1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
|
/* Simulator Floating-point support.
Copyright (C) 1997 Free Software Foundation, Inc.
Contributed by Cygnus Support.
This file is part of GDB, the GNU debugger.
This program is free software; you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
the Free Software Foundation; either version 2, or (at your option)
any later version.
This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
You should have received a copy of the GNU General Public License along
with this program; if not, write to the Free Software Foundation, Inc.,
59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. */
#ifndef SIM_FPU_H
#define SIM_FPU_H
/* The FPU intermediate type - this object, passed by reference,
should be treated as opaque.
Pragmatics - pass struct by ref:
The alternatives for this object/interface that were considered
were: a packed 64 bit value; an unpacked structure passed by value;
and an unpacked structure passed by reference.
The packed 64 bit value was rejected because: it limited the
precision of intermediate values; reasonable performance would only
be achieved when the sim_fpu package was in-lined allowing repeated
unpacking operations to be eliminated.
For unpacked structures (passed by value and reference), the code
quality of GCC-2.7 (on x86) for each alternative was compared.
Needless to say the results, while better then for a packed 64 bit
object, were still poor (GCC had only limited support for the
optimization of references to structure members). Regardless, the
struct-by-ref alternative achieved better results when compiled
with (better speed) and without (better code density) in-lining.
Here's looking forward to an improved GCC optimizer.
Pragmatics - avoid host FP hardware:
FP operations can be implemented by either: the host's floating
point hardware; or by emulating the FP operations using integer
only routines. This is direct tradeoff between speed, portability
and correctness.
The two principal reasons for selecting portability and correctness
over speed are:
1 - Correctness. The assumption that FP correctness wasn't an
issue for code being run on simulators was wrong. Instead of
running FP tolerant (?) code, simulator users instead typically run
very aggressive FP code sequences. The sole purpose of those
sequences being to test the target ISA's FP implementation.
2 - Portability. The host FP implementation is not predictable. A
simulator modeling aggressive FP code sequences using the hosts FPU
relies heavily on the correctness of the hosts FP implementation.
It turns out that such trust can be misplaced. The behavior of
host FP implementations when handling edge conditions such as SNaNs
and exceptions varied widely.
*/
typedef enum
{
sim_fpu_class_zero,
sim_fpu_class_snan,
sim_fpu_class_qnan,
sim_fpu_class_number,
sim_fpu_class_denorm,
sim_fpu_class_infinity,
} sim_fpu_class;
typedef struct _sim_fpu {
sim_fpu_class class;
int normal_exp;
int result;
int sign;
unsigned64 fraction;
} sim_fpu;
/* Rounding options.
The value zero (sim_fpu_round_default) for ALU operations indicates
that, when possible, rounding should be avoided. */
typedef enum
{
sim_fpu_round_default = 0,
sim_fpu_round_near = 1,
sim_fpu_round_zero = 2,
sim_fpu_round_up = 3,
sim_fpu_round_down = 4,
} sim_fpu_round;
/* Options when handling denormalized numbers. */
typedef enum
{
sim_fpu_denorm_underflow_inexact = 1,
sim_fpu_denorm_zero = 2,
} sim_fpu_denorm;
/* Status values returned by FPU operators.
When checking the result of an FP sequence (ex 32to, add, single,
to32) the caller may either: check the return value of each FP
operator; or form the union (OR) of the returned values and examine
them once at the end.
FIXME: This facility is still being developed. The choice of
status values returned and their exact meaning may changed in the
future. */
typedef enum
{
sim_fpu_status_invalid_snan = 1,
sim_fpu_status_invalid_qnan = 2,
sim_fpu_status_invalid_isi = 4, /* (inf - inf) */
sim_fpu_status_invalid_idi = 8, /* (inf / inf) */
sim_fpu_status_invalid_zdz = 16, /* (0 / 0) */
sim_fpu_status_invalid_imz = 32, /* (inf * 0) */
sim_fpu_status_invalid_cvi = 64, /* convert to integer */
sim_fpu_status_invalid_div0 = 128, /* (X / 0) */
sim_fpu_status_invalid_cmp = 256, /* compare */
sim_fpu_status_invalid_sqrt = 512,
sim_fpu_status_rounded = 1024,
sim_fpu_status_inexact = 2048,
sim_fpu_status_overflow = 4096,
sim_fpu_status_underflow = 8192,
sim_fpu_status_denorm = 16384,
} sim_fpu_status;
/* Directly map between a 32/64 bit register and the sim_fpu internal
type.
When converting from the 32/64 bit packed format to the sim_fpu
internal type, the operation is exact.
