common_settings.cmake

# This is a CMake build file, for more information consult:
# http://en.wikipedia.org/wiki/CMake
# and
# http://www.cmake.org/Wiki/CMake
# http://www.cmake.org/cmake/help/syntax.html
# http://www.cmake.org/Wiki/CMake_Useful_Variables
# http://www.cmake.org/cmake/help/cmake-2-8-docs.html

# This file is intended to be included by other cmake files (see src/applications/*/CMakeLists.txt)

# ----------------------------------------------------------------------

site_name(HOSTNAME)

# Use "gcc -v -Q -march=native -O3 test.c -o test" to see which options the compiler actualy uses
# using -march=native will include all sse options available on the given machine (-msse -msse2 -msse3 -mssse3 -msse4.1 -msse4.2, etc...)
# also using -march=native will imply -mtune=native
# Thus the optimization flags below should work great on all machines
# (O3 is already added by CMAKE_CXX_FLAGS_RELEASE)
set(OPT_CXX_FLAGS "-fopenmp -ffast-math -funroll-loops -march=native")
#set(OPT_CXX_FLAGS "-fopenmp -ffast-math -funroll-loops -march=native -freciprocal-math -funsafe-math-optimizations -fassociative-math -ffinite-math-only -fcx-limited-range")  # cheap -Ofast copy
#set(OPT_CXX_FLAGS "-ffast-math -funroll-loops -march=native") # disabled OpenMp, just for testing
#set(OPT_CXX_FLAGS "-fopenmp -ffast-math -funroll-loops") # disable native compilation, so we can profile with valgrind/callgrind

# enable link time optimization
# http://gcc.gnu.org/onlinedocs/gcc/Optimize-Options.html
#set(OPT_CXX_FLAGS "${OPT_CXX_FLAGS} -flto")
#set(CMAKE_EXE_LINKER_FLAGS "${CMAKE_EXE_LINKER_FLAGS} -flto")
#set(CMAKE_SHARED_LINKER_FLAGS "${CMAKE_SHARED_LINKER_FLAGS} -flto")
#set(CMAKE_MODULE_LINKER_FLAGS "${CMAKE_MODULE_LINKER_FLAGS} -flto")

#option(CUDA_HOST_SHARED_FLAGS OFF) # cuda/nvcc uses a separate set of options than gcc

set(VISICS_MACHINES 
  "vesta" "nereid" # intel vpro
  "enif" # core 2 quad
  "kochab" "oculus"  "izar" "yildun" "watar" "sadr" # intel v7, core i7 860  @ 2.80GHz
  "unuk" # intel v7, cire i7 870 @ 2.93GHz
  "jabbah" "matar" # top of the line cpu and gpu
)


list(FIND VISICS_MACHINES ${HOSTNAME} HOSTED_AT_VISICS)
#message(STATUS "HOSTED_AT_VISICS == ${HOSTED_AT_VISICS}")

if(HOSTED_AT_VISICS GREATER -1)
  message(STATUS "Using ${VISICS_MACHINES} optimisation options")

  # since gcc 4.6 the option -Ofast provides faster than -O3
  set(CMAKE_CXX_FLAGS_RELEASE "-O3 -DNDEBUG")
  #set(CMAKE_CXX_FLAGS_RELEASE "-Ofast -DNDEBUG")

  # add local compiled opencv trunk in the pkg-config paths
  set(PKG_CONFIG_PATH ${PKG_CONFIG_PATH}:/users/visics/rbenenso/no_backup/usr/local/lib/pkgconfig)

  #set(OPT_CXX_FLAGS "-fopenmp -march=native -mtune=native -ffast-math -msse -msse2 -msse3 -mssse3 -msse4.1 -msse4.2") # just for testing
  
  option(USE_GPU "Should the GPU be used ?" ON)
  #option(USE_GPU "Should the GPU be used ?" OFF) # set to false for testing purposes only
  set(CUDA_BUILD_EMULATION OFF CACHE BOOL "enable emulation mode")
  set(CUDA_BUILD_CUBIN OFF)
  set(local_CUDA_CUT_INCLUDE_DIRS "/users/visics/rbenenso/code/references/cuda/cuda_sdk/C/common/inc")
  set(local_CUDA_CUT_LIBRARY_DIRS "/users/visics/rbenenso/code/references/cuda/cuda_sdk/C/lib")
  set(local_CUDA_LIB_DIR "/usr/lib64/nvidia")
  set(local_CUDA_LIB "/usr/lib64/nvidia/libcuda.so")
  set(cuda_LIBS "cuda")
  set(cutil_LIB "cutil")

  # if you get error messages in nvcc-generated files,
  # enable the following line for debugging:
  #set(CUDA_NVCC_FLAGS "${CUDA_NVCC_FLAGS};--keep")
  #set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} --host-compilation c++ --device-compilation c++)
  set(CUDA_NVCC_EXECUTABLE  /users/visics/rbenenso/code/references/cuda/gcc-4.4/nvcc-4.4.sh)
  set(CUDA_SDK_ROOT_DIR  /users/visics/rbenenso/code/references/cuda_sdk_4.0.17/C)
  set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} --compiler-options -D__USE_XOPEN2K8) # black magic required on Visics machines

  set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} -arch=sm_20) # only matar, jabbah and yildun can run current code

  # faster malloc, and a good profiler via http://google-perftools.googlecode.com
  #set(google_perftools_LIBS tcmalloc profiler)
  set(google_perftools_LIBS tcmalloc_and_profiler)

  set(liblinear_INCLUDE_DIRS "/users/visics/rbenenso/code/references/machine_learning/liblinear-1.8")
  set(liblinear_LIBRARY_DIRS "/users/visics/rbenenso/code/references/machine_learning/liblinear-1.8")

 add_definitions("-Dint_p_NULL=((int*)0)")

elseif(${HOSTNAME} STREQUAL  "rodrigob-laptop")
  message(STATUS "Using rodrigob-laptop optimisation options")

  option(USE_GPU "Should the GPU be used ?" FALSE)
  set(CUDA_BUILD_EMULATION ON CACHE BOOL "enable emulation mode")
  set(CUDA_BUILD_CUBIN OFF)
  set(local_CUDA_CUT_INCLUDE_DIRS "/home/rodrigob/work/code/biclop_references/cuda/cuda_sdk/C/common/inc")
  set(local_CUDA_CUT_LIBRARY_DIRS "/home/rodrigob/work/code/biclop_references/cuda/cuda_sdk/C/lib")
  set(cuda_LIBS "")
  set(cutil_LIB "cutil")

  set(GCC44_DIRECTORY "/home/rodrigob/work/code/biclop_references/cuda/gcc-4.4/")
  set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} --compiler-bindir ${GCC44_DIRECTORY})

  # faster malloc, and a good profiler via http://google-perftools.googlecode.com
  set(google_perftools_LIBS tcmalloc profiler)
  set(EUROPA_SVN "/home/rodrigob/work/code/europa_svn/code")

  set(liblinear_INCLUDE_DIRS "/home/rodrigob/work/code/biclop_references/liblinear-1.8")
  set(liblinear_LIBRARY_DIRS "/home/rodrigob/work/code/biclop_references/liblinear-1.8")


elseif(${HOSTNAME} STREQUAL  "visics-gt680r")
  message(STATUS "Using visics-gt680r optimisation options")

  option(USE_GPU "Should the GPU be used ?" TRUE)
  #set(CUDA_BUILD_EMULATION OFF CACHE BOOL "enable emulation mode")
  set(CUDA_BUILD_CUBIN OFF)
  
  # work around to use gcc-4.4 instead of 4.5
  #set(CUDA_NVCC_EXECUTABLE "/home/rodrigob/code/references/cuda/gcc-4.4/nvcc-4.4.sh") 
  set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} -arch=sm_21)

  set(local_CUDA_CUT_INCLUDE_DIRS "/home/rodrigob/code/references/cuda/cuda_sdk/C/common/inc")
  set(local_CUDA_CUT_LIBRARY_DIRS "/home/rodrigob/code/references/cuda/cuda_sdk/C/lib")
  set(local_CUDA_LIB_DIR "/usr/local/cuda/lib64")
  set(cuda_LIBS "")
  set(cutil_LIB "cutil")

  # faster malloc, and a good profiler via http://google-perftools.googlecode.com
  set(google_perftools_LIBS tcmalloc profiler)
  set(EUROPA_SVN "/home/rodrigob/code/europa_svn/code")

  set(liblinear_INCLUDE_DIRS "/home/rodrigob/code/references/liblinear-1.8")
  set(liblinear_LIBRARY_DIRS "/home/rodrigob/code/references/liblinear-1.8")

elseif(${HOSTNAME} STREQUAL  "lap-12-31")
  message(STATUS "Using lap-12-31 optimisation options")

  set(CMAKE_CXX_COMPILER "/usr/bin/g++-4.7")
  set(CMAKE_C_COMPILER "/usr/bin/g++-4.7")
  # since gcc 4.6 -Ofast is faster than -O3
  set(CMAKE_C_FLAGS_RELEASE "-Ofast -DNDEBUG")
  set(CMAKE_CXX_FLAGS_RELEASE "-Ofast -DNDEBUG")

  set(CUDA_PROPAGATE_HOST_FLAGS "OFF") # since nvcc does not handle -Ofast
  set(CUDA_NVCC_FLAGS_RELEASE "-O3 -DNDEBUG")
  set(CUDA_NVCC_FLAGS_RELWITHDEBINFO "-O3 -DNDEBUG")

  # add local compiled opencv trunk in the pkg-config paths
  #set(PKG_CONFIG_PATH ${PKG_CONFIG_PATH}:/home/mfritz/local/lib/pkgconfig)
  #set(PKG_CONFIG_PATH /home/mfritz/local/lib/pkgconfig)

  set(opencv_INCLUDE_DIRS "/usr/local/include/opencv2")
  set(opencv_LIBRARY_DIRS "/usr/local/lib")

  option(USE_GPU "Should the GPU be used ?" TRUE)
  #set(CUDA_BUILD_EMULATION OFF CACHE BOOL "enable emulation mode")
  set(CUDA_BUILD_CUBIN OFF)

  # work around to use gcc-4.4 instead of 4.5
  #set(CUDA_NVCC_EXECUTABLE "/home/rodrigob/code/references/cuda/gcc-4.4/nvcc-4.4.sh")
  set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} -arch=sm_30)

  set(local_CUDA_CUT_INCLUDE_DIRS "/home/rodrigob/code/references/cuda/cuda_sdk/C/common/inc")
  set(local_CUDA_CUT_LIBRARY_DIRS "/home/rodrigob/code/references/cuda/cuda_sdk/C/lib")
  set(local_CUDA_LIB_DIR "/usr/local/cuda/lib64")
  set(cuda_LIBS "")
  #set(cutil_LIB "cutil")
  set(cutil_LIB "") # no cutil

  # faster malloc, and a good profiler via http://google-perftools.googlecode.com
  set(google_perftools_LIBS tcmalloc profiler)
  set(EUROPA_SVN "/home/rodrigob/code/europa_svn/code")

  set(liblinear_INCLUDE_DIRS "/home/rodrigob/code/references/liblinear-1.8")
  set(liblinear_LIBRARY_DIRS "/home/rodrigob/code/references/liblinear-1.8")

  include_directories(
   # "/home/rodrigob/code/references/libwebp-0.3.0-linux-x86-64/include"
  )

  link_directories(
   # "/home/rodrigob/code/references/libwebp-0.3.0-linux-x86-64/lib"
  )

else ()
  message(FATAL_ERROR, "Unknown machine, please add your configuration inside biclop/common_settings.cmake")
  
endif ()

# ----------------------------------------------------------------------
# enable compilation for shared libraries
set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} -Xcompiler -fpic)

if(CMAKE_BUILD_TYPE MATCHES "Debug")
  # enable cuda debug information, to use with cuda-dbg
  set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} -g -G)
else()
# FIXME disabled only for testing
#  set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} -O3 --use_fast_math) # speed up host and device code
endif()

# ----------------------------------------------------------------------
# set default compilation flags and default build

set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -Wall")
#set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -g") # add debug information, even in release mode
set(CMAKE_CXX_FLAGS_RELEASE "${CMAKE_CXX_FLAGS_RELEASE} -Wall -DNDEBUG -DBOOST_DISABLE_ASSERTS ${OPT_CXX_FLAGS}")
#set(CMAKE_CXX_FLAGS_RELWITHDEBINFO "${CMAKE_CXX_FLAGS_RELWITHDEBINFO} ${CMAKE_CXX_FLAGS_RELEASE}")
set(CMAKE_CXX_FLAGS_RELWITHDEBINFO "${CMAKE_CXX_FLAGS_RELEASE} -g")  
set(CMAKE_CXX_FLAGS_DEBUG "${CMAKE_CXX_FLAGS_DEBUG} -DDEBUG") 


if(USE_GPU)
  add_definitions(-DUSE_GPU) 
endif(USE_GPU)


# set default cmake build type (None Debug Release RelWithDebInfo MinSizeRel)
if( NOT CMAKE_BUILD_TYPE )
   set( CMAKE_BUILD_TYPE "Release" )
endif()

# ----------------------------------------------------------------------