| 1 | /* Default profiling support. |
| 2 | Copyright (C) 1996-2015 Free Software Foundation, Inc. |
| 3 | Contributed by Cygnus Support. |
| 4 | |
| 5 | This file is part of GDB, the GNU debugger. |
| 6 | |
| 7 | This program is free software; you can redistribute it and/or modify |
| 8 | it under the terms of the GNU General Public License as published by |
| 9 | the Free Software Foundation; either version 3 of the License, or |
| 10 | (at your option) any later version. |
| 11 | |
| 12 | This program is distributed in the hope that it will be useful, |
| 13 | but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 14 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
| 15 | GNU General Public License for more details. |
| 16 | |
| 17 | You should have received a copy of the GNU General Public License |
| 18 | along with this program. If not, see <http://www.gnu.org/licenses/>. */ |
| 19 | |
| 20 | #include "sim-main.h" |
| 21 | #include "sim-io.h" |
| 22 | #include "sim-options.h" |
| 23 | #include "sim-assert.h" |
| 24 | |
| 25 | #ifdef HAVE_STDLIB_H |
| 26 | #include <stdlib.h> |
| 27 | #endif |
| 28 | |
| 29 | #ifdef HAVE_STRING_H |
| 30 | #include <string.h> |
| 31 | #else |
| 32 | #ifdef HAVE_STRINGS_H |
| 33 | #include <strings.h> |
| 34 | #endif |
| 35 | #endif |
| 36 | #include <ctype.h> |
| 37 | |
| 38 | #if !WITH_PROFILE_PC_P |
| 39 | static unsigned int _profile_stub; |
| 40 | # define PROFILE_PC_FREQ(p) _profile_stub |
| 41 | # define PROFILE_PC_NR_BUCKETS(p) _profile_stub |
| 42 | # define PROFILE_PC_SHIFT(p) _profile_stub |
| 43 | # define PROFILE_PC_START(p) _profile_stub |
| 44 | # define PROFILE_PC_END(p) _profile_stub |
| 45 | # define PROFILE_INSN_COUNT(p) &_profile_stub |
| 46 | #endif |
| 47 | |
| 48 | #define COMMAS(n) sim_add_commas (comma_buf, sizeof (comma_buf), (n)) |
| 49 | |
| 50 | static MODULE_INIT_FN profile_init; |
| 51 | static MODULE_UNINSTALL_FN profile_uninstall; |
| 52 | |
| 53 | static DECLARE_OPTION_HANDLER (profile_option_handler); |
| 54 | |
| 55 | enum { |
| 56 | OPTION_PROFILE_INSN = OPTION_START, |
| 57 | OPTION_PROFILE_MEMORY, |
| 58 | OPTION_PROFILE_MODEL, |
| 59 | OPTION_PROFILE_FILE, |
| 60 | OPTION_PROFILE_CORE, |
| 61 | OPTION_PROFILE_CPU_FREQUENCY, |
| 62 | OPTION_PROFILE_PC, |
| 63 | OPTION_PROFILE_PC_RANGE, |
| 64 | OPTION_PROFILE_PC_GRANULARITY, |
| 65 | OPTION_PROFILE_RANGE, |
| 66 | OPTION_PROFILE_FUNCTION |
| 67 | }; |
| 68 | |
| 69 | static const OPTION profile_options[] = { |
| 70 | { {"profile", optional_argument, NULL, 'p'}, |
| 71 | 'p', "on|off", "Perform profiling", |
| 72 | profile_option_handler, NULL }, |
| 73 | { {"profile-insn", optional_argument, NULL, OPTION_PROFILE_INSN}, |
| 74 | '\0', "on|off", "Perform instruction profiling", |
| 75 | profile_option_handler, NULL }, |
| 76 | { {"profile-memory", optional_argument, NULL, OPTION_PROFILE_MEMORY}, |
| 77 | '\0', "on|off", "Perform memory profiling", |
| 78 | profile_option_handler, NULL }, |
| 79 | { {"profile-core", optional_argument, NULL, OPTION_PROFILE_CORE}, |
| 80 | '\0', "on|off", "Perform CORE profiling", |
| 81 | profile_option_handler, NULL }, |
| 82 | { {"profile-model", optional_argument, NULL, OPTION_PROFILE_MODEL}, |
| 83 | '\0', "on|off", "Perform model profiling", |
| 84 | profile_option_handler, NULL }, |
| 85 | { {"profile-cpu-frequency", required_argument, NULL, |
| 86 | OPTION_PROFILE_CPU_FREQUENCY}, |
| 87 | '\0', "CPU FREQUENCY", "Specify the speed of the simulated cpu clock", |
| 88 | profile_option_handler, NULL }, |
| 89 | |
| 90 | { {"profile-file", required_argument, NULL, OPTION_PROFILE_FILE}, |
| 91 | '\0', "FILE NAME", "Specify profile output file", |
| 92 | profile_option_handler, NULL }, |
| 93 | |
| 94 | { {"profile-pc", optional_argument, NULL, OPTION_PROFILE_PC}, |
| 95 | '\0', "on|off", "Perform PC profiling", |
| 96 | profile_option_handler, NULL }, |
| 97 | { {"profile-pc-frequency", required_argument, NULL, 'F'}, |
| 98 | 'F', "PC PROFILE FREQUENCY", "Specified PC profiling frequency", |
| 99 | profile_option_handler, NULL }, |
| 100 | { {"profile-pc-size", required_argument, NULL, 'S'}, |
| 101 | 'S', "PC PROFILE SIZE", "Specify PC profiling size", |
| 102 | profile_option_handler, NULL }, |
| 103 | { {"profile-pc-granularity", required_argument, NULL, OPTION_PROFILE_PC_GRANULARITY}, |
| 104 | '\0', "PC PROFILE GRANULARITY", "Specify PC profiling sample coverage", |
| 105 | profile_option_handler, NULL }, |
| 106 | { {"profile-pc-range", required_argument, NULL, OPTION_PROFILE_PC_RANGE}, |
| 107 | '\0', "BASE,BOUND", "Specify PC profiling address range", |
| 108 | profile_option_handler, NULL }, |
| 109 | |
| 110 | #ifdef SIM_HAVE_ADDR_RANGE |
| 111 | { {"profile-range", required_argument, NULL, OPTION_PROFILE_RANGE}, |
| 112 | '\0', "START,END", "Specify range of addresses for instruction and model profiling", |
| 113 | profile_option_handler, NULL }, |
| 114 | #if 0 /*wip*/ |
| 115 | { {"profile-function", required_argument, NULL, OPTION_PROFILE_FUNCTION}, |
| 116 | '\0', "FUNCTION", "Specify function to profile", |
| 117 | profile_option_handler, NULL }, |
| 118 | #endif |
| 119 | #endif |
| 120 | |
| 121 | { {NULL, no_argument, NULL, 0}, '\0', NULL, NULL, NULL, NULL } |
| 122 | }; |
| 123 | |
| 124 | /* Set/reset the profile options indicated in MASK. */ |
| 125 | |
| 126 | SIM_RC |
| 127 | set_profile_option_mask (SIM_DESC sd, const char *name, int mask, const char *arg) |
| 128 | { |
| 129 | int profile_nr; |
| 130 | int cpu_nr; |
| 131 | int profile_val = 1; |
| 132 | |
| 133 | if (arg != NULL) |
| 134 | { |
| 135 | if (strcmp (arg, "yes") == 0 |
| 136 | || strcmp (arg, "on") == 0 |
| 137 | || strcmp (arg, "1") == 0) |
| 138 | profile_val = 1; |
| 139 | else if (strcmp (arg, "no") == 0 |
| 140 | || strcmp (arg, "off") == 0 |
| 141 | || strcmp (arg, "0") == 0) |
| 142 | profile_val = 0; |
| 143 | else |
| 144 | { |
| 145 | sim_io_eprintf (sd, "Argument `%s' for `--profile%s' invalid, one of `on', `off', `yes', `no' expected\n", arg, name); |
| 146 | return SIM_RC_FAIL; |
| 147 | } |
| 148 | } |
| 149 | |
| 150 | /* update applicable profile bits */ |
| 151 | for (profile_nr = 0; profile_nr < MAX_PROFILE_VALUES; ++profile_nr) |
| 152 | { |
| 153 | if ((mask & (1 << profile_nr)) == 0) |
| 154 | continue; |
| 155 | |
| 156 | #if 0 /* see sim-trace.c, set flags in STATE here if/when there are any */ |
| 157 | /* Set non-cpu specific values. */ |
| 158 | switch (profile_nr) |
| 159 | { |
| 160 | case ??? : |
| 161 | break; |
| 162 | } |
| 163 | #endif |
| 164 | |
| 165 | /* Set cpu values. */ |
| 166 | for (cpu_nr = 0; cpu_nr < MAX_NR_PROCESSORS; cpu_nr++) |
| 167 | { |
| 168 | CPU_PROFILE_FLAGS (STATE_CPU (sd, cpu_nr))[profile_nr] = profile_val; |
| 169 | } |
| 170 | } |
| 171 | |
| 172 | /* Re-compute the cpu profile summary. */ |
| 173 | if (profile_val) |
| 174 | { |
| 175 | for (cpu_nr = 0; cpu_nr < MAX_NR_PROCESSORS; cpu_nr++) |
| 176 | CPU_PROFILE_DATA (STATE_CPU (sd, cpu_nr))->profile_any_p = 1; |
| 177 | } |
| 178 | else |
| 179 | { |
| 180 | for (cpu_nr = 0; cpu_nr < MAX_NR_PROCESSORS; cpu_nr++) |
| 181 | { |
| 182 | CPU_PROFILE_DATA (STATE_CPU (sd, cpu_nr))->profile_any_p = 0; |
| 183 | for (profile_nr = 0; profile_nr < MAX_PROFILE_VALUES; ++profile_nr) |
| 184 | { |
| 185 | if (CPU_PROFILE_FLAGS (STATE_CPU (sd, cpu_nr))[profile_nr]) |
| 186 | { |
| 187 | CPU_PROFILE_DATA (STATE_CPU (sd, cpu_nr))->profile_any_p = 1; |
| 188 | break; |
| 189 | } |
| 190 | } |
| 191 | } |
| 192 | } |
| 193 | |
| 194 | return SIM_RC_OK; |
| 195 | } |
| 196 | |
| 197 | /* Set one profile option based on its IDX value. |
| 198 | Not static as cgen-scache.c uses it. */ |
| 199 | |
| 200 | SIM_RC |
| 201 | sim_profile_set_option (SIM_DESC sd, const char *name, int idx, const char *arg) |
| 202 | { |
| 203 | return set_profile_option_mask (sd, name, 1 << idx, arg); |
| 204 | } |
| 205 | |
| 206 | static SIM_RC |
| 207 | parse_frequency (SIM_DESC sd, const char *arg, unsigned long *freq) |
| 208 | { |
| 209 | const char *ch; |
| 210 | /* First, parse a decimal number. */ |
| 211 | *freq = 0; |
| 212 | ch = arg; |
| 213 | if (isdigit (*arg)) |
| 214 | { |
| 215 | for (/**/; *ch != '\0'; ++ch) |
| 216 | { |
| 217 | if (! isdigit (*ch)) |
| 218 | break; |
| 219 | *freq = *freq * 10 + (*ch - '0'); |
| 220 | } |
| 221 | |
| 222 | /* Accept KHz, MHz or Hz as a suffix. */ |
| 223 | if (tolower (*ch) == 'm') |
| 224 | { |
| 225 | *freq *= 1000000; |
| 226 | ++ch; |
| 227 | } |
| 228 | else if (tolower (*ch) == 'k') |
| 229 | { |
| 230 | *freq *= 1000; |
| 231 | ++ch; |
| 232 | } |
| 233 | |
| 234 | if (tolower (*ch) == 'h') |
| 235 | { |
| 236 | ++ch; |
| 237 | if (tolower (*ch) == 'z') |
| 238 | ++ch; |
| 239 | } |
| 240 | } |
| 241 | |
| 242 | if (*ch != '\0') |
| 243 | { |
| 244 | sim_io_eprintf (sd, "Invalid argument for --profile-cpu-frequency: %s\n", |
| 245 | arg); |
| 246 | *freq = 0; |
| 247 | return SIM_RC_FAIL; |
| 248 | } |
| 249 | |
| 250 | return SIM_RC_OK; |
| 251 | } |
| 252 | |
| 253 | static SIM_RC |
| 254 | profile_option_handler (SIM_DESC sd, |
| 255 | sim_cpu *cpu, |
| 256 | int opt, |
| 257 | char *arg, |
| 258 | int is_command) |
| 259 | { |
| 260 | int cpu_nr; |
| 261 | |
| 262 | /* FIXME: Need to handle `cpu' arg. */ |
| 263 | |
| 264 | switch (opt) |
| 265 | { |
| 266 | case 'p' : |
| 267 | if (! WITH_PROFILE) |
| 268 | sim_io_eprintf (sd, "Profiling not compiled in, `-p' ignored\n"); |
| 269 | else |
| 270 | return set_profile_option_mask (sd, "profile", PROFILE_USEFUL_MASK, |
| 271 | arg); |
| 272 | break; |
| 273 | |
| 274 | case OPTION_PROFILE_INSN : |
| 275 | if (WITH_PROFILE_INSN_P) |
| 276 | return sim_profile_set_option (sd, "-insn", PROFILE_INSN_IDX, arg); |
| 277 | else |
| 278 | sim_io_eprintf (sd, "Instruction profiling not compiled in, `--profile-insn' ignored\n"); |
| 279 | break; |
| 280 | |
| 281 | case OPTION_PROFILE_MEMORY : |
| 282 | if (WITH_PROFILE_MEMORY_P) |
| 283 | return sim_profile_set_option (sd, "-memory", PROFILE_MEMORY_IDX, arg); |
| 284 | else |
| 285 | sim_io_eprintf (sd, "Memory profiling not compiled in, `--profile-memory' ignored\n"); |
| 286 | break; |
| 287 | |
| 288 | case OPTION_PROFILE_CORE : |
| 289 | if (WITH_PROFILE_CORE_P) |
| 290 | return sim_profile_set_option (sd, "-core", PROFILE_CORE_IDX, arg); |
| 291 | else |
| 292 | sim_io_eprintf (sd, "CORE profiling not compiled in, `--profile-core' ignored\n"); |
| 293 | break; |
| 294 | |
| 295 | case OPTION_PROFILE_MODEL : |
| 296 | if (WITH_PROFILE_MODEL_P) |
| 297 | return sim_profile_set_option (sd, "-model", PROFILE_MODEL_IDX, arg); |
| 298 | else |
| 299 | sim_io_eprintf (sd, "Model profiling not compiled in, `--profile-model' ignored\n"); |
| 300 | break; |
| 301 | |
| 302 | case OPTION_PROFILE_CPU_FREQUENCY : |
| 303 | { |
| 304 | unsigned long val; |
| 305 | SIM_RC rc = parse_frequency (sd, arg, &val); |
| 306 | if (rc == SIM_RC_OK) |
| 307 | { |
| 308 | for (cpu_nr = 0; cpu_nr < MAX_NR_PROCESSORS; ++cpu_nr) |
| 309 | PROFILE_CPU_FREQ (CPU_PROFILE_DATA (STATE_CPU (sd,cpu_nr))) = val; |
| 310 | } |
| 311 | return rc; |
| 312 | } |
| 313 | |
| 314 | case OPTION_PROFILE_FILE : |
| 315 | /* FIXME: Might want this to apply to pc profiling only, |
| 316 | or have two profile file options. */ |
| 317 | if (! WITH_PROFILE) |
| 318 | sim_io_eprintf (sd, "Profiling not compiled in, `--profile-file' ignored\n"); |
| 319 | else |
| 320 | { |
| 321 | FILE *f = fopen (arg, "w"); |
| 322 | |
| 323 | if (f == NULL) |
| 324 | { |
| 325 | sim_io_eprintf (sd, "Unable to open profile output file `%s'\n", arg); |
| 326 | return SIM_RC_FAIL; |
| 327 | } |
| 328 | for (cpu_nr = 0; cpu_nr < MAX_NR_PROCESSORS; ++cpu_nr) |
| 329 | PROFILE_FILE (CPU_PROFILE_DATA (STATE_CPU (sd, cpu_nr))) = f; |
| 330 | } |
