2023-11-29 19:00:37 +00:00
|
|
|
#!/bin/bash
|
2023-11-14 01:20:34 +00:00
|
|
|
# This script is intended to run inside the go generate
|
2024-01-04 17:40:15 +00:00
|
|
|
# working directory must be llm/generate/
|
2023-12-20 18:36:01 +00:00
|
|
|
|
|
|
|
# First we build our default built-in library which will be linked into the CGO
|
|
|
|
# binary as a normal dependency. This default build is CPU based.
|
|
|
|
#
|
|
|
|
# Then we build a CUDA dynamic library (although statically linked with the CUDA
|
|
|
|
# library dependencies for maximum portability)
|
|
|
|
#
|
|
|
|
# Then if we detect ROCm, we build a dynamically loaded ROCm lib. ROCm is particularly
|
|
|
|
# important to be a dynamic lib even if it's the only GPU library detected because
|
|
|
|
# we can't redistribute the objectfiles but must rely on dynamic libraries at
|
|
|
|
# runtime, which could lead the server not to start if not present.
|
2023-11-14 01:20:34 +00:00
|
|
|
|
|
|
|
set -ex
|
|
|
|
set -o pipefail
|
|
|
|
|
2024-01-03 23:12:29 +00:00
|
|
|
# See https://llvm.org/docs/AMDGPUUsage.html#processors for reference
|
|
|
|
amdGPUs() {
|
|
|
|
GPU_LIST=(
|
|
|
|
"gfx803"
|
|
|
|
"gfx900"
|
|
|
|
"gfx906:xnack-"
|
|
|
|
"gfx908:xnack-"
|
|
|
|
"gfx90a:xnack+"
|
|
|
|
"gfx90a:xnack-"
|
|
|
|
"gfx1010"
|
|
|
|
"gfx1012"
|
|
|
|
"gfx1030"
|
|
|
|
"gfx1100"
|
|
|
|
"gfx1101"
|
|
|
|
"gfx1102"
|
|
|
|
)
|
|
|
|
(
|
|
|
|
IFS=$';'
|
|
|
|
echo "'${GPU_LIST[*]}'"
|
|
|
|
)
|
|
|
|
}
|
|
|
|
|
2023-11-14 01:20:34 +00:00
|
|
|
echo "Starting linux generate script"
|
2024-01-09 20:55:36 +00:00
|
|
|
if [ -z "${CUDACXX}" ]; then
|
|
|
|
if [ -x /usr/local/cuda/bin/nvcc ]; then
|
|
|
|
export CUDACXX=/usr/local/cuda/bin/nvcc
|
|
|
|
else
|
|
|
|
# Try the default location in case it exists
|
|
|
|
export CUDACXX=$(command -v nvcc)
|
|
|
|
fi
|
2023-11-29 19:00:37 +00:00
|
|
|
fi
|
2024-01-04 00:22:15 +00:00
|
|
|
COMMON_CMAKE_DEFS="-DCMAKE_POSITION_INDEPENDENT_CODE=on -DLLAMA_NATIVE=off -DLLAMA_AVX=on -DLLAMA_AVX2=off -DLLAMA_AVX512=off -DLLAMA_FMA=off -DLLAMA_F16C=off"
|
2023-11-14 01:20:34 +00:00
|
|
|
source $(dirname $0)/gen_common.sh
|
|
|
|
init_vars
|
|
|
|
git_module_setup
|
|
|
|
apply_patches
|
2023-12-20 18:36:01 +00:00
|
|
|
|
2024-01-05 20:13:08 +00:00
|
|
|
if [ -z "${OLLAMA_SKIP_CPU_GENERATE}" ]; then
|
2024-01-07 23:48:05 +00:00
|
|
|
# Users building from source can tune the exact flags we pass to cmake for configuring
|
|
|
|
# llama.cpp, and we'll build only 1 CPU variant in that case as the default.
|
|
|
|
if [ -n "${OLLAMA_CUSTOM_CPU_DEFS}" ]; then
|
|
|
|
echo "OLLAMA_CUSTOM_CPU_DEFS=\"${OLLAMA_CUSTOM_CPU_DEFS}\""
|
|
|
|
CMAKE_DEFS="${OLLAMA_CUSTOM_CPU_DEFS} -DCMAKE_POSITION_INDEPENDENT_CODE=on ${CMAKE_DEFS}"
|
|
|
|
BUILD_DIR="${LLAMACPP_DIR}/build/linux/cpu"
|
|
|
|
echo "Building custom CPU"
|
|
|
|
build
|
|
|
|
install
|
|
|
|
link_server_lib
|
|
|
|
else
|
|
|
|
# Darwin Rosetta x86 emulation does NOT support AVX, AVX2, AVX512
|
|
|
|
# -DLLAMA_AVX -- 2011 Intel Sandy Bridge & AMD Bulldozer
|
|
|
|
# -DLLAMA_F16C -- 2012 Intel Ivy Bridge & AMD 2011 Bulldozer (No significant improvement over just AVX)
|
|
|
|
# -DLLAMA_AVX2 -- 2013 Intel Haswell & 2015 AMD Excavator / 2017 AMD Zen
|
|
|
|
# -DLLAMA_FMA (FMA3) -- 2013 Intel Haswell & 2012 AMD Piledriver
|
|
|
|
# Note: the following seem to yield slower results than AVX2 - ymmv
|
|
|
|
# -DLLAMA_AVX512 -- 2017 Intel Skylake and High End DeskTop (HEDT)
|
|
|
|
# -DLLAMA_AVX512_VBMI -- 2018 Intel Cannon Lake
|
|
|
|
# -DLLAMA_AVX512_VNNI -- 2021 Intel Alder Lake
|
2023-12-23 19:35:44 +00:00
|
|
|
|
2024-01-07 23:48:05 +00:00
|
|
|
COMMON_CPU_DEFS="-DCMAKE_POSITION_INDEPENDENT_CODE=on -DLLAMA_NATIVE=off"
|
|
|
|
#
|
|
|
|
# CPU first for the default library, set up as lowest common denominator for maximum compatibility (including Rosetta)
|
|
|
|
#
|
|
|
|
CMAKE_DEFS="${COMMON_CPU_DEFS} -DLLAMA_AVX=off -DLLAMA_AVX2=off -DLLAMA_AVX512=off -DLLAMA_FMA=off -DLLAMA_F16C=off ${CMAKE_DEFS}"
|
|
|
|
BUILD_DIR="${LLAMACPP_DIR}/build/linux/cpu"
|
|
|
|
echo "Building LCD CPU"
|
|
|
|
build
|
|
|
|
install
|
|
|
|
link_server_lib
|
2023-12-14 01:26:47 +00:00
|
|
|
|
2024-01-07 23:48:05 +00:00
|
|
|
#
|
|
|
|
# ~2011 CPU Dynamic library with more capabilities turned on to optimize performance
|
|
|
|
# Approximately 400% faster than LCD on same CPU
|
|
|
|
#
|
|
|
|
init_vars
|
|
|
|
CMAKE_DEFS="${COMMON_CPU_DEFS} -DLLAMA_AVX=on -DLLAMA_AVX2=off -DLLAMA_AVX512=off -DLLAMA_FMA=off -DLLAMA_F16C=off ${CMAKE_DEFS}"
|
|
|
|
BUILD_DIR="${LLAMACPP_DIR}/build/linux/cpu_avx"
|
|
|
|
echo "Building AVX CPU"
|
|
|
|
build
|
|
|
|
install
|
|
|
|
link_server_lib
|
|
|
|
|
|
|
|
#
|
|
|
|
# ~2013 CPU Dynamic library
|
|
|
|
# Approximately 10% faster than AVX on same CPU
|
|
|
|
#
|
|
|
|
init_vars
|
|
|
|
CMAKE_DEFS="${COMMON_CPU_DEFS} -DLLAMA_AVX=on -DLLAMA_AVX2=on -DLLAMA_AVX512=off -DLLAMA_FMA=on -DLLAMA_F16C=on ${CMAKE_DEFS}"
|
|
|
|
BUILD_DIR="${LLAMACPP_DIR}/build/linux/cpu_avx2"
|
|
|
|
echo "Building AVX2 CPU"
|
|
|
|
build
|
|
|
|
install
|
|
|
|
link_server_lib
|
|
|
|
fi
|
2024-01-05 20:13:08 +00:00
|
|
|
else
|
|
|
|
echo "Skipping CPU generation step as requested"
|
|
|
|
fi
|
2024-01-04 00:08:34 +00:00
|
|
|
|
2024-01-09 20:55:36 +00:00
|
|
|
if [ -z "${CUDA_LIB_DIR}" ]; then
|
|
|
|
# Try the default location in case it exists
|
|
|
|
CUDA_LIB_DIR=/usr/local/cuda/lib64
|
|
|
|
fi
|
|
|
|
|
|
|
|
if [ -d "${CUDA_LIB_DIR}" ]; then
|
2023-12-20 18:36:01 +00:00
|
|
|
echo "CUDA libraries detected - building dynamic CUDA library"
|
|
|
|
init_vars
|
2024-01-09 20:55:36 +00:00
|
|
|
CUDA_MAJOR=$(ls "${CUDA_LIB_DIR}"/libcudart.so.* | head -1 | cut -f3 -d. || true)
|
2024-01-05 20:13:08 +00:00
|
|
|
if [ -n "${CUDA_MAJOR}" ]; then
|
|
|
|
CUDA_VARIANT=_v${CUDA_MAJOR}
|
|
|
|
fi
|
2023-12-20 18:36:01 +00:00
|
|
|
CMAKE_DEFS="-DLLAMA_CUBLAS=on ${COMMON_CMAKE_DEFS} ${CMAKE_DEFS}"
|
2024-01-05 20:13:08 +00:00
|
|
|
BUILD_DIR="${LLAMACPP_DIR}/build/linux/cuda${CUDA_VARIANT}"
|
2023-12-20 18:36:01 +00:00
|
|
|
build
|
2023-12-23 19:35:44 +00:00
|
|
|
install
|
|
|
|
gcc -fPIC -g -shared -o ${BUILD_DIR}/lib/libext_server.so \
|
2023-12-20 18:36:01 +00:00
|
|
|
-Wl,--whole-archive \
|
2023-12-23 19:35:44 +00:00
|
|
|
${BUILD_DIR}/lib/libext_server.a \
|
|
|
|
${BUILD_DIR}/lib/libcommon.a \
|
|
|
|
${BUILD_DIR}/lib/libllama.a \
|
2023-12-20 18:36:01 +00:00
|
|
|
-Wl,--no-whole-archive \
|
|
|
|
${CUDA_LIB_DIR}/libcudart_static.a \
|
|
|
|
${CUDA_LIB_DIR}/libcublas_static.a \
|
|
|
|
${CUDA_LIB_DIR}/libcublasLt_static.a \
|
|
|
|
${CUDA_LIB_DIR}/libcudadevrt.a \
|
|
|
|
${CUDA_LIB_DIR}/libculibos.a \
|
|
|
|
-lrt -lpthread -ldl -lstdc++ -lm
|
2023-12-14 01:26:47 +00:00
|
|
|
fi
|
2023-11-29 19:00:37 +00:00
|
|
|
|
2023-12-20 18:36:01 +00:00
|
|
|
if [ -z "${ROCM_PATH}" ]; then
|
2023-11-29 19:00:37 +00:00
|
|
|
# Try the default location in case it exists
|
|
|
|
ROCM_PATH=/opt/rocm
|
|
|
|
fi
|
|
|
|
|
2023-12-20 18:36:01 +00:00
|
|
|
if [ -z "${CLBlast_DIR}" ]; then
|
2023-11-29 19:00:37 +00:00
|
|
|
# Try the default location in case it exists
|
|
|
|
if [ -d /usr/lib/cmake/CLBlast ]; then
|
|
|
|
export CLBlast_DIR=/usr/lib/cmake/CLBlast
|
|
|
|
fi
|
|
|
|
fi
|
|
|
|
|
2023-12-20 18:36:01 +00:00
|
|
|
if [ -d "${ROCM_PATH}" ]; then
|
|
|
|
echo "ROCm libraries detected - building dynamic ROCm library"
|
2024-01-05 20:13:08 +00:00
|
|
|
if [ -f ${ROCM_PATH}/lib/librocm_smi64.so.? ]; then
|
|
|
|
ROCM_VARIANT=_v$(ls ${ROCM_PATH}/lib/librocm_smi64.so.? | cut -f3 -d. || true)
|
|
|
|
fi
|
2023-11-29 19:00:37 +00:00
|
|
|
init_vars
|
2024-01-03 23:12:29 +00:00
|
|
|
CMAKE_DEFS="${COMMON_CMAKE_DEFS} ${CMAKE_DEFS} -DLLAMA_HIPBLAS=on -DCMAKE_C_COMPILER=$ROCM_PATH/llvm/bin/clang -DCMAKE_CXX_COMPILER=$ROCM_PATH/llvm/bin/clang++ -DAMDGPU_TARGETS=$(amdGPUs) -DGPU_TARGETS=$(amdGPUs)"
|
2024-01-05 20:13:08 +00:00
|
|
|
BUILD_DIR="${LLAMACPP_DIR}/build/linux/rocm${ROCM_VARIANT}"
|
2023-11-29 19:00:37 +00:00
|
|
|
build
|
2023-12-23 19:35:44 +00:00
|
|
|
install
|
|
|
|
gcc -fPIC -g -shared -o ${BUILD_DIR}/lib/libext_server.so \
|
2023-11-29 19:00:37 +00:00
|
|
|
-Wl,--whole-archive \
|
2023-12-23 19:35:44 +00:00
|
|
|
${BUILD_DIR}/lib/libext_server.a \
|
|
|
|
${BUILD_DIR}/lib/libcommon.a \
|
|
|
|
${BUILD_DIR}/lib/libllama.a \
|
2023-11-29 19:00:37 +00:00
|
|
|
-Wl,--no-whole-archive \
|
|
|
|
-lrt -lpthread -ldl -lstdc++ -lm \
|
|
|
|
-L/opt/rocm/lib -L/opt/amdgpu/lib/x86_64-linux-gnu/ \
|
|
|
|
-Wl,-rpath,/opt/rocm/lib,-rpath,/opt/amdgpu/lib/x86_64-linux-gnu/ \
|
|
|
|
-lhipblas -lrocblas -lamdhip64 -lrocsolver -lamd_comgr -lhsa-runtime64 -lrocsparse -ldrm -ldrm_amdgpu
|
|
|
|
fi
|
2023-12-22 17:51:53 +00:00
|
|
|
|
|
|
|
cleanup
|