继续做一些小的inline，大头的话，还是要动这个scheduler

src/nba/include/nba/common/punning.hpp

@@ -13,7 +13,7 @@
 namespace nba {
 
 template<typename T>
-T read(void const* data, uint offset) {
+inline T read(void const* data, uint offset) {
   T value;
   std::memcpy(&value, (u8*)data + offset, sizeof(T));
   return value;

src/nba/include/nba/scheduler.hpp

@@ -249,6 +249,14 @@ private:
   static constexpr int LeftChild(int n) { return n * 2 + 1; }
   static constexpr int RightChild(int n) { return n * 2 + 2; }
 
+  void ALWAYS_INLINE Swap(int i, int j) {
+    auto tmp = heap[i];
+    heap[i] = heap[j];
+    heap[j] = tmp;
+    heap[i]->handle = i;
+    heap[j]->handle = j;
+  }
+
   void Step(u64 timestamp_next) {
     while(heap[0]->timestamp <= timestamp_next && heap_size > 0) {
       auto event = heap[0];
@@ -273,14 +281,6 @@ private:
     }
   }
 
-  void Swap(int i, int j) {
-    auto tmp = heap[i];
-    heap[i] = heap[j];
-    heap[j] = tmp;
-    heap[i]->handle = i;
-    heap[j]->handle = j;
-  }
-
   void Heapify(int n) {
     int l = LeftChild(n);
     int r = RightChild(n);