Fix build and update to CUDA 12.4

.github/workflows/ci.yml

@@ -132,7 +132,7 @@ jobs:
     runs-on: ${{ matrix.os }}
     strategy:
       matrix:
-        os: [ubuntu-22.04, windows-2019]
+        os: [ubuntu-22.04, windows-2022]
         arch: [auto64]
         include:
         - os: ubuntu-22.04
@@ -183,13 +183,13 @@ jobs:
     runs-on: ${{ matrix.os }}
     strategy:
       matrix:
-        os: [ubuntu-22.04, windows-2019]
+        os: [ubuntu-22.04, windows-2022]
 
     steps:
-      - name: Set up Python 3.9
+      - name: Set up Python 3.10.11
         uses: actions/setup-python@v5
         with:
-          python-version: 3.9
+          python-version: "3.10.11"
 
       - uses: actions/checkout@v4
 
@@ -208,13 +208,13 @@ jobs:
       - name: Install wheel
         if: startsWith(matrix.os, 'ubuntu')
         run: |
-          pip install *cp39*manylinux*x86_64.whl
+          pip install *cp310*manylinux*x86_64.whl
 
       - name: Install wheel
         if: startsWith(matrix.os, 'windows')
         shell: bash
         run: |
-          pip install *cp39*win*.whl
+          pip install *cp310*win*.whl
 
       - name: Run tests
         shell: bash
@@ -228,10 +228,10 @@ jobs:
     steps:
       - uses: actions/checkout@v4
 
-      - name: Set up Python 3.9
+      - name: Set up Python 3.10.11
         uses: actions/setup-python@v5
         with:
-          python-version: 3.9
+          python-version: "3.10.11"
 
       - name: Install dependencies
         run: |
@@ -307,10 +307,10 @@ jobs:
     steps:
       - uses: actions/checkout@v4
 
-      - name: Set up Python 3.9
+      - name: Set up Python 3.10.11
         uses: actions/setup-python@v5
         with:
-          python-version: 3.9
+          python-version: "3.10.11"
 
       - name: Download CTranslate2 wheels
         uses: actions/download-artifact@v4
@@ -321,7 +321,7 @@ jobs:
 
       - name: Install CTranslate2 wheel
         run: |
-          pip install *cp39*manylinux*x86_64.whl
+          pip install *cp310*manylinux*x86_64.whl
 
       - name: Install dependencies to build docs
         working-directory: docs

python/tools/prepare_build_environment_linux.sh

@@ -19,20 +19,21 @@ if [ "$CIBW_ARCHS" == "aarch64" ]; then
     rm -r OpenBLAS-*
 
 else
-    # Install CUDA 12.2:
+    # Install CUDA 12.4:
     yum-config-manager --add-repo https://developer.download.nvidia.com/compute/cuda/repos/rhel8/x86_64/cuda-rhel8.repo
     # error mirrorlist.centos.org doesn't exists anymore.
     sed -i s/mirror.centos.org/vault.centos.org/g /etc/yum.repos.d/*.repo
     sed -i s/^#.*baseurl=http/baseurl=http/g /etc/yum.repos.d/*.repo
     sed -i s/^mirrorlist=http/#mirrorlist=http/g /etc/yum.repos.d/*.repo
     yum install --setopt=obsoletes=0 -y \
-        cuda-nvcc-12-2-12.2.140-1 \
-        cuda-cudart-devel-12-2-12.2.140-1 \
-        libcurand-devel-12-2-10.3.3.141-1 \
+        cuda-nvcc-12-4-12.4.99-1 \
+        cuda-cudart-devel-12-4-12.4.99-1 \
+        libcurand-devel-12-4-10.3.5.119-1 \
         libcudnn9-devel-cuda-12-9.1.0.70-1 \
-        libcublas-devel-12-2-12.2.5.6-1 \
-        libnccl-devel-2.19.3-1+cuda12.2
-    ln -s cuda-12.2 /usr/local/cuda
+        libcublas-devel-12-4-12.4.2.65-1 \
+        libnccl-2.20.5-1+cuda12.4 \
+        libnccl-devel-2.20.5-1+cuda12.4
+    ln -s cuda-12.4 /usr/local/cuda
 
     ONEAPI_VERSION=2023.2.0
     yum-config-manager --add-repo https://yum.repos.intel.com/oneapi

python/tools/prepare_build_environment_macos.sh

@@ -1,9 +1,24 @@
-#! /bin/bash
-
+#!/bin/bash
 set -e
 set -x
 
 pip install "cmake==3.18.4"
+brew install libomp
+
+# Get the actual libomp path
+LIBOMP_PREFIX=$(brew --prefix libomp)
+
+# Set environment variables
+export LDFLAGS="-L${LIBOMP_PREFIX}/lib"
+export CPPFLAGS="-I${LIBOMP_PREFIX}/include"
+export CMAKE_PREFIX_PATH="${LIBOMP_PREFIX}"
+
+# Critical: Set OpenMP flags explicitly for CMake
+export OpenMP_C_FLAGS="-Xpreprocessor -fopenmp -I${LIBOMP_PREFIX}/include"
+export OpenMP_C_LIB_NAMES="omp"
+export OpenMP_CXX_FLAGS="-Xpreprocessor -fopenmp -I${LIBOMP_PREFIX}/include"
+export OpenMP_CXX_LIB_NAMES="omp"
+export OpenMP_omp_LIBRARY="${LIBOMP_PREFIX}/lib/libomp.dylib"
 
 mkdir build-release && cd build-release
 
@@ -26,7 +41,19 @@ else
     wget -q https://github.com/oneapi-src/oneDNN/archive/refs/tags/v${ONEDNN_VERSION}.tar.gz
     tar xf *.tar.gz && rm *.tar.gz
     cd oneDNN-*
-    cmake -DCMAKE_BUILD_TYPE=Release -DONEDNN_LIBRARY_TYPE=STATIC -DONEDNN_BUILD_EXAMPLES=OFF -DONEDNN_BUILD_TESTS=OFF -DONEDNN_ENABLE_WORKLOAD=INFERENCE -DONEDNN_ENABLE_PRIMITIVE="CONVOLUTION;REORDER" -DONEDNN_BUILD_GRAPH=OFF .
+    cmake -DCMAKE_POLICY_VERSION_MINIMUM=3.5 \
+          -DCMAKE_BUILD_TYPE=Release \
+          -DONEDNN_LIBRARY_TYPE=STATIC \
+          -DONEDNN_BUILD_EXAMPLES=OFF \
+          -DONEDNN_BUILD_TESTS=OFF \
+          -DONEDNN_ENABLE_WORKLOAD=INFERENCE \
+          -DONEDNN_ENABLE_PRIMITIVE="CONVOLUTION;REORDER" \
+          -DONEDNN_BUILD_GRAPH=OFF \
+          -DOpenMP_C_FLAGS="${OpenMP_C_FLAGS}" \
+          -DOpenMP_CXX_FLAGS="${OpenMP_CXX_FLAGS}" \
+          -DOpenMP_omp_LIBRARY="${OpenMP_omp_LIBRARY}" \
+          -DCMAKE_C_FLAGS="${CPPFLAGS}" \
+          -DCMAKE_CXX_FLAGS="${CPPFLAGS}" .
     make -j$(sysctl -n hw.physicalcpu_max) install
     cd ..
     rm -r oneDNN-*
@@ -35,7 +62,15 @@ else
 
 fi
 
-cmake -DCMAKE_BUILD_TYPE=Release -DBUILD_CLI=OFF -DCMAKE_INSTALL_RPATH_USE_LINK_PATH=ON $CMAKE_EXTRA_OPTIONS ..
+cmake -DCMAKE_POLICY_VERSION_MINIMUM=3.5 \
+      -DCMAKE_BUILD_TYPE=Release \
+      -DBUILD_CLI=OFF \
+      -DCMAKE_INSTALL_RPATH_USE_LINK_PATH=ON \
+      -DOpenMP_C_FLAGS="${OpenMP_C_FLAGS}" \
+      -DOpenMP_CXX_FLAGS="${OpenMP_CXX_FLAGS}" \
+      -DOpenMP_omp_LIBRARY="${OpenMP_omp_LIBRARY}" \
+      $CMAKE_EXTRA_OPTIONS ..
+
 VERBOSE=1 make -j$(sysctl -n hw.physicalcpu_max) install
 cd ..
 rm -r build-release

python/tools/prepare_build_environment_windows.sh

@@ -3,9 +3,10 @@
 set -e
 set -x
 
-CUDA_ROOT="C:/Program Files/NVIDIA GPU Computing Toolkit/CUDA/v12.2"
-curl --netrc-optional -L -nv -o cuda.exe https://developer.download.nvidia.com/compute/cuda/12.2.2/local_installers/cuda_12.2.2_537.13_windows.exe
-./cuda.exe -s nvcc_12.2 cudart_12.2 cublas_dev_12.2 curand_dev_12.2
+CUDA_ROOT="C:/Program Files/NVIDIA GPU Computing Toolkit/CUDA/v12.4"
+curl --netrc-optional -L -nv -o cuda.exe https://developer.download.nvidia.com/compute/cuda/12.4.0/local_installers/cuda_12.4.0_551.61_windows.exe
+./cuda.exe -s nvcc_12.4 cudart_12.4 cublas_dev_12.4 curand_dev_12.4
+
 rm cuda.exe
 
 CUDNN_ROOT="C:/Program Files/NVIDIA/CUDNN/v9.1"
@@ -30,7 +31,7 @@ cp -r "$CUDNN_ROOT"/* "$CUDA_ROOT"
 rm cudnn.exe
 
 # See https://github.com/oneapi-src/oneapi-ci for installer URLs
-curl --netrc-optional -L -nv -o webimage.exe https://registrationcenter-download.intel.com/akdlm/irc_nas/19078/w_BaseKit_p_2023.0.0.25940_offline.exe
+curl --netrc-optional -L -nv -o webimage.exe https://registrationcenter-download.intel.com/akdlm/IRC_NAS/62641e01-1e8d-4ace-91d6-ae03f7f8a71f/w_BaseKit_p_2024.0.0.49563_offline.exe
 ./webimage.exe -s -x -f webimage_extracted --log extract.log
 rm webimage.exe
 ./webimage_extracted/bootstrapper.exe -s --action install --components="intel.oneapi.win.mkl.devel" --eula=accept -p=NEED_VS2017_INTEGRATION=0 -p=NEED_VS2019_INTEGRATION=0 --log-dir=.
@@ -53,5 +54,5 @@ rm -r build
 
 cp README.md python/
 cp $CTRANSLATE2_ROOT/bin/ctranslate2.dll python/ctranslate2/
-cp "C:/Program Files (x86)/Intel/oneAPI/compiler/latest/windows/redist/intel64_win/compiler/libiomp5md.dll" python/ctranslate2/
+cp "C:/Program Files (x86)/Intel/oneAPI/compiler/2024.0/bin/libiomp5md.dll" python/ctranslate2/
 cp "$CUDA_ROOT/bin/cudnn64_9.dll" python/ctranslate2/