drt: getUpdatedXform(false) elimination #6141
Conversation
Signed-off-by: bernardo <[email protected]>
![github-actions[bot]](https://avatars.githubusercontent.com/in/15368?s=60&v=4){kind=small}
Signed-off-by: bernardo <[email protected]>
bnmfw
force-pushed
the
drt_dbTransform_refactor
branch
from
f68ddee to
2200928
Compare
|
clang-tidy review says "All clean, LGTM! 👍" |
|
clang-tidy review says "All clean, LGTM! 👍" |
bnmfw
force-pushed
the
drt_dbTransform_refactor
branch
from
1e9a755 to
c87a7dd
Compare
|
clang-tidy review says "All clean, LGTM! 👍" |
| @@ -94,6 +95,15 @@ class frInst : public frRef | |||
| void setOrigin(const Point& tmpPoint) override { xform_.setOffset(tmpPoint); } | |||
| dbTransform getTransform() const override { return xform_; } | |||
There was a problem hiding this comment.
Is this used anywhere now?
There was a problem hiding this comment.
Unfortunatly, yes. getTransform() itself is used in the now renamed getNoRotationTransform() and in some other cases. Both setOrigin() and setOrient() still see some uses. I intend to completely eliminate these two functions, but I'm not sure it is possible. (I'm not 100% sure about what I'm about to say) During Pin Access virtual vias have to be created to check for DRVs when considering if a Access Point is valid. These vias exists only on the drt, and don't have an odb counterpart so during checking they need to use and configure the drt Transform.
…en necessary Signed-off-by: bernardo <[email protected]>
Signed-off-by: bernardo <[email protected]>
Signed-off-by: bernardo <[email protected]>
bnmfw
force-pushed
the
drt_dbTransform_refactor
branch
2 times, most recently
from
bc8bf0f to
1bd92bf
Compare
Signed-off-by: bernardo <[email protected]>
|
clang-tidy review says "All clean, LGTM! 👍" |
|
clang-tidy review says "All clean, LGTM! 👍" |
Signed-off-by: bernardo <[email protected]>
|
clang-tidy review says "All clean, LGTM! 👍" |
This PR is ISPD and Secure CI Safe.
The problem
This PR solves part of the conflicts presented in issue #5712. In
io::Parser::setInst()the Transform object fromodb::dbInstis copied and held as a private attribute ofdrt::frInst. This new Transform is also modified to have its origin at the lower left corner, which is not always accurate. Furthermore, an auxiliary function,getUpdatedXform, exists to undo this origin transformation, trying to get back to the original odb Transform. Unfortunately, even this strategy is not error proof, as this patch to get back to the odb inst in done on the privatedrt::frInstwhich might become obsolete if the instance is moved, as the update in the Transform to get the new cell location is not precise.Solutions
getDBTransformandsetDBInstwere created. WhenevergetUpdatedXform()is called,getDBTransformis used as that is what the original function is trying to do. Once the privatedrt::frInstTransform is elimintedgetDBTransformwill be renamed to become simplygetTransform.getUpdatedXform(true)use cases are not eliminated in this PR.DesignCallBack::inDbPreMoveInstwas included to correctly "move" the instance in the R Tree strucute.The elimination of
getUpdatedXform()cases allowed for the elimination ofupdateXform()and simplifications of functions that used it to simply usegetDBTransform().Furthermore,
getUpdatedXform()was refactored to eliminate the now unused functionality. This reduced the function to one that simply gets the Transformation and sets the rotation as R0.getUpdatedXform()was renamed togetNoRotationTransform()accordingly (this is temporary, the objective is to eliminate the function altogether).In a lot of cases I noticed that the code manually got the Transformation and set its rotation to R0. All this cases were substituted by
getNoRotationTransform(). This might help eliminating thedrt::InstTransform, as it reduces many uses to a single function.frInstTerm::getShapesandfrInstTerm::getBBox()had dead code eliminated. Both functions had a mode in which they used thedrt::InstTransform, and a mode where they used the oldgetUpdatedXform()but only used the latter. They were refactored to only have the second case and no alternative mode.Finally some redundant code was eliminated regarding Transform manipulation. When a transform is created it always defaults to having the origin at 0,0 and rotation as R0, specifying that upon creation of the Transform is redundant. Also, ideally there wont even be a way to set this parameters in the future, as odb will handle the Transform, so the less the drt sets this parameters the better.
Fixture::makeInstwas simplified to avoid these redundancies.gc_Test Patch
Now that the creation of an inst requires the association to a
odb::Instsome of the tests on gc_test had to be patched to accommodate this initialization.Fixture::createDummyInstwas created to accommodate for this requirement. The only required part of the instance is the transform which is initialized,