swift-mirror/benchmark/scripts/run_smoke_bench

#!/usr/bin/env python
# -*- coding: utf-8 -*-

# ===--- run_smoke_bench -------------------------------------------------===//
#
#  This source file is part of the Swift.org open source project
#
#  Copyright (c) 2014 - 2017 Apple Inc. and the Swift project authors
#  Licensed under Apache License v2.0 with Runtime Library Exception
#
#  See https://swift.org/LICENSE.txt for license information
#  See https://swift.org/CONTRIBUTORS.txt for the list of Swift project authors
#
# ===---------------------------------------------------------------------===//
#
# Performs a very fast check which benchmarks regressed and improved.
#
# Initially runs the benchmars with a low sample count and just re-runs those
# benchmarks which differ.
# Also reports code size differences.
#
# ===---------------------------------------------------------------------===//

from __future__ import print_function

import argparse
import glob
import os
import subprocess
import sys
from imp import load_source

from compare_perf_tests import LogParser, TestComparator, create_report

# import Benchmark_Driver  # doesn't work because it misses '.py' extension
Benchmark_Driver = load_source(
    "Benchmark_Driver",
    os.path.join(os.path.dirname(os.path.abspath(__file__)), "Benchmark_Driver"),
)
# from Benchmark_Driver import BenchmarkDriver, BenchmarkDoctor, ...
BenchmarkDriver = Benchmark_Driver.BenchmarkDriver
BenchmarkDoctor = Benchmark_Driver.BenchmarkDoctor
MarkdownReportHandler = Benchmark_Driver.MarkdownReportHandler

VERBOSE = False


class DriverArgs(object):
    """Arguments for BenchmarkDriver."""

    def __init__(self, tests, optimization="O"):
        """Initialize with path to the build-dir and optimization level."""
        self.benchmarks = None
        self.filters = None
        self.tests = os.path.join(tests, "bin")
        self.optimization = optimization


def log(msg):
    print(msg)
    sys.stdout.flush()


def main():
    global VERBOSE
    argparser = argparse.ArgumentParser()
    argparser.add_argument(
        "-verbose", action="store_true", help="print verbose messages"
    )
    argparser.add_argument(
        "-O",
        action="append_const",
        const="O",
        dest="opt_levels",
        help="test -O benchmarks",
    )
    argparser.add_argument(
        "-Osize",
        action="append_const",
        const="Osize",
        dest="opt_levels",
        help="test -Osize benchmarks",
    )
    argparser.add_argument(
        "-Onone",
        action="append_const",
        const="Onone",
        dest="opt_levels",
        help="test -Onone benchmarks (except code size)",
    )
    argparser.add_argument(
        "-skip-code-size",
        action="store_true",
        help="Don't report code size differences",
    )
    argparser.add_argument(
        "-skip-performance",
        action="store_true",
        help="Don't report performance differences",
    )
    argparser.add_argument(
        "-skip-check-added",
        action="store_true",
        help="Don't validate newly added benchmarks",
    )
    argparser.add_argument(
        "-o",
        type=str,
        help="In addition to stdout, write the results into a markdown file",
    )
    argparser.add_argument(
        "-threshold",
        type=float,
        help="The performance threshold in %% which triggers a re-run",
        default=5,
    )
    argparser.add_argument(
        "-num-samples",
        type=int,
        help="The (minimum) number of samples to run",
        default=3,
    )
    argparser.add_argument(
        "-num-reruns",
        type=int,
        help="The number of re-runs until it's assumed to be a real change",
        default=8,
    )
    argparser.add_argument(
        "-platform", type=str, help="The benchmark build platform", default="macosx"
    )
    argparser.add_argument(
        "oldbuilddir", nargs=1, type=str, help="old benchmark build directory"
    )
    argparser.add_argument(
        "newbuilddir", nargs=1, type=str, help="new benchmark build directory"
    )
    args = argparser.parse_args()
    VERBOSE = args.verbose

    return test_opt_levels(args)


def test_opt_levels(args):
    output_file = None
    if args.o:
        output_file = open(args.o, "w")

    changes = False
    for opt_level in args.opt_levels or ["O", "Osize", "Onone"]:
        log("####### Testing optimization level -" + opt_level + " #######")
        if not args.skip_performance:
            if test_performance(
                opt_level,
                args.oldbuilddir[0],
                args.newbuilddir[0],
                float(args.threshold) / 100,
                args.num_samples,
                args.num_reruns,
                output_file,
            ):
                changes = True

        # There is no point in reporting code size for Onone.
        if not args.skip_code_size and opt_level != "Onone":
            if report_code_size(
                opt_level,
                args.oldbuilddir[0],
                args.newbuilddir[0],
                args.platform,
                output_file,
            ):
                changes = True

    if not args.skip_code_size:
        if report_code_size(
            "swiftlibs",
            args.oldbuilddir[0],
            args.newbuilddir[0],
            args.platform,
            output_file,
        ):
            changes = True

    if not args.skip_check_added:
        check_added(args, output_file)

    if output_file:
        if changes:
            output_file.write(get_info_text())
        else:
            output_file.write("### No performance and code size changes")
        output_file.close()
    return 0


def measure(driver, tests, i):
    """Log and measure samples of the tests with the given driver.

