25% regression in compile time of issue-20936-deep-vector benchmark

[alexcrichton]

According to perf.rust-lang.org this benchmarked experienced a >25% regressoin in compile time due to #39409

Seems bad!

cc @pnkfelix, @rust-lang/compiler

[nikomatsakis]

triage: P-medium

We got a lot of stuff on our plate, and lacking evidence that this regression is more widespread, we'll consider this p-medium for now. The cause is certainly that EndRegion added a bunch of MIR statements. We can perhaps fix by adding some filtering?

[alexcrichton]

Looking at some data again, it's worth nothing that #39409 regressed almost all benchmarks, much more so than #42808 which is also perceived as a relatively high-priority regression.

[alexcrichton]

FWIW the perf graph here shows that we recovered almost all of the performance here, but there's still a 10% regression from before #39409. @arielb1 do you think it's worth opening another issue for that?