create_array slow for large static arrays

[JakobAsslaender]

Hi,
I would like to auto-generate functions that create large (roughly 10x10) static matrices. However, the call of SymbolicUtils.Code.create_array seems slow as shown in the MWE below. Some brief tests suggest that the problem might be the elems... in

SymbolicUtils.jl/src/code.jl

Line 565 in 835fb0d

SArray{Tuple{dims...}}(elems...)

Many thanks for looking into this!!!

MWE:

using StaticArrays
using Symbolics
using BenchmarkTools 

function symbolics_output(a, b, c, d)
    #= /Users/asslaj01/.julia/packages/SymbolicUtils/8d6hE/src/code.jl:373 =#
    #= /Users/asslaj01/.julia/packages/SymbolicUtils/8d6hE/src/code.jl:374 =#
    #= /Users/asslaj01/.julia/packages/SymbolicUtils/8d6hE/src/code.jl:375 =#
    begin
        #= /Users/asslaj01/.julia/packages/SymbolicUtils/8d6hE/src/code.jl:468 =#
        (SymbolicUtils.Code.create_array)(SArray, nothing, Val{2}(), Val{(8, 8)}(), a, b, c, d, a, b, c, d, a, b, c, d, a, b, c, d, a, b, c, d, a, b, c, d, a, b, c, d, a, b, c, d, a, b, c, d, a, b, c, d, a, b, c, d, a, b, c, d, a, b, c, d, a, b, c, d, a, b, c, d, a, b, c, d)
    end
end

function symbolics_optimized(a, b, c, d)
    SMatrix{8,8,Float64,64}(a, b, c, d, a, b, c, d, a, b, c, d, a, b, c, d, a, b, c, d, a, b, c, d, a, b, c, d, a, b, c, d, a, b, c, d, a, b, c, d, a, b, c, d, a, b, c, d, a, b, c, d, a, b, c, d, a, b, c, d, a, b, c, d)
end

@benchmark symbolics_output(1.0, 2.0, 3.0, 4.0)
@benchmark symbolics_optimized(1.0, 2.0, 3.0, 4.0)

Output:

BenchmarkTools.Trial: 10000 samples with 8 evaluations.
 Range (min … max):  3.114 μs …  12.461 μs  ┊ GC (min … max): 0.00% … 0.00%
 Time  (median):     3.191 μs               ┊ GC (median):    0.00%
 Time  (mean ± σ):   3.221 μs ± 319.398 ns  ┊ GC (mean ± σ):  0.00% ± 0.00%

     ▄█▆▁                                                      
  ▂▃▅████▅▃▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▁▁▂▂▁▁▂▁▂▁▁▁▂▁▁▁▂▂▂▂▂ ▃
  3.11 μs         Histogram: frequency by time        4.04 μs <

 Memory estimate: 3.12 KiB, allocs estimate: 68.

vs.

BenchmarkTools.Trial: 10000 samples with 1000 evaluations.
 Range (min … max):  5.775 ns … 61.289 ns  ┊ GC (min … max): 0.00% … 0.00%
 Time  (median):     5.796 ns              ┊ GC (median):    0.00%
 Time  (mean ± σ):   5.861 ns ±  0.928 ns  ┊ GC (mean ± σ):  0.00% ± 0.00%

  ▄█▃                                                         
  ███▅▃▃▃▃▂▂▂▂▂▂▂▂▂▂▂▁▁▂▁▁▁▁▁▁▂▂▂▁▁▁▂▁▂▁▁▂▂▂▂▂▂▁▁▁▁▁▂▁▁▂▂▂▂▂ ▂
  5.78 ns        Histogram: frequency by time        6.47 ns <

 Memory estimate: 0 bytes, allocs estimate: 0.

[ernestds]

Does simply removing the ... solve the problem? Or does it create other problems somewhere? (Julia noob here)

using StaticArrays
using Symbolics
using BenchmarkTools 

@inline function create_array2(::Type{<:SArray}, ::Nothing, nd::Val, ::Val{dims}, elems...) where dims
    SArray{Tuple{dims...}}(elems)
end
function symbolics_output(a, b, c, d)
    #= /Users/asslaj01/.julia/packages/SymbolicUtils/8d6hE/src/code.jl:373 =#
    #= /Users/asslaj01/.julia/packages/SymbolicUtils/8d6hE/src/code.jl:374 =#
    #= /Users/asslaj01/.julia/packages/SymbolicUtils/8d6hE/src/code.jl:375 =#
    begin
        #= /Users/asslaj01/.julia/packages/SymbolicUtils/8d6hE/src/code.jl:468 =#
        (create_array2)(SArray, nothing, Val{2}(), Val{(8, 8)}(), a, b, c, d, a, b, c, d, a, b, c, d, a, b, c, d, a, b, c, d, a, b, c, d, a, b, c, d, a, b, c, d, a, b, c, d, a, b, c, d, a, b, c, d, a, b, c, d, a, b, c, d, a, b, c, d, a, b, c, d, a, b, c, d)
    end
end

@benchmark symbolics_output(1.0, 2.0, 3.0, 4.0)

BenchmarkTools.Trial: 10000 samples with 1000 evaluations.
 Range (min … max):  3.917 ns … 17.943 ns  ┊ GC (min … max): 0.00% … 0.00%
 Time  (median):     3.967 ns              ┊ GC (median):    0.00%
 Time  (mean ± σ):   4.010 ns ±  0.371 ns  ┊ GC (mean ± σ):  0.00% ± 0.00%

   █                                                          
  ▅█▄▂▂▂▂▂▁▂▁▂▂▁▂▂▂▁▁▂▂▂▂▂▂▂▁▂▂▁▁▂▁▂▁▂▂▂▂▂▁▂▂▁▂▂▂▂▂▂▂▂▂▂▂▂▂▂ ▂
  3.92 ns        Histogram: frequency by time        5.78 ns <

 Memory estimate: 0 bytes, allocs estimate: 0.