Constant static matrices cause compilation issues

this example should not allocate:

using LinearAlgebra, DifferentialEquations, DiffEqGPU
A  = Float32[1. 0  0 -5; 4 -2  4 -3; -4  0  0  1; 5 -2  2  3]
u0 = Array{Float32}(rand(4,2))
tspan = (0.0f0,1.0f0)
function fip(du,u,p,t) 
   mul!(du, A, u)
end
probip = ODEProblem(fip,u0,tspan)
function prob_func(prob, i, repeat)
    prob
end
ensprobip=EnsembleProblem(probip, prob_func=prob_func)
trajs = 123
simgpu = solve(ensprobip, Tsit5(), EnsembleGPUArray(), trajectories=trajs)

i ran it on a linux machine with CUDA 9.0 and several Tesla K80 GPUs with compute capability 3.7.

see also this thread: https://discourse.julialang.org/t/test-diffeqgpu-errors-with-unsupported-call-to-the-julia-runtime/28893/17

with EnsembleCPUArray it works fine, but it fails to compile on GPU with the stack trace:

InvalidIRError: compiling gpu_kernel(Cassette.Context{nametype(Ctx),Nothing,Nothing,getfield(GPUifyLoops, Symbol("##PassType#371")),Nothing,Cassette.DisableHooks}, typeof(DiffEqGPU.gpu_kernel), ODEFunction{true,typeof(fip),UniformScaling{Bool},Nothing,Nothing,Nothing,Nothing,Nothing,Nothing,Nothing,Nothing,Nothing}, CUDAnative.CuDeviceArray{Float32,2,CUDAnative.AS.Global}, CUDAnative.CuDeviceArray{Float32,2,CUDAnative.AS.Global}, CUDAnative.CuDeviceArray{DiffEqBase.NullParameters,2,CUDAnative.AS.Global}, Float32) resulted in invalid LLVM IR
Reason: unsupported dynamic function invocation (call to overdub)
Stacktrace:
 [1] fip at In[1]:10
 [2] ODEFunction at /home/bq_nbecker/isi/julia_pkg/GPUTest/packages/DiffEqBase/8uyX3/src/diffeqfunction.jl:230
 [3] gpu_kernel at /home/bq_nbecker/isi/julia_pkg/GPUTest/packages/DiffEqGPU/QB1WC/src/DiffEqGPU.jl:6
 [4] overdub at /home/bq_nbecker/isi/julia_pkg/GPUTest/packages/Cassette/YCOeN/src/overdub.jl:0

Stacktrace:
 [1] check_ir(::CUDAnative.CompilerJob, ::LLVM.Module) at /home/bq_nbecker/isi/julia_pkg/GPUTest/packages/CUDAnative/UWBIY/src/compiler/validation.jl:114
 [2] macro expansion at /home/bq_nbecker/isi/julia_pkg/GPUTest/packages/CUDAnative/UWBIY/src/compiler/driver.jl:188 [inlined]
 [3] macro expansion at /home/bq_nbecker/isi/julia_pkg/GPUTest/packages/TimerOutputs/7zSea/src/TimerOutput.jl:216 [inlined]
 [4] #codegen#130(::Bool, ::Bool, ::Bool, ::Bool, ::Bool, ::typeof(CUDAnative.codegen), ::Symbol, ::CUDAnative.CompilerJob) at /home/bq_nbecker/isi/julia_pkg/GPUTest/packages/CUDAnative/UWBIY/src/compiler/driver.jl:186
 [5] #codegen at /home/bq_nbecker/isi/julia_pkg/GPUTest/packages/CUDAnative/UWBIY/src/compiler/driver.jl:0 [inlined]
 [6] #compile#129(::Bool, ::Bool, ::Bool, ::Bool, ::Bool, ::typeof(CUDAnative.compile), ::Symbol, ::CUDAnative.CompilerJob) at /home/bq_nbecker/isi/julia_pkg/GPUTest/packages/CUDAnative/UWBIY/src/compiler/driver.jl:47
 [7] #compile at /home/bq_nbecker/isi/julia_pkg/GPUTest/packages/CUDAnative/UWBIY/src/compiler/common.jl:0 [inlined]
 [8] #compile#128 at /home/bq_nbecker/isi/julia_pkg/GPUTest/packages/CUDAnative/UWBIY/src/compiler/driver.jl:28 [inlined]
 [9] #compile at ./none:0 [inlined] (repeats 2 times)
 [10] macro expansion at /home/bq_nbecker/isi/julia_pkg/GPUTest/packages/CUDAnative/UWBIY/src/execution.jl:389 [inlined]
 [11] #cufunction#170(::String, ::Base.Iterators.Pairs{Union{},Union{},Tuple{},NamedTuple{(),Tuple{}}}, ::typeof(CUDAnative.cufunction), ::typeof(Cassette.overdub), ::Type{Tuple{Cassette.Context{nametype(Ctx),Nothing,Nothing,getfield(GPUifyLoops, Symbol("##PassType#371")),Nothing,Cassette.DisableHooks},typeof(DiffEqGPU.gpu_kernel),ODEFunction{true,typeof(fip),UniformScaling{Bool},Nothing,Nothing,Nothing,Nothing,Nothing,Nothing,Nothing,Nothing,Nothing},CUDAnative.CuDeviceArray{Float32,2,CUDAnative.AS.Global},CUDAnative.CuDeviceArray{Float32,2,CUDAnative.AS.Global},CUDAnative.CuDeviceArray{DiffEqBase.NullParameters,2,CUDAnative.AS.Global},Float32}}) at /home/bq_nbecker/isi/julia_pkg/GPUTest/packages/CUDAnative/UWBIY/src/execution.jl:357
 [12] (::getfield(CUDAnative, Symbol("#kw##cufunction")))(::NamedTuple{(:name,),Tuple{String}}, ::typeof(CUDAnative.cufunction), ::Function, ::Type) at ./none:0
 [13] #launch#50(::Base.Iterators.Pairs{Union{},Union{},Tuple{},NamedTuple{(),Tuple{}}}, ::typeof(GPUifyLoops.launch), ::GPUifyLoops.CUDA, ::typeof(DiffEqGPU.gpu_kernel), ::Function, ::Vararg{Any,N} where N) at /home/bq_nbecker/isi/julia_pkg/GPUTest/packages/GPUifyLoops/mjszO/src/GPUifyLoops.jl:125
 [14] launch at /home/bq_nbecker/isi/julia_pkg/GPUTest/packages/GPUifyLoops/mjszO/src/GPUifyLoops.jl:119 [inlined]
 [15] macro expansion at /home/bq_nbecker/isi/julia_pkg/GPUTest/packages/GPUifyLoops/mjszO/src/GPUifyLoops.jl:54 [inlined]
 [16] #12 at /home/bq_nbecker/isi/julia_pkg/GPUTest/packages/DiffEqGPU/QB1WC/src/DiffEqGPU.jl:61 [inlined]
 [17] ODEFunction at /home/bq_nbecker/isi/julia_pkg/GPUTest/packages/DiffEqBase/8uyX3/src/diffeqfunction.jl:230 [inlined]
 [18] initialize!(::OrdinaryDiffEq.ODEIntegrator{Tsit5,true,CuArrays.CuArray{Float32,2},Float32,CuArrays.CuArray{DiffEqBase.NullParameters,2},Float32,Float32,Float32,Array{CuArrays.CuArray{Float32,2},1},ODESolution{Float32,3,Array{CuArrays.CuArray{Float32,2},1},Nothing,Nothing,Array{Float32,1},Array{Array{CuArrays.CuArray{Float32,2},1},1},ODEProblem{CuArrays.CuArray{Float32,2},Tuple{Float32,Float32},true,CuArrays.CuArray{DiffEqBase.NullParameters,2},ODEFunction{true,getfield(DiffEqGPU, Symbol("##12#22")){ODEFunction{true,typeof(fip),UniformScaling{Bool},Nothing,Nothing,Nothing,Nothing,Nothing,Nothing,Nothing,Nothing,Nothing}},UniformScaling{Bool},Nothing,Nothing,Nothing,Nothing,Nothing,Nothing,Nothing,Nothing,Nothing},Base.Iterators.Pairs{Union{},Union{},Tuple{},NamedTuple{(),Tuple{}}},DiffEqBase.StandardODEProblem},Tsit5,OrdinaryDiffEq.InterpolationData{ODEFunction{true,getfield(DiffEqGPU, Symbol("##12#22")){ODEFunction{true,typeof(fip),UniformScaling{Bool},Nothing,Nothing,Nothing,Nothing,Nothing,Nothing,Nothing,Nothing,Nothing}},UniformScaling{Bool},Nothing,Nothing,Nothing,Nothing,Nothing,Nothing,Nothing,Nothing,Nothing},Array{CuArrays.CuArray{Float32,2},1},Array{Float32,1},Array{Array{CuArrays.CuArray{Float32,2},1},1},OrdinaryDiffEq.Tsit5Cache{CuArrays.CuArray{Float32,2},CuArrays.CuArray{Float32,2},CuArrays.CuArray{Float32,2},OrdinaryDiffEq.Tsit5ConstantCache{Float32,Float32}}},DiffEqBase.DEStats},ODEFunction{true,getfield(DiffEqGPU, Symbol("##12#22")){ODEFunction{true,typeof(fip),UniformScaling{Bool},Nothing,Nothing,Nothing,Nothing,Nothing,Nothing,Nothing,Nothing,Nothing}},UniformScaling{Bool},Nothing,Nothing,Nothing,Nothing,Nothing,Nothing,Nothing,Nothing,Nothing},OrdinaryDiffEq.Tsit5Cache{CuArrays.CuArray{Float32,2},CuArrays.CuArray{Float32,2},CuArrays.CuArray{Float32,2},OrdinaryDiffEq.Tsit5ConstantCache{Float32,Float32}},OrdinaryDiffEq.DEOptions{Float32,Float32,Float32,Float32,typeof(DiffEqBase.ODE_DEFAULT_NORM),typeof(opnorm),CallbackSet{Tuple{},Tuple{}},typeof(DiffEqBase.ODE_DEFAULT_ISOUTOFDOMAIN),typeof(DiffEqBase.ODE_DEFAULT_PROG_MESSAGE),typeof(DiffEqBase.ODE_DEFAULT_UNSTABLE_CHECK),DataStructures.BinaryHeap{Float32,DataStructures.LessThan},DataStructures.BinaryHeap{Float32,DataStructures.LessThan},Nothing,Nothing,Int64,Array{Float32,1},Array{Float32,1},Array{Float32,1}},CuArrays.CuArray{Float32,2},Float32,Nothing}, ::OrdinaryDiffEq.Tsit5Cache{CuArrays.CuArray{Float32,2},CuArrays.CuArray{Float32,2},CuArrays.CuArray{Float32,2},OrdinaryDiffEq.Tsit5ConstantCache{Float32,Float32}}) at /home/bq_nbecker/isi/julia_pkg/GPUTest/packages/OrdinaryDiffEq/tQd6p/src/perform_step/low_order_rk_perform_step.jl:623
 [19] #__init#335(::Array{Float32,1}, ::Array{Float32,1}, ::Array{Float32,1}, ::Nothing, ::Bool, ::Bool, ::Bool, ::Bool, ::Nothing, ::Bool, ::Bool, ::Float32, ::Float32, ::Float32, ::Bool, ::Bool, ::Rational{Int64}, ::Nothing, ::Nothing, ::Rational{Int64}, ::Int64, ::Int64, ::Int64, ::Rational{Int64}, ::Bool, ::Int64, ::Nothing, ::Nothing, ::Int64, ::typeof(DiffEqBase.ODE_DEFAULT_NORM), ::typeof(opnorm), ::typeof(DiffEqBase.ODE_DEFAULT_ISOUTOFDOMAIN), ::typeof(DiffEqBase.ODE_DEFAULT_UNSTABLE_CHECK), ::Bool, ::Bool, ::Bool, ::Bool, ::Bool, ::Bool, ::Bool, ::Int64, ::String, ::typeof(DiffEqBase.ODE_DEFAULT_PROG_MESSAGE), ::Nothing, ::Bool, ::Bool, ::Bool, ::Base.Iterators.Pairs{Union{},Union{},Tuple{},NamedTuple{(),Tuple{}}}, ::typeof(DiffEqBase.