monolish_dense.hpp

Go to the documentation of this file.

#pragma once
#include "monolish_matrix.hpp"
#include "monolish_tensor.hpp"
#include <iostream>
#include <memory>
 
namespace monolish {
template <typename Float> class vector;
template <typename TYPE, typename Float> class view1D;
template <typename TYPE, typename Float> class view_Dense;
template <typename TYPE, typename Float> class view_tensor_Dense;
 
namespace tensor {
template <typename Float> class tensor_Dense;
}
 
namespace matrix {
 
template <typename Float> class Dense {
private:
  size_t rowN;
 
  size_t colN;
 
  // size_t nnz;
 
  mutable std::shared_ptr<bool> gpu_status = std::make_shared<bool>(false);
 
  size_t first = 0;
 
public:
  std::shared_ptr<Float> val;
 
  size_t val_nnz = 0;
 
  size_t alloc_nnz = 0;
 
  bool val_create_flag = false;
 
  Dense() { val_create_flag = true; }
 
  void convert(const COO<Float> &coo);
 
  void convert(const Dense<Float> &dense);
 
  Dense(const COO<Float> &coo) {
    val_create_flag = true;
    convert(coo);
  }
 
  Dense(const Dense<Float> &dense);
 
  Dense(const Dense<Float> &dense, Float value);
 
  Dense(const view_Dense<vector<Float>, Float> &dense);
 
  Dense(const view_Dense<matrix::Dense<Float>, Float> &dense);
 
  Dense(const view_Dense<tensor::tensor_Dense<Float>, Float> &dense);
 
  Dense(const size_t M, const size_t N);
 
  Dense(const size_t M, const size_t N, const Float *value);
 
  Dense(const size_t M, const size_t N, const std::vector<Float> &value);
 
  Dense(const size_t M, const size_t N, const vector<Float> &value);
 
  Dense(const size_t M, const size_t N,
        const std::initializer_list<Float> &list);
 
  Dense(const size_t M, const size_t N, const Float min, const Float max);
 
  Dense(const size_t M, const size_t N, const Float min, const Float max,
        const std::uint32_t seed);
 
  Dense(const size_t M, const size_t N, const Float value);
 
  void set_ptr(const size_t M, const size_t N, const std::vector<Float> &value);
 
  void set_ptr(const size_t M, const size_t N, const Float *value);
 
  void set_ptr(const size_t M, const size_t N, const Float value);
 
  [[nodiscard]] size_t get_row() const { return rowN; }
 
  [[nodiscard]] size_t get_col() const { return colN; }
 
  [[nodiscard]] std::shared_ptr<Float> get_val() { return val; }
 
  [[nodiscard]] size_t get_nnz() const { return val_nnz; }
 
  [[nodiscard]] size_t get_alloc_nnz() const { return alloc_nnz; }
 
  [[nodiscard]] size_t get_first() const { return first; }
 
  [[nodiscard]] size_t get_offset() const { return get_first(); }
 
  void set_row(const size_t N) { rowN = N; };
 
  void set_col(const size_t M) { colN = M; };
 
  // void set_nnz(const size_t NZ) { val_nnz = NZ; };
 
  void set_first(size_t i) { first = i; }
 
  [[nodiscard]] std::string type() const { return "Dense"; }
 
  void transpose();
 
  void transpose(const Dense &B);
 
  [[nodiscard]] double get_data_size() const {
    return get_nnz() * sizeof(Float) / 1.0e+9;
  }
 
  [[nodiscard]] Float at(const size_t i) const;
 
  [[nodiscard]] Float at(const size_t i, const size_t j) const;
 
  [[nodiscard]] Float at(const size_t i) {
    return static_cast<const Dense *>(this)->at(i);
  };
 
  [[nodiscard]] Float at(const size_t i, const size_t j) {
    return static_cast<const Dense *>(this)->at(i, j);
  };
 
  void insert(const size_t i, const Float Val);
 
  void insert(const size_t i, const size_t j, 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; }
 
  [[nodiscard]] std::shared_ptr<bool> get_gpu_status() const {
    return gpu_status;
  }
 
  ~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(size_t N, Float Val = 0) {
    if (first + N < alloc_nnz) {
      for (size_t i = val_nnz; i < N; ++i) {
        begin()[i] = Val;
      }
      val_nnz = N;
      return;
    }
    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;
      first = 0;
    } else {
      throw std::runtime_error("Error, not create vector cant use resize");
    }
  }
 
  void move(const tensor::tensor_Dense<Float> &tensor_dense);
 
  void move(const tensor::tensor_Dense<Float> &tensor_dense, int rowN,
            int colN);
 
  void move(const view_tensor_Dense<vector<Float>, Float> &tensor_dense);
 
  void move(const view_tensor_Dense<matrix::Dense<Float>, Float> &tensor_dense);
 
  void move(const view_tensor_Dense<tensor::tensor_Dense<Float>, Float>
                &tensor_dense);
 
  void move(const view_tensor_Dense<vector<Float>, Float> &tensor_dense,
            int rowN, int colN);
 
  void move(const view_tensor_Dense<matrix::Dense<Float>, Float> &tensor_dense,
            int rowN, int colN);
 
  void move(
      const view_tensor_Dense<tensor::tensor_Dense<Float>, Float> &tensor_dense,
      int rowN, int colN);
 
  [[nodiscard]] const Float *begin() const { return data() + get_offset(); }
 
  [[nodiscard]] Float *begin() { return data() + get_offset(); }
 
  [[nodiscard]] const Float *end() const {
    return data() + get_offset() + get_nnz();
  }
 
  [[nodiscard]] Float *end() { return data() + get_offset() + get_nnz(); }
 
 
  void diag(vector<Float> &vec) const;
  void diag(view1D<vector<Float>, Float> &vec) const;
  void diag(view1D<matrix::Dense<Float>, Float> &vec) const;
  void diag(view1D<tensor::tensor_Dense<Float>, Float> &vec) const;
 
  void row(const size_t r, vector<Float> &vec) const;
  void row(const size_t r, view1D<vector<Float>, Float> &vec) const;
  void row(const size_t r, view1D<matrix::Dense<Float>, Float> &vec) const;
  void row(const size_t r,
           view1D<tensor::tensor_Dense<Float>, Float> &vec) const;
 
  void col(const size_t c, vector<Float> &vec) const;
  void col(const size_t c, view1D<vector<Float>, Float> &vec) const;
  void col(const size_t c, view1D<matrix::Dense<Float>, Float> &vec) const;
  void col(const size_t c,
           view1D<tensor::tensor_Dense<Float>, Float> &vec) const;
 
 
  void fill(Float value);
 
  void operator=(const Dense<Float> &mat);
 
  void operator=(const view_Dense<vector<Float>, Float> &mat);
 
  void operator=(const view_Dense<matrix::Dense<Float>, Float> &mat);
 
  void operator=(const view_Dense<tensor::tensor_Dense<Float>, Float> &mat);
 
  [[nodiscard]] Float &operator[](size_t i) {
    if (get_device_mem_stat()) {
      throw std::runtime_error("Error, GPU vector cant use operator[]");
    }
    return data()[first + i];
  }
 
  [[nodiscard]] bool equal(const Dense<Float> &mat,
                           bool compare_cpu_and_device = false) const;
 
  [[nodiscard]] bool operator==(const Dense<Float> &mat) const;
 
  [[nodiscard]] bool operator!=(const Dense<Float> &mat) const;
 
  void reshape(const size_t row, const size_t col);
 
 
  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 matrix
} // namespace monolish