[445] | 1 | /* |
---|
| 2 | * almosmkh.c - User level ALMOS-MKH specific library implementation. |
---|
| 3 | * |
---|
[637] | 4 | * Author Alain Greiner (2016,2017,2018,2019) |
---|
[445] | 5 | * |
---|
| 6 | * Copyright (c) UPMC Sorbonne Universites |
---|
| 7 | * |
---|
| 8 | * This file is part of ALMOS-MKH. |
---|
| 9 | * |
---|
| 10 | * ALMOS-MKH is free software; you can redistribute it and/or modify it |
---|
| 11 | * under the terms of the GNU General Public License as published by |
---|
| 12 | * the Free Software Foundation; version 2.0 of the License. |
---|
| 13 | * |
---|
| 14 | * ALMOS-MKH is distributed in the hope that it will be useful, but |
---|
| 15 | * WITHOUT ANY WARRANTY; without even the implied warranty of |
---|
| 16 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
---|
| 17 | * General Public License for more details. |
---|
| 18 | * |
---|
| 19 | * You should have received a copy of the GNU General Public License |
---|
| 20 | * along with ALMOS-MKH; if not, write to the Free Software Foundation, |
---|
| 21 | * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA |
---|
| 22 | */ |
---|
| 23 | |
---|
| 24 | #include <almosmkh.h> |
---|
[444] | 25 | #include <hal_user.h> |
---|
[637] | 26 | #include <hal_macros.h> |
---|
[457] | 27 | #include <hal_shared_types.h> |
---|
[445] | 28 | #include <syscalls_numbers.h> |
---|
| 29 | #include <string.h> |
---|
[444] | 30 | #include <stdio.h> |
---|
[445] | 31 | #include <stdlib.h> |
---|
| 32 | #include <unistd.h> |
---|
| 33 | #include <mman.h> |
---|
[444] | 34 | |
---|
[637] | 35 | #define DEBUG_REMOTE_MALLOC 0 |
---|
| 36 | #define DEBUG_PTHREAD_PARALLEL 1 |
---|
[573] | 37 | |
---|
[637] | 38 | ////////////////////////////////////////////////////////////////////////////////////// |
---|
| 39 | ///////////// Non standard system calls /////////////////////////////////////// |
---|
| 40 | ////////////////////////////////////////////////////////////////////////////////////// |
---|
[444] | 41 | |
---|
| 42 | ////////////////////////// |
---|
| 43 | int fg( unsigned int pid ) |
---|
| 44 | { |
---|
| 45 | return hal_user_syscall( SYS_FG, |
---|
| 46 | (reg_t)pid, 0, 0, 0 ); |
---|
| 47 | } |
---|
| 48 | |
---|
[457] | 49 | ////////////////////////////// |
---|
| 50 | int is_fg( unsigned int pid, |
---|
| 51 | unsigned int * owner ) |
---|
| 52 | { |
---|
| 53 | return hal_user_syscall( SYS_IS_FG, |
---|
| 54 | (reg_t)pid, |
---|
| 55 | (reg_t)owner, 0, 0 ); |
---|
| 56 | } |
---|
| 57 | |
---|
[444] | 58 | ////////////////////////////////////// |
---|
| 59 | int get_config( unsigned int * x_size, |
---|
| 60 | unsigned int * y_size, |
---|
| 61 | unsigned int * ncores ) |
---|
| 62 | { |
---|
| 63 | return hal_user_syscall( SYS_GET_CONFIG, |
---|
| 64 | (reg_t)x_size, |
---|
| 65 | (reg_t)y_size, |
---|
| 66 | (reg_t)ncores, 0 ); |
---|
| 67 | } |
---|
| 68 | |
---|
[637] | 69 | //////////////////////////////////// |
---|
| 70 | int get_core_id( unsigned int * cxy, |
---|
| 71 | unsigned int * lid ) |
---|
[444] | 72 | { |
---|
[637] | 73 | return hal_user_syscall( SYS_GET_CORE_ID, |
---|
[444] | 74 | (reg_t)cxy, |
---|
| 75 | (reg_t)lid, 0, 0 ); |
---|
| 76 | } |
---|
| 77 | |
---|
[637] | 78 | ///////////////////////////////////// |
---|
| 79 | int get_nb_cores( unsigned int cxy, |
---|
| 80 | unsigned int * ncores ) |
---|
| 81 | { |
---|
| 82 | return hal_user_syscall( SYS_GET_NB_CORES, |
---|
| 83 | (reg_t)cxy, |
---|
| 84 | (reg_t)ncores, 0, 0 ); |
---|
| 85 | } |
---|
| 86 | |
---|
[444] | 87 | /////////////////////////////////////////// |
---|
[637] | 88 | int get_best_core( unsigned int base_cxy, |
---|
| 89 | unsigned int level, |
---|
| 90 | unsigned int * cxy, |
---|
| 91 | unsigned int * lid ) |
---|
| 92 | { |
---|
| 93 | return hal_user_syscall( SYS_GET_BEST_CORE, |
---|
| 94 | (reg_t)base_cxy, |
---|
| 95 | (reg_t)level, |
---|
| 96 | (reg_t)cxy, |
---|
| 97 | (reg_t)lid ); |
---|
| 98 | } |
---|
| 99 | |
---|
| 100 | /////////////////////////////////////////// |
---|
[444] | 101 | int get_cycle( unsigned long long * cycle ) |
---|
| 102 | { |
---|
| 103 | return hal_user_syscall( SYS_GET_CYCLE, |
---|
| 104 | (reg_t)cycle, 0, 0, 0 ); |
---|
| 105 | } |
---|
| 106 | |
---|
[589] | 107 | ////////////////////////////////// |
---|
| 108 | int place_fork( unsigned int cxy ) |
---|
| 109 | { |
---|
| 110 | return hal_user_syscall( SYS_PLACE_FORK, |
---|
| 111 | (reg_t)cxy, 0, 0, 0 ); |
---|
| 112 | } |
---|
| 113 | |
---|
[444] | 114 | ///////////////////////////////// |
---|
| 115 | int utls( unsigned int operation, |
---|
| 116 | unsigned int value ) |
---|
| 117 | { |
---|
| 118 | return hal_user_syscall( SYS_UTLS, |
---|
| 119 | (reg_t)operation, |
---|
| 120 | (reg_t)value, 0, 0 ); |
---|
| 121 | } |
---|
| 122 | |
---|
[580] | 123 | /////////////////////////////// |
---|
| 124 | unsigned int get_uint32( void ) |
---|
[444] | 125 | { |
---|
| 126 | unsigned int i; |
---|
[580] | 127 | int c; // ASCII character value |
---|
[444] | 128 | |
---|
| 129 | unsigned char buf[32]; |
---|
[580] | 130 | |
---|
[444] | 131 | unsigned int save = 0; |
---|
[580] | 132 | unsigned int value = 0; |
---|
[444] | 133 | unsigned int done = 0; |
---|
| 134 | unsigned int overflow = 0; |
---|
| 135 | unsigned int length = 0; |
---|
[580] | 136 | |
---|
[444] | 137 | // get characters |
---|
| 138 | while (done == 0) |
---|
| 139 | { |
---|
| 140 | // read one character |
---|
[580] | 141 | c = getchar(); |
---|
[444] | 142 | |
---|
[580] | 143 | // analyse this character |
---|
| 144 | if ( ((c > 0x2F) && (c < 0x3A)) || // 0 to 9 |
---|
| 145 | ((c > 0x40) && (c < 0x47)) || // A to F |
---|
| 146 | ((c > 0x60) && (c < 0x67)) || // a to f |
---|
| 147 | (((c == 0x58) || (c == 0x78)) && (length == 1)) ) // X or x |
---|
[444] | 148 | { |
---|
[580] | 149 | putchar( c ); // echo |
---|
| 150 | if ( c > 0x60 ) c = c - 0x20; // to upper case |
---|
| 151 | buf[length] = (unsigned char)c; |
---|
[444] | 152 | length++; |
---|
| 153 | } |
---|
[580] | 154 | else if (c == 0x0A) // LF character |
---|
[444] | 155 | { |
---|
| 156 | done = 1; |
---|
| 157 | } |
---|
[580] | 158 | else if ( (c == 0x7F) || // DEL character |
---|
| 159 | (c == 0x08) ) // BS character |
---|
[444] | 160 | { |
---|
| 161 | if ( length > 0 ) |
---|
| 162 | { |
---|
| 163 | length--; |
---|
[580] | 164 | printf("\b \b"); // BS / / BS |
---|
[444] | 165 | } |
---|
| 166 | } |
---|
[580] | 167 | else if ( c == 0 ) // EOF character |
---|
[444] | 168 | { |
---|
| 169 | return -1; |
---|
| 170 | } |
---|
| 171 | |
---|
| 172 | // test buffer overflow |
---|
| 173 | if ( length >= 32 ) |
---|
| 174 | { |
---|
| 175 | overflow = 1; |
---|
| 176 | done = 1; |
---|
| 177 | } |
---|
| 178 | } // end while characters |
---|
| 179 | |
---|
[580] | 180 | // string to int conversion with overflow detection |
---|
[444] | 181 | if ( overflow == 0 ) |
---|
| 182 | { |
---|
[580] | 183 | // test (decimal / hexa) |
---|
| 184 | if( (buf[0] == 0x30) && (buf[1] == 0x58) ) // hexadecimal input |
---|
[444] | 185 | { |
---|
[580] | 186 | for (i = 2; (i < length) && (overflow == 0) ; i++) |
---|
| 187 | { |
---|
| 188 | if( buf[i] < 0x40 ) value = (value << 4) + (buf[i] - 0x30); |
---|
| 189 | else value = (value << 4) + (buf[i] - 0x37); |
---|
| 190 | if (value < save) overflow = 1; |
---|
| 191 | save = value; |
---|
| 192 | } |
---|
[444] | 193 | } |
---|
[580] | 194 | else // decimal input |
---|
| 195 | { |
---|
