Am I not reading the profile right?
By now the single slowest line is “word_sim5: line 4”, about 20%+ (but three rand lines combined, 5, 9 and 16, about 75% of the time).
Do the numbers for that line make sense (it seems already fast, and with “undef” should be O(1), as not initializing):
julia> @time words = Array{Int32}(undef, omega);
0.000003 seconds (1 allocation: 496 bytes)
julia> omega = 100000
100000
julia> @time words = Array{Int32}(undef, omega);
0.000008 seconds (2 allocations: 390.703 KiB)
julia> omega = 1000000000
1000000000
julia> @time words = Array{Int32}(undef, omega);
0.012678 seconds (2 allocations: 3.725 GiB, 99.61% gc time)
julia> @time words = Array{Int32}(undef, omega);
0.004373 seconds (2 allocations: 3.725 GiB, 99.06% gc time)
julia> @time words = Array{Int32}(undef, omega);
0.047590 seconds (2 allocations: 3.725 GiB, 99.92% gc time)
For those extra test I did (unrelated to the program, really), I’m exercising the GC [with @btime I do see the same allocations, but “gc time” is gone, as it takes care of avoiding GC], but why? It seems I’m testing Clibs malloc, and it’s not O(1), even then why GC?
[Funny story, I have 128 GB in mem. so the computer is not swapping, I thought I had 12 GB, and was told, “that can’t be right, we didn’t give you only 12 GB”, I had never seen the + before in top so disregarded a digit cut off:
KiB Mem : 13182060+total, 10671788+free, 14977804 used, 10124916 buff/cache
KiB Swap: 2097148 total, 2097148 free, 0 used. 11454323+avail Mem ]