usbalbin/linear_algebra

Linear algebra library for rust

linear_algebra's Issues

Memory leak

When using an AMD Vega 56 as the computing device the following code seems to create a massive memory leak when doing many iterations. It appears as if the old m1's underlaying buffer is never freed thus burning through device memory in a matter of seconds.

`

  use matrix::*;
  let x = 200;

  let mut m1: Matrix<f32> = Matrix::from_vec(
      (0..(rows * cols)).map(|x| x as f32 / (rows * cols) as f32).collect(),
      x,//Rows
      x//Cols
  );

  let m2: Matrix<f32> = Matrix::from_vec(
      (0..(rows * cols)).map(|x| x as f32 / (rows * cols * rows) as f32).collect(),
      x,//Rows
      x//Cols
  );

  for _ in 0..100000 {
      m1 = &m1 * &m2; // <- Leak here
  }
  println!("{:?}", m1);

`

This issue does not appear when using an Intel i7 3930k as the computing device.

Not thread safe

linear_algebra is currently not thread safe due to multiple threads having mutable access to static data. References to this data is currently handed out from cl_data() which is called for every math operation on vectors or matrices.

Add sort() to Vector

Implement some sort of sorting algorithm for sorting Vector elements

Can only use Vector/Matrices over one type at a time

Currently all kernels are compiled only once. This leads to all kernels being compiled ONLY for the element type of the first math operation, which is very bad.