dh1tw has quit [Quit: My MacBook Pro has gone to sleep. ZZZzzz…]
<MoeIcenowy>
apritzel: I'm trying to add Travis CI to arm-trusted-firmware of Allwinner flavor
leviathanch has joined #linux-sunxi
<KotCzarny>
running pigz -11 is a lesson in patience
<ErwinH>
Try running it at 24MHz
<KotCzarny>
erwinh: compare power usage with 24mhz to 408mhz
tyler-baker has quit [*.net *.split]
<KotCzarny>
or even 480mhz
vpeter has quit [Read error: Connection reset by peer]
<ErwinH>
My powermeter isn't accurate enough to get a good readout. Only shows current with 2 decimals. But: running at 1100uV it's 0.09A@480MHz vs 0.08A@24MHz @5V.
<KotCzarny>
so unless you can undervolt soc at lower than 480mhz, it's not worth it
<ErwinH>
Depends on your usecase :)
<KotCzarny>
what would be 24mhz usecase that would prefer 24mhz?
<ErwinH>
Something that needs internet and gpio but doesn't need cpu power.
<KotCzarny>
but since difference is 10mA +/-10..?
matthias_bgg has joined #linux-sunxi
Amit_T_ has joined #linux-sunxi
<miasma>
ErwinH: try esp8266. it supports both tcp/ip and gpio. you'll save another 88% of energy
tyler-baker has joined #linux-sunxi
<ErwinH>
But it has 2.4G wifi
<buZz>
like >90% of access points, yes :P
<ErwinH>
Don't take this discussion so seriously btw. I don't have any real application for this setup.
<ErwinH>
And that 90% is complaining about the terrible wireless connections :)
<KotCzarny>
like you would get stellar stability with cheapest realtek they could get ;)
Worf has joined #linux-sunxi
lemonzest has joined #linux-sunxi
fl_0 has joined #linux-sunxi
<ErwinH>
Therefor I skip wireless connectivity unless.
<ErwinH>
unless there's a decent chip/driver.
<buZz>
ErwinH: ESP32 has Ethernet ;)
<buZz>
(and wifi and BT)
<ErwinH>
Hehe :)
<MoeIcenowy>
playing with chips needs some creativity
mzki has joined #linux-sunxi
LargePrime has quit [Ping timeout: 268 seconds]
Amit_T_ has quit [Ping timeout: 260 seconds]
LargePrime has joined #linux-sunxi
<ErwinH>
Must say, the H5 with mainline kernel is running smoothly. DVFS, THS, Ethernet and USB.
<jelle>
H5 with mainline arm64?!
msevwork has quit [Quit: Leaving]
<ErwinH>
With a few patches :)
<ErwinH>
1368MHz@1.4V at 48C
DullTube has joined #linux-sunxi
<jelle>
ErwinH: ah nice, I guess apritzel's work
<ErwinH>
Apritzel and a bit from Megi
<jelle>
sounds like after christmas todo
<KotCzarny>
erwinh: what? 1.37ghz?
<KotCzarny>
did they change chip process or what?
<ErwinH>
Haven't tried cpuburn yet btw... :)
<KotCzarny>
go for it
<KotCzarny>
:)
apritzel has joined #linux-sunxi
enrico__ has joined #linux-sunxi
<ErwinH>
cpuburn-a53, no sweat :) Running at approx 70C, between 480MHz and 816Mhz.
bugzc has quit [Ping timeout: 260 seconds]
<jelle>
ErwinH: orange pi pc2 I assume
<ErwinH>
Is there any other H5 atm?
lkcl has quit [Ping timeout: 265 seconds]
<jelle>
oh no idea ;-)
<jelle>
ErwinH: there are so many orangepi's :P
<ErwinH>
Fair point, but as far as I know the other H5 boards from OrangePi are still in development.
<scelestic>
too bad the only h5 ones have 1gb memory :(
<KotCzarny>
ErwinH: disable cpu throttling
<montjoie>
yes need more than 4:)
<ErwinH>
Why?
<KotCzarny>
because otherwise you won't see if its stable at that freq
<ErwinH>
But it doesn't have to be stable at 95C, since you want it to throttle.
kaspter has quit [Ping timeout: 240 seconds]
<KotCzarny>
nope, you want it to run cool and stable no matter what freq
<KotCzarny>
and throttling is just a safety valve
<ErwinH>
But running cool at 1.3GHz@Full load is not realistic.
