cachepc-linux

Fork of AMDESE/linux with modifications for CachePC side-channel attack
git clone https://git.sinitax.com/sinitax/cachepc-linux
Log | Files | Refs | README | LICENSE | sfeed.txt

op-1.h (9449B)


      1/* Software floating-point emulation.
      2   Basic one-word fraction declaration and manipulation.
      3   Copyright (C) 1997,1998,1999 Free Software Foundation, Inc.
      4   This file is part of the GNU C Library.
      5   Contributed by Richard Henderson (rth@cygnus.com),
      6		  Jakub Jelinek (jj@ultra.linux.cz),
      7		  David S. Miller (davem@redhat.com) and
      8		  Peter Maydell (pmaydell@chiark.greenend.org.uk).
      9
     10   The GNU C Library is free software; you can redistribute it and/or
     11   modify it under the terms of the GNU Library General Public License as
     12   published by the Free Software Foundation; either version 2 of the
     13   License, or (at your option) any later version.
     14
     15   The GNU C Library is distributed in the hope that it will be useful,
     16   but WITHOUT ANY WARRANTY; without even the implied warranty of
     17   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
     18   Library General Public License for more details.
     19
     20   You should have received a copy of the GNU Library General Public
     21   License along with the GNU C Library; see the file COPYING.LIB.  If
     22   not, write to the Free Software Foundation, Inc.,
     23   59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.  */
     24
     25#ifndef    __MATH_EMU_OP_1_H__
     26#define    __MATH_EMU_OP_1_H__
     27
     28#define _FP_FRAC_DECL_1(X)	_FP_W_TYPE X##_f=0
     29#define _FP_FRAC_COPY_1(D,S)	(D##_f = S##_f)
     30#define _FP_FRAC_SET_1(X,I)	(X##_f = I)
     31#define _FP_FRAC_HIGH_1(X)	(X##_f)
     32#define _FP_FRAC_LOW_1(X)	(X##_f)
     33#define _FP_FRAC_WORD_1(X,w)	(X##_f)
     34
     35#define _FP_FRAC_ADDI_1(X,I)	(X##_f += I)
     36#define _FP_FRAC_SLL_1(X,N)			\
     37  do {						\
     38    if (__builtin_constant_p(N) && (N) == 1)	\
     39      X##_f += X##_f;				\
     40    else					\
     41      X##_f <<= (N);				\
     42  } while (0)
     43#define _FP_FRAC_SRL_1(X,N)	(X##_f >>= N)
     44
     45/* Right shift with sticky-lsb.  */
     46#define _FP_FRAC_SRS_1(X,N,sz)	__FP_FRAC_SRS_1(X##_f, N, sz)
     47
     48#define __FP_FRAC_SRS_1(X,N,sz)						\
     49   (X = (X >> (N) | (__builtin_constant_p(N) && (N) == 1		\
     50		     ? X & 1 : (X << (_FP_W_TYPE_SIZE - (N))) != 0)))
     51
     52#define _FP_FRAC_ADD_1(R,X,Y)	(R##_f = X##_f + Y##_f)
     53#define _FP_FRAC_SUB_1(R,X,Y)	(R##_f = X##_f - Y##_f)
     54#define _FP_FRAC_DEC_1(X,Y)	(X##_f -= Y##_f)
     55#define _FP_FRAC_CLZ_1(z, X)	__FP_CLZ(z, X##_f)
     56
     57/* Predicates */
     58#define _FP_FRAC_NEGP_1(X)	((_FP_WS_TYPE)X##_f < 0)
     59#define _FP_FRAC_ZEROP_1(X)	(X##_f == 0)
     60#define _FP_FRAC_OVERP_1(fs,X)	(X##_f & _FP_OVERFLOW_##fs)
     61#define _FP_FRAC_CLEAR_OVERP_1(fs,X)	(X##_f &= ~_FP_OVERFLOW_##fs)
     62#define _FP_FRAC_EQ_1(X, Y)	(X##_f == Y##_f)
     63#define _FP_FRAC_GE_1(X, Y)	(X##_f >= Y##_f)
     64#define _FP_FRAC_GT_1(X, Y)	(X##_f > Y##_f)
     65
     66#define _FP_ZEROFRAC_1		0
     67#define _FP_MINFRAC_1		1
     68#define _FP_MAXFRAC_1		(~(_FP_WS_TYPE)0)
     69
     70/*
     71 * Unpack the raw bits of a native fp value.  Do not classify or
     72 * normalize the data.
     73 */
     74
     75#define _FP_UNPACK_RAW_1(fs, X, val)				\
     76  do {								\
     77    union _FP_UNION_##fs _flo; _flo.flt = (val);		\
     78								\
     79    X##_f = _flo.bits.frac;					\
     80    X##_e = _flo.bits.exp;					\
     81    X##_s = _flo.bits.sign;					\
     82  } while (0)
     83
     84#define _FP_UNPACK_RAW_1_P(fs, X, val)				\
     85  do {								\
     86    union _FP_UNION_##fs *_flo =				\
     87      (union _FP_UNION_##fs *)(val);				\
     88								\
     89    X##_f = _flo->bits.frac;					\
     90    X##_e = _flo->bits.exp;					\
     91    X##_s = _flo->bits.sign;					\
     92  } while (0)
     93
     94/*
     95 * Repack the raw bits of a native fp value.
     96 */
     97
     98#define _FP_PACK_RAW_1(fs, val, X)				\
     99  do {								\
    100    union _FP_UNION_##fs _flo;					\
    101								\
    102    _flo.bits.frac = X##_f;					\
    103    _flo.bits.exp  = X##_e;					\
    104    _flo.bits.sign = X##_s;					\
    105								\
    106    (val) = _flo.flt;						\
    107  } while (0)
    108
    109#define _FP_PACK_RAW_1_P(fs, val, X)				\
    110  do {								\
    111    union _FP_UNION_##fs *_flo =				\
    112      (union _FP_UNION_##fs *)(val);				\
    113								\
    114    _flo->bits.frac = X##_f;					\
    115    _flo->bits.exp  = X##_e;					\
    116    _flo->bits.sign = X##_s;					\
    117  } while (0)
    118
    119
    120/*
    121 * Multiplication algorithms:
    122 */
    123
    124/* Basic.  Assuming the host word size is >= 2*FRACBITS, we can do the
    125   multiplication immediately.  */
    126
    127#define _FP_MUL_MEAT_1_imm(wfracbits, R, X, Y)				\
    128  do {									\
    129    R##_f = X##_f * Y##_f;						\
    130    /* Normalize since we know where the msb of the multiplicands	\
    131       were (bit B), we know that the msb of the of the product is	\
    132       at either 2B or 2B-1.  */					\
    133    _FP_FRAC_SRS_1(R, wfracbits-1, 2*wfracbits);			\
    134  } while (0)
    135
    136/* Given a 1W * 1W => 2W primitive, do the extended multiplication.  */
    137
    138#define _FP_MUL_MEAT_1_wide(wfracbits, R, X, Y, doit)			\
    139  do {									\
    140    _FP_W_TYPE _Z_f0, _Z_f1;						\
    141    doit(_Z_f1, _Z_f0, X##_f, Y##_f);					\
    142    /* Normalize since we know where the msb of the multiplicands	\
    143       were (bit B), we know that the msb of the of the product is	\
    144       at either 2B or 2B-1.  */					\
    145    _FP_FRAC_SRS_2(_Z, wfracbits-1, 2*wfracbits);			\
    146    R##_f = _Z_f0;							\
    147  } while (0)
    148
    149/* Finally, a simple widening multiply algorithm.  What fun!  */
    150
    151#define _FP_MUL_MEAT_1_hard(wfracbits, R, X, Y)				\
    152  do {									\
    153    _FP_W_TYPE _xh, _xl, _yh, _yl, _z_f0, _z_f1, _a_f0, _a_f1;		\
    154									\
    155    /* split the words in half */					\
    156    _xh = X##_f >> (_FP_W_TYPE_SIZE/2);					\
    157    _xl = X##_f & (((_FP_W_TYPE)1 << (_FP_W_TYPE_SIZE/2)) - 1);		\
    158    _yh = Y##_f >> (_FP_W_TYPE_SIZE/2);					\
    159    _yl = Y##_f & (((_FP_W_TYPE)1 << (_FP_W_TYPE_SIZE/2)) - 1);		\
    160									\
    161    /* multiply the pieces */						\
    162    _z_f0 = _xl * _yl;							\
    163    _a_f0 = _xh * _yl;							\
    164    _a_f1 = _xl * _yh;							\
    165    _z_f1 = _xh * _yh;							\
    166									\
    167    /* reassemble into two full words */				\
    168    if ((_a_f0 += _a_f1) < _a_f1)					\
    169      _z_f1 += (_FP_W_TYPE)1 << (_FP_W_TYPE_SIZE/2);			\
    170    _a_f1 = _a_f0 >> (_FP_W_TYPE_SIZE/2);				\
    171    _a_f0 = _a_f0 << (_FP_W_TYPE_SIZE/2);				\
    172    _FP_FRAC_ADD_2(_z, _z, _a);						\
    173									\
    174    /* normalize */							\
    175    _FP_FRAC_SRS_2(_z, wfracbits - 1, 2*wfracbits);			\
    176    R##_f = _z_f0;							\
    177  } while (0)
    178
    179
    180/*
    181 * Division algorithms:
    182 */
    183
    184/* Basic.  Assuming the host word size is >= 2*FRACBITS, we can do the
    185   division immediately.  Give this macro either _FP_DIV_HELP_imm for
    186   C primitives or _FP_DIV_HELP_ldiv for the ISO function.  Which you
    187   choose will depend on what the compiler does with divrem4.  */
    188
    189#define _FP_DIV_MEAT_1_imm(fs, R, X, Y, doit)		\
    190  do {							\
    191    _FP_W_TYPE _q, _r;					\
    192    X##_f <<= (X##_f < Y##_f				\
    193	       ? R##_e--, _FP_WFRACBITS_##fs		\
    194	       : _FP_WFRACBITS_##fs - 1);		\
    195    doit(_q, _r, X##_f, Y##_f);				\
    196    R##_f = _q | (_r != 0);				\
    197  } while (0)
    198
    199/* GCC's longlong.h defines a 2W / 1W => (1W,1W) primitive udiv_qrnnd
    200   that may be useful in this situation.  This first is for a primitive
    201   that requires normalization, the second for one that does not.  Look
    202   for UDIV_NEEDS_NORMALIZATION to tell which your machine needs.  */
    203
    204#define _FP_DIV_MEAT_1_udiv_norm(fs, R, X, Y)				\
    205  do {									\
    206    _FP_W_TYPE _nh, _nl, _q, _r, _y;					\
    207									\
    208    /* Normalize Y -- i.e. make the most significant bit set.  */	\
    209    _y = Y##_f << _FP_WFRACXBITS_##fs;					\
    210									\
    211    /* Shift X op correspondingly high, that is, up one full word.  */	\
    212    if (X##_f < Y##_f)							\
    213      {									\
    214	R##_e--;							\
    215	_nl = 0;							\
    216	_nh = X##_f;							\
    217      }									\
    218    else								\
    219      {									\
    220	_nl = X##_f << (_FP_W_TYPE_SIZE - 1);				\
    221	_nh = X##_f >> 1;						\
    222      }									\
    223    									\
    224    udiv_qrnnd(_q, _r, _nh, _nl, _y);					\
    225    R##_f = _q | (_r != 0);						\
    226  } while (0)
    227
    228#define _FP_DIV_MEAT_1_udiv(fs, R, X, Y)		\
    229  do {							\
    230    _FP_W_TYPE _nh, _nl, _q, _r;			\
    231    if (X##_f < Y##_f)					\
    232      {							\
    233	R##_e--;					\
    234	_nl = X##_f << _FP_WFRACBITS_##fs;		\
    235	_nh = X##_f >> _FP_WFRACXBITS_##fs;		\
    236      }							\
    237    else						\
    238      {							\
    239	_nl = X##_f << (_FP_WFRACBITS_##fs - 1);	\
    240	_nh = X##_f >> (_FP_WFRACXBITS_##fs + 1);	\
    241      }							\
    242    udiv_qrnnd(_q, _r, _nh, _nl, Y##_f);		\
    243    R##_f = _q | (_r != 0);				\
    244  } while (0)
    245  
    246  
    247/*
    248 * Square root algorithms:
    249 * We have just one right now, maybe Newton approximation
    250 * should be added for those machines where division is fast.
    251 */
    252 
    253#define _FP_SQRT_MEAT_1(R, S, T, X, q)			\
    254  do {							\
    255    while (q != _FP_WORK_ROUND)				\
    256      {							\
    257        T##_f = S##_f + q;				\
    258        if (T##_f <= X##_f)				\
    259          {						\
    260            S##_f = T##_f + q;				\
    261            X##_f -= T##_f;				\
    262            R##_f += q;					\
    263          }						\
    264        _FP_FRAC_SLL_1(X, 1);				\
    265        q >>= 1;					\
    266      }							\
    267    if (X##_f)						\
    268      {							\
    269	if (S##_f < X##_f)				\
    270	  R##_f |= _FP_WORK_ROUND;			\
    271	R##_f |= _FP_WORK_STICKY;			\
    272      }							\
    273  } while (0)
    274
    275/*
    276 * Assembly/disassembly for converting to/from integral types.  
    277 * No shifting or overflow handled here.
    278 */
    279
    280#define _FP_FRAC_ASSEMBLE_1(r, X, rsize)	(r = X##_f)
    281#define _FP_FRAC_DISASSEMBLE_1(X, r, rsize)	(X##_f = r)
    282
    283
    284/*
    285 * Convert FP values between word sizes
    286 */
    287
    288#define _FP_FRAC_CONV_1_1(dfs, sfs, D, S)				\
    289  do {									\
    290    D##_f = S##_f;							\
    291    if (_FP_WFRACBITS_##sfs > _FP_WFRACBITS_##dfs)			\
    292      {									\
    293	if (S##_c != FP_CLS_NAN)					\
    294	  _FP_FRAC_SRS_1(D, (_FP_WFRACBITS_##sfs-_FP_WFRACBITS_##dfs),	\
    295			 _FP_WFRACBITS_##sfs);				\
    296	else								\
    297	  _FP_FRAC_SRL_1(D, (_FP_WFRACBITS_##sfs-_FP_WFRACBITS_##dfs));	\
    298      }									\
    299    else								\
    300      D##_f <<= _FP_WFRACBITS_##dfs - _FP_WFRACBITS_##sfs;		\
    301  } while (0)
    302
    303#endif /* __MATH_EMU_OP_1_H__ */