pixman: Branch 'master'

[GitLab Mirror]

 test/lowlevel_blt_bench_compare.py |  177 +++++++++++++++++++++++++++++++++++++
 1 file changed, 177 insertions(+)

New commits:
commit c53b6c827c6c623910b83d49400ab58e60dc3b65
Author: Marek Pikuła <m.pikula at partner.samsung.com>
Date:   Thu Apr 10 02:10:52 2025 +0200

    test: Add lowlevel-blt-bench result comparison script
    
    The script parses and compares results from the `lowlevel-blt-bench`.
    For example, it can be used to compare different backends when running
    the benchmark with different `PIXMAN_DISABLE`.
    
    Additional features:
    
    - save the results in a CSV file (both single result and comparison),
    - print the results as a table with colors, allowing for easy
      identification of the best and worst results.
    
    Signed-off-by: Marek Pikuła <m.pikula at partner.samsung.com>

diff --git a/test/lowlevel_blt_bench_compare.py b/test/lowlevel_blt_bench_compare.py
new file mode 100755
index 0000000..7bb4953
--- /dev/null
+++ b/test/lowlevel_blt_bench_compare.py
@@ -0,0 +1,177 @@
+#!/usr/bin/env python3
+"""
+This script parses benchmark log files generated by the lowlevel-blt-bench benchmark and
+compares the results. It can also export the parsed results to a CSV file.
+
+Usage:
+    python3 lowlevel_blt_bench_compare <log_file> [<log_file2>]
+        [--export-csv <csv_file>] [--quiet] [--color] [--color-breaks <breaks>]
+        [--disable-columns <columns>]
+
+Author: Marek Pikuła <m.pikula at partner.samsung.com>
+"""
+
+import argparse
+import re
+
+import pandas as pd
+from colorama import Fore, Style
+from tabulate import tabulate
+
+SPEEDUP_COLORS = [
+    Fore.RED,
+    Fore.LIGHTRED_EX,
+    Fore.YELLOW,
+    Fore.LIGHTYELLOW_EX,
+    Fore.LIGHTGREEN_EX,
+    Fore.GREEN,
+    Fore.BLUE,
+]
+
+
+def colorize_speedup(value: float, breaks: list[float]) -> str:
+    """Colorize the speedup value depending on the range of its value."""
+    color = SPEEDUP_COLORS[-1]
+    for c, b in zip(SPEEDUP_COLORS, breaks):
+        if value <= b:
+            color = c
+            break
+
+    return f"{color}{value}{Style.RESET_ALL}"
+
+
+def print_df(df: pd.DataFrame, color: bool, breaks: list[float], drop_cols: list[str]):
+    """Print the DataFrame with colorized speedup values."""
+    df.loc["Average"] = df.mean(axis=0)
+    df.drop(columns=drop_cols, errors="ignore", inplace=True)
+
+    table = tabulate(
+        df.map(colorize_speedup, breaks=breaks) if color else df,
+        floatfmt="2.3f",
+        headers="keys",
+        showindex=True,
+    )
+
+    # Print the table
+    print(table)
+
+
+def parse_benchmark_log(log_file: str):
+    """Parse a benchmark log file and return a DataFrame with the results."""
+
+    # Regular expression to match benchmark lines.
+    benchmark_regex = re.compile(
+        r"^\s*(\S+)\s*=\s*L1:\s*([\d.]+)\s*L2:\s*([\d.]+)\s*M:\s*([\d.]+).*"
+        r"HT:\s*([\d.]+)\s*VT:\s*([\d.]+)\s*R:\s*([\d.]+)\s*RT:\s*([\d.]+)\s*"
+        r"\(\s*([\d.]+).*\)"
+    )
+
+    # Read the log file and parse benchmark results using list comprehension.
+    with open(log_file, "r", encoding="utf-8") as file:
+        parsed_lines = tuple(
+            (
+                match.group(1),
+                map(float, match.groups()[1:]),
+            )
+            for line in file
+            if (match := benchmark_regex.match(line))
+        )
+
+        # Unpack parsed lines into functions and metrics.
+        functions, metrics = zip(*parsed_lines) if parsed_lines else ([], [])
+
+        # Create a DataFrame from the parsed data.
+        out = pd.DataFrame(
+            metrics,
+            index=functions,
+            columns=("L1", "L2", "M", "HT", "VT", "R", "RT", "Kops/s"),
+        )
+        out["Avg"] = out.mean(axis=1)
+        return out
+
+    return pd.DataFrame()
+
+
+if __name__ == "__main__":
+    # Set up argument parser.
+    parser = argparse.ArgumentParser(
+        description="Parse and compare lowlevel-blt-bench benchmark results.",
+    )
+    parser.add_argument(
+        "log_file",
+        help="Path to the first benchmark log file.",
+    )
+    parser.add_argument(
+        "log_file2",
+        nargs="?",
+        help="Path to the second benchmark log file (optional).",
+    )
+    parser.add_argument(
+        "--export-csv",
+        "-e",
+        metavar="CSV_FILE",
+        help="Export the parsed results to a CSV file.",
+    )
+    parser.add_argument(
+        "--quiet",
+        "-q",
+        action="store_true",
+        help="Don't print results (useful with --export-csv).",
+    )
+    parser.add_argument(
+        "--color",
+        "-c",
+        action="store_true",
+        help="Print table in color.",
+    )
+    parser.add_argument(
+        "--disable-columns",
+        "-d",
+        metavar="COLUMNS",
+        help="Comma-separated list of columns to disable (e.g., 'L1,L2,M').",
+    )
+    parser.add_argument(
+        "--color-breaks",
+        "-b",
+        metavar="BREAKS",
+        default="0.8,0.9,1.1,1.5,3.0,5.0",
+        help="Speedup values for color breaks (up to 6).",
+    )
+    args = parser.parse_args()
+
+    # Don't truncate DataFrame output.
+    pd.set_option("display.max_rows", None)
+
+    # Parse list arguments.
+    disabled_columns: list[str] = (
+        args.disable_columns.split(",") if args.disable_columns else []
+    )
+    color_breaks: list[float] = list(map(float, args.color_breaks.split(",")))
+
+    # Parse the first log file.
+    df1 = parse_benchmark_log(args.log_file)
+    to_export = df1
+
+    if args.log_file2:
+        # Parse the second log file and calculate speedup
+        df2 = parse_benchmark_log(args.log_file2)
+
+        # Align the two DataFrames based on their indices
+        df1, df2 = df1.align(df2, join="inner")
+
+        speedup = (df2 / df1) * 100000 // 100 / 1000
+
+        if not args.quiet:
+            print(f'Speedup between "{args.log_file}" and "{args.log_file2}":\n')
+            print_df(speedup, args.color, color_breaks, disabled_columns)
+
+        to_export = speedup
+    elif not args.quiet:
+        # Print the parsed DataFrame
+        print_df(df1, args.color, color_breaks, disabled_columns)
+
+    # Export parsed results to CSV if requested
+    if args.export_csv:
+        to_export.to_csv(args.export_csv)
+        if not args.quiet:
+            print(f"Parsed results exported to {args.export_csv}")