ntuple for inference

[ChrisRackauckas]

On this example:

using DifferentialEquations, SnoopCompile

function lorenz(du,u,p,t)
 du[1] = 10.0(u[2]-u[1])
 du[2] = u[1]*(28.0-u[3]) - u[2]
 du[3] = u[1]*u[2] - (8/3)*u[3]
end

u0 = [1.0;0.0;0.0]
tspan = (0.0,100.0)
prob = ODEProblem(lorenz,u0,tspan)
alg = Rodas5(chunk_size = Val{3}())
tinf = @snoopi_deep solve(prob,alg)
solve(prob,alg)

we saw just two inference triggers:

julia> itrigs = inference_triggers(tinf)
2-element Vector{InferenceTrigger}:
 Inference triggered to call setindex!(::Vector{Tuple{Float64, Float64, Float64}}, ::Tuple{Bool, Bool, Bool}, ::Int64) from generate_chunked_partials (C:\Users\accou\.julia\dev\SparseDiffTools\src\differentiation\compute_jacobian_ad.jl:75) with specialization SparseDiffTools.generate_chunked_partials(::Vector{Float64}, ::UnitRange{Int64}, ::Val{3})
 Inference triggered to call OrdinaryDiffEq.jacobian2W!(::Matrix{Float64}, ::LinearAlgebra.UniformScaling{Bool}, ::Float64, ::Matrix{Float64}, ::Bool) called from toplevel

The second one is clear: okay, just use ntuple on like 75 right? So I did that, and it made the function more type-unstable? Look at the itrigs after that:

julia> itrigs = inference_triggers(tinf)
6-element Vector{InferenceTrigger}:
 Inference triggered to call hcat(::BitMatrix, ::BitMatrix) from generate_chunked_partials (C:\Users\accou\.julia\dev\SparseDiffTools\src\differentiation\compute_jacobian_ad.jl:67) with specialization SparseDiffTools.generate_chunked_partials(::Vector{Float64}, ::UnitRange{Int64}, ::Val{3})
 Inference triggered to call Vector{Tuple{Float64, Float64, Float64}}(::UndefInitializer, ::Int64) from generate_chunked_partials (C:\Users\accou\.julia\dev\SparseDiffTools\src\differentiation\compute_jacobian_ad.jl:73) with specialization SparseDiffTools.generate_chunked_partials(::Vector{Float64}, ::UnitRange{Int64}, ::Val{3})
 Inference triggered to call ntuple(::SparseDiffTools.var"#15#16"{3, Int64, Int64}, ::Val{3}) from generate_chunked_partials (C:\Users\accou\.julia\dev\SparseDiffTools\src\differentiation\compute_jacobian_ad.jl:75) with specialization SparseDiffTools.generate_chunked_partials(::Vector{Float64}, ::UnitRange{Int64}, ::Val{3})
 Inference triggered to call setindex!(::Vector{Tuple{Float64, Float64, Float64}}, ::Tuple{Bool, Bool, Bool}, ::Int64) from generate_chunked_partials (C:\Users\accou\.julia\dev\SparseDiffTools\src\differentiation\compute_jacobian_ad.jl:75) with specialization SparseDiffTools.generate_chunked_partials(::Vector{Float64}, ::UnitRange{Int64}, ::Val{3})
 Inference triggered to call setindex!(::Vector{Vector{Tuple{Float64, Float64, Float64}}}, ::Vector{Tuple{Float64, Float64, Float64}}, ::Int64) from generate_chunked_partials (C:\Users\accou\.julia\dev\SparseDiffTools\src\differentiation\compute_jacobian_ad.jl:77) with specialization SparseDiffTools.generate_chunked_partials(::Vector{Float64}, ::UnitRange{Int64}, ::Val{3})
 Inference triggered to call OrdinaryDiffEq.jacobian2W!(::Matrix{Float64}, ::LinearAlgebra.UniformScaling{Bool}, ::Float64, ::Matrix{Float64}, ::Bool) called from toplevel

@chriselrod or @yingboma could I get some Cthulhu magic to look at what's going on there?

[codecov-commenter]

Codecov Report

Merging #168 (8691112) into master (f10cda4) will increase coverage by 0.02%. The diff coverage is 100.00%.

@@            Coverage Diff             @@
##           master     #168      +/-   ##
==========================================
+ Coverage   78.86%   78.89%   +0.02%     
==========================================
  Files          14       14              
  Lines         743      744       +1     
==========================================
+ Hits          586      587       +1     
  Misses        157      157              

Impacted Files	Coverage Δ
src/differentiation/compute_jacobian_ad.jl	94.17% <100.00%> (+0.02%)	:arrow_up:

---

Continue to review full report at Codecov.

Legend - Click here to learn more Δ = absolute <relative> (impact), ø = not affected, ? = missing data Powered by Codecov. Last update f10cda4...8691112. Read the comment docs.

[ChrisRackauckas]

Nope, @YingboMa that also causes:

julia> itrigs = inference_triggers(tinf)
6-element Vector{InferenceTrigger}:
 Inference triggered to call hcat(::BitMatrix, ::BitMatrix) from generate_chunked_partials (C:\Users\accou\.julia\dev\SparseDiffTools\src\differentiation\compute_jacobian_ad.jl:67) with specialization SparseDiffTools.generate_chunked_partials(::Vector{Float64}, ::UnitRange{Int64}, ::Val{3})
 Inference triggered to call Vector{Tuple{Float64, Float64, Float64}}(::UndefInitializer, ::Int64) from generate_chunked_partials (C:\Users\accou\.julia\dev\SparseDiffTools\src\differentiation\compute_jacobian_ad.jl:73) with specialization SparseDiffTools.generate_chunked_partials(::Vector{Float64}, ::UnitRange{Int64}, ::Val{3})      
 Inference triggered to call ntuple(::SparseDiffTools.var"#15#16"{3, Int64, Int64}, ::Val{3}) from generate_chunked_partials (C:\Users\accou\.julia\dev\SparseDiffTools\src\differentiation\compute_jacobian_ad.jl:75) with specialization SparseDiffTools.generate_chunked_partials(::Vector{Float64}, ::UnitRange{Int64}, ::Val{3})
 Inference triggered to call setindex!(::Vector{Tuple{Float64, Float64, Float64}}, ::Tuple{Bool, Bool, Bool}, ::Int64) from generate_chunked_partials (C:\Users\accou\.julia\dev\SparseDiffTools\src\differentiation\compute_jacobian_ad.jl:75) with specialization SparseDiffTools.generate_chunked_partials(::Vector{Float64}, ::UnitRange{Int64}, ::Val{3})
 Inference triggered to call setindex!(::Vector{Vector{Tuple{Float64, Float64, Float64}}}, ::Vector{Tuple{Float64, Float64, Float64}}, ::Int64) from generate_chunked_partials (C:\Users\accou\.julia\dev\SparseDiffTools\src\differentiation\compute_jacobian_ad.jl:77) with specialization SparseDiffTools.generate_chunked_partials(::Vector{Float64}, ::UnitRange{Int64}, ::Val{3})
 Inference triggered to call OrdinaryDiffEq.jacobian2W!(::Matrix{Float64}, ::UniformScaling{Bool}, ::Float64, ::Matrix{Float64}, ::Bool) called from toplevel

[YingboMa]

I'll try to play with this locally and submit a PR.

[chriselrod]

Actually, this seems better: https://github.com/JuliaDiff/SparseDiffTools.jl/pull/169/files