[OPENMP][NVPTX]Reduce number of barriers in reductions.

After the fix for the syncthreads we don't need to generate extra
barriers for the parallel reductions.

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@350530 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/CodeGen/CGOpenMPRuntimeNVPTX.cpp b/lib/CodeGen/CGOpenMPRuntimeNVPTX.cpp
index 21911c96f36ad89f56a99574e1b97cf8206ce9bb..403aefbef9ae8bc4fff9d9b7c052257857762145 100644 (file)
--- a/lib/CodeGen/CGOpenMPRuntimeNVPTX.cpp
+++ b/lib/CodeGen/CGOpenMPRuntimeNVPTX.cpp
@@ -3124,7 +3124,6 @@ static void emitReductionListCopy(
 ///     sync
 ///     if (I am the first warp)
 ///       Copy smem[thread_id] to my local D
-///     sync
 static llvm::Value *emitInterWarpCopyFunction(CodeGenModule &CGM,
                                               ArrayRef<const Expr *> Privates,
                                               QualType ReductionArrayTy,
@@ -3337,12 +3336,6 @@ static llvm::Value *emitInterWarpCopyFunction(CodeGenModule &CGM,
 
       CGF.EmitBlock(W0MergeBB);
 
-      // While warp 0 copies values from transfer medium, all other warps must
-      // wait.
-      // kmpc_barrier.
-      CGM.getOpenMPRuntime().emitBarrierCall(CGF, Loc, OMPD_unknown,
-                                             /*EmitChecks=*/false,
-                                             /*ForceSimpleCall=*/true);
       if (NumIters > 1) {
         Cnt = Bld.CreateNSWAdd(Cnt, llvm::ConstantInt::get(CGM.IntTy, /*V=*/1));
         CGF.EmitStoreOfScalar(Cnt, CntAddr, /*Volatile=*/false, C.IntTy);

diff --git a/test/OpenMP/nvptx_target_parallel_reduction_codegen.cpp b/test/OpenMP/nvptx_target_parallel_reduction_codegen.cpp
index 4f06f33e685050108034979430a39cdb20c70456..c23f275bd1a4e4010575547e0b13d5b5743ab104 100644 (file)
--- a/test/OpenMP/nvptx_target_parallel_reduction_codegen.cpp
+++ b/test/OpenMP/nvptx_target_parallel_reduction_codegen.cpp
@@ -231,7 +231,6 @@ int bar(int n){
   // CHECK: br label {{%?}}[[READ_CONT]]
   //
   // CHECK: [[READ_CONT]]
-  // CHECK: call void @__kmpc_barrier(%struct.ident_t* @
   // CHECK: [[NEXT:%.+]] = add nsw i32 [[CNT]], 1
   // CHECK: store i32 [[NEXT]], i32* [[CNT_ADDR]],
   // CHECK: br label
@@ -468,7 +467,6 @@ int bar(int n){
   //
   // CHECK: [[READ_CONT]]
   // CHECK: call void @__kmpc_barrier(%struct.ident_t* @
-  // CHECK: call void @__kmpc_barrier(%struct.ident_t* @
   // CHECK: [[IS_WARP_MASTER:%.+]] = icmp eq i32 [[LANEID]], 0
   // CHECK: br i1 [[IS_WARP_MASTER]], label {{%?}}[[DO_COPY:.+]], label {{%?}}[[COPY_ELSE:.+]]
   //
@@ -507,7 +505,6 @@ int bar(int n){
   // CHECK: br label {{%?}}[[READ_CONT]]
   //
   // CHECK: [[READ_CONT]]
-  // CHECK: call void @__kmpc_barrier(%struct.ident_t* @
   // CHECK: ret
 
 
@@ -823,7 +820,6 @@ int bar(int n){
   // CHECK: br label {{%?}}[[READ_CONT]]
   //
   // CHECK: [[READ_CONT]]
-  // CHECK: call void @__kmpc_barrier(%struct.ident_t* @
   // CHECK: ret
 
 #endif

diff --git a/test/OpenMP/nvptx_teams_reduction_codegen.cpp b/test/OpenMP/nvptx_teams_reduction_codegen.cpp
index b0246e2ee80e8426fb0ca14e90259bcc9801e4ec..0de25295a7a9873c616c6204f19fd5adb7602c49 100644 (file)
--- a/test/OpenMP/nvptx_teams_reduction_codegen.cpp
+++ b/test/OpenMP/nvptx_teams_reduction_codegen.cpp
@@ -473,7 +473,6 @@ int bar(int n){
   // CHECK: br label {{%?}}[[READ_CONT]]
   //
   // CHECK: [[READ_CONT]]
-  // CHECK: call void @__kmpc_barrier(%struct.ident_t* @
   // CHECK: ret
 
 #endif