5
# Tools for building CUDA C files: libraries and build dependencies.
7
# This script locates the NVIDIA CUDA C tools. It should work on linux,
8
# windows, and mac and should be reasonably up to date with CUDA C
11
# This script makes use of the standard find_package arguments of
12
# <VERSION>, REQUIRED and QUIET. CUDA_FOUND will report if an
13
# acceptable version of CUDA was found.
15
# The script will prompt the user to specify CUDA_TOOLKIT_ROOT_DIR if
16
# the prefix cannot be determined by the location of nvcc in the system
17
# path and REQUIRED is specified to find_package(). To use a different
18
# installed version of the toolkit set the environment variable
19
# CUDA_BIN_PATH before running cmake (e.g.
20
# CUDA_BIN_PATH=/usr/local/cuda1.0 instead of the default
21
# /usr/local/cuda) or set CUDA_TOOLKIT_ROOT_DIR after configuring. If
22
# you change the value of CUDA_TOOLKIT_ROOT_DIR, various components that
23
# depend on the path will be relocated.
25
# It might be necessary to set CUDA_TOOLKIT_ROOT_DIR manually on certain
26
# platforms, or to use a cuda runtime not installed in the default
27
# location. In newer versions of the toolkit the cuda library is
28
# included with the graphics driver- be sure that the driver version
29
# matches what is needed by the cuda runtime version.
31
# The following variables affect the behavior of the macros in the
32
# script (in alphebetical order). Note that any of these flags can be
33
# changed multiple times in the same directory before calling
34
# CUDA_ADD_EXECUTABLE, CUDA_ADD_LIBRARY, CUDA_COMPILE, CUDA_COMPILE_PTX,
35
# CUDA_COMPILE_FATBIN, CUDA_COMPILE_CUBIN or CUDA_WRAP_SRCS::
37
# CUDA_64_BIT_DEVICE_CODE (Default matches host bit size)
38
# -- Set to ON to compile for 64 bit device code, OFF for 32 bit device code.
39
# Note that making this different from the host code when generating object
40
# or C files from CUDA code just won't work, because size_t gets defined by
41
# nvcc in the generated source. If you compile to PTX and then load the
42
# file yourself, you can mix bit sizes between device and host.
44
# CUDA_ATTACH_VS_BUILD_RULE_TO_CUDA_FILE (Default ON)
45
# -- Set to ON if you want the custom build rule to be attached to the source
46
# file in Visual Studio. Turn OFF if you add the same cuda file to multiple
49
# This allows the user to build the target from the CUDA file; however, bad
50
# things can happen if the CUDA source file is added to multiple targets.
51
# When performing parallel builds it is possible for the custom build
52
# command to be run more than once and in parallel causing cryptic build
53
# errors. VS runs the rules for every source file in the target, and a
54
# source can have only one rule no matter how many projects it is added to.
55
# When the rule is run from multiple targets race conditions can occur on
56
# the generated file. Eventually everything will get built, but if the user
57
# is unaware of this behavior, there may be confusion. It would be nice if
58
# this script could detect the reuse of source files across multiple targets
59
# and turn the option off for the user, but no good solution could be found.
61
# CUDA_BUILD_CUBIN (Default OFF)
62
# -- Set to ON to enable and extra compilation pass with the -cubin option in
63
# Device mode. The output is parsed and register, shared memory usage is
64
# printed during build.
66
# CUDA_BUILD_EMULATION (Default OFF for device mode)
67
# -- Set to ON for Emulation mode. -D_DEVICEEMU is defined for CUDA C files
68
# when CUDA_BUILD_EMULATION is TRUE.
70
# CUDA_GENERATED_OUTPUT_DIR (Default CMAKE_CURRENT_BINARY_DIR)
71
# -- Set to the path you wish to have the generated files placed. If it is
72
# blank output files will be placed in CMAKE_CURRENT_BINARY_DIR.
73
# Intermediate files will always be placed in
74
# CMAKE_CURRENT_BINARY_DIR/CMakeFiles.
76
# CUDA_HOST_COMPILATION_CPP (Default ON)
77
# -- Set to OFF for C compilation of host code.
79
# CUDA_HOST_COMPILER (Default CMAKE_C_COMPILER, $(VCInstallDir)/bin for VS)
80
# -- Set the host compiler to be used by nvcc. Ignored if -ccbin or
81
# --compiler-bindir is already present in the CUDA_NVCC_FLAGS or
82
# CUDA_NVCC_FLAGS_<CONFIG> variables. For Visual Studio targets
83
# $(VCInstallDir)/bin is a special value that expands out to the path when
84
# the command is run from withing VS.
87
# CUDA_NVCC_FLAGS_<CONFIG>
88
# -- Additional NVCC command line arguments. NOTE: multiple arguments must be
89
# semi-colon delimited (e.g. --compiler-options;-Wall)
91
# CUDA_PROPAGATE_HOST_FLAGS (Default ON)
92
# -- Set to ON to propagate CMAKE_{C,CXX}_FLAGS and their configuration
93
# dependent counterparts (e.g. CMAKE_C_FLAGS_DEBUG) automatically to the
94
# host compiler through nvcc's -Xcompiler flag. This helps make the
95
# generated host code match the rest of the system better. Sometimes
96
# certain flags give nvcc problems, and this will help you turn the flag
97
# propagation off. This does not affect the flags supplied directly to nvcc
98
# via CUDA_NVCC_FLAGS or through the OPTION flags specified through
99
# CUDA_ADD_LIBRARY, CUDA_ADD_EXECUTABLE, or CUDA_WRAP_SRCS. Flags used for
100
# shared library compilation are not affected by this flag.
102
# CUDA_SEPARABLE_COMPILATION (Default OFF)
103
# -- If set this will enable separable compilation for all CUDA runtime object
104
# files. If used outside of CUDA_ADD_EXECUTABLE and CUDA_ADD_LIBRARY
105
# (e.g. calling CUDA_WRAP_SRCS directly),
106
# CUDA_COMPUTE_SEPARABLE_COMPILATION_OBJECT_FILE_NAME and
107
# CUDA_LINK_SEPARABLE_COMPILATION_OBJECTS should be called.
109
# CUDA_VERBOSE_BUILD (Default OFF)
110
# -- Set to ON to see all the commands used when building the CUDA file. When
111
# using a Makefile generator the value defaults to VERBOSE (run make
112
# VERBOSE=1 to see output), although setting CUDA_VERBOSE_BUILD to ON will
113
# always print the output.
115
# The script creates the following macros (in alphebetical order)::
117
# CUDA_ADD_CUFFT_TO_TARGET( cuda_target )
118
# -- Adds the cufft library to the target (can be any target). Handles whether
119
# you are in emulation mode or not.
121
# CUDA_ADD_CUBLAS_TO_TARGET( cuda_target )
122
# -- Adds the cublas library to the target (can be any target). Handles
123
# whether you are in emulation mode or not.
125
# CUDA_ADD_EXECUTABLE( cuda_target file0 file1 ...
126
# [WIN32] [MACOSX_BUNDLE] [EXCLUDE_FROM_ALL] [OPTIONS ...] )
127
# -- Creates an executable "cuda_target" which is made up of the files
128
# specified. All of the non CUDA C files are compiled using the standard
129
# build rules specified by CMAKE and the cuda files are compiled to object
130
# files using nvcc and the host compiler. In addition CUDA_INCLUDE_DIRS is
131
# added automatically to include_directories(). Some standard CMake target
132
# calls can be used on the target after calling this macro
133
# (e.g. set_target_properties and target_link_libraries), but setting
134
# properties that adjust compilation flags will not affect code compiled by
135
# nvcc. Such flags should be modified before calling CUDA_ADD_EXECUTABLE,
136
# CUDA_ADD_LIBRARY or CUDA_WRAP_SRCS.
138
# CUDA_ADD_LIBRARY( cuda_target file0 file1 ...
139
# [STATIC | SHARED | MODULE] [EXCLUDE_FROM_ALL] [OPTIONS ...] )
140
# -- Same as CUDA_ADD_EXECUTABLE except that a library is created.
142
# CUDA_BUILD_CLEAN_TARGET()
143
# -- Creates a convience target that deletes all the dependency files
144
# generated. You should make clean after running this target to ensure the
145
# dependency files get regenerated.
147
# CUDA_COMPILE( generated_files file0 file1 ... [STATIC | SHARED | MODULE]
149
# -- Returns a list of generated files from the input source files to be used
150
# with ADD_LIBRARY or ADD_EXECUTABLE.
152
# CUDA_COMPILE_PTX( generated_files file0 file1 ... [OPTIONS ...] )
153
# -- Returns a list of PTX files generated from the input source files.
155
# CUDA_COMPILE_FATBIN( generated_files file0 file1 ... [OPTIONS ...] )
156
# -- Returns a list of FATBIN files generated from the input source files.
158
# CUDA_COMPILE_CUBIN( generated_files file0 file1 ... [OPTIONS ...] )
159
# -- Returns a list of CUBIN files generated from the input source files.
161
# CUDA_COMPUTE_SEPARABLE_COMPILATION_OBJECT_FILE_NAME( output_file_var
164
# -- Compute the name of the intermediate link file used for separable
165
# compilation. This file name is typically passed into
166
# CUDA_LINK_SEPARABLE_COMPILATION_OBJECTS. output_file_var is produced
167
# based on cuda_target the list of objects files that need separable
168
# compilation as specified by object_files. If the object_files list is
169
# empty, then output_file_var will be empty. This function is called
170
# automatically for CUDA_ADD_LIBRARY and CUDA_ADD_EXECUTABLE. Note that
171
# this is a function and not a macro.
173
# CUDA_INCLUDE_DIRECTORIES( path0 path1 ... )
174
# -- Sets the directories that should be passed to nvcc
175
# (e.g. nvcc -Ipath0 -Ipath1 ... ). These paths usually contain other .cu
180
# CUDA_LINK_SEPARABLE_COMPILATION_OBJECTS( output_file_var cuda_target
181
# nvcc_flags object_files)
183
# -- Generates the link object required by separable compilation from the given
184
# object files. This is called automatically for CUDA_ADD_EXECUTABLE and
185
# CUDA_ADD_LIBRARY, but can be called manually when using CUDA_WRAP_SRCS
186
# directly. When called from CUDA_ADD_LIBRARY or CUDA_ADD_EXECUTABLE the
187
# nvcc_flags passed in are the same as the flags passed in via the OPTIONS
188
# argument. The only nvcc flag added automatically is the bitness flag as
189
