DRT: Stubborn Tiles flow #6213
DRT: Stubborn Tiles flow #6213
Conversation
Signed-off-by: osamahammad21 <[email protected]>
Signed-off-by: osamahammad21 <[email protected]>
Signed-off-by: osamahammad21 <[email protected]>
Signed-off-by: osamahammad21 <[email protected]>
Signed-off-by: osamahammad21 <[email protected]>
Signed-off-by: osamahammad21 <[email protected]>
Signed-off-by: osamahammad21 <[email protected]>
Signed-off-by: osamahammad21 <[email protected]>
Signed-off-by: osamahammad21 <[email protected]>
Signed-off-by: Osama <[email protected]>
|
clang-tidy review says "All clean, LGTM! 👍" |
| IterationProgress iter_prog; | ||
| auto block = getDesign()->getTopBlock(); | ||
| const auto num_drvs = block->getNumMarkers(); | ||
| const bool stubborn_flow = num_drvs <= 11 && ripupMode != RipUpMode::ALL |
There was a problem hiding this comment.
why 11? would this be better as a function of DRCs per unit area (it seems like 11 would be a large number for small design and potentially a small number for a large design)?
| std::unique_ptr<FlexDRWorker> createWorker(int x_offset, | ||
| int y_offset, | ||
| const SearchRepairArgs& args, | ||
| Rect routeBox = Rect(0, 0, 0, 0)); |
There was a problem hiding this comment.
Take by reference? Rect() is 0,0,0,0 so no need to specify.
| bool isEqualIgnoringSizeAndOffset(const SearchRepairArgs& other) const | ||
| { | ||
| return (mazeEndIter == other.mazeEndIter | ||
| && workerDRCCost == other.workerDRCCost | ||
| && workerMarkerCost == other.workerMarkerCost | ||
| && workerFixedShapeCost == other.workerFixedShapeCost | ||
| && std::fabs(workerMarkerDecay - other.workerMarkerDecay) < 1e-6 | ||
| && ripupMode == other.ripupMode | ||
| && followGuide == other.followGuide); | ||
| } |
There was a problem hiding this comment.
Overly long for an inline in a header. Not performance critical so out-of-line.
| if (iter_ > router_cfg_->END_ITERATION) { | ||
| break; | ||
| } |
There was a problem hiding this comment.
Previously this was after searchRepair so it seems one more iteration would have been done.
| ProfileTask profile(fmt::format("DR:searchRepair{}", iter_).c_str()); | ||
|
|
||
| if (dist_on_) { | ||
| if ((iter_ % 10 == 0 && iter_ != 60) || iter_ == 3 || iter_ == 15) { |
There was a problem hiding this comment.
What does this mean? Please give some motivating logic.
| std::vector<Rect> merge_boxes; | ||
| for (auto& box : drv_boxes) { | ||
| merge_boxes.emplace_back(box); | ||
| } |
There was a problem hiding this comment.
std::vector<Rect> merge_boxes{drv_boxes}; compact and more efficient.
| for (auto it2 = it1 + 1; it2 != merge_boxes.end(); ++it2) { | ||
| Rect merge_box = (*it1); | ||
| merge_box.merge((*it2)); | ||
| if (merge_box.dx() > 4 || merge_box.dy() > 4) { |
There was a problem hiding this comment.
smaller than or equal to 5x5 GCells seems to conflict
| continue; | ||
| } | ||
| (*it1).merge((*it2)); | ||
| merge_boxes.erase(it2); |
There was a problem hiding this comment.
This invalidates both your iterators.
| const int rect_id) | ||
| { | ||
| return std::any_of(grid.begin() + rect.xMin(), | ||
| grid.begin() + rect.xMax() + 1, |
There was a problem hiding this comment.
Are you certain that this will never be beyond the end of the vector? Likewise in the inner any_of
| case frDirEnum::E: | ||
| max_idx.addX(1); |
There was a problem hiding this comment.
If you expand W you check for 0 but when expanding E you don't check for the right edge?
|
Where does this prevent abutting workers from running simultaneously? |
When the number of DRVs is less than or equal to 11 DRVs, we start this new flow where we do the following:
Improvement in ISPD designs:
Improvement in public ORFS designs (designs with no change are hidden):