Deleted Added
full compact
ieeefp.h (175178) ieeefp.h (175179)
1/*-
2 * Copyright (c) 2003 Peter Wemm.
3 * Copyright (c) 1990 Andrew Moore, Talke Studio
4 * All rights reserved.
5 *
6 * Redistribution and use in source and binary forms, with or without
7 * modification, are permitted provided that the following conditions
8 * are met:

--- 18 unchanged lines hidden (view full) ---

27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32 * SUCH DAMAGE.
33 *
34 * from: @(#) ieeefp.h 1.0 (Berkeley) 9/23/93
1/*-
2 * Copyright (c) 2003 Peter Wemm.
3 * Copyright (c) 1990 Andrew Moore, Talke Studio
4 * All rights reserved.
5 *
6 * Redistribution and use in source and binary forms, with or without
7 * modification, are permitted provided that the following conditions
8 * are met:

--- 18 unchanged lines hidden (view full) ---

27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32 * SUCH DAMAGE.
33 *
34 * from: @(#) ieeefp.h 1.0 (Berkeley) 9/23/93
35 * $FreeBSD: head/sys/amd64/include/ieeefp.h 175178 2008-01-09 10:42:47Z bde $
35 * $FreeBSD: head/sys/amd64/include/ieeefp.h 175179 2008-01-09 13:35:31Z bde $
36 */
37
38#ifndef _MACHINE_IEEEFP_H_
39#define _MACHINE_IEEEFP_H_
40
41/*
42 * IEEE floating point type, constant and function definitions.
43 */

--- 72 unchanged lines hidden (view full) ---

116 * SSE mxcsr register bit field offsets
117 */
118#define SSE_STKY_OFF 0 /* exception flags offset */
119#define SSE_DAZ_OFF 6 /* DAZ exception mask offset */
120#define SSE_MSKS_OFF 7 /* other exception masks offset */
121#define SSE_RND_OFF 13 /* rounding control offset */
122#define SSE_FZ_OFF 15 /* flush to zero offset */
123
36 */
37
38#ifndef _MACHINE_IEEEFP_H_
39#define _MACHINE_IEEEFP_H_
40
41/*
42 * IEEE floating point type, constant and function definitions.
43 */

--- 72 unchanged lines hidden (view full) ---

116 * SSE mxcsr register bit field offsets
117 */
118#define SSE_STKY_OFF 0 /* exception flags offset */
119#define SSE_DAZ_OFF 6 /* DAZ exception mask offset */
120#define SSE_MSKS_OFF 7 /* other exception masks offset */
121#define SSE_RND_OFF 13 /* rounding control offset */
122#define SSE_FZ_OFF 15 /* flush to zero offset */
123
124#if defined(__GNUCLIKE_ASM) && defined(__CC_SUPPORTS___INLINE__) \
124#if defined(__GNUCLIKE_ASM) && defined(__CC_SUPPORTS___INLINE) \
125 && !defined(__cplusplus)
126
125 && !defined(__cplusplus)
126
127#define __fldenv(addr) __asm __volatile("fldenv %0" : : "m" (*(addr)))
128#define __fnstenv(addr) __asm __volatile("fnstenv %0" : "=m" (*(addr)))
129#define __fldcw(addr) __asm __volatile("fldcw %0" : : "m" (*(addr)))
127#define __fldcw(addr) __asm __volatile("fldcw %0" : : "m" (*(addr)))
128#define __fldenv(addr) __asm __volatile("fldenv %0" : : "m" (*(addr)))
130#define __fnstcw(addr) __asm __volatile("fnstcw %0" : "=m" (*(addr)))
129#define __fnstcw(addr) __asm __volatile("fnstcw %0" : "=m" (*(addr)))
130#define __fnstenv(addr) __asm __volatile("fnstenv %0" : "=m" (*(addr)))
131#define __fnstsw(addr) __asm __volatile("fnstsw %0" : "=m" (*(addr)))
132#define __ldmxcsr(addr) __asm __volatile("ldmxcsr %0" : : "m" (*(addr)))
133#define __stmxcsr(addr) __asm __volatile("stmxcsr %0" : "=m" (*(addr)))
134
135/*
136 * General notes about conflicting SSE vs FP status bits.
137 * This code assumes that software will not fiddle with the control
138 * bits of the SSE and x87 in such a way to get them out of sync and
139 * still expect this to work. Break this at your peril.
140 * Because I based this on the i386 port, the x87 state is used for
141 * the fpget*() functions, and is shadowed into the SSE state for
142 * the fpset*() functions. For dual source fpget*() functions, I
143 * merge the two together. I think.
144 */
145
131#define __fnstsw(addr) __asm __volatile("fnstsw %0" : "=m" (*(addr)))
132#define __ldmxcsr(addr) __asm __volatile("ldmxcsr %0" : : "m" (*(addr)))
133#define __stmxcsr(addr) __asm __volatile("stmxcsr %0" : "=m" (*(addr)))
134
135/*
136 * General notes about conflicting SSE vs FP status bits.
137 * This code assumes that software will not fiddle with the control
138 * bits of the SSE and x87 in such a way to get them out of sync and
139 * still expect this to work. Break this at your peril.
140 * Because I based this on the i386 port, the x87 state is used for
141 * the fpget*() functions, and is shadowed into the SSE state for
142 * the fpset*() functions. For dual source fpget*() functions, I
143 * merge the two together. I think.
144 */
145
146static __inline__ fp_rnd_t
146static __inline fp_rnd_t
147__fpgetround(void)
148{
149 unsigned short _cw;
150
151 __fnstcw(&_cw);
152 return ((_cw & FP_RND_FLD) >> FP_RND_OFF);
153}
154
147__fpgetround(void)
148{
149 unsigned short _cw;
150
151 __fnstcw(&_cw);
152 return ((_cw & FP_RND_FLD) >> FP_RND_OFF);
153}
154
155static __inline__ fp_rnd_t
155static __inline fp_rnd_t
156__fpsetround(fp_rnd_t _m)
157{
158 unsigned short _cw;
156__fpsetround(fp_rnd_t _m)
157{
158 unsigned short _cw;
159 unsigned int _mxcsr;
159 unsigned _mxcsr;
160 fp_rnd_t _p;
161
162 __fnstcw(&_cw);
163 _p = (_cw & FP_RND_FLD) >> FP_RND_OFF;
164 _cw &= ~FP_RND_FLD;
165 _cw |= (_m << FP_RND_OFF) & FP_RND_FLD;
166 __fldcw(&_cw);
167 __stmxcsr(&_mxcsr);
168 _mxcsr &= ~SSE_RND_FLD;
169 _mxcsr |= (_m << SSE_RND_OFF) & SSE_RND_FLD;
170 __ldmxcsr(&_mxcsr);
171 return (_p);
172}
173
174/*
175 * Get or set the rounding precision for x87 arithmetic operations.
176 * There is no equivalent SSE mode or control.
177 */
178
160 fp_rnd_t _p;
161
162 __fnstcw(&_cw);
163 _p = (_cw & FP_RND_FLD) >> FP_RND_OFF;
164 _cw &= ~FP_RND_FLD;
165 _cw |= (_m << FP_RND_OFF) & FP_RND_FLD;
166 __fldcw(&_cw);
167 __stmxcsr(&_mxcsr);
168 _mxcsr &= ~SSE_RND_FLD;
169 _mxcsr |= (_m << SSE_RND_OFF) & SSE_RND_FLD;
170 __ldmxcsr(&_mxcsr);
171 return (_p);
172}
173
174/*
175 * Get or set the rounding precision for x87 arithmetic operations.
176 * There is no equivalent SSE mode or control.
177 */
178
179static __inline__ fp_prec_t
179static __inline fp_prec_t
180__fpgetprec(void)
181{
182 unsigned short _cw;
183
184 __fnstcw(&_cw);
185 return ((_cw & FP_PRC_FLD) >> FP_PRC_OFF);
186}
187
180__fpgetprec(void)
181{
182 unsigned short _cw;
183
184 __fnstcw(&_cw);
185 return ((_cw & FP_PRC_FLD) >> FP_PRC_OFF);
186}
187
188static __inline__ fp_prec_t
188static __inline fp_prec_t
189__fpsetprec(fp_rnd_t _m)
190{
189__fpsetprec(fp_rnd_t _m)
190{
191 unsigned short _cw;
192 fp_prec_t _p;
191 fp_prec_t _p;
192 unsigned short _cw;
193
194 __fnstcw(&_cw);
195 _p = (_cw & FP_PRC_FLD) >> FP_PRC_OFF;
196 _cw &= ~FP_PRC_FLD;
197 _cw |= (_m << FP_PRC_OFF) & FP_PRC_FLD;
198 __fldcw(&_cw);
199 return (_p);
200}
201
202/*
203 * Get or set the exception mask.
204 * Note that the x87 mask bits are inverted by the API -- a mask bit of 1
205 * means disable for x87 and SSE, but for fp*mask() it means enable.
206 */
207
193
194 __fnstcw(&_cw);
195 _p = (_cw & FP_PRC_FLD) >> FP_PRC_OFF;
196 _cw &= ~FP_PRC_FLD;
197 _cw |= (_m << FP_PRC_OFF) & FP_PRC_FLD;
198 __fldcw(&_cw);
199 return (_p);
200}
201
202/*
203 * Get or set the exception mask.
204 * Note that the x87 mask bits are inverted by the API -- a mask bit of 1
205 * means disable for x87 and SSE, but for fp*mask() it means enable.
206 */
207
208static __inline__ fp_except_t
208static __inline fp_except_t
209__fpgetmask(void)
210{
211 unsigned short _cw;
212
213 __fnstcw(&_cw);
209__fpgetmask(void)
210{
211 unsigned short _cw;
212
213 __fnstcw(&_cw);
214 return ((~_cw) & FP_MSKS_FLD);
214 return ((~_cw & FP_MSKS_FLD) >> FP_MSKS_OFF);
215}
216
215}
216
217static __inline__ fp_except_t
217static __inline fp_except_t
218__fpsetmask(fp_except_t _m)
219{
218__fpsetmask(fp_except_t _m)
219{
220 unsigned short _cw;
221 unsigned int _mxcsr;
222 fp_except_t _p;
220 fp_except_t _p;
221 unsigned _mxcsr;
222 unsigned short _cw;
223
224 __fnstcw(&_cw);
223
224 __fnstcw(&_cw);
225 _p = (~_cw) & FP_MSKS_FLD;
225 _p = (~_cw & FP_MSKS_FLD) >> FP_MSKS_OFF;
226 _cw &= ~FP_MSKS_FLD;
226 _cw &= ~FP_MSKS_FLD;
227 _cw |= (~_m) & FP_MSKS_FLD;
227 _cw |= (~_m >> FP_MSKS_OFF) & FP_MSKS_FLD;
228 __fldcw(&_cw);
229 __stmxcsr(&_mxcsr);
230 /* XXX should we clear non-ieee SSE_DAZ_FLD and SSE_FZ_FLD ? */
231 _mxcsr &= ~SSE_MSKS_FLD;
228 __fldcw(&_cw);
229 __stmxcsr(&_mxcsr);
230 /* XXX should we clear non-ieee SSE_DAZ_FLD and SSE_FZ_FLD ? */
231 _mxcsr &= ~SSE_MSKS_FLD;
232 _mxcsr |= ((~_m) << SSE_MSKS_OFF) & SSE_MSKS_FLD;
232 _mxcsr |= (~_m << SSE_MSKS_OFF) & SSE_MSKS_FLD;
233 __ldmxcsr(&_mxcsr);
234 return (_p);
235}
236
233 __ldmxcsr(&_mxcsr);
234 return (_p);
235}
236
237static __inline__ fp_except_t
237static __inline fp_except_t
238__fpgetsticky(void)
239{
238__fpgetsticky(void)
239{
240 unsigned short _sw;
241 unsigned int _mxcsr;
242 fp_except_t _ex;
240 fp_except_t _ex;
241 unsigned _mxcsr;
242 unsigned short _sw;
243
244 __fnstsw(&_sw);
243
244 __fnstsw(&_sw);
245 _ex = _sw & FP_STKY_FLD;
245 _ex = (_sw & FP_STKY_FLD) >> FP_STKY_OFF;
246 __stmxcsr(&_mxcsr);
246 __stmxcsr(&_mxcsr);
247 _ex |= _mxcsr & SSE_STKY_FLD;
247 _ex |= (_mxcsr & SSE_STKY_FLD) >> SSE_STKY_OFF;
248 return (_ex);
249}
250
248 return (_ex);
249}
250
251#endif /* __GNUCLIKE_ASM && __CC_SUPPORTS___INLINE__ && !__cplusplus */
251#endif /* __GNUCLIKE_ASM && __CC_SUPPORTS___INLINE && !__cplusplus */
252
252
253#if !defined(__IEEEFP_NOINLINES__) && !defined(__cplusplus) \
254 && defined(__GNUCLIKE_ASM) && defined(__CC_SUPPORTS___INLINE__)
253#if !defined(__IEEEFP_NOINLINES__) && !defined(__cplusplus) && \
254 defined(__GNUCLIKE_ASM) && defined(__CC_SUPPORTS___INLINE)
255
256#define fpgetround() __fpgetround()
257#define fpsetround(_m) __fpsetround(_m)
258#define fpgetprec() __fpgetprec()
259#define fpsetprec(_m) __fpsetprec(_m)
260#define fpgetmask() __fpgetmask()
261#define fpsetmask(_m) __fpsetmask(_m)
262#define fpgetsticky() __fpgetsticky()
263
264/* Suppress prototypes in the MI header. */
265#define _IEEEFP_INLINED_ 1
266
255
256#define fpgetround() __fpgetround()
257#define fpsetround(_m) __fpsetround(_m)
258#define fpgetprec() __fpgetprec()
259#define fpsetprec(_m) __fpsetprec(_m)
260#define fpgetmask() __fpgetmask()
261#define fpsetmask(_m) __fpsetmask(_m)
262#define fpgetsticky() __fpgetsticky()
263
264/* Suppress prototypes in the MI header. */
265#define _IEEEFP_INLINED_ 1
266
267#else /* !(!__IEEEFP_NOINLINES__ && !__cplusplus && __GNUCLIKE_ASM
268 && __CC_SUPPORTS___INLINE__) */
267#else /* !(!__IEEEFP_NOINLINES__ && !__cplusplus && __GNUCLIKE_ASM &&
268 __CC_SUPPORTS___INLINE) */
269
270/* Augment the userland declarations. */
271__BEGIN_DECLS
269
270/* Augment the userland declarations. */
271__BEGIN_DECLS
272extern fp_prec_t fpgetprec(void);
273extern fp_prec_t fpsetprec(fp_prec_t);
272fp_prec_t fpgetprec(void);
273fp_prec_t fpsetprec(fp_prec_t);
274__END_DECLS
275
274__END_DECLS
275
276#endif /* !__IEEEFP_NOINLINES__ && !__cplusplus && __GNUCLIKE_ASM
277 && __CC_SUPPORTS___INLINE__ */
276#endif /* !__IEEEFP_NOINLINES__ && !__cplusplus && __GNUCLIKE_ASM &&
277 __CC_SUPPORTS___INLINE */
278
279#endif /* !_MACHINE_IEEEFP_H_ */
278
279#endif /* !_MACHINE_IEEEFP_H_ */