| 196 | for (i = 0; (i < length) && (overflow == 0) ; i++) |
---|
| 197 | { |
---|
| 198 | value = (value * 10) + (buf[i] - 0x30); |
---|
| 199 | if (value < save) overflow = 1; |
---|
| 200 | save = value; |
---|
| 201 | } |
---|
| 202 | } |
---|
[444] | 203 | } |
---|
| 204 | |
---|
| 205 | // final evaluation |
---|
| 206 | if ( overflow == 0 ) |
---|
| 207 | { |
---|
| 208 | // return value |
---|
[580] | 209 | return value; |
---|
[444] | 210 | } |
---|
| 211 | else |
---|
| 212 | { |
---|
| 213 | // cancel all echo characters |
---|
| 214 | for (i = 0; i < length ; i++) |
---|
| 215 | { |
---|
| 216 | printf("\b \b"); // BS / / BS |
---|
| 217 | } |
---|
| 218 | |
---|
| 219 | // echo character '0' |
---|
| 220 | putchar( '0' ); |
---|
| 221 | |
---|
| 222 | // return 0 value |
---|
| 223 | return 0; |
---|
| 224 | } |
---|
[580] | 225 | } // end get_uint32() |
---|
[444] | 226 | |
---|
[626] | 227 | ////////////////////////////// |
---|
| 228 | int get_string( char * string, |
---|
| 229 | int maxlen ) |
---|
| 230 | { |
---|
| 231 | int c; |
---|
| 232 | int done = 0; |
---|
| 233 | int length = 0; |
---|
[445] | 234 | |
---|
[626] | 235 | while( done == 0 ) |
---|
| 236 | { |
---|
| 237 | // check buffer overflow |
---|
| 238 | if( length >= maxlen-1 ) |
---|
| 239 | { |
---|
| 240 | return -1; // return failure |
---|
| 241 | } |
---|
| 242 | |
---|
| 243 | // read one character |
---|
| 244 | c = getchar(); |
---|
| 245 | |
---|
| 246 | // analyse this character |
---|
| 247 | if ( (c >= 0x20) && (c < 0x7F) ) // printable character |
---|
| 248 | { |
---|
| 249 | putchar( c ); // echo |
---|
| 250 | string[length] = (char)c; // register character in string |
---|
| 251 | length++; // update length |
---|
| 252 | } |
---|
| 253 | else if (c == 0x0A) // LF character marks end of string |
---|
| 254 | { |
---|
| 255 | done = 1; |
---|
| 256 | } |
---|
| 257 | else if ( (c == 0x7F) || // DEL character |
---|
| 258 | (c == 0x08) ) // BS character |
---|
| 259 | { |
---|
| 260 | if ( length > 0 ) |
---|
| 261 | { |
---|
| 262 | length--; |
---|
| 263 | printf("\b \b"); // BS / / BS |
---|
| 264 | } |
---|
| 265 | } |
---|
| 266 | else if ( c == 0 ) // EOF character |
---|
| 267 | { |
---|
| 268 | return -1; // return failure |
---|
| 269 | } |
---|
| 270 | } |
---|
| 271 | |
---|
| 272 | // set NUL character in string and return success |
---|
| 273 | string[length] = 0; |
---|
| 274 | return 0; |
---|
| 275 | |
---|
| 276 | } // end get_string() |
---|
| 277 | |
---|
[637] | 278 | ////////////////////////////////////////////////////////////////////////////////////// |
---|
| 279 | /////////////// non standard debug functions /////////////////////////////////// |
---|
| 280 | ////////////////////////////////////////////////////////////////////////////////////// |
---|
[626] | 281 | |
---|
[445] | 282 | //////////////////////////////////// |
---|
| 283 | void display_string( char * string ) |
---|
| 284 | { |
---|
| 285 | hal_user_syscall( SYS_DISPLAY, |
---|
| 286 | DISPLAY_STRING, |
---|
| 287 | (reg_t)string, 0, 0 ); |
---|
| 288 | } |
---|
| 289 | |
---|
[580] | 290 | ///////////////////////////////////////////////////// |
---|
[445] | 291 | int display_vmm( unsigned int cxy, unsigned int pid ) |
---|
| 292 | { |
---|
| 293 | return hal_user_syscall( SYS_DISPLAY, |
---|
| 294 | DISPLAY_VMM, |
---|
[580] | 295 | (reg_t)cxy, |
---|
| 296 | (reg_t)pid, 0 ); |
---|
[445] | 297 | } |
---|
| 298 | |
---|
[597] | 299 | //////////////////////////////////// |
---|
[445] | 300 | int display_sched( unsigned int cxy, |
---|
| 301 | unsigned int lid ) |
---|
| 302 | { |
---|
| 303 | return hal_user_syscall( SYS_DISPLAY, |
---|
| 304 | DISPLAY_SCHED, |
---|
| 305 | (reg_t)cxy, |
---|
| 306 | (reg_t)lid, 0 ); |
---|
| 307 | } |
---|
| 308 | |
---|
[597] | 309 | //////////////////////////////////////////////// |
---|
[589] | 310 | int display_cluster_processes( unsigned int cxy, |
---|
| 311 | unsigned int owned ) |
---|
[445] | 312 | { |
---|
| 313 | return hal_user_syscall( SYS_DISPLAY, |
---|
| 314 | DISPLAY_CLUSTER_PROCESSES, |
---|
[589] | 315 | (reg_t)cxy, |
---|
| 316 | (reg_t)owned, 0 ); |
---|
[445] | 317 | } |
---|
| 318 | |
---|
[580] | 319 | //////////////////////////////////////// |
---|
| 320 | int display_busylocks( unsigned int pid, |
---|
| 321 | unsigned int trdid ) |
---|
| 322 | { |
---|
| 323 | return hal_user_syscall( SYS_DISPLAY, |
---|
| 324 | DISPLAY_BUSYLOCKS, |
---|
| 325 | (reg_t)pid, |
---|
| 326 | (reg_t)trdid, 0 ); |
---|
| 327 | } |
---|
| 328 | |
---|
| 329 | ///////////////////////// |
---|
[478] | 330 | int display_chdev( void ) |
---|
[445] | 331 | { |
---|
| 332 | return hal_user_syscall( SYS_DISPLAY, |
---|
| 333 | DISPLAY_CHDEV, 0, 0, 0 ); |
---|
| 334 | } |
---|
| 335 | |
---|
[573] | 336 | /////////////////////// |
---|
[478] | 337 | int display_vfs( void ) |
---|
[445] | 338 | { |
---|
| 339 | return hal_user_syscall( SYS_DISPLAY, |
---|
| 340 | DISPLAY_VFS, 0, 0, 0 ); |
---|
| 341 | } |
---|
| 342 | |
---|
| 343 | //////////////////////////////////////////////// |
---|
| 344 | int display_txt_processes( unsigned int txt_id ) |
---|
| 345 | { |
---|
| 346 | return hal_user_syscall( SYS_DISPLAY, |
---|
| 347 | DISPLAY_TXT_PROCESSES, |
---|
| 348 | (reg_t)txt_id, 0, 0 ); |
---|
| 349 | } |
---|
| 350 | |
---|
[597] | 351 | //////////////////////// |
---|
| 352 | int display_dqdt( void ) |
---|
| 353 | { |
---|
| 354 | return hal_user_syscall( SYS_DISPLAY, |
---|
| 355 | DISPLAY_DQDT, 0, 0, 0 ); |
---|
| 356 | } |
---|
| 357 | |
---|
[611] | 358 | /////////////////////////////////////// |
---|
| 359 | int display_mapper( char * path, |
---|
| 360 | unsigned int page_id, |
---|
| 361 | unsigned int nbytes) |
---|
| 362 | { |
---|
| 363 | return hal_user_syscall( SYS_DISPLAY, |
---|
| 364 | DISPLAY_MAPPER, |
---|
| 365 | (reg_t)path, |
---|
| 366 | (reg_t)page_id, |
---|
| 367 | (reg_t)nbytes ); |
---|
| 368 | } |
---|
| 369 | |
---|
[623] | 370 | /////////////////////////////////////// |
---|
| 371 | int display_barrier( unsigned int pid ) |
---|
| 372 | { |
---|
| 373 | return hal_user_syscall( SYS_DISPLAY, |
---|
| 374 | DISPLAY_BARRIER, |
---|
| 375 | (reg_t)pid, 0, 0 ); |
---|
| 376 | } |
---|
| 377 | |
---|
[626] | 378 | /////////////////////////////////////// |
---|
| 379 | int display_fat( unsigned int page_id, |
---|
| 380 | unsigned int nb_entries ) |
---|
| 381 | { |
---|
| 382 | return hal_user_syscall( SYS_DISPLAY, |
---|
| 383 | DISPLAY_FAT, |
---|
| 384 | (reg_t)page_id, |
---|
| 385 | (reg_t)nb_entries, 0 ); |
---|
| 386 | } |
---|
| 387 | |
---|
[573] | 388 | /////////////////////////////// |
---|
[445] | 389 | int trace( unsigned int active, |
---|
[457] | 390 | unsigned int cxy, |
---|
[445] | 391 | unsigned int lid ) |
---|
| 392 | { |
---|
| 393 | return hal_user_syscall( SYS_TRACE, |
---|
| 394 | (reg_t)active, |
---|
[457] | 395 | (reg_t)cxy, |
---|
[445] | 396 | (reg_t)lid, 0 ); |
---|
| 397 | } |
---|
| 398 | |
---|
[573] | 399 | ///////////////// |
---|
[478] | 400 | void idbg( void ) |
---|
[444] | 401 | { |
---|
| 402 | char cmd; |
---|
| 403 | |
---|
| 404 | while( 1 ) |
---|
| 405 | { |
---|
[626] | 406 | // display prompt |
---|
[444] | 407 | printf("\n[idbg] cmd = "); |
---|
[626] | 408 | |
---|
| 409 | // get a one character command |
---|
[444] | 410 | cmd = (char)getchar(); |
---|
| 411 | |
---|
[626] | 412 | // display all busylocks owned by thread(pid,trdid) |
---|
| 413 | if( cmd == 'b' ) |
---|
[444] | 414 | { |
---|
