Show description

C -I . c -I. –I/opt/AMDAPP/include –L/opt/AMDAPP/lib/x86_64 –lOpenCL Regardless of the platform, a binary executable user_buffer would be deposited locally. Running the application on both platforms, we would get the following result: Number of OpenCL platforms found: 1 Kernel name: hello with arity: 1 About to create command queue and enqueue this kernel... Task has been enqueued successfully! Check passed! How it works… The application created a million of the UserData objects on the host. Refer to the following code snippet: /* Prepare an array of UserData via dynamic memory allocation */ UserData* ud_in = (UserData*) malloc( sizeof(UserData) * DATA_SIZE); // input to device UserData* ud_out = (UserData*) malloc( sizeof(UserData) * DATA_SIZE); // output from device for( int i = 0; i < DATA_SIZE; ++i) { (ud_in + i)->x = i; (ud_in + i)->y = i; (ud_in + i)->z = i; (ud_in + i)->w = 3 * i; } 47 Understanding OpenCL Data Transfer and Partitioning The application then sends it to the device for computation after the program and kernel objects have been initialized, and we assign the recently created UDObj memory object to the kernel as its argument.

I/opt/AMDAPP/include –L/opt/AMDAPP/lib/x86_64 –lOpenCL Regardless of the platform, a binary executable user_buffer would be deposited locally. Running the application on both platforms, we would get the following result: Number of OpenCL platforms found: 1 Kernel name: hello with arity: 1 About to create command queue and enqueue this kernel... Task has been enqueued successfully! Check passed! How it works… The application created a million of the UserData objects on the host. Refer to the following code snippet: /* Prepare an array of UserData via dynamic memory allocation */ UserData* ud_in = (UserData*) malloc( sizeof(UserData) * DATA_SIZE); // input to device UserData* ud_out = (UserData*) malloc( sizeof(UserData) * DATA_SIZE); // output from device for( int i = 0; i < DATA_SIZE; ++i) { (ud_in + i)->x = i; (ud_in + i)->y = i; (ud_in + i)->z = i; (ud_in + i)->w = 3 * i; } 47 Understanding OpenCL Data Transfer and Partitioning The application then sends it to the device for computation after the program and kernel objects have been initialized, and we assign the recently created UDObj memory object to the kernel as its argument.

This API returns a cl_kernel, which represents the kernel object when successful. This API provides the programmer with an option of not transforming every kernel function in the program into actual OpenCL kernel objects ready for execution. But if you wish to simply transform all kernel functions in the program into kernel objects, then clCreateKernelsInProgram is the API to use: cl_int clCreateKernelsInProgram(cl_program program, cl_uint num_kernels, cl_kernel* kernels, cl_uint* num_kernels_ret) You use this API to ask OpenCL to create and load the kernels into the kernels argument, and you hint to the OpenCL compiler how many kernels you're expecting with the num_kernels argument.