/*
* Swap half-rows of 2^n * (2*2^n) matrix.
* FORWARD_CYCLE: even/odd permutation of the halfrows.
* BACKWARD_CYCLE: reverse the even/odd permutation.
*/
static int
swap_halfrows_pow2(mpd_uint_t *matrix, mpd_size_t rows, mpd_size_t cols, int dir)
{
mpd_uint_t buf1[BUFSIZE];
mpd_uint_t buf2[BUFSIZE];
mpd_uint_t *readbuf, *writebuf, *hp;
mpd_size_t *done, dbits;
mpd_size_t b = BUFSIZE, stride;
mpd_size_t hn, hmax; /* halfrow number */
mpd_size_t m, r=0;
mpd_size_t offset;
mpd_size_t next;
assert(cols == mul_size_t(2, rows));
if (dir == FORWARD_CYCLE) {
r = rows;
}
else if (dir == BACKWARD_CYCLE) {
r = 2;
}
else {
abort(); /* GCOV_NOT_REACHED */
}
m = cols - 1;
hmax = rows; /* cycles start at odd halfrows */
dbits = 8 * sizeof *done;
if ((done = mpd_calloc(hmax/(sizeof *done) + 1, sizeof *done)) == NULL) {
return 0;
}
for (hn = 1; hn <= hmax; hn += 2) {
if (done[hn/dbits] & mpd_bits[hn%dbits]) {
continue;
}
readbuf = buf1; writebuf = buf2;
for (offset = 0; offset < cols/2; offset += b) {
stride = (offset + b < cols/2) ? b : cols/2-offset;
hp = matrix + hn*cols/2;
memcpy(readbuf, hp+offset, stride*(sizeof *readbuf));
pointerswap(&readbuf, &writebuf);
next = mulmod_size_t(hn, r, m);
hp = matrix + next*cols/2;
while (next != hn) {
memcpy(readbuf, hp+offset, stride*(sizeof *readbuf));
memcpy(hp+offset, writebuf, stride*(sizeof *writebuf));
pointerswap(&readbuf, &writebuf);
done[next/dbits] |= mpd_bits[next%dbits];
next = mulmod_size_t(next, r, m);
hp = matrix + next*cols/2;
}
memcpy(hp+offset, writebuf, stride*(sizeof *writebuf));
done[hn/dbits] |= mpd_bits[hn%dbits];
}
}
mpd_free(done);
return 1;
}
/*
* Knuth, TAOCP Volume 2, 4.3.1:
* q, r := quotient and remainder of uconst (len nplusm)
* divided by vconst (len n)
* nplusm > n
*
* If r is not NULL, r will contain the remainder. If r is NULL, the
* return value indicates if there is a remainder: 1 for true, 0 for
* false. A return value of -1 indicates an error.
*/
int
_mpd_basedivmod(mpd_uint_t *q, mpd_uint_t *r,
const mpd_uint_t *uconst, const mpd_uint_t *vconst,
mpd_size_t nplusm, mpd_size_t n)
{
mpd_uint_t ustatic[MPD_MINALLOC_MAX];
mpd_uint_t vstatic[MPD_MINALLOC_MAX];
mpd_uint_t *u = ustatic;
mpd_uint_t *v = vstatic;
mpd_uint_t d, qhat, rhat, w2[2];
mpd_uint_t hi, lo, x;
mpd_uint_t carry;
mpd_size_t i, j, m;
int retval = 0;
assert(n > 1 && nplusm >= n);
m = sub_size_t(nplusm, n);
/* D1: normalize */
d = MPD_RADIX / (vconst[n-1] + 1);
if (nplusm >= MPD_MINALLOC_MAX) {
if ((u = mpd_calloc(nplusm+1, sizeof *u)) == NULL) {
return -1;
}
}
if (n >= MPD_MINALLOC_MAX) {
if ((v = mpd_calloc(n+1, sizeof *v)) == NULL) {
mpd_free(u);
return -1;
}
}
_mpd_shortmul(u, uconst, nplusm, d);
_mpd_shortmul(v, vconst, n, d);
/* D2: loop */
rhat = 0;
for (j=m; j != MPD_SIZE_MAX; j--) {
/* D3: calculate qhat and rhat */
rhat = _mpd_shortdiv(w2, u+j+n-1, 2, v[n-1]);
qhat = w2[1] * MPD_RADIX + w2[0];
while (1) {
if (qhat < MPD_RADIX) {
_mpd_singlemul(w2, qhat, v[n-2]);
if (w2[1] <= rhat) {
if (w2[1] != rhat || w2[0] <= u[j+n-2]) {
break;
}
}
}
qhat -= 1;
rhat += v[n-1];
if (rhat < v[n-1] || rhat >= MPD_RADIX) {
break;
}
}
/* D4: multiply and subtract */
carry = 0;
for (i=0; i <= n; i++) {
_mpd_mul_words(&hi, &lo, qhat, v[i]);
lo = carry + lo;
if (lo < carry) hi++;
_mpd_div_words_r(&hi, &lo, hi, lo);
x = u[i+j] - lo;
carry = (u[i+j] < x);
u[i+j] = carry ? x+MPD_RADIX : x;
carry += hi;
}
q[j] = qhat;
/* D5: test remainder */
if (carry) {
q[j] -= 1;
/* D6: add back */
(void)_mpd_baseadd(u+j, u+j, v, n+1, n);
}
}
/* D8: unnormalize */
if (r != NULL) {
_mpd_shortdiv(r, u, n, d);
/* we are not interested in the return value here */
retval = 0;
}
else {
retval = !_mpd_isallzero(u, n);
}
if (u != ustatic) mpd_free(u);
if (v != vstatic) mpd_free(v);
return retval;
}