rmp: Simulated annealing #8437

kbieganski · 2025-09-25T16:47:42Z

New resynthesis strategy for rmp.
It searches for an ABC script that improves timing via simulated annealing.
The script is a series of operations on ABC's internal AIG data structure.
A neighboring solution is a script with one operation added, removed, or two operations swapped.
The optimization function is defined as the worst slack.

github-actions · 2025-09-25T16:58:57Z

clang-tidy review says "All clean, LGTM! 👍"

github-actions · 2025-09-25T17:04:41Z

clang-tidy review says "All clean, LGTM! 👍"

maliberty · 2025-09-25T22:03:10Z

Please share results showing the effectiveness of this change.

kbieganski · 2025-09-29T10:39:38Z

Some results are in the tests: here, here, and here.

github-actions · 2025-09-29T10:40:27Z

clang-tidy review says "All clean, LGTM! 👍"

povik

One request for rewording the documentation. The rest are nits or have to do with pre-existing code

src/rmp/README.md

povik · 2025-09-29T11:02:55Z

src/rmp/README.md

+| `-corner` | Process corner to use. |
+| `-slack_threshold` | Specifies a (setup) timing slack value below which timing paths need to be analyzed for restructuring. The default value is `0`, and the allowed values are floats `[0, MAX_FLOAT]`. |
+| `-seed` | Seed to use for randomness in simulated annealing. |
+| `-temp` | Initial temperature for simulated annealing. The default is the required of the worst slack path. |


Is this the required arrival time on the endpoint for the worst slack path?

Based on the code it is. I think it would be good to say it's the required time on the endpoint in the description.

It think users will find required time to be a strange way to express their intent.

@maliberty They're specifying temperature but the default value for it is taken from the required time at the end of the worst path.

povik · 2025-09-29T11:08:55Z

src/rmp/src/utils.cpp

+  for (sta::Vertex* vertex : *sta->endpoints()) {
+    sta::Pin* pin = vertex->pin();
+    sta::PortDirection* direction = network->direction(vertex->pin());
+    if (!direction->isInput()) {


What's the reason for this check? I think we disqualify top module outputs here.

I see this is code moved from elsewhere but still I wonder about the purpose of the check

@QuantamHD Do you know?

I believe this was to defend against endpoints that are unconstrained. It's probably overly simplistic.

It's not the right check

src/rmp/src/annealing_strategy.cpp

github-actions · 2025-09-30T18:07:06Z

clang-tidy review says "All clean, LGTM! 👍"

github-actions · 2025-09-30T18:16:48Z

clang-tidy review says "All clean, LGTM! 👍"

github-actions · 2025-09-30T21:31:08Z

clang-tidy review says "All clean, LGTM! 👍"

povik

Codewise this looks fine

src/rmp/README.md

github-actions · 2025-10-01T16:27:22Z

clang-tidy review says "All clean, LGTM! 👍"

github-actions · 2025-10-01T16:58:55Z

clang-tidy review says "All clean, LGTM! 👍"

github-actions · 2025-10-01T17:04:15Z

clang-tidy review says "All clean, LGTM! 👍"

github-actions · 2025-10-01T19:55:59Z

clang-tidy review says "All clean, LGTM! 👍"

maliberty · 2025-10-03T05:36:02Z

Please merge master which should fix the test failure

maliberty · 2025-10-03T05:38:26Z

src/rmp/src/annealing_strategy.cpp

+
+  // GIA ops as lambdas
+  // All the magic numbers are defaults from abc/src/base/abci/abc.c
+  // Or from the ORFS qbc_speed script


Suggested change

// Or from the ORFS qbc_speed script

// Or from the ORFS abc_speed script

maliberty · 2025-10-03T05:43:28Z

src/rmp/src/annealing_strategy.cpp

+      worse_iters = 0;
+    }
+
+    if (i > 0 && (i + 1) % 10 == 0) {


Redundant - if i == 0 then (i + 1) % 10 != 0 so no need to check i > 0

Signed-off-by: Krzysztof Bieganski <[email protected]>

I give up Signed-off-by: Krzysztof Bieganski <[email protected]>

Signed-off-by: Krzysztof Bieganski <[email protected]>

Signed-off-by: Ryszard Rozak <[email protected]>

github-actions · 2025-10-03T11:39:49Z

clang-tidy review says "All clean, LGTM! 👍"

github-actions · 2025-10-03T11:45:29Z

clang-tidy review says "All clean, LGTM! 👍"

maliberty · 2025-10-03T14:32:02Z

I didn't want to hold this PR up further but I think it would be good to add and eqy step to prove equivalence.

