blob: ba226f8237a696cd7310d81a8cbe9bc1e5e75ebf [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/* Software floating-point emulation.
2 Basic four-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_4_H__
26#define __MATH_EMU_OP_4_H__
27
28#define _FP_FRAC_DECL_4(X) _FP_W_TYPE X##_f[4]
29#define _FP_FRAC_COPY_4(D,S) \
30 (D##_f[0] = S##_f[0], D##_f[1] = S##_f[1], \
31 D##_f[2] = S##_f[2], D##_f[3] = S##_f[3])
32#define _FP_FRAC_SET_4(X,I) __FP_FRAC_SET_4(X, I)
33#define _FP_FRAC_HIGH_4(X) (X##_f[3])
34#define _FP_FRAC_LOW_4(X) (X##_f[0])
35#define _FP_FRAC_WORD_4(X,w) (X##_f[w])
36
37#define _FP_FRAC_SLL_4(X,N) \
38 do { \
39 _FP_I_TYPE _up, _down, _skip, _i; \
40 _skip = (N) / _FP_W_TYPE_SIZE; \
41 _up = (N) % _FP_W_TYPE_SIZE; \
42 _down = _FP_W_TYPE_SIZE - _up; \
43 if (!_up) \
44 for (_i = 3; _i >= _skip; --_i) \
45 X##_f[_i] = X##_f[_i-_skip]; \
46 else \
47 { \
48 for (_i = 3; _i > _skip; --_i) \
49 X##_f[_i] = X##_f[_i-_skip] << _up \
50 | X##_f[_i-_skip-1] >> _down; \
51 X##_f[_i--] = X##_f[0] << _up; \
52 } \
53 for (; _i >= 0; --_i) \
54 X##_f[_i] = 0; \
55 } while (0)
56
57/* This one was broken too */
58#define _FP_FRAC_SRL_4(X,N) \
59 do { \
60 _FP_I_TYPE _up, _down, _skip, _i; \
61 _skip = (N) / _FP_W_TYPE_SIZE; \
62 _down = (N) % _FP_W_TYPE_SIZE; \
63 _up = _FP_W_TYPE_SIZE - _down; \
64 if (!_down) \
65 for (_i = 0; _i <= 3-_skip; ++_i) \
66 X##_f[_i] = X##_f[_i+_skip]; \
67 else \
68 { \
69 for (_i = 0; _i < 3-_skip; ++_i) \
70 X##_f[_i] = X##_f[_i+_skip] >> _down \
71 | X##_f[_i+_skip+1] << _up; \
72 X##_f[_i++] = X##_f[3] >> _down; \
73 } \
74 for (; _i < 4; ++_i) \
75 X##_f[_i] = 0; \
76 } while (0)
77
78
79/* Right shift with sticky-lsb.
80 * What this actually means is that we do a standard right-shift,
81 * but that if any of the bits that fall off the right hand side
82 * were one then we always set the LSbit.
83 */
84#define _FP_FRAC_SRS_4(X,N,size) \
85 do { \
86 _FP_I_TYPE _up, _down, _skip, _i; \
87 _FP_W_TYPE _s; \
88 _skip = (N) / _FP_W_TYPE_SIZE; \
89 _down = (N) % _FP_W_TYPE_SIZE; \
90 _up = _FP_W_TYPE_SIZE - _down; \
91 for (_s = _i = 0; _i < _skip; ++_i) \
92 _s |= X##_f[_i]; \
93 _s |= X##_f[_i] << _up; \
94/* s is now != 0 if we want to set the LSbit */ \
95 if (!_down) \
96 for (_i = 0; _i <= 3-_skip; ++_i) \
97 X##_f[_i] = X##_f[_i+_skip]; \
98 else \
99 { \
100 for (_i = 0; _i < 3-_skip; ++_i) \
101 X##_f[_i] = X##_f[_i+_skip] >> _down \
102 | X##_f[_i+_skip+1] << _up; \
103 X##_f[_i++] = X##_f[3] >> _down; \
104 } \
105 for (; _i < 4; ++_i) \
106 X##_f[_i] = 0; \
107 /* don't fix the LSB until the very end when we're sure f[0] is stable */ \
108 X##_f[0] |= (_s != 0); \
109 } while (0)
110
111#define _FP_FRAC_ADD_4(R,X,Y) \
112 __FP_FRAC_ADD_4(R##_f[3], R##_f[2], R##_f[1], R##_f[0], \
113 X##_f[3], X##_f[2], X##_f[1], X##_f[0], \
114 Y##_f[3], Y##_f[2], Y##_f[1], Y##_f[0])
115
116#define _FP_FRAC_SUB_4(R,X,Y) \
117 __FP_FRAC_SUB_4(R##_f[3], R##_f[2], R##_f[1], R##_f[0], \
118 X##_f[3], X##_f[2], X##_f[1], X##_f[0], \
119 Y##_f[3], Y##_f[2], Y##_f[1], Y##_f[0])
120
121#define _FP_FRAC_DEC_4(X,Y) \
122 __FP_FRAC_DEC_4(X##_f[3], X##_f[2], X##_f[1], X##_f[0], \
123 Y##_f[3], Y##_f[2], Y##_f[1], Y##_f[0])
124
125#define _FP_FRAC_ADDI_4(X,I) \
126 __FP_FRAC_ADDI_4(X##_f[3], X##_f[2], X##_f[1], X##_f[0], I)
127
128#define _FP_ZEROFRAC_4 0,0,0,0
129#define _FP_MINFRAC_4 0,0,0,1
130#define _FP_MAXFRAC_4 (~(_FP_WS_TYPE)0), (~(_FP_WS_TYPE)0), (~(_FP_WS_TYPE)0), (~(_FP_WS_TYPE)0)
131
132#define _FP_FRAC_ZEROP_4(X) ((X##_f[0] | X##_f[1] | X##_f[2] | X##_f[3]) == 0)
133#define _FP_FRAC_NEGP_4(X) ((_FP_WS_TYPE)X##_f[3] < 0)
134#define _FP_FRAC_OVERP_4(fs,X) (_FP_FRAC_HIGH_##fs(X) & _FP_OVERFLOW_##fs)
135#define _FP_FRAC_CLEAR_OVERP_4(fs,X) (_FP_FRAC_HIGH_##fs(X) &= ~_FP_OVERFLOW_##fs)
136
137#define _FP_FRAC_EQ_4(X,Y) \
138 (X##_f[0] == Y##_f[0] && X##_f[1] == Y##_f[1] \
139 && X##_f[2] == Y##_f[2] && X##_f[3] == Y##_f[3])
140
141#define _FP_FRAC_GT_4(X,Y) \
142 (X##_f[3] > Y##_f[3] || \
143 (X##_f[3] == Y##_f[3] && (X##_f[2] > Y##_f[2] || \
144 (X##_f[2] == Y##_f[2] && (X##_f[1] > Y##_f[1] || \
145 (X##_f[1] == Y##_f[1] && X##_f[0] > Y##_f[0]) \
146 )) \
147 )) \
148 )
149
150#define _FP_FRAC_GE_4(X,Y) \
151 (X##_f[3] > Y##_f[3] || \
152 (X##_f[3] == Y##_f[3] && (X##_f[2] > Y##_f[2] || \
153 (X##_f[2] == Y##_f[2] && (X##_f[1] > Y##_f[1] || \
154 (X##_f[1] == Y##_f[1] && X##_f[0] >= Y##_f[0]) \
155 )) \
156 )) \
157 )
158
159
160#define _FP_FRAC_CLZ_4(R,X) \
161 do { \
162 if (X##_f[3]) \
163 { \
164 __FP_CLZ(R,X##_f[3]); \
165 } \
166 else if (X##_f[2]) \
167 { \
168 __FP_CLZ(R,X##_f[2]); \
169 R += _FP_W_TYPE_SIZE; \
170 } \
171 else if (X##_f[1]) \
172 { \
173 __FP_CLZ(R,X##_f[2]); \
174 R += _FP_W_TYPE_SIZE*2; \
175 } \
176 else \
177 { \
178 __FP_CLZ(R,X##_f[0]); \
179 R += _FP_W_TYPE_SIZE*3; \
180 } \
181 } while(0)
182
183
184#define _FP_UNPACK_RAW_4(fs, X, val) \
185 do { \
186 union _FP_UNION_##fs _flo; _flo.flt = (val); \
187 X##_f[0] = _flo.bits.frac0; \
188 X##_f[1] = _flo.bits.frac1; \
189 X##_f[2] = _flo.bits.frac2; \
190 X##_f[3] = _flo.bits.frac3; \
191 X##_e = _flo.bits.exp; \
192 X##_s = _flo.bits.sign; \
193 } while (0)
194
195#define _FP_UNPACK_RAW_4_P(fs, X, val) \
196 do { \
197 union _FP_UNION_##fs *_flo = \
198 (union _FP_UNION_##fs *)(val); \
199 \
200 X##_f[0] = _flo->bits.frac0; \
201 X##_f[1] = _flo->bits.frac1; \
202 X##_f[2] = _flo->bits.frac2; \
203 X##_f[3] = _flo->bits.frac3; \
204 X##_e = _flo->bits.exp; \
205 X##_s = _flo->bits.sign; \
206 } while (0)
207
208#define _FP_PACK_RAW_4(fs, val, X) \
209 do { \
210 union _FP_UNION_##fs _flo; \
211 _flo.bits.frac0 = X##_f[0]; \
212 _flo.bits.frac1 = X##_f[1]; \
213 _flo.bits.frac2 = X##_f[2]; \
214 _flo.bits.frac3 = X##_f[3]; \
215 _flo.bits.exp = X##_e; \
216 _flo.bits.sign = X##_s; \
217 (val) = _flo.flt; \
218 } while (0)
219
220#define _FP_PACK_RAW_4_P(fs, val, X) \
221 do { \
222 union _FP_UNION_##fs *_flo = \
223 (union _FP_UNION_##fs *)(val); \
224 \
225 _flo->bits.frac0 = X##_f[0]; \
226 _flo->bits.frac1 = X##_f[1]; \
227 _flo->bits.frac2 = X##_f[2]; \
228 _flo->bits.frac3 = X##_f[3]; \
229 _flo->bits.exp = X##_e; \
230 _flo->bits.sign = X##_s; \
231 } while (0)
232
233/*
234 * Multiplication algorithms:
235 */
236
237/* Given a 1W * 1W => 2W primitive, do the extended multiplication. */
238
239#define _FP_MUL_MEAT_4_wide(wfracbits, R, X, Y, doit) \
240 do { \
241 _FP_FRAC_DECL_8(_z); _FP_FRAC_DECL_2(_b); _FP_FRAC_DECL_2(_c); \
242 _FP_FRAC_DECL_2(_d); _FP_FRAC_DECL_2(_e); _FP_FRAC_DECL_2(_f); \
243 \
244 doit(_FP_FRAC_WORD_8(_z,1), _FP_FRAC_WORD_8(_z,0), X##_f[0], Y##_f[0]); \
245 doit(_b_f1, _b_f0, X##_f[0], Y##_f[1]); \
246 doit(_c_f1, _c_f0, X##_f[1], Y##_f[0]); \
247 doit(_d_f1, _d_f0, X##_f[1], Y##_f[1]); \
248 doit(_e_f1, _e_f0, X##_f[0], Y##_f[2]); \
249 doit(_f_f1, _f_f0, X##_f[2], Y##_f[0]); \
250 __FP_FRAC_ADD_3(_FP_FRAC_WORD_8(_z,3),_FP_FRAC_WORD_8(_z,2), \
251 _FP_FRAC_WORD_8(_z,1), 0,_b_f1,_b_f0, \
252 0,0,_FP_FRAC_WORD_8(_z,1)); \
253 __FP_FRAC_ADD_3(_FP_FRAC_WORD_8(_z,3),_FP_FRAC_WORD_8(_z,2), \
254 _FP_FRAC_WORD_8(_z,1), 0,_c_f1,_c_f0, \
255 _FP_FRAC_WORD_8(_z,3),_FP_FRAC_WORD_8(_z,2), \
256 _FP_FRAC_WORD_8(_z,1)); \
257 __FP_FRAC_ADD_3(_FP_FRAC_WORD_8(_z,4),_FP_FRAC_WORD_8(_z,3), \
258 _FP_FRAC_WORD_8(_z,2), 0,_d_f1,_d_f0, \
259 0,_FP_FRAC_WORD_8(_z,3),_FP_FRAC_WORD_8(_z,2)); \
260 __FP_FRAC_ADD_3(_FP_FRAC_WORD_8(_z,4),_FP_FRAC_WORD_8(_z,3), \
261 _FP_FRAC_WORD_8(_z,2), 0,_e_f1,_e_f0, \
262 _FP_FRAC_WORD_8(_z,4),_FP_FRAC_WORD_8(_z,3), \
263 _FP_FRAC_WORD_8(_z,2)); \
264 __FP_FRAC_ADD_3(_FP_FRAC_WORD_8(_z,4),_FP_FRAC_WORD_8(_z,3), \
265 _FP_FRAC_WORD_8(_z,2), 0,_f_f1,_f_f0, \
266 _FP_FRAC_WORD_8(_z,4),_FP_FRAC_WORD_8(_z,3), \
267 _FP_FRAC_WORD_8(_z,2)); \
268 doit(_b_f1, _b_f0, X##_f[0], Y##_f[3]); \
269 doit(_c_f1, _c_f0, X##_f[3], Y##_f[0]); \
270 doit(_d_f1, _d_f0, X##_f[1], Y##_f[2]); \
271 doit(_e_f1, _e_f0, X##_f[2], Y##_f[1]); \
272 __FP_FRAC_ADD_3(_FP_FRAC_WORD_8(_z,5),_FP_FRAC_WORD_8(_z,4), \
273 _FP_FRAC_WORD_8(_z,3), 0,_b_f1,_b_f0, \
274 0,_FP_FRAC_WORD_8(_z,4),_FP_FRAC_WORD_8(_z,3)); \
275 __FP_FRAC_ADD_3(_FP_FRAC_WORD_8(_z,5),_FP_FRAC_WORD_8(_z,4), \
276 _FP_FRAC_WORD_8(_z,3), 0,_c_f1,_c_f0, \
277 _FP_FRAC_WORD_8(_z,5),_FP_FRAC_WORD_8(_z,4), \
278 _FP_FRAC_WORD_8(_z,3)); \
279 __FP_FRAC_ADD_3(_FP_FRAC_WORD_8(_z,5),_FP_FRAC_WORD_8(_z,4), \
280 _FP_FRAC_WORD_8(_z,3), 0,_d_f1,_d_f0, \
281 _FP_FRAC_WORD_8(_z,5),_FP_FRAC_WORD_8(_z,4), \
282 _FP_FRAC_WORD_8(_z,3)); \
283 __FP_FRAC_ADD_3(_FP_FRAC_WORD_8(_z,5),_FP_FRAC_WORD_8(_z,4), \
284 _FP_FRAC_WORD_8(_z,3), 0,_e_f1,_e_f0, \
285 _FP_FRAC_WORD_8(_z,5),_FP_FRAC_WORD_8(_z,4), \
286 _FP_FRAC_WORD_8(_z,3)); \
287 doit(_b_f1, _b_f0, X##_f[2], Y##_f[2]); \
288 doit(_c_f1, _c_f0, X##_f[1], Y##_f[3]); \
289 doit(_d_f1, _d_f0, X##_f[3], Y##_f[1]); \
290 doit(_e_f1, _e_f0, X##_f[2], Y##_f[3]); \
291 doit(_f_f1, _f_f0, X##_f[3], Y##_f[2]); \
292 __FP_FRAC_ADD_3(_FP_FRAC_WORD_8(_z,6),_FP_FRAC_WORD_8(_z,5), \
293 _FP_FRAC_WORD_8(_z,4), 0,_b_f1,_b_f0, \
294 0,_FP_FRAC_WORD_8(_z,5),_FP_FRAC_WORD_8(_z,4)); \
295 __FP_FRAC_ADD_3(_FP_FRAC_WORD_8(_z,6),_FP_FRAC_WORD_8(_z,5), \
296 _FP_FRAC_WORD_8(_z,4), 0,_c_f1,_c_f0, \
297 _FP_FRAC_WORD_8(_z,6),_FP_FRAC_WORD_8(_z,5), \
298 _FP_FRAC_WORD_8(_z,4)); \
299 __FP_FRAC_ADD_3(_FP_FRAC_WORD_8(_z,6),_FP_FRAC_WORD_8(_z,5), \
300 _FP_FRAC_WORD_8(_z,4), 0,_d_f1,_d_f0, \
301 _FP_FRAC_WORD_8(_z,6),_FP_FRAC_WORD_8(_z,5), \
302 _FP_FRAC_WORD_8(_z,4)); \
303 __FP_FRAC_ADD_3(_FP_FRAC_WORD_8(_z,7),_FP_FRAC_WORD_8(_z,6), \
304 _FP_FRAC_WORD_8(_z,5), 0,_e_f1,_e_f0, \
305 0,_FP_FRAC_WORD_8(_z,6),_FP_FRAC_WORD_8(_z,5)); \
306 __FP_FRAC_ADD_3(_FP_FRAC_WORD_8(_z,7),_FP_FRAC_WORD_8(_z,6), \
307 _FP_FRAC_WORD_8(_z,5), 0,_f_f1,_f_f0, \
308 _FP_FRAC_WORD_8(_z,7),_FP_FRAC_WORD_8(_z,6), \
309 _FP_FRAC_WORD_8(_z,5)); \
310 doit(_b_f1, _b_f0, X##_f[3], Y##_f[3]); \
311 __FP_FRAC_ADD_2(_FP_FRAC_WORD_8(_z,7),_FP_FRAC_WORD_8(_z,6), \
312 _b_f1,_b_f0, \
313 _FP_FRAC_WORD_8(_z,7),_FP_FRAC_WORD_8(_z,6)); \
314 \
315 /* Normalize since we know where the msb of the multiplicands \
316 were (bit B), we know that the msb of the of the product is \
317 at either 2B or 2B-1. */ \
318 _FP_FRAC_SRS_8(_z, wfracbits-1, 2*wfracbits); \
319 __FP_FRAC_SET_4(R, _FP_FRAC_WORD_8(_z,3), _FP_FRAC_WORD_8(_z,2), \
320 _FP_FRAC_WORD_8(_z,1), _FP_FRAC_WORD_8(_z,0)); \
321 } while (0)
322
323#define _FP_MUL_MEAT_4_gmp(wfracbits, R, X, Y) \
324 do { \
325 _FP_FRAC_DECL_8(_z); \
326 \
327 mpn_mul_n(_z_f, _x_f, _y_f, 4); \
328 \
329 /* Normalize since we know where the msb of the multiplicands \
330 were (bit B), we know that the msb of the of the product is \
331 at either 2B or 2B-1. */ \
332 _FP_FRAC_SRS_8(_z, wfracbits-1, 2*wfracbits); \
333 __FP_FRAC_SET_4(R, _FP_FRAC_WORD_8(_z,3), _FP_FRAC_WORD_8(_z,2), \
334 _FP_FRAC_WORD_8(_z,1), _FP_FRAC_WORD_8(_z,0)); \
335 } while (0)
336
337/*
338 * Helper utility for _FP_DIV_MEAT_4_udiv:
339 * pppp = m * nnn
340 */
341#define umul_ppppmnnn(p3,p2,p1,p0,m,n2,n1,n0) \
342 do { \
343 UWtype _t; \
344 umul_ppmm(p1,p0,m,n0); \
345 umul_ppmm(p2,_t,m,n1); \
346 __FP_FRAC_ADDI_2(p2,p1,_t); \
347 umul_ppmm(p3,_t,m,n2); \
348 __FP_FRAC_ADDI_2(p3,p2,_t); \
349 } while (0)
350
351/*
352 * Division algorithms:
353 */
354
355#define _FP_DIV_MEAT_4_udiv(fs, R, X, Y) \
356 do { \
357 int _i; \
358 _FP_FRAC_DECL_4(_n); _FP_FRAC_DECL_4(_m); \
359 _FP_FRAC_SET_4(_n, _FP_ZEROFRAC_4); \
360 if (_FP_FRAC_GT_4(X, Y)) \
361 { \
362 _n_f[3] = X##_f[0] << (_FP_W_TYPE_SIZE - 1); \
363 _FP_FRAC_SRL_4(X, 1); \
364 } \
365 else \
366 R##_e--; \
367 \
368 /* Normalize, i.e. make the most significant bit of the \
369 denominator set. */ \
370 _FP_FRAC_SLL_4(Y, _FP_WFRACXBITS_##fs); \
371 \
372 for (_i = 3; ; _i--) \
373 { \
374 if (X##_f[3] == Y##_f[3]) \
375 { \
376 /* This is a special case, not an optimization \
377 (X##_f[3]/Y##_f[3] would not fit into UWtype). \
378 As X## is guaranteed to be < Y, R##_f[_i] can be either \
379 (UWtype)-1 or (UWtype)-2. */ \
380 R##_f[_i] = -1; \
381 if (!_i) \
382 break; \
383 __FP_FRAC_SUB_4(X##_f[3], X##_f[2], X##_f[1], X##_f[0], \
384 Y##_f[2], Y##_f[1], Y##_f[0], 0, \
385 X##_f[2], X##_f[1], X##_f[0], _n_f[_i]); \
386 _FP_FRAC_SUB_4(X, Y, X); \
387 if (X##_f[3] > Y##_f[3]) \
388 { \
389 R##_f[_i] = -2; \
390 _FP_FRAC_ADD_4(X, Y, X); \
391 } \
392 } \
393 else \
394 { \
395 udiv_qrnnd(R##_f[_i], X##_f[3], X##_f[3], X##_f[2], Y##_f[3]); \
396 umul_ppppmnnn(_m_f[3], _m_f[2], _m_f[1], _m_f[0], \
397 R##_f[_i], Y##_f[2], Y##_f[1], Y##_f[0]); \
398 X##_f[2] = X##_f[1]; \
399 X##_f[1] = X##_f[0]; \
400 X##_f[0] = _n_f[_i]; \
401 if (_FP_FRAC_GT_4(_m, X)) \
402 { \
403 R##_f[_i]--; \
404 _FP_FRAC_ADD_4(X, Y, X); \
405 if (_FP_FRAC_GE_4(X, Y) && _FP_FRAC_GT_4(_m, X)) \
406 { \
407 R##_f[_i]--; \
408 _FP_FRAC_ADD_4(X, Y, X); \
409 } \
410 } \
411 _FP_FRAC_DEC_4(X, _m); \
412 if (!_i) \
413 { \
414 if (!_FP_FRAC_EQ_4(X, _m)) \
415 R##_f[0] |= _FP_WORK_STICKY; \
416 break; \
417 } \
418 } \
419 } \
420 } while (0)
421
422
423/*
424 * Square root algorithms:
425 * We have just one right now, maybe Newton approximation
426 * should be added for those machines where division is fast.
