Skip to content

Commit 322e3dc

Browse files
kzhuravldsalinas_amdeng
kzhuravl
authored and
dsalinas_amdeng
committed
Adjust device-libs search (llvm#4517)
1 parent e28322a commit 322e3dc

File tree

1 file changed

+232
-0
lines changed

1 file changed

+232
-0
lines changed

‎offload/DeviceRTL/CMakeLists.txt‎

Lines changed: 232 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,232 @@
1+
set(LIBOMPTARGET_BUILD_DEVICERTL_BCLIB TRUE CACHE BOOL
2+
"Can be set to false to disable building this library.")
3+
4+
if (NOT LIBOMPTARGET_BUILD_DEVICERTL_BCLIB)
5+
message(STATUS "Not building DeviceRTL: Disabled by LIBOMPTARGET_BUILD_DEVICERTL_BCLIB")
6+
return()
7+
endif()
8+
9+
# Check to ensure the host system is a supported host architecture.
10+
if(NOT ${CMAKE_SIZEOF_VOID_P} EQUAL "8")
11+
message(STATUS "Not building DeviceRTL: Runtime does not support 32-bit hosts")
12+
return()
13+
endif()
14+
15+
if (LLVM_DIR)
16+
# Builds that use pre-installed LLVM have LLVM_DIR set.
17+
# A standalone or LLVM_ENABLE_RUNTIMES=openmp build takes this route
18+
find_program(CLANG_TOOL clang PATHS ${LLVM_TOOLS_BINARY_DIR} NO_DEFAULT_PATH)
19+
elseif (LLVM_TOOL_CLANG_BUILD AND NOT CMAKE_CROSSCOMPILING AND NOT OPENMP_STANDALONE_BUILD)
20+
# LLVM in-tree builds may use CMake target names to discover the tools.
21+
# A LLVM_ENABLE_PROJECTS=openmp build takes this route
22+
set(CLANG_TOOL $<TARGET_FILE:clang>)
23+
else()
24+
message(STATUS "Not building DeviceRTL. No appropriate clang found")
25+
return()
26+
endif()
27+
28+
set(devicertl_base_directory ${CMAKE_CURRENT_SOURCE_DIR})
29+
set(include_directory ${devicertl_base_directory}/include)
30+
set(source_directory ${devicertl_base_directory}/src)
31+
32+
if(OFFLOAD_ENABLE_EMISSARY_APIS)
33+
set(emissary_includes ${include_directory}/EmissaryIds.h)
34+
set(emissary_sources src/EmissaryFortrt.cpp src/EmissaryPrint.cpp)
35+
endif()
36+
37+
set(include_files
38+
${include_directory}/Allocator.h
39+
${include_directory}/Configuration.h
40+
${include_directory}/Platform.h
41+
${include_directory}/Debug.h
42+
${include_directory}/Interface.h
43+
${include_directory}/LibC.h
44+
${include_directory}/Mapping.h
45+
${include_directory}/Profiling.h
46+
${include_directory}/State.h
47+
${include_directory}/Synchronization.h
48+
${include_directory}/DeviceTypes.h
49+
${include_directory}/DeviceUtils.h
50+
${include_directory}/Xteamr.h
51+
${include_directory}/Xteams.h
52+
${include_directory}/Workshare.h
53+
${emissary_includes}
54+
)
55+
56+
set(src_files
57+
${source_directory}/Allocator.cpp
58+
${source_directory}/Configuration.cpp
59+
${source_directory}/Debug.cpp
60+
${source_directory}/Kernel.cpp
61+
${source_directory}/LibC.cpp
62+
${source_directory}/LibM.cpp
63+
${source_directory}/Mapping.cpp
64+
${source_directory}/Misc.cpp
65+
${source_directory}/Parallelism.cpp
66+
${source_directory}/Profiling.cpp
67+
${source_directory}/Reduction.cpp
68+
${source_directory}/State.cpp
69+
${source_directory}/Synchronization.cpp
70+
${source_directory}/Tasking.cpp
71+
${source_directory}/DeviceUtils.cpp
72+
${source_directory}/Workshare.cpp
73+
${source_directory}/ExtraMapping.cpp
74+
${source_directory}/Xteamr.cpp
75+
${source_directory}/Memory.cpp
76+
${source_directory}/Xteams.cpp
77+
${emissary_sources}
78+
)
79+
80+
# We disable the slp vectorizer during the runtime optimization to avoid
81+
# vectorized accesses to the shared state. Generally, those are "good" but
82+
# the optimizer pipeline (esp. Attributor) does not fully support vectorized
83+
# instructions yet and we end up missing out on way more important constant
84+
# propagation. That said, we will run the vectorizer again after the runtime
85+
# has been linked into the user program.
86+
set(clang_opt_flags -O3 -mllvm -openmp-opt-disable -DSHARED_SCRATCHPAD_SIZE=512 -mllvm -vectorize-slp=false )
87+
88+
# If the user built with the GPU C library enabled we will use that instead.
89+
if(${LIBOMPTARGET_GPU_LIBC_SUPPORT})
90+
list(APPEND clang_opt_flags -DOMPTARGET_HAS_LIBC)
91+
endif()
92+
93+
# Set flags for LLVM Bitcode compilation.
94+
set(bc_flags -c -flto -std=c++17 -fvisibility=hidden
95+
${clang_opt_flags} -nogpulib -nostdlibinc
96+
-fno-rtti -fno-exceptions -fconvergent-functions
97+
-Wno-unknown-cuda-version
98+
-DOMPTARGET_DEVICE_RUNTIME
99+
-I${include_directory}
100+
-I${devicertl_base_directory}/../include
101+
-I${devicertl_base_directory}/../../libc
102+
)
103+
104+
if(${LIBOMPTARGET_DEVICE_DEBUG})
105+
list(APPEND bc_flags -DOMPTARGET_DEBUG=-1)
106+
else()
107+
list(APPEND bc_flags -DOMPTARGET_DEBUG=0)
108+
endif()
109+
110+
# first create an object target
111+
add_library(omptarget.devicertl.all_objs OBJECT IMPORTED)
112+
function(compileDeviceRTLLibrary target_name target_triple)
113+
set(target_bc_flags ${ARGN})
114+
115+
if(${target_name} MATCHES "amdgpu")
116+
find_package(AMDDeviceLibs REQUIRED CONFIG
117+
HINTS ${CMAKE_BINARY_DIR}/../../tools/rocm-device-libs
118+
${CMAKE_BINARY_DIR}/../rocm-device-libs-prefix/src/rocm-device-libs-build
119+
${CMAKE_INSTALL_PREFIX}
120+
)
121+
get_target_property(_ocml_bc ocml IMPORTED_LOCATION)
122+
get_target_property(_ockl_bc ockl IMPORTED_LOCATION)
123+
if(NOT _ockl_bc)
124+
message(FATAL_ERROR "Could not find ockl.bc")
125+
endif()
126+
if(NOT _ocml_bc)
127+
message(FATAL_ERROR "Could not find ocml.bc")
128+
endif()
129+
list(APPEND target_bc_flags -Xclang -mlink-builtin-bitcode -Xclang ${_ockl_bc})
130+
list(APPEND target_bc_flags -Xclang -mlink-builtin-bitcode -Xclang ${_ocml_bc})
131+
endif()
132+
133+
foreach(src ${src_files})
134+
get_filename_component(infile ${src} ABSOLUTE)
135+
get_filename_component(outfile ${src} NAME)
136+
set(outfile "${outfile}-${target_name}.o")
137+
set(depfile "${outfile}.d")
138+
139+
# Passing an empty CPU to -march= suppressed target specific metadata.
140+
add_custom_command(OUTPUT ${outfile}
141+
COMMAND ${CLANG_TOOL}
142+
${bc_flags}
143+
--target=${target_triple}
144+
${target_bc_flags}
145+
-MD -MF ${depfile}
146+
${infile} -o ${outfile}
147+
DEPENDS ${infile}
148+
DEPFILE ${depfile}
149+
COMMENT "Building LLVM bitcode ${outfile}"
150+
VERBATIM
151+
)
152+
if(TARGET clang)
153+
# Add a file-level dependency to ensure that clang is up-to-date.
154+
# By default, add_custom_command only builds clang if the
155+
# executable is missing.
156+
add_custom_command(OUTPUT ${outfile}
157+
DEPENDS clang
158+
APPEND
159+
)
160+
endif()
161+
set_property(DIRECTORY APPEND PROPERTY ADDITIONAL_MAKE_CLEAN_FILES ${outfile})
162+
163+
list(APPEND obj_files ${CMAKE_CURRENT_BINARY_DIR}/${outfile})
164+
endforeach()
165+
# Trick to combine these into a bitcode file via the linker's LTO pass. This
166+
# is used to provide the legacy `libomptarget-<name>.bc` files. Hack this
167+
