[AMD] Enable reordering across ops with nested regions

[karthik-man]

Re-enable reordering across ops with nested regions.

• This PR reverts the changes in PR5203 that prevents reordering across scf ops
• Passes omitUsesFromAbove=false to mlir::getBackwardSlice(). This ensures that the backward slice captures the values used from above, in ops with nested regions, like scf ops.

New contributor declaration

• I am not making a trivial change, such as fixing a typo in a comment.

• I have written a PR description following these
  rules.

• I have run pre-commit run --from-ref origin/main --to-ref HEAD.

• Select one of the following.

  • I have added tests.
    • /test for lit tests
    • /unittest for C++ tests
    • /python/test for end-to-end tests
  • This PR does not need a test because the lit test added in PR5203 already makes sure that we don't hoist scf ops above its dependencies

• Select one of the following.

  • [x ] I have not added any lit tests.
  • The lit tests I have added follow these best practices,
    including the "tests should be minimal" section. (Usually running Python code
    and using the instructions it generates is not minimal.)

[karthik-man]

@antiagainst @sjw36 This is the fix we discussed in PR5203. Kindly let me know if it looks good.

[sjw36]

Thanks @karthik-man ! This feature does the right thing!

FYI, it does however enable more reorders, so it may cause changes that affect performance.

[karthik-man]

Thanks @karthik-man ! This feature does the right thing!

FYI, it does however enable more reorders, so it may cause changes that affect performance.

Thanks for the review. I don't think that this enables more reorders when compared to the state of this pass before PR5203.
Prior to PR5203, ops with regions were being reordered without taking into consideration the values flowing into the regions. This change only allows reorderings that respect the data dependencies, including nested regions.

[sjw36]

Thanks for the review. I don't think that this enables more reorders when compared to the state of this pass before PR5203. Prior to PR5203, ops with regions were being reordered without taking into consideration the values flowing into the regions. This change only allows reorderings that respect the data dependencies, including nested regions.

Right, just the cases that used to crash, now will pass and reorder those cases.