I’ll copy the StructReinterpret.jl README:
This package provides an array type similar to ReinterpretArray for reinterpreting bits type element arrays as an array of a homogeneous struct. It works with NTuple wrapper structs as well, such as StaticArrays.
Since the scope of this package is limited to homogeneous structs, we are able to provide a faster implementation of getindex.
The reinterpreted views are constructed using the function struct_reinterpret(::Type{StructType}, A::Array).
Quick start
julia> using StructReinterpret
julia> using BenchmarkTools
julia> using StaticArrays
julia> a = rand(16,16)
16×16 Array{Float64,2}:
0.875014 0.416685 0.621561 0.384859 0.667835 … 0.216662 0.374557 0.917396 0.130807
0.246099 0.0342423 0.954619 0.97971 0.965025 0.102531 0.0383284 0.209974 0.510821
0.216308 0.419426 0.0832236 0.255366 0.946569 0.0964819 0.866634 0.976113 0.881321
0.809495 0.978105 0.794213 0.0227254 0.546804 0.035234 0.0291488 0.892203 0.561338
0.508886 0.715384 0.298849 0.73234 0.868665 0.370543 0.398822 0.853815 0.763415
0.516966 0.13554 0.357667 0.270016 0.294258 … 0.612717 0.40947 0.87473 0.722321
0.0588623 0.49227 0.677219 0.513642 0.399244 0.00378207 0.335834 0.575709 0.560295
0.40336 0.112596 0.65482 0.110796 0.501041 0.824756 0.602597 0.41101 0.148798
0.416314 0.900824 0.495721 0.0565835 0.303076 0.042943 0.40373 0.730327 0.216735
0.227081 0.464685 0.853119 0.739154 0.899333 0.430394 0.444928 0.670904 0.633727
0.369514 0.879112 0.223657 0.138067 0.0621263 … 0.331263 0.7227 0.682614 0.487437
0.460371 0.780633 0.524148 0.923904 0.493046 0.630204 0.452069 0.0684457 0.462189
0.00884589 0.737021 0.0282085 0.777078 0.255077 0.857888 0.211491 0.286567 0.698837
0.132446 0.973928 0.206144 0.766223 0.767426 0.211002 0.0129859 0.120143 0.887541
0.91351 0.814305 0.142757 0.897146 0.819569 0.924263 0.350141 0.617936 0.133594
0.72665 0.998746 0.116852 0.502043 0.205864 … 0.790677 0.0130923 0.745223 0.931052
julia> sr = struct_reinterpret(SVector{2,Float64},a) # reinterpret function provided by this package
8×16 StructReinterpret.StructReinterpretDenseArray{SArray{Tuple{2},Float64,1,2},2,Array{Float64,2}}:
[0.875014, 0.246099] [0.416685, 0.0342423] … [0.917396, 0.209974] [0.130807, 0.510821]
[0.216308, 0.809495] [0.419426, 0.978105] [0.976113, 0.892203] [0.881321, 0.561338]
[0.508886, 0.516966] [0.715384, 0.13554] [0.853815, 0.87473] [0.763415, 0.722321]
[0.0588623, 0.40336] [0.49227, 0.112596] [0.575709, 0.41101] [0.560295, 0.148798]
[0.416314, 0.227081] [0.900824, 0.464685] [0.730327, 0.670904] [0.216735, 0.633727]
[0.369514, 0.460371] [0.879112, 0.780633] … [0.682614, 0.0684457] [0.487437, 0.462189]
[0.00884589, 0.132446] [0.737021, 0.973928] [0.286567, 0.120143] [0.698837, 0.887541]
[0.91351, 0.72665] [0.814305, 0.998746] [0.617936, 0.745223] [0.133594, 0.931052]
julia> r = reinterpret(SVector{2,Float64},a)
8×16 reinterpret(SArray{Tuple{2},Float64,1,2}, ::Array{Float64,2}):
[0.875014, 0.246099] [0.416685, 0.0342423] … [0.917396, 0.209974] [0.130807, 0.510821]
[0.216308, 0.809495] [0.419426, 0.978105] [0.976113, 0.892203] [0.881321, 0.561338]
[0.508886, 0.516966] [0.715384, 0.13554] [0.853815, 0.87473] [0.763415, 0.722321]
[0.0588623, 0.40336] [0.49227, 0.112596] [0.575709, 0.41101] [0.560295, 0.148798]
[0.416314, 0.227081] [0.900824, 0.464685] [0.730327, 0.670904] [0.216735, 0.633727]
[0.369514, 0.460371] [0.879112, 0.780633] … [0.682614, 0.0684457] [0.487437, 0.462189]
[0.00884589, 0.132446] [0.737021, 0.973928] [0.286567, 0.120143] [0.698837, 0.887541]
[0.91351, 0.72665] [0.814305, 0.998746] [0.617936, 0.745223] [0.133594, 0.931052]
julia> nr = Array(r)
8×16 Array{SArray{Tuple{2},Float64,1,2},2}:
[0.875014, 0.246099] [0.416685, 0.0342423] … [0.917396, 0.209974] [0.130807, 0.510821]
[0.216308, 0.809495] [0.419426, 0.978105] [0.976113, 0.892203] [0.881321, 0.561338]
[0.508886, 0.516966] [0.715384, 0.13554] [0.853815, 0.87473] [0.763415, 0.722321]
[0.0588623, 0.40336] [0.49227, 0.112596] [0.575709, 0.41101] [0.560295, 0.148798]
[0.416314, 0.227081] [0.900824, 0.464685] [0.730327, 0.670904] [0.216735, 0.633727]
[0.369514, 0.460371] [0.879112, 0.780633] … [0.682614, 0.0684457] [0.487437, 0.462189]
[0.00884589, 0.132446] [0.737021, 0.973928] [0.286567, 0.120143] [0.698837, 0.887541]
[0.91351, 0.72665] [0.814305, 0.998746] [0.617936, 0.745223] [0.133594, 0.931052]
julia> @benchmark sum($nr) # for base comparision
BenchmarkTools.Trial:
memory estimate: 0 bytes
allocs estimate: 0
--------------
minimum time: 95.649 ns (0.00% GC)
median time: 95.782 ns (0.00% GC)
mean time: 96.023 ns (0.00% GC)
maximum time: 155.224 ns (0.00% GC)
--------------
samples: 10000
evals/sample: 950
julia> @benchmark sum($r) # julia's built-in reinterpret array
BenchmarkTools.Trial:
memory estimate: 0 bytes
allocs estimate: 0
--------------
minimum time: 120.977 ns (0.00% GC)
median time: 121.432 ns (0.00% GC)
mean time: 122.028 ns (0.00% GC)
maximum time: 550.409 ns (0.00% GC)
--------------
samples: 10000
evals/sample: 907
julia> @benchmark sum($sr) # this packages implementation of reinterpret array
BenchmarkTools.Trial:
memory estimate: 0 bytes
allocs estimate: 0
--------------
minimum time: 96.732 ns (0.00% GC)
median time: 97.040 ns (0.00% GC)
mean time: 97.705 ns (0.00% GC)
maximum time: 246.368 ns (0.00% GC)
--------------
samples: 10000
evals/sample: 949
I’m not registering this package right now because I made a PR to base to add this special-cased implementation of getindex to the built-in ReinterpretArray.
While that is not merged you can use it right now by cloning the repository.
If for some reason that PR is not merged, then I’ll properly register this package.