How to copy a big array to memory and use it in OpenCL kernel?

Question

I have an array of uint8_t. The size of the array is about 2.000.000. I need to do some calculations on these values, but after I call the kernel and copy the modified values back, it returns only zeros.

I'm creating the array, the "row" and "columns" are int.

uint8_t arrayIn[rows * columns];
uint8_t arrayOut[rows * columns];

I'm creating the cl_mem objects and copy the array data into.

arrayInMem = clCreateBuffer(context, CL_MEM_READ_ONLY, rows * columns * sizeof(uint8_t), NULL, &err);
arrayOutMem = clCreateBuffer(context, CL_MEM_WRITE_ONLY, rows * columns * sizeof(uint8_t), NULL, &err);
err = clEnqueueWriteBuffer(img_cmd_queue, arrayInMem, CL_TRUE, 0, rows * columns * sizeof(uint8_t), arrayIn, 0, NULL, NULL);

Setting the kernel arg like this.

err = clSetKernelArg(kernel, 0, sizeof(cl_mem), (void *)&arrayInMem);
err = clSetKernelArg(kernel, 1, sizeof(cl_mem), (void *)&arrayOutMem);

Reading back to the host the modified array.

err = clEnqueueReadBuffer(img_cmd_queue, arrayOutMem, CL_TRUE, 0, MEM_SIZE * sizeof(uint8_t), arrayOut, 0, NULL, NULL);

The kernel signature look like this:

__kernel void calculate(__global uchar * arrayInKernel, __global uchar * arrayOutKernel){
//do some calculation like this eg. 
//int gid = get_global_id(0);
//arrayOutKernel[gid] = 2 * arrayInKernel[gid];
}

Could somebody help, what am I missing out?

Answer 1

Your code is fine, assuming MEM_SIZE = rows * columns . The argument order in clEnqueueReadBuffer also is correct.

I could imagine that you forgot to call clFinish(img_cmd_queue); after clEnqueueWriteBuffer , clEnqueueNDRangeKernel and clEnqueueReadBuffer and before you check the results in arrayOut . All these commands end up in a queue and without clFinish the queue may be executed after you checked results.