xref: /openbsd/sys/lib/libkern/softfloat.h (revision e5dd7070)
1 /*	$OpenBSD: softfloat.h,v 1.5 2014/06/10 04:16:57 deraadt Exp $	*/
2 /*	$NetBSD: softfloat.h,v 1.1 2001/04/26 03:10:48 ross Exp $	*/
3 
4 /* This is a derivative work. */
5 
6 /*-
7  * Copyright (c) 2001 The NetBSD Foundation, Inc.
8  * All rights reserved.
9  *
10  * This code is derived from software contributed to The NetBSD Foundation
11  * by Ross Harvey.
12  *
13  * Redistribution and use in source and binary forms, with or without
14  * modification, are permitted provided that the following conditions
15  * are met:
16  * 1. Redistributions of source code must retain the above copyright
17  *    notice, this list of conditions and the following disclaimer.
18  * 2. Redistributions in binary form must reproduce the above copyright
19  *    notice, this list of conditions and the following disclaimer in the
20  *    documentation and/or other materials provided with the distribution.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
23  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
24  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
25  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
26  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
27  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
28  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
29  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
30  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
31  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
32  * POSSIBILITY OF SUCH DAMAGE.
33  */
34 
35 /*
36 ===============================================================================
37 
38 This C header file is part of the SoftFloat IEC/IEEE Floating-point
39 Arithmetic Package, Release 2a.
40 
41 Written by John R. Hauser.  This work was made possible in part by the
42 International Computer Science Institute, located at Suite 600, 1947 Center
43 Street, Berkeley, California 94704.  Funding was partially provided by the
44 National Science Foundation under grant MIP-9311980.  The original version
45 of this code was written as part of a project to build a fixed-point vector
46 processor in collaboration with the University of California at Berkeley,
47 overseen by Profs. Nelson Morgan and John Wawrzynek.  More information
48 is available through the Web page `http://HTTP.CS.Berkeley.EDU/~jhauser/
49 arithmetic/SoftFloat.html'.
50 
51 THIS SOFTWARE IS DISTRIBUTED AS IS, FOR FREE.  Although reasonable
52 effort has been made to avoid it, THIS SOFTWARE MAY CONTAIN FAULTS THAT
53 WILL AT TIMES RESULT IN INCORRECT BEHAVIOR.  USE OF THIS SOFTWARE IS
54 RESTRICTED TO PERSONS AND ORGANIZATIONS WHO CAN AND WILL TAKE FULL
55 RESPONSIBILITY FOR ALL LOSSES, COSTS, OR OTHER PROBLEMS ARISING FROM
56 THEIR OWN USE OF THE SOFTWARE, AND WHO ALSO EFFECTIVELY INDEMNIFY
57 (possibly via similar legal warning) JOHN HAUSER AND THE INTERNATIONAL
58 COMPUTER SCIENCE INSTITUTE AGAINST ALL LOSSES, COSTS, OR OTHER PROBLEMS
59 ARISING FROM THE USE OF THE SOFTWARE BY THEIR CUSTOMERS AND CLIENTS.
60 
61 Derivative works are acceptable, even for commercial purposes, so long as
62 (1) they include prominent notice that the work is derivative, and (2) they
63 include prominent notice akin to these four paragraphs for those parts of
64 this code that are retained.
65 
66 ===============================================================================
67 */
68 
69 #include <sys/types.h>
70 
71 #include "machine/ieeefp.h"
72 #include <sys/endian.h>
73 
74 /*
75 -------------------------------------------------------------------------------
76 The macro `FLOATX80' must be defined to enable the extended double-precision
77 floating-point format `floatx80'.  If this macro is not defined, the
78 `floatx80' type will not be defined, and none of the functions that either
79 input or output the `floatx80' type will be defined.  The same applies to
80 the `FLOAT128' macro and the quadruple-precision format `float128'.
81 -------------------------------------------------------------------------------
82 */
83 /* #define FLOATX80 */
84 /* #define FLOAT128 */
85 
86 /*
87 -------------------------------------------------------------------------------
88 Software IEC/IEEE floating-point types.
89 -------------------------------------------------------------------------------
90 */
91 typedef u_int32_t float32;
92 typedef u_int64_t float64;
93 #ifdef FLOATX80
94 typedef struct {
95 #if BYTE_ORDER == BIG_ENDIAN
96     u_int16_t high;
97     u_int64_t low;
98 #else
99     u_int64_t low;
100     u_int16_t high;
101 #endif
102 } floatx80;
103 #endif
104 #ifdef FLOAT128
105 typedef struct {
106     u_int64_t high, low;
107 } float128;
108 #endif
109 
110 /*
111  * Some of the global variables that used to be here have been removed for
112  * fairly obvious (defopt-MULTIPROCESSOR) reasons.  The rest (which don't
113  * change dynamically) will be removed later. [ross]
114  */
115 
116 #define float_rounding_mode() fpgetround()
117 
118 /*
119 -------------------------------------------------------------------------------
