AuroraMiddleware/skia2
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity
5ebaf19444
skia2/infra/bots/recipes/perf.expected/cpu_scale_failed.json

887 lines
42 KiB
JSON
Raw Normal View History

Make CPU scaling code cleaner This also makes *sure* the CPU frequency we set the device to actually "takes". Along the way, I learned if scaling_max_freq is < the frequency we set, the scaling_max_freq will be used instead, which was happening to the PixelCs and AndroidOnes. As a result, this may make those two Test- configs faster. Bug: skia: Change-Id: I10c98d37e296a19e1cf67bfe7269bb59cdd912d5 Reviewed-on: https://skia-review.googlesource.com/74360 Commit-Queue: Kevin Lubick <kjlubick@google.com> Reviewed-by: Ben Wagner <benjaminwagner@google.com> Reviewed-by: Eric Boren <borenet@google.com>
2017-11-21 21:47:16 +00:00
[
{
"cmd": [
Use full paths for adb Add logic to run on Nexus 5xs in Golo. Bug:788839 Change-Id: I12290d11a0b1b0f012ada216da3e5b2599979c5e Reviewed-on: https://skia-review.googlesource.com/81861 Commit-Queue: Kevin Lubick <kjlubick@google.com> Reviewed-by: Eric Boren <borenet@google.com>
2017-12-07 16:19:31 +00:00
"/usr/bin/adb.1.0.35",
Make CPU scaling code cleaner This also makes *sure* the CPU frequency we set the device to actually "takes". Along the way, I learned if scaling_max_freq is < the frequency we set, the scaling_max_freq will be used instead, which was happening to the PixelCs and AndroidOnes. As a result, this may make those two Test- configs faster. Bug: skia: Change-Id: I10c98d37e296a19e1cf67bfe7269bb59cdd912d5 Reviewed-on: https://skia-review.googlesource.com/74360 Commit-Queue: Kevin Lubick <kjlubick@google.com> Reviewed-by: Ben Wagner <benjaminwagner@google.com> Reviewed-by: Eric Boren <borenet@google.com>
2017-11-21 21:47:16 +00:00
"shell",
"mkdir",
"-p",
"/sdcard/revenge_of_the_skiabot/resources"
],
"cwd": "[START_DIR]/skia",
"env": {
"BUILDTYPE": "Debug",
"CHROME_HEADLESS": "1",
"PATH": "<PATH>:RECIPE_PACKAGE_REPO[depot_tools]",
"SKIA_OUT": "[START_DIR]/out"
},
"infra_step": true,
"name": "mkdir /sdcard/revenge_of_the_skiabot/resources"
},
{
"cmd": [
"python",
"-u",
Use full paths for adb Add logic to run on Nexus 5xs in Golo. Bug:788839 Change-Id: I12290d11a0b1b0f012ada216da3e5b2599979c5e Reviewed-on: https://skia-review.googlesource.com/81861 Commit-Queue: Kevin Lubick <kjlubick@google.com> Reviewed-by: Eric Boren <borenet@google.com>
2017-12-07 16:19:31 +00:00
"\nimport os\nimport subprocess\nimport sys\nhost = sys.argv[1]\ndevice = sys.argv[2]\nfor d, _, fs in os.walk(host):\n p = os.path.relpath(d, host)\n if p != '.' and p.startswith('.'):\n continue\n for f in fs:\n print os.path.join(p,f)\n subprocess.check_call(['/usr/bin/adb.1.0.35', 'push',\n os.path.realpath(os.path.join(host, p, f)),\n os.path.join(device, p, f)])\n",
Make CPU scaling code cleaner This also makes *sure* the CPU frequency we set the device to actually "takes". Along the way, I learned if scaling_max_freq is < the frequency we set, the scaling_max_freq will be used instead, which was happening to the PixelCs and AndroidOnes. As a result, this may make those two Test- configs faster. Bug: skia: Change-Id: I10c98d37e296a19e1cf67bfe7269bb59cdd912d5 Reviewed-on: https://skia-review.googlesource.com/74360 Commit-Queue: Kevin Lubick <kjlubick@google.com> Reviewed-by: Ben Wagner <benjaminwagner@google.com> Reviewed-by: Eric Boren <borenet@google.com>
2017-11-21 21:47:16 +00:00
"[START_DIR]/skia/resources",
"/sdcard/revenge_of_the_skiabot/resources"
],
"env": {
"BUILDTYPE": "Debug",
"CHROME_HEADLESS": "1",
"PATH": "<PATH>:RECIPE_PACKAGE_REPO[depot_tools]",
"SKIA_OUT": "[START_DIR]/out"
},
"infra_step": true,
"name": "push [START_DIR]/skia/resources/* /sdcard/revenge_of_the_skiabot/resources",
"~followup_annotations": [
"@@@STEP_LOG_LINE@python.inline@@@@",
"@@@STEP_LOG_LINE@python.inline@import os@@@",
"@@@STEP_LOG_LINE@python.inline@import subprocess@@@",
"@@@STEP_LOG_LINE@python.inline@import sys@@@",
"@@@STEP_LOG_LINE@python.inline@host = sys.argv[1]@@@",
"@@@STEP_LOG_LINE@python.inline@device = sys.argv[2]@@@",
"@@@STEP_LOG_LINE@python.inline@for d, _, fs in os.walk(host):@@@",
"@@@STEP_LOG_LINE@python.inline@ p = os.path.relpath(d, host)@@@",
"@@@STEP_LOG_LINE@python.inline@ if p != '.' and p.startswith('.'):@@@",
"@@@STEP_LOG_LINE@python.inline@ continue@@@",
"@@@STEP_LOG_LINE@python.inline@ for f in fs:@@@",
"@@@STEP_LOG_LINE@python.inline@ print os.path.join(p,f)@@@",
Use full paths for adb Add logic to run on Nexus 5xs in Golo. Bug:788839 Change-Id: I12290d11a0b1b0f012ada216da3e5b2599979c5e Reviewed-on: https://skia-review.googlesource.com/81861 Commit-Queue: Kevin Lubick <kjlubick@google.com> Reviewed-by: Eric Boren <borenet@google.com>
2017-12-07 16:19:31 +00:00
"@@@STEP_LOG_LINE@python.inline@ subprocess.check_call(['/usr/bin/adb.1.0.35', 'push',@@@",
Make CPU scaling code cleaner This also makes *sure* the CPU frequency we set the device to actually "takes". Along the way, I learned if scaling_max_freq is < the frequency we set, the scaling_max_freq will be used instead, which was happening to the PixelCs and AndroidOnes. As a result, this may make those two Test- configs faster. Bug: skia: Change-Id: I10c98d37e296a19e1cf67bfe7269bb59cdd912d5 Reviewed-on: https://skia-review.googlesource.com/74360 Commit-Queue: Kevin Lubick <kjlubick@google.com> Reviewed-by: Ben Wagner <benjaminwagner@google.com> Reviewed-by: Eric Boren <borenet@google.com>
2017-11-21 21:47:16 +00:00
"@@@STEP_LOG_LINE@python.inline@ os.path.realpath(os.path.join(host, p, f)),@@@",
"@@@STEP_LOG_LINE@python.inline@ os.path.join(device, p, f)])@@@",
"@@@STEP_LOG_END@python.inline@@@"
]
},
{
"cmd": [
"python",
"-u",
"RECIPE_MODULE[recipe_engine::file]/resources/fileutil.py",
"--json-output",
"/path/to/tmp/json",
"copy",
"[START_DIR]/skia/infra/bots/assets/skp/VERSION",
"/path/to/tmp/"
],
"infra_step": true,
"name": "Get downloaded SKP VERSION"
},
{
"cmd": [
"python",
"-u",
"RECIPE_MODULE[recipe_engine::file]/resources/fileutil.py",
"--json-output",
"/path/to/tmp/json",
"copy",
"42",
"[START_DIR]/tmp/SKP_VERSION"
],
"infra_step": true,
"name": "write SKP_VERSION"
},
{
"cmd": [
Use full paths for adb Add logic to run on Nexus 5xs in Golo. Bug:788839 Change-Id: I12290d11a0b1b0f012ada216da3e5b2599979c5e Reviewed-on: https://skia-review.googlesource.com/81861 Commit-Queue: Kevin Lubick <kjlubick@google.com> Reviewed-by: Eric Boren <borenet@google.com>
2017-12-07 16:19:31 +00:00
"/usr/bin/adb.1.0.35",
Make CPU scaling code cleaner This also makes *sure* the CPU frequency we set the device to actually "takes". Along the way, I learned if scaling_max_freq is < the frequency we set, the scaling_max_freq will be used instead, which was happening to the PixelCs and AndroidOnes. As a result, this may make those two Test- configs faster. Bug: skia: Change-Id: I10c98d37e296a19e1cf67bfe7269bb59cdd912d5 Reviewed-on: https://skia-review.googlesource.com/74360 Commit-Queue: Kevin Lubick <kjlubick@google.com> Reviewed-by: Ben Wagner <benjaminwagner@google.com> Reviewed-by: Eric Boren <borenet@google.com>
2017-11-21 21:47:16 +00:00
"shell",
"cat",
"/sdcard/revenge_of_the_skiabot/SKP_VERSION"
],
"cwd": "[START_DIR]/skia",
"env": {
"BUILDTYPE": "Debug",
"CHROME_HEADLESS": "1",
"PATH": "<PATH>:RECIPE_PACKAGE_REPO[depot_tools]",
"SKIA_OUT": "[START_DIR]/out"
},
"infra_step": true,
"name": "read /sdcard/revenge_of_the_skiabot/SKP_VERSION",
"stdout": "/path/to/tmp/"
},
{
"cmd": [
Use full paths for adb Add logic to run on Nexus 5xs in Golo. Bug:788839 Change-Id: I12290d11a0b1b0f012ada216da3e5b2599979c5e Reviewed-on: https://skia-review.googlesource.com/81861 Commit-Queue: Kevin Lubick <kjlubick@google.com> Reviewed-by: Eric Boren <borenet@google.com>
2017-12-07 16:19:31 +00:00
"/usr/bin/adb.1.0.35",
Make CPU scaling code cleaner This also makes *sure* the CPU frequency we set the device to actually "takes". Along the way, I learned if scaling_max_freq is < the frequency we set, the scaling_max_freq will be used instead, which was happening to the PixelCs and AndroidOnes. As a result, this may make those two Test- configs faster. Bug: skia: Change-Id: I10c98d37e296a19e1cf67bfe7269bb59cdd912d5 Reviewed-on: https://skia-review.googlesource.com/74360 Commit-Queue: Kevin Lubick <kjlubick@google.com> Reviewed-by: Ben Wagner <benjaminwagner@google.com> Reviewed-by: Eric Boren <borenet@google.com>
2017-11-21 21:47:16 +00:00
"shell",
"rm",
"-f",
"/sdcard/revenge_of_the_skiabot/SKP_VERSION"
],
"cwd": "[START_DIR]/skia",
"env": {
"BUILDTYPE": "Debug",
"CHROME_HEADLESS": "1",
"PATH": "<PATH>:RECIPE_PACKAGE_REPO[depot_tools]",
"SKIA_OUT": "[START_DIR]/out"
},
"infra_step": true,
"name": "rm /sdcard/revenge_of_the_skiabot/SKP_VERSION"
},
{
"cmd": [
Use full paths for adb Add logic to run on Nexus 5xs in Golo. Bug:788839 Change-Id: I12290d11a0b1b0f012ada216da3e5b2599979c5e Reviewed-on: https://skia-review.googlesource.com/81861 Commit-Queue: Kevin Lubick <kjlubick@google.com> Reviewed-by: Eric Boren <borenet@google.com>
2017-12-07 16:19:31 +00:00
"/usr/bin/adb.1.0.35",
Make CPU scaling code cleaner This also makes *sure* the CPU frequency we set the device to actually "takes". Along the way, I learned if scaling_max_freq is < the frequency we set, the scaling_max_freq will be used instead, which was happening to the PixelCs and AndroidOnes. As a result, this may make those two Test- configs faster. Bug: skia: Change-Id: I10c98d37e296a19e1cf67bfe7269bb59cdd912d5 Reviewed-on: https://skia-review.googlesource.com/74360 Commit-Queue: Kevin Lubick <kjlubick@google.com> Reviewed-by: Ben Wagner <benjaminwagner@google.com> Reviewed-by: Eric Boren <borenet@google.com>
2017-11-21 21:47:16 +00:00
"shell",
"rm",
"-rf",
"/sdcard/revenge_of_the_skiabot/skps"
],
"cwd": "[START_DIR]/skia",
"env": {
"BUILDTYPE": "Debug",
"CHROME_HEADLESS": "1",
"PATH": "<PATH>:RECIPE_PACKAGE_REPO[depot_tools]",
"SKIA_OUT": "[START_DIR]/out"
},
"infra_step": true,
"name": "rm /sdcard/revenge_of_the_skiabot/skps"
},
{
"cmd": [
Use full paths for adb Add logic to run on Nexus 5xs in Golo. Bug:788839 Change-Id: I12290d11a0b1b0f012ada216da3e5b2599979c5e Reviewed-on: https://skia-review.googlesource.com/81861 Commit-Queue: Kevin Lubick <kjlubick@google.com> Reviewed-by: Eric Boren <borenet@google.com>
2017-12-07 16:19:31 +00:00
"/usr/bin/adb.1.0.35",
Make CPU scaling code cleaner This also makes *sure* the CPU frequency we set the device to actually "takes". Along the way, I learned if scaling_max_freq is < the frequency we set, the scaling_max_freq will be used instead, which was happening to the PixelCs and AndroidOnes. As a result, this may make those two Test- configs faster. Bug: skia: Change-Id: I10c98d37e296a19e1cf67bfe7269bb59cdd912d5 Reviewed-on: https://skia-review.googlesource.com/74360 Commit-Queue: Kevin Lubick <kjlubick@google.com> Reviewed-by: Ben Wagner <benjaminwagner@google.com> Reviewed-by: Eric Boren <borenet@google.com>
2017-11-21 21:47:16 +00:00
"shell",
"mkdir",
"-p",
"/sdcard/revenge_of_the_skiabot/skps"
],
"cwd": "[START_DIR]/skia",
"env": {
"BUILDTYPE": "Debug",
"CHROME_HEADLESS": "1",
"PATH": "<PATH>:RECIPE_PACKAGE_REPO[depot_tools]",
"SKIA_OUT": "[START_DIR]/out"
},
"infra_step": true,
"name": "mkdir /sdcard/revenge_of_the_skiabot/skps"
},
{
"cmd": [
"python",
"-u",
Use full paths for adb Add logic to run on Nexus 5xs in Golo. Bug:788839 Change-Id: I12290d11a0b1b0f012ada216da3e5b2599979c5e Reviewed-on: https://skia-review.googlesource.com/81861 Commit-Queue: Kevin Lubick <kjlubick@google.com> Reviewed-by: Eric Boren <borenet@google.com>
2017-12-07 16:19:31 +00:00
"\nimport os\nimport subprocess\nimport sys\nhost = sys.argv[1]\ndevice = sys.argv[2]\nfor d, _, fs in os.walk(host):\n p = os.path.relpath(d, host)\n if p != '.' and p.startswith('.'):\n continue\n for f in fs:\n print os.path.join(p,f)\n subprocess.check_call(['/usr/bin/adb.1.0.35', 'push',\n os.path.realpath(os.path.join(host, p, f)),\n os.path.join(device, p, f)])\n",
Make CPU scaling code cleaner This also makes *sure* the CPU frequency we set the device to actually "takes". Along the way, I learned if scaling_max_freq is < the frequency we set, the scaling_max_freq will be used instead, which was happening to the PixelCs and AndroidOnes. As a result, this may make those two Test- configs faster. Bug: skia: Change-Id: I10c98d37e296a19e1cf67bfe7269bb59cdd912d5 Reviewed-on: https://skia-review.googlesource.com/74360 Commit-Queue: Kevin Lubick <kjlubick@google.com> Reviewed-by: Ben Wagner <benjaminwagner@google.com> Reviewed-by: Eric Boren <borenet@google.com>
2017-11-21 21:47:16 +00:00
"[START_DIR]/skp",
"/sdcard/revenge_of_the_skiabot/skps"
],
"env": {
"BUILDTYPE": "Debug",
"CHROME_HEADLESS": "1",
"PATH": "<PATH>:RECIPE_PACKAGE_REPO[depot_tools]",
"SKIA_OUT": "[START_DIR]/out"
},
"infra_step": true,
"name": "push [START_DIR]/skp/* /sdcard/revenge_of_the_skiabot/skps",
"~followup_annotations": [
"@@@STEP_LOG_LINE@python.inline@@@@",
"@@@STEP_LOG_LINE@python.inline@import os@@@",
"@@@STEP_LOG_LINE@python.inline@import subprocess@@@",
"@@@STEP_LOG_LINE@python.inline@import sys@@@",
"@@@STEP_LOG_LINE@python.inline@host = sys.argv[1]@@@",
"@@@STEP_LOG_LINE@python.inline@device = sys.argv[2]@@@",
"@@@STEP_LOG_LINE@python.inline@for d, _, fs in os.walk(host):@@@",
"@@@STEP_LOG_LINE@python.inline@ p = os.path.relpath(d, host)@@@",
"@@@STEP_LOG_LINE@python.inline@ if p != '.' and p.startswith('.'):@@@",
"@@@STEP_LOG_LINE@python.inline@ continue@@@",
"@@@STEP_LOG_LINE@python.inline@ for f in fs:@@@",
"@@@STEP_LOG_LINE@python.inline@ print os.path.join(p,f)@@@",
Use full paths for adb Add logic to run on Nexus 5xs in Golo. Bug:788839 Change-Id: I12290d11a0b1b0f012ada216da3e5b2599979c5e Reviewed-on: https://skia-review.googlesource.com/81861 Commit-Queue: Kevin Lubick <kjlubick@google.com> Reviewed-by: Eric Boren <borenet@google.com>
2017-12-07 16:19:31 +00:00
"@@@STEP_LOG_LINE@python.inline@ subprocess.check_call(['/usr/bin/adb.1.0.35', 'push',@@@",
Make CPU scaling code cleaner This also makes *sure* the CPU frequency we set the device to actually "takes". Along the way, I learned if scaling_max_freq is < the frequency we set, the scaling_max_freq will be used instead, which was happening to the PixelCs and AndroidOnes. As a result, this may make those two Test- configs faster. Bug: skia: Change-Id: I10c98d37e296a19e1cf67bfe7269bb59cdd912d5 Reviewed-on: https://skia-review.googlesource.com/74360 Commit-Queue: Kevin Lubick <kjlubick@google.com> Reviewed-by: Ben Wagner <benjaminwagner@google.com> Reviewed-by: Eric Boren <borenet@google.com>
2017-11-21 21:47:16 +00:00
"@@@STEP_LOG_LINE@python.inline@ os.path.realpath(os.path.join(host, p, f)),@@@",
"@@@STEP_LOG_LINE@python.inline@ os.path.join(device, p, f)])@@@",
"@@@STEP_LOG_END@python.inline@@@"
]
},
{
"cmd": [
Use full paths for adb Add logic to run on Nexus 5xs in Golo. Bug:788839 Change-Id: I12290d11a0b1b0f012ada216da3e5b2599979c5e Reviewed-on: https://skia-review.googlesource.com/81861 Commit-Queue: Kevin Lubick <kjlubick@google.com> Reviewed-by: Eric Boren <borenet@google.com>
2017-12-07 16:19:31 +00:00
"/usr/bin/adb.1.0.35",
Make CPU scaling code cleaner This also makes *sure* the CPU frequency we set the device to actually "takes". Along the way, I learned if scaling_max_freq is < the frequency we set, the scaling_max_freq will be used instead, which was happening to the PixelCs and AndroidOnes. As a result, this may make those two Test- configs faster. Bug: skia: Change-Id: I10c98d37e296a19e1cf67bfe7269bb59cdd912d5 Reviewed-on: https://skia-review.googlesource.com/74360 Commit-Queue: Kevin Lubick <kjlubick@google.com> Reviewed-by: Ben Wagner <benjaminwagner@google.com> Reviewed-by: Eric Boren <borenet@google.com>
2017-11-21 21:47:16 +00:00
"push",
"[START_DIR]/tmp/SKP_VERSION",
"/sdcard/revenge_of_the_skiabot/SKP_VERSION"
],
"cwd": "[START_DIR]/skia",
"env": {
"BUILDTYPE": "Debug",
"CHROME_HEADLESS": "1",
"PATH": "<PATH>:RECIPE_PACKAGE_REPO[depot_tools]",
"SKIA_OUT": "[START_DIR]/out"
},
"infra_step": true,
"name": "push [START_DIR]/tmp/SKP_VERSION /sdcard/revenge_of_the_skiabot/SKP_VERSION"
},
{
"cmd": [
"python",
"-u",
"RECIPE_MODULE[recipe_engine::file]/resources/fileutil.py",
"--json-output",
"/path/to/tmp/json",
"copy",
"[START_DIR]/skia/infra/bots/assets/skimage/VERSION",
"/path/to/tmp/"
],
"infra_step": true,
"name": "Get downloaded skimage VERSION"
},
{
"cmd": [
"python",
"-u",
"RECIPE_MODULE[recipe_engine::file]/resources/fileutil.py",
"--json-output",
"/path/to/tmp/json",
"copy",
"42",
"[START_DIR]/tmp/SK_IMAGE_VERSION"
],
"infra_step": true,
"name": "write SK_IMAGE_VERSION"
},
{
"cmd": [
Use full paths for adb Add logic to run on Nexus 5xs in Golo. Bug:788839 Change-Id: I12290d11a0b1b0f012ada216da3e5b2599979c5e Reviewed-on: https://skia-review.googlesource.com/81861 Commit-Queue: Kevin Lubick <kjlubick@google.com> Reviewed-by: Eric Boren <borenet@google.com>
2017-12-07 16:19:31 +00:00
"/usr/bin/adb.1.0.35",
Make CPU scaling code cleaner This also makes *sure* the CPU frequency we set the device to actually "takes". Along the way, I learned if scaling_max_freq is < the frequency we set, the scaling_max_freq will be used instead, which was happening to the PixelCs and AndroidOnes. As a result, this may make those two Test- configs faster. Bug: skia: Change-Id: I10c98d37e296a19e1cf67bfe7269bb59cdd912d5 Reviewed-on: https://skia-review.googlesource.com/74360 Commit-Queue: Kevin Lubick <kjlubick@google.com> Reviewed-by: Ben Wagner <benjaminwagner@google.com> Reviewed-by: Eric Boren <borenet@google.com>
2017-11-21 21:47:16 +00:00
"shell",
"cat",
"/sdcard/revenge_of_the_skiabot/SK_IMAGE_VERSION"
],
"cwd": "[START_DIR]/skia",
"env": {
"BUILDTYPE": "Debug",
"CHROME_HEADLESS": "1",
"PATH": "<PATH>:RECIPE_PACKAGE_REPO[depot_tools]",
"SKIA_OUT": "[START_DIR]/out"
},
"infra_step": true,
"name": "read /sdcard/revenge_of_the_skiabot/SK_IMAGE_VERSION",
"stdout": "/path/to/tmp/"
},
{
"cmd": [
Use full paths for adb Add logic to run on Nexus 5xs in Golo. Bug:788839 Change-Id: I12290d11a0b1b0f012ada216da3e5b2599979c5e Reviewed-on: https://skia-review.googlesource.com/81861 Commit-Queue: Kevin Lubick <kjlubick@google.com> Reviewed-by: Eric Boren <borenet@google.com>
2017-12-07 16:19:31 +00:00
"/usr/bin/adb.1.0.35",
Make CPU scaling code cleaner This also makes *sure* the CPU frequency we set the device to actually "takes". Along the way, I learned if scaling_max_freq is < the frequency we set, the scaling_max_freq will be used instead, which was happening to the PixelCs and AndroidOnes. As a result, this may make those two Test- configs faster. Bug: skia: Change-Id: I10c98d37e296a19e1cf67bfe7269bb59cdd912d5 Reviewed-on: https://skia-review.googlesource.com/74360 Commit-Queue: Kevin Lubick <kjlubick@google.com> Reviewed-by: Ben Wagner <benjaminwagner@google.com> Reviewed-by: Eric Boren <borenet@google.com>
2017-11-21 21:47:16 +00:00
"shell",
"rm",
"-f",
"/sdcard/revenge_of_the_skiabot/SK_IMAGE_VERSION"
],
"cwd": "[START_DIR]/skia",
"env": {
"BUILDTYPE": "Debug",
"CHROME_HEADLESS": "1",
"PATH": "<PATH>:RECIPE_PACKAGE_REPO[depot_tools]",
"SKIA_OUT": "[START_DIR]/out"
},
"infra_step": true,
"name": "rm /sdcard/revenge_of_the_skiabot/SK_IMAGE_VERSION"
},
{
"cmd": [
Use full paths for adb Add logic to run on Nexus 5xs in Golo. Bug:788839 Change-Id: I12290d11a0b1b0f012ada216da3e5b2599979c5e Reviewed-on: https://skia-review.googlesource.com/81861 Commit-Queue: Kevin Lubick <kjlubick@google.com> Reviewed-by: Eric Boren <borenet@google.com>
2017-12-07 16:19:31 +00:00
"/usr/bin/adb.1.0.35",
Make CPU scaling code cleaner This also makes *sure* the CPU frequency we set the device to actually "takes". Along the way, I learned if scaling_max_freq is < the frequency we set, the scaling_max_freq will be used instead, which was happening to the PixelCs and AndroidOnes. As a result, this may make those two Test- configs faster. Bug: skia: Change-Id: I10c98d37e296a19e1cf67bfe7269bb59cdd912d5 Reviewed-on: https://skia-review.googlesource.com/74360 Commit-Queue: Kevin Lubick <kjlubick@google.com> Reviewed-by: Ben Wagner <benjaminwagner@google.com> Reviewed-by: Eric Boren <borenet@google.com>
2017-11-21 21:47:16 +00:00
"shell",
"rm",
"-rf",
"/sdcard/revenge_of_the_skiabot/images"
],
"cwd": "[START_DIR]/skia",
"env": {
"BUILDTYPE": "Debug",
"CHROME_HEADLESS": "1",
"PATH": "<PATH>:RECIPE_PACKAGE_REPO[depot_tools]",
"SKIA_OUT": "[START_DIR]/out"
},
"infra_step": true,
"name": "rm /sdcard/revenge_of_the_skiabot/images"
},
{
"cmd": [
Use full paths for adb Add logic to run on Nexus 5xs in Golo. Bug:788839 Change-Id: I12290d11a0b1b0f012ada216da3e5b2599979c5e Reviewed-on: https://skia-review.googlesource.com/81861 Commit-Queue: Kevin Lubick <kjlubick@google.com> Reviewed-by: Eric Boren <borenet@google.com>
2017-12-07 16:19:31 +00:00
"/usr/bin/adb.1.0.35",
Make CPU scaling code cleaner This also makes *sure* the CPU frequency we set the device to actually "takes". Along the way, I learned if scaling_max_freq is < the frequency we set, the scaling_max_freq will be used instead, which was happening to the PixelCs and AndroidOnes. As a result, this may make those two Test- configs faster. Bug: skia: Change-Id: I10c98d37e296a19e1cf67bfe7269bb59cdd912d5 Reviewed-on: https://skia-review.googlesource.com/74360 Commit-Queue: Kevin Lubick <kjlubick@google.com> Reviewed-by: Ben Wagner <benjaminwagner@google.com> Reviewed-by: Eric Boren <borenet@google.com>
2017-11-21 21:47:16 +00:00
"shell",
"mkdir",
"-p",
"/sdcard/revenge_of_the_skiabot/images"
],
"cwd": "[START_DIR]/skia",
"env": {
"BUILDTYPE": "Debug",
"CHROME_HEADLESS": "1",
"PATH": "<PATH>:RECIPE_PACKAGE_REPO[depot_tools]",
"SKIA_OUT": "[START_DIR]/out"
},
"infra_step": true,
"name": "mkdir /sdcard/revenge_of_the_skiabot/images"
},
{
"cmd": [
"python",
"-u",
Use full paths for adb Add logic to run on Nexus 5xs in Golo. Bug:788839 Change-Id: I12290d11a0b1b0f012ada216da3e5b2599979c5e Reviewed-on: https://skia-review.googlesource.com/81861 Commit-Queue: Kevin Lubick <kjlubick@google.com> Reviewed-by: Eric Boren <borenet@google.com>
2017-12-07 16:19:31 +00:00
"\nimport os\nimport subprocess\nimport sys\nhost = sys.argv[1]\ndevice = sys.argv[2]\nfor d, _, fs in os.walk(host):\n p = os.path.relpath(d, host)\n if p != '.' and p.startswith('.'):\n continue\n for f in fs:\n print os.path.join(p,f)\n subprocess.check_call(['/usr/bin/adb.1.0.35', 'push',\n os.path.realpath(os.path.join(host, p, f)),\n os.path.join(device, p, f)])\n",
Make CPU scaling code cleaner This also makes *sure* the CPU frequency we set the device to actually "takes". Along the way, I learned if scaling_max_freq is < the frequency we set, the scaling_max_freq will be used instead, which was happening to the PixelCs and AndroidOnes. As a result, this may make those two Test- configs faster. Bug: skia: Change-Id: I10c98d37e296a19e1cf67bfe7269bb59cdd912d5 Reviewed-on: https://skia-review.googlesource.com/74360 Commit-Queue: Kevin Lubick <kjlubick@google.com> Reviewed-by: Ben Wagner <benjaminwagner@google.com> Reviewed-by: Eric Boren <borenet@google.com>
2017-11-21 21:47:16 +00:00
"[START_DIR]/skimage",
"/sdcard/revenge_of_the_skiabot/images"
],
"env": {
"BUILDTYPE": "Debug",
"CHROME_HEADLESS": "1",
"PATH": "<PATH>:RECIPE_PACKAGE_REPO[depot_tools]",
"SKIA_OUT": "[START_DIR]/out"
},
"infra_step": true,
"name": "push [START_DIR]/skimage/* /sdcard/revenge_of_the_skiabot/images",
"~followup_annotations": [
"@@@STEP_LOG_LINE@python.inline@@@@",
"@@@STEP_LOG_LINE@python.inline@import os@@@",
"@@@STEP_LOG_LINE@python.inline@import subprocess@@@",
"@@@STEP_LOG_LINE@python.inline@import sys@@@",
"@@@STEP_LOG_LINE@python.inline@host = sys.argv[1]@@@",
"@@@STEP_LOG_LINE@python.inline@device = sys.argv[2]@@@",
"@@@STEP_LOG_LINE@python.inline@for d, _, fs in os.walk(host):@@@",
"@@@STEP_LOG_LINE@python.inline@ p = os.path.relpath(d, host)@@@",
"@@@STEP_LOG_LINE@python.inline@ if p != '.' and p.startswith('.'):@@@",
"@@@STEP_LOG_LINE@python.inline@ continue@@@",
"@@@STEP_LOG_LINE@python.inline@ for f in fs:@@@",
"@@@STEP_LOG_LINE@python.inline@ print os.path.join(p,f)@@@",
Use full paths for adb Add logic to run on Nexus 5xs in Golo. Bug:788839 Change-Id: I12290d11a0b1b0f012ada216da3e5b2599979c5e Reviewed-on: https://skia-review.googlesource.com/81861 Commit-Queue: Kevin Lubick <kjlubick@google.com> Reviewed-by: Eric Boren <borenet@google.com>
2017-12-07 16:19:31 +00:00
"@@@STEP_LOG_LINE@python.inline@ subprocess.check_call(['/usr/bin/adb.1.0.35', 'push',@@@",
Make CPU scaling code cleaner This also makes *sure* the CPU frequency we set the device to actually "takes". Along the way, I learned if scaling_max_freq is < the frequency we set, the scaling_max_freq will be used instead, which was happening to the PixelCs and AndroidOnes. As a result, this may make those two Test- configs faster. Bug: skia: Change-Id: I10c98d37e296a19e1cf67bfe7269bb59cdd912d5 Reviewed-on: https://skia-review.googlesource.com/74360 Commit-Queue: Kevin Lubick <kjlubick@google.com> Reviewed-by: Ben Wagner <benjaminwagner@google.com> Reviewed-by: Eric Boren <borenet@google.com>
2017-11-21 21:47:16 +00:00
"@@@STEP_LOG_LINE@python.inline@ os.path.realpath(os.path.join(host, p, f)),@@@",
"@@@STEP_LOG_LINE@python.inline@ os.path.join(device, p, f)])@@@",
"@@@STEP_LOG_END@python.inline@@@"
]
},
{
"cmd": [
Use full paths for adb Add logic to run on Nexus 5xs in Golo. Bug:788839 Change-Id: I12290d11a0b1b0f012ada216da3e5b2599979c5e Reviewed-on: https://skia-review.googlesource.com/81861 Commit-Queue: Kevin Lubick <kjlubick@google.com> Reviewed-by: Eric Boren <borenet@google.com>
2017-12-07 16:19:31 +00:00
"/usr/bin/adb.1.0.35",
Make CPU scaling code cleaner This also makes *sure* the CPU frequency we set the device to actually "takes". Along the way, I learned if scaling_max_freq is < the frequency we set, the scaling_max_freq will be used instead, which was happening to the PixelCs and AndroidOnes. As a result, this may make those two Test- configs faster. Bug: skia: Change-Id: I10c98d37e296a19e1cf67bfe7269bb59cdd912d5 Reviewed-on: https://skia-review.googlesource.com/74360 Commit-Queue: Kevin Lubick <kjlubick@google.com> Reviewed-by: Ben Wagner <benjaminwagner@google.com> Reviewed-by: Eric Boren <borenet@google.com>
2017-11-21 21:47:16 +00:00
"push",
"[START_DIR]/tmp/SK_IMAGE_VERSION",
"/sdcard/revenge_of_the_skiabot/SK_IMAGE_VERSION"
],
"cwd": "[START_DIR]/skia",
"env": {
"BUILDTYPE": "Debug",
"CHROME_HEADLESS": "1",
"PATH": "<PATH>:RECIPE_PACKAGE_REPO[depot_tools]",
"SKIA_OUT": "[START_DIR]/out"
},
"infra_step": true,
"name": "push [START_DIR]/tmp/SK_IMAGE_VERSION /sdcard/revenge_of_the_skiabot/SK_IMAGE_VERSION"
},
{
"cmd": [
"python",
"-u",
"RECIPE_MODULE[recipe_engine::file]/resources/fileutil.py",
"--json-output",
"/path/to/tmp/json",
"copy",
"[START_DIR]/skia/infra/bots/assets/svg/VERSION",
"/path/to/tmp/"
],
"infra_step": true,
"name": "Get downloaded SVG VERSION"
},
{
"cmd": [
"python",
"-u",
"RECIPE_MODULE[recipe_engine::file]/resources/fileutil.py",
"--json-output",
"/path/to/tmp/json",
"copy",
"42",
"[START_DIR]/tmp/SVG_VERSION"
],
"infra_step": true,
"name": "write SVG_VERSION"
},
{
"cmd": [
Use full paths for adb Add logic to run on Nexus 5xs in Golo. Bug:788839 Change-Id: I12290d11a0b1b0f012ada216da3e5b2599979c5e Reviewed-on: https://skia-review.googlesource.com/81861 Commit-Queue: Kevin Lubick <kjlubick@google.com> Reviewed-by: Eric Boren <borenet@google.com>
2017-12-07 16:19:31 +00:00
"/usr/bin/adb.1.0.35",
Make CPU scaling code cleaner This also makes *sure* the CPU frequency we set the device to actually "takes". Along the way, I learned if scaling_max_freq is < the frequency we set, the scaling_max_freq will be used instead, which was happening to the PixelCs and AndroidOnes. As a result, this may make those two Test- configs faster. Bug: skia: Change-Id: I10c98d37e296a19e1cf67bfe7269bb59cdd912d5 Reviewed-on: https://skia-review.googlesource.com/74360 Commit-Queue: Kevin Lubick <kjlubick@google.com> Reviewed-by: Ben Wagner <benjaminwagner@google.com> Reviewed-by: Eric Boren <borenet@google.com>
2017-11-21 21:47:16 +00:00
"shell",
"cat",
"/sdcard/revenge_of_the_skiabot/SVG_VERSION"
],
"cwd": "[START_DIR]/skia",
"env": {
"BUILDTYPE": "Debug",
"CHROME_HEADLESS": "1",
"PATH": "<PATH>:RECIPE_PACKAGE_REPO[depot_tools]",
"SKIA_OUT": "[START_DIR]/out"
},
"infra_step": true,
"name": "read /sdcard/revenge_of_the_skiabot/SVG_VERSION",
"stdout": "/path/to/tmp/"
},
{
"cmd": [
Use full paths for adb Add logic to run on Nexus 5xs in Golo. Bug:788839 Change-Id: I12290d11a0b1b0f012ada216da3e5b2599979c5e Reviewed-on: https://skia-review.googlesource.com/81861 Commit-Queue: Kevin Lubick <kjlubick@google.com> Reviewed-by: Eric Boren <borenet@google.com>
2017-12-07 16:19:31 +00:00
"/usr/bin/adb.1.0.35",
Make CPU scaling code cleaner This also makes *sure* the CPU frequency we set the device to actually "takes". Along the way, I learned if scaling_max_freq is < the frequency we set, the scaling_max_freq will be used instead, which was happening to the PixelCs and AndroidOnes. As a result, this may make those two Test- configs faster. Bug: skia: Change-Id: I10c98d37e296a19e1cf67bfe7269bb59cdd912d5 Reviewed-on: https://skia-review.googlesource.com/74360 Commit-Queue: Kevin Lubick <kjlubick@google.com> Reviewed-by: Ben Wagner <benjaminwagner@google.com> Reviewed-by: Eric Boren <borenet@google.com>
2017-11-21 21:47:16 +00:00
"shell",
"rm",
"-f",
"/sdcard/revenge_of_the_skiabot/SVG_VERSION"
],
"cwd": "[START_DIR]/skia",
"env": {
"BUILDTYPE": "Debug",
"CHROME_HEADLESS": "1",
"PATH": "<PATH>:RECIPE_PACKAGE_REPO[depot_tools]",
"SKIA_OUT": "[START_DIR]/out"
},
"infra_step": true,
"name": "rm /sdcard/revenge_of_the_skiabot/SVG_VERSION"
},
{
"cmd": [
Use full paths for adb Add logic to run on Nexus 5xs in Golo. Bug:788839 Change-Id: I12290d11a0b1b0f012ada216da3e5b2599979c5e Reviewed-on: https://skia-review.googlesource.com/81861 Commit-Queue: Kevin Lubick <kjlubick@google.com> Reviewed-by: Eric Boren <borenet@google.com>
2017-12-07 16:19:31 +00:00
"/usr/bin/adb.1.0.35",
Make CPU scaling code cleaner This also makes *sure* the CPU frequency we set the device to actually "takes". Along the way, I learned if scaling_max_freq is < the frequency we set, the scaling_max_freq will be used instead, which was happening to the PixelCs and AndroidOnes. As a result, this may make those two Test- configs faster. Bug: skia: Change-Id: I10c98d37e296a19e1cf67bfe7269bb59cdd912d5 Reviewed-on: https://skia-review.googlesource.com/74360 Commit-Queue: Kevin Lubick <kjlubick@google.com> Reviewed-by: Ben Wagner <benjaminwagner@google.com> Reviewed-by: Eric Boren <borenet@google.com>
2017-11-21 21:47:16 +00:00
"shell",
"rm",
"-rf",
"/sdcard/revenge_of_the_skiabot/svgs"
],
"cwd": "[START_DIR]/skia",
"env": {
"BUILDTYPE": "Debug",
"CHROME_HEADLESS": "1",
"PATH": "<PATH>:RECIPE_PACKAGE_REPO[depot_tools]",
"SKIA_OUT": "[START_DIR]/out"
},
"infra_step": true,
"name": "rm /sdcard/revenge_of_the_skiabot/svgs"
},
{
"cmd": [
