[all-commits] [llvm/llvm-project] e87625: [OpenMP][libomptarget] Enable parallel copies via ...

[Jan Patrick Lehr via All-commits]

  Branch: refs/heads/main
  Home:   https://github.com/llvm/llvm-project
  Commit: e876250b636522d1eb05a908f2e1cd451feab001
      https://github.com/llvm/llvm-project/commit/e876250b636522d1eb05a908f2e1cd451feab001
  Author: Jan Patrick Lehr <jplehr at users.noreply.github.com>
  Date:   2023-11-14 (Tue, 14 Nov 2023)

  Changed paths:
    M openmp/libomptarget/plugins-nextgen/amdgpu/src/rtl.cpp

  Log Message:
  -----------
  [OpenMP][libomptarget] Enable parallel copies via multiple SDMA engines (#71801)

This enables the AMDGPU plugin to use a new ROCm 5.7 interface to
dispatch asynchronous data transfers across SDMA engines.

The default functionality stays unchanged, meaning that all data
transfers are enqueued into a H2D queue or an D2H queue, depending on
transfer direction, via the HSA interface used previously.

The new interface can be enabled via the environment variable
`LIBOMPTARGET_AMDGPU_USE_MULTIPLE_SDMA_ENGINES=true` when libomptarget
is built against a recent ROCm version (5.7 and later).
As of now, requests are distributed in a round-robin fashion across
available SDMA engines.