1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722
|
/*
PsychToolbox3/Source/Linux/Base/PsychTimeGlue.c
AUTHORS:
mario.kleiner at tuebingen.mpg.de mk
PLATFORMS: GNU/Linux Only
PROJECTS:
HISTORY:
2/20/06 mk Wrote it. Derived from Windows version.
1/03/09 mk Add generic Mutex locking support as service to ptb modules. Add PsychYieldIntervalSeconds().
DESCRIPTION:
Functions for querying system time and for waiting for either a
specified amount of time or until a specified point in time.
Also returns timer ticks and resolution of timers.
TO DO:
*/
#include "Psych.h"
#include <time.h>
#include <errno.h>
#include <sched.h>
// utsname for uname() so we can find out on which kernel we're running:
#include <sys/utsname.h>
/*
* file local state variables
*/
static double precisionTimerAdjustmentFactor=1;
static double estimatedGetSecsValueAtTickCountZero;
static psych_bool isKernelTimebaseFrequencyHzInitialized=FALSE;
static double kernelTimebaseFrequencyHz;
static double sleepwait_threshold = 0.01;
static double clockinc = 0;
void PsychWaitUntilSeconds(double whenSecs)
{
struct timespec rqtp;
double targettime;
static unsigned int missed_count=0;
double now=0.0;
int rc;
// Get current time:
PsychGetPrecisionTimerSeconds(&now);
// If the deadline has already passed, we do nothing and return immediately:
if (now >= whenSecs) return;
// Waiting stage 1: If we have more than sleepwait_threshold seconds left
// until the deadline, we call the OS usleep() function, so the
// CPU gets released for (difference - sleepwait_threshold) seconds to other processes and threads.
// -> Good for general system behaviour and for lowered power-consumption (longer battery runtime for
// Laptops) as the CPU can go idle if nothing else to do...
// Set an absolute deadline of whenSecs - sleepwait_threshold. We busy-wait the last few microseconds
// to take scheduling jitter/delays gracefully into account:
targettime = whenSecs - sleepwait_threshold;
// Convert targettime to timespec for the Posix clock functions:
rqtp.tv_sec = (unsigned long long) targettime;
rqtp.tv_nsec = ((targettime - (double) rqtp.tv_sec) * (double) 1e9);
// Use clock_nanosleep() to high-res sleep until targettime, repeat if that gets
// prematurely interrupted for whatever reason...
while(now < targettime) {
// MK: Oldstyle - obsolete: usleep((unsigned long)((whenSecs - now - sleepwait_threshold) * 1000000.0f));
// Starting in 2008, we use high-precision/high-resolution POSIX realtime timers for precise waiting:
// Call clock_nanosleep, use the realtime wall clock instead of the monotonic clock -- monotonic would
// by theoretically a bit better as NTP time adjustments couldn't mess with our sleep, but that would
// cause inconsistencies to other times reported by different useful system services which all measure
// against wall clock, and in practice, the effect of NTP adjustments is minimal or negligible, as these
// never create backwards running time or large timewarps, only 1 ppm level adjustments per second, ie,
// the effect is way below the sleepwait_threshold for any reasonable sleep time -- easily compensated by
// our hybrid approach...
// We use TIMER_ABSTIME, so we are totally drift-free and restartable in case our sleep gets interrupted by
// signals. If clock_nanosleep gets EINTR - Interrupted by a posix signal, we simply loop and restart the
// sleep. If it returns a different error condition, we abort sleep iteration -- something would be seriously
// wrong...
if ((rc = clock_nanosleep(CLOCK_REALTIME, TIMER_ABSTIME, &rqtp, NULL)) && (rc != EINTR)) break;
// Update our 'now' time for reiterating or continuing with busy-sleep...
PsychGetPrecisionTimerSeconds(&now);
}
// Waiting stage 2: We are less than sleepwait_threshold seconds away from deadline.
