xf.li | bdd93d5 | 2023-05-12 07:10:14 -0700 | [diff] [blame^] | 1 | /* Copyright (C) 1997-2016 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, see |
| 17 | <http://www.gnu.org/licenses/>. */ |
| 18 | |
| 19 | #include <stdio.h> |
| 20 | #include <math.h> |
| 21 | #include <gmp.h> |
| 22 | #include <string.h> |
| 23 | #include <limits.h> |
| 24 | #include <assert.h> |
| 25 | |
| 26 | #define PRINT_ERRORS 0 |
| 27 | |
| 28 | #define N 0 |
| 29 | #define N2 20 |
| 30 | #define FRAC (32 * 4) |
| 31 | |
| 32 | #define mpbpl (CHAR_BIT * sizeof (mp_limb_t)) |
| 33 | #define SZ (FRAC / mpbpl + 1) |
| 34 | typedef mp_limb_t mp1[SZ], mp2[SZ * 2]; |
| 35 | |
| 36 | /* These strings have exactly 100 hex digits in them. */ |
| 37 | static const char sin1[101] = |
| 38 | "d76aa47848677020c6e9e909c50f3c3289e511132f518b4def" |
| 39 | "b6ca5fd6c649bdfb0bd9ff1edcd4577655b5826a3d3b50c264"; |
| 40 | static const char cos1[101] = |
| 41 | "8a51407da8345c91c2466d976871bd29a2373a894f96c3b7f2" |
| 42 | "300240b760e6fa96a94430a52d0e9e43f3450e3b8ff99bc934"; |
| 43 | static const char hexdig[] = "0123456789abcdef"; |
| 44 | |
| 45 | static void |
| 46 | print_mpn_hex (const mp_limb_t *x, unsigned size) |
| 47 | { |
| 48 | char value[size + 1]; |
| 49 | unsigned i; |
| 50 | const unsigned final = (size * 4 > SZ * mpbpl) ? SZ * mpbpl / 4 : size; |
| 51 | |
| 52 | memset (value, '0', size); |
| 53 | |
| 54 | for (i = 0; i < final ; i++) |
| 55 | value[size-1-i] = hexdig[x[i * 4 / mpbpl] >> (i * 4) % mpbpl & 0xf]; |
| 56 | |
| 57 | value[size] = '\0'; |
| 58 | fputs (value, stdout); |
| 59 | } |
| 60 | |
| 61 | static void |
| 62 | sincosx_mpn (mp1 si, mp1 co, mp1 xx, mp1 ix) |
| 63 | { |
| 64 | int i; |
| 65 | mp2 s[4], c[4]; |
| 66 | mp1 tmp, x; |
| 67 | |
| 68 | if (ix == NULL) |
| 69 | { |
| 70 | memset (si, 0, sizeof (mp1)); |
| 71 | memset (co, 0, sizeof (mp1)); |
| 72 | co[SZ-1] = 1; |
| 73 | memcpy (x, xx, sizeof (mp1)); |
| 74 | } |
| 75 | else |
| 76 | mpn_sub_n (x, xx, ix, SZ); |
| 77 | |
| 78 | for (i = 0; i < 1 << N; i++) |
| 79 | { |
| 80 | #define add_shift_mulh(d,x,s1,s2,sh,n) \ |
| 81 | do { \ |
| 82 | if (s2 != NULL) { \ |
| 83 | if (sh > 0) { \ |
| 84 | assert (sh < mpbpl); \ |
| 85 | mpn_lshift (tmp, s1, SZ, sh); \ |
| 86 | if (n) \ |
| 87 | mpn_sub_n (tmp,tmp,s2+FRAC/mpbpl,SZ); \ |
| 88 | else \ |
| 89 | mpn_add_n (tmp,tmp,s2+FRAC/mpbpl,SZ); \ |
