[llvm] [AArch64] Add flag to control unrolling for small multi-exit loops (PR #131998)

[David Sherwood via llvm-commits]

david-arm wrote:

Following reviewer suggestions I've just gone ahead and enabled this for neoverse-v1 and neoverse-v2. The testing I've done doesn't suggest any downsides in terms of performance for SPEC2017 or any significant increase in code size. I saw less than 0.1% increase in code size for SPEC2017 and I observed a 0.3% increase for the LLVM test suite. However, when building SPEC2017 with libc++ I saw about 6% and 9% performance improvements for xalancbmk on neoverse-v1 and neoverse-v2, respectively.

https://github.com/llvm/llvm-project/pull/131998