PX4-Autopilot/Debug/poor-mans-profiler.sh
Pavel Kirienko 2b2c307eac Performance audit (intentionally duplicates #6829) (#6847)
* UAVCAN ESC output: removing ESC output channels from published message that are always zero. This allows the UAVCAN stack to always transfer only the minimum number of output values, avoiding redundant zeroes and the associated increase in bus load and CPU time

* Added a separate mixer file for CAN quadrotor

* Sampling profiler improvements

* PMSP: Output more endpoints

* Matrix update

* libc usage workaround

* Removed UAVCAN perfcounters

* Matrix submodule update
2017-03-18 04:16:20 -10:00

271 lines
8.0 KiB
Bash
Executable File

#!/bin/bash
#
# Author: Pavel Kirienko <pavel.kirienko@zubax.com>
#
# Poor man's sampling profiler for NuttX.
#
# Usage: Install flamegraph.pl in your PATH, configure your .gdbinit, run the script with proper arguments and go
# have a coffee. When you're back, you'll see the flamegraph. Note that frequent calls to GDB significantly
# interfere with normal operation of the target, which means that you can't profile real-time tasks with it.
# For best results, ensure that the PC is not overloaded, the USB host controller to which the debugger is
# connected is not congested. You should also allow the current user to set negative nice values.
#
# The FlameGraph script can be downloaded from https://github.com/brendangregg/FlameGraph. Thanks Mr. Gregg.
#
# Requirements: ARM GDB with Python support. You can get one by downloading the sources from
# https://launchpad.net/gcc-arm-embedded and building them with correct flags.
# Note that Python support is not required if no per-task sampling is needed.
#
set -e
root=$(dirname $0)/..
function die()
{
echo "$@"
exit 1
}
function usage()
{
echo "Invalid usage. Supported options:"
cat $0 | sed -n 's/^\s*--\([^)\*]*\).*/\1/p' # Don't try this at home.
exit 1
}
which flamegraph.pl > /dev/null || die "Install flamegraph.pl first"
#
# Parsing the arguments. Read this section for usage info.
#
nsamples=0
sleeptime=0.1 # Doctors recommend 7-8 hours a day
taskname=
elf=
append=0
fgfontsize=10
fgwidth=1900
for i in "$@"
do
case $i in
--nsamples=*)
nsamples="${i#*=}"
;;
--sleeptime=*)
sleeptime="${i#*=}"
;;
--taskname=*)
taskname="${i#*=}"
;;
--elf=*)
elf="${i#*=}"
;;
--append)
append=1
;;
--fgfontsize=*)
fgfontsize="${i#*=}"
;;
--fgwidth=*)
fgwidth="${i#*=}"
;;
*)
usage
;;
esac
shift
done
[[ -z "$elf" ]] && die "Please specify the ELF file location, e.g.: build_px4fmu-v4_default/src/firmware/nuttx/firmware_nuttx"
#
# Temporary files
#
stacksfile=/tmp/pmpn-stacks.log
foldfile=/tmp/pmpn-folded.txt
graphfile=/tmp/pmpn-flamegraph.svg
gdberrfile=/tmp/pmpn-gdberr.log
#
# Sampling if requested. Note that if $append is true, the stack file will not be rewritten.
#
cd $root
if [[ $nsamples > 0 ]]
then
[[ $append = 0 ]] && (rm -f $stacksfile; echo "Old stacks removed")
echo "Sampling the task '$taskname'..."
for x in $(seq 1 $nsamples)
do
if [[ "$taskname" = "" ]]
then
arm-none-eabi-gdb $elf --batch -ex "set print asm-demangle on" -ex bt \
2> $gdberrfile \
| sed -n 's/\(#.*\)/\1/p' \
>> $stacksfile
else
arm-none-eabi-gdb $elf --batch -ex "set print asm-demangle on" \
-ex "source $root/Debug/Nuttx.py" \
-ex "show mybt $taskname" \
2> $gdberrfile \
| sed -n 's/0\.0:\(#.*\)/\1/p' \
>> $stacksfile
fi
echo -e '\n\n' >> $stacksfile
echo -ne "\r$x/$nsamples"
sleep $sleeptime
done
echo
echo "Stacks saved to $stacksfile"
else
echo "Sampling skipped - set 'nsamples' to re-sample."
fi
#
# Folding the stacks.
#
[ -f $stacksfile ] || die "Where are the stack samples?"
cat << 'EOF' > /tmp/pmpn-folder.py
#
# This stack folder correctly handles C++ types.
#
from __future__ import print_function, division
