[CUDAGraph]Add debug func (#3616)

* add print dot files

* refine code

fastdeploy/model_executor/graph_optimization/cudagraph_piecewise_backend.py

@@ -32,7 +32,7 @@ class ConcreteSizeEntry:
     """Record the concrete information corresponding to the current shape(num_tokens)"""
 
     # Concrete shape
-    runtime_bs: int
+    real_shape: int
     # The size is in cudagraph_capture_sizes
     use_cudagraph: bool = True
     # Has runtime-bs been captured before
@@ -114,6 +114,9 @@ class CudaGraphPiecewiseBackend:
             output._clear
 
             paddle.device.synchronize()
+
+            # For CUDAGraph debug
+            # self._save_cudagrpah_dot_files(entry)
             logger.debug(f"[CUDA GRAPH] CUDAGraph captured for real shape {padding_real_shape}")
 
         # Replay
@@ -127,7 +130,7 @@ class CudaGraphPiecewiseBackend:
         self.concrete_size_entries: Dict[int, ConcreteSizeEntry] = {}
 
         for shape in self.cudagraph_capture_sizes:
-            self.concrete_size_entries[shape] = ConcreteSizeEntry(runtime_bs=shape)
+            self.concrete_size_entries[shape] = ConcreteSizeEntry(real_shape=shape)
 
         logger.info(
             f"[CUDA GRAPH] CUDAGraph capture list {self.cudagraph_capture_sizes}, " "Created all real shape entry."
@@ -146,3 +149,11 @@ class CudaGraphPiecewiseBackend:
 
         # Create new entrys
         self._create_entry_dict()
+
+    def _save_cudagrpah_dot_files(self, entry):
+        """Print CUDAGrpah to dot files"""
+        if entry.cuda_graph:
+            entry.cuda_graph.print_to_dot_files(
+                f"./log/GraphDotFiles/backend{id(self)}_shape{entry.real_shape}",
+                1 << 0,
+            )