mpl: ignore the macro being placed in overlap checks

[naveenvenk17]

Fixes #11269.

place_macro sets the target macro location before checking for overlaps. The overlap scan then finds that same placed macro and reports a self-overlap, so a valid placement fails unless overlap checking is disabled.

Skip the instance under placement while keeping the checks against every other placed macro. The existing right-angle placement regression now marks the macro placed before calling place_macro, covering the failing path.

Validation on the clean OpenROAD VM:

• Targeted placed-macro regression: passed
• Full MPL suite: running on the VM

[gemini-code-assist]

Code Review

This pull request prevents a macro from checking overlap with itself in findOverlappedMacros inside rtl_mp.cpp. It also updates Bazel test targets and introduces several new tests, including place_macro_right_angle_rotation. The review feedback points out that the golden file place_macro_right_angle_rotation.defok was placed in the wrong directory (testcases/), which will cause Bazel test failures, and should be moved to the parent test directory.

[gemini-code-assist]

The golden file place_macro_right_angle_rotation.defok is added under src/mpl/test/testcases/, but this diff_files call expects it in the current directory (src/mpl/test/). Additionally, the Bazel glob in src/mpl/test/BUILD (glob(["place_macro_right_angle_rotation.*"])) will not find it inside testcases/, which will cause the Bazel test to fail. To resolve this and align with the repository's conventions, please move the golden file from src/mpl/test/testcases/place_macro_right_angle_rotation.defok to src/mpl/test/place_macro_right_angle_rotation.defok.

[naveenvenk17]

Updated for current master. The golden output now comes from the existing parent test location, so there is no separate misplaced .defok file.

The regression now first marks the target macro as placed, then moves it. That exercises the self-overlap path directly. I also squashed the branch into one signed commit.

Validation on the GCP VM:

• //src/mpl/test:place_macro_right_angle_rotation-tcl_test passed three consecutive runs
• //src/mpl/test:all passed 42 of 42 tests
• clang-format-18 --dry-run --Werror and git diff --check passed

[AcKoucher]

@naveenvenk17 Thanks for fixing this!

[naveenvenk17]

Thanks, I split that setup out.

place_macro_right_angle_rotation is back to checking only the R90 placement. The new place_macro_already_placed_overlap test first marks the macro placed, then requests the same legal R90 placement without allowing overlap. That path only succeeds when the target macro is excluded from its own overlap search.

The new regression has its own expected log and DEF files, plus Bazel and CMake registration.

I rebased this on current master and validated it on the GCP VM:

• both focused placement tests passed three uncached runs
• the full //src/mpl/test:all suite passed 43 of 43 targets
• clang-format --dry-run --Werror and git diff --check passed

I also had the final test split reviewed separately before the last test run; there were no blocking findings.

[AcKoucher]

@naveenvenk17 Please, do not force-push after a review, because we lose history.

[AcKoucher]

Please, use an input DEF that does not require setting -orientation that way we keep the test case narrower.

[naveenvenk17]

You are right. I rebased and force-pushed to keep the branch current, but that removed review history. I will keep any further updates as additive commits and will not force-push this PR again.