[ViewVC] Diff of: cvs/OpenCL/OpenCL.pm

Comparing OpenCL/OpenCL.pm (file contents):
Revision 1.53 by root, Tue Apr 24 14:30:57 2012 UTC vs.
Revision 1.57 by root, Tue Apr 24 23:58:34 2012 UTC

…		…
43		43
44	OpenCL::Event objects are used to signal when something is complete.	44	OpenCL::Event objects are used to signal when something is complete.
45		45
46	=head2 HELPFUL RESOURCES	46	=head2 HELPFUL RESOURCES
47		47
48	The OpenCL spec used to develop this module (1.2 spec was available, but	48	The OpenCL specs used to develop this module:
49	no implementation was available to me :).
50		49
51	http://www.khronos.org/registry/cl/specs/opencl-1.1.pdf	50	http://www.khronos.org/registry/cl/specs/opencl-1.1.pdf
		51	http://www.khronos.org/registry/cl/specs/opencl-1.2.pdf
		52	http://www.khronos.org/registry/cl/specs/opencl-1.2-extensions.pdf
52		53
53	OpenCL manpages:	54	OpenCL manpages:
54		55
55	http://www.khronos.org/registry/cl/sdk/1.1/docs/man/xhtml/	56	http://www.khronos.org/registry/cl/sdk/1.1/docs/man/xhtml/
		57	http://www.khronos.org/registry/cl/sdk/1.2/docs/man/xhtml/
56		58
57	If you are into UML class diagrams, the following diagram might help - if	59	If you are into UML class diagrams, the following diagram might help - if
58	not, it will be mildly cobfusing:	60	not, it will be mildly confusing (also, the class hierarchy of this module
		61	is much more fine-grained):
59		62
60	http://www.khronos.org/registry/cl/sdk/1.1/docs/man/xhtml/classDiagram.html	63	http://www.khronos.org/registry/cl/sdk/1.2/docs/man/xhtml/classDiagram.html
61		64
62	Here's a tutorial from AMD (very AMD-centric, too), not sure how useful it	65	Here's a tutorial from AMD (very AMD-centric, too), not sure how useful it
63	is, but at least it's free of charge:	66	is, but at least it's free of charge:
64		67
65	http://developer.amd.com/zones/OpenCLZone/courses/Documents/Introduction_to_OpenCL_Programming%20Training_Guide%20%28201005%29.pdf	68	http://developer.amd.com/zones/OpenCLZone/courses/Documents/Introduction_to_OpenCL_Programming%20Training_Guide%20%28201005%29.pdf
…		…
376		379
377	For this to work, the OpenGL library must be loaded, a GLX context must	380	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	381	have been created and be made current, and C<dlsym> must be available and
379	capable of finding the function via C<RTLD_DEFAULT>.	382	capable of finding the function via C<RTLD_DEFAULT>.
380		383
		384	=head2 EVENT SYSTEM
		385
		386	OpenCL can generate a number of (potentially) asynchronous events, for
		387	example, after compiling a program, to signal a context-related error or,
		388	perhaps most important, to signal completion of queued jobs (by setting
		389	callbacks on OpenCL::Event objects).
		390
		391	To facilitate this, this module maintains an event queue - each
		392	time an asynchronous event happens, it is queued, and perl will be
		393	interrupted. This is implemented via the L<Async::Interrupt> module. In
		394	addition, this module has L<AnyEvent> support, so it can seamlessly
		395	integrate itself into many event loops.
		396
		397	Since this module is a bit hard to understand, here are some case examples:
		398
		399	=head3 Don't use callbacks.
		400
		401	When your program never uses any callbacks, then there will never be any
		402	notifications you need to take care of, and therefore no need to worry
		403	about all this.
		404
		405	You can achieve a great deal by explicitly waiting for events, or using
		406	barriers and flush calls. In many programs, there is no need at all to
		407	tinker with asynchronous events.
		408
		409	=head3 Use AnyEvent
		410
		411	This module automatically registers a watcher that invokes all outstanding
		412	event callbacks when AnyEvent is initialised (and block asynchronous
		413	interruptions). Using this mode of operations is the safest and most
		414	recommended one.
		415
		416	To use this, simply use AnyEvent and this module normally, make sure you
		417	have an event loop running:
		418
		419	use Gtk2 -init;
		420	use AnyEvent;
		421
		422	# initialise AnyEvent, by creating a watcher, or:
		423	AnyEvent::detect;
		424
		425	my $e = $queue->enqueue_marker;
		426	$e->cb (sub {
		427	warn "opencl is finished\n";
		428	})
		429
		430	main Gtk2;
		431
		432	Note that this module will not initialise AnyEvent for you. Before
		433	AnyEvent is initialised, the module will asynchronously interrupt perl
		434	instead. To avoid any surprises, it's best to explicitly initialise
		435	AnyEvent.
		436
		437	You can temporarily enable asynchronous interruptions (see next paragraph)
		438	by calling C<$OpenCL::INTERRUPT->unblock> and disable them again by
		439	calling C<$OpenCL::INTERRUPT->block>.
		440
		441	=head3 Let yourself be interrupted at any time
		442
		443	This mode is the default unless AnyEvent is loaded and initialised. In
		444	this mode, OpenCL asynchronously interrupts a running perl program. The
		445	emphasis is on both I<asynchronously> and I<running> here.
		446
		447	Asynchronously means that perl might execute your callbacks at any
		448	time. For example, in the following code (I<THAT YOU SHOULD NOT COPY>),
		449	the C<until> loop following the marker call will be interrupted by the
		450	callback:
		451
		452	my $e = $queue->enqueue_marker;
		453	my $flag;
		454	$e->cb (sub { $flag = 1 });
		455	1 until $flag;
		456	# $flag is now 1
		457
		458	The reason why you shouldn't blindly copy the above code is that
		459	busy waiting is a really really bad thing, and really really bad for
		460	performance.
		461
		462	While at first this asynchronous business might look exciting, it can be
		463	really hard, because you need to be prepared for the callback code to be
		464	executed at any time, which limits the amount of things the callback code
		465	can do safely.
		466
		467	This can be mitigated somewhat by using C<<
		468	$OpenCL::INTERRUPT->scope_block >> (see the L<Async::Interrupt>
		469	documentation for details).
		470
		471	The other problem is that your program must be actively I<running> to be
		472	interrupted. When you calculate stuff, your program is running. When you
		473	hang in some C functions or other block execution (by calling C<sleep>,
		474	C<select>, running an event loop and so on), your program is waiting, not
		475	running.
		476
		477	One way around that would be to attach a read watcher to your event loop,
		478	listening for events on C<< $OpenCL::INTERRUPT->pipe_fileno >>, using a
		479	dummy callback (C<sub { }>) to temporarily execute some perl code.
		480
		481	That is then awfully close to using the built-in AnyEvent support above,
		482	though, so consider that one instead.
		483
		484	=head3 Be creative
		485
		486	OpenCL exports the L<Async::Interrupt> object it uses in the global
		487	variable C<$OpenCL::INTERRUPT>. You can configure it in any way you like.
		488
		489	So if you want to feel like a real pro, err, wait, if you feel no risk
		490	menas no fun, you can experiment by implementing your own mode of
		491	operations.
		492
381	=cut	493	=cut
382		494
383	package OpenCL;	495	package OpenCL;
384		496
385	use common::sense;	497	use common::sense;
		498	use Async::Interrupt ();
		499
		500	our $POLL_FUNC; # set by XS
386		501
387	BEGIN {	502	BEGIN {
388	our $VERSION = '0.96';	503	our $VERSION = '0.97';
389		504
390	require XSLoader;	505	require XSLoader;
391	XSLoader::load (__PACKAGE__, $VERSION);	506	XSLoader::load (__PACKAGE__, $VERSION);
392		507
393	@OpenCL::Platform::ISA =	508	@OpenCL::Platform::ISA =
…		…
440		555
441	Returns all available OpenCL::Platform objects.	556	Returns all available OpenCL::Platform objects.
442		557
443	L<http://www.khronos.org/registry/cl/sdk/1.1/docs/man/xhtml/clGetPlatformIDs.html>	558	L<http://www.khronos.org/registry/cl/sdk/1.1/docs/man/xhtml/clGetPlatformIDs.html>
444		559
445	=item $ctx = OpenCL::context_from_type $properties, $type = OpenCL::DEVICE_TYPE_DEFAULT, $notify = undef	560	=item $ctx = OpenCL::context_from_type $properties, $type = OpenCL::DEVICE_TYPE_DEFAULT, $callback->($err, $pvt) = $print_stderr
446		561
447	Tries to create a context from a default device and platform - never worked for me.	562	Tries to create a context from a default device and platform type - never worked for me.
