Open AKKamath opened 2 years ago
Hi, I am experiencing the same issue. I created a Docker image of the Lonestar application, and it runs on a cluster with a Quadro RTX 6000 GPU. Even when I run the small sample, the BFS and some other applications hang. Have you resolved the issue?
FROM ubuntu:20.04
ENV BUILD_DIR=/galois/build
ENV SRC_DIR=/galois
WORKDIR /galois
RUN apt-get update \
&& apt-get install -qy \
apt-transport-https \
ca-certificates \
curl \
wget \
gnupg \
software-properties-common \
&& curl -fL https://apt.llvm.org/llvm-snapshot.gpg.key | apt-key add - \
&& apt-add-repository -y 'deb http://apt.llvm.org/focal/ llvm-toolchain-focal main' \
&& apt-get update
RUN apt-get install -qy \
ccache \
clang++-10 \
clang-10 \
clang-format-10 \
clang-tidy-10 \
# clang++ \
# clang \
# clang-format \
# clang-tidy \
cmake \
# g++-8 \
# gcc-8 \
g++-7 \
gcc-7 \
git \
gosu \
libfmt-dev \
libopenmpi-dev \
libboost-all-dev\
libnuma-dev\
llvm-10-dev \
libpapi-dev\
mpich \
libeigen3-dev\
python3-pip \
python-is-python3 \
# && update-alternatives --verbose --install /usr/bin/gcc gcc /usr/bin/gcc-8 90 \
# && update-alternatives --verbose --install /usr/bin/g++ g++ /usr/bin/g++-8 90 \
&& update-alternatives --verbose --install /usr/bin/gcc gcc /usr/bin/gcc-7 90 \
&& update-alternatives --verbose --install /usr/bin/g++ g++ /usr/bin/g++-7 90 \
&& update-alternatives --verbose --install /usr/bin/clang clang /usr/bin/clang-10 90 \
&& update-alternatives --verbose --install /usr/bin/clang++ clang++ /usr/bin/clang++-10 90 \
&& rm -rf /var/lib/apt/lists/*
# RUN apt-get install llvm \
# libpapi-dev \
# mpich \
# libeigen3-dev
RUN pip3 install --upgrade --no-cache-dir pip setuptools \
&& pip3 install --no-cache-dir conan==1.24
RUN wget https://developer.download.nvidia.com/compute/cuda/10.2/Prod/local_installers/cuda_10.2.89_440.33.01_linux.run \
&& chmod +x cuda_10.2.89_440.33.01_linux.run \
&& ./cuda_10.2.89_440.33.01_linux.run --silent --toolkit --override --toolkitpath=/usr/local/cuda \
&& rm -rf cuda_10.2.89_440.33.01_linux.run
# RUN echo "export PS1='\\W$ '" >> /root/.bashrc
ENV HOME=/root
COPY . .
RUN mkdir -p $BUILD_DIR
WORKDIR ${BUILD_DIR}
RUN export CUDACXX=/usr/local/cuda/bin/nvcc \
&& cmake $SRC_DIR -DGALOIS_CUDA_CAPABILITY="7.5" -DCMAKE_BUILD_TYPE=Debug
# RUN cmake -S $SRC_DIR -B $BUILD_DIR -DCMAKE_BUILD_TYPE=Release
RUN cd $BUILD_DIR/inputs && mkdir small_inputs
RUN mv $SRC_DIR/small_inputs_for_lonestar_test.tar.gz $BUILD_DIR/inputs/small_inputs
RUN cd $BUILD_DIR/inputs/small_inputs \
&& tar -xzvf small_inputs_for_lonestar_test.tar.gz
WORKDIR $BUILD_DIR
# RUN make -j
# RUN make install
WORKDIR /galois
RUN chmod +x /galois/entry.sh
# ENTRYPOINT ["sh", "-c", "cd /galois/build && make test && /bin/sh"]
ENTRYPOINT ["/galois/entry.sh"]
# CMD ["/bin/sh"]
# ENTRYPOINT ["/entry.sh"]
I ran LonestarGPU's implementation of BFS on Hollywood input data set (https://suitesparse-collection-website.herokuapp.com/MM/LAW/hollywood-2009.tar.gz). The application seems to hang. I added printfs to the code and it seems to hang at the "bfs_kernel".
Full output is listed below: