[OpenMP][libomptarget] Fix __tgt_rtl_run_target_team_region_async API with missing parameter

[carlo.bertolli]

Summary

I missed the async info parameter in the first version of this API.

[carlo.bertolli]

Add context.

[JonChesterfield]

LG. Bit surprised the previous version built cleanly, extern C function type different to header should have been rejected

[JonChesterfield]

openmp/libomptarget/plugins/amdgpu/src/rtl.cpp

2289

Ah, that's why. No extern C here so no error checking, just defined an unrelated function

[carlo.bertolli]

It looks like https://reviews.llvm.org/D115823 will take a bit longer to be merged. I will merge this in to fix the API error, as it can go in first.

[JonChesterfield]

Yep, I need to revise that patch. Don't like how easy it is to misspell or mistype these functions, going to look for a fix for that

openmp/libomptarget/plugins/amdgpu/src/rtl.cpp

int32_t __tgt_rtl_run_target_region(int32_t device_id, void *tgt_entry_ptr,
   // fix thread num
   int32_t team_num = 1;
   int32_t thread_limit = 0; // use default
   return __tgt_rtl_run_target_team_region(device_id, tgt_entry_ptr, tgt_args,
                                           tgt_offsets, arg_num, team_num,
                                           thread_limit, 0);
 }
 
 int32_t __tgt_rtl_run_target_team_region_async(
     int32_t device_id, void *tgt_entry_ptr, void **tgt_args,
     ptrdiff_t *tgt_offsets, int32_t arg_num, int32_t num_teams,
-    int32_t thread_limit, uint64_t loop_tripcount) {
+    int32_t thread_limit, uint64_t loop_tripcount,
+    __tgt_async_info *AsyncInfo) {
+  assert(AsyncInfo && "AsyncInfo is nullptr");
+  initAsyncInfo(AsyncInfo);
 
   DeviceInfo.load_run_lock.lock_shared();
   int32_t res =
       runRegionLocked(device_id, tgt_entry_ptr, tgt_args, tgt_offsets, arg_num,
                       num_teams, thread_limit, loop_tripcount);
 
   DeviceInfo.load_run_lock.unlock_shared();
   return res;
 }
 
 int32_t __tgt_rtl_run_target_region_async(int32_t device_id,
                                           void *tgt_entry_ptr, void **tgt_args,
                                           ptrdiff_t *tgt_offsets,
                                           int32_t arg_num,
                                           __tgt_async_info *AsyncInfo) {
-  assert(AsyncInfo && "AsyncInfo is nullptr");
-  initAsyncInfo(AsyncInfo);
-
   // use one team and one thread
   // fix thread num
   int32_t team_num = 1;
   int32_t thread_limit = 0; // use default
-  return __tgt_rtl_run_target_team_region_async(device_id, tgt_entry_ptr,
-                                                tgt_args, tgt_offsets, arg_num,
-                                                team_num, thread_limit, 0);
+  return __tgt_rtl_run_target_team_region_async(
+      device_id, tgt_entry_ptr, tgt_args, tgt_offsets, arg_num, team_num,
+      thread_limit, 0, AsyncInfo);
 }
 
 int32_t __tgt_rtl_synchronize(int32_t device_id, __tgt_async_info *AsyncInfo) {
   assert(AsyncInfo && "AsyncInfo is nullptr");
 
   // Cuda asserts that AsyncInfo->Queue is non-null, but this invariant
   // is not ensured by devices.cpp for amdgcn
   // assert(AsyncInfo->Queue && "AsyncInfo->Queue is nullptr");