint_randinit(state);
_flint_rand_init_gmp(state);
for (depth = 6; depth <= 12; depth++)
{
for (w = 1; w <= 5; w++)
{
mp_size_t n = (1UL<<depth);
mp_size_t limbs = (n*w)/GMP_LIMB_BITS;
mp_size_t size = limbs + 1;
mp_size_t i;
mp_limb_t * ptr;
mp_limb_t ** ii, ** jj, *t1, *t2;
ii = flint_malloc((2*(n + n*size) + 2*size)*sizeof(mp_limb_t));
for (i = 0, ptr = (mp_limb_t *) ii + 2*n; i < 2*n; i++, ptr += size)
{
ii[i] = ptr;
random_fermat(ii[i], state, limbs);
}
t1 = ptr;
t2 = t1 + size;
for (i = 0; i < 2*n; i++)
mpn_normmod_2expp1(ii[i], limbs);
jj = flint_malloc(2*(n + n*size)*sizeof(mp_limb_t));
for (i = 0, ptr = (mp_limb_t *) jj + 2*n; i < 2*n; i++, ptr += size)
{
jj[i] = ptr;
mpn_copyi(jj[i], ii[i], size);
}
fft_radix2(ii, n, w, &t1, &t2);
ifft_radix2(ii, n, w, &t1, &t2);
for (i = 0; i < 2*n; i++)
{
mpn_div_2expmod_2expp1(ii[i], ii[i], limbs, depth + 1);
mpn_normmod_2expp1(ii[i], limbs);
}
for (i = 0; i < 2*n; i++)
{
if (mpn_cmp(ii[i], jj[i], size) != 0)
{
printf("FAIL:\n");
printf("Error in entry %ld\n", i);
abort();
}
}
flint_free(ii);
flint_free(jj);
}
}
flint_randclear(state);
printf("PASS\n");
return 0;
}
void mpir_fft_mulmod_2expp1(mp_ptr r1, mp_srcptr i1, mp_srcptr i2,
mp_size_t r_limbs, mp_bitcnt_t depth, mp_bitcnt_t w)
{
mp_size_t n = (((mp_size_t)1)<<depth);
mp_bitcnt_t bits1 = (r_limbs*GMP_LIMB_BITS)/(2*n);
mp_size_t limb_add, limbs = (n*w)/GMP_LIMB_BITS;
mp_size_t size = limbs + 1;
mp_size_t i, j, ll;
mp_limb_t * ptr;
mp_limb_t ** ii, ** jj, *tt, *t1, *t2, *s1, *r, *ii0, *jj0;
mp_limb_t c;
TMP_DECL;
TMP_MARK;
ii = TMP_BALLOC_MP_PTRS(2*(n + n*size) + 4*n + 5*size);
for (i = 0, ptr = (mp_ptr) ii + 2*n; i < 2*n; i++, ptr += size)
{
ii[i] = ptr;
}
ii0 = ptr;
t1 = ii0 + 2*n;
t2 = t1 + size;
s1 = t2 + size;
r = s1 + size;
tt = r + 2*n;
if (i1 != i2)
{
jj = TMP_BALLOC_MP_PTRS(2*(n + n*size) + 2*n);
for (i = 0, ptr = (mp_ptr) jj + 2*n; i < 2*n; i++, ptr += size)
{
jj[i] = ptr;
}
jj0 = ptr;
} else
{
jj = ii;
jj0 = ii0;
}
j = mpir_fft_split_bits(ii, i1, r_limbs, bits1, limbs);
for ( ; j < 2*n; j++)
mpn_zero(ii[j], limbs + 1);
for (i = 0; i < 2*n; i++)
ii0[i] = ii[i][0];
mpir_fft_negacyclic(ii, n, w, &t1, &t2, &s1);
for (j = 0; j < 2*n; j++)
mpn_normmod_2expp1(ii[j], limbs);
if (i1 != i2)
{
j = mpir_fft_split_bits(jj, i2, r_limbs, bits1, limbs);
for ( ; j < 2*n; j++)
mpn_zero(jj[j], limbs + 1);
for (i = 0; i < 2*n; i++)
jj0[i] = jj[i][0];
mpir_fft_negacyclic(jj, n, w, &t1, &t2, &s1);
}
for (j = 0; j < 2*n; j++)
{
if (i1 != i2) mpn_normmod_2expp1(jj[j], limbs);
c = 2*ii[j][limbs] + jj[j][limbs];
ii[j][limbs] = mpn_mulmod_2expp1_basecase(ii[j], ii[j], jj[j], c, n*w, tt);
}
mpir_ifft_negacyclic(ii, n, w, &t1, &t2, &s1);
mpir_fft_naive_convolution_1(r, ii0, jj0, 2*n);
for (j = 0; j < 2*n; j++)
{
mp_limb_t t, cy2;
mpn_div_2expmod_2expp1(ii[j], ii[j], limbs, depth + 1);
