1
2 /*
3 ===============================================================================
4
5 This C header file is part of the SoftFloat IEC/IEEE Floating-point
6 Arithmetic Package, Release 2.
7
8 Written by John R. Hauser. This work was made possible in part by the
9 International Computer Science Institute, located at Suite 600, 1947 Center
10 Street, Berkeley, California 94704. Funding was partially provided by the
11 National Science Foundation under grant MIP-9311980. The original version
12 of this code was written as part of a project to build a fixed-point vector
13 processor in collaboration with the University of California at Berkeley,
14 overseen by Profs. Nelson Morgan and John Wawrzynek. More information
15 is available through the Web page `http://HTTP.CS.Berkeley.EDU/~jhauser/
16 arithmetic/softfloat.html'.
17
18 THIS SOFTWARE IS DISTRIBUTED AS IS, FOR FREE. Although reasonable effort
19 has been made to avoid it, THIS SOFTWARE MAY CONTAIN FAULTS THAT WILL AT
20 TIMES RESULT IN INCORRECT BEHAVIOR. USE OF THIS SOFTWARE IS RESTRICTED TO
21 PERSONS AND ORGANIZATIONS WHO CAN AND WILL TAKE FULL RESPONSIBILITY FOR ANY
22 AND ALL LOSSES, COSTS, OR OTHER PROBLEMS ARISING FROM ITS USE.
23
24 Derivative works are acceptable, even for commercial purposes, so long as
25 (1) they include prominent notice that the work is derivative, and (2) they
26 include prominent notice akin to these three paragraphs for those parts of
27 this code that are retained.
28
29 ===============================================================================
30 */
31
32 #ifndef __SOFTFLOAT_H__
33 #define __SOFTFLOAT_H__
34
35 /*
36 -------------------------------------------------------------------------------
37 The macro `FLOATX80' must be defined to enable the extended double-precision
38 floating-point format `floatx80'. If this macro is not defined, the
39 `floatx80' type will not be defined, and none of the functions that either
40 input or output the `floatx80' type will be defined.
41 -------------------------------------------------------------------------------
42 */
43 #ifdef CONFIG_FPE_NWFPE_XP
44 #define FLOATX80
45 #endif
46
47 /*
48 -------------------------------------------------------------------------------
49 Software IEC/IEEE floating-point types.
50 -------------------------------------------------------------------------------
51 */
52 typedef unsigned long int float32;
53 typedef unsigned long long float64;
54 typedef struct {
55 unsigned short high;
56 unsigned long long low;
57 } floatx80;
58
59 /*
60 -------------------------------------------------------------------------------
61 Software IEC/IEEE floating-point underflow tininess-detection mode.
62 -------------------------------------------------------------------------------
63 */
64 extern signed char float_detect_tininess;
65 enum {
66 float_tininess_after_rounding = 0,
67 float_tininess_before_rounding = 1
68 };
69
70 /*
71 -------------------------------------------------------------------------------
72 Software IEC/IEEE floating-point rounding mode.
73 -------------------------------------------------------------------------------
74 */
75 extern signed char float_rounding_mode;
76 enum {
77 float_round_nearest_even = 0,
78 float_round_to_zero = 1,
79 float_round_down = 2,
80 float_round_up = 3
81 };
82
83 /*
84 -------------------------------------------------------------------------------
85 Software IEC/IEEE floating-point exception flags.
86 -------------------------------------------------------------------------------
87 extern signed char float_exception_flags;
88 enum {
89 float_flag_inexact = 1,
90 float_flag_underflow = 2,
91 float_flag_overflow = 4,
92 float_flag_divbyzero = 8,
93 float_flag_invalid = 16
94 };
95
96 ScottB: November 4, 1998
97 Changed the enumeration to match the bit order in the FPA11.
98 */
99
100 extern signed char float_exception_flags;
101 enum {
102 float_flag_invalid = 1,
103 float_flag_divbyzero = 2,
104 float_flag_overflow = 4,
105 float_flag_underflow = 8,
106 float_flag_inexact = 16
107 };
108
109 /*
110 -------------------------------------------------------------------------------
111 Routine to raise any or all of the software IEC/IEEE floating-point
112 exception flags.
113 -------------------------------------------------------------------------------
114 */
115 void float_raise( signed char );
116
117 /*
118 -------------------------------------------------------------------------------
119 Software IEC/IEEE integer-to-floating-point conversion routines.
120 -------------------------------------------------------------------------------
121 */
122 float32 int32_to_float32( signed int );
123 float64 int32_to_float64( signed int );
124 #ifdef FLOATX80
125 floatx80 int32_to_floatx80( signed int );
126 #endif
127
128 /*
129 -------------------------------------------------------------------------------
130 Software IEC/IEEE single-precision conversion routines.
131 -------------------------------------------------------------------------------
132 */
133 signed int float32_to_int32( float32 );
134 signed int float32_to_int32_round_to_zero( float32 );
135 float64 float32_to_float64( float32 );
136 #ifdef FLOATX80
137 floatx80 float32_to_floatx80( float32 );
138 #endif
139
140 /*
141 -------------------------------------------------------------------------------
