Merge pull request #1670 from jet47:cuda-cmake-fix

This commit is contained in:
Roman Donchenko 2013-10-29 13:24:14 +04:00 committed by OpenCV Buildbot
commit 957c85e9c4
5 changed files with 2286 additions and 46 deletions

1792
cmake/FindCUDA.cmake Normal file

File diff suppressed because it is too large Load Diff

View File

@ -0,0 +1,93 @@
# James Bigler, NVIDIA Corp (nvidia.com - jbigler)
# Abe Stephens, SCI Institute -- http://www.sci.utah.edu/~abe/FindCuda.html
#
# Copyright (c) 2008 - 2009 NVIDIA Corporation. All rights reserved.
#
# Copyright (c) 2007-2009
# Scientific Computing and Imaging Institute, University of Utah
#
# This code is licensed under the MIT License. See the FindCUDA.cmake script
# for the text of the license.
# The MIT License
#
# License for the specific language governing rights and limitations under
# Permission is hereby granted, free of charge, to any person obtaining a
# copy of this software and associated documentation files (the "Software"),
# to deal in the Software without restriction, including without limitation
# the rights to use, copy, modify, merge, publish, distribute, sublicense,
# and/or sell copies of the Software, and to permit persons to whom the
# Software is furnished to do so, subject to the following conditions:
#
# The above copyright notice and this permission notice shall be included
# in all copies or substantial portions of the Software.
#
# THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
# OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
# FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
# THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
# LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
# FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
# DEALINGS IN THE SOFTWARE.
#
#######################################################################
# This converts a file written in makefile syntax into one that can be included
# by CMake.
file(READ ${input_file} depend_text)
if (${depend_text} MATCHES ".+")
# message("FOUND DEPENDS")
# Remember, four backslashes is escaped to one backslash in the string.
string(REGEX REPLACE "\\\\ " " " depend_text ${depend_text})
# This works for the nvcc -M generated dependency files.
string(REGEX REPLACE "^.* : " "" depend_text ${depend_text})
string(REGEX REPLACE "[ \\\\]*\n" ";" depend_text ${depend_text})
set(dependency_list "")
foreach(file ${depend_text})
string(REGEX REPLACE "^ +" "" file ${file})
# OK, now if we had a UNC path, nvcc has a tendency to only output the first '/'
# instead of '//'. Here we will test to see if the file exists, if it doesn't then
# try to prepend another '/' to the path and test again. If it still fails remove the
# path.
if(NOT EXISTS "${file}")
if (EXISTS "/${file}")
set(file "/${file}")
else()
message(WARNING " Removing non-existent dependency file: ${file}")
set(file "")
endif()
endif()
if(NOT IS_DIRECTORY "${file}")
# If softlinks start to matter, we should change this to REALPATH. For now we need
# to flatten paths, because nvcc can generate stuff like /bin/../include instead of
# just /include.
get_filename_component(file_absolute "${file}" ABSOLUTE)
list(APPEND dependency_list "${file_absolute}")
endif()
endforeach()
else()
# message("FOUND NO DEPENDS")
endif()
# Remove the duplicate entries and sort them.
list(REMOVE_DUPLICATES dependency_list)
list(SORT dependency_list)
foreach(file ${dependency_list})
set(cuda_nvcc_depend "${cuda_nvcc_depend} \"${file}\"\n")
endforeach()
file(WRITE ${output_file} "# Generated by: make2cmake.cmake\nSET(CUDA_NVCC_DEPEND\n ${cuda_nvcc_depend})\n\n")

