kernel void compute_accuracy_per_output(
global float *outputs,
global float *expected_outputs,
global float *accuracies,
int count
) {
int index = get_global_id(0)
if (index >= count) {
return }
float expected_output = (float)expected_outputs[index] float output = (float)outputs[index] accuracies[index] = 1.0f - fabs(output - expected_output) / fmax(expected_output, output)}