updating ctx for launch_heuristic

Author: leios

src/gpuarrays.jl

@@ -10,11 +10,15 @@ import KernelAbstractions: Backend
 
 ## execution
 
-struct oneArrayBackend <: Backend end
-
-@inline function GPUArrays.launch_heuristic(::oneArrayBackend, f::F, args::Vararg{Any,N};
+@inline function GPUArrays.launch_heuristic(::oneAPIBackend, f::F, args::Vararg{Any,N};
                                              elements::Int, elements_per_thread::Int) where {F,N}
-    kernel = @oneapi launch=false f(oneKernelContext(), args...)
+    ndrange, workgroupsize, iterspace, dynamic = KA.launch_config(obj, nothing,
+                                                                  nothing)
+
+    # this might not be the final context, since we may tune the workgroupsize
+    ctx = KA.mkcontext(obj, ndrange, iterspace)
+
+    kernel = @oneapi launch=false f(ctx, args...)
 
     items = launch_configuration(kernel)
     # XXX: how many groups is a good number? the API doesn't tell us.

src/oneAPI.jl

@@ -67,14 +67,14 @@ end
 # integrations and specialized functionality
 include("broadcast.jl")
 include("mapreduce.jl")
-include("gpuarrays.jl")
-include("random.jl")
-include("utils.jl")
-
 include("oneAPIKernels.jl")
 import .oneAPIKernels: oneAPIBackend
 export oneAPIBackend
 
+include("gpuarrays.jl")
+include("random.jl")
+include("utils.jl")
+
 function __init__()
     precompiling = ccall(:jl_generating_output, Cint, ()) != 0
     precompiling && return