GPIO.output still extremely slow in V1.07 on Stretch

[MrBillium]

I am still struggling to get this library working well on Stretch.
With a brand new load of latest Stretch IOT image ( 11-4-17) and no changes except your prerequisites and installation, I am getting a 900mS run time for setting a pin 100 times. The same code take 3mS running on Jessie. This problem is independent of Python version and I have tried both BBB and BBG.

Bill

`import Adafruit_BBIO.GPIO as GPIO
import timeit

pin = 'P9_14'

GPIO.setup(pin, GPIO.OUT)

def setoutput():
for x in range(1, 10):
GPIO.output(pin, True)
GPIO.output(pin, False)

run 5 times, 10 x 2 x 5 = 100 outputs

result = timeit.Timer(setoutput).timeit(5)
print('time to complete 100 sets = ', result)
GPIO.cleanup()`

[pdp7]

@MrBillium thanks for the feedback. I'll look into this issue.

[pdp7]

Here is the Python test I've run:

import Adafruit_BBIO.GPIO as GPIO
import timeit

pin = 'P9_14'

GPIO.setup(pin, GPIO.OUT)

def setoutput():
  for x in range(1, 10):
    GPIO.output(pin, True)
    GPIO.output(pin, False)
    #run 5 times, 10 x 2 x 5 = 100 outputs

result = timeit.Timer(setoutput).timeit(5)
print('time to complete 100 sets = ', result)
GPIO.cleanup()

Here is the output for Debian 9.2 with 4.14.0-rc7-ti-r6 kernel:

debian@beaglebone:~$ sudo python time-test.py
('time to complete 100 sets = ', 1.4771080017089844)

output of sudo /opt/scripts/tools/version.sh:

git:/opt/scripts/:[6c2688b0be448b7bb9ca18849b430d496a84acb4]
eeprom:[A335BNLT000C3014BBBK1316]
model:[TI_AM335x_BeagleBone_Black]
dogtag:[BeagleBoard.org Debian Image 2017-10-17]
bootloader:[microSD-(push-button)]:[/dev/mmcblk0]:[U-Boot 2017.09-00002-g0f3f1c7907]
bootloader:[eMMC-(default)]:[/dev/mmcblk1]:[U-Boot 2017.09-rc2-00002-g84a7f2]
kernel:[4.14.0-rc7-ti-r6]
nodejs:[v6.11.5]
uboot_overlay_options:[enable_uboot_overlays=1]
uboot_overlay_options:[uboot_overlay_pru=/lib/firmware/AM335X-PRU-RPROC-4-4-TI-00A0.dtbo]
uboot_overlay_options:[enable_uboot_cape_universal=1]
pkg:[bb-cape-overlays]:[4.4.20171027.1-0rcnee1~stretch+20171027]
pkg:[bb-wl18xx-firmware]:[1.20170829-0rcnee1~stretch+20170829]
pkg:[firmware-ti-connectivity]:[20170823-1rcnee0~stretch+20170830]

[pdp7]

With a brand new load of latest Stretch IOT image ( 11-4-17) and no changes except your prerequisites and installation, I am getting a 900mS run time for setting a pin 100 times. The same code take 3mS running on Jessie

@MrBillium:
What kernel is running on the Jessie image that you are using?
Are you running the same version of Adafruit_BBIO on both Jessie and Stretch?

[pdp7]

One possible inefficiency is this check for board type inside gpio_set_value():
https://github.com/adafruit/adafruit-beaglebone-io-python/blob/master/source/event_gpio.c#L405

The function beaglebone_blue() executes popen() to run grep:
https://github.com/adafruit/adafruit-beaglebone-io-python/blob/master/source/common.c#L602

This should be cached so that it only needs to be done once when Adafruit_BBIO is initialized.

[MrBillium]

Just to be clear, when I say BBB I mean BB Black not Blue. Also for both test cases the setting of the pin does actually work in that I see the output switching.

First a BBB with brand new 11-4 image and no changes except Adafruit install and requirements: Runs in 1.3 seconds.
root@beaglebone:/home# pip3 show adafruit_bbio
Name: Adafruit-BBIO
Version: 1.0.7
Summary: A module to control BeagleBone IO channels
Home-page: https://github.com/adafruit/adafruit-beaglebone-io-python/
Author: Justin Cooper
Author-email: justin@adafruit.com
License: MIT
Location: /usr/local/lib/python3.5/dist-packages
Requires:

root@beaglebone:/home# sudo /opt/scripts/tools/version.sh
git:/opt/scripts/:[a5d5f437baeaa7f8e14d5ae037268abf7a67e88e]
eeprom:[A335BNLT00C05114BBBK195C]
model:[TI_AM335x_BeagleBone_Black]
dogtag:[BeagleBoard.org Debian Image 2017-11-05]
bootloader:[eMMC-(default)]:[/dev/mmcblk1]:[U-Boot 2017.09-00002-g0f3f1c7907]
kernel:[4.4.91-ti-r136]
nodejs:[v6.11.5]
uboot_overlay_options:[enable_uboot_overlays=1]
uboot_overlay_options:[disable_uboot_overlay_video=1]
uboot_overlay_options:[disable_uboot_overlay_audio=1]
uboot_overlay_options:[disable_uboot_overlay_adc=1]
uboot_overlay_options:[uboot_overlay_pru=/lib/firmware/AM335X-PRU-RPROC-4-4-TI-00A0.dtbo]
uboot_overlay_options:[enable_uboot_cape_universal=1]
pkg:[bb-cape-overlays]:[4.4.20171101.0-0rcnee1stretch+20171101]
pkg:[bb-wl18xx-firmware]:[1.20170829-0rcnee1stretch+20170829]
pkg:[firmware-ti-connectivity]:[20170823-1rcnee0~stretch+20170830]

root@beaglebone:/home# sudo python3 timetest.py
time to complete 100 sets = 1.387603264000063
root@beaglebone:/home#

Next a BBG with my production load of Jessie with overlay changes . BBIO library installed in Python 3.6 Venv. Runs in 0.05 seconds!

(py360) root@beaglebone:/home# pip3 show adafruit_bbio
Name: Adafruit-BBIO
Version: 1.0.3
Summary: A module to control BeagleBone IO channels
Home-page: https://github.com/adafruit/adafruit-beaglebone-io-python/
Author: Justin Cooper
Author-email: justin@adafruit.com
License: MIT
Location: /home/venvs/py360/lib/python3.6/site-packages
Requires:

(py360) root@beaglebone:/home# sudo /opt/scripts/tools/version.sh
git:/opt/scripts/:[e869b11af4400bfba2a333f00f0f06e1d145d1d2]
eeprom:[A335BNLTBBG1BBG217014819]
dogtag:[BeagleBoard.org Debian Image 2017-03-26]
bootloader:[/dev/mmcblk1]:[U-Boot 2017.03-00002-gd12b1519b4]
kernel:[4.4.55-ti-r94]
nodejs:[v4.8.1]
(py360) root@beaglebone:/home# python3 timetest.py
time to complete 100 sets = 0.005994903000100749
(py360) root@beaglebone:/home#

I have several BB's on my bench so can run any tests you think may help.

Bill

[pdp7]

@MrBillium It appears that Adafruit_BBIO version 1.0.3 is on your Jessie setup. Could you try the latest Adafruit_BBIO version (v1.0.7) on Jessie and see if the performance regression occurs?

[pdp7]

@MrBillium please try out the latest master branch.

Timing test shows large performance improvement after commit 3281479

debian@beaglebone:~/ssh/adafruit-beaglebone-io-python$ python3 ~/time-test.py 
time to complete 100 sets =  0.006628266011830419

debian@beaglebone:~/ssh/adafruit-beaglebone-io-python$ python ~/time-test.py 
('time to complete 100 sets = ', 0.0067021846771240234)

Results before commit 3281479 with performance regression:

debian@beaglebone:~/ssh/adafruit-beaglebone-io-python$ python ~/time-test.py 
('time to complete 100 sets = ', 1.5183091163635254)

debian@beaglebone:~/ssh/adafruit-beaglebone-io-python$ python3 ~/time-test.py 
time to complete 100 sets =  1.3909540939785074

[MrBillium]

On Stretch image, rolling BBIO back to 1.0.3 fixes the issue
1.0.7 time to complete = 1.37
1.0.3 time to complete = 0.006

Bill

[pdp7]

@MrBillium were you able to try with the latest master? I think commit 3281479 should help.

[pdp7]

@MrBillium please try version 1.0.8:
https://github.com/adafruit/adafruit-beaglebone-io-python/releases/tag/1.0.8

[MrBillium]

Yes it works! I am getting 0.006S run time using 1.0.8 and my LCD which uses Adafruit_charlcd is now writing whole screen at a time instead of one character at a time – which is why I went down this rabbit hole. Thanks again Bill From: Drew Fustini [mailto:notifications@github.com] Sent: Sunday, November 19, 2017 5:30 PM To: adafruit/adafruit-beaglebone-io-python <adafruit-beaglebone-io-python@noreply.github.com> Cc: MrBillium <bilmar19@gmail.com>; Mention <mention@noreply.github.com> Subject: Re: [adafruit/adafruit-beaglebone-io-python] GPIO.output still extremely slow in V1.07 on Stretch (#190) @MrBillium <https://github.com/mrbillium> were you able to try with the latest master? I think commit <3281479> 3281479 should help. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#190 (comment)> , or mute the thread <https://github.com/notifications/unsubscribe-auth/ANnql7qRhy62fg5PUAdcGCdVgVuqdHXMks5s4KvogaJpZM4QbGNP> . <https://github.com/notifications/beacon/ANnql_gNGgjU4ajwAIl54aLWYXVEqzfnks5s4KvogaJpZM4QbGNP.gif>

[pdp7]

Good to hear! Thanks for pointing out this performance regression so that it could be fixed.