+2006-04-25 Bruno Haible <bruno@clisp.org>
+ Richard B. Kreckel <kreckel@ginac.de>
+
+ Make it theoretically possible to use bignums and long-floats with
+ more than 2^32 significant digits or bits.
+ * doc/cln.tex (logcount): Change return type to uintC.
+ (struct cl_byte): Change elements to uintC.
+ (integer_length, ord2, power2p): Change return type to uintC.
+ (scale_float): Change argument type to sintC.
+ (float_digits, float_precision): Change return type to uintC.
+ * examples/atan_recip.cc: Use uintC instead of uintL where appropriate.
+ * examples/atanh_recip.cc: Likewise.
+ * include/cln/GV.h: Likewise.
+ * include/cln/GV_complex.h: Likewise.
+ * include/cln/GV_integer.h: Likewise.
+ * include/cln/GV_modinteger.h: Likewise.
+ * include/cln/GV_number.h: Likewise.
+ * include/cln/GV_rational.h: Likewise.
+ * include/cln/GV_real.h: Likewise.
+ * include/cln/SV.h: Likewise.
+ * include/cln/SV_complex.h: Likewise.
+ * include/cln/SV_integer.h: Likewise.
+ * include/cln/SV_number.h: Likewise.
+ * include/cln/SV_rational.h: Likewise.
+ * include/cln/SV_real.h: Likewise.
+ * include/cln/SV_ringelt.h: Likewise.
+ * include/cln/dfloat.h: Likewise.
+ * include/cln/ffloat.h: Likewise.
+ * include/cln/float.h: Likewise.
+ * include/cln/integer.h: Likewise.
+ * include/cln/lfloat.h: Likewise.
+ * include/cln/modinteger.h: Likewise.
+ * include/cln/sfloat.h: Likewise.
+ * src/base/cl_low.h (integerlengthC): New macro.
+ * src/base/digitseq/cl_2DS_div.cc: Use uintC instead of uintL where
+ appropriate.
+ * src/base/digitseq/cl_2DS_recip.cc: Likewise.
+ * src/base/digitseq/cl_DS.h: Likewise.
+ * src/base/digitseq/cl_DS_mul.c: Likewise.
+ * src/base/digitseq/cl_DS_mul_fftc.h: Likewise.
+ * src/base/digitseq/cl_DS_mul_fftcs.h: Likewise.
+ * src/base/digitseq/cl_DS_mul_fftm.h: Likewise.
+ * src/base/digitseq/cl_DS_mul_fftp.h: Likewise.
+ * src/base/digitseq/cl_DS_mul_fftp3.h: Likewise.
+ * src/base/digitseq/cl_DS_mul_fftp3m.h: Likewise.
+ * src/base/digitseq/cl_DS_mul_fftr.h: Likewise.
+ * src/base/digitseq/cl_DS_mul_kara.h: Likewise.
+ * src/base/digitseq/cl_DS_mul_nuss.h: Likewise.
+ * src/base/digitseq/cl_DS_recip.cc: Likewise.
+ * src/base/digitseq/cl_DS_recipsqrt.cc: Likewise.
+ * src/base/digitseq/cl_DS_sqrt.cc: Likewise.
+ * src/base/digitseq/cl_DS_trandom.cc: Likewise.
+ * src/complex/input/cl_N_read.cc: Likewise.
+ * src/complex/transcendental/cl_C_asinh_aux.cc: Likewise.
+ * src/complex/transcendental/cl_C_expt_C.cc: Likewise.
+ * src/float/cl_F.h: Likewise.
+ * src/float/conv/cl_F_from_F_f.cc: Likewise.
+ * src/float/conv/cl_F_from_I_f.cc: Likewise.
+ * src/float/conv/cl_F_from_RA_f.cc: Likewise.
+ * src/float/dfloat/conv/cl_I_to_double.cc: Likewise.
+ * src/float/dfloat/conv/cl_RA_to_double.cc: Likewise.
+ * src/float/dfloat/elem/cl_DF_from_I.cc: Likewise.
+ * src/float/dfloat/elem/cl_DF_from_RA.cc: Likewise.
+ * src/float/dfloat/elem/cl_DF_scale.cc: Likewise.
+ * src/float/dfloat/misc/cl_DF_digits.cc: Likewise.
+ * src/float/dfloat/misc/cl_DF_precision.cc: Likewise.
+ * src/float/elem/cl_F_scale.cc: Likewise.
+ * src/float/ffloat/conv/cl_I_to_float.cc: Likewise.
+ * src/float/ffloat/conv/cl_RA_to_float.cc: Likewise.
+ * src/float/ffloat/elem/cl_FF_from_I.cc: Likewise.
+ * src/float/ffloat/elem/cl_FF_from_RA.cc: Likewise.
+ * src/float/ffloat/elem/cl_FF_scale.cc: Likewise.
+ * src/float/ffloat/misc/cl_FF_digits.cc: Likewise.
+ * src/float/ffloat/misc/cl_FF_precision.cc: Likewise.
+ * src/float/input/cl_F_read.cc: Likewise.
+ * src/float/lfloat/algebraic/cl_LF_sqrt.cc: Likewise.
+ * src/float/lfloat/elem/cl_LF_1plus.cc: Likewise.
+ * src/float/lfloat/elem/cl_LF_I_div.cc: Likewise.
+ * src/float/lfloat/elem/cl_LF_I_mul.cc: Likewise.
+ * src/float/lfloat/elem/cl_LF_div.cc: Likewise.
+ * src/float/lfloat/elem/cl_LF_from_I.cc: Likewise.
+ * src/float/lfloat/elem/cl_LF_from_RA.cc: Likewise.
+ * src/float/lfloat/elem/cl_LF_fround.cc: Likewise.
+ * src/float/lfloat/elem/cl_LF_ftrunc.cc: Likewise.
+ * src/float/lfloat/elem/cl_LF_futrunc.cc: Likewise.
+ * src/float/lfloat/elem/cl_LF_scale.cc: Likewise.
+ * src/float/lfloat/elem/cl_LF_to_I.cc: Likewise.
+ * src/float/lfloat/misc/cl_LF_digits.cc: Likewise.
+ * src/float/lfloat/misc/cl_LF_idecode.cc: Likewise.
+ * src/float/lfloat/misc/cl_LF_leninc.cc: Likewise.
+ * src/float/lfloat/misc/cl_LF_lenincx.cc: Likewise.
+ * src/float/lfloat/misc/cl_LF_precision.cc: Likewise.
+ * src/float/lfloat/misc/cl_LF_shortenrel.cc: Likewise.
+ * src/float/lfloat/misc/cl_LF_shortenwith.cc: Likewise.
+ * src/float/misc/cl_F_digits.cc: Likewise.
+ * src/float/misc/cl_F_epsneg.cc: Likewise.
+ * src/float/misc/cl_F_epspos.cc: Likewise.
+ * src/float/misc/cl_F_leastneg.cc: Likewise.
+ * src/float/misc/cl_F_leastpos.cc: Likewise.
+ * src/float/misc/cl_F_mostneg.cc: Likewise.
+ * src/float/misc/cl_F_mostpos.cc: Likewise.
+ * src/float/misc/cl_F_precision.cc: Likewise.
+ * src/float/misc/cl_F_rational.cc: Likewise.
+ * src/float/misc/cl_F_shortenrel.cc: Likewise.
+ * src/float/output/cl_F_dprint.cc: Likewise.
+ * src/float/random/cl_F_random.cc: Likewise.
+ * src/float/sfloat/elem/cl_SF_from_I.cc: Likewise.
+ * src/float/sfloat/elem/cl_SF_from_RA.cc: Likewise.
+ * src/float/sfloat/elem/cl_SF_scale.cc: Likewise.
+ * src/float/sfloat/misc/cl_SF_digits.cc: Likewise.
+ * src/float/sfloat/misc/cl_SF_precision.cc: Likewise.
+ * src/float/transcendental/cl_F_atanhx.cc: Likewise.
+ * src/float/transcendental/cl_F_atanx.cc: Likewise.
+ * src/float/transcendental/cl_F_catalanconst_f.cc: Likewise.
+ * src/float/transcendental/cl_F_cos.cc: Likewise.
+ * src/float/transcendental/cl_F_cosh.cc: Likewise.
+ * src/float/transcendental/cl_F_coshsinh.cc: Likewise.
+ * src/float/transcendental/cl_F_cossin.cc: Likewise.
+ * src/float/transcendental/cl_F_eulerconst_f.cc: Likewise.
+ * src/float/transcendental/cl_F_exp1_f.cc: Likewise.
+ * src/float/transcendental/cl_F_expx.cc: Likewise.
+ * src/float/transcendental/cl_F_ln10_f.cc: Likewise.
+ * src/float/transcendental/cl_F_ln2_f.cc: Likewise.
+ * src/float/transcendental/cl_F_lnx.cc: Likewise.
+ * src/float/transcendental/cl_F_pi_f.cc: Likewise.
+ * src/float/transcendental/cl_F_sin.cc: Likewise.
+ * src/float/transcendental/cl_F_sinh.cc: Likewise.
+ * src/float/transcendental/cl_F_sinhx.cc: Likewise.
+ * src/float/transcendental/cl_F_sinx.cc: Likewise.
+ * src/float/transcendental/cl_F_tran.h: Likewise.
+ * src/float/transcendental/cl_F_zeta_int_f.cc: Likewise.
+ * src/float/transcendental/cl_LF_atan_recip.cc: Likewise.
+ * src/float/transcendental/cl_LF_atanh_recip.cc: Likewise.
+ * src/float/transcendental/cl_LF_catalanconst.cc: Likewise.
+ * src/float/transcendental/cl_LF_coshsinh_aux.cc: Likewise.
+ * src/float/transcendental/cl_LF_cossin_aux.cc: Likewise.
+ * src/float/transcendental/cl_LF_eulerconst.cc: Likewise.
+ * src/float/transcendental/cl_LF_exp1.cc: Likewise.
+ * src/float/transcendental/cl_LF_exp_aux.cc: Likewise.
+ * src/float/transcendental/cl_LF_pi.cc: Likewise.
+ * src/float/transcendental/cl_LF_ratseries.cc: Likewise.
+ * src/float/transcendental/cl_LF_ratseries_.cc: Likewise.
+ * src/float/transcendental/cl_LF_ratseries_a.cc: Likewise.
+ * src/float/transcendental/cl_LF_ratseries_ab.cc: Likewise.
+ * src/float/transcendental/cl_LF_ratseries_b.cc: Likewise.
+ * src/float/transcendental/cl_LF_ratseries_p.cc: Likewise.
+ * src/float/transcendental/cl_LF_ratseries_pa.cc: Likewise.
+ * src/float/transcendental/cl_LF_ratseries_pab.cc: Likewise.
+ * src/float/transcendental/cl_LF_ratseries_pb.cc: Likewise.
+ * src/float/transcendental/cl_LF_ratseries_pq.cc: Likewise.
+ * src/float/transcendental/cl_LF_ratseries_pqa.cc: Likewise.
+ * src/float/transcendental/cl_LF_ratseries_pqab.cc: Likewise.
+ * src/float/transcendental/cl_LF_ratseries_pqb.cc: Likewise.
+ * src/float/transcendental/cl_LF_ratseries_q.cc: Likewise.
+ * src/float/transcendental/cl_LF_ratseries_qa.cc: Likewise.
+ * src/float/transcendental/cl_LF_ratseries_qab.cc: Likewise.
+ * src/float/transcendental/cl_LF_ratseries_qb.cc: Likewise.
+ * src/float/transcendental/cl_LF_ratseries_stream_pq.cc: Likewise.
+ * src/float/transcendental/cl_LF_ratseries_stream_pqa.cc: Likewise.
+ * src/float/transcendental/cl_LF_ratseries_stream_pqab.cc: Likewise.
+ * src/float/transcendental/cl_LF_ratseries_stream_pqb.cc: Likewise.
+ * src/float/transcendental/cl_LF_ratsumseries_pqcd.cc: Likewise.
+ * src/float/transcendental/cl_LF_ratsumseries_pqcd_aux.cc: Likewise.
+ * src/float/transcendental/cl_LF_ratsumseries_pqd.cc: Likewise.
+ * src/float/transcendental/cl_LF_ratsumseries_pqd_aux.cc: Likewise.
+ * src/float/transcendental/cl_LF_tran.h: Likewise.
+ * src/float/transcendental/cl_LF_zeta3.cc: Likewise.
+ * src/float/transcendental/cl_LF_zeta_int.cc: Likewise.
+ * src/integer/algebraic/cl_I_rootp_I.cc: Likewise.
+ * src/integer/algebraic/cl_I_rootp_aux.cc: Likewise.
+ * src/integer/bitwise/cl_I_ash.cc: Likewise.
+ * src/integer/bitwise/cl_I_ash_I.cc: Likewise.
+ * src/integer/bitwise/cl_I_byte.h: Likewise.
+ * src/integer/bitwise/cl_I_fullbyte.cc: Likewise.
+ * src/integer/bitwise/cl_I_ilength.cc: Likewise.
+ * src/integer/bitwise/cl_I_ldb.cc: Likewise.
+ * src/integer/bitwise/cl_I_ldbtest.cc: Likewise.
+ * src/integer/bitwise/cl_I_ldbx.cc: Likewise.
+ * src/integer/bitwise/cl_I_ldbxtest.cc: Likewise.
+ * src/integer/bitwise/cl_I_logbitp.cc: Likewise.
+ * src/integer/bitwise/cl_I_logbitp_I.cc: Likewise.
+ * src/integer/bitwise/cl_I_logcount.cc: Likewise.
+ * src/integer/bitwise/cl_I_mkf.cc: Likewise.
+ * src/integer/bitwise/cl_I_mkfx.cc: Likewise.
+ * src/integer/cl_I.h: Likewise.
+ * src/integer/conv/cl_I_to_digits.cc: Likewise.
+ * src/integer/conv/cl_I_digits_need.cc: Likewise.
+ * src/integer/conv/cl_I_from_digits.cc: Likewise.
+ * src/integer/gcd/cl_I_gcd.cc: Likewise.
+ * src/integer/gcd/cl_I_xgcd.cc: Likewise.
+ * src/integer/misc/cl_I_eqhashcode.cc: Likewise.
+ * src/integer/misc/cl_I_ord2.cc: Likewise.
+ * src/integer/misc/cl_I_power2p.cc: Likewise.
+ * src/integer/output/cl_I_cached_power.h (cached_power_table): allow
+ for 40 elements.
+ * src/integer/output/cl_I_decstring.cc: Use uintC instead of uintL
+ where appropriate.
+ * src/integer/output/cl_I_print.cc: Likewise.
+ * src/integer/output/cl_I_print_string.cc: Likewise.
+ * src/modinteger/cl_MI.cc: Likewise.
+ * src/modinteger/cl_MI_lshift.cc: Likewise.
+ * src/modinteger/cl_MI_montgom.h: Likewise.
+ * src/modinteger/cl_MI_pow2.h: Likewise.
+ * src/modinteger/cl_MI_pow2m1.h: Likewise.
+ * src/modinteger/cl_MI_pow2p1.h: Likewise.
+ * src/modinteger/cl_MI_rshift.cc: Likewise.
+ * src/modinteger/cl_MI_std.h: Likewise.
+ * src/numtheory/cl_IF_millerrabin.cc: Likewise.
+ * src/numtheory/cl_nt_isprobprime.cc: Likewise.
+ * src/numtheory/cl_nt_sqrtmodp.cc: Likewise.
+ * src/polynomial/elem/cl_UP_GF2.h: Likewise.
+ * src/real/conv/cl_F_from_R_f.cc: Likewise.
+ * src/real/format-output/cl_fmt_floatstring.cc: Likewise.
+ * src/real/input/cl_R_read.cc: Likewise.
+ * src/vector/cl_GV_I.cc: Likewise.
+ * src/vector/cl_GV_I_copy.cc: Likewise.
+ * src/vector/cl_GV_number.cc: Likewise.
+ * src/vector/cl_GV_number_copy.cc: Likewise.
+ * src/vector/cl_SV_copy.cc: Likewise.
+ * src/vector/cl_SV_number.cc: Likewise.
+ * src/vector/cl_SV_ringelt.cc: Likewise.
+ * tests/main.cc: Likewise.
+ * tests/test_I_ilength.cc: Likewise.
+ * tests/test_I_ord2.cc: Likewise.
+
2006-04-19 Bruno Haible <bruno@clisp.org>
Prepare for autoconf-2.60.
Returns true if the @code{n}th bit (from the right) of @code{x} is set.
Bit 0 is the least significant bit.
-@item uintL logcount (const cl_I& x)
+@item uintC logcount (const cl_I& x)
@cindex @code{logcount ()}
Returns the number of one bits in @code{x}, if @code{x} >= 0, or
the number of zero bits in @code{x}, if @code{x} < 0.
The following functions operate on intervals of bits in integers.
The type
@example
-struct cl_byte @{ uintL size; uintL position; @};
+struct cl_byte @{ uintC size; uintC position; @};
@end example
@cindex @code{cl_byte}
represents the bit interval containing the bits
by @code{-y} bits to the right (if @code{y}<=0). In other words, this
returns @code{floor(x * expt(2,y))}.
-@item uintL integer_length (const cl_I& x)
+@item uintC integer_length (const cl_I& x)
@cindex @code{integer_length ()}
Returns the number of bits (excluding the sign bit) needed to represent @code{x}
in two's complement notation. This is the smallest n >= 0 such that
-2^n <= x < 2^n. If x > 0, this is the unique n > 0 such that
2^(n-1) <= x < 2^n.
-@item uintL ord2 (const cl_I& x)
+@item uintC ord2 (const cl_I& x)
@cindex @code{ord2 ()}
@code{x} must be non-zero. This function returns the number of 0 bits at the
right of @code{x} in two's complement notation. This is the largest n >= 0
such that 2^n divides @code{x}.
-@item uintL power2p (const cl_I& x)
+@item uintC power2p (const cl_I& x)
@cindex @code{power2p ()}
@code{x} must be > 0. This function checks whether @code{x} is a power of 2.
If @code{x} = 2^(n-1), it returns n. Else it returns 0.
defines the following operations.
@table @code
-@item @var{type} scale_float (const @var{type}& x, sintL delta)
+@item @var{type} scale_float (const @var{type}& x, sintC delta)
@cindex @code{scale_float ()}
@itemx @var{type} scale_float (const @var{type}& x, const cl_I& delta)
Returns @code{x*2^delta}. This is more efficient than an explicit multiplication
Returns the sign @code{s} of @code{x} as a float. The value is 1 for
@code{x} >= 0, -1 for @code{x} < 0.
-@item uintL float_digits (const @var{type}& x)
+@item uintC float_digits (const @var{type}& x)
@cindex @code{float_digits ()}
Returns the number of mantissa bits in the floating-point representation
of @code{x}, including the hidden bit. The value only depends on the type
of @code{x}, not on its value.
-@item uintL float_precision (const @var{type}& x)
+@item uintC float_precision (const @var{type}& x)
@cindex @code{float_precision ()}
Returns the number of significant mantissa bits in the floating-point
representation of @code{x}. Since denormalized numbers are not supported,
const cl_LF atan_recip_1a (cl_I m, uintC len)
{
var uintC actuallen = len + 1;
- var cl_LF eps = scale_float(cl_I_to_LF(1,actuallen),-intDsize*(sintL)actuallen);
+ var cl_LF eps = scale_float(cl_I_to_LF(1,actuallen),-intDsize*(sintC)actuallen);
var cl_I m2 = m*m;
var cl_LF fterm = cl_I_to_LF(1,actuallen)/m;
var cl_LF fsum = fterm;
- for (var uintL n = 1; fterm >= eps; n++) {
+ for (var uintC n = 1; fterm >= eps; n++) {
fterm = fterm/m2;
fterm = cl_LF_shortenwith(fterm,eps);
if ((n % 2) == 0)
var cl_I m2 = m*m;
var cl_I fterm = floor1((cl_I)1 << (intDsize*actuallen), m);
var cl_I fsum = fterm;
- for (var uintL n = 1; fterm > 0; n++) {
+ for (var uintC n = 1; fterm > 0; n++) {
fterm = floor1(fterm,m2);
if ((n % 2) == 0)
fsum = fsum + floor1(fterm,2*n+1);
else
fsum = fsum - floor1(fterm,2*n+1);
}
- return scale_float(cl_I_to_LF(fsum,len),-intDsize*(sintL)actuallen);
+ return scale_float(cl_I_to_LF(fsum,len),-intDsize*(sintC)actuallen);
}
const cl_LF atan_recip_1c (cl_I m, uintC len)
{
var uintC actuallen = len + 1;
var cl_I m2 = m*m;
- var sintL N = (sintL)(0.69314718*intDsize/2*actuallen/log(double_approx(m))) + 1;
+ var sintC N = (sintC)(0.69314718*intDsize/2*actuallen/log(double_approx(m))) + 1;
var cl_I num = 0, den = 1; // "lazy rational number"
- for (sintL n = N-1; n>=0; n--) {
+ for (sintC n = N-1; n>=0; n--) {
// Multiply sum with 1/m^2:
den = den * m2;
// Add (-1)^n/(2n+1):
{
var uintC actuallen = len + 1;
var cl_I m2 = m*m;
- var uintL N = (uintL)(0.69314718*intDsize/2*actuallen/log(double_approx(m))) + 1;
+ var uintC N = (uintC)(0.69314718*intDsize/2*actuallen/log(double_approx(m))) + 1;
CL_ALLOCA_STACK;
var cl_I* bv = (cl_I*) cl_alloca(N*sizeof(cl_I));
var cl_I* qv = (cl_I*) cl_alloca(N*sizeof(cl_I));
- var uintL n;
+ var uintC n;
for (n = 0; n < N; n++) {
new (&bv[n]) cl_I ((n % 2) == 0 ? (cl_I)(2*n+1) : -(cl_I)(2*n+1));
new (&qv[n]) cl_I (n==0 ? m : m2);
const cl_LF atan_recip_2a (cl_I m, uintC len)
{
var uintC actuallen = len + 1;
- var cl_LF eps = scale_float(cl_I_to_LF(1,actuallen),-intDsize*(sintL)actuallen);
+ var cl_LF eps = scale_float(cl_I_to_LF(1,actuallen),-intDsize*(sintC)actuallen);
var cl_I m2 = m*m+1;
var cl_LF fterm = cl_I_to_LF(m,actuallen)/m2;
var cl_LF fsum = fterm;
- for (var uintL n = 1; fterm >= eps; n++) {
+ for (var uintC n = 1; fterm >= eps; n++) {
fterm = The(cl_LF)((2*n)*fterm)/((2*n+1)*m2);
fterm = cl_LF_shortenwith(fterm,eps);
fsum = fsum + LF_to_LF(fterm,actuallen);
var cl_I m2 = m*m+1;
var cl_I fterm = floor1((cl_I)m << (intDsize*actuallen), m2);
var cl_I fsum = fterm;
- for (var uintL n = 1; fterm > 0; n++) {
+ for (var uintC n = 1; fterm > 0; n++) {
fterm = floor1((2*n)*fterm,(2*n+1)*m2);
fsum = fsum + fterm;
}
- return scale_float(cl_I_to_LF(fsum,len),-intDsize*(sintL)actuallen);
+ return scale_float(cl_I_to_LF(fsum,len),-intDsize*(sintC)actuallen);
}
const cl_LF atan_recip_2c (cl_I m, uintC len)
{
var uintC actuallen = len + 1;
var cl_I m2 = m*m+1;
- var uintL N = (uintL)(0.69314718*intDsize*actuallen/log(double_approx(m2))) + 1;
+ var uintC N = (uintC)(0.69314718*intDsize*actuallen/log(double_approx(m2))) + 1;
var cl_I num = 0, den = 1; // "lazy rational number"
- for (uintL n = N; n>0; n--) {
+ for (uintC n = N; n>0; n--) {
// Multiply sum with (2n)/(2n+1)(m^2+1):
num = num * (2*n);
den = den * ((2*n+1)*m2);
{
var uintC actuallen = len + 1;
var cl_I m2 = m*m+1;
- var uintL N = (uintL)(0.69314718*intDsize*actuallen/log(double_approx(m2))) + 1;
+ var uintC N = (uintC)(0.69314718*intDsize*actuallen/log(double_approx(m2))) + 1;
CL_ALLOCA_STACK;
var cl_I* pv = (cl_I*) cl_alloca(N*sizeof(cl_I));
var cl_I* qv = (cl_I*) cl_alloca(N*sizeof(cl_I));
- var uintL n;
+ var uintC n;
new (&pv[0]) cl_I (m);
new (&qv[0]) cl_I (m2);
for (n = 1; n < N; n++) {
if (argc < 2)
exit(1);
cl_I m = (cl_I)argv[1];
- uintL len = atoi(argv[2]);
+ uintC len = atol(argv[2]);
cl_LF p;
ln(cl_I_to_LF(1000,len+10)); // fill cache
// Method 1.
const cl_LF atanh_recip_1a (cl_I m, uintC len)
{
var uintC actuallen = len + 1;
- var cl_LF eps = scale_float(cl_I_to_LF(1,actuallen),-intDsize*(sintL)actuallen);
+ var cl_LF eps = scale_float(cl_I_to_LF(1,actuallen),-intDsize*(sintC)actuallen);
var cl_I m2 = m*m;
var cl_LF fterm = cl_I_to_LF(1,actuallen)/m;
var cl_LF fsum = fterm;
- for (var uintL n = 1; fterm >= eps; n++) {
+ for (var uintC n = 1; fterm >= eps; n++) {
fterm = fterm/m2;
fterm = cl_LF_shortenwith(fterm,eps);
fsum = fsum + LF_to_LF(fterm/(2*n+1),actuallen);
var cl_I m2 = m*m;
var cl_I fterm = floor1((cl_I)1 << (intDsize*actuallen), m);
var cl_I fsum = fterm;
- for (var uintL n = 1; fterm > 0; n++) {
+ for (var uintC n = 1; fterm > 0; n++) {
fterm = floor1(fterm,m2);
fsum = fsum + floor1(fterm,2*n+1);
}
- return scale_float(cl_I_to_LF(fsum,len),-intDsize*(sintL)actuallen);
+ return scale_float(cl_I_to_LF(fsum,len),-intDsize*(sintC)actuallen);
}
const cl_LF atanh_recip_1c (cl_I m, uintC len)
{
var uintC actuallen = len + 1;
var cl_I m2 = m*m;
- var sintL N = (sintL)(0.69314718*intDsize/2*actuallen/log(double_approx(m))) + 1;
+ var sintC N = (sintC)(0.69314718*intDsize/2*actuallen/log(double_approx(m))) + 1;
var cl_I num = 0, den = 1; // "lazy rational number"
- for (sintL n = N-1; n>=0; n--) {
+ for (sintC n = N-1; n>=0; n--) {
// Multiply sum with 1/m^2:
den = den * m2;
// Add 1/(2n+1):
{
var uintC actuallen = len + 1;
var cl_I m2 = m*m;
- var uintL N = (uintL)(0.69314718*intDsize/2*actuallen/log(double_approx(m))) + 1;
+ var uintC N = (uintC)(0.69314718*intDsize/2*actuallen/log(double_approx(m))) + 1;
CL_ALLOCA_STACK;
var cl_I* bv = (cl_I*) cl_alloca(N*sizeof(cl_I));
var cl_I* qv = (cl_I*) cl_alloca(N*sizeof(cl_I));
- var uintL n;
+ var uintC n;
for (n = 0; n < N; n++) {
new (&bv[n]) cl_I ((cl_I)(2*n+1));
new (&qv[n]) cl_I (n==0 ? m : m2);
const cl_LF atanh_recip_2a (cl_I m, uintC len)
{
var uintC actuallen = len + 1;
- var cl_LF eps = scale_float(cl_I_to_LF(1,actuallen),-intDsize*(sintL)actuallen);
+ var cl_LF eps = scale_float(cl_I_to_LF(1,actuallen),-intDsize*(sintC)actuallen);
var cl_I m2 = m*m-1;
var cl_LF fterm = cl_I_to_LF(m,actuallen)/m2;
var cl_LF fsum = fterm;
- for (var uintL n = 1; fterm >= eps; n++) {
+ for (var uintC n = 1; fterm >= eps; n++) {
fterm = The(cl_LF)((2*n)*fterm)/((2*n+1)*m2);
fterm = cl_LF_shortenwith(fterm,eps);
if ((n % 2) == 0)
var cl_I m2 = m*m-1;
var cl_I fterm = floor1((cl_I)m << (intDsize*actuallen), m2);
var cl_I fsum = fterm;
- for (var uintL n = 1; fterm > 0; n++) {
+ for (var uintC n = 1; fterm > 0; n++) {
fterm = floor1((2*n)*fterm,(2*n+1)*m2);
if ((n % 2) == 0)
fsum = fsum + fterm;
else
fsum = fsum - fterm;
}
- return scale_float(cl_I_to_LF(fsum,len),-intDsize*(sintL)actuallen);
+ return scale_float(cl_I_to_LF(fsum,len),-intDsize*(sintC)actuallen);
}
const cl_LF atanh_recip_2c (cl_I m, uintC len)
{
var uintC actuallen = len + 1;
var cl_I m2 = m*m-1;
- var uintL N = (uintL)(0.69314718*intDsize*actuallen/log(double_approx(m2))) + 1;
+ var uintC N = (uintC)(0.69314718*intDsize*actuallen/log(double_approx(m2))) + 1;
var cl_I num = 0, den = 1; // "lazy rational number"
- for (uintL n = N; n>0; n--) {
+ for (uintC n = N; n>0; n--) {
// Multiply sum with -(2n)/(2n+1)(m^2+1):
num = num * (2*n);
den = - den * ((2*n+1)*m2);
{
var uintC actuallen = len + 1;
var cl_I m2 = m*m-1;
- var uintL N = (uintL)(0.69314718*intDsize*actuallen/log(double_approx(m2))) + 1;
+ var uintC N = (uintC)(0.69314718*intDsize*actuallen/log(double_approx(m2))) + 1;
CL_ALLOCA_STACK;
var cl_I* pv = (cl_I*) cl_alloca(N*sizeof(cl_I));
var cl_I* qv = (cl_I*) cl_alloca(N*sizeof(cl_I));
- var uintL n;
+ var uintC n;
new (&pv[0]) cl_I (m);
new (&qv[0]) cl_I (m2);
for (n = 1; n < N; n++) {
if (argc < 2)
exit(1);
cl_I m = (cl_I)argv[1];
- uintL len = atoi(argv[2]);
+ uintC len = atol(argv[2]);
cl_LF p;
ln(cl_I_to_LF(1000,len+10)); // fill cache
// Method 1.
template <class T>
class cl_GV_inner {
protected:
- uintL len; // number of elements
+ uintC len; // number of elements
public:
- uintL length () const; // number of elements
+ uintC length () const; // number of elements
cl_GV_vectorops<T>* vectorops; // get/set element
const cl_GV_index<T> operator[] (unsigned long index);
const cl_GV_constindex<T> operator[] (unsigned long index) const;
const cl_GV_constindex<T> operator[] (int index) const;
public: /* ugh */
// Constructor.
- cl_GV_inner (uintL l, cl_GV_vectorops<T>* ops) : len (l), vectorops (ops) {}
+ cl_GV_inner (uintC l, cl_GV_vectorops<T>* ops) : len (l), vectorops (ops) {}
public:
// Destructor.
~cl_GV_inner ();
// through [].
public:
cl_GV_inner<T>* vec;
- uintL index;
+ uintC index;
operator T () const;
// Constructor:
- cl_GV_index (cl_GV_inner<T>* v, uintL i) : vec (v), index (i) {}
+ cl_GV_index (cl_GV_inner<T>* v, uintC i) : vec (v), index (i) {}
// Assignment operator.
void operator= (const T& x) const;
#if (defined(__sparc__) || defined(__sparc64__) || defined(__mips__) || defined(__mips64__)) && !defined(__GNUC__) // maybe an SGI CC and Sun CC bug?
// through []. It lacks the assignment operator.
public:
const cl_GV_inner<T>* vec;
- uintL index;
+ uintC index;
operator T () const;
// Constructor:
- cl_GV_constindex (const cl_GV_inner<T>* v, uintL i) : vec (v), index (i) {}
+ cl_GV_constindex (const cl_GV_inner<T>* v, uintC i) : vec (v), index (i) {}
private:
// No default constructor, assignment operator.
cl_GV_constindex ();
template <class T>
struct cl_GV_vectorops {
- const T (*element) (const cl_GV_inner<T>* vec, uintL index);
- void (*set_element) (cl_GV_inner<T>* vec, uintL index, const T& x);
+ const T (*element) (const cl_GV_inner<T>* vec, uintC index);
+ void (*set_element) (cl_GV_inner<T>* vec, uintC index, const T& x);
void (*do_delete) (cl_GV_inner<T>* vec);
- void (*copy_elements) (const cl_GV_inner<T>* srcvec, uintL srcindex, cl_GV_inner<T>* destvec, uintL destindex, uintL count);
+ void (*copy_elements) (const cl_GV_inner<T>* srcvec, uintC srcindex, cl_GV_inner<T>* destvec, uintC destindex, uintC count);
};
// All member functions are inline.
template <class T>
-inline uintL cl_GV_inner<T>::length () const
+inline uintC cl_GV_inner<T>::length () const
{
return len;
}
struct cl_GV : public BASE {
public:
// Length.
- uintL length () const
+ uintC length () const
{
return ((const cl_heap_GV<T> *) this->pointer)->v.length();
}
cl_GV& operator= (const cl_GV&);
// Copy a piece of a vector into another vector.
// (Both vectors must be of the same type. Overlapping not allowed.)
- static void copy_elements (const cl_GV& src, uintL srcindex, cl_GV& dest, uintL destindex, uintL count)
+ static void copy_elements (const cl_GV& src, uintC srcindex, cl_GV& dest, uintC destindex, uintC count)
{
const cl_heap_GV<T> * hsrc = (const cl_heap_GV<T> *) src.pointer;
cl_heap_GV<T> * hdest = (cl_heap_GV<T> *) dest.pointer;
// Constructors.
cl_GV_N ();
cl_GV_N (const cl_GV_N&);
- explicit cl_GV_N (uintL len);
+ explicit cl_GV_N (uintC len);
// Assignment operators.
cl_GV_N& operator= (const cl_GV_N&);
// Private pointer manipulations.
};
inline cl_GV_N::cl_GV_N (const cl_GV_N& x) : cl_GV<cl_N,cl_GV_number> (as_cl_private_thing(x)) {}
CL_DEFINE_ASSIGNMENT_OPERATOR(cl_GV_N,cl_GV_N)
-inline cl_GV_N::cl_GV_N (uintL len)
+inline cl_GV_N::cl_GV_N (uintC len)
: cl_GV<cl_N,cl_GV_number> ((cl_heap_GV_N*) cl_make_heap_GV_number(len)) {}
inline cl_GV_N::cl_GV_N ()
: cl_GV<cl_N,cl_GV_number> ((cl_heap_GV_N*) (cl_heap_GV_number*) cl_null_GV_number) {}
struct cl_heap_GV<cl_I> : cl_heap {
cl_GV_inner<cl_I> v;
// here room for the elements
- sintL maxbits () const;
+ sintC maxbits () const;
};
typedef cl_heap_GV<cl_I> cl_heap_GV_I;
cl_GV_I ();
cl_GV_I (const cl_GV_I&);
// Create a vector of unconstrained integers.
- explicit cl_GV_I (uintL len);
+ explicit cl_GV_I (uintC len);
// Create a vector of m-bit integers (>=0, <2^m).
- cl_GV_I (uintL len, sintL m);
+ cl_GV_I (uintC len, sintC m);
// Assignment operators.
cl_GV_I& operator= (const cl_GV_I&);
// Number m of bits allowed per element (-1 if unconstrained).
- sintL maxbits () const
+ sintC maxbits () const
{
return ((const cl_heap_GV_I *) pointer)->maxbits();
}
};
inline cl_GV_I::cl_GV_I (const cl_GV_I& x) : cl_GV<cl_I,cl_GV_RA> (as_cl_private_thing(x)) {}
CL_DEFINE_ASSIGNMENT_OPERATOR(cl_GV_I,cl_GV_I)
-extern cl_heap_GV_I* cl_make_heap_GV_I (uintL len);
-inline cl_GV_I::cl_GV_I (uintL len)
+extern cl_heap_GV_I* cl_make_heap_GV_I (uintC len);
+inline cl_GV_I::cl_GV_I (uintC len)
: cl_GV<cl_I,cl_GV_RA> (cl_make_heap_GV_I(len)) {}
-extern cl_heap_GV_I* cl_make_heap_GV_I (uintL len, sintL m);
-inline cl_GV_I::cl_GV_I (uintL len, sintL m)
+extern cl_heap_GV_I* cl_make_heap_GV_I (uintC len, sintC m);
+inline cl_GV_I::cl_GV_I (uintC len, sintC m)
: cl_GV<cl_I,cl_GV_RA> (cl_make_heap_GV_I(len,m)) {}
// Private pointer manipulations. Never throw away a `struct cl_heap_GV_I *'!
cl_GV_MI ();
cl_GV_MI (const cl_GV_MI&);
// Create a vector of modular integers.
- cl_GV_MI (uintL len, cl_heap_modint_ring* R);
+ cl_GV_MI (uintC len, cl_heap_modint_ring* R);
// Assignment operators.
cl_GV_MI& operator= (const cl_GV_MI&);
// Number m of bits allowed per element (-1 if unconstrained).
- sintL maxbits () const
+ sintC maxbits () const
{
return ((const cl_heap_GV_I *) pointer)->maxbits();
}
CL_DEFINE_ASSIGNMENT_OPERATOR(cl_GV_MI,cl_GV_MI)
inline cl_GV_MI::cl_GV_MI ()
: cl_GV<_cl_MI,cl_GV_any> ((cl_heap_GV_MI*) (cl_heap_GV_I*) cl_null_GV_I) {}
-inline cl_GV_MI::cl_GV_MI (uintL len, cl_heap_modint_ring* R)
+inline cl_GV_MI::cl_GV_MI (uintC len, cl_heap_modint_ring* R)
: cl_GV<_cl_MI,cl_GV_any> ((cl_heap_GV_MI*) cl_make_heap_GV_I(len,R->bits)) {}
// Copy a vector.
// Constructors.
cl_GV_number ();
cl_GV_number (const cl_GV_number&);
- explicit cl_GV_number (uintL len);
+ explicit cl_GV_number (uintC len);
// Assignment operators.
cl_GV_number& operator= (const cl_GV_number&);
// Private pointer manipulations.
};
inline cl_GV_number::cl_GV_number (const cl_GV_number& x) : cl_GV<cl_number,cl_GV_any> (as_cl_private_thing(x)) {}
CL_DEFINE_ASSIGNMENT_OPERATOR(cl_GV_number,cl_GV_number)
-extern cl_heap_GV_number* cl_make_heap_GV_number (uintL len);
-inline cl_GV_number::cl_GV_number (uintL len)
+extern cl_heap_GV_number* cl_make_heap_GV_number (uintC len);
+inline cl_GV_number::cl_GV_number (uintC len)
: cl_GV<cl_number,cl_GV_any> (cl_make_heap_GV_number(len)) {}
// Private pointer manipulations. Never throw away a `struct cl_heap_GV_number *'!
// Constructors.
cl_GV_RA ();
cl_GV_RA (const cl_GV_RA&);
- explicit cl_GV_RA (uintL len);
+ explicit cl_GV_RA (uintC len);
// Assignment operators.
cl_GV_RA& operator= (const cl_GV_RA&);
// Private pointer manipulations.
};
inline cl_GV_RA::cl_GV_RA (const cl_GV_RA& x) : cl_GV<cl_RA,cl_GV_R> (as_cl_private_thing(x)) {}
CL_DEFINE_ASSIGNMENT_OPERATOR(cl_GV_RA,cl_GV_RA)
-inline cl_GV_RA::cl_GV_RA (uintL len)
+inline cl_GV_RA::cl_GV_RA (uintC len)
: cl_GV<cl_RA,cl_GV_R> ((cl_heap_GV_RA*) cl_make_heap_GV_number(len)) {}
inline cl_GV_RA::cl_GV_RA ()
: cl_GV<cl_RA,cl_GV_R> ((cl_heap_GV_RA*) (cl_heap_GV_number*) cl_null_GV_number) {}
// Constructors.
cl_GV_R ();
cl_GV_R (const cl_GV_R&);
- explicit cl_GV_R (uintL len);
+ explicit cl_GV_R (uintC len);
// Assignment operators.
cl_GV_R& operator= (const cl_GV_R&);
// Private pointer manipulations.
};
inline cl_GV_R::cl_GV_R (const cl_GV_R& x) : cl_GV<cl_R,cl_GV_N> (as_cl_private_thing(x)) {}
CL_DEFINE_ASSIGNMENT_OPERATOR(cl_GV_R,cl_GV_R)
-inline cl_GV_R::cl_GV_R (uintL len)
+inline cl_GV_R::cl_GV_R (uintC len)
: cl_GV<cl_R,cl_GV_N> ((cl_heap_GV_R*) cl_make_heap_GV_number(len)) {}
inline cl_GV_R::cl_GV_R ()
: cl_GV<cl_R,cl_GV_N> ((cl_heap_GV_R*) (cl_heap_GV_number*) cl_null_GV_number) {}
template <class T>
class cl_SV_inner {
protected:
- uintL len; // number of elements
+ uintC len; // number of elements
private:
// T data[]; // the elements
T * data() { return (T *) (this+1); }
const T * data() const { return (const T *) (this+1); }
public:
- uintL length () const { return len; } // number of elements
+ uintC length () const { return len; } // number of elements
const T & operator[] (unsigned long index) const
{
#ifndef CL_SV_NO_RANGECHECKS
{ return operator[]((unsigned long)index); }
public: /* ugh */
// Constructor.
- cl_SV_inner (uintL l) : len (l) {}
+ cl_SV_inner (uintC l) : len (l) {}
public:
// Destructor.
~cl_SV_inner ();
template <class T>
inline cl_SV_inner<T>::~cl_SV_inner ()
{
- uintL i = len;
+ uintC i = len;
while (i > 0) {
i--;
data()[i].~T();
struct cl_SV : public BASE {
public:
// Length.
- uintL length () const
+ uintC length () const
{
return ((const cl_heap_SV<T> *) this->pointer)->v.length();
}
// Constructors.
cl_SV_N () : cl_SV<cl_N,cl_SV_number> ((cl_heap_SV_N*) (cl_heap_SV_number*) cl_null_SV_number) {};
cl_SV_N (const cl_SV_N&);
- explicit cl_SV_N (uintL len) : cl_SV<cl_N,cl_SV_number> ((cl_heap_SV_N*) cl_make_heap_SV_number(len)) {};
+ explicit cl_SV_N (uintC len) : cl_SV<cl_N,cl_SV_number> ((cl_heap_SV_N*) cl_make_heap_SV_number(len)) {};
// Assignment operators.
cl_SV_N& operator= (const cl_SV_N&);
// Private pointer manipulations.
