hysdn: No longer broken on SMP.
[linux-2.6] / include / asm-sparc / sfp-machine.h
1 /* Machine-dependent software floating-point definitions.
2    Sparc userland (_Q_*) version.
3    Copyright (C) 1997,1998,1999 Free Software Foundation, Inc.
4    This file is part of the GNU C Library.
5    Contributed by Richard Henderson (rth@cygnus.com),
6                   Jakub Jelinek (jj@ultra.linux.cz),
7                   David S. Miller (davem@redhat.com) and
8                   Peter Maydell (pmaydell@chiark.greenend.org.uk).
9
10    The GNU C Library is free software; you can redistribute it and/or
11    modify it under the terms of the GNU Library General Public License as
12    published by the Free Software Foundation; either version 2 of the
13    License, or (at your option) any later version.
14
15    The GNU C Library is distributed in the hope that it will be useful,
16    but WITHOUT ANY WARRANTY; without even the implied warranty of
17    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
18    Library General Public License for more details.
19
20    You should have received a copy of the GNU Library General Public
21    License along with the GNU C Library; see the file COPYING.LIB.  If
22    not, write to the Free Software Foundation, Inc.,
23    59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.  */
24
25 #ifndef _SFP_MACHINE_H
26 #define _SFP_MACHINE_H
27
28    
29 #define _FP_W_TYPE_SIZE         32
30 #define _FP_W_TYPE              unsigned long
31 #define _FP_WS_TYPE             signed long
32 #define _FP_I_TYPE              long
33
34 #define _FP_MUL_MEAT_S(R,X,Y)                                   \
35   _FP_MUL_MEAT_1_wide(_FP_WFRACBITS_S,R,X,Y,umul_ppmm)
36 #define _FP_MUL_MEAT_D(R,X,Y)                                   \
37   _FP_MUL_MEAT_2_wide(_FP_WFRACBITS_D,R,X,Y,umul_ppmm)
38 #define _FP_MUL_MEAT_Q(R,X,Y)                                   \
39   _FP_MUL_MEAT_4_wide(_FP_WFRACBITS_Q,R,X,Y,umul_ppmm)
40
41 #define _FP_DIV_MEAT_S(R,X,Y)   _FP_DIV_MEAT_1_udiv(S,R,X,Y)
42 #define _FP_DIV_MEAT_D(R,X,Y)   _FP_DIV_MEAT_2_udiv(D,R,X,Y)
43 #define _FP_DIV_MEAT_Q(R,X,Y)   _FP_DIV_MEAT_4_udiv(Q,R,X,Y)
44
45 #define _FP_NANFRAC_S           ((_FP_QNANBIT_S << 1) - 1)
46 #define _FP_NANFRAC_D           ((_FP_QNANBIT_D << 1) - 1), -1
47 #define _FP_NANFRAC_Q           ((_FP_QNANBIT_Q << 1) - 1), -1, -1, -1
48 #define _FP_NANSIGN_S           0
49 #define _FP_NANSIGN_D           0
50 #define _FP_NANSIGN_Q           0
51
52 #define _FP_KEEPNANFRACP 1
53
54 /* If one NaN is signaling and the other is not,
55  * we choose that one, otherwise we choose X.
56  */
57 /* For _Qp_* and _Q_*, this should prefer X, for
58  * CPU instruction emulation this should prefer Y.
59  * (see SPAMv9 B.2.2 section).
60  */
61 #define _FP_CHOOSENAN(fs, wc, R, X, Y, OP)                      \
62   do {                                                          \
63     if ((_FP_FRAC_HIGH_RAW_##fs(Y) & _FP_QNANBIT_##fs)          \
64         && !(_FP_FRAC_HIGH_RAW_##fs(X) & _FP_QNANBIT_##fs))     \
65       {                                                         \
66         R##_s = X##_s;                                          \
67         _FP_FRAC_COPY_##wc(R,X);                                \
68       }                                                         \
69     else                                                        \
70       {                                                         \
71         R##_s = Y##_s;                                          \
72         _FP_FRAC_COPY_##wc(R,Y);                                \
73       }                                                         \
74     R##_c = FP_CLS_NAN;                                         \
75   } while (0)
76
77 /* Some assembly to speed things up. */
78 #define __FP_FRAC_ADD_3(r2,r1,r0,x2,x1,x0,y2,y1,y0)                     \
79   __asm__ ("addcc %r7,%8,%2\n\t"                                        \
80            "addxcc %r5,%6,%1\n\t"                                       \
81            "addx %r3,%4,%0\n"                                           \
82            : "=r" ((USItype)(r2)),                                      \
83              "=&r" ((USItype)(r1)),                                     \
84              "=&r" ((USItype)(r0))                                      \
85            : "%rJ" ((USItype)(x2)),                                     \
86              "rI" ((USItype)(y2)),                                      \
87              "%rJ" ((USItype)(x1)),                                     \
88              "rI" ((USItype)(y1)),                                      \
89              "%rJ" ((USItype)(x0)),                                     \
90              "rI" ((USItype)(y0))                                       \
91            : "cc")
92
93 #define __FP_FRAC_SUB_3(r2,r1,r0,x2,x1,x0,y2,y1,y0)                     \
94   __asm__ ("subcc %r7,%8,%2\n\t"                                        \
95             "subxcc %r5,%6,%1\n\t"                                      \
96             "subx %r3,%4,%0\n"                                          \
97            : "=r" ((USItype)(r2)),                                      \
98              "=&r" ((USItype)(r1)),                                     \
99              "=&r" ((USItype)(r0))                                      \
100            : "%rJ" ((USItype)(x2)),                                     \
101              "rI" ((USItype)(y2)),                                      \
102              "%rJ" ((USItype)(x1)),                                     \
103              "rI" ((USItype)(y1)),                                      \
104              "%rJ" ((USItype)(x0)),                                     \
105              "rI" ((USItype)(y0))                                       \
106            : "cc")
107
108 #define __FP_FRAC_ADD_4(r3,r2,r1,r0,x3,x2,x1,x0,y3,y2,y1,y0)            \
109   do {                                                                  \
110     /* We need to fool gcc,  as we need to pass more than 10            \
111        input/outputs.  */                                               \
112     register USItype _t1 __asm__ ("g1"), _t2 __asm__ ("g2");            \
113     __asm__ __volatile__ (                                              \
114             "addcc %r8,%9,%1\n\t"                                       \
115             "addxcc %r6,%7,%0\n\t"                                      \
116             "addxcc %r4,%5,%%g2\n\t"                                    \
117             "addx %r2,%3,%%g1\n\t"                                      \
118            : "=&r" ((USItype)(r1)),                                     \
119              "=&r" ((USItype)(r0))                                      \
120            : "%rJ" ((USItype)(x3)),                                     \
121              "rI" ((USItype)(y3)),                                      \
122              "%rJ" ((USItype)(x2)),                                     \
123              "rI" ((USItype)(y2)),                                      \
124              "%rJ" ((USItype)(x1)),                                     \
125              "rI" ((USItype)(y1)),                                      \
126              "%rJ" ((USItype)(x0)),                                     \
