| 1 |
/* Copyright (C) 2007-2025 Free Software Foundation, Inc. |
| 2 |
|
| 3 |
This file is part of GCC. |
| 4 |
|
| 5 |
GCC is free software; you can redistribute it and/or modify it under |
| 6 |
the terms of the GNU General Public License as published by the Free |
| 7 |
Software Foundation; either version 3, or (at your option) any later |
| 8 |
version. |
| 9 |
|
| 10 |
GCC is distributed in the hope that it will be useful, but WITHOUT ANY |
| 11 |
WARRANTY; without even the implied warranty of MERCHANTABILITY or |
| 12 |
FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License |
| 13 |
for more details. |
| 14 |
|
| 15 |
Under Section 7 of GPL version 3, you are granted additional |
| 16 |
permissions described in the GCC Runtime Library Exception, version |
| 17 |
3.1, as published by the Free Software Foundation. |
| 18 |
|
| 19 |
You should have received a copy of the GNU General Public License and |
| 20 |
a copy of the GCC Runtime Library Exception along with this program; |
| 21 |
see the files COPYING3 and COPYING.RUNTIME respectively. If not, see |
| 22 |
<http://www.gnu.org/licenses/>. */ |
| 23 |
|
| 24 |
#include "bid_internal.h" |
| 25 |
|
| 26 |
/***************************************************************************** |
| 27 |
* BID128_to_int64_rnint |
| 28 |
****************************************************************************/ |
| 29 |
|
| 30 |
BID128_FUNCTION_ARG1_NORND_CUSTOMRESTYPE (SINT64, bid128_to_int64_rnint, |
| 31 |
x) |
| 32 |
|
| 33 |
SINT64 res; |
| 34 |
UINT64 x_sign; |
| 35 |
UINT64 x_exp; |
| 36 |
int exp; // unbiased exponent |
| 37 |
// Note: C1.w[1], C1.w[0] represent x_signif_hi, x_signif_lo (all are UINT64) |
| 38 |
UINT64 tmp64; |
| 39 |
BID_UI64DOUBLE tmp1; |
| 40 |
unsigned int x_nr_bits; |
| 41 |
int q, ind, shift; |
| 42 |
UINT128 C1, C; |
| 43 |
UINT128 Cstar; // C* represents up to 34 decimal digits ~ 113 bits |
| 44 |
UINT256 fstar; |
| 45 |
UINT256 P256; |
| 46 |
|
| 47 |
// unpack x |
| 48 |
x_sign = x.w[1] & MASK_SIGN; // 0 for positive, MASK_SIGN for negative |
| 49 |
x_exp = x.w[1] & MASK_EXP; // biased and shifted left 49 bit positions |
| 50 |
C1.w[1] = x.w[1] & MASK_COEFF; |
| 51 |
C1.w[0] = x.w[0]; |
| 52 |
|
| 53 |
// check for NaN or Infinity |
| 54 |
if ((x.w[1] & MASK_SPECIAL) == MASK_SPECIAL) { |
| 55 |
// x is special |
| 56 |
if ((x.w[1] & MASK_NAN) == MASK_NAN) { // x is NAN |
| 57 |
if ((x.w[1] & MASK_SNAN) == MASK_SNAN) { // x is SNAN |
| 58 |
// set invalid flag |
| 59 |
*pfpsf |= INVALID_EXCEPTION; |
| 60 |
// return Integer Indefinite |
| 61 |
res = 0x8000000000000000ull; |
| 62 |
} else { // x is QNaN |
| 63 |
// set invalid flag |
| 64 |
*pfpsf |= INVALID_EXCEPTION; |
| 65 |
// return Integer Indefinite |
| 66 |
res = 0x8000000000000000ull; |
| 67 |
} |
| 68 |
BID_RETURN (res); |
| 69 |
} else { // x is not a NaN, so it must be infinity |
| 70 |
if (!x_sign) { // x is +inf |
| 71 |
// set invalid flag |
| 72 |
*pfpsf |= INVALID_EXCEPTION; |
| 73 |
// return Integer Indefinite |
| 74 |
res = 0x8000000000000000ull; |
| 75 |
} else { // x is -inf |
| 76 |
// set invalid flag |
| 77 |
*pfpsf |= INVALID_EXCEPTION; |
| 78 |
// return Integer Indefinite |
| 79 |
res = 0x8000000000000000ull; |
| 80 |
} |
| 81 |
BID_RETURN (res); |
| 82 |
} |
| 83 |
} |
| 84 |
// check for non-canonical values (after the check for special values) |
| 85 |
if ((C1.w[1] > 0x0001ed09bead87c0ull) || |
| 86 |
(C1.w[1] == 0x0001ed09bead87c0ull |
| 87 |
&& (C1.w[0] > 0x378d8e63ffffffffull)) |
| 88 |
|| ((x.w[1] & 0x6000000000000000ull) == 0x6000000000000000ull)) { |
| 89 |
res = 0x0000000000000000ull; |
| 90 |
BID_RETURN (res); |
| 91 |
} else if ((C1.w[1] == 0x0ull) && (C1.w[0] == 0x0ull)) { |
| 92 |
// x is 0 |
| 93 |
res = 0x0000000000000000ull; |
| 94 |
BID_RETURN (res); |
| 95 |
} else { // x is not special and is not zero |
| 96 |
|
| 97 |
// q = nr. of decimal digits in x |
| 98 |
// determine first the nr. of bits in x |
| 99 |
if (C1.w[1] == 0) { |
| 100 |
if (C1.w[0] >= 0x0020000000000000ull) { // x >= 2^53 |
| 101 |
// split the 64-bit value in two 32-bit halves to avoid rounding errors |
| 102 |
if (C1.w[0] >= 0x0000000100000000ull) { // x >= 2^32 |
| 103 |
tmp1.d = (double) (C1.w[0] >> 32); // exact conversion |
| 104 |
x_nr_bits = |
| 105 |
33 + ((((unsigned int) (tmp1.ui64 >> 52)) & 0x7ff) - 0x3ff); |
| 106 |
} else { // x < 2^32 |
| 107 |
tmp1.d = (double) (C1.w[0]); // exact conversion |
| 108 |
x_nr_bits = |
| 109 |
1 + ((((unsigned int) (tmp1.ui64 >> 52)) & 0x7ff) - 0x3ff); |
| 110 |
} |
| 111 |
} else { // if x < 2^53 |
| 112 |
tmp1.d = (double) C1.w[0]; // exact conversion |
| 113 |
x_nr_bits = |
| 114 |
1 + ((((unsigned int) (tmp1.ui64 >> 52)) & 0x7ff) - 0x3ff); |
| 115 |
} |
| 116 |
} else { // C1.w[1] != 0 => nr. bits = 64 + nr_bits (C1.w[1]) |
| 117 |
tmp1.d = (double) C1.w[1]; // exact conversion |
| 118 |
x_nr_bits = |
| 119 |
65 + ((((unsigned int) (tmp1.ui64 >> 52)) & 0x7ff) - 0x3ff); |
| 120 |
} |
| 121 |
q = nr_digits[x_nr_bits - 1].digits; |
| 122 |
if (q == 0) { |
| 123 |
q = nr_digits[x_nr_bits - 1].digits1; |
| 124 |
if (C1.w[1] > nr_digits[x_nr_bits - 1].threshold_hi |
| 125 |
|| (C1.w[1] == nr_digits[x_nr_bits - 1].threshold_hi |
| 126 |
&& C1.w[0] >= nr_digits[x_nr_bits - 1].threshold_lo)) |
| 127 |
q++; |
| 128 |
} |
| 129 |
exp = (x_exp >> 49) - 6176; |
| 130 |
if ((q + exp) > 19) { // x >= 10^19 ~= 2^63.11... (cannot fit in SINT64) |
| 131 |
// set invalid flag |
| 132 |
*pfpsf |= INVALID_EXCEPTION; |
| 133 |
// return Integer Indefinite |
| 134 |
res = 0x8000000000000000ull; |
| 135 |
BID_RETURN (res); |
| 136 |
} else if ((q + exp) == 19) { // x = c(0)c(1)...c(18).c(19)...c(q-1) |
| 137 |
// in this case 2^63.11... ~= 10^19 <= x < 10^20 ~= 2^66.43... |
| 138 |
// so x rounded to an integer may or may not fit in a signed 64-bit int |
| 139 |
// the cases that do not fit are identified here; the ones that fit |
| 140 |
// fall through and will be handled with other cases further, |
| 141 |
// under '1 <= q + exp <= 19' |
| 142 |
if (x_sign) { // if n < 0 and q + exp = 19 |
| 143 |
// if n < -2^63 - 1/2 then n is too large |
| 144 |
// too large if c(0)c(1)...c(18).c(19)...c(q-1) > 2^63+1/2 |
| 145 |
// <=> 0.c(0)c(1)...c(q-1) * 10^20 > 5*(2^64+1), 1<=q<=34 |
| 146 |
// <=> 0.c(0)c(1)...c(q-1) * 10^20 > 0x50000000000000005, 1<=q<=34 |
| 147 |
C.w[1] = 0x0000000000000005ull; |
| 148 |
C.w[0] = 0000000000000005ull; |
| 149 |
if (q <= 19) { // 1 <= q <= 19 => 1 <= 20-q <= 19 => |
| 150 |
// 10^(20-q) is 64-bit, and so is C1 |
| 151 |
__mul_64x64_to_128MACH (C1, C1.w[0], ten2k64[20 - q]); |
| 152 |
} else if (q == 20) { |
| 153 |
; // C1 * 10^0 = C1 |
| 154 |
} else { // if 21 <= q <= 34 |
| 155 |
__mul_128x64_to_128 (C, ten2k64[q - 20], C); // max 47-bit x 67-bit |
| 156 |
} |
| 157 |
if (C1.w[1] > C.w[1] || (C1.w[1] == C.w[1] && C1.w[0] > C.w[0])) { |
| 158 |
// set invalid flag |
| 159 |
*pfpsf |= INVALID_EXCEPTION; |
| 160 |
// return Integer Indefinite |
| 161 |
res = 0x8000000000000000ull; |
| 162 |
BID_RETURN (res); |
| 163 |
} |
| 164 |
// else cases that can be rounded to a 64-bit int fall through |
| 165 |
// to '1 <= q + exp <= 19' |
| 166 |
} else { // if n > 0 and q + exp = 19 |
| 167 |
// if n >= 2^63 - 1/2 then n is too large |
| 168 |
// too large if c(0)c(1)...c(18).c(19)...c(q-1) >= 2^63-1/2 |
| 169 |
// <=> if 0.c(0)c(1)...c(q-1) * 10^20 >= 5*(2^64-1), 1<=q<=34 |
| 170 |
// <=> if 0.c(0)c(1)...c(q-1) * 10^20 >= 0x4fffffffffffffffb, 1<=q<=34 |
| 171 |
C.w[1] = 0x0000000000000004ull; |
| 172 |
C.w[0] = 0xfffffffffffffffbull; |
| 173 |
if (q <= 19) { // 1 <= q <= 19 => 1 <= 20-q <= 19 => |
| 174 |
// 10^(20-q) is 64-bit, and so is C1 |
| 175 |
__mul_64x64_to_128MACH (C1, C1.w[0], ten2k64[20 - q]); |
| 176 |
} else if (q == 20) { |
| 177 |
; // C1 * 10^0 = C1 |
| 178 |
} else { // if 21 <= q <= 34 |
| 179 |
__mul_128x64_to_128 (C, ten2k64[q - 20], C); // max 47-bit x 67-bit |
| 180 |
} |
| 181 |
if (C1.w[1] > C.w[1] || (C1.w[1] == C.w[1] && C1.w[0] >= C.w[0])) { |
| 182 |
// set invalid flag |
| 183 |
*pfpsf |= INVALID_EXCEPTION; |
| 184 |
// return Integer Indefinite |
| 185 |
res = 0x8000000000000000ull; |
| 186 |
BID_RETURN (res); |
| 187 |
} |
| 188 |
// else cases that can be rounded to a 64-bit int fall through |
| 189 |
// to '1 <= q + exp <= 19' |
| 190 |
} |
| 191 |
} |
| 192 |
// n is not too large to be converted to int64: -2^63-1/2 <= n < 2^63-1/2 |
| 193 |
// Note: some of the cases tested for above fall through to this point |
| 194 |
// Restore C1 which may have been modified above |
| 195 |
C1.w[1] = x.w[1] & MASK_COEFF; |
| 196 |
C1.w[0] = x.w[0]; |
| 197 |
if ((q + exp) < 0) { // n = +/-0.0...c(0)c(1)...c(q-1) |
| 198 |
// return 0 |
| 199 |
res = 0x0000000000000000ull; |
| 200 |
BID_RETURN (res); |
| 201 |
} else if ((q + exp) == 0) { // n = +/-0.c(0)c(1)...c(q-1) |
| 202 |
// if 0.c(0)c(1)...c(q-1) <= 0.5 <=> c(0)c(1)...c(q-1) <= 5 * 10^(q-1) |
| 203 |
// res = 0 |
| 204 |
// else |
| 205 |
// res = +/-1 |
| 206 |
ind = q - 1; |
| 207 |
if (ind <= 18) { // 0 <= ind <= 18 |
| 208 |
if ((C1.w[1] == 0) && (C1.w[0] <= midpoint64[ind])) { |
| 209 |
res = 0x0000000000000000ull; // return 0 |
| 210 |
} else if (x_sign) { // n < 0 |
| 211 |
res = 0xffffffffffffffffull; // return -1 |
| 212 |
} else { // n > 0 |
| 213 |
res = 0x0000000000000001ull; // return +1 |
| 214 |
} |
| 215 |
} else { // 19 <= ind <= 33 |
| 216 |
if ((C1.w[1] < midpoint128[ind - 19].w[1]) |
| 217 |
|| ((C1.w[1] == midpoint128[ind - 19].w[1]) |
| 218 |
&& (C1.w[0] <= midpoint128[ind - 19].w[0]))) { |
| 219 |
res = 0x0000000000000000ull; // return 0 |
| 220 |
} else if (x_sign) { // n < 0 |
| 221 |
res = 0xffffffffffffffffull; // return -1 |
| 222 |
} else { // n > 0 |
| 223 |
res = 0x0000000000000001ull; // return +1 |
| 224 |
} |
| 225 |
} |
| 226 |
} else { // if (1 <= q + exp <= 19, 1 <= q <= 34, -33 <= exp <= 18) |
| 227 |
// -2^63-1/2 <= x <= -1 or 1 <= x < 2^63-1/2 so x can be rounded |
| 228 |
// to nearest to a 64-bit signed integer |
| 229 |
if (exp < 0) { // 2 <= q <= 34, -33 <= exp <= -1, 1 <= q + exp <= 19 |
| 230 |
ind = -exp; // 1 <= ind <= 33; ind is a synonym for 'x' |
| 231 |
// chop off ind digits from the lower part of C1 |
| 232 |
// C1 = C1 + 1/2 * 10^ind where the result C1 fits in 127 bits |
| 233 |
tmp64 = C1.w[0]; |
| 234 |
if (ind <= 19) { |
| 235 |
C1.w[0] = C1.w[0] + midpoint64[ind - 1]; |
| 236 |
} else { |
| 237 |
C1.w[0] = C1.w[0] + midpoint128[ind - 20].w[0]; |
| 238 |
C1.w[1] = C1.w[1] + midpoint128[ind - 20].w[1]; |
| 239 |
} |
| 240 |
if (C1.w[0] < tmp64) |
| 241 |
C1.w[1]++; |
| 242 |
// calculate C* and f* |
| 243 |
// C* is actually floor(C*) in this case |
| 244 |
// C* and f* need shifting and masking, as shown by |
| 245 |
// shiftright128[] and maskhigh128[] |
| 246 |
// 1 <= x <= 33 |
| 247 |
// kx = 10^(-x) = ten2mk128[ind - 1] |
| 248 |
// C* = (C1 + 1/2 * 10^x) * 10^(-x) |
| 249 |
// the approximation of 10^(-x) was rounded up to 118 bits |
| 250 |
__mul_128x128_to_256 (P256, C1, ten2mk128[ind - 1]); |
| 251 |
if (ind - 1 <= 21) { // 0 <= ind - 1 <= 21 |
| 252 |
Cstar.w[1] = P256.w[3]; |
| 253 |
Cstar.w[0] = P256.w[2]; |
| 254 |
fstar.w[3] = 0; |
| 255 |
fstar.w[2] = P256.w[2] & maskhigh128[ind - 1]; |
| 256 |
fstar.w[1] = P256.w[1]; |
| 257 |
fstar.w[0] = P256.w[0]; |
| 258 |
} else { // 22 <= ind - 1 <= 33 |
| 259 |
Cstar.w[1] = 0; |
| 260 |
Cstar.w[0] = P256.w[3]; |
| 261 |
fstar.w[3] = P256.w[3] & maskhigh128[ind - 1]; |
| 262 |
fstar.w[2] = P256.w[2]; |
| 263 |
fstar.w[1] = P256.w[1]; |
| 264 |
fstar.w[0] = P256.w[0]; |
| 265 |
} |
| 266 |
// the top Ex bits of 10^(-x) are T* = ten2mk128trunc[ind], e.g. |
| 267 |
// if x=1, T*=ten2mk128trunc[0]=0x19999999999999999999999999999999 |
| 268 |
// if (0 < f* < 10^(-x)) then the result is a midpoint |
| 269 |
// if floor(C*) is even then C* = floor(C*) - logical right |
| 270 |
// shift; C* has p decimal digits, correct by Prop. 1) |
| 271 |
// else if floor(C*) is odd C* = floor(C*)-1 (logical right |
| 272 |
// shift; C* has p decimal digits, correct by Pr. 1) |
| 273 |
// else |
| 274 |
// C* = floor(C*) (logical right shift; C has p decimal digits, |
| 275 |
// correct by Property 1) |
| 276 |
// n = C* * 10^(e+x) |
| 277 |
|
| 278 |
// shift right C* by Ex-128 = shiftright128[ind] |
| 279 |
shift = shiftright128[ind - 1]; // 0 <= shift <= 102 |
| 280 |
if (ind - 1 <= 21) { // 0 <= ind - 1 <= 21 |
| 281 |
Cstar.w[0] = |
| 282 |
(Cstar.w[0] >> shift) | (Cstar.w[1] << (64 - shift)); |
| 283 |
// redundant, it will be 0! Cstar.w[1] = (Cstar.w[1] >> shift); |
| 284 |
} else { // 22 <= ind - 1 <= 33 |
| 285 |
Cstar.w[0] = (Cstar.w[0] >> (shift - 64)); // 2 <= shift - 64 <= 38 |
| 286 |
} |
| 287 |
// if the result was a midpoint it was rounded away from zero, so |
| 288 |
// it will need a correction |
| 289 |
// check for midpoints |
| 290 |
if ((fstar.w[3] == 0) && (fstar.w[2] == 0) && |
| 291 |
(fstar.w[1] || fstar.w[0]) && |
| 292 |
(fstar.w[1] < ten2mk128trunc[ind - 1].w[1] || |
| 293 |
(fstar.w[1] == ten2mk128trunc[ind - 1].w[1] && |
| 294 |
fstar.w[0] <= ten2mk128trunc[ind - 1].w[0]))) { |
| 295 |
// the result is a midpoint; round to nearest |
| 296 |
if (Cstar.w[0] & 0x01) { // Cstar.w[0] is odd; MP in [EVEN, ODD] |
| 297 |
// if floor(C*) is odd C = floor(C*) - 1; the result >= 1 |
| 298 |
Cstar.w[0]--; // Cstar.w[0] is now even |
| 299 |
} // else MP in [ODD, EVEN] |
| 300 |
} |
| 301 |
if (x_sign) |
| 302 |
res = -Cstar.w[0]; |
| 303 |
else |
| 304 |
res = Cstar.w[0]; |
| 305 |
} else if (exp == 0) { |
| 306 |
// 1 <= q <= 19 |
| 307 |
// res = +/-C (exact) |
| 308 |
if (x_sign) |
| 309 |
res = -C1.w[0]; |
| 310 |
else |
| 311 |
res = C1.w[0]; |
| 312 |
} else { // if (exp>0) => 1 <= exp <= 18, 1 <= q < 18, 2 <= q + exp <= 19 |
| 313 |
// res = +/-C * 10^exp (exact) where this fits in 64-bit integer |
| 314 |
if (x_sign) |
| 315 |
res = -C1.w[0] * ten2k64[exp]; |
| 316 |
else |
| 317 |
res = C1.w[0] * ten2k64[exp]; |
| 318 |
} |
| 319 |
} |
| 320 |
} |
| 321 |
|
| 322 |
BID_RETURN (res); |
| 323 |
} |
| 324 |
|
| 325 |
/***************************************************************************** |
| 326 |
* BID128_to_int64_xrnint |
| 327 |
****************************************************************************/ |
| 328 |
|
| 329 |
BID128_FUNCTION_ARG1_NORND_CUSTOMRESTYPE (SINT64, |
| 330 |
bid128_to_int64_xrnint, x) |
| 331 |
|
| 332 |
SINT64 res; |
| 333 |
UINT64 x_sign; |
| 334 |
UINT64 x_exp; |
| 335 |
int exp; // unbiased exponent |
| 336 |
// Note: C1.w[1], C1.w[0] represent x_signif_hi, x_signif_lo (all are UINT64) |
| 337 |
UINT64 tmp64, tmp64A; |
| 338 |
BID_UI64DOUBLE tmp1; |
| 339 |
unsigned int x_nr_bits; |
| 340 |
int q, ind, shift; |
| 341 |
UINT128 C1, C; |
| 342 |
UINT128 Cstar; // C* represents up to 34 decimal digits ~ 113 bits |
| 343 |
UINT256 fstar; |
| 344 |
UINT256 P256; |
| 345 |
|
| 346 |
// unpack x |
| 347 |
x_sign = x.w[1] & MASK_SIGN; // 0 for positive, MASK_SIGN for negative |
| 348 |
x_exp = x.w[1] & MASK_EXP; // biased and shifted left 49 bit positions |
| 349 |
C1.w[1] = x.w[1] & MASK_COEFF; |
| 350 |
C1.w[0] = x.w[0]; |
| 351 |
|
| 352 |
// check for NaN or Infinity |
| 353 |
if ((x.w[1] & MASK_SPECIAL) == MASK_SPECIAL) { |
| 354 |
// x is special |
| 355 |
if ((x.w[1] & MASK_NAN) == MASK_NAN) { // x is NAN |
| 356 |
if ((x.w[1] & MASK_SNAN) == MASK_SNAN) { // x is SNAN |
| 357 |
// set invalid flag |
| 358 |
*pfpsf |= INVALID_EXCEPTION; |
| 359 |
// return Integer Indefinite |
| 360 |
res = 0x8000000000000000ull; |
| 361 |
} else { // x is QNaN |
| 362 |
// set invalid flag |
| 363 |
*pfpsf |= INVALID_EXCEPTION; |
| 364 |
// return Integer Indefinite |
| 365 |
res = 0x8000000000000000ull; |
| 366 |
} |
| 367 |
BID_RETURN (res); |
| 368 |
} else { // x is not a NaN, so it must be infinity |
| 369 |
if (!x_sign) { // x is +inf |
| 370 |
// set invalid flag |
| 371 |
*pfpsf |= INVALID_EXCEPTION; |
| 372 |
// return Integer Indefinite |
| 373 |
res = 0x8000000000000000ull; |
| 374 |
} else { // x is -inf |
| 375 |
// set invalid flag |
| 376 |
*pfpsf |= INVALID_EXCEPTION; |
| 377 |
// return Integer Indefinite |
| 378 |
res = 0x8000000000000000ull; |
| 379 |
} |
| 380 |
BID_RETURN (res); |
