2 # This file is part of the GROMACS molecular simulation package.
4 # Copyright (c) 2012,2013, by the GROMACS development team, led by
5 # Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
6 # and including many others, as listed in the AUTHORS file in the
7 # top-level source directory and at http://www.gromacs.org.
9 # GROMACS is free software; you can redistribute it and/or
10 # modify it under the terms of the GNU Lesser General Public License
11 # as published by the Free Software Foundation; either version 2.1
12 # of the License, or (at your option) any later version.
14 # GROMACS is distributed in the hope that it will be useful,
15 # but WITHOUT ANY WARRANTY; without even the implied warranty of
16 # MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
17 # Lesser General Public License for more details.
19 # You should have received a copy of the GNU Lesser General Public
20 # License along with GROMACS; if not, see
21 # http://www.gnu.org/licenses, or write to the Free Software Foundation,
22 # Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
24 # If you want to redistribute modifications to GROMACS, please
25 # consider that scientific software is very special. Version
26 # control is crucial - bugs must be traceable. We will be happy to
27 # consider code for inclusion in the official distribution, but
28 # derived work must not be called official GROMACS. Details are found
29 # in the README & COPYING files - if they are missing, get the
30 # official version at http://www.gromacs.org.
32 # To help us fund GROMACS development, we humbly ask that you cite
33 # the research papers on the package. Check out http://www.gromacs.org.
35 # If the user did not set GMX_GPU we'll consider this option to be
36 # in "auto" mode meaning that we will:
37 # - search for CUDA and set GMX_GPU=ON we find it
38 # - check whether GPUs are present
39 # - if CUDA is not found but GPUs were detected issue a warning
40 if (NOT DEFINED GMX_GPU)
41 set(GMX_GPU_AUTO TRUE CACHE INTERNAL "GPU acceleration will be selected automatically")
43 option(GMX_GPU "Enable GPU acceleration" OFF)
45 if(GMX_GPU AND GMX_DOUBLE)
46 message(FATAL_ERROR "GPU acceleration is not available in double precision!")
48 if(GMX_GPU_AUTO AND GMX_DOUBLE)
49 message(WARNING "GPU acceleration is not available in double precision, disabled!")
50 set_property(CACHE GMX_GPU PROPERTY VALUE OFF)
51 set_property(CACHE GMX_GPU_AUTO PROPERTY VALUE OFF)
54 # detect GPUs in the build host machine
55 if ((GMX_GPU OR GMX_GPU_AUTO) AND NOT GMX_GPU_DETECTION_DONE)
60 # We need to call find_package even when we've already done the detection/setup
61 if(GMX_GPU OR GMX_GPU_AUTO)
62 if(NOT GMX_GPU AND NOT GMX_DETECT_GPU_AVAILABLE)
63 # Stay quiet when detection has occured and found no GPU.
64 # Noise is acceptable when there is a GPU or the user required one.
65 set(FIND_CUDA_QUIETLY QUIET)
67 # We support CUDA >=v3.2 on *nix, but <= v4.1 doesn't work with MSVC
69 find_package(CUDA 4.1 ${FIND_CUDA_QUIETLY})
71 find_package(CUDA 3.2 ${FIND_CUDA_QUIETLY})
75 # Depending on the current vale of GMX_GPU and GMX_GPU_AUTO:
76 # - OFF, FALSE: Will skip this detection/setup.
77 # - OFF, TRUE : Will keep GMX_GPU=OFF if no CUDA is detected, but will assemble
78 # a warning message which will be issued at the end of the
79 # configuration if GPU(s) were found in the build system.
80 # - ON , FALSE: The user requested GPU build and this requires CUDA, so we will
81 # fail if it is not available.
82 # - ON , TRUE : Can't happen (GMX_GPU=ON can only be user-set at this point)
83 if((GMX_GPU OR GMX_GPU_AUTO) AND NOT GMX_GPU_DETECTION_DONE)
84 if (EXISTS ${CUDA_TOOLKIT_ROOT_DIR})
85 set(CUDA_FOUND TRUE CACHE INTERNAL "Whether the CUDA toolkit was found" FORCE)
87 set(CUDA_FOUND FALSE CACHE INTERNAL "Whether the CUDA toolkit was found" FORCE)
90 # assemble warning/error message
91 if (GMX_DETECT_GPU_AVAILABLE)
93 ${GMX_DETECT_GPU_COUNT} NVIDIA GPU(s) found in the system")
96 if (NOT GMX_DETECT_GPU_INFO STREQUAL "")
98 foreach(gpu ${GMX_DETECT_GPU_INFO})
104 # TODO remove the second part of the message when we'll have compute
105 # capability information from the detection.
