Linaro seem to done some amazing optimisation - Android General

So it would seem that Linaro have done some toolchain optimisation and come up with a version of 4.0.4 AOSP that runs significantly faster than stock AOSP. I searched a little and found all the threads that say that optimised toolchains don't actually do anything, but this seems to be definitive proof that they do.
Here's the link

Related

New 200 line kernel patch in the news

Curious if any of the nuts and bolts kernel devs have comments on this new kernel patch?
http://www.pcworld.com/businesscent...x_kernel_patch_delivers_huge_speed_boost.html
What a difference 233 lines of code can make. That's the size of a small new patch to the Linux kernel's scheduler that has been found to reduce the average latency of the desktop by about 60 times. It's a small patch with a really, really big gain for desktop users of the open source operating system, in other words.
Click to expand...
Click to collapse
Actual patch:
http://lkml.org/lkml/2010/10/19/123
What a blessing this would be...
Someone submitted patch to Cyan...
Doesn't look good though..
http://review.cyanogenmod.com/#change,417
I actually added the changes to ezterry's kernel last night and left it compiling this morning on my way out the door. My intution says this change wont help much, I already saw a ton of cgroup related code in the kernel... Never know until you try though.
Edit: After reading all the comments I probably won't even try booting the kernel at this point, lol!
The idea behind the patch is that it doesn't increase speed, just usability, which is something Android certainly needs. This was something tested on a 4 core CPU on Linux, so I don't know how much of a benefit it'll be for Android platforms.
Really, using BFS or BFQ scheduler would increase performance more on our phones. Sadly, neither is really stable, and cause memory leaks (on our phones at least)
Why isn't Jit stable yet? It's always disabled, and considered bad to enable.
Jit is actually stable for android. just not used on the G1 and magic because it uses to much memory compared to the bennifits.
on devices with larger RAM it is actually working quite well

Linaro Toolchain

Hey Guys, a question:
With all this hoo-hah about the linaro toolchain improving performance, I was wondering what it is exactly and whether it is possible to somehow make it a flashable zip so we can all flash it to our phones and live life in the fast lane lol.
Thoughts? Ideas? Info? Anyone done it?
It's set when compiling the source, it cannot be added on afterward. The ROM/kernel itself is either built with it or with something else. So far there has been little success performance wise in comparison to the claims, however some kernels and ROMs have been built with it (check out the dev section). Don't expect mind boggling performance out of anything claiming so because of linaro.
cheers mate, cleared many misconceptions. how good was state of origin aye?
No worries
I've been going through exams here so didn't get a chance to watch game 2. I should check out the highlights.