| 381 |
} |
| 382 |
} |
| 383 |
// check for non-canonical values (after the check for special values) |
| 384 |
if ((C1.w[1] > 0x0001ed09bead87c0ull) |
| 385 |
|| (C1.w[1] == 0x0001ed09bead87c0ull |
| 386 |
&& (C1.w[0] > 0x378d8e63ffffffffull)) |
| 387 |
|| ((x.w[1] & 0x6000000000000000ull) == 0x6000000000000000ull)) { |
| 388 |
res = 0x0000000000000000ull; |
| 389 |
BID_RETURN (res); |
| 390 |
} else if ((C1.w[1] == 0x0ull) && (C1.w[0] == 0x0ull)) { |
| 391 |
// x is 0 |
| 392 |
res = 0x0000000000000000ull; |
| 393 |
BID_RETURN (res); |
| 394 |
} else { // x is not special and is not zero |
| 395 |
|
| 396 |
// q = nr. of decimal digits in x |
| 397 |
// determine first the nr. of bits in x |
| 398 |
if (C1.w[1] == 0) { |
| 399 |
if (C1.w[0] >= 0x0020000000000000ull) { // x >= 2^53 |
| 400 |
// split the 64-bit value in two 32-bit halves to avoid rounding errors |
| 401 |
if (C1.w[0] >= 0x0000000100000000ull) { // x >= 2^32 |
| 402 |
tmp1.d = (double) (C1.w[0] >> 32); // exact conversion |
| 403 |
x_nr_bits = |
| 404 |
33 + ((((unsigned int) (tmp1.ui64 >> 52)) & 0x7ff) - 0x3ff); |
| 405 |
} else { // x < 2^32 |
| 406 |
tmp1.d = (double) (C1.w[0]); // exact conversion |
| 407 |
x_nr_bits = |
| 408 |
1 + ((((unsigned int) (tmp1.ui64 >> 52)) & 0x7ff) - 0x3ff); |
| 409 |
} |
| 410 |
} else { // if x < 2^53 |
| 411 |
tmp1.d = (double) C1.w[0]; // exact conversion |
| 412 |
x_nr_bits = |
| 413 |
1 + ((((unsigned int) (tmp1.ui64 >> 52)) & 0x7ff) - 0x3ff); |
| 414 |
} |
| 415 |
} else { // C1.w[1] != 0 => nr. bits = 64 + nr_bits (C1.w[1]) |
| 416 |
tmp1.d = (double) C1.w[1]; // exact conversion |
| 417 |
x_nr_bits = |
| 418 |
65 + ((((unsigned int) (tmp1.ui64 >> 52)) & 0x7ff) - 0x3ff); |
| 419 |
} |
| 420 |
q = nr_digits[x_nr_bits - 1].digits; |
| 421 |
if (q == 0) { |
| 422 |
q = nr_digits[x_nr_bits - 1].digits1; |
| 423 |
if (C1.w[1] > nr_digits[x_nr_bits - 1].threshold_hi |
| 424 |
|| (C1.w[1] == nr_digits[x_nr_bits - 1].threshold_hi |
| 425 |
&& C1.w[0] >= nr_digits[x_nr_bits - 1].threshold_lo)) |
| 426 |
q++; |
| 427 |
} |
| 428 |
exp = (x_exp >> 49) - 6176; |
| 429 |
if ((q + exp) > 19) { // x >= 10^19 ~= 2^63.11... (cannot fit in SINT64) |
| 430 |
// set invalid flag |
| 431 |
*pfpsf |= INVALID_EXCEPTION; |
| 432 |
// return Integer Indefinite |
| 433 |
res = 0x8000000000000000ull; |
| 434 |
BID_RETURN (res); |
| 435 |
} else if ((q + exp) == 19) { // x = c(0)c(1)...c(18).c(19)...c(q-1) |
| 436 |
// in this case 2^63.11... ~= 10^19 <= x < 10^20 ~= 2^66.43... |
| 437 |
// so x rounded to an integer may or may not fit in a signed 64-bit int |
| 438 |
// the cases that do not fit are identified here; the ones that fit |
| 439 |
// fall through and will be handled with other cases further, |
| 440 |
// under '1 <= q + exp <= 19' |
| 441 |
if (x_sign) { // if n < 0 and q + exp = 19 |
| 442 |
// if n < -2^63 - 1/2 then n is too large |
| 443 |
// too large if c(0)c(1)...c(18).c(19)...c(q-1) > 2^63+1/2 |
| 444 |
// <=> 0.c(0)c(1)...c(q-1) * 10^20 > 5*(2^64+1), 1<=q<=34 |
| 445 |
// <=> 0.c(0)c(1)...c(q-1) * 10^20 > 0x50000000000000005, 1<=q<=34 |
| 446 |
C.w[1] = 0x0000000000000005ull; |
| 447 |
C.w[0] = 0000000000000005ull; |
| 448 |
if (q <= 19) { // 1 <= q <= 19 => 1 <= 20-q <= 19 => |
| 449 |
// 10^(20-q) is 64-bit, and so is C1 |
| 450 |
__mul_64x64_to_128MACH (C1, C1.w[0], ten2k64[20 - q]); |
| 451 |
} else if (q == 20) { |
| 452 |
; // C1 * 10^0 = C1 |
| 453 |
} else { // if 21 <= q <= 34 |
| 454 |
__mul_128x64_to_128 (C, ten2k64[q - 20], C); // max 47-bit x 67-bit |
| 455 |
} |
| 456 |
if (C1.w[1] > C.w[1] || (C1.w[1] == C.w[1] && C1.w[0] > C.w[0])) { |
| 457 |
// set invalid flag |
| 458 |
*pfpsf |= INVALID_EXCEPTION; |
| 459 |
// return Integer Indefinite |
| 460 |
res = 0x8000000000000000ull; |
| 461 |
BID_RETURN (res); |
| 462 |
} |
| 463 |
// else cases that can be rounded to a 64-bit int fall through |
| 464 |
// to '1 <= q + exp <= 19' |
| 465 |
} else { // if n > 0 and q + exp = 19 |
| 466 |
// if n >= 2^63 - 1/2 then n is too large |
| 467 |
// too large if c(0)c(1)...c(18).c(19)...c(q-1) >= 2^63-1/2 |
| 468 |
// <=> if 0.c(0)c(1)...c(q-1) * 10^20 >= 5*(2^64-1), 1<=q<=34 |
| 469 |
// <=> if 0.c(0)c(1)...c(q-1) * 10^20 >= 0x4fffffffffffffffb, 1<=q<=34 |
| 470 |
C.w[1] = 0x0000000000000004ull; |
| 471 |
C.w[0] = 0xfffffffffffffffbull; |
| 472 |
if (q <= 19) { // 1 <= q <= 19 => 1 <= 20-q <= 19 => |
| 473 |
// 10^(20-q) is 64-bit, and so is C1 |
| 474 |
__mul_64x64_to_128MACH (C1, C1.w[0], ten2k64[20 - q]); |
| 475 |
} else if (q == 20) { |
| 476 |
; // C1 * 10^0 = C1 |
| 477 |
} else { // if 21 <= q <= 34 |
| 478 |
__mul_128x64_to_128 (C, ten2k64[q - 20], C); // max 47-bit x 67-bit |
| 479 |
} |
| 480 |
if (C1.w[1] > C.w[1] || (C1.w[1] == C.w[1] && C1.w[0] >= C.w[0])) { |
| 481 |
// set invalid flag |
| 482 |
*pfpsf |= INVALID_EXCEPTION; |
| 483 |
// return Integer Indefinite |
| 484 |
res = 0x8000000000000000ull; |
| 485 |
BID_RETURN (res); |
| 486 |
} |
| 487 |
// else cases that can be rounded to a 64-bit int fall through |
| 488 |
// to '1 <= q + exp <= 19' |
| 489 |
} |
| 490 |
} |
| 491 |
// n is not too large to be converted to int64: -2^63-1/2 <= n < 2^63-1/2 |
| 492 |
// Note: some of the cases tested for above fall through to this point |
| 493 |
// Restore C1 which may have been modified above |
| 494 |
C1.w[1] = x.w[1] & MASK_COEFF; |
| 495 |
C1.w[0] = x.w[0]; |
| 496 |
if ((q + exp) < 0) { // n = +/-0.0...c(0)c(1)...c(q-1) |
| 497 |
// set inexact flag |
| 498 |
*pfpsf |= INEXACT_EXCEPTION; |
| 499 |
// return 0 |
| 500 |
res = 0x0000000000000000ull; |
| 501 |
BID_RETURN (res); |
| 502 |
} else if ((q + exp) == 0) { // n = +/-0.c(0)c(1)...c(q-1) |
| 503 |
// if 0.c(0)c(1)...c(q-1) <= 0.5 <=> c(0)c(1)...c(q-1) <= 5 * 10^(q-1) |
| 504 |
// res = 0 |
| 505 |
// else |
| 506 |
// res = +/-1 |
| 507 |
ind = q - 1; |
| 508 |
if (ind <= 18) { // 0 <= ind <= 18 |
| 509 |
if ((C1.w[1] == 0) && (C1.w[0] <= midpoint64[ind])) { |
| 510 |
res = 0x0000000000000000ull; // return 0 |
| 511 |
} else if (x_sign) { // n < 0 |
| 512 |
res = 0xffffffffffffffffull; // return -1 |
| 513 |
} else { // n > 0 |
| 514 |
res = 0x0000000000000001ull; // return +1 |
| 515 |
} |
| 516 |
} else { // 19 <= ind <= 33 |
| 517 |
if ((C1.w[1] < midpoint128[ind - 19].w[1]) |
| 518 |
|| ((C1.w[1] == midpoint128[ind - 19].w[1]) |
| 519 |
&& (C1.w[0] <= midpoint128[ind - 19].w[0]))) { |
| 520 |
res = 0x0000000000000000ull; // return 0 |
| 521 |
} else if (x_sign) { // n < 0 |
| 522 |
res = 0xffffffffffffffffull; // return -1 |
| 523 |
} else { // n > 0 |
| 524 |
res = 0x0000000000000001ull; // return +1 |
| 525 |
} |
| 526 |
} |
| 527 |
// set inexact flag |
| 528 |
*pfpsf |= INEXACT_EXCEPTION; |
| 529 |
} else { // if (1 <= q + exp <= 19, 1 <= q <= 34, -33 <= exp <= 18) |
| 530 |
// -2^63-1/2 <= x <= -1 or 1 <= x < 2^63-1/2 so x can be rounded |
| 531 |
// to nearest to a 64-bit signed integer |
| 532 |
if (exp < 0) { // 2 <= q <= 34, -33 <= exp <= -1, 1 <= q + exp <= 19 |
| 533 |
ind = -exp; // 1 <= ind <= 33; ind is a synonym for 'x' |
| 534 |
// chop off ind digits from the lower part of C1 |
| 535 |
// C1 = C1 + 1/2 * 10^ind where the result C1 fits in 127 bits |
| 536 |
tmp64 = C1.w[0]; |
| 537 |
if (ind <= 19) { |
| 538 |
C1.w[0] = C1.w[0] + midpoint64[ind - 1]; |
| 539 |
} else { |
| 540 |
C1.w[0] = C1.w[0] + midpoint128[ind - 20].w[0]; |
| 541 |
C1.w[1] = C1.w[1] + midpoint128[ind - 20].w[1]; |
| 542 |
} |
| 543 |
if (C1.w[0] < tmp64) |
| 544 |
C1.w[1]++; |
| 545 |
// calculate C* and f* |
| 546 |
// C* is actually floor(C*) in this case |
| 547 |
// C* and f* need shifting and masking, as shown by |
| 548 |
// shiftright128[] and maskhigh128[] |
| 549 |
// 1 <= x <= 33 |
| 550 |
// kx = 10^(-x) = ten2mk128[ind - 1] |
| 551 |
// C* = (C1 + 1/2 * 10^x) * 10^(-x) |
| 552 |
// the approximation of 10^(-x) was rounded up to 118 bits |
| 553 |
__mul_128x128_to_256 (P256, C1, ten2mk128[ind - 1]); |
| 554 |
if (ind - 1 <= 21) { // 0 <= ind - 1 <= 21 |
| 555 |
Cstar.w[1] = P256.w[3]; |
| 556 |
Cstar.w[0] = P256.w[2]; |
| 557 |
fstar.w[3] = 0; |
| 558 |
fstar.w[2] = P256.w[2] & maskhigh128[ind - 1]; |
| 559 |
fstar.w[1] = P256.w[1]; |
| 560 |
fstar.w[0] = P256.w[0]; |
| 561 |
} else { // 22 <= ind - 1 <= 33 |
| 562 |
Cstar.w[1] = 0; |
| 563 |
Cstar.w[0] = P256.w[3]; |
| 564 |
fstar.w[3] = P256.w[3] & maskhigh128[ind - 1]; |
| 565 |
fstar.w[2] = P256.w[2]; |
| 566 |
fstar.w[1] = P256.w[1]; |
| 567 |
fstar.w[0] = P256.w[0]; |
| 568 |
} |
| 569 |
// the top Ex bits of 10^(-x) are T* = ten2mk128trunc[ind], e.g. |
| 570 |
// if x=1, T*=ten2mk128trunc[0]=0x19999999999999999999999999999999 |
| 571 |
// if (0 < f* < 10^(-x)) then the result is a midpoint |
| 572 |
// if floor(C*) is even then C* = floor(C*) - logical right |
| 573 |
// shift; C* has p decimal digits, correct by Prop. 1) |
| 574 |
// else if floor(C*) is odd C* = floor(C*)-1 (logical right |
| 575 |
// shift; C* has p decimal digits, correct by Pr. 1) |
| 576 |
// else |
| 577 |
// C* = floor(C*) (logical right shift; C has p decimal digits, |
| 578 |
// correct by Property 1) |
| 579 |
// n = C* * 10^(e+x) |
| 580 |
|
| 581 |
// shift right C* by Ex-128 = shiftright128[ind] |
| 582 |
shift = shiftright128[ind - 1]; // 0 <= shift <= 102 |
| 583 |
if (ind - 1 <= 21) { // 0 <= ind - 1 <= 21 |
| 584 |
Cstar.w[0] = |
| 585 |
(Cstar.w[0] >> shift) | (Cstar.w[1] << (64 - shift)); |
| 586 |
// redundant, it will be 0! Cstar.w[1] = (Cstar.w[1] >> shift); |
| 587 |
} else { // 22 <= ind - 1 <= 33 |
| 588 |
Cstar.w[0] = (Cstar.w[0] >> (shift - 64)); // 2 <= shift - 64 <= 38 |
| 589 |
} |
| 590 |
// determine inexactness of the rounding of C* |
| 591 |
// if (0 < f* - 1/2 < 10^(-x)) then |
| 592 |
// the result is exact |
| 593 |
// else // if (f* - 1/2 > T*) then |
| 594 |
// the result is inexact |
| 595 |
if (ind - 1 <= 2) { |
| 596 |
if (fstar.w[1] > 0x8000000000000000ull || |
| 597 |
(fstar.w[1] == 0x8000000000000000ull |
| 598 |
&& fstar.w[0] > 0x0ull)) { |
| 599 |
// f* > 1/2 and the result may be exact |
| 600 |
tmp64 = fstar.w[1] - 0x8000000000000000ull; // f* - 1/2 |
| 601 |
if (tmp64 > ten2mk128trunc[ind - 1].w[1] |
| 602 |
|| (tmp64 == ten2mk128trunc[ind - 1].w[1] |
| 603 |
&& fstar.w[0] >= ten2mk128trunc[ind - 1].w[0])) { |
| 604 |
// set the inexact flag |
| 605 |
*pfpsf |= INEXACT_EXCEPTION; |
| 606 |
} // else the result is exact |
| 607 |
} else { // the result is inexact; f2* <= 1/2 |
| 608 |
// set the inexact flag |
| 609 |
*pfpsf |= INEXACT_EXCEPTION; |
| 610 |
} |
| 611 |
} else if (ind - 1 <= 21) { // if 3 <= ind <= 21 |
| 612 |
if (fstar.w[3] > 0x0 || |
| 613 |
(fstar.w[3] == 0x0 && fstar.w[2] > onehalf128[ind - 1]) || |
| 614 |
(fstar.w[3] == 0x0 && fstar.w[2] == onehalf128[ind - 1] && |
| 615 |
(fstar.w[1] || fstar.w[0]))) { |
| 616 |
// f2* > 1/2 and the result may be exact |
| 617 |
// Calculate f2* - 1/2 |
| 618 |
tmp64 = fstar.w[2] - onehalf128[ind - 1]; |
| 619 |
tmp64A = fstar.w[3]; |
| 620 |
if (tmp64 > fstar.w[2]) |
| 621 |
tmp64A--; |
| 622 |
if (tmp64A || tmp64 |
| 623 |
|| fstar.w[1] > ten2mk128trunc[ind - 1].w[1] |
| 624 |
|| (fstar.w[1] == ten2mk128trunc[ind - 1].w[1] |
| 625 |
&& fstar.w[0] > ten2mk128trunc[ind - 1].w[0])) { |
| 626 |
// set the inexact flag |
| 627 |
*pfpsf |= INEXACT_EXCEPTION; |
| 628 |
} // else the result is exact |
| 629 |
} else { // the result is inexact; f2* <= 1/2 |
| 630 |
// set the inexact flag |
| 631 |
*pfpsf |= INEXACT_EXCEPTION; |
| 632 |
} |
| 633 |
} else { // if 22 <= ind <= 33 |
| 634 |
if (fstar.w[3] > onehalf128[ind - 1] || |
| 635 |
(fstar.w[3] == onehalf128[ind - 1] && |
| 636 |
(fstar.w[2] || fstar.w[1] || fstar.w[0]))) { |
| 637 |
// f2* > 1/2 and the result may be exact |
| 638 |
// Calculate f2* - 1/2 |
| 639 |
tmp64 = fstar.w[3] - onehalf128[ind - 1]; |
| 640 |
if (tmp64 || fstar.w[2] |
| 641 |
|| fstar.w[1] > ten2mk128trunc[ind - 1].w[1] |
| 642 |
|| (fstar.w[1] == ten2mk128trunc[ind - 1].w[1] |
| 643 |
&& fstar.w[0] > ten2mk128trunc[ind - 1].w[0])) { |
| 644 |
// set the inexact flag |
| 645 |
*pfpsf |= INEXACT_EXCEPTION; |
| 646 |
} // else the result is exact |
| 647 |
} else { // the result is inexact; f2* <= 1/2 |
| 648 |
// set the inexact flag |
| 649 |
*pfpsf |= INEXACT_EXCEPTION; |
| 650 |
} |
| 651 |
} |
| 652 |
|
| 653 |
// if the result was a midpoint it was rounded away from zero, so |
| 654 |
// it will need a correction |
| 655 |
// check for midpoints |
| 656 |
if ((fstar.w[3] == 0) && (fstar.w[2] == 0) && |
| 657 |
(fstar.w[1] || fstar.w[0]) && |
| 658 |
(fstar.w[1] < ten2mk128trunc[ind - 1].w[1] || |
| 659 |
(fstar.w[1] == ten2mk128trunc[ind - 1].w[1] && |
| 660 |
fstar.w[0] <= ten2mk128trunc[ind - 1].w[0]))) { |
| 661 |
// the result is a midpoint; round to nearest |
| 662 |
if (Cstar.w[0] & 0x01) { // Cstar.w[0] is odd; MP in [EVEN, ODD] |
| 663 |
// if floor(C*) is odd C = floor(C*) - 1; the result >= 1 |
| 664 |
Cstar.w[0]--; // Cstar.w[0] is now even |
| 665 |
} // else MP in [ODD, EVEN] |
| 666 |
} |
| 667 |
if (x_sign) |
| 668 |
res = -Cstar.w[0]; |
| 669 |
else |
| 670 |
res = Cstar.w[0]; |
| 671 |
} else if (exp == 0) { |
| 672 |
// 1 <= q <= 19 |
| 673 |
// res = +/-C (exact) |
| 674 |
if (x_sign) |
| 675 |
res = -C1.w[0]; |
| 676 |
else |
| 677 |
res = C1.w[0]; |
| 678 |
} else { // if (exp>0) => 1 <= exp <= 18, 1 <= q < 18, 2 <= q + exp <= 19 |
| 679 |
// res = +/-C * 10^exp (exact) where this fits in 64-bit integer |
| 680 |
if (x_sign) |
| 681 |
res = -C1.w[0] * ten2k64[exp]; |
| 682 |
else |
| 683 |
res = C1.w[0] * ten2k64[exp]; |
| 684 |
} |
| 685 |
} |
| 686 |
} |
| 687 |
|
| 688 |
BID_RETURN (res); |
| 689 |
} |
| 690 |
|
| 691 |
/***************************************************************************** |
| 692 |
* BID128_to_int64_floor |
| 693 |
****************************************************************************/ |
| 694 |
|
| 695 |
BID128_FUNCTION_ARG1_NORND_CUSTOMRESTYPE (SINT64, bid128_to_int64_floor, |
| 696 |
x) |
| 697 |
|
| 698 |
SINT64 res; |
| 699 |
UINT64 x_sign; |
| 700 |
UINT64 x_exp; |
| 701 |
int exp; // unbiased exponent |
| 702 |
// Note: C1.w[1], C1.w[0] represent x_signif_hi, x_signif_lo (all are UINT64) |
| 703 |
BID_UI64DOUBLE tmp1; |
| 704 |
unsigned int x_nr_bits; |
| 705 |
int q, ind, shift; |
| 706 |
UINT128 C1, C; |
| 707 |
UINT128 Cstar; // C* represents up to 34 decimal digits ~ 113 bits |
| 708 |
UINT256 fstar; |
| 709 |
UINT256 P256; |
| 710 |
|
| 711 |
// unpack x |
| 712 |
x_sign = x.w[1] & MASK_SIGN; // 0 for positive, MASK_SIGN for negative |
| 713 |
x_exp = x.w[1] & MASK_EXP; // biased and shifted left 49 bit positions |
| 714 |
C1.w[1] = x.w[1] & MASK_COEFF; |
| 715 |
C1.w[0] = x.w[0]; |
| 716 |
|
| 717 |
// check for NaN or Infinity |
| 718 |
if ((x.w[1] & MASK_SPECIAL) == MASK_SPECIAL) { |
| 719 |
// x is special |
| 720 |
if ((x.w[1] & MASK_NAN) == MASK_NAN) { // x is NAN |
| 721 |
if ((x.w[1] & MASK_SNAN) == MASK_SNAN) { // x is SNAN |
| 722 |
// set invalid flag |
| 723 |
*pfpsf |= INVALID_EXCEPTION; |
| 724 |
// return Integer Indefinite |
| 725 |
res = 0x8000000000000000ull; |
| 726 |
} else { // x is QNaN |
| 727 |
// set invalid flag |
| 728 |
*pfpsf |= INVALID_EXCEPTION; |
| 729 |
// return Integer Indefinite |
| 730 |
res = 0x8000000000000000ull; |
| 731 |
} |
| 732 |
BID_RETURN (res); |
| 733 |
} else { // x is not a NaN, so it must be infinity |
| 734 |
if (!x_sign) { // x is +inf |
| 735 |
// set invalid flag |
| 736 |
*pfpsf |= INVALID_EXCEPTION; |
| 737 |
// return Integer Indefinite |
| 738 |
res = 0x8000000000000000ull; |
| 739 |
} else { // x is -inf |
| 740 |
// set invalid flag |
| 741 |
*pfpsf |= INVALID_EXCEPTION; |
| 742 |
// return Integer Indefinite |
| 743 |
res = 0x8000000000000000ull; |
| 744 |
} |
| 745 |
BID_RETURN (res); |
| 746 |
} |
| 747 |
} |
| 748 |
// check for non-canonical values (after the check for special values) |
| 749 |
if ((C1.w[1] > 0x0001ed09bead87c0ull) |
| 750 |
|| (C1.w[1] == 0x0001ed09bead87c0ull |
| 751 |
&& (C1.w[0] > 0x378d8e63ffffffffull)) |
| 752 |
|| ((x.w[1] & 0x6000000000000000ull) == 0x6000000000000000ull)) { |
| 753 |
res = 0x0000000000000000ull; |
| 754 |
BID_RETURN (res); |
| 755 |
} else if ((C1.w[1] == 0x0ull) && (C1.w[0] == 0x0ull)) { |
| 756 |
// x is 0 |
| 757 |
res = 0x0000000000000000ull; |
| 758 |
BID_RETURN (res); |
| 759 |
} else { // x is not special and is not zero |
| 760 |
|
| 761 |
// q = nr. of decimal digits in x |
| 762 |
// determine first the nr. of bits in x |
| 763 |
if (C1.w[1] == 0) { |
| 764 |
if (C1.w[0] >= 0x0020000000000000ull) { // x >= 2^53 |
| 765 |
// split the 64-bit value in two 32-bit halves to avoid rounding errors |
| 766 |
if (C1.w[0] >= 0x0000000100000000ull) { // x >= 2^32 |
| 767 |
tmp1.d = (double) (C1.w[0] >> 32); // exact conversion |
| 768 |
x_nr_bits = |
| 769 |
33 + ((((unsigned int) (tmp1.ui64 >> 52)) & 0x7ff) - 0x3ff); |
| 770 |
} else { // x < 2^32 |
| 771 |
tmp1.d = (double) (C1.w[0]); // exact conversion |
| 772 |
x_nr_bits = |
| 773 |
1 + ((((unsigned int) (tmp1.ui64 >> 52)) & 0x7ff) - 0x3ff); |
| 774 |
} |
| 775 |
} else { // if x < 2^53 |
| 776 |
tmp1.d = (double) C1.w[0]; // exact conversion |
| 777 |
x_nr_bits = |
| 778 |
1 + ((((unsigned int) (tmp1.ui64 >> 52)) & 0x7ff) - 0x3ff); |
| 779 |
} |
| 780 |
} else { // C1.w[1] != 0 => nr. bits = 64 + nr_bits (C1.w[1]) |
| 781 |
tmp1.d = (double) C1.w[1]; // exact conversion |
| 782 |
x_nr_bits = |
| 783 |
65 + ((((unsigned int) (tmp1.ui64 >> 52)) & 0x7ff) - 0x3ff); |
| 784 |
} |
| 785 |
q = nr_digits[x_nr_bits - 1].digits; |
| 786 |
if (q == 0) { |
| 787 |
q = nr_digits[x_nr_bits - 1].digits1; |
| 788 |
if (C1.w[1] > nr_digits[x_nr_bits - 1].threshold_hi |
| 789 |
|| (C1.w[1] == nr_digits[x_nr_bits - 1].threshold_hi |
| 790 |
&& C1.w[0] >= nr_digits[x_nr_bits - 1].threshold_lo)) |
| 791 |
q++; |
| 792 |
} |
| 793 |
exp = (x_exp >> 49) - 6176; |
| 794 |
|
| 795 |
if ((q + exp) > 19) { // x >= 10^19 ~= 2^63.11... (cannot fit in SINT64) |
| 796 |
// set invalid flag |
| 797 |
*pfpsf |= INVALID_EXCEPTION; |
| 798 |
// return Integer Indefinite |
| 799 |
res = 0x8000000000000000ull; |
| 800 |
BID_RETURN (res); |
| 801 |
} else if ((q + exp) == 19) { // x = c(0)c(1)...c(18).c(19)...c(q-1) |
| 802 |
// in this case 2^63.11... ~= 10^19 <= x < 10^20 ~= 2^66.43... |
| 803 |
// so x rounded to an integer may or may not fit in a signed 64-bit int |
| 804 |
// the cases that do not fit are identified here; the ones that fit |
| 805 |
// fall through and will be handled with other cases further, |
| 806 |
// under '1 <= q + exp <= 19' |
| 807 |
if (x_sign) { // if n < 0 and q + exp = 19 |
| 808 |
// if n < -2^63 then n is too large |
| 809 |
// too large if c(0)c(1)...c(18).c(19)...c(q-1) > 2^63 |
| 810 |
// <=> 0.c(0)c(1)...c(q-1) * 10^20 > 10*2^63, 1<=q<=34 |
| 811 |
// <=> 0.c(0)c(1)...c(q-1) * 10^20 > 0x50000000000000000, 1<=q<=34 |
| 812 |
C.w[1] = 0x0000000000000005ull; |
| 813 |
C.w[0] = 0x0000000000000000ull; |
| 814 |
if (q <= 19) { // 1 <= q <= 19 => 1 <= 20-q <= 19 => |
| 815 |
// 10^(20-q) is 64-bit, and so is C1 |
| 816 |
__mul_64x64_to_128MACH (C1, C1.w[0], ten2k64[20 - q]); |
| 817 |
} else if (q == 20) { |
| 818 |
; // C1 * 10^0 = C1 |
| 819 |
} else { // if 21 <= q <= 34 |
| 820 |
__mul_128x64_to_128 (C, ten2k64[q - 20], C); // max 47-bit x 67-bit |
| 821 |
} |
| 822 |
if (C1.w[1] > C.w[1] || (C1.w[1] == C.w[1] && C1.w[0] > C.w[0])) { |
| 823 |
// set invalid flag |
| 824 |
*pfpsf |= INVALID_EXCEPTION; |
| 825 |
// return Integer Indefinite |
| 826 |
res = 0x8000000000000000ull; |
| 827 |
BID_RETURN (res); |
| 828 |
} |
| 829 |
// else cases that can be rounded to a 64-bit int fall through |
| 830 |
// to '1 <= q + exp <= 19' |
| 831 |
} else { // if n > 0 and q + exp = 19 |
| 832 |
// if n >= 2^63 then n is too large |
| 833 |
// too large if c(0)c(1)...c(18).c(19)...c(q-1) >= 2^63 |
| 834 |
// <=> if 0.c(0)c(1)...c(q-1) * 10^20 >= 5*2^64, 1<=q<=34 |
| 835 |
// <=> if 0.c(0)c(1)...c(q-1) * 10^20 >= 0x50000000000000000, 1<=q<=34 |
| 836 |
C.w[1] = 0x0000000000000005ull; |
| 837 |
C.w[0] = 0x0000000000000000ull; |
| 838 |
if (q <= 19) { // 1 <= q <= 19 => 1 <= 20-q <= 19 => |
| 839 |
// 10^(20-q) is 64-bit, and so is C1 |
| 840 |
__mul_64x64_to_128MACH (C1, C1.w[0], ten2k64[20 - q]); |
| 841 |
} else if (q == 20) { |
| 842 |
; // C1 * 10^0 = C1 |
| 843 |
} else { // if 21 <= q <= 34 |
| 844 |
__mul_128x64_to_128 (C, ten2k64[q - 20], C); // max 47-bit x 67-bit |
| 845 |
} |
| 846 |
if (C1.w[1] > C.w[1] || (C1.w[1] == C.w[1] && C1.w[0] >= C.w[0])) { |
| 847 |
// set invalid flag |
| 848 |
*pfpsf |= INVALID_EXCEPTION; |
| 849 |
// return Integer Indefinite |
| 850 |
res = 0x8000000000000000ull; |
| 851 |
BID_RETURN (res); |
| 852 |
} |
| 853 |
// else cases that can be rounded to a 64-bit int fall through |
| 854 |
// to '1 <= q + exp <= 19' |
| 855 |
} |
| 856 |
} |
| 857 |
// n is not too large to be converted to int64: -2^63-1 < n < 2^63 |
| 858 |
// Note: some of the cases tested for above fall through to this point |
| 859 |
// Restore C1 which may have been modified above |
| 860 |
C1.w[1] = x.w[1] & MASK_COEFF; |
| 861 |
C1.w[0] = x.w[0]; |
| 862 |
if ((q + exp) <= 0) { // n = +/-0.[0...0]c(0)c(1)...c(q-1) |
| 863 |
// return -1 or 0 |
| 864 |
if (x_sign) |
| 865 |
res = 0xffffffffffffffffull; |
| 866 |
else |
| 867 |
res = 0x0000000000000000ull; |
| 868 |
BID_RETURN (res); |
| 869 |
} else { // if (1 <= q + exp <= 19, 1 <= q <= 34, -33 <= exp <= 18) |
| 870 |
// -2^63 <= x <= -1 or 1 <= x < 2^63 so x can be rounded |
| 871 |
// toward zero to a 64-bit signed integer |
| 872 |
if (exp < 0) { // 2 <= q <= 34, -33 <= exp <= -1, 1 <= q + exp <= 19 |
| 873 |
ind = -exp; // 1 <= ind <= 33; ind is a synonym for 'x' |
| 874 |
// chop off ind digits from the lower part of C1 |
| 875 |
// C1 fits in 127 bits |
| 876 |
// calculate C* and f* |
| 877 |
// C* is actually floor(C*) in this case |
| 878 |
// C* and f* need shifting and masking, as shown by |
| 879 |
// shiftright128[] and maskhigh128[] |
| 880 |
// 1 <= x <= 33 |
| 881 |
// kx = 10^(-x) = ten2mk128[ind - 1] |
| 882 |
// C* = C1 * 10^(-x) |
| 883 |
// the approximation of 10^(-x) was rounded up to 118 bits |
| 884 |
__mul_128x128_to_256 (P256, C1, ten2mk128[ind - 1]); |
| 885 |
if (ind - 1 <= 21) { // 0 <= ind - 1 <= 21 |
| 886 |
Cstar.w[1] = P256.w[3]; |
| 887 |
Cstar.w[0] = P256.w[2]; |
| 888 |
fstar.w[3] = 0; |
| 889 |
fstar.w[2] = P256.w[2] & maskhigh128[ind - 1]; |
| 890 |
fstar.w[1] = P256.w[1]; |
| 891 |
fstar.w[0] = P256.w[0]; |
| 892 |
} else { // 22 <= ind - 1 <= 33 |
| 893 |
Cstar.w[1] = 0; |
| 894 |
Cstar.w[0] = P256.w[3]; |
| 895 |
fstar.w[3] = P256.w[3] & maskhigh128[ind - 1]; |
| 896 |
fstar.w[2] = P256.w[2]; |
| 897 |
fstar.w[1] = P256.w[1]; |
| 898 |
fstar.w[0] = P256.w[0]; |
| 899 |
} |
| 900 |
// the top Ex bits of 10^(-x) are T* = ten2mk128trunc[ind], e.g. |
| 901 |
// if x=1, T*=ten2mk128trunc[0]=0x19999999999999999999999999999999 |
| 902 |
// C* = floor(C*) (logical right shift; C has p decimal digits, |
| 903 |
// correct by Property 1) |
| 904 |
// n = C* * 10^(e+x) |
| 905 |
|
| 906 |
// shift right C* by Ex-128 = shiftright128[ind] |
| 907 |
shift = shiftright128[ind - 1]; // 0 <= shift <= 102 |
| 908 |
if (ind - 1 <= 21) { // 0 <= ind - 1 <= 21 |
| 909 |
Cstar.w[0] = |
| 910 |
(Cstar.w[0] >> shift) | (Cstar.w[1] << (64 - shift)); |
| 911 |
// redundant, it will be 0! Cstar.w[1] = (Cstar.w[1] >> shift); |
| 912 |
} else { // 22 <= ind - 1 <= 33 |
| 913 |
Cstar.w[0] = (Cstar.w[0] >> (shift - 64)); // 2 <= shift - 64 <= 38 |
| 914 |
} |
| 915 |
// if the result is negative and inexact, need to add 1 to it |
| 916 |
|
| 917 |
// determine inexactness of the rounding of C* |
| 918 |
// if (0 < f* < 10^(-x)) then |
| 919 |
// the result is exact |
| 920 |
// else // if (f* > T*) then |
| 921 |
// the result is inexact |
| 922 |
if (ind - 1 <= 2) { |
| 923 |
if (fstar.w[1] > ten2mk128trunc[ind - 1].w[1] || |
| 924 |
(fstar.w[1] == ten2mk128trunc[ind - 1].w[1] && |
| 925 |
fstar.w[0] > ten2mk128trunc[ind - 1].w[0])) { |
| 926 |
if (x_sign) { // positive and inexact |
| 927 |
Cstar.w[0]++; |
| 928 |
if (Cstar.w[0] == 0x0) |
| 929 |
Cstar.w[1]++; |
| 930 |
} |
| 931 |
} // else the result is exact |
| 932 |
} else if (ind - 1 <= 21) { // if 3 <= ind <= 21 |
| 933 |
if (fstar.w[2] || fstar.w[1] > ten2mk128trunc[ind - 1].w[1] |
| 934 |
|| (fstar.w[1] == ten2mk128trunc[ind - 1].w[1] |
| 935 |
&& fstar.w[0] > ten2mk128trunc[ind - 1].w[0])) { |
| 936 |
if (x_sign) { // positive and inexact |
| 937 |
Cstar.w[0]++; |
| 938 |
if (Cstar.w[0] == 0x0) |
| 939 |
Cstar.w[1]++; |
| 940 |
} |
| 941 |
} // else the result is exact |
| 942 |
} else { // if 22 <= ind <= 33 |
| 943 |
if (fstar.w[3] || fstar.w[2] |
| 944 |
|| fstar.w[1] > ten2mk128trunc[ind - 1].w[1] |
| 945 |
|| (fstar.w[1] == ten2mk128trunc[ind - 1].w[1] |
| 946 |
&& fstar.w[0] > ten2mk128trunc[ind - 1].w[0])) { |
| 947 |
if (x_sign) { // positive and inexact |
| 948 |
Cstar.w[0]++; |
| 949 |
if (Cstar.w[0] == 0x0) |
| 950 |
Cstar.w[1]++; |
| 951 |
} |
| 952 |
} // else the result is exact |
| 953 |
} |
| 954 |
|
| 955 |
if (x_sign) |
| 956 |
res = -Cstar.w[0]; |
| 957 |
else |
| 958 |
res = Cstar.w[0]; |
| 959 |
} else if (exp == 0) { |
| 960 |
// 1 <= q <= 19 |
| 961 |
// res = +/-C (exact) |
| 962 |
if (x_sign) |
| 963 |
res = -C1.w[0]; |
| 964 |
else |
| 965 |
res = C1.w[0]; |
| 966 |
} else { // if (exp>0) => 1 <= exp <= 18, 1 <= q < 18, 2 <= q + exp <= 19 |
| 967 |
// res = +/-C * 10^exp (exact) where this fits in 64-bit integer |
| 968 |
if (x_sign) |
| 969 |
res = -C1.w[0] * ten2k64[exp]; |
| 970 |
else |
| 971 |
res = C1.w[0] * ten2k64[exp]; |
| 972 |
} |
| 973 |
} |
| 974 |
} |
| 975 |
|
| 976 |
BID_RETURN (res); |
| 977 |
} |
| 978 |
|
| 979 |
/***************************************************************************** |
| 980 |
* BID128_to_int64_xfloor |
| 981 |
****************************************************************************/ |
| 982 |
|
| 983 |
BID128_FUNCTION_ARG1_NORND_CUSTOMRESTYPE (SINT64, |
| 984 |
bid128_to_int64_xfloor, x) |
| 985 |
|
| 986 |
SINT64 res; |
| 987 |
UINT64 x_sign; |
| 988 |
UINT64 x_exp; |
| 989 |
int exp; // unbiased exponent |
| 990 |
// Note: C1.w[1], C1.w[0] represent x_signif_hi, x_signif_lo (all are UINT64) |
| 991 |
BID_UI64DOUBLE tmp1; |
| 992 |
unsigned int x_nr_bits; |
| 993 |
int q, ind, shift; |
| 994 |
UINT128 C1, C; |
| 995 |
UINT128 Cstar; // C* represents up to 34 decimal digits ~ 113 bits |
| 996 |
UINT256 fstar; |
| 997 |
UINT256 P256; |
| 998 |
|
| 999 |
// unpack x |
| 1000 |
x_sign = x.w[1] & MASK_SIGN; // 0 for positive, MASK_SIGN for negative |
| 1001 |
x_exp = x.w[1] & MASK_EXP; // biased and shifted left 49 bit positions |
| 1002 |
C1.w[1] = x.w[1] & MASK_COEFF; |
| 1003 |
C1.w[0] = x.w[0]; |
| 1004 |
|
| 1005 |
// check for NaN or Infinity |
| 1006 |
if ((x.w[1] & MASK_SPECIAL) == MASK_SPECIAL) { |
| 1007 |
// x is special |
| 1008 |
if ((x.w[1] & MASK_NAN) == MASK_NAN) { // x is NAN |
| 1009 |
if ((x.w[1] & MASK_SNAN) == MASK_SNAN) { // x is SNAN |
| 1010 |
// set invalid flag |
| 1011 |
*pfpsf |= INVALID_EXCEPTION; |
| 1012 |
// return Integer Indefinite |
| 1013 |
res = 0x8000000000000000ull; |
| 1014 |
} else { // x is QNaN |
| 1015 |
// set invalid flag |
| 1016 |
*pfpsf |= INVALID_EXCEPTION; |
| 1017 |
// return Integer Indefinite |
| 1018 |
res = 0x8000000000000000ull; |
| 1019 |
} |
| 1020 |
BID_RETURN (res); |
| 1021 |
} else { // x is not a NaN, so it must be infinity |
| 1022 |
if (!x_sign) { // x is +inf |
| 1023 |
// set invalid flag |
| 1024 |
*pfpsf |= INVALID_EXCEPTION; |
| 1025 |
// return Integer Indefinite |
| 1026 |
res = 0x8000000000000000ull; |
| 1027 |
} else { // x is -inf |
| 1028 |
// set invalid flag |
| 1029 |
*pfpsf |= INVALID_EXCEPTION; |
| 1030 |
// return Integer Indefinite |
| 1031 |
res = 0x8000000000000000ull; |
| 1032 |
} |
| 1033 |
BID_RETURN (res); |
| 1034 |
} |
| 1035 |
} |
| 1036 |
// check for non-canonical values (after the check for special values) |
| 1037 |
if ((C1.w[1] > 0x0001ed09bead87c0ull) |
| 1038 |
|| (C1.w[1] == 0x0001ed09bead87c0ull |
| 1039 |
&& (C1.w[0] > 0x378d8e63ffffffffull)) |
| 1040 |
|| ((x.w[1] & 0x6000000000000000ull) == 0x6000000000000000ull)) { |
| 1041 |
res = 0x0000000000000000ull; |
| 1042 |
BID_RETURN (res); |
| 1043 |
} else if ((C1.w[1] == 0x0ull) && (C1.w[0] == 0x0ull)) { |
| 1044 |
// x is 0 |
| 1045 |
res = 0x0000000000000000ull; |
| 1046 |
BID_RETURN (res); |
| 1047 |
} else { // x is not special and is not zero |
| 1048 |
|
| 1049 |
// q = nr. of decimal digits in x |
| 1050 |
// determine first the nr. of bits in x |
| 1051 |
if (C1.w[1] == 0) { |
| 1052 |
if (C1.w[0] >= 0x0020000000000000ull) { // x >= 2^53 |
| 1053 |
// split the 64-bit value in two 32-bit halves to avoid rounding errors |
| 1054 |
if (C1.w[0] >= 0x0000000100000000ull) { // x >= 2^32 |
| 1055 |
tmp1.d = (double) (C1.w[0] >> 32); // exact conversion |
| 1056 |
x_nr_bits = |
| 1057 |
33 + ((((unsigned int) (tmp1.ui64 >> 52)) & 0x7ff) - 0x3ff); |
| 1058 |
} else { // x < 2^32 |
| 1059 |
tmp1.d = (double) (C1.w[0]); // exact conversion |
| 1060 |
x_nr_bits = |
| 1061 |
1 + ((((unsigned int) (tmp1.ui64 >> 52)) & 0x7ff) - 0x3ff); |
| 1062 |
} |
| 1063 |
} else { // if x < 2^53 |
| 1064 |
tmp1.d = (double) C1.w[0]; // exact conversion |
| 1065 |
x_nr_bits = |
| 1066 |
1 + ((((unsigned int) (tmp1.ui64 >> 52)) & 0x7ff) - 0x3ff); |
| 1067 |
} |
| 1068 |
} else { // C1.w[1] != 0 => nr. bits = 64 + nr_bits (C1.w[1]) |
| 1069 |
tmp1.d = (double) C1.w[1]; // exact conversion |
| 1070 |
x_nr_bits = |
| 1071 |
65 + ((((unsigned int) (tmp1.ui64 >> 52)) & 0x7ff) - 0x3ff); |
| 1072 |
} |
| 1073 |
q = nr_digits[x_nr_bits - 1].digits; |
| 1074 |
if (q == 0) { |
| 1075 |
q = nr_digits[x_nr_bits - 1].digits1; |
| 1076 |
if (C1.w[1] > nr_digits[x_nr_bits - 1].threshold_hi |
| 1077 |
|| (C1.w[1] == nr_digits[x_nr_bits - 1].threshold_hi |
| 1078 |
&& C1.w[0] >= nr_digits[x_nr_bits - 1].threshold_lo)) |
| 1079 |
q++; |
| 1080 |
} |
| 1081 |
exp = (x_exp >> 49) - 6176; |
| 1082 |
if ((q + exp) > 19) { // x >= 10^19 ~= 2^63.11... (cannot fit in SINT64) |
| 1083 |
// set invalid flag |
| 1084 |
*pfpsf |= INVALID_EXCEPTION; |
| 1085 |
// return Integer Indefinite |
| 1086 |
res = 0x8000000000000000ull; |
| 1087 |
BID_RETURN (res); |
| 1088 |
} else if ((q + exp) == 19) { // x = c(0)c(1)...c(18).c(19)...c(q-1) |
| 1089 |
// in this case 2^63.11... ~= 10^19 <= x < 10^20 ~= 2^66.43... |
| 1090 |
// so x rounded to an integer may or may not fit in a signed 64-bit int |
| 1091 |
// the cases that do not fit are identified here; the ones that fit |
| 1092 |
// fall through and will be handled with other cases further, |
| 1093 |
// under '1 <= q + exp <= 19' |
| 1094 |
if (x_sign) { // if n < 0 and q + exp = 19 |
| 1095 |
// if n < -2^63 then n is too large |
| 1096 |
// too large if c(0)c(1)...c(18).c(19)...c(q-1) > 2^63 |
| 1097 |
// <=> 0.c(0)c(1)...c(q-1) * 10^20 > 10*2^63, 1<=q<=34 |
| 1098 |
// <=> 0.c(0)c(1)...c(q-1) * 10^20 > 0x50000000000000000, 1<=q<=34 |
| 1099 |
C.w[1] = 0x0000000000000005ull; |
| 1100 |
C.w[0] = 0x0000000000000000ull; |
| 1101 |
if (q <= 19) { // 1 <= q <= 19 => 1 <= 20-q <= 19 => |
| 1102 |
// 10^(20-q) is 64-bit, and so is C1 |
| 1103 |
__mul_64x64_to_128MACH (C1, C1.w[0], ten2k64[20 - q]); |
| 1104 |
} else if (q == 20) { |
| 1105 |
; // C1 * 10^0 = C1 |
| 1106 |
} else { // if 21 <= q <= 34 |
| 1107 |
__mul_128x64_to_128 (C, ten2k64[q - 20], C); // max 47-bit x 67-bit |
| 1108 |
} |
| 1109 |
if (C1.w[1] > C.w[1] || (C1.w[1] == C.w[1] && C1.w[0] > C.w[0])) { |
| 1110 |
// set invalid flag |
| 1111 |
*pfpsf |= INVALID_EXCEPTION; |
| 1112 |
// return Integer Indefinite |
| 1113 |
res = 0x8000000000000000ull; |
| 1114 |
BID_RETURN (res); |
| 1115 |
} |
| 1116 |
// else cases that can be rounded to a 64-bit int fall through |
| 1117 |
// to '1 <= q + exp <= 19' |
| 1118 |
} else { // if n > 0 and q + exp = 19 |
| 1119 |
// if n >= 2^63 then n is too large |
| 1120 |
// too large if c(0)c(1)...c(18).c(19)...c(q-1) >= 2^63 |
| 1121 |
// <=> if 0.c(0)c(1)...c(q-1) * 10^20 >= 5*2^64, 1<=q<=34 |
| 1122 |
// <=> if 0.c(0)c(1)...c(q-1) * 10^20 >= 0x50000000000000000, 1<=q<=34 |
| 1123 |
C.w[1] = 0x0000000000000005ull; |
| 1124 |
C.w[0] = 0x0000000000000000ull; |
| 1125 |
if (q <= 19) { // 1 <= q <= 19 => 1 <= 20-q <= 19 => |
| 1126 |
// 10^(20-q) is 64-bit, and so is C1 |
| 1127 |
__mul_64x64_to_128MACH (C1, C1.w[0], ten2k64[20 - q]); |
| 1128 |
