Arch Linux User Repository

I'm getting the following error:

-- Build files have been written to: /tmp/makepkg/tensorrt/src/build
make: Entering directory '/tmp/makepkg/tensorrt/src/build'
[  2%] Built target third_party.protobuf
[  2%] Built target gen_onnx_proto
[  2%] Built target caffe_proto
[  2%] Built target gen_onnx_operators_proto
[  2%] Built target gen_onnx_data_proto
[  4%] Built target onnx_proto
[  8%] Built target nvcaffeparser_static
[ 12%] Built target nvcaffeparser
[ 17%] Built target nvonnxparser_static
[ 17%] Built target nvonnxparser
make[2]: *** No rule to make target '/opt/cuda/lib/libcudart_static.a', needed by 'libnvinfer_plugin.so.8.5.3'.  Stop.
make[1]: *** [CMakeFiles/Makefile2:1166: plugin/CMakeFiles/nvinfer_plugin.dir/all] Error 2
make[1]: *** Waiting for unfinished jobs....
[ 58%] Built target nvinfer_plugin_static
make: *** [Makefile:156: all] Error 2
make: Leaving directory '/tmp/makepkg/tensorrt/src/build'
==> ERROR: A failure occurred in build().
    Aborting...

I've downgraded CUDA and CUDNN to match the tensorRT version, but now this is happening.. Any hints ?

/opt/cuda/lib

does not exits, but instead there is lib64

Even if I create a symlink to lib, it does not work... Thank you

Oh ok. Look like I've been very unlucky and decided to install this on the wrong day. CUDA was just updated in the repos, yesterday's version is working fine. I downgraded my system using the Arch Linux Archive.

I'm getting this error:

[ 24%] Building CXX object plugin/CMakeFiles/nvinfer_plugin.dir/bertQKVToContextPlugin/qkvToContextPlugin.cpp.o
[ 24%] Building CXX object plugin/CMakeFiles/nvinfer_plugin.dir/embLayerNormPlugin/embLayerNormPlugin.cpp.o
[ 24%] Building CXX object plugin/CMakeFiles/nvinfer_plugin.dir/embLayerNormPlugin/embLayerNormVarSeqlenPlugin.cpp.o
[ 24%] Building CXX object plugin/CMakeFiles/nvinfer_plugin.dir/fcPlugin/fcPlugin.cpp.o
In file included from /home/smoolak/.cache/yay/tensorrt/src/TensorRT/plugin/common/bertCommon.h:26,
                 from /home/smoolak/.cache/yay/tensorrt/src/TensorRT/plugin/fcPlugin/fcPlugin.h:27,
                 from /home/smoolak/.cache/yay/tensorrt/src/TensorRT/plugin/fcPlugin/fcPlugin.cpp:24:
/home/smoolak/.cache/yay/tensorrt/src/TensorRT/plugin/fcPlugin/fcPlugin.h: In member function ‘void nvinfer1::plugin::bert::AlgoProps::populate(const cublasLtMatmulAlgo_t&)’:
/home/smoolak/.cache/yay/tensorrt/src/TensorRT/plugin/fcPlugin/fcPlugin.h:379:25: error: ‘CUBLASLT_ALGO_CAP_MATHMODE_IMPL’ was not declared in this scope; did you mean ‘CUBLASLT_ALGO_CAP_TILE_IDS’?
  379 |             matmulAlgo, CUBLASLT_ALGO_CAP_MATHMODE_IMPL, &mathMode, sizeof(mathMode), nullptr));
      |                         ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
/home/smoolak/.cache/yay/tensorrt/src/TensorRT/plugin/common/checkMacrosPlugin.h:215:19: note: in definition of macro ‘PLUGIN_CUBLASASSERT’
  215 |         auto s_ = status_;                                                                                             \
      |                   ^~~~~~~
/home/smoolak/.cache/yay/tensorrt/src/TensorRT/plugin/fcPlugin/fcPlugin.cpp: In function ‘void LtGemmSearch(cublasLtHandle_t, cublasOperation_t, cublasOperation_t, const int&, const int&, const int&, const void*, const void*, const int&, const void*, const int&, const void*, void*, const int&, void*, size_t, cublasComputeType_t, cudaDataType_t, cudaDataType_t, cudaDataType_t, cudaDataType_t, std::vector<nvinfer1::plugin::bert::customMatMultPerfType_t>&)’:
/home/smoolak/.cache/yay/tensorrt/src/TensorRT/plugin/fcPlugin/fcPlugin.cpp:171:21: error: ‘CUBLASLT_MATMUL_PREF_MATH_MODE_MASK’ was not declared in this scope; did you mean ‘CUBLASLT_MATMUL_PREF_IMPL_MASK’?
  171 |         preference, CUBLASLT_MATMUL_PREF_MATH_MODE_MASK, &mathMode, sizeof(mathMode)));
      |                     ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
/home/smoolak/.cache/yay/tensorrt/src/TensorRT/plugin/common/checkMacrosPlugin.h:215:19: note: in definition of macro ‘PLUGIN_CUBLASASSERT’
  215 |         auto s_ = status_;                                                                                             \
      |                   ^~~~~~~
/home/smoolak/.cache/yay/tensorrt/src/TensorRT/plugin/fcPlugin/fcPlugin.cpp:218:20: error: ‘CUBLASLT_ALGO_CAP_MATHMODE_IMPL’ was not declared in this scope; did you mean ‘CUBLASLT_ALGO_CAP_TILE_IDS’?
  218 |             &algo, CUBLASLT_ALGO_CAP_MATHMODE_IMPL, &mathMode, sizeof(mathMode), nullptr));
      |                    ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
/home/smoolak/.cache/yay/tensorrt/src/TensorRT/plugin/common/checkMacrosPlugin.h:215:19: note: in definition of macro ‘PLUGIN_CUBLASASSERT’
  215 |         auto s_ = status_;                                                                                             \
      |                   ^~~~~~~
make[2]: *** [plugin/CMakeFiles/nvinfer_plugin.dir/build.make:2260: plugin/CMakeFiles/nvinfer_plugin.dir/fcPlugin/fcPlugin.cpp.o] Error 1
make[1]: *** [CMakeFiles/Makefile2:1166: plugin/CMakeFiles/nvinfer_plugin.dir/all] Error 2
make: *** [Makefile:156: all] Error 2
make: Leaving directory '/home/smoolak/.cache/yay/tensorrt/src/build'
==> ERROR: A failure occurred in build().
    Aborting...
 -> error making: tensorrt

@thoth You can refer to my workaround solution below.

Mine fails with:

HEAD is now at 914c06fb chore: set to version 2.9.2
/tmp/build/tensorrt/src/TensorRT/python/build /tmp/build/tensorrt/src/TensorRT/python
-- The CXX compiler identification is GNU 12.2.0
-- The C compiler identification is GNU 12.2.0
-- Detecting CXX compiler ABI info
-- Detecting CXX compiler ABI info - done
-- Check for working CXX compiler: /usr/bin/c++ - skipped
-- Detecting CXX compile features
-- Detecting CXX compile features - done
-- Detecting C compiler ABI info
-- Detecting C compiler ABI info - done
-- Check for working C compiler: /usr/bin/cc - skipped
-- Detecting C compile features
-- Detecting C compile features - done
-- Configuring done
CMake Error: The following variables are used in this project, but they are set to NOTFOUND.
Please set them or make sure they are set and tested correctly in the CMake files:
/tmp/build/tensorrt/src/TensorRT/python/PY_CONFIG_INCLUDE
   used as include directory in directory /tmp/build/tensorrt/src/TensorRT/python
/tmp/build/tensorrt/src/TensorRT/python/PY_INCLUDE
   used as include directory in directory /tmp/build/tensorrt/src/TensorRT/python

-- Generating done
CMake Generate step failed.  Build files cannot be regenerated correctly.
==> ERROR: A failure occurred in build().
    Aborting...

@otakutyrant license prohibits the package distribution. Unless Nvidia grants the package distribution within Arch Linux, we cannot do it.

@Baytars It may be. You can always build the package in a clean chroot to make sure that it builds fine. This is how all official repository packages are built, and I build all my packages in this way.

This package is more unpopular than I thought. It should be in official packages.

@dbermond Is it related to using virtual environment in Anaconda? The output log showed that, instead of the system Python distribution, the Python path it found was that managed by Anaconda, which I use routinely. So despite the existence of makedepends, PKGBUILD could not find the dependencies in the virtual environment.