hanchenye-llvm-project/openmp/runtime/CMakeLists.txt

965 lines
40 KiB
CMake
Raw Normal View History

#
#//===----------------------------------------------------------------------===//
#//
#// The LLVM Compiler Infrastructure
#//
#// This file is dual licensed under the MIT and the University of Illinois Open
#// Source Licenses. See LICENSE.txt for details.
#//
#//===----------------------------------------------------------------------===//
#
################
# CMAKE libiomp5
cmake_minimum_required(VERSION 2.8 FATAL_ERROR)
#########
# GLOBALS
set(GLOBAL_DEBUG 0)
# Add cmake directory to search for custom cmake functions
set(CMAKE_MODULE_PATH ${CMAKE_CURRENT_SOURCE_DIR}/cmake ${CMAKE_MODULE_PATH})
#######################################################################
# Standalone build or part of LLVM?
set(LIBOMP_STANDALONE_BUILD FALSE)
if("${CMAKE_SOURCE_DIR}" STREQUAL "${CMAKE_CURRENT_SOURCE_DIR}" OR
"${CMAKE_SOURCE_DIR}/runtime" STREQUAL "${CMAKE_CURRENT_SOURCE_DIR}")
project(libiomp C CXX)
set(LIBOMP_STANDALONE_BUILD TRUE)
endif()
# These include files are in cmake/ subdirectory except for FindPerl which is a cmake standard module
include(HelperFunctions)
include(Definitions) # -D definitions when compiling
include(CommonFlags) # compiler, assembler, fortran, linker flags common for all compilers
include(SourceFiles) # source files to compile
include(PerlFlags) # Perl flags for generate-def.pl and expand-vars.pl
include(FindPerl) # Standard cmake module to check for Perl
include(GetArchitecture) # get_architecture()
####################################################################
# CONFIGURATION
#
# * Any variable/value that is CACHE-ed can be changed after the initial run of cmake
# through the file, CMakeCache.txt which is in the build directory.
# * If you change any value in CMakeCache.txt, then just run cmake ..
# and the changed will be picked up. One can also use -DVARIABLE=VALUE
# when calling cmake to changed configuration values.
# * CMAKE_C_COMPILER, CMAKE_CXX_COMPILER, CMAKE_ASM_MASM_COMPILER, CMAKE_ASM_COMPILER,
# CMAKE_Fortran_COMPILER can only by specified on the initial run of cmake.
# This means you cannot specify -DCMAKE_C_COMPILER= on a subsequent run of cmake
# in the same build directory until that build directory is emptied.
# If you want to change the compiler, then empty the build directory and rerun cmake.
# Build Configuration
set(os_possible_values lin mac win)
set(arch_possible_values 32e 32 arm ppc64 ppc64le aarch64 mic)
set(build_type_possible_values release debug relwithdebinfo)
set(omp_version_possible_values 41 40 30)
set(lib_type_possible_values normal profile stubs)
set(mic_arch_possible_values knf knc)
# Below, cmake will try and determine the operating system and architecture for you.
# These values are set in CMakeCache.txt when cmake is first run (-Dvar_name=... will take precedence)
# parameter | default value
# ----------------------------
# Right now, this build system considers os=lin to mean "Unix-like that is not MAC"
# Apple goes first because CMake considers Mac to be a Unix based
# operating system, while libiomp5 considers it a special case
if(${APPLE})
set(temp_os mac)
elseif(${UNIX})
set(temp_os lin)
elseif(${WIN32})
set(temp_os win)
else()
set(temp_os lin)
endif()
# If adding a new architecture, take a look at cmake/GetArchitecture.cmake
get_architecture(detected_arch)
set(LIBOMP_OS ${temp_os} CACHE STRING
"The operating system to build for (lin/mac/win)")
set(LIBOMP_ARCH ${detected_arch} CACHE STRING
"The architecture to build for (32e/32/arm/ppc64/ppc64le/aarch64/mic). 32e is Intel(R) 64 architecture, 32 is IA-32 architecture")
set(LIBOMP_LIB_TYPE normal CACHE STRING
"Performance,Profiling,Stubs library (normal/profile/stubs)")
set(LIBOMP_VERSION 5 CACHE STRING
"Produce libguide (version 4) or libiomp5 (version 5)")
set(LIBOMP_OMP_VERSION 41 CACHE STRING
"The OpenMP version (41/40/30)")
set(LIBOMP_MIC_ARCH knc CACHE STRING
"Intel(R) Many Integrated Core Architecture (Intel(R) MIC Architecture) (knf/knc). Ignored if not Intel(R) MIC Architecture build.")
set(LIBOMP_FORTRAN_MODULES false CACHE BOOL
"Create Fortran module files? (requires fortran compiler)")
# - These tests are little tests performed after the library is formed.
# - The library won't be copied to the exports directory
# until it has passed/skipped all below tests
# - To skip these tests, just pass -DLIBOMP_MICRO_TESTS=OFF to cmake
set(LIBOMP_TEST_TOUCH true CACHE BOOL
"Perform a small touch test?")
set(LIBOMP_TEST_RELO true CACHE BOOL
"Perform a relocation test for dynamic libraries?")
set(LIBOMP_TEST_EXECSTACK true CACHE BOOL
"Perform a execstack test for linux dynamic libraries?")
set(LIBOMP_TEST_INSTR true CACHE BOOL
"Perform an instruction test for Intel(R) MIC Architecture libraries?")
set(LIBOMP_TEST_DEPS true CACHE BOOL
"Perform a library dependency test?")
set(LIBOMP_MICRO_TESTS false CACHE BOOL
"Perform touch, relo, execstack, instr, and deps tests?")
# - stats-gathering enables OpenMP stats where things like the number of
# parallel regions, clock ticks spent in particular openmp regions are recorded.
set(LIBOMP_STATS false CACHE BOOL
"Stats-Gathering functionality?")
# OMPT-support
set(LIBOMP_OMPT_SUPPORT false CACHE BOOL
"OMPT-support?")
set(LIBOMP_OMPT_BLAME true CACHE BOOL
"OMPT-blame?")
set(LIBOMP_OMPT_TRACE true CACHE BOOL
"OMPT-trace?")
# User specified flags. These are appended to the predetermined flags found
# in CommonFlags.cmake and ${CMAKE_C_COMPILER_ID}/*Flags.cmake (e.g., GNU/CFlags.cmake)
set(LIBOMP_CFLAGS "" CACHE STRING
"Appended user specified C compiler flags.")
set(LIBOMP_CXXFLAGS "" CACHE STRING
"Appended user specified C++ compiler flags.")
set(LIBOMP_CPPFLAGS "" CACHE STRING
"Appended user specified C preprocessor flags.")
set(LIBOMP_ASMFLAGS "" CACHE STRING
"Appended user specified assembler flags.")
set(LIBOMP_LDFLAGS "" CACHE STRING
"Appended user specified linker flags.")
set(LIBOMP_LIBFLAGS "" CACHE STRING
"Appended user specified linked libs flags. (e.g., -lm)")
set(LIBOMP_FFLAGS "" CACHE STRING
"Appended user specified Fortran compiler flags. These are only used if LIBOMP_FORTRAN_MODULES==true.")
# Should the libiomp5 library and generated headers be copied into the original source exports/ directory
# Turning this to false aids parallel builds to not interfere with each other.
set(LIBOMP_COPY_EXPORTS true CACHE STRING
"Should exports be copied into source exports/ directory?")
# - Allow three build types: Release, Debug, RelWithDebInfo (these relate to build.pl's release, debug, and diag settings respectively)
# - default is Release (when CMAKE_BUILD_TYPE is not defined)
# - CMAKE_BUILD_TYPE affects the -O and -g flags (CMake magically includes correct version of them on per compiler basis)
# - typical: Release = -O3 -DNDEBUG
# RelWithDebInfo = -O2 -g -DNDEBUG
# Debug = -g
if(CMAKE_BUILD_TYPE)
# CMAKE_BUILD_TYPE was defined, check for validity
string(TOLOWER "${CMAKE_BUILD_TYPE}" cmake_build_type_lowercase)
check_variable(cmake_build_type_lowercase "${build_type_possible_values}")
else()
# CMAKE_BUILD_TYPE was not defined, set default to Release
unset(CMAKE_BUILD_TYPE CACHE)
set(CMAKE_BUILD_TYPE Release CACHE STRING
"Choose the type of build, options are: Release/Debug/RelWithDebInfo")
string(TOLOWER "${CMAKE_BUILD_TYPE}" cmake_build_type_lowercase)
check_variable(cmake_build_type_lowercase "${build_type_possible_values}")
endif()
# Allow user to choose a suffix for the installation directory, or if part of
# LLVM build then just use LLVM_LIBDIR_SUFFIX
if(${LIBOMP_STANDALONE_BUILD})
set(LIBOMP_LIBDIR_SUFFIX "" CACHE STRING
"suffix of lib installation directory e.g., 64 => lib64")
else()
set(LIBOMP_LIBDIR_SUFFIX ${LLVM_LIBDIR_SUFFIX})
endif()
# Check valid values
check_variable(LIBOMP_OS "${os_possible_values}")
check_variable(LIBOMP_ARCH "${arch_possible_values}")
check_variable(LIBOMP_OMP_VERSION "${omp_version_possible_values}")
check_variable(LIBOMP_LIB_TYPE "${lib_type_possible_values}")
if("${LIBOMP_ARCH}" STREQUAL "mic")
check_variable(LIBOMP_MIC_ARCH "${mic_arch_possible_values}")
endif()
# Get the build number from kmp_version.c
get_build_number("${CMAKE_CURRENT_SOURCE_DIR}" build_number)
# Getting time and date
# As of now, no timestamp will be created.
set(date "No Timestamp")
#################################################################
# Set some useful flags variables for other parts of cmake to use
# Operating System
set(LINUX FALSE)
set(MAC FALSE)
set(WINDOWS FALSE)
set(MIC FALSE)
set(FREEBSD FALSE)
if("${LIBOMP_OS}" STREQUAL "lin")
set(LINUX TRUE)
set(real_os lin)
elseif("${LIBOMP_OS}" STREQUAL "mac")
set(MAC TRUE)
set(real_os mac)
elseif("${LIBOMP_OS}" STREQUAL "win")
set(WINDOWS TRUE)
set(real_os win)
endif()
if("${CMAKE_SYSTEM_NAME}" STREQUAL "FreeBSD")
set(FREEBSD TRUE)
endif()
# Architecture
set(IA32 FALSE)
set(INTEL64 FALSE)
set(ARM FALSE)
set(AARCH64 FALSE)
set(PPC64BE FALSE)
set(PPC64LE FALSE)
set(PPC64 FALSE)
if("${LIBOMP_ARCH}" STREQUAL "32") # IA-32 architecture
set(IA32 TRUE)
elseif("${LIBOMP_ARCH}" STREQUAL "32e") # Intel(R) 64 architecture
set(INTEL64 TRUE)
elseif("${LIBOMP_ARCH}" STREQUAL "arm") # ARM architecture
set(ARM TRUE)
elseif("${LIBOMP_ARCH}" STREQUAL "ppc64") # PPC64BE architecture
set(PPC64BE TRUE)
set(PPC64 TRUE)
elseif("${LIBOMP_ARCH}" STREQUAL "ppc64le") # PPC64LE architecture
set(PPC64LE TRUE)
set(PPC64 TRUE)
elseif("${LIBOMP_ARCH}" STREQUAL "aarch64") # AARCH64 architecture
set(AARCH64 TRUE)
elseif("${LIBOMP_ARCH}" STREQUAL "mic") # Intel(R) Many Integrated Core Architecture
set(MIC TRUE)
endif()
# Set some flags based on build_type
# cmake_build_type_lowercase is based off of CMAKE_BUILD_TYPE, just put in lowercase.
set(RELEASE_BUILD FALSE)
set(DEBUG_BUILD FALSE)
set(RELWITHDEBINFO_BUILD FALSE)
if("${cmake_build_type_lowercase}" STREQUAL "release")
set(RELEASE_BUILD TRUE)
elseif("${cmake_build_type_lowercase}" STREQUAL "debug")
set(DEBUG_BUILD TRUE)
elseif("${cmake_build_type_lowercase}" STREQUAL "relwithdebinfo")
set(RELWITHDEBINFO_BUILD TRUE)
endif()
# Include itt notify interface? Right now, always.
set(LIBOMP_USE_ITT_NOTIFY TRUE)
# normal, profile, stubs library.
set(NORMAL_LIBRARY FALSE)
set(STUBS_LIBRARY FALSE)
set(PROFILE_LIBRARY FALSE)
if("${LIBOMP_LIB_TYPE}" STREQUAL "normal")
set(NORMAL_LIBRARY TRUE)
elseif("${LIBOMP_LIB_TYPE}" STREQUAL "profile")
set(PROFILE_LIBRARY TRUE)
elseif("${LIBOMP_LIB_TYPE}" STREQUAL "stubs")
set(STUBS_LIBRARY TRUE)
endif()
###############################################
# Features for compilation and build in general
# - Does the compiler support a 128-bit floating point data type? Default is false
# - If a compiler does, then change it in the CMakeCache.txt file (or using the cmake GUI)
# or send to cmake -DCOMPILER_SUPPORTS_QUAD_PRECISION=true
# - If COMPILER_SUPPORTS_QUAD_PRECISION is true, then a corresponding COMPILER_QUAD_TYPE must be given
# This is the compiler's quad-precision data type.
# ** TODO: This isn't complete yet. Finish it. Requires changing macros in kmp_os.h **
set(LIBOMP_COMPILER_SUPPORTS_QUAD_PRECISION false CACHE BOOL
"*INCOMPLETE* Does the compiler support a 128-bit floating point type?")
set(LIBOMP_COMPILER_QUAD_TYPE "" CACHE STRING
"*INCOMPLETE* The quad precision data type (e.g., for gcc, __float128)")
# - Should the orignal build rules for builds be used? (cmake/OriginalBuildRules.cmake). This setting is off by default.
# - This always compiles with -g. And if it is a release build, the debug info is stripped out via objcopy and put into libiomp5.dbg.
set(LIBOMP_USE_BUILDPL_RULES false CACHE BOOL
"Should the build follow build.pl rules/recipes?")
# - Should the build use the predefined linker flags (OS-dependent) in CommonFlags.cmake?
# - these predefined linker flags should work for Windows, Mac, and True Linux for the most popular compilers/linkers
set(LIBOMP_USE_PREDEFINED_LINKER_FLAGS true CACHE BOOL
"Should the build use the predefined linker flags in CommonFlags.cmake?")
I apologise in advance for the size of this check-in. At Intel we do understand that this is not friendly, and are working to change our internal code-development to make it easier to make development features available more frequently and in finer (more functional) chunks. Unfortunately we haven't got that in place yet, and unpicking this into multiple separate check-ins would be non-trivial, so please bear with me on this one. We should be better in the future. Apologies over, what do we have here? GGC 4.9 compatibility -------------------- * We have implemented the new entrypoints used by code compiled by GCC 4.9 to implement the same functionality in gcc 4.8. Therefore code compiled with gcc 4.9 that used to work will continue to do so. However, there are some other new entrypoints (associated with task cancellation) which are not implemented. Therefore user code compiled by gcc 4.9 that uses these new features will not link against the LLVM runtime. (It remains unclear how to handle those entrypoints, since the GCC interface has potentially unpleasant performance implications for join barriers even when cancellation is not used) --- new parallel entry points --- new entry points that aren't OpenMP 4.0 related These are implemented fully :- GOMP_parallel_loop_dynamic() GOMP_parallel_loop_guided() GOMP_parallel_loop_runtime() GOMP_parallel_loop_static() GOMP_parallel_sections() GOMP_parallel() --- cancellation entry points --- Currently, these only give a runtime error if OMP_CANCELLATION is true because our plain barriers don't check for cancellation while waiting GOMP_barrier_cancel() GOMP_cancel() GOMP_cancellation_point() GOMP_loop_end_cancel() GOMP_sections_end_cancel() --- taskgroup entry points --- These are implemented fully. GOMP_taskgroup_start() GOMP_taskgroup_end() --- target entry points --- These are empty (as they are in libgomp) GOMP_target() GOMP_target_data() GOMP_target_end_data() GOMP_target_update() GOMP_teams() Improvements in Barriers and Fork/Join -------------------------------------- * Barrier and fork/join code is now in its own file (which makes it easier to understand and modify). * Wait/release code is now templated and in its own file; suspend/resume code is also templated * There's a new, hierarchical, barrier, which exploits the cache-hierarchy of the Intel(r) Xeon Phi(tm) coprocessor to improve fork/join and barrier performance. ***BEWARE*** the new source files have *not* been added to the legacy Cmake build system. If you want to use that fixes wil be required. Statistics Collection Code -------------------------- * New code has been added to collect application statistics (if this is enabled at library compile time; by default it is not). The statistics code itself is generally useful, the lightweight timing code uses the X86 rdtsc instruction, so will require changes for other architectures. The intent of this code is not for users to tune their codes but rather 1) For timing code-paths inside the runtime 2) For gathering general properties of OpenMP codes to focus attention on which OpenMP features are most used. Nested Hot Teams ---------------- * The runtime now maintains more state to reduce the overhead of creating and destroying inner parallel teams. This improves the performance of code that repeatedly uses nested parallelism with the same resource allocation. Set the new KMP_HOT_TEAMS_MAX_LEVEL envirable to a depth to enable this (and, of course, OMP_NESTED=true to enable nested parallelism at all). Improved Intel(r) VTune(Tm) Amplifier support --------------------------------------------- * The runtime provides additional information to Vtune via the itt_notify interface to allow it to display better OpenMP specific analyses of load-imbalance. Support for OpenMP Composite Statements --------------------------------------- * Implement new entrypoints required by some of the OpenMP 4.1 composite statements. Improved ifdefs --------------- * More separation of concepts ("Does this platform do X?") from platforms ("Are we compiling for platform Y?"), which should simplify future porting. ScaleMP* contribution --------------------- Stack padding to improve the performance in their environment where cross-node coherency is managed at the page level. Redesign of wait and release code --------------------------------- The code is simplified and performance improved. Bug Fixes --------- *Fixes for Windows multiple processor groups. *Fix Fortran module build on Linux: offload attribute added. *Fix entry names for distribute-parallel-loop construct to be consistent with the compiler codegen. *Fix an inconsistent error message for KMP_PLACE_THREADS environment variable. llvm-svn: 219214
2014-10-08 00:25:50 +08:00
# - On multinode systems, larger alignment is desired to avoid false sharing
set(LIBOMP_USE_INTERNODE_ALIGNMENT false CACHE BOOL
"Should larger alignment (4096 bytes) be used for some locks and data structures?")
I apologise in advance for the size of this check-in. At Intel we do understand that this is not friendly, and are working to change our internal code-development to make it easier to make development features available more frequently and in finer (more functional) chunks. Unfortunately we haven't got that in place yet, and unpicking this into multiple separate check-ins would be non-trivial, so please bear with me on this one. We should be better in the future. Apologies over, what do we have here? GGC 4.9 compatibility -------------------- * We have implemented the new entrypoints used by code compiled by GCC 4.9 to implement the same functionality in gcc 4.8. Therefore code compiled with gcc 4.9 that used to work will continue to do so. However, there are some other new entrypoints (associated with task cancellation) which are not implemented. Therefore user code compiled by gcc 4.9 that uses these new features will not link against the LLVM runtime. (It remains unclear how to handle those entrypoints, since the GCC interface has potentially unpleasant performance implications for join barriers even when cancellation is not used) --- new parallel entry points --- new entry points that aren't OpenMP 4.0 related These are implemented fully :- GOMP_parallel_loop_dynamic() GOMP_parallel_loop_guided() GOMP_parallel_loop_runtime() GOMP_parallel_loop_static() GOMP_parallel_sections() GOMP_parallel() --- cancellation entry points --- Currently, these only give a runtime error if OMP_CANCELLATION is true because our plain barriers don't check for cancellation while waiting GOMP_barrier_cancel() GOMP_cancel() GOMP_cancellation_point() GOMP_loop_end_cancel() GOMP_sections_end_cancel() --- taskgroup entry points --- These are implemented fully. GOMP_taskgroup_start() GOMP_taskgroup_end() --- target entry points --- These are empty (as they are in libgomp) GOMP_target() GOMP_target_data() GOMP_target_end_data() GOMP_target_update() GOMP_teams() Improvements in Barriers and Fork/Join -------------------------------------- * Barrier and fork/join code is now in its own file (which makes it easier to understand and modify). * Wait/release code is now templated and in its own file; suspend/resume code is also templated * There's a new, hierarchical, barrier, which exploits the cache-hierarchy of the Intel(r) Xeon Phi(tm) coprocessor to improve fork/join and barrier performance. ***BEWARE*** the new source files have *not* been added to the legacy Cmake build system. If you want to use that fixes wil be required. Statistics Collection Code -------------------------- * New code has been added to collect application statistics (if this is enabled at library compile time; by default it is not). The statistics code itself is generally useful, the lightweight timing code uses the X86 rdtsc instruction, so will require changes for other architectures. The intent of this code is not for users to tune their codes but rather 1) For timing code-paths inside the runtime 2) For gathering general properties of OpenMP codes to focus attention on which OpenMP features are most used. Nested Hot Teams ---------------- * The runtime now maintains more state to reduce the overhead of creating and destroying inner parallel teams. This improves the performance of code that repeatedly uses nested parallelism with the same resource allocation. Set the new KMP_HOT_TEAMS_MAX_LEVEL envirable to a depth to enable this (and, of course, OMP_NESTED=true to enable nested parallelism at all). Improved Intel(r) VTune(Tm) Amplifier support --------------------------------------------- * The runtime provides additional information to Vtune via the itt_notify interface to allow it to display better OpenMP specific analyses of load-imbalance. Support for OpenMP Composite Statements --------------------------------------- * Implement new entrypoints required by some of the OpenMP 4.1 composite statements. Improved ifdefs --------------- * More separation of concepts ("Does this platform do X?") from platforms ("Are we compiling for platform Y?"), which should simplify future porting. ScaleMP* contribution --------------------- Stack padding to improve the performance in their environment where cross-node coherency is managed at the page level. Redesign of wait and release code --------------------------------- The code is simplified and performance improved. Bug Fixes --------- *Fixes for Windows multiple processor groups. *Fix Fortran module build on Linux: offload attribute added. *Fix entry names for distribute-parallel-loop construct to be consistent with the compiler codegen. *Fix an inconsistent error message for KMP_PLACE_THREADS environment variable. llvm-svn: 219214
2014-10-08 00:25:50 +08:00
# - libgomp drop-in compatibility
if(${LINUX} AND NOT ${PPC64})
set(LIBOMP_USE_VERSION_SYMBOLS true CACHE BOOL
"Should version symbols be used? These provide binary compatibility with libgomp.")
I apologise in advance for the size of this check-in. At Intel we do understand that this is not friendly, and are working to change our internal code-development to make it easier to make development features available more frequently and in finer (more functional) chunks. Unfortunately we haven't got that in place yet, and unpicking this into multiple separate check-ins would be non-trivial, so please bear with me on this one. We should be better in the future. Apologies over, what do we have here? GGC 4.9 compatibility -------------------- * We have implemented the new entrypoints used by code compiled by GCC 4.9 to implement the same functionality in gcc 4.8. Therefore code compiled with gcc 4.9 that used to work will continue to do so. However, there are some other new entrypoints (associated with task cancellation) which are not implemented. Therefore user code compiled by gcc 4.9 that uses these new features will not link against the LLVM runtime. (It remains unclear how to handle those entrypoints, since the GCC interface has potentially unpleasant performance implications for join barriers even when cancellation is not used) --- new parallel entry points --- new entry points that aren't OpenMP 4.0 related These are implemented fully :- GOMP_parallel_loop_dynamic() GOMP_parallel_loop_guided() GOMP_parallel_loop_runtime() GOMP_parallel_loop_static() GOMP_parallel_sections() GOMP_parallel() --- cancellation entry points --- Currently, these only give a runtime error if OMP_CANCELLATION is true because our plain barriers don't check for cancellation while waiting GOMP_barrier_cancel() GOMP_cancel() GOMP_cancellation_point() GOMP_loop_end_cancel() GOMP_sections_end_cancel() --- taskgroup entry points --- These are implemented fully. GOMP_taskgroup_start() GOMP_taskgroup_end() --- target entry points --- These are empty (as they are in libgomp) GOMP_target() GOMP_target_data() GOMP_target_end_data() GOMP_target_update() GOMP_teams() Improvements in Barriers and Fork/Join -------------------------------------- * Barrier and fork/join code is now in its own file (which makes it easier to understand and modify). * Wait/release code is now templated and in its own file; suspend/resume code is also templated * There's a new, hierarchical, barrier, which exploits the cache-hierarchy of the Intel(r) Xeon Phi(tm) coprocessor to improve fork/join and barrier performance. ***BEWARE*** the new source files have *not* been added to the legacy Cmake build system. If you want to use that fixes wil be required. Statistics Collection Code -------------------------- * New code has been added to collect application statistics (if this is enabled at library compile time; by default it is not). The statistics code itself is generally useful, the lightweight timing code uses the X86 rdtsc instruction, so will require changes for other architectures. The intent of this code is not for users to tune their codes but rather 1) For timing code-paths inside the runtime 2) For gathering general properties of OpenMP codes to focus attention on which OpenMP features are most used. Nested Hot Teams ---------------- * The runtime now maintains more state to reduce the overhead of creating and destroying inner parallel teams. This improves the performance of code that repeatedly uses nested parallelism with the same resource allocation. Set the new KMP_HOT_TEAMS_MAX_LEVEL envirable to a depth to enable this (and, of course, OMP_NESTED=true to enable nested parallelism at all). Improved Intel(r) VTune(Tm) Amplifier support --------------------------------------------- * The runtime provides additional information to Vtune via the itt_notify interface to allow it to display better OpenMP specific analyses of load-imbalance. Support for OpenMP Composite Statements --------------------------------------- * Implement new entrypoints required by some of the OpenMP 4.1 composite statements. Improved ifdefs --------------- * More separation of concepts ("Does this platform do X?") from platforms ("Are we compiling for platform Y?"), which should simplify future porting. ScaleMP* contribution --------------------- Stack padding to improve the performance in their environment where cross-node coherency is managed at the page level. Redesign of wait and release code --------------------------------- The code is simplified and performance improved. Bug Fixes --------- *Fixes for Windows multiple processor groups. *Fix Fortran module build on Linux: offload attribute added. *Fix entry names for distribute-parallel-loop construct to be consistent with the compiler codegen. *Fix an inconsistent error message for KMP_PLACE_THREADS environment variable. llvm-svn: 219214
2014-10-08 00:25:50 +08:00
else()
set(LIBOMP_USE_VERSION_SYMBOLS false)
I apologise in advance for the size of this check-in. At Intel we do understand that this is not friendly, and are working to change our internal code-development to make it easier to make development features available more frequently and in finer (more functional) chunks. Unfortunately we haven't got that in place yet, and unpicking this into multiple separate check-ins would be non-trivial, so please bear with me on this one. We should be better in the future. Apologies over, what do we have here? GGC 4.9 compatibility -------------------- * We have implemented the new entrypoints used by code compiled by GCC 4.9 to implement the same functionality in gcc 4.8. Therefore code compiled with gcc 4.9 that used to work will continue to do so. However, there are some other new entrypoints (associated with task cancellation) which are not implemented. Therefore user code compiled by gcc 4.9 that uses these new features will not link against the LLVM runtime. (It remains unclear how to handle those entrypoints, since the GCC interface has potentially unpleasant performance implications for join barriers even when cancellation is not used) --- new parallel entry points --- new entry points that aren't OpenMP 4.0 related These are implemented fully :- GOMP_parallel_loop_dynamic() GOMP_parallel_loop_guided() GOMP_parallel_loop_runtime() GOMP_parallel_loop_static() GOMP_parallel_sections() GOMP_parallel() --- cancellation entry points --- Currently, these only give a runtime error if OMP_CANCELLATION is true because our plain barriers don't check for cancellation while waiting GOMP_barrier_cancel() GOMP_cancel() GOMP_cancellation_point() GOMP_loop_end_cancel() GOMP_sections_end_cancel() --- taskgroup entry points --- These are implemented fully. GOMP_taskgroup_start() GOMP_taskgroup_end() --- target entry points --- These are empty (as they are in libgomp) GOMP_target() GOMP_target_data() GOMP_target_end_data() GOMP_target_update() GOMP_teams() Improvements in Barriers and Fork/Join -------------------------------------- * Barrier and fork/join code is now in its own file (which makes it easier to understand and modify). * Wait/release code is now templated and in its own file; suspend/resume code is also templated * There's a new, hierarchical, barrier, which exploits the cache-hierarchy of the Intel(r) Xeon Phi(tm) coprocessor to improve fork/join and barrier performance. ***BEWARE*** the new source files have *not* been added to the legacy Cmake build system. If you want to use that fixes wil be required. Statistics Collection Code -------------------------- * New code has been added to collect application statistics (if this is enabled at library compile time; by default it is not). The statistics code itself is generally useful, the lightweight timing code uses the X86 rdtsc instruction, so will require changes for other architectures. The intent of this code is not for users to tune their codes but rather 1) For timing code-paths inside the runtime 2) For gathering general properties of OpenMP codes to focus attention on which OpenMP features are most used. Nested Hot Teams ---------------- * The runtime now maintains more state to reduce the overhead of creating and destroying inner parallel teams. This improves the performance of code that repeatedly uses nested parallelism with the same resource allocation. Set the new KMP_HOT_TEAMS_MAX_LEVEL envirable to a depth to enable this (and, of course, OMP_NESTED=true to enable nested parallelism at all). Improved Intel(r) VTune(Tm) Amplifier support --------------------------------------------- * The runtime provides additional information to Vtune via the itt_notify interface to allow it to display better OpenMP specific analyses of load-imbalance. Support for OpenMP Composite Statements --------------------------------------- * Implement new entrypoints required by some of the OpenMP 4.1 composite statements. Improved ifdefs --------------- * More separation of concepts ("Does this platform do X?") from platforms ("Are we compiling for platform Y?"), which should simplify future porting. ScaleMP* contribution --------------------- Stack padding to improve the performance in their environment where cross-node coherency is managed at the page level. Redesign of wait and release code --------------------------------- The code is simplified and performance improved. Bug Fixes --------- *Fixes for Windows multiple processor groups. *Fix Fortran module build on Linux: offload attribute added. *Fix entry names for distribute-parallel-loop construct to be consistent with the compiler codegen. *Fix an inconsistent error message for KMP_PLACE_THREADS environment variable. llvm-svn: 219214
2014-10-08 00:25:50 +08:00
endif()
# - TSX based locks have __asm code which can be troublesome for some compilers. This feature is also x86 specific.
I apologise in advance for the size of this check-in. At Intel we do understand that this is not friendly, and are working to change our internal code-development to make it easier to make development features available more frequently and in finer (more functional) chunks. Unfortunately we haven't got that in place yet, and unpicking this into multiple separate check-ins would be non-trivial, so please bear with me on this one. We should be better in the future. Apologies over, what do we have here? GGC 4.9 compatibility -------------------- * We have implemented the new entrypoints used by code compiled by GCC 4.9 to implement the same functionality in gcc 4.8. Therefore code compiled with gcc 4.9 that used to work will continue to do so. However, there are some other new entrypoints (associated with task cancellation) which are not implemented. Therefore user code compiled by gcc 4.9 that uses these new features will not link against the LLVM runtime. (It remains unclear how to handle those entrypoints, since the GCC interface has potentially unpleasant performance implications for join barriers even when cancellation is not used) --- new parallel entry points --- new entry points that aren't OpenMP 4.0 related These are implemented fully :- GOMP_parallel_loop_dynamic() GOMP_parallel_loop_guided() GOMP_parallel_loop_runtime() GOMP_parallel_loop_static() GOMP_parallel_sections() GOMP_parallel() --- cancellation entry points --- Currently, these only give a runtime error if OMP_CANCELLATION is true because our plain barriers don't check for cancellation while waiting GOMP_barrier_cancel() GOMP_cancel() GOMP_cancellation_point() GOMP_loop_end_cancel() GOMP_sections_end_cancel() --- taskgroup entry points --- These are implemented fully. GOMP_taskgroup_start() GOMP_taskgroup_end() --- target entry points --- These are empty (as they are in libgomp) GOMP_target() GOMP_target_data() GOMP_target_end_data() GOMP_target_update() GOMP_teams() Improvements in Barriers and Fork/Join -------------------------------------- * Barrier and fork/join code is now in its own file (which makes it easier to understand and modify). * Wait/release code is now templated and in its own file; suspend/resume code is also templated * There's a new, hierarchical, barrier, which exploits the cache-hierarchy of the Intel(r) Xeon Phi(tm) coprocessor to improve fork/join and barrier performance. ***BEWARE*** the new source files have *not* been added to the legacy Cmake build system. If you want to use that fixes wil be required. Statistics Collection Code -------------------------- * New code has been added to collect application statistics (if this is enabled at library compile time; by default it is not). The statistics code itself is generally useful, the lightweight timing code uses the X86 rdtsc instruction, so will require changes for other architectures. The intent of this code is not for users to tune their codes but rather 1) For timing code-paths inside the runtime 2) For gathering general properties of OpenMP codes to focus attention on which OpenMP features are most used. Nested Hot Teams ---------------- * The runtime now maintains more state to reduce the overhead of creating and destroying inner parallel teams. This improves the performance of code that repeatedly uses nested parallelism with the same resource allocation. Set the new KMP_HOT_TEAMS_MAX_LEVEL envirable to a depth to enable this (and, of course, OMP_NESTED=true to enable nested parallelism at all). Improved Intel(r) VTune(Tm) Amplifier support --------------------------------------------- * The runtime provides additional information to Vtune via the itt_notify interface to allow it to display better OpenMP specific analyses of load-imbalance. Support for OpenMP Composite Statements --------------------------------------- * Implement new entrypoints required by some of the OpenMP 4.1 composite statements. Improved ifdefs --------------- * More separation of concepts ("Does this platform do X?") from platforms ("Are we compiling for platform Y?"), which should simplify future porting. ScaleMP* contribution --------------------- Stack padding to improve the performance in their environment where cross-node coherency is managed at the page level. Redesign of wait and release code --------------------------------- The code is simplified and performance improved. Bug Fixes --------- *Fixes for Windows multiple processor groups. *Fix Fortran module build on Linux: offload attribute added. *Fix entry names for distribute-parallel-loop construct to be consistent with the compiler codegen. *Fix an inconsistent error message for KMP_PLACE_THREADS environment variable. llvm-svn: 219214
2014-10-08 00:25:50 +08:00
if(${IA32} OR ${INTEL64})
set(LIBOMP_USE_ADAPTIVE_LOCKS true CACHE BOOL
"Should TSX-based lock be compiled (adaptive lock in kmp_lock.cpp). These are x86 specific.")
else()
set(LIBOMP_USE_ADAPTIVE_LOCKS false)
endif()
##################################
# Error checking the configuration
if(${LIBOMP_STATS} AND (${WINDOWS} OR ${MAC}))
error_say("Stats-gathering functionality is only supported on x86-Linux and Intel(R) MIC Architecture")
endif()
if(${LIBOMP_STATS} AND NOT (${IA32} OR ${INTEL64} OR ${MIC}))
error_say("Stats-gathering functionality is only supported on x86-Linux and Intel(R) MIC Architecture")
endif()
if(${LIBOMP_USE_ADAPTIVE_LOCKS} AND NOT(${IA32} OR ${INTEL64}))
error_say("Adaptive locks (TSX) functionality is only supported on x86 Architecture")
endif()
###############################################
# - Create the suffix for the export directory
# - Only add to suffix when not a default value
# - Example suffix: .deb.30.s1
# final export directory: exports/lin_32e.deb.30.s1/lib
# - These suffixes imply the build is a Debug, OpenMP 3.0, Stats-Gathering version of the library
if(NOT "${cmake_build_type_lowercase}" STREQUAL "release")
string(SUBSTRING "${cmake_build_type_lowercase}" 0 3 build_type_suffix)
set(suffix "${suffix}.${build_type_suffix}")
endif()
if(NOT "${LIBOMP_OMP_VERSION}" STREQUAL "41")
set(suffix "${suffix}.${LIBOMP_OMP_VERSION}")
endif()
if(${LIBOMP_STATS})
set(suffix "${suffix}.s1")
endif()
if(${LIBOMP_OMPT_SUPPORT})
set(suffix "${suffix}.ompt")
if(NOT ${LIBOMP_OMPT_BLAME})
set(suffix "${suffix}.no-ompt-blame")
endif()
if(NOT ${LIBOMP_OMPT_TRACE})
set(suffix "${suffix}.no-ompt-trace")
endif()
endif()
####################################
# Setting file extensions / suffixes
set(obj ${CMAKE_C_OUTPUT_EXTENSION})
set(lib ${CMAKE_STATIC_LIBRARY_SUFFIX})
set(dll ${CMAKE_SHARED_LIBRARY_SUFFIX})
set(exe ${CMAKE_EXECUTABLE_SUFFIX})
######################
# Find perl executable
# Perl is used to create omp.h (and other headers) along with kmp_i18n_id.inc and kmp_i18n_default.inc (see below in Rules section)
if(NOT "${PERL_FOUND}") # variable is defined in FindPerl Standard CMake Module
error_say("Error: Could not find valid perl")
endif()
#########################
# Setting directory names
if(${MIC})
set(platform "${real_os}_${LIBOMP_MIC_ARCH}") # e.g., lin_knf, lin_knc
else()
set(platform "${real_os}_${LIBOMP_ARCH}") # e.g., lin_32e, mac_32
endif()
# build directory (Where CMakeCache.txt is created, build files generated)
set(build_dir "${CMAKE_CURRENT_BINARY_DIR}")
set(src_dir "${CMAKE_CURRENT_SOURCE_DIR}/src")
set(tools_dir "${CMAKE_CURRENT_SOURCE_DIR}/tools")
set(export_dir "${CMAKE_CURRENT_SOURCE_DIR}/exports")
set(export_ptf_dir "${export_dir}/${platform}${suffix}")
set(export_cmn_dir1 "${export_dir}/common${suffix}/include")
set(export_cmn_dir2 "${export_dir}/common${suffix}/include_compat")
set(export_inc_dir "${export_ptf_dir}/include")
set(export_mod_dir "${export_ptf_dir}/include_compat")
_export_lib_dir(${platform} export_lib_dir) # set exports directory (relative to build_dir) e.g., ../exports/lin_32e/lib/
# or ../exports/mac_32e/lib.thin/ for mac
if(${MAC})
# macs use lib.thin/ subdirectory for non-fat libraries that only contain one architecture
# macs use lib/ subdirectory for fat libraries that contain both IA-32 architecture and Intel(R) 64 architecture code.
_export_lib_fat_dir(${platform} export_lib_fat_dir)
endif()
set(inc_dir "${src_dir}/include/${LIBOMP_OMP_VERSION}")
############################
# Setting final library name
set(lib_item "libiomp")
if(${PROFILE_LIBRARY})
set(lib_item "${lib_item}prof")
endif()
if(${STUBS_LIBRARY})
set(lib_item "${lib_item}stubs")
endif()
set(lib_item "${lib_item}${LIBOMP_VERSION}")
if(${WINDOWS})
set(lib_item "${lib_item}md")
endif()
set(LIBOMP_LIB_NAME "${lib_item}" CACHE STRING "OMP library name")
set(lib_ext "${dll}")
# ${lib_file} is real library name:
# libiomp5.so for Linux
# libiomp5.dylib for Mac
# libiomp5md.dll for Windows
set(lib_file "${LIBOMP_LIB_NAME}${lib_ext}")
########################################
# Setting export file names
if(${WINDOWS})
set(imp_file "${lib_item}${lib}") # this is exported (libiomp5md.lib)
set(def_file "${lib_item}.def") # this is not exported
set(rc_file "${lib_item}.rc") # this is not exported
if("${CMAKE_BUILD_TYPE}" STREQUAL "Debug" OR
"${CMAKE_BUILD_TYPE}" STREQUAL "RelWithDebInfo" OR
${LIBOMP_USE_BUILDPL_RULES})
set(pdb_file "${lib_file}.pdb") # this is exported if it exists (libiomp5md.dll.pdb)
endif()
endif()
set(export_lib_files "${lib_file}" "${imp_file}" "${pdb_file}")
set(export_inc_files "iomp_lib.h")
set(export_mod_files "omp_lib.mod" "omp_lib_kinds.mod")
set(export_cmn_files1 "omp.h" "omp_lib.h" "omp_lib.f" "omp_lib.f90")
set(export_cmn_files2 "iomp.h")
if(${LIBOMP_OMPT_SUPPORT})
set(export_cmn_files1 ${export_cmn_files1} "ompt.h")
endif()
set(export_cmn_files "${export_cmn_files1}" "${export_cmn_files2}")
if("${export_lib_fat_dir}")
set(export_lib_fat_files "${lib_file}" "${imp_file}")
endif()
#########################
# Getting legal type/arch
set_legal_type(legal_type)
set_legal_arch(legal_arch)
#################################################
# Preprocessor Definitions (cmake/Definitions.cmake)
# Preprocessor Includes
# Compiler (C/C++) Flags (cmake/CommonFlags.cmake)
# Assembler Flags (cmake/CommonFlags.cmake)
# Fortran Flags (cmake/CommonFlags.cmake)
# Linker Flags (cmake/CommonFlags.cmake)
# Archiver Flags (cmake/CommonFlags.cmake)
# Helper Perl Script Flags (cmake/PerlFlags.cmake)
# * Inside the cmake/CommonFlags.cmake file, the LIBOMP_*FLAGS are added.
# * Cannot use CMAKE_*_FLAGS directly because -x c++ is put in the linker command and mangles the linking phase.
# preprocessor flags (-D definitions and -I includes)
# Grab environment variable CPPFLAGS and append those to definitions
set(include_dirs ${CMAKE_CURRENT_BINARY_DIR} ${src_dir} ${src_dir}/i18n ${inc_dir} ${src_dir}/thirdparty/ittnotify)
include_directories(${include_dirs})
# Grab compiler-dependent flags
# Cmake will look for cmake/${CMAKE_C_COMPILER_ID}/CFlags.cmake to append additional c, cxx, and linker flags.
set(CMAKE_MODULE_PATH ${CMAKE_CURRENT_SOURCE_DIR}/cmake/${CMAKE_C_COMPILER_ID} ${CMAKE_MODULE_PATH})
find_file(compiler_specific_include_file_found CFlags.cmake ${CMAKE_MODULE_PATH})
if(compiler_specific_include_file_found)
include(CFlags) # COMPILER_SUPPORTS_QUAD_PRECISION changed in here
append_compiler_specific_c_and_cxx_flags(C_FLAGS CXX_FLAGS)
append_compiler_specific_linker_flags(LD_FLAGS LD_LIB_FLAGS)
else()
warning_say("Could not find cmake/${CMAKE_C_COMPILER_ID}/CFlags.cmake: will only use default flags")
endif()
# Grab assembler-dependent flags
# CMake will look for cmake/${CMAKE_ASM_COMPILER_ID}/AsmFlags.cmake to append additional assembler flags.
if(${WINDOWS})
# Windows based systems use CMAKE_ASM_MASM_COMPILER
# The windows assembly files are in MASM format, and they require a tool that can handle MASM syntax (ml.exe or ml64.exe typically)
enable_language(ASM_MASM)
set(CMAKE_MODULE_PATH ${CMAKE_CURRENT_SOURCE_DIR}/cmake/${CMAKE_ASM_MASM_COMPILER_ID} ${CMAKE_MODULE_PATH})
find_file(assembler_specific_include_file_found AsmFlags.cmake ${CMAKE_MODULE_PATH})
if(assembler_specific_include_file_found)
include(AsmFlags)
append_assembler_specific_asm_flags(ASM_FLAGS)
else()
warning_say("Could not find cmake/${CMAKE_ASM_MASM_COMPILER_ID}/AsmFlags.cmake: will only use default flags")
endif()
else()
# Unix (including Mac) based systems use CMAKE_ASM_COMPILER
# Unix assembly files can be handled by compiler usually.
find_file(assembler_specific_include_file_found AsmFlags.cmake ${CMAKE_MODULE_PATH})
if(assembler_specific_include_file_found)
include(AsmFlags)
append_assembler_specific_asm_flags(ASM_FLAGS)
else()
warning_say("Could not find cmake/${CMAKE_ASM_COMPILER_ID}/AsmFlags.cmake: will only use default flags")
endif()
endif()
# Grab all the compiler-independent flags
append_c_and_cxx_flags_common(C_FLAGS CXX_FLAGS)
append_asm_flags_common(ASM_FLAGS)
append_fort_flags_common(F_FLAGS)
append_linker_flags_common(LD_FLAGS LD_LIB_FLAGS)
append_archiver_flags_common(AR_FLAGS)
append_cpp_flags(DEFINITIONS_FLAGS)
# Setup the flags correctly for cmake (covert to string)
# Pretty them up (STRIP any beginning and trailing whitespace)
list_to_string("${DEFINITIONS_FLAGS}" DEFINITIONS_FLAGS)
list_to_string("${C_FLAGS}" C_FLAGS)
list_to_string("${CXX_FLAGS}" CXX_FLAGS)
list_to_string("${ASM_FLAGS}" ASM_FLAGS)
list_to_string("${LD_FLAGS}" LD_FLAGS)
list_to_string("${LD_LIB_FLAGS}" LD_LIB_FLAGS)
# Windows specific for creating import library
list_to_string("${AR_FLAGS}" AR_FLAGS)
string(STRIP "${DEFINITIONS_FLAGS}" DEFINITIONS_FLAGS)
string(STRIP "${C_FLAGS}" C_FLAGS)
string(STRIP "${CXX_FLAGS}" CXX_FLAGS)
string(STRIP "${ASM_FLAGS}" ASM_FLAGS)
string(STRIP "${LD_FLAGS}" LD_FLAGS)
string(STRIP "${LD_LIB_FLAGS}" LD_LIB_FLAGS)
# Windows specific for creating import library
string(STRIP "${AR_FLAGS}" AR_FLAGS)
# Grab the Perl flags
set_ev_flags(ev_flags) # expand-vars.pl flags
set_gd_flags(gd_flags) # generate-def.pl flags (Windows only)
set(oa_opts "--os=${real_os}" "--arch=${LIBOMP_ARCH}") # sent to the perl scripts
#########################################################
# Getting correct source files (cmake/SourceFiles.cmake)
set_c_files(lib_c_items)
set_cpp_files(lib_cxx_items)
set_asm_files(lib_asm_items)
set_imp_c_files(imp_c_items) # Windows-specific
###################################
# Setting all source file variables
set(lib_src_files "${lib_c_items}" "${lib_cxx_items}" "${lib_asm_items}")
set(imp_src_files "${imp_c_items}")
add_prefix("${src_dir}/" lib_src_files)
add_prefix("${src_dir}/" imp_src_files) # Windows-specific
add_prefix("${src_dir}/" lib_c_items)
add_prefix("${src_dir}/" lib_cxx_items)
add_prefix("${src_dir}/" lib_asm_items)
add_prefix("${src_dir}/" imp_c_items) # Windows-specific
#####################################################################
# Debug print outs. Will print "variable = ${variable}" if GLOBAL_DEBUG == 1
if(GLOBAL_DEBUG)
include(CMakePrintSystemInformation)
endif()
debug_say_var(CMAKE_ASM_COMPILE_OBJECT)
debug_say_var(CMAKE_RC_COMPILER)
debug_say_var(CMAKE_C_COMPILER_ID)
debug_say_var(date)
debug_say_var(LIBOMP_STATS)
debug_say_var(lib_file)
debug_say_var(export_lib_files)
debug_say_var(DEFINITIONS_FLAGS)
debug_say_var(C_FLAGS)
debug_say_var(CXX_FLAGS)
debug_say_var(ASM_FLAGS)
debug_say_var(F_FLAGS)
debug_say_var(LD_FLAGS)
debug_say_var(LD_LIB_FLAGS)
debug_say_var(AR_FLAGS)
debug_say_var(ev_flags)
debug_say_var(gd_flags)
debug_say_var(oa_opts)
debug_say_var(lib_c_items)
debug_say_var(lib_cxx_items)
debug_say_var(lib_asm_items)
debug_say_var(imp_c_items)
debug_say_var(lib_src_files)
debug_say_var(imp_src_files)
####################################################################
# --------------------- #
# --- Rules/Recipes --- #
# --------------------- #
####################################################################
# Below, ${ldeps} always stands for "local dependencies" for the
# next immediate target to be created via add_custom_target() or
# add_custom_command()
####################
# --- Create all ---
add_custom_target(lib ALL DEPENDS iomp5)
add_custom_target(inc ALL DEPENDS ${export_inc_files})
if(${LIBOMP_FORTRAN_MODULES})
add_custom_target(mod ALL DEPENDS ${export_mod_files})
endif()
#############################
# --- Create Common Files ---
add_custom_target(common ALL DEPENDS ${export_cmn_files})
add_custom_target(clean-common COMMAND ${CMAKE_COMMAND} -E remove -f ${export_cmn_files})
# --- Put headers in convenient locations post build ---
if(${LIBOMP_COPY_EXPORTS})
add_custom_command(TARGET common POST_BUILD
COMMAND ${CMAKE_COMMAND} -E make_directory ${export_cmn_dir1}
COMMAND ${CMAKE_COMMAND} -E make_directory ${export_cmn_dir2}
COMMAND ${CMAKE_COMMAND} -E copy omp.h ${export_cmn_dir1}
COMMAND ${CMAKE_COMMAND} -E copy omp_lib.h ${export_cmn_dir1}
COMMAND ${CMAKE_COMMAND} -E copy omp_lib.f ${export_cmn_dir1}
COMMAND ${CMAKE_COMMAND} -E copy omp_lib.f90 ${export_cmn_dir1}
COMMAND ${CMAKE_COMMAND} -E copy iomp.h ${export_cmn_dir2}
)
if(${LIBOMP_OMPT_SUPPORT})
add_custom_command(TARGET common POST_BUILD
COMMAND ${CMAKE_COMMAND} -E make_directory ${export_cmn_dir1}
COMMAND ${CMAKE_COMMAND} -E copy ompt.h ${export_cmn_dir1}
)
endif()
add_custom_command(TARGET inc POST_BUILD
COMMAND ${CMAKE_COMMAND} -E make_directory ${export_inc_dir}
COMMAND ${CMAKE_COMMAND} -E copy iomp_lib.h ${export_inc_dir}
)
if(${LIBOMP_FORTRAN_MODULES})
add_custom_command(TARGET mod POST_BUILD
COMMAND ${CMAKE_COMMAND} -E make_directory ${export_mod_dir}
COMMAND ${CMAKE_COMMAND} -E copy omp_lib.mod ${export_mod_dir}
COMMAND ${CMAKE_COMMAND} -E copy omp_lib_kinds.mod ${export_mod_dir}
)
endif()
endif()
######################################################
# --- Build the main library ---
# $(lib_file) <== Main library file to create
# objects depend on : .inc files and omp.h
# This way the *.inc and omp.h are generated before any compilations take place
add_custom_target(needed-headers DEPENDS ${build_dir}/kmp_i18n_id.inc ${build_dir}/kmp_i18n_default.inc ${build_dir}/omp.h)
# For Windows, there is a definitions file (.def) and resource file (.res) created using generate-def.pl and rc.exe respectively.
