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