# GPU Kernel Compilation Error -- Error in sharing functions in memory?

**URL:** <https://discourse.julialang.org/t/gpu-kernel-compilation-error-error-in-sharing-functions-in-memory/88046>\
**Category:** New to Julia\
**Tags:** gpu\
**Created:** [September 30, 2022, 4:03pm UTC](https://discourse.julialang.org/t/gpu-kernel-compilation-error-error-in-sharing-functions-in-memory/88046 "2022-09-30T16:03:40Z")\
**Posts on this page:** 2\
**Page:** 1

<div class="post-metadata">

**Author:** ![yoshi](https://sea2.discourse-cdn.com/julialang/user_avatar/discourse.julialang.org/yoshi/32/38259_2.png) [@yoshi](https://discourse.julialang.org/u/yoshi)\
**Post date:** [September 30, 2022, 4:03pm UTC](https://discourse.julialang.org/t/gpu-kernel-compilation-error-error-in-sharing-functions-in-memory/88046/1 "2022-09-30T16:03:40Z")

</div>

I’m struggling to troubleshoot my code – I’m receiving a GPU kernel compilation error. Is there some problem with how my functions are defined? (Note how the errors seem to appear on my “current equations” – seen in the code below as Il, IK, and ICa).

```julia
ERROR: InvalidIRError: compiling kernel #gpu_gpu_kernel(KernelAbstractions.CompilerMetadata{KernelAbstractions.NDIte
ration.DynamicSize, KernelAbstractions.NDIteration.DynamicCheck, Nothing, CartesianIndices{1, Tuple{Base.OneTo{Int6$
}}}, KernelAbstractions.NDIteration.NDRange{1, KernelAbstractions.NDIteration.DynamicSize, KernelAbstractions.NDIte$
ation.DynamicSize, CartesianIndices{1, Tuple{Base.OneTo{Int64}}}, CartesianIndices{1, Tuple{Base.OneTo{Int64}}}}}, $
ypeof(f), CuDeviceMatrix{Float64, 1}, CuDeviceMatrix{Float64, 1}, CuDeviceMatrix{Float64, 1}, Float32) resulted in $
nvalid LLVM IR                                                                                                      
Reason: unsupported dynamic function invocation (call to -)                                                         
Stacktrace:                                                                                                         
 [1] Il                                                                                                             
   @ ./REPL[80]:1                                                                                                   
 [2] f                                                                                                              
   @ ./REPL[81]:5                                                                                                   
 [3] macro expansion                                                                                                
   @ ~/.julia/packages/DiffEqGPU/CiiCq/src/DiffEqGPU.jl:24
 [4] gpu_gpu_kernel
   @ ~/.julia/packages/KernelAbstractions/DqITC/src/macros.jl:81
 [5] gpu_gpu_kernel
   @ ./none:0
Reason: unsupported dynamic function invocation (call to *) [364/1886]
Stacktrace:
 [1] Il
   @ ./REPL[80]:1
 [2] f
   @ ./REPL[81]:5
 [3] macro expansion
   @ ~/.julia/packages/DiffEqGPU/CiiCq/src/DiffEqGPU.jl:24
 [4] gpu_gpu_kernel
   @ ~/.julia/packages/KernelAbstractions/DqITC/src/macros.jl:81
 [5] gpu_gpu_kernel
   @ ./none:0
Reason: unsupported dynamic function invocation (call to -)
Stacktrace:
 [1] Ik
   @ ./REPL[79]:1
 [2] f
   @ ./REPL[81]:5
 [3] macro expansion
   @ ~/.julia/packages/DiffEqGPU/CiiCq/src/DiffEqGPU.jl:24
 [4] gpu_gpu_kernel
   @ ~/.julia/packages/KernelAbstractions/DqITC/src/macros.jl:81
 [5] gpu_gpu_kernel
   @ ./none:0
Reason: unsupported dynamic function invocation (call to *) [340/1886]
Stacktrace:
 [1] *
   @ ./operators.jl:591
 [2] Ik
   @ ./REPL[79]:1
 [3] f
   @ ./REPL[81]:5
 [4] macro expansion
   @ ~/.julia/packages/DiffEqGPU/CiiCq/src/DiffEqGPU.jl:24
 [5] gpu_gpu_kernel
   @ ~/.julia/packages/KernelAbstractions/DqITC/src/macros.jl:81
 [6] gpu_gpu_kernel
   @ ./none:0
Reason: unsupported dynamic function invocation (call to *)
Stacktrace:
 [1] f
   @ ./REPL[81]:5
 [2] macro expansion
   @ ~/.julia/packages/DiffEqGPU/CiiCq/src/DiffEqGPU.jl:24
 [3] gpu_gpu_kernel
   @ ~/.julia/packages/KernelAbstractions/DqITC/src/macros.jl:81
 [4] gpu_gpu_kernel
   @ ./none:0
Reason: unsupported dynamic function invocation (call to +) [316/1886]
Stacktrace:
 [1] f
   @ ./REPL[81]:5
 [2] macro expansion
   @ ~/.julia/packages/DiffEqGPU/CiiCq/src/DiffEqGPU.jl:24
 [3] gpu_gpu_kernel
   @ ~/.julia/packages/KernelAbstractions/DqITC/src/macros.jl:81
 [4] gpu_gpu_kernel
   @ ./none:0
Reason: unsupported dynamic function invocation (call to /)
Stacktrace:
 [1] f
   @ ./REPL[81]:5
 [2] macro expansion
   @ ~/.julia/packages/DiffEqGPU/CiiCq/src/DiffEqGPU.jl:24
 [3] gpu_gpu_kernel
   @ ~/.julia/packages/KernelAbstractions/DqITC/src/macros.jl:81
 [4] gpu_gpu_kernel
   @ ./none:0
Reason: unsupported dynamic function invocation (call to convert)
Stacktrace:
 [1] setindex!
   @ ~/.julia/packages/CUDA/DfvRa/src/device/array.jl:194
 [2] setindex!
   @ ./subarray.jl:347
 [3] f
[4] macro expansion [288/1886]
   @ ~/.julia/packages/DiffEqGPU/CiiCq/src/DiffEqGPU.jl:24
 [5] gpu_gpu_kernel
   @ ~/.julia/packages/KernelAbstractions/DqITC/src/macros.jl:81
 [6] gpu_gpu_kernel
   @ ./none:0
Reason: unsupported dynamic function invocation (call to *)
Stacktrace:
 [1] f
   @ ./REPL[81]:9
 [2] macro expansion
   @ ~/.julia/packages/DiffEqGPU/CiiCq/src/DiffEqGPU.jl:24
 [3] gpu_gpu_kernel
   @ ~/.julia/packages/KernelAbstractions/DqITC/src/macros.jl:81
 [4] gpu_gpu_kernel
   @ ./none:0
Reason: unsupported dynamic function invocation (call to -)
Stacktrace:
 [1] f
   @ ./REPL[81]:9
 [2] macro expansion
   @ ~/.julia/packages/DiffEqGPU/CiiCq/src/DiffEqGPU.jl:24
 [3] gpu_gpu_kernel
   @ ~/.julia/packages/KernelAbstractions/DqITC/src/macros.jl:81
 [4] gpu_gpu_kernel
   @ ./none:0
Reason: unsupported dynamic function invocation (call to convert)
Stacktrace:
 [1] setindex!
   @ ~/.julia/packages/CUDA/DfvRa/src/device/array.jl:194
 [2] setindex!
   @ ./subarray.jl:347
 [3] f
   @ ./REPL[81]:9
 [4] macro expansion
   @ ~/.julia/packages/DiffEqGPU/CiiCq/src/DiffEqGPU.jl:24
 [5] gpu_gpu_kernel
   @ ~/.julia/packages/KernelAbstractions/DqITC/src/macros.jl:81
 [6] gpu_gpu_kernel
   @ ./none:0
Reason: unsupported dynamic function invocation (call to -)
Stacktrace:
 [1] Ica
   @ ./REPL[78]:2
 [2] multiple call sites
   @ unknown:0
Reason: unsupported dynamic function invocation (call to *)
Stacktrace:
 [1] *
   @ ./operators.jl:591
 [2] Ica
   @ ./REPL[78]:2
 [3] multiple call sites
   @ unknown:0
Hint: catch this exception as `err` and call `code_typed(err; interactive = true)` to introspect the erronous code with Cthulhu.jl

