… | |
… | |
376 | |
376 | |
377 | For this to work, the OpenGL library must be loaded, a GLX context must |
377 | For this to work, the OpenGL library must be loaded, a GLX context must |
378 | have been created and be made current, and C<dlsym> must be available and |
378 | have been created and be made current, and C<dlsym> must be available and |
379 | capable of finding the function via C<RTLD_DEFAULT>. |
379 | capable of finding the function via C<RTLD_DEFAULT>. |
380 | |
380 | |
|
|
381 | =head2 EVENT SYSTEM |
|
|
382 | |
|
|
383 | OpenCL can generate a number of (potentially) asynchronous events, for |
|
|
384 | example, after compiling a program, to signal a context-related error or, |
|
|
385 | perhaps most important, to signal completion of queued jobs (by setting |
|
|
386 | callbacks on OpenCL::Event objects). |
|
|
387 | |
|
|
388 | To facilitate this, this module maintains an event queue - each |
|
|
389 | time an asynchronous event happens, it is queued, and perl will be |
|
|
390 | interrupted. This is implemented via the L<Async::Interrupt> module. In |
|
|
391 | addition, this module has L<AnyEvent> support, so it can seamlessly |
|
|
392 | integrate itself into many event loops. |
|
|
393 | |
|
|
394 | Since this module is a bit hard to understand, here are some case examples: |
|
|
395 | |
|
|
396 | =head3 Don't use callbacks. |
|
|
397 | |
|
|
398 | When your program never uses any callbacks, then there will never be any |
|
|
399 | notifications you need to take care of, and therefore no need to worry |
|
|
400 | about all this. |
|
|
401 | |
|
|
402 | You can achieve a great deal by explicitly waiting for events, or using |
|
|
403 | barriers and flush calls. In many programs, there is no need at all to |
|
|
404 | tinker with asynchronous events. |
|
|
405 | |
|
|
406 | =head3 Use AnyEvent |
|
|
407 | |
|
|
408 | This module automatically registers a watcher that invokes all outstanding |
|
|
409 | event callbacks when AnyEvent is initialised (and block asynchronous |
|
|
410 | interruptions). Using this mode of operations is the safest and most |
|
|
411 | recommended one. |
|
|
412 | |
|
|
413 | To use this, simply use AnyEvent and this module normally, make sure you |
|
|
414 | have an event loop running: |
|
|
415 | |
|
|
416 | use Gtk2 -init; |
|
|
417 | use AnyEvent; |
|
|
418 | |
|
|
419 | # initialise AnyEvent, by creating a watcher, or: |
|
|
420 | AnyEvent::detect; |
|
|
421 | |
|
|
422 | my $e = $queue->enqueue_marker; |
|
|
423 | $e->cb (sub { |
|
|
424 | warn "opencl is finished\n"; |
|
|
425 | }) |
|
|
426 | |
|
|
427 | main Gtk2; |
|
|
428 | |
|
|
429 | Note that this module will not initialise AnyEvent for you. Before |
|
|
430 | AnyEvent is initialised, the module will asynchronously interrupt perl |
|
|
431 | instead. To avoid any surprises, it's best to explicitly initialise |
|
|
432 | AnyEvent. |
|
|
433 | |
|
|
434 | You can temporarily enable asynchronous interruptions (see next paragraph) |
|
|
435 | by calling C<$OpenCL::INTERRUPT->unblock> and disable them again by |
|
|
436 | calling C<$OpenCL::INTERRUPT->block>. |
|
|
437 | |
|
|
438 | =head3 Let yourself be interrupted at any time |
|
|
439 | |
|
|
