Modulized the bm_*.py functions

5 files changed, 250 insertions, 153 deletions

diff --git a/tools/profiling/microbenchmarks/bm_build.py b/tools/profiling/microbenchmarks/bm_build.py
new file mode 100755
index 0000000000..13ca1dfd1b
--- /dev/null
+++ b/tools/profiling/microbenchmarks/bm_build.py
@@ -0,0 +1,68 @@
+#!/usr/bin/env python2.7
+# Copyright 2017, Google Inc.
+# All rights reserved.
+#
+# Redistribution and use in source and binary forms, with or without
+# modification, are permitted provided that the following conditions are
+# met:
+#
+#     * Redistributions of source code must retain the above copyright
+# notice, this list of conditions and the following disclaimer.
+#     * Redistributions in binary form must reproduce the above
+# copyright notice, this list of conditions and the following disclaimer
+# in the documentation and/or other materials provided with the
+# distribution.
+#     * Neither the name of Google Inc. nor the names of its
+# contributors may be used to endorse or promote products derived from
+# this software without specific prior written permission.
+#
+# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+# "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+# LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+# A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+# OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+# SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+# LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+# DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+# THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+# (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+### Python utility to build opt and counters benchmarks """
+
+import bm_constants
+
+import argparse
+import subprocess
+import multiprocessing
+import os
+import shutil
+
+def _args():
+  argp = argparse.ArgumentParser(description='Builds microbenchmarks')
+  argp.add_argument('-b', '--benchmarks', nargs='+', choices=bm_constants._AVAILABLE_BENCHMARK_TESTS, default=bm_constants._AVAILABLE_BENCHMARK_TESTS)
+  argp.add_argument('-j', '--jobs', type=int, default=multiprocessing.cpu_count())
+  argp.add_argument('-n', '--name', type=str, help='Unique name of this build')
+  return argp.parse_args()
+
+def _make_cmd(cfg, jobs, benchmarks):
+  return ['make'] + benchmarks + [
+      'CONFIG=%s' % cfg, '-j', '%d' % jobs]
+
+def build(name, jobs, benchmarks):
+  shutil.rmtree('bm_diff_%s' % name, ignore_errors=True)
+  subprocess.check_call(['git', 'submodule', 'update'])
+  try:
+    subprocess.check_call(_make_cmd('opt', jobs, benchmarks))
+    subprocess.check_call(_make_cmd('counters', jobs, benchmarks))
+  except subprocess.CalledProcessError, e:
+    subprocess.check_call(['make', 'clean'])
+    subprocess.check_call(_make_cmd('opt', jobs, benchmarks))
+    subprocess.check_call(_make_cmd('counters', jobs, benchmarks))
+  os.rename('bins', 'bm_diff_%s' % name, )
+
+if __name__ == '__main__':
+  args = _args()
+  build(args.name, args.jobs, args.benchmarks)
+
+
diff --git a/tools/profiling/microbenchmarks/bm_constants.py b/tools/profiling/microbenchmarks/bm_constants.py
new file mode 100644
index 0000000000..ada1e32e72
--- /dev/null
+++ b/tools/profiling/microbenchmarks/bm_constants.py
@@ -0,0 +1,56 @@
+#!/usr/bin/env python2.7
+# Copyright 2017, Google Inc.
+# All rights reserved.
+#
+# Redistribution and use in source and binary forms, with or without
+# modification, are permitted provided that the following conditions are
+# met:
+#
+#     * Redistributions of source code must retain the above copyright
+# notice, this list of conditions and the following disclaimer.
+#     * Redistributions in binary form must reproduce the above
+# copyright notice, this list of conditions and the following disclaimer
+# in the documentation and/or other materials provided with the
+# distribution.
+#     * Neither the name of Google Inc. nor the names of its
+# contributors may be used to endorse or promote products derived from
+# this software without specific prior written permission.
+#
+# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+# "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+# LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+# A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+# OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+# SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+# LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+# DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+# THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+# (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+### Configurable constants for the bm_*.py family """
+
+_AVAILABLE_BENCHMARK_TESTS = ['bm_fullstack_unary_ping_pong',
+                              'bm_fullstack_streaming_ping_pong',
+                              'bm_fullstack_streaming_pump',
+                              'bm_closure',
+                              'bm_cq',
+                              'bm_call_create',
+                              'bm_error',
+                              'bm_chttp2_hpack',
+                              'bm_chttp2_transport',
+                              'bm_pollset',
+                              'bm_metadata',
+                              'bm_fullstack_trickle']
+
+
+_INTERESTING = (
+  'cpu_time',
+  'real_time',
+  'locks_per_iteration',
+  'allocs_per_iteration',
+  'writes_per_iteration',
+  'atm_cas_per_iteration',
+  'atm_add_per_iteration',
+  'nows_per_iteration',
+)
diff --git a/tools/profiling/microbenchmarks/bm_diff.py b/tools/profiling/microbenchmarks/bm_diff.py
index 299abb5fdb..4de318fec7 100755
--- a/tools/profiling/microbenchmarks/bm_diff.py
+++ b/tools/profiling/microbenchmarks/bm_diff.py
@@ -28,47 +28,18 @@
 # (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
 # OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
-import sys
-import json
+""" Computes the diff between two bm runs and outputs significant results """
+
 import bm_json
+import bm_constants
+import bm_speedup
+
+import json
 import tabulate
 import argparse
-from scipy import stats
-import subprocess
-import multiprocessing
 import collections
-import pipes
-import os
-sys.path.append(os.path.join(os.path.dirname(sys.argv[0]), '..', '..', 'run_tests', 'python_utils'))
-import comment_on_pr
-import jobset
-import itertools
-import speedup
-import random
-import shutil
-import errno
-
-_INTERESTING = (
-  'cpu_time',
-  'real_time',
-  'locks_per_iteration',
-  'allocs_per_iteration',
-  'writes_per_iteration',
-  'atm_cas_per_iteration',
-  'atm_add_per_iteration',
-  'cli_transport_stalls_per_iteration',
-  'cli_stream_stalls_per_iteration',
-  'svr_transport_stalls_per_iteration',
-  'svr_stream_stalls_per_iteration'
-  'nows_per_iteration',
-)
 
-def changed_ratio(n, o):
-  if float(o) <= .0001: o = 0
-  if float(n) <= .0001: n = 0
-  if o == 0 and n == 0: return 0
-  if o == 0: return 100
-  return (float(n)-float(o))/float(o)
+verbose = False
 
 def median(ary):
   ary = sorted(ary)
@@ -78,91 +49,27 @@ def median(ary):
   else:
     return ary[n/2]
 
-def min_change(pct):
-  return lambda n, o: abs(changed_ratio(n,o)) > pct/100.0
-
-_AVAILABLE_BENCHMARK_TESTS = ['bm_fullstack_unary_ping_pong',
-                              'bm_fullstack_streaming_ping_pong',
-                              'bm_fullstack_streaming_pump',
-                              'bm_closure',
-                              'bm_cq',
-                              'bm_call_create',
-                              'bm_error',
-                              'bm_chttp2_hpack',
-                              'bm_chttp2_transport',
-                              'bm_pollset',
-                              'bm_metadata',
-                              'bm_fullstack_trickle']
-
-argp = argparse.ArgumentParser(description='Perform diff on microbenchmarks')
-argp.add_argument('-t', '--track',
-                  choices=sorted(_INTERESTING),
-                  nargs='+',
-                  default=sorted(_INTERESTING),
-                  help='Which metrics to track')
-argp.add_argument('-b', '--benchmarks', nargs='+', choices=_AVAILABLE_BENCHMARK_TESTS, default=['bm_cq'])
-argp.add_argument('-d', '--diff_base', type=str)
-argp.add_argument('-r', '--repetitions', type=int, default=1)
-argp.add_argument('-l', '--loops', type=int, default=20)
-argp.add_argument('-j', '--jobs', type=int, default=multiprocessing.cpu_count())
-args = argp.parse_args()
-
-assert args.diff_base
-
-def avg(lst):
-  sum = 0.0
-  n = 0.0
-  for el in lst:
-    sum += el
-    n += 1
-  return sum / n
-
-def make_cmd(cfg):
-  return ['make'] + args.benchmarks + [
-      'CONFIG=%s' % cfg, '-j', '%d' % args.jobs]
-
-def build(dest):
-  shutil.rmtree('bm_diff_%s' % dest, ignore_errors=True)
-  subprocess.check_call(['git', 'submodule', 'update'])
-  try:
-    subprocess.check_call(make_cmd('opt'))
-    subprocess.check_call(make_cmd('counters'))
-  except subprocess.CalledProcessError, e:
-    subprocess.check_call(['make', 'clean'])
-    subprocess.check_call(make_cmd('opt'))
-    subprocess.check_call(make_cmd('counters'))
-  os.rename('bins', 'bm_diff_%s' % dest)
-
-def collect1(bm, cfg, ver, idx):
-  cmd = ['bm_diff_%s/%s/%s' % (ver, cfg, bm),
-         '--benchmark_out=%s.%s.%s.%d.json' % (bm, cfg, ver, idx),
-         '--benchmark_out_format=json',
-         '--benchmark_repetitions=%d' % (args.repetitions)
-         ]
-  return jobset.JobSpec(cmd, shortname='%s %s %s %d/%d' % (bm, cfg, ver, idx+1, args.loops),
-                             verbose_success=True, timeout_seconds=None)
-
-build('new')
-
-where_am_i = subprocess.check_output(['git', 'rev-parse', '--abbrev-ref', 'HEAD']).strip()
-subprocess.check_call(['git', 'checkout', args.diff_base])
-try:
-  build('old')
-finally:
-  subprocess.check_call(['git', 'checkout', where_am_i])
-  subprocess.check_call(['git', 'submodule', 'update'])
-
-jobs = []
-for loop in range(0, args.loops):
-  jobs.extend(x for x in itertools.chain(
-    (collect1(bm, 'opt', 'new', loop) for bm in args.benchmarks),
-    (collect1(bm, 'counters', 'new', loop) for bm in args.benchmarks),
-    (collect1(bm, 'opt', 'old', loop) for bm in args.benchmarks),
-    (collect1(bm, 'counters', 'old', loop) for bm in args.benchmarks),
-  ))
-random.shuffle(jobs, random.SystemRandom().random)
-
-jobset.run(jobs, maxjobs=args.jobs)
+def _args():
+  argp = argparse.ArgumentParser(description='Perform diff on microbenchmarks')
+  argp.add_argument('-t', '--track',
+                    choices=sorted(bm_constants._INTERESTING),
+                    nargs='+',
+                    default=sorted(bm_constants._INTERESTING),
+                    help='Which metrics to track')
+  argp.add_argument('-b', '--benchmarks', nargs='+', choices=bm_constants._AVAILABLE_BENCHMARK_TESTS, default=bm_constants._AVAILABLE_BENCHMARK_TESTS)
+  argp.add_argument('-l', '--loops', type=int, default=20)
+  argp.add_argument('-n', '--new', type=str, help='New benchmark name')
+  argp.add_argument('-o', '--old', type=str, help='Old benchmark name')
+  argp.add_argument('-v', '--verbose', type=bool, help='print details of before/after')
+  args = argp.parse_args()
+  global verbose
+  if args.verbose: verbose = True
+  assert args.new
+  assert args.old
+  return args
+
+def maybe_print(str):
+  if verbose: print str
 
