[Fusion] [Graph] Add qknorm rope fusion operator (#4711)

### What this PR does / why we need it?
This PR add `qkv_rmsnorm_rope` operator and introduces a graph fusion
pass for `qknorm_rope` operations. The implementation includes a new
configuration flag, a pattern matching pass using
`torch._inductor.pattern_matcher`, and a custom Triton kernel for the
fused operation.

Co-authored-by: Angazenn
[supperccell@163.com](mailto:supperccell@163.com)

### Does this PR introduce _any_ user-facing change?
Yes, add new additional_config

- vLLM version: v0.12.0
- vLLM main:
ad32e3e19c

---------

Signed-off-by: wxsIcey <1790571317@qq.com>

vllm_ascend/worker/model_runner_v1.py

@@ -84,12 +84,6 @@ from vllm.v1.worker.utils import AttentionGroup
 
 import vllm_ascend.envs as envs_ascend
 from vllm_ascend.ascend_config import get_ascend_config
-from vllm_ascend.ascend_forward_context import (MoECommType,
-                                                get_mc2_tokens_capacity,
-                                                select_moe_comm_method,
-                                                set_ascend_forward_context,
-                                                set_cos_and_sin, set_mc2_mask,
-                                                set_mc2_tokens_capacity)
 from vllm_ascend.attention.attention_mask import AttentionMaskBuilder
 from vllm_ascend.attention.attention_v1 import AscendAttentionState
 from vllm_ascend.attention.utils import (AscendCommonAttentionMetadata,
@@ -111,6 +105,7 @@ from vllm_ascend.eplb.core.eplb_utils import EPLBParamUtils
 from vllm_ascend.eplb.core.eplb_worker import EplbProcess
 from vllm_ascend.eplb.eplb_updator import EplbUpdator
 from vllm_ascend.eplb.utils import model_register
+from vllm_ascend.ops.rotary_embedding import set_cos_and_sin, update_cos_sin
 from vllm_ascend.ops.weight_prefetch import WeightPrefetchMethod
 from vllm_ascend.patch.worker.patch_module import patch_torch_npu_argsort
 from vllm_ascend.sample.logits_processor import build_logitsprocs
@@ -125,6 +120,10 @@ from vllm_ascend.utils import (ACL_FORMAT_FRACTAL_ND, ACL_FORMAT_FRACTAL_NZ,
                                is_moe_model, lmhead_tp_enable, vllm_version_is)
 from vllm_ascend.worker.npu_input_batch import NPUInputBatch
 
+from vllm_ascend.ascend_forward_context import (  # isort: skip
+    MoECommType, get_mc2_tokens_capacity, select_moe_comm_method,
+    set_ascend_forward_context, set_mc2_mask, set_mc2_tokens_capacity)
+
 if TYPE_CHECKING:
     import xgrammar as xgr  # type: ignore[import-untyped]
     from vllm.v1.core.sched.output import GrammarOutput, SchedulerOutput
@@ -1122,6 +1121,9 @@ class NPUModelRunner(GPUModelRunner):
                 for layer_name in attn_group.layer_names:
                     attn_metadata[layer_name] = attn_metadata_i
 
+        # update global cos, sin
+        update_cos_sin(positions)
+
         if lmhead_tp_enable():
             max_num_reqs_across_dp = self.max_num_reqs * self.uniform_decode_query_len
             logits_indices = nn.functional.pad(
@@ -2084,6 +2086,9 @@ class NPUModelRunner(GPUModelRunner):
             else:
                 positions = self.positions.gpu[:num_tokens_padded]
 
+            # update global cos, sin
+            update_cos_sin(positions)
+
             if get_pp_group().is_first_rank:
                 intermediate_tensors = None
             else: