CMakeLists.txt

if("x${CMAKE_SOURCE_DIR}" STREQUAL "x${CMAKE_BINARY_DIR}")
  message(FATAL_ERROR "\
In-source build is not a good practice.
Please use:
  mkdir build
  cd build
  cmake ..
to build this project"
  )
endif()

cmake_minimum_required(VERSION 3.8 FATAL_ERROR)

message(STATUS "CMAKE_VERSION: ${CMAKE_VERSION}")

# see https://cmake.org/cmake/help/latest/policy/CMP0111.html
#
# This is to suppress the warnings for importing PyTorch.
if(POLICY CMP0111)
  cmake_policy(SET CMP0111 OLD)
endif()
if(POLICY CMP0074)
  cmake_policy(SET CMP0074 NEW)
endif()

set(CMAKE_DISABLE_FIND_PACKAGE_MKL TRUE)

set(languages CXX)
set(_K2_WITH_CUDA ON)

find_program(K2_HAS_NVCC nvcc)
if(NOT K2_HAS_NVCC AND "$ENV{CUDACXX}" STREQUAL "")
  message(STATUS "No NVCC detected. Disable CUDA support")
  set(_K2_WITH_CUDA OFF)
endif()

if(APPLE OR (DEFINED K2_WITH_CUDA AND NOT K2_WITH_CUDA))
  if(_K2_WITH_CUDA)
    message(STATUS "Disable CUDA support")
    set(_K2_WITH_CUDA OFF)
  endif()
endif()

if(_K2_WITH_CUDA)
  set(languages ${languages} CUDA)
endif()

message(STATUS "Enabled languages: ${languages}")

project(k2 ${languages})

set(K2_VERSION "1.24.4")

# ----------------- Supported build types for K2 project -----------------
set(K2_ALLOWABLE_BUILD_TYPES Debug Release RelWithDebInfo MinSizeRel)
set(K2_DEFAULT_BUILD_TYPE "Release")
set_property(CACHE CMAKE_BUILD_TYPE PROPERTY STRINGS "${K2_ALLOWABLE_BUILD_TYPES}")
if(NOT CMAKE_BUILD_TYPE AND NOT CMAKE_CONFIGURATION_TYPES)
  # CMAKE_CONFIGURATION_TYPES: with config type values from other generators (IDE).
  message(STATUS "No CMAKE_BUILD_TYPE given, default to ${K2_DEFAULT_BUILD_TYPE}")
  set(CMAKE_BUILD_TYPE "${K2_DEFAULT_BUILD_TYPE}")
elseif(NOT CMAKE_BUILD_TYPE IN_LIST K2_ALLOWABLE_BUILD_TYPES)
  message(FATAL_ERROR "Invalid build type: ${CMAKE_BUILD_TYPE}, \
    choose one from ${K2_ALLOWABLE_BUILD_TYPES}")
endif()

string(TOUPPER ${CMAKE_BUILD_TYPE} CMAKE_BUILD_TYPE_UPPERCASE)
if("${CMAKE_BUILD_TYPE_UPPERCASE}" STREQUAL "DEBUG")
  # refer to https://docs.nvidia.com/cuda/cuda-memcheck/index.html#compilation-options
  # The two options are to make cuda-memcheck's stack backtrace feature more useful.
  set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} --compiler-options -rdynamic --compiler-options -lineinfo")
endif()

set(CMAKE_EXPORT_COMPILE_COMMANDS ON)

option(BUILD_SHARED_LIBS "Whether to build shared or static lib" ON)
option(K2_USE_PYTORCH "Whether to build with PyTorch" ON)
option(K2_ENABLE_BENCHMARK "Whether to enable benchmark" ON)
option(K2_WITH_CUDA "Whether to build k2 with CUDA" ${_K2_WITH_CUDA})
option(K2_ENABLE_NVTX "Whether to build k2 with the NVTX library" ON)
option(K2_ENABLE_TESTS "Whether to build tests" ON)

# You have to enable this option if you will run k2 on a machine different from
# the one you used to build k2 and the two machines have different types of GPUs
#
# We set it to OFF by default since most users will use the same machine to build
# and run k2. If you enable it, it will take much longer to build k2 and the
# resulting shared library also becomes larger in filesize.
option(K2_BUILD_FOR_ALL_ARCHS "Whether to build k2 for all GPU architectures" OFF)

if(NOT K2_WITH_CUDA)
  message(STATUS "Set K2_ENABLE_NVTX to OFF since K2_WITH_CUDA is OFF")
  set(K2_ENABLE_NVTX OFF CACHE BOOL "" FORCE)
endif()

if(NOT K2_USE_PYTORCH)
  message(FATAL_ERROR "\
    Please set K2_USE_PYTORCH to ON.
    Support for other frameworks will be added later")
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")

if(BUILD_SHARED_LIBS AND MSVC)
  set(CMAKE_WINDOWS_EXPORT_ALL_SYMBOLS ON)
endif()

set(CMAKE_SKIP_BUILD_RPATH FALSE)
set(BUILD_RPATH_USE_ORIGIN TRUE)
set(CMAKE_INSTALL_RPATH_USE_LINK_PATH TRUE)

if(NOT APPLE)
  set(k2_rpath_origin "$ORIGIN")
else()
  set(k2_rpath_origin "@loader_path")
endif()

set(CMAKE_INSTALL_RPATH ${k2_rpath_origin})
set(CMAKE_BUILD_RPATH ${k2_rpath_origin})

if(UNIX AND NOT APPLE)
  execute_process(COMMAND
    lsb_release -sd
    OUTPUT_VARIABLE K2_OS
    OUTPUT_STRIP_TRAILING_WHITESPACE
  )
elseif(APPLE)
  execute_process(COMMAND
    sw_vers -productName
    OUTPUT_VARIABLE _product_name
    OUTPUT_STRIP_TRAILING_WHITESPACE
  )

  execute_process(COMMAND
    sw_vers -productVersion
    OUTPUT_VARIABLE _product_version
    OUTPUT_STRIP_TRAILING_WHITESPACE
  )

  execute_process(COMMAND
    sw_vers -buildVersion
    OUTPUT_VARIABLE _build_version
    OUTPUT_STRIP_TRAILING_WHITESPACE
  )
  set(K2_OS "${_product_name} ${_product_version} ${_build_version}")
elseif(WIN32)
  execute_process(COMMAND
    wmic os get caption,version
    OUTPUT_VARIABLE K2_OS_TWO_LINES
    OUTPUT_STRIP_TRAILING_WHITESPACE
  )
  # Now K2_OS_TWO_LINES contains something like
  #  Caption                          Version
  #  Microsoft Windows 10 Pro         10.0.18362
  string(REPLACE "\n" ";" K2_OS_LIST ${K2_OS_TWO_LINES})
  list(GET K2_OS_LIST 1 K2_OS)
else()
  set(K2_OS "Unknown")
endif()

string(REGEX REPLACE "^\"+|\"+$" "" K2_OS "${K2_OS}")
message(STATUS "K2_OS: ${K2_OS}")

find_package(Git REQUIRED)
execute_process(COMMAND
  "${GIT_EXECUTABLE}" describe --always --abbrev=40
  WORKING_DIRECTORY "${CMAKE_SOURCE_DIR}"
  OUTPUT_VARIABLE K2_GIT_SHA1
  ERROR_QUIET OUTPUT_STRIP_TRAILING_WHITESPACE
)

execute_process(COMMAND
  "${GIT_EXECUTABLE}" log -1 --format=%ad --date=local
  WORKING_DIRECTORY "${CMAKE_SOURCE_DIR}"
  OUTPUT_VARIABLE K2_GIT_DATE
  ERROR_QUIET OUTPUT_STRIP_TRAILING_WHITESPACE
)

if(NOT CMAKE_CXX_STANDARD)
  set(CMAKE_CXX_STANDARD 14 CACHE STRING "The C++ version to be used.")
endif()
set(CMAKE_CXX_EXTENSIONS OFF)

message(STATUS "C++ Standard version: ${CMAKE_CXX_STANDARD}")

include(CheckIncludeFileCXX)
check_include_file_cxx(cxxabi.h K2_HAVE_CXXABI_H)
check_include_file_cxx(execinfo.h K2_HAVE_EXECINFO_H)

include(CheckCXXCompilerFlag)
if(NOT WIN32)
  check_cxx_compiler_flag("-std=c++${CMAKE_CXX_STANDARD}" K2_COMPILER_SUPPORTS_CXX${CMAKE_CXX_STANDARD})
else()
  # windows x86 or x86_64
  check_cxx_compiler_flag("/std:c++${CMAKE_CXX_STANDARD}" K2_COMPILER_SUPPORTS_CXX${CMAKE_CXX_STANDARD})
endif()
if(NOT K2_COMPILER_SUPPORTS_CXX${CMAKE_CXX_STANDARD})
  message(FATAL_ERROR "
    k2 requires a compiler supporting at least C++${CMAKE_CXX_STANDARD}.
    If you are using GCC, please upgrade it to at least version 7.0.
    If you are using Clang, please upgrade it to at least version 3.4.")
endif()

# ========= Settings for CUB begin =========
# the following settings are modified from cub/CMakeLists.txt

if(K2_WITH_CUDA)
  # Force CUDA C++ standard to be the same as the C++ standard used.
  #
  # Now, CMake is unaligned with reality on standard versions: https://gitlab.kitware.com/cmake/cmake/issues/18597
  # which means that using standard CMake methods, it's impossible to actually sync the CXX and CUDA versions for pre-11
  # versions of C++; CUDA accepts 98 but translates that to 03, while CXX doesn't accept 03 (and doesn't translate that to 03).
  # In case this gives You, dear user, any trouble, please escalate the above CMake bug, so we can support reality properly.
  if(DEFINED CMAKE_CUDA_STANDARD)
    message(WARNING "You've set CMAKE_CUDA_STANDARD; please note that this variable is ignored, and CMAKE_CXX_STANDARD"
      " is used as the C++ standard version for both C++ and CUDA.")
  endif()


  unset(CMAKE_CUDA_STANDARD CACHE)
  set(CMAKE_CUDA_STANDARD ${CMAKE_CXX_STANDARD})

  include(cmake/select_compute_arch.cmake)
  cuda_select_nvcc_arch_flags(K2_COMPUTE_ARCH_FLAGS)
  message(STATUS "K2_COMPUTE_ARCH_FLAGS: ${K2_COMPUTE_ARCH_FLAGS}")

  # set(K2_COMPUTE_ARCHS 30 32 35 50 52 53 60 61 62 70 72)
  # message(WARNING "arch 62/72 are not supported for now")

  # see https://arnon.dk/matching-sm-architectures-arch-and-gencode-for-various-nvidia-cards/
  # https://www.myzhar.com/blog/tutorials/tutorial-nvidia-gpu-cuda-compute-capability/
  set(K2_COMPUTE_ARCH_CANDIDATES 35 50 60 61 70 75)
  if(CUDA_VERSION VERSION_GREATER "11.0")
      list(APPEND K2_COMPUTE_ARCH_CANDIDATES 80 86)
      if(WIN32)
        # To fix the following warning from PyTorch:
        # c10/util/TypeCast.h(39): warning : calling a constexpr __host__ function from a
        # __host__ __device__ function is not allowed. The experimental flag '--expt-relaxed-constexpr'
        # can be used to allow this
        string(APPEND CMAKE_CUDA_FLAGS " --expt-relaxed-constexpr ")
      endif()

      string(APPEND CMAKE_CUDA_FLAGS " -Wno-deprecated-gpu-targets ")
  endif()

  if(CUDA_VERSION VERSION_GREATER_EQUAL "11.8")
    # https://docs.nvidia.com/cuda/archive/11.8.0/pdf/CUDA_Toolkit_Release_Notes.pdf
    # added support for Hopper (90)
    list(APPEND K2_COMPUTE_ARCH_CANDIDATES 90)
  endif()

  if(CUDA_VERSION VERSION_GREATER_EQUAL "12.0")
    # https://docs.nvidia.com/cuda/archive/12.0.0/pdf/CUDA_Toolkit_Release_Notes.pdf
    # support for 35, 37 compute capabilities removed in CUDA 12.0
    list(REMOVE_ITEM K2_COMPUTE_ARCH_CANDIDATES 35 37)
  endif()

  message(STATUS "K2_COMPUTE_ARCH_CANDIDATES ${K2_COMPUTE_ARCH_CANDIDATES}")

  if(NOT K2_BUILD_FOR_ALL_ARCHS)
    set(K2_COMPUTE_ARCHS)

    foreach(COMPUTE_ARCH IN LISTS K2_COMPUTE_ARCH_CANDIDATES)
      if("${K2_COMPUTE_ARCH_FLAGS}" MATCHES ${COMPUTE_ARCH})
        message(STATUS "Adding arch ${COMPUTE_ARCH}")
        list(APPEND K2_COMPUTE_ARCHS ${COMPUTE_ARCH})
      else()
        message(STATUS "Skipping arch ${COMPUTE_ARCH}")
      endif()
    endforeach()

    if(NOT K2_COMPUTE_ARCHS)
      set(K2_COMPUTE_ARCHS ${K2_COMPUTE_ARCH_CANDIDATES})
    endif()
  else()
      message(STATUS "Build k2 for all GPU architectures")
      set(K2_COMPUTE_ARCHS ${K2_COMPUTE_ARCH_CANDIDATES})
  endif()

  message(STATUS "K2_COMPUTE_ARCHS: ${K2_COMPUTE_ARCHS}")

  # Flags for moderngpu
  set(K2_MODERN_GPU_FLAGS " -lineinfo --expt-extended-lambda -use_fast_math -Xptxas=-w ")

  foreach(COMPUTE_ARCH IN LISTS K2_COMPUTE_ARCHS)
    set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} ${K2_MODERN_GPU_FLAGS} --expt-extended-lambda -gencode arch=compute_${COMPUTE_ARCH},code=sm_${COMPUTE_ARCH}")
    set(CMAKE_CUDA_ARCHITECTURES "${COMPUTE_ARCH}-real;${COMPUTE_ARCH}-virtual;${CMAKE_CUDA_ARCHITECTURES}")
  endforeach()
# ========= Settings for CUB end=========
endif()

list(APPEND CMAKE_MODULE_PATH ${CMAKE_SOURCE_DIR}/cmake/Modules)
find_package(Valgrind)
if(Valgrind_FOUND)
  find_program(MEMORYCHECK_COMMAND NAMES ${Valgrind_EXECUTABLE})
  set(MEMORYCHECK_COMMAND_OPTIONS "--suppressions=${CMAKE_SOURCE_DIR}/scripts/valgrind.supp --leak-check=full")
  include(Dart)
  message(STATUS "To check memory, run `ctest -R <NAME> -D ExperimentalMemCheck`")
endif()

list(APPEND CMAKE_MODULE_PATH ${CMAKE_SOURCE_DIR}/cmake)


include(pybind11)
if(K2_USE_PYTORCH)
  add_definitions(-DK2_USE_PYTORCH)
  include(torch)
  configure_file(
    ${PROJECT_SOURCE_DIR}/k2/python/k2/torch_version.py.in
    ${PROJECT_SOURCE_DIR}/k2/python/k2/torch_version.py @ONLY
  )
  message(STATUS "Generated ${PROJECT_BINARY_DIR}/torch_version.py")
endif()

if(K2_WITH_CUDA)
  add_definitions(-DK2_WITH_CUDA)
endif()

if(WIN32)
  add_definitions(-DNOMINMAX) # Otherwise, std::max() and std::min() won't work
endif()

if(K2_WITH_CUDA AND CUDA_VERSION VERSION_LESS 11.0)
  # CUB is included in CUDA toolkit 11.0 and above
  include(cub)
endif()

if(K2_WITH_CUDA)
  include(moderngpu)
endif()

if(K2_ENABLE_TESTS)
  enable_testing()
  include(googletest)
endif()


if(K2_WITH_CUDA AND NOT WIN32)
  string(APPEND CMAKE_CUDA_FLAGS " --compiler-options -Wall ")
  string(APPEND CMAKE_CUDA_FLAGS " --compiler-options -Wno-strict-overflow ")
  string(APPEND CMAKE_CUDA_FLAGS " --compiler-options -Wno-unknown-pragmas ")
  message(STATUS "CMAKE_CUDA_FLAGS: ${CMAKE_CUDA_FLAGS}")
endif()


if(NOT WIN32)
  string(APPEND CMAKE_CXX_FLAGS " -Wno-unused-variable ")
  string(APPEND CMAKE_CXX_FLAGS " -Wno-strict-overflow ")
endif()

if(WIN32)
  # disable various warnings for MSVC
  # NOTE: Most of the warnings are from PyTorch C++ APIs
  # 4005: macro redefinition
  # 4018: signed/unsigned mismatch
  # 4067: unexpected tokens following preprocessor directive
  # 4068: unknown pragma "unroll"
  # 4099: type name first seen using 'class' now seen using 'struct'
  # 4101: 'identifier' : unreferenced local variable
  # 4190: 'identifier1' has C-linkage specified, but returns UDT 'identifier2' which is incompatible with C
  # 4224: conversion from 'int64_t' to 'int32_t', possible loss of data
  # 4244: conversion from 'const M' to 'const FloatType'
  # 4251: 'type' : class 'type1' needs to have dll-interface to be used by clients of class 'type2'
  # 4267: conversion from 'size_t' to 'I', possible loss of data
  # 4275: non - DLL-interface class 'class_1' used as base for DLL-interface class 'class_2'
  # 4305: truncation from 'int' to 'bool'
  # 4522: 'class' : multiple assignment operators specified
  # 4551: function call missing argument list
  # 4624: destructor was implicitly defined as deleted
  # 4700: uninitialized local variable 'device' used
  # 4722: destructor never returns
  # 4805: '|': unsafe mix of type 'uintptr_t' and type 'bool' in operation
  # 4819: The file contains a character that cannot be presented in the current code page.
  # 4838: conversion from 'type_1' to 'type_2' requires a narrowing conversion
  # 4996: "getenv": This function is unsafe
  set(disabled_warnings
      /wd4005
      /wd4018
      /wd4067
      /wd4068
      /wd4099
      /wd4101
      /wd4190
      /wd4224
      /wd4244
      /wd4251
      /wd4267
      /wd4275
      /wd4305
      /wd4522
      /wd4551
      /wd4624
      /wd4700
      /wd4722
      /wd4805
      /wd4819
      /wd4838
      /wd4996
  )
  message(STATUS "Disabled warnings: ${disabled_warnings}")
  foreach(w IN LISTS disabled_warnings)
    string(APPEND CMAKE_CXX_FLAGS " ${w} ")
    string(APPEND CMAKE_CUDA_FLAGS " --compiler-options ${w} ")
  endforeach()
  string(APPEND CMAKE_CXX_FLAGS " /bigobj ")
endif()

message(STATUS "CMAKE_CXX_FLAGS: ${CMAKE_CXX_FLAGS}")
message(STATUS "CMAKE_CUDA_FLAGS: ${CMAKE_CUDA_FLAGS}")

include(GNUInstallDirs)

add_subdirectory(k2)

configure_file(
  ${PROJECT_SOURCE_DIR}/cmake/k2ConfigVersion.cmake.in
  ${PROJECT_BINARY_DIR}/k2ConfigVersion.cmake
  @ONLY
)

configure_file(
  ${PROJECT_SOURCE_DIR}/cmake/k2Config.cmake.in
  ${PROJECT_BINARY_DIR}/k2Config.cmake
  @ONLY
)

install(FILES
  ${PROJECT_BINARY_DIR}/k2ConfigVersion.cmake
  ${PROJECT_BINARY_DIR}/k2Config.cmake
  DESTINATION share/cmake/k2
)

install(FILES
  ${PROJECT_SOURCE_DIR}/k2/python/k2/torch_version.py
  DESTINATION ./
)