[469] | 1 | /*************************************************************************/ |
---|
| 2 | /* */ |
---|
| 3 | /* Copyright (c) 1994 Stanford University */ |
---|
| 4 | /* */ |
---|
| 5 | /* All rights reserved. */ |
---|
| 6 | /* */ |
---|
| 7 | /* Permission is given to use, copy, and modify this software for any */ |
---|
| 8 | /* non-commercial purpose as long as this copyright notice is not */ |
---|
| 9 | /* removed. All other uses, including redistribution in whole or in */ |
---|
| 10 | /* part, are forbidden without prior written permission. */ |
---|
| 11 | /* */ |
---|
| 12 | /* This software is provided with absolutely no warranty and no */ |
---|
| 13 | /* support. */ |
---|
| 14 | /* */ |
---|
| 15 | /*************************************************************************/ |
---|
| 16 | |
---|
| 17 | /////////////////////////////////////////////////////////////////////////// |
---|
| 18 | // This port of the SPLASH FFT benchmark on the ALMOS-MKH OS has been |
---|
| 19 | // done by Alain Greiner (august 2018). |
---|
| 20 | // |
---|
| 21 | // This application performs the 1D fast Fourier transfom for an array |
---|
| 22 | // of N complex points, using the Cooley-Tuckey FFT method. |
---|
| 23 | // The N data points are seen as a 2D array (rootN rows * rootN columns). |
---|
| 24 | // Each thread handle (rootN / nthreads) rows. The N input data points |
---|
| 25 | // be initialised in three different modes: |
---|
| 26 | // - CONSTANT : all data points have the same [1,0] value |
---|
| 27 | // - COSIN : data point n has [cos(n/N) , sin(n/N)] values |
---|
| 28 | // - RANDOM : data points have pseudo random values |
---|
| 29 | // |
---|
[628] | 30 | // The main parameters for this generic application are the following: |
---|
| 31 | // - M : N = 2**M = number of data points / M must be an even number. |
---|
| 32 | // - T : nthreads = ncores defined by the hardware / must be power of 2. |
---|
| 33 | // |
---|
| 34 | // This application uses 4 shared data arrays, that are dynamically |
---|
| 35 | // allocated an distributed, using the remote_malloc() function, with |
---|
| 36 | // one sub-buffer per cluster: |
---|
[469] | 37 | // - data[N] contains N input data points, with 2 double per point. |
---|
| 38 | // - trans[N] contains N intermediate data points, 2 double per point. |
---|
| 39 | // - umain[rootN] contains rootN coefs required for a rootN points FFT. |
---|
| 40 | // - twid[N] contains N coefs : exp(2*pi*i*j/N) / i and j in [0,rootN-1]. |
---|
| 41 | // For data, trans, twid, each sub-buffer contains (N/nclusters) points. |
---|
| 42 | // For umain, each sub-buffer contains (rootN/nclusters) points. |
---|
| 43 | // |
---|
| 44 | // There is one thread per core. |
---|
| 45 | // The max number of clusters is defined by (X_MAX * Y_MAX). |
---|
| 46 | // The max number of cores per cluster is defined by CORES_MAX. |
---|
| 47 | // |
---|
| 48 | // Several configuration parameters can be defined below: |
---|
[574] | 49 | // - PRINT_ARRAY : Print out complex data points arrays. |
---|
| 50 | // - CHECK : Perform both FFT and inverse FFT to check output/input. |
---|
| 51 | // - DEBUG_MAIN : Display intermediate results in main() |
---|
| 52 | // - DEBUG_FFT1D : Display intermediate results in FFT1D() |
---|
[628] | 53 | // - DEBUG_ROW : Display intermedite results in FFTrow() |
---|
[469] | 54 | // |
---|
| 55 | // Regarding final instrumentation: |
---|
| 56 | // - the sequencial initialisation time (init_time) is computed |
---|
| 57 | // by the main thread in the main() function. |
---|
| 58 | // - The parallel execution time (parallel_time[i]) is computed by each |
---|
| 59 | // thread(i) in the slave() function. |
---|
| 60 | // - The synchronisation time related to the barriers (sync_time[i]) |
---|
| 61 | // is computed by each thread(i) in the slave() function. |
---|
| 62 | // The results are displayed on the TXT terminal, and registered on disk. |
---|
| 63 | /////////////////////////////////////////////////////////////////////////// |
---|
| 64 | |
---|
| 65 | #include <math.h> |
---|
| 66 | #include <stdio.h> |
---|
| 67 | #include <stdlib.h> |
---|
| 68 | #include <fcntl.h> |
---|
| 69 | #include <unistd.h> |
---|
| 70 | #include <pthread.h> |
---|
| 71 | #include <almosmkh.h> |
---|
| 72 | #include <hal_macros.h> |
---|
| 73 | |
---|
| 74 | // constants |
---|
| 75 | |
---|
| 76 | #define PI 3.14159265359 |
---|
| 77 | #define PAGE_SIZE 4096 |
---|
| 78 | #define X_MAX 16 // max number of clusters in a row |
---|
| 79 | #define Y_MAX 16 // max number of clusters in a column |
---|
| 80 | #define CORES_MAX 4 // max number of cores in a cluster |
---|
| 81 | #define CLUSTERS_MAX X_MAX * Y_MAX |
---|
| 82 | #define THREADS_MAX CLUSTERS_MAX * CORES_MAX |
---|
| 83 | #define RANDOM 0 |
---|
| 84 | #define COSIN 1 |
---|
| 85 | #define CONSTANT 2 |
---|
| 86 | |
---|
| 87 | // parameters |
---|
| 88 | |
---|
[636] | 89 | #define DEFAULT_M 12 // 4096 data points |
---|
[628] | 90 | #define USE_DQT_BARRIER 0 // use DDT barrier if non zero |
---|
| 91 | #define MODE COSIN // DATA array initialisation mode |
---|
| 92 | #define CHECK 0 |
---|
[630] | 93 | #define DEBUG_MAIN 0 // trace main() function (detailed if odd) |
---|
[628] | 94 | #define DEBUG_SLAVE 0 // trace slave() function (detailed if odd) |
---|
| 95 | #define DEBUG_FFT1D 0 // trace FFT1D() function (detailed if odd) |
---|
[574] | 96 | #define DEBUG_ROW 0 // trace FFTRow() function (detailed if odd) |
---|
| 97 | #define PRINT_ARRAY 0 |
---|
[469] | 98 | |
---|
| 99 | // macro to swap two variables |
---|
| 100 | #define SWAP(a,b) { double tmp; tmp = a; a = b; b = tmp; } |
---|
| 101 | |
---|
[588] | 102 | ///////////////////////////////////////////////////////////////////////////////////// |
---|
| 103 | // structure containing the arguments for the slave() function |
---|
| 104 | ///////////////////////////////////////////////////////////////////////////////////// |
---|
| 105 | |
---|
| 106 | typedef struct args_s |
---|
| 107 | { |
---|
| 108 | unsigned int tid; // thread continuous index |
---|
| 109 | unsigned int main_tid; // main thread continuous index |
---|
| 110 | } |
---|
| 111 | args_t; |
---|
| 112 | |
---|
| 113 | ///////////////////////////////////////////////////////////////////////////////////// |
---|
[469] | 114 | // global variables |
---|
[588] | 115 | ///////////////////////////////////////////////////////////////////////////////////// |
---|
[469] | 116 | |
---|
| 117 | unsigned int x_size; // number of clusters per row in the mesh |
---|
| 118 | unsigned int y_size; // number of clusters per column in the mesh |
---|
| 119 | unsigned int ncores; // number of cores per cluster |
---|
[473] | 120 | unsigned int nthreads; // total number of threads (one thread per core) |
---|
| 121 | unsigned int nclusters; // total number of clusters |
---|
| 122 | unsigned int M = DEFAULT_M; // log2(number of points) |
---|
| 123 | unsigned int N; // number of points (N = 2^M) |
---|
| 124 | unsigned int rootN; // rootN = 2^M/2 |
---|
| 125 | unsigned int rows_per_thread; // number of data "rows" handled by a single thread |
---|
| 126 | unsigned int points_per_cluster; // number of data points per cluster |
---|
[469] | 127 | |
---|
| 128 | // arrays of pointers on distributed buffers (one sub-buffer per cluster) |
---|
