ERROR: InvalidIRError: compiling kernel broadcast_kernel; Reason unsupported dynamic function invocation

[chriselrod]

Reason: unsupported dynamic function invocation (call to emit_printf(::Val{fmt}, argspec...) where fmt in oneAPI at /home/chriselrod/.julia/packages/oneAPI/bEvNc/src/device/opencl/printf.jl:27)

julia> using oneAPI

julia> a = oneArray(rand(2,2))
2×2 oneArray{Float64, 2}:
 0.94417   0.947858
 0.463884  0.31073

julia> a .+ 1
WARNING: both LLVM and ExprTools export "parameters"; uses of it in module oneAPI must be qualified
ERROR: InvalidIRError: compiling kernel broadcast_kernel(oneAPI.oneKernelContext, oneDeviceMatrix{Float64, 1}, Base.Broadcast.Broadcasted{Nothing, Tuple{Base.OneTo{Int64}, Base.OneTo{Int64}}, typeof(+), Tuple{Base.Broadcast.Extruded{oneDeviceMatrix{Float64, 1}, Tuple{Bool, Bool}, Tuple{Int64, Int64}}, Int64}}, Int64) resulted in invalid LLVM IR
Reason: unsupported dynamic function invocation (call to emit_printf(::Val{fmt}, argspec...) where fmt in oneAPI at /home/chriselrod/.julia/packages/oneAPI/bEvNc/src/device/opencl/printf.jl:27)
Stacktrace:
  [1] macro expansion
    @ ~/.julia/packages/oneAPI/bEvNc/src/device/opencl/printf.jl:129
  [2] _print
    @ ~/.julia/packages/oneAPI/bEvNc/src/device/opencl/printf.jl:91
  [3] macro expansion
    @ ~/.julia/packages/oneAPI/bEvNc/src/device/opencl/printf.jl:178
  [4] throw_boundserror
    @ ~/.julia/packages/oneAPI/bEvNc/src/device/quirks.jl:3
  [5] getindex
    @ range.jl:702
  [6] _broadcast_getindex_evalf
    @ broadcast.jl:648
  [7] _broadcast_getindex
    @ broadcast.jl:621
  [8] #19
    @ broadcast.jl:1098
  [9] ntuple
    @ ntuple.jl:49
 [10] copy
    @ broadcast.jl:1098
 [11] materialize
    @ broadcast.jl:883
 [12] getindex
    @ multidimensional.jl:353
 [13] _getindex
    @ abstractarray.jl:1209
 [14] getindex
    @ abstractarray.jl:1170
 [15] macro expansion
    @ ~/.julia/packages/GPUArrays/8dzSJ/src/device/indexing.jl:81
 [16] broadcast_kernel
    @ ~/.julia/packages/GPUArrays/8dzSJ/src/host/broadcast.jl:58
Stacktrace:
  [1] check_ir(job::GPUCompiler.CompilerJob{GPUCompiler.SPIRVCompilerTarget, oneAPI.oneAPICompilerParams, GPUCompiler.FunctionSpec{GPUArrays.var"#broadcast_kernel#16", Tuple{oneAPI.oneKernelContext, oneDeviceMatrix{Float64, 1}, Base.Broadcast.Broadcasted{Nothing, Tuple{Base.OneTo{Int64}, Base.OneTo{Int64}}, typeof(+), Tuple{Base.Broadcast.Extruded{oneDeviceMatrix{Float64, 1}, Tuple{Bool, Bool}, Tuple{Int64, Int64}}, Int64}}, Int64}}}, args::LLVM.Module)
    @ GPUCompiler ~/.julia/packages/GPUCompiler/QjFdA/src/validation.jl:111
  [2] macro expansion
    @ ~/.julia/packages/GPUCompiler/QjFdA/src/driver.jl:319 [inlined]
  [3] macro expansion
    @ ~/.julia/packages/TimerOutputs/ZQ0rt/src/TimerOutput.jl:236 [inlined]
  [4] macro expansion
    @ ~/.julia/packages/GPUCompiler/QjFdA/src/driver.jl:317 [inlined]
  [5] emit_asm(job::GPUCompiler.CompilerJob, ir::LLVM.Module; strip::Bool, validate::Bool, format::LLVM.API.LLVMCodeGenFileType)
    @ GPUCompiler ~/.julia/packages/GPUCompiler/QjFdA/src/utils.jl:62
  [6] zefunction_compile(job::GPUCompiler.CompilerJob)
    @ oneAPI ~/.julia/packages/oneAPI/bEvNc/src/compiler/execution.jl:131
  [7] cached_compilation(cache::Dict{UInt64, Any}, job::GPUCompiler.CompilerJob, compiler::typeof(oneAPI.zefunction_compile), linker::typeof(oneAPI.zefunction_link))
    @ GPUCompiler ~/.julia/packages/GPUCompiler/QjFdA/src/cache.jl:89
  [8] zefunction(f::Function, tt::Type; name::Nothing, kwargs::Base.Iterators.Pairs{Union{}, Union{}, Tuple{}, NamedTuple{(), Tuple{}}})
    @ oneAPI ~/.julia/packages/oneAPI/bEvNc/src/compiler/execution.jl:122
  [9] macro expansion
    @ ~/.julia/packages/oneAPI/bEvNc/src/compiler/execution.jl:34 [inlined]
 [10] #gpu_call#77
    @ ~/.julia/packages/oneAPI/bEvNc/src/gpuarrays.jl:29 [inlined]
 [11] gpu_call(::GPUArrays.var"#broadcast_kernel#16", ::oneArray{Float64, 2}, ::Base.Broadcast.Broadcasted{Nothing, Tuple{Base.OneTo{Int64}, Base.OneTo{Int64}}, typeof(+), Tuple{Base.Broadcast.Extruded{oneArray{Float64, 2}, Tuple{Bool, Bool}, Tuple{Int64, Int64}}, Int64}}, ::Int64; target::oneArray{Float64, 2}, total_threads::Nothing, threads::Int64, blocks::Int64, name::Nothing)
    @ GPUArrays ~/.julia/packages/GPUArrays/8dzSJ/src/device/execution.jl:67
 [12] copyto!
    @ ~/.julia/packages/GPUArrays/8dzSJ/src/host/broadcast.jl:65 [inlined]
 [13] copyto!
    @ ./broadcast.jl:936 [inlined]
 [14] copy
    @ ~/.julia/packages/GPUArrays/8dzSJ/src/host/broadcast.jl:47 [inlined]
 [15] materialize(bc::Base.Broadcast.Broadcasted{oneAPI.oneArrayStyle{2}, Nothing, typeof(+), Tuple{oneArray{Float64, 2}, Int64}})
    @ Base.Broadcast ./broadcast.jl:883
 [16] top-level scope
    @ REPL[3]:1

julia> versioninfo()
Julia Version 1.6.3-pre.1
Commit 7c45ff0e94* (2021-07-16 20:20 UTC)
Platform Info:
  OS: Linux (x86_64-redhat-linux)
  CPU: 11th Gen Intel(R) Core(TM) i7-1165G7 @ 2.80GHz
  WORD_SIZE: 64
  LIBM: libopenlibm
  LLVM: libLLVM-11.0.1 (ORCJIT, tigerlake)
Environment:
  JULIA_NUM_THREADS = 8

julia> using oneAPI.oneL0

julia> drv = first(drivers());

julia> dev = first(devices(drv))
ZeDevice(GPU, vendor 0x8086, device 0x9a49): Intel(R) Iris(R) Xe Graphics [0x9a49]

julia> compute_properties(dev)
(maxTotalGroupSize = 512, maxGroupSizeX = 512, maxGroupSizeY = 512, maxGroupSizeZ = 512, maxGroupCountX = 4294967295, maxGroupCountY = 4294967295, maxGroupCountZ = 4294967295, maxSharedLocalMemory = 65536, subGroupSizes = (8, 16, 32))

julia> ctx = ZeContext(drv);

julia> queue = ZeCommandQueue(ctx, dev);

julia> execute!(queue) do list
         append_barrier!(list)
       end

julia> function kernel()
         barrier()
         return
       end
kernel (generic function with 1 method)

julia> @oneapi items=1 kernel()

julia> @device_code_llvm @oneapi items=1 kernel()
; CompilerJob of kernel kernel() for GPUCompiler.SPIRVCompilerTarget
define spir_kernel void @_Z17julia_kernel_4443() local_unnamed_addr {
entry:
;  @ REPL[18]:2 within `kernel'
; ┌ @ /home/chriselrod/.julia/packages/oneAPI/bEvNc/src/device/opencl/synchronization.jl:9 within `barrier' @ /home/chriselrod/.julia/packages/oneAPI/bEvNc/src/device/opencl/synchronization.jl:9
; │┌ @ /home/chriselrod/.julia/packages/oneAPI/bEvNc/src/device/utils.jl:47 within `macro expansion'
    call void @_Z7barrierj(i32 0)
    ret void
; └└
}

[maleadt]

This is JuliaGPU/GPUCompiler.jl#214, due to that ambiguity the implementation of printf doesn't infer (but wants to throw a run-time error).

Shouldn't happen on the master branch, and I'll be tagging a release soon!

[chriselrod]

As an update:

julia> a = oneArray(rand(Float32, 2,2))
2×2 oneArray{Float32, 2}:
 0.787535  0.849568
 0.418914  0.621363

julia> a .+ 1f0
2×2 oneArray{Float32, 2}:
 1.78754  1.84957
 1.41891  1.62136

julia> a = oneArray(rand(2,2))
2×2 oneArray{Float64, 2}:
 0.130347  0.95715
 0.450905  0.869891

julia> a .+ 1
┌ Error: Module compilation failed:
│
│ error: double type is not supported on this platform

Perhaps worth making the example use Float32 by default?
Iris(R) Xe Graphics are common in laptop iGPUs over the past year.

[maleadt]

Yeah that makes sense, I've updated the README.
(My IGP did have Float64 support, is why I didn't run into it earlier.)