[llvm] `dwarf2yaml.cpp` optimizations (PR #179048)

[Derek Schuff via llvm-commits]

dschuff wrote:

> @jh7370 Yes. The dwarf2yaml part was significant but definitely not the only optimization. How do you want me to benchmark? As mentioned, this LLVM version seems newer than what the modified Binaryen version was based on and contains more copies than it does, although the Binayen version has extra copies other places. Do you want me to add some of the LLVM copies to the Binaryen version of the main branch, although that will not be entirely accurate, or just compare with the version from the main branch, which is less representative for this PR? Or maybe I can use some LLVM tool to test instead?

If I understand the request, I think @jh7370 is saying, does wasm-emscripten-finalize still get noticeably faster if you apply only the dwarf2yaml.cpp changes, but not the wasm-debug.cpp changes. Just to have some evidence that this change is valuable on its own.
IMO it shouldn't be necessary to try to hack up the Binaryen fork of this code to get as close as possible; I think that if you run your build with just the dwarf2yaml.cpp changes from your Binrayen PR and you still get any significant fraction of the 3s speedup that you measured overall, that ought to be enough to motivate this change.

https://github.com/llvm/llvm-project/pull/179048