1 #!/usr/bin/python3
2 # Generate Intel taken branches Linux perf event script for autofdo profiling.
3
4 # Copyright (C) 2016-2023 Free Software Foundation, Inc.
5 #
6 # GCC is free software; you can redistribute it and/or modify it under
7 # the terms of the GNU General Public License as published by the Free
8 # Software Foundation; either version 3, or (at your option) any later
9 # version.
10 #
11 # GCC is distributed in the hope that it will be useful, but WITHOUT ANY
12 # WARRANTY; without even the implied warranty of MERCHANTABILITY or
13 # FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
14 # for more details.
15 #
16 # You should have received a copy of the GNU General Public License
17 # along with GCC; see the file COPYING3. If not see
18 # <http://www.gnu.org/licenses/>. */
19
20 # Run it with perf record -b -e EVENT program ...
21 # The Linux Kernel needs to support the PMU of the current CPU, and
22 # It will likely not work in VMs.
23 # Add --all to print for all cpus, otherwise for current cpu.
24 # Add --script to generate shell script to run correct event.
25 #
26 # Requires internet (https) access. This may require setting up a proxy
27 # with export https_proxy=...
28 #
29 import urllib.request
30 import sys
31 import json
32 import argparse
33 import collections
34 import os
35
36 baseurl = "https://download.01.org/perfmon"
37
38 target_events = ('BR_INST_RETIRED.NEAR_TAKEN',
39 'BR_INST_EXEC.TAKEN',
40 'BR_INST_RETIRED.TAKEN_JCC',
41 'BR_INST_TYPE_RETIRED.COND_TAKEN')
42
43 ap = argparse.ArgumentParser()
44 ap.add_argument('--all', '-a', help='Print for all CPUs', action='store_true')
45 ap.add_argument('--script', help='Generate shell script', action='store_true')
46 args = ap.parse_args()
47
48 eventmap = collections.defaultdict(list)
49
50 def get_cpustr():
51 cpuinfo = os.getenv("CPUINFO")
52 if cpuinfo is None:
53 cpuinfo = '/proc/cpuinfo'
54 f = open(cpuinfo, 'r')
55 cpu = [None, None, None, None]
56 for j in f:
57 n = j.split()
58 if n[0] == 'vendor_id':
59 cpu[0] = n[2]
60 elif n[0] == 'model' and n[1] == ':':
61 cpu[2] = int(n[2])
62 elif n[0] == 'cpu' and n[1] == 'family':
63 cpu[1] = int(n[3])
64 elif n[0] == 'stepping' and n[1] == ':':
65 cpu[3] = int(n[2])
66 if all(v is not None for v in cpu):
67 break
68 # stepping for SKX only
69 stepping = cpu[0] == "GenuineIntel" and cpu[1] == 6 and cpu[2] == 0x55
70 if stepping:
71 return "%s-%d-%X-%X" % tuple(cpu)
72 return "%s-%d-%X" % tuple(cpu)[:3]
73
74 def find_event(eventurl, model):
75 print("Downloading", eventurl, file = sys.stderr)
76 u = urllib.request.urlopen(eventurl)
77 events = json.loads(u.read())
78 u.close()
79
80 found = 0
81 for j in events:
82 if j['EventName'] in target_events:
83 event = "cpu/event=%s,umask=%s/" % (j['EventCode'], j['UMask'])
84 if 'PEBS' in j and int(j['PEBS']) > 0:
85 event += "p"
86 if args.script:
87 eventmap[event].append(model)
88 else:
89 print(j['EventName'], "event for model", model, "is", event)
90 found += 1
91 return found
92
93 if not args.all:
94 cpu = get_cpustr()
95 if not cpu:
96 sys.exit("Unknown CPU type")
97
98 url = baseurl + "/mapfile.csv"
99 print("Downloading", url, file = sys.stderr)
100 u = urllib.request.urlopen(url)
101 found = 0
102 cpufound = 0
103 for j in u:
104 n = j.rstrip().decode().split(',')
105 if len(n) >= 4 and (args.all or n[0] == cpu) and n[3] == "core":
106 components = n[0].split("-")
107 model = components[2]
108 model = int(model, 16)
109 cpufound += 1
110 found += find_event(baseurl + n[2], model)
111 u.close()
112
113 if args.script:
114 print('''#!/bin/sh
115 # Profile workload for gcc profile feedback (autofdo) using Linux perf.
116 # Auto generated. To regenerate for new CPUs run
117 # contrib/gen_autofdo_event.py --script --all in gcc source
118
119 # usages:
120 # gcc-auto-profile program (profile program and children)
121 # gcc-auto-profile -a sleep X (profile all for X secs, may need root)
122 # gcc-auto-profile -p PID sleep X (profile PID)
123 # gcc-auto-profile --kernel -a sleep X (profile kernel)
124 # gcc-auto-profile --all -a sleep X (profile kernel and user space)
125
126 # Identify branches taken event for CPU.
127 #
128
129 FLAGS=u
130
131 if [ "$1" = "--kernel" ] ; then
132 FLAGS=k
133 shift
134 fi
135 if [ "$1" = "--all" ] ; then
136 FLAGS=uk
137 shift
138 fi
139
140 if ! grep -q Intel /proc/cpuinfo ; then
141 echo >&2 "Only Intel CPUs supported"
142 exit 1
143 fi
144
145 if grep -q hypervisor /proc/cpuinfo ; then
146 echo >&2 "Warning: branch profiling may not be functional in VMs"
147 fi
148
149 case `grep -E -q "^cpu family\s*: 6" /proc/cpuinfo &&
150 grep -E "^model\s*:" /proc/cpuinfo | head -n1` in''')
151 for event, mod in eventmap.items():
152 for m in mod[:-1]:
153 print("model*:\ %s|\\" % m)
154 print('model*:\ %s) E="%s$FLAGS" ;;' % (mod[-1], event))
155 print('''*)
156 echo >&2 "Unknown CPU. Run contrib/gen_autofdo_event.py --all --script to update script."
157 exit 1 ;;''')
158 print("esac")
159 print("set -x")
160 print('if ! perf record -e $E -b "$@" ; then')
161 print(' # PEBS may not actually be working even if the processor supports it')
162 print(' # (e.g., in a virtual machine). Trying to run without /p.')
163 print(' set +x')
164 print(' echo >&2 "Retrying without /p."')
165 print(' E="$(echo "${E}" | sed -e \'s/\/p/\//\')"')
166 print(' set -x')
167 print(' exec perf record -e $E -b "$@"')
168 print(' set +x')
169 print('fi')
170
171 if cpufound == 0 and not args.all:
172 sys.exit('CPU %s not found' % cpu)
173
174 if found == 0:
175 sys.exit('Branch event not found')