[llvm] [AArch64] Implement latency costs for loads (PR #201567)

[David Sherwood via llvm-commits]

david-arm wrote:

It looks like this PR caused a 1.5% regression in the SPEC2026 ocio benchmark on graviton 4, which is specifically related to the SelectOptimize pass. I'm still investigating, but I'm guessing we may have to either:

1. Make all latencies more accurate for all instructions to match the more accurate load latency (and hence get a balanced view of cost), or
2. Revert the change for neoverse-v2.

https://github.com/llvm/llvm-project/pull/201567