bertrik has quit [Remote host closed the connection]
Andy-D has joined #linux-sunxi
ricardocrudo has quit [Remote host closed the connection]
<HeavyMetal>
hi I am wondering if anyone got Linux to boot from the NAND of an A10 device? I am trying to get the MK802ii to boot from the NAND but so far it is not happening
<Turl>
HeavyMetal: yes, it can be done, and many have done so. It's even documented on the wiki if my memory serves me right
<ikeeki>
and it works with some very basic games, but it seems that the .bin files inside the roms are most of the times missing or misnamed,,, I've tried the P. bay looking for some fixed roms, no luck. If you know some better sources for that pliz tell me. If anybody could suggest some fex file that could help in some way or any tips I'd be grateful too. In short, any help or tip will be appreciated, it does not matter what
<rm>
you what mate
<rm>
"roms are misnamed" => "fex file must be the solution"?
<ikeeki>
most of them lack of some files inside the .zip
<ikeeki>
regarding fex:
<ikeeki>
Im on x11, and I wondered if anybody knows some special tip kind of framebuffer_number=4 that could improve performance or stability, as Im experiencing some little blackscreen flashes too while launching mole (a 46 kb 80's game)
<ikeeki>
x11=x windows, sorry
<ikeeki>
(xorg)
<Turl>
ikeeki: what board?
<ikeeki>
Turl: A20/CT. Im trying to get all the info available about game emulation
<Turl>
ikeeki: not much to say, other than make sure you're using fbturbo
<ikeeki>
Im moving to game emulation, as I think I cannot optimize more the video playing in aRUNTU (1080p in 1920x1080p60 video output)
<ikeeki>
yep, I'll try to experiment with different fbturbo options via xorg...
<Turl>
ikeeki: you may want to lower the resolution or refresh rate if you want more performance
<ikeeki>
NEVER
<ikeeki>
jajajaja
<Turl>
I doubt you'll ever notice if you run 1080p50 :)
rz2k has joined #linux-sunxi
<ikeeki>
Well... I'd say that I can notice in video playing (cedrus, mpv)
<Turl>
you mean for content that's 30fps or less? :)
<ikeeki>
Yep, surprisingly I notice a better movement of the objects in a same screen
ganbold_ has quit [Ping timeout: 240 seconds]
ganbold_ has joined #linux-sunxi
<ikeeki>
Kind of less stuttering
<ikeeki>
have to go, thanks Turl
ikeeki has quit [Quit: Saliendo]
jemk has joined #linux-sunxi
Gerwin_J has quit [Ping timeout: 240 seconds]
bgal has joined #linux-sunxi
prahal has joined #linux-sunxi
physis has joined #linux-sunxi
kivutar has quit [Ping timeout: 240 seconds]
ecelis has quit [Ping timeout: 240 seconds]
alexst has joined #linux-sunxi
alexst has quit [Ping timeout: 256 seconds]
Andy-D has joined #linux-sunxi
bgal has quit [Ping timeout: 240 seconds]
ecelis has joined #linux-sunxi
akaizen_ has quit [Remote host closed the connection]
akaizen has joined #linux-sunxi
akaizen has quit [Ping timeout: 240 seconds]
alexst has joined #linux-sunxi
Zboonet has joined #linux-sunxi
ninolein has quit [Ping timeout: 240 seconds]
ninolein has joined #linux-sunxi
alexst has quit [Ping timeout: 260 seconds]
<Turl>
(for reference, from #olimex) ssvb_: iirc you explained here some time ago that if you need data from a different section of RAM, it needs to be switched somehow and that generates delay
<Turl>
ssvb_: is that the case?
<ssvb_>
Turl: yes, there are 8 banks, each having size 4K on the systems with the 32-bit memory interface
physis has quit [Remote host closed the connection]
<ssvb_>
Turl: the pages, which belongs to the same bank can't be active at the same time
<Turl>
ssvb_: so, if the dma engine can do burst of up to 8 32bit ops, that would be the most optimal then?
<ssvb_>
Turl: the practical implication of this banks stuff is that if you copy from one buffer into another and they are exactly 32K apart, then you get really poor performance because of bank conflicts and switching all the time
<ssvb_>
Turl: the larger bursts are always better
<ssvb_>
Turl: especially in this worst case of banks switching scenario
physis has joined #linux-sunxi
<ssvb_>
Turl: how fast is the dma working for you now?
<Turl>
ssvb_: I noticed empirically :) bursts of 1x8b gave ~3-4MB/s, and 8x32b gives ~12MB/s
tomcheng76 has joined #linux-sunxi
<Turl>
(on memcpy ops)
<ssvb_>
Turl: this sounds suspiciously slow
<Turl>
ssvb_: if I thread several requests I can get like 14MB/s, so maybe rewriting the way it issues requests so there's less idle time could yield a bit better speeds
tomcheng86 has quit [Ping timeout: 240 seconds]
<Turl>
ssvb_: yeah, it's certainly not a speed demon
<ssvb_>
Turl: the CPU memcpy is much faster than that (by a factor of 10x or more), even if you are reading uncached memory
<ssvb_>
Turl: how large are the buffers that you used in these tests? maybe the setup overhead is very high and dwarfs the time spent actually copying the data?
<Turl>
ssvb_: I tried w/ 60K (61440) buffers
<Turl>
ssvb_: dmatest also then goes and reads the buffers and compares contents
<Turl>
so maybe that competes with the engine
<ssvb_>
Turl: try using larger buffers (8MB or more)
<Turl>
ah cool, noverify
<ssvb_>
yeah, and disabling verification should clearly help :)
<Turl>
I disabled verification and got ~71MB/s
<ssvb_>
what is the exact difference between the source and destination pointers?
<Turl>
you should ask dmatest that :p
<ssvb_>
71MB/s is not so great either
<ssvb_>
something closer to 1GB/s would be good
<ssvb_>
something below 500MB/s is already under-performing
<ssvb_>
but if you hit bank conflicts, this probably could explain the performance drop
netlynx has quit [Remote host closed the connection]
<Turl>
ssvb_: max buffer size is 128K :/
<Turl>
ssvb_: this dma engine was certainly not made to be high performance
Quarx has joined #linux-sunxi
<Turl>
ssvb_: it cannot even use a linked list
physis has quit [Remote host closed the connection]
<Turl>
you can add ethernet on USB - most tabs have a host port
Black_Horseman has joined #linux-sunxi
Black_Horseman has joined #linux-sunxi
<Turl>
the sunxi netbooks have ethernet too
<Black_Horseman>
hola
<Turl>
ohai
avsm has joined #linux-sunxi
indy has joined #linux-sunxi
jemk has quit [Quit: leaving]
akaizen has quit [Remote host closed the connection]
akaizen_ has joined #linux-sunxi
<CaptHindsight>
the ethernet is for real time machine control, so you can't use ethernet on USB, so that might still be the problem with ethernet on an ARM soc tablet
<CaptHindsight>
if they used a USB ethernet MAC vs on the ARM SOC
kivutar has joined #linux-sunxi
akaizen_ has quit [Remote host closed the connection]
<Turl>
CaptHindsight: IP is already best-effort, is it really used for machine control?
akaizen_ has joined #linux-sunxi
kelvan has joined #linux-sunxi
<CaptHindsight>
Turl: yes, it works well, but it needs a dedicated point to point connection
<CaptHindsight>
a tablet is a low cost way to have a user interface to a machine controller
<Turl>
CaptHindsight: you could get one of those netbooks with sunxi SoCs
<CaptHindsight>
and the A20 is fast enough to handle both real time muilti-axis synchronized motion control and the UI
<Turl>
and get a keyboard on it,
<CaptHindsight>
I have one of the first with the A10, the mainboard was held on place by two sided foam tape :)
<CaptHindsight>
maybe we can use the onchip ethernet for real time and use an extra USB ethernet dongle for internet
<CaptHindsight>
can ARM-based systems handle atomic stores to volatile doubles? Looks like NO
<CaptHindsight>
arm does not guarantee atomic updates of larger-than-32-bit variables, and we are currently using 64-bit floating point numbers in our application
<CaptHindsight>
the only way to get atomic 64-bit memory accesses on ARM is to use ldrex/strex with a loop (ldrd and strd instructions are not guaranteed to appear atomic).
<CaptHindsight>
some arms have atomic 64 bit ops (strexd) but are unlikely to generate them for mere stores to volatile doubles.
<CaptHindsight>
anyone else run into this when porting apps from x86 to ARM and what they did for fixes?