flashinfer-ai / flashinfer

FlashInfer: Kernel Library for LLM Serving
https://flashinfer.ai
Apache License 2.0
768 stars 64 forks source link

bugfix: suppress alignment warning of sampling kernels #297

Closed yzh119 closed 3 weeks ago

yzh119 commented 3 weeks ago

We declare multiple kernels inside the sampling.cuh and they use dynamic shared memory (with the same extern variable name) with different alignment requirements (e.g. some are alignof 4, some are alignof 64).

In this PR we use different names for extern variable that have different alignment requirements to suppress the warning.