[llvm] [AArch64] Generate DOT instructions from matching IR (PR #69583)

[Graham Hunter via llvm-commits]

huntergr-arm wrote:

This pass works with a change to LoopVectorize (https://github.com/llvm/llvm-project/pull/69587) to match IR to SVE SDOT/UDOT instructions and increase the effective bandwidth of the loop.

It's downstream code I wrote 4 years ago that we're trying to upstream, so if a completely different approach for generating the dot instructions automatically would be preferable please let me know.

https://github.com/llvm/llvm-project/pull/69583