This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: [PATCH] Replace old AWK script (utilizing bc) with Python implementation
- From: Martin LiÅka <mliska at suse dot cz>
- To: Matthias Klose <doko at ubuntu dot com>, GCC Patches <gcc-patches at gcc dot gnu dot org>
- Cc: Jan Hubicka <hubicka at ucw dot cz>
- Date: Tue, 26 Apr 2016 15:58:03 +0200
- Subject: Re: [PATCH] Replace old AWK script (utilizing bc) with Python implementation
- Authentication-results: sourceware.org; auth=none
- References: <571E3052 dot 8040909 at suse dot cz> <571E698F dot 8060503 at ubuntu dot com>
On 04/25/2016 09:01 PM, Matthias Klose wrote:
> please could you make the shebang python3? Not sure if it's good to replace one old implementation with a soon to become old implementation.
>
> Matthias
Sure, thanks for pointing out.
Attaching v2, where I changed:
+ switched from python2 to python3
+ added a new column with better readable coverage values:
HEURISTICS BRANCHES (REL) HITRATE COVERAGE COVERAGE (REL)
loop iv compare 70 0.1% 59.31% / 71.45% 391732444 391.73M 0.0%
unconditional jump 252 0.2% 100.00% / 100.00% 62269 62.27K 0.0%
guess loop iv compare 362 0.3% 89.02% / 90.08% 2703184164 2.70G 0.2%
...
As I've just tested the patch, I runs significantly faster for SPEC2006 profile dumps:
real 0m1.162s vs. real 0m17.962s
Martin
>From 38ee7451629e5fe7d9d2468bf24e1929193be2a6 Mon Sep 17 00:00:00 2001
From: marxin <mliska@suse.cz>
Date: Mon, 25 Apr 2016 16:42:42 +0200
Subject: [PATCH] Replace AWK script with the python script.
contrib/ChangeLog:
2016-04-25 Martin Liska <mliska@suse.cz>
* analyze_brprob: Remove.
* analyze_brprob.py: New file.
---
contrib/analyze_brprob | 147 ----------------------------------------------
contrib/analyze_brprob.py | 136 ++++++++++++++++++++++++++++++++++++++++++
2 files changed, 136 insertions(+), 147 deletions(-)
delete mode 100755 contrib/analyze_brprob
create mode 100755 contrib/analyze_brprob.py
diff --git a/contrib/analyze_brprob b/contrib/analyze_brprob
deleted file mode 100755
index 5702834..0000000
--- a/contrib/analyze_brprob
+++ /dev/null
@@ -1,147 +0,0 @@
-#!/usr/bin/awk -f
-# Script to analyze experimental results of our branch prediction heuristics
-# Contributed by Jan Hubicka, SuSE Inc.
-# Copyright (C) 2001, 2003 Free Software Foundation, Inc.
-#
-# 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 COPYING. If not, write to
-# the Free Software Foundation, 51 Franklin Street, Fifth Floor,
-# Boston, MA 02110-1301, USA.
-#
-#
-# 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.c.
-#
-# Usage:
-# Step 1: Compile and profile your program. You need to use -fprofile-arcs
-# flag to get the profiles
-# Step 2: Generate log files. The information about given heuristics are
-# saved into ipa-profile dumps. You need to pass the -fdimp-ipa-profile switch
-# to the compiler as well
-# as -fbranch-probabilities to get the results of profiling noted in the dumps.
-# Ensure that there are no "Arc profiling: some edge counts were bad." warnings.
-# Step 3: Run this script to concatenate all *.profile files:
-# analyze_brprob `find . -name *.profile`
-# the information is collected and print once all files are parsed. This
-# may take a while.
-# Note that the script does use bc to perform long arithmetic.
-# Step 4: Read the results. Basically the following table is printed:
-# (this is just an example from a very early stage of branch prediction pass
-# development, so please don't take these numbers seriously)
-#
-#HEURISTICS BRANCHES (REL) HITRATE COVERAGE (REL)
-#opcode 2889 83.7% 94.96%/ 97.62% 7516383 75.3%
-#pointer 246 7.1% 99.69%/ 99.86% 118791 1.2%
-#loop header 449 13.0% 98.32%/ 99.07% 43553 0.4%
-#first match 3450 100.0% 89.92%/ 97.27% 9979782 100.0%
-#loop exit 924 26.8% 88.95%/ 95.58% 9026266 90.4%
-#error return 150 4.3% 64.48%/ 86.81% 453542 4.5%
-#call 803 23.3% 51.66%/ 98.61% 3614037 36.2%
-#loop branch 51 1.5% 99.26%/ 99.27% 26854 0.3%
-#noreturn call 951 27.6% 100.00%/100.00% 1759809 17.6%
-#
-# The heuristic called "first match" is a heuristic used by GCC branch
-# prediction pass and it predicts 89.92% branches correctly.
-#
-# The quality of heuristics can be rated using both, coverage and hitrate
-# parameters. For example "loop branch" heuristics (predicting loopback edge
-# as taken) have both very high hitrate and coverage, so it is very useful.
-# On the other hand, "exit block" heuristics (predicting exit edges as not
-# taken) have good hitrate, but poor coverage, so only 3 branches have been
-# predicted. The "loop header" heuristic has problems, since it tends to
-# misspredict.
-#
-# The implementation of this script is somewhat brute force. My awk skills
-# are limited.
-
-function longeval(e)
-{
- e = "echo \"scale = 2 ;"e"\" | bc"
- e | getline res
- close (e)
- return res
-}
-
-BEGIN {nnames = 0}
-
-/^ .* heuristics: .*.$/ {
- name=$0
- sub (/^ /,"",name)
- sub (/ heuristics: .*.$/,"",name)
- if (!(name in branches))
- {
- names[nnames] = name
- branches[name]=0
- counts[name]=0
- hits[name]=0
- phits[name]=0
- nnames++
- }
- branches[name]+=1
- }
-
-/^ .* heuristics: .*. exec [0-9]* hit [0-9]* (.*.)$/ {
- name=$0
- sub (/^ /,"",name)
- sub (/ heuristics: .*. exec [0-9]* hit [0-9]* (.*.)$/,"",name)
- pred=$0
- sub (/^ .* heuristics: /,"",pred)
- sub (/. exec [0-9]* hit [0-9]* (.*.)$/,"",pred)
- count=$0
- sub (/^ .* heuristics: .*. exec /,"",count)
- sub (/ hit [0-9]* (.*.)$/,"",count)
- hit=$0
- sub (/^ .* heuristics: .*. exec [0-9]* hit /,"",hit)
- sub (/ (.*.)$/,"",hit)
-
- if (int(pred) < 50.0)
- {
- hit = count"-"hit;
- }
- counts[name]=counts[name] "+" count
- hits[name]=hits[name] "+" hit
- phits[name]=phits[name] "+(("hit")<"count"/2)*("count"-("hit"))+(("hit")>="count"/2)*("hit")"
-
- #BC crashes on long strings. Irritating.
- if (length(counts[name]) > 2000)
- counts[name] = longeval(counts[name])
- if (length(hits[name]) > 2000)
- hits[name] = longeval(hits[name])
- if (length(phits[name]) > 2000)
- phits[name] = longeval(phits[name])
- }
-END {
- # Heuristics called combined predicts just everything.
- maxcounts = longeval(counts["combined"])
- maxbranches = branches["combined"]
- max = names["combined"]
- printf("HEURISTICS BRANCHES (REL) HITRATE COVERAGE (REL)\n")
- for (i = 0; i < nnames ; i++)
- {
- name = names[i]
- counts[name] = longeval(counts[name])
- printf ("%-26s %8i %5.1f%% %6s%% / %6s%% %12s %5.1f%%\n",
- name,
- branches[name], branches[name] * 100 / maxbranches,
- longeval("("hits[name]") * 100 /(" counts[name]"-0.00001)"),
- longeval("("phits[name]") * 100 /(" counts[name]"-0.00001)"),
- counts[name], longeval(counts[name]" * 100 / ("maxcounts"-0.00001)"))
- }
-}
diff --git a/contrib/analyze_brprob.py b/contrib/analyze_brprob.py
new file mode 100755
index 0000000..36371ff
--- /dev/null
+++ b/contrib/analyze_brprob.py
@@ -0,0 +1,136 @@
+#!/usr/bin/env python3
+#
+# 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
+# <http://www.gnu.org/licenses/>. */
+#
+#
+#
+# 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
+
+def percentage(a, b):
+ return 100.0 * a / b
+
+class Summary:
+ def __init__(self, name):
+ self.name = name
+ self.branches = 0
+ self.count = 0
+ self.hits = 0
+ self.fits = 0
+
+ def count_formatted(self):
+ v = self.count
+ for unit in ['','K','M','G','T','P','E','Z']:
+ if v < 1000:
+ return "%3.2f%s" % (v, unit)
+ v /= 1000.0
+ return "%.1f%s" % (v, 'Y')
+
+class Profile:
+ def __init__(self, filename):
+ self.filename = filename
+ self.heuristics = {}
+
+ def add(self, name, prediction, count, hits):
+ if not name in self.heuristics:
+ self.heuristics[name] = Summary(name)
+
+ s = self.heuristics[name]
+ s.branches += 1
+ s.count += count
+ if prediction < 50:
+ hits = count - hits
+ s.hits += hits
+ s.fits += max(hits, count - hits)
+
+ def branches_max(self):
+ return max([v.branches for k, v in self.heuristics.items()])
+
+ def count_max(self):
+ return max([v.count for k, v in self.heuristics.items()])
+
+ def dump(self):
+ print('%-36s %8s %6s %-16s %14s %8s %6s' % ('HEURISTICS', 'BRANCHES', '(REL)',
+ 'HITRATE', 'COVERAGE', 'COVERAGE', '(REL)'))
+ for (k, v) in sorted(self.heuristics.items(), key = lambda x: x[1].branches):
+ print('%-36s %8i %5.1f%% %6.2f%% / %6.2f%% %14i %8s %5.1f%%' %
+ (k, v.branches, percentage(v.branches, self.branches_max ()),
+ percentage(v.hits, v.count), percentage(v.fits, v.count),
+ v.count, v.count_formatted(), percentage(v.count, self.count_max()) ))
+
+if len(sys.argv) != 2:
+ print('Usage: ./analyze_brprob.py dump_file')
+ exit(1)
+
+profile = Profile(sys.argv[1])
+r = re.compile(' (.*) heuristics: (.*)%.*exec ([0-9]*) hit ([0-9]*)')
+for l in open(profile.filename).readlines():
+ m = r.match(l)
+ if m != None:
+ name = m.group(1)
+ prediction = float(m.group(2))
+ count = int(m.group(3))
+ hits = int(m.group(4))
+
+ profile.add(name, prediction, count, hits)
+
+profile.dump()
--
2.8.1