xref: /freebsd/lib/msun/src/math_private.h (revision 830940567b49bb0c08dfaed40418999e76616909)
1 /*
2  * ====================================================
3  * Copyright (C) 1993 by Sun Microsystems, Inc. All rights reserved.
4  *
5  * Developed at SunPro, a Sun Microsystems, Inc. business.
6  * Permission to use, copy, modify, and distribute this
7  * software is freely granted, provided that this notice
8  * is preserved.
9  * ====================================================
10  */
11 
12 /*
13  * from: @(#)fdlibm.h 5.1 93/09/24
14  * $FreeBSD$
15  */
16 
17 #ifndef _MATH_PRIVATE_H_
18 #define	_MATH_PRIVATE_H_
19 
20 #include <sys/types.h>
21 #include <machine/endian.h>
22 
23 /*
24  * The original fdlibm code used statements like:
25  *	n0 = ((*(int*)&one)>>29)^1;		* index of high word *
26  *	ix0 = *(n0+(int*)&x);			* high word of x *
27  *	ix1 = *((1-n0)+(int*)&x);		* low word of x *
28  * to dig two 32 bit words out of the 64 bit IEEE floating point
29  * value.  That is non-ANSI, and, moreover, the gcc instruction
30  * scheduler gets it wrong.  We instead use the following macros.
31  * Unlike the original code, we determine the endianness at compile
32  * time, not at run time; I don't see much benefit to selecting
33  * endianness at run time.
34  */
35 
36 /*
37  * A union which permits us to convert between a double and two 32 bit
38  * ints.
39  */
40 
41 #ifdef __arm__
42 #if defined(__VFP_FP__)
43 #define	IEEE_WORD_ORDER	BYTE_ORDER
44 #else
45 #define	IEEE_WORD_ORDER	BIG_ENDIAN
46 #endif
47 #else /* __arm__ */
48 #define	IEEE_WORD_ORDER	BYTE_ORDER
49 #endif
50 
51 #if IEEE_WORD_ORDER == BIG_ENDIAN
52 
53 typedef union
54 {
55   double value;
56   struct
57   {
58     u_int32_t msw;
59     u_int32_t lsw;
60   } parts;
61 } ieee_double_shape_type;
62 
63 #endif
64 
65 #if IEEE_WORD_ORDER == LITTLE_ENDIAN
66 
67 typedef union
68 {
69   double value;
70   struct
71   {
72     u_int32_t lsw;
73     u_int32_t msw;
74   } parts;
75 } ieee_double_shape_type;
76 
77 #endif
78 
79 /* Get two 32 bit ints from a double.  */
80 
81 #define EXTRACT_WORDS(ix0,ix1,d)				\
82 do {								\
83   ieee_double_shape_type ew_u;					\
84   ew_u.value = (d);						\
85   (ix0) = ew_u.parts.msw;					\
86   (ix1) = ew_u.parts.lsw;					\
87 } while (0)
88 
89 /* Get the more significant 32 bit int from a double.  */
90 
91 #define GET_HIGH_WORD(i,d)					\
92 do {								\
93   ieee_double_shape_type gh_u;					\
94   gh_u.value = (d);						\
95   (i) = gh_u.parts.msw;						\
96 } while (0)
97 
98 /* Get the less significant 32 bit int from a double.  */
99 
100 #define GET_LOW_WORD(i,d)					\
101 do {								\
102   ieee_double_shape_type gl_u;					\
103   gl_u.value = (d);						\
104   (i) = gl_u.parts.lsw;						\
105 } while (0)
106 
107 /* Set a double from two 32 bit ints.  */
108 
109 #define INSERT_WORDS(d,ix0,ix1)					\
110 do {								\
111   ieee_double_shape_type iw_u;					\
112   iw_u.parts.msw = (ix0);					\
113   iw_u.parts.lsw = (ix1);					\
114   (d) = iw_u.value;						\
115 } while (0)
116 
117 /* Set the more significant 32 bits of a double from an int.  */
118 
119 #define SET_HIGH_WORD(d,v)					\
120 do {								\
121   ieee_double_shape_type sh_u;					\
122   sh_u.value = (d);						\
123   sh_u.parts.msw = (v);						\
124   (d) = sh_u.value;						\
125 } while (0)
126 
127 /* Set the less significant 32 bits of a double from an int.  */
128 
129 #define SET_LOW_WORD(d,v)					\
130 do {								\
131   ieee_double_shape_type sl_u;					\
132   sl_u.value = (d);						\
133   sl_u.parts.lsw = (v);						\
134   (d) = sl_u.value;						\
135 } while (0)
136 
137 /*
138  * A union which permits us to convert between a float and a 32 bit
139  * int.