| 129 | double * data[CLUSTERS_MAX]; // original time-domain data |
---|
| 130 | double * trans[CLUSTERS_MAX]; // used as auxiliary space for transpose |
---|
| 131 | double * bloup[CLUSTERS_MAX]; // used as auxiliary space for DFT |
---|
| 132 | double * umain[CLUSTERS_MAX]; // roots of unity used fo rootN points FFT |
---|
| 133 | double * twid[CLUSTERS_MAX]; // twiddle factor : exp(-2iPI*k*n/N) |
---|
| 134 | |
---|
| 135 | // instrumentation counters |
---|
[588] | 136 | unsigned int parallel_time[THREADS_MAX]; // total computation time (per thread) |
---|
| 137 | unsigned int sync_time[THREADS_MAX]; // cumulated waiting time in barriers (per thread) |
---|
| 138 | unsigned int init_time; // initialisation time (in main) |
---|
[469] | 139 | |
---|
| 140 | // synchronisation barrier (all threads) |
---|
| 141 | pthread_barrier_t barrier; |
---|
[628] | 142 | pthread_barrierattr_t barrier_attr; |
---|
[469] | 143 | |
---|
| 144 | // threads identifiers, attributes, and arguments |
---|
| 145 | pthread_t trdid[THREADS_MAX]; // kernel threads identifiers |
---|
| 146 | pthread_attr_t attr[THREADS_MAX]; // POSIX thread attributes |
---|
[588] | 147 | args_t args[THREADS_MAX]; // slave function arguments |
---|
[469] | 148 | |
---|
| 149 | ///////////////////////////////////////////////////////////////////////////////// |
---|
| 150 | // functions declaration |
---|
| 151 | ///////////////////////////////////////////////////////////////////////////////// |
---|
| 152 | |
---|
[588] | 153 | void slave( args_t * args ); |
---|
[469] | 154 | |
---|
[503] | 155 | double CheckSum( void ); |
---|
[469] | 156 | |
---|
| 157 | void InitX(double ** x , unsigned int mode); |
---|
| 158 | |
---|
| 159 | void InitU(double ** u); |
---|
| 160 | |
---|
| 161 | void InitT(double ** u); |
---|
| 162 | |
---|
[473] | 163 | unsigned int BitReverse( unsigned int k ); |
---|
[469] | 164 | |
---|
[473] | 165 | void FFT1D( int direction, |
---|
| 166 | double ** x, |
---|
| 167 | double ** tmp, |
---|
| 168 | double * upriv, |
---|
| 169 | double ** twid, |
---|
| 170 | unsigned int MyNum, |
---|
| 171 | unsigned int MyFirst, |
---|
| 172 | unsigned int MyLast ); |
---|
[469] | 173 | |
---|
[473] | 174 | void TwiddleOneCol( int direction, |
---|
| 175 | unsigned int j, |
---|
| 176 | double ** u, |
---|
| 177 | double ** x, |
---|
| 178 | unsigned int offset_x ); |
---|
[469] | 179 | |
---|
[473] | 180 | void Scale( double ** x, |
---|
| 181 | unsigned int offset_x ); |
---|
[469] | 182 | |
---|
[473] | 183 | void Transpose( double ** src, |
---|
| 184 | double ** dest, |
---|
| 185 | unsigned int MyFirst, |
---|
| 186 | unsigned int MyLast ); |
---|
[469] | 187 | |
---|
[473] | 188 | void Copy( double ** src, |
---|
| 189 | double ** dest, |
---|
| 190 | unsigned int MyFirst, |
---|
| 191 | unsigned int MyLast ); |
---|
[469] | 192 | |
---|
[473] | 193 | void Reverse( double ** x, |
---|
| 194 | unsigned int offset_x ); |
---|
[469] | 195 | |
---|
[574] | 196 | void FFTRow( int direction, |
---|
[473] | 197 | double * u, |
---|
| 198 | double ** x, |
---|
| 199 | unsigned int offset_x ); |
---|
[469] | 200 | |
---|
[473] | 201 | void PrintArray( double ** x, |
---|
| 202 | unsigned int size ); |
---|
[469] | 203 | |
---|
[473] | 204 | void SimpleDft( int direction, |
---|
| 205 | unsigned int size, |
---|
| 206 | double ** src, |
---|
| 207 | unsigned int src_offset, |
---|
| 208 | double ** dst, |
---|
| 209 | unsigned int dst_offset ); |
---|
[469] | 210 | |
---|
| 211 | /////////////////////////////////////////////////////////////////// |
---|
| 212 | // This main() function execute the sequencial initialisation |
---|
| 213 | // launch the parallel execution, and makes the instrumentation. |
---|
| 214 | /////////////////////////////////////////////////////////////////// |
---|
[503] | 215 | void main ( void ) |
---|
[469] | 216 | { |
---|
[628] | 217 | int error; |
---|
| 218 | |
---|
[469] | 219 | unsigned int main_cxy; // main thread cluster |
---|
| 220 | unsigned int main_x; // main thread X coordinate |
---|
| 221 | unsigned int main_y; // main thread y coordinate |
---|
| 222 | unsigned int main_lid; // main thread local core index |
---|
| 223 | unsigned int main_tid; // main thread continuous index |
---|
| 224 | |
---|
| 225 | unsigned int x; // current index for cluster X coordinate |
---|
| 226 | unsigned int y; // current index for cluster Y coordinate |
---|
| 227 | unsigned int lid; // current index for core in a cluster |
---|
| 228 | unsigned int ci; // continuous cluster index (from x,y) |
---|
| 229 | unsigned int cxy; // hardware specific cluster identifier |
---|
| 230 | unsigned int tid; // continuous thread index |
---|
| 231 | |
---|
| 232 | unsigned long long start_init_cycle; |
---|
[588] | 233 | unsigned long long end_init_cycle; |
---|
[469] | 234 | |
---|
| 235 | #if CHECK |
---|
| 236 | double ck1; // for input/output checking |
---|
| 237 | double ck3; // for input/output checking |
---|
| 238 | #endif |
---|
| 239 | |
---|
| 240 | // get FFT application start cycle |
---|
[588] | 241 | get_cycle( &start_init_cycle ); |
---|
[469] | 242 | |
---|
| 243 | // get platform parameters to compute nthreads & nclusters |
---|
| 244 | if( get_config( &x_size , &y_size , &ncores ) ) |
---|
| 245 | { |
---|
[628] | 246 | printf("\n[fft error] cannot get hardware configuration\n"); |
---|
[469] | 247 | exit( 0 ); |
---|
| 248 | } |
---|
| 249 | |
---|
| 250 | // check ncores |
---|
| 251 | if( (ncores != 1) && (ncores != 2) && (ncores != 4) ) |
---|
| 252 | { |
---|
[628] | 253 | printf("\n[fft error] number of cores per cluster must be 1/2/4\n"); |
---|
[469] | 254 | exit( 0 ); |
---|
| 255 | } |
---|
| 256 | |
---|
| 257 | // check x_size |
---|
| 258 | if( (x_size != 1) && (x_size != 2) && (x_size != 4) && (x_size != 8) && (x_size != 16) ) |
---|
| 259 | { |
---|
[628] | 260 | printf("\n[fft error] x_size must be 1/2/4/8/16\n"); |
---|
[469] | 261 | exit( 0 ); |
---|
| 262 | } |
---|
| 263 | |
---|
| 264 | // check y_size |
---|
| 265 | if( (y_size != 1) && (y_size != 2) && (y_size != 4) && (y_size != 8) && (y_size != 16) ) |
---|
| 266 | { |
---|
[628] | 267 | printf("\n[fft error] y_size must be 1/2/4/8/16\n"); |
---|
[469] | 268 | exit( 0 ); |
---|
| 269 | } |
---|
| 270 | |
---|
| 271 | nthreads = x_size * y_size * ncores; |
---|
| 272 | nclusters = x_size * y_size; |
---|
| 273 | |
---|
| 274 | // compute various constants depending on N and T |
---|
| 275 | N = 1 << M; |
---|
| 276 | rootN = 1 << (M / 2); |
---|
| 277 | rows_per_thread = rootN / nthreads; |
---|
| 278 | points_per_cluster = N / nclusters; |
---|
| 279 | |
---|
| 280 | // check N versus T |
---|
| 281 | if( rootN < nthreads ) |
---|
| 282 | { |
---|
[628] | 283 | printf("\n[fft error] sqrt(N) must be larger than T\n"); |
---|
[469] | 284 | exit( 0 ); |
---|
| 285 | } |
---|
| 286 | |
---|
| 287 | // get main thread coordinates (main_x, main_y, main_lid) |
---|
| 288 | get_core( &main_cxy , &main_lid ); |
---|
| 289 | main_x = HAL_X_FROM_CXY( main_cxy ); |
---|
| 290 | main_y = HAL_Y_FROM_CXY( main_cxy ); |
---|
| 291 | main_tid = (((main_x * y_size) + main_y) * ncores) + main_lid; |
---|
| 292 | |
---|
[635] | 293 | printf("\n[fft] starts / core[%x,%d] / %d points / %d thread(s) / PID %x / cycle %d\n", |
---|
| 294 | main_cxy, main_lid, N, nthreads, getpid(), (unsigned int)start_init_cycle ); |
---|
[469] | 295 | |
---|
| 296 | // allocate memory for the distributed data[i], trans[i], umain[i], twid[i] buffers |
---|
| 297 | // the index (i) is a continuous cluster index |
---|
[473] | 298 | unsigned int data_size = (N / nclusters) * 2 * sizeof(double); |
---|
| 299 | unsigned int coefs_size = (rootN / nclusters) * 2 * sizeof(double); |
---|
[469] | 300 | for (x = 0 ; x < x_size ; x++) |
---|
| 301 | { |
---|
| 302 | for (y = 0 ; y < y_size ; y++) |
---|
| 303 | { |
---|
| 304 | ci = x * y_size + y; |
---|
| 305 | cxy = HAL_CXY_FROM_XY( x , y ); |
---|
| 306 | data[ci] = (double *)remote_malloc( data_size , cxy ); |
---|
| 307 | trans[ci] = (double *)remote_malloc( data_size , cxy ); |
---|
| 308 | bloup[ci] = (double *)remote_malloc( data_size , cxy ); |
---|
| 309 | umain[ci] = (double *)remote_malloc( coefs_size , cxy ); |
---|
| 310 | twid[ci] = (double *)remote_malloc( data_size , cxy ); |
---|
| 311 | } |
---|
| 312 | } |
---|
| 313 | |
---|
[635] | 314 | #if DEBUG_MAIN |
---|
| 315 | printf("\n[fft] main completes remote_malloc\n"); |
---|
| 316 | #endif |
---|
[574] | 317 | |
---|
[469] | 318 | // arrays initialisation |
---|
| 319 | InitX( data , MODE ); |
---|
| 320 | InitU( umain ); |
---|
| 321 | InitT( twid ); |
---|
| 322 | |
---|
[635] | 323 | #if DEBUG_MAIN |
---|
| 324 | printf("\n[fft] main completes arrays init\n"); |
---|
| 325 | #endif |
---|
[574] | 326 | |
---|
[469] | 327 | #if CHECK |
---|
[473] | 328 | ck1 = CheckSum(); |
---|
[469] | 329 | #endif |
---|
| 330 | |
---|
[574] | 331 | #if PRINT_ARRAY |
---|
[469] | 332 | printf("\nData values / base = %x\n", &data[0][0] ); |
---|
| 333 | PrintArray( data , N ); |
---|
| 334 | |
---|
| 335 | printf("\nTwiddle values / base = %x\n", &twid[0][0] ); |
---|
| 336 | PrintArray( twid , N ); |
---|
| 337 | |
---|
| 338 | SimpleDft( 1 , N , data , 0 , bloup , 0 ); |
---|
| 339 | |
---|
| 340 | printf("\nExpected results / base = %x\n", &bloup[0][0] ); |
---|
| 341 | PrintArray( bloup , N ); |
---|
| 342 | #endif |
---|
| 343 | |
---|
[582] | 344 | // initialise barrier |
---|
[628] | 345 | if( USE_DQT_BARRIER ) |
---|
[473] | 346 | { |
---|
[628] | 347 | barrier_attr.x_size = x_size; |
---|
| 348 | barrier_attr.y_size = y_size; |
---|
| 349 | barrier_attr.nthreads = ncores; |
---|
| 350 | error = pthread_barrier_init( &barrier, &barrier_attr , nthreads ); |
---|
| 351 | } |
---|
| 352 | else |
---|
| 353 | { |
---|
| 354 | error = pthread_barrier_init( &barrier, NULL , nthreads ); |
---|
| 355 | } |
---|
| 356 | |
---|
| 357 | if( error ) |
---|
| 358 | { |
---|
| 359 | printf("\n[fft error] cannot initialize barrier\n"); |
---|
[473] | 360 | exit( 0 ); |
---|
| 361 | } |
---|
[469] | 362 | |
---|
[635] | 363 | #if DEBUG_MAIN |
---|
| 364 | printf("\n[fft] main completes barrier init\n"); |
---|
| 365 | #endif |
---|
[473] | 366 | |
---|
[469] | 367 | // launch other threads to execute the slave() function |
---|
| 368 | // on cores other than the core running the main thread |
---|
| 369 | for (x = 0 ; x < x_size ; x++) |
---|
| 370 | { |
---|
| 371 | for (y = 0 ; y < y_size ; y++) |
---|
| 372 | { |
---|
[588] | 373 | // compute cluster identifier |
---|
| 374 | cxy = HAL_CXY_FROM_XY( x , y ); |
---|
| 375 | |
---|
[469] | 376 | for ( lid = 0 ; lid < ncores ; lid++ ) |
---|
| 377 | { |
---|
[582] | 378 | // compute thread user index (continuous index) |
---|
[469] | 379 | tid = (((x * y_size) + y) * ncores) + lid; |
---|
| 380 | |
---|
| 381 | // set thread attributes |
---|
| 382 | attr[tid].attributes = PT_ATTR_CLUSTER_DEFINED | PT_ATTR_CORE_DEFINED; |
---|
[588] | 383 | attr[tid].cxy = cxy; |
---|
[469] | 384 | attr[tid].lid = lid; |
---|
| 385 | |
---|
| 386 | // set slave function argument |
---|
[588] | 387 | args[tid].tid = tid; |
---|
| 388 | args[tid].main_tid = main_tid; |
---|
[469] | 389 | |
---|
| 390 | // create thread |
---|
| 391 | if( tid != main_tid ) |
---|
| 392 | { |
---|
| 393 | if ( pthread_create( &trdid[tid], // pointer on kernel identifier |
---|
| 394 | &attr[tid], // pointer on thread attributes |
---|
| 395 | &slave, // pointer on function |
---|
| 396 | &args[tid]) ) // pointer on function arguments |
---|
| 397 | { |
---|
[628] | 398 | printf("\n[fft error] creating thread %x\n", tid ); |
---|
[469] | 399 | exit( 0 ); |
---|
| 400 | } |
---|
[635] | 401 | |
---|
[628] | 402 | #if (DEBUG_MAIN & 1) |
---|
[588] | 403 | unsigned long long debug_cycle; |
---|
| 404 | get_cycle( &debug_cycle ); |
---|
[628] | 405 | printf("\n[fft] main created thread %d on core[%x,%d] / cycle %d\n", |
---|
[588] | 406 | tid, cxy, lid, (unsigned int)debug_cycle ); |
---|
[469] | 407 | #endif |
---|
| 408 | } |
---|
| 409 | } |
---|
| 410 | } |
---|
| 411 | } |
---|
| 412 | |
---|
[635] | 413 | #if DEBUG_MAIN |
---|
| 414 | printf("\n[fft] main completes threads creation\n"); |
---|
| 415 | #endif |
---|
[628] | 416 | |
---|
[588] | 417 | get_cycle( &end_init_cycle ); |
---|
[630] | 418 | |
---|
| 419 | // register sequencial time |
---|
[588] | 420 | init_time = (unsigned int)(end_init_cycle - start_init_cycle); |
---|
[469] | 421 | |
---|
[588] | 422 | // main itself executes the slave() function |
---|
[469] | 423 | slave( &args[main_tid] ); |
---|
| 424 | |
---|
| 425 | // wait other threads completion |
---|
| 426 | for (x = 0 ; x < x_size ; x++) |
---|
| 427 | { |
---|
| 428 | for (y = 0 ; y < y_size ; y++) |
---|
| 429 | { |
---|
| 430 | for ( lid = 0 ; lid < ncores ; lid++ ) |
---|
| 431 | { |
---|
| 432 | // compute thread continuous index |
---|
[473] | 433 | tid = (((x * y_size) + y) * ncores) + lid; |
---|
[469] | 434 | |
---|
| 435 | if( tid != main_tid ) |
---|
| 436 | { |
---|
| 437 | if( pthread_join( trdid[tid] , NULL ) ) |
---|
| 438 | { |
---|
[628] | 439 | printf("\n[fft error] in main thread joining thread %x\n", tid ); |
---|
[469] | 440 | exit( 0 ); |
---|
| 441 | } |
---|
[588] | 442 | |
---|
[628] | 443 | #if (DEBUG_MAIN & 1) |
---|
| 444 | printf("\n[fft] main thread %d joined thread %d\n", main_tid, tid ); |
---|
[588] | 445 | #endif |
---|
[469] | 446 | |
---|
| 447 | } |
---|
| 448 | } |
---|
| 449 | } |
---|
| 450 | } |
---|
| 451 | |
---|
[574] | 452 | #if PRINT_ARRAY |
---|
[469] | 453 | printf("\nData values after FFT:\n"); |
---|
| 454 | PrintArray( data , N ); |
---|
| 455 | #endif |
---|
| 456 | |
---|
| 457 | #if CHECK |
---|
[473] | 458 | ck3 = CheckSum(); |
---|
[469] | 459 | printf("\n*** Results ***\n"); |
---|
| 460 | printf("Checksum difference is %f (%f, %f)\n", ck1 - ck3, ck1, ck3); |
---|
| 461 | if (fabs(ck1 - ck3) < 0.001) printf("Results OK\n"); |
---|
| 462 | else printf("Results KO\n"); |
---|
| 463 | #endif |
---|
| 464 | |
---|
| 465 | // instrumentation |
---|
[628] | 466 | char name[64]; |
---|
| 467 | char path[128]; |
---|
[469] | 468 | char string[256]; |
---|
[628] | 469 | int ret; |
---|
[469] | 470 | |
---|
[628] | 471 | // build file name |
---|
| 472 | if( USE_DQT_BARRIER ) |
---|
[636] | 473 | snprintf( name , 64 , "fft_dqt_%d_%d_%d", N , x_size * y_size , ncores ); |
---|
[628] | 474 | else |
---|
[636] | 475 | snprintf( name , 64 , "fft_smp_%d_%d_%d", N , x_size * y_size , ncores ); |
---|
[469] | 476 | |
---|
[628] | 477 | // build pathname |
---|
| 478 | snprintf( path , 128 , "/home/%s", name ); |
---|
| 479 | |
---|
[469] | 480 | // open instrumentation file |
---|
[628] | 481 | FILE * f = fopen( path , NULL ); |
---|
| 482 | if ( f == NULL ) |
---|
| 483 | { |
---|
| 484 | printf("\n[fft error] cannot open instrumentation file <%s>\n", path ); |
---|
| 485 | exit( 0 ); |
---|
| 486 | } |
---|
| 487 | printf("\n[fft] file <%s> open\n", path ); |
---|
[469] | 488 | |
---|
[628] | 489 | // display header on terminal, and save to file |
---|
| 490 | printf("\n----- %s -----\n", name ); |
---|
[469] | 491 | |
---|
[628] | 492 | ret = fprintf( f , "\n----- %s -----\n", name ); |
---|
| 493 | if( ret < 0 ) |
---|
| 494 | { |
---|
| 495 | printf("\n[fft error] cannot write header to file <%s>\n", path ); |
---|
| 496 | exit(0); |
---|
| 497 | } |
---|
[469] | 498 | |
---|
[628] | 499 | // display results for each thread on terminal, and save to file |
---|
[588] | 500 | for (tid = 0 ; tid < nthreads ; tid++) |
---|
| 501 | { |
---|
[628] | 502 | snprintf( string , 256 , "- tid %d : Sequencial %d / Parallel %d / Barrier %d\n", |
---|
[588] | 503 | tid, init_time, parallel_time[tid], sync_time[tid] ); |
---|
| 504 | |
---|
| 505 | // display on terminal, and save to instrumentation file |
---|
| 506 | printf("%s" , string ); |
---|
[628] | 507 | fprintf( f , "%s" , string ); |
---|
| 508 | if( ret < 0 ) |
---|
| 509 | { |
---|
| 510 | printf("\n[fft error] cannot write thread %d to file <%s>\n", tid, path ); |
---|
| 511 | exit(0); |
---|
| 512 | } |
---|
[588] | 513 | } |
---|
| 514 | |
---|
[628] | 515 | // display MIN/MAX values on terminal and save to file |
---|
| 516 | unsigned int min_para = parallel_time[0]; |
---|
| 517 | unsigned int max_para = parallel_time[0]; |
---|
| 518 | unsigned int min_sync = sync_time[0]; |
---|
| 519 | unsigned int max_sync = sync_time[0]; |
---|
[588] | 520 | |
---|
[630] | 521 | for (tid = 0 ; tid < nthreads ; tid++) |
---|
[469] | 522 | { |
---|
| 523 | if (parallel_time[tid] > max_para) max_para = parallel_time[tid]; |
---|
| 524 | if (parallel_time[tid] < min_para) min_para = parallel_time[tid]; |
---|
| 525 | if (sync_time[tid] > max_sync) max_sync = sync_time[tid]; |
---|
| 526 | if (sync_time[tid] < min_sync) min_sync = sync_time[tid]; |
---|
| 527 | } |
---|
| 528 | |
---|
[628] | 529 | snprintf( string , 256 , "\n Sequencial Parallel Barrier\n" |
---|
[588] | 530 | "MIN : %d\t | %d\t | %d\t (cycles)\n" |
---|
| 531 | "MAX : %d\t | %d\t | %d\t (cycles)\n", |
---|
[469] | 532 | (int)init_time, (int)min_para, (int)min_sync, |
---|
| 533 | (int)init_time, (int)max_para, (int)max_sync ); |
---|
[628] | 534 | printf("%s", string ); |
---|
| 535 | ret = fprintf( f , "%s", string ); |
---|
| 536 | if( ret < 0 ) |
---|
| 537 | { |
---|
| 538 | printf("\n[fft error] cannot write MIN/MAX to file <%s>\n", path ); |
---|
| 539 | exit(0); |
---|
| 540 | } |
---|
[469] | 541 | |
---|
[628] | 542 | // close instrumentation file |
---|
| 543 | ret = fclose( f ); |
---|
| 544 | if( ret ) |
---|
| 545 | { |
---|
| 546 | printf("\n[fft error] cannot close file <%s>\n", path ); |
---|
| 547 | exit(0); |
---|
| 548 | } |
---|
[629] | 549 | printf("\n[fft] file <%s> closed\n", path ); |
---|
[469] | 550 | |
---|
[628] | 551 | exit( 0 ); |
---|
| 552 | |
---|
[469] | 553 | } // end main() |
---|
| 554 | |
---|
| 555 | /////////////////////////////////////////////////////////////// |
---|
| 556 | // This function is executed in parallel by all threads. |
---|
| 557 | /////////////////////////////////////////////////////////////// |
---|
[588] | 558 | void slave( args_t * args ) |
---|
[469] | 559 | { |
---|
[473] | 560 | unsigned int i; |
---|
[588] | 561 | unsigned int MyNum; // this thread index |
---|
| 562 | unsigned int MainNum; // main thread index |
---|
[473] | 563 | unsigned int MyFirst; // index first row allocated to thread |
---|
| 564 | unsigned int MyLast; // index last row allocated to thread |
---|
| 565 | double * upriv; |
---|
| 566 | unsigned int c_id; |
---|
| 567 | unsigned int c_offset; |
---|
[469] | 568 | |
---|
| 569 | unsigned long long parallel_start; |
---|
| 570 | unsigned long long parallel_stop; |
---|
| 571 | unsigned long long barrier_start; |
---|
| 572 | unsigned long long barrier_stop; |
---|
| 573 | |
---|
[588] | 574 | MyNum = args->tid; |
---|
| 575 | MainNum = args->main_tid; |
---|
[469] | 576 | |
---|
| 577 | get_cycle( ¶llel_start ); |
---|
| 578 | |
---|
[588] | 579 | #if DEBUG_SLAVE |
---|
[629] | 580 | printf("\n[fft] %s : thread %d enter / cycle %d\n", |
---|
[588] | 581 | __FUNCTION__, MyNum, (unsigned int)parallel_start ); |
---|
| 582 | #endif |
---|
| 583 | |
---|
[629] | 584 | // BARRIER |
---|
| 585 | get_cycle( &barrier_start ); |
---|
| 586 | pthread_barrier_wait( &barrier ); |
---|
| 587 | get_cycle( &barrier_stop ); |
---|
[630] | 588 | sync_time[MyNum] += (unsigned int)(barrier_stop - barrier_start); |
---|
[629] | 589 | |
---|
[630] | 590 | #if DEBUG_SLAVE |
---|
| 591 | printf("\n[@@@] %s : thread %d exit first barrier / cycle %d\n", |
---|
| 592 | __FUNCTION__, MyNum, (unsigned int)barrier_stop ); |
---|
| 593 | #endif |
---|
[629] | 594 | |
---|
[469] | 595 | // allocate and initialise local array upriv[] |
---|
| 596 | // that is a local copy of the rootN coefs defined in umain[] |
---|
| 597 | upriv = (double *)malloc(2 * (rootN - 1) * sizeof(double)); |
---|
| 598 | for ( i = 0 ; i < (rootN - 1) ; i++) |
---|
| 599 | { |
---|
| 600 | c_id = i / (rootN / nclusters); |
---|
| 601 | c_offset = i % (rootN / nclusters); |
---|
| 602 | upriv[2*i] = umain[c_id][2*c_offset]; |
---|
| 603 | upriv[2*i+1] = umain[c_id][2*c_offset+1]; |
---|
| 604 | } |
---|
| 605 | |
---|
| 606 | // compute first and last rows handled by the thread |
---|
| 607 | MyFirst = rootN * MyNum / nthreads; |
---|
| 608 | MyLast = rootN * (MyNum + 1) / nthreads; |
---|
| 609 | |
---|
| 610 | // perform forward FFT |
---|
| 611 | FFT1D( 1 , data , trans , upriv , twid , MyNum , MyFirst , MyLast ); |
---|
| 612 | |
---|
| 613 | #if CHECK |
---|
| 614 | get_cycle( &barrier_start ); |
---|
| 615 | pthread_barrier_wait( &barrier ); |
---|
| 616 | get_cycle( &barrier_stop ); |
---|
[630] | 617 | sync_time[MyNum] += (unsigned int)(barrier_stop - barrier_start); |
---|
[469] | 618 | FFT1D( -1 , data , trans , upriv , twid , MyNum , MyFirst , MyLast ); |
---|
| 619 | #endif |
---|
| 620 | |
---|
| 621 | get_cycle( ¶llel_stop ); |
---|
| 622 | |
---|
[630] | 623 | // register parallel time |
---|
| 624 | parallel_time[MyNum] = (unsigned int)(parallel_stop - parallel_start); |
---|
| 625 | |
---|
[588] | 626 | #if DEBUG_SLAVE |
---|
[630] | 627 | printf("\n[fft] %s : thread %x completes fft / p_start %d / p_stop %d\n", |
---|
| 628 | __FUNCTION__, MyNum, (unsigned int)parallel_start, (unsigned int)parallel_stop ); |
---|
| 629 | int tid; |
---|
| 630 | for (tid = 0 ; tid < nthreads ; tid++) |
---|
| 631 | { |
---|
| 632 | printf("- tid %d : Sequencial %d / Parallel %d / Barrier %d\n", |
---|
| 633 | tid , init_time, parallel_time[tid], sync_time[tid] ); |
---|
| 634 | } |
---|
[588] | 635 | #endif |
---|
[469] | 636 | |
---|
[588] | 637 | // exit only if MyNum != MainNum |
---|
| 638 | if( MyNum != MainNum ) pthread_exit( NULL ); |
---|
| 639 | |
---|
[469] | 640 | } // end slave() |
---|
| 641 | |
---|
| 642 | //////////////////////////////////////////////////////////////////////////////////////// |
---|
| 643 | // This function makes the DFT from the src[nclusters][points_per_cluster] distributed |
---|
| 644 | // buffer, to the dst[nclusters][points_per_cluster] distributed buffer. |
---|
| 645 | //////////////////////////////////////////////////////////////////////////////////////// |
---|
[473] | 646 | void SimpleDft( int direction, // 1 direct / -1 reverse |
---|
| 647 | unsigned int size, // number of points |
---|
| 648 | double ** src, // source distributed buffer |
---|
| 649 | unsigned int src_offset, // offset in source array |
---|
| 650 | double ** dst, // destination distributed buffer |
---|
| 651 | unsigned int dst_offset ) // offset in destination array |
---|
[469] | 652 | { |
---|
[473] | 653 | unsigned int n , k; |
---|
| 654 | double phi; // 2*PI*n*k/N |
---|
| 655 | double u_r; // cos( phi ) |
---|
| 656 | double u_c; // sin( phi ) |
---|
| 657 | double d_r; // Re(data[n]) |
---|
| 658 | double d_c; // Im(data[n]) |
---|
| 659 | double accu_r; // Re(accu) |
---|
| 660 | double accu_c; // Im(accu) |
---|
| 661 | unsigned int c_id; // distributed buffer cluster index |
---|
| 662 | unsigned int c_offset; // offset in distributed buffer |
---|
[469] | 663 | |
---|
| 664 | for ( k = 0 ; k < size ; k++ ) // loop on the output data points |
---|
| 665 | { |
---|
| 666 | // initialise accu |
---|
| 667 | accu_r = 0; |
---|
| 668 | accu_c = 0; |
---|
| 669 | |
---|
| 670 | for ( n = 0 ; n < size ; n++ ) // loop on the input data points |
---|
| 671 | { |
---|
| 672 | // compute coef |
---|
| 673 | phi = (double)(2*PI*n*k) / size; |
---|
| 674 | u_r = cos( phi ); |
---|
| 675 | u_c = -sin( phi ) * direction; |
---|
| 676 | |
---|
| 677 | // get input data point |
---|
| 678 | c_id = (src_offset + n) / (points_per_cluster); |
---|
| 679 | c_offset = (src_offset + n) % (points_per_cluster); |
---|
[473] | 680 | d_r = src[c_id][2*c_offset]; |
---|
| 681 | d_c = src[c_id][2*c_offset+1]; |
---|
[469] | 682 | |
---|
| 683 | // increment accu |
---|
| 684 | accu_r += ((u_r*d_r) - (u_c*d_c)); |
---|
| 685 | accu_c += ((u_r*d_c) + (u_c*d_r)); |
---|
| 686 | } |
---|
| 687 | |
---|
| 688 | // scale for inverse DFT |
---|
| 689 | if ( direction == -1 ) |
---|
| 690 | { |
---|
| 691 | accu_r /= size; |
---|
| 692 | accu_c /= size; |
---|
| 693 | } |
---|
| 694 | |
---|
| 695 | // set output data point |
---|
| 696 | c_id = (dst_offset + k) / (points_per_cluster); |
---|
| 697 | c_offset = (dst_offset + k) % (points_per_cluster); |
---|
| 698 | dst[c_id][2*c_offset] = accu_r; |
---|
| 699 | dst[c_id][2*c_offset+1] = accu_c; |
---|
| 700 | } |
---|
| 701 | |
---|
| 702 | } // end SimpleDft() |
---|
| 703 | |
---|
[582] | 704 | /////////////////////// |
---|
[503] | 705 | double CheckSum( void ) |
---|
[469] | 706 | { |
---|
[473] | 707 | unsigned int i , j; |
---|
| 708 | unsigned int c_id; |
---|
| 709 | unsigned int c_offset; |
---|
[629] | 710 | double cks; |
---|
[469] | 711 | |
---|
| 712 | cks = 0.0; |
---|
| 713 | for (j = 0; j < rootN ; j++) |
---|
| 714 | { |
---|
| 715 | for (i = 0; i < rootN ; i++) |
---|
| 716 | { |
---|
| 717 | c_id = (rootN * j + i) / (points_per_cluster); |
---|
| 718 | c_offset = (rootN * j + i) % (points_per_cluster); |
---|
| 719 | |
---|
| 720 | cks += data[c_id][2*c_offset] + data[c_id][2*c_offset+1]; |
---|
| 721 | } |
---|
| 722 | } |
---|
| 723 | return(cks); |
---|
| 724 | } |
---|
| 725 | |
---|
| 726 | |
---|
| 727 | //////////////////////////// |
---|
| 728 | void InitX(double ** x, |
---|
| 729 | unsigned int mode ) |
---|
| 730 | { |
---|
[473] | 731 | unsigned int i , j; |
---|
| 732 | unsigned int c_id; |
---|
| 733 | unsigned int c_offset; |
---|
| 734 | unsigned int index; |
---|
[469] | 735 | |
---|
| 736 | for ( j = 0 ; j < rootN ; j++ ) // loop on row index |
---|
| 737 | { |
---|
| 738 | for ( i = 0 ; i < rootN ; i++ ) // loop on point in a row |
---|
| 739 | { |
---|
| 740 | index = j * rootN + i; |
---|
| 741 | c_id = index / (points_per_cluster); |
---|
| 742 | c_offset = index % (points_per_cluster); |
---|
| 743 | |
---|
| 744 | // complex input signal is random |
---|
| 745 | if ( mode == RANDOM ) |
---|
| 746 | { |
---|
[473] | 747 | x[c_id][2*c_offset] = ( (double)rand() ) / 65536; |
---|
| 748 | x[c_id][2*c_offset+1] = ( (double)rand() ) / 65536; |
---|
[469] | 749 | } |
---|
| 750 | |
---|
| 751 | |
---|
| 752 | // complex input signal is cos(n/N) / sin(n/N) |
---|
| 753 | if ( mode == COSIN ) |
---|
| 754 | { |
---|
| 755 | double phi = (double)( 2 * PI * index) / N; |
---|
[473] | 756 | x[c_id][2*c_offset] = cos( phi ); |
---|
| 757 | x[c_id][2*c_offset+1] = sin( phi ); |
---|
[469] | 758 | } |
---|
| 759 | |
---|
| 760 | // complex input signal is constant |
---|
| 761 | if ( mode == CONSTANT ) |
---|
| 762 | { |
---|
[473] | 763 | x[c_id][2*c_offset] = 1.0; |
---|
| 764 | x[c_id][2*c_offset+1] = 0.0; |
---|
[469] | 765 | } |
---|
| 766 | } |
---|
| 767 | } |
---|
| 768 | } |
---|
| 769 | |
---|
| 770 | ///////////////////////// |
---|
| 771 | void InitU( double ** u ) |
---|
| 772 | { |
---|
[473] | 773 | unsigned int q; |
---|
| 774 | unsigned int j; |
---|
| 775 | unsigned int base; |
---|
| 776 | unsigned int n1; |
---|
| 777 | unsigned int c_id; |
---|
| 778 | unsigned int c_offset; |
---|
[469] | 779 | double phi; |
---|
[473] | 780 | unsigned int stop = 0; |
---|
[469] | 781 | |
---|
[473] | 782 | for (q = 0 ; ((unsigned int)(1 << q) < N) && (stop == 0) ; q++) |
---|
[469] | 783 | { |
---|
| 784 | n1 = 1 << q; |
---|
| 785 | base = n1 - 1; |
---|
| 786 | for (j = 0; (j < n1) && (stop == 0) ; j++) |
---|
| 787 | { |
---|
| 788 | if (base + j > rootN - 1) return; |
---|
| 789 | |
---|
| 790 | c_id = (base + j) / (rootN / nclusters); |
---|
| 791 | c_offset = (base + j) % (rootN / nclusters); |
---|
| 792 | phi = (double)(2.0 * PI * j) / (2 * n1); |
---|
| 793 | u[c_id][2*c_offset] = cos( phi ); |
---|
| 794 | u[c_id][2*c_offset+1] = -sin( phi ); |
---|
| 795 | } |
---|
| 796 | } |
---|
| 797 | } |
---|
| 798 | |
---|
| 799 | ////////////////////////// |
---|
| 800 | void InitT( double ** u ) |
---|
| 801 | { |
---|
[473] | 802 | unsigned int i, j; |
---|
| 803 | unsigned int index; |
---|
| 804 | unsigned int c_id; |
---|
| 805 | unsigned int c_offset; |
---|
[469] | 806 | double phi; |
---|
| 807 | |
---|
| 808 | for ( j = 0 ; j < rootN ; j++ ) // loop on row index |
---|
| 809 | { |
---|
| 810 | for ( i = 0 ; i < rootN ; i++ ) // loop on points in a row |
---|
| 811 | { |
---|
| 812 | index = j * rootN + i; |
---|
| 813 | c_id = index / (points_per_cluster); |
---|
| 814 | c_offset = index % (points_per_cluster); |
---|
| 815 | |
---|
| 816 | phi = (double)(2.0 * PI * i * j) / N; |
---|
| 817 | u[c_id][2*c_offset] = cos( phi ); |
---|
| 818 | u[c_id][2*c_offset+1] = -sin( phi ); |
---|
| 819 | } |
---|
| 820 | } |
---|
| 821 | } |
---|
| 822 | |
---|
| 823 | //////////////////////////////////////////////////////////////////////////////////////// |
---|
| 824 | // This function returns an index value that is the bit reverse of the input value. |
---|
| 825 | //////////////////////////////////////////////////////////////////////////////////////// |
---|
[473] | 826 | unsigned int BitReverse( unsigned int k ) |
---|
[469] | 827 | { |
---|
[473] | 828 | unsigned int i; |
---|
| 829 | unsigned int j; |
---|
| 830 | unsigned int tmp; |
---|
[469] | 831 | |
---|
| 832 | j = 0; |
---|
| 833 | tmp = k; |
---|
| 834 | for (i = 0; i < M/2 ; i++) |
---|
| 835 | { |
---|
| 836 | j = 2 * j + (tmp & 0x1); |
---|
| 837 | tmp = tmp >> 1; |
---|
| 838 | } |
---|
| 839 | return j; |
---|
| 840 | } |
---|
| 841 | |
---|
| 842 | //////////////////////////////////////////////////////////////////////////////////////// |
---|
| 843 | // This function perform the in place (direct or inverse) FFT on the N data points |
---|
| 844 | // contained in the distributed buffers x[nclusters][points_per_cluster]. |
---|
| 845 | // It handles the (N) points 1D array as a (rootN*rootN) points 2D array. |
---|
| 846 | // 1) it transpose (rootN/nthreads ) rows from x to tmp. |
---|
| 847 | // 2) it make (rootN/nthreads) FFT on the tmp rows and apply the twiddle factor. |
---|
| 848 | // 3) it transpose (rootN/nthreads) columns from tmp to x. |
---|
| 849 | // 4) it make (rootN/nthreads) FFT on the x rows. |
---|
[574] | 850 | // It calls the FFTRow() 2*(rootN/nthreads) times to perform the in place FFT |
---|
[469] | 851 | // on the rootN points contained in a row. |
---|
| 852 | //////////////////////////////////////////////////////////////////////////////////////// |
---|
[473] | 853 | void FFT1D( int direction, // direct 1 / inverse -1 |
---|
| 854 | double ** x, // input & output distributed data points array |
---|
| 855 | double ** tmp, // auxiliary distributed data points array |
---|
| 856 | double * upriv, // local array containing coefs for rootN FFT |
---|
| 857 | double ** twid, // distributed arrays containing N twiddle factors |
---|
| 858 | unsigned int MyNum, // thread continuous index |
---|
| 859 | unsigned int MyFirst, |
---|
| 860 | unsigned int MyLast ) |
---|
[469] | 861 | { |
---|
[473] | 862 | unsigned int j; |
---|
[469] | 863 | unsigned long long barrier_start; |
---|
| 864 | unsigned long long barrier_stop; |
---|
| 865 | |
---|
[574] | 866 | #if DEBUG_FFT1D |
---|
[588] | 867 | unsigned long long cycle; |
---|
| 868 | get_cycle( &cycle ); |
---|
[629] | 869 | printf("\n[fft] %s : thread %d enter / first %d / last %d / cycle %d\n", |
---|
[588] | 870 | __FUNCTION__, MyNum, MyFirst, MyLast, (unsigned int)cycle ); |
---|
[574] | 871 | #endif |
---|
| 872 | |
---|
[469] | 873 | // transpose (rootN/nthreads) rows from x to tmp |
---|
| 874 | Transpose( x , tmp , MyFirst , MyLast ); |
---|
| 875 | |
---|
[574] | 876 | #if( DEBUG_FFT1D & 1 ) |
---|
| 877 | get_cycle( &cycle ); |
---|
[629] | 878 | printf("\n[fft] %s : thread %d after first transpose / cycle %d\n", |
---|
[574] | 879 | __FUNCTION__, MyNum, (unsigned int)cycle ); |
---|
| 880 | if( PRINT_ARRAY ) PrintArray( tmp , N ); |
---|
[469] | 881 | #endif |
---|
| 882 | |
---|
| 883 | // BARRIER |
---|
| 884 | get_cycle( &barrier_start ); |
---|
| 885 | pthread_barrier_wait( &barrier ); |
---|
| 886 | get_cycle( &barrier_stop ); |
---|
[630] | 887 | sync_time[MyNum] = (unsigned int)(barrier_stop - barrier_start); |
---|
[469] | 888 | |
---|
[574] | 889 | #if( DEBUG_FFT1D & 1 ) |
---|
| 890 | get_cycle( &cycle ); |
---|
[629] | 891 | printf("\n[fft] %s : thread %d exit barrier after first transpose / cycle %d\n", |
---|
[574] | 892 | __FUNCTION__, MyNum, (unsigned int)cycle ); |
---|
| 893 | #endif |
---|
| 894 | |
---|
[469] | 895 | // do FFTs on rows of tmp (i.e. columns of x) and apply twiddle factor |
---|
| 896 | for (j = MyFirst; j < MyLast; j++) |
---|
| 897 | { |
---|
[574] | 898 | FFTRow( direction , upriv , tmp , j * rootN ); |
---|
| 899 | |
---|
[469] | 900 | TwiddleOneCol( direction , j , twid , tmp , j * rootN ); |
---|
| 901 | } |
---|
| 902 | |
---|
[574] | 903 | #if( DEBUG_FFT1D & 1 ) |
---|
[629] | 904 | printf("\n[fft] %s : thread %d after first twiddle\n", __FUNCTION__, MyNum); |
---|
[574] | 905 | if( PRINT_ARRAY ) PrintArray( tmp , N ); |
---|
[469] | 906 | #endif |
---|
| 907 | |
---|
| 908 | // BARRIER |
---|
| 909 | get_cycle( &barrier_start ); |
---|
| 910 | pthread_barrier_wait( &barrier ); |
---|
| 911 | get_cycle( &barrier_stop ); |
---|
| 912 | |
---|
[574] | 913 | #if( DEBUG_FFT1D & 1 ) |
---|
[629] | 914 | printf("\n[fft] %s : thread %d exit barrier after first twiddle\n", __FUNCTION__, MyNum); |
---|
[574] | 915 | #endif |
---|
| 916 | |
---|
[630] | 917 | sync_time[MyNum] += (unsigned int)(barrier_stop - barrier_start); |
---|
[469] | 918 | |
---|
| 919 | // transpose tmp to x |
---|
| 920 | Transpose( tmp , x , MyFirst , MyLast ); |
---|
| 921 | |
---|
[574] | 922 | #if( DEBUG_FFT1D & 1 ) |
---|
[629] | 923 | printf("\n[fft] %s : thread %d after second transpose\n", __FUNCTION__, MyNum); |
---|
[574] | 924 | if( PRINT_ARRAY ) PrintArray( x , N ); |
---|
[469] | 925 | #endif |
---|
| 926 | |
---|
| 927 | // BARRIER |
---|
| 928 | get_cycle( &barrier_start ); |
---|
| 929 | pthread_barrier_wait( &barrier ); |
---|
| 930 | get_cycle( &barrier_stop ); |
---|
| 931 | |
---|
[574] | 932 | #if( DEBUG_FFT1D & 1 ) |
---|
[629] | 933 | printf("\n[fft] %s : thread %d exit barrier after second transpose\n", __FUNCTION__, MyNum); |
---|
[574] | 934 | #endif |
---|
| 935 | |
---|
[630] | 936 | sync_time[MyNum] += (unsigned int)(barrier_stop - barrier_start); |
---|
[469] | 937 | |
---|
| 938 | // do FFTs on rows of x and apply the scaling factor |
---|
| 939 | for (j = MyFirst; j < MyLast; j++) |
---|
| 940 | { |
---|
[574] | 941 | FFTRow( direction , upriv , x , j * rootN ); |
---|
[469] | 942 | if (direction == -1) Scale( x , j * rootN ); |
---|
| 943 | } |
---|
| 944 | |
---|
[574] | 945 | #if( DEBUG_FFT1D & 1 ) |
---|
[629] | 946 | printf("\n[fft] %s : thread %d after FFT on rows\n", __FUNCTION__, MyNum); |
---|
[574] | 947 | if( PRINT_ARRAY ) PrintArray( x , N ); |
---|
[469] | 948 | #endif |
---|
| 949 | |
---|
| 950 | // BARRIER |
---|
| 951 | get_cycle( &barrier_start ); |
---|
| 952 | pthread_barrier_wait( &barrier ); |
---|
| 953 | get_cycle( &barrier_stop ); |
---|
| 954 | |
---|
[574] | 955 | #if( DEBUG_FFT1D & 1 ) |
---|
[629] | 956 | printf("\n[fft] %s : thread %d exit barrier after FFT on rows\n", __FUNCTION__, MyNum); |
---|
[574] | 957 | #endif |
---|
[630] | 958 | sync_time[MyNum] += (unsigned int)(barrier_stop - barrier_start); |
---|
[469] | 959 | |
---|
| 960 | // transpose x to tmp |
---|
| 961 | Transpose( x , tmp , MyFirst , MyLast ); |
---|
| 962 | |
---|
[574] | 963 | #if( DEBUG_FFT1D & 1 ) |
---|
[628] | 964 | printf("\n[fft] %s : thread %x after third transpose\n", __FUNCTION__, MyNum); |
---|
[574] | 965 | if( PRINT_ARRAY ) PrintArray( x , N ); |
---|
[469] | 966 | #endif |
---|
| 967 | |
---|
| 968 | // BARRIER |
---|
| 969 | get_cycle( &barrier_start ); |
---|
| 970 | pthread_barrier_wait( &barrier ); |
---|
| 971 | get_cycle( &barrier_stop ); |
---|
| 972 | |
---|
[574] | 973 | #if( DEBUG_FFT1D & 1 ) |
---|
[629] | 974 | printf("\n[fft] %s : thread %d exit barrier after third transpose\n", __FUNCTION__, MyNum); |
---|
[574] | 975 | #endif |
---|
| 976 | |
---|
[630] | 977 | sync_time[MyNum] += (unsigned int)(barrier_stop - barrier_start); |
---|
[469] | 978 | sync_time[MyNum] += (long)(barrier_stop - barrier_start); |
---|
| 979 | |
---|
| 980 | // copy tmp to x |
---|
| 981 | Copy( tmp , x , MyFirst , MyLast ); |
---|
| 982 | |
---|
| 983 | #if DEBUG_FFT1D |
---|
[629] | 984 | printf("\n[fft] %s : thread %d completed\n", __FUNCTION__, MyNum); |
---|
[574] | 985 | if( PRINT_ARRAY ) PrintArray( x , N ); |
---|
[469] | 986 | #endif |
---|
| 987 | |
---|
| 988 | } // end FFT1D() |
---|
| 989 | |
---|
| 990 | ///////////////////////////////////////////////////////////////////////////////////// |
---|
| 991 | // This function multiply all points contained in a row (rootN points) of the |
---|
| 992 | // x[] array by the corresponding twiddle factor, contained in the u[] array. |
---|
| 993 | ///////////////////////////////////////////////////////////////////////////////////// |
---|
[473] | 994 | void TwiddleOneCol( int direction, |
---|
| 995 | unsigned int j, // y coordinate in 2D view of coef array |
---|
| 996 | double ** u, // coef array base address |
---|
| 997 | double ** x, // data array base address |
---|
| 998 | unsigned int offset_x ) // first point in N points data array |
---|
[469] | 999 | { |
---|
[473] | 1000 | unsigned int i; |
---|
[469] | 1001 | double omega_r; |
---|
| 1002 | double omega_c; |
---|
| 1003 | double x_r; |
---|
| 1004 | double x_c; |
---|
[473] | 1005 | unsigned int c_id; |
---|
| 1006 | unsigned int c_offset; |
---|
[469] | 1007 | |
---|
| 1008 | for (i = 0; i < rootN ; i++) // loop on the rootN points |
---|
| 1009 | { |
---|
| 1010 | // get coef |
---|
| 1011 | c_id = (j * rootN + i) / (points_per_cluster); |
---|
| 1012 | c_offset = (j * rootN + i) % (points_per_cluster); |
---|
| 1013 | omega_r = u[c_id][2*c_offset]; |
---|
| 1014 | omega_c = direction * u[c_id][2*c_offset+1]; |
---|
| 1015 | |
---|
| 1016 | // access data |
---|
| 1017 | c_id = (offset_x + i) / (points_per_cluster); |
---|
| 1018 | c_offset = (offset_x + i) % (points_per_cluster); |
---|
| 1019 | x_r = x[c_id][2*c_offset]; |
---|
| 1020 | x_c = x[c_id][2*c_offset+1]; |
---|
| 1021 | |
---|
| 1022 | x[c_id][2*c_offset] = omega_r*x_r - omega_c * x_c; |
---|
| 1023 | x[c_id][2*c_offset+1] = omega_r*x_c + omega_c * x_r; |
---|
| 1024 | } |
---|
| 1025 | } // end TwiddleOneCol() |
---|
| 1026 | |
---|
[473] | 1027 | //////////////////////////// |
---|
| 1028 | void Scale( double ** x, // data array base address |
---|
| 1029 | unsigned int offset_x ) // first point of the row to be scaled |
---|
[469] | 1030 | { |
---|
[473] | 1031 | unsigned int i; |
---|
| 1032 | unsigned int c_id; |
---|
| 1033 | unsigned int c_offset; |
---|
[469] | 1034 | |
---|
| 1035 | for (i = 0; i < rootN ; i++) |
---|
| 1036 | { |
---|
| 1037 | c_id = (offset_x + i) / (points_per_cluster); |
---|
| 1038 | c_offset = (offset_x + i) % (points_per_cluster); |
---|
[473] | 1039 | x[c_id][2*c_offset] /= N; |
---|
| 1040 | x[c_id][2*c_offset + 1] /= N; |
---|
[469] | 1041 | } |
---|
| 1042 | } |
---|
| 1043 | |
---|
[473] | 1044 | /////////////////////////////////// |
---|
| 1045 | void Transpose( double ** src, // source buffer (array of pointers) |
---|
| 1046 | double ** dest, // destination buffer (array of pointers) |
---|
| 1047 | unsigned int MyFirst, // first row allocated to the thread |
---|
| 1048 | unsigned int MyLast ) // last row allocated to the thread |
---|
[469] | 1049 | { |
---|
[473] | 1050 | unsigned int row; // row index |
---|
| 1051 | unsigned int point; // data point index in a row |
---|
[469] | 1052 | |
---|
[473] | 1053 | unsigned int index_src; // absolute index in the source N points array |
---|
| 1054 | unsigned int c_id_src; // cluster for the source buffer |
---|
| 1055 | unsigned int c_offset_src; // offset in the source buffer |
---|
[469] | 1056 | |
---|
[473] | 1057 | unsigned int index_dst; // absolute index in the dest N points array |
---|
| 1058 | unsigned int c_id_dst; // cluster for the dest buffer |
---|
| 1059 | unsigned int c_offset_dst; // offset in the dest buffer |
---|
[469] | 1060 | |
---|
| 1061 | |
---|
| 1062 | // scan all data points allocated to the thread |
---|
| 1063 | // (between MyFirst row and MyLast row) from the source buffer |
---|
| 1064 | // and write these points to the destination buffer |
---|
| 1065 | for ( row = MyFirst ; row < MyLast ; row++ ) // loop on the rows |
---|
| 1066 | { |
---|
| 1067 | for ( point = 0 ; point < rootN ; point++ ) // loop on points in row |
---|
| 1068 | { |
---|
| 1069 | index_src = row * rootN + point; |
---|
| 1070 | c_id_src = index_src / (points_per_cluster); |
---|
| 1071 | c_offset_src = index_src % (points_per_cluster); |
---|
| 1072 | |
---|
| 1073 | index_dst = point * rootN + row; |
---|
| 1074 | c_id_dst = index_dst / (points_per_cluster); |
---|
| 1075 | c_offset_dst = index_dst % (points_per_cluster); |
---|
| 1076 | |
---|
| 1077 | dest[c_id_dst][2*c_offset_dst] = src[c_id_src][2*c_offset_src]; |
---|
| 1078 | dest[c_id_dst][2*c_offset_dst+1] = src[c_id_src][2*c_offset_src+1]; |
---|
| 1079 | } |
---|
| 1080 | } |
---|
| 1081 | } // end Transpose() |
---|
| 1082 | |
---|
[473] | 1083 | ////////////////////////////// |
---|
| 1084 | void Copy( double ** src, // source buffer (array of pointers) |
---|
| 1085 | double ** dest, // destination buffer (array of pointers) |
---|
| 1086 | unsigned int MyFirst, // first row allocated to the thread |
---|
| 1087 | unsigned int MyLast ) // last row allocated to the thread |
---|
[469] | 1088 | { |
---|
[473] | 1089 | unsigned int row; // row index |
---|
| 1090 | unsigned int point; // data point index in a row |
---|
[469] | 1091 | |
---|
[473] | 1092 | unsigned int index; // absolute index in the N points array |
---|
| 1093 | unsigned int c_id; // cluster index |
---|
| 1094 | unsigned int c_offset; // offset in local buffer |
---|
[469] | 1095 | |
---|
| 1096 | // scan all data points allocated to the thread |
---|
| 1097 | for ( row = MyFirst ; row < MyLast ; row++ ) // loop on the rows |
---|
| 1098 | { |
---|
| 1099 | for ( point = 0 ; point < rootN ; point++ ) // loop on points in row |
---|
| 1100 | { |
---|
| 1101 | index = row * rootN + point; |
---|
| 1102 | c_id = index / (points_per_cluster); |
---|
| 1103 | c_offset = index % (points_per_cluster); |
---|
| 1104 | |
---|
| 1105 | dest[c_id][2*c_offset] = src[c_id][2*c_offset]; |
---|
| 1106 | dest[c_id][2*c_offset+1] = src[c_id][2*c_offset+1]; |
---|
| 1107 | } |
---|
| 1108 | } |
---|
| 1109 | } // end Copy() |
---|
| 1110 | |
---|
[473] | 1111 | /////////////////////////////// |
---|
| 1112 | void Reverse( double ** x, |
---|
| 1113 | unsigned int offset_x ) |
---|
[469] | 1114 | { |
---|
[473] | 1115 | unsigned int j, k; |
---|
| 1116 | unsigned int c_id_j; |
---|
| 1117 | unsigned int c_offset_j; |
---|
| 1118 | unsigned int c_id_k; |
---|
| 1119 | unsigned int c_offset_k; |
---|
[469] | 1120 | |
---|
| 1121 | for (k = 0 ; k < rootN ; k++) |
---|
| 1122 | { |
---|
| 1123 | j = BitReverse( k ); |
---|
| 1124 | if (j > k) |
---|
| 1125 | { |
---|
| 1126 | c_id_j = (offset_x + j) / (points_per_cluster); |
---|
| 1127 | c_offset_j = (offset_x + j) % (points_per_cluster); |
---|
| 1128 | c_id_k = (offset_x + k) / (points_per_cluster); |
---|
| 1129 | c_offset_k = (offset_x + k) % (points_per_cluster); |
---|
| 1130 | |
---|
| 1131 | SWAP(x[c_id_j][2*c_offset_j] , x[c_id_k][2*c_offset_k]); |
---|
| 1132 | SWAP(x[c_id_j][2*c_offset_j+1], x[c_id_k][2*c_offset_k+1]); |
---|
| 1133 | } |
---|
| 1134 | } |
---|
| 1135 | } |
---|
| 1136 | |
---|
| 1137 | ///////////////////////////////////////////////////////////////////////////// |
---|
| 1138 | // This function makes the in-place FFT on all points contained in a row |
---|
| 1139 | // (i.e. rootN points) of the x[nclusters][points_per_cluster] array. |
---|
| 1140 | ///////////////////////////////////////////////////////////////////////////// |
---|
[574] | 1141 | void FFTRow( int direction, // 1 direct / -1 inverse |