Use full paths for adb Add logic to run on Nexus 5xs in Golo. Bug:788839 Change-Id: I12290d11a0b1b0f012ada216da3e5b2599979c5e Reviewed-on: https://skia-review.googlesource.com/81861 Commit-Queue: Kevin Lubick <kjlubick@google.com> Reviewed-by: Eric Boren <borenet@google.com>
2017-12-07 16:19:31 +00:00
"/usr/bin/adb.1.0.35",
Make CPU scaling code cleaner This also makes *sure* the CPU frequency we set the device to actually "takes". Along the way, I learned if scaling_max_freq is < the frequency we set, the scaling_max_freq will be used instead, which was happening to the PixelCs and AndroidOnes. As a result, this may make those two Test- configs faster. Bug: skia: Change-Id: I10c98d37e296a19e1cf67bfe7269bb59cdd912d5 Reviewed-on: https://skia-review.googlesource.com/74360 Commit-Queue: Kevin Lubick <kjlubick@google.com> Reviewed-by: Ben Wagner <benjaminwagner@google.com> Reviewed-by: Eric Boren <borenet@google.com>
2017-11-21 21:47:16 +00:00
"shell",
"mkdir",
"-p",
"/sdcard/revenge_of_the_skiabot/svgs"
],
"cwd": "[START_DIR]/skia",
"env": {
"BUILDTYPE": "Debug",
"CHROME_HEADLESS": "1",
"PATH": "<PATH>:RECIPE_PACKAGE_REPO[depot_tools]",
"SKIA_OUT": "[START_DIR]/out"
},
"infra_step": true,
"name": "mkdir /sdcard/revenge_of_the_skiabot/svgs"
},
{
"cmd": [
"python",
"-u",
Use full paths for adb Add logic to run on Nexus 5xs in Golo. Bug:788839 Change-Id: I12290d11a0b1b0f012ada216da3e5b2599979c5e Reviewed-on: https://skia-review.googlesource.com/81861 Commit-Queue: Kevin Lubick <kjlubick@google.com> Reviewed-by: Eric Boren <borenet@google.com>
2017-12-07 16:19:31 +00:00
"\nimport os\nimport subprocess\nimport sys\nhost = sys.argv[1]\ndevice = sys.argv[2]\nfor d, _, fs in os.walk(host):\n p = os.path.relpath(d, host)\n if p != '.' and p.startswith('.'):\n continue\n for f in fs:\n print os.path.join(p,f)\n subprocess.check_call(['/usr/bin/adb.1.0.35', 'push',\n os.path.realpath(os.path.join(host, p, f)),\n os.path.join(device, p, f)])\n",
Make CPU scaling code cleaner This also makes *sure* the CPU frequency we set the device to actually "takes". Along the way, I learned if scaling_max_freq is < the frequency we set, the scaling_max_freq will be used instead, which was happening to the PixelCs and AndroidOnes. As a result, this may make those two Test- configs faster. Bug: skia: Change-Id: I10c98d37e296a19e1cf67bfe7269bb59cdd912d5 Reviewed-on: https://skia-review.googlesource.com/74360 Commit-Queue: Kevin Lubick <kjlubick@google.com> Reviewed-by: Ben Wagner <benjaminwagner@google.com> Reviewed-by: Eric Boren <borenet@google.com>
2017-11-21 21:47:16 +00:00
"[START_DIR]/svg",
"/sdcard/revenge_of_the_skiabot/svgs"
],
"env": {
"BUILDTYPE": "Debug",
"CHROME_HEADLESS": "1",
"PATH": "<PATH>:RECIPE_PACKAGE_REPO[depot_tools]",
"SKIA_OUT": "[START_DIR]/out"
},
"infra_step": true,
"name": "push [START_DIR]/svg/* /sdcard/revenge_of_the_skiabot/svgs",
"~followup_annotations": [
"@@@STEP_LOG_LINE@python.inline@@@@",
"@@@STEP_LOG_LINE@python.inline@import os@@@",
"@@@STEP_LOG_LINE@python.inline@import subprocess@@@",
"@@@STEP_LOG_LINE@python.inline@import sys@@@",
"@@@STEP_LOG_LINE@python.inline@host = sys.argv[1]@@@",
"@@@STEP_LOG_LINE@python.inline@device = sys.argv[2]@@@",
"@@@STEP_LOG_LINE@python.inline@for d, _, fs in os.walk(host):@@@",
"@@@STEP_LOG_LINE@python.inline@ p = os.path.relpath(d, host)@@@",
"@@@STEP_LOG_LINE@python.inline@ if p != '.' and p.startswith('.'):@@@",
"@@@STEP_LOG_LINE@python.inline@ continue@@@",
"@@@STEP_LOG_LINE@python.inline@ for f in fs:@@@",
"@@@STEP_LOG_LINE@python.inline@ print os.path.join(p,f)@@@",
Use full paths for adb Add logic to run on Nexus 5xs in Golo. Bug:788839 Change-Id: I12290d11a0b1b0f012ada216da3e5b2599979c5e Reviewed-on: https://skia-review.googlesource.com/81861 Commit-Queue: Kevin Lubick <kjlubick@google.com> Reviewed-by: Eric Boren <borenet@google.com>
2017-12-07 16:19:31 +00:00
"@@@STEP_LOG_LINE@python.inline@ subprocess.check_call(['/usr/bin/adb.1.0.35', 'push',@@@",
Make CPU scaling code cleaner This also makes *sure* the CPU frequency we set the device to actually "takes". Along the way, I learned if scaling_max_freq is < the frequency we set, the scaling_max_freq will be used instead, which was happening to the PixelCs and AndroidOnes. As a result, this may make those two Test- configs faster. Bug: skia: Change-Id: I10c98d37e296a19e1cf67bfe7269bb59cdd912d5 Reviewed-on: https://skia-review.googlesource.com/74360 Commit-Queue: Kevin Lubick <kjlubick@google.com> Reviewed-by: Ben Wagner <benjaminwagner@google.com> Reviewed-by: Eric Boren <borenet@google.com>
2017-11-21 21:47:16 +00:00
"@@@STEP_LOG_LINE@python.inline@ os.path.realpath(os.path.join(host, p, f)),@@@",
"@@@STEP_LOG_LINE@python.inline@ os.path.join(device, p, f)])@@@",
"@@@STEP_LOG_END@python.inline@@@"
]
},
{
"cmd": [
Use full paths for adb Add logic to run on Nexus 5xs in Golo. Bug:788839 Change-Id: I12290d11a0b1b0f012ada216da3e5b2599979c5e Reviewed-on: https://skia-review.googlesource.com/81861 Commit-Queue: Kevin Lubick <kjlubick@google.com> Reviewed-by: Eric Boren <borenet@google.com>
2017-12-07 16:19:31 +00:00
"/usr/bin/adb.1.0.35",
Make CPU scaling code cleaner This also makes *sure* the CPU frequency we set the device to actually "takes". Along the way, I learned if scaling_max_freq is < the frequency we set, the scaling_max_freq will be used instead, which was happening to the PixelCs and AndroidOnes. As a result, this may make those two Test- configs faster. Bug: skia: Change-Id: I10c98d37e296a19e1cf67bfe7269bb59cdd912d5 Reviewed-on: https://skia-review.googlesource.com/74360 Commit-Queue: Kevin Lubick <kjlubick@google.com> Reviewed-by: Ben Wagner <benjaminwagner@google.com> Reviewed-by: Eric Boren <borenet@google.com>
2017-11-21 21:47:16 +00:00
"push",
"[START_DIR]/tmp/SVG_VERSION",
"/sdcard/revenge_of_the_skiabot/SVG_VERSION"
],
"cwd": "[START_DIR]/skia",
"env": {
"BUILDTYPE": "Debug",
"CHROME_HEADLESS": "1",
"PATH": "<PATH>:RECIPE_PACKAGE_REPO[depot_tools]",
"SKIA_OUT": "[START_DIR]/out"
},
"infra_step": true,
"name": "push [START_DIR]/tmp/SVG_VERSION /sdcard/revenge_of_the_skiabot/SVG_VERSION"
},
{
"cmd": [
"python",
"-u",
"import os\nprint os.environ.get('SWARMING_BOT_ID', '')\n"
],
"name": "get swarming bot id",
"stdout": "/path/to/tmp/",
"~followup_annotations": [
"@@@STEP_LOG_LINE@python.inline@import os@@@",
"@@@STEP_LOG_LINE@python.inline@print os.environ.get('SWARMING_BOT_ID', '')@@@",
"@@@STEP_LOG_END@python.inline@@@"
]
},
{
"cmd": [
"python",
"-u",
"import os\nprint os.environ.get('SWARMING_TASK_ID', '')\n"
],
"name": "get swarming task id",
"stdout": "/path/to/tmp/",
"~followup_annotations": [
"@@@STEP_LOG_LINE@python.inline@import os@@@",
"@@@STEP_LOG_LINE@python.inline@print os.environ.get('SWARMING_TASK_ID', '')@@@",
"@@@STEP_LOG_END@python.inline@@@"
]
},
{
"cmd": [
"python",
"-u",
Fix scaling logic for NexusPlayer and AndroidOne They seem to be their own special cases for what governors they support. Bug: skia: NOTRY=true Change-Id: I7bb220e1d3ba6851c17c7e6ef327aab24ffdba42 Reviewed-on: https://skia-review.googlesource.com/83900 Reviewed-by: Kevin Lubick <kjlubick@google.com> Commit-Queue: Kevin Lubick <kjlubick@google.com>
2017-12-12 13:14:19 +00:00
"\nimport os\nimport subprocess\nimport sys\nimport time\nADB = sys.argv[1]\ncpu = int(sys.argv[2])\ngov = sys.argv[3]\n\nlog = subprocess.check_output([ADB, 'root'])\n# check for message like 'adbd cannot run as root in production builds'\nprint log\nif 'cannot' in log:\n raise Exception('adb root failed')\n\nsubprocess.check_output([ADB, 'shell', 'echo \"%s\" > '\n '/sys/devices/system/cpu/cpu%d/cpufreq/scaling_governor' % (gov, cpu)])\nactual_gov = subprocess.check_output([ADB, 'shell', 'cat '\n '/sys/devices/system/cpu/cpu%d/cpufreq/scaling_governor' % cpu]).strip()\nif actual_gov != gov:\n raise Exception('(actual, expected) (%s, %s)'\n % (actual_gov, gov))\n",
Use full paths for adb Add logic to run on Nexus 5xs in Golo. Bug:788839 Change-Id: I12290d11a0b1b0f012ada216da3e5b2599979c5e Reviewed-on: https://skia-review.googlesource.com/81861 Commit-Queue: Kevin Lubick <kjlubick@google.com> Reviewed-by: Eric Boren <borenet@google.com>
2017-12-07 16:19:31 +00:00
"/usr/bin/adb.1.0.35",
Scale extra CPUs to powersave during nanobench. This scaling logic correctly accounts for some devices which have multiple CPUs. Previously, we were scaling the smaller of these CPUs, which likely had a negative impact on nanobench, given nanobench was single threaded and the CPUs weren't allowed to idle much (because we set the CPU). This CL sets those additional CPUs to powersave when we run nanobench and then correctly scales down the beefier CPU we want to run nanobench on. For DM, we just run it in ondemand mode, which will hopefully be "as fast as possible", but allow the CPU governor to scale down if overheating becomes a problem. Bug: skia:7378 notry=TRUE Change-Id: I45ca5d9fb32182233d1b2d094842c879f2b84da4 Reviewed-on: https://skia-review.googlesource.com/83240 Commit-Queue: Kevin Lubick <kjlubick@google.com> Reviewed-by: Mike Klein <mtklein@chromium.org>
2017-12-12 12:33:48 +00:00
"0",
"userspace"
],
"env": {
"BUILDTYPE": "Debug",
"CHROME_HEADLESS": "1",
"PATH": "<PATH>:RECIPE_PACKAGE_REPO[depot_tools]",
"SKIA_OUT": "[START_DIR]/out"
},
"infra_step": true,
Experiment with turning CPUs off during perf Perf was nice and flat after https://skia-review.googlesource.com/c/skia/+/83240 but there was a 4-5x slowdown on the benchmarks themselves, indicating that perhaps we were running nanobench on the LITTLE cores (now on powersave) instead of the big cores we thought (which were recently scaled down). This experiment will show us FOR CERTAIN that we are running on core 0 or 1 which is at .6 max frequency. We expect to see a speedup from the previous results. This is leading to turning of CPUs we don't need to make sure nanobench is running on the ones we expect. Bug: skia: NOTRY=true Change-Id: Ida65181e4d90e778e65e3f22d761288b9ade64f6 Reviewed-on: https://skia-review.googlesource.com/84201 Reviewed-by: Mike Klein <mtklein@chromium.org> Commit-Queue: Kevin Lubick <kjlubick@google.com>
2017-12-12 20:57:20 +00:00
"name": "Set CPU 0's governor to userspace",
Scale extra CPUs to powersave during nanobench. This scaling logic correctly accounts for some devices which have multiple CPUs. Previously, we were scaling the smaller of these CPUs, which likely had a negative impact on nanobench, given nanobench was single threaded and the CPUs weren't allowed to idle much (because we set the CPU). This CL sets those additional CPUs to powersave when we run nanobench and then correctly scales down the beefier CPU we want to run nanobench on. For DM, we just run it in ondemand mode, which will hopefully be "as fast as possible", but allow the CPU governor to scale down if overheating becomes a problem. Bug: skia:7378 notry=TRUE Change-Id: I45ca5d9fb32182233d1b2d094842c879f2b84da4 Reviewed-on: https://skia-review.googlesource.com/83240 Commit-Queue: Kevin Lubick <kjlubick@google.com> Reviewed-by: Mike Klein <mtklein@chromium.org>
2017-12-12 12:33:48 +00:00
"timeout": 30,
"~followup_annotations": [
"@@@STEP_LOG_LINE@python.inline@@@@",
"@@@STEP_LOG_LINE@python.inline@import os@@@",
"@@@STEP_LOG_LINE@python.inline@import subprocess@@@",
"@@@STEP_LOG_LINE@python.inline@import sys@@@",
"@@@STEP_LOG_LINE@python.inline@import time@@@",
"@@@STEP_LOG_LINE@python.inline@ADB = sys.argv[1]@@@",
"@@@STEP_LOG_LINE@python.inline@cpu = int(sys.argv[2])@@@",
"@@@STEP_LOG_LINE@python.inline@gov = sys.argv[3]@@@",
"@@@STEP_LOG_LINE@python.inline@@@@",
"@@@STEP_LOG_LINE@python.inline@log = subprocess.check_output([ADB, 'root'])@@@",
"@@@STEP_LOG_LINE@python.inline@# check for message like 'adbd cannot run as root in production builds'@@@",
"@@@STEP_LOG_LINE@python.inline@print log@@@",
"@@@STEP_LOG_LINE@python.inline@if 'cannot' in log:@@@",
"@@@STEP_LOG_LINE@python.inline@ raise Exception('adb root failed')@@@",
"@@@STEP_LOG_LINE@python.inline@@@@",
"@@@STEP_LOG_LINE@python.inline@subprocess.check_output([ADB, 'shell', 'echo \"%s\" > '@@@",
"@@@STEP_LOG_LINE@python.inline@ '/sys/devices/system/cpu/cpu%d/cpufreq/scaling_governor' % (gov, cpu)])@@@",
"@@@STEP_LOG_LINE@python.inline@actual_gov = subprocess.check_output([ADB, 'shell', 'cat '@@@",
"@@@STEP_LOG_LINE@python.inline@ '/sys/devices/system/cpu/cpu%d/cpufreq/scaling_governor' % cpu]).strip()@@@",
Fix scaling logic for NexusPlayer and AndroidOne They seem to be their own special cases for what governors they support. Bug: skia: NOTRY=true Change-Id: I7bb220e1d3ba6851c17c7e6ef327aab24ffdba42 Reviewed-on: https://skia-review.googlesource.com/83900 Reviewed-by: Kevin Lubick <kjlubick@google.com> Commit-Queue: Kevin Lubick <kjlubick@google.com>
2017-12-12 13:14:19 +00:00
"@@@STEP_LOG_LINE@python.inline@if actual_gov != gov:@@@",
"@@@STEP_LOG_LINE@python.inline@ raise Exception('(actual, expected) (%s, %s)'@@@",
Scale extra CPUs to powersave during nanobench. This scaling logic correctly accounts for some devices which have multiple CPUs. Previously, we were scaling the smaller of these CPUs, which likely had a negative impact on nanobench, given nanobench was single threaded and the CPUs weren't allowed to idle much (because we set the CPU). This CL sets those additional CPUs to powersave when we run nanobench and then correctly scales down the beefier CPU we want to run nanobench on. For DM, we just run it in ondemand mode, which will hopefully be "as fast as possible", but allow the CPU governor to scale down if overheating becomes a problem. Bug: skia:7378 notry=TRUE Change-Id: I45ca5d9fb32182233d1b2d094842c879f2b84da4 Reviewed-on: https://skia-review.googlesource.com/83240 Commit-Queue: Kevin Lubick <kjlubick@google.com> Reviewed-by: Mike Klein <mtklein@chromium.org>
2017-12-12 12:33:48 +00:00
"@@@STEP_LOG_LINE@python.inline@ % (actual_gov, gov))@@@",
"@@@STEP_LOG_END@python.inline@@@"
]
},
{
"cmd": [
"python",
"-u",
"\nimport os\nimport subprocess\nimport sys\nimport time\nADB = sys.argv[1]\ntarget_percent = float(sys.argv[2])\ncpu = int(sys.argv[3])\nlog = subprocess.check_output([ADB, 'root'])\n# check for message like 'adbd cannot run as root in production builds'\nprint log\nif 'cannot' in log:\n raise Exception('adb root failed')\n\nroot = '/sys/devices/system/cpu/cpu%d/cpufreq' %cpu\n\n# All devices we test on give a list of their available frequencies.\navailable_freqs = subprocess.check_output([ADB, 'shell',\n 'cat %s/scaling_available_frequencies' % root])\n\n# Check for message like '/system/bin/sh: file not found'\nif available_freqs and '/system/bin/sh' not in available_freqs:\n available_freqs = sorted(\n int(i) for i in available_freqs.strip().split())\nelse:\n raise Exception('Could not get list of available frequencies: %s' %\n available_freqs)\n\nmaxfreq = available_freqs[-1]\ntarget = int(round(maxfreq * target_percent))\nfreq = maxfreq\nfor f in reversed(available_freqs):\n if f <= target:\n freq = f\n break\n\nprint 'Setting frequency to %d' % freq\n\n# If scaling_max_freq is lower than our attempted setting, it won't take.\n# We must set min first, because if we try to set max to be less than min\n# (which sometimes happens after certain devices reboot) it returns a\n# perplexing permissions error.\nsubprocess.check_output([ADB, 'shell', 'echo 0 > '\n '%s/scaling_min_freq' % root])\nsubprocess.check_output([ADB, 'shell', 'echo %d > '\n '%s/scaling_max_freq' % (freq, root)])\nsubprocess.check_output([ADB, 'shell', 'echo %d > '\n '%s/scaling_setspeed' % (freq, root)])\ntime.sleep(5)\nactual_freq = subprocess.check_output([ADB, 'shell', 'cat '\n '%s/scaling_cur_freq' % root]).strip()\nif actual_freq != str(freq):\n raise Exception('(actual, expected) (%s, %d)'\n % (actual_freq, freq))\n",
