Author: Joseph Huber
Date: 2021-10-08T22:21:59-04:00
New Revision: bad44d5f39bc29c1062f0c4ce8b71720734ad3cb

URL: 
https://github.com/llvm/llvm-project/commit/bad44d5f39bc29c1062f0c4ce8b71720734ad3cb
DIFF: 
https://github.com/llvm/llvm-project/commit/bad44d5f39bc29c1062f0c4ce8b71720734ad3cb.diff

LOG: [OpenMP] Add RTL function for getting number of threads in block.

This patch adds support for the
`__kmpc_get_hardware_num_threads_in_block` function that returns the
number of threads. This was missing in the new runtime and was used by
the AMDGPU plugin which prevented it from using the new runtime. This
patchs also unified the interface for getting the thread numbers in the
frontend.

Originally authored by jdoerfert.

Reviewed By: JonChesterfield

Differential Revision: https://reviews.llvm.org/D111475

Added: 
    

Modified: 
    clang/lib/CodeGen/CGOpenMPRuntimeAMDGCN.cpp
    clang/lib/CodeGen/CGOpenMPRuntimeAMDGCN.h
    clang/lib/CodeGen/CGOpenMPRuntimeGPU.cpp
    clang/lib/CodeGen/CGOpenMPRuntimeGPU.h
    clang/lib/CodeGen/CGOpenMPRuntimeNVPTX.cpp
    clang/lib/CodeGen/CGOpenMPRuntimeNVPTX.h
    clang/test/OpenMP/nvptx_distribute_parallel_generic_mode_codegen.cpp
    clang/test/OpenMP/nvptx_parallel_codegen.cpp
    clang/test/OpenMP/nvptx_target_teams_distribute_parallel_for_codegen.cpp
    
clang/test/OpenMP/nvptx_target_teams_distribute_parallel_for_generic_mode_codegen.cpp
    
clang/test/OpenMP/nvptx_target_teams_distribute_parallel_for_simd_codegen.cpp
    openmp/libomptarget/DeviceRTL/include/Interface.h
    openmp/libomptarget/DeviceRTL/src/Mapping.cpp
    openmp/libomptarget/DeviceRTL/src/Utils.cpp

Removed: 
    


################################################################################
diff  --git a/clang/lib/CodeGen/CGOpenMPRuntimeAMDGCN.cpp 
b/clang/lib/CodeGen/CGOpenMPRuntimeAMDGCN.cpp
index cac5faaa8d0f2..cf6f6bc3654c0 100644
--- a/clang/lib/CodeGen/CGOpenMPRuntimeAMDGCN.cpp
+++ b/clang/lib/CodeGen/CGOpenMPRuntimeAMDGCN.cpp
@@ -46,16 +46,3 @@ llvm::Value 
*CGOpenMPRuntimeAMDGCN::getGPUThreadID(CodeGenFunction &CGF) {
       CGF.CGM.getIntrinsic(llvm::Intrinsic::amdgcn_workitem_id_x);
   return Bld.CreateCall(F, llvm::None, "nvptx_tid");
 }
-
-llvm::Value *CGOpenMPRuntimeAMDGCN::getGPUNumThreads(CodeGenFunction &CGF) {
-  CGBuilderTy &Bld = CGF.Builder;
-  llvm::Module *M = &CGF.CGM.getModule();
-  const char *LocSize = "__kmpc_amdgcn_gpu_num_threads";
-  llvm::Function *F = M->getFunction(LocSize);
-  if (!F) {
-    F = llvm::Function::Create(
-        llvm::FunctionType::get(CGF.Int32Ty, llvm::None, false),
-        llvm::GlobalVariable::ExternalLinkage, LocSize, &CGF.CGM.getModule());
-  }
-  return Bld.CreateCall(F, llvm::None, "nvptx_num_threads");
-}

diff  --git a/clang/lib/CodeGen/CGOpenMPRuntimeAMDGCN.h 
b/clang/lib/CodeGen/CGOpenMPRuntimeAMDGCN.h
index c1421261bfc19..cecbcaf3c3d36 100644
--- a/clang/lib/CodeGen/CGOpenMPRuntimeAMDGCN.h
+++ b/clang/lib/CodeGen/CGOpenMPRuntimeAMDGCN.h
@@ -32,9 +32,6 @@ class CGOpenMPRuntimeAMDGCN final : public CGOpenMPRuntimeGPU 
{
 
   /// Get the id of the current thread on the GPU.
   llvm::Value *getGPUThreadID(CodeGenFunction &CGF) override;
-
-  /// Get the maximum number of threads in a block of the GPU.
-  llvm::Value *getGPUNumThreads(CodeGenFunction &CGF) override;
 };
 
 } // namespace CodeGen

diff  --git a/clang/lib/CodeGen/CGOpenMPRuntimeGPU.cpp 
b/clang/lib/CodeGen/CGOpenMPRuntimeGPU.cpp
index be47ccbbbc58a..366482085d4fa 100644
--- a/clang/lib/CodeGen/CGOpenMPRuntimeGPU.cpp
+++ b/clang/lib/CodeGen/CGOpenMPRuntimeGPU.cpp
@@ -3947,3 +3947,16 @@ void CGOpenMPRuntimeGPU::clear() {
   }
   CGOpenMPRuntime::clear();
 }
+
+llvm::Value *CGOpenMPRuntimeGPU::getGPUNumThreads(CodeGenFunction &CGF) {
+  CGBuilderTy &Bld = CGF.Builder;
+  llvm::Module *M = &CGF.CGM.getModule();
+  const char *LocSize = "__kmpc_get_hardware_num_threads_in_block";
+  llvm::Function *F = M->getFunction(LocSize);
+  if (!F) {
+    F = llvm::Function::Create(
+        llvm::FunctionType::get(CGF.Int32Ty, llvm::None, false),
+        llvm::GlobalVariable::ExternalLinkage, LocSize, &CGF.CGM.getModule());
+  }
+  return Bld.CreateCall(F, llvm::None, "nvptx_num_threads");
+}

diff  --git a/clang/lib/CodeGen/CGOpenMPRuntimeGPU.h 
b/clang/lib/CodeGen/CGOpenMPRuntimeGPU.h
index 5d3b711e6d4b5..cc069844660b4 100644
--- a/clang/lib/CodeGen/CGOpenMPRuntimeGPU.h
+++ b/clang/lib/CodeGen/CGOpenMPRuntimeGPU.h
@@ -182,7 +182,7 @@ class CGOpenMPRuntimeGPU : public CGOpenMPRuntime {
   virtual llvm::Value *getGPUThreadID(CodeGenFunction &CGF) = 0;
 
   /// Get the maximum number of threads in a block of the GPU.
-  virtual llvm::Value *getGPUNumThreads(CodeGenFunction &CGF) = 0;
+  llvm::Value *getGPUNumThreads(CodeGenFunction &CGF);
 
   /// Emit call to void __kmpc_push_proc_bind(ident_t *loc, kmp_int32
   /// global_tid, int proc_bind) to generate code for 'proc_bind' clause.

diff  --git a/clang/lib/CodeGen/CGOpenMPRuntimeNVPTX.cpp 
b/clang/lib/CodeGen/CGOpenMPRuntimeNVPTX.cpp
index 1688d07b90b6e..ae07ffd3c93fb 100644
--- a/clang/lib/CodeGen/CGOpenMPRuntimeNVPTX.cpp
+++ b/clang/lib/CodeGen/CGOpenMPRuntimeNVPTX.cpp
@@ -46,11 +46,3 @@ llvm::Value 
*CGOpenMPRuntimeNVPTX::getGPUThreadID(CodeGenFunction &CGF) {
       &CGF.CGM.getModule(), llvm::Intrinsic::nvvm_read_ptx_sreg_tid_x);
   return Bld.CreateCall(F, llvm::None, "nvptx_tid");
 }
-
-llvm::Value *CGOpenMPRuntimeNVPTX::getGPUNumThreads(CodeGenFunction &CGF) {
-  CGBuilderTy &Bld = CGF.Builder;
-  llvm::Function *F;
-  F = llvm::Intrinsic::getDeclaration(
-      &CGF.CGM.getModule(), llvm::Intrinsic::nvvm_read_ptx_sreg_ntid_x);
-  return Bld.CreateCall(F, llvm::None, "nvptx_num_threads");
-}

diff  --git a/clang/lib/CodeGen/CGOpenMPRuntimeNVPTX.h 
b/clang/lib/CodeGen/CGOpenMPRuntimeNVPTX.h
index 5f16029592665..ddb4fc2c59b2c 100644
--- a/clang/lib/CodeGen/CGOpenMPRuntimeNVPTX.h
+++ b/clang/lib/CodeGen/CGOpenMPRuntimeNVPTX.h
@@ -32,9 +32,6 @@ class CGOpenMPRuntimeNVPTX final : public CGOpenMPRuntimeGPU {
 
   /// Get the id of the current thread on the GPU.
   llvm::Value *getGPUThreadID(CodeGenFunction &CGF) override;
-
-  /// Get the maximum number of threads in a block of the GPU.
-  llvm::Value *getGPUNumThreads(CodeGenFunction &CGF) override;
 };
 
 } // CodeGen namespace.

