[ROM][5.0.2/LRX22G][AOSP][LINARO/OPTIMIZED] FML: Fork My Life (2015/01/08)

[#Intro]
Oh, and now there's these Lollipops I'm handing out. They are free as always, and they are very very delicious.


[#Why]


[#Info]

I need people testing and finding bugs if anything is going to be fixed. I might not have the Bluetooth devices you have, I might not use the camera as much as you, I might not use mobile data as much as you; you get the idea.
NOTE: FML is built and tested on the Verizon Galaxy Nexus variant, also known as toro. The GSM Galaxy Nexus variant (the one pertaining to the forum section you're currently in), also known as maguro, is extremely similar in hardware to the Verizon Galaxy Nexus, however I'm not able to test these builds personally. There aren't any code changes between builds for toro and maguro so that I know there shouldn't be any huge issues, but things specific to maguro I'll need feedback on to make sure they're working OK.


[#HowTo]


KitKat/ROM Stable: omni-4.4.4-20140705-maguro-FML.zip (159.07 MB)
KitKat/TWRP Stable: fml-twrp-2.7.1.0-20140705-maguro.img (8.58 MB)
KitKat/ROM Beta: omni-4.4.4-20141015-maguro-FML.zip (160.13 MB)
KitKat/ROM Beta Hotfix: boot.img (4.83MB)
Hotfix Info: The latest beta had a bug on maguro causing the radio to be a jerk and not work. Flashing this boot.img will resolve that issue. If you'd rather not flash it with fastboot like is normally done for .img files, you can use the Flashify app for it, or you can open up the ROM zip and replace the boot.img in there with the hotfix.
Lollipop/ROM Beta: FML-AOSP-5.0-20150108-maguro.zip (194.45 MB)

LOLLIPOP INFO:
USE THESE GAPPS: FML-GApps-5.0.x-20150101-tuna.zip (167.81 MB)
They are based on PA's GApps, huge thank you to them!
Known Issues:
- Camera can be a little touchy, but it generally works.
- Long SMS messages fail to send, fix is on-the-way though.
Lollipop Changelogs can be found in the post below!

(KitKat) BETA INFO:
Beta builds are using updated GPU drivers (and an updated kernel to go with them) courtesy of @Ziyan, as well as being up to date with the latest stuff from OmniROM. NOTE: YOU CANNOT FLASH A DIFFERENT KERNEL WITH THESE BUILDS.
Currently broken with the new GPU drivers:
- Hardware Video Decoding FIXED 2014/10/06
- Hardware Video Encoding(?) FIXED 2014/10/08
- Camera (PARTIALLY) FIXED 2014/10/08
- The stock camera app (as well as Google Camera) has this weird quirk of crashing when pressing the shutter button to take a picture, however many different camera apps on the Play Store (in particular Camera ZOOM FX) work flawlessly. Video recording is generally OK, however I've been getting some reports of the audio and video being a little out of sync, so your mileage may vary...
See the Changelog post below for..... well..... changelogs.


[#GApps]


- The screenrecord command does not work directly, however it does work via the Power menu.
- There can be a slight (noticeable, but not huge) delay when pressing the Recents or Home button.


[#Thanks]


[#Donations]

XDA:DevDB Information
FML: Fork My Life, ROM for the Samsung Galaxy Nexus

Contributors
MWisBest
Source Code: https://github.com/MWisBest/

ROM OS Version: 5.0.x Lollipop
ROM Kernel: Linux 3.0.x
Based On: AOSP

Version Information
Status: Beta
Stable Release Date: 2014-07-05
Beta Release Date: 2015-01-08

Created 2014-05-27
Last Updated 2015-01-08

Lollipop Beta Changelogs:
2015/01/08
- Updated the camera HAL.
This is going to be a little... fragile, at first, but in the long run it's needed. Stock camera app saves pictures now at least.
You may also notice that there's a 5.1MP resolution now (clear your camera app's data if you don't see it), despite our camera supposedly being 5.0MP all these years. Turns out the sensor's native resolution is indeed a whopping 16 pixels taller and 16 pixels wider than we've been using.

