StaticArrays?

[rafaqz]

I have a strange recurring bug with static arrays where I am getting a bounds error running
accessing an SMatrix on an immutable struct using a cuda gpu, where it runs fine on the CPU.

What is strange is that it was initially broken, then worked for a few weeks on the GPU and now doesn't again after updating packages, or something.

[8] #19 at /home/raf/.julia/packages/GPUArrays/3E1qk/src/broadcast.jl:15
Reason: unsupported call to the Julia runtime (jl_bounds_error_unboxed_int)
Stacktrace:
 [1] getindex at tuple.jl:24
 [2] getindex at /home/raf/.julia/packages/StaticArrays/Ze5H3/src/SMatrix.jl:76
 [3] macro expansion at /home/raf/.julia/packages/StaticArrays/Ze5H3/src/indexing.jl:28
 [4] _getindex_scalar at /home/raf/.julia/packages/StaticArrays/Ze5H3/src/indexing.jl:16
 [5] getindex at /home/raf/.julia/packages/StaticArrays/Ze5H3/src/indexing.jl:12

If the compiler can infer the indices it works fine, but otherwise not at all, even
if you bound them to well within the matrix.

[SimonDanisch]

Minimal working example please :)

[rafaqz]

Knew that was coming as soon as I posted :)

struct Test{S}
    matrix::S
end
test = Test(SMatrix{5,5}(ones(Int64, 5,5)))
a = CuArray(zeros(5))

inferit(a, x, y, test) = test.matrix[1, 5]  
broadcast(inferit, a, (1,2,3,4,5), (1,2,3,4,5), (test,))

getit(a, x, y, test) = test.matrix[y, x]  
broadcast(getit, a, (1,2,3,4,5), (1,2,3,4,5), (test,))

inferit works, get it errors with:

julia> broadcast(getit, a, (1,2,3,4,5), (1,2,3,4,5), (test,))                                      
ERROR: GPU compilation failed, try inspecting generated code with any of the @device_code_... macros
InvalidIRError: compiling #19(CuArrays.CuKernelState, CuDeviceArray{Int64,1,CUDAnative.AS.Global}, Base.Broadcast.Broadcasted{Nothing,Tuple{Base.OneTo{Int64}},typeof(getit),Tuple{Base.Broadcast.Extruded{CuDeviceArray{Float64,1,CUDAnative.AS.Global},Tuple{Bool},Tuple{Int64}},NTuple{5,Int64},NTuple{5,Int64},Tuple{Test{SArray{Tuple{5,5},Int64,2,25}}}}}) resulted in invalid LLVM IR
Reason: unsupported call to the Julia runtime (jl_bounds_error_unboxed_int)
Stacktrace:
 [1] getindex at tuple.jl:24
 [2] getindex at /home/raf/.julia/packages/StaticArrays/Ze5H3/src/SMatrix.jl:76
 [3] macro expansion at /home/raf/.julia/packages/StaticArrays/Ze5H3/src/indexing.jl:28
 [4] _getindex_scalar at /home/raf/.julia/packages/StaticArrays/Ze5H3/src/indexing.jl:16
 [5] getindex at /home/raf/.julia/packages/StaticArrays/Ze5H3/src/indexing.jl:12
 [6] getit at REPL[38]:1
 [7] _broadcast_getindex_evalf at broadcast.jl:574
 [8] _broadcast_getindex at broadcast.jl:547
 [9] getindex at broadcast.jl:507
 [10] macro expansion at /home/raf/.julia/packages/GPUArrays/3E1qk/src/abstract_gpu_interface.jl:69
 [11] #19 at /home/raf/.julia/packages/GPUArrays/3E1qk/src/broadcast.jl:15
Stacktrace:
 [1] check_ir at /home/raf/.julia/packages/CUDAnative/FzmMm/src/validation.jl:131 [inlined]        
 [2] macro expansion at ./logging.jl:322 [inlined]                                                 
 [3] #compile_function#69(::Bool, ::Function, ::CUDAnative.CompilerContext) at /home/raf/.julia/packages/CUDAnative/FzmMm/src/compiler.jl:651
 [4] compile_function at /home/raf/.julia/packages/CUDAnative/FzmMm/src/compiler.jl:651 [inlined]  
 [5] #cufunction#70(::Base.Iterators.Pairs{Symbol,getfield(GPUArrays, Symbol("##19#20")),Tuple{Symbol},NamedTuple{(:inner_f,),Tuple{getfield(GPUArrays, Symbol("##19#20"))}}}, ::Function, ::CUDAdrv.CuDevice, ::Any, ::Any) at /home/raf/.julia/packages/CUDAnative/FzmMm/src/compiler.jl:714             
 [6] (::getfield(CUDAnative, Symbol("#kw##cufunction")))(::NamedTuple{(:inner_f,),Tuple{getfield(GPUArrays, Symbol("##19#20"))}}, ::typeof(cufunction), ::CUDAdrv.CuDevice, ::Function, ::Type) at ./none:0                                                                                                
 [7] macro expansion at /home/raf/.julia/packages/CUDAnative/FzmMm/src/execution.jl:219 [inlined]  
 [8] _cuda(::CUDAnative.KernelWrapper{getfield(GPUArrays, Symbol("##19#20"))}, ::getfield(GPUArrays, Symbol("##19#20")), ::Tuple{}, ::NamedTuple{(:blocks, :threads),Tuple{Tuple{Int64},Tuple{Int64}}}, 
::CuArrays.CuKernelState, ::CuDeviceArray{Int64,1,CUDAnative.AS.Global}, ::Base.Broadcast.Broadcasted{Nothing,Tuple{Base.OneTo{Int64}},typeof(getit),Tuple{Base.Broadcast.Extruded{CuDeviceArray{Float64
,1,CUDAnative.AS.Global},Tuple{Bool},Tuple{Int64}},NTuple{5,Int64},NTuple{5,Int64},Tuple{Test{SArray{Tuple{5,5},Int64,2,25}}}}}) at /home/raf/.julia/packages/CUDAnative/FzmMm/src/execution.jl:177    
 [9] _gpu_call(::Function, ::CuArray{Int64,1}, ::Tuple{CuArray{Int64,1},Base.Broadcast.Broadcasted{Nothing,Tuple{Base.OneTo{Int64}},typeof(getit),Tuple{Base.Broadcast.Extruded{CuArray{Float64,1},Tuple{Bool},Tuple{Int64}},NTuple{5,Int64},NTuple{5,Int64},Tuple{Test{SArray{Tuple{5,5},Int64,2,25}}}}}}, ::Tuple{Tuple{Int64},Tuple{Int64}}) at ./gcutils.jl:87                                             
 [10] gpu_call(::Function, ::CuArray{Int64,1}, ::Tuple{CuArray{Int64,1},Base.Broadcast.Broadcasted{Nothing,Tuple{Base.OneTo{Int64}},typeof(getit),Tuple{Base.Broadcast.Extruded{CuArray{Float64,1},Tuple{Bool},Tuple{Int64}},NTuple{5,Int64},NTuple{5,Int64},Tuple{Test{SArray{Tuple{5,5},Int64,2,25}}}}}}, ::Int64) at /home/raf/.julia/packages/GPUArrays/3E1qk/src/abstract_gpu_interface.jl:151            
 [11] gpu_call at /home/raf/.julia/packages/GPUArrays/3E1qk/src/abstract_gpu_interface.jl:128 [inlined]                                                                                                
 [12] copyto! at /home/raf/.julia/packages/GPUArrays/3E1qk/src/broadcast.jl:14 [inlined]           
 [13] copyto! at ./broadcast.jl:768 [inlined]                                                      
 [14] copy at ./broadcast.jl:744 [inlined]                                                         
 [15] materialize(::Base.Broadcast.Broadcasted{Base.Broadcast.ArrayStyle{CuArray},Nothing,typeof(getit),Tuple{CuArray{Float64,1},NTuple{5,Int64},NTuple{5,Int64},Tuple{Test{SArray{Tuple{5,5},Int64,2,25}}}}}) at ./broadcast.jl:724                                                                       
 [16] broadcast(::typeof(getit), ::CuArray{Float64,1}, ::NTuple{5,Int64}, ::NTuple{5,Int64}, ::Vararg{Any,N} where N) at ./broadcast.jl:702                                                            
 [17] top-level scope at none:0                                                                    

[SimonDanisch]

did you try putting @inbounds into inferit and getit?

