TensorFlow manylinux

A manylinux compatible Docker image to build the mrphys TensorFlow ops. It is based on manylinux and multipython Docker images created by the TensorFlow team. Then the following changes are applied.

• Install TensorFlow on all Python versions.
• Patch auditwheel to whitelist the TensorFlow framework library.
• Install some system dependencies for mrphys or third-party packages.
• Install dependencies for docs.
• Install latest Git version.
• Compile FFTW3 for generic x86-64 architectures.
• Patch TensorFlow installations to add CUDA headers.
• Install patchelf 0.12 from source to fix an issue with auditwheel.
• Install spiral_waveform.
• Add custom extensions for sphinx.

TensorFlow Versions

Each manylinux image has a specific TensorFlow version as detailed below:

TensorFlow manylinux	TensorFlow
v1.7	v2.6
v1.8	v2.7
v1.9	v2.8
v1.10	v2.8
v1.11	v2.9
v1.12	v2.9
v1.13	v2.9
v1.14	v2.10
v1.15	v2.11