448		563
449	L<http://www.khronos.org/registry/cl/sdk/1.1/docs/man/xhtml/clCreateContextFromType.html>	564	L<http://www.khronos.org/registry/cl/sdk/1.1/docs/man/xhtml/clCreateContextFromType.html>
450		565
		566	=item $ctx = OpenCL::context $properties, \@devices, $callback->($err, $pvt) = $print_stderr)
		567
		568	Create a new OpenCL::Context object using the given device object(s). This
		569	function isn't implemented yet, use C<< $platform->context >> instead.
		570
		571	L<http://www.khronos.org/registry/cl/sdk/1.1/docs/man/xhtml/clCreateContext.html>
		572
451	=item OpenCL::wait_for_events $wait_events...	573	=item OpenCL::wait_for_events $wait_events...
452		574
453	Waits for all events to complete.	575	Waits for all events to complete.
454		576
455	L<http://www.khronos.org/registry/cl/sdk/1.1/docs/man/xhtml/clWaitForEvents.html>	577	L<http://www.khronos.org/registry/cl/sdk/1.1/docs/man/xhtml/clWaitForEvents.html>
		578
		579	=item OpenCL::poll
		580
		581	Checks if there are any outstanding events (see L<EVENT SYSTEM>) and
		582	invokes their callbacks.
		583
		584	=item $OpenCL::INTERRUPT
		585
		586	The L<Async::Interrupt> object used to signal asynchronous events (see
		587	L<EVENT SYSTEM>).
		588
		589	=cut
		590
		591	our $INTERRUPT = new Async::Interrupt c_cb => [$POLL_FUNC, 0];
		592
		593	&_eq_initialise ($INTERRUPT->signal_func);
		594
		595	=item $OpenCL::WATCHER
		596
		597	The L<AnyEvent> watcher object used to watch for asynchronous events (see
		598	L<EVENT SYSTEM>). This variable is C<undef> until L<AnyEvent> has been
		599	loaded I<and> initialised (e.g. by calling C<AnyEvent::detect>).
		600
		601	=cut
		602
		603	our $WATCHER;
		604
		605	sub _init_anyevent {
		606	$INTERRUPT->block;
		607	$WATCHER = AE::io ($INTERRUPT->pipe_fileno, 0, sub { $INTERRUPT->handle });
		608	}
		609
		610	if (defined $AnyEvent::MODEL) {
		611	_init_anyevent;
		612	} else {
		613	push @AnyEvent::post_detect, \&_init_anyevent;
		614	}
456		615
457	=back	616	=back
458		617
459	=head2 THE OpenCL::Object CLASS	618	=head2 THE OpenCL::Object CLASS
460		619
…		…
472	C<IV> type in your code and cast that to the correct type.	631	C<IV> type in your code and cast that to the correct type.
473		632
474	=cut	633	=cut
475		634
476	sub OpenCL::Object::id {	635	sub OpenCL::Object::id {
		636	ref $_[0] eq "SCALAR"
477	${$_[0]}	637	? ${ $_[0] }
		638	: $_[0][0]
478	}	639	}
479		640
480	=back	641	=back
481		642
482	=head2 THE OpenCL::Platform CLASS	643	=head2 THE OpenCL::Platform CLASS
…		…
485		646
486	=item @devices = $platform->devices ($type = OpenCL::DEVICE_TYPE_ALL)	647	=item @devices = $platform->devices ($type = OpenCL::DEVICE_TYPE_ALL)
487		648
488	Returns a list of matching OpenCL::Device objects.	649	Returns a list of matching OpenCL::Device objects.
489		650
490	=item $ctx = $platform->context_from_type ($properties, $type = OpenCL::DEVICE_TYPE_DEFAULT, $notify = undef)	651	=item $ctx = $platform->context_from_type ($properties, $type = OpenCL::DEVICE_TYPE_DEFAULT, $callback->($err, $pvt) = $print_stderr)
491		652
492	Tries to create a context. Never worked for me, and you need devices explicitly anyway.	653	Tries to create a context. Never worked for me, and you need devices explicitly anyway.
493		654
494	L<http://www.khronos.org/registry/cl/sdk/1.1/docs/man/xhtml/clCreateContextFromType.html>	655	L<http://www.khronos.org/registry/cl/sdk/1.1/docs/man/xhtml/clCreateContextFromType.html>
495		656
496	=item $ctx = $platform->context ($properties = undef, @$devices, $notify = undef)	657	=item $ctx = $platform->context ($properties, \@devices, $callback->($err, $pvt) = $print_stderr)
497		658
498	Create a new OpenCL::Context object using the given device object(s)- a	659	Create a new OpenCL::Context object using the given device object(s)- a
499	CL_CONTEXT_PLATFORM property is supplied automatically.	660	CL_CONTEXT_PLATFORM property is supplied automatically.
500		661
501	L<http://www.khronos.org/registry/cl/sdk/1.1/docs/man/xhtml/clCreateContext.html>	662	L<http://www.khronos.org/registry/cl/sdk/1.1/docs/man/xhtml/clCreateContext.html>
…		…
841	=cut	1002	=cut
842		1003
843	sub OpenCL::Context::build_program {	1004	sub OpenCL::Context::build_program {
844	my ($self, $prog, $options) = @_;	1005	my ($self, $prog, $options) = @_;
845		1006
		1007	require Carp;
		1008
846	$prog = $self->program_with_source ($prog)	1009	$prog = $self->program_with_source ($prog)
847	unless ref $prog;	1010	unless ref $prog;
848		1011
		1012	# we build separately per device so we instantly know which one failed
849	for my $dev ($self->devices) {	1013	for my $dev ($self->devices) {
850	eval { $prog->build ($dev, $options); 1 }	1014	eval { $prog->build ([$dev], $options); 1 }
851	or Carp::croak "Building OpenCL program for device '" . $dev->name . "' failed:\n"	1015	or Carp::croak ("Building OpenCL program for device '" . $dev->name . "' failed:\n"
852	. $prog->build_log ($dev);	1016	. $prog->build_log ($dev));
853	}	1017	}
854		1018
855	$prog	1019	$prog
856	}	1020	}
857		1021
…		…
1080		1244
1081	=item $ev = $queue->enqueue_task ($kernel, $wait_events...)	1245	=item $ev = $queue->enqueue_task ($kernel, $wait_events...)
1082		1246
1083	L<http://www.khronos.org/registry/cl/sdk/1.1/docs/man/xhtml/clEnqueueTask.html>	1247	L<http://www.khronos.org/registry/cl/sdk/1.1/docs/man/xhtml/clEnqueueTask.html>
1084		1248
1085	=item $ev = $queue->enqueue_nd_range_kernel ($kernel, @$global_work_offset, @$global_work_size, @$local_work_size, $wait_events...)	1249	=item $ev = $queue->enqueue_nd_range_kernel ($kernel, \@global_work_offset, \@global_work_size, \@local_work_size, $wait_events...)
1086		1250
1087	Enqueues a kernel execution.	1251	Enqueues a kernel execution.
1088		1252
1089	@$global_work_size must be specified as a reference to an array of	1253	\@global_work_size must be specified as a reference to an array of
1090	integers specifying the work sizes (element counts).	1254	integers specifying the work sizes (element counts).
1091		1255
1092	@$global_work_offset must be either C<undef> (in which case all offsets	1256	\@global_work_offset must be either C<undef> (in which case all offsets
1093	are C<0>), or a reference to an array of work offsets, with the same number	1257	are C<0>), or a reference to an array of work offsets, with the same number
1094	of elements as @$global_work_size.	1258	of elements as \@global_work_size.
1095		1259
1096	@$local_work_size must be either C<undef> (in which case the	1260	\@local_work_size must be either C<undef> (in which case the
1097	implementation is supposed to choose good local work sizes), or a	1261	implementation is supposed to choose good local work sizes), or a
1098	reference to an array of local work sizes, with the same number of	1262	reference to an array of local work sizes, with the same number of
1099	elements as @$global_work_size.	1263	elements as \@global_work_size.
1100		1264
1101	L<http://www.khronos.org/registry/cl/sdk/1.1/docs/man/xhtml/clEnqueueNDRangeKernel.html>	1265	L<http://www.khronos.org/registry/cl/sdk/1.1/docs/man/xhtml/clEnqueueNDRangeKernel.html>
1102		1266
1103	=item $ev = $queue->enqueue_acquire_gl_objects ([object, ...], $wait_events...)	1267	=item $ev = $queue->enqueue_acquire_gl_objects ([object, ...], $wait_events...)
1104		1268
…		…
1349		1513
1350	=head2 THE OpenCL::Program CLASS	1514	=head2 THE OpenCL::Program CLASS
1351		1515
1352	=over 4	1516	=over 4
1353		1517
1354	=item $program->build ($device, $options = "")	1518	=item $program->build (\@devices = undef, $options = "", $cb->($program) = undef)
1355		1519
1356	Tries to build the program with the given options. See also the	1520	Tries to build the program with the given options. See also the
1357	C<$ctx->build> convenience function.	1521	C<$ctx->build> convenience function.
1358		1522
		1523	If a callback is specified, then it will be called when compilation is
		1524	finished. Note that many OpenCL implementations block your program while
		1525	compiling whether you use a callback or not. See C<build_async> if you
		1526	want to make sure the build is done in the background.
		1527
		1528	Note that some OpenCL implementations atc up badly, and don't call the
		1529	callback in some error cases (but call it in others). This implementation
		1530	assumes the callback will always be called, and leaks memory if this is
		1531	not so. So best make sure you don't pass in invalid values.
		1532
1359	L<http://www.khronos.org/registry/cl/sdk/1.1/docs/man/xhtml/clBuildProgram.html>	1533	L<http://www.khronos.org/registry/cl/sdk/1.1/docs/man/xhtml/clBuildProgram.html>
		1534
		1535	=item $program->build_async (\@devices = undef, $options = "", $cb->($program) = undef)
		1536
		1537	Similar to C<< ->build >>, except it starts a thread, and never fails (you
		1538	need to check the compilation status form the callback, or by polling).
1360		1539
1361	=item $packed_value = $program->build_info ($device, $name)	1540	=item $packed_value = $program->build_info ($device, $name)
1362		1541
1363	Similar to C<< $platform->info >>, but returns build info for a previous	1542	Similar to C<< $platform->info >>, but returns build info for a previous
1364	build attempt for the given device.	1543	build attempt for the given device.
…		…
1541		1720
1542	Waits for the event to complete.	1721	Waits for the event to complete.
1543		1722
1544	L<http://www.khronos.org/registry/cl/sdk/1.1/docs/man/xhtml/clWaitForEvents.html>	1723	L<http://www.khronos.org/registry/cl/sdk/1.1/docs/man/xhtml/clWaitForEvents.html>
1545		1724
		1725	=item $ev->cb ($exec_callback_type, $callback->($event, $event_command_exec_status))
		1726
		1727	Adds a callback to the callback stack for the given event type. There is
		1728	no way to remove a callback again.
		1729
		1730	L<http://www.khronos.org/registry/cl/sdk/1.1/docs/man/xhtml/clSetEventCallback.html>
		1731
1546	=item $packed_value = $ev->info ($name)	1732	=item $packed_value = $ev->info ($name)
1547		1733
1548	See C<< $platform->info >> for details.	1734	See C<< $platform->info >> for details.
1549		1735
1550	L<http://www.khronos.org/registry/cl/sdk/1.1/docs/man/xhtml/clGetEventInfo.html>	1736	L<http://www.khronos.org/registry/cl/sdk/1.1/docs/man/xhtml/clGetEventInfo.html>
…		…
1610		1796
1611	=over 4	1797	=over 4
1612		1798
1613	=item $ev->set_status ($execution_status)	1799	=item $ev->set_status ($execution_status)
1614		1800
		1801	Sets the execution status of the user event. Can only be called once,
		1802	either with OpenCL::COMPLETE or a negative number as status.
		1803
1615	L<http://www.khronos.org/registry/cl/sdk/1.1/docs/man/xhtml/clSetUserEventStatus.html>	1804	L<http://www.khronos.org/registry/cl/sdk/1.1/docs/man/xhtml/clSetUserEventStatus.html>
1616		1805
1617	=back	1806	=back
1618		1807
1619	=cut	1808	=cut

Diff Legend

-–
+Removed lines
-+
+Added lines
-<
+Changed lines
->
+Changed lines

Comparing OpenCL/OpenCL.pm (file contents): Revision 1.53 by root, Tue Apr 24 14:30:57 2012 UTC vs. Revision 1.57 by root, Tue Apr 24 23:58:34 2012 UTC

Diff Legend

Comparing OpenCL/OpenCL.pm (file contents):
Revision 1.53 by root, Tue Apr 24 14:30:57 2012 UTC vs.
Revision 1.57 by root, Tue Apr 24 23:58:34 2012 UTC