[platform] Move get_cu_count to utils (#27005)

Signed-off-by: wangxiyuan <wangxiyuan1007@gmail.com>

vllm/platforms/interface.py

@@ -545,13 +545,6 @@ class Platform:
             cls._global_graph_pool = self.graph_pool_handle()
         return cls._global_graph_pool
 
-    @classmethod
-    def get_cu_count(cls, device_id: int = 0) -> int:
-        """
-        Returns the total number of compute units (CU) on single GPU.
-        """
-        raise NotImplementedError
-
     @classmethod
     def get_static_graph_wrapper_cls(cls) -> str:
         """

vllm/platforms/rocm.py

@@ -423,10 +423,6 @@ class RocmPlatform(Platform):
     def opaque_attention_op(cls) -> bool:
         return True
 
-    @classmethod
-    def get_cu_count(cls, device_id: int = 0) -> int:
-        return torch.cuda.get_device_properties(device_id).multi_processor_count
-
     @classmethod
     def is_navi(cls) -> bool:
         return "gfx1" in torch.cuda.get_device_properties(0).gcnArchName