107 Compute capability information not available, consult the NVIDIA website:
108 https://developer.nvidia.com/cuda-gpus
112 set(CUDA_NOTFOUND_MESSAGE "
113 mdrun supports native GPU acceleration on NVIDIA hardware with compute
114 capability >=2.0 (Fermi or later). This requires the NVIDIA CUDA toolkit,
115 which was not found. Its location can be hinted by setting the
116 CUDA_TOOLKIT_ROOT_DIR CMake option (does not work as an environment variable).
117 The typical location would be /usr/local/cuda[-version].
118 Note that CPU or GPU acceleration can be selected at runtime!
125 # Disable GPU acceleration in auto mode
126 message(STATUS "No compatible CUDA toolkit found (v3.2+), disabling native GPU acceleration")
127 set_property(CACHE GMX_GPU PROPERTY VALUE OFF)
128 set(CUDA_NOTFOUND_AUTO ON)
130 # the user requested CUDA, but it wasn't found
131 message(FATAL_ERROR "${CUDA_NOTFOUND_MESSAGE}")
135 message(STATUS "Enabling native GPU acceleration")
136 set_property(CACHE GMX_GPU PROPERTY VALUE ON)
138 endif() # NOT CUDA_FOUND
140 # Annoyingly enough, FindCUDA leaves a few variables behind as non-advanced.
141 # We need to mark these advanced outside the conditional, otherwise, if the
142 # user turns GMX_GPU=OFF after a failed cmake pass, these variables will be
143 # left behind in the cache.
144 mark_as_advanced(CUDA_BUILD_CUBIN CUDA_BUILD_EMULATION CUDA_SDK_ROOT_DIR CUDA_VERBOSE_BUILD)
146 mark_as_advanced(CUDA_TOOLKIT_ROOT_DIR)
149 # Try to execute ${CUDA_NVCC_EXECUTABLE} --version and set the output
150 # (or an error string) in the argument variable.
151 # Note that semicolon is used as separator for nvcc.
154 # COMPILER_INFO - [output variable] string with compiler path, ID and
155 # some compiler-provided information
156 # COMPILER_FLAGS - [output variable] flags for the compiler
158 macro(get_cuda_compiler_info COMPILER_INFO COMPILER_FLAGS)
159 if(CUDA_NVCC_EXECUTABLE)
161 # Get the nvcc version string. This is multi-line, but since it is only 4 lines
162 # and might change in the future it is better to store than trying to parse out
163 # the version from the current format.
164 execute_process(COMMAND ${CUDA_NVCC_EXECUTABLE} --version
165 RESULT_VARIABLE _nvcc_version_res
166 OUTPUT_VARIABLE _nvcc_version_out
167 ERROR_VARIABLE _nvcc_version_err
168 OUTPUT_STRIP_TRAILING_WHITESPACE)
169 if (${_nvcc_version_res} EQUAL 0)
170 # Fix multi-line mess: Replace newline with ";" so we can use it in a define
171 string(REPLACE "\n" ";" _nvcc_info_singleline ${_nvcc_version_out})
172 SET(${COMPILER_INFO} "${CUDA_NVCC_EXECUTABLE} ${_nvcc_info_singleline}")
173 string(TOUPPER ${CMAKE_BUILD_TYPE} _build_type)
174 SET(_compiler_flags "${CUDA_NVCC_FLAGS_${_build_type}}")
175 if(CUDA_PROPAGATE_HOST_FLAGS)
176 string(REGEX REPLACE "[ ]+" ";" _cxx_flags_nospace "${BUILD_CXXFLAGS}")
178 SET(${COMPILER_FLAGS} "${CUDA_NVCC_FLAGS}${CUDA_NVCC_FLAGS_${_build_type}}; ${_cxx_flags_nospace}")
180 SET(${COMPILER_INFO} "N/A")
181 SET(${COMPILER_FLAGS} "N/A")
187 # set up nvcc options
188 include(gmxManageNvccConfig)
190 # Atomic operations used for polling wait for GPU
191 # (to avoid the cudaStreamSynchronize + ECC bug).
192 # ThreadMPI is now always included. Thus, we don't check for Atomics anymore here.
194 # no OpenMP is no good!
197 To use GPU acceleration efficiently, mdrun requires OpenMP multi-threading.
198 Without OpenMP a single CPU core can be used with a GPU which is not optimal.
199 Note that with MPI multiple processes can be forced to use a single GPU, but this
200 typically inefficient. Note that you need to set both C and C++ compilers that
201 support OpenMP (CC and CXX environment variables, respectively) when using GPUs.")