Preallocating Vector of Vectors

General Usage Performance
,
20

Interesting. Here, with 1.7, the loop version is improved (considering the correction in makearray3 pointed above):


julia> @btime makearr2(Float64, 500, 1000);
  195.416 μs (1000 allocations: 3.89 MiB)

julia> @btime makearr3(Float64, 500, 1000);
  83.967 μs (501 allocations: 3.88 MiB)

julia> @btime makearr4(Val(Float64), 500, 1000);
  85.357 μs (501 allocations: 3.88 MiB)

In Julia 1.6 I have:

julia> @btime makearr3(Float64, 500, 1000);
  189.542 μs (1001 allocations: 3.89 MiB)

Thus, the gain associated to the specialization in 1.6 is given in 1.7 for some reason.

1 Like