Repository navigation
Conversation
The macro grid halo (2.0) equaled MACRO_ROWS_HALO_X/Y (2). Rows snap down to the site grid, so the followpin on the row edge next to a macro straddles the halo boundary and every via onto it is obstructed by the macro grid. When that row sits entirely under the macro (a pocket between macros, as in swerv_wrapper) the rail cannot be connected anywhere and pdngen fails with PDN-0179. Shrink the macro grid halo to 1.9 so it stays inside the row halo. Signed-off-by: Matt Liberty <mliberty@precisioninno.com>
Codex Review SummaryThis comment shows the latest Codex review activity on this pull request.
ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings. |
There was a problem hiding this comment.
Code Review
This pull request reduces the macro grid halo values from 2.0 to 1.9 in the OpenROAD PDN grid strategy script for ASAP7 to prevent blocking the followpin and vias on the row edge next to a macro. The review feedback recommends dynamically calculating these halo values from the MACRO_ROWS_HALO_X and MACRO_ROWS_HALO_Y environment variables rather than hardcoding them, ensuring the script remains robust if those variables are customized.
🔍 QoR checkMetrics reflect the PR merge build — i.e. what will land on the target branch. Commit 62 design(s) checked — 0 with regression(s), 0 without a comparable baseline. |
|
all designs passed, pr-merge failed in reporting. |
0b6c450
into
The-OpenROAD-Project:master
Problem
asap7/swerv_wrapperfails in2_4_floorplan_pdn:Cause
The asap7 macro grid halo (2.0 um) equals
MACRO_ROWS_HALO_X/Y(2 um).u38955(bottom edge y=76.256), the top row ends at y=74.25.The VDD followpin on that row edge (74.241–74.259) crosses into the claimed area. Any M2→M5 via stack on it (about 74.226–74.273) is rejected as obstructed. Normally the rail runs on past the macro and gets a via there. Here the pocket between
u49927andu49934is narrower thanu38955, so no via position is left and the rail stays floating. The repair channel is stretched up over the macro because it reaches for the next M6 VDD strap. That stretch is a symptom, not the cause.Fix
Change the macro grid halo to 1.9 um so it stays inside the 2 um row halo.
With the change, swerv_wrapper's
do-2_4_floorplan_pdnpasses, and its remaining PDN warnings are a subset of the ones it gave before.This script is the default asap7
PDN_TCL, so coralnpu, cva6 and riscv32i are affected as well. Their metrics may move.