// Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
//     http://www.apache.org/licenses/LICENSE-2.0
//
// 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 "paddle/phi/kernels/eig_kernel.h"
#include "paddle/phi/kernels/cpu/eig.h"

#include "paddle/phi/core/kernel_registry.h"

namespace phi {

template <typename T, typename Context>
void EigKernel(const Context& dev_ctx,
               const DenseTensor& x,
               DenseTensor* out_w,
               DenseTensor* out_v) {
  dev_ctx.template Alloc<dtype::Complex<T>>(out_w);
  dev_ctx.template Alloc<dtype::Complex<T>>(out_v);

  if (x.numel() == 0) {
    return;
  }

  if (!IsComplexType(x.dtype())) {
    int batch_count = BatchCount(x);
    int order = static_cast<int>(x.dims(-1));

    PADDLE_ENFORCE_LT(0,
                      order,
                      errors::InvalidArgument(
                          "The order of Input(X) should be greater than 0."));

    DenseTensor out_w_real;
    DenseTensor out_v_real;

    // double the size of out_w_real, the first half stores the real part,
    // the next half stores the imag part
    std::vector<int64_t> real_w_dims = vectorize<int64_t>(out_w->dims());
    real_w_dims.back() *= 2;
    out_w_real.Resize(real_w_dims);
    dev_ctx.template Alloc<dtype::Real<T>>(&out_w_real);
    out_v_real.Resize(x.dims());
    dev_ctx.template Alloc<dtype::Real<T>>(&out_v_real);

    ApplyEigKernel<dtype::Real<T>, Context>(
        x, &out_w_real, &out_v_real, dev_ctx);

    // 1. extract real part & imag part from out_w_real
    DenseTensor out_w_real_part =
        funcs::Slice<T>(dev_ctx, out_w_real, {-1}, {0}, {order});
    DenseTensor out_w_imag_part =
        funcs::Slice<T>(dev_ctx, out_w_real, {-1}, {order}, {order * 2});

    // 2. construct complex values
    auto* out_w_real_part_ptr = out_w_real_part.data<dtype::Real<T>>();
    auto* out_w_imag_part_ptr = out_w_imag_part.data<dtype::Real<T>>();
    int out_w_numel = static_cast<int>(out_w->numel());

    funcs::ForRange<Context> for_range(dev_ctx, out_w_numel);
    funcs::RealImagToComplexFunctor<dtype::Complex<T>> functor(
        out_w_real_part_ptr,
        out_w_imag_part_ptr,
        dev_ctx.template Alloc<dtype::Complex<T>>(out_w),
        out_w_numel);

    for_range(functor);

    // 3. construct complex vectors
    DenseTensor out_v_real_trans = TransposeLast2Dim<T>(dev_ctx, out_v_real);
    DenseTensor out_v_trans;
    out_v_trans.Resize(x.dims());
    dev_ctx.template Alloc<dtype::Complex<T>>(&out_v_trans);
    ConstructComplexVectors<dtype::Real<T>, dtype::Complex<T>, Context>(
        &out_v_trans, *out_w, out_v_real_trans, dev_ctx, batch_count, order);
    TransposeTwoAxis<dtype::Complex<T>, Context>(
        out_v_trans, out_v, x.dims().size() - 1, x.dims().size() - 2, dev_ctx);
  } else {
    ApplyEigKernel<T, Context>(x, out_w, out_v, dev_ctx);
  }
}

}  // namespace phi

PD_REGISTER_KERNEL(eig,
                   CPU,
                   ALL_LAYOUT,
                   phi::EigKernel,
                   float,
                   double,
                   phi::complex64,
                   phi::complex128) {
  if (kernel_key.dtype() == phi::DataType::FLOAT32 ||
      kernel_key.dtype() == phi::DataType::FLOAT64) {
    kernel->OutputAt(0).SetDataType(phi::dtype::ToComplex(kernel_key.dtype()));
    kernel->OutputAt(1).SetDataType(phi::dtype::ToComplex(kernel_key.dtype()));
  }
}