2016-04-28 12:02:37 +00:00
|
|
|
#!/usr/bin/env python3
|
2022-12-14 12:50:03 +00:00
|
|
|
|
2024-01-03 11:19:35 +00:00
|
|
|
# Copyright (C) 2016-2024 Free Software Foundation, Inc.
|
2016-04-28 12:02:37 +00:00
|
|
|
#
|
|
|
|
# Script to analyze results of our branch prediction heuristics
|
|
|
|
#
|
|
|
|
# This file is part of GCC.
|
|
|
|
#
|
|
|
|
# GCC is free software; you can redistribute it and/or modify it under
|
|
|
|
# the terms of the GNU General Public License as published by the Free
|
|
|
|
# Software Foundation; either version 3, or (at your option) any later
|
|
|
|
# version.
|
|
|
|
#
|
|
|
|
# GCC is distributed in the hope that it will be useful, but WITHOUT ANY
|
|
|
|
# WARRANTY; without even the implied warranty of MERCHANTABILITY or
|
|
|
|
# FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
|
|
|
|
# for more details.
|
|
|
|
#
|
|
|
|
# You should have received a copy of the GNU General Public License
|
|
|
|
# along with GCC; see the file COPYING3. If not see
|
2024-01-04 15:01:20 +00:00
|
|
|
# <http://www.gnu.org/licenses/>.
|
2016-04-28 12:02:37 +00:00
|
|
|
#
|
|
|
|
#
|
|
|
|
#
|
|
|
|
# This script is used to calculate two basic properties of the branch prediction
|
|
|
|
# heuristics - coverage and hitrate. Coverage is number of executions
|
|
|
|
# of a given branch matched by the heuristics and hitrate is probability
|
|
|
|
# that once branch is predicted as taken it is really taken.
|
|
|
|
#
|
|
|
|
# These values are useful to determine the quality of given heuristics.
|
|
|
|
# Hitrate may be directly used in predict.def.
|
|
|
|
#
|
|
|
|
# Usage:
|
|
|
|
# Step 1: Compile and profile your program. You need to use -fprofile-generate
|
|
|
|
# flag to get the profiles.
|
|
|
|
# Step 2: Make a reference run of the intrumented application.
|
|
|
|
# Step 3: Compile the program with collected profile and dump IPA profiles
|
|
|
|
# (-fprofile-use -fdump-ipa-profile-details)
|
|
|
|
# Step 4: Collect all generated dump files:
|
|
|
|
# find . -name '*.profile' | xargs cat > dump_file
|
|
|
|
# Step 5: Run the script:
|
|
|
|
# ./analyze_brprob.py dump_file
|
|
|
|
# and read results. Basically the following table is printed:
|
|
|
|
#
|
|
|
|
# HEURISTICS BRANCHES (REL) HITRATE COVERAGE (REL)
|
|
|
|
# early return (on trees) 3 0.2% 35.83% / 93.64% 66360 0.0%
|
|
|
|
# guess loop iv compare 8 0.6% 53.35% / 53.73% 11183344 0.0%
|
|
|
|
# call 18 1.4% 31.95% / 69.95% 51880179 0.2%
|
|
|
|
# loop guard 23 1.8% 84.13% / 84.85% 13749065956 42.2%
|
|
|
|
# opcode values positive (on trees) 42 3.3% 15.71% / 84.81% 6771097902 20.8%
|
|
|
|
# opcode values nonequal (on trees) 226 17.6% 72.48% / 72.84% 844753864 2.6%
|
|
|
|
# loop exit 231 18.0% 86.97% / 86.98% 8952666897 27.5%
|
|
|
|
# loop iterations 239 18.6% 91.10% / 91.10% 3062707264 9.4%
|
|
|
|
# DS theory 281 21.9% 82.08% / 83.39% 7787264075 23.9%
|
|
|
|
# no prediction 293 22.9% 46.92% / 70.70% 2293267840 7.0%
|
|
|
|
# guessed loop iterations 313 24.4% 76.41% / 76.41% 10782750177 33.1%
|
|
|
|
# first match 708 55.2% 82.30% / 82.31% 22489588691 69.0%
|
|
|
|
# combined 1282 100.0% 79.76% / 81.75% 32570120606 100.0%
|
|
|
|
#
|
|
|
|
#
|
|
|
|
# The heuristics called "first match" is a heuristics used by GCC branch
|
|
|
|
# prediction pass and it predicts 55.2% branches correctly. As you can,
|
|
|
|
# the heuristics has very good covertage (69.05%). On the other hand,
|
|
|
|
# "opcode values nonequal (on trees)" heuristics has good hirate, but poor
|
|
|
|
# coverage.
|
|
|
|
|
|
|
|
import sys
|
|
|
|
import os
|
|
|
|
import re
|
2016-06-09 11:38:23 +00:00
|
|
|
import argparse
|
2016-04-28 12:02:37 +00:00
|
|
|
|
2016-06-24 16:22:44 +00:00
|
|
|
from math import *
|
|
|
|
|
2016-12-14 10:30:24 +00:00
|
|
|
counter_aggregates = set(['combined', 'first match', 'DS theory',
|
|
|
|
'no prediction'])
|
2018-01-19 12:03:24 +00:00
|
|
|
hot_threshold = 10
|
2016-12-14 10:30:24 +00:00
|
|
|
|
2016-04-28 12:02:37 +00:00
|
|
|
def percentage(a, b):
|
|
|
|
return 100.0 * a / b
|
|
|
|
|
2016-06-24 16:22:44 +00:00
|
|
|
def average(values):
|
|
|
|
return 1.0 * sum(values) / len(values)
|
|
|
|
|
|
|
|
def average_cutoff(values, cut):
|
|
|
|
l = len(values)
|
|
|
|
skip = floor(l * cut / 2)
|
|
|
|
if skip > 0:
|
|
|
|
values.sort()
|
|
|
|
values = values[skip:-skip]
|
|
|
|
return average(values)
|
|
|
|
|
|
|
|
def median(values):
|
|
|
|
values.sort()
|
|
|
|
return values[int(len(values) / 2)]
|
|
|
|
|
2017-05-30 07:16:31 +00:00
|
|
|
class PredictDefFile:
|
|
|
|
def __init__(self, path):
|
|
|
|
self.path = path
|
|
|
|
self.predictors = {}
|
|
|
|
|
|
|
|
def parse_and_modify(self, heuristics, write_def_file):
|
|
|
|
lines = [x.rstrip() for x in open(self.path).readlines()]
|
|
|
|
|
|
|
|
p = None
|
|
|
|
modified_lines = []
|
2022-01-31 15:39:02 +00:00
|
|
|
for i, l in enumerate(lines):
|
2017-05-30 07:16:31 +00:00
|
|
|
if l.startswith('DEF_PREDICTOR'):
|
2022-01-31 15:39:02 +00:00
|
|
|
next_line = lines[i + 1]
|
|
|
|
if l.endswith(','):
|
|
|
|
l += next_line
|
2017-05-30 07:16:31 +00:00
|
|
|
m = re.match('.*"(.*)".*', l)
|
|
|
|
p = m.group(1)
|
|
|
|
elif l == '':
|
|
|
|
p = None
|
|
|
|
|
|
|
|
if p != None:
|
|
|
|
heuristic = [x for x in heuristics if x.name == p]
|
|
|
|
heuristic = heuristic[0] if len(heuristic) == 1 else None
|
|
|
|
|
|
|
|
m = re.match('.*HITRATE \(([^)]*)\).*', l)
|
|
|
|
if (m != None):
|
|
|
|
self.predictors[p] = int(m.group(1))
|
|
|
|
|
|
|
|
# modify the line
|
|
|
|
if heuristic != None:
|
|
|
|
new_line = (l[:m.start(1)]
|
|
|
|
+ str(round(heuristic.get_hitrate()))
|
|
|
|
+ l[m.end(1):])
|
|
|
|
l = new_line
|
|
|
|
p = None
|
|
|
|
elif 'PROB_VERY_LIKELY' in l:
|
|
|
|
self.predictors[p] = 100
|
|
|
|
modified_lines.append(l)
|
|
|
|
|
|
|
|
# save the file
|
|
|
|
if write_def_file:
|
|
|
|
with open(self.path, 'w+') as f:
|
|
|
|
for l in modified_lines:
|
|
|
|
f.write(l + '\n')
|
2018-01-19 12:03:24 +00:00
|
|
|
class Heuristics:
|
|
|
|
def __init__(self, count, hits, fits):
|
|
|
|
self.count = count
|
|
|
|
self.hits = hits
|
|
|
|
self.fits = fits
|
2017-05-30 07:16:31 +00:00
|
|
|
|
2016-04-28 12:02:37 +00:00
|
|
|
class Summary:
|
|
|
|
def __init__(self, name):
|
|
|
|
self.name = name
|
2018-01-19 12:03:24 +00:00
|
|
|
self.edges= []
|
|
|
|
|
|
|
|
def branches(self):
|
|
|
|
return len(self.edges)
|
|
|
|
|
|
|
|
def hits(self):
|
|
|
|
return sum([x.hits for x in self.edges])
|
|
|
|
|
|
|
|
def fits(self):
|
|
|
|
return sum([x.fits for x in self.edges])
|
|
|
|
|
|
|
|
def count(self):
|
|
|
|
return sum([x.count for x in self.edges])
|
|
|
|
|
|
|
|
def successfull_branches(self):
|
|
|
|
return len([x for x in self.edges if 2 * x.hits >= x.count])
|
2016-04-28 12:02:37 +00:00
|
|
|
|
2016-06-09 11:38:23 +00:00
|
|
|
def get_hitrate(self):
|
2018-01-19 12:03:24 +00:00
|
|
|
return 100.0 * self.hits() / self.count()
|
2016-12-14 10:30:24 +00:00
|
|
|
|
|
|
|
def get_branch_hitrate(self):
|
2018-01-19 12:03:24 +00:00
|
|
|
return 100.0 * self.successfull_branches() / self.branches()
|
2016-06-09 11:38:23 +00:00
|
|
|
|
2016-04-28 12:02:37 +00:00
|
|
|
def count_formatted(self):
|
2018-01-19 12:03:24 +00:00
|
|
|
v = self.count()
|
2017-11-08 07:24:14 +00:00
|
|
|
for unit in ['', 'k', 'M', 'G', 'T', 'P', 'E', 'Z', 'Y']:
|
2016-04-28 12:02:37 +00:00
|
|
|
if v < 1000:
|
|
|
|
return "%3.2f%s" % (v, unit)
|
|
|
|
v /= 1000.0
|
|
|
|
return "%.1f%s" % (v, 'Y')
|
|
|
|
|
2018-01-19 12:03:24 +00:00
|
|
|
def count(self):
|
|
|
|
return sum([x.count for x in self.edges])
|
|
|
|
|
2017-05-30 07:16:31 +00:00
|
|
|
def print(self, branches_max, count_max, predict_def):
|
2018-01-19 12:03:24 +00:00
|
|
|
# filter out most hot edges (if requested)
|
|
|
|
self.edges = sorted(self.edges, reverse = True, key = lambda x: x.count)
|
|
|
|
if args.coverage_threshold != None:
|
|
|
|
threshold = args.coverage_threshold * self.count() / 100
|
|
|
|
edges = [x for x in self.edges if x.count < threshold]
|
|
|
|
if len(edges) != 0:
|
|
|
|
self.edges = edges
|
|
|
|
|
2017-05-30 07:16:31 +00:00
|
|
|
predicted_as = None
|
|
|
|
if predict_def != None and self.name in predict_def.predictors:
|
|
|
|
predicted_as = predict_def.predictors[self.name]
|
|
|
|
|
2016-12-14 10:30:24 +00:00
|
|
|
print('%-40s %8i %5.1f%% %11.2f%% %7.2f%% / %6.2f%% %14i %8s %5.1f%%' %
|
2018-01-19 12:03:24 +00:00
|
|
|
(self.name, self.branches(),
|
|
|
|
percentage(self.branches(), branches_max),
|
2016-12-14 10:30:24 +00:00
|
|
|
self.get_branch_hitrate(),
|
|
|
|
self.get_hitrate(),
|
2018-01-19 12:03:24 +00:00
|
|
|
percentage(self.fits(), self.count()),
|
|
|
|
self.count(), self.count_formatted(),
|
|
|
|
percentage(self.count(), count_max)), end = '')
|
2017-05-30 07:16:31 +00:00
|
|
|
|
|
|
|
if predicted_as != None:
|
|
|
|
print('%12i%% %5.1f%%' % (predicted_as,
|
|
|
|
self.get_hitrate() - predicted_as), end = '')
|
2018-01-19 12:03:24 +00:00
|
|
|
else:
|
|
|
|
print(' ' * 20, end = '')
|
|
|
|
|
|
|
|
# print details about the most important edges
|
|
|
|
if args.coverage_threshold == None:
|
|
|
|
edges = [x for x in self.edges[:100] if x.count * hot_threshold > self.count()]
|
|
|
|
if args.verbose:
|
|
|
|
for c in edges:
|
|
|
|
r = 100.0 * c.count / self.count()
|
|
|
|
print(' %.0f%%:%d' % (r, c.count), end = '')
|
|
|
|
elif len(edges) > 0:
|
|
|
|
print(' %0.0f%%:%d' % (100.0 * sum([x.count for x in edges]) / self.count(), len(edges)), end = '')
|
|
|
|
|
2017-05-30 07:16:31 +00:00
|
|
|
print()
|
2016-12-14 10:30:24 +00:00
|
|
|
|
2016-04-28 12:02:37 +00:00
|
|
|
class Profile:
|
|
|
|
def __init__(self, filename):
|
|
|
|
self.filename = filename
|
|
|
|
self.heuristics = {}
|
2016-06-24 16:22:44 +00:00
|
|
|
self.niter_vector = []
|
2016-04-28 12:02:37 +00:00
|
|
|
|
|
|
|
def add(self, name, prediction, count, hits):
|
|
|
|
if not name in self.heuristics:
|
|
|
|
self.heuristics[name] = Summary(name)
|
|
|
|
|
|
|
|
s = self.heuristics[name]
|
2016-12-14 10:30:24 +00:00
|
|
|
|
2016-04-28 12:02:37 +00:00
|
|
|
if prediction < 50:
|
|
|
|
hits = count - hits
|
2016-12-14 10:30:24 +00:00
|
|
|
remaining = count - hits
|
2018-01-19 12:03:24 +00:00
|
|
|
fits = max(hits, remaining)
|
2016-12-14 10:30:24 +00:00
|
|
|
|
2018-01-19 12:03:24 +00:00
|
|
|
s.edges.append(Heuristics(count, hits, fits))
|
2016-04-28 12:02:37 +00:00
|
|
|
|
2016-06-24 16:22:44 +00:00
|
|
|
def add_loop_niter(self, niter):
|
|
|
|
if niter > 0:
|
|
|
|
self.niter_vector.append(niter)
|
|
|
|
|
2016-04-28 12:02:37 +00:00
|
|
|
def branches_max(self):
|
2018-01-19 12:03:24 +00:00
|
|
|
return max([v.branches() for k, v in self.heuristics.items()])
|
2016-04-28 12:02:37 +00:00
|
|
|
|
|
|
|
def count_max(self):
|
2018-01-19 12:03:24 +00:00
|
|
|
return max([v.count() for k, v in self.heuristics.items()])
|
2016-04-28 12:02:37 +00:00
|
|
|
|
2017-05-30 07:16:31 +00:00
|
|
|
def print_group(self, sorting, group_name, heuristics, predict_def):
|
2016-12-14 10:30:24 +00:00
|
|
|
count_max = self.count_max()
|
|
|
|
branches_max = self.branches_max()
|
|
|
|
|
2018-01-19 12:03:24 +00:00
|
|
|
sorter = lambda x: x.branches()
|
2016-12-14 10:30:24 +00:00
|
|
|
if sorting == 'branch-hitrate':
|
|
|
|
sorter = lambda x: x.get_branch_hitrate()
|
|
|
|
elif sorting == 'hitrate':
|
|
|
|
sorter = lambda x: x.get_hitrate()
|
2016-06-09 11:38:23 +00:00
|
|
|
elif sorting == 'coverage':
|
2016-12-14 10:30:24 +00:00
|
|
|
sorter = lambda x: x.count
|
|
|
|
elif sorting == 'name':
|
|
|
|
sorter = lambda x: x.name.lower()
|
|
|
|
|
2018-01-19 12:03:24 +00:00
|
|
|
print('%-40s %8s %6s %12s %18s %14s %8s %6s %12s %6s %s' %
|
2016-12-14 10:30:24 +00:00
|
|
|
('HEURISTICS', 'BRANCHES', '(REL)',
|
2017-05-30 07:16:31 +00:00
|
|
|
'BR. HITRATE', 'HITRATE', 'COVERAGE', 'COVERAGE', '(REL)',
|
2018-01-19 12:03:24 +00:00
|
|
|
'predict.def', '(REL)', 'HOT branches (>%d%%)' % hot_threshold))
|
2016-12-14 10:30:24 +00:00
|
|
|
for h in sorted(heuristics, key = sorter):
|
2017-05-30 07:16:31 +00:00
|
|
|
h.print(branches_max, count_max, predict_def)
|
2016-12-14 10:30:24 +00:00
|
|
|
|
|
|
|
def dump(self, sorting):
|
|
|
|
heuristics = self.heuristics.values()
|
|
|
|
if len(heuristics) == 0:
|
|
|
|
print('No heuristics available')
|
|
|
|
return
|
|
|
|
|
2017-05-30 07:16:31 +00:00
