| 1 |
19 |
jeremybenn |
/* profiler.c -- profiling utility
|
| 2 |
|
|
|
| 3 |
|
|
Copyright (C) 2001 Marko Mlinar, markom@opencores.org
|
| 4 |
|
|
Copyright (C) 2008 Embecosm Limited
|
| 5 |
|
|
|
| 6 |
|
|
Contributor Jeremy Bennett <jeremy.bennett@embecosm.com>
|
| 7 |
|
|
|
| 8 |
|
|
This file is part of Or1ksim, the OpenRISC 1000 Architectural Simulator.
|
| 9 |
|
|
|
| 10 |
|
|
This program is free software; you can redistribute it and/or modify it
|
| 11 |
|
|
under the terms of the GNU General Public License as published by the Free
|
| 12 |
|
|
Software Foundation; either version 3 of the License, or (at your option)
|
| 13 |
|
|
any later version.
|
| 14 |
|
|
|
| 15 |
|
|
This program is distributed in the hope that it will be useful, but WITHOUT
|
| 16 |
|
|
ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
|
| 17 |
|
|
FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
|
| 18 |
|
|
more details.
|
| 19 |
|
|
|
| 20 |
|
|
You should have received a copy of the GNU General Public License along
|
| 21 |
|
|
with this program. If not, see <http://www.gnu.org/licenses/>. */
|
| 22 |
|
|
|
| 23 |
|
|
/* This program is commented throughout in a fashion suitable for processing
|
| 24 |
|
|
with Doxygen. */
|
| 25 |
|
|
|
| 26 |
|
|
/* Command line utility, that displays profiling information, generated
|
| 27 |
|
|
by or1ksim. (use profile command interactively, when running or1ksim, or
|
| 28 |
|
|
separate psim command). */
|
| 29 |
|
|
|
| 30 |
100 |
julius |
#define PROF_DEBUG 0
|
| 31 |
19 |
jeremybenn |
|
| 32 |
|
|
/* Autoconf and/or portability configuration */
|
| 33 |
|
|
#include "config.h"
|
| 34 |
|
|
#include "port.h"
|
| 35 |
|
|
|
| 36 |
|
|
/* Package includes */
|
| 37 |
|
|
#include "profiler.h"
|
| 38 |
|
|
#include "sim-config.h"
|
| 39 |
|
|
#include "argtable2.h"
|
| 40 |
|
|
|
| 41 |
|
|
/*! Maximum stack frames that can be profiled */
|
| 42 |
100 |
julius |
#define MAX_STACK 262144
|
| 43 |
19 |
jeremybenn |
|
| 44 |
|
|
/*! Data structure representing information about a stack frame */
|
| 45 |
|
|
struct stack_struct
|
| 46 |
|
|
{
|
| 47 |
|
|
unsigned int addr; /*!< Function address */
|
| 48 |
|
|
unsigned int cycles; /*!< Cycles of func start; subfuncs added later */
|
| 49 |
|
|
unsigned int raddr; /*!< Return address */
|
| 50 |
|
|
char name[33]; /*!< Name of the function */
|
| 51 |
|
|
};
|
| 52 |
|
|
|
| 53 |
|
|
/*! Global: data about functions */
|
| 54 |
|
|
struct func_struct prof_func[MAX_FUNCS];
|
| 55 |
|
|
|
| 56 |
|
|
/*! Global: total number of functions */
|
| 57 |
|
|
int prof_nfuncs = 0;
|
| 58 |
|
|
|
| 59 |
|
|
/*! Global: current cycles */
