[GitHub] marcoabreu commented on a change in pull request #9542: Dockerfile to create Jetson TX1 and TX2 compatible builds.

[GitBox <gi...@apache.org>]

marcoabreu commented on a change in pull request #9542: Dockerfile to create Jetson TX1 and TX2 compatible builds.
URL: https://github.com/apache/incubator-mxnet/pull/9542#discussion_r163539980
 
 

 ##########
 File path: docker_multiarch/Dockerfile.build.jetson
 ##########
 @@ -0,0 +1,74 @@
+# -*- mode: dockerfile -*-
+# dockerfile to build libmxnet.so for armv7
+
+FROM nvidia/cuda:8.0-cudnn5-devel as cudabuilder
+
+FROM dockcross/linux-arm64
+
+ENV ARCH aarch64
+ENV NVCCFLAGS "-m64"
+ENV CUDA_ARCH "-gencode arch=compute_53,code=sm_53 -gencode arch=compute_62,code=sm_62"
+ENV BUILD_OPTS "USE_OPENCV=0 USE_BLAS=openblas USE_SSE=0 USE_CUDA=1 USE_CUDNN=0 ENABLE_CUDA_RTC=0 USE_NCCL=0 USE_CUDA_PATH=/usr/local/cuda/"
+ENV CC /usr/bin/aarch64-linux-gnu-gcc
+ENV CXX /usr/bin/aarch64-linux-gnu-g++
+ENV FC /usr/bin/aarch64-linux-gnu-gfortran-4.9
+ENV HOSTCC gcc
+
+WORKDIR /work
+
+# Build OpenBLAS
+ADD https://api.github.com/repos/xianyi/OpenBLAS/git/refs/heads/master /tmp/openblas_version.json
+RUN git clone https://github.com/xianyi/OpenBLAS.git && \
+    cd OpenBLAS && \
+    make -j$(nproc) TARGET=ARMV8 && \
+    make install && \
+    ln -s /opt/OpenBLAS/lib/libopenblas.so /usr/lib/libopenblas.so && \
+    ln -s /opt/OpenBLAS/lib/libopenblas.a /usr/lib/libopenblas.a && \
+    ln -s /opt/OpenBLAS/lib/libopenblas.a /usr/lib/liblapack.a
+
+ENV LD_LIBRARY_PATH $LD_LIBRARY_PATH:/opt/OpenBLAS/lib
+ENV CPLUS_INCLUDE_PATH /opt/OpenBLAS/include
+
+# Setup CUDA build env (including configuring and copying nvcc)
+COPY --from=cudabuilder /usr/local/cuda /usr/local/cuda
+ENV PATH $PATH:/usr/local/cuda/bin
+ENV TARGET_ARCH aarch64
+ENV TARGET_OS linux
+
+# Install ARM depedencies based on Jetpack 3.1
+RUN wget http://developer.download.nvidia.com/devzone/devcenter/mobile/jetpack_l4t/013/linux-x64/cuda-repo-l4t-8-0-local_8.0.84-1_arm64.deb && \
+    wget http://developer.download.nvidia.com/devzone/devcenter/mobile/jetpack_l4t/013/linux-x64/libcudnn6_6.0.21-1+cuda8.0_arm64.deb && \
+    dpkg -i cuda-repo-l4t-8-0-local_8.0.84-1_arm64.deb && \
+    dpkg -i libcudnn6_6.0.21-1+cuda8.0_arm64.deb && \
+    apt update -y && \
+    apt install cuda-cudart-cross-aarch64-8-0 cuda-cublas-cross-aarch64-8-0 \
+    cuda-nvml-cross-aarch64-8-0 cuda-nvrtc-cross-aarch64-8-0 cuda-cufft-cross-aarch64-8-0 \
+    cuda-curand-cross-aarch64-8-0 cuda-cusolver-cross-aarch64-8-0 cuda-cusparse-cross-aarch64-8-0 \
+    cuda-misc-headers-cross-aarch64-8-0 cuda-npp-cross-aarch64-8-0 libcudnn6 -y && \
+    cp /usr/local/cuda-8.0/targets/aarch64-linux/lib/*.so /usr/local/cuda/lib64/ && \
+    cp /usr/local/cuda-8.0/targets/aarch64-linux/lib/stubs/*.so /usr/local/cuda/lib64/stubs/ && \
+    cp -r /usr/local/cuda-8.0/targets/aarch64-linux/include/ /usr/local/cuda/include/ && \
+    rm cuda-repo-l4t-8-0-local_8.0.84-1_arm64.deb && rm libcudnn6_6.0.21-1+cuda8.0_arm64.deb
+
+# Build MXNet
+#ADD https://api.github.com/repos/apache/incubator-mxnet/git/refs/heads/master mxnet_version.json
+#RUN git clone --recursive https://github.com/apache/incubator-mxnet.git mxnet
+ADD mxnet mxnet
+
+WORKDIR /work/mxnet
+
+# Add ARM specific settings
+ADD arm.crosscompile.mk make/config.mk
+
+# Build and link
+RUN make -j$(nproc) $BUILD_OPTS
+
+# Create a binary wheel for easy installation.
+# When using tool.py output will be in the jetson folder.
+# Scp the .whl file to your target device, and install via
+# pip install
+WORKDIR /work/mxnet/python
+RUN python setup.py bdist_wheel
+
+# Copy build artifacts to output folder for tool.py script
+RUN mkdir -p /work/build & cp dist/mxnet-1.0.1-py2-none-any.whl /work/build && cp ../lib/* /work/build
 
 Review comment:
   Could the 1.0.1 in the filename be replaced with a dynamic variable in order to be compatible with version changes in the future?

----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on GitHub and use the
URL above to go to the specific comment.
 
For queries about this service, please contact Infrastructure at:
users@infra.apache.org


With regards,
Apache Git Services