perf: optimize CGRANode::isOccupied with O(1) periodic lookup (~7.5x speedup)

src/CGRANode.cpp

@@ -377,13 +377,15 @@ bool CGRANode::canOccupy(DFGNode* t_opt, int t_cycle, int t_II) {
 }
 
 bool CGRANode::isOccupied(int t_cycle, int t_II) {
-  for (int cycle=t_cycle; cycle<m_cycleBoundary; cycle+=t_II) {
-    for (pair<DFGNode*, int> p: *(m_dfgNodesWithOccupyStatus[cycle])) {
-      // If DVFS is supported, the entire tile is occupied before the current multi-cycle operation
-      // completes. Otherwise, the next operation can start before the current one completes.
-      if (p.second == START_PIPE_OCCUPY or p.second == SINGLE_OCCUPY or m_supportDVFS) {
-        return true;
-      }
+  // Optimization: setDFGNode() has already materialized the DFG node mapping 
+  // across all cycles with II interval (populating cycle, cycle+II, etc.).
+  // Therefore, checking the specific canonical cycle here is sufficient 
+  // and equivalent to checking the entire loop.
+  for (pair<DFGNode*, int> p: *(m_dfgNodesWithOccupyStatus[t_II + (t_cycle % t_II)])){
+    // If DVFS is supported, the entire tile is occupied before the current multi-cycle operation
+    // completes. Otherwise, the next operation can start before the current one completes.
+    if (p.second == START_PIPE_OCCUPY or p.second == SINGLE_OCCUPY or m_supportDVFS) {
+      return true;
     }
   }
   return false;