drt: fix 3D pin access via generation from bottom layers

[openroad-ci]

fix the issue: #7138

[gemini-code-assist]

Code Review

This pull request fixes an issue with 3D pin access generation by correctly handling vias from bottom layers. The changes in genViaEnclosedCoords ensure the correct via bounding box is used depending on the layer, and genAPEnclosedBoundary is updated to search for vias on lower layers, not just the layer above.

The changes look correct and address the problem. I have a couple of suggestions to improve code robustness and maintainability.

[gemini-code-assist]

The current logic assumes that if layer_num is not via_def->getLayer1Num(), it must be via_def->getLayer2Num(). This is generally true for vias, but it's safer to explicitly check for it to make the code more robust against unexpected inputs. Adding an else if for the second layer and a fallback case for logging a warning would be a good improvement.

  if (via_def->getLayer1Num() == layer_num) {
    box = via.getLayer1BBox();
  } else if (via_def->getLayer2Num() == layer_num) {
    box = via.getLayer2BBox();
  } else {
    logger_->warn(DRT, 0, "genViaEnclosedCoords: layer_num {} is not part of via {}", layer_num, via_def->getName());
    return;
  }

[gemini-code-assist]

The logic for finding vias upwards and downwards is very similar, leading to code duplication. This can be refactored into a helper lambda function to improve maintainability and readability. The lambda can encapsulate the logic of iterating through via definitions for a given cut layer, generating coordinates, and checking against max_num_via_trial.

Additionally, the current implementation for finding 'up' vias uses operator[] on layer_num_to_via_defs_, which can insert an element if the key doesn't exist. Using find and at (as done for 'down' vias) is safer. The suggested refactoring addresses both points.

  auto find_and_gen_vias = [&](int cut_layer, int& count) {
    if (layer_num_to_via_defs_.find(cut_layer) == layer_num_to_via_defs_.end()) {
      return false;
    }
    for (auto& [tup, via] : layer_num_to_via_defs_.at(cut_layer)[1]) {
      genViaEnclosedCoords(coords, rect, via, layer_num, is_curr_layer_horz);
      count++;
      if (count >= max_num_via_trial) {
        return true;
      }
    }
    return false;
  };

  if (layer_num + 1 <= getDesign()->getTech()->getTopLayerNum()) {
    find_and_gen_vias(layer_num + 1, cnt);
  }

  int down_via_cnt = 0;
  for (int l = layer_num - 1;
       l >= layer_num - 3 && l >= getDesign()->getTech()->getBottomLayerNum();
       l--) {
    if (find_and_gen_vias(l, down_via_cnt)) {
      break;
    }
  }

[github-actions]

clang-tidy review says "All clean, LGTM! 👍"

[github-actions]

clang-tidy review says "All clean, LGTM! 👍"

[github-actions]

clang-tidy review says "All clean, LGTM! 👍"

[github-actions]

clang-tidy review says "All clean, LGTM! 👍"

[jferreiraOpenRoad]

This PR introduces enhancements to FlexPA to improve access point generation robustness, specifically addressing connectivity for Block Terminals (BTerms) in 3D/top-level contexts and ensuring compliance with LEF58_RECTONLY layer constraints.

Key Changes:

Smart Handling for RECTONLY Constraints:

In createMultipleAccessPoints, added logic to verify if a pin shape aligns with the layer width when LEF58_RECTONLY is present.
If the shape is not width-aligned (which would cause a DRC violation for planar routing), planar access is automatically disabled.
Fallback Strategy: Via access is explicitly enabled in these scenarios, forcing the router to escape to an adjacent layer to ensure connectivity without violating rules.
Downward Via Generation for BTerms:

Introduced the is_bterm parameter in genAPsFromRect, genAPEnclosedBoundary, and genAPCosted to distinguish between standard cell pins and block terminals.
Modified genAPEnclosedBoundary to enable the exploration of downward vias (checking up to 3 layers below) specifically for BTerms. This resolves connectivity issues for pins on upper metal layers that cannot access higher layers (e.g., in 3D-IC integration).

[osamahammad21]

@jferreiraOpenRoad Could you add a test case with one bterm that needs access from below?

[jferreiraOpenRoad]

https://drive.google.com/file/d/1bNeVX3PoVS3VcCSjwQGLioJHcPKss-5h/view?usp=sharing

You need to modify the OpenRoad binary path. ../Godfort/build/gcd/job0/route.detailed/0/replay.sh

[github-actions]

clang-tidy review says "All clean, LGTM! 👍"

[github-actions]

clang-tidy review says "All clean, LGTM! 👍"