grt: cache per-layer wire resistance

[rafaelmoresco]

Summary

Chached dbuToMicron and layer resistance to avoid unecessary calls during res aware operations. In a VTune test these calls were costing a bit over 10% of Fast Route time

Type of Change

• Refactoring

Impact

Reduce Res Aware runtime.

Verification

• I have verified that the local build succeeds (./etc/Build.sh).
• I have run the relevant tests and they pass.
• My code follows the repository's formatting guidelines.
• I have signed my commits (DCO).

[gemini-code-assist]

Code Review

This pull request optimizes performance in FastRouteCore by caching dbu_per_micron_ and layer_res_per_micron_ during tech layer preprocessing, reducing database lookups in getWireResistance and dbuToMicrons. The review feedback highlights several critical safety concerns, including potential out-of-bounds vector access, null pointer dereferences, and division-by-zero vulnerabilities in the new caching logic.

[gemini-code-assist]

Accessing layer_res_per_micron_[layer] without verifying that layer is within the valid bounds of the vector can lead to undefined behavior or segmentation faults. Adding bounds checking ensures safety.

Suggested change

	if (layer < net->getMinLayer() || layer > net->getMaxLayer()) {
	return BIG_INT;
	}
	if (layer < 0 || layer >= layer_res_per_micron_.size() || layer < net->getMinLayer() || layer > net->getMaxLayer()) {
	return BIG_INT;
	}

[gemini-code-assist]

Several potential issues exist here:

1. getTechLayer could return nullptr.
2. ndr->getLayerRule(db_layer) can return nullptr if the NDR does not define a rule for the given layer.
3. layer_width could be 0.0f, leading to a division-by-zero error.

Adding defensive checks for these cases prevents potential crashes and undefined behavior.

  // If net has NDR, get the correct width value
  odb::dbTechNonDefaultRule* ndr = net->getDbNet()->getNonDefaultRule();
  if (ndr != nullptr) {
    odb::dbTechLayer* db_layer = getTechLayer(layer, false);
    if (db_layer != nullptr) {
      odb::dbTechLayerRule* layerRule = ndr->getLayerRule(db_layer);
      if (layerRule != nullptr) {
        const float layer_width = dbuToMicrons(layerRule->getWidth());
        if (layer_width > 0.0f) {
          res_ohm_per_micron = db_layer->getResistance() / layer_width;
        }
      }
    }
  }

[gemini-code-assist]

If layer_width is 0.0f (e.g., due to an invalid or uninitialized layer width in the tech file), dividing by it will result in a division-by-zero error. It is safer to guard against this by checking if layer_width is greater than zero.

Suggested change

	// Keep layer_width as float: other rounding perturbs getWireCost() results
	const float layer_width = dbuToMicrons(db_layer->getWidth());
	layer_res_per_micron_[layer] = db_layer->getResistance() / layer_width;
	// Keep layer_width as float: other rounding perturbs getWireCost() results
	const float layer_width = dbuToMicrons(db_layer->getWidth());
	layer_res_per_micron_[layer] = layer_width > 0.0f ? db_layer->getResistance() / layer_width : 0.0f;

[gemini-code-assist]

If dbu_per_micron_ is 0.0 (e.g., if dbuToMicrons is called before preProcessTechLayers initializes it, or if the database tech units are uninitialized), this will result in a division-by-zero error. Falling back to the original database-level implementation when dbu_per_micron_ is 0.0 provides a safe and robust fallback.

double FastRouteCore::dbuToMicrons(const int dbu)
{
  if (dbu_per_micron_ == 0.0) {
    return db_->getChip()->getBlock()->dbuToMicrons(dbu);
  }
  return dbu / dbu_per_micron_;
}