fix formatting

src/nditeration.jl

@@ -138,18 +138,14 @@ needs to perform dynamic bounds-checking.
 """
 @inline function partition(ndrange, __workgroupsize)
     @assert length(__workgroupsize) <= length(ndrange)
-    if length(__workgroupsize) < length(ndrange)
-        # pad workgroupsize with ones
-        workgroupsize = ntuple(Val(length(ndrange))) do I
-            Base.@_inline_meta
-            if I > length(__workgroupsize)
-                return 1
-            else
-                return __workgroupsize[I]
-            end
+    # pad workgroupsize with ones
+    workgroupsize = ntuple(Val(length(ndrange))) do I
+        Base.@_inline_meta
+        if I > length(__workgroupsize) || __workgroupsize[I] == 0
+            return 1
+        else
+            return __workgroupsize[I]
         end
-    else
-        workgroupsize = __workgroupsize
     end
     let workgroupsize = workgroupsize
         dynamic = Ref(false)

src/pocl/backend.jl

@@ -25,26 +25,26 @@ KA.allocate(::POCLBackend, ::Type{T}, dims::Tuple) where {T} = Array{T}(undef, d
 
 function KA.zeros(backend::POCLBackend, ::Type{T}, dims::Tuple) where {T}
     arr = KA.allocate(backend, T, dims)
-    kernel = init_kernel(backend)
+    kernel = KA.init_kernel(backend)
     kernel(arr, zero, T, ndrange = length(arr))
     return arr
 end
 function KA.ones(backend::POCLBackend, ::Type{T}, dims::Tuple) where {T}
     arr = KA.allocate(backend, T, dims)
-    kernel = init_kernel(backend)
+    kernel = KA.init_kernel(backend)
     kernel(arr, one, T; ndrange = length(arr))
     return arr
 end
 
 function KA.copyto!(backend::POCLBackend, A, B)
-    if get_backend(A) == get_backend(B) && get_backend(A) isa POCLBackend
+    if KA.get_backend(A) == KA.get_backend(B) && KA.get_backend(A) isa POCLBackend
         if length(A) != length(B)
             error("Arrays must match in length")
         end
         if Base.mightalias(A, B)
             error("Arrays may not alias")
         end
-        kernel = copy_kernel(backend)
+        kernel = KA.copy_kernel(backend)
         kernel(A, B, ndrange = length(A))
         return A
     else
@@ -131,9 +131,9 @@ function (obj::KA.Kernel{POCLBackend})(args...; ndrange = nothing, workgroupsize
     # Launch kernel
     global_size = groups * items
     local_size = items
-    kernel(ctx, args...; global_size, local_size)
-
-    cl.finish(cl.queue()) # TODO, would waiting on an event be cheaper?
+    event = kernel(ctx, args...; global_size, local_size)
+    wait(event)
+    cl.clReleaseEvent(event)
     return nothing
 end
 

src/pocl/compiler/compilation.jl

@@ -2,18 +2,20 @@
 
 struct OpenCLCompilerParams <: AbstractCompilerParams end
 const OpenCLCompilerConfig = CompilerConfig{SPIRVCompilerTarget, OpenCLCompilerParams}
-const OpenCLCompilerJob = CompilerJob{SPIRVCompilerTarget,OpenCLCompilerParams}
+const OpenCLCompilerJob = CompilerJob{SPIRVCompilerTarget, OpenCLCompilerParams}
 
-GPUCompiler.runtime_module(::CompilerJob{<:Any,OpenCLCompilerParams}) = POCL
+GPUCompiler.runtime_module(::CompilerJob{<:Any, OpenCLCompilerParams}) = POCL
 
 GPUCompiler.method_table(::OpenCLCompilerJob) = method_table
 
 # filter out OpenCL built-ins
 # TODO: eagerly lower these using the translator API
 GPUCompiler.isintrinsic(job::OpenCLCompilerJob, fn::String) =
-    invoke(GPUCompiler.isintrinsic,
-           Tuple{CompilerJob{SPIRVCompilerTarget}, typeof(fn)},
-           job, fn) ||
+    invoke(
+    GPUCompiler.isintrinsic,
+    Tuple{CompilerJob{SPIRVCompilerTarget}, typeof(fn)},
+    job, fn
+) ||
     in(fn, opencl_builtins)
 
 
@@ -42,14 +44,14 @@ function compiler_config(dev::cl.Device; kwargs...)
     end
     return config
 end
-@noinline function _compiler_config(dev; kernel=true, name=nothing, always_inline=false, kwargs...)
+@noinline function _compiler_config(dev; kernel = true, name = nothing, always_inline = false, kwargs...)
     supports_fp16 = "cl_khr_fp16" in dev.extensions
     supports_fp64 = "cl_khr_fp64" in dev.extensions
 
     # create GPUCompiler objects
     target = SPIRVCompilerTarget(; supports_fp16, supports_fp64, kwargs...)
     params = OpenCLCompilerParams()
-    CompilerConfig(target, params; kernel, name, always_inline)
+    return CompilerConfig(target, params; kernel, name, always_inline)
 end
 
