devectorizing matrices

#!/usr/bin/julia

#julia matrix test program.

v = rand(5)
m = rand(5,5)
r = zeros(5)

println("right product")
tic()
	for idx=1:100
		r = m * v
	end
toc()  #==> 0.25 seconds

v = v'

println("left product")
#uses memory-appropriate ordering, but performs worse.
tic()
	for idx=1:100
		r = v * m
	end
toc()  #==> 0.28 seconds

println("poorly devectorized product")
#a first, naive shot at devectorization
v = rand(5)
m = rand(25)
tic()
	for idx=1:100
		r = zeros(5)
		for jdx = 1:5
			q = jdx * 5
			r += m[q - 4 : q] .* v
		end
	end
toc()  #==> 0.50 seconds


println("devectorized product")
# reduce number of operations in the matrix multiplication loop.
# use short-cycle addition operations only.
tic()
	for idx=1:100
		r = zeros(5)
		a = 1
		b = 5
		for jdx = 1:5
			r += m[a : b] .* v
			a += 5
			b += 5
		end
	end
toc()  #==> 0.01 seconds!