IF (APPLE)
    cmake_minimum_required(VERSION 3.4)
ELSE()
    cmake_minimum_required(VERSION 2.8)
ENDIF()

project(ctc_release)

IF (NOT APPLE)
    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -O2")
ENDIF()

IF (APPLE)
    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -std=c++14 -O2")
    add_definitions(-DAPPLE)
ENDIF()

include_directories(include)

FIND_PACKAGE(CUDA 6.5)
MESSAGE(STATUS "cuda found ${CUDA_FOUND}")

option(WITH_GPU "compile warp-ctc with cuda." ${CUDA_FOUND})
option(WITH_OMP "compile warp-ctc with openmp." ON)

if(NOT WITH_OMP)
    add_definitions(-DCTC_DISABLE_OMP)
endif()
if (WITH_OMP)
    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -fopenmp")
    set(CUDA_NVCC_FLAGS "${CUDA_NVCC_FLAGS} -Xcompiler -fopenmp")
endif()

# need to be at least 30 or __shfl_down in reduce wont compile
set(CUDA_NVCC_FLAGS "${CUDA_NVCC_FLAGS} -gencode arch=compute_50,code=sm_50")
set(CUDA_NVCC_FLAGS "${CUDA_NVCC_FLAGS} -gencode arch=compute_52,code=sm_52")
IF(CMAKE_CXX_COMPILER_VERSION VERSION_GREATER 5)
  SET(CUDA_NVCC_FLAGS "${CUDA_NVCC_FLAGS} -D_MWAITXINTRIN_H_INCLUDED -D_FORCE_INLINES")
ENDIF()

IF (CUDA_VERSION GREATER 7.6)
    set(CUDA_NVCC_FLAGS "${CUDA_NVCC_FLAGS} -gencode arch=compute_60,code=sm_60")
    set(CUDA_NVCC_FLAGS "${CUDA_NVCC_FLAGS} -gencode arch=compute_61,code=sm_61")
    set(CUDA_NVCC_FLAGS "${CUDA_NVCC_FLAGS} -gencode arch=compute_62,code=sm_62")
ENDIF()

IF (CUDA_VERSION GREATER 8.9)
    set(CUDA_NVCC_FLAGS "${CUDA_NVCC_FLAGS} -gencode arch=compute_70,code=sm_70")
    set(CUDA_NVCC_FLAGS "${CUDA_NVCC_FLAGS} -gencode arch=compute_75,code=sm_75")
    set(CUDA_NVCC_FLAGS "${CUDA_NVCC_FLAGS} -gencode arch=compute_80,code=sm_80")
    set(CUDA_NVCC_FLAGS "${CUDA_NVCC_FLAGS} -gencode arch=compute_90,code=sm_90")
ENDIF()

if (NOT APPLE)
    set(CUDA_NVCC_FLAGS "${CUDA_NVCC_FLAGS} --std=c++14")
    set(CUDA_NVCC_FLAGS "${CUDA_NVCC_FLAGS}")
ENDIF()

IF (APPLE)
    EXEC_PROGRAM(uname ARGS -v  OUTPUT_VARIABLE DARWIN_VERSION)
    STRING(REGEX MATCH "[0-9]+" DARWIN_VERSION ${DARWIN_VERSION})
    MESSAGE(STATUS "DARWIN_VERSION=${DARWIN_VERSION}")

    #for el capitain have to use rpath

    IF (DARWIN_VERSION LESS 15)
        set(CMAKE_SKIP_RPATH TRUE)
    ENDIF ()

ELSE()
    #always skip for linux
    set(CMAKE_SKIP_RPATH TRUE)
ENDIF()


IF (WITH_GPU)

    MESSAGE(STATUS "Building shared library with GPU support")

    CUDA_ADD_LIBRARY(warpctc SHARED src/ctc_entrypoint.cu src/reduce.cu)
    IF (!Torch_FOUND)
        TARGET_LINK_LIBRARIES(warpctc ${CUDA_curand_LIBRARY})
    ENDIF()

    add_executable(test_cpu tests/test_cpu.cpp tests/random.cpp )
    TARGET_LINK_LIBRARIES(test_cpu warpctc)
    SET_TARGET_PROPERTIES(test_cpu PROPERTIES COMPILE_FLAGS "${CMAKE_CXX_FLAGS} --std=c++14")

    cuda_add_executable(test_gpu tests/test_gpu.cu tests/random.cpp )
    TARGET_LINK_LIBRARIES(test_gpu warpctc ${CUDA_curand_LIBRARY})
    SET_TARGET_PROPERTIES(test_gpu PROPERTIES COMPILE_FLAGS "${CMAKE_CXX_FLAGS} --std=c++14")

    INSTALL(TARGETS warpctc
            RUNTIME DESTINATION "bin"
            LIBRARY DESTINATION "lib"
            ARCHIVE DESTINATION "lib")

    INSTALL(FILES include/ctc.h DESTINATION "include")
ELSE()
    MESSAGE(STATUS "Building shared library with no GPU support")

    if (NOT APPLE)
        set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -std=c++14 -O2")
    ENDIF()

    ADD_LIBRARY(warpctc SHARED src/ctc_entrypoint.cpp)

    add_executable(test_cpu tests/test_cpu.cpp tests/random.cpp )
    TARGET_LINK_LIBRARIES(test_cpu warpctc)
    SET_TARGET_PROPERTIES(test_cpu PROPERTIES COMPILE_FLAGS "${CMAKE_CXX_FLAGS} --std=c++14")

    INSTALL(TARGETS warpctc
            RUNTIME DESTINATION "bin"
            LIBRARY DESTINATION "lib"
            ARCHIVE DESTINATION "lib")

    INSTALL(FILES include/ctc.h DESTINATION "include")
ENDIF()