<KotCzarny>
why not? it's only a matter of running stable at some voltage
lkcl has joined #linux-sunxi
<KotCzarny>
h3 chips already dont throttle with 1296mhz
<apritzel>
KotCzarny: well, you might want to run a single threaded program at full frequency, while the other cores are mostly idle
<KotCzarny>
apritzel, that's true
<KotCzarny>
maybe armbian should provide additional throttling profile that prefers freq over cpu count
<KotCzarny>
igorpec ^
<apritzel>
actually I guess it should work like this: you test those OPPs with good cooling to get the table, so to prove that frequency x is stable at voltage y
<apritzel>
and then you install a temperature monitor to prevent the chip running too hot
<apritzel>
so that people get the best experience with their actual cooling solution
<NiteHawk>
apritzel: to be honest i haven't looked into that in depth - i still need to get a SPI chip and hook it up to my pine64...
<apritzel>
NiteHawk: IC, it's just that there are boards now with SPI flash soldered and initialising them via FEL is very tempting
<NiteHawk>
i suspect with the opi zero the SPI functionality will become increasingly useful, but its probably up to ssvb to decide when the code is ready for integration into the main repo
vpeter has joined #linux-sunxi
<KotCzarny>
nitehawk: looks like its ok on my opipc
<KotCzarny>
maybe you should add minimal usleep in that while
<KotCzarny>
10 or something
Da_Coynul has joined #linux-sunxi
<NiteHawk>
Allwinner's code doesn't do that either, as it's controlled by the hardware. we just have to wait for bit 1 to flip back to 0 (indicating that the read is complete)
<KotCzarny>
i know, but i hate busyloops without minimal usleep
<KotCzarny>
and aw code isnt good coding practices source ;)
<NiteHawk>
hmm - maybe "wfi"? (wait for interrupt)
<KotCzarny>
out of curiosity i've added single printf there and at most i got three chars yesterday
<KotCzarny>
could be more on running system (ie. not fel)
<NiteHawk>
and uart0-helloworld is bare metal code, there's no usleep library routine available - we'd have to do our own wait loop anyway
<KotCzarny>
oh well, then nvm
<apritzel>
and frankly: what would a bare metal usleep implemention do anyway?
<NiteHawk>
KotCzarny: you mean 'serial' printf right before the wait loop? because after it, the SoC should be responsive again
<KotCzarny>
nitehawk: inside that empty {}
<apritzel>
I guess enabling interrupts and using a timer and wfi is a bit overkill for that ...
<NiteHawk>
okay, i take that as "WFI without interrupts would be pointless"?
<NiteHawk>
otoh, i'd expect FEL mode to have I/O and interrupts enabled
<KotCzarny>
nitehawk, disregard my comment, busyloop is good enough
dh1tw has quit [Quit: My MacBook Pro has gone to sleep. ZZZzzz…]
<NiteHawk>
KotCzarny: probably. it should be a very minuscule timespan anyway
<apritzel>
NiteHawk: does that code run with the MMU on? that depends on the SoC, right?
<apritzel>
so as long as the memory region is non-cacheable, that should be fine
<apritzel>
if you don't mind an endless loop in case of some error
pietrushnic`away has joined #linux-sunxi
<apritzel>
NiteHawk: and for the interrupts: can you read 0x1c81000 and 0x1c82000 and check the lower bits?
<NiteHawk>
right. but the only thing it does is to read the SID_PRCTL - following AW's reference implementation that they use in their U-Boot code
<apritzel>
NiteHawk: these are the GIC control registers, my hunch is that they are disabled
<apritzel>
NiteHawk: well, that's probably just bike shedding, but if for some reason the bit never gets set, it blocks
<NiteHawk>
true. would a "wfi" be harmful (i.e. deadlock), if there's no interrupt generated?
<apritzel>
and my only concern was that if this region is for whatever reason marked as cacheable (with the MMU on) you just read from the L1 cache ...
<apritzel>
should be much faster this way, though :-D
leviathanch has quit [Remote host closed the connection]
<NiteHawk>
would "dsb" help with the L1 cache?
<apritzel>
no
<apritzel>
but frankly the MMU setup should mark the MMIO space as device non-cacheable
Pepe has joined #linux-sunxi
<apritzel>
I guess it does that
<NiteHawk>
yes, anything else would be extremely counter-intuitive
<apritzel>
dsb doesn't help because you are: a) single core and b) there is a dependency on the result of the read, so the CPU can't "optimise" anyway
pietrushnic`away is now known as pietrushnic
hojnikb has joined #linux-sunxi
<hojnikb>
ErwinH: can you test stable voltages for 1200Mhz and 1300Mhz on H5 ? 1.4V seems a bit high.
<hojnikb>
especially if it's indeed fabricated at 28nmž
Da_Coynul has quit [Quit: My Mac has gone to sleep. ZZZzzz…]
hojnikb has quit [Quit: Page closed]
yann-kaelig has joined #linux-sunxi
tkaiser has joined #linux-sunxi
<tkaiser>
hojnikb: I doubt H5 is 28nm. I bet it's 40 nm like most recent Allwinner SoCs (A80 and A83T being the 28nm expections)
<ErwinH>
If you like I'll test the stability, shouldn't be to hard.
<tkaiser>
ErwinH: If you want to do it correctly you need a large heatsink and a lot of airflow ;)
<tkaiser>
ErwinH: This OpenBLAS based Linpack mentioned above did the trick. Very realiable regarding undervoltage situations.
<ErwinH>
What type of heatsink did you use?
<tkaiser>
Rather insufficient ones. One 20x20x5 mm (googling for '20x20 cubie' should give an idea) you can see three comments above at the github issue. Low price, low profile, low performance.
<tkaiser>
zoobab_: I'm always surprised how enthousiastic people are doing silly things. Take a rather slow device and start to overclock it. Why? Why not taking a fast device when needed?
<scelestic>
i remember it being usefull but this was with a pentium-233mhz overclocked to 300mhz with a high busspeed
<tkaiser>
And if I want to squeeze out the max from a RPi 3 then I throw Raspbian in the bin and try to use ARMv8 code built with a modern compiler version. Overclocking is so stupid IMO.
paulk-collins has quit [Ping timeout: 265 seconds]
nekomona has joined #linux-sunxi
<zoobab_>
raspbian is built with an old version of gcc?
<KotCzarny>
raspbian is built with an old version of arm core
<wens>
raspian is built for armv6
leviathanch has joined #linux-sunxi
libv_ is now known as libv
<tkaiser>
with an outdated GCC version. So the 'best' of two worlds combined ;)
<tkaiser>
IIRC it's GCC 4.9 now and by simply letting some code run through 5.4 for example certain tasks finish faster. Same with ARMv6 code running on an ARMv8 CPU. Just stupid if performance is what you want.
<tkaiser>
So instead of adding copper shims I would add some brain to the problem.
<scelestic>
guess i should update my build machine, i think debian stable still uses gcc 4.9
<ErwinH>
Is that the reason why sysbench performs way faster on the H5 than on the Raspberry Pi 3? (7 seconds vs 120 second) Armv6 vs armv8?
<tkaiser>
ErwinH: Exactly. This test is doing prime numbers and that can use some other and more efficient instructions when being able to use ARMv8 instructions.
<jelle>
would probably be faster with armv7h too
<tkaiser>
In other words: sysbench is not a benchmark but a joke. But that applies to most commonly used benchmarks
<tkaiser>
jelle: Sure. And GCC version (Jessie 4.9, Ubuntu Xenial 5.4) makes approx 30 percent difference with this 'benchmark'
<tkaiser>
KotCzarny: Huh? The 'best ones' since there exists one single use case for compression in the world?
<zoobab_>
zstd not in the graph?
<KotCzarny>
zoobab, is there debian package for it?
<ErwinH>
I'll have to prepare the tests for tomorrow. Don't have the right tools atm.
<tkaiser>
KotCzarny: And solely using technology from last century? Are you kidding?
<tkaiser>
KotCzarny: apt-get install zstd
<KotCzarny>
no banana
<KotCzarny>
debian jessie
<tkaiser>
KotCzarny: But when you're after performance you don't use distro packages. Since you want performance. Don't use overclocking, heatsink, fans and horribly outdated Debian derivates. Use your brain ;)
<KotCzarny>
tkaiser, but using distro packages mean 'same compiler, same os'
<wens>
KotCzarny: time to upgrade to unstable
<KotCzarny>
so results are comparable, not absolute
Axl_ has joined #linux-sunxi
<tkaiser>
KotCzarny: Nope. Shitty settings remain shitty settings. If you want to make a performance comparison you do differently.
<KotCzarny>
for one, rzip -z -9 resulted in smallest file
<tkaiser>
To benefit from latest compiler technology. How can it be that zstd running on a single core outperforms lbzip2? Since it's technology from this and not last century.
<KotCzarny>
i mean lrzip
<KotCzarny>
and once i add zstd i'll update the graph
<tkaiser>
KotCzarny: Especially when we're talking about SBC it's moronic to only consider stuff that has been developed with CPUs from 30 years ago. Way too inefficient.
Andy-D_ has joined #linux-sunxi
<KotCzarny>
funny that zstd is in sid/stretch and not earlier
paulk-collins has joined #linux-sunxi
<zoobab_>
zstd kill them all
<tkaiser>
zoobab_: while running on a single CPU core and the others utilizing 4 ;)
KB3VGW has joined #linux-sunxi
<zoobab_>
problem is that FB might a patent behind it
<zoobab_>
so adoption/reimplementation might be hindered
<zoobab_>
that was one of the reason behind the creation of Gzip actually
<tkaiser>
zoobab_: Well, I understand https://github.com/facebook/zstd/issues/335 differently but I'm no lawyer. I only know that with zstd it's possible to create device backups on slow ARM devices in reasonable time. Since it's sooo lightweight regarding CPU ressources compared to the usual compressors everyone only thinks about.
<KotCzarny>
gotta redo tests later on arm cores later too, now i'm having problems to use pzstd to use data from stdin and write to stdout
KB3VGW has quit [Read error: Connection reset by peer]
KB3VGW has joined #linux-sunxi
<KotCzarny>
umkay. let's roll.
<KotCzarny>
1 to 22
<zoobab_>
@tkaiser reading from the discussion, it does not look like FB applied for any patent on this
<KotCzarny>
zoobab: ..yet? ;)
<KotCzarny>
anyway, compression ratio isnt that great
<zoobab_>
in the US there is also this stupid 18 months period of uncertainty
<zoobab_>
the best would be to have confirmation from FB themselves
<KotCzarny>
same as before, pixz/plzip are better when compression ratio is wanted, and are fast enough
JohnDoe71rus has quit [Quit: Miranda IM! Smaller, Faster, Easier. http://miranda-im.org]
likewise has quit [Client Quit]
likewise has joined #linux-sunxi
KB3VGW has quit [Read error: Connection reset by peer]
KB3VGW has joined #linux-sunxi
likewise has left #linux-sunxi [#linux-sunxi]
likewise has joined #linux-sunxi
<likewise>
Are any meta-sunxi maintainers/users lingering here?
<jelle>
just ask
<buZz>
it always amazes me to see OpenEmbedded still being used :O
popolon has joined #linux-sunxi
<scelestic>
oh that brings back memories, gpe, opie and ipaqs
<scelestic>
though i think OpenEmbedded came after that
<KotCzarny>
tkaiser, what pzstd compression level are you using?
<KotCzarny>
because i see better time/compression options at almost any step
DullTube has quit [Quit: Leaving]
fkluknav has quit [Remote host closed the connection]
fraveydank has quit [Quit: My MacBook has gone to sleep. ZZZzzz…]
<tkaiser>
KotCzarny: I don't use outdated Debian Jessie. No idea what's going on there. Don't care either. :) If you're the only one on this planet where zstd fails maybe it's related to your setup?
<KotCzarny>
sure, but tell me, did you try pixz in your setup?
fraveydank has joined #linux-sunxi
<tkaiser>
Sure, way too resource hungry. But of course better compression result with highest compression.
<likewise>
MoeIcenowy: where did you source this device?
<MoeIcenowy>
Taobao
<MoeIcenowy>
but it is still not generally available
<MoeIcenowy>
and it easily run out of stock
ErwinH has quit [Ping timeout: 258 seconds]
massi has quit [Quit: Leaving]
Axl_ has quit [Ping timeout: 240 seconds]
tkaiser has joined #linux-sunxi
<tkaiser>
MoeIcenowy: This Lichee Pi is lacking AXP209?
<NiteHawk>
MoeIcenowy: "The debugging UART is the UART1, located at the top-right corner, silked with RX and TX." Shouldn't that be bottom-right according to your image?
