[gentoo-soc] Weekly Report: Portage Powered Android

[Gunwant Jain]

Hello everyone,

I am sending my weekly report this late because I figured that I had
already discussed my findings this week on the mailing list. But my
mentor asked me to write it nevertheless.

My task was to build AOSP's LLVM toolchain, but for AArch64 host. 
Patching the build scripts for that was the easy part. But running the
script over and over became tedious at some point. Many times, my sshfs
caused problems with the CMakeCache, essentially forcing me to clean
build. Other times, the OOM killer got me. 

Nevertheless I can report that llvm's stage1 builds for the AArch64 host
and CMake is accepting the stage1 compilers to build the final llvm
stage2. I have yet to fully build llvm stage2, but I am sure that my
Mentor's kind offering of an arm64 server would check off that task.

I have already explained llvm stage1 and stage2, in some depth, in the 
previous weekly report's reply chain. I'll put a brief explanation here
anyways.
Bascially, AOSP packages its own LLVM distribution. The idiomatic way to
build LLVM in this case would be to build a "stage 1" compiler with your
host toolchain and then build the "stage 2" compiler using the "stage 1"
compiler. In AOSP's case, that stage2 compiler is used :   
    - in the AOSP tree as the prebuilt toolchain, used to build AOSP.
    - as the stage1 compiler to build a version bumped LLVM toolchain
      (later used as the prebuilt toolchain in the AOSP tree).

Therefore my next task would be to use that llvm toolchain (built for
AArch64 host) in the AOSP tree and build Android on arm64 unlike the
conventional cross-compiling method.

Regards,
Gunwant