// Perform busy-waiting until deadline reached:
while(now < whenSecs) PsychGetPrecisionTimerSeconds(&now);
// Check for deadline-miss of more than 0.1 ms:
if (now - whenSecs > 0.0001) {
// Deadline missed by over 0.1 ms.
missed_count++;
// As long as the threshold is below a msec, immediately increase by 100 microsecs...
if (sleepwait_threshold < 0.001) sleepwait_threshold+=0.0001;
// If threshold has reached 1 msec, we require multiple consecutive misses before increasing any further:
if (missed_count>5) {
// Too many consecutive misses. Increase our threshold for sleep-waiting
// by 0.1 ms until it reaches max. 10 ms.
if (sleepwait_threshold < 0.01) sleepwait_threshold+=0.0001;
printf("PTB-WARNING: Wait-Deadline missed for %i consecutive times (Last miss %lf ms). New sleepwait_threshold is %lf ms.\n",
missed_count, (now - whenSecs)*1000.0f, sleepwait_threshold*1000.0f);
}
}
else {
// No miss detected. Reset counter...
missed_count=0;
}
// Ready.
return;
}
void PsychWaitIntervalSeconds(double delaySecs)
{
double deadline;
if (delaySecs <= 0) return;
// Get current time:
PsychGetPrecisionTimerSeconds(&deadline);
// Compute deadline in absolute system time:
deadline+=delaySecs;
// Wait until deadline reached:
PsychWaitUntilSeconds(deadline);
return;
}
/* PsychYieldIntervalSeconds() - Yield the cpu for given 'delaySecs'
*
* PsychYieldIntervalSeconds() differs from PsychWaitIntervalSeconds() in that
* it is supposed to release the cpu to other threads or processes for *at least*
* the given amount of time 'delaySecs', instead of *exactly* 'delaySecs'.
*
* If one wants to wait an exact amount of time, one uses PsychWaitIntervalSeconds().
* If one just "has nothing to do" for some minimum amount of time, and wants to
* play nice to other threads/processes and exact timing is not crucial, then
* this is the routine of choice. Typical use is within polling loops, where one
* wants to pause between polling cycles and it doesn't matter if the pause takes
* a bit longer.
*
* A 'delaySecs' of <= zero will just release the cpu for the remainder of
* the current scheduling timeslice. If you don't know what to do, choose a
* zero setting.
*
*/
void PsychYieldIntervalSeconds(double delaySecs)
{
if (delaySecs <= 0) {
// Yield cpu for remainder of this timeslice:
sched_yield();
}
else {
// On Linux we use standard wait ops - they're good enough for us.
// However, we make sure that the wait lasts at least 2x the sleepwait_threshold,
// so the cpu gets certainly released to other threads, instead of getting hogged
// by busy-waiting for too short delaySecs intervals - which would be detrimental
// to the goals of PsychYieldIntervalSeconds():
delaySecs = (delaySecs > 2.0 * sleepwait_threshold) ? delaySecs : (2.0 * sleepwait_threshold);
PsychWaitIntervalSeconds(delaySecs);
}
}
double PsychGetKernelTimebaseFrequencyHz(void)
{
if(!isKernelTimebaseFrequencyHzInitialized){
isKernelTimebaseFrequencyHzInitialized=TRUE;
PsychGetPrecisionTimerTicksPerSecond(&kernelTimebaseFrequencyHz);
}
return((double)kernelTimebaseFrequencyHz);
}
void PsychInitTimeGlue(void)
{
// TODO: Add Mutex init code for the timeglue mutex!
// Set this, although its totally pointless on our implementation...
PsychEstimateGetSecsValueAtTickCountZero();
}
/* Called at module shutdown/jettison time: */
void PsychExitTimeGlue(void)
{
// TODO: Add Mutex teardown code for the timeglue mutex!
return;
}
void PsychGetPrecisionTimerTicks(psych_uint64 *ticks)
{
double secs;
// MK: Simply map current systemtime to microseconds...
