[UR][L0] Correctly wait on barrier on urEnqueueEventsWaitWithBarrier (#11541)

When event list is null, a barrier is still needed for all previous
commands, so fix it.

---------

Signed-off-by: Jaime Arteaga <[email protected]>
Co-authored-by: Kenneth Benzie (Benie) <[email protected]>

Author: Jaime Arteaga

sycl/plugins/unified_runtime/CMakeLists.txt

@@ -57,13 +57,13 @@ if(SYCL_PI_UR_USE_FETCH_CONTENT)
   include(FetchContent)
 
   set(UNIFIED_RUNTIME_REPO "https://github.com/oneapi-src/unified-runtime.git")
-  # commit ce4acbc4e479c3e8c591f345f7ba30345a8a2a40
-  # Merge: 76aaf05c 28590a82
+  # commit e69ed21468e04ed6e832accf162422ed11736446
+  # Merge: 20fa0b5f 7fd9dafd
   # Author: Kenneth Benzie (Benie) <[email protected]>
-  # Date:   Wed Dec 6 17:13:51 2023 +0000
-  #     Merge pull request #1099 from jandres742/largeallocations
-  #     [UR][L0] Unify use of large allocation in L0 adapter
-  set(UNIFIED_RUNTIME_TAG ce4acbc4e479c3e8c591f345f7ba30345a8a2a40)
+  # Date:   Fri Dec 8 12:18:51 2023 +0000
+  #     Merge pull request #962 from jandres742/fixwaitbarrierwithevent
+  #     [UR][L0] Correctly wait on barrier on urEnqueueEventsWaitWithBarrier
+  set(UNIFIED_RUNTIME_TAG e69ed21468e04ed6e832accf162422ed11736446)
 
   if(SYCL_PI_UR_OVERRIDE_FETCH_CONTENT_REPO)
     set(UNIFIED_RUNTIME_REPO "${SYCL_PI_UR_OVERRIDE_FETCH_CONTENT_REPO}")

sycl/test-e2e/Plugin/level_zero_barrier_optimization.cpp

@@ -10,6 +10,8 @@
 int main() {
   sycl::queue Q1({sycl::property::queue::in_order{}});
   sycl::queue Q2({sycl::property::queue::in_order{}});
+  sycl::queue Q3({sycl::property::queue::in_order{},
+                  sycl::property::queue::enable_profiling{}});
 
   // Test case 1 - events in the barrier's waitlist are from different queues.
   std::cout << "Test1" << std::endl;
@@ -126,5 +128,30 @@ int main() {
   assert(Event11.get_info<sycl::info::event::command_execution_status>() ==
          sycl::info::event_command_status::complete);
 
+  // Test case 6 - submit barrier after queue sync with profiling enabled, i.e.
+  // last event = nullptr.
+  std::cout << "Test3" << std::endl;
+  auto Event12 = Q3.submit(
+      [&](sycl::handler &cgh) { cgh.single_task<class kernel12>([]() {}); });
+  auto Event13 = Q3.submit(
+      [&](sycl::handler &cgh) { cgh.single_task<class kernel13>([]() {}); });
+  Q3.wait();
+
+  // CHECK: Test3
+  // CHECK: ---> piEnqueueEventsWaitWithBarrier(
+  // CHECK: ZE ---> zeEventCreate
+  // CHECK-NOT: ZE ---> zeCommandListAppendWaitOnEvents
+  // CHECK-NOT: ZE ---> zeCommandListAppendSignalEvent
+  // CHECK: ZE ---> zeCommandListAppendBarrier
+  // CHECK: ) ---> 	pi_result : PI_SUCCESS
+  auto BarrierEvent6 = Q3.ext_oneapi_submit_barrier({Event12, Event13});
+  BarrierEvent6.wait();
+
+  // Check that kernel events are completed after waiting for barrier event.
+  assert(Event12.get_info<sycl::info::event::command_execution_status>() ==
+         sycl::info::event_command_status::complete);
+  assert(Event13.get_info<sycl::info::event::command_execution_status>() ==
+         sycl::info::event_command_status::complete);
+
   return 0;
 }