ChuanqiXu9 wrote: After I took a quick look at https://github.com/llvm/llvm-project/pull/99285, I feel this is not what I thought when I heard the idea. Correct me if I misunderstand it.
The problems are: 1. It looks like all the `.noalloc` variant are emitted all the time. This is absolutely not good. It emits duplicated code (especially currently its linkage is not internal). What I had in mind is to create `.noalloc` variant on need or lazily. 2. The original ramp function and the `.noalloc` variant shares a lot of codes if I am not mistaken. Then it is pretty bad for the code size. What I had in mind is, after we generate the `.noalloc` variant, we will rewrite the ramp function too and the ramp function will call the `.noalloc` function after it allocates the coroutine frame. https://github.com/llvm/llvm-project/pull/99283 _______________________________________________ llvm-branch-commits mailing list llvm-branch-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
