matrice

Column major matrix data type in pure C with bindings

Features

• Basic vector and matrix type for int (int32), int64, float (float32) and double (float64).
• Factory functions: vector_ones,vector_zeros, vector_fill, vector_arange, matrix_ones, matrix_zeros, matrix_fill and matrix_arange.
• Elementwise vector and matrix operations vector_add, vector_sub, vector_mul, vector_div, matrix_add, matrix_sub, matrix_mul, and matrix_div.
• Matrix multiplication using auto vectorization
• Bindings for C++ with classes.
• Bindings for OCaml

Roadmap

• [] Matrix view
• [] Matrix transpose
• [] Bindings for Python.
• [] Matrix decomposition methods : QR, LU, Cholesky
• [] Determinant
• [] Inverse of a matrix

Data types and operations

• Data types

matrice	numpy	pytorch	Eigen
int; int32	np.int	torch.int32; torch.int	int
float; float32	np.float32	torch.float32; torch.float	float
double; float64	np.float64	torch.float64	double

• Empty, ones and zeros

matrice	numpy	pytorch	Eigen
matrix_make(type, shape);vector_make(type, size)	np.empty	torch.empty
matrix_zeros(type, shape);vector_zeros(type, size)	np.zeros	torch.zeros
matrix_ones(type, shape);vector_ones(type, size)	np.ones	torch.ones

Build and install

The following command will build and install the C shared library and header files.

make main
make install

The C++ bindings can be installed by running

make install-cpp

The OCaml bindings can be installed by typing

cd bindings/ocaml/matrice
dune build
dune install

Examples

• Uninitialized matrix

#include "matrix.h"

int main() {
    uint32_t shape[2] = {2, 3};
    matrix* mat = matrix_make(kfloat, shape);
    matrix_print(mat);
    matrix_free(mat);
}

• Matrix initialized with range

#include "matrix.h"

int main() {
    uint32_t shape[2] = {2, 3};
    float value = 1.f;
    matrix* mat = matrix_arange(kfloat, shape, &value);
    matrix_print(mat);
    matrix_free(mat);
}

• Example of using matrix_set and matrix_get

#include "matrix.h"

int main() {
   uint32_t shape[2] = {3, 3};
   matrix *mat = matrix_make(kfloat, shape);
   float x = 1.f;
   for (uint32_t j = 0; j < matrix_cols(mat); j++) {
      for (uint32_t i = 0; i < matrix_rows(mat); i++) {
         matrix_set(mat, i, j, &x);
         x += 1.f;
      }
   }
   matrix_print(mat);
   for (uint32_t j = 0; j < matrix_cols(mat); j++) {
      for (uint32_t i = 0; i < matrix_rows(mat); i++) {
         float value;
         matrix_get(mat, i, j, &value);
         printf("m[%i, %i] = %f\n", i, j, value);
      }
   }
   matrix_free(mat);
}

C++ Bindings

In C++, the classes matrice::vector and matrice::matrix are defined.

A basic example is the following:

#include <matrice/matrice.hxx>

int main() {
    using namespace matrice;
    matrix x = arange(4, 4, 1.0f);
    matrix y = arange(4, 4, 1.0f);
    auto z = matmul(x, y);
    std::cout << "shape = " << z.rows() << "x" << z.cols() << std::endl;
    std::cout << "size = " << z.size() << std::endl;
    std::cout << z;
    // Access indices with ()
    std::cout << z(0, 1) << std::endl;
    // Modify a value
    z.mutable_data()[0] = 1.0f;
    // Or modify it by its indices
    z(0,0) = 2.0f;
    // Access indices with the pointer to the data
    std::cout << z.data()[0] << std::endl;
}

OCaml Bindins

In OCaml, two records Vector.vector and Matrix.matrix are defined.

A working example is the following:

open Matrice

(*Create an uninitialized matrix of float*)
let rows = 3;;
let cols = 4;;
let a = Matrix.make Float32 rows cols;;
Matrix.print_endline a;;

(*Create an uninitialized vector of int*)
let b = Vector.make Int32 10;;
(*Set the values*)
for i = 0 to 9 do
    Vector.set_int32 b (Int32.of_int i)
done
Vector.print_endline b;;

(*Matrix multiplication*)
let x = Matrix.arange_float 4 4 1.0;;
let y = Matrix.arange_float 4 4 1.0;;
Matrix.print_endline matmul x y;;