add new interface function for FPGA track close #1479

Optimize general col2im to speed up transpose conv

gpu 1x1 optimise close #1506

modify pridecttime judge

Fix no effect if setting thread count in another thread

Resize feed and fetch variables before infer shape, and reimplement scale kernel according to fluid

remove unimpled gpu op in load fix #1498

Fix compile with opencl

remove mali fix #1498