skia2/tools/skpbench/skpbench.py
Adlai Holler d37a085cd0 Enable DAG reordering on almost all bots
We exclude the NUC5PPYH just to get desktop coverage,
and we exclude the Pixel4, 4XL, 5 because of an apparent driver
bug I'm in the progress of tracking down.

Bug: skia:10877
Change-Id: Ic925cc7434c5228bcc2ee07ae752f89229db55c7
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/399742
Reviewed-by: Robert Phillips <robertphillips@google.com>
Reviewed-by: Ravi Mistry <rmistry@google.com>
Commit-Queue: Adlai Holler <adlai@google.com>
2021-04-22 16:55:39 +00:00

397 lines
14 KiB
Python
Executable File

#!/usr/bin/env python
# Copyright 2016 Google Inc.
#
# Use of this source code is governed by a BSD-style license that can be
# found in the LICENSE file.
from __future__ import print_function
from _adb import Adb
from _benchresult import BenchResult
from _hardware import HardwareException, Hardware
from argparse import ArgumentParser
from multiprocessing import Queue
from threading import Thread, Timer
import collections
import glob
import math
import re
import subprocess
import sys
import time
__argparse = ArgumentParser(description="""
Executes the skpbench binary with various configs and skps.
Also monitors the output in order to filter out and re-run results that have an
unacceptable stddev.
""")
__argparse.add_argument('skpbench',
help="path to the skpbench binary")
__argparse.add_argument('--adb',
action='store_true', help="execute skpbench over adb")
__argparse.add_argument('--adb_binary', default='adb',
help="The name of the adb binary to use.")
__argparse.add_argument('-s', '--device-serial',
help="if using adb, ID of the specific device to target "
"(only required if more than 1 device is attached)")
__argparse.add_argument('-m', '--max-stddev',
type=float, default=4,
help="initial max allowable relative standard deviation")
__argparse.add_argument('-x', '--suffix',
help="suffix to append on config (e.g. '_before', '_after')")
__argparse.add_argument('-w','--write-path',
help="directory to save .png proofs to disk.")
__argparse.add_argument('-v','--verbosity',
type=int, default=1, help="level of verbosity (0=none to 5=debug)")
__argparse.add_argument('-d', '--duration',
type=int, help="number of milliseconds to run each benchmark")
__argparse.add_argument('-l', '--sample-ms',
type=int, help="duration of a sample (minimum)")
__argparse.add_argument('--gpu',
action='store_true',
help="perform timing on the gpu clock instead of cpu (gpu work only)")
__argparse.add_argument('--fps',
action='store_true', help="use fps instead of ms")
__argparse.add_argument('--pr',
help="comma- or space-separated list of GPU path renderers, including: "
"[[~]all [~]default [~]dashline [~]msaa [~]aaconvex "
"[~]aalinearizing [~]small [~]tess]")
__argparse.add_argument('--cc',
action='store_true', help="allow coverage counting shortcuts to render paths")
__argparse.add_argument('--nocache',
action='store_true', help="disable caching of path mask textures")
__argparse.add_argument('--allPathsVolatile',
action='store_true',
help="Causes all GPU paths to be processed as if 'setIsVolatile' had been called.")
__argparse.add_argument('-c', '--config',
default='gl', help="comma- or space-separated list of GPU configs")
__argparse.add_argument('-a', '--resultsfile',
help="optional file to append results into")
__argparse.add_argument('--ddl',
action='store_true', help="record the skp into DDLs before rendering")
__argparse.add_argument('--lock-clocks',
action='store_true', help="Put device in benchmarking mode (locked clocks, no other processes)")
__argparse.add_argument('--clock-speed',
type=float, default=66.0, help="A number between 0 and 100 indicating how fast to lock the CPU and GPU clock."
"Valid speeds are chosen from their respective available frequencies list.")
__argparse.add_argument('--ddlNumRecordingThreads',
type=int, default=0,
help="number of DDL recording threads (0=num_cores)")
__argparse.add_argument('--ddlTilingWidthHeight',
type=int, default=0, help="number of tiles along one edge when in DDL mode")
__argparse.add_argument('--dontReduceOpsTaskSplitting',
action='store_true', help="don't reorder GPU tasks to reduce render target swaps")
__argparse.add_argument('--gpuThreads',
type=int, default=-1,
help="Create this many extra threads to assist with GPU work, including"
" software path rendering. Defaults to two.")
__argparse.add_argument('--internalSamples',
type=int, default=-1,
help="Number of samples for internal draws that use MSAA or mixed samples.")
__argparse.add_argument('srcs',
nargs='+',
help=".skp files or directories to expand for .skp files, and/or .svg files")
__argparse.add_argument('--gpuResourceCacheLimit',
type=int, default=-1,
help="Maximum number of bytes to use for budgeted GPU resources.")
FLAGS = __argparse.parse_args()
if FLAGS.adb:
import _adb_path as _path
_path.init(FLAGS.device_serial, FLAGS.adb_binary)
else:
import _os_path as _path
def dump_commandline_if_verbose(commandline):
if FLAGS.verbosity >= 5:
quoted = ['\'%s\'' % re.sub(r'([\\\'])', r'\\\1', x) for x in commandline]
print(' '.join(quoted), file=sys.stderr)
class StddevException(Exception):
pass
class Message:
READLINE = 0,
POLL_HARDWARE = 1,
EXIT = 2
def __init__(self, message, value=None):
self.message = message
self.value = value
class SubprocessMonitor(Thread):
def __init__(self, queue, proc):
self._queue = queue
self._proc = proc
Thread.__init__(self)
def run(self):
"""Runs on the background thread."""
for line in iter(self._proc.stdout.readline, b''):
self._queue.put(Message(Message.READLINE, line.decode('utf-8').rstrip()))
self._queue.put(Message(Message.EXIT))
class SKPBench:
ARGV = [FLAGS.skpbench, '--verbosity', str(FLAGS.verbosity)]
if FLAGS.duration:
ARGV.extend(['--duration', str(FLAGS.duration)])
if FLAGS.sample_ms:
ARGV.extend(['--sampleMs', str(FLAGS.sample_ms)])
if FLAGS.gpu:
ARGV.extend(['--gpuClock', 'true'])
if FLAGS.fps:
ARGV.extend(['--fps', 'true'])
if FLAGS.pr:
ARGV.extend(['--pr'] + re.split(r'[ ,]', FLAGS.pr))
if FLAGS.cc:
ARGV.extend(['--cc', 'true'])
if FLAGS.nocache:
ARGV.extend(['--cachePathMasks', 'false'])
if FLAGS.allPathsVolatile:
ARGV.extend(['--allPathsVolatile', 'true'])
if FLAGS.gpuThreads != -1:
ARGV.extend(['--gpuThreads', str(FLAGS.gpuThreads)])
if FLAGS.internalSamples != -1:
ARGV.extend(['--internalSamples', str(FLAGS.internalSamples)])
# DDL parameters
if FLAGS.ddl:
ARGV.extend(['--ddl', 'true'])
if FLAGS.ddlNumRecordingThreads:
ARGV.extend(['--ddlNumRecordingThreads',
str(FLAGS.ddlNumRecordingThreads)])
if FLAGS.ddlTilingWidthHeight:
ARGV.extend(['--ddlTilingWidthHeight', str(FLAGS.ddlTilingWidthHeight)])
if FLAGS.dontReduceOpsTaskSplitting:
ARGV.extend(['--dontReduceOpsTaskSplitting'])
if FLAGS.gpuResourceCacheLimit:
ARGV.extend(['--gpuResourceCacheLimit', str(FLAGS.gpuResourceCacheLimit)])
if FLAGS.adb:
if FLAGS.device_serial is None:
ARGV[:0] = [FLAGS.adb_binary, 'shell']
else:
ARGV[:0] = [FLAGS.adb_binary, '-s', FLAGS.device_serial, 'shell']
@classmethod
def get_header(cls, outfile=sys.stdout):
commandline = cls.ARGV + ['--duration', '0']
dump_commandline_if_verbose(commandline)
out = subprocess.check_output(commandline, stderr=subprocess.STDOUT)
return out.rstrip()
@classmethod
def run_warmup(cls, warmup_time, config):
if not warmup_time:
return
print('running %i second warmup...' % warmup_time, file=sys.stderr)
commandline = cls.ARGV + ['--duration', str(warmup_time * 1000),
'--config', config,
'--src', 'warmup']
dump_commandline_if_verbose(commandline)
output = subprocess.check_output(commandline, stderr=subprocess.STDOUT)
# validate the warmup run output.
for line in output.decode('utf-8').split('\n'):
match = BenchResult.match(line.rstrip())
if match and match.bench == 'warmup':
return
raise Exception('Invalid warmup output:\n%s' % output)
def __init__(self, src, config, max_stddev, best_result=None):
self.src = src
self.config = config
self.max_stddev = max_stddev
self.best_result = best_result
self._queue = Queue()
self._proc = None
self._monitor = None
self._hw_poll_timer = None
def __enter__(self):
return self
def __exit__(self, exception_type, exception_value, traceback):
if self._proc:
self.terminate()
if self._hw_poll_timer:
self._hw_poll_timer.cancel()
def execute(self, hardware):
hardware.sanity_check()
self._schedule_hardware_poll()
commandline = self.ARGV + ['--config', self.config,
'--src', self.src,
'--suppressHeader', 'true']
if FLAGS.write_path:
pngfile = _path.join(FLAGS.write_path, self.config,
_path.basename(self.src) + '.png')
commandline.extend(['--png', pngfile])
dump_commandline_if_verbose(commandline)
self._proc = subprocess.Popen(commandline, stdout=subprocess.PIPE,
stderr=subprocess.STDOUT)
self._monitor = SubprocessMonitor(self._queue, self._proc)
self._monitor.start()
while True:
message = self._queue.get()
if message.message == Message.READLINE:
result = BenchResult.match(message.value)
if result:
hardware.sanity_check()
self._process_result(result)
elif hardware.filter_line(message.value):
print(message.value, file=sys.stderr)
continue
if message.message == Message.POLL_HARDWARE:
hardware.sanity_check()
self._schedule_hardware_poll()
continue
if message.message == Message.EXIT:
self._monitor.join()
self._proc.wait()
if self._proc.returncode != 0:
raise Exception("skpbench exited with nonzero exit code %i" %
self._proc.returncode)
self._proc = None
break
def _schedule_hardware_poll(self):
if self._hw_poll_timer:
self._hw_poll_timer.cancel()
self._hw_poll_timer = \
Timer(1, lambda: self._queue.put(Message(Message.POLL_HARDWARE)))
self._hw_poll_timer.start()
def _process_result(self, result):
if not self.best_result or result.stddev <= self.best_result.stddev:
self.best_result = result
elif FLAGS.verbosity >= 2:
print("reusing previous result for %s/%s with lower stddev "
"(%s%% instead of %s%%)." %
(result.config, result.bench, self.best_result.stddev,
result.stddev), file=sys.stderr)
if self.max_stddev and self.best_result.stddev > self.max_stddev:
raise StddevException()
def terminate(self):
if self._proc:
self._proc.terminate()
self._monitor.join()
self._proc.wait()
self._proc = None
def emit_result(line, resultsfile=None):
print(line)
sys.stdout.flush()
if resultsfile:
print(line, file=resultsfile)
resultsfile.flush()
def run_benchmarks(configs, srcs, hardware, resultsfile=None):
hasheader = False
benches = collections.deque([(src, config, FLAGS.max_stddev)
for src in srcs
for config in configs])
while benches:
try:
with hardware:
SKPBench.run_warmup(hardware.warmup_time, configs[0])
if not hasheader:
emit_result(SKPBench.get_header(), resultsfile)
hasheader = True
while benches:
benchargs = benches.popleft()
with SKPBench(*benchargs) as skpbench:
try:
skpbench.execute(hardware)
if skpbench.best_result:
emit_result(skpbench.best_result.format(FLAGS.suffix),
resultsfile)
else:
print("WARNING: no result for %s with config %s" %
(skpbench.src, skpbench.config), file=sys.stderr)
except StddevException:
retry_max_stddev = skpbench.max_stddev * math.sqrt(2)
if FLAGS.verbosity >= 1:
print("stddev is too high for %s/%s (%s%%, max=%.2f%%), "
"re-queuing with max=%.2f%%." %
(skpbench.best_result.config, skpbench.best_result.bench,
skpbench.best_result.stddev, skpbench.max_stddev,
retry_max_stddev),
file=sys.stderr)
benches.append((skpbench.src, skpbench.config, retry_max_stddev,
skpbench.best_result))
except HardwareException as exception:
skpbench.terminate()
if FLAGS.verbosity >= 4:
hardware.print_debug_diagnostics()
if FLAGS.verbosity >= 1:
print("%s; rebooting and taking a %i second nap..." %
(exception.message, exception.sleeptime), file=sys.stderr)
benches.appendleft(benchargs) # retry the same bench next time.
raise # wake hw up from benchmarking mode before the nap.
except HardwareException as exception:
time.sleep(exception.sleeptime)
def main():
# Delimiter is ',' or ' ', skip if nested inside parens (e.g. gpu(a=b,c=d)).
DELIMITER = r'[, ](?!(?:[^(]*\([^)]*\))*[^()]*\))'
configs = re.split(DELIMITER, FLAGS.config)
srcs = _path.find_skps(FLAGS.srcs)
assert srcs
if FLAGS.adb:
adb = Adb(FLAGS.device_serial, FLAGS.adb_binary,
echo=(FLAGS.verbosity >= 5))
from _hardware_android import HardwareAndroid
model = adb.check('getprop ro.product.model').strip()
if model == 'Pixel C':
from _hardware_pixel_c import HardwarePixelC
hardware = HardwarePixelC(adb)
elif model == 'Pixel' or model == "Pixel XL":
from _hardware_pixel import HardwarePixel
hardware = HardwarePixel(adb)
elif model == 'Pixel 2':
from _hardware_pixel2 import HardwarePixel2
hardware = HardwarePixel2(adb)
elif model == 'Nexus 6P':
from _hardware_nexus_6p import HardwareNexus6P
hardware = HardwareNexus6P(adb)
else:
print("WARNING: %s: don't know how to monitor this hardware; results "
"may be unreliable." % model, file=sys.stderr)
hardware = HardwareAndroid(adb)
if FLAGS.lock_clocks:
hardware.__enter__()
print("Entered benchmarking mode, not running benchmarks. Reboot to restore.");
return;
if FLAGS.clock_speed:
hardware.setDesiredClock(FLAGS.clock_speed)
else:
hardware = Hardware()
if FLAGS.resultsfile:
with open(FLAGS.resultsfile, mode='a+') as resultsfile:
run_benchmarks(configs, srcs, hardware, resultsfile=resultsfile)
else:
run_benchmarks(configs, srcs, hardware)
if __name__ == '__main__':
main()