427 */
428
429#define _FP_SQRT_MEAT_4(R, S, T, X, q) \
430 do { \
431 while (q) \
432 { \
433 T##_f[3] = S##_f[3] + q; \
434 if (T##_f[3] <= X##_f[3]) \
435 { \
436 S##_f[3] = T##_f[3] + q; \
437 X##_f[3] -= T##_f[3]; \
438 R##_f[3] += q; \
439 } \
440 _FP_FRAC_SLL_4(X, 1); \
441 q >>= 1; \
442 } \
443 q = (_FP_W_TYPE)1 << (_FP_W_TYPE_SIZE - 1); \
444 while (q) \
445 { \
446 T##_f[2] = S##_f[2] + q; \
447 T##_f[3] = S##_f[3]; \
448 if (T##_f[3] < X##_f[3] || \
449 (T##_f[3] == X##_f[3] && T##_f[2] <= X##_f[2])) \
450 { \
451 S##_f[2] = T##_f[2] + q; \
452 S##_f[3] += (T##_f[2] > S##_f[2]); \
453 __FP_FRAC_DEC_2(X##_f[3], X##_f[2], \
454 T##_f[3], T##_f[2]); \
455 R##_f[2] += q; \
456 } \
457 _FP_FRAC_SLL_4(X, 1); \
458 q >>= 1; \
459 } \
460 q = (_FP_W_TYPE)1 << (_FP_W_TYPE_SIZE - 1); \
461 while (q) \
462 { \
463 T##_f[1] = S##_f[1] + q; \
464 T##_f[2] = S##_f[2]; \
465 T##_f[3] = S##_f[3]; \
466 if (T##_f[3] < X##_f[3] || \
467 (T##_f[3] == X##_f[3] && (T##_f[2] < X##_f[2] || \
468 (T##_f[2] == X##_f[2] && T##_f[1] <= X##_f[1])))) \
469 { \
470 S##_f[1] = T##_f[1] + q; \
471 S##_f[2] += (T##_f[1] > S##_f[1]); \
472 S##_f[3] += (T##_f[2] > S##_f[2]); \
473 __FP_FRAC_DEC_3(X##_f[3], X##_f[2], X##_f[1], \
474 T##_f[3], T##_f[2], T##_f[1]); \
475 R##_f[1] += q; \
476 } \
477 _FP_FRAC_SLL_4(X, 1); \
478 q >>= 1; \
479 } \
480 q = (_FP_W_TYPE)1 << (_FP_W_TYPE_SIZE - 1); \
481 while (q != _FP_WORK_ROUND) \
482 { \
483 T##_f[0] = S##_f[0] + q; \
484 T##_f[1] = S##_f[1]; \
485 T##_f[2] = S##_f[2]; \
486 T##_f[3] = S##_f[3]; \
487 if (_FP_FRAC_GE_4(X,T)) \
488 { \
489 S##_f[0] = T##_f[0] + q; \
490 S##_f[1] += (T##_f[0] > S##_f[0]); \
491 S##_f[2] += (T##_f[1] > S##_f[1]); \
492 S##_f[3] += (T##_f[2] > S##_f[2]); \
493 _FP_FRAC_DEC_4(X, T); \
494 R##_f[0] += q; \
495 } \
496 _FP_FRAC_SLL_4(X, 1); \
497 q >>= 1; \
498 } \
499 if (!_FP_FRAC_ZEROP_4(X)) \
500 { \
501 if (_FP_FRAC_GT_4(X,S)) \
502 R##_f[0] |= _FP_WORK_ROUND; \
503 R##_f[0] |= _FP_WORK_STICKY; \
504 } \
505 } while (0)
506
507
508/*
509 * Internals
510 */
511
512#define __FP_FRAC_SET_4(X,I3,I2,I1,I0) \
513 (X##_f[3] = I3, X##_f[2] = I2, X##_f[1] = I1, X##_f[0] = I0)
514
515#ifndef __FP_FRAC_ADD_3
516#define __FP_FRAC_ADD_3(r2,r1,r0,x2,x1,x0,y2,y1,y0) \
517 do { \
518 int _c1, _c2; \
519 r0 = x0 + y0; \
520 _c1 = r0 < x0; \
