Thanks! I still don't understand why something this simple, even in the global scope, can't be optimized.
Also, any idea why the regression in run time from 0.3 to 0.4? On Friday, October 10, 2014 11:05:02 PM UTC-5, Patrick O'Leary wrote: > > > http://julia.readthedocs.org/en/release-0.3/manual/performance-tips/#avoid-global-variables > > On Friday, October 10, 2014 10:56:36 PM UTC-5, David Smith wrote: >> >> This code runs in 0.5 sec in v0.3.2, but takes 0.64 s in v0.4: >> tic() >> N = 256 >> n = 80 >> x = rand(N,N) + 1im*randn(N,N) >> f = zeros(Complex128, N, N, n) >> for t = 1:n >> f[:,:,t] = fft(x + float(t)) >> end >> toc() >> >> If I enclose it in a function, as in the following, and then run it by >> calling the function, it runs in 0.3 s in both v0.3.2 and v0.4. >> function bench() >> tic() >> N = 256 >> n = 80 >> x = rand(N,N) + 1im*randn(N,N) >> f = zeros(Complex128, N, N, n) >> for t = 1:n >> f[:,:,t] = fft(x + float(t)) >> end >> toc() >> end >> bench() >> >> >> What is going on? >> >> >> >>
