[mlir] [flang] [llvm] [clang] [MLIR][LLVM] Add Continuous Loop Peeling transform to SCF (PR #77328)

[via cfe-commits]

muneebkhan85 wrote:

@matthias-springer I had to move the pull request https://github.com/llvm/llvm-project/pull/71555 here due to an erroneous force push.

I have addressed all the comments you had pointed out in the original review. Most importantly 

1) I have re-written the logic for the comment https://github.com/llvm/llvm-project/pull/71555#discussion_r1444714750
 so that instead of rewriting ops after running `rewritePeeledMinMaxOp`, I instead correct the loop step in the map at the time the new loop is being created (by cloning) inside `splitLoopHelper`. This also means that there's no need for a change to the return type of `rewritePeeledMinMaxOp `as I had committed earlier. If this looks good to you, I can revert the changes to the return type of `rewritePeeledMinMaxOp`.

2) I have added a new test case, so that both cases `usePowerSplit = false` and `usePowerSplit = true` are tested https://github.com/llvm/llvm-project/pull/71555#discussion_r1444699098



https://github.com/llvm/llvm-project/pull/77328