[all-commits] [llvm/llvm-project] 7634c6: [OpenMP][AMDGPU] Use DS_Max_Warp_Number instead of...

pdhaliwal-amd via All-commits all-commits at lists.llvm.org
Mon Sep 7 02:15:49 PDT 2020


  Branch: refs/heads/master
  Home:   https://github.com/llvm/llvm-project
  Commit: 7634c64b6121ba61a6c72c6b45e3561ad8cf345e
      https://github.com/llvm/llvm-project/commit/7634c64b6121ba61a6c72c6b45e3561ad8cf345e
  Author: Pushpinder Singh <Pushpinder.Singh at amd.com>
  Date:   2020-09-07 (Mon, 07 Sep 2020)

  Changed paths:
    M openmp/libomptarget/deviceRTLs/common/omptarget.h
    M openmp/libomptarget/deviceRTLs/common/src/data_sharing.cu

  Log Message:
  -----------
  [OpenMP][AMDGPU] Use DS_Max_Warp_Number instead of WARPSIZE

The size of worker_rootS should have been DS_Max_Warp_Number.
This reduces memory usage by deviceRTL on AMDGPU from around 2.3GB
to around 770MB.

Reviewed By: JonChesterfield, jdoerfert

Differential Revision: https://reviews.llvm.org/D87084




More information about the All-commits mailing list