matvec.jl

using BenchmarkTools
BLAS.set_num_threads(1)

function BLAS2(A,x)
    A*x
end
function loop_mn(A,x)
    m,n = size(A)
    y = zeros(m)
    @inbounds @simd for i=1:m
        for j=1:n
            y[i] += A[i,j]*x[j]
        end
    end
end
function loop_nm(A,x)
    m,n = size(A)
    y = zeros(m)
    @inbounds @simd for j=1:n
        for i=1:m
            y[i] += A[i,j]*x[j]
        end
    end
end
function views_mn(A,x)
    m,n = size(A)
    y = zeros(m)
    @inbounds for i=1:m
        @views y[i] = dot(A[i,:],x)
    end
end
function noviews_mn(A,x)
    m,n = size(A)
    y = zeros(m)
    @inbounds for i=1:m
        y[i] = dot(A[i,:],x)
    end
end
function views_nm(A,x)
    m,n = size(A)
    y = zeros(m)
    @inbounds for i=1:n
        @views y .+= x[i]*A[:,i]
    end
end
function noviews_nm(A,x)
    m,n = size(A)
    y = zeros(m)
    @inbounds for i=1:n
        y .+= x[i]*A[:,i]
    end
end
m = 10
n = 10000
A = randn(m,n)
x = randn(n)
@btime BLAS2(A,x)
@btime loop_mn(A,x)
@btime loop_nm(A,x)
@btime views_mn(A,x)
@btime views_nm(A,x)
@btime noviews_mn(A,x)
@btime noviews_nm(A,x)

A = [randn(n) for i=1:m]
function array_of_arrays(A,x)
    y = zeros(size(A,1))
    for i = 1:size(A,1)
        y[i] = dot(A[i],x)
    end
end
@btime array_of_arrays(A,x)