[gpl] Include internal power in MBFF clustering cost#10075
Conversation
|
@mikesinouye please test this on your internal designs. I'm seeing improvement on some of our designs. |
![gemini-code-assist[bot]](https://avatars.githubusercontent.com/in/956858?s=60&v=4){kind=small}
There was a problem hiding this comment.
Code Review
This pull request integrates clock pin internal power into the MBFF cost metric, ensuring more accurate power estimation during clustering. The changes include adding a method to calculate clock pin energy, updating the cost calculation logic, and caching clock periods. I have identified a potential division-by-zero risk in the cost normalization logic, a performance concern regarding redundant STA lookups within the SetVars loop, and a violation of the style guide regarding the use of magic numbers for clock activity calculations.
|
clang-tidy review says "All clean, LGTM! 👍" |
openroad-ci
force-pushed
the
mbff-improve-power-modeling
branch
from
c194db0 to
eee27b3
Compare
maliberty
changed the title
|
clang-tidy review says "All clean, LGTM! 👍" |
|
@codex review |
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: eee27b3ea3
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
openroad-ci
force-pushed
the
mbff-improve-power-modeling
branch
2 times, most recently
from
68e9145 to
2961689
Compare
|
clang-tidy review says "All clean, LGTM! 👍" |
1 similar comment
|
clang-tidy review says "All clean, LGTM! 👍" |
openroad-ci
force-pushed
the
mbff-improve-power-modeling
branch
from
2961689 to
3f3b92b
Compare
![github-actions[bot]](https://avatars.githubusercontent.com/in/15368?s=60&v=4){kind=small}
The MBFF algorithm previously used only leakage power to decide whether to replace single-bit flip-flops with multi-bit cells. For flip-flops, internal power dominates total power, and MBFF cells share scan (SE/SI) and clock structures across bits, giving 40-90% savings on those pins. Ignoring internal power caused clustering to increase total power on some PDKs. Add getInternalEnergy() which sums average internal energy across all pins (CK, D, Q, SE, SI) from Liberty internal_power tables. Use this alongside leakage in three places: - SetRatios: norm_power_ uses total estimated power (leakage + internal_energy * clock_activity) so the ILP cost function reflects total power, not just leakage. - ReadLibs: select best tray per size by minimum total estimated power instead of minimum leakage, so cells with lower total power (e.g. SVT over LVT) are preferred even when their leakage is higher. - SetVars: select the single-bit baseline cell by lowest total estimated power, with both leakage and internal energy paired from the same cell. Clock period is obtained from SDC before ReadLibs runs so tray selection can account for internal power. Falls back to leakage-only when no clock is defined or no internal_power tables exist. Signed-off-by: Matt Liberty <mliberty@precisioninno.com>
openroad-ci
force-pushed
the
mbff-improve-power-modeling
branch
from
3f3b92b to
ae5c013
Compare
|
clang-tidy review says "All clean, LGTM! 👍" |
maliberty
merged commit aec50a4
into
The-OpenROAD-Project:master
3 participants
Summary
The MBFF algorithm previously used only leakage power to decide whether
to replace single-bit flip-flops with multi-bit cells. For flip-flops,
internal power dominates total power, and MBFF cells share scan (SE/SI)
and clock structures across bits, giving 40-90% savings on those pins.
Ignoring internal power caused clustering to increase total power on
some PDKs.
Add getInternalEnergy() which sums average internal energy across all
pins (CK, D, Q, SE, SI) from Liberty internal_power tables. Use this
alongside leakage in three places:
internal_energy * clock_activity) so the ILP cost function reflects
total power, not just leakage.
instead of minimum leakage, so cells with lower total power (e.g. SVT
over LVT) are preferred even when their leakage is higher.
power, with both leakage and internal energy paired from the same cell.
Clock period is obtained from SDC before ReadLibs runs so tray selection
can account for internal power. Falls back to leakage-only when no
clock is defined or no internal_power tables exist.
Type of Change
Verification
./etc/Build.sh).