Pad SpinLock to avoid false sharing (#261)

* Pad SpinLock to avoid false sharing

* Fix a compilation error

Author: zhengyu123

ddprof-lib/src/main/cpp/arch_dd.h

@@ -5,6 +5,8 @@
 
 #include <stddef.h>
 
+constexpr int DEFAULT_CACHE_LINE_SIZE = 64;
+
 static inline long long atomicInc(volatile long long &var,
                                   long long increment = 1) {
   return __sync_fetch_and_add(&var, increment);

ddprof-lib/src/main/cpp/spinLock.h

@@ -21,15 +21,17 @@
 
 // Cannot use regular mutexes inside signal handler.
 // This lock is based on CAS busy loop. GCC atomic builtins imply full barrier.
-class SpinLock {
+class alignas(DEFAULT_CACHE_LINE_SIZE) SpinLock {
 private:
   //  0 - unlocked
   //  1 - exclusive lock
   // <0 - shared lock
   volatile int _lock;
-
+  char _padding[DEFAULT_CACHE_LINE_SIZE - sizeof(_lock)];
 public:
-  constexpr SpinLock(int initial_state = 0) : _lock(initial_state) {}
+  constexpr SpinLock(int initial_state = 0) : _lock(initial_state), _padding() {
+    static_assert(sizeof(SpinLock) == DEFAULT_CACHE_LINE_SIZE);
+  }
 
   void reset() { _lock = 0; }