/*
* arccosh(x) == log [ x + sqrt(x^2 - 1) ]
*
* x >= 1.0
*/
void m_apm_arccosh(M_APM rr, int places, M_APM aa)
{
M_APM tmp1, tmp2;
int ii;
ii = m_apm_compare(aa, MM_One);
if (ii == -1) /* x < 1 */
{
M_apm_log_error_msg(M_APM_RETURN, "\'m_apm_arccosh\', Argument < 1");
M_set_to_zero(rr);
return;
}
tmp1 = M_get_stack_var();
tmp2 = M_get_stack_var();
m_apm_multiply(tmp1, aa, aa);
m_apm_subtract(tmp2, tmp1, MM_One);
m_apm_sqrt(tmp1, (places + 6), tmp2);
m_apm_add(tmp2, aa, tmp1);
m_apm_log(rr, places, tmp2);
M_restore_stack(2);
}
void M_apm_round_fixpt(M_APM btmp, int places, M_APM atmp)
{
int xp, ii;
xp = atmp->m_apm_exponent;
ii = xp + places - 1;
M_set_to_zero(btmp); /* assume number is too small so the net result is 0 */
if (ii >= 0)
{
m_apm_round(btmp, ii, atmp);
}
else
{
if (ii == -1) /* next digit is significant which may round up */
{
if (atmp->m_apm_data[0] >= 50) /* digit >= 5, round up */
{
m_apm_copy(btmp, atmp);
btmp->m_apm_data[0] = 10;
btmp->m_apm_exponent += 1;
btmp->m_apm_datalength = 1;
M_apm_normalize(btmp);
}
}
}
}
void m_apm_integer_divide(M_APM rr, M_APM aa, M_APM bb)
{
/*
* we must use this divide function since the
* faster divide function using the reciprocal
* will round the result (possibly changing
* nnm.999999... --> nn(m+1).0000 which would
* invalidate the 'integer_divide' goal).
*/
if (aa->m_apm_error || bb->m_apm_error)
{
M_set_to_error(rr);
return;
}
M_apm_sdivide(rr, 4, aa, bb);
if (rr->m_apm_exponent <= 0) /* result is 0 */
{
M_set_to_zero(rr);
}
else
{
if (rr->m_apm_datalength > rr->m_apm_exponent)
{
rr->m_apm_datalength = rr->m_apm_exponent;
M_apm_normalize(rr);
}
}
}
/*
Calculate arctan using the identity :
x
arctan (x) == arcsin [ --------------- ]
sqrt(1 + x^2)
*/
void m_apm_arctan(M_APM rr, int places, M_APM xx)
{
M_APM tmp8, tmp9;
if (xx->m_apm_sign == 0) /* input == 0 ?? */
{
M_set_to_zero(rr);
return;
}
if (xx->m_apm_exponent <= -4) /* input close to 0 ?? */
{
M_arctan_near_0(rr, places, xx);
return;
}
if (xx->m_apm_exponent >= 4) /* large input */
{
M_arctan_large_input(rr, places, xx);
return;
}
tmp8 = M_get_stack_var();
tmp9 = M_get_stack_var();
m_apm_multiply(tmp9, xx, xx);
m_apm_add(tmp8, tmp9, MM_One);
m_apm_sqrt(tmp9, (places + 6), tmp8);
m_apm_divide(tmp8, (places + 6), xx, tmp9);
m_apm_arcsin(rr, places, tmp8);
M_restore_stack(2);
}
/*
* arctanh(x) == 0.5 * log [ (1 + x) / (1 - x) ]
*
* |x| < 1.0
*/
void m_apm_arctanh(M_APM rr, int places, M_APM aa)
{
M_APM tmp1, tmp2, tmp3;
int ii, local_precision;
tmp1 = M_get_stack_var();
m_apm_absolute_value(tmp1, aa);
ii = m_apm_compare(tmp1, MM_One);
if (ii >= 0) /* |x| >= 1.0 */
{
M_apm_log_error_msg(M_APM_RETURN, "\'m_apm_arctanh\', |Argument| >= 1");
M_set_to_zero(rr);
M_restore_stack(1);
return;
}
tmp2 = M_get_stack_var();
tmp3 = M_get_stack_var();
local_precision = places + 8;
m_apm_add(tmp1, MM_One, aa);
m_apm_subtract(tmp2, MM_One, aa);
m_apm_divide(tmp3, local_precision, tmp1, tmp2);
m_apm_log(tmp2, local_precision, tmp3);
m_apm_multiply(tmp1, tmp2, MM_0_5);
m_apm_round(rr, places, tmp1);
M_restore_stack(3);
}
/*
* arcsinh(x) == log [ x + sqrt(x^2 + 1) ]
*
* also, use arcsinh(-x) == -arcsinh(x)
*/
void m_apm_arcsinh(M_APM rr, int places, M_APM aa)
{
M_APM tmp0, tmp1, tmp2;
/* result is 0 if input is 0 */
if (aa->m_apm_sign == 0)
{
M_set_to_zero(rr);
return;
}
tmp0 = M_get_stack_var();
tmp1 = M_get_stack_var();
tmp2 = M_get_stack_var();
m_apm_absolute_value(tmp0, aa);
m_apm_multiply(tmp1, tmp0, tmp0);
m_apm_add(tmp2, tmp1, MM_One);
m_apm_sqrt(tmp1, (places + 6), tmp2);
m_apm_add(tmp2, tmp0, tmp1);
m_apm_log(rr, places, tmp2);
rr->m_apm_sign = aa->m_apm_sign; /* fix final sign */
M_restore_stack(3);
}
void m_apm_divide(M_APM rr, int places, M_APM aa, M_APM bb)
{
M_APM tmp0, tmp1;
int sn, nexp, dplaces;
sn = aa->m_apm_sign * bb->m_apm_sign;
if (sn == 0) /* one number is zero, result is zero */
{
if (bb->m_apm_sign == 0)
{
M_apm_log_error_msg(M_APM_RETURN,
"Warning! ... \'m_apm_divide\', Divide by 0");
}
M_set_to_zero(rr);
return;
}
/*
* Use the original 'Knuth' method for smaller divides. On the
* author's system, this was the *approx* break even point before
* the reciprocal method used below became faster.
*/
if (places < 250)
{
M_apm_sdivide(rr, places, aa, bb);
return;
}
/* mimic the decimal place behavior of the original divide */
nexp = aa->m_apm_exponent - bb->m_apm_exponent;
if (nexp > 0)
dplaces = nexp + places;
else
dplaces = places;
tmp0 = M_get_stack_var();
tmp1 = M_get_stack_var();
m_apm_reciprocal(tmp0, (dplaces + 8), bb);
m_apm_multiply(tmp1, tmp0, aa);
m_apm_round(rr, dplaces, tmp1);
M_restore_stack(2);
}
/*
* return the nearest integer <= input
*/
void m_apm_floor(M_APM bb, M_APM aa)
{
M_APM mtmp;
m_apm_copy(bb, aa);
if (m_apm_is_integer(bb)) /* if integer, we're done */
return;
if (bb->m_apm_exponent <= 0) /* if |bb| < 1, result is -1 or 0 */
{
if (bb->m_apm_sign < 0)
m_apm_negate(bb, MM_One);
else
M_set_to_zero(bb);
return;
}
if (bb->m_apm_sign < 0)
{
mtmp = M_get_stack_var();
m_apm_negate(mtmp, bb);
mtmp->m_apm_datalength = mtmp->m_apm_exponent;
M_apm_normalize(mtmp);
m_apm_add(bb, mtmp, MM_One);
bb->m_apm_sign = -1;
M_restore_stack(1);
}
else
{
bb->m_apm_datalength = bb->m_apm_exponent;
M_apm_normalize(bb);
}
}
void m_apm_reciprocal(M_APM rr, int places, M_APM aa)
{
M_APM last_x, guess, tmpN, tmp1, tmp2;
int ii, bflag, dplaces, nexp, tolerance;
if (aa->m_apm_sign == 0)
{
M_apm_log_error_msg(M_APM_RETURN,
"Warning! ... \'m_apm_reciprocal\', Input = 0");
M_set_to_zero(rr);
return;
}
last_x = M_get_stack_var();
guess = M_get_stack_var();
tmpN = M_get_stack_var();
tmp1 = M_get_stack_var();
tmp2 = M_get_stack_var();
m_apm_absolute_value(tmpN, aa);
/*
normalize the input number (make the exponent 0) so
the 'guess' below will not over/under flow on large
magnitude exponents.
*/
nexp = aa->m_apm_exponent;
tmpN->m_apm_exponent -= nexp;
m_apm_set_double(guess, (1.0 / m_apm_get_double(tmpN)));
tolerance = places + 4;
dplaces = places + 16;
bflag = FALSE;
m_apm_negate(last_x, MM_Ten);
/* Use the following iteration to calculate the reciprocal :
X = X * [ 2 - N * X ]
n+1
*/
ii = 0;
while (TRUE)
{
m_apm_multiply(tmp1, tmpN, guess);
m_apm_subtract(tmp2, MM_Two, tmp1);
m_apm_multiply(tmp1, tmp2, guess);
if (bflag)
break;
m_apm_round(guess, dplaces, tmp1);
/* force at least 2 iterations so 'last_x' has valid data */
if (ii != 0)
{
m_apm_subtract(tmp2, guess, last_x);
if (tmp2->m_apm_sign == 0)
break;
/*
* if we are within a factor of 4 on the error term,
* we will be accurate enough after the *next* iteration
* is complete.
