Efficient coordinate collection from LineStrings using reinterpret

[asinghvi17]

import GeoFormatTypes as GFT, WellKnownGeometry as WKG, GeoInterface as GI

tups = tuple.(rand(300_000), rand(300_000))
geoms_as_points = GFT.val.(WKG.getwkb.(GI.Point.(tups)))
geoms_as_linestring = GFT.val(WKG.getwkb(GI.LineString(tups)))

# parse geoms as points, each WKB individually
@benchmark begin
    map($(geoms_as_points)) do geom
        only(reinterpret(Tuple{Float64, Float64}, view(geom, (5+1):length(geom)))) # this has to be parsed differently in EWKB and GeoPkg wkb
    end
end # median time 3ms v/s regular parsing 300ms

# parse the whole linestring
@benchmark collect(reinterpret(Tuple{Float64, Float64}, view($ls_as_wkb, (1+4+4+1):length($ls_as_wkb))))
# median time 290ns vs unknown (still running) 
# we could even drop collect here

[evetion]

Fun thing, doing the above works fine, but ofcourse changes the return type to an NTuple. That sorta implies we got to change the return types of the other (point) methods too, which actually could slow down things:

function GI.getcoord(::GI.PointTrait, geom::WKBtype)
    offset = 1
    ncoord = GI.ncoord(geom)
    data = @view geom.val[headersize+offset:headersize+offset+sizeof(Float64)*ncoord-1]

    reinterpret(Float64, data)  # current 211.961 ns (2 allocations: 112 bytes)
    only(reinterpret(NTuple{ncoord,Float64}, data))  #   580.569 ns (7 allocations: 320 bytes)
    GC.@preserve geom unsafe_load(Ptr{NTuple{ncoord,Float64}}(pointer(geom.val, headersize + offset)))  # 466.837 ns (6 allocations: 224 bytes)
end

[rafaqz]

Have you profiled this? That just looks like type instability to me

[evetion]

Those three different variants were run using btime yes.

[rafaqz]

No I mean ProfileView... Yours will have some red

[evetion]

With the linked PR

import GeoFormatTypes as GFT, WellKnownGeometry as WKG, GeoInterface as GI

tups = tuple.(rand(300_000), rand(300_000))
geoms_as_linestring = WKG.getwkb(GI.LineString(tups))
@btime GI.coordinates(geoms_as_linestring);
98.625 μs (5 allocations: 4.58 MiB)