Vendor import of llvm-project main llvmorg-15-init-16436-g18a6ab5b8d1f.vendor/llvm-project/llvmorg-15-init-16436-g18a6ab5b8d1f

1 files changed, 41 insertions, 99 deletions

diff --git a/clang/lib/CodeGen/CGOpenMPRuntime.cpp b/clang/lib/CodeGen/CGOpenMPRuntime.cpp
index 305040b01c08..091eb9da5af4 100644
--- a/clang/lib/CodeGen/CGOpenMPRuntime.cpp
+++ b/clang/lib/CodeGen/CGOpenMPRuntime.cpp
@@ -6717,11 +6717,9 @@ llvm::Value *CGOpenMPRuntime::emitNumTeamsForTargetDirective(
     default:
       break;
     }
-  } else if (DefaultNT == -1) {
-    return nullptr;
   }
 
-  return Bld.getInt32(DefaultNT);
+  return llvm::ConstantInt::get(CGF.Int32Ty, DefaultNT);
 }
 
 static llvm::Value *getNumThreads(CodeGenFunction &CGF, const CapturedStmt *CS,
@@ -10189,9 +10187,8 @@ llvm::Function *CGOpenMPRuntime::getOrCreateUserDefinedMapperFunc(
   return UDMMap.lookup(D);
 }
 
-void CGOpenMPRuntime::emitTargetNumIterationsCall(
+llvm::Value *CGOpenMPRuntime::emitTargetNumIterationsCall(
     CodeGenFunction &CGF, const OMPExecutableDirective &D,
-    llvm::Value *DeviceID,
     llvm::function_ref<llvm::Value *(CodeGenFunction &CGF,
                                      const OMPLoopDirective &D)>
         SizeEmitter) {
@@ -10201,20 +10198,12 @@ void CGOpenMPRuntime::emitTargetNumIterationsCall(
   if (!isOpenMPDistributeDirective(Kind) || !isOpenMPTeamsDirective(Kind))
     TD = getNestedDistributeDirective(CGM.getContext(), D);
   if (!TD)
-    return;
+    return llvm::ConstantInt::get(CGF.Int64Ty, 0);
+
   const auto *LD = cast<OMPLoopDirective>(TD);
-  auto &&CodeGen = [LD, DeviceID, SizeEmitter, &D, this](CodeGenFunction &CGF,
-                                                         PrePostActionTy &) {
-    if (llvm::Value *NumIterations = SizeEmitter(CGF, *LD)) {
-      llvm::Value *RTLoc = emitUpdateLocation(CGF, D.getBeginLoc());
-      llvm::Value *Args[] = {RTLoc, DeviceID, NumIterations};
-      CGF.EmitRuntimeCall(
-          OMPBuilder.getOrCreateRuntimeFunction(
-              CGM.getModule(), OMPRTL___kmpc_push_target_tripcount_mapper),
-          Args);
-    }
-  };
-  emitInlinedDirective(CGF, OMPD_unknown, CodeGen);
+  if (llvm::Value *NumIterations = SizeEmitter(CGF, *LD))
+    return NumIterations;
+  return llvm::ConstantInt::get(CGF.Int64Ty, 0);
 }
 
 void CGOpenMPRuntime::emitTargetCall(
@@ -10308,26 +10297,34 @@ void CGOpenMPRuntime::emitTargetCall(
     // Source location for the ident struct
     llvm::Value *RTLoc = emitUpdateLocation(CGF, D.getBeginLoc());
 
-    // Emit tripcount for the target loop-based directive.
-    emitTargetNumIterationsCall(CGF, D, DeviceID, SizeEmitter);
+    // Get tripcount for the target loop-based directive.
+    llvm::Value *NumIterations =
+        emitTargetNumIterationsCall(CGF, D, SizeEmitter);
+
+    // Arguments for the target kernel.
+    SmallVector<llvm::Value *> KernelArgs{
+        CGF.Builder.getInt32(/* Version */ 1),
+        PointerNum,
+        InputInfo.BasePointersArray.getPointer(),
+        InputInfo.PointersArray.getPointer(),
+        InputInfo.SizesArray.getPointer(),
+        MapTypesArray,
+        MapNamesArray,
+        InputInfo.MappersArray.getPointer(),
+        NumIterations};
+
+    // Arguments passed to the 'nowait' variant.
+    SmallVector<llvm::Value *> NoWaitKernelArgs{
+        CGF.Builder.getInt32(0),
+        llvm::ConstantPointerNull::get(CGM.VoidPtrTy),
+        CGF.Builder.getInt32(0),
+        llvm::ConstantPointerNull::get(CGM.VoidPtrTy),
+    };
+
+    bool HasNoWait = D.hasClausesOfKind<OMPNowaitClause>();
 
