[AMDGPU] Add amdgpu_gfx calling convention

[Flakebi]

Summary

Add a calling convention called amdgpu_gfx for real function calls
within graphics shaders. For the moment, this uses the same calling
convention as other calls in amdgpu, with registers excluded for return
address, stack pointer and stack buffer descriptor.

[arsenm]

One problem is you seem to be assuming this is the one calling convention usable from both CS and other shader types. However, the default FP mode is assumed to be different between these (we assume IEEE mode is enabled for CS and not for other shader types). These would all need to be agree to avoid inserting mode switches.

I also don't see any tests for the inreg SGPR arguments. I think this is the most difficult part to handle since there isn't a straightforward way to insert the necessary waterfall loops, so it probably will require many later patches.

Can you also add a note to the AMDGPUUsage section since this is a different ABI with different register counts

llvm/include/llvm/IR/CallingConv.h

245

I don't like the name. "Callable" doesn't feel like it belongs in the name. amdgpu_gfx_func would be better, but I don't think that's great either

llvm/lib/Target/AMDGPU/AMDGPUCallLowering.cpp

462–463

I think shaders should refer to the entry points, so IsCallableShader wouldn't make sense as a name

llvm/lib/Target/AMDGPU/SIISelLowering.cpp

3120

This should probably be a positive hsa check

llvm/lib/Target/AMDGPU/Utils/AMDGPUBaseInfo.cpp

1014

I think this should return false, it's not an entry point

llvm/lib/Target/AMDGPU/Utils/AMDGPUPALMetadata.cpp

737

llvm_unreachable

llvm/test/CodeGen/AMDGPU/unsupported-calls.ll

72

This is very fragile negative check.

It's also questionable if this should be allowed to call C calling conventions. It would need an FP mode switch since the default IEEE mode bit is different (although it's currently not the backend's responsibility to insert it)

[Flakebi]

Update from internal review comments.

Rename calling convention to amdgpu_gfx.

[Flakebi]

Disallow calls with C calling convention from shaders

[Flakebi]

llvm/test/CodeGen/AMDGPU/unsupported-calls.ll

72

C calling convention is now disallowed. I made the check more generic but I didn’t find a great way to make it a positive check that would not break checks for the other test functions.

[nhaehnle]

I noticed what I believe is one more correctness issues, plus a few assorted comments.

llvm/lib/AsmParser/LLParser.cpp

2078–2080

This could be aligned with the rest of the case statements again now.

llvm/lib/IR/AsmWriter.cpp

391–393

Same here wrt alignment to the other case statements.

llvm/lib/Target/AMDGPU/AMDGPUCallLowering.cpp

473–476

IsGraphics is true for amdgpu_gfx functions, right? Why are we using SI_RETURN_TO_EPILOG in this case?

llvm/lib/Target/AMDGPU/SIISelLowering.cpp

2957–2963

I assume you'll add support for calls from amdgpu_cs to amdgpu_gfx in a follow-up change?

llvm/lib/Target/AMDGPU/SIInstrInfo.cpp

5090 ↗

(On Diff #299346)

I believe this will break radeonsi because it actually ends up settings OS == unknown. We'll have to follow-up on that, but in the meantime, couldn't this use AMDGPU::isGraphics?

[nhaehnle]

llvm/lib/Target/AMDGPU/SIInstrInfo.cpp

5090 ↗

(On Diff #299346)

Okay, technically I suppose it shouldn't break radeonsi, but regress it.

[Flakebi]

Fix comments

llvm/lib/Target/AMDGPU/SIISelLowering.cpp

2957–2963

Calls from any shader type to amdgpu_gfx should be allowed with this. Calls with any calling convention that is not amdgpu_gfx will fail.

“Fail” in this case means it silently prints a message if output is enabled and compiles successful anyway (just omitting the call).

[Flakebi]

Rebased

[Flakebi]

Fix lld test failure

[Flakebi]

ping :)

[arsenm]

Needs globalisel tests

llvm/lib/Target/AMDGPU/SIISelLowering.cpp

1787–2258

Missing equivalent GlobalISel changes

[sebastian-ne]

llvm/lib/Target/AMDGPU/SIISelLowering.cpp

1787–2258

What would that be?
The change here shouldn’t change behavior apart from checking the assert for amdgpu_gfx and I can’t find the same structure in AMDGPUCallLowering.cpp.

[arsenm]

llvm/lib/Target/AMDGPU/SIISelLowering.cpp

1787–2258

Should at least add the tests or run line for it

[sebastian-ne]

Fixed a bug for GlobalISel and add tests.
Thanks for pushing for tests Matt.

llvm/include/llvm/IR/CallingConv.h

enum {
     /// Calling convention between AArch64 SVE functions
     AArch64_SVE_VectorCall = 98,
 
     /// Calling convention for emscripten __invoke_* functions. The first
     /// argument is required to be the function ptr being indirectly called.
     /// The remainder matches the regular calling convention.
     WASM_EmscriptenInvoke = 99,
 
+    /// Calling convention used for AMDPAL callable shaders.
+    AMDGPU_GFX_CALLABLE = 100,
+
     /// The highest possible calling convention ID. Must be some 2^k - 1.
     MaxID = 1023
   };
 
 } // end namespace CallingConv
 
 } // end namespace llvm
 
 #endif // LLVM_IR_CALLINGCONV_H

llvm/lib/AsmParser/LLLexer.cpp

#define KEYWORD(STR)                                                           \
   KEYWORD(amdgpu_vs);
   KEYWORD(amdgpu_ls);
   KEYWORD(amdgpu_hs);
   KEYWORD(amdgpu_es);
   KEYWORD(amdgpu_gs);
   KEYWORD(amdgpu_ps);
   KEYWORD(amdgpu_cs);
   KEYWORD(amdgpu_kernel);
+  KEYWORD(amdgpu_gfx_callable);
   KEYWORD(tailcc);
 
   KEYWORD(cc);
   KEYWORD(c);
 
   KEYWORD(attributes);
 
   KEYWORD(alwaysinline);

llvm/lib/AsmParser/LLParser.cpp

bool LLParser::ParseOptionalCallingConv(unsigned &CC) {
   case lltok::kw_preserve_allcc: CC = CallingConv::PreserveAll; break;
   case lltok::kw_ghccc:          CC = CallingConv::GHC; break;
   case lltok::kw_swiftcc:        CC = CallingConv::Swift; break;
   case lltok::kw_x86_intrcc:     CC = CallingConv::X86_INTR; break;
   case lltok::kw_hhvmcc:         CC = CallingConv::HHVM; break;
   case lltok::kw_hhvm_ccc:       CC = CallingConv::HHVM_C; break;
   case lltok::kw_cxx_fast_tlscc: CC = CallingConv::CXX_FAST_TLS; break;
   case lltok::kw_amdgpu_vs:      CC = CallingConv::AMDGPU_VS; break;
+  case lltok::kw_amdgpu_gfx_callable:
+    CC = CallingConv::AMDGPU_GFX_CALLABLE;
+    break;
   case lltok::kw_amdgpu_ls:      CC = CallingConv::AMDGPU_LS; break;
   case lltok::kw_amdgpu_hs:      CC = CallingConv::AMDGPU_HS; break;
   case lltok::kw_amdgpu_es:      CC = CallingConv::AMDGPU_ES; break;
   case lltok::kw_amdgpu_gs:      CC = CallingConv::AMDGPU_GS; break;
   case lltok::kw_amdgpu_ps:      CC = CallingConv::AMDGPU_PS; break;
   case lltok::kw_amdgpu_cs:      CC = CallingConv::AMDGPU_CS; break;
   case lltok::kw_amdgpu_kernel:  CC = CallingConv::AMDGPU_KERNEL; break;
   case lltok::kw_tailcc:         CC = CallingConv::Tail; break;

llvm/lib/AsmParser/LLToken.h

enum Kind {
   kw_amdgpu_vs,
   kw_amdgpu_ls,
   kw_amdgpu_hs,
   kw_amdgpu_es,
   kw_amdgpu_gs,
   kw_amdgpu_ps,
   kw_amdgpu_cs,
   kw_amdgpu_kernel,
+  kw_amdgpu_gfx_callable,
   kw_tailcc,
 
   // Attributes:
   kw_attributes,
   kw_allocsize,
   kw_alwaysinline,
   kw_argmemonly,
   kw_sanitize_address,

llvm/lib/IR/AsmWriter.cpp

static void PrintCallingConv(unsigned cc, raw_ostream &Out) {
   case CallingConv::AMDGPU_VS:     Out << "amdgpu_vs"; break;
   case CallingConv::AMDGPU_LS:     Out << "amdgpu_ls"; break;
   case CallingConv::AMDGPU_HS:     Out << "amdgpu_hs"; break;
   case CallingConv::AMDGPU_ES:     Out << "amdgpu_es"; break;
   case CallingConv::AMDGPU_GS:     Out << "amdgpu_gs"; break;
   case CallingConv::AMDGPU_PS:     Out << "amdgpu_ps"; break;
   case CallingConv::AMDGPU_CS:     Out << "amdgpu_cs"; break;
   case CallingConv::AMDGPU_KERNEL: Out << "amdgpu_kernel"; break;
+  case CallingConv::AMDGPU_GFX_CALLABLE:
+    Out << "amdgpu_gfx_callable";
+    break;
   }
 }
 
 enum PrefixType {
   GlobalPrefix,
   ComdatPrefix,
   LabelPrefix,
   LocalPrefix,

llvm/lib/Target/AMDGPU/AMDGPUAsmPrinter.cpp

bool AMDGPUAsmPrinter::runOnMachineFunction(MachineFunction &MF) {
   } else {
     auto I = CallGraphResourceInfo.insert(
       std::make_pair(&MF.getFunction(), SIFunctionResourceInfo()));
     SIFunctionResourceInfo &Info = I.first->second;
     assert(I.second && "should only be called once per function");
     Info = analyzeResourceUsage(MF);
   }
 
-  if (STM.isAmdPalOS())
+  if (STM.isAmdPalOS() && MFI->isEntryFunction())
     EmitPALMetadata(MF, CurrentProgramInfo);
   else if (!STM.isAmdHsaOS()) {
     EmitProgramInfoSI(MF, CurrentProgramInfo);
   }
 
   DumpCodeInstEmitter = nullptr;
   if (STM.dumpCode()) {
     // For -dumpcode, get the assembler out of the streamer, even if it does

llvm/lib/Target/AMDGPU/AMDGPUCallLowering.cpp

bool AMDGPUCallLowering::lowerReturn(MachineIRBuilder &B,
   MachineFunction &MF = B.getMF();
   MachineRegisterInfo &MRI = MF.getRegInfo();
   SIMachineFunctionInfo *MFI = MF.getInfo<SIMachineFunctionInfo>();
   MFI->setIfReturnsVoid(!Val);
 
   assert(!Val == VRegs.empty() && "Return value without a vreg");
 
   CallingConv::ID CC = B.getMF().getFunction().getCallingConv();
   const bool IsShader = AMDGPU::isShader(CC);
+  const bool IsCallableShader = CC == CallingConv::AMDGPU_GFX_CALLABLE;
   const bool IsWaveEnd = (IsShader && MFI->returnsVoid()) ||
                          AMDGPU::isKernel(CC);
   if (IsWaveEnd) {
     B.buildInstr(AMDGPU::S_ENDPGM)
       .addImm(0);
     return true;
   }
 
   auto const &ST = MF.getSubtarget<GCNSubtarget>();
 
-  unsigned ReturnOpc =
-      IsShader ? AMDGPU::SI_RETURN_TO_EPILOG : AMDGPU::S_SETPC_B64_return;
+  unsigned ReturnOpc = IsShader && !IsCallableShader
+                           ? AMDGPU::SI_RETURN_TO_EPILOG
+                           : AMDGPU::S_SETPC_B64_return;
 
   auto Ret = B.buildInstrNoInsert(ReturnOpc);
   Register ReturnAddrVReg;
   if (ReturnOpc == AMDGPU::S_SETPC_B64_return) {
     ReturnAddrVReg = MRI.createVirtualRegister(&AMDGPU::CCR_SGPR_64RegClass);
     Ret.addUse(ReturnAddrVReg);
   }
 

llvm/lib/Target/AMDGPU/AMDGPUCallingConv.td

 
 // Inversion of CCIfInReg
 class CCIfNotInReg<CCAction A> : CCIf<"!ArgFlags.isInReg()", A> {}
 class CCIfExtend<CCAction A>
   : CCIf<"ArgFlags.isSExt() || ArgFlags.isZExt()", A>;
 
 // Calling convention for SI
 def CC_SI : CallingConv<[
+  // 0-3 are reserved for the stack buffer descriptor
+  // 30-31 are reserved for the return address
+  // 32 is reserved for the stack pointer
+  CCIfInReg<CCIfType<[f32, i32, f16, i16, v2i16, v2f16] , CCAssignToReg<[
+    SGPR4, SGPR5, SGPR6, SGPR7,
+    SGPR8, SGPR9, SGPR10, SGPR11, SGPR12, SGPR13, SGPR14, SGPR15,
+    SGPR16, SGPR17, SGPR18, SGPR19, SGPR20, SGPR21, SGPR22, SGPR23,
+    SGPR24, SGPR25, SGPR26, SGPR27, SGPR28, SGPR29,
+    SGPR33, SGPR34, SGPR35, SGPR36, SGPR37, SGPR38, SGPR39,
+    SGPR40, SGPR41, SGPR42, SGPR43
+  ]>>>,
+
+  CCIfNotInReg<CCIfType<[f32, i32, f16, i16, v2i16, v2f16] , CCAssignToReg<[
+    VGPR0, VGPR1, VGPR2, VGPR3, VGPR4, VGPR5, VGPR6, VGPR7,
+    VGPR8, VGPR9, VGPR10, VGPR11, VGPR12, VGPR13, VGPR14, VGPR15,
+    VGPR16, VGPR17, VGPR18, VGPR19, VGPR20, VGPR21, VGPR22, VGPR23,
+    VGPR24, VGPR25, VGPR26, VGPR27, VGPR28, VGPR29, VGPR30, VGPR31,
+    VGPR32, VGPR33, VGPR34, VGPR35, VGPR36, VGPR37, VGPR38, VGPR39,
+    VGPR40, VGPR41, VGPR42, VGPR43, VGPR44, VGPR45, VGPR46, VGPR47,
+    VGPR48, VGPR49, VGPR50, VGPR51, VGPR52, VGPR53, VGPR54, VGPR55,
+    VGPR56, VGPR57, VGPR58, VGPR59, VGPR60, VGPR61, VGPR62, VGPR63,
+    VGPR64, VGPR65, VGPR66, VGPR67, VGPR68, VGPR69, VGPR70, VGPR71,
+    VGPR72, VGPR73, VGPR74, VGPR75, VGPR76, VGPR77, VGPR78, VGPR79,
+    VGPR80, VGPR81, VGPR82, VGPR83, VGPR84, VGPR85, VGPR86, VGPR87,
+    VGPR88, VGPR89, VGPR90, VGPR91, VGPR92, VGPR93, VGPR94, VGPR95,
+    VGPR96, VGPR97, VGPR98, VGPR99, VGPR100, VGPR101, VGPR102, VGPR103,
+    VGPR104, VGPR105, VGPR106, VGPR107, VGPR108, VGPR109, VGPR110, VGPR111,
+    VGPR112, VGPR113, VGPR114, VGPR115, VGPR116, VGPR117, VGPR118, VGPR119,
+    VGPR120, VGPR121, VGPR122, VGPR123, VGPR124, VGPR125, VGPR126, VGPR127,
+    VGPR128, VGPR129, VGPR130, VGPR131, VGPR132, VGPR133, VGPR134, VGPR135
+  ]>>>,
+
+  CCIfType<[i32, f32, v2i16, v2f16, i16, f16, i1], CCAssignToStack<4, 4>>,
+  CCIfType<[i64, f64, v2i32, v2f32], CCAssignToStack<8, 4>>,
+  CCIfType<[v3i32, v3f32], CCAssignToStack<12, 4>>,
+  CCIfType<[v4i32, v4f32, v2i64, v2f64], CCAssignToStack<16, 4>>,
+  CCIfType<[v5i32, v5f32], CCAssignToStack<20, 4>>,
+  CCIfType<[v8i32, v8f32], CCAssignToStack<32, 4>>,
+  CCIfType<[v16i32, v16f32], CCAssignToStack<64, 4>>
+]>;
+
+def RetCC_SI : CallingConv<[
+  // 0-3 are reserved for the stack buffer descriptor
+  // 32 is reserved for the stack pointer
+  CCIfInReg<CCIfType<[f32, i32, f16, i16, v2i16, v2f16] , CCAssignToReg<[
+    SGPR4, SGPR5, SGPR6, SGPR7,
+    SGPR8, SGPR9, SGPR10, SGPR11, SGPR12, SGPR13, SGPR14, SGPR15,
+    SGPR16, SGPR17, SGPR18, SGPR19, SGPR20, SGPR21, SGPR22, SGPR23,
+    SGPR24, SGPR25, SGPR26, SGPR27, SGPR28, SGPR29, SGPR30, SGPR31,
+    SGPR33, SGPR34, SGPR35, SGPR36, SGPR37, SGPR38, SGPR39,
+    SGPR40, SGPR41, SGPR42, SGPR43
+  ]>>>,
+
+  CCIfNotInReg<CCIfType<[f32, i32, f16, i16, v2i16, v2f16] , CCAssignToReg<[
+    VGPR0, VGPR1, VGPR2, VGPR3, VGPR4, VGPR5, VGPR6, VGPR7,
+    VGPR8, VGPR9, VGPR10, VGPR11, VGPR12, VGPR13, VGPR14, VGPR15,
+    VGPR16, VGPR17, VGPR18, VGPR19, VGPR20, VGPR21, VGPR22, VGPR23,
+    VGPR24, VGPR25, VGPR26, VGPR27, VGPR28, VGPR29, VGPR30, VGPR31,
+    VGPR32, VGPR33, VGPR34, VGPR35, VGPR36, VGPR37, VGPR38, VGPR39,
+    VGPR40, VGPR41, VGPR42, VGPR43, VGPR44, VGPR45, VGPR46, VGPR47,
+    VGPR48, VGPR49, VGPR50, VGPR51, VGPR52, VGPR53, VGPR54, VGPR55,
+    VGPR56, VGPR57, VGPR58, VGPR59, VGPR60, VGPR61, VGPR62, VGPR63,
+    VGPR64, VGPR65, VGPR66, VGPR67, VGPR68, VGPR69, VGPR70, VGPR71,
+    VGPR72, VGPR73, VGPR74, VGPR75, VGPR76, VGPR77, VGPR78, VGPR79,
+    VGPR80, VGPR81, VGPR82, VGPR83, VGPR84, VGPR85, VGPR86, VGPR87,
+    VGPR88, VGPR89, VGPR90, VGPR91, VGPR92, VGPR93, VGPR94, VGPR95,
+    VGPR96, VGPR97, VGPR98, VGPR99, VGPR100, VGPR101, VGPR102, VGPR103,
+    VGPR104, VGPR105, VGPR106, VGPR107, VGPR108, VGPR109, VGPR110, VGPR111,
+    VGPR112, VGPR113, VGPR114, VGPR115, VGPR116, VGPR117, VGPR118, VGPR119,
+    VGPR120, VGPR121, VGPR122, VGPR123, VGPR124, VGPR125, VGPR126, VGPR127,
+    VGPR128, VGPR129, VGPR130, VGPR131, VGPR132, VGPR133, VGPR134, VGPR135
+  ]>>>,
+
+  CCIfType<[i32, f32, v2i16, v2f16, i16, f16, i1], CCAssignToStack<4, 4>>,
+  CCIfType<[i64, f64, v2i32, v2f32], CCAssignToStack<8, 4>>,
+  CCIfType<[v3i32, v3f32], CCAssignToStack<12, 4>>,
+  CCIfType<[v4i32, v4f32, v2i64, v2f64], CCAssignToStack<16, 4>>,
+  CCIfType<[v5i32, v5f32], CCAssignToStack<20, 4>>,
+  CCIfType<[v8i32, v8f32], CCAssignToStack<32, 4>>,
+  CCIfType<[v16i32, v16f32], CCAssignToStack<64, 4>>
+]>;
+
+def CC_SI_SHADER : CallingConv<[
 
   CCIfInReg<CCIfType<[f32, i32, f16, i16, v2i16, v2f16] , CCAssignToReg<[
     SGPR0, SGPR1, SGPR2, SGPR3, SGPR4, SGPR5, SGPR6, SGPR7,
     SGPR8, SGPR9, SGPR10, SGPR11, SGPR12, SGPR13, SGPR14, SGPR15,
     SGPR16, SGPR17, SGPR18, SGPR19, SGPR20, SGPR21, SGPR22, SGPR23,
     SGPR24, SGPR25, SGPR26, SGPR27, SGPR28, SGPR29, SGPR30, SGPR31,
     SGPR32, SGPR33, SGPR34, SGPR35, SGPR36, SGPR37, SGPR38, SGPR39,
     SGPR40, SGPR41, SGPR42, SGPR43
CCIfType<[i32, f32, i16, f16, v2i16, v2f16], CCAssignToReg<[
     VGPR16, VGPR17, VGPR18, VGPR19, VGPR20, VGPR21, VGPR22, VGPR23,
     VGPR24, VGPR25, VGPR26, VGPR27, VGPR28, VGPR29, VGPR30, VGPR31]>>,
 ]>;
 
 def CC_AMDGPU : CallingConv<[
    CCIf<"static_cast<const GCNSubtarget&>"
          "(State.getMachineFunction().getSubtarget()).getGeneration() >= "
            "AMDGPUSubtarget::SOUTHERN_ISLANDS",
-        CCDelegateTo<CC_SI>>,
+        CCDelegateTo<CC_SI_SHADER>>,
    CCIf<"static_cast<const GCNSubtarget&>"
          "(State.getMachineFunction().getSubtarget()).getGeneration() >= "
            "AMDGPUSubtarget::SOUTHERN_ISLANDS && State.getCallingConv() == CallingConv::C",
         CCDelegateTo<CC_AMDGPU_Func>>
 ]>;

llvm/lib/Target/AMDGPU/AMDGPUISelLowering.cpp

CCAssignFn *AMDGPUCallLowering::CCAssignFnForCall(CallingConv::ID CC,
   case CallingConv::AMDGPU_HS:
   case CallingConv::AMDGPU_ES:
   case CallingConv::AMDGPU_LS:
     return CC_AMDGPU;
   case CallingConv::C:
   case CallingConv::Fast:
   case CallingConv::Cold:
     return CC_AMDGPU_Func;
+  case CallingConv::AMDGPU_GFX_CALLABLE:
+    return CC_SI;
   case CallingConv::AMDGPU_KERNEL:
   case CallingConv::SPIR_KERNEL:
   default:
     report_fatal_error("Unsupported calling convention for call");
   }
 }
 
 CCAssignFn *AMDGPUCallLowering::CCAssignFnForReturn(CallingConv::ID CC,
                                                     bool IsVarArg) {
   switch (CC) {
   case CallingConv::AMDGPU_KERNEL:
   case CallingConv::SPIR_KERNEL:
     llvm_unreachable("kernels should not be handled here");
   case CallingConv::AMDGPU_VS:
   case CallingConv::AMDGPU_GS:
   case CallingConv::AMDGPU_PS:
   case CallingConv::AMDGPU_CS:
   case CallingConv::AMDGPU_HS:
   case CallingConv::AMDGPU_ES:
   case CallingConv::AMDGPU_LS:
     return RetCC_SI_Shader;
+  case CallingConv::AMDGPU_GFX_CALLABLE:
+    return RetCC_SI;
   case CallingConv::C:
   case CallingConv::Fast:
   case CallingConv::Cold:
     return RetCC_AMDGPU_Func;
   default:
     report_fatal_error("Unsupported calling convention.");
   }
 }

llvm/lib/Target/AMDGPU/AMDGPUTargetTransformInfo.cpp

case CallingConv::SPIR_KERNEL:
     return true;
   case CallingConv::AMDGPU_VS:
   case CallingConv::AMDGPU_LS:
   case CallingConv::AMDGPU_HS:
   case CallingConv::AMDGPU_ES:
   case CallingConv::AMDGPU_GS:
   case CallingConv::AMDGPU_PS:
   case CallingConv::AMDGPU_CS:
+  case CallingConv::AMDGPU_GFX_CALLABLE:
     // For non-compute shaders, SGPR inputs are marked with either inreg.
     // Everything else is in VGPRs.
     return F->getAttributes().hasParamAttribute(A->getArgNo(), Attribute::InReg);
   default:
     // TODO: Should calls support inreg for SGPR inputs?
     return false;
   }
 }

llvm/lib/Target/AMDGPU/SIISelLowering.cpp

SDValue SITargetLowering::getPreloadedValue(SelectionDAG &DAG,
   return CreateLiveInRegister(DAG, RC, Reg->getRegister(), VT);
 }
 
 static void processShaderInputArgs(SmallVectorImpl<ISD::InputArg> &Splits,
                                    CallingConv::ID CallConv,
                                    ArrayRef<ISD::InputArg> Ins,
                                    BitVector &Skipped,
                                    FunctionType *FType,
                                    SIMachineFunctionInfo *Info) {
   for (unsigned I = 0, E = Ins.size(), PSInputNum = 0; I != E; ++I) {
     const ISD::InputArg *Arg = &Ins[I];
 
     assert((!Arg->VT.isVector() || Arg->VT.getScalarSizeInBits() == 16) &&
            "vector type argument should have been split");
 
     // First check if it's a PS input addr.
     if (CallConv == CallingConv::AMDGPU_PS &&
         !Arg->Flags.isInReg() && PSInputNum <= 15) {
       bool SkipArg = !Arg->Used && !Info->isPSInputAllocated(PSInputNum);
 
       // Inconveniently only the first part of the split is marked as isSplit,
       // so skip to the end. We only want to increment PSInputNum once for the
       // entire split argument.
       if (Arg->Flags.isSplit()) {
         while (!Arg->Flags.isSplitEnd()) {
           assert((!Arg->VT.isVector() ||
                   Arg->VT.getScalarSizeInBits() == 16) &&
                  "unexpected vector split in ps argument type");
           if (!SkipArg)
             Splits.push_back(*Arg);
           Arg = &Ins[++I];
         }
       }
 
       if (SkipArg) {
         // We can safely skip PS inputs.
         Skipped.set(Arg->getOrigArgIndex());
         ++PSInputNum;
         continue;
       }
 
       Info->markPSInputAllocated(PSInputNum);
       if (Arg->Used)
         Info->markPSInputEnabled(PSInputNum);
 
       ++PSInputNum;
     }
 
     Splits.push_back(*Arg);
   }
 }
 
 // Allocate special inputs passed in VGPRs.
 void SITargetLowering::allocateSpecialEntryInputVGPRs(CCState &CCInfo,
                                                       MachineFunction &MF,
                                                       const SIRegisterInfo &TRI,
                                                       SIMachineFunctionInfo &Info) const {
   const LLT S32 = LLT::scalar(32);
   MachineRegisterInfo &MRI = MF.getRegInfo();
 
   if (Info.hasWorkItemIDX()) {
     Register Reg = AMDGPU::VGPR0;
     MRI.setType(MF.addLiveIn(Reg, &AMDGPU::VGPR_32RegClass), S32);
 
     CCInfo.AllocateReg(Reg);
     Info.setWorkItemIDX(ArgDescriptor::createRegister(Reg));
   }
 
   if (Info.hasWorkItemIDY()) {
     Register Reg = AMDGPU::VGPR1;
     MRI.setType(MF.addLiveIn(Reg, &AMDGPU::VGPR_32RegClass), S32);
 
     CCInfo.AllocateReg(Reg);
     Info.setWorkItemIDY(ArgDescriptor::createRegister(Reg));
   }
 
   if (Info.hasWorkItemIDZ()) {
     Register Reg = AMDGPU::VGPR2;
     MRI.setType(MF.addLiveIn(Reg, &AMDGPU::VGPR_32RegClass), S32);
 
     CCInfo.AllocateReg(Reg);
     Info.setWorkItemIDZ(ArgDescriptor::createRegister(Reg));
   }
 }
 
 // Try to allocate a VGPR at the end of the argument list, or if no argument
 // VGPRs are left allocating a stack slot.
 // If \p Mask is is given it indicates bitfield position in the register.
 // If \p Arg is given use it with new ]p Mask instead of allocating new.
 static ArgDescriptor allocateVGPR32Input(CCState &CCInfo, unsigned Mask = ~0u,
                                          ArgDescriptor Arg = ArgDescriptor()) {
   if (Arg.isSet())
     return ArgDescriptor::createArg(Arg, Mask);
 
   ArrayRef<MCPhysReg> ArgVGPRs
     = makeArrayRef(AMDGPU::VGPR_32RegClass.begin(), 32);
   unsigned RegIdx = CCInfo.getFirstUnallocated(ArgVGPRs);
   if (RegIdx == ArgVGPRs.size()) {
     // Spill to stack required.
     int64_t Offset = CCInfo.AllocateStack(4, Align(4));
 
     return ArgDescriptor::createStack(Offset, Mask);
   }
 
   unsigned Reg = ArgVGPRs[RegIdx];
   Reg = CCInfo.AllocateReg(Reg);
   assert(Reg != AMDGPU::NoRegister);
 
   MachineFunction &MF = CCInfo.getMachineFunction();
   Register LiveInVReg = MF.addLiveIn(Reg, &AMDGPU::VGPR_32RegClass);
   MF.getRegInfo().setType(LiveInVReg, LLT::scalar(32));
   return ArgDescriptor::createRegister(Reg, Mask);
 }
 
 static ArgDescriptor allocateSGPR32InputImpl(CCState &CCInfo,
                                              const TargetRegisterClass *RC,
                                              unsigned NumArgRegs) {
   ArrayRef<MCPhysReg> ArgSGPRs = makeArrayRef(RC->begin(), 32);
   unsigned RegIdx = CCInfo.getFirstUnallocated(ArgSGPRs);
   if (RegIdx == ArgSGPRs.size())
     report_fatal_error("ran out of SGPRs for arguments");
 
   unsigned Reg = ArgSGPRs[RegIdx];
   Reg = CCInfo.AllocateReg(Reg);
   assert(Reg != AMDGPU::NoRegister);
 
   MachineFunction &MF = CCInfo.getMachineFunction();
   MF.addLiveIn(Reg, RC);
   return ArgDescriptor::createRegister(Reg);
 }
 
 static ArgDescriptor allocateSGPR32Input(CCState &CCInfo) {
   return allocateSGPR32InputImpl(CCInfo, &AMDGPU::SGPR_32RegClass, 32);
 }
 
 static ArgDescriptor allocateSGPR64Input(CCState &CCInfo) {
   return allocateSGPR32InputImpl(CCInfo, &AMDGPU::SGPR_64RegClass, 16);
 }
 
 /// Allocate implicit function VGPR arguments at the end of allocated user
 /// arguments.
 void SITargetLowering::allocateSpecialInputVGPRs(
   CCState &CCInfo, MachineFunction &MF,
   const SIRegisterInfo &TRI, SIMachineFunctionInfo &Info) const {
   const unsigned Mask = 0x3ff;
   ArgDescriptor Arg;
 
   if (Info.hasWorkItemIDX()) {
     Arg = allocateVGPR32Input(CCInfo, Mask);
     Info.setWorkItemIDX(Arg);
   }
 
   if (Info.hasWorkItemIDY()) {
     Arg = allocateVGPR32Input(CCInfo, Mask << 10, Arg);
     Info.setWorkItemIDY(Arg);
   }
 
   if (Info.hasWorkItemIDZ())
     Info.setWorkItemIDZ(allocateVGPR32Input(CCInfo, Mask << 20, Arg));
 }
 
 /// Allocate implicit function VGPR arguments in fixed registers.
 void SITargetLowering::allocateSpecialInputVGPRsFixed(
   CCState &CCInfo, MachineFunction &MF,
   const SIRegisterInfo &TRI, SIMachineFunctionInfo &Info) const {
   Register Reg = CCInfo.AllocateReg(AMDGPU::VGPR31);
   if (!Reg)
     report_fatal_error("failed to allocated VGPR for implicit arguments");
 
   const unsigned Mask = 0x3ff;
   Info.setWorkItemIDX(ArgDescriptor::createRegister(Reg, Mask));
   Info.setWorkItemIDY(ArgDescriptor::createRegister(Reg, Mask << 10));
   Info.setWorkItemIDZ(ArgDescriptor::createRegister(Reg, Mask << 20));
 }
 
 void SITargetLowering::allocateSpecialInputSGPRs(
   CCState &CCInfo,
   MachineFunction &MF,
   const SIRegisterInfo &TRI,
   SIMachineFunctionInfo &Info) const {
   auto &ArgInfo = Info.getArgInfo();
 
   // TODO: Unify handling with private memory pointers.
 
   if (Info.hasDispatchPtr())
     ArgInfo.DispatchPtr = allocateSGPR64Input(CCInfo);
 
   if (Info.hasQueuePtr())
     ArgInfo.QueuePtr = allocateSGPR64Input(CCInfo);
 
   // Implicit arg ptr takes the place of the kernarg segment pointer. This is a
   // constant offset from the kernarg segment.
   if (Info.hasImplicitArgPtr())
     ArgInfo.ImplicitArgPtr = allocateSGPR64Input(CCInfo);
 
   if (Info.hasDispatchID())
     ArgInfo.DispatchID = allocateSGPR64Input(CCInfo);
 
   // flat_scratch_init is not applicable for non-kernel functions.
 
   if (Info.hasWorkGroupIDX())
     ArgInfo.WorkGroupIDX = allocateSGPR32Input(CCInfo);
 
   if (Info.hasWorkGroupIDY())
     ArgInfo.WorkGroupIDY = allocateSGPR32Input(CCInfo);
 
   if (Info.hasWorkGroupIDZ())
     ArgInfo.WorkGroupIDZ = allocateSGPR32Input(CCInfo);
 }
 
 // Allocate special inputs passed in user SGPRs.
 void SITargetLowering::allocateHSAUserSGPRs(CCState &CCInfo,
                                             MachineFunction &MF,
                                             const SIRegisterInfo &TRI,
                                             SIMachineFunctionInfo &Info) const {
   if (Info.hasImplicitBufferPtr()) {
     Register ImplicitBufferPtrReg = Info.addImplicitBufferPtr(TRI);
     MF.addLiveIn(ImplicitBufferPtrReg, &AMDGPU::SGPR_64RegClass);
     CCInfo.AllocateReg(ImplicitBufferPtrReg);
   }
 
   // FIXME: How should these inputs interact with inreg / custom SGPR inputs?
   if (Info.hasPrivateSegmentBuffer()) {
     Register PrivateSegmentBufferReg = Info.addPrivateSegmentBuffer(TRI);
     MF.addLiveIn(PrivateSegmentBufferReg, &AMDGPU::SGPR_128RegClass);
     CCInfo.AllocateReg(PrivateSegmentBufferReg);
   }
 
   if (Info.hasDispatchPtr()) {
     Register DispatchPtrReg = Info.addDispatchPtr(TRI);
     MF.addLiveIn(DispatchPtrReg, &AMDGPU::SGPR_64RegClass);
     CCInfo.AllocateReg(DispatchPtrReg);
   }
 
   if (Info.hasQueuePtr()) {
     Register QueuePtrReg = Info.addQueuePtr(TRI);
     MF.addLiveIn(QueuePtrReg, &AMDGPU::SGPR_64RegClass);
     CCInfo.AllocateReg(QueuePtrReg);
   }
 
   if (Info.hasKernargSegmentPtr()) {
     MachineRegisterInfo &MRI = MF.getRegInfo();
     Register InputPtrReg = Info.addKernargSegmentPtr(TRI);
     CCInfo.AllocateReg(InputPtrReg);
 
     Register VReg = MF.addLiveIn(InputPtrReg, &AMDGPU::SGPR_64RegClass);
     MRI.setType(VReg, LLT::pointer(AMDGPUAS::CONSTANT_ADDRESS, 64));
   }
 
   if (Info.hasDispatchID()) {
     Register DispatchIDReg = Info.addDispatchID(TRI);
     MF.addLiveIn(DispatchIDReg, &AMDGPU::SGPR_64RegClass);
     CCInfo.AllocateReg(DispatchIDReg);
   }
 
   if (Info.hasFlatScratchInit()) {
     Register FlatScratchInitReg = Info.addFlatScratchInit(TRI);
     MF.addLiveIn(FlatScratchInitReg, &AMDGPU::SGPR_64RegClass);
     CCInfo.AllocateReg(FlatScratchInitReg);
   }
 
   // TODO: Add GridWorkGroupCount user SGPRs when used. For now with HSA we read
   // these from the dispatch pointer.
 }
 
 // Allocate special input registers that are initialized per-wave.
 void SITargetLowering::allocateSystemSGPRs(CCState &CCInfo,
                                            MachineFunction &MF,
                                            SIMachineFunctionInfo &Info,
                                            CallingConv::ID CallConv,
                                            bool IsShader) const {
   if (Info.hasWorkGroupIDX()) {
     Register Reg = Info.addWorkGroupIDX();
     MF.addLiveIn(Reg, &AMDGPU::SGPR_32RegClass);
     CCInfo.AllocateReg(Reg);
   }
 
   if (Info.hasWorkGroupIDY()) {
     Register Reg = Info.addWorkGroupIDY();
     MF.addLiveIn(Reg, &AMDGPU::SGPR_32RegClass);
     CCInfo.AllocateReg(Reg);
   }
 
   if (Info.hasWorkGroupIDZ()) {
     Register Reg = Info.addWorkGroupIDZ();
     MF.addLiveIn(Reg, &AMDGPU::SGPR_32RegClass);
     CCInfo.AllocateReg(Reg);
   }
 
   if (Info.hasWorkGroupInfo()) {
     Register Reg = Info.addWorkGroupInfo();
     MF.addLiveIn(Reg, &AMDGPU::SGPR_32RegClass);
     CCInfo.AllocateReg(Reg);
   }
 
   if (Info.hasPrivateSegmentWaveByteOffset()) {
     // Scratch wave offset passed in system SGPR.
     unsigned PrivateSegmentWaveByteOffsetReg;
 
-    if (IsShader) {
+    if (IsShader && CallConv != CallingConv::AMDGPU_GFX_CALLABLE) {
       PrivateSegmentWaveByteOffsetReg =
         Info.getPrivateSegmentWaveByteOffsetSystemSGPR();
 
       // This is true if the scratch wave byte offset doesn't have a fixed
       // location.
       if (PrivateSegmentWaveByteOffsetReg == AMDGPU::NoRegister) {
         PrivateSegmentWaveByteOffsetReg = findFirstFreeSGPR(CCInfo);
         Info.setPrivateSegmentWaveByteOffset(PrivateSegmentWaveByteOffsetReg);
       }
     } else
       PrivateSegmentWaveByteOffsetReg = Info.addPrivateSegmentWaveByteOffset();
 
     MF.addLiveIn(PrivateSegmentWaveByteOffsetReg, &AMDGPU::SGPR_32RegClass);
     CCInfo.AllocateReg(PrivateSegmentWaveByteOffsetReg);
   }
 }
 
 static void reservePrivateMemoryRegs(const TargetMachine &TM,
                                      MachineFunction &MF,
                                      const SIRegisterInfo &TRI,
                                      SIMachineFunctionInfo &Info) {
   // Now that we've figured out where the scratch register inputs are, see if
   // should reserve the arguments and use them directly.
   MachineFrameInfo &MFI = MF.getFrameInfo();
   bool HasStackObjects = MFI.hasStackObjects();
   const GCNSubtarget &ST = MF.getSubtarget<GCNSubtarget>();
 
   // Record that we know we have non-spill stack objects so we don't need to
   // check all stack objects later.
   if (HasStackObjects)
     Info.setHasNonSpillStackObjects(true);
 
   // Everything live out of a block is spilled with fast regalloc, so it's
   // almost certain that spilling will be required.
   if (TM.getOptLevel() == CodeGenOpt::None)
     HasStackObjects = true;
 
   // For now assume stack access is needed in any callee functions, so we need
   // the scratch registers to pass in.
   bool RequiresStackAccess = HasStackObjects || MFI.hasCalls();
 
   if (RequiresStackAccess && ST.isAmdHsaOrMesa(MF.getFunction())) {
     // If we have stack objects, we unquestionably need the private buffer
     // resource. For the Code Object V2 ABI, this will be the first 4 user
     // SGPR inputs. We can reserve those and use them directly.
 
     Register PrivateSegmentBufferReg =
         Info.getPreloadedReg(AMDGPUFunctionArgInfo::PRIVATE_SEGMENT_BUFFER);
     Info.setScratchRSrcReg(PrivateSegmentBufferReg);
   } else {
     unsigned ReservedBufferReg = TRI.reservedPrivateSegmentBufferReg(MF);
     // We tentatively reserve the last registers (skipping the last registers
     // which may contain VCC, FLAT_SCR, and XNACK). After register allocation,
     // we'll replace these with the ones immediately after those which were
     // really allocated. In the prologue copies will be inserted from the
     // argument to these reserved registers.
 
     // Without HSA, relocations are used for the scratch pointer and the
     // buffer resource setup is always inserted in the prologue. Scratch wave
     // offset is still in an input SGPR.
     Info.setScratchRSrcReg(ReservedBufferReg);
   }
 
   MachineRegisterInfo &MRI = MF.getRegInfo();
 
   // For entry functions we have to set up the stack pointer if we use it,
   // whereas non-entry functions get this "for free". This means there is no
   // intrinsic advantage to using S32 over S34 in cases where we do not have
   // calls but do need a frame pointer (i.e. if we are requested to have one
   // because frame pointer elimination is disabled). To keep things simple we
   // only ever use S32 as the call ABI stack pointer, and so using it does not
   // imply we need a separate frame pointer.
   //
   // Try to use s32 as the SP, but move it if it would interfere with input
   // arguments. This won't work with calls though.
   //
   // FIXME: Move SP to avoid any possible inputs, or find a way to spill input
   // registers.
   if (!MRI.isLiveIn(AMDGPU::SGPR32)) {
     Info.setStackPtrOffsetReg(AMDGPU::SGPR32);
   } else {
     assert(AMDGPU::isShader(MF.getFunction().getCallingConv()));
 
     if (MFI.hasCalls())
       report_fatal_error("call in graphics shader with too many input SGPRs");
 
     for (unsigned Reg : AMDGPU::SGPR_32RegClass) {
       if (!MRI.isLiveIn(Reg)) {
         Info.setStackPtrOffsetReg(Reg);
         break;
       }
     }
 
     if (Info.getStackPtrOffsetReg() == AMDGPU::SP_REG)
       report_fatal_error("failed to find register for SP");
   }
 
   // hasFP should be accurate for entry functions even before the frame is
   // finalized, because it does not rely on the known stack size, only
   // properties like whether variable sized objects are present.
   if (ST.getFrameLowering()->hasFP(MF)) {
     Info.setFrameOffsetReg(AMDGPU::SGPR33);
   }
 }
 
 bool SITargetLowering::supportSplitCSR(MachineFunction *MF) const {
   const SIMachineFunctionInfo *Info = MF->getInfo<SIMachineFunctionInfo>();
   return !Info->isEntryFunction();
 }
 
 void SITargetLowering::initializeSplitCSR(MachineBasicBlock *Entry) const {
 
 }
 
 void SITargetLowering::insertCopiesSplitCSR(
   MachineBasicBlock *Entry,
   const SmallVectorImpl<MachineBasicBlock *> &Exits) const {
   const SIRegisterInfo *TRI = getSubtarget()->getRegisterInfo();
 
   const MCPhysReg *IStart = TRI->getCalleeSavedRegsViaCopy(Entry->getParent());
   if (!IStart)
     return;
 
   const TargetInstrInfo *TII = Subtarget->getInstrInfo();
   MachineRegisterInfo *MRI = &Entry->getParent()->getRegInfo();
   MachineBasicBlock::iterator MBBI = Entry->begin();
   for (const MCPhysReg *I = IStart; *I; ++I) {
     const TargetRegisterClass *RC = nullptr;
     if (AMDGPU::SReg_64RegClass.contains(*I))
       RC = &AMDGPU::SGPR_64RegClass;
     else if (AMDGPU::SReg_32RegClass.contains(*I))
       RC = &AMDGPU::SGPR_32RegClass;
     else
       llvm_unreachable("Unexpected register class in CSRsViaCopy!");
 
     Register NewVR = MRI->createVirtualRegister(RC);
     // Create copy from CSR to a virtual register.
     Entry->addLiveIn(*I);
     BuildMI(*Entry, MBBI, DebugLoc(), TII->get(TargetOpcode::COPY), NewVR)
       .addReg(*I);
 
     // Insert the copy-back instructions right before the terminator.
     for (auto *Exit : Exits)
       BuildMI(*Exit, Exit->getFirstTerminator(), DebugLoc(),
               TII->get(TargetOpcode::COPY), *I)
         .addReg(NewVR);
   }
 }
 
 SDValue SITargetLowering::LowerFormalArguments(
     SDValue Chain, CallingConv::ID CallConv, bool isVarArg,
     const SmallVectorImpl<ISD::InputArg> &Ins, const SDLoc &DL,
     SelectionDAG &DAG, SmallVectorImpl<SDValue> &InVals) const {
   const SIRegisterInfo *TRI = getSubtarget()->getRegisterInfo();
 
   MachineFunction &MF = DAG.getMachineFunction();
   const Function &Fn = MF.getFunction();
   FunctionType *FType = MF.getFunction().getFunctionType();
   SIMachineFunctionInfo *Info = MF.getInfo<SIMachineFunctionInfo>();
 
   if (Subtarget->isAmdHsaOS() && AMDGPU::isShader(CallConv)) {
     DiagnosticInfoUnsupported NoGraphicsHSA(
         Fn, "unsupported non-compute shaders with HSA", DL.getDebugLoc());
     DAG.getContext()->diagnose(NoGraphicsHSA);
     return DAG.getEntryNode();
   }
 
   SmallVector<ISD::InputArg, 16> Splits;
   SmallVector<CCValAssign, 16> ArgLocs;
   BitVector Skipped(Ins.size());
   CCState CCInfo(CallConv, isVarArg, DAG.getMachineFunction(), ArgLocs,
                  *DAG.getContext());
 
   bool IsShader = AMDGPU::isShader(CallConv);
   bool IsKernel = AMDGPU::isKernel(CallConv);
   bool IsEntryFunc = AMDGPU::isEntryFunctionCC(CallConv);
 
   if (IsShader) {
     processShaderInputArgs(Splits, CallConv, Ins, Skipped, FType, Info);
 
     // At least one interpolation mode must be enabled or else the GPU will
     // hang.
     //
     // Check PSInputAddr instead of PSInputEnable. The idea is that if the user
     // set PSInputAddr, the user wants to enable some bits after the compilation
     // based on run-time states. Since we can't know what the final PSInputEna
     // will look like, so we shouldn't do anything here and the user should take
     // responsibility for the correct programming.
SITargetLowering::LowerReturn(SDValue Chain, CallingConv::ID CallConv,
   SIMachineFunctionInfo *Info = MF.getInfo<SIMachineFunctionInfo>();
 
   if (AMDGPU::isKernel(CallConv)) {
     return AMDGPUTargetLowering::LowerReturn(Chain, CallConv, isVarArg, Outs,
                                              OutVals, DL, DAG);
   }
 
   bool IsShader = AMDGPU::isShader(CallConv);
+  bool IsCallableShader = CallConv == CallingConv::AMDGPU_GFX_CALLABLE;
 
   Info->setIfReturnsVoid(Outs.empty());
-  bool IsWaveEnd = Info->returnsVoid() && IsShader;
+  bool IsWaveEnd = Info->returnsVoid() && IsShader && !IsCallableShader;
 
   // CCValAssign - represent the assignment of the return value to a location.
   SmallVector<CCValAssign, 48> RVLocs;
   SmallVector<ISD::OutputArg, 48> Splits;
 
   // CCState - Info about the registers and stack slots.
   CCState CCInfo(CallConv, isVarArg, DAG.getMachineFunction(), RVLocs,
                  *DAG.getContext());
SITargetLowering::LowerReturn(SDValue Chain, CallingConv::ID CallConv,
 
   // Update chain and glue.
   RetOps[0] = Chain;
   if (Flag.getNode())
     RetOps.push_back(Flag);
 
   unsigned Opc = AMDGPUISD::ENDPGM;
   if (!IsWaveEnd)
-    Opc = IsShader ? AMDGPUISD::RETURN_TO_EPILOG : AMDGPUISD::RET_FLAG;
+    Opc = IsShader && !IsCallableShader ? AMDGPUISD::RETURN_TO_EPILOG
+                                        : AMDGPUISD::RET_FLAG;
   return DAG.getNode(Opc, DL, MVT::Other, RetOps);
 }
 
 SDValue SITargetLowering::LowerCallResult(
     SDValue Chain, SDValue InFlag, CallingConv::ID CallConv, bool IsVarArg,
     const SmallVectorImpl<ISD::InputArg> &Ins, const SDLoc &DL,
     SelectionDAG &DAG, SmallVectorImpl<SDValue> &InVals, bool IsThisReturn,
     SDValue ThisVal) const {
for (auto InputID : InputRegs) {
     assert(IncomingArgRC == ArgRC);
 
     // All special arguments are ints for now.
     EVT ArgVT = TRI->getSpillSize(*ArgRC) == 8 ? MVT::i64 : MVT::i32;
     SDValue InputReg;
 
     if (IncomingArg) {
       InputReg = loadInputValue(DAG, ArgRC, ArgVT, DL, *IncomingArg);
-    } else {
+    } else if (InputID == AMDGPUFunctionArgInfo::IMPLICIT_ARG_PTR) {
       // The implicit arg ptr is special because it doesn't have a corresponding
       // input for kernels, and is computed from the kernarg segment pointer.
-      assert(InputID == AMDGPUFunctionArgInfo::IMPLICIT_ARG_PTR);
       InputReg = getImplicitArgPtr(DAG, DL);
+    } else {
+      // Argument is not passed to the function
+      continue;
     }
 
     if (OutgoingArg->isRegister()) {
       RegsToPass.emplace_back(OutgoingArg->getRegister(), InputReg);
       if (!CCInfo.AllocateReg(OutgoingArg->getRegister()))
         report_fatal_error("failed to allocate implicit input argument");
     } else {
       unsigned SpecialArgOffset =
if (IsVarArg) {
     return lowerUnhandledCall(CLI, InVals,
                               "unsupported call to variadic function ");
   }
 
   if (!CLI.CB)
     report_fatal_error("unsupported libcall legalization");
 
   if (!AMDGPUTargetMachine::EnableFixedFunctionABI &&
-      !CLI.CB->getCalledFunction()) {
+      !CLI.CB->getCalledFunction() &&
+      CallConv != CallingConv::AMDGPU_GFX_CALLABLE) {
     return lowerUnhandledCall(CLI, InVals,
                               "unsupported indirect call to function ");
   }
 
   if (IsTailCall && MF.getTarget().Options.GuaranteedTailCallOpt) {
     return lowerUnhandledCall(CLI, InVals,
                               "unsupported required tail call to function ");
   }
 
-  if (AMDGPU::isShader(MF.getFunction().getCallingConv())) {
+  if (AMDGPU::isShader(CallConv) &&
+      CallConv != CallingConv::AMDGPU_GFX_CALLABLE) {
     // Note the issue is with the CC of the calling function, not of the call
     // itself.
     return lowerUnhandledCall(CLI, InVals,
                           "unsupported call from graphics shader of function ");
   }
 
   if (IsTailCall) {
     IsTailCall = isEligibleForTailCallOptimization(
       Callee, CallConv, IsVarArg, Outs, OutVals, Ins, DAG);
     if (!IsTailCall && CLI.CB && CLI.CB->isMustTailCall()) {
       report_fatal_error("failed to perform tail call elimination on a call "
                          "site marked musttail");
     }
 
     bool TailCallOpt = MF.getTarget().Options.GuaranteedTailCallOpt;
 
     // A sibling call is one where we're under the usual C ABI and not planning
     // to change that but can still do a tail call:
     if (!TailCallOpt && IsTailCall)
       IsSibCall = true;
 
if (VA.isRegLoc()) {
       } else {
         SDValue Store =
             DAG.getStore(Chain, DL, Arg, DstAddr, DstInfo, Alignment);
         MemOpChains.push_back(Store);
       }
     }
   }
 
-  if (!AMDGPUTargetMachine::EnableFixedFunctionABI) {
+  if (!AMDGPUTargetMachine::EnableFixedFunctionABI &&
+      !Subtarget->isAmdPalOS()) {
     // Copy special input registers after user input arguments.
     passSpecialInputs(CLI, CCInfo, *Info, RegsToPass, MemOpChains, Chain);
   }
 
   if (!MemOpChains.empty())
     Chain = DAG.getNode(ISD::TokenFactor, DL, MVT::Other, MemOpChains);
 
   // Build a sequence of copy-to-reg nodes chained together with token chain

llvm/lib/Target/AMDGPU/SILowerSGPRSpills.cpp

static bool lowerShiftReservedVGPR(MachineFunction &MF,
 
   assert(ReservedVGPRInfoItr != FuncInfo->getSGPRSpillVGPRs().end());
   auto Index =
       std::distance(FuncInfo->getSGPRSpillVGPRs().begin(), ReservedVGPRInfoItr);
 
   FuncInfo->setSGPRSpillVGPRs(LowestAvailableVGPR, FI, Index);
 
   for (MachineBasicBlock &MBB : MF) {
+    assert(LowestAvailableVGPR.isValid() && "Did not find an available VGPR");
     MBB.addLiveIn(LowestAvailableVGPR);
     MBB.sortUniqueLiveIns();
   }
 
   return true;
 }
 
 bool SILowerSGPRSpills::runOnMachineFunction(MachineFunction &MF) {

llvm/lib/Target/AMDGPU/SIRegisterInfo.cpp

 // Forced to be here by one .inc
 const MCPhysReg *SIRegisterInfo::getCalleeSavedRegs(
   const MachineFunction *MF) const {
   CallingConv::ID CC = MF->getFunction().getCallingConv();
   switch (CC) {
   case CallingConv::C:
   case CallingConv::Fast:
   case CallingConv::Cold:
+  case CallingConv::AMDGPU_GFX_CALLABLE:
     return CSR_AMDGPU_HighRegs_SaveList;
   default: {
     // Dummy to not crash RegisterClassInfo.
     static const MCPhysReg NoCalleeSavedReg = AMDGPU::NoRegister;
     return &NoCalleeSavedReg;
   }
   }
 }
 
 const MCPhysReg *
 SIRegisterInfo::getCalleeSavedRegsViaCopy(const MachineFunction *MF) const {
   return nullptr;
 }
 
 const uint32_t *SIRegisterInfo::getCallPreservedMask(const MachineFunction &MF,
                                                      CallingConv::ID CC) const {
   switch (CC) {
   case CallingConv::C:
   case CallingConv::Fast:
   case CallingConv::Cold:
+  case CallingConv::AMDGPU_GFX_CALLABLE:
     return CSR_AMDGPU_HighRegs_RegMask;
   default:
     return nullptr;
   }
 }
 
 Register SIRegisterInfo::getFrameRegister(const MachineFunction &MF) const {
   const SIFrameLowering *TFI =

llvm/lib/Target/AMDGPU/Utils/AMDGPUBaseInfo.cpp

bool isShader(CallingConv::ID cc) {
   switch(cc) {
     case CallingConv::AMDGPU_VS:
     case CallingConv::AMDGPU_LS:
     case CallingConv::AMDGPU_HS:
     case CallingConv::AMDGPU_ES:
     case CallingConv::AMDGPU_GS:
     case CallingConv::AMDGPU_PS:
     case CallingConv::AMDGPU_CS:
+    case CallingConv::AMDGPU_GFX_CALLABLE:
       return true;
     default:
       return false;
   }
 }
 
 bool isCompute(CallingConv::ID cc) {
   return !isShader(cc) || cc == CallingConv::AMDGPU_CS;
case CallingConv::SPIR_KERNEL:
     return true;
   case CallingConv::AMDGPU_VS:
   case CallingConv::AMDGPU_LS:
   case CallingConv::AMDGPU_HS:
   case CallingConv::AMDGPU_ES:
   case CallingConv::AMDGPU_GS:
   case CallingConv::AMDGPU_PS:
   case CallingConv::AMDGPU_CS:
+  case CallingConv::AMDGPU_GFX_CALLABLE:
     // For non-compute shaders, SGPR inputs are marked with either inreg or byval.
     // Everything else is in VGPRs.
     return F->getAttributes().hasParamAttribute(A->getArgNo(), Attribute::InReg) ||
            F->getAttributes().hasParamAttribute(A->getArgNo(), Attribute::ByVal);
   default:
     // TODO: Should calls support inreg for SGPR inputs?
     return false;
   }

llvm/lib/Target/AMDGPU/Utils/AMDGPUPALMetadata.cpp

static const char *getStageName(CallingConv::ID CC) {
   case CallingConv::AMDGPU_GS:
     return ".gs";
   case CallingConv::AMDGPU_ES:
     return ".es";
   case CallingConv::AMDGPU_HS:
     return ".hs";
   case CallingConv::AMDGPU_LS:
     return ".ls";
+  case CallingConv::AMDGPU_GFX_CALLABLE:
+    assert(false && "Callable shader has no hardware stage");
   default:
     return ".cs";
   }
 }
 
 // Get (create if necessary) the .hardware_stages entry for the given calling
 // convention.
 msgpack::MapDocNode AMDGPUPALMetadata::getHwStage(unsigned CC) {

llvm/test/Bitcode/compatibility.ll

 declare cc89 void @f.cc89()
 ; CHECK: declare amdgpu_ps void @f.cc89()
 declare amdgpu_ps void @f.amdgpu_ps()
 ; CHECK: declare amdgpu_ps void @f.amdgpu_ps()
 declare cc90 void @f.cc90()
 ; CHECK: declare amdgpu_cs void @f.cc90()
 declare amdgpu_cs void @f.amdgpu_cs()
 ; CHECK: declare amdgpu_cs void @f.amdgpu_cs()
+declare amdgpu_gfx_callable void @f.amdgpu_gfx_callable()
+; CHECK: declare amdgpu_gfx_callable void @f.amdgpu_gfx_callable()
 declare cc91 void @f.cc91()
 ; CHECK: declare amdgpu_kernel void @f.cc91()
 declare amdgpu_kernel void @f.amdgpu_kernel()
 ; CHECK: declare amdgpu_kernel void @f.amdgpu_kernel()
 declare cc93 void @f.cc93()
 ; CHECK: declare amdgpu_hs void @f.cc93()
 declare amdgpu_hs void @f.amdgpu_hs()
 ; CHECK: declare amdgpu_hs void @f.amdgpu_hs()

llvm/test/CodeGen/AMDGPU/amdpal-callable.ll

+; RUN: llc -mtriple=amdgcn--amdpal -verify-machineinstrs < %s | FileCheck -check-prefix=GCN -check-prefix=SI -enable-var-scope %s
+; RUN: llc -mtriple=amdgcn--amdpal -mcpu=tonga -verify-machineinstrs < %s | FileCheck -check-prefix=GCN -check-prefix=VI -enable-var-scope %s
+; RUN: llc -mtriple=amdgcn--amdpal -mcpu=gfx900 -verify-machineinstrs < %s | FileCheck -check-prefix=GCN -check-prefix=GFX9 -enable-var-scope %s
+
+; GCN-LABEL: {{^}}gfx_callable_amdpal:
+; GCN: .amd_amdgpu_pal_metadata {{$}}
+define amdgpu_gfx_callable half @gfx_callable_amdpal(half %arg0) {
+  %add = fadd half %arg0, 1.0
+  ret half %add
+}