 # compile to executable machine code
@@ -59,7 +61,7 @@ function compile(@nospecialize(job::CompilerJob))
         GPUCompiler.compile(:obj, job)
     end
 
-    (;obj, entry=LLVM.name(meta.entry))
+    return (; obj, entry = LLVM.name(meta.entry))
 end
 
 # link into an executable kernel
@@ -70,5 +72,5 @@ function link(@nospecialize(job::CompilerJob), compiled)
         error("Your device does not support SPIR-V, which is currently required for native execution.")
     end
     cl.build!(prog)
-    cl.Kernel(prog, compiled.entry)
+    return cl.Kernel(prog, compiled.entry)
 end

src/pocl/compiler/execution.jl

@@ -9,7 +9,7 @@ const LAUNCH_KWARGS = [:global_size, :local_size, :queue]
 
 macro opencl(ex...)
     call = ex[end]
-    kwargs = map(ex[1:end-1]) do kwarg
+    kwargs = map(ex[1:(end - 1)]) do kwarg
         if kwarg isa Symbol
             :($kwarg = $kwarg)
         elseif Meta.isexpr(kwarg, :(=))
@@ -31,14 +31,14 @@ macro opencl(ex...)
     macro_kwargs, compiler_kwargs, call_kwargs, other_kwargs =
         split_kwargs(kwargs, MACRO_KWARGS, COMPILER_KWARGS, LAUNCH_KWARGS)
     if !isempty(other_kwargs)
-        key,val = first(other_kwargs).args
+        key, val = first(other_kwargs).args
         throw(ArgumentError("Unsupported keyword argument '$key'"))
     end
 
     # handle keyword arguments that influence the macro's behavior
     launch = true
     for kwarg in macro_kwargs
-        key,val = kwarg.args
+        key, val = kwarg.args
         if key == :launch
             isa(val, Bool) || throw(ArgumentError("`launch` keyword argument to @opencl should be a constant value"))
             launch = val::Bool
@@ -56,7 +56,8 @@ macro opencl(ex...)
 
     # convert the arguments, call the compiler and launch the kernel
     # while keeping the original arguments alive
-    push!(code.args,
+    push!(
+        code.args,
         quote
             $f_var = $f
             GC.@preserve $(vars...) $f_var begin
@@ -69,13 +70,16 @@ macro opencl(ex...)
                 end
                 $kernel
             end
-         end)
+        end
+    )
 
-    return esc(quote
-        let
-            $code
+    return esc(
+        quote
+            let
+                $code
+            end
         end
-    end)
+    )
 end
 
 
@@ -101,21 +105,23 @@ end
 # Base.RefValue isn't GPU compatible, so provide a compatible alternative
 # TODO: port improvements from CUDA.jl
 struct CLRefValue{T} <: Ref{T}
-  x::T
+    x::T
 end
 Base.getindex(r::CLRefValue) = r.x
 Adapt.adapt_structure(to::KernelAdaptor, r::Base.RefValue) = CLRefValue(adapt(to, r[]))
 
 # broadcast sometimes passes a ref(type), resulting in a GPU-incompatible DataType box.
 # avoid that by using a special kind of ref that knows about the boxed type.
 struct CLRefType{T} <: Ref{DataType} end
-Base.getindex(r::CLRefType{T}) where T = T
-Adapt.adapt_structure(to::KernelAdaptor, r::Base.RefValue{<:Union{DataType,Type}}) =
+Base.getindex(r::CLRefType{T}) where {T} = T
+Adapt.adapt_structure(to::KernelAdaptor, r::Base.RefValue{<:Union{DataType, Type}}) =
     CLRefType{r[]}()
 
 # case where type is the function being broadcasted
-Adapt.adapt_structure(to::KernelAdaptor,
-                      bc::Broadcast.Broadcasted{Style, <:Any, Type{T}}) where {Style, T} =
+Adapt.adapt_structure(
+    to::KernelAdaptor,
+    bc::Broadcast.Broadcasted{Style, <:Any, Type{T}}
+) where {Style, T} =
     Broadcast.Broadcasted{Style}((x...) -> T(x...), adapt(to, bc.args), bc.axes)
 
 """
@@ -131,29 +137,30 @@ register methods for the the `OpenCL.KernelAdaptor` type.
 The `pointers` argument is used to collect pointers to indirect SVM buffers, which need to
 be registered with OpenCL before invoking the kernel.
 """
-function clconvert(arg, pointers::Vector{Ptr{Cvoid}}=Ptr{Cvoid}[])
-    adapt(KernelAdaptor(pointers), arg)
+function clconvert(arg, pointers::Vector{Ptr{Cvoid}} = Ptr{Cvoid}[])
+    return adapt(KernelAdaptor(pointers), arg)
 end
 
 
-
 ## abstract kernel functionality
 
-abstract type AbstractKernel{F,TT} end
+abstract type AbstractKernel{F, TT} end
 
-@inline @generated function (kernel::AbstractKernel{F,TT})(args...;
-                                                           call_kwargs...) where {F,TT}
+@inline @generated function (kernel::AbstractKernel{F, TT})(
+        args...;
+        call_kwargs...
+    ) where {F, TT}
     sig = Tuple{F, TT.parameters...}    # Base.signature_type with a function type
-    args = (:(kernel.f), (:( clconvert(args[$i], svm_pointers) ) for i in 1:length(args))...)
+    args = (:(kernel.f), (:(clconvert(args[$i], svm_pointers)) for i in 1:length(args))...)
 