|
| 60 |
|
|
int prof_cycles = 0;
|
| 61 |
|
|
|
| 62 |
|
|
/*! Representation of the stack */
|
| 63 |
|
|
static struct stack_struct stack[MAX_STACK];
|
| 64 |
|
|
|
| 65 |
|
|
/*! Current depth */
|
| 66 |
|
|
static int nstack = 0;
|
| 67 |
|
|
|
| 68 |
|
|
/*! Max depth */
|
| 69 |
|
|
static int maxstack = 0;
|
| 70 |
|
|
|
| 71 |
|
|
/*! Number of total calls */
|
| 72 |
|
|
static int ntotcalls = 0;
|
| 73 |
|
|
|
| 74 |
|
|
/*! Number of covered calls */
|
| 75 |
|
|
static int nfunccalls = 0;
|
| 76 |
|
|
|
| 77 |
|
|
/*! Whether we are in cumulative mode */
|
| 78 |
|
|
static int cumulative = 0;
|
| 79 |
|
|
|
| 80 |
|
|
/*! Whether we should not report warnings */
|
| 81 |
|
|
static int quiet = 0;
|
| 82 |
|
|
|
| 83 |
|
|
/*! File to read from */
|
| 84 |
|
|
static FILE *fprof = 0;
|
| 85 |
|
|
|
| 86 |
|
|
|
| 87 |
|
|
/*---------------------------------------------------------------------------*/
|
| 88 |
|
|
/*! Acquire data from profiler file
|
| 89 |
|
|
|
| 90 |
|
|
@param[in] fprofname Data file to analyse
|
| 91 |
|
|
|
| 92 |
|
|
@return 0 on success, return code otherwise */
|
| 93 |
|
|
/*---------------------------------------------------------------------------*/
|
| 94 |
|
|
int
|
| 95 |
|
|
prof_acquire (const char *fprofname)
|
| 96 |
|
|
{
|
| 97 |
|
|
int line = 0;
|
| 98 |
|
|
int reopened = 0;
|
| 99 |
|
|
|
| 100 |
|
|
if (runtime.sim.fprof)
|
| 101 |
|
|
{
|
| 102 |
|
|
fprof = runtime.sim.fprof;
|
| 103 |
|
|
reopened = 1;
|
| 104 |
100 |
julius |
if (PROF_DEBUG) printf("reopened=1\n");
|
| 105 |
19 |
jeremybenn |
rewind (fprof);
|
| 106 |
|
|
}
|
| 107 |
|
|
else
|
| 108 |
|
|
fprof = fopen (fprofname, "rt");
|
| 109 |
|
|
|
| 110 |
|
|
if (!fprof)
|
| 111 |
|
|
{
|
| 112 |
|
|
fprintf (stderr, "Cannot open profile file: %s\n", fprofname);
|
| 113 |
|
|
return 1;
|
| 114 |
|
|
}
|
| 115 |
100 |
julius |
int ctr =0;
|
| 116 |
19 |
jeremybenn |
while (1)
|
| 117 |
|
|
{
|
| 118 |
100 |
julius |
if (PROF_DEBUG) printf("%d ",ctr++);
|
| 119 |
19 |
jeremybenn |
char dir = fgetc (fprof);
|
| 120 |
|
|
line++;
|
| 121 |
|
|
if (dir == '+')
|
| 122 |
|
|
{
|
| 123 |
|
|
if (fscanf
|
| 124 |
|
|
(fprof, "%08X %08X %08X %s\n", &stack[nstack].cycles,
|
| 125 |
|
|
&stack[nstack].raddr, &stack[nstack].addr,
|
| 126 |
|
|
&stack[nstack].name[0]) != 4)
|
| 127 |
|
|
fprintf (stderr, "Error reading line #%i\n", line);
|
| 128 |
|
|
else
|
| 129 |
|
|
{
|
| 130 |
|
|
prof_cycles = stack[nstack].cycles;
|
| 131 |
|
|
nstack++;
|
| 132 |
100 |
julius |
if (PROF_DEBUG) printf("+ 0x%.8x nstack %d\n",stack[nstack-1].raddr, nstack);
|
| 133 |
19 |
jeremybenn |
if (nstack > maxstack)
|