llvm/test/CodeGen/AMDGPU/gfx-callable-argument-types.ll

+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc -mtriple=amdgcn--amdpal -mcpu=gfx900 -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN,GFX9 %s
+; RUN: llc -mtriple=amdgcn--amdpal -mcpu=gfx1010 -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN,GFX10 %s
+
+declare hidden amdgpu_gfx_callable void @external_void_func_i1(i1) #0
+declare hidden amdgpu_gfx_callable void @external_void_func_i1_signext(i1 signext) #0
+declare hidden amdgpu_gfx_callable void @external_void_func_i1_zeroext(i1 zeroext) #0
+
+declare hidden amdgpu_gfx_callable void @external_void_func_i8(i8) #0
+declare hidden amdgpu_gfx_callable void @external_void_func_i8_signext(i8 signext) #0
+declare hidden amdgpu_gfx_callable void @external_void_func_i8_zeroext(i8 zeroext) #0
+
+declare hidden amdgpu_gfx_callable void @external_void_func_i16(i16) #0
+declare hidden amdgpu_gfx_callable void @external_void_func_i16_signext(i16 signext) #0
+declare hidden amdgpu_gfx_callable void @external_void_func_i16_zeroext(i16 zeroext) #0
+
+declare hidden amdgpu_gfx_callable void @external_void_func_i32(i32) #0
+declare hidden amdgpu_gfx_callable void @external_void_func_i64(i64) #0
+declare hidden amdgpu_gfx_callable void @external_void_func_v2i64(<2 x i64>) #0
+declare hidden amdgpu_gfx_callable void @external_void_func_v3i64(<3 x i64>) #0
+declare hidden amdgpu_gfx_callable void @external_void_func_v4i64(<4 x i64>) #0
+
+declare hidden amdgpu_gfx_callable void @external_void_func_f16(half) #0
+declare hidden amdgpu_gfx_callable void @external_void_func_f32(float) #0
+declare hidden amdgpu_gfx_callable void @external_void_func_f64(double) #0
+declare hidden amdgpu_gfx_callable void @external_void_func_v2f32(<2 x float>) #0
+declare hidden amdgpu_gfx_callable void @external_void_func_v2f64(<2 x double>) #0
+declare hidden amdgpu_gfx_callable void @external_void_func_v3f32(<3 x float>) #0
+declare hidden amdgpu_gfx_callable void @external_void_func_v3f64(<3 x double>) #0
+declare hidden amdgpu_gfx_callable void @external_void_func_v5f32(<5 x float>) #0
+
+declare hidden amdgpu_gfx_callable void @external_void_func_v2i16(<2 x i16>) #0
+declare hidden amdgpu_gfx_callable void @external_void_func_v2f16(<2 x half>) #0
+declare hidden amdgpu_gfx_callable void @external_void_func_v3i16(<3 x i16>) #0
+declare hidden amdgpu_gfx_callable void @external_void_func_v3f16(<3 x half>) #0
+declare hidden amdgpu_gfx_callable void @external_void_func_v4i16(<4 x i16>) #0
+declare hidden amdgpu_gfx_callable void @external_void_func_v4f16(<4 x half>) #0
+
+declare hidden amdgpu_gfx_callable void @external_void_func_v2i32(<2 x i32>) #0
+declare hidden amdgpu_gfx_callable void @external_void_func_v3i32(<3 x i32>) #0
+declare hidden amdgpu_gfx_callable void @external_void_func_v3i32_i32(<3 x i32>, i32) #0
+declare hidden amdgpu_gfx_callable void @external_void_func_v4i32(<4 x i32>) #0
+declare hidden amdgpu_gfx_callable void @external_void_func_v5i32(<5 x i32>) #0
+declare hidden amdgpu_gfx_callable void @external_void_func_v8i32(<8 x i32>) #0
+declare hidden amdgpu_gfx_callable void @external_void_func_v16i32(<16 x i32>) #0
+declare hidden amdgpu_gfx_callable void @external_void_func_v32i32(<32 x i32>) #0
+declare hidden amdgpu_gfx_callable void @external_void_func_v32i32_i32(<32 x i32>, i32) #0
+
+declare hidden amdgpu_gfx_callable void @external_void_func_i1_inreg(i1 inreg) #0
+declare hidden amdgpu_gfx_callable void @external_void_func_i8_inreg(i8 inreg) #0
+declare hidden amdgpu_gfx_callable void @external_void_func_i16_inreg(i16 inreg) #0
+declare hidden amdgpu_gfx_callable void @external_void_func_i32_inreg(i32 inreg) #0
+declare hidden amdgpu_gfx_callable void @external_void_func_i64_inreg(i64 inreg) #0
+declare hidden amdgpu_gfx_callable void @external_void_func_v2i64_inreg(<2 x i64> inreg) #0
+declare hidden amdgpu_gfx_callable void @external_void_func_v3i64_inreg(<3 x i64> inreg) #0
+declare hidden amdgpu_gfx_callable void @external_void_func_v4i64_inreg(<4 x i64> inreg) #0
+
+declare hidden amdgpu_gfx_callable void @external_void_func_f16_inreg(half inreg) #0
+declare hidden amdgpu_gfx_callable void @external_void_func_f32_inreg(float inreg) #0
+declare hidden amdgpu_gfx_callable void @external_void_func_f64_inreg(double inreg) #0
+declare hidden amdgpu_gfx_callable void @external_void_func_v2f32_inreg(<2 x float> inreg) #0
+declare hidden amdgpu_gfx_callable void @external_void_func_v2f64_inreg(<2 x double> inreg) #0
+declare hidden amdgpu_gfx_callable void @external_void_func_v3f32_inreg(<3 x float> inreg) #0
+declare hidden amdgpu_gfx_callable void @external_void_func_v3f64_inreg(<3 x double> inreg) #0
+declare hidden amdgpu_gfx_callable void @external_void_func_v5f32_inreg(<5 x float> inreg) #0
+
+declare hidden amdgpu_gfx_callable void @external_void_func_v2i16_inreg(<2 x i16> inreg) #0
+declare hidden amdgpu_gfx_callable void @external_void_func_v2f16_inreg(<2 x half> inreg) #0
+declare hidden amdgpu_gfx_callable void @external_void_func_v3i16_inreg(<3 x i16> inreg) #0
+declare hidden amdgpu_gfx_callable void @external_void_func_v3f16_inreg(<3 x half> inreg) #0
+declare hidden amdgpu_gfx_callable void @external_void_func_v4i16_inreg(<4 x i16> inreg) #0
+declare hidden amdgpu_gfx_callable void @external_void_func_v4f16_inreg(<4 x half> inreg) #0
+
+declare hidden amdgpu_gfx_callable void @external_void_func_v2i32_inreg(<2 x i32> inreg) #0
+declare hidden amdgpu_gfx_callable void @external_void_func_v3i32_inreg(<3 x i32> inreg) #0
+declare hidden amdgpu_gfx_callable void @external_void_func_v3i32_i32_inreg(<3 x i32> inreg, i32 inreg) #0
+declare hidden amdgpu_gfx_callable void @external_void_func_v4i32_inreg(<4 x i32> inreg) #0
+declare hidden amdgpu_gfx_callable void @external_void_func_v5i32_inreg(<5 x i32> inreg) #0
+declare hidden amdgpu_gfx_callable void @external_void_func_v8i32_inreg(<8 x i32> inreg) #0
+declare hidden amdgpu_gfx_callable void @external_void_func_v16i32_inreg(<16 x i32> inreg) #0
+declare hidden amdgpu_gfx_callable void @external_void_func_v32i32_inreg(<32 x i32> inreg) #0
+declare hidden amdgpu_gfx_callable void @external_void_func_v32i32_i32_inreg(<32 x i32> inreg, i32 inreg) #0
+
+; return value and argument
+declare hidden amdgpu_gfx_callable i32 @external_i32_func_i32(i32) #0
+
+; Structs
+declare hidden amdgpu_gfx_callable void @external_void_func_struct_i8_i32({ i8, i32 }) #0
+declare hidden amdgpu_gfx_callable void @external_void_func_byval_struct_i8_i32({ i8, i32 } addrspace(5)* byval) #0
+declare hidden amdgpu_gfx_callable void @external_void_func_sret_struct_i8_i32_byval_struct_i8_i32({ i8, i32 } addrspace(5)* sret, { i8, i32 } addrspace(5)* byval) #0
+
+declare hidden amdgpu_gfx_callable void @external_void_func_v16i8(<16 x i8>) #0
+
+define amdgpu_gfx_callable void @test_call_external_void_func_i1_imm() #0 {
+; GFX9-LABEL: test_call_external_void_func_i1_imm:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    s_or_saveexec_b64 s[4:5], -1
+; GFX9-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX9-NEXT:    s_mov_b64 exec, s[4:5]
+; GFX9-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX9-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX9-NEXT:    s_mov_b32 s33, s32
+; GFX9-NEXT:    s_add_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_mov_b32_e32 v0, 1
+; GFX9-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX9-NEXT:    s_getpc_b64 s[4:5]
+; GFX9-NEXT:    s_add_u32 s4, s4, external_void_func_i1@rel32@lo+4
+; GFX9-NEXT:    s_addc_u32 s5, s5, external_void_func_i1@rel32@hi+12
+; GFX9-NEXT:    buffer_store_byte v0, off, s[0:3], s32
+; GFX9-NEXT:    s_swappc_b64 s[30:31], s[4:5]
+; GFX9-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX9-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX9-NEXT:    s_sub_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX9-NEXT:    s_or_saveexec_b64 s[6:7], -1
+; GFX9-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX9-NEXT:    s_mov_b64 exec, s[6:7]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    s_setpc_b64 s[4:5]
+;
+; GFX10-LABEL: test_call_external_void_func_i1_imm:
+; GFX10:       ; %bb.0:
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    s_or_saveexec_b32 s4, -1
+; GFX10-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s4
+; GFX10-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX10-NEXT:    v_mov_b32_e32 v0, 1
+; GFX10-NEXT:    s_mov_b32 s33, s32
+; GFX10-NEXT:    s_add_u32 s32, s32, 0x200
+; GFX10-NEXT:    s_getpc_b64 s[4:5]
+; GFX10-NEXT:    s_add_u32 s4, s4, external_void_func_i1@rel32@lo+4
+; GFX10-NEXT:    s_addc_u32 s5, s5, external_void_func_i1@rel32@hi+12
+; GFX10-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX10-NEXT:    buffer_store_byte v0, off, s[0:3], s32
+; GFX10-NEXT:    s_swappc_b64 s[30:31], s[4:5]
+; GFX10-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX10-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX10-NEXT:    s_sub_u32 s32, s32, 0x200
+; GFX10-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX10-NEXT:    s_or_saveexec_b32 s6, -1
+; GFX10-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s6
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    s_setpc_b64 s[4:5]
+  call amdgpu_gfx_callable void @external_void_func_i1(i1 true)
+  ret void
+}
+
+define amdgpu_gfx_callable void @test_call_external_void_func_i1_signext(i32) #0 {
+; GFX9-LABEL: test_call_external_void_func_i1_signext:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    s_or_saveexec_b64 s[4:5], -1
+; GFX9-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX9-NEXT:    s_mov_b64 exec, s[4:5]
+; GFX9-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX9-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX9-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX9-NEXT:    global_load_ubyte v0, v[0:1], off
+; GFX9-NEXT:    s_mov_b32 s33, s32
+; GFX9-NEXT:    s_add_u32 s32, s32, 0x400
+; GFX9-NEXT:    s_getpc_b64 s[4:5]
+; GFX9-NEXT:    s_add_u32 s4, s4, external_void_func_i1_signext@rel32@lo+4
+; GFX9-NEXT:    s_addc_u32 s5, s5, external_void_func_i1_signext@rel32@hi+12
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    v_and_b32_e32 v0, 1, v0
+; GFX9-NEXT:    buffer_store_byte v0, off, s[0:3], s32
+; GFX9-NEXT:    s_swappc_b64 s[30:31], s[4:5]
+; GFX9-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX9-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX9-NEXT:    s_sub_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX9-NEXT:    s_or_saveexec_b64 s[6:7], -1
+; GFX9-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX9-NEXT:    s_mov_b64 exec, s[6:7]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    s_setpc_b64 s[4:5]
+;
+; GFX10-LABEL: test_call_external_void_func_i1_signext:
+; GFX10:       ; %bb.0:
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    s_or_saveexec_b32 s4, -1
+; GFX10-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s4
+; GFX10-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX10-NEXT:    s_mov_b32 s33, s32
+; GFX10-NEXT:    s_add_u32 s32, s32, 0x200
+; GFX10-NEXT:    s_getpc_b64 s[4:5]
+; GFX10-NEXT:    s_add_u32 s4, s4, external_void_func_i1_signext@rel32@lo+4
+; GFX10-NEXT:    s_addc_u32 s5, s5, external_void_func_i1_signext@rel32@hi+12
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX10-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX10-NEXT:    global_load_ubyte v0, v[0:1], off
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    v_and_b32_e32 v0, 1, v0
+; GFX10-NEXT:    buffer_store_byte v0, off, s[0:3], s32
+; GFX10-NEXT:    s_swappc_b64 s[30:31], s[4:5]
+; GFX10-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX10-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX10-NEXT:    s_sub_u32 s32, s32, 0x200
+; GFX10-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX10-NEXT:    s_or_saveexec_b32 s6, -1
+; GFX10-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s6
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    s_setpc_b64 s[4:5]
+  %var = load volatile i1, i1 addrspace(1)* undef
+  call amdgpu_gfx_callable void @external_void_func_i1_signext(i1 %var)
+  ret void
+}
+
+define amdgpu_gfx_callable void @test_call_external_void_func_i1_zeroext(i32) #0 {
+; GFX9-LABEL: test_call_external_void_func_i1_zeroext:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    s_or_saveexec_b64 s[4:5], -1
+; GFX9-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX9-NEXT:    s_mov_b64 exec, s[4:5]
+; GFX9-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX9-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX9-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX9-NEXT:    global_load_ubyte v0, v[0:1], off
+; GFX9-NEXT:    s_mov_b32 s33, s32
+; GFX9-NEXT:    s_add_u32 s32, s32, 0x400
+; GFX9-NEXT:    s_getpc_b64 s[4:5]
+; GFX9-NEXT:    s_add_u32 s4, s4, external_void_func_i1_zeroext@rel32@lo+4
+; GFX9-NEXT:    s_addc_u32 s5, s5, external_void_func_i1_zeroext@rel32@hi+12
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    v_and_b32_e32 v0, 1, v0
+; GFX9-NEXT:    buffer_store_byte v0, off, s[0:3], s32
+; GFX9-NEXT:    s_swappc_b64 s[30:31], s[4:5]
+; GFX9-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX9-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX9-NEXT:    s_sub_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX9-NEXT:    s_or_saveexec_b64 s[6:7], -1
+; GFX9-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX9-NEXT:    s_mov_b64 exec, s[6:7]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    s_setpc_b64 s[4:5]
+;
+; GFX10-LABEL: test_call_external_void_func_i1_zeroext:
+; GFX10:       ; %bb.0:
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    s_or_saveexec_b32 s4, -1
+; GFX10-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s4
+; GFX10-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX10-NEXT:    s_mov_b32 s33, s32
+; GFX10-NEXT:    s_add_u32 s32, s32, 0x200
+; GFX10-NEXT:    s_getpc_b64 s[4:5]
+; GFX10-NEXT:    s_add_u32 s4, s4, external_void_func_i1_zeroext@rel32@lo+4
+; GFX10-NEXT:    s_addc_u32 s5, s5, external_void_func_i1_zeroext@rel32@hi+12
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX10-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX10-NEXT:    global_load_ubyte v0, v[0:1], off
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    v_and_b32_e32 v0, 1, v0
+; GFX10-NEXT:    buffer_store_byte v0, off, s[0:3], s32
+; GFX10-NEXT:    s_swappc_b64 s[30:31], s[4:5]
+; GFX10-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX10-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX10-NEXT:    s_sub_u32 s32, s32, 0x200
+; GFX10-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX10-NEXT:    s_or_saveexec_b32 s6, -1
+; GFX10-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s6
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    s_setpc_b64 s[4:5]
+  %var = load volatile i1, i1 addrspace(1)* undef
+  call amdgpu_gfx_callable void @external_void_func_i1_zeroext(i1 %var)
+  ret void
+}
+
+define amdgpu_gfx_callable void @test_call_external_void_func_i8_imm(i32) #0 {
+; GFX9-LABEL: test_call_external_void_func_i8_imm:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    s_or_saveexec_b64 s[4:5], -1
+; GFX9-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX9-NEXT:    s_mov_b64 exec, s[4:5]
+; GFX9-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX9-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX9-NEXT:    s_mov_b32 s33, s32
+; GFX9-NEXT:    s_add_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_mov_b32_e32 v0, 0x7b
+; GFX9-NEXT:    s_getpc_b64 s[4:5]
+; GFX9-NEXT:    s_add_u32 s4, s4, external_void_func_i8@rel32@lo+4
+; GFX9-NEXT:    s_addc_u32 s5, s5, external_void_func_i8@rel32@hi+12
+; GFX9-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX9-NEXT:    s_swappc_b64 s[30:31], s[4:5]
+; GFX9-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX9-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX9-NEXT:    s_sub_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX9-NEXT:    s_or_saveexec_b64 s[6:7], -1
+; GFX9-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX9-NEXT:    s_mov_b64 exec, s[6:7]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    s_setpc_b64 s[4:5]
+;
+; GFX10-LABEL: test_call_external_void_func_i8_imm:
+; GFX10:       ; %bb.0:
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    s_or_saveexec_b32 s4, -1
+; GFX10-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s4
+; GFX10-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX10-NEXT:    v_mov_b32_e32 v0, 0x7b
+; GFX10-NEXT:    s_mov_b32 s33, s32
+; GFX10-NEXT:    s_add_u32 s32, s32, 0x200
+; GFX10-NEXT:    s_getpc_b64 s[4:5]
+; GFX10-NEXT:    s_add_u32 s4, s4, external_void_func_i8@rel32@lo+4
+; GFX10-NEXT:    s_addc_u32 s5, s5, external_void_func_i8@rel32@hi+12
+; GFX10-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX10-NEXT:    s_swappc_b64 s[30:31], s[4:5]
+; GFX10-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX10-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX10-NEXT:    s_sub_u32 s32, s32, 0x200
+; GFX10-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX10-NEXT:    s_or_saveexec_b32 s6, -1
+; GFX10-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s6
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    s_setpc_b64 s[4:5]
+  call amdgpu_gfx_callable void @external_void_func_i8(i8 123)
+  ret void
+}
+
+define amdgpu_gfx_callable void @test_call_external_void_func_i8_signext(i32) #0 {
+; GFX9-LABEL: test_call_external_void_func_i8_signext:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    s_or_saveexec_b64 s[4:5], -1
+; GFX9-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX9-NEXT:    s_mov_b64 exec, s[4:5]
+; GFX9-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX9-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX9-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX9-NEXT:    global_load_sbyte v0, v[0:1], off
+; GFX9-NEXT:    s_mov_b32 s33, s32
+; GFX9-NEXT:    s_add_u32 s32, s32, 0x400
+; GFX9-NEXT:    s_getpc_b64 s[4:5]
+; GFX9-NEXT:    s_add_u32 s4, s4, external_void_func_i8_signext@rel32@lo+4
+; GFX9-NEXT:    s_addc_u32 s5, s5, external_void_func_i8_signext@rel32@hi+12
+; GFX9-NEXT:    s_swappc_b64 s[30:31], s[4:5]
+; GFX9-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX9-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX9-NEXT:    s_sub_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX9-NEXT:    s_or_saveexec_b64 s[6:7], -1
+; GFX9-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX9-NEXT:    s_mov_b64 exec, s[6:7]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    s_setpc_b64 s[4:5]
+;
+; GFX10-LABEL: test_call_external_void_func_i8_signext:
+; GFX10:       ; %bb.0:
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    s_or_saveexec_b32 s4, -1
+; GFX10-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s4
+; GFX10-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX10-NEXT:    s_mov_b32 s33, s32
+; GFX10-NEXT:    s_add_u32 s32, s32, 0x200
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    s_getpc_b64 s[4:5]
+; GFX10-NEXT:    s_add_u32 s4, s4, external_void_func_i8_signext@rel32@lo+4
+; GFX10-NEXT:    s_addc_u32 s5, s5, external_void_func_i8_signext@rel32@hi+12
+; GFX10-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX10-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX10-NEXT:    global_load_sbyte v0, v[0:1], off
+; GFX10-NEXT:    s_swappc_b64 s[30:31], s[4:5]
+; GFX10-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX10-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX10-NEXT:    s_sub_u32 s32, s32, 0x200
+; GFX10-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX10-NEXT:    s_or_saveexec_b32 s6, -1
+; GFX10-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s6
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    s_setpc_b64 s[4:5]
+  %var = load volatile i8, i8 addrspace(1)* undef
+  call amdgpu_gfx_callable void @external_void_func_i8_signext(i8 %var)
+  ret void
+}
+
+define amdgpu_gfx_callable void @test_call_external_void_func_i8_zeroext(i32) #0 {
+; GFX9-LABEL: test_call_external_void_func_i8_zeroext:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    s_or_saveexec_b64 s[4:5], -1
+; GFX9-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX9-NEXT:    s_mov_b64 exec, s[4:5]
+; GFX9-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX9-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX9-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX9-NEXT:    global_load_ubyte v0, v[0:1], off
+; GFX9-NEXT:    s_mov_b32 s33, s32
+; GFX9-NEXT:    s_add_u32 s32, s32, 0x400
+; GFX9-NEXT:    s_getpc_b64 s[4:5]
+; GFX9-NEXT:    s_add_u32 s4, s4, external_void_func_i8_zeroext@rel32@lo+4
+; GFX9-NEXT:    s_addc_u32 s5, s5, external_void_func_i8_zeroext@rel32@hi+12
+; GFX9-NEXT:    s_swappc_b64 s[30:31], s[4:5]
+; GFX9-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX9-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX9-NEXT:    s_sub_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX9-NEXT:    s_or_saveexec_b64 s[6:7], -1
+; GFX9-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX9-NEXT:    s_mov_b64 exec, s[6:7]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    s_setpc_b64 s[4:5]
+;
+; GFX10-LABEL: test_call_external_void_func_i8_zeroext:
+; GFX10:       ; %bb.0:
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    s_or_saveexec_b32 s4, -1
+; GFX10-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s4
+; GFX10-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX10-NEXT:    s_mov_b32 s33, s32
+; GFX10-NEXT:    s_add_u32 s32, s32, 0x200
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    s_getpc_b64 s[4:5]
+; GFX10-NEXT:    s_add_u32 s4, s4, external_void_func_i8_zeroext@rel32@lo+4
+; GFX10-NEXT:    s_addc_u32 s5, s5, external_void_func_i8_zeroext@rel32@hi+12
+; GFX10-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX10-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX10-NEXT:    global_load_ubyte v0, v[0:1], off
+; GFX10-NEXT:    s_swappc_b64 s[30:31], s[4:5]
+; GFX10-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX10-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX10-NEXT:    s_sub_u32 s32, s32, 0x200
+; GFX10-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX10-NEXT:    s_or_saveexec_b32 s6, -1
+; GFX10-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s6
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    s_setpc_b64 s[4:5]
+  %var = load volatile i8, i8 addrspace(1)* undef
+  call amdgpu_gfx_callable void @external_void_func_i8_zeroext(i8 %var)
+  ret void
+}
+
+define amdgpu_gfx_callable void @test_call_external_void_func_i16_imm() #0 {
+; GFX9-LABEL: test_call_external_void_func_i16_imm:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    s_or_saveexec_b64 s[4:5], -1
+; GFX9-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX9-NEXT:    s_mov_b64 exec, s[4:5]
+; GFX9-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX9-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX9-NEXT:    s_mov_b32 s33, s32
+; GFX9-NEXT:    s_add_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_mov_b32_e32 v0, 0x7b
+; GFX9-NEXT:    s_getpc_b64 s[4:5]
+; GFX9-NEXT:    s_add_u32 s4, s4, external_void_func_i16@rel32@lo+4
+; GFX9-NEXT:    s_addc_u32 s5, s5, external_void_func_i16@rel32@hi+12
+; GFX9-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX9-NEXT:    s_swappc_b64 s[30:31], s[4:5]
+; GFX9-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX9-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX9-NEXT:    s_sub_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX9-NEXT:    s_or_saveexec_b64 s[6:7], -1
+; GFX9-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX9-NEXT:    s_mov_b64 exec, s[6:7]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    s_setpc_b64 s[4:5]
+;
+; GFX10-LABEL: test_call_external_void_func_i16_imm:
+; GFX10:       ; %bb.0:
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    s_or_saveexec_b32 s4, -1
+; GFX10-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s4
+; GFX10-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX10-NEXT:    v_mov_b32_e32 v0, 0x7b
+; GFX10-NEXT:    s_mov_b32 s33, s32
+; GFX10-NEXT:    s_add_u32 s32, s32, 0x200
+; GFX10-NEXT:    s_getpc_b64 s[4:5]
+; GFX10-NEXT:    s_add_u32 s4, s4, external_void_func_i16@rel32@lo+4
+; GFX10-NEXT:    s_addc_u32 s5, s5, external_void_func_i16@rel32@hi+12
+; GFX10-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX10-NEXT:    s_swappc_b64 s[30:31], s[4:5]
+; GFX10-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX10-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX10-NEXT:    s_sub_u32 s32, s32, 0x200
+; GFX10-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX10-NEXT:    s_or_saveexec_b32 s6, -1
+; GFX10-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s6
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    s_setpc_b64 s[4:5]
+  call amdgpu_gfx_callable void @external_void_func_i16(i16 123)
+  ret void
+}
+
+define amdgpu_gfx_callable void @test_call_external_void_func_i16_signext(i32) #0 {
+; GFX9-LABEL: test_call_external_void_func_i16_signext:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    s_or_saveexec_b64 s[4:5], -1
+; GFX9-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX9-NEXT:    s_mov_b64 exec, s[4:5]
+; GFX9-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX9-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX9-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX9-NEXT:    global_load_ushort v0, v[0:1], off
+; GFX9-NEXT:    s_mov_b32 s33, s32
+; GFX9-NEXT:    s_add_u32 s32, s32, 0x400
+; GFX9-NEXT:    s_getpc_b64 s[4:5]
+; GFX9-NEXT:    s_add_u32 s4, s4, external_void_func_i16_signext@rel32@lo+4
+; GFX9-NEXT:    s_addc_u32 s5, s5, external_void_func_i16_signext@rel32@hi+12
+; GFX9-NEXT:    s_swappc_b64 s[30:31], s[4:5]
+; GFX9-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX9-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX9-NEXT:    s_sub_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX9-NEXT:    s_or_saveexec_b64 s[6:7], -1
+; GFX9-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX9-NEXT:    s_mov_b64 exec, s[6:7]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    s_setpc_b64 s[4:5]
+;
+; GFX10-LABEL: test_call_external_void_func_i16_signext:
+; GFX10:       ; %bb.0:
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    s_or_saveexec_b32 s4, -1
+; GFX10-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s4
+; GFX10-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX10-NEXT:    s_mov_b32 s33, s32
+; GFX10-NEXT:    s_add_u32 s32, s32, 0x200
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    s_getpc_b64 s[4:5]
+; GFX10-NEXT:    s_add_u32 s4, s4, external_void_func_i16_signext@rel32@lo+4
+; GFX10-NEXT:    s_addc_u32 s5, s5, external_void_func_i16_signext@rel32@hi+12
+; GFX10-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX10-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX10-NEXT:    global_load_ushort v0, v[0:1], off
+; GFX10-NEXT:    s_swappc_b64 s[30:31], s[4:5]
+; GFX10-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX10-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX10-NEXT:    s_sub_u32 s32, s32, 0x200
+; GFX10-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX10-NEXT:    s_or_saveexec_b32 s6, -1
+; GFX10-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s6
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    s_setpc_b64 s[4:5]
+  %var = load volatile i16, i16 addrspace(1)* undef
+  call amdgpu_gfx_callable void @external_void_func_i16_signext(i16 %var)
+  ret void
+}
+
+define amdgpu_gfx_callable void @test_call_external_void_func_i16_zeroext(i32) #0 {
+; GFX9-LABEL: test_call_external_void_func_i16_zeroext:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    s_or_saveexec_b64 s[4:5], -1
+; GFX9-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX9-NEXT:    s_mov_b64 exec, s[4:5]
+; GFX9-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX9-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX9-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX9-NEXT:    global_load_ushort v0, v[0:1], off
+; GFX9-NEXT:    s_mov_b32 s33, s32
+; GFX9-NEXT:    s_add_u32 s32, s32, 0x400
+; GFX9-NEXT:    s_getpc_b64 s[4:5]
+; GFX9-NEXT:    s_add_u32 s4, s4, external_void_func_i16_zeroext@rel32@lo+4
+; GFX9-NEXT:    s_addc_u32 s5, s5, external_void_func_i16_zeroext@rel32@hi+12
+; GFX9-NEXT:    s_swappc_b64 s[30:31], s[4:5]
+; GFX9-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX9-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX9-NEXT:    s_sub_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX9-NEXT:    s_or_saveexec_b64 s[6:7], -1
+; GFX9-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX9-NEXT:    s_mov_b64 exec, s[6:7]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    s_setpc_b64 s[4:5]
+;
+; GFX10-LABEL: test_call_external_void_func_i16_zeroext:
+; GFX10:       ; %bb.0:
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    s_or_saveexec_b32 s4, -1
+; GFX10-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s4
+; GFX10-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX10-NEXT:    s_mov_b32 s33, s32
+; GFX10-NEXT:    s_add_u32 s32, s32, 0x200
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    s_getpc_b64 s[4:5]
+; GFX10-NEXT:    s_add_u32 s4, s4, external_void_func_i16_zeroext@rel32@lo+4
+; GFX10-NEXT:    s_addc_u32 s5, s5, external_void_func_i16_zeroext@rel32@hi+12
+; GFX10-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX10-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX10-NEXT:    global_load_ushort v0, v[0:1], off
+; GFX10-NEXT:    s_swappc_b64 s[30:31], s[4:5]
+; GFX10-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX10-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX10-NEXT:    s_sub_u32 s32, s32, 0x200
+; GFX10-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX10-NEXT:    s_or_saveexec_b32 s6, -1
+; GFX10-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s6
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    s_setpc_b64 s[4:5]
+  %var = load volatile i16, i16 addrspace(1)* undef
+  call amdgpu_gfx_callable void @external_void_func_i16_zeroext(i16 %var)
+  ret void
+}
+
+define amdgpu_gfx_callable void @test_call_external_void_func_i32_imm(i32) #0 {
+; GFX9-LABEL: test_call_external_void_func_i32_imm:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    s_or_saveexec_b64 s[4:5], -1
+; GFX9-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX9-NEXT:    s_mov_b64 exec, s[4:5]
+; GFX9-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX9-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX9-NEXT:    s_mov_b32 s33, s32
+; GFX9-NEXT:    s_add_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_mov_b32_e32 v0, 42
+; GFX9-NEXT:    s_getpc_b64 s[4:5]
+; GFX9-NEXT:    s_add_u32 s4, s4, external_void_func_i32@rel32@lo+4
+; GFX9-NEXT:    s_addc_u32 s5, s5, external_void_func_i32@rel32@hi+12
+; GFX9-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX9-NEXT:    s_swappc_b64 s[30:31], s[4:5]
+; GFX9-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX9-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX9-NEXT:    s_sub_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX9-NEXT:    s_or_saveexec_b64 s[6:7], -1
+; GFX9-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX9-NEXT:    s_mov_b64 exec, s[6:7]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    s_setpc_b64 s[4:5]
+;
+; GFX10-LABEL: test_call_external_void_func_i32_imm:
+; GFX10:       ; %bb.0:
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    s_or_saveexec_b32 s4, -1
+; GFX10-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s4
+; GFX10-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX10-NEXT:    v_mov_b32_e32 v0, 42
+; GFX10-NEXT:    s_mov_b32 s33, s32
+; GFX10-NEXT:    s_add_u32 s32, s32, 0x200
+; GFX10-NEXT:    s_getpc_b64 s[4:5]
+; GFX10-NEXT:    s_add_u32 s4, s4, external_void_func_i32@rel32@lo+4
+; GFX10-NEXT:    s_addc_u32 s5, s5, external_void_func_i32@rel32@hi+12
+; GFX10-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX10-NEXT:    s_swappc_b64 s[30:31], s[4:5]
+; GFX10-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX10-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX10-NEXT:    s_sub_u32 s32, s32, 0x200
+; GFX10-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX10-NEXT:    s_or_saveexec_b32 s6, -1
+; GFX10-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s6
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    s_setpc_b64 s[4:5]
+  call amdgpu_gfx_callable void @external_void_func_i32(i32 42)
+  ret void
+}
+
+define amdgpu_gfx_callable void @test_call_external_void_func_i64_imm() #0 {
+; GFX9-LABEL: test_call_external_void_func_i64_imm:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    s_or_saveexec_b64 s[4:5], -1
+; GFX9-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX9-NEXT:    s_mov_b64 exec, s[4:5]
+; GFX9-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX9-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX9-NEXT:    s_mov_b32 s33, s32
+; GFX9-NEXT:    s_add_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_mov_b32_e32 v0, 0x7b
+; GFX9-NEXT:    v_mov_b32_e32 v1, 0
+; GFX9-NEXT:    s_getpc_b64 s[4:5]
+; GFX9-NEXT:    s_add_u32 s4, s4, external_void_func_i64@rel32@lo+4
+; GFX9-NEXT:    s_addc_u32 s5, s5, external_void_func_i64@rel32@hi+12
+; GFX9-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX9-NEXT:    s_swappc_b64 s[30:31], s[4:5]
+; GFX9-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX9-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX9-NEXT:    s_sub_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX9-NEXT:    s_or_saveexec_b64 s[6:7], -1
+; GFX9-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX9-NEXT:    s_mov_b64 exec, s[6:7]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    s_setpc_b64 s[4:5]
+;
+; GFX10-LABEL: test_call_external_void_func_i64_imm:
+; GFX10:       ; %bb.0:
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    s_or_saveexec_b32 s4, -1
+; GFX10-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s4
+; GFX10-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX10-NEXT:    v_mov_b32_e32 v0, 0x7b
+; GFX10-NEXT:    v_mov_b32_e32 v1, 0
+; GFX10-NEXT:    s_mov_b32 s33, s32
+; GFX10-NEXT:    s_add_u32 s32, s32, 0x200
+; GFX10-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX10-NEXT:    s_getpc_b64 s[4:5]
+; GFX10-NEXT:    s_add_u32 s4, s4, external_void_func_i64@rel32@lo+4
+; GFX10-NEXT:    s_addc_u32 s5, s5, external_void_func_i64@rel32@hi+12
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX10-NEXT:    s_swappc_b64 s[30:31], s[4:5]
+; GFX10-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX10-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX10-NEXT:    s_sub_u32 s32, s32, 0x200
+; GFX10-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX10-NEXT:    s_or_saveexec_b32 s6, -1
+; GFX10-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s6
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    s_setpc_b64 s[4:5]
+  call amdgpu_gfx_callable void @external_void_func_i64(i64 123)
+  ret void
+}
+
+define amdgpu_gfx_callable void @test_call_external_void_func_v2i64() #0 {
+; GFX9-LABEL: test_call_external_void_func_v2i64:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    s_or_saveexec_b64 s[4:5], -1
+; GFX9-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX9-NEXT:    s_mov_b64 exec, s[4:5]
+; GFX9-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX9-NEXT:    v_mov_b32_e32 v0, 0
+; GFX9-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX9-NEXT:    v_mov_b32_e32 v1, 0
+; GFX9-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX9-NEXT:    global_load_dwordx4 v[0:3], v[0:1], off
+; GFX9-NEXT:    s_mov_b32 s33, s32
+; GFX9-NEXT:    s_add_u32 s32, s32, 0x400
+; GFX9-NEXT:    s_getpc_b64 s[4:5]
+; GFX9-NEXT:    s_add_u32 s4, s4, external_void_func_v2i64@rel32@lo+4
+; GFX9-NEXT:    s_addc_u32 s5, s5, external_void_func_v2i64@rel32@hi+12
+; GFX9-NEXT:    s_swappc_b64 s[30:31], s[4:5]
+; GFX9-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX9-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX9-NEXT:    s_sub_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX9-NEXT:    s_or_saveexec_b64 s[6:7], -1
+; GFX9-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX9-NEXT:    s_mov_b64 exec, s[6:7]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    s_setpc_b64 s[4:5]
+;
+; GFX10-LABEL: test_call_external_void_func_v2i64:
+; GFX10:       ; %bb.0:
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    s_or_saveexec_b32 s4, -1
+; GFX10-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s4
+; GFX10-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX10-NEXT:    v_mov_b32_e32 v0, 0
+; GFX10-NEXT:    v_mov_b32_e32 v1, 0
+; GFX10-NEXT:    s_mov_b32 s33, s32
+; GFX10-NEXT:    s_add_u32 s32, s32, 0x200
+; GFX10-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    s_getpc_b64 s[4:5]
+; GFX10-NEXT:    s_add_u32 s4, s4, external_void_func_v2i64@rel32@lo+4
+; GFX10-NEXT:    s_addc_u32 s5, s5, external_void_func_v2i64@rel32@hi+12
+; GFX10-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX10-NEXT:    global_load_dwordx4 v[0:3], v[0:1], off
+; GFX10-NEXT:    s_swappc_b64 s[30:31], s[4:5]
+; GFX10-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX10-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX10-NEXT:    s_sub_u32 s32, s32, 0x200
+; GFX10-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX10-NEXT:    s_or_saveexec_b32 s6, -1
+; GFX10-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s6
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    s_setpc_b64 s[4:5]
+  %val = load <2 x i64>, <2 x i64> addrspace(1)* null
+  call amdgpu_gfx_callable void @external_void_func_v2i64(<2 x i64> %val)
+  ret void
+}
+
+define amdgpu_gfx_callable void @test_call_external_void_func_v2i64_imm() #0 {
+; GFX9-LABEL: test_call_external_void_func_v2i64_imm:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    s_or_saveexec_b64 s[4:5], -1
+; GFX9-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX9-NEXT:    s_mov_b64 exec, s[4:5]
+; GFX9-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX9-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX9-NEXT:    s_mov_b32 s33, s32
+; GFX9-NEXT:    s_add_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_mov_b32_e32 v0, 1
+; GFX9-NEXT:    v_mov_b32_e32 v1, 2
+; GFX9-NEXT:    v_mov_b32_e32 v2, 3
+; GFX9-NEXT:    v_mov_b32_e32 v3, 4
+; GFX9-NEXT:    s_getpc_b64 s[4:5]
+; GFX9-NEXT:    s_add_u32 s4, s4, external_void_func_v2i64@rel32@lo+4
+; GFX9-NEXT:    s_addc_u32 s5, s5, external_void_func_v2i64@rel32@hi+12
+; GFX9-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX9-NEXT:    s_swappc_b64 s[30:31], s[4:5]
+; GFX9-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX9-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX9-NEXT:    s_sub_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX9-NEXT:    s_or_saveexec_b64 s[6:7], -1
+; GFX9-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX9-NEXT:    s_mov_b64 exec, s[6:7]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    s_setpc_b64 s[4:5]
+;
+; GFX10-LABEL: test_call_external_void_func_v2i64_imm:
+; GFX10:       ; %bb.0:
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    s_or_saveexec_b32 s4, -1
+; GFX10-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s4
+; GFX10-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX10-NEXT:    v_mov_b32_e32 v0, 1
+; GFX10-NEXT:    v_mov_b32_e32 v1, 2
+; GFX10-NEXT:    v_mov_b32_e32 v2, 3
+; GFX10-NEXT:    v_mov_b32_e32 v3, 4
+; GFX10-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX10-NEXT:    s_mov_b32 s33, s32
+; GFX10-NEXT:    s_add_u32 s32, s32, 0x200
+; GFX10-NEXT:    s_getpc_b64 s[4:5]
+; GFX10-NEXT:    s_add_u32 s4, s4, external_void_func_v2i64@rel32@lo+4
+; GFX10-NEXT:    s_addc_u32 s5, s5, external_void_func_v2i64@rel32@hi+12
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX10-NEXT:    s_swappc_b64 s[30:31], s[4:5]
+; GFX10-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX10-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX10-NEXT:    s_sub_u32 s32, s32, 0x200
+; GFX10-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX10-NEXT:    s_or_saveexec_b32 s6, -1
+; GFX10-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s6
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    s_setpc_b64 s[4:5]
+  call amdgpu_gfx_callable void @external_void_func_v2i64(<2 x i64> <i64 8589934593, i64 17179869187>)
+  ret void
+}
+
+define amdgpu_gfx_callable void @test_call_external_void_func_v3i64() #0 {
+; GFX9-LABEL: test_call_external_void_func_v3i64:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    s_or_saveexec_b64 s[4:5], -1
+; GFX9-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX9-NEXT:    s_mov_b64 exec, s[4:5]
+; GFX9-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX9-NEXT:    v_mov_b32_e32 v0, 0
+; GFX9-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX9-NEXT:    v_mov_b32_e32 v1, 0
+; GFX9-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX9-NEXT:    global_load_dwordx4 v[0:3], v[0:1], off
+; GFX9-NEXT:    s_mov_b32 s33, s32
+; GFX9-NEXT:    s_add_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_mov_b32_e32 v4, 1
+; GFX9-NEXT:    v_mov_b32_e32 v5, 2
+; GFX9-NEXT:    s_getpc_b64 s[4:5]
+; GFX9-NEXT:    s_add_u32 s4, s4, external_void_func_v3i64@rel32@lo+4
+; GFX9-NEXT:    s_addc_u32 s5, s5, external_void_func_v3i64@rel32@hi+12
+; GFX9-NEXT:    s_swappc_b64 s[30:31], s[4:5]
+; GFX9-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX9-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX9-NEXT:    s_sub_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX9-NEXT:    s_or_saveexec_b64 s[6:7], -1
+; GFX9-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX9-NEXT:    s_mov_b64 exec, s[6:7]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    s_setpc_b64 s[4:5]
+;
+; GFX10-LABEL: test_call_external_void_func_v3i64:
+; GFX10:       ; %bb.0:
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    s_or_saveexec_b32 s4, -1
+; GFX10-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s4
+; GFX10-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX10-NEXT:    v_mov_b32_e32 v0, 0
+; GFX10-NEXT:    v_mov_b32_e32 v1, 0
+; GFX10-NEXT:    v_mov_b32_e32 v4, 1
+; GFX10-NEXT:    v_mov_b32_e32 v5, 2
+; GFX10-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX10-NEXT:    s_mov_b32 s33, s32
+; GFX10-NEXT:    s_add_u32 s32, s32, 0x200
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    s_getpc_b64 s[4:5]
+; GFX10-NEXT:    s_add_u32 s4, s4, external_void_func_v3i64@rel32@lo+4
+; GFX10-NEXT:    s_addc_u32 s5, s5, external_void_func_v3i64@rel32@hi+12
+; GFX10-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX10-NEXT:    global_load_dwordx4 v[0:3], v[0:1], off
+; GFX10-NEXT:    s_swappc_b64 s[30:31], s[4:5]
+; GFX10-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX10-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX10-NEXT:    s_sub_u32 s32, s32, 0x200
+; GFX10-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX10-NEXT:    s_or_saveexec_b32 s6, -1
+; GFX10-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s6
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    s_setpc_b64 s[4:5]
+  %load = load <2 x i64>, <2 x i64> addrspace(1)* null
+  %val = shufflevector <2 x i64> %load, <2 x i64> <i64 8589934593, i64 undef>, <3 x i32> <i32 0, i32 1, i32 2>
+
+  call amdgpu_gfx_callable void @external_void_func_v3i64(<3 x i64> %val)
+  ret void
+}
+
+define amdgpu_gfx_callable void @test_call_external_void_func_v4i64() #0 {
+; GFX9-LABEL: test_call_external_void_func_v4i64:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    s_or_saveexec_b64 s[4:5], -1
+; GFX9-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX9-NEXT:    s_mov_b64 exec, s[4:5]
+; GFX9-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX9-NEXT:    v_mov_b32_e32 v0, 0
+; GFX9-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX9-NEXT:    v_mov_b32_e32 v1, 0
+; GFX9-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX9-NEXT:    global_load_dwordx4 v[0:3], v[0:1], off
+; GFX9-NEXT:    s_mov_b32 s33, s32
+; GFX9-NEXT:    s_add_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_mov_b32_e32 v4, 1
+; GFX9-NEXT:    v_mov_b32_e32 v5, 2
+; GFX9-NEXT:    v_mov_b32_e32 v6, 3
+; GFX9-NEXT:    v_mov_b32_e32 v7, 4
+; GFX9-NEXT:    s_getpc_b64 s[4:5]
+; GFX9-NEXT:    s_add_u32 s4, s4, external_void_func_v4i64@rel32@lo+4
+; GFX9-NEXT:    s_addc_u32 s5, s5, external_void_func_v4i64@rel32@hi+12
+; GFX9-NEXT:    s_swappc_b64 s[30:31], s[4:5]
+; GFX9-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX9-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX9-NEXT:    s_sub_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX9-NEXT:    s_or_saveexec_b64 s[6:7], -1
+; GFX9-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX9-NEXT:    s_mov_b64 exec, s[6:7]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    s_setpc_b64 s[4:5]
+;
+; GFX10-LABEL: test_call_external_void_func_v4i64:
+; GFX10:       ; %bb.0:
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    s_or_saveexec_b32 s4, -1
+; GFX10-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s4
+; GFX10-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX10-NEXT:    v_mov_b32_e32 v0, 0
+; GFX10-NEXT:    v_mov_b32_e32 v1, 0
+; GFX10-NEXT:    v_mov_b32_e32 v4, 1
+; GFX10-NEXT:    v_mov_b32_e32 v5, 2
+; GFX10-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX10-NEXT:    v_mov_b32_e32 v6, 3
+; GFX10-NEXT:    v_mov_b32_e32 v7, 4
+; GFX10-NEXT:    s_mov_b32 s33, s32
+; GFX10-NEXT:    s_add_u32 s32, s32, 0x200
+; GFX10-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX10-NEXT:    global_load_dwordx4 v[0:3], v[0:1], off
+; GFX10-NEXT:    s_getpc_b64 s[4:5]
+; GFX10-NEXT:    s_add_u32 s4, s4, external_void_func_v4i64@rel32@lo+4
+; GFX10-NEXT:    s_addc_u32 s5, s5, external_void_func_v4i64@rel32@hi+12
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    s_swappc_b64 s[30:31], s[4:5]
+; GFX10-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX10-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX10-NEXT:    s_sub_u32 s32, s32, 0x200
+; GFX10-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX10-NEXT:    s_or_saveexec_b32 s6, -1
+; GFX10-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s6
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    s_setpc_b64 s[4:5]
+  %load = load <2 x i64>, <2 x i64> addrspace(1)* null
+  %val = shufflevector <2 x i64> %load, <2 x i64> <i64 8589934593, i64 17179869187>, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  call amdgpu_gfx_callable void @external_void_func_v4i64(<4 x i64> %val)
+  ret void
+}
+
+define amdgpu_gfx_callable void @test_call_external_void_func_f16_imm() #0 {
+; GFX9-LABEL: test_call_external_void_func_f16_imm:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    s_or_saveexec_b64 s[4:5], -1
+; GFX9-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX9-NEXT:    s_mov_b64 exec, s[4:5]
+; GFX9-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX9-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX9-NEXT:    s_mov_b32 s33, s32
+; GFX9-NEXT:    s_add_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_mov_b32_e32 v0, 0x4400
+; GFX9-NEXT:    s_getpc_b64 s[4:5]
+; GFX9-NEXT:    s_add_u32 s4, s4, external_void_func_f16@rel32@lo+4
+; GFX9-NEXT:    s_addc_u32 s5, s5, external_void_func_f16@rel32@hi+12
+; GFX9-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX9-NEXT:    s_swappc_b64 s[30:31], s[4:5]
+; GFX9-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX9-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX9-NEXT:    s_sub_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX9-NEXT:    s_or_saveexec_b64 s[6:7], -1
+; GFX9-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX9-NEXT:    s_mov_b64 exec, s[6:7]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    s_setpc_b64 s[4:5]
+;
+; GFX10-LABEL: test_call_external_void_func_f16_imm:
+; GFX10:       ; %bb.0:
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    s_or_saveexec_b32 s4, -1
+; GFX10-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s4
+; GFX10-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX10-NEXT:    v_mov_b32_e32 v0, 0x4400
+; GFX10-NEXT:    s_mov_b32 s33, s32
+; GFX10-NEXT:    s_add_u32 s32, s32, 0x200
+; GFX10-NEXT:    s_getpc_b64 s[4:5]
+; GFX10-NEXT:    s_add_u32 s4, s4, external_void_func_f16@rel32@lo+4
+; GFX10-NEXT:    s_addc_u32 s5, s5, external_void_func_f16@rel32@hi+12
+; GFX10-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX10-NEXT:    s_swappc_b64 s[30:31], s[4:5]
+; GFX10-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX10-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX10-NEXT:    s_sub_u32 s32, s32, 0x200
+; GFX10-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX10-NEXT:    s_or_saveexec_b32 s6, -1
+; GFX10-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s6
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    s_setpc_b64 s[4:5]
+  call amdgpu_gfx_callable void @external_void_func_f16(half 4.0)
+  ret void
+}
+
+define amdgpu_gfx_callable void @test_call_external_void_func_f32_imm() #0 {
+; GFX9-LABEL: test_call_external_void_func_f32_imm:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    s_or_saveexec_b64 s[4:5], -1
+; GFX9-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX9-NEXT:    s_mov_b64 exec, s[4:5]
+; GFX9-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX9-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX9-NEXT:    s_mov_b32 s33, s32
+; GFX9-NEXT:    s_add_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_mov_b32_e32 v0, 4.0
+; GFX9-NEXT:    s_getpc_b64 s[4:5]
+; GFX9-NEXT:    s_add_u32 s4, s4, external_void_func_f32@rel32@lo+4
+; GFX9-NEXT:    s_addc_u32 s5, s5, external_void_func_f32@rel32@hi+12
+; GFX9-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX9-NEXT:    s_swappc_b64 s[30:31], s[4:5]
+; GFX9-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX9-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX9-NEXT:    s_sub_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX9-NEXT:    s_or_saveexec_b64 s[6:7], -1
+; GFX9-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX9-NEXT:    s_mov_b64 exec, s[6:7]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    s_setpc_b64 s[4:5]
+;
+; GFX10-LABEL: test_call_external_void_func_f32_imm:
+; GFX10:       ; %bb.0:
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    s_or_saveexec_b32 s4, -1
+; GFX10-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s4
+; GFX10-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX10-NEXT:    v_mov_b32_e32 v0, 4.0
+; GFX10-NEXT:    s_mov_b32 s33, s32
+; GFX10-NEXT:    s_add_u32 s32, s32, 0x200
+; GFX10-NEXT:    s_getpc_b64 s[4:5]
+; GFX10-NEXT:    s_add_u32 s4, s4, external_void_func_f32@rel32@lo+4
+; GFX10-NEXT:    s_addc_u32 s5, s5, external_void_func_f32@rel32@hi+12
+; GFX10-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX10-NEXT:    s_swappc_b64 s[30:31], s[4:5]
+; GFX10-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX10-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX10-NEXT:    s_sub_u32 s32, s32, 0x200
+; GFX10-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX10-NEXT:    s_or_saveexec_b32 s6, -1
+; GFX10-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s6
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    s_setpc_b64 s[4:5]
+  call amdgpu_gfx_callable void @external_void_func_f32(float 4.0)
+  ret void
+}
+
+define amdgpu_gfx_callable void @test_call_external_void_func_v2f32_imm() #0 {
+; GFX9-LABEL: test_call_external_void_func_v2f32_imm:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    s_or_saveexec_b64 s[4:5], -1
+; GFX9-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX9-NEXT:    s_mov_b64 exec, s[4:5]
+; GFX9-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX9-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX9-NEXT:    s_mov_b32 s33, s32
+; GFX9-NEXT:    s_add_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_mov_b32_e32 v0, 1.0
+; GFX9-NEXT:    v_mov_b32_e32 v1, 2.0
+; GFX9-NEXT:    s_getpc_b64 s[4:5]
+; GFX9-NEXT:    s_add_u32 s4, s4, external_void_func_v2f32@rel32@lo+4
+; GFX9-NEXT:    s_addc_u32 s5, s5, external_void_func_v2f32@rel32@hi+12
+; GFX9-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX9-NEXT:    s_swappc_b64 s[30:31], s[4:5]
+; GFX9-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX9-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX9-NEXT:    s_sub_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX9-NEXT:    s_or_saveexec_b64 s[6:7], -1
+; GFX9-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX9-NEXT:    s_mov_b64 exec, s[6:7]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    s_setpc_b64 s[4:5]
+;
+; GFX10-LABEL: test_call_external_void_func_v2f32_imm:
+; GFX10:       ; %bb.0:
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    s_or_saveexec_b32 s4, -1
+; GFX10-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s4
+; GFX10-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX10-NEXT:    v_mov_b32_e32 v0, 1.0
+; GFX10-NEXT:    v_mov_b32_e32 v1, 2.0
+; GFX10-NEXT:    s_mov_b32 s33, s32
+; GFX10-NEXT:    s_add_u32 s32, s32, 0x200
+; GFX10-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX10-NEXT:    s_getpc_b64 s[4:5]
+; GFX10-NEXT:    s_add_u32 s4, s4, external_void_func_v2f32@rel32@lo+4
+; GFX10-NEXT:    s_addc_u32 s5, s5, external_void_func_v2f32@rel32@hi+12
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX10-NEXT:    s_swappc_b64 s[30:31], s[4:5]
+; GFX10-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX10-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX10-NEXT:    s_sub_u32 s32, s32, 0x200
+; GFX10-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX10-NEXT:    s_or_saveexec_b32 s6, -1
+; GFX10-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s6
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    s_setpc_b64 s[4:5]
+  call amdgpu_gfx_callable void @external_void_func_v2f32(<2 x float> <float 1.0, float 2.0>)
+  ret void
+}
+
+define amdgpu_gfx_callable void @test_call_external_void_func_v3f32_imm() #0 {
+; GFX9-LABEL: test_call_external_void_func_v3f32_imm:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    s_or_saveexec_b64 s[4:5], -1
+; GFX9-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX9-NEXT:    s_mov_b64 exec, s[4:5]
+; GFX9-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX9-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX9-NEXT:    s_mov_b32 s33, s32
+; GFX9-NEXT:    s_add_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_mov_b32_e32 v0, 1.0
+; GFX9-NEXT:    v_mov_b32_e32 v1, 2.0
+; GFX9-NEXT:    v_mov_b32_e32 v2, 4.0
+; GFX9-NEXT:    s_getpc_b64 s[4:5]
+; GFX9-NEXT:    s_add_u32 s4, s4, external_void_func_v3f32@rel32@lo+4
+; GFX9-NEXT:    s_addc_u32 s5, s5, external_void_func_v3f32@rel32@hi+12
+; GFX9-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX9-NEXT:    s_swappc_b64 s[30:31], s[4:5]
+; GFX9-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX9-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX9-NEXT:    s_sub_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX9-NEXT:    s_or_saveexec_b64 s[6:7], -1
+; GFX9-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX9-NEXT:    s_mov_b64 exec, s[6:7]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    s_setpc_b64 s[4:5]
+;
+; GFX10-LABEL: test_call_external_void_func_v3f32_imm:
+; GFX10:       ; %bb.0:
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    s_or_saveexec_b32 s4, -1
+; GFX10-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s4
+; GFX10-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX10-NEXT:    v_mov_b32_e32 v0, 1.0
+; GFX10-NEXT:    v_mov_b32_e32 v1, 2.0
+; GFX10-NEXT:    v_mov_b32_e32 v2, 4.0
+; GFX10-NEXT:    s_mov_b32 s33, s32
+; GFX10-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX10-NEXT:    s_add_u32 s32, s32, 0x200
+; GFX10-NEXT:    s_getpc_b64 s[4:5]
+; GFX10-NEXT:    s_add_u32 s4, s4, external_void_func_v3f32@rel32@lo+4
+; GFX10-NEXT:    s_addc_u32 s5, s5, external_void_func_v3f32@rel32@hi+12
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX10-NEXT:    s_swappc_b64 s[30:31], s[4:5]
+; GFX10-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX10-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX10-NEXT:    s_sub_u32 s32, s32, 0x200
+; GFX10-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX10-NEXT:    s_or_saveexec_b32 s6, -1
+; GFX10-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s6
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    s_setpc_b64 s[4:5]
+  call amdgpu_gfx_callable void @external_void_func_v3f32(<3 x float> <float 1.0, float 2.0, float 4.0>)
+  ret void
+}
+
+define amdgpu_gfx_callable void @test_call_external_void_func_v5f32_imm() #0 {
+; GFX9-LABEL: test_call_external_void_func_v5f32_imm:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    s_or_saveexec_b64 s[4:5], -1
+; GFX9-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX9-NEXT:    s_mov_b64 exec, s[4:5]
+; GFX9-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX9-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX9-NEXT:    s_mov_b32 s33, s32
+; GFX9-NEXT:    s_add_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_mov_b32_e32 v0, 1.0
+; GFX9-NEXT:    v_mov_b32_e32 v1, 2.0
+; GFX9-NEXT:    v_mov_b32_e32 v2, 4.0
+; GFX9-NEXT:    v_mov_b32_e32 v3, -1.0
+; GFX9-NEXT:    v_mov_b32_e32 v4, 0.5
+; GFX9-NEXT:    s_getpc_b64 s[4:5]
+; GFX9-NEXT:    s_add_u32 s4, s4, external_void_func_v5f32@rel32@lo+4
+; GFX9-NEXT:    s_addc_u32 s5, s5, external_void_func_v5f32@rel32@hi+12
+; GFX9-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX9-NEXT:    s_swappc_b64 s[30:31], s[4:5]
+; GFX9-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX9-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX9-NEXT:    s_sub_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX9-NEXT:    s_or_saveexec_b64 s[6:7], -1
+; GFX9-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX9-NEXT:    s_mov_b64 exec, s[6:7]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    s_setpc_b64 s[4:5]
+;
+; GFX10-LABEL: test_call_external_void_func_v5f32_imm:
+; GFX10:       ; %bb.0:
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    s_or_saveexec_b32 s4, -1
+; GFX10-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s4
+; GFX10-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX10-NEXT:    v_mov_b32_e32 v0, 1.0
+; GFX10-NEXT:    v_mov_b32_e32 v1, 2.0
+; GFX10-NEXT:    v_mov_b32_e32 v2, 4.0
+; GFX10-NEXT:    v_mov_b32_e32 v3, -1.0
+; GFX10-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX10-NEXT:    v_mov_b32_e32 v4, 0.5
+; GFX10-NEXT:    s_mov_b32 s33, s32
+; GFX10-NEXT:    s_add_u32 s32, s32, 0x200
+; GFX10-NEXT:    s_getpc_b64 s[4:5]
+; GFX10-NEXT:    s_add_u32 s4, s4, external_void_func_v5f32@rel32@lo+4
+; GFX10-NEXT:    s_addc_u32 s5, s5, external_void_func_v5f32@rel32@hi+12
+; GFX10-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    s_swappc_b64 s[30:31], s[4:5]
+; GFX10-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX10-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX10-NEXT:    s_sub_u32 s32, s32, 0x200
+; GFX10-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX10-NEXT:    s_or_saveexec_b32 s6, -1
+; GFX10-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s6
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    s_setpc_b64 s[4:5]
+  call amdgpu_gfx_callable void @external_void_func_v5f32(<5 x float> <float 1.0, float 2.0, float 4.0, float -1.0, float 0.5>)
+  ret void
+}
+
+define amdgpu_gfx_callable void @test_call_external_void_func_f64_imm() #0 {
+; GFX9-LABEL: test_call_external_void_func_f64_imm:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    s_or_saveexec_b64 s[4:5], -1
+; GFX9-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX9-NEXT:    s_mov_b64 exec, s[4:5]
+; GFX9-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX9-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX9-NEXT:    s_mov_b32 s33, s32
+; GFX9-NEXT:    s_add_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_mov_b32_e32 v0, 0
+; GFX9-NEXT:    v_mov_b32_e32 v1, 0x40100000
+; GFX9-NEXT:    s_getpc_b64 s[4:5]
+; GFX9-NEXT:    s_add_u32 s4, s4, external_void_func_f64@rel32@lo+4
+; GFX9-NEXT:    s_addc_u32 s5, s5, external_void_func_f64@rel32@hi+12
+; GFX9-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX9-NEXT:    s_swappc_b64 s[30:31], s[4:5]
+; GFX9-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX9-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX9-NEXT:    s_sub_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX9-NEXT:    s_or_saveexec_b64 s[6:7], -1
+; GFX9-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX9-NEXT:    s_mov_b64 exec, s[6:7]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    s_setpc_b64 s[4:5]
+;
+; GFX10-LABEL: test_call_external_void_func_f64_imm:
+; GFX10:       ; %bb.0:
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    s_or_saveexec_b32 s4, -1
+; GFX10-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s4
+; GFX10-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX10-NEXT:    v_mov_b32_e32 v0, 0
+; GFX10-NEXT:    v_mov_b32_e32 v1, 0x40100000
+; GFX10-NEXT:    s_mov_b32 s33, s32
+; GFX10-NEXT:    s_add_u32 s32, s32, 0x200
+; GFX10-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX10-NEXT:    s_getpc_b64 s[4:5]
+; GFX10-NEXT:    s_add_u32 s4, s4, external_void_func_f64@rel32@lo+4
+; GFX10-NEXT:    s_addc_u32 s5, s5, external_void_func_f64@rel32@hi+12
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX10-NEXT:    s_swappc_b64 s[30:31], s[4:5]
+; GFX10-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX10-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX10-NEXT:    s_sub_u32 s32, s32, 0x200
+; GFX10-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX10-NEXT:    s_or_saveexec_b32 s6, -1
+; GFX10-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s6
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    s_setpc_b64 s[4:5]
+  call amdgpu_gfx_callable void @external_void_func_f64(double 4.0)
+  ret void
+}
+
+define amdgpu_gfx_callable void @test_call_external_void_func_v2f64_imm() #0 {
+; GFX9-LABEL: test_call_external_void_func_v2f64_imm:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    s_or_saveexec_b64 s[4:5], -1
+; GFX9-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX9-NEXT:    s_mov_b64 exec, s[4:5]
+; GFX9-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX9-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX9-NEXT:    s_mov_b32 s33, s32
+; GFX9-NEXT:    s_add_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_mov_b32_e32 v0, 0
+; GFX9-NEXT:    v_mov_b32_e32 v1, 2.0
+; GFX9-NEXT:    v_mov_b32_e32 v2, 0
+; GFX9-NEXT:    v_mov_b32_e32 v3, 0x40100000
+; GFX9-NEXT:    s_getpc_b64 s[4:5]
+; GFX9-NEXT:    s_add_u32 s4, s4, external_void_func_v2f64@rel32@lo+4
+; GFX9-NEXT:    s_addc_u32 s5, s5, external_void_func_v2f64@rel32@hi+12
+; GFX9-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX9-NEXT:    s_swappc_b64 s[30:31], s[4:5]
+; GFX9-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX9-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX9-NEXT:    s_sub_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX9-NEXT:    s_or_saveexec_b64 s[6:7], -1
+; GFX9-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX9-NEXT:    s_mov_b64 exec, s[6:7]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    s_setpc_b64 s[4:5]
+;
+; GFX10-LABEL: test_call_external_void_func_v2f64_imm:
+; GFX10:       ; %bb.0:
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    s_or_saveexec_b32 s4, -1
+; GFX10-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s4
+; GFX10-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX10-NEXT:    v_mov_b32_e32 v0, 0
+; GFX10-NEXT:    v_mov_b32_e32 v1, 2.0
+; GFX10-NEXT:    v_mov_b32_e32 v2, 0
+; GFX10-NEXT:    v_mov_b32_e32 v3, 0x40100000
+; GFX10-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX10-NEXT:    s_mov_b32 s33, s32
+; GFX10-NEXT:    s_add_u32 s32, s32, 0x200
+; GFX10-NEXT:    s_getpc_b64 s[4:5]
+; GFX10-NEXT:    s_add_u32 s4, s4, external_void_func_v2f64@rel32@lo+4
+; GFX10-NEXT:    s_addc_u32 s5, s5, external_void_func_v2f64@rel32@hi+12
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX10-NEXT:    s_swappc_b64 s[30:31], s[4:5]
+; GFX10-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX10-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX10-NEXT:    s_sub_u32 s32, s32, 0x200
+; GFX10-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX10-NEXT:    s_or_saveexec_b32 s6, -1
+; GFX10-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s6
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    s_setpc_b64 s[4:5]
+  call amdgpu_gfx_callable void @external_void_func_v2f64(<2 x double> <double 2.0, double 4.0>)
+  ret void
+}
+
+define amdgpu_gfx_callable void @test_call_external_void_func_v3f64_imm() #0 {
+; GFX9-LABEL: test_call_external_void_func_v3f64_imm:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    s_or_saveexec_b64 s[4:5], -1
+; GFX9-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX9-NEXT:    s_mov_b64 exec, s[4:5]
+; GFX9-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX9-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX9-NEXT:    s_mov_b32 s33, s32
+; GFX9-NEXT:    s_add_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_mov_b32_e32 v0, 0
+; GFX9-NEXT:    v_mov_b32_e32 v1, 2.0
+; GFX9-NEXT:    v_mov_b32_e32 v2, 0