```

The code is given below.

```julia
using DiffEqGPU, OrdinaryDiffEq, GPUInspector

#################################################
##### ODE params

# initialize parameter vector
p = fill(NaN,8);

# Time Constants for Homeostasis -- (g_Ca, g_K, nHalf)
p[1] = 500; p[2] = 500; p[3] = 500;

# C_T & Delta
p[4] = 20; p[5] = 5; 

# Sigmoid Ranges -- (g_Ca, g_K, nHalf)
p[6] = 3; p[7] = 6; p[8] = 20;

# Calcium Handling
A = -1; k = -1/.1;

I_app = 0;
C = 1; g_L = .5;
E_L = -50; E_Ca = 100; E_K = -70;

#################################################
## Initial Condition Testing -- Tests for Multistability

IC_V = -75.0:15.0:75.0;
IC_n = 0:.2:1;

IC_gCa = 0:.3:1.5;
IC_gK = 0:.6:3;

IC_Ca = 0.0:50.0:300.0;
IC_nHalf = 0:p[8]/5:p[8];

AllICs = collect.(Iterators.product(IC_V,IC_n,IC_gCa,IC_gK,IC_Ca,IC_nHalf));

#################################################

# Define functions
sigmoid(x) = 1/(1+exp(-x));
tau(V, halfN) = 3/cosh(((V-halfN)-10)/29);
Ninf(V, halfN) = sigmoid(((V-halfN)-10)/7.25);

# Define Currents
Ica(g_Ca,V) = g_Ca*(sigmoid((V+1)/7.5)+.1)*(V-E_Ca);
Ik(g_K,n,V) = g_K*n*(V-E_K);
Il(V) = g_L*(V-E_L);

# u[1] - V cell 1, u[2] - slow var cell, u[3] - g_Ca, u[4] - g_K, u[5] - [Ca], u[6] - halfN
# p[1] - tau_L.gCa , p[2] - tau_L.gK, p[3] - tau_L.nHalf, p[4] - C_T, p[5] - Delta, p[6] - g_Ca, p[7] - g_K, p[8] - nHalf
function f(du,u,p,t)
    du[1] = (-1*Il(u[1]) + -1*Ica(u[3],u[1]) + -1*Ik(u[4],u[2],u[1]) + I_app)/C
    du[2] = (Ninf(u[1],u[6])-u[2])/tau(u[1],u[6])
    du[3] = (p[6]*sigmoid((p[4]-u[5])/p[5]) - u[3])/p[1]
    du[4] = (p[7]*sigmoid((u[5]-p[4])/p[5]) - u[4])/p[2]
    du[5] = -1*k*(A*Ica(u[3],u[1]) - u[5])
    du[6] = (p[8]*(sigmoid((p[4]-u[5])/p[5])) - u[6])/p[3]
end

##############

#Trajs = length(AllICs);
Trajs = 997;
u0 = Float32[0.0;0.0;0.0;0.0;0.0;0.0]
tspan = (0.0f0,10000.0f0)
SaveTimes = [7000:.1:7150;7150.1];
prob = ODEProblem(f,u0,tspan,p);
function prob_func(prob,i,repeat)
  remake(prob,u0=collect(AllICs[i]))
end

ensemble_prob = EnsembleProblem(prob, prob_func = prob_func, safetycopy=false)

monitoring_start()    
@time sol = solve(ensemble_prob,Tsit5(),EnsembleGPUArray(),trajectories=Trajs,saveat=SaveTimes)
results = monitoring_stop();
plot_monitoring_results(results)

```

---

<div class="post-metadata">

**Author:** ![yoshi](https://sea2.discourse-cdn.com/julialang/user_avatar/discourse.julialang.org/yoshi/32/38259_2.png) [@yoshi](https://discourse.julialang.org/u/yoshi)\
**Post date:** [October 3, 2022, 9:17pm UTC](https://discourse.julialang.org/t/gpu-kernel-compilation-error-error-in-sharing-functions-in-memory/88046/2 "2022-10-03T21:17:43Z")

</div>

The problem here was that all parameters need to passed into the function call through the parameter vector p. They can’t be defined earlier then be implicitly passed into the ODE.

For instance:  
`du[5] = -1*k*(A*Ica(u[3],u[1]) - u[5])`  
Must be  
`du[5] = -1*p[11]*(p[12]*Ica(u[3],u[1]) - u[5])`