# specified by CUDA_64_BIT_DEVICE_CODE. Note that this is a function
190
# instead of a macro.
192
# CUDA_WRAP_SRCS ( cuda_target format generated_files file0 file1 ...
193
# [STATIC | SHARED | MODULE] [OPTIONS ...] )
194
# -- This is where all the magic happens. CUDA_ADD_EXECUTABLE,
195
# CUDA_ADD_LIBRARY, CUDA_COMPILE, and CUDA_COMPILE_PTX all call this
196
# function under the hood.
198
# Given the list of files (file0 file1 ... fileN) this macro generates
199
# custom commands that generate either PTX or linkable objects (use "PTX" or
200
# "OBJ" for the format argument to switch). Files that don't end with .cu
201
# or have the HEADER_FILE_ONLY property are ignored.
203
# The arguments passed in after OPTIONS are extra command line options to
204
# give to nvcc. You can also specify per configuration options by
205
# specifying the name of the configuration followed by the options. General
206
# options must preceed configuration specific options. Not all
207
# configurations need to be specified, only the ones provided will be used.
209
# OPTIONS -DFLAG=2 "-DFLAG_OTHER=space in flag"
211
# RELEASE --use_fast_math
212
# RELWITHDEBINFO --use_fast_math;-g
213
# MINSIZEREL --use_fast_math
215
# For certain configurations (namely VS generating object files with
216
# CUDA_ATTACH_VS_BUILD_RULE_TO_CUDA_FILE set to ON), no generated file will
217
# be produced for the given cuda file. This is because when you add the
218
# cuda file to Visual Studio it knows that this file produces an object file
219
# and will link in the resulting object file automatically.
221
# This script will also generate a separate cmake script that is used at
222
# build time to invoke nvcc. This is for several reasons.
224
# 1. nvcc can return negative numbers as return values which confuses
225
# Visual Studio into thinking that the command succeeded. The script now
226
# checks the error codes and produces errors when there was a problem.
228
# 2. nvcc has been known to not delete incomplete results when it
229
# encounters problems. This confuses build systems into thinking the
230
# target was generated when in fact an unusable file exists. The script
231
# now deletes the output files if there was an error.
233
# 3. By putting all the options that affect the build into a file and then
234
# make the build rule dependent on the file, the output files will be
235
# regenerated when the options change.
237
# This script also looks at optional arguments STATIC, SHARED, or MODULE to
238
# determine when to target the object compilation for a shared library.
239
# BUILD_SHARED_LIBS is ignored in CUDA_WRAP_SRCS, but it is respected in
240
# CUDA_ADD_LIBRARY. On some systems special flags are added for building
241
# objects intended for shared libraries. A preprocessor macro,
242
# <target_name>_EXPORTS is defined when a shared library compilation is
245
# Flags passed into add_definitions with -D or /D are passed along to nvcc.
249
# The script defines the following variables::
251
# CUDA_VERSION_MAJOR -- The major version of cuda as reported by nvcc.
252
# CUDA_VERSION_MINOR -- The minor version.
254
# CUDA_VERSION_STRING -- CUDA_VERSION_MAJOR.CUDA_VERSION_MINOR
256
# CUDA_TOOLKIT_ROOT_DIR -- Path to the CUDA Toolkit (defined if not set).
257
# CUDA_SDK_ROOT_DIR -- Path to the CUDA SDK. Use this to find files in the
258
# SDK. This script will not directly support finding
259
# specific libraries or headers, as that isn't
260
# supported by NVIDIA. If you want to change
261
# libraries when the path changes see the
262
# FindCUDA.cmake script for an example of how to clear
263
# these variables. There are also examples of how to
264
# use the CUDA_SDK_ROOT_DIR to locate headers or
265
# libraries, if you so choose (at your own risk).
266
# CUDA_INCLUDE_DIRS -- Include directory for cuda headers. Added automatically
267
# for CUDA_ADD_EXECUTABLE and CUDA_ADD_LIBRARY.
268
# CUDA_LIBRARIES -- Cuda RT library.
269
# CUDA_CUFFT_LIBRARIES -- Device or emulation library for the Cuda FFT
270
# implementation (alternative to:
271
# CUDA_ADD_CUFFT_TO_TARGET macro)
272
# CUDA_CUBLAS_LIBRARIES -- Device or emulation library for the Cuda BLAS
273
# implementation (alterative to:
274
# CUDA_ADD_CUBLAS_TO_TARGET macro).
275
# CUDA_cupti_LIBRARY -- CUDA Profiling Tools Interface library.
276
# Only available for CUDA version 4.0+.
277
# CUDA_curand_LIBRARY -- CUDA Random Number Generation library.
278
# Only available for CUDA version 3.2+.
279
# CUDA_cusparse_LIBRARY -- CUDA Sparse Matrix library.
280
# Only available for CUDA version 3.2+.
281
# CUDA_npp_LIBRARY -- NVIDIA Performance Primitives lib.
282
# Only available for CUDA version 4.0+.
283
# CUDA_nppc_LIBRARY -- NVIDIA Performance Primitives lib (core).
284
# Only available for CUDA version 5.5+.
285
# CUDA_nppi_LIBRARY -- NVIDIA Performance Primitives lib (image processing).
286
# Only available for CUDA version 5.5+.
287
# CUDA_npps_LIBRARY -- NVIDIA Performance Primitives lib (signal processing).
288
# Only available for CUDA version 5.5+.
289
# CUDA_nvcuvenc_LIBRARY -- CUDA Video Encoder library.
290
# Only available for CUDA version 3.2+.
292
# CUDA_nvcuvid_LIBRARY -- CUDA Video Decoder library.
293
# Only available for CUDA version 3.2+.
297
# James Bigler, NVIDIA Corp (nvidia.com - jbigler)
298
# Abe Stephens, SCI Institute -- http://www.sci.utah.edu/~abe/FindCuda.html
300
# Copyright (c) 2008 - 2009 NVIDIA Corporation. All rights reserved.
302
# Copyright (c) 2007-2009
303
# Scientific Computing and Imaging Institute, University of Utah
305
# This code is licensed under the MIT License. See the FindCUDA.cmake script
306
# for the text of the license.
310
# License for the specific language governing rights and limitations under
311
# Permission is hereby granted, free of charge, to any person obtaining a
312
# copy of this software and associated documentation files (the "Software"),
313
# to deal in the Software without restriction, including without limitation
314
# the rights to use, copy, modify, merge, publish, distribute, sublicense,
315
# and/or sell copies of the Software, and to permit persons to whom the
316
# Software is furnished to do so, subject to the following conditions:
318
# The above copyright notice and this permission notice shall be included
319
# in all copies or substantial portions of the Software.
321
# THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
322
# OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
323
# FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
324
# THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
325
# LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
326
# FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
327
# DEALINGS IN THE SOFTWARE.
329
###############################################################################
333
# This macro helps us find the location of helper files we will need the full path to
334
macro(CUDA_FIND_HELPER_FILE _name _extension)
335
set(_full_name "${_name}.${_extension}")
336
# CMAKE_CURRENT_LIST_FILE contains the full path to the file currently being
337
# processed. Using this variable, we can pull out the current path, and
338
# provide a way to get access to the other files we need local to here.
339
get_filename_component(CMAKE_CURRENT_LIST_DIR "${CMAKE_CURRENT_LIST_FILE}" PATH)
340
set(CUDA_${_name} "${CMAKE_CURRENT_LIST_DIR}/FindCUDA/${_full_name}")
341
if(NOT EXISTS "${CUDA_${_name}}")
342
set(error_message "${_full_name} not found in ${CMAKE_CURRENT_LIST_DIR}/FindCUDA")
343
if(CUDA_FIND_REQUIRED)
344
message(FATAL_ERROR "${error_message}")
346
if(NOT CUDA_FIND_QUIETLY)
347
message(STATUS "${error_message}")
351
# Set this variable as internal, so the user isn't bugged with it.
352
set(CUDA_${_name} ${CUDA_${_name}} CACHE INTERNAL "Location of ${_full_name}" FORCE)
355
#####################################################################
356
## CUDA_INCLUDE_NVCC_DEPENDENCIES
359
# So we want to try and include the dependency file if it exists. If
360
# it doesn't exist then we need to create an empty one, so we can
363
# If it does exist, then we need to check to see if all the files it
364
# depends on exist. If they don't then we should clear the dependency
365
# file and regenerate it later. This covers the case where a header
366
# file has disappeared or moved.
368
macro(CUDA_INCLUDE_NVCC_DEPENDENCIES dependency_file)
369
set(CUDA_NVCC_DEPEND)
370
set(CUDA_NVCC_DEPEND_REGENERATE FALSE)
373
# Include the dependency file. Create it first if it doesn't exist . The
374
# INCLUDE puts a dependency that will force CMake to rerun and bring in the
375
# new info when it changes. DO NOT REMOVE THIS (as I did and spent a few
376
# hours figuring out why it didn't work.
377
if(NOT EXISTS ${dependency_file})
378
file(WRITE ${dependency_file} "#FindCUDA.cmake generated file. Do not edit.\n")
380
# Always include this file to force CMake to run again next
381
# invocation and rebuild the dependencies.
382
#message("including dependency_file = ${dependency_file}")
383
include(${dependency_file})
385
# Now we need to verify the existence of all the included files
386
# here. If they aren't there we need to just blank this variable and
387
# make the file regenerate again.
388
# if(DEFINED CUDA_NVCC_DEPEND)
389
# message("CUDA_NVCC_DEPEND set")
391
# message("CUDA_NVCC_DEPEND NOT set")
394
#message("CUDA_NVCC_DEPEND found")
395
foreach(f ${CUDA_NVCC_DEPEND})
396
# message("searching for ${f}")
398
#message("file ${f} not found")
399
set(CUDA_NVCC_DEPEND_REGENERATE TRUE)
403
#message("CUDA_NVCC_DEPEND false")
