[llvm] [RISCV] Use vsetvli instead of vlenb in Prologue/Epilogue (PR #113756)
Pengcheng Wang via llvm-commits
llvm-commits at lists.llvm.org
Mon Oct 28 20:35:53 PDT 2024
wangpc-pp wrote:
I think we should make it a feature. For Xiangshan, IIUC, `csrr a0, vlenb` will be decoded as `addi a0, zero, $vlenb`, and with fusion/wide issue width, it won't be slower than `vsetvli`. There are too many things we can do with the implementation of `vsetvli` and as a consequence we can't assume the behavior here.
https://github.com/llvm/llvm-project/pull/113756
More information about the llvm-commits
mailing list