| 331 | break; |
| 332 | |
| 333 | case OPTION_PROFILE_PC: |
| 334 | if (WITH_PROFILE_PC_P) |
| 335 | return sim_profile_set_option (sd, "-pc", PROFILE_PC_IDX, arg); |
| 336 | else |
| 337 | sim_io_eprintf (sd, "PC profiling not compiled in, `--profile-pc' ignored\n"); |
| 338 | break; |
| 339 | |
| 340 | case 'F' : |
| 341 | if (WITH_PROFILE_PC_P) |
| 342 | { |
| 343 | /* FIXME: Validate arg. */ |
| 344 | int val = atoi (arg); |
| 345 | for (cpu_nr = 0; cpu_nr < MAX_NR_PROCESSORS; ++cpu_nr) |
| 346 | PROFILE_PC_FREQ (CPU_PROFILE_DATA (STATE_CPU (sd, cpu_nr))) = val; |
| 347 | for (cpu_nr = 0; cpu_nr < MAX_NR_PROCESSORS; ++cpu_nr) |
| 348 | CPU_PROFILE_FLAGS (STATE_CPU (sd, cpu_nr))[PROFILE_PC_IDX] = 1; |
| 349 | } |
| 350 | else |
| 351 | sim_io_eprintf (sd, "PC profiling not compiled in, `--profile-pc-frequency' ignored\n"); |
| 352 | break; |
| 353 | |
| 354 | case 'S' : |
| 355 | if (WITH_PROFILE_PC_P) |
| 356 | { |
| 357 | /* FIXME: Validate arg. */ |
| 358 | int val = atoi (arg); |
| 359 | for (cpu_nr = 0; cpu_nr < MAX_NR_PROCESSORS; ++cpu_nr) |
| 360 | PROFILE_PC_NR_BUCKETS (CPU_PROFILE_DATA (STATE_CPU (sd, cpu_nr))) = val; |
| 361 | for (cpu_nr = 0; cpu_nr < MAX_NR_PROCESSORS; ++cpu_nr) |
| 362 | CPU_PROFILE_FLAGS (STATE_CPU (sd, cpu_nr))[PROFILE_PC_IDX] = 1; |
| 363 | } |
| 364 | else |
| 365 | sim_io_eprintf (sd, "PC profiling not compiled in, `--profile-pc-size' ignored\n"); |
| 366 | break; |
| 367 | |
| 368 | case OPTION_PROFILE_PC_GRANULARITY: |
| 369 | if (WITH_PROFILE_PC_P) |
| 370 | { |
| 371 | int shift; |
| 372 | int val = atoi (arg); |
| 373 | /* check that the granularity is a power of two */ |
| 374 | shift = 0; |
| 375 | while (val > (1 << shift)) |
| 376 | { |
| 377 | shift += 1; |
| 378 | } |
| 379 | if (val != (1 << shift)) |
| 380 | { |
| 381 | sim_io_eprintf (sd, "PC profiling granularity not a power of two\n"); |
| 382 | return SIM_RC_FAIL; |
| 383 | } |
| 384 | if (shift == 0) |
| 385 | { |
| 386 | sim_io_eprintf (sd, "PC profiling granularity too small"); |
| 387 | return SIM_RC_FAIL; |
| 388 | } |
| 389 | for (cpu_nr = 0; cpu_nr < MAX_NR_PROCESSORS; ++cpu_nr) |
| 390 | PROFILE_PC_SHIFT (CPU_PROFILE_DATA (STATE_CPU (sd, cpu_nr))) = shift; |
| 391 | for (cpu_nr = 0; cpu_nr < MAX_NR_PROCESSORS; ++cpu_nr) |
| 392 | CPU_PROFILE_FLAGS (STATE_CPU (sd, cpu_nr))[PROFILE_PC_IDX] = 1; |
| 393 | } |
| 394 | else |
| 395 | sim_io_eprintf (sd, "PC profiling not compiled in, `--profile-pc-granularity' ignored\n"); |
| 396 | break; |
| 397 | |
| 398 | case OPTION_PROFILE_PC_RANGE: |
| 399 | if (WITH_PROFILE_PC_P) |
| 400 | { |
| 401 | /* FIXME: Validate args */ |
| 402 | char *chp = arg; |
| 403 | unsigned long base; |
| 404 | unsigned long bound; |
| 405 | base = strtoul (chp, &chp, 0); |
| 406 | if (*chp != ',') |
| 407 | { |
| 408 | sim_io_eprintf (sd, "--profile-pc-range missing BOUND argument\n"); |
| 409 | return SIM_RC_FAIL; |
| 410 | } |
| 411 | bound = strtoul (chp + 1, NULL, 0); |
| 412 | for (cpu_nr = 0; cpu_nr < MAX_NR_PROCESSORS; ++cpu_nr) |
| 413 | { |
| 414 | PROFILE_PC_START (CPU_PROFILE_DATA (STATE_CPU (sd, cpu_nr))) = base; |
| 415 | PROFILE_PC_END (CPU_PROFILE_DATA (STATE_CPU (sd, cpu_nr))) = bound; |
| 416 | } |
| 417 | for (cpu_nr = 0; cpu_nr < MAX_NR_PROCESSORS; ++cpu_nr) |
| 418 | CPU_PROFILE_FLAGS (STATE_CPU (sd, cpu_nr))[PROFILE_PC_IDX] = 1; |
| 419 | } |
| 420 | else |
| 421 | sim_io_eprintf (sd, "PC profiling not compiled in, `--profile-pc-range' ignored\n"); |
| 422 | break; |
| 423 | |
| 424 | #ifdef SIM_HAVE_ADDR_RANGE |
| 425 | case OPTION_PROFILE_RANGE : |
| 426 | if (WITH_PROFILE) |
| 427 | { |
| 428 | char *chp = arg; |
| 429 | unsigned long start,end; |
| 430 | start = strtoul (chp, &chp, 0); |
| 431 | if (*chp != ',') |
| 432 | { |
| 433 | sim_io_eprintf (sd, "--profile-range missing END argument\n"); |
| 434 | return SIM_RC_FAIL; |
| 435 | } |
| 436 | end = strtoul (chp + 1, NULL, 0); |
| 437 | /* FIXME: Argument validation. */ |
| 438 | if (cpu != NULL) |
| 439 | sim_addr_range_add (PROFILE_RANGE (CPU_PROFILE_DATA (cpu)), |
| 440 | start, end); |
| 441 | else |
| 442 | for (cpu_nr = 0; cpu_nr < MAX_NR_PROCESSORS; ++cpu_nr) |
| 443 | sim_addr_range_add (PROFILE_RANGE (CPU_PROFILE_DATA (STATE_CPU (sd, cpu_nr))), |
| 444 | start, end); |
| 445 | } |
| 446 | else |
| 447 | sim_io_eprintf (sd, "Profiling not compiled in, `--profile-range' ignored\n"); |
| 448 | break; |
| 449 | |
| 450 | case OPTION_PROFILE_FUNCTION : |
| 451 | if (WITH_PROFILE) |
| 452 | { |
| 453 | /*wip: need to compute function range given name*/ |
| 454 | } |
| 455 | else |
| 456 | sim_io_eprintf (sd, "Profiling not compiled in, `--profile-function' ignored\n"); |
| 457 | break; |
| 458 | #endif /* SIM_HAVE_ADDR_RANGE */ |
| 459 | } |
| 460 | |
| 461 | return SIM_RC_OK; |
| 462 | } |
| 463 | \f |
| 464 | /* Profiling output hooks. */ |
| 465 | |
| 466 | static void |
| 467 | profile_vprintf (SIM_DESC sd, sim_cpu *cpu, const char *fmt, va_list ap) |
| 468 | { |
| 469 | FILE *fp = PROFILE_FILE (CPU_PROFILE_DATA (cpu)); |
| 470 | |
| 471 | /* If an output file was given, redirect output to that. */ |
| 472 | if (fp != NULL) |
| 473 | vfprintf (fp, fmt, ap); |
| 474 | else |
| 475 | sim_io_evprintf (sd, fmt, ap); |
| 476 | } |
| 477 | |
| 478 | __attribute__ ((format (printf, 3, 4))) |
| 479 | static void |
| 480 | profile_printf (SIM_DESC sd, sim_cpu *cpu, const char *fmt, ...) |
| 481 | { |
| 482 | va_list ap; |
| 483 | |
| 484 | va_start (ap, fmt); |
| 485 | profile_vprintf (sd, cpu, fmt, ap); |
| 486 | va_end (ap); |
| 487 | } |
| 488 | \f |
| 489 | /* PC profiling support */ |
| 490 | |
| 491 | #if WITH_PROFILE_PC_P |
| 492 | |
| 493 | static void |
| 494 | profile_pc_cleanup (SIM_DESC sd) |
| 495 | { |
| 496 | int n; |
| 497 | for (n = 0; n < MAX_NR_PROCESSORS; n++) |
| 498 | { |
| 499 | sim_cpu *cpu = STATE_CPU (sd, n); |
| 500 | PROFILE_DATA *data = CPU_PROFILE_DATA (cpu); |
