ppc64le - undefined symbol: _ZN4cudf2io6detail12cufile_input10read_asyncEmmPhN3rmm16cuda_stream_viewE

[rnukala1]

I build cuda version of conda package for cuDF(21.10.00a cuda) and also dependent cuda version of conda packages of pyarrow and rmm on ppc64le (POWER) architecture. When I run a sample with simple "import cudf" I see the following error

(testcudfp39) [opence@waldevmlcerce04 cudf]$ python cudf_sample2.py Traceback (most recent call last): File "/rcebuild/rajesh/sample_test_programs/rapids/cudf/cudf_sample2.py", line 1, in import cudf File "/local/opence/anaconda3/envs/testcudfp39/lib/python3.9/site-packages/cudf/init.py", line 12, in from cudf import api, core, datasets, testing File "/local/opence/anaconda3/envs/testcudfp39/lib/python3.9/site-packages/cudf/datasets.py", line 5, in from cudf._lib.transform import bools_to_mask File "/local/opence/anaconda3/envs/testcudfp39/lib/python3.9/site-packages/cudf/_lib/init.py", line 4, in from . import ( ImportError: /local/opence/anaconda3/envs/testcudfp39/lib/python3.9/site-packages/cudf/_lib/../../../../libcudf.so: undefined symbol: _ZN4cudf2io6detail12cufile_input10read_asyncEmmPhN3rmm16cuda_stream_viewE

Below is the package list installed in the conda environment (testcudfp39) [opence@waldevmlcerce04 cudf]$ conda list

packages in environment at /local/opence/anaconda3/envs/testcudfp39:

Name Version Build Channel

_libgcc_mutex 0.1 main _openmp_mutex 5.1 1_gnu abseil-cpp 20211102.0 h8ec6b75_0 arrow-cpp 5.0.0 py39h926b4ae_34_cuda local arrow-cpp-proc 3.0.0 cuda local aws-c-common 0.4.57 he6710b0_1 aws-c-event-stream 0.1.6 h29c3540_5 aws-checksums 0.1.9 he6710b0_0 aws-sdk-cpp 1.8.185 h6205695_0 bottleneck 1.3.4 py39hfc2be9b_0 brotli 1.0.9 he6710b0_2 bzip2 1.0.8 h7b6447c_0 c-ares 1.18.1 h140841e_0 ca-certificates 2022.4.26 h6ffa863_0 cachetools 4.2.2 pyhd3eb1b0_0 certifi 2022.5.18.1 py39h6ffa863_0 cudatoolkit 11.2.2 haa41c22_9 rocketce cudf 21.10.00a cuda_11.2_py39_ge05bd4bf3c_345 local cupy 10.5.0 py39h0b15df5_0 local dlpack 0.5 h3b9df90_0 conda-forge fastavro 1.4.7 py39h140841e_0 fastrlock 0.6 py39h29c3540_0 fsspec 2022.3.0 py39h6ffa863_0 gflags 2.2.2 he6710b0_0 glog 0.5.0 h29c3540_0 grpc-cpp 1.36.4 h60f4467_pb3.14_2 rocketce krb5 1.19.2 h6205695_0 ld_impl_linux-ppc64le 2.38 hec883e6_1 libblas 3.9.0 11_linuxppc64le_openblas conda-forge libcblas 3.9.0 11_linuxppc64le_openblas conda-forge libcudf 21.10.00a cuda11.2_ge05bd4bf3c_345 local libcurl 7.82.0 ha47bf17_0 libedit 3.1.20210910 h140841e_0 libev 4.33 h140841e_1 libevent 2.1.12 hb567c45_0 libffi 3.3 he6710b0_2 libgcc-ng 11.2.0 h1234567_1 libgomp 11.2.0 h1234567_1 liblapack 3.9.0 11_linuxppc64le_openblas conda-forge libllvm11 11.1.0 h621ed2f_1 libnghttp2 1.46.0 hedb86c2_0 libopenblas 0.3.17 1 rocketce libprotobuf 3.14.0 h5f94dde_0 librmm 21.10.00a cuda11.2_gae27a572_42 local libssh2 1.10.0 hb567c45_0 libstdcxx-ng 11.2.0 h1234567_1 libthrift 0.15.0 heb2aae8_0 libutf8proc 2.6.1 h140841e_0 llvmlite 0.38.0 py39h5f94dde_0 lz4-c 1.9.3 h29c3540_1 ncurses 6.3 h140841e_2 numba 0.55.1 py39haab0e66_0 numexpr 2.8.1 py39h6e379ce_0 numpy 1.21.3 py39he089932_1 conda-forge nvtx 0.2.3 py39ha810350_1 conda-forge openssl 1.1.1o h140841e_0 orc 1.6.5 hb77ef19_2 packaging 21.3 pyhd3eb1b0_0 pandas 1.3.5 py39h724cb3c_0 parquet-cpp 1.5.1 h34088ae_4 pip 21.2.4 py39h6ffa863_0 protobuf 3.14.0 py39h29c3540_1 pyarrow 5.0.0 py39h926b4ae_34_cuda local pyparsing 3.0.4 pyhd3eb1b0_0 python 3.9.12 h836d2c2_1 python-dateutil 2.8.2 pyhd3eb1b0_0 python_abi 3.9 2_cp39 conda-forge pytz 2022.1 py39h6ffa863_0 re2 2022.04.01 h29c3540_0 readline 8.1.2 h140841e_1 rmm 21.10.00a cuda_11.2_py39_gae27a572_42 local setuptools 61.2.0 py39h6ffa863_0 six 1.16.0 pyhd3eb1b0_1 snappy 1.1.9 h29c3540_0 spdlog 1.10.0 h06f31f1_0 conda-forge sqlite 3.38.3 hd7247d8_0 tbb 2021.5.0 h66086b3_0 tk 8.6.12 h7e00dab_0 typing_extensions 4.1.1 pyh06a4308_0 tzdata 2022a hda174b7_0 wheel 0.37.1 pyhd3eb1b0_0 xz 5.2.5 h140841e_0 zlib 1.2.12 h140841e_2 zstd 1.4.9 hc52992f_0

From the documentation of cuda I understood that cuFile is not supported in ppc64le architecture. How can we build the conda package by removing the dependency of cuFile. Quick help on this is appreciated. Thank you.

[quasiben]

Can you try with -DUSE_GDS=ON (https://github.com/rapidsai/cudf/blob/branch-22.08/build.sh#L169) ?

[rnukala1]

Looks like for ppc64le architecture there is no support for cuFile. So, I am not able to install GDS in my power machine. I tried using the argument -DUSE_GDS=ON as well as -DUSE_GDS=OFF and I see the same error.

[github-actions[bot]]

This issue has been labeled inactive-30d due to no recent activity in the past 30 days. Please close this issue if no further response or action is needed. Otherwise, please respond with a comment indicating any updates or changes to the original issue and/or confirm this issue still needs to be addressed. This issue will be labeled inactive-90d if there is no activity in the next 60 days.

[GregoryKimball]

@rnukala1 is this issue still relevant?

[github-actions[bot]]

This issue has been labeled inactive-30d due to no recent activity in the past 30 days. Please close this issue if no further response or action is needed. Otherwise, please respond with a comment indicating any updates or changes to the original issue and/or confirm this issue still needs to be addressed. This issue will be labeled inactive-90d if there is no activity in the next 60 days.

[vyasr]

I'm going to close this issue for now, but please reopen it if the issues are still present.