[SYCL] Tests for events caching mode in the L0 plugin (intel#1121)

Author: againull

SYCL/Plugin/level_zero_events_caching.cpp

@@ -0,0 +1,27 @@
+// REQUIRES: gpu, level_zero
+// TODO: There is a known issue that ZE_DEBUG=4 produces flaky output on
+// Windows.
+// UNSUPPORTED: windows
+
+// RUN: %clangxx -fsycl -fsycl-targets=%sycl_triple %level_zero_options %s -o %t.out
+// RUN: env SYCL_PI_LEVEL_ZERO_DEVICE_SCOPE_EVENTS=0 ZE_DEBUG=4 %GPU_RUN_PLACEHOLDER %t.out 2>&1 | FileCheck --check-prefixes=CACHING-ENABLED %s
+// RUN: env SYCL_PI_LEVEL_ZERO_DEVICE_SCOPE_EVENTS=0 SYCL_PI_LEVEL_ZERO_DISABLE_EVENTS_CACHING=1 ZE_DEBUG=4 %GPU_RUN_PLACEHOLDER %t.out 2>&1 | FileCheck --check-prefixes=CACHING-DISABLED %s
+
+// CACHING-ENABLED: zeEventCreate = 1
+// CACHING-DISABLED: zeEventCreate = 256
+
+// Check event caching modes in the L0 plugin.
+
+#include <CL/sycl.hpp>
+
+int main() {
+  cl::sycl::queue deviceQueue;
+
+  for (int i = 0; i < 256; i++) {
+    auto Event = deviceQueue.submit([&](cl::sycl::handler &cgh) {
+      cgh.single_task<class SimpleKernel>([=]() {});
+    });
+    Event.wait();
+  }
+  return 0;
+}

SYCL/Plugin/level_zero_events_caching_leak.cpp

@@ -0,0 +1,41 @@
+// REQUIRES: gpu, level_zero
+// TODO: There is a known issue that ZE_DEBUG=4 produces flaky output on
+// Windows.
+// UNSUPPORTED: windows
+
+// RUN: %clangxx -fsycl -fsycl-targets=%sycl_triple %level_zero_options %s -o %t.out
+// RUN: env ZE_MAX_NUMBER_OF_EVENTS_PER_EVENT_POOL=4 ZE_DEBUG=4 %GPU_RUN_PLACEHOLDER %t.out
+// RUN: env SYCL_PI_LEVEL_ZERO_DISABLE_EVENTS_CACHING=1 ZE_MAX_NUMBER_OF_EVENTS_PER_EVENT_POOL=4 ZE_DEBUG=4 %GPU_RUN_PLACEHOLDER %t.out
+
+// Check that events and pools are not leaked when event caching is
+// enabled/disabled.
+
+#include <CL/sycl.hpp>
+#include <array>
+
+constexpr cl::sycl::access::mode sycl_read = cl::sycl::access::mode::read;
+constexpr cl::sycl::access::mode sycl_write = cl::sycl::access::mode::write;
+
+int main() {
+  cl::sycl::queue deviceQueue;
+
+  const size_t array_size = 4;
+  std::array<int, array_size> A = {{1, 2, 3, 4}}, B = {{1, 2, 3, 4}}, C;
+  cl::sycl::range<1> numOfItems{array_size};
+  cl::sycl::buffer<int, 1> bufferA(A.data(), numOfItems);
+  cl::sycl::buffer<int, 1> bufferB(B.data(), numOfItems);
+  cl::sycl::buffer<int, 1> bufferC(C.data(), numOfItems);
+
+  for (int i = 0; i < 256; i++) {
+    deviceQueue.submit([&](cl::sycl::handler &cgh) {
+      auto accessorA = bufferA.get_access<sycl_read>(cgh);
+      auto accessorB = bufferB.get_access<sycl_read>(cgh);
+      auto accessorC = bufferC.get_access<sycl_write>(cgh);
+
+      cgh.parallel_for<class SimpleVadd>(numOfItems, [=](cl::sycl::id<1> wiID) {
+        accessorC[wiID] = accessorA[wiID] + accessorB[wiID];
+      });
+    });
+  }
+  return 0;
+}