[WIP] Router Lookahead Profiler

[nedsels]

There is a new class LookaheadProfiler which records a variety of data that can be used to evaluate the quality of the router lookahead.

Description

Using the branch reconstructed in RouteTree::add_subtree_from_heap(), LookaheadProfiler records the following data into a .csv file for every node in the branch:

• iteration no.
• source node id (the first node in the branch; not necessarily a SOURCE)
• sink node id
• sink block name
• sink atom block model
• sink cluster block type
• sink cluster tile height
• sink cluster tile width
• current node
• node type
• node length
• num. nodes from sink
• delta x, delta y
• actual cost, delay, and congestion from current node to sink
• predicted cost, delay, and congestion from current node to sink (predicted by the lookahead)
• criticality

The profiler is enabled by specifying CMAKE_PARAMS="DVPR_PROFILE_LOOKAHEAD=on" when building VPR, and also using the command-line option --profile_router_lookahead [output_file_name.csv].

This data can be analyzed using the script vtr_flow/scripts/profiling_utils/parse_lookahead_data.py, which produces a data summary file and a number of scatter plots, bar graphs, pie charts, and heatmaps. (TBD: Currently, using -h gives detailed instructions, but a README will be added as well.)

Motivation and Context

This was used to verify issues that led to #2639. It can be used to identify current/future issues with the lookahead.

How Has This Been Tested?

There is a new CI test that ensures that VPR builds properly with PROFILE_LOOKAHEAD defined.

Currently, even with PROFILE_LOOKAHEAD undefined, there is a degradation in QoR. These results are averages over 3 runs each:

	Routing Runtime (s)	CPD (ns)	Wirelength	Memory Footprint
vtr (old)	0.94	10.00	27433.36	153.79
vtr (new)	0.91	10.00	27520.35	153.59
titan (old)	294.51	17.18	3200094.23	5494.64
titan (new)	294.30	17.48	3252463.91	5486.61
titan_large* (old)	618.20	17.72	6430833.54	8865.55
titan_large* (new)	614.95	18.14	6532452.49	8854.25

For some reason, there is a 1.6% wirelength increase on titan (and titan_large) circuits and a 2.4% CPD increase on titan_large circuits.

*10 largest titan circuits.

Types of changes

• [ ] Bug fix (change which fixes an issue)
• [x] New feature (change which adds functionality)
• [ ] Breaking change (fix or feature that would cause existing functionality to change)

Checklist:

• [x] My change requires a change to the documentation
• [ ] I have updated the documentation accordingly
• [x] I have added tests to cover my changes
• [ ] All new and existing tests passed