Skip to content

Latest commit

 

History

History
117 lines (96 loc) · 4.9 KB

README_x86.md

File metadata and controls

117 lines (96 loc) · 4.9 KB
Raw

Instructions for x86_64 platforms

All demos in this repository, with minor tweaks, should also work on x86_64 platforms with NVIDIA GPU(s). Here is a list of required modifications if you'd like to run the demos on an x86_64 PC/server.

Demo #1 (GoogLeNet) and #2 (MTCNN)

  1. When compiling create_engine.cpp, set CUDA_VER, INCPATHS and LIBPATHS based on how you've installed CUDA and TensorRT on your system. For example, I have CUDA 10.0 installed at /usr/local/cuda and TensorRT 5.1.5.0 at /usr/local/TensorRT-5.1.5.0. So I'd modify the following lines in common/Makefile.config:

    ......
CUDA_VER=cuda-10.0
......
INCPATHS    =-I"/usr/local/TensorRT-5.1.5.0/include" -I"$(CUDA_INSTALL_DIR)/include" -I"/usr/local/include" -I"$(CUDNN_INSTALL_DIR)/include" $(TGT_INCLUDES) -I"../common"
LIBPATHS    =-L"/usr/local/TensorRT-5.1.5.0/lib" -L"$(LOCALLIB)" -L"$(CUDA_INSTALL_DIR)/targets/$(TRIPLE)/$(CUDA_LIBDIR)" -L"/usr/local/lib" -L"$(CUDA_INSTALL_DIR)/$(CUDA_LIBDIR)" -L"$(CUDNN_INSTALL_DIR)/$(CUDNN_LIBDIR)" $(TGT_LIBS)
......

  2. Modify setup.py. Add 'include' and 'library' paths for TensorRT. For example, I had to add these 2 lines in setup.py.

     ......
 library_dirs = [
     '/usr/local/cuda/lib64',
+    '/usr/local/TensorRT-5.1.5.0/lib',
     '/usr/local/lib',
 ]
 ......
 include_dirs = [
     ......
     '-I/usr/local/cuda/include',
+    '-I/usr/local/TensorRT-5.1.5.0/include',
     '-I/usr/local/include',
 ]
 ......

  3. Follow the steps in the original README.md, and the demos should work on x86_64 as well.

Demo #3 (SSD)

  1. Install tensorrt, uff and graphsurgeon python3 packages. For example, I do the following on my x86_64 PC since I have TensorRT-5.1.5.0 installed at /usr/local.

    $ export TRT_DIR=/usr/local/TensorRT-5.1.5.0
$ sudo pip3 install ${TRT_DIR}/python/tensorrt-5.1.5.0-cp36-none-linux_x86_64.whl
$ sudo pip3 install ${TRT_DIR}/uff/uff-0.6.3-py2.py3-none-any.whl
$ sudo pip3 install ${TRT_DIR}/graphsurgeon/graphsurgeon-0.4.1-py2.py3-none-any.whl

  2. Patch /usr/local/lib/python3.?/dist-packages/graphsurgeon/node_manipulation.py by adding the following line (around line #42):

     def shape(node):
     ......
     node.name = name or node.name
     node.op = op or node.op or node.name
+    node.attr["dtype"].type = 1
     for key, val in kwargs.items():
     ......

  3. Add TensorRT library path into LD_LIBRARY_PATH environment variable. For example, I add the following line at the end of my ${HOME}/.bashrc and log out/in again.

    export LD_LIBRARY_PATH=/usr/local/TensorRT-5.1.5.0/lib\
                       ${LD_LIBRARY_PATH:+:${LD_LIBRARY_PATH}}

  4. Re-build libflattenconcat.so from TensorRT's 'python/uff_ssd' sample source code. For example,

    $ mkdir -p ${HOME}/src/TensorRT-5.1.5.0
$ cp -r /usr/local/TensorRT-5.1.5.0/samples ${HOME}/src/TensorRT-5.1.5.0
$ cd ${HOME}/src/TensorRT-5.1.5.0/samples/python/uff_ssd
$ mkdir build
$ cd build
$ cmake -D NVINFER_LIB=/usr/local/TensorRT-5.1.5.0/lib/libnvinfer.so \
        -D TRT_INCLUDE=/usr/local/TensorRT-5.1.5.0/include ..
$ make
$ cp libflattenconcat.so ${HOME}/project/tensorrt_demos/ssd/

  5. Install PyCUDA.

    $ sudo apt-get install -y build-essential python-dev
$ sudo apt-get install -y libboost-python-dev libboost-thread-dev
$ sudo pip3 install setuptools
$ export boost_pylib=$(basename /usr/lib/x86_64-linux-gnu/libboost_python3-py3?.so)
$ export boost_pylibname=${boost_pylib%.so}
$ export boost_pyname=${boost_pylibname/lib/}
$ cd ${HOME}/src
$ wget https://files.pythonhosted.org/packages/5e/3f/5658c38579b41866ba21ee1b5020b8225cec86fe717e4b1c5c972de0a33c/pycuda-2019.1.2.tar.gz
$ tar xzvf pycuda-2019.1.2.tar.gz
$ cd pycuda-2019.1.2
$ ./configure.py --python-exe=/usr/bin/python3 \
                 --cuda-root=/usr/local/cuda \
                 --cudadrv-lib-dir=/usr/lib/x86_64-linux-gnu \
                 --boost-inc-dir=/usr/include \
                 --boost-lib-dir=/usr/lib/x86_64-linux-gnu \
                 --boost-python-libname=${boost_pyname} \
                 --boost-thread-libname=boost_thread \
                 --no-use-shipped-boost
$ make -j4
$ python3 setup.py build
$ sudo python3 setup.py install
$ python3 -c "import pycuda; print('pycuda version:', pycuda.VERSION)"

  6. Follow the steps in the original README.md but skip install.sh. You should be able to build the SSD TensorRT engines and run them on on x86_64 as well.

Demo #4 (YOLOv3)

This demo runs on x86 platforms directly. No modification is required.