[all-commits] [llvm/llvm-project] 840d0b: [amdgpu] D2D memcpy via streams and HSA (#69977)

[Jon Chesterfield via All-commits]

  Branch: refs/heads/main
  Home:   https://github.com/llvm/llvm-project
  Commit: 840d0b7e03045c30b052dce8d0bb82f3fa0c5d6d
      https://github.com/llvm/llvm-project/commit/840d0b7e03045c30b052dce8d0bb82f3fa0c5d6d
  Author: Jon Chesterfield <JonChesterfield at users.noreply.github.com>
  Date:   2023-10-24 (Tue, 24 Oct 2023)

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

  Log Message:
  -----------
  [amdgpu] D2D memcpy via streams and HSA (#69977)

hsa_amd_memory_async_copy can handle device to device copies if passed
the corresponding parameters.

No functional change - currently D2D copy goes through a fallback in
libomptarget that stages through a host malloc, after this it goes
directly through HSA.

Works under exactly the situations that HSA works. Verified locally on a
performance benchmark. Hoping to attract further testing from internal
developers after it lands.