Rev 2479 | Go to most recent revision | 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<mul.h> |
||
37 | #include<comparison.h> |
||
829 | cejka | 38 | #include<common.h> |
731 | cejka | 39 | |
40 | /** Multiply two 32 bit float numbers |
||
41 | * |
||
42 | */ |
||
43 | float32 mulFloat32(float32 a, float32 b) |
||
44 | { |
||
45 | float32 result; |
||
1031 | cejka | 46 | uint64_t frac1, frac2; |
47 | int32_t exp; |
||
731 | cejka | 48 | |
49 | result.parts.sign = a.parts.sign ^ b.parts.sign; |
||
50 | |||
737 | cejka | 51 | if (isFloat32NaN(a) || isFloat32NaN(b) ) { |
731 | cejka | 52 | /* TODO: fix SigNaNs */ |
53 | if (isFloat32SigNaN(a)) { |
||
804 | cejka | 54 | result.parts.fraction = a.parts.fraction; |
731 | cejka | 55 | result.parts.exp = a.parts.exp; |
56 | return result; |
||
57 | }; |
||
58 | if (isFloat32SigNaN(b)) { /* TODO: fix SigNaN */ |
||
804 | cejka | 59 | result.parts.fraction = b.parts.fraction; |
731 | cejka | 60 | result.parts.exp = b.parts.exp; |
61 | return result; |
||
62 | }; |
||
63 | /* set NaN as result */ |
||
737 | cejka | 64 | result.binary = FLOAT32_NAN; |
731 | cejka | 65 | return result; |
66 | }; |
||
67 | |||
68 | if (isFloat32Infinity(a)) { |
||
69 | if (isFloat32Zero(b)) { |
||
70 | /* FIXME: zero * infinity */ |
||
737 | cejka | 71 | result.binary = FLOAT32_NAN; |
731 | cejka | 72 | return result; |
73 | } |
||
804 | cejka | 74 | result.parts.fraction = a.parts.fraction; |
731 | cejka | 75 | result.parts.exp = a.parts.exp; |
76 | return result; |
||
77 | } |
||
78 | |||
79 | if (isFloat32Infinity(b)) { |
||
80 | if (isFloat32Zero(a)) { |
||
81 | /* FIXME: zero * infinity */ |
||
737 | cejka | 82 | result.binary = FLOAT32_NAN; |
731 | cejka | 83 | return result; |
84 | } |
||
804 | cejka | 85 | result.parts.fraction = b.parts.fraction; |
731 | cejka | 86 | result.parts.exp = b.parts.exp; |
87 | return result; |
||
88 | } |
||
89 | |||
90 | /* exp is signed so we can easy detect underflow */ |
||
91 | exp = a.parts.exp + b.parts.exp; |
||
92 | exp -= FLOAT32_BIAS; |
||
93 | |||
737 | cejka | 94 | if (exp >= FLOAT32_MAX_EXPONENT) { |
731 | cejka | 95 | /* FIXME: overflow */ |
96 | /* set infinity as result */ |
||
737 | cejka | 97 | result.binary = FLOAT32_INF; |
98 | result.parts.sign = a.parts.sign ^ b.parts.sign; |
||
731 | cejka | 99 | return result; |
100 | }; |
||
101 | |||
102 | if (exp < 0) { |
||
103 | /* FIXME: underflow */ |
||
104 | /* return signed zero */ |
||
804 | cejka | 105 | result.parts.fraction = 0x0; |
731 | cejka | 106 | result.parts.exp = 0x0; |
107 | return result; |
||
108 | }; |
||
109 | |||
804 | cejka | 110 | frac1 = a.parts.fraction; |
737 | cejka | 111 | if (a.parts.exp > 0) { |
804 | cejka | 112 | frac1 |= FLOAT32_HIDDEN_BIT_MASK; |
731 | cejka | 113 | } else { |
114 | ++exp; |
||
115 | }; |
||
116 | |||
804 | cejka | 117 | frac2 = b.parts.fraction; |
737 | cejka | 118 | |
119 | if (b.parts.exp > 0) { |
||
804 | cejka | 120 | frac2 |= FLOAT32_HIDDEN_BIT_MASK; |
731 | cejka | 121 | } else { |
122 | ++exp; |
||
123 | }; |
||
124 | |||
804 | cejka | 125 | frac1 <<= 1; /* one bit space for rounding */ |
731 | cejka | 126 | |
804 | cejka | 127 | frac1 = frac1 * frac2; |
731 | cejka | 128 | /* round and return */ |
129 | |||
804 | cejka | 130 | while ((exp < FLOAT32_MAX_EXPONENT) && (frac1 >= ( 1 << (FLOAT32_FRACTION_SIZE + 2)))) { |
131 | /* 23 bits of fraction + one more for hidden bit (all shifted 1 bit left)*/ |
||
731 | cejka | 132 | ++exp; |
804 | cejka | 133 | frac1 >>= 1; |
731 | cejka | 134 | }; |
135 | |||
136 | /* rounding */ |
||
804 | cejka | 137 | /* ++frac1; FIXME: not works - without it is ok */ |
138 | frac1 >>= 1; /* shift off rounding space */ |
||
731 | cejka | 139 | |
804 | cejka | 140 | if ((exp < FLOAT32_MAX_EXPONENT) && (frac1 >= (1 << (FLOAT32_FRACTION_SIZE + 1)))) { |
731 | cejka | 141 | ++exp; |
804 | cejka | 142 | frac1 >>= 1; |
731 | cejka | 143 | }; |
144 | |||
737 | cejka | 145 | if (exp >= FLOAT32_MAX_EXPONENT ) { |
731 | cejka | 146 | /* TODO: fix overflow */ |
147 | /* return infinity*/ |
||
737 | cejka | 148 | result.parts.exp = FLOAT32_MAX_EXPONENT; |
804 | cejka | 149 | result.parts.fraction = 0x0; |
731 | cejka | 150 | return result; |
151 | } |
||
152 | |||
804 | cejka | 153 | exp -= FLOAT32_FRACTION_SIZE; |
731 | cejka | 154 | |
804 | cejka | 155 | if (exp <= FLOAT32_FRACTION_SIZE) { |
731 | cejka | 156 | /* denormalized number */ |
804 | cejka | 157 | frac1 >>= 1; /* denormalize */ |
158 | while ((frac1 > 0) && (exp < 0)) { |
||
159 | frac1 >>= 1; |
||
731 | cejka | 160 | ++exp; |
161 | }; |
||
804 | cejka | 162 | if (frac1 == 0) { |
731 | cejka | 163 | /* FIXME : underflow */ |
164 | result.parts.exp = 0; |
||
804 | cejka | 165 | result.parts.fraction = 0; |
731 | cejka | 166 | return result; |
167 | }; |
||
168 | }; |
||
169 | result.parts.exp = exp; |
||
804 | cejka | 170 | result.parts.fraction = frac1 & ( (1 << FLOAT32_FRACTION_SIZE) - 1); |
731 | cejka | 171 | |
172 | return result; |
||
173 | |||
174 | } |
||
175 | |||
737 | cejka | 176 | /** Multiply two 64 bit float numbers |
177 | * |
||
178 | */ |
||
179 | float64 mulFloat64(float64 a, float64 b) |
||
180 | { |
||
181 | float64 result; |
||
1031 | cejka | 182 | uint64_t frac1, frac2; |
183 | int32_t exp; |
||
731 | cejka | 184 | |
737 | cejka | 185 | result.parts.sign = a.parts.sign ^ b.parts.sign; |
186 | |||
187 | if (isFloat64NaN(a) || isFloat64NaN(b) ) { |
||
188 | /* TODO: fix SigNaNs */ |
||
189 | if (isFloat64SigNaN(a)) { |
||
804 | cejka | 190 | result.parts.fraction = a.parts.fraction; |
737 | cejka | 191 | result.parts.exp = a.parts.exp; |
192 | return result; |
||
193 | }; |
||
194 | if (isFloat64SigNaN(b)) { /* TODO: fix SigNaN */ |
||
804 | cejka | 195 | result.parts.fraction = b.parts.fraction; |
737 | cejka | 196 | result.parts.exp = b.parts.exp; |
197 | return result; |
||
198 | }; |
||
199 | /* set NaN as result */ |
||
200 | result.binary = FLOAT64_NAN; |
||
201 | return result; |
||
202 | }; |
||
203 | |||
204 | if (isFloat64Infinity(a)) { |
||
205 | if (isFloat64Zero(b)) { |
||
206 | /* FIXME: zero * infinity */ |
||
207 | result.binary = FLOAT64_NAN; |
||
208 | return result; |
||
209 | } |
||
804 | cejka | 210 | result.parts.fraction = a.parts.fraction; |
737 | cejka | 211 | result.parts.exp = a.parts.exp; |
212 | return result; |
||
213 | } |
||
731 | cejka | 214 | |
737 | cejka | 215 | if (isFloat64Infinity(b)) { |
216 | if (isFloat64Zero(a)) { |
||
217 | /* FIXME: zero * infinity */ |
||
218 | result.binary = FLOAT64_NAN; |
||
219 | return result; |
||
220 | } |
||
804 | cejka | 221 | result.parts.fraction = b.parts.fraction; |
737 | cejka | 222 | result.parts.exp = b.parts.exp; |
223 | return result; |
||
224 | } |
||
225 | |||
226 | /* exp is signed so we can easy detect underflow */ |
||
829 | cejka | 227 | exp = a.parts.exp + b.parts.exp - FLOAT64_BIAS; |
737 | cejka | 228 | |
804 | cejka | 229 | frac1 = a.parts.fraction; |
829 | cejka | 230 | |
737 | cejka | 231 | if (a.parts.exp > 0) { |
804 | cejka | 232 | frac1 |= FLOAT64_HIDDEN_BIT_MASK; |
737 | cejka | 233 | } else { |
234 | ++exp; |
||
235 | }; |
||
236 | |||
804 | cejka | 237 | frac2 = b.parts.fraction; |
737 | cejka | 238 | |
239 | if (b.parts.exp > 0) { |
||
804 | cejka | 240 | frac2 |= FLOAT64_HIDDEN_BIT_MASK; |
737 | cejka | 241 | } else { |
242 | ++exp; |
||
243 | }; |
||
244 | |||
829 | cejka | 245 | frac1 <<= (64 - FLOAT64_FRACTION_SIZE - 1); |
246 | frac2 <<= (64 - FLOAT64_FRACTION_SIZE - 2); |
||
737 | cejka | 247 | |
804 | cejka | 248 | mul64integers(frac1, frac2, &frac1, &frac2); |
737 | cejka | 249 | |
829 | cejka | 250 | frac2 |= (frac1 != 0); |
251 | if (frac2 & (0x1ll << 62)) { |
||
252 | frac2 <<= 1; |
||
253 | exp--; |
||
737 | cejka | 254 | } |
255 | |||
829 | cejka | 256 | result = finishFloat64(exp, frac2, result.parts.sign); |
257 | return result; |
||
737 | cejka | 258 | } |
259 | |||
260 | /** Multiply two 64 bit numbers and return result in two parts |
||
261 | * @param a first operand |
||
262 | * @param b second operand |
||
263 | * @param lo lower part from result |
||
264 | * @param hi higher part of result |
||
265 | */ |
||
1031 | cejka | 266 | void mul64integers(uint64_t a,uint64_t b, uint64_t *lo, uint64_t *hi) |
737 | cejka | 267 | { |
1031 | cejka | 268 | uint64_t low, high, middle1, middle2; |
269 | uint32_t alow, blow; |
||
829 | cejka | 270 | |
737 | cejka | 271 | alow = a & 0xFFFFFFFF; |
272 | blow = b & 0xFFFFFFFF; |
||
273 | |||
828 | cejka | 274 | a >>= 32; |
275 | b >>= 32; |
||
737 | cejka | 276 | |
1031 | cejka | 277 | low = ((uint64_t)alow) * blow; |
737 | cejka | 278 | middle1 = a * blow; |
279 | middle2 = alow * b; |
||
280 | high = a * b; |
||
281 | |||
282 | middle1 += middle2; |
||
1031 | cejka | 283 | high += (((uint64_t)(middle1 < middle2)) << 32) + (middle1 >> 32); |
804 | cejka | 284 | middle1 <<= 32; |
737 | cejka | 285 | low += middle1; |
286 | high += (low < middle1); |
||
287 | *lo = low; |
||
288 | *hi = high; |
||
828 | cejka | 289 | |
737 | cejka | 290 | return; |
291 | } |
||
292 | |||
1740 | jermar | 293 | /** @} |
1657 | cejka | 294 | */ |