0e3ecd7a2a4c544de6228ed466c29992f65def8e,cuda_functional.py,SRU_Compute_GPU,backward,#SRU_Compute_GPU#Any#Any#,539

Before Change

        init_ = x.new(ncols).zero_() if init is None else init
        grad_u = u.new(*u.size()).zero_()
        grad_wc = x.new(2, batch, d*bidir)
        grad_bias = x.new(2, batch, d*bidir)
        grad_init = x.new(batch, d*bidir)

        //  For DEBUG
        //  size = (length, batch, x.size(-1)) if x.dim() == 3 else (batch, x.size(-1))
        //  grad_x = x.new(*x.size()) if k_ == 3 else x.new(*size).zero_()

        //  Normal use
        grad_x = x.new(*x.size()).zero_() if skip_type > 0 and k_ == 3 else None

        if skip_type > 0 and k_ == 3:
            x_ptr = x.contiguous()*scale_x if scale_x != 1 else x.contiguous()
            x_ptr = x_ptr.data_ptr()
        else:
            x_ptr = 0

        stream, _, _, bwd_func, bibwd_func = self.get_functions()
        FUNC = bwd_func if not self.bidirectional else bibwd_func
        FUNC(args=[
            u.contiguous().data_ptr(),
            x_ptr,
            weight_c.data_ptr(),
            bias.data_ptr(),
            init_.contiguous().data_ptr(),
            mask_h.data_ptr() if mask_h is not None else 0,
            mask_pad.data_ptr() if mask_pad is not None else 0,
            c.data_ptr(),
            grad_h.contiguous().data_ptr(),
            grad_last.contiguous().data_ptr(),
            length,
            batch,
            d,
            k_,
            grad_u.data_ptr(),
            grad_x.data_ptr() if skip_type > 0 and k_ == 3 else 0,
            grad_wc.data_ptr(),
            grad_bias.data_ptr(),
            grad_init.data_ptr(),
            self.activation_type,
            skip_type],
            block=(thread_per_block, 1, 1),
            grid=(num_block, 1, 1),
            stream=stream
        )

        if skip_type > 0 and k_ == 3 and scale_x != 1:
            grad_x.mul_(scale_x)
        return grad_u, grad_x, grad_wc.sum(1).view(-1), grad_bias.sum(1).view(-1), grad_init, None

After Change

        init_ = x.new(ncols).zero_() if init is None else init
        grad_u = u.new(*u.size()).zero_()
        grad_wc = x.new(2*bidir*d).zero_()
        grad_bias = x.new(2*bidir*d).zero_()
        grad_init = x.new(batch, d*bidir)
        grad_x = x.new(*x.size()).zero_() if skip_type > 0 and k_ == 3 else None

        if skip_type > 0 and k_ == 3:
            x_ptr = x.contiguous()*scale_x if scale_x != 1 else x.contiguous()
            x_ptr = x_ptr.data_ptr()
        else:
            x_ptr = 0

        stream, _, _, bwd_func, bibwd_func = self.get_functions()
        FUNC = bwd_func if not self.bidirectional else bibwd_func
        FUNC(args=[
            u.contiguous().data_ptr(),
            x_ptr,
            weight_c.data_ptr(),
            bias.data_ptr(),
            init_.contiguous().data_ptr(),
            mask_h.data_ptr() if mask_h is not None else 0,
            mask_pad.data_ptr() if mask_pad is not None else 0,
            c.data_ptr(),
            grad_h.contiguous().data_ptr(),
            grad_last.contiguous().data_ptr(),
            length,
            batch,
            d,
            k_,
            grad_u.data_ptr(),
            grad_x.data_ptr() if skip_type > 0 and k_ == 3 else 0,
            grad_wc.data_ptr(),
            grad_bias.data_ptr(),
            grad_init.data_ptr(),
            self.activation_type,
            skip_type],
            block=(thread_per_block, 1, 1),
            grid=(num_block, 1, 1),
            stream=stream
        )

        if skip_type > 0 and k_ == 3 and scale_x != 1:
            grad_x.mul_(scale_x)
        return grad_u, grad_x, grad_wc, grad_bias, grad_init, None

Instances