| 501 | if (PROFILE_PC_COUNT (data) != NULL) |
| 502 | free (PROFILE_PC_COUNT (data)); |
| 503 | PROFILE_PC_COUNT (data) = NULL; |
| 504 | if (PROFILE_PC_EVENT (data) != NULL) |
| 505 | sim_events_deschedule (sd, PROFILE_PC_EVENT (data)); |
| 506 | PROFILE_PC_EVENT (data) = NULL; |
| 507 | } |
| 508 | } |
| 509 | |
| 510 | |
| 511 | static void |
| 512 | profile_pc_uninstall (SIM_DESC sd) |
| 513 | { |
| 514 | profile_pc_cleanup (sd); |
| 515 | } |
| 516 | |
| 517 | static void |
| 518 | profile_pc_event (SIM_DESC sd, |
| 519 | void *data) |
| 520 | { |
| 521 | sim_cpu *cpu = (sim_cpu*) data; |
| 522 | PROFILE_DATA *profile = CPU_PROFILE_DATA (cpu); |
| 523 | address_word pc; |
| 524 | unsigned i; |
| 525 | switch (STATE_WATCHPOINTS (sd)->sizeof_pc) |
| 526 | { |
| 527 | case 2: pc = *(unsigned_2*)(STATE_WATCHPOINTS (sd)->pc) ; break; |
| 528 | case 4: pc = *(unsigned_4*)(STATE_WATCHPOINTS (sd)->pc) ; break; |
| 529 | case 8: pc = *(unsigned_8*)(STATE_WATCHPOINTS (sd)->pc) ; break; |
| 530 | default: pc = 0; |
| 531 | } |
| 532 | i = (pc - PROFILE_PC_START (profile)) >> PROFILE_PC_SHIFT (profile); |
| 533 | if (i < PROFILE_PC_NR_BUCKETS (profile)) |
| 534 | PROFILE_PC_COUNT (profile) [i] += 1; /* Overflow? */ |
| 535 | else |
| 536 | PROFILE_PC_COUNT (profile) [PROFILE_PC_NR_BUCKETS (profile)] += 1; |
| 537 | PROFILE_PC_EVENT (profile) = |
| 538 | sim_events_schedule (sd, PROFILE_PC_FREQ (profile), profile_pc_event, cpu); |
| 539 | } |
| 540 | |
| 541 | static SIM_RC |
| 542 | profile_pc_init (SIM_DESC sd) |
| 543 | { |
| 544 | int n; |
| 545 | profile_pc_cleanup (sd); |
| 546 | for (n = 0; n < MAX_NR_PROCESSORS; n++) |
| 547 | { |
| 548 | sim_cpu *cpu = STATE_CPU (sd, n); |
| 549 | PROFILE_DATA *data = CPU_PROFILE_DATA (cpu); |
| 550 | if (CPU_PROFILE_FLAGS (STATE_CPU (sd, n))[PROFILE_PC_IDX] |
| 551 | && STATE_WATCHPOINTS (sd)->pc != NULL) |
| 552 | { |
| 553 | int bucket_size; |
| 554 | /* fill in the frequency if not specified */ |
| 555 | if (PROFILE_PC_FREQ (data) == 0) |
| 556 | PROFILE_PC_FREQ (data) = 257; |
| 557 | /* fill in the start/end if not specified */ |
| 558 | if (PROFILE_PC_END (data) == 0) |
| 559 | { |
| 560 | PROFILE_PC_START (data) = STATE_TEXT_START (sd); |
| 561 | PROFILE_PC_END (data) = STATE_TEXT_END (sd); |
| 562 | } |
| 563 | /* Compute the number of buckets if not specified. */ |
| 564 | if (PROFILE_PC_NR_BUCKETS (data) == 0) |
| 565 | { |
| 566 | if (PROFILE_PC_BUCKET_SIZE (data) == 0) |
| 567 | PROFILE_PC_NR_BUCKETS (data) = 16; |
| 568 | else |
| 569 | { |
| 570 | if (PROFILE_PC_END (data) == 0) |
| 571 | { |
| 572 | /* nr_buckets = (full-address-range / 2) / (bucket_size / 2) */ |
| 573 | PROFILE_PC_NR_BUCKETS (data) = |
| 574 | ((1 << (STATE_WATCHPOINTS (sd)->sizeof_pc) * (8 - 1)) |
| 575 | / (PROFILE_PC_BUCKET_SIZE (data) / 2)); |
| 576 | } |
| 577 | else |
| 578 | { |
| 579 | PROFILE_PC_NR_BUCKETS (data) = |
| 580 | ((PROFILE_PC_END (data) |
| 581 | - PROFILE_PC_START (data) |
| 582 | + PROFILE_PC_BUCKET_SIZE (data) - 1) |
| 583 | / PROFILE_PC_BUCKET_SIZE (data)); |
| 584 | } |
| 585 | } |
| 586 | } |
| 587 | /* Compute the bucket size if not specified. Ensure that it |
| 588 | is rounded up to the next power of two */ |
| 589 | if (PROFILE_PC_BUCKET_SIZE (data) == 0) |
| 590 | { |
| 591 | if (PROFILE_PC_END (data) == 0) |
| 592 | /* bucket_size = (full-address-range / 2) / (nr_buckets / 2) */ |
| 593 | bucket_size = ((1 << ((STATE_WATCHPOINTS (sd)->sizeof_pc * 8) - 1)) |
| 594 | / (PROFILE_PC_NR_BUCKETS (data) / 2)); |
| 595 | else |
| 596 | bucket_size = ((PROFILE_PC_END (data) |
| 597 | - PROFILE_PC_START (data) |
| 598 | + PROFILE_PC_NR_BUCKETS (data) - 1) |
| 599 | / PROFILE_PC_NR_BUCKETS (data)); |
| 600 | PROFILE_PC_SHIFT (data) = 0; |
| 601 | while (bucket_size > PROFILE_PC_BUCKET_SIZE (data)) |
| 602 | { |
| 603 | PROFILE_PC_SHIFT (data) += 1; |
| 604 | } |
| 605 | } |
| 606 | /* Align the end address with bucket size */ |
| 607 | if (PROFILE_PC_END (data) != 0) |
| 608 | PROFILE_PC_END (data) = (PROFILE_PC_START (data) |
| 609 | + (PROFILE_PC_BUCKET_SIZE (data) |
| 610 | * PROFILE_PC_NR_BUCKETS (data))); |
| 611 | /* create the relevant buffers */ |
| 612 | PROFILE_PC_COUNT (data) = |
| 613 | NZALLOC (unsigned, PROFILE_PC_NR_BUCKETS (data) + 1); |
| 614 | PROFILE_PC_EVENT (data) = |
| 615 | sim_events_schedule (sd, |
| 616 | PROFILE_PC_FREQ (data), |
| 617 | profile_pc_event, |
| 618 | cpu); |
| 619 | } |
| 620 | } |
| 621 | return SIM_RC_OK; |
| 622 | } |
| 623 | |
| 624 | static void |
| 625 | profile_print_pc (sim_cpu *cpu, int verbose) |
| 626 | { |
| 627 | SIM_DESC sd = CPU_STATE (cpu); |
| 628 | PROFILE_DATA *profile = CPU_PROFILE_DATA (cpu); |
| 629 | char comma_buf[20]; |
| 630 | unsigned max_val; |
| 631 | unsigned total; |
| 632 | unsigned i; |
| 633 | |
| 634 | if (PROFILE_PC_COUNT (profile) == 0) |
| 635 | return; |
| 636 | |
| 637 | profile_printf (sd, cpu, "Program Counter Statistics:\n\n"); |
| 638 | |
| 639 | /* First pass over data computes various things. */ |
| 640 | max_val = 0; |
| 641 | total = 0; |
| 642 | for (i = 0; i <= PROFILE_PC_NR_BUCKETS (profile); ++i) |
| 643 | { |
| 644 | total += PROFILE_PC_COUNT (profile) [i]; |
| 645 | if (PROFILE_PC_COUNT (profile) [i] > max_val) |
| 646 | max_val = PROFILE_PC_COUNT (profile) [i]; |
| 647 | } |
| 648 | |
| 649 | profile_printf (sd, cpu, " Total samples: %s\n", |
| 650 | COMMAS (total)); |
| 651 | profile_printf (sd, cpu, " Granularity: %s bytes per bucket\n", |
| 652 | COMMAS (PROFILE_PC_BUCKET_SIZE (profile))); |
| 653 | profile_printf (sd, cpu, " Size: %s buckets\n", |
| 654 | COMMAS (PROFILE_PC_NR_BUCKETS (profile))); |
| 655 | profile_printf (sd, cpu, " Frequency: %s cycles per sample\n", |
| 656 | COMMAS (PROFILE_PC_FREQ (profile))); |
| 657 | |
| 658 | if (PROFILE_PC_END (profile) != 0) |
| 659 | profile_printf (sd, cpu, " Range: 0x%lx 0x%lx\n", |
| 660 | (long) PROFILE_PC_START (profile), |
| 661 | (long) PROFILE_PC_END (profile)); |
| 662 | |
| 663 | if (verbose && max_val != 0) |
| 664 | { |
| 665 | /* Now we can print the histogram. */ |
| 666 | profile_printf (sd, cpu, "\n"); |
| 667 | for (i = 0; i <= PROFILE_PC_NR_BUCKETS (profile); ++i) |
| 668 | { |
| 669 | if (PROFILE_PC_COUNT (profile) [i] != 0) |
| 670 | { |
| 671 | profile_printf (sd, cpu, " "); |
| 672 | if (i == PROFILE_PC_NR_BUCKETS (profile)) |
| 673 | profile_printf (sd, cpu, "%10s:", "overflow"); |
| 674 | else |
| 675 | profile_printf (sd, cpu, "0x%08lx:", |
| 676 | (long) (PROFILE_PC_START (profile) |
| 677 | + (i * PROFILE_PC_BUCKET_SIZE (profile)))); |
| 678 | profile_printf (sd, cpu, " %*s", |
| 679 | max_val < 10000 ? 5 : 10, |
| 680 | COMMAS (PROFILE_PC_COUNT (profile) [i])); |
| 681 | profile_printf (sd, cpu, " %4.1f", |
| 682 | (PROFILE_PC_COUNT (profile) [i] * 100.0) / total); |
| 683 | profile_printf (sd, cpu, ": "); |
| 684 | sim_profile_print_bar (sd, cpu, PROFILE_HISTOGRAM_WIDTH, |
| 685 | PROFILE_PC_COUNT (profile) [i], |
| 686 | max_val); |
| 687 | profile_printf (sd, cpu, "\n"); |
| 688 | } |
| 689 | } |
| 690 | } |
| 691 | |
| 692 | /* dump the histogram to the file "gmon.out" using BSD's gprof file |
| 693 | format */ |
| 694 | /* Since a profile data file is in the native format of the host on |
| 695 | which the profile is being, endian issues are not considered in |
| 696 | the code below. */ |
| 697 | /* FIXME: Is this the best place for this code? */ |
| 698 | { |
| 699 | FILE *pf = fopen ("gmon.out", "wb"); |
| 700 | |
| 701 | if (pf == NULL) |
| 702 | sim_io_eprintf (sd, "Failed to open \"gmon.out\" profile file\n"); |
| 703 | else |
| 704 | { |
| 705 | int ok; |
| 706 | /* FIXME: what if the target has a 64 bit PC? */ |
| 707 | unsigned32 header[3]; |
| 708 | unsigned loop; |
| 709 | if (PROFILE_PC_END (profile) != 0) |
| 710 | { |
| 711 | header[0] = PROFILE_PC_START (profile); |
| 712 | header[1] = PROFILE_PC_END (profile); |
| 713 | } |
| 714 | else |
| 715 | { |
| 716 | header[0] = 0; |
| 717 | header[1] = 0; |
| 718 | } |
| 719 | /* size of sample buffer (+ header) */ |
| 720 | header[2] = PROFILE_PC_NR_BUCKETS (profile) * 2 + sizeof (header); |
| 721 | |
| 722 | /* Header must be written out in target byte order. */ |
| 723 | H2T (header[0]); |
| 724 | H2T (header[1]); |
| 725 | H2T (header[2]); |
| 726 | |
| 727 | ok = fwrite (&header, sizeof (header), 1, pf); |
| 728 | for (loop = 0; |
| 729 | ok && (loop < PROFILE_PC_NR_BUCKETS (profile)); |
| 730 | loop++) |
| 731 | { |
| 732 | signed16 sample; |
| 733 | if (PROFILE_PC_COUNT (profile) [loop] >= 0xffff) |
| 734 | sample = 0xffff; |
| 735 | else |
| 736 | sample = PROFILE_PC_COUNT (profile) [loop]; |
| 737 | H2T (sample); |
| 738 | ok = fwrite (&sample, sizeof (sample), 1, pf); |
| 739 | } |
| 740 | if (ok == 0) |
| 741 | sim_io_eprintf (sd, "Failed to write to \"gmon.out\" profile file\n"); |
| 742 | fclose (pf); |
| 743 | } |
| 744 | } |
| 745 | |
| 746 | profile_printf (sd, cpu, "\n"); |
| 747 | } |
| 748 | |
| 749 | #endif |
| 750 | \f |
| 751 | /* Summary printing support. */ |
| 752 | |
| 753 | #if WITH_PROFILE_INSN_P |
| 754 | |
| 755 | static SIM_RC |
| 756 | profile_insn_init (SIM_DESC sd) |
| 757 | { |
| 758 | int c; |
| 759 | |
| 760 | for (c = 0; c < MAX_NR_PROCESSORS; ++c) |
| 761 | { |
| 762 | sim_cpu *cpu = STATE_CPU (sd, c); |
| 763 | |
| 764 | if (CPU_MAX_INSNS (cpu) > 0) |
| 765 | PROFILE_INSN_COUNT (CPU_PROFILE_DATA (cpu)) = NZALLOC (unsigned int, CPU_MAX_INSNS (cpu)); |
| 766 | } |
| 767 | |
| 768 | return SIM_RC_OK; |
| 769 | } |
| 770 | |
| 771 | static void |
| 772 | profile_print_insn (sim_cpu *cpu, int verbose) |
| 773 | { |
| 774 | unsigned int i, n, total, max_val, max_name_len; |
| 775 | SIM_DESC sd = CPU_STATE (cpu); |
| 776 | PROFILE_DATA *data = CPU_PROFILE_DATA (cpu); |
| 777 | char comma_buf[20]; |
| 778 | |
| 779 | /* If MAX_INSNS not set, insn profiling isn't supported. */ |
| 780 | if (CPU_MAX_INSNS (cpu) == 0) |
| 781 | return; |
| 782 | |
| 783 | profile_printf (sd, cpu, "Instruction Statistics"); |
| 784 | #ifdef SIM_HAVE_ADDR_RANGE |
| 785 | if (PROFILE_RANGE (data)->ranges) |
| 786 | profile_printf (sd, cpu, " (for selected address range(s))"); |
| 787 | #endif |
| 788 | profile_printf (sd, cpu, "\n\n"); |
| 789 | |
| 790 | /* First pass over data computes various things. */ |
| 791 | max_val = 0; |
| 792 | total = 0; |
| 793 | max_name_len = 0; |
| 794 | for (i = 0; i < CPU_MAX_INSNS (cpu); ++i) |
| 795 | { |
| 796 | const char *name = (*CPU_INSN_NAME (cpu)) (cpu, i); |
| 797 | |
| 798 | if (name == NULL) |
| 799 | continue; |
| 800 | total += PROFILE_INSN_COUNT (data) [i]; |
| 801 | if (PROFILE_INSN_COUNT (data) [i] > max_val) |
| 802 | max_val = PROFILE_INSN_COUNT (data) [i]; |
| 803 | n = strlen (name); |
| 804 | if (n > max_name_len) |
| 805 | max_name_len = n; |
| 806 | } |
| 807 | /* set the total insn count, in case client is being lazy */ |
| 808 | if (! PROFILE_TOTAL_INSN_COUNT (data)) |
| 809 | PROFILE_TOTAL_INSN_COUNT (data) = total; |
| 810 | |
| 811 | profile_printf (sd, cpu, " Total: %s insns\n", COMMAS (total)); |
| 812 | |
| 813 | if (verbose && max_val != 0) |
| 814 | { |
| 815 | /* Now we can print the histogram. */ |
| 816 | profile_printf (sd, cpu, "\n"); |
| 817 | for (i = 0; i < CPU_MAX_INSNS (cpu); ++i) |
| 818 | { |
| 819 | const char *name = (*CPU_INSN_NAME (cpu)) (cpu, i); |
| 820 | |
| 821 | if (name == NULL) |
| 822 | continue; |
| 823 | if (PROFILE_INSN_COUNT (data) [i] != 0) |
| 824 | { |
| 825 | profile_printf (sd, cpu, " %*s: %*s: ", |
| 826 | max_name_len, name, |
| 827 | max_val < 10000 ? 5 : 10, |
| 828 | COMMAS (PROFILE_INSN_COUNT (data) [i])); |
| 829 | sim_profile_print_bar (sd, cpu, PROFILE_HISTOGRAM_WIDTH, |
| 830 | PROFILE_INSN_COUNT (data) [i], |
| 831 | max_val); |
