The Raspberry Pi 2 Q&A thread

[tvjon]

paddr=3f000000, baddr=C0000000

I'm afraid that means there has been another change.

DMA is not working as it used to work.

May be of interest?

http://infocenter.arm.com/help/index.js ... index.html

[dom]

Ref: Jazelle on Pi1, afaik, this is not enabled, and therefore unused. I think its a licensing issue rather than anything technical, but not sure.

I don't believe there's any licensing issue. I believe a half decent jit outperforms Jazelle, so no one supports Jazelle.

[plugwash]

AIUI sun/arm never released the specs for jazelle so no opensource vm can use it. The oracle VM we use on the pi doesn't either.

I think the only things that used it were some old J2ME VMs on phones.

[guzunty]

I am seeing a difference in behaviour of usleep between a B+ and Pi2.

Using a bitscope on non pi host I see 1ms on B+ but around 3ms on Pi2.

This is with a call to usleep(1000).

Has anyone else seen this or have I missed something?

[EDIT] Correction, I see 1ms on B+ and 28ms(!) on Pi2.

[solar3000]

Single-threaded compute performance for tight loops (i.e. C or assembly routines) would be 1.8-2x faster. Compute performance for stuff with a larger memory footprint (interpreted languages like Java, Javascript, Ruby/Python) will benefit proportionally more due to the L2 cache.

How do you get those numbers? 700MHz vs 900MHz isn't that different. What else would be making single threads go 2x?

[croston]

I am seeing a difference in behaviour of usleep between a B+ and Pi2.

Using a bitscope on non pi host I see 1ms on B+ but around 3ms on Pi2.

This is with a call to usleep(1000).

Has anyone else seen this or have I missed something?

[EDIT] Correction, I see 1ms on B+ and 28ms(!) on Pi2.

This is the reason I have not released a new RPi.GPIO yet. I've not had time to investigate yet. Has anyone got any ideas?

[RaTTuS]

thread locking via different CPU cores ?

[solar3000]

So far no one has asked about youtube.
Pis A/B run youtube at one frame every 60 seconds.
How does pi 2 do?

[jdb]

Single-threaded compute performance for tight loops (i.e. C or assembly routines) would be 1.8-2x faster. Compute performance for stuff with a larger memory footprint (interpreted languages like Java, Javascript, Ruby/Python) will benefit proportionally more due to the L2 cache.

How do you get those numbers? 700MHz vs 900MHz isn't that different. What else would be making single threads go 2x?

The main reason is that Cortex A7s have dual-issue instruction parallelism - Superscalar execution (but still in-order execution). Two instructions can be executed per clock cycle as long as there are no register or load/store dependencies on the preceeding instruction(s). Compilers targeting the core know this and optimise instruction scheduling to a certain extent. This plus the clock speed differential gives you ~2x for single-threaded and tight loops.

There are other things that are "better" - larger TLBs for example.
http://infocenter.arm.com/help/index.js ... diifa.html
vs
http://infocenter.arm.com/help/index.js ... DIJBD.html

[guzunty]

thread locking via different CPU cores ?

I didn't start any threads myself. I used Gert and Doms demo program here:

http://elinux.org/RPi_Low-level_peripherals#C_2

to make the measurements.

I just changed the sleep(1) to usleep(1000) and modified for the new peripheral base address.

Of course, something else could be contending for the GPIO or the timer.

[jdb]

thread locking via different CPU cores ?

I didn't start any threads myself. I used Gert and Doms demo program here:

http://elinux.org/RPi_Low-level_peripherals#C_2

to make the measurements.

I just changed the sleep(1) to usleep(1000) and modified for the new peripheral base address.

Of course, something else could be contending for the GPIO or the timer.

We moved both Pi1 and Pi2 to a 3.18 kernel base. There's nothing telling in the kernel configs between each platform that would result in timer discrepancies.

Can you run the same test on a Pi1 that is fully up to date?

[Leopold]

What is the meaning of the Revision shown in /proc/cpuinfo? For past models the final four hex digits were the revision number as indicated at http://elinux.org/RPi_HardwareHistory#B ... on_History but for the 2B it seems there is now more to it?

[gkreidl]

So far no one has asked about youtube.
Pis A/B run youtube at one frame every 60 seconds.
How does pi 2 do?

Run web videos using the GPU (either with epiphany/kweb3 using gstreamer-1.0 or with kweb(3) + omxplayer(GUI)). They will play with full speed (and even full screen with the second solution). That works on all Pis. I don't believe, that the ARM7 can decode them by software alone.

[rurwin]

Ref: Jazelle on Pi1, afaik, this is not enabled, and therefore unused. I think its a licensing issue rather than anything technical, but not sure.

I don't believe there's any licensing issue. I believe a half decent jit outperforms Jazelle, so no one supports Jazelle.

The published specifications are very incomplete, being only sufficient for writing operating system code that can support a JVM that uses Jazelle[citation needed]. The declared intent is that only the JVM software needs to (or is allowed to) depend on the hardware interface details. This tight binding facilitates that the hardware and JVM can evolve together without affecting other software. In effect, this gives ARM Holdings considerable control over which JVMs are able to exploit Jazelle[citation needed]. It also prevents open source JVMs from using Jazelle. These issues do not apply to the ARMv7 ThumbEE environment, the nominal successor to Jazelle DBX.

That is also the answer I received way back in 2012 when I asked about an optimised JVM.

Jazelle v1 Architecture Reference Manual
This is a placeholder for a restricted document that is not available from this site. Please contact ARM for more information.

[solar3000]

So far no one has asked about youtube.
Pis A/B run youtube at one frame every 60 seconds.
How does pi 2 do?

Run web videos using the GPU (either with epiphany/kweb3 using gstreamer-1.0 or with kweb(3) + omxplayer(GUI)). They will play with full speed (and even full screen with the second solution). That works on all Pis. I don't believe, that the ARM7 can decode them by software alone.

yes, I've run youtube and youtube-dl.
I meant in a browser without special tools, just flash or html5.

[gadgetoid]

I'm also seeing timing issues on the Pi 2, I've boiled them down to this example which takes 20sec! on a Pi 2

This affects usleep, and subsequently Python's sleep command, which makes code like our Display-o-Tron 3000 examples run in sloooow moootion.

Code: Select all

#include  <unistd.h>

int main()
{
 int x = 0;
 for(x = 0; x < 1000; x++){
   usleep(1000);
 }
}

And when timed:

Code: Select all

pi@raspberrypi ~ $ time ./a.out

real    0m20.094s
user    0m0.000s
sys     0m0.020s

Same SD card placed in a +B:

Code: Select all

pi@raspberrypi ~ $ time ./a.out

real    0m1.095s
user    0m0.000s
sys     0m0.020s

[gadgetoid]

What is the meaning of the Revision shown in /proc/cpuinfo? For past models the final four hex digits were the revision number as indicated at http://elinux.org/RPi_HardwareHistory#B ... on_History but for the 2B it seems there is now more to it?

It's a bitfield that contains more information such as memory amount, processor, Pi model, manufacturer and revision.

I've whipped up a Python module to read it: https://github.com/pimoroni/RPi.version

[gkreidl]

yes, I've run youtube and youtube-dl.
I meant in a browser without special tools, just flash or html5.

No flash video (except with the help of youtube-dl). HTML5 video should profit from the better CPU (often still a bit shaky in the browser right now), but only in a browser that uses GPU HW acceleration (epiphany/kweb3).

[RaTTuS]

What is the meaning of the Revision shown in /proc/cpuinfo? For past models the final four hex digits were the revision number as indicated at http://elinux.org/RPi_HardwareHistory#B ... on_History but for the 2B it seems there is now more to it?

Code: Select all

32 bits
NEW                   23: will be 1 for the new scheme, 0 for the old scheme
MEMSIZE             20: 0=256M 1=512M 2=1G
MANUFACTURER  16: 0=SONY 1=EGOMAN
PROCESSOR         12: 0=2835 1=2836
TYPE                   04: 0=MODELA 1=MODELB 2=MODELA+ 3=MODELB+ 4=Pi2 MODEL B 5=ALPHA 6=CM
REV                     00: 0=REV0 1=REV1 2=REV2

pi2 = 1<<23 | 2<<20 | 1<<12 | 4<<4 = 0xa01040

[pi-man-uk]

Hi, just a question on the upgrade process, I’m planning to transition my main server from a model b to my new pi2 tonight. I’ve just apt-get updated/upgraded and noticed /boot got full up of new kernels etc on the pi1, rebooted it and it came back - so I figure I’m good to go. Sooooo, next step, I plan tonight to shut it down, stick the 4 gig full size SD card in my macbook and use dd to create an image file from it on the mac, then I plan to use dd again to write that image file to a 4 gig class 10 *micro SD* card – stick this in the new pi2, boto and it’s off to the races. Will it work?

[DirkS]

Will it work?

Yes, but don't blame me if it doesn't

[Ste123]

FYI - this may help others.

I had a problem booting with my new Pi2B+. The boot would only go so far before it hung. I had run the update command etc,

I finally worked it out I had to boot into the old B+ and reset the overclocking to Normal using raspi-config. Only after doing this could I boot into the Pi2 (which I then overclocked to 1GHz with no problems)

Running fine now - uptime just over 24 hours so far....

[Massi]

And when timed:

Code: Select all

pi@raspberrypi ~ $ time ./a.out

real    0m20.094s
user    0m0.000s
sys     0m0.020s

Same SD card placed in a +B:

Code: Select all

pi@raspberrypi ~ $ time ./a.out

real    0m1.095s
user    0m0.000s
sys     0m0.020s

O_O

[pi-man-uk]

Will it work?

Yes, but don't blame me if it doesn't

ok thanks, I like your answer, I'll go for it. Then I need to swap out all my other Pi with Pi2! (mostly openelec) now I have found the promised land I can't sleep at night with the old model, need lots of 2amp PSUs and new cases also, as well as microSD cards... oh gawd...

[jamesh]

And when timed:

Code: Select all

pi@raspberrypi ~ $ time ./a.out

real    0m20.094s
user    0m0.000s
sys     0m0.020s

Same SD card placed in a +B:

Code: Select all

pi@raspberrypi ~ $ time ./a.out

real    0m1.095s
user    0m0.000s
sys     0m0.020s

O_O

Did we mention it was faster?