[KERNEL][AOSP4.4/5.1/6.0/7.1] dkp - d2vzw & d2usc - 2/4/18

Welcome to decimalman's kernel playground!​
As the name suggests, dkp is a hodgepodge of features and tweaks that I wanted to play with. It should get excellent battery life without feeling sluggish. It doesn't come with its own tuner app, so pick your favorite. Personally, I like Trickster MOD and Kernel Adiutor, so I go out of my way to make things work in them. Most other apps should work, too.
Features:
Overclocking up to 2.1 GHz, but you'll need to increase your voltages to get there (if you can get there at all)
Underclocking down to 54 MHz, with stability improvements
Undervolting compatible with most apps
Fast charge without unplugging first
Glorious animations for the notification and softkey LEDs
Well-integrated erandom means you don't need CrossBreeder or Seeder (recent AOSP builds use ISAAC instead)
freelunch and tierservative governors for optimal battery life without sacrificing responsiveness
Automatic mpdecision and auto-hotplug are only enabled when needed
Adjustable minimum voltage for stability on finicky processors
Optimized UKSM to free up some extra memory
Code optimizations for size and speed
Compiler optimizations (-O3, LTO, and more) because faster is better
Donors: Thanks, everyone! Your generosity is much appreciated. :good:
drpenguino, 0xScott, vmancini3 (twice! :good, Ch4m3l30n, rompnit, Mystique, ryandubbz, techdog, ElwOOd_CbGp, ScOULaris, ZipAddict
Remember:
Nandroid!
last_kmsg and/or logcat or it didn't happen.
Other kernels have their own threads or forums. Discuss them there.
Image dumps (settings, battery life, whatever) belong inside [HIDE][/HIDE] (that's HIDE, if you're on the mobile app) tags.
Be silly. We're here to have fun.
Installation:
Reboot to recovery. I recommend that one recovery...you know, the one that flashes zips? I forget what it's called.
Flash dkp. Optionally, rename and flash dkp-vmin-XXX.zip (see below).
Reboot.
Undervolting:
Undervolting on dkp is more complex than other kernels. Some processors get unstable at lower voltages, so (like the stock kernel) dkp keeps the processor voltage above 1150 mV by default. I refer to this limit as the minimum voltage. In order to undervolt, you'll need to lower the minimum voltage: if you use Trickster MOD or Kernel Adiutor, just disable "Override Minimum Voltage", otherwise rename dkp-vmin-XXX.zip to e.g. dkp-vmin-600.zip (which would apply a 600 mV minimum voltage) and flash it. If this causes instability (crashes, audio/video glitches, etc.), try using dkp-vmin-XXX.zip to apply a higher minimum voltage (somewhere between 950 and 1050 mV seems to work well for most people).
Downloads:
MediaFire:
All Downloads
dkp-vmin-XXX.zip
Solidfiles (Make sure you have an adblocker!):
All Downloads
dkp-vmin-XXX.zip
Source: I'm always happy to see my code used, so cherry-pick away. I'll even put together feature patches if you ask nicely.
Bugs:
Let me know.
Stable changelog:
3/3/13: Initial release for d2spr. Didn't get around to making threads for other carriers.
4/8/13 (3.0):
FauxSound support
Strip more useless stuff
A few bonus optimizations
4/8/13 (3.4):
Port everything except erandom from 3.0
Enhance cpufreq for easier configuration
4/24/13 (3.4):
Bugfixes: better support for tuner apps, fixed potential SOD bugs, automatic mpdecision fixups, etc.
Lots of CM/CAF/Linux updates
Working AssWax governor
Trinity colors support
sio, zen I/O schedulers
erandom is back!
Built with a super-fancy Linaro GCC 4.8.1-dev compiler toolchain for maximum -O3 goodness
Probably lots more, but there's hundreds of commits to sort through...
5/29/13 (3.4):
Bugfixes: better overclocking support, better hwrng support, etc.
Updates: new CM updates, Linux 3.4.47, updated FauxSound driver, added invisiblek's new panel colors interface
Automatic auto-hotplug
New optimizations, including link-time optimization and an updated GNU+Linaro GCC 4.8.1-dev toolchain
6/14/13 (3.4):
Bugfixes: fix several critical bugs in the 5/29 release.
9/7/13 (3.4):
Fixes for OC, UV, auto-hotplug.
A few new optimizations.
Synced up with CM.
9/20/13 (TW):
Ported everything from AOSP to TW.
9/20/13 (4.3):
Merged 4.3 from CM into the existing 4.2 code.
Current experimental branches:
Nothing interesting at the moment.
Goodies:
dkp doesn't come with its own splash screen. However, the dkp installer (i.e. the install zip) is smarter than you think, and can apply a custom splash screen for you. Here's how:
Create a folder on your internal storage named "dkp"
Copy a PNG image into the directory, and rename it "splash.png". Alternatively, copy an RLE image (i.e. from a flashable custom splash screen zip) and rename it "splash.rle". Ideally, the image should be roughly 1280x720 to begin with, since it won't be resized.
The image will be used as your splash screen whenever you flash dkp. Reflash to apply initially.
mikedavis120 has put together a how-to video that covers tweaking dkp for optimal battery life. If you're new to dkp, take a look! He also put together a zipped collection of apps that will come in handy while tuning dkp. It also includes a flashable zip, "dkp-debug_v1.zip". After flashing it, running
Code:
su
dkp
from a terminal emulator will collect lots of useful debug information that will make it much easier for me to track down the issue you're having. :good: mikedavis120 recommends installing SuperSU (included in the zip) instead of what's included in you ROM.
sysfs:
It's possible to adjust all the settings available in dkp without using apps. Because they show up as files, settings can be adjusted with file managers, terminal emulators, adb and initscripts. Here's the most interesting files inside sysfs:
/sys/devices/platform/mipi_samsung_oled.513/lcd/panel/panel_colors (not available on newer AOSP builds): display tint (0 = very red, 2 = default, 4 = trinity colors)
/sys/class/misc/gammacontrol (only available on newer AOSP builds): various color controls. See this post for details on enabling Trinity colors on builds that use these controls.
/sys/devices/system/cpu/cpu<N>/cpufreq/UV_mV_table: voltage table
/sys/devices/system/cpu/cpu<N>/cpufreq/scaling_...: scaling_governor is the governor, scaling_min_freq and scaling_max_freq are the minimum and maximum frequencies, scaling_available_governors and scaling_available_frequencies show the available governors and frequencies
/sys/kernel/dkp/force_fast_charge: fast charge
/sys/kernel/dkp/link_core_settings: when linked (the default), frequency settings and some governors are automatically copied to the other core
/sys/kernel/dkp/vmin: minimum processor voltage in mV
/sys/kernel/mm/uksm/run: activate UKSM
auto-hotplug tuners:
These show up in the governor settings for any governor that doesn't do its own hotplugging. They only take effect when using auto-hotplug, so you'll probably need to disable mpdecision in Trickster.
hotplug_intpulse: when set to 1, automatically turns core 2 on whenever the screen/buttons/whatever is pressed. Default is 0.
hotplug_sampling_periods: number of samples to use for average number of running tasks. Default is 15.
hotplug_sampling_rate: number of 'jiffies' (currently 1 jiffy = 10 ms) between each sample of running tasks. Default is 20 (0.2 sec).
hotplug_enable_one_threshold: the average number of running tasks required to turn core 2 on, multiplied by 100. Default is 125 (1.25 tasks on average).
hotplug_disable_one_threshold: the average number of running tasks required to keep core 2 on, multiplied by 100. Default is 250 (2.5 tasks on average).
freelunch/nanolunch tuners:
freelunch and nanolunch aren't materially based on other governors, so their configuration is quite different than other governors. There's lots of tuners, since I haven't really decided on an ideal tuning. I encourage experimentation! I'll explain a bit of how these governors work before actually listing the tuners.
Generally speaking, there are two modes: in "normal" mode, sampling is done occasionally and frequency is generally increased slowly; in "interactive" mode, sampling is done much more quickly, and frequency increases much more quickly. "Interactive" mode ends after several samples of very low usage. The idea of a "hispeed" frequency is used in lots of governors, and it refers to the frequency that the CPU will jump to when more CPU usage is needed; generally, it's a generous estimate of how much CPU will be needed. Here, the hispeed frequency is adjusted on-the-fly, increasing when more CPU is needed and gradually decreasing when the CPU is idle. In "interactive" mode, the hispeed frequency is kept fairly high so that everything will feel snappy.
Hotplugging is taken care of in the least complicated (and in my opinion, most reasonable) way possible: if core 1 is using lots of CPU, and there are several tasks running (in other words, if it's likely that core 2 will have something to do), core 2 is turned on; if either core isn't doing much except using power, core 2 is turned off.
sampling_rate: the usual
hotplug_up_cycles: number of consecutive heavily-loaded samples before core 2 is turned on
hotplug_down_cycles: number of consecutive lightly-loaded samples before core 2 is turned off
hotplug_up_load: number of running tasks required to bring core 2 online
hotplug_up_usage: number of used CPU cycles (in thousands per second) required to bring core 2 online
hotplug_down_usage: number of used CPU cycles (in thousands per second) required on both cores to keep core 2 online
overestimate_khz: number of CPU cycles to overshoot usage by in "normal" mode
hispeed_thresh: if CPU usage is within this many cycles (in thousands per second) of the maximum frequency, frequency will be increased to the hispeed frequency. Generally, hispeed is pretty low in "normal" mode, and fairly high in "interactive" mode.
hispeed_decrease: when the CPU is sitting idle, the hispeed frequency is decreased by this amount each sample (this isn't ideal, but it works)
interaction_hispeed: the initial hispeed frequency when switching to "interactive" mode
interaction_return_cycles: number of consecutive lightly-loaded samples before returning to "normal" mode
interaction_return_usage: number of used CPU cycles (in thousands per second) required to stay in "interactive" mode
interaction_panic (nanolunch only): when set to 1, allows aggressively jumping past the current hispeed frequency under some circumstances
interaction_sampling_rate/overestimate_khz: equivalent to the "normal" versions of the tuners, these take effect in "interactive" mode
This looks great - especially excited about your custom governor! Thanks!!
Sent from my SCH-I535 using xda app-developers app
Glad we have another kernel dev. But do you plan on releasing a tw kernel? I run tw 98% of the time and would love to try it out. Thanks for your work regardless
PsiPhiDan said:
This looks great - especially excited about your custom governor! Thanks!!
Sent from my SCH-I535 using xda app-developers app
Click to expand...
Click to collapse
freelunch is absurdly simple, but does its job well. Let me know what you think. Happy flashing!
aypeeootrek said:
Glad we have another kernel dev. But do you plan on releasing a tw kernel? I run tw 98% of the time and would love to try it out. Thanks for your work regardless
Click to expand...
Click to collapse
I don't run TW, so I don't have any plans to release a TW kernel. If there's enough interest, I suppose I could get to work on one though.
Interesting. Looks cool man, will check it out
Sent from my SCH-I535 using Tapatalk 2
I'm excited to give this a shot!
Sent from my SCH-I535 using xda app-developers app
flashaholic commencing flash!
Anyone using trickster mod with this kernel?
Sent from my SCH-I535 using Tapatalk 2
Love this kernel! Freelunch is amazing
Sent from my SCH-I535 using xda premium
This kernel run pretty good!!
Sent from my SCH-I535 using xda app-developers app
masri1987 said:
Anyone using trickster mod with this kernel?
Sent from my SCH-I535 using Tapatalk 2
Click to expand...
Click to collapse
It's the recommended tuner app, and plenty of people are using it (mostly on Sprint and AT&T). It shows some options that are completely bogus (like GPU governor), but I've added a few extras that only Trickster supports.
Thanks for the encouraging feedback, everyone! It's much appreciated. :good:
Kudos on this kernel. It has been perfect for me so far.
Running good on my d2usc! What scheduler do you recommend?
Sent from my SCH-R530U using Tapatalk 2
New 3.4 experimental builds are going up!
governors is building now, and has sio and zen. It'll eventually have a few new governors, but I haven't gotten everything ported yet.
gcc480 is up now, and is a test to make sure GCC 4.8.0 isn't causing any crazy bugs. It's also got sio and zen.
Edit: I forgot about you, trvbone. I don't have a preferred scheduler. I usually just use sio. I've never had a positive experience with ROW, but that might just be bad luck on my part.
More edit: I'm building a 3.0 kernel with trinity colors support and GCC 4.8.0 now. It should be up shortly.
decimalman said:
New 3.4 experimental builds are going up!
governors is building now, and has sio and zen. It'll eventually have a few new governors, but I haven't gotten everything ported yet.
gcc480 is up now, and is a test to make sure GCC 4.8.0 isn't causing any crazy bugs. It's also got sio and zen.
Edit: I forgot about you, trvbone. I don't have a preferred scheduler. I usually just use sio. I've never had a positive experience with ROW, but that might just be bad luck on my part.
Click to expand...
Click to collapse
I'm no noob when it comes to kernel lingo, but I'm not sure what the dif is here between the kernels listed as governors and gcc480. Are these two different kernels? I'm not aware of what gcc480 is I guess. Sorry until recently I was a dedicated lean kernel user and don't know all of the terminology since lk was pretty plain Jane.
I'm not a huge fan of row, but that's what I'm using right now cause it's quick just a little sketchy sometimes.
Sent from my SCH-R530U using Tapatalk 2
trvbone said:
I'm no noob when it comes to kernel lingo, but I'm not sure what the dif is here between the kernels listed as governors and gcc480. Are these two different kernels? I'm not aware of what gcc480 is I guess. Sorry until recently I was a dedicated lean kernel user and don't know all of the terminology since lk was pretty plain Jane.
I'm not a huge fan of row, but that's what I'm using right now cause it's quick just a little sketchy sometimes.
Sent from my SCH-R530U using Tapatalk 2
Click to expand...
Click to collapse
I have a tendency to describe things in the least understandable way possible. Sorry about that.
Hopefully, there's no noticeable differences between the two builds. I think the gcc480 build might benchmark a bit better, but I haven't really compared. The only actual difference is which compiler I used to build the kernel, which probably doesn't matter for 99% of users.
The governors build is built with the same ol' Linaro nightly toolchains that I've been using, so I expect it to be pretty stable.
The gcc480 build uses a newer, fancier compiler: GCC 4.8.0. I've found at least one new bug with the new compiler (it's fixed in the uploaded builds, don't worry ). I'm not ready to call the gcc480 build stable since it's gotten so little testing, but it's been running great for me all day. I'd love to hear feedback from anyone who flashes it.
Eventually, most other kernels will switch to GCC 4.8.0 (probably once Linaro releases a full-featured build). I think gideonx (of BMS fame) is planning to switch sometime soon, and I would expect ktoonsez to switch pretty soon too.
decimalman said:
I have a tendency to describe things in the least understandable way possible. Sorry about that.
Hopefully, there's no noticeable differences between the two builds. I think the gcc480 build might benchmark a bit better, but I haven't really compared. The only actual difference is which compiler I used to build the kernel, which probably doesn't matter for 99% of users.
The governors build is built with the same ol' Linaro nightly toolchains that I've been using, so I expect it to be pretty stable.
The gcc480 build uses a newer, fancier compiler: GCC 4.8.0. I've found at least one new bug with the new compiler (it's fixed in the uploaded builds, don't worry ). I'm not ready to call the gcc480 build stable since it's gotten so little testing, but it's been running great for me all day. I'd love to hear feedback from anyone who flashes it.
Eventually, most other kernels will switch to GCC 4.8.0 (probably once Linaro releases a full-featured build). I think gideonx (of BMS fame) is planning to switch sometime soon, and I would expect ktoonsez to switch pretty soon too.
Click to expand...
Click to collapse
Thanks for the expl I'll flash tomorrow morning and give it a whirl! Great job BTW. This is now my daily driver!
Sent from my SCH-R530U using Tapatalk 2
decimalman said:
I have a tendency to describe things in the least understandable way possible. Sorry about that.
Hopefully, there's no noticeable differences between the two builds. I think the gcc480 build might benchmark a bit better, but I haven't really compared. The only actual difference is which compiler I used to build the kernel, which probably doesn't matter for 99% of users.
The governors build is built with the same ol' Linaro nightly toolchains that I've been using, so I expect it to be pretty stable.
The gcc480 build uses a newer, fancier compiler: GCC 4.8.0. I've found at least one new bug with the new compiler (it's fixed in the uploaded builds, don't worry ). I'm not ready to call the gcc480 build stable since it's gotten so little testing, but it's been running great for me all day. I'd love to hear feedback from anyone who flashes it.
Eventually, most other kernels will switch to GCC 4.8.0 (probably once Linaro releases a full-featured build). I think gideonx (of BMS fame) is planning to switch sometime soon, and I would expect ktoonsez to switch pretty soon too.
Click to expand...
Click to collapse
Giving it a go now. Do you have any test methods you use to test the stability of a kernel other than the everyday use approach? Thanks for the support, looks sweet:good::good::good:

[Q] What is the buzz about Linaro?

First I saw a linaro kernel and now im seeing linaro aosp roms. What is the benefits of Linaro(either rom or kernel or both) over stock, or basic aosp or even the big roms like aokp and cm?
Linaro is a toolchain used in the compilation of roms. Its supposed to be extra efficient. Some kernels are compiled with linaro too. Kernel devs seem divided on whether it adds any benefit in a kernel-only build or not.
I tend to not waste energy thinking about it.
-----------------------
Sent via tapatalk.
I do NOT reply to support queries over PM. Please keep support queries to the Q&A section, so that others may benefit
rootSU said:
Linaro is a toolchain used in the compilation of roms. Its supposed to be extra efficient. Some kernels are compiled with linaro too. Kernel devs seem divided on whether it adds any benefit in a kernel-only build or not.
I tend to not waste energy thinking about it.
Click to expand...
Click to collapse
you dont sound too convinced
is it supposed to be better for battery life? better with background services? faster?
i saw one benchmark of linaro with dalvik patches and they're getting the same score i did on cataclysm with bionic/dalvik optimization patches
Enddo said:
you dont sound too convinced
Click to expand...
Click to collapse
Well devs are divided on whether building only a kernel with linaro has benefits or not. I know less than kernel devs so I haven't got an opinion. All I can say is choosing a kernel build with linaro is not on my priorities list.
http://en.m.wikipedia.org/wiki/Linaro
-----------------------
Sent via tapatalk.
I do NOT reply to support queries over PM. Please keep support queries to the Q&A section, so that others may benefit
Enddo said:
you dont sound too convinced
is it supposed to be better for battery life? better with background services? faster?
i saw one benchmark of linaro with dalvik patches and they're getting the same score i did on cataclysm with bionic/dalvik optimization patches
Click to expand...
Click to collapse
it has nothing to do with battery life and background services. all it is is a toolchain used to build roms and kernels. just because the developers of the linaro toolchain published that their toolchain improved performance two years ago, some people now believe it improves performance. compared to what they used as a comparison, it did. but they used an old toolchain to compare it to. but the google toolchain does just as good as the linaro toolchain, if not better. linaro is just a buzzword, nothing more. people add it into their titles for their roms or kernels to make people think that their rom or kernel is special, but it doesnt make a difference in reality. it is just a buzzword.
thanks for the help guys. im sorry if i came off noobish

About Uber TC

There are many recognised Rom developers for this phone. My research so far From what I have read has shown that the best roms are the ones with Uber TC. Just wanted to know if this Uber toolchain thing is kernel dependant or Rom dependant.

Categories

Resources