    Collect increasing number of samples, depending on the iteration.
    """
    num_samples = min(i + 3, 10)
    msg = "    Iteration {0} for {1}: num samples = {2}, ".format(
        i, driver.args.tests, num_samples
    )
    msg += (
        "running all tests"
        if driver.all_tests == tests
        else "re-testing {0} tests".format(len(tests))
    )
    log(msg)
    driver.tests = tests
    return driver.run(num_samples=num_samples, sample_time=0.0025)


def merge(results, other_results):
    """"Merge the other PreformanceTestResults into the first dictionary."""
    for test, result in other_results.items():
        results[test].merge(result)
    return results


def test_performance(
    opt_level, old_dir, new_dir, threshold, num_samples, num_reruns, output_file
):
    """Detect performance changes in benchmarks.

    Start fast with few samples per benchmark and gradually spend more time
    gathering more precise measurements of the change candidates.
    """

    i, unchanged_length_count = 0, 0
    old, new = [
        BenchmarkDriver(DriverArgs(dir, optimization=opt_level))
        for dir in [old_dir, new_dir]
    ]
    results = [measure(driver, driver.tests, i) for driver in [old, new]]
    tests = TestComparator(results[0], results[1], threshold)
    changed = tests.decreased + tests.increased

    while len(changed) > 0 and unchanged_length_count < num_reruns:
        i += 1
        if VERBOSE:
            log("        test again: " + str([test.name for test in changed]))
        results = [
            merge(the_results, measure(driver, [test.name for test in changed], i))
            for the_results, driver in zip(results, [old, new])
        ]
        tests = TestComparator(results[0], results[1], threshold)
        changed = tests.decreased + tests.increased

        if len(old.tests) == len(changed):
            unchanged_length_count += 1
        else:
            unchanged_length_count = 0

    log("")
    return report_results(
        "Performance: -" + opt_level, None, None, threshold * 1.4, output_file, *results
    )


def report_code_size(opt_level, old_dir, new_dir, platform, output_file):
    if opt_level == "swiftlibs":
        files = glob.glob(os.path.join(old_dir, "lib", "swift", platform, "*.dylib"))
    else:
        files = glob.glob(
            os.path.join(old_dir, opt_level + "-*" + platform + "*", "*.o")
        )

    idx = 1
    old_lines = ""
    new_lines = ""
    for oldfile in files:
        newfile = oldfile.replace(old_dir, new_dir, 1)
        if os.path.isfile(newfile):
            oldsize = get_codesize(oldfile)
            newsize = get_codesize(newfile)
            bname = os.path.basename(oldfile)

            def result_line(value):
                v = "," + str(value)
                return str(idx) + "," + bname + ",1" + (v * 3) + ",0" + v + "\n"

            old_lines += result_line(oldsize)
            new_lines += result_line(newsize)
            idx += 1

    return report_results(
        "Code size: -" + opt_level, old_lines, new_lines, 0.01, output_file
    )


def get_codesize(filename):
    output = subprocess.check_output(["size", filename]).splitlines()
    header_line = output[0]
    data_line = output[1]
    if header_line.find("__TEXT") != 0:
        sys.exit("unexpected output from size command:\n" + output)
    return int(data_line.split("\t")[0])


def report_results(
    title,
    old_lines,
    new_lines,
    threshold,
    output_file,
    old_results=None,
    new_results=None,
):
    old_results = old_results or LogParser.results_from_string(old_lines)
    new_results = new_results or LogParser.results_from_string(new_lines)

    print("------- " + title + " -------")
    print(create_report(old_results, new_results, threshold, "git"))

    if output_file:
        report = create_report(old_results, new_results, threshold, "markdown")
        if report != "":
            output_file.write("### " + title + "\n")
            output_file.write(report)
            output_file.write("\n")
            return True
    return False


def get_info_text():
    text = """
<details>
  <summary><strong>How to read the data</strong></summary>
The tables contain differences in performance which are larger than 8% and
differences in code size which are larger than 1%.

If you see any unexpected regressions, you should consider fixing the
regressions before you merge the PR.

**Noise**: Sometimes the performance results (not code size!) contain false
alarms. Unexpected regressions which are marked with '(?)' are probably noise.
If you see regressions which you cannot explain you can try to run the
benchmarks again. If regressions still show up, please consult with the
performance team (@eeckstein).
</details>

<details>
  <summary><strong>Hardware Overview</strong></summary>

"""
    po = subprocess.check_output(["system_profiler", "SPHardwareDataType"])
    for line in po.splitlines():
        selection = [
            "Model Name",
            "Model Identifier",
            "Processor Name",
            "Processor Speed",
            "Number of Processors",
            "Total Number of Cores",
            "L2 Cache",
            "L3 Cache",
            "Memory:",
        ]

        if any(s in line for s in selection):
            text += line + "\n"

    text += """
</details>"""
    return text


def check_added(args, output_file=None):
    old = BenchmarkDriver(DriverArgs(args.oldbuilddir[0]))
    new = BenchmarkDriver(DriverArgs(args.newbuilddir[0]))
    added = set(new.tests).difference(set(old.tests))
    new.tests = list(added)
    doctor = BenchmarkDoctor(args, driver=new)
    if added and output_file:
        doctor.log.addHandler(MarkdownReportHandler(output_file))
    doctor.check()


if __name__ == "__main__":
    sys.exit(main())