feature: events deadlock detection in validation layer

Related-To: NEO-12810 Signed-off-by: Chandio, Bibrak Qamar <[email protected]>
oneapi-src · Nov 12, 2024 · abde7db · abde7db
1 parent 3091196
commit abde7db
Show file tree

Hide file tree

Showing 8 changed files with 13 additions and 744 deletions.
diff --git a/source/layers/validation/README.md b/source/layers/validation/README.md
@@ -18,7 +18,7 @@ By default, no validation modes will be enabled. The individual validation modes
 
 - `ZE_ENABLE_PARAMETER_VALIDATION`
 - `ZE_ENABLE_HANDLE_LIFETIME`
-- `ZEL_ENABLE_EVENTSDEADLOCK_CHECKER`
+- `ZEL_ENABLE_EVENTS_CHECKER`
 - `ZE_ENABLE_MEMORY_TRACKER` (Not yet Implemeneted)
 - `ZE_ENABLE_THREADING_VALIDATION` (Not yet Implemeneted)
 
@@ -49,9 +49,11 @@ This mode maintains an internal mapping of each handle type to a state structure
 - Additional per handle state checks added as needed
     - Example - Check ze_cmdlist_handle_t open or closed
 
-### `ZEL_ENABLE_EVENTSDEADLOCK_CHECKER`
+### `ZEL_ENABLE_EVENTS_CHECKER`
 
-The Events Deadlock Checker is designed to detect potential deadlocks that might occur due to improper event usage in the Level Zero API. It prints out wairning messages for user when it detects a potential deadlock.
+The Events Checker validates usage of events. 
+- It is designed to detect potential deadlocks that might occur due to improper event usage in the Level Zero API. It prints out warning messages for user when it detects a potential deadlock.
+- In some cases it may also detect whether an event is being used more than once without being reset. Consider a case in which a single event is signaled from twice.
 
 ### `ZE_ENABLE_THREADING_VALIDATION` (Not yet Implemeneted)
 

diff --git a/source/layers/validation/checkers/CMakeLists.txt b/source/layers/validation/checkers/CMakeLists.txt
@@ -1,3 +1,3 @@
-add_subdirectory(events_deadlock)
+add_subdirectory(events_checker)
 add_subdirectory(parameter_validation)
 add_subdirectory(template)
diff --git a/source/layers/validation/checkers/events_deadlock/CMakeLists.txt b/source/layers/validation/checkers/events_deadlock/CMakeLists.txt
diff --git a/source/layers/validation/checkers/events_deadlock/DESIGN.md b/source/layers/validation/checkers/events_deadlock/DESIGN.md
diff --git a/source/layers/validation/checkers/events_deadlock/zel_events_deadlock_checker.cpp b/source/layers/validation/checkers/events_deadlock/zel_events_deadlock_checker.cpp
diff --git a/source/layers/validation/checkers/events_deadlock/zel_events_deadlock_checker.h b/source/layers/validation/checkers/events_deadlock/zel_events_deadlock_checker.h
diff --git a/test/CMakeLists.txt b/test/CMakeLists.txt
@@ -40,8 +40,8 @@ set_property(TEST tests_both_gpu PROPERTY ENVIRONMENT "ZE_ENABLE_NULL_DRIVER=1")
 add_test(NAME tests_both_npu COMMAND tests --gtest_filter=*GivenLevelZeroLoaderPresentWhenCallingzeInitThenZeInitDriversThenBothCallsSucceedWithNPUTypes*)
 set_property(TEST tests_both_npu PROPERTY ENVIRONMENT "ZE_ENABLE_NULL_DRIVER=1")
 add_test(NAME tests_event_deadlock COMMAND tests --gtest_filter=*GivenLevelZeroLoaderPresentWhenCallingzeCommandListAppendMemoryCopyWithCircularDependencyOnEventsThenValidationLayerPrintsWarningOfDeadlock*)
-set_property(TEST tests_event_deadlock PROPERTY ENVIRONMENT "ZE_ENABLE_NULL_DRIVER=1;ZE_ENABLE_VALIDATION_LAYER=1;ZEL_ENABLE_EVENTSDEADLOCK_CHECKER=1")
+set_property(TEST tests_event_deadlock PROPERTY ENVIRONMENT "ZE_ENABLE_NULL_DRIVER=1;ZE_ENABLE_VALIDATION_LAYER=1;ZEL_ENABLE_EVENTS_CHECKER=1")
 add_test(NAME tests_event_deadlock_reset COMMAND tests --gtest_filter=*GivenLevelZeroLoaderPresentWhenCallingzeCommandListAppendMemoryCopyWithCircularDependencyOnEventsAndExplicitCallzeEventHostSignalThenValidationLayerPrintsWarningOfIllegalUsage*)
-set_property(TEST tests_event_deadlock_reset PROPERTY ENVIRONMENT "ZE_ENABLE_NULL_DRIVER=1;ZE_ENABLE_VALIDATION_LAYER=1;ZEL_ENABLE_EVENTSDEADLOCK_CHECKER=1")
+set_property(TEST tests_event_deadlock_reset PROPERTY ENVIRONMENT "ZE_ENABLE_NULL_DRIVER=1;ZE_ENABLE_VALIDATION_LAYER=1;ZEL_ENABLE_EVENTS_CHECKER=1")
 add_test(NAME tests_event_reset_reuse COMMAND tests --gtest_filter=*GivenLevelZeroLoaderPresentWhenCallingzeEventHostResetWithAlreadySignaledEventThenUsingEventAgainThenValidationLayerDoesNotPrintsWarningOfIllegalUsage*)
-set_property(TEST tests_event_reset_reuse PROPERTY ENVIRONMENT "ZE_ENABLE_NULL_DRIVER=1;ZE_ENABLE_VALIDATION_LAYER=1;ZEL_ENABLE_EVENTSDEADLOCK_CHECKER=1")
+set_property(TEST tests_event_reset_reuse PROPERTY ENVIRONMENT "ZE_ENABLE_NULL_DRIVER=1;ZE_ENABLE_VALIDATION_LAYER=1;ZEL_ENABLE_EVENTS_CHECKER=1")
diff --git a/third_party/xla/graphcycles.cc b/third_party/xla/graphcycles.cc
@@ -142,7 +142,7 @@ int32_t GraphCycles::NewNode() {
   if (rep_->free_nodes_.empty()) {
     Node n;
     n.visited = false;
-    n.rank = rep_->nodes_.size();
+    n.rank = static_cast<int32_t>(rep_->nodes_.size());
     rep_->nodes_.emplace_back(n);
     rep_->node_io_.emplace_back();
     rep_->node_data_.push_back(nullptr);
@@ -522,8 +522,9 @@ std::vector<int32_t> GraphCycles::AllNodesInPostOrder() const {
   all_nodes.reserve(rep_->nodes_.size() - free_nodes_set.size());
 
   for (int64_t i = 0, e = rep_->nodes_.size(); i < e; i++) {
-    if (!contains(free_nodes_set, i)) {
-      all_nodes.push_back(i);
+    int32_t index = static_cast<int32_t>(i);
+    if (!contains(free_nodes_set, index)) {
+      all_nodes.push_back(index);
     }
   }