BSWAP can use the VREV instruction on MVE to produce better results than expanding.
Details
Details
Diff Detail
Diff Detail
- Repository
- rL LLVM
Event Timeline
Comment Actions
I guess an i8 bswap isn't a thing that makes sense! That's good.
Can you add a 64bit test. We have a vrev64.8 too if you want to try and add the codegen for it.
Comment Actions
LGTM
llvm/test/CodeGen/Thumb2/mve-bswap.ll | ||
---|---|---|
27 | We get this one for free through a shuffle vector? That's good! |
We get this one for free through a shuffle vector? That's good!