Add multiple CPU variants for Intel Mac

This also refines the build process for the ext_server build.

llm/generate/gen_darwin.sh

@@ -9,16 +9,52 @@ set -o pipefail
 echo "Starting darwin generate script"
 source $(dirname $0)/gen_common.sh
 init_vars
-CMAKE_DEFS="-DCMAKE_OSX_DEPLOYMENT_TARGET=11.0 -DCMAKE_SYSTEM_NAME=Darwin -DLLAMA_ACCELERATE=on ${CMAKE_DEFS}"
-BUILD_DIR="${LLAMACPP_DIR}/build/darwin/metal"
+git_module_setup
+apply_patches
+
+COMMON_DARWIN_DEFS="-DCMAKE_OSX_DEPLOYMENT_TARGET=11.0 -DCMAKE_SYSTEM_NAME=Darwin -DLLAMA_ACCELERATE=off"
+
 case "${GOARCH}" in
 "amd64")
-    CMAKE_DEFS="-DCMAKE_SYSTEM_PROCESSOR=x86_64 -DCMAKE_OSX_ARCHITECTURES=x86_64 -DLLAMA_METAL=off -DLLAMA_NATIVE=off -DLLAMA_AVX=on -DLLAMA_AVX2=off -DLLAMA_AVX512=off -DLLAMA_FMA=off -DLLAMA_F16C=off ${CMAKE_DEFS}"
-    ARCH="x86_64"
+    COMMON_CPU_DEFS="${COMMON_DARWIN_DEFS} -DCMAKE_SYSTEM_PROCESSOR=${ARCH} -DCMAKE_OSX_ARCHITECTURES=${ARCH} -DLLAMA_METAL=off -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/darwin/${ARCH}/cpu"
+    echo "Building LCD CPU"
+    build
+    compress_libs
+
+    #
+    # ~2011 CPU Dynamic library with more capabilities turned on to optimize performance
+    # Approximately 400% faster than LCD on same CPU
+    #
+    init_vars
+    CMAKE_DEFS="${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/darwin/${ARCH}/cpu_avx"
+    echo "Building AVX CPU"
+    build
+    compress_libs
+
+    #
+    # ~2013 CPU Dynamic library
+    # Approximately 10% faster than AVX on same CPU
+    #
+    init_vars
+    CMAKE_DEFS="${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/darwin/${ARCH}/cpu_avx2"
+    echo "Building AVX2 CPU"
+    build
+    compress_libs
     ;;
 "arm64")
-    CMAKE_DEFS="-DCMAKE_SYSTEM_PROCESSOR=arm64 -DCMAKE_OSX_ARCHITECTURES=arm64 -DLLAMA_METAL=on ${CMAKE_DEFS}"
-    ARCH="arm64"
+    CMAKE_DEFS="${COMMON_DARWIN_DEFS} -DCMAKE_SYSTEM_PROCESSOR=${ARCH} -DCMAKE_OSX_ARCHITECTURES=${ARCH} -DLLAMA_METAL=on -DLLAMA_ACCELERATE=on ${CMAKE_DEFS}"
+    BUILD_DIR="${LLAMACPP_DIR}/build/darwin/${ARCH}/metal"
+    EXTRA_LIBS="${EXTRA_LIBS} -framework Accelerate -framework Foundation -framework Metal -framework MetalKit -framework MetalPerformanceShaders"
+    build
+    compress_libs
     ;;
 *)
     echo "GOARCH must be set"
@@ -27,21 +63,4 @@ case "${GOARCH}" in
     ;;
 esac
 
-git_module_setup
-apply_patches
-build
-install
-gcc -fPIC -g -shared -o ${BUILD_DIR}/lib/libext_server.so \
-    -arch ${ARCH} \
-    -Wl,-force_load ${BUILD_DIR}/lib/libext_server.a \
-    ${BUILD_DIR}/lib/libcommon.a \
-    ${BUILD_DIR}/lib/libllama.a \
-    ${BUILD_DIR}/lib/libggml_static.a \
-    -lpthread -ldl -lm -lc++ \
-    -framework Accelerate \
-    -framework Foundation \
-    -framework Metal \
-    -framework MetalKit \
-    -framework MetalPerformanceShaders
-
 cleanup