#include "op_plugin/AclOpsInterface.h"
#include "op_plugin/OpApiInterface.h"
#include "op_plugin/utils/op_api_common.h"
namespace op_api {
using npu_preparation = at_npu::native::OpPreparation;
at::Tensor& smooth_l1_loss_backward_out(
const at::Tensor& grad_out,
const at::Tensor& self,
const at::Tensor& target,
int64_t reduction,
double beta,
at::Tensor& grad_input)
{
DO_COMPATIBILITY(aclnnSmoothL1LossBackward,
acl_op::smooth_l1_loss_backward_out(grad_out, self, target, reduction, beta, grad_input));
auto mid_shape = op_infer::broadcast_ops_npu_output_size(self.sizes(), target.sizes());
auto output_size = op_infer::broadcast_ops_npu_output_size(mid_shape, grad_out.sizes());
npu_preparation::check_tensor({grad_out, self, target}, grad_input, grad_input.scalar_type(), output_size);
float sigma = static_cast<float>(beta);
EXEC_NPU_CMD(aclnnSmoothL1LossBackward, grad_out, self, target, reduction, sigma, grad_input);
return grad_input;
}
at::Tensor smooth_l1_loss_backward(
const at::Tensor& grad_out,
const at::Tensor& self,
const at::Tensor& target,
int64_t reduction,
double beta)
{
DO_COMPATIBILITY(aclnnSmoothL1LossBackward,
acl_op::smooth_l1_loss_backward(grad_out, self, target, reduction, beta));
auto mid_shape = op_infer::broadcast_ops_npu_output_size(self.sizes(), target.sizes());
auto output_size = op_infer::broadcast_ops_npu_output_size(mid_shape, grad_out.sizes());
at::Tensor grad_input = npu_preparation::apply_tensor_without_format(self, output_size);
float sigma = static_cast<float>(beta);
EXEC_NPU_CMD(aclnnSmoothL1LossBackward, grad_out, self, target, reduction, sigma, grad_input);
return grad_input;
}
}