440 | This mode is the default unless AnyEvent is loaded and initialised. In |
|
|
441 | this mode, OpenCL asynchronously interrupts a running perl program. The |
|
|
442 | emphasis is on both I<asynchronously> and I<running> here. |
|
|
443 | |
|
|
444 | Asynchronously means that perl might execute your callbacks at any |
|
|
445 | time. For example, in the following code (I<THAT YOU SHOULD NOT COPY>), |
|
|
446 | the C<until> loop following the marker call will be interrupted by the |
|
|
447 | callback: |
|
|
448 | |
|
|
449 | my $e = $queue->enqueue_marker; |
|
|
450 | my $flag; |
|
|
451 | $e->cb (sub { $flag = 1 }); |
|
|
452 | 1 until $flag; |
|
|
453 | # $flag is now 1 |
|
|
454 | |
|
|
455 | The reason why you shouldn't blindly copy the above code is that |
|
|
456 | busy waiting is a really really bad thing, and really really bad for |
|
|
457 | performance. |
|
|
458 | |
|
|
459 | While at first this asynchronous business might look exciting, it can be |
|
|
460 | really hard, because you need to be prepared for the callback code to be |
|
|
461 | executed at any time, which limits the amount of things the callback code |
|
|
462 | can do safely. |
|
|
463 | |
|
|
464 | This can be mitigated somewhat by using C<< |
|
|
465 | $OpenCL::INTERRUPT->scope_block >> (see the L<Async::Interrupt> |
|
|
466 | documentation for details). |
|
|
467 | |
|
|
468 | The other problem is that your program must be actively I<running> to be |
|
|
469 | interrupted. When you calculate stuff, your program is running. When you |
|
|
470 | hang in some C functions or other block execution (by calling C<sleep>, |
|
|
471 | C<select>, running an event loop and so on), your program is waiting, not |
|
|
472 | running. |
|
|
473 | |
|
|
474 | One way around that would be to attach a read watcher to your event loop, |
|
|
475 | listening for events on C<< $OpenCL::INTERRUPT->pipe_fileno >>, using a |
|
|
476 | dummy callback (C<sub { }>) to temporarily execute some perl code. |
|
|
477 | |
|
|
478 | That is then awfully close to using the built-in AnyEvent support above, |
|
|
479 | though, so consider that one instead. |
|
|
480 | |
|
|
481 | =head3 Be creative |
|
|
482 | |
|
|
483 | OpenCL exports the L<Async::Interrupt> object it uses in the global |
|
|
484 | variable C<$OpenCL::INTERRUPT>. You can configure it in any way you like. |
|
|
485 | |
|
|
486 | So if you want to feel like a real pro, err, wait, if you feel no risk |
|
|
487 | menas no fun, you can experiment by implementing your own mode of |
|
|
488 | operations. |
|
|
489 | |
381 | =cut |
490 | =cut |
382 | |
491 | |
383 | package OpenCL; |
492 | package OpenCL; |
384 | |
493 | |
385 | use common::sense; |
494 | use common::sense; |
|
|
495 | use Async::Interrupt (); |
|
|
496 | |
|
|
497 | our $POLL_FUNC; # set by XS |
386 | |
498 | |
387 | BEGIN { |
499 | BEGIN { |
388 | our $VERSION = '0.97'; |
500 | our $VERSION = '0.97'; |
389 | |
501 | |
390 | require XSLoader; |
502 | require XSLoader; |
… | |
… | |
440 | |
552 | |
441 | Returns all available OpenCL::Platform objects. |
553 | Returns all available OpenCL::Platform objects. |
442 | |
554 | |
443 | L<http://www.khronos.org/registry/cl/sdk/1.1/docs/man/xhtml/clGetPlatformIDs.html> |
