Using only implicit copying - works fine with CPU but not with GPU

solveWithDeviceMemory.java

 public synchronized float[] solveWithDeviceMemory() {
  	// Allocate native (device) memory for the input data
	CLBuffer<Float> bufIn = _context.createFloatBuffer(CLMem.Usage.Input, _input.length);
	// Allocate native (device) memory for the output data
	CLBuffer<Float> bufOut = _context.createFloatBuffer(CLMem.Usage.Output, _input.length);

	// Copy input data directly to device memory
	Pointer<Float> ptrIn = bufIn.map(_queue, CLMem.MapFlags.Write);
	ptrIn.setFloats(_input);
	bufIn.unmap(_queue, ptrIn);

	// Setup the method arguments for the kernel
	_testKernel.setArg(0, bufIn);
	_testKernel.setArg(1, bufOut);
	_testKernel.setArg(2, _input.length);

	// Enqueue execution of the kernel
	CLEvent completion = _testKernel.enqueueNDRange(_queue, new int[] { _input.length });

	// wait for completion of the event
	completion.waitFor();
	
	// map output buffer so that we can read from it
      	Pointer<Float> ptrOut = bufOut.map(_queue, CLMem.MapFlags.Read);
	float[] returnVal = ptrOut.getFloats();
	bufOut.unmap(_queue, ptrOut);
	
	return returnVal;
}