| 832 | profile_printf (sd, cpu, "\n"); |
| 833 | } |
| 834 | } |
| 835 | } |
| 836 | |
| 837 | profile_printf (sd, cpu, "\n"); |
| 838 | } |
| 839 | |
| 840 | #endif |
| 841 | |
| 842 | #if WITH_PROFILE_MEMORY_P |
| 843 | |
| 844 | static void |
| 845 | profile_print_memory (sim_cpu *cpu, int verbose) |
| 846 | { |
| 847 | unsigned int i, n; |
| 848 | unsigned int total_read, total_write; |
| 849 | unsigned int max_val, max_name_len; |
| 850 | /* FIXME: Need to add smp support. */ |
| 851 | SIM_DESC sd = CPU_STATE (cpu); |
| 852 | PROFILE_DATA *data = CPU_PROFILE_DATA (cpu); |
| 853 | char comma_buf[20]; |
| 854 | |
| 855 | profile_printf (sd, cpu, "Memory Access Statistics\n\n"); |
| 856 | |
| 857 | /* First pass over data computes various things. */ |
| 858 | max_val = total_read = total_write = max_name_len = 0; |
| 859 | for (i = 0; i < MODE_TARGET_MAX; ++i) |
| 860 | { |
| 861 | total_read += PROFILE_READ_COUNT (data) [i]; |
| 862 | total_write += PROFILE_WRITE_COUNT (data) [i]; |
| 863 | if (PROFILE_READ_COUNT (data) [i] > max_val) |
| 864 | max_val = PROFILE_READ_COUNT (data) [i]; |
| 865 | if (PROFILE_WRITE_COUNT (data) [i] > max_val) |
| 866 | max_val = PROFILE_WRITE_COUNT (data) [i]; |
| 867 | n = strlen (MODE_NAME (i)); |
| 868 | if (n > max_name_len) |
| 869 | max_name_len = n; |
| 870 | } |
| 871 | |
| 872 | /* One could use PROFILE_LABEL_WIDTH here. I chose not to. */ |
| 873 | profile_printf (sd, cpu, " Total read: %s accesses\n", |
| 874 | COMMAS (total_read)); |
| 875 | profile_printf (sd, cpu, " Total write: %s accesses\n", |
| 876 | COMMAS (total_write)); |
| 877 | |
| 878 | if (verbose && max_val != 0) |
| 879 | { |
| 880 | /* FIXME: Need to separate instruction fetches from data fetches |
| 881 | as the former swamps the latter. */ |
| 882 | /* Now we can print the histogram. */ |
| 883 | profile_printf (sd, cpu, "\n"); |
| 884 | for (i = 0; i < MODE_TARGET_MAX; ++i) |
| 885 | { |
| 886 | if (PROFILE_READ_COUNT (data) [i] != 0) |
| 887 | { |
| 888 | profile_printf (sd, cpu, " %*s read: %*s: ", |
| 889 | max_name_len, MODE_NAME (i), |
| 890 | max_val < 10000 ? 5 : 10, |
| 891 | COMMAS (PROFILE_READ_COUNT (data) [i])); |
| 892 | sim_profile_print_bar (sd, cpu, PROFILE_HISTOGRAM_WIDTH, |
| 893 | PROFILE_READ_COUNT (data) [i], |
| 894 | max_val); |
| 895 | profile_printf (sd, cpu, "\n"); |
| 896 | } |
| 897 | if (PROFILE_WRITE_COUNT (data) [i] != 0) |
| 898 | { |
| 899 | profile_printf (sd, cpu, " %*s write: %*s: ", |
| 900 | max_name_len, MODE_NAME (i), |
| 901 | max_val < 10000 ? 5 : 10, |
| 902 | COMMAS (PROFILE_WRITE_COUNT (data) [i])); |
| 903 | sim_profile_print_bar (sd, cpu, PROFILE_HISTOGRAM_WIDTH, |
| 904 | PROFILE_WRITE_COUNT (data) [i], |
| 905 | max_val); |
| 906 | profile_printf (sd, cpu, "\n"); |
| 907 | } |
| 908 | } |
| 909 | } |
| 910 | |
| 911 | profile_printf (sd, cpu, "\n"); |
| 912 | } |
| 913 | |
| 914 | #endif |
| 915 | |
| 916 | #if WITH_PROFILE_CORE_P |
| 917 | |
| 918 | static void |
| 919 | profile_print_core (sim_cpu *cpu, int verbose) |
| 920 | { |
| 921 | unsigned int total; |
| 922 | unsigned int max_val; |
| 923 | /* FIXME: Need to add smp support. */ |
| 924 | SIM_DESC sd = CPU_STATE (cpu); |
| 925 | PROFILE_DATA *data = CPU_PROFILE_DATA (cpu); |
| 926 | char comma_buf[20]; |
| 927 | |
| 928 | profile_printf (sd, cpu, "CORE Statistics\n\n"); |
| 929 | |
| 930 | /* First pass over data computes various things. */ |
| 931 | { |
| 932 | unsigned map; |
| 933 | total = 0; |
| 934 | max_val = 0; |
| 935 | for (map = 0; map < nr_maps; map++) |
| 936 | { |
| 937 | total += PROFILE_CORE_COUNT (data) [map]; |
| 938 | if (PROFILE_CORE_COUNT (data) [map] > max_val) |
| 939 | max_val = PROFILE_CORE_COUNT (data) [map]; |
| 940 | } |
| 941 | } |
| 942 | |
| 943 | /* One could use PROFILE_LABEL_WIDTH here. I chose not to. */ |
| 944 | profile_printf (sd, cpu, " Total: %s accesses\n", |
| 945 | COMMAS (total)); |
| 946 | |
| 947 | if (verbose && max_val != 0) |
| 948 | { |
| 949 | unsigned map; |
| 950 | /* Now we can print the histogram. */ |
| 951 | profile_printf (sd, cpu, "\n"); |
| 952 | for (map = 0; map < nr_maps; map++) |
| 953 | { |
| 954 | if (PROFILE_CORE_COUNT (data) [map] != 0) |
| 955 | { |
| 956 | profile_printf (sd, cpu, "%10s:", map_to_str (map)); |
| 957 | profile_printf (sd, cpu, "%*s: ", |
| 958 | max_val < 10000 ? 5 : 10, |
| 959 | COMMAS (PROFILE_CORE_COUNT (data) [map])); |
| 960 | sim_profile_print_bar (sd, cpu, PROFILE_HISTOGRAM_WIDTH, |
| 961 | PROFILE_CORE_COUNT (data) [map], |
| 962 | max_val); |
| 963 | profile_printf (sd, cpu, "\n"); |
| 964 | } |
| 965 | } |
| 966 | } |
| 967 | |
| 968 | profile_printf (sd, cpu, "\n"); |
| 969 | } |
| 970 | |
| 971 | #endif |
| 972 | |
| 973 | #if WITH_PROFILE_MODEL_P |
| 974 | |
| 975 | static void |
| 976 | profile_print_model (sim_cpu *cpu, int verbose) |
| 977 | { |
| 978 | SIM_DESC sd = CPU_STATE (cpu); |
| 979 | PROFILE_DATA *data = CPU_PROFILE_DATA (cpu); |
| 980 | unsigned long cti_stall_cycles = PROFILE_MODEL_CTI_STALL_CYCLES (data); |
| 981 | unsigned long load_stall_cycles = PROFILE_MODEL_LOAD_STALL_CYCLES (data); |
| 982 | unsigned long total_cycles = PROFILE_MODEL_TOTAL_CYCLES (data); |
| 983 | char comma_buf[20]; |
| 984 | |
| 985 | profile_printf (sd, cpu, "Model %s Timing Information", |
| 986 | MODEL_NAME (CPU_MODEL (cpu))); |
| 987 | #ifdef SIM_HAVE_ADDR_RANGE |
| 988 | if (PROFILE_RANGE (data)->ranges) |
| 989 | profile_printf (sd, cpu, " (for selected address range(s))"); |
| 990 | #endif |
| 991 | profile_printf (sd, cpu, "\n\n"); |
| 992 | profile_printf (sd, cpu, " %-*s %s\n", |
| 993 | PROFILE_LABEL_WIDTH, "Taken branches:", |
| 994 | COMMAS (PROFILE_MODEL_TAKEN_COUNT (data))); |
| 995 | profile_printf (sd, cpu, " %-*s %s\n", |
| 996 | PROFILE_LABEL_WIDTH, "Untaken branches:", |
| 997 | COMMAS (PROFILE_MODEL_UNTAKEN_COUNT (data))); |
| 998 | profile_printf (sd, cpu, " %-*s %s\n", |
| 999 | PROFILE_LABEL_WIDTH, "Cycles stalled due to branches:", |
| 1000 | COMMAS (cti_stall_cycles)); |
| 1001 | profile_printf (sd, cpu, " %-*s %s\n", |
| 1002 | PROFILE_LABEL_WIDTH, "Cycles stalled due to loads:", |
| 1003 | COMMAS (load_stall_cycles)); |
| 1004 | profile_printf (sd, cpu, " %-*s %s\n", |
| 1005 | PROFILE_LABEL_WIDTH, "Total cycles (*approximate*):", |
| 1006 | COMMAS (total_cycles)); |
| 1007 | profile_printf (sd, cpu, "\n"); |
| 1008 | } |
| 1009 | |
| 1010 | #endif |
| 1011 | |
| 1012 | void |
| 1013 | sim_profile_print_bar (SIM_DESC sd, sim_cpu *cpu, unsigned int width, |
| 1014 | unsigned int val, unsigned int max_val) |
| 1015 | { |
| 1016 | unsigned int i, count; |
| 1017 | |
| 1018 | count = ((double) val / (double) max_val) * (double) width; |
| 1019 | |
| 1020 | for (i = 0; i < count; ++i) |
| 1021 | profile_printf (sd, cpu, "*"); |
| 1022 | } |
| 1023 | |
| 1024 | /* Print the simulator's execution speed for CPU. */ |
| 1025 | |
| 1026 | static void |
| 1027 | profile_print_speed (sim_cpu *cpu) |
| 1028 | { |
| 1029 | SIM_DESC sd = CPU_STATE (cpu); |
| 1030 | PROFILE_DATA *data = CPU_PROFILE_DATA (cpu); |
| 1031 | unsigned long milliseconds = sim_events_elapsed_time (sd); |
| 1032 | unsigned long total = PROFILE_TOTAL_INSN_COUNT (data); |
| 1033 | double clock; |
| 1034 | double secs; |
| 1035 | char comma_buf[20]; |
| 1036 | |
| 1037 | profile_printf (sd, cpu, "Simulator Execution Speed\n\n"); |
| 1038 | |
| 1039 | if (total != 0) |
| 1040 | profile_printf (sd, cpu, " Total instructions: %s\n", COMMAS (total)); |
| 1041 | |
| 1042 | if (milliseconds < 1000) |
| 1043 | profile_printf (sd, cpu, " Total execution time: < 1 second\n\n"); |
| 1044 | else |
| 1045 | { |
| 1046 | /* The printing of the time rounded to 2 decimal places makes the speed |
| 1047 | calculation seem incorrect [even though it is correct]. So round |
| 1048 | MILLISECONDS first. This can marginally affect the result, but it's |
| 1049 | better that the user not perceive there's a math error. */ |
| 1050 | secs = (double) milliseconds / 1000; |
| 1051 | secs = ((double) (unsigned long) (secs * 100 + .5)) / 100; |
| 1052 | profile_printf (sd, cpu, " Total execution time : %.2f seconds\n", secs); |
| 1053 | /* Don't confuse things with data that isn't useful. |
| 1054 | If we ran for less than 2 seconds, only use the data if we |
| 1055 | executed more than 100,000 insns. */ |
| 1056 | if (secs >= 2 || total >= 100000) |
| 1057 | profile_printf (sd, cpu, " Simulator speed: %s insns/second\n", |
| 1058 | COMMAS ((unsigned long) ((double) total / secs))); |
| 1059 | } |
| 1060 | |
| 1061 | /* Print simulated execution time if the cpu frequency has been specified. */ |
| 1062 | clock = PROFILE_CPU_FREQ (data); |
| 1063 | if (clock != 0) |
| 1064 | { |
| 1065 | if (clock >= 1000000) |
| 1066 | profile_printf (sd, cpu, " Simulated cpu frequency: %.2f MHz\n", |
| 1067 | clock / 1000000); |
| 1068 | else |
| 1069 | profile_printf (sd, cpu, " Simulated cpu frequency: %.2f Hz\n", clock); |
| 1070 | |
| 1071 | #if WITH_PROFILE_MODEL_P |
| 1072 | if (PROFILE_FLAGS (data) [PROFILE_MODEL_IDX]) |
| 1073 | { |
| 1074 | /* The printing of the time rounded to 2 decimal places makes the |
| 1075 | speed calculation seem incorrect [even though it is correct]. |
| 1076 | So round SECS first. This can marginally affect the result, |
| 1077 | but it's better that the user not perceive there's a math |
| 1078 | error. */ |
| 1079 | secs = PROFILE_MODEL_TOTAL_CYCLES (data) / clock; |
| 1080 | secs = ((double) (unsigned long) (secs * 100 + .5)) / 100; |
| 1081 | profile_printf (sd, cpu, " Simulated execution time: %.2f seconds\n", |
| 1082 | secs); |
| 1083 | } |
| 1084 | #endif /* WITH_PROFILE_MODEL_P */ |
| 1085 | } |
| 1086 | } |
| 1087 | |
| 1088 | #ifdef SIM_HAVE_ADDR_RANGE |
| 1089 | /* Print selected address ranges. */ |
| 1090 | |
| 1091 | static void |
| 1092 | profile_print_addr_ranges (sim_cpu *cpu) |
| 1093 | { |
| 1094 | ADDR_SUBRANGE *asr = PROFILE_RANGE (CPU_PROFILE_DATA (cpu))->ranges; |
| 1095 | SIM_DESC sd = CPU_STATE (cpu); |
| 1096 | |
| 1097 | if (asr) |
| 1098 | { |
| 1099 | profile_printf (sd, cpu, "Selected address ranges\n\n"); |
| 1100 | while (asr != NULL) |
| 1101 | { |
| 1102 | profile_printf (sd, cpu, " 0x%lx - 0x%lx\n", |
| 1103 | (long) asr->start, (long) asr->end); |
| 1104 | asr = asr->next; |
| 1105 | } |
| 1106 | profile_printf (sd, cpu, "\n"); |
| 1107 | } |
| 1108 | } |
| 1109 | #endif |
| 1110 | |
| 1111 | /* Top level function to print all summary profile information. |
| 1112 | It is [currently] intended that all such data is printed by this function. |
| 1113 | I'd rather keep it all in one place for now. To that end, MISC_CPU and |
| 1114 | MISC are callbacks used to print any miscellaneous data. |
| 1115 | |
| 1116 | One might want to add a user option that allows printing by type or by cpu |
| 1117 | (i.e. print all insn data for each cpu first, or print data cpu by cpu). |
| 1118 | This may be a case of featuritis so it's currently left out. |
| 1119 | |
| 1120 | Note that results are indented two spaces to distinguish them from |
| 1121 | section titles. */ |
| 1122 | |
| 1123 | static void |
| 1124 | profile_info (SIM_DESC sd, int verbose) |
| 1125 | { |
| 1126 | int i,c; |
| 1127 | int print_title_p = 0; |
| 1128 | |
| 1129 | /* Only print the title if some data has been collected. */ |
| 1130 | /* ??? Why don't we just exit if no data collected? */ |
| 1131 | /* FIXME: If the number of processors can be selected on the command line, |
| 1132 | then MAX_NR_PROCESSORS will need to take an argument of `sd'. */ |
| 1133 | |
| 1134 | for (c = 0; c < MAX_NR_PROCESSORS && !print_title_p; ++c) |
| 1135 | { |
| 1136 | sim_cpu *cpu = STATE_CPU (sd, c); |
| 1137 | PROFILE_DATA *data = CPU_PROFILE_DATA (cpu); |
| 1138 | |
| 1139 | for (i = 0; i < MAX_PROFILE_VALUES; ++i) |
