paddlepaddle--paddle/paddle/phi/core/meta_tensor.cc

/* 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/core/meta_tensor.h"

#include "glog/logging.h"

#include "paddle/phi/core/dense_tensor.h"
#include "paddle/phi/core/distributed/auto_parallel/dist_tensor.h"
#include "paddle/phi/core/enforce.h"
#include "paddle/phi/core/selected_rows.h"
#include "paddle/phi/core/string_tensor.h"
#include "paddle/phi/core/string_tensor_utils.h"
#include "paddle/phi/core/tensor_utils.h"

namespace phi {

static inline void ValidCheck(const MetaTensor& meta_tensor) {
  PADDLE_ENFORCE_EQ(meta_tensor.initialized(),
                    true,
                    common::errors::InvalidArgument(
                        "The current MetaTensor is not initialized."));
}

int64_t MetaTensor::numel() const {
  ValidCheck(*this);
  return tensor_->numel();
}

DDim MetaTensor::dims() const {
  ValidCheck(*this);
  if (SelectedRows::classof(tensor_)) {
    return static_cast<SelectedRows*>(tensor_)->GetCompleteDims();
  } else {
    return tensor_->dims();
  }
}

size_t MetaTensor::size() const {
  ValidCheck(*this);
  PADDLE_ENFORCE_EQ(
      TensorArray::classof(tensor_),
      true,
      common::errors::InvalidArgument(
          "The current MetaTensor is not initialized by TensorArray."));
  TensorArray* tensor_array = static_cast<TensorArray*>(tensor_);
  return tensor_array->size();
}

DDim MetaTensor::dims(int64_t index) const {
  ValidCheck(*this);
  PADDLE_ENFORCE_EQ(
      TensorArray::classof(tensor_),
      true,
      common::errors::InvalidArgument(
          "The current MetaTensor is not initialized by TensorArray."));
  TensorArray* tensor_array = static_cast<TensorArray*>(tensor_);
  return tensor_array->at(index).dims();
}

DDim MetaTensor::strides() const {
  ValidCheck(*this);
  if (dynamic_cast<DenseTensor*>(tensor_)) {
    return dynamic_cast<DenseTensor*>(tensor_)->strides();
  }
  return DDim();
}

DataType MetaTensor::dtype() const {
  ValidCheck(*this);
  return tensor_->dtype();
}

DataLayout MetaTensor::layout() const {
  ValidCheck(*this);
  return tensor_->layout();
}

void MetaTensor::set_dims(const DDim& dims) {
  ValidCheck(*this);
  if (DenseTensor::classof(tensor_)) {
    auto meta =
        DenseTensorUtils::GetMutableMeta(static_cast<DenseTensor*>(tensor_));
    meta->dims = dims;
    if (!strided_kernel_used_) {
      meta->strides = meta->calc_strides(dims);
    }
  } else if (StringTensor::classof(tensor_)) {
    StringTensorUtils::GetMutableMeta(static_cast<StringTensor*>(tensor_))
        ->dims = dims;
  } else if (SelectedRows::classof(tensor_)) {
    static_cast<SelectedRows*>(tensor_)->set_height(dims[0]);
  } else if (SparseCooTensor::classof(tensor_)) {
    DenseTensorUtils::GetMutableMeta(static_cast<SparseCooTensor*>(tensor_))
        ->dims = dims;
  } else if (SparseCsrTensor::classof(tensor_)) {
    DenseTensorUtils::GetMutableMeta(static_cast<SparseCsrTensor*>(tensor_))
        ->dims = dims;
  } else if (distributed::DistTensor::classof(tensor_)) {
    static_cast<distributed::DistTensor*>(tensor_)->unsafe_set_dims(dims);
  } else {
    PADDLE_THROW(common::errors::Unimplemented(
        "Unsupported setting dims for `%s`.", tensor_->type_info().name()));
  }
}

void MetaTensor::set_strides(const DDim& strides) {
  ValidCheck(*this);
  if (dynamic_cast<DenseTensor*>(tensor_)) {
    DenseTensorUtils::GetMutableMeta(dynamic_cast<DenseTensor*>(tensor_))
        ->strides = strides;
  }
}

