Skip to content

xiangshan: the parent repairs setup at global route, 3,276 to 2,406 ps - #1212

Merged
oharboe merged 1 commit into
mainfrom
xiangshan-parent-repair-at-grt
Oct 9, 2026
Merged

oharboe merged 1 commit into
mainfrom
xiangshan-parent-repair-at-grt

Conversation

@oharboe

@oharboe oharboe commented Oct 9, 2026

Copy link
Copy Markdown
Collaborator

The parent's setup repair at global route, with the route's parasitics: 3,276 to 2,406 ps (-26.6 %), the design period with it (the slowest block is MemBlock at 1,870).

before (row 14) after (row 15)
parent reg2reg 3,276 ps 2,406 ps
global route 19 min, 83 GB 84 min, 102 GB
blocks unchanged unchanged

Why: the parent's worst paths are a mass of CtrlBlock control, not one path (a false-path peel: about 9 ps a register over 40 registers), and every repair before global route times them with estimated wires. The repair: 1,525 VT swaps, 74 buffers, 34 pin swaps. The new worst path is CtrlBlock's flush redirect into Frontend.

The change, in the parent only:

  • SKIP_INCREMENTAL_REPAIR dropped, so ORFS's grt stage repairs.
  • HOLD_SLACK_MARGIN = -1000 in the grt stage only: hold stays with CTS's repair; this stage's hold repair took an earlier run to 120 GB.

Measured on #1187's commit f681c30 (row 14's build), not on main: KPI row 15 is in flight like rows 13 and 14. A confirmation build of this branch on main is running and will add its row here.

Not tested locally beyond fix_lint.

🤖 Generated with Claude Code

The parent's worst paths are a mass of CtrlBlock control (a false-path
peel: about 9 ps a register over 40), and the repairs before global
route time them with estimated wires. With the route's parasitics the
stage's own setup repair takes the parent from 3,276 to 2,406 ps on
row 14's build (#1187 at f681c30): 1,525 VT swaps, 74 buffers, 34 pin
swaps; its worst path is now CtrlBlock's flush redirect into Frontend.
The blocks are unchanged, so the design period is the parent's, 2,406.

SKIP_INCREMENTAL_REPAIR goes from the parent; HOLD_SLACK_MARGIN -1000 in
its grt stage leaves hold to CTS's repair, as this stage's hold repair
took an earlier run of it to 120 GB. Global route costs 84 min against
19 and peaks at 102 GB against 83.

KPI row 15 is that measurement, in flight like rows 13 and 14: it was
built on #1187's commit, not on main.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Signed-off-by: Øyvind Harboe <oyvind.harboe@zylin.com>
@oharboe
oharboe merged commit 3d7f302 into main Oct 9, 2026
1 check passed
@oharboe
oharboe deleted the xiangshan-parent-repair-at-grt branch October 9, 2026 20:06
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant