[Openmp-commits] [PATCH] D95584: [OpenMP][NFC] Added release note for new `deviceRTLs` and hidden helper task

[Shilei Tian via Phabricator via Openmp-commits]

tianshilei1992 created this revision.
tianshilei1992 added a reviewer: jdoerfert.
Herald added subscribers: guansong, yaxunl.
tianshilei1992 requested review of this revision.
Herald added subscribers: openmp-commits, sstefan1.
Herald added a project: OpenMP.

Added release note for new `deviceRTLs` and hidden helper task for LLVM
12.


Repository:
  rG LLVM Github Monorepo

https://reviews.llvm.org/D95584

Files:
  openmp/docs/ReleaseNotes.rst


Index: openmp/docs/ReleaseNotes.rst
===================================================================
--- openmp/docs/ReleaseNotes.rst
+++ openmp/docs/ReleaseNotes.rst
@@ -7,7 +7,7 @@
    These are in-progress notes for the upcoming LLVM 12.0.0 release.
    Release notes for previous releases can be found on
    `the Download Page <https://releases.llvm.org/download.html>`_.
-   
+
 
 Introduction
 ============
@@ -44,3 +44,23 @@
   ``LIBOMPTARGET_INFO`` allows the user to request certain information from the
   ``libomptarget`` runtime using a 32-bit field. A full description of each
   environment variable is described :ref:`here <libopenmptarget_environment_vars>`.
+
+- ``target nowait`` was supported via hidden helper task, which is a task not
+  bound to any parallel region. A hidden helper team with a number of threads is
+  created when the first hidden helper task is encountered. The number of threads
+  can be configured via the environment variable
+  ``LIBOMP_NUM_HIDDEN_HELPER_THREADS``. By default it is 8. If
+  ``LIBOMP_NUM_HIDDEN_HELPER_THREADS=0``, hidden helper task is disabled and
+  falls back to a regular OpenMP task. It can also be disabled by setting the
+  environment variable ``LIBOMP_USE_HIDDEN_HELPER_TASK=OFF``.
+
+- ``deviceRTLs`` for NVPTX platform is CUDA free now. It is generally OpenMP code.
+  Target dependent parts are implemented Clang/LLVM/NVVM intrinsics. CUDA SDK is
+  also dropped as a dependence to build the device runtime, which means device
+  runtime can also be built on a non-CUDA system. However, it is disabled by
+  default. Set the CMake variable ``LIBOMPTARGET_BUILD_NVPTX_BCLIB=ON`` to enable
+  the build of NVPTX device runtime on a non-CUDA system.
+  ``gcc-multilib`` and ``g++-multilib`` are required. Meanwhile, static NVPTX
+  device runtime library was also dropped. Bitcode library is required to build
+  an OpenMP program. If the library is not found in ``LIBRARY_PATH``, an error
+  will be raised. User can also specify it via ``--libomptarget-nvptx-bc-path=``.


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D95584.319740.patch
Type: text/x-patch
Size: 2050 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/openmp-commits/attachments/20210128/4d886f2b/attachment.bin>