[Cleanup] Cleanup unused CMake options (#5470)

* cleanup unused cmake options

* disable BUILD_TORCH for cugraph

* resolve comments

CMakeLists.txt

@@ -33,6 +33,7 @@ dgl_option(USE_HDFS "Build with HDFS support" OFF) # Set env HADOOP_HDFS_HOME if
 dgl_option(REBUILD_LIBXSMM "Clean LIBXSMM build cache at every build" OFF) # Set env HADOOP_HDFS_HOME if needed
 dgl_option(USE_EPOLL "Build with epoll for socket communicator" ON)
 dgl_option(TP_BUILD_LIBUV "Build libuv together with tensorpipe (only impacts Linux)" ON)
+dgl_option(BUILD_TORCH "Build the PyTorch plugin" OFF)
 dgl_option(BUILD_SPARSE "Build DGL sparse library" ON)
 dgl_option(TORCH_PYTHON_INTERPS "Python interpreter used to build tensoradapter and DGL sparse library" python3)
 

benchmarks/scripts/build_dgl_asv.sh

@@ -11,11 +11,7 @@ pip install -r /asv/torch_gpu_pip.txt
 # 'CUDA_TOOLKIT_ROOT_DIR' is always required for sparse build as torch1.13.1+cu116 is installed.
 CMAKE_VARS="-DUSE_OPENMP=ON -DBUILD_TORCH=ON -DCUDA_TOOLKIT_ROOT_DIR=/usr/local/cuda"
 if [[ $DEVICE == "gpu" ]]; then
-    CMAKE_VARS="-DUSE_CUDA=ON -DUSE_NCCL=ON $CMAKE_VARS"
-fi
-arch=`uname -m`
-if [[ $arch == *"x86"* ]]; then
-  CMAKE_VARS="-DUSE_AVX=ON $CMAKE_VARS"
+    CMAKE_VARS="-DUSE_CUDA=ON $CMAKE_VARS"
 fi
 mkdir -p build
 pushd build

cmake/config.cmake

@@ -40,14 +40,11 @@ set(BUILD_CPP_TEST OFF)
 # Whether to enable OpenMP.
 set(USE_OPENMP ON)
 
-# Whether to enable Intel's avx optimized kernel.
-set(USE_AVX ON)
-
 # Whether to build PyTorch plugins.
 set(BUILD_TORCH ON)
 
 # Whether to enable CUDA kernels compiled with TVM.
 set(USE_TVM OFF)
 
-# Whether to enable fp16 to support mixed precision training.
-set(USE_FP16 OFF)
+# Whether to build DGL sparse library.
+set(BUILD_SPARSE ON)

docs/source/guide_ko/mixed_precision.rst

@@ -5,28 +5,7 @@
 
 :ref:`(English Version) <guide-mixed_precision>`
 
-DGL은 mixed precision 학습을 위해서 `PyTorch's automatic mixed precision package <https://pytorch.org/docs/stable/amp.html>`_ 와 호환된다. 따라서, 학습 시간 및 GPU 메모리 사용량을 절약할 수 있다. 이 기능을 활성화하기 위해서는, PyTorch 1.6+, python 3.7+을 설치하고, ``float16`` 데이터 타입 지원을 위해서 DGL을 소스 파일을 사용해서 빌드해야 한다. (이 기능은 아직 베타 단계이고, pre-built pip wheel 형태로 제공하지 않는다.)
-
-설치
----------
-
-우선 DGL 소스 코드를 GitHub에서 다운로드하고, ``USE_FP16=ON`` 플래그를 사용해서 shared library를 빌드한다.
-
-.. code:: bash
-
-   git clone --recurse-submodules https://github.com/dmlc/dgl.git
-   cd dgl
-   mkdir build
-   cd build
-   cmake -DUSE_CUDA=ON -DUSE_FP16=ON ..
-   make -j
-
-다음으로 Python 바인딩을 설치한다.
-
-.. code:: bash
-
-   cd ../python
-   python setup.py install
+DGL은 mixed precision 학습을 위해서 `PyTorch's automatic mixed precision package <https://pytorch.org/docs/stable/amp.html>`_ 와 호환된다. 따라서, 학습 시간 및 GPU 메모리 사용량을 절약할 수 있다. 
 
 Half precision을 사용한 메시지 전달
 ------------------------------

docs/source/install/index.rst

@@ -115,7 +115,7 @@ install the Python binding for DGL.
 
    mkdir build
    cd build
-   cmake -DUSE_OPENMP=off -DCMAKE_C_FLAGS='-DXBYAK_DONT_USE_MAP_JIT' -DCMAKE_CXX_FLAGS='-DXBYAK_DONT_USE_MAP_JIT' -DUSE_AVX=OFF -DUSE_LIBXSMM=OFF ..
+   cmake -DUSE_OPENMP=off -DCMAKE_C_FLAGS='-DXBYAK_DONT_USE_MAP_JIT' -DCMAKE_CXX_FLAGS='-DXBYAK_DONT_USE_MAP_JIT' -DUSE_LIBXSMM=OFF ..
    make -j4
    cd ../python
    python setup.py install

script/build_dgl.sh

@@ -10,7 +10,7 @@ examples:
   Start a CUDA build: bash $0 -g
   Build incrementally: bash $0
   Remove all intermediate output and restart a CPU only build: bash $0 -c -r
-  Build with extra cmake arguments: bash $0 -c -e '-DUSE_AVX=ON'
+  Build with extra cmake arguments: bash $0 -c -e '-DBUILD_TORCH=ON'
 
 Build DGL. By default, build incrementally on top of the current state.
 

tests/scripts/build_dgl.bat

@@ -13,7 +13,7 @@ SET TEMP=%WORKSPACE%\tmp
 SET TMPDIR=%WORKSPACE%\tmp
 
 PUSHD build
-cmake -DCMAKE_CXX_FLAGS="/DDGL_EXPORTS" -DUSE_AVX=ON -DUSE_OPENMP=ON -DBUILD_TORCH=ON -Dgtest_force_shared_crt=ON -DDMLC_FORCE_SHARED_CRT=ON -DBUILD_CPP_TEST=1 -DCMAKE_CONFIGURATION_TYPES="Release" -DTORCH_PYTHON_INTERPS=python -DBUILD_SPARSE=ON .. -G "Visual Studio 16 2019" || EXIT /B 1
+cmake -DCMAKE_CXX_FLAGS="/DDGL_EXPORTS" -DUSE_OPENMP=ON -DBUILD_TORCH=ON -Dgtest_force_shared_crt=ON -DDMLC_FORCE_SHARED_CRT=ON -DBUILD_CPP_TEST=1 -DCMAKE_CONFIGURATION_TYPES="Release" -DTORCH_PYTHON_INTERPS=python -DBUILD_SPARSE=ON .. -G "Visual Studio 16 2019" || EXIT /B 1
 msbuild dgl.sln /m /nr:false || EXIT /B 1
 COPY /Y Release\runUnitTests.exe .
 POPD

tests/scripts/build_dgl.sh

@@ -18,13 +18,7 @@ if [ "$1" != "cugraph" ]; then
     CMAKE_VARS="$CMAKE_VARS -DBUILD_TORCH=ON -DTORCH_PYTHON_INTERPS=/opt/conda/envs/pytorch-ci/bin/python"
 else
     # Disable sparse build as cugraph docker image lacks cuDNN.
-    CMAKE_VARS="$CMAKE_VARS -DBUILD_SPARSE=OFF"
-fi
-
-#This is implemented to detect underlying architecture and enable arch specific optimization.
-arch=`uname -m`
-if [[ $arch == *"x86"* ]]; then
-  CMAKE_VARS="-DUSE_AVX=ON $CMAKE_VARS"
+    CMAKE_VARS="$CMAKE_VARS -DBUILD_TORCH=OFF -DBUILD_SPARSE=OFF"
 fi
 
 if [[ $1 != "cpu" ]]; then