CODEWITHSUNDEEP
parallel-processingopencl
khalil
khalil

Reputation: 25

multiple kernel invocations in opencl

When a kernel argument is set using clSetKernelArg(memory_object) and call the kernel multiple times using clEnqueueNDKernelRangeKernel( ), will the memory_object that you had previously set as the argument in clSetKernelArg( ) for the kernel, be transferred from the host to the GPU, for every invocation of the kernel? 

    status = clSetKernelArg(kernel, 0, sizeof(cl_mem), &inbuf);



    size_t num_total_wi = M * N;
    size_t num_wg = num_elem/N + ((num_elem/N == 0) ? 0 : 1); 
    size_t num_iter = num_elem/(num_total_wi) + (((num_elem % (num_total_wi)) == 0) ?     0 : 1); 
    for(i = 0; i < num_iter; i++) {
            size_t global_work_size[3] = {num_total_wi, 1, 1};    
            size_t local_ws[3] = {N, 1, 1}; 
            size_t global_offset[3] = {i * num_total_wi, 0, 0}; 

            clEnqueueNDRangeKernel(queue, kernel, 1, global_offset, global_work_size, local_ws, 0, NULL, NULL);

    }

Upvotes: 1

Views: 464

Answers (1)

matthias
matthias

Reputation: 2181

A memory transfer is initiated by a call to clEnqueueWriteBuffer or clEnqueueWriteImage but will most likely be started when clEnqueueNDRangeKernel is called. Once the data is transferred, you can re-use the memory objects as often as you like and it won't be transferred again if you don't state it explicitly again.

Upvotes: 1

Related Questions