// 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/masked_select_kernel.h"

#include "glog/logging.h"

#include "paddle/phi/backends/xpu/enforce_xpu.h"
#include "paddle/phi/core/kernel_registry.h"

#include "paddle/phi/common/memory_utils.h"
#include "paddle/phi/kernels/expand_kernel.h"
#include "paddle/phi/kernels/funcs/common_shape.h"

namespace phi {

template <typename T, typename Context>
void MaskedSelectKernel(const Context& dev_ctx,
                        const DenseTensor& x,
                        const DenseTensor& mask,
                        DenseTensor* out) {
  using XPUType = typename XPUTypeTrait<T>::Type;
  if (x.numel() == 0 || mask.numel() == 0) {
    out->Resize({0});
    dev_ctx.template Alloc<T>(out);
    return;
  }

  auto expanded_size = funcs::MatrixGetBroadcastBatchPortion(
      vectorize(x.dims()), vectorize(mask.dims()));
  DDim expand_dims = make_ddim(expanded_size);
  DenseTensor mask_expand;
  DenseTensor x_expand;
  if (mask.dims() != expand_dims) {
    ExpandKernel<bool, Context>(
        dev_ctx, mask, IntArray(expanded_size), &mask_expand);
  } else {
    mask_expand = mask;
  }

  if (x.dims() != expand_dims) {
    ExpandKernel<T, Context>(dev_ctx, x, IntArray(expanded_size), &x_expand);
  } else {
    x_expand = x;
  }

  auto* mask_data = mask_expand.data<bool>();
  auto* input_data = reinterpret_cast<const XPUType*>(x_expand.data<T>());
  auto input_dim = x_expand.dims();
  auto mask_dim = mask_expand.dims();

  PADDLE_ENFORCE_EQ(input_dim,
                    mask_dim,
                    common::errors::InvalidArgument(
                        "The dim size of input and mask in OP(masked_selected) "
                        "must be equal, but got input dim:(%ld), mask dim: "
                        "(%ld). Please check input "
                        "value.",
                        input_dim,
                        mask_dim));
  xpu::ctx_guard RAII_GUARD(dev_ctx.x_context());
  int64_t* out_size = RAII_GUARD.alloc_l3_or_gm<int64_t>(1);
  int64_t out_size_cpu;

  PADDLE_ENFORCE_XDNN_SUCCESS(
      xpu::nonzero_count(
          dev_ctx.x_context(), mask_data, out_size, mask.numel()),
      "nonzero_count ");
  memory_utils::Copy(CPUPlace(),
                     static_cast<void*>(&out_size_cpu),
                     mask.place(),
                     static_cast<void*>(out_size),
                     sizeof(int64_t));
  if (std::getenv("XPUSIM_SKIP_RUN") &&
      std::strcmp(std::getenv("XPUSIM_SKIP_RUN"), "1") == 0) {
    VLOG(3) << "WARNING: In the simulator mode, the variable out_size_cpu "
               "stores an uninitialized value. To avoid allocating a memory of "
               "random size, we assign numel to out_size_cpu";
    out_size_cpu = mask.numel();
  }
  DDim out_dim{out_size_cpu};
  out->Resize(out_dim);
  auto out_data = reinterpret_cast<XPUType*>(dev_ctx.template Alloc<T>(out));

  auto input_shape = vectorize<int64_t>(input_dim);
  auto mask_shape = vectorize<int64_t>(mask_dim);
  if (input_dim.size() == 0) {
    input_shape = std::vector<int64_t>({1});
  }
  if (mask_dim.size() == 0) {
    mask_shape = std::vector<int64_t>({1});
  }

  if (out_size_cpu > 0) {
    PADDLE_ENFORCE_XDNN_SUCCESS(xpu::masked_select(dev_ctx.x_context(),
                                                   input_data,
                                                   mask_data,
                                                   out_data,
                                                   input_shape,
                                                   mask_shape,
                                                   out_size_cpu),
                                "masked_select");
  }
}

}  // namespace phi

PD_REGISTER_KERNEL(masked_select,
                   XPU,
                   ALL_LAYOUT,
                   phi::MaskedSelectKernel,
                   float,
                   phi::float16,
                   phi::bfloat16,
                   int,
                   int64_t) {
  kernel->InputAt(1).SetDataType(phi::DataType::BOOL);
}