| 134 |
|
|
maxstack = nstack;
|
| 135 |
|
|
}
|
| 136 |
|
|
ntotcalls++;
|
| 137 |
|
|
}
|
| 138 |
|
|
else if (dir == '-')
|
| 139 |
|
|
{
|
| 140 |
|
|
struct stack_struct s;
|
| 141 |
|
|
if (fscanf (fprof, "%08X %08X\n", &s.cycles, &s.raddr) != 2)
|
| 142 |
|
|
fprintf (stderr, "Error reading line #%i\n", line);
|
| 143 |
|
|
else
|
| 144 |
|
|
{
|
| 145 |
|
|
int i;
|
| 146 |
|
|
prof_cycles = s.cycles;
|
| 147 |
100 |
julius |
if (PROF_DEBUG) printf("- 0x%.8x nstack %d\n",s.raddr ,nstack);
|
| 148 |
19 |
jeremybenn |
for (i = nstack - 1; i >= 0; i--)
|
| 149 |
|
|
if (stack[i].raddr == s.raddr)
|
| 150 |
|
|
break;
|
| 151 |
|
|
if (i >= 0)
|
| 152 |
|
|
{
|
| 153 |
|
|
/* pop everything above current from stack,
|
| 154 |
|
|
if more than one, something went wrong */
|
| 155 |
|
|
while (nstack > i)
|
| 156 |
|
|
{
|
| 157 |
|
|
int j;
|
| 158 |
|
|
long time;
|
| 159 |
|
|
nstack--;
|
| 160 |
|
|
time = s.cycles - stack[nstack].cycles;
|
| 161 |
|
|
if (!quiet && time < 0)
|
| 162 |
|
|
{
|
| 163 |
|
|
fprintf (stderr,
|
| 164 |
|
|
"WARNING: Negative time at %s (return addr = %08X).\n",
|
| 165 |
|
|
stack[i].name, stack[i].raddr);
|
| 166 |
|
|
time = 0;
|
| 167 |
|
|
}
|
| 168 |
|
|
|
| 169 |
|
|
/* Whether in normal mode, we must substract called function from execution time. */
|
| 170 |
|
|
if (!cumulative)
|
| 171 |
|
|
for (j = 0; j < nstack; j++)
|
| 172 |
|
|
stack[j].cycles += time;
|
| 173 |
|
|
|
| 174 |
|
|
if (!quiet && i != nstack)
|
| 175 |
|
|
fprintf (stderr,
|
| 176 |
|
|
"WARNING: Missaligned return call for %s (%08X) (found %s @ %08X), closing.\n",
|
| 177 |
|
|
stack[nstack].name, stack[nstack].raddr,
|
| 178 |
|
|
stack[i].name, stack[i].raddr);
|
| 179 |
|
|
|
| 180 |
|
|
for (j = 0; j < prof_nfuncs; j++)
|
| 181 |
|
|
if (stack[nstack].addr == prof_func[j].addr)
|
| 182 |
|
|
{ /* function exists, append. */
|
| 183 |
|
|
prof_func[j].cum_cycles += time;
|
| 184 |
|
|
prof_func[j].calls++;
|
| 185 |
|
|
nfunccalls++;
|
| 186 |
|
|
break;
|
| 187 |
|
|
}
|
| 188 |
|
|
if (j >= prof_nfuncs)
|
| 189 |
|
|
{ /* function does not yet exist, create new. */
|
| 190 |
|
|
prof_func[prof_nfuncs].cum_cycles = time;
|
| 191 |
|
|
prof_func[prof_nfuncs].calls = 1;
|
| 192 |
|
|
nfunccalls++;
|
| 193 |
|
|
prof_func[prof_nfuncs].addr = stack[nstack].addr;
|
| 194 |
|
|
strcpy (prof_func[prof_nfuncs].name,
|
| 195 |
|
|
stack[nstack].name);
|
| 196 |
|
|
prof_nfuncs++;
|
| 197 |
|
|
}
|
| 198 |
|
|
}
|
| 199 |
|
|
}
|
| 200 |
|
|
else if (!quiet)
|
| 201 |
|
|
fprintf (stderr,
|
| 202 |
|
|
"WARNING: Cannot find return call for (%08X), ignoring.\n",
|
| 203 |
|
|
s.raddr);
|
| 204 |
|
|
}
|
| 205 |
|
|
}
|
| 206 |
|
|
else
|
| 207 |
|
|
break;
|
| 208 |
|
|
}
|
| 209 |
|
|
|
| 210 |
|
|
/* If we have reopened the file, we need to add end of "[outside functions]" */
|
| 211 |
|
|
if (reopened)
|
| 212 |
100 |
julius |
{
|
| 213 |
19 |
jeremybenn |
prof_cycles = runtime.sim.cycles;
|
| 214 |
|
|
/* pop everything above current from stack,
|
| 215 |
|
|
if more than one, something went wrong */
|
| 216 |
|
|
while (nstack > 0)
|
| 217 |
|
|
{
|
| 218 |
|
|
int j;
|
| 219 |
|
|
long time;
|
| 220 |
|
|
nstack--;
|
| 221 |
|
|
time = runtime.sim.cycles - stack[nstack].cycles;
|
| 222 |
|
|
/* Whether in normal mode, we must substract called function from execution time. */
|
| 223 |
|
|
if (!cumulative)
|
| 224 |
|
|
for (j = 0; j < nstack; j++)
|
| 225 |
|
|
stack[j].cycles += time;
|
| 226 |
|
|
|
| 227 |
|
|
for (j = 0; j < prof_nfuncs; j++)
|
| 228 |
|
|
if (stack[nstack].addr == prof_func[j].addr)
|
| 229 |
|
|
{ /* function exists, append. */
|
| 230 |
|
|
prof_func[j].cum_cycles += time;
|
| 231 |
|
|
prof_func[j].calls++;
|
| 232 |
|
|
nfunccalls++;
|
| 233 |
|
|
break;
|
| 234 |
|
|
}
|
| 235 |
|
|
if (j >= prof_nfuncs)
|
| 236 |
|
|
{ /* function does not yet exist, create new. */
|
| 237 |
|
|
prof_func[prof_nfuncs].cum_cycles = time;
|
| 238 |
|
|
prof_func[prof_nfuncs].calls = 1;
|
| 239 |
|
|
nfunccalls++;
|
| 240 |
|
|
prof_func[prof_nfuncs].addr = stack[nstack].addr;
|
| 241 |
|
|
strcpy (prof_func[prof_nfuncs].name, stack[nstack].name);
|
| 242 |
|
|
prof_nfuncs++;
|
| 243 |
|
|
}
|
| 244 |
|
|
}
|
| 245 |
|
|
}
|
| 246 |
|
|
else
|
| 247 |
|
|
fclose (fprof);
|
| 248 |
|
|
return 0;
|
| 249 |
|
|
}
|
| 250 |
|
|
|
| 251 |
|
|
/* Print out profiling data */
|
| 252 |
|
|
static void
|
| 253 |
|
|
prof_print ()
|
| 254 |
|
|
{
|
| 255 |
|
|
int i, j;
|
| 256 |
|
|
if (cumulative)
|
| 257 |
|
|
PRINTF ("CUMULATIVE TIMES\n");
|
| 258 |
|
|
PRINTF
|
| 259 |
|
|
("---------------------------------------------------------------------------\n");
|
| 260 |
|
|
PRINTF
|
| 261 |
|
|
("|function name |addr |# calls |avg cycles |total cyles |\n");
|
| 262 |
|
|
PRINTF
|
| 263 |
|
|
("|-------------------------+--------+--------+------------+----------------|\n");
|
| 264 |
|
|
for (j = 0; j < prof_nfuncs; j++)
|
| 265 |
|
|
{
|
| 266 |
|
|
int bestcyc = 0, besti = 0;
|
| 267 |
|
|
for (i = 0; i < prof_nfuncs; i++)
|
| 268 |
|
|
if (prof_func[i].cum_cycles > bestcyc)
|
| 269 |
|
|
{
|
| 270 |
|
|
bestcyc = prof_func[i].cum_cycles;
|
| 271 |
|
|
besti = i;
|
| 272 |
|
|
}
|
| 273 |
|
|
i = besti;
|
| 274 |
|
|
PRINTF ("| %-24s|%08X|%8li|%12.1f|%11li,%3.0f%%|\n",
|
| 275 |
|
|
