happyz
/
llama.cpp
mirror of https://github.com/ggerganov/llama.cpp.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

[WIP] llama: ABI split builds five .so artifacts. 

However, all .so are performing on SVE level
This commit is contained in:
Han Yin 2025-05-12 15:44:22 -07:00
parent eab502a735
commit 75d1abe24a
2 changed files with 60 additions and 69 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

117
examples/llama.android/llama/src/main/cpp/CMakeLists.txt
View File

@ -1,92 +1,83 @@
# Sets the minimum CMake version required for this project.
cmake_minimum_required(VERSION 3.22.1) cmake_minimum_required(VERSION 3.22.1)
project("llama_android" VERSION 1.0.0 LANGUAGES C CXX) project("llama_android" VERSION 1.0.0 LANGUAGES C CXX)
# --------------------------------------------------------------------
# 0. Global language & toolchain options
# --------------------------------------------------------------------
set(CMAKE_C_STANDARD 11) set(CMAKE_C_STANDARD 11)
set(CMAKE_C_STANDARD_REQUIRED true) set(CMAKE_C_STANDARD_REQUIRED true)
set(CMAKE_CXX_STANDARD 17) set(CMAKE_CXX_STANDARD 17)
set(CMAKE_CXX_STANDARD_REQUIRED true) set(CMAKE_CXX_STANDARD_REQUIRED true)
set(GGML_CPU_KLEIDIAI ON CACHE BOOL "" FORCE) ## ---- ggml / llama feature switches (configure-time) ----------------
#set(GGML_CPU_KLEIDIAI ON CACHE BOOL "" FORCE)
set(GGML_LLAMAFILE OFF CACHE BOOL "" FORCE) set(GGML_LLAMAFILE OFF CACHE BOOL "" FORCE)
set(LLAMA_BUILD_COMMON ON CACHE BOOL "" FORCE) set(LLAMA_BUILD_COMMON ON CACHE BOOL "" FORCE)
set(LLAMA_CURL OFF CACHE BOOL "" FORCE) set(LLAMA_CURL OFF CACHE BOOL "" FORCE)
string(APPEND CMAKE_C_FLAGS " -march=armv8.7-a+dotprod+i8mm+sve") # baseline ISA used for every source that is NOT in a tier wrapper
string(APPEND CMAKE_CXX_FLAGS " -march=armv8.7-a+dotprod+i8mm+sve") string(APPEND CMAKE_C_FLAGS " -march=armv9.2-a+dotprod+i8mm+sve+sve2+sme")
string(APPEND CMAKE_CXX_FLAGS " -march=armv9.2-a+dotprod+i8mm+sve+sve2+sme")
set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS}" CACHE STRING "" FORCE) set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS}" CACHE STRING "" FORCE)
set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS}" CACHE STRING "" FORCE) set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS}" CACHE STRING "" FORCE)
## CMake definitions # NOT NEEDED WHEN CMAKE_BUILD_TYPE PASSED IN BY GRADLE
#add_compile_definitions( #message(DEBUG "CMAKE_BUILD_TYPE: ${CMAKE_BUILD_TYPE}")
# GGML_CPU_KLEIDIAI=ON #if(CMAKE_BUILD_TYPE STREQUAL "Debug")
# GGML_LLAMAFILE=OFF # string(APPEND CMAKE_C_FLAGS " -O3 -DNDEBUG")
# LLAMA_BUILD_COMMON=ON # string(APPEND CMAKE_CXX_FLAGS " -O3 -DNDEBUG")
# LLAMA_CURL=OFF #endif()
#)
# Force Release only when Gradle didn't specify one # default to Release if Gradle hasn’t set a variant
if(NOT DEFINED CMAKE_BUILD_TYPE) if(NOT DEFINED CMAKE_BUILD_TYPE)
set(CMAKE_BUILD_TYPE Release CACHE STRING "" FORCE) set(CMAKE_BUILD_TYPE Release CACHE STRING "" FORCE)
endif() endif()
## Also provides "common" ### Also provides "common"
#include(FetchContent) ##include(FetchContent)
#FetchContent_Declare( ##FetchContent_Declare(
# llama ## llama
# GIT_REPOSITORY https://github.com/ggml-org/llama.cpp ## GIT_REPOSITORY https://github.com/ggml-org/llama.cpp
# GIT_TAG master ## GIT_TAG master
#) ##)
#FetchContent_MakeAvailable(llama) ##FetchContent_MakeAvailable(llama)
# Add local llama.cpp as subdirectory # --------------------------------------------------------------------
add_subdirectory(../../../../../../ build-llama) # 1. Pull in llama.cpp (creates targets "llama" and "common")
# --------------------------------------------------------------------
set(LLAMA_SRC ${CMAKE_CURRENT_LIST_DIR}/../../../../../../)
add_subdirectory(${LLAMA_SRC} build-llama)
#function(build_llama_tier tier march_flag) # -----------------------------------------------------------
# set(LIB_NAME "${CMAKE_PROJECT_NAME}_${tier}") # 2. Helper to create one wrapper .so for a hardware tier
# # -----------------------------------------------------------
## set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -march=armv8.7-a+i8mm") function(build_llama_tier tier march)
## set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -march=armv8.7-a+i8mm") set(wrapper llama_android_${tier})
# add_library(${wrapper} SHARED llama-android.cpp)
# add_library(${LIB_NAME} SHARED llama-android.cpp)
#
# MIGHT NOT WORK???? # ISA tuning that belongs only to this tier
# target_include_directories(${LIB_NAME} PRIVATE target_compile_options(${wrapper} PRIVATE "-march=${march}")
# ${CMAKE_CURRENT_SOURCE_DIR}/../../../../../../common)
# ?????
# target_compile_options(${LIB_NAME} PRIVATE "-march=${march_flag}") # NOT NEEDED!
# # # sampling.h lives in common/, ggml-impl.h lives in ggml/src
# target_link_libraries(${LIB_NAME} # target_include_directories(${wrapper} PRIVATE
# llama # ${LLAMA_SRC}/common
# common # ${LLAMA_SRC}/ggml/src)
# android
# log
# )
#
# set_target_properties(${LIB_NAME} PROPERTIES
# OUTPUT_NAME "llama_android_${tier}"
# )
#endfunction()
#
#build_llama_tier(t0 "armv8-a+simd")
#build_llama_tier(t1 "armv8.2-a+dotprod")
#build_llama_tier(t2 "armv8.6-a+dotprod+i8mm")
#build_llama_tier(t3 "armv9-a+dotprod+i8mm+sve+sve2")
#build_llama_tier(t4 "armv9.2-a+dotprod+i8mm+sve+sve2+sme")
target_link_libraries(${wrapper} PRIVATE
add_library(${CMAKE_PROJECT_NAME} SHARED
# List C/C++ source files with relative paths to this CMakeLists.txt.
llama-android.cpp)
target_link_libraries(${CMAKE_PROJECT_NAME}
# List libraries link to the target library
llama llama
common common
android android
log) log)
set_target_properties(${wrapper} PROPERTIES OUTPUT_NAME ${wrapper})
endfunction()
# -----------------------------------------------------------
# 3. Build the five tiers (no KleidiAI for now)
# -----------------------------------------------------------
build_llama_tier(t0 "armv8-a+simd")
build_llama_tier(t1 "armv8.2-a+dotprod")
build_llama_tier(t2 "armv8.6-a+dotprod+i8mm")
build_llama_tier(t3 "armv9-a+dotprod+i8mm+sve+sve2")
build_llama_tier(t4 "armv9.2-a+dotprod+i8mm+sve+sve2+sme")

2
examples/llama.android/llama/src/main/java/android/llama/cpp/LLamaAndroid.kt
View File

@ -240,7 +240,7 @@ class LLamaAndroid private constructor() : InferenceEngine {
private val TAG = LLamaAndroid::class.simpleName private val TAG = LLamaAndroid::class.simpleName
// TODO-han.yin: replace with dynamic loader // TODO-han.yin: replace with dynamic loader
private const val LIB_LLAMA_ANDROID = "llama_android" private const val LIB_LLAMA_ANDROID = "llama_android_t4"
// Enforce only one instance of Llm. // Enforce only one instance of Llm.
private val _instance: LLamaAndroid = LLamaAndroid() private val _instance: LLamaAndroid = LLamaAndroid()