Introduce powerful simulation script and provide an example

2023-07-26 16:27:08 +02:00
parent 0b88161640
commit 26b418124e
4 changed files with 590 additions and 0 deletions
--- a/scripts/simulation/example_ddr3_simulations.json
+++ b/scripts/simulation/example_ddr3_simulations.json
@@ -0,0 +1,142 @@
+{
+    "simulation": {
+        "addressmapping": {
+            "BANK_BIT": [
+                6,
+                7,
+                8
+            ],
+            "BYTE_BIT": [
+                0,
+                1,
+                2
+            ],
+            "COLUMN_BIT": [
+                3,
+                4,
+                5,
+                9,
+                10,
+                11,
+                12,
+                13,
+                14,
+                15
+            ],
+            "ROW_BIT": [
+                16,
+                17,
+                18,
+                19,
+                20,
+                21,
+                22,
+                23,
+                24,
+                25,
+                26,
+                27,
+                28,
+                29
+            ]
+        },
+        "mcconfig": {
+            "PagePolicy": "<page_policy>",
+            "SchedulerBuffer": "<scheduler_buffer>",
+            "RequestBufferSize": <scheduler_buffer_size>,
+            "RefreshPolicy": "<refresh_policy>",
+            "RefreshMaxPostponed": 0,
+            "RefreshMaxPulledin": 0,
+            "RefreshManagement": false,
+            "Arbiter": "Simple",
+            "CmdMux": "Oldest",
+            "MaxActiveTransactions": 128,
+            "PowerDownPolicy": "NoPowerDown",
+            "RespQueue": "Fifo",
+            "Scheduler": "FrFcfs"
+        },
+        "memspec": {
+            "memarchitecturespec": {
+                "burstLength": 8,
+                "dataRate": 2,
+                "nbrOfBanks": 8,
+                "nbrOfChannels": 1,
+                "nbrOfColumns": 1024,
+                "nbrOfDevices": 8,
+                "nbrOfRanks": 1,
+                "nbrOfRows": 16384,
+                "width": 8
+            },
+            "memoryId": "MICRON_1Gb_DDR3-1600_8bit_G",
+            "memoryType": "DDR3",
+            "mempowerspec": {
+                "idd0": 70.0,
+                "idd2n": 45.0,
+                "idd2p0": 12.0,
+                "idd2p1": 30.0,
+                "idd3n": 45.0,
+                "idd3p0": 35.0,
+                "idd3p1": 35.0,
+                "idd4r": 140.0,
+                "idd4w": 145.0,
+                "idd5": 170.0,
+                "idd6": 8.0,
+                "vdd": 1.5
+            },
+            "memtimingspec": {
+                "ACTPDEN": 1,
+                "AL": 0,
+                "CCD": 4,
+                "CKE": 3,
+                "CKESR": 4,
+                "CL": 10,
+                "DQSCK": 0,
+                "FAW": 24,
+                "PRPDEN": 1,
+                "RAS": 28,
+                "RC": 38,
+                "RCD": 10,
+                "REFI": 6240,
+                "REFPDEN": 1,
+                "RFC": 88,
+                "RL": 10,
+                "RP": 10,
+                "RRD": 5,
+                "RTP": 6,
+                "RTRS": 1,
+                "WL": 8,
+                "WR": 12,
+                "WTR": 6,
+                "XP": 6,
+                "XPDLL": 20,
+                "XS": 96,
+                "XSDLL": 512,
+                "clkMhz": 800
+            }
+        },
+        "simconfig": {
+            "AddressOffset": 0,
+            "CheckTLM2Protocol": false,
+            "DatabaseRecording": true,
+            "Debug": false,
+            "EnableWindowing": false,
+            "PowerAnalysis": false,
+            "SimulationName": "example",
+            "SimulationProgressBar": false,
+            "StoreMode": "NoStorage",
+            "UseMalloc": false,
+            "WindowSize": 1000
+        },
+        "simulationid": "ddr3-example",
+        "tracesetup": [
+            {
+                "clkMhz": 2000,
+                "type": "generator",
+                "name": "gen1",
+                "numRequests": 20000,
+                "rwRatio": 0.85,
+                "addressDistribution": "<pattern>"
+            }
+        ]
+    }
+}
--- a/scripts/simulation/example_ddr3_simulations.py
+++ b/scripts/simulation/example_ddr3_simulations.py
@@ -0,0 +1,114 @@
+#!/usr/bin/env python3
+
+import os
+import matplotlib.pyplot as plt
+
+from dataclasses import dataclass
+from simulation import (
+    Simulation,
+    Configuration,
+    get_options_from_args,
+    simulation_results,
+)
+
+script_directory = os.path.dirname(os.path.abspath(__file__))
+base_config = os.path.join(script_directory, "example_ddr3_simulations.json")
+
+simulations: list[Simulation] = []
+
+
+@dataclass(frozen=True)
+class SchedulerBuffer:
+    buffer: str
+    size: int
+
+
+refresh_policies = ["NoRefresh", "AllBank"]
+patterns = ["sequential", "random"]
+scheduler_buffers = [
+    SchedulerBuffer("Bankwise", 8),
+    SchedulerBuffer("Bankwise", 16),
+]
+
+for refresh_policy in refresh_policies:
+    for scheduler_buffer in scheduler_buffers:
+        for pattern in patterns:
+            PAGE_POLICY = "Open" if pattern == "sequential" else "Closed"
+
+            config = Configuration(
+                f"{refresh_policy}_{scheduler_buffer.buffer}_{scheduler_buffer.size}_{pattern}",
+                {
+                    "refresh_policy": refresh_policy,
+                    "pattern": pattern,
+                    "page_policy": PAGE_POLICY,
+                    "scheduler_buffer": scheduler_buffer.buffer,
+                    "scheduler_buffer_size": scheduler_buffer.size,
+                },
+            )
+
+            simulations.append(Simulation(config))
+
+
+options = get_options_from_args()
+
+if options.base_config is None:
+    options.base_config = base_config
+
+dataframe = simulation_results(options, simulations)
+
+# Generate interesting plots!
+for refresh_policy in refresh_policies:
+    refresh_policy_filter = dataframe["refresh_policy"] == refresh_policy
+
+    for scheduler_buffer in scheduler_buffers:
+        scheduler_buffer_filter = (
+            dataframe["scheduler_buffer"] == scheduler_buffer.buffer
+        )
+        scheduler_buffer_size_filter = (
+            dataframe["scheduler_buffer_size"] == scheduler_buffer.size
+        )
+
+        filters = (
+            refresh_policy_filter
+            & scheduler_buffer_filter
+            & scheduler_buffer_size_filter
+        )
+
+        TEMP_DATAFRAME = dataframe[filters]
+
+        for pattern in patterns:
+            pattern_filter = dataframe["pattern"] == pattern
+
+            plt.figure(figsize=(8.0, 12.0))
+
+            filters = (
+                refresh_policy_filter
+                & scheduler_buffer_filter
+                & scheduler_buffer_size_filter
+            )
+
+            TEMP_DATAFRAME = dataframe[filters]
+
+            plt.bar(
+                "pattern",
+                "bandwidth",
+                data=TEMP_DATAFRAME,
+            )
+
+        # Plot MAX line
+        plt.bar(
+            "MAX",
+            "max_bandwidth",
+            data=TEMP_DATAFRAME,
+            color="darkgray",
+        )
+
+        title = f"{refresh_policy} {scheduler_buffer.buffer} {scheduler_buffer.size}"
+
+        plt.title(title)
+        plt.axis(ymin=0, ymax=120)
+        plt.xlabel("Access Pattern")
+        plt.ylabel("Bandwidth [Gb/s]")
+
+        plt.savefig(f"{options.out_dir}/{title}.png")
+        plt.close()
--- a/scripts/simulation/requirements.txt
+++ b/scripts/simulation/requirements.txt
@@ -0,0 +1,3 @@
+tqdm
+pandas
+matplotlib
--- a/scripts/simulation/simulation.py
+++ b/scripts/simulation/simulation.py
@@ -0,0 +1,331 @@
+import argparse
+import subprocess
+import sys
+import sqlite3
+import json
+import os
+import re
+
+from pathlib import Path
+from dataclasses import dataclass, fields
+from typing import Optional
+from multiprocessing.pool import ThreadPool
+from tqdm import tqdm
+
+import pandas as pd
+
