/* Software floating-point emulation. Basic two-word fraction declaration and manipulation. Copyright (C) 1997,1998,1999 Free Software Foundation, Inc. This file is part of the GNU C Library. Contributed by Richard Henderson (rth@cygnus.com), Jakub Jelinek (jj@ultra.linux.cz), David S. Miller (davem@redhat.com) and Peter Maydell (pmaydell@chiark.greenend.org.uk).
The GNU C Library is free software; you can redistribute it and/or modify it under the terms of the GNU Library General Public License as published by the Free Software Foundation; either version 2 of the License, or (at your option) any later version.
The GNU C Library 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 Library General Public License for more details.
You should have received a copy of the GNU Library General Public License along with the GNU C Library; see the file COPYING.LIB. If not, write to the Free Software Foundation, Inc.,
59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. */
/* Given a 1W * 1W => 2W primitive, do the extended multiplication. */
#define _FP_MUL_MEAT_2_wide(wfracbits, R, X, Y, doit) \ do { \
_FP_FRAC_DECL_4(_z); _FP_FRAC_DECL_2(_b); _FP_FRAC_DECL_2(_c); \
\
doit(_FP_FRAC_WORD_4(_z,1), _FP_FRAC_WORD_4(_z,0), X##_f0, Y##_f0); \
doit(_b_f1, _b_f0, X##_f0, Y##_f1); \
doit(_c_f1, _c_f0, X##_f1, Y##_f0); \
doit(_FP_FRAC_WORD_4(_z,3), _FP_FRAC_WORD_4(_z,2), X##_f1, Y##_f1); \
\
__FP_FRAC_ADD_3(_FP_FRAC_WORD_4(_z,3),_FP_FRAC_WORD_4(_z,2), \
_FP_FRAC_WORD_4(_z,1), 0, _b_f1, _b_f0, \
_FP_FRAC_WORD_4(_z,3),_FP_FRAC_WORD_4(_z,2), \
_FP_FRAC_WORD_4(_z,1)); \
__FP_FRAC_ADD_3(_FP_FRAC_WORD_4(_z,3),_FP_FRAC_WORD_4(_z,2), \
_FP_FRAC_WORD_4(_z,1), 0, _c_f1, _c_f0, \
_FP_FRAC_WORD_4(_z,3),_FP_FRAC_WORD_4(_z,2), \
_FP_FRAC_WORD_4(_z,1)); \
\ /* Normalize since we know where the msb of the multiplicands \ were (bit B), we know that the msb of the of the product is \
at either 2B or 2B-1. */
_FP_FRAC_SRS_4(_z, wfracbits-1, 2*wfracbits); \
R##_f0 = _FP_FRAC_WORD_4(_z,0); \
R##_f1 = _FP_FRAC_WORD_4(_z,1); \
} while (0)
/* Given a 1W * 1W => 2W primitive, do the extended multiplication. Do only 3 multiplications instead of four. This one is for machines
where multiplication is much more expensive than subtraction. */
#define _FP_MUL_MEAT_2_wide_3mul(wfracbits, R, X, Y, doit) \ do { \
_FP_FRAC_DECL_4(_z); _FP_FRAC_DECL_2(_b); _FP_FRAC_DECL_2(_c); \
_FP_W_TYPE _d; \ int _c1, _c2; \
\
_b_f0 = X##_f0 + X##_f1; \
_c1 = _b_f0 < X##_f0; \
_b_f1 = Y##_f0 + Y##_f1; \
_c2 = _b_f1 < Y##_f0; \
doit(_d, _FP_FRAC_WORD_4(_z,0), X##_f0, Y##_f0); \
doit(_FP_FRAC_WORD_4(_z,2), _FP_FRAC_WORD_4(_z,1), _b_f0, _b_f1); \
doit(_c_f1, _c_f0, X##_f1, Y##_f1); \
\
_b_f0 &= -_c2; \
_b_f1 &= -_c1; \
__FP_FRAC_ADD_3(_FP_FRAC_WORD_4(_z,3),_FP_FRAC_WORD_4(_z,2), \
_FP_FRAC_WORD_4(_z,1), (_c1 & _c2), 0, _d, \
0, _FP_FRAC_WORD_4(_z,2), _FP_FRAC_WORD_4(_z,1)); \
__FP_FRAC_ADDI_2(_FP_FRAC_WORD_4(_z,3),_FP_FRAC_WORD_4(_z,2), \
_b_f0); \
__FP_FRAC_ADDI_2(_FP_FRAC_WORD_4(_z,3),_FP_FRAC_WORD_4(_z,2), \
_b_f1); \
__FP_FRAC_DEC_3(_FP_FRAC_WORD_4(_z,3),_FP_FRAC_WORD_4(_z,2), \
_FP_FRAC_WORD_4(_z,1), \
0, _d, _FP_FRAC_WORD_4(_z,0)); \
__FP_FRAC_DEC_3(_FP_FRAC_WORD_4(_z,3),_FP_FRAC_WORD_4(_z,2), \
_FP_FRAC_WORD_4(_z,1), 0, _c_f1, _c_f0); \