if(${WINDOWS})
add_custom_target(needed-windows-files DEPENDS ${build_dir}/${def_file} ${build_dir}/${rc_file})
list(APPEND lib_src_files ${build_dir}/${rc_file})
endif()
# Remove any cmake-automatic linking of libraries by linker, This is so linux
# and mac don't include libstdc++ just because we compile c++ files.
if(${LIBOMP_USE_PREDEFINED_LINKER_FLAGS})
set(CMAKE_C_IMPLICIT_LINK_LIBRARIES "")
set(CMAKE_CXX_IMPLICIT_LINK_LIBRARIES "")
set(CMAKE_ASM_IMPLICIT_LINK_LIBRARIES "")
endif()
# --- ${lib_file} rule ---
add_library(iomp5 SHARED ${lib_src_files})
set_target_properties(iomp5 PROPERTIES
PREFIX "" SUFFIX "" # Take control
OUTPUT_NAME "${lib_file}" # of output name
LINK_FLAGS "${LD_FLAGS}"
LINKER_LANGUAGE C # use C Compiler for linking step
SKIP_BUILD_RPATH true # have Mac linker -install_name just be "-install_name libiomp5.dylib"
)
# --- Copy libiomp into exports directory post build ---
if(${WINDOWS})
get_target_property(LIBOMP_OUTPUT_DIRECTORY iomp5 RUNTIME_OUTPUT_DIRECTORY)
else()
get_target_property(LIBOMP_OUTPUT_DIRECTORY iomp5 LIBRARY_OUTPUT_DIRECTORY)
endif()
if(NOT LIBOMP_OUTPUT_DIRECTORY)
set(LIBOMP_OUTPUT_DIRECTORY ${build_dir})
endif()
if(${LIBOMP_COPY_EXPORTS})
add_custom_command(TARGET iomp5 POST_BUILD
COMMAND ${CMAKE_COMMAND} -E make_directory ${export_lib_dir}
COMMAND ${CMAKE_COMMAND} -E copy ${LIBOMP_OUTPUT_DIRECTORY}/${lib_file} ${export_lib_dir}
)
endif()
# Linking command will include libraries in LD_LIB_FLAGS
target_link_libraries(iomp5 ${LD_LIB_FLAGS} ${CMAKE_DL_LIBS})
# Create *.inc and omp.h before compiling any sources
add_dependencies(iomp5 needed-headers)
if(${WINDOWS})
# Create .def and .rc file before compiling any sources
add_dependencies(iomp5 needed-windows-files)
endif()
# Set the compiler flags for each type of source
set_source_files_properties(${lib_c_items}
${imp_c_items} PROPERTIES COMPILE_FLAGS "${C_FLAGS}")
set_source_files_properties(${lib_cxx_items} PROPERTIES COMPILE_FLAGS "${CXX_FLAGS}")
if(${WINDOWS})
# Windows operating system has to use MASM assembler
set_source_files_properties(${lib_asm_items} PROPERTIES COMPILE_FLAGS "${ASM_FLAGS}")
else()
# Non-Windows operating systems can use compiler to assemble the assembly files
set_source_files_properties(${lib_asm_items} PROPERTIES COMPILE_FLAGS "${ASM_FLAGS}" LANGUAGE C)
endif()
# Set the -D definitions for all sources
add_definitions(${DEFINITIONS_FLAGS})
# If creating a build that imitates build.pl's rules then set USE_BUILDPL_RULES to true
if(${LIBOMP_USE_BUILDPL_RULES})
include(BuildPLRules)
endif()
######################################################
# --- Source file specific flags ---
# kmp_version.o : -D _KMP_BUILD_TIME="\"$(date)}\""
set_source_files_properties(${src_dir}/kmp_version.c PROPERTIES COMPILE_DEFINITIONS "_KMP_BUILD_TIME=\"\\\"${date}\\\"\"")
# z_Linux_asm.o : -D KMP_ARCH_*
if(${ARM})
set_source_files_properties(${src_dir}/z_Linux_asm.s PROPERTIES COMPILE_DEFINITIONS "KMP_ARCH_ARM")
elseif(${AARCH64})
set_source_files_properties(${src_dir}/z_Linux_asm.s PROPERTIES COMPILE_DEFINITIONS "KMP_ARCH_AARCH64")
elseif(${INTEL64} OR ${MIC})
set_source_files_properties(${src_dir}/z_Linux_asm.s PROPERTIES COMPILE_DEFINITIONS "KMP_ARCH_X86_64")
elseif(${IA32})
set_source_files_properties(${src_dir}/z_Linux_asm.s PROPERTIES COMPILE_DEFINITIONS "KMP_ARCH_X86")
elseif(${PPC64})
set_source_files_properties(${src_dir}/z_Linux_asm.s PROPERTIES COMPILE_DEFINITIONS "KMP_ARCH_PPC64")
endif()
if(${WINDOWS})
set_source_files_properties(${src_dir}/thirdparty/ittnotify/ittnotify_static.c PROPERTIES COMPILE_DEFINITIONS "UNICODE")
endif()
######################################################
# MAC specific build rules
if(${MAC})
# fat library rules
if(${INTEL64})
_export_lib_fat_dir("mac_32e" export_fat_mac_32e)
_export_lib_dir("mac_32" export_mac_32)
_export_lib_dir("mac_32e" export_mac_32e)
add_custom_target(fat
COMMAND ${CMAKE_COMMAND} -E echo Building 32 and 32e fat libraries from ${export_mac_32}/${lib_file} and ${export_mac_32e}/${lib_file}
COMMAND ${CMAKE_COMMAND} -E echo Will put fat library in ${export_fat_mac_32e} directory
COMMAND ${CMAKE_COMMAND} -E make_directory ${export_fat_mac_32e}
COMMAND lipo -create -output ${export_fat_mac_32e}/${lib_file} ${export_mac_32}/${lib_file} ${export_mac_32e}/${lib_file}
)
endif()
endif()
######################################################
# Windows specific build rules
if(${WINDOWS})
# --- Create $(imp_file) ---
# This file is first created in the unstripped/${lib_file} creation step.
# It is then "re-linked" to include kmp_import.c which prevents linking of both Visual Studio OpenMP and newly built OpenMP
if(NOT "${imp_file}" STREQUAL "")
set(generated_import_file ${lib_file}${lib})
add_library(iomp5imp STATIC ${generated_import_file} ${imp_src_files})
set_source_files_properties(${generated_import_file} PROPERTIES GENERATED TRUE EXTERNAL_OBJECT TRUE)
set_target_properties(iomp5imp PROPERTIES
PREFIX "" SUFFIX ""
OUTPUT_NAME "${imp_file}"
STATIC_LIBRARY_FLAGS "${AR_FLAGS}"
LINKER_LANGUAGE C
SKIP_BUILD_RPATH true
)
add_custom_command(TARGET iomp5imp PRE_BUILD COMMAND ${CMAKE_COMMAND} -E remove -f ${imp_file})
add_dependencies(iomp5imp iomp5)
get_target_property(LIBOMPIMP_OUTPUT_DIRECTORY iomp5imp ARCHIVE_OUTPUT_DIRECTORY)
if(NOT LIBOMPIMP_OUTPUT_DIRECTORY)
set(LIBOMPIMP_OUTPUT_DIRECTORY ${build_dir})
endif()
if(${LIBOMP_COPY_EXPORTS})
add_custom_command(TARGET iomp5imp POST_BUILD
COMMAND ${CMAKE_COMMAND} -E make_directory ${export_lib_dir}
COMMAND ${CMAKE_COMMAND} -E copy ${LIBOMPIMP_OUTPUT_DIRECTORY}/${imp_file} ${export_lib_dir}
)
endif()
endif()
# --- Create $(def_file) ---
if(NOT "${def_file}" STREQUAL "")
string_to_list("${gd_flags}" gd_flags)
add_custom_command(
OUTPUT ${def_file}
COMMAND ${PERL_EXECUTABLE} ${tools_dir}/generate-def.pl ${gd_flags} -o ${def_file} ${src_dir}/dllexports
DEPENDS ${src_dir}/dllexports ${tools_dir}/generate-def.pl
)
endif()
# --- Create $(rc_file) ---
if(NOT "${rc_file}" STREQUAL "")
add_custom_command(
OUTPUT ${rc_file}
COMMAND ${CMAKE_COMMAND} -E copy libiomp.rc ${rc_file}
DEPENDS libiomp.rc
)
endif()
endif()
######################################################
# kmp_i18n_id.inc and kmp_i18n_default.inc
set(perlcmd "${PERL_EXECUTABLE}" "${tools_dir}/message-converter.pl" "${oa_opts}" "--prefix=kmp_i18n" "--enum=kmp_i18n_id.inc" "${src_dir}/i18n/en_US.txt")
add_custom_command(
OUTPUT ${build_dir}/kmp_i18n_id.inc
COMMAND ${perlcmd}
DEPENDS ${src_dir}/i18n/en_US.txt ${tools_dir}/message-converter.pl
)
set(perlcmd "${PERL_EXECUTABLE}" "${tools_dir}/message-converter.pl" "${oa_opts}" "--prefix=kmp_i18n" "--default=kmp_i18n_default.inc" "${src_dir}/i18n/en_US.txt")
add_custom_command(
OUTPUT ${build_dir}/kmp_i18n_default.inc
COMMAND ${perlcmd}
DEPENDS ${src_dir}/i18n/en_US.txt ${tools_dir}/message-converter.pl
)
