Revert "[opt] remove trivially dead instructions in mandatory combine"

[xedin]

Reverts #30429

Testing to see whether this is a cause of SIL failure in source compatibility project Kingfisher.

[xedin]

@swift-ci please test source compatibility

[xedin]

Reverting this PR fixes Kingfisher project failures in debug configuration.

[xedin]

/cc @zoecarver @gottesmm

[xedin]

@swift-ci please smoke test

[xedin]

Looks like there is an unrelated break now related to MovieSwiftUI...

[zoecarver]

@xedin we tested fairly extensively (source compatibility and device filecheck preset). What's the error?

[xedin]

@zoecarver https://ci.swift.org/view/Source%20Compatibility/job/swift-master-source-compat-suite-debug/2848/artifact/swift-source-compat-suite/ (Kingfisher fails in 4 configurations).

[zoecarver]

Hmm I see. It might have somehow removed the destroy_addr. I'll play around with it a bit and see if I can make a reproducer.

[zoecarver]

Looks like the kingfisher build is now succeeding. Still no reproducer but, it was something to do with one of the enum operators.

[zoecarver]

@xedin I'll keep trying to debug why the crash happened but I'm beginning to suspect it wasn't this commit, actually. When building Kingfisher from master, it doesn't crash anymore but, when I checkout a5d1214220b8ddf29acf7c08c8bb42993ed56c54 and try to build I get the crash. Anyway, as I said, I'll keep trying to figure out what the cause is.

[zoecarver]

Ignore the last comment. I tracked down the issue, it was this commit (oops). Here's the pattern that we incorrectly remove (unchecked_take_enum_data_addr):

%96 = alloc_stack $KingfisherOptionsInfoItem    // users: %111, %105, %98, %97
  copy_addr %13 to [initialization] %96 : $*KingfisherOptionsInfoItem // id: %97
  %98 = unchecked_take_enum_data_addr %96 : $*KingfisherOptionsInfoItem, #KingfisherOptionsInfoItem.downloadPriority!enumelt // user: %99
  %99 = load [trivial] %98 : $*Float