142 Software IEC/IEEE single-precision operations.
143 -------------------------------------------------------------------------------
144 */
145 float32 float32_round_to_int( float32 );
146 float32 float32_add( float32, float32 );
147 float32 float32_sub( float32, float32 );
148 float32 float32_mul( float32, float32 );
149 float32 float32_div( float32, float32 );
150 float32 float32_rem( float32, float32 );
151 float32 float32_sqrt( float32 );
152 char float32_eq( float32, float32 );
153 char float32_le( float32, float32 );
154 char float32_lt( float32, float32 );
155 char float32_eq_signaling( float32, float32 );
156 char float32_le_quiet( float32, float32 );
157 char float32_lt_quiet( float32, float32 );
158 char float32_is_signaling_nan( float32 );
159
160 /*
161 -------------------------------------------------------------------------------
162 Software IEC/IEEE double-precision conversion routines.
163 -------------------------------------------------------------------------------
164 */
165 signed int float64_to_int32( float64 );
166 signed int float64_to_int32_round_to_zero( float64 );
167 float32 float64_to_float32( float64 );
168 #ifdef FLOATX80
169 floatx80 float64_to_floatx80( float64 );
170 #endif
171
172 /*
173 -------------------------------------------------------------------------------
174 Software IEC/IEEE double-precision operations.
175 -------------------------------------------------------------------------------
176 */
177 float64 float64_round_to_int( float64 );
178 float64 float64_add( float64, float64 );
179 float64 float64_sub( float64, float64 );
180 float64 float64_mul( float64, float64 );
181 float64 float64_div( float64, float64 );
182 float64 float64_rem( float64, float64 );
183 float64 float64_sqrt( float64 );
184 char float64_eq( float64, float64 );
185 char float64_le( float64, float64 );
186 char float64_lt( float64, float64 );
187 char float64_eq_signaling( float64, float64 );
188 char float64_le_quiet( float64, float64 );
189 char float64_lt_quiet( float64, float64 );
190 char float64_is_signaling_nan( float64 );
191
192 #ifdef FLOATX80
193
194 /*
195 -------------------------------------------------------------------------------
196 Software IEC/IEEE extended double-precision conversion routines.
197 -------------------------------------------------------------------------------
198 */
199 signed int floatx80_to_int32( floatx80 );
200 signed int floatx80_to_int32_round_to_zero( floatx80 );
201 float32 floatx80_to_float32( floatx80 );
202 float64 floatx80_to_float64( floatx80 );
203
204 /*
205 -------------------------------------------------------------------------------
206 Software IEC/IEEE extended double-precision rounding precision. Valid
207 values are 32, 64, and 80.
208 -------------------------------------------------------------------------------
209 */
210 extern signed char floatx80_rounding_precision;
211
212 /*
213 -------------------------------------------------------------------------------
214 Software IEC/IEEE extended double-precision operations.
215 -------------------------------------------------------------------------------
216 */
217 floatx80 floatx80_round_to_int( floatx80 );
218 floatx80 floatx80_add( floatx80, floatx80 );
219 floatx80 floatx80_sub( floatx80, floatx80 );
220 floatx80 floatx80_mul( floatx80, floatx80 );
221 floatx80 floatx80_div( floatx80, floatx80 );
222 floatx80 floatx80_rem( floatx80, floatx80 );
223 floatx80 floatx80_sqrt( floatx80 );
224 char floatx80_eq( floatx80, floatx80 );
225 char floatx80_le( floatx80, floatx80 );
226 char floatx80_lt( floatx80, floatx80 );
227 char floatx80_eq_signaling( floatx80, floatx80 );
228 char floatx80_le_quiet( floatx80, floatx80 );
229 char floatx80_lt_quiet( floatx80, floatx80 );
230 char floatx80_is_signaling_nan( floatx80 );
231
232 #endif
233
extractFloat32Sign(float32 a)234 static inline flag extractFloat32Sign(float32 a)
235 {
236 return a >> 31;
237 }
238
float32_eq_nocheck(float32 a,float32 b)239 static inline flag float32_eq_nocheck(float32 a, float32 b)
240 {
241 return (a == b) || ((bits32) ((a | b) << 1) == 0);
242 }
243
float32_lt_nocheck(float32 a,float32 b)244 static inline flag float32_lt_nocheck(float32 a, float32 b)
245 {
246 flag aSign, bSign;
247
248 aSign = extractFloat32Sign(a);
249 bSign = extractFloat32Sign(b);
250 if (aSign != bSign)
251 return aSign && ((bits32) ((a | b) << 1) != 0);
252 return (a != b) && (aSign ^ (a < b));
253 }
254
extractFloat64Sign(float64 a)255 static inline flag extractFloat64Sign(float64 a)
256 {
257 return a >> 63;
258 }
259
float64_eq_nocheck(float64 a,float64 b)260 static inline flag float64_eq_nocheck(float64 a, float64 b)
261 {
262 return (a == b) || ((bits64) ((a | b) << 1) == 0);
263 }
264
float64_lt_nocheck(float64 a,float64 b)265 static inline flag float64_lt_nocheck(float64 a, float64 b)
266 {
267 flag aSign, bSign;
268
269 aSign = extractFloat64Sign(a);
270 bSign = extractFloat64Sign(b);
271 if (aSign != bSign)
272 return aSign && ((bits64) ((a | b) << 1) != 0);
273 return (a != b) && (aSign ^ (a < b));
274 }
275
276 #endif
277