diff  --git 
a/clang/test/OpenMP/nvptx_distribute_parallel_generic_mode_codegen.cpp 
b/clang/test/OpenMP/nvptx_distribute_parallel_generic_mode_codegen.cpp
index c24a4fa39cdf2..676d5a4ded4c5 100644
--- a/clang/test/OpenMP/nvptx_distribute_parallel_generic_mode_codegen.cpp
+++ b/clang/test/OpenMP/nvptx_distribute_parallel_generic_mode_codegen.cpp
@@ -3005,7 +3005,7 @@ int main(int argc, char **argv) {
 // CHECK4-NEXT:    store i32 [[TMP6]], i32* [[CONV1]], align 4
 // CHECK4-NEXT:    [[TMP7:%.*]] = load i64, i64* [[ARGC_CASTED]], align 8
 // CHECK4-NEXT:    store i32 [[TMP5]], i32* [[DOTTHREADID_TEMP_]], align 4
-// CHECK4-NEXT:    call void @__omp_outlined__(i32* [[DOTTHREADID_TEMP_]], 
i32* [[DOTZERO_ADDR]], [10 x i32]* [[TMP0]], [10 x i32]* [[TMP1]], i32* 
[[TMP2]], i64 [[TMP7]], [10 x i32]* [[TMP3]]) #[[ATTR5:[0-9]+]]
+// CHECK4-NEXT:    call void @__omp_outlined__(i32* [[DOTTHREADID_TEMP_]], 
i32* [[DOTZERO_ADDR]], [10 x i32]* [[TMP0]], [10 x i32]* [[TMP1]], i32* 
[[TMP2]], i64 [[TMP7]], [10 x i32]* [[TMP3]]) #[[ATTR4:[0-9]+]]
 // CHECK4-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* 
@[[GLOB1]], i8 2, i1 false)
 // CHECK4-NEXT:    ret void
 // CHECK4:       worker.exit:
@@ -3068,7 +3068,7 @@ int main(int argc, char **argv) {
 // CHECK4-NEXT:    [[TMP8:%.*]] = bitcast [10 x i32]* [[B4]] to i8*
 // CHECK4-NEXT:    [[TMP9:%.*]] = bitcast [10 x i32]* [[TMP0]] to i8*
 // CHECK4-NEXT:    call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 4 [[TMP8]], 
i8* align 4 [[TMP9]], i64 40, i1 false)
-// CHECK4-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 
@llvm.nvvm.read.ptx.sreg.ntid.x()
+// CHECK4-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 
@__kmpc_get_hardware_num_threads_in_block()
 // CHECK4-NEXT:    [[TMP10:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], 
align 8
 // CHECK4-NEXT:    [[TMP11:%.*]] = load i32, i32* [[TMP10]], align 4
 // CHECK4-NEXT:    call void @__kmpc_distribute_static_init_4(%struct.ident_t* 
@[[GLOB2:[0-9]+]], i32 [[TMP11]], i32 91, i32* [[DOTOMP_IS_LAST]], i32* 
[[DOTOMP_COMB_LB]], i32* [[DOTOMP_COMB_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 
[[NVPTX_NUM_THREADS]])
@@ -3259,23 +3259,23 @@ int main(int argc, char **argv) {
 // CHECK4-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP18]], 1
 // CHECK4-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
 // CHECK4-NEXT:    store i32 [[ADD]], i32* [[I6]], align 4
-// CHECK4-NEXT:    [[CALL:%.*]] = call i32 @_Z3fooPi(i32* [[I6]]) 
#[[ATTR7:[0-9]+]]
-// CHECK4-NEXT:    [[CALL9:%.*]] = call i32 @_Z3fooPi(i32* [[TMP1]]) #[[ATTR7]]
+// CHECK4-NEXT:    [[CALL:%.*]] = call i32 @_Z3fooPi(i32* [[I6]]) 
#[[ATTR6:[0-9]+]]
+// CHECK4-NEXT:    [[CALL9:%.*]] = call i32 @_Z3fooPi(i32* [[TMP1]]) #[[ATTR6]]
 // CHECK4-NEXT:    [[ADD10:%.*]] = add nsw i32 [[CALL]], [[CALL9]]
 // CHECK4-NEXT:    [[TMP19:%.*]] = load i32, i32* [[I6]], align 4
 // CHECK4-NEXT:    [[IDXPROM:%.*]] = sext i32 [[TMP19]] to i64
 // CHECK4-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [10 x i32], [10 x 
i32]* [[B4]], i64 0, i64 [[IDXPROM]]
-// CHECK4-NEXT:    [[CALL11:%.*]] = call i32 @_Z3fooPi(i32* [[ARRAYIDX]]) 
#[[ATTR7]]
+// CHECK4-NEXT:    [[CALL11:%.*]] = call i32 @_Z3fooPi(i32* [[ARRAYIDX]]) 
#[[ATTR6]]
 // CHECK4-NEXT:    [[ADD12:%.*]] = add nsw i32 [[ADD10]], [[CALL11]]
 // CHECK4-NEXT:    [[TMP20:%.*]] = load i32, i32* [[I6]], align 4
 // CHECK4-NEXT:    [[IDXPROM13:%.*]] = sext i32 [[TMP20]] to i64
 // CHECK4-NEXT:    [[ARRAYIDX14:%.*]] = getelementptr inbounds [10 x i32], [10 
x i32]* [[C5]], i64 0, i64 [[IDXPROM13]]
-// CHECK4-NEXT:    [[CALL15:%.*]] = call i32 @_Z3fooPi(i32* [[ARRAYIDX14]]) 
#[[ATTR7]]
+// CHECK4-NEXT:    [[CALL15:%.*]] = call i32 @_Z3fooPi(i32* [[ARRAYIDX14]]) 
#[[ATTR6]]
 // CHECK4-NEXT:    [[ADD16:%.*]] = add nsw i32 [[ADD12]], [[CALL15]]
 // CHECK4-NEXT:    [[TMP21:%.*]] = load i32, i32* [[I6]], align 4
 // CHECK4-NEXT:    [[IDXPROM17:%.*]] = sext i32 [[TMP21]] to i64
 // CHECK4-NEXT:    [[ARRAYIDX18:%.*]] = getelementptr inbounds [10 x i32], [10 
x i32]* [[TMP4]], i64 0, i64 [[IDXPROM17]]
-// CHECK4-NEXT:    [[CALL19:%.*]] = call i32 @_Z3fooPi(i32* [[ARRAYIDX18]]) 
#[[ATTR7]]
+// CHECK4-NEXT:    [[CALL19:%.*]] = call i32 @_Z3fooPi(i32* [[ARRAYIDX18]]) 
#[[ATTR6]]
 // CHECK4-NEXT:    [[ADD20:%.*]] = add nsw i32 [[ADD16]], [[CALL19]]
 // CHECK4-NEXT:    store i32 [[ADD20]], i32* [[TMP1]], align 4
 // CHECK4-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
@@ -3337,7 +3337,7 @@ int main(int argc, char **argv) {
 // CHECK5-NEXT:    store i32 [[TMP6]], i32* [[ARGC_CASTED]], align 4
 // CHECK5-NEXT:    [[TMP7:%.*]] = load i32, i32* [[ARGC_CASTED]], align 4
 // CHECK5-NEXT:    store i32 [[TMP5]], i32* [[DOTTHREADID_TEMP_]], align 4
-// CHECK5-NEXT:    call void @__omp_outlined__(i32* [[DOTTHREADID_TEMP_]], 
i32* [[DOTZERO_ADDR]], [10 x i32]* [[TMP0]], [10 x i32]* [[TMP1]], i32* 
[[TMP2]], i32 [[TMP7]], [10 x i32]* [[TMP3]]) #[[ATTR5:[0-9]+]]
+// CHECK5-NEXT:    call void @__omp_outlined__(i32* [[DOTTHREADID_TEMP_]], 
i32* [[DOTZERO_ADDR]], [10 x i32]* [[TMP0]], [10 x i32]* [[TMP1]], i32* 
[[TMP2]], i32 [[TMP7]], [10 x i32]* [[TMP3]]) #[[ATTR4:[0-9]+]]
 // CHECK5-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* 
@[[GLOB1]], i8 2, i1 false)
 // CHECK5-NEXT:    ret void
 // CHECK5:       worker.exit:
@@ -3399,7 +3399,7 @@ int main(int argc, char **argv) {
 // CHECK5-NEXT:    [[TMP8:%.*]] = bitcast [10 x i32]* [[B4]] to i8*
 // CHECK5-NEXT:    [[TMP9:%.*]] = bitcast [10 x i32]* [[TMP0]] to i8*
 // CHECK5-NEXT:    call void @llvm.memcpy.p0i8.p0i8.i32(i8* align 4 [[TMP8]], 
i8* align 4 [[TMP9]], i32 40, i1 false)
-// CHECK5-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 
@llvm.nvvm.read.ptx.sreg.ntid.x()
+// CHECK5-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 
@__kmpc_get_hardware_num_threads_in_block()
 // CHECK5-NEXT:    [[TMP10:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], 
align 4
 // CHECK5-NEXT:    [[TMP11:%.*]] = load i32, i32* [[TMP10]], align 4
 // CHECK5-NEXT:    call void @__kmpc_distribute_static_init_4(%struct.ident_t* 
@[[GLOB2:[0-9]+]], i32 [[TMP11]], i32 91, i32* [[DOTOMP_IS_LAST]], i32* 
[[DOTOMP_COMB_LB]], i32* [[DOTOMP_COMB_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 
[[NVPTX_NUM_THREADS]])
@@ -3585,20 +3585,20 @@ int main(int argc, char **argv) {
 // CHECK5-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP18]], 1
 // CHECK5-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
 // CHECK5-NEXT:    store i32 [[ADD]], i32* [[I5]], align 4
-// CHECK5-NEXT:    [[CALL:%.*]] = call i32 @_Z3fooPi(i32* [[I5]]) 
#[[ATTR7:[0-9]+]]
-// CHECK5-NEXT:    [[CALL7:%.*]] = call i32 @_Z3fooPi(i32* [[TMP1]]) #[[ATTR7]]
+// CHECK5-NEXT:    [[CALL:%.*]] = call i32 @_Z3fooPi(i32* [[I5]]) 
#[[ATTR6:[0-9]+]]
+// CHECK5-NEXT:    [[CALL7:%.*]] = call i32 @_Z3fooPi(i32* [[TMP1]]) #[[ATTR6]]
 // CHECK5-NEXT:    [[ADD8:%.*]] = add nsw i32 [[CALL]], [[CALL7]]
 // CHECK5-NEXT:    [[TMP19:%.*]] = load i32, i32* [[I5]], align 4
 // CHECK5-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [10 x i32], [10 x 
i32]* [[B3]], i32 0, i32 [[TMP19]]
-// CHECK5-NEXT:    [[CALL9:%.*]] = call i32 @_Z3fooPi(i32* [[ARRAYIDX]]) 
#[[ATTR7]]
+// CHECK5-NEXT:    [[CALL9:%.*]] = call i32 @_Z3fooPi(i32* [[ARRAYIDX]]) 
#[[ATTR6]]
 // CHECK5-NEXT:    [[ADD10:%.*]] = add nsw i32 [[ADD8]], [[CALL9]]
 // CHECK5-NEXT:    [[TMP20:%.*]] = load i32, i32* [[I5]], align 4
 // CHECK5-NEXT:    [[ARRAYIDX11:%.*]] = getelementptr inbounds [10 x i32], [10 
x i32]* [[C4]], i32 0, i32 [[TMP20]]
-// CHECK5-NEXT:    [[CALL12:%.*]] = call i32 @_Z3fooPi(i32* [[ARRAYIDX11]]) 
#[[ATTR7]]
+// CHECK5-NEXT:    [[CALL12:%.*]] = call i32 @_Z3fooPi(i32* [[ARRAYIDX11]]) 
#[[ATTR6]]
 // CHECK5-NEXT:    [[ADD13:%.*]] = add nsw i32 [[ADD10]], [[CALL12]]
 // CHECK5-NEXT:    [[TMP21:%.*]] = load i32, i32* [[I5]], align 4
 // CHECK5-NEXT:    [[ARRAYIDX14:%.*]] = getelementptr inbounds [10 x i32], [10 
x i32]* [[TMP4]], i32 0, i32 [[TMP21]]
-// CHECK5-NEXT:    [[CALL15:%.*]] = call i32 @_Z3fooPi(i32* [[ARRAYIDX14]]) 
#[[ATTR7]]
+// CHECK5-NEXT:    [[CALL15:%.*]] = call i32 @_Z3fooPi(i32* [[ARRAYIDX14]]) 
#[[ATTR6]]
 // CHECK5-NEXT:    [[ADD16:%.*]] = add nsw i32 [[ADD13]], [[CALL15]]
 // CHECK5-NEXT:    store i32 [[ADD16]], i32* [[TMP1]], align 4
 // CHECK5-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
@@ -3660,7 +3660,7 @@ int main(int argc, char **argv) {
 // CHECK6-NEXT:    store i32 [[TMP6]], i32* [[ARGC_CASTED]], align 4
 // CHECK6-NEXT:    [[TMP7:%.*]] = load i32, i32* [[ARGC_CASTED]], align 4
 // CHECK6-NEXT:    store i32 [[TMP5]], i32* [[DOTTHREADID_TEMP_]], align 4
-// CHECK6-NEXT:    call void @__omp_outlined__(i32* [[DOTTHREADID_TEMP_]], 
i32* [[DOTZERO_ADDR]], [10 x i32]* [[TMP0]], [10 x i32]* [[TMP1]], i32* 
[[TMP2]], i32 [[TMP7]], [10 x i32]* [[TMP3]]) #[[ATTR5:[0-9]+]]
+// CHECK6-NEXT:    call void @__omp_outlined__(i32* [[DOTTHREADID_TEMP_]], 
i32* [[DOTZERO_ADDR]], [10 x i32]* [[TMP0]], [10 x i32]* [[TMP1]], i32* 
[[TMP2]], i32 [[TMP7]], [10 x i32]* [[TMP3]]) #[[ATTR4:[0-9]+]]
 // CHECK6-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* 
@[[GLOB1]], i8 2, i1 false)
 // CHECK6-NEXT:    ret void
 // CHECK6:       worker.exit:
@@ -3722,7 +3722,7 @@ int main(int argc, char **argv) {
 // CHECK6-NEXT:    [[TMP8:%.*]] = bitcast [10 x i32]* [[B4]] to i8*
 // CHECK6-NEXT:    [[TMP9:%.*]] = bitcast [10 x i32]* [[TMP0]] to i8*
 // CHECK6-NEXT:    call void @llvm.memcpy.p0i8.p0i8.i32(i8* align 4 [[TMP8]], 
i8* align 4 [[TMP9]], i32 40, i1 false)
-// CHECK6-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 
@llvm.nvvm.read.ptx.sreg.ntid.x()
+// CHECK6-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 
@__kmpc_get_hardware_num_threads_in_block()
 // CHECK6-NEXT:    [[TMP10:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], 
align 4
 // CHECK6-NEXT:    [[TMP11:%.*]] = load i32, i32* [[TMP10]], align 4
 // CHECK6-NEXT:    call void @__kmpc_distribute_static_init_4(%struct.ident_t* 
@[[GLOB2:[0-9]+]], i32 [[TMP11]], i32 91, i32* [[DOTOMP_IS_LAST]], i32* 
[[DOTOMP_COMB_LB]], i32* [[DOTOMP_COMB_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 
[[NVPTX_NUM_THREADS]])
@@ -3908,20 +3908,20 @@ int main(int argc, char **argv) {
 // CHECK6-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP18]], 1
 // CHECK6-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
 // CHECK6-NEXT:    store i32 [[ADD]], i32* [[I5]], align 4
-// CHECK6-NEXT:    [[CALL:%.*]] = call i32 @_Z3fooPi(i32* [[I5]]) 
#[[ATTR7:[0-9]+]]
-// CHECK6-NEXT:    [[CALL7:%.*]] = call i32 @_Z3fooPi(i32* [[TMP1]]) #[[ATTR7]]
+// CHECK6-NEXT:    [[CALL:%.*]] = call i32 @_Z3fooPi(i32* [[I5]]) 
#[[ATTR6:[0-9]+]]
+// CHECK6-NEXT:    [[CALL7:%.*]] = call i32 @_Z3fooPi(i32* [[TMP1]]) #[[ATTR6]]
 // CHECK6-NEXT:    [[ADD8:%.*]] = add nsw i32 [[CALL]], [[CALL7]]
 // CHECK6-NEXT:    [[TMP19:%.*]] = load i32, i32* [[I5]], align 4
 // CHECK6-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [10 x i32], [10 x 
i32]* [[B3]], i32 0, i32 [[TMP19]]
-// CHECK6-NEXT:    [[CALL9:%.*]] = call i32 @_Z3fooPi(i32* [[ARRAYIDX]]) 
#[[ATTR7]]
+// CHECK6-NEXT:    [[CALL9:%.*]] = call i32 @_Z3fooPi(i32* [[ARRAYIDX]]) 
#[[ATTR6]]
 // CHECK6-NEXT:    [[ADD10:%.*]] = add nsw i32 [[ADD8]], [[CALL9]]
 // CHECK6-NEXT:    [[TMP20:%.*]] = load i32, i32* [[I5]], align 4
 // CHECK6-NEXT:    [[ARRAYIDX11:%.*]] = getelementptr inbounds [10 x i32], [10 
x i32]* [[C4]], i32 0, i32 [[TMP20]]
-// CHECK6-NEXT:    [[CALL12:%.*]] = call i32 @_Z3fooPi(i32* [[ARRAYIDX11]]) 
#[[ATTR7]]
+// CHECK6-NEXT:    [[CALL12:%.*]] = call i32 @_Z3fooPi(i32* [[ARRAYIDX11]]) 
#[[ATTR6]]
 // CHECK6-NEXT:    [[ADD13:%.*]] = add nsw i32 [[ADD10]], [[CALL12]]
 // CHECK6-NEXT:    [[TMP21:%.*]] = load i32, i32* [[I5]], align 4
 // CHECK6-NEXT:    [[ARRAYIDX14:%.*]] = getelementptr inbounds [10 x i32], [10 
x i32]* [[TMP4]], i32 0, i32 [[TMP21]]
-// CHECK6-NEXT:    [[CALL15:%.*]] = call i32 @_Z3fooPi(i32* [[ARRAYIDX14]]) 
#[[ATTR7]]
+// CHECK6-NEXT:    [[CALL15:%.*]] = call i32 @_Z3fooPi(i32* [[ARRAYIDX14]]) 
#[[ATTR6]]
 // CHECK6-NEXT:    [[ADD16:%.*]] = add nsw i32 [[ADD13]], [[CALL15]]
 // CHECK6-NEXT:    store i32 [[ADD16]], i32* [[TMP1]], align 4
 // CHECK6-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]

diff  --git a/clang/test/OpenMP/nvptx_parallel_codegen.cpp 
b/clang/test/OpenMP/nvptx_parallel_codegen.cpp
index 35d6813d25267..96be6915cdbec 100644
--- a/clang/test/OpenMP/nvptx_parallel_codegen.cpp
+++ b/clang/test/OpenMP/nvptx_parallel_codegen.cpp
@@ -1665,7 +1665,7 @@ int bar(int n){
 // CHECK1-NEXT:    [[TMP0:%.*]] = load i32*, i32** [[A_ADDR]], align 8
 // CHECK1-NEXT:    [[TMP1:%.*]] = call i64 @__kmpc_warp_active_thread_mask()
 // CHECK1-NEXT:    [[NVPTX_TID:%.*]] = call i32 
@llvm.nvvm.read.ptx.sreg.tid.x()
-// CHECK1-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 
@llvm.nvvm.read.ptx.sreg.ntid.x()
+// CHECK1-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 
@__kmpc_get_hardware_num_threads_in_block()
 // CHECK1-NEXT:    store i32 0, i32* [[CRITICAL_COUNTER]], align 4
 // CHECK1-NEXT:    br label [[OMP_CRITICAL_LOOP:%.*]]
 // CHECK1:       omp.critical.loop:
@@ -1937,7 +1937,7 @@ int bar(int n){
 // CHECK2-NEXT:    [[TMP0:%.*]] = load i32*, i32** [[A_ADDR]], align 4
 // CHECK2-NEXT:    [[TMP1:%.*]] = call i64 @__kmpc_warp_active_thread_mask()
 // CHECK2-NEXT:    [[NVPTX_TID:%.*]] = call i32 
@llvm.nvvm.read.ptx.sreg.tid.x()
-// CHECK2-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 
@llvm.nvvm.read.ptx.sreg.ntid.x()
+// CHECK2-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 
@__kmpc_get_hardware_num_threads_in_block()
 // CHECK2-NEXT:    store i32 0, i32* [[CRITICAL_COUNTER]], align 4
 // CHECK2-NEXT:    br label [[OMP_CRITICAL_LOOP:%.*]]
 // CHECK2:       omp.critical.loop:

