make use of Event(f, args)

Author: vchuravy

src/backends/cpu.jl

@@ -39,13 +39,11 @@ function wait(::CPU, ev::CPUEvent, progress=nothing)
     end
 end
 
-function async_copy!(::CPU, A, B; dependencies=nothing)
-    wait(CPU(), MultiEvent(dependencies), yield)
-    copyto!(A, B)
-    return CPUEvent(nothing)
+function async_copy!(::CPU, A, B; dependencies=nothing, progress=yield)
+    Event(copyto!, A, B, dependencies=dependencies, progress=progress)
 end
 
-function (obj::Kernel{CPU})(args...; ndrange=nothing, workgroupsize=nothing, dependencies=nothing)
+function (obj::Kernel{CPU})(args...; ndrange=nothing, workgroupsize=nothing, dependencies=nothing, progress=yield)
     if ndrange isa Integer
         ndrange = (ndrange,)
     end
@@ -65,13 +63,12 @@ function (obj::Kernel{CPU})(args...; ndrange=nothing, workgroupsize=nothing, dep
         ndrange = nothing
     end
 
-    t = Threads.@spawn __run(obj, ndrange, iterspace, args, dependencies, Val(dynamic))
-    return CPUEvent(t)
+    Event(__run, obj, ndrange, iterspace, args, Val(dynamic),
+          dependencies=dependencies, progress=progress)
 end
 
 # Inference barriers
-function __run(obj, ndrange, iterspace, args, dependencies, ::Val{dynamic}) where dynamic
-    wait(CPU(), MultiEvent(dependencies), yield)
+function __run(obj, ndrange, iterspace, args, ::Val{dynamic}) where dynamic
     N = length(iterspace)
     Nthreads = Threads.nthreads()
     if Nthreads == 1

src/backends/cuda.jl

@@ -109,12 +109,12 @@ function __pin!(a)
     return nothing
 end
 
-function async_copy!(::CUDA, A, B; dependencies=nothing)
+function async_copy!(::CUDA, A, B; dependencies=nothing, progress=yield)
     A isa Array && __pin!(A)
     B isa Array && __pin!(B)
 
     stream = next_stream()
-    wait(CUDA(), MultiEvent(dependencies), yield, stream)
+    wait(CUDA(), MultiEvent(dependencies), progress, stream)
     event = CuEvent(CUDAdrv.EVENT_DISABLE_TIMING)
     GC.@preserve A B begin
         destptr = pointer(A)
@@ -133,7 +133,7 @@ end
 ###
 # Kernel launch
 ###
-function (obj::Kernel{CUDA})(args...; ndrange=nothing, dependencies=nothing, workgroupsize=nothing)
+function (obj::Kernel{CUDA})(args...; ndrange=nothing, dependencies=nothing, workgroupsize=nothing, progress=yield)
     if ndrange isa Integer
         ndrange = (ndrange,)
     end
@@ -142,7 +142,7 @@ function (obj::Kernel{CUDA})(args...; ndrange=nothing, dependencies=nothing, wor
     end
 
     stream = next_stream()
-    wait(CUDA(), MultiEvent(dependencies), yield, stream)
+    wait(CUDA(), MultiEvent(dependencies), progress, stream)
 
     if KernelAbstractions.workgroupsize(obj) <: DynamicSize && workgroupsize === nothing
         # TODO: allow for NDRange{1, DynamicSize, DynamicSize}(nothing, nothing)