2 # This file is part of the GROMACS molecular simulation package.
4 # Copyright (c) 2012,2013,2014,2015,2016,2017,2018, 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 # Manage CUDA nvcc compilation configuration, try to be smart to ease the users'
36 # pain as much as possible:
37 # - use the CUDA_HOST_COMPILER if defined by the user, otherwise
38 # - check if nvcc works with CUDA_HOST_COMPILER and the generated nvcc and C++ flags
40 # - (advanced) variables set:
41 # * CUDA_HOST_COMPILER_OPTIONS - the full host-compiler related option list passed to nvcc
43 # Note that from CMake 2.8.10 FindCUDA defines CUDA_HOST_COMPILER internally,
44 # so we won't set it ourselves, but hope that the module does a good job.
46 # glibc 2.23 changed string.h in a way that breaks CUDA compilation in
47 # many projects, but which has a trivial workaround. It would be nicer
48 # to compile with nvcc and see that the workaround is necessary and
49 # effective, but it is unclear how to do that. Also, grepping in the
50 # glibc source shows that _FORCE_INLINES is only used in this string.h
51 # feature and performance of memcpy variants is unimportant for CUDA
52 # code in GROMACS. So this workaround is good enough to keep problems
53 # away from users installing GROMACS. See Redmine 1942.
54 function(work_around_glibc_2_23)
55 try_compile(IS_GLIBC_2_23_OR_HIGHER ${CMAKE_BINARY_DIR} ${CMAKE_SOURCE_DIR}/cmake/TestGlibcVersion.cpp)
56 if(IS_GLIBC_2_23_OR_HIGHER)
57 message(STATUS "Adding work-around for issue compiling CUDA code with glibc 2.23 string.h")
58 list(APPEND CUDA_HOST_COMPILER_OPTIONS "-D_FORCE_INLINES")
59 set(CUDA_HOST_COMPILER_OPTIONS ${CUDA_HOST_COMPILER_OPTIONS} PARENT_SCOPE)
63 gmx_check_if_changed(CUDA_HOST_COMPILER_CHANGED CUDA_HOST_COMPILER)
65 # set up host compiler and its options
66 if(CUDA_HOST_COMPILER_CHANGED)
67 set(CUDA_HOST_COMPILER_OPTIONS "")
69 if(APPLE AND CMAKE_C_COMPILER_ID MATCHES "GNU")
70 # Some versions of gcc-4.8 and gcc-4.9 produce errors (in particular on OS X)
71 # if we do not use -D__STRICT_ANSI__. It is harmless, so we might as well add it for all versions.
72 list(APPEND CUDA_HOST_COMPILER_OPTIONS "-D__STRICT_ANSI__")
75 work_around_glibc_2_23()
77 set(CUDA_HOST_COMPILER_OPTIONS "${CUDA_HOST_COMPILER_OPTIONS}"
78 CACHE STRING "Options for nvcc host compiler (do not edit!).")
80 mark_as_advanced(CUDA_HOST_COMPILER CUDA_HOST_COMPILER_OPTIONS)
83 # If any of these manual override variables for target CUDA GPU architectures
84 # or virtual architecture is set, parse the values and assemble the nvcc
85 # command line for these. Otherwise use our defaults.
86 # Note that the manual override variables require a semicolon separated
87 # architectures codes.
88 if (GMX_CUDA_TARGET_SM OR GMX_CUDA_TARGET_COMPUTE)
89 set(GMX_CUDA_NVCC_GENCODE_FLAGS)
90 set(_target_sm_list ${GMX_CUDA_TARGET_SM})
91 foreach(_target ${_target_sm_list})
92 list(APPEND GMX_CUDA_NVCC_GENCODE_FLAGS "-gencode;arch=compute_${_target},code=sm_${_target}")
94 set(_target_compute_list ${GMX_CUDA_TARGET_COMPUTE})
95 foreach(_target ${_target_compute_list})
96 list(APPEND GMX_CUDA_NVCC_GENCODE_FLAGS "-gencode;arch=compute_${_target},code=compute_${_target}")
99 # Set the CUDA GPU architectures to compile for:
100 # - with CUDA >=5.0 <6.5: CC <=3.5 is supported
101 # => compile sm_30, sm_35 SASS, and compute_35 PTX
102 # - with CUDA ==6.5: CC <=3.7 and 5.0 are supported
103 # => compile sm_30, sm_35, sm_37 sm_50, SASS, and compute_50 PTX
104 # - with CUDA >=7.0 CC 5.2 is supported (5.3, Tegra X1 we don't generate code for)
105 # => compile sm_30, sm_35, sm_37, sm_50, & sm_52 SASS, and compute_52 PTX
106 # - with CUDA >=8.0 CC 6.0-6.2 is supported (but we know nothing about CC 6.2, so we won't generate code or it)
107 # => compile sm_30, sm_35, sm_37, sm_50, sm_52, sm_60, sm_61 SASS, and compute_60 and compute_61 PTX
108 # - with CUDA >=9.0 CC 7.0 is supported and CC 2.0 is no longer supported
109 # => compile sm_30, sm_35, sm_37, sm_50, sm_52, sm_60, sm_61, sm_70 SASS, and compute_70 PTX
111 # Note that CUDA 6.5.19 second patch release supports cc 5.2 too, but
112 # CUDA_VERSION does not contain patch version and having PTX 5.0 JIT-ed is
113 # equally fast as compiling with sm_5.2 anyway.
115 # First add flags that trigger SASS (binary) code generation for physical arch
116 list (APPEND GMX_CUDA_NVCC_GENCODE_FLAGS "-gencode;arch=compute_30,code=sm_30")
117 list (APPEND GMX_CUDA_NVCC_GENCODE_FLAGS "-gencode;arch=compute_35,code=sm_35")
119 if(NOT CUDA_VERSION VERSION_LESS "6.5") # >= 6.5
120 list (APPEND GMX_CUDA_NVCC_GENCODE_FLAGS "-gencode;arch=compute_37,code=sm_37")
121 list (APPEND GMX_CUDA_NVCC_GENCODE_FLAGS "-gencode;arch=compute_50,code=sm_50")
123 if(NOT CUDA_VERSION VERSION_LESS "7.0") # >= 7.0
124 list (APPEND GMX_CUDA_NVCC_GENCODE_FLAGS "-gencode;arch=compute_52,code=sm_52")
126 if(NOT CUDA_VERSION VERSION_LESS "8.0") # >= 8.0
127 list (APPEND GMX_CUDA_NVCC_GENCODE_FLAGS "-gencode;arch=compute_60,code=sm_60")
128 list (APPEND GMX_CUDA_NVCC_GENCODE_FLAGS "-gencode;arch=compute_61,code=sm_61")
130 if(NOT CUDA_VERSION VERSION_LESS "9.0") # >= 9.0
131 list (APPEND GMX_CUDA_NVCC_GENCODE_FLAGS "-gencode;arch=compute_70,code=sm_70")
134 # Next add flags that trigger PTX code generation for the newest supported virtual arch
135 # that's useful to JIT to future architectures
136 if(CUDA_VERSION VERSION_LESS "6.5")
137 list (APPEND GMX_CUDA_NVCC_GENCODE_FLAGS "-gencode;arch=compute_35,code=compute_35")
138 elseif(CUDA_VERSION VERSION_LESS "7.0")
139 list (APPEND GMX_CUDA_NVCC_GENCODE_FLAGS "-gencode;arch=compute_50,code=compute_50")
140 elseif(CUDA_VERSION VERSION_LESS "8.0")
