[KERNEL] [Y300/G510] Stock Huawei

[Guest]

Well, the changes to OnDemand definitely got an impact on scores:

 

Before it behaved like this:

• idle? -> scale to MIN frequency
• any kind of load? -> immediately squeeze anything out of the CPU it got and bump everything to max, no matter how much energy it does cost

Now it does: scale better through all available frequency ranges, which leads to an more frequent selection of frequencies below MAX.

 

Before: it only jumped between 320 MHz and 1008 MHz

Now: frequencies 480 MHz and 700 MHz are getting used as well

 

I'm trying to optimize the scaling algorithm(s) a bit.

 

Here's an screenshot of the frequency utilization AFTER those changes:

 

Edited July 24, 2014 by Guest

[Guest luca020400]

Good work !!

 

I can't see the screen 

Edited July 24, 2014 by luca020400

[Guest]

I can't see the screen

You need to click on "spoiler"

EDIT: you are right. On my mobile browser it doesn't get shown.

Edited July 25, 2014 by Guest

[Guest]

Update of yesterday's TEST_BUILD: http://www26.zippyshare.com/v/41036313/file.html -> http://www1.zippyshare.com/v/3102165/file.html

 

• commented out 1 line of the new ondemand governor (which is supposed to downscale the frequency faster)
• compressed the kernel once again with LZ4 (I still don't notice any difference during boot)
• removed 2 commits
• changed kernel's config a tiny bit

 

This build seems quite responsive and first bench runs seems to perform better than yesterday - even with the new OnDemand. I still fear the new algorithm for target frequency determination causes l0wer benchmark scores though. They are not back at the level they have been once.

 

 

(same build compiled with GCC 4.6.4 -> http://www44.zippyshare.com/v/84522078/file.html ; seems like 3D scores are better on this GCC version. in my oppinion GCC 4.9.1 is still the better choice)

 

 

 

EDIT: In the updated version

• PMEM got removed
• reduced MAX charging voltage from 4350 mV to 4300 mV to increase battery longevity and to prevent overcharging
• enabled 2nd ZRAM device (total number of ZRAM devices is now: 2)
• some minor config/code changes not worth mentioning

Edited July 26, 2014 by Guest

[Guest fonz93]

If i want to use an init.d script to start frandom at boot i have to write only "start frandom"?

[Guest]

Yes, because on this kernel build there's already an frandom-script included within the ramdisk.

 

Optionally you can use any other frandom-script of your choice, too. It doesn't matter. It might the more safe to use "start frandom" in case I'll auto-enable frandom one day again :P

[Guest fonz93]

I remember that you posted (or Victod maybe) an executable command in terminal emulator to see if it is working, can you link that? i can't find it, thank you

Edited July 24, 2014 by fonz93

[Guest]

If it worked you will have 2 new files under /dev :

/dev/random.orig

/dev/urandom.orig

[Guest jor1196]

and when using the gcc maclaw?

[Guest Romagnolo1973]

No issue here with this new beta Gcc 4.9.1 Ondemand + sio

every new version is a step approaching the perfection 

[Guest fonz93]

If it worked you will have 2 new files under /dev :

/dev/random.orig

/dev/urandom.orig

 

i have random, random.orig, urandom, urandom.orig, it's ok?right?

 

 

With this kernel version, i have a problem of no "enough space available" (SCREENSHOT) (6 mb only) in /data, but i have 350 mb left, this happens only if i start frandom at boot

 

i simply added this line:

 

# Frandom

start frandom;

 

in my init.d script, where i am wrong?

 

 

Edit: After 2 reboots, the problem just disappeared and now it seems that "start frandom" via init.d works, really strange

Edited July 24, 2014 by fonz93

[Guest]

@jor: I assume "mclaw" is a nickname? If this should be the case: he seems to use linaro GCC 4.7.4

[Guest jor1196]

@jor: I assume "mclaw" is a nickname? If this should be the case: he seems to use linaro GCC 4.7.4

 

NovaTp GCC 4.9

[Guest luca020400]

Update of yesterday's TEST_BUILD: http://www26.zippyshare.com/v/41036313/file.html

 

• commented out 1 line of the new ondemand governor (which is supposed to downscale the frequency faster)
• compressed the kernel once again with LZ4 (I still don't notice any difference during boot)
• removed 2 commits
• changed kernel's config a tiny bit

 

This build seems quite responsive and first bench runs seems to perform better than yesterday - even with the new OnDemand. I still fear the new algorithm for target frequency determination causes l0wer benchmark scores though. They are not back at the level they have been once.

 

 

(same build compiled with GCC 4.6.4 -> http://www44.zippyshare.com/v/84522078/file.html ; seems like 3D scores are better on this GCC version. in my oppinion GCC 4.9.1 is still the better choice)

Thanks 

Only a question : I tried to build your kernel in mine CarbonRom 4.2.2 with GCC 4.6.4 but got a lot of error

How have you compiled this kernel with this gcc version ?

Thanks 

[Guest]

By supressing all warnings about uninitialized variables. Most of the time it's a false positive.

I don't have any other kinds of errors. Everything got fixed.

[Guest SH3H1]

Is 118MB ZRAM (Disksize?) too much?? :Huh:

@Fonz93

Can you make a script with this? Since the app halfs the disksize 1/2 every reboot :wacko:

[Guest SH3H1]

-_- Pissed with UC !

Edited July 25, 2014 by SH3H1

[Guest]

NovaTp GCC 4.9

 

found it: https://github.com/NovaFusion/novatp_arm-eabi-4.9

 

But I'm not sure yet why it should be different?

[Guest]

Is 118MB ZRAM (Disksize?) too much??

 

It shouldn't be more than 10% of total RAM available in general. For our device I wouldn't recommend more than 50 - 75 MB totally.

[Guest SH3H1]

It shouldn't be more than 10% of total RAM available in general. For our device I wouldn't recommend more than 50 - 75 MB totally.

Okay thank you...!!!

So i merely ask fonz to make one with 75MB

I wan't maximum possible since my phone slows down whenever the ZRAM block is full :O

[Guest SH3H1]

Never mind....!!

Fonz i don't wanna bother ya...there is already a script :)

Edited July 25, 2014 by SH3H1

[Guest]

Which kernel build (date) are you using? You shouldn't expect any slow-downs

[Guest barnir]

What build do you think to be the best atm?

[Guest]

What build do you think to be the best atm?

 

The updated build of today: http://www1.zippyshare.com/v/3102165/file.html

Edited July 26, 2014 by Guest

[Guest]

Some technical information about the RAM usage on our phone.
 
My device got an total amount of 463 MB RAM.
 
A part of it is getting statically allocated for our hardware:
 

MDP (video/camera): 28 MB
ADSP (QDSP audio/multimedia chip): 18 MB
Audio: 2 MB
"Unknown" (don't know yet): 2 MB
---
= 50 MB "dead" memory not useable for anything else

 
 
 
This leaves 413 MB as maximum available for the kernel, which is getting split into:
 

Memory block #1: 251 MB
Memory block #2: 0 MB
Memory block #3: 1 MB
Memory block #4: 160 MB
 
---
 
Number of free pages: 393 MB
Reserved pages (for buffers and other stuff): 70 MB
Highmem pages: 0 MB

This leaves a real amount of 323 MB memory for Android.


It's possible to reduce the size of the MDP & ADSP buffers. This would increase the amount of free RAM at cost of functionality. We could only experimentally try how low we can go.

Edited July 26, 2014 by Guest