1 # - Tools for building CUDA C files: libraries and build dependencies.
2 # This script locates the NVIDIA CUDA C tools. It should work on linux, windows,
3 # and mac and should be reasonably up to date with CUDA C releases.
5 # This script makes use of the standard find_package arguments of <VERSION>,
6 # REQUIRED and QUIET. CUDA_FOUND will report if an acceptable version of CUDA
9 # The script will prompt the user to specify CUDA_TOOLKIT_ROOT_DIR if the prefix
10 # cannot be determined by the location of nvcc in the system path and REQUIRED
11 # is specified to find_package(). To use a different installed version of the
12 # toolkit set the environment variable CUDA_BIN_PATH before running cmake
13 # (e.g. CUDA_BIN_PATH=/usr/local/cuda1.0 instead of the default /usr/local/cuda)
14 # or set CUDA_TOOLKIT_ROOT_DIR after configuring. If you change the value of
15 # CUDA_TOOLKIT_ROOT_DIR, various components that depend on the path will be
18 # It might be necessary to set CUDA_TOOLKIT_ROOT_DIR manually on certain
19 # platforms, or to use a cuda runtime not installed in the default location. In
20 # newer versions of the toolkit the cuda library is included with the graphics
21 # driver- be sure that the driver version matches what is needed by the cuda
24 # The following variables affect the behavior of the macros in the script (in
25 # alphebetical order). Note that any of these flags can be changed multiple
26 # times in the same directory before calling CUDA_ADD_EXECUTABLE,
27 # CUDA_ADD_LIBRARY, CUDA_COMPILE, CUDA_COMPILE_PTX or CUDA_WRAP_SRCS.
29 # CUDA_64_BIT_DEVICE_CODE (Default matches host bit size)
30 # -- Set to ON to compile for 64 bit device code, OFF for 32 bit device code.
31 # Note that making this different from the host code when generating object
32 # or C files from CUDA code just won't work, because size_t gets defined by
33 # nvcc in the generated source. If you compile to PTX and then load the
34 # file yourself, you can mix bit sizes between device and host.
36 # CUDA_ATTACH_VS_BUILD_RULE_TO_CUDA_FILE (Default ON)
37 # -- Set to ON if you want the custom build rule to be attached to the source
38 # file in Visual Studio. Turn OFF if you add the same cuda file to multiple
41 # This allows the user to build the target from the CUDA file; however, bad
42 # things can happen if the CUDA source file is added to multiple targets.
43 # When performing parallel builds it is possible for the custom build
44 # command to be run more than once and in parallel causing cryptic build
45 # errors. VS runs the rules for every source file in the target, and a
46 # source can have only one rule no matter how many projects it is added to.
47 # When the rule is run from multiple targets race conditions can occur on
48 # the generated file. Eventually everything will get built, but if the user
49 # is unaware of this behavior, there may be confusion. It would be nice if
50 # this script could detect the reuse of source files across multiple targets
51 # and turn the option off for the user, but no good solution could be found.
53 # CUDA_BUILD_CUBIN (Default OFF)
54 # -- Set to ON to enable and extra compilation pass with the -cubin option in
55 # Device mode. The output is parsed and register, shared memory usage is
56 # printed during build.
58 # CUDA_BUILD_EMULATION (Default OFF for device mode)
59 # -- Set to ON for Emulation mode. -D_DEVICEEMU is defined for CUDA C files
60 # when CUDA_BUILD_EMULATION is TRUE.
62 # CUDA_GENERATED_OUTPUT_DIR (Default CMAKE_CURRENT_BINARY_DIR)
63 # -- Set to the path you wish to have the generated files placed. If it is
64 # blank output files will be placed in CMAKE_CURRENT_BINARY_DIR.
65 # Intermediate files will always be placed in
66 # CMAKE_CURRENT_BINARY_DIR/CMakeFiles.
68 # CUDA_HOST_COMPILATION_CPP (Default ON)
69 # -- Set to OFF for C compilation of host code.
72 # CUDA_NVCC_FLAGS_<CONFIG>
73 # -- Additional NVCC command line arguments. NOTE: multiple arguments must be
74 # semi-colon delimited (e.g. --compiler-options;-Wall)
76 # CUDA_PROPAGATE_HOST_FLAGS (Default ON)
77 # -- Set to ON to propagate CMAKE_{C,CXX}_FLAGS and their configuration
78 # dependent counterparts (e.g. CMAKE_C_FLAGS_DEBUG) automatically to the
79 # host compiler through nvcc's -Xcompiler flag. This helps make the
80 # generated host code match the rest of the system better. Sometimes
81 # certain flags give nvcc problems, and this will help you turn the flag
82 # propagation off. This does not affect the flags supplied directly to nvcc
83 # via CUDA_NVCC_FLAGS or through the OPTION flags specified through
84 # CUDA_ADD_LIBRARY, CUDA_ADD_EXECUTABLE, or CUDA_WRAP_SRCS. Flags used for
85 # shared library compilation are not affected by this flag.
87 # CUDA_VERBOSE_BUILD (Default OFF)
88 # -- Set to ON to see all the commands used when building the CUDA file. When
89 # using a Makefile generator the value defaults to VERBOSE (run make
90 # VERBOSE=1 to see output), although setting CUDA_VERBOSE_BUILD to ON will
91 # always print the output.
93 # The script creates the following macros (in alphebetical order):
95 # CUDA_ADD_CUFFT_TO_TARGET( cuda_target )
96 # -- Adds the cufft library to the target (can be any target). Handles whether
97 # you are in emulation mode or not.
99 # CUDA_ADD_CUBLAS_TO_TARGET( cuda_target )
100 # -- Adds the cublas library to the target (can be any target). Handles
101 # whether you are in emulation mode or not.
103 # CUDA_ADD_EXECUTABLE( cuda_target file0 file1 ...
104 # [WIN32] [MACOSX_BUNDLE] [EXCLUDE_FROM_ALL] [OPTIONS ...] )
105 # -- Creates an executable "cuda_target" which is made up of the files
106 # specified. All of the non CUDA C files are compiled using the standard
107 # build rules specified by CMAKE and the cuda files are compiled to object
108 # files using nvcc and the host compiler. In addition CUDA_INCLUDE_DIRS is
109 # added automatically to include_directories(). Some standard CMake target
110 # calls can be used on the target after calling this macro
111 # (e.g. set_target_properties and target_link_libraries), but setting
112 # properties that adjust compilation flags will not affect code compiled by
113 # nvcc. Such flags should be modified before calling CUDA_ADD_EXECUTABLE,
114 # CUDA_ADD_LIBRARY or CUDA_WRAP_SRCS.
116 # CUDA_ADD_LIBRARY( cuda_target file0 file1 ...
117 # [STATIC | SHARED | MODULE] [EXCLUDE_FROM_ALL] [OPTIONS ...] )
118 # -- Same as CUDA_ADD_EXECUTABLE except that a library is created.
120 # CUDA_BUILD_CLEAN_TARGET()
121 # -- Creates a convience target that deletes all the dependency files
122 # generated. You should make clean after running this target to ensure the
123 # dependency files get regenerated.
125 # CUDA_COMPILE( generated_files file0 file1 ... [STATIC | SHARED | MODULE]
127 # -- Returns a list of generated files from the input source files to be used
128 # with ADD_LIBRARY or ADD_EXECUTABLE.
130 # CUDA_COMPILE_PTX( generated_files file0 file1 ... [OPTIONS ...] )
131 # -- Returns a list of PTX files generated from the input source files.
133 # CUDA_INCLUDE_DIRECTORIES( path0 path1 ... )
134 # -- Sets the directories that should be passed to nvcc
135 # (e.g. nvcc -Ipath0 -Ipath1 ... ). These paths usually contain other .cu
138 # CUDA_WRAP_SRCS ( cuda_target format generated_files file0 file1 ...
139 # [STATIC | SHARED | MODULE] [OPTIONS ...] )
140 # -- This is where all the magic happens. CUDA_ADD_EXECUTABLE,
141 # CUDA_ADD_LIBRARY, CUDA_COMPILE, and CUDA_COMPILE_PTX all call this
142 # function under the hood.
144 # Given the list of files (file0 file1 ... fileN) this macro generates
145 # custom commands that generate either PTX or linkable objects (use "PTX" or
146 # "OBJ" for the format argument to switch). Files that don't end with .cu
147 # or have the HEADER_FILE_ONLY property are ignored.
149 # The arguments passed in after OPTIONS are extra command line options to
150 # give to nvcc. You can also specify per configuration options by
151 # specifying the name of the configuration followed by the options. General
152 # options must preceed configuration specific options. Not all
153 # configurations need to be specified, only the ones provided will be used.
155 # OPTIONS -DFLAG=2 "-DFLAG_OTHER=space in flag"
157 # RELEASE --use_fast_math
158 # RELWITHDEBINFO --use_fast_math;-g
159 # MINSIZEREL --use_fast_math
161 # For certain configurations (namely VS generating object files with
162 # CUDA_ATTACH_VS_BUILD_RULE_TO_CUDA_FILE set to ON), no generated file will
163 # be produced for the given cuda file. This is because when you add the
164 # cuda file to Visual Studio it knows that this file produces an object file
165 # and will link in the resulting object file automatically.
167 # This script will also generate a separate cmake script that is used at
168 # build time to invoke nvcc. This is for serveral reasons.
170 # 1. nvcc can return negative numbers as return values which confuses
171 # Visual Studio into thinking that the command succeeded. The script now
172 # checks the error codes and produces errors when there was a problem.
174 # 2. nvcc has been known to not delete incomplete results when it
175 # encounters problems. This confuses build systems into thinking the
176 # target was generated when in fact an unusable file exists. The script
177 # now deletes the output files if there was an error.
179 # 3. By putting all the options that affect the build into a file and then
180 # make the build rule dependent on the file, the output files will be
181 # regenerated when the options change.
183 # This script also looks at optional arguments STATIC, SHARED, or MODULE to
184 # determine when to target the object compilation for a shared library.
185 # BUILD_SHARED_LIBS is ignored in CUDA_WRAP_SRCS, but it is respected in
186 # CUDA_ADD_LIBRARY. On some systems special flags are added for building
187 # objects intended for shared libraries. A preprocessor macro,
188 # <target_name>_EXPORTS is defined when a shared library compilation is
191 # Flags passed into add_definitions with -D or /D are passed along to nvcc.
193 # The script defines the following variables:
195 # CUDA_VERSION_MAJOR -- The major version of cuda as reported by nvcc.
196 # CUDA_VERSION_MINOR -- The minor version.
198 # CUDA_VERSION_STRING -- CUDA_VERSION_MAJOR.CUDA_VERSION_MINOR
200 # CUDA_TOOLKIT_ROOT_DIR -- Path to the CUDA Toolkit (defined if not set).
201 # CUDA_SDK_ROOT_DIR -- Path to the CUDA SDK. Use this to find files in the
202 # SDK. This script will not directly support finding
203 # specific libraries or headers, as that isn't
204 # supported by NVIDIA. If you want to change
205 # libraries when the path changes see the
206 # FindCUDA.cmake script for an example of how to clear
207 # these variables. There are also examples of how to
208 # use the CUDA_SDK_ROOT_DIR to locate headers or
209 # libraries, if you so choose (at your own risk).
210 # CUDA_INCLUDE_DIRS -- Include directory for cuda headers. Added automatically
211 # for CUDA_ADD_EXECUTABLE and CUDA_ADD_LIBRARY.
212 # CUDA_LIBRARIES -- Cuda RT library.
213 # CUDA_CUFFT_LIBRARIES -- Device or emulation library for the Cuda FFT
214 # implementation (alternative to:
215 # CUDA_ADD_CUFFT_TO_TARGET macro)
216 # CUDA_CUBLAS_LIBRARIES -- Device or emulation library for the Cuda BLAS
217 # implementation (alterative to:
218 # CUDA_ADD_CUBLAS_TO_TARGET macro).
221 # James Bigler, NVIDIA Corp (nvidia.com - jbigler)
222 # Abe Stephens, SCI Institute -- http://www.sci.utah.edu/~abe/FindCuda.html
224 # Copyright (c) 2008 - 2009 NVIDIA Corporation. All rights reserved.
226 # Copyright (c) 2007-2009
227 # Scientific Computing and Imaging Institute, University of Utah
229 # This code is licensed under the MIT License. See the FindCUDA.cmake script
230 # for the text of the license.
234 # License for the specific language governing rights and limitations under
235 # Permission is hereby granted, free of charge, to any person obtaining a
236 # copy of this software and associated documentation files (the "Software"),
237 # to deal in the Software without restriction, including without limitation
238 # the rights to use, copy, modify, merge, publish, distribute, sublicense,
239 # and/or sell copies of the Software, and to permit persons to whom the
240 # Software is furnished to do so, subject to the following conditions:
242 # The above copyright notice and this permission notice shall be included
243 # in all copies or substantial portions of the Software.
245 # THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
246 # OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
247 # FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
248 # THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
249 # LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
250 # FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
251 # DEALINGS IN THE SOFTWARE.
253 ###############################################################################
257 # We need to have at least this version to support the VERSION_LESS argument to 'if' (2.6.2) and unset (2.6.3)
259 cmake_minimum_required(VERSION 2.6.3)
262 # This macro helps us find the location of helper files we will need the full path to
263 macro(CUDA_FIND_HELPER_FILE _name _extension)
264 set(_full_name "${_name}.${_extension}")
265 # CMAKE_CURRENT_LIST_FILE contains the full path to the file currently being
266 # processed. Using this variable, we can pull out the current path, and
267 # provide a way to get access to the other files we need local to here.
268 get_filename_component(CMAKE_CURRENT_LIST_DIR "${CMAKE_CURRENT_LIST_FILE}" PATH)
269 find_file(CUDA_${_name} ${_full_name} PATHS ${CMAKE_CURRENT_LIST_DIR}/FindCUDA NO_DEFAULT_PATH)
270 if(NOT CUDA_${_name})
271 set(error_message "${_full_name} not found in CMAKE_MODULE_PATH")
272 if(CUDA_FIND_REQUIRED)
273 message(FATAL_ERROR "${error_message}")
274 else(CUDA_FIND_REQUIRED)
275 if(NOT CUDA_FIND_QUIETLY)
276 message(STATUS "${error_message}")
277 endif(NOT CUDA_FIND_QUIETLY)
278 endif(CUDA_FIND_REQUIRED)
279 endif(NOT CUDA_${_name})
280 # Set this variable as internal, so the user isn't bugged with it.
281 set(CUDA_${_name} ${CUDA_${_name}} CACHE INTERNAL "Location of ${_full_name}" FORCE)
282 endmacro(CUDA_FIND_HELPER_FILE)
284 #####################################################################
285 ## CUDA_INCLUDE_NVCC_DEPENDENCIES
288 # So we want to try and include the dependency file if it exists. If
289 # it doesn't exist then we need to create an empty one, so we can
292 # If it does exist, then we need to check to see if all the files it
293 # depends on exist. If they don't then we should clear the dependency
294 # file and regenerate it later. This covers the case where a header
295 # file has disappeared or moved.
297 macro(CUDA_INCLUDE_NVCC_DEPENDENCIES dependency_file)
298 set(CUDA_NVCC_DEPEND)
299 set(CUDA_NVCC_DEPEND_REGENERATE FALSE)
302 # Include the dependency file. Create it first if it doesn't exist . The
303 # INCLUDE puts a dependency that will force CMake to rerun and bring in the
304 # new info when it changes. DO NOT REMOVE THIS (as I did and spent a few
305 # hours figuring out why it didn't work.
306 if(NOT EXISTS ${dependency_file})
307 file(WRITE ${dependency_file} "#FindCUDA.cmake generated file. Do not edit.\n")
309 # Always include this file to force CMake to run again next
310 # invocation and rebuild the dependencies.
311 #message("including dependency_file = ${dependency_file}")
312 include(${dependency_file})
314 # Now we need to verify the existence of all the included files
315 # here. If they aren't there we need to just blank this variable and
316 # make the file regenerate again.
317 # if(DEFINED CUDA_NVCC_DEPEND)
318 # message("CUDA_NVCC_DEPEND set")
320 # message("CUDA_NVCC_DEPEND NOT set")
323 #message("CUDA_NVCC_DEPEND true")
324 foreach(f ${CUDA_NVCC_DEPEND})
325 #message("searching for ${f}")
327 #message("file ${f} not found")
328 set(CUDA_NVCC_DEPEND_REGENERATE TRUE)
331 else(CUDA_NVCC_DEPEND)
332 #message("CUDA_NVCC_DEPEND false")
333 # No dependencies, so regenerate the file.
334 set(CUDA_NVCC_DEPEND_REGENERATE TRUE)
335 endif(CUDA_NVCC_DEPEND)
337 #message("CUDA_NVCC_DEPEND_REGENERATE = ${CUDA_NVCC_DEPEND_REGENERATE}")
338 # No incoming dependencies, so we need to generate them. Make the
339 # output depend on the dependency file itself, which should cause the
341 if(CUDA_NVCC_DEPEND_REGENERATE)
342 file(WRITE ${dependency_file} "#FindCUDA.cmake generated file. Do not edit.\n")
343 endif(CUDA_NVCC_DEPEND_REGENERATE)
345 endmacro(CUDA_INCLUDE_NVCC_DEPENDENCIES)
347 ###############################################################################
348 ###############################################################################
349 # Setup variables' defaults
350 ###############################################################################
351 ###############################################################################
353 # Allow the user to specify if the device code is supposed to be 32 or 64 bit.
354 if(CMAKE_SIZEOF_VOID_P EQUAL 8)
355 set(CUDA_64_BIT_DEVICE_CODE_DEFAULT ON)
357 set(CUDA_64_BIT_DEVICE_CODE_DEFAULT OFF)
359 option(CUDA_64_BIT_DEVICE_CODE "Compile device code in 64 bit mode" ${CUDA_64_BIT_DEVICE_CODE_DEFAULT})
361 # Attach the build rule to the source file in VS. This option
362 option(CUDA_ATTACH_VS_BUILD_RULE_TO_CUDA_FILE "Attach the build rule to the CUDA source file. Enable only when the CUDA source file is added to at most one target." ON)
364 # Prints out extra information about the cuda file during compilation
365 option(CUDA_BUILD_CUBIN "Generate and parse .cubin files in Device mode." OFF)
367 # Set whether we are using emulation or device mode.
368 option(CUDA_BUILD_EMULATION "Build in Emulation mode" OFF)
370 # Where to put the generated output.
371 set(CUDA_GENERATED_OUTPUT_DIR "" CACHE PATH "Directory to put all the output files. If blank it will default to the CMAKE_CURRENT_BINARY_DIR")
373 # Parse HOST_COMPILATION mode.
374 option(CUDA_HOST_COMPILATION_CPP "Generated file extension" ON)
376 # Extra user settable flags
377 set(CUDA_NVCC_FLAGS "" CACHE STRING "Semi-colon delimit multiple arguments.")
379 # Propagate the host flags to the host compiler via -Xcompiler
380 option(CUDA_PROPAGATE_HOST_FLAGS "Propage C/CXX_FLAGS and friends to the host compiler via -Xcompile" ON)
382 # Specifies whether the commands used when compiling the .cu file will be printed out.
383 option(CUDA_VERBOSE_BUILD "Print out the commands run while compiling the CUDA source file. With the Makefile generator this defaults to VERBOSE variable specified on the command line, but can be forced on with this option." OFF)
386 CUDA_64_BIT_DEVICE_CODE
387 CUDA_ATTACH_VS_BUILD_RULE_TO_CUDA_FILE
388 CUDA_GENERATED_OUTPUT_DIR
389 CUDA_HOST_COMPILATION_CPP
391 CUDA_PROPAGATE_HOST_FLAGS
394 # Makefile and similar generators don't define CMAKE_CONFIGURATION_TYPES, so we
395 # need to add another entry for the CMAKE_BUILD_TYPE. We also need to add the
396 # standerd set of 4 build types (Debug, MinSizeRel, Release, and RelWithDebInfo)
397 # for completeness. We need run this loop in order to accomodate the addition
398 # of extra configuration types. Duplicate entries will be removed by
400 set(CUDA_configuration_types ${CMAKE_CONFIGURATION_TYPES} ${CMAKE_BUILD_TYPE} Debug MinSizeRel Release RelWithDebInfo)
401 list(REMOVE_DUPLICATES CUDA_configuration_types)
402 foreach(config ${CUDA_configuration_types})
403 string(TOUPPER ${config} config_upper)
404 set(CUDA_NVCC_FLAGS_${config_upper} "" CACHE STRING "Semi-colon delimit multiple arguments.")
405 mark_as_advanced(CUDA_NVCC_FLAGS_${config_upper})
408 ###############################################################################
409 ###############################################################################
410 # Locate CUDA, Set Build Type, etc.
411 ###############################################################################
412 ###############################################################################
414 # Check to see if the CUDA_TOOLKIT_ROOT_DIR and CUDA_SDK_ROOT_DIR have changed,
415 # if they have then clear the cache variables, so that will be detected again.
416 if(NOT "${CUDA_TOOLKIT_ROOT_DIR}" STREQUAL "${CUDA_TOOLKIT_ROOT_DIR_INTERNAL}")
417 unset(CUDA_NVCC_EXECUTABLE CACHE)
418 unset(CUDA_VERSION CACHE)
419 unset(CUDA_TOOLKIT_INCLUDE CACHE)
420 unset(CUDA_CUDART_LIBRARY CACHE)
421 if(CUDA_VERSION VERSION_EQUAL "3.0")
422 # This only existed in the 3.0 version of the CUDA toolkit
423 unset(CUDA_CUDARTEMU_LIBRARY CACHE)
425 unset(CUDA_CUDA_LIBRARY CACHE)
426 unset(CUDA_cublas_LIBRARY CACHE)
427 unset(CUDA_cublasemu_LIBRARY CACHE)
428 unset(CUDA_cufft_LIBRARY CACHE)
429 unset(CUDA_cufftemu_LIBRARY CACHE)
432 if(NOT "${CUDA_SDK_ROOT_DIR}" STREQUAL "${CUDA_SDK_ROOT_DIR_INTERNAL}")
433 # No specific variables to catch. Use this kind of code before calling
434 # find_package(CUDA) to clean up any variables that may depend on this path.
436 # unset(MY_SPECIAL_CUDA_SDK_INCLUDE_DIR CACHE)
437 # unset(MY_SPECIAL_CUDA_SDK_LIBRARY CACHE)
440 # Search for the cuda distribution.
441 if(NOT CUDA_TOOLKIT_ROOT_DIR)
443 # Search in the CUDA_BIN_PATH first.
444 find_path(CUDA_TOOLKIT_ROOT_DIR
446 PATHS ENV CUDA_BIN_PATH
447 DOC "Toolkit location."
450 # Now search default paths
451 find_path(CUDA_TOOLKIT_ROOT_DIR
455 DOC "Toolkit location."
458 if (CUDA_TOOLKIT_ROOT_DIR)
459 string(REGEX REPLACE "[/\\\\]?bin[64]*[/\\\\]?$" "" CUDA_TOOLKIT_ROOT_DIR ${CUDA_TOOLKIT_ROOT_DIR})
460 # We need to force this back into the cache.
461 set(CUDA_TOOLKIT_ROOT_DIR ${CUDA_TOOLKIT_ROOT_DIR} CACHE PATH "Toolkit location." FORCE)
462 endif(CUDA_TOOLKIT_ROOT_DIR)
463 if (NOT EXISTS ${CUDA_TOOLKIT_ROOT_DIR})
464 if(CUDA_FIND_REQUIRED)
465 message(FATAL_ERROR "Specify CUDA_TOOLKIT_ROOT_DIR")
466 elseif(NOT CUDA_FIND_QUIETLY)
467 message("CUDA_TOOLKIT_ROOT_DIR not found or specified")
469 endif (NOT EXISTS ${CUDA_TOOLKIT_ROOT_DIR})
470 endif (NOT CUDA_TOOLKIT_ROOT_DIR)
472 # CUDA_NVCC_EXECUTABLE
473 find_program(CUDA_NVCC_EXECUTABLE
475 PATHS "${CUDA_TOOLKIT_ROOT_DIR}/bin"
476 "${CUDA_TOOLKIT_ROOT_DIR}/bin64"
480 # Search default search paths, after we search our own set of paths.
481 find_program(CUDA_NVCC_EXECUTABLE nvcc)
482 mark_as_advanced(CUDA_NVCC_EXECUTABLE)
484 if(CUDA_NVCC_EXECUTABLE AND NOT CUDA_VERSION)
485 # Compute the version.
486 execute_process (COMMAND ${CUDA_NVCC_EXECUTABLE} "--version" OUTPUT_VARIABLE NVCC_OUT)
487 string(REGEX REPLACE ".*release ([0-9]+)\\.([0-9]+).*" "\\1" CUDA_VERSION_MAJOR ${NVCC_OUT})
488 string(REGEX REPLACE ".*release ([0-9]+)\\.([0-9]+).*" "\\2" CUDA_VERSION_MINOR ${NVCC_OUT})
489 set(CUDA_VERSION "${CUDA_VERSION_MAJOR}.${CUDA_VERSION_MINOR}" CACHE STRING "Version of CUDA as computed from nvcc.")
490 mark_as_advanced(CUDA_VERSION)
492 # Need to set these based off of the cached value
493 string(REGEX REPLACE "([0-9]+)\\.([0-9]+).*" "\\1" CUDA_VERSION_MAJOR "${CUDA_VERSION}")
494 string(REGEX REPLACE "([0-9]+)\\.([0-9]+).*" "\\2" CUDA_VERSION_MINOR "${CUDA_VERSION}")
497 # Always set this convenience variable
498 set(CUDA_VERSION_STRING "${CUDA_VERSION}")
500 # Here we need to determine if the version we found is acceptable. We will
501 # assume that is unless CUDA_FIND_VERSION_EXACT or CUDA_FIND_VERSION is
502 # specified. The presence of either of these options checks the version
503 # string and signals if the version is acceptable or not.
504 set(_cuda_version_acceptable TRUE)
506 if(CUDA_FIND_VERSION_EXACT AND NOT CUDA_VERSION VERSION_EQUAL CUDA_FIND_VERSION)
507 set(_cuda_version_acceptable FALSE)
510 if(CUDA_FIND_VERSION AND CUDA_VERSION VERSION_LESS CUDA_FIND_VERSION)
511 set(_cuda_version_acceptable FALSE)
514 if(NOT _cuda_version_acceptable)
515 set(_cuda_error_message "Requested CUDA version ${CUDA_FIND_VERSION}, but found unacceptable version ${CUDA_VERSION}")
516 if(CUDA_FIND_REQUIRED)
517 message("${_cuda_error_message}")
518 elseif(NOT CUDA_FIND_QUIETLY)
519 message("${_cuda_error_message}")
523 # CUDA_TOOLKIT_INCLUDE
524 find_path(CUDA_TOOLKIT_INCLUDE
525 device_functions.h # Header included in toolkit
526 PATHS "${CUDA_TOOLKIT_ROOT_DIR}/include"
530 # Search default search paths, after we search our own set of paths.
531 find_path(CUDA_TOOLKIT_INCLUDE device_functions.h)
532 mark_as_advanced(CUDA_TOOLKIT_INCLUDE)
534 # Set the user list of include dir to nothing to initialize it.
535 set (CUDA_NVCC_INCLUDE_ARGS_USER "")
536 set (CUDA_INCLUDE_DIRS ${CUDA_TOOLKIT_INCLUDE})
538 macro(FIND_LIBRARY_LOCAL_FIRST _var _names _doc)
539 if(CMAKE_SIZEOF_VOID_P EQUAL 8)
540 set(_cuda_64bit_lib_dir "${CUDA_TOOLKIT_ROOT_DIR}/lib64")
544 PATHS ${_cuda_64bit_lib_dir}
545 "${CUDA_TOOLKIT_ROOT_DIR}/lib"
550 # Search default search paths, after we search our own set of paths.
551 find_library(${_var} NAMES ${_names} DOC ${_doc})
555 find_library_local_first(CUDA_CUDART_LIBRARY cudart "\"cudart\" library")
556 if(CUDA_VERSION VERSION_EQUAL "3.0")
557 # The cudartemu library only existed for the 3.0 version of CUDA.
558 find_library_local_first(CUDA_CUDARTEMU_LIBRARY cudartemu "\"cudartemu\" library")
560 CUDA_CUDARTEMU_LIBRARY
563 # If we are using emulation mode and we found the cudartemu library then use
564 # that one instead of cudart.
565 if(CUDA_BUILD_EMULATION AND CUDA_CUDARTEMU_LIBRARY)
566 set(CUDA_LIBRARIES ${CUDA_CUDARTEMU_LIBRARY})
568 set(CUDA_LIBRARIES ${CUDA_CUDART_LIBRARY})
571 # We need to add the path to cudart to the linker using rpath, since the
572 # library name for the cuda libraries is prepended with @rpath.
573 if(CUDA_BUILD_EMULATION AND CUDA_CUDARTEMU_LIBRARY)
574 get_filename_component(_cuda_path_to_cudart "${CUDA_CUDARTEMU_LIBRARY}" PATH)
576 get_filename_component(_cuda_path_to_cudart "${CUDA_CUDART_LIBRARY}" PATH)
578 if(_cuda_path_to_cudart)
579 list(APPEND CUDA_LIBRARIES -Wl,-rpath "-Wl,${_cuda_path_to_cudart}")
583 # 1.1 toolkit on linux doesn't appear to have a separate library on
585 find_library_local_first(CUDA_CUDA_LIBRARY cuda "\"cuda\" library (older versions only).")
587 # Add cuda library to the link line only if it is found.
588 if (CUDA_CUDA_LIBRARY)
589 set(CUDA_LIBRARIES ${CUDA_LIBRARIES} ${CUDA_CUDA_LIBRARY})
590 endif(CUDA_CUDA_LIBRARY)
597 #######################
598 # Look for some of the toolkit helper libraries
599 macro(FIND_CUDA_HELPER_LIBS _name)
600 find_library_local_first(CUDA_${_name}_LIBRARY ${_name} "\"${_name}\" library")
601 mark_as_advanced(CUDA_${_name}_LIBRARY)
602 endmacro(FIND_CUDA_HELPER_LIBS)
604 # Search for cufft and cublas libraries.
605 find_cuda_helper_libs(cufftemu)
606 find_cuda_helper_libs(cublasemu)
607 find_cuda_helper_libs(cufft)
608 find_cuda_helper_libs(cublas)
610 if (CUDA_BUILD_EMULATION)
611 set(CUDA_CUFFT_LIBRARIES ${CUDA_cufftemu_LIBRARY})
612 set(CUDA_CUBLAS_LIBRARIES ${CUDA_cublasemu_LIBRARY})
614 set(CUDA_CUFFT_LIBRARIES ${CUDA_cufft_LIBRARY})
615 set(CUDA_CUBLAS_LIBRARIES ${CUDA_cublas_LIBRARY})
618 ########################
619 # Look for the SDK stuff
620 find_path(CUDA_SDK_ROOT_DIR common/inc/cutil.h
621 "$ENV{NVSDKCUDA_ROOT}"
622 "[HKEY_LOCAL_MACHINE\\SOFTWARE\\NVIDIA Corporation\\Installed Products\\NVIDIA SDK 10\\Compute;InstallDir]"
623 "/Developer/GPU\ Computing/C"
626 # Keep the CUDA_SDK_ROOT_DIR first in order to be able to override the
627 # environment variables.
628 set(CUDA_SDK_SEARCH_PATH
629 "${CUDA_SDK_ROOT_DIR}"
630 "${CUDA_TOOLKIT_ROOT_DIR}/local/NVSDK0.2"
631 "${CUDA_TOOLKIT_ROOT_DIR}/NVSDK0.2"
632 "${CUDA_TOOLKIT_ROOT_DIR}/NV_CUDA_SDK"
633 "$ENV{HOME}/NVIDIA_CUDA_SDK"
634 "$ENV{HOME}/NVIDIA_CUDA_SDK_MACOSX"
638 # Example of how to find an include file from the CUDA_SDK_ROOT_DIR
640 # find_path(CUDA_CUT_INCLUDE_DIR
642 # PATHS ${CUDA_SDK_SEARCH_PATH}
643 # PATH_SUFFIXES "common/inc"
644 # DOC "Location of cutil.h"
647 # # Now search system paths
648 # find_path(CUDA_CUT_INCLUDE_DIR cutil.h DOC "Location of cutil.h")
650 # mark_as_advanced(CUDA_CUT_INCLUDE_DIR)
653 # Example of how to find a library in the CUDA_SDK_ROOT_DIR
655 # # cutil library is called cutil64 for 64 bit builds on windows. We don't want
656 # # to get these confused, so we are setting the name based on the word size of
659 # if(CMAKE_SIZEOF_VOID_P EQUAL 8)
660 # set(cuda_cutil_name cutil64)
661 # else(CMAKE_SIZEOF_VOID_P EQUAL 8)
662 # set(cuda_cutil_name cutil32)
663 # endif(CMAKE_SIZEOF_VOID_P EQUAL 8)
665 # find_library(CUDA_CUT_LIBRARY
666 # NAMES cutil ${cuda_cutil_name}
667 # PATHS ${CUDA_SDK_SEARCH_PATH}
668 # # The new version of the sdk shows up in common/lib, but the old one is in lib
669 # PATH_SUFFIXES "common/lib" "lib"
670 # DOC "Location of cutil library"
673 # # Now search system paths
674 # find_library(CUDA_CUT_LIBRARY NAMES cutil ${cuda_cutil_name} DOC "Location of cutil library")
675 # mark_as_advanced(CUDA_CUT_LIBRARY)
676 # set(CUDA_CUT_LIBRARIES ${CUDA_CUT_LIBRARY})
680 #############################
681 # Check for required components
684 set(CUDA_TOOLKIT_ROOT_DIR_INTERNAL "${CUDA_TOOLKIT_ROOT_DIR}" CACHE INTERNAL
685 "This is the value of the last time CUDA_TOOLKIT_ROOT_DIR was set successfully." FORCE)
686 set(CUDA_SDK_ROOT_DIR_INTERNAL "${CUDA_SDK_ROOT_DIR}" CACHE INTERNAL
687 "This is the value of the last time CUDA_SDK_ROOT_DIR was set successfully." FORCE)
689 include(FindPackageHandleStandardArgs)
690 find_package_handle_standard_args(CUDA DEFAULT_MSG
691 CUDA_TOOLKIT_ROOT_DIR
695 _cuda_version_acceptable
700 ###############################################################################
701 ###############################################################################
703 ###############################################################################
704 ###############################################################################
706 ###############################################################################
707 # Add include directories to pass to the nvcc command.
708 macro(CUDA_INCLUDE_DIRECTORIES)
710 list(APPEND CUDA_NVCC_INCLUDE_ARGS_USER "-I${dir}")
711 endforeach(dir ${ARGN})
712 endmacro(CUDA_INCLUDE_DIRECTORIES)
715 ##############################################################################
716 cuda_find_helper_file(parse_cubin cmake)
717 cuda_find_helper_file(make2cmake cmake)
718 cuda_find_helper_file(run_nvcc cmake)
720 ##############################################################################
721 # Separate the OPTIONS out from the sources
723 macro(CUDA_GET_SOURCES_AND_OPTIONS _sources _cmake_options _options)
725 set( ${_cmake_options} )
727 set( _found_options FALSE )
729 if(arg STREQUAL "OPTIONS")
730 set( _found_options TRUE )
732 arg STREQUAL "WIN32" OR
733 arg STREQUAL "MACOSX_BUNDLE" OR
734 arg STREQUAL "EXCLUDE_FROM_ALL" OR
735 arg STREQUAL "STATIC" OR
736 arg STREQUAL "SHARED" OR
737 arg STREQUAL "MODULE"
739 list(APPEND ${_cmake_options} "${arg}")
741 if ( _found_options )
742 list(APPEND ${_options} "${arg}")
744 # Assume this is a file
745 list(APPEND ${_sources} "${arg}")
751 ##############################################################################
752 # Parse the OPTIONS from ARGN and set the variables prefixed by _option_prefix
754 macro(CUDA_PARSE_NVCC_OPTIONS _option_prefix)
757 # Determine if we are dealing with a perconfiguration flag
758 foreach(config ${CUDA_configuration_types})
759 string(TOUPPER ${config} config_upper)
760 if (arg STREQUAL "${config_upper}")
761 set( _found_config _${arg})
762 # Set arg to nothing to keep it from being processed further
768 list(APPEND ${_option_prefix}${_found_config} "${arg}")
773 ##############################################################################
774 # Helper to add the include directory for CUDA only once
775 function(CUDA_ADD_CUDA_INCLUDE_ONCE)
776 get_directory_property(_include_directories INCLUDE_DIRECTORIES)
778 if(_include_directories)
779 foreach(dir ${_include_directories})
780 if("${dir}" STREQUAL "${CUDA_INCLUDE_DIRS}")
786 include_directories(${CUDA_INCLUDE_DIRS})
790 function(CUDA_BUILD_SHARED_LIBRARY shared_flag)
791 set(cmake_args ${ARGN})
792 # If SHARED, MODULE, or STATIC aren't already in the list of arguments, then
793 # add SHARED or STATIC based on the value of BUILD_SHARED_LIBS.
794 list(FIND cmake_args SHARED _cuda_found_SHARED)
795 list(FIND cmake_args MODULE _cuda_found_MODULE)
796 list(FIND cmake_args STATIC _cuda_found_STATIC)
797 if( _cuda_found_SHARED GREATER -1 OR
798 _cuda_found_MODULE GREATER -1 OR
799 _cuda_found_STATIC GREATER -1)
800 set(_cuda_build_shared_libs)
802 if (BUILD_SHARED_LIBS)
803 set(_cuda_build_shared_libs SHARED)
805 set(_cuda_build_shared_libs STATIC)
808 set(${shared_flag} ${_cuda_build_shared_libs} PARENT_SCOPE)
811 ##############################################################################
812 # This helper macro populates the following variables and setups up custom
813 # commands and targets to invoke the nvcc compiler to generate C or PTX source
814 # dependant upon the format parameter. The compiler is invoked once with -M
815 # to generate a dependency file and a second time with -cuda or -ptx to generate
816 # a .cpp or .ptx file.
818 # cuda_target - Target name
819 # format - PTX or OBJ
820 # FILE1 .. FILEN - The remaining arguments are the sources to be wrapped.
821 # OPTIONS - Extra options to NVCC
823 # generated_files - List of generated files
824 ##############################################################################
825 ##############################################################################
827 macro(CUDA_WRAP_SRCS cuda_target format generated_files)
829 if( ${format} MATCHES "PTX" )
830 set( compile_to_ptx ON )
831 elseif( ${format} MATCHES "OBJ")
832 set( compile_to_ptx OFF )
834 message( FATAL_ERROR "Invalid format flag passed to CUDA_WRAP_SRCS: '${format}'. Use OBJ or PTX.")
837 # Set up all the command line flags here, so that they can be overriden on a per target basis.
841 # Emulation if the card isn't present.
842 if (CUDA_BUILD_EMULATION)
844 set(nvcc_flags ${nvcc_flags} --device-emulation -D_DEVICEEMU -g)
845 else(CUDA_BUILD_EMULATION)
846 # Device mode. No flags necessary.
847 endif(CUDA_BUILD_EMULATION)
849 if(CUDA_HOST_COMPILATION_CPP)
850 set(CUDA_C_OR_CXX CXX)
851 else(CUDA_HOST_COMPILATION_CPP)
852 if(CUDA_VERSION VERSION_LESS "3.0")
853 set(nvcc_flags ${nvcc_flags} --host-compilation C)
855 message(WARNING "--host-compilation flag is deprecated in CUDA version >= 3.0. Removing --host-compilation C flag" )
858 endif(CUDA_HOST_COMPILATION_CPP)
860 set(generated_extension ${CMAKE_${CUDA_C_OR_CXX}_OUTPUT_EXTENSION})
862 if(CUDA_64_BIT_DEVICE_CODE)
863 set(nvcc_flags ${nvcc_flags} -m64)
865 set(nvcc_flags ${nvcc_flags} -m32)
868 # This needs to be passed in at this stage, because VS needs to fill out the
869 # value of VCInstallDir from within VS.
870 if(CMAKE_GENERATOR MATCHES "Visual Studio")
871 if( CMAKE_SIZEOF_VOID_P EQUAL 8 )
872 # Add nvcc flag for 64b Windows
873 set(ccbin_flags -D "\"CCBIN:PATH=$(VCInstallDir)bin\"" )
877 # Figure out which configure we will use and pass that in as an argument to
878 # the script. We need to defer the decision until compilation time, because
879 # for VS projects we won't know if we are making a debug or release build
881 if(CMAKE_GENERATOR MATCHES "Visual Studio")
882 set( CUDA_build_configuration "$(ConfigurationName)" )
884 set( CUDA_build_configuration "${CMAKE_BUILD_TYPE}")
887 # Initialize our list of includes with the user ones followed by the CUDA system ones.
888 set(CUDA_NVCC_INCLUDE_ARGS ${CUDA_NVCC_INCLUDE_ARGS_USER} "-I${CUDA_INCLUDE_DIRS}")
889 # Get the include directories for this directory and use them for our nvcc command.
890 get_directory_property(CUDA_NVCC_INCLUDE_DIRECTORIES INCLUDE_DIRECTORIES)
891 if(CUDA_NVCC_INCLUDE_DIRECTORIES)
892 foreach(dir ${CUDA_NVCC_INCLUDE_DIRECTORIES})
893 list(APPEND CUDA_NVCC_INCLUDE_ARGS "-I${dir}")
897 # Reset these variables
898 set(CUDA_WRAP_OPTION_NVCC_FLAGS)
899 foreach(config ${CUDA_configuration_types})
900 string(TOUPPER ${config} config_upper)
901 set(CUDA_WRAP_OPTION_NVCC_FLAGS_${config_upper})
904 CUDA_GET_SOURCES_AND_OPTIONS(_cuda_wrap_sources _cuda_wrap_cmake_options _cuda_wrap_options ${ARGN})
905 CUDA_PARSE_NVCC_OPTIONS(CUDA_WRAP_OPTION_NVCC_FLAGS ${_cuda_wrap_options})
907 # Figure out if we are building a shared library. BUILD_SHARED_LIBS is
908 # respected in CUDA_ADD_LIBRARY.
909 set(_cuda_build_shared_libs FALSE)
911 list(FIND _cuda_wrap_cmake_options SHARED _cuda_found_SHARED)
912 list(FIND _cuda_wrap_cmake_options MODULE _cuda_found_MODULE)
913 if(_cuda_found_SHARED GREATER -1 OR _cuda_found_MODULE GREATER -1)
914 set(_cuda_build_shared_libs TRUE)
917 list(FIND _cuda_wrap_cmake_options STATIC _cuda_found_STATIC)
918 if(_cuda_found_STATIC GREATER -1)
919 set(_cuda_build_shared_libs FALSE)
923 if(_cuda_build_shared_libs)
924 # If we are setting up code for a shared library, then we need to add extra flags for
925 # compiling objects for shared libraries.
926 set(CUDA_HOST_SHARED_FLAGS ${CMAKE_SHARED_LIBRARY_${CUDA_C_OR_CXX}_FLAGS})
928 set(CUDA_HOST_SHARED_FLAGS)
930 # Only add the CMAKE_{C,CXX}_FLAGS if we are propagating host flags. We
931 # always need to set the SHARED_FLAGS, though.
932 if(CUDA_PROPAGATE_HOST_FLAGS)
933 set(CUDA_HOST_FLAGS "set(CMAKE_HOST_FLAGS ${CMAKE_${CUDA_C_OR_CXX}_FLAGS} ${CUDA_HOST_SHARED_FLAGS})")
935 set(CUDA_HOST_FLAGS "set(CMAKE_HOST_FLAGS ${CUDA_HOST_SHARED_FLAGS})")
938 set(CUDA_NVCC_FLAGS_CONFIG "# Build specific configuration flags")
939 # Loop over all the configuration types to generate appropriate flags for run_nvcc.cmake
940 foreach(config ${CUDA_configuration_types})
941 string(TOUPPER ${config} config_upper)
942 # CMAKE_FLAGS are strings and not lists. By not putting quotes around CMAKE_FLAGS
943 # we convert the strings to lists (like we want).
945 if(CUDA_PROPAGATE_HOST_FLAGS)
946 # nvcc chokes on -g3, so replace it with -g
947 if(CMAKE_COMPILER_IS_GNUCC)
948 string(REPLACE "-g3" "-g" _cuda_C_FLAGS "${CMAKE_${CUDA_C_OR_CXX}_FLAGS_${config_upper}}")
950 set(_cuda_C_FLAGS "${CMAKE_${CUDA_C_OR_CXX}_FLAGS_${config_upper}}")
953 set(CUDA_HOST_FLAGS "${CUDA_HOST_FLAGS}\nset(CMAKE_HOST_FLAGS_${config_upper} ${_cuda_C_FLAGS})")
956 # Note that if we ever want CUDA_NVCC_FLAGS_<CONFIG> to be string (instead of a list
957 # like it is currently), we can remove the quotes around the
958 # ${CUDA_NVCC_FLAGS_${config_upper}} variable like the CMAKE_HOST_FLAGS_<CONFIG> variable.
959 set(CUDA_NVCC_FLAGS_CONFIG "${CUDA_NVCC_FLAGS_CONFIG}\nset(CUDA_NVCC_FLAGS_${config_upper} \"${CUDA_NVCC_FLAGS_${config_upper}};;${CUDA_WRAP_OPTION_NVCC_FLAGS_${config_upper}}\")")
963 # Don't use any of the host compilation flags for PTX targets.
965 set(CUDA_NVCC_FLAGS_CONFIG)
968 # Get the list of definitions from the directory property
969 get_directory_property(CUDA_NVCC_DEFINITIONS COMPILE_DEFINITIONS)
970 if(CUDA_NVCC_DEFINITIONS)
971 foreach(_definition ${CUDA_NVCC_DEFINITIONS})
972 list(APPEND nvcc_flags "-D${_definition}")
976 if(_cuda_build_shared_libs)
977 list(APPEND nvcc_flags "-D${cuda_target}_EXPORTS")
980 # Determine output directory
981 if(CUDA_GENERATED_OUTPUT_DIR)
982 set(cuda_compile_output_dir "${CUDA_GENERATED_OUTPUT_DIR}")
984 set(cuda_compile_output_dir "${CMAKE_CURRENT_BINARY_DIR}")
987 # Reset the output variable
988 set(_cuda_wrap_generated_files "")
990 # Iterate over the macro arguments and create custom
991 # commands for all the .cu files.
992 foreach(file ${ARGN})
993 # Ignore any file marked as a HEADER_FILE_ONLY
994 get_source_file_property(_is_header ${file} HEADER_FILE_ONLY)
995 if(${file} MATCHES ".*\\.cu$" AND NOT _is_header)
997 # Add a custom target to generate a c or ptx file. ######################
999 get_filename_component( basename ${file} NAME )
1000 if( compile_to_ptx )
1001 set(generated_file_path "${cuda_compile_output_dir}")
1002 set(generated_file_basename "${cuda_target}_generated_${basename}.ptx")
1003 set(format_flag "-ptx")
1004 file(MAKE_DIRECTORY "${cuda_compile_output_dir}")
1005 else( compile_to_ptx )
1006 set(generated_file_path "${cuda_compile_output_dir}/${CMAKE_CFG_INTDIR}")
1007 set(generated_file_basename "${cuda_target}_generated_${basename}${generated_extension}")
1008 set(format_flag "-c")
1009 endif( compile_to_ptx )
1011 # Set all of our file names. Make sure that whatever filenames that have
1012 # generated_file_path in them get passed in through as a command line
1013 # argument, so that the ${CMAKE_CFG_INTDIR} gets expanded at run time
1014 # instead of configure time.
1015 set(generated_file "${generated_file_path}/${generated_file_basename}")
1016 set(cmake_dependency_file "${CMAKE_CURRENT_BINARY_DIR}/CMakeFiles/${generated_file_basename}.depend")
1017 set(NVCC_generated_dependency_file "${CMAKE_CURRENT_BINARY_DIR}/CMakeFiles/${generated_file_basename}.NVCC-depend")
1018 set(generated_cubin_file "${generated_file_path}/${generated_file_basename}.cubin.txt")
1019 set(custom_target_script "${CMAKE_CURRENT_BINARY_DIR}/CMakeFiles/${generated_file_basename}.cmake")
1021 # Setup properties for obj files:
1022 if( NOT compile_to_ptx )
1023 set_source_files_properties("${generated_file}"
1025 EXTERNAL_OBJECT true # This is an object file not to be compiled, but only be linked.
1029 # Don't add CMAKE_CURRENT_SOURCE_DIR if the path is already an absolute path.
1030 get_filename_component(file_path "${file}" PATH)
1031 if(IS_ABSOLUTE "${file_path}")
1032 set(source_file "${file}")
1034 set(source_file "${CMAKE_CURRENT_SOURCE_DIR}/${file}")
1037 # Bring in the dependencies. Creates a variable CUDA_NVCC_DEPEND #######
1038 cuda_include_nvcc_dependencies(${cmake_dependency_file})
1040 # Convience string for output ###########################################
1041 if(CUDA_BUILD_EMULATION)
1042 set(cuda_build_type "Emulation")
1043 else(CUDA_BUILD_EMULATION)
1044 set(cuda_build_type "Device")
1045 endif(CUDA_BUILD_EMULATION)
1047 # Build the NVCC made dependency file ###################################
1048 set(build_cubin OFF)
1049 if ( NOT CUDA_BUILD_EMULATION AND CUDA_BUILD_CUBIN )
1050 if ( NOT compile_to_ptx )
1051 set ( build_cubin ON )
1052 endif( NOT compile_to_ptx )
1053 endif( NOT CUDA_BUILD_EMULATION AND CUDA_BUILD_CUBIN )
1055 # Configure the build script
1056 configure_file("${CUDA_run_nvcc}" "${custom_target_script}" @ONLY)
1058 # So if a user specifies the same cuda file as input more than once, you
1059 # can have bad things happen with dependencies. Here we check an option
1060 # to see if this is the behavior they want.
1061 if(CUDA_ATTACH_VS_BUILD_RULE_TO_CUDA_FILE)
1062 set(main_dep MAIN_DEPENDENCY ${source_file})
1064 set(main_dep DEPENDS ${source_file})
1067 if(CUDA_VERBOSE_BUILD)
1068 set(verbose_output ON)
1069 elseif(CMAKE_GENERATOR MATCHES "Makefiles")
1070 set(verbose_output "$(VERBOSE)")
1072 set(verbose_output OFF)
1075 # Create up the comment string
1076 file(RELATIVE_PATH generated_file_relative_path "${CMAKE_BINARY_DIR}" "${generated_file}")
1078 set(cuda_build_comment_string "Building NVCC ptx file ${generated_file_relative_path}")
1080 set(cuda_build_comment_string "Building NVCC (${cuda_build_type}) object ${generated_file_relative_path}")
1083 # Build the generated file and dependency file ##########################
1085 OUTPUT ${generated_file}
1086 # These output files depend on the source_file and the contents of cmake_dependency_file
1088 DEPENDS ${CUDA_NVCC_DEPEND}
1089 DEPENDS ${custom_target_script}
1090 # Make sure the output directory exists before trying to write to it.
1091 COMMAND ${CMAKE_COMMAND} -E make_directory "${generated_file_path}"
1092 COMMAND ${CMAKE_COMMAND} ARGS
1093 -D verbose:BOOL=${verbose_output}
1095 -D build_configuration:STRING=${CUDA_build_configuration}
1096 -D "generated_file:STRING=${generated_file}"
1097 -D "generated_cubin_file:STRING=${generated_cubin_file}"
1098 -P "${custom_target_script}"
1099 COMMENT "${cuda_build_comment_string}"
1102 # Make sure the build system knows the file is generated.
