activation.py 文件源码

def forward(ctx, input, lower, upper, train, inplace):
        ctx.lower = lower
        ctx.upper = upper
        ctx.train = train
        ctx.inplace = inplace
        ctx._backend = type2backend[type(input)]
        if ctx.inplace:
            ctx.mark_dirty(input)
            output = input
        else:
            output = input.new(input.size())
        ctx.noise = input.new()
        ctx._backend.RReLU_updateOutput(
            ctx._backend.library_state,
            input,
            output,
            ctx.noise,
            ctx.lower,
            ctx.upper,
            ctx.train,
            ctx.inplace,
            torch.default_generator if not input.is_cuda else 0
        )
        ctx.save_for_backward(input)
        return output