fix a bug in format_conv_data function

Fix quantize kernel while pad != 0, implement the fusion kernel of dequantize, elementwise_add, batch norm and Relu

faster gemm_int8, max speedup can be 2(int8 / float), add gemm_with_bias and add gemm_with_relu_bias.

fix a bug in feed kernel

Optimize im2col and implement winograd conv, bring 35% speed up for googlenet

add tanh kernel in FPGA v1 fixed #1314

fix #1310  update android development document

add split and transpose kernels for FPGA track close #1312

imp mobilenetssd

add kernels for V1 for FPGA track

imp concat kernel and update readme