address comments

Signed-off-by: Kunshang Ji <[email protected]>

Author: jikunshang

vllm/compilation/fix_functionalization.py

@@ -27,16 +27,19 @@ class FixFunctionalizationPass(VllmInductorPass):
     """
 
     def __call__(self, graph: torch.fx.Graph):
+        # XPU does not support auto-functionalization yet.
+        # Will enable this when switch to vllm-xpu-kernels.
+        if current_platform.is_xpu():
+            logger.debug("XPU platform does not support fix functionality"
+                         "pass currently.")
+            return
+
         self.begin()
         self.dump_graph(graph, "before_fix_functionalization")
 
         self.nodes_to_remove: list[torch.fx.Node] = []
         count = 0
         for node in graph.nodes:
-            # XPU does not support auto-functionalization yet.
-            # Will enable this when switch to vllm-xpu-kernels.
-            if current_platform.is_xpu():
-                continue
             if not is_func(node, auto_functionalized):
                 continue  # Avoid deep if-elif nesting
 

vllm/platforms/xpu.py

@@ -2,7 +2,7 @@
 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project
 
 import os
-from typing import TYPE_CHECKING, Optional
+from typing import TYPE_CHECKING, Any, Optional
 
 import torch
 
@@ -78,10 +78,6 @@ def get_device_total_memory(cls, device_id: int = 0) -> int:
     def is_async_output_supported(cls, enforce_eager: Optional[bool]) -> bool:
         return True
 
-    @classmethod
-    def get_piecewise_backend_cls(cls) -> str:
-        return "vllm.compilation.cuda_piecewise_backend.CUDAPiecewiseBackend"  # noqa
-
     @classmethod
     def inference_mode(cls):
         return torch.no_grad()
@@ -201,3 +197,9 @@ def supports_v1(cls, model_config: ModelConfig) -> bool:
     @classmethod
     def device_count(cls) -> int:
         return torch.xpu.device_count()
+
+    def get_global_graph_pool(self) -> Any:
+        """
+        Currently xpu does NOT support Graph model.
+        """
+        return None