140  */
141 
142 typedef union
143 {
144   float value;
145   /* FIXME: Assumes 32 bit int.  */
146   unsigned int word;
147 } ieee_float_shape_type;
148 
149 /* Get a 32 bit int from a float.  */
150 
151 #define GET_FLOAT_WORD(i,d)					\
152 do {								\
153   ieee_float_shape_type gf_u;					\
154   gf_u.value = (d);						\
155   (i) = gf_u.word;						\
156 } while (0)
157 
158 /* Set a float from a 32 bit int.  */
159 
160 #define SET_FLOAT_WORD(d,i)					\
161 do {								\
162   ieee_float_shape_type sf_u;					\
163   sf_u.word = (i);						\
164   (d) = sf_u.value;						\
165 } while (0)
166 
167 #ifdef FLT_EVAL_METHOD
168 /*
169  * Attempt to get strict C99 semantics for assignment with non-C99 compilers.
170  */
171 #if FLT_EVAL_METHOD == 0 || __GNUC__ == 0
172 #define	STRICT_ASSIGN(type, lval, rval)	((lval) = (rval))
173 #else
174 #define	STRICT_ASSIGN(type, lval, rval) do {	\
175 	volatile type __lval;			\
176 						\
177 	if (sizeof(type) >= sizeof(double))	\
178 		(lval) = (rval);		\
179 	else {					\
180 		__lval = (rval);		\
181 		(lval) = __lval;		\
182 	}					\
183 } while (0)
184 #endif
185 #endif
186 
187 /*
188  * Common routine to process the arguments to nan(), nanf(), and nanl().
189  */
190 void _scan_nan(uint32_t *__words, int __num_words, const char *__s);
191 
192 #ifdef _COMPLEX_H
193 
194 /*
195  * C99 specifies that complex numbers have the same representation as
196  * an array of two elements, where the first element is the real part
197  * and the second element is the imaginary part.
198  */
199 typedef union {
200 	float complex f;
201 	float a[2];
202 } float_complex;
203 typedef union {
204 	double complex f;
205 	double a[2];
206 } double_complex;
207 typedef union {
208 	long double complex f;
209 	long double a[2];
210 } long_double_complex;
211 #define	REALPART(z)	((z).a[0])
212 #define	IMAGPART(z)	((z).a[1])
213 
214 /*
215  * Inline functions that can be used to construct complex values.
216  *
217  * The C99 standard intends x+I*y to be used for this, but x+I*y is
218  * currently unusable in general since gcc introduces many overflow,
219  * underflow, sign and efficiency bugs by rewriting I*y as
220  * (0.0+I)*(y+0.0*I) and laboriously computing the full complex product.
221  * In particular, I*Inf is corrupted to NaN+I*Inf, and I*-0 is corrupted
222  * to -0.0+I*0.0.
223  */
224 static __inline float complex
225 cpackf(float x, float y)
226 {
227 	float_complex z;
228 
229 	REALPART(z) = x;
230 	IMAGPART(z) = y;
231 	return (z.f);
232 }
233 
234 static __inline double complex
235 cpack(double x, double y)
236 {
237 	double_complex z;
238 
239 	REALPART(z) = x;
240 	IMAGPART(z) = y;
241 	return (z.f);
242 }
243 
244 static __inline long double complex
245 cpackl(long double x, long double y)
246 {
247 	long_double_complex z;
248 
249 	REALPART(z) = x;
250 	IMAGPART(z) = y;
251 	return (z.f);
252 }
253 #endif /* _COMPLEX_H */
254 
255 #ifdef __GNUCLIKE_ASM
256 
257 /* Asm versions of some functions. */
258 
259 #ifdef __amd64__
260 static __inline int
261 irint(double x)
262 {
263 	int n;
264 
265 	asm("cvtsd2si %1,%0" : "=r" (n) : "x" (x));
266 	return (n);
267 }
268 #define	HAVE_EFFICIENT_IRINT
269 #endif
270 
271 #ifdef __i386__
272 static __inline int
273 irint(double x)
274 {
275 	int n;
276 
277 	asm("fistl %0" : "=m" (n) : "t" (x));
278 	return (n);
279 }
280 #define	HAVE_EFFICIENT_IRINT
281 #endif
282 
283 #endif /* __GNUCLIKE_ASM */
284 
285 /*
286  * ieee style elementary functions
287  *
288  * We rename functions here to improve other sources' diffability
289  * against fdlibm.
