VPR Test `test_odd_even_routing` Failed

[ueqri]

The VPR test test_odd_even_routing (as shown below) has consistently failed in CI (U: C++ Unit Tests) since 2b6a935 (included) in the master branch.

vtr-verilog-to-routing/vpr/test/test_odd_even_routing.cpp

Lines 228 to 264 in 3e53a93

	SECTION("Test case where multiple traffic flows are router, and routes are checked for turn legality and deadlock freedom.") {
	std::random_device device;
	std::mt19937 rand_num_gen(device());
	std::uniform_int_distribution<std::mt19937::result_type> dist(0, 99);
	NocTrafficFlows traffic_flow_storage;
	for (int i = 0; i < 100; i++) {
	auto src_blk_id = (ClusterBlockId)dist(rand_num_gen);
	ClusterBlockId dst_blk_id;
	do {
	dst_blk_id = (ClusterBlockId)dist(rand_num_gen);
	} while (src_blk_id == dst_blk_id);
	traffic_flow_storage.create_noc_traffic_flow("dummy_name_1", "dummy_name_2", src_blk_id, dst_blk_id, 1, 1, 1);
	}
	traffic_flow_storage.finished_noc_traffic_flows_setup();
	vtr::vector<NocTrafficFlowId, std::vector<NocLinkId>> traffic_flow_routes(traffic_flow_storage.get_number_of_traffic_flows());
	for (const auto& [id, traffic_flow] : traffic_flow_storage.get_all_traffic_flows().pairs()) {
	NocRouterId src_router_id = noc_model.get_router_at_grid_location(block_locs[traffic_flow.source_router_cluster_id].loc);
	NocRouterId dst_router_id = noc_model.get_router_at_grid_location(block_locs[traffic_flow.sink_router_cluster_id].loc);
	REQUIRE_NOTHROW(routing_algorithm.route_flow(src_router_id, dst_router_id,
	id, traffic_flow_routes[id], noc_model));
	}
	ChannelDependencyGraph cdg(noc_model, traffic_flow_storage, traffic_flow_routes, block_locs);
	REQUIRE(cdg.has_cycles() == false);
	check_turn_legality(traffic_flow_routes, noc_model, routing_algorithm);
	}

@AlexandreSinger @soheilshahrouz It seems that Alex's commit 2b6a935 has nothing to do with either NoC (or odd even routing). I am not sure why this test started failing since then.

The corresponding logs are here:

-------------------------------------------------------------------------------
test_route_flow
  Test case where multiple traffic flows are router, and routes are checked for
  turn legality and deadlock freedom.
-------------------------------------------------------------------------------
/home/runner/work/vtr-verilog-to-routing/vtr-verilog-to-routing/vpr/test/test_odd_even_routing.cpp:228
...............................................................................

/home/runner/work/vtr-verilog-to-routing/vtr-verilog-to-routing/vpr/test/test_odd_even_routing.cpp:228: FAILED:
due to a fatal error condition:
  SIGSEGV - Segmentation violation signal

Showed in https://github.com/verilog-to-routing/vtr-verilog-to-routing/actions/runs/11077555018/job/30783077163?pr=2720.

-------------------------------------------------------------------------------
test_route_flow
  Test case where multiple traffic flows are router, and routes are checked for
  turn legality and deadlock freedom.
-------------------------------------------------------------------------------
/home/runner/work/vtr-verilog-to-routing/vtr-verilog-to-routing/vpr/test/test_odd_even_routing.cpp:228
...............................................................................

/home/runner/work/vtr-verilog-to-routing/vtr-verilog-to-routing/vpr/test/test_odd_even_routing.cpp:228: FAILED:
due to unexpected exception with message:
  std::bad_alloc

Showed in https://github.com/verilog-to-routing/vtr-verilog-to-routing/actions/runs/11072279016/job/30766140332 and same in https://github.com/verilog-to-routing/vtr-verilog-to-routing/actions/runs/11070489508/job/30760275852.

[AlexandreSinger]

@ueqri Thanks for creating an issue on this. Soheil and I have been informally talking about this since Monday. It started since around when I made the ClusterLegalizer change. @soheilshahrouz has been looking into it, but is saying that he cannot replicate it locally. It seems to be something with the exact environment of the CI which is causing this test to randomly fail. I am not familiar with this Unit Test to know why.

[ueqri]

Gotcha! Thanks, Alex and Soheil! I was just trying to make CI happy in the router PR and found this error that I have no idea how to resolve. 😂

[AlexandreSinger]

Yeah, I am thinking of disabling this test while we figure this out; but you can just try to rerun the test.

[soheilshahrouz]

Resolved by mergin #2735.