<NiteHawk>
ah, sorry - ignore that. the image shows the board rotated by 90 degrees
<MoeIcenowy>
tkaiser: not lacking
<MoeIcenowy>
NitHawk: See the "Lichee Pi" silkprint
Mr__Anderson has quit [Remote host closed the connection]
KB3VGW has quit [Read error: Connection reset by peer]
KB3VGW has joined #linux-sunxi
nekomona has quit [Quit: Leaving]
zzeroo has joined #linux-sunxi
fraveydank has joined #linux-sunxi
fraveydank has quit [Quit: My MacBook has gone to sleep. ZZZzzz…]
fraveydank has joined #linux-sunxi
Amit_T_ has quit [Remote host closed the connection]
enrico__ has quit [Quit: Bye]
fraveydank has quit [Quit: My MacBook has gone to sleep. ZZZzzz…]
Amit_T_ has joined #linux-sunxi
leviathanch has quit [Remote host closed the connection]
fraveydank has joined #linux-sunxi
leviathanch has joined #linux-sunxi
fraveydank has quit [Client Quit]
fraveydank has joined #linux-sunxi
fraveydank has quit [Client Quit]
fraveydank has joined #linux-sunxi
fraveydank has quit [Client Quit]
yann-kaelig has quit [Quit: Leaving]
fraveydank has joined #linux-sunxi
yann-kaelig has joined #linux-sunxi
fraveydank has quit [Client Quit]
Amit_T_ has quit [Remote host closed the connection]
IgorPec has quit [Ping timeout: 250 seconds]
netlynx has joined #linux-sunxi
netlynx has quit [Changing host]
netlynx has joined #linux-sunxi
dh1tw has joined #linux-sunxi
KB3VGW has quit [Read error: Connection reset by peer]
KB3VGW has joined #linux-sunxi
reinforce has quit [Remote host closed the connection]
fraveydank has joined #linux-sunxi
jstein_ has joined #linux-sunxi
ErwinH has joined #linux-sunxi
apritzel has quit [Ping timeout: 246 seconds]
jstein_ is now known as jstein
ErwinH has quit [Ping timeout: 265 seconds]
Amit_T_ has joined #linux-sunxi
Amit_T_ has quit [Client Quit]
<montjoie>
does someone recently compiled uboot ? it failed with "SyntaxError: Missing parentheses in call to 'print'"
<likewise>
montjoie: what uboot GIT?
<MoeIcenowy>
I'm now trying to do some development on u-boot master
buZz has quit [Ping timeout: 256 seconds]
<montjoie>
uboot git master yes
KB3VGW has quit [Read error: Connection reset by peer]
Amit_T_ has joined #linux-sunxi
MXfive has quit [Quit: Sleep Quit.]
<likewise>
montjoie: upstream master, right?
<montjoie>
yes
<montjoie>
checkout v2016.11 make the issue disapear
KB3VGW has joined #linux-sunxi
<rookieone>
e
fraveydank has quit [Read error: Connection reset by peer]
florianH has quit [Quit: Connection closed for inactivity]
jstein_ has joined #linux-sunxi
jstein is now known as Guest56026
jernej has joined #linux-sunxi
jstein_ is now known as jstein
Guest56026 has quit [Ping timeout: 265 seconds]
MXfive has joined #linux-sunxi
Amit_T_ has quit [Quit: ChatZilla 0.9.93 [Firefox 46.0.1/20160511224619]]
MXfive has quit [Quit: Sleep Quit.]
foxx_ has quit [Ping timeout: 250 seconds]
KB3VGW has quit [Read error: Connection reset by peer]
KB3VGW has joined #linux-sunxi
jstein has quit [Remote host closed the connection]
<MoeIcenowy>
mripard: your ARM SUNXI part in U-Boot didn't include boards/sunxi/ now...
buZz has joined #linux-sunxi
buZz is now known as Guest74603
terra854 has quit [Quit: Connection closed for inactivity]
Mr__Anderson has joined #linux-sunxi
ErwinH has joined #linux-sunxi
KB3VGW has quit [Read error: Connection reset by peer]
KB3VGW has joined #linux-sunxi
IgorPec has joined #linux-sunxi
Andy-D_ has joined #linux-sunxi
leviathanch has quit [Remote host closed the connection]
<FrostyBytes>
it is mainline linux (actually linux-libre). sunxi_defconfig works, but when I change things to modules usb is broken
<FrostyBytes>
probably I'm forgetting to load a module
<FrostyBytes>
oh also usbhid I am loading that
<FrostyBytes>
lsusb also shows no output
<FrostyBytes>
I went through everything usb-related that is on in defconfig, and tried to determine the related module and loaded it
ErwinH has joined #linux-sunxi
<FrostyBytes>
when I do lsmod, I get ohci_platform, ohci_hcd, ehci_platform, ehci_hcd, sunxi, phy_sun4i_usb, musb_hdrc, usbcore, nls_base, phy_generic, udc_core, usb_common, phy_sun9i_usb
cptG_ has joined #linux-sunxi
ErwinH has quit [Ping timeout: 240 seconds]
cptG has quit [Ping timeout: 268 seconds]
Da_Coynul has joined #linux-sunxi
<FrostyBytes>
let's see what happens when I reinject the =y's from sunxi_defconfig onto my config
Da_Coynul has quit [Read error: Connection reset by peer]