290  */
291 #define	__ieee754_sqrt	sqrt
292 #define	__ieee754_acos	acos
293 #define	__ieee754_acosh	acosh
294 #define	__ieee754_log	log
295 #define	__ieee754_atanh	atanh
296 #define	__ieee754_asin	asin
297 #define	__ieee754_atan2	atan2
298 #define	__ieee754_exp	exp
299 #define	__ieee754_cosh	cosh
300 #define	__ieee754_fmod	fmod
301 #define	__ieee754_pow	pow
302 #define	__ieee754_lgamma lgamma
303 #define	__ieee754_gamma	gamma
304 #define	__ieee754_lgamma_r lgamma_r
305 #define	__ieee754_gamma_r gamma_r
306 #define	__ieee754_log10	log10
307 #define	__ieee754_sinh	sinh
308 #define	__ieee754_hypot	hypot
309 #define	__ieee754_j0	j0
310 #define	__ieee754_j1	j1
311 #define	__ieee754_y0	y0
312 #define	__ieee754_y1	y1
313 #define	__ieee754_jn	jn
314 #define	__ieee754_yn	yn
315 #define	__ieee754_remainder remainder
316 #define	__ieee754_scalb	scalb
317 #define	__ieee754_sqrtf	sqrtf
318 #define	__ieee754_acosf	acosf
319 #define	__ieee754_acoshf acoshf
320 #define	__ieee754_logf	logf
321 #define	__ieee754_atanhf atanhf
322 #define	__ieee754_asinf	asinf
323 #define	__ieee754_atan2f atan2f
324 #define	__ieee754_expf	expf
325 #define	__ieee754_coshf	coshf
326 #define	__ieee754_fmodf	fmodf
327 #define	__ieee754_powf	powf
328 #define	__ieee754_lgammaf lgammaf
329 #define	__ieee754_gammaf gammaf
330 #define	__ieee754_lgammaf_r lgammaf_r
331 #define	__ieee754_gammaf_r gammaf_r
332 #define	__ieee754_log10f log10f
333 #define	__ieee754_sinhf	sinhf
334 #define	__ieee754_hypotf hypotf
335 #define	__ieee754_j0f	j0f
336 #define	__ieee754_j1f	j1f
337 #define	__ieee754_y0f	y0f
338 #define	__ieee754_y1f	y1f
339 #define	__ieee754_jnf	jnf
340 #define	__ieee754_ynf	ynf
341 #define	__ieee754_remainderf remainderf
342 #define	__ieee754_scalbf scalbf
343 
344 /* fdlibm kernel function */
345 int	__kernel_rem_pio2(double*,double*,int,int,int);
346 
347 /* double precision kernel functions */
348 #ifdef INLINE_REM_PIO2
349 __inline
350 #endif
351 int	__ieee754_rem_pio2(double,double*);
352 double	__kernel_sin(double,double,int);
353 double	__kernel_cos(double,double);
354 double	__kernel_tan(double,double,int);
355 
356 /* float precision kernel functions */
357 #ifdef INLINE_REM_PIO2F
358 __inline
359 #endif
360 int	__ieee754_rem_pio2f(float,double*);
361 #ifdef INLINE_KERNEL_SINDF
362 __inline
363 #endif
364 float	__kernel_sindf(double);
365 #ifdef INLINE_KERNEL_COSDF
366 __inline
367 #endif
368 float	__kernel_cosdf(double);
369 #ifdef INLINE_KERNEL_TANDF
370 __inline
371 #endif
372 float	__kernel_tandf(double,int);
373 
374 /* long double precision kernel functions */
375 long double __kernel_sinl(long double, long double, int);
376 long double __kernel_cosl(long double, long double);
377 long double __kernel_tanl(long double, long double, int);
378 
379 #endif /* !_MATH_PRIVATE_H_ */
380