Rev 1031 | 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 | |||
1657 | cejka | 29 | /** @addtogroup softfloat |
30 | * @{ |
||
31 | */ |
||
32 | /** @file |
||
33 | */ |
||
34 | |||
734 | cejka | 35 | #include<sftypes.h> |
36 | #include<sub.h> |
||
37 | #include<comparison.h> |
||
731 | cejka | 38 | |
39 | /** Subtract two float32 numbers with same signs |
||
40 | */ |
||
41 | float32 subFloat32(float32 a, float32 b) |
||
42 | { |
||
43 | int expdiff; |
||
1031 | cejka | 44 | uint32_t exp1, exp2, frac1, frac2; |
731 | cejka | 45 | float32 result; |
46 | |||
47 | result.f = 0; |
||
48 | |||
734 | cejka | 49 | expdiff = a.parts.exp - b.parts.exp; |
804 | cejka | 50 | if ((expdiff < 0 ) || ((expdiff == 0) && (a.parts.fraction < b.parts.fraction))) { |
731 | cejka | 51 | if (isFloat32NaN(b)) { |
804 | cejka | 52 | /* TODO: fix SigNaN */ |
731 | cejka | 53 | if (isFloat32SigNaN(b)) { |
54 | }; |
||
55 | return b; |
||
56 | }; |
||
57 | |||
734 | cejka | 58 | if (b.parts.exp == FLOAT32_MAX_EXPONENT) { |
59 | b.parts.sign = !b.parts.sign; /* num -(+-inf) = -+inf */ |
||
731 | cejka | 60 | return b; |
61 | } |
||
62 | |||
63 | result.parts.sign = !a.parts.sign; |
||
64 | |||
804 | cejka | 65 | frac1 = b.parts.fraction; |
734 | cejka | 66 | exp1 = b.parts.exp; |
804 | cejka | 67 | frac2 = a.parts.fraction; |
734 | cejka | 68 | exp2 = a.parts.exp; |
69 | expdiff *= -1; |
||
731 | cejka | 70 | } else { |
71 | if (isFloat32NaN(a)) { |
||
804 | cejka | 72 | /* TODO: fix SigNaN */ |
734 | cejka | 73 | if (isFloat32SigNaN(a) || isFloat32SigNaN(b)) { |
731 | cejka | 74 | }; |
75 | return a; |
||
76 | }; |
||
77 | |||
734 | cejka | 78 | if (a.parts.exp == FLOAT32_MAX_EXPONENT) { |
79 | if (b.parts.exp == FLOAT32_MAX_EXPONENT) { |
||
731 | cejka | 80 | /* inf - inf => nan */ |
804 | cejka | 81 | /* TODO: fix exception */ |
731 | cejka | 82 | result.binary = FLOAT32_NAN; |
83 | return result; |
||
84 | }; |
||
85 | return a; |
||
86 | } |
||
87 | |||
88 | result.parts.sign = a.parts.sign; |
||
89 | |||
804 | cejka | 90 | frac1 = a.parts.fraction; |
734 | cejka | 91 | exp1 = a.parts.exp; |
804 | cejka | 92 | frac2 = b.parts.fraction; |
734 | cejka | 93 | exp2 = b.parts.exp; |
731 | cejka | 94 | }; |
95 | |||
734 | cejka | 96 | if (exp1 == 0) { |
804 | cejka | 97 | /* both are denormalized */ |
98 | result.parts.fraction = frac1-frac2; |
||
99 | if (result.parts.fraction > frac1) { |
||
100 | /* TODO: underflow exception */ |
||
731 | cejka | 101 | return result; |
102 | }; |
||
734 | cejka | 103 | result.parts.exp = 0; |
731 | cejka | 104 | return result; |
105 | }; |
||
734 | cejka | 106 | |
107 | /* add hidden bit */ |
||
804 | cejka | 108 | frac1 |= FLOAT32_HIDDEN_BIT_MASK; |
731 | cejka | 109 | |
