[llvm] Reapply "[SPARC] Use umulxhi to do extending 64x64->128 multiply when we have VIS3" (#135897) (PR #136475)

[Sergei Barannikov via llvm-commits]

s-barannikov wrote:

> > Did you figure out why the cause of the buildbot failure? CI tests passed, IIRC (Just making sure it won't repeat again)
> 
> Yes, the argument filling order to `multi3` gets reversed. In the old patch it was done the upper half first ([signed](https://github.com/llvm/llvm-project/pull/135714/files#diff-fd1bc1ece510604886dbf175ea3c349afe4526b3b863464cbfec3e347fa3c8dfR9-R13), [unsigned](https://github.com/llvm/llvm-project/pull/135714/files#diff-fd1bc1ece510604886dbf175ea3c349afe4526b3b863464cbfec3e347fa3c8dfR40-R44)), now it's the lower half first ([signed](https://github.com/llvm/llvm-project/pull/136475/files#diff-fd1bc1ece510604886dbf175ea3c349afe4526b3b863464cbfec3e347fa3c8dfR9-R13), [unsigned](https://github.com/llvm/llvm-project/pull/136475/files#diff-fd1bc1ece510604886dbf175ea3c349afe4526b3b863464cbfec3e347fa3c8dfR40-R44)).


Yes, but the pattern is the same, no? The original commit did pass pre-commit checks, but failed on a buildbot due to changed instruction order. Why was the order different between CI / build bot?


https://github.com/llvm/llvm-project/pull/136475