[626] | 415 | printf("b / pid = "); |
---|
| 416 | unsigned int pid = get_uint32(); |
---|
| 417 | printf(" / trdid = "); |
---|
| 418 | unsigned int trdid = get_uint32(); |
---|
| 419 | display_busylocks( pid , trdid ); |
---|
[444] | 420 | } |
---|
[626] | 421 | // return to calling process |
---|
| 422 | else if( cmd == 'c' ) |
---|
| 423 | { |
---|
| 424 | printf("c\n"); |
---|
| 425 | break; |
---|
| 426 | } |
---|
| 427 | // display FAT mapper(page,entries) |
---|
| 428 | else if( cmd == 'f' ) |
---|
| 429 | { |
---|
| 430 | printf("f / page = "); |
---|
| 431 | unsigned int page = get_uint32(); |
---|
| 432 | printf(" / entries = "); |
---|
| 433 | unsigned int entries = get_uint32(); |
---|
| 434 | display_fat( page , entries ); |
---|
| 435 | } |
---|
| 436 | // list all supported commands |
---|
| 437 | else if( cmd == 'h' ) |
---|
| 438 | { |
---|
| 439 | printf("h\n" |
---|
| 440 | "- b : display on TXT0 busylocks taken by thread[pid,trdid]\n" |
---|
| 441 | "- c : resume calling process execution\n" |
---|
| 442 | "- f : display on TXT0 FAT mapper[page,entries]\n" |
---|
| 443 | "- h : list of supported commands\n" |
---|
| 444 | "- m : display on TXT0 mapper[path,page,nbytes]\n" |
---|
| 445 | "- p : display on TXT0 process descriptors in cluster[cxy]\n" |
---|
| 446 | "- q : display on TXT0 DQDT state\n" |
---|
| 447 | "- s : display on TXT0 scheduler state for core[cxy,lid]\n" |
---|
| 448 | "- t : display on TXT0 process decriptors attached to TXT[tid]\n" |
---|
| 449 | "- v : display on TXT0 VMM state for process[cxy,pid]\n" |
---|
| 450 | "- x : force calling process to exit\n" |
---|
| 451 | "- y : activate/desactivate trace for core[cxy,lid]\n" |
---|
| 452 | ); |
---|
| 453 | } |
---|
| 454 | // display MAPPER(path,page,nbytes) |
---|
| 455 | else if( cmd == 'm' ) |
---|
| 456 | { |
---|
| 457 | char path[128]; |
---|
| 458 | printf("m / path = "); |
---|
| 459 | int error = get_string( path , 128 ); |
---|
| 460 | printf(" / page = "); |
---|
| 461 | unsigned int page = get_uint32(); |
---|
| 462 | printf(" / nbytes = "); |
---|
| 463 | unsigned int nbytes = get_uint32(); |
---|
| 464 | if( error == 0 ) display_mapper( path , page , nbytes ); |
---|
| 465 | } |
---|
| 466 | // display all processes in cluster(cxy) |
---|
[444] | 467 | else if( cmd == 'p' ) |
---|
| 468 | { |
---|
| 469 | printf("p / cxy = "); |
---|
[626] | 470 | unsigned int cxy = get_uint32(); |
---|
[589] | 471 | display_cluster_processes( cxy , 0 ); |
---|
[444] | 472 | } |
---|
[626] | 473 | // display DQDT |
---|
| 474 | else if( cmd == 'q' ) |
---|
| 475 | { |
---|
| 476 | printf("q\n"); |
---|
| 477 | display_dqdt(); |
---|
| 478 | } |
---|
| 479 | // display scheduler state for core(cxy,lid) |
---|
[444] | 480 | else if( cmd == 's' ) |
---|
| 481 | { |
---|
| 482 | printf("s / cxy = "); |
---|
[626] | 483 | unsigned int cxy = get_uint32(); |
---|
[444] | 484 | printf(" / lid = "); |
---|
[626] | 485 | unsigned int lid = get_uint32(); |
---|
[444] | 486 | display_sched( cxy , lid ); |
---|
| 487 | } |
---|
[626] | 488 | // display all processes attached to TXT(txt_id) |
---|
| 489 | else if( cmd == 't' ) |
---|
| 490 | { |
---|
| 491 | printf("t / txt_id = "); |
---|
| 492 | unsigned int txt_id = get_uint32(); |
---|
| 493 | display_txt_processes( txt_id ); |
---|
| 494 | } |
---|
| 495 | // display vmm state for process(cxy, pid) |
---|
[444] | 496 | else if( cmd == 'v' ) |
---|
| 497 | { |
---|
| 498 | printf("v / cxy = "); |
---|
[626] | 499 | unsigned int cxy = get_uint32(); |
---|
[580] | 500 | printf(" / pid = "); |
---|
[626] | 501 | unsigned int pid = get_uint32(); |
---|
[580] | 502 | display_vmm( cxy , pid ); |
---|
[444] | 503 | } |
---|
[626] | 504 | // force the calling process to exit |
---|
| 505 | else if( cmd == 'x' ) |
---|
[444] | 506 | { |
---|
[626] | 507 | printf("x\n"); |
---|
| 508 | exit( 0 ); |
---|
[444] | 509 | } |
---|
[626] | 510 | // activate scheduler trace for core(cxy,lid) |
---|
[444] | 511 | else if( cmd == 'y' ) |
---|
| 512 | { |
---|
| 513 | printf("y / active = "); |
---|
[626] | 514 | unsigned int active = get_uint32(); |
---|
[444] | 515 | printf(" / cxy = "); |
---|
[626] | 516 | unsigned int cxy = get_uint32(); |
---|
[444] | 517 | printf(" / lid = "); |
---|
[626] | 518 | unsigned int lid = get_uint32(); |
---|
[444] | 519 | trace( active , cxy , lid ); |
---|
| 520 | } |
---|
[626] | 521 | } // en while |
---|
[444] | 522 | } // end idbg() |
---|
| 523 | |
---|
| 524 | |
---|
[637] | 525 | ///////////////////////////////////////////////////////////////////////////////////////// |
---|
| 526 | /////////////// non standard remote_malloc //////////////////////////////////////// |
---|
| 527 | ///////////////////////////////////////////////////////////////////////////////////////// |
---|
[444] | 528 | |
---|
[445] | 529 | ///////////////////////////////////////////////////////////////////////////////////////// |
---|
| 530 | // Global variable defining the allocator array (one per cluster) |
---|
| 531 | // This array (about 16 Kbytes ) will be stored in the data segment |
---|
[637] | 532 | // of any application linked with this libray. |
---|
[445] | 533 | ///////////////////////////////////////////////////////////////////////////////////////// |
---|
| 534 | |
---|
| 535 | malloc_store_t store[MALLOC_MAX_CLUSTERS]; |
---|
| 536 | |
---|
| 537 | // Macro returning the smallest power of 2 larger or equal to size value |
---|
| 538 | |
---|
| 539 | #define GET_SIZE_INDEX(size) (size <= 0x00000001) ? 0 :\ |
---|
| 540 | (size <= 0x00000002) ? 1 :\ |
---|
| 541 | (size <= 0x00000004) ? 2 :\ |
---|
| 542 | (size <= 0x00000008) ? 3 :\ |
---|
| 543 | (size <= 0x00000010) ? 4 :\ |
---|
| 544 | (size <= 0x00000020) ? 5 :\ |
---|
| 545 | (size <= 0x00000040) ? 6 :\ |
---|
| 546 | (size <= 0x00000080) ? 7 :\ |
---|
| 547 | (size <= 0x00000100) ? 8 :\ |
---|
| 548 | (size <= 0x00000200) ? 9 :\ |
---|
| 549 | (size <= 0x00000400) ? 10 :\ |
---|
| 550 | (size <= 0x00000800) ? 11 :\ |
---|
| 551 | (size <= 0x00001000) ? 12 :\ |
---|
| 552 | (size <= 0x00002000) ? 13 :\ |
---|
| 553 | (size <= 0x00004000) ? 14 :\ |
---|
| 554 | (size <= 0x00008000) ? 15 :\ |
---|
| 555 | (size <= 0x00010000) ? 16 :\ |
---|
| 556 | (size <= 0x00020000) ? 17 :\ |
---|
| 557 | (size <= 0x00040000) ? 18 :\ |
---|
| 558 | (size <= 0x00080000) ? 19 :\ |
---|
| 559 | (size <= 0x00100000) ? 20 :\ |
---|
| 560 | (size <= 0x00200000) ? 21 :\ |
---|
| 561 | (size <= 0x00400000) ? 22 :\ |
---|
| 562 | (size <= 0x00800000) ? 23 :\ |
---|
| 563 | (size <= 0x01000000) ? 24 :\ |
---|
| 564 | (size <= 0x02000000) ? 25 :\ |
---|
| 565 | (size <= 0x04000000) ? 26 :\ |
---|
| 566 | (size <= 0x08000000) ? 27 :\ |
---|
| 567 | (size <= 0x10000000) ? 28 :\ |
---|
| 568 | (size <= 0x20000000) ? 29 :\ |
---|
| 569 | (size <= 0x40000000) ? 30 :\ |
---|
| 570 | (size <= 0x80000000) ? 31 :\ |
---|
| 571 | 32 |
---|
| 572 | |
---|
| 573 | //////////////////////////////////////////////////////////////////////////////////////////// |
---|
| 574 | // This static function display the current state of the allocator in cluster <cxy>. |
---|
| 575 | //////////////////////////////////////////////////////////////////////////////////////////// |
---|
| 576 | |
---|
[637] | 577 | #if DEBUG_REMOTE_MALLOC |
---|
[445] | 578 | static void display_free_array( unsigned int cxy ) |
---|
| 579 | { |
---|
| 580 | unsigned int next; |
---|
| 581 | unsigned int id; |
---|
| 582 | unsigned int iter; |
---|
| 583 | |
---|
| 584 | printf("\n***** store[%x] base = %x / size = %x\n", |
---|
| 585 | cxy , store[cxy].store_base, store[cxy].store_size ); |
---|
| 586 | for ( id = 0 ; id < 32 ; id++ ) |
---|
| 587 | { |
---|
| 588 | next = store[cxy].free[id]; |
---|
| 589 | printf(" - free[%d] = " , id ); |
---|
| 590 | iter = 0; |
---|
| 591 | while ( next != 0 ) |
---|
| 592 | { |
---|
| 593 | printf("%x | ", next ); |
---|
| 594 | next = (*(unsigned int*)next); |
---|
| 595 | iter++; |
---|
| 596 | } |
---|
| 597 | printf("0\n"); |
---|
| 598 | } |
---|
| 599 | } // end display_free_array() |
---|
| 600 | #endif |
---|
| 601 | |
---|
| 602 | |
---|
| 603 | ////////////////////////////////////////////////////////////////////i////////////////////// |
---|
| 604 | // This static function initialises the store in the cluster identified by the <cxy> |
---|
| 605 | // arguments. It is called by the malloc() or remote_malloc when a specific store(x,y) |
---|
| 606 | // is accessed for the first time by a remote() or remote_malloc() request. |
---|
| 607 | // It uses the mmap( MAP_REMOTE ) syscall to allocate a new vseg mapped in cluster (cxy). |
---|
| 608 | ////////////////////////////////////////////////////////////////////i////////////////////// |
---|
| 609 | // @ cxy : target cluster identifier (fixed format). |
---|
| 610 | // @ store_size : store size (bytes). |
---|
| 611 | // # return without setting the initialized field in store(cxy) if failure. |
---|
| 612 | ////////////////////////////////////////////////////////////////////i////////////////////// |
---|
| 613 | static void store_init( unsigned int cxy, |
---|
| 614 | unsigned int store_size ) |
---|
| 615 | { |
---|
| 616 | unsigned int store_base; // store base address |
---|
| 617 | unsigned int free_index; // index in free[array] |
---|
| 618 | |
---|
| 619 | unsigned int alloc_base; // alloc[] array base |
---|
| 620 | unsigned int alloc_size; // alloc[] array size |
---|
| 621 | unsigned int alloc_index; // index in alloc[array] |
---|
| 622 | |
---|
| 623 | unsigned int iter; // iterator |
---|
| 624 | |
---|
[637] | 625 | #if DEBUG_REMOTE_MALLOC |
---|
| 626 | unsigned int core_cxy; |
---|
| 627 | unsigned int core_lid; |
---|
| 628 | get_core_id( &core_cxy , &core_lid ); |
---|
| 629 | printf("\n[%s] core[%x,%d] enter for store[%x] / size = %x\n", |
---|
| 630 | __FUNCTION__, core_cxy, core_lid, cxy, store_size ); |
---|
[445] | 631 | #endif |
---|
| 632 | |
---|
| 633 | // get index in free[] array from size |
---|
| 634 | free_index = GET_SIZE_INDEX( store_size ); |
---|
| 635 | |
---|
| 636 | // check store size power of 2 |
---|
[473] | 637 | if( store_size != (unsigned int)(1<<free_index) ) |
---|
[445] | 638 | { |
---|
| 639 | printf("\n[ERROR] in %s : store[%x] size not power of 2 / size = %x\n", |
---|
| 640 | __FUNCTION__, cxy , store_size ); |
---|
| 641 | return; |
---|
| 642 | } |
---|
| 643 | |
---|
| 644 | // allocate store in virtual space |
---|
| 645 | void * vadr = mmap( NULL, // MAP_FIXED not supported |
---|
| 646 | store_size, |
---|
| 647 | PROT_READ | PROT_WRITE, |
---|
| 648 | MAP_REMOTE| MAP_SHARED, |
---|
| 649 | cxy, // fd is cluster identifier |
---|
| 650 | 0 ); // offset unused |
---|
| 651 | |
---|
| 652 | if( vadr == NULL ) |
---|
| 653 | { |
---|
| 654 | printf("\n[ERROR] in %s : cannot mmap store[%x]\n", |
---|
| 655 | __FUNCTION__, cxy ); |
---|
| 656 | return; |
---|
| 657 | } |
---|
| 658 | |
---|
| 659 | store_base = (unsigned int)vadr; |
---|
| 660 | |
---|
| 661 | // check allocated store alignment |
---|
| 662 | if( store_base % store_size ) |
---|
| 663 | { |
---|
| 664 | printf("\n[ERROR] in %s : store[%x] not aligned / base = %x / size = %x\n", |
---|
| 665 | __FUNCTION__, cxy , store_base , store_size ); |
---|
| 666 | return; |
---|
| 667 | } |
---|
| 668 | |
---|
[637] | 669 | #if DEBUG_REMOTE_MALLOC |
---|
| 670 | printf("\n[%s] core[%x,%d] created vseg %x for store[%x]\n", |
---|
| 671 | __FUNCTION__, core_cxy, core_lid, store_base, cxy ); |
---|
[445] | 672 | #endif |
---|
| 673 | |
---|
| 674 | // compute size of block containing alloc[] array |
---|
| 675 | alloc_size = store_size / MALLOC_MIN_BLOCK_SIZE; |
---|
| 676 | if ( alloc_size < MALLOC_MIN_BLOCK_SIZE) alloc_size = MALLOC_MIN_BLOCK_SIZE; |
---|
| 677 | |
---|
| 678 | // get index for the corresponding block |
---|
| 679 | alloc_index = GET_SIZE_INDEX( alloc_size ); |
---|
| 680 | |
---|
| 681 | // compute alloc[] array base address |
---|
| 682 | alloc_base = store_base + store_size - alloc_size; |
---|
| 683 | |
---|
| 684 | // reset the free[] array |
---|
| 685 | for ( iter = 0 ; iter < 32 ; iter++ ) |
---|
| 686 | { |
---|
| 687 | store[cxy].free[iter] = 0; |
---|
| 688 | } |
---|
| 689 | |
---|
| 690 | // split the store into various sizes blocks, |
---|
| 691 | // initializes the free[] array and NEXT pointers |
---|
| 692 | // base is the block base address |
---|
| 693 | unsigned int base = store_base; |
---|
| 694 | unsigned int * ptr; |
---|
| 695 | for ( iter = free_index-1 ; iter >= alloc_index ; iter-- ) |
---|
| 696 | { |
---|
| 697 | store[cxy].free[iter] = base; |
---|
| 698 | ptr = (unsigned int*)base; |
---|
| 699 | *ptr = 0; |
---|
| 700 | base = base + (1<<iter); |
---|
| 701 | } |
---|
| 702 | |
---|
| 703 | // initialize store mutex |
---|
| 704 | if( pthread_mutex_init( &store[cxy].mutex , NULL ) ) |
---|
| 705 | { |
---|
| 706 | printf("\n[ERROR] in %s : cannot initialize mutex for store[%x]\n", |
---|
| 707 | __FUNCTION__, cxy ); |
---|
| 708 | return; |
---|
| 709 | } |
---|
| 710 | |
---|
| 711 | store[cxy].cxy = cxy; |
---|
| 712 | store[cxy].store_base = store_base; |
---|
| 713 | store[cxy].store_size = store_size; |
---|
| 714 | store[cxy].alloc_size = alloc_size; |
---|
| 715 | store[cxy].alloc_base = alloc_base; |
---|
| 716 | store[cxy].initialized = MALLOC_INITIALIZED; |
---|
| 717 | |
---|
| 718 | |
---|
[637] | 719 | #if DEBUG_REMOTE_MALLOC |
---|
| 720 | printf("\n[%s] core[%x,%d] completed store[%x] initialisation\n", |
---|
| 721 | __FUNCTION__, core_cxy, core_lid, cxy ); |
---|
| 722 | #endif |
---|
[445] | 723 | |
---|
[637] | 724 | #if (DEBUG_REMOTE_MALLOC & 1) |
---|
[445] | 725 | display_free_array( cxy ); |
---|
| 726 | #endif |
---|
| 727 | |
---|
| 728 | } // end store_init() |
---|
| 729 | |
---|
| 730 | //////////////////////////////////////////////////////// |
---|
| 731 | static unsigned int split_block( malloc_store_t * store, |
---|
| 732 | unsigned int vaddr, |
---|
| 733 | unsigned int searched_index, |
---|
| 734 | unsigned int requested_index ) |
---|
| 735 | { |
---|
| 736 | // push the upper half block into free[searched_index-1] |
---|
| 737 | unsigned int* new = (unsigned int*)(vaddr + (1<<(searched_index-1))); |
---|
| 738 | *new = store->free[searched_index-1]; |
---|
| 739 | store->free[searched_index-1] = (unsigned int)new; |
---|
| 740 | |
---|
| 741 | if ( searched_index == requested_index + 1 ) // terminal case: return lower half block |
---|
| 742 | { |
---|
| 743 | return vaddr; |
---|
| 744 | } |
---|
| 745 | else // non terminal case : lower half block must be split again |
---|
| 746 | { |
---|
| 747 | return split_block( store, vaddr, searched_index-1, requested_index ); |
---|
| 748 | } |
---|
| 749 | } // end split_block() |
---|
| 750 | |
---|
| 751 | ////////////////////////////////////////////////////// |
---|
| 752 | static unsigned int get_block( malloc_store_t * store, |
---|
| 753 | unsigned int searched_index, |