734 | cejka | 110 | if (exp2 == 0) { |
111 | /* denormalized */ |
||
112 | --expdiff; |
||
113 | } else { |
||
114 | /* normalized */ |
||
804 | cejka | 115 | frac2 |= FLOAT32_HIDDEN_BIT_MASK; |
734 | cejka | 116 | }; |
731 | cejka | 117 | |
734 | cejka | 118 | /* create some space for rounding */ |
804 | cejka | 119 | frac1 <<= 6; |
120 | frac2 <<= 6; |
||
731 | cejka | 121 | |
804 | cejka | 122 | if (expdiff > FLOAT32_FRACTION_SIZE + 1) { |
734 | cejka | 123 | goto done; |
124 | }; |
||
731 | cejka | 125 | |
804 | cejka | 126 | frac1 = frac1 - (frac2 >> expdiff); |
734 | cejka | 127 | done: |
804 | cejka | 128 | /* TODO: find first nonzero digit and shift result and detect possibly underflow */ |
129 | while ((exp1 > 0) && (!(frac1 & (FLOAT32_HIDDEN_BIT_MASK << 6 )))) { |
||
734 | cejka | 130 | --exp1; |
804 | cejka | 131 | frac1 <<= 1; |
132 | /* TODO: fix underflow - frac1 == 0 does not necessary means underflow... */ |
||
734 | cejka | 133 | }; |
134 | |||
804 | cejka | 135 | /* rounding - if first bit after fraction is set then round up */ |
136 | frac1 += 0x20; |
||
734 | cejka | 137 | |
804 | cejka | 138 | if (frac1 & (FLOAT32_HIDDEN_BIT_MASK << 7)) { |
734 | cejka | 139 | ++exp1; |
804 | cejka | 140 | frac1 >>= 1; |
734 | cejka | 141 | }; |
142 | |||
143 | /*Clear hidden bit and shift */ |
||
804 | cejka | 144 | result.parts.fraction = ((frac1 >> 6) & (~FLOAT32_HIDDEN_BIT_MASK)); |
734 | cejka | 145 | result.parts.exp = exp1; |
146 | |||
147 | return result; |
||
148 | } |
||
149 | |||
150 | /** Subtract two float64 numbers with same signs |
||
151 | */ |
||
152 | float64 subFloat64(float64 a, float64 b) |
||
153 | { |
||
154 | int expdiff; |
||
1031 | cejka | 155 | uint32_t exp1, exp2; |
156 | uint64_t frac1, frac2; |
||
734 | cejka | 157 | float64 result; |
158 | |||
159 | result.d = 0; |
||
160 | |||
161 | expdiff = a.parts.exp - b.parts.exp; |
||
804 | cejka | 162 | if ((expdiff < 0 ) || ((expdiff == 0) && (a.parts.fraction < b.parts.fraction))) { |
734 | cejka | 163 | if (isFloat64NaN(b)) { |
804 | cejka | 164 | /* TODO: fix SigNaN */ |
734 | cejka | 165 | if (isFloat64SigNaN(b)) { |
166 | }; |
||
167 | return b; |
||
168 | }; |
||
169 | |||
170 | if (b.parts.exp == FLOAT64_MAX_EXPONENT) { |
||
171 | b.parts.sign = !b.parts.sign; /* num -(+-inf) = -+inf */ |
||
172 | return b; |
||
173 | } |
||
174 | |||
175 | result.parts.sign = !a.parts.sign; |
||
176 | |||
804 | cejka | 177 | frac1 = b.parts.fraction; |
734 | cejka | 178 | exp1 = b.parts.exp; |
804 | cejka | 179 | frac2 = a.parts.fraction; |
734 | cejka | 180 | exp2 = a.parts.exp; |
181 | expdiff *= -1; |
||
182 | } else { |
||
183 | if (isFloat64NaN(a)) { |
||
804 | cejka | 184 | /* TODO: fix SigNaN */ |
734 | cejka | 185 | if (isFloat64SigNaN(a) || isFloat64SigNaN(b)) { |
186 | }; |
||
187 | return a; |
||
188 | }; |
||
189 | |||
190 | if (a.parts.exp == FLOAT64_MAX_EXPONENT) { |
||
191 | if (b.parts.exp == FLOAT64_MAX_EXPONENT) { |
||
192 | /* inf - inf => nan */ |
||
804 | cejka | 193 | /* TODO: fix exception */ |
734 | cejka | 194 | result.binary = FLOAT64_NAN; |
195 | return result; |
||
196 | }; |
||
197 | return a; |
||
198 | } |
||
199 | |||
200 | result.parts.sign = a.parts.sign; |
||
201 | |||
804 | cejka | 202 | frac1 = a.parts.fraction; |
734 | cejka | 203 | exp1 = a.parts.exp; |
804 | cejka | 204 | frac2 = b.parts.fraction; |
734 | cejka | 205 | exp2 = b.parts.exp; |
206 | }; |
||
207 | |||
208 | if (exp1 == 0) { |
||
804 | cejka | 209 | /* both are denormalized */ |
210 | result.parts.fraction = frac1 - frac2; |
||
211 | if (result.parts.fraction > frac1) { |
||
212 | /* TODO: underflow exception */ |
||
734 | cejka | 213 | return result; |
214 | }; |
||
215 | result.parts.exp = 0; |
||
216 | return result; |
||
217 | }; |
||
218 | |||
219 | /* add hidden bit */ |
||
804 | cejka | 220 | frac1 |= FLOAT64_HIDDEN_BIT_MASK; |
734 | cejka | 221 | |
222 | if (exp2 == 0) { |
||
223 | /* denormalized */ |
||
731 | cejka | 224 | --expdiff; |
225 | } else { |
||
734 | cejka | 226 | /* normalized */ |
804 | cejka | 227 | frac2 |= FLOAT64_HIDDEN_BIT_MASK; |
731 | cejka | 228 | }; |
229 | |||
734 | cejka | 230 | /* create some space for rounding */ |
804 | cejka | 231 | frac1 <<= 6; |
232 | frac2 <<= 6; |
||
734 | cejka | 233 | |
804 | cejka | 234 | if (expdiff > FLOAT64_FRACTION_SIZE + 1) { |
731 | cejka | 235 | goto done; |
236 | }; |
||
237 | |||
804 | cejka | 238 | frac1 = frac1 - (frac2 >> expdiff); |
731 | cejka | 239 | done: |
804 | cejka | 240 | /* TODO: find first nonzero digit and shift result and detect possibly underflow */ |
241 | while ((exp1 > 0) && (!(frac1 & (FLOAT64_HIDDEN_BIT_MASK << 6 )))) { |
||
734 | cejka | 242 | --exp1; |
804 | cejka | 243 | frac1 <<= 1; |
244 | /* TODO: fix underflow - frac1 == 0 does not necessary means underflow... */ |
||
731 | cejka | 245 | }; |
246 | |||
804 | cejka | 247 | /* rounding - if first bit after fraction is set then round up */ |
248 | frac1 += 0x20; |
||
731 | cejka | 249 | |
804 | cejka | 250 | if (frac1 & (FLOAT64_HIDDEN_BIT_MASK << 7)) { |
731 | cejka | 251 | ++exp1; |
804 | cejka | 252 | frac1 >>= 1; |
731 | cejka | 253 | }; |
254 | |||
734 | cejka | 255 | /*Clear hidden bit and shift */ |
804 | cejka | 256 | result.parts.fraction = ((frac1 >> 6) & (~FLOAT64_HIDDEN_BIT_MASK)); |
731 | cejka | 257 | result.parts.exp = exp1; |
258 | |||
259 | return result; |
||
734 | cejka | 260 | } |
731 | cejka | 261 | |
1657 | cejka | 262 | |
263 | /** @} |
||
264 | */ |
||
265 |