} else if (q == 20) { |
| 1129 |
; // C1 * 10^0 = C1 |
| 1130 |
} else { // if 21 <= q <= 34 |
| 1131 |
__mul_128x64_to_128 (C, ten2k64[q - 20], C); // max 47-bit x 67-bit |
| 1132 |
} |
| 1133 |
if (C1.w[1] > C.w[1] || (C1.w[1] == C.w[1] && C1.w[0] >= C.w[0])) { |
| 1134 |
// set invalid flag |
| 1135 |
*pfpsf |= INVALID_EXCEPTION; |
| 1136 |
// return Integer Indefinite |
| 1137 |
res = 0x8000000000000000ull; |
| 1138 |
BID_RETURN (res); |
| 1139 |
} |
| 1140 |
// else cases that can be rounded to a 64-bit int fall through |
| 1141 |
// to '1 <= q + exp <= 19' |
| 1142 |
} |
| 1143 |
} |
| 1144 |
// n is not too large to be converted to int64: -2^63-1 < n < 2^63 |
| 1145 |
// Note: some of the cases tested for above fall through to this point |
| 1146 |
// Restore C1 which may have been modified above |
| 1147 |
C1.w[1] = x.w[1] & MASK_COEFF; |
| 1148 |
C1.w[0] = x.w[0]; |
| 1149 |
if ((q + exp) <= 0) { // n = +/-0.[0...0]c(0)c(1)...c(q-1) |
| 1150 |
// set inexact flag |
| 1151 |
*pfpsf |= INEXACT_EXCEPTION; |
| 1152 |
// return -1 or 0 |
| 1153 |
if (x_sign) |
| 1154 |
res = 0xffffffffffffffffull; |
| 1155 |
else |
| 1156 |
res = 0x0000000000000000ull; |
| 1157 |
BID_RETURN (res); |
| 1158 |
} else { // if (1 <= q + exp <= 19, 1 <= q <= 34, -33 <= exp <= 18) |
| 1159 |
// -2^63 <= x <= -1 or 1 <= x < 2^63 so x can be rounded |
| 1160 |
// toward zero to a 64-bit signed integer |
| 1161 |
if (exp < 0) { // 2 <= q <= 34, -33 <= exp <= -1, 1 <= q + exp <= 19 |
| 1162 |
ind = -exp; // 1 <= ind <= 33; ind is a synonym for 'x' |
| 1163 |
// chop off ind digits from the lower part of C1 |
| 1164 |
// C1 fits in 127 bits |
| 1165 |
// calculate C* and f* |
| 1166 |
// C* is actually floor(C*) in this case |
| 1167 |
// C* and f* need shifting and masking, as shown by |
| 1168 |
// shiftright128[] and maskhigh128[] |
| 1169 |
// 1 <= x <= 33 |
| 1170 |
// kx = 10^(-x) = ten2mk128[ind - 1] |
| 1171 |
// C* = C1 * 10^(-x) |
| 1172 |
// the approximation of 10^(-x) was rounded up to 118 bits |
| 1173 |
__mul_128x128_to_256 (P256, C1, ten2mk128[ind - 1]); |
| 1174 |
if (ind - 1 <= 21) { // 0 <= ind - 1 <= 21 |
| 1175 |
Cstar.w[1] = P256.w[3]; |
| 1176 |
Cstar.w[0] = P256.w[2]; |
| 1177 |
fstar.w[3] = 0; |
| 1178 |
fstar.w[2] = P256.w[2] & maskhigh128[ind - 1]; |
| 1179 |
fstar.w[1] = P256.w[1]; |
| 1180 |
fstar.w[0] = P256.w[0]; |
| 1181 |
} else { // 22 <= ind - 1 <= 33 |
| 1182 |
Cstar.w[1] = 0; |
| 1183 |
Cstar.w[0] = P256.w[3]; |
| 1184 |
fstar.w[3] = P256.w[3] & maskhigh128[ind - 1]; |
| 1185 |
fstar.w[2] = P256.w[2]; |
| 1186 |
fstar.w[1] = P256.w[1]; |
| 1187 |
fstar.w[0] = P256.w[0]; |
| 1188 |
} |
| 1189 |
// the top Ex bits of 10^(-x) are T* = ten2mk128trunc[ind], e.g. |
| 1190 |
// if x=1, T*=ten2mk128trunc[0]=0x19999999999999999999999999999999 |
| 1191 |
// C* = floor(C*) (logical right shift; C has p decimal digits, |
| 1192 |
// correct by Property 1) |
| 1193 |
// n = C* * 10^(e+x) |
| 1194 |
|
| 1195 |
// shift right C* by Ex-128 = shiftright128[ind] |
| 1196 |
shift = shiftright128[ind - 1]; // 0 <= shift <= 102 |
| 1197 |
if (ind - 1 <= 21) { // 0 <= ind - 1 <= 21 |
| 1198 |
Cstar.w[0] = |
| 1199 |
(Cstar.w[0] >> shift) | (Cstar.w[1] << (64 - shift)); |
| 1200 |
// redundant, it will be 0! Cstar.w[1] = (Cstar.w[1] >> shift); |
| 1201 |
} else { // 22 <= ind - 1 <= 33 |
| 1202 |
Cstar.w[0] = (Cstar.w[0] >> (shift - 64)); // 2 <= shift - 64 <= 38 |
| 1203 |
} |
| 1204 |
// if the result is negative and inexact, need to add 1 to it |
| 1205 |
|
| 1206 |
// determine inexactness of the rounding of C* |
| 1207 |
// if (0 < f* < 10^(-x)) then |
| 1208 |
// the result is exact |
| 1209 |
// else // if (f* > T*) then |
| 1210 |
// the result is inexact |
| 1211 |
if (ind - 1 <= 2) { |
| 1212 |
if (fstar.w[1] > ten2mk128trunc[ind - 1].w[1] || |
| 1213 |
(fstar.w[1] == ten2mk128trunc[ind - 1].w[1] && |
| 1214 |
fstar.w[0] > ten2mk128trunc[ind - 1].w[0])) { |
| 1215 |
if (x_sign) { // positive and inexact |
| 1216 |
Cstar.w[0]++; |
| 1217 |
if (Cstar.w[0] == 0x0) |
| 1218 |
Cstar.w[1]++; |
| 1219 |
} |
| 1220 |
// set the inexact flag |
| 1221 |
*pfpsf |= INEXACT_EXCEPTION; |
| 1222 |
} // else the result is exact |
| 1223 |
} else if (ind - 1 <= 21) { // if 3 <= ind <= 21 |
| 1224 |
if (fstar.w[2] || fstar.w[1] > ten2mk128trunc[ind - 1].w[1] |
| 1225 |
|| (fstar.w[1] == ten2mk128trunc[ind - 1].w[1] |
| 1226 |
&& fstar.w[0] > ten2mk128trunc[ind - 1].w[0])) { |
| 1227 |
if (x_sign) { // positive and inexact |
| 1228 |
Cstar.w[0]++; |
| 1229 |
if (Cstar.w[0] == 0x0) |
| 1230 |
Cstar.w[1]++; |
| 1231 |
} |
| 1232 |
// set the inexact flag |
| 1233 |
*pfpsf |= INEXACT_EXCEPTION; |
| 1234 |
} // else the result is exact |
| 1235 |
} else { // if 22 <= ind <= 33 |
| 1236 |
if (fstar.w[3] || fstar.w[2] |
| 1237 |
|| fstar.w[1] > ten2mk128trunc[ind - 1].w[1] |
| 1238 |
|| (fstar.w[1] == ten2mk128trunc[ind - 1].w[1] |
| 1239 |
&& fstar.w[0] > ten2mk128trunc[ind - 1].w[0])) { |
| 1240 |
if (x_sign) { // positive and inexact |
| 1241 |
Cstar.w[0]++; |
| 1242 |
if (Cstar.w[0] == 0x0) |
| 1243 |
Cstar.w[1]++; |
| 1244 |
} |
| 1245 |
// set the inexact flag |
| 1246 |
*pfpsf |= INEXACT_EXCEPTION; |
| 1247 |
} // else the result is exact |
| 1248 |
} |
| 1249 |
|
| 1250 |
if (x_sign) |
| 1251 |
res = -Cstar.w[0]; |
| 1252 |
else |
| 1253 |
res = Cstar.w[0]; |
| 1254 |
} else if (exp == 0) { |
| 1255 |
// 1 <= q <= 19 |
| 1256 |
// res = +/-C (exact) |
| 1257 |
if (x_sign) |
| 1258 |
res = -C1.w[0]; |
| 1259 |
else |
| 1260 |
res = C1.w[0]; |
| 1261 |
} else { // if (exp>0) => 1 <= exp <= 18, 1 <= q < 18, 2 <= q + exp <= 19 |
| 1262 |
// res = +/-C * 10^exp (exact) where this fits in 64-bit integer |
| 1263 |
if (x_sign) |
| 1264 |
res = -C1.w[0] * ten2k64[exp]; |
| 1265 |
else |
| 1266 |
res = C1.w[0] * ten2k64[exp]; |
| 1267 |
} |
| 1268 |
} |
| 1269 |
} |
| 1270 |
|
| 1271 |
BID_RETURN (res); |
| 1272 |
} |
| 1273 |
|
| 1274 |
/***************************************************************************** |
| 1275 |
* BID128_to_int64_ceil |
| 1276 |
****************************************************************************/ |
| 1277 |
|
| 1278 |
BID128_FUNCTION_ARG1_NORND_CUSTOMRESTYPE (SINT64, bid128_to_int64_ceil, |
| 1279 |
x) |
| 1280 |
|
| 1281 |
SINT64 res; |
| 1282 |
UINT64 x_sign; |
| 1283 |
UINT64 x_exp; |
| 1284 |
int exp; // unbiased exponent |
| 1285 |
// Note: C1.w[1], C1.w[0] represent x_signif_hi, x_signif_lo (all are UINT64) |
| 1286 |
BID_UI64DOUBLE tmp1; |
| 1287 |
unsigned int x_nr_bits; |
| 1288 |
int q, ind, shift; |
| 1289 |
UINT128 C1, C; |
| 1290 |
UINT128 Cstar; // C* represents up to 34 decimal digits ~ 113 bits |
| 1291 |
UINT256 fstar; |
| 1292 |
UINT256 P256; |
| 1293 |
|
| 1294 |
// unpack x |
| 1295 |
x_sign = x.w[1] & MASK_SIGN; // 0 for positive, MASK_SIGN for negative |
| 1296 |
x_exp = x.w[1] & MASK_EXP; // biased and shifted left 49 bit positions |
| 1297 |
C1.w[1] = x.w[1] & MASK_COEFF; |
| 1298 |
C1.w[0] = x.w[0]; |
| 1299 |
|
| 1300 |
// check for NaN or Infinity |
| 1301 |
if ((x.w[1] & MASK_SPECIAL) == MASK_SPECIAL) { |
| 1302 |
// x is special |
| 1303 |
if ((x.w[1] & MASK_NAN) == MASK_NAN) { // x is NAN |
| 1304 |
if ((x.w[1] & MASK_SNAN) == MASK_SNAN) { // x is SNAN |
| 1305 |
// set invalid flag |
| 1306 |
*pfpsf |= INVALID_EXCEPTION; |
| 1307 |
// return Integer Indefinite |
| 1308 |
res = 0x8000000000000000ull; |
| 1309 |
} else { // x is QNaN |
| 1310 |
// set invalid flag |
| 1311 |
*pfpsf |= INVALID_EXCEPTION; |
| 1312 |
// return Integer Indefinite |
| 1313 |
res = 0x8000000000000000ull; |
| 1314 |
} |
| 1315 |
BID_RETURN (res); |
| 1316 |
} else { // x is not a NaN, so it must be infinity |
| 1317 |
if (!x_sign) { // x is +inf |
| 1318 |
// set invalid flag |
| 1319 |
*pfpsf |= INVALID_EXCEPTION; |
| 1320 |
// return Integer Indefinite |
| 1321 |
res = 0x8000000000000000ull; |
| 1322 |
} else { // x is -inf |
| 1323 |
// set invalid flag |
| 1324 |
*pfpsf |= INVALID_EXCEPTION; |
| 1325 |
// return Integer Indefinite |
| 1326 |
res = 0x8000000000000000ull; |
| 1327 |
} |
| 1328 |
BID_RETURN (res); |
| 1329 |
} |
| 1330 |
} |
| 1331 |
// check for non-canonical values (after the check for special values) |
| 1332 |
if ((C1.w[1] > 0x0001ed09bead87c0ull) |
| 1333 |
|| (C1.w[1] == 0x0001ed09bead87c0ull |
| 1334 |
&& (C1.w[0] > 0x378d8e63ffffffffull)) |
| 1335 |
|| ((x.w[1] & 0x6000000000000000ull) == 0x6000000000000000ull)) { |
| 1336 |
res = 0x0000000000000000ull; |
| 1337 |
BID_RETURN (res); |
| 1338 |
} else if ((C1.w[1] == 0x0ull) && (C1.w[0] == 0x0ull)) { |
| 1339 |
// x is 0 |
| 1340 |
res = 0x0000000000000000ull; |
| 1341 |
BID_RETURN (res); |
| 1342 |
} else { // x is not special and is not zero |
| 1343 |
|
| 1344 |
// q = nr. of decimal digits in x |
| 1345 |
// determine first the nr. of bits in x |
| 1346 |
if (C1.w[1] == 0) { |
| 1347 |
if (C1.w[0] >= 0x0020000000000000ull) { // x >= 2^53 |
| 1348 |
// split the 64-bit value in two 32-bit halves to avoid rounding errors |
| 1349 |
if (C1.w[0] >= 0x0000000100000000ull) { // x >= 2^32 |
| 1350 |
tmp1.d = (double) (C1.w[0] >> 32); // exact conversion |
| 1351 |
x_nr_bits = |
| 1352 |
33 + ((((unsigned int) (tmp1.ui64 >> 52)) & 0x7ff) - 0x3ff); |
| 1353 |
} else { // x < 2^32 |
| 1354 |
tmp1.d = (double) (C1.w[0]); // exact conversion |
| 1355 |
x_nr_bits = |
| 1356 |
1 + ((((unsigned int) (tmp1.ui64 >> 52)) & 0x7ff) - 0x3ff); |
| 1357 |
} |
| 1358 |
} else { // if x < 2^53 |
| 1359 |
tmp1.d = (double) C1.w[0]; // exact conversion |
| 1360 |
x_nr_bits = |
| 1361 |
1 + ((((unsigned int) (tmp1.ui64 >> 52)) & 0x7ff) - 0x3ff); |
| 1362 |
} |
| 1363 |
} else { // C1.w[1] != 0 => nr. bits = 64 + nr_bits (C1.w[1]) |
| 1364 |
tmp1.d = (double) C1.w[1]; // exact conversion |
| 1365 |
x_nr_bits = |
| 1366 |
65 + ((((unsigned int) (tmp1.ui64 >> 52)) & 0x7ff) - 0x3ff); |
| 1367 |
} |
| 1368 |
q = nr_digits[x_nr_bits - 1].digits; |
| 1369 |
if (q == 0) { |
| 1370 |
q = nr_digits[x_nr_bits - 1].digits1; |
| 1371 |
if (C1.w[1] > nr_digits[x_nr_bits - 1].threshold_hi |
| 1372 |
|| (C1.w[1] == nr_digits[x_nr_bits - 1].threshold_hi |
| 1373 |
&& C1.w[0] >= nr_digits[x_nr_bits - 1].threshold_lo)) |
| 1374 |
q++; |
| 1375 |
} |
| 1376 |
exp = (x_exp >> 49) - 6176; |
| 1377 |
if ((q + exp) > 19) { // x >= 10^19 ~= 2^63.11... (cannot fit in SINT64) |
| 1378 |
// set invalid flag |
| 1379 |
*pfpsf |= INVALID_EXCEPTION; |
| 1380 |
// return Integer Indefinite |
| 1381 |
res = 0x8000000000000000ull; |
| 1382 |
BID_RETURN (res); |
| 1383 |
} else if ((q + exp) == 19) { // x = c(0)c(1)...c(18).c(19)...c(q-1) |
| 1384 |
// in this case 2^63.11... ~= 10^19 <= x < 10^20 ~= 2^66.43... |
| 1385 |
// so x rounded to an integer may or may not fit in a signed 64-bit int |
| 1386 |
// the cases that do not fit are identified here; the ones that fit |
| 1387 |
// fall through and will be handled with other cases further, |
| 1388 |
// under '1 <= q + exp <= 19' |
| 1389 |
if (x_sign) { // if n < 0 and q + exp = 19 |
| 1390 |
// if n <= -2^63 - 1 then n is too large |
| 1391 |
// too large if c(0)c(1)...c(18).c(19)...c(q-1) >= 2^63+1 |
| 1392 |
// <=> 0.c(0)c(1)...c(q-1) * 10^20 > 5*(2^64+2), 1<=q<=34 |
| 1393 |
// <=> 0.c(0)c(1)...c(q-1) * 10^20 > 0x5000000000000000a, 1<=q<=34 |
| 1394 |
C.w[1] = 0x0000000000000005ull; |
| 1395 |
C.w[0] = 0x000000000000000aull; |
| 1396 |
if (q <= 19) { // 1 <= q <= 19 => 1 <= 20-q <= 19 => |
| 1397 |
// 10^(20-q) is 64-bit, and so is C1 |
| 1398 |
__mul_64x64_to_128MACH (C1, C1.w[0], ten2k64[20 - q]); |
| 1399 |
} else if (q == 20) { |
| 1400 |
; // C1 * 10^0 = C1 |
| 1401 |
} else { // if 21 <= q <= 34 |
| 1402 |
__mul_128x64_to_128 (C, ten2k64[q - 20], C); // max 47-bit x 67-bit |
| 1403 |
} |
| 1404 |
if (C1.w[1] > C.w[1] || (C1.w[1] == C.w[1] && C1.w[0] >= C.w[0])) { |
| 1405 |
// set invalid flag |
| 1406 |
*pfpsf |= INVALID_EXCEPTION; |
| 1407 |
// return Integer Indefinite |
| 1408 |
res = 0x8000000000000000ull; |
| 1409 |
BID_RETURN (res); |
| 1410 |
} |
| 1411 |
// else cases that can be rounded to a 64-bit int fall through |
| 1412 |
// to '1 <= q + exp <= 19' |
| 1413 |
} else { // if n > 0 and q + exp = 19 |
| 1414 |
// if n > 2^63 - 1 then n is too large |
| 1415 |
// too large if c(0)c(1)...c(18).c(19)...c(q-1) > 2^63 - 1 |
| 1416 |
// <=> if 0.c(0)c(1)...c(q-1) * 10^20 > 10*(2^63-1), 1<=q<=34 |
| 1417 |
// <=> if 0.c(0)c(1)...c(q-1) * 10^20 > 0x4fffffffffffffff6, 1<=q<=34 |
| 1418 |
C.w[1] = 0x0000000000000004ull; |
| 1419 |
C.w[0] = 0xfffffffffffffff6ull; |
| 1420 |
if (q <= 19) { // 1 <= q <= 19 => 1 <= 20-q <= 19 => |
| 1421 |
// 10^(20-q) is 64-bit, and so is C1 |
| 1422 |
__mul_64x64_to_128MACH (C1, C1.w[0], ten2k64[20 - q]); |
| 1423 |
} else if (q == 20) { |
| 1424 |
; // C1 * 10^0 = C1 |
| 1425 |
} else { // if 21 <= q <= 34 |
| 1426 |
__mul_128x64_to_128 (C, ten2k64[q - 20], C); // max 47-bit x 67-bit |
| 1427 |
} |
| 1428 |
if (C1.w[1] > C.w[1] || (C1.w[1] == C.w[1] && C1.w[0] > C.w[0])) { |
| 1429 |
// set invalid flag |
| 1430 |
*pfpsf |= INVALID_EXCEPTION; |
| 1431 |
// return Integer Indefinite |
| 1432 |
res = 0x8000000000000000ull; |
| 1433 |
BID_RETURN (res); |
| 1434 |
} |
| 1435 |
// else cases that can be rounded to a 64-bit int fall through |
| 1436 |
// to '1 <= q + exp <= 19' |
| 1437 |
} |
| 1438 |
} |
| 1439 |
// n is not too large to be converted to int64: -2^63-1 < n <= 2^63 - 1 |
| 1440 |
// Note: some of the cases tested for above fall through to this point |
| 1441 |
// Restore C1 which may have been modified above |
| 1442 |
C1.w[1] = x.w[1] & MASK_COEFF; |
| 1443 |
C1.w[0] = x.w[0]; |
| 1444 |
if ((q + exp) <= 0) { // n = +/-0.[0...0]c(0)c(1)...c(q-1) |
| 1445 |
// return 0 or 1 |
| 1446 |
if (x_sign) |
| 1447 |
res = 0x0000000000000000ull; |
| 1448 |
else |
| 1449 |
res = 0x0000000000000001ull; |
| 1450 |
BID_RETURN (res); |
| 1451 |
} else { // if (1 <= q + exp <= 19, 1 <= q <= 34, -33 <= exp <= 18) |
| 1452 |
// -2^63-1 < x <= -1 or 1 <= x <= 2^63 - 1 so x can be rounded |
| 1453 |
// up to a 64-bit signed integer |
| 1454 |
if (exp < 0) { // 2 <= q <= 34, -33 <= exp <= -1, 1 <= q + exp <= 19 |
| 1455 |
ind = -exp; // 1 <= ind <= 33; ind is a synonym for 'x' |
| 1456 |
// chop off ind digits from the lower part of C1 |
| 1457 |
// C1 fits in 127 bits |
| 1458 |
// calculate C* and f* |
| 1459 |
// C* is actually floor(C*) in this case |
| 1460 |
// C* and f* need shifting and masking, as shown by |
| 1461 |
// shiftright128[] and maskhigh128[] |
| 1462 |
// 1 <= x <= 33 |
| 1463 |
// kx = 10^(-x) = ten2mk128[ind - 1] |
| 1464 |
// C* = C1 * 10^(-x) |
| 1465 |
// the approximation of 10^(-x) was rounded up to 118 bits |
| 1466 |
__mul_128x128_to_256 (P256, C1, ten2mk128[ind - 1]); |
| 1467 |
if (ind - 1 <= 21) { // 0 <= ind - 1 <= 21 |
| 1468 |
Cstar.w[1] = P256.w[3]; |
| 1469 |
Cstar.w[0] = P256.w[2]; |
| 1470 |
fstar.w[3] = 0; |
| 1471 |
fstar.w[2] = P256.w[2] & maskhigh128[ind - 1]; |
| 1472 |
fstar.w[1] = P256.w[1]; |
| 1473 |
fstar.w[0] = P256.w[0]; |
| 1474 |
} else { // 22 <= ind - 1 <= 33 |
| 1475 |
Cstar.w[1] = 0; |
| 1476 |
Cstar.w[0] = P256.w[3]; |
| 1477 |
fstar.w[3] = P256.w[3] & maskhigh128[ind - 1]; |
| 1478 |
fstar.w[2] = P256.w[2]; |
| 1479 |
fstar.w[1] = P256.w[1]; |
| 1480 |
fstar.w[0] = P256.w[0]; |
| 1481 |
} |
| 1482 |
// the top Ex bits of 10^(-x) are T* = ten2mk128trunc[ind], e.g. |
| 1483 |
// if x=1, T*=ten2mk128trunc[0]=0x19999999999999999999999999999999 |
| 1484 |
// C* = floor(C*) (logical right shift; C has p decimal digits, |
| 1485 |
// correct by Property 1) |
| 1486 |
// n = C* * 10^(e+x) |
| 1487 |
|
| 1488 |
// shift right C* by Ex-128 = shiftright128[ind] |
| 1489 |
shift = shiftright128[ind - 1]; // 0 <= shift <= 102 |
| 1490 |
if (ind - 1 <= 21) { // 0 <= ind - 1 <= 21 |
| 1491 |
Cstar.w[0] = |
| 1492 |
(Cstar.w[0] >> shift) | (Cstar.w[1] << (64 - shift)); |
| 1493 |
// redundant, it will be 0! Cstar.w[1] = (Cstar.w[1] >> shift); |
| 1494 |
} else { // 22 <= ind - 1 <= 33 |
| 1495 |
Cstar.w[0] = (Cstar.w[0] >> (shift - 64)); // 2 <= shift - 64 <= 38 |
| 1496 |
} |
| 1497 |
// if the result is positive and inexact, need to add 1 to it |
| 1498 |
|
| 1499 |
// determine inexactness of the rounding of C* |
| 1500 |
// if (0 < f* < 10^(-x)) then |
