unsupported call to an unknown function (call to __nv_fabsf)

[kshyatt]

MWE:

using CUDA                                                                                                                                                                                                           
                                                                                                                                                                                                                     
A = CUDA.cuRAND.randn(64)                                                                                                                                                                                            
                                                                                                                                                                                                                     
B = abs.(A)

Full trace:

ERROR: LoadError: InvalidIRError: compiling MethodInstance for (::GPUArrays.var"#gpu_broadcast_kernel_linear#_copyto!##1")(::KernelAbstractions.CompilerMetadata{KernelAbstractions.NDIteration.DynamicSize, KernelAbstractions.NDIteration.DynamicCheck, Nothing, CartesianIndices{1, Tuple{Base.OneTo{Int64}}}, KernelAbstractions.NDIteration.NDRange{1, KernelAbstractions.NDIteration.DynamicSize, KernelAbstractions.NDIteration.DynamicSize, CartesianIndices{1, Tuple{Base.OneTo{Int64}}}, CartesianIndices{1, Tuple{Base.OneTo{Int64}}}}}, ::CuDeviceVector{Float32, 1}, ::Base.Broadcast.Broadcasted{CUDACore.CuArrayStyle{1, CUDACore.DeviceMemory}, Tuple{Base.OneTo{Int64}}, typeof(abs), Tuple{Base.Broadcast.Extruded{CuDeviceVector{Float32, 1}, Tuple{Bool}, Tuple{Int64}}}}) resulted in invalid LLVM IR                                                          
Reason: unsupported call to an unknown function (call to __nv_fabsf)                                                                                                                                                 
Stacktrace:                                                                                                                                                                                                          
 [1] abs                                                                                                                                                                                                             
   @ ~/.julia/dev/CUDA/CUDACore/src/device/intrinsics/math.jl:316                                                                                                                                                    
 [2] _broadcast_getindex_evalf                                                                                                                                                                                       
   @ ./broadcast.jl:699                                                                                                                                                                                              
 [3] _broadcast_getindex                                                                                                                                                                                             
   @ ./broadcast.jl:672                                                                                                                                                                                              
 [4] _getindex                                                                                                                                                                                                       
   @ ./broadcast.jl:620                                                                                                                                                                                              
 [5] getindex                                                                                                                                                                                                        
   @ ./broadcast.jl:616                                                                                                                                                                                              
 [6] macro expansion                                                                                                                                                                                                 
   @ ~/.julia/packages/KernelAbstractions/ecO4B/src/macros.jl:332                                                                                                                                                    
 [7] gpu_broadcast_kernel_linear                                                                                                                                                                                     
   @ ./none:0                          

[vchuravy]

Which CUDA version?

[maleadt]

Which CUDA version?

and LLVM version, and GPUCompiler version, etc. That's all relevant here.

[kshyatt]

Sorry I went to lunch. It’s main on cyclops, I’ll post the rest soon

…

On Tue, May 19, 2026 at 1:04 PM Tim Besard ***@***.***> wrote: *maleadt* left a comment (JuliaGPU/CUDA.jl#3147) <#3147 (comment)> Which CUDA version? and LLVM version, and GPUCompiler version, etc. That's all relevant here. — Reply to this email directly, view it on GitHub <#3147 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAGKJY5XYFBPYWIYAZBWMXL43Q5VHAVCNFSM6AAAAACZEC45FOVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHM2DIOBXGEYDSMRYGU> . Triage notifications on the go with GitHub Mobile for iOS <https://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675> or Android <https://play.google.com/store/apps/details?id=com.github.android&referrer=utm_campaign%3Dnotification-email%26utm_medium%3Demail%26utm_source%3Dgithub>. You are receiving this because you authored the thread.Message ID: ***@***.***>

[kshyatt]

julia> using CUDA; CUDA.versioninfo()
CUDA toolchain: 
- runtime 13.2.0, artifact installation
- driver 595.58.3 for 13.2
- compiler 13.2.51, artifact installation

CUDA libraries: 
- cuBLAS: 13.4.0
- cuSPARSE: 12.7.10
- cuSOLVER: 12.2.0
- cuFFT: 12.2.0
- cuRAND: 10.4.2
- CUPTI: 2026.1.1 (API 13.2.1)
- NVML: 13.0.0+595.58.3

Julia packages: 
- CUDACore: 6.1.0
- GPUArrays: 11.5.2
- GPUCompiler: 1.9.1
- KernelAbstractions: 0.9.41
- CUDA_Driver_jll: 13.2.1+0
- CUDA_Compiler_jll: 0.4.3+0
- CUDA_Runtime_jll: 0.21.0+1

Toolchain:
- Julia: 1.12.6
- LLVM: 18.1.7

1 device:
  0: NVIDIA GeForce RTX 5080 (sm_120, 15.467 GiB / 15.921 GiB available)

Julia version 1.12.6

[maleadt]

Your GPUCompiler.jl is too old for CUDA.jl#main:

CUDA.jl/CUDACore/Project.toml

Line 56 in 0bd53ac

GPUCompiler = "1.10"

[kshyatt]

D: pkg noooooo why didn't you save me