rsz: Global sizing by erendn · Pull Request #10599 · The-OpenROAD-Project/OpenROAD

erendn · 2026-06-05T20:06:28Z

Summary

This PR introduces a new repair timing policy, Lagrangian relaxation-based global sizing.

Type of Change

New feature

Impact

The new global sizing policy is disabled by default now. Initial QoR experiments show promising results when enabled with repair_timing -phases "GLOBAL_SIZING LEGACY LAST_GASP".

Verification

I have verified that the local build succeeds (./etc/Build.sh).
I have run the relevant tests and they pass.
My code follows the repository's formatting guidelines.
I have included tests to prevent regressions.
I have signed my commits (DCO).

Signed-off-by: Eren Dogan <erendogan@google.com>

gemini-code-assist

Code Review

This pull request introduces a Lagrangian-Relaxation-driven global sizing and Vt assignment optimization phase (GLOBAL_SIZING) to the resizer. It adds the LRSubproblem class to evaluate per-gate Lagrangian subproblems concurrently and the GlobalSizingPolicy class to drive the outer optimization loop. It also introduces worker-safe overloads for gate delay calculations in Resizer to support parallel evaluation. The review comments correctly identify critical null-pointer dereference risks in both LRSubproblem::isDataArc and GlobalSizingPolicy::isDataArc where edge->role() can be null.

gemini-code-assist · 2026-06-05T20:08:46Z

+  const sta::TimingRole* role = edge->role();
+  if (role->isTimingCheck()) {
+    return false;
+  }


The edge->role() pointer can be nullptr for standard combinational arcs that do not have a specific timing role. Calling role->isTimingCheck() without a null check will cause a segmentation fault. Please add a null check for role before calling isTimingCheck().

Suggested change

const sta::TimingRole* role = edge->role();

if (role->isTimingCheck()) {

return false;

}

const sta::TimingRole* role = edge->role();

if (role != nullptr && role->isTimingCheck()) {

return false;

}

gemini-code-assist · 2026-06-05T20:08:46Z

+  const sta::TimingRole* role = edge->role();
+  if (role->isTimingCheck()) {
+    return false;
+  }


The edge->role() pointer can be nullptr for standard combinational arcs that do not have a specific timing role. Calling role->isTimingCheck() without a null check will cause a segmentation fault. Please add a null check for role before calling isTimingCheck().

Suggested change

const sta::TimingRole* role = edge->role();

if (role->isTimingCheck()) {

return false;

}

const sta::TimingRole* role = edge->role();

if (role != nullptr && role->isTimingCheck()) {

return false;

}

erendn · 2026-06-05T20:57:37Z

@QuantamHD @mguthaus FYI

Signed-off-by: Eren Dogan <erendogan@google.com>

maliberty · 2026-06-06T03:08:27Z

I've merged with master and started a secure CI where the repair_timing in cts.tcl uses the above phases.

maliberty · 2026-06-06T03:08:53Z

@codex review

chatgpt-codex-connector

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 4c3e619379

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".

chatgpt-codex-connector · 2026-06-06T03:15:56Z

+        = r.best_is_downsize ? downsize_accept_tol : upsize_accept_tol;
+    if (r.best_cost < r.baseline_cost * (1.0f - tol)) {
+      sta::LibertyCell* prev = network_->libertyCell(r.inst);
+      if (subproblem_->applyReplacement(r.inst, r.best_cell)) {


Re-check DRC before applying stale Jacobi decisions

When a driver and one of its fanout loads are both resized in the same Jacobi sweep, each GateDecision was DRC-filtered against the pre-sweep snapshot, but this serial apply step installs the cell without revalidating against the live design after earlier replacements. A load upsized earlier in the loop can increase the driver's actual output load after the driver's snapshot passed checkOutputMaxCap/slew, so GLOBAL_SIZING can leave max-cap/max-slew violations that repair_timing does not check at finish.

Useful? React with 👍 / 👎.

chatgpt-codex-connector · 2026-06-06T03:15:56Z

+        = r.best_is_downsize ? downsize_accept_tol : upsize_accept_tol;
+    if (r.best_cost < r.baseline_cost * (1.0f - tol)) {
+      sta::LibertyCell* prev = network_->libertyCell(r.inst);
+      if (subproblem_->applyReplacement(r.inst, r.best_cell)) {


Stop applying upsizes once max utilization is reached

For repair_timing -setup -max_utilization ... -phases GLOBAL_SIZING, this loop can keep accepting area-increasing replacements without any resizer_.overMaxArea() guard or rollback. Other setup policies stop when the area limit is hit, but here the phase may run past the requested utilization and then fail in OptimizationPolicy::finish() with RSZ-0025, even though a partial set of replacements could have respected the user limit.

Useful? React with 👍 / 👎.

maliberty · 2026-06-06T11:27:07Z

The two significant issues in the public CI are cva6/asap7 crashed:

[2026-06-06T04:06:53.732Z] /usr/include/c++/11/bits/stl_vector.h:1045: std::vector<_Tp, _Alloc>::reference std::vector<_Tp, _Alloc>::operator[](std::vector<_Tp, _Alloc>::size_type) [with _Tp = float; _Alloc = std::allocator<float>; std::vector<_Tp, _Alloc>::reference = float&; std::vector<_Tp, _Alloc>::size_type = long unsigned int]: Assertion '__n < this->size()' failed.