######################################################
# Micro test rules for after library has been built (cmake/MicroTests.cmake)
# - Only perform if ${tests} == true (specify when invoking: cmake -Dtests=on ...)
if(${LIBOMP_MICRO_TESTS})
include(MicroTests)
add_custom_target(libiomp-micro-tests)
if(NOT ${MIC} AND ${LIBOMP_TEST_TOUCH})
add_dependencies(libiomp-micro-tests libiomp-test-touch)
endif()
if(${LINUX} AND ${LIBOMP_TEST_RELO})
add_dependencies(libiomp-micro-tests libiomp-test-relo)
endif()
if(${LINUX} AND ${LIBOMP_TEST_EXECSTACK})
add_dependencies(libiomp-micro-tests libiomp-test-execstack)
endif()
if(${MIC} AND ${LIBOMP_TEST_INSTR})
add_dependencies(libiomp-micro-tests libiomp-test-instr)
endif()
if(${LIBOMP_TEST_DEPS})
add_dependencies(libiomp-micro-tests libiomp-test-deps)
endif()
endif()
######################################################
# --- Create Fortran Files ---
# omp_lib.mod
if(${LIBOMP_FORTRAN_MODULES})
# Grab fortran-compiler-dependent flags
# Cmake will look for cmake/${CMAKE_Fortran_COMPILER_ID}/FortranFlags.cmake to append additional fortran flags.
enable_language(Fortran)
set(CMAKE_MODULE_PATH ${CMAKE_CURRENT_SOURCE_DIR}/cmake/${CMAKE_Fortran_COMPILER_ID} ${CMAKE_MODULE_PATH})
find_file(fortran_specific_include_file_found FortranFlags.cmake ${CMAKE_MODULE_PATH})
if(fortran_specific_include_file_found)
include(FortranFlags)
append_fortran_compiler_specific_fort_flags(F_FLAGS)
else()
warning_say("Could not find cmake/${CMAKE_Fortran_COMPILER_ID}/FortranFlags.cmake: will only use default flags in CommonFlags.cmake")
endif()
set(omp_lib_f "omp_lib.f90" )
add_custom_command(
OUTPUT "omp_lib.mod"
COMMAND ${CMAKE_Fortran_COMPILER} -c ${F_FLAGS} ${omp_lib_f}
DEPENDS ${omp_lib_f}
)
add_custom_command(
OUTPUT "omp_lib_kinds.mod"
COMMAND ${CMAKE_Fortran_COMPILER} -c ${F_FLAGS} ${omp_lib_f}
DEPENDS ${omp_lib_f}
)
# clean omp_lib.o from build directory when "make clean"
set_directory_properties(PROPERTIES ADDITIONAL_MAKE_CLEAN_FILES omp_lib${obj})
endif()
###############################################################
# --- Using expand-vars.pl to generate files ---
# - 'file' is generated using expand-vars.pl and 'file'.var
# - Any .h .f .f90 .rc files should be created with this recipe
macro(expand_vars_recipe file_dir filename)
get_source_file_property(extra_ev_flags ${filename} COMPILE_DEFINITIONS)
if("${extra_ev_flags}" MATCHES "NOTFOUND")
set(extra_ev_flags)
else()
string_to_list("${extra_ev_flags}" extra_ev_flags)
endif()
if(NOT "${filename}" STREQUAL "")
add_custom_command(
OUTPUT ${filename}
COMMAND ${PERL_EXECUTABLE} ${tools_dir}/expand-vars.pl --strict ${ev_flags} ${extra_ev_flags} ${file_dir}/${filename}.var ${filename}
DEPENDS ${file_dir}/${filename}.var ${src_dir}/kmp_version.c ${tools_dir}/expand-vars.pl
)
endif()
endmacro()
string_to_list("${ev_flags}" ev_flags)
# omp_lib.h : ev-flags += -D KMP_INT_PTR_KIND="int_ptr_kind()"
set_source_files_properties(omp_lib.h PROPERTIES COMPILE_DEFINITIONS "-D KMP_INT_PTR_KIND=\"int_ptr_kind()\"")
# iomp_lib.h : ev-flags += -D KMP_INT_PTR_KIND=$(if $(filter 32,$(arch)),4,8)
if(${IA32}) # 32 bit archs
set_source_files_properties(iomp_lib.h PROPERTIES COMPILE_DEFINITIONS "-D KMP_INT_PTR_KIND=4")
else()
set_source_files_properties(iomp_lib.h PROPERTIES COMPILE_DEFINITIONS "-D KMP_INT_PTR_KIND=8")
endif()
# libiomp.rc : ev-flags += -D KMP_FILE=$(lib_file)
set_source_files_properties(libiomp.rc PROPERTIES COMPILE_DEFINITIONS "-D KMP_FILE=${lib_file}")
expand_vars_recipe(${src_dir}/include/${LIBOMP_OMP_VERSION} omp.h)
expand_vars_recipe(${src_dir}/include/${LIBOMP_OMP_VERSION} ompt.h)
expand_vars_recipe(${src_dir}/include/${LIBOMP_OMP_VERSION} omp_lib.h)
expand_vars_recipe(${src_dir}/include/${LIBOMP_OMP_VERSION} omp_lib.f)
expand_vars_recipe(${src_dir}/include/${LIBOMP_OMP_VERSION} omp_lib.f90)
expand_vars_recipe(${src_dir}/include/${LIBOMP_OMP_VERSION} iomp.h)
expand_vars_recipe(${src_dir}/include/${LIBOMP_OMP_VERSION} iomp_lib.h)
expand_vars_recipe(${src_dir} libiomp.rc)
####################################################################
# Print configuration after all variables are set.
if(${LIBOMP_STANDALONE_BUILD})
say("LIBOMP: Operating System -- ${LIBOMP_OS}")
say("LIBOMP: Target Architecture -- ${LIBOMP_ARCH}")
if(${MIC})
say("LIBOMP: Intel(R) MIC Architecture -- ${LIBOMP_MIC_ARCH}")
endif()
say("LIBOMP: Build Type -- ${CMAKE_BUILD_TYPE}")
say("LIBOMP: OpenMP Version -- ${LIBOMP_OMP_VERSION}")
say("LIBOMP: Lib Type -- ${LIBOMP_LIB_TYPE}")
say("LIBOMP: Fortran Modules -- ${LIBOMP_FORTRAN_MODULES}")
# will say development if all zeros
if("${build_number}" STREQUAL "00000000")
set(build "development")
else()
set(build "${build_number}")
endif()
say("LIBOMP: Build -- ${build}")
say("LIBOMP: Stats-Gathering -- ${LIBOMP_STATS}")
say("LIBOMP: OMPT-support -- ${LIBOMP_OMPT_SUPPORT}")
if(${LIBOMP_OMPT_SUPPORT})
say("LIBOMP: OMPT-blame -- ${LIBOMP_OMPT_BLAME}")
say("LIBOMP: OMPT-trace -- ${LIBOMP_OMPT_TRACE}")
endif()
say("LIBOMP: Use build.pl rules -- ${LIBOMP_USE_BUILDPL_RULES}")
say("LIBOMP: Adaptive locks -- ${LIBOMP_USE_ADAPTIVE_LOCKS}")
say("LIBOMP: Use predefined linker flags -- ${LIBOMP_USE_PREDEFINED_LINKER_FLAGS}")
say("LIBOMP: Compiler supports quad precision -- ${LIBOMP_COMPILER_SUPPORTS_QUAD_PRECISION}")
endif()
####################################################################
# Install rules
# We want to install libiomp5 in DESTDIR/CMAKE_INSTALL_PREFIX/lib
# We want to install headers in DESTDIR/CMAKE_INSTALL_PREFIX/include
if(${LIBOMP_STANDALONE_BUILD})
set(LIBOMP_HEADERS_INSTALL_PATH include)
else()
string(REGEX MATCH "[0-9]+\\.[0-9]+(\\.[0-9]+)?" CLANG_VERSION ${PACKAGE_VERSION})
set(LIBOMP_HEADERS_INSTALL_PATH lib${LIBOMP_LIBDIR_SUFFIX}/clang/${CLANG_VERSION}/include)
endif()
if(${WINDOWS})
install(TARGETS iomp5 RUNTIME DESTINATION bin)
if(NOT "${imp_file}" STREQUAL "")
install(TARGETS iomp5imp ARCHIVE DESTINATION lib${LIBOMP_LIBDIR_SUFFIX})
endif()
else()
install(TARGETS iomp5 LIBRARY DESTINATION lib${LIBOMP_LIBDIR_SUFFIX})
endif()
install(
FILES
${build_dir}/omp.h
${build_dir}/iomp.h
DESTINATION ${LIBOMP_HEADERS_INSTALL_PATH}
)
if(${LIBOMP_OMPT_SUPPORT})
install(FILES ${build_dir}/ompt.h DESTINATION ${LIBOMP_HEADERS_INSTALL_PATH})
endif()