When converting from the sim_fpu internal type to 32/64 bit packed
format, the operation may result in a loss of precision. The
configuration macro WITH_FPU_CONVERSION controls this. By default,
silent round to nearest is performed. Alternativly, round up,
round down and round to zero can be performed. In a simulator
emulating exact FPU behavour, sim_fpu_round_{32,64} should be
called before packing the sim_fpu value. */
INLINE_SIM_FPU (void) sim_fpu_32to (sim_fpu *f, unsigned32 s);
INLINE_SIM_FPU (void) sim_fpu_232to (sim_fpu *f, unsigned32 h, unsigned32 l);
INLINE_SIM_FPU (void) sim_fpu_64to (sim_fpu *f, unsigned64 d);
INLINE_SIM_FPU (void) sim_fpu_to32 (unsigned32 *s, const sim_fpu *f);
INLINE_SIM_FPU (void) sim_fpu_to232 (unsigned32 *h, unsigned32 *l, const sim_fpu *f);
INLINE_SIM_FPU (void) sim_fpu_to64 (unsigned64 *d, const sim_fpu *f);
#if WITH_TARGET_FLOATING_POINT_BITSIZE == 32
#define sim_fpu_tofp sim_fpu_to32
#define sim_fpu_fpto sim_fpu_32to
#define sim_fpu_round_fp sim_fpu_round_32
#endif
#if WITH_TARGET_FLOATING_POINT_BITSIZE == 64
#define sim_fpu_tofp sim_fpu_to64
#define sim_fpu_fpto sim_fpu_64to
#define sim_fpu_round_fp sim_fpu_round_64
#endif
/* Rounding operators.
Force an intermediate result to an exact 32/64 bit
representation. */
INLINE_SIM_FPU (int) sim_fpu_round_32 (sim_fpu *f,
sim_fpu_round round,
sim_fpu_denorm denorm);
INLINE_SIM_FPU (int) sim_fpu_round_64 (sim_fpu *f,
sim_fpu_round round,
sim_fpu_denorm denorm);
/* Arrithmetic operators.
FIXME: In the future, additional arguments ROUNDING and BITSIZE may
be added. */
INLINE_SIM_FPU (int) sim_fpu_add (sim_fpu *f,
const sim_fpu *l, const sim_fpu *r);
INLINE_SIM_FPU (int) sim_fpu_sub (sim_fpu *f,
const sim_fpu *l, const sim_fpu *r);
INLINE_SIM_FPU (int) sim_fpu_mul (sim_fpu *f,
const sim_fpu *l, const sim_fpu *r);
INLINE_SIM_FPU (int) sim_fpu_div (sim_fpu *f,
const sim_fpu *l, const sim_fpu *r);
INLINE_SIM_FPU (int) sim_fpu_neg (sim_fpu *f,
const sim_fpu *a);
INLINE_SIM_FPU (int) sim_fpu_abs (sim_fpu *f,
const sim_fpu *a);
INLINE_SIM_FPU (int) sim_fpu_inv (sim_fpu *f,
const sim_fpu *a);
INLINE_SIM_FPU (int) sim_fpu_sqrt (sim_fpu *f,
const sim_fpu *sqr);
/* Conversion of integer <-> floating point. */
INLINE_SIM_FPU (int) sim_fpu_i32to (sim_fpu *f, signed32 i,
sim_fpu_round round);
INLINE_SIM_FPU (int) sim_fpu_u32to (sim_fpu *f, unsigned32 u,
sim_fpu_round round);
INLINE_SIM_FPU (int) sim_fpu_i64to (sim_fpu *f, signed64 i,
sim_fpu_round round);
INLINE_SIM_FPU (int) sim_fpu_u64to (sim_fpu *f, unsigned64 u,
sim_fpu_round round);
INLINE_SIM_FPU (int) sim_fpu_i232to (sim_fpu *f, signed32 h, signed32 l,
sim_fpu_round round);
INLINE_SIM_FPU (int) sim_fpu_u232to (sim_fpu *f, unsigned32 h, unsigned32 l,
sim_fpu_round round);
INLINE_SIM_FPU (int) sim_fpu_to32i (signed32 *i, const sim_fpu *f,
sim_fpu_round round);
INLINE_SIM_FPU (int) sim_fpu_to32u (unsigned32 *u, const sim_fpu *f,
sim_fpu_round round);
INLINE_SIM_FPU (int) sim_fpu_to64i (signed64 *i, const sim_fpu *f,
sim_fpu_round round);
INLINE_SIM_FPU (int) sim_fpu_to64u (unsigned64 *u, const sim_fpu *f,
sim_fpu_round round);
INLINE_SIM_FPU (int) sim_fpu_to232i (signed64 *h, signed64 *l, const sim_fpu *f,
sim_fpu_round round);
INLINE_SIM_FPU (int) sim_fpu_to232u (unsigned64 *h, unsigned64 *l, const sim_fpu *f,
sim_fpu_round round);
/* Conversion of internal sim_fpu type to host double format.
For debuging/tracing only. A SNaN is never returned. */
/* INLINE_SIM_FPU (float) sim_fpu_2f (const sim_fpu *f); */
INLINE_SIM_FPU (double) sim_fpu_2d (const sim_fpu *d);
/* INLINE_SIM_FPU (void) sim_fpu_f2 (sim_fpu *f, float s); */
INLINE_SIM_FPU (void) sim_fpu_d2 (sim_fpu *f, double d);
/* Specific number classes.
