#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;
std::tuple<at::Tensor&, at::Tensor&> topk_out(
const at::Tensor& self,
int64_t k,
int64_t dim,
bool largest,
bool sorted,
at::Tensor& values,
at::Tensor& indices)
{
if (self.dim() == 0 && k == 0) {
TORCH_CHECK(values.dtype() == at::ScalarType::Long, "Expected out tensor to have dtype long int, bug got ",
values.dtype(), "instead.", OPS_ERROR(ErrCode::PARAM));
TORCH_CHECK(indices.dtype() == at::ScalarType::Long, "Expected out tensor to have dtype long int, bug got ",
indices.dtype(), "instead.", OPS_ERROR(ErrCode::PARAM));
values.resize_as_(self).fill_(self);
indices.resize_as_(self).fill_(0);
return std::tuple<at::Tensor&, at::Tensor&>(values, indices);
}
DO_COMPATIBILITY(aclnnTopk, acl_op::topk_out(self, k, dim, largest, sorted, values, indices));
auto output_size = op_infer::topk_npu_output_size(self, k, dim);
npu_preparation::check_tensor({self}, values, self.scalar_type(), output_size);
npu_preparation::check_tensor({self}, indices, at::ScalarType::Long, output_size);
EXEC_NPU_CMD(aclnnTopk, self, k, dim, largest, sorted, values, indices);
return std::tuple<at::Tensor&, at::Tensor&>(values, indices);
}
std::tuple<at::Tensor, at::Tensor> topk(
const at::Tensor& self,
int64_t k,
int64_t dim,
bool largest,
bool sorted)
{
DO_COMPATIBILITY(aclnnTopk, acl_op::topk(self, k, dim, largest, sorted));
auto output_size = op_infer::topk_npu_output_size(self, k, dim);
at::Tensor values;
at::Tensor indices;
if (self.dim() == 0 && k == 0) {
values = at::zeros_like(self).fill_(self);
indices = at::zeros_like(self, at::kLong);
return std::tuple<at::Tensor, at::Tensor>(values, indices);
}
values = npu_preparation::apply_tensor_without_format(output_size, self.options());
indices = npu_preparation::apply_tensor_without_format(output_size, self.options().dtype(at::kLong));
EXEC_NPU_CMD(aclnnTopk, self, k, dim, largest, sorted, values, indices);
return std::tuple<at::Tensor, at::Tensor>(values, indices);
}
}