---|
| 754 | unsigned int requested_index ) |
---|
| 755 | { |
---|
| 756 | // test terminal case |
---|
[473] | 757 | if ( (unsigned int)(1<<searched_index) > store->store_size ) // failure |
---|
[445] | 758 | { |
---|
| 759 | return 0; |
---|
| 760 | } |
---|
| 761 | else // search a block in free[searched_index] |
---|
| 762 | { |
---|
| 763 | unsigned int vaddr = store->free[searched_index]; |
---|
| 764 | if ( vaddr == 0 ) // block not found : search in free[searched_index+1] |
---|
| 765 | { |
---|
| 766 | return get_block( store, searched_index+1, requested_index ); |
---|
| 767 | } |
---|
| 768 | else // block found : pop it from free[searched_index] |
---|
| 769 | { |
---|
| 770 | // pop the block from free[searched_index] |
---|
| 771 | unsigned int next = *((unsigned int*)vaddr); |
---|
| 772 | store->free[searched_index] = next; |
---|
| 773 | |
---|
| 774 | // test if the block must be split |
---|
| 775 | if ( searched_index == requested_index ) // no split required |
---|
| 776 | { |
---|
| 777 | return vaddr; |
---|
| 778 | } |
---|
| 779 | else // split is required |
---|
| 780 | { |
---|
| 781 | return split_block( store, vaddr, searched_index, requested_index ); |
---|
| 782 | } |
---|
| 783 | } |
---|
| 784 | } |
---|
| 785 | } // end get_block() |
---|
| 786 | |
---|
| 787 | //////////////////////////////////////// |
---|
| 788 | void * remote_malloc( unsigned int size, |
---|
| 789 | unsigned int cxy ) |
---|
| 790 | { |
---|
[573] | 791 | int error; |
---|
[445] | 792 | |
---|
[637] | 793 | #if DEBUG_REMOTE_MALLOC |
---|
| 794 | unsigned int core_cxy; |
---|
| 795 | unsigned int core_lid; |
---|
| 796 | get_core_id( &core_cxy , &core_lid ); |
---|
| 797 | printf("\n[%s] core[%x,%d] enter for size = %x / target_cxy = %x\n", |
---|
| 798 | __FUNCTION__ , core_cxy, core_lid, size , cxy ); |
---|
[445] | 799 | #endif |
---|
| 800 | |
---|
| 801 | // check arguments |
---|
| 802 | if( size == 0 ) |
---|
| 803 | { |
---|
| 804 | printf("\n[ERROR] in %s : requested size = 0 \n", |
---|
| 805 | __FUNCTION__ ); |
---|
| 806 | return NULL; |
---|
| 807 | } |
---|
| 808 | if( cxy >= MALLOC_MAX_CLUSTERS ) |
---|
| 809 | { |
---|
| 810 | printf("\n[ERROR] in %s : illegal cluster %x\n", |
---|
| 811 | __FUNCTION__ , cxy ); |
---|
| 812 | return NULL; |
---|
| 813 | } |
---|
| 814 | |
---|
| 815 | // initializes target store if required |
---|
| 816 | if( store[cxy].initialized != MALLOC_INITIALIZED ) |
---|
| 817 | { |
---|
| 818 | store_init( cxy , MALLOC_LOCAL_STORE_SIZE ); |
---|
| 819 | |
---|
| 820 | if( store[cxy].initialized != MALLOC_INITIALIZED ) |
---|
| 821 | { |
---|
| 822 | printf("\n[ERROR] in %s : cannot allocate store in cluster %x\n", |
---|
| 823 | __FUNCTION__ , cxy ); |
---|
| 824 | return NULL; |
---|
| 825 | } |
---|
| 826 | } |
---|
| 827 | |
---|
| 828 | // normalize size |
---|
| 829 | if ( size < MALLOC_MIN_BLOCK_SIZE ) size = MALLOC_MIN_BLOCK_SIZE; |
---|
| 830 | |
---|
| 831 | // compute requested_index for the free[] array |
---|
| 832 | unsigned int requested_index = GET_SIZE_INDEX( size ); |
---|
| 833 | |
---|
| 834 | // take the lock protecting access to store[cxy] |
---|
[573] | 835 | error = pthread_mutex_lock( &store[cxy].mutex ); |
---|
[445] | 836 | |
---|
[573] | 837 | if( error ) |
---|
| 838 | { |
---|
| 839 | printf("\n[ERROR] in %s : cannot take the lock protecting store in cluster %x\n", |
---|
| 840 | __FUNCTION__ , cxy ); |
---|
| 841 | return NULL; |
---|
| 842 | } |
---|
| 843 | |
---|
[445] | 844 | // call the recursive function get_block |
---|
| 845 | unsigned int base = get_block( &store[cxy], |
---|
| 846 | requested_index, |
---|
| 847 | requested_index ); |
---|
| 848 | |
---|
| 849 | // check block found |
---|
| 850 | if (base == 0) |
---|
| 851 | { |
---|
| 852 | pthread_mutex_unlock( &store[cxy].mutex ); |
---|
| 853 | printf("\n[ERROR] in %s : no more space in cluster %x\n", |
---|
| 854 | __FUNCTION__ , cxy ); |
---|
| 855 | return NULL; |
---|
| 856 | } |
---|
| 857 | |
---|
| 858 | // compute pointer in alloc[] array |
---|
| 859 | unsigned offset = (base - store[cxy].store_base) / MALLOC_MIN_BLOCK_SIZE; |
---|
| 860 | unsigned char * ptr = (unsigned char*)(store[cxy].alloc_base + offset); |
---|
| 861 | |
---|
| 862 | // update alloc_array |
---|
| 863 | *ptr = requested_index; |
---|
| 864 | |
---|
| 865 | // release the lock |
---|
| 866 | pthread_mutex_unlock( &store[cxy].mutex ); |
---|
| 867 | |
---|
[637] | 868 | #if DEBUG_REMOTE_MALLOC |
---|
| 869 | printf("\n[%s] core[%x,%d] exit / base = %x / size = %x / from store[%x]\n", |
---|
| 870 | __FUNCTION__, core_cxy, core_lid, base , size , cxy ); |
---|
[445] | 871 | #endif |
---|
| 872 | |
---|
| 873 | return (void*) base; |
---|
| 874 | |
---|
| 875 | } // end remote_malloc() |
---|
| 876 | |
---|
| 877 | ////////////////////////////////////////// |
---|
| 878 | void * remote_calloc ( unsigned int count, |
---|
| 879 | unsigned int size, |
---|
| 880 | unsigned int cxy ) |
---|
| 881 | { |
---|
| 882 | void * ptr = remote_malloc( count * size , cxy ); |
---|
| 883 | memset( ptr , 0 , count * size ); |
---|
| 884 | return ptr; |
---|
| 885 | } |
---|
| 886 | |
---|
| 887 | ////////////////////////////////// |
---|
| 888 | void * remote_realloc( void * ptr, |
---|
| 889 | unsigned int size, |
---|
| 890 | unsigned int cxy ) |
---|
| 891 | { |
---|
| 892 | // simple allocation when (ptr == NULL) |
---|
| 893 | if( ptr == NULL ) |
---|
| 894 | { |
---|
| 895 | return remote_malloc( size , cxy ); |
---|
| 896 | } |
---|
| 897 | |
---|
| 898 | // simple free when (size == 0) |
---|
| 899 | if( size == 0 ) |
---|
| 900 | { |
---|
| 901 | remote_free( ptr , cxy ); |
---|
| 902 | return NULL; |
---|
| 903 | } |
---|
| 904 | |
---|
| 905 | // check cxy and ptr in general case |
---|
| 906 | if( cxy >= MALLOC_MAX_CLUSTERS ) |
---|
| 907 | { |
---|
| 908 | printf("\n[ERROR] in %s : illegal cluster index %x\n", |
---|
| 909 | __FUNCTION__ , cxy ); |
---|
| 910 | return NULL; |
---|
| 911 | } |
---|
| 912 | |
---|
| 913 | unsigned int base = (unsigned int)ptr; |
---|
| 914 | |
---|
| 915 | if( (base < store[cxy].store_base) || |
---|
| 916 | (base >= (store[cxy].store_base + store[cxy].store_size)) ) |
---|
| 917 | { |
---|
| 918 | printf("\n[ERROR] in %s : illegal pointer = %x\n", |
---|
| 919 | __FUNCTION__, ptr ); |
---|
| 920 | return NULL; |
---|
| 921 | } |
---|
| 922 | |
---|
| 923 | // compute index in free[] array |
---|
| 924 | int index = (base - store[cxy].store_base) / MALLOC_MIN_BLOCK_SIZE; |
---|
| 925 | |
---|
| 926 | // compute old size |
---|
[473] | 927 | char * pchar = (char *) (store[cxy].alloc_base + index); |
---|
| 928 | unsigned int old_size = (unsigned int)(1 << ((int) *pchar)); |
---|
[445] | 929 | |
---|
| 930 | // allocate a new block |
---|
| 931 | void * new_ptr = remote_malloc( size , cxy ); |
---|
| 932 | |
---|
| 933 | // save old data to new block |
---|
[473] | 934 | int min_size = (int)((size < old_size) ? size : old_size); |
---|
[445] | 935 | memcpy( new_ptr, ptr, min_size ); |
---|
| 936 | |
---|
| 937 | // release old block |
---|
| 938 | remote_free( ptr , cxy ); |
---|
| 939 | |
---|
| 940 | return new_ptr; |
---|
| 941 | |
---|
[637] | 942 | } // end remote_realloc() |
---|
| 943 | |
---|
| 944 | |
---|
[445] | 945 | ////////////////////////////////////////////////////// |
---|
| 946 | static void update_free_array( malloc_store_t * store, |
---|
| 947 | unsigned int base, |
---|