// Constructors.
cl_SV_I () : cl_SV<cl_I,cl_SV_RA> ((cl_heap_SV_I*) (cl_heap_SV_number*) cl_null_SV_number) {};
cl_SV_I (const cl_SV_I&);
- explicit cl_SV_I (uintL len) : cl_SV<cl_I,cl_SV_RA> ((cl_heap_SV_I*) cl_make_heap_SV_number(len)) {};
+ explicit cl_SV_I (uintC len) : cl_SV<cl_I,cl_SV_RA> ((cl_heap_SV_I*) cl_make_heap_SV_number(len)) {};
// Assignment operators.
cl_SV_I& operator= (const cl_SV_I&);
};
// Constructors.
cl_SV_number ();
cl_SV_number (const cl_SV_number&);
- explicit cl_SV_number (uintL len);
+ explicit cl_SV_number (uintC len);
// Assignment operators.
cl_SV_number& operator= (const cl_SV_number&);
// Private pointer manipulations.
inline cl_SV_number::cl_SV_number (const cl_SV_number& x) : cl_SV<cl_number,cl_SV_any> (as_cl_private_thing(x)) {}
CL_DEFINE_ASSIGNMENT_OPERATOR(cl_SV_number,cl_SV_number)
// Returns a new simple vector with uninitialized contents.
-extern cl_heap_SV_number* cl_make_heap_SV_number_uninit (uintL len);
+extern cl_heap_SV_number* cl_make_heap_SV_number_uninit (uintC len);
// Returns a new simple vector with all elements initialized to 0.
-extern cl_heap_SV_number* cl_make_heap_SV_number (uintL len);
-inline cl_SV_number::cl_SV_number (uintL len)
+extern cl_heap_SV_number* cl_make_heap_SV_number (uintC len);
+inline cl_SV_number::cl_SV_number (uintC len)
: cl_SV<cl_number,cl_SV_any> (cl_make_heap_SV_number(len)) {}
// Private pointer manipulations. Never throw away a `struct cl_heap_SV_number *'!
// Constructors.
cl_SV_RA () : cl_SV<cl_RA,cl_SV_R> ((cl_heap_SV_RA*) (cl_heap_SV_number*) cl_null_SV_number) {};
cl_SV_RA (const cl_SV_RA&);
- explicit cl_SV_RA (uintL len) : cl_SV<cl_RA,cl_SV_R> ((cl_heap_SV_RA*) cl_make_heap_SV_number(len)) {};
+ explicit cl_SV_RA (uintC len) : cl_SV<cl_RA,cl_SV_R> ((cl_heap_SV_RA*) cl_make_heap_SV_number(len)) {};
// Assignment operators.
cl_SV_RA& operator= (const cl_SV_RA&);
// Private pointer manipulations.
// Constructors.
cl_SV_R () : cl_SV<cl_R,cl_SV_N> ((cl_heap_SV_R*) (cl_heap_SV_number*) cl_null_SV_number) {};
cl_SV_R (const cl_SV_R&);
- explicit cl_SV_R (uintL len) : cl_SV<cl_R,cl_SV_N> ((cl_heap_SV_R*) cl_make_heap_SV_number(len)) {};
+ explicit cl_SV_R (uintC len) : cl_SV<cl_R,cl_SV_N> ((cl_heap_SV_R*) cl_make_heap_SV_number(len)) {};
// Assignment operators.
cl_SV_R& operator= (const cl_SV_R&);
// Private pointer manipulations.
// Constructors.
cl_SV_ringelt ();
cl_SV_ringelt (const cl_SV_ringelt&);
- explicit cl_SV_ringelt (uintL len);
+ explicit cl_SV_ringelt (uintC len);
// Assignment operators.
cl_SV_ringelt& operator= (const cl_SV_ringelt&);
// Private pointer manipulations.
inline cl_SV_ringelt::cl_SV_ringelt (const cl_SV_ringelt& x) : cl_SV<_cl_ring_element,cl_SV_any> (as_cl_private_thing(x)) {}
CL_DEFINE_ASSIGNMENT_OPERATOR(cl_SV_ringelt,cl_SV_ringelt)
// Returns a new simple vector with uninitialized contents.
-extern cl_heap_SV_ringelt* cl_make_heap_SV_ringelt_uninit (uintL len);
+extern cl_heap_SV_ringelt* cl_make_heap_SV_ringelt_uninit (uintC len);
// Returns a new simple vector with all elements initialized to some value.
-extern cl_heap_SV_ringelt* cl_make_heap_SV_ringelt (uintL len);
-inline cl_SV_ringelt::cl_SV_ringelt (uintL len)
+extern cl_heap_SV_ringelt* cl_make_heap_SV_ringelt (uintC len);
+inline cl_SV_ringelt::cl_SV_ringelt (uintC len)
: cl_SV<_cl_ring_element,cl_SV_any> (cl_make_heap_SV_ringelt(len)) {}
// Private pointer manipulations.
extern const cl_DF float_sign (const cl_DF& x);
// float_digits(x) liefert (float-digits x), wo x ein Float ist.
-// < ergebnis: ein uintL >0
-extern uintL float_digits (const cl_DF& x);
+// < ergebnis: ein uintC >0
+extern uintC float_digits (const cl_DF& x);
// float_precision(x) liefert (float-precision x), wo x ein Float ist.
-// < ergebnis: ein uintL >=0
-extern uintL float_precision (const cl_DF& x);
+// < ergebnis: ein uintC >=0
+extern uintC float_precision (const cl_DF& x);
// integer_decode_float(x) liefert zu einem Float x: (integer-decode-float x).
// scale_float(x,delta) liefert x*2^delta, wo x ein DF ist.
-extern const cl_DF scale_float (const cl_DF& x, sintL delta);
+extern const cl_DF scale_float (const cl_DF& x, sintC delta);
extern const cl_DF scale_float (const cl_DF& x, const cl_I& delta);
extern const cl_FF float_sign (const cl_FF& x);
// float_digits(x) liefert (float-digits x), wo x ein Float ist.
-// < ergebnis: ein uintL >0
-extern uintL float_digits (const cl_FF& x);
+// < ergebnis: ein uintC >0
+extern uintC float_digits (const cl_FF& x);
// float_precision(x) liefert (float-precision x), wo x ein Float ist.
-// < ergebnis: ein uintL >=0
-extern uintL float_precision (const cl_FF& x);
+// < ergebnis: ein uintC >=0
+extern uintC float_precision (const cl_FF& x);
// integer_decode_float(x) liefert zu einem Float x: (integer-decode-float x).
// scale_float(x,delta) liefert x*2^delta, wo x ein FF ist.
-extern const cl_FF scale_float (const cl_FF& x, sintL delta);
+extern const cl_FF scale_float (const cl_FF& x, sintC delta);
extern const cl_FF scale_float (const cl_FF& x, const cl_I& delta);
extern const cl_F float_sign (const cl_F& x, const cl_F& y);
// float_digits(x) liefert (float-digits x), wo x ein Float ist.
-// < ergebnis: ein uintL >0
-extern uintL float_digits (const cl_F& x);
+// < ergebnis: ein uintC >0
+extern uintC float_digits (const cl_F& x);
// float_precision(x) liefert (float-precision x), wo x ein Float ist.
-// < ergebnis: ein uintL >=0
-extern uintL float_precision (const cl_F& x);
+// < ergebnis: ein uintC >=0
+extern uintC float_precision (const cl_F& x);
// Returns the floating point format of a float.
inline float_format_t float_format (const cl_F& x)
// scale_float(x,delta) liefert x*2^delta, wo x ein Float ist.
-extern const cl_F scale_float (const cl_F& x, sintL delta);
+extern const cl_F scale_float (const cl_F& x, sintC delta);
extern const cl_F scale_float (const cl_F& x, const cl_I& delta);
// Prüft, ob (LOGBITP x y), wo x und y Integers sind.
// Ergebnis: /=0, wenn ja; =0, wenn nein.
-extern cl_boolean logbitp (uintL x, const cl_I& y);
+extern cl_boolean logbitp (uintC x, const cl_I& y);
extern cl_boolean logbitp (const cl_I& x, const cl_I& y);
// Prüft, ob (ODDP x), wo x ein Integer ist.
{ return (cl_boolean) (!oddp(x)); }
// (ASH x y), wo x und y Integers sind. Ergebnis Integer.
-extern const cl_I ash (const cl_I& x, sintL y);
+extern const cl_I ash (const cl_I& x, sintC y);
extern const cl_I ash (const cl_I& x, const cl_I& y);
// (LOGCOUNT x), wo x ein Integer ist. Ergebnis uintL.
-extern uintL logcount (const cl_I& x);
+extern uintC logcount (const cl_I& x);
// (INTEGER-LENGTH x), wo x ein Integer ist. Ergebnis uintL.
-extern uintL integer_length (const cl_I& x);
+extern uintC integer_length (const cl_I& x);
// (ORD2 x) = max{n>=0: 2^n | x }, wo x ein Integer /=0 ist. Ergebnis uintL.
-extern uintL ord2 (const cl_I& x);
+extern uintC ord2 (const cl_I& x);
// power2p(x) stellt fest, ob ein Integer x>0 eine Zweierpotenz ist.
// Ergebnis: n>0, wenn x=2^(n-1), 0 sonst.
-extern uintL power2p (const cl_I& x);
+extern uintC power2p (const cl_I& x);
inline const cl_I operator| (const cl_I& x, const cl_I& y)
{ return logior(x,y); }
extern const cl_I abs (const cl_I& x);
// Shifts.
-inline const cl_I operator<< (const cl_I& x, sintL y) // assume 0 <= y < 2^31
+inline const cl_I operator<< (const cl_I& x, sintC y) // assume 0 <= y < 2^(intCsize-1)
{ return ash(x,y); }
inline const cl_I operator<< (const cl_I& x, const cl_I& y) // assume y >= 0
{ return ash(x,y); }
-inline const cl_I operator>> (const cl_I& x, sintL y) // assume 0 <= y < 2^31
+inline const cl_I operator>> (const cl_I& x, sintC y) // assume 0 <= y < 2^(intCsize-1)
{ return ash(x,-y); }
inline const cl_I operator>> (const cl_I& x, const cl_I& y) // assume y >= 0
{ return ash(x,-y); }
// BYTE-Operationen auf Integers
struct cl_byte {
- uintL size;
- uintL position;
+ uintC size;
+ uintC position;
// Konstruktor:
- cl_byte (unsigned int s, unsigned int p) : size (s), position (p) {}
+ cl_byte (uintC s, uintC p) : size (s), position (p) {}
};
// (LDB byte n), wo n ein Integer ist.
inline cl_I& operator-- /* prefix */ (cl_I& x) { return x = minus1(x); }
inline void operator-- /* postfix */ (cl_I& x, int dummy) { (void)dummy; x = minus1(x); }
inline cl_I& operator*= (cl_I& x, const cl_I& y) { return x = x * y; }
-inline cl_I& operator<<= (cl_I& x, sintL y) // assume 0 <= y < 2^31
+inline cl_I& operator<<= (cl_I& x, sintC y) // assume 0 <= y < 2^(intCsize-1)
{ return x = x << y; }
inline cl_I& operator<<= (cl_I& x, const cl_I& y) // assume y >= 0
{ return x = x << y; }
-inline cl_I& operator>>= (cl_I& x, sintL y) // assume 0 <= y < 2^31
+inline cl_I& operator>>= (cl_I& x, sintC y) // assume 0 <= y < 2^(intCsize-1)
{ return x = x >> y; }
inline cl_I& operator>>= (cl_I& x, const cl_I& y) // assume y >= 0
{ return x = x >> y; }
extern const cl_LF float_sign (const cl_LF& x);
// float_digits(x) liefert (float-digits x), wo x ein Float ist.
-// < ergebnis: ein uintL >0
-extern uintL float_digits (const cl_LF& x);
+// < ergebnis: ein uintC >0
+extern uintC float_digits (const cl_LF& x);
// float_precision(x) liefert (float-precision x), wo x ein Float ist.
-// < ergebnis: ein uintL >=0
-extern uintL float_precision (const cl_LF& x);
+// < ergebnis: ein uintC >=0
+extern uintC float_precision (const cl_LF& x);
// integer_decode_float(x) liefert zu einem Float x: (integer-decode-float x).
// scale_float(x,delta) liefert x*2^delta, wo x ein LF ist.
-extern const cl_LF scale_float (const cl_LF& x, sintL delta);
+extern const cl_LF scale_float (const cl_LF& x, sintC delta);
extern const cl_LF scale_float (const cl_LF& x, const cl_I& delta);
return _retract(x);
}
// Miscellaneous.
- sintL bits; // number of bits needed to represent a representative, or -1
+ sintC bits; // number of bits needed to represent a representative, or -1
int log2_bits; // log_2(bits), or -1
// Property operations.
cl_property* get_property (const cl_symbol& key)
{ return y.ring()->minus(y.ring()->canonhom(x),y); }
// Shifts.
-extern const cl_MI operator<< (const cl_MI& x, sintL y); // assume 0 <= y < 2^31
-extern const cl_MI operator>> (const cl_MI& x, sintL y); // assume m odd, 0 <= y < 2^31
+extern const cl_MI operator<< (const cl_MI& x, sintC y); // assume 0 <= y < 2^(intCsize-1)
+extern const cl_MI operator>> (const cl_MI& x, sintC y); // assume m odd, 0 <= y < 2^(intCsize-1)
// Equality.
inline bool operator== (const cl_MI& x, const cl_MI& y)
extern const cl_SF float_sign (const cl_SF& x);
// float_digits(x) liefert (float-digits x), wo x ein Float ist.
-// < ergebnis: ein uintL >0
-extern uintL float_digits (const cl_SF& x);
+// < ergebnis: ein uintC >0
+extern uintC float_digits (const cl_SF& x);
// float_precision(x) liefert (float-precision x), wo x ein Float ist.
-// < ergebnis: ein uintL >=0
-extern uintL float_precision (const cl_SF& x);
+// < ergebnis: ein uintC >=0
+extern uintC float_precision (const cl_SF& x);
// integer_decode_float(x) liefert zu einem Float x: (integer-decode-float x).
// scale_float(x,delta) liefert x*2^delta, wo x ein SF ist.
-extern const cl_SF scale_float (const cl_SF& x, sintL delta);
+extern const cl_SF scale_float (const cl_SF& x, sintC delta);
extern const cl_SF scale_float (const cl_SF& x, const cl_I& delta);
#define signean_minus -1
// Integer type used for counters.
+// Constraint: sizeof(uintC) >= sizeof(uintL)
#if (defined(HAVE_FAST_LONGLONG) && (defined(__alpha__) || defined(__ia64__) || defined(__powerpc64__) || defined(__x86_64__)))
#define intCsize long_bitsize
typedef long sintC;
}
#endif
+// Bits einer uintC-Zahl zählen:
+// integerlengthC(digit,size=);
+// setzt size auf die höchste in digit vorkommende Bitnummer.
+// > digit: ein uintC >0
+// < size: >0, <=intCsize, mit 2^(size-1) <= digit < 2^size
+ #if (intCsize==32)
+ #define integerlengthC integerlength32
+ #endif
+ #if (intCsize==64)
+ #define integerlengthC integerlength64
+ #endif
+
// Hintere Nullbits eines 32-Bit-Wortes zählen:
// ord2_32(digit,count=);
// setzt size auf die kleinste in digit vorkommende Bitnummer.
//
// Break-even-point. When in doubt, prefer to choose the standard algorithm.
#if CL_USE_GMP
- static inline cl_boolean cl_recip_suitable (uintL m, uintL n) // n <= m
+ static inline cl_boolean cl_recip_suitable (uintC m, uintC n) // n <= m
{ if (n < 2000)
return cl_false;
else // when n >= 4400/(m/n)^2, i.e. (m/66)^2 > n
- { var uintL mq = floor(m,66);
- if ((mq >= bit(16)) || ((uintL)(mq*mq) > n))
+ { var uintC mq = floor(m,66);
+ if ((mq >= bit(intCsize/2)) || (mq*mq > n))
return cl_true;
else
return cl_false;
// 1.8*N / N : Newton for N >= 500
// 1.9*N / N : Newton for N >= 500
// 2.0*N / N : Newton for N >= 500
- static inline cl_boolean cl_recip_suitable (uintL m, uintL n) // n <= m
+ static inline cl_boolean cl_recip_suitable (uintC m, uintC n) // n <= m
{ if (n < 500)
return cl_false;
else // when n >= 2100/(m/n)^2, i.e. (m/46)^2 > n
- { var uintL mq = floor(m,46);
- if ((mq >= bit(16)) || ((uintL)(mq*mq) > n))
+ { var uintC mq = floor(m,46);
+ if ((mq >= bit(intCsize/2)) || (mq*mq > n))
return cl_true;
else
return cl_false;
// return 2*b-a*b^2 mod 2^(intDsize*2*n).
CL_ALLOCA_STACK;
var uintL k = 0; // number of Newton steps
- var uintL n = len;
+ var uintC n = len;
while (n >= recip2adic_threshold) {
n = ceiling(n,2);
k++;
var uintD* b2_LSDptr;
var uintD* prod_LSDptr;
num_stack_alloc(len+1,,b2_LSDptr=);
- num_stack_alloc(2*(uintL)len,,prod_LSDptr=);
+ num_stack_alloc(2*len,,prod_LSDptr=);
do {
// n = ceiling(len/2^k)
// Compute n2 = ceiling(len/2^(k-1)),
// then n = ceiling(n2/2).
k--;
- var uintL n2 = ((len-1)>>k)+1; // = 2*n or = 2*n-1
+ var uintC n2 = ((len-1)>>k)+1; // = 2*n or = 2*n-1
// Set b := 2*b-a*b^2 mod 2^(intDsize*n2)
cl_UDS_mul_square(dest_LSDptr,n,b2_LSDptr); // b^2
cl_UDS_mul(b2_LSDptr,n2,a_LSDptr,n2,prod_LSDptr); // a*b^2
// zusätzlich belegt wird.
#define num_stack_array(need,low_zuweisung,high_zuweisung) \
- {var uintL __need = (uintL)(need); \
+ {var uintC __need = (uintC)(need); \
var uintD* __array = cl_alloc_array(uintD,__need); \
unused (low_zuweisung &__array[0]); unused (high_zuweisung &__array[__need]); \
}
#define num_stack_small_array(need,low_zuweisung,high_zuweisung) \
- {var uintL __need = (uintL)(need); \
+ {var uintC __need = (uintC)(need); \
var uintD* __array = cl_small_alloc_array(uintD,__need); \
unused (low_zuweisung &__array[0]); unused (high_zuweisung &__array[__need]); \
}
#define num_stack_small_alloc(need,MSDptr_zuweisung,LSDptr_zuweisung) \
num_stack_small_array(need,MSDptr_zuweisung,LSDptr_zuweisung)
#define num_stack_alloc_1(need,MSDptr_zuweisung,LSDptr_zuweisung) \
- num_stack_array((uintL)(need)+1,MSDptr_zuweisung 1 + ,LSDptr_zuweisung)
+ num_stack_array((uintC)(need)+1,MSDptr_zuweisung 1 + ,LSDptr_zuweisung)
#define num_stack_small_alloc_1(need,MSDptr_zuweisung,LSDptr_zuweisung) \
- num_stack_small_array((uintL)(need)+1,MSDptr_zuweisung 1 + ,LSDptr_zuweisung)
+ num_stack_small_array((uintC)(need)+1,MSDptr_zuweisung 1 + ,LSDptr_zuweisung)
#else
#define num_stack_alloc(need,MSDptr_zuweisung,LSDptr_zuweisung) \
num_stack_array(need,LSDptr_zuweisung,MSDptr_zuweisung)
#define num_stack_small_alloc(need,MSDptr_zuweisung,LSDptr_zuweisung) \
num_stack_small_array(need,LSDptr_zuweisung,MSDptr_zuweisung)
#define num_stack_alloc_1(need,MSDptr_zuweisung,LSDptr_zuweisung) \
- num_stack_array((uintL)(need)+1,LSDptr_zuweisung,MSDptr_zuweisung -1 + )
+ num_stack_array((uintC)(need)+1,LSDptr_zuweisung,MSDptr_zuweisung -1 + )
#define num_stack_small_alloc_1(need,MSDptr_zuweisung,LSDptr_zuweisung) \
- num_stack_small_array((uintL)(need)+1,LSDptr_zuweisung,MSDptr_zuweisung -1 + )
+ num_stack_small_array((uintC)(need)+1,LSDptr_zuweisung,MSDptr_zuweisung -1 + )
#endif
// Ergebnis ist die UDS MSDptr/len/LSDptr, mit len=len1+len2, im Stack.
// Dabei wird num_stack erniedrigt.
#define UDS_UDS_mul_UDS(len1,LSDptr1,len2,LSDptr2, MSDptr_zuweisung,len_zuweisung,LSDptr_zuweisung) \
- var uintL CONCAT(len_from_UDSmul_,__LINE__) = (uintL)(len1) + (uintL)(len2); \
+ var uintC CONCAT(len_from_UDSmul_,__LINE__) = (uintC)(len1) + (uintC)(len2); \
var uintD* CONCAT(LSDptr_from_UDSmul_,__LINE__); \
unused (len_zuweisung CONCAT(len_from_UDSmul_,__LINE__)); \
num_stack_alloc(CONCAT(len_from_UDSmul_,__LINE__),MSDptr_zuweisung,LSDptr_zuweisung CONCAT(LSDptr_from_UDSmul_,__LINE__) =); \
#define DS_DS_mul_DS(MSDptr1,len1,LSDptr1,MSDptr2,len2,LSDptr2, MSDptr_zuweisung,len_zuweisung,LSDptr_zuweisung) \
var uintD* MSDptr0; \
var uintD* LSDptr0; \
- var uintL len_from_DSmal = (uintL)(len1) + (uintL)(len2); \
+ var uintC len_from_DSmal = (uintC)(len1) + (uintC)(len2); \
unused (len_zuweisung len_from_DSmal); \
num_stack_alloc(len_from_DSmal,MSDptr_zuweisung MSDptr0 =,LSDptr_zuweisung LSDptr0 =); \
var uintD MSD1_from_DSmal = mspref(MSDptr1,0); \
var uintD MSD2_from_DSmal = mspref(MSDptr2,0); \
- var uintL len1_from_DSmal = (len1); \
- var uintL len2_from_DSmal = (len2); \
+ var uintC len1_from_DSmal = (len1); \
+ var uintC len2_from_DSmal = (len2); \
if (MSD1_from_DSmal==0) { msprefnext(MSDptr0) = 0; len1_from_DSmal--; } \
if (MSD2_from_DSmal==0) { msprefnext(MSDptr0) = 0; len2_from_DSmal--; } \
cl_UDS_mul((LSDptr1),len1_from_DSmal,(LSDptr2),len2_from_DSmal,LSDptr0); \
const unsigned int cl_fftm_threshold2 = 2*cl_fftm_threshold;
// len1 > cl_fftm_threshold1 && len2 > cl_fftm_threshold2
// && len1 >= cl_fftm_threshold1 + cl_fftm_threshold/(len2-cl_fftm_threshold1)*(cl_fftm_threshold-cl_fftm_threshold1).
- static inline cl_boolean cl_fftm_suitable (uintL len1, uintL len2)
+ static inline cl_boolean cl_fftm_suitable (uintC len1, uintC len2)
{ if (len1 >= cl_fftm_threshold)
return cl_true;
if (len1 > cl_fftm_threshold1)
if (len2 > cl_fftm_threshold2)
- { var uint32 hi;
+ { const unsigned int prod_threshold = cl_fftm_threshold*(cl_fftm_threshold-cl_fftm_threshold1);
+ if (len1-cl_fftm_threshold1 >= prod_threshold)
+ return cl_true;
+ if (len2-cl_fftm_threshold1 >= prod_threshold)
+ return cl_true;
+ var uint32 hi;
var uint32 lo;
mulu32(len1-cl_fftm_threshold1,len2-cl_fftm_threshold1, hi=,lo=);
- if (hi > 0 || lo >= cl_fftm_threshold*(cl_fftm_threshold-cl_fftm_threshold1))
+ if (hi > 0 || lo >= prod_threshold)
return cl_true;
}
return cl_false;
#ifndef _BIT_REVERSE
#define _BIT_REVERSE
// Reverse an n-bit number x. n>0.
-static uintL bit_reverse (uintL n, uintL x)
+static uintC bit_reverse (uintL n, uintC x)
{
- var uintL y = 0;
+ var uintC y = 0;
do {
y <<= 1;
y |= (x & 1);
#endif
// Compute a complex convolution using FFT: z[0..N-1] := x[0..N-1] * y[0..N-1].
-static void fftc_convolution (const uintL n, const uintL N, // N = 2^n
+static void fftc_convolution (const uintL n, const uintC N, // N = 2^n
fftc_complex * x, // N numbers
fftc_complex * y, // N numbers
fftc_complex * z // N numbers result
#else
var fftc_complex* const w = cl_alloc_array(fftc_complex,(N>>1)+1);
#endif
- var uintL i;
+ var uintC i;
// Initialize w[i] to w^i, w a primitive N-th root of unity.
w[0] = fftc_roots_of_1[0];
#if (FFTC_BACKWARD == RECIPROOT) || defined(DEBUG_FFTC)
{
var sintL l;
/* l = n-1 */ {
- var const uintL tmax = N>>1; // tmax = 2^(n-1)
- for (var uintL t = 0; t < tmax; t++) {
- var uintL i1 = t;
- var uintL i2 = i1 + tmax;
+ var const uintC tmax = N>>1; // tmax = 2^(n-1)
+ for (var uintC t = 0; t < tmax; t++) {
+ var uintC i1 = t;
+ var uintC i2 = i1 + tmax;
// Butterfly: replace (x(i1),x(i2)) by
// (x(i1) + x(i2), x(i1) - x(i2)).
var fftc_complex tmp;
}
}
for (l = n-2; l>=0; l--) {
- var const uintL smax = (uintL)1 << (n-1-l);
- var const uintL tmax = (uintL)1 << l;
- for (var uintL s = 0; s < smax; s++) {
- var uintL exp = bit_reverse(n-1-l,s) << l;
- for (var uintL t = 0; t < tmax; t++) {
- var uintL i1 = (s << (l+1)) + t;
- var uintL i2 = i1 + tmax;
+ var const uintC smax = (uintC)1 << (n-1-l);
+ var const uintC tmax = (uintC)1 << l;
+ for (var uintC s = 0; s < smax; s++) {
+ var uintC exp = bit_reverse(n-1-l,s) << l;
+ for (var uintC t = 0; t < tmax; t++) {
+ var uintC i1 = (s << (l+1)) + t;
+ var uintC i2 = i1 + tmax;
// Butterfly: replace (x(i1),x(i2)) by
// (x(i1) + w^exp*x(i2), x(i1) - w^exp*x(i2)).
var fftc_complex tmp;
if (!squaring) {
var sintL l;
/* l = n-1 */ {
- var uintL const tmax = N>>1; // tmax = 2^(n-1)
- for (var uintL t = 0; t < tmax; t++) {
- var uintL i1 = t;
- var uintL i2 = i1 + tmax;
+ var uintC const tmax = N>>1; // tmax = 2^(n-1)
+ for (var uintC t = 0; t < tmax; t++) {
+ var uintC i1 = t;
+ var uintC i2 = i1 + tmax;
// Butterfly: replace (y(i1),y(i2)) by
// (y(i1) + y(i2), y(i1) - y(i2)).
var fftc_complex tmp;
}
}
for (l = n-2; l>=0; l--) {
- var const uintL smax = (uintL)1 << (n-1-l);
- var const uintL tmax = (uintL)1 << l;
- for (var uintL s = 0; s < smax; s++) {
- var uintL exp = bit_reverse(n-1-l,s) << l;
- for (var uintL t = 0; t < tmax; t++) {
- var uintL i1 = (s << (l+1)) + t;
- var uintL i2 = i1 + tmax;
+ var const uintC smax = (uintC)1 << (n-1-l);
+ var const uintC tmax = (uintC)1 << l;
+ for (var uintC s = 0; s < smax; s++) {
+ var uintC exp = bit_reverse(n-1-l,s) << l;
+ for (var uintC t = 0; t < tmax; t++) {
+ var uintC i1 = (s << (l+1)) + t;
+ var uintC i2 = i1 + tmax;
// Butterfly: replace (y(i1),y(i2)) by
// (y(i1) + w^exp*y(i2), y(i1) - w^exp*y(i2)).
var fftc_complex tmp;
{
var uintL l;
for (l = 0; l < n-1; l++) {
- var const uintL smax = (uintL)1 << (n-1-l);
- var const uintL tmax = (uintL)1 << l;
+ var const uintC smax = (uintC)1 << (n-1-l);
+ var const uintC tmax = (uintC)1 << l;
#if FFTC_BACKWARD != CLEVER
- for (var uintL s = 0; s < smax; s++) {
- var uintL exp = bit_reverse(n-1-l,s) << l;
+ for (var uintC s = 0; s < smax; s++) {
+ var uintC exp = bit_reverse(n-1-l,s) << l;
#if FFTC_BACKWARD == RECIPROOT
if (exp > 0)
exp = N - exp; // negate exp (use w^-1 instead of w)
#endif
- for (var uintL t = 0; t < tmax; t++) {
- var uintL i1 = (s << (l+1)) + t;
- var uintL i2 = i1 + tmax;
+ for (var uintC t = 0; t < tmax; t++) {
+ var uintC i1 = (s << (l+1)) + t;
+ var uintC i2 = i1 + tmax;
// Inverse Butterfly: replace (z(i1),z(i2)) by
// ((z(i1)+z(i2))/2, (z(i1)-z(i2))/(2*w^exp)).
// Do the division by 2 later.
}
#else // FFTC_BACKWARD == CLEVER: clever handling of negative exponents
/* s = 0, exp = 0 */ {
- for (var uintL t = 0; t < tmax; t++) {
- var uintL i1 = t;
- var uintL i2 = i1 + tmax;
+ for (var uintC t = 0; t < tmax; t++) {
+ var uintC i1 = t;
+ var uintC i2 = i1 + tmax;
// Inverse Butterfly: replace (z(i1),z(i2)) by
// ((z(i1)+z(i2))/2, (z(i1)-z(i2))/(2*w^exp)),
// with exp <-- 0.
z[i2] = diff;
}
}
- for (var uintL s = 1; s < smax; s++) {
- var uintL exp = bit_reverse(n-1-l,s) << l;
+ for (var uintC s = 1; s < smax; s++) {
+ var uintC exp = bit_reverse(n-1-l,s) << l;
exp = (N>>1) - exp; // negate exp (use w^-1 instead of w)
- for (var uintL t = 0; t < tmax; t++) {
- var uintL i1 = (s << (l+1)) + t;
- var uintL i2 = i1 + tmax;
+ for (var uintC t = 0; t < tmax; t++) {
+ var uintC i1 = (s << (l+1)) + t;
+ var uintC i2 = i1 + tmax;
// Inverse Butterfly: replace (z(i1),z(i2)) by
// ((z(i1)+z(i2))/2, (z(i1)-z(i2))/(2*w^exp)),
// with exp <-- (N/2 - exp).
}
/* l = n-1 */ {
var const fftc_real pow2 = (fftc_real)1 / (fftc_real)N;
- var const uintL tmax = N>>1; // tmax = 2^(n-1)
- for (var uintL t = 0; t < tmax; t++) {
- var uintL i1 = t;
- var uintL i2 = i1 + tmax;
+ var const uintC tmax = N>>1; // tmax = 2^(n-1)
+ for (var uintC t = 0; t < tmax; t++) {
+ var uintC i1 = t;
+ var uintC i2 = i1 + tmax;
// Inverse Butterfly: replace (z(i1),z(i2)) by
// ((z(i1)+z(i2))/2, (z(i1)-z(i2))/2).
// Do all the divisions by 2 now.
// Split len uintD's below sourceptr into chunks of l bits, thus filling
// N complex numbers at x.
-static void fill_factor (uintL N, fftc_complex* x, uintL l,
- const uintD* sourceptr, uintL len)
+static void fill_factor (uintC N, fftc_complex* x, uintL l,
+ const uintD* sourceptr, uintC len)
{
- var uintL i;
+ var uintC i;
if (max_l(2) > intDsize && l > intDsize) {
// l > intDsize
if (max_l(2) > 64 && l > 64) {
// The z[i] are known to be approximately integers >= 0, < N*2^(2*l).
// Assumes room for floor(N*l/intDsize)+(1+ceiling((n+2*l)/intDsize)) uintD's
// below destptr. Fills len digits and returns (destptr lspop len).
-static uintD* unfill_product (uintL n, uintL N, // N = 2^n
+static uintD* unfill_product (uintL n, uintC N, // N = 2^n
const fftc_complex * z, uintL l,
uintD* destptr)
{
- var uintL i;
+ var uintC i;
if (n + 2*l <= intDsize) {
// 2-digit carry is sufficient, l < intDsize
var uintD carry0 = 0;
// is 2*len1*intDsize/l_max - 1 <= 2^k.
{
var const int l = max_l(2);
- var uintL lhs = 2*ceiling((uintL)len1*intDsize,l) - 1; // >=1
+ var uintC lhs = 2*ceiling(len1*intDsize,l) - 1; // >=1
if (lhs < 3)
k = 2;
else
- integerlength32(lhs-1, k=); // k>=2
+ integerlengthC(lhs-1, k=); // k>=2
}
// Try whether this k is ok or whether we have to increase k.
for ( ; ; k++) {
fprint(std::cerr, "FFT problem: numbers too big, floating point precision not sufficient\n");
cl_abort();
}
- if (2*ceiling((uintL)len1*intDsize,max_l_table[k])-1 <= ((uintL)1 << k))
+ if (2*ceiling(len1*intDsize,max_l_table[k])-1 <= ((uintC)1 << k))
break;
}
// We could try to reduce l, keeping the same k. But why should we?
// Calculate the number of pieces in which source2 will have to be
// split. Each of the pieces must satisfy
// ceiling(len1*intDsize/l) + ceiling(len2*intDsize/l) - 1 <= 2^k,
- var uintL len2p;
+ var uintC len2p;
// Try once with k, once with k+1. Compare them.
{
- var uintL remaining_k = ((uintL)1 << k) + 1 - ceiling((uintL)len1*intDsize,max_l_table[k]);
- var uintL max_piecelen_k = floor(remaining_k*max_l_table[k],intDsize);
- var uintL numpieces_k = ceiling(len2,max_piecelen_k);
- var uintL remaining_k1 = ((uintL)1 << (k+1)) + 1 - ceiling((uintL)len1*intDsize,max_l_table[k+1]);
- var uintL max_piecelen_k1 = floor(remaining_k1*max_l_table[k+1],intDsize);
- var uintL numpieces_k1 = ceiling(len2,max_piecelen_k1);
+ var uintC remaining_k = ((uintC)1 << k) + 1 - ceiling(len1*intDsize,max_l_table[k]);
+ var uintC max_piecelen_k = floor(remaining_k*max_l_table[k],intDsize);
+ var uintC numpieces_k = ceiling(len2,max_piecelen_k);
+ var uintC remaining_k1 = ((uintC)1 << (k+1)) + 1 - ceiling(len1*intDsize,max_l_table[k+1]);
+ var uintC max_piecelen_k1 = floor(remaining_k1*max_l_table[k+1],intDsize);
+ var uintC numpieces_k1 = ceiling(len2,max_piecelen_k1);
if (numpieces_k <= 2*numpieces_k1) {
// keep k
len2p = max_piecelen_k;
}
var const uintL l = max_l_table[k];
var const uintL n = k;
- var const uintL N = (uintL)1 << n;
+ var const uintC N = (uintC)1 << n;
CL_ALLOCA_STACK;
var fftc_complex* const x = cl_alloc_array(fftc_complex,N);
var fftc_complex* const y = cl_alloc_array(fftc_complex,N);
var fftc_complex* const z = x; // put z in place of x - saves memory
#endif
var uintD* const tmpprod1 = cl_alloc_array(uintD,len1+1);
- var uintL tmpprod_len = floor(l<<n,intDsize)+6;
+ var uintC tmpprod_len = floor(l<<n,intDsize)+6;
var uintD* const tmpprod = cl_alloc_array(uintD,tmpprod_len);
- var uintL destlen = len1+len2;
+ var uintC destlen = len1+len2;
clear_loop_lsp(destptr,destlen);
do {
if (len2p > len2)
var fftc_real re_hi_limit = (fftc_real)N * fftc_pow2_table[l] * fftc_pow2_table[l] + (fftc_real)0.5;
var fftc_real im_lo_limit = (fftc_real)(-0.5);
var fftc_real im_hi_limit = (fftc_real)0.5;
- for (var uintL i = 0; i < N; i++) {
+ for (var uintC i = 0; i < N; i++) {
if (!(z[i].im > im_lo_limit
&& z[i].im < im_hi_limit))
cl_abort();
#endif
var uintD* tmpLSDptr = arrayLSDptr(tmpprod,tmpprod_len);
var uintD* tmpMSDptr = unfill_product(n,N,z,l,tmpLSDptr);
- var uintL tmplen =
+ var uintC tmplen =
#if CL_DS_BIG_ENDIAN_P
tmpLSDptr - tmpMSDptr;
#else
r.im = r_im;
}
-static uintL shuffle (uintL n, uintL x)
+static uintC shuffle (uintL n, uintC x)
{
- var uintL y = 0;
- var sintL v = 1;
+ var uintC y = 0;
+ var sintC v = 1;
// Invariant: y + v*shuffle(n,x).
do {
if (x & 1)
}
#if 0 // unused
-static uintL invshuffle (uintL n, uintL x)
+static uintC invshuffle (uintL n, uintC x)
{
- var uintL y = 0;
- var uintL v = 1;
+ var uintC y = 0;
+ var uintC v = 1;
// Invariant: y + v*invshuffle(n,x).
do {
- if (x == ((uintL)1 << (n-1)))
+ if (x == ((uintC)1 << (n-1)))
return y + v;
- else if (x > ((uintL)1 << (n-1))) {
- x = ((uintL)1 << n) - x;
+ else if (x > ((uintC)1 << (n-1))) {
+ x = ((uintC)1 << n) - x;
y = y+v;
}
v <<= 1;
#endif
// Compute a real convolution using FFT: z[0..N-1] := x[0..N-1] * y[0..N-1].
-static void fftcs_convolution (const uintL n, const uintL N, // N = 2^n
+static void fftcs_convolution (const uintL n, const uintC N, // N = 2^n
fftcs_real * x, // N numbers
fftcs_real * y, // N numbers
fftcs_real * z // N numbers result
{
CL_ALLOCA_STACK;
var fftcs_complex* const w = cl_alloc_array(fftcs_complex,N>>2);
- var uintL i;
+ var uintC i;
// Initialize w[i] to w^i, w a primitive N-th root of unity.
w[0] = fftcs_roots_of_1[0];
{
var uintL l;
for (l = n-1; l > 0; l--) {
/* s = 0 */ {
- var const uintL tmax = (uintL)1 << l;
- for (var uintL t = 0; t < tmax; t++) {
- var uintL i1 = t;
- var uintL i2 = i1 + tmax;
+ var const uintC tmax = (uintC)1 << l;
+ for (var uintC t = 0; t < tmax; t++) {
+ var uintC i1 = t;
+ var uintC i2 = i1 + tmax;
// replace (x[i1],x[i2]) by
// (x[i1] + x[i2], x[i1] - x[i2])
var fftcs_real tmp;
x[i1] = x[i1] + tmp;
}
}
- var const uintL smax = (uintL)1 << (n-1-l);
- var const uintL tmax = (uintL)1 << (l-1);
- for (var uintL s = 1; s < smax; s++) {
- var uintL exp = shuffle(n-1-l,s) << (l-1);
- for (var uintL t = 0; t < tmax; t++) {
- var uintL i1 = (s << (l+1)) + t;
- var uintL i2 = i1 + tmax;
- var uintL i3 = i2 + tmax;
- var uintL i4 = i3 + tmax;
+ var const uintC smax = (uintC)1 << (n-1-l);
+ var const uintC tmax = (uintC)1 << (l-1);
+ for (var uintC s = 1; s < smax; s++) {
+ var uintC exp = shuffle(n-1-l,s) << (l-1);
+ for (var uintC t = 0; t < tmax; t++) {
+ var uintC i1 = (s << (l+1)) + t;
+ var uintC i2 = i1 + tmax;
+ var uintC i3 = i2 + tmax;
+ var uintC i4 = i3 + tmax;
// replace (x[i1],x[i2],x[i3],x[i4]) by
// (x[i1] + x[i2]*Re(w^exp) - x[i4]*Im(w^exp),
// x[i3] + x[i4]*Re(w^exp) + x[i2]*Im(w^exp),
var uintL l;
for (l = n-1; l > 0; l--) {
/* s = 0 */ {
- var const uintL tmax = (uintL)1 << l;
- for (var uintL t = 0; t < tmax; t++) {
- var uintL i1 = t;
- var uintL i2 = i1 + tmax;
+ var const uintC tmax = (uintC)1 << l;
+ for (var uintC t = 0; t < tmax; t++) {
+ var uintC i1 = t;
+ var uintC i2 = i1 + tmax;
// replace (y[i1],y[i2]) by
// (y[i1] + y[i2], y[i1] - y[i2])
var fftcs_real tmp;
y[i1] = y[i1] + tmp;
}
}
- var const uintL smax = (uintL)1 << (n-1-l);
- var const uintL tmax = (uintL)1 << (l-1);
- for (var uintL s = 1; s < smax; s++) {
- var uintL exp = shuffle(n-1-l,s) << (l-1);
- for (var uintL t = 0; t < tmax; t++) {
- var uintL i1 = (s << (l+1)) + t;
- var uintL i2 = i1 + tmax;
- var uintL i3 = i2 + tmax;
- var uintL i4 = i3 + tmax;
+ var const uintC smax = (uintC)1 << (n-1-l);
+ var const uintC tmax = (uintC)1 << (l-1);
+ for (var uintC s = 1; s < smax; s++) {
+ var uintC exp = shuffle(n-1-l,s) << (l-1);
+ for (var uintC t = 0; t < tmax; t++) {
+ var uintC i1 = (s << (l+1)) + t;
+ var uintC i2 = i1 + tmax;
+ var uintC i3 = i2 + tmax;
+ var uintC i4 = i3 + tmax;
// replace (y[i1],y[i2],y[i3],y[i4]) by
// (y[i1] + y[i2]*Re(w^exp) - y[i4]*Im(w^exp),
// y[i3] + y[i4]*Re(w^exp) + y[i2]*Im(w^exp),
}
for (l = 1; l < n; l++) {
/* s = 0 */ {
- var const uintL tmax = (uintL)1 << l;
- for (var uintL t = 0; t < tmax; t++) {
- var uintL i1 = t;
- var uintL i2 = i1 + tmax;
+ var const uintC tmax = (uintC)1 << l;
+ for (var uintC t = 0; t < tmax; t++) {
+ var uintC i1 = t;
+ var uintC i2 = i1 + tmax;
// replace (z[i1],z[i2]) by
// ((z[i1]+z[i2])/2, (z[i1]-z[i2])/2)
// Do the division by 2 later.
z[i1] = z[i1] + tmp;
}
}
- var const uintL smax = (uintL)1 << (n-1-l);
- var const uintL tmax = (uintL)1 << (l-1);
- for (var uintL s = 1; s < smax; s++) {
- var uintL exp = shuffle(n-1-l,s) << (l-1);
- for (var uintL t = 0; t < tmax; t++) {
- var uintL i1 = (s << (l+1)) + t;
- var uintL i2 = i1 + tmax;
- var uintL i3 = i2 + tmax;
- var uintL i4 = i3 + tmax;
+ var const uintC smax = (uintC)1 << (n-1-l);
+ var const uintC tmax = (uintC)1 << (l-1);
+ for (var uintC s = 1; s < smax; s++) {
+ var uintC exp = shuffle(n-1-l,s) << (l-1);
+ for (var uintC t = 0; t < tmax; t++) {
+ var uintC i1 = (s << (l+1)) + t;
+ var uintC i2 = i1 + tmax;
+ var uintC i3 = i2 + tmax;
+ var uintC i4 = i3 + tmax;
// replace (z[i1],z[i2],z[i3],z[i4]) by
// ((z[i1]+z[i3])/2,
// (z[i1]-z[i3])/2*Re(w^exp)+(z[i2]+z[i4])/2*Im(w^exp),
// Split len uintD's below sourceptr into chunks of l bits, thus filling
// N real numbers at x.
