Function std::rocm_bias_channels

int rocm_bias_channels(
    struct RocmBuf y,
    struct RocmBuf bias,
    unsigned long batch,
    unsigned long channels,
    unsigned long plane,
    int relu,
)

NCHW per-channel bias, in place: y[b][c][i] += bias[c] for every one of the 'plane' (= height*width) elements, then max(0, .) if relu != 0. The epilogue gpu_miopen.h runs after a convolution.