[OPENMP][NVPTX]Mark barrier functions calls as convergent.

Added convergent attribute to the barrier functions calls for correct
optimizations.

llvm-svn: 366437

diff --git a/clang/lib/CodeGen/CGOpenMPRuntimeNVPTX.cpp b/clang/lib/CodeGen/CGOpenMPRuntimeNVPTX.cpp
index 48dcbbf..8aa03de 100644 (file)
--- a/clang/lib/CodeGen/CGOpenMPRuntimeNVPTX.cpp
+++ b/clang/lib/CodeGen/CGOpenMPRuntimeNVPTX.cpp
@@ -2669,8 +2669,9 @@ void CGOpenMPRuntimeNVPTX::syncCTAThreads(CodeGenFunction &CGF) {
       llvm::ConstantPointerNull::get(
           cast<llvm::PointerType>(getIdentTyPointerTy())),
       llvm::ConstantInt::get(CGF.Int32Ty, /*V=*/0, /*isSigned=*/true)};
-  CGF.EmitRuntimeCall(
+  llvm::CallInst *Call = CGF.EmitRuntimeCall(
       createNVPTXRuntimeFunction(OMPRTL__kmpc_barrier_simple_spmd), Args);
+  Call->setConvergent();
 }
 
 void CGOpenMPRuntimeNVPTX::emitBarrierCall(CodeGenFunction &CGF,
@@ -2684,7 +2685,9 @@ void CGOpenMPRuntimeNVPTX::emitBarrierCall(CodeGenFunction &CGF,
   unsigned Flags = getDefaultFlagsForBarriers(Kind);
   llvm::Value *Args[] = {emitUpdateLocation(CGF, Loc, Flags),
                          getThreadID(CGF, Loc)};
-  CGF.EmitRuntimeCall(createNVPTXRuntimeFunction(OMPRTL__kmpc_barrier), Args);
+  llvm::CallInst *Call = CGF.EmitRuntimeCall(
+      createNVPTXRuntimeFunction(OMPRTL__kmpc_barrier), Args);
+  Call->setConvergent();
 }
 
 void CGOpenMPRuntimeNVPTX::emitCriticalRegion(

diff --git a/clang/test/OpenMP/nvptx_parallel_codegen.cpp b/clang/test/OpenMP/nvptx_parallel_codegen.cpp
index cdbc887..155ffc5 100644 (file)
--- a/clang/test/OpenMP/nvptx_parallel_codegen.cpp
+++ b/clang/test/OpenMP/nvptx_parallel_codegen.cpp
@@ -88,7 +88,7 @@ int bar(int n){
 // CHECK: br label {{%?}}[[AWAIT_WORK:.+]]
 //
 // CHECK: [[AWAIT_WORK]]
-// CHECK: call void @__kmpc_barrier_simple_spmd(%struct.ident_t* null, i32 0)
+// CHECK: call void @__kmpc_barrier_simple_spmd(%struct.ident_t* null, i32 0) #[[#BARRIER_ATTRS:]]
 // CHECK: [[KPR:%.+]] = call i1 @__kmpc_kernel_parallel(i8** [[OMP_WORK_FN]]
 // CHECK: [[KPRB:%.+]] = zext i1 [[KPR]] to i8
 // store i8 [[KPRB]], i8* [[OMP_EXEC_STATUS]], align 1
@@ -318,10 +318,10 @@ int bar(int n){
 // CHECK: define internal void [[PARALLEL_FN4]](
 // CHECK: [[A:%.+]] = alloca i[[SZ:32|64]],
 // CHECK: store i[[SZ]] 45, i[[SZ]]* %a,
-// CHECK: call void @__kmpc_barrier(%struct.ident_t* @{{.+}}, i32 %{{.+}})
+// CHECK: call void @__kmpc_barrier(%struct.ident_t* @{{.+}}, i32 %{{.+}}) #[[#BARRIER_ATTRS]]
 // CHECK: ret void
 
-// CHECK: declare void @__kmpc_barrier(%struct.ident_t*, i32) #[[BARRIER_ATTRS:.+]]
+// CHECK: declare void @__kmpc_barrier(%struct.ident_t*, i32) #[[#BARRIER_ATTRS]]
 
 // CHECK-LABEL: define {{.*}}void {{@__omp_offloading_.+template.+l55}}_worker()
 // CHECK-LABEL: define {{.*}}void {{@__omp_offloading_.+template.+l55}}(
@@ -367,6 +367,6 @@ int bar(int n){
 // CHECK:  store i32 [[NEW_CC_VAL]], i32* [[CC]],
 // CHECK:  br label
 
-// CHECK: attributes #[[BARRIER_ATTRS]] = {{.*}} convergent {{.*}}
+// CHECK: attributes #[[#BARRIER_ATTRS]] = {{.*}} convergent {{.*}}
 
 #endif