Asking JIT not to align loops triggers multiple asserts

[filipnavara]

Try to compile code on ARM64 with JitAlignLoopBoundary=4 option. At least three different asserts are hit.

runtime/src/coreclr/jit/emit.cpp

Line 5732 in 516e81e

assert(emitLastIns->idIns() == INS_align);

runtime/src/coreclr/jit/emit.cpp

Line 6067 in 516e81e

assert(markedLastLoop && markedCurrLoop);

and one more that can be suppressed with JitHideAlignBehindJmp=0. I forgot the exact location.

[EgorBo]

сс @kunalspathak @amanasifkhalid

[kunalspathak]

JitAlignLoopBoundary was an experimental config that was part of non-adaptive loop alignment and is untested.

[kunalspathak]

Is there a specific scenario where you are not seeing the desired alignment and hence trying out loop boundary option?

[filipnavara]

Is there a specific scenario where you are not seeing the desired alignment and hence trying out loop boundary option?

I was not aware of the JitAlignLoops=0 option and I tried to disable alignment to test the scenario in #107284.

[kunalspathak]

@filipnavara are you unblocked or do you want anything out of this issue?