NB: When either a 32/64 bit floating points is converted to
internal format, or an internal format number is rounded to 32/64
bit precision, a special marker is retained that indicates that the
value was normalized. For such numbers both is_number and
is_denorm will return true. */
INLINE_SIM_FPU (int) sim_fpu_is_nan (const sim_fpu *s); /* 1 => SNaN or QNaN */
INLINE_SIM_FPU (int) sim_fpu_is_snan (const sim_fpu *s); /* 1 => SNaN */
INLINE_SIM_FPU (int) sim_fpu_is_qnan (const sim_fpu *s); /* 1 => QNaN */
INLINE_SIM_FPU (int) sim_fpu_is_zero (const sim_fpu *s);
INLINE_SIM_FPU (int) sim_fpu_is_infinity (const sim_fpu *s);
INLINE_SIM_FPU (int) sim_fpu_is_number (const sim_fpu *s); /* !zero */
INLINE_SIM_FPU (int) sim_fpu_is_denorm (const sim_fpu *s); /* !zero */
/* Specific comparison operators
The comparison operators set *IS to zero and return a nonzero
result for NaNs et.al. */
INLINE_SIM_FPU (int) sim_fpu_lt (int *is, const sim_fpu *l, const sim_fpu *r);
INLINE_SIM_FPU (int) sim_fpu_le (int *is, const sim_fpu *l, const sim_fpu *r);
INLINE_SIM_FPU (int) sim_fpu_eq (int *is, const sim_fpu *l, const sim_fpu *r);
INLINE_SIM_FPU (int) sim_fpu_ne (int *is, const sim_fpu *l, const sim_fpu *r);
INLINE_SIM_FPU (int) sim_fpu_ge (int *is, const sim_fpu *l, const sim_fpu *r);
INLINE_SIM_FPU (int) sim_fpu_gt (int *is, const sim_fpu *l, const sim_fpu *r);
INLINE_SIM_FPU (int) sim_fpu_is_lt (const sim_fpu *l, const sim_fpu *r);
INLINE_SIM_FPU (int) sim_fpu_is_le (const sim_fpu *l, const sim_fpu *r);
INLINE_SIM_FPU (int) sim_fpu_is_eq (const sim_fpu *l, const sim_fpu *r);
INLINE_SIM_FPU (int) sim_fpu_is_ne (const sim_fpu *l, const sim_fpu *r);
INLINE_SIM_FPU (int) sim_fpu_is_ge (const sim_fpu *l, const sim_fpu *r);
INLINE_SIM_FPU (int) sim_fpu_is_gt (const sim_fpu *l, const sim_fpu *r);
/* General number class and comparison operators.
The result of the comparison is indicated by returning one of the
values below. Efficient emulation of a target FP compare
instruction can be achieved by redefining the values below to match
corresponding target FP status bits.
For instance. SIM_FPU_QNAN may be redefined to be the bit
`INVALID' while SIM_FPU_NINF might be redefined as the bits
`NEGATIVE | INFINITY | VALID'. */
#ifndef SIM_FPU_IS_SNAN
enum {
SIM_FPU_IS_SNAN = 1, /* Noisy not-a-number */
SIM_FPU_IS_QNAN = 2, /* Quite not-a-number */
SIM_FPU_IS_NINF = 3, /* -infinity */
SIM_FPU_IS_PINF = 4, /* +infinity */
SIM_FPU_IS_NNUMBER = 5, /* -number - [ -MAX .. -MIN ] */
SIM_FPU_IS_PNUMBER = 6, /* +number - [ +MIN .. +MAX ] */
SIM_FPU_IS_NDENORM = 7, /* -denorm - ( MIN .. 0 ) */
SIM_FPU_IS_PDENORM = 8, /* +denorm - ( 0 .. MIN ) */
SIM_FPU_IS_NZERO = 9, /* -0 */
SIM_FPU_IS_PZERO = 10, /* +0 */
};
#endif
INLINE_SIM_FPU (int) sim_fpu_is (const sim_fpu *l);
INLINE_SIM_FPU (int) sim_fpu_cmp (const sim_fpu *l, const sim_fpu *r);
/* A constant of useful numbers */
extern const sim_fpu sim_fpu_zero;
extern const sim_fpu sim_fpu_one;
extern const sim_fpu sim_fpu_two;
extern const sim_fpu sim_fpu_qnan;
/* For debugging */
typedef void sim_fpu_print_func (void *, char *, ...);
INLINE_SIM_FPU (void) sim_fpu_print_fpu (const sim_fpu *f,
sim_fpu_print_func *print,
void *arg);
INLINE_SIM_FPU (void) sim_fpu_print_status (int status,
sim_fpu_print_func *print,
void *arg);
#endif
|