521 r1 = x1 + y1; \
522 _c2 = r1 < x1; \
523 r1 += _c1; \
524 _c2 |= r1 < _c1; \
525 r2 = x2 + y2 + _c2; \
526 } while (0)
527#endif
528
529#ifndef __FP_FRAC_ADD_4
530#define __FP_FRAC_ADD_4(r3,r2,r1,r0,x3,x2,x1,x0,y3,y2,y1,y0) \
531 do { \
532 int _c1, _c2, _c3; \
533 r0 = x0 + y0; \
534 _c1 = r0 < x0; \
535 r1 = x1 + y1; \
536 _c2 = r1 < x1; \
537 r1 += _c1; \
538 _c2 |= r1 < _c1; \
539 r2 = x2 + y2; \
540 _c3 = r2 < x2; \
541 r2 += _c2; \
542 _c3 |= r2 < _c2; \
543 r3 = x3 + y3 + _c3; \
544 } while (0)
545#endif
546
547#ifndef __FP_FRAC_SUB_3
548#define __FP_FRAC_SUB_3(r2,r1,r0,x2,x1,x0,y2,y1,y0) \
549 do { \
550 int _c1, _c2; \
551 r0 = x0 - y0; \
552 _c1 = r0 > x0; \
553 r1 = x1 - y1; \
554 _c2 = r1 > x1; \
555 r1 -= _c1; \
556 _c2 |= r1 > _c1; \
557 r2 = x2 - y2 - _c2; \
558 } while (0)
559#endif
560
561#ifndef __FP_FRAC_SUB_4
562#define __FP_FRAC_SUB_4(r3,r2,r1,r0,x3,x2,x1,x0,y3,y2,y1,y0) \
563 do { \
564 int _c1, _c2, _c3; \
565 r0 = x0 - y0; \
566 _c1 = r0 > x0; \
567 r1 = x1 - y1; \
568 _c2 = r1 > x1; \
569 r1 -= _c1; \
570 _c2 |= r1 > _c1; \
571 r2 = x2 - y2; \
572 _c3 = r2 > x2; \
573 r2 -= _c2; \
574 _c3 |= r2 > _c2; \
575 r3 = x3 - y3 - _c3; \
576 } while (0)
577#endif
578
579#ifndef __FP_FRAC_DEC_3
580#define __FP_FRAC_DEC_3(x2,x1,x0,y2,y1,y0) \
581 do { \
582 UWtype _t0, _t1, _t2; \
583 _t0 = x0, _t1 = x1, _t2 = x2; \
584 __FP_FRAC_SUB_3 (x2, x1, x0, _t2, _t1, _t0, y2, y1, y0); \
585 } while (0)
586#endif
587
588#ifndef __FP_FRAC_DEC_4
589#define __FP_FRAC_DEC_4(x3,x2,x1,x0,y3,y2,y1,y0) \
590 do { \
591 UWtype _t0, _t1, _t2, _t3; \
592 _t0 = x0, _t1 = x1, _t2 = x2, _t3 = x3; \
593 __FP_FRAC_SUB_4 (x3,x2,x1,x0,_t3,_t2,_t1,_t0, y3,y2,y1,y0); \
594 } while (0)
595#endif
596
597#ifndef __FP_FRAC_ADDI_4
598#define __FP_FRAC_ADDI_4(x3,x2,x1,x0,i) \
599 do { \
600 UWtype _t; \
601 _t = ((x0 += i) < i); \
602 x1 += _t; _t = (x1 < _t); \
603 x2 += _t; _t = (x2 < _t); \
604 x3 += _t; \
605 } while (0)
606#endif
607
608/* Convert FP values between word sizes. This appears to be more
609 * complicated than I'd have expected it to be, so these might be
610 * wrong... These macros are in any case somewhat bogus because they
611 * use information about what various FRAC_n variables look like
612 * internally [eg, that 2 word vars are X_f0 and x_f1]. But so do
613 * the ones in op-2.h and op-1.h.
614 */
615#define _FP_FRAC_CONV_1_4(dfs, sfs, D, S) \
616 do { \
617 if (S##_c != FP_CLS_NAN) \
618 _FP_FRAC_SRS_4(S, (_FP_WFRACBITS_##sfs - _FP_WFRACBITS_##dfs), \
619 _FP_WFRACBITS_##sfs); \
620 else \
621 _FP_FRAC_SRL_4(S, (_FP_WFRACBITS_##sfs - _FP_WFRACBITS_##dfs)); \
622 D##_f = S##_f[0]; \
623 } while (0)
624
625#define _FP_FRAC_CONV_2_4(dfs, sfs, D, S) \
626 do { \
627 if (S##_c != FP_CLS_NAN) \
628 _FP_FRAC_SRS_4(S, (_FP_WFRACBITS_##sfs - _FP_WFRACBITS_##dfs), \
629 _FP_WFRACBITS_##sfs); \
630 else \
631 _FP_FRAC_SRL_4(S, (_FP_WFRACBITS_##sfs - _FP_WFRACBITS_##dfs)); \
632 D##_f0 = S##_f[0]; \
633 D##_f1 = S##_f[1]; \
634 } while (0)
635
636/* Assembly/disassembly for converting to/from integral types.
637 * No shifting or overflow handled here.
638 */
639/* Put the FP value X into r, which is an integer of size rsize. */
640#define _FP_FRAC_ASSEMBLE_4(r, X, rsize) \
641 do { \
642 if (rsize <= _FP_W_TYPE_SIZE) \
643 r = X##_f[0]; \
644 else if (rsize <= 2*_FP_W_TYPE_SIZE) \
645 { \
646 r = X##_f[1]; \
647 r <<= _FP_W_TYPE_SIZE; \
648 r += X##_f[0]; \
649 } \
650 else \
651 { \
652 /* I'm feeling lazy so we deal with int == 3words (implausible)*/ \
653 /* and int == 4words as a single case. */ \
654 r = X##_f[3]; \
655 r <<= _FP_W_TYPE_SIZE; \
656 r += X##_f[2]; \
657 r <<= _FP_W_TYPE_SIZE; \
658 r += X##_f[1]; \
659 r <<= _FP_W_TYPE_SIZE; \
660 r += X##_f[0]; \
661 } \
662 } while (0)
663
664/* "No disassemble Number Five!" */
665/* move an integer of size rsize into X's fractional part. We rely on
666 * the _f[] array consisting of words of size _FP_W_TYPE_SIZE to avoid
667 * having to mask the values we store into it.
668 */
669#define _FP_FRAC_DISASSEMBLE_4(X, r, rsize) \
670 do { \
671 X##_f[0] = r; \
672 X##_f[1] = (rsize <= _FP_W_TYPE_SIZE ? 0 : r >> _FP_W_TYPE_SIZE); \
673 X##_f[2] = (rsize <= 2*_FP_W_TYPE_SIZE ? 0 : r >> 2*_FP_W_TYPE_SIZE); \
674 X##_f[3] = (rsize <= 3*_FP_W_TYPE_SIZE ? 0 : r >> 3*_FP_W_TYPE_SIZE); \
675 } while (0)
676
677#define _FP_FRAC_CONV_4_1(dfs, sfs, D, S) \
678 do { \
679 D##_f[0] = S##_f; \
680 D##_f[1] = D##_f[2] = D##_f[3] = 0; \
681 _FP_FRAC_SLL_4(D, (_FP_WFRACBITS_##dfs - _FP_WFRACBITS_##sfs)); \
682 } while (0)
683
684#define _FP_FRAC_CONV_4_2(dfs, sfs, D, S) \
685 do { \
686 D##_f[0] = S##_f0; \
687 D##_f[1] = S##_f1; \
688 D##_f[2] = D##_f[3] = 0; \
689 _FP_FRAC_SLL_4(D, (_FP_WFRACBITS_##dfs - _FP_WFRACBITS_##sfs)); \
690 } while (0)
691
692#endif