__init), ::ODEProblem{CuArrays.CuArray{Float32,2},Tuple{Float32,Float32},true,CuArrays.CuArray{DiffEqBase.NullParameters,2},ODEFunction{true,getfield(DiffEqGPU, Symbol("##12#22")){ODEFunction{true,typeof(fip),UniformScaling{Bool},Nothing,Nothing,Nothing,Nothing,Nothing,Nothing,Nothing,Nothing,Nothing}},UniformScaling{Bool},Nothing,Nothing,Nothing,Nothing,Nothing,Nothing,Nothing,Nothing,Nothing},Base.Iterators.Pairs{Union{},Union{},Tuple{},NamedTuple{(),Tuple{}}},DiffEqBase.StandardODEProblem}, ::Tsit5, ::Array{CuArrays.CuArray{Float32,2},1}, ::Array{Float32,1}, ::Array{Any,1}, ::Type{Val{true}}) at /home/bq_nbecker/isi/julia_pkg/GPUTest/packages/OrdinaryDiffEq/tQd6p/src/solve.jl:352
 [20] __init(::ODEProblem{CuArrays.CuArray{Float32,2},Tuple{Float32,Float32},true,CuArrays.CuArray{DiffEqBase.NullParameters,2},ODEFunction{true,getfield(DiffEqGPU, Symbol("##12#22")){ODEFunction{true,typeof(fip),UniformScaling{Bool},Nothing,Nothing,Nothing,Nothing,Nothing,Nothing,Nothing,Nothing,Nothing}},UniformScaling{Bool},Nothing,Nothing,Nothing,Nothing,Nothing,Nothing,Nothing,Nothing,Nothing},Base.Iterators.Pairs{Union{},Union{},Tuple{},NamedTuple{(),Tuple{}}},DiffEqBase.StandardODEProblem}, ::Tsit5, ::Array{CuArrays.CuArray{Float32,2},1}, ::Array{Float32,1}, ::Array{Any,1}, ::Type{Val{true}}) at /home/bq_nbecker/isi/julia_pkg/GPUTest/packages/OrdinaryDiffEq/tQd6p/src/solve.jl:66 (repeats 4 times)
 [21] #__solve#334 at /home/bq_nbecker/isi/julia_pkg/GPUTest/packages/OrdinaryDiffEq/tQd6p/src/solve.jl:4 [inlined]
 [22] __solve at /home/bq_nbecker/isi/julia_pkg/GPUTest/packages/OrdinaryDiffEq/tQd6p/src/solve.jl:4 [inlined]
 [23] #solve_call#425(::Base.Iterators.Pairs{Union{},Union{},Tuple{},NamedTuple{(),Tuple{}}}, ::typeof(DiffEqBase.solve_call), ::ODEProblem{CuArrays.CuArray{Float32,2},Tuple{Float32,Float32},true,CuArrays.CuArray{DiffEqBase.NullParameters,2},ODEFunction{true,getfield(DiffEqGPU, Symbol("##12#22")){ODEFunction{true,typeof(fip),UniformScaling{Bool},Nothing,Nothing,Nothing,Nothing,Nothing,Nothing,Nothing,Nothing,Nothing}},UniformScaling{Bool},Nothing,Nothing,Nothing,Nothing,Nothing,Nothing,Nothing,Nothing,Nothing},Base.Iterators.Pairs{Union{},Union{},Tuple{},NamedTuple{(),Tuple{}}},DiffEqBase.StandardODEProblem}, ::Tsit5) at /home/bq_nbecker/isi/julia_pkg/GPUTest/packages/DiffEqBase/8uyX3/src/solve.jl:40
 [24] solve_call at /home/bq_nbecker/isi/julia_pkg/GPUTest/packages/DiffEqBase/8uyX3/src/solve.jl:37 [inlined]
 [25] #solve#426 at /home/bq_nbecker/isi/julia_pkg/GPUTest/packages/DiffEqBase/8uyX3/src/solve.jl:57 [inlined]