PsychGetPrecisionTimerSeconds(&secs);
*ticks = (psych_uint64) (secs * 1000000.0 + 0.5);
return;
}
void PsychGetPrecisionTimerTicksPerSecond(double *frequency)
{
// MK: Ok, set this to 1 million. Our timesource is gettimeofday(), which
// resolves time at microsecond resolution, so one can think of it as a
// virtual timer with a tickrate of 1 Mhz
*frequency=1000000.0f;
return;
}
void PsychGetPrecisionTimerTicksMinimumDelta(psych_uint32 *delta)
{
struct timespec res;
// We return the real clock tick resolution in microseconds, as 1 tick == 1 microsec
// in our implementation.
clock_getres(CLOCK_REALTIME, &res);
*delta = (psych_uint32) ((((double) res.tv_sec) + ((double) res.tv_nsec / 1e9)) * 1e6);
}
/* PsychOSGetLinuxMonotonicTime() -- Linux only.
*
* Return CLOCK_MONOTONIC time (usually system uptime) in seconds.
* Return zero on failure.
*
* Some subsystems return time not in gettimeofday() time aka CLOCK_REALTIME time,
* but in CLOCK_MONOTONIC time. In such cases we need to query this time to compute
* proper offsets for remapping into the gettimeofday() timebase which is used
* everywhere in PTB.
*
* An example is ALSA audio support in PsychPortAudio: ALSA drivers are free to
* return their audio timestamps in CLOCK_REALTIME time or CLOCK_MONOTONIC time,
* so we need to dynamically check, adapt and remap if neccessary.
*
*/
double PsychOSGetLinuxMonotonicTime(void)
{
struct timespec ts;
if (0!= clock_gettime(CLOCK_MONOTONIC, &ts)) return(0.0);
return((double) ts.tv_sec + ((double) ts.tv_nsec / (double) 1e9));
}
/* PsychOSMonotonicToRefTime(t)
*
* Map given input time value monotonicTime to PTB reference time if
* neccessary, pass-through otherwise.
*
* Can conditionally convert from CLOCK_MONOTONIC time to reftime, e.g.,
* to CLOCK_REALTIME aka gettimeofday().
*
*/
double PsychOSMonotonicToRefTime(double monotonicTime)
{
double now, now2, tMonotonic;
// Get current reftime:
PsychGetAdjustedPrecisionTimerSeconds(&now);
// Get current CLOCK_MONOTONIC time:
tMonotonic = PsychOSGetLinuxMonotonicTime();
// Given input monotonicTime time value closer to tMonotonic than to GetSecs time?
if (fabs(monotonicTime - tMonotonic) < fabs(monotonicTime - now)) {
// Timestamps are in monotonic time! Need to remap.
// Requery reference and monotonic time in a retry-loop
// to make sure remapping error is tighlty bounded to max. 20 usecs:
do {
// Get current reftime:
PsychGetAdjustedPrecisionTimerSeconds(&now);
// Get current CLOCK_MONOTONIC time:
tMonotonic = PsychOSGetLinuxMonotonicTime();
// Requery to make sure mapping is tight:
PsychGetAdjustedPrecisionTimerSeconds(&now2);
} while (now2 - now > 0.000020);
// Computer average of both timestamps to get best estimate of "now":
now = (now + now2) / 2;
// tMonotonic shall be the offset between GetSecs and monotonic time,
// i.e., the offset that needs to be added to monotonic timestamps to
// remap them to GetSecs time:
tMonotonic = now - tMonotonic;
// Correct timestamp by adding corrective offset:
monotonicTime += tMonotonic;
}
return(monotonicTime);
}
void PsychGetPrecisionTimerSeconds(double *secs)
{
static psych_bool firstTime = TRUE;
struct timespec res;
// First time invocation?
if (firstTime) {
// We query the real clock tick resolution in secs and store in global clockinc.