void MetaTensor::set_dtype(DataType dtype) {
  ValidCheck(*this);
  if (DenseTensor::classof(tensor_)) {
    DenseTensorUtils::GetMutableMeta(static_cast<DenseTensor*>(tensor_))
        ->dtype = dtype;
  } else if (TensorArray::classof(tensor_)) {
    static_cast<TensorArray*>(tensor_)->set_type(dtype);
  } else if (StringTensor::classof(tensor_)) {
    // No need to set dtype
  } else if (SelectedRows::classof(tensor_)) {
    DenseTensorUtils::GetMutableMeta(
        static_cast<SelectedRows*>(tensor_)->mutable_value())
        ->dtype = dtype;
  } else if (SparseCooTensor::classof(tensor_)) {
    DenseTensorUtils::GetMutableMeta(static_cast<SparseCooTensor*>(tensor_))
        ->dtype = dtype;
  } else if (SparseCsrTensor::classof(tensor_)) {
    DenseTensorUtils::GetMutableMeta(static_cast<SparseCsrTensor*>(tensor_))
        ->dtype = dtype;
  } else if (distributed::DistTensor::classof(tensor_)) {
    // For pipeline parallelism, DistTensor holds an uninitialized DenseTensor,
    // But kernel launch needs to get it's placement, dtype and layout.
    VLOG(3) << "DistTensor set dtype: " << dtype;
    DenseTensorUtils::GetMutableMeta(
        static_cast<distributed::DistTensor*>(tensor_)->unsafe_mutable_value())
        ->dtype = dtype;
  } else {
    PADDLE_THROW(common::errors::Unimplemented(
        "Unsupported setting dtype for `%s`.", tensor_->type_info().name()));
  }
}

void MetaTensor::set_layout(DataLayout layout) {
  ValidCheck(*this);
  if (DenseTensor::classof(tensor_)) {
    auto meta =
        DenseTensorUtils::GetMutableMeta(static_cast<DenseTensor*>(tensor_));
    meta->layout = layout;
    if (!strided_kernel_used_) {
      meta->strides = meta->calc_strides(meta->dims);
    }
  } else if (TensorArray::classof(tensor_)) {
    static_cast<TensorArray*>(tensor_)->set_layout(layout);
  } else if (StringTensor::classof(tensor_)) {
    // No need to set layout
  } else if (SelectedRows::classof(tensor_)) {
    auto meta = DenseTensorUtils::GetMutableMeta(
        static_cast<SelectedRows*>(tensor_)->mutable_value());
    meta->layout = layout;
    if (!strided_kernel_used_) {
      meta->strides = meta->calc_strides(meta->dims);
    }
  } else if (SparseCooTensor::classof(tensor_)) {
    DenseTensorUtils::GetMutableMeta(static_cast<SparseCooTensor*>(tensor_))
        ->layout = layout;
  } else if (SparseCsrTensor::classof(tensor_)) {
    DenseTensorUtils::GetMutableMeta(static_cast<SparseCsrTensor*>(tensor_))
        ->layout = layout;
  } else if (distributed::DistTensor::classof(tensor_)) {
    VLOG(3) << "DistTensor set layout: " << layout;
    DenseTensorUtils::GetMutableMeta(
        static_cast<distributed::DistTensor*>(tensor_)->unsafe_mutable_value())
        ->layout = layout;
  } else {
    PADDLE_THROW(common::errors::Unimplemented(
        "Unsupported setting layout for `%s`.", tensor_->type_info().name()));
  }
}

void MetaTensor::share_lod(const MetaTensor& meta_tensor) {
  ValidCheck(*this);
  ValidCheck(meta_tensor);
  if (SparseCooTensor::classof(tensor_) || SparseCsrTensor::classof(tensor_) ||
      distributed::DistTensor::classof(tensor_)) {
    return;
  }
  if (meta_tensor.lod().empty()) {
    // no need share
    return;
  }
  if (DenseTensor::classof(tensor_)) {
    DenseTensorUtils::GetMutableMeta(static_cast<DenseTensor*>(tensor_))
        ->legacy_lod = meta_tensor.lod();
  } else if (SelectedRows::classof(tensor_)) {
    DenseTensorUtils::GetMutableMeta(
        static_cast<SelectedRows*>(tensor_)->mutable_value())
        ->legacy_lod = meta_tensor.lod();
  } else {
    PADDLE_THROW(common::errors::Unimplemented(
        "Unsupported sharing lod inplace for `%s`.",
        tensor_->type_info().name()));
  }
}

void MetaTensor::share_lod(const LegacyLoD& legacy_lod) {
  ValidCheck(*this);
  if (SparseCooTensor::classof(tensor_) || SparseCsrTensor::classof(tensor_) ||
      distributed::DistTensor::classof(tensor_)) {
    return;
  }
  if (legacy_lod.empty()) {
    // no need share
    return;
  }
  if (DenseTensor::classof(tensor_)) {
    DenseTensorUtils::GetMutableMeta(static_cast<DenseTensor*>(tensor_))
        ->legacy_lod = legacy_lod;
  } else if (SelectedRows::classof(tensor_)) {
    DenseTensorUtils::GetMutableMeta(
        static_cast<SelectedRows*>(tensor_)->mutable_value())
        ->legacy_lod = legacy_lod;
  } else {
    PADDLE_THROW(common::errors::Unimplemented(
        "Unsupported sharing lod inplace for `%s`.",
        tensor_->type_info().name()));
  }
}

void MetaTensor::share_lod(const MetaTensor& meta_tensor, int64_t index) {
  ValidCheck(meta_tensor);
  PADDLE_ENFORCE_EQ(
      meta_tensor.is_tensor_array(),
      true,
      common::errors::InvalidArgument(
          "The current MetaTensor is not initialized by TensorArray."));
  if (meta_tensor.lod(index).empty()) {
    // no need share
    return;
  }
  share_lod(meta_tensor.lod(index));
}

void MetaTensor::share_meta(const MetaTensor& meta_tensor) {
  ValidCheck(*this);
  if (DenseTensor::classof(tensor_) || SelectedRows::classof(tensor_) ||
      SparseCooTensor::classof(tensor_) || SparseCsrTensor::classof(tensor_) ||
      distributed::DistTensor::classof(tensor_)) {
    share_dims(meta_tensor);
    set_dtype(meta_tensor.dtype());
    set_layout(meta_tensor.layout());
    share_lod(meta_tensor);
  } else {
    PADDLE_THROW(common::errors::Unimplemented(
        "Unsupported sharing meta for `%s`.", tensor_->type_info().name()));
  }
}

TensorBase* MetaTensor::tensor() const { return tensor_; }

bool MetaTensor::is_dense() const { return DenseTensor::classof(tensor_); }
bool MetaTensor::is_selected_rows() const {
  return SelectedRows::classof(tensor_);
}
bool MetaTensor::is_dist() const {
  return distributed::DistTensor::classof(tensor_);
}
bool MetaTensor::is_tensor_array() const {
  return TensorArray::classof(tensor_);
}

bool MetaTensor::is_same_tensor(const MetaTensor& meta_tensor) const {
  return tensor_ != nullptr && tensor_ == meta_tensor.tensor();
}

void MetaTensor::share_dims(const MetaTensor& meta_tensor) {
  ValidCheck(*this);
  bool is_dense_tensor = DenseTensor::classof(tensor_);
  bool is_selected_rows = SelectedRows::classof(tensor_);
  bool is_sparse_coo = SparseCooTensor::classof(tensor_);
  bool is_sparse_csr = SparseCsrTensor::classof(tensor_);
  bool is_dist_tensor = distributed::DistTensor::classof(tensor_);
  if (is_dense_tensor || is_selected_rows || is_sparse_coo || is_sparse_csr ||
      is_dist_tensor) {
    if (is_selected_rows) {
      const auto in_tensor_base = meta_tensor.tensor();
      PADDLE_ENFORCE_EQ(
          SelectedRows::classof(in_tensor_base),
          true,
          errors::InvalidArgument("The input MetaTensor is SelectedRows, but "
                                  "the output MetaTensor is not this type."));
      auto* selected_rows_out = static_cast<SelectedRows*>(tensor_);
      auto* selected_rows_in = static_cast<SelectedRows*>(in_tensor_base);
      selected_rows_out->set_rows(selected_rows_in->rows());
      selected_rows_out->set_height(selected_rows_in->height());
      auto meta = DenseTensorUtils::GetMutableMeta(
          static_cast<SelectedRows*>(tensor_)->mutable_value());
      meta->dims = selected_rows_in->mutable_value()->dims();
      if (!strided_kernel_used_) {
        meta->strides = meta->calc_strides(meta->dims);
      }
    } else {
      set_dims(meta_tensor.dims());
    }
  } else {
    PADDLE_THROW(common::errors::Unimplemented(
        "Unsupported sharing dims for `%s`.", tensor_->type_info().name()));
  }
}

void MetaTensor::share_strides(const MetaTensor& meta_tensor) {
  ValidCheck(*this);
  if (DenseTensor::classof(tensor_)) {
    set_strides(meta_tensor.strides());
  }
}

bool MetaTensor::initialized() const { return tensor_ != nullptr; }

// Private Member Methods

const LegacyLoD& MetaTensor::lod() const {
  if (DenseTensor::classof(tensor_)) {
    return static_cast<DenseTensor*>(tensor_)->lod();
  } else if (SelectedRows::classof(tensor_)) {
    return static_cast<SelectedRows*>(tensor_)->value().lod();
  } else if (SparseCooTensor::classof(tensor_)) {
    return static_cast<SparseCooTensor*>(tensor_)->non_zero_elements().lod();
  } else if (SparseCsrTensor::classof(tensor_)) {
    return static_cast<SparseCsrTensor*>(tensor_)->non_zero_elements().lod();
  } else {
    PADDLE_THROW(common::errors::Unimplemented(
        "Unsupported getting lod of `%s`.", tensor_->type_info().name()));
  }
}

const LegacyLoD& MetaTensor::lod(int64_t index) const {
  ValidCheck(*this);
  PADDLE_ENFORCE_EQ(
      is_tensor_array(),
      true,
      common::errors::InvalidArgument(
          "The current MetaTensor is not initialized by TensorArray."));
  TensorArray* tensor_array = static_cast<TensorArray*>(tensor_);
  return tensor_array->at(index).lod();
}

}  // namespace phi