diff --git a/llvm/lib/Transforms/IPO/OpenMPOpt.cpp b/llvm/lib/Transforms/IPO/OpenMPOpt.cpp
--- a/llvm/lib/Transforms/IPO/OpenMPOpt.cpp
+++ b/llvm/lib/Transforms/IPO/OpenMPOpt.cpp
@@ -33,6 +33,8 @@
 #include "llvm/IR/GlobalValue.h"
 #include "llvm/IR/Instruction.h"
 #include "llvm/IR/IntrinsicInst.h"
+#include "llvm/IR/IntrinsicsAMDGPU.h"
+#include "llvm/IR/IntrinsicsNVPTX.h"
 #include "llvm/InitializePasses.h"
 #include "llvm/Support/CommandLine.h"
 #include "llvm/Transforms/IPO.h"
@@ -2554,9 +2556,8 @@
   auto &OMPInfoCache = static_cast<OMPInformationCache &>(A.getInfoCache());
   auto &RFI = OMPInfoCache.RFIs[OMPRTL___kmpc_target_init];
 
-  // Check if the edge into the successor block compares the __kmpc_target_init
-  // result with -1. If we are in non-SPMD-mode that signals only the main
-  // thread will execute the edge.
+  // Check if the edge into the successor block contains a condition that only
+  // lets the main thread execute it.
   auto IsInitialThreadOnly = [&](BranchInst *Edge, BasicBlock *SuccessorBB) {
     if (!Edge || !Edge->isConditional())
       return false;
@@ -2571,7 +2572,7 @@
     if (!C)
       return false;
 
-    // Match:  -1 == __kmpc_target_init (for non-SPMD kernels only!)
+    // Match: -1 == __kmpc_target_init (for non-SPMD kernels only!)
     if (C->isAllOnesValue()) {
       auto *CB = dyn_cast<CallBase>(Cmp->getOperand(0));
       CB = CB ? OpenMPOpt::getCallIfRegularCall(*CB, &RFI) : nullptr;
@@ -2583,6 +2584,18 @@
       return IsSPMDModeCI && IsSPMDModeCI->isZero();
     }
 
+    if (C->isZero()) {
+      // Match: 0 == llvm.nvvm.read.ptx.sreg.tid.x()
+      if (auto *II = dyn_cast<IntrinsicInst>(Cmp->getOperand(0)))
+        if (II->getIntrinsicID() == Intrinsic::nvvm_read_ptx_sreg_tid_x)
+          return true;
+
+      // Match: 0 == llvm.amdgcn.workitem.id.x()
+      if (auto *II = dyn_cast<IntrinsicInst>(Cmp->getOperand(0)))
+        if (II->getIntrinsicID() == Intrinsic::amdgcn_workitem_id_x)
+          return true;
+    }
+
     return false;
   };
 
diff --git a/llvm/test/Transforms/OpenMP/single_threaded_execution.ll b/llvm/test/Transforms/OpenMP/single_threaded_execution.ll
--- a/llvm/test/Transforms/OpenMP/single_threaded_execution.ll
+++ b/llvm/test/Transforms/OpenMP/single_threaded_execution.ll
@@ -18,8 +18,6 @@
   %cmp = icmp eq i32 %call, -1
   br i1 %cmp, label %if.then, label %if.else
 if.then:
-  call void @nvptx()
-  call void @amdgcn()
   br label %if.end
 if.else:
   br label %if.end
@@ -31,13 +29,15 @@
 ; REMARKS: remark: single_threaded_execution.c:1:0: Could not internalize function. Some optimizations may not be possible.
 ; REMARKS-NOT: remark: single_threaded_execution.c:1:0: Could not internalize function. Some optimizations may not be possible.
 
-; CHECK-DAG: [openmp-opt] Basic block @nvptx entry is executed by a single thread.
+; CHECK-NOT: [openmp-opt] Basic block @nvptx entry is executed by a single thread.
 ; CHECK-DAG: [openmp-opt] Basic block @nvptx if.then is executed by a single thread.
-; CHECK-DAG: [openmp-opt] Basic block @nvptx if.end is executed by a single thread.
+; CHECK-NOT: [openmp-opt] Basic block @nvptx if.end is executed by a single thread.
 ; Function Attrs: noinline
-define internal void @nvptx() {
+define void @nvptx() {
 entry:
-  br i1 true, label %if.then, label %if.end
+  %call = call i32 @llvm.nvvm.read.ptx.sreg.tid.x()
+  %cmp = icmp eq i32 %call, 0
+  br i1 %cmp, label %if.then, label %if.end
 
 if.then:
   call void @foo()
@@ -50,13 +50,15 @@
   ret void
 }
 
-; CHECK-DAG: [openmp-opt] Basic block @amdgcn entry is executed by a single thread.
+; CHECK-NOT: [openmp-opt] Basic block @amdgcn entry is executed by a single thread.
 ; CHECK-DAG: [openmp-opt] Basic block @amdgcn if.then is executed by a single thread.
-; CHECK-DAG: [openmp-opt] Basic block @amdgcn if.end is executed by a single thread.
+; CHECK-NOT: [openmp-opt] Basic block @amdgcn if.end is executed by a single thread.
 ; Function Attrs: noinline
-define internal void @amdgcn() {
+define void @amdgcn() {
 entry:
-  br i1 false, label %if.then, label %if.end
+  %call = call i32 @llvm.amdgcn.workitem.id.x()
+  %cmp = icmp eq i32 %call, 0
+  br i1 %cmp, label %if.then, label %if.end
 
 if.then:
   call void @foo()
@@ -104,6 +106,7 @@
 declare void @__kmpc_kernel_prepare_parallel(i8*)
 
 declare i32 @__kmpc_target_init(%struct.ident_t*, i1, i1, i1)
+
 declare void @__kmpc_target_deinit(%struct.ident_t*, i1, i1)
 
 attributes #0 = { cold noinline }