1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
|
# Find CUDA
if(MSVC)
# Setting this to true brakes Visual Studio builds.
set(CUDA_ATTACH_VS_BUILD_RULE_TO_CUDA_FILE OFF CACHE BOOL "CUDA_ATTACH_VS_BUILD_RULE_TO_CUDA_FILE")
endif()
set(CUDA_FIND_QUIETLY TRUE)
find_package(CUDA)
if(CUDA_FOUND)
message(STATUS "Found CUDA Toolkit v${CUDA_VERSION_STRING}")
set(HAVE_CUDA TRUE)
if(${CUDA_VERSION_STRING} VERSION_LESS "7.5")
# Recent versions of cmake set CUDA_HOST_COMPILER to CMAKE_C_COMPILER which
# on OSX defaults to clang (/usr/bin/cc), but this is not a supported cuda
# compiler. So, here we will preemptively set CUDA_HOST_COMPILER to gcc if
# that compiler exists in /usr/bin. This will not override an existing cache
# value if the user has passed CUDA_HOST_COMPILER on the command line.
if (NOT DEFINED CUDA_HOST_COMPILER AND CMAKE_C_COMPILER_ID STREQUAL "Clang" AND EXISTS /usr/bin/gcc)
set(CUDA_HOST_COMPILER /usr/bin/gcc CACHE FILEPATH "Host side compiler used by NVCC")
message(STATUS "Setting CMAKE_HOST_COMPILER to /usr/bin/gcc instead of ${CMAKE_C_COMPILER}. See http://dev.pointclouds.org/issues/979")
endif()
# Send a warning if CUDA_HOST_COMPILER is set to a compiler that is known
# to be unsupported.
if (CUDA_HOST_COMPILER STREQUAL CMAKE_C_COMPILER AND CMAKE_C_COMPILER_ID STREQUAL "Clang")
message(WARNING "CUDA_HOST_COMPILER is set to an unsupported compiler: ${CMAKE_C_COMPILER}. See http://dev.pointclouds.org/issues/979")
endif()
endif()
# CUDA_ARCH_BIN is a space separated list of versions to include in output so-file. So you can set CUDA_ARCH_BIN = 10 11 12 13 20
# Also user can specify virtual arch in parenthesis to limit instructions set,
# for example CUDA_ARCH_BIN = 11(11) 12(11) 13(11) 20(11) 21(11) -> forces using only sm_11 instructions.
# The CMake scripts interpret XX as XX (XX). This allows user to omit parenthesis.
# Arch 21 is an exceptional case since it doesn't have own sm_21 instructions set.
# So 21 = 21(21) is an invalid configuration and user has to explicitly force previous sm_20 instruction set via 21(20).
# CUDA_ARCH_BIN adds support of only listed GPUs. As alternative CMake scripts also parse 'CUDA_ARCH_PTX' variable,
# which is a list of intermediate PTX codes to include in final so-file. The PTX code can/will be JIT compiled for any current or future GPU.
# To add support of older GPU for kinfu, I would embed PTX 11 and 12 into so-file. GPU with sm_13 will run PTX 12 code (no difference for kinfu)
# Find a complete list for CUDA compute capabilities at http://developer.nvidia.com/cuda-gpus
if(NOT ${CUDA_VERSION_STRING} VERSION_LESS "10.0")
set(__cuda_arch_bin "3.0 3.5 5.0 5.2 5.3 6.0 6.1 7.0 7.2 7.5")
elseif(NOT ${CUDA_VERSION_STRING} VERSION_LESS "9.1")
set(__cuda_arch_bin "3.0 3.5 5.0 5.2 5.3 6.0 6.1 7.0 7.2")
elseif(NOT ${CUDA_VERSION_STRING} VERSION_LESS "9.0")
set(__cuda_arch_bin "3.0 3.5 5.0 5.2 5.3 6.0 6.1 7.0")
elseif(NOT ${CUDA_VERSION_STRING} VERSION_LESS "8.0")
set(__cuda_arch_bin "2.0 2.1(2.0) 3.0 3.5 5.0 5.2 5.3 6.0 6.1")
elseif(NOT ${CUDA_VERSION_STRING} VERSION_LESS "6.5")
set(__cuda_arch_bin "2.0 2.1(2.0) 3.0 3.5 5.0 5.2")
elseif(NOT ${CUDA_VERSION_STRING} VERSION_LESS "6.0")
set(__cuda_arch_bin "2.0 2.1(2.0) 3.0 3.5 5.0")
elseif(NOT ${CUDA_VERSION_STRING} VERSION_LESS "5.0")
set(__cuda_arch_bin "2.0 2.1(2.0) 3.0 3.5")
elseif(${CUDA_VERSION_STRING} VERSION_GREATER "4.1")
set(__cuda_arch_bin "2.0 2.1(2.0) 3.0")
else()
set(__cuda_arch_bin "2.0 2.1(2.0)")
endif()
set(CUDA_ARCH_BIN ${__cuda_arch_bin} CACHE STRING "Specify 'real' GPU architectures to build binaries for, BIN(PTX) format is supported")
set(CUDA_ARCH_PTX "" CACHE STRING "Specify 'virtual' PTX arch to build PTX intermediate code for. Example: 1.0 1.2 or 10 12")
#set(CUDA_ARCH_PTX "1.1 1.2" CACHE STRING "Specify 'virtual' PTX arch to build PTX intermediate code for. Example: 1.0 1.2 or 10 12")
# Guess this macros will be included in cmake distributive
include(${PCL_SOURCE_DIR}/cmake/CudaComputeTargetFlags.cmake)
APPEND_TARGET_ARCH_FLAGS()
# Prevent compilation issues between recent gcc versions and old CUDA versions
list(APPEND CUDA_NVCC_FLAGS "-D_FORCE_INLINES")
endif()
|
