-
Notifications
You must be signed in to change notification settings - Fork 551
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Substantial slowdown during resizer invocation post-grt #4121
Comments
Is this something that used to work? We've always considered post-grt rsz to be in development. It is not enabled in ORFS. |
If I add -verbose I can see it is progressing slowly. |
We've had post-GRT resizing for as long as I remember, and it worked fine as recently as 0a6d0fd |
Is there any timeframe for an investigation into this? It's been a very useful feature; we didn't know it was in development and we've been using the resizer post-GRT since August 2021 The-OpenROAD-Project/OpenLane#531 |
Its in the queue but I haven't started to look at it yet. |
The older version does run faster. It is a substantial slowdown rather than a hang from what I see. The window is two months wide. Do you have any narrower scope of the problem? There have been a lot of a lot changes in that time. |
@luis201420 @eder-matheus any guesses on the source of this slowdown? |
git bisect shows: @arthurjolo thoughts on the performance implications of this change? |
I verified that if I change the default critical_nets_percentage_ to 0 at the head of master then it runs much faster. |
If the design runs a large number of iteration to remove overflow the run time should have a large increase, since it is updating the slack for most of the nets. |
@arthurjolo this is happening from inside repair_design |
@maliberty I haven't seen much of repair_design, I will take a look into it. |
repair_design will make some netlist edits and then call for updated parasitics (Resizer::ensureWireParasitic). Since we are post-grt that will call for incremental global routing on the modified nets. In that tight inner loop I don't think we want to any timing updates. |
Fixes The-OpenROAD-Project#4121 Signed-off-by: Matt Liberty <mliberty@precisioninno.com>
Fixes The-OpenROAD-Project#4121 Signed-off-by: Matt Liberty <mliberty@precisioninno.com>
Thank you so much for the fast turnaround on this- appreciated |
* Update OpenROAD to `0a584d1` * Fixes issue where post-GRT resizing run-time and memory consumption got out of hand: see The-OpenROAD-Project/OpenROAD#4121 (comment) for one example * Disabled latch linting for `salsa20` * Enable post-GRT resizer for `aes_core`, `spm` and `aes` * Added hidden debug options `--force-design-dir`, `--force-run-dir`
Describe the bug
CI printed
2023-10-11T00:27:13.3062451Z [INFO RSZ-0058] Using max wire length 4459um.
then timed outExpected Behavior
Resizer step should pass
Environment
bdc8e94
To Reproduce
https://drive.google.com/file/d/1Q1IQOyUUbxX26my9kEVG9M7-MK7qs4zt/view?usp=sharing
Untar then run
./run.sh
Relevant log output
Screenshots
No response
Additional Context
No response
The text was updated successfully, but these errors were encountered: