Crosstoolchain discussion for TF700t!

[LetMeKnow]

I hope that _that will lead us in this topic because he seems to know away more than I do. I am here to learn and feel free to discuss anything that you like. No restrictions so we can get all the input from other users....

 

[LetMeKnow]

What is a toolchain?

After discussion with a few users, it is a mixed of toolchain types that they use.. According to my research, androideabi is targetting ROM build and optimize for the ROMs' binaries. It is fine when you use it to compile your kernel source but it is not optimized for the kernel compilation.

For kernel compiling, you should use the gnueabi toolchain because it uses the kernel's source for a specific kernel version during the toolcchain compiling for a better compatibility, I guess... However, some users reported that it was fine to use for ROM build also...

So the question is it matter what types of toolchains we are using? What are the benefits between the two? Does anyone see any difference between the two with users' experiences?

 

[lj50036]

Here I will take this spot and fill it with useful info and links about what I have found on the web .... :good:

1. MythBusters XDA Edition: “Optimized” Compiler Toolchains
   
   
2. USING THE ANDROID TOOLCHAIN AS A STANDALONE COMPILER
   
   
3. ELinuxToolchains
   
   
4. The GNU Toolchain for ARM targets
   
   
5. ARM

 

[LetMeKnow]

Here I will take this spot and fill it with useful info and links about what I have found on the web .... :good:


MythBusters XDA Edition: “Optimized” Compiler Toolchains

It is great that you are joining the discussion because I have a lot of questions and some good optimizations while I tested with these toolchains. I will give what I know a long the way when the questions come up and hope we will have a better understanding what to use and not to use...

 

[hardslog]

In the olden days, I used the 4.6.2 linaro toolchains and I have heard that a lot of people swear by DoomLord's prebuilts.

Just wanted to throw that out there. I personally have not tried anything above 4.7 yet but now I am tempted to

 

[LetMeKnow]

In the olden days, I used the 4.6.2 linaro toolchains and I have heard that a lot of people swear by DoomLord's prebuilts.

Just wanted to throw that out there. I personally have not tried anything above 4.7 yet but now I am tempted to

Adding to your comment, I do see a performance improvement with different toolchains but some users said it is just a placebo...:crying: I am one of the trials and errors users with testing so nothing is going to stop me until proving by testing and users' experiences, haha...

BTW, I could not get the gcc-4.8/4.9 to work on our tf700 chipset yet because there are some graphical problems on linux kernel v3.1.10. I hope that someone can figure it out so we can test it...

 

[hardslog]

There is a PAC rom in the TF300 forums that claims they are using SaberMod 4.8 without issues. http://xdaforums.com/showthread.php?t=2501869

Furthermore there is a kernel (no longer in development it seems) in the TF300 forums that claims to use linaro 4.8 toolchains http://xdaforums.com/showthread.php?t=2625580

 

[LetMeKnow]

There is a PAC rom in the TF300 forums that claims they are using SaberMod 4.8 without issues. http://xdaforums.com/showthread.php?t=2501869

Furthermore there is a kernel (no longer in development it seems) in the TF300 forums that claims to use linaro 4.8 toolchains http://xdaforums.com/showthread.php?t=2625580

Thanks for the information...:good: I will look more into it when I have more time..

BTW, You should try the linaro toolchain for your kernel compilation but you should use the right kernel version that you intend to run. It is running very smooth... It takes less than 10 minutes to compile and test it out..

 

[sziggins]

Cross Compiler Toolchains [Linaro GCC]

Hi,
Interesting thread but in my humble opinion should be in TF700's development section. So, I just used Christopher83's Toolchain for compiling _that's that10 kernel and flash it in CROMBi-kk RC3. As we have Tegra 3 Soc I used the toolchain with arm-cortex_a9-linux-gnueabi prefix which is optimized for Cortex-A9 cpu with Neon-VFPv3. I tested all the latest versions: 4.9 doesn't work at all (the TF700 was vibrating continuously!), the 4.8 had visual glitches but with 4.7 is working with no problems at all! Finally, from the same thread krislibaeer clarifies a bit the linaro prebuilt toolchains

here a little explanation:

arm-eabi toolchain: is for kernels
arm-linux-androideabi: is for rom building

so you use the arm-eabi toolchain for your kernels and the arm-linux-androideabi for roms
hope it helps a bit

so recommend is the arm-eabi toolchain for kernels

Hope that helps the discussion.
Cheers.