-static void fill_factor (uintL N, fftcs_real* x, uintL l,
- const uintD* sourceptr, uintL len)
+static void fill_factor (uintC N, fftcs_real* x, uintL l,
+ const uintD* sourceptr, uintC len)
{
- var uintL i;
+ var uintC i;
if (max_l(2) > intDsize && l > intDsize) {
// l > intDsize
if (max_l(2) > 64 && l > 64) {
// The z[i] are known to be approximately integers >= 0, < N*2^(2*l).
// Assumes room for floor(N*l/intDsize)+(1+ceiling((n+2*l)/intDsize)) uintD's
// below destptr. Fills len digits and returns (destptr lspop len).
-static uintD* unfill_product (uintL n, uintL N, // N = 2^n
+static uintD* unfill_product (uintL n, uintC N, // N = 2^n
const fftcs_real * z, uintL l,
uintD* destptr)
{
- var uintL i;
+ var uintC i;
if (n + 2*l <= intDsize) {
// 2-digit carry is sufficient, l < intDsize
var uintD carry0 = 0;
// is 2*len1*intDsize/l_max - 1 <= 2^k.
{
var const int l = max_l(2);
- var uintL lhs = 2*ceiling((uintL)len1*intDsize,l) - 1; // >=1
+ var uintC lhs = 2*ceiling(len1*intDsize,l) - 1; // >=1
if (lhs < 3)
k = 2;
else
- integerlength32(lhs-1, k=); // k>=2
+ integerlengthC(lhs-1, k=); // k>=2
}
// Try whether this k is ok or whether we have to increase k.
for ( ; ; k++) {
fprint(std::cerr, "FFT problem: numbers too big, floating point precision not sufficient\n");
cl_abort();
}
- if (2*ceiling((uintL)len1*intDsize,max_l_table[k])-1 <= ((uintL)1 << k))
+ if (2*ceiling(len1*intDsize,max_l_table[k])-1 <= ((uintC)1 << k))
break;
}
// We could try to reduce l, keeping the same k. But why should we?
// Calculate the number of pieces in which source2 will have to be
// split. Each of the pieces must satisfy
// ceiling(len1*intDsize/l) + ceiling(len2*intDsize/l) - 1 <= 2^k,
- var uintL len2p;
+ var uintC len2p;
// Try once with k, once with k+1. Compare them.
{
- var uintL remaining_k = ((uintL)1 << k) + 1 - ceiling((uintL)len1*intDsize,max_l_table[k]);
- var uintL max_piecelen_k = floor(remaining_k*max_l_table[k],intDsize);
- var uintL numpieces_k = ceiling(len2,max_piecelen_k);
- var uintL remaining_k1 = ((uintL)1 << (k+1)) + 1 - ceiling((uintL)len1*intDsize,max_l_table[k+1]);
- var uintL max_piecelen_k1 = floor(remaining_k1*max_l_table[k+1],intDsize);
- var uintL numpieces_k1 = ceiling(len2,max_piecelen_k1);
+ var uintC remaining_k = ((uintC)1 << k) + 1 - ceiling(len1*intDsize,max_l_table[k]);
+ var uintC max_piecelen_k = floor(remaining_k*max_l_table[k],intDsize);
+ var uintC numpieces_k = ceiling(len2,max_piecelen_k);
+ var uintC remaining_k1 = ((uintC)1 << (k+1)) + 1 - ceiling(len1*intDsize,max_l_table[k+1]);
+ var uintC max_piecelen_k1 = floor(remaining_k1*max_l_table[k+1],intDsize);
+ var uintC numpieces_k1 = ceiling(len2,max_piecelen_k1);
if (numpieces_k <= 2*numpieces_k1) {
// keep k
len2p = max_piecelen_k;
}
var const uintL l = max_l_table[k];
var const uintL n = k;
- var const uintL N = (uintL)1 << n;
+ var const uintC N = (uintC)1 << n;
CL_ALLOCA_STACK;
var fftcs_real* const x = cl_alloc_array(fftcs_real,N);
var fftcs_real* const y = cl_alloc_array(fftcs_real,N);
var fftcs_real* const z = x; // put z in place of x - saves memory
#endif
var uintD* const tmpprod1 = cl_alloc_array(uintD,len1+1);
- var uintL tmpprod_len = floor(l<<n,intDsize)+6;
+ var uintC tmpprod_len = floor(l<<n,intDsize)+6;
var uintD* const tmpprod = cl_alloc_array(uintD,tmpprod_len);
- var uintL destlen = len1+len2;
+ var uintC destlen = len1+len2;
clear_loop_lsp(destptr,destlen);
do {
if (len2p > len2)
{
var fftcs_real re_lo_limit = (fftcs_real)(-0.5);
var fftcs_real re_hi_limit = (fftcs_real)N * fftcs_pow2_table[l] * fftcs_pow2_table[l] + (fftcs_real)0.5;
- for (var uintL i = 0; i < N; i++)
+ for (var uintC i = 0; i < N; i++)
if (!(z[i] > re_lo_limit
&& z[i] < re_hi_limit))
cl_abort();
#endif
var uintD* tmpLSDptr = arrayLSDptr(tmpprod,tmpprod_len);
var uintD* tmpMSDptr = unfill_product(n,N,z,l,tmpLSDptr);
- var uintL tmplen =
+ var uintC tmplen =
#if CL_DS_BIG_ENDIAN_P
tmpLSDptr - tmpMSDptr;
#else
// Operations modulo p = 2^R+1, each chunk represented as chlen words
// (chlen = floor(R/intDsize)+1).
-static inline void assign (const uintL R, const uintL chlen,
+static inline void assign (const uintC R, const uintC chlen,
const uintD* a, uintD* r)
{
unused R;
}
// r := (a + b) mod p
-static void addm (const uintL R, const uintL chlen,
+static void addm (const uintC R, const uintC chlen,
const uintD* a, const uintD* b, uintD* r)
{
unused R;
}
// r := (a - b) mod p
-static void subm (const uintL R, const uintL chlen,
+static void subm (const uintC R, const uintC chlen,
const uintD* a, const uintD* b, uintD* r)
{
unused R;
// r := (a << s) mod p (0 <= s < R).
// Assume that a and r don't overlap.
-static void shiftleftm (const uintL R, const uintL chlen,
+static void shiftleftm (const uintC R, const uintC chlen,
const uintD* a, uintL s, uintD* r)
{
// Write a = 2^(R-s)*b + c, then
}
// r := (a * b) mod p
-static void mulm (const uintL R, const uintL chlen,
+static void mulm (const uintC R, const uintC chlen,
const uintD* a, const uintD* b, uintD* r)
{
unused R;
}
// b := (a / 2) mod p
-static void shiftm (const uintL R, const uintL chlen,
+static void shiftm (const uintC R, const uintC chlen,
const uintD* a, uintD* b)
{
unused R;
#ifndef _BIT_REVERSE
#define _BIT_REVERSE
// Reverse an n-bit number x. n>0.
-static uintL bit_reverse (uintL n, uintL x)
+static uintC bit_reverse (uintL n, uintC x)
{
- var uintL y = 0;
+ var uintC y = 0;
do {
y <<= 1;
y |= (x & 1);
}
#endif
-static void mulu_fftm (const uintL r, const uintL R, // R = 2^r
- const uintL m, const uintL M, // M = 2^m
- const uintL k, // K = intDsize*k
+static void mulu_fftm (const uintL r, const uintC R, // R = 2^r
+ const uintL m, const uintC M, // M = 2^m
+ const uintC k, // K = intDsize*k
const uintD* sourceptr1, uintC len1,
const uintD* sourceptr2, uintC len2,
uintD* destptr)
// R = 2^r, M = 2^m, M | 2R.
// m > 0.
{
- var const uintL chlen = floor(R,intDsize)+1; // chunk length (in words)
+ var const uintC chlen = floor(R,intDsize)+1; // chunk length (in words)
CL_ALLOCA_STACK;
var uintD* const arrX = cl_alloc_array(uintD,chlen<<m);
var uintD* const arrY = cl_alloc_array(uintD,chlen<<m);
num_stack_alloc(chlen,,sum=);
num_stack_alloc(chlen,,diff=);
var bool squaring = ((sourceptr1 == sourceptr2) && (len1 == len2));
- var uintL i;
+ var uintC i;
// Initialize factors X(i) and Y(i).
{
var const uintD* sptr = sourceptr1;
- var uintL slen = len1;
+ var uintC slen = len1;
for (i = 0; i < M; i++) {
var uintD* ptr = X(i);
if (slen >= k) {
}
if (!squaring) {
var const uintD* sptr = sourceptr2;
- var uintL slen = len2;
+ var uintC slen = len2;
for (i = 0; i < M; i++) {
var uintD* ptr = Y(i);
if (slen >= k) {
{
var sintL l;
/* l = m-1 */ {
- var const uintL tmax = M>>1; // tmax = 2^(m-1)
- for (var uintL t = 0; t < tmax; t++) {
- var uintL i1 = t;
- var uintL i2 = i1 + tmax;
+ var const uintC tmax = M>>1; // tmax = 2^(m-1)
+ for (var uintC t = 0; t < tmax; t++) {
+ var uintC i1 = t;
+ var uintC i2 = i1 + tmax;
// Butterfly: replace (X(i1),X(i2)) by
// (X(i1) + X(i2), X(i1) - X(i2)).
assign(R,chlen, X(i2), tmp);
}
}
for (l = m-2; l>=0; l--) {
- var const uintL smax = (uintL)1 << (m-1-l);
- var const uintL tmax = (uintL)1 << l;
- for (var uintL s = 0; s < smax; s++) {
+ var const uintC smax = (uintC)1 << (m-1-l);
+ var const uintC tmax = (uintC)1 << l;
+ for (var uintC s = 0; s < smax; s++) {
// w^exp = 2^(exp << (r+1-m)).
- var uintL exp = bit_reverse(m-1-l,s) << (r-(m-1-l));
- for (var uintL t = 0; t < tmax; t++) {
- var uintL i1 = (s << (l+1)) + t;
- var uintL i2 = i1 + tmax;
+ var uintC exp = bit_reverse(m-1-l,s) << (r-(m-1-l));
+ for (var uintC t = 0; t < tmax; t++) {
+ var uintC i1 = (s << (l+1)) + t;
+ var uintC i2 = i1 + tmax;
// Butterfly: replace (X(i1),X(i2)) by
// (X(i1) + w^exp*X(i2), X(i1) - w^exp*X(i2)).
shiftleftm(R,chlen, X(i2),exp, tmp);
if (!squaring) {
var sintL l;
/* l = m-1 */ {
- var const uintL tmax = M>>1; // tmax = 2^(m-1)
- for (var uintL t = 0; t < tmax; t++) {
- var uintL i1 = t;
- var uintL i2 = i1 + tmax;
+ var const uintC tmax = M>>1; // tmax = 2^(m-1)
+ for (var uintC t = 0; t < tmax; t++) {
+ var uintC i1 = t;
+ var uintC i2 = i1 + tmax;
// Butterfly: replace (Y(i1),Y(i2)) by
// (Y(i1) + Y(i2), Y(i1) - Y(i2)).
assign(R,chlen, Y(i2), tmp);
}
}
for (l = m-2; l>=0; l--) {
- var const uintL smax = (uintL)1 << (m-1-l);
- var const uintL tmax = (uintL)1 << l;
- for (var uintL s = 0; s < smax; s++) {
+ var const uintC smax = (uintC)1 << (m-1-l);
+ var const uintC tmax = (uintC)1 << l;
+ for (var uintC s = 0; s < smax; s++) {
// w^exp = 2^(exp << (r+1-m)).
- var uintL exp = bit_reverse(m-1-l,s) << (r-(m-1-l));
- for (var uintL t = 0; t < tmax; t++) {
- var uintL i1 = (s << (l+1)) + t;
- var uintL i2 = i1 + tmax;
+ var uintC exp = bit_reverse(m-1-l,s) << (r-(m-1-l));
+ for (var uintC t = 0; t < tmax; t++) {
+ var uintC i1 = (s << (l+1)) + t;
+ var uintC i2 = i1 + tmax;
// Butterfly: replace (Y(i1),Y(i2)) by
// (Y(i1) + w^exp*Y(i2), Y(i1) - w^exp*Y(i2)).
shiftleftm(R,chlen, Y(i2),exp, tmp);
{
var uintL l;
for (l = 0; l < m-1; l++) {
- var const uintL smax = (uintL)1 << (m-1-l);
- var const uintL tmax = (uintL)1 << l;
+ var const uintC smax = (uintC)1 << (m-1-l);
+ var const uintC tmax = (uintC)1 << l;
/* s = 0, exp = 0 */ {
- for (var uintL t = 0; t < tmax; t++) {
- var uintL i1 = t;
- var uintL i2 = i1 + tmax;
+ for (var uintC t = 0; t < tmax; t++) {
+ var uintC i1 = t;
+ var uintC i2 = i1 + tmax;
// Inverse Butterfly: replace (Z(i1),Z(i2)) by
// ((Z(i1)+Z(i2))/2, (Z(i1)-Z(i2))/(2*w^exp)),
// with exp <-- 0.
shiftm(R,chlen, diff, Z(i2));
}
}
- for (var uintL s = 1; s < smax; s++) {
+ for (var uintC s = 1; s < smax; s++) {
// w^exp = 2^(exp << (r+1-m)).
- var uintL exp = bit_reverse(m-1-l,s) << (r-(m-1-l));
+ var uintC exp = bit_reverse(m-1-l,s) << (r-(m-1-l));
exp = R - exp; // negate exp (use w^-1 instead of w)
- for (var uintL t = 0; t < tmax; t++) {
- var uintL i1 = (s << (l+1)) + t;
- var uintL i2 = i1 + tmax;
+ for (var uintC t = 0; t < tmax; t++) {
+ var uintC i1 = (s << (l+1)) + t;
+ var uintC i2 = i1 + tmax;
// Inverse Butterfly: replace (Z(i1),Z(i2)) by
// ((Z(i1)+Z(i2))/2, (Z(i1)-Z(i2))/(2*w^exp)),
// with exp <-- (M/2 - exp).
}
}
/* l = m-1 */ {
- var const uintL tmax = M>>1; // tmax = 2^(m-1)
- for (var uintL t = 0; t < tmax; t++) {
- var uintL i1 = t;
- var uintL i2 = i1 + tmax;
+ var const uintC tmax = M>>1; // tmax = 2^(m-1)
+ for (var uintC t = 0; t < tmax; t++) {
+ var uintC i1 = t;
+ var uintC i2 = i1 + tmax;
// Inverse Butterfly: replace (Z(i1),Z(i2)) by
// ((Z(i1)+Z(i2))/2, (Z(i1)-Z(i2))/2).
addm(R,chlen, Z(i1),Z(i2), sum);
static bool okfor (uintL r, uintL m, uintC len1, uintC len2)
{
- var uintL R = (uintL)1 << r;
- var uintL M = (uintL)1 << m;
- var uintL k = floor(R-m,2*intDsize);
+ var uintC R = (uintC)1 << r;
+ var uintC M = (uintC)1 << m;
+ var uintC k = floor(R-m,2*intDsize);
return (ceiling(len1,k)+ceiling(len2,k) <= M+1);
}
-static uintL numpieces (uintL r, uintL m, uintC len1, uintC len2)
+static uintC numpieces (uintL r, uintL m, uintC len1, uintC len2)
{
- var uintL R = (uintL)1 << r;
- var uintL M = (uintL)1 << m;
- var uintL k = floor(R-m,2*intDsize);
- var uintL piecelen2 = (M+1-ceiling(len1,k))*k;
+ var uintC R = (uintC)1 << r;
+ var uintC M = (uintC)1 << m;
+ var uintC k = floor(R-m,2*intDsize);
+ var uintC piecelen2 = (M+1-ceiling(len1,k))*k;
#ifdef DEBUG_FFTM
- if ((sintL)piecelen2 <= 0)
+ if ((sintC)piecelen2 <= 0)
cl_abort();
#endif
return ceiling(len2,piecelen2);
// Called only with 6 <= len1 <= len2.
{
var uint32 n;
- integerlength32(len1-1, n=); // 2^(n-1) < len1 <= 2^n
+ integerlengthC(len1-1, n=); // 2^(n-1) < len1 <= 2^n
var uintL r;
var uintL m;
r = ceiling(log2_intDsize+1+n,2);
if (r < log2_intDsize+2)
r = log2_intDsize+2;
retry: {
- var uintL k = floor(((uintL)1 << r) - (r+1), 2*intDsize);
- var uintL M = 2*ceiling(len1,k)-1;
- integerlength32(M, m=);
+ var uintC k = floor(((uintC)1 << r) - (r+1), 2*intDsize);
+ var uintC M = 2*ceiling(len1,k)-1;
+ integerlengthC(M, m=);
if (m == 0)
m = 1;
if (m > r+1) {
if (!(sourceptr1 == sourceptr2 && len1 == len2)) // when squaring, keep one piece
r--;
} else {
- var uintL q1 = numpieces(r,m,len1,len2);
+ var uintC q1 = numpieces(r,m,len1,len2);
if (m <= r) {
- var uintL q2 = numpieces(r,m+1,len1,len2);
+ var uintC q2 = numpieces(r,m+1,len1,len2);
if (2*q2 <= q1)
m++;
} else {
- var uintL q2 = numpieces(r+1,m,len1,len2);
+ var uintC q2 = numpieces(r+1,m,len1,len2);
if (3*q2 <= q1)
r++;
}
}
- var uintL R = (uintL)1 << r;
- var uintL M = (uintL)1 << m;
- var uintL k = floor(R-m,2*intDsize);
- var uintL piecelen2 = (M+1-ceiling(len1,k))*k;
+ var uintC R = (uintC)1 << r;
+ var uintC M = (uintC)1 << m;
+ var uintC k = floor(R-m,2*intDsize);
+ var uintC piecelen2 = (M+1-ceiling(len1,k))*k;
if (piecelen2 >= len2) {
// One piece only.
mulu_fftm(r,R, m,M, k, sourceptr1,len1, sourceptr2,len2, destptr);
CL_ALLOCA_STACK;
var uintD* tmpptr;
num_stack_alloc(len1+piecelen2,,tmpptr=);
- var uintL destlen = len1+len2;
+ var uintC destlen = len1+len2;
clear_loop_lsp(destptr,destlen);
do {
- var uintL len2p; // length of a piece of source2
+ var uintC len2p; // length of a piece of source2
len2p = piecelen2;
if (len2p > len2)
len2p = len2;
// len2p = min(piecelen2,len2).
- var uintL destlenp = len1 + len2p;
+ var uintC destlenp = len1 + len2p;
// destlenp = min(len1+piecelen2,destlen).
// Use tmpptr[-destlenp..-1].
if (len2p == 1) {
#ifndef _BIT_REVERSE
#define _BIT_REVERSE
// Reverse an n-bit number x. n>0.
-static uintL bit_reverse (uintL n, uintL x)
+static uintC bit_reverse (uintL n, uintC x)
{
- var uintL y = 0;
+ var uintC y = 0;
do {
y <<= 1;
y |= (x & 1);
#endif
// Compute an convolution mod p using FFT: z[0..N-1] := x[0..N-1] * y[0..N-1].
-static void fftp_convolution (const uintL n, const uintL N, // N = 2^n
+static void fftp_convolution (const uintL n, const uintC N, // N = 2^n
fftp_word * x, // N words
fftp_word * y, // N words
fftp_word * z // N words result
#else
var fftp_word* const w = cl_alloc_array(fftp_word,(N>>1)+1);
#endif
- var uintL i;
+ var uintC i;
// Initialize w[i] to w^i, w a primitive N-th root of unity.
w[0] = fftp_roots_of_1[0];
w[1] = fftp_roots_of_1[n];
{
var sintL l;
/* l = n-1 */ {
- var const uintL tmax = N>>1; // tmax = 2^(n-1)
- for (var uintL t = 0; t < tmax; t++) {
- var uintL i1 = t;
- var uintL i2 = i1 + tmax;
+ var const uintC tmax = N>>1; // tmax = 2^(n-1)
+ for (var uintC t = 0; t < tmax; t++) {
+ var uintC i1 = t;
+ var uintC i2 = i1 + tmax;
// Butterfly: replace (x(i1),x(i2)) by
// (x(i1) + x(i2), x(i1) - x(i2)).
var fftp_word tmp;
}
}
for (l = n-2; l>=0; l--) {
- var const uintL smax = (uintL)1 << (n-1-l);
- var const uintL tmax = (uintL)1 << l;
- for (var uintL s = 0; s < smax; s++) {
- var uintL exp = bit_reverse(n-1-l,s) << l;
- for (var uintL t = 0; t < tmax; t++) {
- var uintL i1 = (s << (l+1)) + t;
- var uintL i2 = i1 + tmax;
+ var const uintC smax = (uintC)1 << (n-1-l);
+ var const uintC tmax = (uintC)1 << l;
+ for (var uintC s = 0; s < smax; s++) {
+ var uintC exp = bit_reverse(n-1-l,s) << l;
+ for (var uintC t = 0; t < tmax; t++) {
+ var uintC i1 = (s << (l+1)) + t;
+ var uintC i2 = i1 + tmax;
// Butterfly: replace (x(i1),x(i2)) by
// (x(i1) + w^exp*x(i2), x(i1) - w^exp*x(i2)).
var fftp_word tmp;
if (!squaring) {
var sintL l;
/* l = n-1 */ {
- var uintL const tmax = N>>1; // tmax = 2^(n-1)
- for (var uintL t = 0; t < tmax; t++) {
- var uintL i1 = t;
- var uintL i2 = i1 + tmax;
+ var uintC const tmax = N>>1; // tmax = 2^(n-1)
+ for (var uintC t = 0; t < tmax; t++) {
+ var uintC i1 = t;
+ var uintC i2 = i1 + tmax;
// Butterfly: replace (y(i1),y(i2)) by
// (y(i1) + y(i2), y(i1) - y(i2)).
var fftp_word tmp;
}
}
for (l = n-2; l>=0; l--) {
- var const uintL smax = (uintL)1 << (n-1-l);
- var const uintL tmax = (uintL)1 << l;
- for (var uintL s = 0; s < smax; s++) {
- var uintL exp = bit_reverse(n-1-l,s) << l;
- for (var uintL t = 0; t < tmax; t++) {
- var uintL i1 = (s << (l+1)) + t;
- var uintL i2 = i1 + tmax;
+ var const uintC smax = (uintC)1 << (n-1-l);
+ var const uintC tmax = (uintC)1 << l;
+ for (var uintC s = 0; s < smax; s++) {
+ var uintC exp = bit_reverse(n-1-l,s) << l;
+ for (var uintC t = 0; t < tmax; t++) {
+ var uintC i1 = (s << (l+1)) + t;
+ var uintC i2 = i1 + tmax;
// Butterfly: replace (y(i1),y(i2)) by
// (y(i1) + w^exp*y(i2), y(i1) - w^exp*y(i2)).
var fftp_word tmp;
{
var uintL l;
for (l = 0; l < n-1; l++) {
- var const uintL smax = (uintL)1 << (n-1-l);
- var const uintL tmax = (uintL)1 << l;
+ var const uintC smax = (uintC)1 << (n-1-l);
+ var const uintC tmax = (uintC)1 << l;
#if FFTP_BACKWARD != CLEVER
- for (var uintL s = 0; s < smax; s++) {
- var uintL exp = bit_reverse(n-1-l,s) << l;
+ for (var uintC s = 0; s < smax; s++) {
+ var uintC exp = bit_reverse(n-1-l,s) << l;
#if FFTP_BACKWARD == RECIPROOT
if (exp > 0)
exp = N - exp; // negate exp (use w^-1 instead of w)
#endif
- for (var uintL t = 0; t < tmax; t++) {
- var uintL i1 = (s << (l+1)) + t;
- var uintL i2 = i1 + tmax;
+ for (var uintC t = 0; t < tmax; t++) {
+ var uintC i1 = (s << (l+1)) + t;
+ var uintC i2 = i1 + tmax;
// Inverse Butterfly: replace (z(i1),z(i2)) by
// ((z(i1)+z(i2))/2, (z(i1)-z(i2))/(2*w^exp)).
var fftp_word sum;
}
#else // FFTP_BACKWARD == CLEVER: clever handling of negative exponents
/* s = 0, exp = 0 */ {
- for (var uintL t = 0; t < tmax; t++) {
- var uintL i1 = t;
- var uintL i2 = i1 + tmax;
+ for (var uintC t = 0; t < tmax; t++) {
+ var uintC i1 = t;
+ var uintC i2 = i1 + tmax;
// Inverse Butterfly: replace (z(i1),z(i2)) by
// ((z(i1)+z(i2))/2, (z(i1)-z(i2))/(2*w^exp)),
// with exp <-- 0.
shiftp(diff, z[i2]);
}
}
- for (var uintL s = 1; s < smax; s++) {
- var uintL exp = bit_reverse(n-1-l,s) << l;
+ for (var uintC s = 1; s < smax; s++) {
+ var uintC exp = bit_reverse(n-1-l,s) << l;
exp = (N>>1) - exp; // negate exp (use w^-1 instead of w)
- for (var uintL t = 0; t < tmax; t++) {
- var uintL i1 = (s << (l+1)) + t;
- var uintL i2 = i1 + tmax;
+ for (var uintC t = 0; t < tmax; t++) {
+ var uintC i1 = (s << (l+1)) + t;
+ var uintC i2 = i1 + tmax;
// Inverse Butterfly: replace (z(i1),z(i2)) by
// ((z(i1)+z(i2))/2, (z(i1)-z(i2))/(2*w^exp)),
// with exp <-- (N/2 - exp).
#endif
}
/* l = n-1 */ {
- var const uintL tmax = N>>1; // tmax = 2^(n-1)
- for (var uintL t = 0; t < tmax; t++) {
- var uintL i1 = t;
- var uintL i2 = i1 + tmax;
+ var const uintC tmax = N>>1; // tmax = 2^(n-1)
+ for (var uintC t = 0; t < tmax; t++) {
+ var uintC i1 = t;
+ var uintC i2 = i1 + tmax;
// Inverse Butterfly: replace (z(i1),z(i2)) by
// ((z(i1)+z(i2))/2, (z(i1)-z(i2))/2).
var fftp_word sum;
// sum(i=0..N-1, x_i T^i), sum(i=0..N-1, y_i T^i)
// multipliziert, und zwar durch Fourier-Transformation (s.o.).
var uint32 n;
- integerlength32(len1-1, n=); // 2^(n-1) < len1 <= 2^n
- var uintL len = (uintL)1 << n; // kleinste Zweierpotenz >= len1
+ integerlengthC(len1-1, n=); // 2^(n-1) < len1 <= 2^n
+ var uintC len = (uintC)1 << n; // kleinste Zweierpotenz >= len1
// Wählt man N = len, so hat man ceiling(len2/(len-len1+1)) * FFT(len).
// Wählt man N = 2*len, so hat man ceiling(len2/(2*len-len1+1)) * FFT(2*len).
// Wir wählen das billigere von beiden:
n = n+1;
len = len << 1;
}
- var const uintL N = len; // N = 2^n
+ var const uintC N = len; // N = 2^n
CL_ALLOCA_STACK;
var fftp_word* const x = cl_alloc_array(fftp_word,N);
var fftp_word* const y = cl_alloc_array(fftp_word,N);
#endif
var uintD* const tmpprod = cl_alloc_array(uintD,len1+1);
var uintP i;
- var uintL destlen = len1+len2;
+ var uintC destlen = len1+len2;
clear_loop_lsp(destptr,destlen);
do {
- var uintL len2p; // length of a piece of source2
+ var uintC len2p; // length of a piece of source2
len2p = N - len1 + 1;
if (len2p > len2)
len2p = len2;
if (inc_loop_lsp(destptr lspop (len1+1),destlen-(len1+1)))
cl_abort();
} else {
- var uintL destlenp = len1 + len2p - 1;
+ var uintC destlenp = len1 + len2p - 1;
// destlenp = min(N,destlen-1).
var bool squaring = ((sourceptr1 == sourceptr2) && (len1 == len2p));
// Fill factor x.
#ifndef _BIT_REVERSE
#define _BIT_REVERSE
// Reverse an n-bit number x. n>0.
-static uintL bit_reverse (uintL n, uintL x)
+static uintC bit_reverse (uintL n, uintC x)
{
- var uintL y = 0;
+ var uintC y = 0;
do {
y <<= 1;
y |= (x & 1);
#endif
// Compute an convolution mod p using FFT: z[0..N-1] := x[0..N-1] * y[0..N-1].
-static void fftp3_convolution (const uintL n, const uintL N, // N = 2^n
+static void fftp3_convolution (const uintL n, const uintC N, // N = 2^n
fftp3_word * x, // N words
fftp3_word * y, // N words
fftp3_word * z // N words result
#else
var fftp3_word* const w = cl_alloc_array(fftp3_word,(N>>1)+1);
#endif
- var uintL i;
+ var uintC i;
// Initialize w[i] to w^i, w a primitive N-th root of unity.
w[0] = fftp3_roots_of_1[0];
w[1] = fftp3_roots_of_1[n];
{
var sintL l;
/* l = n-1 */ {
- var const uintL tmax = N>>1; // tmax = 2^(n-1)
- for (var uintL t = 0; t < tmax; t++) {
- var uintL i1 = t;
- var uintL i2 = i1 + tmax;
+ var const uintC tmax = N>>1; // tmax = 2^(n-1)
+ for (var uintC t = 0; t < tmax; t++) {
+ var uintC i1 = t;
+ var uintC i2 = i1 + tmax;
// Butterfly: replace (x(i1),x(i2)) by
// (x(i1) + x(i2), x(i1) - x(i2)).
var fftp3_word tmp;
}
}
for (l = n-2; l>=0; l--) {
- var const uintL smax = (uintL)1 << (n-1-l);
- var const uintL tmax = (uintL)1 << l;
- for (var uintL s = 0; s < smax; s++) {
- var uintL exp = bit_reverse(n-1-l,s) << l;
- for (var uintL t = 0; t < tmax; t++) {
- var uintL i1 = (s << (l+1)) + t;
- var uintL i2 = i1 + tmax;
+ var const uintC smax = (uintC)1 << (n-1-l);
+ var const uintC tmax = (uintC)1 << l;
+ for (var uintC s = 0; s < smax; s++) {
+ var uintC exp = bit_reverse(n-1-l,s) << l;
+ for (var uintC t = 0; t < tmax; t++) {
+ var uintC i1 = (s << (l+1)) + t;
+ var uintC i2 = i1 + tmax;
// Butterfly: replace (x(i1),x(i2)) by
// (x(i1) + w^exp*x(i2), x(i1) - w^exp*x(i2)).
var fftp3_word tmp;
if (!squaring) {
var sintL l;
/* l = n-1 */ {
- var uintL const tmax = N>>1; // tmax = 2^(n-1)
- for (var uintL t = 0; t < tmax; t++) {
- var uintL i1 = t;
- var uintL i2 = i1 + tmax;
+ var uintC const tmax = N>>1; // tmax = 2^(n-1)
+ for (var uintC t = 0; t < tmax; t++) {
+ var uintC i1 = t;
+ var uintC i2 = i1 + tmax;
// Butterfly: replace (y(i1),y(i2)) by
// (y(i1) + y(i2), y(i1) - y(i2)).
var fftp3_word tmp;
}
}
for (l = n-2; l>=0; l--) {
- var const uintL smax = (uintL)1 << (n-1-l);
- var const uintL tmax = (uintL)1 << l;
- for (var uintL s = 0; s < smax; s++) {
- var uintL exp = bit_reverse(n-1-l,s) << l;
- for (var uintL t = 0; t < tmax; t++) {
- var uintL i1 = (s << (l+1)) + t;
- var uintL i2 = i1 + tmax;
+ var const uintC smax = (uintC)1 << (n-1-l);
+ var const uintC tmax = (uintC)1 << l;
+ for (var uintC s = 0; s < smax; s++) {
+ var uintC exp = bit_reverse(n-1-l,s) << l;
+ for (var uintC t = 0; t < tmax; t++) {
+ var uintC i1 = (s << (l+1)) + t;
+ var uintC i2 = i1 + tmax;
// Butterfly: replace (y(i1),y(i2)) by
// (y(i1) + w^exp*y(i2), y(i1) - w^exp*y(i2)).
var fftp3_word tmp;
{
var uintL l;
for (l = 0; l < n-1; l++) {
- var const uintL smax = (uintL)1 << (n-1-l);
- var const uintL tmax = (uintL)1 << l;
+ var const uintC smax = (uintC)1 << (n-1-l);
+ var const uintC tmax = (uintC)1 << l;
#if FFTP3_BACKWARD != CLEVER
- for (var uintL s = 0; s < smax; s++) {
- var uintL exp = bit_reverse(n-1-l,s) << l;
+ for (var uintC s = 0; s < smax; s++) {
+ var uintC exp = bit_reverse(n-1-l,s) << l;
#if FFTP3_BACKWARD == RECIPROOT
if (exp > 0)
exp = N - exp; // negate exp (use w^-1 instead of w)
#endif
- for (var uintL t = 0; t < tmax; t++) {
- var uintL i1 = (s << (l+1)) + t;
- var uintL i2 = i1 + tmax;
+ for (var uintC t = 0; t < tmax; t++) {
+ var uintC i1 = (s << (l+1)) + t;
+ var uintC i2 = i1 + tmax;
// Inverse Butterfly: replace (z(i1),z(i2)) by
// ((z(i1)+z(i2))/2, (z(i1)-z(i2))/(2*w^exp)).
var fftp3_word sum;
}
#else // FFTP3_BACKWARD == CLEVER: clever handling of negative exponents
/* s = 0, exp = 0 */ {
- for (var uintL t = 0; t < tmax; t++) {
- var uintL i1 = t;
- var uintL i2 = i1 + tmax;
+ for (var uintC t = 0; t < tmax; t++) {
+ var uintC i1 = t;
+ var uintC i2 = i1 + tmax;
// Inverse Butterfly: replace (z(i1),z(i2)) by
// ((z(i1)+z(i2))/2, (z(i1)-z(i2))/(2*w^exp)),
// with exp <-- 0.
shiftp3(diff, z[i2]);
}
}
- for (var uintL s = 1; s < smax; s++) {
- var uintL exp = bit_reverse(n-1-l,s) << l;
+ for (var uintC s = 1; s < smax; s++) {
+ var uintC exp = bit_reverse(n-1-l,s) << l;
exp = (N>>1) - exp; // negate exp (use w^-1 instead of w)
- for (var uintL t = 0; t < tmax; t++) {
- var uintL i1 = (s << (l+1)) + t;
- var uintL i2 = i1 + tmax;
+ for (var uintC t = 0; t < tmax; t++) {
+ var uintC i1 = (s << (l+1)) + t;
+ var uintC i2 = i1 + tmax;
// Inverse Butterfly: replace (z(i1),z(i2)) by
// ((z(i1)+z(i2))/2, (z(i1)-z(i2))/(2*w^exp)),
// with exp <-- (N/2 - exp).
#endif
}
/* l = n-1 */ {
- var const uintL tmax = N>>1; // tmax = 2^(n-1)
- for (var uintL t = 0; t < tmax; t++) {
- var uintL i1 = t;
- var uintL i2 = i1 + tmax;
+ var const uintC tmax = N>>1; // tmax = 2^(n-1)
+ for (var uintC t = 0; t < tmax; t++) {
+ var uintC i1 = t;
+ var uintC i2 = i1 + tmax;
// Inverse Butterfly: replace (z(i1),z(i2)) by
// ((z(i1)+z(i2))/2, (z(i1)-z(i2))/2).
var fftp3_word sum;
// sum(i=0..N-1, x_i T^i), sum(i=0..N-1, y_i T^i)
// multipliziert, und zwar durch Fourier-Transformation (s.o.).
var uint32 n;
- integerlength32(len1-1, n=); // 2^(n-1) < len1 <= 2^n
- var uintL len = (uintL)1 << n; // kleinste Zweierpotenz >= len1
+ integerlengthC(len1-1, n=); // 2^(n-1) < len1 <= 2^n
+ var uintC len = (uintC)1 << n; // kleinste Zweierpotenz >= len1
// Wählt man N = len, so hat man ceiling(len2/(len-len1+1)) * FFT(len).
// Wählt man N = 2*len, so hat man ceiling(len2/(2*len-len1+1)) * FFT(2*len).
// Wir wählen das billigere von beiden:
n = n+1;
len = len << 1;
}
- var const uintL N = len; // N = 2^n
+ var const uintC N = len; // N = 2^n
CL_ALLOCA_STACK;
var fftp3_word* const x = cl_alloc_array(fftp3_word,N);
var fftp3_word* const y = cl_alloc_array(fftp3_word,N);
#endif
var uintD* const tmpprod = cl_alloc_array(uintD,len1+1);
var uintP i;
- var uintL destlen = len1+len2;
+ var uintC destlen = len1+len2;
clear_loop_lsp(destptr,destlen);
do {
- var uintL len2p; // length of a piece of source2
+ var uintC len2p; // length of a piece of source2
len2p = N - len1 + 1;
if (len2p > len2)
len2p = len2;
if (inc_loop_lsp(destptr lspop (len1+1),destlen-(len1+1)))
cl_abort();
} else {
- var uintL destlenp = len1 + len2p - 1;
+ var uintC destlenp = len1 + len2p - 1;
// destlenp = min(N,destlen-1).
var bool squaring = ((sourceptr1 == sourceptr2) && (len1 == len2p));
// Fill factor x.
#ifndef _BIT_REVERSE
#define _BIT_REVERSE
// Reverse an n-bit number x. n>0.
-static uintL bit_reverse (uintL n, uintL x)
+static uintC bit_reverse (uintL n, uintC x)
{
- var uintL y = 0;
+ var uintC y = 0;
do {
y <<= 1;
y |= (x & 1);
#endif
// Compute an convolution mod p using FFT: z[0..N-1] := x[0..N-1] * y[0..N-1].
-static void fftp3m_convolution (const uintL n, const uintL N, // N = 2^n
+static void fftp3m_convolution (const uintL n, const uintC N, // N = 2^n
fftp3m_word * x, // N words
fftp3m_word * y, // N words
fftp3m_word * z // N words result
#else
var fftp3m_word* const w = cl_alloc_array(fftp3m_word,(N>>1)+1);
#endif
- var uintL i;
+ var uintC i;
// Initialize w[i] to w^i, w a primitive N-th root of unity.
w[0] = fftp3m_roots_of_1[0];
w[1] = fftp3m_roots_of_1[n];
{
var sintL l;
/* l = n-1 */ {
- var const uintL tmax = N>>1; // tmax = 2^(n-1)
- for (var uintL t = 0; t < tmax; t++) {
- var uintL i1 = t;
- var uintL i2 = i1 + tmax;
+ var const uintC tmax = N>>1; // tmax = 2^(n-1)
+ for (var uintC t = 0; t < tmax; t++) {
+ var uintC i1 = t;
+ var uintC i2 = i1 + tmax;
// Butterfly: replace (x(i1),x(i2)) by
// (x(i1) + x(i2), x(i1) - x(i2)).
var fftp3m_word tmp;
}
}
for (l = n-2; l>=0; l--) {
- var const uintL smax = (uintL)1 << (n-1-l);
- var const uintL tmax = (uintL)1 << l;
- for (var uintL s = 0; s < smax; s++) {
- var uintL exp = bit_reverse(n-1-l,s) << l;
- for (var uintL t = 0; t < tmax; t++) {
- var uintL i1 = (s << (l+1)) + t;
- var uintL i2 = i1 + tmax;
+ var const uintC smax = (uintC)1 << (n-1-l);
+ var const uintC tmax = (uintC)1 << l;
+ for (var uintC s = 0; s < smax; s++) {
+ var uintC exp = bit_reverse(n-1-l,s) << l;
+ for (var uintC t = 0; t < tmax; t++) {
+ var uintC i1 = (s << (l+1)) + t;
+ var uintC i2 = i1 + tmax;
// Butterfly: replace (x(i1),x(i2)) by
// (x(i1) + w^exp*x(i2), x(i1) - w^exp*x(i2)).
var fftp3m_word tmp;
if (!squaring) {
var sintL l;
/* l = n-1 */ {
- var uintL const tmax = N>>1; // tmax = 2^(n-1)
- for (var uintL t = 0; t < tmax; t++) {
- var uintL i1 = t;
- var uintL i2 = i1 + tmax;
+ var uintC const tmax = N>>1; // tmax = 2^(n-1)
+ for (var uintC t = 0; t < tmax; t++) {
+ var uintC i1 = t;
+ var uintC i2 = i1 + tmax;
// Butterfly: replace (y(i1),y(i2)) by
// (y(i1) + y(i2), y(i1) - y(i2)).
var fftp3m_word tmp;
}
}
for (l = n-2; l>=0; l--) {
- var const uintL smax = (uintL)1 << (n-1-l);
- var const uintL tmax = (uintL)1 << l;
- for (var uintL s = 0; s < smax; s++) {
- var uintL exp = bit_reverse(n-1-l,s) << l;
- for (var uintL t = 0; t < tmax; t++) {
- var uintL i1 = (s << (l+1)) + t;
- var uintL i2 = i1 + tmax;
+ var const uintC smax = (uintC)1 << (n-1-l);
+ var const uintC tmax = (uintC)1 << l;
+ for (var uintC s = 0; s < smax; s++) {
+ var uintC exp = bit_reverse(n-1-l,s) << l;
+ for (var uintC t = 0; t < tmax; t++) {
+ var uintC i1 = (s << (l+1)) + t;
+ var uintC i2 = i1 + tmax;
// Butterfly: replace (y(i1),y(i2)) by
// (y(i1) + w^exp*y(i2), y(i1) - w^exp*y(i2)).
var fftp3m_word tmp;
{
var uintL l;
for (l = 0; l < n-1; l++) {
- var const uintL smax = (uintL)1 << (n-1-l);
- var const uintL tmax = (uintL)1 << l;
+ var const uintC smax = (uintC)1 << (n-1-l);
+ var const uintC tmax = (uintC)1 << l;
#if FFTP3M_BACKWARD != CLEVER
- for (var uintL s = 0; s < smax; s++) {
- var uintL exp = bit_reverse(n-1-l,s) << l;
+ for (var uintC s = 0; s < smax; s++) {
+ var uintC exp = bit_reverse(n-1-l,s) << l;
#if FFTP3M_BACKWARD == RECIPROOT
if (exp > 0)
exp = N - exp; // negate exp (use w^-1 instead of w)
#endif
- for (var uintL t = 0; t < tmax; t++) {
- var uintL i1 = (s << (l+1)) + t;
- var uintL i2 = i1 + tmax;
+ for (var uintC t = 0; t < tmax; t++) {
+ var uintC i1 = (s << (l+1)) + t;
+ var uintC i2 = i1 + tmax;
// Inverse Butterfly: replace (z(i1),z(i2)) by
// ((z(i1)+z(i2))/2, (z(i1)-z(i2))/(2*w^exp)).
var fftp3m_word sum;
}
#else // FFTP3M_BACKWARD == CLEVER: clever handling of negative exponents
/* s = 0, exp = 0 */ {
- for (var uintL t = 0; t < tmax; t++) {
- var uintL i1 = t;
- var uintL i2 = i1 + tmax;
+ for (var uintC t = 0; t < tmax; t++) {
+ var uintC i1 = t;
+ var uintC i2 = i1 + tmax;
// Inverse Butterfly: replace (z(i1),z(i2)) by
// ((z(i1)+z(i2))/2, (z(i1)-z(i2))/(2*w^exp)),
// with exp <-- 0.
shiftp3m(diff, z[i2]);
}
}
- for (var uintL s = 1; s < smax; s++) {
- var uintL exp = bit_reverse(n-1-l,s) << l;
+ for (var uintC s = 1; s < smax; s++) {
+ var uintC exp = bit_reverse(n-1-l,s) << l;
exp = (N>>1) - exp; // negate exp (use w^-1 instead of w)
- for (var uintL t = 0; t < tmax; t++) {
- var uintL i1 = (s << (l+1)) + t;
- var uintL i2 = i1 + tmax;
+ for (var uintC t = 0; t < tmax; t++) {
+ var uintC i1 = (s << (l+1)) + t;
+ var uintC i2 = i1 + tmax;
// Inverse Butterfly: replace (z(i1),z(i2)) by
// ((z(i1)+z(i2))/2, (z(i1)-z(i2))/(2*w^exp)),
// with exp <-- (N/2 - exp).
