gpl: extend max iter cap by routability internal iterations#7772
Conversation
…s performed during routability modify the final iteration on the log to match existing table include report of gpl iteartions during routability include warning to the maximum iterations reached, probable undetected divergence some variables renaming Signed-off-by: Augusto Berndt <augusto.berndt@precisioninno.com>
Signed-off-by: Augusto Berndt <augusto.berndt@precisioninno.com>
Signed-off-by: Augusto Berndt <augusto.berndt@precisioninno.com>
Signed-off-by: Augusto Berndt <augusto.berndt@precisioninno.com>
Signed-off-by: Augusto Berndt <augusto.berndt@precisioninno.com>
Signed-off-by: Augusto Berndt <augusto.berndt@precisioninno.com>
|
clang-tidy review says "All clean, LGTM! 👍" |
Signed-off-by: Augusto Berndt <augusto.berndt@precisioninno.com>
|
clang-tidy review says "All clean, LGTM! 👍" |
Signed-off-by: Augusto Berndt <augusto.berndt@precisioninno.com>
Signed-off-by: Augusto Berndt <augusto.berndt@precisioninno.com>
|
clang-tidy review says "All clean, LGTM! 👍" |
1 similar comment
|
clang-tidy review says "All clean, LGTM! 👍" |
Signed-off-by: Augusto Berndt <augusto.berndt@precisioninno.com>
|
clang-tidy review says "All clean, LGTM! 👍" |
Signed-off-by: Augusto Berndt <augusto.berndt@precisioninno.com>
Signed-off-by: Augusto Berndt <augusto.berndt@precisioninno.com>
|
clang-tidy review says "All clean, LGTM! 👍" |
|
clang-tidy review says "All clean, LGTM! 👍" |
Signed-off-by: Augusto Berndt <augusto.berndt@precisioninno.com>
Signed-off-by: Augusto Berndt <augusto.berndt@precisioninno.com>
|
clang-tidy review says "All clean, LGTM! 👍" |
|
CI failed only on bazel with: |
|
The real issue is mock-array The "too big" is more minor. |
…o multiple conflicts Signed-off-by: Augusto Berndt <augusto.berndt@precisioninno.com>
|
clang-tidy review says "All clean, LGTM! 👍" |
1 similar comment
|
clang-tidy review says "All clean, LGTM! 👍" |
Signed-off-by: Augusto Berndt <augusto.berndt@precisioninno.com>
this test falls into a corner case because it starts with low overflow Signed-off-by: Augusto Berndt <augusto.berndt@precisioninno.com>
…re off Signed-off-by: Augusto Berndt <augusto.berndt@precisioninno.com>
Signed-off-by: Augusto Berndt <augusto.berndt@precisioninno.com>
Signed-off-by: Augusto Berndt <augusto.berndt@precisioninno.com>
I realized why this is happening. I made a small logic change to detect whether routability mode is active: instead of relying solely on the overflow values, I added a check for whether the snapshot has already been saved. As a result, TD is now allowed to run even before the snapshot is saved and overflow is low. This change only affects designs that start with a low overflow (e.g., below 0.60). |
|
clang-tidy review says "All clean, LGTM! 👍" |
Signed-off-by: Augusto Berndt <augusto.berndt@precisioninno.com>
Signed-off-by: Augusto Berndt <augusto.berndt@precisioninno.com>
|
clang-tidy review says "All clean, LGTM! 👍" |
Signed-off-by: Augusto Berndt <augusto.berndt@precisioninno.com>
|
clang-tidy review says "All clean, LGTM! 👍" |
Signed-off-by: Augusto Berndt <augusto.berndt@precisioninno.com>
Signed-off-by: Augusto Berndt <augusto.berndt@precisioninno.com>
|
clang-tidy review says "All clean, LGTM! 👍" |
|
secure-CI is green in The-OpenROAD-Project/OpenROAD-flow-scripts#3355 |
Increase the maximum number of nesterov iterations by the number of iterations performed during routability mode (i.e., the section between 0.6 and 0.3 overflow, where snapshot saving and revert triggering occur). This acts as a global stopping criteria for Nesterov.
This change addresses the behavior observed in issue #7581, where a design exhibited very small but consistent reductions in congestion during routability. As a result, it performed an unusually high number of routability reverts (~30), eventually hitting the existing 5000-iteration limit.
This PR also introduces a new routability stopping criteria: a fixed cap of 50 routability reverts.
Log changes: