Raspberry Pi 4

@ Mike

No, not an ssd

I am using a 32GB Samung Evo+ SDHC

Could be SD writes? I have the 4GB version, so no need to access swap.
Prob could throw some read/writes into the mix to emulate next time.
I'm looking at using rt-tests to do a latency trace. I'll post that up shortly.

Cheers
MrGreg

Hello

Following on from previous

rt-tests can apparently do a latency trace.
See here foe info/example:
taste.tuxfamily.org/wiki/index.php?title..._source_of_latencies
And man page
manpages.debian.org/stretch/rt-tests/cyclictest.8.en.html

Just tried it
It works on Gene's RT kernel

Just a "simple" matter of finding and eliminating them.....
Well....Might be helpful? TBD

Cheers
MrGreg

Thx Andy,
I will read up on mmap and the way hostmot... handles gpio pins
I had thought of using an array og bytes, one byte per pin, with bits for each 'feature'
But i hant though of how users enter the information.
There a re a LOT of pins to describe on these various devices.
thx
tomp

MrGreg
I tested Genes and my own kernel a lot over the last 2 days
There's not a lot of difference between them
I built mine from JT's instruction on Feb9 ( so had some newer pkgs than JT had )
I have different monitor than Gene
I have the npt fix from samco
see
and i force 600mhz AFTER booting ( not automated yet)
sudo sh -c "echo -n 600000 > /sys/devices/system/cpu/cpufreq/policy0/scaling_min_freq"
sudo sh -c "echo -n 600000 > /sys/devices/system/cpu/cpufreq/policy0/scaling_max_freq"
sudo sh -c "echo -n performance > /sys/devices/system/cpu/cpufreq/policy0/scaling_governor"

My system is responsive and videos smooth enough for me (infrequent blips i cant blame on anyone in particular)

here what you can expect:
run
latency-test period
then
IF you really beat on the system
like 4glxgears, 2 resized to larger
and chromium browser showing youtube video full screen, then back to default size so you can see the katency dialog
>I< get 106 usecs ( very long test )
IF you are nice to it ))
like 4 glxgears, no resizing, hands in pocketses
>I< get 50uS ( 13 hrs )

and here's some pix...

conclusion: MY OWN
iF you use it like a 1 bladed swiss army knif, you wont be developing and surfing and gaming
and you rely on the DRO ( not gui ) during cutting
THEN you can use a rpi3 or 4 to controll a slower stepp using gpio
a medium speed stepper using PiCNC
a high speed stepper using SPI with Mesa cards

YMMV but not a lot
HTH

tomp

Hello

Firstly
@ tjtr33 I would like to have a discussion with you about Picnc. PM or open a new thread, if that's OK?

Secondly:
I also did an extended latency test(s) over last night in through the morning. Different approach and different (ish) results

First test part 1
Reason for approach: To test if internal Pi h/keeping etc was causing a big latency by itself?
Gene's RT cold boot.
No other progs /processes started IE none!
Start latency-test period
Initially and for a minute or so
16uS ..... ( glimpsed magic pink unicorn )
23uS 10 mins
58uS 9 hours
"Thinks" cause for optimism?

First test part 2
Without reboot or other intervention before starting next phase (latency test still running)
Glxgears @1080p
Web browser (Firefox) with heavy java on 2 tabs
202uS
File(s) transfer of circa 100MB on SD ( to / from SD)
202uS
Left to simmer on load for further 30 mins
243uS
Left to simmer for further 90 mins
243uS

Hmmm Curious?
Same load conditions and time (2hrs) as previous tests, but applied differently gives quite different results. Also added a Dir/file transfer in to see if that had an effect.
But no, stuck solidly at 243uS for circa 2+hrs.
A little short of 12hrs in total IE pt1 & pt2

Test 2 ( repeat original test)
Glxgears + webserver with x2 heave java script pages ( same as orig previous tests)
restart latency-test period
Initially 98uS
30mins 320uS
2hrs 340uS

Not quite what i was seeing before. Similar start, but no big hike in latency after 90mins

Seems not what you do but the way you do it?

The npt thing may well be worth a look. However if it using npdt? I think that is considered depreciated and may only have a limited future? Hmmm. Prob best I take a look at it.

Re your comments:
Yes, in the real world, I would not be running anything more than the cnc and possibly an editor to tinker with gcode. Nothing else.
I'm not confident about using soft stepping from gpio atm.
Mesa or perhaps Picnc? 7kHz per axis is a minimum/must So is the "Axis" Gui !!
For me, 14kHz per axis, 4axis max is enough. I currently achieve that with a Pi2-Picnc combo.
So yes, both prob doable on Pi4 - Buster.

Thanks for the info & pics
BTW look forward to chat about picnc if poss?

Cheers
MrGreg

Nobody told which version of the RPI4 (size of memory used) and cooling.
There is a huge difference between the RPI4 with 1Go and 4Go of RAM.
Due to thermal throttling i've seen a huge difference when i installed a fan with a heater.
I don't care about latency since i'm using an FPGA. The raspberry is not the Beaglebone, there is no PRU. So using RPI GPIO for Fast I/O is limited i think.
The cheapest way to have fast I/O on RPI could be a STM32 in SPI using the Hostmot2 RPSPI driver as start.
github.com/LinuxCNC/linuxcnc/blob/master...hostmot2/hm2_rpspi.c