--- OpenCL/OpenCL.pm 2011/11/15 09:24:40 1.2 +++ OpenCL/OpenCL.pm 2011/11/17 02:54:14 1.11 @@ -1,6 +1,6 @@ =head1 NAME -OpenCL - bindings to, well, OpenCL +OpenCL - Open Computing Language Bindings =head1 SYNOPSIS @@ -8,28 +8,636 @@ =head1 DESCRIPTION -This is an early release which is not useful yet. +This is an early release which might be useful, but hasn't seen much testing. -Enumerate all devices and get contexts for them; +=head2 OpenCL FROM 10000 FEET HEIGHT + +Here is a high level overview of OpenCL: + +First you need to find one or more OpenCL::Platforms (kind of like +vendors) - usually there is only one. + +Each platform gives you access to a number of OpenCL::Device objects, e.g. +your graphics card. + +From a platform and some device(s), you create an OpenCL::Context, which is +a very central object in OpenCL: Once you have a context you can create +most other objects: + +OpenCL::Program objects, which store source code and, after building for a +specific device ("compiling and linking"), also binary programs. For each +kernel function in a program you can then create an OpenCL::Kernel object +which represents basically a function call with argument values. + +OpenCL::Memory objects of various flavours: OpenCL::Buffers objects (flat +memory areas, think array) and OpenCL::Image objects (think 2d or 3d +array) for bulk data and input and output for kernels. + +OpenCL::Sampler objects, which are kind of like texture filter modes in +OpenGL. + +OpenCL::Queue objects - command queues, which allow you to submit memory +reads, writes and copies, as well as kernel calls to your devices. They +also offer a variety of methods to synchronise request execution, for +example with barriers or OpenCL::Event objects. + +OpenCL::Event objects are used to signal when something is complete. + +=head2 HELPFUL RESOURCES + +The OpenCL spec used to develop this module (1.2 spec was available, but +no implementation was available to me :). + + http://www.khronos.org/registry/cl/specs/opencl-1.1.pdf + +OpenCL manpages: + + http://www.khronos.org/registry/cl/sdk/1.1/docs/man/xhtml/ + +=head1 BASIC WORKFLOW + +To get something done, you basically have to do this once (refer to the +examples below for actual code, this is just a high-level description): + +Find some platform (e.g. the first one) and some device(s) (e.g. the first +device of the platform), and create a context from those. + +Create program objects from your OpenCL source code, then build (compile) +the programs for each device you want to run them on. + +Create kernel objects for all kernels you want to use (surprisingly, these +are not device-specific). + +Then, to execute stuff, you repeat these steps, possibly resuing or +sharing some buffers: + +Create some input and output buffers from your context. Set these as +arguments to your kernel. + +Enqueue buffer writes to initialise your input buffers (when not +initialised at creation time). + +Enqueue the kernel execution. + +Enqueue buffer reads for your output buffer to read results. + +=head1 EXAMPLES + +=head2 Enumerate all devices and get contexts for them. + +Best run this once to get a feel for the platforms and devices in your +system. for my $platform (OpenCL::platforms) { - warn $platform->info (OpenCL::PLATFORM_NAME); - warn $platform->info (OpenCL::PLATFORM_EXTENSIONS); + printf "platform: %s\n", $platform->info (OpenCL::PLATFORM_NAME); + printf "extensions: %s\n", $platform->info (OpenCL::PLATFORM_EXTENSIONS); for my $device ($platform->devices) { - warn $device->info (OpenCL::DEVICE_NAME); - my $ctx = $device->context_simple; + printf "+ device: %s\n", $device->info (OpenCL::DEVICE_NAME); + my $ctx = $device->context; # do stuff } } -Get a useful context and a command queue: +=head2 Get a useful context and a command queue. + +This is a useful boilerplate for any OpenCL program that only wants to use +one device, + + my ($platform) = OpenCL::platforms; # find first platform + my ($dev) = $platform->devices; # find first device of platform + my $ctx = $platform->context (undef, [$dev]); # create context out of those + my $queue = $ctx->queue ($dev); # create a command queue for the device + +=head2 Print all supported image formats of a context. + +Best run this once for your context, to see whats available and how to +gather information. + + for my $type (OpenCL::MEM_OBJECT_IMAGE2D, OpenCL::MEM_OBJECT_IMAGE3D) { + print "supported image formats for ", OpenCL::enum2str $type, "\n"; + + for my $f ($ctx->supported_image_formats (0, $type)) { + printf " %-10s %-20s\n", OpenCL::enum2str $f->[0], OpenCL::enum2str $f->[1]; + } + } + +=head2 Create a buffer with some predefined data, read it back synchronously, +then asynchronously. + + my $buf = $ctx->buffer_sv (OpenCL::MEM_COPY_HOST_PTR, "helmut"); + + $queue->enqueue_read_buffer ($buf, 1, 1, 3, my $data); + print "$data\n"; + + my $ev = $queue->enqueue_read_buffer ($buf, 0, 1, 3, my $data); + $ev->wait; + print "$data\n"; # prints "elm" + +=head2 Create and build a program, then create a kernel out of one of its +functions. + + my $src = ' + __kernel void + squareit (__global float *input, __global float *output) + { + size_t id = get_global_id (0); + output [id] = input [id] * input [id]; + } + '; + + my $prog = $ctx->program_with_source ($src); + + # build croaks on compile errors, so catch it and print the compile errors + eval { $prog->build ($dev); 1 } + or die $prog->build_info ($dev, OpenCL::PROGRAM_BUILD_LOG); + + my $kernel = $prog->kernel ("squareit"); + +=head2 Create some input and output float buffers, then call the +'squareit' kernel on them. + + my $input = $ctx->buffer_sv (OpenCL::MEM_COPY_HOST_PTR, pack "f*", 1, 2, 3, 4.5); + my $output = $ctx->buffer (0, OpenCL::SIZEOF_FLOAT * 5); + + # set buffer + $kernel->set_buffer (0, $input); + $kernel->set_buffer (1, $output); + + # execute it for all 4 numbers + $queue->enqueue_nd_range_kernel ($kernel, undef, [4], undef); + + # enqueue a synchronous read + $queue->enqueue_read_buffer ($output, 1, 0, OpenCL::SIZEOF_FLOAT * 4, my $data); + + # print the results: + printf "%s\n", join ", ", unpack "f*", $data; + +=head2 The same enqueue operations as before, but assuming an out-of-order queue, +showing off barriers. + + # execute it for all 4 numbers + $queue->enqueue_nd_range_kernel ($kernel, undef, [4], undef); + + # enqueue a barrier to ensure in-order execution + $queue->enqueue_barrier; + + # enqueue an async read + $queue->enqueue_read_buffer ($output, 0, 0, OpenCL::SIZEOF_FLOAT * 4, my $data); + + # wait for all requests to finish + $queue->finish; + +=head2 The same enqueue operations as before, but assuming an out-of-order queue, +showing off event objects and wait lists. + + # execute it for all 4 numbers + my $ev = $queue->enqueue_nd_range_kernel ($kernel, undef, [4], undef); + + # enqueue an async read + $ev = $queue->enqueue_read_buffer ($output, 0, 0, OpenCL::SIZEOF_FLOAT * 4, my $data, $ev); + + # wait for the last event to complete + $ev->wait; + +=head1 DOCUMENTATION + +=head2 BASIC CONVENTIONS + +This is not a 1:1 C-style translation of OpenCL to Perl - instead I +attempted to make the interface as type-safe as possible and introducing +object syntax where it makes sense. There are a number of important +differences between the OpenCL C API and this module: + +=over 4 + +=item * Object lifetime managament is automatic - there is no need +to free objects explicitly (C), the release function +is called automatically once all Perl references to it go away. + +=item * OpenCL uses CamelCase for function names (C), +while this module uses underscores as word separator and often leaves out +prefixes (C<< $platform->info >>). + +=item * OpenCL often specifies fixed vector function arguments as short +arrays (C), while this module explicitly expects the +components as separate arguments- + +=item * Where possible, one of the pitch values is calculated from the +perl scalar length and need not be specified. + +=item * When enqueuing commands, the wait list is specified by adding +extra arguments to the function - anywhere a C<$wait_events...> argument +is documented this can be any number of event objects. + +=item * When enqueuing commands, if the enqueue method is called in void +context, no event is created. In all other contexts an event is returned +by the method. + +=item * This module expects all functions to return C. If any +other status is returned the function will throw an exception, so you +don't normally have to to any error checking. + +=back + +=head2 PERL AND OPENCL TYPES + +This handy(?) table lists OpenCL types and their perl, PDL and pack/unpack +format equivalents: + + OpenCL perl PDL pack/unpack + char IV - c + uchar IV byte C + short IV short s + ushort IV ushort S + int IV long? l + uint IV - L + long IV longlong q + ulong IV - Q + float NV float f + half IV ushort S + double NV double d + +=head2 THE OpenCL PACKAGE + +=over 4 + +=item $int = OpenCL::errno + +The last error returned by a function - it's only valid after an error occured +and before calling another OpenCL function. + +=item $str = OpenCL::err2str $errval + +Comverts an error value into a human readable string. + +=item $str = OpenCL::enum2str $enum + +Converts most enum values (inof parameter names, image format constants, +object types, addressing and filter modes, command types etc.) into a +human readbale string. When confronted with some random integer it can be +very helpful to pass it through this function to maybe get some readable +string out of it. + +=item @platforms = OpenCL::platforms + +Returns all available OpenCL::Platform objects. + +L + +=item $ctx = OpenCL::context_from_type $properties, $type = OpenCL::DEVICE_TYPE_DEFAULT, $notify = undef + +Tries to create a context from a default device and platform - never worked for me. + +L + +=item OpenCL::wait_for_events $wait_events... + +Waits for all events to complete. + +L + +=back + +=head2 THE OpenCL::Platform CLASS + +=over 4 + +=item $packed_value = $platform->info ($name) + +Calls C and returns the packed, raw value - for +strings, this will be the string, for other values you probably need to +use the correct C. This might get improved in the future. Hopefully. + +L + +=item @devices = $platform->devices ($type = OpenCL::DEVICE_TYPE_ALL) + +Returns a list of matching OpenCL::Device objects. + +=item $ctx = $platform->context_from_type ($properties, $type = OpenCL::DEVICE_TYPE_DEFAULT, $notify = undef) + +Tries to create a context. Never worked for me, and you need devices explitly anyway. + +L + +=item $ctx = $device->context ($properties = undef, @$devices, $notify = undef) + +Create a new OpenCL::Context object using the given device object(s)- a +CL_CONTEXT_PLATFORM property is supplied automatically. + +L + +=back + +=head2 THE OpenCL::Device CLASS + +=over 4 + +=item $packed_value = $device->info ($name) + +See C<< $platform->info >> for details. + +L + +=back + +=head2 THE OpenCL::Context CLASS + +=over 4 + +=item $packed_value = $ctx->info ($name) + +See C<< $platform->info >> for details. + +L + +=item $queue = $ctx->queue ($device, $properties) + +Create a new OpenCL::Queue object from the context and the given device. + +L + +=item $ev = $ctx->user_event + +Creates a new OpenCL::UserEvent object. + +L + +=item $buf = $ctx->buffer ($flags, $len) + +Creates a new OpenCL::Buffer object with the given flags and octet-size. + +L + +=item $buf = $ctx->buffer_sv ($flags, $data) + +Creates a new OpenCL::Buffer object and initialise it with the given data values. + +=item $img = $ctx->image2d ($flags, $channel_order, $channel_type, $width, $height, $data) + +Creates a new OpenCL::Image2D object and optionally initialises it with the given data values. + +L + +=item $img = $ctx->image3d ($flags, $channel_order, $channel_type, $width, $height, $depth, $slice_pitch, $data) - my $dev = ((OpenCL::platforms)[0]->devices)[0]; - my $ctx = $dev->context_simple; - my $queue = $ctx->command_queue_simple ($dev); +Creates a new OpenCL::Image3D object and optionally initialises it with the given data values. + +L + +=item @formats = $ctx->supported_image_formats ($flags, $image_type) + +Returns a list of matching image formats - each format is an arrayref with +two values, $channel_order and $channel_type, in it. + +L + +=item $sampler = $ctx->sampler ($normalized_coords, $addressing_mode, $filter_mode) + +Creates a new OpenCL::Sampler object. + +L + +=item $program = $ctx->program_with_source ($string) + +Creates a new OpenCL::Program object from the given source code. + +L + +=back + +=head2 THE OpenCL::Queue CLASS + +An OpenCL::Queue represents an execution queue for OpenCL. You execute +requests by calling their respective C method and waitinf for +it to complete in some way. + +All the enqueue methods return an event object that can be used to wait +for completion, unless the method is called in void context, in which case +no event object is created. + +They also allow you to specify any number of other event objects that this +request has to wait for before it starts executing, by simply passing the +event objects as extra parameters to the enqueue methods. + +Queues execute in-order by default, without any parallelism, so in most +cases (i.e. you use only one queue) it's not necessary to wait for or +create event objects. + +=over 4 + +=item $packed_value = $ctx->info ($name) + +See C<< $platform->info >> for details. + +L + +=item $ev = $queue->enqueue_read_buffer ($buffer, $blocking, $offset, $len, $data, $wait_events...) + +Reads data from buffer into the given string. + +L + +=item $ev = $queue->enqueue_write_buffer ($buffer, $blocking, $offset, $data, $wait_events...) + +Writes data to buffer from the given string. + +L + +=item $ev = $queue->enqueue_copy_buffer ($src, $dst, $src_offset, $dst_offset, $len, $wait_events...) + +L + +=item $ev = $queue->enqueue_read_image ($src, $blocking, $x, $y, $z, $width, $height, $depth, $row_pitch, $slice_pitch, $data, $wait_events...) + +L + +=item $ev = $queue->enqueue_write_image ($src, $blocking, $x, $y, $z, $width, $height, $depth, $row_pitch, $data, $wait_events...) + +L + +=item $ev = $queue->enqueue_copy_buffer_rect ($src, $dst, $src_x, $src_y, $src_z, $dst_x, $dst_y, $dst_z, $width, $height, $depth, $src_row_pitch, $src_slice_pitch, 4dst_row_pitch, $dst_slice_pitch, $ait_event...) + +Yeah. + +L + +=item $ev = $queue->enqueue_copy_buffer_to_image (OpenCL::Buffer src, OpenCL::Image dst, size_t src_offset, size_t dst_x, size_t dst_y, size_t dst_z, size_t width, size_t height, size_t depth, ...) + +L. + +=item $ev = $queue->enqueue_copy_image (OpenCL::Image src, OpenCL::Buffer dst, size_t src_x, size_t src_y, size_t src_z, size_t dst_x, size_t dst_y, size_t dst_z, size_t width, size_t height, size_t depth, ...) + +L + +=item $ev = $queue->enqueue_copy_image_to_buffer (OpenCL::Image src, OpenCL::Buffer dst, size_t src_x, size_t src_y, size_t src_z, size_t width, size_t height, size_t depth, size_t dst_offset, ...) + +L + +=item $ev = $queue->enqueue_task ($kernel, $wait_events...) + +L + +=item $ev = $queue->enqueue_nd_range_kernel ($kernel, @$global_work_offset, @$global_work_size, @$local_work_size, $wait_events...) + +Enqueues a kernel execution. + +@$global_work_size must be specified as a reference to an array of +integers specifying the work sizes (element counts). + +@$global_work_offset must be either C (in which case all offsets +are C<0>), or a reference to an array of work offsets, with the same number +of elements as @$global_work_size. + +@$local_work_size must be either C (in which case the +implementation is supposed to choose good local work sizes), or a +reference to an array of local work sizes, with the same number of +elements as @$global_work_size. + +L + +=item $ev = $queue->enqueue_marker + +L + +=item $ev = $queue->enqueue_wait_for_events ($wait_events...) + +L + +=item $queue->enqueue_barrier + +L + +=item $queue->flush + +L + +=item $queue->finish + +L + +=back + +=head2 THE OpenCL::Memory CLASS + +This the superclass of all memory objects - OpenCL::Buffer, OpenCL::Image, +OpenCL::Image2D and OpenCL::Image3D. The subclasses of this class +currently only exist to allow type-checking. + +=over 4 + +=item $packed_value = $memory->info ($name) + +See C<< $platform->info >> for details. + +L + +=back + +=head2 THE OpenCL::Sampler CLASS =over 4 +=item $packed_value = $sampler->info ($name) + +See C<< $platform->info >> for details. + +L + +=back + +=head2 THE OpenCL::Program CLASS + +=over 4 + +=item $packed_value = $program->info ($name) + +See C<< $platform->info >> for details. + +L + +=item $program->build ($device, $options = "") + +Tries to build the program with the givne options. + +L + +=item $packed_value = $program->build_info ($device, $name) + +Similar to C<< $platform->info >>, but returns build info for a previous +build attempt for the given device. + +L + +=item $kernel = $program->kernel ($function_name) + +Creates an OpenCL::Kernel object out of the named C<__kernel> function in +the program. + +L + +=back + +=head2 THE OpenCL::Kernel CLASS + +=over 4 + +=item $packed_value = $kernel->info ($name) + +See C<< $platform->info >> for details. + +L + +=item $kernel->set_TYPE ($index, $value) + +This is a family of methods to set the kernel argument with the number C<$index> to the give C<$value>. + +TYPE is one of C, C, C, C, C, C, +C, C, C, C, C, C, C, +C, C, C or C. + +Chars and integers (including the half type) are specified as integers, +float and double as floating point values, memory/buffer/image2d/image3d +must be an object of that type or C, and sampler and event must be +objects of that type. + +L + +=back + +=head2 THE OpenCL::Event CLASS + +This is the superclass for all event objects (including OpenCL::UserEvent +objects). + +=over 4 + +=item $packed_value = $ev->info ($name) + +See C<< $platform->info >> for details. + +L + +=item $ev->wait + +Waits for the event to complete. + +L + +=back + +=head2 THE OpenCL::UserEvent CLASS + +This is a subclass of OpenCL::Event. + +=over 4 + +=item $ev->set_status ($execution_status) + +L + +=back + =cut package OpenCL; @@ -37,16 +645,22 @@ use common::sense; BEGIN { - our $VERSION = '0.01'; + our $VERSION = '0.03'; require XSLoader; XSLoader::load (__PACKAGE__, $VERSION); + + @OpenCL::Buffer::ISA = + @OpenCL::Image::ISA = OpenCL::Memory::; + + @OpenCL::Image2D::ISA = + @OpenCL::Image3D::ISA = OpenCL::Image::; + + @OpenCL::UserEvent::ISA = OpenCL::Event::; } 1; -=back - =head1 AUTHOR Marc Lehmann