#endif
}
/* l = n-1 */ {
- var const uintL tmax = N>>1; // tmax = 2^(n-1)
- for (var uintL t = 0; t < tmax; t++) {
- var uintL i1 = t;
- var uintL i2 = i1 + tmax;
+ var const uintC tmax = N>>1; // tmax = 2^(n-1)
+ for (var uintC t = 0; t < tmax; t++) {
+ var uintC i1 = t;
+ var uintC i2 = i1 + tmax;
// Inverse Butterfly: replace (z(i1),z(i2)) by
// ((z(i1)+z(i2))/2, (z(i1)-z(i2))/2).
var fftp3m_word sum;
// sum(i=0..N-1, x_i T^i), sum(i=0..N-1, y_i T^i)
// multipliziert, und zwar durch Fourier-Transformation (s.o.).
var uint32 n;
- integerlength32(len1-1, n=); // 2^(n-1) < len1 <= 2^n
- var uintL len = (uintL)1 << n; // kleinste Zweierpotenz >= len1
+ integerlengthC(len1-1, n=); // 2^(n-1) < len1 <= 2^n
+ var uintC len = (uintC)1 << n; // kleinste Zweierpotenz >= len1
// Wählt man N = len, so hat man ceiling(len2/(len-len1+1)) * FFT(len).
// Wählt man N = 2*len, so hat man ceiling(len2/(2*len-len1+1)) * FFT(2*len).
// Wir wählen das billigere von beiden:
n = n+1;
len = len << 1;
}
- var const uintL N = len; // N = 2^n
+ var const uintC N = len; // N = 2^n
CL_ALLOCA_STACK;
var fftp3m_word* const x = cl_alloc_array(fftp3m_word,N);
var fftp3m_word* const y = cl_alloc_array(fftp3m_word,N);
#endif
var uintD* const tmpprod = cl_alloc_array(uintD,len1+1);
var uintP i;
- var uintL destlen = len1+len2;
+ var uintC destlen = len1+len2;
clear_loop_lsp(destptr,destlen);
do {
- var uintL len2p; // length of a piece of source2
+ var uintC len2p; // length of a piece of source2
len2p = N - len1 + 1;
if (len2p > len2)
len2p = len2;
if (inc_loop_lsp(destptr lspop (len1+1),destlen-(len1+1)))
cl_abort();
} else {
- var uintL destlenp = len1 + len2p - 1;
+ var uintC destlenp = len1 + len2p - 1;
// destlenp = min(N,destlen-1).
var bool squaring = ((sourceptr1 == sourceptr2) && (len1 == len2p));
// Fill factor x.
r.im = r_im;
}
-static uintL shuffle (uintL n, uintL x)
+static uintC shuffle (uintL n, uintC x)
{
- var uintL y = 0;
- var sintL v = 1;
+ var uintC y = 0;
+ var sintC v = 1;
// Invariant: y + v*shuffle(n,x).
do {
if (x & 1)
}
#if 0 // unused
-static uintL invshuffle (uintL n, uintL x)
+static uintC invshuffle (uintL n, uintC x)
{
- var uintL y = 0;
- var uintL v = 1;
+ var uintC y = 0;
+ var uintC v = 1;
// Invariant: y + v*invshuffle(n,x).
do {
- if (x == ((uintL)1 << (n-1)))
+ if (x == ((uintC)1 << (n-1)))
return y + v;
- else if (x > ((uintL)1 << (n-1))) {
- x = ((uintL)1 << n) - x;
+ else if (x > ((uintC)1 << (n-1))) {
+ x = ((uintC)1 << n) - x;
y = y+v;
}
v <<= 1;
#endif
// Compute a real convolution using FFT: z[0..N-1] := x[0..N-1] * y[0..N-1].
-static void fftr_convolution (const uintL n, const uintL N, // N = 2^n
+static void fftr_convolution (const uintL n, const uintC N, // N = 2^n
fftr_real * x, // N numbers
fftr_real * y, // N numbers
fftr_real * z // N numbers result
{
CL_ALLOCA_STACK;
var fftr_cosinus* const w = cl_alloc_array(fftr_cosinus,N>>2);
- var uintL i;
+ var uintC i;
// Initialize w[exp].c to exp(2 pi i exp/N).
w[0].c = fftr_roots_of_1[0];
{
var uintL l;
for (l = n-1; l > 0; l--) {
/* s = 0 */ {
- var const uintL tmax = (uintL)1 << l;
- for (var uintL t = 0; t < tmax; t++) {
- var uintL i1 = t;
- var uintL i2 = i1 + tmax;
+ var const uintC tmax = (uintC)1 << l;
+ for (var uintC t = 0; t < tmax; t++) {
+ var uintC i1 = t;
+ var uintC i2 = i1 + tmax;
// replace (x[i1],x[i2]) by
// (x[i1] + x[i2], x[i1] - x[i2])
var fftr_real tmp;
x[i1] = x[i1] + tmp;
}
}
- var const uintL smax = (uintL)1 << (n-1-l);
- var const uintL tmax = (uintL)1 << (l-1);
- for (var uintL s = 1; s < smax; s++) {
- var uintL exp = shuffle(n-1-l,s) << (l-1);
- for (var uintL t = 0; t < tmax; t++) {
- var uintL i1 = (s << (l+1)) + t;
- var uintL i2 = i1 + tmax;
- var uintL i3 = i2 + tmax;
- var uintL i4 = i3 + tmax;
+ var const uintC smax = (uintC)1 << (n-1-l);
+ var const uintC tmax = (uintC)1 << (l-1);
+ for (var uintC s = 1; s < smax; s++) {
+ var uintC exp = shuffle(n-1-l,s) << (l-1);
+ for (var uintC t = 0; t < tmax; t++) {
+ var uintC i1 = (s << (l+1)) + t;
+ var uintC i2 = i1 + tmax;
+ var uintC i3 = i2 + tmax;
+ var uintC i4 = i3 + tmax;
// replace (x[i1],x[i2],x[i3],x[i4]) by
// (x[i1]-x[i3] - x[i4]*gam,
// x[i3]*gam + x[i2]+x[i4]*(gam^2-1),
var uintL l;
for (l = n-1; l > 0; l--) {
/* s = 0 */ {
- var const uintL tmax = (uintL)1 << l;
- for (var uintL t = 0; t < tmax; t++) {
- var uintL i1 = t;
- var uintL i2 = i1 + tmax;
+ var const uintC tmax = (uintC)1 << l;
+ for (var uintC t = 0; t < tmax; t++) {
+ var uintC i1 = t;
+ var uintC i2 = i1 + tmax;
// replace (y[i1],y[i2]) by
// (y[i1] + y[i2], y[i1] - y[i2])
var fftr_real tmp;
y[i1] = y[i1] + tmp;
}
}
- var const uintL smax = (uintL)1 << (n-1-l);
- var const uintL tmax = (uintL)1 << (l-1);
- for (var uintL s = 1; s < smax; s++) {
- var uintL exp = shuffle(n-1-l,s) << (l-1);
- for (var uintL t = 0; t < tmax; t++) {
- var uintL i1 = (s << (l+1)) + t;
- var uintL i2 = i1 + tmax;
- var uintL i3 = i2 + tmax;
- var uintL i4 = i3 + tmax;
+ var const uintC smax = (uintC)1 << (n-1-l);
+ var const uintC tmax = (uintC)1 << (l-1);
+ for (var uintC s = 1; s < smax; s++) {
+ var uintC exp = shuffle(n-1-l,s) << (l-1);
+ for (var uintC t = 0; t < tmax; t++) {
+ var uintC i1 = (s << (l+1)) + t;
+ var uintC i2 = i1 + tmax;
+ var uintC i3 = i2 + tmax;
+ var uintC i4 = i3 + tmax;
// replace (y[i1],y[i2],y[i3],y[i4]) by
// (y[i1]-y[i3] - y[i4]*gam,
// y[i3]*gam + y[i2]+y[i4]*(gam^2-1),
// Multiplication mod (z+1).
z[1] = x[1] * y[1];
#if 0 // This needs w[1..2^(n-1)-1].
- var const uintL smax = (uintL)1 << (n-1);
- for (var uintL s = 1; s < smax; s++) {
+ var const uintC smax = (uintC)1 << (n-1);
+ for (var uintC s = 1; s < smax; s++) {
// Multiplication mod (z^2 - 2 cos(2 pi j/2^n) z + 1)
// with j = shuffle(n-1,s).
- var uintL exp = shuffle(n-1,s);
+ var uintC exp = shuffle(n-1,s);
var fftr_real tmp0 = x[2*s] * y[2*s];
var fftr_real tmp1 = x[2*s] * y[2*s+1] + x[2*s+1] * y[2*s];
var fftr_real tmp2 = x[2*s+1] * y[2*s+1];
z[2] = tmp0 - tmp2;
z[3] = tmp1;
}
- var const uintL smax = (uintL)1 << (n-2);
- for (var uintL s = 1; s < smax; s++) {
- var uintL exp = shuffle(n-2,s);
+ var const uintC smax = (uintC)1 << (n-2);
+ for (var uintC s = 1; s < smax; s++) {
+ var uintC exp = shuffle(n-2,s);
// Multiplication mod (z^2 - 2 cos(2 pi j/2^n) z + 1)
// with j = shuffle(n-1,2*s) = shuffle(n-2,s).
var fftr_real gam = w[exp].gam.d;
}
for (l = 1; l < n; l++) {
/* s = 0 */ {
- var const uintL tmax = (uintL)1 << l;
- for (var uintL t = 0; t < tmax; t++) {
- var uintL i1 = t;
- var uintL i2 = i1 + tmax;
+ var const uintC tmax = (uintC)1 << l;
+ for (var uintC t = 0; t < tmax; t++) {
+ var uintC i1 = t;
+ var uintC i2 = i1 + tmax;
// replace (z[i1],z[i2]) by
// ((z[i1]+z[i2])/2, (z[i1]-z[i2])/2)
// Do the division by 2 later.
z[i1] = z[i1] + tmp;
}
}
- var const uintL smax = (uintL)1 << (n-1-l);
- var const uintL tmax = (uintL)1 << (l-1);
- for (var uintL s = 1; s < smax; s++) {
- var uintL exp = shuffle(n-1-l,s) << (l-1);
- for (var uintL t = 0; t < tmax; t++) {
- var uintL i1 = (s << (l+1)) + t;
- var uintL i2 = i1 + tmax;
- var uintL i3 = i2 + tmax;
- var uintL i4 = i3 + tmax;
+ var const uintC smax = (uintC)1 << (n-1-l);
+ var const uintC tmax = (uintC)1 << (l-1);
+ for (var uintC s = 1; s < smax; s++) {
+ var uintC exp = shuffle(n-1-l,s) << (l-1);
+ for (var uintC t = 0; t < tmax; t++) {
+ var uintC i1 = (s << (l+1)) + t;
+ var uintC i2 = i1 + tmax;
+ var uintC i3 = i2 + tmax;
+ var uintC i4 = i3 + tmax;
// replace (z[i1],z[i2],z[i3],z[i4]) by
// ((z[i1]+z[i3]+(z[i2]-z[i4])/gam)/2,
// (z[i2]+z[i4]-(z[i3]-z[i1])/gam*(gam^2-1))/2,
// Split len uintD's below sourceptr into chunks of l bits, thus filling
// N real numbers at x.
-static void fill_factor (uintL N, fftr_real* x, uintL l,
- const uintD* sourceptr, uintL len)
+static void fill_factor (uintC N, fftr_real* x, uintL l,
+ const uintD* sourceptr, uintC len)
{
- var uintL i;
+ var uintC i;
if (max_l(2) > intDsize && l > intDsize) {
// l > intDsize
if (max_l(2) > 64 && l > 64) {
// The z[i] are known to be approximately integers >= 0, < N*2^(2*l).
// Assumes room for floor(N*l/intDsize)+(1+ceiling((n+2*l)/intDsize)) uintD's
// below destptr. Fills len digits and returns (destptr lspop len).
-static uintD* unfill_product (uintL n, uintL N, // N = 2^n
+static uintD* unfill_product (uintL n, uintC N, // N = 2^n
const fftr_real * z, uintL l,
uintD* destptr)
{
- var uintL i;
+ var uintC i;
if (n + 2*l <= intDsize) {
// 2-digit carry is sufficient, l < intDsize
var uintD carry0 = 0;
// is 2*len1*intDsize/l_max - 1 <= 2^k.
{
var const int l = max_l(2);
- var uintL lhs = 2*ceiling((uintL)len1*intDsize,l) - 1; // >=1
+ var uintC lhs = 2*ceiling(len1*intDsize,l) - 1; // >=1
if (lhs < 3)
k = 2;
else
- integerlength32(lhs-1, k=); // k>=2
+ integerlengthC(lhs-1, k=); // k>=2
}
// Try whether this k is ok or whether we have to increase k.
for ( ; ; k++) {
fprint(std::cerr, "FFT problem: numbers too big, floating point precision not sufficient\n");
cl_abort();
}
- if (2*ceiling((uintL)len1*intDsize,max_l_table[k])-1 <= ((uintL)1 << k))
+ if (2*ceiling(len1*intDsize,max_l_table[k])-1 <= ((uintC)1 << k))
break;
}
// We could try to reduce l, keeping the same k. But why should we?
// Calculate the number of pieces in which source2 will have to be
// split. Each of the pieces must satisfy
// ceiling(len1*intDsize/l) + ceiling(len2*intDsize/l) - 1 <= 2^k,
- var uintL len2p;
+ var uintC len2p;
// Try once with k, once with k+1. Compare them.
{
- var uintL remaining_k = ((uintL)1 << k) + 1 - ceiling((uintL)len1*intDsize,max_l_table[k]);
- var uintL max_piecelen_k = floor(remaining_k*max_l_table[k],intDsize);
- var uintL numpieces_k = ceiling(len2,max_piecelen_k);
- var uintL remaining_k1 = ((uintL)1 << (k+1)) + 1 - ceiling((uintL)len1*intDsize,max_l_table[k+1]);
- var uintL max_piecelen_k1 = floor(remaining_k1*max_l_table[k+1],intDsize);
- var uintL numpieces_k1 = ceiling(len2,max_piecelen_k1);
+ var uintC remaining_k = ((uintC)1 << k) + 1 - ceiling(len1*intDsize,max_l_table[k]);
+ var uintC max_piecelen_k = floor(remaining_k*max_l_table[k],intDsize);
+ var uintC numpieces_k = ceiling(len2,max_piecelen_k);
+ var uintC remaining_k1 = ((uintC)1 << (k+1)) + 1 - ceiling(len1*intDsize,max_l_table[k+1]);
+ var uintC max_piecelen_k1 = floor(remaining_k1*max_l_table[k+1],intDsize);
+ var uintC numpieces_k1 = ceiling(len2,max_piecelen_k1);
if (numpieces_k <= 2*numpieces_k1) {
// keep k
len2p = max_piecelen_k;
}
var const uintL l = max_l_table[k];
var const uintL n = k;
- var const uintL N = (uintL)1 << n;
+ var const uintC N = (uintC)1 << n;
CL_ALLOCA_STACK;
var fftr_real* const x = cl_alloc_array(fftr_real,N);
var fftr_real* const y = cl_alloc_array(fftr_real,N);
var fftr_real* const z = x; // put z in place of x - saves memory
#endif
var uintD* const tmpprod1 = cl_alloc_array(uintD,len1+1);
- var uintL tmpprod_len = floor(l<<n,intDsize)+6;
+ var uintC tmpprod_len = floor(l<<n,intDsize)+6;
var uintD* const tmpprod = cl_alloc_array(uintD,tmpprod_len);
- var uintL destlen = len1+len2;
+ var uintC destlen = len1+len2;
clear_loop_lsp(destptr,destlen);
do {
if (len2p > len2)
{
var fftr_real re_lo_limit = (fftr_real)(-0.5);
var fftr_real re_hi_limit = (fftr_real)N * fftr_pow2_table[l] * fftr_pow2_table[l] + (fftr_real)0.5;
- for (var uintL i = 0; i < N; i++)
+ for (var uintC i = 0; i < N; i++)
if (!(z[i] > re_lo_limit
&& z[i] < re_hi_limit))
cl_abort();
#endif
var uintD* tmpLSDptr = arrayLSDptr(tmpprod,tmpprod_len);
var uintD* tmpMSDptr = unfill_product(n,N,z,l,tmpLSDptr);
- var uintL tmplen =
+ var uintC tmplen =
#if CL_DS_BIG_ENDIAN_P
tmpLSDptr - tmpMSDptr;
#else
{ var uintD* prod_MSDptr;
var uintD* prod_LSDptr;
var uintD* prodhi_LSDptr;
- num_stack_small_alloc(2*(uintL)len1,prod_MSDptr=,prod_LSDptr=);
+ num_stack_small_alloc(2*len1,prod_MSDptr=,prod_LSDptr=);
prodhi_LSDptr = prod_LSDptr lspop 2*k_lo;
// prod_MSDptr/2*len1/prod_LSDptr wird zuerst die beiden
// Produkte x1*y1 in prod_MSDptr/2*k_hi/prodhi_LSDptr
var uintD* prod_MSDptr;
var uintC prod_len = len1+len2;
var uintD* prod_LSDptr;
- num_stack_small_alloc((uintL)prod_len,prod_MSDptr=,prod_LSDptr=);
+ num_stack_small_alloc(prod_len,prod_MSDptr=,prod_LSDptr=);
{ var uintC k_hi = floor(len2,2); // Länge der High-Teile: floor(len2/2) >0
var uintC k_lo = len2 - k_hi; // Länge der Low-Teile: ceiling(len2/2) >0
// Es gilt k_hi <= k_lo <= len1 <= len2, k_lo + k_hi = len2.
#ifndef _BIT_REVERSE
#define _BIT_REVERSE
// Reverse an n-bit number x. n>0.
-static uintL bit_reverse (uintL n, uintL x)
+static uintC bit_reverse (uintL n, uintC x)
{
- var uintL y = 0;
+ var uintC y = 0;
do {
y <<= 1;
y |= (x & 1);
// threshold1 = 3: 25.6 sec 16.6 sec
// threshold1 = 4: 25.7 sec 17.6 sec
// threshold1 = 5: 26.1 sec 18.0 sec
-const uintL cl_nuss_threshold1 = 3;
+const uintC cl_nuss_threshold1 = 3;
// Threshold for recursion base in mulu_nuss_cyclic().
-const uintL cl_nuss_threshold2 = 1;
+const uintC cl_nuss_threshold2 = 1;
// Computes z[k] := sum(i+j==k mod N, x[i]*y[j]*(-1)^((i+j-k)/N))
// for all k=0..N-1.
-static void mulu_nuss_negacyclic (const uintL n, const uintL N, // N = 2^n
+static void mulu_nuss_negacyclic (const uintL n, const uintC N, // N = 2^n
const nuss_inword * x, // N words
const nuss_inword * y, // N words
nuss_outword * z // N words result
}
// 1 < n <= cl_nuss_threshold1.
#if 0 // straightforward, but slow
- var uintL k;
+ var uintC k;
for (k = 0; k < N; k++) {
- var uintL i;
+ var uintC i;
var nuss_outword accu;
mul(x[0],y[k], accu);
for (i = 1; i <= k; i++) {
z[k] = accu;
}
#else
- var const uintL M = (uintL)1 << (n-1); // M = N/2
- var uintL i, j, k;
+ var const uintC M = (uintC)1 << (n-1); // M = N/2
+ var uintC i, j, k;
for (k = 0; k < N; k++)
zero(z[k]);
for (i = 0; i < M; i++) {
- var uintL iM = i+M;
+ var uintC iM = i+M;
for (j = 0; j < M-i; j++) {
- var uintL jM = j+M;
+ var uintC jM = j+M;
// z[i+j] += x[i] (y[j] + y[j+M]) - (x[i] + x[i+M]) y[j+M]
// z[i+j+M] += x[i] (y[j] + y[j+M]) + (x[i+M] - x[i]) y[j]
var nuss_inword x_sum;
mul(x_sum,y[j], second); add(first,second, temp); add(z[i+j+M],temp, z[i+j+M]);
}
for (j = M-i; j < M; j++) {
- var uintL jM = j+M;
+ var uintC jM = j+M;
// z[i+j] += x[i] (y[j] + y[j+M]) - (x[i] + x[i+M]) y[j+M]
// z[i+j-M] -= x[i] (y[j] + y[j+M]) + (x[i+M] - x[i]) y[j]
var nuss_inword x_sum;
// Recursive FFT.
var const uintL m = n >> 1; // floor(n/2)
var const uintL r = n - m; // ceiling(n/2)
- var const uintL M = (uintL)1 << m; // M = 2^m
- var const uintL R = (uintL)1 << r; // R = 2^r
+ var const uintC M = (uintC)1 << m; // M = 2^m
+ var const uintC R = (uintC)1 << r; // R = 2^r
CL_ALLOCA_STACK;
var nuss_inword* const auX = cl_alloc_array(nuss_inword,2*N);
var nuss_inword* const auY = cl_alloc_array(nuss_inword,2*N);
var nuss_inword* const tmp2 = cl_alloc_array(nuss_inword,R);
var nuss_outword* const tmpZ = cl_alloc_array(nuss_outword,R);
var bool squaring = (x == y);
- var uintL i, j;
+ var uintC i, j;
// Initialize polynomials X(i) and Y(i).
for (i = 0; i < M; i++) {
{
X(i+M,j) = X(i,j);
// Level l = m-1..0:
for (l = m-1; l>=0; l--) {
- var const uintL smax = (uintL)1 << (m-l);
- var const uintL tmax = (uintL)1 << l;
- for (var uintL s = 0; s < smax; s++) {
- var uintL exp = bit_reverse(m-l,s) << (l + r-m);
- for (var uintL t = 0; t < tmax; t++) {
- var uintL i1 = (s << (l+1)) + t;
- var uintL i2 = i1 + tmax;
+ var const uintC smax = (uintC)1 << (m-l);
+ var const uintC tmax = (uintC)1 << l;
+ for (var uintC s = 0; s < smax; s++) {
+ var uintC exp = bit_reverse(m-l,s) << (l + r-m);
+ for (var uintC t = 0; t < tmax; t++) {
+ var uintC i1 = (s << (l+1)) + t;
+ var uintC i2 = i1 + tmax;
// Butterfly: replace (X(i1),X(i2)) by
// (X(i1) + w^exp*X(i2), X(i1) - w^exp*X(i2)).
for (j = 0; j < exp; j++) {
Y(i+M,j) = Y(i,j);
// Level l = m-1..0:
for (l = m-1; l>=0; l--) {
- var const uintL smax = (uintL)1 << (m-l);
- var const uintL tmax = (uintL)1 << l;
- for (var uintL s = 0; s < smax; s++) {
- var uintL exp = bit_reverse(m-l,s) << (l + r-m);
- for (var uintL t = 0; t < tmax; t++) {
- var uintL i1 = (s << (l+1)) + t;
- var uintL i2 = i1 + tmax;
+ var const uintC smax = (uintC)1 << (m-l);
+ var const uintC tmax = (uintC)1 << l;
+ for (var uintC s = 0; s < smax; s++) {
+ var uintC exp = bit_reverse(m-l,s) << (l + r-m);
+ for (var uintC t = 0; t < tmax; t++) {
+ var uintC i1 = (s << (l+1)) + t;
+ var uintC i2 = i1 + tmax;
// Butterfly: replace (Y(i1),Y(i2)) by
// (Y(i1) + w^exp*Y(i2), Y(i1) - w^exp*Y(i2)).
for (j = 0; j < exp; j++) {
var uintL l;
// Level l = 0..m-1:
for (l = 0; l < m; l++) {
- var const uintL smax = (uintL)1 << (m-l);
- var const uintL tmax = (uintL)1 << l;
- for (var uintL s = 0; s < smax; s++) {
- var uintL exp = bit_reverse(m-l,s) << (l + r-m);
- for (var uintL t = 0; t < tmax; t++) {
- var uintL i1 = (s << (l+1)) + t;
- var uintL i2 = i1 + tmax;
+ var const uintC smax = (uintC)1 << (m-l);
+ var const uintC tmax = (uintC)1 << l;
+ for (var uintC s = 0; s < smax; s++) {
+ var uintC exp = bit_reverse(m-l,s) << (l + r-m);
+ for (var uintC t = 0; t < tmax; t++) {
+ var uintC i1 = (s << (l+1)) + t;
+ var uintC i2 = i1 + tmax;
// Inverse Butterfly: replace (Z(i1),Z(i2)) by
// ((Z(i1)+Z(i2))/2, (Z(i1)-Z(i2))/(2*w^exp)).
for (j = 0; j < exp; j++)
}
// Level l=m:
for (i = 0; i < M; i++) {
- var uintL i1 = i;
- var uintL i2 = i1 + M;
+ var uintC i1 = i;
+ var uintC i2 = i1 + M;
// Inverse Butterfly: replace (Z(i1),Z(i2)) by
// ((Z(i1)+Z(i2))/2, (Z(i1)-Z(i2))/2).
for (j = 0; j < R; j++) {
// Computes z[k] := sum(i+j==k mod N, x[i]*y[j])
// for all k=0..N-1.
-static void mulu_nuss_cyclic (const uintL n, const uintL N, // N = 2^n
+static void mulu_nuss_cyclic (const uintL n, const uintC N, // N = 2^n
nuss_inword * x, // N words, modified!
nuss_inword * y, // N words, modified!
nuss_outword * z // N words result
#if 0 // useless code because cl_nuss_threshold2 == 1
if (n <= cl_nuss_threshold2) {
#if 0 // straightforward, but slow
- var uintL k;
+ var uintC k;
for (k = 0; k < N; k++) {
- var uintL i;
+ var uintC i;
var nuss_outword accu;
mul(x[0],y[k], accu);
for (i = 1; i <= k; i++) {
z[k] = accu;
}
#else
- var const uintL M = (uintL)1 << (n-1); // M = N/2
- var uintL i, j, k;
+ var const uintC M = (uintC)1 << (n-1); // M = N/2
+ var uintC i, j, k;
for (k = 0; k < N; k++)
zero(z[k]);
for (i = 0; i < M; i++) {
- var uintL iM = i+M;
+ var uintC iM = i+M;
for (j = 0; j < M; j++) {
- var uintL jM = j+M;
+ var uintC jM = j+M;
// z[i+j] += ((x[i] + x[i+M]) (y[j] + y[j+M]) + (x[i] - x[i+M]) (y[j] - y[j+M])) / 2
// z[i+j+M] += ((x[i] + x[i+M]) (y[j] + y[j+M]) - (x[i] - x[i+M]) (y[j] - y[j+M])) / 2
var nuss_inword x_sum;
mul(x_sum,y_sum, first);
mul(x_diff,y_diff, second);
add(first,second, temp); add(z[i+j],temp, z[i+j]);
- var uintL ijM = (i+j+M) & (N-1);
+ var uintC ijM = (i+j+M) & (N-1);
sub(first,second, temp); add(z[ijM],temp, z[ijM]);
}
}
}
#endif
var const uintL m = n-1;
- var const uintL M = (uintL)1 << m; // M = 2^m = N/2
- var uintL i;
+ var const uintC M = (uintC)1 << m; // M = 2^m = N/2
+ var uintC i;
// Chinese remainder theorem: u^N-1 = (u^M-1)*(u^M+1)
for (i = 0; i < M; i++) {
// Butterfly: replace (x(i),x(i+M))
// sum(i=0..N-1, x_i T^i), sum(i=0..N-1, y_i T^i)
// multipliziert, und zwar durch Fourier-Transformation (s.o.).
var uint32 n;
- integerlength32(len1-1, n=); // 2^(n-1) < len1 <= 2^n
- var uintL len = (uintL)1 << n; // kleinste Zweierpotenz >= len1
+ integerlengthC(len1-1, n=); // 2^(n-1) < len1 <= 2^n
+ var uintC len = (uintC)1 << n; // kleinste Zweierpotenz >= len1
// Wählt man N = len, so hat man ceiling(len2/(len-len1+1)) * FFT(len).
// Wählt man N = 2*len, so hat man ceiling(len2/(2*len-len1+1)) * FFT(2*len).
// Wir wählen das billigere von beiden:
n = n+1;
len = len << 1;
}
- var const uintL N = len; // N = 2^n
+ var const uintC N = len; // N = 2^n
CL_ALLOCA_STACK;
var nuss_inword* const x = cl_alloc_array(nuss_inword,N);
var nuss_inword* const y = cl_alloc_array(nuss_inword,N);
var nuss_outword* const z = cl_alloc_array(nuss_outword,N);
var uintD* const tmpprod = cl_alloc_array(uintD,len1+1);
var uintP i;
- var uintL destlen = len1+len2;
+ var uintC destlen = len1+len2;
clear_loop_lsp(destptr,destlen);
do {
- var uintL len2p; // length of a piece of source2
+ var uintC len2p; // length of a piece of source2
len2p = N - len1 + 1;
if (len2p > len2)
len2p = len2;
if (inc_loop_lsp(destptr lspop (len1+1),destlen-(len1+1)))
cl_abort();
} else {
- var uintL destlenp = len1 + len2p - 1;
+ var uintC destlenp = len1 + len2p - 1;
// destlenp = min(N,destlen-1).
var bool squaring = ((sourceptr1 == sourceptr2) && (len1 == len2p));
// Fill factor x.
}
// Other steps.
var int k;
- integerlength32((uint32)b_len-1,k=);
+ integerlengthC(b_len-1,k=);
// 2^(k-1) < b_len <= 2^k, so we need k steps.
var uintC n = 1;
for (; k>0; k--)
}
// Other steps.
var int k;
- integerlength32((uint32)b_len-1,k=);
+ integerlengthC(b_len-1,k=);
// 2^(k-1) < b_len <= 2^k, so we need k steps, plus one
// one more step at the beginning (because step 1 was not complete).
var uintC n = 0;
// Newton faster for 3200<N Newton faster for 2750<N
// When in doubt, prefer to choose the standard algorithm.
#if CL_USE_GMP
- static inline cl_boolean cl_recipsqrt_suitable (uintL n)
+ static inline cl_boolean cl_recipsqrt_suitable (uintC n)
{ return (cl_boolean)(n >= 3200); }
#else
// Use the old default values from CLN version <= 1.0.3 as a crude estimate.
// 5000 24.1 10.7
// 10000 98 23.2
// -----> Newton faster for 1570 <= N <= 1790 and for N >= 2100.
- static inline cl_boolean cl_recipsqrt_suitable (uintL n)
+ static inline cl_boolean cl_recipsqrt_suitable (uintC n)
{ return (cl_boolean)(n >= 2100); }
#endif
// A um 2s Bits nach links verschoben kopieren:
var uintD* new_a_MSDptr;
{ var uintD* new_a_LSDptr;
- num_stack_alloc(2*(uintL)n,new_a_MSDptr=,new_a_LSDptr=); // 2n Digits Platz belegen
+ num_stack_alloc(2*n,new_a_MSDptr=,new_a_LSDptr=); // 2n Digits Platz belegen
{var uintL shiftcount = 2*s;
if (!((a_len & bit(0)) ==0)) // a_len ungerade?
{ s += intDsize/2; lsprefnext(new_a_LSDptr) = 0; } // ja -> ein Nulldigit einschieben
// Idea from Torbjörn Granlund, see his "random2.c" file in gmp 2.0.
var uintD* ptr = MSDptr mspop len;
DS_clear_loop(MSDptr,len,ptr);
- var uintL bit_pos = 0;
+ var uintC bit_pos = 0;
var uint32 ran = 0;
var uintC ran_bits = 0;
- while (bit_pos < intDsize*(uintL)len)
+ while (bit_pos < intDsize*len)
{ if (ran_bits < log2_intDsize+1)
{ ran = random32(randomstate); ran_bits = 32; }
var uintL n_bits = (ran >> 1) % intDsize + 1; // number of bits
if (ran & 1)
{ // put in a bit string of n_bits bits at position bit_pos.
- if (bit_pos + n_bits > intDsize*(uintL)len)
- { n_bits = intDsize*(uintL)len - bit_pos; }
+ if (bit_pos + n_bits > intDsize*len)
+ { n_bits = intDsize*len - bit_pos; }
if (bit_pos / intDsize == (bit_pos + n_bits - 1) / intDsize)
{ // need to modify one digit
lspref(ptr,bit_pos/intDsize) |= (((uintD)1 << n_bits) - 1) << (bit_pos%intDsize);
ptr_after_prec = skip_digits(ptr,string_limit,10);
if (ptr_after_prec == ptr) goto not_float_syntax;
var cl_I prec1 = digits_to_I(ptr,ptr_after_prec-ptr,10);
- var uintL prec2 = cl_I_to_UL(prec1);
+ var uintC prec2 = cl_I_to_ulong(prec1);
prec = (float_base==10 ? float_format(prec2)
- : (float_format_t)((uintL)((1+prec2)*::log((double)float_base)*1.442695041)+1)
+ : (float_format_t)((uintC)((1+prec2)*::log((double)float_base)*1.442695041)+1)
);
} else {
switch (exponent_marker) {
// Count the number of significant digits.
ptr = ptr_after_sign;
while (ptr < ptr_after_fracpart && (*ptr == '0' || *ptr == '.')) ptr++;
- var uintL num_significant_digits =
+ var uintC num_significant_digits =
(ptr_after_fracpart - ptr) - (ptr_before_fracpart > ptr ? 1 : 0);
- var uintL prec2 = (num_significant_digits>=2 ? num_significant_digits-2 : 0);
+ var uintC prec2 = (num_significant_digits>=2 ? num_significant_digits-2 : 0);
var float_format_t precx =
(float_base==10 ? float_format(prec2)
- : (float_format_t)((uintL)((1+prec2)*::log((double)float_base)*1.442695041)+1)
+ : (float_format_t)((uintC)((1+prec2)*::log((double)float_base)*1.442695041)+1)
);
- if ((uintL)precx > (uintL)prec)
+ if ((uintC)precx > (uintC)prec)
prec = precx;
}
}
var cl_F& y = yf;
if (zerop(y)) // y=0.0 -> arcsin(y) = y als Ergebnis
return cl_C_R(0,y);
- if (float_exponent(y) <= (-(sintL)float_digits(y))>>1)
+ if (float_exponent(y) <= (-(sintC)float_digits(y))>>1)
// e <= -d/2 <==> e <= -ceiling(d/2)
return cl_C_R(0,y);
var cl_F temp = 1-square(y);
if (cl_false) {
complex_rational:
// x in Q(i)
- var uintL k = power2p(n);
+ var uintC k = power2p(n);
if (k) {
// n Zweierpotenz = 2^(k-1). n>1, also k>1
Mutable(cl_N,x);
{ if ((value) <= float_format_sfloat) { SF_statement } \
elif ((value) <= float_format_ffloat) { FF_statement } \
elif ((value) <= float_format_dfloat) { DF_statement } \
- else { var uintL len = ceiling((uintL)(value),intDsize); LF_statement } \
+ else { var uintC len = ceiling((uintC)(value),intDsize); LF_statement } \
}
} // namespace cln
const cl_F cl_float (const cl_F& x, float_format_t f)
{
- floatformatcase((uintL)f
+ floatformatcase((uintC)f
, return cl_F_to_SF(x);
, return cl_F_to_FF(x);
, return cl_F_to_DF(x);
const cl_F cl_float (const cl_I& x, float_format_t f)
{
- floatformatcase((uintL)f
+ floatformatcase((uintC)f
, return cl_I_to_SF(x);
, return cl_I_to_FF(x);
, return cl_I_to_DF(x);
const cl_F cl_float (const cl_RA& x, float_format_t f)
{
- floatformatcase((uintL)f
+ floatformatcase((uintC)f
, return cl_RA_to_SF(x);
, return cl_RA_to_FF(x);
, return cl_RA_to_DF(x);
if (eq(x,0)) { return 0.0; }
var cl_signean sign = -(cl_signean)minusp(x); // Vorzeichen
var cl_I abs_x = (sign==0 ? x : -x);
- var uintL exp = integer_length(abs_x); // (integer-length x)
+ var uintC exp = integer_length(abs_x); // (integer-length x)
// NDS zu |x|>0 bilden:
var const uintD* MSDptr;
var uintC len;
var const cl_I& b = denominator(x); // b
var cl_signean sign = -(cl_signean)minusp(a); // Vorzeichen
if (!(sign==0)) { a = -a; } // Betrag nehmen, liefert a
- var sintL lendiff = (sintL)integer_length(a) // (integer-length a)
- - (sintL)integer_length(b); // (integer-length b)
+ var sintC lendiff = (sintC)integer_length(a) // (integer-length a)
+ - (sintC)integer_length(b); // (integer-length b)
if (lendiff > DF_exp_high-DF_exp_mid) // Exponent >= n-m > Obergrenze ?
{
#if (cl_word_size==64)
if (eq(x,0)) { return cl_DF_0; }
var cl_signean sign = -(cl_signean)minusp(x); // Vorzeichen
var cl_I abs_x = (sign==0 ? x : -x);
- var uintL exp = integer_length(abs_x); // (integer-length x)
+ var uintC exp = integer_length(abs_x); // (integer-length x)
// NDS zu |x|>0 bilden:
var const uintD* MSDptr;
var uintC len;
var const cl_I& b = denominator(x); // b
var cl_signean sign = -(cl_signean)minusp(a); // Vorzeichen
if (!(sign==0)) { a = -a; } // Betrag nehmen, liefert a
- var sintL lendiff = (sintL)integer_length(a) // (integer-length a)
- - (sintL)integer_length(b); // (integer-length b)
+ var sintC lendiff = (sintC)integer_length(a) // (integer-length a)
+ - (sintC)integer_length(b); // (integer-length b)
if (lendiff > DF_exp_high-DF_exp_mid) // Exponent >= n-m > Obergrenze ?
{ cl_error_floating_point_overflow(); } // -> Overflow
if (lendiff < DF_exp_low-DF_exp_mid-2) // Exponent <= n-m+2 < Untergrenze ?
namespace cln {
-const cl_DF scale_float (const cl_DF& x, sintL delta)
+const cl_DF scale_float (const cl_DF& x, sintC delta)
{
// Methode:
// x=0.0 -> x als Ergebnis
#endif
if (delta >= 0)
// delta>=0
- { var uintL udelta = delta;
+ { var uintC udelta = delta;
if (udelta <= (uintL)(DF_exp_high-DF_exp_low))
{ exp = exp+udelta;
#if (cl_word_size==64)
}
else
// delta<0
- { var uintL udelta = -delta;
+ { var uintC udelta = -delta;
if (udelta <= (uintL)(DF_exp_high-DF_exp_low))
{ exp = exp-udelta;
#if (cl_word_size==64)
namespace cln {
MAYBE_INLINE
-uintL float_digits (const cl_DF& x)
+uintC float_digits (const cl_DF& x)
{
unused x;
return DF_mant_len+1; // 53
namespace cln {
MAYBE_INLINE2
-uintL float_precision (const cl_DF& x)
+uintC float_precision (const cl_DF& x)
{
if (zerop(x)) return 0;
return DF_mant_len+1; // 53
namespace cln {
-const cl_F scale_float (const cl_F& x, sintL delta)
+const cl_F scale_float (const cl_F& x, sintC delta)
{
floatcase(x
, return scale_float(x,delta);
if (eq(x,0)) { return 0.0; }
var cl_signean sign = -(cl_signean)minusp(x); // Vorzeichen
var cl_I abs_x = (sign==0 ? x : -x);
- var uintL exp = integer_length(abs_x); // (integer-length x)
+ var uintC exp = integer_length(abs_x); // (integer-length x)
// NDS zu |x|>0 bilden:
var const uintD* MSDptr;
var uintC len;
var const cl_I& b = denominator(x); // b
var cl_signean sign = -(cl_signean)minusp(a); // Vorzeichen
if (!(sign==0)) { a = -a; } // Betrag nehmen, liefert a
- var sintL lendiff = (sintL)integer_length(a) // (integer-length a)
- - (sintL)integer_length(b); // (integer-length b)
+ var sintC lendiff = (sintC)integer_length(a) // (integer-length a)
+ - (sintC)integer_length(b); // (integer-length b)
if (lendiff > FF_exp_high-FF_exp_mid) // Exponent >= n-m > Obergrenze ?
{ u.eksplicit = make_FF_word(sign,bit(FF_exp_len)-1,0); // Infinity
return u.machine_float;
if (eq(x,0)) { return cl_FF_0; }
var cl_signean sign = -(cl_signean)minusp(x); // Vorzeichen
var cl_I abs_x = (sign==0 ? x : -x);
- var uintL exp = integer_length(abs_x); // (integer-length x)
+ var uintC exp = integer_length(abs_x); // (integer-length x)
// NDS zu |x|>0 bilden:
var const uintD* MSDptr;
var uintC len;
var const cl_I& b = denominator(x); // b
var cl_signean sign = -(cl_signean)minusp(a); // Vorzeichen
if (!(sign==0)) { a = -a; } // Betrag nehmen, liefert a
- var sintL lendiff = (sintL)integer_length(a) // (integer-length a)
- - (sintL)integer_length(b); // (integer-length b)
+ var sintC lendiff = (sintC)integer_length(a) // (integer-length a)
+ - (sintC)integer_length(b); // (integer-length b)
if (lendiff > FF_exp_high-FF_exp_mid) // Exponent >= n-m > Obergrenze ?
{ cl_error_floating_point_overflow(); } // -> Overflow
if (lendiff < FF_exp_low-FF_exp_mid-2) // Exponent <= n-m+2 < Untergrenze ?
namespace cln {
-const cl_FF scale_float (const cl_FF& x, sintL delta)
+const cl_FF scale_float (const cl_FF& x, sintC delta)
{
// Methode:
// x=0.0 -> x als Ergebnis
FF_decode(x, { return x; }, sign=,exp=,mant=);
if (delta >= 0)
// delta>=0
- { var uintL udelta = delta;
+ { var uintC udelta = delta;
if (udelta <= (uintL)(FF_exp_high-FF_exp_low))
{ exp = exp+udelta;
return encode_FF(sign,exp,mant);
}
else
// delta<0
- { var uintL udelta = -delta;
+ { var uintC udelta = -delta;
if (udelta <= (uintL)(FF_exp_high-FF_exp_low))
{ exp = exp-udelta;
return encode_FF(sign,exp,mant);
namespace cln {
MAYBE_INLINE
-uintL float_digits (const cl_FF& x)
+uintC float_digits (const cl_FF& x)
{
unused x;
return FF_mant_len+1; // 24
namespace cln {
MAYBE_INLINE2
-uintL float_precision (const cl_FF& x)
+uintC float_precision (const cl_FF& x)
{
if (zerop(x)) return 0;
return FF_mant_len+1; // 24
ptr_after_prec = skip_digits(ptr,string_limit,10);
if (ptr_after_prec == ptr) goto not_float_syntax;
var cl_I prec1 = digits_to_I(ptr,ptr_after_prec-ptr,10);
- var uintL prec2 = cl_I_to_UL(prec1);
+ var uintC prec2 = cl_I_to_UL(prec1);
prec = (float_base==10 ? float_format(prec2)
- : (float_format_t)((uintL)((1+prec2)*::log((double)float_base)*1.442695041)+1)
+ : (float_format_t)((uintC)((1+prec2)*::log((double)float_base)*1.442695041)+1)
);
} else {
switch (exponent_marker) {
// Count the number of significant digits.
ptr = ptr_after_sign;
while (ptr < ptr_after_fracpart && (*ptr == '0' || *ptr == '.')) ptr++;
- var uintL num_significant_digits =
+ var uintC num_significant_digits =
(ptr_after_fracpart - ptr) - (ptr_before_fracpart > ptr ? 1 : 0);
- var uintL prec2 = (num_significant_digits>=2 ? num_significant_digits-2 : 0);
+ var uintC prec2 = (num_significant_digits>=2 ? num_significant_digits-2 : 0);
var float_format_t precx =
(float_base==10 ? float_format(prec2)
- : (float_format_t)((uintL)((1+prec2)*::log((double)float_base)*1.442695041)+1)
+ : (float_format_t)((uintC)((1+prec2)*::log((double)float_base)*1.442695041)+1)
);
- if ((uintL)precx > (uintL)prec)
+ if ((uintC)precx > (uintC)prec)
prec = precx;
}
}
CL_ALLOCA_STACK;
var uintD* r_MSDptr;
var uintD* r_LSDptr;
- var uintL r_len = 2*(uintL)len+2; // L�ge des Radikanden
+ var uintC r_len = 2*len+2; // Länge des Radikanden
num_stack_alloc(r_len, r_MSDptr=,r_LSDptr=);
if ((uexp & bit(0)) == (LF_exp_mid & bit(0)))
// Exponent gerade
swap(uintL, uexp1,uexp2);
}
}
- if (expdiff >= intDsize * (uintL)len + 2) // e1-e2 >= 16n+2 ?
+ if (expdiff >= intDsize * len + 2) // e1-e2 >= 16n+2 ?
{ return x1; } // ja -> x1 als Ergebnis
// neues Long-Float allozieren:
var Lfloat y = allocate_lfloat(len,uexp1,TheLfloat(x1)->sign);
}
// UDS y_mantMSDptr/len/y_mantLSDptr/rounding_bits normalisieren:
{var uintD* ptr = y_mantMSDptr;
- var uintL k = 0;
+ var uintC k = 0;
var uintC count;
dotimesC(count,len,
{ if (!(mspref(ptr,0)==0)) goto nonzero_found;
}
var cl_signean sign = -(cl_signean)minusp(y); // Vorzeichen von y
var cl_I abs_y = (sign==0 ? y : -y);
- var uintL y_exp = integer_length(abs_y);
+ var uintC y_exp = integer_length(abs_y);
var uintC len = TheLfloat(x)->len;
#ifndef CL_LF_PEDANTIC
if (ceiling(y_exp,intDsize) > len)
// y nicht zu einer NUDS normalisieren! (Damit ein Bit Spielraum ist.)
// Zähler bilden: x * 2^(intDsize*y_len)
var uintD* z_MSDptr;
- var uintL z_len;
+ var uintC z_len;
var uintD* z_LSDptr;
- z_len = (uintL)len + (uintL)y_len;
+ z_len = len + y_len;
num_stack_alloc(z_len, z_MSDptr=,z_LSDptr=);
{ var uintD* ptr = copy_loop_msp(arrayMSDptr(TheLfloat(x)->data,len),z_MSDptr,len); // len Digits
clear_loop_msp(ptr,y_len); // und y_len Null-Digits
if (TheLfloat(x)->expo == 0) { return x; }
var cl_signean sign = -(cl_signean)minusp(y); // Vorzeichen von y
var cl_I abs_y = (sign==0 ? y : -y);
- var uintL y_exp = integer_length(abs_y);
+ var uintC y_exp = integer_length(abs_y);
var uintC len = TheLfloat(x)->len;
#ifndef CL_LF_PEDANTIC
if (ceiling(y_exp,intDsize) > len)
TheLfloat(x1)->sign ^ TheLfloat(x2)->sign // Vorzeichen kombinieren
);
// Nenner bilden:
- var uintL n_len;
+ var uintC n_len;
n_len = len2;
#ifndef CL_LF_PEDANTIC
if (n_len > len) { n_len = len+1; }
// Zähler bilden:
CL_ALLOCA_STACK;
var uintD* z_MSDptr;
- var uintL z_len;
+ var uintC z_len;
var uintD* z_LSDptr;
z_len = n_len + len + 1;
num_stack_alloc(z_len, z_MSDptr=,z_LSDptr=);
if (eq(x,0)) { return encode_LF0(len); } // x=0 -> Ergebnis 0.0
var cl_signean sign = -(cl_signean)minusp(x); // Vorzeichen von x
var cl_I abs_x = (sign==0 ? x : -x);
- var uintL exp = integer_length(abs_x); // (integer-length x) < intDsize*2^intCsize
+ var uintC exp = integer_length(abs_x); // (integer-length x) < intDsize*2^intCsize
// Teste, ob exp <= LF_exp_high-LF_exp_mid :
if ( (log2_intDsize+intCsize < 32)
&& ((uintL)(intDsize*bitc(intCsize)-1) <= (uintL)(LF_exp_high-LF_exp_mid))
var const cl_I& b = denominator(x); // b
var cl_signean sign = -(cl_signean)minusp(a); // Vorzeichen
if (!(sign==0)) { a = -a; } // Betrag nehmen, liefert a
- var sintL lendiff = (sintL)integer_length(a) // (integer-length a)
- - (sintL)integer_length(b); // (integer-length b)
+ var sintC lendiff = (sintC)integer_length(a) // (integer-length a)
+ - (sintC)integer_length(b); // (integer-length b)
// |lendiff| < intDsize*2^intCsize. Da für LF-Exponenten ein sintL zur
// Verfügung steht, braucht man keinen Test auf Overflow oder Underflow.
- var uintL difflimit = intDsize*(uintL)len + 1; // 16n+1
+ var uintC difflimit = intDsize*len + 1; // 16n+1
var cl_I zaehler;
var cl_I nenner;
- if (lendiff > (sintL)difflimit)
+ if (lendiff > (sintC)difflimit)
// 0 <= k-m-16n-1 < k < intDsize*2^intCsize
- { nenner = ash(b,(uintL)(lendiff - difflimit));
+ { nenner = ash(b,(uintC)(lendiff - difflimit));
zaehler = a;
}
else
// 0 < -k+m+16n+1 <= m+1 + 16n < intDsize*2^intCsize + intDsize*2^intCsize
- { zaehler = ash(a,(uintL)(difflimit - lendiff)); // (ash a -k+m+16n+1)
+ { zaehler = ash(a,(uintC)(difflimit - lendiff)); // (ash a -k+m+16n+1)
nenner = b; // b
}
// Division zaehler/nenner durchführen:
var uintC mantlen;
LF_decode(x, { return x; }, sign=,exp=,mantMSDptr=,mantlen=,);
if (exp<0) { return encode_LF0(mantlen); } // e<0 -> Ergebnis 0.0
- if ((uintL)exp >= intDsize*(uintL)mantlen) // e>=16n -> x als Ergebnis
+ if ((uintL)exp >= intDsize*mantlen) // e>=16n -> x als Ergebnis
{ return x; }
else
// 0 <= e < 16n
return encode_LF0(len); // e<0 -> Ergebnis 0.0
}
var uintL exp = uexp - LF_exp_mid;
- if (exp >= intDsize*(uintL)len) // e>=16n -> x als Ergebnis
+ if (exp >= intDsize*len) // e>=16n -> x als Ergebnis
{ return x; }
// 0 <= e < 16n
// alle hinteren 16n-e Bits wegrunden:
var uintC mantlen;
LF_decode(x, { return x; }, sign=,exp=,mantMSDptr=,mantlen=,);
if (exp<=0) { return encode_LF0(mantlen); } // e<=0 -> Ergebnis 0.0
- if ((uintL)exp >= intDsize*(uintL)mantlen) // e>=16n -> x als Ergebnis
+ if ((uintL)exp >= intDsize*mantlen) // e>=16n -> x als Ergebnis
{ return x; }
else
// 0 < e < 16n
return encode_LF0(len); // e<=0 -> Ergebnis 0.0
}
var uintL exp = uexp - LF_exp_mid;
- if (exp >= intDsize*(uintL)len) // e>=16n -> x als Ergebnis
+ if (exp >= intDsize*len) // e>=16n -> x als Ergebnis
{ return x; }
// 0 < e < 16n
var Lfloat y = allocate_lfloat(len,uexp,TheLfloat(x)->sign); // neues Long-Float
var uintC mantlen;
LF_decode(x, { return x; }, sign=,exp=,mantMSDptr=,mantlen=,);
if (exp<=0) { return encode_LF1s(sign,mantlen); } // e<=0 -> Ergebnis +-1.0
- if ((uintL)exp >= intDsize*(uintL)mantlen) // e>=16n -> x als Ergebnis
+ if ((uintL)exp >= intDsize*mantlen) // e>=16n -> x als Ergebnis
{ return x; }
else
// 0 < e < 16n
return encode_LF1s(TheLfloat(x)->sign,len); // e<=0 -> Ergebnis +-1.0
}
var uintL exp = uexp - LF_exp_mid;
- if (exp >= intDsize*(uintL)len) // e>=16n -> x als Ergebnis
+ if (exp >= intDsize*len) // e>=16n -> x als Ergebnis
{ return x; }
// 0 < e < 16n
// Testen, ob alle hinteren 16n-e Bits =0 sind:
namespace cln {
-const cl_LF scale_float (const cl_LF& x, sintL delta)
+const cl_LF scale_float (const cl_LF& x, sintC delta)
{
// Methode:
// delta=0 -> x als Ergebnis
if (delta == 0) { return x; } // delta=0 -> x als Ergebnis
var uintL uexp = TheLfloat(x)->expo;
if (uexp==0) { return x; }
- var uintL udelta = delta;
+ var uintC udelta = delta;
if (delta >= 0) {
// udelta = delta >=0
if ( ((uexp = uexp+udelta) < udelta) // Exponent-Überlauf?
)
{ cl_error_floating_point_overflow(); }
} else {
- // delta <0, udelta = 2^32+delta
- if ( ((uexp = uexp+udelta) >= udelta) // oder Exponent-Unterlauf?
+ // delta <0, udelta = 2^intCsize+delta
+ if ( ((uintL)(-(uexp = uexp+udelta)) <= (uintC)(-udelta)) // oder Exponent-Unterlauf?
|| (uexp < LF_exp_low) // oder Exponent zu klein?
)
{ cl_error_floating_point_underflow(); }
// Vorzeichen*Mantisse ist die DS MSDptr/len1/LSDptr.
// (ASH Vorzeichen*Mantisse (- e 16n)) durchführen:
return ash(DS_to_I(MSDptr,len1),
- minus(uexp, LF_exp_mid + intDsize*(uintL)len)
+ minus(uexp, LF_exp_mid + intDsize*len)
);
}
namespace cln {
MAYBE_INLINE
-uintL float_digits (const cl_LF& x)
+uintC float_digits (const cl_LF& x)
{
- return intDsize*(uintL)(TheLfloat(x)->len);
+ return intDsize*(uintC)(TheLfloat(x)->len);
}
} // namespace cln
// Mantisse
mant,
// e-intDsize*n = uexp-LF_exp_mid-intDsize*n als Integer
- minus(uexp, LF_exp_mid + intDsize*(uintL)len),
+ minus(uexp, LF_exp_mid + intDsize*len),
(sign>=0 ? cl_I(1) : cl_I(-1)) // (-1)^s erzeugen
);
}
// <==> intDsize*n < intDsize^2*inc^2 - 4*intDsize*inc + 4
// <==> n <= intDsize*inc^2 - 4*inc
return
- #define NMAX(k) (uintL)((intDsize*(k)-4)*(k))
+ #define NMAX(k) (uintC)((intDsize*(k)-4)*(k))
#define FITS(n,k) ((n) <= NMAX(k))
- #define n_max (uintL)(bitm(intCsize)-1)
+ #define n_max (uintC)(bitm(intCsize)-1)
#define TRYINC(inc) FITS(n_max,inc) || FITS(n,inc) ? RETINC(inc) :
#define RETINC(inc) \
/* at this point we know n <= NMAX(inc) */ \
// <==> intDsize*n < intDsize^2*inc^2 - 66*intDsize*inc + 1089
// <==> n <= intDsize*inc^2 - 66*inc + floor(1089/intDsize)
return
- #define NMAX(k) (uintL)((intDsize*(k)-66)*(k)+floor(1089,intDsize))
+ #define NMAX(k) (uintC)((intDsize*(k)-66)*(k)+floor(1089,intDsize))
#define FITS(n,k) ((intDsize*(k) > 33) && ((n) <= NMAX(k)))
- #define n_max (uintL)(bitm(intCsize)-1)
+ #define n_max (uintC)(bitm(intCsize)-1)
#define TRYINC(inc) FITS(n_max,inc) || FITS(n,inc) ? RETINC(inc) :
#define RETINC(inc) \
/* at this point we know n <= NMAX(inc) */ \
namespace cln {
MAYBE_INLINE2
-uintL float_precision (const cl_LF& x)
+uintC float_precision (const cl_LF& x)
{
if (zerop(x)) return 0;
- return intDsize*(uintL)(TheLfloat(x)->len);
+ return intDsize*(uintC)(TheLfloat(x)->len);
}
} // namespace cln
// 1 ulp(x) = 2^(ex-dx), 1 ulp(y) = 2^(ey-dy).
// Falls ex-dx < ey-dy, x von Precision dx auf dy-ey+ex verkürzen.
var sintL ey = float_exponent(y);
- var sintL dy = float_precision(y);
+ var sintC dy = float_precision(y);
if (dy==0) // zerop(y) ?
cl_abort();
var sintL ex = float_exponent(x);
- var sintL dx = float_precision(x);
+ var sintC dx = float_precision(x);
if (dx==0) // zerop(x) ?
return x;
var sintL d = ex - ey;
return LF_to_LF(x,LF_minlen);
if (d >= dx - dy)
return x;
- var uintL new_dx = dy + d;
- var uintL len = ceiling(new_dx,intDsize);
+ var uintC new_dx = dy + d;
+ var uintC len = ceiling(new_dx,intDsize);
if (len < LF_minlen)
len = LF_minlen;
- if (intDsize*len < (uintL)dx)
+ if (intDsize*len < (uintC)dx)
return shorten(x,len);
else
return x;
// Falls ex-dx < ey, x von Precision dx auf ex-ey verkürzen.
var sintL ey = float_exponent(y);
var sintL ex = float_exponent(x);
- var uintL dx = float_precision(x);
+ var uintC dx = float_precision(x);
if (dx==0) // zerop(x) ?
return x;
var sintL ulpx = ex - dx;
namespace cln {
-uintL float_digits (const cl_F& x)
+uintC float_digits (const cl_F& x)
{
floatcase(x
, return float_digits(x);
inline const cl_LF LF_negative_epsilon (uintC len)
{
- var Lfloat erg = allocate_lfloat(len,LF_exp_mid-intDsize*(uintL)len,0);
+ var Lfloat erg = allocate_lfloat(len,LF_exp_mid-intDsize*len,0);
var uintD* ptr = &TheLfloat(erg)->data[0];
#if CL_DS_BIG_ENDIAN_P
*ptr++ = bit(intDsize-1);
const cl_F float_negative_epsilon (float_format_t f)
{
- floatformatcase((uintL)f
+ floatformatcase((uintC)f
, return SF_negative_epsilon;
, return FF_negative_epsilon;
, return DF_negative_epsilon;
inline const cl_LF LF_epsilon (uintC len)
{
- var Lfloat erg = allocate_lfloat(len,LF_exp_mid+1-intDsize*(uintL)len,0);
+ var Lfloat erg = allocate_lfloat(len,LF_exp_mid+1-intDsize*len,0);
var uintD* ptr = &TheLfloat(erg)->data[0];
#if CL_DS_BIG_ENDIAN_P
*ptr++ = bit(intDsize-1);
const cl_F float_epsilon (float_format_t f)
{
- floatformatcase((uintL)f
+ floatformatcase((uintC)f
, return SF_epsilon;
, return FF_epsilon;
, return DF_epsilon;
const cl_F least_negative_float (float_format_t f)
{
- floatformatcase((uintL)f
+ floatformatcase((uintC)f
, return least_negative_SF;
, return least_negative_FF;
, return least_negative_DF;
const cl_F least_positive_float (float_format_t f)
{
- floatformatcase((uintL)f
+ floatformatcase((uintC)f
, return least_positive_SF;
, return least_positive_FF;
, return least_positive_DF;
const cl_F most_negative_float (float_format_t f)
{
- floatformatcase((uintL)f
+ floatformatcase((uintC)f
, return most_negative_SF;
, return most_negative_FF;
, return most_negative_DF;
const cl_F most_positive_float (float_format_t f)
{
- floatformatcase((uintL)f
+ floatformatcase((uintC)f
, return most_positive_SF;
, return most_positive_FF;
, return most_positive_DF;
namespace cln {
-uintL float_precision (const cl_F& x)
+uintC float_precision (const cl_F& x)
{
floatcase(x
, return float_precision(x);
// 0 < -e <= LF_exp_mid-LF_exp_low + intDsize*len < 2^32,
var cl_I minus_e = -e;
var uintL _e = cl_I_to_UL(minus_e); // daher kein Überlauf
- var uintL k = ord2(n);
+ var uintC k = ord2(n);
if (k >= _e)
// Kürze mit 2^(-e).
return ash(n,e);
else
// Kürze mit 2^k, 0 <= k < -e.
- return I_I_to_RT(ash(n,-(sintL)k),
- ash(1,minus_e-UL_to_I(k)));
+ return I_I_to_RT(ash(n,-(sintC)k),
+ ash(1,minus_e-(cl_I)(unsigned long)k));
#endif
}
}
// 1 ulp(x) = 2^(ex-dx), 1 ulp(y) = 2^(ey-dy).
// Falls ex-dx < ey-dy, x von Precision dx auf dy-ey+ex verkürzen.
var sintL ey = float_exponent(y);
- var sintL dy = float_precision(y);
+ var sintC dy = float_precision(y);
if (dy==0) // zerop(y) ?
cl_abort();
var sintL ex = float_exponent(x);
- var sintL dx = float_precision(x);
+ var sintC dx = float_precision(x);
if (dx==0) // zerop(x) ?
return x;
var sintL d = ex - ey;
return cl_F_to_SF(x);
if (d >= dx - dy)
return x;
- var uintL new_dx = dy + d;
+ var uintC new_dx = dy + d;
floatformatcase(new_dx
, return cl_F_to_SF(x);
, return cl_F_to_FF(x);
, return cl_F_to_DF(x);
- , if (intDsize*len < (uintL)dx)
+ , if (intDsize*len < (uintC)dx)
return shorten(The(cl_LF)(x),len);
else
return x;
// x/=0, also ist sign das Vorzeichen von x und
// |x| = 2^binexpo * float(binmant,x) . Ab jetzt oBdA x>0.
// Also x = 2^binexpo * float(binmant,x) .
- var uintL l = integer_length(binmant); // Anzahl der Bits von binmant, >=3
+ var uintC l = integer_length(binmant); // Anzahl der Bits von binmant, >=3
var cl_I binmant2 = ash(binmant,1); // 2*binmant
var cl_I oben = plus1(binmant2); // obere Intervallgrenze ist
// (x+x2)/2 = 2^(binexpo-1) * oben
// Wandle dazu 2^e := 2^(binexpo-1) ins Dezimalsystem um.
var cl_I e = binexpo - 1;
var bool e_gross = (abs(e) > ash(l,1)); // Ist |e| recht groß, >2*l ?
- var uintL g; // Hilfsvariablen für den Fall, daß |e| groß ist
+ var uintC g; // Hilfsvariablen für den Fall, daß |e| groß ist
var cl_I f; //
var cl_I zehn_d; // Hilfsvariable 10^|d| für den Fall, daß |e| klein ist
var cl_I d; // Ergebnisvariablen
const cl_F random_F (random_state& randomstate, const cl_F& n)
{
- var uintL d = float_digits(n); // d = (float-digits n) > 0
+ var uintC d = float_digits(n); // d = (float-digits n) > 0
// Bilde neue UDS mit d Zufallsbits:
CL_ALLOCA_STACK;
- var uintL len = ceiling(d,intDsize);
+ var uintC len = ceiling(d,intDsize);
var uintD* MSDptr;
num_stack_alloc_1(len,MSDptr=,);
random_UDS(randomstate,MSDptr,len); // len (>0) Zufallsdigits
var cl_I mant = UDS_to_I(MSDptr,len);
// Bilde Zufalls-Float zwischen 0 und 1
// = (scale-float (float Zufalls-Integer,d_Bits n) (- d)) :
- var cl_F result = scale_float(cl_float(mant,n),-(sintL)d) * n;
+ var cl_F result = scale_float(cl_float(mant,n),-(sintC)d) * n;
// result ist ein Zufalls-Float >=0, <=n.
if (result == n)
// falls (durch Rundung) result=n, durch 0 ersetzen:
if (eq(x,0)) { return SF_0; }
var cl_signean sign = -(cl_signean)minusp(x); // Vorzeichen
var cl_I abs_x = (sign==0 ? x : -x);
- var uintL exp = integer_length(abs_x); // (integer-length x)
+ var uintC exp = integer_length(abs_x); // (integer-length x)
// NDS zu |x|>0 bilden:
var const uintD* MSDptr;
var uintC len;
var const cl_I& b = denominator(x); // b
var cl_signean sign = -(cl_signean)minusp(a); // Vorzeichen
if (!(sign==0)) { a = -a; } // Betrag nehmen, liefert a
- var sintL lendiff = (sintL)integer_length(a) // (integer-length a)
- - (sintL)integer_length(b); // (integer-length b)
+ var sintC lendiff = (sintC)integer_length(a) // (integer-length a)
+ - (sintC)integer_length(b); // (integer-length b)
if (lendiff > SF_exp_high-SF_exp_mid) // Exponent >= n-m > Obergrenze ?
{ cl_error_floating_point_overflow(); } // -> Overflow
if (lendiff < SF_exp_low-SF_exp_mid-2) // Exponent <= n-m+2 < Untergrenze ?
namespace cln {
-const cl_SF scale_float (const cl_SF& x, sintL delta)
+const cl_SF scale_float (const cl_SF& x, sintC delta)
{
// Methode:
// x=0.0 -> x als Ergebnis
SF_decode(x, { return x; }, sign=,exp=,mant=);
if (delta >= 0)
// delta>=0
- { var uintL udelta = delta;
+ { var uintC udelta = delta;
if (udelta <= (uintL)(SF_exp_high-SF_exp_low))
{ exp = exp+udelta;
return encode_SF(sign,exp,mant);
}
else
// delta<0
- { var uintL udelta = -delta;
+ { var uintC udelta = -delta;
if (udelta <= (uintL)(SF_exp_high-SF_exp_low))
{ exp = exp-udelta;
return encode_SF(sign,exp,mant);
namespace cln {
MAYBE_INLINE
-uintL float_digits (const cl_SF& x)
+uintC float_digits (const cl_SF& x)
{
unused x;
return SF_mant_len+1; // 17
namespace cln {
MAYBE_INLINE2
-uintL float_precision (const cl_SF& x)
+uintC float_precision (const cl_SF& x)
{
if (zerop(x)) return 0;
return SF_mant_len+1; // 17
{
if (zerop(x))
return x;
- var uintL actuallen = TheLfloat(x)->len;
- var uintL d = float_digits(x);
+ var uintC actuallen = TheLfloat(x)->len;
+ var uintC d = float_digits(x);
var sintL e = float_exponent(x);
- if (e <= (sintL)(-d)>>1) // e <= -d/2 <==> e <= -ceiling(d/2)
+ if (e <= (sintC)(-d)>>1) // e <= -d/2 <==> e <= -ceiling(d/2)
return x; // ja -> x als Ergebnis
if (actuallen >= 34) {
DeclareType(cl_LF,x);
} else {
// naive2:
// floating-point representation with smooth precision reduction
- var cl_LF eps = scale_float(b,-(sintL)d-10);
+ var cl_LF eps = scale_float(b,-(sintC)d-10);
loop {
var cl_LF new_sum = sum + LF_to_LF(b/(cl_I)i,actuallen); // (+ sum (/ b i))
if (new_sum == sum) // = sum ?
}
if (zerop(x))
return x;
- var uintL d = float_digits(x);
+ var uintC d = float_digits(x);
var sintL e = float_exponent(x);
- if (e <= (sintL)(-d)>>1) // e <= -d/2 <==> e <= -ceiling(d/2)
+ if (e <= (sintC)(-d)>>1) // e <= -d/2 <==> e <= -ceiling(d/2)
return x; // ja -> x als Ergebnis
var uintL k = 0; // Rekursionszähler k:=0
// Bei e <= -1-limit_slope*floor(sqrt(d)) kann die Potenzreihe
{
if (zerop(x))
return x;
- var uintL actuallen = TheLfloat(x)->len;
- var uintL d = float_digits(x);
+ var uintC actuallen = TheLfloat(x)->len;
+ var uintC d = float_digits(x);
var sintL e = float_exponent(x);
- if (e <= (sintL)(-d)>>1) // e <= -d/2 <==> e <= -ceiling(d/2)
+ if (e <= (sintC)(-d)>>1) // e <= -d/2 <==> e <= -ceiling(d/2)
return x; // ja -> x als Ergebnis
var uintL k = 0; // Rekursionszähler k:=0
// Bei e <= -1-limit_slope*floor(sqrt(d)) kann die Potenzreihe
} else {
// naive2:
// floating-point representation with smooth precision reduction
- var cl_LF eps = scale_float(b,-(sintL)d-10);
+ var cl_LF eps = scale_float(b,-(sintC)d-10);
loop {
var cl_LF new_sum = sum + LF_to_LF(b/(cl_I)i,actuallen); // (+ sum (/ b i))
if (new_sum == sum) // = sum ?
{
if (zerop(x))
return x;
- var uintL d = float_digits(x);
+ var uintC d = float_digits(x);
var sintL e = float_exponent(x);
- if (e <= (sintL)(-d)>>1) // e <= -d/2 <==> e <= -ceiling(d/2)
+ if (e <= (sintC)(-d)>>1) // e <= -d/2 <==> e <= -ceiling(d/2)
return x; // ja -> x als Ergebnis
var uintL k = 0; // Rekursionszähler k:=0
var uintL sqrt_d = floor(isqrt(d),2); // limit_slope*floor(sqrt(d))
// z' = truncate(y*2^(2n))/2^(2n).
// Set z := z + z' and x+i*y := (x+i*y)*exp(-i*z').
var uintC len = TheLfloat(t)->len;
- var uintL d = intDsize*(uintL)len;
- if (zerop(t) || (float_exponent(t) <= (sintL)(-d)>>1))
+ var uintC d = intDsize*len;
+ if (zerop(t) || (float_exponent(t) <= (sintC)(-d)>>1))
return t;
var cl_LF x = recip(sqrt(cl_I_to_LF(1,len) + square(t)));
var cl_LF y = t*x;
var cl_LF z = cl_I_to_LF(0,len);
loop {
- if (zerop(y) || (float_exponent(y) <= (sintL)(-d)>>1))
+ if (zerop(y) || (float_exponent(y) <= (sintC)(-d)>>1))
break;
var cl_idecoded_float y_ = integer_decode_float(y);
// y = (-1)^sign * 2^exponent * mantissa
- var uintL lm = integer_length(y_.mantissa);
+ var uintC lm = integer_length(y_.mantissa);
var uintL me = cl_I_to_UL(- y_.exponent);
var cl_I p;
- var uintL lq;
+ var uintC lq;
var cl_boolean last_step = cl_false;
if (lm >= me) { // |y| >= 1/2 ?
p = y_.sign; // 1 or -1
if (2*n >= lm)
last_step = cl_true;
}
- z = z + scale_float(cl_I_to_LF(p,len),-(sintL)lq);
+ z = z + scale_float(cl_I_to_LF(p,len),-(sintC)lq);
if (last_step)
break;
var cl_LF_cos_sin_t cis_z = cl_cossin_aux(-p,lq,len);
const cl_F catalanconst (float_format_t f)
{
- floatformatcase((uintL)f
+ floatformatcase((uintC)f
, return cl_SF_catalanconst;
, return cl_FF_catalanconst;
, return cl_DF_catalanconst;
var cl_F_div_t q_r = cl_round_pi(cl_F_extendsqrt(x));
var cl_I& q = q_r.quotient;
var cl_LF r = The(cl_LF)(q_r.remainder);
- if (zerop(r) || (float_exponent(r) <= (-(sintL)float_digits(r))>>1))
+ if (zerop(r) || (float_exponent(r) <= (-(sintC)float_digits(r))>>1))
cos_r = cl_float(1,x); // (cos r) = 1.0
else {
var cl_LF s = scale_float(r,-1); // s := r/2
var cl_F_div_t q_r = cl_round_pi(cl_F_extendsqrt(x));
var cl_I& q = q_r.quotient;
var cl_F& r = q_r.remainder;
- if (zerop(r) || (float_exponent(r) <= (-(sintL)float_digits(r))>>1))
+ if (zerop(r) || (float_exponent(r) <= (-(sintC)float_digits(r))>>1))
cos_r = cl_float(1,x); // (cos r) = 1.0
else {
var cl_F s = scale_float(r,-1); // s := r/2
// e<0
if (zerop(x))
return cl_float(1,x);
- var uintL d = float_digits(x);
- if (e <= (1-(sintL)d)>>1) // e <= (1-d)/2 <==> e <= -ceiling((d-1)/2) ?
+ var uintC d = float_digits(x);
+ if (e <= (1-(sintC)d)>>1) // e <= (1-d)/2 <==> e <= -ceiling((d-1)/2) ?
return cl_float(1,x); // ja -> 1.0 als Ergebnis
// Rechengenauigkeit erhöhen
if (longfloatp(x)) {
var sintL e = float_exponent(x);
if (e < 0) { // Exponent e abtesten
// e<0
- if (zerop(x) || (e <= (1-(sintL)float_digits(x))>>1))
+ if (zerop(x) || (e <= (1-(sintC)float_digits(x))>>1))
// e <= (1-d)/2 <==> e <= -ceiling((d-1)/2)
return cosh_sinh_t(cl_float(1,x),x);
// Rechengenauigkeit erhöhen
var cl_LF r = The(cl_LF)(q_r.remainder);
var cl_LF y = sinx_naive(r); // y := sin(r)^2
// erste Komponente cos(r) berechnen:
- if (zerop(r) || (float_exponent(r) <= (-(sintL)float_digits(r))>>1))
+ if (zerop(r) || (float_exponent(r) <= (-(sintC)float_digits(r))>>1))
cos_r = cl_float(1,x); // cos(r) = 1.0
else
cos_r = cl_float(sqrt(1-y),x); // cos(r) = sqrt(1-y)
var cl_F& r = q_r.remainder;
var cl_F y = sinxbyx_naive(r); // y := (sin(r)/r)^2
// erste Komponente cos(r) berechnen:
- if (zerop(r) || (float_exponent(r) <= (-(sintL)float_digits(r))>>1))
+ if (zerop(r) || (float_exponent(r) <= (-(sintC)float_digits(r))>>1))
cos_r = cl_float(1,x); // cos(r) = 1.0
else
cos_r = cl_float(sqrt(1 - square(r)*y),x); // sqrt(1-r^2*y)
const cl_F eulerconst (float_format_t f)
{
- floatformatcase((uintL)f
+ floatformatcase((uintC)f
, return cl_SF_eulerconst;
, return cl_FF_eulerconst;
, return cl_DF_eulerconst;
const cl_F exp1 (float_format_t f)
{
- floatformatcase((uintL)f
+ floatformatcase((uintC)f
, return cl_SF_exp1;
, return cl_FF_exp1;
, return cl_DF_exp1;
if (zerop(x))
return cl_float(1,x);
var uintL actuallen = TheLfloat(x)->len;
- var uintL d = float_digits(x);
+ var uintC d = float_digits(x);
var sintL e = float_exponent(x);
- if (e < -(sintL)d) // e < -d ?
+ if (e < -(sintC)d) // e < -d ?
return cl_float(1,x); // ja -> 1.0 als Ergebnis
{ Mutable(cl_LF,x);
var uintL k = 0; // Rekursionszähler k:=0
// Potenzreihe anwenden:
var int i = 0;
var cl_LF b = cl_float(1,x); // b := (float 1 x)
- var cl_LF eps = scale_float(b,-(sintL)d-10);
+ var cl_LF eps = scale_float(b,-(sintC)d-10);
var cl_LF sum = cl_float(0,x); // sum := (float 0 x)
loop {
var cl_LF new_sum = sum + LF_to_LF(b,actuallen);
}
if (zerop(x))
return cl_float(1,x);
- var uintL d = float_digits(x);
+ var uintC d = float_digits(x);
var sintL e = float_exponent(x);
- if (e < -(sintL)d) // e < -d ?
+ if (e < -(sintC)d) // e < -d ?
return cl_float(1,x); // ja -> 1.0 als Ergebnis
{ Mutable(cl_F,x);
var uintL k = 0; // Rekursionszähler k:=0
const cl_F cl_ln10 (float_format_t f)
{
- floatformatcase((uintL)f
+ floatformatcase((uintC)f
, return cl_SF_ln10;
, return cl_FF_ln10;
, return cl_DF_ln10;
const cl_F cl_ln2 (float_format_t f)
{
- floatformatcase((uintL)f
+ floatformatcase((uintC)f
, return cl_SF_ln2;
, return cl_FF_ln2;
, return cl_DF_ln2;
if (zerop(y)) // y=0.0 -> y als Ergebnis
return y;
var uintL actuallen = TheLfloat(x)->len;
- var uintL d = float_digits(x);
+ var uintC d = float_digits(x);
var sintL e = float_exponent(y);
- if (e <= -(sintL)d) // e <= -d ?
+ if (e <= -(sintC)d) // e <= -d ?
return y; // ja -> y als Ergebnis
{ Mutable(cl_LF,x);
var uintL k = 0; // Rekursionszähler k:=0
} else {
// naive2:
// floating-point representation with smooth precision reduction
- var cl_LF eps = scale_float(b,-(sintL)d-10);
+ var cl_LF eps = scale_float(b,-(sintC)d-10);
loop {
var cl_LF new_sum = sum + LF_to_LF(b/(cl_I)i,actuallen); // (+ sum (/ b i))
if (new_sum == sum) // = sum ?
} else {
// naive2:
// floating-point representation with smooth precision reduction
- var cl_LF eps = scale_float(b,-(sintL)d-10);
+ var cl_LF eps = scale_float(b,-(sintC)d-10);
loop {
var cl_LF new_sum = sum + LF_to_LF(b/(cl_I)i,actuallen); // (+ sum (/ b i))
if (new_sum == sum) // = sum ?
var cl_F y = x-cl_float(1,x);
if (zerop(y)) // y=0.0 -> y als Ergebnis
return y;
- var uintL d = float_digits(x);
+ var uintC d = float_digits(x);
var sintL e = float_exponent(y);
- if (e <= -(sintL)d) // e <= -d ?
+ if (e <= -(sintC)d) // e <= -d ?
return y; // ja -> y als Ergebnis
{ Mutable(cl_F,x);
var uintL k = 0; // Rekursionszähler k:=0
// x1 = (-1)^sign * 2^exponent * mantissa
if (zerop(x1_.mantissa))
break;
- var uintL lm = integer_length(x1_.mantissa);
+ var uintC lm = integer_length(x1_.mantissa);
var uintL me = cl_I_to_UL(- x1_.exponent);
var cl_I p;
- var uintL lq;
+ var uintC lq;
var cl_boolean last_step = cl_false;
if (lm >= me) { // |x'| >= 1/2 ?
p = x1_.sign; // 1 or -1
if (2*n >= lm)
last_step = cl_true;
}
- y = y + scale_float(cl_I_to_LF(p,len),-(sintL)lq);
+ y = y + scale_float(cl_I_to_LF(p,len),-(sintC)lq);
if (last_step)
break;
x = x * cl_exp_aux(-p,lq,len);
const cl_F pi (float_format_t f)
{
- floatformatcase((uintL)f
+ floatformatcase((uintC)f
, return cl_SF_pi;
, return cl_FF_pi;
, return cl_DF_pi;
z = -z;
} else {
// cos(r) berechnen:
- if (zerop(r) || (float_exponent(r) <= (-(sintL)float_digits(r))>>1))
+ if (zerop(r) || (float_exponent(r) <= (-(sintC)float_digits(r))>>1))
z = cl_float(1,x); // cos(r) = 1.0
else
z = cl_float(sqrt(1 - y),x); // sqrt(1-y)
z = cl_float(r*sqrt(y),x);
} else {
// cos(r) berechnen:
- if (zerop(r) || (float_exponent(r) <= (-(sintL)float_digits(r))>>1))
+ if (zerop(r) || (float_exponent(r) <= (-(sintC)float_digits(r))>>1))
z = cl_float(1,x); // cos(r) = 1.0
else
z = cl_float(sqrt(1 - square(r)*y),x); // sqrt(1-r^2*y)
} else
#endif
if ((TheLfloat(x)->len >= 500)
- && (float_exponent(x) > (-(sintL)float_digits(x))>>1)) {
+ && (float_exponent(x) > (-(sintC)float_digits(x))>>1)) {
// verwende exp(x), schneller als cl_coshsinh_ratseries
// (aber nur bei 0 > e > -d/2, denn wir müssen, um
// Auslöschung zu verhindern, |e| Bits dazunehmen)
if (zerop(x))
return cl_float(1,x);
- var uintL d = float_digits(x);
+ var uintC d = float_digits(x);
var sintL e = float_exponent(x);
- if (e <= (1-(sintL)d)>>1) // e <= (1-d)/2 <==> e <= -ceiling((d-1)/2) ?
+ if (e <= (1-(sintC)d)>>1) // e <= (1-d)/2 <==> e <= -ceiling((d-1)/2) ?
return cl_float(1,x); // ja -> 1.0 als Ergebnis
{ Mutable(cl_F,x);
// Bei e <= -1-limit_slope*floor(sqrt(d)) kann die Potenzreihe
if (zerop(x))
return x;
var uintL actuallen = TheLfloat(x)->len;
- var uintL d = float_digits(x);
+ var uintC d = float_digits(x);
var sintL e = float_exponent(x);
- if (e <= (1-(sintL)d)>>1) // e <= (1-d)/2 <==> e <= -ceiling((d-1)/2) ?
+ if (e <= (1-(sintC)d)>>1) // e <= (1-d)/2 <==> e <= -ceiling((d-1)/2) ?
return square(x); // ja -> x^2 als Ergebnis
{ Mutable(cl_LF,x);
var sintL ee = e;
b = round1(round1(The(cl_LF)(b*a)),(cl_I)((i+1)*(i+2)));
i = i+2;
}
- powser_value = scale_float(cl_float(sum,x),-d);
+ powser_value = scale_float(cl_float(sum,x),-(sintC)d);
} else if (actuallen <= 7) { // Break-even-Point before extendsqrt: N<=6
// naive2:
// floating-point representation
// naive3:
// floating-point representation with smooth precision reduction
var cl_LF b = x; // b := x
- var cl_LF eps = scale_float(b,-(sintL)d-10);
+ var cl_LF eps = scale_float(b,-(sintC)d-10);
var cl_LF sum = cl_float(0,x); // sum := (float 0 x)
loop {
var cl_LF new_sum = sum + LF_to_LF(b,actuallen);
if (zerop(x))
return cl_float(1,x);
- var uintL d = float_digits(x);
+ var uintC d = float_digits(x);
var sintL e = float_exponent(x);
- if (e <= (-(sintL)d)>>1) // e <= (-d)/2 <==> e <= -ceiling(d/2) ?
+ if (e <= (-(sintC)d)>>1) // e <= (-d)/2 <==> e <= -ceiling(d/2) ?
return cl_float(1,x); // ja -> 1.0 als Ergebnis
{ Mutable(cl_F,x);
// Bei e <= -1-limit_slope*floor(sqrt(d)) kann die Potenzreihe
if (zerop(x))
return x;
var uintL actuallen = TheLfloat(x)->len;
- var uintL d = float_digits(x);
+ var uintC d = float_digits(x);
var sintL e = float_exponent(x);
- if (e <= (-(sintL)d)>>1) // e <= (-d)/2 <==> e <= -ceiling(d/2) ?
+ if (e <= (-(sintC)d)>>1) // e <= (-d)/2 <==> e <= -ceiling(d/2) ?
return square(x); // ja -> x^2 als Ergebnis
{ Mutable(cl_LF,x);
var sintL ee = e;
b = round1(round1(The(cl_LF)(b*a)),(cl_I)((i+1)*(i+2)));
i = i+2;
}
- powser_value = scale_float(cl_float(sum,x),-d);
+ powser_value = scale_float(cl_float(sum,x),-(sintC)d);
} else if (actuallen <= 7) { // Break-even-Point before extendsqrt: N<=6
// naive2:
// floating-point representation
// naive3:
// floating-point representation with smooth precision reduction
var cl_LF b = x; // b := x
- var cl_LF eps = scale_float(b,-(sintL)d-10);
+ var cl_LF eps = scale_float(b,-(sintC)d-10);
var cl_LF sum = cl_float(0,x); // sum := (float 0 x)
loop {
var cl_LF new_sum = sum + LF_to_LF(b,actuallen);
// cl_exp_aux(p,lq,len) liefert die Zahl exp(p/2^lq) mit len Digits.
// 0 < |p| < 2^lq.
// Es sollte |p|^2 < 2^lq sein, sonst ist das nicht effizient.
-extern const cl_LF cl_exp_aux (const cl_I& p, uintL lq, uintC len);
+extern const cl_LF cl_exp_aux (const cl_I& p, uintC lq, uintC len);
// cl_cossin_aux(p,lq,len) liefert cos(p/2^lq) und sin(p/2^lq) mit len Digits.
// 0 < |p| < 2^lq.
cl_LF_cos_sin_t (const cl_LF& u, const cl_LF& v) : cos (u), sin (v) {}
cl_LF_cos_sin_t () {}
};
-extern const cl_LF_cos_sin_t cl_cossin_aux (const cl_I& p, uintL lq, uintC len);
+extern const cl_LF_cos_sin_t cl_cossin_aux (const cl_I& p, uintC lq, uintC len);
// cl_coshsinh_aux(p,lq,len) liefert cosh(p/2^lq) und sinh(p/2^lq) mit len Digits.
// 0 < |p| < 2^lq.
cl_LF_cosh_sinh_t (const cl_LF& u, const cl_LF& v) : cosh (u), sinh (v) {}
cl_LF_cosh_sinh_t () {}
};
-extern const cl_LF_cosh_sinh_t cl_coshsinh_aux (const cl_I& p, uintL lq, uintC len);
+extern const cl_LF_cosh_sinh_t cl_coshsinh_aux (const cl_I& p, uintC lq, uintC len);
// atanhx(x) liefert zu einem Float x (betragsmäßig <1/2) atanh(x) als Float.
extern const cl_F atanhx (const cl_F& x);
const cl_F zeta (int s, float_format_t f)
{
- floatformatcase((uintL)f
+ floatformatcase((uintC)f
, return cl_LF_to_SF(zeta(s,LF_minlen));
, return cl_LF_to_FF(zeta(s,LF_minlen));
, return cl_LF_to_DF(zeta(s,LF_minlen));
{
var uintC actuallen = len + 1;
var cl_I m2 = m*m+1;
- var uintL N = (uintL)(0.69314718*intDsize*actuallen/::log(double_approx(m2))) + 1;
+ var uintC N = (uintC)(0.69314718*intDsize*actuallen/::log(double_approx(m2))) + 1;
CL_ALLOCA_STACK;
var cl_I* pv = (cl_I*) cl_alloca(N*sizeof(cl_I));
var cl_I* qv = (cl_I*) cl_alloca(N*sizeof(cl_I));
- var uintL n;
+ var uintC n;
new (&pv[0]) cl_I (m);
new (&qv[0]) cl_I (m2);
for (n = 1; n < N; n++) {
{
var uintC actuallen = len + 1;
var cl_I m2 = m*m;
- var uintL N = (uintL)(0.69314718*intDsize/2*actuallen/::log(double_approx(m))) + 1;
+ var uintC N = (uintC)(0.69314718*intDsize/2*actuallen/::log(double_approx(m))) + 1;
CL_ALLOCA_STACK;
var cl_I* bv = (cl_I*) cl_alloca(N*sizeof(cl_I));
var cl_I* qv = (cl_I*) cl_alloca(N*sizeof(cl_I));
- var uintL n;
+ var uintC n;
for (n = 0; n < N; n++) {
new (&bv[n]) cl_I ((cl_I)(2*n+1));
new (&qv[n]) cl_I (n==0 ? m : m2);
// Every summand gives 0.6 new decimal digits in precision.
// The sum is best evaluated using fixed-point arithmetic,
// so that the precision is reduced for the later summands.
- var uintL actuallen = len + 2; // 2 Schutz-Digits
- var sintL scale = intDsize*actuallen;
+ var uintC actuallen = len + 2; // 2 Schutz-Digits
+ var sintC scale = intDsize*actuallen;
var cl_I sum = 0;
var cl_I n = 0;
var cl_I factor = ash(1,scale);
{
// Same formula as above, using a binary splitting evaluation.
// See [Borwein, Borwein, section 10.2.3].
- var uintL actuallen = len + 2; // 2 Schutz-Digits
+ var uintC actuallen = len + 2; // 2 Schutz-Digits
// Evaluate a sum(0 <= n < N, a(n)/b(n) * (p(0)...p(n))/(q(0)...q(n)))
// with appropriate N, and
// a(n) = 1, b(n) = 2*n+1,
// p(n) = n for n>0, q(n) = 2*(2*n+1) for n>0.
- var uintL N = (intDsize/2)*actuallen;
+ var uintC N = (intDsize/2)*actuallen;
// 4^-N <= 2^(-intDsize*actuallen).
CL_ALLOCA_STACK;
var cl_I* bv = (cl_I*) cl_alloca(N*sizeof(cl_I));
var cl_I* pv = (cl_I*) cl_alloca(N*sizeof(cl_I));
var cl_I* qv = (cl_I*) cl_alloca(N*sizeof(cl_I));
- var uintL n;
+ var uintC n;
init1(cl_I, bv[0]) (1);
init1(cl_I, pv[0]) (1);
init1(cl_I, qv[0]) (1);
{
// We compute f(x) classically and g(x) using the partial sums of f(x).
var uintC actuallen = len+2; // 2 Schutz-Digits
- var uintL x = (uintL)(0.693148*intDsize*actuallen)+1;
- var uintL N = (uintL)(2.718281828*x);
+ var uintC x = (uintC)(0.693148*intDsize*actuallen)+1;
+ var uintC N = (uintC)(2.718281828*x);
var cl_LF fterm = cl_I_to_LF(1,actuallen);
var cl_LF fsum = fterm;
var cl_LF gterm = fterm;
var cl_LF gsum = gterm;
- var uintL n;
+ var uintC n;
// After n loops
// fterm = x^n/n!, fsum = 1 + x/1! + ... + x^n/n!,
// gterm = S_n*x^n/n!, gsum = S_0*x^0/0! + ... + S_n*x^n/n!.
const cl_LF compute_catalanconst_expintegral2 (uintC len)
{
var uintC actuallen = len+2; // 2 Schutz-Digits
- var uintL x = (uintL)(0.693148*intDsize*actuallen)+1;
- var uintL N = (uintL)(2.718281828*x);
+ var uintC x = (uintC)(0.693148*intDsize*actuallen)+1;
+ var uintC N = (uintC)(2.718281828*x);
CL_ALLOCA_STACK;
var cl_pqd_series_term* args = (cl_pqd_series_term*) cl_alloca(N*sizeof(cl_pqd_series_term));
- var uintL n;
+ var uintC n;
for (n = 0; n < N; n++) {
if (n==0) {
init1(cl_I, args[n].p) (1);
const cl_LF compute_catalanconst_cvz1 (uintC len)
{
var uintC actuallen = len+2; // 2 Schutz-Digits
- var uintL N = (uintL)(0.39321985*intDsize*actuallen)+1;
+ var uintC N = (uintC)(0.39321985*intDsize*actuallen)+1;
#if 0
var cl_LF fterm = cl_I_to_LF(2*(cl_I)N*(cl_I)N,actuallen);
var cl_LF fsum = fterm;
var cl_LF gterm = fterm;
var cl_LF gsum = gterm;
- var uintL n;
+ var uintC n;
// After n loops
// fterm = (N+n)!N/(2n+2)!(N-n-1)!*2^(2n+2), fsum = ... + fterm,
// gterm = S_n*fterm, gsum = ... + gterm.
var cl_I fsum = fterm;
var cl_LF gterm = cl_I_to_LF(fterm,actuallen);
var cl_LF gsum = gterm;
- var uintL n;
+ var uintC n;
// After n loops
// fterm = (N+n)!N/(2n+2)!(N-n-1)!*2^(2n+2), fsum = ... + fterm,
// gterm = S_n*fterm, gsum = ... + gterm.
const cl_LF compute_catalanconst_cvz2 (uintC len)
{
var uintC actuallen = len+2; // 2 Schutz-Digits
- var uintL N = (uintL)(0.39321985*intDsize*actuallen)+1;
+ var uintC N = (uintC)(0.39321985*intDsize*actuallen)+1;
CL_ALLOCA_STACK;
var cl_pqd_series_term* args = (cl_pqd_series_term*) cl_alloca(N*sizeof(cl_pqd_series_term));
- var uintL n;
+ var uintC n;
for (n = 0; n < N; n++) {
init1(cl_I, args[n].p) (2*(cl_I)(N-n)*(cl_I)(N+n));
init1(cl_I, args[n].q) ((cl_I)(2*n+1)*(cl_I)(n+1));
// by a power series evaluation brings 20% speedup, even more for small lengths.
#define TRIVIAL_SPEEDUP
-const cl_LF_cosh_sinh_t cl_coshsinh_aux (const cl_I& p, uintL lq, uintC len)
+const cl_LF_cosh_sinh_t cl_coshsinh_aux (const cl_I& p, uintC lq, uintC len)
{
{ Mutable(cl_I,p);
- var uintL lp = integer_length(p); // now |p| < 2^lp.
+ var uintC lp = integer_length(p); // now |p| < 2^lp.
if (!(lp <= lq)) cl_abort();
lp = lq - lp; // now |p/2^lq| < 2^-lp.
// Minimize lq (saves computation time).
{
- var uintL lp2 = ord2(p);
+ var uintC lp2 = ord2(p);
if (lp2 > 0) {
p = p >> lp2;
lq = lq - lp2;
// Third approximation:
// N2 = ceiling(M*log(2)/(log(N1)-1+lp*log(2))), slightly too large.
// N = N2+2, two more terms for safety.
- var uintL N0 = intDsize*actuallen;
- var uintL N1 = (uintL)(0.693147*intDsize*actuallen/(::log((double)N0)-1.0+0.693148*lp));
- var uintL N2 = (uintL)(0.693148*intDsize*actuallen/(::log((double)N1)-1.0+0.693147*lp))+1;
- var uintL N = N2+2;
+ var uintC N0 = intDsize*actuallen;
+ var uintC N1 = (uintC)(0.693147*intDsize*actuallen/(::log((double)N0)-1.0+0.693148*lp));
+ var uintC N2 = (uintC)(0.693148*intDsize*actuallen/(::log((double)N1)-1.0+0.693147*lp))+1;
+ var uintC N = N2+2;
N = ceiling(N,2);
CL_ALLOCA_STACK;
var cl_I* pv = (cl_I*) cl_alloca(N*sizeof(cl_I));
var cl_I* qv = (cl_I*) cl_alloca(N*sizeof(cl_I));
- var uintL* qsv = (uintL*) cl_alloca(N*sizeof(uintL));
- var uintL n;
+ var uintC* qsv = (uintC*) cl_alloca(N*sizeof(uintC));
+ var uintC n;
var cl_I p2 = square(p);
var cl_LF sinhsum;
{
// by a power series evaluation brings 20% speedup, even more for small lengths.
#define TRIVIAL_SPEEDUP
-const cl_LF_cos_sin_t cl_cossin_aux (const cl_I& p, uintL lq, uintC len)
+const cl_LF_cos_sin_t cl_cossin_aux (const cl_I& p, uintC lq, uintC len)
{
{ Mutable(cl_I,p);
- var uintL lp = integer_length(p); // now |p| < 2^lp.
+ var uintC lp = integer_length(p); // now |p| < 2^lp.
if (!(lp <= lq)) cl_abort();
lp = lq - lp; // now |p/2^lq| < 2^-lp.
// Minimize lq (saves computation time).
{
- var uintL lp2 = ord2(p);
+ var uintC lp2 = ord2(p);
if (lp2 > 0) {
p = p >> lp2;
lq = lq - lp2;
// Third approximation:
// N2 = ceiling(M*log(2)/(log(N1)-1+lp*log(2))), slightly too large.
// N = N2+2, two more terms for safety.
- var uintL N0 = intDsize*actuallen;
- var uintL N1 = (uintL)(0.693147*intDsize*actuallen/(::log((double)N0)-1.0+0.693148*lp));
- var uintL N2 = (uintL)(0.693148*intDsize*actuallen/(::log((double)N1)-1.0+0.693147*lp))+1;
- var uintL N = N2+2;
+ var uintC N0 = intDsize*actuallen;
+ var uintC N1 = (uintC)(0.693147*intDsize*actuallen/(::log((double)N0)-1.0+0.693148*lp));
+ var uintC N2 = (uintC)(0.693148*intDsize*actuallen/(::log((double)N1)-1.0+0.693147*lp))+1;
+ var uintC N = N2+2;
N = ceiling(N,2);
CL_ALLOCA_STACK;
var cl_I* pv = (cl_I*) cl_alloca(N*sizeof(cl_I));
var cl_I* qv = (cl_I*) cl_alloca(N*sizeof(cl_I));
- var uintL* qsv = (uintL*) cl_alloca(N*sizeof(uintL));
- var uintL n;
+ var uintC* qsv = (uintC*) cl_alloca(N*sizeof(uintC));
+ var uintC n;
var cl_I p2 = -square(p);
var cl_LF sinsum;
{
// to more than double the floating-point precision because of the large
// extinction which takes place. But luckily we compute with integers.
var uintC actuallen = len+1; // 1 Schutz-Digit
- var uintL z = (uintL)(0.693148*intDsize*actuallen)+1;
- var uintL N = (uintL)(3.591121477*z);
+ var uintC z = (uintC)(0.693148*intDsize*actuallen)+1;
+ var uintC N = (uintC)(3.591121477*z);
CL_ALLOCA_STACK;
var cl_I* bv = (cl_I*) cl_alloca(N*sizeof(cl_I));
var cl_I* pv = (cl_I*) cl_alloca(N*sizeof(cl_I));
var cl_I* qv = (cl_I*) cl_alloca(N*sizeof(cl_I));
- var uintL n;
+ var uintC n;
for (n = 0; n < N; n++) {
init1(cl_I, bv[n]) (n+1);
init1(cl_I, pv[n]) (n==0 ? (cl_I)z : -(cl_I)z);
// each.
// We compute f(x) classically and g(x) using the partial sums of f(x).
var uintC actuallen = len+2; // 2 Schutz-Digits
- var uintL x = (uintL)(0.693148*intDsize*actuallen)+1;
- var uintL N = (uintL)(2.718281828*x);
+ var uintC x = (uintC)(0.693148*intDsize*actuallen)+1;
+ var uintC N = (uintC)(2.718281828*x);
var cl_LF one = cl_I_to_LF(1,actuallen);
var cl_LF fterm = one;
var cl_LF fsum = fterm;
var cl_LF gterm = cl_I_to_LF(0,actuallen);
var cl_LF gsum = gterm;
- var uintL n;
+ var uintC n;
// After n loops
// fterm = x^n/n!, fsum = 1 + x/1! + ... + x^n/n!,
// gterm = H_n*x^n/n!, gsum = H_1*x/1! + ... + H_n*x^n/n!.
const cl_LF compute_eulerconst_expintegral2 (uintC len)
{
var uintC actuallen = len+2; // 2 Schutz-Digits
- var uintL x = (uintL)(0.693148*intDsize*actuallen)+1;
- var uintL N = (uintL)(2.718281828*x);
+ var uintC x = (uintC)(0.693148*intDsize*actuallen)+1;
+ var uintC N = (uintC)(2.718281828*x);
CL_ALLOCA_STACK;
var cl_pqd_series_term* args = (cl_pqd_series_term*) cl_alloca(N*sizeof(cl_pqd_series_term));
- var uintL n;
+ var uintC n;
for (n = 0; n < N; n++) {
init1(cl_I, args[n].p) (x);
init1(cl_I, args[n].q) (n+1);
{
// We compute f(x) classically and g(x) using the partial sums of f(x).
var uintC actuallen = len+1; // 1 Schutz-Digit
- var uintL sx = (uintL)(0.25*0.693148*intDsize*actuallen)+1;
- var uintL N = (uintL)(3.591121477*sx);
+ var uintC sx = (uintC)(0.25*0.693148*intDsize*actuallen)+1;
+ var uintC N = (uintC)(3.591121477*sx);
var cl_I x = square((cl_I)sx);
- var cl_LF eps = scale_float(cl_I_to_LF(1,LF_minlen),-(sintL)(sx*2.88539+10));
+ var cl_LF eps = scale_float(cl_I_to_LF(1,LF_minlen),-(sintC)(sx*2.88539+10));
var cl_LF fterm = cl_I_to_LF(1,actuallen);
var cl_LF fsum = fterm;
var cl_LF gterm = cl_I_to_LF(0,actuallen);
var cl_LF gsum = gterm;
- var uintL n;
+ var uintC n;
// After n loops
// fterm = x^n/n!^2, fsum = 1 + x/1!^2 + ... + x^n/n!^2,
// gterm = H_n*x^n/n!^2, gsum = H_1*x/1!^2 + ... + H_n*x^n/n!^2.
// (Because HD_n grows like exp(n), hence HN_n grows like exp(n) as
// well, and we store all HN_n values in an array!)
var uintC actuallen = len+1; // 1 Schutz-Digit
- var uintL sx = (uintL)(0.25*0.693148*intDsize*actuallen)+1;
- var uintL N = (uintL)(3.591121477*sx);
+ var uintC sx = (uintC)(0.25*0.693148*intDsize*actuallen)+1;
+ var uintC N = (uintC)(3.591121477*sx);
var cl_I x = square((cl_I)sx);
CL_ALLOCA_STACK;
var cl_I* av = (cl_I*) cl_alloca(N*sizeof(cl_I));
var cl_I* pv = (cl_I*) cl_alloca(N*sizeof(cl_I));
var cl_I* qv = (cl_I*) cl_alloca(N*sizeof(cl_I));
- var uintL n;
+ var uintC n;
// Evaluate f(x).
init1(cl_I, pv[0]) (1);
init1(cl_I, qv[0]) (1);
const cl_LF compute_eulerconst_besselintegral3 (uintC len)
{
var uintC actuallen = len+1; // 1 Schutz-Digit
- var uintL sx = (uintL)(0.25*0.693148*intDsize*actuallen)+1;
- var uintL N = (uintL)(3.591121477*sx);
+ var uintC sx = (uintC)(0.25*0.693148*intDsize*actuallen)+1;
+ var uintC N = (uintC)(3.591121477*sx);
var cl_I x = square((cl_I)sx);
CL_ALLOCA_STACK;
var cl_I* pv = (cl_I*) cl_alloca(N*sizeof(cl_I));
var cl_I* qv = (cl_I*) cl_alloca(N*sizeof(cl_I));
- var uintL n;
+ var uintC n;
// Evaluate f(x).
init1(cl_I, pv[0]) (1);
init1(cl_I, qv[0]) (1);
const cl_LF compute_eulerconst_besselintegral4 (uintC len)
{
var uintC actuallen = len+2; // 2 Schutz-Digits
- var uintL sx = (uintL)(0.25*0.693148*intDsize*actuallen)+1;
- var uintL N = (uintL)(3.591121477*sx);
+ var uintC sx = (uintC)(0.25*0.693148*intDsize*actuallen)+1;
+ var uintC N = (uintC)(3.591121477*sx);
var cl_I x = square((cl_I)sx);
CL_ALLOCA_STACK;
var cl_pqd_series_term* args = (cl_pqd_series_term*) cl_alloca(N*sizeof(cl_pqd_series_term));
- var uintL n;
+ var uintC n;
for (n = 0; n < N; n++) {
init1(cl_I, args[n].p) (x);
init1(cl_I, args[n].q) (square((cl_I)(n+1)));
// Third approximation:
// N2 = ceiling(M*log(2)/(log(N1)-1)), slightly too large.
// N = N2+2, two more terms for safety.
- var uintL N0 = intDsize*actuallen;
- var uintL N1 = (uintL)(0.693147*intDsize*actuallen/(::log((double)N0)-1.0));
- var uintL N2 = (uintL)(0.693148*intDsize*actuallen/(::log((double)N1)-1.0))+1;
- var uintL N = N2+2;
+ var uintC N0 = intDsize*actuallen;
+ var uintC N1 = (uintC)(0.693147*intDsize*actuallen/(::log((double)N0)-1.0));
+ var uintC N2 = (uintC)(0.693148*intDsize*actuallen/(::log((double)N1)-1.0))+1;
+ var uintC N = N2+2;
CL_ALLOCA_STACK;
var cl_I* qv = (cl_I*) cl_alloca(N*sizeof(cl_I));
- var uintL* qsv = (uintL*) cl_alloca(N*sizeof(uintL));
- var uintL n;
+ var uintC* qsv = (uintC*) cl_alloca(N*sizeof(uintC));
+ var uintC n;
for (n = 0; n < N; n++) {
init1(cl_I, qv[n]) (n==0 ? 1 : n);
}
namespace cln {
-const cl_LF cl_exp_aux (const cl_I& p, uintL lq, uintC len)
+const cl_LF cl_exp_aux (const cl_I& p, uintC lq, uintC len)
{
{ Mutable(cl_I,p);
- var uintL lp = integer_length(p); // now |p| < 2^lp.
+ var uintC lp = integer_length(p); // now |p| < 2^lp.
if (!(lp <= lq)) cl_abort();
lp = lq - lp; // now |p/2^lq| < 2^-lp.
// Minimize lq (saves computation time).
{
- var uintL lp2 = ord2(p);
+ var uintC lp2 = ord2(p);
if (lp2 > 0) {
p = p >> lp2;
lq = lq - lp2;
// Third approximation:
// N2 = ceiling(M*log(2)/(log(N1)-1+lp*log(2))), slightly too large.
// N = N2+2, two more terms for safety.
- var uintL N0 = intDsize*actuallen;
- var uintL N1 = (uintL)(0.693147*intDsize*actuallen/(::log((double)N0)-1.0+0.693148*lp));
- var uintL N2 = (uintL)(0.693148*intDsize*actuallen/(::log((double)N1)-1.0+0.693147*lp))+1;
- var uintL N = N2+2;
+ var uintC N0 = intDsize*actuallen;
+ var uintC N1 = (uintC)(0.693147*intDsize*actuallen/(::log((double)N0)-1.0+0.693148*lp));
+ var uintC N2 = (uintC)(0.693148*intDsize*actuallen/(::log((double)N1)-1.0+0.693147*lp))+1;
+ var uintC N = N2+2;
CL_ALLOCA_STACK;
var cl_I* pv = (cl_I*) cl_alloca(N*sizeof(cl_I));
var cl_I* qv = (cl_I*) cl_alloca(N*sizeof(cl_I));
- var uintL* qsv = (uintL*) cl_alloca(N*sizeof(uintL));
- var uintL n;
+ var uintC* qsv = (uintC*) cl_alloca(N*sizeof(uintC));
+ var uintC n;
init1(cl_I, pv[0]) (1);
init1(cl_I, qv[0]) (1);
for (n = 1; n < N; n++) {
// (/ (expt a 2) t)
// )
var uintC actuallen = len + 1; // 1 Schutz-Digit
- var uintL uexp_limit = LF_exp_mid - intDsize*(uintL)len;
+ var uintC uexp_limit = LF_exp_mid - intDsize*len;
// Ein Long-Float ist genau dann betragsmäßig <2^-n, wenn
// sein Exponent < LF_exp_mid-n = uexp_limit ist.
var cl_LF a = cl_I_to_LF(1,actuallen);
// Hence,
// pi = AGM(1,1/sqrt(2))^2 * 1/(1/2 - sum(k even, 2^k*[....])).
var uintC actuallen = len + 1; // 1 Schutz-Digit
- var uintL uexp_limit = LF_exp_mid - intDsize*(uintL)len;
+ var uintC uexp_limit = LF_exp_mid - intDsize*len;
var cl_LF one = cl_I_to_LF(1,actuallen);
var cl_LF a = one;
var cl_LF wa = one;
// in precision.
// The sum is best evaluated using fixed-point arithmetic,
// so that the precision is reduced for the later summands.
- var uintL actuallen = len + 4; // 4 Schutz-Digits
- var sintL scale = intDsize*actuallen;
+ var uintC actuallen = len + 4; // 4 Schutz-Digits
+ var sintC scale = intDsize*actuallen;
static const cl_I A = "163096908";
static const cl_I B = "6541681608";
//static const cl_I J1 = "10939058860032000"; // 72*abs(J)
{
// Same formula as above, using a binary splitting evaluation.
// See [Borwein, Borwein, section 10.2.3].
- var uintL actuallen = len + 4; // 4 Schutz-Digits
+ var uintC actuallen = len + 4; // 4 Schutz-Digits
static const cl_I A = "163096908";
static const cl_I B = "6541681608";
static const cl_I J1 = "10939058860032000"; // 72*abs(J)
// a(n) = A+n*B, b(n) = 1,
// p(n) = -(6n-5)(2n-1)(6n-1) for n>0,
// q(n) = 72*|J|*n^3 for n>0.
- var const uintL n_slope = (uintL)(intDsize*32*0.02122673)+1;
+ var const uintC n_slope = (uintC)(intDsize*32*0.02122673)+1;
// n_slope >= 32*intDsize*log(2)/log(|J|), normally n_slope = 22.
- var uintL N = (n_slope*actuallen)/32 + 1;
+ var uintC N = (n_slope*actuallen)/32 + 1;
// N > intDsize*log(2)/log(|J|) * actuallen, hence
// |J|^-N < 2^(-intDsize*actuallen).
CL_ALLOCA_STACK;
var cl_I* av = (cl_I*) cl_alloca(N*sizeof(cl_I));
var cl_I* pv = (cl_I*) cl_alloca(N*sizeof(cl_I));
var cl_I* qv = (cl_I*) cl_alloca(N*sizeof(cl_I));
- var uintL* qsv = (uintL*) cl_alloca(N*sizeof(uintL));
- var uintL n;
+ var uintC* qsv = (uintC*) cl_alloca(N*sizeof(uintC));
+ var uintC n;
for (n = 0; n < N; n++) {
init1(cl_I, av[n]) (A+n*B);
if (n==0) {
// and T = B*Q*S (all integers). On entry N1 < N2.
// P will not be computed if a NULL pointer is passed.
-const cl_LF eval_rational_series (uintL N, const cl_rational_series& args, uintC len)
+const cl_LF eval_rational_series (uintC N, const cl_rational_series& args, uintC len)
{
if (args.pv)
if (args.qv)
// and T = B*Q*S (all integers). On entry N1 < N2.
// P will not be computed if a NULL pointer is passed.
-static inline void eval__series_aux (uintL N1, uintL N2,
+static inline void eval__series_aux (uintC N1, uintC N2,
const cl__series& args,
cl_I* T)
{
*T = N2-N1;
}
-const cl_LF eval_rational_series (uintL N, const cl__series& args, uintC len)
+const cl_LF eval_rational_series (uintC N, const cl__series& args, uintC len)
{
if (N==0)
return cl_I_to_LF(0,len);
// and T = B*Q*S (all integers). On entry N1 < N2.
// P will not be computed if a NULL pointer is passed.
-static void eval_a_series_aux (uintL N1, uintL N2,
+static void eval_a_series_aux (uintC N1, uintC N2,
const cl_a_series& args,
cl_I* T)
{
break;
}
default: {
- var uintL Nm = (N1+N2)/2; // midpoint
+ var uintC Nm = (N1+N2)/2; // midpoint
// Compute left part.
var cl_I LT;
eval_a_series_aux(N1,Nm,args,<);
}
}
-const cl_LF eval_rational_series (uintL N, const cl_a_series& args, uintC len)
+const cl_LF eval_rational_series (uintC N, const cl_a_series& args, uintC len)
{
if (N==0)
return cl_I_to_LF(0,len);
// and T = B*Q*S (all integers). On entry N1 < N2.
// P will not be computed if a NULL pointer is passed.
-static void eval_ab_series_aux (uintL N1, uintL N2,
+static void eval_ab_series_aux (uintC N1, uintC N2,
const cl_ab_series& args,
cl_I* B, cl_I* T)
{
break;
}
default: {
- var uintL Nm = (N1+N2)/2; // midpoint
+ var uintC Nm = (N1+N2)/2; // midpoint
// Compute left part.
var cl_I LB, LT;
eval_ab_series_aux(N1,Nm,args,&LB,<);
}
}
-const cl_LF eval_rational_series (uintL N, const cl_ab_series& args, uintC len)
+const cl_LF eval_rational_series (uintC N, const cl_ab_series& args, uintC len)
{
if (N==0)
return cl_I_to_LF(0,len);
// and T = B*Q*S (all integers). On entry N1 < N2.
// P will not be computed if a NULL pointer is passed.
-static void eval_b_series_aux (uintL N1, uintL N2,
+static void eval_b_series_aux (uintC N1, uintC N2,
const cl_b_series& args,
cl_I* B, cl_I* T)
{
break;
}
default: {
- var uintL Nm = (N1+N2)/2; // midpoint
+ var uintC Nm = (N1+N2)/2; // midpoint
// Compute left part.
var cl_I LB, LT;
eval_b_series_aux(N1,Nm,args,&LB,<);
}
}
-const cl_LF eval_rational_series (uintL N, const cl_b_series& args, uintC len)
+const cl_LF eval_rational_series (uintC N, const cl_b_series& args, uintC len)
{
if (N==0)
return cl_I_to_LF(0,len);
// and T = B*Q*S (all integers). On entry N1 < N2.
// P will not be computed if a NULL pointer is passed.
-static void eval_p_series_aux (uintL N1, uintL N2,
+static void eval_p_series_aux (uintC N1, uintC N2,
const cl_p_series& args,
cl_I* P, cl_I* T)
{
break;
}
default: {
- var uintL Nm = (N1+N2)/2; // midpoint
+ var uintC Nm = (N1+N2)/2; // midpoint
// Compute left part.
var cl_I LP, LT;
eval_p_series_aux(N1,Nm,args,&LP,<);
}
}
-const cl_LF eval_rational_series (uintL N, const cl_p_series& args, uintC len)
+const cl_LF eval_rational_series (uintC N, const cl_p_series& args, uintC len)
{
if (N==0)
return cl_I_to_LF(0,len);
// and T = B*Q*S (all integers). On entry N1 < N2.
// P will not be computed if a NULL pointer is passed.
-static void eval_pa_series_aux (uintL N1, uintL N2,
+static void eval_pa_series_aux (uintC N1, uintC N2,
const cl_pa_series& args,
cl_I* P, cl_I* T)
{
break;
}
default: {
- var uintL Nm = (N1+N2)/2; // midpoint
+ var uintC Nm = (N1+N2)/2; // midpoint
// Compute left part.
var cl_I LP, LT;
eval_pa_series_aux(N1,Nm,args,&LP,<);
}
}
-const cl_LF eval_rational_series (uintL N, const cl_pa_series& args, uintC len)
+const cl_LF eval_rational_series (uintC N, const cl_pa_series& args, uintC len)
{
if (N==0)
return cl_I_to_LF(0,len);
// and T = B*Q*S (all integers). On entry N1 < N2.
// P will not be computed if a NULL pointer is passed.
-static void eval_pab_series_aux (uintL N1, uintL N2,
+static void eval_pab_series_aux (uintC N1, uintC N2,
const cl_pab_series& args,
cl_I* P, cl_I* B, cl_I* T)
{
break;
}
default: {
- var uintL Nm = (N1+N2)/2; // midpoint
+ var uintC Nm = (N1+N2)/2; // midpoint
// Compute left part.
var cl_I LP, LB, LT;
eval_pab_series_aux(N1,Nm,args,&LP,&LB,<);
}
}
-const cl_LF eval_rational_series (uintL N, const cl_pab_series& args, uintC len)
+const cl_LF eval_rational_series (uintC N, const cl_pab_series& args, uintC len)
{
if (N==0)
return cl_I_to_LF(0,len);
// and T = B*Q*S (all integers). On entry N1 < N2.
// P will not be computed if a NULL pointer is passed.
-static void eval_pb_series_aux (uintL N1, uintL N2,
+static void eval_pb_series_aux (uintC N1, uintC N2,
const cl_pb_series& args,
cl_I* P, cl_I* B, cl_I* T)
{
break;
}
default: {
- var uintL Nm = (N1+N2)/2; // midpoint
+ var uintC Nm = (N1+N2)/2; // midpoint
// Compute left part.
var cl_I LP, LB, LT;
eval_pb_series_aux(N1,Nm,args,&LP,&LB,<);
}
}
-const cl_LF eval_rational_series (uintL N, const cl_pb_series& args, uintC len)
+const cl_LF eval_rational_series (uintC N, const cl_pb_series& args, uintC len)
{
if (N==0)
return cl_I_to_LF(0,len);
// and T = B*Q*S (all integers). On entry N1 < N2.
// P will not be computed if a NULL pointer is passed.
-static void eval_pq_series_aux (uintL N1, uintL N2,
+static void eval_pq_series_aux (uintC N1, uintC N2,
const cl_pq_series& args,
cl_I* P, cl_I* Q, cl_I* T)
{
break;
}
default: {
- var uintL Nm = (N1+N2)/2; // midpoint
+ var uintC Nm = (N1+N2)/2; // midpoint
// Compute left part.
var cl_I LP, LQ, LT;
eval_pq_series_aux(N1,Nm,args,&LP,&LQ,<);
}
}
-static void eval_pqs_series_aux (uintL N1, uintL N2,
+static void eval_pqs_series_aux (uintC N1, uintC N2,
const cl_pq_series& args,
- cl_I* P, cl_I* Q, uintL* QS, cl_I* T)
+ cl_I* P, cl_I* Q, uintC* QS, cl_I* T)
{
switch (N2 - N1) {
case 0:
break;
}
default: {
- var uintL Nm = (N1+N2)/2; // midpoint
+ var uintC Nm = (N1+N2)/2; // midpoint
// Compute left part.
var cl_I LP, LQ, LT;
- var uintL LQS;
+ var uintC LQS;
eval_pqs_series_aux(N1,Nm,args,&LP,&LQ,&LQS,<);
// Compute right part.
var cl_I RP, RQ, RT;
- var uintL RQS;
+ var uintC RQS;
eval_pqs_series_aux(Nm,N2,args,(P?&RP:(cl_I*)0),&RQ,&RQS,&RT);
// Put together partial results.
if (P) { *P = LP*RP; }
}
}
-const cl_LF eval_rational_series (uintL N, const cl_pq_series& args, uintC len)
+const cl_LF eval_rational_series (uintC N, const cl_pq_series& args, uintC len)
{
if (N==0)
return cl_I_to_LF(0,len);
// Split qv[n] into qv[n]*2^qsv[n].
{
var cl_I* qp = args.qv;
- var uintL* qsp = args.qsv;
- for (var uintL n = 0; n < N; n++, qp++, qsp++) {
+ var uintC* qsp = args.qsv;
+ for (var uintC n = 0; n < N; n++, qp++, qsp++) {
// Pull out maximal power of 2 out of *qp = args.qv[n].
- var uintL qs = 0;
+ var uintC qs = 0;
if (!zerop(*qp)) {
qs = ord2(*qp);
if (qs > 0)
}
}
// Main computation.
- var uintL QS;
+ var uintC QS;
eval_pqs_series_aux(0,N,args,NULL,&Q,&QS,&T);
return cl_I_to_LF(T,len) / scale_float(cl_I_to_LF(Q,len),QS);
}
// and T = B*Q*S (all integers). On entry N1 < N2.
// P will not be computed if a NULL pointer is passed.
-static void eval_pqa_series_aux (uintL N1, uintL N2,
+static void eval_pqa_series_aux (uintC N1, uintC N2,
const cl_pqa_series& args,
cl_I* P, cl_I* Q, cl_I* T)
{
break;
}
default: {
- var uintL Nm = (N1+N2)/2; // midpoint
+ var uintC Nm = (N1+N2)/2; // midpoint
// Compute left part.
var cl_I LP, LQ, LT;
eval_pqa_series_aux(N1,Nm,args,&LP,&LQ,<);
}
}
-static void eval_pqsa_series_aux (uintL N1, uintL N2,
+static void eval_pqsa_series_aux (uintC N1, uintC N2,
const cl_pqa_series& args,
- cl_I* P, cl_I* Q, uintL* QS, cl_I* T)
+ cl_I* P, cl_I* Q, uintC* QS, cl_I* T)
{
switch (N2 - N1) {
case 0:
break;
}
default: {
- var uintL Nm = (N1+N2)/2; // midpoint
+ var uintC Nm = (N1+N2)/2; // midpoint
// Compute left part.
var cl_I LP, LQ, LT;
- var uintL LQS;
+ var uintC LQS;
eval_pqsa_series_aux(N1,Nm,args,&LP,&LQ,&LQS,<);
// Compute right part.
var cl_I RP, RQ, RT;
- var uintL RQS;
+ var uintC RQS;
eval_pqsa_series_aux(Nm,N2,args,(P?&RP:(cl_I*)0),&RQ,&RQS,&RT);
// Put together partial results.
if (P) { *P = LP*RP; }
}
}
-const cl_LF eval_rational_series (uintL N, const cl_pqa_series& args, uintC len)
+const cl_LF eval_rational_series (uintC N, const cl_pqa_series& args, uintC len)
{
if (N==0)
return cl_I_to_LF(0,len);
// Split qv[n] into qv[n]*2^qsv[n].
{
var cl_I* qp = args.qv;
- var uintL* qsp = args.qsv;
- for (var uintL n = 0; n < N; n++, qp++, qsp++) {
+ var uintC* qsp = args.qsv;
+ for (var uintC n = 0; n < N; n++, qp++, qsp++) {
// Pull out maximal power of 2 out of *qp = args.qv[n].
- var uintL qs = 0;
+ var uintC qs = 0;
if (!zerop(*qp)) {
qs = ord2(*qp);
if (qs > 0)
}
}
// Main computation.
- var uintL QS;
+ var uintC QS;
eval_pqsa_series_aux(0,N,args,NULL,&Q,&QS,&T);
return cl_I_to_LF(T,len) / scale_float(cl_I_to_LF(Q,len),QS);
}
// and T = B*Q*S (all integers). On entry N1 < N2.
// P will not be computed if a NULL pointer is passed.
-static void eval_pqab_series_aux (uintL N1, uintL N2,
+static void eval_pqab_series_aux (uintC N1, uintC N2,
const cl_pqab_series& args,
cl_I* P, cl_I* Q, cl_I* B, cl_I* T)
{
break;
}
default: {
- var uintL Nm = (N1+N2)/2; // midpoint
+ var uintC Nm = (N1+N2)/2; // midpoint
// Compute left part.
var cl_I LP, LQ, LB, LT;
eval_pqab_series_aux(N1,Nm,args,&LP,&LQ,&LB,<);
}
}
-static void eval_pqsab_series_aux (uintL N1, uintL N2,
+static void eval_pqsab_series_aux (uintC N1, uintC N2,
const cl_pqab_series& args,
- cl_I* P, cl_I* Q, uintL* QS, cl_I* B, cl_I* T)
+ cl_I* P, cl_I* Q, uintC* QS, cl_I* B, cl_I* T)
{
switch (N2 - N1) {
case 0:
break;
}
default: {
- var uintL Nm = (N1+N2)/2; // midpoint
+ var uintC Nm = (N1+N2)/2; // midpoint
// Compute left part.
var cl_I LP, LQ, LB, LT;
- var uintL LQS;
+ var uintC LQS;
eval_pqsab_series_aux(N1,Nm,args,&LP,&LQ,&LQS,&LB,<);
// Compute right part.
var cl_I RP, RQ, RB, RT;
- var uintL RQS;
+ var uintC RQS;
eval_pqsab_series_aux(Nm,N2,args,(P?&RP:(cl_I*)0),&RQ,&RQS,&RB,&RT);
// Put together partial results.
if (P) { *P = LP*RP; }
}
}
-const cl_LF eval_rational_series (uintL N, const cl_pqab_series& args, uintC len)
+const cl_LF eval_rational_series (uintC N, const cl_pqab_series& args, uintC len)
{
if (N==0)
return cl_I_to_LF(0,len);
// Split qv[n] into qv[n]*2^qsv[n].
{
var cl_I* qp = args.qv;
- var uintL* qsp = args.qsv;
- for (var uintL n = 0; n < N; n++, qp++, qsp++) {
+ var uintC* qsp = args.qsv;
+ for (var uintC n = 0; n < N; n++, qp++, qsp++) {
// Pull out maximal power of 2 out of *qp = args.qv[n].
- var uintL qs = 0;
+ var uintC qs = 0;
if (!zerop(*qp)) {
qs = ord2(*qp);
if (qs > 0)
}
}
// Main computation.
- var uintL QS;
+ var uintC QS;
eval_pqsab_series_aux(0,N,args,NULL,&Q,&QS,&B,&T);
return cl_I_to_LF(T,len) / scale_float(cl_I_to_LF(B*Q,len),QS);
}
// and T = B*Q*S (all integers). On entry N1 < N2.
// P will not be computed if a NULL pointer is passed.
-static void eval_pqb_series_aux (uintL N1, uintL N2,
+static void eval_pqb_series_aux (uintC N1, uintC N2,
const cl_pqb_series& args,
cl_I* P, cl_I* Q, cl_I* B, cl_I* T)
{
break;
}
default: {
- var uintL Nm = (N1+N2)/2; // midpoint
+ var uintC Nm = (N1+N2)/2; // midpoint
// Compute left part.
var cl_I LP, LQ, LB, LT;
eval_pqb_series_aux(N1,Nm,args,&LP,&LQ,&LB,<);
}
}
-static void eval_pqsb_series_aux (uintL N1, uintL N2,
+static void eval_pqsb_series_aux (uintC N1, uintC N2,
const cl_pqb_series& args,
- cl_I* P, cl_I* Q, uintL* QS, cl_I* B, cl_I* T)
+ cl_I* P, cl_I* Q, uintC* QS, cl_I* B, cl_I* T)
{
switch (N2 - N1) {
case 0:
break;
}
default: {
- var uintL Nm = (N1+N2)/2; // midpoint
+ var uintC Nm = (N1+N2)/2; // midpoint
// Compute left part.
var cl_I LP, LQ, LB, LT;
- var uintL LQS;
+ var uintC LQS;
eval_pqsb_series_aux(N1,Nm,args,&LP,&LQ,&LQS,&LB,<);
// Compute right part.
var cl_I RP, RQ, RB, RT;
- var uintL RQS;
+ var uintC RQS;
eval_pqsb_series_aux(Nm,N2,args,(P?&RP:(cl_I*)0),&RQ,&RQS,&RB,&RT);
// Put together partial results.
if (P) { *P = LP*RP; }
}
}
-const cl_LF eval_rational_series (uintL N, const cl_pqb_series& args, uintC len)
+const cl_LF eval_rational_series (uintC N, const cl_pqb_series& args, uintC len)
{
if (N==0)
return cl_I_to_LF(0,len);
// Split qv[n] into qv[n]*2^qsv[n].
{
var cl_I* qp = args.qv;
- var uintL* qsp = args.qsv;
- for (var uintL n = 0; n < N; n++, qp++, qsp++) {
+ var uintC* qsp = args.qsv;
+ for (var uintC n = 0; n < N; n++, qp++, qsp++) {
// Pull out maximal power of 2 out of *qp = args.qv[n].
- var uintL qs = 0;
+ var uintC qs = 0;
if (!zerop(*qp)) {
qs = ord2(*qp);
if (qs > 0)
}
}
// Main computation.
- var uintL QS;
+ var uintC QS;
eval_pqsb_series_aux(0,N,args,NULL,&Q,&QS,&B,&T);
return cl_I_to_LF(T,len) / scale_float(cl_I_to_LF(B*Q,len),QS);
}
// and T = B*Q*S (all integers). On entry N1 < N2.
// P will not be computed if a NULL pointer is passed.
-static void eval_q_series_aux (uintL N1, uintL N2,
+static void eval_q_series_aux (uintC N1, uintC N2,
const cl_q_series& args,
cl_I* Q, cl_I* T)
{
break;
}
default: {
- var uintL Nm = (N1+N2)/2; // midpoint
+ var uintC Nm = (N1+N2)/2; // midpoint
// Compute left part.
var cl_I LQ, LT;
eval_q_series_aux(N1,Nm,args,&LQ,<);
}
}
-const cl_LF eval_rational_series (uintL N, const cl_q_series& args, uintC len)
+const cl_LF eval_rational_series (uintC N, const cl_q_series& args, uintC len)
{
if (N==0)
return cl_I_to_LF(0,len);
// and T = B*Q*S (all integers). On entry N1 < N2.
// P will not be computed if a NULL pointer is passed.
-static void eval_qa_series_aux (uintL N1, uintL N2,
+static void eval_qa_series_aux (uintC N1, uintC N2,
const cl_qa_series& args,
cl_I* Q, cl_I* T)
{
break;
}
default: {
- var uintL Nm = (N1+N2)/2; // midpoint
+ var uintC Nm = (N1+N2)/2; // midpoint
// Compute left part.
var cl_I LQ, LT;
eval_qa_series_aux(N1,Nm,args,&LQ,<);
}
}
-const cl_LF eval_rational_series (uintL N, const cl_qa_series& args, uintC len)
+const cl_LF eval_rational_series (uintC N, const cl_qa_series& args, uintC len)
{
if (N==0)
return cl_I_to_LF(0,len);
// and T = B*Q*S (all integers). On entry N1 < N2.
// P will not be computed if a NULL pointer is passed.
-static void eval_qab_series_aux (uintL N1, uintL N2,
+static void eval_qab_series_aux (uintC N1, uintC N2,
const cl_qab_series& args,
cl_I* Q, cl_I* B, cl_I* T)
{
break;
}
default: {
- var uintL Nm = (N1+N2)/2; // midpoint
+ var uintC Nm = (N1+N2)/2; // midpoint
// Compute left part.
var cl_I LQ, LB, LT;
eval_qab_series_aux(N1,Nm,args,&LQ,&LB,<);
}
}
-const cl_LF eval_rational_series (uintL N, const cl_qab_series& args, uintC len)
+const cl_LF eval_rational_series (uintC N, const cl_qab_series& args, uintC len)
{
if (N==0)
return cl_I_to_LF(0,len);
// and T = B*Q*S (all integers). On entry N1 < N2.
// P will not be computed if a NULL pointer is passed.
-static void eval_qb_series_aux (uintL N1, uintL N2,
+static void eval_qb_series_aux (uintC N1, uintC N2,
const cl_qb_series& args,
cl_I* Q, cl_I* B, cl_I* T)
{
break;
}
default: {
- var uintL Nm = (N1+N2)/2; // midpoint
+ var uintC Nm = (N1+N2)/2; // midpoint
// Compute left part.
var cl_I LQ, LB, LT;
eval_qb_series_aux(N1,Nm,args,&LQ,&LB,<);
}
}
-const cl_LF eval_rational_series (uintL N, const cl_qb_series& args, uintC len)
+const cl_LF eval_rational_series (uintC N, const cl_qb_series& args, uintC len)
{
if (N==0)
return cl_I_to_LF(0,len);
namespace cln {
-static void eval_pq_series_aux (uintL N1, uintL N2,
+static void eval_pq_series_aux (uintC N1, uintC N2,
cl_pq_series_stream& args,
cl_I* P, cl_I* Q, cl_I* T)
{
break;
}
default: {
- var uintL Nm = (N1+N2)/2; // midpoint
+ var uintC Nm = (N1+N2)/2; // midpoint
// Compute left part.
var cl_I LP, LQ, LT;
eval_pq_series_aux(N1,Nm,args,&LP,&LQ,<);
}
}
-const cl_LF eval_rational_series (uintL N, cl_pq_series_stream& args, uintC len)
+const cl_LF eval_rational_series (uintC N, cl_pq_series_stream& args, uintC len)
{
if (N==0)
return cl_I_to_LF(0,len);
namespace cln {
-static void eval_pqa_series_aux (uintL N1, uintL N2,
+static void eval_pqa_series_aux (uintC N1, uintC N2,
cl_pqa_series_stream& args,
cl_I* P, cl_I* Q, cl_I* T)
{
break;
}
default: {
- var uintL Nm = (N1+N2)/2; // midpoint
+ var uintC Nm = (N1+N2)/2; // midpoint
// Compute left part.
var cl_I LP, LQ, LT;
eval_pqa_series_aux(N1,Nm,args,&LP,&LQ,<);
}
}
-const cl_LF eval_rational_series (uintL N, cl_pqa_series_stream& args, uintC len)
+const cl_LF eval_rational_series (uintC N, cl_pqa_series_stream& args, uintC len)
{
if (N==0)
return cl_I_to_LF(0,len);
namespace cln {
-static void eval_pqab_series_aux (uintL N1, uintL N2,
+static void eval_pqab_series_aux (uintC N1, uintC N2,
cl_pqab_series_stream& args,
cl_I* P, cl_I* Q, cl_I* B, cl_I* T)
{
break;
}
default: {
- var uintL Nm = (N1+N2)/2; // midpoint
+ var uintC Nm = (N1+N2)/2; // midpoint
// Compute left part.
var cl_I LP, LQ, LB, LT;
eval_pqab_series_aux(N1,Nm,args,&LP,&LQ,&LB,<);
}
}
-const cl_LF eval_rational_series (uintL N, cl_pqab_series_stream& args, uintC len)
+const cl_LF eval_rational_series (uintC N, cl_pqab_series_stream& args, uintC len)
{
if (N==0)
return cl_I_to_LF(0,len);
namespace cln {
-static void eval_pqb_series_aux (uintL N1, uintL N2,
+static void eval_pqb_series_aux (uintC N1, uintC N2,
cl_pqb_series_stream& args,
cl_I* P, cl_I* Q, cl_I* B, cl_I* T)
{
break;
}
default: {
- var uintL Nm = (N1+N2)/2; // midpoint
+ var uintC Nm = (N1+N2)/2; // midpoint
// Compute left part.
var cl_I LP, LQ, LB, LT;
eval_pqb_series_aux(N1,Nm,args,&LP,&LQ,&LB,<);
}
}
-const cl_LF eval_rational_series (uintL N, cl_pqb_series_stream& args, uintC len)
+const cl_LF eval_rational_series (uintC N, cl_pqb_series_stream& args, uintC len)
{
if (N==0)
return cl_I_to_LF(0,len);
namespace cln {
-const cl_LF eval_pqcd_series (uintL N, cl_pqcd_series_term* args, uintC len)
+const cl_LF eval_pqcd_series (uintC N, cl_pqcd_series_term* args, uintC len)
{
if (N==0)
return cl_I_to_LF(0,len);
namespace cln {
-void eval_pqcd_series_aux (uintL N, cl_pqcd_series_term* args, cl_pqcd_series_result& Z, cl_boolean rightmost)
+void eval_pqcd_series_aux (uintC N, cl_pqcd_series_term* args, cl_pqcd_series_result& Z, cl_boolean rightmost)
{
// N = N2-N1
switch (N) {
break;
}
default: {
- var uintL Nm = N/2; // midpoint
+ var uintC Nm = N/2; // midpoint
// Compute left part.
var cl_pqcd_series_result L;
eval_pqcd_series_aux(Nm,args+0,L,cl_false);
namespace cln {
-const cl_LF eval_pqd_series (uintL N, cl_pqd_series_term* args, uintC len)
+const cl_LF eval_pqd_series (uintC N, cl_pqd_series_term* args, uintC len)
{
if (N==0)
return cl_I_to_LF(0,len);
namespace cln {
-void eval_pqd_series_aux (uintL N, cl_pqd_series_term* args, cl_pqd_series_result& Z, cl_boolean rightmost)
+void eval_pqd_series_aux (uintC N, cl_pqd_series_term* args, cl_pqd_series_result& Z, cl_boolean rightmost)
{
// N = N2-N1
switch (N) {
break;
}
default: {
- var uintL Nm = N/2; // midpoint
+ var uintC Nm = N/2; // midpoint
// Compute left part.
var cl_pqd_series_result L;
eval_pqd_series_aux(Nm,args+0,L,cl_false);
cl_I* qv;
const cl_I* av;
const cl_I* bv;
- uintL* qsv;
+ uintC* qsv;
};
-extern const cl_LF eval_rational_series (uintL N, const cl_rational_series& args, uintC len);
+extern const cl_LF eval_rational_series (uintC N, const cl_rational_series& args, uintC len);
// In each the special cases below, none of (a,b,p,q) can be NULL. But qs can
// still be given or NULL.
cl_I* qv;
const cl_I* av;
const cl_I* bv;
- uintL* qsv;
+ uintC* qsv;
};
-extern const cl_LF eval_rational_series (uintL N, const cl_pqab_series& args, uintC len);
+extern const cl_LF eval_rational_series (uintC N, const cl_pqab_series& args, uintC len);
struct cl_pqb_series {
const cl_I* pv;
cl_I* qv;
const cl_I* bv;
- uintL* qsv;
+ uintC* qsv;
};
-extern const cl_LF eval_rational_series (uintL N, const cl_pqb_series& args, uintC len);
+extern const cl_LF eval_rational_series (uintC N, const cl_pqb_series& args, uintC len);
struct cl_pqa_series {
const cl_I* pv;
cl_I* qv;
const cl_I* av;
- uintL* qsv;
+ uintC* qsv;
};
-extern const cl_LF eval_rational_series (uintL N, const cl_pqa_series& args, uintC len);
+extern const cl_LF eval_rational_series (uintC N, const cl_pqa_series& args, uintC len);
struct cl_pq_series {
const cl_I* pv;
cl_I* qv;
- uintL* qsv;
+ uintC* qsv;
};
-extern const cl_LF eval_rational_series (uintL N, const cl_pq_series& args, uintC len);
+extern const cl_LF eval_rational_series (uintC N, const cl_pq_series& args, uintC len);
struct cl_pab_series {
const cl_I* pv;
const cl_I* av;
const cl_I* bv;
};
-extern const cl_LF eval_rational_series (uintL N, const cl_pab_series& args, uintC len);
+extern const cl_LF eval_rational_series (uintC N, const cl_pab_series& args, uintC len);
struct cl_pb_series {
const cl_I* pv;
const cl_I* bv;
};
-extern const cl_LF eval_rational_series (uintL N, const cl_pb_series& args, uintC len);
+extern const cl_LF eval_rational_series (uintC N, const cl_pb_series& args, uintC len);
struct cl_pa_series {
const cl_I* pv;
const cl_I* av;
};
-extern const cl_LF eval_rational_series (uintL N, const cl_pa_series& args, uintC len);
+extern const cl_LF eval_rational_series (uintC N, const cl_pa_series& args, uintC len);
struct cl_p_series {
const cl_I* pv;
};
-extern const cl_LF eval_rational_series (uintL N, const cl_p_series& args, uintC len);
+extern const cl_LF eval_rational_series (uintC N, const cl_p_series& args, uintC len);
struct cl_qab_series {
cl_I* qv;
const cl_I* av;
const cl_I* bv;
- uintL* qsv;
+ uintC* qsv;
};
-extern const cl_LF eval_rational_series (uintL N, const cl_qab_series& args, uintC len);
+extern const cl_LF eval_rational_series (uintC N, const cl_qab_series& args, uintC len);
struct cl_qb_series {
cl_I* qv;
const cl_I* bv;
- uintL* qsv;
+ uintC* qsv;
};
-extern const cl_LF eval_rational_series (uintL N, const cl_qb_series& args, uintC len);
+extern const cl_LF eval_rational_series (uintC N, const cl_qb_series& args, uintC len);
struct cl_qa_series {
cl_I* qv;
const cl_I* av;
- uintL* qsv;
+ uintC* qsv;
};
-extern const cl_LF eval_rational_series (uintL N, const cl_qa_series& args, uintC len);
+extern const cl_LF eval_rational_series (uintC N, const cl_qa_series& args, uintC len);
struct cl_q_series {
cl_I* qv;
- uintL* qsv;
+ uintC* qsv;
};
-extern const cl_LF eval_rational_series (uintL N, const cl_q_series& args, uintC len);
+extern const cl_LF eval_rational_series (uintC N, const cl_q_series& args, uintC len);
struct cl_ab_series {
const cl_I* av;
const cl_I* bv;
};
-extern const cl_LF eval_rational_series (uintL N, const cl_ab_series& args, uintC len);
+extern const cl_LF eval_rational_series (uintC N, const cl_ab_series& args, uintC len);
struct cl_b_series {
const cl_I* bv;
};
-extern const cl_LF eval_rational_series (uintL N, const cl_b_series& args, uintC len);
+extern const cl_LF eval_rational_series (uintC N, const cl_b_series& args, uintC len);
struct cl_a_series {
const cl_I* av;
};
-extern const cl_LF eval_rational_series (uintL N, const cl_a_series& args, uintC len);
+extern const cl_LF eval_rational_series (uintC N, const cl_a_series& args, uintC len);
struct cl__series {
};
-extern const cl_LF eval_rational_series (uintL N, const cl__series& args, uintC len);
+extern const cl_LF eval_rational_series (uintC N, const cl__series& args, uintC len);
// In this alternate implementation the series is not represented as a couple
// Constructor.
cl_pqab_series_stream (cl_pqab_series_term (*n)(cl_pqab_series_stream&)) : nextfn (n) {}
};
-extern const cl_LF eval_rational_series (uintL N, cl_pqab_series_stream& args, uintC len);
+extern const cl_LF eval_rational_series (uintC N, cl_pqab_series_stream& args, uintC len);
struct cl_pqb_series_term {
cl_I p;
// Constructor.
cl_pqb_series_stream (cl_pqb_series_term (*n)(cl_pqb_series_stream&)) : nextfn (n) {}
};
-extern const cl_LF eval_rational_series (uintL N, cl_pqb_series_stream& args, uintC len);
+extern const cl_LF eval_rational_series (uintC N, cl_pqb_series_stream& args, uintC len);
struct cl_pqa_series_term {
cl_I p;
// Constructor.
cl_pqa_series_stream (cl_pqa_series_term (*n)(cl_pqa_series_stream&)) : nextfn (n) {}
};
-extern const cl_LF eval_rational_series (uintL N, cl_pqa_series_stream& args, uintC len);
+extern const cl_LF eval_rational_series (uintC N, cl_pqa_series_stream& args, uintC len);
struct cl_pq_series_term {
cl_I p;
// Constructor.
cl_pq_series_stream (cl_pq_series_term (*n)(cl_pq_series_stream&)) : nextfn (n) {}
};
-extern const cl_LF eval_rational_series (uintL N, cl_pq_series_stream& args, uintC len);
+extern const cl_LF eval_rational_series (uintC N, cl_pq_series_stream& args, uintC len);
// [Generalization.]
cl_I D;
cl_I V;
};
-extern void eval_pqcd_series_aux (uintL N, cl_pqcd_series_term* args, cl_pqcd_series_result& Z, cl_boolean rightmost = cl_true);
+extern void eval_pqcd_series_aux (uintC N, cl_pqcd_series_term* args, cl_pqcd_series_result& Z, cl_boolean rightmost = cl_true);
// Ditto, but returns U/S.
-extern const cl_LF eval_pqcd_series (uintL N, cl_pqcd_series_term* args, uintC len);
+extern const cl_LF eval_pqcd_series (uintC N, cl_pqcd_series_term* args, uintC len);
// [Special case c(n)=1.]
// Subroutine:
cl_I D;
cl_I V;
};
-extern void eval_pqd_series_aux (uintL N, cl_pqd_series_term* args, cl_pqd_series_result& Z, cl_boolean rightmost = cl_true);
+extern void eval_pqd_series_aux (uintC N, cl_pqd_series_term* args, cl_pqd_series_result& Z, cl_boolean rightmost = cl_true);
// Ditto, but returns U/S.
-extern const cl_LF eval_pqd_series (uintL N, cl_pqd_series_term* args, uintC len);
+extern const cl_LF eval_pqd_series (uintC N, cl_pqd_series_term* args, uintC len);
} // namespace cln
// a(n) = 205*n^2+250*n+77, b(n) = 1,
// p(0) = 1, p(n) = -n^5 for n>0, q(n) = 32*(2n+1)^5.
var uintC actuallen = len+2; // 2 Schutz-Digits
- var uintL N = ceiling(actuallen*intDsize,10);
+ var uintC N = ceiling(actuallen*intDsize,10);
// 1024^-N <= 2^(-intDsize*actuallen).
CL_ALLOCA_STACK;
var cl_I* av = (cl_I*) cl_alloca(N*sizeof(cl_I));
var cl_I* pv = (cl_I*) cl_alloca(N*sizeof(cl_I));
var cl_I* qv = (cl_I*) cl_alloca(N*sizeof(cl_I));
- var uintL n;
+ var uintC n;
for (n = 0; n < N; n++) {
init1(cl_I, av[n]) (205*square((cl_I)n) + 250*(cl_I)n + 77);
if (n==0)
// with convergence acceleration through exp(x), and evaluated
// using the binary-splitting algorithm.
var uintC actuallen = len+2; // 2 Schutz-Digits
- var uintL x = (uintL)(0.693148*intDsize*actuallen)+1;
- var uintL N = (uintL)(2.718281828*x);
+ var uintC x = (uintC)(0.693148*intDsize*actuallen)+1;
+ var uintC N = (uintC)(2.718281828*x);
CL_ALLOCA_STACK;
var cl_pqd_series_term* args = (cl_pqd_series_term*) cl_alloca(N*sizeof(cl_pqd_series_term));
- var uintL n;
+ var uintC n;
for (n = 0; n < N; n++) {
if (n==0) {
init1(cl_I, args[n].p) (1);
// zeta(s) = 1/(1-2^(1-s)) sum(n=0..infty, (-1)^n/(n+1)^s),
// with Cohen-Villegas-Zagier convergence acceleration.
var uintC actuallen = len+2; // 2 Schutz-Digits
- var uintL N = (uintL)(0.39321985*intDsize*actuallen)+1;
+ var uintC N = (uintC)(0.39321985*intDsize*actuallen)+1;
var cl_I fterm = 2*(cl_I)N*(cl_I)N;
var cl_I fsum = fterm;
var cl_LF gterm = cl_I_to_LF(fterm,actuallen);
var cl_LF gsum = gterm;
- var uintL n;
+ var uintC n;
// After n loops
// fterm = (N+n)!N/(2n+2)!(N-n-1)!*2^(2n+2), fsum = ... + fterm,
// gterm = S_n*fterm, gsum = ... + gterm.
// with Cohen-Villegas-Zagier convergence acceleration, and
// evaluated using the binary splitting algorithm.
var uintC actuallen = len+2; // 2 Schutz-Digits
- var uintL N = (uintL)(0.39321985*intDsize*actuallen)+1;
+ var uintC N = (uintC)(0.39321985*intDsize*actuallen)+1;
CL_ALLOCA_STACK;
var cl_pqd_series_term* args = (cl_pqd_series_term*) cl_alloca(N*sizeof(cl_pqd_series_term));
- var uintL n;
+ var uintC n;
for (n = 0; n < N; n++) {
init1(cl_I, args[n].p) (2*(cl_I)(N-n)*(cl_I)(N+n));
init1(cl_I, args[n].q) ((cl_I)(2*n+1)*(cl_I)(n+1));
{
if (eq(x,0) || eq(x,1)) // x=0 oder x=1 ?
{ *w = x; return cl_true; } // ja -> x als Ergebnis
- if (n >= UL_to_I(integer_length(x)))
+ if (n >= (cl_I)(unsigned long)integer_length(x))
{ return cl_false; }
- // Nun ist n < (integer-length x). Also paßt n in ein uintL.
- return cl_rootp_aux(x,cl_I_to_UL(n),w);
+ // Nun ist n < (integer-length x). Also paßt n in ein uintC.
+ return cl_rootp_aux(x,cl_I_to_ulong(n),w);
}
} // namespace cln
}
// Nun ist n ungerade.
if (n==1) { *w = x; return cl_true; } // n=1 -> x als Ergebnis
- var uintL oq = 0; // Shift von y am Schluß
- {var uintL o = ord2(x);
+ var uintC oq = 0; // Shift von y am Schluß
+ {var uintC o = ord2(x);
if (!(o==0))
{var uintL o_r; divu_3232_3232(o,n, oq=,o_r=); // o_r = o mod n
if (!(o_r==0)) { return cl_false; } // o nicht durch n teilbar -> fertig
// oq = o/n.
// dividiere x durch 2^o:
- x = ash(x,-(sintL)o);
+ x = ash(x,-(sintC)o);
} }
// Nun ist n ungerade, x ungerade.
CL_ALLOCA_STACK;
y_lsd_ok:
ASSERT(expt_pos(y_lsd,n)==x_lsd);
// Nun ist y_lsd^n == x_lsd mod beta=2^intDsize.
- { var uintL m = ceiling((uintL)x_len,n); // für y nötige Länge, >0, <=x_len
+ { var uintC m = ceiling(x_len,n); // für y nötige Länge, >0, <=x_len
var uintD* y_LSDptr;
{ var uintD* z1_LSDptr;
var uintD* z2_LSDptr;
var uintD* z3_LSDptr;
num_stack_alloc_1(m, ,y_LSDptr=); // Platz für y
- {var uintL need = 2*m+(32/intDsize-1); // >= max(2*m,m+32/intDsize)
+ {var uintC need = 2*m+(32/intDsize-1); // >= max(2*m,m+32/intDsize)
num_stack_alloc(need, ,z1_LSDptr=); // Platz für Rechenregister 1
num_stack_alloc(need, ,z2_LSDptr=); // Platz für Rechenregister 2
num_stack_alloc(need, ,z3_LSDptr=); // Platz für Rechenregister 3
}
- {var uintL k = 1; // y ist bisher mod beta^k bekannt
+ {var uintC k = 1; // y ist bisher mod beta^k bekannt
lspref(y_LSDptr,0) = y_lsd; // Startwert von y
until (k==m)
- { var uintL k2 = 2*k; if (k2>m) { k2=m; } // k2 = min(2*k,m) > k
+ { var uintC k2 = 2*k; if (k2>m) { k2=m; } // k2 = min(2*k,m) > k
// bisheriges y mod beta^k2 mit n-1 potenzieren:
// Methode für z := y^(n-1) :
// zz:=y, e:=n-1.
namespace cln {
-const cl_I ash (const cl_I& x, sintL y)
+const cl_I ash (const cl_I& x, sintC y)
{
// Methode:
// x = 0 -> 0 als Ergebnis
CL_ALLOCA_STACK;
if (y >= 0) {
// y>0
- var uintL y_ = (uintL)y;
+ var uintC y_ = (uintC)y;
var uintL i = y_%intDsize; // i = y mod intDsize, >=0, <intDsize
- var uintL k = floor(y_,intDsize); // k = y div intDsize, >=0, <2^intCsize
+ var uintC k = floor(y_,intDsize); // k = y div intDsize, >=0, <2^intCsize
var uintD* LSDptr;
var uintC len;
var const uintD* x_LSDptr;
I_to_NDS_nocopy(x, ,len=,x_LSDptr=,cl_false,); // DS zu x bilden.
- if (k >= (uintC)(~(uintC)len)) // kann len+k+1 Überlauf geben?
+ if (k >= (uintC)(~len)) // kann len+k+1 Überlauf geben?
{ cl_ash_error(y); } // ja -> Fehler
num_stack_alloc_1(len+k,,LSDptr=);
LSDptr = clear_loop_lsp(LSDptr,k); // k Nulldigits
}
} else {
// y<0
- var uintL y_ = (uintL)(-y); // Wert von -y, >0
+ var uintC y_ = (uintC)(-y); // Wert von -y, >0
var uintL i = y_%intDsize; // i = (-y) mod intDsize, >=0, <intDsize
- var uintL k = floor(y_,intDsize); // k = (-y) div intDsize, >=0
+ var uintC k = floor(y_,intDsize); // k = (-y) div intDsize, >=0
// DS zu x bilden:
var uintD* MSDptr;
var uintC len;
if (!minusp(y)) {
// y>=0
var uintL i; // i = y mod intDsize, >=0, <intDsize
- var cl_uint k; // k = y div intDsize, >=0, <2^intCsize
+ var uintC k; // k = y div intDsize, >=0, <2^intCsize
if (bignump(y)) {
#if (log2_intDsize+intCsize <= cl_value_len-1)
// y >= 2^(cl_value_len-1) >= intDsize*2^intCsize
var uintC len;
var const uintD* x_LSDptr;
I_to_NDS_nocopy(x, ,len=,x_LSDptr=,cl_false,); // DS zu x bilden.
- if (k >= (uintC)(~(uintC)len)) // kann len+k+1 Überlauf geben?
+ if (k >= (uintC)(~len)) // kann len+k+1 Überlauf geben?
{ cl_ash_error(y); } // ja -> Fehler
num_stack_alloc_1(len+k,,LSDptr=);
LSDptr = clear_loop_lsp(LSDptr,k); // k Nulldigits
} else {
// y<0
var uintL i; // i = (-y) mod intDsize, >=0, <intDsize
- var cl_uint k; // k = (-y) div intDsize, >=0, <2^intCsize
+ var uintC k; // k = (-y) div intDsize, >=0, <2^intCsize
if (bignump(y)) {
#if (log2_intDsize+intCsize <= cl_value_len-1)
// -y-1 >= 2^(cl_value_len-1) >= intDsize*2^intCsize
// cl_fullbyte(p,q) liefert zu p,q die Zahl 2^q-2^p als Integer,
// wobei p und q uintL sind. Bei p<=q ist das Ergebnis also
// ein Integer >=0, bei dem genau die Bits p,...,q-1 gesetzt sind.
-extern const cl_I cl_fullbyte (uintL p, uintL q);
+extern const cl_I cl_fullbyte (uintC p, uintC q);
// Extrahiere die Bits p,...,q-1 der Zahl x,
// wobei 0 <= p <= q <= l = (integer-length x).
// Ergebnis (wie bei LDB) ein Integer >=0.
-extern const cl_I ldb_extract (const cl_I& x, uintL p, uintL q);
+extern const cl_I ldb_extract (const cl_I& x, uintC p, uintC q);
// Teste, ob eines der Bits p,...,q-1 der Zahl x /=0 ist,
// wobei 0 <= p <= q <= l = (integer-length x).
// Ergebnis (wie bei LDB-TEST) cl_false wenn nein, cl_true wenn ja.
-extern cl_boolean ldb_extract_test (const cl_I& x, uintL p, uintL q);
+extern cl_boolean ldb_extract_test (const cl_I& x, uintC p, uintC q);
// Extrahiere die Bits p,...,q-1 der Zahl x,
// wobei 0 <= p <= q <= l = (integer-length x).
// Ergebnis (wie bei MASK-FIELD) ein Integer >=0.
-extern const cl_I mkf_extract (const cl_I& x, uintL p, uintL q);
+extern const cl_I mkf_extract (const cl_I& x, uintC p, uintC q);
} // namespace cln
namespace cln {
-const cl_I cl_fullbyte (uintL p, uintL q)
+const cl_I cl_fullbyte (uintC p, uintC q)
{
if (p==q)
return 0;
else
- return ash(-1,UL_to_I(p)) + ash(1,UL_to_I(q));
+ return ash(-1,(cl_I)(unsigned long)p) + ash(1,(cl_I)(unsigned long)q);
}
} // namespace cln
namespace cln {
-uintL integer_length (const cl_I& x)
+uintC integer_length (const cl_I& x)
{
if (fixnump(x))
{ var uintL bitcount = 0;
{ var const uintD* MSDptr;
var uintC len;
BN_to_NDS_nocopy(x, MSDptr=,len=,); // normalisierte DS zu x bilden.
- var uintL bitcount = intDsize*(uintL)(len-1); // Anzahl Digits mal intDsize
+ var uintC bitcount = intDsize*(len-1); // Anzahl Digits mal intDsize
// MSDigit nehmen, testen, welches das höchste Bit ist, das vom
// Vorzeichenbit abweicht:
var uintD msd = mspref(MSDptr,0); // MSDigit
// q:=min(p+s,l).
// Extrahiere die Bits p,...,q-1 von n.
// Falls p+s>l und n<0, füge p+s-l Einsenbits an (addiere 2^s-2^(l-p)).
- var uintL s = b.size;
- var uintL p = b.position;
- var uintL l = integer_length(n); // l = (integer-length n)
+ var uintC s = b.size;
+ var uintC p = b.position;
+ var uintC l = integer_length(n); // l = (integer-length n)
if (l<=p)
// l<=p
if (!minusp(n))
else
// l>p
{ var cl_I erg = ldb_extract(n,p,(p+s<l ? p+s : l));
- var uintL lp = l-p;
+ var uintC lp = l-p;
if ((s>lp) && minusp(n)) // s>l-p und n<0 ?
{ return logior(erg,cl_fullbyte(lp,s)); } // setze Bits l-p,...,s-1
// (logisches Exklusiv-Oder oder Addition ginge auch)
// und bei n<0 sind Bits p+s-1..l =1.
// Falls p+s<=l,
// extrahiere die Bits p,...,p+s-1 von n und teste sie.
- var uintL s = b.size;
- var uintL p = b.position;
+ var uintC s = b.size;
+ var uintC p = b.position;
if (s==0) return cl_false;
- var uintL l = integer_length(n); // l = (integer-length n)
+ var uintC l = integer_length(n); // l = (integer-length n)
if (l<=p)
// l<=p
if (!minusp(n))
return cl_true; // n<0
else
// l>p
- { var uintL ps = p+s;
+ { var uintC ps = p+s;
if (ps>l) // p+s>l ?
return cl_true;
// Bits p,...,q-1 mit q = min(p+s,l) = p+s extrahieren und testen:
namespace cln {
-const cl_I ldb_extract (const cl_I& x, uintL p, uintL q)
+const cl_I ldb_extract (const cl_I& x, uintC p, uintC q)
{ CL_ALLOCA_STACK;
var const uintD* MSDptr;
var uintC len;
var const uintD* LSDptr;
I_to_NDS_nocopy(x, MSDptr=,len=,LSDptr=,cl_true, { return 0; } ); // NDS zu x bilden
// MSDptr erhöhen und len erniedrigen, so daß len = ceiling(q/intDsize) wird:
- { var uintL qD = ceiling(q,intDsize); // ceiling(q/intDsize)
+ { var uintC qD = ceiling(q,intDsize); // ceiling(q/intDsize)
// wegen q<=l ist qD = ceiling(q/intDsize) <= ceiling((l+1)/intDsize) = len, also
// paßt qD ebenso wie len in ein uintC.
- MSDptr = MSDptr mspop ((uintL)len - qD); // MSDptr um len-qD Digits erhöhen
+ MSDptr = MSDptr mspop (len - qD); // MSDptr um len-qD Digits erhöhen
len = qD; // len um len-qD erniedrigen
}
// LSDptr und len um floor(p/intDsize) erniedrigen:
- { var uintL pD = floor(p,intDsize); // floor(p/intDsize)
+ { var uintC pD = floor(p,intDsize); // floor(p/intDsize)
LSDptr = LSDptr lspop pD;
len -= pD;
}
var uintD* newMSDptr;
{ var uintL i = p%intDsize; // p mod intDsize
// Kopiere sie und schiebe sie dabei um i Bits nach rechts:
- num_stack_alloc_1((uintL)len, newMSDptr=,); // neue UDS newMSDptr/len/..
+ num_stack_alloc_1(len, newMSDptr=,); // neue UDS newMSDptr/len/..
if (i==0)
{ copy_loop_msp(MSDptr,newMSDptr,len); }
else
}
// newMSDptr/len/.. = geschobene Kopie der maßgeblichen Digits
// Ausblenden der Bits mit Nummern >= q-p:
- { var uintL bitcount = intDsize*(uintL)len - (q-p);
+ { var uintC bitcount = intDsize*len - (q-p);
// Anzahl vorne auszublendender Bits ( >=0, <= intDsize-1 + intDsize-1 )
if (bitcount>=intDsize)
{ bitcount -= intDsize; msshrink(newMSDptr); len -= 1; } // intDsize Bits ausblenden
namespace cln {
-cl_boolean ldb_extract_test (const cl_I& x, uintL p, uintL q)
+cl_boolean ldb_extract_test (const cl_I& x, uintC p, uintC q)
{ var const uintD* MSDptr;
var uintC len;
var const uintD* LSDptr;
I_to_NDS_nocopy(x, MSDptr=,len=,LSDptr=,cl_true, { return cl_false; } ); // NDS zu x bilden
// MSDptr erhöhen und len erniedrigen, so daß len = ceiling(q/intDsize) wird:
- { var uintL qD = ceiling(q,intDsize); // ceiling(q/intDsize)
+ { var uintC qD = ceiling(q,intDsize); // ceiling(q/intDsize)
// wegen q<=l ist qD = ceiling(q/intDsize) <= ceiling((l+1)/intDsize) = len, also
// paßt qD ebenso wie len in ein uintC.
- MSDptr = MSDptr mspop ((uintL)len - qD); // MSDptr um len-qD Digits erhöhen
+ MSDptr = MSDptr mspop (len - qD); // MSDptr um len-qD Digits erhöhen
len = qD; // len um len-qD erniedrigen
}
// LSDptr und len um floor(p/intDsize) erniedrigen:
- { var uintL pD = p/intDsize; // floor(p/intDsize)
+ { var uintC pD = p/intDsize; // floor(p/intDsize)
LSDptr = LSDptr lspop pD;
len -= pD;
}
namespace cln {
-cl_boolean logbitp (uintL x, const cl_I& y)
+cl_boolean logbitp (uintC x, const cl_I& y)
{
// Methode:
// Falls x>=intDsize*Länge(y), teste Vorzeichen von y.
var uintC ylen;
var const uintD* yLSDptr;
I_to_NDS_nocopy(y, yMSDptr=,ylen=,yLSDptr=,cl_true, { return cl_false; } ); // DS zu y
- if (x < intDsize*(uintL)ylen)
+ if (x < intDsize*ylen)
// x ist >=0, < intDsize*ylen
{ if (lspref(yLSDptr,floor(x,intDsize)) & bit(x%intDsize))
return cl_true;
var uintC ylen;
var const uintD* yLSDptr;
I_to_NDS_nocopy(y, ,ylen=,yLSDptr=,cl_true, { return cl_false; } ); // DS zu y
- if (x_ < intDsize*(uintL)ylen)
+ if (x_ < intDsize*ylen)
// x ist ein Fixnum >=0, < intDsize*ylen
{ if (lspref(yLSDptr,floor(x_,intDsize)) & bit(x_%intDsize))
return cl_true;
namespace cln {
-uintL logcount (const cl_I& x)
+uintC logcount (const cl_I& x)
{
if (fixnump(x))
{ var uintV x32 = FN_to_V(x); // x als intDsize-Bit-Zahl
{ var const uintD* MSDptr;
var uintC len;
BN_to_NDS_nocopy(x, MSDptr=,len=,); // DS zu x bilden, len>0.
- var uintL bitcount = 0; // Bitzähler
+ var uintC bitcount = 0; // Bitzähler
var const uintD* ptr = MSDptr; // läuft durch die Digits durch
var uintD sign = sign_of_sintD(mspref(ptr,0)); // Vorzeichen
dotimespC(len,len,
- { bitcount += (uintL)logcountD(msprefnext(ptr) ^ sign); });
+ { bitcount += (uintC)logcountD(msprefnext(ptr) ^ sign); });
// 0 <= bitcount < intDsize*2^intCsize.
return bitcount;
}
// q:=min(p+s,l).
// Extrahiere die Bits p,...,q-1 von n.
// Falls p+s>l und n<0, füge p+s-l Einsenbits an (addiere 2^(p+s)-2^l).
- var uintL s = b.size;
- var uintL p = b.position;
- {var uintL ps = p+s;
- var uintL l = integer_length(n); // l = (integer-length n)
+ var uintC s = b.size;
+ var uintC p = b.position;
+ {var uintC ps = p+s;
+ var uintC l = integer_length(n); // l = (integer-length n)
if (l<=p)
// l<=p
if (!minusp(n))
namespace cln {
-const cl_I mkf_extract (const cl_I& x, uintL p, uintL q)
+const cl_I mkf_extract (const cl_I& x, uintC p, uintC q)
{
CL_ALLOCA_STACK;
var const uintD* MSDptr;
var const uintD* LSDptr;
I_to_NDS_nocopy(x, MSDptr=,len=,LSDptr=,cl_true, { return 0; } ); // NDS zu x bilden
// MSDptr erhöhen und len erniedrigen, so daß len = ceiling(q/intDsize) wird:
- { var uintL qD = ceiling(q,intDsize); // ceiling(q/intDsize)
+ { var uintC qD = ceiling(q,intDsize); // ceiling(q/intDsize)
// wegen q<=l ist qD = ceiling(q/intDsize) <= ceiling((l+1)/intDsize) = len, also
// paßt qD ebenso wie len in ein uintC.
- MSDptr = MSDptr mspop ((uintL)len - qD); // MSDptr um len-qD Digits erhöhen
+ MSDptr = MSDptr mspop (len - qD); // MSDptr um len-qD Digits erhöhen
len = qD; // len um len-qD erniedrigen
}
// Platz (len Digits) für die neue UDS bereitstellen:
var uintD* newMSDptr;
- num_stack_alloc_1((uintL)len, newMSDptr = ,); // Platz belegen
- {var uintL pD = p/intDsize; // floor(p/intDsize), paßt in ein uintC
+ num_stack_alloc_1(len, newMSDptr = ,); // Platz belegen
+ {var uintC pD = p/intDsize; // floor(p/intDsize), paßt in ein uintC
// Kopiere len-pD Digits aus der DS zu x heraus:
- var uintD* midptr = copy_loop_msp(MSDptr,newMSDptr,len-(uintC)pD);
+ var uintD* midptr = copy_loop_msp(MSDptr,newMSDptr,len-pD);
// Lösche p-intDsize*floor(p/intDsize) Bits im Digit unterhalb von midptr:
- {var uintL p_D = p%intDsize;
+ {var uintC p_D = p%intDsize;
if (!(p_D==0)) { lspref(midptr,0) &= minus_bit(p_D); }
}
// Lösche pD Digits darüber:
// Bignums.
struct cl_heap_bignum : cl_heap {
- unsigned int length; // length (in digits)
+ uintC length; // length (in digits)
uintD data[1]; // number in two's complement representation
};
inline cl_heap_bignum* TheBignum (const cl_number& obj)
{ return (cl_heap_bignum*)(obj.pointer); }
-inline cl_heap_bignum* allocate_bignum (unsigned int length)
+inline cl_heap_bignum* allocate_bignum (uintC length)
{
cl_heap_bignum* p = (cl_heap_bignum*) malloc_hook(offsetofa(cl_heap_bignum,data)+sizeof(uintD)*length);
p->refcount = 1;
// > MSBptr/len/..: Ziffernfolge, bestehend aus Punkten (werden überlesen)
// und Ziffern/Buchstaben mit Wert < base.
// < ergebnis: der dargestellte Integer >=0
- extern const cl_I digits_to_I (const char * MSBptr, uintL len, uintD base);
+ extern const cl_I digits_to_I (const char * MSBptr, uintC len, uintD base);
// Hilfsfunktion zur Ausgabe von Integers
// cl_digits_need(len,base) liefert eine obere Abschätzung für die Anzahl der
// Ziffern im Stellenwertsystem der Basis base, die x >= 0 braucht.
- extern uintL cl_digits_need (const cl_I& x, uintL base);
+ extern uintC cl_digits_need (const cl_I& x, uintL base);
// Wandelt ein Integer in ein Stellensystem um.
// I_to_digits(x,base, &ergebnis);
// > base: Stellensystem-Basis, 2 <= base <= 36.
// > ergebnis.LSBptr: darunter ist mindestens digits_need(len) Bytes Platz
// < ergebnis: fertige Folge MSBptr/len/LSBptr von Ziffern
- typedef struct { uintB* MSBptr; uintL len; uintB* LSBptr; } cl_digits;
+ typedef struct { uintB* MSBptr; uintC len; uintB* LSBptr; } cl_digits;
extern void I_to_digits (const cl_I& x, uintD base, cl_digits* erg);
};\r
\r
struct cached_power_table {\r
- cached_power_table_entry element[30];\r
+ cached_power_table_entry element[40];\r
// Constructor and destructor - nothing special.\r
cached_power_table () {}\r
~cached_power_table () {}\r
namespace cln {
-uintL cl_digits_need (const cl_I& x, uintL base)
+uintC cl_digits_need (const cl_I& x, uintL base)
{
if (fixnump(x))
{ return cl_value_len; } // x < 2^cl_value_len, base >= 2, also reicht das
else
{ var uintC len = TheBignum(x)->length;
// 1+ceiling(len * intDsize*log(2)/log(base)) Bytes oder etwas mehr
- var uintL need = 1+floor(len,1024/intDsize); // > ceiling(len*intDsize/1024) >= 0
+ var uintC need = 1+floor(len,1024/intDsize); // > ceiling(len*intDsize/1024) >= 0
switch (base) // need mit ceiling(1024*log(2)/log(base)) multiplizieren:
{ case 2: need = 1024*need; break;
case 3: need = 647*need; break;
\r
namespace cln {\r
\r
-static const cl_I digits_to_I_base2 (const char * MSBptr, uintL len, uintD base)\r
+static const cl_I digits_to_I_base2 (const char * MSBptr, uintC len, uintD base)\r
{\r
// base is a power of two: write the digits from least significant\r
// to most significant into the result NUDS. Result needs\r
return NUDS_to_I(erg_MSDptr,erg_len);\r
}\r
\r
-static const cl_I digits_to_I_baseN (const char * MSBptr, uintL len, uintD base)\r
+static const cl_I digits_to_I_baseN (const char * MSBptr, uintC len, uintD base)\r
{\r
// base is not a power of two: Add digits one by one. Result nees\r
// 1+ceiling(len*log(base)/(intDsize*log(2))) or some more digits.\r
var uintD* erg_MSDptr;\r
var uintC erg_len;\r
var uintD* erg_LSDptr;\r
- var uintL need = 1+floor(len,intDsize*256); // > len/(intDsize*256) >=0\r
+ var uintC need = 1+floor(len,intDsize*256); // > len/(intDsize*256) >=0\r
switch (base) { // multiply need with ceiling(256*log(base)/log(2)):\r
case 2: need = 256*need; break;\r
case 3: need = 406*need; break;\r
return NUDS_to_I(erg_MSDptr,erg_len);\r
}\r
\r
-const cl_I digits_to_I (const char * MSBptr, uintL len, uintD base)\r
+const cl_I digits_to_I (const char * MSBptr, uintC len, uintD base)\r
{\r
if ((base & (base-1)) == 0) {\r
return digits_to_I_base2(MSBptr, len, base);\r
const int cl_digits_algo = 1;
// like I_to_digits, except that the result has exactly erg_len characters.
