[llvm] [LoopVectorize] Improve Vectorization of Low Trip Count Loops (PR #195823)

[Jack Styles via llvm-commits]

Stylie777 wrote:

@fhahn To ensure the MicroBenchmarks do not regress, I have introduced an option here to limit the loop size that this will take effect on, or at least consider vectorisation for as we cannot accurately determine it for small loop bodies. This stops the regressions for the MicroBenchmarks and I am going to update the MicroBenchmark I added to [llvm-test-suite](https://github.com/llvm/llvm-test-suite/pull/404) to test forced vectorisation, auto vectorisation and scalar performance. It does mean that for some examples we do not see the benefits, but this should satisfy your concerns regarding some examples that do show worsening performance with this patch.

I'd like to get this merged, so please let me know your thoughts. Thanks! 

https://github.com/llvm/llvm-project/pull/195823