// This is useful as a constraint on sleepwait_threshold etc. for our sleep routines...
clock_getres(CLOCK_REALTIME, &res);
clockinc = ((double) res.tv_sec) + ((double) res.tv_nsec / 1.e9);
// sleepwait_threshold should be significantly higher than the granularity of
// the underlying system clock, say 100x the resolution, but no higher than 10 msecs,
// and no lower than 100 microseconds. We start with optimistic 250 microseconds...
sleepwait_threshold = 0.00025;
if (sleepwait_threshold < 100 * clockinc) sleepwait_threshold = 100 * clockinc;
if (sleepwait_threshold > 0.010) sleepwait_threshold = 0.010;
// Only output info about sleepwait threshold and clock resolution if we consider the
// clock rather low res, ie. increments bigger 20 microseconds:
if (clockinc > 0.00002) printf("PTB-INFO: Real resolution of (rather low resolution!) system clock is %1.4f microseconds, dynamic sleepwait_threshold starts with %lf msecs...\n", clockinc * 1e6, sleepwait_threshold * 1e3);
firstTime = FALSE;
}
// We use gettimeofday() - It works with microsecond resolution and
// is implemented via the highest precision time source on each
// Linux system, e.g., the processors performance counters on
// Intel Pentium systems. Actually, the resolution of the underlying
// clocksource is often much better than 1 microsecond, e.g., nanoseconds,
// but Linux chooses always the highest precision reliable source, so in
// case TSC's are broken and HPET's are not available and ACPI PM-Timers
// aren't available, it could be a worse than 1 usec source, although this
// is extremely unlikely...
static double oldss = -1;
double ss;
struct timeval tv;
gettimeofday(&tv, NULL);
ss = ((double) tv.tv_sec) + (((double) tv.tv_usec) / 1000000.0);
// Some correctness checks against last queried value, if initialized:
if (oldss > -1) {
// Old reference available. We check for monotonicity, ie. if time
// is not going backwards. That's all we can do, as we don't have access
// to a reference clock. We can't check for clock halts either, because
// given a fast machine, or concurrent calls from multiple threads and a
// low granularity clocksource, it could happen that multiple queries report
// the same time. However, i (MK) checked the source code of current Linux2.6.22
// kernels and found that Linux itself contains plenty of checks and measures to
// make sure the clock is working correctly and to compensate for any conceivable
// failure. We couldn't do a better job than the kernel in any case. This check
// is just to spot idiots in front of the keyboard that manually override Linux
// clocksource selection and basically try to shoot themselves into the leg.
// It may also be useful is somebody is running a very old Linux kernel without
// sophisticated checking and for testing/debugging PTB and its error-handling itself by
// fault-injection...
// MK: DISABLED FOR THIS RELEASE: Gives false alarms due to some race-condition when
// function is called from multiple concurrent threads. Proper fix is known, but i
// want to get a beta out now and not in a week...
// if (ss < oldss) {
if (FALSE) {
// Time warp detected! Time going backwards!!! Nothing we can do, only report
// it:
printf("\n\nPTB-CRITICAL-ERROR: Your systems clock is reporting time to run backwards!!!\n");
printf("PTB-CRITICAL-ERROR: (Delta %lf secs). This is impossible and indicates some\n", ss - oldss);
printf("PTB-CRITICAL-ERROR: broken clock hardware or Linux setup!! Stop using this machine\n");
printf("PTB-CRITICAL-ERROR: for psychophysics immmediately and resolve the problem!!!\n\n");
fflush(NULL);
}
}
// Init reference timestamp for checking in next call:
oldss = ss;
// Assign final time value:
*secs= ss;
}
void PsychGetAdjustedPrecisionTimerSeconds(double *secs)
{
double rawSecs, factor;
PsychGetPrecisionTimerSeconds(&rawSecs);
PsychGetPrecisionTimerAdjustmentFactor(&factor);
*secs=rawSecs * precisionTimerAdjustmentFactor;
}
void PsychGetPrecisionTimerAdjustmentFactor(double *factor)
{
*factor=precisionTimerAdjustmentFactor;
}
void PsychSetPrecisionTimerAdjustmentFactor(double *factor)
{
precisionTimerAdjustmentFactor=*factor;
}
/*
PsychEstimateGetSecsValueAtTickCountZero()
*/
void PsychEstimateGetSecsValueAtTickCountZero(void)
{
// MK: Todo - Implement GetTickCout().
// double nowTicks, nowSecs;
// nowTicks=(double) GetTickCount();
// PsychGetAdjustedPrecisionTimerSeconds(&nowSecs);
// estimatedGetSecsValueAtTickCountZero=nowSecs - nowTicks * (1/1000.0f);
// This is zero by definition of our counters...
estimatedGetSecsValueAtTickCountZero=0;
}
double PsychGetEstimatedSecsValueAtTickCountZero(void)
{
return(estimatedGetSecsValueAtTickCountZero);
}
/* Init a Mutex: */
int PsychInitMutex(psych_mutex* mutex)
{
int rc;
// Use mutex attributes:
pthread_mutexattr_t attr;
// Set them to default settings, except for...
pthread_mutexattr_init(&attr);
// ... priority inheritance: We absolutely want it for extra
// good realtime behaviour - Avoidance of priority inversion
// at lock contention points:
pthread_mutexattr_setprotocol(&attr, PTHREAD_PRIO_INHERIT);
// Create mutex with attributes in attr:
rc = pthread_mutex_init(mutex, &attr);
if (rc != 0) {
printf("\n\nPTB-CRITICAL: PsychInitMutex(): Mutex initialization failed [%s]! Expect huge trouble and serious malfunctions!!!\n", strerror(rc));
printf("PTB-CRITICAL: PsychInitMutex(): Set a breakpoint on your debugger on pthread_mutexattr_destroy() to debug this.\n\n");
}
// Done with it:
pthread_mutexattr_destroy(&attr);
return(rc);
}
/* Deinit and destroy a Mutex: */
int PsychDestroyMutex(psych_mutex* mutex)
{
return(pthread_mutex_destroy(mutex));
}
/* Lock a Mutex, blocking until mutex is available if it isn't available: */
int PsychLockMutex(psych_mutex* mutex)
{
return(pthread_mutex_lock(mutex));
}
/* Try to lock a Mutex, returning immediately, with a return code that tells if mutex could be locked or not: */
int PsychTryLockMutex(psych_mutex* mutex)
{
return(pthread_mutex_trylock(mutex));
}
/* Unlock a Mutex: */
int PsychUnlockMutex(psych_mutex* mutex)
{
return(pthread_mutex_unlock(mutex));
}
/* Create a parallel thread of execution, invoke its main routine: */
int PsychCreateThread(psych_thread* threadhandle, void* threadparams, void *(*start_routine)(void *), void *arg)
{
// threadparams not yet used, this line just to make compiler happy:
(void) threadparams;
// Return result code of pthread_create - We're a really thin wrapper around this Posix call:
return( pthread_create(threadhandle, NULL, start_routine, arg) );
}
/* Join a parallel thread - Wait for its termination, then return its result code: */
int PsychDeleteThread(psych_thread* threadhandle)
{
// Join on the thread, wait for termination:
int rc = pthread_join(*threadhandle, NULL);
// Null out now invalid thread handle of dead thread:
*threadhandle = 0;
// Return return code of joined thread:
return(rc);
}
/* Send abort request to thread: */
int PsychAbortThread(psych_thread* threadhandle)
{
return( pthread_cancel(*threadhandle) );
}
/* Check for abort request to thread: Exit thread gracefully if abort requested: */
void PsychTestCancelThread(psych_thread* threadhandle)
{
// threadhandle unused on POSIX: This line just to make compiler happy:
(void) threadhandle;
// Test for cancellation, cancel if so:
pthread_testcancel();
}
/* Return handle of calling thread: */
psych_threadid PsychGetThreadId(void)
{
return( pthread_self() );
}
/* Check if two given thread handles do refer to the same thread: */
int PsychIsThreadEqual(psych_thread threadOne, psych_thread threadTwo)
{
return( pthread_equal(threadOne, threadTwo) );
}
/* Check if current (invoking) thread has an id equal to given threadid: */
int PsychIsCurrentThreadEqualToId(psych_threadid threadId)
{
return( pthread_equal(PsychGetThreadId(), threadId) );
}
/* Check if current (invoking) thread is equal to given threadhandle: */
int PsychIsCurrentThreadEqualToPsychThread(psych_thread threadhandle)
{
return( pthread_equal(PsychGetThreadId(), threadhandle) );
}
/* Change priority for thread 'threadhandle', or for the calling thread if 'threadhandle' == NULL.
* threadhandle == 0x1 means "Main Psychtoolbox thread" and may incur special treatment.
* 'basePriority' can be 0 for normal scheduling, 1 for higher priority and 2 for highest priority.
* 'tweakPriority' modulates more fine-grained within the category given by 'basepriority'. It
* can be anywhere between 0 and some big value where bigger means more priority.
*
* Returns zero on success, non-zero on failure to set new priority.
*/
int PsychSetThreadPriority(psych_thread* threadhandle, int basePriority, int tweakPriority)
{
int rc = 0;
int policy;
struct sched_param sp;
pthread_t thread;
if ((NULL != threadhandle) && ((psych_thread*) 0x1 != threadhandle)) {
// Retrieve thread handle of thread to change:
thread = *threadhandle;
}
else {
// Retrieve handle of calling thread:
thread = pthread_self();
}
// Retrieve current scheduling policy and parameters:
pthread_getschedparam(thread, &policy, &sp);
switch(basePriority) {
case 0: // Normal priority. No change to scheduling priority:
policy = SCHED_OTHER;
sp.sched_priority = 0;
break;
case 1: // High priority / Round robin realtime.
policy = SCHED_RR;
sp.sched_priority = sp.sched_priority + tweakPriority;
break;
case 2: // Highest priority: FIFO scheduling
case 10: // Multimedia class scheduling emulation for non-Windows:
policy = SCHED_FIFO;
sp.sched_priority = sp.sched_priority + tweakPriority;
break;
default:
printf("PTB-CRITICAL: In call to PsychSetThreadPriority(): Invalid/Unknown basePriority %i provided!\n", basePriority);
rc = 2;
}
// Try to apply new priority and scheduling method:
if (rc == 0) {
// Make sure we have at least prio level 1 for RT scheduling policies:
if ((policy != SCHED_OTHER) && (sp.sched_priority < 1)) sp.sched_priority = 1;
rc = pthread_setschedparam(thread, policy, &sp);
if (rc != 0) printf("PTB-CRITICAL: In call to PsychSetThreadPriority(): Failed to set new basePriority %i, tweakPriority %i, effective %i [%s] for thread %p provided!\n",
basePriority, tweakPriority, sp.sched_priority, (policy != SCHED_OTHER) ? "REALTIME" : "NORMAL", (void*) threadhandle);
}
// rc is either zero for success, or 2 for invalid arg, or some other non-zero failure code:
return(rc);
}
/* Assign a name to a thread, for debugging: */
void PsychSetThreadName(const char *name)
{
# if defined(__GNU_LIBRARY__) && defined(__GLIBC__) && defined(__GLIBC_MINOR__) && \
(__GLIBC__ >= 3 || (__GLIBC__ == 2 && __GLIBC_MINOR__ >= 12))
pthread_setname_np(pthread_self(), name);
# endif
(void)name;
}
/* Initialize condition variable:
* CAUTION: Use of condition_attribute is non-portable! Code using it will not work properly
* on MS-Windows as this attribute is unsupported there! Pass NULL for this argument for
* portable operation!