diff  --git 
a/clang/test/OpenMP/nvptx_target_teams_distribute_parallel_for_codegen.cpp 
b/clang/test/OpenMP/nvptx_target_teams_distribute_parallel_for_codegen.cpp
index b71cc83fbe784..c415a7493f813 100644
--- a/clang/test/OpenMP/nvptx_target_teams_distribute_parallel_for_codegen.cpp
+++ b/clang/test/OpenMP/nvptx_target_teams_distribute_parallel_for_codegen.cpp
@@ -18866,7 +18866,7 @@ int bar(int n){
 // CHECK1-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_COMB_UB]], align 4
 // CHECK1-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
 // CHECK1-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
-// CHECK1-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 
@llvm.nvvm.read.ptx.sreg.ntid.x()
+// CHECK1-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 
@__kmpc_get_hardware_num_threads_in_block()
 // CHECK1-NEXT:    [[TMP5:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], 
align 8
 // CHECK1-NEXT:    [[TMP6:%.*]] = load i32, i32* [[TMP5]], align 4
 // CHECK1-NEXT:    call void @__kmpc_distribute_static_init_4(%struct.ident_t* 
@[[GLOB2]], i32 [[TMP6]], i32 91, i32* [[DOTOMP_IS_LAST]], i32* 
[[DOTOMP_COMB_LB]], i32* [[DOTOMP_COMB_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 
[[NVPTX_NUM_THREADS]])
@@ -19097,7 +19097,7 @@ int bar(int n){
 // CHECK1-NEXT:    store i32 9, i32* [[DOTOMP_COMB_UB]], align 4
 // CHECK1-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
 // CHECK1-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
-// CHECK1-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 
@llvm.nvvm.read.ptx.sreg.ntid.x()
+// CHECK1-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 
@__kmpc_get_hardware_num_threads_in_block()
 // CHECK1-NEXT:    [[TMP1:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], 
align 8
 // CHECK1-NEXT:    [[TMP2:%.*]] = load i32, i32* [[TMP1]], align 4
 // CHECK1-NEXT:    call void @__kmpc_distribute_static_init_4(%struct.ident_t* 
@[[GLOB2]], i32 [[TMP2]], i32 91, i32* [[DOTOMP_IS_LAST]], i32* 
[[DOTOMP_COMB_LB]], i32* [[DOTOMP_COMB_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 
[[NVPTX_NUM_THREADS]])
@@ -19299,7 +19299,7 @@ int bar(int n){
 // CHECK1-NEXT:    store i32 99, i32* [[DOTOMP_COMB_UB]], align 4
 // CHECK1-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
 // CHECK1-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
-// CHECK1-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 
@llvm.nvvm.read.ptx.sreg.ntid.x()
+// CHECK1-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 
@__kmpc_get_hardware_num_threads_in_block()
 // CHECK1-NEXT:    [[TMP1:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], 
align 8
 // CHECK1-NEXT:    [[TMP2:%.*]] = load i32, i32* [[TMP1]], align 4
 // CHECK1-NEXT:    call void @__kmpc_distribute_static_init_4(%struct.ident_t* 
@[[GLOB2]], i32 [[TMP2]], i32 91, i32* [[DOTOMP_IS_LAST]], i32* 
[[DOTOMP_COMB_LB]], i32* [[DOTOMP_COMB_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 
[[NVPTX_NUM_THREADS]])
@@ -19562,7 +19562,7 @@ int bar(int n){
 // CHECK1-NEXT:    store i64 [[TMP7]], i64* [[DOTOMP_COMB_UB]], align 8
 // CHECK1-NEXT:    store i64 1, i64* [[DOTOMP_STRIDE]], align 8
 // CHECK1-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
-// CHECK1-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 
@llvm.nvvm.read.ptx.sreg.ntid.x()
+// CHECK1-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 
@__kmpc_get_hardware_num_threads_in_block()
 // CHECK1-NEXT:    [[CONV12:%.*]] = zext i32 [[NVPTX_NUM_THREADS]] to i64
 // CHECK1-NEXT:    [[TMP8:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], 
align 8
 // CHECK1-NEXT:    [[TMP9:%.*]] = load i32, i32* [[TMP8]], align 4
@@ -19867,7 +19867,7 @@ int bar(int n){
 // CHECK1-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_COMB_UB]], align 4
 // CHECK1-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
 // CHECK1-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
-// CHECK1-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 
@llvm.nvvm.read.ptx.sreg.ntid.x()
+// CHECK1-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 
@__kmpc_get_hardware_num_threads_in_block()
 // CHECK1-NEXT:    [[TMP5:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], 
align 8
 // CHECK1-NEXT:    [[TMP6:%.*]] = load i32, i32* [[TMP5]], align 4
 // CHECK1-NEXT:    call void @__kmpc_distribute_static_init_4(%struct.ident_t* 
@[[GLOB2]], i32 [[TMP6]], i32 91, i32* [[DOTOMP_IS_LAST]], i32* 
[[DOTOMP_COMB_LB]], i32* [[DOTOMP_COMB_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 
[[NVPTX_NUM_THREADS]])
@@ -20460,7 +20460,7 @@ int bar(int n){
 // CHECK2-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_COMB_UB]], align 4
 // CHECK2-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
 // CHECK2-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
-// CHECK2-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 
@llvm.nvvm.read.ptx.sreg.ntid.x()
+// CHECK2-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 
@__kmpc_get_hardware_num_threads_in_block()
 // CHECK2-NEXT:    [[TMP5:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], 
align 8
 // CHECK2-NEXT:    [[TMP6:%.*]] = load i32, i32* [[TMP5]], align 4
 // CHECK2-NEXT:    call void @__kmpc_distribute_static_init_4(%struct.ident_t* 
@[[GLOB2]], i32 [[TMP6]], i32 91, i32* [[DOTOMP_IS_LAST]], i32* 
[[DOTOMP_COMB_LB]], i32* [[DOTOMP_COMB_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 
[[NVPTX_NUM_THREADS]])
@@ -20691,7 +20691,7 @@ int bar(int n){
 // CHECK2-NEXT:    store i32 9, i32* [[DOTOMP_COMB_UB]], align 4
 // CHECK2-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
 // CHECK2-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
-// CHECK2-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 
@llvm.nvvm.read.ptx.sreg.ntid.x()
+// CHECK2-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 
@__kmpc_get_hardware_num_threads_in_block()
 // CHECK2-NEXT:    [[TMP1:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], 
align 8
 // CHECK2-NEXT:    [[TMP2:%.*]] = load i32, i32* [[TMP1]], align 4
 // CHECK2-NEXT:    call void @__kmpc_distribute_static_init_4(%struct.ident_t* 
@[[GLOB2]], i32 [[TMP2]], i32 91, i32* [[DOTOMP_IS_LAST]], i32* 
[[DOTOMP_COMB_LB]], i32* [[DOTOMP_COMB_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 
[[NVPTX_NUM_THREADS]])
@@ -20893,7 +20893,7 @@ int bar(int n){
 // CHECK2-NEXT:    store i32 99, i32* [[DOTOMP_COMB_UB]], align 4
 // CHECK2-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
 // CHECK2-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
-// CHECK2-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 
@llvm.nvvm.read.ptx.sreg.ntid.x()
+// CHECK2-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 
@__kmpc_get_hardware_num_threads_in_block()
 // CHECK2-NEXT:    [[TMP1:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], 
align 8
 // CHECK2-NEXT:    [[TMP2:%.*]] = load i32, i32* [[TMP1]], align 4
 // CHECK2-NEXT:    call void @__kmpc_distribute_static_init_4(%struct.ident_t* 
@[[GLOB2]], i32 [[TMP2]], i32 91, i32* [[DOTOMP_IS_LAST]], i32* 
[[DOTOMP_COMB_LB]], i32* [[DOTOMP_COMB_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 
[[NVPTX_NUM_THREADS]])
@@ -21154,7 +21154,7 @@ int bar(int n){
 // CHECK2-NEXT:    store i32 [[TMP7]], i32* [[DOTOMP_COMB_UB]], align 4
 // CHECK2-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
 // CHECK2-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
-// CHECK2-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 
@llvm.nvvm.read.ptx.sreg.ntid.x()
+// CHECK2-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 
@__kmpc_get_hardware_num_threads_in_block()
 // CHECK2-NEXT:    [[TMP8:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], 
align 8
 // CHECK2-NEXT:    [[TMP9:%.*]] = load i32, i32* [[TMP8]], align 4
 // CHECK2-NEXT:    call void @__kmpc_distribute_static_init_4(%struct.ident_t* 
@[[GLOB2]], i32 [[TMP9]], i32 91, i32* [[DOTOMP_IS_LAST]], i32* 
[[DOTOMP_COMB_LB]], i32* [[DOTOMP_COMB_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 
[[NVPTX_NUM_THREADS]])
@@ -21456,7 +21456,7 @@ int bar(int n){
 // CHECK2-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_COMB_UB]], align 4
 // CHECK2-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
 // CHECK2-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
-// CHECK2-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 
@llvm.nvvm.read.ptx.sreg.ntid.x()
+// CHECK2-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 
@__kmpc_get_hardware_num_threads_in_block()
 // CHECK2-NEXT:    [[TMP5:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], 
align 8
 // CHECK2-NEXT:    [[TMP6:%.*]] = load i32, i32* [[TMP5]], align 4
 // CHECK2-NEXT:    call void @__kmpc_distribute_static_init_4(%struct.ident_t* 
@[[GLOB2]], i32 [[TMP6]], i32 91, i32* [[DOTOMP_IS_LAST]], i32* 
[[DOTOMP_COMB_LB]], i32* [[DOTOMP_COMB_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 
[[NVPTX_NUM_THREADS]])
@@ -22029,7 +22029,7 @@ int bar(int n){
 // CHECK3-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_COMB_UB]], align 4
 // CHECK3-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
 // CHECK3-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
-// CHECK3-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 
@llvm.nvvm.read.ptx.sreg.ntid.x()
+// CHECK3-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 
@__kmpc_get_hardware_num_threads_in_block()
 // CHECK3-NEXT:    [[TMP5:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], 
align 4
 // CHECK3-NEXT:    [[TMP6:%.*]] = load i32, i32* [[TMP5]], align 4
 // CHECK3-NEXT:    call void @__kmpc_distribute_static_init_4(%struct.ident_t* 
@[[GLOB2]], i32 [[TMP6]], i32 91, i32* [[DOTOMP_IS_LAST]], i32* 
[[DOTOMP_COMB_LB]], i32* [[DOTOMP_COMB_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 
[[NVPTX_NUM_THREADS]])
@@ -22252,7 +22252,7 @@ int bar(int n){
 // CHECK3-NEXT:    store i32 9, i32* [[DOTOMP_COMB_UB]], align 4
 // CHECK3-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
 // CHECK3-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
-// CHECK3-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 
@llvm.nvvm.read.ptx.sreg.ntid.x()
+// CHECK3-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 
@__kmpc_get_hardware_num_threads_in_block()
 // CHECK3-NEXT:    [[TMP1:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], 
align 4
 // CHECK3-NEXT:    [[TMP2:%.*]] = load i32, i32* [[TMP1]], align 4
 // CHECK3-NEXT:    call void @__kmpc_distribute_static_init_4(%struct.ident_t* 
@[[GLOB2]], i32 [[TMP2]], i32 91, i32* [[DOTOMP_IS_LAST]], i32* 
[[DOTOMP_COMB_LB]], i32* [[DOTOMP_COMB_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 
[[NVPTX_NUM_THREADS]])
@@ -22445,7 +22445,7 @@ int bar(int n){
 // CHECK3-NEXT:    store i32 99, i32* [[DOTOMP_COMB_UB]], align 4
 // CHECK3-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
 // CHECK3-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
-// CHECK3-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 
@llvm.nvvm.read.ptx.sreg.ntid.x()
+// CHECK3-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 
@__kmpc_get_hardware_num_threads_in_block()
 // CHECK3-NEXT:    [[TMP1:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], 
align 4
 // CHECK3-NEXT:    [[TMP2:%.*]] = load i32, i32* [[TMP1]], align 4
 // CHECK3-NEXT:    call void @__kmpc_distribute_static_init_4(%struct.ident_t* 
@[[GLOB2]], i32 [[TMP2]], i32 91, i32* [[DOTOMP_IS_LAST]], i32* 
[[DOTOMP_COMB_LB]], i32* [[DOTOMP_COMB_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 
[[NVPTX_NUM_THREADS]])
@@ -22696,7 +22696,7 @@ int bar(int n){
 // CHECK3-NEXT:    store i64 [[TMP7]], i64* [[DOTOMP_COMB_UB]], align 8
 // CHECK3-NEXT:    store i64 1, i64* [[DOTOMP_STRIDE]], align 8
 // CHECK3-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
-// CHECK3-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 
@llvm.nvvm.read.ptx.sreg.ntid.x()
+// CHECK3-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 
@__kmpc_get_hardware_num_threads_in_block()
 // CHECK3-NEXT:    [[CONV11:%.*]] = zext i32 [[NVPTX_NUM_THREADS]] to i64
 // CHECK3-NEXT:    [[TMP8:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], 
align 4
 // CHECK3-NEXT:    [[TMP9:%.*]] = load i32, i32* [[TMP8]], align 4
@@ -22999,7 +22999,7 @@ int bar(int n){
 // CHECK3-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_COMB_UB]], align 4
 // CHECK3-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
 // CHECK3-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
-// CHECK3-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 
@llvm.nvvm.read.ptx.sreg.ntid.x()
+// CHECK3-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 
@__kmpc_get_hardware_num_threads_in_block()
 // CHECK3-NEXT:    [[TMP5:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], 
align 4
 // CHECK3-NEXT:    [[TMP6:%.*]] = load i32, i32* [[TMP5]], align 4
 // CHECK3-NEXT:    call void @__kmpc_distribute_static_init_4(%struct.ident_t* 
@[[GLOB2]], i32 [[TMP6]], i32 91, i32* [[DOTOMP_IS_LAST]], i32* 
[[DOTOMP_COMB_LB]], i32* [[DOTOMP_COMB_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 
[[NVPTX_NUM_THREADS]])
@@ -23563,7 +23563,7 @@ int bar(int n){
 // CHECK4-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_COMB_UB]], align 4
 // CHECK4-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
 // CHECK4-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
-// CHECK4-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 
@llvm.nvvm.read.ptx.sreg.ntid.x()
+// CHECK4-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 
@__kmpc_get_hardware_num_threads_in_block()
 // CHECK4-NEXT:    [[TMP5:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], 
align 4
 // CHECK4-NEXT:    [[TMP6:%.*]] = load i32, i32* [[TMP5]], align 4
 // CHECK4-NEXT:    call void @__kmpc_distribute_static_init_4(%struct.ident_t* 
@[[GLOB2]], i32 [[TMP6]], i32 91, i32* [[DOTOMP_IS_LAST]], i32* 
[[DOTOMP_COMB_LB]], i32* [[DOTOMP_COMB_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 
[[NVPTX_NUM_THREADS]])
@@ -23786,7 +23786,7 @@ int bar(int n){
 // CHECK4-NEXT:    store i32 9, i32* [[DOTOMP_COMB_UB]], align 4
 // CHECK4-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
 // CHECK4-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
-// CHECK4-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 
@llvm.nvvm.read.ptx.sreg.ntid.x()
+// CHECK4-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 
@__kmpc_get_hardware_num_threads_in_block()
 // CHECK4-NEXT:    [[TMP1:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], 
align 4
 // CHECK4-NEXT:    [[TMP2:%.*]] = load i32, i32* [[TMP1]], align 4
 // CHECK4-NEXT:    call void @__kmpc_distribute_static_init_4(%struct.ident_t* 
@[[GLOB2]], i32 [[TMP2]], i32 91, i32* [[DOTOMP_IS_LAST]], i32* 
[[DOTOMP_COMB_LB]], i32* [[DOTOMP_COMB_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 
[[NVPTX_NUM_THREADS]])
@@ -23979,7 +23979,7 @@ int bar(int n){
 // CHECK4-NEXT:    store i32 99, i32* [[DOTOMP_COMB_UB]], align 4
 // CHECK4-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
 // CHECK4-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
-// CHECK4-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 
@llvm.nvvm.read.ptx.sreg.ntid.x()
+// CHECK4-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 
@__kmpc_get_hardware_num_threads_in_block()
 // CHECK4-NEXT:    [[TMP1:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], 
align 4
 // CHECK4-NEXT:    [[TMP2:%.*]] = load i32, i32* [[TMP1]], align 4
 // CHECK4-NEXT:    call void @__kmpc_distribute_static_init_4(%struct.ident_t* 
@[[GLOB2]], i32 [[TMP2]], i32 91, i32* [[DOTOMP_IS_LAST]], i32* 
[[DOTOMP_COMB_LB]], i32* [[DOTOMP_COMB_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 
[[NVPTX_NUM_THREADS]])
@@ -24230,7 +24230,7 @@ int bar(int n){
 // CHECK4-NEXT:    store i64 [[TMP7]], i64* [[DOTOMP_COMB_UB]], align 8
 // CHECK4-NEXT:    store i64 1, i64* [[DOTOMP_STRIDE]], align 8
 // CHECK4-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
-// CHECK4-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 
@llvm.nvvm.read.ptx.sreg.ntid.x()
+// CHECK4-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 
@__kmpc_get_hardware_num_threads_in_block()
 // CHECK4-NEXT:    [[CONV11:%.*]] = zext i32 [[NVPTX_NUM_THREADS]] to i64
 // CHECK4-NEXT:    [[TMP8:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], 
align 4
 // CHECK4-NEXT:    [[TMP9:%.*]] = load i32, i32* [[TMP8]], align 4
@@ -24533,7 +24533,7 @@ int bar(int n){
 // CHECK4-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_COMB_UB]], align 4
 // CHECK4-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
 // CHECK4-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
-// CHECK4-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 
@llvm.nvvm.read.ptx.sreg.ntid.x()
+// CHECK4-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 
@__kmpc_get_hardware_num_threads_in_block()
 // CHECK4-NEXT:    [[TMP5:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], 
align 4
 // CHECK4-NEXT:    [[TMP6:%.*]] = load i32, i32* [[TMP5]], align 4
 // CHECK4-NEXT:    call void @__kmpc_distribute_static_init_4(%struct.ident_t* 
@[[GLOB2]], i32 [[TMP6]], i32 91, i32* [[DOTOMP_IS_LAST]], i32* 
[[DOTOMP_COMB_LB]], i32* [[DOTOMP_COMB_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 
[[NVPTX_NUM_THREADS]])

diff  --git 
a/clang/test/OpenMP/nvptx_target_teams_distribute_parallel_for_generic_mode_codegen.cpp
 
b/clang/test/OpenMP/nvptx_target_teams_distribute_parallel_for_generic_mode_codegen.cpp
index ea13b9312cb1c..261b277cdc497 100644
--- 
a/clang/test/OpenMP/nvptx_target_teams_distribute_parallel_for_generic_mode_codegen.cpp
+++ 
b/clang/test/OpenMP/nvptx_target_teams_distribute_parallel_for_generic_mode_codegen.cpp
@@ -60,7 +60,7 @@ int main(int argc, char **argv) {
 // CHECK1-NEXT:    store i32 [[TMP5]], i32* [[CONV3]], align 4
 // CHECK1-NEXT:    [[TMP6:%.*]] = load i64, i64* [[DOTCAPTURE_EXPR__CASTED]], 
align 8
 // CHECK1-NEXT:    store i32 [[TMP2]], i32* [[DOTTHREADID_TEMP_]], align 4
-// CHECK1-NEXT:    call void @__omp_outlined__(i32* [[DOTTHREADID_TEMP_]], 
i32* [[DOTZERO_ADDR]], i64 [[TMP4]], i32* [[TMP0]], i64 [[TMP6]]) 
#[[ATTR3:[0-9]+]]
+// CHECK1-NEXT:    call void @__omp_outlined__(i32* [[DOTTHREADID_TEMP_]], 
i32* [[DOTZERO_ADDR]], i64 [[TMP4]], i32* [[TMP0]], i64 [[TMP6]]) 
#[[ATTR2:[0-9]+]]
 // CHECK1-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* 
@[[GLOB1]], i8 2, i1 false)
 // CHECK1-NEXT:    ret void
 // CHECK1:       worker.exit:
@@ -113,7 +113,7 @@ int main(int argc, char **argv) {
 // CHECK1-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_COMB_UB]], align 4
 // CHECK1-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
 // CHECK1-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
-// CHECK1-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 
@llvm.nvvm.read.ptx.sreg.ntid.x()
+// CHECK1-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 
@__kmpc_get_hardware_num_threads_in_block()
 // CHECK1-NEXT:    [[TMP5:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], 
align 8
 // CHECK1-NEXT:    [[TMP6:%.*]] = load i32, i32* [[TMP5]], align 4
 // CHECK1-NEXT:    call void @__kmpc_distribute_static_init_4(%struct.ident_t* 
@[[GLOB2:[0-9]+]], i32 [[TMP6]], i32 91, i32* [[DOTOMP_IS_LAST]], i32* 
[[DOTOMP_COMB_LB]], i32* [[DOTOMP_COMB_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 
[[NVPTX_NUM_THREADS]])
@@ -304,10 +304,10 @@ int main(int argc, char **argv) {
 // CHECK1-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP19]], 1
 // CHECK1-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
 // CHECK1-NEXT:    store i32 [[ADD]], i32* [[I7]], align 4
-// CHECK1-NEXT:    [[CALL:%.*]] = call i32 @_Z3fooPi(i32* [[I7]]) 
#[[ATTR5:[0-9]+]]
-// CHECK1-NEXT:    [[CALL13:%.*]] = call i32 @_Z3fooPi(i32* [[TMP0]]) 
#[[ATTR5]]
+// CHECK1-NEXT:    [[CALL:%.*]] = call i32 @_Z3fooPi(i32* [[I7]]) 
#[[ATTR4:[0-9]+]]
+// CHECK1-NEXT:    [[CALL13:%.*]] = call i32 @_Z3fooPi(i32* [[TMP0]]) 
#[[ATTR4]]
 // CHECK1-NEXT:    [[ADD14:%.*]] = add nsw i32 [[CALL]], [[CALL13]]
-// CHECK1-NEXT:    [[CALL15:%.*]] = call i32 @_Z3fooPi(i32* [[CONV]]) 
#[[ATTR5]]
+// CHECK1-NEXT:    [[CALL15:%.*]] = call i32 @_Z3fooPi(i32* [[CONV]]) 
#[[ATTR4]]
 // CHECK1-NEXT:    [[ADD16:%.*]] = add nsw i32 [[ADD14]], [[CALL15]]
 // CHECK1-NEXT:    store i32 [[ADD16]], i32* [[TMP0]], align 4
 // CHECK1-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
@@ -366,7 +366,7 @@ int main(int argc, char **argv) {
 // CHECK2-NEXT:    store i32 [[TMP5]], i32* [[DOTCAPTURE_EXPR__CASTED]], align 
4
 // CHECK2-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR__CASTED]], 
align 4
 // CHECK2-NEXT:    store i32 [[TMP2]], i32* [[DOTTHREADID_TEMP_]], align 4
-// CHECK2-NEXT:    call void @__omp_outlined__(i32* [[DOTTHREADID_TEMP_]], 
i32* [[DOTZERO_ADDR]], i32 [[TMP4]], i32* [[TMP0]], i32 [[TMP6]]) 
#[[ATTR3:[0-9]+]]
+// CHECK2-NEXT:    call void @__omp_outlined__(i32* [[DOTTHREADID_TEMP_]], 
i32* [[DOTZERO_ADDR]], i32 [[TMP4]], i32* [[TMP0]], i32 [[TMP6]]) 
#[[ATTR2:[0-9]+]]
 // CHECK2-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* 
@[[GLOB1]], i8 2, i1 false)
 // CHECK2-NEXT:    ret void
 // CHECK2:       worker.exit:
@@ -417,7 +417,7 @@ int main(int argc, char **argv) {
 // CHECK2-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_COMB_UB]], align 4
 // CHECK2-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
 // CHECK2-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
-// CHECK2-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 
@llvm.nvvm.read.ptx.sreg.ntid.x()
+// CHECK2-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 
@__kmpc_get_hardware_num_threads_in_block()
 // CHECK2-NEXT:    [[TMP5:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], 
align 4
 // CHECK2-NEXT:    [[TMP6:%.*]] = load i32, i32* [[TMP5]], align 4
 // CHECK2-NEXT:    call void @__kmpc_distribute_static_init_4(%struct.ident_t* 
@[[GLOB2:[0-9]+]], i32 [[TMP6]], i32 91, i32* [[DOTOMP_IS_LAST]], i32* 
[[DOTOMP_COMB_LB]], i32* [[DOTOMP_COMB_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 
[[NVPTX_NUM_THREADS]])
@@ -598,10 +598,10 @@ int main(int argc, char **argv) {
 // CHECK2-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP19]], 1
 // CHECK2-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
 // CHECK2-NEXT:    store i32 [[ADD]], i32* [[I4]], align 4
-// CHECK2-NEXT:    [[CALL:%.*]] = call i32 @_Z3fooPi(i32* [[I4]]) 
#[[ATTR5:[0-9]+]]
-// CHECK2-NEXT:    [[CALL8:%.*]] = call i32 @_Z3fooPi(i32* [[TMP0]]) #[[ATTR5]]
+// CHECK2-NEXT:    [[CALL:%.*]] = call i32 @_Z3fooPi(i32* [[I4]]) 
#[[ATTR4:[0-9]+]]
+// CHECK2-NEXT:    [[CALL8:%.*]] = call i32 @_Z3fooPi(i32* [[TMP0]]) #[[ATTR4]]
 // CHECK2-NEXT:    [[ADD9:%.*]] = add nsw i32 [[CALL]], [[CALL8]]
-// CHECK2-NEXT:    [[CALL10:%.*]] = call i32 @_Z3fooPi(i32* [[ARGC_ADDR]]) 
#[[ATTR5]]
+// CHECK2-NEXT:    [[CALL10:%.*]] = call i32 @_Z3fooPi(i32* [[ARGC_ADDR]]) 
#[[ATTR4]]
 // CHECK2-NEXT:    [[ADD11:%.*]] = add nsw i32 [[ADD9]], [[CALL10]]
 // CHECK2-NEXT:    store i32 [[ADD11]], i32* [[TMP0]], align 4
 // CHECK2-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
@@ -660,7 +660,7 @@ int main(int argc, char **argv) {
 // CHECK3-NEXT:    store i32 [[TMP5]], i32* [[DOTCAPTURE_EXPR__CASTED]], align 
4
 // CHECK3-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR__CASTED]], 
align 4
 // CHECK3-NEXT:    store i32 [[TMP2]], i32* [[DOTTHREADID_TEMP_]], align 4
-// CHECK3-NEXT:    call void @__omp_outlined__(i32* [[DOTTHREADID_TEMP_]], 
i32* [[DOTZERO_ADDR]], i32 [[TMP4]], i32* [[TMP0]], i32 [[TMP6]]) 
#[[ATTR3:[0-9]+]]
+// CHECK3-NEXT:    call void @__omp_outlined__(i32* [[DOTTHREADID_TEMP_]], 
i32* [[DOTZERO_ADDR]], i32 [[TMP4]], i32* [[TMP0]], i32 [[TMP6]]) 
#[[ATTR2:[0-9]+]]
 // CHECK3-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* 
@[[GLOB1]], i8 2, i1 false)
 // CHECK3-NEXT:    ret void
 // CHECK3:       worker.exit:
@@ -711,7 +711,7 @@ int main(int argc, char **argv) {
 // CHECK3-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_COMB_UB]], align 4
 // CHECK3-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
 // CHECK3-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
-// CHECK3-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 
@llvm.nvvm.read.ptx.sreg.ntid.x()
+// CHECK3-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 
@__kmpc_get_hardware_num_threads_in_block()
 // CHECK3-NEXT:    [[TMP5:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], 
align 4
 // CHECK3-NEXT:    [[TMP6:%.*]] = load i32, i32* [[TMP5]], align 4
 // CHECK3-NEXT:    call void @__kmpc_distribute_static_init_4(%struct.ident_t* 
@[[GLOB2:[0-9]+]], i32 [[TMP6]], i32 91, i32* [[DOTOMP_IS_LAST]], i32* 
[[DOTOMP_COMB_LB]], i32* [[DOTOMP_COMB_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 
[[NVPTX_NUM_THREADS]])
@@ -892,10 +892,10 @@ int main(int argc, char **argv) {
 // CHECK3-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP19]], 1
 // CHECK3-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
 // CHECK3-NEXT:    store i32 [[ADD]], i32* [[I4]], align 4
-// CHECK3-NEXT:    [[CALL:%.*]] = call i32 @_Z3fooPi(i32* [[I4]]) 
#[[ATTR5:[0-9]+]]
-// CHECK3-NEXT:    [[CALL8:%.*]] = call i32 @_Z3fooPi(i32* [[TMP0]]) #[[ATTR5]]
+// CHECK3-NEXT:    [[CALL:%.*]] = call i32 @_Z3fooPi(i32* [[I4]]) 
#[[ATTR4:[0-9]+]]
+// CHECK3-NEXT:    [[CALL8:%.*]] = call i32 @_Z3fooPi(i32* [[TMP0]]) #[[ATTR4]]
 // CHECK3-NEXT:    [[ADD9:%.*]] = add nsw i32 [[CALL]], [[CALL8]]
-// CHECK3-NEXT:    [[CALL10:%.*]] = call i32 @_Z3fooPi(i32* [[ARGC_ADDR]]) 
#[[ATTR5]]
+// CHECK3-NEXT:    [[CALL10:%.*]] = call i32 @_Z3fooPi(i32* [[ARGC_ADDR]]) 
#[[ATTR4]]
