Skip to content

Information on compiling addons on IBM PPC system, under Conda #2407

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Closed
paulobaptista opened this issue Mar 2, 2021 · 5 comments
Closed

Comments

@paulobaptista
Copy link

I wanted to shared the steps to compile addons with a Conda environment that loads TF, CUDA, CUDNN

Cuda Driver and Hardware
Driver Version: 440.33.01
CUDA Version: 10.2
GPU: Tesla V100
Conda Environment
tensorflow 2.1.3
tensorflow-gpu 2.1.3
cudatoolkit 10.2.89
cudatoolkit-dev 10.2.89
cudnn 7.6.5_10.2
bazel 0.29.1

Process
to meet addons requirements based on versions of installed CUDA, Tensorflow versiosn, we need to install 0.9.1 version of addons
activate conda environment with proper versions of tensorflow, cuda, cudnn
git clone https://github.com/tensorflow/addons.git
cd addons
git checkout tags/v0.9.1
set environment variables
export TF_NEED_CUDA="1"
export TF_CUDA_PATHS=$CONDA_PREFIX/cuda <---(this was the critical step)
export TF_CUDA_VERSION="10"
export TF_CUDNN_VERSION="7"
export CUDA_TOOLKIT_PATH=$CONDA_PREFIX
export CUDNN_INSTALL_PATH=$CONDA_PREFIX
python3 ./configure.py

The following applies for v0.9.1 of the build...

edit build_deps/toolchains/gpu/crosstool/BUILD.tpl
add the "ppc" line to the following block
cc_toolchain_suite(
name = "toolchain",
toolchains = {
"local|compiler": ":cc-compiler-local",
"ppc": ":cc-compiler-local",
"k8": ":cc-compiler-local",
},
)

run build and install

bazel build --enable_runfiles build_pip_pkg
bazel-bin/build_pip_pkg artifacts
pip install artifacts/tensorflow_addons-*.whl

@bhack
Copy link
Contributor

bhack commented Mar 10, 2021

Thanks, but we don't officially support conda here and in TF. @WindQAQ @seanpmorgan Can we collect this info somewhere?

@seanpmorgan
Copy link
Member

There was an early hope to package on conda forge:
#201

The rest of the tf ecosystem hasn't done it either (except TF core which is managed by Anaconda). I'm interested but concerned for the amount of work

@bhack
Copy link
Contributor

bhack commented Mar 10, 2021

Esri has an Addons conda package. Is It still maintained?
https://anaconda.org/esri/tensorflow-addons

@bhack
Copy link
Contributor

bhack commented Mar 10, 2021

Some other conda packages https://anaconda.org/search?q=Addons+tensorflow

@seanpmorgan
Copy link
Member

I didn't know of those packaging channels that support Addons. Given our availability I think it's best we let the community handle Conda builds (same thing TF core does, but Anaconda does it). Closing this as we have no way to test CI for a power PC build.

Thanks for posting this though I'm sure it'll be helpful for others. I wouldn't be opposed to a PR which adds this either since it can't really hurt the existing builds. Closing for now but happy to review PR if there's community interest.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

3 participants