Impossible to cast a float into an integer

[albop]

It currently seems impossible (to me?) to convert a float into an integer inside a oneapi kernel.
I tried with several functions like round(Int, ...), trunc(Int, ...) and also with convert(Int(rount(...))) but I always get a compilation error.

Here is a self contained example with the corresponding error log.

using oneAPI
using KernelAbstractions
using Adapt
const KA = KernelAbstractions

@kernel function round_kernel(A,B)
    I = @index(Global)
    # rounding operation
    A[I] = trunc(Int,B[I])
end

A = zeros(Int8, 1024, 1024)
B = rand(Float32, 1024, 1024)*10
A_gpu = adapt(oneArray, A)
B_gpu = adapt(oneArray, B)
dev = get_backend(A_gpu)
round_kernel(dev, 64)(A_gpu, B_gpu, ndrange=size(A_gpu))
KA.synchronize(dev)

ERROR: InvalidIRError: compiling MethodInstance for gpu_round_kernel(::KernelAbstractions.CompilerMetadata{…}, ::oneDeviceMatrix{…}, ::oneDeviceMatrix{…}) resulted in invalid LLVM IR
Reason: unsupported call to an unknown function (call to gpu_malloc)
Stacktrace:
 [1] malloc
   @ ~/.julia/packages/GPUCompiler/kqxyC/src/runtime.jl:88
 [2] macro expansion
   @ ~/.julia/packages/GPUCompiler/kqxyC/src/runtime.jl:183
 [3] macro expansion
   @ ./none:0
 [4] box
  @ ./none:0
 [5] box_float32
   @ ~/.julia/packages/GPUCompiler/kqxyC/src/runtime.jl:212
 [6] trunc
   @ ./float.jl:905
 [7] macro expansion
   @ ~/Econforge/Dolo.jl/misc/test_oneapi.jl:173
 [8] gpu_round_kernel
   @ ~/.julia/packages/KernelAbstractions/sZvJo/src/macros.jl:95
 [9] gpu_round_kernel
   @ ./none:0
Hint: catch this exception as `err` and call `code_typed(err; interactive = true)` to introspect the erronous code with Cthulhu.jl
Stacktrace:
  [1] check_ir(job::GPUCompiler.CompilerJob{GPUCompiler.SPIRVCompilerTarget, oneAPI.oneAPICompilerParams}, args::LLVM.Module)
    @ GPUCompiler ~/.julia/packages/GPUCompiler/kqxyC/src/validation.jl:147
  [2] macro expansion
    @ ~/.julia/packages/GPUCompiler/kqxyC/src/driver.jl:445 [inlined]
  [3] macro expansion
    @ ~/.julia/packages/TimerOutputs/Lw5SP/src/TimerOutput.jl:253 [inlined]
  [4] macro expansion
    @ ~/.julia/packages/GPUCompiler/kqxyC/src/driver.jl:444 [inlined]
  [5] 
    @ GPUCompiler ~/.julia/packages/GPUCompiler/kqxyC/src/utils.jl:92
  [6] emit_llvm
    @ ~/.julia/packages/GPUCompiler/kqxyC/src/utils.jl:86 [inlined]
  [7] 
    @ GPUCompiler ~/.julia/packages/GPUCompiler/kqxyC/src/driver.jl:134
  [8] codegen
    @ ~/.julia/packages/GPUCompiler/kqxyC/src/driver.jl:115 [inlined]
  [9] 
    @ GPUCompiler ~/.julia/packages/GPUCompiler/kqxyC/src/driver.jl:111
 [10] compile
    @ ~/.julia/packages/GPUCompiler/kqxyC/src/driver.jl:103 [inlined]
 [11] #58
    @ ~/.julia/dev/oneAPI/src/compiler/compilation.jl:81 [inlined]
 [12] JuliaContext(f::oneAPI.var"#58#59"{GPUCompiler.CompilerJob{…}}; kwargs::@Kwargs{})
    @ GPUCompiler ~/.julia/packages/GPUCompiler/kqxyC/src/driver.jl:52
 [13] JuliaContext(f::Function)
    @ GPUCompiler ~/.julia/packages/GPUCompiler/kqxyC/src/driver.jl:42
 [14] compile(job::GPUCompiler.CompilerJob)
    @ oneAPI ~/.julia/dev/oneAPI/src/compiler/compilation.jl:80
 [15] actual_compilation(cache::Dict{…}, src::Core.MethodInstance, world::UInt64, cfg::GPUCompiler.CompilerConfig{…}, compiler::typeof(oneAPI.compile), linker::typeof(oneAPI.link))
    @ GPUCompiler ~/.julia/packages/GPUCompiler/kqxyC/src/execution.jl:128
 [16] cached_compilation(cache::Dict{…}, src::Core.MethodInstance, cfg::GPUCompiler.CompilerConfig{…}, compiler::Function, linker::Function)
    @ GPUCompiler ~/.julia/packages/GPUCompiler/kqxyC/src/execution.jl:103
 [17] macro expansion
    @ ~/.julia/dev/oneAPI/src/compiler/execution.jl:203 [inlined]
 [18] macro expansion
    @ ./lock.jl:267 [inlined]
 [19] zefunction(f::typeof(gpu_round_kernel), tt::Type{Tuple{…}}; kwargs::@Kwargs{})
    @ oneAPI ~/.julia/dev/oneAPI/src/compiler/execution.jl:198
 [20] zefunction(f::typeof(gpu_round_kernel), tt::Type{Tuple{…}})
    @ oneAPI ~/.julia/dev/oneAPI/src/compiler/execution.jl:195
 [21] macro expansion
    @ ~/.julia/dev/oneAPI/src/compiler/execution.jl:66 [inlined]
 [22] (::KernelAbstractions.Kernel{…})(::oneArray{…}, ::Vararg{…}; ndrange::Tuple{…}, workgroupsize::Nothing)
    @ oneAPI.oneAPIKernels ~/.julia/dev/oneAPI/src/oneAPIKernels.jl:89
 [23] top-level scope
    @ ~/Econforge/Dolo.jl/misc/test_oneapi.jl:182
Some type information was truncated. Use `show(err)` to see complete types.

[albop]

Actually, I have just found a way:

@kernel function round_kernel(A,B)
  I = @index(Global)
  v = trunc(B[I])
  A[I] = (v::Int8)
end

Not sure what to do with this issue. I guess the GPU definitions of trunc, round, ceil, could potentially be enhanced.

[albop]

The comment above is wrong: the kernel above doesn' t raise an exception but doesn' t do anything either. I guess the instruction v::Int8 is just an assertion, which invalidates silently the whole kernel.

[maleadt]

This is a known limitation, tracked in #65.
In your case, use unsafe_trunc instead, which doesn't generate an exception (and thus doesn't require an allocation). It's unfortunate, but adding overlay definitions to trunc is much too invasive, so time is better spent working on a device-side allocator fixing #65.

[albop]

Thank you ! Le jeu. 23 mai 2024, 10:27, Tim Besard ***@***.***> a écrit :

…

This is a known limitation, tracked in #65 <#65>. In your case, use unsafe_trunc instead, which doesn't generate an exception (and thus doesn't require an allocation). It's unfortunate, but adding overlay definitions to trunc is much too invasive, so time is better spent working on a device-side allocator fixing #65 <#65>. — Reply to this email directly, view it on GitHub <#441 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AACDSKJM2TKXKDW5PL3K2M3ZDWSAPAVCNFSM6AAAAABIEYBLTGVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDCMRWGUZDKOJUGA> . You are receiving this because you authored the thread.Message ID: ***@***.***>