[Openmp-commits] [openmp] dc400f8 - [libomptarget] Deprecate old method for setting the tripcount

[Joseph Huber via Openmp-commits]

Author: Joseph Huber
Date: 2022-08-29T20:08:26-05:00
New Revision: dc400f8612a2c7e85fe297c1cde4f0ed01376611

URL: https://github.com/llvm/llvm-project/commit/dc400f8612a2c7e85fe297c1cde4f0ed01376611
DIFF: https://github.com/llvm/llvm-project/commit/dc400f8612a2c7e85fe297c1cde4f0ed01376611.diff

LOG: [libomptarget] Deprecate old method for setting the tripcount

Previously, the tripcount was set by a push call. We moved away from
this with the new interface that added the tripcount to the kernel
arguments struct, but kept around the old interface for legacy purposes
for the LLVM 15 release. This patch removes the support for the legacy
method.

This removes the support for the old method, but does not break
backwards compatibility. This will result in applications using the old
interface being slower when run on the device.

Reviewed By: jdoerfert

Differential Revision: https://reviews.llvm.org/D132885

Added: 
    

Modified: 
    openmp/libomptarget/include/device.h
    openmp/libomptarget/src/LegacyAPI.cpp
    openmp/libomptarget/src/omptarget.cpp

Removed: 
    


################################################################################
diff  --git a/openmp/libomptarget/include/device.h b/openmp/libomptarget/include/device.h
index 7f854ac5b4f9f..2e4ebf5846f8e 100644
--- a/openmp/libomptarget/include/device.h
+++ b/openmp/libomptarget/include/device.h
@@ -334,10 +334,6 @@ struct DeviceTy {
 
   std::mutex PendingGlobalsMtx, ShadowMtx;
 
-  // NOTE: Once libomp gains full target-task support, this state should be
-  // moved into the target task in libomp.
-  std::map<int32_t, uint64_t> LoopTripCnt;
-
   DeviceTy(RTLInfoTy *RTL);
   // DeviceTy is not copyable
   DeviceTy(const DeviceTy &D) = delete;

diff  --git a/openmp/libomptarget/src/LegacyAPI.cpp b/openmp/libomptarget/src/LegacyAPI.cpp
index d963be9d23059..654cfdcd4575e 100644
--- a/openmp/libomptarget/src/LegacyAPI.cpp
+++ b/openmp/libomptarget/src/LegacyAPI.cpp
@@ -77,7 +77,7 @@ EXTERN int __tgt_target_mapper(ident_t *Loc, int64_t DeviceId, void *HostPtr,
                                map_var_info_t *ArgNames, void **ArgMappers) {
   TIMESCOPE_WITH_IDENT(Loc);
   __tgt_kernel_arguments KernelArgs{
-      1, ArgNum, ArgsBase, Args, ArgSizes, ArgTypes, ArgNames, ArgMappers, -1};
+      1, ArgNum, ArgsBase, Args, ArgSizes, ArgTypes, ArgNames, ArgMappers, 0};
   return __tgt_target_kernel(Loc, DeviceId, -1, -1, HostPtr, &KernelArgs);
 }
 
@@ -120,7 +120,7 @@ EXTERN int __tgt_target_teams_mapper(ident_t *Loc, int64_t DeviceId,
   TIMESCOPE_WITH_IDENT(Loc);
 
   __tgt_kernel_arguments KernelArgs{
-      1, ArgNum, ArgsBase, Args, ArgSizes, ArgTypes, ArgNames, ArgMappers, -1};
+      1, ArgNum, ArgsBase, Args, ArgSizes, ArgTypes, ArgNames, ArgMappers, 0};
   return __tgt_target_kernel(Loc, DeviceId, NumTeams, ThreadLimit, HostPtr,
                              &KernelArgs);
 }
@@ -165,17 +165,7 @@ EXTERN int __tgt_target_teams_nowait_mapper(
 EXTERN void __kmpc_push_target_tripcount_mapper(ident_t *Loc, int64_t DeviceId,
                                                 uint64_t LoopTripcount) {
   TIMESCOPE_WITH_IDENT(Loc);
-  if (checkDeviceAndCtors(DeviceId, Loc)) {
-    DP("Not offloading to device %" PRId64 "\n", DeviceId);
-    return;
-  }
-
-  DP("__kmpc_push_target_tripcount(%" PRId64 ", %" PRIu64 ")\n", DeviceId,
-     LoopTripcount);
-  PM->TblMapMtx.lock();
-  PM->Devices[DeviceId]->LoopTripCnt[__kmpc_global_thread_num(NULL)] =
-      LoopTripcount;
-  PM->TblMapMtx.unlock();
+  DP("WARNING: __kmpc_push_target_tripcount has been deprecated and is a noop");
 }
 
 EXTERN void __kmpc_push_target_tripcount(int64_t DeviceId,

diff  --git a/openmp/libomptarget/src/omptarget.cpp b/openmp/libomptarget/src/omptarget.cpp
index 4615331948d1a..197a5489c11d6 100644
--- a/openmp/libomptarget/src/omptarget.cpp
+++ b/openmp/libomptarget/src/omptarget.cpp
@@ -1130,26 +1130,6 @@ TableMap *getTableMap(void *HostPtr) {
   return nullptr;
 }
 
-/// Get loop trip count
-/// FIXME: This function will not work right if calling
-/// __kmpc_push_target_tripcount_mapper in one thread but doing offloading in
-/// another thread, which might occur when we call task yield.
-uint64_t getLoopTripCount(int64_t DeviceId) {
-  DeviceTy &Device = *PM->Devices[DeviceId];
-  uint64_t LoopTripCount = 0;
-
-  {
-    std::lock_guard<std::mutex> TblMapLock(PM->TblMapMtx);
-    auto I = Device.LoopTripCnt.find(__kmpc_global_thread_num(NULL));
-    if (I != Device.LoopTripCnt.end()) {
-      LoopTripCount = I->second;
-      Device.LoopTripCnt.erase(I);
-    }
-  }
-
-  return LoopTripCount;
-}
-
 /// A class manages private arguments in a target region.
 class PrivateArgumentManagerTy {
   /// A data structure for the information of first-private arguments. We can
@@ -1512,8 +1492,6 @@ int target(ident_t *Loc, DeviceTy &Device, void *HostPtr, int32_t ArgNum,
   }
   assert(TargetTable && "Global data has not been mapped\n");
 
-  // FIXME: Use legacy tripcount method if it is '-1'.
-  Tripcount = Tripcount == -1UL ? getLoopTripCount(DeviceId) : Tripcount;
   DP("loop trip count is %" PRIu64 ".\n", Tripcount);
 
   // We need to keep bases and offsets separate. Sometimes (e.g. in OpenCL) we