1 | /* Copyright (C) 1997, 1998, 2000 Free Software Foundation, Inc.
|
---|
2 | This file is part of the GNU C Library.
|
---|
3 | Contributed by Geoffrey Keating <Geoff.Keating@anu.edu.au>, 1997.
|
---|
4 |
|
---|
5 | The GNU C Library is free software; you can redistribute it and/or
|
---|
6 | modify it under the terms of the GNU Lesser General Public
|
---|
7 | License as published by the Free Software Foundation; either
|
---|
8 | version 2.1 of the License, or (at your option) any later version.
|
---|
9 |
|
---|
10 | The GNU C Library is distributed in the hope that it will be useful,
|
---|
11 | but WITHOUT ANY WARRANTY; without even the implied warranty of
|
---|
12 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
|
---|
13 | Lesser General Public License for more details.
|
---|
14 |
|
---|
15 | You should have received a copy of the GNU Lesser General Public
|
---|
16 | License along with the GNU C Library; if not, write to the Free
|
---|
17 | Software Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA
|
---|
18 | 02111-1307 USA. */
|
---|
19 |
|
---|
20 | #include <stdio.h>
|
---|
21 | #include <math.h>
|
---|
22 | #include <gmp.h>
|
---|
23 | #include <string.h>
|
---|
24 | #include <limits.h>
|
---|
25 | #include <assert.h>
|
---|
26 |
|
---|
27 | #define PRINT_ERRORS 0
|
---|
28 |
|
---|
29 | #define N 0
|
---|
30 | #define N2 20
|
---|
31 | #define FRAC (32 * 4)
|
---|
32 |
|
---|
33 | #define mpbpl (CHAR_BIT * sizeof (mp_limb_t))
|
---|
34 | #define SZ (FRAC / mpbpl + 1)
|
---|
35 | typedef mp_limb_t mp1[SZ], mp2[SZ * 2];
|
---|
36 |
|
---|
37 | /* These strings have exactly 100 hex digits in them. */
|
---|
38 | static const char sin1[101] =
|
---|
39 | "d76aa47848677020c6e9e909c50f3c3289e511132f518b4def"
|
---|
40 | "b6ca5fd6c649bdfb0bd9ff1edcd4577655b5826a3d3b50c264";
|
---|
41 | static const char cos1[101] =
|
---|
42 | "8a51407da8345c91c2466d976871bd29a2373a894f96c3b7f2"
|
---|
43 | "300240b760e6fa96a94430a52d0e9e43f3450e3b8ff99bc934";
|
---|
44 | static const char hexdig[] = "0123456789abcdef";
|
---|
45 |
|
---|
46 | static void
|
---|
47 | print_mpn_hex (const mp_limb_t *x, unsigned size)
|
---|
48 | {
|
---|
49 | char value[size + 1];
|
---|
50 | unsigned i;
|
---|
51 | const unsigned final = (size * 4 > SZ * mpbpl) ? SZ * mpbpl / 4 : size;
|
---|
52 |
|
---|
53 | memset (value, '0', size);
|
---|
54 |
|
---|
55 | for (i = 0; i < final ; i++)
|
---|
56 | value[size-1-i] = hexdig[x[i * 4 / mpbpl] >> (i * 4) % mpbpl & 0xf];
|
---|
57 |
|
---|
58 | value[size] = '\0';
|
---|
59 | fputs (value, stdout);
|
---|
60 | }
|
---|
61 |
|
---|
62 | static void
|
---|
63 | sincosx_mpn (mp1 si, mp1 co, mp1 xx, mp1 ix)
|
---|
64 | {
|
---|
65 | int i;
|
---|
66 | mp2 s[4], c[4];
|
---|
67 | mp1 tmp, x;
|
