Mercurial > hg > sv-dependency-builds

annotate src/fftw-3.3.5/tests/bench.c @ 56:af97cad61ff0

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
Add updated build of PortAudio for OSX
author Chris Cannam <cannam@all-day-breakfast.com>
date Tue, 03 Jan 2017 15:10:52 +0000
parents 2cd0e3b3e1fd
children
rev   line source
Chris@42 1 /**************************************************************************/
Chris@42 2 /* NOTE to users: this is the FFTW self-test and benchmark program.
Chris@42 3 It is probably NOT a good place to learn FFTW usage, since it has a
Chris@42 4 lot of added complexity in order to exercise and test the full API,
Chris@42 5 etcetera. We suggest reading the manual.
Chris@42 6
Chris@42 7 (Some of the self-test code is split off into fftw-bench.c and
Chris@42 8 hook.c.) */
Chris@42 9 /**************************************************************************/
Chris@42 10
Chris@42 11 #include <math.h>
Chris@42 12 #include <stdio.h>
Chris@42 13 #include <string.h>
Chris@42 14 #include "fftw-bench.h"
Chris@42 15
Chris@42 16 static const char *mkversion(void) { return FFTW(version); }
Chris@42 17 static const char *mkcc(void) { return FFTW(cc); }
Chris@42 18 static const char *mkcodelet_optim(void) { return FFTW(codelet_optim); }
Chris@42 19
Chris@42 20 BEGIN_BENCH_DOC
Chris@42 21 BENCH_DOC("name", "fftw3")
Chris@42 22 BENCH_DOCF("version", mkversion)
Chris@42 23 BENCH_DOCF("cc", mkcc)
Chris@42 24 BENCH_DOCF("codelet-optim", mkcodelet_optim)
Chris@42 25 END_BENCH_DOC
Chris@42 26
Chris@42 27 static FFTW(iodim) *bench_tensor_to_fftw_iodim(bench_tensor *t)
Chris@42 28 {
Chris@42 29 FFTW(iodim) *d;
Chris@42 30 int i;
Chris@42 31
Chris@42 32 BENCH_ASSERT(t->rnk >= 0);
Chris@42 33 if (t->rnk == 0) return 0;
Chris@42 34
Chris@42 35 d = (FFTW(iodim) *)bench_malloc(sizeof(FFTW(iodim)) * t->rnk);
Chris@42 36 for (i = 0; i < t->rnk; ++i) {
Chris@42 37 d[i].n = t->dims[i].n;
Chris@42 38 d[i].is = t->dims[i].is;
Chris@42 39 d[i].os = t->dims[i].os;
Chris@42 40 }
Chris@42 41
Chris@42 42 return d;
Chris@42 43 }
Chris@42 44
Chris@42 45 static void extract_reim_split(int sign, int size, bench_real *p,
Chris@42 46 bench_real **r, bench_real **i)
Chris@42 47 {
Chris@42 48 if (sign == FFTW_FORWARD) {
Chris@42 49 *r = p + 0;
Chris@42 50 *i = p + size;
Chris@42 51 } else {
Chris@42 52 *r = p + size;
Chris@42 53 *i = p + 0;
Chris@42 54 }
Chris@42 55 }
Chris@42 56
Chris@42 57 static int sizeof_problem(bench_problem *p)
Chris@42 58 {
Chris@42 59 return tensor_sz(p->sz) * tensor_sz(p->vecsz);
Chris@42 60 }
Chris@42 61
Chris@42 62 /* ouch */
Chris@42 63 static int expressible_as_api_many(bench_tensor *t)
Chris@42 64 {
Chris@42 65 int i;
Chris@42 66
Chris@42 67 BENCH_ASSERT(BENCH_FINITE_RNK(t->rnk));
Chris@42 68
Chris@42 69 i = t->rnk - 1;
Chris@42 70 while (--i >= 0) {
Chris@42 71 bench_iodim *d = t->dims + i;
Chris@42 72 if (d[0].is % d[1].is) return 0;
Chris@42 73 if (d[0].os % d[1].os) return 0;
Chris@42 74 }
Chris@42 75 return 1;
Chris@42 76 }
Chris@42 77
Chris@42 78 static int *mkn(bench_tensor *t)
Chris@42 79 {
Chris@42 80 int *n = (int *) bench_malloc(sizeof(int *) * t->rnk);
Chris@42 81 int i;
Chris@42 82 for (i = 0; i < t->rnk; ++i)
Chris@42 83 n[i] = t->dims[i].n;
Chris@42 84 return n;
Chris@42 85 }
Chris@42 86
Chris@42 87 static void mknembed_many(bench_tensor *t, int **inembedp, int **onembedp)
Chris@42 88 {
Chris@42 89 int i;
Chris@42 90 bench_iodim *d;
Chris@42 91 int *inembed = (int *) bench_malloc(sizeof(int *) * t->rnk);
Chris@42 92 int *onembed = (int *) bench_malloc(sizeof(int *) * t->rnk);
Chris@42 93
Chris@42 94 BENCH_ASSERT(BENCH_FINITE_RNK(t->rnk));
Chris@42 95 *inembedp = inembed; *onembedp = onembed;
Chris@42 96
Chris@42 97 i = t->rnk - 1;
Chris@42 98 while (--i >= 0) {
Chris@42 99 d = t->dims + i;
Chris@42 100 inembed[i+1] = d[0].is / d[1].is;
Chris@42 101 onembed[i+1] = d[0].os / d[1].os;
Chris@42 102 }
Chris@42 103 }
Chris@42 104
Chris@42 105 /* try to use the most appropriate API function. Big mess. */
Chris@42 106
Chris@42 107 static int imax(int a, int b) { return (a > b ? a : b); }
Chris@42 108
Chris@42 109 static int halfish_sizeof_problem(bench_problem *p)
Chris@42 110 {
Chris@42 111 int n2 = sizeof_problem(p);
Chris@42 112 if (BENCH_FINITE_RNK(p->sz->rnk) && p->sz->rnk > 0)
Chris@42 113 n2 = (n2 / imax(p->sz->dims[p->sz->rnk - 1].n, 1)) *
Chris@42 114 (p->sz->dims[p->sz->rnk - 1].n / 2 + 1);
Chris@42 115 return n2;
Chris@42 116 }
Chris@42 117
Chris@42 118 static FFTW(plan) mkplan_real_split(bench_problem *p, unsigned flags)
Chris@42 119 {
Chris@42 120 FFTW(plan) pln;
Chris@42 121 bench_tensor *sz = p->sz, *vecsz = p->vecsz;
Chris@42 122 FFTW(iodim) *dims, *howmany_dims;
Chris@42 123 bench_real *ri, *ii, *ro, *io;
Chris@42 124 int n2 = halfish_sizeof_problem(p);
Chris@42 125
Chris@42 126 extract_reim_split(FFTW_FORWARD, n2, (bench_real *) p->in, &ri, &ii);
Chris@42 127 extract_reim_split(FFTW_FORWARD, n2, (bench_real *) p->out, &ro, &io);
Chris@42 128
Chris@42 129 dims = bench_tensor_to_fftw_iodim(sz);
Chris@42 130 howmany_dims = bench_tensor_to_fftw_iodim(vecsz);
Chris@42 131 if (p->sign < 0) {
Chris@42 132 if (verbose > 2) printf("using plan_guru_split_dft_r2c\n");
Chris@42 133 pln = FFTW(plan_guru_split_dft_r2c)(sz->rnk, dims,
Chris@42 134 vecsz->rnk, howmany_dims,
Chris@42 135 ri, ro, io, flags);
Chris@42 136 }
Chris@42 137 else {
Chris@42 138 if (verbose > 2) printf("using plan_guru_split_dft_c2r\n");
Chris@42 139 pln = FFTW(plan_guru_split_dft_c2r)(sz->rnk, dims,
Chris@42 140 vecsz->rnk, howmany_dims,
Chris@42 141 ri, ii, ro, flags);
Chris@42 142 }
Chris@42 143 bench_free(dims);
Chris@42 144 bench_free(howmany_dims);
Chris@42 145 return pln;
Chris@42 146 }
Chris@42 147
Chris@42 148 static FFTW(plan) mkplan_real_interleaved(bench_problem *p, unsigned flags)
Chris@42 149 {
Chris@42 150 FFTW(plan) pln;