import fileinput, collections, os, sys
def enforce(x, msg='Invalid input'):
if not x:
raise Exception(msg)
def split_first_part_with_parens(line):
LBRACES = {'(':'()', '<':'<>', '[':'[]', '{':'{}'}
RBRACES = {')':'()', '>':'<>', ']':'[]', '}':'{}'}
QUOTES = set(['"', "'"])
quotes = collections.defaultdict(bool)
braces = collections.defaultdict(int)
out = ''
for ch in line:
out += ch
# escape character cancels further processing
if ch == '\\':
continue
# special cases
if out.endswith('operator>') or out.endswith('operator>>') or out.endswith('operator->'): # gotta love c++
braces['<>'] += 1
if out.endswith('operator<') or out.endswith('operator<<'):
braces['<>'] -= 1
# switching quotes
if ch in QUOTES:
quotes[ch] = not quotes[ch]
# counting parens only when outside quotes
if sum(quotes.values()) == 0:
if ch in LBRACES.keys():
braces[LBRACES[ch]] += 1
if ch in RBRACES.keys():
braces[RBRACES[ch]] -= 1
# sanity check
for v in braces.values():
enforce(v >= 0, 'Unaligned braces: ' + str(dict(braces)))
# termination condition
if ch == ' ' and sum(braces.values()) == 0:
break
out = out.strip()
return out, line[len(out):]
def parse(line):
def take_path(line, output):
line = line.strip()
if line.startswith('at '):
line = line[3:].strip()
if line:
output['file_full_path'] = line.rsplit(':', 1)[0].strip()
output['file_base_name'] = os.path.basename(output['file_full_path'])
output['line'] = int(line.rsplit(':', 1)[1])
return output
def take_args(line, output):
line = line.lstrip()
if line[0] == '(':
output['args'], line = split_first_part_with_parens(line)
return take_path(line.lstrip(), output)
def take_function(line, output):
output['function'], line = split_first_part_with_parens(line.lstrip())
return take_args(line.lstrip(), output)
def take_mem_loc(line, output):
line = line.lstrip()
if line.startswith('0x'):
end = line.find(' ')
num = line[:end]
output['memloc'] = int(num, 16)
line = line[end:].lstrip()
end = line.find(' ')
enforce(line[:end] == 'in')
line = line[end:].lstrip()
return take_function(line, output)
def take_frame_num(line, output):
line = line.lstrip()
enforce(line[0] == '#')
end = line.find(' ')
num = line[1:end]
output['frame_num'] = int(num)
return take_mem_loc(line[end:], output)
return take_frame_num(line, {})
stacks = collections.defaultdict(int)
current = ''
stack_tops = collections.defaultdict(int)
num_stack_frames = 0
for idx,line in enumerate(fileinput.input()):
try:
line = line.strip()
if line:
inf = parse(line)
fun = inf['function']
current = (fun + ';' + current) if current else fun
if inf['frame_num'] == 0:
num_stack_frames += 1
stack_tops[fun] += 1
elif current:
stacks[current] += 1
current = ''
except Exception, ex:
print('ERROR (line %d):' % (idx + 1), ex, file=sys.stderr)
for s, f in sorted(stacks.items(), key=lambda (s, f): s):
print(s, f)
print('Total stack frames:', num_stack_frames, file=sys.stderr)
print('Top consumers (distribution of the stack tops):', file=sys.stderr)
for name,num in sorted(stack_tops.items(), key=lambda (name, num): num, reverse=True)[:300]:
print('% 7.3f%% ' % (100 * num / num_stack_frames), name, file=sys.stderr)
EOF
cat $stacksfile | python /tmp/pmpn-folder.py > $foldfile
echo "Folded stacks saved to $foldfile"
#
# Graphing.
#
cat $foldfile | flamegraph.pl --fontsize=$fgfontsize --width=$fgwidth > $graphfile
echo "FlameGraph saved to $graphfile"
# On KDE, xdg-open prefers Gwenview by default, which doesn't handle interactive SVGs, so we need a browser.
# The current implementation is hackish and stupid. Somebody, please do something about it.
opener=xdg-open
which firefox > /dev/null && opener=firefox
which google-chrome > /dev/null && opener=google-chrome
$opener $graphfile