"/usr/bin/adb.1.0.35",
"0.6",
Experiment with turning CPUs off during perf Perf was nice and flat after https://skia-review.googlesource.com/c/skia/+/83240 but there was a 4-5x slowdown on the benchmarks themselves, indicating that perhaps we were running nanobench on the LITTLE cores (now on powersave) instead of the big cores we thought (which were recently scaled down). This experiment will show us FOR CERTAIN that we are running on core 0 or 1 which is at .6 max frequency. We expect to see a speedup from the previous results. This is leading to turning of CPUs we don't need to make sure nanobench is running on the ones we expect. Bug: skia: NOTRY=true Change-Id: Ida65181e4d90e778e65e3f22d761288b9ade64f6 Reviewed-on: https://skia-review.googlesource.com/84201 Reviewed-by: Mike Klein <mtklein@chromium.org> Commit-Queue: Kevin Lubick <kjlubick@google.com>
2017-12-12 20:57:20 +00:00
"0"
Scale extra CPUs to powersave during nanobench. This scaling logic correctly accounts for some devices which have multiple CPUs. Previously, we were scaling the smaller of these CPUs, which likely had a negative impact on nanobench, given nanobench was single threaded and the CPUs weren't allowed to idle much (because we set the CPU). This CL sets those additional CPUs to powersave when we run nanobench and then correctly scales down the beefier CPU we want to run nanobench on. For DM, we just run it in ondemand mode, which will hopefully be "as fast as possible", but allow the CPU governor to scale down if overheating becomes a problem. Bug: skia:7378 notry=TRUE Change-Id: I45ca5d9fb32182233d1b2d094842c879f2b84da4 Reviewed-on: https://skia-review.googlesource.com/83240 Commit-Queue: Kevin Lubick <kjlubick@google.com> Reviewed-by: Mike Klein <mtklein@chromium.org>
2017-12-12 12:33:48 +00:00
],
"env": {
"BUILDTYPE": "Debug",
"CHROME_HEADLESS": "1",
"PATH": "<PATH>:RECIPE_PACKAGE_REPO[depot_tools]",
"SKIA_OUT": "[START_DIR]/out"
},
"infra_step": true,
Experiment with turning CPUs off during perf Perf was nice and flat after https://skia-review.googlesource.com/c/skia/+/83240 but there was a 4-5x slowdown on the benchmarks themselves, indicating that perhaps we were running nanobench on the LITTLE cores (now on powersave) instead of the big cores we thought (which were recently scaled down). This experiment will show us FOR CERTAIN that we are running on core 0 or 1 which is at .6 max frequency. We expect to see a speedup from the previous results. This is leading to turning of CPUs we don't need to make sure nanobench is running on the ones we expect. Bug: skia: NOTRY=true Change-Id: Ida65181e4d90e778e65e3f22d761288b9ade64f6 Reviewed-on: https://skia-review.googlesource.com/84201 Reviewed-by: Mike Klein <mtklein@chromium.org> Commit-Queue: Kevin Lubick <kjlubick@google.com>
2017-12-12 20:57:20 +00:00
"name": "Scale CPU 0 to 0.600000",
Make CPU scaling code cleaner This also makes *sure* the CPU frequency we set the device to actually "takes". Along the way, I learned if scaling_max_freq is < the frequency we set, the scaling_max_freq will be used instead, which was happening to the PixelCs and AndroidOnes. As a result, this may make those two Test- configs faster. Bug: skia: Change-Id: I10c98d37e296a19e1cf67bfe7269bb59cdd912d5 Reviewed-on: https://skia-review.googlesource.com/74360 Commit-Queue: Kevin Lubick <kjlubick@google.com> Reviewed-by: Ben Wagner <benjaminwagner@google.com> Reviewed-by: Eric Boren <borenet@google.com>
2017-11-21 21:47:16 +00:00
"timeout": 30,
"~followup_annotations": [
"step returned non-zero exit code: 1",
"@@@STEP_LOG_LINE@python.inline@@@@",
"@@@STEP_LOG_LINE@python.inline@import os@@@",
"@@@STEP_LOG_LINE@python.inline@import subprocess@@@",
"@@@STEP_LOG_LINE@python.inline@import sys@@@",
Make cpu scaling more robust Some bots, especially the Nexus 7s, seemed to occasionally fail when setting the CPU frequency. I was unable to repro this behavior, so this is a shotgun approach. We add a 5 second delay between setting and checking, checking frequency using scaling_cur_freq instead of scaling_setspeed, set the min_freq as well as max_freq, and retry up to 3 times if setting cpu frequency fails. NOTRY=true Bug: skia: Change-Id: Id4d85d8d509c9dba8e3a0e06b5992f5adadf36d2 Reviewed-on: https://skia-review.googlesource.com/78140 Commit-Queue: Kevin Lubick <kjlubick@google.com> Reviewed-by: Eric Boren <borenet@google.com>
2017-11-30 13:35:29 +00:00
"@@@STEP_LOG_LINE@python.inline@import time@@@",
Make CPU scaling code cleaner This also makes *sure* the CPU frequency we set the device to actually "takes". Along the way, I learned if scaling_max_freq is < the frequency we set, the scaling_max_freq will be used instead, which was happening to the PixelCs and AndroidOnes. As a result, this may make those two Test- configs faster. Bug: skia: Change-Id: I10c98d37e296a19e1cf67bfe7269bb59cdd912d5 Reviewed-on: https://skia-review.googlesource.com/74360 Commit-Queue: Kevin Lubick <kjlubick@google.com> Reviewed-by: Ben Wagner <benjaminwagner@google.com> Reviewed-by: Eric Boren <borenet@google.com>
2017-11-21 21:47:16 +00:00
"@@@STEP_LOG_LINE@python.inline@ADB = sys.argv[1]@@@",
Scale extra CPUs to powersave during nanobench. This scaling logic correctly accounts for some devices which have multiple CPUs. Previously, we were scaling the smaller of these CPUs, which likely had a negative impact on nanobench, given nanobench was single threaded and the CPUs weren't allowed to idle much (because we set the CPU). This CL sets those additional CPUs to powersave when we run nanobench and then correctly scales down the beefier CPU we want to run nanobench on. For DM, we just run it in ondemand mode, which will hopefully be "as fast as possible", but allow the CPU governor to scale down if overheating becomes a problem. Bug: skia:7378 notry=TRUE Change-Id: I45ca5d9fb32182233d1b2d094842c879f2b84da4 Reviewed-on: https://skia-review.googlesource.com/83240 Commit-Queue: Kevin Lubick <kjlubick@google.com> Reviewed-by: Mike Klein <mtklein@chromium.org>
2017-12-12 12:33:48 +00:00
"@@@STEP_LOG_LINE@python.inline@target_percent = float(sys.argv[2])@@@",
"@@@STEP_LOG_LINE@python.inline@cpu = int(sys.argv[3])@@@",
Make CPU scaling code cleaner This also makes *sure* the CPU frequency we set the device to actually "takes". Along the way, I learned if scaling_max_freq is < the frequency we set, the scaling_max_freq will be used instead, which was happening to the PixelCs and AndroidOnes. As a result, this may make those two Test- configs faster. Bug: skia: Change-Id: I10c98d37e296a19e1cf67bfe7269bb59cdd912d5 Reviewed-on: https://skia-review.googlesource.com/74360 Commit-Queue: Kevin Lubick <kjlubick@google.com> Reviewed-by: Ben Wagner <benjaminwagner@google.com> Reviewed-by: Eric Boren <borenet@google.com>
2017-11-21 21:47:16 +00:00
"@@@STEP_LOG_LINE@python.inline@log = subprocess.check_output([ADB, 'root'])@@@",
"@@@STEP_LOG_LINE@python.inline@# check for message like 'adbd cannot run as root in production builds'@@@",
Fix scaling issues for some devices (e.g. Pixel) Bug: skia: NOTRY=true Change-Id: Ia6a80ecb97b5d30197932ec7d340163624e315bd Reviewed-on: https://skia-review.googlesource.com/82682 Reviewed-by: Ravi Mistry <rmistry@google.com> Commit-Queue: Kevin Lubick <kjlubick@google.com>
2017-12-08 15:26:56 +00:00
"@@@STEP_LOG_LINE@python.inline@print log@@@",
Make CPU scaling code cleaner This also makes *sure* the CPU frequency we set the device to actually "takes". Along the way, I learned if scaling_max_freq is < the frequency we set, the scaling_max_freq will be used instead, which was happening to the PixelCs and AndroidOnes. As a result, this may make those two Test- configs faster. Bug: skia: Change-Id: I10c98d37e296a19e1cf67bfe7269bb59cdd912d5 Reviewed-on: https://skia-review.googlesource.com/74360 Commit-Queue: Kevin Lubick <kjlubick@google.com> Reviewed-by: Ben Wagner <benjaminwagner@google.com> Reviewed-by: Eric Boren <borenet@google.com>
2017-11-21 21:47:16 +00:00
"@@@STEP_LOG_LINE@python.inline@if 'cannot' in log:@@@",
"@@@STEP_LOG_LINE@python.inline@ raise Exception('adb root failed')@@@",
"@@@STEP_LOG_LINE@python.inline@@@@",
Scale extra CPUs to powersave during nanobench. This scaling logic correctly accounts for some devices which have multiple CPUs. Previously, we were scaling the smaller of these CPUs, which likely had a negative impact on nanobench, given nanobench was single threaded and the CPUs weren't allowed to idle much (because we set the CPU). This CL sets those additional CPUs to powersave when we run nanobench and then correctly scales down the beefier CPU we want to run nanobench on. For DM, we just run it in ondemand mode, which will hopefully be "as fast as possible", but allow the CPU governor to scale down if overheating becomes a problem. Bug: skia:7378 notry=TRUE Change-Id: I45ca5d9fb32182233d1b2d094842c879f2b84da4 Reviewed-on: https://skia-review.googlesource.com/83240 Commit-Queue: Kevin Lubick <kjlubick@google.com> Reviewed-by: Mike Klein <mtklein@chromium.org>
2017-12-12 12:33:48 +00:00
"@@@STEP_LOG_LINE@python.inline@root = '/sys/devices/system/cpu/cpu%d/cpufreq' %cpu@@@",
"@@@STEP_LOG_LINE@python.inline@@@@",
"@@@STEP_LOG_LINE@python.inline@# All devices we test on give a list of their available frequencies.@@@",
"@@@STEP_LOG_LINE@python.inline@available_freqs = subprocess.check_output([ADB, 'shell',@@@",
"@@@STEP_LOG_LINE@python.inline@ 'cat %s/scaling_available_frequencies' % root])@@@",
"@@@STEP_LOG_LINE@python.inline@@@@",
"@@@STEP_LOG_LINE@python.inline@# Check for message like '/system/bin/sh: file not found'@@@",
"@@@STEP_LOG_LINE@python.inline@if available_freqs and '/system/bin/sh' not in available_freqs:@@@",
"@@@STEP_LOG_LINE@python.inline@ available_freqs = sorted(@@@",
"@@@STEP_LOG_LINE@python.inline@ int(i) for i in available_freqs.strip().split())@@@",
Make CPU scaling code cleaner This also makes *sure* the CPU frequency we set the device to actually "takes". Along the way, I learned if scaling_max_freq is < the frequency we set, the scaling_max_freq will be used instead, which was happening to the PixelCs and AndroidOnes. As a result, this may make those two Test- configs faster. Bug: skia: Change-Id: I10c98d37e296a19e1cf67bfe7269bb59cdd912d5 Reviewed-on: https://skia-review.googlesource.com/74360 Commit-Queue: Kevin Lubick <kjlubick@google.com> Reviewed-by: Ben Wagner <benjaminwagner@google.com> Reviewed-by: Eric Boren <borenet@google.com>
2017-11-21 21:47:16 +00:00
"@@@STEP_LOG_LINE@python.inline@else:@@@",
Scale extra CPUs to powersave during nanobench. This scaling logic correctly accounts for some devices which have multiple CPUs. Previously, we were scaling the smaller of these CPUs, which likely had a negative impact on nanobench, given nanobench was single threaded and the CPUs weren't allowed to idle much (because we set the CPU). This CL sets those additional CPUs to powersave when we run nanobench and then correctly scales down the beefier CPU we want to run nanobench on. For DM, we just run it in ondemand mode, which will hopefully be "as fast as possible", but allow the CPU governor to scale down if overheating becomes a problem. Bug: skia:7378 notry=TRUE Change-Id: I45ca5d9fb32182233d1b2d094842c879f2b84da4 Reviewed-on: https://skia-review.googlesource.com/83240 Commit-Queue: Kevin Lubick <kjlubick@google.com> Reviewed-by: Mike Klein <mtklein@chromium.org>
2017-12-12 12:33:48 +00:00
"@@@STEP_LOG_LINE@python.inline@ raise Exception('Could not get list of available frequencies: %s' %@@@",
"@@@STEP_LOG_LINE@python.inline@ available_freqs)@@@",
Make CPU scaling code cleaner This also makes *sure* the CPU frequency we set the device to actually "takes". Along the way, I learned if scaling_max_freq is < the frequency we set, the scaling_max_freq will be used instead, which was happening to the PixelCs and AndroidOnes. As a result, this may make those two Test- configs faster. Bug: skia: Change-Id: I10c98d37e296a19e1cf67bfe7269bb59cdd912d5 Reviewed-on: https://skia-review.googlesource.com/74360 Commit-Queue: Kevin Lubick <kjlubick@google.com> Reviewed-by: Ben Wagner <benjaminwagner@google.com> Reviewed-by: Eric Boren <borenet@google.com>
2017-11-21 21:47:16 +00:00
"@@@STEP_LOG_LINE@python.inline@@@@",
"@@@STEP_LOG_LINE@python.inline@maxfreq = available_freqs[-1]@@@",
"@@@STEP_LOG_LINE@python.inline@target = int(round(maxfreq * target_percent))@@@",
"@@@STEP_LOG_LINE@python.inline@freq = maxfreq@@@",
"@@@STEP_LOG_LINE@python.inline@for f in reversed(available_freqs):@@@",
"@@@STEP_LOG_LINE@python.inline@ if f <= target:@@@",
"@@@STEP_LOG_LINE@python.inline@ freq = f@@@",
"@@@STEP_LOG_LINE@python.inline@ break@@@",
"@@@STEP_LOG_LINE@python.inline@@@@",
"@@@STEP_LOG_LINE@python.inline@print 'Setting frequency to %d' % freq@@@",
"@@@STEP_LOG_LINE@python.inline@@@@",
"@@@STEP_LOG_LINE@python.inline@# If scaling_max_freq is lower than our attempted setting, it won't take.@@@",
Fix scaling issues for some devices (e.g. Pixel) Bug: skia: NOTRY=true Change-Id: Ia6a80ecb97b5d30197932ec7d340163624e315bd Reviewed-on: https://skia-review.googlesource.com/82682 Reviewed-by: Ravi Mistry <rmistry@google.com> Commit-Queue: Kevin Lubick <kjlubick@google.com>
2017-12-08 15:26:56 +00:00
"@@@STEP_LOG_LINE@python.inline@# We must set min first, because if we try to set max to be less than min@@@",
"@@@STEP_LOG_LINE@python.inline@# (which sometimes happens after certain devices reboot) it returns a@@@",
"@@@STEP_LOG_LINE@python.inline@# perplexing permissions error.@@@",
Don't set min frequency for Android devices Bug: skia:7378 NOTRY=true No-Tree-Checks: true Change-Id: Iaacbc8eadaa4da9832a53d4794a2214625bef9b6 Reviewed-on: https://skia-review.googlesource.com/80842 Commit-Queue: Kevin Lubick <kjlubick@google.com> Reviewed-by: Kevin Lubick <kjlubick@google.com>
2017-12-05 20:54:19 +00:00
"@@@STEP_LOG_LINE@python.inline@subprocess.check_output([ADB, 'shell', 'echo 0 > '@@@",
Scale extra CPUs to powersave during nanobench. This scaling logic correctly accounts for some devices which have multiple CPUs. Previously, we were scaling the smaller of these CPUs, which likely had a negative impact on nanobench, given nanobench was single threaded and the CPUs weren't allowed to idle much (because we set the CPU). This CL sets those additional CPUs to powersave when we run nanobench and then correctly scales down the beefier CPU we want to run nanobench on. For DM, we just run it in ondemand mode, which will hopefully be "as fast as possible", but allow the CPU governor to scale down if overheating becomes a problem. Bug: skia:7378 notry=TRUE Change-Id: I45ca5d9fb32182233d1b2d094842c879f2b84da4 Reviewed-on: https://skia-review.googlesource.com/83240 Commit-Queue: Kevin Lubick <kjlubick@google.com> Reviewed-by: Mike Klein <mtklein@chromium.org>
2017-12-12 12:33:48 +00:00
"@@@STEP_LOG_LINE@python.inline@ '%s/scaling_min_freq' % root])@@@",
Make cpu scaling more robust Some bots, especially the Nexus 7s, seemed to occasionally fail when setting the CPU frequency. I was unable to repro this behavior, so this is a shotgun approach. We add a 5 second delay between setting and checking, checking frequency using scaling_cur_freq instead of scaling_setspeed, set the min_freq as well as max_freq, and retry up to 3 times if setting cpu frequency fails. NOTRY=true Bug: skia: Change-Id: Id4d85d8d509c9dba8e3a0e06b5992f5adadf36d2 Reviewed-on: https://skia-review.googlesource.com/78140 Commit-Queue: Kevin Lubick <kjlubick@google.com> Reviewed-by: Eric Boren <borenet@google.com>
2017-11-30 13:35:29 +00:00
"@@@STEP_LOG_LINE@python.inline@subprocess.check_output([ADB, 'shell', 'echo %d > '@@@",
Scale extra CPUs to powersave during nanobench. This scaling logic correctly accounts for some devices which have multiple CPUs. Previously, we were scaling the smaller of these CPUs, which likely had a negative impact on nanobench, given nanobench was single threaded and the CPUs weren't allowed to idle much (because we set the CPU). This CL sets those additional CPUs to powersave when we run nanobench and then correctly scales down the beefier CPU we want to run nanobench on. For DM, we just run it in ondemand mode, which will hopefully be "as fast as possible", but allow the CPU governor to scale down if overheating becomes a problem. Bug: skia:7378 notry=TRUE Change-Id: I45ca5d9fb32182233d1b2d094842c879f2b84da4 Reviewed-on: https://skia-review.googlesource.com/83240 Commit-Queue: Kevin Lubick <kjlubick@google.com> Reviewed-by: Mike Klein <mtklein@chromium.org>
2017-12-12 12:33:48 +00:00
"@@@STEP_LOG_LINE@python.inline@ '%s/scaling_max_freq' % (freq, root)])@@@",
Fix scaling issues for some devices (e.g. Pixel) Bug: skia: NOTRY=true Change-Id: Ia6a80ecb97b5d30197932ec7d340163624e315bd Reviewed-on: https://skia-review.googlesource.com/82682 Reviewed-by: Ravi Mistry <rmistry@google.com> Commit-Queue: Kevin Lubick <kjlubick@google.com>
2017-12-08 15:26:56 +00:00
"@@@STEP_LOG_LINE@python.inline@subprocess.check_output([ADB, 'shell', 'echo %d > '@@@",
Scale extra CPUs to powersave during nanobench. This scaling logic correctly accounts for some devices which have multiple CPUs. Previously, we were scaling the smaller of these CPUs, which likely had a negative impact on nanobench, given nanobench was single threaded and the CPUs weren't allowed to idle much (because we set the CPU). This CL sets those additional CPUs to powersave when we run nanobench and then correctly scales down the beefier CPU we want to run nanobench on. For DM, we just run it in ondemand mode, which will hopefully be "as fast as possible", but allow the CPU governor to scale down if overheating becomes a problem. Bug: skia:7378 notry=TRUE Change-Id: I45ca5d9fb32182233d1b2d094842c879f2b84da4 Reviewed-on: https://skia-review.googlesource.com/83240 Commit-Queue: Kevin Lubick <kjlubick@google.com> Reviewed-by: Mike Klein <mtklein@chromium.org>
2017-12-12 12:33:48 +00:00
"@@@STEP_LOG_LINE@python.inline@ '%s/scaling_setspeed' % (freq, root)])@@@",
Make cpu scaling more robust Some bots, especially the Nexus 7s, seemed to occasionally fail when setting the CPU frequency. I was unable to repro this behavior, so this is a shotgun approach. We add a 5 second delay between setting and checking, checking frequency using scaling_cur_freq instead of scaling_setspeed, set the min_freq as well as max_freq, and retry up to 3 times if setting cpu frequency fails. NOTRY=true Bug: skia: Change-Id: Id4d85d8d509c9dba8e3a0e06b5992f5adadf36d2 Reviewed-on: https://skia-review.googlesource.com/78140 Commit-Queue: Kevin Lubick <kjlubick@google.com> Reviewed-by: Eric Boren <borenet@google.com>
2017-11-30 13:35:29 +00:00
"@@@STEP_LOG_LINE@python.inline@time.sleep(5)@@@",
"@@@STEP_LOG_LINE@python.inline@actual_freq = subprocess.check_output([ADB, 'shell', 'cat '@@@",
Scale extra CPUs to powersave during nanobench. This scaling logic correctly accounts for some devices which have multiple CPUs. Previously, we were scaling the smaller of these CPUs, which likely had a negative impact on nanobench, given nanobench was single threaded and the CPUs weren't allowed to idle much (because we set the CPU). This CL sets those additional CPUs to powersave when we run nanobench and then correctly scales down the beefier CPU we want to run nanobench on. For DM, we just run it in ondemand mode, which will hopefully be "as fast as possible", but allow the CPU governor to scale down if overheating becomes a problem. Bug: skia:7378 notry=TRUE Change-Id: I45ca5d9fb32182233d1b2d094842c879f2b84da4 Reviewed-on: https://skia-review.googlesource.com/83240 Commit-Queue: Kevin Lubick <kjlubick@google.com> Reviewed-by: Mike Klein <mtklein@chromium.org>
2017-12-12 12:33:48 +00:00
"@@@STEP_LOG_LINE@python.inline@ '%s/scaling_cur_freq' % root]).strip()@@@",
Make cpu scaling more robust Some bots, especially the Nexus 7s, seemed to occasionally fail when setting the CPU frequency. I was unable to repro this behavior, so this is a shotgun approach. We add a 5 second delay between setting and checking, checking frequency using scaling_cur_freq instead of scaling_setspeed, set the min_freq as well as max_freq, and retry up to 3 times if setting cpu frequency fails. NOTRY=true Bug: skia: Change-Id: Id4d85d8d509c9dba8e3a0e06b5992f5adadf36d2 Reviewed-on: https://skia-review.googlesource.com/78140 Commit-Queue: Kevin Lubick <kjlubick@google.com> Reviewed-by: Eric Boren <borenet@google.com>
2017-11-30 13:35:29 +00:00
"@@@STEP_LOG_LINE@python.inline@if actual_freq != str(freq):@@@",
"@@@STEP_LOG_LINE@python.inline@ raise Exception('(actual, expected) (%s, %d)'@@@",
"@@@STEP_LOG_LINE@python.inline@ % (actual_freq, freq))@@@",
"@@@STEP_LOG_END@python.inline@@@",
"@@@STEP_EXCEPTION@@@"
]
},
{
"cmd": [
"python",
"-u",
Scale extra CPUs to powersave during nanobench. This scaling logic correctly accounts for some devices which have multiple CPUs. Previously, we were scaling the smaller of these CPUs, which likely had a negative impact on nanobench, given nanobench was single threaded and the CPUs weren't allowed to idle much (because we set the CPU). This CL sets those additional CPUs to powersave when we run nanobench and then correctly scales down the beefier CPU we want to run nanobench on. For DM, we just run it in ondemand mode, which will hopefully be "as fast as possible", but allow the CPU governor to scale down if overheating becomes a problem. Bug: skia:7378 notry=TRUE Change-Id: I45ca5d9fb32182233d1b2d094842c879f2b84da4 Reviewed-on: https://skia-review.googlesource.com/83240 Commit-Queue: Kevin Lubick <kjlubick@google.com> Reviewed-by: Mike Klein <mtklein@chromium.org>
2017-12-12 12:33:48 +00:00
"\nimport os\nimport subprocess\nimport sys\nimport time\nADB = sys.argv[1]\ntarget_percent = float(sys.argv[2])\ncpu = int(sys.argv[3])\nlog = subprocess.check_output([ADB, 'root'])\n# check for message like 'adbd cannot run as root in production builds'\nprint log\nif 'cannot' in log:\n raise Exception('adb root failed')\n\nroot = '/sys/devices/system/cpu/cpu%d/cpufreq' %cpu\n\n# All devices we test on give a list of their available frequencies.\navailable_freqs = subprocess.check_output([ADB, 'shell',\n 'cat %s/scaling_available_frequencies' % root])\n\n# Check for message like '/system/bin/sh: file not found'\nif available_freqs and '/system/bin/sh' not in available_freqs:\n available_freqs = sorted(\n int(i) for i in available_freqs.strip().split())\nelse:\n raise Exception('Could not get list of available frequencies: %s' %\n available_freqs)\n\nmaxfreq = available_freqs[-1]\ntarget = int(round(maxfreq * target_percent))\nfreq = maxfreq\nfor f in reversed(available_freqs):\n if f <= target:\n freq = f\n break\n\nprint 'Setting frequency to %d' % freq\n\n# If scaling_max_freq is lower than our attempted setting, it won't take.\n# We must set min first, because if we try to set max to be less than min\n# (which sometimes happens after certain devices reboot) it returns a\n# perplexing permissions error.\nsubprocess.check_output([ADB, 'shell', 'echo 0 > '\n '%s/scaling_min_freq' % root])\nsubprocess.check_output([ADB, 'shell', 'echo %d > '\n '%s/scaling_max_freq' % (freq, root)])\nsubprocess.check_output([ADB, 'shell', 'echo %d > '\n '%s/scaling_setspeed' % (freq, root)])\ntime.sleep(5)\nactual_freq = subprocess.check_output([ADB, 'shell', 'cat '\n '%s/scaling_cur_freq' % root]).strip()\nif actual_freq != str(freq):\n raise Exception('(actual, expected) (%s, %d)'\n % (actual_freq, freq))\n",
Use full paths for adb Add logic to run on Nexus 5xs in Golo. Bug:788839 Change-Id: I12290d11a0b1b0f012ada216da3e5b2599979c5e Reviewed-on: https://skia-review.googlesource.com/81861 Commit-Queue: Kevin Lubick <kjlubick@google.com> Reviewed-by: Eric Boren <borenet@google.com>
2017-12-07 16:19:31 +00:00
"/usr/bin/adb.1.0.35",
Scale extra CPUs to powersave during nanobench. This scaling logic correctly accounts for some devices which have multiple CPUs. Previously, we were scaling the smaller of these CPUs, which likely had a negative impact on nanobench, given nanobench was single threaded and the CPUs weren't allowed to idle much (because we set the CPU). This CL sets those additional CPUs to powersave when we run nanobench and then correctly scales down the beefier CPU we want to run nanobench on. For DM, we just run it in ondemand mode, which will hopefully be "as fast as possible", but allow the CPU governor to scale down if overheating becomes a problem. Bug: skia:7378 notry=TRUE Change-Id: I45ca5d9fb32182233d1b2d094842c879f2b84da4 Reviewed-on: https://skia-review.googlesource.com/83240 Commit-Queue: Kevin Lubick <kjlubick@google.com> Reviewed-by: Mike Klein <mtklein@chromium.org>
2017-12-12 12:33:48 +00:00
"0.6",
Experiment with turning CPUs off during perf Perf was nice and flat after https://skia-review.googlesource.com/c/skia/+/83240 but there was a 4-5x slowdown on the benchmarks themselves, indicating that perhaps we were running nanobench on the LITTLE cores (now on powersave) instead of the big cores we thought (which were recently scaled down). This experiment will show us FOR CERTAIN that we are running on core 0 or 1 which is at .6 max frequency. We expect to see a speedup from the previous results. This is leading to turning of CPUs we don't need to make sure nanobench is running on the ones we expect. Bug: skia: NOTRY=true Change-Id: Ida65181e4d90e778e65e3f22d761288b9ade64f6 Reviewed-on: https://skia-review.googlesource.com/84201 Reviewed-by: Mike Klein <mtklein@chromium.org> Commit-Queue: Kevin Lubick <kjlubick@google.com>
2017-12-12 20:57:20 +00:00
"0"
Make cpu scaling more robust Some bots, especially the Nexus 7s, seemed to occasionally fail when setting the CPU frequency. I was unable to repro this behavior, so this is a shotgun approach. We add a 5 second delay between setting and checking, checking frequency using scaling_cur_freq instead of scaling_setspeed, set the min_freq as well as max_freq, and retry up to 3 times if setting cpu frequency fails. NOTRY=true Bug: skia: Change-Id: Id4d85d8d509c9dba8e3a0e06b5992f5adadf36d2 Reviewed-on: https://skia-review.googlesource.com/78140 Commit-Queue: Kevin Lubick <kjlubick@google.com> Reviewed-by: Eric Boren <borenet@google.com>
2017-11-30 13:35:29 +00:00
],
"env": {
"BUILDTYPE": "Debug",
"CHROME_HEADLESS": "1",
"PATH": "<PATH>:RECIPE_PACKAGE_REPO[depot_tools]",
"SKIA_OUT": "[START_DIR]/out"
},
"infra_step": true,
Experiment with turning CPUs off during perf Perf was nice and flat after https://skia-review.googlesource.com/c/skia/+/83240 but there was a 4-5x slowdown on the benchmarks themselves, indicating that perhaps we were running nanobench on the LITTLE cores (now on powersave) instead of the big cores we thought (which were recently scaled down). This experiment will show us FOR CERTAIN that we are running on core 0 or 1 which is at .6 max frequency. We expect to see a speedup from the previous results. This is leading to turning of CPUs we don't need to make sure nanobench is running on the ones we expect. Bug: skia: NOTRY=true Change-Id: Ida65181e4d90e778e65e3f22d761288b9ade64f6 Reviewed-on: https://skia-review.googlesource.com/84201 Reviewed-by: Mike Klein <mtklein@chromium.org> Commit-Queue: Kevin Lubick <kjlubick@google.com>
2017-12-12 20:57:20 +00:00
"name": "Scale CPU 0 to 0.600000 (attempt 2)",
Make cpu scaling more robust Some bots, especially the Nexus 7s, seemed to occasionally fail when setting the CPU frequency. I was unable to repro this behavior, so this is a shotgun approach. We add a 5 second delay between setting and checking, checking frequency using scaling_cur_freq instead of scaling_setspeed, set the min_freq as well as max_freq, and retry up to 3 times if setting cpu frequency fails. NOTRY=true Bug: skia: Change-Id: Id4d85d8d509c9dba8e3a0e06b5992f5adadf36d2 Reviewed-on: https://skia-review.googlesource.com/78140 Commit-Queue: Kevin Lubick <kjlubick@google.com> Reviewed-by: Eric Boren <borenet@google.com>
2017-11-30 13:35:29 +00:00
"timeout": 30,
"~followup_annotations": [
"step returned non-zero exit code: 1",
"@@@STEP_LOG_LINE@python.inline@@@@",
"@@@STEP_LOG_LINE@python.inline@import os@@@",
"@@@STEP_LOG_LINE@python.inline@import subprocess@@@",
"@@@STEP_LOG_LINE@python.inline@import sys@@@",
"@@@STEP_LOG_LINE@python.inline@import time@@@",
"@@@STEP_LOG_LINE@python.inline@ADB = sys.argv[1]@@@",
Scale extra CPUs to powersave during nanobench. This scaling logic correctly accounts for some devices which have multiple CPUs. Previously, we were scaling the smaller of these CPUs, which likely had a negative impact on nanobench, given nanobench was single threaded and the CPUs weren't allowed to idle much (because we set the CPU). This CL sets those additional CPUs to powersave when we run nanobench and then correctly scales down the beefier CPU we want to run nanobench on. For DM, we just run it in ondemand mode, which will hopefully be "as fast as possible", but allow the CPU governor to scale down if overheating becomes a problem. Bug: skia:7378 notry=TRUE Change-Id: I45ca5d9fb32182233d1b2d094842c879f2b84da4 Reviewed-on: https://skia-review.googlesource.com/83240 Commit-Queue: Kevin Lubick <kjlubick@google.com> Reviewed-by: Mike Klein <mtklein@chromium.org>
2017-12-12 12:33:48 +00:00
"@@@STEP_LOG_LINE@python.inline@target_percent = float(sys.argv[2])@@@",
"@@@STEP_LOG_LINE@python.inline@cpu = int(sys.argv[3])@@@",
Make cpu scaling more robust Some bots, especially the Nexus 7s, seemed to occasionally fail when setting the CPU frequency. I was unable to repro this behavior, so this is a shotgun approach. We add a 5 second delay between setting and checking, checking frequency using scaling_cur_freq instead of scaling_setspeed, set the min_freq as well as max_freq, and retry up to 3 times if setting cpu frequency fails. NOTRY=true Bug: skia: Change-Id: Id4d85d8d509c9dba8e3a0e06b5992f5adadf36d2 Reviewed-on: https://skia-review.googlesource.com/78140 Commit-Queue: Kevin Lubick <kjlubick@google.com> Reviewed-by: Eric Boren <borenet@google.com>
2017-11-30 13:35:29 +00:00
"@@@STEP_LOG_LINE@python.inline@log = subprocess.check_output([ADB, 'root'])@@@",
"@@@STEP_LOG_LINE@python.inline@# check for message like 'adbd cannot run as root in production builds'@@@",
Fix scaling issues for some devices (e.g. Pixel) Bug: skia: NOTRY=true Change-Id: Ia6a80ecb97b5d30197932ec7d340163624e315bd Reviewed-on: https://skia-review.googlesource.com/82682 Reviewed-by: Ravi Mistry <rmistry@google.com> Commit-Queue: Kevin Lubick <kjlubick@google.com>
2017-12-08 15:26:56 +00:00
"@@@STEP_LOG_LINE@python.inline@print log@@@",
Make cpu scaling more robust Some bots, especially the Nexus 7s, seemed to occasionally fail when setting the CPU frequency. I was unable to repro this behavior, so this is a shotgun approach. We add a 5 second delay between setting and checking, checking frequency using scaling_cur_freq instead of scaling_setspeed, set the min_freq as well as max_freq, and retry up to 3 times if setting cpu frequency fails. NOTRY=true Bug: skia: Change-Id: Id4d85d8d509c9dba8e3a0e06b5992f5adadf36d2 Reviewed-on: https://skia-review.googlesource.com/78140 Commit-Queue: Kevin Lubick <kjlubick@google.com> Reviewed-by: Eric Boren <borenet@google.com>
2017-11-30 13:35:29 +00:00
"@@@STEP_LOG_LINE@python.inline@if 'cannot' in log:@@@",
"@@@STEP_LOG_LINE@python.inline@ raise Exception('adb root failed')@@@",
"@@@STEP_LOG_LINE@python.inline@@@@",
Scale extra CPUs to powersave during nanobench. This scaling logic correctly accounts for some devices which have multiple CPUs. Previously, we were scaling the smaller of these CPUs, which likely had a negative impact on nanobench, given nanobench was single threaded and the CPUs weren't allowed to idle much (because we set the CPU). This CL sets those additional CPUs to powersave when we run nanobench and then correctly scales down the beefier CPU we want to run nanobench on. For DM, we just run it in ondemand mode, which will hopefully be "as fast as possible", but allow the CPU governor to scale down if overheating becomes a problem. Bug: skia:7378 notry=TRUE Change-Id: I45ca5d9fb32182233d1b2d094842c879f2b84da4 Reviewed-on: https://skia-review.googlesource.com/83240 Commit-Queue: Kevin Lubick <kjlubick@google.com> Reviewed-by: Mike Klein <mtklein@chromium.org>
2017-12-12 12:33:48 +00:00
"@@@STEP_LOG_LINE@python.inline@root = '/sys/devices/system/cpu/cpu%d/cpufreq' %cpu@@@",
"@@@STEP_LOG_LINE@python.inline@@@@",
"@@@STEP_LOG_LINE@python.inline@# All devices we test on give a list of their available frequencies.@@@",
"@@@STEP_LOG_LINE@python.inline@available_freqs = subprocess.check_output([ADB, 'shell',@@@",
"@@@STEP_LOG_LINE@python.inline@ 'cat %s/scaling_available_frequencies' % root])@@@",
"@@@STEP_LOG_LINE@python.inline@@@@",
"@@@STEP_LOG_LINE@python.inline@# Check for message like '/system/bin/sh: file not found'@@@",
"@@@STEP_LOG_LINE@python.inline@if available_freqs and '/system/bin/sh' not in available_freqs:@@@",
"@@@STEP_LOG_LINE@python.inline@ available_freqs = sorted(@@@",
"@@@STEP_LOG_LINE@python.inline@ int(i) for i in available_freqs.strip().split())@@@",
Make cpu scaling more robust Some bots, especially the Nexus 7s, seemed to occasionally fail when setting the CPU frequency. I was unable to repro this behavior, so this is a shotgun approach. We add a 5 second delay between setting and checking, checking frequency using scaling_cur_freq instead of scaling_setspeed, set the min_freq as well as max_freq, and retry up to 3 times if setting cpu frequency fails. NOTRY=true Bug: skia: Change-Id: Id4d85d8d509c9dba8e3a0e06b5992f5adadf36d2 Reviewed-on: https://skia-review.googlesource.com/78140 Commit-Queue: Kevin Lubick <kjlubick@google.com> Reviewed-by: Eric Boren <borenet@google.com>
2017-11-30 13:35:29 +00:00
"@@@STEP_LOG_LINE@python.inline@else:@@@",
Scale extra CPUs to powersave during nanobench. This scaling logic correctly accounts for some devices which have multiple CPUs. Previously, we were scaling the smaller of these CPUs, which likely had a negative impact on nanobench, given nanobench was single threaded and the CPUs weren't allowed to idle much (because we set the CPU). This CL sets those additional CPUs to powersave when we run nanobench and then correctly scales down the beefier CPU we want to run nanobench on. For DM, we just run it in ondemand mode, which will hopefully be "as fast as possible", but allow the CPU governor to scale down if overheating becomes a problem. Bug: skia:7378 notry=TRUE Change-Id: I45ca5d9fb32182233d1b2d094842c879f2b84da4 Reviewed-on: https://skia-review.googlesource.com/83240 Commit-Queue: Kevin Lubick <kjlubick@google.com> Reviewed-by: Mike Klein <mtklein@chromium.org>
2017-12-12 12:33:48 +00:00
"@@@STEP_LOG_LINE@python.inline@ raise Exception('Could not get list of available frequencies: %s' %@@@",
"@@@STEP_LOG_LINE@python.inline@ available_freqs)@@@",
Make cpu scaling more robust Some bots, especially the Nexus 7s, seemed to occasionally fail when setting the CPU frequency. I was unable to repro this behavior, so this is a shotgun approach. We add a 5 second delay between setting and checking, checking frequency using scaling_cur_freq instead of scaling_setspeed, set the min_freq as well as max_freq, and retry up to 3 times if setting cpu frequency fails. NOTRY=true Bug: skia: Change-Id: Id4d85d8d509c9dba8e3a0e06b5992f5adadf36d2 Reviewed-on: https://skia-review.googlesource.com/78140 Commit-Queue: Kevin Lubick <kjlubick@google.com> Reviewed-by: Eric Boren <borenet@google.com>
2017-11-30 13:35:29 +00:00
"@@@STEP_LOG_LINE@python.inline@@@@",
"@@@STEP_LOG_LINE@python.inline@maxfreq = available_freqs[-1]@@@",
"@@@STEP_LOG_LINE@python.inline@target = int(round(maxfreq * target_percent))@@@",
"@@@STEP_LOG_LINE@python.inline@freq = maxfreq@@@",
"@@@STEP_LOG_LINE@python.inline@for f in reversed(available_freqs):@@@",
"@@@STEP_LOG_LINE@python.inline@ if f <= target:@@@",
"@@@STEP_LOG_LINE@python.inline@ freq = f@@@",
"@@@STEP_LOG_LINE@python.inline@ break@@@",
"@@@STEP_LOG_LINE@python.inline@@@@",
"@@@STEP_LOG_LINE@python.inline@print 'Setting frequency to %d' % freq@@@",
"@@@STEP_LOG_LINE@python.inline@@@@",
"@@@STEP_LOG_LINE@python.inline@# If scaling_max_freq is lower than our attempted setting, it won't take.@@@",
Fix scaling issues for some devices (e.g. Pixel) Bug: skia: NOTRY=true Change-Id: Ia6a80ecb97b5d30197932ec7d340163624e315bd Reviewed-on: https://skia-review.googlesource.com/82682 Reviewed-by: Ravi Mistry <rmistry@google.com> Commit-Queue: Kevin Lubick <kjlubick@google.com>
2017-12-08 15:26:56 +00:00
"@@@STEP_LOG_LINE@python.inline@# We must set min first, because if we try to set max to be less than min@@@",
"@@@STEP_LOG_LINE@python.inline@# (which sometimes happens after certain devices reboot) it returns a@@@",
"@@@STEP_LOG_LINE@python.inline@# perplexing permissions error.@@@",
Don't set min frequency for Android devices Bug: skia:7378 NOTRY=true No-Tree-Checks: true Change-Id: Iaacbc8eadaa4da9832a53d4794a2214625bef9b6 Reviewed-on: https://skia-review.googlesource.com/80842 Commit-Queue: Kevin Lubick <kjlubick@google.com> Reviewed-by: Kevin Lubick <kjlubick@google.com>
2017-12-05 20:54:19 +00:00
"@@@STEP_LOG_LINE@python.inline@subprocess.check_output([ADB, 'shell', 'echo 0 > '@@@",
Scale extra CPUs to powersave during nanobench. This scaling logic correctly accounts for some devices which have multiple CPUs. Previously, we were scaling the smaller of these CPUs, which likely had a negative impact on nanobench, given nanobench was single threaded and the CPUs weren't allowed to idle much (because we set the CPU). This CL sets those additional CPUs to powersave when we run nanobench and then correctly scales down the beefier CPU we want to run nanobench on. For DM, we just run it in ondemand mode, which will hopefully be "as fast as possible", but allow the CPU governor to scale down if overheating becomes a problem. Bug: skia:7378 notry=TRUE Change-Id: I45ca5d9fb32182233d1b2d094842c879f2b84da4 Reviewed-on: https://skia-review.googlesource.com/83240 Commit-Queue: Kevin Lubick <kjlubick@google.com> Reviewed-by: Mike Klein <mtklein@chromium.org>
2017-12-12 12:33:48 +00:00
"@@@STEP_LOG_LINE@python.inline@ '%s/scaling_min_freq' % root])@@@",
Make cpu scaling more robust Some bots, especially the Nexus 7s, seemed to occasionally fail when setting the CPU frequency. I was unable to repro this behavior, so this is a shotgun approach. We add a 5 second delay between setting and checking, checking frequency using scaling_cur_freq instead of scaling_setspeed, set the min_freq as well as max_freq, and retry up to 3 times if setting cpu frequency fails. NOTRY=true Bug: skia: Change-Id: Id4d85d8d509c9dba8e3a0e06b5992f5adadf36d2 Reviewed-on: https://skia-review.googlesource.com/78140 Commit-Queue: Kevin Lubick <kjlubick@google.com> Reviewed-by: Eric Boren <borenet@google.com>
2017-11-30 13:35:29 +00:00
"@@@STEP_LOG_LINE@python.inline@subprocess.check_output([ADB, 'shell', 'echo %d > '@@@",
Scale extra CPUs to powersave during nanobench. This scaling logic correctly accounts for some devices which have multiple CPUs. Previously, we were scaling the smaller of these CPUs, which likely had a negative impact on nanobench, given nanobench was single threaded and the CPUs weren't allowed to idle much (because we set the CPU). This CL sets those additional CPUs to powersave when we run nanobench and then correctly scales down the beefier CPU we want to run nanobench on. For DM, we just run it in ondemand mode, which will hopefully be "as fast as possible", but allow the CPU governor to scale down if overheating becomes a problem. Bug: skia:7378 notry=TRUE Change-Id: I45ca5d9fb32182233d1b2d094842c879f2b84da4 Reviewed-on: https://skia-review.googlesource.com/83240 Commit-Queue: Kevin Lubick <kjlubick@google.com> Reviewed-by: Mike Klein <mtklein@chromium.org>
2017-12-12 12:33:48 +00:00
"@@@STEP_LOG_LINE@python.inline@ '%s/scaling_max_freq' % (freq, root)])@@@",
Fix scaling issues for some devices (e.g. Pixel) Bug: skia: NOTRY=true Change-Id: Ia6a80ecb97b5d30197932ec7d340163624e315bd Reviewed-on: https://skia-review.googlesource.com/82682 Reviewed-by: Ravi Mistry <rmistry@google.com> Commit-Queue: Kevin Lubick <kjlubick@google.com>
2017-12-08 15:26:56 +00:00
"@@@STEP_LOG_LINE@python.inline@subprocess.check_output([ADB, 'shell', 'echo %d > '@@@",
Scale extra CPUs to powersave during nanobench. This scaling logic correctly accounts for some devices which have multiple CPUs. Previously, we were scaling the smaller of these CPUs, which likely had a negative impact on nanobench, given nanobench was single threaded and the CPUs weren't allowed to idle much (because we set the CPU). This CL sets those additional CPUs to powersave when we run nanobench and then correctly scales down the beefier CPU we want to run nanobench on. For DM, we just run it in ondemand mode, which will hopefully be "as fast as possible", but allow the CPU governor to scale down if overheating becomes a problem. Bug: skia:7378 notry=TRUE Change-Id: I45ca5d9fb32182233d1b2d094842c879f2b84da4 Reviewed-on: https://skia-review.googlesource.com/83240 Commit-Queue: Kevin Lubick <kjlubick@google.com> Reviewed-by: Mike Klein <mtklein@chromium.org>
2017-12-12 12:33:48 +00:00
"@@@STEP_LOG_LINE@python.inline@ '%s/scaling_setspeed' % (freq, root)])@@@",
Make cpu scaling more robust Some bots, especially the Nexus 7s, seemed to occasionally fail when setting the CPU frequency. I was unable to repro this behavior, so this is a shotgun approach. We add a 5 second delay between setting and checking, checking frequency using scaling_cur_freq instead of scaling_setspeed, set the min_freq as well as max_freq, and retry up to 3 times if setting cpu frequency fails. NOTRY=true Bug: skia: Change-Id: Id4d85d8d509c9dba8e3a0e06b5992f5adadf36d2 Reviewed-on: https://skia-review.googlesource.com/78140 Commit-Queue: Kevin Lubick <kjlubick@google.com> Reviewed-by: Eric Boren <borenet@google.com>
2017-11-30 13:35:29 +00:00
"@@@STEP_LOG_LINE@python.inline@time.sleep(5)@@@",
"@@@STEP_LOG_LINE@python.inline@actual_freq = subprocess.check_output([ADB, 'shell', 'cat '@@@",
Scale extra CPUs to powersave during nanobench. This scaling logic correctly accounts for some devices which have multiple CPUs. Previously, we were scaling the smaller of these CPUs, which likely had a negative impact on nanobench, given nanobench was single threaded and the CPUs weren't allowed to idle much (because we set the CPU). This CL sets those additional CPUs to powersave when we run nanobench and then correctly scales down the beefier CPU we want to run nanobench on. For DM, we just run it in ondemand mode, which will hopefully be "as fast as possible", but allow the CPU governor to scale down if overheating becomes a problem. Bug: skia:7378 notry=TRUE Change-Id: I45ca5d9fb32182233d1b2d094842c879f2b84da4 Reviewed-on: https://skia-review.googlesource.com/83240 Commit-Queue: Kevin Lubick <kjlubick@google.com> Reviewed-by: Mike Klein <mtklein@chromium.org>
2017-12-12 12:33:48 +00:00
"@@@STEP_LOG_LINE@python.inline@ '%s/scaling_cur_freq' % root]).strip()@@@",
Make cpu scaling more robust Some bots, especially the Nexus 7s, seemed to occasionally fail when setting the CPU frequency. I was unable to repro this behavior, so this is a shotgun approach. We add a 5 second delay between setting and checking, checking frequency using scaling_cur_freq instead of scaling_setspeed, set the min_freq as well as max_freq, and retry up to 3 times if setting cpu frequency fails. NOTRY=true Bug: skia: Change-Id: Id4d85d8d509c9dba8e3a0e06b5992f5adadf36d2 Reviewed-on: https://skia-review.googlesource.com/78140 Commit-Queue: Kevin Lubick <kjlubick@google.com> Reviewed-by: Eric Boren <borenet@google.com>
2017-11-30 13:35:29 +00:00
"@@@STEP_LOG_LINE@python.inline@if actual_freq != str(freq):@@@",
"@@@STEP_LOG_LINE@python.inline@ raise Exception('(actual, expected) (%s, %d)'@@@",
"@@@STEP_LOG_LINE@python.inline@ % (actual_freq, freq))@@@",
"@@@STEP_LOG_END@python.inline@@@",
"@@@STEP_EXCEPTION@@@"
]
},
{
"cmd": [
"python",
"-u",
Scale extra CPUs to powersave during nanobench. This scaling logic correctly accounts for some devices which have multiple CPUs. Previously, we were scaling the smaller of these CPUs, which likely had a negative impact on nanobench, given nanobench was single threaded and the CPUs weren't allowed to idle much (because we set the CPU). This CL sets those additional CPUs to powersave when we run nanobench and then correctly scales down the beefier CPU we want to run nanobench on. For DM, we just run it in ondemand mode, which will hopefully be "as fast as possible", but allow the CPU governor to scale down if overheating becomes a problem. Bug: skia:7378 notry=TRUE Change-Id: I45ca5d9fb32182233d1b2d094842c879f2b84da4 Reviewed-on: https://skia-review.googlesource.com/83240 Commit-Queue: Kevin Lubick <kjlubick@google.com> Reviewed-by: Mike Klein <mtklein@chromium.org>
2017-12-12 12:33:48 +00:00
"\nimport os\nimport subprocess\nimport sys\nimport time\nADB = sys.argv[1]\ntarget_percent = float(sys.argv[2])\ncpu = int(sys.argv[3])\nlog = subprocess.check_output([ADB, 'root'])\n# check for message like 'adbd cannot run as root in production builds'\nprint log\nif 'cannot' in log:\n raise Exception('adb root failed')\n\nroot = '/sys/devices/system/cpu/cpu%d/cpufreq' %cpu\n\n# All devices we test on give a list of their available frequencies.\navailable_freqs = subprocess.check_output([ADB, 'shell',\n 'cat %s/scaling_available_frequencies' % root])\n\n# Check for message like '/system/bin/sh: file not found'\nif available_freqs and '/system/bin/sh' not in available_freqs:\n available_freqs = sorted(\n int(i) for i in available_freqs.strip().split())\nelse:\n raise Exception('Could not get list of available frequencies: %s' %\n available_freqs)\n\nmaxfreq = available_freqs[-1]\ntarget = int(round(maxfreq * target_percent))\nfreq = maxfreq\nfor f in reversed(available_freqs):\n if f <= target:\n freq = f\n break\n\nprint 'Setting frequency to %d' % freq\n\n# If scaling_max_freq is lower than our attempted setting, it won't take.\n# We must set min first, because if we try to set max to be less than min\n# (which sometimes happens after certain devices reboot) it returns a\n# perplexing permissions error.\nsubprocess.check_output([ADB, 'shell', 'echo 0 > '\n '%s/scaling_min_freq' % root])\nsubprocess.check_output([ADB, 'shell', 'echo %d > '\n '%s/scaling_max_freq' % (freq, root)])\nsubprocess.check_output([ADB, 'shell', 'echo %d > '\n '%s/scaling_setspeed' % (freq, root)])\ntime.sleep(5)\nactual_freq = subprocess.check_output([ADB, 'shell', 'cat '\n '%s/scaling_cur_freq' % root]).strip()\nif actual_freq != str(freq):\n raise Exception('(actual, expected) (%s, %d)'\n % (actual_freq, freq))\n",