 class Benchmark:
 
@@ -173,18 +80,18 @@ class Benchmark:
     }
     self.final = {}
 
-  def add_sample(self, data, new):
-    for f in args.track:
+  def add_sample(self, track, data, new):
+    for f in track:
       if f in data:
         self.samples[new][f].append(float(data[f]))
 
-  def process(self):
-    for f in sorted(args.track):
+  def process(self, track):
+    for f in sorted(track):
       new = self.samples[True][f]
       old = self.samples[False][f]
       if not new or not old: continue
       mdn_diff = abs(median(new) - median(old))
-      print '%s: new=%r old=%r mdn_diff=%r' % (f, new, old, mdn_diff)
+      maybe_print('%s: new=%r old=%r mdn_diff=%r' % (f, new, old, mdn_diff))
       s = speedup.speedup(new, old)
       if abs(s) > 3 and mdn_diff > 0.5:
         self.final[f] = '%+d%%' % s
@@ -196,29 +103,17 @@ class Benchmark:
   def row(self, flds):
     return [self.final[f] if f in self.final else '' for f in flds]
 
-
-def eintr_be_gone(fn):
-  """Run fn until it doesn't stop because of EINTR"""
-  while True:
-    try:
-      return fn()
-    except IOError, e:
-      if e.errno != errno.EINTR:
-        raise
-
-
 def read_json(filename):
   try:
     with open(filename) as f: return json.loads(f.read())
   except ValueError, e:
     return None
 
-
-def finalize():
+def finalize(bms, loops, track):
   benchmarks = collections.defaultdict(Benchmark)
 
-  for bm in args.benchmarks:
-    for loop in range(0, args.loops):
+  for bm in bms:
+    for loop in range(0, loops):
       js_new_ctr = read_json('%s.counters.new.%d.json' % (bm, loop))
       js_new_opt = read_json('%s.opt.new.%d.json' % (bm, loop))
       js_old_ctr = read_json('%s.counters.old.%d.json' % (bm, loop))
@@ -226,22 +121,20 @@ def finalize():
 
       if js_new_ctr:
         for row in bm_json.expand_json(js_new_ctr, js_new_opt):
-          print row
           name = row['cpp_name']
           if name.endswith('_mean') or name.endswith('_stddev'): continue
-          benchmarks[name].add_sample(row, True)
+          benchmarks[name].add_sample(track, row, True)
       if js_old_ctr:
         for row in bm_json.expand_json(js_old_ctr, js_old_opt):
-          print row
           name = row['cpp_name']
           if name.endswith('_mean') or name.endswith('_stddev'): continue
-          benchmarks[name].add_sample(row, False)
+          benchmarks[name].add_sample(track, row, False)
 
