[CPU][Perf] Accelerate unquantized MoE for AArch64

[fadara01]

Purpose

Accelerate unquantized MoE for AArch64

• Enable FusedMoE kernel for AArch64
• Implement AdvSIMD BFMMLA interface to accelerate w13 and w2 GEMMs
• Extend generic micro kernel interface and MoE kernel to support packing input matrix
• Abstract sleef.h includes and tanh symbol for x86 under the AVX vectorizer class

Performance

1.96x higher throughput for gpt-oss and 2.18x higher throughput for gemma4 with benchmark below and 64 Neoverse-V2 cores

MODEL=unsloth/gpt-oss-20b-BF16
#MODEL=google/gemma-4-26B-A4B-it
# gemma4 needs this as attention currently hangs without it.
#export VLLM_CPU_ATTN_SPLIT_KV=0 
vllm bench throughput \
  --num-prompts 128 \
  --seed 0 \
  --dataset-name sonnet \
  --dataset-path /home/fadara01/vllm-moe/vllm/benchmarks/sonnet.txt \
  --input-len 256 \
  --output-len 256 \
  --max-model-len 4096 \
  --max-num-batched-tokens 4096 \
  --model $MODEL \
  --tensor-parallel-size 1 \
  --no-enable-prefix-caching \
  --num-warmups 5

Test Plan

CI

Test Result

---

Essential Elements of an Effective PR Description Checklist

• The purpose of the PR, such as "Fix some issue (link existing issues this PR will resolve)".
• The test plan, such as providing test command.
• The test results, such as pasting the results comparison before and after, or e2e results
• (Optional) The necessary documentation update, such as updating supported_models.md and examples for a new model.

[fadara01]

Hi @mgoin @bigPYJ1151 :)

Could you please have a look at this?

[mergify]

This pull request has merge conflicts that must be resolved before it can be
merged. Please rebase the PR, @fadara01.

https://docs.github.com/en/pull-requests/collaborating-with-pull-requests/working-with-forks/syncing-a-fork