+; GFX9-NEXT:    v_mov_b32_e32 v3, 0x40100000
+; GFX9-NEXT:    v_mov_b32_e32 v4, 0
+; GFX9-NEXT:    v_mov_b32_e32 v5, 0x40200000
+; GFX9-NEXT:    s_getpc_b64 s[4:5]
+; GFX9-NEXT:    s_add_u32 s4, s4, external_void_func_v3f64@rel32@lo+4
+; GFX9-NEXT:    s_addc_u32 s5, s5, external_void_func_v3f64@rel32@hi+12
+; GFX9-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX9-NEXT:    s_swappc_b64 s[30:31], s[4:5]
+; GFX9-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX9-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX9-NEXT:    s_sub_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX9-NEXT:    s_or_saveexec_b64 s[6:7], -1
+; GFX9-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX9-NEXT:    s_mov_b64 exec, s[6:7]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    s_setpc_b64 s[4:5]
+;
+; GFX10-LABEL: test_call_external_void_func_v3f64_imm:
+; GFX10:       ; %bb.0:
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    s_or_saveexec_b32 s4, -1
+; GFX10-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s4
+; GFX10-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX10-NEXT:    v_mov_b32_e32 v0, 0
+; GFX10-NEXT:    v_mov_b32_e32 v1, 2.0
+; GFX10-NEXT:    v_mov_b32_e32 v2, 0
+; GFX10-NEXT:    v_mov_b32_e32 v3, 0x40100000
+; GFX10-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX10-NEXT:    v_mov_b32_e32 v4, 0
+; GFX10-NEXT:    v_mov_b32_e32 v5, 0x40200000
+; GFX10-NEXT:    s_mov_b32 s33, s32
+; GFX10-NEXT:    s_add_u32 s32, s32, 0x200
+; GFX10-NEXT:    s_getpc_b64 s[4:5]
+; GFX10-NEXT:    s_add_u32 s4, s4, external_void_func_v3f64@rel32@lo+4
+; GFX10-NEXT:    s_addc_u32 s5, s5, external_void_func_v3f64@rel32@hi+12
+; GFX10-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    s_swappc_b64 s[30:31], s[4:5]
+; GFX10-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX10-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX10-NEXT:    s_sub_u32 s32, s32, 0x200
+; GFX10-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX10-NEXT:    s_or_saveexec_b32 s6, -1
+; GFX10-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s6
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    s_setpc_b64 s[4:5]
+  call amdgpu_gfx_callable void @external_void_func_v3f64(<3 x double> <double 2.0, double 4.0, double 8.0>)
+  ret void
+}
+
+define amdgpu_gfx_callable void @test_call_external_void_func_v2i16() #0 {
+; GFX9-LABEL: test_call_external_void_func_v2i16:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    s_or_saveexec_b64 s[4:5], -1
+; GFX9-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX9-NEXT:    s_mov_b64 exec, s[4:5]
+; GFX9-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX9-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX9-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX9-NEXT:    global_load_dword v0, v[0:1], off
+; GFX9-NEXT:    s_mov_b32 s33, s32
+; GFX9-NEXT:    s_add_u32 s32, s32, 0x400
+; GFX9-NEXT:    s_getpc_b64 s[4:5]
+; GFX9-NEXT:    s_add_u32 s4, s4, external_void_func_v2i16@rel32@lo+4
+; GFX9-NEXT:    s_addc_u32 s5, s5, external_void_func_v2i16@rel32@hi+12
+; GFX9-NEXT:    s_swappc_b64 s[30:31], s[4:5]
+; GFX9-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX9-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX9-NEXT:    s_sub_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX9-NEXT:    s_or_saveexec_b64 s[6:7], -1
+; GFX9-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX9-NEXT:    s_mov_b64 exec, s[6:7]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    s_setpc_b64 s[4:5]
+;
+; GFX10-LABEL: test_call_external_void_func_v2i16:
+; GFX10:       ; %bb.0:
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    s_or_saveexec_b32 s4, -1
+; GFX10-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s4
+; GFX10-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX10-NEXT:    s_mov_b32 s33, s32
+; GFX10-NEXT:    s_add_u32 s32, s32, 0x200
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    s_getpc_b64 s[4:5]
+; GFX10-NEXT:    s_add_u32 s4, s4, external_void_func_v2i16@rel32@lo+4
+; GFX10-NEXT:    s_addc_u32 s5, s5, external_void_func_v2i16@rel32@hi+12
+; GFX10-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX10-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX10-NEXT:    global_load_dword v0, v[0:1], off
+; GFX10-NEXT:    s_swappc_b64 s[30:31], s[4:5]
+; GFX10-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX10-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX10-NEXT:    s_sub_u32 s32, s32, 0x200
+; GFX10-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX10-NEXT:    s_or_saveexec_b32 s6, -1
+; GFX10-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s6
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    s_setpc_b64 s[4:5]
+  %val = load <2 x i16>, <2 x i16> addrspace(1)* undef
+  call amdgpu_gfx_callable void @external_void_func_v2i16(<2 x i16> %val)
+  ret void
+}
+
+define amdgpu_gfx_callable void @test_call_external_void_func_v3i16() #0 {
+; GFX9-LABEL: test_call_external_void_func_v3i16:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    s_or_saveexec_b64 s[4:5], -1
+; GFX9-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX9-NEXT:    s_mov_b64 exec, s[4:5]
+; GFX9-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX9-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX9-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX9-NEXT:    global_load_dwordx2 v[0:1], v[0:1], off
+; GFX9-NEXT:    s_mov_b32 s33, s32
+; GFX9-NEXT:    s_add_u32 s32, s32, 0x400
+; GFX9-NEXT:    s_getpc_b64 s[4:5]
+; GFX9-NEXT:    s_add_u32 s4, s4, external_void_func_v3i16@rel32@lo+4
+; GFX9-NEXT:    s_addc_u32 s5, s5, external_void_func_v3i16@rel32@hi+12
+; GFX9-NEXT:    s_swappc_b64 s[30:31], s[4:5]
+; GFX9-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX9-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX9-NEXT:    s_sub_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX9-NEXT:    s_or_saveexec_b64 s[6:7], -1
+; GFX9-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX9-NEXT:    s_mov_b64 exec, s[6:7]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    s_setpc_b64 s[4:5]
+;
+; GFX10-LABEL: test_call_external_void_func_v3i16:
+; GFX10:       ; %bb.0:
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    s_or_saveexec_b32 s4, -1
+; GFX10-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s4
+; GFX10-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX10-NEXT:    s_mov_b32 s33, s32
+; GFX10-NEXT:    s_add_u32 s32, s32, 0x200
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    s_getpc_b64 s[4:5]
+; GFX10-NEXT:    s_add_u32 s4, s4, external_void_func_v3i16@rel32@lo+4
+; GFX10-NEXT:    s_addc_u32 s5, s5, external_void_func_v3i16@rel32@hi+12
+; GFX10-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX10-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX10-NEXT:    global_load_dwordx2 v[0:1], v[0:1], off
+; GFX10-NEXT:    s_swappc_b64 s[30:31], s[4:5]
+; GFX10-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX10-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX10-NEXT:    s_sub_u32 s32, s32, 0x200
+; GFX10-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX10-NEXT:    s_or_saveexec_b32 s6, -1
+; GFX10-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s6
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    s_setpc_b64 s[4:5]
+  %val = load <3 x i16>, <3 x i16> addrspace(1)* undef
+  call amdgpu_gfx_callable void @external_void_func_v3i16(<3 x i16> %val)
+  ret void
+}
+
+define amdgpu_gfx_callable void @test_call_external_void_func_v3f16() #0 {
+; GFX9-LABEL: test_call_external_void_func_v3f16:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    s_or_saveexec_b64 s[4:5], -1
+; GFX9-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX9-NEXT:    s_mov_b64 exec, s[4:5]
+; GFX9-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX9-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX9-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX9-NEXT:    global_load_dwordx2 v[0:1], v[0:1], off
+; GFX9-NEXT:    s_mov_b32 s33, s32
+; GFX9-NEXT:    s_add_u32 s32, s32, 0x400
+; GFX9-NEXT:    s_getpc_b64 s[4:5]
+; GFX9-NEXT:    s_add_u32 s4, s4, external_void_func_v3f16@rel32@lo+4
+; GFX9-NEXT:    s_addc_u32 s5, s5, external_void_func_v3f16@rel32@hi+12
+; GFX9-NEXT:    s_swappc_b64 s[30:31], s[4:5]
+; GFX9-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX9-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX9-NEXT:    s_sub_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX9-NEXT:    s_or_saveexec_b64 s[6:7], -1
+; GFX9-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX9-NEXT:    s_mov_b64 exec, s[6:7]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    s_setpc_b64 s[4:5]
+;
+; GFX10-LABEL: test_call_external_void_func_v3f16:
+; GFX10:       ; %bb.0:
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    s_or_saveexec_b32 s4, -1
+; GFX10-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s4
+; GFX10-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX10-NEXT:    s_mov_b32 s33, s32
+; GFX10-NEXT:    s_add_u32 s32, s32, 0x200
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    s_getpc_b64 s[4:5]
+; GFX10-NEXT:    s_add_u32 s4, s4, external_void_func_v3f16@rel32@lo+4
+; GFX10-NEXT:    s_addc_u32 s5, s5, external_void_func_v3f16@rel32@hi+12
+; GFX10-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX10-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX10-NEXT:    global_load_dwordx2 v[0:1], v[0:1], off
+; GFX10-NEXT:    s_swappc_b64 s[30:31], s[4:5]
+; GFX10-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX10-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX10-NEXT:    s_sub_u32 s32, s32, 0x200
+; GFX10-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX10-NEXT:    s_or_saveexec_b32 s6, -1
+; GFX10-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s6
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    s_setpc_b64 s[4:5]
+  %val = load <3 x half>, <3 x half> addrspace(1)* undef
+  call amdgpu_gfx_callable void @external_void_func_v3f16(<3 x half> %val)
+  ret void
+}
+
+define amdgpu_gfx_callable void @test_call_external_void_func_v3i16_imm() #0 {
+; GFX9-LABEL: test_call_external_void_func_v3i16_imm:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    s_or_saveexec_b64 s[4:5], -1
+; GFX9-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX9-NEXT:    s_mov_b64 exec, s[4:5]
+; GFX9-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX9-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX9-NEXT:    s_mov_b32 s33, s32
+; GFX9-NEXT:    s_add_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_mov_b32_e32 v0, 0x20001
+; GFX9-NEXT:    v_mov_b32_e32 v1, 3
+; GFX9-NEXT:    s_getpc_b64 s[4:5]
+; GFX9-NEXT:    s_add_u32 s4, s4, external_void_func_v3i16@rel32@lo+4
+; GFX9-NEXT:    s_addc_u32 s5, s5, external_void_func_v3i16@rel32@hi+12
+; GFX9-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX9-NEXT:    s_swappc_b64 s[30:31], s[4:5]
+; GFX9-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX9-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX9-NEXT:    s_sub_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX9-NEXT:    s_or_saveexec_b64 s[6:7], -1
+; GFX9-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX9-NEXT:    s_mov_b64 exec, s[6:7]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    s_setpc_b64 s[4:5]
+;
+; GFX10-LABEL: test_call_external_void_func_v3i16_imm:
+; GFX10:       ; %bb.0:
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    s_or_saveexec_b32 s4, -1
+; GFX10-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s4
+; GFX10-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX10-NEXT:    v_mov_b32_e32 v0, 0x20001
+; GFX10-NEXT:    v_mov_b32_e32 v1, 3
+; GFX10-NEXT:    s_mov_b32 s33, s32
+; GFX10-NEXT:    s_add_u32 s32, s32, 0x200
+; GFX10-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX10-NEXT:    s_getpc_b64 s[4:5]
+; GFX10-NEXT:    s_add_u32 s4, s4, external_void_func_v3i16@rel32@lo+4
+; GFX10-NEXT:    s_addc_u32 s5, s5, external_void_func_v3i16@rel32@hi+12
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX10-NEXT:    s_swappc_b64 s[30:31], s[4:5]
+; GFX10-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX10-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX10-NEXT:    s_sub_u32 s32, s32, 0x200
+; GFX10-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX10-NEXT:    s_or_saveexec_b32 s6, -1
+; GFX10-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s6
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    s_setpc_b64 s[4:5]
+  call amdgpu_gfx_callable void @external_void_func_v3i16(<3 x i16> <i16 1, i16 2, i16 3>)
+  ret void
+}
+
+define amdgpu_gfx_callable void @test_call_external_void_func_v3f16_imm() #0 {
+; GFX9-LABEL: test_call_external_void_func_v3f16_imm:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    s_or_saveexec_b64 s[4:5], -1
+; GFX9-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX9-NEXT:    s_mov_b64 exec, s[4:5]
+; GFX9-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX9-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX9-NEXT:    s_mov_b32 s33, s32
+; GFX9-NEXT:    s_add_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_mov_b32_e32 v0, 0x40003c00
+; GFX9-NEXT:    v_mov_b32_e32 v1, 0x4400
+; GFX9-NEXT:    s_getpc_b64 s[4:5]
+; GFX9-NEXT:    s_add_u32 s4, s4, external_void_func_v3f16@rel32@lo+4
+; GFX9-NEXT:    s_addc_u32 s5, s5, external_void_func_v3f16@rel32@hi+12
+; GFX9-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX9-NEXT:    s_swappc_b64 s[30:31], s[4:5]
+; GFX9-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX9-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX9-NEXT:    s_sub_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX9-NEXT:    s_or_saveexec_b64 s[6:7], -1
+; GFX9-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX9-NEXT:    s_mov_b64 exec, s[6:7]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    s_setpc_b64 s[4:5]
+;
+; GFX10-LABEL: test_call_external_void_func_v3f16_imm:
+; GFX10:       ; %bb.0:
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    s_or_saveexec_b32 s4, -1
+; GFX10-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s4
+; GFX10-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX10-NEXT:    v_mov_b32_e32 v0, 0x40003c00
+; GFX10-NEXT:    v_mov_b32_e32 v1, 0x4400
+; GFX10-NEXT:    s_mov_b32 s33, s32
+; GFX10-NEXT:    s_add_u32 s32, s32, 0x200
+; GFX10-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX10-NEXT:    s_getpc_b64 s[4:5]
+; GFX10-NEXT:    s_add_u32 s4, s4, external_void_func_v3f16@rel32@lo+4
+; GFX10-NEXT:    s_addc_u32 s5, s5, external_void_func_v3f16@rel32@hi+12
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX10-NEXT:    s_swappc_b64 s[30:31], s[4:5]
+; GFX10-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX10-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX10-NEXT:    s_sub_u32 s32, s32, 0x200
+; GFX10-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX10-NEXT:    s_or_saveexec_b32 s6, -1
+; GFX10-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s6
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    s_setpc_b64 s[4:5]
+  call amdgpu_gfx_callable void @external_void_func_v3f16(<3 x half> <half 1.0, half 2.0, half 4.0>)
+  ret void
+}
+
+define amdgpu_gfx_callable void @test_call_external_void_func_v4i16() #0 {
+; GFX9-LABEL: test_call_external_void_func_v4i16:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    s_or_saveexec_b64 s[4:5], -1
+; GFX9-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX9-NEXT:    s_mov_b64 exec, s[4:5]
+; GFX9-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX9-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX9-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX9-NEXT:    global_load_dwordx2 v[0:1], v[0:1], off
+; GFX9-NEXT:    s_mov_b32 s33, s32
+; GFX9-NEXT:    s_add_u32 s32, s32, 0x400
+; GFX9-NEXT:    s_getpc_b64 s[4:5]
+; GFX9-NEXT:    s_add_u32 s4, s4, external_void_func_v4i16@rel32@lo+4
+; GFX9-NEXT:    s_addc_u32 s5, s5, external_void_func_v4i16@rel32@hi+12
+; GFX9-NEXT:    s_swappc_b64 s[30:31], s[4:5]
+; GFX9-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX9-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX9-NEXT:    s_sub_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX9-NEXT:    s_or_saveexec_b64 s[6:7], -1
+; GFX9-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX9-NEXT:    s_mov_b64 exec, s[6:7]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    s_setpc_b64 s[4:5]
+;
+; GFX10-LABEL: test_call_external_void_func_v4i16:
+; GFX10:       ; %bb.0:
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    s_or_saveexec_b32 s4, -1
+; GFX10-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s4
+; GFX10-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX10-NEXT:    s_mov_b32 s33, s32
+; GFX10-NEXT:    s_add_u32 s32, s32, 0x200
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    s_getpc_b64 s[4:5]
+; GFX10-NEXT:    s_add_u32 s4, s4, external_void_func_v4i16@rel32@lo+4
+; GFX10-NEXT:    s_addc_u32 s5, s5, external_void_func_v4i16@rel32@hi+12
+; GFX10-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX10-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX10-NEXT:    global_load_dwordx2 v[0:1], v[0:1], off
+; GFX10-NEXT:    s_swappc_b64 s[30:31], s[4:5]
+; GFX10-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX10-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX10-NEXT:    s_sub_u32 s32, s32, 0x200
+; GFX10-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX10-NEXT:    s_or_saveexec_b32 s6, -1
+; GFX10-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s6
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    s_setpc_b64 s[4:5]
+  %val = load <4 x i16>, <4 x i16> addrspace(1)* undef
+  call amdgpu_gfx_callable void @external_void_func_v4i16(<4 x i16> %val)
+  ret void
+}
+
+define amdgpu_gfx_callable void @test_call_external_void_func_v4i16_imm() #0 {
+; GFX9-LABEL: test_call_external_void_func_v4i16_imm:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    s_or_saveexec_b64 s[4:5], -1
+; GFX9-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX9-NEXT:    s_mov_b64 exec, s[4:5]
+; GFX9-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX9-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX9-NEXT:    s_mov_b32 s33, s32
+; GFX9-NEXT:    s_add_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_mov_b32_e32 v0, 0x20001
+; GFX9-NEXT:    v_mov_b32_e32 v1, 0x40003
+; GFX9-NEXT:    s_getpc_b64 s[4:5]
+; GFX9-NEXT:    s_add_u32 s4, s4, external_void_func_v4i16@rel32@lo+4
+; GFX9-NEXT:    s_addc_u32 s5, s5, external_void_func_v4i16@rel32@hi+12
+; GFX9-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX9-NEXT:    s_swappc_b64 s[30:31], s[4:5]
+; GFX9-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX9-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX9-NEXT:    s_sub_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX9-NEXT:    s_or_saveexec_b64 s[6:7], -1
+; GFX9-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX9-NEXT:    s_mov_b64 exec, s[6:7]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    s_setpc_b64 s[4:5]
+;
+; GFX10-LABEL: test_call_external_void_func_v4i16_imm:
+; GFX10:       ; %bb.0:
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    s_or_saveexec_b32 s4, -1
+; GFX10-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s4
+; GFX10-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX10-NEXT:    v_mov_b32_e32 v0, 0x20001
+; GFX10-NEXT:    v_mov_b32_e32 v1, 0x40003
+; GFX10-NEXT:    s_mov_b32 s33, s32
+; GFX10-NEXT:    s_add_u32 s32, s32, 0x200
+; GFX10-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX10-NEXT:    s_getpc_b64 s[4:5]
+; GFX10-NEXT:    s_add_u32 s4, s4, external_void_func_v4i16@rel32@lo+4
+; GFX10-NEXT:    s_addc_u32 s5, s5, external_void_func_v4i16@rel32@hi+12
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX10-NEXT:    s_swappc_b64 s[30:31], s[4:5]
+; GFX10-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX10-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX10-NEXT:    s_sub_u32 s32, s32, 0x200
+; GFX10-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX10-NEXT:    s_or_saveexec_b32 s6, -1
+; GFX10-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s6
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    s_setpc_b64 s[4:5]
+  call amdgpu_gfx_callable void @external_void_func_v4i16(<4 x i16> <i16 1, i16 2, i16 3, i16 4>)
+  ret void
+}
+
+define amdgpu_gfx_callable void @test_call_external_void_func_v2f16() #0 {
+; GFX9-LABEL: test_call_external_void_func_v2f16:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    s_or_saveexec_b64 s[4:5], -1
+; GFX9-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX9-NEXT:    s_mov_b64 exec, s[4:5]
+; GFX9-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX9-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX9-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX9-NEXT:    global_load_dword v0, v[0:1], off
+; GFX9-NEXT:    s_mov_b32 s33, s32
+; GFX9-NEXT:    s_add_u32 s32, s32, 0x400
+; GFX9-NEXT:    s_getpc_b64 s[4:5]
+; GFX9-NEXT:    s_add_u32 s4, s4, external_void_func_v2f16@rel32@lo+4
+; GFX9-NEXT:    s_addc_u32 s5, s5, external_void_func_v2f16@rel32@hi+12
+; GFX9-NEXT:    s_swappc_b64 s[30:31], s[4:5]
+; GFX9-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX9-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX9-NEXT:    s_sub_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX9-NEXT:    s_or_saveexec_b64 s[6:7], -1
+; GFX9-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX9-NEXT:    s_mov_b64 exec, s[6:7]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    s_setpc_b64 s[4:5]
+;
+; GFX10-LABEL: test_call_external_void_func_v2f16:
+; GFX10:       ; %bb.0:
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    s_or_saveexec_b32 s4, -1
+; GFX10-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s4
+; GFX10-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX10-NEXT:    s_mov_b32 s33, s32
+; GFX10-NEXT:    s_add_u32 s32, s32, 0x200
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    s_getpc_b64 s[4:5]
+; GFX10-NEXT:    s_add_u32 s4, s4, external_void_func_v2f16@rel32@lo+4
+; GFX10-NEXT:    s_addc_u32 s5, s5, external_void_func_v2f16@rel32@hi+12
+; GFX10-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX10-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX10-NEXT:    global_load_dword v0, v[0:1], off
+; GFX10-NEXT:    s_swappc_b64 s[30:31], s[4:5]
+; GFX10-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX10-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX10-NEXT:    s_sub_u32 s32, s32, 0x200
+; GFX10-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX10-NEXT:    s_or_saveexec_b32 s6, -1
+; GFX10-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s6
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    s_setpc_b64 s[4:5]
+  %val = load <2 x half>, <2 x half> addrspace(1)* undef
+  call amdgpu_gfx_callable void @external_void_func_v2f16(<2 x half> %val)
+  ret void
+}
+
+define amdgpu_gfx_callable void @test_call_external_void_func_v2i32() #0 {
+; GFX9-LABEL: test_call_external_void_func_v2i32:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    s_or_saveexec_b64 s[4:5], -1
+; GFX9-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX9-NEXT:    s_mov_b64 exec, s[4:5]
+; GFX9-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX9-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX9-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX9-NEXT:    global_load_dwordx2 v[0:1], v[0:1], off
+; GFX9-NEXT:    s_mov_b32 s33, s32
+; GFX9-NEXT:    s_add_u32 s32, s32, 0x400
+; GFX9-NEXT:    s_getpc_b64 s[4:5]
+; GFX9-NEXT:    s_add_u32 s4, s4, external_void_func_v2i32@rel32@lo+4
+; GFX9-NEXT:    s_addc_u32 s5, s5, external_void_func_v2i32@rel32@hi+12
+; GFX9-NEXT:    s_swappc_b64 s[30:31], s[4:5]
+; GFX9-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX9-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX9-NEXT:    s_sub_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX9-NEXT:    s_or_saveexec_b64 s[6:7], -1
+; GFX9-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX9-NEXT:    s_mov_b64 exec, s[6:7]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    s_setpc_b64 s[4:5]
+;
+; GFX10-LABEL: test_call_external_void_func_v2i32:
+; GFX10:       ; %bb.0:
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    s_or_saveexec_b32 s4, -1
+; GFX10-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s4
+; GFX10-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX10-NEXT:    s_mov_b32 s33, s32
+; GFX10-NEXT:    s_add_u32 s32, s32, 0x200
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    s_getpc_b64 s[4:5]
+; GFX10-NEXT:    s_add_u32 s4, s4, external_void_func_v2i32@rel32@lo+4
+; GFX10-NEXT:    s_addc_u32 s5, s5, external_void_func_v2i32@rel32@hi+12
+; GFX10-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX10-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX10-NEXT:    global_load_dwordx2 v[0:1], v[0:1], off
+; GFX10-NEXT:    s_swappc_b64 s[30:31], s[4:5]
+; GFX10-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX10-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX10-NEXT:    s_sub_u32 s32, s32, 0x200
+; GFX10-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX10-NEXT:    s_or_saveexec_b32 s6, -1
+; GFX10-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s6
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    s_setpc_b64 s[4:5]
+  %val = load <2 x i32>, <2 x i32> addrspace(1)* undef
+  call amdgpu_gfx_callable void @external_void_func_v2i32(<2 x i32> %val)
+  ret void
+}
+
+define amdgpu_gfx_callable void @test_call_external_void_func_v2i32_imm() #0 {
+; GFX9-LABEL: test_call_external_void_func_v2i32_imm:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    s_or_saveexec_b64 s[4:5], -1
+; GFX9-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX9-NEXT:    s_mov_b64 exec, s[4:5]
+; GFX9-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX9-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX9-NEXT:    s_mov_b32 s33, s32
+; GFX9-NEXT:    s_add_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_mov_b32_e32 v0, 1
+; GFX9-NEXT:    v_mov_b32_e32 v1, 2
+; GFX9-NEXT:    s_getpc_b64 s[4:5]
+; GFX9-NEXT:    s_add_u32 s4, s4, external_void_func_v2i32@rel32@lo+4
+; GFX9-NEXT:    s_addc_u32 s5, s5, external_void_func_v2i32@rel32@hi+12
+; GFX9-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX9-NEXT:    s_swappc_b64 s[30:31], s[4:5]
+; GFX9-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX9-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX9-NEXT:    s_sub_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX9-NEXT:    s_or_saveexec_b64 s[6:7], -1
+; GFX9-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX9-NEXT:    s_mov_b64 exec, s[6:7]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    s_setpc_b64 s[4:5]
+;
+; GFX10-LABEL: test_call_external_void_func_v2i32_imm:
+; GFX10:       ; %bb.0:
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    s_or_saveexec_b32 s4, -1
+; GFX10-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s4
+; GFX10-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX10-NEXT:    v_mov_b32_e32 v0, 1
+; GFX10-NEXT:    v_mov_b32_e32 v1, 2
+; GFX10-NEXT:    s_mov_b32 s33, s32
+; GFX10-NEXT:    s_add_u32 s32, s32, 0x200
+; GFX10-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX10-NEXT:    s_getpc_b64 s[4:5]
+; GFX10-NEXT:    s_add_u32 s4, s4, external_void_func_v2i32@rel32@lo+4
+; GFX10-NEXT:    s_addc_u32 s5, s5, external_void_func_v2i32@rel32@hi+12
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX10-NEXT:    s_swappc_b64 s[30:31], s[4:5]
+; GFX10-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX10-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX10-NEXT:    s_sub_u32 s32, s32, 0x200
+; GFX10-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX10-NEXT:    s_or_saveexec_b32 s6, -1
+; GFX10-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s6
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    s_setpc_b64 s[4:5]
+  call amdgpu_gfx_callable void @external_void_func_v2i32(<2 x i32> <i32 1, i32 2>)
+  ret void
+}
+
+define amdgpu_gfx_callable void @test_call_external_void_func_v3i32_imm(i32) #0 {
+; GFX9-LABEL: test_call_external_void_func_v3i32_imm:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    s_or_saveexec_b64 s[4:5], -1
+; GFX9-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX9-NEXT:    s_mov_b64 exec, s[4:5]
+; GFX9-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX9-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX9-NEXT:    s_mov_b32 s33, s32
+; GFX9-NEXT:    s_add_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_mov_b32_e32 v0, 3
+; GFX9-NEXT:    v_mov_b32_e32 v1, 4
+; GFX9-NEXT:    v_mov_b32_e32 v2, 5
+; GFX9-NEXT:    s_getpc_b64 s[4:5]
+; GFX9-NEXT:    s_add_u32 s4, s4, external_void_func_v3i32@rel32@lo+4
+; GFX9-NEXT:    s_addc_u32 s5, s5, external_void_func_v3i32@rel32@hi+12
+; GFX9-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX9-NEXT:    s_swappc_b64 s[30:31], s[4:5]
+; GFX9-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX9-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX9-NEXT:    s_sub_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX9-NEXT:    s_or_saveexec_b64 s[6:7], -1
+; GFX9-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX9-NEXT:    s_mov_b64 exec, s[6:7]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    s_setpc_b64 s[4:5]
+;
+; GFX10-LABEL: test_call_external_void_func_v3i32_imm:
+; GFX10:       ; %bb.0:
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    s_or_saveexec_b32 s4, -1
+; GFX10-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s4
+; GFX10-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX10-NEXT:    v_mov_b32_e32 v0, 3
+; GFX10-NEXT:    v_mov_b32_e32 v1, 4
+; GFX10-NEXT:    v_mov_b32_e32 v2, 5
+; GFX10-NEXT:    s_mov_b32 s33, s32
+; GFX10-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX10-NEXT:    s_add_u32 s32, s32, 0x200
+; GFX10-NEXT:    s_getpc_b64 s[4:5]
+; GFX10-NEXT:    s_add_u32 s4, s4, external_void_func_v3i32@rel32@lo+4
+; GFX10-NEXT:    s_addc_u32 s5, s5, external_void_func_v3i32@rel32@hi+12
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX10-NEXT:    s_swappc_b64 s[30:31], s[4:5]
+; GFX10-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX10-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX10-NEXT:    s_sub_u32 s32, s32, 0x200
+; GFX10-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX10-NEXT:    s_or_saveexec_b32 s6, -1
+; GFX10-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s6
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    s_setpc_b64 s[4:5]
+  call amdgpu_gfx_callable void @external_void_func_v3i32(<3 x i32> <i32 3, i32 4, i32 5>)
+  ret void
+}
+
+define amdgpu_gfx_callable void @test_call_external_void_func_v3i32_i32(i32) #0 {
+; GFX9-LABEL: test_call_external_void_func_v3i32_i32:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    s_or_saveexec_b64 s[4:5], -1
+; GFX9-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX9-NEXT:    s_mov_b64 exec, s[4:5]
+; GFX9-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX9-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX9-NEXT:    s_mov_b32 s33, s32
+; GFX9-NEXT:    s_add_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_mov_b32_e32 v0, 3
+; GFX9-NEXT:    v_mov_b32_e32 v1, 4
+; GFX9-NEXT:    v_mov_b32_e32 v2, 5
+; GFX9-NEXT:    v_mov_b32_e32 v3, 6
+; GFX9-NEXT:    s_getpc_b64 s[4:5]
+; GFX9-NEXT:    s_add_u32 s4, s4, external_void_func_v3i32_i32@rel32@lo+4
+; GFX9-NEXT:    s_addc_u32 s5, s5, external_void_func_v3i32_i32@rel32@hi+12
+; GFX9-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX9-NEXT:    s_swappc_b64 s[30:31], s[4:5]
+; GFX9-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX9-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX9-NEXT:    s_sub_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX9-NEXT:    s_or_saveexec_b64 s[6:7], -1
+; GFX9-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX9-NEXT:    s_mov_b64 exec, s[6:7]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    s_setpc_b64 s[4:5]
+;
+; GFX10-LABEL: test_call_external_void_func_v3i32_i32:
+; GFX10:       ; %bb.0:
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    s_or_saveexec_b32 s4, -1
+; GFX10-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s4
+; GFX10-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX10-NEXT:    v_mov_b32_e32 v0, 3
+; GFX10-NEXT:    v_mov_b32_e32 v1, 4
+; GFX10-NEXT:    v_mov_b32_e32 v2, 5
+; GFX10-NEXT:    v_mov_b32_e32 v3, 6
+; GFX10-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX10-NEXT:    s_mov_b32 s33, s32
+; GFX10-NEXT:    s_add_u32 s32, s32, 0x200
+; GFX10-NEXT:    s_getpc_b64 s[4:5]
+; GFX10-NEXT:    s_add_u32 s4, s4, external_void_func_v3i32_i32@rel32@lo+4
+; GFX10-NEXT:    s_addc_u32 s5, s5, external_void_func_v3i32_i32@rel32@hi+12
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX10-NEXT:    s_swappc_b64 s[30:31], s[4:5]
+; GFX10-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX10-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX10-NEXT:    s_sub_u32 s32, s32, 0x200
+; GFX10-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX10-NEXT:    s_or_saveexec_b32 s6, -1
+; GFX10-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s6
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    s_setpc_b64 s[4:5]
+  call amdgpu_gfx_callable void @external_void_func_v3i32_i32(<3 x i32> <i32 3, i32 4, i32 5>, i32 6)
+  ret void
+}
+
+define amdgpu_gfx_callable void @test_call_external_void_func_v4i32() #0 {
+; GFX9-LABEL: test_call_external_void_func_v4i32:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    s_or_saveexec_b64 s[4:5], -1
+; GFX9-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX9-NEXT:    s_mov_b64 exec, s[4:5]
+; GFX9-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX9-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX9-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX9-NEXT:    global_load_dwordx4 v[0:3], v[0:1], off
+; GFX9-NEXT:    s_mov_b32 s33, s32
+; GFX9-NEXT:    s_add_u32 s32, s32, 0x400
+; GFX9-NEXT:    s_getpc_b64 s[4:5]
+; GFX9-NEXT:    s_add_u32 s4, s4, external_void_func_v4i32@rel32@lo+4
+; GFX9-NEXT:    s_addc_u32 s5, s5, external_void_func_v4i32@rel32@hi+12
+; GFX9-NEXT:    s_swappc_b64 s[30:31], s[4:5]
+; GFX9-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX9-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX9-NEXT:    s_sub_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX9-NEXT:    s_or_saveexec_b64 s[6:7], -1
+; GFX9-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX9-NEXT:    s_mov_b64 exec, s[6:7]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    s_setpc_b64 s[4:5]
+;
+; GFX10-LABEL: test_call_external_void_func_v4i32:
+; GFX10:       ; %bb.0:
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    s_or_saveexec_b32 s4, -1
+; GFX10-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s4
+; GFX10-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX10-NEXT:    s_mov_b32 s33, s32
+; GFX10-NEXT:    s_add_u32 s32, s32, 0x200
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    s_getpc_b64 s[4:5]
+; GFX10-NEXT:    s_add_u32 s4, s4, external_void_func_v4i32@rel32@lo+4
+; GFX10-NEXT:    s_addc_u32 s5, s5, external_void_func_v4i32@rel32@hi+12
+; GFX10-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX10-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX10-NEXT:    global_load_dwordx4 v[0:3], v[0:1], off
+; GFX10-NEXT:    s_swappc_b64 s[30:31], s[4:5]
+; GFX10-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX10-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX10-NEXT:    s_sub_u32 s32, s32, 0x200
+; GFX10-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX10-NEXT:    s_or_saveexec_b32 s6, -1
+; GFX10-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s6
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    s_setpc_b64 s[4:5]
+  %val = load <4 x i32>, <4 x i32> addrspace(1)* undef
+  call amdgpu_gfx_callable void @external_void_func_v4i32(<4 x i32> %val)
+  ret void
+}
+
+define amdgpu_gfx_callable void @test_call_external_void_func_v4i32_imm() #0 {
+; GFX9-LABEL: test_call_external_void_func_v4i32_imm:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    s_or_saveexec_b64 s[4:5], -1
+; GFX9-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX9-NEXT:    s_mov_b64 exec, s[4:5]
+; GFX9-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX9-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX9-NEXT:    s_mov_b32 s33, s32
+; GFX9-NEXT:    s_add_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_mov_b32_e32 v0, 1
+; GFX9-NEXT:    v_mov_b32_e32 v1, 2
+; GFX9-NEXT:    v_mov_b32_e32 v2, 3
+; GFX9-NEXT:    v_mov_b32_e32 v3, 4
+; GFX9-NEXT:    s_getpc_b64 s[4:5]
+; GFX9-NEXT:    s_add_u32 s4, s4, external_void_func_v4i32@rel32@lo+4
+; GFX9-NEXT:    s_addc_u32 s5, s5, external_void_func_v4i32@rel32@hi+12
+; GFX9-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX9-NEXT:    s_swappc_b64 s[30:31], s[4:5]
+; GFX9-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX9-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX9-NEXT:    s_sub_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX9-NEXT:    s_or_saveexec_b64 s[6:7], -1
+; GFX9-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX9-NEXT:    s_mov_b64 exec, s[6:7]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    s_setpc_b64 s[4:5]
+;
+; GFX10-LABEL: test_call_external_void_func_v4i32_imm:
+; GFX10:       ; %bb.0:
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    s_or_saveexec_b32 s4, -1
+; GFX10-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s4
+; GFX10-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX10-NEXT:    v_mov_b32_e32 v0, 1
+; GFX10-NEXT:    v_mov_b32_e32 v1, 2
+; GFX10-NEXT:    v_mov_b32_e32 v2, 3
+; GFX10-NEXT:    v_mov_b32_e32 v3, 4
+; GFX10-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX10-NEXT:    s_mov_b32 s33, s32
+; GFX10-NEXT:    s_add_u32 s32, s32, 0x200
+; GFX10-NEXT:    s_getpc_b64 s[4:5]
+; GFX10-NEXT:    s_add_u32 s4, s4, external_void_func_v4i32@rel32@lo+4
+; GFX10-NEXT:    s_addc_u32 s5, s5, external_void_func_v4i32@rel32@hi+12
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX10-NEXT:    s_swappc_b64 s[30:31], s[4:5]
+; GFX10-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX10-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX10-NEXT:    s_sub_u32 s32, s32, 0x200
+; GFX10-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX10-NEXT:    s_or_saveexec_b32 s6, -1
+; GFX10-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s6
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    s_setpc_b64 s[4:5]
+  call amdgpu_gfx_callable void @external_void_func_v4i32(<4 x i32> <i32 1, i32 2, i32 3, i32 4>)
+  ret void
+}
+
+define amdgpu_gfx_callable void @test_call_external_void_func_v5i32_imm() #0 {
+; GFX9-LABEL: test_call_external_void_func_v5i32_imm:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    s_or_saveexec_b64 s[4:5], -1
+; GFX9-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX9-NEXT:    s_mov_b64 exec, s[4:5]
+; GFX9-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX9-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX9-NEXT:    s_mov_b32 s33, s32
+; GFX9-NEXT:    s_add_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_mov_b32_e32 v0, 1
+; GFX9-NEXT:    v_mov_b32_e32 v1, 2
+; GFX9-NEXT:    v_mov_b32_e32 v2, 3
+; GFX9-NEXT:    v_mov_b32_e32 v3, 4
+; GFX9-NEXT:    v_mov_b32_e32 v4, 5
+; GFX9-NEXT:    s_getpc_b64 s[4:5]
+; GFX9-NEXT:    s_add_u32 s4, s4, external_void_func_v5i32@rel32@lo+4
+; GFX9-NEXT:    s_addc_u32 s5, s5, external_void_func_v5i32@rel32@hi+12
+; GFX9-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX9-NEXT:    s_swappc_b64 s[30:31], s[4:5]
+; GFX9-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX9-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX9-NEXT:    s_sub_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX9-NEXT:    s_or_saveexec_b64 s[6:7], -1
+; GFX9-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX9-NEXT:    s_mov_b64 exec, s[6:7]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    s_setpc_b64 s[4:5]
+;
+; GFX10-LABEL: test_call_external_void_func_v5i32_imm:
+; GFX10:       ; %bb.0:
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    s_or_saveexec_b32 s4, -1
+; GFX10-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s4
+; GFX10-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX10-NEXT:    v_mov_b32_e32 v0, 1
+; GFX10-NEXT:    v_mov_b32_e32 v1, 2
+; GFX10-NEXT:    v_mov_b32_e32 v2, 3
+; GFX10-NEXT:    v_mov_b32_e32 v3, 4
+; GFX10-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX10-NEXT:    v_mov_b32_e32 v4, 5
+; GFX10-NEXT:    s_mov_b32 s33, s32
+; GFX10-NEXT:    s_add_u32 s32, s32, 0x200
+; GFX10-NEXT:    s_getpc_b64 s[4:5]
+; GFX10-NEXT:    s_add_u32 s4, s4, external_void_func_v5i32@rel32@lo+4
+; GFX10-NEXT:    s_addc_u32 s5, s5, external_void_func_v5i32@rel32@hi+12
+; GFX10-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    s_swappc_b64 s[30:31], s[4:5]
+; GFX10-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX10-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX10-NEXT:    s_sub_u32 s32, s32, 0x200
+; GFX10-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX10-NEXT:    s_or_saveexec_b32 s6, -1
+; GFX10-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s6
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    s_setpc_b64 s[4:5]
+  call amdgpu_gfx_callable void @external_void_func_v5i32(<5 x i32> <i32 1, i32 2, i32 3, i32 4, i32 5>)
+  ret void
+}
+
+define amdgpu_gfx_callable void @test_call_external_void_func_v8i32() #0 {
+; GFX9-LABEL: test_call_external_void_func_v8i32:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    s_or_saveexec_b64 s[4:5], -1
+; GFX9-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX9-NEXT:    s_mov_b64 exec, s[4:5]
+; GFX9-NEXT:    s_load_dwordx2 s[4:5], s[4:5], 0x0
+; GFX9-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX9-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX9-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX9-NEXT:    s_mov_b32 s33, s32
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    v_mov_b32_e32 v4, s4
+; GFX9-NEXT:    v_mov_b32_e32 v5, s5
+; GFX9-NEXT:    global_load_dwordx4 v[0:3], v[4:5], off
+; GFX9-NEXT:    global_load_dwordx4 v[4:7], v[4:5], off offset:16
+; GFX9-NEXT:    s_add_u32 s32, s32, 0x400
+; GFX9-NEXT:    s_getpc_b64 s[4:5]
+; GFX9-NEXT:    s_add_u32 s4, s4, external_void_func_v8i32@rel32@lo+4
+; GFX9-NEXT:    s_addc_u32 s5, s5, external_void_func_v8i32@rel32@hi+12
+; GFX9-NEXT:    s_swappc_b64 s[30:31], s[4:5]
+; GFX9-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX9-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX9-NEXT:    s_sub_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX9-NEXT:    s_or_saveexec_b64 s[6:7], -1
+; GFX9-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX9-NEXT:    s_mov_b64 exec, s[6:7]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    s_setpc_b64 s[4:5]
+;
+; GFX10-LABEL: test_call_external_void_func_v8i32:
+; GFX10:       ; %bb.0:
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    s_or_saveexec_b32 s4, -1
+; GFX10-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s4
+; GFX10-NEXT:    s_load_dwordx2 s[4:5], s[4:5], 0x0
+; GFX10-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX10-NEXT:    s_mov_b32 s33, s32
+; GFX10-NEXT:    s_add_u32 s32, s32, 0x200
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX10-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX10-NEXT:    v_mov_b32_e32 v4, s4
+; GFX10-NEXT:    v_mov_b32_e32 v5, s5
+; GFX10-NEXT:    s_getpc_b64 s[4:5]
+; GFX10-NEXT:    s_add_u32 s4, s4, external_void_func_v8i32@rel32@lo+4
+; GFX10-NEXT:    s_addc_u32 s5, s5, external_void_func_v8i32@rel32@hi+12
+; GFX10-NEXT:    s_clause 0x1
+; GFX10-NEXT:    global_load_dwordx4 v[0:3], v[4:5], off
+; GFX10-NEXT:    global_load_dwordx4 v[4:7], v[4:5], off offset:16
+; GFX10-NEXT:    s_swappc_b64 s[30:31], s[4:5]
+; GFX10-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX10-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX10-NEXT:    s_sub_u32 s32, s32, 0x200
+; GFX10-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX10-NEXT:    s_or_saveexec_b32 s6, -1
+; GFX10-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s6
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    s_setpc_b64 s[4:5]
+  %ptr = load <8 x i32> addrspace(1)*, <8 x i32> addrspace(1)* addrspace(4)* undef
+  %val = load <8 x i32>, <8 x i32> addrspace(1)* %ptr
+  call amdgpu_gfx_callable void @external_void_func_v8i32(<8 x i32> %val)
+  ret void
+}
+
+define amdgpu_gfx_callable void @test_call_external_void_func_v8i32_imm() #0 {
+; GFX9-LABEL: test_call_external_void_func_v8i32_imm:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    s_or_saveexec_b64 s[4:5], -1
+; GFX9-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX9-NEXT:    s_mov_b64 exec, s[4:5]
+; GFX9-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX9-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX9-NEXT:    s_mov_b32 s33, s32
+; GFX9-NEXT:    s_add_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_mov_b32_e32 v0, 1
+; GFX9-NEXT:    v_mov_b32_e32 v1, 2
+; GFX9-NEXT:    v_mov_b32_e32 v2, 3
+; GFX9-NEXT:    v_mov_b32_e32 v3, 4
+; GFX9-NEXT:    v_mov_b32_e32 v4, 5
+; GFX9-NEXT:    v_mov_b32_e32 v5, 6
+; GFX9-NEXT:    v_mov_b32_e32 v6, 7
+; GFX9-NEXT:    v_mov_b32_e32 v7, 8
+; GFX9-NEXT:    s_getpc_b64 s[4:5]
+; GFX9-NEXT:    s_add_u32 s4, s4, external_void_func_v8i32@rel32@lo+4
+; GFX9-NEXT:    s_addc_u32 s5, s5, external_void_func_v8i32@rel32@hi+12
+; GFX9-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX9-NEXT:    s_swappc_b64 s[30:31], s[4:5]
+; GFX9-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX9-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX9-NEXT:    s_sub_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX9-NEXT:    s_or_saveexec_b64 s[6:7], -1
+; GFX9-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX9-NEXT:    s_mov_b64 exec, s[6:7]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    s_setpc_b64 s[4:5]
+;
+; GFX10-LABEL: test_call_external_void_func_v8i32_imm:
+; GFX10:       ; %bb.0:
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    s_or_saveexec_b32 s4, -1
+; GFX10-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s4
+; GFX10-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX10-NEXT:    v_mov_b32_e32 v0, 1
+; GFX10-NEXT:    v_mov_b32_e32 v1, 2
+; GFX10-NEXT:    v_mov_b32_e32 v2, 3
+; GFX10-NEXT:    v_mov_b32_e32 v3, 4
+; GFX10-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX10-NEXT:    v_mov_b32_e32 v4, 5
+; GFX10-NEXT:    v_mov_b32_e32 v5, 6
+; GFX10-NEXT:    v_mov_b32_e32 v6, 7
+; GFX10-NEXT:    v_mov_b32_e32 v7, 8
+; GFX10-NEXT:    s_mov_b32 s33, s32
+; GFX10-NEXT:    s_add_u32 s32, s32, 0x200
+; GFX10-NEXT:    s_getpc_b64 s[4:5]
+; GFX10-NEXT:    s_add_u32 s4, s4, external_void_func_v8i32@rel32@lo+4
+; GFX10-NEXT:    s_addc_u32 s5, s5, external_void_func_v8i32@rel32@hi+12
+; GFX10-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    s_swappc_b64 s[30:31], s[4:5]
+; GFX10-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX10-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX10-NEXT:    s_sub_u32 s32, s32, 0x200
+; GFX10-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX10-NEXT:    s_or_saveexec_b32 s6, -1
+; GFX10-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s6
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    s_setpc_b64 s[4:5]
+  call amdgpu_gfx_callable void @external_void_func_v8i32(<8 x i32> <i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8>)
+  ret void
+}
+
+define amdgpu_gfx_callable void @test_call_external_void_func_v16i32() #0 {
+; GFX9-LABEL: test_call_external_void_func_v16i32:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    s_or_saveexec_b64 s[4:5], -1
+; GFX9-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX9-NEXT:    s_mov_b64 exec, s[4:5]
+; GFX9-NEXT:    s_load_dwordx2 s[4:5], s[4:5], 0x0
+; GFX9-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX9-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX9-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX9-NEXT:    s_mov_b32 s33, s32
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    v_mov_b32_e32 v13, s5
+; GFX9-NEXT:    v_mov_b32_e32 v12, s4
+; GFX9-NEXT:    global_load_dwordx4 v[0:3], v[12:13], off
+; GFX9-NEXT:    global_load_dwordx4 v[4:7], v[12:13], off offset:16
+; GFX9-NEXT:    global_load_dwordx4 v[8:11], v[12:13], off offset:32
+; GFX9-NEXT:    global_load_dwordx4 v[12:15], v[12:13], off offset:48
+; GFX9-NEXT:    s_add_u32 s32, s32, 0x400
+; GFX9-NEXT:    s_getpc_b64 s[4:5]
+; GFX9-NEXT:    s_add_u32 s4, s4, external_void_func_v16i32@rel32@lo+4
+; GFX9-NEXT:    s_addc_u32 s5, s5, external_void_func_v16i32@rel32@hi+12
+; GFX9-NEXT:    s_swappc_b64 s[30:31], s[4:5]
+; GFX9-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX9-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX9-NEXT:    s_sub_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX9-NEXT:    s_or_saveexec_b64 s[6:7], -1
+; GFX9-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX9-NEXT:    s_mov_b64 exec, s[6:7]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    s_setpc_b64 s[4:5]
+;
+; GFX10-LABEL: test_call_external_void_func_v16i32:
+; GFX10:       ; %bb.0:
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    s_or_saveexec_b32 s4, -1
+; GFX10-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s4
+; GFX10-NEXT:    s_load_dwordx2 s[4:5], s[4:5], 0x0
+; GFX10-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX10-NEXT:    s_mov_b32 s33, s32
+; GFX10-NEXT:    s_add_u32 s32, s32, 0x200
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX10-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX10-NEXT:    v_mov_b32_e32 v13, s5
+; GFX10-NEXT:    v_mov_b32_e32 v12, s4
+; GFX10-NEXT:    s_getpc_b64 s[4:5]
+; GFX10-NEXT:    s_add_u32 s4, s4, external_void_func_v16i32@rel32@lo+4
+; GFX10-NEXT:    s_addc_u32 s5, s5, external_void_func_v16i32@rel32@hi+12
+; GFX10-NEXT:    s_clause 0x3
+; GFX10-NEXT:    global_load_dwordx4 v[0:3], v[12:13], off
+; GFX10-NEXT:    global_load_dwordx4 v[4:7], v[12:13], off offset:16
+; GFX10-NEXT:    global_load_dwordx4 v[8:11], v[12:13], off offset:32
+; GFX10-NEXT:    global_load_dwordx4 v[12:15], v[12:13], off offset:48
+; GFX10-NEXT:    s_swappc_b64 s[30:31], s[4:5]
+; GFX10-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX10-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX10-NEXT:    s_sub_u32 s32, s32, 0x200
+; GFX10-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX10-NEXT:    s_or_saveexec_b32 s6, -1
+; GFX10-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s6
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    s_setpc_b64 s[4:5]
+  %ptr = load <16 x i32> addrspace(1)*, <16 x i32> addrspace(1)* addrspace(4)* undef
+  %val = load <16 x i32>, <16 x i32> addrspace(1)* %ptr
+  call amdgpu_gfx_callable void @external_void_func_v16i32(<16 x i32> %val)
+  ret void
+}
+
+define amdgpu_gfx_callable void @test_call_external_void_func_v32i32() #0 {
+; GFX9-LABEL: test_call_external_void_func_v32i32:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    s_or_saveexec_b64 s[4:5], -1
+; GFX9-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX9-NEXT:    s_mov_b64 exec, s[4:5]
+; GFX9-NEXT:    s_load_dwordx2 s[4:5], s[4:5], 0x0
+; GFX9-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX9-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX9-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX9-NEXT:    s_mov_b32 s33, s32
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    v_mov_b32_e32 v29, s5
+; GFX9-NEXT:    v_mov_b32_e32 v28, s4
+; GFX9-NEXT:    global_load_dwordx4 v[0:3], v[28:29], off
+; GFX9-NEXT:    global_load_dwordx4 v[4:7], v[28:29], off offset:16
+; GFX9-NEXT:    global_load_dwordx4 v[8:11], v[28:29], off offset:32
+; GFX9-NEXT:    global_load_dwordx4 v[12:15], v[28:29], off offset:48
+; GFX9-NEXT:    global_load_dwordx4 v[16:19], v[28:29], off offset:64
+; GFX9-NEXT:    global_load_dwordx4 v[20:23], v[28:29], off offset:80
+; GFX9-NEXT:    global_load_dwordx4 v[24:27], v[28:29], off offset:96
+; GFX9-NEXT:    global_load_dwordx4 v[28:31], v[28:29], off offset:112
+; GFX9-NEXT:    s_add_u32 s32, s32, 0x400
+; GFX9-NEXT:    s_getpc_b64 s[4:5]
+; GFX9-NEXT:    s_add_u32 s4, s4, external_void_func_v32i32@rel32@lo+4
+; GFX9-NEXT:    s_addc_u32 s5, s5, external_void_func_v32i32@rel32@hi+12
+; GFX9-NEXT:    s_swappc_b64 s[30:31], s[4:5]
+; GFX9-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX9-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX9-NEXT:    s_sub_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX9-NEXT:    s_or_saveexec_b64 s[6:7], -1
+; GFX9-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX9-NEXT:    s_mov_b64 exec, s[6:7]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    s_setpc_b64 s[4:5]
+;
+; GFX10-LABEL: test_call_external_void_func_v32i32:
+; GFX10:       ; %bb.0:
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    s_or_saveexec_b32 s4, -1
+; GFX10-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s4
+; GFX10-NEXT:    s_load_dwordx2 s[4:5], s[4:5], 0x0
+; GFX10-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX10-NEXT:    s_mov_b32 s33, s32
+; GFX10-NEXT:    s_add_u32 s32, s32, 0x200
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX10-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX10-NEXT:    v_mov_b32_e32 v29, s5
+; GFX10-NEXT:    v_mov_b32_e32 v28, s4
+; GFX10-NEXT:    s_getpc_b64 s[4:5]
+; GFX10-NEXT:    s_add_u32 s4, s4, external_void_func_v32i32@rel32@lo+4
+; GFX10-NEXT:    s_addc_u32 s5, s5, external_void_func_v32i32@rel32@hi+12
+; GFX10-NEXT:    s_clause 0x7
+; GFX10-NEXT:    global_load_dwordx4 v[0:3], v[28:29], off
+; GFX10-NEXT:    global_load_dwordx4 v[4:7], v[28:29], off offset:16
+; GFX10-NEXT:    global_load_dwordx4 v[8:11], v[28:29], off offset:32
+; GFX10-NEXT:    global_load_dwordx4 v[12:15], v[28:29], off offset:48
+; GFX10-NEXT:    global_load_dwordx4 v[16:19], v[28:29], off offset:64
+; GFX10-NEXT:    global_load_dwordx4 v[20:23], v[28:29], off offset:80
+; GFX10-NEXT:    global_load_dwordx4 v[24:27], v[28:29], off offset:96
+; GFX10-NEXT:    global_load_dwordx4 v[28:31], v[28:29], off offset:112
+; GFX10-NEXT:    s_swappc_b64 s[30:31], s[4:5]
+; GFX10-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX10-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX10-NEXT:    s_sub_u32 s32, s32, 0x200
+; GFX10-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX10-NEXT:    s_or_saveexec_b32 s6, -1
+; GFX10-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s6
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    s_setpc_b64 s[4:5]
+  %ptr = load <32 x i32> addrspace(1)*, <32 x i32> addrspace(1)* addrspace(4)* undef
+  %val = load <32 x i32>, <32 x i32> addrspace(1)* %ptr
+  call amdgpu_gfx_callable void @external_void_func_v32i32(<32 x i32> %val)
+  ret void
+}
+
+define amdgpu_gfx_callable void @test_call_external_void_func_v32i32_i32(i32) #0 {
+; GFX9-LABEL: test_call_external_void_func_v32i32_i32:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    s_or_saveexec_b64 s[4:5], -1
+; GFX9-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX9-NEXT:    s_mov_b64 exec, s[4:5]
+; GFX9-NEXT:    s_load_dwordx2 s[4:5], s[4:5], 0x0
+; GFX9-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX9-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX9-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX9-NEXT:    global_load_dword v32, v[0:1], off
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    v_mov_b32_e32 v29, s5
+; GFX9-NEXT:    v_mov_b32_e32 v28, s4
+; GFX9-NEXT:    global_load_dwordx4 v[0:3], v[28:29], off
+; GFX9-NEXT:    global_load_dwordx4 v[4:7], v[28:29], off offset:16
+; GFX9-NEXT:    global_load_dwordx4 v[8:11], v[28:29], off offset:32
+; GFX9-NEXT:    global_load_dwordx4 v[12:15], v[28:29], off offset:48
+; GFX9-NEXT:    global_load_dwordx4 v[16:19], v[28:29], off offset:64
+; GFX9-NEXT:    global_load_dwordx4 v[20:23], v[28:29], off offset:80
+; GFX9-NEXT:    global_load_dwordx4 v[24:27], v[28:29], off offset:96
+; GFX9-NEXT:    global_load_dwordx4 v[28:31], v[28:29], off offset:112
+; GFX9-NEXT:    s_mov_b32 s33, s32
+; GFX9-NEXT:    s_add_u32 s32, s32, 0x400
+; GFX9-NEXT:    s_getpc_b64 s[6:7]
+; GFX9-NEXT:    s_add_u32 s6, s6, external_void_func_v32i32_i32@rel32@lo+4
+; GFX9-NEXT:    s_addc_u32 s7, s7, external_void_func_v32i32_i32@rel32@hi+12
+; GFX9-NEXT:    s_swappc_b64 s[30:31], s[6:7]
+; GFX9-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX9-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX9-NEXT:    s_sub_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX9-NEXT:    s_or_saveexec_b64 s[6:7], -1
+; GFX9-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX9-NEXT:    s_mov_b64 exec, s[6:7]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    s_setpc_b64 s[4:5]
+;
+; GFX10-LABEL: test_call_external_void_func_v32i32_i32:
+; GFX10:       ; %bb.0:
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    s_or_saveexec_b32 s4, -1
+; GFX10-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s4
+; GFX10-NEXT:    s_load_dwordx2 s[4:5], s[4:5], 0x0
+; GFX10-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX10-NEXT:    s_mov_b32 s33, s32
+; GFX10-NEXT:    s_add_u32 s32, s32, 0x200
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX10-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX10-NEXT:    v_mov_b32_e32 v29, s5
+; GFX10-NEXT:    v_mov_b32_e32 v28, s4
+; GFX10-NEXT:    s_getpc_b64 s[4:5]
+; GFX10-NEXT:    s_add_u32 s4, s4, external_void_func_v32i32_i32@rel32@lo+4
+; GFX10-NEXT:    s_addc_u32 s5, s5, external_void_func_v32i32_i32@rel32@hi+12
+; GFX10-NEXT:    s_clause 0x7
+; GFX10-NEXT:    global_load_dwordx4 v[0:3], v[28:29], off
+; GFX10-NEXT:    global_load_dwordx4 v[4:7], v[28:29], off offset:16
+; GFX10-NEXT:    global_load_dwordx4 v[8:11], v[28:29], off offset:32
+; GFX10-NEXT:    global_load_dwordx4 v[12:15], v[28:29], off offset:48
+; GFX10-NEXT:    global_load_dwordx4 v[16:19], v[28:29], off offset:64
+; GFX10-NEXT:    global_load_dwordx4 v[20:23], v[28:29], off offset:80
+; GFX10-NEXT:    global_load_dwordx4 v[24:27], v[28:29], off offset:96
+; GFX10-NEXT:    global_load_dwordx4 v[28:31], v[28:29], off offset:112
+; GFX10-NEXT:    s_waitcnt vmcnt(7)
+; GFX10-NEXT:    global_load_dword v32, v[0:1], off
+; GFX10-NEXT:    s_swappc_b64 s[30:31], s[4:5]
+; GFX10-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX10-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX10-NEXT:    s_sub_u32 s32, s32, 0x200
+; GFX10-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX10-NEXT:    s_or_saveexec_b32 s6, -1
+; GFX10-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s6
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    s_setpc_b64 s[4:5]
+  %ptr0 = load <32 x i32> addrspace(1)*, <32 x i32> addrspace(1)* addrspace(4)* undef
+  %val0 = load <32 x i32>, <32 x i32> addrspace(1)* %ptr0
+  %val1 = load i32, i32 addrspace(1)* undef
+  call amdgpu_gfx_callable void @external_void_func_v32i32_i32(<32 x i32> %val0, i32 %val1)
+  ret void
+}
+
+define amdgpu_gfx_callable void @test_call_external_i32_func_i32_imm(i32 addrspace(1)* %out) #0 {
+; GFX9-LABEL: test_call_external_i32_func_i32_imm:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    s_or_saveexec_b64 s[4:5], -1
+; GFX9-NEXT:    buffer_store_dword v42, off, s[0:3], s32 offset:8 ; 4-byte Folded Spill
+; GFX9-NEXT:    s_mov_b64 exec, s[4:5]
+; GFX9-NEXT:    v_writelane_b32 v42, s33, 2
+; GFX9-NEXT:    s_mov_b32 s33, s32
+; GFX9-NEXT:    s_nop 2
+; GFX9-NEXT:    buffer_store_dword v40, off, s[0:3], s33 offset:4 ; 4-byte Folded Spill
+; GFX9-NEXT:    buffer_store_dword v41, off, s[0:3], s33 ; 4-byte Folded Spill
+; GFX9-NEXT:    v_writelane_b32 v42, s30, 0
+; GFX9-NEXT:    s_add_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_mov_b32_e32 v40, v0
+; GFX9-NEXT:    v_mov_b32_e32 v0, 42
+; GFX9-NEXT:    s_getpc_b64 s[4:5]
+; GFX9-NEXT:    s_add_u32 s4, s4, external_i32_func_i32@rel32@lo+4
+; GFX9-NEXT:    s_addc_u32 s5, s5, external_i32_func_i32@rel32@hi+12
+; GFX9-NEXT:    v_writelane_b32 v42, s31, 1
+; GFX9-NEXT:    v_mov_b32_e32 v41, v1
+; GFX9-NEXT:    s_swappc_b64 s[30:31], s[4:5]
+; GFX9-NEXT:    global_store_dword v[40:41], v0, off
+; GFX9-NEXT:    v_readlane_b32 s4, v42, 0
+; GFX9-NEXT:    v_readlane_b32 s5, v42, 1
+; GFX9-NEXT:    buffer_load_dword v41, off, s[0:3], s33 ; 4-byte Folded Reload
+; GFX9-NEXT:    buffer_load_dword v40, off, s[0:3], s33 offset:4 ; 4-byte Folded Reload
+; GFX9-NEXT:    s_sub_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_readlane_b32 s33, v42, 2
+; GFX9-NEXT:    s_or_saveexec_b64 s[6:7], -1
+; GFX9-NEXT:    buffer_load_dword v42, off, s[0:3], s32 offset:8 ; 4-byte Folded Reload
+; GFX9-NEXT:    s_mov_b64 exec, s[6:7]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    s_setpc_b64 s[4:5]
+;
+; GFX10-LABEL: test_call_external_i32_func_i32_imm:
+; GFX10:       ; %bb.0:
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    s_or_saveexec_b32 s4, -1
+; GFX10-NEXT:    buffer_store_dword v42, off, s[0:3], s32 offset:8 ; 4-byte Folded Spill
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s4
+; GFX10-NEXT:    v_writelane_b32 v42, s33, 2
+; GFX10-NEXT:    s_mov_b32 s33, s32
+; GFX10-NEXT:    s_add_u32 s32, s32, 0x200
+; GFX10-NEXT:    s_nop 1
+; GFX10-NEXT:    buffer_store_dword v40, off, s[0:3], s33 offset:4 ; 4-byte Folded Spill
+; GFX10-NEXT:    buffer_store_dword v41, off, s[0:3], s33 ; 4-byte Folded Spill
+; GFX10-NEXT:    v_mov_b32_e32 v40, v0
+; GFX10-NEXT:    v_writelane_b32 v42, s30, 0
+; GFX10-NEXT:    v_mov_b32_e32 v0, 42
+; GFX10-NEXT:    s_getpc_b64 s[4:5]
+; GFX10-NEXT:    s_add_u32 s4, s4, external_i32_func_i32@rel32@lo+4
+; GFX10-NEXT:    s_addc_u32 s5, s5, external_i32_func_i32@rel32@hi+12
+; GFX10-NEXT:    v_mov_b32_e32 v41, v1
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    v_writelane_b32 v42, s31, 1
+; GFX10-NEXT:    s_swappc_b64 s[30:31], s[4:5]
+; GFX10-NEXT:    global_store_dword v[40:41], v0, off
+; GFX10-NEXT:    v_readlane_b32 s4, v42, 0
+; GFX10-NEXT:    v_readlane_b32 s5, v42, 1
+; GFX10-NEXT:    s_clause 0x1
+; GFX10-NEXT:    buffer_load_dword v41, off, s[0:3], s33
+; GFX10-NEXT:    buffer_load_dword v40, off, s[0:3], s33 offset:4
+; GFX10-NEXT:    s_sub_u32 s32, s32, 0x200
+; GFX10-NEXT:    v_readlane_b32 s33, v42, 2
+; GFX10-NEXT:    s_or_saveexec_b32 s6, -1
+; GFX10-NEXT:    buffer_load_dword v42, off, s[0:3], s32 offset:8 ; 4-byte Folded Reload
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s6
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    s_setpc_b64 s[4:5]
+  %val = call amdgpu_gfx_callable i32 @external_i32_func_i32(i32 42)
+  store volatile i32 %val, i32 addrspace(1)* %out
+  ret void
+}
+
+define amdgpu_gfx_callable void @test_call_external_void_func_struct_i8_i32() #0 {
+; GFX9-LABEL: test_call_external_void_func_struct_i8_i32:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    s_or_saveexec_b64 s[4:5], -1
+; GFX9-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX9-NEXT:    s_mov_b64 exec, s[4:5]
+; GFX9-NEXT:    s_load_dwordx2 s[4:5], s[4:5], 0x0
+; GFX9-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX9-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX9-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX9-NEXT:    s_mov_b32 s33, s32
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    v_mov_b32_e32 v2, s4
+; GFX9-NEXT:    v_mov_b32_e32 v3, s5
+; GFX9-NEXT:    global_load_dword v1, v[2:3], off offset:4
+; GFX9-NEXT:    global_load_ubyte v0, v[2:3], off
+; GFX9-NEXT:    s_add_u32 s32, s32, 0x400
+; GFX9-NEXT:    s_getpc_b64 s[4:5]
+; GFX9-NEXT:    s_add_u32 s4, s4, external_void_func_struct_i8_i32@rel32@lo+4
+; GFX9-NEXT:    s_addc_u32 s5, s5, external_void_func_struct_i8_i32@rel32@hi+12
+; GFX9-NEXT:    s_swappc_b64 s[30:31], s[4:5]
+; GFX9-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX9-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX9-NEXT:    s_sub_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX9-NEXT:    s_or_saveexec_b64 s[6:7], -1
+; GFX9-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX9-NEXT:    s_mov_b64 exec, s[6:7]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    s_setpc_b64 s[4:5]
+;
+; GFX10-LABEL: test_call_external_void_func_struct_i8_i32:
+; GFX10:       ; %bb.0:
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    s_or_saveexec_b32 s4, -1
+; GFX10-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s4
+; GFX10-NEXT:    s_load_dwordx2 s[4:5], s[4:5], 0x0
+; GFX10-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX10-NEXT:    s_mov_b32 s33, s32
+; GFX10-NEXT:    s_add_u32 s32, s32, 0x200
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX10-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX10-NEXT:    v_mov_b32_e32 v1, s4
+; GFX10-NEXT:    v_mov_b32_e32 v2, s5
+; GFX10-NEXT:    s_getpc_b64 s[4:5]
+; GFX10-NEXT:    s_add_u32 s4, s4, external_void_func_struct_i8_i32@rel32@lo+4
+; GFX10-NEXT:    s_addc_u32 s5, s5, external_void_func_struct_i8_i32@rel32@hi+12
+; GFX10-NEXT:    s_clause 0x1
+; GFX10-NEXT:    global_load_ubyte v0, v[1:2], off
+; GFX10-NEXT:    global_load_dword v1, v[1:2], off offset:4
+; GFX10-NEXT:    s_swappc_b64 s[30:31], s[4:5]
+; GFX10-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX10-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX10-NEXT:    s_sub_u32 s32, s32, 0x200
+; GFX10-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX10-NEXT:    s_or_saveexec_b32 s6, -1
+; GFX10-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s6
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    s_setpc_b64 s[4:5]
+  %ptr0 = load { i8, i32 } addrspace(1)*, { i8, i32 } addrspace(1)* addrspace(4)* undef
+  %val = load { i8, i32 }, { i8, i32 } addrspace(1)* %ptr0
+  call amdgpu_gfx_callable void @external_void_func_struct_i8_i32({ i8, i32 } %val)
+  ret void
+}
+
+define amdgpu_gfx_callable void @test_call_external_void_func_byval_struct_i8_i32() #0 {
+; GFX9-LABEL: test_call_external_void_func_byval_struct_i8_i32:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    s_or_saveexec_b64 s[4:5], -1
+; GFX9-NEXT:    buffer_store_dword v40, off, s[0:3], s32 offset:8 ; 4-byte Folded Spill
+; GFX9-NEXT:    s_mov_b64 exec, s[4:5]
+; GFX9-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX9-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX9-NEXT:    s_mov_b32 s33, s32
+; GFX9-NEXT:    v_mov_b32_e32 v0, 3
+; GFX9-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX9-NEXT:    buffer_store_byte v0, off, s[0:3], s33
+; GFX9-NEXT:    v_mov_b32_e32 v0, 8
+; GFX9-NEXT:    s_add_u32 s32, s32, 0x400
+; GFX9-NEXT:    buffer_store_dword v0, off, s[0:3], s33 offset:4
+; GFX9-NEXT:    v_lshrrev_b32_e64 v0, 6, s33
+; GFX9-NEXT:    s_getpc_b64 s[4:5]
+; GFX9-NEXT:    s_add_u32 s4, s4, external_void_func_byval_struct_i8_i32@rel32@lo+4
+; GFX9-NEXT:    s_addc_u32 s5, s5, external_void_func_byval_struct_i8_i32@rel32@hi+12
+; GFX9-NEXT:    s_swappc_b64 s[30:31], s[4:5]
+; GFX9-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX9-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX9-NEXT:    s_sub_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX9-NEXT:    s_or_saveexec_b64 s[6:7], -1
+; GFX9-NEXT:    buffer_load_dword v40, off, s[0:3], s32 offset:8 ; 4-byte Folded Reload
+; GFX9-NEXT:    s_mov_b64 exec, s[6:7]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    s_setpc_b64 s[4:5]
+;
+; GFX10-LABEL: test_call_external_void_func_byval_struct_i8_i32:
+; GFX10:       ; %bb.0:
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    s_or_saveexec_b32 s4, -1
+; GFX10-NEXT:    buffer_store_dword v40, off, s[0:3], s32 offset:8 ; 4-byte Folded Spill
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s4
+; GFX10-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX10-NEXT:    v_mov_b32_e32 v0, 3
+; GFX10-NEXT:    v_mov_b32_e32 v1, 8
+; GFX10-NEXT:    s_mov_b32 s33, s32
+; GFX10-NEXT:    s_add_u32 s32, s32, 0x200
+; GFX10-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    s_getpc_b64 s[4:5]
+; GFX10-NEXT:    s_add_u32 s4, s4, external_void_func_byval_struct_i8_i32@rel32@lo+4
+; GFX10-NEXT:    s_addc_u32 s5, s5, external_void_func_byval_struct_i8_i32@rel32@hi+12
+; GFX10-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX10-NEXT:    buffer_store_byte v0, off, s[0:3], s33
+; GFX10-NEXT:    buffer_store_dword v1, off, s[0:3], s33 offset:4
+; GFX10-NEXT:    v_lshrrev_b32_e64 v0, 5, s33
+; GFX10-NEXT:    s_swappc_b64 s[30:31], s[4:5]
+; GFX10-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX10-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX10-NEXT:    s_sub_u32 s32, s32, 0x200
+; GFX10-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX10-NEXT:    s_or_saveexec_b32 s6, -1
+; GFX10-NEXT:    buffer_load_dword v40, off, s[0:3], s32 offset:8 ; 4-byte Folded Reload
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s6
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    s_setpc_b64 s[4:5]
+  %val = alloca { i8, i32 }, align 4, addrspace(5)
+  %gep0 = getelementptr inbounds { i8, i32 }, { i8, i32 } addrspace(5)* %val, i32 0, i32 0
+  %gep1 = getelementptr inbounds { i8, i32 }, { i8, i32 } addrspace(5)* %val, i32 0, i32 1
+  store i8 3, i8 addrspace(5)* %gep0
+  store i32 8, i32 addrspace(5)* %gep1
+  call amdgpu_gfx_callable void @external_void_func_byval_struct_i8_i32({ i8, i32 } addrspace(5)* %val)
+  ret void
+}
+
+define amdgpu_gfx_callable void @test_call_external_void_func_sret_struct_i8_i32_byval_struct_i8_i32(i32) #0 {
+; GFX9-LABEL: test_call_external_void_func_sret_struct_i8_i32_byval_struct_i8_i32:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    s_or_saveexec_b64 s[4:5], -1
+; GFX9-NEXT:    buffer_store_dword v40, off, s[0:3], s32 offset:16 ; 4-byte Folded Spill
+; GFX9-NEXT:    s_mov_b64 exec, s[4:5]
+; GFX9-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX9-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX9-NEXT:    s_mov_b32 s33, s32
+; GFX9-NEXT:    v_mov_b32_e32 v0, 3
+; GFX9-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX9-NEXT:    buffer_store_byte v0, off, s[0:3], s33
+; GFX9-NEXT:    v_mov_b32_e32 v0, 8
+; GFX9-NEXT:    buffer_store_dword v0, off, s[0:3], s33 offset:4
+; GFX9-NEXT:    v_lshrrev_b32_e64 v0, 6, s33
+; GFX9-NEXT:    s_add_u32 s32, s32, 0x800
+; GFX9-NEXT:    v_add_u32_e32 v0, 8, v0
+; GFX9-NEXT:    v_lshrrev_b32_e64 v1, 6, s33
+; GFX9-NEXT:    s_getpc_b64 s[4:5]
+; GFX9-NEXT:    s_add_u32 s4, s4, external_void_func_sret_struct_i8_i32_byval_struct_i8_i32@rel32@lo+4
+; GFX9-NEXT:    s_addc_u32 s5, s5, external_void_func_sret_struct_i8_i32_byval_struct_i8_i32@rel32@hi+12
+; GFX9-NEXT:    s_swappc_b64 s[30:31], s[4:5]
+; GFX9-NEXT:    buffer_load_ubyte v0, off, s[0:3], s33 offset:8
+; GFX9-NEXT:    buffer_load_dword v1, off, s[0:3], s33 offset:12
+; GFX9-NEXT:    s_sub_u32 s32, s32, 0x800
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    global_store_byte v[0:1], v0, off
+; GFX9-NEXT:    global_store_dword v[0:1], v1, off
+; GFX9-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX9-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX9-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX9-NEXT:    s_or_saveexec_b64 s[6:7], -1
+; GFX9-NEXT:    buffer_load_dword v40, off, s[0:3], s32 offset:16 ; 4-byte Folded Reload
+; GFX9-NEXT:    s_mov_b64 exec, s[6:7]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    s_setpc_b64 s[4:5]
+;
+; GFX10-LABEL: test_call_external_void_func_sret_struct_i8_i32_byval_struct_i8_i32:
+; GFX10:       ; %bb.0:
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    s_or_saveexec_b32 s4, -1
+; GFX10-NEXT:    buffer_store_dword v40, off, s[0:3], s32 offset:16 ; 4-byte Folded Spill
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s4
+; GFX10-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX10-NEXT:    v_mov_b32_e32 v0, 3
+; GFX10-NEXT:    v_mov_b32_e32 v1, 8
+; GFX10-NEXT:    s_mov_b32 s33, s32
+; GFX10-NEXT:    s_add_u32 s32, s32, 0x400
+; GFX10-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    s_getpc_b64 s[4:5]
+; GFX10-NEXT:    s_add_u32 s4, s4, external_void_func_sret_struct_i8_i32_byval_struct_i8_i32@rel32@lo+4
+; GFX10-NEXT:    s_addc_u32 s5, s5, external_void_func_sret_struct_i8_i32_byval_struct_i8_i32@rel32@hi+12
+; GFX10-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX10-NEXT:    buffer_store_byte v0, off, s[0:3], s33
+; GFX10-NEXT:    buffer_store_dword v1, off, s[0:3], s33 offset:4
+; GFX10-NEXT:    v_lshrrev_b32_e64 v0, 5, s33
+; GFX10-NEXT:    v_lshrrev_b32_e64 v1, 5, s33
+; GFX10-NEXT:    v_add_nc_u32_e32 v0, 8, v0
+; GFX10-NEXT:    s_swappc_b64 s[30:31], s[4:5]
+; GFX10-NEXT:    s_clause 0x1
+; GFX10-NEXT:    buffer_load_ubyte v0, off, s[0:3], s33 offset:8
+; GFX10-NEXT:    buffer_load_dword v1, off, s[0:3], s33 offset:12
+; GFX10-NEXT:    s_sub_u32 s32, s32, 0x400
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    global_store_byte v[0:1], v0, off
+; GFX10-NEXT:    global_store_dword v[0:1], v1, off
+; GFX10-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX10-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX10-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX10-NEXT:    s_or_saveexec_b32 s6, -1
+; GFX10-NEXT:    buffer_load_dword v40, off, s[0:3], s32 offset:16 ; 4-byte Folded Reload
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s6
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    s_setpc_b64 s[4:5]
+  %in.val = alloca { i8, i32 }, align 4, addrspace(5)
+  %out.val = alloca { i8, i32 }, align 4, addrspace(5)
+  %in.gep0 = getelementptr inbounds { i8, i32 }, { i8, i32 } addrspace(5)* %in.val, i32 0, i32 0
+  %in.gep1 = getelementptr inbounds { i8, i32 }, { i8, i32 } addrspace(5)* %in.val, i32 0, i32 1
+  store i8 3, i8 addrspace(5)* %in.gep0
+  store i32 8, i32 addrspace(5)* %in.gep1
+  call amdgpu_gfx_callable void @external_void_func_sret_struct_i8_i32_byval_struct_i8_i32({ i8, i32 } addrspace(5)* %out.val, { i8, i32 } addrspace(5)* %in.val)
+  %out.gep0 = getelementptr inbounds { i8, i32 }, { i8, i32 } addrspace(5)* %out.val, i32 0, i32 0
+  %out.gep1 = getelementptr inbounds { i8, i32 }, { i8, i32 } addrspace(5)* %out.val, i32 0, i32 1
+  %out.val0 = load i8, i8 addrspace(5)* %out.gep0
+  %out.val1 = load i32, i32 addrspace(5)* %out.gep1
+
+  store volatile i8 %out.val0, i8 addrspace(1)* undef
+  store volatile i32 %out.val1, i32 addrspace(1)* undef
+  ret void
+}
+
+define amdgpu_gfx_callable void @test_call_external_void_func_v16i8() #0 {
+; GFX9-LABEL: test_call_external_void_func_v16i8:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    s_or_saveexec_b64 s[4:5], -1
+; GFX9-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX9-NEXT:    s_mov_b64 exec, s[4:5]
+; GFX9-NEXT:    s_load_dwordx2 s[4:5], s[4:5], 0x0
+; GFX9-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX9-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX9-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX9-NEXT:    s_mov_b32 s33, s32
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    v_mov_b32_e32 v0, s4
+; GFX9-NEXT:    v_mov_b32_e32 v1, s5
+; GFX9-NEXT:    global_load_dwordx4 v[0:3], v[0:1], off
+; GFX9-NEXT:    s_add_u32 s32, s32, 0x400
+; GFX9-NEXT:    s_getpc_b64 s[4:5]
+; GFX9-NEXT:    s_add_u32 s4, s4, external_void_func_v16i8@rel32@lo+4
+; GFX9-NEXT:    s_addc_u32 s5, s5, external_void_func_v16i8@rel32@hi+12
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    v_lshrrev_b32_e32 v16, 8, v0
+; GFX9-NEXT:    v_lshrrev_b32_e32 v17, 16, v0
+; GFX9-NEXT:    v_lshrrev_b32_e32 v18, 24, v0
+; GFX9-NEXT:    v_lshrrev_b32_e32 v5, 8, v1
+; GFX9-NEXT:    v_lshrrev_b32_e32 v6, 16, v1
+; GFX9-NEXT:    v_lshrrev_b32_e32 v7, 24, v1
+; GFX9-NEXT:    v_mov_b32_e32 v4, v1
+; GFX9-NEXT:    v_lshrrev_b32_e32 v9, 8, v2
+; GFX9-NEXT:    v_lshrrev_b32_e32 v10, 16, v2
+; GFX9-NEXT:    v_lshrrev_b32_e32 v11, 24, v2
+; GFX9-NEXT:    v_mov_b32_e32 v8, v2
+; GFX9-NEXT:    v_lshrrev_b32_e32 v13, 8, v3
+; GFX9-NEXT:    v_lshrrev_b32_e32 v14, 16, v3
+; GFX9-NEXT:    v_lshrrev_b32_e32 v15, 24, v3
+; GFX9-NEXT:    v_mov_b32_e32 v12, v3
+; GFX9-NEXT:    v_mov_b32_e32 v1, v16
+; GFX9-NEXT:    v_mov_b32_e32 v2, v17
+; GFX9-NEXT:    v_mov_b32_e32 v3, v18
+; GFX9-NEXT:    s_swappc_b64 s[30:31], s[4:5]
+; GFX9-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX9-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX9-NEXT:    s_sub_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX9-NEXT:    s_or_saveexec_b64 s[6:7], -1
+; GFX9-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX9-NEXT:    s_mov_b64 exec, s[6:7]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    s_setpc_b64 s[4:5]
+;
+; GFX10-LABEL: test_call_external_void_func_v16i8:
+; GFX10:       ; %bb.0:
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    s_or_saveexec_b32 s4, -1
+; GFX10-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s4
+; GFX10-NEXT:    s_load_dwordx2 s[4:5], s[4:5], 0x0
+; GFX10-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX10-NEXT:    s_mov_b32 s33, s32
+; GFX10-NEXT:    s_add_u32 s32, s32, 0x200
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX10-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX10-NEXT:    v_mov_b32_e32 v0, s4
+; GFX10-NEXT:    v_mov_b32_e32 v1, s5
+; GFX10-NEXT:    s_getpc_b64 s[4:5]
+; GFX10-NEXT:    s_add_u32 s4, s4, external_void_func_v16i8@rel32@lo+4
+; GFX10-NEXT:    s_addc_u32 s5, s5, external_void_func_v16i8@rel32@hi+12
+; GFX10-NEXT:    global_load_dwordx4 v[0:3], v[0:1], off
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    v_lshrrev_b32_e32 v16, 8, v0
+; GFX10-NEXT:    v_lshrrev_b32_e32 v17, 16, v0
+; GFX10-NEXT:    v_lshrrev_b32_e32 v18, 24, v0
+; GFX10-NEXT:    v_lshrrev_b32_e32 v5, 8, v1
+; GFX10-NEXT:    v_lshrrev_b32_e32 v6, 16, v1
+; GFX10-NEXT:    v_lshrrev_b32_e32 v7, 24, v1
+; GFX10-NEXT:    v_mov_b32_e32 v4, v1
+; GFX10-NEXT:    v_lshrrev_b32_e32 v9, 8, v2
+; GFX10-NEXT:    v_lshrrev_b32_e32 v10, 16, v2
+; GFX10-NEXT:    v_lshrrev_b32_e32 v11, 24, v2
+; GFX10-NEXT:    v_mov_b32_e32 v8, v2
+; GFX10-NEXT:    v_lshrrev_b32_e32 v13, 8, v3
+; GFX10-NEXT:    v_lshrrev_b32_e32 v14, 16, v3
+; GFX10-NEXT:    v_lshrrev_b32_e32 v15, 24, v3
+; GFX10-NEXT:    v_mov_b32_e32 v12, v3
+; GFX10-NEXT:    v_mov_b32_e32 v1, v16
+; GFX10-NEXT:    v_mov_b32_e32 v2, v17
+; GFX10-NEXT:    v_mov_b32_e32 v3, v18
+; GFX10-NEXT:    s_swappc_b64 s[30:31], s[4:5]
+; GFX10-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX10-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX10-NEXT:    s_sub_u32 s32, s32, 0x200
+; GFX10-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX10-NEXT:    s_or_saveexec_b32 s6, -1
+; GFX10-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s6
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    s_setpc_b64 s[4:5]
+  %ptr = load <16 x i8> addrspace(1)*, <16 x i8> addrspace(1)* addrspace(4)* undef
+  %val = load <16 x i8>, <16 x i8> addrspace(1)* %ptr
+  call amdgpu_gfx_callable void @external_void_func_v16i8(<16 x i8> %val)
+  ret void
+}
+
+define void @tail_call_byval_align16(<32 x i32> %val, double %tmp) #0 {
+; GFX9-LABEL: tail_call_byval_align16:
+; GFX9:       ; %bb.0: ; %entry
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    s_or_saveexec_b64 s[4:5], -1
+; GFX9-NEXT:    buffer_store_dword v40, off, s[0:3], s32 offset:16 ; 4-byte Folded Spill
+; GFX9-NEXT:    s_mov_b64 exec, s[4:5]
+; GFX9-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX9-NEXT:    s_mov_b32 s33, s32
+; GFX9-NEXT:    v_lshrrev_b32_e64 v32, 6, s33
+; GFX9-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX9-NEXT:    s_add_u32 s32, s32, 0x800
+; GFX9-NEXT:    v_add_u32_e32 v32, 8, v32
+; GFX9-NEXT:    s_getpc_b64 s[4:5]
+; GFX9-NEXT:    s_add_u32 s4, s4, byval_align16_f64_arg@rel32@lo+4
+; GFX9-NEXT:    s_addc_u32 s5, s5, byval_align16_f64_arg@rel32@hi+12
+; GFX9-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX9-NEXT:    s_swappc_b64 s[30:31], s[4:5]
+; GFX9-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX9-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX9-NEXT:    s_sub_u32 s32, s32, 0x800
+; GFX9-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX9-NEXT:    s_or_saveexec_b64 s[6:7], -1
+; GFX9-NEXT:    buffer_load_dword v40, off, s[0:3], s32 offset:16 ; 4-byte Folded Reload
+; GFX9-NEXT:    s_mov_b64 exec, s[6:7]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    s_setpc_b64 s[4:5]
+;
+; GFX10-LABEL: tail_call_byval_align16:
+; GFX10:       ; %bb.0: ; %entry
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    s_or_saveexec_b32 s4, -1
+; GFX10-NEXT:    buffer_store_dword v40, off, s[0:3], s32 offset:16 ; 4-byte Folded Spill
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s4
+; GFX10-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX10-NEXT:    s_mov_b32 s33, s32
+; GFX10-NEXT:    s_add_u32 s32, s32, 0x400
+; GFX10-NEXT:    v_lshrrev_b32_e64 v32, 5, s33
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX10-NEXT:    s_getpc_b64 s[4:5]
+; GFX10-NEXT:    s_add_u32 s4, s4, byval_align16_f64_arg@rel32@lo+4
+; GFX10-NEXT:    s_addc_u32 s5, s5, byval_align16_f64_arg@rel32@hi+12
+; GFX10-NEXT:    v_add_nc_u32_e32 v32, 8, v32
+; GFX10-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX10-NEXT:    s_swappc_b64 s[30:31], s[4:5]
+; GFX10-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX10-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX10-NEXT:    s_sub_u32 s32, s32, 0x400
+; GFX10-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX10-NEXT:    s_or_saveexec_b32 s6, -1
+; GFX10-NEXT:    buffer_load_dword v40, off, s[0:3], s32 offset:16 ; 4-byte Folded Reload
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s6
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    s_setpc_b64 s[4:5]
+entry:
+  %alloca = alloca double, align 8, addrspace(5)
+  tail call amdgpu_gfx_callable void @byval_align16_f64_arg(<32 x i32> %val, double addrspace(5)* byval align 16 %alloca)
+  ret void
+}
+
+; inreg arguments are put in sgprs
+define amdgpu_gfx_callable void @test_call_external_void_func_i1_imm_inreg() #0 {
+; GFX9-LABEL: test_call_external_void_func_i1_imm_inreg:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    s_or_saveexec_b64 s[4:5], -1
+; GFX9-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX9-NEXT:    s_mov_b64 exec, s[4:5]
+; GFX9-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX9-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX9-NEXT:    s_mov_b32 s33, s32
+; GFX9-NEXT:    s_add_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_mov_b32_e32 v0, 1
+; GFX9-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX9-NEXT:    s_getpc_b64 s[4:5]
+; GFX9-NEXT:    s_add_u32 s4, s4, external_void_func_i1_inreg@rel32@lo+4
+; GFX9-NEXT:    s_addc_u32 s5, s5, external_void_func_i1_inreg@rel32@hi+12
+; GFX9-NEXT:    buffer_store_byte v0, off, s[0:3], s32
+; GFX9-NEXT:    s_swappc_b64 s[30:31], s[4:5]
+; GFX9-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX9-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX9-NEXT:    s_sub_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX9-NEXT:    s_or_saveexec_b64 s[6:7], -1
+; GFX9-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX9-NEXT:    s_mov_b64 exec, s[6:7]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    s_setpc_b64 s[4:5]
+;
+; GFX10-LABEL: test_call_external_void_func_i1_imm_inreg:
+; GFX10:       ; %bb.0:
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    s_or_saveexec_b32 s4, -1
+; GFX10-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s4
+; GFX10-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX10-NEXT:    v_mov_b32_e32 v0, 1
+; GFX10-NEXT:    s_mov_b32 s33, s32
+; GFX10-NEXT:    s_add_u32 s32, s32, 0x200
+; GFX10-NEXT:    s_getpc_b64 s[4:5]
+; GFX10-NEXT:    s_add_u32 s4, s4, external_void_func_i1_inreg@rel32@lo+4
+; GFX10-NEXT:    s_addc_u32 s5, s5, external_void_func_i1_inreg@rel32@hi+12
+; GFX10-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX10-NEXT:    buffer_store_byte v0, off, s[0:3], s32
+; GFX10-NEXT:    s_swappc_b64 s[30:31], s[4:5]
+; GFX10-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX10-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX10-NEXT:    s_sub_u32 s32, s32, 0x200
+; GFX10-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX10-NEXT:    s_or_saveexec_b32 s6, -1
+; GFX10-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s6
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    s_setpc_b64 s[4:5]
+  call amdgpu_gfx_callable void @external_void_func_i1_inreg(i1 true)
+  ret void
+}
+
+define amdgpu_gfx_callable void @test_call_external_void_func_i8_imm_inreg(i32) #0 {
+; GFX9-LABEL: test_call_external_void_func_i8_imm_inreg:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    s_or_saveexec_b64 s[4:5], -1
+; GFX9-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX9-NEXT:    s_mov_b64 exec, s[4:5]
+; GFX9-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX9-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX9-NEXT:    s_mov_b32 s33, s32
+; GFX9-NEXT:    s_add_u32 s32, s32, 0x400
+; GFX9-NEXT:    s_movk_i32 s4, 0x7b
+; GFX9-NEXT:    s_getpc_b64 s[6:7]
+; GFX9-NEXT:    s_add_u32 s6, s6, external_void_func_i8_inreg@rel32@lo+4
+; GFX9-NEXT:    s_addc_u32 s7, s7, external_void_func_i8_inreg@rel32@hi+12
+; GFX9-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX9-NEXT:    s_swappc_b64 s[30:31], s[6:7]
+; GFX9-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX9-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX9-NEXT:    s_sub_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX9-NEXT:    s_or_saveexec_b64 s[6:7], -1
+; GFX9-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX9-NEXT:    s_mov_b64 exec, s[6:7]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    s_setpc_b64 s[4:5]
+;
+; GFX10-LABEL: test_call_external_void_func_i8_imm_inreg:
+; GFX10:       ; %bb.0:
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    s_or_saveexec_b32 s4, -1
+; GFX10-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s4
+; GFX10-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX10-NEXT:    s_movk_i32 s4, 0x7b
+; GFX10-NEXT:    s_mov_b32 s33, s32
+; GFX10-NEXT:    s_add_u32 s32, s32, 0x200
+; GFX10-NEXT:    s_getpc_b64 s[6:7]
+; GFX10-NEXT:    s_add_u32 s6, s6, external_void_func_i8_inreg@rel32@lo+4
+; GFX10-NEXT:    s_addc_u32 s7, s7, external_void_func_i8_inreg@rel32@hi+12
+; GFX10-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX10-NEXT:    s_swappc_b64 s[30:31], s[6:7]
+; GFX10-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX10-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX10-NEXT:    s_sub_u32 s32, s32, 0x200
+; GFX10-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX10-NEXT:    s_or_saveexec_b32 s6, -1
+; GFX10-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s6
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    s_setpc_b64 s[4:5]
+  call amdgpu_gfx_callable void @external_void_func_i8_inreg(i8 123)
+  ret void
+}
+
+define amdgpu_gfx_callable void @test_call_external_void_func_i16_imm_inreg() #0 {
+; GFX9-LABEL: test_call_external_void_func_i16_imm_inreg:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    s_or_saveexec_b64 s[4:5], -1
+; GFX9-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX9-NEXT:    s_mov_b64 exec, s[4:5]
+; GFX9-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX9-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX9-NEXT:    s_mov_b32 s33, s32
+; GFX9-NEXT:    s_add_u32 s32, s32, 0x400
+; GFX9-NEXT:    s_movk_i32 s4, 0x7b
+; GFX9-NEXT:    s_getpc_b64 s[6:7]
+; GFX9-NEXT:    s_add_u32 s6, s6, external_void_func_i16_inreg@rel32@lo+4
+; GFX9-NEXT:    s_addc_u32 s7, s7, external_void_func_i16_inreg@rel32@hi+12
+; GFX9-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX9-NEXT:    s_swappc_b64 s[30:31], s[6:7]
+; GFX9-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX9-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX9-NEXT:    s_sub_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX9-NEXT:    s_or_saveexec_b64 s[6:7], -1
+; GFX9-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX9-NEXT:    s_mov_b64 exec, s[6:7]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    s_setpc_b64 s[4:5]
+;
+; GFX10-LABEL: test_call_external_void_func_i16_imm_inreg:
+; GFX10:       ; %bb.0:
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    s_or_saveexec_b32 s4, -1
+; GFX10-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s4
+; GFX10-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX10-NEXT:    s_movk_i32 s4, 0x7b
+; GFX10-NEXT:    s_mov_b32 s33, s32
+; GFX10-NEXT:    s_add_u32 s32, s32, 0x200
+; GFX10-NEXT:    s_getpc_b64 s[6:7]
+; GFX10-NEXT:    s_add_u32 s6, s6, external_void_func_i16_inreg@rel32@lo+4
+; GFX10-NEXT:    s_addc_u32 s7, s7, external_void_func_i16_inreg@rel32@hi+12
+; GFX10-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX10-NEXT:    s_swappc_b64 s[30:31], s[6:7]
+; GFX10-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX10-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX10-NEXT:    s_sub_u32 s32, s32, 0x200
+; GFX10-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX10-NEXT:    s_or_saveexec_b32 s6, -1
+; GFX10-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s6
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    s_setpc_b64 s[4:5]
+  call amdgpu_gfx_callable void @external_void_func_i16_inreg(i16 123)
+  ret void
+}
+
+define amdgpu_gfx_callable void @test_call_external_void_func_i32_imm_inreg(i32) #0 {
+; GFX9-LABEL: test_call_external_void_func_i32_imm_inreg:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    s_or_saveexec_b64 s[4:5], -1
+; GFX9-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX9-NEXT:    s_mov_b64 exec, s[4:5]
+; GFX9-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX9-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX9-NEXT:    s_mov_b32 s33, s32
+; GFX9-NEXT:    s_add_u32 s32, s32, 0x400
+; GFX9-NEXT:    s_mov_b32 s4, 42
+; GFX9-NEXT:    s_getpc_b64 s[6:7]
+; GFX9-NEXT:    s_add_u32 s6, s6, external_void_func_i32_inreg@rel32@lo+4
+; GFX9-NEXT:    s_addc_u32 s7, s7, external_void_func_i32_inreg@rel32@hi+12
+; GFX9-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX9-NEXT:    s_swappc_b64 s[30:31], s[6:7]
+; GFX9-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX9-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX9-NEXT:    s_sub_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX9-NEXT:    s_or_saveexec_b64 s[6:7], -1
+; GFX9-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX9-NEXT:    s_mov_b64 exec, s[6:7]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    s_setpc_b64 s[4:5]
+;
+; GFX10-LABEL: test_call_external_void_func_i32_imm_inreg:
+; GFX10:       ; %bb.0:
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    s_or_saveexec_b32 s4, -1
+; GFX10-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s4
+; GFX10-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX10-NEXT:    s_mov_b32 s4, 42
+; GFX10-NEXT:    s_mov_b32 s33, s32
+; GFX10-NEXT:    s_add_u32 s32, s32, 0x200
+; GFX10-NEXT:    s_getpc_b64 s[6:7]
+; GFX10-NEXT:    s_add_u32 s6, s6, external_void_func_i32_inreg@rel32@lo+4
+; GFX10-NEXT:    s_addc_u32 s7, s7, external_void_func_i32_inreg@rel32@hi+12
+; GFX10-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX10-NEXT:    s_swappc_b64 s[30:31], s[6:7]
+; GFX10-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX10-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX10-NEXT:    s_sub_u32 s32, s32, 0x200
+; GFX10-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX10-NEXT:    s_or_saveexec_b32 s6, -1
+; GFX10-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s6
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    s_setpc_b64 s[4:5]
+  call amdgpu_gfx_callable void @external_void_func_i32_inreg(i32 42)
+  ret void
+}
+
+define amdgpu_gfx_callable void @test_call_external_void_func_i64_imm_inreg() #0 {
+; GFX9-LABEL: test_call_external_void_func_i64_imm_inreg:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    s_or_saveexec_b64 s[4:5], -1
+; GFX9-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX9-NEXT:    s_mov_b64 exec, s[4:5]
+; GFX9-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX9-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX9-NEXT:    s_mov_b32 s33, s32
+; GFX9-NEXT:    s_add_u32 s32, s32, 0x400
+; GFX9-NEXT:    s_movk_i32 s4, 0x7b
+; GFX9-NEXT:    s_mov_b32 s5, 0
+; GFX9-NEXT:    s_getpc_b64 s[6:7]
+; GFX9-NEXT:    s_add_u32 s6, s6, external_void_func_i64_inreg@rel32@lo+4
+; GFX9-NEXT:    s_addc_u32 s7, s7, external_void_func_i64_inreg@rel32@hi+12
+; GFX9-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX9-NEXT:    s_swappc_b64 s[30:31], s[6:7]
+; GFX9-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX9-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX9-NEXT:    s_sub_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX9-NEXT:    s_or_saveexec_b64 s[6:7], -1
+; GFX9-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX9-NEXT:    s_mov_b64 exec, s[6:7]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    s_setpc_b64 s[4:5]
+;
+; GFX10-LABEL: test_call_external_void_func_i64_imm_inreg:
+; GFX10:       ; %bb.0:
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    s_or_saveexec_b32 s4, -1
+; GFX10-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s4
+; GFX10-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX10-NEXT:    s_movk_i32 s4, 0x7b
+; GFX10-NEXT:    s_mov_b32 s5, 0
+; GFX10-NEXT:    s_mov_b32 s33, s32
+; GFX10-NEXT:    s_add_u32 s32, s32, 0x200
+; GFX10-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX10-NEXT:    s_getpc_b64 s[6:7]
+; GFX10-NEXT:    s_add_u32 s6, s6, external_void_func_i64_inreg@rel32@lo+4
+; GFX10-NEXT:    s_addc_u32 s7, s7, external_void_func_i64_inreg@rel32@hi+12
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX10-NEXT:    s_swappc_b64 s[30:31], s[6:7]
+; GFX10-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX10-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX10-NEXT:    s_sub_u32 s32, s32, 0x200
+; GFX10-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX10-NEXT:    s_or_saveexec_b32 s6, -1
+; GFX10-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s6
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    s_setpc_b64 s[4:5]
+  call amdgpu_gfx_callable void @external_void_func_i64_inreg(i64 123)
+  ret void
+}
+
+define amdgpu_gfx_callable void @test_call_external_void_func_v2i64_inreg() #0 {
+; GFX9-LABEL: test_call_external_void_func_v2i64_inreg:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    s_or_saveexec_b64 s[4:5], -1
+; GFX9-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX9-NEXT:    s_mov_b64 exec, s[4:5]
+; GFX9-NEXT:    s_mov_b64 s[4:5], 0
+; GFX9-NEXT:    s_load_dwordx4 s[4:7], s[4:5], 0x0
+; GFX9-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX9-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX9-NEXT:    s_mov_b32 s33, s32
+; GFX9-NEXT:    s_add_u32 s32, s32, 0x400
+; GFX9-NEXT:    s_getpc_b64 s[8:9]
+; GFX9-NEXT:    s_add_u32 s8, s8, external_void_func_v2i64_inreg@rel32@lo+4
+; GFX9-NEXT:    s_addc_u32 s9, s9, external_void_func_v2i64_inreg@rel32@hi+12
+; GFX9-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX9-NEXT:    s_swappc_b64 s[30:31], s[8:9]
+; GFX9-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX9-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX9-NEXT:    s_sub_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX9-NEXT:    s_or_saveexec_b64 s[6:7], -1
+; GFX9-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX9-NEXT:    s_mov_b64 exec, s[6:7]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    s_setpc_b64 s[4:5]
+;
+; GFX10-LABEL: test_call_external_void_func_v2i64_inreg:
+; GFX10:       ; %bb.0:
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    s_or_saveexec_b32 s4, -1
+; GFX10-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s4
+; GFX10-NEXT:    s_mov_b64 s[4:5], 0
+; GFX10-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX10-NEXT:    s_load_dwordx4 s[4:7], s[4:5], 0x0
+; GFX10-NEXT:    s_mov_b32 s33, s32
+; GFX10-NEXT:    s_add_u32 s32, s32, 0x200
+; GFX10-NEXT:    s_getpc_b64 s[8:9]
+; GFX10-NEXT:    s_add_u32 s8, s8, external_void_func_v2i64_inreg@rel32@lo+4
+; GFX10-NEXT:    s_addc_u32 s9, s9, external_void_func_v2i64_inreg@rel32@hi+12
+; GFX10-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX10-NEXT:    s_swappc_b64 s[30:31], s[8:9]
+; GFX10-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX10-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX10-NEXT:    s_sub_u32 s32, s32, 0x200
+; GFX10-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX10-NEXT:    s_or_saveexec_b32 s6, -1
+; GFX10-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s6
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    s_setpc_b64 s[4:5]
+  %val = load <2 x i64>, <2 x i64> addrspace(4)* null
+  call amdgpu_gfx_callable void @external_void_func_v2i64_inreg(<2 x i64> %val)
+  ret void
+}
+
+define amdgpu_gfx_callable void @test_call_external_void_func_v2i64_imm_inreg() #0 {
+; GFX9-LABEL: test_call_external_void_func_v2i64_imm_inreg:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    s_or_saveexec_b64 s[4:5], -1
+; GFX9-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX9-NEXT:    s_mov_b64 exec, s[4:5]
+; GFX9-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX9-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX9-NEXT:    s_mov_b32 s33, s32
+; GFX9-NEXT:    s_add_u32 s32, s32, 0x400
+; GFX9-NEXT:    s_mov_b32 s4, 1
+; GFX9-NEXT:    s_mov_b32 s5, 2
+; GFX9-NEXT:    s_mov_b32 s6, 3
+; GFX9-NEXT:    s_mov_b32 s7, 4
+; GFX9-NEXT:    s_getpc_b64 s[8:9]
+; GFX9-NEXT:    s_add_u32 s8, s8, external_void_func_v2i64_inreg@rel32@lo+4
+; GFX9-NEXT:    s_addc_u32 s9, s9, external_void_func_v2i64_inreg@rel32@hi+12
+; GFX9-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX9-NEXT:    s_swappc_b64 s[30:31], s[8:9]
+; GFX9-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX9-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX9-NEXT:    s_sub_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX9-NEXT:    s_or_saveexec_b64 s[6:7], -1