   really_interesting = set()
   for name, bm in benchmarks.items():
-    print name
-    really_interesting.update(bm.process())
-  fields = [f for f in args.track if f in really_interesting]
+    maybe_print(name)
+    really_interesting.update(bm.process(track))
+  fields = [f for f in track if f in really_interesting]
 
   headers = ['Benchmark'] + fields
   rows = []
@@ -249,11 +142,12 @@ def finalize():
     if benchmarks[name].skip(): continue
     rows.append([name] + benchmarks[name].row(fields))
   if rows:
-    text = 'Performance differences noted:\n' + tabulate.tabulate(rows, headers=headers, floatfmt='+.2f')
+    return tabulate.tabulate(rows, headers=headers, floatfmt='+.2f')
   else:
-    text = 'No significant performance differences'
-  print text
-  comment_on_pr.comment_on_pr('```\n%s\n```' % text)
+    return None
+
+if __name__ == '__main__':
+  args = _args()
+  print finalize(args.benchmarks, args.loops, args.track)
 
 
-eintr_be_gone(finalize)
diff --git a/tools/profiling/microbenchmarks/bm_run.py b/tools/profiling/microbenchmarks/bm_run.py
new file mode 100755
index 0000000000..458e319440
--- /dev/null
+++ b/tools/profiling/microbenchmarks/bm_run.py
@@ -0,0 +1,77 @@
+#!/usr/bin/env python2.7
+# Copyright 2017, Google Inc.
+# All rights reserved.
+#
+# Redistribution and use in source and binary forms, with or without
+# modification, are permitted provided that the following conditions are
+# met:
+#
+#     * Redistributions of source code must retain the above copyright
+# notice, this list of conditions and the following disclaimer.
+#     * Redistributions in binary form must reproduce the above
+# copyright notice, this list of conditions and the following disclaimer
+# in the documentation and/or other materials provided with the
+# distribution.
+#     * Neither the name of Google Inc. nor the names of its
+# contributors may be used to endorse or promote products derived from
+# this software without specific prior written permission.
+#
+# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+# "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+# LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+# A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+# OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+# SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+# LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+# DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+# THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+# (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+### Python utility to run opt and counters benchmarks and save json output """
+
+import bm_constants
+
+import argparse
+import multiprocessing
+import random
+import itertools
+import sys
+import os
+
+sys.path.append(os.path.join(os.path.dirname(sys.argv[0]), '..', '..', 'run_tests', 'python_utils'))
+import jobset
+
+def _args():
+  argp = argparse.ArgumentParser(description='Runs microbenchmarks')
+  argp.add_argument('-b', '--benchmarks', nargs='+', choices=bm_constants._AVAILABLE_BENCHMARK_TESTS, default=bm_constants._AVAILABLE_BENCHMARK_TESTS)
+  argp.add_argument('-j', '--jobs', type=int, default=multiprocessing.cpu_count())
+  argp.add_argument('-n', '--name', type=str, help='Unique name of this build')
+  argp.add_argument('-r', '--repetitions', type=int, default=1)
+  argp.add_argument('-l', '--loops', type=int, default=20)
+  return argp.parse_args()
+
+def _collect_bm_data(bm, cfg, name, reps, idx):
+  cmd = ['bm_diff_%s/%s/%s' % (name, cfg, bm),
+         '--benchmark_out=%s.%s.%s.%d.json' % (bm, cfg, name, idx),
+         '--benchmark_out_format=json',
+         '--benchmark_repetitions=%d' % (reps)
+         ]
+  return jobset.JobSpec(cmd, shortname='%s %s %s %d/%d' % (bm, cfg, name, idx+1, args.loops),
+                             verbose_success=True, timeout_seconds=None)
+
+def _run_bms(benchmarks, name, loops, reps):
+  jobs = []
+  for loop in range(0, loops):
+    jobs.extend(x for x in itertools.chain(
+      (_collect_bm_data(bm, 'opt', name, reps, loop) for bm in benchmarks),
+      (_collect_bm_data(bm, 'counters', name, reps, loop) for bm in benchmarks),
+    ))
+  random.shuffle(jobs, random.SystemRandom().random)
+
+  jobset.run(jobs, maxjobs=args.jobs)
+
+if __name__ == '__main__':
+  args = _args()
+  assert args.name
+  _run_bms(args.benchmarks, args.name, args.loops, args.repetitions)
diff --git a/tools/profiling/microbenchmarks/speedup.py b/tools/profiling/microbenchmarks/bm_speedup.py
index 8af0066c9d..9e395a782e 100644
--- a/tools/profiling/microbenchmarks/speedup.py
+++ b/tools/profiling/microbenchmarks/bm_speedup.py
@@ -27,6 +27,8 @@
 # (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
 # OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
+""" The math behind the diff functionality """
+
 from scipy import stats
 import math