FMGPU: Fast Matrix operations of the GPU

This library allows to efficiently carry out matrix operations using GPU acceleration. The package is available on npm as js-fmgpu. The following functions are currently available:

Todo

The following functions are to be implemented next:

• det (GPU accelerated)
• map
• inverse

scale

Multiply input matrix by a scalar value. B = fmgpu.scale(A, alpha)

add

Add two matrices together. C = fmgpu.add(A, B)

sub

Subtract second matrix from first one C = fmgpu.sub(A, B)

transpose

Transpose the given matrix B = fmgpu.transpose(A)

hadamard

Returns Hadamard product, also known as pointwise product between two matrices C = fmgpu.hadamard(A, B)

dot

Returns the dot product between two matrices. C = fmgpu.dot(A, B)

detJS (pure JS)

Returns the determinant of a given matrix. Note that this function is not yet GPU-accelerated. B = fmgpu.detJS(A)

equals

Returns true if two matrices are equal (within some epsilon value) eq = fmgpu.equals(A, B, 1e-6)

solveLinearSystemSmall (pure JS)

Solves a linear system of equations using pure JS code (less overhead, better for smaller matrices, of size < 200) x = fmgpu.solveLinearSystemSmall(A, b)

solveLinearSystem

Solves a linear system of equations using GPU-accelerated code. (Adapted for matrices of size N > 200) x = fmgpu.solveLinearSystem(A, b)