 // CHECK3-NEXT:    [[ADD11:%.*]] = add nsw i32 [[ADD9]], [[CALL10]]
 // CHECK3-NEXT:    store i32 [[ADD11]], i32* [[TMP0]], align 4
 // CHECK3-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
@@ -958,7 +958,7 @@ int main(int argc, char **argv) {
 // CHECK4-NEXT:    store i32 [[TMP5]], i32* [[CONV3]], align 4
 // CHECK4-NEXT:    [[TMP6:%.*]] = load i64, i64* [[DOTCAPTURE_EXPR__CASTED]], 
align 8
 // CHECK4-NEXT:    store i32 [[TMP2]], i32* [[DOTTHREADID_TEMP_]], align 4
-// CHECK4-NEXT:    call void @__omp_outlined__(i32* [[DOTTHREADID_TEMP_]], 
i32* [[DOTZERO_ADDR]], i64 [[TMP4]], i32* [[TMP0]], i64 [[TMP6]]) 
#[[ATTR3:[0-9]+]]
+// CHECK4-NEXT:    call void @__omp_outlined__(i32* [[DOTTHREADID_TEMP_]], 
i32* [[DOTZERO_ADDR]], i64 [[TMP4]], i32* [[TMP0]], i64 [[TMP6]]) 
#[[ATTR2:[0-9]+]]
 // CHECK4-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* 
@[[GLOB1]], i8 2, i1 false)
 // CHECK4-NEXT:    ret void
 // CHECK4:       worker.exit:
@@ -1011,7 +1011,7 @@ int main(int argc, char **argv) {
 // CHECK4-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_COMB_UB]], align 4
 // CHECK4-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
 // CHECK4-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
-// CHECK4-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 
@llvm.nvvm.read.ptx.sreg.ntid.x()
+// CHECK4-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 
@__kmpc_get_hardware_num_threads_in_block()
 // CHECK4-NEXT:    [[TMP5:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], 
align 8
 // CHECK4-NEXT:    [[TMP6:%.*]] = load i32, i32* [[TMP5]], align 4
 // CHECK4-NEXT:    call void @__kmpc_distribute_static_init_4(%struct.ident_t* 
@[[GLOB2:[0-9]+]], i32 [[TMP6]], i32 91, i32* [[DOTOMP_IS_LAST]], i32* 
[[DOTOMP_COMB_LB]], i32* [[DOTOMP_COMB_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 
[[NVPTX_NUM_THREADS]])
@@ -1202,10 +1202,10 @@ int main(int argc, char **argv) {
 // CHECK4-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP19]], 1
 // CHECK4-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
 // CHECK4-NEXT:    store i32 [[ADD]], i32* [[I7]], align 4
-// CHECK4-NEXT:    [[CALL:%.*]] = call i32 @_Z3fooPi(i32* [[I7]]) 
#[[ATTR5:[0-9]+]]
-// CHECK4-NEXT:    [[CALL13:%.*]] = call i32 @_Z3fooPi(i32* [[TMP0]]) 
#[[ATTR5]]
+// CHECK4-NEXT:    [[CALL:%.*]] = call i32 @_Z3fooPi(i32* [[I7]]) 
#[[ATTR4:[0-9]+]]
+// CHECK4-NEXT:    [[CALL13:%.*]] = call i32 @_Z3fooPi(i32* [[TMP0]]) 
#[[ATTR4]]
 // CHECK4-NEXT:    [[ADD14:%.*]] = add nsw i32 [[CALL]], [[CALL13]]
-// CHECK4-NEXT:    [[CALL15:%.*]] = call i32 @_Z3fooPi(i32* [[CONV]]) 
#[[ATTR5]]
+// CHECK4-NEXT:    [[CALL15:%.*]] = call i32 @_Z3fooPi(i32* [[CONV]]) 
#[[ATTR4]]
 // CHECK4-NEXT:    [[ADD16:%.*]] = add nsw i32 [[ADD14]], [[CALL15]]
 // CHECK4-NEXT:    store i32 [[ADD16]], i32* [[TMP0]], align 4
 // CHECK4-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
@@ -1264,7 +1264,7 @@ int main(int argc, char **argv) {
 // CHECK5-NEXT:    store i32 [[TMP5]], i32* [[DOTCAPTURE_EXPR__CASTED]], align 
4
 // CHECK5-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR__CASTED]], 
align 4
 // CHECK5-NEXT:    store i32 [[TMP2]], i32* [[DOTTHREADID_TEMP_]], align 4
-// CHECK5-NEXT:    call void @__omp_outlined__(i32* [[DOTTHREADID_TEMP_]], 
i32* [[DOTZERO_ADDR]], i32 [[TMP4]], i32* [[TMP0]], i32 [[TMP6]]) 
#[[ATTR3:[0-9]+]]
+// CHECK5-NEXT:    call void @__omp_outlined__(i32* [[DOTTHREADID_TEMP_]], 
i32* [[DOTZERO_ADDR]], i32 [[TMP4]], i32* [[TMP0]], i32 [[TMP6]]) 
#[[ATTR2:[0-9]+]]
 // CHECK5-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* 
@[[GLOB1]], i8 2, i1 false)
 // CHECK5-NEXT:    ret void
 // CHECK5:       worker.exit:
@@ -1315,7 +1315,7 @@ int main(int argc, char **argv) {
 // CHECK5-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_COMB_UB]], align 4
 // CHECK5-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
 // CHECK5-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
-// CHECK5-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 
@llvm.nvvm.read.ptx.sreg.ntid.x()
+// CHECK5-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 
@__kmpc_get_hardware_num_threads_in_block()
 // CHECK5-NEXT:    [[TMP5:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], 
align 4
 // CHECK5-NEXT:    [[TMP6:%.*]] = load i32, i32* [[TMP5]], align 4
 // CHECK5-NEXT:    call void @__kmpc_distribute_static_init_4(%struct.ident_t* 
@[[GLOB2:[0-9]+]], i32 [[TMP6]], i32 91, i32* [[DOTOMP_IS_LAST]], i32* 
[[DOTOMP_COMB_LB]], i32* [[DOTOMP_COMB_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 
[[NVPTX_NUM_THREADS]])
@@ -1496,10 +1496,10 @@ int main(int argc, char **argv) {
 // CHECK5-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP19]], 1
 // CHECK5-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
 // CHECK5-NEXT:    store i32 [[ADD]], i32* [[I4]], align 4
-// CHECK5-NEXT:    [[CALL:%.*]] = call i32 @_Z3fooPi(i32* [[I4]]) 
#[[ATTR5:[0-9]+]]
-// CHECK5-NEXT:    [[CALL8:%.*]] = call i32 @_Z3fooPi(i32* [[TMP0]]) #[[ATTR5]]
+// CHECK5-NEXT:    [[CALL:%.*]] = call i32 @_Z3fooPi(i32* [[I4]]) 
#[[ATTR4:[0-9]+]]
+// CHECK5-NEXT:    [[CALL8:%.*]] = call i32 @_Z3fooPi(i32* [[TMP0]]) #[[ATTR4]]
 // CHECK5-NEXT:    [[ADD9:%.*]] = add nsw i32 [[CALL]], [[CALL8]]
-// CHECK5-NEXT:    [[CALL10:%.*]] = call i32 @_Z3fooPi(i32* [[ARGC_ADDR]]) 
#[[ATTR5]]
+// CHECK5-NEXT:    [[CALL10:%.*]] = call i32 @_Z3fooPi(i32* [[ARGC_ADDR]]) 
#[[ATTR4]]
 // CHECK5-NEXT:    [[ADD11:%.*]] = add nsw i32 [[ADD9]], [[CALL10]]
 // CHECK5-NEXT:    store i32 [[ADD11]], i32* [[TMP0]], align 4
 // CHECK5-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
@@ -1558,7 +1558,7 @@ int main(int argc, char **argv) {
 // CHECK6-NEXT:    store i32 [[TMP5]], i32* [[DOTCAPTURE_EXPR__CASTED]], align 
4
 // CHECK6-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR__CASTED]], 
align 4
 // CHECK6-NEXT:    store i32 [[TMP2]], i32* [[DOTTHREADID_TEMP_]], align 4
-// CHECK6-NEXT:    call void @__omp_outlined__(i32* [[DOTTHREADID_TEMP_]], 
i32* [[DOTZERO_ADDR]], i32 [[TMP4]], i32* [[TMP0]], i32 [[TMP6]]) 
#[[ATTR3:[0-9]+]]
+// CHECK6-NEXT:    call void @__omp_outlined__(i32* [[DOTTHREADID_TEMP_]], 
i32* [[DOTZERO_ADDR]], i32 [[TMP4]], i32* [[TMP0]], i32 [[TMP6]]) 
#[[ATTR2:[0-9]+]]
 // CHECK6-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* 
@[[GLOB1]], i8 2, i1 false)
 // CHECK6-NEXT:    ret void
 // CHECK6:       worker.exit:
@@ -1609,7 +1609,7 @@ int main(int argc, char **argv) {
 // CHECK6-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_COMB_UB]], align 4
 // CHECK6-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
 // CHECK6-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
-// CHECK6-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 
@llvm.nvvm.read.ptx.sreg.ntid.x()
+// CHECK6-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 
@__kmpc_get_hardware_num_threads_in_block()
 // CHECK6-NEXT:    [[TMP5:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], 
align 4
 // CHECK6-NEXT:    [[TMP6:%.*]] = load i32, i32* [[TMP5]], align 4
 // CHECK6-NEXT:    call void @__kmpc_distribute_static_init_4(%struct.ident_t* 
@[[GLOB2:[0-9]+]], i32 [[TMP6]], i32 91, i32* [[DOTOMP_IS_LAST]], i32* 
[[DOTOMP_COMB_LB]], i32* [[DOTOMP_COMB_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 
[[NVPTX_NUM_THREADS]])
@@ -1790,10 +1790,10 @@ int main(int argc, char **argv) {
 // CHECK6-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP19]], 1
 // CHECK6-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
 // CHECK6-NEXT:    store i32 [[ADD]], i32* [[I4]], align 4
-// CHECK6-NEXT:    [[CALL:%.*]] = call i32 @_Z3fooPi(i32* [[I4]]) 
#[[ATTR5:[0-9]+]]
-// CHECK6-NEXT:    [[CALL8:%.*]] = call i32 @_Z3fooPi(i32* [[TMP0]]) #[[ATTR5]]
+// CHECK6-NEXT:    [[CALL:%.*]] = call i32 @_Z3fooPi(i32* [[I4]]) 
#[[ATTR4:[0-9]+]]
+// CHECK6-NEXT:    [[CALL8:%.*]] = call i32 @_Z3fooPi(i32* [[TMP0]]) #[[ATTR4]]
 // CHECK6-NEXT:    [[ADD9:%.*]] = add nsw i32 [[CALL]], [[CALL8]]
-// CHECK6-NEXT:    [[CALL10:%.*]] = call i32 @_Z3fooPi(i32* [[ARGC_ADDR]]) 
#[[ATTR5]]
+// CHECK6-NEXT:    [[CALL10:%.*]] = call i32 @_Z3fooPi(i32* [[ARGC_ADDR]]) 
#[[ATTR4]]
 // CHECK6-NEXT:    [[ADD11:%.*]] = add nsw i32 [[ADD9]], [[CALL10]]
 // CHECK6-NEXT:    store i32 [[ADD11]], i32* [[TMP0]], align 4
 // CHECK6-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]

