rsz: preserve buffers on hierarchical module boundaries in canRemoveBuffer

[oharboe]

Summary

When global placement performs timing-driven buffer removal (Resizer::removeBuffers), canRemoveBuffer checks whether a candidate buffer can be safely removed. If the buffer is situated on a hierarchical module boundary where its input pin and output pin belong to different dbModNets, removing the buffer flattens the net without updating hierarchical module port/net bindings, leaving downstream module inputs disconnected.

Fix

Check whether input_modnet != output_modnet in Resizer::canRemoveBuffer and return false to preserve buffers located on hierarchical module boundaries.

[gemini-code-assist]

Code Review

This pull request modifies Resizer::canRemoveBuffer to prevent buffer removal when the input and output hierarchical nets differ. The reviewer correctly identifies a critical side-effect where this change causes Resizer::removeBuffer to merge a hierarchical net with itself, potentially leading to undefined behavior or crashes in OpenDB. The reviewer suggests updating the merging logic to only merge distinct nets.

[oharboe]

@precisionmoon Heads up! I changed reviewer because reviewer was automatically assigned and this is a continuation of a conversation between me and @jhkim-pii

[oharboe]

/gemini review

[gemini-code-assist]

Code Review

This pull request modifies Resizer.cc to prevent buffer removal when the input and output hierarchical nets differ, and updates the buffer removal logic to avoid merging identical hierarchical nets. The reviewer pointed out that the updated condition in removeBuffer causes a fall-through to the else if branch when both survivor_modnet and removed_modnet are identical and non-null, leading to a redundant call to connectTermsOf. Nesting the inequality check inside the non-null check is suggested to resolve this issue.

[gemini-code-assist]

If survivor_modnet == removed_modnet (and both are non-null), the current condition survivor_modnet != removed_modnet evaluates to false. This causes the execution to fall through to the else if (survivor_modnet != nullptr) branch, which will then execute survivor_modnet->connectTermsOf(db_removed). Since both nets are already part of the same hierarchical module net, calling connectTermsOf is redundant and potentially problematic.

Nesting the inequality check inside the if (survivor_modnet != nullptr && removed_modnet != nullptr) block avoids falling through to the else if branches when both modnets are identical.

  if (survivor_modnet != nullptr && removed_modnet != nullptr) {
    if (survivor_modnet != removed_modnet) {
      survivor_modnet->mergeModNet(removed_modnet);
    }
  } else if (survivor_modnet != nullptr) {

[oharboe]

@jhkim-pii I don't know what I'm doing here... I don't know if I should follow Gemini's directions... Please advice.