// Copyright (c) 2023 Huawei Technologies Co., Ltd
// Copyright (c) 2019, Facebook CORPORATION.
// All rights reserved.
//
// Licensed under the BSD 3-Clause License  (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
// https://opensource.org/licenses/BSD-3-Clause
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.

#include "op_plugin/AclOpsInterface.h"
#include "op_plugin/utils/OpAdapter.h"

namespace acl_op {
using npu_preparation = at_npu::native::OpPreparation;

std::tuple<at::Tensor, at::Tensor> _ctc_loss(const at::Tensor &log_probs, const at::Tensor &targets,
                                             at::IntArrayRef input_lengths_list, at::IntArrayRef target_lengths_list,
                                             int64_t blank, bool zero_infinity)
{
    TORCH_CHECK(log_probs.dim() == 2 || log_probs.dim() == 3,
                "log_probs has to be a 2D or 3D Tensor, but got Tensor of dimension ", log_probs.dim(),
                OPS_ERROR(ErrCode::PARAM));
    at::Tensor log_probs_cast = log_probs;
    if (log_probs.scalar_type() == at::kHalf) {
        log_probs_cast = at_npu::native::custom_ops::_npu_dtype_cast(log_probs_cast, at::kFloat);
    }

    int64_t max_length = 0;
    for (auto &i : target_lengths_list) {
        if (i > max_length) {
            max_length = i;
        }
    }
    // add max_length info
    auto shape = log_probs.sizes();
    blank = blank + max_length * shape[2];

    auto output_sizes = op_infer::ctc_loss_npu_output_size(log_probs, max_length);
    at::Tensor neg_log_likelihood = npu_preparation::apply_tensor_with_format(
        std::get<0>(output_sizes), log_probs_cast.options(), npu_preparation::get_tensor_npu_format(log_probs_cast));

    at::Tensor log_alpha = npu_preparation::apply_tensor_with_format(
        std::get<1>(output_sizes), log_probs_cast.options(), npu_preparation::get_tensor_npu_format(log_probs_cast));

    if (log_probs.dim() == 2) {
        c10::SmallVector<int64_t, N> log_probs_shape = op_infer::array_to_small_vector(log_probs.sizes());
        c10::SmallVector<int64_t, N> log_probs_shape_3d = {log_probs_shape[0], 1, log_probs_shape[1]};
        log_probs_cast = log_probs_cast.reshape(log_probs_shape_3d);
    }
    at_npu::native::OpCommand cmd;
    cmd.Name("CTCLossV2")
        .Input(log_probs_cast)
        .Input(targets)
        .Input(input_lengths_list)
        .Input(target_lengths_list)
        .Output(neg_log_likelihood)
        .Output(log_alpha)
        .Attr("blank", blank)
        .Attr("zero_infinity", zero_infinity)
        .Run();

    if (log_probs.scalar_type() == at::kHalf) {
        neg_log_likelihood = at_npu::native::custom_ops::_npu_dtype_cast(neg_log_likelihood, at::kHalf);
        log_alpha = at_npu::native::custom_ops::_npu_dtype_cast(log_alpha, at::kHalf);
    }

    return std::tie(neg_log_likelihood, log_alpha);
}

} // namespace acl_op