[Openmp-commits] [PATCH] D136641: [OPENMP] Fast cross-team reduction (xteamr) helper functions.

[Greg Rodgers via Phabricator via Openmp-commits]

gregrodgers updated this revision to Diff 471196.
gregrodgers added a comment.
Herald added a reviewer: sscalpone.

4th update to original D136631 <https://reviews.llvm.org/D136631> review. Changes:

- Remove unnecessary thread sync following LDS reduction.
- Replaced the bool LDS __is_last_team with LDS copy of the team counter value returned from atomic inc.  This removes a few more scalar registers. Performance and resource utilization noticably improved from all updates to this review.
- Change parameter name inival to "rnv" (Reduction Null Value) for consistency.
- Improved header docs and comment block for main function _xteamr_reduction.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D136641/new/

https://reviews.llvm.org/D136641

Files:
  openmp/libomptarget/DeviceRTL/CMakeLists.txt
  openmp/libomptarget/DeviceRTL/include/Interface.h
  openmp/libomptarget/DeviceRTL/include/Xteamr.h
  openmp/libomptarget/DeviceRTL/src/Xteamr.cpp
  openmp/libomptarget/test/lit.cfg
  openmp/libomptarget/test/xteamr/test_xteamr.cpp
  openmp/libomptarget/test/xteamr/test_xteamr.h
  openmp/libomptarget/test/xteamr/test_xteamr.sh

-------------- next part --------------
A non-text attachment was scrubbed...
Name: D136641.471196.patch
Type: text/x-patch
Size: 189893 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/openmp-commits/attachments/20221027/89502d59/attachment-0001.bin>