120 Software IEC/IEEE floating-point underflow tininess-detection mode.
121 -------------------------------------------------------------------------------
122 */
123 
124 extern int float_detect_tininess;
125 enum {
126     float_tininess_after_rounding  = 1,
127     float_tininess_before_rounding = 0
128 };
129 
130 /*
131 -------------------------------------------------------------------------------
132 Software IEC/IEEE floating-point rounding mode.
133 -------------------------------------------------------------------------------
134 */
135 
136 enum {
137     float_round_nearest_even = FP_RN,
138     float_round_to_zero      = FP_RZ,
139     float_round_down         = FP_RM,
140     float_round_up           = FP_RP
141 };
142 
143 /*
144 -------------------------------------------------------------------------------
145 Software IEC/IEEE floating-point exception flags.
146 -------------------------------------------------------------------------------
147 */
148 
149 enum {
150     float_flag_inexact   =  FP_X_IMP,
151     float_flag_underflow =  FP_X_UFL,
152     float_flag_overflow  =  FP_X_OFL,
153     float_flag_divbyzero =  FP_X_DZ,
154     float_flag_invalid   =  FP_X_INV
155 };
156 
157 /*
158 -------------------------------------------------------------------------------
159 Software IEC/IEEE integer-to-floating-point conversion routines.
160 -------------------------------------------------------------------------------
161 */
162 float32 int32_to_float32( int );
163 float64 int32_to_float64( int );
164 #ifdef FLOATX80
165 floatx80 int32_to_floatx80( int );
166 #endif
167 #ifdef FLOAT128
168 float128 int32_to_float128( int );
169 #endif
170 #ifndef SOFTFLOAT_FOR_GCC /* __floatdi?f is in libgcc2.c */
171 float32 int64_to_float32( int64_t );
172 float64 int64_to_float64( int64_t );
173 #ifdef FLOATX80
174 floatx80 int64_to_floatx80( int64_t );
175 #endif
176 #ifdef FLOAT128
177 float128 int64_to_float128( int64_t );
178 #endif
179 #endif
180 
181 /*
182 -------------------------------------------------------------------------------
183 Software IEC/IEEE single-precision conversion routines.
184 -------------------------------------------------------------------------------
185 */
186 int float32_to_int32( float32 );
187 int float32_to_int32_round_to_zero( float32 );
188 #ifndef SOFTFLOAT_FOR_GCC /* __fix?fdi provided by libgcc2.c */
189 int64_t float32_to_int64( float32 );
190 int64_t float32_to_int64_round_to_zero( float32 );
191 #endif
192 float64 float32_to_float64( float32 );
193 #ifdef FLOATX80
194 floatx80 float32_to_floatx80( float32 );
195 #endif
196 #ifdef FLOAT128
197 float128 float32_to_float128( float32 );
198 #endif
199 
200 /*
201 -------------------------------------------------------------------------------
202 Software IEC/IEEE single-precision operations.
203 -------------------------------------------------------------------------------
204 */
205 float32 float32_round_to_int( float32 );
206 float32 float32_add( float32, float32 );
207 float32 float32_sub( float32, float32 );
208 float32 float32_mul( float32, float32 );
209 float32 float32_div( float32, float32 );
210 float32 float32_rem( float32, float32 );
211 float32 float32_sqrt( float32 );
212 int float32_eq( float32, float32 );
213 int float32_le( float32, float32 );
214 int float32_lt( float32, float32 );
215 int float32_eq_signaling( float32, float32 );
216 int float32_le_quiet( float32, float32 );
217 int float32_lt_quiet( float32, float32 );
218 #ifndef SOFTFLOAT_FOR_GCC
219 int float32_is_signaling_nan( float32 );
220 #endif
221 
222 /*
223 -------------------------------------------------------------------------------
224 Software IEC/IEEE double-precision conversion routines.
225 -------------------------------------------------------------------------------
226 */
227 int float64_to_int32( float64 );
228 int float64_to_int32_round_to_zero( float64 );
229 #ifndef SOFTFLOAT_FOR_GCC /* __fix?fdi provided by libgcc2.c */
230 int64_t float64_to_int64( float64 );
231 #ifdef __alpha__
232 int64_t float64_to_int64_no_overflow( float64 );
233 #endif /* __alpha__ */
234 int64_t float64_to_int64_round_to_zero( float64 );
235 #endif
236 float32 float64_to_float32( float64 );
237 #ifdef FLOATX80
238 floatx80 float64_to_floatx80( float64 );
239 #endif
240 #ifdef FLOAT128
241 float128 float64_to_float128( float64 );
242 #endif
243 
244 /*
245 -------------------------------------------------------------------------------
