[llvm-branch-commits] [llvm] release/19.x: [AArch64] Add streaming-mode stack hazard optimization remarks (#101695) (PR #102168)

[David Green via llvm-branch-commits]

davemgreen wrote:

> The patch here is pretty big in size, but it seems to only affects the remarks, on the other hand it doesn't seem to really fix anything and in that case I feel like RC3 might be the wrong time to merge this. Is there a huge upside to take this this late in the process?

Thanks - I wasn't sure what state the branch was in. As @jroelofs points out the issues this is attempting to help with can pretty performance-sensitive and a hard to diagnose without assistance. The issue is that when and where spills happen can occur quite chaotically out of the register allocator, and so users need to be using the same compiler to diagnose the issues as they will use in practice. Having to provide patches and for users to build the compiler themselves is quite difficult compared to having this on the branch.

All the code (meaningfully) changed in this patch needs to be enabled with both `-Rpass-analysis=sme` and a backend `-mllvm -aarch64-stack-hazard-remark-size=XYZ` (or `-mllvm -aarch64-stack-hazard-size=xyz`), so the chance of it breaking anything else should be very low.

https://github.com/llvm/llvm-project/pull/102168