Use full paths for adb Add logic to run on Nexus 5xs in Golo. Bug:788839 Change-Id: I12290d11a0b1b0f012ada216da3e5b2599979c5e Reviewed-on: https://skia-review.googlesource.com/81861 Commit-Queue: Kevin Lubick <kjlubick@google.com> Reviewed-by: Eric Boren <borenet@google.com>
2017-12-07 16:19:31 +00:00
"/usr/bin/adb.1.0.35",
Scale extra CPUs to powersave during nanobench. This scaling logic correctly accounts for some devices which have multiple CPUs. Previously, we were scaling the smaller of these CPUs, which likely had a negative impact on nanobench, given nanobench was single threaded and the CPUs weren't allowed to idle much (because we set the CPU). This CL sets those additional CPUs to powersave when we run nanobench and then correctly scales down the beefier CPU we want to run nanobench on. For DM, we just run it in ondemand mode, which will hopefully be "as fast as possible", but allow the CPU governor to scale down if overheating becomes a problem. Bug: skia:7378 notry=TRUE Change-Id: I45ca5d9fb32182233d1b2d094842c879f2b84da4 Reviewed-on: https://skia-review.googlesource.com/83240 Commit-Queue: Kevin Lubick <kjlubick@google.com> Reviewed-by: Mike Klein <mtklein@chromium.org>
2017-12-12 12:33:48 +00:00
"0.6",
Experiment with turning CPUs off during perf Perf was nice and flat after https://skia-review.googlesource.com/c/skia/+/83240 but there was a 4-5x slowdown on the benchmarks themselves, indicating that perhaps we were running nanobench on the LITTLE cores (now on powersave) instead of the big cores we thought (which were recently scaled down). This experiment will show us FOR CERTAIN that we are running on core 0 or 1 which is at .6 max frequency. We expect to see a speedup from the previous results. This is leading to turning of CPUs we don't need to make sure nanobench is running on the ones we expect. Bug: skia: NOTRY=true Change-Id: Ida65181e4d90e778e65e3f22d761288b9ade64f6 Reviewed-on: https://skia-review.googlesource.com/84201 Reviewed-by: Mike Klein <mtklein@chromium.org> Commit-Queue: Kevin Lubick <kjlubick@google.com>
2017-12-12 20:57:20 +00:00
"0"
Make cpu scaling more robust Some bots, especially the Nexus 7s, seemed to occasionally fail when setting the CPU frequency. I was unable to repro this behavior, so this is a shotgun approach. We add a 5 second delay between setting and checking, checking frequency using scaling_cur_freq instead of scaling_setspeed, set the min_freq as well as max_freq, and retry up to 3 times if setting cpu frequency fails. NOTRY=true Bug: skia: Change-Id: Id4d85d8d509c9dba8e3a0e06b5992f5adadf36d2 Reviewed-on: https://skia-review.googlesource.com/78140 Commit-Queue: Kevin Lubick <kjlubick@google.com> Reviewed-by: Eric Boren <borenet@google.com>
2017-11-30 13:35:29 +00:00
],
"env": {
"BUILDTYPE": "Debug",
"CHROME_HEADLESS": "1",
"PATH": "<PATH>:RECIPE_PACKAGE_REPO[depot_tools]",
"SKIA_OUT": "[START_DIR]/out"
},
"infra_step": true,
Experiment with turning CPUs off during perf Perf was nice and flat after https://skia-review.googlesource.com/c/skia/+/83240 but there was a 4-5x slowdown on the benchmarks themselves, indicating that perhaps we were running nanobench on the LITTLE cores (now on powersave) instead of the big cores we thought (which were recently scaled down). This experiment will show us FOR CERTAIN that we are running on core 0 or 1 which is at .6 max frequency. We expect to see a speedup from the previous results. This is leading to turning of CPUs we don't need to make sure nanobench is running on the ones we expect. Bug: skia: NOTRY=true Change-Id: Ida65181e4d90e778e65e3f22d761288b9ade64f6 Reviewed-on: https://skia-review.googlesource.com/84201 Reviewed-by: Mike Klein <mtklein@chromium.org> Commit-Queue: Kevin Lubick <kjlubick@google.com>
2017-12-12 20:57:20 +00:00
"name": "Scale CPU 0 to 0.600000 (attempt 3)",
Make cpu scaling more robust Some bots, especially the Nexus 7s, seemed to occasionally fail when setting the CPU frequency. I was unable to repro this behavior, so this is a shotgun approach. We add a 5 second delay between setting and checking, checking frequency using scaling_cur_freq instead of scaling_setspeed, set the min_freq as well as max_freq, and retry up to 3 times if setting cpu frequency fails. NOTRY=true Bug: skia: Change-Id: Id4d85d8d509c9dba8e3a0e06b5992f5adadf36d2 Reviewed-on: https://skia-review.googlesource.com/78140 Commit-Queue: Kevin Lubick <kjlubick@google.com> Reviewed-by: Eric Boren <borenet@google.com>
2017-11-30 13:35:29 +00:00
"timeout": 30,
"~followup_annotations": [
"step returned non-zero exit code: 1",
"@@@STEP_LOG_LINE@python.inline@@@@",
"@@@STEP_LOG_LINE@python.inline@import os@@@",
"@@@STEP_LOG_LINE@python.inline@import subprocess@@@",
"@@@STEP_LOG_LINE@python.inline@import sys@@@",
"@@@STEP_LOG_LINE@python.inline@import time@@@",
"@@@STEP_LOG_LINE@python.inline@ADB = sys.argv[1]@@@",
Scale extra CPUs to powersave during nanobench. This scaling logic correctly accounts for some devices which have multiple CPUs. Previously, we were scaling the smaller of these CPUs, which likely had a negative impact on nanobench, given nanobench was single threaded and the CPUs weren't allowed to idle much (because we set the CPU). This CL sets those additional CPUs to powersave when we run nanobench and then correctly scales down the beefier CPU we want to run nanobench on. For DM, we just run it in ondemand mode, which will hopefully be "as fast as possible", but allow the CPU governor to scale down if overheating becomes a problem. Bug: skia:7378 notry=TRUE Change-Id: I45ca5d9fb32182233d1b2d094842c879f2b84da4 Reviewed-on: https://skia-review.googlesource.com/83240 Commit-Queue: Kevin Lubick <kjlubick@google.com> Reviewed-by: Mike Klein <mtklein@chromium.org>
2017-12-12 12:33:48 +00:00
"@@@STEP_LOG_LINE@python.inline@target_percent = float(sys.argv[2])@@@",
"@@@STEP_LOG_LINE@python.inline@cpu = int(sys.argv[3])@@@",
Make cpu scaling more robust Some bots, especially the Nexus 7s, seemed to occasionally fail when setting the CPU frequency. I was unable to repro this behavior, so this is a shotgun approach. We add a 5 second delay between setting and checking, checking frequency using scaling_cur_freq instead of scaling_setspeed, set the min_freq as well as max_freq, and retry up to 3 times if setting cpu frequency fails. NOTRY=true Bug: skia: Change-Id: Id4d85d8d509c9dba8e3a0e06b5992f5adadf36d2 Reviewed-on: https://skia-review.googlesource.com/78140 Commit-Queue: Kevin Lubick <kjlubick@google.com> Reviewed-by: Eric Boren <borenet@google.com>
2017-11-30 13:35:29 +00:00
"@@@STEP_LOG_LINE@python.inline@log = subprocess.check_output([ADB, 'root'])@@@",
"@@@STEP_LOG_LINE@python.inline@# check for message like 'adbd cannot run as root in production builds'@@@",
Fix scaling issues for some devices (e.g. Pixel) Bug: skia: NOTRY=true Change-Id: Ia6a80ecb97b5d30197932ec7d340163624e315bd Reviewed-on: https://skia-review.googlesource.com/82682 Reviewed-by: Ravi Mistry <rmistry@google.com> Commit-Queue: Kevin Lubick <kjlubick@google.com>
2017-12-08 15:26:56 +00:00
"@@@STEP_LOG_LINE@python.inline@print log@@@",
Make cpu scaling more robust Some bots, especially the Nexus 7s, seemed to occasionally fail when setting the CPU frequency. I was unable to repro this behavior, so this is a shotgun approach. We add a 5 second delay between setting and checking, checking frequency using scaling_cur_freq instead of scaling_setspeed, set the min_freq as well as max_freq, and retry up to 3 times if setting cpu frequency fails. NOTRY=true Bug: skia: Change-Id: Id4d85d8d509c9dba8e3a0e06b5992f5adadf36d2 Reviewed-on: https://skia-review.googlesource.com/78140 Commit-Queue: Kevin Lubick <kjlubick@google.com> Reviewed-by: Eric Boren <borenet@google.com>
2017-11-30 13:35:29 +00:00
"@@@STEP_LOG_LINE@python.inline@if 'cannot' in log:@@@",
"@@@STEP_LOG_LINE@python.inline@ raise Exception('adb root failed')@@@",
"@@@STEP_LOG_LINE@python.inline@@@@",
Scale extra CPUs to powersave during nanobench. This scaling logic correctly accounts for some devices which have multiple CPUs. Previously, we were scaling the smaller of these CPUs, which likely had a negative impact on nanobench, given nanobench was single threaded and the CPUs weren't allowed to idle much (because we set the CPU). This CL sets those additional CPUs to powersave when we run nanobench and then correctly scales down the beefier CPU we want to run nanobench on. For DM, we just run it in ondemand mode, which will hopefully be "as fast as possible", but allow the CPU governor to scale down if overheating becomes a problem. Bug: skia:7378 notry=TRUE Change-Id: I45ca5d9fb32182233d1b2d094842c879f2b84da4 Reviewed-on: https://skia-review.googlesource.com/83240 Commit-Queue: Kevin Lubick <kjlubick@google.com> Reviewed-by: Mike Klein <mtklein@chromium.org>
2017-12-12 12:33:48 +00:00
"@@@STEP_LOG_LINE@python.inline@root = '/sys/devices/system/cpu/cpu%d/cpufreq' %cpu@@@",
"@@@STEP_LOG_LINE@python.inline@@@@",
"@@@STEP_LOG_LINE@python.inline@# All devices we test on give a list of their available frequencies.@@@",
"@@@STEP_LOG_LINE@python.inline@available_freqs = subprocess.check_output([ADB, 'shell',@@@",
"@@@STEP_LOG_LINE@python.inline@ 'cat %s/scaling_available_frequencies' % root])@@@",
"@@@STEP_LOG_LINE@python.inline@@@@",
"@@@STEP_LOG_LINE@python.inline@# Check for message like '/system/bin/sh: file not found'@@@",
"@@@STEP_LOG_LINE@python.inline@if available_freqs and '/system/bin/sh' not in available_freqs:@@@",
"@@@STEP_LOG_LINE@python.inline@ available_freqs = sorted(@@@",
"@@@STEP_LOG_LINE@python.inline@ int(i) for i in available_freqs.strip().split())@@@",
Make cpu scaling more robust Some bots, especially the Nexus 7s, seemed to occasionally fail when setting the CPU frequency. I was unable to repro this behavior, so this is a shotgun approach. We add a 5 second delay between setting and checking, checking frequency using scaling_cur_freq instead of scaling_setspeed, set the min_freq as well as max_freq, and retry up to 3 times if setting cpu frequency fails. NOTRY=true Bug: skia: Change-Id: Id4d85d8d509c9dba8e3a0e06b5992f5adadf36d2 Reviewed-on: https://skia-review.googlesource.com/78140 Commit-Queue: Kevin Lubick <kjlubick@google.com> Reviewed-by: Eric Boren <borenet@google.com>
2017-11-30 13:35:29 +00:00
"@@@STEP_LOG_LINE@python.inline@else:@@@",
Scale extra CPUs to powersave during nanobench. This scaling logic correctly accounts for some devices which have multiple CPUs. Previously, we were scaling the smaller of these CPUs, which likely had a negative impact on nanobench, given nanobench was single threaded and the CPUs weren't allowed to idle much (because we set the CPU). This CL sets those additional CPUs to powersave when we run nanobench and then correctly scales down the beefier CPU we want to run nanobench on. For DM, we just run it in ondemand mode, which will hopefully be "as fast as possible", but allow the CPU governor to scale down if overheating becomes a problem. Bug: skia:7378 notry=TRUE Change-Id: I45ca5d9fb32182233d1b2d094842c879f2b84da4 Reviewed-on: https://skia-review.googlesource.com/83240 Commit-Queue: Kevin Lubick <kjlubick@google.com> Reviewed-by: Mike Klein <mtklein@chromium.org>
2017-12-12 12:33:48 +00:00
"@@@STEP_LOG_LINE@python.inline@ raise Exception('Could not get list of available frequencies: %s' %@@@",
"@@@STEP_LOG_LINE@python.inline@ available_freqs)@@@",
Make cpu scaling more robust Some bots, especially the Nexus 7s, seemed to occasionally fail when setting the CPU frequency. I was unable to repro this behavior, so this is a shotgun approach. We add a 5 second delay between setting and checking, checking frequency using scaling_cur_freq instead of scaling_setspeed, set the min_freq as well as max_freq, and retry up to 3 times if setting cpu frequency fails. NOTRY=true Bug: skia: Change-Id: Id4d85d8d509c9dba8e3a0e06b5992f5adadf36d2 Reviewed-on: https://skia-review.googlesource.com/78140 Commit-Queue: Kevin Lubick <kjlubick@google.com> Reviewed-by: Eric Boren <borenet@google.com>
2017-11-30 13:35:29 +00:00
"@@@STEP_LOG_LINE@python.inline@@@@",
"@@@STEP_LOG_LINE@python.inline@maxfreq = available_freqs[-1]@@@",
"@@@STEP_LOG_LINE@python.inline@target = int(round(maxfreq * target_percent))@@@",
"@@@STEP_LOG_LINE@python.inline@freq = maxfreq@@@",
"@@@STEP_LOG_LINE@python.inline@for f in reversed(available_freqs):@@@",
"@@@STEP_LOG_LINE@python.inline@ if f <= target:@@@",
"@@@STEP_LOG_LINE@python.inline@ freq = f@@@",
"@@@STEP_LOG_LINE@python.inline@ break@@@",
"@@@STEP_LOG_LINE@python.inline@@@@",
"@@@STEP_LOG_LINE@python.inline@print 'Setting frequency to %d' % freq@@@",
"@@@STEP_LOG_LINE@python.inline@@@@",
"@@@STEP_LOG_LINE@python.inline@# If scaling_max_freq is lower than our attempted setting, it won't take.@@@",
Fix scaling issues for some devices (e.g. Pixel) Bug: skia: NOTRY=true Change-Id: Ia6a80ecb97b5d30197932ec7d340163624e315bd Reviewed-on: https://skia-review.googlesource.com/82682 Reviewed-by: Ravi Mistry <rmistry@google.com> Commit-Queue: Kevin Lubick <kjlubick@google.com>
2017-12-08 15:26:56 +00:00
"@@@STEP_LOG_LINE@python.inline@# We must set min first, because if we try to set max to be less than min@@@",
"@@@STEP_LOG_LINE@python.inline@# (which sometimes happens after certain devices reboot) it returns a@@@",
"@@@STEP_LOG_LINE@python.inline@# perplexing permissions error.@@@",
Don't set min frequency for Android devices Bug: skia:7378 NOTRY=true No-Tree-Checks: true Change-Id: Iaacbc8eadaa4da9832a53d4794a2214625bef9b6 Reviewed-on: https://skia-review.googlesource.com/80842 Commit-Queue: Kevin Lubick <kjlubick@google.com> Reviewed-by: Kevin Lubick <kjlubick@google.com>
2017-12-05 20:54:19 +00:00
"@@@STEP_LOG_LINE@python.inline@subprocess.check_output([ADB, 'shell', 'echo 0 > '@@@",
Scale extra CPUs to powersave during nanobench. This scaling logic correctly accounts for some devices which have multiple CPUs. Previously, we were scaling the smaller of these CPUs, which likely had a negative impact on nanobench, given nanobench was single threaded and the CPUs weren't allowed to idle much (because we set the CPU). This CL sets those additional CPUs to powersave when we run nanobench and then correctly scales down the beefier CPU we want to run nanobench on. For DM, we just run it in ondemand mode, which will hopefully be "as fast as possible", but allow the CPU governor to scale down if overheating becomes a problem. Bug: skia:7378 notry=TRUE Change-Id: I45ca5d9fb32182233d1b2d094842c879f2b84da4 Reviewed-on: https://skia-review.googlesource.com/83240 Commit-Queue: Kevin Lubick <kjlubick@google.com> Reviewed-by: Mike Klein <mtklein@chromium.org>
2017-12-12 12:33:48 +00:00
"@@@STEP_LOG_LINE@python.inline@ '%s/scaling_min_freq' % root])@@@",