     # filter out ghost arguments that shouldn't be passed
     predicate = dt -> GPUCompiler.isghosttype(dt) || Core.Compiler.isconstType(dt)
     to_pass = map(!predicate, sig.parameters)
-    call_t =                  Type[x[1] for x in zip(sig.parameters,  to_pass) if x[2]]
-    call_args = Union{Expr,Symbol}[x[1] for x in zip(args, to_pass)            if x[2]]
+    call_t = Type[x[1] for x in zip(sig.parameters, to_pass) if x[2]]
+    call_args = Union{Expr, Symbol}[x[1] for x in zip(args, to_pass)            if x[2]]
 
     # replace non-isbits arguments (they should be unused, or compilation would have failed)
-    for (i,dt) in enumerate(call_t)
+    for (i, dt) in enumerate(call_t)
         if !isbitstype(dt)
             call_t[i] = Ptr{Any}
             call_args[i] = :C_NULL
@@ -163,17 +170,16 @@ abstract type AbstractKernel{F,TT} end
     # finalize types
     call_tt = Base.to_tuple_type(call_t)
 
-    quote
+    return quote
         svm_pointers = Ptr{Cvoid}[]
         $cl.clcall(kernel.fun, $call_tt, $(call_args...); svm_pointers, call_kwargs...)
     end
 end
 
 
-
 ## host-side kernels
 
-struct HostKernel{F,TT} <: AbstractKernel{F,TT}
+struct HostKernel{F, TT} <: AbstractKernel{F, TT}
     f::F
     fun::cl.Kernel
 end
@@ -183,7 +189,7 @@ end
 
 const clfunction_lock = ReentrantLock()
 
-function clfunction(f::F, tt::TT=Tuple{}; kwargs...) where {F,TT}
+function clfunction(f::F, tt::TT = Tuple{}; kwargs...) where {F, TT}
     ctx = context()
     dev = device()
 
@@ -200,10 +206,10 @@ function clfunction(f::F, tt::TT=Tuple{}; kwargs...) where {F,TT}
         kernel = get(_kernel_instances, h, nothing)
         if kernel === nothing
             # create the kernel state object
-            kernel = HostKernel{F,tt}(f, fun)
+            kernel = HostKernel{F, tt}(f, fun)
             _kernel_instances[h] = kernel
         end
-        return kernel::HostKernel{F,tt}
+        return kernel::HostKernel{F, tt}
     end
 end
 

src/pocl/compiler/reflection.jl

@@ -24,27 +24,28 @@ for method in (:code_typed, :code_warntype, :code_llvm, :code_native)
     args = method == :code_typed ? (:job,) : (:io, :job)
 
     @eval begin
-        function $method(io::IO, @nospecialize(func), @nospecialize(types);
-                         kernel::Bool=false, kwargs...)
+        function $method(
+                io::IO, @nospecialize(func), @nospecialize(types);
+                kernel::Bool = false, kwargs...
+            )
             compiler_kwargs, kwargs = split_kwargs_runtime(kwargs, COMPILER_KWARGS)
             source = methodinstance(typeof(func), Base.to_tuple_type(types))
             config = compiler_config(device(); kernel, compiler_kwargs...)
             job = CompilerJob(source, config)
-            GPUCompiler.$method($(args...); kwargs...)
+            return GPUCompiler.$method($(args...); kwargs...)
         end
         $method(@nospecialize(func), @nospecialize(types); kwargs...) =
             $method(stdout, func, types; kwargs...)
     end
 end
 
 
-
 #
 # @device_code_* functions
 #
 
 export @device_code_lowered, @device_code_typed, @device_code_warntype, @device_code_llvm,
-       @device_code_native, @device_code
+    @device_code_native, @device_code
 
 # forward to GPUCompiler
 @eval $(Symbol("@device_code_lowered")) = $(getfield(GPUCompiler, Symbol("@device_code_lowered")))
@@ -70,5 +71,5 @@ function return_type(@nospecialize(func), @nospecialize(tt))
     job = CompilerJob(source, config)
     interp = GPUCompiler.get_interpreter(job)
     sig = Base.signature_type(func, tt)
-    Core.Compiler.return_type(interp, sig)
+    return Core.Compiler.return_type(interp, sig)
 end