rsz: Fix and clean up size move preliminaries#8237
Conversation
Make sure we don't attempt to size an instance which isn't a standard cell. Also clean up the clone move condition. This condition looks to have been introduced in commit 533c037 as a workaround for undo limitations (mistakenly with flipped logic) and carried over to other places. Signed-off-by: Martin Povišer <povik@cutebit.org>
|
FYI @mguthaus @precisionmoon regarding the clone move cleanup |
|
clang-tidy review says "All clean, LGTM! 👍" |
| // We always size the cloned gates for some reason, but it would be good if we | ||
| // also down-sized here instead since we might want smaller original. | ||
| if (!resizer_->dontTouch(drvr) | ||
| || resizer_->clone_move_->hasPendingMoves(drvr)) { |
There was a problem hiding this comment.
Removing the clone_move_->hasPendingMoves check should only matter for instances with dont touch. For that reason I'm expecting none or few metric changes. I can run a secure CI if we don't want to chance it.
| if (resizer_->dontTouch(load_inst) | ||
| || !resizer_->isLogicStdCell(load_inst)) { |
There was a problem hiding this comment.
I see similar checks in various places. Perhaps you could do a refactor similar to okToBufferNet like okToResizeInstance ?
There was a problem hiding this comment.
Do you want to do this here or in another?
There was a problem hiding this comment.
I'd rather do this elsewhere. We could fold isLogicStdCell into dontTouch as I'm not sure there's any transformation we want the resizer to apply to macros.
3 participants
Make sure we don't attempt to size an instance which isn't a standard cell. Also clean up the clone move condition. This condition looks to have been introduced in commit 533c037 as a workaround for undo limitations (mistakenly with flipped logic) and carried over to other places.