Building LeMaker Android 4.2 from Source

129 31179
liab  
The reason for the fool cpu core is to search in lichee/linux-3.4. I used the linux-3.4 from the good matthuisman/android an now the second cpu is also in lemaker busy.

root@android:/ # [   24.092414] init: untracked pid 1546 exited
mpstat -P ALL
Linux 3.4.39 (localhost)        12/05/14        _armv7l_        (2 CPU)

18:12:06     CPU    %usr   %nice    %sys %iowait    %irq   %soft  %steal  %guest   %idle
18:12:06     all   14.23    0.00    7.88   44.04    0.00    0.86    0.00    0.00   32.99
18:12:06       0    9.21    0.00    8.27   31.36    0.00    1.73    0.00    0.00   49.44
18:12:06       1   19.25    0.00    7.50   56.67    0.00    0.00    0.00    0.00   16.59
root@android:/ #

mattrix  
Edited by mattrix at Fri Dec 5, 2014 13:10

So, Lemaker Source is OK then?


Oh, I see. You swapped out the Kernel and then it worked.

liab  
Edited by liab at Fri Dec 5, 2014 14:27

Unfortunately it's not useful to take the other linux, the lemaker image remains slow.
Streamig with vu+ in hd will crash after a short time.

The second in this compilation is, LAN seems ok, but will not work. *) It has a MAC address. WLAN works.

So the best image for me is matthuisman/android.

*) Edit: A ping via LAN is ok, but starting an app needing network makes an error.

ChrisP  
Edited by ChrisP at Sat Dec 6, 2014 18:48

Here is the result I had from Antutu Benchmark on various image. The first one is the first version from Mattrix derived from the official Android 4.2.2. The second one is  a test on image generated from LeMaker SDK. The third was on image generated from SDK2. The last one was the one done after I believed to have successfully implemented r5p0 mali driver.



  
memory
  
  
integer
  
  
float
  
  
score2d
  
  
scored3d
  
  
database
  
  
sram
  
  
svm
  
  
srs
  
  
smt
  
  
snand
  
  
ints
  
  
floats
  
  
score
  
  
date
  
  
cpu
  
  
model
  
  
584
  
  
554
  
  
694
  
  
710
  
  
2538
  
  
445
  
  
988
  
  
798
  
  
0
  
  
2059
  
  
411
  
  
703
  
  
761
  
  
11245
  
  
2014-11-28 02:40
  
  
1056 MHz (x2)
  
    
  
567
  
  
543
  
  
671
  
  
671
  
  
2393
  
  
325
  
  
900
  
  
750
  
  
0
  
  
1946
  
  
427
  
  
699
  
  
769
  
  
10661
  
  
2014-12-04 16:52
  
  
1056 MHz (x2)
  
    
  
506
  
  
478
  
  
598
  
  
698
  
  
2587
  
  
405
  
  
961
  
  
676
  
  
0
  
  
1644
  
  
352
  
  
606
  
  
663
  
  
10174
  
  
2014-11-28 01:54
  
  
1056 MHz (x2)
  
    
  
501
  
  
475
  
  
595
  
  
787
  
  
2834
  
  
435
  
  
1099
  
  
683
  
  
0
  
  
1777
  
  
303
  
  
606
  
  
663
  
  
10758
  
  
2014-11-28 07:11
  
  
1056 MHz (x2)
  
    
5775466667752785325101076701858467700769112452014-12-07 01:451056 MHz (x2)Build 4.2 from  LeMaker SDK\bpi_android-RAM480.img

ChrisP  
Edited by ChrisP at Sat Dec 6, 2014 00:06

I think I found the most important part that explain different behavior regarding memory/integer/float:

LeMaker sources have this value in fex file:
dram_clk            = 432

where SDK 2 had:
dram_clk            = 480

But I don't get where could the video increase in my last SDK 2 image come from. I'll republish a fixed version for investigation.

It was maybe a little precipitated to remove old SDK2 git repository, it could have been usefull for investigation :/

liab  
Edited by liab at Fri Dec 5, 2014 19:10

Just I found out, a fresh burned lemaker image works in fact with 2 CPU's. But after 1 minute
after it automatic installs something one CPU is lost! What is that?

root@android:/ # mpstat -P ALL                                                
Linux 3.4.39 (localhost)        12/06/14        _armv7l_        (2 CPU)

00:14:05     CPU    %usr   %nice    %sys %iowait    %irq   %soft  %steal  %guest   %idle
00:14:05     all   18.68    4.51   15.46   30.30    0.00    1.94    0.00    0.00   29.10
00:14:05       0   16.36    3.41   18.01   21.70    0.00    3.81    0.00    0.00   36.71
00:14:05       1   21.08    5.64   12.84   39.19    0.00    0.01    0.00    0.00   21.23
root@android:/ # mpstat -P ALL                                                
Linux 3.4.39 (localhost)        12/06/14        _armv7l_        (1 CPU)

00:15:27     CPU    %usr   %nice    %sys %iowait    %irq   %soft  %steal  %guest   %idle
00:15:27     all   17.46    4.40   14.33   31.06    0.00    1.95    0.00    0.00   30.80
00:15:27       0   15.61    3.44   16.85   22.10    0.00    3.84    0.00    0.00   38.14
root@android:/ #

Edit: Now I saw suddenly there are 2 CPU's again. Perhaps it have something to do with
the powermanagement. If nothing to do 1 CPU and if it's needed 2 CPU's are working.

ChrisP replied at Fri Dec 5, 2014 17:00
I think I found the most important part that explain different behavior regarding memory/integer/flo ...

In order to be more stable, I set the dram frequency to 432MHz. Though the 480 can result better performance.

Edited by tony_zhang at Fri Dec 5, 2014 21:33
liab replied at Fri Dec 5, 2014 18:17
Just I found out, a fresh burned lemaker image works in fact with 2 CPU's. But after 1 minute
after  ...


I think the Android power management auto select cpu.

ChrisP  
Just updated the benchmark list, which show that both cores are working fine with this SDK. Going to 480 MHz allow to go back to hugher performance of the best past images.

But 1080p is still no go

dlanor  
I'm not sure exactly what you mean by 'no go', but it is an unavoidable speed problem that 1080p  involves 2.25 times the amount of pixels per screen, as compared to 720p. And thus it also involves 2.25 times the amount of screen RAM to be processed in each full screen update. That will always have a huge impact on overall speed, reducing it drastically. So 720p is probably the highest resolution that will give us satisfactory speed.

Best regards: dlanor

You have to log in before you can reply Login | Sign Up

Points Rules