[clang] [llvm] [MC] Make UseAssemblerInfoForParsing mostly true (PR #91082)

[Fangrui Song via llvm-commits]

MaskRay wrote:

> > @gulfemsavrun Commit [245491a](https://github.com/llvm/llvm-project/commit/245491a9f384e4c53421196533c2a2b693efaf8d) (`[MC] Disable MCAssembler based constant folding for DwarfDebug`; independent of this patch) should have fixed the compile time regression. Might be worth testing the commit before and at [245491a](https://github.com/llvm/llvm-project/commit/245491a9f384e4c53421196533c2a2b693efaf8d) to see the time decrease.
> 
> @MaskRay I already checked with that patch, and we still have the same issue. We see this issue in llvm ToT. For example, this is a run from today: [luci-milo.appspot.com/ui/p/fuchsia/builders/prod/clang-mac-x64/b8746512510687347537/overview](https://luci-milo.appspot.com/ui/p/fuchsia/builders/prod/clang-mac-x64/b8746512510687347537/overview)

This change exposed a latent performance issue (primarily due to -O1 or above + -g) in  https://reviews.llvm.org/D153096 , which should be properly fixed by b06e736982a3568fe2bcea8688550f9e393b7450 (just landed). Can you verify?

245491a9f384e4c53421196533c2a2b693efaf8d was a workaround, but I believe it effectively addressed the performance issue, and I am unable to detect compile time regression. I am curious if you can find a source file that demonstrates a slowdown with 245491a9f384e4c53421196533c2a2b693efaf8d applied.  The `AttemptToFoldSymbolOffsetDifference` function is likely related.

(The aforementioned commits modify different files and are independent.)

https://github.com/llvm/llvm-project/pull/91082