add linux/windows build workflow (#10)

.github/workflows/linux-x64-gpu.yml

@@ -0,0 +1,61 @@
+name: linux-x64-gpu
+on:
+  push:
+    paths:
+      - '.github/workflows/linux-x64-gpu.yml'
+      - 'src/**'
+      - 'CMakeLists.txt'
+  pull_request:
+    paths:
+      - '.github/workflows/linux-x64-gpu.yml'
+      - 'src/**'
+      - 'CMakeLists.txt'
+concurrency:
+  group: linux-x64-gpu-${{ github.ref }}
+  cancel-in-progress: true
+permissions:
+  contents: read
+
+jobs:
+  build:
+    strategy:
+      matrix:
+        cudaver: [11.8, 12.1]
+    name: cuda-${{ matrix.cudaver }}
+    runs-on: ubuntu-latest
+    steps:
+      - name: Free disk space
+        uses: jlumbroso/free-disk-space@main
+        with:
+          # This might remove tools that are actually needed, if set to "true" but frees about 6 GB
+          tool-cache: false
+          docker-images: false
+          # All of these default to true, but feel free to set to "false" if necessary for your workflow
+          android: true
+          dotnet: true
+          haskell: true
+          large-packages: true
+          swap-storage: false
+      - name: Checkout repository
+        uses: actions/checkout@v3
+      - name: Build
+        uses: addnab/docker-run-action@v3
+        with:
+          image: openmmlab/lmdeploy-builder:cuda${{ matrix.cudaver }}
+          options: -v ${{ github.workspace }}:/work
+          run: |
+            cd /work
+            source /opt/conda/bin/activate
+            conda activate py38
+            mkdir build && cd build
+            cmake .. \
+              -DCMAKE_BUILD_TYPE=RelWithDebInfo \
+              -DCMAKE_EXPORT_COMPILE_COMMANDS=1 \
+              -DCMAKE_INSTALL_PREFIX=./install \
+              -DCMAKE_CUDA_FLAGS="-lineinfo" \
+              -DUSE_NVTX=ON \
+              -DBUILD_TEST=ON
+            make -j$(nproc) && make install
+            cd ..
+            rm -rf build
+            python setup.py bdist_wheel --plat-name manylinux2014_x86_64 -d /tmp

.github/workflows/windows-x64-gpu.yml

@@ -0,0 +1,57 @@
+name: windows-x64-gpu
+on:
+  push:
+    paths:
+      - '.github/workflows/windows-x64-gpu.yml'
+      - 'src/**'
+      - 'CMakeLists.txt'
+  pull_request:
+    paths:
+      - '.github/workflows/windows-x64-gpu.yml'
+      - 'src/**'
+      - 'CMakeLists.txt'
+concurrency:
+  group: windows-x64-gpu-${{ github.ref }}
+  cancel-in-progress: true
+permissions:
+  contents: read
+
+jobs:
+  build:
+    strategy:
+      matrix:
+        cudaver: [11.8.0, 12.1.0]
+    name: cuda-${{ matrix.cudaver }}
+    runs-on: windows-latest
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@v3
+      - name: Set up python
+        uses: actions/setup-python@v4
+        with:
+          python-version: '3.8'
+      - name: Install python packages
+        run: |
+          pip install -r requirements/build.txt
+          pip install wheel
+      - name: Setup CUDA Toolkit
+        id: cuda-toolkit
+        shell: pwsh
+        run: ./builder/windows/setup_cuda.ps1
+        env:
+            INPUT_CUDA_VERSION: ${{ matrix.cudaver }}
+      - name: Build wheel
+        run: |
+          $env:BUILD_TEST="ON"
+          mkdir build
+          cd build
+          ..\builder\windows\generate.ps1
+          cmake --build . --config Release -- /m /v:q
+          if (-Not $?) {
+            echo "build failed"
+            exit 1
+          }
+          cmake --install . --config Release
+          cd ..
+          rm build -Force -Recurse
+          python setup.py bdist_wheel -d build/wheel

CMakeLists.txt

@@ -12,14 +12,11 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-cmake_minimum_required(VERSION 3.18 FATAL_ERROR)
+cmake_minimum_required(VERSION 3.11 FATAL_ERROR)
 project(TurboMind LANGUAGES CXX CUDA)
 
 find_package(CUDA 11.4 REQUIRED)
 
-
-set(CMAKE_MODULE_PATH ${PROJECT_SOURCE_DIR}/cmake/Modules)
-
 option(BUILD_TEST "Build tests" OFF)
 
 include(FetchContent)
@@ -51,8 +48,6 @@ set(CMAKE_SHARED_LINKER_FLAGS "${CMAKE_SHARED_LINKER_FLAGS} -fuse-ld=gold")
 set(CUDA_PATH ${CUDA_TOOLKIT_ROOT_DIR})
 
 
-list(APPEND CMAKE_MODULE_PATH ${CUDA_PATH}/lib64)
-
 # profiling
 option(USE_NVTX "Whether or not to use nvtx" ON)
 if(USE_NVTX)

generate.sh

@@ -10,10 +10,7 @@ fi
 cmake ${builder} .. \
     -DCMAKE_BUILD_TYPE=RelWithDebInfo \
     -DCMAKE_EXPORT_COMPILE_COMMANDS=1 \
+    -DCMAKE_INSTALL_PREFIX=${WORKSPACE_PATH}/install \
     -DCMAKE_CUDA_FLAGS="-lineinfo" \
     -DUSE_NVTX=ON \
-    -DBUILD_TEST=ON \
-    -DFETCHCONTENT_UPDATES_DISCONNECTED=ON \
-    -DLMDEPLOY_ASAN_ENABLE=OFF \
-    -DLMDEPLOY_UBSAN_ENABLE=OFF \
-    -DCMAKE_CUDA_ARCHITECTURES="80-real"
+    -DBUILD_TEST=OFF

src/turbomind/api/python/linear.cc

@@ -5,6 +5,7 @@
 #include "src/turbomind/kernels/gemm/gemm.h"
 #include "src/turbomind/kernels/gemm/types.h"
 #include "src/turbomind/utils/cuda_utils.h"
+#include "src/turbomind/utils/macro.h"
 #include <cuda_fp16.h>
 #include <cuda_runtime.h>
 #include <fstream>