perf: optimize causal conv1d and mega chunk gdn for qwen3.5.#1999

yingxudeng merged 1 commit into

xLLM-AI:mainxLLM-AI/xllm:mainfrom

Sinle4Cat:prefill_opt_conv1d_gdnSinle4Cat/xllm:prefill_opt_conv1d_gdnCopy head branch name to clipboard