[2026-06-06T04:06:53.732Z] Signal 6 received

[2026-06-06T04:06:53.732Z] Stack trace:

[2026-06-06T04:06:53.732Z]  0# 0x0000000000EDBC37 in /OpenROAD-flow-scripts/tools/install/OpenROAD/bin/openroad

[2026-06-06T04:06:53.732Z]  1# 0x0000000000042520 in /lib/x86_64-linux-gnu/libc.so.6

[2026-06-06T04:06:53.732Z]  2# pthread_kill in /lib/x86_64-linux-gnu/libc.so.6

[2026-06-06T04:06:53.732Z]  3# raise in /lib/x86_64-linux-gnu/libc.so.6

[2026-06-06T04:06:53.732Z]  4# abort in /lib/x86_64-linux-gnu/libc.so.6

[2026-06-06T04:06:53.732Z]  5# 0x0000000000ED1418 in /OpenROAD-flow-scripts/tools/install/OpenROAD/bin/openroad

[2026-06-06T04:06:53.732Z]  6# rsz::GlobalSizingPolicy::projectFlowBalance(rsz::LRParams const&) in /OpenROAD-flow-scripts/tools/install/OpenROAD/bin/openroad

[2026-06-06T04:06:53.732Z]  7# rsz::GlobalSizingPolicy::iterate() in /OpenROAD-flow-scripts/tools/install/OpenROAD/bin/openroad

[2026-06-06T04:06:53.732Z]  8# rsz::Optimizer::run() in /OpenROAD-flow-scripts/tools/install/OpenROAD/bin/openroad

[2026-06-06T04:06:53.732Z]  9# rsz::Resizer::repairSetup(double, double, int, int, int, bool, bool, std::vector<rsz::MoveType, std::allocator<rsz::MoveType> > const&, char const*, bool, bool, bool, bool, bool, bool, bool, bool) in /OpenROAD-flow-scripts/tools/install/OpenROAD/bin/openroad

and ibex/ihp fails in legalization

 [ERROR DPL-0036] Detailed placement failed inside DPL.

To reproduce, use your branch with

diff --git a/flow/scripts/cts.tcl b/flow/scripts/cts.tcl
index 1378c9e7b9..e9cdaea7e5 100644
--- a/flow/scripts/cts.tcl
+++ b/flow/scripts/cts.tcl
@@ -66,7 +66,7 @@ if { !$::env(SKIP_CTS_REPAIR_TIMING) } {
     write_lec_verilog 4_before_rsz_lec.v
   }
 
-  repair_timing_helper
+  repair_timing_helper -phases "GLOBAL_SIZING LEGACY LAST_GASP"

maliberty · 2026-06-06T11:44:04Z

Public results are mixed

maliberty · 2026-06-06T11:46:22Z

Even at CTS where the change is applied

rsz: Add global sizing

9cf00eb

Signed-off-by: Eren Dogan <erendogan@google.com>

github-actions Bot added the size/XL label Jun 5, 2026

gemini-code-assist Bot reviewed Jun 5, 2026

View reviewed changes

rsz: clang-tidy

4c3e619

Signed-off-by: Eren Dogan <erendogan@google.com>

chatgpt-codex-connector Bot reviewed Jun 6, 2026

View reviewed changes

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

rsz: Global sizing#10599

rsz: Global sizing#10599
erendn wants to merge 2 commits into
The-OpenROAD-Project:masterfrom
erendn:global_sizing

erendn commented Jun 5, 2026

Uh oh!

gemini-code-assist Bot left a comment

Uh oh!

gemini-code-assist Bot Jun 5, 2026

Uh oh!

gemini-code-assist Bot Jun 5, 2026

Uh oh!

erendn commented Jun 5, 2026

Uh oh!

maliberty commented Jun 6, 2026

Uh oh!

maliberty commented Jun 6, 2026

Uh oh!

chatgpt-codex-connector Bot left a comment

Uh oh!

chatgpt-codex-connector Bot Jun 6, 2026

Uh oh!

chatgpt-codex-connector Bot Jun 6, 2026

Uh oh!

maliberty commented Jun 6, 2026

Uh oh!

maliberty commented Jun 6, 2026

Uh oh!

maliberty commented Jun 6, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Conversation

erendn commented Jun 5, 2026

Summary

Type of Change

Impact

Verification

Uh oh!

gemini-code-assist Bot left a comment

Choose a reason for hiding this comment

Code Review

Uh oh!

gemini-code-assist Bot Jun 5, 2026

Choose a reason for hiding this comment

Uh oh!

gemini-code-assist Bot Jun 5, 2026

Choose a reason for hiding this comment

Uh oh!

erendn commented Jun 5, 2026

Uh oh!

maliberty commented Jun 6, 2026

Uh oh!

maliberty commented Jun 6, 2026

Uh oh!

chatgpt-codex-connector Bot left a comment

Choose a reason for hiding this comment

💡 Codex Review

Uh oh!

chatgpt-codex-connector Bot Jun 6, 2026

Choose a reason for hiding this comment

Uh oh!

chatgpt-codex-connector Bot Jun 6, 2026

Choose a reason for hiding this comment

Uh oh!

maliberty commented Jun 6, 2026

Uh oh!

maliberty commented Jun 6, 2026

Uh oh!

maliberty commented Jun 6, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants