[Refactor] 4/N Distinguish the branches based on the applicable scenarios of PA and FIA Ops. (#5081)

RFC: https://github.com/vllm-project/vllm-ascend/issues/4629

Reason:

We distinguish the branches based on the applicable scenarios of
pagedAttention and fusedInferAttention, making the code more clear.

At the same time, it is convenient for the subsequent iterations of
sliding_window and sinks and removePA ops after FIA is ready.

Todo:

remove PA ops after FIA is ready
add slidingwindow and ops for gpt_oss
replace FIA with FIA_v2
- vLLM version: v0.12.0
- vLLM main:
ad32e3e19c

---------

Signed-off-by: weijinqian_v1 <weijinqian@huawei.com>
Co-authored-by: weijinqian_v1 <weijinqian@huawei.com>

vllm_ascend/attention/utils.py

@@ -10,7 +10,8 @@ from vllm.distributed.kv_transfer import (get_kv_transfer_group,
                                           is_v1_kv_transfer_group)
 from vllm.forward_context import ForwardContext, get_forward_context
 
-from vllm_ascend.utils import get_ascend_config
+from vllm_ascend.utils import (AscendDeviceType, get_ascend_config,
+                               get_ascend_device_type)
 
 
 @lru_cache
@@ -18,8 +19,11 @@ def using_paged_attention(runtime_shape: int) -> bool:
     vllm_config = get_current_vllm_config()
     if vllm_config.speculative_config is not None:
         return False
+    if get_ascend_device_type() == AscendDeviceType.A5:
+        return False
     from vllm.config.compilation import CUDAGraphMode
-    if vllm_config.compilation_config.cudagraph_mode != CUDAGraphMode.FULL_DECODE_ONLY:
+    cudagraph_mode = vllm_config.compilation_config.cudagraph_mode
+    if cudagraph_mode != CUDAGraphMode.FULL_DECODE_ONLY:
         return False
 
     return runtime_shape in get_ascend_config().pa_shape_list