Rev 835 | Rev 1657 | Go to most recent revision | Details | Compare with Previous | Last modification | View Log | RSS feed
Rev | Author | Line No. | Line |
---|---|---|---|
731 | cejka | 1 | /* |
2 | * Copyright (C) 2005 Josef 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 | |||
29 | #include<sftypes.h> |
||
30 | #include<add.h> |
||
828 | cejka | 31 | #include<div.h> |
731 | cejka | 32 | #include<comparison.h> |
828 | cejka | 33 | #include<mul.h> |
829 | cejka | 34 | #include<common.h> |
731 | cejka | 35 | |
829 | cejka | 36 | |
731 | cejka | 37 | float32 divFloat32(float32 a, float32 b) |
38 | { |
||
804 | cejka | 39 | float32 result; |
1031 | cejka | 40 | int32_t aexp, bexp, cexp; |
41 | uint64_t afrac, bfrac, cfrac; |
||
731 | cejka | 42 | |
804 | cejka | 43 | result.parts.sign = a.parts.sign ^ b.parts.sign; |
44 | |||
45 | if (isFloat32NaN(a)) { |
||
46 | if (isFloat32SigNaN(a)) { |
||
47 | /*FIXME: SigNaN*/ |
||
48 | } |
||
49 | /*NaN*/ |
||
50 | return a; |
||
51 | } |
||
52 | |||
53 | if (isFloat32NaN(b)) { |
||
54 | if (isFloat32SigNaN(b)) { |
||
55 | /*FIXME: SigNaN*/ |
||
56 | } |
||
57 | /*NaN*/ |
||
58 | return b; |
||
59 | } |
||
60 | |||
61 | if (isFloat32Infinity(a)) { |
||
62 | if (isFloat32Infinity(b)) { |
||
63 | /*FIXME: inf / inf */ |
||
64 | result.binary = FLOAT32_NAN; |
||
65 | return result; |
||
66 | } |
||
67 | /* inf / num */ |
||
68 | result.parts.exp = a.parts.exp; |
||
69 | result.parts.fraction = a.parts.fraction; |
||
70 | return result; |
||
71 | } |
||
72 | |||
73 | if (isFloat32Infinity(b)) { |
||
74 | if (isFloat32Zero(a)) { |
||
75 | /* FIXME 0 / inf */ |
||
76 | result.parts.exp = 0; |
||
77 | result.parts.fraction = 0; |
||
78 | return result; |
||
79 | } |
||
80 | /* FIXME: num / inf*/ |
||
81 | result.parts.exp = 0; |
||
82 | result.parts.fraction = 0; |
||
83 | return result; |
||
84 | } |
||
85 | |||
86 | if (isFloat32Zero(b)) { |
||
87 | if (isFloat32Zero(a)) { |
||
88 | /*FIXME: 0 / 0*/ |
||
89 | result.binary = FLOAT32_NAN; |
||
90 | return result; |
||
91 | } |
||
92 | /* FIXME: division by zero */ |
||
93 | result.parts.exp = 0; |
||
94 | result.parts.fraction = 0; |
||
95 | return result; |
||
96 | } |
||
97 | |||
98 | |||
99 | afrac = a.parts.fraction; |
||
100 | aexp = a.parts.exp; |
||
101 | bfrac = b.parts.fraction; |
||
102 | bexp = b.parts.exp; |
||
103 | |||
104 | /* denormalized numbers */ |
||
105 | if (aexp == 0) { |
||
106 | if (afrac == 0) { |
||
107 | result.parts.exp = 0; |
||
108 | result.parts.fraction = 0; |
||
109 | return result; |
||
110 | } |
||
111 | /* normalize it*/ |
||
112 | |||
113 | afrac <<= 1; |
||
114 | /* afrac is nonzero => it must stop */ |
||
115 | while (! (afrac & FLOAT32_HIDDEN_BIT_MASK) ) { |
||
116 | afrac <<= 1; |
||
117 | aexp--; |
||
118 | } |
||
119 | } |
||
120 | |||
121 | if (bexp == 0) { |
||
122 | bfrac <<= 1; |
||
123 | /* bfrac is nonzero => it must stop */ |
||
124 | while (! (bfrac & FLOAT32_HIDDEN_BIT_MASK) ) { |
||
125 | bfrac <<= 1; |
||
126 | bexp--; |
||
127 | } |
||
128 | } |
||
129 | |||
130 | afrac = (afrac | FLOAT32_HIDDEN_BIT_MASK ) << (32 - FLOAT32_FRACTION_SIZE - 1 ); |
||
131 | bfrac = (bfrac | FLOAT32_HIDDEN_BIT_MASK ) << (32 - FLOAT32_FRACTION_SIZE ); |
||
132 | |||
133 | if ( bfrac <= (afrac << 1) ) { |
||
134 | afrac >>= 1; |
||
135 | aexp++; |
||
136 | } |
||
137 | |||
138 | cexp = aexp - bexp + FLOAT32_BIAS - 2; |
||
139 | |||
140 | cfrac = (afrac << 32) / bfrac; |
||
141 | if (( cfrac & 0x3F ) == 0) { |
||
142 | cfrac |= ( bfrac * cfrac != afrac << 32 ); |
||
143 | } |
||
144 | |||
145 | /* pack and round */ |
||
146 | |||
828 | cejka | 147 | /* find first nonzero digit and shift result and detect possibly underflow */ |
804 | cejka | 148 | while ((cexp > 0) && (cfrac) && (!(cfrac & (FLOAT32_HIDDEN_BIT_MASK << 7 )))) { |
149 | cexp--; |
||
150 | cfrac <<= 1; |
||
151 | /* TODO: fix underflow */ |
||
152 | }; |
||
153 | |||
154 | cfrac += (0x1 << 6); /* FIXME: 7 is not sure*/ |
||
155 | |||
156 | if (cfrac & (FLOAT32_HIDDEN_BIT_MASK << 7)) { |
||
157 | ++cexp; |
||
158 | cfrac >>= 1; |
||
159 | } |
||
160 | |||
161 | /* check overflow */ |
||
162 | if (cexp >= FLOAT32_MAX_EXPONENT ) { |
||
163 | /* FIXME: overflow, return infinity */ |
||
164 | result.parts.exp = FLOAT32_MAX_EXPONENT; |
||
165 | result.parts.fraction = 0; |
||
166 | return result; |
||
167 | } |
||
168 | |||
169 | if (cexp < 0) { |
||
170 | /* FIXME: underflow */ |
||
171 | result.parts.exp = 0; |
||
172 | if ((cexp + FLOAT32_FRACTION_SIZE) < 0) { |
||
173 | result.parts.fraction = 0; |
||
174 | return result; |
||
175 | } |
||
176 | cfrac >>= 1; |
||
177 | while (cexp < 0) { |
||
178 | cexp ++; |
||
179 | cfrac >>= 1; |
||
180 | } |
||
181 | |||
182 | } else { |
||
1031 | cejka | 183 | result.parts.exp = (uint32_t)cexp; |
804 | cejka | 184 | } |
185 | |||
186 | result.parts.fraction = ((cfrac >> 6) & (~FLOAT32_HIDDEN_BIT_MASK)); |
||
187 | |||
188 | return result; |
||
731 | cejka | 189 | } |
190 | |||
828 | cejka | 191 | float64 divFloat64(float64 a, float64 b) |
192 | { |
||
193 | float64 result; |
||
1031 | cejka | 194 | int64_t aexp, bexp, cexp; |
195 | uint64_t afrac, bfrac, cfrac; |
||
196 | uint64_t remlo, remhi; |
||
828 | cejka | 197 | |
198 | result.parts.sign = a.parts.sign ^ b.parts.sign; |
||
199 | |||
200 | if (isFloat64NaN(a)) { |
||
835 | cejka | 201 | |
202 | if (isFloat64SigNaN(b)) { |
||
203 | /*FIXME: SigNaN*/ |
||
204 | return b; |
||
205 | } |
||
206 | |||
828 | cejka | 207 | if (isFloat64SigNaN(a)) { |
208 | /*FIXME: SigNaN*/ |
||
209 | } |
||
210 | /*NaN*/ |
||
211 | return a; |
||
212 | } |
||
213 | |||
214 | if (isFloat64NaN(b)) { |
||
215 | if (isFloat64SigNaN(b)) { |
||
216 | /*FIXME: SigNaN*/ |
||
217 | } |
||
218 | /*NaN*/ |
||
219 | return b; |
||
220 | } |
||
221 | |||
222 | if (isFloat64Infinity(a)) { |
||
835 | cejka | 223 | if (isFloat64Infinity(b) || isFloat64Zero(b)) { |
828 | cejka | 224 | /*FIXME: inf / inf */ |
225 | result.binary = FLOAT64_NAN; |
||
226 | return result; |
||
227 | } |
||
228 | /* inf / num */ |
||
229 | result.parts.exp = a.parts.exp; |
||
230 | result.parts.fraction = a.parts.fraction; |
||
231 | return result; |
||
232 | } |
||
233 | |||
234 | if (isFloat64Infinity(b)) { |
||
235 | if (isFloat64Zero(a)) { |
||
236 | /* FIXME 0 / inf */ |
||
237 | result.parts.exp = 0; |
||
238 | result.parts.fraction = 0; |
||
239 | return result; |
||
240 | } |
||
241 | /* FIXME: num / inf*/ |
||
242 | result.parts.exp = 0; |
||
243 | result.parts.fraction = 0; |
||
244 | return result; |
||
245 | } |
||
246 | |||
247 | if (isFloat64Zero(b)) { |
||