|
|
|
predict_def = None
|
|
|
|
if args.def_file != None:
|
|
|
|
predict_def = PredictDefFile(args.def_file)
|
|
|
|
predict_def.parse_and_modify(heuristics, args.write_def_file)
|
|
|
|
|
2016-12-14 10:30:24 +00:00
|
|
|
special = list(filter(lambda x: x.name in counter_aggregates,
|
|
|
|
heuristics))
|
|
|
|
normal = list(filter(lambda x: x.name not in counter_aggregates,
|
|
|
|
heuristics))
|
2016-06-09 11:38:23 +00:00
|
|
|
|
2017-05-30 07:16:31 +00:00
|
|
|
self.print_group(sorting, 'HEURISTICS', normal, predict_def)
|
2016-12-14 10:30:24 +00:00
|
|
|
print()
|
2017-05-30 07:16:31 +00:00
|
|
|
self.print_group(sorting, 'HEURISTIC AGGREGATES', special, predict_def)
|
2016-04-28 12:02:37 +00:00
|
|
|
|
2016-07-21 12:53:24 +00:00
|
|
|
if len(self.niter_vector) > 0:
|
|
|
|
print ('\nLoop count: %d' % len(self.niter_vector)),
|
|
|
|
print(' avg. # of iter: %.2f' % average(self.niter_vector))
|
|
|
|
print(' median # of iter: %.2f' % median(self.niter_vector))
|
|
|
|
for v in [1, 5, 10, 20, 30]:
|
|
|
|
cut = 0.01 * v
|
2016-12-14 10:30:24 +00:00
|
|
|
print(' avg. (%d%% cutoff) # of iter: %.2f'
|
|
|
|
% (v, average_cutoff(self.niter_vector, cut)))
|
2016-06-24 16:22:44 +00:00
|
|
|
|
2016-06-09 11:38:23 +00:00
|
|
|
parser = argparse.ArgumentParser()
|
2016-12-14 10:30:24 +00:00
|
|
|
parser.add_argument('dump_file', metavar = 'dump_file',
|
|
|
|
help = 'IPA profile dump file')
|
|
|
|
parser.add_argument('-s', '--sorting', dest = 'sorting',
|
|
|
|
choices = ['branches', 'branch-hitrate', 'hitrate', 'coverage', 'name'],
|
|
|
|
default = 'branches')
|
2017-05-30 07:16:31 +00:00
|
|
|
parser.add_argument('-d', '--def-file', help = 'path to predict.def')
|
|
|
|
parser.add_argument('-w', '--write-def-file', action = 'store_true',
|
|
|
|
help = 'Modify predict.def file in order to set new numbers')
|
2018-01-19 12:03:24 +00:00
|
|
|
parser.add_argument('-c', '--coverage-threshold', type = int,
|
|
|
|
help = 'Ignore edges that have percentage coverage >= coverage-threshold')
|
|
|
|
parser.add_argument('-v', '--verbose', action = 'store_true', help = 'Print verbose informations')
|
2016-06-09 11:38:23 +00:00
|
|
|
|
|
|
|
args = parser.parse_args()
|
2016-04-28 12:02:37 +00:00
|
|
|
|
2017-05-30 07:16:31 +00:00
|
|
|
profile = Profile(args.dump_file)
|
2016-06-24 16:22:44 +00:00
|
|
|
loop_niter_str = ';; profile-based iteration count: '
|
2018-01-19 12:03:24 +00:00
|
|
|
|
2017-05-30 07:16:31 +00:00
|
|
|
for l in open(args.dump_file):
|
2018-01-19 12:03:24 +00:00
|
|
|
if l.startswith(';;heuristics;'):
|
|
|
|
parts = l.strip().split(';')
|
|
|
|
assert len(parts) == 8
|
|
|
|
name = parts[3]
|
|
|
|
prediction = float(parts[6])
|
|
|
|
count = int(parts[4])
|
|
|
|
hits = int(parts[5])
|
2016-04-28 12:02:37 +00:00
|
|
|
|
|
|
|
profile.add(name, prediction, count, hits)
|
2016-06-24 16:22:44 +00:00
|
|
|
elif l.startswith(loop_niter_str):
|
|
|
|
v = int(l[len(loop_niter_str):])
|
|
|
|
profile.add_loop_niter(v)
|
2016-04-28 12:02:37 +00:00
|
|
|
|
2016-06-09 11:38:23 +00:00
|
|
|
profile.dump(args.sorting)
|