I happened to come across a very interesting article regarding speed improvements for ICS on devices with (TI Pandaboard) OMAP 4430 processors. I believe the NT falls into this category. Supposedly the improvements will practically double the performance of the current build of ICS (4.0.4). The difference is really quite significant. It looks like it is being added to CM9 so we should hopefully see it on the NT!
From what I can gather...
They use a newer compiler and a special toolset built in. The use the linaro android system that has all string operations replaced for a faster build. They changed the way that the build system works. It sounds like the optimization is CPU based, so it should work even if Ducati isn't working. There are a lot of other optimization included, but I didn't really understand what the guy is talking about. You can find a video in the link below.
http://www.androidpolice.com/2012/0...e-and-now-parts-of-it-are-being-added-to-cm9/
https://wiki.linaro.org/Platform/Android
I guess the question is what we need to do to be sure that this gets implemented?
It sounds like some of it is being worked into the CM source. So I'd say we may see some after that unless some one feels like working it into a build themselves.
See the last few posts in this thread.
Hey guys,
just want to know, if the Linaro Toolchain is implemented in some kind of ROM or something like that.
The Optimus 2x got it already. And really -> I have never seen some smoother Android Experience... This is nutz!!!
Would be nice if someone could implement this in some SGS2 ROMs.
I've heard that it might be added to CyanogenMod.
Hello,
I have tried to search kernels which are compiled with Linaro toolchain but I cannot orient in it. Can please anyone post here kernels which are compiled by Linaro toolchains here? (Jelly bean only).
festr said:
Hello,
I have tried to search kernels which are compiled with Linaro toolchain but I cannot orient in it. Can please anyone post here kernels which are compiled by Linaro toolchains here? (Jelly bean only).
Click to expand...
Click to collapse
Probably none. It's better/easier to use the toolchain from Google, that way you can be sure there won't be any problems compiling and you won't get any bugs from using an experimental toolchain.
Linaro's changes are being merged/considered for merging upstream. So as Google updates their toolchain we'll get the optimizations made by Linaro (and others).
If you want to live on the bleeding edge you're welcome to build with Linaro's toolchain. I'd suggest building Android as well as the kernel if you do that though.
I have used kernel on SGS 1 compiled by Linaro toolchain and the performance gain was outstanding. So thats the main reason I'm asking if someone already made experimental kernel compiled by Linaro for SGS 3. The SGS1 was pretty stable and I still have it on my old SGS 1.
Actually the performance gain from the Linaro TC is minimal, nothing outstanding. So don't even sweat it dude. Go on with the stock toolchains.
At least it was noticable difference on SGS 1 with linaro / without linaro. But it is only my expirience without any science background.
You have tried Linaro compiled kernel on SGS 3?
b-eock said:
Actually the performance gain from the Linaro TC is minimal, nothing outstanding. So don't even sweat it dude. Go on with the stock toolchains.
Click to expand...
Click to collapse
Yes exactly, the benefits were minimal at best.
Outstanding is waaaayyyyy off the mark.
(I've used linaro kernels on gnex, the exact same source (Franco's kernel) one built using linaro tool chain, the other using the Google tool chain.)The difference was negligible.
Sent from my GT-I9300 using xda premium
Thank you for sharing expirience. That is totally in oposite to my expirience on SGS1
Perseus at least is compiled with linaro toolchain.
b-eock said:
Actually the performance gain from the Linaro TC is minimal, nothing outstanding. So don't even sweat it dude. Go on with the stock toolchains.
Click to expand...
Click to collapse
The performance in SIII is minimal, ok. BUT USE LESS CPU THAN THE USUAL KERNEL.
and this is GOOD FOR NOT SPEND TO MUCH BATTERY...
Striatum_bdr said:
Perseus at least is compiled with linaro toolchain.
Click to expand...
Click to collapse
I actually reverted to using the standard toolchain, there were some bugs with the Linaro causing games to crash. Anyway lets get this myth busted once and for all: kernel compilation optimizations are almost useless. The kernel is there for system management and control, there is no performance critical parts in it or large computations being done. These are all parts of the operating system, so your ROM needs to be compiled with these optimizations to see any effect, especially graphics drivers and things like that.
The kernel actually do a lot of stuff. Graphic drivers is also as kernel module. I have heard from other developers that compiling android with LInaro do not bring any noticable gain. Its intended to speedup the kernel.
festr said:
The kernel actually do a lot of stuff. Graphic drivers is also as kernel module. I have heard from other developers that compiling android with LInaro do not bring any noticable gain. Its intended to speedup the kernel.
Click to expand...
Click to collapse
Graphics drivers are not part of the kernel, the GPU platform drivers are. There are no graphical computations or complex things done on the kernel, the kernel just does low level management stuff.
Actually to argue on that I have to learn more. All in all, this speeks for itself:
http://www.youtube.com/watch?v=mrQRYmYip6Q
But anyway - it seems that some work of Linaro has been adopted by Goodle for Jelly bean.
festr said:
Actually to argue on that I have to learn more. All in all, this speeks for itself:
http://www.youtube.com/watch?v=mrQRYmYip6Q
Click to expand...
Click to collapse
That was the entire system built with the linaro toolchaim (ROM + Kernel) and its doing synthetic benchmarks, take from that what you will.
Also you are arguing with the wrong guy here, if anyone knows about this stuff its Andreilux.
Sent from my GT-I9300 using xda premium
According to +Jean-Baptiste Queru, We've already merged some of Linaro's changes into AOSP (and therefore into Google's internal master branch), a few weeks ago. Some more are pending review and look promising.
Can any of the existing Tbolt kernels be compiled using a newer toolchain for better performance?
Looks like Linaro 4.6 works on 2.3.7.
http://www.linaro.org/downloads/1111/
I love leankernel, zoom, etc., but maybe we could squeeze a bit more out of them?
So, this is something I've been trying to find an answer for, here and elsewhere, but found nothing conclusive:
Between Linaro and SaberMod toolchains, which one produces faster / more optimized binaries? I'm aware of Linaro's improvements over the standard Google toolchain, but it's not clear if SaberMod also follows the same line, and which one should be better regarding binary performance.
If you were to rephrase the question, your answer would be "42".
dbolivar said:
So, this is something I've been trying to find an answer for, here and elsewhere, but found nothing conclusive:
Between Linaro and SaberMod toolchains, which one produces faster / more optimized binaries? I'm aware of Linaro's improvements over the standard Google toolchain, but it's not clear if SaberMod also follows the same line, and which one should be better regarding binary performance.
Click to expand...
Click to collapse
SaberMod is more based on GNU GCC over aosp and generally linaro as well. Also take into consideration that SaberMod has things like graphite enabled where linaro or Google does not.