[llvm] [RISCV] Remove vmv.s.x and vmv.x.s lmul pseudo variants (PR #71501)

Luke Lau via llvm-commits llvm-commits at lists.llvm.org
Tue Dec 12 01:21:57 PST 2023


lukel97 wrote:

Rebased now that #74040 is landed, test diffs should be smaller now. 

The one remaining regression is where we lose the x0, x0 form when a vmv.s.x/vmv.x.s is the first instruction in the basic block: https://github.com/llvm/llvm-project/pull/71501/files#diff-10cadf020a900ddef6073743714540c0dd895bb1eabc9555c9fe6a0004f9d15bL358-L361

I think the correct fix for this is to perform SEW/LMUL ratio adjustment in the backwards pass (Or generally improve how we emit the x0, x0 form). But this doesn't seem to be a common occurrence, and I would prefer to leave it to a follow up patch.

https://github.com/llvm/llvm-project/pull/71501


More information about the llvm-commits mailing list