---|
68 | mp_limb_t chk, round;
|
---|
69 |
|
---|
70 | if (ix == NULL)
|
---|
71 | {
|
---|
72 | memset (si, 0, sizeof (mp1));
|
---|
73 | memset (co, 0, sizeof (mp1));
|
---|
74 | co[SZ-1] = 1;
|
---|
75 | memcpy (x, xx, sizeof (mp1));
|
---|
76 | }
|
---|
77 | else
|
---|
78 | mpn_sub_n (x, xx, ix, SZ);
|
---|
79 |
|
---|
80 | for (i = 0; i < 1 << N; i++)
|
---|
81 | {
|
---|
82 | #define add_shift_mulh(d,x,s1,s2,sh,n) \
|
---|
83 | /* d = (n ? -1 : 1) * (s1 + (s2>>sh)) * x / (1>>N); */ \
|
---|
84 | do { \
|
---|
85 | if (s2 != NULL) { \
|
---|
86 | if (sh > 0) { \
|
---|
87 | assert (sh < mpbpl); \
|
---|
88 | mpn_lshift (tmp, s1, SZ, sh); \
|
---|
89 | chk = (n ? mpn_sub_n : mpn_add_n)(tmp,tmp,s2+FRAC/mpbpl,SZ); \
|
---|
90 | } else \
|
---|
91 | chk = (n ? mpn_sub_n : mpn_add_n)(tmp,s1,s2+FRAC/mpbpl,SZ); \
|
---|
92 | /* assert(chk == 0); */ \
|
---|
93 | mpn_mul_n(d,tmp,x,SZ); \
|
---|
94 | } else \
|
---|
95 | mpn_mul_n(d,s1,x,SZ); \
|
---|
96 | /* assert(d[SZ*2-1] == 0); */ \
|
---|
97 | assert(N+sh < mpbpl); \
|
---|
98 | if (N+sh > 0) mpn_rshift(d,d,2*SZ,N+sh); \
|
---|
99 | } while(0)
|
---|
100 | #define summ(d,ss,s,n) \
|
---|
101 | /* d = ss +/- (s[0]+2*s[1]+2*s[2]+s[3])/6; */ \
|
---|
102 | do { \
|
---|
103 | chk = mpn_add_n(tmp,s[1]+FRAC/mpbpl,s[2]+FRAC/mpbpl,SZ); \
|
---|
104 | mpn_lshift(tmp,tmp,SZ,1); \
|
---|
105 | chk |= mpn_add_n(tmp,tmp,s[0]+FRAC/mpbpl,SZ); \
|
---|
106 | chk |= mpn_add_n(tmp,tmp,s[3]+FRAC/mpbpl,SZ); \
|
---|
107 | round = mpn_divmod_1(tmp,tmp,SZ,6); \
|
---|
108 | /* chk |= mpn_add_1(tmp,tmp,SZ, (round > 3) ); */ \
|
---|
109 | chk |= (n ? mpn_sub_n : mpn_add_n)(d,ss,tmp,SZ); \
|
---|
110 | /* assert(chk == 0); */ \
|
---|
111 | } while (0)
|
---|
112 |
|
---|
113 | add_shift_mulh (s[0], x, co, NULL, 0, 0); /* s0 = h * c; */
|
---|
114 | add_shift_mulh (c[0], x, si, NULL, 0, 0); /* c0 = h * s; */
|
---|
115 | add_shift_mulh (s[1], x, co, c[0], 1, 1); /* s1 = h * (c - c0/2); */
|
---|
116 | add_shift_mulh (c[1], x, si, s[0], 1, 0); /* c1 = h * (s + s0/2); */
|
---|
117 | add_shift_mulh (s[2], x, co, c[1], 1, 1); /* s2 = h * (c - c1/2); */
|
---|
118 | add_shift_mulh (c[2], x, si, s[1], 1, 0); /* c2 = h * (s + s1/2); */
|
---|
119 | add_shift_mulh (s[3], x, co, c[2], 0, 1); /* s3 = h * (c - c2); */
|
---|
120 | add_shift_mulh (c[3], x, si, s[2], 0, 0); /* c3 = h * (s + s2); */
|
---|
121 | summ (si, si, s, 0); /* s = s + (s0+2*s1+2*s2+s3)/6; */
|
---|
122 | summ (co, co, c, 1); /* c = c - (c0+2*c1+2*c2+c3)/6; */
|
---|
123 | }
|
---|
124 | #undef add_shift_mulh
|
---|
125 | #undef summ
|
---|
126 | }
|
---|
127 |
|
---|
128 | static int
|
---|
129 | mpn_bitsize (const mp_limb_t *SRC_PTR, mp_size_t SIZE)
|
---|
130 | {
|
---|
131 | int i, j;
|
---|
132 | for (i = SIZE - 1; i > 0; i--)
