paddlepaddle--paddle/paddle/phi/kernels/funcs/indexing.h

// Copyright (c) 2025 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.

#pragma once

#include <vector>
#include "paddle/common/array.h"
#include "paddle/phi/backends/context_pool.h"
#include "paddle/phi/common/int_array.h"
#include "paddle/phi/common/memory_utils.h"
#include "paddle/phi/common/place.h"
#include "paddle/phi/core/dense_tensor.h"
#include "paddle/phi/kernels/cast_kernel.h"
#include "paddle/phi/kernels/expand_kernel.h"
#include "paddle/phi/kernels/nonzero_kernel.h"
#include "paddle/phi/kernels/reshape_kernel.h"
#include "paddle/phi/kernels/slice_kernel.h"
#include "paddle/phi/kernels/split_kernel.h"

#if defined(__NVCC__) || defined(__HIPCC__)
#ifdef __NVCC__
#include <cuda.h>
#include <cuda_runtime.h>
#elif defined(__HIPCC__)
#include <hip/hip_runtime.h>
#endif
#endif

namespace phi {

namespace funcs {

static inline common::DDim InferSizeSymdimvector(const common::DDim& a,
                                                 const common::DDim& b) {
  auto dimsA = a.size();
  auto dimsB = b.size();
  auto ndim = dimsA > dimsB ? dimsA : dimsB;
  common::DDim expandedSizes = make_ddim(std::vector<int64_t>(ndim, 0));

  for (int64_t i = ndim - 1; i >= 0; --i) {
    int64_t offset = ndim - 1 - i;
    int64_t dimA = dimsA - 1 - offset;
    int64_t dimB = dimsB - 1 - offset;
    auto sizeA = (dimA >= 0) ? a[dimA] : 1;
    auto sizeB = (dimB >= 0) ? b[dimB] : 1;

    PADDLE_ENFORCE_EQ(
        sizeA == sizeB || sizeA == 1 || sizeB == 1,
        true,
        common::errors::Fatal(
            "The size of tensor a (%d) must match the size of tensor b "
            "(%d) at non-singleton dimension %d",
            sizeA,
            sizeB,
            i));

    expandedSizes[i] = sizeA == 1 ? sizeB : sizeA;
  }

  return expandedSizes;
}

template <typename T, typename Context>
std::vector<DenseTensor*> ExpandTensors(
    const Context& dev_ctx,
    const std::vector<std::unique_ptr<DenseTensor>>& indices) {
  std::vector<DenseTensor*> result;
  for (auto& index : indices) {
    if (index->dtype() == DataType::BOOL) {
      DenseTensor bool_2_idx(DataType::INT64);
      NonZeroKernel<bool, Context>(dev_ctx, *index, &bool_2_idx);
      if (bool_2_idx.numel() == 0) {
        std::vector<DenseTensor*> empty_result;
        return empty_result;
      }
      for (int j = 0; j < index->dims().size(); j++) {
        SliceKernel<int64_t, Context>(
            dev_ctx, bool_2_idx, {1}, {j}, {j + 1}, {1}, {1}, index.get());
        result.emplace_back(index.get());
      }
    } else {
      result.emplace_back(index.get());
    }
  }
  return result;
}

template <typename T, typename Context>
std::vector<DenseTensor*> ExpandOutplace(
    const Context& dev_ctx, const std::vector<DenseTensor*>& to_expand) {
  bool first = true;
  common::DDim sizes;
  for (size_t i = 0; i < to_expand.size(); i++) {
    if (!to_expand[i]->initialized()) {
      continue;
    } else if (first) {
      sizes = to_expand[i]->dims();
      first = false;
    } else {
      sizes = InferSizeSymdimvector(sizes, to_expand[i]->dims());
    }
  }

  std::vector<DenseTensor*> result(to_expand.size());
  for (size_t i = 0; i < to_expand.size(); i++) {
    if (!to_expand[i]->initialized()) {
      continue;
    } else if (to_expand[i]->dims() == sizes) {
      result[i] = to_expand[i];
    } else {
      if (to_expand[i]->dtype() == DataType::INT32) {
        DenseTensor tmp_idx(DataType::INT64);
        ExpandKernel<int32_t, Context>(dev_ctx,
                                       *(to_expand[i]),
                                       IntArray(vectorize<int32_t>(sizes)),
                                       &tmp_idx);
        *(to_expand[i]) = tmp_idx;
        result[i] = to_expand[i];
      } else if (to_expand[i]->dtype() == DataType::INT64) {
        DenseTensor tmp_idx(DataType::INT64);
        ExpandKernel<int64_t, Context>(dev_ctx,
                                       *(to_expand[i]),
                                       IntArray(vectorize<int64_t>(sizes)),
                                       &tmp_idx);
        *(to_expand[i]) = tmp_idx;
        result[i] = to_expand[i];
      } else {
        PADDLE_THROW(::common::errors::Unimplemented(
            "Index in Stride Mechanism must be int32_t, int64_t or bool"));
      }
    }
  }
  return result;
}

