Use SnoopPrecompile?

[mkitti]

We may be able to accelerate GR.jl further by using SnoopPrecompile.

using SnoopPrecompile
function __init__()
    @precompile_setup begin
        x = [1, 3, 2, 4, 5, 7, 6, 9, 8]
        y = Float64[1, 3, 2, 4, 5, 7, 6, 9, 8]
        @precompile_all_calls begin
            plot(x)
            plot(y)
            plot(x,y)
        end
    end
end

julia> using SnoopPrecompile

julia> SnoopPrecompile.verbose[] = true
true


julia> using GR
[ Info: Precompiling GR [28b8d3ca-fb5f-59d9-8090-bfdbd6d07a71]
Warning: Ignoring XDG_SESSION_TYPE=wayland on Gnome. Use QT_QPA_PLATFORM=wayland to run on Wayland anyway.
MethodInstance for GR.jlgr.isrowvec(::Vector{Int64})
MethodInstance for GR.jlgr.isvector(::StepRangeLen{Float64, Base.TwicePrecision{Float64}, Base.TwicePrecision{Float64}, Int64})
MethodInstance for vec(::StepRangeLen{Float64, Base.TwicePrecision{Float64}, Base.TwicePrecision{Float64}, Int64})
MethodInstance for GR.jlgr.isvector(::Vector{Int64})
MethodInstance for GR.jlgr.set_viewport(::Symbol, ::Vector{Int64}, ::GR.jlgr.PlotObject)
MethodInstance for /(::Float64, ::Int32)
MethodInstance for GR.setwsviewport(::Int64, ::Float64, ::Int64, ::Float64)
MethodInstance for GR.setwswindow(::Int64, ::Int64, ::Int64, ::Float64)
MethodInstance for GR.jlgr.set_window(::Symbol, ::GR.jlgr.PlotObject)
MethodInstance for GR.jlgr.Extrema64(::StepRangeLen{Float64, Base.TwicePrecision{Float64}, Base.TwicePrecision{Float64}, Int64})
MethodInstance for GR.jlgr.Extrema64(::Vector{Int64})
MethodInstance for GR.jlgr.fix_minmax(::Float64, ::Float64)
MethodInstance for GR.jlgr.fix_minmax(::Int64, ::Int64)
MethodInstance for GR.adjustlimits(::Float64, ::Float64)
MethodInstance for GR.jlgr.auto_tick(::Float64, ::Float64)
MethodInstance for GR.setwindow(::Float64, ::Float64, ::Float64, ::Float64)
MethodInstance for GR.jlgr.draw_axes(::Symbol)
MethodInstance for Base.indexed_iterate(::Tuple{Float64, Tuple{Float64, Float64}, Int64}, ::Int64)
MethodInstance for Base.literal_pow(::typeof(^), ::Float64, ::Val{2})
MethodInstance for GR.setcharheight(::Float64)
MethodInstance for GR.grid(::Float64, ::Float64, ::Int64, ::Int64, ::Int64, ::Int64)
MethodInstance for GR.axes(::Float64, ::Float64, ::Float64, ::Float64, ::Int64, ::Int64, ::Float64)
MethodInstance for GR.polyline(::StepRangeLen{Float64, Base.TwicePrecision{Float64}, Base.TwicePrecision{Float64}, Int64}, ::Vector{Int64})
MethodInstance for GR.jlgr.isrowvec(::Vector{Float64})
MethodInstance for GR.jlgr.isvector(::Vector{Float64})
MethodInstance for GR.jlgr.Extrema64(::Vector{Float64})
MethodInstance for GR.polyline(::StepRangeLen{Float64, Base.TwicePrecision{Float64}, Base.TwicePrecision{Float64}, Int64}, ::Vector{Float64})
MethodInstance for GR.jlgr.var"#plot#34"(::Base.Pairs{Symbol, Union{}, Tuple{}, NamedTuple{(), Tuple{}}}, ::typeof(GR.jlgr.plot), ::Vector{Int64}, ::Vararg{Union{AbstractString, Function, AbstractVector, AbstractMatrix}})
MethodInstance for GR.polyline(::Vector{Int64}, ::Vector{Float64})

@t-bltg @sjkelly I'm a bit confused about the state of precompilation between here and Plots.jl. I was under the impression we already have done the above, but I see precompile.jl is quite limited.

[sjkelly]

The precomile sparsity is deliberate. In the pre-1.8 world I had done some benchmarks with such statements and it was not helpful to overall load+compile time. I presume on 1.8+ that will not be the case, but it may be helpful to version block.

[mkitti]

Thanks for clarifying. As @jheinen pointed out, we are seeing some significant acceleration with Julia 1.9 due to native code caching. I suspect we can do even better with more precompilation. Perhaps, we should restrict it to Julia 1.9+.