RRozak · 2025-10-08T08:53:57Z

Could you tell how it should be added? I wanted to add similar commands to tests we added in this PR and put the annealing between them:

OpenROAD/src/rsz/test/eqy_repair_setup2.tcl

Lines 14 to 15 in 75f3458

    
           write_verilog_for_eqy eqy_repair_setup2 before "None" 
        
           run_equivalence_test eqy_repair_setup2 ./Nangate45/work_around_yosys/ "None"

Unfortunately it hangs. I also tried to run eqy_repair_setup2.tcl with EQUIVALENCE_CHECK env variable set to 1 and it didn't complete in 17h.

maliberty · 2025-10-08T16:29:20Z

I have no expertise with those tools and find them obscure. I suggest asking on their GH for help.

kbieganski force-pushed the rmp-simulated-annealing branch from d320176 to 7024860 Compare September 25, 2025 16:53

maliberty requested a review from povik September 25, 2025 22:02

povik self-assigned this Sep 26, 2025

kbieganski force-pushed the rmp-simulated-annealing branch from 7024860 to 56cddd0 Compare September 29, 2025 10:28

povik reviewed Sep 29, 2025

View reviewed changes

kbieganski force-pushed the rmp-simulated-annealing branch from 88ddf1e to 5f4ef6b Compare September 30, 2025 21:20

povik approved these changes Oct 1, 2025

View reviewed changes

src/rmp/README.md Outdated Show resolved Hide resolved

kbieganski force-pushed the rmp-simulated-annealing branch 2 times, most recently from 725c336 to 41cbfd0 Compare October 1, 2025 16:51

kbieganski force-pushed the rmp-simulated-annealing branch from 41cbfd0 to 3383d74 Compare October 1, 2025 19:45

maliberty requested changes Oct 3, 2025

View reviewed changes

kbieganski added 6 commits October 3, 2025 11:47

rmp: Simulated annealing on ABC script

50619e7

Signed-off-by: Krzysztof Bieganski <[email protected]>

rmp: Factor out common code

b94a5d4

Signed-off-by: Krzysztof Bieganski <[email protected]>

rmp: Annealing - add docs

dba9367

Signed-off-by: Krzysztof Bieganski <[email protected]>

rmp: Annealing - always keep the best result

53171b1

Signed-off-by: Krzysztof Bieganski <[email protected]>

rmp: Annealing - More verbose logs

b1fd826

Signed-off-by: Krzysztof Bieganski <[email protected]>

rmp: Annealing - Address review

01e674a

Signed-off-by: Krzysztof Bieganski <[email protected]>

kbieganski and others added 8 commits October 3, 2025 11:47

rmp: Annealing - fix command name

8aa0341

Signed-off-by: Krzysztof Bieganski <[email protected]>

rmp: Annealing - Add reverts

41a396d

Signed-off-by: Krzysztof Bieganski <[email protected]>

rmp: Remove unnecessary headers

25d0252

Signed-off-by: Krzysztof Bieganski <[email protected]>

rmp: Format code

b58e800

Signed-off-by: Krzysztof Bieganski <[email protected]>

rmp: Allow redundant declarations in ABC headers

f7d0d35

I give up Signed-off-by: Krzysztof Bieganski <[email protected]>

rmp: Fix Bazel tests

de23f1a

Signed-off-by: Krzysztof Bieganski <[email protected]>

rmp: Annealing - fix typo

3474f90

Signed-off-by: Krzysztof Bieganski <[email protected]>

Remove redundant condition

39e886e

Signed-off-by: Ryszard Rozak <[email protected]>

RRozak force-pushed the rmp-simulated-annealing branch from 3383d74 to 39e886e Compare October 3, 2025 11:29

Fix typo

2389ab5

Signed-off-by: Ryszard Rozak <[email protected]>

maliberty approved these changes Oct 3, 2025

View reviewed changes

maliberty merged commit 7b5fbf8 into The-OpenROAD-Project:master Oct 3, 2025
13 checks passed

RRozak deleted the rmp-simulated-annealing branch October 6, 2025 06:52

	// Or from the ORFS qbc_speed script
	// Or from the ORFS abc_speed script

rmp: Simulated annealing #8437

rmp: Simulated annealing #8437

Uh oh!

Conversation

kbieganski commented Sep 25, 2025

Uh oh!

github-actions bot commented Sep 25, 2025

Uh oh!

github-actions bot commented Sep 25, 2025

Uh oh!

maliberty commented Sep 25, 2025

Uh oh!

kbieganski commented Sep 29, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

github-actions bot commented Sep 29, 2025

Uh oh!

povik left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

github-actions bot commented Sep 30, 2025

Uh oh!

github-actions bot commented Sep 30, 2025

Uh oh!

github-actions bot commented Sep 30, 2025

Uh oh!

povik left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

github-actions bot commented Oct 1, 2025

Uh oh!

github-actions bot commented Oct 1, 2025

Uh oh!

github-actions bot commented Oct 1, 2025

Uh oh!

github-actions bot commented Oct 1, 2025

Uh oh!

maliberty commented Oct 3, 2025

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

github-actions bot commented Oct 3, 2025

Uh oh!

github-actions bot commented Oct 3, 2025

Uh oh!

Uh oh!

maliberty commented Oct 3, 2025

Uh oh!

RRozak commented Oct 8, 2025

Uh oh!

maliberty commented Oct 8, 2025

kbieganski commented Sep 29, 2025 •

edited

Loading