*/
if ((-4 * tmp2->m_apm_exponent) > tolerance)
bflag = TRUE;
}
m_apm_copy(last_x, guess);
ii++;
}
m_apm_round(rr, places, tmp1);
rr->m_apm_exponent -= nexp;
rr->m_apm_sign = aa->m_apm_sign;
M_restore_stack(5);
}
void m_apm_arccos(M_APM r, int places, M_APM x)
{
M_APM tmp0, tmp1, tmp2, tmp3, current_x;
int ii, maxiter, maxp, tolerance, local_precision;
current_x = M_get_stack_var();
tmp0 = M_get_stack_var();
tmp1 = M_get_stack_var();
tmp2 = M_get_stack_var();
tmp3 = M_get_stack_var();
m_apm_absolute_value(tmp0, x);
ii = m_apm_compare(tmp0, MM_One);
if (ii == 1) /* |x| > 1 */
{
M_apm_log_error_msg(M_APM_RETURN, "\'m_apm_arccos\', |Argument| > 1");
M_set_to_zero(r);
M_restore_stack(5);
return;
}
if (ii == 0) /* |x| == 1, arccos = 0, PI */
{
if (x->m_apm_sign == 1)
{
M_set_to_zero(r);
}
else
{
M_check_PI_places(places);
m_apm_round(r, places, MM_lc_PI);
}
M_restore_stack(5);
return;
}
if (m_apm_compare(tmp0, MM_0_85) == 1) /* check if > 0.85 */
{
M_cos_to_sin(tmp2, (places + 4), x);
if (x->m_apm_sign == 1)
{
m_apm_arcsin(r, places, tmp2);
}
else
{
M_check_PI_places(places);
m_apm_arcsin(tmp3, (places + 4), tmp2);
m_apm_subtract(tmp1, MM_lc_PI, tmp3);
m_apm_round(r, places, tmp1);
}
M_restore_stack(5);
return;
}
if (x->m_apm_sign == 0) /* input == 0 ?? */
{
M_check_PI_places(places);
m_apm_round(r, places, MM_lc_HALF_PI);
M_restore_stack(5);
return;
}
if (x->m_apm_exponent <= -4) /* input close to 0 ?? */
{
M_arccos_near_0(r, places, x);
M_restore_stack(5);
return;
}
tolerance = -(places + 4);
maxp = places + 8;
local_precision = 18;
/*
* compute the maximum number of iterations
* that should be needed to calculate to
* the desired accuracy. [ constant below ~= 1 / log(2) ]
*/
maxiter = (int)(log((double)(places + 2)) * 1.442695) + 3;
if (maxiter < 5)
maxiter = 5;
M_get_acos_guess(current_x, x);
/* Use the following iteration to solve for arc-cos :
cos(X) - N
X = X + ------------
n+1 sin(X)
*/
ii = 0;
while (TRUE)
{
M_4x_cos(tmp1, local_precision, current_x);
M_cos_to_sin(tmp2, local_precision, tmp1);
if (tmp2->m_apm_sign != 0)
tmp2->m_apm_sign = current_x->m_apm_sign;
m_apm_subtract(tmp3, tmp1, x);
m_apm_divide(tmp0, local_precision, tmp3, tmp2);
m_apm_add(tmp2, current_x, tmp0);
m_apm_copy(current_x, tmp2);
if (ii != 0)
{
if (((2 * tmp0->m_apm_exponent) < tolerance) || (tmp0->m_apm_sign == 0))
break;
}
if (++ii == maxiter)
{
M_apm_log_error_msg(M_APM_RETURN,
"\'m_apm_arccos\', max iteration count reached");
break;
}
local_precision *= 2;
if (local_precision > maxp)
local_precision = maxp;
}
m_apm_round(r, places, current_x);
M_restore_stack(5);
}
void M_apm_sdivide(M_APM r, int places, M_APM a, M_APM b)
{
int j, k, m, b0, sign, nexp, indexr, icompare, iterations;
long trial_numer;
void *vp;
if (M_div_firsttime)
{
M_div_firsttime = FALSE;
M_div_worka = m_apm_init();
M_div_workb = m_apm_init();
M_div_tmp7 = m_apm_init();
M_div_tmp8 = m_apm_init();
M_div_tmp9 = m_apm_init();
}
sign = a->m_apm_sign * b->m_apm_sign;
if (sign == 0) /* one number is zero, result is zero */
{
if (b->m_apm_sign == 0)
{
M_apm_log_error_msg(M_APM_RETURN, "\'M_apm_sdivide\', Divide by 0");
}
M_set_to_zero(r);
return;
}
/*
* Knuth step D1. Since base = 100, base / 2 = 50.
* (also make the working copies positive)
*/
if (b->m_apm_data[0] >= 50)
{
m_apm_absolute_value(M_div_worka, a);
m_apm_absolute_value(M_div_workb, b);
}
else /* 'normal' step D1 */
{
k = 100 / (b->m_apm_data[0] + 1);
m_apm_set_long(M_div_tmp9, (long)k);
m_apm_multiply(M_div_worka, M_div_tmp9, a);
m_apm_multiply(M_div_workb, M_div_tmp9, b);
M_div_worka->m_apm_sign = 1;
M_div_workb->m_apm_sign = 1;
}
/* setup trial denominator for step D3 */
b0 = 100 * (int)M_div_workb->m_apm_data[0];
if (M_div_workb->m_apm_datalength >= 3)
b0 += M_div_workb->m_apm_data[1];
nexp = M_div_worka->m_apm_exponent - M_div_workb->m_apm_exponent;
if (nexp > 0)
iterations = nexp + places + 1;
else
iterations = places + 1;
k = (iterations + 1) >> 1; /* required size of result, in bytes */
if (k > r->m_apm_malloclength)
{
if ((vp = MAPM_REALLOC(r->m_apm_data, (k + 32))) == NULL)
{
/* fatal, this does not return */
M_apm_log_error_msg(M_APM_FATAL, "\'M_apm_sdivide\', Out of memory");
}
r->m_apm_malloclength = k + 28;
r->m_apm_data = (UCHAR *)vp;
}
/* clear the exponent in the working copies */
M_div_worka->m_apm_exponent = 0;
M_div_workb->m_apm_exponent = 0;
/* if numbers are equal, ratio == 1.00000... */
if ((icompare = m_apm_compare(M_div_worka, M_div_workb)) == 0)
{
iterations = 1;
r->m_apm_data[0] = 10;
nexp++;
}
else /* ratio not 1, do the real division */
{
if (icompare == 1) /* numerator > denominator */
{
nexp++; /* to adjust the final exponent */
M_div_worka->m_apm_exponent += 1; /* multiply numerator by 10 */
}
else /* numerator < denominator */
{
M_div_worka->m_apm_exponent += 2; /* multiply numerator by 100 */
}
indexr = 0;
m = 0;
while (TRUE)
{
/*
* Knuth step D3. Only use the 3rd -> 6th digits if the number
* actually has that many digits.
*/
trial_numer = 10000L * (long)M_div_worka->m_apm_data[0];
if (M_div_worka->m_apm_datalength >= 5)
{
trial_numer += 100 * M_div_worka->m_apm_data[1]
+ M_div_worka->m_apm_data[2];
}
else
{
if (M_div_worka->m_apm_datalength >= 3)
trial_numer += 100 * M_div_worka->m_apm_data[1];
}
j = (int)(trial_numer / b0);
/*
* Since the library 'normalizes' all the results, we need
* to look at the exponent of the number to decide if we
* have a lead in 0n or 00.
*/
if ((k = 2 - M_div_worka->m_apm_exponent) > 0)
{
while (TRUE)
{
j /= 10;
if (--k == 0)
break;
}
}
if (j == 100) /* qhat == base ?? */
j = 99; /* if so, decrease by 1 */
m_apm_set_long(M_div_tmp8, (long)j);
m_apm_multiply(M_div_tmp7, M_div_tmp8, M_div_workb);
/*
* Compare our q-hat (j) against the desired number.
* j is either correct, 1 too large, or 2 too large
* per Theorem B on pg 272 of Art of Compter Programming,
* Volume 2, 3rd Edition.
*
* The above statement is only true if using the 2 leading
* digits of the numerator and the leading digit of the
* denominator. Since we are using the (3) leading digits
* of the numerator and the (2) leading digits of the
* denominator, we eliminate the case where our q-hat is
* 2 too large, (and q-hat being 1 too large is quite remote).
*/
if (m_apm_compare(M_div_tmp7, M_div_worka) == 1)
{
j--;
m_apm_subtract(M_div_tmp8, M_div_tmp7, M_div_workb);
m_apm_copy(M_div_tmp7, M_div_tmp8);
}
/*
* Since we know q-hat is correct, step D6 is unnecessary.
*
* Store q-hat, step D5. Since D6 is unnecessary, we can
* do D5 before D4 and decide if we are done.
*/
r->m_apm_data[indexr++] = (UCHAR)j; /* j == 'qhat' */
m += 2;
if (m >= iterations)
break;
/* step D4 */
m_apm_subtract(M_div_tmp9, M_div_worka, M_div_tmp7);
/*
* if the subtraction yields zero, the division is exact
* and we are done early.
*/
if (M_div_tmp9->m_apm_sign == 0)
{
iterations = m;
break;
}
/* multiply by 100 and re-save */
M_div_tmp9->m_apm_exponent += 2;
m_apm_copy(M_div_worka, M_div_tmp9);
}
}
r->m_apm_sign = sign;
r->m_apm_exponent = nexp;
r->m_apm_datalength = iterations;
M_apm_normalize(r);
}
void m_apm_arctan2(M_APM rr, int places, M_APM yy, M_APM xx)
{
M_APM tmp5, tmp6, tmp7;
int ix, iy;
iy = yy->m_apm_sign;
ix = xx->m_apm_sign;
if (ix == 0) /* x == 0 */
{
if (iy == 0) /* y == 0 */
{
M_apm_log_error_msg(M_APM_RETURN, "\'m_apm_arctan2\', Both Inputs = 0");
M_set_to_zero(rr);
return;
}
M_check_PI_places(places);
m_apm_round(rr, places, MM_lc_HALF_PI);
rr->m_apm_sign = iy;
return;
}
if (iy == 0)
{
if (ix == 1)
{
M_set_to_zero(rr);
}
else
{
M_check_PI_places(places);
m_apm_round(rr, places, MM_lc_PI);
}
return;
}
/*
* the special cases have been handled, now do the real work
*/
tmp5 = M_get_stack_var();
tmp6 = M_get_stack_var();
tmp7 = M_get_stack_var();
m_apm_divide(tmp6, (places + 6), yy, xx);
m_apm_arctan(tmp5, (places + 6), tmp6);
if (ix == 1) /* 'x' is positive */
{
m_apm_round(rr, places, tmp5);
}
else /* 'x' is negative */
{
M_check_PI_places(places);
if (iy == 1) /* 'y' is positive */
{
m_apm_add(tmp7, tmp5, MM_lc_PI);
m_apm_round(rr, places, tmp7);
}
else /* 'y' is negative */
{
m_apm_subtract(tmp7, tmp5, MM_lc_PI);
m_apm_round(rr, places, tmp7);
}
}
M_restore_stack(3);
}
/*
* From Knuth, The Art of Computer Programming:
*
* This is the binary GCD algorithm as described
* in the book (Algorithm B)
*/
void m_apm_gcd(M_APM r, M_APM u, M_APM v)
{
M_APM tmpM, tmpN, tmpT, tmpU, tmpV;
int kk, kr, mm;
long pow_2;
/* 'is_integer' will return 0 || 1 */
if ((m_apm_is_integer(u) + m_apm_is_integer(v)) != 2)
{
M_apm_log_error_msg(M_APM_RETURN,
"Warning! \'m_apm_gcd\', Non-integer input");
M_set_to_zero(r);
return;
}
if (u->m_apm_sign == 0)
{
m_apm_absolute_value(r, v);
return;
}
if (v->m_apm_sign == 0)
{
m_apm_absolute_value(r, u);
return;
}
tmpM = M_get_stack_var();
tmpN = M_get_stack_var();
tmpT = M_get_stack_var();
tmpU = M_get_stack_var();
tmpV = M_get_stack_var();
m_apm_absolute_value(tmpU, u);
m_apm_absolute_value(tmpV, v);
/* Step B1 */
kk = 0;
while (TRUE)
{
mm = 1;
if (m_apm_is_odd(tmpU))
break;
mm = 0;
if (m_apm_is_odd(tmpV))
break;
m_apm_multiply(tmpN, MM_0_5, tmpU);
m_apm_copy(tmpU, tmpN);
m_apm_multiply(tmpN, MM_0_5, tmpV);
m_apm_copy(tmpV, tmpN);
kk++;
}
/* Step B2 */
if (mm)
{
m_apm_negate(tmpT, tmpV);
goto B4;
}
m_apm_copy(tmpT, tmpU);
/* Step: */
B3:
m_apm_multiply(tmpN, MM_0_5, tmpT);
m_apm_copy(tmpT, tmpN);
/* Step: */
B4:
if (m_apm_is_even(tmpT))
goto B3;
/* Step B5 */
if (tmpT->m_apm_sign == 1)
m_apm_copy(tmpU, tmpT);
else
m_apm_negate(tmpV, tmpT);
/* Step B6 */
m_apm_subtract(tmpT, tmpU, tmpV);
if (tmpT->m_apm_sign != 0)
goto B3;
/*
* result = U * 2 ^ kk
*/
if (kk == 0)
m_apm_copy(r, tmpU);
else
{
if (kk == 1)
m_apm_multiply(r, tmpU, MM_Two);
if (kk == 2)
m_apm_multiply(r, tmpU, MM_Four);
if (kk >= 3)
{
mm = kk / 28;
kr = kk % 28;
pow_2 = 1L << kr;
if (mm == 0)
{
m_apm_set_long(tmpN, pow_2);
m_apm_multiply(r, tmpU, tmpN);
}
else
{
m_apm_copy(tmpN, MM_One);
m_apm_set_long(tmpM, 0x10000000L); /* 2 ^ 28 */
while (TRUE)
{
m_apm_multiply(tmpT, tmpN, tmpM);
m_apm_copy(tmpN, tmpT);
if (--mm == 0)
break;
}
if (kr == 0)
{
m_apm_multiply(r, tmpU, tmpN);
}
else
{
m_apm_set_long(tmpM, pow_2);
m_apm_multiply(tmpT, tmpN, tmpM);
m_apm_multiply(r, tmpU, tmpT);
}
}
}
}
M_restore_stack(5);
}
void m_apm_sqrt(M_APM rr, int places, M_APM aa)
{
M_APM last_x, guess, tmpN, tmp7, tmp8, tmp9;
int ii, bflag, nexp, tolerance, dplaces;
if (aa->m_apm_sign <= 0)
{
if (aa->m_apm_sign == -1)
{
M_apm_log_error_msg(M_APM_RETURN, "\'m_apm_sqrt\', Negative argument");
}
M_set_to_zero(rr);
return;
}
last_x = M_get_stack_var();
guess = M_get_stack_var();
tmpN = M_get_stack_var();
tmp7 = M_get_stack_var();
tmp8 = M_get_stack_var();
tmp9 = M_get_stack_var();
m_apm_copy(tmpN, aa);
/*
normalize the input number (make the exponent near 0) so
the 'guess' function will not over/under flow on large
magnitude exponents.
*/
nexp = aa->m_apm_exponent / 2;
tmpN->m_apm_exponent -= 2 * nexp;
M_get_sqrt_guess(guess, tmpN); /* actually gets 1/sqrt guess */
tolerance = places + 4;
dplaces = places + 16;
bflag = FALSE;
m_apm_negate(last_x, MM_Ten);
/* Use the following iteration to calculate 1 / sqrt(N) :
X = 0.5 * X * [ 3 - N * X^2 ]
n+1
*/
ii = 0;
while (TRUE)
{
m_apm_multiply(tmp9, tmpN, guess);
m_apm_multiply(tmp8, tmp9, guess);
m_apm_round(tmp7, dplaces, tmp8);
m_apm_subtract(tmp9, MM_Three, tmp7);
m_apm_multiply(tmp8, tmp9, guess);
m_apm_multiply(tmp9, tmp8, MM_0_5);
if (bflag)
break;
m_apm_round(guess, dplaces, tmp9);
/* force at least 2 iterations so 'last_x' has valid data */
if (ii != 0)
{
m_apm_subtract(tmp7, guess, last_x);
if (tmp7->m_apm_sign == 0)
break;
/*
* if we are within a factor of 4 on the error term,
* we will be accurate enough after the *next* iteration
* is complete. (note that the sign of the exponent on
* the error term will be a negative number).
*/
if ((-4 * tmp7->m_apm_exponent) > tolerance)
bflag = TRUE;
}
m_apm_copy(last_x, guess);
ii++;
}
/*
* multiply by the starting number to get the final
* sqrt and then adjust the exponent since we found
* the sqrt of the normalized number.
*/
m_apm_multiply(tmp8, tmp9, tmpN);
m_apm_round(rr, places, tmp8);
rr->m_apm_exponent += nexp;
M_restore_stack(6);
}
void m_apm_exp(M_APM r, int places, M_APM x)
{
M_APM tmp7, tmp8, tmp9;
int dplaces, nn, ii;
if (MM_firsttime1)
{
MM_firsttime1 = FALSE;
MM_exp_log2R = m_apm_init();
MM_exp_512R = m_apm_init();
m_apm_set_string(MM_exp_log2R, "1.44269504089"); /* ~ 1 / log(2) */
m_apm_set_string(MM_exp_512R, "1.953125E-3"); /* 1 / 512 */
}
tmp7 = M_get_stack_var();
tmp8 = M_get_stack_var();
tmp9 = M_get_stack_var();
if (x->m_apm_sign == 0) /* if input == 0, return '1' */
{
m_apm_copy(r, MM_One);
M_restore_stack(3);
return;
}
if (x->m_apm_exponent <= -3) /* already small enough so call _raw directly */
{
M_raw_exp(tmp9, (places + 6), x);
m_apm_round(r, places, tmp9);
M_restore_stack(3);
return;
}
/*
From David H. Bailey's MPFUN Fortran package :
exp (t) = (1 + r + r^2 / 2! + r^3 / 3! + r^4 / 4! ...) ^ q * 2 ^ n
where q = 256, r = t' / q, t' = t - n Log(2) and where n is chosen so
that -0.5 Log(2) < t' <= 0.5 Log(2). Reducing t mod Log(2) and
dividing by 256 insures that -0.001 < r <= 0.001, which accelerates
convergence in the above series.
I use q = 512 and also limit how small 'r' can become. The 'r' used
here is limited in magnitude from 1.95E-4 < |r| < 1.35E-3. Forcing
'r' into a narrow range keeps the algorithm 'well behaved'.
( the range is [0.1 / 512] to [log(2) / 512] )
*/
if (M_exp_compute_nn(&nn, tmp7, x) != 0)
{
M_apm_log_error_msg(M_APM_RETURN,
"\'m_apm_exp\', Input too large, Overflow");
M_set_to_zero(r);
M_restore_stack(3);
return;
}
dplaces = places + 8;
/* check to make sure our log(2) is accurate enough */
M_check_log_places(dplaces);
m_apm_multiply(tmp8, tmp7, MM_lc_log2);
m_apm_subtract(tmp7, x, tmp8);
/*
* guarantee that |tmp7| is between 0.1 and 0.9999999....
* (in practice, the upper limit only reaches log(2), 0.693... )
*/
while (TRUE)
{
if (tmp7->m_apm_sign != 0)
{
if (tmp7->m_apm_exponent == 0)
break;
}
if (tmp7->m_apm_sign >= 0)
{
nn++;
m_apm_subtract(tmp8, tmp7, MM_lc_log2);
m_apm_copy(tmp7, tmp8);
}
else
{
nn--;
m_apm_add(tmp8, tmp7, MM_lc_log2);
m_apm_copy(tmp7, tmp8);
}
}
m_apm_multiply(tmp9, tmp7, MM_exp_512R);
/* perform the series expansion ... */
M_raw_exp(tmp8, dplaces, tmp9);
/*
* raise result to the 512 power
*
* note : x ^ 512 = (((x ^ 2) ^ 2) ^ 2) ... 9 times
*/
ii = 9;
while (TRUE)
{
m_apm_multiply(tmp9, tmp8, tmp8);
m_apm_round(tmp8, dplaces, tmp9);
if (--ii == 0)
break;
}
/* now compute 2 ^ N */
m_apm_integer_pow(tmp7, dplaces, MM_Two, nn);
m_apm_multiply(tmp9, tmp7, tmp8);
m_apm_round(r, places, tmp9);
M_restore_stack(3); /* restore the 3 locals we used here */
}
void m_apm_multiply(M_APM r, M_APM a, M_APM b)
{
int ai, itmp, sign, nexp, ii, jj, indexa, indexb, index0, numdigits;
UCHAR *cp, *cpr, *cp_div, *cp_rem;
void *vp;
sign = a->m_apm_sign * b->m_apm_sign;
nexp = a->m_apm_exponent + b->m_apm_exponent;
if (sign == 0) /* one number is zero, result is zero */
{
M_set_to_zero(r);
return;
}
numdigits = a->m_apm_datalength + b->m_apm_datalength;
indexa = (a->m_apm_datalength + 1) >> 1;
indexb = (b->m_apm_datalength + 1) >> 1;
/*
* If we are multiplying 2 'big' numbers, use the fast algorithm.
*
* This is a **very** approx break even point between this algorithm
* and the FFT multiply. Note that different CPU's, operating systems,
* and compiler's may yield a different break even point. This point
* (~96 decimal digits) is how the test came out on the author's system.
*/
if (indexa >= 48 && indexb >= 48)
{
M_fast_multiply(r, a, b);
return;
}
ii = (numdigits + 1) >> 1; /* required size of result, in bytes */
if (ii > r->m_apm_malloclength)
{
if ((vp = MAPM_REALLOC(r->m_apm_data, (ii + 32))) == NULL)
{
/* fatal, this does not return */
M_apm_log_error_msg(M_APM_FATAL, "\'m_apm_multiply\', Out of memory");
}
r->m_apm_malloclength = ii + 28;
r->m_apm_data = (UCHAR *)vp;
}
M_get_div_rem_addr(&cp_div, &cp_rem);
index0 = indexa + indexb;
cp = r->m_apm_data;
memset(cp, 0, index0);
ii = indexa;
while (TRUE)
{
index0--;
cpr = cp + index0;
jj = indexb;
ai = (int)a->m_apm_data[--ii];
while (TRUE)
{
itmp = ai * b->m_apm_data[--jj];
*(cpr-1) += cp_div[itmp];
*cpr += cp_rem[itmp];
if (*cpr >= 100)
{
*cpr -= 100;
*(cpr-1) += 1;
}
cpr--;
if (*cpr >= 100)
{
*cpr -= 100;
*(cpr-1) += 1;
}
if (jj == 0)
break;
}
if (ii == 0)
break;
}
r->m_apm_sign = sign;
r->m_apm_exponent = nexp;
r->m_apm_datalength = numdigits;
M_apm_normalize(r);
}
void m_apm_log(M_APM r, int places, M_APM a)
{
M_APM tmp0, tmp1, tmp2;
int mexp, dplaces;
if (a->m_apm_sign <= 0)
{
M_apm_log_error_msg(M_APM_RETURN,
"Warning! ... \'m_apm_log\', Negative argument");
M_set_to_zero(r);
return;
}
tmp0 = M_get_stack_var();
tmp1 = M_get_stack_var();
tmp2 = M_get_stack_var();
dplaces = places + 8;
/*
* if the input is real close to 1, use the series expansion
* to compute the log.
*
* 0.9999 < a < 1.0001
*/
m_apm_subtract(tmp0, a, MM_One);
if (tmp0->m_apm_sign == 0) /* is input exactly 1 ?? */
{ /* if so, result is 0 */
M_set_to_zero(r);
M_restore_stack(3);
return;
}
if (tmp0->m_apm_exponent <= -4)
{
M_log_near_1(r, places, tmp0);
M_restore_stack(3);
return;
}
/* make sure our log(10) is accurate enough for this calculation */
/* (and log(2) which is called from M_log_basic_iteration) */
M_check_log_places(dplaces + 25);
mexp = a->m_apm_exponent;
if (mexp >= -4 && mexp <= 4)
{
M_log_basic_iteration(r, places, a);
}
else
{
/*
* use log (x * y) = log(x) + log(y)
*
* here we use y = exponent of our base 10 number.
*
* let 'C' = log(10) = 2.3025850929940....
*
* then log(x * y) = log(x) + ( C * base_10_exponent )
*/
m_apm_copy(tmp2, a);
mexp = tmp2->m_apm_exponent - 2;
tmp2->m_apm_exponent = 2; /* force number between 10 & 100 */
M_log_basic_iteration(tmp0, dplaces, tmp2);
m_apm_set_long(tmp1, (long)mexp);
m_apm_multiply(tmp2, tmp1, MM_lc_log10);
m_apm_add(tmp1, tmp2, tmp0);
m_apm_round(r, places, tmp1);
}
M_restore_stack(3); /* restore the 3 locals we used here */
}
void m_apm_integer_pow(M_APM rr, int places, M_APM aa, int mexp)
{
M_APM tmp0, tmpy, tmpz;
int nexp, ii, signflag, local_precision;
if (mexp == 0)
{
m_apm_copy(rr, MM_One);
return;
}
else
{
if (mexp > 0)
{
signflag = 0;
nexp = mexp;
}
else
{
signflag = 1;
nexp = -mexp;
}
}
if (aa->m_apm_sign == 0)
{
M_set_to_zero(rr);
return;
}
tmp0 = M_get_stack_var();
tmpy = M_get_stack_var();
tmpz = M_get_stack_var();
local_precision = places + 8;
m_apm_copy(tmpy, MM_One);
m_apm_copy(tmpz, aa);
while (TRUE)
{
ii = nexp & 1;
nexp = nexp >> 1;
if (ii != 0) /* exponent -was- odd */
{
m_apm_multiply(tmp0, tmpy, tmpz);
m_apm_round(tmpy, local_precision, tmp0);
if (nexp == 0)
break;
}
m_apm_multiply(tmp0, tmpz, tmpz);
m_apm_round(tmpz, local_precision, tmp0);
}
if (signflag)
{
m_apm_reciprocal(rr, places, tmpy);
}
else
{
m_apm_round(rr, places, tmpy);
}
M_restore_stack(3);
}
void m_apm_integer_pow_nr(M_APM rr, M_APM aa, int mexp)
{
M_APM tmp0, tmpy, tmpz;
int nexp, ii;
if (mexp == 0)
{
m_apm_copy(rr, MM_One);
return;
}
else
{
if (mexp < 0)
{
M_apm_log_error_msg(M_APM_RETURN,
"Warning! ... \'m_apm_integer_pow_nr\', Negative exponent");
M_set_to_zero(rr);
return;
}
}
if (mexp == 1)
{
m_apm_copy(rr, aa);
return;
}
if (mexp == 2)
{
m_apm_multiply(rr, aa, aa);
return;
}
nexp = mexp;
if (aa->m_apm_sign == 0)
{
M_set_to_zero(rr);
return;
}
tmp0 = M_get_stack_var();
tmpy = M_get_stack_var();
tmpz = M_get_stack_var();
m_apm_copy(tmpy, MM_One);
m_apm_copy(tmpz, aa);
while (TRUE)
{
ii = nexp & 1;
nexp = nexp >> 1;
if (ii != 0) /* exponent -was- odd */
{
m_apm_multiply(tmp0, tmpy, tmpz);
if (nexp == 0)
break;
m_apm_copy(tmpy, tmp0);
}
m_apm_multiply(tmp0, tmpz, tmpz);
m_apm_copy(tmpz, tmp0);
}
m_apm_copy(rr, tmp0);
M_restore_stack(3);
}
/*
Calculate the POW function by calling EXP :
Y A
X = e where A = Y * log(X)
*/
void m_apm_pow(M_APM rr, int places, M_APM xx, M_APM yy)
{
int iflag, pflag;
char sbuf[64];
M_APM tmp8, tmp9;
/* if yy == 0, return 1 */
if (yy->m_apm_sign == 0)
{
m_apm_copy(rr, MM_One);
return;
}
/* if xx == 0, return 0 */
if (xx->m_apm_sign == 0)
{
M_set_to_zero(rr);
return;
}
if (M_size_flag == 0) /* init locals on first call */
{
M_size_flag = M_get_sizeof_int();
M_last_log_digits = 0;
M_last_xx_input = m_apm_init();
M_last_xx_log = m_apm_init();
}
/*
* if 'yy' is a small enough integer, call the more
* efficient _integer_pow function.
*/
if (m_apm_is_integer(yy))
{
iflag = FALSE;
if (M_size_flag == 2) /* 16 bit compilers */
{
if (yy->m_apm_exponent <= 4)
iflag = TRUE;
}
else /* >= 32 bit compilers */
{
if (yy->m_apm_exponent <= 7)
iflag = TRUE;
}
if (iflag)
{
m_apm_to_integer_string(sbuf, yy);
m_apm_integer_pow(rr, places, xx, atoi(sbuf));
return;
}
}
tmp8 = M_get_stack_var();
tmp9 = M_get_stack_var();
/*
* If parameter 'X' is the same this call as it
* was the previous call, re-use the saved log
* calculation from last time.
*/
pflag = FALSE;
if (M_last_log_digits >= places)
{
if (m_apm_compare(xx, M_last_xx_input) == 0)
pflag = TRUE;
}
if (pflag)
{
m_apm_round(tmp9, (places + 8), M_last_xx_log);
}
else
{
m_apm_log(tmp9, (places + 8), xx);
M_last_log_digits = places + 2;
/* save the 'X' input value and the log calculation */
m_apm_copy(M_last_xx_input, xx);
m_apm_copy(M_last_xx_log, tmp9);
}
m_apm_multiply(tmp8, tmp9, yy);
m_apm_exp(rr, places, tmp8);
M_restore_stack(2); /* restore the 2 locals we used here */
}