2015/01/04
- Added back some of the classic FML optimizations and more.
Built with -fstrict-aliasing flag, C++11, and the Linaro GCC 4.9 toolchain.
- Updated to AOSP "android-5.0.2_r1" tag.
As usual, the only real change was them bumping the version number.
- Improved flashing.
The speed of flashing the ROM zip is now much faster.
- Superuser is now built-in.
There's no need to flash SuperSU. You can find Superuser in the Settings app. You may have to enable the "Development Settings" menu to see it.
Some apps are a little sketchy with this Superuser though, most notably Titanium Backup. If you have issues, try flashing SuperSU.
- Video playback should be fixed.
YouTube in particular seems OK. If you have some weird format you're playing back locally I cannot make any guarantees though.
- The camera situation has improved and also regressed.
The stock camera is kinda back to the point of where it was with the KitKat Beta builds:
---- Preview works. Saving photos does not work. Video recording kinda works.
--- However:
---- The camera calibration is a little messed up. Long story, but it'll be fixed soon. So your white balance may look kinda funky, among other things.
- Audio quality is improved, with support for 192kHz FLAC as well.
This is possible thanks to a new audio resampler, which doesn't impose the same sample rate restrictions of the stock Android ones. It also has far better quality than the stock Android resamplers.
- WiiMotes can now be paired via Bluetooth.
I used to be a bit of a Wii hacking enthusiast... I think I still need to add some more stuff to make it useable as a gamepad or something, but yeah.
- Fixed location services issues.
- (toro) Fixed switching between 4G/LTE and 3G/CDMA network settings.
- Kernel changes:
Reclaimed 38MB of RAM from the carveout stuff.
Removed HDMI's framebuffer, saving an additional ~16MB of RAM. With Lollipop, MHL/HDMI out isn't working with our old HWComposer, so might as well save some RAM in the mean time.
Added overclocking support, as well as adjusted the stock frequencies a little.
Added the GPU kernel driver back into the kernel, rather than building it as a separate module.
Switched to LZO compression for the kernel. It results in a slightly larger size to the kernel, but it boots faster.
Added the "purple tint fix".
Added compiler flag to tune code for Cortex-A9 CPU.
Built with Linaro GCC 4.9 toolchain.
(v3): Fixed VYL00M eMMC bootlooping.

Older Builds:


KitKat Beta Changelogs:
2014/10/15
- Cleaned up the kernel.
Ziyan went over all his work not long ago and cleaned up the commit history and whatnot. He's letting me do the task of merging in the GPU driver stuff, so I'm currently just getting the kernel to a "clean slate" of sorts for other kernel devs to fork and build off of. Basically this should hopefully be the new "stock" or "Google" kernel. As a result, there's some bells and whistles missing from this build. It does however include some new audio updates that weren't in the previous kernel, which leads me to this...
- Changed audio sampling rate to 48kHz.
Previously the sampling rate has been 44.1kHz. The Galaxy Nexus supports 48kHz though! Here's what mainly sparked this change: since KitKat, the UI audio effects (e.x. touch sounds) have been switching over to 48kHz. With the Galaxy Nexus still using 44.1kHz, UI sound playback became kinda sluggish, and this was a big contributor to it. The UI sounds use the "fast track" audio path (where it tries to do as little processing as it can to the audio in order to play it with a minimum amount of latency), however a requirement of using this is that no resampling (e.g. converting 48kHz to 44.1kHz, like it has been doing currently with KitKat) can be required on the audio being played. Therefore the UI sounds were taking the "deep buffer" audio path, which has a delay to it. In switching to 48kHz, the "fast track" path is actually even faster now than it has ever been previously, and the Galaxy Nexus was already TOP OF THE LINE when it comes to audio latency believe it or not!
Another thing about this is that there is now support for playback of 96kHz audio files.
- Some updates to the camera HAL.
This hasn't fixed the situation of stock and stock-based camera apps crashing when taking a picture, but it's a start on it though.