diff  --git 
a/clang/test/OpenMP/nvptx_target_teams_distribute_parallel_for_simd_codegen.cpp 
b/clang/test/OpenMP/nvptx_target_teams_distribute_parallel_for_simd_codegen.cpp
index 14172267626b8..89b808d037686 100644
--- 
a/clang/test/OpenMP/nvptx_target_teams_distribute_parallel_for_simd_codegen.cpp
+++ 
b/clang/test/OpenMP/nvptx_target_teams_distribute_parallel_for_simd_codegen.cpp
@@ -9803,7 +9803,7 @@ int bar(int n){
 // CHECK1-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_COMB_UB]], align 4
 // CHECK1-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
 // CHECK1-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
-// CHECK1-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 
@llvm.nvvm.read.ptx.sreg.ntid.x()
+// CHECK1-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 
@__kmpc_get_hardware_num_threads_in_block()
 // CHECK1-NEXT:    [[TMP5:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], 
align 8
 // CHECK1-NEXT:    [[TMP6:%.*]] = load i32, i32* [[TMP5]], align 4
 // CHECK1-NEXT:    call void @__kmpc_distribute_static_init_4(%struct.ident_t* 
@[[GLOB2]], i32 [[TMP6]], i32 91, i32* [[DOTOMP_IS_LAST]], i32* 
[[DOTOMP_COMB_LB]], i32* [[DOTOMP_COMB_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 
[[NVPTX_NUM_THREADS]])
@@ -10058,7 +10058,7 @@ int bar(int n){
 // CHECK1-NEXT:    store i32 9, i32* [[DOTOMP_COMB_UB]], align 4
 // CHECK1-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
 // CHECK1-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
-// CHECK1-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 
@llvm.nvvm.read.ptx.sreg.ntid.x()
+// CHECK1-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 
@__kmpc_get_hardware_num_threads_in_block()
 // CHECK1-NEXT:    [[TMP1:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], 
align 8
 // CHECK1-NEXT:    [[TMP2:%.*]] = load i32, i32* [[TMP1]], align 4
 // CHECK1-NEXT:    call void @__kmpc_distribute_static_init_4(%struct.ident_t* 
@[[GLOB2]], i32 [[TMP2]], i32 91, i32* [[DOTOMP_IS_LAST]], i32* 
[[DOTOMP_COMB_LB]], i32* [[DOTOMP_COMB_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 
[[NVPTX_NUM_THREADS]])
@@ -10274,7 +10274,7 @@ int bar(int n){
 // CHECK1-NEXT:    store i32 99, i32* [[DOTOMP_COMB_UB]], align 4
 // CHECK1-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
 // CHECK1-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
-// CHECK1-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 
@llvm.nvvm.read.ptx.sreg.ntid.x()
+// CHECK1-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 
@__kmpc_get_hardware_num_threads_in_block()
 // CHECK1-NEXT:    [[TMP1:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], 
align 8
 // CHECK1-NEXT:    [[TMP2:%.*]] = load i32, i32* [[TMP1]], align 4
 // CHECK1-NEXT:    call void @__kmpc_distribute_static_init_4(%struct.ident_t* 
@[[GLOB2]], i32 [[TMP2]], i32 91, i32* [[DOTOMP_IS_LAST]], i32* 
[[DOTOMP_COMB_LB]], i32* [[DOTOMP_COMB_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 
[[NVPTX_NUM_THREADS]])
@@ -10867,7 +10867,7 @@ int bar(int n){
 // CHECK2-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_COMB_UB]], align 4
 // CHECK2-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
 // CHECK2-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
-// CHECK2-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 
@llvm.nvvm.read.ptx.sreg.ntid.x()
+// CHECK2-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 
@__kmpc_get_hardware_num_threads_in_block()
 // CHECK2-NEXT:    [[TMP5:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], 
align 4
 // CHECK2-NEXT:    [[TMP6:%.*]] = load i32, i32* [[TMP5]], align 4
 // CHECK2-NEXT:    call void @__kmpc_distribute_static_init_4(%struct.ident_t* 
@[[GLOB2]], i32 [[TMP6]], i32 91, i32* [[DOTOMP_IS_LAST]], i32* 
[[DOTOMP_COMB_LB]], i32* [[DOTOMP_COMB_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 
[[NVPTX_NUM_THREADS]])
@@ -11114,7 +11114,7 @@ int bar(int n){
 // CHECK2-NEXT:    store i32 9, i32* [[DOTOMP_COMB_UB]], align 4
 // CHECK2-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
 // CHECK2-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
-// CHECK2-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 
@llvm.nvvm.read.ptx.sreg.ntid.x()
+// CHECK2-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 
@__kmpc_get_hardware_num_threads_in_block()
 // CHECK2-NEXT:    [[TMP1:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], 
align 4
 // CHECK2-NEXT:    [[TMP2:%.*]] = load i32, i32* [[TMP1]], align 4
 // CHECK2-NEXT:    call void @__kmpc_distribute_static_init_4(%struct.ident_t* 
@[[GLOB2]], i32 [[TMP2]], i32 91, i32* [[DOTOMP_IS_LAST]], i32* 
[[DOTOMP_COMB_LB]], i32* [[DOTOMP_COMB_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 
[[NVPTX_NUM_THREADS]])
@@ -11321,7 +11321,7 @@ int bar(int n){
 // CHECK2-NEXT:    store i32 99, i32* [[DOTOMP_COMB_UB]], align 4
 // CHECK2-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
 // CHECK2-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
-// CHECK2-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 
@llvm.nvvm.read.ptx.sreg.ntid.x()
+// CHECK2-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 
@__kmpc_get_hardware_num_threads_in_block()
 // CHECK2-NEXT:    [[TMP1:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], 
align 4
 // CHECK2-NEXT:    [[TMP2:%.*]] = load i32, i32* [[TMP1]], align 4
 // CHECK2-NEXT:    call void @__kmpc_distribute_static_init_4(%struct.ident_t* 
@[[GLOB2]], i32 [[TMP2]], i32 91, i32* [[DOTOMP_IS_LAST]], i32* 
[[DOTOMP_COMB_LB]], i32* [[DOTOMP_COMB_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 
[[NVPTX_NUM_THREADS]])
@@ -11905,7 +11905,7 @@ int bar(int n){
 // CHECK3-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_COMB_UB]], align 4
 // CHECK3-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
 // CHECK3-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
-// CHECK3-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 
@llvm.nvvm.read.ptx.sreg.ntid.x()
+// CHECK3-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 
@__kmpc_get_hardware_num_threads_in_block()
 // CHECK3-NEXT:    [[TMP5:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], 
align 4
 // CHECK3-NEXT:    [[TMP6:%.*]] = load i32, i32* [[TMP5]], align 4
 // CHECK3-NEXT:    call void @__kmpc_distribute_static_init_4(%struct.ident_t* 
@[[GLOB2]], i32 [[TMP6]], i32 91, i32* [[DOTOMP_IS_LAST]], i32* 
[[DOTOMP_COMB_LB]], i32* [[DOTOMP_COMB_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 
[[NVPTX_NUM_THREADS]])
@@ -12152,7 +12152,7 @@ int bar(int n){
 // CHECK3-NEXT:    store i32 9, i32* [[DOTOMP_COMB_UB]], align 4
 // CHECK3-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
 // CHECK3-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
-// CHECK3-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 
@llvm.nvvm.read.ptx.sreg.ntid.x()
+// CHECK3-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 
@__kmpc_get_hardware_num_threads_in_block()
 // CHECK3-NEXT:    [[TMP1:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], 
align 4
 // CHECK3-NEXT:    [[TMP2:%.*]] = load i32, i32* [[TMP1]], align 4
 // CHECK3-NEXT:    call void @__kmpc_distribute_static_init_4(%struct.ident_t* 
@[[GLOB2]], i32 [[TMP2]], i32 91, i32* [[DOTOMP_IS_LAST]], i32* 
[[DOTOMP_COMB_LB]], i32* [[DOTOMP_COMB_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 
[[NVPTX_NUM_THREADS]])
@@ -12359,7 +12359,7 @@ int bar(int n){
 // CHECK3-NEXT:    store i32 99, i32* [[DOTOMP_COMB_UB]], align 4
 // CHECK3-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
 // CHECK3-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
-// CHECK3-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 
@llvm.nvvm.read.ptx.sreg.ntid.x()
+// CHECK3-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 
@__kmpc_get_hardware_num_threads_in_block()
 // CHECK3-NEXT:    [[TMP1:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], 
align 4
 // CHECK3-NEXT:    [[TMP2:%.*]] = load i32, i32* [[TMP1]], align 4
 // CHECK3-NEXT:    call void @__kmpc_distribute_static_init_4(%struct.ident_t* 
@[[GLOB2]], i32 [[TMP2]], i32 91, i32* [[DOTOMP_IS_LAST]], i32* 
[[DOTOMP_COMB_LB]], i32* [[DOTOMP_COMB_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 
[[NVPTX_NUM_THREADS]])

