Integrate atomic exercise as a final task in race exercise.

exercises/CMakeLists.txt

@@ -15,7 +15,6 @@ endif()
 # Include the exercises that (should) work on all platforms.
 add_subdirectory( hello )
 add_subdirectory( asan )
-add_subdirectory( atomic )
 add_subdirectory( basicTypes )
 add_subdirectory( callgrind )
 add_subdirectory( condition_variable )

exercises/ExercisesCheatSheet.md

@@ -344,13 +344,9 @@ Concurrency Exercises
 
 Typical race condition where a simple mutex and lock_guard "solves" the problem.
 
-The second step is to look at the execution time and find out that it's not really a solution. One could then try an atomic and see the difference, although I do not introduce them in the course
+The second step is to look at the execution time and find out that it's not really a solution.
 
-### Atomicity (directory: [`atomic`](atomic))
-
-Exactly the same race condition as above. Fix them using an `atomic<int>`.
-
-*Optional*: Compare run times for lock and atomic solution. Those are likely not very different, as many locks are implemented using atomics.
+Try then to use an `atomic<int>` to solve the issue and compare the execution time with the lock solution.
 
 ### Condition variables (directory: [`condition_variable`](condition_variable))
 

exercises/race/CMakeLists.txt

@@ -13,6 +13,10 @@ add_executable( racing "racing.cpp" )
 target_link_libraries( racing PRIVATE Threads::Threads )
 
 # Create the "solution executable".
-add_executable( racing.sol EXCLUDE_FROM_ALL "solution/racing.sol.cpp" )
-target_link_libraries( racing.sol PRIVATE Threads::Threads )
-add_dependencies( solution racing.sol )
+add_executable( racing.sol1 EXCLUDE_FROM_ALL "solution/racing.sol1.cpp" )
+target_link_libraries( racing.sol1 PRIVATE Threads::Threads )
+add_dependencies( solution1 racing.sol1 )
+add_executable( racing.sol2 EXCLUDE_FROM_ALL "solution/racing.sol2.cpp" )
+target_link_libraries( racing.sol2 PRIVATE Threads::Threads )
+add_dependencies( solution2 racing.sol2 )
+add_dependencies( solution solution1 solution2 )

exercises/race/Makefile

@@ -1,14 +1,17 @@
 PROGRAM_NAME=racing
 
 all: $(PROGRAM_NAME)
-solution: $(PROGRAM_NAME).sol
+solution: $(PROGRAM_NAME).sol1 $(PROGRAM_NAME).sol2
 
 
 clean:
-	rm -f *o $(PROGRAM_NAME) *~ core $(PROGRAM_NAME).sol
+	rm -f *o $(PROGRAM_NAME) *~ core $(PROGRAM_NAME).sol?
 
 $(PROGRAM_NAME) : $(PROGRAM_NAME).cpp
 	${CXX} ${CXXFLAGS} -g -std=c++17 -O2 -pthread -Wall -Wextra -L. -o $@ $<
 
-$(PROGRAM_NAME).sol : solution/$(PROGRAM_NAME).sol.cpp
+$(PROGRAM_NAME).sol1 : solution/$(PROGRAM_NAME).sol1.cpp
+	${CXX} ${CXXFLAGS} -g -std=c++17 -O2 -pthread -Wall -Wextra -L. -o $@ $<
+
+$(PROGRAM_NAME).sol2 : solution/$(PROGRAM_NAME).sol2.cpp
 	${CXX} ${CXXFLAGS} -g -std=c++17 -O2 -pthread -Wall -Wextra -L. -o $@ $<

exercises/race/README.md

@@ -1,12 +1,16 @@
 
 ## Instructions
 
-* Compile and run the executable, see if it races
-    * If you have a bash shell, try `./run ./racing`, which keeps invoking the executable
-      until a race condition is detected
-* (Optional) You can use `valgrind --tool=helgrind ./racing` to prove your assumption
-* (Optional) If your operating system supports it, recompile with thread sanitizer.
+The program `racing.cpp` is incrementing a shared integer many times, within several threads, which should lead to race conditions if no specific protection is used. The values of the global parameters `nThread`, `nInc` and `nRepeat` can be custommized for your own computer.
+
+Tasks
+- Compile and run the executable, check it races.
+- If you have a bash shell, try `./run ./racing`, which keeps invoking the executable until a race condition is detected.
+- (Optional) You can use `valgrind --tool=helgrind ./racing` to prove your assumption
+- (Optional) If your operating system supports it, recompile with thread sanitizer.
   With Makefile, use e.g. `make CXXFLAGS="-fsanitize=thread"`
-* Use a mutex to fix the issue
-* See the difference in execution time
-* (Optional) Check again with `valgrind` or thread sanitizer if the problem is fixed
+- Use a `std::mutex` to fix the issue.
+- See the difference in execution time, for example with `time ./racing`.
+  You might have to increase `nRepeat` if it completes too fast, or lower it if it takes too long.
+- (Optional) Check again with `valgrind` or thread sanitizer if the problem is fixed.
+- Try to use `std::atomic` instead of the mutex, and compare the execution time.

