adding flaky CPU front-ent and tests

Author: leios

src/KernelAbstractions.jl

@@ -319,7 +319,7 @@ This is a unified atomic interface
 
 macro atomic(ex)
     quote
-        return $__atomic(ex)
+        $__atomic($ex)
     end
 end
 

src/cpu.jl

@@ -269,3 +269,11 @@ end
 
 # Argument conversion
 KernelAbstractions.argconvert(k::Kernel{CPU}, arg) = arg
+
+###
+# CPU implementation of atomic macro
+###
+
+@inline function Cassette.overdub(::CUDACtx, ::typeof(__atomic), ex)
+    println(ex)
+end

test/atomic_test.jl

@@ -0,0 +1,18 @@
+using KernelAbstractions, Test
+
+@kernel function kernel_atomics(input)
+  I = @index(Global)
+  KernelAbstractions.@atomic input[1] += 1
+end
+
+function atomics_testsuite(backend)
+
+    @testset "atomic test" begin
+        A = ArrayT{Int}(zeros(64))
+
+        kernel = kernel_atomic(backend(), 4)
+        wait(kernel(ndrange=(length(A),)))
+
+        @test A[1] == length(A)
+    end
+end

test/testsuite.jl

@@ -15,6 +15,7 @@ include("compiler.jl")
 include("reflection.jl")
 include("examples.jl")
 include("convert.jl")
+include("atomic_test.jl")
 
 function testsuite(backend, backend_str, backend_mod, AT, DAT)
     @testset "Unittests" begin
@@ -69,6 +70,12 @@ function testsuite(backend, backend_str, backend_mod, AT, DAT)
         convert_testsuite(backend, AT)
     end
 
+    if backend_str != "ROCM"
+        @testset "Atomics" begin
+            atomics_testsuite(backend, AT)
+        end
+    end
+
     if backend_str == "CUDA"
         @testset "Examples" begin
             examples_testsuite()