555 | L<http://www.khronos.org/registry/cl/sdk/1.1/docs/man/xhtml/clGetPlatformIDs.html> |
444 | |
556 | |
445 | =item $ctx = OpenCL::context_from_type $properties, $type = OpenCL::DEVICE_TYPE_DEFAULT, $notify = undef |
557 | =item $ctx = OpenCL::context_from_type $properties, $type = OpenCL::DEVICE_TYPE_DEFAULT, $notify = $print_stderr |
446 | |
558 | |
447 | Tries to create a context from a default device and platform - never worked for me. |
559 | Tries to create a context from a default device and platform - never worked for me. |
448 | |
560 | |
449 | L<http://www.khronos.org/registry/cl/sdk/1.1/docs/man/xhtml/clCreateContextFromType.html> |
561 | L<http://www.khronos.org/registry/cl/sdk/1.1/docs/man/xhtml/clCreateContextFromType.html> |
450 | |
562 | |
451 | =item OpenCL::wait_for_events $wait_events... |
563 | =item OpenCL::wait_for_events $wait_events... |
452 | |
564 | |
453 | Waits for all events to complete. |
565 | Waits for all events to complete. |
454 | |
566 | |
455 | L<http://www.khronos.org/registry/cl/sdk/1.1/docs/man/xhtml/clWaitForEvents.html> |
567 | L<http://www.khronos.org/registry/cl/sdk/1.1/docs/man/xhtml/clWaitForEvents.html> |
|
|
568 | |
|
|
569 | =item OpenCL::poll |
|
|
570 | |
|
|
571 | Checks if there are any outstanding events (see L<EVENT SYSTEM>) and |
|
|
572 | invokes their callbacks. |
|
|
573 | |
|
|
574 | =item $OpenCL::INTERRUPT |
|
|
575 | |
|
|
576 | The L<Async::Interrupt> object used to signal asynchronous events (see |
|
|
577 | L<EVENT SYSTEM>). |
|
|
578 | |
|
|
579 | =cut |
|
|
580 | |
|
|
581 | our $INTERRUPT = new Async::Interrupt c_cb => [$POLL_FUNC, 0]; |
|
|
582 | |
|
|
583 | &_eq_initialise ($INTERRUPT->signal_func); |
|
|
584 | |
|
|
585 | =item $OpenCL::WATCHER |
|
|
586 | |
|
|
587 | The L<AnyEvent> watcher object used to watch for asynchronous events (see |
|
|
588 | L<EVENT SYSTEM>). This variable is C<undef> until L<AnyEvent> has been |
|
|
589 | loaded I<and> initialised (e.g. by calling C<AnyEvent::detect>). |
|
|
590 | |
|
|
591 | =cut |
|
|
592 | |
|
|
593 | our $WATCHER; |
|
|
594 | |
|
|
595 | sub _init_anyevent { |
|
|
596 | $INTERRUPT->block; |
|
|
597 | $WATCHER = AE::io ($INTERRUPT->pipe_fileno, 0, sub { $INTERRUPT->handle }); |
|
|
598 | } |
|
|
599 | |
|
|
600 | if (defined $AnyEvent::MODEL) { |
|
|
601 | _init_anyevent; |
|
|
602 | } else { |
|
|
603 | push @AnyEvent::post_detect, \&_init_anyevent; |
|
|
604 | } |
456 | |
605 | |
457 | =back |
606 | =back |
458 | |
607 | |
459 | =head2 THE OpenCL::Object CLASS |
608 | =head2 THE OpenCL::Object CLASS |
460 | |
609 | |
… | |
… | |
472 | C<IV> type in your code and cast that to the correct type. |
621 | C<IV> type in your code and cast that to the correct type. |
473 | |
622 | |
474 | =cut |
623 | =cut |
475 | |
624 | |
476 | sub OpenCL::Object::id { |
625 | sub OpenCL::Object::id { |
|
|
626 | ref $_[0] eq "SCALAR" |
477 | ${$_[0]} |
627 | ? ${ $_[0] } |
|
|
628 | : $_[0][0] |
478 | } |
629 | } |
479 | |
630 | |
480 | =back |
631 | =back |
481 | |
632 | |
482 | =head2 THE OpenCL::Platform CLASS |
633 | =head2 THE OpenCL::Platform CLASS |
… | |
… | |
485 | |
636 | |
486 | =item @devices = $platform->devices ($type = OpenCL::DEVICE_TYPE_ALL) |
637 | =item @devices = $platform->devices ($type = OpenCL::DEVICE_TYPE_ALL) |
487 | |
638 | |
488 | Returns a list of matching OpenCL::Device objects. |
639 | Returns a list of matching OpenCL::Device objects. |
489 | |
640 | |
490 | =item $ctx = $platform->context_from_type ($properties, $type = OpenCL::DEVICE_TYPE_DEFAULT, $notify = undef) |
641 | =item $ctx = $platform->context_from_type ($properties, $type = OpenCL::DEVICE_TYPE_DEFAULT, $notify = $print_stderr) |
491 | |
642 | |
492 | Tries to create a context. Never worked for me, and you need devices explicitly anyway. |
643 | Tries to create a context. Never worked for me, and you need devices explicitly anyway. |
493 | |
644 | |
494 | L<http://www.khronos.org/registry/cl/sdk/1.1/docs/man/xhtml/clCreateContextFromType.html> |
645 | L<http://www.khronos.org/registry/cl/sdk/1.1/docs/man/xhtml/clCreateContextFromType.html> |
495 | |
646 | |
496 | =item $ctx = $platform->context ($properties = undef, @$devices, $notify = undef) |
647 | =item $ctx = $platform->context ($properties, @$devices, $notify = $print_stderr) |
497 | |
648 | |
498 | Create a new OpenCL::Context object using the given device object(s)- a |
649 | Create a new OpenCL::Context object using the given device object(s)- a |
499 | CL_CONTEXT_PLATFORM property is supplied automatically. |
650 | CL_CONTEXT_PLATFORM property is supplied automatically. |
500 | |
651 | |
501 | L<http://www.khronos.org/registry/cl/sdk/1.1/docs/man/xhtml/clCreateContext.html> |
652 | L<http://www.khronos.org/registry/cl/sdk/1.1/docs/man/xhtml/clCreateContext.html> |
… | |
… | |
846 | require Carp; |
997 | require Carp; |
847 | |
998 | |
848 | $prog = $self->program_with_source ($prog) |
999 | $prog = $self->program_with_source ($prog) |
849 | unless ref $prog; |
1000 | unless ref $prog; |
850 | |
1001 | |
|
|
1002 | # we build separately per device so we instantly know which one failed |
851 | for my $dev ($self->devices) { |
1003 | for my $dev ($self->devices) { |
852 | eval { $prog->build ($dev, $options); 1 } |
1004 | eval { $prog->build ([$dev], $options); 1 } |
853 | or Carp::croak ("Building OpenCL program for device '" . $dev->name . "' failed:\n" |
1005 | or Carp::croak ("Building OpenCL program for device '" . $dev->name . "' failed:\n" |
854 | . $prog->build_log ($dev)); |
1006 | . $prog->build_log ($dev)); |
855 | } |
1007 | } |
856 | |
1008 | |
857 | $prog |
1009 | $prog |
858 | } |
1010 | } |
859 | |
1011 | |
… | |
… | |
1351 | |
1503 | |
1352 | =head2 THE OpenCL::Program CLASS |
1504 | =head2 THE OpenCL::Program CLASS |
1353 | |
1505 | |
1354 | =over 4 |
1506 | =over 4 |
1355 | |
1507 | |
1356 | =item $program->build ($device, $options = "") |
1508 | =item $program->build (\@devices = undef, $options = "", $cb->($program) = undef) |
1357 | |
1509 | |
1358 | Tries to build the program with the given options. See also the |
1510 | Tries to build the program with the given options. See also the |
1359 | C<$ctx->build> convenience function. |
1511 | C<$ctx->build> convenience function. |
1360 | |
1512 | |
|
|
1513 | If a callback is specified, then it will be called when compilation is |
|
|
1514 | finished. Note that many OpenCL implementations block your program while |