template <typename T, typename Context>
struct AdvancedIndex {
  AdvancedIndex(const Context& dev_ctx,
                const DenseTensor& self,
                const std::vector<const DenseTensor*>& orig);
  ~AdvancedIndex() = default;
  DenseTensor src;
  std::vector<std::unique_ptr<DenseTensor>> tmp_indices;
  std::vector<const DenseTensor*> indices;
  std::vector<int64_t> indexed_sizes;
  std::vector<int64_t> indexed_strides;
  int64_t dims_before;
  int64_t dims_after;
  bool bool_case;
  bool empty_index = false;
};

inline static void RestrideSrc(const DenseTensor& self,
                               const int64_t& dims_before,
                               const int64_t& dims_indexed,
                               const std::vector<int64_t>& replacement_shape,
                               DenseTensor* view_src) {
  std::vector<int64_t> shape_vec = (vectorize<int64_t>(self.dims()));
  std::vector<int64_t> strides_vec = (vectorize<int64_t>(self.strides()));
  std::vector<int64_t>* shape = &shape_vec;
  std::vector<int64_t>* strides = &strides_vec;
  int64_t end = dims_before + dims_indexed;
  shape->erase(shape->begin() + dims_before, shape->begin() + end);
  strides->erase(strides->begin() + dims_before, strides->begin() + end);
  shape->insert(shape->begin() + dims_before,
                replacement_shape.begin(),
                replacement_shape.end());
  strides->insert(strides->begin() + dims_before, replacement_shape.size(), 0);
  auto meta = self.meta();
  meta.dims = make_ddim(*shape);
  meta.strides = make_ddim(*strides);
  meta.offset = self.offset();
  view_src->set_meta(meta);
  view_src->ResetHolder(self.Holder());
  view_src->ShareInplaceVersionCounterWith(self);
}

inline static void ReshapeIndexer(DenseTensor* index,
                                  const int64_t& dims_before,
                                  const int64_t& dims_after) {
  auto orig_shape = vectorize<int64_t>(index->dims());
  auto shape = std::vector<int64_t>{};
  shape.insert(shape.end(), dims_before, 1);
  shape.insert(shape.end(), orig_shape.begin(), orig_shape.end());
  shape.insert(shape.end(), dims_after, 1);
  index->Resize(shape);
}

template <typename T, typename Context>
inline AdvancedIndex<T, Context>::AdvancedIndex(
    const Context& dev_ctx,
    const DenseTensor& self,
    const std::vector<const DenseTensor*>& orig) {
  for (int i = 0; i < orig.size(); i++) {
    tmp_indices.emplace_back(std::make_unique<DenseTensor>());
    *(tmp_indices.back()) = *(const_cast<DenseTensor*>(orig[i]));
  }

  auto indices = ExpandTensors<T, Context>(dev_ctx, this->tmp_indices);
  if (indices.empty()) {
    empty_index = true;
    return;
  }

  indices = ExpandOutplace<T, Context>(dev_ctx, indices);
  while (indices.size() < static_cast<size_t>(self.dims().size())) {
    indices.emplace_back();
  }

  std::vector<DenseTensor*> indices_int64;
  for (auto& indice : indices) {
    if (indice && indice->dtype() == DataType::INT32) {
      *indice = Cast<int, Context>(dev_ctx, *indice, DataType::INT64);
    }
    indices_int64.push_back(indice);
  }

  std::vector<DenseTensor*> indices_list = indices_int64;

  uint32_t element_size_bytes = phi::SizeOf(self.dtype());
  int64_t dims_before = 0, dims_after = 0, dims_indexed = 0;
  std::vector<int64_t> shape_vec = vectorize<int64_t>(self.dims());
  std::vector<int64_t> stride_vec = vectorize<int64_t>(self.strides());
  std::vector<int64_t> replacement_shape;
  std::vector<int64_t> idx_shape_vec = {};
  std::vector<int64_t> idx_stride_vec = {};
  for (size_t dim = 0; dim < indices_list.size(); dim++) {
    if (!indices_list[dim]) {
      if (dims_indexed == 0) {
        dims_before++;
      } else {
        dims_after++;
      }
    } else {
      dims_indexed++;
      replacement_shape = vectorize<int64_t>(indices_list[dim]->dims());

      indexed_sizes.push_back(shape_vec[dim]);
      indexed_strides.push_back(stride_vec[dim] * element_size_bytes);
    }
  }

  this->dims_before = dims_before;
  this->dims_after = dims_after;
  RestrideSrc(self, dims_before, dims_indexed, replacement_shape, &(this->src));

  for (auto& index : indices_list) {
    if (index) {
      ReshapeIndexer(index, dims_before, dims_after);
      this->indices.push_back(index);
    }
  }
}

}  // namespace funcs
}  // namespace phi