sind and cosd not working with Float32 oneArrays

[pvillacorta]

Supose 4 different arrays:

using CUDA, oneAPI

a = CuArray([0.1, 0.2, 0.3])
b = CuArray([0.1f0, 0.2f0, 0.3f0])
c = oneArray([0.1, 0.2, 0.3])
d = oneArray([0.1f0, 0.2f0, 0.3f0])

We calculate sind by broadcasting:

julia> sind.(a) #works
3-element CuArray{Float64, 1, CUDA.DeviceMemory}:
 0.001745328365898309
 0.0034906514152237326
 0.00523596383141958

julia> sind.(b) #works
3-element CuArray{Float32, 1, CUDA.DeviceMemory}:
 0.0017453284
 0.0034906515
 0.005235964

julia> sind.(c) #works
3-element oneArray{Float64, 1, oneAPI.oneL0.DeviceBuffer}:
 0.001745328365898309
 0.0034906514152237326
 0.00523596383141958

These three work. However, when we try with a oneArray composed by Float32 elements, it throws the following error:

julia> sind.(d)
ERROR: InvalidIRError: compiling MethodInstance for (::GPUArrays.var"#34#36")(::oneAPI.oneKernelContext, ::oneDeviceVector{…}, ::Base.Broadcast.Broadcasted{…}, ::Int64) resulted in invalid LLVM IR
Reason: unsupported call to an unknown function (call to gpu_malloc)
Stacktrace:
  [1] malloc
    @ ~/.julia/packages/GPUCompiler/Y4hSX/src/runtime.jl:88
  [2] macro expansion
    @ ~/.julia/packages/GPUCompiler/Y4hSX/src/runtime.jl:183
  [3] macro expansion
    @ ./none:0
  [4] box
    @ ./none:0
  [5] box_float32
    @ ~/.julia/packages/GPUCompiler/Y4hSX/src/runtime.jl:212
  [6] sind
    @ ./special/trig.jl:1183
  [7] _broadcast_getindex_evalf
    @ ./broadcast.jl:709
  [8] _broadcast_getindex
    @ ./broadcast.jl:682
  [9] getindex
    @ ./broadcast.jl:636
 [10] #34
    @ ~/.julia/packages/GPUArrays/qt4ax/src/host/broadcast.jl:59
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/Y4hSX/src/validation.jl:147
  [2] macro expansion
    @ ~/.julia/packages/GPUCompiler/Y4hSX/src/driver.jl:458 [inlined]
  [3] macro expansion
    @ ~/.julia/packages/TimerOutputs/Lw5SP/src/TimerOutput.jl:253 [inlined]
  [4] macro expansion
    @ ~/.julia/packages/GPUCompiler/Y4hSX/src/driver.jl:457 [inlined]
  [5] 
    @ GPUCompiler ~/.julia/packages/GPUCompiler/Y4hSX/src/utils.jl:103
  [6] emit_llvm
    @ ~/.julia/packages/GPUCompiler/Y4hSX/src/utils.jl:97 [inlined]
  [7] 
    @ GPUCompiler ~/.julia/packages/GPUCompiler/Y4hSX/src/driver.jl:136
  [8] codegen
    @ ~/.julia/packages/GPUCompiler/Y4hSX/src/driver.jl:115 [inlined]
  [9] 
    @ GPUCompiler ~/.julia/packages/GPUCompiler/Y4hSX/src/driver.jl:111
 [10] compile
    @ ~/.julia/packages/GPUCompiler/Y4hSX/src/driver.jl:103 [inlined]
 [11] #58
    @ ~/.julia/packages/oneAPI/z4Axk/src/compiler/compilation.jl:81 [inlined]
 [12] JuliaContext(f::oneAPI.var"#58#59"{GPUCompiler.CompilerJob{…}}; kwargs::@Kwargs{})
    @ GPUCompiler ~/.julia/packages/GPUCompiler/Y4hSX/src/driver.jl:52
 [13] JuliaContext(f::Function)
    @ GPUCompiler ~/.julia/packages/GPUCompiler/Y4hSX/src/driver.jl:42
 [14] compile(job::GPUCompiler.CompilerJob)
    @ oneAPI ~/.julia/packages/oneAPI/z4Axk/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/Y4hSX/src/execution.jl:237
 [16] cached_compilation(cache::Dict{…}, src::Core.MethodInstance, cfg::GPUCompiler.CompilerConfig{…}, compiler::Function, linker::Function)
    @ GPUCompiler ~/.julia/packages/GPUCompiler/Y4hSX/src/execution.jl:151
 [17] macro expansion
    @ ~/.julia/packages/oneAPI/z4Axk/src/compiler/execution.jl:203 [inlined]
 [18] macro expansion
    @ ./lock.jl:267 [inlined]
 [19] zefunction(f::GPUArrays.var"#34#36", tt::Type{Tuple{…}}; kwargs::@Kwargs{})
    @ oneAPI ~/.julia/packages/oneAPI/z4Axk/src/compiler/execution.jl:198
 [20] zefunction
    @ ~/.julia/packages/oneAPI/z4Axk/src/compiler/execution.jl:195 [inlined]
 [21] macro expansion
    @ ~/.julia/packages/oneAPI/z4Axk/src/compiler/execution.jl:66 [inlined]
 [22] #launch_heuristic#93
    @ ~/.julia/packages/oneAPI/z4Axk/src/gpuarrays.jl:17 [inlined]
 [23] launch_heuristic
    @ ~/.julia/packages/oneAPI/z4Axk/src/gpuarrays.jl:15 [inlined]
 [24] _copyto!
    @ ~/.julia/packages/GPUArrays/qt4ax/src/host/broadcast.jl:78 [inlined]
 [25] copyto!
    @ ~/.julia/packages/GPUArrays/qt4ax/src/host/broadcast.jl:44 [inlined]
 [26] copy
    @ ~/.julia/packages/GPUArrays/qt4ax/src/host/broadcast.jl:29 [inlined]
 [27] materialize(bc::Base.Broadcast.Broadcasted{oneAPI.oneArrayStyle{…}, Nothing, typeof(sind), Tuple{…}})
    @ Base.Broadcast ./broadcast.jl:903
 [28] top-level scope
    @ REPL[10]:1
Some type information was truncated. Use `show(err)` to see complete types.

The same happens with cosd.
I do not know if it also happens with AMD and/or Metal.jl, I am not able to test it.
Thank you!

[maleadt]

Yeah, this is somewhat known: #65
We currently do not support device-side exceptions, when they require allocations. We ought to implement a bump pointer allocator or something, but that hasn't happened.

So I think we can close this in favor of #65?