[Graph][Fusion] Integrating inductor pass and npugraph ex pass

[wxsIcey]

What this PR does / why we need it?

Integrating inductor pass and npugraph ex pass, see RFC: #6347

Does this PR introduce any user-facing change?

N/A

How was this patch tested?

all tests passed.

• vLLM version: v0.14.1
• vLLM main: vllm-project/vllm@dc917cc

[github-actions]

👋 Hi! Thank you for contributing to the vLLM Ascend project. The following points will speed up your PR merge:‌‌

• A PR should do only one thing, smaller PRs enable faster reviews.
• Every PR should include unit tests and end-to-end tests ‌to ensure it works and is not broken by other future PRs.
• Write the commit message by fulfilling the PR description to help reviewer and future developers understand.

If CI fails, you can run linting and testing checks locally according Contributing and Testing.

[gemini-code-assist]

Code Review

This pull request refactors the graph fusion passes by introducing a BasePattern class to unify pattern registration for both the inductor and npugraph_ex backends. This is a good architectural improvement. However, the current implementation contains two critical bugs that will cause runtime errors: the new BasePattern calls a method that is not defined, and a class is removed without updating its usage elsewhere. My review includes specific comments and suggestions to address these critical issues.

I am having trouble creating individual review comments. Click here to see my feedback.

vllm_ascend/compilation/passes/base_pattern.py (27-43)

The call to self.get_extra_check() on line 31 will raise an AttributeError at runtime as the method is not defined in BasePattern. Additionally, the extra_check variable is unused. To fix the runtime error and improve consistency, I suggest removing the call to the undefined method and applying the existing extra_stream_scope_check to both pm.register_replacement and torchair.register_replacement.

    def register(self, pm_pass: PatternMatcherPass) -> None:
        pattern_fn = self.get_pattern()
        replacement_fn = self.get_replacement()
        example_inputs = self.get_example_inputs()
        
        pm.register_replacement(
            pattern_fn, replacement_fn, example_inputs, 
            pm.fwd_only, pm_pass, extra_check=extra_stream_scope_check
        )
        
        torchair.register_replacement(
            search_fn=pattern_fn,
            replace_fn=replacement_fn,
            example_inputs=example_inputs,
            extra_check=extra_stream_scope_check,
        )

vllm_ascend/compilation/npugraph_ex_passes/graphex_norm_quant_fusion_pass.py (28-87)

Removing the GraphEXAddRMSNormQuantPattern class introduces a NameError at runtime because GraphEXAddRMSNormFusionPass in this same file still attempts to use it on line 242.

Since AddRMSNormQuantPattern (which now uses BasePattern) handles registration for both torchair and inductor, the call to GraphEXAddRMSNormQuantPattern(...).register() is likely no longer needed and should be removed from GraphEXAddRMSNormFusionPass.__init__.

[github-actions]

This pull request has conflicts, please resolve those before we can evaluate the pull request.

[wxsIcey]

please help to verify npugraph ex is available @ChenCangtao

[github-actions]

This pull request has conflicts, please resolve those before we can evaluate the pull request.

[github-actions]

This pull request has conflicts, please resolve those before we can evaluate the pull request.