Index: lib/Target/AMDGPU/AMDGPUOpenCLEnqueuedBlockLowering.cpp
===================================================================
--- lib/Target/AMDGPU/AMDGPUOpenCLEnqueuedBlockLowering.cpp
+++ lib/Target/AMDGPU/AMDGPUOpenCLEnqueuedBlockLowering.cpp
@@ -89,6 +89,21 @@
   }
 }
 
+/// If \p U is instruction or constant, collect functions which directly or
+/// indirectly use it.
+static void collectFunctionUsers(User *U, DenseSet<Function *> &Funcs) {
+  if (auto *I = dyn_cast<Instruction>(U)) {
+    auto *F = I->getParent()->getParent();
+    Funcs.insert(F);
+    collectCallers(F, Funcs);
+    return;
+  }
+  if (!isa<Constant>(U))
+    return;
+  for (auto UU : U->users())
+    collectFunctionUsers(&*UU, Funcs);
+}
+
 bool AMDGPUOpenCLEnqueuedBlockLowering::runOnModule(Module &M) {
   DenseSet<Function *> Callers;
   auto &C = M.getContext();
@@ -101,6 +116,7 @@
                                    M.getDataLayout());
         F.setName(Name);
       }
+      DEBUG(dbgs() << "found enqueued kernel: " << F.getName() << '\n');
       auto RuntimeHandle = (F.getName() + ".runtime_handle").str();
       auto *GV = new GlobalVariable(
           M, Type::getInt8Ty(C)->getPointerTo(AMDGPUAS::GLOBAL_ADDRESS),
@@ -111,22 +127,15 @@
       DEBUG(dbgs() << "runtime handle created: " << *GV << '\n');
 
       for (auto U : F.users()) {
-        if (!isa<ConstantExpr>(&*U))
+        auto *UU = &*U;
+        if (!isa<ConstantExpr>(UU))
           continue;
-        auto *BitCast = cast<ConstantExpr>(&*U);
+        collectFunctionUsers(UU, Callers);
+        auto *BitCast = cast<ConstantExpr>(UU);
         auto *NewPtr = ConstantExpr::getPointerCast(GV, BitCast->getType());
         BitCast->replaceAllUsesWith(NewPtr);
         F.addFnAttr("runtime-handle", RuntimeHandle);
         F.setLinkage(GlobalValue::ExternalLinkage);
-
-        // Collect direct or indirect callers of enqueue_kernel.
-        for (auto U : NewPtr->users()) {
-          if (auto *I = dyn_cast<Instruction>(&*U)) {
-            auto *F = I->getParent()->getParent();
-            Callers.insert(F);
-            collectCallers(F, Callers);
-          }
-        }
         Changed = true;
       }
     }
@@ -136,6 +145,7 @@
     if (F->getCallingConv() != CallingConv::AMDGPU_KERNEL)
       continue;
     F->addFnAttr("calls-enqueue-kernel");
+    DEBUG(dbgs() << "mark enqueue_kernel caller:" << F->getName() << '\n');
   }
   return Changed;
 }
Index: test/CodeGen/AMDGPU/enqueue-kernel.ll
===================================================================
--- test/CodeGen/AMDGPU/enqueue-kernel.ll
+++ test/CodeGen/AMDGPU/enqueue-kernel.ll
@@ -1,9 +1,9 @@
 ; RUN: opt -data-layout=A5 -amdgpu-lower-enqueued-block -S < %s | FileCheck %s
 
-; CHECK: @__test_block_invoke_kernel.runtime_handle = external addrspace(1) externally_initialized constant i8 addrspace(1)*
-; CHECK: @__test_block_invoke_2_kernel.runtime_handle = external addrspace(1) externally_initialized constant i8 addrspace(1)*
-; CHECK: @__amdgpu_enqueued_kernel.runtime_handle = external addrspace(1) externally_initialized constant i8 addrspace(1)*
-; CHECK: @__amdgpu_enqueued_kernel.1.runtime_handle = external addrspace(1) externally_initialized constant i8 addrspace(1)*
+; CHECK: @__test_block_invoke_kernel.runtime_handle = addrspace(1) global i8 addrspace(1)* null
+; CHECK: @__test_block_invoke_2_kernel.runtime_handle = addrspace(1) global i8 addrspace(1)* null
+; CHECK: @__amdgpu_enqueued_kernel.runtime_handle = addrspace(1) global i8 addrspace(1)* null
+; CHECK: @__amdgpu_enqueued_kernel.1.runtime_handle = addrspace(1) global i8 addrspace(1)* null
 
 %struct.ndrange_t = type { i32 }
 %opencl.queue_t = type opaque
@@ -80,6 +80,19 @@
   ret void
 }
 
+; __enqueue_kernel* functions may get inlined
+; CHECK-LABEL: define amdgpu_kernel void @inlined_caller
+; CHECK-SAME: #[[AT_CALLER]]
+; CHECK-NOT: @__test_block_invoke_kernel
+; CHECK: load i64, i64 addrspace(1)* bitcast (i8 addrspace(1)* addrspace(1)* @__test_block_invoke_kernel.runtime_handle to i64 addrspace(1)*)
+define amdgpu_kernel void @inlined_caller(i8 addrspace(1)* %a, i8 %b, i64 addrspace(1)* %c, i64 %d) local_unnamed_addr
+  !kernel_arg_addr_space !3 !kernel_arg_access_qual !4 !kernel_arg_type !5 !kernel_arg_base_type !5 !kernel_arg_type_qual !6 {
+entry:
+  %tmp = load i64, i64 addrspace(1)* addrspacecast (i64* bitcast (void (<{ i32, i32, i8 addrspace(1)*, i8 }>)* @__test_block_invoke_kernel to i64*) to i64 addrspace(1)*)
+  store i64 %tmp, i64 addrspace(1)* %c
+  ret void
+}
+
 ; CHECK-LABEL: define dso_local amdgpu_kernel void @__test_block_invoke_kernel
 ; CHECK-SAME: #[[AT1:[0-9]+]]
 define internal amdgpu_kernel void @__test_block_invoke_kernel(<{ i32, i32, i8 addrspace(1)*, i8 }> %arg) #0