mpn_normmod_2expp1(ii[j], limbs);
t = ii[j][limbs];
ii[j][limbs] = r[j] - ii[j][0];
cy2 = mpn_add_1(ii[j], ii[j], limbs + 1, ii[j][limbs]);
add_ssaaaa(r[j], ii[j][limbs], 0, ii[j][limbs], 0, t);
if (cy2) r[j]++;
}
mpn_zero(r1, r_limbs + 1);
mpir_fft_combine_bits(r1, ii, 2*n - 1, bits1, limbs + 1, r_limbs + 1);
/*
as the negacyclic convolution has effectively done subtractions
some of the coefficients will be negative, so need to subtract p
*/
ll = 0;
limb_add = bits1/GMP_LIMB_BITS;
for (j = 0; j < 2*n - 2; j++)
{
if (r[j])
mpn_sub_1(r1 + ll + 1, r1 + ll + 1, r_limbs - ll, 1);
else if ((mp_limb_signed_t) ii[j][limbs] < 0) /* coefficient was -ve */
{
mpn_sub_1(r1 + ll + 1, r1 + ll + 1, r_limbs - ll, 1);
mpn_sub_1(r1 + ll + limbs + 1, r1 + ll + limbs + 1, r_limbs - limbs - ll, 1);
}
ll += limb_add;
}
/* penultimate coefficient, top bit was already ignored */
if (r[j] || (mp_limb_signed_t) ii[j][limbs] < 0) /* coefficient was -ve */
mpn_sub_1(r1 + ll + 1, r1 + ll + 1, r_limbs - ll, 1);
/* final coefficient wraps around */
if (limb_add)
r1[r_limbs] += mpn_add_n(r1 + r_limbs - limb_add, r1 + r_limbs - limb_add, ii[2*n - 1], limb_add);
c = mpn_sub_n(r1, r1, ii[2*n - 1] + limb_add, limbs + 1 - limb_add);
mpn_addmod_2expp1_1(r1 + limbs + 1 - limb_add, r_limbs - limbs - 1 + limb_add, -c);
mpn_normmod_2expp1(r1, r_limbs);
TMP_FREE;
}
int
main(void)
{
mp_bitcnt_t bits;
mp_size_t j, k, n, w, limbs;
mp_limb_t * nn;
mpz_t p, m1, m2;
FLINT_TEST_INIT(state);
flint_printf("normmod_2expp1....");
fflush(stdout);
_flint_rand_init_gmp(state);
mpz_init(m1);
mpz_init(m2);
mpz_init(p);
/* normalisation mod p = 2^wn + 1 where B divides nw and n is a power of 2 */
for (bits = FLINT_BITS; bits < 32*FLINT_BITS; bits += FLINT_BITS)
{
for (j = 1; j < 32; j++)
{
for (k = 1; k <= GMP_NUMB_BITS; k <<= 1)
{
n = bits/k;
w = j*k;
limbs = (n*w)/GMP_LIMB_BITS;
nn = flint_malloc((limbs + 1)*sizeof(mp_limb_t));
random_fermat(nn, state, limbs);
fermat_to_mpz(m1, nn, limbs);
set_p(p, n, w);
mpn_normmod_2expp1(nn, limbs);
fermat_to_mpz(m2, nn, limbs);
mpz_mod(m1, m1, p);
if (mpz_cmp(m1, m2) != 0)
{
flint_printf("FAIL:\n");
flint_printf("mpn_normmod_2expp1 error\n");
gmp_printf("want %Zx\n\n", m1);
gmp_printf("got %Zx\n", m2);
abort();
}
flint_free(nn);
}
}
}
mpz_clear(m2);
mpz_clear(m1);
mpz_clear(p);
FLINT_TEST_CLEANUP(state);
flint_printf("PASS\n");
return 0;
}
int
main(void)
{
mp_bitcnt_t depth, w;
gmp_randstate_t state;
tests_start();
fflush(stdout);
gmp_randinit_default(state);
for (depth = 6; depth <= 12; depth++)
{
for (w = 1; w <= 5; w++)
{
mp_size_t n = (((mp_limb_t)1)<<depth);
mp_size_t limbs = (n*w)/GMP_LIMB_BITS;
mp_size_t size = limbs + 1;
mp_size_t i;
mp_limb_t * ptr;
mp_limb_t ** ii, ** jj, * t1, * t2, * s1;
ii = malloc((2*(n + n*size) + 3*size)*sizeof(mp_limb_t));
