#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_out(const at::Tensor &self, const at::Tensor &target, int64_t reduction, double beta,
at::Tensor &out)
{
DO_COMPATIBILITY(aclnnSmoothL1Loss, acl_op::smooth_l1_loss_out(self, target, reduction, beta, out));
auto outputSize = op_infer::smooth_l1_loss_npu_output_size(self, reduction);
npu_preparation::check_tensor({self, target}, out, out.scalar_type(), outputSize);
npu_preparation::check_memory({self, target}, {out});
float sigma = static_cast<float>(beta);
EXEC_NPU_CMD(aclnnSmoothL1Loss, self, target, reduction, sigma, out);
return out;
}
at::Tensor smooth_l1_loss(const at::Tensor &self, const at::Tensor &target, int64_t reduction, double beta)
{
DO_COMPATIBILITY(aclnnSmoothL1Loss, acl_op::smooth_l1_loss(self, target, reduction, beta));
auto outputSize = op_infer::smooth_l1_loss_npu_output_size(self, reduction);
at::ScalarType high_type = at::native::result_type(self, target);
at::Tensor result = npu_preparation::apply_tensor_without_format(outputSize, self.options().dtype(high_type));
float sigma = static_cast<float>(beta);
EXEC_NPU_CMD(aclnnSmoothL1Loss, self, target, reduction, sigma, result);
return result;
}
}