ChampSim/branch_comparison.py at baseline · includealex/ChampSim · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
import argparse
import numpy as np
import os
import re
import matplotlib.pyplot as plt

from pathlib import Path
from scipy.stats import gmean

RUN_REGEXP=re.compile(r"run_(\d+)")
MPKI_REGEXP=re.compile(r"MPKI: (?P<mpki>\d+\.\d+) ")
IPC_REGEXP=re.compile(r"CPU 0 cumulative IPC: (?P<ipc>\d+\.\d+) instructions: (?P<instrs>\d+) cycles: (?P<cycles>\d+)")

INCORRECT_IPC=-1
INCORRECT_MPKI=-1

def extract_number(filename: str):
    match = re.search(RUN_REGEXP, filename)
    return int(match.group(1)) if match else float('inf')

def parse_mpki_and_ipc(filename: Path):
    with open(filename, "r") as ifile:
        ilines=ifile.readlines()

    cur_ipc = INCORRECT_IPC
    cur_mpki = INCORRECT_MPKI

    for iline in ilines:
        if (match:=re.search(IPC_REGEXP,iline)):
            cur_ipc = float(match["ipc"])
            continue

        # mpki info is after ipc
        if (match := re.search(MPKI_REGEXP, iline)):
            cur_mpki=float(match["mpki"])
            break

    return cur_ipc, cur_mpki


def plot_results(all_results, labels):
    fig, (ax1, ax2) = plt.subplots(2, 1, figsize=(12, 10), sharex=True)

    num_labels = len(labels)
    file_idxs = sorted(list(set(el["file_idx"] for results in all_results for el in results)))
    bar_width = 0.8 / num_labels

    for i, (results, label) in enumerate(zip(all_results, labels)):
        mpki_dict = {el["file_idx"]: el["mpki"] for el in results}
        ipc_dict = {el["file_idx"]: el["ipc"] for el in results}

        mpkis = [mpki_dict.get(idx, 0) for idx in file_idxs]
        ipcs = [ipc_dict.get(idx, 0) for idx in file_idxs]

        res_gmean_mpki = np.round(gmean([v for v in mpkis if v > 0]), 3)
        res_gmean_ipc = np.round(gmean([v for v in ipcs if v > 0]), 3)

        positions = np.arange(len(file_idxs)) + i * bar_width

        ax1.bar(positions, mpkis, bar_width, label=f"{label} (gmean MPKI={res_gmean_mpki})")
        ax2.bar(positions, ipcs, bar_width, label=f"{label} (gmean IPC={res_gmean_ipc})")

    ax1.set_ylabel("MPKI")
    ax2.set_ylabel("IPC")
    ax2.set_xlabel("File Index")
    ax1.set_title("MPKI Comparison")
    ax2.set_title("IPC Comparison")
    ax2.set_xticks(np.arange(len(file_idxs)) + bar_width * (num_labels - 1) / 2)
    ax2.set_xticklabels(file_idxs)
    ax1.legend()
    ax2.legend()
    ax1.grid(True)
    ax2.grid(True)
    plt.tight_layout()
    plt.savefig("branch_comparison.png")
    plt.close()

def run_analysis(input_dir: Path):
    input_files = os.listdir(input_dir)
    input_files = sorted(input_files, key=extract_number)

    results = []
    for ifile in input_files:
        cur_ipc, cur_mpki = parse_mpki_and_ipc(Path(f"{input_dir}/{ifile}"))
        if cur_ipc == INCORRECT_IPC or cur_mpki == INCORRECT_MPKI:
            continue
        if (file_idx := str(extract_number(ifile))) == float('inf'):
            continue

        results.append({
            "file_idx": file_idx,
            "ipc": cur_ipc,
            "mpki": cur_mpki
        })

    return results

if __name__ == "__main__":
    parser = argparse.ArgumentParser()
    parser.add_argument("--runs-dirs", type=str, help="comma-separated list of directories with results")
    args = parser.parse_args()

    runs_dirs = [Path(p.strip()) for p in args.runs_dirs.split(",")]

    all_results = []
    labels = []

    labels_correct={"baseline_runs": "bimodal",
                    "markov_predictor_runs": "markov",
                    "markov_probable_runs": "markov with probability"}

    for dir_path in runs_dirs:
        results = run_analysis(dir_path)
        all_results.append(results)
        labels.append(labels_correct[dir_path.name] if dir_path.name in labels_correct.keys() else dir_path.name)

    plot_results(all_results, labels)