248 | if (isFloat64Zero(a)) { |
||
249 | /*FIXME: 0 / 0*/ |
||
250 | result.binary = FLOAT64_NAN; |
||
251 | return result; |
||
252 | } |
||
253 | /* FIXME: division by zero */ |
||
254 | result.parts.exp = 0; |
||
255 | result.parts.fraction = 0; |
||
256 | return result; |
||
257 | } |
||
258 | |||
259 | |||
260 | afrac = a.parts.fraction; |
||
261 | aexp = a.parts.exp; |
||
262 | bfrac = b.parts.fraction; |
||
263 | bexp = b.parts.exp; |
||
264 | |||
265 | /* denormalized numbers */ |
||
266 | if (aexp == 0) { |
||
267 | if (afrac == 0) { |
||
835 | cejka | 268 | result.parts.exp = 0; |
269 | result.parts.fraction = 0; |
||
270 | return result; |
||
828 | cejka | 271 | } |
272 | /* normalize it*/ |
||
273 | |||
835 | cejka | 274 | aexp++; |
828 | cejka | 275 | /* afrac is nonzero => it must stop */ |
276 | while (! (afrac & FLOAT64_HIDDEN_BIT_MASK) ) { |
||
277 | afrac <<= 1; |
||
278 | aexp--; |
||
279 | } |
||
280 | } |
||
281 | |||
282 | if (bexp == 0) { |
||
835 | cejka | 283 | bexp++; |
828 | cejka | 284 | /* bfrac is nonzero => it must stop */ |
285 | while (! (bfrac & FLOAT64_HIDDEN_BIT_MASK) ) { |
||
286 | bfrac <<= 1; |
||
287 | bexp--; |
||
288 | } |
||
289 | } |
||
290 | |||
291 | afrac = (afrac | FLOAT64_HIDDEN_BIT_MASK ) << (64 - FLOAT64_FRACTION_SIZE - 2 ); |
||
292 | bfrac = (bfrac | FLOAT64_HIDDEN_BIT_MASK ) << (64 - FLOAT64_FRACTION_SIZE - 1); |
||
293 | |||
294 | if ( bfrac <= (afrac << 1) ) { |
||
295 | afrac >>= 1; |
||
296 | aexp++; |
||
297 | } |
||
298 | |||
299 | cexp = aexp - bexp + FLOAT64_BIAS - 2; |
||
300 | |||
301 | cfrac = divFloat64estim(afrac, bfrac); |
||
302 | |||
303 | if (( cfrac & 0x1FF ) <= 2) { /*FIXME:?? */ |
||
304 | mul64integers( bfrac, cfrac, &remlo, &remhi); |
||
305 | /* (__u128)afrac << 64 - ( ((__u128)remhi<<64) + (__u128)remlo )*/ |
||
306 | remhi = afrac - remhi - ( remlo > 0); |
||
307 | remlo = - remlo; |
||
308 | |||
1031 | cejka | 309 | while ((int64_t) remhi < 0) { |
828 | cejka | 310 | cfrac--; |
311 | remlo += bfrac; |
||
312 | remhi += ( remlo < bfrac ); |
||
313 | } |
||
314 | cfrac |= ( remlo != 0 ); |
||
315 | } |
||
316 | |||
829 | cejka | 317 | /* round and shift */ |
318 | result = finishFloat64(cexp, cfrac, result.parts.sign); |
||
319 | return result; |
||
828 | cejka | 320 | |
321 | } |
||
322 | |||
1031 | cejka | 323 | uint64_t divFloat64estim(uint64_t a, uint64_t b) |
828 | cejka | 324 | { |
1031 | cejka | 325 | uint64_t bhi; |
326 | uint64_t remhi, remlo; |
||
327 | uint64_t result; |
||
828 | cejka | 328 | |
329 | if ( b <= a ) { |
||
330 | return 0xFFFFFFFFFFFFFFFFull; |
||
331 | } |
||
332 | |||
333 | bhi = b >> 32; |
||
334 | result = ((bhi << 32) <= a) ?( 0xFFFFFFFFull << 32) : ( a / bhi) << 32; |
||
335 | mul64integers(b, result, &remlo, &remhi); |
||
336 | |||
337 | remhi = a - remhi - (remlo > 0); |
||
338 | remlo = - remlo; |
||
339 | |||
340 | b <<= 32; |
||
1031 | cejka | 341 | while ( (int64_t) remhi < 0 ) { |
828 | cejka | 342 | result -= 0x1ll << 32; |
343 | remlo += b; |
||
344 | remhi += bhi + ( remlo < b ); |
||
345 | } |
||
346 | remhi = (remhi << 32) | (remlo >> 32); |
||
347 | if (( bhi << 32) <= remhi) { |
||
348 | result |= 0xFFFFFFFF; |
||
349 | } else { |
||
350 | result |= remhi / bhi; |
||
351 | } |
||
352 | |||
353 | |||
354 | return result; |
||
355 | } |
||
356 |