Clusterer feasibility changes #1641
Conversation
…tersections in the clusterer
…n start_new_cluster
…ally during clustering
| @@ -164,6 +164,7 @@ enum e_block_pack_status { | |||
| BLK_PASSED, | |||
| BLK_FAILED_FEASIBLE, | |||
| BLK_FAILED_ROUTE, | |||
| BLK_FAILED_FLOORPLANNING, | |||
There was a problem hiding this comment.
think about whether this enum can be in a lower header file. If it's only used in the clusterer, can have a cluster_utilities header file and put this enum in there. This can be a cleanup on a separate pull request
There was a problem hiding this comment.
in pack subdirectory - is there a cluster utilities type header file? check, could possibly put there
…of unnecessary warning messages during clustering
… creating a cluster
|
@litghost nightly has been taking a long time on this. |
Kokoro failed to start this job correct, and it will eventually die. This is due to a kokoro bug. |
…_region is now in region class. Added a print function for vtr::Rect
…r_feasibility_changes' of https://github.com/verilog-to-routing/vtr-verilog-to-routing into clusterer_feasibility_changes
|
@litghost Does nightly seem to be running normally here? |
|
@litghost It says nightly failed but I can't see anything on the invocation details. Did kokoro fail to start it properly again, or were there some QoR failures? |
|
@litghost I started nightly again yesterday and it seems like it failed again. Are there some qor failures or other issues present? |
|
qor_results_compare.xlsx |
|
Please attach the QoR data. Note that the reviews by Sarah on her code were really joint reviews we did over zoom, and that she typed in. I'm happy with the code. |
…er_feasibility_changes' of https://github.com/verilog-to-routing/vtr-verilog-to-routing into clusterer_feasibility_changes
With the changes in this branch, the clusterer will not pack molecules together that have conflicting PartitionRegions (meaning they have different floorplanning constraints).
Description
Most changes were made in /vpr/src/pack/cluster.cpp. The following changes were made in this file:
-In do_clustering, a new check was added for whether the block failed a floorplanning feasibility check.
-In start_new_cluster, the cluster is assigned an empty PartitionRegion, and that PartitionRegion gets updated to match the seed atom, if the seed atom has its own non-empty PartitionRegion
-In try_pack_molecule a new feasibility check for floorplanning feasibility was added
-A new routine, intersect_atom_cluster_part_regions is called in try_pack_molecule to check compatibility between the atom and cluster
Related Issue
This change adds functionality needed to address the changes wanted in issue #932