Support for exceptions

[nicolamos]

For a simple situation where I need to take the floor of a number and convert it to a integer, the compilation step through GPUCompiler fails with

ERROR: InvalidIRError: compiling kernel #36#37(MtlDeviceVector{Float32, 1}) resulted in invalid LLVM IR
Reason: unsupported call to an unknown function (call to gpu_malloc)

The failing code is the following:

import Metal
Metal.@sync Metal.@metal (xs-> (x = xs[1]; k = Int(x); xs[1] = k; nothing))(Metal.MtlVector(Float32[1,2,3]))

[maleadt]

Yeah, this is currently expected, due to Metal.jl not supporting exceptions yet. You can work around this by doing a non-throwing conversion, e.g. trunc(Int, x) or x % Int.

[nicolamos]

Thank you for the suggestion!

[nicolamos]

Sorry, but I'm getting the same error.

• x % Int is not defined when x <: AbstractFloat;
• trunc(Int, x) does throw as well;
• unsafe_trunc(Int, x) can not be ported to the GPU, I guess;
• MT.trunc_fast does not accept an output type argument.

The error is the following:

ERROR: InvalidIRError: compiling kernel #55#56(Metal.MtlDeviceVector{Float32, 1}) resulted in invalid LLVM IR
Reason: unsupported call to an unknown function (call to gpu_malloc)
Stacktrace:
 [1] malloc
   @ ~/.julia/packages/GPUCompiler/hi5Wg/src/runtime.jl:89
 [2] macro expansion
   @ ~/.julia/packages/GPUCompiler/hi5Wg/src/runtime.jl:184
 [3] macro expansion
   @ ./none:0
 [4] box
   @ ./none:0
 [5] box_float32
   @ ~/.julia/packages/GPUCompiler/hi5Wg/src/runtime.jl:213
 [6] Int64
   @ ./float.jl:788
 [7] convert
   @ ./number.jl:7
 [8] #55
   @ ./REPL[32]:1
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.MetalCompilerTarget, Metal.MetalCompilerParams, GPUCompiler.FunctionSpec{var"#55#56", Tuple{Metal.MtlDeviceVector{Float32, 1}}}}, args::LLVM.Module)
    @ GPUCompiler ~/.julia/packages/GPUCompiler/hi5Wg/src/validation.jl:141
  [2] macro expansion
    @ ~/.julia/packages/GPUCompiler/hi5Wg/src/driver.jl:418 [inlined]
  [3] macro expansion
    @ ~/.julia/packages/TimerOutputs/LHjFw/src/TimerOutput.jl:253 [inlined]
  [4] macro expansion
    @ ~/.julia/packages/GPUCompiler/hi5Wg/src/driver.jl:416 [inlined]
  [5] emit_asm(job::GPUCompiler.CompilerJob, ir::LLVM.Module; strip::Bool, validate::Bool, format::LLVM.API.LLVMCodeGenFileType)
    @ GPUCompiler ~/.julia/packages/GPUCompiler/hi5Wg/src/utils.jl:68
  [6] mtlfunction_compile(job::GPUCompiler.CompilerJob, ctx::LLVM.Context)
    @ Metal ~/.julia/packages/Metal/pfCxO/src/compiler/execution.jl:168
  [7] #32
    @ ~/.julia/packages/Metal/pfCxO/src/compiler/execution.jl:161 [inlined]
  [8] JuliaContext(f::Metal.var"#32#33"{GPUCompiler.CompilerJob{GPUCompiler.MetalCompilerTarget, Metal.MetalCompilerParams, GPUCompiler.FunctionSpec{var"#55#56", Tuple{Metal.MtlDeviceVector{Float32, 1}}}}})
    @ GPUCompiler ~/.julia/packages/GPUCompiler/hi5Wg/src/driver.jl:76
  [9] mtlfunction_compile(job::GPUCompiler.CompilerJob)
    @ Metal ~/.julia/packages/Metal/pfCxO/src/compiler/execution.jl:160
 [10] cached_compilation(cache::Dict{UInt64, Any}, job::GPUCompiler.CompilerJob, compiler::typeof(Metal.mtlfunction_compile), linker::typeof(Metal.mtlfunction_link))
    @ GPUCompiler ~/.julia/packages/GPUCompiler/hi5Wg/src/cache.jl:90
 [11] mtlfunction(f::var"#55#56", tt::Type{Tuple{Metal.MtlDeviceVector{Float32, 1}}}; name::Nothing, kwargs::Base.Pairs{Symbol, Union{}, Tuple{}, NamedTuple{(), Tuple{}}})
    @ Metal ~/.julia/packages/Metal/pfCxO/src/compiler/execution.jl:148
 [12] mtlfunction(f::var"#55#56", tt::Type{Tuple{Metal.MtlDeviceVector{Float32, 1}}})
    @ Metal ~/.julia/packages/Metal/pfCxO/src/compiler/execution.jl:141
 [13] macro expansion
    @ ~/.julia/packages/Metal/pfCxO/src/compiler/execution.jl:64 [inlined]
 [14] top-level scope
    @ ~/.julia/packages/GPUCompiler/hi5Wg/src/reflection.jl:201

[maleadt]

unsafe_trunc(Int, x) can not be ported to the GPU, I guess;

Sorry, that's the one I meant, not plain trunc. Why does that not work for you on the GPU? k = unsafe_trunc(Int, x) does the job here.

[nicolamos]

Yes sorry, it works. I don’t know what was the issue… maybe I forgot the Int argument somewhere.

I also have another issue. I don’t really understand how can I use a function from another package. If I define the function in the same module (Main in this case), it works; but if import it from another package Metal complains that it is not defined.

[maleadt]

I also have another issue. I don’t really understand how can I use a function from another package. If I define the function in the same module (Main in this case), it works; but if import it from another package Metal complains that it is not defined.

That obviously should work fine. The bug tracker is not really suited for such queries though, but you can open a thread on Discourse with some actual code to demonstrate the issue.