Merge branch 'main' of github.com:TILOS-AI-Institute/MacroPlacement into main

.gitignore

@@ -4,11 +4,34 @@ Flows/job
 Flows/util/__pycache__
 CodeElements/*/*/__pycache__
 CodeElements/Plc_client/test/
-CodeElements/Plc_client/test/*/*
-CodeElements/Plc_client/plc_client_os.py
+CodeElements/Plc_client/test/*
 CodeElements/Plc_client/__pycache__/*
 CodeElements/Plc_client/proto_reader.py
 CodeElements/Plc_client/plc_client.py
+CodeElements/failed_proxy_plc/*
+CodeElements/EvalCT/test/g657_ub5_nruns10_c5_r3_v3_rc1
+CodeElements/EvalCT/snapshot*
+CodeElements/EvalCT/circuit_training
+CodeElements/EvalCT/__pycache__/
+CodeElements/EvalCT/eval_run*.plc
+CodeElements/EvalCT/eval_run*.plc
+CodeElements/EvalCT/saved_policy/run_00/111/train/*
+CodeElements/EvalCT/saved_policy/run_00/111/snapshot*.plc
+CodeElements/EvalCT/saved_policy/run_00/111/rl*.plc
+CodeElements/EvalCT/saved_policy/run_os_64128_g657_ub5_nruns10_c5_r3_v3_rc1
+CodeElements/epoch_*.plc
+CodeElements/FDPlacement/test/*
+CodeElements/Plc_client/placement_util*.py
+CodeElements/StatTest/test/ariane/*_vs_*
+CodeElements/StatTest/test/flow2_68_1.3_ct/*
+CodeElements/StatTest/test/ariane/*.plc
+CodeElements/*/__pycache__/*
+CodeElements/*.png
+CodeElements/StatTest/*.csv
+CodeElements/os.txt
+CodeElements/EvalCT/test/ariane/*.plc
+CodeElements/FDPlacement/fd.txt
+CodeElements/StatTest/test/ariane/*.pb.txt
 CodeElements/Plc_client/failed_proxy_plc/*
 __pycache__/
 __pycache__

CodeElements/EvalCT/README.md

+# Circuit Training Model Evaulation
+
+## Quick Start
+First, make sure you are under `CodeElements/EvalCT/`. To run `eval_ct.py`, we need to download `circuit_training` code under the same directory. Run the following command to get the stable version that we have tested.
+```
+# Compatible with Revision #90. Latest Version might cause errors
+svn export -r 90 --force https://github.com/google-research/circuit_training.git/trunk/circuit_training
+```
+
+Next, we need to prepare your trained policy and the testcase you want to evaluate on. Assume you have trained models (which usually can be found under `./logs`), copy the run folder into `saved_model` folder. Make sure your testcase is under `./test`. The `ckptID` is the policy checkpoint ID saved after each iteration.
+
+Finally, run the following command with path to netlist file, initial placement file and model run directory path.
+```
+$ python3 -m eval_ct --netlist ./test/ariane/netlist.pb.txt\
+            --plc ./test/ariane/initial.plc\
+            --rundir run_00\
+            --ckptID policy_checkpoint_0000103984
+```
+The placement will be stored under `CodeElements/EvalCT/` and named as `eval_[RUN_DIR]_to_[TESTCASE].plc`.
+
+## Trained Policy
+We are providing one of the run we trained from scratch using Google's Ariane testcase. **This should not be taken as representing the potential of Circuit Training**. We are only providing these trained weights here for the sake of testing. Please feel free to load any of your own trained weights. You may find similar file structure under `./logs` after training.
+
+## View Your Result
+You can view the result by supplying this placement file into the open-sourced Plc_client testbench and use the `display_canvas` function.

CodeElements/EvalCT/eval_ct.py

+import collections
+import functools
+import os
+import time
+from typing import Text
+import statistics
+import re
+
+from absl import app
+from absl import flags
+from absl.flags import argparse_flags
+from circuit_training.environment import environment
+from circuit_training.environment import placement_util
+from tf_agents.experimental.distributed import reverb_variable_container
+from tf_agents.metrics import py_metric
+from tf_agents.metrics import py_metrics
+from tf_agents.policies import greedy_policy  # pylint: disable=unused-import
+from tf_agents.policies import py_tf_eager_policy
+from tf_agents.policies import policy_loader
+from tf_agents.train import actor
+from tf_agents.train import learner
+from tf_agents.train.utils import train_utils
+from tf_agents.trajectories import trajectory
+from tf_agents.utils import common
+from tf_agents.policies import greedy_policy  # pylint: disable=unused-import
+from tf_agents.system import system_multiprocessing as multiprocessing
+
+"""
+Example
+    At ./MacroPlacement/CodeElement/EvalCT, run the following command:
+
+        $ cd EvalCT && python3 -m eval_ct --netlist ./test/ariane/netlist.pb.txt\
+            --plc ./test/ariane/initial.plc\
+            --rundir run_00\
+            --ckptID policy_checkpoint_0000103984\
+            && cd -
+
+"""
+# InfoMetric_wirelength = 0.09238692254177283
+# 		 InfoMetric_congestion = 0.9468230846211636
+# 		 InfoMetric_density = 0.5462616496124097
+
+
+class InfoMetric(py_metric.PyStepMetric):
+    """Observer for graphing the environment info metrics."""
+
+    def __init__(
+        self,
+        env,
+        info_metric_key: Text,
+        buffer_size: int = 1,
+        name: Text = 'InfoMetric',
+    ):
+        """Observer reporting TensorBoard metrics at the end of each episode.
+
+        Args:
+        env: environment.
+        info_metric_key: a string key from the environment info to report,
+            e.g. wirelength, density, congestion.
+        buffer_size: size of the buffer for calculating the aggregated metrics.
+        name: name of the observer object.
+        """
+        super(InfoMetric, self).__init__(name + '_' + info_metric_key)
+
+        self._env = env
+        self._info_metric_key = info_metric_key
+        self._buffer = collections.deque(maxlen=buffer_size)
+
+    def call(self, traj: trajectory.Trajectory):
+        """Report the requested metrics at the end of each episode."""
+
+        # We collect the metrics from the info from the environment instead.
+        # The traj argument is kept to be compatible with the actor/learner API
+        # for metrics.
+        del traj
+
+        if self._env.done:
+            # placement_util.save_placement(self._env._plc, './reload_weight.plc', '')
+            metric_value = self._env.get_info()[self._info_metric_key]
+            self._buffer.append(metric_value)
+
+    def result(self):
+        return statistics.mean(self._buffer)
+
+    def reset(self):
+        self._buffer.clear()
+
+def evaulate(model_dir, ckpt_id, create_env_fn):
+    # Create the path for the serialized greedy policy.
+    policy_saved_model_path = os.path.join(model_dir,
+                                           learner.POLICY_SAVED_MODEL_DIR,
+                                           learner.GREEDY_POLICY_SAVED_MODEL_DIR)
+    try:
+        assert os.path.isdir(policy_saved_model_path) 
+        print("#[POLICY SAVED MODEL PATH] " + policy_saved_model_path)
+    except AssertionError:
+        print("[ERROR POLICY SAVED MODEL PATH NOT FOUND] " + policy_saved_model_path)
+        exit(0)
+
+    policy_saved_chkpt_path = os.path.join(model_dir,
+                                         learner.POLICY_SAVED_MODEL_DIR,
+                                         "checkpoints", ckpt_id)
+    try:
+        assert os.path.isdir(policy_saved_chkpt_path)
+        print("#[POLICY SAVED CHECKPOINT PATH] " + policy_saved_chkpt_path)
+    except AssertionError:
+        print("[ERROR POLICY SAVED CHECKPOINT PATH NOT FOUND] " + policy_saved_chkpt_path)
+        exit(0)
+
+    saved_model_pb_path = os.path.join(policy_saved_model_path, 'saved_model.pb')
+
+    try:
+        assert os.path.isfile(saved_model_pb_path)
+        print("#[SAVED MODEL PB PATH] " + saved_model_pb_path)
+    except AssertionError:
+        print("[ERROR SAVE MODEL PB PATH NOT FOUND] " + saved_model_pb_path)
+        exit(0)
+
+    policy = policy_loader.load(policy_saved_model_path, policy_saved_chkpt_path)
+
+    policy.get_initial_state()
+
+    print(policy.variables()[0].numpy())
+
+    train_step = train_utils.create_train_step()
+
+    # Create the environment.
+    env = create_env_fn()
+
+    # Create the evaluator actor.
+    info_metrics = [
+        InfoMetric(env, 'wirelength'),
+        InfoMetric(env, 'congestion'),
+        InfoMetric(env, 'density'),
+    ]
+
+    eval_actor = actor.Actor(
+      env,
+      policy,
+      train_step,
+      episodes_per_run=1,
+      summary_dir=os.path.join(model_dir, learner.TRAIN_DIR, 'eval'),
+      metrics=[
+          py_metrics.NumberOfEpisodes(),
+          py_metrics.EnvironmentSteps(),
+          py_metrics.AverageReturnMetric(
+              name='eval_episode_return', buffer_size=1),
+          py_metrics.AverageEpisodeLengthMetric(buffer_size=1),
+      ] + info_metrics,
+      name='performance')
+
+    eval_actor.run_and_log()
+
+
+def main(args):
+    NETLIST_FILE = args.netlist
+    INIT_PLACEMENT = args.plc
+    POLICY_CHECKPOINT_ID = args.ckptID
+    GLOBAL_SEED = 111
+    CD_RUNTIME = False
+    RUN_NAME = args.rundir
+
+    # extract eval testcase name
+    EVAL_TESTCASE = re.search("/test/(.+?)/netlist.pb.txt", NETLIST_FILE).group(1)
+
+    create_env_fn = functools.partial(
+        environment.create_circuit_environment,
+        netlist_file=NETLIST_FILE,
+        init_placement=INIT_PLACEMENT,
+        is_eval=True,
+        save_best_cost=True,
+        output_plc_file=str('./eval_' + RUN_NAME + '_to_' + EVAL_TESTCASE + '.plc'),
+        global_seed=GLOBAL_SEED,
+        cd_finetune=CD_RUNTIME
+    )
+
+    evaulate(model_dir=os.path.join("./saved_policy", RUN_NAME, str(GLOBAL_SEED)), 
+                ckpt_id=POLICY_CHECKPOINT_ID, create_env_fn=create_env_fn) 
+
+def parse_flags(argv):
+    parser = argparse_flags.ArgumentParser(
+        description='An argparse + app.run example')
+    parser.add_argument("--netlist", required=True,
+                        help="Path to netlist in pb.txt")
+    parser.add_argument("--plc", required=True,
+                        help="Path to plc in .plc")
+    parser.add_argument("--rundir", required=True,
+                        help="Path to run directory that contains saved policies")
+    parser.add_argument("--ckptID", required=True,
+                        help="Policy checkpoint ID")
+
+    return parser.parse_args(argv[1:])
+
+if __name__ == '__main__':
+    app.run(main, flags_parser=parse_flags)
\ No newline at end of file