+sys.path.append("extensions/apps/traceAnalyzer/scripts")
+
+from metrics import (
+    average_response_latency_in_ns,
+    max_response_latency_in_ns,
+    memory_active_in_percent,
+    maximum_data_rate,
+)
+
+
+@dataclass
+class Options:
+    dramsys: Path
+    override: bool
+    out_dir: Path
+    simulate: bool
+    metrics: bool
+    base_config: Path | None
+    resource_dir: Path | None = None
+    jobs: int | None = None
+
+
+@dataclass(frozen=True)
+class SubConfig:
+    name: str
+    parameters: dict[str, str]
+
+
+@dataclass(frozen=True)
+class Statistics:
+    filename: str
+    databus_utilization: float
+    bandwidth: float
+    max_bandwidth: float
+    avg_latency: float
+    max_latency: float
+
+
+@dataclass(frozen=True)
+class Configuration:
+    name: str
+    tokens: dict[str, str | int]
+
+
+@dataclass
+class Simulation:
+    config: Configuration
+    directory: Optional[str] = None
+    statistics: Optional[list[Statistics]] = None
+
+
+def run_dramsys(dramsys: Path, simulation_dir: Path, resource_dir: Path | None):
+    with open(f"{simulation_dir}/out.txt", "w", encoding="utf-8") as output_file:
+        command = [dramsys.absolute(), "config.json"]
+
+        if resource_dir is not None:
+            command.append(resource_dir)
+
+        subprocess.run(command, cwd=simulation_dir, stdout=output_file, check=True)
+
+
+def calculate_simulation_metrics(simulation: Simulation):
+    simulation_dir = simulation.directory
+    stats: list[Statistics] = []
+
+    for file in os.listdir(simulation_dir):
+        if file.endswith(".tdb"):
+            connection = sqlite3.connect(f"{simulation_dir}/{file}")
+
+            try:
+                max_bandwidth = maximum_data_rate(connection) / 1000
+                avg_latency = average_response_latency_in_ns(connection)
+                max_latency = max_response_latency_in_ns(connection)
+                databus_utilization = memory_active_in_percent(connection) / 100
+                bandwidth = databus_utilization * max_bandwidth
+
+                stats.append(
+                    Statistics(
+                        file,
+                        databus_utilization,
+                        bandwidth,
+                        max_bandwidth,
+                        avg_latency,
+                        max_latency,
+                    )
+                )
+
+            except Exception as error:
+                print(
+                    f"Warning: Could not calculate metrics for {simulation_dir}/{file}: {error}"
+                )
+
+    simulation.statistics = stats
+
+
+# Replace placeholders with actual values
+def replace_placeholders(config_json: str, tokens: dict) -> str:
+    for key, value in tokens.items():
+        placeholder = f"<{key}>"
+
+        config_json = config_json.replace(placeholder, str(value))
+
+    return config_json
+
+
+@dataclass
+class WorkItem:
+    dramsys: Path
+    simulation: Simulation
+    base_config: Path
+    resource_dir: Path | None
+
+
+def simulate(
+    work_item: WorkItem,
+):
+    simulation_dir = work_item.simulation.directory
+
+    json_config = None
+    with open(work_item.base_config, encoding="utf-8") as config_file:
+        config_string = config_file.read()
+        config_string = replace_placeholders(
+            config_string, work_item.simulation.config.tokens
+        )
+
+        json_config = json.loads(config_string)
+
+    simulation_json = simulation_dir / "config.json"
+
+    # Save config besides simulation directory
+    with open(simulation_json, "w", encoding="utf-8") as config_file:
+        json.dump(json_config, config_file, indent=4)
+
+    run_dramsys(work_item.dramsys, simulation_dir, work_item.resource_dir)
+    calculate_simulation_metrics(work_item.simulation)
+
+
+def generate_dataframe(simulations: list[Simulation], out_dir: str) -> pd.DataFrame:
+    # Pack results in a panda dataframe
+    labels = ["name", "channel"]
+    statistic_labels = list(map(lambda field: field.name, fields(Statistics)))
+
+    # Get one simulation...
+    config_keys, _ = zip(*simulations[0].config.tokens.items())
+
+    labels.extend(config_keys)
+    labels.extend(statistic_labels)
+
+    entries = []
+
+    for simulation in simulations:
+        _, config_values = zip(*simulation.config.tokens.items())
+
+        for stat in simulation.statistics:
+            channel_pattern = re.compile("(?<=ch)[0-9]+")
+            channel = int(channel_pattern.search(stat.filename)[0])
+
+            entries.append(
+                [
+                    simulation.config.name,
+                    channel,
+                    *config_values,
+                    stat.filename,
+                    stat.databus_utilization,
+                    stat.bandwidth,
+                    stat.max_bandwidth,
+                    stat.avg_latency,
+                    stat.max_latency,
+                ]
+            )
+
+    dataframe = pd.DataFrame(data=entries, columns=labels)
+    dataframe.to_csv(f"{out_dir}/statistics.csv", sep=";")
+
+    return dataframe
+
+
+def populate_simulation_directories(
+    simulations: list[Simulation], out_dir: str, override: bool
+):
+    for simulation in simulations:
+        simulation_dir = Path(f"{out_dir}/simulations/{simulation.config.name}")
+
+        try:
+            simulation_dir.mkdir(parents=True, exist_ok=override)
+        except FileExistsError:
+            print(
+                "Previous simulations artifacts found. To continue, enable the force override flag."
+            )
+            sys.exit(-1)
+
+        simulation.directory = simulation_dir
+
+
+def calculate_metrics(
+    simulations: list[Simulation], out_dir: str, jobs: int | None
+) -> pd.DataFrame:
+    populate_simulation_directories(simulations, out_dir, override=True)
+
+    with ThreadPool(jobs) as thread_pool:
+        for _ in tqdm(
+            thread_pool.imap_unordered(calculate_simulation_metrics, simulations),
+            total=len(simulations),
+        ):
+            pass
+
+    return generate_dataframe(simulations, out_dir)
+
+
+def run_simulations(simulations: list[Simulation], options: Options) -> pd.DataFrame:
+    if len(simulations) == 0:
+        print("Must specify at least one simulation configuration!")
+        sys.exit(-1)
+
+    if options.base_config is None:
+        print("Must specify a base config")
+        sys.exit(-1)
+
+    print("Create simulation directories...")
+    populate_simulation_directories(simulations, options.out_dir, options.override)
+
+    print("Run simulations...")
+    with ThreadPool(options.jobs) as thread_pool:
+        args = list(
+            WorkItem(
+                options.dramsys, simulation, options.base_config, options.resource_dir
+            )
+            for simulation in simulations
+        )
+
+        for _ in tqdm(thread_pool.imap_unordered(simulate, args), total=len(args)):
+            pass
+
+    print("Calculate metrics...")
+    return calculate_metrics(simulations, options.out_dir, options.jobs)
+
+
+def get_options_from_args() -> Options:
+    parser = argparse.ArgumentParser(description="DRAMSys simulation utility")
+    parser.add_argument("dramsys", type=Path, help="path to the DRAMSys executable")
+    parser.add_argument(
+        "--simulate",
+        default=False,
+        action="store_true",
+        help="run the simulations generating simulation artifacts",
+    )
+    parser.add_argument(
+        "--metrics",
+        default=False,
+        action="store_true",
+        help="calculate the metrics from existing simulation artifacts",
+    )
+    parser.add_argument(
+        "-f",
+        "--force",
+        default=False,
+        action="store_true",
+        help="force override existing simulation artifacts",
+    )
+    parser.add_argument(
+        "--out-dir",
+        type=Path,
+        default="out",
+        help="path to the output directory",
+    )
+    parser.add_argument(
+        "--base-config",
+        type=Path,
+        help="path to the base configuration file",
+    )
+    parser.add_argument(
+        "--resource-dir",
+        type=Path,
+        help="path to the resource directory",
+    )
+    parser.add_argument(
+        "-j",
+        "--jobs",
+        metavar="N",
+        type=int,
+        default=None,
+        help="run N jobs in parallel",
+    )
+
+    arguments = parser.parse_args()
+    return Options(
+        arguments.dramsys,
+        arguments.force,
+        arguments.out_dir,
+        arguments.simulate,
+        arguments.metrics,
+        arguments.base_config,
+        arguments.resource_dir,
+        arguments.jobs,
+    )
+
+
+def simulation_results(
+    options: Options,
+    simulations: list[Simulation],
+) -> pd.DataFrame:
+    if options.simulate:
+        return run_simulations(simulations, options)
+
+    if options.metrics:
+        return calculate_metrics(simulations, options.out_dir, options.jobs)
+
+    print("Summarizing simulation results in statistics.csv...")
+
+    statistics_file = f"{options.out_dir}/statistics.csv"
+    if not os.path.isfile(statistics_file):
+        print("Run the simulations first to generate simulation artifacts")
+        sys.exit(-1)
+
+    return pd.read_csv(f"{options.out_dir}/statistics.csv", sep=";")