Repository navigation
gpl: add option to compute density subtracting macro area from the denominator - #12111
Conversation
Signed-off-by: Arthur Koucher <arthurkoucher@precisioninno.com>
There was a problem hiding this comment.
Code Review
This pull request introduces the -unplaced_macros_as_fixed option to the get_global_placement_uniform_density command, allowing unplaced macros to be treated as fixed instances during density calculations. However, several issues were identified in the review: the new flag is parsed in Tcl but not passed to the underlying C++ implementation, making it ignored; unplaced macros assigned to specific regions are incorrectly excluded from those regions' area calculations; and the regression test passes coincidentally without actually validating the new flag's functionality.
Signed-off-by: Arthur Koucher <arthurkoucher@precisioninno.com>
|
The wording of the feature is not exactly accurate. I'm marking this as a draft and will ping when it's ready to review. |
Signed-off-by: Arthur Koucher <arthurkoucher@precisioninno.com>
Signed-off-by: Arthur Koucher <arthurkoucher@precisioninno.com>
|
@LucasYuki Ready for review. |
Summary
Currently in ORFS we use GPL's uniform density as "std cell utilization" for MPL (which is just an odd name for std cell density). As GPL is capable of placing every cell in the design including the macros, the uniform density is computed accordingly i.e., as if it were going to place them:
The fact that we use such density for std cells in MPL causes a discrepancy between the utilization that MPL and GPL see, because, for MPL, the density should refer strictly to the std cells. That means the area of the macros must be treated as area that is not movable by GPL when computing the uniform density. We achieve that by simply subtracting the macro area from the denominator:
Note that there is already an innate discrepancy between the density that these tools see, because MPL precisely bloats the std cell clusters with the value set by the user and GPL may not reach that target value.
Observation regarding the naming idea here: initially I thought of
-unfixed_macros_as_fixed/placed, however, that is not precise, because a placed/fixed macro has a location in the die, which is definitely not the case. After some thinking, I decided that being a bit more literal for this case helps us more than harms us.That said, this PR:
get_global_placement_uniform_densitya proper command rather than a "developer" command which I think makes sense as it is used in production on ORFS.Type of Change
Impact
None. Results will change when the option is used in ORFS.
Verification
./etc/Build.sh).Related Issues
Found when working on #11592