Files
ik_llama.cpp/CMakeLists.txt
Kawrakow 519405dc97 Async compute graph evaluation (2 or more GPUs) (#1089)
* WIP: absorb adding input into std_attn and std_ffn

* WIP: NCCL infra

* WIP: add reduce and fake_cpy ops

* WIP

* WIP: graph appears to work, layer is broken

* WIP: Qwen3-MoE works with graph, layer still broken

* WIP: GLM-4.5 graph works

* WIP: fix sm layer (dense)

* WIP: fix sm layer (MoE)

* WIP: fast PP with bespoke 4-GPU NCCL

I guess, I'm not using NCCL the right way as PP is very
low with a single communicator group for 3 or more GPUs.
But if I create 4 communicator groups for pairs of GPUs
(0,1, 2,3, 0,2, 1,3) and use that, PP is fast: I'm hitting
1500 t/s for L3-70B on the 4x3090 system, which is
~20% better than the previous sm graph without NCCL.
But that cannot be the solution (I cannot be creating pairwise
communicators and associated logic for every possible number of GPUs).

* WIP: Cohere2

* Explicitely set device

* Bespoke 3-GPU case

* WIP

* Do not repeat get_rows multiple times

* Fix 3 GPUs

* OK, let's leave it in

* Simple async

* This sync seems enough

* Only do async for 4 or more backends

With 2 GPUs (so, 3 backends) not using async is slightly faster

* Scheduler changes

* Use OpenMP if available

Surprisingly (at least to me), this is quite a bit faster than
std::thread and std::barrier. GLM-4.5-AIR with 4 GPUs is now
at 105 t/s at zero context!

* Do not use OpenMP if there are tensor overrides

* Set omp max active levels

* Be more careful with having set the device before using a stream

* Command line option to turn on async. Set to false by defualt for now

---------

Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2025-12-27 08:18:06 +01:00

245 lines
7.5 KiB
CMake

cmake_minimum_required(VERSION 3.14) # for add_link_options and implicit target directories.
project("llama.cpp" C CXX)
include(CheckIncludeFileCXX)
#set(CMAKE_WARN_DEPRECATED YES)
set(CMAKE_WARN_UNUSED_CLI YES)
set(CMAKE_EXPORT_COMPILE_COMMANDS ON)
set(CMAKE_CXX_STANDARD 20)
set(CMAKE_CXX_STANDARD_REQUIRED true)
set(CMAKE_CUDA_USE_RESPONSE_FILE_FOR_INCLUDES 0)
set(CMAKE_CUDA_USE_RESPONSE_FILE_FOR_LIBRARIES 0)
set(CMAKE_CUDA_USE_RESPONSE_FILE_FOR_OBJECTS 0)
if (NOT XCODE AND NOT MSVC AND NOT CMAKE_BUILD_TYPE)
set(CMAKE_BUILD_TYPE Release CACHE STRING "Build type" FORCE)
set_property(CACHE CMAKE_BUILD_TYPE PROPERTY STRINGS "Debug" "Release" "MinSizeRel" "RelWithDebInfo")
endif()
# Add path to modules
list(APPEND CMAKE_MODULE_PATH "${CMAKE_CURRENT_SOURCE_DIR}/cmake/")
set(CMAKE_RUNTIME_OUTPUT_DIRECTORY ${CMAKE_BINARY_DIR}/bin)
if (CMAKE_SOURCE_DIR STREQUAL CMAKE_CURRENT_SOURCE_DIR)
set(LLAMA_STANDALONE ON)
include(git-vars)
# configure project version
# TODO
else()
set(LLAMA_STANDALONE OFF)
endif()
if (EMSCRIPTEN)
set(BUILD_SHARED_LIBS_DEFAULT OFF)
option(LLAMA_WASM_SINGLE_FILE "llama: embed WASM inside the generated llama.js" ON)
else()
if (MINGW)
set(BUILD_SHARED_LIBS_DEFAULT OFF)
else()
set(BUILD_SHARED_LIBS_DEFAULT ON)
endif()
endif()
option(BUILD_SHARED_LIBS "build shared libraries" ${BUILD_SHARED_LIBS_DEFAULT})
if (WIN32)
add_compile_definitions(_CRT_SECURE_NO_WARNINGS)
endif()
# force MSVC compiler charset to utf-8
if (MSVC)
add_compile_options("$<$<COMPILE_LANGUAGE:C>:/utf-8>")
add_compile_options("$<$<COMPILE_LANGUAGE:CXX>:/utf-8>")
add_compile_options("$<$<COMPILE_LANGUAGE:C>:/bigobj>")
add_compile_options("$<$<COMPILE_LANGUAGE:CXX>:/bigobj>")
endif()
#
# option list
#
# debug
option(LLAMA_ALL_WARNINGS "llama: enable all compiler warnings" ON)
option(LLAMA_ALL_WARNINGS_3RD_PARTY "llama: enable all compiler warnings in 3rd party libs" OFF)
# build
option(LLAMA_FATAL_WARNINGS "llama: enable -Werror flag" OFF)
# sanitizers
option(LLAMA_SANITIZE_THREAD "llama: enable thread sanitizer" OFF)
option(LLAMA_SANITIZE_ADDRESS "llama: enable address sanitizer" OFF)
option(LLAMA_SANITIZE_UNDEFINED "llama: enable undefined sanitizer" OFF)
# extra artifacts
option(LLAMA_BUILD_TESTS "llama: build tests" ${LLAMA_STANDALONE})
option(LLAMA_BUILD_EXAMPLES "llama: build examples" ${LLAMA_STANDALONE})
option(LLAMA_BUILD_SERVER "llama: build server example" ${LLAMA_STANDALONE})
# 3rd party libs
option(LLAMA_CURL "llama: use libcurl to download model from an URL" OFF)
option(LLAMA_LLGUIDANCE "llama-common: include LLGuidance library for structured output in common utils" OFF)
# Required for relocatable CMake package
include(${CMAKE_CURRENT_SOURCE_DIR}/cmake/build-info.cmake)
# override ggml options
set(GGML_SANITIZE_THREAD ${LLAMA_SANITIZE_THREAD})
set(GGML_SANITIZE_ADDRESS ${LLAMA_SANITIZE_ADDRESS})
set(GGML_SANITIZE_UNDEFINED ${LLAMA_SANITIZE_UNDEFINED})
set(GGML_ALL_WARNINGS ${LLAMA_ALL_WARNINGS})
set(GGML_FATAL_WARNINGS ${LLAMA_FATAL_WARNINGS})
# change the default for these ggml options
if (NOT DEFINED GGML_LLAMAFILE)
set(GGML_LLAMAFILE ON)
endif()
if (NOT DEFINED GGML_CUDA_USE_GRAPHS)
set(GGML_CUDA_USE_GRAPHS ON)
endif()
# transition helpers
function (llama_option_depr TYPE OLD NEW)
if (${OLD})
message(${TYPE} "${OLD} is deprecated and will be removed in the future.\nUse ${NEW} instead\n")
set(${NEW} ON PARENT_SCOPE)
endif()
endfunction()
llama_option_depr(FATAL_ERROR LLAMA_CUBLAS GGML_CUDA)
llama_option_depr(WARNING LLAMA_CUDA GGML_CUDA)
llama_option_depr(WARNING LLAMA_KOMPUTE GGML_KOMPUTE)
llama_option_depr(WARNING LLAMA_METAL GGML_METAL)
llama_option_depr(WARNING LLAMA_METAL_EMBED_LIBRARY GGML_METAL_EMBED_LIBRARY)
llama_option_depr(WARNING LLAMA_NATIVE GGML_NATIVE)
llama_option_depr(WARNING LLAMA_RPC GGML_RPC)
llama_option_depr(WARNING LLAMA_SYCL GGML_SYCL)
llama_option_depr(WARNING LLAMA_SYCL_F16 GGML_SYCL_F16)
llama_option_depr(WARNING LLAMA_CANN GGML_CANN)
if (NOT MSVC)
if (LLAMA_SANITIZE_THREAD)
message(STATUS "Using -fsanitize=thread")
add_compile_options(-fsanitize=thread)
link_libraries (-fsanitize=thread)
endif()
if (LLAMA_SANITIZE_ADDRESS)
message(STATUS "Using -fsanitize=address")
add_compile_options(-fsanitize=address -fno-omit-frame-pointer)
link_libraries (-fsanitize=address)
endif()
if (LLAMA_SANITIZE_UNDEFINED)
message(STATUS "Using -fsanitize=undefined")
add_compile_options(-fsanitize=undefined)
link_libraries (-fsanitize=undefined)
endif()
endif()
#
# build the library
#
if (NOT TARGET ggml)
add_subdirectory(ggml)
# ... otherwise assume ggml is added by a parent CMakeLists.txt
endif()
#
# build the library
#
add_subdirectory(src)
#
# install
#
include(GNUInstallDirs)
include(CMakePackageConfigHelpers)
set(LLAMA_BUILD_NUMBER ${BUILD_NUMBER})
set(LLAMA_BUILD_COMMIT ${BUILD_COMMIT})
set(LLAMA_INSTALL_VERSION 0.0.${BUILD_NUMBER})
set(LLAMA_INCLUDE_INSTALL_DIR ${CMAKE_INSTALL_INCLUDEDIR} CACHE PATH "Location of header files")
set(LLAMA_LIB_INSTALL_DIR ${CMAKE_INSTALL_LIBDIR} CACHE PATH "Location of library files")
set(LLAMA_BIN_INSTALL_DIR ${CMAKE_INSTALL_BINDIR} CACHE PATH "Location of binary files")
# At the moment some compile definitions are placed within the ggml/src
# directory but not exported on the `ggml` target. This could be improved by
# determining _precisely_ which defines are necessary for the llama-config
# package.
#
get_target_property(GGML_DIRECTORY ggml SOURCE_DIR)
get_directory_property(GGML_DIR_DEFINES DIRECTORY ${GGML_DIRECTORY} COMPILE_DEFINITIONS)
get_target_property(GGML_TARGET_DEFINES ggml COMPILE_DEFINITIONS)
set(GGML_TRANSIENT_DEFINES ${GGML_TARGET_DEFINES} ${GGML_DIR_DEFINES})
get_target_property(GGML_LINK_LIBRARIES ggml LINK_LIBRARIES)
set_target_properties(llama PROPERTIES PUBLIC_HEADER ${CMAKE_CURRENT_SOURCE_DIR}/include/llama.h)
install(TARGETS llama LIBRARY PUBLIC_HEADER)
configure_package_config_file(
${CMAKE_CURRENT_SOURCE_DIR}/cmake/llama-config.cmake.in
${CMAKE_CURRENT_BINARY_DIR}/llama-config.cmake
INSTALL_DESTINATION ${CMAKE_INSTALL_LIBDIR}/cmake/llama
PATH_VARS LLAMA_INCLUDE_INSTALL_DIR
LLAMA_LIB_INSTALL_DIR
LLAMA_BIN_INSTALL_DIR )
write_basic_package_version_file(
${CMAKE_CURRENT_BINARY_DIR}/llama-version.cmake
VERSION ${LLAMA_INSTALL_VERSION}
COMPATIBILITY SameMajorVersion)
install(FILES ${CMAKE_CURRENT_BINARY_DIR}/llama-config.cmake
${CMAKE_CURRENT_BINARY_DIR}/llama-version.cmake
DESTINATION ${CMAKE_INSTALL_LIBDIR}/cmake/llama)
install(
FILES convert_hf_to_gguf.py
PERMISSIONS
OWNER_READ
OWNER_WRITE
OWNER_EXECUTE
GROUP_READ
GROUP_EXECUTE
WORLD_READ
WORLD_EXECUTE
DESTINATION ${CMAKE_INSTALL_BINDIR})
configure_file(cmake/llama.pc.in
"${CMAKE_CURRENT_BINARY_DIR}/llama.pc"
@ONLY)
install(FILES "${CMAKE_CURRENT_BINARY_DIR}/llama.pc"
DESTINATION lib/pkgconfig)
#
# programs, examples and tests
#
add_subdirectory(common)
if (LLAMA_BUILD_TESTS AND NOT CMAKE_JS_VERSION)
include(CTest)
add_subdirectory(tests)
endif ()
if (LLAMA_BUILD_EXAMPLES)
add_subdirectory(examples)
add_subdirectory(pocs)
endif()