prof_func[i].name, prof_func[i].addr, prof_func[i].calls,
|
| 276 |
|
|
((double) prof_func[i].cum_cycles / prof_func[i].calls),
|
| 277 |
|
|
prof_func[i].cum_cycles,
|
| 278 |
|
|
(100. * prof_func[i].cum_cycles / prof_cycles));
|
| 279 |
|
|
prof_func[i].cum_cycles = -1;
|
| 280 |
|
|
}
|
| 281 |
|
|
PRINTF
|
| 282 |
|
|
("---------------------------------------------------------------------------\n");
|
| 283 |
|
|
PRINTF ("Total %i functions, %i cycles.\n", prof_nfuncs, prof_cycles);
|
| 284 |
|
|
PRINTF ("Total function calls %i/%i (max depth %i).\n", nfunccalls,
|
| 285 |
|
|
ntotcalls, maxstack);
|
| 286 |
|
|
}
|
| 287 |
|
|
|
| 288 |
|
|
/* Set options */
|
| 289 |
|
|
void
|
| 290 |
|
|
prof_set (int _quiet, int _cumulative)
|
| 291 |
|
|
{
|
| 292 |
|
|
quiet = _quiet;
|
| 293 |
|
|
cumulative = _cumulative;
|
| 294 |
|
|
}
|
| 295 |
|
|
|
| 296 |
|
|
/*---------------------------------------------------------------------------*/
|
| 297 |
|
|
/*! Parse the arguments for the profiling utility
|
| 298 |
|
|
|
| 299 |
|
|
Updated by Jeremy Bennett to use argtable2. Also has an option just to
|
| 300 |
|
|
print help, for use with the CLI.
|
| 301 |
|
|
|
| 302 |
|
|
@param[in] argc Number of command args
|
| 303 |
|
|
@param[in] argv Vector of the command args
|
| 304 |
|
|
@param[in] just_help If 1 (true), ignore argc & argv and just print out
|
| 305 |
|
|
the help message without parsing args
|
| 306 |
|
|
|
| 307 |
|
|
@return 0 on success, 1 on failure */
|
| 308 |
|
|
/*---------------------------------------------------------------------------*/
|
| 309 |
|
|
int
|
| 310 |
|
|
main_profiler (int argc, char *argv[], int just_help)
|
| 311 |
|
|
{
|
| 312 |
|
|
struct arg_lit *vercop;
|
| 313 |
|
|
struct arg_lit *help;
|
| 314 |
|
|
struct arg_lit *cum_arg;
|
| 315 |
|
|
struct arg_lit *quiet_arg;
|
| 316 |
|
|
struct arg_file *gen_file;
|
| 317 |
|
|
struct arg_end *end;
|
| 318 |
|
|
|
| 319 |
|
|
void *argtab[6];
|
| 320 |
|
|
int nerrors;
|
| 321 |
|
|
|
| 322 |
|
|
/* Specify each argument, with fallback values */
|
| 323 |
|
|
vercop = arg_lit0 ("v", "version", "version and copyright notice");
|
| 324 |
|
|
help = arg_lit0 ("h", "help", "print this help message");
|
| 325 |
|
|
cum_arg = arg_lit0 ("c", "cumulative",
|
| 326 |
|
|
"cumulative sum of cycles in functions");
|
| 327 |
|
|
quiet_arg = arg_lit0 ("q", "quiet", "suppress messages");
|
| 328 |
|
|
gen_file = arg_file0 ("g", "generate", "<file>",
|
| 329 |
|
|
"data file to analyse (default " "sim.profile)");
|
| 330 |
|
|
gen_file->filename[0] = "sim.profile";
|
| 331 |
|
|
end = arg_end (20);
|
| 332 |
|
|
|
| 333 |
|
|
