Index: lib/Target/AMDGPU/AMDGPUAsmPrinter.h
===================================================================
--- lib/Target/AMDGPU/AMDGPUAsmPrinter.h
+++ lib/Target/AMDGPU/AMDGPUAsmPrinter.h
@@ -41,7 +41,7 @@
     // the end are tracked separately.
     int32_t NumVGPR = 0;
     int32_t NumExplicitSGPR = 0;
-    uint32_t PrivateSegmentSize = 0;
+    uint64_t PrivateSegmentSize = 0;
     bool UsesVCC = false;
     bool UsesFlatScratch = false;
     bool HasDynamicallySizedStack = false;
@@ -61,7 +61,7 @@
     uint32_t DX10Clamp = 0;
     uint32_t DebugMode = 0;
     uint32_t IEEEMode = 0;
-    uint32_t ScratchSize = 0;
+    uint64_t ScratchSize = 0;
 
     uint64_t ComputePGMRSrc1 = 0;
 
@@ -144,7 +144,7 @@
                        const SIProgramInfo &KernelInfo);
   void emitCommonFunctionComments(uint32_t NumVGPR,
                                   uint32_t NumSGPR,
-                                  uint32_t ScratchSize,
+                                  uint64_t ScratchSize,
                                   uint64_t CodeSize);
 
 public:
Index: lib/Target/AMDGPU/AMDGPUAsmPrinter.cpp
===================================================================
--- lib/Target/AMDGPU/AMDGPUAsmPrinter.cpp
+++ lib/Target/AMDGPU/AMDGPUAsmPrinter.cpp
@@ -262,7 +262,7 @@
 void AMDGPUAsmPrinter::emitCommonFunctionComments(
   uint32_t NumVGPR,
   uint32_t NumSGPR,
-  uint32_t ScratchSize,
+  uint64_t ScratchSize,
   uint64_t CodeSize) {
   OutStreamer->emitRawComment(" codeLenInByte = " + Twine(CodeSize), false);
   OutStreamer->emitRawComment(" NumSgprs: " + Twine(NumSGPR), false);
@@ -600,7 +600,7 @@
 
   int32_t MaxVGPR = -1;
   int32_t MaxSGPR = -1;
-  uint32_t CalleeFrameSize = 0;
+  uint64_t CalleeFrameSize = 0;
 
   for (const MachineBasicBlock &MBB : MF) {
     for (const MachineInstr &MI : MBB) {
@@ -718,7 +718,7 @@
           MaxSGPR = std::max(MaxSGPR, MaxSGPRGuess);
           MaxVGPR = std::max(MaxVGPR, 23);
 
-          CalleeFrameSize = std::max(CalleeFrameSize, 16384u);
+          CalleeFrameSize = std::max(CalleeFrameSize, UINT64_C(16384));
           Info.UsesVCC = true;
           Info.UsesFlatScratch = ST.hasFlatAddressSpace();
           Info.HasDynamicallySizedStack = true;
@@ -763,6 +763,12 @@
   ProgInfo.FlatUsed = Info.UsesFlatScratch;
   ProgInfo.DynamicCallStack = Info.HasDynamicallySizedStack || Info.HasRecursion;
 
+  if (!isUInt<32>(ProgInfo.ScratchSize)) {
+    DiagnosticInfoStackSize DiagStackSize(*MF.getFunction(),
+                                          ProgInfo.ScratchSize, DS_Error);
+    MF.getFunction()->getContext().diagnose(DiagStackSize);
+  }
+
   const SISubtarget &STM = MF.getSubtarget<SISubtarget>();
   const SIMachineFunctionInfo *MFI = MF.getInfo<SIMachineFunctionInfo>();
   const SIInstrInfo *TII = STM.getInstrInfo();
Index: test/CodeGen/AMDGPU/stack-size-overflow.ll
===================================================================
--- /dev/null
+++ test/CodeGen/AMDGPU/stack-size-overflow.ll
@@ -0,0 +1,14 @@
+; RUN: not llc -march=amdgcn < %s 2>&1 | FileCheck -check-prefix=ERROR %s
+; RUN: not llc -march=amdgcn < %s | FileCheck -check-prefix=GCN %s
+
+declare void @llvm.memset.p0i8.i32(i8* nocapture, i8, i32, i32, i1) #1
+
+; ERROR: error: stack size limit exceeded (4294967296) in stack_size_limit
+; GCN: ; ScratchSize: 4294967296
+define amdgpu_kernel void @stack_size_limit() #0 {
+entry:
+  %alloca = alloca [1073741823 x i32], align 4
+  %bc = bitcast [1073741823 x i32]* %alloca to i8*
+  call void @llvm.memset.p0i8.i32(i8* %bc, i8 9, i32 1073741823, i32 1, i1 true)
+  ret void
+}