Global hostcalls when setting a view, e.g. x[boolean_mask] .= 0

[Alexander-Barth]

I am getting this error when using AMDGPU trying to replace all NaNs by another value:

julia> import AMDGPU
julia> x0 = AMDGPU.roc([1,2,3,NaN32]);
julia> x0[isnan.(x0)] .= 0;
┌ Warning: Global hostcalls detected!
│ - Source: MethodInstance for AcceleratedKernels.gpu__accumulate_block!(::KernelAbstractions.CompilerMetadata{KernelAbstractions.NDIteration.DynamicSize, KernelAbstractions.NDIteration.DynamicCheck, Nothing, CartesianIndices{1, Tuple{Base.OneTo{Int64}}}, KernelAbstractions.NDIteration.NDRange{1, KernelAbstractions.NDIteration.DynamicSize, KernelAbstractions.NDIteration.StaticSize{(256,)}, CartesianIndices{1, Tuple{Base.OneTo{Int64}}}, Nothing}}, ::typeof(+), ::AMDGPU.Device.ROCDeviceVector{Bool, 1}, ::Bool, ::Bool, ::Bool, ::AMDGPU.Device.ROCDeviceVector{Int8, 1}, ::AMDGPU.Device.ROCDeviceVector{Bool, 1})
│ - Hostcalls: [:malloc_hostcall]
│ 
│ Use `AMDGPU.synchronize(; stop_hostcalls=false)` to synchronize and stop them.
│ Otherwise, performance might degrade if they keep running in the background.
└ @ AMDGPU.Compiler /tmp/julia-depot-DINDiff-barthale/packages/AMDGPU/BgSqf/src/compiler/codegen.jl:208
Memory access fault by GPU node-4 (Agent handle: 0x10f0140) on address 0x152f8c650000. Reason: Unknown.

[115426] signal 6 (-6): Aborted
in expression starting at REPL[5]:1
Allocations: 28864586 (Pool: 28863762; Big: 824); GC: 21

I am using AMDGPU 1.2.3 and julia 1.11.3.

julia> AMDGPU.versioninfo()
[ Info: AMDGPU versioninfo
┌───────────┬──────────────────┬───────────┬─────────────────────────────────────────────────────────────────────────────────────────────────────┐
│ Available │ Name             │ Version   │ Path                                                                                                │
├───────────┼──────────────────┼───────────┼─────────────────────────────────────────────────────────────────────────────────────────────────────┤
│     +     │ LLD              │ -         │ /opt/rocm-6.0.3/llvm/bin/ld.lld                                                                     │
│     +     │ Device Libraries │ -         │ /tmp/julia-depot-DINDiff-barthale/artifacts/5ad5ecb46e3c334821f54c1feecc6c152b7b6a45/amdgcn/bitcode │
│     +     │ HIP              │ 6.0.32831 │ /opt/rocm-6.0.3/lib/libamdhip64.so                                                                  │
│     +     │ rocBLAS          │ 4.0.0     │ /opt/rocm-6.0.3/lib/librocblas.so                                                                   │
│     +     │ rocSOLVER        │ 3.24.0    │ /opt/rocm-6.0.3/lib/librocsolver.so                                                                 │
│     +     │ rocSPARSE        │ 3.0.2     │ /opt/rocm-6.0.3/lib/librocsparse.so                                                                 │
│     +     │ rocRAND          │ 2.10.5    │ /opt/rocm-6.0.3/lib/librocrand.so                                                                   │
│     +     │ rocFFT           │ 1.0.27    │ /opt/rocm-6.0.3/lib/librocfft.so                                                                    │
│     +     │ MIOpen           │ 3.0.0     │ /opt/rocm-6.0.3/lib/libMIOpen.so                                                                    │
└───────────┴──────────────────┴───────────┴─────────────────────────────────────────────────────────────────────────────────────────────────────┘

[ Info: AMDGPU devices
┌────┬─────────────────────┬────────────────────────┬───────────┬────────────┬───────────────┐
│ Id │                Name │               GCN arch │ Wavefront │     Memory │ Shared Memory │
├────┼─────────────────────┼────────────────────────┼───────────┼────────────┼───────────────┤
│  1 │ AMD Instinct MI250X │ gfx90a:sramecc+:xnack- │        64 │ 63.984 GiB │    64.000 KiB │
└────┴─────────────────────┴────────────────────────┴───────────┴────────────┴───────────────┘

This is not a big problem for me as I can use x0 = ifelse.(isnan.(x0),zero(eltype(x0)),x0); instead.

Thank you for this great packages. I would not be able to use some "serious" compute cluster for my research without this package :-)

[pxl-th]

That's a regression from #730... Not sure how it was not caught before.

MWE:

julia> x = AMDGPU.rand(Bool, 16);
julia> cumsum(m)

If this is a blocker for you, you can use [email protected]. It'll be fixed soon.

[pxl-th]

Thank you for this great packages. I would not be able to use some "serious" compute cluster for my research without this package :-)

Thanks!

[pxl-th]

Should be fixed by JuliaGPU/AcceleratedKernels.jl#24

[Alexander-Barth]

This is great! Thank you so much!

[pxl-th]

New version of AcceleratedKernels is released, so should be enough to just run ]up command in your project.
I'll close the issue.