|
---|
133 | if (SRC_PTR[i] != 0)
|
---|
134 | break;
|
---|
135 | for (j = mpbpl - 1; j >= 0; j--)
|
---|
136 | if ((SRC_PTR[i] & (mp_limb_t)1 << j) != 0)
|
---|
137 | break;
|
---|
138 |
|
---|
139 | return i * mpbpl + j;
|
---|
140 | }
|
---|
141 |
|
---|
142 | int
|
---|
143 | main (void)
|
---|
144 | {
|
---|
145 | mp1 si, co, x, ox, xt, s2, c2, s3, c3;
|
---|
146 | int i;
|
---|
147 | int sin_errors = 0, cos_errors = 0;
|
---|
148 | int sin_failures = 0, cos_failures = 0;
|
---|
149 | mp1 sin_maxerror, cos_maxerror;
|
---|
150 | int sin_maxerror_s = 0, cos_maxerror_s = 0;
|
---|
151 | const double sf = pow (2, mpbpl);
|
---|
152 |
|
---|
153 | /* assert(mpbpl == mp_bits_per_limb); */
|
---|
154 | assert(FRAC / mpbpl * mpbpl == FRAC);
|
---|
155 |
|
---|
156 | memset (sin_maxerror, 0, sizeof (mp1));
|
---|
157 | memset (cos_maxerror, 0, sizeof (mp1));
|
---|
158 | memset (xt, 0, sizeof (mp1));
|
---|
159 | xt[(FRAC - N2) / mpbpl] = (mp_limb_t)1 << (FRAC - N2) % mpbpl;
|
---|
160 |
|
---|
161 | for (i = 0; i < 1 << N2; i++)
|
---|
162 | {
|
---|
163 | int s2s, s3s, c2s, c3s, j;
|
---|
164 | double ds2,dc2;
|
---|
165 |
|
---|
166 | mpn_mul_1 (x, xt, SZ, i);
|
---|
167 | sincosx_mpn (si, co, x, i == 0 ? NULL : ox);
|
---|
168 | memcpy (ox, x, sizeof (mp1));
|
---|
169 | ds2 = sin (i / (double) (1 << N2));
|
---|
170 | dc2 = cos (i / (double) (1 << N2));
|
---|
171 | for (j = SZ-1; j >= 0; j--)
|
---|
172 | {
|
---|
173 | s2[j] = (mp_limb_t) ds2;
|
---|
174 | ds2 = (ds2 - s2[j]) * sf;
|
---|
175 | c2[j] = (mp_limb_t) dc2;
|
---|
176 | dc2 = (dc2 - c2[j]) * sf;
|
---|
177 | }
|
---|
178 | if (mpn_cmp (si, s2, SZ) >= 0)
|
---|
179 | mpn_sub_n (s3, si, s2, SZ);
|
---|
180 | else
|
---|
181 | mpn_sub_n (s3, s2, si, SZ);
|
---|
182 | if (mpn_cmp (co, c2, SZ) >= 0)
|
---|
183 | mpn_sub_n (c3, co, c2, SZ);
|
---|
184 | else
|
---|
185 | mpn_sub_n (c3, c2, co, SZ);
|
---|
186 |
|
---|
187 | s2s = mpn_bitsize (s2, SZ);
|
---|
188 | s3s = mpn_bitsize (s3, SZ);
|
---|
189 | c2s = mpn_bitsize (c2, SZ);
|
---|
190 | c3s = mpn_bitsize (c3, SZ);
|
---|
191 | if ((s3s >= 0 && s2s - s3s < 54)
|
---|
192 | || (c3s >= 0 && c2s - c3s < 54)
|
---|
193 | || 0)
|
---|
194 | {
|
---|
195 | #if PRINT_ERRORS
|
---|
196 | printf ("%06x ", i * (0x100000 / (1 << N2)));
|
---|
197 | print_mpn_hex(si, (FRAC / 4) + 1);
|
---|
198 | putchar (' ');
|
---|
199 | print_mpn_hex (co, (FRAC / 4) + 1);
|
---|
200 | putchar ('\n');
|
---|
201 | fputs (" ", stdout);
|
---|
202 | print_mpn_hex (s2, (FRAC / 4) + 1);
|
---|
203 | putchar (' ');
|
---|
204 | print_mpn_hex (c2, (FRAC / 4) + 1);
|
---|
205 | putchar ('\n');
|
---|
206 | printf (" %c%c ",
|
---|
207 | s3s >= 0 && s2s-s3s < 54 ? s2s - s3s == 53 ? 'e' : 'F' : 'P',
|
---|
208 | c3s >= 0 && c2s-c3s < 54 ? c2s - c3s == 53 ? 'e' : 'F' : 'P');
