tsan, asan

Author: galabovaa

.bazelrc

@@ -1,6 +1,5 @@
 build:asan --strip=never
 build:asan --copt -fsanitize=address
-build:asan --copt -daddress_sanitizer
 build:asan --copt -O1
 build:asan --copt -g
 build:asan --copt -fno-omit-frame-pointer

.github/workflows/sanitizers-bazel.yml

@@ -21,7 +21,7 @@ jobs:
         run: bazel build -c dbg --config=tsan //...
 
       - name: Bazel test
-        run: bazel test -c dbg --config=tsan //...
+        run: bazel test -c dbg --config=tsan --runs_per_test 100 //...
 
       - name: Upload bazel-testlogs
         uses: actions/upload-artifact@v4

check/TestHighsParallel.cpp

@@ -218,26 +218,26 @@ TEST_CASE("MatrixMultHighs", "[parallel]") {
   matrix_multiplication("highs", parallel::num_threads(), 1);
 }
 
-// TEST_CASE("FibonacciTasksHighs", "[parallel]") {
-//   HighsTaskExecutor::shutdown();
-//   auto beg = std::chrono::high_resolution_clock::now();
-//   parallel::initialize_scheduler(numThreads);
-//   int64_t result = fib(41);
-//   auto end = std::chrono::high_resolution_clock::now();
-
-//   if (dev_run)
-//     std::cout << "time elapsed for fib(41) with HiGHS work stealing: "
-//               << (std::chrono::duration_cast<std::chrono::microseconds>(end -
-//                                                                         beg)
-//                       .count() /
-//                   1e3)
-//               << "ms" << std::endl;
-
-//   // REQUIRE(result == 4807526976);
-//   // REQUIRE(result == 20365011074);
-//   // fib 41
-//   REQUIRE(result == 267914296);
-// }
+TEST_CASE("FibonacciTasksHighs", "[parallel]") {
+  HighsTaskExecutor::shutdown();
+  auto beg = std::chrono::high_resolution_clock::now();
+  parallel::initialize_scheduler(numThreads);
+  int64_t result = fib(41);
+  auto end = std::chrono::high_resolution_clock::now();
+
+  if (dev_run)
+    std::cout << "time elapsed for fib(41) with HiGHS work stealing: "
+              << (std::chrono::duration_cast<std::chrono::microseconds>(end -
+                                                                        beg)
+                      .count() /
+                  1e3)
+              << "ms" << std::endl;
+
+  // REQUIRE(result == 4807526976);
+  // REQUIRE(result == 20365011074);
+  // fib 41
+  REQUIRE(result == 267914296);
+}
 
 #if 0
 TEST_CASE("MatrixMultOmp", "[parallel]") {

highs/parallel/HighsSplitDeque.h

@@ -156,9 +156,6 @@ class HighsSplitDeque {
 
     void publishWork(HighsSplitDeque* localDeque) {
       HighsSplitDeque* sleeper = popSleeper(localDeque);
-
-      // TSAN_ANNOTATE_HAPPENS_BEFORE(sleeper);
-
       while (sleeper) {
         uint32_t t = localDeque->selfStealAndGetTail();
         if (t == localDeque->ownerData.splitCopy) {
@@ -168,7 +165,6 @@ class HighsSplitDeque {
                                                     std::memory_order_relaxed);
             haveJobs.fetch_add(-1, std::memory_order_release);
           }
-          // TSAN_ANNOTATE_HAPPENS_AFTER(sleeper);
           pushSleeper(sleeper);
           return;
         } else {
@@ -185,23 +181,20 @@ class HighsSplitDeque {
           return;
         }
 
-        // TSAN_ANNOTATE_HAPPENS_AFTER(sleeper);
         sleeper = popSleeper(localDeque);
       }
     }
 
     HighsTask* waitForNewTask(HighsSplitDeque* localDeque) {
       pushSleeper(localDeque);
       localDeque->stealerData.semaphore.acquire();
-      // std::cout << " return injected task read " << &localDeque->stealerData.injectedTask << std::endl;
+
+      TSAN_ANNOTATE_HAPPENS_AFTER(&localDeque->stealerData.injectedTask);
+
       return localDeque->stealerData.injectedTask;
     }
   };
 
-  // const HighsTask * getInjectedTask() {
-  //   return stealerData.injectedTask;
-  // }
-
  private:
   static_assert(sizeof(OwnerData) <= 64,
                 "sizeof(OwnerData) exceeds cache line size");
@@ -472,11 +465,7 @@ class HighsSplitDeque {
   void injectTaskAndNotify(HighsTask* t) {
     stealerData.injectedTask = t;
 
-    // std::cout << "     injected task write " << &stealerData.injectedTask << std::endl;
-
-
-    // TSAN_ANNOTATE_HAPPENS_BEFORE(&stealerData.injectedTask);
-    // TSAN_ANNOTATE_HAPPENS_AFTER(&stealerData);
+    TSAN_ANNOTATE_HAPPENS_BEFORE(&stealerData.injectedTask);
 
     stealerData.semaphore.release();
   }

highs/parallel/HighsTaskExecutor.h

@@ -20,15 +20,12 @@
 #include <thread>
 #include <vector>
 
-#include <iostream>
-
 #include "parallel/HighsCacheAlign.h"
 #include "parallel/HighsSchedulerConstants.h"
 #include "parallel/HighsSplitDeque.h"
 #include "util/HighsInt.h"
 #include "util/HighsRandom.h"
 
-
 class HighsTaskExecutor {
  public:
   using cache_aligned = highs::cache_aligned;
@@ -37,11 +34,7 @@ class HighsTaskExecutor {
     bool isMain{false};
 
     void dispose();
-    ~ExecutorHandle() { 
-
-      // TSAN_ANNOTATE_HAPPENS_AFTER(&workerBunk)
-      // dispose(); 
-      }
+    ~ExecutorHandle() { dispose(); }
   };
 
  private:
@@ -106,16 +99,8 @@ class HighsTaskExecutor {
       threadLocalWorkerDeque() = localDeque;
 
       HighsTask* currentTask = ptr->workerBunk->waitForNewTask(localDeque);
-
-      // auto ptr = localDeque->getInjectedTask();
-      // TSAN_ANNOTATE_HAPPENS_BEFORE(localDeque->getInjectedTask());
-
       while (currentTask != nullptr) {
         localDeque->runStolenTask(currentTask);
-        // TSAN_ANNOTATE_HAPPENS_BEFORE(currentTask);
-
-        // TSAN_ANNOTATE_HAPPENS_AFTER(currentTask);
-
 
         currentTask = ptr->random_steal_loop(localDeque);
         if (currentTask != nullptr) continue;
@@ -133,35 +118,18 @@ class HighsTaskExecutor {
 
     workerDeques.resize(numThreads);
     workerBunk = cache_aligned::make_shared<HighsSplitDeque::WorkerBunk>();
-#ifdef __has_feature
-std::cout << "__has_feature is ON" << std::endl;
-#else
-std::cout << "__has_feature is OFF" << std::endl;
-#endif
-#ifdef TSAN_ENABLED
-std::cout << "TSAN is ON" << std::endl;
-#else
-std::cout << "TSAN is OFF" << std::endl;
-#endif
-    for (int i = 0; i < numThreads; ++i) {
+    for (int i = 0; i < numThreads; ++i)
       workerDeques[i] = cache_aligned::make_unique<HighsSplitDeque>(
           workerBunk, workerDeques.data(), i, numThreads);
 
-
-      // TSAN_ANNOTATE_HAPPENS_AFTER(&workerBunk);
-    }
-
     threadLocalWorkerDeque() = workerDeques[0].get();
     workerThreads.reserve(numThreads - 1);
     referenceCount.store(numThreads);
 
     for (int i = 1, numThreads = static_cast<int>(workerDeques.size());
          i < numThreads; ++i) {
       workerThreads.emplace_back(&HighsTaskExecutor::run_worker, i, this);
-      // TSAN_ANNOTATE_HAPPENS_BEFORE(&workerThreads[i]);
-      // std::cout <<"happens before thread " << i << ": " << &workerThreads[i] << std::endl;
     }
-
   }
 
   void stopWorkerThreads(bool blocking = false) {
@@ -172,21 +140,15 @@ std::cout << "TSAN is OFF" << std::endl;
     // now inject the null task as termination signal to every worker
     for (auto& workerDeque : workerDeques) {
       workerDeque->injectTaskAndNotify(nullptr);
-      // TSAN_ANNOTATE_HAPPENS_AFTER(workerDeque->getInjectedTask());
     }
 
     // only block if called on main thread, otherwise deadlock may occur
     if (blocking && executorHandle.isMain) {
       for (auto& workerThread : workerThreads) {
-        // TSAN_ANNOTATE_HAPPENS_AFTER(&workerThread);
-        // std::cout <<"happens after thread blocking and isMain" << ": " << &workerThread << std::endl;
         workerThread.join();
-        // std::cout <<"happens after join" << std::endl;
       }
     } else {
       for (auto& workerThread : workerThreads) {
-        // TSAN_ANNOTATE_HAPPENS_AFTER(&workerThread);
-        // std::cout <<"happens after thread " << ": " << &workerThread << std::endl;
         workerThread.detach();
       }
     }