Analyze the output "nvcc -time ..."

nvcc-time-analyzer.py

#!/usr/bin/env python

import argparse
import pandas as pd

TO_MS = {
    " s"  : 1000,
    " ms" : 1,
    " us" : 0.001,
    " ns" : 0.000001,
}


def parse_args():
    argparser = argparse.ArgumentParser("`nvcc -time` visualizer")
    argparser.add_argument("csv", type=str,
                           help="CSV file containing the output of `nvcc -time`")
    args = argparser.parse_args()
    return args


def phasewise(df):
    df_by_phase = df.groupby(" phase name ")
    phases = {
        "phase"    : [],
        "time(ms)" : [],
    }
    for phase, grp in df_by_phase:
        phases["phase"].append(phase)
        phases["time(ms)"].append(grp["time(ms)"].sum())
    phase = pd.DataFrame(phases)
    phase.sort_values(by="time(ms)", ascending=False, inplace=True)
    print("Total cpu time (in ms) spent phase-wise:")
    for ind in phase.index:
        print("%32s : %13.3f" % (phase["phase"][ind], phase["time(ms)"][ind]))
    pass


def filewise(df):
    df_by_file = df.groupby("source file name ")
    files = {
        "file"     : [],
        "time(ms)" : [],
    }
    for file, grp in df_by_file:
        files["file"].append(file)
        files["time(ms)"].append(grp["time(ms)"].sum())
    file = pd.DataFrame(files)
    file.sort_values(by="time(ms)", ascending=False, inplace=True)
    breakdown = {}
    print("Total cpu time (in ms) spent file-wise, across the top 10 files:")
    for id, ind in enumerate(file.index):
        f = file["file"][ind]
        print("%13.3f : %s" % (file["time(ms)"][ind], f))
        breakdown[f] = df_by_file.get_group(f).sort_values(by="time(ms)", ascending=False)
        if id > 10:
            break
    print("Breakdown of total cpu time (in ms) spent file-wise, across the top 10 files:")
    for id, ind in enumerate(file.index):
        f = file["file"][ind]
        print("  %s:" % f)
        for j in breakdown[f].index:
            print("    %32s : %13.3f" % (breakdown[f][" phase name "][j], breakdown[f]["time(ms)"][j]))
        if id > 10:
            break
    pass


def main():
    args = parse_args()
    df = pd.read_csv(args.csv)
    df["time(ms)"]  = df.apply(lambda row: row[" metric "] * TO_MS[row[" unit"]], axis=1)
    phasewise(df)
    filewise(df)


if __name__ == "__main__":
    main()