 [26] solve(::ODEProblem{CuArrays.CuArray{Float32,2},Tuple{Float32,Float32},true,CuArrays.CuArray{DiffEqBase.NullParameters,2},ODEFunction{true,getfield(DiffEqGPU, Symbol("##12#22")){ODEFunction{true,typeof(fip),UniformScaling{Bool},Nothing,Nothing,Nothing,Nothing,Nothing,Nothing,Nothing,Nothing,Nothing}},UniformScaling{Bool},Nothing,Nothing,Nothing,Nothing,Nothing,Nothing,Nothing,Nothing,Nothing},Base.Iterators.Pairs{Union{},Union{},Tuple{},NamedTuple{(),Tuple{}}},DiffEqBase.StandardODEProblem}, ::Tsit5) at /home/bq_nbecker/isi/julia_pkg/GPUTest/packages/DiffEqBase/8uyX3/src/solve.jl:45
 [27] #batch_solve#5(::Base.Iterators.Pairs{Union{},Union{},Tuple{},NamedTuple{(),Tuple{}}}, ::typeof(DiffEqGPU.batch_solve), ::EnsembleProblem{ODEProblem{Array{Float32,2},Tuple{Float32,Float32},true,DiffEqBase.NullParameters,ODEFunction{true,typeof(fip),UniformScaling{Bool},Nothing,Nothing,Nothing,Nothing,Nothing,Nothing,Nothing,Nothing,Nothing},Base.Iterators.Pairs{Union{},Union{},Tuple{},NamedTuple{(),Tuple{}}},DiffEqBase.StandardODEProblem},typeof(prob_func),getfield(DiffEqBase, Symbol("##332#338")),getfield(DiffEqBase, Symbol("##334#340")),Array{Any,1}}, ::Tsit5, ::EnsembleGPUArray, ::UnitRange{Int64}) at /home/bq_nbecker/isi/julia_pkg/GPUTest/packages/DiffEqGPU/QB1WC/src/DiffEqGPU.jl:66
 [28] batch_solve at /home/bq_nbecker/isi/julia_pkg/GPUTest/packages/DiffEqGPU/QB1WC/src/DiffEqGPU.jl:45 [inlined]
 [29] (::getfield(DiffEqGPU, Symbol("##3#4")){Int64,Int64,Base.Iterators.Pairs{Union{},Union{},Tuple{},NamedTuple{(),Tuple{}}},EnsembleProblem{ODEProblem{Array{Float32,2},Tuple{Float32,Float32},true,DiffEqBase.NullParameters,ODEFunction{true,typeof(fip),UniformScaling{Bool},Nothing,Nothing,Nothing,Nothing,Nothing,Nothing,Nothing,Nothing,Nothing},Base.Iterators.Pairs{Union{},Union{},Tuple{},NamedTuple{(),Tuple{}}},DiffEqBase.StandardODEProblem},typeof(prob_func),getfield(DiffEqBase, Symbol("##332#338")),getfield(DiffEqBase, Symbol("##334#340")),Array{Any,1}},Tsit5,EnsembleGPUArray})(::Int64) at /home/bq_nbecker/isi/julia_pkg/GPUTest/packages/DiffEqGPU/QB1WC/src/DiffEqGPU.jl:37
 [30] iterate at ./generator.jl:47 [inlined]
 [31] _collect(::UnitRange{Int64}, ::Base.Generator{UnitRange{Int64},getfield(DiffEqGPU, Symbol("##3#4")){Int64,Int64,Base.Iterators.Pairs{Union{},Union{},Tuple{},NamedTuple{(),Tuple{}}},EnsembleProblem{ODEProblem{Array{Float32,2},Tuple{Float32,Float32},true,DiffEqBase.NullParameters,ODEFunction{true,typeof(fip),UniformScaling{Bool},Nothing,Nothing,Nothing,Nothing,Nothing,Nothing,Nothing,Nothing,Nothing},Base.Iterators.Pairs{Union{},Union{},Tuple{},NamedTuple{(),Tuple{}}},DiffEqBase.StandardODEProblem},typeof(prob_func),getfield(DiffEqBase, Symbol("##332#338")),getfield(DiffEqBase, Symbol("##334#340")),Array{Any,1}},Tsit5,EnsembleGPUArray}}, ::Base.EltypeUnknown, ::Base.HasShape{1}) at ./array.jl:619