__FP_FRAC_ADD_2(_FP_FRAC_WORD_4(_z,3), _FP_FRAC_WORD_4(_z,2), \
_c_f1, _c_f0, \
_FP_FRAC_WORD_4(_z,3), _FP_FRAC_WORD_4(_z,2)); \
\ /* Normalize since we know where the msb of the multiplicands \ were (bit B), we know that the msb of the of the product is \
at either 2B or 2B-1. */
_FP_FRAC_SRS_4(_z, wfracbits-1, 2*wfracbits); \
R##_f0 = _FP_FRAC_WORD_4(_z,0); \
R##_f1 = _FP_FRAC_WORD_4(_z,1); \
} while (0)
#define _FP_MUL_MEAT_2_gmp(wfracbits, R, X, Y) \ do { \
_FP_FRAC_DECL_4(_z); \
_FP_W_TYPE _x[2], _y[2]; \
_x[0] = X##_f0; _x[1] = X##_f1; \
_y[0] = Y##_f0; _y[1] = Y##_f1; \
\
mpn_mul_n(_z_f, _x, _y, 2); \
\ /* Normalize since we know where the msb of the multiplicands \ were (bit B), we know that the msb of the of the product is \
at either 2B or 2B-1. */
_FP_FRAC_SRS_4(_z, wfracbits-1, 2*wfracbits); \
R##_f0 = _z_f[0]; \
R##_f1 = _z_f[1]; \
} while (0)
/* Do at most 120x120=240 bits multiplication using double floating point multiplication. This is useful if floating point multiplication has much bigger throughput than integer multiply. It is supposed to work for _FP_W_TYPE_SIZE 64 and wfracbits between 106 and 120 only. Caller guarantees that X and Y has (1LLL << (wfracbits - 1)) set. SETFETZ is a macro which will disable all FPU exceptions and set rounding
towards zero, RESETFE should optionally reset it back. */
#define _FP_DIV_MEAT_2_udiv(fs, R, X, Y) \ do { \
_FP_W_TYPE _n_f2, _n_f1, _n_f0, _r_f1, _r_f0, _m_f1, _m_f0; \ if (_FP_FRAC_GT_2(X, Y)) \
{ \
_n_f2 = X##_f1 >> 1; \
_n_f1 = X##_f1 << (_FP_W_TYPE_SIZE - 1) | X##_f0 >> 1; \
_n_f0 = X##_f0 << (_FP_W_TYPE_SIZE - 1); \
} \ else \
{ \
R##_e--; \
_n_f2 = X##_f1; \
_n_f1 = X##_f0; \
_n_f0 = 0; \
} \
\ /* Normalize, i.e. make the most significant bit of the \
denominator set. */
_FP_FRAC_SLL_2(Y, _FP_WFRACXBITS_##fs); \
\
udiv_qrnnd(R##_f1, _r_f1, _n_f2, _n_f1, Y##_f1); \
umul_ppmm(_m_f1, _m_f0, R##_f1, Y##_f0); \
_r_f0 = _n_f0; \ if (_FP_FRAC_GT_2(_m, _r)) \
{ \
R##_f1--; \
_FP_FRAC_ADD_2(_r, Y, _r); \ if (_FP_FRAC_GE_2(_r, Y) && _FP_FRAC_GT_2(_m, _r)) \
{ \
R##_f1--; \
_FP_FRAC_ADD_2(_r, Y, _r); \
} \
} \
_FP_FRAC_DEC_2(_r, _m); \
\ if (_r_f1 == Y##_f1) \
{ \ /* This is a special case, not an optimization \ (_r/Y##_f1 would not fit into UWtype). \ As _r is guaranteed to be < Y, R##_f0 can be either \ (UWtype)-1 or (UWtype)-2. But as we know what kind \ of bits it is (sticky, guard, round), we don't care. \ We also don't care what the reminder is, because the \
guard bit will be set anyway. -jj */
R##_f0 = -1; \
} \ else \
{ \
udiv_qrnnd(R##_f0, _r_f1, _r_f1, _r_f0, Y##_f1); \
umul_ppmm(_m_f1, _m_f0, R##_f0, Y##_f0); \
_r_f0 = 0; \ if (_FP_FRAC_GT_2(_m, _r)) \
{ \
R##_f0--; \
_FP_FRAC_ADD_2(_r, Y, _r); \ if (_FP_FRAC_GE_2(_r, Y) && _FP_FRAC_GT_2(_m, _r)) \
{ \
R##_f0--; \
_FP_FRAC_ADD_2(_r, Y, _r); \
} \
} \ if (!_FP_FRAC_EQ_2(_r, _m)) \
R##_f0 |= _FP_WORK_STICKY; \
} \
} while (0)
Die Informationen auf dieser Webseite wurden
nach bestem Wissen sorgfältig zusammengestellt. Es wird jedoch weder Vollständigkeit, noch Richtigkeit,
noch Qualität der bereit gestellten Informationen zugesichert.
Bemerkung:
Die farbliche Syntaxdarstellung und die Messung sind noch experimentell.