 

[LetMeKnow]

Hi,
Interesting thread but in my humble opinion should be in TF700's development section. So, I just used Christopher83's Toolchain for compiling _that's that10 kernel and flash it in CROMBi-kk RC3. As we have Tegra 3 Soc I used the toolchain with arm-cortex_a9-linux-gnueabi prefix which is optimized for Cortex-A9 cpu with Neon-VFPv3. I tested all the latest versions: 4.9 doesn't work at all (the TF700 was vibrating continuously!), the 4.8 had visual glitches but with 4.7 is working with no problems at all! Finally, from the same thread krislibaeer clarifies a bit the linaro prebuilt toolchains
Hope that helps the discussion.
Cheers.

There are a few things that you need to pay attention to.
1. Neon-VFPv3 is for Cortex-a8 and not for a9. You may want to flag it as neon-fp16..
2. I believed that your toolchain is targetting linux kernel version 3.4.x or something but not for version 3.1.10.
3. I have the same issues with my owm builds gcc-4.8/4.9 without any solution.
4. Try some of -Ofast flag to see the improvement on v3.1.10

Good luck....:fingers-crossed:

 

[hardslog]

There are a few things that you need to pay attention to.
1. Neon-VFPv3 is for Cortex-a8 and not for a9. You may want to flag it as neon-fp16..
2. I believed that your toolchain is targetting linux kernel version 3.4.x or something but not for version 3.1.10.
3. I have the same issues with my owm builds gcc-4.8/4.9 without any solution.
4. Try some of -Ofast flag to see the improvement on v3.1.10

Good luck....:fingers-crossed:

Just an FYI

I took the plunge and tried a new toolchain. Ended up trying a 4.9 linaro one for the Grimlock Kernel. Works like a champ on my TF300t. HOWEVER for some reason it will not even boot on a TF700. I'm told it vibrates and the screen goes all white or something. So here is the question:

Why would new toolchains work fine on a TF300 but not on a TF700? One of the transformers' great mysteries :laugh:

 

[LetMeKnow]

Just an FYI

I took the plunge and tried a new toolchain. Ended up trying a 4.9 linaro one for the Grimlock Kernel. Works like a champ on my TF300t. HOWEVER for some reason it will not even boot on a TF700. I'm told it vibrates and the screen goes all white or something. So here is the question:

Why would new toolchains work fine on a TF300 but not on a TF700? One of the transformers' great mysteries :laugh:

Thanks for the information and very good quedtion....:good:

Here is my wild guess because the chipset is using in the tf700t, cortex-a9 t33... I checked the diffs on gcc4.7 and gcc4.9 and tried to match all libraries in hope that I could narrow down the bug but it was failed. There was one time that I succeeded boot into the tf700 with my compiled gcc4.9 and thought that I found the bug but if I rebooted it, it got back to the graphical issue, flicking screen... If I rebooted a few more times then the tf700 was working again. I did all my best to figure out the bug but it was a big failure at the end. That is how far it goes as of today... I don't know enough to solve the mysteries and hope that someone else will....:fingers-crossed:

 

[hardslog]

Thanks for the information and very good quedtion....:good:

Here is my wild guess because the chipset is using in the tf700t, cortex-a9 t33... I checked the diffs on gcc4.7 and gcc4.9 and tried to match all libraries in hope that I could narrow down the bug but it was failed. There was one time that I succeeded boot into the tf700 with my compiled gcc4.9 and thought that I found the bug but if I rebooted it, it got back to the graphical issue, flicking screen... If I rebooted a few more times then the tf700 was working again. I did all my best to figure out the bug but it was a big failure at the end. That is how far it goes as of today... I don't know enough to solve the mysteries and hope that someone else will....:fingers-crossed:

Have you tried to compile a stock TF700 kernel with a 4.8 or 4.9 toolchain? I'm asking because _that kernel and Grimlock kernel actually change the cpu_speedo_id of the TF700 from 5 to 12

For reference check this commit: https://github.com/Hardslog/grimlock_kernel_asus_tegra3_unified/commit/50a19d0f6d6d03e6187a8fa7273be77755d72324#diff-c8f9ec2e1535a394abdd70e576a02ed7R160

I can only go so far with testing as I don't own a TF700........

 

[LetMeKnow]

Have you tried to compile a stock TF700 kernel with a 4.8 or 4.9 toolchain? I'm asking because _that kernel and Grimlock kernel actually change the cpu_speedo_id of the TF700 from 5 to 12

For reference check this commit: https://github.com/Hardslog/grimlock_kernel_asus_tegra3_unified/commit/50a19d0f6d6d03e6187a8fa7273be77755d72324#diff-c8f9ec2e1535a394abdd70e576a02ed7R160

I can only go so far with testing as I don't own a TF700........

No, I have not but it is a good idea to try out. I have a few more days before leaving for two weeks... I will report back before the weekend, thanks again...:highfive:

BTW, have you try some -Ofast flags, not the -Ofast itself? Some of them are working very well with tf700 kernel..

Update: I don't have time to try your recommendation because I am preparing for my business trip. I will give it a test when I am back...