Chris@42 151 bench_tensor *sz = p->sz, *vecsz = p->vecsz;
Chris@42 152
Chris@42 153 if (vecsz->rnk == 0 && tensor_unitstridep(sz)
Chris@42 154 && tensor_real_rowmajorp(sz, p->sign, p->in_place))
Chris@42 155 goto api_simple;
Chris@42 156
Chris@42 157 if (vecsz->rnk == 1 && expressible_as_api_many(sz))
Chris@42 158 goto api_many;
Chris@42 159
Chris@42 160 goto api_guru;
Chris@42 161
Chris@42 162 api_simple:
Chris@42 163 switch (sz->rnk) {
Chris@42 164 case 1:
Chris@42 165 if (p->sign < 0) {
Chris@42 166 if (verbose > 2) printf("using plan_dft_r2c_1d\n");
Chris@42 167 return FFTW(plan_dft_r2c_1d)(sz->dims[0].n,
Chris@42 168 (bench_real *) p->in,
Chris@42 169 (bench_complex *) p->out,
Chris@42 170 flags);
Chris@42 171 }
Chris@42 172 else {
Chris@42 173 if (verbose > 2) printf("using plan_dft_c2r_1d\n");
Chris@42 174 return FFTW(plan_dft_c2r_1d)(sz->dims[0].n,
Chris@42 175 (bench_complex *) p->in,
Chris@42 176 (bench_real *) p->out,
Chris@42 177 flags);
Chris@42 178 }
Chris@42 179 break;
Chris@42 180 case 2:
Chris@42 181 if (p->sign < 0) {
Chris@42 182 if (verbose > 2) printf("using plan_dft_r2c_2d\n");
Chris@42 183 return FFTW(plan_dft_r2c_2d)(sz->dims[0].n, sz->dims[1].n,
Chris@42 184 (bench_real *) p->in,
Chris@42 185 (bench_complex *) p->out,
Chris@42 186 flags);
Chris@42 187 }
Chris@42 188 else {
Chris@42 189 if (verbose > 2) printf("using plan_dft_c2r_2d\n");
Chris@42 190 return FFTW(plan_dft_c2r_2d)(sz->dims[0].n, sz->dims[1].n,
Chris@42 191 (bench_complex *) p->in,
Chris@42 192 (bench_real *) p->out,
Chris@42 193 flags);
Chris@42 194 }
Chris@42 195 break;
Chris@42 196 case 3:
Chris@42 197 if (p->sign < 0) {
Chris@42 198 if (verbose > 2) printf("using plan_dft_r2c_3d\n");
Chris@42 199 return FFTW(plan_dft_r2c_3d)(
Chris@42 200 sz->dims[0].n, sz->dims[1].n, sz->dims[2].n,
Chris@42 201 (bench_real *) p->in, (bench_complex *) p->out,
Chris@42 202 flags);
Chris@42 203 }
Chris@42 204 else {
Chris@42 205 if (verbose > 2) printf("using plan_dft_c2r_3d\n");
Chris@42 206 return FFTW(plan_dft_c2r_3d)(
Chris@42 207 sz->dims[0].n, sz->dims[1].n, sz->dims[2].n,
Chris@42 208 (bench_complex *) p->in, (bench_real *) p->out,
Chris@42 209 flags);
Chris@42 210 }
Chris@42 211 break;
Chris@42 212 default: {
Chris@42 213 int *n = mkn(sz);
Chris@42 214 if (p->sign < 0) {
Chris@42 215 if (verbose > 2) printf("using plan_dft_r2c\n");
Chris@42 216 pln = FFTW(plan_dft_r2c)(sz->rnk, n,
Chris@42 217 (bench_real *) p->in,
Chris@42 218 (bench_complex *) p->out,
Chris@42 219 flags);
Chris@42 220 }
Chris@42 221 else {
Chris@42 222 if (verbose > 2) printf("using plan_dft_c2r\n");
Chris@42 223 pln = FFTW(plan_dft_c2r)(sz->rnk, n,
Chris@42 224 (bench_complex *) p->in,
Chris@42 225 (bench_real *) p->out,
Chris@42 226 flags);
Chris@42 227 }
Chris@42 228 bench_free(n);
Chris@42 229 return pln;
Chris@42 230 }
Chris@42 231 }
Chris@42 232
Chris@42 233 api_many:
Chris@42 234 {
