Delay emutls deallocation for one round

With Android/Bionic, delay deallocation to round 2 of 4. It must run after
C++ thread_local destructors have been called, but before the final 2
rounds, because emutls calls free, and jemalloc then needs another 2
rounds to free its thread-specific data.

Bug: https://github.com/android-ndk/ndk/issues/687
Bug: b/78022094
Test: manual
Test: ./run_tests.py --rebuild --filter emutls-dealloc
Change-Id: I01bd634d97b7d22161b5cc8ca71b3cb94064a03e
1 file changed
tree: 3e862c668a71a00e479fc9facb6c8f54ed1848f5
  1. OWNERS
  2. README.md
  3. README.version
  4. build-gcc.sh
  5. build.py
  6. compiler_wrapper
  7. gcc-4.8.3/
  8. gcc-4.9/
  9. update-prebuilts.py
README.md

Building GCC for Android

The following process is used to build the GCC that is used by both the Android platfrom and the NDK.

Both Linux and Windows toolchains are built on Linux machines. Windows host binaries are built with mingw. Building binaries for Mac OS X should be built using 10.8 to ensure compatibility with Android's minimum supported hosts.

Prerequisites

  • Android GCC Repository

    • Check out the branch gcc

      repo init -u https://android.googlesource.com/platform/manifest -b gcc
      
      # Googlers, use
      repo init -u \
          persistent-https://android.git.corp.google.com/platform/manifest \
          -b gcc
      
  • Additional Linux Dependencies (available from apt):

    • texinfo
    • gcc-mingw32
    • bison
    • flex
    • libtool
  • Mac OS X also requires Xcode.

Host/Target prebuilts

For Linux or Darwin:

# Additional options and toolchain names will be show with --help.
$ python build.py --toolchain TOOLCHAIN_NAME

For Windows, from Linux:

$ python build.py --system windows TOOLCHAIN_NAME