NAME OpenCL - Open Computing Language Bindings SYNOPSIS use OpenCL; DESCRIPTION This is an early release which might be useful, but hasn't seen much testing. 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/ EXAMPLES Enumerate all devices and get contexts for them. for my $platform (OpenCL::platforms) { warn $platform->info (OpenCL::PLATFORM_NAME); warn $platform->info (OpenCL::PLATFORM_EXTENSIONS); for my $device ($platform->devices) { warn $device->info (OpenCL::DEVICE_NAME); my $ctx = $device->context_simple; # do stuff } } Get a useful context and a command queue. my $dev = ((OpenCL::platforms)[0]->devices)[0]; my $ctx = $dev->context_simple; my $queue = $ctx->command_queue_simple ($dev); Print all supported image formats of a context. for my $type (OpenCL::MEM_OBJECT_IMAGE2D, OpenCL::MEM_OBJECT_IMAGE3D) { say "supported image formats for ", OpenCL::enum2str $type; for my $f ($ctx->supported_image_formats (0, $type)) { printf " %-10s %-20s\n", OpenCL::enum2str $f->[0], OpenCL::enum2str $f->[1]; } } 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); warn $data; my $ev = $queue->enqueue_read_buffer ($buf, 0, 1, 3, my $data); $ev->wait; warn $data; 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); eval { $prog->build ($dev); 1 } or die $prog->build_info ($dev, OpenCL::PROGRAM_BUILD_LOG); my $kernel = $prog->kernel ("squareit"); Create some input and output float buffers, then call squareit 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: say join ", ", unpack "f*", $data; 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; 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 $ev->wait; DOCUMENTATION BASIC CONVENTIONS This is not a 1:1 C-style translation of OpenCL to Perl - instead I attempted to make the interface as type-safe as possible and introducing object syntax where it makes sense. There are a number of important differences between the OpenCL C API and this module: * Object lifetime managament is automatic - there is no need to free objects explicitly ("clReleaseXXX"), the release function is called automatically once all Perl references to it go away. * OpenCL uses CamelCase for function names ("clGetPlatformInfo"), while this module uses underscores as word separator and often leaves out prefixes ("$platform->info"). * OpenCL often specifies fixed vector function arguments as short arrays ("size_t origin[3]"), while this module explicitly expects the components as separate arguments- * Where possible, the row_pitch value is calculated from the perl scalar length and need not be specified. * When enqueuing commands, the wait list is specified by adding extra arguments to the function - everywhere a "$wait_events..." argument is documented this can be any number of event objects. * 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. * This module expects all functions to return "CL_SUCCESS". If any other status is returned the function will throw an exception, so you don't normally have to to any error checking. PERL AND OPENCL TYPES This handy(?) table lists OpenCL types and their perl and pack/unpack format equivalents: OpenCL perl pack/unpack char IV c uchar IV C short IV s ushort IV S int IV l uint IV L long IV q ulong IV Q float NV f half IV S double NV d THE OpenCL PACKAGE $int = OpenCL::errno The last error returned by a function - it's only changed on errors. $str = OpenCL::err2str $errval Comverts an error value into a human readable string. $str = OpenCL::err2str $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. @platforms = OpenCL::platforms Returns all available OpenCL::Platform objects. $ctx = OpenCL::context_from_type_simple $type = OpenCL::DEVICE_TYPE_DEFAULT Tries to create a context from a default device and platform - never worked for me. OpenCL::wait_for_events $wait_events... Waits for all events to complete. THE OpenCL::Platform CLASS $packed_value = $platform->info ($name) Calls "clGetPlatformInfo" and returns the packed, raw value - for strings, this will be the string, for other values you probably need to use the correct "unpack". This might get improved in the future. Hopefully. @devices = $platform->devices ($type = OpenCL::DEVICE_TYPE_ALL) Returns a list of matching OpenCL::Device objects. $ctx = $platform->context_from_type_simple ($type = OpenCL::DEVICE_TYPE_DEFAULT) Tries to create a context. Never worked for me. THE OpenCL::Device CLASS $packed_value = $device->info ($name) See "$platform->info" for details. $ctx = $device->context_simple Convenience function to create a new OpenCL::Context object. THE OpenCL::Context CLASS $packed_value = $ctx->info ($name) See "$platform->info" for details. $queue = $ctx->command_queue_simple ($device) Convenience function to create a new OpenCL::Queue object from the context and the given device. $ev = $ctx->user_event Creates a new OpenCL::UserEvent object. $buf = $ctx->buffer ($flags, $len) Creates a new OpenCL::Buffer object with the given flags and octet-size. $buf = $ctx->buffer_sv ($flags, $data) Creates a new OpenCL::Buffer object and initialise it with the given data values. $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. $img = $ctx->image3d ($flags, $channel_order, $channel_type, $width, $height, $depth, $slice_pitch, $data) Creates a new OpenCL::Image3D object and optionally initialises it with the given data values. @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. $sampler = $ctx->sampler ($normalized_coords, $addressing_mode, $filter_mode) Creates a new OpenCL::Sampler object. $program = $ctx->program_with_source ($string) Creates a new OpenCL::Program object from the given source code. THE OpenCL::Queue CLASS An OpenCL::Queue represents an execution queue for OpenCL. You execute requests by calling their respective "enqueue_xxx" 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. $packed_value = $ctx->info ($name) See "$platform->info" for details. $ev = $queue->enqueue_read_buffer ($buffer, $blocking, $offset, $len, $data, $wait_events...) Reads data from buffer into the given string. $ev = $queue->enqueue_write_buffer ($buffer, $blocking, $offset, $data, $wait_events...) Writes data to buffer from the given string. $ev = $queue->enqueue_copy_buffer ($src, $dst, $src_offset, $dst_offset, $len, $wait_events...) $ev = $queue->enqueue_read_image ($src, $blocking, $x, $y, $z, $width, $height, $depth, $row_pitch, $slice_pitch, $data, $wait_events...) $ev = $queue->enqueue_write_image ($src, $blocking, $x, $y, $z, $width, $height, $depth, $row_pitch, $data, $wait_events...) $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. $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, ...) . $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, ...) $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, ...) $ev = $queue->enqueue_task ($kernel, $wait_events...) $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). @$global_work_offset must be either "undef" (in which case all offsets are 0), or a reference to an array of work offsets, with the same number of elements as @$global_work_size. @$local_work_size must be either "undef" (in which case the implementation is supposed to choose good local work sizes), or a reference to an array of local work sizes, with the same number of elements as @$global_work_size. $ev = $queue->enqueue_marker $ev = $queue->enqueue_wait_for_events ($wait_events...) $queue->enqueue_barrier $queue->flush $queue->finish 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. $packed_value = $memory->info ($name) See "$platform->info" for details. THE OpenCL::Sampler CLASS $packed_value = $sampler->info ($name) See "$platform->info" for details. THE OpenCL::Program CLASS $packed_value = $program->info ($name) See "$platform->info" for details. $program->build ($device, $options = "") Tries to build the program with the givne options. $packed_value = $program->build_info ($device, $name) Similar to "$platform->info", but returns build info for a previous build attempt for the given device. $kernel = $program->kernel ($function_name) Creates an OpenCL::Kernel object out of the named "__kernel" function in the program. THE OpenCL::Kernel CLASS $packed_value = $kernel->info ($name) See "$platform->info" for details. $kernel->set_TYPE ($index, $value) This is a family of methods to set the kernel argument with the number $index to the give $value. TYPE is one of "char", "uchar", "short", "ushort", "int", "uint", "long", "ulong", "half", "float", "double", "memory", "buffer", "image2d", "image3d", "sampler" or "event". 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 "undef", and sampler and event must be objects of that type. THE OpenCL::Event CLASS This is the superclass for all event objects (including OpenCL::UserEvent objects). $packed_value = $ev->info ($name) See "$platform->info" for details. $ev->wait Waits for the event to complete. THE OpenCL::UserEvent CLASS This is a subclass of OpenCL::Event. $ev->set_status ($execution_status) AUTHOR Marc Lehmann http://home.schmorp.de/