127              "rI" ((USItype)(y0))                                       \
128            : "cc", "g1", "g2");                                         \
129     __asm__ __volatile__ ("" : "=r" (_t1), "=r" (_t2));                 \
130     r3 = _t1; r2 = _t2;                                                 \
131   } while (0)
132
133 #define __FP_FRAC_SUB_4(r3,r2,r1,r0,x3,x2,x1,x0,y3,y2,y1,y0)            \
134   do {                                                                  \
135     /* We need to fool gcc,  as we need to pass more than 10            \
136        input/outputs.  */                                               \
137     register USItype _t1 __asm__ ("g1"), _t2 __asm__ ("g2");            \
138     __asm__ __volatile__ (                                              \
139             "subcc %r8,%9,%1\n\t"                                       \
140             "subxcc %r6,%7,%0\n\t"                                      \
141             "subxcc %r4,%5,%%g2\n\t"                                    \
142             "subx %r2,%3,%%g1\n\t"                                      \
143            : "=&r" ((USItype)(r1)),                                     \
144              "=&r" ((USItype)(r0))                                      \
145            : "%rJ" ((USItype)(x3)),                                     \
146              "rI" ((USItype)(y3)),                                      \
147              "%rJ" ((USItype)(x2)),                                     \
148              "rI" ((USItype)(y2)),                                      \
149              "%rJ" ((USItype)(x1)),                                     \
150              "rI" ((USItype)(y1)),                                      \
151              "%rJ" ((USItype)(x0)),                                     \
152              "rI" ((USItype)(y0))                                       \
153            : "cc", "g1", "g2");                                         \
154     __asm__ __volatile__ ("" : "=r" (_t1), "=r" (_t2));                 \
155     r3 = _t1; r2 = _t2;                                                 \
156   } while (0)
157
158 #define __FP_FRAC_DEC_3(x2,x1,x0,y2,y1,y0) __FP_FRAC_SUB_3(x2,x1,x0,x2,x1,x0,y2,y1,y0)
159
160 #define __FP_FRAC_DEC_4(x3,x2,x1,x0,y3,y2,y1,y0) __FP_FRAC_SUB_4(x3,x2,x1,x0,x3,x2,x1,x0,y3,y2,y1,y0)
161
162 #define __FP_FRAC_ADDI_4(x3,x2,x1,x0,i)                                 \
163   __asm__ ("addcc %3,%4,%3\n\t"                                         \
164            "addxcc %2,%%g0,%2\n\t"                                      \
165            "addxcc %1,%%g0,%1\n\t"                                      \
166            "addx %0,%%g0,%0\n\t"                                        \
167            : "=&r" ((USItype)(x3)),                                     \
168              "=&r" ((USItype)(x2)),                                     \
169              "=&r" ((USItype)(x1)),                                     \
170              "=&r" ((USItype)(x0))                                      \
171            : "rI" ((USItype)(i)),                                       \
172              "0" ((USItype)(x3)),                                       \
173              "1" ((USItype)(x2)),                                       \
174              "2" ((USItype)(x1)),                                       \
175              "3" ((USItype)(x0))                                        \
176            : "cc")
177
178 #ifndef CONFIG_SMP
179 extern struct task_struct *last_task_used_math;
180 #endif
181
182 /* Obtain the current rounding mode. */
183 #ifndef FP_ROUNDMODE
184 #ifdef CONFIG_SMP
185 #define FP_ROUNDMODE    ((current->thread.fsr >> 30) & 0x3)
186 #else
187 #define FP_ROUNDMODE    ((last_task_used_math->thread.fsr >> 30) & 0x3)
188 #endif
189 #endif
190
191 /* Exception flags. */
192 #define FP_EX_INVALID           (1 << 4)
193 #define FP_EX_OVERFLOW          (1 << 3)
194 #define FP_EX_UNDERFLOW         (1 << 2)
195 #define FP_EX_DIVZERO           (1 << 1)
196 #define FP_EX_INEXACT           (1 << 0)
197
198 #define FP_HANDLE_EXCEPTIONS return _fex
199
200 #ifdef CONFIG_SMP
201 #define FP_INHIBIT_RESULTS ((current->thread.fsr >> 23) & _fex)
202 #else
203 #define FP_INHIBIT_RESULTS ((last_task_used_math->thread.fsr >> 23) & _fex)
204 #endif
205
206 #ifdef CONFIG_SMP
207 #define FP_TRAPPING_EXCEPTIONS ((current->thread.fsr >> 23) & 0x1f)
208 #else
209 #define FP_TRAPPING_EXCEPTIONS ((last_task_used_math->thread.fsr >> 23) & 0x1f)
210 #endif
211
212 #endif