1 Support, Getting Involved, and FAQ
2 ==================================
4 Please do not hesitate to reach out to us via openmp-dev@lists.llvm.org or join
5 one of our :ref:`regular calls <calls>`. Some common questions are answered in
13 OpenMP in LLVM Technical Call
14 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
16 - Development updates on OpenMP (and OpenACC) in the LLVM Project, including Clang, optimization, and runtime work.
17 - Join `OpenMP in LLVM Technical Call <https://bluejeans.com/544112769//webrtc>`__.
18 - Time: Weekly call on every Wednesday 7:00 AM Pacific time.
19 - Meeting minutes are `here <https://docs.google.com/document/d/1Tz8WFN13n7yJ-SCE0Qjqf9LmjGUw0dWO9Ts1ss4YOdg/edit>`__.
20 - Status tracking `page <https://openmp.llvm.org/docs>`__.
23 OpenMP in Flang Technical Call
24 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
25 - Development updates on OpenMP and OpenACC in the Flang Project.
26 - Join `OpenMP in Flang Technical Call <https://bit.ly/39eQW3o>`_
27 - Time: Weekly call on every Thursdays 8:00 AM Pacific time.
28 - Meeting minutes are `here <https://docs.google.com/document/d/1yA-MeJf6RYY-ZXpdol0t7YoDoqtwAyBhFLr5thu5pFI>`__.
29 - Status tracking `page <https://docs.google.com/spreadsheets/d/1FvHPuSkGbl4mQZRAwCIndvQx9dQboffiD-xD0oqxgU0/edit#gid=0>`__.
38 The FAQ is a work in progress and most of the expected content is not
39 yet available. While you can expect changes, we always welcome feedback and
40 additions. Please contact, e.g., through ``openmp-dev@lists.llvm.org``.
43 Q: How to contribute a patch to the webpage or any other part?
44 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
46 All patches go through the regular `LLVM review process
47 <https://llvm.org/docs/Contributing.html#how-to-submit-a-patch>`_.
50 .. _build_offload_capable_compiler:
52 Q: How to build an OpenMP GPU offload capable compiler?
53 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
54 To build an *effective* OpenMP offload capable compiler, only one extra CMake
55 option, `LLVM_ENABLE_RUNTIMES="openmp"`, is needed when building LLVM (Generic
56 information about building LLVM is available `here
57 <https://llvm.org/docs/GettingStarted.html>`__.). Make sure all backends that
58 are targeted by OpenMP to be enabled. By default, Clang will be built with all
59 backends enabled. When building with `LLVM_ENABLE_RUNTIMES="openmp"` OpenMP
60 should not be enabled in `LLVM_ENABLE_PROJECTS` because it is enabled by
63 For Nvidia offload, please see :ref:`build_nvidia_offload_capable_compiler`.
64 For AMDGPU offload, please see :ref:`build_amdgpu_offload_capable_compiler`.
67 The compiler that generates the offload code should be the same (version) as
68 the compiler that builds the OpenMP device runtimes. The OpenMP host runtime
69 can be built by a different compiler.
71 .. _advanced_builds: https://llvm.org//docs/AdvancedBuilds.html
73 .. _build_nvidia_offload_capable_compiler:
75 Q: How to build an OpenMP NVidia offload capable compiler?
76 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
77 The Cuda SDK is required on the machine that will execute the openmp application.
79 If your build machine is not the target machine or automatic detection of the
80 available GPUs failed, you should also set:
82 - `CLANG_OPENMP_NVPTX_DEFAULT_ARCH=sm_XX` where `XX` is the architecture of your GPU, e.g, 80.
83 - `LIBOMPTARGET_NVPTX_COMPUTE_CAPABILITIES=YY` where `YY` is the numeric compute capacity of your GPU, e.g., 75.
86 .. _build_amdgpu_offload_capable_compiler:
88 Q: How to build an OpenMP AMDGPU offload capable compiler?
89 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
90 A subset of the `ROCm <https://github.com/radeonopencompute>`_ toolchain is
91 required to build the LLVM toolchain and to execute the openmp application.
92 Either install ROCm somewhere that cmake's find_package can locate it, or
93 build the required subcomponents ROCt and ROCr from source.
95 The two components used are ROCT-Thunk-Interface, roct, and ROCR-Runtime, rocr.
96 Roct is the userspace part of the linux driver. It calls into the driver which
97 ships with the linux kernel. It is an implementation detail of Rocr from
98 OpenMP's perspective. Rocr is an implementation of `HSA
99 <http://www.hsafoundation.com>`_.
103 SOURCE_DIR=same-as-llvm-source # e.g. the checkout of llvm-project, next to openmp
105 INSTALL_PREFIX=same-as-llvm-install
108 git clone git@github.com:RadeonOpenCompute/ROCT-Thunk-Interface.git -b roc-4.2.x \
110 git clone git@github.com:RadeonOpenCompute/ROCR-Runtime.git -b rocm-4.2.x \
113 cd $BUILD_DIR && mkdir roct && cd roct
114 cmake $SOURCE_DIR/ROCT-Thunk-Interface/ -DCMAKE_INSTALL_PREFIX=$INSTALL_PREFIX \
115 -DCMAKE_BUILD_TYPE=Release -DBUILD_SHARED_LIBS=OFF
118 cd $BUILD_DIR && mkdir rocr && cd rocr
119 cmake $SOURCE_DIR/ROCR-Runtime/src -DIMAGE_SUPPORT=OFF \
120 -DCMAKE_INSTALL_PREFIX=$INSTALL_PREFIX -DCMAKE_BUILD_TYPE=Release \
121 -DBUILD_SHARED_LIBS=ON
124 ``IMAGE_SUPPORT`` requires building rocr with clang and is not used by openmp.
126 Provided cmake's find_package can find the ROCR-Runtime package, LLVM will
127 build a tool ``bin/amdgpu-arch`` which will print a string like ``gfx906`` when
128 run if it recognises a GPU on the local system. LLVM will also build a shared
129 library, libomptarget.rtl.amdgpu.so, which is linked against rocr.
131 With those libraries installed, then LLVM build and installed, try:
133 .. code-block:: shell
135 clang -O2 -fopenmp -fopenmp-targets=amdgcn-amd-amdhsa example.c -o example && ./example
137 Q: What are the known limitations of OpenMP AMDGPU offload?
138 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
139 LD_LIBRARY_PATH or rpath/runpath are required to find libomp.so and libomptarget.so
141 There is no libc. That is, malloc and printf do not exist. Libm is implemented in terms
142 of the rocm device library, which will be searched for if linking with '-lm'.
144 Some versions of the driver for the radeon vii (gfx906) will error unless the
145 environment variable 'export HSA_IGNORE_SRAMECC_MISREPORT=1' is set.
147 It is a recent addition to LLVM and the implementation differs from that which
148 has been shipping in ROCm and AOMP for some time. Early adopters will encounter
151 Q: What are the LLVM components used in offloading and how are they found?
152 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
153 The libraries used by an executable compiled for target offloading are:
154 - ``libomp.so`` (or similar), the host openmp runtime
155 - ``libomptarget.so``, the target-agnostic target offloading openmp runtime
156 - plugins loaded by libomptarget.so:
157 - ``libomptarget.rtl.amdgpu.so``
158 - ``libomptarget.rtl.cuda.so``
159 - ``libomptarget.rtl.x86_64.so``
160 - ``libomptarget.rtl.ve.so``
162 - dependencies of those plugins, e.g. cuda/rocr for nvptx/amdgpu
164 The compiled executable is dynamically linked against a host runtime, e.g.
165 ``libomp.so``, and against the target offloading runtime, ``libomptarget.so``. These
166 are found like any other dynamic library, by setting rpath or runpath on the
167 executable, by setting ``LD_LIBRARY_PATH``, or by adding them to the system search.
169 ``libomptarget.so`` has rpath or runpath (whichever the system default is) set to
170 ``$ORIGIN``, and the plugins are located next to it, so it will find the plugins
171 without any environment variables set. If ``LD_LIBRARY_PATH`` is set, whether it
172 overrides which plugin is found depends on whether your system treats ``-Wl,-rpath``
175 The plugins will try to find their dependencies in plugin-dependent fashion.
177 The cuda plugin is dynamically linked against libcuda if cmake found it at
178 compiler build time. Otherwise it will attempt to dlopen ``libcuda.so``. It does
181 The amdgpu plugin is linked against ROCr if cmake found it at compiler build
182 time. Otherwise it will attempt to dlopen ``libhsa-runtime64.so``. It has rpath
183 set to ``$ORIGIN``, so installing ``libhsa-runtime64.so`` in the same directory is a
184 way to locate it without environment variables.
186 In addition to those, there is a compiler runtime library called deviceRTL.
187 This is compiled from mostly common code into an architecture specific
188 bitcode library, e.g. ``libomptarget-nvptx-sm_70.bc``.
190 Clang and the deviceRTL need to match closely as the interface between them
191 changes frequently. Using both from the same monorepo checkout is strongly
194 Unlike the host side which lets environment variables select components, the
195 deviceRTL that is located in the clang lib directory is preferred. Only if
196 it is absent, the ``LIBRARY_PATH`` environment variable is searched to find a
197 bitcode file with the right name. This can be overridden by passing a clang
198 flag, ``--libomptarget-nvptx-bc-path`` or ``--libomptarget-amdgcn-bc-path``. That
199 can specify a directory or an exact bitcode file to use.
