I hope to use the GPU on a mobile phone to directly and quickly access and synchronize memory data, which the CPU will also read and modify in real-time.

On a mobile phone, this is a simple example, where the printed result is “1 7” instead of “7 7”. It’s known that active synchronization can be achieved using methods like enqueueReadBuffer or enqueueNDRangeKernel, but is there a way to achieve automatic synchronization? Ideally, the GPU should be able to read the real-time updated values of data directly after modification.

   data = (int*)mmap(NULL, 0x1000, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
    data[0] = 1;
    
    cl::Buffer inputBuffer = cl::Buffer(context, CL_MEM_READ_WRITE | CL_MEM_USE_HOST_PTR , sizeof(int) * dataSize, data, &err);
    cl::Buffer outputBuffer = cl::Buffer(context, CL_MEM_WRITE_ONLY, sizeof(int) * dataSize, NULL, &err);
    
    cl::Program::Sources source(1, std::make_pair(kernelSource2, strlen(kernelSource2) + 1));
    cl::Program program(context, source);
    err = program.build(devices);
    cl::Kernel kernel(program, "square", &err);
    kernel.setArg(0, inputBuffer);
    kernel.setArg(1, outputBuffer);
    
    data[0] = 7;
    size_t globalSize = dataSize;
    cl::Event event;
    err = queue.enqueueNDRangeKernel(kernel, cl::NullRange, cl::NDRange(globalSize), cl::NullRange, nullptr, &event);
    int *result = (int*)malloc(sizeof(int) * dataSize);
    queue.enqueueReadBuffer(outputBuffer, CL_TRUE, 0, sizeof(int) * dataSize, result);
    printf("%d %dn", result[0], data[0])