caffe2/operators/reshape_op.h

#ifndef CAFFE2_OPERATORS_RESHAPE_OP_H_
#define CAFFE2_OPERATORS_RESHAPE_OP_H_

#include "caffe2/core/common_omp.h"
#include "caffe2/core/context.h"
#include "caffe2/core/logging.h"
#include "caffe2/core/operator.h"
#include "caffe2/utils/math.h"
#include "c10/util/irange.h"

namespace caffe2 {

// Takes a shape and data tensor and reshapes it
template <typename F, class Context>
class ReshapeOp : public Operator<Context> {
 public:
  USE_OPERATOR_CONTEXT_FUNCTIONS;
  template <class... Args>
  explicit ReshapeOp(Args&&... args)
      : Operator<Context>(std::forward<Args>(args)...),
        new_shape_(this->template GetRepeatedArgument<int64_t>("shape")) {}

  bool RunOnDevice() override {
    if (InputSize() == 2) {
      return DispatchHelper<TensorTypes<int, int64_t>>::call(this, Input(1));
    }
    CAFFE_ENFORCE(
        OperatorBase::HasArgument("shape"), "Argument `shape` is missing.");
    return this->template DoRunWithType<int64_t>();
  }

  template <typename T>
  bool DoRunWithType() {
    DoRunWithTypeImpl<T>(Input(0), Output(0));
    return true;
  }

 protected:
  template <typename T>
  void DoRunWithTypeImpl(const Tensor& input, Tensor* output) {
    vector<int64_t> actual_new_shape = new_shape_;
    if (InputSize() == 2) {
      CAFFE_ENFORCE(
          !OperatorBase::HasArgument("shape"),
          "New shape is specified by the input blob, do not pass in "
          "the argument `shape`.");

      // Shape should be always stored only on CPU
      // Just in case if for some reason shape is on GPU
      if (this->InputIsTensorType(1, CPU)) {
        // originally, shape input must be in CPU context
        auto& shape = this->template Input<Tensor>(1, CPU);
        CAFFE_ENFORCE_EQ(
            shape.dim(),
            1,
            "When input_as_shape is true, the input must be a 1D tensor of "
            "data type int64_t");
        CAFFE_ENFORCE(shape.numel() > 0);
        auto* shape_data = shape.template data<T>();
        actual_new_shape.insert(
            actual_new_shape.end(), shape_data, shape_data + shape.dim32(0));
      } else {
        auto& shape = Input(1);
        CAFFE_ENFORCE_EQ(
            shape.dim(),
            1,
            "When input_as_shape is true, the input must be a 1D tensor of "
            "data type int64_t");
        CAFFE_ENFORCE(shape.numel() > 0);
        auto* shape_data = shape.template data<T>();
        // Fetch copy from
        std::unique_ptr<T[]> shape_data_copy =
            std::make_unique<T[]>(shape.dim32(0));
        context_.template CopyToCPU<T>(
            shape.dim32(0), shape_data, shape_data_copy.get());
        actual_new_shape.insert(
            actual_new_shape.end(),
            shape_data_copy.get(),
            shape_data_copy.get() + shape.dim32(0));
      }
    }

    // Checks if the new shape is valid and fills in the missing dimension
    // specified by -1.
    // NOTE: At most one dimension can be -1.
    auto total_size = input.numel();
    T size = 1;

    // NOTE: support for legacy caffe1 syntax
    // Copy over the dimensions for those that are specified zero.
    if (total_size != 0) {
      // NOLINTNEXTLINE(clang-diagnostic-sign-compare)
      for (size_t i = 0; i < actual_new_shape.size() && i < input.dim(); ++i) {
        if (actual_new_shape[i] == 0) {
          actual_new_shape[i] = input.size(i);
        }
      }
    }

    int unknown_idx = -1;
    for (const auto i : c10::irange(actual_new_shape.size())) {
      const auto dim = actual_new_shape[i];
      if (dim == -1) {
        CAFFE_ENFORCE(
            unknown_idx == -1,
            "Argument `shape` has more than one missing dimension.");
        unknown_idx = i;
      } else {
        size *= dim;
      }
    }
    if (size == 0 && total_size != 0) {
      CAFFE_THROW(
          "Can not reshape a non-zero size (",
          total_size,
          ") tensor to zero size.");
    }
    if (total_size != 0) {
      // if tensor is not empty, infer the size of the unknown index
      if (unknown_idx != -1) {
        CAFFE_ENFORCE_NE(
            size,
            0,
            "New shape at dim ",
            unknown_idx,
            " can not be inferred since new size is zero.");
        CAFFE_ENFORCE(
            total_size % size == 0,
            "Argument `shape` does not agree with the input data.",
            " (",
            total_size,
            " vs ",
            size,
            ")");
        actual_new_shape[unknown_idx] = total_size / size;
      } else {
        CAFFE_ENFORCE_EQ(
            total_size,
            size,
            "Argument `shape` does not agree with the input data.",
            " (",
            total_size,
            " != ",
            size,
            ")");
      }
    } else if (unknown_idx != -1) {
      // if size is empty, then set unknown index to be 0 (empty tensor)
      actual_new_shape[unknown_idx] = 0;
    }

    // Write the original shape to the second output.
    auto* old_shape = this->template Output<Tensor>(1, CPU);
    old_shape->Resize(input.sizes().size());
    T* old_shape_data = old_shape->template mutable_data<T>();
    std::vector<T> old_shape_vector(input.sizes().begin(), input.sizes().end());
    for (const auto i : c10::irange(old_shape_vector.size())) {
      old_shape_data[i] = old_shape_vector[i];
    }

    output->Resize(actual_new_shape);
    if (output != &input) {
      // If we are not doing in-place computation, a copy is needed.
      context_.CopyItemsSameDevice(
          input.dtype(),
          input.numel(),
          input.raw_data(),
          output->raw_mutable_data(input.dtype()));
    }
  }

 private:
  vector<int64_t> new_shape_;
};

} // namespace caffe2

#endif // CAFFE2_OPERATORS_RESHAPE_OP_H_