fpu_emulate.h revision 1.10 1 /* $NetBSD: fpu_emulate.h,v 1.10 2003/09/22 14:18:35 cl Exp $ */
2
3 /*
4 * Copyright (c) 1995 Gordon Ross
5 * Copyright (c) 1995 Ken Nakata
6 * All rights reserved.
7 *
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
10 * are met:
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 * 3. The name of the author may not be used to endorse or promote products
17 * derived from this software without specific prior written permission.
18 * 4. All advertising materials mentioning features or use of this software
19 * must display the following acknowledgement:
20 * This product includes software developed by Gordon Ross
21 *
22 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
23 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
24 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
25 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
26 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
27 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
28 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
29 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
30 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
31 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
32 */
33
34 #ifndef _FPU_EMULATE_H_
35 #define _FPU_EMULATE_H_
36
37 #include <sys/types.h>
38 #include <sys/signal.h>
39 #include <sys/signalvar.h>
40 #include <sys/siginfo.h>
41
42 /*
43 * Floating point emulator (tailored for SPARC/modified for m68k, but
44 * structurally machine-independent).
45 *
46 * Floating point numbers are carried around internally in an `expanded'
47 * or `unpacked' form consisting of:
48 * - sign
49 * - unbiased exponent
50 * - mantissa (`1.' + 80-bit fraction + guard + round)
51 * - sticky bit
52 * Any implied `1' bit is inserted, giving a 81-bit mantissa that is
53 * always nonzero. Additional low-order `guard' and `round' bits are
54 * scrunched in, making the entire mantissa 83 bits long. This is divided
55 * into three 32-bit words, with `spare' bits left over in the upper part
56 * of the top word (the high bits of fp_mant[0]). An internal `exploded'
57 * number is thus kept within the half-open interval [1.0,2.0) (but see
58 * the `number classes' below). This holds even for denormalized numbers:
59 * when we explode an external denorm, we normalize it, introducing low-order
60 * zero bits, so that the rest of the code always sees normalized values.
61 *
62 * Note that a number of our algorithms use the `spare' bits at the top.
63 * The most demanding algorithm---the one for sqrt---depends on two such
64 * bits, so that it can represent values up to (but not including) 8.0,
65 * and then it needs a carry on top of that, so that we need three `spares'.
66 *
67 * The sticky-word is 32 bits so that we can use `OR' operators to goosh
68 * whole words from the mantissa into it.
69 *
70 * All operations are done in this internal extended precision. According
71 * to Hennesey & Patterson, Appendix A, rounding can be repeated---that is,
72 * it is OK to do a+b in extended precision and then round the result to
73 * single precision---provided single, double, and extended precisions are
74 * `far enough apart' (they always are), but we will try to avoid any such
75 * extra work where possible.
76 */
77 struct fpn {
78 int fp_class; /* see below */
79 int fp_sign; /* 0 => positive, 1 => negative */
80 int fp_exp; /* exponent (unbiased) */
81 int fp_sticky; /* nonzero bits lost at right end */
82 u_int fp_mant[3]; /* 83-bit mantissa */
83 };
84
85 #define FP_NMANT 83 /* total bits in mantissa (incl g,r) */
86 #define FP_NG 2 /* number of low-order guard bits */
87 #define FP_LG ((FP_NMANT - 1) & 31) /* log2(1.0) for fp_mant[0] */
88 #define FP_QUIETBIT (1 << (FP_LG - 1)) /* Quiet bit in NaNs (0.5) */
89 #define FP_1 (1 << FP_LG) /* 1.0 in fp_mant[0] */
90 #define FP_2 (1 << (FP_LG + 1)) /* 2.0 in fp_mant[0] */
91
92 #define CPYFPN(dst, src) \
93 if ((dst) != (src)) { \
94 (dst)->fp_class = (src)->fp_class; \
95 (dst)->fp_sign = (src)->fp_sign; \
96 (dst)->fp_exp = (src)->fp_exp; \
97 (dst)->fp_sticky = (src)->fp_sticky; \
98 (dst)->fp_mant[0] = (src)->fp_mant[0]; \
99 (dst)->fp_mant[1] = (src)->fp_mant[1]; \
100 (dst)->fp_mant[2] = (src)->fp_mant[2]; \
101 }
102
103 /*
104 * Number classes. Since zero, Inf, and NaN cannot be represented using
105 * the above layout, we distinguish these from other numbers via a class.
106 */
107 #define FPC_SNAN -2 /* signalling NaN (sign irrelevant) */
108 #define FPC_QNAN -1 /* quiet NaN (sign irrelevant) */
109 #define FPC_ZERO 0 /* zero (sign matters) */
110 #define FPC_NUM 1 /* number (sign matters) */
111 #define FPC_INF 2 /* infinity (sign matters) */
112
113 #define ISNAN(fp) ((fp)->fp_class < 0)
114 #define ISZERO(fp) ((fp)->fp_class == 0)
115 #define ISINF(fp) ((fp)->fp_class == FPC_INF)
116
117 /*
118 * ORDER(x,y) `sorts' a pair of `fpn *'s so that the right operand (y) points
119 * to the `more significant' operand for our purposes. Appendix N says that
120 * the result of a computation involving two numbers are:
121 *
122 * If both are SNaN: operand 2, converted to Quiet
123 * If only one is SNaN: the SNaN operand, converted to Quiet
124 * If both are QNaN: operand 2
125 * If only one is QNaN: the QNaN operand
126 *
127 * In addition, in operations with an Inf operand, the result is usually
128 * Inf. The class numbers are carefully arranged so that if
129 * (unsigned)class(op1) > (unsigned)class(op2)
130 * then op1 is the one we want; otherwise op2 is the one we want.
131 */
132 #define ORDER(x, y) { \
133 if ((u_int)(x)->fp_class > (u_int)(y)->fp_class) \
134 SWAP(x, y); \
135 }
136 #define SWAP(x, y) { \
137 register struct fpn *swap; \
138 swap = (x), (x) = (y), (y) = swap; \
139 }
140
141 /*
142 * Emulator state.
143 */
144 struct fpemu {
145 struct frame *fe_frame; /* integer regs, etc */
146 struct fpframe *fe_fpframe; /* FP registers, etc */
147 u_int fe_fpsr; /* fpsr copy (modified during op) */
148 u_int fe_fpcr; /* fpcr copy */
149 struct fpn fe_f1; /* operand 1 */
150 struct fpn fe_f2; /* operand 2, if required */
151 struct fpn fe_f3; /* available storage for result */
152 };
153
154 /*****************************************************************************
155 * End of definitions derived from Sparc FPE
156 *****************************************************************************/
157
158 /*
159 * Internal info about a decoded effective address.
160 */
161 struct insn_ea {
162 int ea_regnum;
163 int ea_ext[3]; /* extension words if any */
164 int ea_flags; /* flags == 0 means mode 2: An@ */
165 #define EA_DIRECT 0x001 /* mode [01]: Dn or An */
166 #define EA_PREDECR 0x002 /* mode 4: An@- */
167 #define EA_POSTINCR 0x004 /* mode 3: An@+ */
168 #define EA_OFFSET 0x008 /* mode 5 or (7,2): APC@(d16) */
169 #define EA_INDEXED 0x010 /* mode 6 or (7,3): APC@(Xn:*:*,d8) etc */
170 #define EA_ABS 0x020 /* mode (7,[01]): abs */
171 #define EA_PC_REL 0x040 /* mode (7,[23]): PC@(d16) etc */
172 #define EA_IMMED 0x080 /* mode (7,4): #immed */
173 #define EA_MEM_INDIR 0x100 /* mode 6 or (7,3): APC@(Xn:*:*,*)@(*) etc */
174 #define EA_BASE_SUPPRSS 0x200 /* mode 6 or (7,3): base register suppressed */
175 #define EA_FRAME_EA 0x400 /* MC68LC040 only: precalculated EA from
176 format 4 stack frame */
177 int ea_moffs; /* offset used for fmoveMulti */
178 };
179
180 #define ea_offset ea_ext[0] /* mode 5: offset word */
181 #define ea_absaddr ea_ext[0] /* mode (7,[01]): absolute address */
182 #define ea_immed ea_ext /* mode (7,4): immediate value */
183 #define ea_basedisp ea_ext[0] /* mode 6: base displacement */
184 #define ea_outerdisp ea_ext[1] /* mode 6: outer displacement */
185 #define ea_idxreg ea_ext[2] /* mode 6: index register number */
186 #define ea_fea ea_ext[0] /* MC68LC040 only: frame EA */
187
188 struct instruction {
189 u_int is_pc; /* insn's address */
190 u_int is_nextpc; /* next PC */
191 int is_advance; /* length of instruction */
192 int is_datasize; /* size of memory operand */
193 int is_opcode; /* opcode word */
194 int is_word1; /* second word */
195 struct insn_ea is_ea; /* decoded effective address mode */
196 };
197
198 /*
199 * FP data types
200 */
201 #define FTYPE_LNG 0 /* Long Word Integer */
202 #define FTYPE_SNG 1 /* Single Prec */
203 #define FTYPE_EXT 2 /* Extended Prec */
204 #define FTYPE_BCD 3 /* Packed BCD */
205 #define FTYPE_WRD 4 /* Word Integer */
206 #define FTYPE_DBL 5 /* Double Prec */
207 #define FTYPE_BYT 6 /* Byte Integer */
208
209 /*
210 * MC68881/68882 FPcr bit definitions (should these go to <m68k/reg.h>
211 * or <m68k/fpu.h> or something?)