exercises/race/racing.cpp

@@ -1,37 +1,40 @@
+
 #include <iostream>
 #include <thread>
 #include <vector>
 
 /*
- * This program tries to increment an integer 100 times from multiple threads.
- * If the result comes out at 100*nThread, it stays silent, but it will print
+ * This program tries to increment an integer `nInc` times in `nThread` threads.
+ * If the result comes out at `nInc*nThread`, it stays silent, but it will print
  * an error if a race condition is detected.
  * If you don't see it racing, try ./run ./racing, which keeps invoking the
  * executable until a race condition is detected.
  */
 
-constexpr unsigned int nThread = 2;
+constexpr std::size_t nThread = 10;
+constexpr std::size_t nInc = 1000;
+constexpr std::size_t nRepeat = 1000;
 
 int main() {
   int nError = 0;
 
-  for (int j = 0; j < 1000; j++) {
+  for (std::size_t j = 0; j < nRepeat; j++) {
     int a = 0;
 
     // Increment the variable a 100 times:
-    auto inc100 = [&a](){
-      for (int i = 0; i < 100; ++i) {
+    auto increment = [&a](){
+      for (std::size_t i = 0; i < nInc; ++i) {
         a++;
       }
     };
 
-    // Start up all threads:
+    // Start up all threads
     std::vector<std::thread> threads;
-    for (unsigned int i = 0; i < nThread; ++i) threads.emplace_back(inc100);
+    for (std::size_t i = 0; i < nThread; ++i) threads.emplace_back(increment);
     for (auto & thread : threads) thread.join();
 
     // Check
-    if (a != nThread * 100) {
+    if (a != nThread * nInc) {
       std::cerr << "Race detected! Result: " << a << '\n';
       nError++;
     }

exercises/race/solution/racing.sol1.cpp

@@ -0,0 +1,39 @@
+
+#include <iostream>
+#include <vector>
+#include <thread>
+#include <mutex>
+
+constexpr std::size_t nThread = 10;
+constexpr std::size_t nInc = 1000;
+constexpr std::size_t nRepeat = 1000;
+
+int main() {
+  int nError = 0;
+
+  for (std::size_t j = 0; j < nRepeat; j++) {
+    int a = 0;
+    std::mutex aMutex;
+
+    // Increment the variable a 100 times:
+    auto increment = [&a,&aMutex](){
+      for (std::size_t i = 0; i < nInc; ++i) {
+        std::scoped_lock lock{aMutex};
+        a++;
+      }
+    };
+
+    // Start up all threads:
+    std::vector<std::thread> threads;
+    for (std::size_t i = 0; i < nThread; ++i) threads.emplace_back(increment);
+    for (auto & thread : threads) thread.join();
+
+    // Check
+    if (a != nThread * nInc) {
+      std::cerr << "Race detected! Result: " << a << '\n';
+      nError++;
+    }
+  }
+
+  return nError;
+}

exercises/race/solution/racing.sol2.cpp

@@ -0,0 +1,37 @@
+
+#include <iostream>
+#include <vector>
+#include <thread>
+#include <atomic>
+
+constexpr std::size_t nThread = 10;
+constexpr std::size_t nInc = 1000;
+constexpr std::size_t nRepeat = 1000;
+
+int main() {
+  int nError = 0;
+
+  for (std::size_t j = 0; j < nRepeat; j++) {
+    std::atomic<int> a{0};
+
+    // Increment the variable a 100 times:
+    auto increment = [&a](){
+      for (std::size_t i = 0; i < nInc; ++i) {
+        a++;
+      }
+    };
+
+    // Start up all threads
+    std::vector<std::thread> threads;
+    for (std::size_t i = 0; i < nThread; ++i) threads.emplace_back(increment);
+    for (auto & thread : threads) thread.join();
+
+    // Check
+    if (a != nThread * nInc) {
+      std::cerr << "Race detected! Result: " << a << '\n';
+      nError++;
+    }
+  }
+
+  return nError;
+}