diff  --git a/openmp/libomptarget/DeviceRTL/include/Interface.h 
b/openmp/libomptarget/DeviceRTL/include/Interface.h
index ff4f929345651..c7bfede8bbd42 100644
--- a/openmp/libomptarget/DeviceRTL/include/Interface.h
+++ b/openmp/libomptarget/DeviceRTL/include/Interface.h
@@ -203,6 +203,9 @@ void __kmpc_get_shared_variables(void ***GlobalArgs);
 /// External interface to get the thread ID.
 uint32_t __kmpc_get_hardware_thread_id_in_block();
 
+/// External interface to get the number of threads.
+uint32_t __kmpc_get_hardware_num_threads_in_block();
+
 /// Kernel
 ///
 ///{

diff  --git a/openmp/libomptarget/DeviceRTL/src/Mapping.cpp 
b/openmp/libomptarget/DeviceRTL/src/Mapping.cpp
index 66089eaa7496c..c2041856b6cc6 100644
--- a/openmp/libomptarget/DeviceRTL/src/Mapping.cpp
+++ b/openmp/libomptarget/DeviceRTL/src/Mapping.cpp
@@ -84,8 +84,7 @@ uint32_t getNumberOfBlocks() {
 }
 
 uint32_t getNumberOfProcessorElements() {
-  // TODO
-  return mapping::getBlockSize();
+  return getBlockSize();
 }
 
 uint32_t getWarpId() {
@@ -234,5 +233,9 @@ extern "C" {
 __attribute__((noinline)) uint32_t __kmpc_get_hardware_thread_id_in_block() {
   return mapping::getThreadIdInBlock();
 }
+
+__attribute__((noinline)) uint32_t __kmpc_get_hardware_num_threads_in_block() {
+  return mapping::getNumberOfProcessorElements();
+}
 }
 #pragma omp end declare target

diff  --git a/openmp/libomptarget/DeviceRTL/src/Utils.cpp 
b/openmp/libomptarget/DeviceRTL/src/Utils.cpp
index 1317bc69fc512..f11c54ed1c497 100644
--- a/openmp/libomptarget/DeviceRTL/src/Utils.cpp
+++ b/openmp/libomptarget/DeviceRTL/src/Utils.cpp
@@ -22,6 +22,7 @@ namespace _OMP {
 /// Helper to keep code alive without introducing a performance penalty.
 __attribute__((used, weak, optnone)) void keepAlive() {
   __kmpc_get_hardware_thread_id_in_block();
+  __kmpc_get_hardware_num_threads_in_block();
   __kmpc_barrier_simple_spmd(nullptr, 0);
 }
 } // namespace _OMP


        
_______________________________________________
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

Reply via email to