212 */
213
214 /* fpsr */
215 #define FPSR_CCB 0xff000000
216 # define FPSR_NEG 0x08000000
217 # define FPSR_ZERO 0x04000000
218 # define FPSR_INF 0x02000000
219 # define FPSR_NAN 0x01000000
220 #define FPSR_QTT 0x00ff0000
221 # define FPSR_QSG 0x00800000
222 # define FPSR_QUO 0x007f0000
223 #define FPSR_EXCP 0x0000ff00
224 # define FPSR_BSUN 0x00008000
225 # define FPSR_SNAN 0x00004000
226 # define FPSR_OPERR 0x00002000
227 # define FPSR_OVFL 0x00001000
228 # define FPSR_UNFL 0x00000800
229 # define FPSR_DZ 0x00000400
230 # define FPSR_INEX2 0x00000200
231 # define FPSR_INEX1 0x00000100
232 #define FPSR_AEX 0x000000ff
233 # define FPSR_AIOP 0x00000080
234 # define FPSR_AOVFL 0x00000040
235 # define FPSR_AUNFL 0x00000020
236 # define FPSR_ADZ 0x00000010
237 # define FPSR_AINEX 0x00000008
238
239 /* fpcr */
240 #define FPCR_EXCP FPSR_EXCP
241 # define FPCR_BSUN FPSR_BSUN
242 # define FPCR_SNAN FPSR_SNAN
243 # define FPCR_OPERR FPSR_OPERR
244 # define FPCR_OVFL FPSR_OVFL
245 # define FPCR_UNFL FPSR_UNFL
246 # define FPCR_DZ FPSR_DZ
247 # define FPCR_INEX2 FPSR_INEX2
248 # define FPCR_INEX1 FPSR_INEX1
249 #define FPCR_MODE 0x000000ff
250 # define FPCR_PREC 0x000000c0
251 # define FPCR_EXTD 0x00000000
252 # define FPCR_SNGL 0x00000040
253 # define FPCR_DBL 0x00000080
254 # define FPCR_ROUND 0x00000030
255 # define FPCR_NEAR 0x00000000
256 # define FPCR_ZERO 0x00000010
257 # define FPCR_MINF 0x00000020
258 # define FPCR_PINF 0x00000030
259
260 /*
261 * Other functions.
262 */
263
264 /* Build a new Quiet NaN (sign=0, frac=all 1's). */
265 struct fpn *fpu_newnan __P((struct fpemu *fe));
266
267 /*
268 * Shift a number right some number of bits, taking care of round/sticky.
269 * Note that the result is probably not a well-formed number (it will lack
270 * the normal 1-bit mant[0]&FP_1).
271 */
272 int fpu_shr __P((struct fpn * fp, int shr));
273 /*
274 * Round a number according to the round mode in FPCR
275 */
276 int fpu_round __P((register struct fpemu *fe, register struct fpn *fp));
277
278 /* type conversion */
279 void fpu_explode __P((struct fpemu *fe, struct fpn *fp, int t, u_int *src));
280 void fpu_implode __P((struct fpemu *fe, struct fpn *fp, int t, u_int *dst));
281
282 /*
283 * non-static emulation functions
284 */
285 /* type 0 */
286 int fpu_emul_fmovecr __P((struct fpemu *fe, struct instruction *insn));
287 int fpu_emul_fstore __P((struct fpemu *fe, struct instruction *insn));
288 int fpu_emul_fscale __P((struct fpemu *fe, struct instruction *insn));
289
290 /*
291 * include function declarations of those which are called by fpu_emul_arith()
292 */
293 #include "fpu_arith_proto.h"
294
295 int fpu_emulate __P((struct frame *frame, struct fpframe *fpf, ksiginfo_t *ksi));
296
297 /*
298 * "helper" functions
299 */
300 /* return values from constant rom */
301 struct fpn *fpu_const __P((struct fpn *fp, u_int offset));
302 /* update exceptions and FPSR */
303 int fpu_upd_excp __P((struct fpemu *fe));
304 u_int fpu_upd_fpsr __P((struct fpemu *fe, struct fpn *fp));
305
306 /* address mode decoder, and load/store */
307 int fpu_decode_ea __P((struct frame *frame, struct instruction *insn,
308 struct insn_ea *ea, int modreg));
309 int fpu_load_ea __P((struct frame *frame, struct instruction *insn,
310 struct insn_ea *ea, char *dst));
311 int fpu_store_ea __P((struct frame *frame, struct instruction *insn,
312 struct insn_ea *ea, char *src));
313
314 /* fpu_subr.c */
315 void fpu_norm __P((register struct fpn *fp));
316
317 #if !defined(FPE_DEBUG)
318 # define FPE_DEBUG 0
319 #endif
320
321 #endif /* _FPU_EMULATE_H_ */
322