/* Set up the argument table */
|
| 334 |
|
|
argtab[0] = vercop;
|
| 335 |
|
|
argtab[1] = help;
|
| 336 |
|
|
argtab[2] = cum_arg;
|
| 337 |
|
|
argtab[3] = quiet_arg;
|
| 338 |
|
|
argtab[4] = gen_file;
|
| 339 |
|
|
argtab[5] = end;
|
| 340 |
|
|
|
| 341 |
|
|
/* If we are just asked for a help message, then we don't parse the
|
| 342 |
|
|
args. This is used to implement the help function from the CLI. */
|
| 343 |
|
|
if (just_help)
|
| 344 |
|
|
{
|
| 345 |
|
|
printf ("profile");
|
| 346 |
|
|
arg_print_syntax (stdout, argtab, "\n");
|
| 347 |
|
|
arg_print_glossary (stdout, argtab, " %-25s %s\n");
|
| 348 |
|
|
|
| 349 |
|
|
arg_freetable (argtab, sizeof (argtab) / sizeof (argtab[0]));
|
| 350 |
|
|
return 0;
|
| 351 |
|
|
}
|
| 352 |
|
|
|
| 353 |
|
|
/* Parse */
|
| 354 |
|
|
nerrors = arg_parse (argc, argv, argtab);
|
| 355 |
|
|
|
| 356 |
|
|
/* Special case here is if help or version is specified, we ignore any other
|
| 357 |
|
|
errors and just print the help or version information and then give up. */
|
| 358 |
|
|
if (vercop->count > 0)
|
| 359 |
|
|
{
|
| 360 |
|
|
PRINTF ("OpenRISC 1000 Profiling Utility, version %s\n",
|
| 361 |
|
|
PACKAGE_VERSION);
|
| 362 |
|
|
|
| 363 |
|
|
arg_freetable (argtab, sizeof (argtab) / sizeof (argtab[0]));
|
| 364 |
|
|
return 0;
|
| 365 |
|
|
}
|
| 366 |
|
|
|
| 367 |
|
|
if (help->count > 0)
|
| 368 |
|
|
{
|
| 369 |
|
|
printf ("Usage: %s ", argv[0]);
|
| 370 |
|
|
arg_print_syntax (stdout, argtab, "\n");
|
| 371 |
|
|
arg_print_glossary (stdout, argtab, " %-25s %s\n");
|
| 372 |
|
|
|
| 373 |
|
|
arg_freetable (argtab, sizeof (argtab) / sizeof (argtab[0]));
|
| 374 |
|
|
return 0;
|
| 375 |
|
|
}
|
| 376 |
|
|
|
| 377 |
|
|
/* Deal with any errors */
|
| 378 |
|
|
if (0 != nerrors)
|
| 379 |
|
|
{
|
| 380 |
|
|
arg_print_errors (stderr, end, "profile");
|
| 381 |
|
|
fprintf (stderr, "Usage: %s ", argv[0]);
|
| 382 |
|
|
arg_print_syntaxv (stderr, argtab, "\n");
|
| 383 |
|
|
|
| 384 |
|
|
arg_freetable (argtab, sizeof (argtab) / sizeof (argtab[0]));
|
| 385 |
|
|
return 1;
|
| 386 |
|
|
}
|
| 387 |
|
|
|
| 388 |
|
|
/* Cumulative result wanted? */
|
| 389 |
|
|
cumulative = cum_arg->count;
|
| 390 |
|
|
|
| 391 |
|
|
/* Suppress messages? */
|
| 392 |
|
|
quiet = quiet_arg->count;
|
| 393 |
|
|
|
| 394 |
|
|
/* Get the profile from the file */
|
| 395 |
|
|
prof_acquire (gen_file->filename[0]);
|
| 396 |
|
|
|
| 397 |
|
|
/* Now we have all data acquired. Print out. */
|
| 398 |
|
|
prof_print ();
|
| 399 |
|
|
|
| 400 |
|
|
arg_freetable (argtab, sizeof (argtab) / sizeof (argtab[0]));
|
| 401 |
|
|
return 0;
|
| 402 |
|
|
|
| 403 |
|
|
} /* main_profiler() */
|