Issue with nested device arrays

is the below supposed a bug, incorrectly implemented by me, or just not possible?

```julia
using KernelAbstractions
using pocl_jll, OpenCL
using OpenCL: CLArray

cl.device!(cl.devices(cl.platforms()[1])[1])

backend = OpenCL.OpenCLBackend()

KernelAbstractions.@kernel _dummy_kernel() = nothing
function to_device(backend, arr::CLArray)
kernel = _dummy_kernel(backend)
return KernelAbstractions.argconvert(kernel, arr)
end

# A composite element type (like RGBSpectrum)
struct MyColor
c::NTuple{4, Float32}
end
MyColor(r, g, b) = MyColor((r, g, b, 1.0f0))

# A wrapper type (like Texture)
struct MyTexture{T}
data::T
default::MyColor
isconst::Bool
end

# A material type (like MatteMaterial)
struct MyMaterial{T}
albedo::MyTexture{T}
end

# Create test data
cpu_colors = [MyColor(1f0, 0f0, 0f0) MyColor(0f0, 1f0, 0f0);
MyColor(0f0, 0f0, 1f0) MyColor(1f0, 1f0, 1f0)]
gpu_colors = CLArray(cpu_colors)

# Create material with GPU texture
gpu_tex = MyTexture(to_device(backend, gpu_colors), MyColor(0f0, 0f0, 0f0), false)
gpu_mat = MyMaterial(gpu_tex)

# Store material in a vector - THIS IS THE KEY
gpu_mats = CLArray([gpu_mat])

# Kernel that accesses nested data
@kernel function mwe_kernel!(result, materials)
i = @index(Global)
mat = materials[1] # Extract material from vector
tex = mat.albedo # Get texture wrapper
data = tex.data # Get device matrix
elem = data[1, 1] # Get element (MyColor)
result[i] = elem.c[1] # Access field of element - FAILS
end

kernel! = mwe_kernel!(backend)
result = CLArray([0.0f0])

kernel!(result, gpu_mats; ndrange=1)
```

```
OpenCL.Device(gfx1100 on AMD Accelerated Parallel Processing @0x000000000f75e3d0)

error: line 234: Result type (OpTypePointer) does not match the type that results from indexing into the composite (OpTypePointer).
%_fca_0_0_0_extract = OpCompositeExtract %_ptr_CrossWorkgroup_float %145 0 0

ERROR: Failed to validate generated SPIR-V.
If you think this is a bug, please file an issue and attach /tmp/jl_PE7wxaHevW.bc and /tmp/jl_rwR6vK5ABG.spv.
Stacktrace:
[1] error(s::String)
@ Base ./error.jl:44
[2] mcgen(job::GPUCompiler.CompilerJob{…}, mod::LLVM.Module, format::LLVM.API.LLVMCodeGenFileType)
@ GPUCompiler ~/.julia/packages/GPUCompiler/j4HFa/src/spirv.jl:184
[3] emit_asm(job::GPUCompiler.CompilerJob, ir::LLVM.Module, format::LLVM.API.LLVMCodeGenFileType)
@ GPUCompiler ~/.julia/packages/GPUCompiler/j4HFa/src/driver.jl:438
[4] compile_unhooked(output::Symbol, job::GPUCompiler.CompilerJob; kwargs::@Kwargs{})
@ GPUCompiler ~/.julia/packages/GPUCompiler/j4HFa/src/driver.jl:115
[5] compile_unhooked
@ ~/.julia/packages/GPUCompiler/j4HFa/src/driver.jl:80 [inlined]
[6] compile(target::Symbol, job::GPUCompiler.CompilerJob; kwargs::@Kwargs{})
@ GPUCompiler ~/.julia/packages/GPUCompiler/j4HFa/src/driver.jl:67
[7] compile
@ ~/.julia/packages/GPUCompiler/j4HFa/src/driver.jl:55 [inlined]
[8] (::OpenCL.var"#compile##0#compile##1"{GPUCompiler.CompilerJob{…}})(ctx::LLVM.Context)
@ OpenCL ~/.julia/packages/OpenCL/kWFms/src/compiler/compilation.jl:183
