[llvm] [AArch64] Improve getPartialReductionCost for fixed-width VFs (PR #126538)

[David Sherwood via llvm-commits]

david-arm wrote:

Hi @SamTebbs33, after rebasing the patch I noticed some failing tests due to the tests assuming fixed-width partial reductions were legal for 64-bit accumulators. I've tried to fix the epilogue test by changing the accumulator to 32 bits instead, but I'm not sure if it's still testing the thing you want to.

https://github.com/llvm/llvm-project/pull/126538