|
|
1515 | compiling whether you use a callback or not. See C<build_async> if you |
|
|
1516 | want to make sure the build is done in the background. |
|
|
1517 | |
|
|
1518 | Note that some OpenCL implementations atc up badly, and don't call the |
|
|
1519 | callback in some error cases (but call it in others). This implementation |
|
|
1520 | assumes the callback will always be called, and leaks memory if this is |
|
|
1521 | not so. So best make sure you don't pass in invalid values. |
|
|
1522 | |
1361 | L<http://www.khronos.org/registry/cl/sdk/1.1/docs/man/xhtml/clBuildProgram.html> |
1523 | L<http://www.khronos.org/registry/cl/sdk/1.1/docs/man/xhtml/clBuildProgram.html> |
|
|
1524 | |
|
|
1525 | =item $program->build_async (\@devices = undef, $options = "", $cb->($program) = undef) |
|
|
1526 | |
|
|
1527 | Similar to C<< ->build >>, except it starts a thread, and never fails (you |
|
|
1528 | need to check the compilation status form the callback, or by polling). |
1362 | |
1529 | |
1363 | =item $packed_value = $program->build_info ($device, $name) |
1530 | =item $packed_value = $program->build_info ($device, $name) |
1364 | |
1531 | |
1365 | Similar to C<< $platform->info >>, but returns build info for a previous |
1532 | Similar to C<< $platform->info >>, but returns build info for a previous |
1366 | build attempt for the given device. |
1533 | build attempt for the given device. |
… | |
… | |
1543 | |
1710 | |
1544 | Waits for the event to complete. |
1711 | Waits for the event to complete. |
1545 | |
1712 | |
1546 | L<http://www.khronos.org/registry/cl/sdk/1.1/docs/man/xhtml/clWaitForEvents.html> |
1713 | L<http://www.khronos.org/registry/cl/sdk/1.1/docs/man/xhtml/clWaitForEvents.html> |
1547 | |
1714 | |
|
|
1715 | =item $ev->cb ($exec_callback_type, $callback->($event, $event_command_exec_status)) |
|
|
1716 | |
|
|
1717 | Adds a callback to the callback stack for the given event type. There is |
|
|
1718 | no way to remove a callback again. |
|
|
1719 | |
|
|
1720 | L<http://www.khronos.org/registry/cl/sdk/1.1/docs/man/xhtml/clSetEventCallback.html> |
|
|
1721 | |
1548 | =item $packed_value = $ev->info ($name) |
1722 | =item $packed_value = $ev->info ($name) |
1549 | |
1723 | |
1550 | See C<< $platform->info >> for details. |
1724 | See C<< $platform->info >> for details. |
1551 | |
1725 | |
1552 | L<http://www.khronos.org/registry/cl/sdk/1.1/docs/man/xhtml/clGetEventInfo.html> |
1726 | L<http://www.khronos.org/registry/cl/sdk/1.1/docs/man/xhtml/clGetEventInfo.html> |
… | |
… | |
1612 | |
1786 | |
1613 | =over 4 |
1787 | =over 4 |
1614 | |
1788 | |
1615 | =item $ev->set_status ($execution_status) |
1789 | =item $ev->set_status ($execution_status) |
1616 | |
1790 | |
|
|
1791 | Sets the execution status of the user event. Can only be called once, |
|
|
1792 | either with OpenCL::COMPLETE or a negative number as status. |
|
|
1793 | |
1617 | L<http://www.khronos.org/registry/cl/sdk/1.1/docs/man/xhtml/clSetUserEventStatus.html> |
1794 | L<http://www.khronos.org/registry/cl/sdk/1.1/docs/man/xhtml/clSetUserEventStatus.html> |
1618 | |
1795 | |
1619 | =back |
1796 | =back |
1620 | |
1797 | |
1621 | =cut |
1798 | =cut |