| 948 | unsigned int size_index ) |
---|
| 949 | { |
---|
| 950 | // This recursive function try to merge the released block |
---|
| 951 | // with the companion block if this companion block is free. |
---|
| 952 | // This companion has the same size, and almost the same address |
---|
| 953 | // (only one address bit is different) |
---|
| 954 | // - If the companion is not in free[size_index], |
---|
| 955 | // the released block is pushed in free[size_index]. |
---|
| 956 | // - If the companion is found, it is evicted from free[size_index] |
---|
| 957 | // and the merged bloc is pushed in the free[size_index+1]. |
---|
| 958 | |
---|
| 959 | |
---|
| 960 | // compute released block size |
---|
| 961 | unsigned int size = 1<<size_index; |
---|
| 962 | |
---|
| 963 | // compute companion block and merged block base addresses |
---|
| 964 | unsigned int companion_base; |
---|
| 965 | unsigned int merged_base; |
---|
| 966 | |
---|
| 967 | if ( (base & size) == 0 ) // the released block is aligned on (2*size) |
---|
| 968 | { |
---|
| 969 | companion_base = base + size; |
---|
| 970 | merged_base = base; |
---|
| 971 | } |
---|
| 972 | else |
---|
| 973 | { |
---|
| 974 | companion_base = base - size; |
---|
| 975 | merged_base = base - size; |
---|
| 976 | } |
---|
| 977 | |
---|
| 978 | // scan all blocks in free[size_index] |
---|
| 979 | // the iter & prev variables are actually addresses |
---|
| 980 | unsigned int found = 0; |
---|
| 981 | unsigned int iter = store->free[size_index]; |
---|
| 982 | unsigned int prev = (unsigned int)&store->free[size_index]; |
---|
| 983 | while ( iter ) |
---|
| 984 | { |
---|
| 985 | if ( iter == companion_base ) |
---|
| 986 | { |
---|
| 987 | found = 1; |
---|
| 988 | break; |
---|
| 989 | } |
---|
| 990 | prev = iter; |
---|
| 991 | iter = *(unsigned int*)iter; |
---|
| 992 | } |
---|
| 993 | |
---|
| 994 | if ( found == 0 ) // Companion not found => push in free[size_index] |
---|
| 995 | { |
---|
| 996 | *(unsigned int*)base = store->free[size_index]; |
---|
| 997 | store->free[size_index] = base; |
---|
| 998 | } |
---|
| 999 | else // Companion found : merge |
---|
| 1000 | { |
---|
| 1001 | // evict the searched block from free[size_index] |
---|
| 1002 | *(unsigned int*)prev = *(unsigned int*)iter; |
---|
| 1003 | |
---|
| 1004 | // call the update_free() function for free[size_index+1] |
---|
| 1005 | update_free_array( store, merged_base , size_index+1 ); |
---|
| 1006 | } |
---|
| 1007 | } // end update_free_array() |
---|
| 1008 | |
---|
| 1009 | //////////////////////////////////// |
---|
| 1010 | void remote_free( void * ptr, |
---|
| 1011 | unsigned int cxy ) |
---|
| 1012 | { |
---|
| 1013 | |
---|
[637] | 1014 | #if DEBUG_REMOTE_MALLOC |
---|
[445] | 1015 | printf("\n[MALLOC] %s : enter for block = %x / cxy = %x\n", |
---|
| 1016 | __FUNCTION__, ptr, cxy ); |
---|
| 1017 | #endif |
---|
| 1018 | |
---|
| 1019 | unsigned int base = (unsigned int)ptr; |
---|
| 1020 | |
---|
| 1021 | // check cxy value |
---|
| 1022 | if( cxy >= MALLOC_MAX_CLUSTERS ) |
---|
| 1023 | { |
---|
| 1024 | printf("\n[ERROR] in %s : illegal cluster index %x\n", |
---|
| 1025 | __FUNCTION__ , cxy ); |
---|
| 1026 | return; |
---|
| 1027 | } |
---|
| 1028 | |
---|
| 1029 | // check ptr value |
---|
| 1030 | if( (base < store[cxy].store_base) || |
---|
| 1031 | (base >= (store[cxy].store_base + store[cxy].store_size)) ) |
---|
| 1032 | { |
---|
| 1033 | printf("\n[ERROR] in %s : illegal pointer for released block = %x\n", |
---|
| 1034 | __FUNCTION__, ptr ); |
---|
| 1035 | return; |
---|
| 1036 | } |
---|
| 1037 | |
---|
| 1038 | // get the lock protecting store[cxy] |
---|
| 1039 | pthread_mutex_lock( &store[cxy].mutex ); |
---|
| 1040 | |
---|
| 1041 | // compute released block index in alloc[] array |
---|
| 1042 | unsigned index = (base - store[cxy].store_base ) / MALLOC_MIN_BLOCK_SIZE; |
---|
| 1043 | |
---|
| 1044 | // get the released block size_index |
---|
| 1045 | unsigned char* pchar = (unsigned char*)(store[cxy].alloc_base + index); |
---|
| 1046 | unsigned int size_index = (unsigned int)*pchar; |
---|
| 1047 | |
---|
| 1048 | // check block is allocated |
---|
| 1049 | if ( size_index == 0 ) |
---|
| 1050 | { |
---|
| 1051 | pthread_mutex_unlock( &store[cxy].mutex ); |
---|
| 1052 | printf("\n[ERROR] in %s : released block not allocated / ptr = %x\n", |
---|
| 1053 | __FUNCTION__, ptr ); |
---|
| 1054 | return; |
---|
| 1055 | } |
---|
| 1056 | |
---|
| 1057 | // check released block alignment |
---|
| 1058 | if ( base % (1 << size_index) ) |
---|
| 1059 | { |
---|
| 1060 | pthread_mutex_unlock( &store[cxy].mutex ); |
---|
| 1061 | printf("\n[ERROR] in %s : released block not aligned / ptr = %x\n", |
---|
| 1062 | __FUNCTION__, ptr ); |
---|
| 1063 | return; |
---|
| 1064 | } |
---|
| 1065 | |
---|
| 1066 | // reset the alloc[index] entry |
---|
| 1067 | *pchar = 0; |
---|
| 1068 | |
---|
| 1069 | // call the recursive function update_free_array() |
---|
| 1070 | update_free_array( &store[cxy], base, size_index ); |
---|
| 1071 | |
---|
| 1072 | // release the lock |
---|
| 1073 | pthread_mutex_unlock( &store[cxy].mutex ); |
---|
| 1074 | |
---|
[637] | 1075 | #if DEBUG_REMOTE_MALLOC |
---|
[445] | 1076 | printf("\n[MALLOC] %s : conmpletes for block = %x / cxy = %x\n", |
---|
| 1077 | __FUNCTION__, ptr, cxy ); |
---|
| 1078 | #endif |
---|
| 1079 | |
---|
| 1080 | } // end remote_free() |
---|
| 1081 | |
---|
[637] | 1082 | ///////////////////////////////////////////////////////////////////////////////////////// |
---|
| 1083 | /////////////// non standard pthread_parallel_create ////////////////////////////// |
---|
| 1084 | ///////////////////////////////////////////////////////////////////////////////////////// |
---|
| 1085 | |
---|
| 1086 | #define X_MAX 16 // max number of clusters in a row |
---|
| 1087 | #define Y_MAX 16 // max number of clusters in a column |
---|
| 1088 | #define CLUSTERS_MAX X_MAX * Y_MAX |
---|
| 1089 | #define LEVEL_MAX 5 |
---|
| 1090 | #define CORES_MAX 4 // max number of cores per cluster |
---|
| 1091 | |
---|
| 1092 | typedef struct build_args_s |
---|
| 1093 | { |
---|
| 1094 | unsigned char cxy; // this thread cluster identifier |
---|
| 1095 | unsigned char level; // this thread level in quad-tree |
---|
| 1096 | unsigned char parent_cxy; // parent thread cluster identifier |
---|
| 1097 | unsigned char root_level; // quad-tree root level |
---|
| 1098 | void * work_func; // pointer on work function pointer |
---|
| 1099 | void * work_args_array; // pointer on 2D array of pointers |
---|
| 1100 | pthread_barrier_t * parent_barriers_array; // pointer on 1D array of barriers |
---|
| 1101 | unsigned int error; // return value : 0 if success |
---|
| 1102 | } |
---|
| 1103 | build_args_t; |
---|
| 1104 | |
---|
| 1105 | ///////////////////////////////////////////////////////////////////////////////////////// |
---|
| 1106 | // Global variables used for inter-thread communications |
---|
| 1107 | ///////////////////////////////////////////////////////////////////////////////////////// |
---|
| 1108 | |
---|
| 1109 | pthread_attr_t build_attr [CLUSTERS_MAX][LEVEL_MAX]; // POSIX thread attributes |
---|
| 1110 | |
---|
| 1111 | build_args_t build_args [CLUSTERS_MAX][LEVEL_MAX]; // build function arguments |
---|
| 1112 | |
---|
| 1113 | pthread_barrier_t build_barrier[CLUSTERS_MAX][LEVEL_MAX]; // parent/child synchro |
---|
| 1114 | |
---|
| 1115 | pthread_attr_t work_attr [CLUSTERS_MAX][CORES_MAX]; // POSIX thread attributes |
---|
| 1116 | |
---|
| 1117 | ////////////////////////////////////////////////////////// |
---|
| 1118 | static void pthread_recursive_build( build_args_t * args ) |
---|
| 1119 | { |
---|
| 1120 | unsigned int trdid; // unused (required by pthread_create() |
---|
| 1121 | |
---|
| 1122 | // get arguments |
---|
| 1123 | unsigned int cxy = args->cxy; |
---|
| 1124 | unsigned int level = args->level; |
---|
| 1125 | unsigned int parent_cxy = args->parent_cxy; |
---|
| 1126 | unsigned int root_level = args->root_level; |
---|
| 1127 | void * work_func = args->work_func; |
---|
| 1128 | void * work_args_array = args->work_args_array; |
---|
| 1129 | pthread_barrier_t * parent_barriers_array = args->parent_barriers_array; |
---|
| 1130 | |
---|
| 1131 | // set error default value |
---|
| 1132 | build_args[cxy][level].error = 0; |
---|
| 1133 | |
---|
| 1134 | /////////////////////////////////////////////////////////// |
---|
| 1135 | if( level == 0 ) // children are "work" threads |
---|
| 1136 | { |
---|
| 1137 | unsigned int lid; // core local index |
---|
| 1138 | unsigned int ncores; // number of cores in a cluster |
---|
| 1139 | |
---|
| 1140 | // get number of cores per cluster |
---|
| 1141 | get_nb_cores( cxy , &ncores ); |
---|
| 1142 | |
---|
| 1143 | // kill process if no active core in cluster |
---|
| 1144 | // TODO this "if" should be replaced by an "assert" [AG] |
---|
| 1145 | if( ncores == 0 ) |
---|
| 1146 | { |
---|
| 1147 | printf("\n[PANIC] in %s : no active core in cluster %x\n", |
---|
| 1148 | __FUNCTION__ , cxy ); |
---|
| 1149 | |
---|
| 1150 | // report error to parent |
---|
| 1151 | build_args[parent_cxy][level+1].error = 1; |
---|
| 1152 | |
---|
| 1153 | // kill process |
---|
| 1154 | exit( EXIT_FAILURE ); |
---|
| 1155 | } |
---|
| 1156 | |
---|
| 1157 | // initialize the parent_barrier |
---|
| 1158 | if( pthread_barrier_init( &parent_barriers_array[cxy] , NULL , ncores + 1 ) ) |
---|
| 1159 | { |
---|
| 1160 | printf("\n[ERROR] in %s : cannot initialise barrier for build thread[%x][%d]\n", |
---|
| 1161 | __FUNCTION__ , cxy , level ); |
---|
| 1162 | |
---|
| 1163 | // report error to parent |
---|
| 1164 | build_args[parent_cxy][level+1].error = 1; |
---|
| 1165 | } |
---|
| 1166 | |
---|
| 1167 | #if DEBUG_PTHREAD_PARALLEL |
---|
| 1168 | printf("\n[%s] <build> thread[%x][%d] created barrier / %d children\n", |
---|
| 1169 | __FUNCTION__, cxy, level, ncores + 1 ); |
---|
| 1170 | #endif |
---|
| 1171 | // create (ncores) "work" threads |
---|
| 1172 | for ( lid = 0 ; lid < ncores ; lid++ ) |
---|
| 1173 | { |
---|
| 1174 | // set attributes for thread[cxy][lid] |
---|
| 1175 | work_attr[cxy][lid].attributes = PT_ATTR_DETACH | |
---|
| 1176 | PT_ATTR_CLUSTER_DEFINED | |
---|
| 1177 | PT_ATTR_CORE_DEFINED; |
---|
| 1178 | work_attr[cxy][lid].cxy = cxy; |
---|
| 1179 | work_attr[cxy][lid].lid = lid; |
---|
| 1180 | |
---|
| 1181 | // compute pointer on thread[cxy][lid] arguments |
---|
| 1182 | void * work_args = *((void **)work_args_array + (cxy * CORES_MAX) + lid); |
---|
| 1183 | |
---|
| 1184 | // create thread |
---|
| 1185 | if ( pthread_create( &trdid, // unused |
---|
| 1186 | &work_attr[cxy][lid], |
---|
| 1187 | work_func, |
---|
| 1188 | work_args ) ) |
---|
| 1189 | { |
---|
| 1190 | printf("\n[ERROR] in %s : cannot create work thread[%x,%x]\n", |
---|
| 1191 | __FUNCTION__ , cxy , lid ); |
---|
| 1192 | |
---|
| 1193 | // report error to parent |
---|
| 1194 | build_args[parent_cxy][level+1].error = 1; |
---|
| 1195 | } |
---|
| 1196 | |
---|
| 1197 | #if DEBUG_PTHREAD_PARALLEL |
---|
| 1198 | printf("\n[%s] <build> thread[%x][%d] created <work> thread[%x][%d]\n", |
---|
| 1199 | __FUNCTION__, cxy, level, cxy, lid ); |
---|
| 1200 | #endif |
---|
| 1201 | } |
---|
| 1202 | |
---|
| 1203 | // wait on barrier until "work" children threads completed |
---|
| 1204 | if( pthread_barrier_wait( &parent_barriers_array[cxy] ) ) |
---|
| 1205 | { |
---|
| 1206 | printf("\n[ERROR] in %s / first barrier for <build> thread[%x][%d]\n", |
---|
| 1207 | __FUNCTION__ , cxy , level ); |
---|
| 1208 | |
---|
| 1209 | // report error to parent |
---|
| 1210 | build_args[parent_cxy][level+1].error = 1; |
---|
| 1211 | } |
---|
| 1212 | |
---|
| 1213 | #if DEBUG_PTHREAD_PARALLEL |
---|
| 1214 | printf("\n[%s] <build> thread[%x][%d] resume after children completion\n", |
---|
| 1215 | __FUNCTION__, cxy, level ); |
---|
| 1216 | #endif |
---|
| 1217 | |
---|
| 1218 | } // end level == 0 |
---|
| 1219 | |
---|
| 1220 | //////////////////////////////////////////////////////////// |
---|
| 1221 | else // children are "build" threads |
---|
| 1222 | { |
---|
| 1223 | // the 4 children threads can be created in any core of each quarters |
---|
| 1224 | // of the parent macro-cluster |
---|
| 1225 | |
---|
| 1226 | unsigned int parent_x; // X coordinate of parent macro-cluster |
---|
| 1227 | unsigned int parent_y; // Y coordinate of parent macro-cluster |
---|
| 1228 | unsigned int child_x; // X coordinate of child macro-cluster |
---|
| 1229 | unsigned int child_y; // Y coordinate of child macro-cluster |
---|
| 1230 | unsigned int child_cxy[2][2]; // selected cluster for child thread |
---|
| 1231 | unsigned int child_lid[2][2]; // selected core index for child thread |
---|
| 1232 | int child_sts[2][2]; // -1 if error / 0 if success / +1 if not found |
---|
| 1233 | unsigned int x; // X loop index for children |
---|
| 1234 | unsigned int y; // Y loop index for children |
---|
| 1235 | |
---|
| 1236 | unsigned int nb_children = 0; |
---|
| 1237 | |
---|
| 1238 | // get parent macro-cluster mask and half-size from level |
---|
| 1239 | unsigned int mask = (1 << level) - 1; |
---|
| 1240 | unsigned int half = (level > 0) ? (1 << (level - 1)) : 0; |
---|
| 1241 | |
---|
| 1242 | // get parent macro-cluster coordinates |
---|
| 1243 | parent_x = HAL_X_FROM_CXY( cxy ) & ~mask; |
---|
| 1244 | parent_y = HAL_Y_FROM_CXY( cxy ) & ~mask; |
---|
| 1245 | |
---|
| 1246 | // get child_cxy and child_lid for up to 4 children threads : 00 / 01 / 10 / 11 |
---|
| 1247 | for (x = 0 ; x < 2 ; x++) |
---|
| 1248 | { |
---|
| 1249 | // compute child macro-cluster X coordinate |
---|
| 1250 | child_x = (x == 0) ? parent_x : (parent_x + half); |
---|
| 1251 | |
---|
| 1252 | for (y = 0 ; y < 2 ; y++) |
---|
| 1253 | { |
---|
| 1254 | // compute child macro-cluster Y coordinate |
---|
| 1255 | child_y = (y == 0) ? parent_y : (parent_y + half); |
---|
| 1256 | |
---|
| 1257 | // select the best core in macro-cluster |
---|
| 1258 | child_sts[x][y] = get_best_core( HAL_CXY_FROM_XY( child_x , child_y ), |
---|
| 1259 | level-1, |
---|
| 1260 | &child_cxy[x][y], |
---|
| 1261 | &child_lid[x][y] ); |
---|
| 1262 | |
---|
| 1263 | if( child_sts[x][y] < 0 ) // failure => report error |
---|
| 1264 | { |
---|
| 1265 | printf("\n[ERROR] in %s : illegal arguments for <build> thread[%x,%x]\n", |
---|
| 1266 | __FUNCTION__ , cxy , level ); |
---|
| 1267 | |
---|
| 1268 | // report error to parent |
---|
| 1269 | build_args[parent_cxy][level+1].error = 1; |
---|
| 1270 | } |
---|
| 1271 | else if (child_sts[x][y] > 0 ) // macro-cluster undefined => does nothing |
---|
| 1272 | { |
---|
| 1273 | } |
---|
| 1274 | else // core found |
---|
| 1275 | { |
---|
| 1276 | nb_children++; |
---|
| 1277 | } |
---|
| 1278 | } // end for y |
---|
| 1279 | } // end for x |
---|
| 1280 | |
---|
| 1281 | // kill process if no active core in cluster |
---|
| 1282 | // TODO this "if" should be replaced by an "assert" [AG] |
---|
| 1283 | if( nb_children == 0 ) |
---|
| 1284 | { |
---|
| 1285 | printf("\n[PANIC] in %s : no active core in macro cluster [%x,%d]\n", |
---|
| 1286 | __FUNCTION__ , cxy , level ); |
---|
| 1287 | |
---|
| 1288 | // report error to parent |
---|
| 1289 | build_args[parent_cxy][level+1].error = 1; |
---|
| 1290 | |
---|
| 1291 | // kill process |
---|
| 1292 | exit( EXIT_FAILURE ); |
---|
| 1293 | } |
---|
| 1294 | |
---|
| 1295 | // initialize the barrier for (nb_children + 1) |
---|
| 1296 | if( pthread_barrier_init( &build_barrier[cxy][level], NULL , nb_children + 1 ) ) |
---|
| 1297 | { |
---|
| 1298 | printf("\n[error] in %s : cannot initialise barrier for build thread[%x][%d]\n", |
---|
| 1299 | __FUNCTION__ , cxy , level ); |
---|
| 1300 | |
---|
| 1301 | // report error to parent |
---|
| 1302 | build_args[parent_cxy][level+1].error = 1; |
---|
| 1303 | } |
---|
| 1304 | |
---|
| 1305 | #if DEBUG_PTHREAD_PARALLEL |
---|
| 1306 | printf("\n[%s] <build> thread[%x][%d] created barrier / %d children\n", |
---|
| 1307 | __FUNCTION__, cxy, level, nb_children + 1 ); |
---|
| 1308 | #endif |
---|
| 1309 | // create 1 to 4 children threads |
---|
| 1310 | for (x = 0 ; x < 2 ; x++) |
---|
| 1311 | { |
---|
| 1312 | for (y = 0 ; y < 2 ; y++) |
---|
| 1313 | { |
---|
| 1314 | // thread is created only if macro-cluster is active |
---|
| 1315 | if( child_sts[x][y] == 0 ) |
---|
| 1316 | { |
---|
| 1317 | unsigned int tgt_cxy = child_cxy[x][y]; |
---|
| 1318 | unsigned int tgt_lid = child_lid[x][y]; |
---|
| 1319 | |
---|
| 1320 | // set child thread attributes |
---|
| 1321 | build_attr[tgt_cxy][level-1].attributes = PT_ATTR_DETACH | |
---|
| 1322 | PT_ATTR_CLUSTER_DEFINED | |
---|
| 1323 | PT_ATTR_CORE_DEFINED; |
---|
| 1324 | build_attr[tgt_cxy][level-1].cxy = tgt_cxy; |
---|
| 1325 | build_attr[tgt_cxy][level-1].lid = tgt_lid; |
---|
| 1326 | |
---|
| 1327 | // propagate build function arguments |
---|
| 1328 | build_args[tgt_cxy][level-1].cxy = child_cxy[x][y]; |
---|
| 1329 | build_args[tgt_cxy][level-1].level = level-1; |
---|
| 1330 | build_args[tgt_cxy][level-1].parent_cxy = cxy; |
---|
| 1331 | build_args[tgt_cxy][level-1].root_level = root_level; |
---|
| 1332 | build_args[tgt_cxy][level-1].work_func = work_func; |
---|
| 1333 | build_args[tgt_cxy][level-1].work_args_array = work_args_array; |
---|
| 1334 | build_args[tgt_cxy][level-1].parent_barriers_array = parent_barriers_array; |
---|
| 1335 | |
---|
| 1336 | // create thread |
---|
| 1337 | if( pthread_create( &trdid, |
---|
| 1338 | &build_attr[tgt_cxy][level-1], |
---|
| 1339 | &pthread_recursive_build, |
---|
| 1340 | &build_args[tgt_cxy][level-1] ) ) |
---|
| 1341 | { |
---|
| 1342 | printf("\n[ERROR] in %s : cannot create build thread[%x][%d]\n", |
---|
| 1343 | __FUNCTION__ , child_cxy , level -1 ); |
---|
| 1344 | |
---|
| 1345 | // report error to parent |
---|
| 1346 | build_args[parent_cxy][level+1].error = 1; |
---|
| 1347 | } |
---|
| 1348 | |
---|
| 1349 | #if DEBUG_PTHREAD_PARALLEL |
---|
| 1350 | printf("\n[%s] <build> thread[%x][%d] created <build> thread[%x][%d] on core[%x,%d]\n", |
---|
| 1351 | __FUNCTION__, cxy, level, tgt_cxy, level - 1, tgt_cxy, tgt_lid ); |
---|
| 1352 | #endif |
---|
| 1353 | } //end if sts[x][y] |
---|
| 1354 | } // end for y |
---|
| 1355 | } // end for x |
---|
| 1356 | |
---|
| 1357 | // wait on barrier until "build" children threads completed |
---|
| 1358 | if( pthread_barrier_wait( &build_barrier[cxy][level] ) ) |
---|
| 1359 | { |
---|
| 1360 | printf("\n[ERROR] in %s / first barrier for <build> thread[%x][%d]\n", |
---|
| 1361 | __FUNCTION__ , cxy , level ); |
---|
| 1362 | |
---|
| 1363 | // report error to parent |
---|
| 1364 | build_args[parent_cxy][level+1].error = 1; |
---|
| 1365 | } |
---|
| 1366 | |
---|
| 1367 | #if DEBUG_PTHREAD_PARALLEL |
---|
| 1368 | printf("\n[%s] <build> thread[%x][%d] resume after children completion\n", |
---|
| 1369 | __FUNCTION__, cxy, level ); |
---|
| 1370 | #endif |
---|
| 1371 | |
---|
| 1372 | } // end level > 0 |
---|
| 1373 | |
---|
| 1374 | // report error to parent when required |
---|
| 1375 | if( build_args[cxy][level].error ) |
---|
| 1376 | { |
---|
| 1377 | build_args[parent_cxy][level+1].error = 1; |
---|
| 1378 | } |
---|
| 1379 | |
---|
| 1380 | // all <build> threads - but the root - |
---|
| 1381 | // signal completion to parent thread and exit |
---|
| 1382 | if( level < root_level ) |
---|
| 1383 | { |
---|
| 1384 | if( pthread_barrier_wait( &build_barrier[parent_cxy][level+1] ) ) |
---|
| 1385 | { |
---|
| 1386 | printf("\n[ERROR] in %s / second barrier for <build> thread[%x][%d]\n", |
---|
| 1387 | __FUNCTION__ , cxy , level ); |
---|
| 1388 | |
---|
| 1389 | // report error to parent |
---|
| 1390 | build_args[parent_cxy][level+1].error = 1; |
---|
| 1391 | } |
---|
| 1392 | |
---|
| 1393 | #if DEBUG_PTHREAD_PARALLEL |
---|
| 1394 | printf("\n[%s] <build> thread[%x][%d] exit\n", |
---|
| 1395 | __FUNCTION__, cxy , level ); |
---|
| 1396 | #endif |
---|
| 1397 | // "build" thread exit |
---|
| 1398 | pthread_exit( NULL ); |
---|
| 1399 | } |
---|
| 1400 | } // end pthread_recursive_build() |
---|
| 1401 | |
---|
| 1402 | /////////////////////////////////////////////////////// |
---|
| 1403 | int pthread_parallel_create( unsigned int root_level, |
---|
| 1404 | void * work_func, |
---|
| 1405 | void * work_args_array, |
---|
| 1406 | void * parent_barriers_array ) |
---|
| 1407 | { |
---|
| 1408 | unsigned int root_cxy; |
---|
| 1409 | unsigned int root_lid; // unused, but required by get_core_id() |
---|
| 1410 | |
---|
| 1411 | #if DEBUG_PTHREAD_PARALLEL |
---|
| 1412 | printf("\n[%s] enter / root_level %d / func %x / args %x / barriers %x\n", |
---|
| 1413 | __FUNCTION__, root_level, work_func, work_args_array, parent_barriers_array ); |
---|
| 1414 | #endif |
---|
| 1415 | |
---|
| 1416 | // get calling thread cluster |
---|
| 1417 | get_core_id( &root_cxy , &root_lid ); |
---|
| 1418 | |
---|
| 1419 | // set the build function arguments for the root <build> thread |
---|
| 1420 | build_args[root_cxy][root_level].cxy = root_cxy; |
---|
| 1421 | build_args[root_cxy][root_level].level = root_level; |
---|
| 1422 | build_args[root_cxy][root_level].root_level = root_level; |
---|
| 1423 | build_args[root_cxy][root_level].work_func = work_func; |
---|
| 1424 | build_args[root_cxy][root_level].work_args_array = work_args_array; |
---|
| 1425 | build_args[root_cxy][root_level].parent_barriers_array = parent_barriers_array; |
---|
| 1426 | |
---|
| 1427 | // call the recursive build function |
---|
| 1428 | pthread_recursive_build( &build_args[root_cxy][root_level] ); |
---|
| 1429 | |
---|
| 1430 | // check error |
---|
| 1431 | if( build_args[root_cxy][root_level].error ) |
---|
| 1432 | { |
---|
| 1433 | printf("\n[error] in %s\n", __FUNCTION__ ); |
---|
| 1434 | return -1; |
---|
| 1435 | } |
---|
| 1436 | |
---|
| 1437 | return 0; |
---|
| 1438 | |
---|
| 1439 | } // end pthread_parallel_create() |
---|
| 1440 | |
---|
| 1441 | |
---|
| 1442 | |
---|
[445] | 1443 | // Local Variables: |
---|
| 1444 | // tab-width: 4 |
---|
| 1445 | // c-basic-offset: 4 |
---|
| 1446 | // c-file-offsets:((innamespace . 0)(inline-open . 0)) |
---|
| 1447 | // indent-tabs-mode: nil |
---|
| 1448 | // End: |
---|
| 1449 | // vim: filetype=c:expandtab:shiftwidth=4:tabstop=4:softtabstop=4 |
---|
| 1450 | |
---|
| 1451 | |
---|
| 1452 | |
---|