URL
https://opencores.org/ocsvn/openrisc_me/openrisc_me/trunk
Subversion Repositories openrisc_me
[/] [openrisc/] [trunk/] [gnu-src/] [gcc-4.5.1/] [libgcc/] [config/] [libbid/] [bid64_round_integral.c] - Rev 407
Go to most recent revision | Compare with Previous | Blame | View Log
/* Copyright (C) 2007, 2009 Free Software Foundation, Inc. This file is part of GCC. GCC is free software; you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation; either version 3, or (at your option) any later version. GCC is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details. Under Section 7 of GPL version 3, you are granted additional permissions described in the GCC Runtime Library Exception, version 3.1, as published by the Free Software Foundation. You should have received a copy of the GNU General Public License and a copy of the GCC Runtime Library Exception along with this program; see the files COPYING3 and COPYING.RUNTIME respectively. If not, see <http://www.gnu.org/licenses/>. */ #include "bid_internal.h" /***************************************************************************** * BID64_round_integral_exact ****************************************************************************/ #if DECIMAL_CALL_BY_REFERENCE void bid64_round_integral_exact (UINT64 * pres, UINT64 * px _RND_MODE_PARAM _EXC_FLAGS_PARAM _EXC_MASKS_PARAM _EXC_INFO_PARAM) { UINT64 x = *px; #if !DECIMAL_GLOBAL_ROUNDING unsigned int rnd_mode = *prnd_mode; #endif #else UINT64 bid64_round_integral_exact (UINT64 x _RND_MODE_PARAM _EXC_FLAGS_PARAM _EXC_MASKS_PARAM _EXC_INFO_PARAM) { #endif UINT64 res = 0xbaddbaddbaddbaddull; UINT64 x_sign; int exp; // unbiased exponent // Note: C1 represents the significand (UINT64) BID_UI64DOUBLE tmp1; int x_nr_bits; int q, ind, shift; UINT64 C1; // UINT64 res is C* at first - represents up to 16 decimal digits <= 54 bits UINT128 fstar = { {0x0ull, 0x0ull} }; UINT128 P128; x_sign = x & MASK_SIGN; // 0 for positive, MASK_SIGN for negative // check for NaNs and infinities if ((x & MASK_NAN) == MASK_NAN) { // check for NaN if ((x & 0x0003ffffffffffffull) > 999999999999999ull) x = x & 0xfe00000000000000ull; // clear G6-G12 and the payload bits else x = x & 0xfe03ffffffffffffull; // clear G6-G12 if ((x & MASK_SNAN) == MASK_SNAN) { // SNaN // set invalid flag *pfpsf |= INVALID_EXCEPTION; // return quiet (SNaN) res = x & 0xfdffffffffffffffull; } else { // QNaN res = x; } BID_RETURN (res); } else if ((x & MASK_INF) == MASK_INF) { // check for Infinity res = x_sign | 0x7800000000000000ull; BID_RETURN (res); } // unpack x if ((x & MASK_STEERING_BITS) == MASK_STEERING_BITS) { // if the steering bits are 11 (condition will be 0), then // the exponent is G[0:w+1] exp = ((x & MASK_BINARY_EXPONENT2) >> 51) - 398; C1 = (x & MASK_BINARY_SIG2) | MASK_BINARY_OR2; if (C1 > 9999999999999999ull) { // non-canonical C1 = 0; } } else { // if ((x & MASK_STEERING_BITS) != MASK_STEERING_BITS) exp = ((x & MASK_BINARY_EXPONENT1) >> 53) - 398; C1 = (x & MASK_BINARY_SIG1); } // if x is 0 or non-canonical return 0 preserving the sign bit and // the preferred exponent of MAX(Q(x), 0) if (C1 == 0) { if (exp < 0) exp = 0; res = x_sign | (((UINT64) exp + 398) << 53); BID_RETURN (res); } // x is a finite non-zero number (not 0, non-canonical, or special) switch (rnd_mode) { case ROUNDING_TO_NEAREST: case ROUNDING_TIES_AWAY: // return 0 if (exp <= -(p+1)) if (exp <= -17) { res = x_sign | 0x31c0000000000000ull; *pfpsf |= INEXACT_EXCEPTION; BID_RETURN (res); } break; case ROUNDING_DOWN: // return 0 if (exp <= -p) if (exp <= -16) { if (x_sign) { res = 0xb1c0000000000001ull; } else { res = 0x31c0000000000000ull; } *pfpsf |= INEXACT_EXCEPTION; BID_RETURN (res); } break; case ROUNDING_UP: // return 0 if (exp <= -p) if (exp <= -16) { if (x_sign) { res = 0xb1c0000000000000ull; } else { res = 0x31c0000000000001ull; } *pfpsf |= INEXACT_EXCEPTION; BID_RETURN (res); } break; case ROUNDING_TO_ZERO: // return 0 if (exp <= -p) if (exp <= -16) { res = x_sign | 0x31c0000000000000ull; *pfpsf |= INEXACT_EXCEPTION; BID_RETURN (res); } break; } // end switch () // q = nr. of decimal digits in x (1 <= q <= 54) // determine first the nr. of bits in x if (C1 >= 0x0020000000000000ull) { // x >= 2^53 q = 16; } else { // if x < 2^53 tmp1.d = (double) C1; // exact conversion x_nr_bits = 1 + ((((unsigned int) (tmp1.ui64 >> 52)) & 0x7ff) - 0x3ff); q = nr_digits[x_nr_bits - 1].digits; if (q == 0) { q = nr_digits[x_nr_bits - 1].digits1; if (C1 >= nr_digits[x_nr_bits - 1].threshold_lo) q++; } } if (exp >= 0) { // -exp <= 0 // the argument is an integer already res = x; BID_RETURN (res); } switch (rnd_mode) { case ROUNDING_TO_NEAREST: if ((q + exp) >= 0) { // exp < 0 and 1 <= -exp <= q // need to shift right -exp digits from the coefficient; exp will be 0 ind = -exp; // 1 <= ind <= 16; ind is a synonym for 'x' // chop off ind digits from the lower part of C1 // C1 = C1 + 1/2 * 10^x where the result C1 fits in 64 bits // FOR ROUND_TO_NEAREST, WE ADD 1/2 ULP(y) then truncate C1 = C1 + midpoint64[ind - 1]; // calculate C* and f* // C* is actually floor(C*) in this case // C* and f* need shifting and masking, as shown by // shiftright128[] and maskhigh128[] // 1 <= x <= 16 // kx = 10^(-x) = ten2mk64[ind - 1] // C* = (C1 + 1/2 * 10^x) * 10^(-x) // the approximation of 10^(-x) was rounded up to 64 bits __mul_64x64_to_128 (P128, C1, ten2mk64[ind - 1]); // if (0 < f* < 10^(-x)) then the result is a midpoint // if floor(C*) is even then C* = floor(C*) - logical right // shift; C* has p decimal digits, correct by Prop. 1) // else if floor(C*) is odd C* = floor(C*)-1 (logical right // shift; C* has p decimal digits, correct by Pr. 1) // else // C* = floor(C*) (logical right shift; C has p decimal digits, // correct by Property 1) // n = C* * 10^(e+x) if (ind - 1 <= 2) { // 0 <= ind - 1 <= 2 => shift = 0 res = P128.w[1]; fstar.w[1] = 0; fstar.w[0] = P128.w[0]; } else if (ind - 1 <= 21) { // 3 <= ind - 1 <= 21 => 3 <= shift <= 63 shift = shiftright128[ind - 1]; // 3 <= shift <= 63 res = (P128.w[1] >> shift); fstar.w[1] = P128.w[1] & maskhigh128[ind - 1]; fstar.w[0] = P128.w[0]; } // if (0 < f* < 10^(-x)) then the result is a midpoint // since round_to_even, subtract 1 if current result is odd if ((res & 0x0000000000000001ull) && (fstar.w[1] == 0) && (fstar.w[0] < ten2mk64[ind - 1])) { res--; } // determine inexactness of the rounding of C* // if (0 < f* - 1/2 < 10^(-x)) then // the result is exact // else // if (f* - 1/2 > T*) then // the result is inexact if (ind - 1 <= 2) { if (fstar.w[0] > 0x8000000000000000ull) { // f* > 1/2 and the result may be exact // fstar.w[0] - 0x8000000000000000ull is f* - 1/2 if ((fstar.w[0] - 0x8000000000000000ull) > ten2mk64[ind - 1]) { // set the inexact flag *pfpsf |= INEXACT_EXCEPTION; } // else the result is exact } else { // the result is inexact; f2* <= 1/2 // set the inexact flag *pfpsf |= INEXACT_EXCEPTION; } } else { // if 3 <= ind - 1 <= 21 if (fstar.w[1] > onehalf128[ind - 1] || (fstar.w[1] == onehalf128[ind - 1] && fstar.w[0])) { // f2* > 1/2 and the result may be exact // Calculate f2* - 1/2 if (fstar.w[1] > onehalf128[ind - 1] || fstar.w[0] > ten2mk64[ind - 1]) { // set the inexact flag *pfpsf |= INEXACT_EXCEPTION; } // else the result is exact } else { // the result is inexact; f2* <= 1/2 // set the inexact flag *pfpsf |= INEXACT_EXCEPTION; } } // set exponent to zero as it was negative before. res = x_sign | 0x31c0000000000000ull | res; BID_RETURN (res); } else { // if exp < 0 and q + exp < 0 // the result is +0 or -0 res = x_sign | 0x31c0000000000000ull; *pfpsf |= INEXACT_EXCEPTION; BID_RETURN (res); } break; case ROUNDING_TIES_AWAY: if ((q + exp) >= 0) { // exp < 0 and 1 <= -exp <= q // need to shift right -exp digits from the coefficient; exp will be 0 ind = -exp; // 1 <= ind <= 16; ind is a synonym for 'x' // chop off ind digits from the lower part of C1 // C1 = C1 + 1/2 * 10^x where the result C1 fits in 64 bits // FOR ROUND_TO_NEAREST, WE ADD 1/2 ULP(y) then truncate C1 = C1 + midpoint64[ind - 1]; // calculate C* and f* // C* is actually floor(C*) in this case // C* and f* need shifting and masking, as shown by // shiftright128[] and maskhigh128[] // 1 <= x <= 16 // kx = 10^(-x) = ten2mk64[ind - 1] // C* = (C1 + 1/2 * 10^x) * 10^(-x) // the approximation of 10^(-x) was rounded up to 64 bits __mul_64x64_to_128 (P128, C1, ten2mk64[ind - 1]); // if (0 < f* < 10^(-x)) then the result is a midpoint // C* = floor(C*) - logical right shift; C* has p decimal digits, // correct by Prop. 1) // else // C* = floor(C*) (logical right shift; C has p decimal digits, // correct by Property 1) // n = C* * 10^(e+x) if (ind - 1 <= 2) { // 0 <= ind - 1 <= 2 => shift = 0 res = P128.w[1]; fstar.w[1] = 0; fstar.w[0] = P128.w[0]; } else if (ind - 1 <= 21) { // 3 <= ind - 1 <= 21 => 3 <= shift <= 63 shift = shiftright128[ind - 1]; // 3 <= shift <= 63 res = (P128.w[1] >> shift); fstar.w[1] = P128.w[1] & maskhigh128[ind - 1]; fstar.w[0] = P128.w[0]; } // midpoints are already rounded correctly // determine inexactness of the rounding of C* // if (0 < f* - 1/2 < 10^(-x)) then // the result is exact // else // if (f* - 1/2 > T*) then // the result is inexact if (ind - 1 <= 2) { if (fstar.w[0] > 0x8000000000000000ull) { // f* > 1/2 and the result may be exact // fstar.w[0] - 0x8000000000000000ull is f* - 1/2 if ((fstar.w[0] - 0x8000000000000000ull) > ten2mk64[ind - 1]) { // set the inexact flag *pfpsf |= INEXACT_EXCEPTION; } // else the result is exact } else { // the result is inexact; f2* <= 1/2 // set the inexact flag *pfpsf |= INEXACT_EXCEPTION; } } else { // if 3 <= ind - 1 <= 21 if (fstar.w[1] > onehalf128[ind - 1] || (fstar.w[1] == onehalf128[ind - 1] && fstar.w[0])) { // f2* > 1/2 and the result may be exact // Calculate f2* - 1/2 if (fstar.w[1] > onehalf128[ind - 1] || fstar.w[0] > ten2mk64[ind - 1]) { // set the inexact flag *pfpsf |= INEXACT_EXCEPTION; } // else the result is exact } else { // the result is inexact; f2* <= 1/2 // set the inexact flag *pfpsf |= INEXACT_EXCEPTION; } } // set exponent to zero as it was negative before. res = x_sign | 0x31c0000000000000ull | res; BID_RETURN (res); } else { // if exp < 0 and q + exp < 0 // the result is +0 or -0 res = x_sign | 0x31c0000000000000ull; *pfpsf |= INEXACT_EXCEPTION; BID_RETURN (res); } break; case ROUNDING_DOWN: if ((q + exp) > 0) { // exp < 0 and 1 <= -exp < q // need to shift right -exp digits from the coefficient; exp will be 0 ind = -exp; // 1 <= ind <= 16; ind is a synonym for 'x' // chop off ind digits from the lower part of C1 // C1 fits in 64 bits // calculate C* and f* // C* is actually floor(C*) in this case // C* and f* need shifting and masking, as shown by // shiftright128[] and maskhigh128[] // 1 <= x <= 16 // kx = 10^(-x) = ten2mk64[ind - 1] // C* = C1 * 10^(-x) // the approximation of 10^(-x) was rounded up to 64 bits __mul_64x64_to_128 (P128, C1, ten2mk64[ind - 1]); // C* = floor(C*) (logical right shift; C has p decimal digits, // correct by Property 1) // if (0 < f* < 10^(-x)) then the result is exact // n = C* * 10^(e+x) if (ind - 1 <= 2) { // 0 <= ind - 1 <= 2 => shift = 0 res = P128.w[1]; fstar.w[1] = 0; fstar.w[0] = P128.w[0]; } else if (ind - 1 <= 21) { // 3 <= ind - 1 <= 21 => 3 <= shift <= 63 shift = shiftright128[ind - 1]; // 3 <= shift <= 63 res = (P128.w[1] >> shift); fstar.w[1] = P128.w[1] & maskhigh128[ind - 1]; fstar.w[0] = P128.w[0]; } // if (f* > 10^(-x)) then the result is inexact if ((fstar.w[1] != 0) || (fstar.w[0] >= ten2mk64[ind - 1])) { if (x_sign) { // if negative and not exact, increment magnitude res++; } *pfpsf |= INEXACT_EXCEPTION; } // set exponent to zero as it was negative before. res = x_sign | 0x31c0000000000000ull | res; BID_RETURN (res); } else { // if exp < 0 and q + exp <= 0 // the result is +0 or -1 if (x_sign) { res = 0xb1c0000000000001ull; } else { res = 0x31c0000000000000ull; } *pfpsf |= INEXACT_EXCEPTION; BID_RETURN (res); } break; case ROUNDING_UP: if ((q + exp) > 0) { // exp < 0 and 1 <= -exp < q // need to shift right -exp digits from the coefficient; exp will be 0 ind = -exp; // 1 <= ind <= 16; ind is a synonym for 'x' // chop off ind digits from the lower part of C1 // C1 fits in 64 bits // calculate C* and f* // C* is actually floor(C*) in this case // C* and f* need shifting and masking, as shown by // shiftright128[] and maskhigh128[] // 1 <= x <= 16 // kx = 10^(-x) = ten2mk64[ind - 1] // C* = C1 * 10^(-x) // the approximation of 10^(-x) was rounded up to 64 bits __mul_64x64_to_128 (P128, C1, ten2mk64[ind - 1]); // C* = floor(C*) (logical right shift; C has p decimal digits, // correct by Property 1) // if (0 < f* < 10^(-x)) then the result is exact // n = C* * 10^(e+x) if (ind - 1 <= 2) { // 0 <= ind - 1 <= 2 => shift = 0 res = P128.w[1]; fstar.w[1] = 0; fstar.w[0] = P128.w[0]; } else if (ind - 1 <= 21) { // 3 <= ind - 1 <= 21 => 3 <= shift <= 63 shift = shiftright128[ind - 1]; // 3 <= shift <= 63 res = (P128.w[1] >> shift); fstar.w[1] = P128.w[1] & maskhigh128[ind - 1]; fstar.w[0] = P128.w[0]; } // if (f* > 10^(-x)) then the result is inexact if ((fstar.w[1] != 0) || (fstar.w[0] >= ten2mk64[ind - 1])) { if (!x_sign) { // if positive and not exact, increment magnitude res++; } *pfpsf |= INEXACT_EXCEPTION; } // set exponent to zero as it was negative before. res = x_sign | 0x31c0000000000000ull | res; BID_RETURN (res); } else { // if exp < 0 and q + exp <= 0 // the result is -0 or +1 if (x_sign) { res = 0xb1c0000000000000ull; } else { res = 0x31c0000000000001ull; } *pfpsf |= INEXACT_EXCEPTION; BID_RETURN (res); } break; case ROUNDING_TO_ZERO: if ((q + exp) >= 0) { // exp < 0 and 1 <= -exp <= q // need to shift right -exp digits from the coefficient; exp will be 0 ind = -exp; // 1 <= ind <= 16; ind is a synonym for 'x' // chop off