| 1140 | if (PROFILE_FLAGS (data) [i]) |
| 1141 | { |
| 1142 | profile_printf (sd, cpu, "Summary profiling results:\n\n"); |
| 1143 | print_title_p = 1; |
| 1144 | break; |
| 1145 | } |
| 1146 | } |
| 1147 | |
| 1148 | /* Loop, cpu by cpu, printing results. */ |
| 1149 | |
| 1150 | for (c = 0; c < MAX_NR_PROCESSORS; ++c) |
| 1151 | { |
| 1152 | sim_cpu *cpu = STATE_CPU (sd, c); |
| 1153 | PROFILE_DATA *data = CPU_PROFILE_DATA (cpu); |
| 1154 | |
| 1155 | if (MAX_NR_PROCESSORS > 1 |
| 1156 | && (0 |
| 1157 | #if WITH_PROFILE_INSN_P |
| 1158 | || PROFILE_FLAGS (data) [PROFILE_INSN_IDX] |
| 1159 | #endif |
| 1160 | #if WITH_PROFILE_MEMORY_P |
| 1161 | || PROFILE_FLAGS (data) [PROFILE_MEMORY_IDX] |
| 1162 | #endif |
| 1163 | #if WITH_PROFILE_CORE_P |
| 1164 | || PROFILE_FLAGS (data) [PROFILE_CORE_IDX] |
| 1165 | #endif |
| 1166 | #if WITH_PROFILE_MODEL_P |
| 1167 | || PROFILE_FLAGS (data) [PROFILE_MODEL_IDX] |
| 1168 | #endif |
| 1169 | #if WITH_PROFILE_SCACHE_P && WITH_SCACHE |
| 1170 | || PROFILE_FLAGS (data) [PROFILE_SCACHE_IDX] |
| 1171 | #endif |
| 1172 | #if WITH_PROFILE_PC_P |
| 1173 | || PROFILE_FLAGS (data) [PROFILE_PC_IDX] |
| 1174 | #endif |
| 1175 | )) |
| 1176 | { |
| 1177 | profile_printf (sd, cpu, "CPU %d\n\n", c); |
| 1178 | } |
| 1179 | |
| 1180 | #ifdef SIM_HAVE_ADDR_RANGE |
| 1181 | if (print_title_p |
| 1182 | && (PROFILE_INSN_P (cpu) |
| 1183 | || PROFILE_MODEL_P (cpu))) |
| 1184 | profile_print_addr_ranges (cpu); |
| 1185 | #endif |
| 1186 | |
| 1187 | #if WITH_PROFILE_INSN_P |
| 1188 | if (PROFILE_FLAGS (data) [PROFILE_INSN_IDX]) |
| 1189 | profile_print_insn (cpu, verbose); |
| 1190 | #endif |
| 1191 | |
| 1192 | #if WITH_PROFILE_MEMORY_P |
| 1193 | if (PROFILE_FLAGS (data) [PROFILE_MEMORY_IDX]) |
| 1194 | profile_print_memory (cpu, verbose); |
| 1195 | #endif |
| 1196 | |
| 1197 | #if WITH_PROFILE_CORE_P |
| 1198 | if (PROFILE_FLAGS (data) [PROFILE_CORE_IDX]) |
| 1199 | profile_print_core (cpu, verbose); |
| 1200 | #endif |
| 1201 | |
| 1202 | #if WITH_PROFILE_MODEL_P |
| 1203 | if (PROFILE_FLAGS (data) [PROFILE_MODEL_IDX]) |
| 1204 | profile_print_model (cpu, verbose); |
| 1205 | #endif |
| 1206 | |
| 1207 | #if WITH_PROFILE_SCACHE_P && WITH_SCACHE |
| 1208 | if (PROFILE_FLAGS (data) [PROFILE_SCACHE_IDX]) |
| 1209 | scache_print_profile (cpu, verbose); |
| 1210 | #endif |
| 1211 | |
| 1212 | #if WITH_PROFILE_PC_P |
| 1213 | if (PROFILE_FLAGS (data) [PROFILE_PC_IDX]) |
| 1214 | profile_print_pc (cpu, verbose); |
| 1215 | #endif |
| 1216 | |
| 1217 | /* Print cpu-specific data before the execution speed. */ |
| 1218 | if (PROFILE_INFO_CPU_CALLBACK (data) != NULL) |
| 1219 | PROFILE_INFO_CPU_CALLBACK (data) (cpu, verbose); |
| 1220 | |
| 1221 | /* Always try to print execution time and speed. */ |
| 1222 | if (verbose |
| 1223 | || PROFILE_FLAGS (data) [PROFILE_INSN_IDX]) |
| 1224 | profile_print_speed (cpu); |
| 1225 | } |
| 1226 | |
| 1227 | /* Finally print non-cpu specific miscellaneous data. */ |
| 1228 | if (STATE_PROFILE_INFO_CALLBACK (sd)) |
| 1229 | STATE_PROFILE_INFO_CALLBACK (sd) (sd, verbose); |
| 1230 | |
| 1231 | } |
| 1232 | \f |
| 1233 | /* Install profiling support in the simulator. */ |
| 1234 | |
| 1235 | SIM_RC |
| 1236 | profile_install (SIM_DESC sd) |
| 1237 | { |
| 1238 | int i; |
| 1239 | |
| 1240 | SIM_ASSERT (STATE_MAGIC (sd) == SIM_MAGIC_NUMBER); |
| 1241 | sim_add_option_table (sd, NULL, profile_options); |
| 1242 | for (i = 0; i < MAX_NR_PROCESSORS; ++i) |
| 1243 | memset (CPU_PROFILE_DATA (STATE_CPU (sd, i)), 0, |
| 1244 | sizeof (* CPU_PROFILE_DATA (STATE_CPU (sd, i)))); |
| 1245 | #if WITH_PROFILE_INSN_P |
| 1246 | sim_module_add_init_fn (sd, profile_insn_init); |
| 1247 | #endif |
| 1248 | #if WITH_PROFILE_PC_P |
| 1249 | sim_module_add_uninstall_fn (sd, profile_pc_uninstall); |
| 1250 | sim_module_add_init_fn (sd, profile_pc_init); |
| 1251 | #endif |
| 1252 | sim_module_add_init_fn (sd, profile_init); |
| 1253 | sim_module_add_uninstall_fn (sd, profile_uninstall); |
| 1254 | sim_module_add_info_fn (sd, profile_info); |
| 1255 | return SIM_RC_OK; |
| 1256 | } |
| 1257 | |
| 1258 | static SIM_RC |
| 1259 | profile_init (SIM_DESC sd) |
| 1260 | { |
| 1261 | #ifdef SIM_HAVE_ADDR_RANGE |
| 1262 | /* Check if a range has been specified without specifying what to |
| 1263 | collect. */ |
| 1264 | { |
| 1265 | int i; |
| 1266 | |
| 1267 | for (i = 0; i < MAX_NR_PROCESSORS; ++i) |
| 1268 | { |
| 1269 | sim_cpu *cpu = STATE_CPU (sd, i); |
| 1270 | |
| 1271 | if (ADDR_RANGE_RANGES (PROFILE_RANGE (CPU_PROFILE_DATA (cpu))) |
| 1272 | && ! (PROFILE_INSN_P (cpu) |
| 1273 | || PROFILE_MODEL_P (cpu))) |
| 1274 | { |
| 1275 | sim_io_eprintf_cpu (cpu, "Profiling address range specified without --profile-insn or --profile-model.\n"); |
| 1276 | sim_io_eprintf_cpu (cpu, "Address range ignored.\n"); |
| 1277 | sim_addr_range_delete (PROFILE_RANGE (CPU_PROFILE_DATA (cpu)), |
| 1278 | 0, ~ (address_word) 0); |
| 1279 | } |
| 1280 | } |
| 1281 | } |
| 1282 | #endif |
| 1283 | |
| 1284 | return SIM_RC_OK; |
| 1285 | } |
| 1286 | |
| 1287 | static void |
| 1288 | profile_uninstall (SIM_DESC sd) |
| 1289 | { |
| 1290 | int i,j; |
| 1291 | |
| 1292 | for (i = 0; i < MAX_NR_PROCESSORS; ++i) |
| 1293 | { |
| 1294 | sim_cpu *cpu = STATE_CPU (sd, i); |
| 1295 | PROFILE_DATA *data = CPU_PROFILE_DATA (cpu); |
| 1296 | |
| 1297 | if (PROFILE_FILE (data) != NULL) |
| 1298 | { |
| 1299 | /* If output from different cpus is going to the same file, |
| 1300 | avoid closing the file twice. */ |
| 1301 | for (j = 0; j < i; ++j) |
| 1302 | if (PROFILE_FILE (CPU_PROFILE_DATA (STATE_CPU (sd, j))) |
| 1303 | == PROFILE_FILE (data)) |
| 1304 | break; |
| 1305 | if (i == j) |
| 1306 | fclose (PROFILE_FILE (data)); |
| 1307 | } |
| 1308 | |
| 1309 | if (PROFILE_INSN_COUNT (data) != NULL) |
| 1310 | free (PROFILE_INSN_COUNT (data)); |
| 1311 | } |
| 1312 | } |