NOTICE: Poweramp (and probably most other media players that use their own native code for audio playback rather than straight-up using the built-in Android APIs) is being pissy about the sampling rate change. If you experience issues with music playing, especially things like popping or crackling, please try Play Music or Apollo and see if your issue is still present. I know the UIs for Play Music and Apollo are pretty awful, but I can't write my own media player because I'm dealing with this camera stuff!

KitKat Stable Changelogs:
2014/07/05 (Operation: Streamline)
ROM: Synced with OmniROM's latest changes as of around 6:30 AM 2014/07/05 UTC.
ROM/Build: Fully updated to AOSP 4.4.4 (specifically, the android-4.4.4_r1 tag), which really doesn't change much though...
ROM/Build: Stopped including the unused (as far as I can tell) dock.png in /system/vendor/res/images/dock/
ROM/Build: Leveraged a feature added to updater-script creation by OmniROM which coincidentally makes the ROM flashable on any format of /system partition, beit F2FS, EXT4, exFAT, whatever.
ROM/Build: Stopped including Voice Dialer. Voice Dialer is just an unpolished piece of junk which really isn't used ever since Google Now.
ROM/Build: Stopped including 0xBenchmark.
ROM/Core: A number of changes added for completely seamless and simultaneous F2FS and EXT4 support.
ROM/General: Switched to the Android KitKat boot animation, which takes up nearly 4MB less space than OmniROM's boot animation.
ROM/General: Used OptiPNG heavily on numerous things in an effort to save space.
ROM/Kernel: Added F2FS support, nearly 500 commits were merged in for this.
ROM/Kernel: Relaxed BIGMEM a bit to hopefully fix Camera crashing for some users. (Only a 4MB difference BTW)
ROM/Kernel: Optimized CPU L2 cache settings slightly.
TWRP: Added support for seamless and simultaneous F2FS and EXT4.

I'm forgetting a number of things and I'm not going into as much detail on some of this as I'd like to. Frankly, I'm exhausted. Maybe I'll expand on this tomorrow. Maybe.
Oh I also submitted 8 things to the OmniROM Gerrit. One has been merged so far, the others probably will probably be merged in the next day or two.


Older Builds:

Slim down the build by putting less used stock applications into a separate flashable .zip, such as Browser.


Creating Black Holes with my phone's ridiculously awesome speed.

This is a little section I'm gonna set up explaining things in more technicalish and "down-and-dirty" details of sorts for developers interested in this project and potentially incorporating it into their projects.

The only thing I ask is to make a little "Thank You" section in your main post like I have here and credit at least me and Linaro, and also credit anybody else's work I have used if you use it as well. I'd also appreciate it if you could maybe link my name to this thread or my user profile here on XDA, but that part isn't a requirement.


All of my work can be found on my GitHub. Please note that any commits on my GitHub that are after the most recent build of FML should be considered experimental and potentially not working at all. I develop on the fly and often times things on my GitHub aren't finished and fully tested unless they have made their way into an official build of FML.
Please pay no attention to where it says a repository was forked from. Often times I'll have forks that I just re-use to avoid duplicate and unnecessary extra repos. For example, in repos forked from CyanogenMod you might notice the default branch is actually something like "omni-4.4" indicating that branch is based from OmniROM and not CyanogenMod.
The best place to keep track of what parts of the Android source code that needs patches is the manifest.


One of the big things Linaro does with improving Android's performance is fixing violations of what's known as "the strict aliasing rule."
A pointer is said to alias another pointer when they both refer to the same location of memory. This is OK and not an uncommon thing to do. The strict aliasing rule is that pointers of different types should never refer to the same location of memory (aka alias each other). Things like this are just fine and dandy:
That's alright, as foo and bar are the same type. Note that it's also OK if the only difference between foo and bar is signedness (e.x. uint32_t and int32_t).
Now this...
...this is a problem. foo and bar are NOT the same type. This is a violation of strict aliasing.
Strict aliasing allows a compiler to make some assumptions when compiling and optimizing code that it otherwise couldn't. This is a good read about the benefits of it.
Here's a few examples of fixing strict aliasing violations:
DSPManager
frameworks/av
bionic
Note that not everything is fixable, or worth fixing. Sometimes you'll just have to add -fno-strict-aliasing to the problematic section and call it a day:
frameworks/base