-static inline void I_to_digits_noshrink (const cl_I& X, uintD base, uintL erg_len, cl_digits* erg)
+static inline void I_to_digits_noshrink (const cl_I& X, uintD base, uintC erg_len, cl_digits* erg)
{
I_to_digits(X,base,erg);
if (erg->len > erg_len) cl_abort();
- var uintL count = erg_len - erg->len;
+ var uintC count = erg_len - erg->len;
if (count > 0)
{ var uintB* ptr = erg->MSBptr;
do { *--ptr = '0'; } while (--count > 0);
// for k*2^i characters, convert X1 to string. (Have to convert
// X0 first because the conversion may temporarily prepend some
// zero characters.)
- var uintL ilen_X = integer_length(X);
+ var uintC ilen_X = integer_length(X);
var const cached_power_table_entry * p;
- var uintL ilen_B;
+ var uintC ilen_B;
var uintL i;
for (i = 0; ; i++)
{ p = cached_power(base,i);
{var uintD* divroomptr; // Platz für Divisionsergebnis
var uintD* c_LSDptr;
var uintD* d_LSDptr;
- {var uintL c_len = (uintL)(a_len>=b_len ? a_len : b_len) + 1;
+ {var uintC c_len = (a_len>=b_len ? a_len : b_len) + 1;
num_stack_alloc(c_len,divroomptr=,c_LSDptr=);
num_stack_alloc(c_len,,d_LSDptr=);
// Jetzt ist ../c_len/c_LSDptr, ../c_len/d_LSDptr frei.
// uBb = uBb.MSDptr/uBb.len/uBb.LSDptr,
// alles NUDS.
// Platz für zwei Rechenregister besorgen, mit je max(a_len,b_len)+1 Digits:
- {var uintL c_len = (uintL)(a_len>=b_len ? a_len : b_len) + 1;
+ {var uintC c_len = (a_len>=b_len ? a_len : b_len) + 1;
num_stack_alloc(c_len,,c_LSDptr=);
num_stack_alloc(c_len,divroomptr=,d_LSDptr=);
// Jetzt ist ../c_len/c_LSDptr, ../c_len/d_LSDptr frei.
} else {
sign = 0;
}
- var sintL exp = (uintL)len * intDsize;
+ var sintC exp = len * intDsize;
// Nicht alle führenden 65 Bits sind =0.
if (msd==0) {
msd = msd2;
} else {
sign = 0;
}
- var sintL exp = (uintL)len * intDsize;
+ var sintC exp = len * intDsize;
// Nicht alle führenden intDsize+1 Bits sind =0.
// Wegen intDsize<=32: Nicht alle führenden 33 Bits sind =0.
if (msd==0) {
// (ord2 x) = intDsize * Anzahl der Nulldigits am Schluß
// + Anzahl der Nullbits am Ende des letzten Digits /=0.
-uintL ord2 (const cl_I& x) // x /= 0
+uintC ord2 (const cl_I& x) // x /= 0
{
if (fixnump(x))
{ var uintV x_ = FN_to_V(x); // x als intVsize-Bit-Zahl
#endif
}
else
- { var uintL bitcount = 0;
+ { var uintC bitcount = 0;
var const uintD* ptr;
BN_to_NDS_nocopy(x, ,,ptr=); // normalisierte DS zu x bilden.
while (lspref(ptr,0) == 0) { lsshrink(ptr); bitcount += intDsize; } // Nulldigits abzählen
namespace cln {
-uintL power2p (const cl_I& x) // x > 0
+uintC power2p (const cl_I& x) // x > 0
{
// Methode 1: Wenn ord2(x) = integer_length(x)-1.
// Methode 2: Wenn logand(x,x-1) = 0.
if (DS_test_loop(MSDptr mspop 1,len-1,LSDptr)) return 0; // danach alles Nullen
{var uintL msdlen;
integerlengthD(msd, msdlen=);
- return intDsize*(uintL)(len-1) + msdlen; // integer_length(x) als Ergebnis
+ return intDsize*(len-1) + msdlen; // integer_length(x) als Ergebnis
}}
}
char * cl_decimal_string (const cl_I& x)
{
CL_ALLOCA_STACK;
- var uintL need = cl_digits_need(x,10);
+ var uintC need = cl_digits_need(x,10);
var uintB* ziffern = cl_alloc_array(uintB,need); // Platz für die Ziffern
var cl_digits erg; erg.LSBptr = &ziffern[need];
I_to_digits(x,10,&erg); // Umwandlung in Ziffern
} else
abs_z = z;
CL_ALLOCA_STACK;
- var uintL need = cl_digits_need(abs_z,base);
+ var uintC need = cl_digits_need(abs_z,base);
var uintB* ziffern = cl_alloc_array(uintB,need); // Platz für die Ziffern
var cl_digits erg; erg.LSBptr = &ziffern[need];
I_to_digits(abs_z,(uintD)base,&erg); // Umwandlung in Ziffern
// Ziffern ausgeben:
{
var uintB* ptr = erg.MSBptr;
- var uintL count = erg.len;
+ var uintC count = erg.len;
do { fprintchar(stream,*ptr++); } until (--count==0);
}
}
} else
abs_z = z;
CL_ALLOCA_STACK;
- var uintL need = 1+cl_digits_need(abs_z,base);
+ var uintC need = 1+cl_digits_need(abs_z,base);
var uintB* ziffern = cl_alloc_array(uintB,need); // Platz für die Ziffern
var cl_digits erg; erg.LSBptr = &ziffern[need];
I_to_digits(abs_z,(uintD)base,&erg); // Umwandlung in Ziffern
type = &cl_class_modint_ring;
if (minusp(m)) cl_abort();
if (!cln::zerop(m)) {
- var uintL b = integer_length(m-1);
+ var uintC b = integer_length(m-1);
// m <= 2^b, hence one needs b bits for a representative mod m.
if (b <= 1) {
log2_bits = 0; bits = 1;
} else if (b <= cl_word_size) {
var uintL bb;
- integerlength32(b-1,bb=); // b <= 2^bb with bb minimal
+ integerlengthC(b-1,bb=); // b <= 2^bb with bb minimal
log2_bits = bb; bits = 1<<bb;
} else {
log2_bits = -1; bits = -1;
return new cl_heap_modint_ring_int();
// Now m > 0.
{
- var uintL log2_m = power2p(m);
+ var uintC log2_m = power2p(m);
if (log2_m)
return new cl_heap_modint_ring_pow2(m,log2_m-1);
}
// Now m > 1.
{
- var uintL log2_m = integer_length(m); // = integerlength(m-1)
+ var uintC log2_m = integer_length(m); // = integerlength(m-1)
if (log2_m < 16) // m < 0x10000
return new cl_heap_modint_ring_fix16(m);
#if (cl_value_len <= 32)
#endif
}
{
- var uintL m1 = power2p(m+1);
+ var uintC m1 = power2p(m+1);
if (m1)
return new cl_heap_modint_ring_pow2m1(m,m1-1);
}
{
- var uintL m1 = power2p(m-1);
+ var uintC m1 = power2p(m-1);
if (m1)
return new cl_heap_modint_ring_pow2p1(m,m1-1);
}
namespace cln {
-const cl_MI operator<< (const cl_MI& x, sintL y) // assume 0 <= y < 2^31
+const cl_MI operator<< (const cl_MI& x, sintC y) // assume 0 <= y < 2^(intCsize-1)
{
if (y == 0)
return x;
{
if (!oddp(M))
return NULL;
- var uintL m = integer_length(M);
+ var uintC m = integer_length(M);
CL_ALLOCA_STACK;
var uintC len;
var const uintD* M_LSDptr;
recip2adic(len,M_LSDptr,U_LSDptr);
// Look at U's bits.
#define U_bit(i) (lspref(U_LSDptr,floor(i,intDsize)) & ((uintD)1 << ((i)%intDsize)))
- var uintL i_min;
- var uintL i_max;
- var uintL i = floor(m,2);
+ var uintC i_min;
+ var uintC i_max;
+ var uintC i = floor(m,2);
var cl_boolean negative;
if (U_bit(i)) {
for (; --i > 0; )
// OK, all the bits i_max-1..i_min of U are equal.
if (i_max - i_min <= floor(m,2))
return NULL;
- var uintL n = i_max;
+ var uintC n = i_max;
// Turn U (mod 2^n) into a signed integer.
if (n % intDsize) {
if (negative)
else
lspref(U_LSDptr,floor(n,intDsize)) &= ((uintD)1 << (n % intDsize)) - 1;
}
- var uintL U_len = ceiling(n,intDsize);
+ var uintC U_len = ceiling(n,intDsize);
var cl_I U = DS_to_I(U_LSDptr lspop U_len,U_len);
var cl_I V_N = 1 - U*M;
if (ldb_test(V_N,cl_byte(n,0)))
SUBCLASS_cl_heap_modint_ring()
public:
// Constructor.
- cl_heap_modint_ring_pow2 (const cl_I& m, uintL m1); // m = 2^m1
+ cl_heap_modint_ring_pow2 (const cl_I& m, uintC m1); // m = 2^m1
// Destructor.
~cl_heap_modint_ring_pow2 () {}
// Additional information.
- uintL m1;
+ uintC m1;
};
static inline const cl_I pow2_reduce_modulo (cl_heap_modint_ring* _R, const cl_I& x)
};
// Constructor.
-inline cl_heap_modint_ring_pow2::cl_heap_modint_ring_pow2 (const cl_I& m, uintL _m1)
+inline cl_heap_modint_ring_pow2::cl_heap_modint_ring_pow2 (const cl_I& m, uintC _m1)
: cl_heap_modint_ring (m, &std_setops, &pow2_addops, &pow2_mulops), m1 (_m1)
{
type = &cl_class_modint_ring_pow2;
SUBCLASS_cl_heap_modint_ring()
public:
// Constructor.
- cl_heap_modint_ring_pow2m1 (const cl_I& m, uintL m1); // m = 2^m1 - 1
+ cl_heap_modint_ring_pow2m1 (const cl_I& m, uintC m1); // m = 2^m1 - 1
// Destructor.
~cl_heap_modint_ring_pow2m1 () {}
// Additional information.
- uintL m1;
+ uintC m1;
};
static inline const cl_I pow2m1_reduce_modulo (cl_heap_modint_ring* _R, const cl_I& x)
{ Mutable(cl_I,x);
var cl_boolean sign = minusp(x);
if (sign) { x = lognot(x); }
- var const uintL m1 = R->m1;
+ var const uintC m1 = R->m1;
if (x >= R->modulus) {
x = plus1(x); // avoid staying at x = m
do {
- var uintL xlen = integer_length(x);
+ var uintC xlen = integer_length(x);
var cl_I y = ldb(x,cl_byte(m1,0));
- for (var uintL i = m1; i < xlen; i += m1)
+ for (var uintC i = m1; i < xlen; i += m1)
y = y + ldb(x,cl_byte(m1,i));
x = y;
} while (x > R->modulus);
static const _cl_MI pow2m1_mul (cl_heap_modint_ring* _R, const _cl_MI& x, const _cl_MI& y)
{
var cl_heap_modint_ring_pow2m1* R = (cl_heap_modint_ring_pow2m1*)_R;
- var const uintL m1 = R->m1;
+ var const uintC m1 = R->m1;
var cl_I zr = x.rep * y.rep;
zr = ldb(zr,cl_byte(m1,m1)) + ldb(zr,cl_byte(m1,0));
return _cl_MI(R, zr >= R->modulus ? zr - R->modulus : zr);
static const _cl_MI pow2m1_square (cl_heap_modint_ring* _R, const _cl_MI& x)
{
var cl_heap_modint_ring_pow2m1* R = (cl_heap_modint_ring_pow2m1*)_R;
- var const uintL m1 = R->m1;
+ var const uintC m1 = R->m1;
var cl_I zr = square(x.rep);
zr = ldb(zr,cl_byte(m1,m1)) + ldb(zr,cl_byte(m1,0));
return _cl_MI(R, zr >= R->modulus ? zr - R->modulus : zr);
};
// Constructor.
-inline cl_heap_modint_ring_pow2m1::cl_heap_modint_ring_pow2m1 (const cl_I& m, uintL _m1)
+inline cl_heap_modint_ring_pow2m1::cl_heap_modint_ring_pow2m1 (const cl_I& m, uintC _m1)
: cl_heap_modint_ring (m, &std_setops, &pow2m1_addops, &pow2m1_mulops), m1 (_m1)
{
type = &cl_class_modint_ring_pow2m1;
SUBCLASS_cl_heap_modint_ring()
public:
// Constructor.
- cl_heap_modint_ring_pow2p1 (const cl_I& m, uintL m1); // m = 2^m1 + 1
+ cl_heap_modint_ring_pow2p1 (const cl_I& m, uintC m1); // m = 2^m1 + 1
// Destructor.
~cl_heap_modint_ring_pow2p1 () {}
// Additional information.
- uintL m1;
+ uintC m1;
};
static inline const cl_I pow2p1_reduce_modulo (cl_heap_modint_ring* _R, const cl_I& x)
{ Mutable(cl_I,x);
var bool sign = minusp(x);
if (sign) { x = lognot(x); }
- var const uintL m1 = R->m1;
+ var const uintC m1 = R->m1;
while (x >= R->modulus) {
- var uintL xlen = integer_length(x);
+ var uintC xlen = integer_length(x);
var cl_I y = ldb(x,cl_byte(m1,0));
- for (var uintL i = m1; ; ) {
+ for (var uintC i = m1; ; ) {
y = y - ldb(x,cl_byte(m1,i));
i += m1;
if (i >= xlen)
static const _cl_MI pow2p1_mul (cl_heap_modint_ring* _R, const _cl_MI& x, const _cl_MI& y)
{
var cl_heap_modint_ring_pow2p1* R = (cl_heap_modint_ring_pow2p1*)_R;
- var const uintL m1 = R->m1;
+ var const uintC m1 = R->m1;
var cl_I zr = x.rep * y.rep;
// Now 0 <= zr <= 2^(2*m1).
zr = ldb(zr,cl_byte(1,2*m1)) - ldb(zr,cl_byte(m1,m1)) + ldb(zr,cl_byte(m1,0));
static const _cl_MI pow2p1_square (cl_heap_modint_ring* _R, const _cl_MI& x)
{
var cl_heap_modint_ring_pow2p1* R = (cl_heap_modint_ring_pow2p1*)_R;
- var const uintL m1 = R->m1;
+ var const uintC m1 = R->m1;
var cl_I zr = square(x.rep);
// Now 0 <= zr <= 2^(2*m1).
zr = ldb(zr,cl_byte(1,2*m1)) - ldb(zr,cl_byte(m1,m1)) + ldb(zr,cl_byte(m1,0));
};
// Constructor.
-inline cl_heap_modint_ring_pow2p1::cl_heap_modint_ring_pow2p1 (const cl_I& m, uintL _m1)
+inline cl_heap_modint_ring_pow2p1::cl_heap_modint_ring_pow2p1 (const cl_I& m, uintC _m1)
: cl_heap_modint_ring (m, &std_setops, &pow2p1_addops, &pow2p1_mulops), m1 (_m1)
{
type = &cl_class_modint_ring_pow2p1;
namespace cln {
-const cl_MI operator>> (const cl_MI& x, sintL y) // assume 0 <= y < 2^31
+const cl_MI operator>> (const cl_MI& x, sintC y) // assume 0 <= y < 2^(intCsize-1)
{
if (y == 0)
return x;
// k = 3 for nn <= 69.8
// ... k for nn <= k*(k+1)*2^(2k)/(2^(k+1)-k-2)
var cl_I n = y;
- var uintL nn = integer_length(n);
+ var uintC nn = integer_length(n);
// n has nn bits.
if (nn <= 8) {
// k = 1, normal Left-Right Binary algorithm.
else if (nn <= 1721160) k = 14;
else if (nn <= 3933180) k = 15;
else /* if (nn <= 8914120) */ k = 16;
- var uintL nnk = ceiling(nn,k); // number of base-2^k digits in n
+ var uintC nnk = ceiling(nn,k); // number of base-2^k digits in n
var uint16* n_digits = cl_alloc_array(uint16,nnk);
// Split n into base-2^k digits.
{
var const uintL k_mask = bit(k)-1;
var uintD carry = 0;
var unsigned int carrybits = 0;
- for (var uintL i = 0; i < nnk; i++) {
+ for (var uintC i = 0; i < nnk; i++) {
if (carrybits >= k) {
n_digits[i] = carry & k_mask;
carry = carry >> k;
// Compute maximum odd base-2^k digit.
var uintL maxodd = 1;
if (k <= 8) {
- for (var uintL i = 0; i < nnk; i++) {
+ for (var uintC i = 0; i < nnk; i++) {
var uintL d = n_digits[i];
if (d > 0) {
d = odd_table[d];
}
}
} else {
- for (var uintL i = 0; i < nnk; i++) {
+ for (var uintC i = 0; i < nnk; i++) {
var uintL d = n_digits[i];
if (d > 0) {
var uintL d2; ord2_32(d,d2=);
for ( ; d2>0; d2--)
a = R->_square(a);
}
- for (var sintL i = nnk-2; i >= 0; i--) {
+ for (var sintC i = nnk-2; i >= 0; i--) {
// Compute a := a^(2^k) * x^n_digits[i].
var uintL d = n_digits[i];
var uintL d2;
// [Cohen], section 8.2, algorithm 8.2.2.
var cl_modint_ring R = find_modint_ring(n); // Z/nZ
var cl_I m = n-1;
- var uintL e = ord2(m);
+ var uintC e = ord2(m);
m = m>>e;
// n-1 = 2^e*m
var cl_MI one = R->one();
var cl_MI b = R->expt_pos(a,m); // b = a^m
if (b == one)
goto passed;
- for (uintL s = e; s > 0; s--) {
+ for (uintC s = e; s > 0; s--) {
if (b == minusone)
goto passed;
var cl_MI new_b = R->square(b);
var int count = 50;
// Step 1: Trial division (rules out 87% of all numbers quickly).
const uint32 trialdivide_limit = 70;
- var uintL l = integer_length(n);
+ var uintC l = integer_length(n);
if (l <= 32) {
var uint32 nn = cl_I_to_UL(n);
if (nn <= cl_small_prime_table_limit) {
// take h in G_(j+1) \ G_j, so that h^2 in G_j \ G_(j-1), and
// a^-1*b^2*h^2 is in G_(j-1). So multiply b with h.
var cl_I& p = R->modulus;
- var uintL e = ord2(p-1);
+ var uintC e = ord2(p-1);
var cl_I m = (p-1) >> e;
// p-1 = 2^e*m, m odd.
// We will have the invariant c = a^-1*b^2 in G/G_j.
- var uintL j = e;
+ var uintC j = e;
// Initialize b = a^((m+1)/2), c = a^m, but avoid to divide by a.
var cl_MI c = R->expt_pos(a,(m-1)>>1);
var cl_MI b = R->mul(a,c);
do {
// Now c = a^-1*b^2 in G_j, h in G_j \ G_(j-1).
// Determine the smallest i such that c in G_i.
- var uintL i = 0;
+ var uintC i = 0;
var cl_MI ci = c; // c_i = c^(2^i)
for ( ; i < j; i++, ci = R->square(ci))
if (ci == R->one())
// Indicates that p is not prime.
return new cl_composite_condition(p);
// OK, i < j.
- for (var uintL count = j-i-1; count > 0; count--)
+ for (var uintC count = j-i-1; count > 0; count--)
h = R->square(h);
// Now h in G_(i+1) \ G_i.
b = R->mul(b,h);
else
return sqrt_mod_p_t(2,R->canonhom(x1),R->canonhom(x2));
}
- var uintL l = integer_length(p);
- var uintL e = ord2(p-1);
+ var uintC l = integer_length(p);
+ var uintC e = ord2(p-1);
//if (e > 30 && e > l/(::log((double)l)*0.72-1))
if (e > 30 && e > l/(::log((double)l)*0.92-2.41))
// Algorithm 2.
return cl_MI(R, x[0]);
else {
var uintL count = ceiling(len,intDsize);
- var uintL bitcount = 0;
+ var uintC bitcount = 0;
do {
count--;
bitcount += logcountD(xv->data[count]);
const cl_F cl_float (const cl_R& x, float_format_t f)
{
- floatformatcase((uintL)f
+ floatformatcase((uintC)f
, return cl_R_to_SF(x);
, return cl_R_to_FF(x);
, return cl_R_to_DF(x);
var cl_idecoded_float decoded = integer_decode_float(arg);
var const cl_I& significand = decoded.mantissa;
var const cl_I& expon = decoded.exponent;
- var uintL mantprec = float_digits(arg)-1;
+ var uintC mantprec = float_digits(arg)-1;
var cl_I numerator = significand;
var cl_I denominator = 1;
var cl_I abrund_einh = 1; // Abrundungseinheit:
ptr_after_prec = skip_digits(ptr,string_limit,10);
if (ptr_after_prec == ptr) goto not_float_syntax;
var cl_I prec1 = digits_to_I(ptr,ptr_after_prec-ptr,10);
- var uintL prec2 = cl_I_to_UL(prec1);
+ var uintC prec2 = cl_I_to_ulong(prec1);
prec = (float_base==10 ? float_format(prec2)
- : (float_format_t)((uintL)((1+prec2)*::log((double)float_base)*1.442695041)+1)
+ : (float_format_t)((uintC)((1+prec2)*::log((double)float_base)*1.442695041)+1)
);
} else {
switch (exponent_marker) {
// Count the number of significant digits.
ptr = ptr_after_sign;
while (ptr < ptr_after_fracpart && (*ptr == '0' || *ptr == '.')) ptr++;
- var uintL num_significant_digits =
+ var uintC num_significant_digits =
(ptr_after_fracpart - ptr) - (ptr_before_fracpart > ptr ? 1 : 0);
- var uintL prec2 = (num_significant_digits>=2 ? num_significant_digits-2 : 0);
+ var uintC prec2 = (num_significant_digits>=2 ? num_significant_digits-2 : 0);
var float_format_t precx =
(float_base==10 ? float_format(prec2)
- : (float_format_t)((uintL)((1+prec2)*::log((double)float_base)*1.442695041)+1)
+ : (float_format_t)((uintC)((1+prec2)*::log((double)float_base)*1.442695041)+1)
);
- if ((uintL)precx > (uintL)prec)
+ if ((uintC)precx > (uintC)prec)
prec = precx;
}
}
struct cl_GV_I_vectorops {
cl_GV_vectorops<cl_I> ops;
- sintL m; // for maxbits
+ sintC m; // for maxbits
};
static inline cl_GV_I_vectorops* outcast (cl_GV_vectorops<cl_I>* vectorops)
cl_heap_GV_I_general ();
};
-static const cl_I general_element (const cl_GV_inner<cl_I>* vec, uintL index)
+static const cl_I general_element (const cl_GV_inner<cl_I>* vec, uintC index)
{
return ((const cl_heap_GV_I_general *) outcast(vec))->data[index];
}
-static void general_set_element (cl_GV_inner<cl_I>* vec, uintL index, const cl_I& x)
+static void general_set_element (cl_GV_inner<cl_I>* vec, uintC index, const cl_I& x)
{
((cl_heap_GV_I_general *) outcast(vec))->data[index] = x;
}
static void general_do_delete (cl_GV_inner<cl_I>* vec)
{
var cl_heap_GV_I_general* hv = (cl_heap_GV_I_general *) outcast(vec);
- var uintL len = hv->v.length();
- for (var uintL i = 0; i < len; i++)
+ var uintC len = hv->v.length();
+ for (var uintC i = 0; i < len; i++)
hv->data[i].~cl_I();
}
-static void general_copy_elements (const cl_GV_inner<cl_I>* srcvec, uintL srcindex, cl_GV_inner<cl_I>* destvec, uintL destindex, uintL count)
+static void general_copy_elements (const cl_GV_inner<cl_I>* srcvec, uintC srcindex, cl_GV_inner<cl_I>* destvec, uintC destindex, uintC count)
{
if (count > 0) {
var const cl_heap_GV_I_general* srcv =
(const cl_heap_GV_I_general *) outcast(srcvec);
var cl_heap_GV_I_general* destv =
(cl_heap_GV_I_general *) outcast(destvec);
- var uintL srclen = srcv->v.length();
- var uintL destlen = destv->v.length();
+ var uintC srclen = srcv->v.length();
+ var uintC destlen = destv->v.length();
if (!(srcindex <= srcindex+count && srcindex+count <= srclen))
cl_abort();
if (!(destindex <= destindex+count && destindex+count <= destlen))
-1
};
-cl_heap_GV_I* cl_make_heap_GV_I (uintL len)
+cl_heap_GV_I* cl_make_heap_GV_I (uintC len)
{
var cl_heap_GV_I_general* hv = (cl_heap_GV_I_general*) malloc_hook(offsetofa(cl_heap_GV_I_general,data)+sizeof(cl_I)*len);
hv->refcount = 1;
hv->type = &cl_class_gvector_integer;
new (&hv->v) cl_GV_inner<cl_I> (len,&general_vectorops.ops);
- for (var uintL i = 0; i < len; i++)
+ for (var uintC i = 0; i < len; i++)
init1(cl_I, hv->data[i]) ();
return hv;
}
/* No default constructor. */ \
cl_heap_GV_I_bits##m (); \
}; \
-static const cl_I bits##m##_element (const cl_GV_inner<cl_I>* vec, uintL index); \
-static void bits##m##_set_element (cl_GV_inner<cl_I>* vec, uintL index, const cl_I& x); \
-static void bits##m##_copy_elements (const cl_GV_inner<cl_I>* srcvec, uintL srcindex, cl_GV_inner<cl_I>* destvec, uintL destindex, uintL count) \
+static const cl_I bits##m##_element (const cl_GV_inner<cl_I>* vec, uintC index); \
+static void bits##m##_set_element (cl_GV_inner<cl_I>* vec, uintC index, const cl_I& x); \
+static void bits##m##_copy_elements (const cl_GV_inner<cl_I>* srcvec, uintC srcindex, cl_GV_inner<cl_I>* destvec, uintC destindex, uintC count) \
{ \
if (count > 0) { \
var const cl_heap_GV_I_bits##m * srcv = \
(const cl_heap_GV_I_bits##m *) outcast(srcvec); \
var cl_heap_GV_I_bits##m * destv = \
(cl_heap_GV_I_bits##m *) outcast(destvec); \
- var uintL srclen = srcv->v.length(); \
- var uintL destlen = destv->v.length(); \
+ var uintC srclen = srcv->v.length(); \
+ var uintC destlen = destv->v.length(); \
if (!(srcindex <= srcindex+count && srcindex+count <= srclen)) \
cl_abort(); \
if (!(destindex <= destindex+count && destindex+count <= destlen)) \
// Copy bits srcptr.bits[srcindex..srcindex+count-1] into destptr.bits[destindex..destindex+count-1].
// Assumes that all range checks have already been performed.
-static void bits_copy (const uintD* srcptr, uintL srcindex, uintD* destptr, uintL destindex, uintL count)
+static void bits_copy (const uintD* srcptr, uintC srcindex, uintD* destptr, uintC destindex, uintC count)
{
srcptr += floor(srcindex,intDsize);
destptr += floor(destindex,intDsize);
count -= intDsize-srcindex;
}
// Now srcindex and destindex can be assumed to be 0.
- var uintL count1 = count%intDsize;
+ var uintC count1 = count%intDsize;
count = floor(count,intDsize);
if (count > 0) {
do {
*destptr ^= (*destptr ^ *srcptr) & (uintD)(bit(count1)-1);
}
} else {
- var uintL i = destindex - srcindex;
+ var uintC i = destindex - srcindex;
var uintD tmp;
if (destindex >= srcindex) { // i > 0
if (count <= intDsize-destindex) {
}
srcptr++;
// tmp now contains the low i bits to be put into *destptr.
- var uintL count1 = count%intDsize;
+ var uintC count1 = count%intDsize;
count = floor(count,intDsize);
var uintD lastdest;
if (count == 0)
DEFINE_cl_heap_GV_I_bits(1,uintD)
-static const cl_I bits1_element (const cl_GV_inner<cl_I>* vec, uintL index)
+static const cl_I bits1_element (const cl_GV_inner<cl_I>* vec, uintC index)
{
return (unsigned int)((((const cl_heap_GV_I_bits1 *) outcast(vec))->data[index/intDsize] >> (index%intDsize)) & 0x1);
}
-static void bits1_set_element (cl_GV_inner<cl_I>* vec, uintL index, const cl_I& x)
+static void bits1_set_element (cl_GV_inner<cl_I>* vec, uintC index, const cl_I& x)
{
var uintV xval;
if (fixnump(x)) {
DEFINE_cl_heap_GV_I_bits(2,uintD)
-static const cl_I bits2_element (const cl_GV_inner<cl_I>* vec, uintL index)
+static const cl_I bits2_element (const cl_GV_inner<cl_I>* vec, uintC index)
{
return (unsigned int)((((const cl_heap_GV_I_bits2 *) outcast(vec))->data[index/(intDsize/2)] >> (2*(index%(intDsize/2)))) & 0x3);
}
-static void bits2_set_element (cl_GV_inner<cl_I>* vec, uintL index, const cl_I& x)
+static void bits2_set_element (cl_GV_inner<cl_I>* vec, uintC index, const cl_I& x)
{
var uintV xval;
if (fixnump(x)) {
DEFINE_cl_heap_GV_I_bits(4,uintD)
-static const cl_I bits4_element (const cl_GV_inner<cl_I>* vec, uintL index)
+static const cl_I bits4_element (const cl_GV_inner<cl_I>* vec, uintC index)
{
return (unsigned int)((((const cl_heap_GV_I_bits4 *) outcast(vec))->data[index/(intDsize/4)] >> (4*(index%(intDsize/4)))) & 0xF);
}
-static void bits4_set_element (cl_GV_inner<cl_I>* vec, uintL index, const cl_I& x)
+static void bits4_set_element (cl_GV_inner<cl_I>* vec, uintC index, const cl_I& x)
{
var uintV xval;
if (fixnump(x)) {
DEFINE_cl_heap_GV_I_bits(8,uintD)
-static const cl_I bits8_element (const cl_GV_inner<cl_I>* vec, uintL index)
+static const cl_I bits8_element (const cl_GV_inner<cl_I>* vec, uintC index)
{
#if CL_CPU_BIG_ENDIAN_P
return (unsigned int)((((const cl_heap_GV_I_bits8 *) outcast(vec))->data[index/(intDsize/8)] >> (8*(index%(intDsize/8)))) & 0xFF);
return (unsigned int)(((uint8*)(((const cl_heap_GV_I_bits8 *) outcast(vec))->data))[index]);
#endif
}
-static void bits8_set_element (cl_GV_inner<cl_I>* vec, uintL index, const cl_I& x)
+static void bits8_set_element (cl_GV_inner<cl_I>* vec, uintC index, const cl_I& x)
{
var uintV xval;
if (fixnump(x)) {
DEFINE_cl_heap_GV_I_bits(16,uintD)
-static const cl_I bits16_element (const cl_GV_inner<cl_I>* vec, uintL index)
+static const cl_I bits16_element (const cl_GV_inner<cl_I>* vec, uintC index)
{
#if CL_CPU_BIG_ENDIAN_P
return (unsigned int)((((const cl_heap_GV_I_bits16 *) outcast(vec))->data[index/(intDsize/16)] >> (16*(index%(intDsize/16)))) & 0xFFFF);
return (unsigned int)(((uint16*)(((const cl_heap_GV_I_bits16 *) outcast(vec))->data))[index]);
#endif
}
-static void bits16_set_element (cl_GV_inner<cl_I>* vec, uintL index, const cl_I& x)
+static void bits16_set_element (cl_GV_inner<cl_I>* vec, uintC index, const cl_I& x)
{
var uintV xval;
if (fixnump(x)) {
DEFINE_cl_heap_GV_I_bits(32,uintD)
-static const cl_I bits32_element (const cl_GV_inner<cl_I>* vec, uintL index)
+static const cl_I bits32_element (const cl_GV_inner<cl_I>* vec, uintC index)
{
#if (intDsize==32)
return (unsigned long)(((const cl_heap_GV_I_bits32 *) outcast(vec))->data[index]);
return (unsigned long)(((uint32*)(((const cl_heap_GV_I_bits32 *) outcast(vec))->data))[index]);
#endif
}
-static void bits32_set_element (cl_GV_inner<cl_I>* vec, uintL index, const cl_I& x)
+static void bits32_set_element (cl_GV_inner<cl_I>* vec, uintC index, const cl_I& x)
{
var uint32 xval = cl_I_to_UL(x);
#if (intDsize==32)
&bits32_vectorops
};
-cl_heap_GV_I* cl_make_heap_GV_I (uintL len, sintL m)
+cl_heap_GV_I* cl_make_heap_GV_I (uintC len, sintC m)
{
// Determine log2(bits).
var uintL log2_bits;
return cl_make_heap_GV_I(len);
}
// For room allocation purposes, be pessimistic: assume the uintD case (since intDsize>=32).
- var uintL words = // ceiling(len*2^log2_bits,intDsize)
- (((sintL)len-1)>>(log2_intDsize-log2_bits))+1;
+ var uintC words = // ceiling(len*2^log2_bits,intDsize)
+ (((sintC)len-1)>>(log2_intDsize-log2_bits))+1;
var cl_heap_GV_I_bits32* hv = (cl_heap_GV_I_bits32*) malloc_hook(offsetofa(cl_heap_GV_I_bits32,data)+sizeof(uintD)*words);
hv->refcount = 1;
hv->type = &cl_class_gvector_integer;
new (&hv->v) cl_GV_inner<cl_I> (len,&bits_vectorops[log2_bits]->ops);
var uintD* ptr = (uintD*)(hv->data);
- for (var uintL i = 0; i < words; i++)
+ for (var uintC i = 0; i < words; i++)
ptr[i] = 0;
return (cl_heap_GV_I*) hv;
}
-sintL cl_heap_GV_I::maxbits () const
+sintC cl_heap_GV_I::maxbits () const
{
return outcast(v.vectorops)->m;
}
// An empty vector.
-const cl_GV_I cl_null_GV_I = cl_GV_I((uintL)0);
+const cl_GV_I cl_null_GV_I = cl_GV_I((uintC)0);
} // namespace cln
const cl_GV_I copy (const cl_GV_I& v)
{
- var uintL len = v.length();
+ var uintC len = v.length();
var cl_GV_I w = cl_GV_I(len,v.maxbits());
cl_GV_I::copy_elements(v,0,w,0,len);
return w;
cl_heap_GV_number_general ();
};
-static const cl_number general_element (const cl_GV_inner<cl_number>* vec, uintL index)
+static const cl_number general_element (const cl_GV_inner<cl_number>* vec, uintC index)
{
return ((const cl_heap_GV_number_general *) outcast(vec))->data[index];
}
-static void general_set_element (cl_GV_inner<cl_number>* vec, uintL index, const cl_number& x)
+static void general_set_element (cl_GV_inner<cl_number>* vec, uintC index, const cl_number& x)
{
((cl_heap_GV_number_general *) outcast(vec))->data[index] = x;
}
static void general_do_delete (cl_GV_inner<cl_number>* vec)
{
var cl_heap_GV_number_general* hv = (cl_heap_GV_number_general *) outcast(vec);
- var uintL len = hv->v.length();
- for (var uintL i = 0; i < len; i++)
+ var uintC len = hv->v.length();
+ for (var uintC i = 0; i < len; i++)
hv->data[i].~cl_number();
}
-static void general_copy_elements (const cl_GV_inner<cl_number>* srcvec, uintL srcindex, cl_GV_inner<cl_number>* destvec, uintL destindex, uintL count)
+static void general_copy_elements (const cl_GV_inner<cl_number>* srcvec, uintC srcindex, cl_GV_inner<cl_number>* destvec, uintC destindex, uintC count)
{
if (count > 0) {
var const cl_heap_GV_number_general* srcv =
(const cl_heap_GV_number_general *) outcast(srcvec);
var cl_heap_GV_number_general* destv =
(cl_heap_GV_number_general *) outcast(destvec);
- var uintL srclen = srcv->v.length();
- var uintL destlen = destv->v.length();
+ var uintC srclen = srcv->v.length();
+ var uintC destlen = destv->v.length();
if (!(srcindex <= srcindex+count && srcindex+count <= srclen))
cl_abort();
if (!(destindex <= destindex+count && destindex+count <= destlen))
general_copy_elements
};
-cl_heap_GV_number* cl_make_heap_GV_number (uintL len)
+cl_heap_GV_number* cl_make_heap_GV_number (uintC len)
{
var cl_heap_GV_number_general* hv = (cl_heap_GV_number_general*) malloc_hook(offsetofa(cl_heap_GV_number_general,data)+sizeof(cl_number)*len);
hv->refcount = 1;
hv->type = &cl_class_gvector_number;
new (&hv->v) cl_GV_inner<cl_number> (len,&general_vectorops);
- for (var uintL i = 0; i < len; i++)
+ for (var uintC i = 0; i < len; i++)
init1(cl_number, hv->data[i]) ();
return hv;
}
// An empty vector.
-const cl_GV_number cl_null_GV_number = cl_GV_number((uintL)0);
+const cl_GV_number cl_null_GV_number = cl_GV_number((uintC)0);
} // namespace cln
const cl_GV_number copy (const cl_GV_number& v)
{
- var uintL len = v.length();
+ var uintC len = v.length();
var cl_GV_number w = cl_GV_number(len);
cl_GV_number::copy_elements(v,0,w,0,len);
return w;
const cl_SV_any copy (const cl_SV_any& src)
{
- var uintL len = src.length();
+ var uintC len = src.length();
var cl_heap_SV_any* hv = (cl_heap_SV_any*) malloc_hook(sizeof(cl_heap_SV_any)+sizeof(cl_gcobject)*len);
hv->refcount = 1;
hv->type = src.pointer_type();
new (&hv->v) cl_SV_inner<cl_gcobject> (len);
- for (var uintL i = 0; i < len; i++)
+ for (var uintC i = 0; i < len; i++)
init1(cl_gcobject, hv->v[i]) (src[i]);
return hv;
}
0
};
-cl_heap_SV_number* cl_make_heap_SV_number_uninit (uintL len)
+cl_heap_SV_number* cl_make_heap_SV_number_uninit (uintC len)
{
var cl_heap_SV_number* hv = (cl_heap_SV_number*) malloc_hook(sizeof(cl_heap_SV_number)+sizeof(cl_number)*len);
hv->refcount = 1;
return hv;
}
-cl_heap_SV_number* cl_make_heap_SV_number (uintL len)
+cl_heap_SV_number* cl_make_heap_SV_number (uintC len)
{
var cl_heap_SV_number* hv = (cl_heap_SV_number*) malloc_hook(sizeof(cl_heap_SV_number)+sizeof(cl_number)*len);
hv->refcount = 1;
hv->type = &cl_class_svector_number;
new (&hv->v) cl_SV_inner<cl_number> (len);
- for (var uintL i = 0; i < len; i++)
+ for (var uintC i = 0; i < len; i++)
init1(cl_number, hv->v[i]) (0);
return hv;
}
// An empty vector.
-const cl_SV_number cl_null_SV_number = cl_SV_number((uintL)0);
+const cl_SV_number cl_null_SV_number = cl_SV_number((uintC)0);
} // namespace cln
0
};
-cl_heap_SV_ringelt* cl_make_heap_SV_ringelt_uninit (uintL len)
+cl_heap_SV_ringelt* cl_make_heap_SV_ringelt_uninit (uintC len)
{
var cl_heap_SV_ringelt* hv = (cl_heap_SV_ringelt*) malloc_hook(sizeof(cl_heap_SV_ringelt)+sizeof(_cl_ring_element)*len);
hv->refcount = 1;
return hv;
}
-cl_heap_SV_ringelt* cl_make_heap_SV_ringelt (uintL len)
+cl_heap_SV_ringelt* cl_make_heap_SV_ringelt (uintC len)
{
var cl_heap_SV_ringelt* hv = (cl_heap_SV_ringelt*) malloc_hook(sizeof(cl_heap_SV_ringelt)+sizeof(_cl_ring_element)*len);
hv->refcount = 1;
hv->type = &cl_class_svector_ringelt;
new (&hv->v) cl_SV_inner<_cl_ring_element> (len);
- for (var uintL i = 0; i < len; i++)
+ for (var uintC i = 0; i < len; i++)
init1(_cl_ring_element, hv->v[i]) ();
return hv;
}
// An empty vector.
-const cl_SV_ringelt cl_null_SV_ringelt = cl_SV_ringelt((uintL)0);
+const cl_SV_ringelt cl_null_SV_ringelt = cl_SV_ringelt((uintC)0);
} // namespace cln
#if 0
float_format_t f = float_format(atoi(argv[1]));
extern cl_LF zeta3 (uintC len);
- uintC len = (uintL)f/intDsize+1;
+ uintC len = (uintC)f/intDsize+1;
{ CL_TIMING; cout << zeta(2,f) << endl; }
{ CL_TIMING; cout << expt(pi(f),2)/6 << endl; }
{ CL_TIMING; cout << zeta(3,f) << endl; }
// Check against ash.
for (i = iterations; i > 0; i--) {
cl_I a = testrandom_I();
- uintL l = integer_length(a);
+ uintC l = integer_length(a);
if (a >= 0) {
int b = 0;
if (a < ash(1,l)) {
for (i = iterations; i > 0; i--) {
cl_I a = testrandom_I();
if (a != 0) {
- uintL n = ord2(a);
- cl_I b = ash(a,-(sintL)n);
+ uintC n = ord2(a);
+ cl_I b = ash(a,-(sintC)n);
ASSERT1(oddp(b), a);
- ASSERT1(a == ash(b,(sintL)n), a);
+ ASSERT1(a == ash(b,(sintC)n), a);
}
}
return error;
projects / cln.git / commitdiff