-    bool HasNowait = D.hasClausesOfKind<OMPNowaitClause>();
     // The target region is an outlined function launched by the runtime
-    // via calls __tgt_target() or __tgt_target_teams().
-    //
-    // __tgt_target() launches a target region with one team and one thread,
-    // executing a serial region.  This master thread may in turn launch
-    // more threads within its team upon encountering a parallel region,
-    // however, no additional teams can be launched on the device.
-    //
-    // __tgt_target_teams() launches a target region with one or more teams,
-    // each with one or more threads.  This call is required for target
-    // constructs such as:
-    //  'target teams'
-    //  'target' / 'teams'
-    //  'target teams distribute parallel for'
-    //  'target parallel'
-    // and so on.
+    // via calls to __tgt_target_kernel().
     //
     // Note that on the host and CPU targets, the runtime implementation of
     // these calls simply call the outlined function without forking threads.
@@ -10338,70 +10335,15 @@ void CGOpenMPRuntime::emitTargetCall(
     // In contrast, on the NVPTX target, the implementation of
     // __tgt_target_teams() launches a GPU kernel with the requested number
     // of teams and threads so no additional calls to the runtime are required.
-    if (NumTeams) {
-      // If we have NumTeams defined this means that we have an enclosed teams
-      // region. Therefore we also expect to have NumThreads defined. These two
-      // values should be defined in the presence of a teams directive,
-      // regardless of having any clauses associated. If the user is using teams
-      // but no clauses, these two values will be the default that should be
-      // passed to the runtime library - a 32-bit integer with the value zero.
-      assert(NumThreads && "Thread limit expression should be available along "
-                           "with number of teams.");
-      SmallVector<llvm::Value *> OffloadingArgs = {
-          RTLoc,
-          DeviceID,
-          OutlinedFnID,
-          PointerNum,
-          InputInfo.BasePointersArray.getPointer(),
-          InputInfo.PointersArray.getPointer(),
-          InputInfo.SizesArray.getPointer(),
-          MapTypesArray,
-          MapNamesArray,
-          InputInfo.MappersArray.getPointer(),
-          NumTeams,
-          NumThreads};
-      if (HasNowait) {
-        // Add int32_t depNum = 0, void *depList = nullptr, int32_t
-        // noAliasDepNum = 0, void *noAliasDepList = nullptr.
-        OffloadingArgs.push_back(CGF.Builder.getInt32(0));
-        OffloadingArgs.push_back(llvm::ConstantPointerNull::get(CGM.VoidPtrTy));
-        OffloadingArgs.push_back(CGF.Builder.getInt32(0));
-        OffloadingArgs.push_back(llvm::ConstantPointerNull::get(CGM.VoidPtrTy));
-      }
-      Return = CGF.EmitRuntimeCall(
-          OMPBuilder.getOrCreateRuntimeFunction(
-              CGM.getModule(), HasNowait
-                                   ? OMPRTL___tgt_target_teams_nowait_mapper
-                                   : OMPRTL___tgt_target_teams_mapper),
-          OffloadingArgs);
-    } else {
-      SmallVector<llvm::Value *> OffloadingArgs = {
-          RTLoc,
-          DeviceID,
-          OutlinedFnID,
-          PointerNum,
-          InputInfo.BasePointersArray.getPointer(),
-          InputInfo.PointersArray.getPointer(),
-          InputInfo.SizesArray.getPointer(),
-          MapTypesArray,
-          MapNamesArray,
-          InputInfo.MappersArray.getPointer()};
-      if (HasNowait) {
-        // Add int32_t depNum = 0, void *depList = nullptr, int32_t
-        // noAliasDepNum = 0, void *noAliasDepList = nullptr.
-        OffloadingArgs.push_back(CGF.Builder.getInt32(0));
-        OffloadingArgs.push_back(llvm::ConstantPointerNull::get(CGM.VoidPtrTy));
-        OffloadingArgs.push_back(CGF.Builder.getInt32(0));
-        OffloadingArgs.push_back(llvm::ConstantPointerNull::get(CGM.VoidPtrTy));
-      }
-      Return = CGF.EmitRuntimeCall(
-          OMPBuilder.getOrCreateRuntimeFunction(
-              CGM.getModule(), HasNowait ? OMPRTL___tgt_target_nowait_mapper
-                                         : OMPRTL___tgt_target_mapper),
-          OffloadingArgs);
-    }
-
     // Check the error code and execute the host version if required.
+    CGF.Builder.restoreIP(
+        HasNoWait ? OMPBuilder.emitTargetKernel(
+                        CGF.Builder, Return, RTLoc, DeviceID, NumTeams,
+                        NumThreads, OutlinedFnID, KernelArgs, NoWaitKernelArgs)
+                  : OMPBuilder.emitTargetKernel(CGF.Builder, Return, RTLoc,
+                                                DeviceID, NumTeams, NumThreads,
+                                                OutlinedFnID, KernelArgs));
+
     llvm::BasicBlock *OffloadFailedBlock =
         CGF.createBasicBlock("omp_offload.failed");
     llvm::BasicBlock *OffloadContBlock =