 [32] collect_similar(::UnitRange{Int64}, ::Base.Generator{UnitRange{Int64},getfield(DiffEqGPU, Symbol("##3#4")){Int64,Int64,Base.Iterators.Pairs{Union{},Union{},Tuple{},NamedTuple{(),Tuple{}}},EnsembleProblem{ODEProblem{Array{Float32,2},Tuple{Float32,Float32},true,DiffEqBase.NullParameters,ODEFunction{true,typeof(fip),UniformScaling{Bool},Nothing,Nothing,Nothing,Nothing,Nothing,Nothing,Nothing,Nothing,Nothing},Base.Iterators.Pairs{Union{},Union{},Tuple{},NamedTuple{(),Tuple{}}},DiffEqBase.StandardODEProblem},typeof(prob_func),getfield(DiffEqBase, Symbol("##332#338")),getfield(DiffEqBase, Symbol("##334#340")),Array{Any,1}},Tsit5,EnsembleGPUArray}}) at ./array.jl:548
 [33] map(::Function, ::UnitRange{Int64}) at ./abstractarray.jl:2073
 [34] macro expansion at /home/bq_nbecker/isi/julia_pkg/GPUTest/packages/DiffEqGPU/QB1WC/src/DiffEqGPU.jl:31 [inlined]
 [35] macro expansion at ./util.jl:213 [inlined]
 [36] #__solve#2(::Int64, ::Int64, ::Base.Iterators.Pairs{Union{},Union{},Tuple{},NamedTuple{(),Tuple{}}}, ::typeof(DiffEqBase.__solve), ::EnsembleProblem{ODEProblem{Array{Float32,2},Tuple{Float32,Float32},true,DiffEqBase.NullParameters,ODEFunction{true,typeof(fip),UniformScaling{Bool},Nothing,Nothing,Nothing,Nothing,Nothing,Nothing,Nothing,Nothing,Nothing},Base.Iterators.Pairs{Union{},Union{},Tuple{},NamedTuple{(),Tuple{}}},DiffEqBase.StandardODEProblem},typeof(prob_func),getfield(DiffEqBase, Symbol("##332#338")),getfield(DiffEqBase, Symbol("##334#340")),Array{Any,1}}, ::Tsit5, ::EnsembleGPUArray) at /home/bq_nbecker/isi/julia_pkg/GPUTest/packages/DiffEqGPU/QB1WC/src/DiffEqGPU.jl:30
 [37] #__solve at ./none:0 [inlined]
 [38] #solve#427 at /home/bq_nbecker/isi/julia_pkg/GPUTest/packages/DiffEqBase/8uyX3/src/solve.jl:64 [inlined]
 [39] (::getfield(DiffEqBase, Symbol("#kw##solve")))(::NamedTuple{(:trajectories,),Tuple{Int64}}, ::typeof(solve), ::EnsembleProblem{ODEProblem{Array{Float32,2},Tuple{Float32,Float32},true,DiffEqBase.NullParameters,ODEFunction{true,typeof(fip),UniformScaling{Bool},Nothing,Nothing,Nothing,Nothing,Nothing,Nothing,Nothing,Nothing,Nothing},Base.Iterators.Pairs{Union{},Union{},Tuple{},NamedTuple{(),Tuple{}}},DiffEqBase.StandardODEProblem},typeof(prob_func),getfield(DiffEqBase, Symbol("##332#338")),getfield(DiffEqBase, Symbol("##334#340")),Array{Any,1}}, ::Tsit5, ::EnsembleGPUArray) at ./none:0
 [40] top-level scope at In[3]:15

Start by reproducing the supplied EnsembleGPUArray example and compare it with EnsembleCPUArray. Read the GPU entry points in DiffEqGPU.jl at the stack-trace locations around lines 6 and 61, along with the CUDAnative validation error. Done means the static-matrix example compiles and runs on the GPU without the reported invalid LLVM IR error.