[SYCL] E2E test for Level Zero device-scope events

SYCL/Plugin/level_zero_device_scope_events.cpp

@@ -0,0 +1,40 @@
+// REQUIRES: gpu, level_zero
+
+// RUN: %clangxx -fsycl -fsycl-targets=%sycl_triple  %s -o %t.out
+// RUN: env SYCL_PI_LEVEL_ZERO_DEVICE_SCOPE_EVENTS=1 SYCL_PI_TRACE=2 ZE_DEBUG=1 %GPU_RUN_PLACEHOLDER %t.out 2>&1 %GPU_CHECK_PLACEHOLDER
+
+// Checks that with L0 device-scope events enabled the only host-visible L0
+// event created is at the end of all kernels submission, when host waits for
+// the last kernel's event.
+//
+// CHECK-LABEL: Submitted all kernels
+// CHECK: ---> piEventsWait(
+// CHECK-NEXT:        <unknown> : 1
+// CHECK: ZE ---> zeEventCreate(ZeEventPool, &ZeEventDesc, &ZeHostVisibleEvent)
+// CHECK: ZE ---> zeCommandListAppendWaitOnEvents(CommandList->first, 1, &ZeEvent)
+// CHECK-NEXT: ZE ---> zeCommandListAppendSignalEvent(CommandList->first,
+// ZeHostVisibleEvent)
+// CHECK: Completed all kernels
+
+#include <CL/sycl.hpp>
+
+int main(int argc, char **argv) {
+  cl::sycl::gpu_selector device_selector;
+  cl::sycl::queue queue(device_selector);
+
+  int N = (argc >= 2 ? std::atoi(argv[1]) : 100);
+  std::cout << N << " kernels" << std::endl;
+
+  cl::sycl::event e; // completed event
+  for (int i = 0; i < N; i++) {
+    e = queue.submit([&](cl::sycl::handler &h) {
+      h.depends_on(e);
+      h.single_task<class kernel>([=] {});
+    });
+  } // for
+
+  std::cout << "Submitted all kernels" << std::endl;
+  e.wait(); // Waits for the last kernel to complete.
+  std::cout << "Completed all kernels" << std::endl;
+  return 0;
+}