1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 
22 /*
23  * Copyright 2006 Sun Microsystems, Inc.  All rights reserved.
24  * Use is subject to license terms.
25  */
26 
27 /*
28  * Copyright 2011, Richard Lowe.
29  */
30 
31 /* Functions in this file are duplicated in locallibm.il.  Keep them in sync */
32 
33 #ifndef _LIBM_INLINES_H
34 #define	_LIBM_INLINES_H
35 
36 #ifdef __GNUC__
37 
38 #include <sys/types.h>
39 #include <sys/ieeefp.h>
40 
41 #ifdef __cplusplus
42 extern "C" {
43 #endif
44 
45 extern __GNU_INLINE enum fp_class_type
46 fp_classf(float f)
47 {
48 	enum fp_class_type ret;
49 	int fint;		/* scratch for f as int */
50 	uint64_t tmp;
51 
52 	__asm__ __volatile__(
53 	    "fabss  %3,%3\n\t"
54 	    "st	    %3,%1\n\t"
55 	    "ld	    %1,%0\n\t"
56 	    "orcc   %%g0,%0,%%g0\n\t"
57 	    "be,pn  %%icc,2f\n\t"
58 	    "nop\n\t"
59 	    "1:\n\t"
60 	    "sethi  %%hi(0x7f800000),%2\n\t"
61 	    "andcc  %0,%2,%%g0\n\t"
62 	    "bne,pt %%icc,1f\n\t"
63 	    "nop\n\t"
64 	    "or	    %%g0,1,%0\n\t"
65 	    "ba	    2f\n\t"	/* subnormal */
66 	    "nop\n\t"
67 	    "1:\n\t"
68 	    "subcc  %0,%2,%%g0\n\t"
69 	    "bge,pn %%icc,1f\n\t"
70 	    "nop\n\t"
71 	    "or	    %%g0,2,%0\n\t"
72 	    "ba	    2f\n\t"	/* normal */
73 	    "nop\n\t"
74 	    "1:\n\t"
75 	    "bg,pn  %%icc,1f\n\t"
76 	    "nop\n\t"
77 	    "or	    %%g0,3,%0\n\t"
78 	    "ba	    2f\n\t"	/* infinity */
79 	    "nop\n\t"
80 	    "1:\n\t"
81 	    "sethi  %%hi(0x00400000),%2\n\t"
82 	    "andcc  %0,%2,%%g0\n\t"
83 	    "or	    %%g0,4,%0\n\t"
84 	    "bne,pt %%icc,2f\n\t" /* quiet NaN */
85 	    "nop\n\t"
86 	    "or	    %%g0,5,%0\n\t" /* signalling NaN */
87 	    "2:\n\t"
88 	    : "=r" (ret), "=m" (fint), "=r" (tmp), "+f" (f)
89 	    :
90 	    : "cc");
91 
92 	return (ret);
93 }
94 
95 extern __GNU_INLINE enum fp_class_type
96 fp_class(double d)
97 {
98 	enum fp_class_type ret;
99 	uint64_t dint;		/* Scratch for d-as-long */
100 	uint64_t tmp;
101 
102 	__asm__ __volatile__(
103 	    "fabsd  %3,%3\n\t"
104 	    "std    %3,%1\n\t"
105 	    "ldx    %1,%0\n\t"
106 	    "orcc   %%g0,%0,%%g0\n\t"
107 	    "be,pn  %%xcc,2f\n\t"
108 	    "nop\n\t"
109 	    "sethi  %%hi(0x7ff00000),%2\n\t"
110 	    "sllx   %2,32,%2\n\t"
111 	    "andcc  %0,%2,%%g0\n\t"
112 	    "bne,pt %%xcc,1f\n\t"
113 	    "nop\n\t"
114 	    "or	    %%g0,1,%0\n\t"
115 	    "ba	    2f\n\t"
116 	    "nop\n\t"
117 	    "1:\n\t"
118 	    "subcc  %0,%2,%%g0\n\t"
119 	    "bge,pn %%xcc,1f\n\t"
120 	    "nop\n\t"
121 	    "or	    %%g0,2,%0\n\t"
122 	    "ba	    2f\n\t"
123 	    "nop\n\t"
124 	    "1:\n\t"
125 	    "andncc %0,%2,%0\n\t"
126 	    "bne,pn %%xcc,1f\n\t"
127 	    "nop\n\t"
128 	    "or	    %%g0,3,%0\n\t"
129 	    "ba	    2f\n\t"
130 	    "nop\n\t"
131 	    "1:\n\t"
132 	    "sethi  %%hi(0x00080000),%2\n\t"
133 	    "sllx   %2,32,%2\n\t"
134 	    "andcc  %0,%2,%%g0\n\t"
135 	    "or	    %%g0,4,%0\n\t"
136 	    "bne,pt %%xcc,2f\n\t"
137 	    "nop\n\t"
138 	    "or	    %%g0,5,%0\n\t"
139 	    "2:\n\t"
140 	    : "=r" (ret), "=m" (dint), "=r" (tmp), "+e" (d)
141 	    :
142 	    : "cc");
143 
144 	return (ret);
145 }
146 
147 extern __GNU_INLINE float
148 __inline_sqrtf(float f)
149 {
150 	float ret;
151 
152 	__asm__ __volatile__("fsqrts %1,%0\n\t" : "=f" (ret) : "f" (f));
153 	return (ret);
154 }
155 
156 extern __GNU_INLINE double
157 __inline_sqrt(double d)
158 {
159 	double ret;
160 
161 	__asm__ __volatile__("fsqrtd %1,%0\n\t" : "=f" (ret) : "f" (d));
162 	return (ret);
163 }
164 
165 extern __GNU_INLINE int
166 __swapEX(int i)
167 {
168 	int ret;
169 	uint32_t fsr;
170 	uint64_t tmp1, tmp2;
171 
172 	__asm__ __volatile__(
173 	    "and  %4,0x1f,%2\n\t"
174 	    "sll  %2,5,%2\n\t"	/* shift input to aexc bit location */
175 	    ".volatile\n\t"
176 	    "st   %%fsr,%1\n\t"
177 	    "ld	  %1,%0\n\t"	/* %0 = fsr */
178 	    "andn %0,0x3e0,%3\n\t"
179 	    "or   %2,%3,%2\n\t"	/* %2 = new fsr */
180 	    "st	  %2,%1\n\t"
181 	    "ld	  %1,%%fsr\n\t"
182 	    "srl  %0,5,%0\n\t"
183 	    "and  %0,0x1f,%0\n\t"
184 	    ".nonvolatile\n\t"
185 	    : "=r" (ret), "=m" (fsr), "=r" (tmp1), "=r" (tmp2)
186 	    : "r" (i)
187 	    : "cc");
188 
189 	return (ret);
190 }
191 
192 /*
193  * On the SPARC, __swapRP is a no-op; always return 0 for backward
194  * compatibility
195  */
196 /* ARGSUSED */
197 extern __GNU_INLINE enum fp_precision_type
198 __swapRP(enum fp_precision_type i)
199 {
200 	return (0);
201 }
202 
203 extern __GNU_INLINE enum fp_direction_type
204 __swapRD(enum fp_direction_type d)
205 {
206 	enum fp_direction_type ret;
207 	uint32_t fsr;
208 	uint64_t tmp1, tmp2, tmp3;
209 
210 	__asm__ __volatile__(
211 	    "and   %5,0x3,%0\n\t"
212 	    "sll   %0,30,%2\n\t"	/* shift input to RD bit location */
213 	    ".volatile\n\t"
214 	    "st    %%fsr,%1\n\t"
215 	    "ld	   %1,%0\n\t"		/* %0 = fsr */
216 	    /* mask of rounding direction bits */
217 	    "sethi %%hi(0xc0000000),%4\n\t"
218 	    "andn  %0,%4,%3\n\t"
219 	    "or    %2,%3,%2\n\t"	/* %2 = new fsr */
220 	    "st	   %2,%1\n\t"
221 	    "ld	   %1,%%fsr\n\t"
222 	    "srl   %0,30,%0\n\t"
223 	    "and   %0,0x3,%0\n\t"
224 	    ".nonvolatile\n\t"
225 	    : "=r" (ret), "=m" (fsr), "=r" (tmp1), "=r" (tmp2), "=r" (tmp3)
226 	    : "r" (d)
227 	    : "cc");
228 
229 	return (ret);
230 }
231 
232 extern __GNU_INLINE int
233 __swapTE(int i)
234 {
235 	int ret;
236 	uint32_t fsr;
237 	uint64_t tmp1, tmp2, tmp3;
238 
239 	__asm__ __volatile__(
240 	    "and   %5,0x1f,%0\n\t"
241 	    "sll   %0,23,%2\n\t"	/* shift input to TEM bit location */
242 	    ".volatile\n\t"
243 	    "st    %%fsr,%1\n\t"
244 	    "ld	   %1,%0\n\t"		/* %0 = fsr */
245 	    /* mask of TEM (Trap Enable Mode bits) */
246 	    "sethi %%hi(0x0f800000),%4\n\t"
247 	    "andn  %0,%4,%3\n\t"
248 	    "or    %2,%3,%2\n\t"	/* %2 = new fsr */
249 	    "st	   %2,%1\n\t"
250 	    "ld	   %1,%%fsr\n\t"
251 	    "srl   %0,23,%0\n\t"
252 	    "and   %0,0x1f,%0\n\t"
253 	    ".nonvolatile\n\t"
254 	    : "=r" (ret), "=m" (fsr), "=r" (tmp1), "=r" (tmp2), "=r" (tmp3)
255 	    : "r" (i)
256 	    : "cc");
257 
258 	return (ret);
259 }
260 
261 
262 extern __GNU_INLINE double
263 sqrt(double d)
264 {
265 	return (__inline_sqrt(d));
266 }
267 
268 extern __GNU_INLINE float
269 sqrtf(float f)
270 {
271 	return (__inline_sqrtf(f));
272 }
273 
274 extern __GNU_INLINE double
275 fabs(double d)
276 {
277 	double ret;
278 
279 	__asm__ __volatile__("fabsd %1,%0\n\t" : "=e" (ret) : "e" (d));
280 	return (ret);
281 }
282 
283 extern __GNU_INLINE float
284 fabsf(float f)
285 {
286 	float ret;
287 
288 	__asm__ __volatile__("fabss %1,%0\n\t" : "=f" (ret) : "f" (f));
289 	return (ret);
290 }
291 
292 #ifdef __cplusplus
293 }
294 #endif
295 
296 #endif  /* __GNUC__ */
297 
298 #endif /* _LIBM_INLINES_H */
299