This section will delve into compilers and flags for them. The "Of Unicorns" part is in reference to the amount of false information, misconceptions, and mythical beliefs regarding these things. One thing in particular is the common belief that throwing every flag possible at the compiler results in better/faster binaries. That couldn't be further from the truth, and is something that actually took me quite some time to properly understand myself (in part because the misconceptions are more common than the actual truth!). In this section I will mainly be referencing the GCC compiler, as that's what is currently used for the majority of Android and most Linux systems as a whole. The other compiler that is making quite a run at GCC is Clang, so first I will talk about GCC vs Clang quickly:

GCC vs. Clang
GCC is currently (~May 2014) the most common compiler used for Linux and Linux-based systems (which includes Android). GCC was first released in 1987, as the "GNU C Compiler." Not long after its release, it was extended to support C++ as well, and over time many different languages and platforms became supported by GCC so it is now called the "GNU Compiler Collection." Over all this time, GCC became more and more difficult to maintain. As time passes on, fewer and fewer people have been able to get their foot in the door to work on GCC as it just became so... bloated.
Eventually, somebody finally got the guts (or resources, rather) to take on GCC and make a competing compiler. Clang was born.
Clang is a front-end for LLVM. Initially, LLVM was going to make use of GCC's front-end for making a C/C++/etc. compiler using LLVM's back-end, however this was just too cumbersome of a task due to GCC's difficult-to-work-with codebase, which was what sparked Clang instead.
Fun fact: Clang 1.0 was released in 2009. It was first open-sourced in 2007. That's 20 years after GCC's inception, but yet Clang has managed to tear GCC's usage apart. However to be fair, LLVM's initial release was 2003, but that's still a decade and a half head-start given to GCC!

As of GCC 4.8 vs. LLVM/Clang 3.4, it's kind of a toss-up between the two. In some cases GCC has better binary results and in other cases Clang has better binary results, however Clang outshines GCC in areas other than the resulting compiled code:
1. Clang is faster and uses fewer resources than GCC when compiling. It's usually a safe bet that Clang is going to be at least 50% (1.5x) faster than GCC when compiling, whilst also somehow using less RAM and disk space than GCC. For those of you that like being eco-friendly, just imagine the amount of energy this saves!
2. Clang has generally been ahead-of-the-game when it comes to supporting C++ standards. Current example: Clang has been C++14 feature-complete since the end of 2013, while GCC (even in 4.9!) is not.

Thanks to the competition from Clang, GCC has also been stepping up its game as well too. All-in-all this has been a win-win for everybody so far.


Flags.
Aaand here we go on compiler flags. For this I'll be referencing the GCC documentation on "GCC command options", here: https://gcc.gnu.org/onlinedocs/gcc/index.html#toc_Invoking-GCC
Of particular interest in this section will be "Options That Control Optimization", but other sections are often overlooked which can be of use, I'll explain those later though.

The first thing you'll see in the "Options That Control Optimization" section are the -O options. These are general optimization levels that give you a sensible default to work with.
First off is -O0. This is the default, and doesn't turn on any optimization options. This is generally only used for debugging code, as some optimizations can interfere with that.
Next would be -O1. This enables some optimizations to reduce code size and improve speed, without increasing the time it takes to compile code significantly.
After that is -O2. This is what is generally used for most program compiling, as it enables plenty of optimizations and these ones generally don't cause problems/bugs with the resulting binaries whilst greatly improving speed.
Then there are some non-numbered ones which I'll go over:
-Os is somewhere between -O1 and -O2. It enables most of -O2's options, but disables some that can increase the size of the resulting binaries. This can be of great use when dealing with smaller embedded systems where space is generally preferred over speed. It's closer to -O2 than -O1...
-Og is useful for debugging purposes. It enables a few optimizations that generally don't effect ability to debug code, so devs don't have to deal with the slowness of -O0 as much as usual.
And then there's the almighty -O3...
-O3 enables some extra optimizations that -O2 does not. The drawbacks are increased compile time, increased binary size, and the possibility for some bugs. In many cases, -O3 might not improve speed over -O2 whatsoever. In other cases it can be helpful, but it is nothing compared to the differences between -O1 and -O2.
Android generally uses -O2 by default. It's safe and fast enough for most cases. -Os is also used for any Thumb instruction-set code for ARM, as Thumb is generally meant for reduced size and complexity from what I can tell.
There's also a flag kinda above -O3, which is -Ofast. The problem with -Ofast is that it can cause huge problems for any code that makes the (correct!) assumption about some math stuff, so it is generally avoided.

Then there are some flags that aren't enabled by any optimization level. This is not without reason: these flags are, generally, useless. They increase compile time by a ridiculous amount, and their effects on the resulting binaries might be absolutely nothing. Zilch. Nadda. Even for something as large as Android, these flags can still do nothing. If these flags can have a speed-up, it's generally not without risk, and they should be kept to specific use-cases rather than used on every single thing.
These flags can be considered about the equivalent of using the "placebo" profile for h264 video encoding. It has a less-than 1% improvement in the resulting quality, yet can take twice as long to render (or, in this case, compile). And they are pretty much exactly what the placebo profile for h264 encoding is, a placebo effect. You will not find a measurable increase in speed, and the risks associated with it are generally not worth it!

There is an exception to that though: -flto. This enables Link-Time Optimization, and can have huge impacts on both the speed of the binaries as well as even a reduction in their size! LTO can be viewed as this: when the compiler is truckin along compiling things, it generally only sees bits and pieces of the project at hand. LTO allows the compiler to view how everything works together rather than just each individual part, and in doing so it can find HUGE improvements! When LTO was in its infancy with GCC, it was pretty unstable, but with GCC 4.8 and above it can be used reliably. It's also advised to use the -fuse-linker-plugin flag when using LTO as well (read the docs on that).
Here is my work on using LTO with Android, it's more involved than simply adding it with all the other flags which is why you'll generally hear the people that are making use of the "unicorn flags" say it doesn't work...
https://github.com/MWisBest/android_build/commit/c1b041c32572b6ee1bbb17b1fa8c038c5e9fde1f
https://github.com/MWisBest/android_build/commit/95bb49b613424b70af3e820748724fb92ef35b5e
https://github.com/MWisBest/android_build/commit/d2c13f1c35cfa9c114a69c74cfb1c2631643eebc
Ignore the things in the last couple commits there that aren't related to LTO...
But like I said, it's more involved than just throwing the flags blindly at the compiler, there are a few other fixes required to get a successful build with LTO:
https://github.com/MWisBest/android_bionic/commit/3b3f59a173a7cc4ff3a1cec4456a99108ce08092
And then it also has to be disabled in ART, that fix courtesy of @metalspring : https://github.com/MWisBest/android_art/commit/9ed3774fcba75077e098720406d261b79bd9baa9

// To be continued.



I thought I'd put some simple little things here that can be immensely helpful to devs. Most of these they'll probably know already, but some won't and when I learned these things they had a profound impact on development. At the very least it can be a helpful reminder/reference sort of thing.
Properly Logging Builds
This is something I finally figured out not long ago that has had a huge impact on debugging problems...
I'm not all that great with bash but I generally understood redirecting the output of a program to a file. However, when I tried the usual: , not everything was going to the file. Eventually I learned that just using ">" only redirects stdout, and not stderr which was where all the warnings and errors went to. To get around that, there's a couple options. One, you can redirect stdout and stderr to separate files, like so: , or the other option is to just lump them all into a single file using:

This post is a WIP...

As you may or may not have read in the OP, this ROM was built and tested on the Verizon Galaxy Nexus only for about 9 months. I don't own a GSM Galaxy Nexus, so I'm not able to fully test these GSM builds, however the differences between the two are minimal and everything should work OK.

