[Openmp-commits] [PATCH] D106559: [libomptarget][amdgpu] Implement dlopen of libhsa

[Jon Chesterfield via Phabricator via Openmp-commits]

JonChesterfield added inline comments.
Herald added a subscriber: sstefan1.


================
Comment at: openmp/libomptarget/plugins/amdgpu/CMakeLists.txt:91
+  ${LIBOMPTARGET_DEP_LIBRARIES}
+  ${CMAKE_DL_LIBS}
   ${LIBOMPTARGET_DEP_LIBELF_LIBRARIES}
----------------
drive by fix, this logic was inspired by the nvptx one, where CMAKE_DL_LIBS is used instead of dl


================
Comment at: openmp/libomptarget/plugins/amdgpu/dynamic_hsa/hsa.h:1
+////////////////////////////////////////////////////////////////////////////////
+//
----------------
This file was semi-automatically reduced from https://github.com/RadeonOpenCompute/ROCR-Runtime/tree/master/src/inc and preserves the header block as written there.

Dumped hsa symbols used by the amdgpu plugin then used grep to find them in the ROCR headers, manually cleaned up, then copied across the type definitions and enums that are currently in use.

The ROCR API is expected to be backwards compatible. Elided the comments on the basis that developers are better off reading the official header files.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D106559/new/

https://reviews.llvm.org/D106559