---|
[473] | 1142 | double * u, // private coefs array |
---|
| 1143 | double ** x, // array of pointers on distributed buffers |
---|
| 1144 | unsigned int offset_x ) // absolute offset in the x array |
---|
[469] | 1145 | { |
---|
[473] | 1146 | unsigned int j; |
---|
| 1147 | unsigned int k; |
---|
| 1148 | unsigned int q; |
---|
| 1149 | unsigned int L; |
---|
| 1150 | unsigned int r; |
---|
| 1151 | unsigned int Lstar; |
---|
[469] | 1152 | double * u1; |
---|
| 1153 | |
---|
[473] | 1154 | unsigned int offset_x1; // index first butterfly input |
---|
| 1155 | unsigned int offset_x2; // index second butterfly output |
---|
[469] | 1156 | |
---|
[473] | 1157 | double omega_r; // real part butterfy coef |
---|
| 1158 | double omega_c; // complex part butterfly coef |
---|
[469] | 1159 | |
---|
[473] | 1160 | double tau_r; |
---|
| 1161 | double tau_c; |
---|
[469] | 1162 | |
---|
[473] | 1163 | double d1_r; // real part first butterfly input |
---|
| 1164 | double d1_c; // imag part first butterfly input |
---|
| 1165 | double d2_r; // real part second butterfly input |
---|
| 1166 | double d2_c; // imag part second butterfly input |
---|
[469] | 1167 | |
---|
[473] | 1168 | unsigned int c_id_1; // cluster index for first butterfly input |
---|
| 1169 | unsigned int c_offset_1; // offset for first butterfly input |
---|
| 1170 | unsigned int c_id_2; // cluster index for second butterfly input |
---|
| 1171 | unsigned int c_offset_2; // offset for second butterfly input |
---|
[469] | 1172 | |
---|
[574] | 1173 | #if DEBUG_ROW |
---|
[469] | 1174 | unsigned int p; |
---|
[628] | 1175 | printf("\n[fft] ROW data in / %d points / offset = %d\n", rootN , offset_x ); |
---|
[574] | 1176 | |
---|
[469] | 1177 | for ( p = 0 ; p < rootN ; p++ ) |
---|
| 1178 | { |
---|
[473] | 1179 | unsigned int index = offset_x + p; |
---|
| 1180 | unsigned int c_id = index / (points_per_cluster); |
---|
| 1181 | unsigned int c_offset = index % (points_per_cluster); |
---|
[469] | 1182 | printf("%f , %f | ", x[c_id][2*c_offset] , x[c_id][2*c_offset+1] ); |
---|
| 1183 | } |
---|
| 1184 | printf("\n"); |
---|
| 1185 | #endif |
---|
| 1186 | |
---|
| 1187 | // This makes the rootN input points reordering |
---|
| 1188 | Reverse( x , offset_x ); |
---|
| 1189 | |
---|
[574] | 1190 | #if DEBUG_ROW |
---|
[628] | 1191 | printf("\n[fft] ROW data after reverse / %d points / offset = %d\n", rootN , offset_x ); |
---|
[574] | 1192 | |
---|
[469] | 1193 | for ( p = 0 ; p < rootN ; p++ ) |
---|
| 1194 | { |
---|
[473] | 1195 | unsigned int index = offset_x + p; |
---|
| 1196 | unsigned int c_id = index / (points_per_cluster); |
---|
| 1197 | unsigned int c_offset = index % (points_per_cluster); |
---|
[469] | 1198 | printf("%f , %f | ", x[c_id][2*c_offset] , x[c_id][2*c_offset+1] ); |
---|
| 1199 | } |
---|
| 1200 | printf("\n"); |
---|
| 1201 | #endif |
---|
| 1202 | |
---|
| 1203 | // This implements the multi-stages, in place Butterfly network |
---|
| 1204 | for (q = 1; q <= M/2 ; q++) // loop on stages |
---|
| 1205 | { |
---|
| 1206 | L = 1 << q; // number of points per subset for current stage |
---|
| 1207 | r = rootN / L; // number of subsets |
---|
| 1208 | Lstar = L / 2; |
---|
| 1209 | u1 = &u[2 * (Lstar - 1)]; |
---|
| 1210 | for (k = 0; k < r; k++) // loop on the subsets |
---|
| 1211 | { |
---|
| 1212 | offset_x1 = offset_x + (k * L); // index first point |
---|
| 1213 | offset_x2 = offset_x + (k * L + Lstar); // index second point |
---|
| 1214 | |
---|
[574] | 1215 | #if (DEBUG_ROW & 1) |
---|
| 1216 | printf("\n ### q = %d / k = %d / x1 = %d / x2 = %d\n", q , k , offset_x1 , offset_x2 ); |
---|
[469] | 1217 | #endif |
---|
| 1218 | // makes all in-place butterfly(s) for subset |
---|
| 1219 | for (j = 0; j < Lstar; j++) |
---|
| 1220 | { |
---|
| 1221 | // get coef |
---|
| 1222 | omega_r = u1[2*j]; |
---|
| 1223 | omega_c = direction * u1[2*j+1]; |
---|
| 1224 | |
---|
| 1225 | // get d[x1] address and value |
---|
| 1226 | c_id_1 = (offset_x1 + j) / (points_per_cluster); |
---|
| 1227 | c_offset_1 = (offset_x1 + j) % (points_per_cluster); |
---|
| 1228 | d1_r = x[c_id_1][2*c_offset_1]; |
---|
| 1229 | d1_c = x[c_id_1][2*c_offset_1+1]; |
---|
| 1230 | |
---|
| 1231 | // get d[x2] address and value |
---|
| 1232 | c_id_2 = (offset_x2 + j) / (points_per_cluster); |
---|
| 1233 | c_offset_2 = (offset_x2 + j) % (points_per_cluster); |
---|
| 1234 | d2_r = x[c_id_2][2*c_offset_2]; |
---|
| 1235 | d2_c = x[c_id_2][2*c_offset_2+1]; |
---|
| 1236 | |
---|
[574] | 1237 | #if (DEBUG_ROW & 1) |
---|
[469] | 1238 | printf("\n ### d1_in = (%f , %f) / d2_in = (%f , %f) / coef = (%f , %f)\n", |
---|
| 1239 | d1_r , d1_c , d2_r , d2_c , omega_r , omega_c); |
---|
| 1240 | #endif |
---|
| 1241 | // tau = omega * d[x2] |
---|
| 1242 | tau_r = omega_r * d2_r - omega_c * d2_c; |
---|
| 1243 | tau_c = omega_r * d2_c + omega_c * d2_r; |
---|
| 1244 | |
---|
| 1245 | // set new value for d[x1] = d[x1] + omega * d[x2] |
---|
| 1246 | x[c_id_1][2*c_offset_1] = d1_r + tau_r; |
---|
| 1247 | x[c_id_1][2*c_offset_1+1] = d1_c + tau_c; |
---|
| 1248 | |
---|
| 1249 | // set new value for d[x2] = d[x1] - omega * d[x2] |
---|
| 1250 | x[c_id_2][2*c_offset_2] = d1_r - tau_r; |
---|
| 1251 | x[c_id_2][2*c_offset_2+1] = d1_c - tau_c; |
---|
| 1252 | |
---|
[574] | 1253 | #if (DEBUG_ROW & 1) |
---|
[469] | 1254 | printf("\n ### d1_out = (%f , %f) / d2_out = (%f , %f)\n", |
---|
| 1255 | d1_r + tau_r , d1_c + tau_c , d2_r - tau_r , d2_c - tau_c ); |
---|
| 1256 | #endif |
---|
| 1257 | } |
---|
| 1258 | } |
---|
| 1259 | } |
---|
| 1260 | |
---|
[574] | 1261 | #if DEBUG_ROW |
---|
[628] | 1262 | printf("\n[fft] ROW data out / %d points / offset = %d\n", rootN , offset_x ); |
---|
[469] | 1263 | for ( p = 0 ; p < rootN ; p++ ) |
---|
| 1264 | { |
---|
[473] | 1265 | unsigned int index = offset_x + p; |
---|
| 1266 | unsigned int c_id = index / (points_per_cluster); |
---|
| 1267 | unsigned int c_offset = index % (points_per_cluster); |
---|
[469] | 1268 | printf("%f , %f | ", x[c_id][2*c_offset] , x[c_id][2*c_offset+1] ); |
---|
| 1269 | } |
---|
| 1270 | printf("\n"); |
---|
| 1271 | #endif |
---|
| 1272 | |
---|
[574] | 1273 | } // end FFTRow() |
---|
[469] | 1274 | |
---|
[473] | 1275 | /////////////////////////////////////// |
---|
| 1276 | void PrintArray( double ** array, |
---|
| 1277 | unsigned int size ) |
---|
[469] | 1278 | { |
---|
[473] | 1279 | unsigned int i; |
---|
| 1280 | unsigned int c_id; |
---|
| 1281 | unsigned int c_offset; |
---|
[469] | 1282 | |
---|
| 1283 | // float display |
---|
| 1284 | for (i = 0; i < size ; i++) |
---|
| 1285 | { |
---|
| 1286 | c_id = i / (points_per_cluster); |
---|
| 1287 | c_offset = i % (points_per_cluster); |
---|
| 1288 | |
---|
| 1289 | printf(" %f %f |", array[c_id][2*c_offset], array[c_id][2*c_offset+1]); |
---|
| 1290 | |
---|
| 1291 | if ( (i+1) % 4 == 0) printf("\n"); |
---|
| 1292 | } |
---|
| 1293 | printf("\n"); |
---|
| 1294 | } |
---|
| 1295 | |
---|
| 1296 | |
---|
| 1297 | // Local Variables: |
---|
| 1298 | // tab-width: 4 |
---|
| 1299 | // c-basic-offset: 4 |
---|
| 1300 | // c-file-offsets:((innamespace . 0)(inline-open . 0)) |
---|
| 1301 | // indent-tabs-mode: nil |
---|
| 1302 | // End: |
---|
| 1303 | |
---|
| 1304 | // vim: filetype=cpp:expandtab:shiftwidth=4:tabstop=4:softtabstop=4 |
---|
| 1305 | |
---|