monolish_tensor_dense.hpp

Go to the documentation of this file.

#pragma once
#include "monolish_matrix.hpp"
#include "monolish_tensor.hpp"
#include "monolish_vector.hpp"
 
namespace monolish {
template <typename Float> class vector;
template <typename TYPE, typename Float> class view1D;
namespace tensor {
template <typename Float> class tensor_COO;
template <typename Float> class tensor_Dense {
private:
  std::vector<size_t> shape;
 
  mutable bool gpu_status = false;
 
public:
  std::shared_ptr<Float> val;
 
  size_t val_nnz = 0;
 
  size_t alloc_nnz = 0;
 
  bool val_create_flag = false;
 
  tensor_Dense() { val_create_flag = true; }
 
  void convert(const tensor::tensor_Dense<Float> &tens);
 
  tensor_Dense(const tensor::tensor_Dense<Float> &tens);
 
  void convert(const tensor::tensor_COO<Float> &tens);
 
  tensor_Dense(const tensor::tensor_COO<Float> &tens) {
    convert(tens);
    val_create_flag = true;
  }
 
  void convert(const matrix::Dense<Float> &dense);
 
  tensor_Dense(const matrix::Dense<Float> &dense) {
    convert(dense);
    val_create_flag = true;
  }
 
  void convert(const vector<Float> &vec);
 
  tensor_Dense(const vector<Float> &vec) {
    convert(vec);
    val_create_flag = true;
  }
 
  tensor_Dense(const std::vector<size_t> &shape);
 
  tensor_Dense(const std::vector<size_t> &shape, const Float *value);
 
  tensor_Dense(const std::vector<size_t> &shape,
               const std::vector<Float> &value);
 
  tensor_Dense(const std::vector<size_t> &shape, const Float min,
               const Float max);
 
  tensor_Dense(const std::vector<size_t> &shape, const Float min,
               const Float max, const std::uint32_t seed);
 
  tensor_Dense(const tensor_Dense<Float> &tens, Float value);
 
  void set_ptr(const std::vector<size_t> &shape,
               const std::vector<Float> &value);
 
  void set_ptr(const std::vector<size_t> &shape, const Float *value);
 
  [[nodiscard]] std::vector<size_t> get_shape() const { return shape; }
 
  [[nodiscard]] size_t get_nnz() const { return val_nnz; }
 
  void set_shape(const std::vector<size_t> &shape) { this->shape = shape; };
 
  [[nodiscard]] std::string type() const { return "tensor_Dense"; }
 
  // TODO
  [[nodiscard]] double get_data_size() const {
    return get_nnz() * sizeof(Float) / 1.0e+9;
  }
 
  [[nodiscard]] Float at(const std::vector<size_t> &pos) const;
 
  template <typename... Args>
  [[nodiscard]] Float at(const std::vector<size_t> &pos, const size_t dim,
                         const Args... args) const {
    std::vector<size_t> pos_copy = pos;
    pos_copy.push_back(dim);
    return this->at(pos_copy, args...);
  };
 
  template <typename... Args>
  [[nodiscard]] Float at(const size_t dim, const Args... args) const {
    std::vector<size_t> pos(1);
    pos[0] = dim;
    return this->at(pos, args...);
  };
 
  [[nodiscard]] Float at(const std::vector<size_t> &pos) {
    return static_cast<const tensor_Dense *>(this)->at(pos);
  };
 
  template <typename... Args>
  [[nodiscard]] Float at(const std::vector<size_t> &pos, const Args... args) {
    return static_cast<const tensor_Dense *>(this)->at(pos, args...);
  };
 
  template <typename... Args>
  [[nodiscard]] Float at(const size_t dim, const Args... args) {
    return static_cast<const tensor_Dense *>(this)->at(dim, args...);
  };
 
  void insert(const std::vector<size_t> &pos, const Float Val);
 
  void print_all(bool force_cpu = false) const;
 
  // communication
  // ///////////////////////////////////////////////////////////////////////////
  void send() const;
 
  void recv();
 
  void nonfree_recv();
 
  void device_free() const;
 
  [[nodiscard]] bool get_device_mem_stat() const { return gpu_status; }
 
  ~tensor_Dense() {
    if (val_create_flag) {
      if (get_device_mem_stat()) {
        device_free();
      }
    }
  }
 
  [[nodiscard]] const Float *data() const { return val.get(); }
 
  [[nodiscard]] Float *data() { return val.get(); }
 
  void resize(const size_t N, Float Val = 0) {
    if (get_device_mem_stat()) {
      throw std::runtime_error("Error, GPU matrix cant use resize");
    }
    if (val_create_flag) {
      std::shared_ptr<Float> tmp(new Float[N], std::default_delete<Float[]>());
      size_t copy_size = std::min(val_nnz, N);
      for (size_t i = 0; i < copy_size; ++i) {
        tmp.get()[i] = data()[i];
      }
      for (size_t i = copy_size; i < N; ++i) {
        tmp.get()[i] = Val;
      }
      val = tmp;
      alloc_nnz = N;
      val_nnz = N;
    } else {
      throw std::runtime_error("Error, not create vector cant use resize");
    }
  }
 
  void resize(const std::vector<size_t> &shape, Float val = 0) {
    size_t N = 1;
    for (auto n : shape) {
      N *= n;
    }
    resize(N, val);
    this->shape = shape;
  }
 
  void move(const matrix::Dense<Float> &dense);
 
  void move(const vector<Float> &vec);
 
  void fill(Float value);
 
  void operator=(const tensor_Dense<Float> &tens);
 
  [[nodiscard]] Float &operator[](size_t i) {
    if (get_device_mem_stat()) {
      throw std::runtime_error("Error, GPU vector cant use operator[]");
    }
    return data()[i];
  }
 
  [[nodiscard]] bool equal(const tensor_Dense<Float> &tens,
                           bool compare_cpu_and_device = false) const;
 
  [[nodiscard]] bool operator==(const tensor_Dense<Float> &tens) const;
 
  [[nodiscard]] bool operator!=(const tensor_Dense<Float> &tens) const;
 
  size_t get_index(const std::vector<size_t> &pos) const {
    if (pos.size() != this->shape.size()) {
      throw std::runtime_error("pos size should be same with the shape");
    }
    size_t ind = 0;
    for (auto i = 0; i < pos.size(); ++i) {
      ind *= this->shape[i];
      ind += pos[i];
    }
    return ind;
  }
 
  std::vector<size_t> get_index(const size_t pos) const {
    std::vector<size_t> ind(this->shape.size(), 0);
    auto pos_copy = pos;
    for (int i = (int)this->shape.size() - 1; i >= 0; --i) {
      ind[i] = pos_copy % this->shape[i];
      pos_copy /= this->shape[i];
    }
    return ind;
  }
 
  void reshape(const std::vector<int> &shape);
 
  template <typename... Args>
  void reshape(const std::vector<int> &shape, const size_t dim,
               const Args... args) {
    std::vector<int> shape_copy = shape;
    shape_copy.push_back(dim);
    reshape(shape_copy, args...);
    return;
  }
 
  template <typename... Args> void reshape(const int dim, const Args... args) {
    std::vector<int> shape(1);
    shape[0] = dim;
    reshape(shape, args...);
    return;
  }
 
 
  void diag_add(const Float alpha);
 
  void diag_sub(const Float alpha);
 
  void diag_mul(const Float alpha);
 
  void diag_div(const Float alpha);
 
  void diag_add(const vector<Float> &vec);
  void diag_add(const view1D<vector<Float>, Float> &vec);
  void diag_add(const view1D<matrix::Dense<Float>, Float> &vec);
  void diag_add(const view1D<tensor::tensor_Dense<Float>, Float> &vec);
 
  void diag_sub(const vector<Float> &vec);
  void diag_sub(const view1D<vector<Float>, Float> &vec);
  void diag_sub(const view1D<matrix::Dense<Float>, Float> &vec);
  void diag_sub(const view1D<tensor::tensor_Dense<Float>, Float> &vec);
 
  void diag_mul(const vector<Float> &vec);
  void diag_mul(const view1D<vector<Float>, Float> &vec);
  void diag_mul(const view1D<matrix::Dense<Float>, Float> &vec);
  void diag_mul(const view1D<tensor::tensor_Dense<Float>, Float> &vec);
 
  void diag_div(const vector<Float> &vec);
  void diag_div(const view1D<vector<Float>, Float> &vec);
  void diag_div(const view1D<matrix::Dense<Float>, Float> &vec);
  void diag_div(const view1D<tensor::tensor_Dense<Float>, Float> &vec);
};
} // namespace tensor
} // namespace monolish