ind digits from the lower part of C1 // C1 fits in 127 bits // calculate C* and f* // C* is actually floor(C*) in this case // C* and f* need shifting and masking, as shown by // shiftright128[] and maskhigh128[] // 1 <= x <= 16 // kx = 10^(-x) = ten2mk64[ind - 1] // C* = C1 * 10^(-x) // the approximation of 10^(-x) was rounded up to 64 bits __mul_64x64_to_128 (P128, C1, ten2mk64[ind - 1]); // C* = floor(C*) (logical right shift; C has p decimal digits, // correct by Property 1) // if (0 < f* < 10^(-x)) then the result is exact // n = C* * 10^(e+x) if (ind - 1 <= 2) { // 0 <= ind - 1 <= 2 => shift = 0 res = P128.w[1]; fstar.w[1] = 0; fstar.w[0] = P128.w[0]; } else if (ind - 1 <= 21) { // 3 <= ind - 1 <= 21 => 3 <= shift <= 63 shift = shiftright128[ind - 1]; // 3 <= shift <= 63 res = (P128.w[1] >> shift); fstar.w[1] = P128.w[1] & maskhigh128[ind - 1]; fstar.w[0] = P128.w[0]; } // if (f* > 10^(-x)) then the result is inexact if ((fstar.w[1] != 0) || (fstar.w[0] >= ten2mk64[ind - 1])) { *pfpsf |= INEXACT_EXCEPTION; } // set exponent to zero as it was negative before. res = x_sign | 0x31c0000000000000ull | res; BID_RETURN (res); } else { // if exp < 0 and q + exp < 0 // the result is +0 or -0 res = x_sign | 0x31c0000000000000ull; *pfpsf |= INEXACT_EXCEPTION; BID_RETURN (res); } break; } // end switch () BID_RETURN (res); } /***************************************************************************** * BID64_round_integral_nearest_even ****************************************************************************/ #if DECIMAL_CALL_BY_REFERENCE void bid64_round_integral_nearest_even (UINT64 * pres, UINT64 * px _EXC_FLAGS_PARAM _EXC_MASKS_PARAM _EXC_INFO_PARAM) { UINT64 x = *px; #else UINT64 bid64_round_integral_nearest_even (UINT64 x _EXC_FLAGS_PARAM _EXC_MASKS_PARAM _EXC_INFO_PARAM) { #endif UINT64 res = 0xbaddbaddbaddbaddull; UINT64 x_sign; int exp; // unbiased exponent // Note: C1.w[1], C1.w[0] represent x_signif_hi, x_signif_lo (all are UINT64) BID_UI64DOUBLE tmp1; int x_nr_bits; int q, ind, shift; UINT64 C1; UINT128 fstar; UINT128 P128; x_sign = x & MASK_SIGN; // 0 for positive, MASK_SIGN for negative // check for NaNs and infinities if ((x & MASK_NAN) == MASK_NAN) { // check for NaN if ((x & 0x0003ffffffffffffull) > 999999999999999ull) x = x & 0xfe00000000000000ull; // clear G6-G12 and the payload bits else x = x & 0xfe03ffffffffffffull; // clear G6-G12 if ((x & MASK_SNAN) == MASK_SNAN) { // SNaN // set invalid flag *pfpsf |= INVALID_EXCEPTION; // return quiet (SNaN) res = x & 0xfdffffffffffffffull; } else { // QNaN res = x; } BID_RETURN (res); } else if ((x & MASK_INF) == MASK_INF) { // check for Infinity res = x_sign | 0x7800000000000000ull; BID_RETURN (res); } // unpack x if ((x & MASK_STEERING_BITS) == MASK_STEERING_BITS) { // if the steering bits are 11 (condition will be 0), then // the exponent is G[0:w+1] exp = ((x & MASK_BINARY_EXPONENT2) >> 51) - 398; C1 = (x & MASK_BINARY_SIG2) | MASK_BINARY_OR2; if (C1 > 9999999999999999ull) { // non-canonical C1 = 0; } } else { // if ((x & MASK_STEERING_BITS) != MASK_STEERING_BITS) exp = ((x & MASK_BINARY_EXPONENT1) >> 53) - 398; C1 = (x & MASK_BINARY_SIG1); } // if x is 0 or non-canonical if (C1 == 0) { if (exp < 0) exp = 0; res = x_sign | (((UINT64) exp + 398) << 53); BID_RETURN (res); } // x is a finite non-zero number (not 0, non-canonical, or special) // return 0 if (exp <= -(p+1)) if (exp <= -17) { res = x_sign | 0x31c0000000000000ull; BID_RETURN (res); } // q = nr. of decimal digits in x (1 <= q <= 54) // determine first the nr. of bits in x if (C1 >= 0x0020000000000000ull) { // x >= 2^53 q = 16; } else { // if x < 2^53 tmp1.d = (double) C1; // exact conversion x_nr_bits = 1 + ((((unsigned int) (tmp1.ui64 >> 52)) & 0x7ff) - 0x3ff); q = nr_digits[x_nr_bits - 1].digits; if (q == 0) { q = nr_digits[x_nr_bits - 1].digits1; if (C1 >= nr_digits[x_nr_bits - 1].threshold_lo) q++; } } if (exp >= 0) { // -exp <= 0 // the argument is an integer already res = x; BID_RETURN (res); } else if ((q + exp) >= 0) { // exp < 0 and 1 <= -exp <= q // need to shift right -exp digits from the coefficient; the exp will be 0 ind = -exp; // 1 <= ind <= 16; ind is a synonym for 'x' // chop off ind digits from the lower part of C1 // C1 = C1 + 1/2 * 10^x where the result C1 fits in 64 bits // FOR ROUND_TO_NEAREST, WE ADD 1/2 ULP(y) then truncate C1 = C1 + midpoint64[ind - 1]; // calculate C* and f* // C* is actually floor(C*) in this case // C* and f* need shifting and masking, as shown by // shiftright128[] and maskhigh128[] // 1 <= x <= 16 // kx = 10^(-x) = ten2mk64[ind - 1] // C* = (C1 + 1/2 * 10^x) * 10^(-x) // the approximation of 10^(-x) was rounded up to 64 bits __mul_64x64_to_128 (P128, C1, ten2mk64[ind - 1]); // if (0 < f* < 10^(-x)) then the result is a midpoint // if floor(C*) is even then C* = floor(C*) - logical right // shift; C* has p decimal digits, correct by Prop. 1) // else if floor(C*) is odd C* = floor(C*)-1 (logical right // shift; C* has p decimal digits, correct by Pr. 1) // else // C* = floor(C*) (logical right shift; C has p decimal digits, // correct by Property 1) // n = C* * 10^(e+x) if (ind - 1 <= 2) { // 0 <= ind - 1 <= 2 => shift = 0 res = P128.w[1]; fstar.w[1] = 0; fstar.w[0] = P128.w[0]; } else if (ind - 1 <= 21) { // 3 <= ind - 1 <= 21 => 3 <= shift <= 63 shift = shiftright128[ind - 1]; // 3 <= shift <= 63 res = (P128.w[1] >> shift); fstar.w[1] = P128.w[1] & maskhigh128[ind - 1]; fstar.w[0] = P128.w[0]; } // if (0 < f* < 10^(-x)) then the result is a midpoint // since round_to_even, subtract 1 if current result is odd if ((res & 0x0000000000000001ull) && (fstar.w[1] == 0) && (fstar.w[0] < ten2mk64[ind - 1])) { res--; } // set exponent to zero as it was negative before. res = x_sign | 0x31c0000000000000ull | res; BID_RETURN (res); } else { // if exp < 0 and q + exp < 0 // the result is +0 or -0 res = x_sign | 0x31c0000000000000ull; BID_RETURN (res); } } /***************************************************************************** * BID64_round_integral_negative *****************************************************************************/ #if DECIMAL_CALL_BY_REFERENCE void bid64_round_integral_negative (UINT64 * pres, UINT64 * px _EXC_FLAGS_PARAM _EXC_MASKS_PARAM _EXC_INFO_PARAM) { UINT64 x = *px; #else UINT64 bid64_round_integral_negative (UINT64 x _EXC_FLAGS_PARAM _EXC_MASKS_PARAM _EXC_INFO_PARAM) { #endif UINT64 res = 0xbaddbaddbaddbaddull; UINT64 x_sign; int exp; // unbiased exponent // Note: C1.w[1], C1.w[0] represent x_signif_hi, x_signif_lo (all are UINT64) BID_UI64DOUBLE tmp1; int x_nr_bits; int q, ind, shift; UINT64 C1; // UINT64 res is C* at first - represents up to 34 decimal digits ~ 113 bits UINT128 fstar; UINT128 P128; x_sign = x & MASK_SIGN; // 0 for positive, MASK_SIGN for negative // check for NaNs and infinities if ((x & MASK_NAN) == MASK_NAN) { // check for NaN if ((x & 0x0003ffffffffffffull) > 999999999999999ull) x = x & 0xfe00000000000000ull; // clear G6-G12 and the payload bits else x = x & 0xfe03ffffffffffffull; // clear G6-G12 if ((x & MASK_SNAN) == MASK_SNAN) { // SNaN // set invalid flag *pfpsf |= INVALID_EXCEPTION; // return quiet (SNaN) res = x & 0xfdffffffffffffffull; } else { // QNaN res = x; } BID_RETURN (res); } else if ((x & MASK_INF) == MASK_INF) { // check for Infinity res = x_sign | 0x7800000000000000ull; BID_RETURN (res); } // unpack x if ((x & MASK_STEERING_BITS) == MASK_STEERING_BITS) { // if the steering bits are 11 (condition will be 0), then // the exponent is G[0:w+1] exp = ((x & MASK_BINARY_EXPONENT2) >> 51) - 398; C1 = (x & MASK_BINARY_SIG2) | MASK_BINARY_OR2; if (C1 > 9999999999999999ull) { // non-canonical C1 = 0; } } else { // if ((x & MASK_STEERING_BITS) != MASK_STEERING_BITS) exp = ((x & MASK_BINARY_EXPONENT1) >> 53) - 398; C1 = (x & MASK_BINARY_SIG1); } // if x is 0 or non-canonical if (C1 == 0) { if (exp < 0) exp = 0; res = x_sign | (((UINT64) exp + 398) << 53); BID_RETURN (res); } // x is a finite non-zero number (not 0, non-canonical, or special) // return 0 if (exp <= -p) if (exp <= -16) { if (x_sign) { res = 0xb1c0000000000001ull; } else { res = 0x31c0000000000000ull; } BID_RETURN (res); } // q = nr. of decimal digits in x (1 <= q <= 54) // determine first the nr. of bits in x if (C1 >= 0x0020000000000000ull) { // x >= 2^53 q = 16; } else { // if x < 2^53 tmp1.d = (double) C1; // exact conversion x_nr_bits = 1 + ((((unsigned int) (tmp1.ui64 >> 52)) & 0x7ff) - 0x3ff); q = nr_digits[x_nr_bits - 1].digits; if (q == 0) { q = nr_digits[x_nr_bits - 1].digits1; if (C1 >= nr_digits[x_nr_bits - 1].threshold_lo) q++; } } if (exp >= 0) { // -exp <= 0 // the argument is an integer already res = x; BID_RETURN (res); } else if ((q + exp) > 0) { // exp < 0 and 1 <= -exp < q // need to shift right -exp digits from the coefficient; the exp will be 0 ind = -exp; // 1 <= ind <= 16; ind is a synonym for 'x' // chop off ind digits from the lower part of C1 // C1 fits in 64 bits // calculate C* and f* // C* is actually floor(C*) in this case // C* and f* need shifting and masking, as shown by // shiftright128[] and maskhigh128[] // 1 <= x <= 16 // kx = 10^(-x) = ten2mk64[ind - 1] // C* = C1 * 10^(-x) // the approximation of 10^(-x) was rounded up to 64 bits __mul_64x64_to_128 (P128, C1, ten2mk64[ind - 1]); // C* = floor(C*) (logical right shift; C has p decimal digits, // correct by Property 1) // if (0 < f* < 10^(-x)) then the result is exact // n = C* * 10^(e+x) if (ind - 1 <= 2) { // 0 <= ind - 1 <= 2 => shift = 0 res = P128.w[1]; fstar.w[1] = 0; fstar.w[0] = P128.w[0]; } else if (ind - 1 <= 21) { // 3 <= ind - 1 <= 21 => 3 <= shift <= 63 shift = shiftright128[ind - 1]; // 3 <= shift <= 63 res = (P128.w[1] >> shift); fstar.w[1] = P128.w[1] & maskhigh128[ind - 1]; fstar.w[0] = P128.w[0]; } // if (f* > 10^(-x)) then the result is inexact if (x_sign && ((fstar.w[1] != 0) || (fstar.w[0] >= ten2mk64[ind - 1]))) { // if negative and not exact, increment magnitude res++; } // set exponent to zero as it was negative before. res = x_sign | 0x31c0000000000000ull | res; BID_RETURN (res); } else { // if exp < 0 and q + exp <= 0 // the result is +0 or -1 if (x_sign) { res = 0xb1c0000000000001ull; } else { res = 0x31c0000000000000ull; } BID_RETURN (res); } } /***************************************************************************** * BID64_round_integral_positive ****************************************************************************/ #if DECIMAL_CALL_BY_REFERENCE void bid64_round_integral_positive (UINT64 * pres, UINT64 * px _EXC_FLAGS_PARAM _EXC_MASKS_PARAM _EXC_INFO_PARAM) { UINT64 x = *px; #else UINT64 bid64_round_integral_positive (UINT64 x _EXC_FLAGS_PARAM _EXC_MASKS_PARAM _EXC_INFO_PARAM) { #endif UINT64 res = 0xbaddbaddbaddbaddull; UINT64 x_sign; int exp; // unbiased exponent // Note: C1.w[1], C1.w[0] represent x_signif_hi, x_signif_lo (all are UINT64) BID_UI64DOUBLE tmp1; int x_nr_bits; int q, ind, shift; UINT64 C1; // UINT64 res is C* at first - represents up to 34 decimal digits ~ 113 bits UINT128 fstar; UINT128 P128; x_sign = x & MASK_SIGN; // 0 for positive, MASK_SIGN for negative // check for NaNs and infinities if ((x & MASK_NAN) == MASK_NAN) { // check for NaN if ((x & 0x0003ffffffffffffull) > 999999999999999ull) x = x & 0xfe00000000000000ull; // clear G6-G12 and the payload bits else x = x & 0xfe03ffffffffffffull; // clear G6-G12 if ((x & MASK_SNAN) == MASK_SNAN) { // SNaN // set invalid flag *pfpsf |= INVALID_EXCEPTION; // return quiet (SNaN) res = x & 0xfdffffffffffffffull; } else { // QNaN res = x; } BID_RETURN (res); } else if ((x & MASK_INF) == MASK_INF) { // check for Infinity res = x_sign | 0x7800000000000000ull; BID_RETURN (res); } // unpack x if ((x & MASK_STEERING_BITS) == MASK_STEERING_BITS) { // if the steering bits are 11 (condition will be 0), then // the exponent is G[0:w+1] exp = ((x & MASK_BINARY_EXPONENT2) >> 51) - 398; C1 = (x & MASK_BINARY_SIG2) | MASK_BINARY_OR2; if (C1 > 9999999999999999ull) { // non-canonical C1 = 0; } } else { // if ((x & MASK_STEERING_BITS) != MASK_STEERING_BITS) exp = ((x & MASK_BINARY_EXPONENT1) >> 53) - 398; C1 = (x & MASK_BINARY_SIG1); } // if x is 0 or non-canonical if (C1 == 0) { if (exp < 0) exp = 0; res = x_sign | (((UINT64) exp + 398) << 53); BID_RETURN (res); } // x is a finite non-zero number (not 0, non-canonical, or special) // return 0 if (exp <= -p) if (exp <= -16) { if (x_sign) { res = 0xb1c0000000000000ull; } else { res = 0x31c0000000000001ull; } BID_RETURN (res); } // q = nr. of decimal digits in x (1 <= q <= 54) // determine first the nr. of bits in x if (C1 >= 0x0020000000000000ull) { // x >= 2^53 q = 16; } else { // if x < 2^53 tmp1.d = (double) C1; // exact conversion x_nr_bits = 1 + ((((unsigned int) (tmp1.ui64 >> 52)) & 0x7ff) - 0x3ff); q = nr_digits[x_nr_bits - 1].digits; if (q == 0) { q = nr_digits[x_nr_bits - 1].digits1; if (C1 >= nr_digits[x_nr_bits - 1].threshold_lo) q++; } } if (exp >= 0) { // -exp <= 0 // the argument is an integer already res = x; BID_RETURN (res); } else if ((q + exp) > 0) { // exp < 0 and 1 <= -exp < q // need to shift right -exp digits from the coefficient; the exp will be 0 ind = -exp; // 1 <= ind <= 16; ind is a synonym for 'x' // chop off ind digits from the lower part of C1 // C1 fits in 64 bits // calculate C* and f* // C* is actually floor(C*) in this case // C* and f* need shifting and masking, as shown by // shiftright128[] and maskhigh128[] // 1 <= x <= 16 // kx = 10^(-x) = ten2mk64[ind - 1] // C* = C1 * 10^(-x) // the approximation of 10^(-x) was rounded up to 64 bits __mul_64x64_to_128 (P128, C1, ten2mk64[ind - 1]); // C* = floor(C*) (logical right shift; C has p decimal digits, // correct by Property 1) // if (0 < f* < 10^(-x)) then the result is exact // n = C* * 10^(e+x) if (ind - 1 <= 2) { // 0 <= ind - 1 <= 2 => shift = 0 res = P128.w[1]; fstar.w[1] = 0; fstar.w[0] = P128.w[0]; } else if (ind - 1 <= 21) { // 3 <= ind - 1 <= 21 => 3 <= shift <= 63 shift = shiftright128[ind - 1]; // 3 <= shift <= 63 res = (P128.w[1] >> shift); fstar.w[1] = P128.w[1] & maskhigh128[ind - 1]; fstar.w[0] = P128.w[0]; } // if (f* > 10^(-x)) then the result is inexact if (!x_sign && ((fstar.w[1] != 0) || (fstar.w[0] >= ten2mk64[ind - 1]))) { // if positive and not exact, increment magnitude res++; } // set exponent to zero as it was negative before. res = x_sign | 0x31c0000000000000ull | res; BID_RETURN (res); } else { // if exp < 0 and q + exp <= 0 // the result is -0 or +1 if (x_sign) { res = 0xb1c0000000000000ull; } else { res = 0x31c0000000000001ull; } BID_RETURN (res); } } /***************************************************************************** * BID64_round_integral_zero ****************************************************************************/ #if DECIMAL_CALL_BY_REFERENCE void bid64_round_integral_zero (UINT64 * pres, UINT64 * px _EXC_FLAGS_PARAM _EXC_MASKS_PARAM _EXC_INFO_PARAM) { UINT64 x = *px; #else UINT64 bid64_round_integral_zero (UINT64 x _EXC_FLAGS_PARAM _EXC_MASKS_PARAM _EXC_INFO_PARAM) { #endif UINT64 res = 0xbaddbaddbaddbaddull; UINT64 x_sign; int exp; // unbiased exponent // Note: C1.w[1], C1.w[0] represent x_signif_hi, x_signif_lo (all are UINT64) BID_UI64DOUBLE tmp1; int x_nr_bits; int q, ind, shift; UINT64 C1; // UINT64 res is C* at first - represents up to 34 decimal digits ~ 113 bits UINT128 P128; x_sign = x & MASK_SIGN; // 0 for positive, MASK_SIGN for negative // check for NaNs and infinities if ((x & MASK_NAN) == MASK_NAN) { // check for NaN if ((x & 0x0003ffffffffffffull) > 999999999999999ull) x = x & 0xfe00000000000000ull; // clear G6-G12 and the payload bits else x = x & 0xfe03ffffffffffffull; // clear G6-G12 if ((x & MASK_SNAN) == MASK_SNAN) { // SNaN // set invalid flag *pfpsf |= INVALID_EXCEPTION; // return quiet (SNaN) res = x & 0xfdffffffffffffffull; } else { // QNaN res = x; } BID_RETURN (res); } else if ((x & MASK_INF) == MASK_INF) { // check for Infinity res = x_sign | 0x7800000000000000ull; BID_RETURN (res); } // unpack x if ((x & MASK_STEERING_BITS) == MASK_STEERING_BITS) { // if the steering bits are 11 (condition will be 0), then // the exponent is G[0:w+1] exp = ((x & MASK_BINARY_EXPONENT2) >> 51) - 398; C1 = (x & MASK_BINARY_SIG2) | MASK_BINARY_OR2; if (C1 > 9999999999999999ull) { // non-canonical C1 = 0; } } else { // if ((x & MASK_STEERING_BITS) != MASK_STEERING_BITS) exp = ((x & MASK_BINARY_EXPONENT1) >> 53) - 398; C1 = (x & MASK_BINARY_SIG1); } // if x is 0 or non-canonical if (C1 == 0) { if (exp < 0) exp = 0; res = x_sign | (((UINT64) exp + 398) << 53); BID_RETURN (res); } // x is a finite non-zero number (not 0, non-canonical, or special) // return 0 if (exp <= -p) if (exp <= -16) { res = x_sign | 0x31c0000000000000ull; BID_RETURN (res); } // q = nr. of decimal digits in x (1 <= q <= 54) // determine first the nr. of bits in x if (C1 >= 0x0020000000000000ull) { // x >= 2^53 q = 16; } else { // if x < 2^53 tmp1.d = (double) C1; // exact conversion x_nr_bits = 1 + ((((unsigned int) (tmp1.ui64 >> 52)) & 0x7ff) - 0x3ff); q = nr_digits[x_nr_bits - 1].digits; if (q == 0) { q = nr_digits[x_nr_bits - 1].digits1; if (C1 >= nr_digits[x_nr_bits - 1].threshold_lo) q++; } } if (exp >= 0) { // -exp <= 0 // the argument is an integer already res = x; BID_RETURN (res); } else if ((q + exp) >= 0) { // exp < 0 and 1 <= -exp <= q // need to shift right -exp digits from the coefficient; the exp will be 0 ind = -exp; // 1 <= ind <= 16; ind is a synonym for 'x' // chop off ind digits from the lower part of C1 // C1 fits in 127 bits // calculate C* and f* // C* is actually floor(C*) in this case // C* and f* need shifting and masking, as shown by // shiftright128[] and maskhigh128[] // 1 <= x <= 16 // kx = 10^(-x) = ten2mk64[ind - 1] // C* = C1 * 10^(-x) // the approximation of 10^(-x) was rounded up to 64 bits __mul_64x64_to_128 (P128, C1, ten2mk64[ind - 1]); // C* = floor(C*) (logical right shift; C has p decimal digits, // correct by Property 1) // if (0 < f* < 10^(-x)) then the result is exact // n = C* * 10^(e+x) if (ind - 1 <= 2) { // 0 <= ind - 1 <= 2 => shift = 0 res = P128.w[1]; // redundant fstar.w[1] = 0; // redundant fstar.w[0] = P128.w[0]; } else if (ind - 1 <= 21) { // 3 <= ind - 1 <= 21 => 3 <= shift <= 63 shift = shiftright128[ind - 1]; // 3 <= shift <= 63 res = (P128.w[1] >> shift); // redundant fstar.w[1] = P128.w[1] & maskhigh128[ind - 1]; // redundant