Make cpu scaling more robust Some bots, especially the Nexus 7s, seemed to occasionally fail when setting the CPU frequency. I was unable to repro this behavior, so this is a shotgun approach. We add a 5 second delay between setting and checking, checking frequency using scaling_cur_freq instead of scaling_setspeed, set the min_freq as well as max_freq, and retry up to 3 times if setting cpu frequency fails. NOTRY=true Bug: skia: Change-Id: Id4d85d8d509c9dba8e3a0e06b5992f5adadf36d2 Reviewed-on: https://skia-review.googlesource.com/78140 Commit-Queue: Kevin Lubick <kjlubick@google.com> Reviewed-by: Eric Boren <borenet@google.com>
2017-11-30 13:35:29 +00:00
"@@@STEP_LOG_LINE@python.inline@subprocess.check_output([ADB, 'shell', 'echo %d > '@@@",
Scale extra CPUs to powersave during nanobench. This scaling logic correctly accounts for some devices which have multiple CPUs. Previously, we were scaling the smaller of these CPUs, which likely had a negative impact on nanobench, given nanobench was single threaded and the CPUs weren't allowed to idle much (because we set the CPU). This CL sets those additional CPUs to powersave when we run nanobench and then correctly scales down the beefier CPU we want to run nanobench on. For DM, we just run it in ondemand mode, which will hopefully be "as fast as possible", but allow the CPU governor to scale down if overheating becomes a problem. Bug: skia:7378 notry=TRUE Change-Id: I45ca5d9fb32182233d1b2d094842c879f2b84da4 Reviewed-on: https://skia-review.googlesource.com/83240 Commit-Queue: Kevin Lubick <kjlubick@google.com> Reviewed-by: Mike Klein <mtklein@chromium.org>
2017-12-12 12:33:48 +00:00
"@@@STEP_LOG_LINE@python.inline@ '%s/scaling_max_freq' % (freq, root)])@@@",
Fix scaling issues for some devices (e.g. Pixel) Bug: skia: NOTRY=true Change-Id: Ia6a80ecb97b5d30197932ec7d340163624e315bd Reviewed-on: https://skia-review.googlesource.com/82682 Reviewed-by: Ravi Mistry <rmistry@google.com> Commit-Queue: Kevin Lubick <kjlubick@google.com>
2017-12-08 15:26:56 +00:00
"@@@STEP_LOG_LINE@python.inline@subprocess.check_output([ADB, 'shell', 'echo %d > '@@@",
Scale extra CPUs to powersave during nanobench. This scaling logic correctly accounts for some devices which have multiple CPUs. Previously, we were scaling the smaller of these CPUs, which likely had a negative impact on nanobench, given nanobench was single threaded and the CPUs weren't allowed to idle much (because we set the CPU). This CL sets those additional CPUs to powersave when we run nanobench and then correctly scales down the beefier CPU we want to run nanobench on. For DM, we just run it in ondemand mode, which will hopefully be "as fast as possible", but allow the CPU governor to scale down if overheating becomes a problem. Bug: skia:7378 notry=TRUE Change-Id: I45ca5d9fb32182233d1b2d094842c879f2b84da4 Reviewed-on: https://skia-review.googlesource.com/83240 Commit-Queue: Kevin Lubick <kjlubick@google.com> Reviewed-by: Mike Klein <mtklein@chromium.org>
2017-12-12 12:33:48 +00:00
"@@@STEP_LOG_LINE@python.inline@ '%s/scaling_setspeed' % (freq, root)])@@@",
Make cpu scaling more robust Some bots, especially the Nexus 7s, seemed to occasionally fail when setting the CPU frequency. I was unable to repro this behavior, so this is a shotgun approach. We add a 5 second delay between setting and checking, checking frequency using scaling_cur_freq instead of scaling_setspeed, set the min_freq as well as max_freq, and retry up to 3 times if setting cpu frequency fails. NOTRY=true Bug: skia: Change-Id: Id4d85d8d509c9dba8e3a0e06b5992f5adadf36d2 Reviewed-on: https://skia-review.googlesource.com/78140 Commit-Queue: Kevin Lubick <kjlubick@google.com> Reviewed-by: Eric Boren <borenet@google.com>
2017-11-30 13:35:29 +00:00
"@@@STEP_LOG_LINE@python.inline@time.sleep(5)@@@",
Make CPU scaling code cleaner This also makes *sure* the CPU frequency we set the device to actually "takes". Along the way, I learned if scaling_max_freq is < the frequency we set, the scaling_max_freq will be used instead, which was happening to the PixelCs and AndroidOnes. As a result, this may make those two Test- configs faster. Bug: skia: Change-Id: I10c98d37e296a19e1cf67bfe7269bb59cdd912d5 Reviewed-on: https://skia-review.googlesource.com/74360 Commit-Queue: Kevin Lubick <kjlubick@google.com> Reviewed-by: Ben Wagner <benjaminwagner@google.com> Reviewed-by: Eric Boren <borenet@google.com>
2017-11-21 21:47:16 +00:00
"@@@STEP_LOG_LINE@python.inline@actual_freq = subprocess.check_output([ADB, 'shell', 'cat '@@@",
Scale extra CPUs to powersave during nanobench. This scaling logic correctly accounts for some devices which have multiple CPUs. Previously, we were scaling the smaller of these CPUs, which likely had a negative impact on nanobench, given nanobench was single threaded and the CPUs weren't allowed to idle much (because we set the CPU). This CL sets those additional CPUs to powersave when we run nanobench and then correctly scales down the beefier CPU we want to run nanobench on. For DM, we just run it in ondemand mode, which will hopefully be "as fast as possible", but allow the CPU governor to scale down if overheating becomes a problem. Bug: skia:7378 notry=TRUE Change-Id: I45ca5d9fb32182233d1b2d094842c879f2b84da4 Reviewed-on: https://skia-review.googlesource.com/83240 Commit-Queue: Kevin Lubick <kjlubick@google.com> Reviewed-by: Mike Klein <mtklein@chromium.org>
2017-12-12 12:33:48 +00:00
"@@@STEP_LOG_LINE@python.inline@ '%s/scaling_cur_freq' % root]).strip()@@@",
Make CPU scaling code cleaner This also makes *sure* the CPU frequency we set the device to actually "takes". Along the way, I learned if scaling_max_freq is < the frequency we set, the scaling_max_freq will be used instead, which was happening to the PixelCs and AndroidOnes. As a result, this may make those two Test- configs faster. Bug: skia: Change-Id: I10c98d37e296a19e1cf67bfe7269bb59cdd912d5 Reviewed-on: https://skia-review.googlesource.com/74360 Commit-Queue: Kevin Lubick <kjlubick@google.com> Reviewed-by: Ben Wagner <benjaminwagner@google.com> Reviewed-by: Eric Boren <borenet@google.com>
2017-11-21 21:47:16 +00:00
"@@@STEP_LOG_LINE@python.inline@if actual_freq != str(freq):@@@",
"@@@STEP_LOG_LINE@python.inline@ raise Exception('(actual, expected) (%s, %d)'@@@",
"@@@STEP_LOG_LINE@python.inline@ % (actual_freq, freq))@@@",
"@@@STEP_LOG_END@python.inline@@@",
"@@@STEP_EXCEPTION@@@"
]
},
{
"cmd": [
"python",
"-u",
Use full paths for adb Add logic to run on Nexus 5xs in Golo. Bug:788839 Change-Id: I12290d11a0b1b0f012ada216da3e5b2599979c5e Reviewed-on: https://skia-review.googlesource.com/81861 Commit-Queue: Kevin Lubick <kjlubick@google.com> Reviewed-by: Eric Boren <borenet@google.com>
2017-12-07 16:19:31 +00:00
"\nimport os\nimport subprocess\nimport sys\nout = sys.argv[1]\nlog = subprocess.check_output(['/usr/bin/adb.1.0.35', 'logcat', '-d'])\nfor line in log.split('\\n'):\n tokens = line.split()\n if len(tokens) == 11 and tokens[-7] == 'F' and tokens[-3] == 'pc':\n addr, path = tokens[-2:]\n local = os.path.join(out, os.path.basename(path))\n if os.path.exists(local):\n sym = subprocess.check_output(['addr2line', '-Cfpe', local, addr])\n line = line.replace(addr, addr + ' ' + sym.strip())\n print line\n",
Make CPU scaling code cleaner This also makes *sure* the CPU frequency we set the device to actually "takes". Along the way, I learned if scaling_max_freq is < the frequency we set, the scaling_max_freq will be used instead, which was happening to the PixelCs and AndroidOnes. As a result, this may make those two Test- configs faster. Bug: skia: Change-Id: I10c98d37e296a19e1cf67bfe7269bb59cdd912d5 Reviewed-on: https://skia-review.googlesource.com/74360 Commit-Queue: Kevin Lubick <kjlubick@google.com> Reviewed-by: Ben Wagner <benjaminwagner@google.com> Reviewed-by: Eric Boren <borenet@google.com>
2017-11-21 21:47:16 +00:00
"[START_DIR]/out/Debug"
],
"env": {
"BUILDTYPE": "Debug",
"CHROME_HEADLESS": "1",
"PATH": "<PATH>:RECIPE_PACKAGE_REPO[depot_tools]",
"SKIA_OUT": "[START_DIR]/out"
},
"infra_step": true,
"name": "dump log",
"timeout": 300,
"~followup_annotations": [
"@@@STEP_LOG_LINE@python.inline@@@@",
"@@@STEP_LOG_LINE@python.inline@import os@@@",
"@@@STEP_LOG_LINE@python.inline@import subprocess@@@",
"@@@STEP_LOG_LINE@python.inline@import sys@@@",
"@@@STEP_LOG_LINE@python.inline@out = sys.argv[1]@@@",
Use full paths for adb Add logic to run on Nexus 5xs in Golo. Bug:788839 Change-Id: I12290d11a0b1b0f012ada216da3e5b2599979c5e Reviewed-on: https://skia-review.googlesource.com/81861 Commit-Queue: Kevin Lubick <kjlubick@google.com> Reviewed-by: Eric Boren <borenet@google.com>
2017-12-07 16:19:31 +00:00
"@@@STEP_LOG_LINE@python.inline@log = subprocess.check_output(['/usr/bin/adb.1.0.35', 'logcat', '-d'])@@@",
Make CPU scaling code cleaner This also makes *sure* the CPU frequency we set the device to actually "takes". Along the way, I learned if scaling_max_freq is < the frequency we set, the scaling_max_freq will be used instead, which was happening to the PixelCs and AndroidOnes. As a result, this may make those two Test- configs faster. Bug: skia: Change-Id: I10c98d37e296a19e1cf67bfe7269bb59cdd912d5 Reviewed-on: https://skia-review.googlesource.com/74360 Commit-Queue: Kevin Lubick <kjlubick@google.com> Reviewed-by: Ben Wagner <benjaminwagner@google.com> Reviewed-by: Eric Boren <borenet@google.com>
2017-11-21 21:47:16 +00:00
"@@@STEP_LOG_LINE@python.inline@for line in log.split('\\n'):@@@",
"@@@STEP_LOG_LINE@python.inline@ tokens = line.split()@@@",
"@@@STEP_LOG_LINE@python.inline@ if len(tokens) == 11 and tokens[-7] == 'F' and tokens[-3] == 'pc':@@@",
"@@@STEP_LOG_LINE@python.inline@ addr, path = tokens[-2:]@@@",
"@@@STEP_LOG_LINE@python.inline@ local = os.path.join(out, os.path.basename(path))@@@",
"@@@STEP_LOG_LINE@python.inline@ if os.path.exists(local):@@@",
"@@@STEP_LOG_LINE@python.inline@ sym = subprocess.check_output(['addr2line', '-Cfpe', local, addr])@@@",
"@@@STEP_LOG_LINE@python.inline@ line = line.replace(addr, addr + ' ' + sym.strip())@@@",
"@@@STEP_LOG_LINE@python.inline@ print line@@@",
"@@@STEP_LOG_END@python.inline@@@"
]
},
{
"cmd": [
Quarantine android bots instead of shutting off device This should make it easier to diagnose and remotely fix the devices/hosts. Only works because of https://chrome-internal-review.googlesource.com/c/infradata/config/+/535542 Bug: skia: Change-Id: Idf11a09da6c7367212a0ac0b0b959cbf49fed85b Reviewed-on: https://skia-review.googlesource.com/89181 Reviewed-by: Ravi Mistry <rmistry@google.com> Commit-Queue: Ravi Mistry <rmistry@google.com>
2017-12-22 15:43:15 +00:00
"python",
"-u",
"RECIPE_MODULE[recipe_engine::file]/resources/fileutil.py",
"--json-output",
"/path/to/tmp/json",
"copy",
" ",
Fix absolute path for recipes Bug: skia: Change-Id: I42c528d86f19d369609b26763334e055501a9b9b Reviewed-on: https://skia-review.googlesource.com/90880 Reviewed-by: Eric Boren <borenet@google.com> Commit-Queue: Kevin Lubick <kjlubick@google.com>
2018-01-04 13:40:27 +00:00
"/home/chrome-bot/force_quarantine"
Make CPU scaling code cleaner This also makes *sure* the CPU frequency we set the device to actually "takes". Along the way, I learned if scaling_max_freq is < the frequency we set, the scaling_max_freq will be used instead, which was happening to the PixelCs and AndroidOnes. As a result, this may make those two Test- configs faster. Bug: skia: Change-Id: I10c98d37e296a19e1cf67bfe7269bb59cdd912d5 Reviewed-on: https://skia-review.googlesource.com/74360 Commit-Queue: Kevin Lubick <kjlubick@google.com> Reviewed-by: Ben Wagner <benjaminwagner@google.com> Reviewed-by: Eric Boren <borenet@google.com>
2017-11-21 21:47:16 +00:00
],
"infra_step": true,
Quarantine android bots instead of shutting off device This should make it easier to diagnose and remotely fix the devices/hosts. Only works because of https://chrome-internal-review.googlesource.com/c/infradata/config/+/535542 Bug: skia: Change-Id: Idf11a09da6c7367212a0ac0b0b959cbf49fed85b Reviewed-on: https://skia-review.googlesource.com/89181 Reviewed-by: Ravi Mistry <rmistry@google.com> Commit-Queue: Ravi Mistry <rmistry@google.com>
2017-12-22 15:43:15 +00:00
"name": "Quarantining Bot"
Make CPU scaling code cleaner This also makes *sure* the CPU frequency we set the device to actually "takes". Along the way, I learned if scaling_max_freq is < the frequency we set, the scaling_max_freq will be used instead, which was happening to the PixelCs and AndroidOnes. As a result, this may make those two Test- configs faster. Bug: skia: Change-Id: I10c98d37e296a19e1cf67bfe7269bb59cdd912d5 Reviewed-on: https://skia-review.googlesource.com/74360 Commit-Queue: Kevin Lubick <kjlubick@google.com> Reviewed-by: Ben Wagner <benjaminwagner@google.com> Reviewed-by: Eric Boren <borenet@google.com>
2017-11-21 21:47:16 +00:00
},
{
"cmd": [
Use full paths for adb Add logic to run on Nexus 5xs in Golo. Bug:788839 Change-Id: I12290d11a0b1b0f012ada216da3e5b2599979c5e Reviewed-on: https://skia-review.googlesource.com/81861 Commit-Queue: Kevin Lubick <kjlubick@google.com> Reviewed-by: Eric Boren <borenet@google.com>
2017-12-07 16:19:31 +00:00
"/usr/bin/adb.1.0.35",
Make CPU scaling code cleaner This also makes *sure* the CPU frequency we set the device to actually "takes". Along the way, I learned if scaling_max_freq is < the frequency we set, the scaling_max_freq will be used instead, which was happening to the PixelCs and AndroidOnes. As a result, this may make those two Test- configs faster. Bug: skia: Change-Id: I10c98d37e296a19e1cf67bfe7269bb59cdd912d5 Reviewed-on: https://skia-review.googlesource.com/74360 Commit-Queue: Kevin Lubick <kjlubick@google.com> Reviewed-by: Ben Wagner <benjaminwagner@google.com> Reviewed-by: Eric Boren <borenet@google.com>
2017-11-21 21:47:16 +00:00
"kill-server"
],
"cwd": "[START_DIR]/skia",
"env": {
"BUILDTYPE": "Debug",
"CHROME_HEADLESS": "1",
"PATH": "<PATH>:RECIPE_PACKAGE_REPO[depot_tools]",
"SKIA_OUT": "[START_DIR]/out"
},
"infra_step": true,
"name": "kill adb server"
},
{
"name": "$result",
Experiment with turning CPUs off during perf Perf was nice and flat after https://skia-review.googlesource.com/c/skia/+/83240 but there was a 4-5x slowdown on the benchmarks themselves, indicating that perhaps we were running nanobench on the LITTLE cores (now on powersave) instead of the big cores we thought (which were recently scaled down). This experiment will show us FOR CERTAIN that we are running on core 0 or 1 which is at .6 max frequency. We expect to see a speedup from the previous results. This is leading to turning of CPUs we don't need to make sure nanobench is running on the ones we expect. Bug: skia: NOTRY=true Change-Id: Ida65181e4d90e778e65e3f22d761288b9ade64f6 Reviewed-on: https://skia-review.googlesource.com/84201 Reviewed-by: Mike Klein <mtklein@chromium.org> Commit-Queue: Kevin Lubick <kjlubick@google.com>
2017-12-12 20:57:20 +00:00
"reason": "Failed build steps: Scale CPU 0 to 0.600000, Scale CPU 0 to 0.600000 (attempt 2), Scale CPU 0 to 0.600000 (attempt 3), Scale CPU 0 to 0.600000 (attempt 3)",
Make CPU scaling code cleaner This also makes *sure* the CPU frequency we set the device to actually "takes". Along the way, I learned if scaling_max_freq is < the frequency we set, the scaling_max_freq will be used instead, which was happening to the PixelCs and AndroidOnes. As a result, this may make those two Test- configs faster. Bug: skia: Change-Id: I10c98d37e296a19e1cf67bfe7269bb59cdd912d5 Reviewed-on: https://skia-review.googlesource.com/74360 Commit-Queue: Kevin Lubick <kjlubick@google.com> Reviewed-by: Ben Wagner <benjaminwagner@google.com> Reviewed-by: Eric Boren <borenet@google.com>
2017-11-21 21:47:16 +00:00
"recipe_result": null,
"status_code": 1
}
]
Reference in New Issue Copy Permalink