[rafaqz]

Like this? getit(a, x, y, test) = @inbounds test.matrix[y, x]

Doesn't seem to change anything...

[SimonDanisch]

like:
getit(a, x, y, test) = @inbounds return test.matrix[y, x]
since x = @inbounds bla; x == nothing
And inferit might need it too

[rafaqz]

inferit works fine however you do it! but that still doesn't help with getit (and I restarted julia just in case)

julia> inferit(a, x, y, test) = test.matrix[1, 5]  
inferit (generic function with 1 method)

julia> broadcast(inferit, a, (1,2,3,4,5), (1,2,3,4,5), (test,))
5-element CuArray{Int64,1}:
 1
 1
 1
 1
 1

julia> getit(a, x, y, test) = @inbounds return test.matrix[y, x]  
getit (generic function with 1 method)

julia> broadcast(getit, a, (1,2,3,4,5), (1,2,3,4,5), (test,))
ERROR: GPU compilation failed, try inspecting generated code with any of the @device_code_... macros
InvalidIRError: compiling #19(CuArrays.CuKernelState, CuDeviceArray{Int64,1,CUDAnative.AS.Global}, Base.Broadcast.Broadcasted{Nothing,Tuple{Base.OneTo{Int64}},typeof(getit),Tuple{Base.Broadcast.Extruded{CuDeviceArray{Float64,1,CUDAnative.AS.Global},Tuple{Bool},Tuple{Int64}},NTuple{5,Int64},NTuple{5,Int64},Tuple{Test{SArray{Tuple{5,5},Int64,2,25}}}}}) resulted in invalid LLVM IR
Reason: unsupported call to the Julia runtime (jl_bounds_error_unboxed_int)
Stacktrace:
 [1] getindex at tuple.jl:24
 [2] getindex at /home/raf/.julia/packages/StaticArrays/Ze5H3/src/SMatrix.jl:76
 [3] macro expansion at /home/raf/.julia/packages/StaticArrays/Ze5H3/src/indexing.jl:28
 [4] _getindex_scalar at /home/raf/.julia/packages/StaticArrays/Ze5H3/src/indexing.jl:16
 [5] getindex at /home/raf/.julia/packages/StaticArrays/Ze5H3/src/indexing.jl:12
 [6] getit at REPL[12]:1
 [7] _broadcast_getindex_evalf at broadcast.jl:574
 [8] _broadcast_getindex at broadcast.jl:547
 [9] getindex at broadcast.jl:507
 [10] macro expansion at /home/raf/.julia/packages/GPUArrays/3E1qk/src/abstract_gpu_interface.jl:69
 [11] #19 at /home/raf/.julia/packages/GPUArrays/3E1qk/src/broadcast.jl:15                         
Stacktrace:                                                                                        
 [1] check_ir at /home/raf/.julia/packages/CUDAnative/FzmMm/src/validation.jl:131 [inlined]        
 [2] macro expansion at ./logging.jl:322 [inlined]
 [3] #compile_function#69(::Bool, ::Function, ::CUDAnative.CompilerContext) at /home/raf/.julia/packages/CUDAnative/FzmMm/src/compiler.jl:651
 [4] compile_function at /home/raf/.julia/packages/CUDAnative/FzmMm/src/compiler.jl:651 [inlined]
 [5] #cufunction#70(::Base.Iterators.Pairs{Symbol,getfield(GPUArrays, Symbol("##19#20")),Tuple{Symbol},NamedTuple{(:inner_f,),Tuple{getfield(GPUArrays, Symbol("##19#20"))}}}, ::Function, ::CUDAdrv.CuDevice, ::Any, ::Any) at /home/raf/.julia/packages/CUDAnative/FzmMm/src/compiler.jl:714             
 [6] (::getfield(CUDAnative, Symbol("#kw##cufunction")))(::NamedTuple{(:inner_f,),Tuple{getfield(GPUArrays, Symbol("##19#20"))}}, ::typeof(cufunction), ::CUDAdrv.CuDevice, ::Function, ::Type) at ./none:0
 [7] macro expansion at /home/raf/.julia/packages/CUDAnative/FzmMm/src/execution.jl:219 [inlined]  
 [8] _cuda(::CUDAnative.KernelWrapper{getfield(GPUArrays, Symbol("##19#20"))}, ::getfield(GPUArrays, Symbol("##19#20")), ::Tuple{}, ::NamedTuple{(:blocks, :threads),Tuple{Tuple{Int64},Tuple{Int64}}}, 
::CuArrays.CuKernelState, ::CuDeviceArray{Int64,1,CUDAnative.AS.Global}, ::Base.Broadcast.Broadcasted{Nothing,Tuple{Base.OneTo{Int64}},typeof(getit),Tuple{Base.Broadcast.Extruded{CuDeviceArray{Float64,1,CUDAnative.AS.Global},Tuple{Bool},Tuple{Int64}},NTuple{5,Int64},NTuple{5,Int64},Tuple{Test{SArray{Tuple{5,5},Int64,2,25}}}}}) at /home/raf/.julia/packages/CUDAnative/FzmMm/src/execution.jl:177    
 [9] _gpu_call(::Function, ::CuArray{Int64,1}, ::Tuple{CuArray{Int64,1},Base.Broadcast.Broadcasted{Nothing,Tuple{Base.OneTo{Int64}},typeof(getit),Tuple{Base.Broadcast.Extruded{CuArray{Float64,1},Tuple
{Bool},Tuple{Int64}},NTuple{5,Int64},NTuple{5,Int64},Tuple{Test{SArray{Tuple{5,5},Int64,2,25}}}}}}, 
::Tuple{Tuple{Int64},Tuple{Int64}}) at ./gcutils.jl:87                                             
 [10] gpu_call(::Function, ::CuArray{Int64,1}, ::Tuple{CuArray{Int64,1},Base.Broadcast.Broadcasted{Nothing,Tuple{Base.OneTo{Int64}},typeof(getit),Tuple{Base.Broadcast.Extruded{CuArray{Float64,1},Tuple
{Bool},Tuple{Int64}},NTuple{5,Int64},NTuple{5,Int64},Tuple{Test{SArray{Tuple{5,5},Int64,2,25}}}}}}, 
::Int64) at /home/raf/.julia/packages/GPUArrays/3E1qk/src/abstract_gpu_interface.jl:151            
 [11] gpu_call at /home/raf/.julia/packages/GPUArrays/3E1qk/src/abstract_gpu_interface.jl:128 [inlined]                                                                                                
 [12] copyto! at /home/raf/.julia/packages/GPUArrays/3E1qk/src/broadcast.jl:14 [inlined]           
 [13] copyto! at ./broadcast.jl:768 [inlined]                                                      
 [14] copy at ./broadcast.jl:744 [inlined]                                                         
 [15] materialize(::Base.Broadcast.Broadcasted{Base.Broadcast.ArrayStyle{CuArray},Nothing,typeof(getit),Tuple{CuArray{Float64,1},NTuple{5,Int64},NTuple{5,Int64},Tuple{Test{SArray{Tuple{5,5},Int64,2,25
}}}}}) at ./broadcast.jl:724                                                                       
 [16] broadcast(::typeof(getit), ::CuArray{Float64,1}, ::NTuple{5,Int64}, ::NTuple{5,Int64}, ::Vararg{Any,N} where N) at ./broadcast.jl:702                                                            
 [17] top-level scope at none:0

[rafaqz]

somehow the compiler has access to the indices but the runtime doesn't, not that I understand whats going on at all.

[SimonDanisch]

https://github.com/JuliaArrays/StaticArrays.jl/blob/master/src/SMatrix.jl#L74
Is missing an Base.@propagate_inbounds

[rafaqz]

Thanks! I'll put in a pull request