*/
int PsychInitCondition(psych_condition* condition, const pthread_condattr_t* condition_attribute)
{
return(pthread_cond_init(condition, condition_attribute));
}
/* Destroy condition variable: */
int PsychDestroyCondition(psych_condition* condition)
{
return(pthread_cond_destroy(condition));
}
/* Signal/wakeup exactly one thread waiting on the given condition variable: */
int PsychSignalCondition(psych_condition* condition)
{
return(pthread_cond_signal(condition));
}
/* Signal/Wakeup all threads waiting on the given condition variable:
* CAUTION: Use of this function is non-portable to MS-Windows for now! Code
* using it will malfunction if used on MS-Windows!
*/
int PsychBroadcastCondition(psych_condition* condition)
{
return(pthread_cond_broadcast(condition));
}
/* Atomically release the 'mutex' lock and go to sleep, waiting for the 'condition' variable
* being signalled, then waking up and trying to re-lock the 'mutex'. Will return with
* mutex locked.
*/
int PsychWaitCondition(psych_condition* condition, psych_mutex* mutex)
{
return(pthread_cond_wait(condition, mutex));
}
/* Atomically release the 'mutex' lock and go to sleep, waiting for the 'condition' variable
* being signalled, then waking up and trying to re-lock the 'mutex'. Will return with
* mutex locked.
*
* Like PsychWaitCondition, but function will timeout if it fails being signalled before
* timeout interval 'maxwaittimesecs' expires. In any case, it will only return after
* reacquiring the mutex. It will retun zero on successfull wait, non-zero (ETIMEDOUT) if
* timeout was triggered without the condition being signalled.
*/
int PsychTimedWaitCondition(psych_condition* condition, psych_mutex* mutex, double maxwaittimesecs)
{
struct timespec abstime;
double tnow;
// Convert relative wait time to absolute system time:
PsychGetAdjustedPrecisionTimerSeconds(&tnow);
maxwaittimesecs+=tnow;
// Split maxwaittimesecs in...
// ... full integral seconds (floor() it)...
abstime.tv_sec = (time_t) maxwaittimesecs;
// ... and fractional seconds, expressed as nanoseconds in (long) format:
abstime.tv_nsec = (long) (((double) maxwaittimesecs - (double) abstime.tv_sec) * (double) (1e9));
// Perform wait with timeout:
return(pthread_cond_timedwait(condition, mutex, &abstime));
}
/* Set thread affinity mask of calling thread to the modules global cpuMask:
*
* 'curCpuMask' is an in/out pointer. If NULL, it is completely ignored. If non-NULL,
* the target variable of the pointer will contain the new cpu mask after a change
* of mask. If the target variable already contains a valid (non-zero) current cpu mask
* and it matches the new target cpu mask, then the actual mask change is omitted, assuming it
* is redundant, thereby saving some system call overhead.
*
* Threads can avoid redundant switches if they keep track of their current cpu mask
* by caching it in the *curCpuMask pointer target. They can pass in a zero value if
* unsure, or a NULL pointer if they are neither interested in caching, nor in the old
* or new value.
*
* Returns the old pre-switch affinity mask as a 64-Bit bitfield.
* Comparing the return value (previous mask) with the *curCpuMask value (new mask)
* allows the caller to check if the affinity mask was actually changed, resulting
* in a possible thread migration.
*
* If this function is called without the time lock held, ie., from outside
* of other timeglue functions, a small race condition exists which may cause
* deferred updated to the real new affinity mask due to
*
*/
psych_uint64 PsychAutoLockThreadToCores(psych_uint64* curCpuMask)
{
(void) curCpuMask;
// No op on Linux.
return(INT64_MAX);
}
/* Report official support status for this operating system release.
* The string "Supported" means supported.
* Other strings describe lack of support.
*/
const char* PsychSupportStatus(void)
{
static char statusString[256];
struct utsname unameresult;
uname(&unameresult);
sprintf(statusString, "Linux %s Supported.", unameresult.release);
return(statusString);
}
|