for (i = 0, ptr = (mp_limb_t *) ii + 2*n; i < 2*n; i++, ptr += size)
{
ii[i] = ptr;
mpir_random_fermat(ii[i], state, limbs);
}
t1 = ptr;
t2 = t1 + size;
s1 = t2 + size;
for (i = 0; i < 2*n; i++)
mpn_normmod_2expp1(ii[i], limbs);
jj = malloc(2*(n + n*size)*sizeof(mp_limb_t));
for (i = 0, ptr = (mp_limb_t *) jj + 2*n; i < 2*n; i++, ptr += size)
{
jj[i] = ptr;
mpn_copyi(jj[i], ii[i], size);
}
mpir_fft_negacyclic(ii, n, w, &t1, &t2, &s1);
mpir_ifft_negacyclic(ii, n, w, &t1, &t2, &s1);
for (i = 0; i < 2*n; i++)
{
mpn_div_2expmod_2expp1(ii[i], ii[i], limbs, depth + 1);
mpn_normmod_2expp1(ii[i], limbs);
}
for (i = 0; i < 2*n; i++)
{
if (mpn_cmp(ii[i], jj[i], size) != 0)
{
printf("FAIL:\n");
printf("Error in entry %ld\n", i);
abort();
}
}
free(ii);
free(jj);
}
}
gmp_randclear(state);
tests_end();
return 0;
}
int
main(void)
{
mp_bitcnt_t depth, w;
int iters;
flint_rand_t state;
printf("mulmod_2expp1....");
fflush(stdout);
flint_randinit(state);
_flint_rand_init_gmp(state);
for (iters = 0; iters < 100; iters++)
{
for (depth = 6; depth <= 18; depth++)
{
for (w = 1; w <= 2; w++)
{
mp_size_t n = (1UL<<depth);
mp_bitcnt_t bits = n*w;
mp_size_t int_limbs = bits/FLINT_BITS;
mp_size_t j;
mp_limb_t c, * i1, * i2, * r1, * r2, * tt;
i1 = flint_malloc(6*(int_limbs+1)*sizeof(mp_limb_t));
i2 = i1 + int_limbs + 1;
r1 = i2 + int_limbs + 1;
r2 = r1 + int_limbs + 1;
tt = r2 + int_limbs + 1;
random_fermat(i1, state, int_limbs);
random_fermat(i2, state, int_limbs);
mpn_normmod_2expp1(i1, int_limbs);
mpn_normmod_2expp1(i2, int_limbs);
fft_mulmod_2expp1(r2, i1, i2, n, w, tt);
c = i1[int_limbs] + 2*i2[int_limbs];
c = mpn_mulmod_2expp1(r1, i1, i2, c, int_limbs*FLINT_BITS, tt);
for (j = 0; j < int_limbs; j++)
{
if (r1[j] != r2[j])
{
printf("error in limb %ld, %lx != %lx\n", j, r1[j], r2[j]);
abort();
}
}
if (c != r2[int_limbs])
{
printf("error in limb %ld, %lx != %lx\n", j, c, r2[j]);
abort();
}
flint_free(i1);
}
}
}
/* test squaring */
for (iters = 0; iters < 100; iters++)
{
for (depth = 6; depth <= 18; depth++)
{
for (w = 1; w <= 2; w++)
{
mp_size_t n = (1UL<<depth);
mp_bitcnt_t bits = n*w;
mp_size_t int_limbs = bits/FLINT_BITS;
mp_size_t j;
mp_limb_t c, * i1, * r1, * r2, * tt;
i1 = flint_malloc(5*(int_limbs+1)*sizeof(mp_limb_t));
r1 = i1 + int_limbs + 1;
r2 = r1 + int_limbs + 1;
tt = r2 + int_limbs + 1;
random_fermat(i1, state, int_limbs);
mpn_normmod_2expp1(i1, int_limbs);
fft_mulmod_2expp1(r2, i1, i1, n, w, tt);
c = i1[int_limbs] + 2*i1[int_limbs];
c = mpn_mulmod_2expp1(r1, i1, i1, c, int_limbs*FLINT_BITS, tt);
for (j = 0; j < int_limbs; j++)
{
if (r1[j] != r2[j])
{
printf("error in limb %ld, %lx != %lx\n", j, r1[j], r2[j]);
abort();
}
}
if (c != r2[int_limbs])
{
printf("error in limb %ld, %lx != %lx\n", j, c, r2[j]);
abort();
}
flint_free(i1);
}
}
}
flint_randclear(state);
printf("PASS\n");
return 0;
}