matrixjuliamatrix-multiplicationmemory-efficient

Julia implementation of mul! when Y = A?

Matrix multiplication Y = A * B can be implemented by mul!(Y, A, B) to save on memory allocations. But mul! can't be used if Y = A. Is there a similarly efficient way to calculate Y *= B? Or if not, what is the most efficient way to do matrix multiplication Y *= B

Small working example:

n = 10
A = rand(n,n)
B = rand(n,n)
Y = zeros(n,n)

#mul! removes allocations 
@allocated Y = A * B   #896
@allocated mul!(Y, A, B) #0    

#mul! can't be applied in this case
@allocated Y *= B #896

#desired function performance
@allocated mul_2!(Y, B)  #0

Thanks in advance for your help!

Solution

  • I don't think you can implement this efficiently (due to how blocking in matrix multiplication works). You're better off just keeping around another matrix of the appropriate size to use as a buffer.