Dear ARM A GNU Toolchain maintainers,
I have noticed that the latest AArch32 Linux GCC release v9.2-2019.12 for ARM A is configured without a default fpu option whereas ARM GCC v8.3-2019.03 and older Linaro toolchains used "--with-fpu=vfpv3-d16". The example ABE manifest file for v9.2 also still contains this option. Could you please explain the background for this configuration change?
Thank you very much,
Andrej
Thank you very much, no idea how I could overlook the neon setting.
Just for the understanding: there are 4 sets of gcc configure sets (gcc{1,2,3,4}_configure) in the manifest you pointed me at in contrast to ABE's stage1/stage2. What is the difference in ARM's approach there?
EDIT: the manifest file lacks glibc configuration information. Do you have a hint where to find it?
Kind regards
HI Andrej,
We have separate build system to ABE, which uses 4 stages to build toolchains which use Glibc (for various reasons mainly because we do more in our build system than ABE). We keep flags in our format to make sure we can reproduce on the same build system.
Our final 4th stage (which corresponds to stage 2 in ABE) is one your should be using to trim your compiler. Stage 1 in both build systems builds enough to go with the gcc-final stage. So I would keep stage 1 in ABE and concentrate on trimming our manifest's stage 4 -> ABE's stage 2.
Sorry, if this is a bit overwhelming.
kind regardsPrzemyslaw Wirkus
Thank you for the information!
Just one more question: where can one look up your Glibc configuration options?
kind regards
Andrej Gantvorg
Dear sir,
Looks like we are missing these flags. Well spotted. I will update build system to include those.
FYI:
glibc/configure --enable-shared --with-tls --disable-profile \ --disable-omitfp --disable-bounded --disable-sanity-checks \ --build=x86_64-unknown-linux-gnu --host=aarch64-none-linux-gnu \ --disable-werror --enable-obsolete-rpc --disable-profile \ --without-gd --without-cvs