Rev 2071 | Details | Compare with Previous | Last modification | View Log | RSS feed
Rev | Author | Line No. | Line |
---|---|---|---|
731 | cejka | 1 | /* |
2071 | jermar | 2 | * Copyright (c) 2005 Josef Cejka |
731 | cejka | 3 | * All rights reserved. |
4 | * |
||
5 | * Redistribution and use in source and binary forms, with or without |
||
6 | * modification, are permitted provided that the following conditions |
||
7 | * are met: |
||
8 | * |
||
9 | * - Redistributions of source code must retain the above copyright |
||
10 | * notice, this list of conditions and the following disclaimer. |
||
11 | * - Redistributions in binary form must reproduce the above copyright |
||
12 | * notice, this list of conditions and the following disclaimer in the |
||
13 | * documentation and/or other materials provided with the distribution. |
||
14 | * - The name of the author may not be used to endorse or promote products |
||
15 | * derived from this software without specific prior written permission. |
||
16 | * |
||
17 | * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR |
||
18 | * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES |
||
19 | * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. |
||
20 | * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, |
||
21 | * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT |
||
22 | * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, |
||
23 | * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY |
||
24 | * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT |
||
25 | * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF |
||
26 | * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. |
||
27 | */ |
||
28 | |||
1740 | jermar | 29 | /** @addtogroup softfloat |
1657 | cejka | 30 | * @{ |
31 | */ |
||
32 | /** @file |
||
33 | */ |
||
34 | |||
731 | cejka | 35 | #include<sftypes.h> |
36 | #include<add.h> |
||
828 | cejka | 37 | #include<div.h> |
731 | cejka | 38 | #include<comparison.h> |
828 | cejka | 39 | #include<mul.h> |
829 | cejka | 40 | #include<common.h> |
731 | cejka | 41 | |
829 | cejka | 42 | |
731 | cejka | 43 | float32 divFloat32(float32 a, float32 b) |
44 | { |
||
804 | cejka | 45 | float32 result; |
1031 | cejka | 46 | int32_t aexp, bexp, cexp; |
47 | uint64_t afrac, bfrac, cfrac; |
||
731 | cejka | 48 | |
804 | cejka | 49 | result.parts.sign = a.parts.sign ^ b.parts.sign; |
50 | |||
51 | if (isFloat32NaN(a)) { |
||
52 | if (isFloat32SigNaN(a)) { |
||
53 | /*FIXME: SigNaN*/ |
||
54 | } |
||
55 | /*NaN*/ |
||
56 | return a; |
||
57 | } |
||
58 | |||
59 | if (isFloat32NaN(b)) { |
||
60 | if (isFloat32SigNaN(b)) { |
||
61 | /*FIXME: SigNaN*/ |
||
62 | } |
||
63 | /*NaN*/ |
||
64 | return b; |
||
65 | } |
||
66 | |||
67 | if (isFloat32Infinity(a)) { |
||
68 | if (isFloat32Infinity(b)) { |
||
69 | /*FIXME: inf / inf */ |
||
70 | result.binary = FLOAT32_NAN; |
||
71 | return result; |
||
72 | } |
||
73 | /* inf / num */ |
||
74 | result.parts.exp = a.parts.exp; |
||
75 | result.parts.fraction = a.parts.fraction; |
||
76 | return result; |
||
77 | } |
||
78 | |||
79 | if (isFloat32Infinity(b)) { |
||
80 | if (isFloat32Zero(a)) { |
||
81 | /* FIXME 0 / inf */ |
||
82 | result.parts.exp = 0; |
||
83 | result.parts.fraction = 0; |
||
84 | return result; |
||
85 | } |
||
86 | /* FIXME: num / inf*/ |
||
87 | result.parts.exp = 0; |
||
88 | result.parts.fraction = 0; |
||
89 | return result; |
||
90 | } |
||
91 | |||
92 | if (isFloat32Zero(b)) { |
||
93 | if (isFloat32Zero(a)) { |
||
94 | /*FIXME: 0 / 0*/ |
||
95 | result.binary = FLOAT32_NAN; |
||
96 | return result; |
||
97 | } |
||
98 | /* FIXME: division by zero */ |
||
99 | result.parts.exp = 0; |
||
100 | result.parts.fraction = 0; |
||
101 | return result; |
||
102 | } |
||
103 | |||
104 | |||
105 | afrac = a.parts.fraction; |
||
106 | aexp = a.parts.exp; |
||
107 | bfrac = b.parts.fraction; |
||
108 | bexp = b.parts.exp; |
||
109 | |||
110 | /* denormalized numbers */ |
||
111 | if (aexp == 0) { |
||
112 | if (afrac == 0) { |
||
113 | result.parts.exp = 0; |
||
114 | result.parts.fraction = 0; |
||
115 | return result; |
||
116 | } |
||
117 | /* normalize it*/ |
||
118 | |||
119 | afrac <<= 1; |
||
120 | /* afrac is nonzero => it must stop */ |
||
121 | while (! (afrac & FLOAT32_HIDDEN_BIT_MASK) ) { |
||
122 | afrac <<= 1; |
||
123 | aexp--; |
||
124 | } |
||
125 | } |
||
126 | |||
127 | if (bexp == 0) { |
||
128 | bfrac <<= 1; |
||
129 | /* bfrac is nonzero => it must stop */ |
||
130 | while (! (bfrac & FLOAT32_HIDDEN_BIT_MASK) ) { |
||
131 | bfrac <<= 1; |
||
132 | bexp--; |
||
133 | } |
||
134 | } |
||
135 | |||
136 | afrac = (afrac | FLOAT32_HIDDEN_BIT_MASK ) << (32 - FLOAT32_FRACTION_SIZE - 1 ); |
||
137 | bfrac = (bfrac | FLOAT32_HIDDEN_BIT_MASK ) << (32 - FLOAT32_FRACTION_SIZE ); |
||
138 | |||
139 | if ( bfrac <= (afrac << 1) ) { |
||
140 | afrac >>= 1; |
||
141 | aexp++; |
||
142 | } |
||
143 | |||
144 | cexp = aexp - bexp + FLOAT32_BIAS - 2; |
||
145 | |||
146 | cfrac = (afrac << 32) / bfrac; |
||
147 | if (( cfrac & 0x3F ) == 0) { |
||
148 | cfrac |= ( bfrac * cfrac != afrac << 32 ); |
||
149 | } |
||
150 | |||
151 | /* pack and round */ |
||
152 | |||
828 | cejka | 153 | /* find first nonzero digit and shift result and detect possibly underflow */ |
804 | cejka | 154 | while ((cexp > 0) && (cfrac) && (!(cfrac & (FLOAT32_HIDDEN_BIT_MASK << 7 )))) { |
155 | cexp--; |
||
156 | cfrac <<= 1; |
||
157 | /* TODO: fix underflow */ |
||
158 | }; |
||
159 | |||
160 | cfrac += (0x1 << 6); /* FIXME: 7 is not sure*/ |
||
161 | |||
162 | if (cfrac & (FLOAT32_HIDDEN_BIT_MASK << 7)) { |
||
163 | ++cexp; |
||
164 | cfrac >>= 1; |
||
165 | } |
||
166 | |||
167 | /* check overflow */ |
||
168 | if (cexp >= FLOAT32_MAX_EXPONENT ) { |
||
169 | /* FIXME: overflow, return infinity */ |
||
170 | result.parts.exp = FLOAT32_MAX_EXPONENT; |
||
171 | result.parts.fraction = 0; |
||
172 | return result; |
||
173 | } |
||
174 | |||
175 | if (cexp < 0) { |
||
176 | /* FIXME: underflow */ |
||
177 | result.parts.exp = 0; |
||
178 | if ((cexp + FLOAT32_FRACTION_SIZE) < 0) { |
||
179 | result.parts.fraction = 0; |
||
180 | return result; |
||
181 | } |
||
182 | cfrac >>= 1; |
||
183 | while (cexp < 0) { |
||
184 | cexp ++; |
||
185 | cfrac >>= 1; |
||
186 | } |
||
187 | |||
188 | } else { |
||
1031 | cejka | 189 | result.parts.exp = (uint32_t)cexp; |
804 | cejka | 190 | } |
191 | |||
192 | result.parts.fraction = ((cfrac >> 6) & (~FLOAT32_HIDDEN_BIT_MASK)); |
||
193 | |||
194 | return result; |
||
731 | cejka | 195 | } |
196 | |||
828 | cejka | 197 | float64 divFloat64(float64 a, float64 b) |
198 | { |
||
199 | float64 result; |
||
1031 | cejka | 200 | int64_t aexp, bexp, cexp; |
201 | uint64_t afrac, bfrac, cfrac; |
||
202 | uint64_t remlo, remhi; |
||
828 | cejka | 203 | |
204 | result.parts.sign = a.parts.sign ^ b.parts.sign; |
||
205 | |||
206 | if (isFloat64NaN(a)) { |
||
835 | cejka | 207 | |
208 | if (isFloat64SigNaN(b)) { |
||
209 | /*FIXME: SigNaN*/ |
||
210 | return b; |
||
211 | } |
||
212 | |||
828 | cejka | 213 | if (isFloat64SigNaN(a)) { |
214 | /*FIXME: SigNaN*/ |
||
215 | } |
||
216 | /*NaN*/ |
||
217 | return a; |
||
218 | } |
||
219 | |||
220 | if (isFloat64NaN(b)) { |
||
221 | if (isFloat64SigNaN(b)) { |
||
222 | /*FIXME: SigNaN*/ |
||
223 | } |
||
224 | /*NaN*/ |
||
225 | return b; |
||
226 | } |
||
227 | |||
228 | if (isFloat64Infinity(a)) { |
||
835 | cejka | 229 | if (isFloat64Infinity(b) || isFloat64Zero(b)) { |
828 | cejka | 230 | /*FIXME: inf / inf */ |
231 | result.binary = FLOAT64_NAN; |
||
232 | return result; |
||
233 | } |
||
234 | /* inf / num */ |
||
235 | result.parts.exp = a.parts.exp; |
||
236 | result.parts.fraction = a.parts.fraction; |
||
237 | return result; |
||
238 | } |
||
239 | |||
240 | if (isFloat64Infinity(b)) { |
||
241 | if (isFloat64Zero(a)) { |
||
242 | /* FIXME 0 / inf */ |
||
243 | result.parts.exp = 0; |
||
244 | result.parts.fraction = 0; |
||
245 | return result; |
||
246 | } |
||
247 | /* FIXME: num / inf*/ |
||
248 | result.parts.exp = 0; |
||
249 | result.parts.fraction = 0; |
||
250 | return result; |
||
251 | } |
||
252 | |||
253 | if (isFloat64Zero(b)) { |
||
254 | if (isFloat64Zero(a)) { |
||
255 | /*FIXME: 0 / 0*/ |
||
256 | result.binary = FLOAT64_NAN; |
||
257 | return result; |
||
258 | } |
||
259 | /* FIXME: division by zero */ |
||
260 | result.parts.exp = 0; |
||
261 | result.parts.fraction = 0; |
||
262 | return result; |
||
263 | } |
||
264 | |||
265 | |||
266 | afrac = a.parts.fraction; |
||
267 | aexp = a.parts.exp; |
||
268 | bfrac = b.parts.fraction; |
||
269 | bexp = b.parts.exp; |
||
270 | |||
271 | /* denormalized numbers */ |
