C++ (Cpp) global_size_round_upの例

コード例 #1

ファイル: device_opencl.cpp プロジェクト: fabiosantoscode/emcycles

	void tonemap(DeviceTask& task)
	{
		/* cast arguments to cl types */
		cl_mem d_data = CL_MEM_PTR(const_mem_map["__data"]->device_pointer);
		cl_mem d_rgba = CL_MEM_PTR(task.rgba);
		cl_mem d_buffer = CL_MEM_PTR(task.buffer);
		cl_int d_x = task.x;
		cl_int d_y = task.y;
		cl_int d_w = task.w;
		cl_int d_h = task.h;
		cl_int d_sample = task.sample;
		cl_int d_resolution = task.resolution;
		cl_int d_offset = task.offset;
		cl_int d_stride = task.stride;

		/* sample arguments */
		int narg = 0;
		ciErr = 0;

		ciErr |= clSetKernelArg(ckFilmConvertKernel, narg++, sizeof(d_data), (void*)&d_data);
		ciErr |= clSetKernelArg(ckFilmConvertKernel, narg++, sizeof(d_rgba), (void*)&d_rgba);
		ciErr |= clSetKernelArg(ckFilmConvertKernel, narg++, sizeof(d_buffer), (void*)&d_buffer);

#define KERNEL_TEX(type, ttype, name) \
	ciErr |= set_kernel_arg_mem(ckFilmConvertKernel, &narg, #name);
#include "kernel_textures.h"

		ciErr |= clSetKernelArg(ckFilmConvertKernel, narg++, sizeof(d_sample), (void*)&d_sample);
		ciErr |= clSetKernelArg(ckFilmConvertKernel, narg++, sizeof(d_resolution), (void*)&d_resolution);
		ciErr |= clSetKernelArg(ckFilmConvertKernel, narg++, sizeof(d_x), (void*)&d_x);
		ciErr |= clSetKernelArg(ckFilmConvertKernel, narg++, sizeof(d_y), (void*)&d_y);
		ciErr |= clSetKernelArg(ckFilmConvertKernel, narg++, sizeof(d_w), (void*)&d_w);
		ciErr |= clSetKernelArg(ckFilmConvertKernel, narg++, sizeof(d_h), (void*)&d_h);
		ciErr |= clSetKernelArg(ckFilmConvertKernel, narg++, sizeof(d_offset), (void*)&d_offset);
		ciErr |= clSetKernelArg(ckFilmConvertKernel, narg++, sizeof(d_stride), (void*)&d_stride);

		opencl_assert(ciErr);

		size_t workgroup_size;

		clGetKernelWorkGroupInfo(ckFilmConvertKernel, cdDevice,
			CL_KERNEL_WORK_GROUP_SIZE, sizeof(size_t), &workgroup_size, NULL);
	
		workgroup_size = max(sqrt((double)workgroup_size), 1.0);

		size_t local_size[2] = {workgroup_size, workgroup_size};
		size_t global_size[2] = {global_size_round_up(local_size[0], d_w), global_size_round_up(local_size[1], d_h)};

		/* run kernel */
		ciErr = clEnqueueNDRangeKernel(cqCommandQueue, ckFilmConvertKernel, 2, NULL, global_size, local_size, 0, NULL, NULL);
		opencl_assert(ciErr);
		opencl_assert(clFinish(cqCommandQueue));
	}

コード例 #2

ファイル: device_opencl.cpp プロジェクト: scorpion81/blender-voro

	void path_trace(RenderTile& rtile, int sample)
	{
		/* cast arguments to cl types */
		cl_mem d_data = CL_MEM_PTR(const_mem_map["__data"]->device_pointer);
		cl_mem d_buffer = CL_MEM_PTR(rtile.buffer);
		cl_mem d_rng_state = CL_MEM_PTR(rtile.rng_state);
		cl_int d_x = rtile.x;
		cl_int d_y = rtile.y;
		cl_int d_w = rtile.w;
		cl_int d_h = rtile.h;
		cl_int d_sample = sample;
		cl_int d_offset = rtile.offset;
		cl_int d_stride = rtile.stride;

		/* sample arguments */
		int narg = 0;
		ciErr = 0;

		ciErr |= clSetKernelArg(ckPathTraceKernel, narg++, sizeof(d_data), (void*)&d_data);
		ciErr |= clSetKernelArg(ckPathTraceKernel, narg++, sizeof(d_buffer), (void*)&d_buffer);
		ciErr |= clSetKernelArg(ckPathTraceKernel, narg++, sizeof(d_rng_state), (void*)&d_rng_state);

#define KERNEL_TEX(type, ttype, name) \
	ciErr |= set_kernel_arg_mem(ckPathTraceKernel, &narg, #name);
#include "kernel_textures.h"

		ciErr |= clSetKernelArg(ckPathTraceKernel, narg++, sizeof(d_sample), (void*)&d_sample);
		ciErr |= clSetKernelArg(ckPathTraceKernel, narg++, sizeof(d_x), (void*)&d_x);
		ciErr |= clSetKernelArg(ckPathTraceKernel, narg++, sizeof(d_y), (void*)&d_y);
		ciErr |= clSetKernelArg(ckPathTraceKernel, narg++, sizeof(d_w), (void*)&d_w);
		ciErr |= clSetKernelArg(ckPathTraceKernel, narg++, sizeof(d_h), (void*)&d_h);
		ciErr |= clSetKernelArg(ckPathTraceKernel, narg++, sizeof(d_offset), (void*)&d_offset);
		ciErr |= clSetKernelArg(ckPathTraceKernel, narg++, sizeof(d_stride), (void*)&d_stride);

		opencl_assert(ciErr);

		size_t workgroup_size;

		clGetKernelWorkGroupInfo(ckPathTraceKernel, cdDevice,
			CL_KERNEL_WORK_GROUP_SIZE, sizeof(size_t), &workgroup_size, NULL);
	
		workgroup_size = max(sqrt((double)workgroup_size), 1.0);

		size_t local_size[2] = {workgroup_size, workgroup_size};
		size_t global_size[2] = {global_size_round_up(local_size[0], d_w), global_size_round_up(local_size[1], d_h)};

		/* run kernel */
		ciErr = clEnqueueNDRangeKernel(cqCommandQueue, ckPathTraceKernel, 2, NULL, global_size, local_size, 0, NULL, NULL);
		opencl_assert(ciErr);
		opencl_assert(clFinish(cqCommandQueue));
	}