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
|
# File 'lib/BOAST/CKernel.rb', line 255
def build_opencl(options)
require 'opencl_ruby_ffi'
platform = nil
platforms = OpenCL::get_platforms
if options[:platform_vendor] then
platforms.each{ |p|
platform = p if p.vendor.match(options[:platform_vendor])
}
else
platform = platforms.first
end
device = nil
type = options[:device_type] ? options[:device_type] : OpenCL::Device::Type::ALL
devices = platform.devices(type)
if options[:device_name] then
devices.each{ |d|
device = d if d.name.match(options[:device_name])
}
else
device = devices.first
end
@context = OpenCL::create_context([device])
program = @context.create_program_with_source([@code.string])
opts = options[:CLFLAGS]
program.build(:options => options[:CLFLAGS])
if options[:verbose] then
program.build_log.each {|dev,log|
STDERR.puts "#{device.name}: #{log}"
}
end
@queue = @context.create_command_queue(device, :properties => OpenCL::CommandQueue::PROFILING_ENABLE)
@kernel = program.create_kernel(@procedure.name)
run_method = "def self.run(*args)\n raise \"Wrong number of arguments \\\#{args.length} for \#{@procedure.parameters.length}\" if args.length > \#{@procedure.parameters.length+1} or args.length < \#{@procedure.parameters.length}\n params = []\n opts = {}\n opts = args.pop if args.length == \#{@procedure.parameters.length+1}\n @procedure.parameters.each_index { |i|\nif @procedure.parameters[i].dimension then\n if @procedure.parameters[i].direction == :in then\n flags = OpenCL::Mem::Flags::READ_ONLY\n elsif @procedure.parameters[i].direction == :out\n flags = OpenCL::Mem::Flags::WRITE_ONLY\n else\n flags = nil\n end\n if @procedure.parameters[i].texture then\n params[i] = @context.create_image_2D( OpenCL::ImageFormat::new( OpenCL::ChannelOrder::R, OpenCL::ChannelType::UNORM_INT8 ), args[i].size * args[i].element_size, 1, :flags => nil )\n if @procedure.parameters[i].direction == :inout or @procedure.parameters[i].direction == :in then\n @queue.enqueue_write_image( params[i], args[i], :blocking => true )\n end\n else\n params[i] = @context.create_buffer( args[i].size * args[i].element_size, :flags => nil )\n if @procedure.parameters[i].direction == :inout or @procedure.parameters[i].direction == :in then\n @queue.enqueue_write_buffer( params[i], args[i], :blocking => true )\n end\n end\nelse\n if @procedure.parameters[i].type.is_a?(Real) then\n params[i] = OpenCL::Half::new(args[i]) if @procedure.parameters[i].type.size == 2\n params[i] = OpenCL::Float::new(args[i]) if @procedure.parameters[i].type.size == 4\n params[i] = OpenCL::Double::new(args[i]) if @procedure.parameters[i].type.size == 8\n elsif @procedure.parameters[i].type.is_a?(Int) then\n if @procedure.parameters[i].type.signed\n params[i] = OpenCL::Char::new(args[i]) if @procedure.parameters[i].type.size == 1\n params[i] = OpenCL::Short::new(args[i]) if @procedure.parameters[i].type.size == 2\n params[i] = OpenCL::Int::new(args[i]) if @procedure.parameters[i].type.size == 4\n params[i] = OpenCL::Long::new(args[i]) if @procedure.parameters[i].type.size == 8\n else\n params[i] = OpenCL::UChar::new(args[i]) if @procedure.parameters[i].type.size == 1\n params[i] = OpenCL::UShort::new(args[i]) if @procedure.parameters[i].type.size == 2\n params[i] = OpenCL::UInt::new(args[i]) if @procedure.parameters[i].type.size == 4\n params[i] = OpenCL::ULong::new(args[i]) if @procedure.parameters[i].type.size == 8\n end\n else\n params[i] = args[i]\n end\nend\n }\n params.each_index{ |i|\[email protected]_arg(i, params[i])\n }\n event = @queue.enqueue_NDrange_kernel(@kernel, opts[:global_work_size], :local_work_size => opts[:local_work_size])\n @procedure.parameters.each_index { |i|\nif @procedure.parameters[i].dimension then\n if @procedure.parameters[i].texture then\n if @procedure.parameters[i].direction == :inout or @procedure.parameters[i].direction == :out then\n @queue.enqueue_read_image( params[i], args[i], :blocking => true )\n end\n else\n if @procedure.parameters[i].direction == :inout or @procedure.parameters[i].direction == :out then\n @queue.enqueue_read_buffer( params[i], args[i], :blocking => true )\n end\n end\nend\n }\n result = {}\n result[:start] = event.profiling_command_start\n result[:end] = event.profiling_command_end\n result[:duration] = (result[:end] - result[:start])/1000000000.0\n return result\nend\n"
eval run_method
return self
end
|