Vpr viewer and flat routing on #3070
Conversation
github-actions
bot
added
VPR
…-verilog-to-routing into vpr_viewer_and_flat_routing_on
There was a problem hiding this comment.
Thanks, Alex, for fixing the issues with the GUI when flat router is used. In the PR description, it would be helpful if you could explain the specific case you tested that was previously incorrect and is now working properly. That would be useful for future reference.
vpr/src/draw/search_bar.h
Outdated
| @@ -17,6 +17,9 @@ | |||
|
|
|||
| #include "ezgl/application.hpp" | |||
|
|
|||
| bool is_net_unrouted(AtomNetId atomic_net_id); | |||
| bool is_net_fully_absorbed(AtomNetId atomic_net_id); | |||
There was a problem hiding this comment.
I think adding Doxygen comments for these two functions would be better.
There was a problem hiding this comment.
I did a bit of digging through the codebase and found that the routing context includes a data structure called net_status, which has a method called is_routed. I think it would be better to use that instead of writing a custom function to determine whether a net is routed.
There was a problem hiding this comment.
The functionality of is_net_fully_absorbed isn't specific to drawing, so I think it would be better to move it to the route_utils file rather than defining it here.
There was a problem hiding this comment.
I did a bit of digging through the codebase and found that the routing context includes a data structure called net_status, which has a method called is_routed. I think it would be better to use that instead of writing a custom function to determine whether a net is routed.
I just tested this approach, and it's not robust. It seems the net_status structure is filled only on --route stage, and never is filled with, let's say if we run vpr viewer in --analysis only stage. Due to this, if we run vpr viewer in analysis mode, the each net id will be reported as not routed if we attempt to use net_status structure.
With respect of this information, could i continue use
route_ctx.route_trees[net_id].has_value()
? I also may add TODO and comment explaining this.
Or we should think how to fix filling net_stats structure for all possible cases? About fixing net_status filling case, i think it more looks like separate issue.
There was a problem hiding this comment.
The functionality of is_net_fully_absorbed isn't specific to drawing, so I think it would be better to move it to the route_utils file rather than defining it here.
done 80675e3
There was a problem hiding this comment.
I think adding Doxygen comments for these two functions would be better.
added 2e9a2d6
There was a problem hiding this comment.
It seems the net_status structure is filled only on --route stage
If you can file an issue on it, that would be good. Thanks!
…not robust for stage except the route. add doxygen doc for is_net_routed and is_net_fully_absorbed
Thanks for review @amin1377! I updated associated issue description with current and proposed behavior. is it ok? If not i may duplicate this information also in the header of this PR. |
|
Looks good. Thanks! |
Related Issue
#3064
How Has This Been Tested?
manual test:
./vpr test_post_verilog_arch.xml unconnected.eblif --device auto --timing_analysis on --constant_net_method route --clock_modeling ideal --exit_before_pack off --circuit_format eblif --absorb_buffer_luts off --route_chan_width 180 --flat_routing on --gen_post_synthesis_netlist on --timing_report_npaths 100 --timing_report_detail netlist --allow_dangling_combinational_nodes on --analysis --disp on
check that visual nodes representation corresponds to nodes listed in unconnected.route file.
Types of changes
Checklist: