[llvm] LoopVectorize: fix phi cost when it is scalar after vectorization (PR #74456)

[Florian Hahn via llvm-commits]

https://github.com/fhahn commented:

> Thanks @fhahn: the patch is now ready. Your comment about replacing the `load` with the `store` makes sense, but can you clarify why you thought `-force-vector-width=4` would work? Also, if you know, what is the difference between sse and sse4.2 in the context of this test?

> Thanks @fhahn: the patch is now ready. Your comment about replacing the load with the store makes sense, but can you clarify why you thought -force-vector-width=4 would work? Also, if you know, what is the difference between sse and sse4.2 in the context of this test?

The reason for not vectorizing was probably the cost-model deciding it is not profitable; `-force-vector-width=4` side-steps that by forcing to vectorize with VF=4 if legally possible.

Not sure about `sse4.2`.


https://github.com/llvm/llvm-project/pull/74456