As the Verizon Galaxy Nexus is pretty much only in the USA, one thing I did to reduce the size of the ROM was to cut back on some of the translations. I've kept what I thought were fairly common languages (including outside the USA). Since I've now expanded to the GSM Galaxy Nexus, which is more international, there's a chance that a language you need might be missing. Please just let me know what you need and I'll be more than happy to include it.

pvkiniyan95 said:

Downloading....perfect timing...i am rom hopping for a while and excited to try this, skimmed last few pages of your toro thread :good:
why not take a look at the recent thumb flag optimisation commits that claimed 6x speeds recently...

Click to expand...

Click to collapse

Yeah I replied to that thread... it's bogus, unsafe, and unprofessional. FML was far beyond that stuff since its inception and it has only gotten better since.

qtoo941 said:

You have written that you have built CM 10.2 FML version. Is it compatible with maguro? Where can I find it and flash it over my phone just to see it?

Click to expand...

Click to collapse

This right here is the first build I've done that's maguro compatible, and there's been lots of improvements since when this ROM was based off of CM so even if it'd work on maguro it's not really worth trying at this point. Sorry!

ahmadairfan said:

Downloading!
What do you recommend guys, Dalvik or ART?
I've been always using ART before and then I read replies in toro thread, most likely chooses Dalvik eh?

Thank you

Click to expand...

Click to collapse

pianistaPL said:

Only ART! Feel the smoothness

Click to expand...

Click to collapse

osm0sis said:

Dalvik! MWis pimped it the **** out!

Click to expand...

Click to collapse

^What he said. At the very least give Dalvik a try.

pianistaPL said:

Ok, end of speaking, phone charged, let's try this ROM!

Wysłane z mojego Galaxy Nexus przy użyciu Tapatalka
@MWisBest Can i let set DexOpt to /cache if i use Banks Core Gapps and i want change to ART?

---------- Post added at 03:29 PM ---------- Previous post was at 03:10 PM ----------

OK, i guess i can't, after boot with ART i get pernamently SystemUI FC So this is a good reason to try this super fast dalvik

Click to expand...

Click to collapse

That sounds like the issue you'd run into with a GApps package that's too large, if you want to use a GApps package that large with ART you'll have to disable the DexOpt to /cache thing before switching to ART.

pvkiniyan95 said:

thanks for that...u saved my time

Click to expand...

Click to collapse

See above.

gautam_nexus said:

gonna flash this B*tch tonite! the changelog looks impressive. one question though do i need to flash the twrp recovery in the OP as well before flashing the ROM. Thanks!

Click to expand...

Click to collapse

The recovery build is not required, I just figured I'd start uploading them as 1. It's built automatically, and 2. I use it myself.

pianistaPL said:

Very fast on dalvik but phone gets mini freeze when i tap recent apps button and when i tap home button (when i have some app opened)
I haven't recent apps buttom freeze/lag only when i am on home screen.

Click to expand...

Click to collapse

I'll certainly have a look at that and see what's causing it. The Recents issue might be due to the code regarding if OmniSwitch should be opened instead of the usual Recents, as I'm not getting that lag with OmniSwitch enabled as Recents.

mosca_ said:

Hi MWisBest, could you please add Spanish language support for this ROM? I'm very impressed with your work, and I'd like to have it translated.

Thank you very much in advance!.

Click to expand...

Click to collapse

Crap! I do have Spanish included, but the furthest I had gone with testing that stuff was to simply make sure they were listed in "Settings --> Language & input --> Language", which does have Español listed, however I clicked on it now after your post and there are some things that are missing translations, most notably: Settings!!
The good news is that I recently figured out how to properly log my builds, and sure enough I've found some warnings about translations not being included, so I should be able to track down what's causing this and have it fixed for the next build.

pianistaPL said:

Polish language doesn't have translated settings too So i'm waiting for new version, thanks

Click to expand...

Click to collapse

Thanks for the heads-up, looks like it's ALL the translations for Settings, dang!