View File

@ -0,0 +1,110 @@
# James Bigler, NVIDIA Corp (nvidia.com - jbigler)
# Abe Stephens, SCI Institute -- http://www.sci.utah.edu/~abe/FindCuda.html
#
# Copyright (c) 2008 - 2009 NVIDIA Corporation. All rights reserved.
#
# Copyright (c) 2007-2009
# Scientific Computing and Imaging Institute, University of Utah
#
# This code is licensed under the MIT License. See the FindCUDA.cmake script
# for the text of the license.
# The MIT License
#
# License for the specific language governing rights and limitations under
# Permission is hereby granted, free of charge, to any person obtaining a
# copy of this software and associated documentation files (the "Software"),
# to deal in the Software without restriction, including without limitation
# the rights to use, copy, modify, merge, publish, distribute, sublicense,
# and/or sell copies of the Software, and to permit persons to whom the
# Software is furnished to do so, subject to the following conditions:
#
# The above copyright notice and this permission notice shall be included
# in all copies or substantial portions of the Software.
#
# THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
# OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
# FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
# THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
# LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
# FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
# DEALINGS IN THE SOFTWARE.
#
#######################################################################
# Parses a .cubin file produced by nvcc and reports statistics about the file.
file(READ ${input_file} file_text)
if (${file_text} MATCHES ".+")
# Remember, four backslashes is escaped to one backslash in the string.
string(REGEX REPLACE ";" "\\\\;" file_text ${file_text})
string(REGEX REPLACE "\ncode" ";code" file_text ${file_text})
list(LENGTH file_text len)
foreach(line ${file_text})
# Only look at "code { }" blocks.
if(line MATCHES "^code")
# Break into individual lines.
string(REGEX REPLACE "\n" ";" line ${line})
foreach(entry ${line})
# Extract kernel names.
if (${entry} MATCHES "[^g]name = ([^ ]+)")
string(REGEX REPLACE ".* = ([^ ]+)" "\\1" entry ${entry})
# Check to see if the kernel name starts with "_"
set(skip FALSE)
# if (${entry} MATCHES "^_")
# Skip the rest of this block.
# message("Skipping ${entry}")
# set(skip TRUE)
# else ()
message("Kernel: ${entry}")
# endif ()
endif()
# Skip the rest of the block if necessary
if(NOT skip)
# Registers
if (${entry} MATCHES "reg([ ]+)=([ ]+)([^ ]+)")
string(REGEX REPLACE ".*([ ]+)=([ ]+)([^ ]+)" "\\3" entry ${entry})
message("Registers: ${entry}")
endif()
# Local memory
if (${entry} MATCHES "lmem([ ]+)=([ ]+)([^ ]+)")
string(REGEX REPLACE ".*([ ]+)=([ ]+)([^ ]+)" "\\3" entry ${entry})
message("Local: ${entry}")
endif()
# Shared memory
if (${entry} MATCHES "smem([ ]+)=([ ]+)([^ ]+)")
string(REGEX REPLACE ".*([ ]+)=([ ]+)([^ ]+)" "\\3" entry ${entry})
message("Shared: ${entry}")
endif()
if (${entry} MATCHES "^}")
message("")
endif()
endif()
endforeach()
endif()
endforeach()
else()
# message("FOUND NO DEPENDS")
endif()