# through as an executable to get it to use the relocatable link.
168+
add_executable(libomptarget-${target_name})
169+
target_sources(libomptarget-${target_name} PRIVATE ${obj_files})
170+
set_target_properties(libomptarget-${target_name} PROPERTIES
171+
RUNTIME_OUTPUT_DIRECTORY ${LIBOMPTARGET_LLVM_LIBRARY_INTDIR}
172+
LINKER_LANGUAGE CXX
173+
BUILD_RPATH ""
174+
INSTALL_RPATH ""
175+
RUNTIME_OUTPUT_NAME libomptarget-${target_name}.bc)
176+
target_compile_options(libomptarget-${target_name} PRIVATE
177+
"--target=${target_triple}" "-fuse-ld=lld" "-march=" "-mcpu="
178+
"-Wno-unused-command-line-argument")
179+
target_link_options(libomptarget-${target_name} PRIVATE
180+
"--target=${target_triple}" "-r" "-nostdlib" "-flto" "-Wl,--lto-emit-llvm"
181+
"-Wl,--lto-newpm-passes=default<O3>" "-Wl,-plugin-opt=-openmp-opt-disable"
182+
"-Wl,-plugin-opt=-attributor-enable=module"
183+
"-Wl,-plugin-opt=-vectorize-slp=false" "-fuse-ld=lld" "-march=" "-mcpu=")
184+
install(TARGETS libomptarget-${target_name}
185+
PERMISSIONS OWNER_WRITE OWNER_READ GROUP_READ WORLD_READ
186+
DESTINATION "lib${LLVM_LIBDIR_SUFFIX}/${target_triple}")
187+
188+
add_library(omptarget.${target_name}.all_objs OBJECT IMPORTED)
189+
set_property(TARGET omptarget.${target_name}.all_objs APPEND PROPERTY IMPORTED_OBJECTS
190+
${LIBOMPTARGET_LLVM_LIBRARY_INTDIR}/libomptarget-${target_name}.bc)
191+
add_dependencies(omptarget.${target_name}.all_objs libomptarget-${target_name})
192+
193+
# Archive all the object files generated above into a static library
194+
add_library(omptarget.${target_name} STATIC)
195+
set_target_properties(omptarget.${target_name} PROPERTIES
196+
ARCHIVE_OUTPUT_DIRECTORY "${LIBOMPTARGET_LLVM_LIBRARY_INTDIR}/${target_triple}"
197+
ARCHIVE_OUTPUT_NAME ompdevice
198+
LINKER_LANGUAGE CXX
199+
)
200+
add_dependencies(omptarget.${target_name} libomptarget-${target_name})
201+
target_link_libraries(omptarget.${target_name} PRIVATE omptarget.${target_name}.all_objs)
202+
target_link_options(omptarget.${target_name} PRIVATE "--target=${target_triple}"
203+
"-Wno-unused-command-line-argument" "-r" "-nostdlib" "-flto"
204+
"-Wl,--lto-emit-llvm" "-fuse-ld=lld" "-march=" "-mcpu=")
205+
206+
install(TARGETS omptarget.${target_name}
207+
ARCHIVE DESTINATION "lib${LLVM_LIBDIR_SUFFIX}/${target_triple}")
208+
209+
if (CMAKE_EXPORT_COMPILE_COMMANDS)
210+
set(ide_target_name omptarget-ide-${target_name})
211+
add_library(${ide_target_name} STATIC EXCLUDE_FROM_ALL ${src_files})
212+
target_compile_options(${ide_target_name} PRIVATE
213+
-fvisibility=hidden --target=${target_triple}
214+
-nogpulib -nostdlibinc -Wno-unknown-cuda-version
215+
)
216+
target_compile_definitions(${ide_target_name} PRIVATE SHARED_SCRATCHPAD_SIZE=512)
217+
target_include_directories(${ide_target_name} PRIVATE
218+
${include_directory}
219+
${devicertl_base_directory}/../../libc
220+
${devicertl_base_directory}/../include
221+
)
222+
install(TARGETS ${ide_target_name} EXCLUDE_FROM_ALL)
223+
endif()
224+
endfunction()
225+
226+
if(NOT LLVM_TARGETS_TO_BUILD OR "AMDGPU" IN_LIST LLVM_TARGETS_TO_BUILD)
227+
compileDeviceRTLLibrary(amdgpu amdgcn-amd-amdhsa -Xclang -mcode-object-version=none)
228+
endif()
229+
230+
if(NOT LLVM_TARGETS_TO_BUILD OR "NVPTX" IN_LIST LLVM_TARGETS_TO_BUILD)
231+
compileDeviceRTLLibrary(nvptx nvptx64-nvidia-cuda --cuda-feature=+ptx63)
232+
endif()

0 commit comments

Comments
 (0)