main() — pytorch Function Reference

Architecture documentation for the main() function in join_results.py from the pytorch codebase.

Function python DynamoBenchmarks Microbenchmarking calls 1

Entity Profile

DynamoBenchmarks→ Microbenchmarking→ main() — pytorch Function Reference

Dependency Diagram

graph TD
  ed3a07d3_3bc2_b1e3_91eb_6fe78c42020a["main()"]
  941c9c45_672f_2175_f07a_216a4fb53eee["longest_common_prefix()"]
  ed3a07d3_3bc2_b1e3_91eb_6fe78c42020a -->|calls| 941c9c45_672f_2175_f07a_216a4fb53eee
  style ed3a07d3_3bc2_b1e3_91eb_6fe78c42020a fill:#6366f1,stroke:#818cf8,color:#fff

Relationship Graph

Source Code

benchmarks/dynamo/join_results.py lines 23–53

def main():
    parser = argparse.ArgumentParser()
    parser.add_argument("--field", "-f", default="speedup", type=str)
    parser.add_argument("--output", "-o", type=str)
    parser.add_argument("inputs", nargs="*")
    args = parser.parse_args()

    prefix = longest_common_prefix([Path(inp).stem for inp in args.inputs])
    frames = []
    fields = []
    for inp in args.inputs:
        field = Path(inp).stem[len(prefix) :]
        fields.append(field)
        frames.append(
            pd.read_csv(inp)
            .filter(["name", args.field])
            .rename(columns={args.field: field})
        )

    df = frames[0]
    for other in frames[1:]:
        df = df.merge(other, how="outer", on="name")
    df = df.fillna(0)

    # drop rows where all backends failed
    df = df[functools.reduce(operator.or_, [df[f] != 0 for f in fields])]

    prefix = prefix.strip("_") or "output"
    output = args.output or f"{prefix}.csv"
    print(f"Writing {output}")
    df.to_csv(output, index=False)