support ollama run on Intel GPUs

This commit is contained in:
Wang,Zhe
2024-05-24 11:18:27 +08:00
parent 7ca71a6b0f
commit fd5971be0b
7 changed files with 615 additions and 32 deletions

View File

@@ -206,6 +206,36 @@ if [ -d "${CUDA_LIB_DIR}" ]; then
fi
if [ -z "${ONEAPI_ROOT}" ]; then
# Try the default location in case it exists
ONEAPI_ROOT=/opt/intel/oneapi
fi
if [ -d "${ONEAPI_ROOT}" ]; then
echo "OneAPI libraries detected - building dynamic OneAPI library"
init_vars
source ${ONEAPI_ROOT}/setvars.sh --force # set up environment variables for oneAPI
CC=icx
CMAKE_DEFS="${COMMON_CMAKE_DEFS} ${CMAKE_DEFS} -DCMAKE_C_COMPILER=icx -DCMAKE_CXX_COMPILER=icpx -DLLAMA_SYCL=ON -DLLAMA_SYCL_F16=OFF"
BUILD_DIR="../build/linux/${ARCH}/oneapi"
EXTRA_LIBS="-fsycl -Wl,-rpath,${ONEAPI_ROOT}/compiler/latest/lib,-rpath,${ONEAPI_ROOT}/mkl/latest/lib,-rpath,${ONEAPI_ROOT}/tbb/latest/lib,-rpath,${ONEAPI_ROOT}/compiler/latest/opt/oclfpga/linux64/lib -lOpenCL -lmkl_core -lmkl_sycl_blas -lmkl_intel_ilp64 -lmkl_tbb_thread -ltbb"
DEBUG_FLAGS="" # icx compiles with -O0 if we pass -g, so we must remove it
build
# copy oneAPI dependencies
for dep in $(ldd "${BUILD_DIR}/bin/ollama_llama_server" | grep "=>" | cut -f2 -d= | cut -f2 -d' ' | grep -e sycl -e mkl -e tbb); do
cp "${dep}" "${BUILD_DIR}/bin/"
done
cp "${ONEAPI_ROOT}/compiler/latest/lib/libOpenCL.so" "${BUILD_DIR}/bin/"
cp "${ONEAPI_ROOT}/compiler/latest/lib/libimf.so" "${BUILD_DIR}/bin/"
cp "${ONEAPI_ROOT}/compiler/latest/lib/libintlc.so.5" "${BUILD_DIR}/bin/"
cp "${ONEAPI_ROOT}/compiler/latest/lib/libirng.so" "${BUILD_DIR}/bin/"
cp "${ONEAPI_ROOT}/compiler/latest/lib/libpi_level_zero.so" "${BUILD_DIR}/bin/"
cp "${ONEAPI_ROOT}/compiler/latest/lib/libsvml.so" "${BUILD_DIR}/bin/"
cp "${ONEAPI_ROOT}/compiler/latest/lib/libur_loader.so.0" "${BUILD_DIR}/bin/"
compress
fi
if [ -z "${ROCM_PATH}" ]; then
# Try the default location in case it exists
ROCM_PATH=/opt/rocm

View File

@@ -289,6 +289,49 @@ function build_cuda() {
}
}
function build_oneapi() {
if ((-not "${env:OLLAMA_SKIP_CUDA_GENERATE}") -and ("${env:ONEAPI_ROOT}")) {
# Get oneAPI version
$script:ONEAPI_VERSION = icpx --version
$script:ONEAPI_VERSION = [regex]::Match($script:ONEAPI_VERSION, '(?<=oneAPI DPC\+\+/C\+\+ Compiler )(?<version>\d+\.\d+\.\d+)').Value
if ($null -ne $script:ONEAPI_VERSION) {
$script:ONEAPI_VARIANT = "_v" + $script:ONEAPI_VERSION
}
init_vars
$script:buildDir = "../build/windows/${script:ARCH}/oneapi$script:ONEAPI_VARIANT"
$script:distDir ="$script:DIST_BASE\oneapi$script:ONEAPI_VARIANT"
$script:cmakeDefs += @(
"-G", "MinGW Makefiles",
"-DLLAMA_SYCL=ON",
"-DCMAKE_C_COMPILER=icx",
"-DCMAKE_CXX_COMPILER=icx",
"-DCMAKE_BUILD_TYPE=Release"
)
Write-Host "Building oneAPI"
build
# Ninja doesn't prefix with config name
if ($null -ne $script:DUMPBIN) {
& "$script:DUMPBIN" /dependents "${script:buildDir}/bin/ollama_llama_server.exe" | Select-String ".dll"
}
sign
install
cp "${env:ONEAPI_ROOT}\compiler\latest\bin\libirngmd.dll" "${script:distDir}"
cp "${env:ONEAPI_ROOT}\compiler\latest\bin\libmmd.dll" "${script:distDir}"
cp "${env:ONEAPI_ROOT}\compiler\latest\bin\pi_level_zero.dll" "${script:distDir}"
cp "${env:ONEAPI_ROOT}\compiler\latest\bin\pi_unified_runtime.dll" "${script:distDir}"
cp "${env:ONEAPI_ROOT}\compiler\latest\bin\pi_win_proxy_loader.dll" "${script:distDir}"
cp "${env:ONEAPI_ROOT}\compiler\latest\bin\svml_dispmd.dll" "${script:distDir}"
cp "${env:ONEAPI_ROOT}\compiler\latest\bin\sycl7.dll" "${script:distDir}"
cp "${env:ONEAPI_ROOT}\mkl\latest\bin\mkl_core.2.dll" "${script:distDir}"
cp "${env:ONEAPI_ROOT}\mkl\latest\bin\mkl_sycl_blas.4.dll" "${script:distDir}"
cp "${env:ONEAPI_ROOT}\mkl\latest\bin\mkl_tbb_thread.2.dll" "${script:distDir}"
} else {
Write-Host "Skipping oneAPI generation step"
}
}
function build_rocm() {
if ((-not "${env:OLLAMA_SKIP_ROCM_GENERATE}") -and ("${env:HIP_PATH}")) {
$script:ROCM_VERSION=(get-item $env:HIP_PATH).Basename
@@ -356,6 +399,7 @@ if ($($args.count) -eq 0) {
build_cpu_avx
build_cpu_avx2
build_cuda
build_oneapi
build_rocm
}