||
272 | if (aexp == 0) { |
||
273 | if (afrac == 0) { |
||
835 | cejka | 274 | result.parts.exp = 0; |
275 | result.parts.fraction = 0; |
||
276 | return result; |
||
828 | cejka | 277 | } |
278 | /* normalize it*/ |
||
279 | |||
835 | cejka | 280 | aexp++; |
828 | cejka | 281 | /* afrac is nonzero => it must stop */ |
282 | while (! (afrac & FLOAT64_HIDDEN_BIT_MASK) ) { |
||
283 | afrac <<= 1; |
||
284 | aexp--; |
||
285 | } |
||
286 | } |
||
287 | |||
288 | if (bexp == 0) { |
||
835 | cejka | 289 | bexp++; |
828 | cejka | 290 | /* bfrac is nonzero => it must stop */ |
291 | while (! (bfrac & FLOAT64_HIDDEN_BIT_MASK) ) { |
||
292 | bfrac <<= 1; |
||
293 | bexp--; |
||
294 | } |
||
295 | } |
||
296 | |||
297 | afrac = (afrac | FLOAT64_HIDDEN_BIT_MASK ) << (64 - FLOAT64_FRACTION_SIZE - 2 ); |
||
298 | bfrac = (bfrac | FLOAT64_HIDDEN_BIT_MASK ) << (64 - FLOAT64_FRACTION_SIZE - 1); |
||
299 | |||
300 | if ( bfrac <= (afrac << 1) ) { |
||
301 | afrac >>= 1; |
||
302 | aexp++; |
||
303 | } |
||
304 | |||
305 | cexp = aexp - bexp + FLOAT64_BIAS - 2; |
||
306 | |||
307 | cfrac = divFloat64estim(afrac, bfrac); |
||
308 | |||
309 | if (( cfrac & 0x1FF ) <= 2) { /*FIXME:?? */ |
||
310 | mul64integers( bfrac, cfrac, &remlo, &remhi); |
||
311 | /* (__u128)afrac << 64 - ( ((__u128)remhi<<64) + (__u128)remlo )*/ |
||
312 | remhi = afrac - remhi - ( remlo > 0); |
||
313 | remlo = - remlo; |
||
314 | |||
1031 | cejka | 315 | while ((int64_t) remhi < 0) { |
828 | cejka | 316 | cfrac--; |
317 | remlo += bfrac; |
||
318 | remhi += ( remlo < bfrac ); |
||
319 | } |
||
320 | cfrac |= ( remlo != 0 ); |
||
321 | } |
||
322 | |||
829 | cejka | 323 | /* round and shift */ |
324 | result = finishFloat64(cexp, cfrac, result.parts.sign); |
||
325 | return result; |
||
828 | cejka | 326 | |
327 | } |
||
328 | |||
1031 | cejka | 329 | uint64_t divFloat64estim(uint64_t a, uint64_t b) |
828 | cejka | 330 | { |
1031 | cejka | 331 | uint64_t bhi; |
332 | uint64_t remhi, remlo; |
||
333 | uint64_t result; |
||
828 | cejka | 334 | |
335 | if ( b <= a ) { |
||
336 | return 0xFFFFFFFFFFFFFFFFull; |
||
337 | } |
||
338 | |||
339 | bhi = b >> 32; |
||
340 | result = ((bhi << 32) <= a) ?( 0xFFFFFFFFull << 32) : ( a / bhi) << 32; |
||
341 | mul64integers(b, result, &remlo, &remhi); |
||
342 | |||
343 | remhi = a - remhi - (remlo > 0); |
||
344 | remlo = - remlo; |
||
345 | |||
346 | b <<= 32; |
||
1031 | cejka | 347 | while ( (int64_t) remhi < 0 ) { |
828 | cejka | 348 | result -= 0x1ll << 32; |
349 | remlo += b; |
||
350 | remhi += bhi + ( remlo < b ); |
||
351 | } |
||
352 | remhi = (remhi << 32) | (remlo >> 32); |
||
353 | if (( bhi << 32) <= remhi) { |
||
354 | result |= 0xFFFFFFFF; |
||
355 | } else { |
||
356 | result |= remhi / bhi; |
||
357 | } |
||
358 | |||
359 | |||
360 | return result; |
||
361 | } |
||
362 | |||
1740 | jermar | 363 | /** @} |
1657 | cejka | 364 | */ |