246 Software IEC/IEEE double-precision operations.
247 -------------------------------------------------------------------------------
248 */
249 #define float64_default_nan 0xFFF8000000000000LL
250 
251 static __inline int
252 float64_is_nan(float64 a)
253 {
254 	return 0xFFE0000000000000LL < a << 1;
255 }
256 
257 static __inline int
258 float64_is_signaling_nan(float64 a)
259 {
260 	return (a >> 51 & 0xFFF) == 0xFFE && (a & 0x0007FFFFFFFFFFFFLL);
261 }
262 
263 float64 float64_round_to_int( float64 );
264 float64 float64_add( float64, float64 );
265 float64 float64_sub( float64, float64 );
266 float64 float64_mul( float64, float64 );
267 float64 float64_div( float64, float64 );
268 float64 float64_rem( float64, float64 );
269 float64 float64_sqrt( float64 );
270 int float64_eq( float64, float64 );
271 int float64_le( float64, float64 );
272 int float64_lt( float64, float64 );
273 int float64_eq_signaling( float64, float64 );
274 int float64_le_quiet( float64, float64 );
275 int float64_lt_quiet( float64, float64 );
276 #ifndef SOFTFLOAT_FOR_GCC
277 int float64_is_signaling_nan( float64 );
278 #endif
279 
280 #ifdef FLOATX80
281 
282 /*
283 -------------------------------------------------------------------------------
284 Software IEC/IEEE extended double-precision conversion routines.
285 -------------------------------------------------------------------------------
286 */
287 int floatx80_to_int32( floatx80 );
288 int floatx80_to_int32_round_to_zero( floatx80 );
289 int64_t floatx80_to_int64( floatx80 );
290 int64_t floatx80_to_int64_round_to_zero( floatx80 );
291 float32 floatx80_to_float32( floatx80 );
292 float64 floatx80_to_float64( floatx80 );
293 #ifdef FLOAT128
294 float128 floatx80_to_float128( floatx80 );
295 #endif
296 
297 /*
298 -------------------------------------------------------------------------------
299 Software IEC/IEEE extended double-precision rounding precision.  Valid
300 values are 32, 64, and 80.
301 -------------------------------------------------------------------------------
302 */
303 extern int floatx80_rounding_precision;
304 
305 /*
306 -------------------------------------------------------------------------------
307 Software IEC/IEEE extended double-precision operations.
308 -------------------------------------------------------------------------------
309 */
310 floatx80 floatx80_round_to_int( floatx80 );
311 floatx80 floatx80_add( floatx80, floatx80 );
312 floatx80 floatx80_sub( floatx80, floatx80 );
313 floatx80 floatx80_mul( floatx80, floatx80 );
314 floatx80 floatx80_div( floatx80, floatx80 );
315 floatx80 floatx80_rem( floatx80, floatx80 );
316 floatx80 floatx80_sqrt( floatx80 );
317 int floatx80_eq( floatx80, floatx80 );
318 int floatx80_le( floatx80, floatx80 );
319 int floatx80_lt( floatx80, floatx80 );
320 int floatx80_eq_signaling( floatx80, floatx80 );
321 int floatx80_le_quiet( floatx80, floatx80 );
322 int floatx80_lt_quiet( floatx80, floatx80 );
323 int floatx80_is_signaling_nan( floatx80 );
324 
325 #endif
326 
327 #ifdef FLOAT128
328 
329 /*
330 -------------------------------------------------------------------------------
331 Software IEC/IEEE quadruple-precision conversion routines.
332 -------------------------------------------------------------------------------
333 */
334 int float128_to_int32( float128 );
335 int float128_to_int32_round_to_zero( float128 );
336 int64_t float128_to_int64( float128 );
337 int64_t float128_to_int64_round_to_zero( float128 );
338 float32 float128_to_float32( float128 );
339 float64 float128_to_float64( float128 );
340 #ifdef FLOATX80
341 floatx80 float128_to_floatx80( float128 );
342 #endif
343 
344 /*
345 -------------------------------------------------------------------------------
346 Software IEC/IEEE quadruple-precision operations.
347 -------------------------------------------------------------------------------
348 */
349 float128 float128_round_to_int( float128 );
350 float128 float128_add( float128, float128 );
351 float128 float128_sub( float128, float128 );
352 float128 float128_mul( float128, float128 );
353 float128 float128_div( float128, float128 );
354 float128 float128_rem( float128, float128 );
355 float128 float128_sqrt( float128 );
356 int float128_eq( float128, float128 );
357 int float128_le( float128, float128 );
358 int float128_lt( float128, float128 );
359 int float128_eq_signaling( float128, float128 );
360 int float128_le_quiet( float128, float128 );
361 int float128_lt_quiet( float128, float128 );
362 int float128_is_signaling_nan( float128 );
363 
364 #endif
365