|
---|
209 | print_mpn_hex (s3, (FRAC / 4) + 1);
|
---|
210 | putchar (' ');
|
---|
211 | print_mpn_hex (c3, (FRAC / 4) + 1);
|
---|
212 | putchar ('\n');
|
---|
213 | #endif
|
---|
214 | sin_errors += s2s - s3s == 53;
|
---|
215 | cos_errors += c2s - c3s == 53;
|
---|
216 | sin_failures += s2s - s3s < 53;
|
---|
217 | cos_failures += c2s - c3s < 53;
|
---|
218 | }
|
---|
219 | if (s3s >= sin_maxerror_s
|
---|
220 | && mpn_cmp (s3, sin_maxerror, SZ) > 0)
|
---|
221 | {
|
---|
222 | memcpy (sin_maxerror, s3, sizeof (mp1));
|
---|
223 | sin_maxerror_s = s3s;
|
---|
224 | }
|
---|
225 | if (c3s >= cos_maxerror_s
|
---|
226 | && mpn_cmp (c3, cos_maxerror, SZ) > 0)
|
---|
227 | {
|
---|
228 | memcpy (cos_maxerror, c3, sizeof (mp1));
|
---|
229 | cos_maxerror_s = c3s;
|
---|
230 | }
|
---|
231 | }
|
---|
232 |
|
---|
233 | /* Check Range-Kutta against precomputed values of sin(1) and cos(1). */
|
---|
234 | memset (x, 0, sizeof (mp1));
|
---|
235 | x[FRAC / mpbpl] = (mp_limb_t)1 << FRAC % mpbpl;
|
---|
236 | sincosx_mpn (si, co, x, ox);
|
---|
237 |
|
---|
238 | memset (s2, 0, sizeof (mp1));
|
---|
239 | memset (c2, 0, sizeof (mp1));
|
---|
240 | for (i = 0; i < 100 && i < FRAC / 4; i++)
|
---|
241 | {
|
---|
242 | s2[(FRAC - i * 4 - 4) / mpbpl] |= ((strchr (hexdig, sin1[i]) - hexdig)
|
---|
243 | << (FRAC - i * 4 - 4) % mpbpl);
|
---|
244 | c2[(FRAC - i * 4 - 4) / mpbpl] |= ((strchr (hexdig, cos1[i]) - hexdig)
|
---|
245 | << (FRAC - i * 4 - 4) % mpbpl);
|
---|
246 | }
|
---|
247 |
|
---|
248 | if (mpn_cmp (si, s2, SZ) >= 0)
|
---|
249 | mpn_sub_n (s3, si, s2, SZ);
|
---|
250 | else
|
---|
251 | mpn_sub_n (s3, s2, si, SZ);
|
---|
252 | if (mpn_cmp (co, c2, SZ) >= 0)
|
---|
253 | mpn_sub_n (c3, co, c2, SZ);
|
---|
254 | else
|
---|
255 | mpn_sub_n (c3, c2, co, SZ);
|
---|
256 |
|
---|
257 | printf ("sin:\n");
|
---|
258 | printf ("%d failures; %d errors; error rate %0.2f%%\n",
|
---|
259 | sin_failures, sin_errors, sin_errors * 100.0 / (double) (1 << N2));
|
---|
260 | fputs ("maximum error: ", stdout);
|
---|
261 | print_mpn_hex (sin_maxerror, (FRAC / 4) + 1);
|
---|
262 | fputs ("\nerror in sin(1): ", stdout);
|
---|
263 | print_mpn_hex (s3, (FRAC / 4) + 1);
|
---|
264 |
|
---|
265 | fputs ("\n\ncos:\n", stdout);
|
---|
266 | printf ("%d failures; %d errors; error rate %0.2f%%\n",
|
---|
267 | cos_failures, cos_errors, cos_errors * 100.0 / (double) (1 << N2));
|
---|
268 | fputs ("maximum error: ", stdout);
|
---|
269 | print_mpn_hex (cos_maxerror, (FRAC / 4) + 1);
|
---|
270 | fputs ("\nerror in cos(1): ", stdout);
|
---|
271 | print_mpn_hex (c3, (FRAC / 4) + 1);
|
---|
272 | putchar ('\n');
|
---|
273 |
|
---|
274 | return (sin_failures == 0 && cos_failures == 0) ? 0 : 1;
|
---|
275 | }
|
---|