[9] JuliaContext(f::OpenCL.var"#compile##0#compile##1"{GPUCompiler.CompilerJob{…}}; kwargs::@Kwargs{})
@ GPUCompiler ~/.julia/packages/GPUCompiler/j4HFa/src/driver.jl:34
[10] JuliaContext(f::Function)
@ GPUCompiler ~/.julia/packages/GPUCompiler/j4HFa/src/driver.jl:25
[11] compile(job::GPUCompiler.CompilerJob)
@ OpenCL ~/.julia/packages/OpenCL/kWFms/src/compiler/compilation.jl:182
[12] actual_compilation(cache::Dict{…}, src::Core.MethodInstance, world::UInt64, cfg::GPUCompiler.CompilerConfig{…}, compiler::typeof(OpenCL.compile), linker::typeof(OpenCL.link))
@ GPUCompiler ~/.julia/packages/GPUCompiler/j4HFa/src/execution.jl:245
[13] cached_compilation(cache::Dict{…}, src::Core.MethodInstance, cfg::GPUCompiler.CompilerConfig{…}, compiler::Function, linker::Function)
@ GPUCompiler ~/.julia/packages/GPUCompiler/j4HFa/src/execution.jl:159
[14] macro expansion
@ ~/.julia/packages/OpenCL/kWFms/src/compiler/execution.jl:191 [inlined]
[15] macro expansion
@ ./lock.jl:376 [inlined]
[16] clfunction(f::typeof(gpu_mwe_kernel!), tt::Type{Tuple{…}}; kwargs::@Kwargs{})
@ OpenCL ~/.julia/packages/OpenCL/kWFms/src/compiler/execution.jl:186
[17] clfunction(f::typeof(gpu_mwe_kernel!), tt::Type{Tuple{…}})
@ OpenCL ~/.julia/packages/OpenCL/kWFms/src/compiler/execution.jl:182
[18] macro expansion
@ ~/.julia/packages/OpenCL/kWFms/src/compiler/execution.jl:66 [inlined]
[19] (::KernelAbstractions.Kernel{…})(::CLArray{…}, ::Vararg{…}; ndrange::Int64, workgroupsize::Nothing)
@ OpenCL.OpenCLKernels ~/.julia/packages/OpenCL/kWFms/src/OpenCLKernels.jl:104
[20] top-level scope
@ /sim/Programmieren/RayTracing/cl_compilation.jl:59

caused by: failed process: Process(setenv(`/home/simon/.julia/artifacts/8769e627f0911e0e6ab9b58f5961768a06de3e61/bin/spirv-val /tmp/jl_rwR6vK5ABG.spv`,["PATH=/home/simon/.julia/artifacts/8769e627f0911e0e6ab9b58f5961768a06de3e61/bin:/home/simon/.config/Code/User/globalStorage/github.copilot-chat/debugCommand:/home/simon/.config/Code/User/globalStorage/github.copilot-chat/copilotCli:/home/simon/.local/bin:/usr/local/bin:/usr/bin:/bin:/snap/bin", "ENV=/usr/share/Modules/init/profile.sh", "CPU=x86_64", "JRE_HOME=/usr/lib64/jvm/java-21-openjdk-21", "FROM_HEADER=", "G_FILENAME_ENCODING=@locale,UTF-8,ISO-8859-15,CP1252", "NO_AT_BRIDGE=1", "XNLSPATH=/usr/share/X11/nls", "POCL_ARGS_CLANG=-fuse-ld=lld;--ld-path=/home/simon/.julia/scratchspaces/627d6b7a-bbe6-5189-83e7-98cc0a5aeadd/bin/lld;-L;/home/simon/.julia/artifacts/dbb2d45cf83dd8f12f3ad2a436243a1ac76cb71a/share/lib;-L;/home/simon/.julia/scratchspaces/627d6b7a-bbe6-5189-83e7-98cc0a5aeadd/lib", "LD_LIBRARY_PATH=/home/simon/.julia/artifacts/8769e627f0911e0e6ab9b58f5961768a06de3e61/lib:/home/simon/.julia/juliaup/julia-1.12.4+0.x64.linux.gnu/bin/../lib/julia:/home/simon/.julia/juliaup/julia-1.12.4+0.x64.linux.gnu/bin/../lib" … "GTK2_RC_FILES=/etc/gtk-2.0/gtkrc:/home/simon/.gtkrc-2.0:/home/simon/.config/gtkrc-2.0", "PYTHONSTARTUP=/etc/pythonstart", "MORE=-sl", "GMAIL_APP_PASSWORD=rvrw cdbm hqfw cuae", "POCL_PATH_LLVM_SPIRV=/home/simon/.julia/scratchspaces/627d6b7a-bbe6-5189-83e7-98cc0a5aeadd/bin/llvm-spirv", "AUDIODRIVER=pulseaudio", "LS_COLORS=", "HOSTTYPE=x86_64", "KDE_SESSION_VERSION=6", "OPENBLAS_MAIN_FREE=1"]), ProcessExited(1)) [1]

Stacktrace:
[1] pipeline_error
@ ./process.jl:597 [inlined]
[2] run(::Cmd; wait::Bool)
@ Base ./process.jl:512
[3] run
@ ./process.jl:509 [inlined]
[4] mcgen(job::GPUCompiler.CompilerJob{…}, mod::LLVM.Module, format::LLVM.API.LLVMCodeGenFileType)
@ GPUCompiler ~/.julia/packages/GPUCompiler/j4HFa/src/spirv.jl:182
[5] emit_asm(job::GPUCompiler.CompilerJob, ir::LLVM.Module, format::LLVM.API.LLVMCodeGenFileType)
@ GPUCompiler ~/.julia/packages/GPUCompiler/j4HFa/src/driver.jl:438
[6] compile_unhooked(output::Symbol, job::GPUCompiler.CompilerJob; kwargs::@Kwargs{})
@ GPUCompiler ~/.julia/packages/GPUCompiler/j4HFa/src/driver.jl:115
[7] compile_unhooked
@ ~/.julia/packages/GPUCompiler/j4HFa/src/driver.jl:80 [inlined]
[8] compile(target::Symbol, job::GPUCompiler.CompilerJob; kwargs::@Kwargs{})
@ GPUCompiler ~/.julia/packages/GPUCompiler/j4HFa/src/driver.jl:67
[9] compile
@ ~/.julia/packages/GPUCompiler/j4HFa/src/driver.jl:55 [inlined]
[10] (::OpenCL.var"#compile##0#compile##1"{GPUCompiler.CompilerJob{…}})(ctx::LLVM.Context)
@ OpenCL ~/.julia/packages/OpenCL/kWFms/src/compiler/compilation.jl:183
[11] JuliaContext(f::OpenCL.var"#compile##0#compile##1"{GPUCompiler.CompilerJob{…}}; kwargs::@Kwargs{})
@ GPUCompiler ~/.julia/packages/GPUCompiler/j4HFa/src/driver.jl:34
[12] JuliaContext(f::Function)
@ GPUCompiler ~/.julia/packages/GPUCompiler/j4HFa/src/driver.jl:25
[13] compile(job::GPUCompiler.CompilerJob)
@ OpenCL ~/.julia/packages/OpenCL/kWFms/src/compiler/compilation.jl:182
[14] actual_compilation(cache::Dict{…}, src::Core.MethodInstance, world::UInt64, cfg::GPUCompiler.CompilerConfig{…}, compiler::typeof(OpenCL.compile), linker::typeof(OpenCL.link))
@ GPUCompiler ~/.julia/packages/GPUCompiler/j4HFa/src/execution.jl:245
[15] cached_compilation(cache::Dict{…}, src::Core.MethodInstance, cfg::GPUCompiler.CompilerConfig{…}, compiler::Function, linker::Function)
@ GPUCompiler ~/.julia/packages/GPUCompiler/j4HFa/src/execution.jl:159
[16] macro expansion
@ ~/.julia/packages/OpenCL/kWFms/src/compiler/execution.jl:191 [inlined]
[17] macro expansion
@ ./lock.jl:376 [inlined]
[18] clfunction(f::typeof(gpu_mwe_kernel!), tt::Type{Tuple{…}}; kwargs::@Kwargs{})
@ OpenCL ~/.julia/packages/OpenCL/kWFms/src/compiler/execution.jl:186
[19] clfunction(f::typeof(gpu_mwe_kernel!), tt::Type{Tuple{…}})
@ OpenCL ~/.julia/packages/OpenCL/kWFms/src/compiler/execution.jl:182
[20] macro expansion
@ ~/.julia/packages/OpenCL/kWFms/src/compiler/execution.jl:66 [inlined]
[21] (::KernelAbstractions.Kernel{…})(::CLArray{…}, ::Vararg{…}; ndrange::Int64, workgroupsize::Nothing)
@ OpenCL.OpenCLKernels ~/.julia/packages/OpenCL/kWFms/src/OpenCLKernels.jl:104
[22] top-level scope
@ /sim/Programmieren/RayTracing/cl_compilation.jl:59
Some type information was truncated. Use `show(err)` to see complete types.
```

The same _does_ work with AMDGPU, pointing to an OpenCL bug.