202 Q: Does OpenMP offloading support work in pre-packaged LLVM releases?
203 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
204 For now, the answer is most likely *no*. Please see :ref:`build_offload_capable_compiler`.
206 Q: Does OpenMP offloading support work in packages distributed as part of my OS?
207 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
208 For now, the answer is most likely *no*. Please see :ref:`build_offload_capable_compiler`.
211 .. _math_and_complex_in_target_regions:
213 Q: Does Clang support `<math.h>` and `<complex.h>` operations in OpenMP target on GPUs?
214 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
216 Yes, LLVM/Clang allows math functions and complex arithmetic inside of OpenMP
217 target regions that are compiled for GPUs.
219 Clang provides a set of wrapper headers that are found first when `math.h` and
220 `complex.h`, for C, `cmath` and `complex`, for C++, or similar headers are
221 included by the application. These wrappers will eventually include the system
222 version of the corresponding header file after setting up a target device
223 specific environment. The fact that the system header is included is important
224 because they differ based on the architecture and operating system and may
225 contain preprocessor, variable, and function definitions that need to be
226 available in the target region regardless of the targeted device architecture.
227 However, various functions may require specialized device versions, e.g.,
228 `sin`, and others are only available on certain devices, e.g., `__umul64hi`. To
229 provide "native" support for math and complex on the respective architecture,
230 Clang will wrap the "native" math functions, e.g., as provided by the device
231 vendor, in an OpenMP begin/end declare variant. These functions will then be
232 picked up instead of the host versions while host only variables and function
233 definitions are still available. Complex arithmetic and functions are support
234 through a similar mechanism. It is worth noting that this support requires
235 `extensions to the OpenMP begin/end declare variant context selector
236 <https://clang.llvm.org/docs/AttributeReference.html#pragma-omp-declare-variant>`__
237 that are exposed through LLVM/Clang to the user as well.
239 Q: What is a way to debug errors from mapping memory to a target device?
240 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
242 An experimental way to debug these errors is to use :ref:`remote process
243 offloading <remote_offloading_plugin>`.
244 By using ``libomptarget.rtl.rpc.so`` and ``openmp-offloading-server``, it is
245 possible to explicitly perform memory transfers between processes on the host
246 CPU and run sanitizers while doing so in order to catch these errors.
248 Q: Why does my application say "Named symbol not found" and abort when I run it?
249 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
251 This is most likely caused by trying to use OpenMP offloading with static
252 libraries. Static libraries do not contain any device code, so when the runtime
253 attempts to execute the target region it will not be found and you will get an
258 CUDA error: Loading '__omp_offloading_fd02_3231c15__Z3foov_l2' Failed
259 CUDA error: named symbol not found
260 Libomptarget error: Unable to generate entries table for device id 0.
262 Currently, the only solution is to change how the application is built and avoid
263 the use of static libraries.
265 Q: Can I use dynamically linked libraries with OpenMP offloading?
266 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
268 Dynamically linked libraries can be only used if there is no device code split
269 between the library and application. Anything declared on the device inside the
270 shared library will not be visible to the application when it's linked.
272 Q: How to build an OpenMP offload capable compiler with an outdated host compiler?
273 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
275 Enabling the OpenMP runtime will perform a two-stage build for you.
276 If your host compiler is different from your system-wide compiler, you may need
277 to set the CMake variable `GCC_INSTALL_PREFIX` so clang will be able to find the
278 correct GCC toolchain in the second stage of the build.
280 For example, if your system-wide GCC installation is too old to build LLVM and
281 you would like to use a newer GCC, set the CMake variable `GCC_INSTALL_PREFIX`
282 to inform clang of the GCC installation you would like to use in the second stage.
284 Q: How can I include OpenMP offloading support in my CMake project?
285 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
287 Currently, there is an experimental CMake find module for OpenMP target
288 offloading provided by LLVM. It will attempt to find OpenMP target offloading
289 support for your compiler. The flags necessary for OpenMP target offloading will
290 be loaded into the ``OpenMPTarget::OpenMPTarget_<device>`` target or the
291 ``OpenMPTarget_<device>_FLAGS`` variable if successful. Currently supported
292 devices are ``AMDGPU`` and ``NVPTX``.
294 To use this module, simply add the path to CMake's current module path and call
295 ``find_package``. The module will be installed with your OpenMP installation by
296 default. Including OpenMP offloading support in an application should now only
297 require a few additions.
299 .. code-block:: cmake
301 cmake_minimum_required(VERSION 3.13.4)
302 project(offloadTest VERSION 1.0 LANGUAGES CXX)
304 list(APPEND CMAKE_MODULE_PATH "${PATH_TO_OPENMP_INSTALL}/lib/cmake/openmp")
306 find_package(OpenMPTarget REQUIRED NVPTX)
308 add_executable(offload)
309 target_link_libraries(offload PRIVATE OpenMPTarget::OpenMPTarget_NVPTX)
310 target_sources(offload PRIVATE ${CMAKE_CURRENT_SOURCE_DIR}/src/Main.cpp)
312 Using this module requires at least CMake version 3.13.4. Supported languages
313 are C and C++ with Fortran support planned in the future. Compiler support is
314 best for Clang but this module should work for other compiler vendors such as