# Copyright (c) 2019-2023, NVIDIA CORPORATION.  All rights reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.

cmake_minimum_required(VERSION 3.11 FATAL_ERROR) # for PyTorch extensions, version should be greater than 3.13
project(TurboMind LANGUAGES CXX CUDA)

find_package(CUDA 10.2 REQUIRED)

find_package(CUDAToolkit REQUIRED)

if(${CUDA_VERSION_MAJOR} VERSION_GREATER_EQUAL "11")
  add_definitions("-DENABLE_BF16")
  message("CUDA_VERSION ${CUDA_VERSION_MAJOR}.${CUDA_VERSION_MINOR} is greater or equal than 11.0, enable -DENABLE_BF16 flag")
endif()

set(CMAKE_MODULE_PATH ${PROJECT_SOURCE_DIR}/cmake/Modules)

option(BUILD_MULTI_GPU "Build multi-gpu support" ON)
option(BUILD_PY_FFI "Build python ffi" ON)
option(BUILD_TEST "Build tests" OFF)

include(FetchContent)
if (BUILD_TEST)
  FetchContent_Declare(
    repo-cutlass
    GIT_REPOSITORY https://github.com/NVIDIA/cutlass.git
    GIT_TAG        6f47420213f757831fae65c686aa471749fa8d60
    GIT_SHALLOW ON
  )

  set(CUTLASS_ENABLE_HEADERS_ONLY ON CACHE BOOL "Enable only the header library")

  FetchContent_MakeAvailable(repo-cutlass)

  set(CUTLASS_HEADER_DIR ${PROJECT_SOURCE_DIR}/3rdparty/cutlass/include)
  set(CUTLASS_EXTENSIONS_DIR ${PROJECT_SOURCE_DIR}/src/turbomind/cutlass_extensions/include)
endif()

FetchContent_Declare(
  yaml-cpp
  GIT_REPOSITORY https://github.com/jbeder/yaml-cpp.git
  GIT_TAG 0.8.0
)
set(YAML_BUILD_SHARED_LIBS OFF CACHE BOOL "Build static library of yaml-cpp")
FetchContent_MakeAvailable(yaml-cpp)


option(SPARSITY_SUPPORT "Build project with Ampere sparsity feature support" OFF)

option(BUILD_FAST_MATH "Build in fast math mode" ON)

# the environment variable
#   ASAN_OPTIONS=protect_shadow_gap=0,intercept_tls_get_addr=0
# must be set at runtime
# https://github.com/google/sanitizers/issues/1322
if (LMDEPLOY_ASAN_ENABLE)
    add_compile_options($<$<COMPILE_LANGUAGE:CXX>:-fsanitize=address>)
    add_link_options(-fsanitize=address)
endif ()

# notice that ubsan has linker issues for ubuntu < 18.04, see
# https://stackoverflow.com/questions/50024731/ld-unrecognized-option-push-state-no-as-needed
if (LMDEPLOY_UBSAN_ENABLE)
    add_compile_options($<$<COMPILE_LANGUAGE:CXX>:-fsanitize=undefined>)
    add_link_options(-fsanitize=undefined)
endif ()

if(BUILD_MULTI_GPU)
    add_definitions("-DBUILD_MULTI_GPU=1")
    set(CMAKE_MODULE_PATH ${PROJECT_SOURCE_DIR}/cmake/Modules)
    find_package(NCCL)
    if (NCCL_FOUND)
        set(USE_NCCL ON)
        add_definitions("-DUSE_NCCL=1")
    endif ()
endif()


set(CXX_STD "17" CACHE STRING "C++ standard")
# enable gold linker for binary and .so
set(CMAKE_EXE_LINKER_FLAGS "${CMAKE_EXE_LINKER_FLAGS} -fuse-ld=gold")
set(CMAKE_SHARED_LINKER_FLAGS "${CMAKE_SHARED_LINKER_FLAGS} -fuse-ld=gold")
set(CUDA_PATH ${CUDA_TOOLKIT_ROOT_DIR})

set(CUSPARSELT_PATH "" CACHE STRING "cuSPARSELt path")

list(APPEND CMAKE_MODULE_PATH ${CUDA_PATH}/lib64)

# profiling
option(USE_NVTX "Whether or not to use nvtx" ON)
if(USE_NVTX)
  message(STATUS "NVTX is enabled.")
  add_definitions("-DUSE_NVTX")
endif()

# setting compiler flags
set(CMAKE_C_FLAGS    "${CMAKE_C_FLAGS}")
set(CMAKE_CXX_STANDARD 17)
set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -Xcompiler -Wall -ldl") # -Xptxas -v

# TODO: build for sm_72 & sm_87 on aarch64 platform (Jetson devices)
if (NOT CMAKE_CUDA_ARCHITECTURES)
  set(CMAKE_CUDA_ARCHITECTURES 70-real 75-real)
  if (${CUDA_VERSION} VERSION_GREATER_EQUAL "11")
    list(APPEND CMAKE_CUDA_ARCHITECTURES 80-real)
  endif ()
  if (${CUDA_VERSION} VERSION_GREATER_EQUAL "11.1")
    list(APPEND CMAKE_CUDA_ARCHITECTURES 86-real)
  endif ()
  if (${CUDA_VERSION} VERSION_GREATER_EQUAL "11.8")
    list(APPEND CMAKE_CUDA_ARCHITECTURES 89-real 90-real)
  endif ()
  if (MSVC)
    list(REMOVE_ITEM CMAKE_CUDA_ARCHITECTURES 80-real 90-real)
  endif ()
endif ()

message(STATUS "Building with CUDA archs: ${CMAKE_CUDA_ARCHITECTURES}")

set(CMAKE_CUDA_RUNTIME_LIBRARY Shared)
set(CMAKE_C_FLAGS_DEBUG    "${CMAKE_C_FLAGS_DEBUG}    -Wall -O0")
set(CMAKE_CXX_FLAGS_DEBUG  "${CMAKE_CXX_FLAGS_DEBUG}  -Wall -O0")
# set(CMAKE_CUDA_FLAGS_DEBUG "${CMAKE_CUDA_FLAGS_DEBUG} -O0 -G -Xcompiler -Wall  --ptxas-options=-v --resource-usage")
set(CMAKE_CUDA_FLAGS_DEBUG "${CMAKE_CUDA_FLAGS_DEBUG} -O0 -G -Xcompiler -Wall -DCUDA_PTX_FP8_F2FP_ENABLED")

set(CMAKE_CXX_STANDARD "${CXX_STD}")
set(CMAKE_CXX_STANDARD_REQUIRED ON)
set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} --expt-extended-lambda")
set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} --expt-relaxed-constexpr")
set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} --std=c++${CXX_STD} -DCUDA_PTX_FP8_F2FP_ENABLED")

set(CMAKE_CXX_FLAGS_RELEASE        "${CMAKE_CXX_FLAGS_RELEASE}        -O3")
set(CMAKE_CXX_FLAGS_RELWITHDEBINFO "${CMAKE_CXX_FLAGS_RELWITHDEBINFO} -O3")
# set(CMAKE_CUDA_FLAGS_RELEASE "${CMAKE_CUDA_FLAGS_RELEASE} -Xcompiler -O3 --ptxas-options=--verbose")
set(CMAKE_CUDA_FLAGS_RELEASE        "${CMAKE_CUDA_FLAGS_RELEASE}        -Xcompiler -O3 -DCUDA_PTX_FP8_F2FP_ENABLED")
set(CMAKE_CUDA_FLAGS_RELWITHDEBINFO "${CMAKE_CUDA_FLAGS_RELWITHDEBINFO} -Xcompiler -O3 -DCUDA_PTX_FP8_F2FP_ENABLED")

if(BUILD_FAST_MATH)
    set(CMAKE_CUDA_FLAGS_RELEASE        "${CMAKE_CUDA_FLAGS_RELEASE}        --use_fast_math")
    set(CMAKE_CUDA_FLAGS_RELWITHDEBINFO "${CMAKE_CUDA_FLAGS_RELWITHDEBINFO} --use_fast_math")
    message("Release build CUDA flags: ${CMAKE_CUDA_FLAGS_RELEASE}")
endif()

set(CMAKE_ARCHIVE_OUTPUT_DIRECTORY ${CMAKE_BINARY_DIR}/lib)
set(CMAKE_LIBRARY_OUTPUT_DIRECTORY ${CMAKE_BINARY_DIR}/lib)
set(CMAKE_RUNTIME_OUTPUT_DIRECTORY ${CMAKE_BINARY_DIR}/bin)

set(COMMON_HEADER_DIRS
  ${PROJECT_SOURCE_DIR}
  ${CUDA_PATH}/include
  ${CUTLASS_HEADER_DIR}
)
message("-- COMMON_HEADER_DIRS: ${COMMON_HEADER_DIRS}")

set(COMMON_LIB_DIRS
  ${CUDA_PATH}/lib64
)

if (SPARSITY_SUPPORT)
  list(APPEND COMMON_HEADER_DIRS ${CUSPARSELT_PATH}/include)
  list(APPEND COMMON_LIB_DIRS ${CUSPARSELT_PATH}/lib64)
  add_definitions(-DSPARSITY_ENABLED=1)
endif()


set(PYTHON_PATH "python" CACHE STRING "Python path")

# turn off warnings on windows
if (MSVC)
  foreach(
    flag_var
    CMAKE_CXX_FLAGS
    CMAKE_CXX_FLAGS_DEBUG
    CMAKE_CXX_FLAGS_RELEASE
    CMAKE_CXX_FLAGS_MINSIZEREL
    CMAKE_CXX_FLAGS_RELWITHDEBINFO
    CMAKE_C_FLAGS
    CMAKE_C_FLAGS_DEBUG
    CMAKE_C_FLAGS_RELEASE
    CMAKE_C_FLAGS_MINSIZEREL
    CMAKE_C_FLAGS_RELWITHDEBINFO
    CMAKE_CUDA_FLAGS
    CMAKE_CUDA_FLAGS_DEBUG
    CMAKE_CUDA_FLAGS_RELEASE
    CMAKE_CUDA_FLAGS_MINSIZEREL
    CMAKE_CUDA_FLAGS_RELWITHDEBINFO)
    string(REGEX REPLACE "-Wall" " /W0 " ${flag_var} "${${flag_var}}")
  endforeach()
endif()

include_directories(
  ${COMMON_HEADER_DIRS}
)

link_directories(
  ${COMMON_LIB_DIRS}
)

# add_subdirectory(3rdparty)
add_subdirectory(src)
# add_subdirectory(examples)

if(BUILD_TEST)
    add_subdirectory(tests/csrc)
endif()

# install python api
if (BUILD_PY_FFI)
  install(TARGETS _turbomind DESTINATION ${CMAKE_SOURCE_DIR}/lmdeploy/lib)
endif ()