/* in1 += x*in2 assuming both in1 and in2 are revbin'd. */
void
_fmpz_vec_add_rev(fmpz * in1, fmpz * in2, long bits)
{
long i;
for (i = 0; i < (1L << bits) - 1; i++)
{
long j = n_revbin(n_revbin(i, bits) + 1, bits);
fmpz_add(in1 + j, in1 + j, in2 + i);
}
}
/*
Switches the coefficients of poly in of length 2^bits into a
poly out of length len.
*/
void
revbin2(fmpz * out, const fmpz * in, long len, long bits)
{
long i;
for (i = 0; i < len; i++)
out[i] = in[n_revbin(i, bits)];
}
/*
Switches the coefficients of poly in of length len into a
poly out of length 2^bits.
*/
void
revbin1(fmpz * out, const fmpz * in, slong len, slong bits)
{
slong i;
for (i = 0; i < len; i++)
out[n_revbin(i, bits)] = in[i];
}
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);
}
}