| 1501 |
// the result is exact |
| 1502 |
// else // if (f* > T*) then |
| 1503 |
// the result is inexact |
| 1504 |
if (ind - 1 <= 2) { |
| 1505 |
if (fstar.w[1] > ten2mk128trunc[ind - 1].w[1] |
| 1506 |
|| (fstar.w[1] == ten2mk128trunc[ind - 1].w[1] |
| 1507 |
&& fstar.w[0] > ten2mk128trunc[ind - 1].w[0])) { |
| 1508 |
if (!x_sign) { // positive and inexact |
| 1509 |
Cstar.w[0]++; |
| 1510 |
if (Cstar.w[0] == 0x0) |
| 1511 |
Cstar.w[1]++; |
| 1512 |
} |
| 1513 |
} // else the result is exact |
| 1514 |
} else if (ind - 1 <= 21) { // if 3 <= ind <= 21 |
| 1515 |
if (fstar.w[2] || fstar.w[1] > ten2mk128trunc[ind - 1].w[1] |
| 1516 |
|| (fstar.w[1] == ten2mk128trunc[ind - 1].w[1] |
| 1517 |
&& fstar.w[0] > ten2mk128trunc[ind - 1].w[0])) { |
| 1518 |
if (!x_sign) { // positive and inexact |
| 1519 |
Cstar.w[0]++; |
| 1520 |
if (Cstar.w[0] == 0x0) |
| 1521 |
Cstar.w[1]++; |
| 1522 |
} |
| 1523 |
} // else the result is exact |
| 1524 |
} else { // if 22 <= ind <= 33 |
| 1525 |
if (fstar.w[3] || fstar.w[2] |
| 1526 |
|| fstar.w[1] > ten2mk128trunc[ind - 1].w[1] |
| 1527 |
|| (fstar.w[1] == ten2mk128trunc[ind - 1].w[1] |
| 1528 |
&& fstar.w[0] > ten2mk128trunc[ind - 1].w[0])) { |
| 1529 |
if (!x_sign) { // positive and inexact |
| 1530 |
Cstar.w[0]++; |
| 1531 |
if (Cstar.w[0] == 0x0) |
| 1532 |
Cstar.w[1]++; |
| 1533 |
} |
| 1534 |
} // else the result is exact |
| 1535 |
} |
| 1536 |
if (x_sign) |
| 1537 |
res = -Cstar.w[0]; |
| 1538 |
else |
| 1539 |
res = Cstar.w[0]; |
| 1540 |
} else if (exp == 0) { |
| 1541 |
// 1 <= q <= 19 |
| 1542 |
// res = +/-C (exact) |
| 1543 |
if (x_sign) |
| 1544 |
res = -C1.w[0]; |
| 1545 |
else |
| 1546 |
res = C1.w[0]; |
| 1547 |
} else { // if (exp>0) => 1 <= exp <= 18, 1 <= q < 18, 2 <= q + exp <= 19 |
| 1548 |
// res = +/-C * 10^exp (exact) where this fits in 64-bit integer |
| 1549 |
if (x_sign) |
| 1550 |
res = -C1.w[0] * ten2k64[exp]; |
| 1551 |
else |
| 1552 |
res = C1.w[0] * ten2k64[exp]; |
| 1553 |
} |
| 1554 |
} |
| 1555 |
} |
| 1556 |
|
| 1557 |
BID_RETURN (res); |
| 1558 |
} |
| 1559 |
|
| 1560 |
/***************************************************************************** |
| 1561 |
* BID128_to_int64_xceil |
| 1562 |
****************************************************************************/ |
| 1563 |
|
| 1564 |
BID128_FUNCTION_ARG1_NORND_CUSTOMRESTYPE (SINT64, bid128_to_int64_xceil, |
| 1565 |
x) |
| 1566 |
|
| 1567 |
SINT64 res; |
| 1568 |
UINT64 x_sign; |
| 1569 |
UINT64 x_exp; |
| 1570 |
int exp; // unbiased exponent |
| 1571 |
// Note: C1.w[1], C1.w[0] represent x_signif_hi, x_signif_lo (all are UINT64) |
| 1572 |
BID_UI64DOUBLE tmp1; |
| 1573 |
unsigned int x_nr_bits; |
| 1574 |
int q, ind, shift; |
| 1575 |
UINT128 C1, C; |
| 1576 |
UINT128 Cstar; // C* represents up to 34 decimal digits ~ 113 bits |
| 1577 |
UINT256 fstar; |
| 1578 |
UINT256 P256; |
| 1579 |
|
| 1580 |
// unpack x |
| 1581 |
x_sign = x.w[1] & MASK_SIGN; // 0 for positive, MASK_SIGN for negative |
| 1582 |
x_exp = x.w[1] & MASK_EXP; // biased and shifted left 49 bit positions |
| 1583 |
C1.w[1] = x.w[1] & MASK_COEFF; |
| 1584 |
C1.w[0] = x.w[0]; |
| 1585 |
|
| 1586 |
// check for NaN or Infinity |
| 1587 |
if ((x.w[1] & MASK_SPECIAL) == MASK_SPECIAL) { |
| 1588 |
// x is special |
| 1589 |
if ((x.w[1] & MASK_NAN) == MASK_NAN) { // x is NAN |
| 1590 |
if ((x.w[1] & MASK_SNAN) == MASK_SNAN) { // x is SNAN |
| 1591 |
// set invalid flag |
| 1592 |
*pfpsf |= INVALID_EXCEPTION; |
| 1593 |
// return Integer Indefinite |
| 1594 |
res = 0x8000000000000000ull; |
| 1595 |
} else { // x is QNaN |
| 1596 |
// set invalid flag |
| 1597 |
*pfpsf |= INVALID_EXCEPTION; |
| 1598 |
// return Integer Indefinite |
| 1599 |
res = 0x8000000000000000ull; |
| 1600 |
} |
| 1601 |
BID_RETURN (res); |
| 1602 |
} else { // x is not a NaN, so it must be infinity |
| 1603 |
if (!x_sign) { // x is +inf |
| 1604 |
// set invalid flag |
| 1605 |
*pfpsf |= INVALID_EXCEPTION; |
| 1606 |
// return Integer Indefinite |
| 1607 |
res = 0x8000000000000000ull; |
| 1608 |
} else { // x is -inf |
| 1609 |
// set invalid flag |
| 1610 |
*pfpsf |= INVALID_EXCEPTION; |
| 1611 |
// return Integer Indefinite |
| 1612 |
res = 0x8000000000000000ull; |
| 1613 |
} |
| 1614 |
BID_RETURN (res); |
| 1615 |
} |
| 1616 |
} |
| 1617 |
// check for non-canonical values (after the check for special values) |
| 1618 |
if ((C1.w[1] > 0x0001ed09bead87c0ull) |
| 1619 |
|| (C1.w[1] == 0x0001ed09bead87c0ull |
| 1620 |
&& (C1.w[0] > 0x378d8e63ffffffffull)) |
| 1621 |
|| ((x.w[1] & 0x6000000000000000ull) == 0x6000000000000000ull)) { |
| 1622 |
res = 0x0000000000000000ull; |
| 1623 |
BID_RETURN (res); |
| 1624 |
} else if ((C1.w[1] == 0x0ull) && (C1.w[0] == 0x0ull)) { |
| 1625 |
// x is 0 |
| 1626 |
res = 0x0000000000000000ull; |
| 1627 |
BID_RETURN (res); |
| 1628 |
} else { // x is not special and is not zero |
| 1629 |
|
| 1630 |
// q = nr. of decimal digits in x |
| 1631 |
// determine first the nr. of bits in x |
| 1632 |
if (C1.w[1] == 0) { |
| 1633 |
if (C1.w[0] >= 0x0020000000000000ull) { // x >= 2^53 |
| 1634 |
// split the 64-bit value in two 32-bit halves to avoid rounding errors |
| 1635 |
if (C1.w[0] >= 0x0000000100000000ull) { // x >= 2^32 |
| 1636 |
tmp1.d = (double) (C1.w[0] >> 32); // exact conversion |
| 1637 |
x_nr_bits = |
| 1638 |
33 + ((((unsigned int) (tmp1.ui64 >> 52)) & 0x7ff) - 0x3ff); |
| 1639 |
} else { // x < 2^32 |
| 1640 |
tmp1.d = (double) (C1.w[0]); // exact conversion |
| 1641 |
x_nr_bits = |
| 1642 |
1 + ((((unsigned int) (tmp1.ui64 >> 52)) & 0x7ff) - 0x3ff); |
| 1643 |
} |
| 1644 |
} else { // if x < 2^53 |
| 1645 |
tmp1.d = (double) C1.w[0]; // exact conversion |
| 1646 |
x_nr_bits = |
| 1647 |
1 + ((((unsigned int) (tmp1.ui64 >> 52)) & 0x7ff) - 0x3ff); |
| 1648 |
} |
| 1649 |
} else { // C1.w[1] != 0 => nr. bits = 64 + nr_bits (C1.w[1]) |
| 1650 |
tmp1.d = (double) C1.w[1]; // exact conversion |
| 1651 |
x_nr_bits = |
| 1652 |
65 + ((((unsigned int) (tmp1.ui64 >> 52)) & 0x7ff) - 0x3ff); |
| 1653 |
} |
| 1654 |
q = nr_digits[x_nr_bits - 1].digits; |
| 1655 |
if (q == 0) { |
| 1656 |
q = nr_digits[x_nr_bits - 1].digits1; |
| 1657 |
if (C1.w[1] > nr_digits[x_nr_bits - 1].threshold_hi |
| 1658 |
|| (C1.w[1] == nr_digits[x_nr_bits - 1].threshold_hi |
| 1659 |
&& C1.w[0] >= nr_digits[x_nr_bits - 1].threshold_lo)) |
| 1660 |
q++; |
| 1661 |
} |
| 1662 |
exp = (x_exp >> 49) - 6176; |
| 1663 |
if ((q + exp) > 19) { // x >= 10^19 ~= 2^63.11... (cannot fit in SINT64) |
| 1664 |
// set invalid flag |
| 1665 |
*pfpsf |= INVALID_EXCEPTION; |
| 1666 |
// return Integer Indefinite |
| 1667 |
res = 0x8000000000000000ull; |
| 1668 |
BID_RETURN (res); |
| 1669 |
} else if ((q + exp) == 19) { // x = c(0)c(1)...c(18).c(19)...c(q-1) |
| 1670 |
// in this case 2^63.11... ~= 10^19 <= x < 10^20 ~= 2^66.43... |
| 1671 |
// so x rounded to an integer may or may not fit in a signed 64-bit int |
| 1672 |
// the cases that do not fit are identified here; the ones that fit |
| 1673 |
// fall through and will be handled with other cases further, |
| 1674 |
// under '1 <= q + exp <= 19' |
| 1675 |
if (x_sign) { // if n < 0 and q + exp = 19 |
| 1676 |
// if n <= -2^63 - 1 then n is too large |
| 1677 |
// too large if c(0)c(1)...c(18).c(19)...c(q-1) >= 2^63+1 |
| 1678 |
// <=> 0.c(0)c(1)...c(q-1) * 10^20 > 5*(2^64+2), 1<=q<=34 |
| 1679 |
// <=> 0.c(0)c(1)...c(q-1) * 10^20 > 0x5000000000000000a, 1<=q<=34 |
| 1680 |
C.w[1] = 0x0000000000000005ull; |
| 1681 |
C.w[0] = 0x000000000000000aull; |
| 1682 |
if (q <= 19) { // 1 <= q <= 19 => 1 <= 20-q <= 19 => |
| 1683 |
// 10^(20-q) is 64-bit, and so is C1 |
| 1684 |
__mul_64x64_to_128MACH (C1, C1.w[0], ten2k64[20 - q]); |
| 1685 |
} else if (q == 20) { |
| 1686 |
; // C1 * 10^0 = C1 |
| 1687 |
} else { // if 21 <= q <= 34 |
| 1688 |
__mul_128x64_to_128 (C, ten2k64[q - 20], C); // max 47-bit x 67-bit |
| 1689 |
} |
| 1690 |
if (C1.w[1] > C.w[1] || (C1.w[1] == C.w[1] && C1.w[0] >= C.w[0])) { |
| 1691 |
// set invalid flag |
| 1692 |
*pfpsf |= INVALID_EXCEPTION; |
| 1693 |
// return Integer Indefinite |
| 1694 |
res = 0x8000000000000000ull; |
| 1695 |
BID_RETURN (res); |
| 1696 |
} |
| 1697 |
// else cases that can be rounded to a 64-bit int fall through |
| 1698 |
// to '1 <= q + exp <= 19' |
| 1699 |
} else { // if n > 0 and q + exp = 19 |
| 1700 |
// if n > 2^63 - 1 then n is too large |
| 1701 |
// too large if c(0)c(1)...c(18).c(19)...c(q-1) > 2^63 - 1 |
| 1702 |
// <=> if 0.c(0)c(1)...c(q-1) * 10^20 > 10*(2^63-1), 1<=q<=34 |
| 1703 |
// <=> if 0.c(0)c(1)...c(q-1) * 10^20 > 0x4fffffffffffffff6, 1<=q<=34 |
| 1704 |
C.w[1] = 0x0000000000000004ull; |
| 1705 |
C.w[0] = 0xfffffffffffffff6ull; |
| 1706 |
if (q <= 19) { // 1 <= q <= 19 => 1 <= 20-q <= 19 => |
| 1707 |
// 10^(20-q) is 64-bit, and so is C1 |
| 1708 |
__mul_64x64_to_128MACH (C1, C1.w[0], ten2k64[20 - q]); |
| 1709 |
} else if (q == 20) { |
| 1710 |
; // C1 * 10^0 = C1 |
| 1711 |
} else { // if 21 <= q <= 34 |
| 1712 |
__mul_128x64_to_128 (C, ten2k64[q - 20], C); // max 47-bit x 67-bit |
| 1713 |
} |
| 1714 |
if (C1.w[1] > C.w[1] || (C1.w[1] == C.w[1] && C1.w[0] > C.w[0])) { |
| 1715 |
// set invalid flag |
| 1716 |
*pfpsf |= INVALID_EXCEPTION; |
| 1717 |
// return Integer Indefinite |
| 1718 |
res = 0x8000000000000000ull; |
| 1719 |
BID_RETURN (res); |
| 1720 |
} |
| 1721 |
// else cases that can be rounded to a 64-bit int fall through |
| 1722 |
// to '1 <= q + exp <= 19' |
| 1723 |
} |
| 1724 |
} |
| 1725 |
// n is not too large to be converted to int64: -2^63-1 < n <= 2^63 - 1 |
| 1726 |
// Note: some of the cases tested for above fall through to this point |
| 1727 |
// Restore C1 which may have been modified above |
| 1728 |
C1.w[1] = x.w[1] & MASK_COEFF; |
| 1729 |
C1.w[0] = x.w[0]; |
| 1730 |
if ((q + exp) <= 0) { // n = +/-0.[0...0]c(0)c(1)...c(q-1) |
| 1731 |
// set inexact flag |
| 1732 |
*pfpsf |= INEXACT_EXCEPTION; |
| 1733 |
// return 0 or 1 |
| 1734 |
if (x_sign) |
| 1735 |
res = 0x0000000000000000ull; |
| 1736 |
else |
| 1737 |
res = 0x0000000000000001ull; |
| 1738 |
BID_RETURN (res); |
| 1739 |
} else { // if (1 <= q + exp <= 19, 1 <= q <= 34, -33 <= exp <= 18) |
| 1740 |
// -2^63-1 < x <= -1 or 1 <= x <= 2^63 - 1 so x can be rounded |
| 1741 |
// up to a 64-bit signed integer |
| 1742 |
if (exp < 0) { // 2 <= q <= 34, -33 <= exp <= -1, 1 <= q + exp <= 19 |
| 1743 |
ind = -exp; // 1 <= ind <= 33; ind is a synonym for 'x' |
| 1744 |
// chop off ind digits from the lower part of C1 |
| 1745 |
// C1 fits in 127 bits |
| 1746 |
// calculate C* and f* |
| 1747 |
// C* is actually floor(C*) in this case |
| 1748 |
// C* and f* need shifting and masking, as shown by |
| 1749 |
// shiftright128[] and maskhigh128[] |
| 1750 |
// 1 <= x <= 33 |
| 1751 |
// kx = 10^(-x) = ten2mk128[ind - 1] |
| 1752 |
// C* = C1 * 10^(-x) |
| 1753 |
// the approximation of 10^(-x) was rounded up to 118 bits |
| 1754 |
__mul_128x128_to_256 (P256, C1, ten2mk128[ind - 1]); |
| 1755 |
if (ind - 1 <= 21) { // 0 <= ind - 1 <= 21 |
| 1756 |
Cstar.w[1] = P256.w[3]; |
| 1757 |
Cstar.w[0] = P256.w[2]; |
| 1758 |
fstar.w[3] = 0; |
| 1759 |
fstar.w[2] = P256.w[2] & maskhigh128[ind - 1]; |
| 1760 |
fstar.w[1] = P256.w[1]; |
| 1761 |
fstar.w[0] = P256.w[0]; |
| 1762 |
} else { // 22 <= ind - 1 <= 33 |
| 1763 |
Cstar.w[1] = 0; |
| 1764 |
Cstar.w[0] = P256.w[3]; |
| 1765 |
fstar.w[3] = P256.w[3] & maskhigh128[ind - 1]; |
| 1766 |
fstar.w[2] = P256.w[2]; |
| 1767 |
fstar.w[1] = P256.w[1]; |
| 1768 |
fstar.w[0] = P256.w[0]; |
| 1769 |
} |
| 1770 |
// the top Ex bits of 10^(-x) are T* = ten2mk128trunc[ind], e.g. |
| 1771 |
// if x=1, T*=ten2mk128trunc[0]=0x19999999999999999999999999999999 |
| 1772 |
// C* = floor(C*) (logical right shift; C has p decimal digits, |
| 1773 |
// correct by Property 1) |
| 1774 |
// n = C* * 10^(e+x) |
| 1775 |
|
| 1776 |
// shift right C* by Ex-128 = shiftright128[ind] |
| 1777 |
shift = shiftright128[ind - 1]; // 0 <= shift <= 102 |
| 1778 |
if (ind - 1 <= 21) { // 0 <= ind - 1 <= 21 |
| 1779 |
Cstar.w[0] = |
| 1780 |
(Cstar.w[0] >> shift) | (Cstar.w[1] << (64 - shift)); |
| 1781 |
// redundant, it will be 0! Cstar.w[1] = (Cstar.w[1] >> shift); |
| 1782 |
} else { // 22 <= ind - 1 <= 33 |
| 1783 |
Cstar.w[0] = (Cstar.w[0] >> (shift - 64)); // 2 <= shift - 64 <= 38 |
| 1784 |
} |
| 1785 |
// if the result is positive and inexact, need to add 1 to it |
| 1786 |
|
| 1787 |
// determine inexactness of the rounding of C* |
| 1788 |
// if (0 < f* < 10^(-x)) then |
| 1789 |
// the result is exact |
| 1790 |
// else // if (f* > T*) then |
| 1791 |
// the result is inexact |
| 1792 |
if (ind - 1 <= 2) { |
| 1793 |
if (fstar.w[1] > ten2mk128trunc[ind - 1].w[1] |
| 1794 |
|| (fstar.w[1] == ten2mk128trunc[ind - 1].w[1] |
| 1795 |
&& fstar.w[0] > ten2mk128trunc[ind - 1].w[0])) { |
| 1796 |
if (!x_sign) { // positive and inexact |
| 1797 |
Cstar.w[0]++; |
| 1798 |
if (Cstar.w[0] == 0x0) |
| 1799 |
Cstar.w[1]++; |
| 1800 |
} |
| 1801 |
// set the inexact flag |
| 1802 |
*pfpsf |= INEXACT_EXCEPTION; |
| 1803 |
} // else the result is exact |
| 1804 |
} else if (ind - 1 <= 21) { // if 3 <= ind <= 21 |
| 1805 |
if (fstar.w[2] || fstar.w[1] > ten2mk128trunc[ind - 1].w[1] |
| 1806 |
|| (fstar.w[1] == ten2mk128trunc[ind - 1].w[1] |
| 1807 |
&& fstar.w[0] > ten2mk128trunc[ind - 1].w[0])) { |
| 1808 |
if (!x_sign) { // positive and inexact |
| 1809 |
Cstar.w[0]++; |
| 1810 |
if (Cstar.w[0] == 0x0) |
| 1811 |
Cstar.w[1]++; |
| 1812 |
} |
| 1813 |
// set the inexact flag |
| 1814 |
*pfpsf |= INEXACT_EXCEPTION; |
| 1815 |
} // else the result is exact |
| 1816 |
} else { // if 22 <= ind <= 33 |
| 1817 |
if (fstar.w[3] || fstar.w[2] |
| 1818 |
|| fstar.w[1] > ten2mk128trunc[ind - 1].w[1] |
| 1819 |
|| (fstar.w[1] == ten2mk128trunc[ind - 1].w[1] |
| 1820 |
&& fstar.w[0] > ten2mk128trunc[ind - 1].w[0])) { |
| 1821 |
if (!x_sign) { // positive and inexact |
| 1822 |
Cstar.w[0]++; |
| 1823 |
if (Cstar.w[0] == 0x0) |
| 1824 |
Cstar.w[1]++; |
| 1825 |
} |
| 1826 |
// set the inexact flag |
| 1827 |
*pfpsf |= INEXACT_EXCEPTION; |
| 1828 |
} // else the result is exact |
| 1829 |
} |
| 1830 |
|
| 1831 |
if (x_sign) |
| 1832 |
res = -Cstar.w[0]; |
| 1833 |
else |
| 1834 |
res = Cstar.w[0]; |
| 1835 |
} else if (exp == 0) { |
| 1836 |
// 1 <= q <= 19 |
| 1837 |
// res = +/-C (exact) |
| 1838 |
if (x_sign) |
| 1839 |
res = -C1.w[0]; |
| 1840 |
else |
| 1841 |
res = C1.w[0]; |
| 1842 |
} else { // if (exp>0) => 1 <= exp <= 18, 1 <= q < 18, 2 <= q + exp <= 19 |
| 1843 |
// res = +/-C * 10^exp (exact) where this fits in 64-bit integer |
| 1844 |
if (x_sign) |
| 1845 |
res = -C1.w[0] * ten2k64[exp]; |
| 1846 |
else |
| 1847 |
res = C1.w[0] * ten2k64[exp]; |
| 1848 |
} |
| 1849 |
} |
| 1850 |
} |
| 1851 |
|
| 1852 |
BID_RETURN (res); |
| 1853 |
} |
| 1854 |
|
| 1855 |
/***************************************************************************** |
| 1856 |
* BID128_to_int64_int |
| 1857 |
****************************************************************************/ |
| 1858 |
|
| 1859 |
BID128_FUNCTION_ARG1_NORND_CUSTOMRESTYPE (SINT64, bid128_to_int64_int, |
| 1860 |
x) |
| 1861 |
|
| 1862 |
SINT64 res; |
| 1863 |
UINT64 x_sign; |
| 1864 |
UINT64 x_exp; |
| 1865 |
int exp; // unbiased exponent |
| 1866 |
// Note: C1.w[1], C1.w[0] represent x_signif_hi, x_signif_lo (all are UINT64) |
| 1867 |
BID_UI64DOUBLE tmp1; |
| 1868 |
unsigned int x_nr_bits; |
| 1869 |
int q, ind, shift; |
| 1870 |
UINT128 C1, C; |
| 1871 |
UINT128 Cstar; // C* represents up to 34 decimal digits ~ 113 bits |
| 1872 |
UINT256 P256; |
| 1873 |
|
| 1874 |
// unpack x |
| 1875 |
x_sign = x.w[1] & MASK_SIGN; // 0 for positive, MASK_SIGN for negative |
| 1876 |
x_exp = x.w[1] & MASK_EXP; // biased and shifted left 49 bit positions |
| 1877 |
C1.w[1] = x.w[1] & MASK_COEFF; |
| 1878 |
C1.w[0] = x.w[0]; |
| 1879 |
|
| 1880 |
// check for NaN or Infinity |
| 1881 |
if ((x.w[1] & MASK_SPECIAL) == MASK_SPECIAL) { |
| 1882 |
// x is special |
| 1883 |
if ((x.w[1] & MASK_NAN) == MASK_NAN) { // x is NAN |
| 1884 |
if ((x.w[1] & MASK_SNAN) == MASK_SNAN) { // x is SNAN |
| 1885 |
// set invalid flag |
| 1886 |
*pfpsf |= INVALID_EXCEPTION; |
| 1887 |
// return Integer Indefinite |
| 1888 |
res = 0x8000000000000000ull; |
| 1889 |
} else { // x is QNaN |
| 1890 |
// set invalid flag |
| 1891 |
*pfpsf |= INVALID_EXCEPTION; |
| 1892 |
// return Integer Indefinite |
| 1893 |
res = 0x8000000000000000ull; |
| 1894 |
} |
| 1895 |
BID_RETURN (res); |
| 1896 |
} else { // x is not a NaN, so it must be infinity |
| 1897 |
if (!x_sign) { // x is +inf |
| 1898 |
// set invalid flag |
| 1899 |
*pfpsf |= INVALID_EXCEPTION; |
| 1900 |
// return Integer Indefinite |
| 1901 |
res = 0x8000000000000000ull; |
| 1902 |
} else { // x is -inf |
| 1903 |
// set invalid flag |
| 1904 |
*pfpsf |= INVALID_EXCEPTION; |
| 1905 |
// return Integer Indefinite |
| 1906 |
res = 0x8000000000000000ull; |
| 1907 |
} |
| 1908 |
BID_RETURN (res); |
| 1909 |
} |
| 1910 |
} |
| 1911 |
// check for non-canonical values (after the check for special values) |
| 1912 |
if ((C1.w[1] > 0x0001ed09bead87c0ull) |
| 1913 |
|| (C1.w[1] == 0x0001ed09bead87c0ull |
| 1914 |
&& (C1.w[0] > 0x378d8e63ffffffffull)) |
| 1915 |
|| ((x.w[1] & 0x6000000000000000ull) == 0x6000000000000000ull)) { |
| 1916 |
res = 0x0000000000000000ull; |
| 1917 |
BID_RETURN (res); |
| 1918 |
} else if ((C1.w[1] == 0x0ull) && (C1.w[0] == 0x0ull)) { |
| 1919 |
// x is 0 |
| 1920 |
res = 0x0000000000000000ull; |
| 1921 |
BID_RETURN (res); |
| 1922 |
} else { // x is not special and is not zero |
| 1923 |
|
| 1924 |
// q = nr. of decimal digits in x |
| 1925 |
// determine first the nr. of bits in x |
| 1926 |
if (C1.w[1] == 0) { |
| 1927 |
if (C1.w[0] >= 0x0020000000000000ull) { // x >= 2^53 |
| 1928 |
// split the 64-bit value in two 32-bit halves to avoid rounding errors |
| 1929 |
if (C1.w[0] >= 0x0000000100000000ull) { // x >= 2^32 |
| 1930 |
tmp1.d = (double) (C1.w[0] >> 32); // exact conversion |
| 1931 |
x_nr_bits = |
| 1932 |
33 + ((((unsigned int) (tmp1.ui64 >> 52)) & 0x7ff) - 0x3ff); |
| 1933 |
} else { // x < 2^32 |
| 1934 |
tmp1.d = (double) (C1.w[0]); // exact conversion |
| 1935 |
x_nr_bits = |
| 1936 |
1 + ((((unsigned int) (tmp1.ui64 >> 52)) & 0x7ff) - 0x3ff); |
| 1937 |
} |
| 1938 |
} else { // if x < 2^53 |
| 1939 |
tmp1.d = (double) C1.w[0]; // exact conversion |
| 1940 |
x_nr_bits = |
| 1941 |
1 + ((((unsigned int) (tmp1.ui64 >> 52)) & 0x7ff) - 0x3ff); |
| 1942 |
} |
| 1943 |
} else { // C1.w[1] != 0 => nr. bits = 64 + nr_bits (C1.w[1]) |
| 1944 |
tmp1.d = (double) C1.w[1]; // exact conversion |
| 1945 |
x_nr_bits = |
| 1946 |
65 + ((((unsigned int) (tmp1.ui64 >> 52)) & 0x7ff) - 0x3ff); |
| 1947 |
} |
| 1948 |
q = nr_digits[x_nr_bits - 1].digits; |
| 1949 |
if (q == 0) { |
| 1950 |
q = nr_digits[x_nr_bits - 1].digits1; |
| 1951 |
if (C1.w[1] > nr_digits[x_nr_bits - 1].threshold_hi |
| 1952 |
|| (C1.w[1] == nr_digits[x_nr_bits - 1].threshold_hi |
| 1953 |
&& C1.w[0] >= nr_digits[x_nr_bits - 1].threshold_lo)) |
| 1954 |
q++; |
| 1955 |
} |
| 1956 |
exp = (x_exp >> 49) - 6176; |
| 1957 |
if ((q + exp) > 19) { // x >= 10^19 ~= 2^63.11... (cannot fit in SINT64) |
| 1958 |
// set invalid flag |
| 1959 |
*pfpsf |= INVALID_EXCEPTION; |
| 1960 |
// return Integer Indefinite |
| 1961 |
res = 0x8000000000000000ull; |
| 1962 |
BID_RETURN (res); |
| 1963 |
} else if ((q + exp) == 19) { // x = c(0)c(1)...c(18).c(19)...c(q-1) |
| 1964 |
// in this case 2^63.11... ~= 10^19 <= x < 10^20 ~= 2^66.43... |
| 1965 |
// so x rounded to an integer may or may not fit in a signed 64-bit int |
| 1966 |
// the cases that do not fit are identified here; the ones that fit |
| 1967 |
// fall through and will be handled with other cases further, |
| 1968 |
// under '1 <= q + exp <= 19' |
| 1969 |
if (x_sign) { // if n < 0 and q + exp = 19 |
| 1970 |
// if n <= -2^63 - 1 then n is too large |
| 1971 |
// too large if c(0)c(1)...c(18).c(19)...c(q-1) >= 2^63+1 |
| 1972 |
// <=> 0.c(0)c(1)...c(q-1) * 10^20 >= 5*(2^64+2), 1<=q<=34 |
| 1973 |
// <=> 0.c(0)c(1)...c(q-1) * 10^20 >= 0x5000000000000000a, 1<=q<=34 |
| 1974 |
C.w[1] = 0x0000000000000005ull; |
| 1975 |
C.w[0] = 0x000000000000000aull; |
| 1976 |
if (q <= 19) { // 1 <= q <= 19 => 1 <= 20-q <= 19 => |
| 1977 |
// 10^(20-q) is 64-bit, and so is C1 |
| 1978 |
__mul_64x64_to_128MACH (C1, C1.w[0], ten2k64[20 - q]); |
| 1979 |
} else if (q == 20) { |
| 1980 |
; // C1 * 10^0 = C1 |
| 1981 |
} else { // if 21 <= q <= 34 |
| 1982 |
__mul_128x64_to_128 (C, ten2k64[q - 20], C); // max 47-bit x 67-bit |
| 1983 |
} |
| 1984 |
if (C1.w[1] > C.w[1] || (C1.w[1] == C.w[1] && C1.w[0] >= C.w[0])) { |
| 1985 |
// set invalid flag |
| 1986 |
*pfpsf |= INVALID_EXCEPTION; |
| 1987 |
// return Integer Indefinite |
| 1988 |
res = 0x8000000000000000ull; |
| 1989 |
BID_RETURN (res); |
| 1990 |
} |
| 1991 |
// else cases that can be rounded to a 64-bit int fall through |
| 1992 |
// to '1 <= q + exp <= 19' |
| 1993 |
} else { // if n > 0 and q + exp = 19 |
| 1994 |
// if n >= 2^63 then n is too large |
| 1995 |
// too large if c(0)c(1)...c(18).c(19)...c(q-1) >= 2^63 |
| 1996 |
// <=> if 0.c(0)c(1)...c(q-1) * 10^20 >= 5*2^64, 1<=q<=34 |
| 1997 |
// <=> if 0.c(0)c(1)...c(q-1) * 10^20 >= 0x50000000000000000, 1<=q<=34 |
| 1998 |
C.w[1] = 0x0000000000000005ull; |
| 1999 |
C.w[0] = 0x0000000000000000ull; |
| 2000 |
if (q <= 19) { // 1 <= q <= 19 => 1 <= 20-q <= 19 => |
| 2001 |
// 10^(20-q) is 64-bit, and so is C1 |
| 2002 |
__mul_64x64_to_128MACH (C1, C1.w[0], ten2k64[20 - q]); |
| 2003 |
} else if (q == 20) { |
| 2004 |
; // C1 * 10^0 = C1 |
| 2005 |
} else { // if 21 <= q <= 34 |
| 2006 |
__mul_128x64_to_128 (C, ten2k64[q - 20], C); // max 47-bit x 67-bit |
| 2007 |
} |
| 2008 |
if (C1.w[1] > C.w[1] || (C1.w[1] == C.w[1] && C1.w[0] >= C.w[0])) { |
| 2009 |
// set invalid flag |
| 2010 |
*pfpsf |= INVALID_EXCEPTION; |
| 2011 |
// return Integer Indefinite |
| 2012 |
res = 0x8000000000000000ull; |
| 2013 |
BID_RETURN (res); |
| 2014 |
} |
| 2015 |
// else cases that can be rounded to a 64-bit int fall through |
| 2016 |
// to '1 <= q + exp <= 19' |
| 2017 |
} |
| 2018 |
} |
| 2019 |
// n is not too large to be converted to int64: -2^63-1 < n < 2^63 |
| 2020 |
// Note: some of the cases tested for above fall through to this point |
| 2021 |
// Restore C1 which may have been modified above |
| 2022 |
C1.w[1] = x.w[1] & MASK_COEFF; |
| 2023 |
C1.w[0] = x.w[0]; |
| 2024 |
if ((q + exp) <= 0) { // n = +/-0.[0...0]c(0)c(1)...c(q-1) |
| 2025 |
// return 0 |
| 2026 |
res = 0x0000000000000000ull; |
| 2027 |
BID_RETURN (res); |
| 2028 |
} else { // if (1 <= q + exp <= 19, 1 <= q <= 34, -33 <= exp <= 18) |
| 2029 |
// -2^63-1 < x <= -1 or 1 <= x < 2^63 so x can be rounded |
| 2030 |
// toward zero to a 64-bit signed integer |
| 2031 |
if (exp < 0) { // 2 <= q <= 34, -33 <= exp <= -1, 1 <= q + exp <= 19 |
| 2032 |
ind = -exp; // 1 <= ind <= 33; ind is a synonym for 'x' |
| 2033 |
// chop off ind digits from the lower part of C1 |
| 2034 |
// C1 fits in 127 bits |
| 2035 |
// calculate C* and f* |
| 2036 |
// C* is actually floor(C*) in this case |
| 2037 |
// C* and f* need shifting and masking, as shown by |
| 2038 |
// shiftright128[] and maskhigh128[] |
| 2039 |
// 1 <= x <= 33 |
| 2040 |
// kx = 10^(-x) = ten2mk128[ind - 1] |
| 2041 |
// C* = C1 * 10^(-x) |
| 2042 |
// the approximation of 10^(-x) was rounded up to 118 bits |
| 2043 |
__mul_128x128_to_256 (P256, C1, ten2mk128[ind - 1]); |
| 2044 |
if (ind - 1 <= 21) { // 0 <= ind - 1 <= 21 |
| 2045 |
Cstar.w[1] = P256.w[3]; |
| 2046 |
Cstar.w[0] = P256.w[2]; |
| 2047 |
} else { // 22 <= ind - 1 <= 33 |
| 2048 |
Cstar.w[1] = 0; |
| 2049 |
Cstar.w[0] = P256.w[3]; |
| 2050 |
} |
| 2051 |
// the top Ex bits of 10^(-x) are T* = ten2mk128trunc[ind], e.g. |
| 2052 |
// if x=1, T*=ten2mk128trunc[0]=0x19999999999999999999999999999999 |
| 2053 |
// C* = floor(C*) (logical right shift; C has p decimal digits, |
| 2054 |
// correct by Property 1) |
| 2055 |
// n = C* * 10^(e+x) |
| 2056 |
|
| 2057 |
// shift right C* by Ex-128 = shiftright128[ind] |
| 2058 |
shift = shiftright128[ind - 1]; // 0 <= shift <= 102 |
| 2059 |
if (ind - 1 <= 21) { // 0 <= ind - 1 <= 21 |
| 2060 |
Cstar.w[0] = |
| 2061 |
(Cstar.w[0] >> shift) | (Cstar.w[1] << (64 - shift)); |
| 2062 |
// redundant, it will be 0! Cstar.w[1] = (Cstar.w[1] >> shift); |
| 2063 |
} else { // 22 <= ind - 1 <= 33 |
| 2064 |
Cstar.w[0] = (Cstar.w[0] >> (shift - 64)); // 2 <= shift - 64 <= 38 |
| 2065 |
} |
| 2066 |
if (x_sign) |
| 2067 |
res = -Cstar.w[0]; |
| 2068 |
else |
| 2069 |
res = Cstar.w[0]; |
| 2070 |
} else if (exp == 0) { |
| 2071 |
// 1 <= q <= 19 |
| 2072 |
// res = +/-C (exact) |
| 2073 |
if (x_sign) |
| 2074 |
res = -C1.w[0]; |
| 2075 |
else |
| 2076 |
res = C1.w[0]; |
| 2077 |
} else { // if (exp>0) => 1 <= exp <= 18, 1 <= q < 18, 2 <= q + exp <= 19 |
| 2078 |
// res = +/-C * 10^exp (exact) where this fits in 64-bit integer |
| 2079 |
if (x_sign) |
| 2080 |
res = -C1.w[0] * ten2k64[exp]; |
| 2081 |
else |
| 2082 |
res = C1.w[0] * ten2k64[exp]; |
| 2083 |
} |
| 2084 |
} |
| 2085 |
} |
| 2086 |
|
| 2087 |
BID_RETURN (res); |
| 2088 |
} |
| 2089 |
|
| 2090 |
/***************************************************************************** |
| 2091 |
* BID128_to_xint64_xint |
| 2092 |
****************************************************************************/ |
| 2093 |
|
| 2094 |
BID128_FUNCTION_ARG1_NORND_CUSTOMRESTYPE (SINT64, bid128_to_int64_xint, |
| 2095 |
x) |
| 2096 |
|
| 2097 |
SINT64 res; |
| 2098 |
UINT64 x_sign; |
| 2099 |
UINT64 x_exp; |
| 2100 |
int exp; // unbiased exponent |
| 2101 |
// Note: C1.w[1], C1.w[0] represent x_signif_hi, x_signif_lo (all are UINT64) |
| 2102 |
BID_UI64DOUBLE tmp1; |
| 2103 |
unsigned int x_nr_bits; |
| 2104 |
int q, ind, shift; |
| 2105 |
UINT128 C1, C; |
| 2106 |
UINT128 Cstar; // C* represents up to 34 decimal digits ~ 113 bits |
| 2107 |
UINT256 fstar; |
| 2108 |
UINT256 P256; |
| 2109 |
|
| 2110 |
// unpack x |
| 2111 |
x_sign = x.w[1] & MASK_SIGN; // 0 for positive, MASK_SIGN for negative |
| 2112 |
x_exp = x.w[1] & MASK_EXP; // biased and shifted left 49 bit positions |
| 2113 |
C1.w[1] = x.w[1] & MASK_COEFF; |
| 2114 |
C1.w[0] = x.w[0]; |
| 2115 |
|
| 2116 |
// check for NaN or Infinity |
| 2117 |
if ((x.w[1] & MASK_SPECIAL) == MASK_SPECIAL) { |
| 2118 |
// x is special |
| 2119 |
if ((x.w[1] & MASK_NAN) == MASK_NAN) { // x is NAN |
| 2120 |
if ((x.w[1] & MASK_SNAN) == MASK_SNAN) { // x is SNAN |
| 2121 |
// set invalid flag |
| 2122 |
*pfpsf |= INVALID_EXCEPTION; |
| 2123 |
// return Integer Indefinite |
| 2124 |
res = 0x8000000000000000ull; |
| 2125 |
} else { // x is QNaN |
| 2126 |
// set invalid flag |
| 2127 |
*pfpsf |= INVALID_EXCEPTION; |
| 2128 |
// return Integer Indefinite |
| 2129 |
res = 0x8000000000000000ull; |
| 2130 |
} |
| 2131 |
BID_RETURN (res); |
| 2132 |
} else { // x is not a NaN, so it must be infinity |
| 2133 |
if (!x_sign) { // x is +inf |
| 2134 |
// set invalid flag |
| 2135 |
*pfpsf |= INVALID_EXCEPTION; |
| 2136 |
// return Integer Indefinite |
| 2137 |
res = 0x8000000000000000ull; |
| 2138 |
} else { // x is -inf |
| 2139 |
// set invalid flag |
| 2140 |
*pfpsf |= INVALID_EXCEPTION; |
| 2141 |
// return Integer Indefinite |
| 2142 |
res = 0x8000000000000000ull; |
| 2143 |
} |
| 2144 |
BID_RETURN (res); |
| 2145 |
} |
| 2146 |
} |
| 2147 |
// check for non-canonical values (after the check for special values) |
| 2148 |
if ((C1.w[1] > 0x0001ed09bead87c0ull) |
| 2149 |
|| (C1.w[1] == 0x0001ed09bead87c0ull |
| 2150 |
&& (C1.w[0] > 0x378d8e63ffffffffull)) |
| 2151 |
|| ((x.w[1] & 0x6000000000000000ull) == 0x6000000000000000ull)) { |
| 2152 |
res = 0x0000000000000000ull; |
| 2153 |
BID_RETURN (res); |
| 2154 |
} else if ((C1.w[1] == 0x0ull) && (C1.w[0] == 0x0ull)) { |
| 2155 |
// x is 0 |
| 2156 |
res = 0x0000000000000000ull; |
| 2157 |
BID_RETURN (res); |
| 2158 |
} else { // x is not special and is not zero |
| 2159 |
|
| 2160 |
// q = nr. of decimal digits in x |
| 2161 |
// determine first the nr. of bits in x |
| 2162 |
if (C1.w[1] == 0) { |
| 2163 |
if (C1.w[0] >= 0x0020000000000000ull) { // x >= 2^53 |
| 2164 |
// split the 64-bit value in two 32-bit halves to avoid rounding errors |
| 2165 |
if (C1.w[0] >= 0x0000000100000000ull) { // x >= 2^32 |
| 2166 |
tmp1.d = (double) (C1.w[0] >> 32); // exact conversion |
| 2167 |
x_nr_bits = |
| 2168 |
33 + ((((unsigned int) (tmp1.ui64 >> 52)) & 0x7ff) - 0x3ff); |
| 2169 |
} else { // x < 2^32 |
| 2170 |
tmp1.d = (double) (C1.w[0]); // exact conversion |
| 2171 |
x_nr_bits = |
| 2172 |
1 + ((((unsigned int) (tmp1.ui64 >> 52)) & 0x7ff) - 0x3ff); |
| 2173 |
} |
| 2174 |
} else { // if x < 2^53 |
| 2175 |
tmp1.d = (double) C1.w[0]; // exact conversion |
| 2176 |
x_nr_bits = |
| 2177 |
1 + ((((unsigned int) (tmp1.ui64 >> 52)) & 0x7ff) - 0x3ff); |
| 2178 |
} |
| 2179 |
} else { // C1.w[1] != 0 => nr. bits = 64 + nr_bits (C1.w[1]) |
| 2180 |
tmp1.d = (double) C1.w[1]; // exact conversion |
| 2181 |
x_nr_bits = |
| 2182 |
65 + ((((unsigned int) (tmp1.ui64 >> 52)) & 0x7ff) - 0x3ff); |
| 2183 |
} |
| 2184 |
q = nr_digits[x_nr_bits - 1].digits; |
| 2185 |
if (q == 0) { |
| 2186 |
q = nr_digits[x_nr_bits - 1].digits1; |
| 2187 |
if (C1.w[1] > nr_digits[x_nr_bits - 1].threshold_hi |
| 2188 |
|| (C1.w[1] == nr_digits[x_nr_bits - 1].threshold_hi |
| 2189 |
&& C1.w[0] >= nr_digits[x_nr_bits - 1].threshold_lo)) |
| 2190 |
q++; |
| 2191 |
} |
| 2192 |
exp = (x_exp >> 49) - 6176; |
| 2193 |
if ((q + exp) > 19) { // x >= 10^19 ~= 2^63.11... (cannot fit in SINT64) |
| 2194 |
// set invalid flag |
| 2195 |
*pfpsf |= INVALID_EXCEPTION; |
| 2196 |
// return Integer Indefinite |
| 2197 |
res = 0x8000000000000000ull; |
| 2198 |
BID_RETURN (res); |
| 2199 |
} else if ((q + exp) == 19) { // x = c(0)c(1)...c(18).c(19)...c(q-1) |
| 2200 |
// in this case 2^63.11... ~= 10^19 <= x < 10^20 ~= 2^66.43... |
| 2201 |
// so x rounded to an integer may or may not fit in a signed 64-bit int |
| 2202 |
// the cases that do not fit are identified here; the ones that fit |
| 2203 |
// fall through and will be handled with other cases further, |
| 2204 |
// under '1 <= q + exp <= 19' |
| 2205 |
if (x_sign) { // if n < 0 and q + exp = 19 |
| 2206 |
// if n <= -2^63 - 1 then n is too large |
| 2207 |
// too large if c(0)c(1)...c(18).c(19)...c(q-1) >= 2^63+1 |
| 2208 |
// <=> 0.c(0)c(1)...c(q-1) * 10^20 >= 5*(2^64+2), 1<=q<=34 |
| 2209 |
// <=> 0.c(0)c(1)...c(q-1) * 10^20 >= 0x5000000000000000a, 1<=q<=34 |
| 2210 |
C.w[1] = 0x0000000000000005ull; |
| 2211 |
C.w[0] = 0x000000000000000aull; |
| 2212 |
if (q <= 19) { // 1 <= q <= 19 => 1 <= 20-q <= 19 => |
| 2213 |
// 10^(20-q) is 64-bit, and so is C1 |
| 2214 |
__mul_64x64_to_128MACH (C1, C1.w[0], ten2k64[20 - q]); |
| 2215 |
} else if (q == 20) { |
| 2216 |
; // C1 * 10^0 = C1 |
| 2217 |
} else { // if 21 <= q <= 34 |
| 2218 |
__mul_128x64_to_128 (C, ten2k64[q - 20], C); // max 47-bit x 67-bit |
| 2219 |
} |
| 2220 |
if (C1.w[1] > C.w[1] || (C1.w[1] == C.w[1] && C1.w[0] >= C.w[0])) { |
| 2221 |
// set invalid flag |
| 2222 |
*pfpsf |= INVALID_EXCEPTION; |
| 2223 |
// return Integer Indefinite |
| 2224 |
res = 0x8000000000000000ull; |
| 2225 |
BID_RETURN (res); |
| 2226 |
} |
| 2227 |
// else cases that can be rounded to a 64-bit int fall through |
| 2228 |
// to '1 <= q + exp <= 19' |
| 2229 |
} else { // if n > 0 and q + exp = 19 |
| 2230 |
// if n >= 2^63 then n is too large |
| 2231 |
// too large if c(0)c(1)...c(18).c(19)...c(q-1) >= 2^63 |
| 2232 |
// <=> if 0.c(0)c(1)...c(q-1) * 10^20 >= 5*2^64, 1<=q<=34 |
| 2233 |
// <=> if 0.c(0)c(1)...c(q-1) * 10^20 >= 0x50000000000000000, 1<=q<=34 |
| 2234 |
C.w[1] = 0x0000000000000005ull; |
| 2235 |
C.w[0] = 0x0000000000000000ull; |
| 2236 |
if (q <= 19) { // 1 <= q <= 19 => 1 <= 20-q <= 19 => |
| 2237 |
// 10^(20-q) is 64-bit, and so is C1 |
| 2238 |
__mul_64x64_to_128MACH (C1, C1.w[0], ten2k64[20 - q]); |
| 2239 |
} else if (q == 20) { |
| 2240 |
; // C1 * 10^0 = C1 |
| 2241 |
} else { // if 21 <= q <= 34 |
| 2242 |
__mul_128x64_to_128 (C, ten2k64[q - 20], C); // max 47-bit x 67-bit |
| 2243 |
} |
| 2244 |
if (C1.w[1] > C.w[1] || (C1.w[1] == C.w[1] && C1.w[0] >= C.w[0])) { |
| 2245 |
// set invalid flag |
| 2246 |
*pfpsf |= INVALID_EXCEPTION; |
| 2247 |
// return Integer Indefinite |
| 2248 |
res = 0x8000000000000000ull; |
| 2249 |
BID_RETURN (res); |
| 2250 |
} |
| 2251 |
// else cases that can be rounded to a 64-bit int fall through |
| 2252 |
// to '1 <= q + exp <= 19' |
| 2253 |
} |
| 2254 |
} |
| 2255 |
// n is not too large to be converted to int64: -2^63-1 < n < 2^63 |
| 2256 |
// Note: some of the cases tested for above fall through to this point |
| 2257 |
// Restore C1 which may have been modified above |
| 2258 |
C1.w[1] = x.w[1] & MASK_COEFF; |
| 2259 |
C1.w[0] = x.w[0]; |
| 2260 |
if ((q + exp) <= 0) { // n = +/-0.[0...0]c(0)c(1)...c(q-1) |
| 2261 |
// set inexact flag |
| 2262 |
*pfpsf |= INEXACT_EXCEPTION; |
| 2263 |
// return 0 |
| 2264 |
res = 0x0000000000000000ull; |
| 2265 |
BID_RETURN (res); |
| 2266 |
} else { // if (1 <= q + exp <= 19, 1 <= q <= 34, -33 <= exp <= 18) |
| 2267 |
// -2^63-1 < x <= -1 or 1 <= x < 2^63 so x can be rounded |
| 2268 |
// toward zero to a 64-bit signed integer |
| 2269 |
if (exp < 0) { // 2 <= q <= 34, -33 <= exp <= -1, 1 <= q + exp <= 19 |
| 2270 |
ind = -exp; // 1 <= ind <= 33; ind is a synonym for 'x' |
| 2271 |
// chop off ind digits from the lower part of C1 |
| 2272 |
// C1 fits in 127 bits |
| 2273 |
// calculate C* and f* |
| 2274 |
// C* is actually floor(C*) in this case |
| 2275 |
// C* and f* need shifting and masking, as shown by |
| 2276 |
// shiftright128[] and maskhigh128[] |
| 2277 |
// 1 <= x <= 33 |
| 2278 |
// kx = 10^(-x) = ten2mk128[ind - 1] |
| 2279 |
// C* = C1 * 10^(-x) |
| 2280 |
// the approximation of 10^(-x) was rounded up to 118 bits |
| 2281 |
__mul_128x128_to_256 (P256, C1, ten2mk128[ind - 1]); |
| 2282 |
if (ind - 1 <= 21) { // 0 <= ind - 1 <= 21 |
| 2283 |
Cstar.w[1] = P256.w[3]; |
| 2284 |
Cstar.w[0] = P256.w[2]; |
| 2285 |
fstar.w[3] = 0; |
| 2286 |
fstar.w[2] = P256.w[2] & maskhigh128[ind - 1]; |
| 2287 |
fstar.w[1] = P256.w[1]; |
| 2288 |
fstar.w[0] = P256.w[0]; |
| 2289 |
} else { // 22 <= ind - 1 <= 33 |
| 2290 |
Cstar.w[1] = 0; |
| 2291 |
Cstar.w[0] = P256.w[3]; |
| 2292 |
fstar.w[3] = P256.w[3] & maskhigh128[ind - 1]; |
| 2293 |
fstar.w[2] = P256.w[2]; |
| 2294 |
fstar.w[1] = P256.w[1]; |
| 2295 |
fstar.w[0] = P256.w[0]; |
| 2296 |
} |
| 2297 |
// the top Ex bits of 10^(-x) are T* = ten2mk128trunc[ind], e.g. |
| 2298 |
// if x=1, T*=ten2mk128trunc[0]=0x19999999999999999999999999999999 |
| 2299 |
// C* = floor(C*) (logical right shift; C has p decimal digits, |
| 2300 |
// correct by Property 1) |
| 2301 |
// n = C* * 10^(e+x) |
| 2302 |
|
| 2303 |
// shift right C* by Ex-128 = shiftright128[ind] |
| 2304 |
shift = shiftright128[ind - 1]; // 0 <= shift <= 102 |
| 2305 |
if (ind - 1 <= 21) { // 0 <= ind - 1 <= 21 |
| 2306 |
Cstar.w[0] = |
| 2307 |
(Cstar.w[0] >> shift) | (Cstar.w[1] << (64 - shift)); |
| 2308 |
// redundant, it will be 0! Cstar.w[1] = (Cstar.w[1] >> shift); |
| 2309 |
} else { // 22 <= ind - 1 <= 33 |
| 2310 |
Cstar.w[0] = (Cstar.w[0] >> (shift - 64)); // 2 <= shift - 64 <= 38 |
| 2311 |
} |
| 2312 |
// determine inexactness of the rounding of C* |
| 2313 |
// if (0 < f* < 10^(-x)) then |
| 2314 |
// the result is exact |
| 2315 |
// else // if (f* > T*) then |
| 2316 |
// the result is inexact |
| 2317 |
if (ind - 1 <= 2) { |
| 2318 |
if (fstar.w[1] > ten2mk128trunc[ind - 1].w[1] |
| 2319 |
|| (fstar.w[1] == ten2mk128trunc[ind - 1].w[1] |
| 2320 |
&& fstar.w[0] > ten2mk128trunc[ind - 1].w[0])) { |
| 2321 |
// set the inexact flag |
| 2322 |
*pfpsf |= INEXACT_EXCEPTION; |
| 2323 |
} // else the result is exact |
| 2324 |
} else if (ind - 1 <= 21) { // if 3 <= ind <= 21 |
| 2325 |
if (fstar.w[2] || fstar.w[1] > ten2mk128trunc[ind - 1].w[1] |
| 2326 |
|| (fstar.w[1] == ten2mk128trunc[ind - 1].w[1] |
| 2327 |
&& fstar.w[0] > ten2mk128trunc[ind - 1].w[0])) { |
| 2328 |
// set the inexact flag |
| 2329 |
*pfpsf |= INEXACT_EXCEPTION; |
| 2330 |
} |
| 2331 |
} else { // if 22 <= ind <= 33 |
| 2332 |
if (fstar.w[3] || fstar.w[2] |
| 2333 |
|| fstar.w[1] > ten2mk128trunc[ind - 1].w[1] |
| 2334 |
|| (fstar.w[1] == ten2mk128trunc[ind - 1].w[1] |
| 2335 |
&& fstar.w[0] > ten2mk128trunc[ind - 1].w[0])) { |
| 2336 |
// set the inexact flag |
| 2337 |
*pfpsf |= INEXACT_EXCEPTION; |
| 2338 |
} |
| 2339 |
} |
| 2340 |
|
| 2341 |
if (x_sign) |
| 2342 |
res = -Cstar.w[0]; |
| 2343 |
else |
| 2344 |
res = Cstar.w[0]; |
| 2345 |
} else if (exp == 0) { |
| 2346 |
// 1 <= q <= 19 |
| 2347 |
// res = +/-C (exact) |
| 2348 |
if (x_sign) |
| 2349 |
res = -C1.w[0]; |
| 2350 |
else |
| 2351 |
res = C1.w[0]; |
| 2352 |
} else { // if (exp>0) => 1 <= exp <= 18, 1 <= q < 18, 2 <= q + exp <= 19 |
| 2353 |
// res = +/-C * 10^exp (exact) where this fits in 64-bit integer |
| 2354 |
if (x_sign) |
| 2355 |
res = -C1.w[0] * ten2k64[exp]; |
| 2356 |
else |
| 2357 |
res = C1.w[0] * ten2k64[exp]; |
| 2358 |
} |
| 2359 |
} |
| 2360 |
} |
| 2361 |
|
| 2362 |
BID_RETURN (res); |
| 2363 |
} |
| 2364 |
|
| 2365 |
/***************************************************************************** |
| 2366 |
* BID128_to_int64_rninta |
| 2367 |
****************************************************************************/ |
| 2368 |
|
| 2369 |
BID128_FUNCTION_ARG1_NORND_CUSTOMRESTYPE (SINT64, |
| 2370 |
bid128_to_int64_rninta, x) |
| 2371 |
|
| 2372 |
SINT64 res; |
| 2373 |
UINT64 x_sign; |
| 2374 |
UINT64 x_exp; |
| 2375 |
int exp; // unbiased exponent |
| 2376 |
// Note: C1.w[1], C1.w[0] represent x_signif_hi, x_signif_lo (all are UINT64) |
| 2377 |
UINT64 tmp64; |
| 2378 |
BID_UI64DOUBLE tmp1; |
| 2379 |
unsigned int x_nr_bits; |
| 2380 |
int q, ind, shift; |
| 2381 |
UINT128 C1, C; |
| 2382 |
UINT128 Cstar; // C* represents up to 34 decimal digits ~ 113 bits |
| 2383 |
UINT256 P256; |
| 2384 |
|
| 2385 |
// unpack x |
| 2386 |
x_sign = x.w[1] & MASK_SIGN; // 0 for positive, MASK_SIGN for negative |
| 2387 |
x_exp = x.w[1] & MASK_EXP; // biased and shifted left 49 bit positions |
| 2388 |
C1.w[1] = x.w[1] & MASK_COEFF; |
| 2389 |
C1.w[0] = x.w[0]; |
| 2390 |
|
| 2391 |
// check for NaN or Infinity |
| 2392 |
if ((x.w[1] & MASK_SPECIAL) == MASK_SPECIAL) { |
| 2393 |
// x is special |
| 2394 |
if ((x.w[1] & MASK_NAN) == MASK_NAN) { // x is NAN |
| 2395 |
if ((x.w[1] & MASK_SNAN) == MASK_SNAN) { // x is SNAN |
| 2396 |
// set invalid flag |
| 2397 |
*pfpsf |= INVALID_EXCEPTION; |
| 2398 |
// return Integer Indefinite |
| 2399 |
res = 0x8000000000000000ull; |
| 2400 |
} else { // x is QNaN |
| 2401 |
// set invalid flag |
| 2402 |
*pfpsf |= INVALID_EXCEPTION; |
| 2403 |
// return Integer Indefinite |
| 2404 |
res = 0x8000000000000000ull; |
| 2405 |
} |
| 2406 |
BID_RETURN (res); |
| 2407 |
} else { // x is not a NaN, so it must be infinity |
| 2408 |
if (!x_sign) { // x is +inf |
| 2409 |
// set invalid flag |
| 2410 |
*pfpsf |= INVALID_EXCEPTION; |
| 2411 |
// return Integer Indefinite |
| 2412 |
res = 0x8000000000000000ull; |
| 2413 |
} else { // x is -inf |
| 2414 |
// set invalid flag |
| 2415 |
*pfpsf |= INVALID_EXCEPTION; |
| 2416 |
// return Integer Indefinite |
| 2417 |
res = 0x8000000000000000ull; |
| 2418 |
} |
| 2419 |
BID_RETURN (res); |
| 2420 |
} |
| 2421 |
} |
| 2422 |
// check for non-canonical values (after the check for special values) |
| 2423 |
if ((C1.w[1] > 0x0001ed09bead87c0ull) |
| 2424 |
|| (C1.w[1] == 0x0001ed09bead87c0ull |
| 2425 |
&& (C1.w[0] > 0x378d8e63ffffffffull)) |
| 2426 |
|| ((x.w[1] & 0x6000000000000000ull) == 0x6000000000000000ull)) { |
| 2427 |
res = 0x0000000000000000ull; |
| 2428 |
BID_RETURN (res); |
| 2429 |
} else if ((C1.w[1] == 0x0ull) && (C1.w[0] == 0x0ull)) { |
| 2430 |
// x is 0 |
| 2431 |
res = 0x0000000000000000ull; |
| 2432 |
BID_RETURN (res); |
| 2433 |
} else { // x is not special and is not zero |
| 2434 |
|
| 2435 |
// q = nr. of decimal digits in x |
| 2436 |
// determine first the nr. of bits in x |
| 2437 |
if (C1.w[1] == 0) { |
| 2438 |
if (C1.w[0] >= 0x0020000000000000ull) { // x >= 2^53 |
| 2439 |
// split the 64-bit value in two 32-bit halves to avoid rounding errors |
| 2440 |
if (C1.w[0] >= 0x0000000100000000ull) { // x >= 2^32 |
| 2441 |
tmp1.d = (double) (C1.w[0] >> 32); // exact conversion |
| 2442 |
x_nr_bits = |
| 2443 |
33 + ((((unsigned int) (tmp1.ui64 >> 52)) & 0x7ff) - 0x3ff); |
| 2444 |
} else { // x < 2^32 |
| 2445 |
tmp1.d = (double) (C1.w[0]); // exact conversion |
| 2446 |
x_nr_bits = |
| 2447 |
1 + ((((unsigned int) (tmp1.ui64 >> 52)) & 0x7ff) - 0x3ff); |
| 2448 |
} |
| 2449 |
} else { // if x < 2^53 |
| 2450 |
tmp1.d = (double) C1.w[0]; // exact conversion |
| 2451 |
x_nr_bits = |
| 2452 |
1 + ((((unsigned int) (tmp1.ui64 >> 52)) & 0x7ff) - 0x3ff); |
| 2453 |
} |
| 2454 |
} else { // C1.w[1] != 0 => nr. bits = 64 + nr_bits (C1.w[1]) |
| 2455 |
tmp1.d = (double) C1.w[1]; // exact conversion |
| 2456 |
x_nr_bits = |
| 2457 |
65 + ((((unsigned int) (tmp1.ui64 >> 52)) & 0x7ff) - 0x3ff); |
| 2458 |
} |
| 2459 |
q = nr_digits[x_nr_bits - 1].digits; |
| 2460 |
if (q == 0) { |
| 2461 |
q = nr_digits[x_nr_bits - 1].digits1; |
| 2462 |
if (C1.w[1] > nr_digits[x_nr_bits - 1].threshold_hi |
| 2463 |
|| (C1.w[1] == nr_digits[x_nr_bits - 1].threshold_hi |
| 2464 |
&& C1.w[0] >= nr_digits[x_nr_bits - 1].threshold_lo)) |
| 2465 |
q++; |
| 2466 |
} |
| 2467 |
exp = (x_exp >> 49) - 6176; |
| 2468 |
if ((q + exp) > 19) { // x >= 10^19 ~= 2^63.11... (cannot fit in SINT64) |
| 2469 |
// set invalid flag |
| 2470 |
*pfpsf |= INVALID_EXCEPTION; |
| 2471 |
// return Integer Indefinite |
| 2472 |
res = 0x8000000000000000ull; |
| 2473 |
BID_RETURN (res); |
| 2474 |
} else if ((q + exp) == 19) { // x = c(0)c(1)...c(18).c(19)...c(q-1) |
| 2475 |
// in this case 2^63.11... ~= 10^19 <= x < 10^20 ~= 2^66.43... |
| 2476 |
// so x rounded to an integer may or may not fit in a signed 64-bit int |
| 2477 |
// the cases that do not fit are identified here; the ones that fit |
| 2478 |
// fall through and will be handled with other cases further, |
| 2479 |
// under '1 <= q + exp <= 19' |
| 2480 |
if (x_sign) { // if n < 0 and q + exp = 19 |
| 2481 |
// if n <= -2^63 - 1/2 then n is too large |
| 2482 |
// too large if c(0)c(1)...c(18).c(19)...c(q-1) >= 2^63+1/2 |
| 2483 |
// <=> 0.c(0)c(1)...c(q-1) * 10^20 >= 5*(2^64+1), 1<=q<=34 |
| 2484 |
// <=> 0.c(0)c(1)...c(q-1) * 10^20 >= 0x50000000000000005, 1<=q<=34 |
| 2485 |
C.w[1] = 0x0000000000000005ull; |
| 2486 |
C.w[0] = 0000000000000005ull; |
| 2487 |
if (q <= 19) { // 1 <= q <= 19 => 1 <= 20-q <= 19 => |
| 2488 |
// 10^(20-q) is 64-bit, and so is C1 |
| 2489 |
__mul_64x64_to_128MACH (C1, C1.w[0], ten2k64[20 - q]); |
| 2490 |
} else if (q == 20) { |
| 2491 |
; // C1 * 10^0 = C1 |
| 2492 |
} else { // if 21 <= q <= 34 |
| 2493 |
__mul_128x64_to_128 (C, ten2k64[q - 20], C); // max 47-bit x 67-bit |
| 2494 |
} |
| 2495 |
if (C1.w[1] > C.w[1] || (C1.w[1] == C.w[1] && C1.w[0] >= C.w[0])) { |
| 2496 |
// set invalid flag |
| 2497 |
*pfpsf |= INVALID_EXCEPTION; |
| 2498 |
// return Integer Indefinite |
| 2499 |
res = 0x8000000000000000ull; |
| 2500 |
BID_RETURN (res); |
| 2501 |
} |
| 2502 |
// else cases that can be rounded to a 64-bit int fall through |
| 2503 |
// to '1 <= q + exp <= 19' |
| 2504 |
} else { // if n > 0 and q + exp = 19 |
| 2505 |
// if n >= 2^63 - 1/2 then n is too large |
| 2506 |
// too large if c(0)c(1)...c(18).c(19)...c(q-1) >= 2^63-1/2 |
| 2507 |
// <=> if 0.c(0)c(1)...c(q-1) * 10^20 >= 5*(2^64-1), 1<=q<=34 |
| 2508 |
// <=> if 0.c(0)c(1)...c(q-1) * 10^20 >= 0x4fffffffffffffffb, 1<=q<=34 |
| 2509 |
C.w[1] = 0x0000000000000004ull; |
| 2510 |
C.w[0] = 0xfffffffffffffffbull; |
| 2511 |
if (q <= 19) { // 1 <= q <= 19 => 1 <= 20-q <= 19 => |
| 2512 |
// 10^(20-q) is 64-bit, and so is C1 |
| 2513 |
__mul_64x64_to_128MACH (C1, C1.w[0], ten2k64[20 - q]); |
| 2514 |
} else if (q == 20) { |
| 2515 |
; // C1 * 10^0 = C1 |
| 2516 |
} else { // if 21 <= q <= 34 |
| 2517 |
__mul_128x64_to_128 (C, ten2k64[q - 20], C); // max 47-bit x 67-bit |
| 2518 |
} |
| 2519 |
if (C1.w[1] > C.w[1] || (C1.w[1] == C.w[1] && C1.w[0] >= C.w[0])) { |
| 2520 |
// set invalid flag |
| 2521 |
*pfpsf |= INVALID_EXCEPTION; |
| 2522 |
// return Integer Indefinite |
| 2523 |
res = 0x8000000000000000ull; |
| 2524 |
BID_RETURN (res); |
| 2525 |
} |
| 2526 |
// else cases that can be rounded to a 64-bit int fall through |
| 2527 |
// to '1 <= q + exp <= 19' |
| 2528 |
} |
| 2529 |
} |
| 2530 |
// n is not too large to be converted to int64: -2^63-1/2 <= n < 2^63-1/2 |
| 2531 |
// Note: some of the cases tested for above fall through to this point |
| 2532 |
// Restore C1 which may have been modified above |
| 2533 |
C1.w[1] = x.w[1] & MASK_COEFF; |
| 2534 |
C1.w[0] = x.w[0]; |
| 2535 |
if ((q + exp) < 0) { // n = +/-0.0...c(0)c(1)...c(q-1) |
| 2536 |
// return 0 |
| 2537 |
res = 0x0000000000000000ull; |
| 2538 |
BID_RETURN (res); |
| 2539 |
} else if ((q + exp) == 0) { // n = +/-0.c(0)c(1)...c(q-1) |
| 2540 |
// if 0.c(0)c(1)...c(q-1) < 0.5 <=> c(0)c(1)...c(q-1) < 5 * 10^(q-1) |
| 2541 |
// res = 0 |
| 2542 |
// else |
| 2543 |
// res = +/-1 |
| 2544 |
ind = q - 1; |
| 2545 |
if (ind <= 18) { // 0 <= ind <= 18 |
| 2546 |
if ((C1.w[1] == 0) && (C1.w[0] < midpoint64[ind])) { |
| 2547 |
res = 0x0000000000000000ull; // return 0 |
| 2548 |
} else if (x_sign) { // n < 0 |
| 2549 |
res = 0xffffffffffffffffull; // return -1 |
| 2550 |
} else { // n > 0 |
| 2551 |
res = 0x0000000000000001ull; // return +1 |
| 2552 |
} |
| 2553 |
} else { // 19 <= ind <= 33 |
| 2554 |
if ((C1.w[1] < midpoint128[ind - 19].w[1]) |
| 2555 |
|| ((C1.w[1] == midpoint128[ind - 19].w[1]) |
| 2556 |
&& (C1.w[0] < midpoint128[ind - 19].w[0]))) { |
| 2557 |
res = 0x0000000000000000ull; // return 0 |
| 2558 |
} else if (x_sign) { // n < 0 |
| 2559 |
res = 0xffffffffffffffffull; // return -1 |
| 2560 |
} else { // n > 0 |
| 2561 |
res = 0x0000000000000001ull; // return +1 |
| 2562 |
} |
| 2563 |
} |
| 2564 |
} else { // if (1 <= q + exp <= 19, 1 <= q <= 34, -33 <= exp <= 18) |
| 2565 |
// -2^63-1/2 <= x <= -1 or 1 <= x < 2^63-1/2 so x can be rounded |
| 2566 |
// to nearest to a 64-bit signed integer |
| 2567 |
if (exp < 0) { // 2 <= q <= 34, -33 <= exp <= -1, 1 <= q + exp <= 19 |
| 2568 |
ind = -exp; // 1 <= ind <= 33; ind is a synonym for 'x' |
| 2569 |
// chop off ind digits from the lower part of C1 |
| 2570 |
// C1 = C1 + 1/2 * 10^ind where the result C1 fits in 127 bits |
| 2571 |
tmp64 = C1.w[0]; |
| 2572 |
if (ind <= 19) { |
| 2573 |
C1.w[0] = C1.w[0] + midpoint64[ind - 1]; |
| 2574 |
} else { |
| 2575 |
C1.w[0] = C1.w[0] + midpoint128[ind - 20].w[0]; |
| 2576 |
C1.w[1] = C1.w[1] + midpoint128[ind - 20].w[1]; |
| 2577 |
} |
| 2578 |
if (C1.w[0] < tmp64) |
| 2579 |
C1.w[1]++; |
| 2580 |
// calculate C* and f* |
| 2581 |
// C* is actually floor(C*) in this case |
| 2582 |
// C* and f* need shifting and masking, as shown by |
| 2583 |
// shiftright128[] and maskhigh128[] |
| 2584 |
// 1 <= x <= 33 |
| 2585 |
// kx = 10^(-x) = ten2mk128[ind - 1] |
| 2586 |
// C* = (C1 + 1/2 * 10^x) * 10^(-x) |
| 2587 |
// the approximation of 10^(-x) was rounded up to 118 bits |
| 2588 |
__mul_128x128_to_256 (P256, C1, ten2mk128[ind - 1]); |
| 2589 |
if (ind - 1 <= 21) { // 0 <= ind - 1 <= 21 |
| 2590 |
Cstar.w[1] = P256.w[3]; |
| 2591 |
Cstar.w[0] = P256.w[2]; |
| 2592 |
} else { // 22 <= ind - 1 <= 33 |
| 2593 |
Cstar.w[1] = 0; |
| 2594 |
Cstar.w[0] = P256.w[3]; |
| 2595 |
} |
| 2596 |
// the top Ex bits of 10^(-x) are T* = ten2mk128trunc[ind], e.g. |
| 2597 |
// if x=1, T*=ten2mk128trunc[0]=0x19999999999999999999999999999999 |
| 2598 |
// if (0 < f* < 10^(-x)) then the result is a midpoint |
| 2599 |
// if floor(C*) is even then C* = floor(C*) - logical right |
| 2600 |
// shift; C* has p decimal digits, correct by Prop. 1) |
| 2601 |
// else if floor(C*) is odd C* = floor(C*)-1 (logical right |
| 2602 |
// shift; C* has p decimal digits, correct by Pr. 1) |
| 2603 |
// else |
| 2604 |
// C* = floor(C*) (logical right shift; C has p decimal digits, |
| 2605 |
// correct by Property 1) |
| 2606 |
// n = C* * 10^(e+x) |
| 2607 |
|
| 2608 |
// shift right C* by Ex-128 = shiftright128[ind] |
| 2609 |
shift = shiftright128[ind - 1]; // 0 <= shift <= 102 |
| 2610 |
if (ind - 1 <= 21) { // 0 <= ind - 1 <= 21 |
| 2611 |
Cstar.w[0] = |
| 2612 |
(Cstar.w[0] >> shift) | (Cstar.w[1] << (64 - shift)); |
| 2613 |
// redundant, it will be 0! Cstar.w[1] = (Cstar.w[1] >> shift); |
| 2614 |
} else { // 22 <= ind - 1 <= 33 |
| 2615 |
Cstar.w[0] = (Cstar.w[0] >> (shift - 64)); // 2 <= shift - 64 <= 38 |
| 2616 |
} |
| 2617 |
|
| 2618 |
// if the result was a midpoint it was rounded away from zero |
| 2619 |
if (x_sign) |
| 2620 |
res = -Cstar.w[0]; |
| 2621 |
else |
| 2622 |
res = Cstar.w[0]; |
| 2623 |
} else if (exp == 0) { |
| 2624 |
// 1 <= q <= 19 |
| 2625 |
// res = +/-C (exact) |
| 2626 |
if (x_sign) |
| 2627 |
res = -C1.w[0]; |
| 2628 |
else |
| 2629 |
res = C1.w[0]; |
| 2630 |
} else { // if (exp>0) => 1 <= exp <= 18, 1 <= q < 18, 2 <= q + exp <= 19 |
| 2631 |
// res = +/-C * 10^exp (exact) where this fits in 64-bit integer |
| 2632 |
if (x_sign) |
| 2633 |
res = -C1.w[0] * ten2k64[exp]; |
| 2634 |
else |
| 2635 |
res = C1.w[0] * ten2k64[exp]; |
| 2636 |
} |
| 2637 |
} |
| 2638 |
} |
| 2639 |
|
| 2640 |
BID_RETURN (res); |
| 2641 |
} |
| 2642 |
|
| 2643 |
/***************************************************************************** |
| 2644 |
* BID128_to_int64_xrninta |
| 2645 |
****************************************************************************/ |
| 2646 |
|
| 2647 |
BID128_FUNCTION_ARG1_NORND_CUSTOMRESTYPE (SINT64, |
| 2648 |
bid128_to_int64_xrninta, x) |
| 2649 |
|
| 2650 |
SINT64 res; |
| 2651 |
UINT64 x_sign; |
| 2652 |
UINT64 x_exp; |
| 2653 |
int exp; // unbiased exponent |
| 2654 |
// Note: C1.w[1], C1.w[0] represent x_signif_hi, x_signif_lo (all are UINT64) |
| 2655 |
UINT64 tmp64, tmp64A; |
| 2656 |
BID_UI64DOUBLE tmp1; |
| 2657 |
unsigned int x_nr_bits; |
| 2658 |
int q, ind, shift; |
| 2659 |
UINT128 C1, C; |
| 2660 |
UINT128 Cstar; // C* represents up to 34 decimal digits ~ 113 bits |
| 2661 |
UINT256 fstar; |
| 2662 |
UINT256 P256; |
| 2663 |
|
| 2664 |
// unpack x |
| 2665 |
x_sign = x.w[1] & MASK_SIGN; // 0 for positive, MASK_SIGN for negative |
| 2666 |
x_exp = x.w[1] & MASK_EXP; // biased and shifted left 49 bit positions |
| 2667 |
C1.w[1] = x.w[1] & MASK_COEFF; |
| 2668 |
C1.w[0] = x.w[0]; |
| 2669 |
|
| 2670 |
// check for NaN or Infinity |
| 2671 |
if ((x.w[1] & MASK_SPECIAL) == MASK_SPECIAL) { |
| 2672 |
// x is special |
| 2673 |
if ((x.w[1] & MASK_NAN) == MASK_NAN) { // x is NAN |
| 2674 |
if ((x.w[1] & MASK_SNAN) == MASK_SNAN) { // x is SNAN |
| 2675 |
// set invalid flag |
| 2676 |
*pfpsf |= INVALID_EXCEPTION; |
| 2677 |
// return Integer Indefinite |
| 2678 |
res = 0x8000000000000000ull; |
| 2679 |
} else { // x is QNaN |
| 2680 |
// set invalid flag |
| 2681 |
*pfpsf |= INVALID_EXCEPTION; |
| 2682 |
// return Integer Indefinite |
| 2683 |
res = 0x8000000000000000ull; |
| 2684 |
} |
| 2685 |
BID_RETURN (res); |
| 2686 |
} else { // x is not a NaN, so it must be infinity |
| 2687 |
if (!x_sign) { // x is +inf |
| 2688 |
// set invalid flag |
| 2689 |
*pfpsf |= INVALID_EXCEPTION; |
| 2690 |
// return Integer Indefinite |
| 2691 |
res = 0x8000000000000000ull; |
| 2692 |
} else { // x is -inf |
| 2693 |
// set invalid flag |
| 2694 |
*pfpsf |= INVALID_EXCEPTION; |
| 2695 |
// return Integer Indefinite |
| 2696 |
res = 0x8000000000000000ull; |
| 2697 |
} |
| 2698 |
BID_RETURN (res); |
| 2699 |
} |
| 2700 |
} |
| 2701 |
// check for non-canonical values (after the check for special values) |
| 2702 |
if ((C1.w[1] > 0x0001ed09bead87c0ull) |
| 2703 |
|| (C1.w[1] == 0x0001ed09bead87c0ull |
| 2704 |
&& (C1.w[0] > 0x378d8e63ffffffffull)) |
| 2705 |
|| ((x.w[1] & 0x6000000000000000ull) == 0x6000000000000000ull)) { |
| 2706 |
res = 0x0000000000000000ull; |
| 2707 |
BID_RETURN (res); |
| 2708 |
} else if ((C1.w[1] == 0x0ull) && (C1.w[0] == 0x0ull)) { |
| 2709 |
// x is 0 |
| 2710 |
res = 0x0000000000000000ull; |
| 2711 |
BID_RETURN (res); |
| 2712 |
} else { // x is not special and is not zero |
| 2713 |
|
| 2714 |
// q = nr. of decimal digits in x |
| 2715 |
// determine first the nr. of bits in x |
| 2716 |
if (C1.w[1] == 0) { |
| 2717 |
if (C1.w[0] >= 0x0020000000000000ull) { // x >= 2^53 |
| 2718 |
// split the 64-bit value in two 32-bit halves to avoid rounding errors |
| 2719 |
if (C1.w[0] >= 0x0000000100000000ull) { // x >= 2^32 |
| 2720 |
tmp1.d = (double) (C1.w[0] >> 32); // exact conversion |
| 2721 |
x_nr_bits = |
| 2722 |
33 + ((((unsigned int) (tmp1.ui64 >> 52)) & 0x7ff) - 0x3ff); |
| 2723 |
} else { // x < 2^32 |
| 2724 |
tmp1.d = (double) (C1.w[0]); // exact conversion |
| 2725 |
x_nr_bits = |
| 2726 |
1 + ((((unsigned int) (tmp1.ui64 >> 52)) & 0x7ff) - 0x3ff); |
| 2727 |
} |
| 2728 |
} else { // if x < 2^53 |
| 2729 |
tmp1.d = (double) C1.w[0]; // exact conversion |
| 2730 |
x_nr_bits = |
| 2731 |
1 + ((((unsigned int) (tmp1.ui64 >> 52)) & 0x7ff) - 0x3ff); |
| 2732 |
} |
| 2733 |
} else { // C1.w[1] != 0 => nr. bits = 64 + nr_bits (C1.w[1]) |
| 2734 |
tmp1.d = (double) C1.w[1]; // exact conversion |
| 2735 |
x_nr_bits = |
| 2736 |
65 + ((((unsigned int) (tmp1.ui64 >> 52)) & 0x7ff) - 0x3ff); |
| 2737 |
} |
| 2738 |
q = nr_digits[x_nr_bits - 1].digits; |
| 2739 |
if (q == 0) { |
| 2740 |
q = nr_digits[x_nr_bits - 1].digits1; |
| 2741 |
if (C1.w[1] > nr_digits[x_nr_bits - 1].threshold_hi |
| 2742 |
|| (C1.w[1] == nr_digits[x_nr_bits - 1].threshold_hi |
| 2743 |
&& C1.w[0] >= nr_digits[x_nr_bits - 1].threshold_lo)) |
| 2744 |
q++; |
| 2745 |
} |
| 2746 |
exp = (x_exp >> 49) - 6176; |
| 2747 |
if ((q + exp) > 19) { // x >= 10^19 ~= 2^63.11... (cannot fit in SINT64) |
| 2748 |
// set invalid flag |
| 2749 |
*pfpsf |= INVALID_EXCEPTION; |
| 2750 |
// return Integer Indefinite |
| 2751 |
res = 0x8000000000000000ull; |
| 2752 |
BID_RETURN (res); |
| 2753 |
} else if ((q + exp) == 19) { // x = c(0)c(1)...c(18).c(19)...c(q-1) |
| 2754 |
// in this case 2^63.11... ~= 10^19 <= x < 10^20 ~= 2^66.43... |
| 2755 |
// so x rounded to an integer may or may not fit in a signed 64-bit int |
| 2756 |
// the cases that do not fit are identified here; the ones that fit |
| 2757 |
// fall through and will be handled with other cases further, |
| 2758 |
// under '1 <= q + exp <= 19' |
| 2759 |
if (x_sign) { // if n < 0 and q + exp = 19 |
| 2760 |
// if n <= -2^63 - 1/2 then n is too large |
| 2761 |
// too large if c(0)c(1)...c(18).c(19)...c(q-1) >= 2^63+1/2 |
| 2762 |
// <=> 0.c(0)c(1)...c(q-1) * 10^20 >= 5*(2^64+1), 1<=q<=34 |
| 2763 |
// <=> 0.c(0)c(1)...c(q-1) * 10^20 >= 0x50000000000000005, 1<=q<=34 |
| 2764 |
C.w[1] = 0x0000000000000005ull; |
| 2765 |
C.w[0] = 0000000000000005ull; |
| 2766 |
if (q <= 19) { // 1 <= q <= 19 => 1 <= 20-q <= 19 => |
| 2767 |
// 10^(20-q) is 64-bit, and so is C1 |
| 2768 |
__mul_64x64_to_128MACH (C1, C1.w[0], ten2k64[20 - q]); |
| 2769 |
} else if (q == 20) { |
| 2770 |
; // C1 * 10^0 = C1 |
| 2771 |
} else { // if 21 <= q <= 34 |
| 2772 |
__mul_128x64_to_128 (C, ten2k64[q - 20], C); // max 47-bit x 67-bit |
| 2773 |
} |
| 2774 |
if (C1.w[1] > C.w[1] || (C1.w[1] == C.w[1] && C1.w[0] >= C.w[0])) { |
| 2775 |
// set invalid flag |
| 2776 |
*pfpsf |= INVALID_EXCEPTION; |
| 2777 |
// return Integer Indefinite |
| 2778 |
res = 0x8000000000000000ull; |
| 2779 |
BID_RETURN (res); |
| 2780 |
} |
| 2781 |
// else cases that can be rounded to a 64-bit int fall through |
| 2782 |
// to '1 <= q + exp <= 19' |
| 2783 |
} else { // if n > 0 and q + exp = 19 |
| 2784 |
// if n >= 2^63 - 1/2 then n is too large |
| 2785 |
// too large if c(0)c(1)...c(18).c(19)...c(q-1) >= 2^63-1/2 |
| 2786 |
// <=> if 0.c(0)c(1)...c(q-1) * 10^20 >= 5*(2^64-1), 1<=q<=34 |
| 2787 |
// <=> if 0.c(0)c(1)...c(q-1) * 10^20 >= 0x4fffffffffffffffb, 1<=q<=34 |
| 2788 |
C.w[1] = 0x0000000000000004ull; |
| 2789 |
C.w[0] = 0xfffffffffffffffbull; |
| 2790 |
if (q <= 19) { // 1 <= q <= 19 => 1 <= 20-q <= 19 => |
| 2791 |
// 10^(20-q) is 64-bit, and so is C1 |
| 2792 |
__mul_64x64_to_128MACH (C1, C1.w[0], ten2k64[20 - q]); |
| 2793 |
} else if (q == 20) { |
| 2794 |
; // C1 * 10^0 = C1 |
| 2795 |
} else { // if 21 <= q <= 34 |
| 2796 |
__mul_128x64_to_128 (C, ten2k64[q - 20], C); // max 47-bit x 67-bit |
| 2797 |
} |
| 2798 |
if (C1.w[1] > C.w[1] || (C1.w[1] == C.w[1] && C1.w[0] >= C.w[0])) { |
| 2799 |
// set invalid flag |
| 2800 |
*pfpsf |= INVALID_EXCEPTION; |
| 2801 |
// return Integer Indefinite |
| 2802 |
res = 0x8000000000000000ull; |
| 2803 |
BID_RETURN (res); |
| 2804 |
} |
| 2805 |
// else cases that can be rounded to a 64-bit int fall through |
| 2806 |
// to '1 <= q + exp <= 19' |
| 2807 |
} |
| 2808 |
} |
| 2809 |
// n is not too large to be converted to int64: -2^63-1/2 <= n < 2^63-1/2 |
| 2810 |
// Note: some of the cases tested for above fall through to this point |
| 2811 |
// Restore C1 which may have been modified above |
| 2812 |
C1.w[1] = x.w[1] & MASK_COEFF; |
| 2813 |
C1.w[0] = x.w[0]; |
| 2814 |
if ((q + exp) < 0) { // n = +/-0.0...c(0)c(1)...c(q-1) |
| 2815 |
// set inexact flag |
| 2816 |
*pfpsf |= INEXACT_EXCEPTION; |
| 2817 |
// return 0 |
| 2818 |
res = 0x0000000000000000ull; |
| 2819 |
BID_RETURN (res); |
| 2820 |
} else if ((q + exp) == 0) { // n = +/-0.c(0)c(1)...c(q-1) |
| 2821 |
// if 0.c(0)c(1)...c(q-1) < 0.5 <=> c(0)c(1)...c(q-1) < 5 * 10^(q-1) |
| 2822 |
// res = 0 |
| 2823 |
// else |
| 2824 |
// res = +/-1 |
| 2825 |
ind = q - 1; |
| 2826 |
if (ind <= 18) { // 0 <= ind <= 18 |
| 2827 |
if ((C1.w[1] == 0) && (C1.w[0] < midpoint64[ind])) { |
| 2828 |
res = 0x0000000000000000ull; // return 0 |
| 2829 |
} else if (x_sign) { // n < 0 |
| 2830 |
res = 0xffffffffffffffffull; // return -1 |
| 2831 |
} else { // n > 0 |
| 2832 |
res = 0x0000000000000001ull; // return +1 |
| 2833 |
} |
| 2834 |
} else { // 19 <= ind <= 33 |
| 2835 |
if ((C1.w[1] < midpoint128[ind - 19].w[1]) |
| 2836 |
|| ((C1.w[1] == midpoint128[ind - 19].w[1]) |
| 2837 |
&& (C1.w[0] < midpoint128[ind - 19].w[0]))) { |
| 2838 |
res = 0x0000000000000000ull; // return 0 |
| 2839 |
} else if (x_sign) { // n < 0 |
| 2840 |
res = 0xffffffffffffffffull; // return -1 |
| 2841 |
} else { // n > 0 |
| 2842 |
res = 0x0000000000000001ull; // return +1 |
| 2843 |
} |
| 2844 |
} |
| 2845 |
// set inexact flag |
| 2846 |
*pfpsf |= INEXACT_EXCEPTION; |
| 2847 |
} else { // if (1 <= q + exp <= 19, 1 <= q <= 34, -33 <= exp <= 18) |
| 2848 |
// -2^63-1/2 <= x <= -1 or 1 <= x < 2^63-1/2 so x can be rounded |
| 2849 |
// to nearest to a 64-bit signed integer |
| 2850 |
if (exp < 0) { // 2 <= q <= 34, -33 <= exp <= -1, 1 <= q + exp <= 19 |
| 2851 |
ind = -exp; // 1 <= ind <= 33; ind is a synonym for 'x' |
| 2852 |
// chop off ind digits from the lower part of C1 |
| 2853 |
// C1 = C1 + 1/2 * 10^ind where the result C1 fits in 127 bits |
| 2854 |
tmp64 = C1.w[0]; |
| 2855 |
if (ind <= 19) { |
| 2856 |
C1.w[0] = C1.w[0] + midpoint64[ind - 1]; |
| 2857 |
} else { |
| 2858 |
C1.w[0] = C1.w[0] + midpoint128[ind - 20].w[0]; |
| 2859 |
C1.w[1] = C1.w[1] + midpoint128[ind - 20].w[1]; |
| 2860 |
} |
| 2861 |
if (C1.w[0] < tmp64) |
| 2862 |
C1.w[1]++; |
| 2863 |
// calculate C* and f* |
| 2864 |
// C* is actually floor(C*) in this case |
| 2865 |
// C* and f* need shifting and masking, as shown by |
| 2866 |
// shiftright128[] and maskhigh128[] |
| 2867 |
// 1 <= x <= 33 |
| 2868 |
// kx = 10^(-x) = ten2mk128[ind - 1] |
| 2869 |
// C* = (C1 + 1/2 * 10^x) * 10^(-x) |
| 2870 |
// the approximation of 10^(-x) was rounded up to 118 bits |
| 2871 |
__mul_128x128_to_256 (P256, C1, ten2mk128[ind - 1]); |
| 2872 |
if (ind - 1 <= 21) { // 0 <= ind - 1 <= 21 |
| 2873 |
Cstar.w[1] = P256.w[3]; |
| 2874 |
Cstar.w[0] = P256.w[2]; |
| 2875 |
fstar.w[3] = 0; |
| 2876 |
fstar.w[2] = P256.w[2] & maskhigh128[ind - 1]; |
| 2877 |
fstar.w[1] = P256.w[1]; |
| 2878 |
fstar.w[0] = P256.w[0]; |
| 2879 |
} else { // 22 <= ind - 1 <= 33 |
| 2880 |
Cstar.w[1] = 0; |
| 2881 |
Cstar.w[0] = P256.w[3]; |
| 2882 |
fstar.w[3] = P256.w[3] & maskhigh128[ind - 1]; |
| 2883 |
fstar.w[2] = P256.w[2]; |
| 2884 |
fstar.w[1] = P256.w[1]; |
| 2885 |
fstar.w[0] = P256.w[0]; |
| 2886 |
} |
| 2887 |
// the top Ex bits of 10^(-x) are T* = ten2mk128trunc[ind], e.g. |
| 2888 |
// if x=1, T*=ten2mk128trunc[0]=0x19999999999999999999999999999999 |
| 2889 |
// if (0 < f* < 10^(-x)) then the result is a midpoint |
| 2890 |
// if floor(C*) is even then C* = floor(C*) - logical right |
| 2891 |
// shift; C* has p decimal digits, correct by Prop. 1) |
| 2892 |
// else if floor(C*) is odd C* = floor(C*)-1 (logical right |
| 2893 |
// shift; C* has p decimal digits, correct by Pr. 1) |
| 2894 |
// else |
| 2895 |
// C* = floor(C*) (logical right shift; C has p decimal digits, |
| 2896 |
// correct by Property 1) |
| 2897 |
// n = C* * 10^(e+x) |
| 2898 |
|
| 2899 |
// shift right C* by Ex-128 = shiftright128[ind] |
| 2900 |
shift = shiftright128[ind - 1]; // 0 <= shift <= 102 |
| 2901 |
if (ind - 1 <= 21) { // 0 <= ind - 1 <= 21 |
| 2902 |
Cstar.w[0] = |
| 2903 |
(Cstar.w[0] >> shift) | (Cstar.w[1] << (64 - shift)); |
| 2904 |
// redundant, it will be 0! Cstar.w[1] = (Cstar.w[1] >> shift); |
| 2905 |
} else { // 22 <= ind - 1 <= 33 |
| 2906 |
Cstar.w[0] = (Cstar.w[0] >> (shift - 64)); // 2 <= shift - 64 <= 38 |
| 2907 |
} |
| 2908 |
// determine inexactness of the rounding of C* |
| 2909 |
// if (0 < f* - 1/2 < 10^(-x)) then |
| 2910 |
// the result is exact |
| 2911 |
// else // if (f* - 1/2 > T*) then |
| 2912 |
// the result is inexact |
| 2913 |
if (ind - 1 <= 2) { |
| 2914 |
if (fstar.w[1] > 0x8000000000000000ull || |
| 2915 |
(fstar.w[1] == 0x8000000000000000ull |
| 2916 |
&& fstar.w[0] > 0x0ull)) { |
| 2917 |
// f* > 1/2 and the result may be exact |
| 2918 |
tmp64 = fstar.w[1] - 0x8000000000000000ull; // f* - 1/2 |
| 2919 |
if (tmp64 > ten2mk128trunc[ind - 1].w[1] |
| 2920 |
|| (tmp64 == ten2mk128trunc[ind - 1].w[1] |
| 2921 |
&& fstar.w[0] >= ten2mk128trunc[ind - 1].w[0])) { |
| 2922 |
// set the inexact flag |
| 2923 |
*pfpsf |= INEXACT_EXCEPTION; |
| 2924 |
} // else the result is exact |
| 2925 |
} else { // the result is inexact; f2* <= 1/2 |
| 2926 |
// set the inexact flag |
| 2927 |
*pfpsf |= INEXACT_EXCEPTION; |
| 2928 |
} |
| 2929 |
} else if (ind - 1 <= 21) { // if 3 <= ind <= 21 |
| 2930 |
if (fstar.w[3] > 0x0 || |
| 2931 |
(fstar.w[3] == 0x0 && fstar.w[2] > onehalf128[ind - 1]) || |
| 2932 |
(fstar.w[3] == 0x0 && fstar.w[2] == onehalf128[ind - 1] && |
| 2933 |
(fstar.w[1] || fstar.w[0]))) { |
| 2934 |
// f2* > 1/2 and the result may be exact |
| 2935 |
// Calculate f2* - 1/2 |
| 2936 |
tmp64 = fstar.w[2] - onehalf128[ind - 1]; |
| 2937 |
tmp64A = fstar.w[3]; |
| 2938 |
if (tmp64 > fstar.w[2]) |
| 2939 |
tmp64A--; |
| 2940 |
if (tmp64A || tmp64 |
| 2941 |
|| fstar.w[1] > ten2mk128trunc[ind - 1].w[1] |
| 2942 |
|| (fstar.w[1] == ten2mk128trunc[ind - 1].w[1] |
| 2943 |
&& fstar.w[0] > ten2mk128trunc[ind - 1].w[0])) { |
| 2944 |
// set the inexact flag |
| 2945 |
*pfpsf |= INEXACT_EXCEPTION; |
| 2946 |
} // else the result is exact |
| 2947 |
} else { // the result is inexact; f2* <= 1/2 |
| 2948 |
// set the inexact flag |
| 2949 |
*pfpsf |= INEXACT_EXCEPTION; |
| 2950 |
} |
| 2951 |
} else { // if 22 <= ind <= 33 |
| 2952 |
if (fstar.w[3] > onehalf128[ind - 1] || |
| 2953 |
(fstar.w[3] == onehalf128[ind - 1] && |
| 2954 |
(fstar.w[2] || fstar.w[1] || fstar.w[0]))) { |
| 2955 |
// f2* > 1/2 and the result may be exact |
| 2956 |
// Calculate f2* - 1/2 |
| 2957 |
tmp64 = fstar.w[3] - onehalf128[ind - 1]; |
| 2958 |
if (tmp64 || fstar.w[2] |
| 2959 |
|| fstar.w[1] > ten2mk128trunc[ind - 1].w[1] |
| 2960 |
|| (fstar.w[1] == ten2mk128trunc[ind - 1].w[1] |
| 2961 |
&& fstar.w[0] > ten2mk128trunc[ind - 1].w[0])) { |
| 2962 |
// set the inexact flag |
| 2963 |
*pfpsf |= INEXACT_EXCEPTION; |
| 2964 |
} // else the result is exact |
| 2965 |
} else { // the result is inexact; f2* <= 1/2 |
| 2966 |
// set the inexact flag |
| 2967 |
*pfpsf |= INEXACT_EXCEPTION; |
| 2968 |
} |
| 2969 |
} |
| 2970 |
|
| 2971 |
// if the result was a midpoint it was rounded away from zero |
| 2972 |
if (x_sign) |
| 2973 |
res = -Cstar.w[0]; |
| 2974 |
else |
| 2975 |
res = Cstar.w[0]; |
| 2976 |
} else if (exp == 0) { |
| 2977 |
// 1 <= q <= 19 |
| 2978 |
// res = +/-C (exact) |
| 2979 |
if (x_sign) |
| 2980 |
res = -C1.w[0]; |
| 2981 |
else |
| 2982 |
res = C1.w[0]; |
| 2983 |
} else { // if (exp>0) => 1 <= exp <= 18, 1 <= q < 18, 2 <= q + exp <= 19 |
| 2984 |
// res = +/-C * 10^exp (exact) where this fits in 64-bit integer |
| 2985 |
if (x_sign) |
| 2986 |
res = -C1.w[0] * ten2k64[exp]; |
| 2987 |
else |
| 2988 |
res = C1.w[0] * ten2k64[exp]; |
| 2989 |
} |
| 2990 |
} |
| 2991 |
} |
| 2992 |
|
| 2993 |
BID_RETURN (res); |
| 2994 |
} |