# Manage CUDA nvcc compilation configuration, try to be smart to ease the users' # pain as much as possible: # - use the CUDA_HOST_COMPILER if defined by the user, otherwise # - auto-detect compatible nvcc host compiler and set nvcc -ccbin (if not MPI wrapper) # - set icc compatibility mode to gcc 4.4/4.5 (CUDA 4.0 is not compatible with gcc >v4.4) # - (advanced) variables set: # * CUDA_HOST_COMPILER - the host compiler for nvcc (only with cmake <2.8.10) # * CUDA_HOST_COMPILER_OPTIONS - the full host-compiler related option list passed to nvcc # # Note that from CMake 2.8.10 FindCUDA defines CUDA_HOST_COMPILER internally, # so we won't set it ourselves, but hope that the module does a good job. if (NOT DEFINED CUDA_NVCC_FLAGS_SET) set(CUDA_NVCC_FLAGS_SET TRUE CACHE INTERNAL "True if NVCC flags have been set" FORCE) # Explicitly set the host compiler for nvcc if the current compiler is # supported and it's not an MPI compiler wrapper, otherwise warn the user. # # Note that even though nvcc compiles host code as C++, we use the # CMAKE_C_COMPILER as host compiler. We do this because CUDA versions # preceding 5.0 only recognize icc, but not icpc. However, both gcc and icc # (i.e. all supported compilers) happily compile C++ code. # # Also note that with MSVC nvcc sets the -compiler-bindir option behind the # scenes; to avoid conflicts we don't set -ccbin automatically. if (NOT DEFINED CUDA_HOST_COMPILER AND NOT MSVC) if (NOT CMAKE_COMPILER_IS_GNUCC AND NOT (CMAKE_C_COMPILER_ID MATCHES "Intel" AND UNIX AND NOT APPLE)) message(WARNING " Will not set the nvcc host compiler because the current C compiler is not compatible with nvcc: ${CMAKE_C_COMPILER} (ID: ${CMAKE_C_COMPILER_ID}) Compatible compilers are: gcc on Linux and Mac OS X, the Intel Compiler on 64-bit Linux and MSVC on Windows. Note that with newer CUDA releases this might change, for up-to-date compatibility information check the NVIDIA documentation. If nothing specified, nvcc will automatically pick the platform-default compiler; Note that mixing compilers can cause errors. To manually set the nvcc host compiler, edit CUDA_NVCC_FLAGS or re-configure setting CUDA_HOST_COMPILER to the full path of a compatible compiler. ") else() # do not use MPI compiler wrappers, as these are prone to brake nvcc if (GMX_MPI AND NOT "${${MPI_PREFIX}_FOUND}" AND # FindMPI-based detection NOT GMX_THREAD_MPI) message(WARNING " Will not set the nvcc host compiler because the current C compiler is an MPI compiler wrapper: ${CMAKE_C_COMPILER} MPI compiler wrappers are prone to not work with nvcc. You might get lucky, but the safest is to use the C compiler that the MPI compiler wrapper uses (if this is compatible). To manually set the nvcc host compiler, edit CUDA_NVCC_FLAGS or re-configure setting CUDA_HOST_COMPILER to the full path of a compatible compiler. ") else() set(CUDA_HOST_COMPILER "${CMAKE_C_COMPILER}") set(CUDA_HOST_COMPILER_AUTOSET TRUE CACHE INTERNAL "True if CUDA_HOST_COMPILER is automatically set" FORCE) endif() endif() endif() if(DEFINED CUDA_HOST_COMPILER) # FindCUDA in CMake 2.8.10 sets the host compiler internally if (CMAKE_VERSION VERSION_LESS "2.8.10") message(STATUS "Setting the nvcc host compiler to: ${CUDA_HOST_COMPILER}") set(CUDA_HOST_COMPILER ${CUDA_HOST_COMPILER} CACHE PATH "Host compiler for nvcc (do not edit!)" FORCE) set(_HOST_COMPILER_OPTION_STRING "-ccbin=${CUDA_HOST_COMPILER};") endif() # On *nix force icc in gcc 4.4 compatibility mode with CUDA 3.2/4.0 and # gcc 4.5 compatibility mode with later CUDA versions. This is needed # as even with icc use as host compiler, when icc's gcc compatibility # mode is higher than the max gcc version officially supported by CUDA, # nvcc will freak out. if (UNIX AND CMAKE_C_COMPILER_ID MATCHES "Intel" AND CUDA_HOST_COMPILER_AUTOSET) if (CUDA_VERSION VERSION_LESS "4.1") message(STATUS "Setting Intel Compiler compatibity mode to gcc 4.4 for nvcc host compilation") set(CUDA_HOST_COMPILER_OPTIONS "${CUDA_HOST_COMPILER_OPTIONS};-Xcompiler;-gcc-version=440;") else() message(STATUS "Setting Intel Compiler compatibity mode to gcc 4.5 for nvcc host compilation") set(CUDA_HOST_COMPILER_OPTIONS "${CUDA_HOST_COMPILER_OPTIONS};-Xcompiler;-gcc-version=450;") endif() endif() set(CUDA_HOST_COMPILER_OPTIONS "${CUDA_HOST_COMPILER_OPTIONS}" CACHE STRING "Options for nvcc host compiler (do not edit!)." FORCE) mark_as_advanced(CUDA_HOST_COMPILER CUDA_HOST_COMPILER_OPTIONS) endif() # on Linux we need to add -fPIC when building shared gmx libs # Note: will add -fPIC for any compiler that supports it as it shouldn't hurt if(BUILD_SHARED_LIBS) GMX_TEST_CXXFLAG(CXXFLAG_FPIC "-fPIC" _FPIC_NVCC_FLAG) if(_FPIC_NVCC_FLAG) set(CUDA_HOST_COMPILER_OPTIONS "${CUDA_HOST_COMPILER_OPTIONS}-Xcompiler;${_FPIC_NVCC_FLAG}") endif() endif() # Set the CUDA GPU architectures to compile for: # - with CUDA >v4.2 compute capability 2.0, 2.1 is, but 3.0 is not supported: # => compile sm_20, sm_21 cubin, and compute_20 PTX # - with CUDA >=4.2 compute capabity <=3.0 is supported: # => compile sm_20, sm_21, sm_30 cubin, and compute_30 PTX # - with CUDA 5.0 compute capabity 3.5 is supported, but generating code # optimized for sm_35 results in lower performance than with sm_30. if(CUDA_VERSION VERSION_LESS "4.2") set(_CUDA_ARCH_STR "-gencode;arch=compute_20,code=sm_20;-gencode;arch=compute_20,code=sm_21;-gencode;arch=compute_20,code=compute_20") else() set(_CUDA_ARCH_STR "-gencode;arch=compute_20,code=sm_20;-gencode;arch=compute_20,code=sm_21;-gencode;arch=compute_30,code=sm_30;-gencode;arch=compute_30,code=compute_30") endif() # finally set the damn flags set(CUDA_NVCC_FLAGS "${_CUDA_ARCH_STR};-use_fast_math;${_HOST_COMPILER_OPTION_STRING}${CUDA_HOST_COMPILER_OPTIONS}" CACHE STRING "Compiler flags for nvcc." FORCE) endif()