Chris@42 235 int *n, *inembed, *onembed;
Chris@42 236 BENCH_ASSERT(vecsz->rnk == 1);
Chris@42 237 n = mkn(sz);
Chris@42 238 mknembed_many(sz, &inembed, &onembed);
Chris@42 239 if (p->sign < 0) {
Chris@42 240 if (verbose > 2) printf("using plan_many_dft_r2c\n");
Chris@42 241 pln = FFTW(plan_many_dft_r2c)(
Chris@42 242 sz->rnk, n, vecsz->dims[0].n,
Chris@42 243 (bench_real *) p->in, inembed,
Chris@42 244 sz->dims[sz->rnk - 1].is, vecsz->dims[0].is,
Chris@42 245 (bench_complex *) p->out, onembed,
Chris@42 246 sz->dims[sz->rnk - 1].os, vecsz->dims[0].os,
Chris@42 247 flags);
Chris@42 248 }
Chris@42 249 else {
Chris@42 250 if (verbose > 2) printf("using plan_many_dft_c2r\n");
Chris@42 251 pln = FFTW(plan_many_dft_c2r)(
Chris@42 252 sz->rnk, n, vecsz->dims[0].n,
Chris@42 253 (bench_complex *) p->in, inembed,
Chris@42 254 sz->dims[sz->rnk - 1].is, vecsz->dims[0].is,
Chris@42 255 (bench_real *) p->out, onembed,
Chris@42 256 sz->dims[sz->rnk - 1].os, vecsz->dims[0].os,
Chris@42 257 flags);
Chris@42 258 }
Chris@42 259 bench_free(n); bench_free(inembed); bench_free(onembed);
Chris@42 260 return pln;
Chris@42 261 }
Chris@42 262
Chris@42 263 api_guru:
Chris@42 264 {
Chris@42 265 FFTW(iodim) *dims, *howmany_dims;
Chris@42 266
Chris@42 267 if (p->sign < 0) {
Chris@42 268 dims = bench_tensor_to_fftw_iodim(sz);
Chris@42 269 howmany_dims = bench_tensor_to_fftw_iodim(vecsz);
Chris@42 270 if (verbose > 2) printf("using plan_guru_dft_r2c\n");
Chris@42 271 pln = FFTW(plan_guru_dft_r2c)(sz->rnk, dims,
Chris@42 272 vecsz->rnk, howmany_dims,
Chris@42 273 (bench_real *) p->in,
Chris@42 274 (bench_complex *) p->out,
Chris@42 275 flags);
Chris@42 276 }
Chris@42 277 else {
Chris@42 278 dims = bench_tensor_to_fftw_iodim(sz);
Chris@42 279 howmany_dims = bench_tensor_to_fftw_iodim(vecsz);
Chris@42 280 if (verbose > 2) printf("using plan_guru_dft_c2r\n");
Chris@42 281 pln = FFTW(plan_guru_dft_c2r)(sz->rnk, dims,
Chris@42 282 vecsz->rnk, howmany_dims,
Chris@42 283 (bench_complex *) p->in,
Chris@42 284 (bench_real *) p->out,
Chris@42 285 flags);
Chris@42 286 }
Chris@42 287 bench_free(dims);
Chris@42 288 bench_free(howmany_dims);
Chris@42 289 return pln;
Chris@42 290 }
Chris@42 291 }
Chris@42 292
Chris@42 293 static FFTW(plan) mkplan_real(bench_problem *p, unsigned flags)
Chris@42 294 {
Chris@42 295 if (p->split)
Chris@42 296 return mkplan_real_split(p, flags);
Chris@42 297 else
Chris@42 298 return mkplan_real_interleaved(p, flags);
Chris@42 299 }
Chris@42 300
Chris@42 301 static FFTW(plan) mkplan_complex_split(bench_problem *p, unsigned flags)
Chris@42 302 {
Chris@42 303 FFTW(plan) pln;
Chris@42 304 bench_tensor *sz = p->sz, *vecsz = p->vecsz;
Chris@42 305 FFTW(iodim) *dims, *howmany_dims;
Chris@42 306 bench_real *ri, *ii, *ro, *io;
Chris@42 307
Chris@42 308 extract_reim_split(p->sign, p->iphyssz, (bench_real *) p->in, &ri, &ii);
Chris@42 309 extract_reim_split(p->sign, p->ophyssz, (bench_real *) p->out, &ro, &io);
