mapreduce over unit ranges on the GPU #33
Description
I was hoping that the following would work
julia> AcceleratedKernels.mapreduce(_->rand(), +, 1:n, ROCBackend(); init=0.0)/n
E.g. using a unit-range and the backend. (Note to self: define get_backend(1:N) = error("") to avoid a stack overflow.)
Of course, this works, but requires an allocation
julia> AcceleratedKernels.mapreduce(_->rand(), +, ROCArray(1:n), ROCBackend(); init=0.0)/n
The error is:
ERROR: GPU compilation of MethodInstance for AcceleratedKernels.gpu__mapreduce_block!(::KernelAbstractions.CompilerMetadata{…}, ::UnitRange{…}, ::SubArray{…}, ::var"#7#8", ::typeof(+), ::Int64) failed
KernelError: passing non-bitstype argument
Argument 4 to your kernel function is of type SubArray{Float64, 1, Vector{Float64}, Tuple{UnitRange{Int64}}, true}, which is not a bitstype:
.parent is of type Vector{Float64} which is not isbits.
.ref is of type MemoryRef{Float64} which is not isbits.
.mem is of type Memory{Float64} which is not isbits.
Only bitstypes, which are "plain data" types that are immutable
and contain no references to other values, can be used in GPU kernels.
For more information, see the `Base.isbitstype` function.
Stacktrace:
[1] check_invocation(job::GPUCompiler.CompilerJob)
@ GPUCompiler ~/.julia/packages/GPUCompiler/3QaEh/src/validation.jl:108
[2] macro expansion
@ ~/.julia/packages/GPUCompiler/3QaEh/src/driver.jl:87 [inlined]
[3] macro expansion
@ ~/.julia/packages/Tracy/GcShf/src/tracepoint.jl:158 [inlined]
[4] compile_unhooked(output::Symbol, job::GPUCompiler.CompilerJob; kwargs::@Kwargs{})
@ GPUCompiler ~/.julia/packages/GPUCompiler/3QaEh/src/driver.jl:85
[5] compile_unhooked
@ ~/.julia/packages/GPUCompiler/3QaEh/src/driver.jl:80 [inlined]
[6] compile(target::Symbol, job::GPUCompiler.CompilerJob; kwargs::@Kwargs{})
@ GPUCompiler ~/.julia/packages/GPUCompiler/3QaEh/src/driver.jl:67
[7] compile
@ ~/.julia/packages/GPUCompiler/3QaEh/src/driver.jl:55 [inlined]
[8] #40
@ ~/.julia/packages/AMDGPU/STpZC/src/compiler/codegen.jl:194 [inlined]
[9] JuliaContext(f::AMDGPU.Compiler.var"#40#41"{GPUCompiler.CompilerJob{GPUCompiler.GCNCompilerTarget, AMDGPU.Compiler.HIPCompilerParams}}; kwargs::@Kwargs{})
@ GPUCompiler ~/.julia/packages/GPUCompiler/3QaEh/src/driver.jl:34
[10] JuliaContext(f::Function)
@ GPUCompiler ~/.julia/packages/GPUCompiler/3QaEh/src/driver.jl:25
[11] hipcompile(job::GPUCompiler.CompilerJob)
@ AMDGPU.Compiler ~/.julia/packages/AMDGPU/STpZC/src/compiler/codegen.jl:193
[12] actual_compilation(cache::Dict{…}, src::Core.MethodInstance, world::UInt64, cfg::GPUCompiler.CompilerConfig{…}, compiler::typeof(AMDGPU.Compiler.hipcompile), linker::typeof(AMDGPU.Compiler.hiplink))
@ GPUCompiler ~/.julia/packages/GPUCompiler/3QaEh/src/execution.jl:245
[13] cached_compilation(cache::Dict{…}, src::Core.MethodInstance, cfg::GPUCompiler.CompilerConfig{…}, compiler::Function, linker::Function)
@ GPUCompiler ~/.julia/packages/GPUCompiler/3QaEh/src/execution.jl:159
[14] macro expansion
@ ~/.julia/packages/AMDGPU/STpZC/src/compiler/codegen.jl:161 [inlined]
[15] macro expansion
@ ./lock.jl:273 [inlined]
[16] hipfunction(f::typeof(AcceleratedKernels.gpu__mapreduce_block!), tt::Type{Tuple{KernelAbstractions.CompilerMetadata{…}, UnitRange{…}, SubArray{…}, var"#7#8", typeof(+), Int64}}; kwargs::@Kwargs{})
@ AMDGPU.Compiler ~/.julia/packages/AMDGPU/STpZC/src/compiler/codegen.jl:155
[17] hipfunction(f::typeof(AcceleratedKernels.gpu__mapreduce_block!), tt::Type{Tuple{KernelAbstractions.CompilerMetadata{…}, UnitRange{…}, SubArray{…}, var"#7#8", typeof(+), Int64}})
@ AMDGPU.Compiler ~/.julia/packages/AMDGPU/STpZC/src/compiler/codegen.jl:154
[18] macro expansion
@ ~/.julia/packages/AMDGPU/STpZC/src/highlevel.jl:155 [inlined]
[19] (::KernelAbstractions.Kernel{…})(::UnitRange{…}, ::Vararg{…}; ndrange::Tuple{…}, workgroupsize::Nothing)
@ AMDGPU.ROCKernels ~/.julia/packages/AMDGPU/STpZC/src/ROCKernels.jl:91
[20] mapreduce_1d(f::var"#7#8", op::typeof(+), src::UnitRange{Int64}, backend::ROCBackend; init::Float64, neutral::Int64, block_size::Int64, temp::Nothing, switch_below::Int64)
@ AcceleratedKernels ~/.julia/packages/AcceleratedKernels/Hf8B2/src/reduce/mapreduce_1d.jl:143
[21] mapreduce_1d
@ ~/.julia/packages/AcceleratedKernels/Hf8B2/src/reduce/mapreduce_1d.jl:102 [inlined]
[22] #_mapreduce_impl#48
@ ~/.julia/packages/AcceleratedKernels/Hf8B2/src/reduce/reduce.jl:252 [inlined]
[23] _mapreduce_impl
@ ~/.julia/packages/AcceleratedKernels/Hf8B2/src/reduce/reduce.jl:234 [inlined]
[24] mapreduce(f::var"#7#8", op::typeof(+), src::UnitRange{…}, backend::ROCBackend; init::Float64, neutral::Int64, dims::Nothing, scheduler::Symbol, max_tasks::Int64, min_elems::Int64, block_size::Int64, temp::Nothing, switch_below::Int64)
@ AcceleratedKernels ~/.julia/packages/AcceleratedKernels/Hf8B2/src/reduce/reduce.jl:219
[25] top-level scope
@ REPL[10]:1
Some type information was truncated. Use `show(err)` to see complete types.