David Green

Thanks for the report. I'll put up a quick fix - it's just a problem with the costmodel not considering that legalized vector types could not be vectors.

Sounds good. Lets try.

/cherry-pick de6dfbbb300e552efa1cd86a023063a39d408b06

/cherry-pick de6dfbbb300e552efa1cd86a023063a39d408b06

I believe that with #58512 this is fixed, even if not optimal. Please let me know if not

Ping - I've rebased and fixed the RISCV tests. Thanks.

Hi this sounds like a good idea to me. Note that the implementation of getHostCPUFeatures isn't amazing for AArch64 at the moment, there was an attempt to fix it up...

It is that bit of code, yeah. I don't know of a way to reproduce this without logging into different machines with different sets of options and trying it. If...

Hi - The last vrev64.32 looks suspicious. There is a call to bitcast here: https://github.com/llvm/llvm-project/blob/74984dee51307779a3eab10a8cd6102be37e1081/llvm/lib/Target/ARM/ARMISelLowering.cpp#L14801, that should probably be an nvcast / ARMISD::VECTOR_REG_CAST instead.

There might well be other places where this is wrong for sure. It shouldn't matter in others though. If the element size is there same then there should not be...