| 90 | } else { \ |
| 91 | if (n) \ |
| 92 | mpn_sub_n (tmp,s1,s2+FRAC/mpbpl,SZ); \ |
| 93 | else \ |
| 94 | mpn_add_n (tmp,s1,s2+FRAC/mpbpl,SZ); \ |
| 95 | } \ |
| 96 | mpn_mul_n(d,tmp,x,SZ); \ |
| 97 | } else \ |
| 98 | mpn_mul_n(d,s1,x,SZ); \ |
| 99 | assert(N+sh < mpbpl); \ |
| 100 | if (N+sh > 0) mpn_rshift(d,d,2*SZ,N+sh); \ |
| 101 | } while(0) |
| 102 | #define summ(d,ss,s,n) \ |
| 103 | do { \ |
| 104 | mpn_add_n(tmp,s[1]+FRAC/mpbpl,s[2]+FRAC/mpbpl,SZ); \ |
| 105 | mpn_lshift(tmp,tmp,SZ,1); \ |
| 106 | mpn_add_n(tmp,tmp,s[0]+FRAC/mpbpl,SZ); \ |
| 107 | mpn_add_n(tmp,tmp,s[3]+FRAC/mpbpl,SZ); \ |
| 108 | mpn_divmod_1(tmp,tmp,SZ,6); \ |
| 109 | if (n) \ |
| 110 | mpn_sub_n (d,ss,tmp,SZ); \ |
| 111 | else \ |
| 112 | mpn_add_n (d,ss,tmp,SZ); \ |
| 113 | } while (0) |
| 114 | |
| 115 | add_shift_mulh (s[0], x, co, NULL, 0, 0); /* s0 = h * c; */ |
| 116 | add_shift_mulh (c[0], x, si, NULL, 0, 0); /* c0 = h * s; */ |
| 117 | add_shift_mulh (s[1], x, co, c[0], 1, 1); /* s1 = h * (c - c0/2); */ |
| 118 | add_shift_mulh (c[1], x, si, s[0], 1, 0); /* c1 = h * (s + s0/2); */ |
| 119 | add_shift_mulh (s[2], x, co, c[1], 1, 1); /* s2 = h * (c - c1/2); */ |
| 120 | add_shift_mulh (c[2], x, si, s[1], 1, 0); /* c2 = h * (s + s1/2); */ |
| 121 | add_shift_mulh (s[3], x, co, c[2], 0, 1); /* s3 = h * (c - c2); */ |
| 122 | add_shift_mulh (c[3], x, si, s[2], 0, 0); /* c3 = h * (s + s2); */ |
| 123 | summ (si, si, s, 0); /* s = s + (s0+2*s1+2*s2+s3)/6; */ |
| 124 | summ (co, co, c, 1); /* c = c - (c0+2*c1+2*c2+c3)/6; */ |
| 125 | } |
| 126 | #undef add_shift_mulh |
| 127 | #undef summ |
| 128 | } |
| 129 | |
| 130 | static int |
| 131 | mpn_bitsize (const mp_limb_t *SRC_PTR, mp_size_t SIZE) |
| 132 | { |
| 133 | int i, j; |
| 134 | for (i = SIZE - 1; i > 0; i--) |
| 135 | if (SRC_PTR[i] != 0) |
| 136 | break; |
| 137 | for (j = mpbpl - 1; j >= 0; j--) |
| 138 | if ((SRC_PTR[i] & (mp_limb_t)1 << j) != 0) |
| 139 | break; |
| 140 | |
| 141 | return i * mpbpl + j; |
| 142 | } |
| 143 | |
| 144 | static int |
| 145 | do_test (void) |
| 146 | { |
| 147 | mp1 si, co, x, ox, xt, s2, c2, s3, c3; |
| 148 | int i; |
| 149 | int sin_errors = 0, cos_errors = 0; |
| 150 | int sin_failures = 0, cos_failures = 0; |
| 151 | mp1 sin_maxerror, cos_maxerror; |
| 152 | int sin_maxerror_s = 0, cos_maxerror_s = 0; |
| 153 | const double sf = pow (2, mpbpl); |
| 154 | |
| 155 | /* assert(mpbpl == mp_bits_per_limb); */ |
| 156 | assert(FRAC / mpbpl * mpbpl == FRAC); |
| 157 | |
| 158 | memset (sin_maxerror, 0, sizeof (mp1)); |
| 159 | memset (cos_maxerror, 0, sizeof (mp1)); |
| 160 | memset (xt, 0, sizeof (mp1)); |
| 161 | xt[(FRAC - N2) / mpbpl] = (mp_limb_t)1 << (FRAC - N2) % mpbpl; |
| 162 | |
| 163 | for (i = 0; i < 1 << N2; i++) |
| 164 | { |
| 165 | int s2s, s3s, c2s, c3s, j; |
| 166 | double ds2,dc2; |
| 167 | |
| 168 | mpn_mul_1 (x, xt, SZ, i); |
| 169 | sincosx_mpn (si, co, x, i == 0 ? NULL : ox); |
| 170 | memcpy (ox, x, sizeof (mp1)); |
| 171 | ds2 = sin (i / (double) (1 << N2)); |
| 172 | dc2 = cos (i / (double) (1 << N2)); |
| 173 | for (j = SZ-1; j >= 0; j--) |
| 174 | { |
| 175 | s2[j] = (mp_limb_t) ds2; |
| 176 | ds2 = (ds2 - s2[j]) * sf; |
| 177 | c2[j] = (mp_limb_t) dc2; |
| 178 | dc2 = (dc2 - c2[j]) * sf; |
| 179 | } |
| 180 | if (mpn_cmp (si, s2, SZ) >= 0) |
| 181 | mpn_sub_n (s3, si, s2, SZ); |
| 182 | else |
| 183 | mpn_sub_n (s3, s2, si, SZ); |
| 184 | if (mpn_cmp (co, c2, SZ) >= 0) |
| 185 | mpn_sub_n (c3, co, c2, SZ); |
| 186 | else |
| 187 | mpn_sub_n (c3, c2, co, SZ); |
| 188 | |
| 189 | s2s = mpn_bitsize (s2, SZ); |
| 190 | s3s = mpn_bitsize (s3, SZ); |
| 191 | c2s = mpn_bitsize (c2, SZ); |
| 192 | c3s = mpn_bitsize (c3, SZ); |
| 193 | if ((s3s >= 0 && s2s - s3s < 54) |
| 194 | || (c3s >= 0 && c2s - c3s < 54) |
| 195 | || 0) |
| 196 | { |
| 197 | #if PRINT_ERRORS |
| 198 | printf ("%06x ", i * (0x100000 / (1 << N2))); |
| 199 | print_mpn_hex(si, (FRAC / 4) + 1); |
| 200 | putchar (' '); |
| 201 | print_mpn_hex (co, (FRAC / 4) + 1); |
| 202 | putchar ('\n'); |
| 203 | fputs (" ", stdout); |
| 204 | print_mpn_hex (s2, (FRAC / 4) + 1); |
| 205 | putchar (' '); |
| 206 | print_mpn_hex (c2, (FRAC / 4) + 1); |
| 207 | putchar ('\n'); |
| 208 | printf (" %c%c ", |
| 209 | s3s >= 0 && s2s-s3s < 54 ? s2s - s3s == 53 ? 'e' : 'F' : 'P', |
| 210 | c3s >= 0 && c2s-c3s < 54 ? c2s - c3s == 53 ? 'e' : 'F' : 'P'); |
| 211 | print_mpn_hex (s3, (FRAC / 4) + 1); |
| 212 | putchar (' '); |
| 213 | print_mpn_hex (c3, (FRAC / 4) + 1); |
| 214 | putchar ('\n'); |
| 215 | #endif |
| 216 | sin_errors += s2s - s3s == 53; |
| 217 | cos_errors += c2s - c3s == 53; |
| 218 | sin_failures += s2s - s3s < 53; |
| 219 | cos_failures += c2s - c3s < 53; |
| 220 | } |
| 221 | if (s3s >= sin_maxerror_s |
| 222 | && mpn_cmp (s3, sin_maxerror, SZ) > 0) |
| 223 | { |
| 224 | memcpy (sin_maxerror, s3, sizeof (mp1)); |
| 225 | sin_maxerror_s = s3s; |
| 226 | } |
| 227 | if (c3s >= cos_maxerror_s |
| 228 | && mpn_cmp (c3, cos_maxerror, SZ) > 0) |
| 229 | { |
| 230 | memcpy (cos_maxerror, c3, sizeof (mp1)); |
| 231 | cos_maxerror_s = c3s; |
| 232 | } |
| 233 | } |
| 234 | |
| 235 | /* Check Range-Kutta against precomputed values of sin(1) and cos(1). */ |
| 236 | memset (x, 0, sizeof (mp1)); |
| 237 | x[FRAC / mpbpl] = (mp_limb_t)1 << FRAC % mpbpl; |
| 238 | sincosx_mpn (si, co, x, ox); |
| 239 | |
| 240 | memset (s2, 0, sizeof (mp1)); |
| 241 | memset (c2, 0, sizeof (mp1)); |
| 242 | for (i = 0; i < 100 && i < FRAC / 4; i++) |
| 243 | { |
| 244 | s2[(FRAC - i * 4 - 4) / mpbpl] |= ((mp_limb_t) (strchr (hexdig, sin1[i]) |
| 245 | - hexdig) |
| 246 | << (FRAC - i * 4 - 4) % mpbpl); |
| 247 | c2[(FRAC - i * 4 - 4) / mpbpl] |= ((mp_limb_t) (strchr (hexdig, cos1[i]) |
| 248 | - hexdig) |
| 249 | << (FRAC - i * 4 - 4) % mpbpl); |
| 250 | } |
| 251 | |
| 252 | if (mpn_cmp (si, s2, SZ) >= 0) |
| 253 | mpn_sub_n (s3, si, s2, SZ); |
| 254 | else |
| 255 | mpn_sub_n (s3, s2, si, SZ); |
| 256 | if (mpn_cmp (co, c2, SZ) >= 0) |
| 257 | mpn_sub_n (c3, co, c2, SZ); |
| 258 | else |
| 259 | mpn_sub_n (c3, c2, co, SZ); |
| 260 | |
| 261 | printf ("sin:\n"); |
| 262 | printf ("%d failures; %d errors; error rate %0.2f%%\n", |
| 263 | sin_failures, sin_errors, sin_errors * 100.0 / (double) (1 << N2)); |
| 264 | fputs ("maximum error: ", stdout); |
| 265 | print_mpn_hex (sin_maxerror, (FRAC / 4) + 1); |
| 266 | fputs ("\nerror in sin(1): ", stdout); |
| 267 | print_mpn_hex (s3, (FRAC / 4) + 1); |
| 268 | |
| 269 | fputs ("\n\ncos:\n", stdout); |
| 270 | printf ("%d failures; %d errors; error rate %0.2f%%\n", |
| 271 | cos_failures, cos_errors, cos_errors * 100.0 / (double) (1 << N2)); |
| 272 | fputs ("maximum error: ", stdout); |
| 273 | print_mpn_hex (cos_maxerror, (FRAC / 4) + 1); |
| 274 | fputs ("\nerror in cos(1): ", stdout); |
| 275 | print_mpn_hex (c3, (FRAC / 4) + 1); |
| 276 | putchar ('\n'); |
| 277 | |
| 278 | return (sin_failures == 0 && cos_failures == 0) ? 0 : 1; |
| 279 | } |
| 280 | |
| 281 | #define TIMEOUT 600 |
| 282 | #define TEST_FUNCTION do_test () |
| 283 | #include "../test-skeleton.c" |