141 list (APPEND GMX_CUDA_NVCC_GENCODE_FLAGS "-gencode;arch=compute_52,code=compute_52")
142 elseif(CUDA_VERSION VERSION_LESS "9.0")
143 list (APPEND GMX_CUDA_NVCC_GENCODE_FLAGS "-gencode;arch=compute_60,code=compute_60")
144 list (APPEND GMX_CUDA_NVCC_GENCODE_FLAGS "-gencode;arch=compute_61,code=compute_61")
145 else() # version >= 9.0
146 list (APPEND GMX_CUDA_NVCC_GENCODE_FLAGS "-gencode;arch=compute_70,code=compute_70")
150 if (GMX_CUDA_TARGET_SM)
151 set_property(CACHE GMX_CUDA_TARGET_SM PROPERTY HELPSTRING "List of CUDA GPU architecture codes to compile for (without the sm_ prefix)")
152 set_property(CACHE GMX_CUDA_TARGET_SM PROPERTY TYPE STRING)
154 if (GMX_CUDA_TARGET_COMPUTE)
155 set_property(CACHE GMX_CUDA_TARGET_COMPUTE PROPERTY HELPSTRING "List of CUDA virtual architecture codes to compile for (without the compute_ prefix)")
156 set_property(CACHE GMX_CUDA_TARGET_COMPUTE PROPERTY TYPE STRING)
159 # assemble the CUDA flags
160 list(APPEND GMX_CUDA_NVCC_FLAGS "${GMX_CUDA_NVCC_GENCODE_FLAGS}")
161 list(APPEND GMX_CUDA_NVCC_FLAGS "-use_fast_math")
163 # assemble the CUDA host compiler flags
164 list(APPEND GMX_CUDA_NVCC_FLAGS "${CUDA_HOST_COMPILER_OPTIONS}")
166 string(TOUPPER "${CMAKE_BUILD_TYPE}" _build_type)
167 gmx_check_if_changed(_cuda_nvcc_executable_or_flags_changed CUDA_NVCC_EXECUTABLE CUDA_NVCC_FLAGS CUDA_NVCC_FLAGS_${_build_type})
169 if(_cuda_nvcc_executable_or_flags_changed OR CUDA_HOST_COMPILER_CHANGED OR NOT GMX_NVCC_WORKS)
170 message(STATUS "Check for working NVCC/C compiler combination")
171 execute_process(COMMAND ${CUDA_NVCC_EXECUTABLE} -ccbin ${CUDA_HOST_COMPILER} -c ${CUDA_NVCC_FLAGS} ${CUDA_NVCC_FLAGS_${_build_type}} ${CMAKE_SOURCE_DIR}/cmake/TestCUDA.cu
172 RESULT_VARIABLE _cuda_test_res
173 OUTPUT_VARIABLE _cuda_test_out
174 ERROR_VARIABLE _cuda_test_err
175 OUTPUT_STRIP_TRAILING_WHITESPACE)
177 if(${_cuda_test_res})
178 message(${_cuda_test_err})
179 message(STATUS "Check for working NVCC/C compiler combination - broken")
180 if(${_cuda_test_err} MATCHES "nsupported")
181 message(FATAL_ERROR "NVCC/C compiler combination does not seem to be supported. CUDA frequently does not support the latest versions of the host compiler, so you might want to try an earlier C/C++ compiler version and make sure your CUDA compiler and driver are as recent as possible.")
183 message(FATAL_ERROR "CUDA compiler does not seem to be functional.")
185 elseif(NOT GMX_CUDA_TEST_COMPILER_QUIETLY)
186 message(STATUS "Check for working NVCC/C compiler combination - works")
187 set(GMX_NVCC_WORKS TRUE CACHE INTERNAL "Nvcc can compile a trivial test program")
189 endif() # GMX_CHECK_NVCC
192 # The flags are set as local variables which shadow the cache variables. The cache variables
193 # (can be set by the user) are appended. This is done in a macro to set the flags when all
194 # host compiler flags are already set.
195 macro(GMX_SET_CUDA_NVCC_FLAGS)
196 set(CUDA_NVCC_FLAGS "${GMX_CUDA_NVCC_FLAGS};${CUDA_NVCC_FLAGS}")