[llvm] [offload] draft POC: verify event synchronization (PR #193927)

[via llvm-commits]

================
@@ -240,6 +241,45 @@ TEST_P(olLaunchKernelLocalMemStaticTest, Success) {
   ASSERT_SUCCESS(olMemFree(Mem));
 }
 
+// The test intends to verify the correctness of the current implementation of
+// the event synchronisation.
+TEST_P(olLaunchKernelSingleCounterSyncEventTest, Success) {
+  LaunchArgs.NumGroups.x = 1;
+  LaunchArgs.DynSharedMemory = 0;
+
+  void *ResNum;
+  ASSERT_SUCCESS(
+      olMemAlloc(Device, OL_ALLOC_TYPE_MANAGED, sizeof(uint32_t), &ResNum));
+
+  // The execution time of the provided kernel should be high enough to ensure
+  // that the read value of the final result is not correct without explicit
+  // synchronization: explicit waiting for the event following the submitted
+  // operation. LoopRange is arbitrarily set with the goal of prolonging the
+  // kernel execution through a high number of loop iterations. In each
+  // iteration, NumberToAdd is added to the final sum, which is stored in
+  // ResNum.
+  int32_t LoopRange = 1000000;
+  int32_t NumberToAdd = 2;
+
+  struct {
+    int32_t InitLoop;
+    int32_t Addend;
+    void *Out;
----------------
EuphoricThinking wrote:

Additionally, the kernel parameters have also changed.

https://github.com/llvm/llvm-project/pull/193927