Chris@42 310
Chris@42 311 dims = bench_tensor_to_fftw_iodim(sz);
Chris@42 312 howmany_dims = bench_tensor_to_fftw_iodim(vecsz);
Chris@42 313 if (verbose > 2) printf("using plan_guru_split_dft\n");
Chris@42 314 pln = FFTW(plan_guru_split_dft)(sz->rnk, dims,
Chris@42 315 vecsz->rnk, howmany_dims,
Chris@42 316 ri, ii, ro, io, flags);
Chris@42 317 bench_free(dims);
Chris@42 318 bench_free(howmany_dims);
Chris@42 319 return pln;
Chris@42 320 }
Chris@42 321
Chris@42 322 static FFTW(plan) mkplan_complex_interleaved(bench_problem *p, unsigned flags)
Chris@42 323 {
Chris@42 324 FFTW(plan) pln;
Chris@42 325 bench_tensor *sz = p->sz, *vecsz = p->vecsz;
Chris@42 326
Chris@42 327 if (vecsz->rnk == 0 && tensor_unitstridep(sz) && tensor_rowmajorp(sz))
Chris@42 328 goto api_simple;
Chris@42 329
Chris@42 330 if (vecsz->rnk == 1 && expressible_as_api_many(sz))
Chris@42 331 goto api_many;
Chris@42 332
Chris@42 333 goto api_guru;
Chris@42 334
Chris@42 335 api_simple:
Chris@42 336 switch (sz->rnk) {
Chris@42 337 case 1:
Chris@42 338 if (verbose > 2) printf("using plan_dft_1d\n");
Chris@42 339 return FFTW(plan_dft_1d)(sz->dims[0].n,
Chris@42 340 (bench_complex *) p->in,
Chris@42 341 (bench_complex *) p->out,
Chris@42 342 p->sign, flags);
Chris@42 343 break;
Chris@42 344 case 2:
Chris@42 345 if (verbose > 2) printf("using plan_dft_2d\n");
Chris@42 346 return FFTW(plan_dft_2d)(sz->dims[0].n, sz->dims[1].n,
Chris@42 347 (bench_complex *) p->in,
Chris@42 348 (bench_complex *) p->out,
Chris@42 349 p->sign, flags);
Chris@42 350 break;
Chris@42 351 case 3:
Chris@42 352 if (verbose > 2) printf("using plan_dft_3d\n");
Chris@42 353 return FFTW(plan_dft_3d)(
Chris@42 354 sz->dims[0].n, sz->dims[1].n, sz->dims[2].n,
Chris@42 355 (bench_complex *) p->in, (bench_complex *) p->out,
Chris@42 356 p->sign, flags);
Chris@42 357 break;
Chris@42 358 default: {
Chris@42 359 int *n = mkn(sz);
Chris@42 360 if (verbose > 2) printf("using plan_dft\n");
Chris@42 361 pln = FFTW(plan_dft)(sz->rnk, n,
Chris@42 362 (bench_complex *) p->in,
Chris@42 363 (bench_complex *) p->out, p->sign, flags);
Chris@42 364 bench_free(n);
Chris@42 365 return pln;
Chris@42 366 }
Chris@42 367 }
Chris@42 368
Chris@42 369 api_many:
Chris@42 370 {
Chris@42 371 int *n, *inembed, *onembed;
Chris@42 372 BENCH_ASSERT(vecsz->rnk == 1);
Chris@42 373 n = mkn(sz);
Chris@42 374 mknembed_many(sz, &inembed, &onembed);
Chris@42 375 if (verbose > 2) printf("using plan_many_dft\n");
Chris@42 376 pln = FFTW(plan_many_dft)(
Chris@42 377 sz->rnk, n, vecsz->dims[0].n,
Chris@42 378 (bench_complex *) p->in,
Chris@42 379 inembed, sz->dims[sz->rnk - 1].is, vecsz->dims[0].is,
Chris@42 380 (bench_complex *) p->out,
Chris@42 381 onembed, sz->dims[sz->rnk - 1].os, vecsz->dims[0].os,
Chris@42 382 p->sign, flags);
Chris@42 383 bench_free(n); bench_free(inembed); bench_free(onembed);
Chris@42 384 return pln;
Chris@42 385 }
Chris@42 386
Chris@42 387 api_guru:
Chris@42 388 {
Chris@42 389 FFTW(iodim) *dims, *howmany_dims;
Chris@42 390
Chris@42 391 dims = bench_tensor_to_fftw_iodim(sz);
Chris@42 392 howmany_dims = bench_tensor_to_fftw_iodim(vecsz);
Chris@42 393 if (verbose > 2) printf("using plan_guru_dft\n");
Chris@42 394 pln = FFTW(plan_guru_dft)(sz->rnk, dims,
Chris@42 395 vecsz->rnk, howmany_dims,
Chris@42 396 (bench_complex *) p->in,
Chris@42 397 (bench_complex *) p->out,
Chris@42 398 p->sign, flags);
Chris@42 399 bench_free(dims);
Chris@42 400 bench_free(howmany_dims);
Chris@42 401 return pln;
Chris@42 402 }
Chris@42 403 }
Chris@42 404
Chris@42 405 static FFTW(plan) mkplan_complex(bench_problem *p, unsigned flags)
Chris@42 406 {
Chris@42 407 if (p->split)
Chris@42 408 return mkplan_complex_split(p, flags);
Chris@42 409 else
Chris@42 410 return mkplan_complex_interleaved(p, flags);
Chris@42 411 }
Chris@42 412
Chris@42 413 static FFTW(plan) mkplan_r2r(bench_problem *p, unsigned flags)
Chris@42 414 {
Chris@42 415 FFTW(plan) pln;
Chris@42 416 bench_tensor *sz = p->sz, *vecsz = p->vecsz;
Chris@42 417 FFTW(r2r_kind) *k;
Chris@42 418
Chris@42 419 k = (FFTW(r2r_kind) *) bench_malloc(sizeof(FFTW(r2r_kind)) * sz->rnk);
Chris@42 420 {
Chris@42 421 int i;
Chris@42 422 for (i = 0; i < sz->rnk; ++i)
Chris@42 423 switch (p->k[i]) {
Chris@42 424 case R2R_R2HC: k[i] = FFTW_R2HC; break;
Chris@42 425 case R2R_HC2R: k[i] = FFTW_HC2R; break;
Chris@42 426 case R2R_DHT: k[i] = FFTW_DHT; break;
Chris@42 427 case R2R_REDFT00: k[i] = FFTW_REDFT00; break;
Chris@42 428 case R2R_REDFT01: k[i] = FFTW_REDFT01; break;
Chris@42 429 case R2R_REDFT10: k[i] = FFTW_REDFT10; break;
Chris@42 430 case R2R_REDFT11: k[i] = FFTW_REDFT11; break;
Chris@42 431 case R2R_RODFT00: k[i] = FFTW_RODFT00; break;
Chris@42 432 case R2R_RODFT01: k[i] = FFTW_RODFT01; break;
Chris@42 433 case R2R_RODFT10: k[i] = FFTW_RODFT10; break;
Chris@42 434 case R2R_RODFT11: k[i] = FFTW_RODFT11; break;
Chris@42 435 default: BENCH_ASSERT(0);
Chris@42 436 }
Chris@42 437 }
Chris@42 438
Chris@42 439 if (vecsz->rnk == 0 && tensor_unitstridep(sz) && tensor_rowmajorp(sz))
Chris@42 440 goto api_simple;
Chris@42 441
Chris@42 442 if (vecsz->rnk == 1 && expressible_as_api_many(sz))
Chris@42 443 goto api_many;
Chris@42 444
Chris@42 445 goto api_guru;
Chris@42 446
Chris@42 447 api_simple:
Chris@42 448 switch (sz->rnk) {
Chris@42 449 case 1:
Chris@42 450 if (verbose > 2) printf("using plan_r2r_1d\n");
Chris@42 451 pln = FFTW(plan_r2r_1d)(sz->dims[0].n,
Chris@42 452 (bench_real *) p->in,
Chris@42 453 (bench_real *) p->out,
Chris@42 454 k[0], flags);
Chris@42 455 goto done;
Chris@42 456 case 2:
Chris@42 457 if (verbose > 2) printf("using plan_r2r_2d\n");
Chris@42 458 pln = FFTW(plan_r2r_2d)(sz->dims[0].n, sz->dims[1].n,
Chris@42 459 (bench_real *) p->in,
Chris@42 460 (bench_real *) p->out,
Chris@42 461 k[0], k[1], flags);
Chris@42 462 goto done;
Chris@42 463 case 3:
Chris@42 464 if (verbose > 2) printf("using plan_r2r_3d\n");
Chris@42 465 pln = FFTW(plan_r2r_3d)(
Chris@42 466 sz->dims[0].n, sz->dims[1].n, sz->dims[2].n,
Chris@42 467 (bench_real *) p->in, (bench_real *) p->out,
Chris@42 468 k[0], k[1], k[2], flags);
Chris@42 469 goto done;
Chris@42 470 default: {
Chris@42 471 int *n = mkn(sz);
Chris@42 472 if (verbose > 2) printf("using plan_r2r\n");
Chris@42 473 pln = FFTW(plan_r2r)(sz->rnk, n,
Chris@42 474 (bench_real *) p->in, (bench_real *) p->out,
Chris@42 475 k, flags);
Chris@42 476 bench_free(n);
Chris@42 477 goto done;
Chris@42 478 }
Chris@42 479 }
Chris@42 480
Chris@42 481 api_many:
Chris@42 482 {
Chris@42 483 int *n, *inembed, *onembed;
Chris@42 484 BENCH_ASSERT(vecsz->rnk == 1);
Chris@42 485 n = mkn(sz);
Chris@42 486 mknembed_many(sz, &inembed, &onembed);
Chris@42 487 if (verbose > 2) printf("using plan_many_r2r\n");
Chris@42 488 pln = FFTW(plan_many_r2r)(
Chris@42 489 sz->rnk, n, vecsz->dims[0].n,
Chris@42 490 (bench_real *) p->in,
Chris@42 491 inembed, sz->dims[sz->rnk - 1].is, vecsz->dims[0].is,
Chris@42 492 (bench_real *) p->out,
Chris@42 493 onembed, sz->dims[sz->rnk - 1].os, vecsz->dims[0].os,
Chris@42 494 k, flags);
Chris@42 495 bench_free(n); bench_free(inembed); bench_free(onembed);
Chris@42 496 goto done;
Chris@42 497 }
Chris@42 498
Chris@42 499 api_guru:
Chris@42 500 {
Chris@42 501 FFTW(iodim) *dims, *howmany_dims;
Chris@42 502
Chris@42 503 dims = bench_tensor_to_fftw_iodim(sz);
Chris@42 504 howmany_dims = bench_tensor_to_fftw_iodim(vecsz);
Chris@42 505 if (verbose > 2) printf("using plan_guru_r2r\n");
Chris@42 506 pln = FFTW(plan_guru_r2r)(sz->rnk, dims,
Chris@42 507 vecsz->rnk, howmany_dims,
Chris@42 508 (bench_real *) p->in,
Chris@42 509 (bench_real *) p->out, k, flags);
Chris@42 510 bench_free(dims);
Chris@42 511 bench_free(howmany_dims);
Chris@42 512 goto done;
Chris@42 513 }
Chris@42 514
Chris@42 515 done:
Chris@42 516 bench_free(k);
Chris@42 517 return pln;
Chris@42 518 }
Chris@42 519
Chris@42 520 FFTW(plan) mkplan(bench_problem *p, unsigned flags)
Chris@42 521 {
Chris@42 522 switch (p->kind) {
Chris@42 523 case PROBLEM_COMPLEX: return mkplan_complex(p, flags);
Chris@42 524 case PROBLEM_REAL: return mkplan_real(p, flags);
Chris@42 525 case PROBLEM_R2R: return mkplan_r2r(p, flags);
Chris@42 526 default: BENCH_ASSERT(0); return 0;
Chris@42 527 }
Chris@42 528 }
Chris@42 529
Chris@42 530 void main_init(int *argc, char ***argv)
Chris@42 531 {
Chris@42 532 UNUSED(argc);
Chris@42 533 UNUSED(argv);
Chris@42 534 }
Chris@42 535
Chris@42 536 void initial_cleanup(void)
Chris@42 537 {
Chris@42 538 }
Chris@42 539
Chris@42 540 void final_cleanup(void)
Chris@42 541 {
Chris@42 542 }
Chris@42 543
Chris@42 544 int import_wisdom(FILE *f)
Chris@42 545 {
Chris@42 546 return FFTW(import_wisdom_from_file)(f);
Chris@42 547 }
Chris@42 548
Chris@42 549 void export_wisdom(FILE *f)
Chris@42 550 {
Chris@42 551 FFTW(export_wisdom_to_file)(f);
Chris@42 552 }