fstar.w[0] = P128.w[0]; } // if (f* > 10^(-x)) then the result is inexact // if ((fstar.w[1] != 0) || (fstar.w[0] >= ten2mk64[ind-1])){ // // redundant // } // set exponent to zero as it was negative before. res = x_sign | 0x31c0000000000000ull | res; BID_RETURN (res); } else { // if exp < 0 and q + exp < 0 // the result is +0 or -0 res = x_sign | 0x31c0000000000000ull; BID_RETURN (res); } } /***************************************************************************** * BID64_round_integral_nearest_away ****************************************************************************/ #if DECIMAL_CALL_BY_REFERENCE void bid64_round_integral_nearest_away (UINT64 * pres, UINT64 * px _EXC_FLAGS_PARAM _EXC_MASKS_PARAM _EXC_INFO_PARAM) { UINT64 x = *px; #else UINT64 bid64_round_integral_nearest_away (UINT64 x _EXC_FLAGS_PARAM _EXC_MASKS_PARAM _EXC_INFO_PARAM) { #endif UINT64 res = 0xbaddbaddbaddbaddull; UINT64 x_sign; int exp; // unbiased exponent // Note: C1.w[1], C1.w[0] represent x_signif_hi, x_signif_lo (all are UINT64) BID_UI64DOUBLE tmp1; int x_nr_bits; int q, ind, shift; UINT64 C1; UINT128 P128; x_sign = x & MASK_SIGN; // 0 for positive, MASK_SIGN for negative // check for NaNs and infinities if ((x & MASK_NAN) == MASK_NAN) { // check for NaN if ((x & 0x0003ffffffffffffull) > 999999999999999ull) x = x & 0xfe00000000000000ull; // clear G6-G12 and the payload bits else x = x & 0xfe03ffffffffffffull; // clear G6-G12 if ((x & MASK_SNAN) == MASK_SNAN) { // SNaN // set invalid flag *pfpsf |= INVALID_EXCEPTION; // return quiet (SNaN) res = x & 0xfdffffffffffffffull; } else { // QNaN res = x; } BID_RETURN (res); } else if ((x & MASK_INF) == MASK_INF) { // check for Infinity res = x_sign | 0x7800000000000000ull; BID_RETURN (res); } // unpack x if ((x & MASK_STEERING_BITS) == MASK_STEERING_BITS) { // if the steering bits are 11 (condition will be 0), then // the exponent is G[0:w+1] exp = ((x & MASK_BINARY_EXPONENT2) >> 51) - 398; C1 = (x & MASK_BINARY_SIG2) | MASK_BINARY_OR2; if (C1 > 9999999999999999ull) { // non-canonical C1 = 0; } } else { // if ((x & MASK_STEERING_BITS) != MASK_STEERING_BITS) exp = ((x & MASK_BINARY_EXPONENT1) >> 53) - 398; C1 = (x & MASK_BINARY_SIG1); } // if x is 0 or non-canonical if (C1 == 0) { if (exp < 0) exp = 0; res = x_sign | (((UINT64) exp + 398) << 53); BID_RETURN (res); } // x is a finite non-zero number (not 0, non-canonical, or special) // return 0 if (exp <= -(p+1)) if (exp <= -17) { res = x_sign | 0x31c0000000000000ull; BID_RETURN (res); } // q = nr. of decimal digits in x (1 <= q <= 54) // determine first the nr. of bits in x if (C1 >= 0x0020000000000000ull) { // x >= 2^53 q = 16; } else { // if x < 2^53 tmp1.d = (double) C1; // exact conversion x_nr_bits = 1 + ((((unsigned int) (tmp1.ui64 >> 52)) & 0x7ff) - 0x3ff); q = nr_digits[x_nr_bits - 1].digits; if (q == 0) { q = nr_digits[x_nr_bits - 1].digits1; if (C1 >= nr_digits[x_nr_bits - 1].threshold_lo) q++; } } if (exp >= 0) { // -exp <= 0 // the argument is an integer already res = x; BID_RETURN (res); } else if ((q + exp) >= 0) { // exp < 0 and 1 <= -exp <= q // need to shift right -exp digits from the coefficient; the exp will be 0 ind = -exp; // 1 <= ind <= 16; ind is a synonym for 'x' // chop off ind digits from the lower part of C1 // C1 = C1 + 1/2 * 10^x where the result C1 fits in 64 bits // FOR ROUND_TO_NEAREST, WE ADD 1/2 ULP(y) then truncate C1 = C1 + midpoint64[ind - 1]; // calculate C* and f* // C* is actually floor(C*) in this case // C* and f* need shifting and masking, as shown by // shiftright128[] and maskhigh128[] // 1 <= x <= 16 // kx = 10^(-x) = ten2mk64[ind - 1] // C* = (C1 + 1/2 * 10^x) * 10^(-x) // the approximation of 10^(-x) was rounded up to 64 bits __mul_64x64_to_128 (P128, C1, ten2mk64[ind - 1]); // if (0 < f* < 10^(-x)) then the result is a midpoint // C* = floor(C*) - logical right shift; C* has p decimal digits, // correct by Prop. 1) // else // C* = floor(C*) (logical right shift; C has p decimal digits, // correct by Property 1) // n = C* * 10^(e+x) if (ind - 1 <= 2) { // 0 <= ind - 1 <= 2 => shift = 0 res = P128.w[1]; } else if (ind - 1 <= 21) { // 3 <= ind - 1 <= 21 => 3 <= shift <= 63 shift = shiftright128[ind - 1]; // 3 <= shift <= 63 res = (P128.w[1] >> shift); } // midpoints are already rounded correctly // set exponent to zero as it was negative before. res = x_sign | 0x31c0000000000000ull | res; BID_RETURN (res); } else { // if exp < 0 and q + exp < 0 // the result is +0 or -0 res = x_sign | 0x31c0000000000000ull; BID_RETURN (res); } }
Go to most recent revision | Compare with Previous | Blame | View Log