@wllama/wllama/CMakeLists.txt

WebAssembly binding for llama.cpp - Enabling on-browser LLM inference

cmake_minimum_required(VERSION 3.14)
project("wllama")

option(WLLAMA_TEST_BACKEND "Build wllama with test-backend-ops included" OFF)

set(MTMD_VIDEO OFF CACHE BOOL "" FORCE)

set(CMAKE_THREAD_LIBS_INIT "-lpthread")
set(CMAKE_HAVE_THREADS_LIBRARY 1)
set(CMAKE_USE_WIN32_THREADS_INIT 0)
set(CMAKE_USE_PTHREADS_INIT 1)
set(THREADS_PREFER_PTHREAD_FLAG ON)

set(WLLAMA_COMPILE_OPTIONS
    -O3 -msimd128 -DNDEBUG
    # -flto=full # note: this breaks things, better to disable
    -frtti
    -pthread
    -gsource-map
)
set(WLLAMA_LINK_OPTIONS
    # -flto=full # note: this breaks things, better to disable
    --no-entry
    -sEXPORT_ES6=0
    -sMODULARIZE=0
    -sINITIAL_MEMORY=128MB
    -sMAXIMUM_MEMORY=4096MB
    -sSTACK_SIZE=5MB
    -sALLOW_MEMORY_GROWTH=1
    -sFORCE_FILESYSTEM=1
    -sEXPORTED_FUNCTIONS=_main,_wllama_malloc,_wllama_start,_wllama_action,_wllama_exit,_wllama_debug
    -sEXPORTED_RUNTIME_METHODS=ccall,cwrap,HEAPU8,MEMFS,FS,mmapAlloc,ENV,wasmMemory
    -sNO_EXIT_RUNTIME=1
    -sIMPORTED_MEMORY=1
    -sPTHREAD_POOL_SIZE=Module[\"pthreadPoolSize\"]
    -sUSE_PTHREADS=1
    -pthread
    -gsource-map
    --emit-symbol-map
    -Wl,--wrap,fopen
    -Wl,--wrap,fclose
    -Wl,--wrap,fread
    -Wl,--wrap,fseek
    -Wl,--wrap,ftell
    -Wl,--wrap,abort
    -Wl,--wrap,ggml_graph_plan # for test-backend-ops
)

if (WLLAMA_COMPAT)
    # no wasm exception (not compatible with asyncify - asyncify is needed for firefox and safari)
    # no mem64 (not compatible with safari)
    list(APPEND WLLAMA_COMPILE_OPTIONS
        -fexceptions
        -pthread
    )
    list(APPEND WLLAMA_LINK_OPTIONS
        -fexceptions
        -sASYNCIFY=1
        -sASYNCIFY_ADD=['wllama_start','wllama_action']
    )
else()
    list(APPEND WLLAMA_COMPILE_OPTIONS
        -sMEMORY64=1
        -fwasm-exceptions
    )
    list(APPEND WLLAMA_LINK_OPTIONS
        -sMEMORY64=1
        -fwasm-exceptions
        -sJSPI
        -sJSPI_EXPORTS=['wllama_start','wllama_action']
    )
endif()

add_compile_options(${WLLAMA_COMPILE_OPTIONS})
add_link_options(${WLLAMA_LINK_OPTIONS})

add_subdirectory(llama.cpp)

set(LLAMA_INSTALL_VERSION 0.0.${LLAMA_BUILD_NUMBER})
add_subdirectory(llama.cpp/tools/mtmd)

file(GLOB_RECURSE LLAMA_COMMON_SRC
    CONFIGURE_DEPENDS
    llama.cpp/common/*.cpp
    llama.cpp/common/*.h
    llama.cpp/common/*.hpp
    llama.cpp/common/jinja/*.cpp
    llama.cpp/common/jinja/*.h
)

list(REMOVE_ITEM LLAMA_COMMON_SRC
    ${CMAKE_CURRENT_SOURCE_DIR}/llama.cpp/common/arg.cpp
    ${CMAKE_CURRENT_SOURCE_DIR}/llama.cpp/common/console.cpp
    ${CMAKE_CURRENT_SOURCE_DIR}/llama.cpp/common/debug.cpp
    ${CMAKE_CURRENT_SOURCE_DIR}/llama.cpp/common/fit.cpp
    ${CMAKE_CURRENT_SOURCE_DIR}/llama.cpp/common/hf-cache.cpp
    ${CMAKE_CURRENT_SOURCE_DIR}/llama.cpp/common/log.cpp
    ${CMAKE_CURRENT_SOURCE_DIR}/llama.cpp/common/preset.cpp
    ${CMAKE_CURRENT_SOURCE_DIR}/llama.cpp/common/download.cpp
)

set(LLAMA_SERVER_SRC llama.cpp/tools/server/server-context.cpp
    llama.cpp/tools/server/server-task.cpp
    llama.cpp/tools/server/server-chat.cpp
    llama.cpp/tools/server/server-common.cpp
    llama.cpp/tools/server/server-schema.cpp
    llama.cpp/tools/server/server-stream.cpp
)

set(WLLAMA_SRC ${LLAMA_COMMON_SRC} ${LLAMA_SERVER_SRC}
    cpp/wllama.cpp
    cpp/glue.hpp
    llama.cpp/include/llama.h)

if(WLLAMA_TEST_BACKEND)
    message(WARNING "Building wllama with test-backend-ops included")
    set(TEST_BACKEND_OPS_SRC ${CMAKE_CURRENT_BINARY_DIR}/test-backend-ops.cpp)
    file(READ ${CMAKE_CURRENT_SOURCE_DIR}/llama.cpp/tests/test-backend-ops.cpp TEST_BACKEND_OPS_CONTENT)
    string(REPLACE "int main(" "int main_test_backend_ops(" TEST_BACKEND_OPS_CONTENT "${TEST_BACKEND_OPS_CONTENT}")
    file(WRITE ${TEST_BACKEND_OPS_SRC} "${TEST_BACKEND_OPS_CONTENT}")
    list(APPEND WLLAMA_SRC ${TEST_BACKEND_OPS_SRC})
    add_compile_definitions(WLLAMA_TEST_BACKEND)
endif()

include_directories(${CMAKE_CURRENT_SOURCE_DIR}/cpp)
include_directories(${CMAKE_CURRENT_SOURCE_DIR}/llama.cpp/include)
include_directories(${CMAKE_CURRENT_SOURCE_DIR}/llama.cpp/common)
include_directories(${CMAKE_CURRENT_SOURCE_DIR}/llama.cpp/vendor)
include_directories(${CMAKE_CURRENT_SOURCE_DIR}/llama.cpp/tools/server)
include_directories(${CMAKE_CURRENT_SOURCE_DIR}/llama.cpp/tools/mtmd)

add_executable(wllama ${WLLAMA_SRC})
target_link_libraries(wllama PRIVATE ggml llama mtmd ${CMAKE_THREAD_LIBS_INIT})