TheRock/rocm-systems/projects/rocprof-trace-decoder/test/diff.py

import csv
import os
import re


def lit_to_signed32(match):
    """Convert a lit(0x...) value to its two's complement signed 32-bit representation."""
    hex_str = match.group(1)
    value = int(hex_str, 16)
    # Convert to signed 32-bit using two's complement
    if value >= 0x80000000:
        value -= 0x100000000
    return str(value)


def normalize_literals(text):
    """Replace lit(0x...) with the signed 32-bit two's complement value, and lit64(0x...) with just 0x..."""
    text = re.sub(r"lit\(0x([0-9a-fA-F]+)\)", lit_to_signed32, text)
    text = re.sub(r"lit64\((0x[0-9a-fA-F]+)\)", r"\1", text)
    text = text.replace("HW_REG_WAVE_SCHED_MODE", "26").replace("HW_REG_SCHED_MODE", "26")
    return text


def normalize_row(row):
    """Apply literal normalization to each cell in a row."""
    return [normalize_literals(cell) for cell in row]


def is_ignorable_row(header, row):
    """Return True for rows with Hitcount=0 and Latency=0 (e.g. padding instructions)."""
    stripped = [h.strip() for h in header]
    hit_idx = stripped.index("Hitcount") if "Hitcount" in stripped else None
    lat_idx = stripped.index("Latency") if "Latency" in stripped else None
    if hit_idx is None or lat_idx is None:
        return False
    if hit_idx >= len(row) or lat_idx >= len(row):
        return False
    return row[hit_idx].strip() == "0" and row[lat_idx].strip() == "0"


def sort_key(header, row):
    """Return a sort key of (CodeObj, Vaddr) as integers for stable sorting."""
    stripped = [h.strip() for h in header]
    codeobj_idx = stripped.index("CodeObj") if "CodeObj" in stripped else None
    vaddr_idx = stripped.index("Vaddr") if "Vaddr" in stripped else None
    codeobj = int(row[codeobj_idx], 0) if codeobj_idx is not None and codeobj_idx < len(row) else 0
    vaddr = int(row[vaddr_idx], 0) if vaddr_idx is not None and vaddr_idx < len(row) else 0
    return (codeobj, vaddr)


def strip_source_column(header, data):
    """Remove the 'Source' column (if present) from the header and all data rows."""
    if not header:
        return header, data
    try:
        idx = [h.strip() for h in header].index("Source")
    except ValueError:
        return header, data
    header = header[:idx] + header[idx + 1:]
    data = [row[:idx] + row[idx + 1:] for row in data]
    return header, data


def diff_csv_files(file1, file2):
    """
    Compare two CSV files and write the differences to an output file.

    :param file1: Path to the first CSV file.
    :param file2: Path to the second CSV file.
    """
    with open(file1, "r", newline="") as f1, open(file2, "r", newline="") as f2:
        reader1 = csv.reader(f1)
        reader2 = csv.reader(f2)

        data1 = list(reader1)
        data2 = list(reader2)

    # Normalize literals before any comparison
    data1 = [normalize_row(row) for row in data1]
    data2 = [normalize_row(row) for row in data2]

    # Strip 'Source' column — it depends on the compiler/toolchain and is not stable
    if data1:
        header1, data1[1:] = strip_source_column(data1[0], data1[1:])
        data1[0] = header1
    if data2:
        header2, data2[1:] = strip_source_column(data2[0], data2[1:])
        data2[0] = header2

    # Filter out rows with zero hitcount and latency (e.g. padding instructions)
    header = data1[0] if data1 else []
    filtered1 = (
        [data1[0]] + [row for row in data1[1:] if not is_ignorable_row(header, row)]
        if data1
        else []
    )
    filtered2 = (
        [data2[0]] + [row for row in data2[1:] if not is_ignorable_row(header, row)]
        if data2
        else []
    )

    # Stable sort data rows by (CodeObj, Vaddr), keeping header in place
    if header and len(filtered1) > 1:
        filtered1[1:] = sorted(filtered1[1:], key=lambda r: sort_key(header, r))
    if header and len(filtered2) > 1:
        filtered2[1:] = sorted(filtered2[1:], key=lambda r: sort_key(header, r))

    for row1, row2 in zip(filtered1, filtered2):
        if row1 != row2:
            print("Stats:", row1)
            print("Compare:", row2)

    for row in filtered2:
        if row not in filtered1:
            print("Missing:", row)


if __name__ == "__main__":
    import argparse

    parser = argparse.ArgumentParser(
        description="Compare two CSV files and output the differences."
    )
    parser.add_argument("file1", type=str, help="Path to the first CSV file.")
    parser.add_argument("file2", type=str, help="Path to the second CSV file.")

    args = parser.parse_args()

    diff_csv_files(args.file1, args.file2)