404
# No dependencies, so regenerate the file.
405
set(CUDA_NVCC_DEPEND_REGENERATE TRUE)
408
#message("CUDA_NVCC_DEPEND_REGENERATE = ${CUDA_NVCC_DEPEND_REGENERATE}")
409
# No incoming dependencies, so we need to generate them. Make the
410
# output depend on the dependency file itself, which should cause the
412
if(CUDA_NVCC_DEPEND_REGENERATE)
413
set(CUDA_NVCC_DEPEND ${dependency_file})
414
#message("Generating an empty dependency_file: ${dependency_file}")
415
file(WRITE ${dependency_file} "#FindCUDA.cmake generated file. Do not edit.\n")
420
###############################################################################
421
###############################################################################
422
# Setup variables' defaults
423
###############################################################################
424
###############################################################################
426
# Allow the user to specify if the device code is supposed to be 32 or 64 bit.
427
if(CMAKE_SIZEOF_VOID_P EQUAL 8)
428
set(CUDA_64_BIT_DEVICE_CODE_DEFAULT ON)
430
set(CUDA_64_BIT_DEVICE_CODE_DEFAULT OFF)
432
option(CUDA_64_BIT_DEVICE_CODE "Compile device code in 64 bit mode" ${CUDA_64_BIT_DEVICE_CODE_DEFAULT})
434
# Attach the build rule to the source file in VS. This option
435
option(CUDA_ATTACH_VS_BUILD_RULE_TO_CUDA_FILE "Attach the build rule to the CUDA source file. Enable only when the CUDA source file is added to at most one target." ON)
437
# Prints out extra information about the cuda file during compilation
438
option(CUDA_BUILD_CUBIN "Generate and parse .cubin files in Device mode." OFF)
440
# Set whether we are using emulation or device mode.
441
option(CUDA_BUILD_EMULATION "Build in Emulation mode" OFF)
443
# Where to put the generated output.
444
set(CUDA_GENERATED_OUTPUT_DIR "" CACHE PATH "Directory to put all the output files. If blank it will default to the CMAKE_CURRENT_BINARY_DIR")
446
# Parse HOST_COMPILATION mode.
447
option(CUDA_HOST_COMPILATION_CPP "Generated file extension" ON)
449
# Extra user settable flags
450
set(CUDA_NVCC_FLAGS "" CACHE STRING "Semi-colon delimit multiple arguments.")
452
if(CMAKE_GENERATOR MATCHES "Visual Studio")
453
set(CUDA_HOST_COMPILER "$(VCInstallDir)bin" CACHE FILEPATH "Host side compiler used by NVCC")
455
# Using cc which is symlink to clang may let NVCC think it is GCC and issue
456
# unhandled -dumpspecs option to clang. Also in case neither
457
# CMAKE_C_COMPILER is defined (project does not use C language) nor
458
# CUDA_HOST_COMPILER is specified manually we should skip -ccbin and let
459
# nvcc use its own default C compiler.
460
if(DEFINED CMAKE_C_COMPILER AND NOT DEFINED CUDA_HOST_COMPILER)
461
get_filename_component(c_compiler_realpath "${CMAKE_C_COMPILER}" REALPATH)
463
set(c_compiler_realpath "")
465
set(CUDA_HOST_COMPILER "${c_compiler_realpath}" CACHE FILEPATH "Host side compiler used by NVCC")
468
# Propagate the host flags to the host compiler via -Xcompiler
469
option(CUDA_PROPAGATE_HOST_FLAGS "Propage C/CXX_FLAGS and friends to the host compiler via -Xcompile" ON)
471
# Enable CUDA_SEPARABLE_COMPILATION
472
option(CUDA_SEPARABLE_COMPILATION "Compile CUDA objects with separable compilation enabled. Requires CUDA 5.0+" OFF)
474
# Specifies whether the commands used when compiling the .cu file will be printed out.
475
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)
478
CUDA_64_BIT_DEVICE_CODE
479
CUDA_ATTACH_VS_BUILD_RULE_TO_CUDA_FILE
480
CUDA_GENERATED_OUTPUT_DIR
481
CUDA_HOST_COMPILATION_CPP
483
CUDA_PROPAGATE_HOST_FLAGS
486
# Makefile and similar generators don't define CMAKE_CONFIGURATION_TYPES, so we
487
# need to add another entry for the CMAKE_BUILD_TYPE. We also need to add the
488
# standerd set of 4 build types (Debug, MinSizeRel, Release, and RelWithDebInfo)
489
# for completeness. We need run this loop in order to accomodate the addition
490
# of extra configuration types. Duplicate entries will be removed by
492
set(CUDA_configuration_types ${CMAKE_CONFIGURATION_TYPES} ${CMAKE_BUILD_TYPE} Debug MinSizeRel Release RelWithDebInfo)
493
list(REMOVE_DUPLICATES CUDA_configuration_types)
494
foreach(config ${CUDA_configuration_types})
495
string(TOUPPER ${config} config_upper)
496
set(CUDA_NVCC_FLAGS_${config_upper} "" CACHE STRING "Semi-colon delimit multiple arguments.")
497
mark_as_advanced(CUDA_NVCC_FLAGS_${config_upper})
500
###############################################################################
501
###############################################################################
502
# Locate CUDA, Set Build Type, etc.
503
###############################################################################
504
###############################################################################
506
macro(cuda_unset_include_and_libraries)
507
unset(CUDA_TOOLKIT_INCLUDE CACHE)
508
unset(CUDA_CUDART_LIBRARY CACHE)
509
unset(CUDA_CUDA_LIBRARY CACHE)
510
# Make sure you run this before you unset CUDA_VERSION.
511
if(CUDA_VERSION VERSION_EQUAL "3.0")
512
# This only existed in the 3.0 version of the CUDA toolkit
513
unset(CUDA_CUDARTEMU_LIBRARY CACHE)
515
unset(CUDA_cupti_LIBRARY CACHE)
516
unset(CUDA_cublas_LIBRARY CACHE)
517
unset(CUDA_cublasemu_LIBRARY CACHE)
518
unset(CUDA_cufft_LIBRARY CACHE)
519
unset(CUDA_cufftemu_LIBRARY CACHE)
520
unset(CUDA_curand_LIBRARY CACHE)
521
unset(CUDA_cusparse_LIBRARY CACHE)
522
unset(CUDA_npp_LIBRARY CACHE)
523
unset(CUDA_nppc_LIBRARY CACHE)
524
unset(CUDA_nppi_LIBRARY CACHE)
525
unset(CUDA_npps_LIBRARY CACHE)
526
unset(CUDA_nvcuvenc_LIBRARY CACHE)
527
unset(CUDA_nvcuvid_LIBRARY CACHE)
530
# Check to see if the CUDA_TOOLKIT_ROOT_DIR and CUDA_SDK_ROOT_DIR have changed,
531
# if they have then clear the cache variables, so that will be detected again.
532
if(DEFINED CUDA_TOOLKIT_ROOT_DIR_INTERNAL AND (NOT "${CUDA_TOOLKIT_ROOT_DIR}" STREQUAL "${CUDA_TOOLKIT_ROOT_DIR_INTERNAL}"))
533
unset(CUDA_TARGET_TRIPLET CACHE)
534
unset(CUDA_TOOLKIT_TARGET_DIR CACHE)
535
unset(CUDA_NVCC_EXECUTABLE CACHE)
536
unset(CUDA_VERSION CACHE)
537
cuda_unset_include_and_libraries()
540
if(DEFINED CUDA_TARGET_TRIPLET_INTERNAL AND (NOT "${CUDA_TARGET_TRIPLET}" STREQUAL "${CUDA_TARGET_TRIPLET_INTERNAL}") OR
541
(DEFINED CUDA_TOOLKIT_TARGET_DIR AND DEFINED CUDA_TOOLKIT_TARGET_DIR_INTERNAL AND NOT "${CUDA_TOOLKIT_TARGET_DIR}" STREQUAL "${CUDA_TOOLKIT_TARGET_DIR_INTERNAL}"))
542
cuda_unset_include_and_libraries()
545
if(NOT "${CUDA_SDK_ROOT_DIR}" STREQUAL "${CUDA_SDK_ROOT_DIR_INTERNAL}")
546
# No specific variables to catch. Use this kind of code before calling
547
# find_package(CUDA) to clean up any variables that may depend on this path.
549
# unset(MY_SPECIAL_CUDA_SDK_INCLUDE_DIR CACHE)
550
# unset(MY_SPECIAL_CUDA_SDK_LIBRARY CACHE)
553
# Search for the cuda distribution.
554
if(NOT CUDA_TOOLKIT_ROOT_DIR)
556
# Search in the CUDA_BIN_PATH first.
557
find_path(CUDA_TOOLKIT_ROOT_DIR
562
PATH_SUFFIXES bin bin64
563
DOC "Toolkit location."
566
# Now search default paths
567
find_path(CUDA_TOOLKIT_ROOT_DIR
571
DOC "Toolkit location."
574
if (CUDA_TOOLKIT_ROOT_DIR)
575
string(REGEX REPLACE "[/\\\\]?bin[64]*[/\\\\]?$" "" CUDA_TOOLKIT_ROOT_DIR ${CUDA_TOOLKIT_ROOT_DIR})
576
# We need to force this back into the cache.
577
set(CUDA_TOOLKIT_ROOT_DIR ${CUDA_TOOLKIT_ROOT_DIR} CACHE PATH "Toolkit location." FORCE)
579
if (NOT EXISTS ${CUDA_TOOLKIT_ROOT_DIR})
580
if(CUDA_FIND_REQUIRED)
581
message(FATAL_ERROR "Specify CUDA_TOOLKIT_ROOT_DIR")
582
elseif(NOT CUDA_FIND_QUIETLY)
583
message("CUDA_TOOLKIT_ROOT_DIR not found or specified")
588
# CUDA_NVCC_EXECUTABLE
589
find_program(CUDA_NVCC_EXECUTABLE
591
PATHS "${CUDA_TOOLKIT_ROOT_DIR}"
594
PATH_SUFFIXES bin bin64
597
# Search default search paths, after we search our own set of paths.
598
find_program(CUDA_NVCC_EXECUTABLE nvcc)
599
mark_as_advanced(CUDA_NVCC_EXECUTABLE)
601
if(CUDA_NVCC_EXECUTABLE AND NOT CUDA_VERSION)
602
# Compute the version.
603
execute_process (COMMAND ${CUDA_NVCC_EXECUTABLE} "--version" OUTPUT_VARIABLE NVCC_OUT)
604
string(REGEX REPLACE ".*release ([0-9]+)\\.([0-9]+).*" "\\1" CUDA_VERSION_MAJOR ${NVCC_OUT})
605
string(REGEX REPLACE ".*release ([0-9]+)\\.([0-9]+).*" "\\2" CUDA_VERSION_MINOR ${NVCC_OUT})
606
set(CUDA_VERSION "${CUDA_VERSION_MAJOR}.${CUDA_VERSION_MINOR}" CACHE STRING "Version of CUDA as computed from nvcc.")
607
mark_as_advanced(CUDA_VERSION)
609
# Need to set these based off of the cached value
610
string(REGEX REPLACE "([0-9]+)\\.([0-9]+).*" "\\1" CUDA_VERSION_MAJOR "${CUDA_VERSION}")
611
string(REGEX REPLACE "([0-9]+)\\.([0-9]+).*" "\\2" CUDA_VERSION_MINOR "${CUDA_VERSION}")
614
# Always set this convenience variable
615
set(CUDA_VERSION_STRING "${CUDA_VERSION}")
617
# Target CPU architecture
618
if(DEFINED CUDA_TARGET_CPU_ARCH)
619
set(_cuda_target_cpu_arch_initial "${CUDA_TARGET_CPU_ARCH}")
620
elseif(CUDA_VERSION VERSION_GREATER "5.0" AND CMAKE_CROSSCOMPILING AND CMAKE_SYSTEM_PROCESSOR MATCHES "^(arm|ARM)")
621
set(_cuda_target_cpu_arch_initial "ARM")
622
elseif(CUDA_VERSION VERSION_GREATER "6.5" AND CMAKE_CROSSCOMPILING AND CMAKE_SYSTEM_PROCESSOR MATCHES "^(aarch64|AARCH64)")
623
set(_cuda_target_cpu_arch_initial "AARCH64")
625
set(_cuda_target_cpu_arch_initial "")
627
set(CUDA_TARGET_CPU_ARCH "${_cuda_target_cpu_arch_initial}" CACHE STRING "Specify the name of the class of CPU architecture for which the input files must be compiled.")
628
mark_as_advanced(CUDA_TARGET_CPU_ARCH)
631
if(DEFINED CUDA_TARGET_OS_VARIANT)
632
set(_cuda_target_os_variant_initial "${CUDA_TARGET_OS_VARIANT}")
634
set(_cuda_target_os_variant_initial "")
636
set(CUDA_TARGET_OS_VARIANT "${_cuda_target_os_variant_initial}" CACHE STRING "Specify the name of the class of OS for which the input files must be compiled.")
637
mark_as_advanced(CUDA_TARGET_OS_VARIANT)
640
if(DEFINED CUDA_TARGET_TRIPLET)
641
set(_cuda_target_triplet_initial "${CUDA_TARGET_TRIPLET}")
642
elseif(CUDA_VERSION VERSION_GREATER "5.0" AND CMAKE_CROSSCOMPILING AND "x${CUDA_TARGET_CPU_ARCH}" STREQUAL "xARM")
643
if("${CUDA_TARGET_OS_VARIANT}" STREQUAL "Android" AND EXISTS "${CUDA_TOOLKIT_ROOT_DIR}/targets/armv7-linux-androideabi")
644
set(_cuda_target_triplet_initial "armv7-linux-androideabi")
645
elseif(EXISTS "${CUDA_TOOLKIT_ROOT_DIR}/targets/armv7-linux-gnueabihf")
646
set(_cuda_target_triplet_initial "armv7-linux-gnueabihf")
648
elseif(CUDA_VERSION VERSION_GREATER "6.5" AND CMAKE_CROSSCOMPILING AND "x${CUDA_TARGET_CPU_ARCH}" STREQUAL "xAARCH64")
649
if("${CUDA_TARGET_OS_VARIANT}" STREQUAL "Android" AND EXISTS "${CUDA_TOOLKIT_ROOT_DIR}/targets/aarch64-linux-androideabi")
650
set(_cuda_target_triplet_initial "aarch64-linux-androideabi")
651
elseif(EXISTS "${CUDA_TOOLKIT_ROOT_DIR}/targets/aarch64-linux-gnueabihf")
652
set(_cuda_target_triplet_initial "aarch64-linux-gnueabihf")
655
set(CUDA_TARGET_TRIPLET "${_cuda_target_triplet_initial}" CACHE STRING "Specify the target triplet for which the input files must be compiled.")
656
file(GLOB __cuda_available_target_tiplets RELATIVE "${CUDA_TOOLKIT_ROOT_DIR}/targets" "${CUDA_TOOLKIT_ROOT_DIR}/targets/*" )
657
set_property(CACHE CUDA_TARGET_TRIPLET PROPERTY STRINGS ${__cuda_available_target_tiplets})
658
mark_as_advanced(CUDA_TARGET_TRIPLET)
661
if(NOT DEFINED CUDA_TOOLKIT_TARGET_DIR AND CUDA_TARGET_TRIPLET AND EXISTS "${CUDA_TOOLKIT_ROOT_DIR}/targets/${CUDA_TARGET_TRIPLET}")
662
set(CUDA_TOOLKIT_TARGET_DIR "${CUDA_TOOLKIT_ROOT_DIR}/targets/${CUDA_TARGET_TRIPLET}")
665
# CUDA_TOOLKIT_INCLUDE
666
find_path(CUDA_TOOLKIT_INCLUDE
667
device_functions.h # Header included in toolkit
668
PATHS "${CUDA_TOOLKIT_TARGET_DIR}" "${CUDA_TOOLKIT_ROOT_DIR}"
671
PATH_SUFFIXES include
674
# Search default search paths, after we search our own set of paths.
675
find_path(CUDA_TOOLKIT_INCLUDE device_functions.h)
676
mark_as_advanced(CUDA_TOOLKIT_INCLUDE)
678
# Set the user list of include dir to nothing to initialize it.
679
set (CUDA_NVCC_INCLUDE_ARGS_USER "")
680
set (CUDA_INCLUDE_DIRS ${CUDA_TOOLKIT_INCLUDE})
682
macro(cuda_find_library_local_first_with_path_ext _var _names _doc _path_ext )
683
if(CMAKE_SIZEOF_VOID_P EQUAL 8)
684
# CUDA 3.2+ on Windows moved the library directories, so we need the new
686
set(_cuda_64bit_lib_dir "${_path_ext}lib/x64" "${_path_ext}lib64" "${_path_ext}libx64" )
688
if(CMAKE_CROSSCOMPILING AND (ARM OR AARCH64))
689
set(_cuda_cross_arm_lib_dir "${_path_ext}lib/stubs")
691
if(CUDA_VERSION VERSION_GREATER "6.0")
692
set(_cuda_static_lib_names "")
693
foreach(name ${_names})
694
list(APPEND _cuda_static_lib_names "${name}_static")
697
# CUDA 3.2+ on Windows moved the library directories, so we need to new
698
# (lib/Win32) and the old path (lib).
700
NAMES ${_names} ${_cuda_static_lib_names}
701
PATHS "${CUDA_TOOLKIT_TARGET_DIR}" "${CUDA_TOOLKIT_ROOT_DIR}"
704
PATH_SUFFIXES ${_cuda_64bit_lib_dir} ${_cuda_cross_arm_lib_dir} "${_path_ext}lib/Win32" "${_path_ext}lib" "${_path_ext}libWin32"
708
# Search default search paths, after we search our own set of paths.
710
NAMES ${_names} ${_cuda_static_lib_names}
711
PATHS "/usr/lib/nvidia-current"
716
macro(cuda_find_library_local_first _var _names _doc)
717
cuda_find_library_local_first_with_path_ext( "${_var}" "${_names}" "${_doc}" "" )
720
macro(find_library_local_first _var _names _doc )
721
cuda_find_library_local_first( "${_var}" "${_names}" "${_doc}" "" )
726
cuda_find_library_local_first(CUDA_CUDART_LIBRARY cudart "\"cudart\" library")
727
if(CUDA_VERSION VERSION_EQUAL "3.0")
728
# The cudartemu library only existed for the 3.0 version of CUDA.
729
cuda_find_library_local_first(CUDA_CUDARTEMU_LIBRARY cudartemu "\"cudartemu\" library")
731
CUDA_CUDARTEMU_LIBRARY
735
# CUPTI library showed up in cuda toolkit 4.0
736
if(NOT CUDA_VERSION VERSION_LESS "4.0")
737
cuda_find_library_local_first_with_path_ext(CUDA_cupti_LIBRARY cupti "\"cupti\" library" "extras/CUPTI/")
738
mark_as_advanced(CUDA_cupti_LIBRARY)
741
# If we are using emulation mode and we found the cudartemu library then use
742
# that one instead of cudart.
743
if(CUDA_BUILD_EMULATION AND CUDA_CUDARTEMU_LIBRARY)
744
set(CUDA_LIBRARIES ${CUDA_CUDARTEMU_LIBRARY})
746
set(CUDA_LIBRARIES ${CUDA_CUDART_LIBRARY})
749
# 1.1 toolkit on linux doesn't appear to have a separate library on
751
cuda_find_library_local_first(CUDA_CUDA_LIBRARY cuda "\"cuda\" library (older versions only).")
758
#######################
759
# Look for some of the toolkit helper libraries
760
macro(FIND_CUDA_HELPER_LIBS _name)
761
cuda_find_library_local_first(CUDA_${_name}_LIBRARY ${_name} "\"${_name}\" library")
762
mark_as_advanced(CUDA_${_name}_LIBRARY)
765
#######################
766
# Disable emulation for v3.1 onward
767
if(CUDA_VERSION VERSION_GREATER "3.0")
768
if(CUDA_BUILD_EMULATION)
769
message(FATAL_ERROR "CUDA_BUILD_EMULATION is not supported in version 3.1 and onwards. You must disable it to proceed. You have version ${CUDA_VERSION}.")
773
# Search for additional CUDA toolkit libraries.
774
if(CUDA_VERSION VERSION_LESS "3.1")
775
# Emulation libraries aren't available in version 3.1 onward.
776
find_cuda_helper_libs(cufftemu)
777
find_cuda_helper_libs(cublasemu)
779
find_cuda_helper_libs(cufft)
780
find_cuda_helper_libs(cublas)
781
if(NOT CUDA_VERSION VERSION_LESS "3.2")
782
# cusparse showed up in version 3.2
783
find_cuda_helper_libs(cusparse)
784
find_cuda_helper_libs(curand)
786
find_cuda_helper_libs(nvcuvenc)
787
find_cuda_helper_libs(nvcuvid)
790
if(CUDA_VERSION VERSION_GREATER "5.0")
791
# In CUDA 5.5 NPP was splitted onto 3 separate libraries.
792
find_cuda_helper_libs(nppc)
793
find_cuda_helper_libs(nppi)
794
find_cuda_helper_libs(npps)
795
set(CUDA_npp_LIBRARY "${CUDA_nppc_LIBRARY};${CUDA_nppi_LIBRARY};${CUDA_npps_LIBRARY}")
796
elseif(NOT CUDA_VERSION VERSION_LESS "4.0")
797
find_cuda_helper_libs(npp)
800
if (CUDA_BUILD_EMULATION)
801
set(CUDA_CUFFT_LIBRARIES ${CUDA_cufftemu_LIBRARY})
802
set(CUDA_CUBLAS_LIBRARIES ${CUDA_cublasemu_LIBRARY})
804
set(CUDA_CUFFT_LIBRARIES ${CUDA_cufft_LIBRARY})
805
set(CUDA_CUBLAS_LIBRARIES ${CUDA_cublas_LIBRARY})
808
########################
809
# Look for the SDK stuff. As of CUDA 3.0 NVSDKCUDA_ROOT has been replaced with
810
# NVSDKCOMPUTE_ROOT with the old CUDA C contents moved into the C subdirectory
811
find_path(CUDA_SDK_ROOT_DIR common/inc/cutil.h
813
"$ENV{NVSDKCOMPUTE_ROOT}/C"
815
"[HKEY_LOCAL_MACHINE\\SOFTWARE\\NVIDIA Corporation\\Installed Products\\NVIDIA SDK 10\\Compute;InstallDir]"
817
"/Developer/GPU\ Computing/C"
820
# Keep the CUDA_SDK_ROOT_DIR first in order to be able to override the
821
# environment variables.
822
set(CUDA_SDK_SEARCH_PATH
823
"${CUDA_SDK_ROOT_DIR}"
824
"${CUDA_TOOLKIT_ROOT_DIR}/local/NVSDK0.2"
825
"${CUDA_TOOLKIT_ROOT_DIR}/NVSDK0.2"
826
"${CUDA_TOOLKIT_ROOT_DIR}/NV_CUDA_SDK"
827
"$ENV{HOME}/NVIDIA_CUDA_SDK"
828
"$ENV{HOME}/NVIDIA_CUDA_SDK_MACOSX"
832
# Example of how to find an include file from the CUDA_SDK_ROOT_DIR
834
# find_path(CUDA_CUT_INCLUDE_DIR
836
# PATHS ${CUDA_SDK_SEARCH_PATH}
837
# PATH_SUFFIXES "common/inc"
838
# DOC "Location of cutil.h"
841
# # Now search system paths
842
# find_path(CUDA_CUT_INCLUDE_DIR cutil.h DOC "Location of cutil.h")
844
# mark_as_advanced(CUDA_CUT_INCLUDE_DIR)
847
# Example of how to find a library in the CUDA_SDK_ROOT_DIR
849
# # cutil library is called cutil64 for 64 bit builds on windows. We don't want
850
# # to get these confused, so we are setting the name based on the word size of
853
# if(CMAKE_SIZEOF_VOID_P EQUAL 8)
854
# set(cuda_cutil_name cutil64)
856
# set(cuda_cutil_name cutil32)
859
# find_library(CUDA_CUT_LIBRARY
860
# NAMES cutil ${cuda_cutil_name}
861
# PATHS ${CUDA_SDK_SEARCH_PATH}
862
# # The new version of the sdk shows up in common/lib, but the old one is in lib
863
# PATH_SUFFIXES "common/lib" "lib"
864
# DOC "Location of cutil library"
867
# # Now search system paths
868
# find_library(CUDA_CUT_LIBRARY NAMES cutil ${cuda_cutil_name} DOC "Location of cutil library")
869
# mark_as_advanced(CUDA_CUT_LIBRARY)
870
# set(CUDA_CUT_LIBRARIES ${CUDA_CUT_LIBRARY})
874
#############################
875
# Check for required components
878
set(CUDA_TOOLKIT_ROOT_DIR_INTERNAL "${CUDA_TOOLKIT_ROOT_DIR}" CACHE INTERNAL
879
"This is the value of the last time CUDA_TOOLKIT_ROOT_DIR was set successfully." FORCE)
880
set(CUDA_TARGET_TRIPLET_INTERNAL "${CUDA_TARGET_TRIPLET}" CACHE INTERNAL
881
"This is the value of the last time CUDA_TARGET_TRIPLET was set successfully." FORCE)
882
set(CUDA_TOOLKIT_TARGET_DIR_INTERNAL "${CUDA_TOOLKIT_TARGET_DIR}" CACHE INTERNAL
883
"This is the value of the last time CUDA_TOOLKIT_TARGET_DIR was set successfully." FORCE)
884
set(CUDA_SDK_ROOT_DIR_INTERNAL "${CUDA_SDK_ROOT_DIR}" CACHE INTERNAL
885
"This is the value of the last time CUDA_SDK_ROOT_DIR was set successfully." FORCE)
887
include(FindPackageHandleStandardArgs)
888
find_package_handle_standard_args(CUDA
890
CUDA_TOOLKIT_ROOT_DIR
900
###############################################################################
901
###############################################################################
903
###############################################################################
904
###############################################################################
906
###############################################################################
907
# Add include directories to pass to the nvcc command.
908
macro(CUDA_INCLUDE_DIRECTORIES)
910
list(APPEND CUDA_NVCC_INCLUDE_ARGS_USER -I${dir})
915
##############################################################################
916
cuda_find_helper_file(parse_cubin cmake)
917
cuda_find_helper_file(make2cmake cmake)
918
cuda_find_helper_file(run_nvcc cmake)
920
##############################################################################
921
# Separate the OPTIONS out from the sources
923
macro(CUDA_GET_SOURCES_AND_OPTIONS _sources _cmake_options _options)
925
set( ${_cmake_options} )
927
set( _found_options FALSE )
929
if("x${arg}" STREQUAL "xOPTIONS")
930
set( _found_options TRUE )
932
"x${arg}" STREQUAL "xWIN32" OR
933
"x${arg}" STREQUAL "xMACOSX_BUNDLE" OR
934
"x${arg}" STREQUAL "xEXCLUDE_FROM_ALL" OR
935
"x${arg}" STREQUAL "xSTATIC" OR
936
"x${arg}" STREQUAL "xSHARED" OR
937
"x${arg}" STREQUAL "xMODULE"
939
list(APPEND ${_cmake_options} ${arg})
941
if ( _found_options )
942
list(APPEND ${_options} ${arg})
944
# Assume this is a file
945
list(APPEND ${_sources} ${arg})
951
##############################################################################
952
# Parse the OPTIONS from ARGN and set the variables prefixed by _option_prefix
954
macro(CUDA_PARSE_NVCC_OPTIONS _option_prefix)
957
# Determine if we are dealing with a perconfiguration flag
958
foreach(config ${CUDA_configuration_types})
959
string(TOUPPER ${config} config_upper)
960
if (arg STREQUAL "${config_upper}")
961
set( _found_config _${arg})
962
# Set arg to nothing to keep it from being processed further
968
list(APPEND ${_option_prefix}${_found_config} "${arg}")
973
##############################################################################
974
# Helper to add the include directory for CUDA only once
975
function(CUDA_ADD_CUDA_INCLUDE_ONCE)
976
get_directory_property(_include_directories INCLUDE_DIRECTORIES)
978
if(_include_directories)
979
foreach(dir ${_include_directories})
980
if("${dir}" STREQUAL "${CUDA_INCLUDE_DIRS}")
986
include_directories(${CUDA_INCLUDE_DIRS})
990
function(CUDA_BUILD_SHARED_LIBRARY shared_flag)
991
set(cmake_args ${ARGN})
992
# If SHARED, MODULE, or STATIC aren't already in the list of arguments, then
993
# add SHARED or STATIC based on the value of BUILD_SHARED_LIBS.
994
list(FIND cmake_args SHARED _cuda_found_SHARED)
995
list(FIND cmake_args MODULE _cuda_found_MODULE)
996
list(FIND cmake_args STATIC _cuda_found_STATIC)
997
if( _cuda_found_SHARED GREATER -1 OR
998
_cuda_found_MODULE GREATER -1 OR
999
_cuda_found_STATIC GREATER -1)
1000
set(_cuda_build_shared_libs)
1002
if (BUILD_SHARED_LIBS)
1003
set(_cuda_build_shared_libs SHARED)
1005
set(_cuda_build_shared_libs STATIC)
1008
set(${shared_flag} ${_cuda_build_shared_libs} PARENT_SCOPE)
1011
##############################################################################
1012
# Helper to avoid clashes of files with the same basename but different paths.
1013
# This doesn't attempt to do exactly what CMake internals do, which is to only
1014
# add this path when there is a conflict, since by the time a second collision
1015
# in names is detected it's already too late to fix the first one. For
1016
# consistency sake the relative path will be added to all files.
1017
function(CUDA_COMPUTE_BUILD_PATH path build_path)
1018
#message("CUDA_COMPUTE_BUILD_PATH([${path}] ${build_path})")
1019
# Only deal with CMake style paths from here on out
1020
file(TO_CMAKE_PATH "${path}" bpath)
1021
if (IS_ABSOLUTE "${bpath}")
1022
# Absolute paths are generally unnessary, especially if something like
1023
# file(GLOB_RECURSE) is used to pick up the files.
1025
string(FIND "${bpath}" "${CMAKE_CURRENT_BINARY_DIR}" _binary_dir_pos)
1026
if (_binary_dir_pos EQUAL 0)
1027
file(RELATIVE_PATH bpath "${CMAKE_CURRENT_BINARY_DIR}" "${bpath}")
1029
file(RELATIVE_PATH bpath "${CMAKE_CURRENT_SOURCE_DIR}" "${bpath}")
1033
# This recipe is from cmLocalGenerator::CreateSafeUniqueObjectFileName in the
1037
string(REGEX REPLACE "^[/]+" "" bpath "${bpath}")
1038
# Avoid absolute paths by removing ':'
1039
string(REPLACE ":" "_" bpath "${bpath}")
1040
# Avoid relative paths that go up the tree
1041
string(REPLACE "../" "__/" bpath "${bpath}")
1043
string(REPLACE " " "_" bpath "${bpath}")
1045
# Strip off the filename. I wait until here to do it, since removin the
1046
# basename can make a path that looked like path/../basename turn into
1047
# path/.. (notice the trailing slash).
1048
get_filename_component(bpath "${bpath}" PATH)
1050
set(${build_path} "${bpath}" PARENT_SCOPE)
1051
#message("${build_path} = ${bpath}")
1054
##############################################################################
1055
# This helper macro populates the following variables and setups up custom
1056
# commands and targets to invoke the nvcc compiler to generate C or PTX source
1057
# dependent upon the format parameter. The compiler is invoked once with -M
1058
# to generate a dependency file and a second time with -cuda or -ptx to generate
1059
# a .cpp or .ptx file.
1061
# cuda_target - Target name
1062
# format - PTX, CUBIN, FATBIN or OBJ
1063
# FILE1 .. FILEN - The remaining arguments are the sources to be wrapped.
1064
# OPTIONS - Extra options to NVCC
1066
# generated_files - List of generated files
1067
##############################################################################
1068
##############################################################################
1070
macro(CUDA_WRAP_SRCS cuda_target format generated_files)
1072
# If CMake doesn't support separable compilation, complain
1073
if(CUDA_SEPARABLE_COMPILATION AND CMAKE_VERSION VERSION_LESS "2.8.10.1")
1074
message(SEND_ERROR "CUDA_SEPARABLE_COMPILATION isn't supported for CMake versions less than 2.8.10.1")
1077
# Set up all the command line flags here, so that they can be overridden on a per target basis.
1081
# Emulation if the card isn't present.
1082
if (CUDA_BUILD_EMULATION)
1084
set(nvcc_flags ${nvcc_flags} --device-emulation -D_DEVICEEMU -g)
1086
# Device mode. No flags necessary.
1089
if(CUDA_HOST_COMPILATION_CPP)
1090
set(CUDA_C_OR_CXX CXX)
1092
if(CUDA_VERSION VERSION_LESS "3.0")
1093
set(nvcc_flags ${nvcc_flags} --host-compilation C)
1095
message(WARNING "--host-compilation flag is deprecated in CUDA version >= 3.0. Removing --host-compilation C flag" )
1097
set(CUDA_C_OR_CXX C)
1100
set(generated_extension ${CMAKE_${CUDA_C_OR_CXX}_OUTPUT_EXTENSION})
1102
if(CUDA_64_BIT_DEVICE_CODE)
1103
set(nvcc_flags ${nvcc_flags} -m64)
1105
set(nvcc_flags ${nvcc_flags} -m32)
1108
if(CUDA_TARGET_CPU_ARCH AND CUDA_VERSION VERSION_LESS "7.0")
1109
# CPU architecture is either ARM or X86. Patch AARCH64 to be ARM
1110
string(REPLACE "AARCH64" "ARM" CUDA_TARGET_CPU_ARCH_patched ${CUDA_TARGET_CPU_ARCH})
1111
set(nvcc_flags ${nvcc_flags} "--target-cpu-architecture=${CUDA_TARGET_CPU_ARCH_patched}")
1114
if(CUDA_TARGET_OS_VARIANT AND CUDA_VERSION VERSION_LESS "7.0")
1115
set(nvcc_flags ${nvcc_flags} "-target-os-variant=${CUDA_TARGET_OS_VARIANT}")
1118
# This needs to be passed in at this stage, because VS needs to fill out the
1119
# value of VCInstallDir from within VS. Note that CCBIN is only used if
1120
# -ccbin or --compiler-bindir isn't used and CUDA_HOST_COMPILER matches
1121
# $(VCInstallDir)/bin.
1122
if(CMAKE_GENERATOR MATCHES "Visual Studio")
1123
set(ccbin_flags -D "\"CCBIN:PATH=$(VCInstallDir)bin\"" )
1128
# Figure out which configure we will use and pass that in as an argument to
1129
# the script. We need to defer the decision until compilation time, because
1130
# for VS projects we won't know if we are making a debug or release build
1132
if(CMAKE_GENERATOR MATCHES "Visual Studio")
1133
set( CUDA_build_configuration "$(ConfigurationName)" )
1135
set( CUDA_build_configuration "${CMAKE_BUILD_TYPE}")
1138
# Initialize our list of includes with the user ones followed by the CUDA system ones.
1139
set(CUDA_NVCC_INCLUDE_ARGS ${CUDA_NVCC_INCLUDE_ARGS_USER} "-I${CUDA_INCLUDE_DIRS}")
1140
# Get the include directories for this directory and use them for our nvcc command.
1141
# Remove duplicate entries which may be present since include_directories
1142
# in CMake >= 2.8.8 does not remove them.
1143
get_directory_property(CUDA_NVCC_INCLUDE_DIRECTORIES INCLUDE_DIRECTORIES)
1144
list(REMOVE_DUPLICATES CUDA_NVCC_INCLUDE_DIRECTORIES)
1145
if(CUDA_NVCC_INCLUDE_DIRECTORIES)
1146
foreach(dir ${CUDA_NVCC_INCLUDE_DIRECTORIES})
1147
list(APPEND CUDA_NVCC_INCLUDE_ARGS -I${dir})
1151
# Reset these variables
1152
set(CUDA_WRAP_OPTION_NVCC_FLAGS)
1153
foreach(config ${CUDA_configuration_types})
1154
string(TOUPPER ${config} config_upper)
1155
set(CUDA_WRAP_OPTION_NVCC_FLAGS_${config_upper})
1158
CUDA_GET_SOURCES_AND_OPTIONS(_cuda_wrap_sources _cuda_wrap_cmake_options _cuda_wrap_options ${ARGN})
1159
CUDA_PARSE_NVCC_OPTIONS(CUDA_WRAP_OPTION_NVCC_FLAGS ${_cuda_wrap_options})
1161
# Figure out if we are building a shared library. BUILD_SHARED_LIBS is
1162
# respected in CUDA_ADD_LIBRARY.
1163
set(_cuda_build_shared_libs FALSE)
1165
list(FIND _cuda_wrap_cmake_options SHARED _cuda_found_SHARED)
1166
list(FIND _cuda_wrap_cmake_options MODULE _cuda_found_MODULE)
1167
if(_cuda_found_SHARED GREATER -1 OR _cuda_found_MODULE GREATER -1)
1168
set(_cuda_build_shared_libs TRUE)
1171
list(FIND _cuda_wrap_cmake_options STATIC _cuda_found_STATIC)
1172
if(_cuda_found_STATIC GREATER -1)
1173
set(_cuda_build_shared_libs FALSE)
1177
if(_cuda_build_shared_libs)
1178
# If we are setting up code for a shared library, then we need to add extra flags for
1179
# compiling objects for shared libraries.
1180
set(CUDA_HOST_SHARED_FLAGS ${CMAKE_SHARED_LIBRARY_${CUDA_C_OR_CXX}_FLAGS})
1182
set(CUDA_HOST_SHARED_FLAGS)
1184
# Only add the CMAKE_{C,CXX}_FLAGS if we are propagating host flags. We
1185
# always need to set the SHARED_FLAGS, though.
1186
if(CUDA_PROPAGATE_HOST_FLAGS)
1187
set(_cuda_host_flags "set(CMAKE_HOST_FLAGS ${CMAKE_${CUDA_C_OR_CXX}_FLAGS} ${CUDA_HOST_SHARED_FLAGS})")
1189
set(_cuda_host_flags "set(CMAKE_HOST_FLAGS ${CUDA_HOST_SHARED_FLAGS})")
1192
set(_cuda_nvcc_flags_config "# Build specific configuration flags")
1193
# Loop over all the configuration types to generate appropriate flags for run_nvcc.cmake
1194
foreach(config ${CUDA_configuration_types})
1195
string(TOUPPER ${config} config_upper)
1196
# CMAKE_FLAGS are strings and not lists. By not putting quotes around CMAKE_FLAGS
1197
# we convert the strings to lists (like we want).
1199
if(CUDA_PROPAGATE_HOST_FLAGS)
1200
# nvcc chokes on -g3 in versions previous to 3.0, so replace it with -g
1201
set(_cuda_fix_g3 FALSE)
1203
if(CMAKE_COMPILER_IS_GNUCC)
1204
if (CUDA_VERSION VERSION_LESS "3.0" OR
1205
CUDA_VERSION VERSION_EQUAL "4.1" OR
1206
CUDA_VERSION VERSION_EQUAL "4.2"
1208
set(_cuda_fix_g3 TRUE)
1212
string(REPLACE "-g3" "-g" _cuda_C_FLAGS "${CMAKE_${CUDA_C_OR_CXX}_FLAGS_${config_upper}}")
1214
set(_cuda_C_FLAGS "${CMAKE_${CUDA_C_OR_CXX}_FLAGS_${config_upper}}")
1217
set(_cuda_host_flags "${_cuda_host_flags}\nset(CMAKE_HOST_FLAGS_${config_upper} ${_cuda_C_FLAGS})")
1220
# Note that if we ever want CUDA_NVCC_FLAGS_<CONFIG> to be string (instead of a list
1221
# like it is currently), we can remove the quotes around the
1222
# ${CUDA_NVCC_FLAGS_${config_upper}} variable like the CMAKE_HOST_FLAGS_<CONFIG> variable.
1223
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}})")
1226
# Get the list of definitions from the directory property
1227
get_directory_property(CUDA_NVCC_DEFINITIONS COMPILE_DEFINITIONS)
1228
if(CUDA_NVCC_DEFINITIONS)
1229
foreach(_definition ${CUDA_NVCC_DEFINITIONS})
1230
list(APPEND nvcc_flags "-D${_definition}")
1234
if(_cuda_build_shared_libs)
1235
list(APPEND nvcc_flags "-D${cuda_target}_EXPORTS")
1238
# Reset the output variable
1239
set(_cuda_wrap_generated_files "")
1241
# Iterate over the macro arguments and create custom
1242
# commands for all the .cu files.
1243
foreach(file ${ARGN})
1244
# Ignore any file marked as a HEADER_FILE_ONLY
1245
get_source_file_property(_is_header ${file} HEADER_FILE_ONLY)
1246
if(${file} MATCHES "\\.cu$" AND NOT _is_header)
1248
# Allow per source file overrides of the format.
1249
get_source_file_property(_cuda_source_format ${file} CUDA_SOURCE_PROPERTY_FORMAT)
1250
if(NOT _cuda_source_format)
1251
set(_cuda_source_format ${format})
1254
if( ${_cuda_source_format} MATCHES "OBJ")
1255
set( cuda_compile_to_external_module OFF )
1257
set( cuda_compile_to_external_module ON )
1258
if( ${_cuda_source_format} MATCHES "PTX" )
1259
set( cuda_compile_to_external_module_type "ptx" )
1260
elseif( ${_cuda_source_format} MATCHES "CUBIN")
1261
set( cuda_compile_to_external_module_type "cubin" )
1262
elseif( ${_cuda_source_format} MATCHES "FATBIN")
1263
set( cuda_compile_to_external_module_type "fatbin" )
1265
message( FATAL_ERROR "Invalid format flag passed to CUDA_WRAP_SRCS for file '${file}': '${_cuda_source_format}'. Use OBJ, PTX, CUBIN or FATBIN.")
1269
if(cuda_compile_to_external_module)
1270
# Don't use any of the host compilation flags for PTX targets.
1271
set(CUDA_HOST_FLAGS)
1272
set(CUDA_NVCC_FLAGS_CONFIG)
1274
set(CUDA_HOST_FLAGS ${_cuda_host_flags})
1275
set(CUDA_NVCC_FLAGS_CONFIG ${_cuda_nvcc_flags_config})
1278
# Determine output directory
1279
cuda_compute_build_path("${file}" cuda_build_path)
1280
set(cuda_compile_intermediate_directory "${CMAKE_CURRENT_BINARY_DIR}/CMakeFiles/${cuda_target}.dir/${cuda_build_path}")
1281
if(CUDA_GENERATED_OUTPUT_DIR)
1282
set(cuda_compile_output_dir "${CUDA_GENERATED_OUTPUT_DIR}")
1284
if ( cuda_compile_to_external_module )
1285
set(cuda_compile_output_dir "${CMAKE_CURRENT_BINARY_DIR}")
1287
set(cuda_compile_output_dir "${cuda_compile_intermediate_directory}")
1291
# Add a custom target to generate a c or ptx file. ######################
1293
get_filename_component( basename ${file} NAME )
1294
if( cuda_compile_to_external_module )
1295
set(generated_file_path "${cuda_compile_output_dir}")
1296
set(generated_file_basename "${cuda_target}_generated_${basename}.${cuda_compile_to_external_module_type}")
1297
set(format_flag "-${cuda_compile_to_external_module_type}")
1298
file(MAKE_DIRECTORY "${cuda_compile_output_dir}")
1300
set(generated_file_path "${cuda_compile_output_dir}/${CMAKE_CFG_INTDIR}")
1301
set(generated_file_basename "${cuda_target}_generated_${basename}${generated_extension}")
1302
if(CUDA_SEPARABLE_COMPILATION)
1303
set(format_flag "-dc")
1305
set(format_flag "-c")
1309
# Set all of our file names. Make sure that whatever filenames that have
1310
# generated_file_path in them get passed in through as a command line
1311
# argument, so that the ${CMAKE_CFG_INTDIR} gets expanded at run time
1312
# instead of configure time.
1313
set(generated_file "${generated_file_path}/${generated_file_basename}")
1314
set(cmake_dependency_file "${cuda_compile_intermediate_directory}/${generated_file_basename}.depend")
1315
set(NVCC_generated_dependency_file "${cuda_compile_intermediate_directory}/${generated_file_basename}.NVCC-depend")
1316
set(generated_cubin_file "${generated_file_path}/${generated_file_basename}.cubin.txt")
1317
set(custom_target_script "${cuda_compile_intermediate_directory}/${generated_file_basename}.cmake")
1319
# Setup properties for obj files:
1320
if( NOT cuda_compile_to_external_module )
1321
set_source_files_properties("${generated_file}"
1323
EXTERNAL_OBJECT true # This is an object file not to be compiled, but only be linked.
1327
# Don't add CMAKE_CURRENT_SOURCE_DIR if the path is already an absolute path.
1328
get_filename_component(file_path "${file}" PATH)
1329
if(IS_ABSOLUTE "${file_path}")
1330
set(source_file "${file}")
1332
set(source_file "${CMAKE_CURRENT_SOURCE_DIR}/${file}")
1335
if( NOT cuda_compile_to_external_module AND CUDA_SEPARABLE_COMPILATION)
1336
list(APPEND ${cuda_target}_SEPARABLE_COMPILATION_OBJECTS "${generated_file}")
1339
# Bring in the dependencies. Creates a variable CUDA_NVCC_DEPEND #######
1340
cuda_include_nvcc_dependencies(${cmake_dependency_file})
1342
# Convience string for output ###########################################
1343
if(CUDA_BUILD_EMULATION)
1344
set(cuda_build_type "Emulation")
1346
set(cuda_build_type "Device")
1349
# Build the NVCC made dependency file ###################################
1350
set(build_cubin OFF)
1351
if ( NOT CUDA_BUILD_EMULATION AND CUDA_BUILD_CUBIN )
1352
if ( NOT cuda_compile_to_external_module )
1353
set ( build_cubin ON )
1357
# Configure the build script
1358
configure_file("${CUDA_run_nvcc}" "${custom_target_script}" @ONLY)
1360
# So if a user specifies the same cuda file as input more than once, you
1361
# can have bad things happen with dependencies. Here we check an option
1362
# to see if this is the behavior they want.
1363
if(CUDA_ATTACH_VS_BUILD_RULE_TO_CUDA_FILE)
1364
set(main_dep MAIN_DEPENDENCY ${source_file})
1366
set(main_dep DEPENDS ${source_file})
1369
if(CUDA_VERBOSE_BUILD)
1370
set(verbose_output ON)
1371
elseif(CMAKE_GENERATOR MATCHES "Makefiles")
1372
set(verbose_output "$(VERBOSE)")
1374
set(verbose_output OFF)
1377
# Create up the comment string
1378
file(RELATIVE_PATH generated_file_relative_path "${CMAKE_BINARY_DIR}" "${generated_file}")
1379
if(cuda_compile_to_external_module)
1380
set(cuda_build_comment_string "Building NVCC ${cuda_compile_to_external_module_type} file ${generated_file_relative_path}")
1382
set(cuda_build_comment_string "Building NVCC (${cuda_build_type}) object ${generated_file_relative_path}")
1385
# Build the generated file and dependency file ##########################
1387
OUTPUT ${generated_file}
1388
# These output files depend on the source_file and the contents of cmake_dependency_file
1390
DEPENDS ${CUDA_NVCC_DEPEND}
1391
DEPENDS ${custom_target_script}
1392
# Make sure the output directory exists before trying to write to it.
1393
COMMAND ${CMAKE_COMMAND} -E make_directory "${generated_file_path}"
1394
COMMAND ${CMAKE_COMMAND} ARGS
1395
-D verbose:BOOL=${verbose_output}
1397
-D build_configuration:STRING=${CUDA_build_configuration}
1398
-D "generated_file:STRING=${generated_file}"
1399
-D "generated_cubin_file:STRING=${generated_cubin_file}"
1400
-P "${custom_target_script}"
1401
WORKING_DIRECTORY "${cuda_compile_intermediate_directory}"
1402
COMMENT "${cuda_build_comment_string}"
1405
# Make sure the build system knows the file is generated.
1406
set_source_files_properties(${generated_file} PROPERTIES GENERATED TRUE)
1408
list(APPEND _cuda_wrap_generated_files ${generated_file})
1410
# Add the other files that we want cmake to clean on a cleanup ##########
1411
list(APPEND CUDA_ADDITIONAL_CLEAN_FILES "${cmake_dependency_file}")
1412
list(REMOVE_DUPLICATES CUDA_ADDITIONAL_CLEAN_FILES)
1413
set(CUDA_ADDITIONAL_CLEAN_FILES ${CUDA_ADDITIONAL_CLEAN_FILES} CACHE INTERNAL "List of intermediate files that are part of the cuda dependency scanning.")
1418
# Set the return parameter
1419
set(${generated_files} ${_cuda_wrap_generated_files})
1422
function(_cuda_get_important_host_flags important_flags flag_string)
1423
if(CMAKE_GENERATOR MATCHES "Visual Studio")
1424
string(REGEX MATCHALL "/M[DT][d]?" flags ${flag_string})
1425
list(APPEND ${important_flags} ${flags})
1427
string(REGEX MATCHALL "-fPIC" flags ${flag_string})
1428
list(APPEND ${important_flags} ${flags})
1430
set(${important_flags} ${${important_flags}} PARENT_SCOPE)
1433
###############################################################################
1434
###############################################################################
1435
# Separable Compilation Link
1436
###############################################################################
1437
###############################################################################
1439
# Compute the filename to be used by CUDA_LINK_SEPARABLE_COMPILATION_OBJECTS
1440
function(CUDA_COMPUTE_SEPARABLE_COMPILATION_OBJECT_FILE_NAME output_file_var cuda_target object_files)
1442
set(generated_extension ${CMAKE_${CUDA_C_OR_CXX}_OUTPUT_EXTENSION})
1443
set(output_file "${CMAKE_CURRENT_BINARY_DIR}/CMakeFiles/${cuda_target}.dir/${CMAKE_CFG_INTDIR}/${cuda_target}_intermediate_link${generated_extension}")
1448
set(${output_file_var} "${output_file}" PARENT_SCOPE)
1451
# Setup the build rule for the separable compilation intermediate link file.
1452
function(CUDA_LINK_SEPARABLE_COMPILATION_OBJECTS output_file cuda_target options object_files)
1455
set_source_files_properties("${output_file}"
1457
EXTERNAL_OBJECT TRUE # This is an object file not to be compiled, but only
1459
GENERATED TRUE # This file is generated during the build
1462
# For now we are ignoring all the configuration specific flags.
1464
CUDA_PARSE_NVCC_OPTIONS(nvcc_flags ${options})
1465
if(CUDA_64_BIT_DEVICE_CODE)
1466
list(APPEND nvcc_flags -m64)
1468
list(APPEND nvcc_flags -m32)
1470
# If -ccbin, --compiler-bindir has been specified, don't do anything. Otherwise add it here.
1471
list( FIND nvcc_flags "-ccbin" ccbin_found0 )
1472
list( FIND nvcc_flags "--compiler-bindir" ccbin_found1 )
1473
if( ccbin_found0 LESS 0 AND ccbin_found1 LESS 0 AND CUDA_HOST_COMPILER )
1474
list(APPEND nvcc_flags -ccbin "\"${CUDA_HOST_COMPILER}\"")
1476
# Create a list of flags specified by CUDA_NVCC_FLAGS_${CONFIG}
1477
set(config_specific_flags)
1479
foreach(config ${CUDA_configuration_types})
1480
string(TOUPPER ${config} config_upper)
1481
# Add config specific flags
1482
foreach(f ${CUDA_NVCC_FLAGS_${config_upper}})
1483
list(APPEND config_specific_flags $<$<CONFIG:${config}>:${f}>)
1485
set(important_host_flags)
1486
_cuda_get_important_host_flags(important_host_flags ${CMAKE_${CUDA_C_OR_CXX}_FLAGS_${config_upper}})
1487
foreach(f ${important_host_flags})
1488
list(APPEND flags $<$<CONFIG:${config}>:-Xcompiler> $<$<CONFIG:${config}>:${f}>)
1491
# Add our general CUDA_NVCC_FLAGS with the configuration specifig flags
1492
set(nvcc_flags ${CUDA_NVCC_FLAGS} ${config_specific_flags} ${nvcc_flags})
1494
file(RELATIVE_PATH output_file_relative_path "${CMAKE_BINARY_DIR}" "${output_file}")
1496
# Some generators don't handle the multiple levels of custom command
1497
# dependencies correctly (obj1 depends on file1, obj2 depends on obj1), so
1498
# we work around that issue by compiling the intermediate link object as a
1499
# pre-link custom command in that situation.
1500
set(do_obj_build_rule TRUE)
1501
if (MSVC_VERSION GREATER 1599)
1502
# VS 2010 and 2012 have this problem. If future versions fix this issue,
1503
# it should still work, it just won't be as nice as the other method.
1504
set(do_obj_build_rule FALSE)
1507
if (do_obj_build_rule)
1509
OUTPUT ${output_file}
1510
DEPENDS ${object_files}
1511
COMMAND ${CUDA_NVCC_EXECUTABLE} ${nvcc_flags} -dlink ${object_files} -o ${output_file}
1513
COMMENT "Building NVCC intermediate link file ${output_file_relative_path}"
1517
TARGET ${cuda_target}
1519
COMMAND ${CMAKE_COMMAND} -E echo "Building NVCC intermediate link file ${output_file_relative_path}"
1520
COMMAND ${CUDA_NVCC_EXECUTABLE} ${nvcc_flags} ${flags} -dlink ${object_files} -o "${output_file}"
1526
###############################################################################
1527
###############################################################################
1529
###############################################################################
1530
###############################################################################
1531
macro(CUDA_ADD_LIBRARY cuda_target)
1533
CUDA_ADD_CUDA_INCLUDE_ONCE()
1535
# Separate the sources from the options
1536
CUDA_GET_SOURCES_AND_OPTIONS(_sources _cmake_options _options ${ARGN})
1537
CUDA_BUILD_SHARED_LIBRARY(_cuda_shared_flag ${ARGN})
1538
# Create custom commands and targets for each file.
1539
CUDA_WRAP_SRCS( ${cuda_target} OBJ _generated_files ${_sources}
1540
${_cmake_options} ${_cuda_shared_flag}
1541
OPTIONS ${_options} )
1543
# Compute the file name of the intermedate link file used for separable
1545
CUDA_COMPUTE_SEPARABLE_COMPILATION_OBJECT_FILE_NAME(link_file ${cuda_target} "${${cuda_target}_SEPARABLE_COMPILATION_OBJECTS}")
1548
add_library(${cuda_target} ${_cmake_options}
1554
# Add a link phase for the separable compilation if it has been enabled. If
1555
# it has been enabled then the ${cuda_target}_SEPARABLE_COMPILATION_OBJECTS
1556
# variable will have been defined.
1557
CUDA_LINK_SEPARABLE_COMPILATION_OBJECTS("${link_file}" ${cuda_target} "${_options}" "${${cuda_target}_SEPARABLE_COMPILATION_OBJECTS}")
1559
target_link_libraries(${cuda_target}
1563
# We need to set the linker language based on what the expected generated file
1564
# would be. CUDA_C_OR_CXX is computed based on CUDA_HOST_COMPILATION_CPP.
1565
set_target_properties(${cuda_target}
1567
LINKER_LANGUAGE ${CUDA_C_OR_CXX}
1573
###############################################################################
1574
###############################################################################
1576
###############################################################################
1577
###############################################################################
1578
macro(CUDA_ADD_EXECUTABLE cuda_target)
1580
CUDA_ADD_CUDA_INCLUDE_ONCE()
1582
# Separate the sources from the options
1583
CUDA_GET_SOURCES_AND_OPTIONS(_sources _cmake_options _options ${ARGN})
1584
# Create custom commands and targets for each file.
1585
CUDA_WRAP_SRCS( ${cuda_target} OBJ _generated_files ${_sources} OPTIONS ${_options} )
1587
# Compute the file name of the intermedate link file used for separable
1589
CUDA_COMPUTE_SEPARABLE_COMPILATION_OBJECT_FILE_NAME(link_file ${cuda_target} "${${cuda_target}_SEPARABLE_COMPILATION_OBJECTS}")
1592
add_executable(${cuda_target} ${_cmake_options}
1598
# Add a link phase for the separable compilation if it has been enabled. If
1599
# it has been enabled then the ${cuda_target}_SEPARABLE_COMPILATION_OBJECTS
1600
# variable will have been defined.
1601
CUDA_LINK_SEPARABLE_COMPILATION_OBJECTS("${link_file}" ${cuda_target} "${_options}" "${${cuda_target}_SEPARABLE_COMPILATION_OBJECTS}")
1603
target_link_libraries(${cuda_target}
1607
# We need to set the linker language based on what the expected generated file
1608
# would be. CUDA_C_OR_CXX is computed based on CUDA_HOST_COMPILATION_CPP.
1609
set_target_properties(${cuda_target}
1611
LINKER_LANGUAGE ${CUDA_C_OR_CXX}
1617
###############################################################################
1618
###############################################################################
1619
# (Internal) helper for manually added cuda source files with specific targets
1620
###############################################################################
1621
###############################################################################
1622
macro(cuda_compile_base cuda_target format generated_files)
1624
# Separate the sources from the options
1625
CUDA_GET_SOURCES_AND_OPTIONS(_sources _cmake_options _options ${ARGN})
1626
# Create custom commands and targets for each file.
1627
CUDA_WRAP_SRCS( ${cuda_target} ${format} _generated_files ${_sources} ${_cmake_options}
1628
OPTIONS ${_options} )
1630
set( ${generated_files} ${_generated_files})
1634
###############################################################################
1635
###############################################################################
1637
###############################################################################
1638
###############################################################################
1639
macro(CUDA_COMPILE generated_files)
1640
cuda_compile_base(cuda_compile OBJ ${generated_files} ${ARGN})
1643
###############################################################################
1644
###############################################################################
1646
###############################################################################
1647
###############################################################################
1648
macro(CUDA_COMPILE_PTX generated_files)
1649
cuda_compile_base(cuda_compile_ptx PTX ${generated_files} ${ARGN})
1652
###############################################################################
1653
###############################################################################
1654
# CUDA COMPILE FATBIN
1655
###############################################################################
1656
###############################################################################
1657
macro(CUDA_COMPILE_FATBIN generated_files)
1658
cuda_compile_base(cuda_compile_fatbin FATBIN ${generated_files} ${ARGN})
1661
###############################################################################
1662
###############################################################################
1663
# CUDA COMPILE CUBIN
1664
###############################################################################
1665
###############################################################################
1666
macro(CUDA_COMPILE_CUBIN generated_files)
1667
cuda_compile_base(cuda_compile_cubin CUBIN ${generated_files} ${ARGN})
1671
###############################################################################
1672
###############################################################################
1673
# CUDA ADD CUFFT TO TARGET
1674
###############################################################################
1675
###############################################################################
1676
macro(CUDA_ADD_CUFFT_TO_TARGET target)
1677
if (CUDA_BUILD_EMULATION)
1678
target_link_libraries(${target} ${CUDA_cufftemu_LIBRARY})
1680
target_link_libraries(${target} ${CUDA_cufft_LIBRARY})
1684
###############################################################################
1685
###############################################################################
1686
# CUDA ADD CUBLAS TO TARGET
1687
###############################################################################
1688
###############################################################################
1689
macro(CUDA_ADD_CUBLAS_TO_TARGET target)
1690
if (CUDA_BUILD_EMULATION)
1691
target_link_libraries(${target} ${CUDA_cublasemu_LIBRARY})
1693
target_link_libraries(${target} ${CUDA_cublas_LIBRARY})
1697
###############################################################################
1698
###############################################################################
1699
# CUDA BUILD CLEAN TARGET
1700
###############################################################################
1701
###############################################################################
1702
macro(CUDA_BUILD_CLEAN_TARGET)
1703
# Call this after you add all your CUDA targets, and you will get a convience
1704
# target. You should also make clean after running this target to get the
1705
# build system to generate all the code again.
1707
set(cuda_clean_target_name clean_cuda_depends)
1708
if (CMAKE_GENERATOR MATCHES "Visual Studio")
1709
string(TOUPPER ${cuda_clean_target_name} cuda_clean_target_name)
1711
add_custom_target(${cuda_clean_target_name}
1712
COMMAND ${CMAKE_COMMAND} -E remove ${CUDA_ADDITIONAL_CLEAN_FILES})
1714
# Clear out the variable, so the next time we configure it will be empty.
1715
# This is useful so that the files won't persist in the list after targets
1716
# have been removed.
1717
set(CUDA_ADDITIONAL_CLEAN_FILES "" CACHE INTERNAL "List of intermediate files that are part of the cuda dependency scanning.")