[Openmp-commits] [PATCH] D111993: [libomptarget][WIP] Patch amdgpu DeviceRTL until it compiles

[Jon Chesterfield via Phabricator via Openmp-commits]

JonChesterfield added inline comments.


================
Comment at: clang/lib/Driver/ToolChains/AMDGPUOpenMP.cpp:255
                          options::OPT_fno_openmp_target_new_runtime, false))
-    BitcodeSuffix = "new-amdgcn-" + GPUArch;
+    BitcodeSuffix = "new-amdgpu-" + GPUArch;
   else
----------------
Naming is somewhat inconsistent. I think 'amdgcn' refers to 'graphics core next', but the gfx10 cards are branded 'radeon dna'. Changing to a new library seems a relatively good point to rename it.


================
Comment at: openmp/libomptarget/DeviceRTL/include/Synchronization.h:48
 /// Atomically read \p Addr with \p Ordering semantics.
 uint32_t read(uint32_t *Addr, int Ordering);
 
----------------
^ would like to call this 'load'


================
Comment at: openmp/libomptarget/DeviceRTL/src/Configuration.cpp:23
 
-extern uint32_t __omp_rtl_debug_kind;
+// extern uint32_t __omp_rtl_debug_kind;
 
----------------
What is this? Symbol is undefined on nvptx as far as I can tell. Amdgpu fails to link without a definition.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D111993/new/

https://reviews.llvm.org/D111993