[vtr_flow] Changed Triggers for Second Run

[AlexandreSinger]

Currently, when running the VPR stage of the vtr_flow script, if you are routing at a fixed channel width and you have "--route" or "--analysis" in the command, it will run the VPR stage twice with the exact same arguments. This is unnecessary and can make some of the runs 2x longer than they need to be.

The purpose of running the VPR flow twice is that the second time would load in the temporary files produced by the first run to ensure they are the same values. It appears as though route and analysis used to be part of this check; however it is no longer being used. I checked the second run code and having the --route or --analysis flag set has no affect on the second run's arguments; so the exact same run is dispatched.

I looked through all the regression tests and only found one testcase that used either of these options; which may be the reason this was never caught.

This is useful for the AP flow since, to use the AP flow, we need to specify the --analytical_place flow which inherently does not perfrom routing. So we have to add --route to our arguments. I noticed that these runs took 2x longer than they should and found this issue.

The code that triggers the second run can be found here:

vtr-verilog-to-routing/vtr_flow/scripts/python_libs/vtr/flow.py

Lines 295 to 338 in 12c4600

	do_second_run = False
	second_run_args = vpr_args
	if (
	"write_rr_graph" in vpr_args
	or "analysis" in vpr_args
	or "route" in vpr_args
	or "write_router_lookahead" in vpr_args
	or "write_intra_cluster_router_lookahead" in vpr_args
	):
	do_second_run = True
	if no_second_run:
	do_second_run = False
	vtr.vpr.run(
	architecture_copy,
	pre_vpr_netlist,
	circuit_copy.stem,
	command_runner=command_runner,
	temp_dir=temp_dir,
	vpr_args=vpr_args,
	)
	if do_second_run:
	# Run vpr again with additional parameters.
	# This is used to ensure that files generated by VPR can be re-loaded by it
	rr_graph_ext = (
	Path(second_run_args["write_rr_graph"]).suffix
	if "write_rr_graph" in second_run_args
	else ".xml"
	)
	if check_place:
	second_run_args["route"] = True
	if check_route:
	second_run_args["analysis"] = True
	vtr.vpr.run_second_time(
	architecture_copy,
	pre_vpr_netlist,
	circuit_copy.stem,
	command_runner=command_runner,
	temp_dir=temp_dir,
	second_run_args=second_run_args,
	rr_graph_ext=rr_graph_ext,
	)

The second run is performed here:

vtr-verilog-to-routing/vtr_flow/scripts/python_libs/vtr/vpr/vpr.py

Lines 276 to 341 in 12c4600

	temp_dir = Path(temp_dir) if not isinstance(temp_dir, Path) else temp_dir
	temp_dir.mkdir(parents=True, exist_ok=True)
	rr_graph_out_file = ""
	if "write_rr_graph" in second_run_args:
	rr_graph_out_file = second_run_args["write_rr_graph"]
	rr_graph_ext = Path(rr_graph_out_file).suffix
	rr_graph_out_file2 = "rr_graph2" + rr_graph_ext
	if "write_rr_graph" in second_run_args:
	second_run_args["read_rr_graph"] = rr_graph_out_file
	second_run_args["write_rr_graph"] = rr_graph_out_file2
	inter_cluster_router_lookahead = ""
	if "write_router_lookahead" in second_run_args:
	inter_cluster_router_lookahead = second_run_args["write_router_lookahead"]
	second_run_args["read_router_lookahead"] = inter_cluster_router_lookahead
	second_run_args["write_router_lookahead"] = "inter_cluster_router_lookahead2.capnp"
	intra_cluster_router_lookahead = ""
	if "write_intra_cluster_router_lookahead" in second_run_args:
	intra_cluster_router_lookahead = second_run_args["write_intra_cluster_router_lookahead"]
	second_run_args["read_intra_cluster_router_lookahead"] = intra_cluster_router_lookahead
	second_run_args[
	"write_intra_cluster_router_lookahead"
	] = "intra_cluster_router_lookahead2.capnp"
	# run VPR
	run(
	architecture,
	circuit,
	circuit_name,
	command_runner,
	temp_dir,
	log_filename="vpr_second_run.out",
	vpr_exec=vpr_exec,
	vpr_args=second_run_args,
	)
	if "write_rr_graph" in second_run_args:
	cmd = ["diff", rr_graph_out_file, rr_graph_out_file2]
	_, diff_result = command_runner.run_system_command(
	cmd, temp_dir, log_filename="diff.rr_graph.out", indent_depth=1
	)
	if diff_result:
	raise InspectError("failed: vpr (RR Graph XML output not consistent when reloaded)")
	if "write_inter_cluster_router_lookahead" in second_run_args:
	cmd = ["diff", inter_cluster_router_lookahead, "inter_cluster_router_lookahead2.capnp"]
	_, diff_result = command_runner.run_system_command(
	cmd, temp_dir, log_filename="diff.inter_cluster_router_lookahead.out", indent_depth=1
	)
	if diff_result:
	raise InspectError(
	"failed: vpr (Inter Cluster Router Lookahead output not consistent when reloaded)"
	)
	if "write_intra_cluster_router_lookahead" in second_run_args:
	cmd = ["diff", intra_cluster_router_lookahead, "intra_cluster_router_lookahead2.capnp"]
	_, diff_result = command_runner.run_system_command(
	cmd, temp_dir, log_filename="diff.intra_cluster_router_lookahead.out", indent_depth=1
	)
	if diff_result:
	raise InspectError(
	"failed: vpr (Intra Cluster Router Lookahead not consistent when reloaded)"
	)

Notice that the second run's args only change based on the router lookahead and rr graph.