1103 set_source_files_properties(${generated_file} PROPERTIES GENERATED TRUE)
1105 # Don't add the object file to the list of generated files if we are using
1106 # visual studio and we are attaching the build rule to the cuda file. VS
1107 # will add our object file to the linker automatically for us.
1108 set(cuda_add_generated_file TRUE)
1110 if(NOT compile_to_ptx AND CMAKE_GENERATOR MATCHES "Visual Studio" AND CUDA_ATTACH_VS_BUILD_RULE_TO_CUDA_FILE)
1111 # Visual Studio 8 crashes when you close the solution when you don't add the object file.
1112 if(NOT CMAKE_GENERATOR MATCHES "Visual Studio 8")
1113 #message("Not adding ${generated_file}")
1114 set(cuda_add_generated_file FALSE)
1118 if(cuda_add_generated_file)
1119 list(APPEND _cuda_wrap_generated_files ${generated_file})
1122 # Add the other files that we want cmake to clean on a cleanup ##########
1123 list(APPEND CUDA_ADDITIONAL_CLEAN_FILES "${cmake_dependency_file}")
1124 list(REMOVE_DUPLICATES CUDA_ADDITIONAL_CLEAN_FILES)
1125 set(CUDA_ADDITIONAL_CLEAN_FILES ${CUDA_ADDITIONAL_CLEAN_FILES} CACHE INTERNAL "List of intermediate files that are part of the cuda dependency scanning.")
1127 endif(${file} MATCHES ".*\\.cu$" AND NOT _is_header)
1130 # Set the return parameter
1131 set(${generated_files} ${_cuda_wrap_generated_files})
1132 endmacro(CUDA_WRAP_SRCS)
1135 ###############################################################################
1136 ###############################################################################
1138 ###############################################################################
1139 ###############################################################################
1140 macro(CUDA_ADD_LIBRARY cuda_target)
1142 CUDA_ADD_CUDA_INCLUDE_ONCE()
1144 # Separate the sources from the options
1145 CUDA_GET_SOURCES_AND_OPTIONS(_sources _cmake_options _options ${ARGN})
1146 CUDA_BUILD_SHARED_LIBRARY(_cuda_shared_flag ${ARGN})
1147 # Create custom commands and targets for each file.
1148 CUDA_WRAP_SRCS( ${cuda_target} OBJ _generated_files ${_sources}
1149 ${_cmake_options} ${_cuda_shared_flag}
1150 OPTIONS ${_options} )
1153 add_library(${cuda_target} ${_cmake_options}
1158 target_link_libraries(${cuda_target}
1162 # We need to set the linker language based on what the expected generated file
1163 # would be. CUDA_C_OR_CXX is computed based on CUDA_HOST_COMPILATION_CPP.
1164 set_target_properties(${cuda_target}
1166 LINKER_LANGUAGE ${CUDA_C_OR_CXX}
1169 endmacro(CUDA_ADD_LIBRARY cuda_target)
1172 ###############################################################################
1173 ###############################################################################
1175 ###############################################################################
1176 ###############################################################################
1177 macro(CUDA_ADD_EXECUTABLE cuda_target)
1179 CUDA_ADD_CUDA_INCLUDE_ONCE()
1181 # Separate the sources from the options
1182 CUDA_GET_SOURCES_AND_OPTIONS(_sources _cmake_options _options ${ARGN})
1183 # Create custom commands and targets for each file.
1184 CUDA_WRAP_SRCS( ${cuda_target} OBJ _generated_files ${_sources} OPTIONS ${_options} )
1187 add_executable(${cuda_target} ${_cmake_options}
1192 target_link_libraries(${cuda_target}
1196 # We need to set the linker language based on what the expected generated file
1197 # would be. CUDA_C_OR_CXX is computed based on CUDA_HOST_COMPILATION_CPP.
1198 set_target_properties(${cuda_target}
1200 LINKER_LANGUAGE ${CUDA_C_OR_CXX}
1203 endmacro(CUDA_ADD_EXECUTABLE cuda_target)
1206 ###############################################################################
1207 ###############################################################################
1209 ###############################################################################
1210 ###############################################################################
1211 macro(CUDA_COMPILE generated_files)
1213 # Separate the sources from the options
1214 CUDA_GET_SOURCES_AND_OPTIONS(_sources _cmake_options _options ${ARGN})
1215 # Create custom commands and targets for each file.
1216 CUDA_WRAP_SRCS( cuda_compile OBJ _generated_files ${_sources} ${_cmake_options}
1217 OPTIONS ${_options} )
1219 set( ${generated_files} ${_generated_files})
1221 endmacro(CUDA_COMPILE)
1224 ###############################################################################
1225 ###############################################################################
1227 ###############################################################################
1228 ###############################################################################
1229 macro(CUDA_COMPILE_PTX generated_files)
1231 # Separate the sources from the options
1232 CUDA_GET_SOURCES_AND_OPTIONS(_sources _cmake_options _options ${ARGN})
1233 # Create custom commands and targets for each file.
1234 CUDA_WRAP_SRCS( cuda_compile_ptx PTX _generated_files ${_sources} ${_cmake_options}
1235 OPTIONS ${_options} )
1237 set( ${generated_files} ${_generated_files})
1239 endmacro(CUDA_COMPILE_PTX)
1241 ###############################################################################
1242 ###############################################################################
1243 # CUDA ADD CUFFT TO TARGET
1244 ###############################################################################
1245 ###############################################################################
1246 macro(CUDA_ADD_CUFFT_TO_TARGET target)
1247 if (CUDA_BUILD_EMULATION)
1248 target_link_libraries(${target} ${CUDA_cufftemu_LIBRARY})
1250 target_link_libraries(${target} ${CUDA_cufft_LIBRARY})
1254 ###############################################################################
1255 ###############################################################################
1256 # CUDA ADD CUBLAS TO TARGET
1257 ###############################################################################
1258 ###############################################################################
1259 macro(CUDA_ADD_CUBLAS_TO_TARGET target)
1260 if (CUDA_BUILD_EMULATION)
1261 target_link_libraries(${target} ${CUDA_cublasemu_LIBRARY})
1263 target_link_libraries(${target} ${CUDA_cublas_LIBRARY})
1267 ###############################################################################
1268 ###############################################################################
1269 # CUDA BUILD CLEAN TARGET
1270 ###############################################################################
1271 ###############################################################################
1272 macro(CUDA_BUILD_CLEAN_TARGET)
1273 # Call this after you add all your CUDA targets, and you will get a convience
1274 # target. You should also make clean after running this target to get the
1275 # build system to generate all the code again.
1277 set(cuda_clean_target_name clean_cuda_depends)
1278 if (CMAKE_GENERATOR MATCHES "Visual Studio")
1279 string(TOUPPER ${cuda_clean_target_name} cuda_clean_target_name)
1281 add_custom_target(${cuda_clean_target_name}
1282 COMMAND ${CMAKE_COMMAND} -E remove ${CUDA_ADDITIONAL_CLEAN_FILES})
1284 # Clear out the variable, so the next time we configure it will be empty.
1285 # This is useful so that the files won't persist in the list after targets
1286 # have been removed.
1287 set(CUDA_ADDITIONAL_CLEAN_FILES "" CACHE INTERNAL "List of intermediate files that are part of the cuda dependency scanning.")
1288 endmacro(CUDA_BUILD_CLEAN_TARGET)