Version 1.7 much faster than 1.6 in a script that calls wrapped external libraries

mkitti · July 20, 2021, 5:39am

The main difference seems to be in the execution of gmt_main.jl:866 at GMT v0.35.0:

Julia 1.6.0

mkitti:

66 @GMT\src\gmt_main.jl:866; grid_init(API::Ptr{Nothing}, X::GMT.GMT_RESOURCE, Grid::GMT.GMTgrid{Float32, 2}, pad::Int64, cube::Bool)
╎  1  @Base\array.jl:801; getindex
╎  33 @Base\operators.jl:560; +

Julia 1.7.0-DEV

Julia 1.6.0 seems to spend much more time on Line 560 of base/operators.jl which involves argument list folding via afoldl:

github.com

JuliaLang/julia/blob/v1.6.0/base/operators.jl#L560


      
              end
              return y
          end
          typeof(afoldl).name.mt.max_args = 18
          
          for op in (:+, :*, :&, :|, :xor, :min, :max, :kron)
              @eval begin
                  # note: these definitions must not cause a dispatch loop when +(a,b) is
                  # not defined, and must only try to call 2-argument definitions, so
                  # that defining +(a,b) is sufficient for full functionality.
                  ($op)(a, b, c, xs...) = afoldl($op, ($op)(($op)(a,b),c), xs...)
                  # a further concern is that it's easy for a type like (Int,Int...)
                  # to match many definitions, so we need to keep the number of
                  # definitions down to avoid losing type information.
              end
          end
          
          function kron! end
          
          const var"'" = adjoint

While that line has not changed in Julia 1.7, afoldl has changed:

It looks like you have @kslimes to thank for this amazing improvement in speed.

Topic		Replies	Views
Understanding Julia performance in simple finite difference code New to Julia	38	4571	June 7, 2018
Julia 1.0 released Announcements	87	13222	September 21, 2018
Any benchmark of Julia v1.0 vs older versions Performance	66	8348	April 3, 2019
Kron vs scalar product speed difference. python code faster? New to Julia question	41	4259	January 14, 2017
Why is this Julia code considerably slower than Matlab New to Julia performance	64	8635	March 5, 2017

Version 1.7 much faster than 1.6 in a script that calls wrapped external libraries

Related topics