Goal:Try to move first loop to GPU
Result:
1,Timing for each single process become unaccessible since everything become paralled.
2,The total time shows greater than default CPU method

Some guess:
1, Maybe I used too many global vars that slowed GPU down
2, Maybe the data set is too small
3, Maybe Matlab already optimized its default loop





########################################################
After some more tests, the experiment is confirmed falled.
Actually, it is paralled on CPU.
Matlab GPU side doesn't support Global vars.