tensorflow: Fail to build from source with gcc 7.3.1

With this workaround, you can compile tensorflow with gcc 6.3

In external/eigen_archive/unsupported/Eigen/CXX11/src/Tensor/TensorReduction.h:801 (you can find the file in a subdirectory of ~/.cache/bazel )

replace

values[i] = internal::InnerMostDimReducer<Self, Op>::reduce(*this, firstIndex + i * num_values_to_reduce, num_values_to_reduce, reducer);

by two instructions instead

Self::CoeffReturnType a = internal::InnerMostDimReducer<Self, Op>::reduce(*this, firstIndex + i * num_values_to_reduce, num_values_to_reduce, reducer); values[i] = a;

Hello,

I can confirm TensorFlow cannot be built anymore on Debian Stable (Stretch) and Ubuntu LTS (Bionic) due to this “internal compiler error”:

external/eigen_archive/unsupported/Eigen/CXX11/src/Tensor/TensorReduction.h: In static member function 'static void std::_Function_handler<void(_ArgTypes ...), _Functor>::_M_invoke(const std::_Any_data&, _ArgTypes&& ...) [with _Functor = Eigen::internal::TensorExecutor<Expression, Eigen::ThreadPoolDevice, Vectorizable, Tileable>::run(const Expression&, const Eigen::ThreadPoolDevice&) [with Expression = const Eigen::TensorAssignOp<Eigen::TensorMap<Eigen::Tensor<std::complex<float>, 0, 1, long int>, 16, Eigen::MakePointer>, const Eigen::TensorReductionOp<Eigen::internal::SumReducer<std::complex<float> >, const Eigen::IndexList<Eigen::type2index<0l> >, const Eigen::TensorMap<Eigen::Tensor<const std::complex<float>, 1, 1, long int>, 16, Eigen::MakePointer>, Eigen::MakePointer> >; bool Vectorizable = true; bool Tileable = false]::<lambda(Eigen::internal::TensorExecutor<const Eigen::TensorAssignOp<Eigen::TensorMap<Eigen::Tensor<std::complex<float>, 0, 1, long int>, 16, Eigen::MakePointer>, const Eigen::TensorReductionOp<Eigen::internal::SumReducer<std::complex<float> >, const Eigen::IndexList<Eigen::type2index<0l> >, const Eigen::TensorMap<Eigen::Tensor<const std::complex<float>, 1, 1, long int>, 16, Eigen::MakePointer>, Eigen::MakePointer> >, Eigen::ThreadPoolDevice, true, false>::StorageIndex, Eigen::internal::TensorExecutor<const Eigen::TensorAssignOp<Eigen::TensorMap<Eigen::Tensor<std::complex<float>, 0, 1, long int>, 16, Eigen::MakePointer>, const Eigen::TensorReductionOp<Eigen::internal::SumReducer<std::complex<float> >, const Eigen::IndexList<Eigen::type2index<0l> >, const Eigen::TensorMap<Eigen::Tensor<const std::complex<float>, 1, 1, long int>, 16, Eigen::MakePointer>, Eigen::MakePointer> >, Eigen::ThreadPoolDevice, true, false>::StorageIndex)>; _ArgTypes = {long int, long int}]':
external/eigen_archive/unsupported/Eigen/CXX11/src/Tensor/TensorReduction.h:801:9: internal compiler error: in emit_move_insn, at expr.c:3547
         values[i] = internal::InnerMostDimReducer<Self, Op>::reduce(*this, firstIndex + i * num_values_to_reduce,
         ^~~~~~

Debian Stable comes with GCC 6.3 while Ubuntu LTS uses 7.3.

Regards,

I did apply the patch from https://gcc.gnu.org/bugzilla/show_bug.cgi?id=89752 and now the compiling works. This patch hasn’t been applied on branch of 7.x or lower version. Please do it manually, and re-compile gcc on arm.

On Wed, Jun 12, 2019 at 11:21 PM Mircho Rodozov notifications@github.com wrote:

yes we are building with 7.4.1 @npanpaliya https://github.com/npanpaliya sure, I might as well try assembler or something 😆

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/tensorflow/tensorflow/issues/25323?email_source=notifications&email_token=AA5BADBJFQCEYV4QRJS7YVLP2EH6FA5CNFSM4GTJ54WKYY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGODXQYLOQ#issuecomment-501319098, or mute the thread https://github.com/notifications/unsubscribe-auth/AA5BADHWYAX2SGJN4LHNHLDP2EH6FANCNFSM4GTJ54WA .

It might because the usable memory of your system is too low compared to the number of parallel Bazel compilation tasks.

You could try setting it to a smaller value using: bazel build --jobs 4 //... or bazel build --ram_utilization_factor 50.

EDIT: wrong solution. Adding --config=opt seems to solve the problem with my gcc 6.3. Not sure why though.

Try with gcc 5.4. It worked for me.

The workaround didn’t work for me.

I’m working from a stretch-based Docker instance, looking to install TF from source.

What worked for me was upgrading to gcc 8.3.0

Instructions

(for other Debian rooks out there like me)

First I added the following line to the bottom of my /etc/apt/sources.list:

deb http://ftp.us.debian.org/debian testing main contrib non-free

Then I ran:

# apt-get update
# apt-get install gcc-8 g++-8

The default gcc for me was 6.3.0:

# gcc --version
gcc (Debian 6.3.0-18+deb9u1) 6.3.0 20170516
Copyright (C) 2016 Free Software Foundation, Inc.
This is free software; see the source for copying conditions.  There is NO
warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.

So I needed to update my gcc to point to the new one:

# update-alternatives --install /usr/bin/gcc gcc /usr/bin/gcc-8 80 --slave /usr/bin/g++ g++ /usr/bin/g++-8 --slave /usr/bin/gcov gcov /usr/bin/gcov-8
update-alternatives: using /usr/bin/gcc-8 to provide /usr/bin/gcc (gcc) in auto mode
# update-alternatives --install /usr/bin/gcc gcc /usr/bin/gcc-6 60 --slave /usr/bin/g++ g++ /usr/bin/g++-6 --slave /usr/bin/gcov gcov /usr/bin/gcov-6

The default is now 8.3.0:

# gcc --version
gcc (Debian 8.3.0-26) 8.3.0
Copyright (C) 2018 Free Software Foundation, Inc.
This is free software; see the source for copying conditions.  There is NO
warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.

If I wanted to default to a different gcc version, I can now use:

# update-alternatives --config gcc

References

• https://stackoverflow.com/questions/43151627/installing-g-7-0-1-on-debian-8-7/43151826#43151826
• https://linuxize.com/post/how-to-install-gcc-compiler-on-ubuntu-18-04/#installing-multiple-gcc-versions

We also run into the same problem trying to compile TensorFlow in our CentOS 7 with devtoolset-7. Does anyone have a solution for this?