Ensure that JIT and other IL compilers preserve order of data dependent memory accesses

[VSadov]

Re: #75790 (comment)

There is a concern about compiler optimization honoring ordering of data-dependent reads (and writes to object fields to a lesser degree). We expect that from the hardware (for writes we insert fences if needed). The question was whether compiler itself may violate such ordering in some optimizations and whether it is costly to prevent that if it does. I think the concern was mostly about reads as writing has obvious sideeffects, which limits what optimizations can do.

Preliminary conclusion was that compiler honors data dependency because the internal representations are tree-based, but it needs to be confirmed.

If that is the case we need to make it more intentional that compilers avoid invalid optimizations in the future.
If compiler actually violate the order and fixing is not feasible, we may consider adjusting the documentation.

This also applies to Mono LLVM-based codegen.

[dotnet-issue-labeler]

I couldn't figure out the best area label to add to this issue. If you have write-permissions please help me learn by adding exactly one area label.

[ghost]

Tagging subscribers to this area: @JulieLeeMSFT, @jakobbotsch, @kunalspathak
See info in area-owners.md if you want to be subscribed.

Issue Details

---

Re: #75790 (comment)

There was a concern about compiler optimization honoring ordering of data-dependent reads (and writes to object fields to a lesser degree). We expect that from the hardware (for writes we insert fences if needed). The question was whether compiler itself may violate such ordering in some optimizations and whether it is costly to prevent that if it does. I think the concern was mostly about reads as writing has obvious sideeffects, which limits what optimizations can do.

Preliminary conclusion was that compiler honors data dependency because the internal representations are tree-based, but it needs to be confirmed.

If that is the case we need to make it more intentional that compilers avoid invalid optimizations in the future.
If compiler actually violate the order and fixing is not feasible, we may consider adjusting the documentation.

This also applies to Mono LLVM-based codegen.

Author:	VSadov
Assignees:	-
Labels:	area-CodeGen-coreclr, untriaged, memory model
Milestone:	-

[JulieLeeMSFT]

cc @dotnet/jit-contrib.