View File

@ -0,0 +1,288 @@
# James Bigler, NVIDIA Corp (nvidia.com - jbigler)
#
# Copyright (c) 2008 - 2009 NVIDIA Corporation. All rights reserved.
#
# This code is licensed under the MIT License. See the FindCUDA.cmake script
# for the text of the license.
# The MIT License
#
# License for the specific language governing rights and limitations under
# Permission is hereby granted, free of charge, to any person obtaining a
# copy of this software and associated documentation files (the "Software"),
# to deal in the Software without restriction, including without limitation
# the rights to use, copy, modify, merge, publish, distribute, sublicense,
# and/or sell copies of the Software, and to permit persons to whom the
# Software is furnished to do so, subject to the following conditions:
#
# The above copyright notice and this permission notice shall be included
# in all copies or substantial portions of the Software.
#
# THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
# OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
# FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
# THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
# LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
# FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
# DEALINGS IN THE SOFTWARE.
##########################################################################
# This file runs the nvcc commands to produce the desired output file along with
# the dependency file needed by CMake to compute dependencies. In addition the
# file checks the output of each command and if the command fails it deletes the
# output files.
# Input variables
#
# verbose:BOOL=<> OFF: Be as quiet as possible (default)
# ON : Describe each step
#
# build_configuration:STRING=<> Typically one of Debug, MinSizeRel, Release, or
# RelWithDebInfo, but it should match one of the
# entries in CUDA_HOST_FLAGS. This is the build
# configuration used when compiling the code. If
# blank or unspecified Debug is assumed as this is
# what CMake does.
#
# generated_file:STRING=<> File to generate. This argument must be passed in.
#
# generated_cubin_file:STRING=<> File to generate. This argument must be passed
# in if build_cubin is true.
if(NOT generated_file)
message(FATAL_ERROR "You must specify generated_file on the command line")
endif()
# Set these up as variables to make reading the generated file easier
set(CMAKE_COMMAND "@CMAKE_COMMAND@") # path
set(source_file "@source_file@") # path
set(NVCC_generated_dependency_file "@NVCC_generated_dependency_file@") # path
set(cmake_dependency_file "@cmake_dependency_file@") # path
set(CUDA_make2cmake "@CUDA_make2cmake@") # path
set(CUDA_parse_cubin "@CUDA_parse_cubin@") # path
set(build_cubin @build_cubin@) # bool
set(CUDA_HOST_COMPILER "@CUDA_HOST_COMPILER@") # bool
# We won't actually use these variables for now, but we need to set this, in
# order to force this file to be run again if it changes.
set(generated_file_path "@generated_file_path@") # path
set(generated_file_internal "@generated_file@") # path
set(generated_cubin_file_internal "@generated_cubin_file@") # path
set(CUDA_NVCC_EXECUTABLE "@CUDA_NVCC_EXECUTABLE@") # path
set(CUDA_NVCC_FLAGS @CUDA_NVCC_FLAGS@ ;; @CUDA_WRAP_OPTION_NVCC_FLAGS@) # list
@CUDA_NVCC_FLAGS_CONFIG@
set(nvcc_flags @nvcc_flags@) # list
set(CUDA_NVCC_INCLUDE_ARGS "@CUDA_NVCC_INCLUDE_ARGS@") # list (needs to be in quotes to handle spaces properly).
set(format_flag "@format_flag@") # string
if(build_cubin AND NOT generated_cubin_file)
message(FATAL_ERROR "You must specify generated_cubin_file on the command line")
endif()
# This is the list of host compilation flags. It C or CXX should already have
# been chosen by FindCUDA.cmake.
@CUDA_HOST_FLAGS@
# Take the compiler flags and package them up to be sent to the compiler via -Xcompiler
set(nvcc_host_compiler_flags "")
# If we weren't given a build_configuration, use Debug.
if(NOT build_configuration)
set(build_configuration Debug)
endif()
string(TOUPPER "${build_configuration}" build_configuration)
#message("CUDA_NVCC_HOST_COMPILER_FLAGS = ${CUDA_NVCC_HOST_COMPILER_FLAGS}")
foreach(flag ${CMAKE_HOST_FLAGS} ${CMAKE_HOST_FLAGS_${build_configuration}})
# Extra quotes are added around each flag to help nvcc parse out flags with spaces.
set(nvcc_host_compiler_flags "${nvcc_host_compiler_flags},\"${flag}\"")
endforeach()
if (nvcc_host_compiler_flags)
set(nvcc_host_compiler_flags "-Xcompiler" ${nvcc_host_compiler_flags})
endif()
#message("nvcc_host_compiler_flags = \"${nvcc_host_compiler_flags}\"")
# Add the build specific configuration flags
list(APPEND CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS_${build_configuration}})
# Any -ccbin existing in CUDA_NVCC_FLAGS gets highest priority
list( FIND CUDA_NVCC_FLAGS "-ccbin" ccbin_found0 )
list( FIND CUDA_NVCC_FLAGS "--compiler-bindir" ccbin_found1 )
if( ccbin_found0 LESS 0 AND ccbin_found1 LESS 0 )
if (CUDA_HOST_COMPILER STREQUAL "$(VCInstallDir)bin" AND DEFINED CCBIN)
set(CCBIN -ccbin "${CCBIN}")
else()
set(CCBIN -ccbin "${CUDA_HOST_COMPILER}")
endif()
endif()
# cuda_execute_process - Executes a command with optional command echo and status message.
#
# status - Status message to print if verbose is true
# command - COMMAND argument from the usual execute_process argument structure
# ARGN - Remaining arguments are the command with arguments
#
# CUDA_result - return value from running the command
#
# Make this a macro instead of a function, so that things like RESULT_VARIABLE
# and other return variables are present after executing the process.
macro(cuda_execute_process status command)
set(_command ${command})
if(NOT _command STREQUAL "COMMAND")
message(FATAL_ERROR "Malformed call to cuda_execute_process. Missing COMMAND as second argument. (command = ${command})")
endif()
if(verbose)
execute_process(COMMAND "${CMAKE_COMMAND}" -E echo -- ${status})
# Now we need to build up our command string. We are accounting for quotes
# and spaces, anything else is left up to the user to fix if they want to
# copy and paste a runnable command line.
set(cuda_execute_process_string)
foreach(arg ${ARGN})
# If there are quotes, excape them, so they come through.
string(REPLACE "\"" "\\\"" arg ${arg})
# Args with spaces need quotes around them to get them to be parsed as a single argument.
if(arg MATCHES " ")
list(APPEND cuda_execute_process_string "\"${arg}\"")
else()
list(APPEND cuda_execute_process_string ${arg})
endif()
endforeach()
# Echo the command
execute_process(COMMAND ${CMAKE_COMMAND} -E echo ${cuda_execute_process_string})
endif()
# Run the command
execute_process(COMMAND ${ARGN} RESULT_VARIABLE CUDA_result )
endmacro()
# Delete the target file
cuda_execute_process(
"Removing ${generated_file}"
COMMAND "${CMAKE_COMMAND}" -E remove "${generated_file}"
)
# For CUDA 2.3 and below, -G -M doesn't work, so remove the -G flag
# for dependency generation and hope for the best.
set(depends_CUDA_NVCC_FLAGS "${CUDA_NVCC_FLAGS}")
set(CUDA_VERSION @CUDA_VERSION@)
if(CUDA_VERSION VERSION_LESS "3.0")
cmake_policy(PUSH)
# CMake policy 0007 NEW states that empty list elements are not
# ignored. I'm just setting it to avoid the warning that's printed.
cmake_policy(SET CMP0007 NEW)
# Note that this will remove all occurances of -G.
list(REMOVE_ITEM depends_CUDA_NVCC_FLAGS "-G")
cmake_policy(POP)
endif()
# nvcc doesn't define __CUDACC__ for some reason when generating dependency files. This
# can cause incorrect dependencies when #including files based on this macro which is
# defined in the generating passes of nvcc invokation. We will go ahead and manually
# define this for now until a future version fixes this bug.
set(CUDACC_DEFINE -D__CUDACC__)
# Generate the dependency file
cuda_execute_process(
"Generating dependency file: ${NVCC_generated_dependency_file}"
COMMAND "${CUDA_NVCC_EXECUTABLE}"
-M
${CUDACC_DEFINE}
"${source_file}"
-o "${NVCC_generated_dependency_file}"
${CCBIN}
${nvcc_flags}
${nvcc_host_compiler_flags}
${depends_CUDA_NVCC_FLAGS}
-DNVCC
${CUDA_NVCC_INCLUDE_ARGS}
)
if(CUDA_result)
message(FATAL_ERROR "Error generating ${generated_file}")
endif()
# Generate the cmake readable dependency file to a temp file. Don't put the
# quotes just around the filenames for the input_file and output_file variables.
# CMake will pass the quotes through and not be able to find the file.
cuda_execute_process(
"Generating temporary cmake readable file: ${cmake_dependency_file}.tmp"
COMMAND "${CMAKE_COMMAND}"
-D "input_file:FILEPATH=${NVCC_generated_dependency_file}"
-D "output_file:FILEPATH=${cmake_dependency_file}.tmp"
-P "${CUDA_make2cmake}"
)
if(CUDA_result)
message(FATAL_ERROR "Error generating ${generated_file}")
endif()
# Copy the file if it is different
cuda_execute_process(
"Copy if different ${cmake_dependency_file}.tmp to ${cmake_dependency_file}"
COMMAND "${CMAKE_COMMAND}" -E copy_if_different "${cmake_dependency_file}.tmp" "${cmake_dependency_file}"
)
if(CUDA_result)
message(FATAL_ERROR "Error generating ${generated_file}")
endif()
# Delete the temporary file
cuda_execute_process(
"Removing ${cmake_dependency_file}.tmp and ${NVCC_generated_dependency_file}"
COMMAND "${CMAKE_COMMAND}" -E remove "${cmake_dependency_file}.tmp" "${NVCC_generated_dependency_file}"
)
if(CUDA_result)
message(FATAL_ERROR "Error generating ${generated_file}")
endif()
# Generate the code
cuda_execute_process(
"Generating ${generated_file}"
COMMAND "${CUDA_NVCC_EXECUTABLE}"
"${source_file}"
${format_flag} -o "${generated_file}"
${CCBIN}
${nvcc_flags}
${nvcc_host_compiler_flags}
${CUDA_NVCC_FLAGS}
-DNVCC
${CUDA_NVCC_INCLUDE_ARGS}
)
if(CUDA_result)
# Since nvcc can sometimes leave half done files make sure that we delete the output file.
cuda_execute_process(
"Removing ${generated_file}"
COMMAND "${CMAKE_COMMAND}" -E remove "${generated_file}"
)
message(FATAL_ERROR "Error generating file ${generated_file}")
else()
if(verbose)
message("Generated ${generated_file} successfully.")
endif()
endif()
# Cubin resource report commands.
if( build_cubin )
# Run with -cubin to produce resource usage report.
cuda_execute_process(
"Generating ${generated_cubin_file}"
COMMAND "${CUDA_NVCC_EXECUTABLE}"
"${source_file}"
${CUDA_NVCC_FLAGS}
${nvcc_flags}
${CCBIN}
${nvcc_host_compiler_flags}
-DNVCC
-cubin
-o "${generated_cubin_file}"
${CUDA_NVCC_INCLUDE_ARGS}
)
# Execute the parser script.
cuda_execute_process(
"Executing the parser script"
COMMAND "${CMAKE_COMMAND}"
-D "input_file:STRING=${generated_cubin_file}"
-P "${CUDA_parse_cubin}"
)
endif()

View File

@ -13,7 +13,9 @@ if(CMAKE_COMPILER_IS_GNUCXX AND NOT APPLE AND CMAKE_CXX_COMPILER_ID STREQUAL "Cl
return()
endif()
find_package(CUDA 4.2 QUIET)
set(CMAKE_MODULE_PATH "${OpenCV_SOURCE_DIR}/cmake" ${CMAKE_MODULE_PATH})
find_host_package(CUDA 4.2 QUIET)
if(CUDA_FOUND)
set(HAVE_CUDA 1)
@ -26,47 +28,6 @@ if(CUDA_FOUND)
set(HAVE_CUBLAS 1)
endif()
if(${CUDA_VERSION} VERSION_LESS "5.5")
find_cuda_helper_libs(npp)
else()
# hack for CUDA 5.5
if(${CMAKE_SYSTEM_PROCESSOR} STREQUAL "arm")
unset(CUDA_TOOLKIT_INCLUDE CACHE)
unset(CUDA_CUDART_LIBRARY CACHE)
unset(CUDA_cublas_LIBRARY CACHE)
unset(CUDA_cufft_LIBRARY CACHE)
unset(CUDA_npp_LIBRARY CACHE)
if(SOFTFP)
set(cuda_arm_path "${CUDA_TOOLKIT_ROOT_DIR}/targets/armv7-linux-gnueabi")
else()
set(cuda_arm_path "${CUDA_TOOLKIT_ROOT_DIR}/targets/armv7-linux-gnueabihf")
endif()
set(CUDA_TOOLKIT_INCLUDE "${cuda_arm_path}/include" CACHE PATH "include path")
set(CUDA_INCLUDE_DIRS ${CUDA_TOOLKIT_INCLUDE})
set(cuda_arm_library_path "${cuda_arm_path}/lib")
set(CUDA_CUDART_LIBRARY "${cuda_arm_library_path}/libcudart.so" CACHE FILEPATH "cudart library")
set(CUDA_LIBRARIES ${CUDA_CUDART_LIBRARY})
set(CUDA_cublas_LIBRARY "${cuda_arm_library_path}/libcublas.so" CACHE FILEPATH "cublas library")
set(CUDA_cufft_LIBRARY "${cuda_arm_library_path}/libcufft.so" CACHE FILEPATH "cufft library")
set(CUDA_nppc_LIBRARY "${cuda_arm_library_path}/libnppc.so" CACHE FILEPATH "nppc library")
set(CUDA_nppi_LIBRARY "${cuda_arm_library_path}/libnppi.so" CACHE FILEPATH "nppi library")
set(CUDA_npps_LIBRARY "${cuda_arm_library_path}/libnpps.so" CACHE FILEPATH "npps library")
set(CUDA_npp_LIBRARY "${CUDA_nppc_LIBRARY};${CUDA_nppi_LIBRARY};${CUDA_npps_LIBRARY}" CACHE STRING "npp library")
else()
unset(CUDA_npp_LIBRARY CACHE)
find_cuda_helper_libs(nppc)
find_cuda_helper_libs(nppi)
find_cuda_helper_libs(npps)
set(CUDA_npp_LIBRARY "${CUDA_nppc_LIBRARY};${CUDA_nppi_LIBRARY};${CUDA_npps_LIBRARY}" CACHE STRING "npp library")
endif()
endif()
if(WITH_NVCUVID)
find_cuda_helper_libs(nvcuvid)
set(HAVE_NVCUVID 1)
@ -168,10 +129,6 @@ if(CUDA_FOUND)
set(OPENCV_CUDA_ARCH_FEATURES "${OPENCV_CUDA_ARCH_FEATURES} ${ARCH}")
endforeach()
if(${CMAKE_SYSTEM_PROCESSOR} STREQUAL "arm")
set(CUDA_NVCC_FLAGS "${CUDA_NVCC_FLAGS} --target-cpu-architecture=ARM")
endif()
# These vars will be processed in other scripts
set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} ${NVCC_FLAGS_EXTRA})
set(OpenCV_CUDA_CC "${NVCC_FLAGS_EXTRA}")