+; GFX9-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX9-NEXT:    s_mov_b64 exec, s[6:7]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    s_setpc_b64 s[4:5]
+;
+; GFX10-LABEL: test_call_external_void_func_v2i64_imm_inreg:
+; GFX10:       ; %bb.0:
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    s_or_saveexec_b32 s4, -1
+; GFX10-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s4
+; GFX10-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX10-NEXT:    s_mov_b32 s4, 1
+; GFX10-NEXT:    s_mov_b32 s5, 2
+; GFX10-NEXT:    s_mov_b32 s6, 3
+; GFX10-NEXT:    s_mov_b32 s7, 4
+; GFX10-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX10-NEXT:    s_mov_b32 s33, s32
+; GFX10-NEXT:    s_add_u32 s32, s32, 0x200
+; GFX10-NEXT:    s_getpc_b64 s[8:9]
+; GFX10-NEXT:    s_add_u32 s8, s8, external_void_func_v2i64_inreg@rel32@lo+4
+; GFX10-NEXT:    s_addc_u32 s9, s9, external_void_func_v2i64_inreg@rel32@hi+12
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX10-NEXT:    s_swappc_b64 s[30:31], s[8:9]
+; GFX10-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX10-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX10-NEXT:    s_sub_u32 s32, s32, 0x200
+; GFX10-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX10-NEXT:    s_or_saveexec_b32 s6, -1
+; GFX10-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s6
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    s_setpc_b64 s[4:5]
+  call amdgpu_gfx_callable void @external_void_func_v2i64_inreg(<2 x i64> <i64 8589934593, i64 17179869187>)
+  ret void
+}
+
+define amdgpu_gfx_callable void @test_call_external_void_func_v3i64_inreg() #0 {
+; GFX9-LABEL: test_call_external_void_func_v3i64_inreg:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    s_or_saveexec_b64 s[4:5], -1
+; GFX9-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX9-NEXT:    s_mov_b64 exec, s[4:5]
+; GFX9-NEXT:    s_mov_b64 s[4:5], 0
+; GFX9-NEXT:    s_load_dwordx4 s[4:7], s[4:5], 0x0
+; GFX9-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX9-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX9-NEXT:    s_mov_b32 s33, s32
+; GFX9-NEXT:    s_add_u32 s32, s32, 0x400
+; GFX9-NEXT:    s_mov_b32 s8, 1
+; GFX9-NEXT:    s_mov_b32 s9, 2
+; GFX9-NEXT:    s_getpc_b64 s[10:11]
+; GFX9-NEXT:    s_add_u32 s10, s10, external_void_func_v3i64_inreg@rel32@lo+4
+; GFX9-NEXT:    s_addc_u32 s11, s11, external_void_func_v3i64_inreg@rel32@hi+12
+; GFX9-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX9-NEXT:    s_swappc_b64 s[30:31], s[10:11]
+; GFX9-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX9-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX9-NEXT:    s_sub_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX9-NEXT:    s_or_saveexec_b64 s[6:7], -1
+; GFX9-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX9-NEXT:    s_mov_b64 exec, s[6:7]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    s_setpc_b64 s[4:5]
+;
+; GFX10-LABEL: test_call_external_void_func_v3i64_inreg:
+; GFX10:       ; %bb.0:
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    s_or_saveexec_b32 s4, -1
+; GFX10-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s4
+; GFX10-NEXT:    s_mov_b64 s[4:5], 0
+; GFX10-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX10-NEXT:    s_load_dwordx4 s[4:7], s[4:5], 0x0
+; GFX10-NEXT:    s_mov_b32 s8, 1
+; GFX10-NEXT:    s_mov_b32 s9, 2
+; GFX10-NEXT:    s_mov_b32 s33, s32
+; GFX10-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX10-NEXT:    s_add_u32 s32, s32, 0x200
+; GFX10-NEXT:    s_getpc_b64 s[10:11]
+; GFX10-NEXT:    s_add_u32 s10, s10, external_void_func_v3i64_inreg@rel32@lo+4
+; GFX10-NEXT:    s_addc_u32 s11, s11, external_void_func_v3i64_inreg@rel32@hi+12
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX10-NEXT:    s_swappc_b64 s[30:31], s[10:11]
+; GFX10-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX10-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX10-NEXT:    s_sub_u32 s32, s32, 0x200
+; GFX10-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX10-NEXT:    s_or_saveexec_b32 s6, -1
+; GFX10-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s6
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    s_setpc_b64 s[4:5]
+  %load = load <2 x i64>, <2 x i64> addrspace(4)* null
+  %val = shufflevector <2 x i64> %load, <2 x i64> <i64 8589934593, i64 undef>, <3 x i32> <i32 0, i32 1, i32 2>
+
+  call amdgpu_gfx_callable void @external_void_func_v3i64_inreg(<3 x i64> %val)
+  ret void
+}
+
+define amdgpu_gfx_callable void @test_call_external_void_func_v4i64_inreg() #0 {
+; GFX9-LABEL: test_call_external_void_func_v4i64_inreg:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    s_or_saveexec_b64 s[4:5], -1
+; GFX9-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX9-NEXT:    s_mov_b64 exec, s[4:5]
+; GFX9-NEXT:    s_mov_b64 s[4:5], 0
+; GFX9-NEXT:    s_load_dwordx4 s[4:7], s[4:5], 0x0
+; GFX9-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX9-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX9-NEXT:    s_mov_b32 s33, s32
+; GFX9-NEXT:    s_add_u32 s32, s32, 0x400
+; GFX9-NEXT:    s_mov_b32 s8, 1
+; GFX9-NEXT:    s_mov_b32 s9, 2
+; GFX9-NEXT:    s_mov_b32 s10, 3
+; GFX9-NEXT:    s_mov_b32 s11, 4
+; GFX9-NEXT:    s_getpc_b64 s[12:13]
+; GFX9-NEXT:    s_add_u32 s12, s12, external_void_func_v4i64_inreg@rel32@lo+4
+; GFX9-NEXT:    s_addc_u32 s13, s13, external_void_func_v4i64_inreg@rel32@hi+12
+; GFX9-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX9-NEXT:    s_swappc_b64 s[30:31], s[12:13]
+; GFX9-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX9-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX9-NEXT:    s_sub_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX9-NEXT:    s_or_saveexec_b64 s[6:7], -1
+; GFX9-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX9-NEXT:    s_mov_b64 exec, s[6:7]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    s_setpc_b64 s[4:5]
+;
+; GFX10-LABEL: test_call_external_void_func_v4i64_inreg:
+; GFX10:       ; %bb.0:
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    s_or_saveexec_b32 s4, -1
+; GFX10-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s4
+; GFX10-NEXT:    s_mov_b64 s[4:5], 0
+; GFX10-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX10-NEXT:    s_load_dwordx4 s[4:7], s[4:5], 0x0
+; GFX10-NEXT:    s_mov_b32 s8, 1
+; GFX10-NEXT:    s_mov_b32 s9, 2
+; GFX10-NEXT:    s_mov_b32 s10, 3
+; GFX10-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX10-NEXT:    s_mov_b32 s11, 4
+; GFX10-NEXT:    s_mov_b32 s33, s32
+; GFX10-NEXT:    s_add_u32 s32, s32, 0x200
+; GFX10-NEXT:    s_getpc_b64 s[12:13]
+; GFX10-NEXT:    s_add_u32 s12, s12, external_void_func_v4i64_inreg@rel32@lo+4
+; GFX10-NEXT:    s_addc_u32 s13, s13, external_void_func_v4i64_inreg@rel32@hi+12
+; GFX10-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    s_swappc_b64 s[30:31], s[12:13]
+; GFX10-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX10-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX10-NEXT:    s_sub_u32 s32, s32, 0x200
+; GFX10-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX10-NEXT:    s_or_saveexec_b32 s6, -1
+; GFX10-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s6
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    s_setpc_b64 s[4:5]
+  %load = load <2 x i64>, <2 x i64> addrspace(4)* null
+  %val = shufflevector <2 x i64> %load, <2 x i64> <i64 8589934593, i64 17179869187>, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  call amdgpu_gfx_callable void @external_void_func_v4i64_inreg(<4 x i64> %val)
+  ret void
+}
+
+define amdgpu_gfx_callable void @test_call_external_void_func_f16_imm_inreg() #0 {
+; GFX9-LABEL: test_call_external_void_func_f16_imm_inreg:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    s_or_saveexec_b64 s[4:5], -1
+; GFX9-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX9-NEXT:    s_mov_b64 exec, s[4:5]
+; GFX9-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX9-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX9-NEXT:    s_mov_b32 s33, s32
+; GFX9-NEXT:    s_add_u32 s32, s32, 0x400
+; GFX9-NEXT:    s_movk_i32 s4, 0x4400
+; GFX9-NEXT:    s_getpc_b64 s[6:7]
+; GFX9-NEXT:    s_add_u32 s6, s6, external_void_func_f16_inreg@rel32@lo+4
+; GFX9-NEXT:    s_addc_u32 s7, s7, external_void_func_f16_inreg@rel32@hi+12
+; GFX9-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX9-NEXT:    s_swappc_b64 s[30:31], s[6:7]
+; GFX9-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX9-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX9-NEXT:    s_sub_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX9-NEXT:    s_or_saveexec_b64 s[6:7], -1
+; GFX9-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX9-NEXT:    s_mov_b64 exec, s[6:7]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    s_setpc_b64 s[4:5]
+;
+; GFX10-LABEL: test_call_external_void_func_f16_imm_inreg:
+; GFX10:       ; %bb.0:
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    s_or_saveexec_b32 s4, -1
+; GFX10-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s4
+; GFX10-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX10-NEXT:    s_movk_i32 s4, 0x4400
+; GFX10-NEXT:    s_mov_b32 s33, s32
+; GFX10-NEXT:    s_add_u32 s32, s32, 0x200
+; GFX10-NEXT:    s_getpc_b64 s[6:7]
+; GFX10-NEXT:    s_add_u32 s6, s6, external_void_func_f16_inreg@rel32@lo+4
+; GFX10-NEXT:    s_addc_u32 s7, s7, external_void_func_f16_inreg@rel32@hi+12
+; GFX10-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX10-NEXT:    s_swappc_b64 s[30:31], s[6:7]
+; GFX10-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX10-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX10-NEXT:    s_sub_u32 s32, s32, 0x200
+; GFX10-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX10-NEXT:    s_or_saveexec_b32 s6, -1
+; GFX10-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s6
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    s_setpc_b64 s[4:5]
+  call amdgpu_gfx_callable void @external_void_func_f16_inreg(half 4.0)
+  ret void
+}
+
+define amdgpu_gfx_callable void @test_call_external_void_func_f32_imm_inreg() #0 {
+; GFX9-LABEL: test_call_external_void_func_f32_imm_inreg:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    s_or_saveexec_b64 s[4:5], -1
+; GFX9-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX9-NEXT:    s_mov_b64 exec, s[4:5]
+; GFX9-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX9-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX9-NEXT:    s_mov_b32 s33, s32
+; GFX9-NEXT:    s_add_u32 s32, s32, 0x400
+; GFX9-NEXT:    s_mov_b32 s4, 4.0
+; GFX9-NEXT:    s_getpc_b64 s[6:7]
+; GFX9-NEXT:    s_add_u32 s6, s6, external_void_func_f32_inreg@rel32@lo+4
+; GFX9-NEXT:    s_addc_u32 s7, s7, external_void_func_f32_inreg@rel32@hi+12
+; GFX9-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX9-NEXT:    s_swappc_b64 s[30:31], s[6:7]
+; GFX9-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX9-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX9-NEXT:    s_sub_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX9-NEXT:    s_or_saveexec_b64 s[6:7], -1
+; GFX9-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX9-NEXT:    s_mov_b64 exec, s[6:7]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    s_setpc_b64 s[4:5]
+;
+; GFX10-LABEL: test_call_external_void_func_f32_imm_inreg:
+; GFX10:       ; %bb.0:
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    s_or_saveexec_b32 s4, -1
+; GFX10-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s4
+; GFX10-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX10-NEXT:    s_mov_b32 s4, 4.0
+; GFX10-NEXT:    s_mov_b32 s33, s32
+; GFX10-NEXT:    s_add_u32 s32, s32, 0x200
+; GFX10-NEXT:    s_getpc_b64 s[6:7]
+; GFX10-NEXT:    s_add_u32 s6, s6, external_void_func_f32_inreg@rel32@lo+4
+; GFX10-NEXT:    s_addc_u32 s7, s7, external_void_func_f32_inreg@rel32@hi+12
+; GFX10-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX10-NEXT:    s_swappc_b64 s[30:31], s[6:7]
+; GFX10-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX10-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX10-NEXT:    s_sub_u32 s32, s32, 0x200
+; GFX10-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX10-NEXT:    s_or_saveexec_b32 s6, -1
+; GFX10-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s6
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    s_setpc_b64 s[4:5]
+  call amdgpu_gfx_callable void @external_void_func_f32_inreg(float 4.0)
+  ret void
+}
+
+define amdgpu_gfx_callable void @test_call_external_void_func_v2f32_imm_inreg() #0 {
+; GFX9-LABEL: test_call_external_void_func_v2f32_imm_inreg:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    s_or_saveexec_b64 s[4:5], -1
+; GFX9-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX9-NEXT:    s_mov_b64 exec, s[4:5]
+; GFX9-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX9-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX9-NEXT:    s_mov_b32 s33, s32
+; GFX9-NEXT:    s_add_u32 s32, s32, 0x400
+; GFX9-NEXT:    s_mov_b32 s4, 1.0
+; GFX9-NEXT:    s_mov_b32 s5, 2.0
+; GFX9-NEXT:    s_getpc_b64 s[6:7]
+; GFX9-NEXT:    s_add_u32 s6, s6, external_void_func_v2f32_inreg@rel32@lo+4
+; GFX9-NEXT:    s_addc_u32 s7, s7, external_void_func_v2f32_inreg@rel32@hi+12
+; GFX9-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX9-NEXT:    s_swappc_b64 s[30:31], s[6:7]
+; GFX9-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX9-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX9-NEXT:    s_sub_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX9-NEXT:    s_or_saveexec_b64 s[6:7], -1
+; GFX9-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX9-NEXT:    s_mov_b64 exec, s[6:7]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    s_setpc_b64 s[4:5]
+;
+; GFX10-LABEL: test_call_external_void_func_v2f32_imm_inreg:
+; GFX10:       ; %bb.0:
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    s_or_saveexec_b32 s4, -1
+; GFX10-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s4
+; GFX10-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX10-NEXT:    s_mov_b32 s4, 1.0
+; GFX10-NEXT:    s_mov_b32 s5, 2.0
+; GFX10-NEXT:    s_mov_b32 s33, s32
+; GFX10-NEXT:    s_add_u32 s32, s32, 0x200
+; GFX10-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX10-NEXT:    s_getpc_b64 s[6:7]
+; GFX10-NEXT:    s_add_u32 s6, s6, external_void_func_v2f32_inreg@rel32@lo+4
+; GFX10-NEXT:    s_addc_u32 s7, s7, external_void_func_v2f32_inreg@rel32@hi+12
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX10-NEXT:    s_swappc_b64 s[30:31], s[6:7]
+; GFX10-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX10-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX10-NEXT:    s_sub_u32 s32, s32, 0x200
+; GFX10-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX10-NEXT:    s_or_saveexec_b32 s6, -1
+; GFX10-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s6
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    s_setpc_b64 s[4:5]
+  call amdgpu_gfx_callable void @external_void_func_v2f32_inreg(<2 x float> <float 1.0, float 2.0>)
+  ret void
+}
+
+define amdgpu_gfx_callable void @test_call_external_void_func_v3f32_imm_inreg() #0 {
+; GFX9-LABEL: test_call_external_void_func_v3f32_imm_inreg:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    s_or_saveexec_b64 s[4:5], -1
+; GFX9-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX9-NEXT:    s_mov_b64 exec, s[4:5]
+; GFX9-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX9-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX9-NEXT:    s_mov_b32 s33, s32
+; GFX9-NEXT:    s_add_u32 s32, s32, 0x400
+; GFX9-NEXT:    s_mov_b32 s4, 1.0
+; GFX9-NEXT:    s_mov_b32 s5, 2.0
+; GFX9-NEXT:    s_mov_b32 s6, 4.0
+; GFX9-NEXT:    s_getpc_b64 s[8:9]
+; GFX9-NEXT:    s_add_u32 s8, s8, external_void_func_v3f32_inreg@rel32@lo+4
+; GFX9-NEXT:    s_addc_u32 s9, s9, external_void_func_v3f32_inreg@rel32@hi+12
+; GFX9-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX9-NEXT:    s_swappc_b64 s[30:31], s[8:9]
+; GFX9-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX9-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX9-NEXT:    s_sub_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX9-NEXT:    s_or_saveexec_b64 s[6:7], -1
+; GFX9-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX9-NEXT:    s_mov_b64 exec, s[6:7]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    s_setpc_b64 s[4:5]
+;
+; GFX10-LABEL: test_call_external_void_func_v3f32_imm_inreg:
+; GFX10:       ; %bb.0:
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    s_or_saveexec_b32 s4, -1
+; GFX10-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s4
+; GFX10-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX10-NEXT:    s_mov_b32 s4, 1.0
+; GFX10-NEXT:    s_mov_b32 s5, 2.0
+; GFX10-NEXT:    s_mov_b32 s6, 4.0
+; GFX10-NEXT:    s_mov_b32 s33, s32
+; GFX10-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX10-NEXT:    s_add_u32 s32, s32, 0x200
+; GFX10-NEXT:    s_getpc_b64 s[8:9]
+; GFX10-NEXT:    s_add_u32 s8, s8, external_void_func_v3f32_inreg@rel32@lo+4
+; GFX10-NEXT:    s_addc_u32 s9, s9, external_void_func_v3f32_inreg@rel32@hi+12
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX10-NEXT:    s_swappc_b64 s[30:31], s[8:9]
+; GFX10-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX10-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX10-NEXT:    s_sub_u32 s32, s32, 0x200
+; GFX10-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX10-NEXT:    s_or_saveexec_b32 s6, -1
+; GFX10-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s6
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    s_setpc_b64 s[4:5]
+  call amdgpu_gfx_callable void @external_void_func_v3f32_inreg(<3 x float> <float 1.0, float 2.0, float 4.0>)
+  ret void
+}
+
+define amdgpu_gfx_callable void @test_call_external_void_func_v5f32_imm_inreg() #0 {
+; GFX9-LABEL: test_call_external_void_func_v5f32_imm_inreg:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    s_or_saveexec_b64 s[4:5], -1
+; GFX9-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX9-NEXT:    s_mov_b64 exec, s[4:5]
+; GFX9-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX9-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX9-NEXT:    s_mov_b32 s33, s32
+; GFX9-NEXT:    s_add_u32 s32, s32, 0x400
+; GFX9-NEXT:    s_mov_b32 s4, 1.0
+; GFX9-NEXT:    s_mov_b32 s5, 2.0
+; GFX9-NEXT:    s_mov_b32 s6, 4.0
+; GFX9-NEXT:    s_mov_b32 s7, -1.0
+; GFX9-NEXT:    s_mov_b32 s8, 0.5
+; GFX9-NEXT:    s_getpc_b64 s[10:11]
+; GFX9-NEXT:    s_add_u32 s10, s10, external_void_func_v5f32_inreg@rel32@lo+4
+; GFX9-NEXT:    s_addc_u32 s11, s11, external_void_func_v5f32_inreg@rel32@hi+12
+; GFX9-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX9-NEXT:    s_swappc_b64 s[30:31], s[10:11]
+; GFX9-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX9-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX9-NEXT:    s_sub_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX9-NEXT:    s_or_saveexec_b64 s[6:7], -1
+; GFX9-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX9-NEXT:    s_mov_b64 exec, s[6:7]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    s_setpc_b64 s[4:5]
+;
+; GFX10-LABEL: test_call_external_void_func_v5f32_imm_inreg:
+; GFX10:       ; %bb.0:
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    s_or_saveexec_b32 s4, -1
+; GFX10-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s4
+; GFX10-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX10-NEXT:    s_mov_b32 s4, 1.0
+; GFX10-NEXT:    s_mov_b32 s5, 2.0
+; GFX10-NEXT:    s_mov_b32 s6, 4.0
+; GFX10-NEXT:    s_mov_b32 s7, -1.0
+; GFX10-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX10-NEXT:    s_mov_b32 s8, 0.5
+; GFX10-NEXT:    s_mov_b32 s33, s32
+; GFX10-NEXT:    s_add_u32 s32, s32, 0x200
+; GFX10-NEXT:    s_getpc_b64 s[10:11]
+; GFX10-NEXT:    s_add_u32 s10, s10, external_void_func_v5f32_inreg@rel32@lo+4
+; GFX10-NEXT:    s_addc_u32 s11, s11, external_void_func_v5f32_inreg@rel32@hi+12
+; GFX10-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    s_swappc_b64 s[30:31], s[10:11]
+; GFX10-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX10-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX10-NEXT:    s_sub_u32 s32, s32, 0x200
+; GFX10-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX10-NEXT:    s_or_saveexec_b32 s6, -1
+; GFX10-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s6
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    s_setpc_b64 s[4:5]
+  call amdgpu_gfx_callable void @external_void_func_v5f32_inreg(<5 x float> <float 1.0, float 2.0, float 4.0, float -1.0, float 0.5>)
+  ret void
+}
+
+define amdgpu_gfx_callable void @test_call_external_void_func_f64_imm_inreg() #0 {
+; GFX9-LABEL: test_call_external_void_func_f64_imm_inreg:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    s_or_saveexec_b64 s[4:5], -1
+; GFX9-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX9-NEXT:    s_mov_b64 exec, s[4:5]
+; GFX9-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX9-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX9-NEXT:    s_mov_b32 s33, s32
+; GFX9-NEXT:    s_add_u32 s32, s32, 0x400
+; GFX9-NEXT:    s_mov_b32 s4, 0
+; GFX9-NEXT:    s_mov_b32 s5, 0x40100000
+; GFX9-NEXT:    s_getpc_b64 s[6:7]
+; GFX9-NEXT:    s_add_u32 s6, s6, external_void_func_f64_inreg@rel32@lo+4
+; GFX9-NEXT:    s_addc_u32 s7, s7, external_void_func_f64_inreg@rel32@hi+12
+; GFX9-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX9-NEXT:    s_swappc_b64 s[30:31], s[6:7]
+; GFX9-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX9-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX9-NEXT:    s_sub_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX9-NEXT:    s_or_saveexec_b64 s[6:7], -1
+; GFX9-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX9-NEXT:    s_mov_b64 exec, s[6:7]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    s_setpc_b64 s[4:5]
+;
+; GFX10-LABEL: test_call_external_void_func_f64_imm_inreg:
+; GFX10:       ; %bb.0:
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    s_or_saveexec_b32 s4, -1
+; GFX10-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s4
+; GFX10-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX10-NEXT:    s_mov_b32 s4, 0
+; GFX10-NEXT:    s_mov_b32 s5, 0x40100000
+; GFX10-NEXT:    s_mov_b32 s33, s32
+; GFX10-NEXT:    s_add_u32 s32, s32, 0x200
+; GFX10-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX10-NEXT:    s_getpc_b64 s[6:7]
+; GFX10-NEXT:    s_add_u32 s6, s6, external_void_func_f64_inreg@rel32@lo+4
+; GFX10-NEXT:    s_addc_u32 s7, s7, external_void_func_f64_inreg@rel32@hi+12
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX10-NEXT:    s_swappc_b64 s[30:31], s[6:7]
+; GFX10-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX10-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX10-NEXT:    s_sub_u32 s32, s32, 0x200
+; GFX10-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX10-NEXT:    s_or_saveexec_b32 s6, -1
+; GFX10-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s6
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    s_setpc_b64 s[4:5]
+  call amdgpu_gfx_callable void @external_void_func_f64_inreg(double 4.0)
+  ret void
+}
+
+define amdgpu_gfx_callable void @test_call_external_void_func_v2f64_imm_inreg() #0 {
+; GFX9-LABEL: test_call_external_void_func_v2f64_imm_inreg:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    s_or_saveexec_b64 s[4:5], -1
+; GFX9-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX9-NEXT:    s_mov_b64 exec, s[4:5]
+; GFX9-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX9-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX9-NEXT:    s_mov_b32 s33, s32
+; GFX9-NEXT:    s_add_u32 s32, s32, 0x400
+; GFX9-NEXT:    s_mov_b32 s4, 0
+; GFX9-NEXT:    s_mov_b32 s5, 2.0
+; GFX9-NEXT:    s_mov_b32 s6, 0
+; GFX9-NEXT:    s_mov_b32 s7, 0x40100000
+; GFX9-NEXT:    s_getpc_b64 s[8:9]
+; GFX9-NEXT:    s_add_u32 s8, s8, external_void_func_v2f64_inreg@rel32@lo+4
+; GFX9-NEXT:    s_addc_u32 s9, s9, external_void_func_v2f64_inreg@rel32@hi+12
+; GFX9-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX9-NEXT:    s_swappc_b64 s[30:31], s[8:9]
+; GFX9-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX9-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX9-NEXT:    s_sub_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX9-NEXT:    s_or_saveexec_b64 s[6:7], -1
+; GFX9-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX9-NEXT:    s_mov_b64 exec, s[6:7]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    s_setpc_b64 s[4:5]
+;
+; GFX10-LABEL: test_call_external_void_func_v2f64_imm_inreg:
+; GFX10:       ; %bb.0:
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    s_or_saveexec_b32 s4, -1
+; GFX10-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s4
+; GFX10-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX10-NEXT:    s_mov_b32 s4, 0
+; GFX10-NEXT:    s_mov_b32 s5, 2.0
+; GFX10-NEXT:    s_mov_b32 s6, 0
+; GFX10-NEXT:    s_mov_b32 s7, 0x40100000
+; GFX10-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX10-NEXT:    s_mov_b32 s33, s32
+; GFX10-NEXT:    s_add_u32 s32, s32, 0x200
+; GFX10-NEXT:    s_getpc_b64 s[8:9]
+; GFX10-NEXT:    s_add_u32 s8, s8, external_void_func_v2f64_inreg@rel32@lo+4
+; GFX10-NEXT:    s_addc_u32 s9, s9, external_void_func_v2f64_inreg@rel32@hi+12
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX10-NEXT:    s_swappc_b64 s[30:31], s[8:9]
+; GFX10-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX10-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX10-NEXT:    s_sub_u32 s32, s32, 0x200
+; GFX10-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX10-NEXT:    s_or_saveexec_b32 s6, -1
+; GFX10-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s6
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    s_setpc_b64 s[4:5]
+  call amdgpu_gfx_callable void @external_void_func_v2f64_inreg(<2 x double> <double 2.0, double 4.0>)
+  ret void
+}
+
+define amdgpu_gfx_callable void @test_call_external_void_func_v3f64_imm_inreg() #0 {
+; GFX9-LABEL: test_call_external_void_func_v3f64_imm_inreg:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    s_or_saveexec_b64 s[4:5], -1
+; GFX9-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX9-NEXT:    s_mov_b64 exec, s[4:5]
+; GFX9-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX9-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX9-NEXT:    s_mov_b32 s33, s32
+; GFX9-NEXT:    s_add_u32 s32, s32, 0x400
+; GFX9-NEXT:    s_mov_b32 s4, 0
+; GFX9-NEXT:    s_mov_b32 s5, 2.0
+; GFX9-NEXT:    s_mov_b32 s6, 0
+; GFX9-NEXT:    s_mov_b32 s7, 0x40100000
+; GFX9-NEXT:    s_mov_b32 s8, 0
+; GFX9-NEXT:    s_mov_b32 s9, 0x40200000
+; GFX9-NEXT:    s_getpc_b64 s[10:11]
+; GFX9-NEXT:    s_add_u32 s10, s10, external_void_func_v3f64_inreg@rel32@lo+4
+; GFX9-NEXT:    s_addc_u32 s11, s11, external_void_func_v3f64_inreg@rel32@hi+12
+; GFX9-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX9-NEXT:    s_swappc_b64 s[30:31], s[10:11]
+; GFX9-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX9-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX9-NEXT:    s_sub_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX9-NEXT:    s_or_saveexec_b64 s[6:7], -1
+; GFX9-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX9-NEXT:    s_mov_b64 exec, s[6:7]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    s_setpc_b64 s[4:5]
+;
+; GFX10-LABEL: test_call_external_void_func_v3f64_imm_inreg:
+; GFX10:       ; %bb.0:
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    s_or_saveexec_b32 s4, -1
+; GFX10-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s4
+; GFX10-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX10-NEXT:    s_mov_b32 s4, 0
+; GFX10-NEXT:    s_mov_b32 s5, 2.0
+; GFX10-NEXT:    s_mov_b32 s6, 0
+; GFX10-NEXT:    s_mov_b32 s7, 0x40100000
+; GFX10-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX10-NEXT:    s_mov_b32 s8, 0
+; GFX10-NEXT:    s_mov_b32 s9, 0x40200000
+; GFX10-NEXT:    s_mov_b32 s33, s32
+; GFX10-NEXT:    s_add_u32 s32, s32, 0x200
+; GFX10-NEXT:    s_getpc_b64 s[10:11]
+; GFX10-NEXT:    s_add_u32 s10, s10, external_void_func_v3f64_inreg@rel32@lo+4
+; GFX10-NEXT:    s_addc_u32 s11, s11, external_void_func_v3f64_inreg@rel32@hi+12
+; GFX10-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    s_swappc_b64 s[30:31], s[10:11]
+; GFX10-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX10-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX10-NEXT:    s_sub_u32 s32, s32, 0x200
+; GFX10-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX10-NEXT:    s_or_saveexec_b32 s6, -1
+; GFX10-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s6
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    s_setpc_b64 s[4:5]
+  call amdgpu_gfx_callable void @external_void_func_v3f64_inreg(<3 x double> <double 2.0, double 4.0, double 8.0>)
+  ret void
+}
+
+define amdgpu_gfx_callable void @test_call_external_void_func_v2i16_inreg() #0 {
+; GFX9-LABEL: test_call_external_void_func_v2i16_inreg:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    s_or_saveexec_b64 s[4:5], -1
+; GFX9-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX9-NEXT:    s_mov_b64 exec, s[4:5]
+; GFX9-NEXT:    s_load_dword s4, s[4:5], 0x0
+; GFX9-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX9-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX9-NEXT:    s_mov_b32 s33, s32
+; GFX9-NEXT:    s_add_u32 s32, s32, 0x400
+; GFX9-NEXT:    s_getpc_b64 s[6:7]
+; GFX9-NEXT:    s_add_u32 s6, s6, external_void_func_v2i16_inreg@rel32@lo+4
+; GFX9-NEXT:    s_addc_u32 s7, s7, external_void_func_v2i16_inreg@rel32@hi+12
+; GFX9-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX9-NEXT:    s_swappc_b64 s[30:31], s[6:7]
+; GFX9-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX9-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX9-NEXT:    s_sub_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX9-NEXT:    s_or_saveexec_b64 s[6:7], -1
+; GFX9-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX9-NEXT:    s_mov_b64 exec, s[6:7]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    s_setpc_b64 s[4:5]
+;
+; GFX10-LABEL: test_call_external_void_func_v2i16_inreg:
+; GFX10:       ; %bb.0:
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    s_or_saveexec_b32 s4, -1
+; GFX10-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s4
+; GFX10-NEXT:    s_load_dword s4, s[4:5], 0x0
+; GFX10-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX10-NEXT:    s_mov_b32 s33, s32
+; GFX10-NEXT:    s_add_u32 s32, s32, 0x200
+; GFX10-NEXT:    s_getpc_b64 s[6:7]
+; GFX10-NEXT:    s_add_u32 s6, s6, external_void_func_v2i16_inreg@rel32@lo+4
+; GFX10-NEXT:    s_addc_u32 s7, s7, external_void_func_v2i16_inreg@rel32@hi+12
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX10-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX10-NEXT:    s_swappc_b64 s[30:31], s[6:7]
+; GFX10-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX10-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX10-NEXT:    s_sub_u32 s32, s32, 0x200
+; GFX10-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX10-NEXT:    s_or_saveexec_b32 s6, -1
+; GFX10-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s6
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    s_setpc_b64 s[4:5]
+  %val = load <2 x i16>, <2 x i16> addrspace(4)* undef
+  call amdgpu_gfx_callable void @external_void_func_v2i16_inreg(<2 x i16> %val)
+  ret void
+}
+
+define amdgpu_gfx_callable void @test_call_external_void_func_v3i16_inreg() #0 {
+; GFX9-LABEL: test_call_external_void_func_v3i16_inreg:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    s_or_saveexec_b64 s[4:5], -1
+; GFX9-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX9-NEXT:    s_mov_b64 exec, s[4:5]
+; GFX9-NEXT:    s_load_dwordx2 s[4:5], s[4:5], 0x0
+; GFX9-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX9-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX9-NEXT:    s_mov_b32 s33, s32
+; GFX9-NEXT:    s_add_u32 s32, s32, 0x400
+; GFX9-NEXT:    s_getpc_b64 s[6:7]
+; GFX9-NEXT:    s_add_u32 s6, s6, external_void_func_v3i16_inreg@rel32@lo+4
+; GFX9-NEXT:    s_addc_u32 s7, s7, external_void_func_v3i16_inreg@rel32@hi+12
+; GFX9-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX9-NEXT:    s_swappc_b64 s[30:31], s[6:7]
+; GFX9-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX9-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX9-NEXT:    s_sub_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX9-NEXT:    s_or_saveexec_b64 s[6:7], -1
+; GFX9-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX9-NEXT:    s_mov_b64 exec, s[6:7]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    s_setpc_b64 s[4:5]
+;
+; GFX10-LABEL: test_call_external_void_func_v3i16_inreg:
+; GFX10:       ; %bb.0:
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    s_or_saveexec_b32 s4, -1
+; GFX10-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s4
+; GFX10-NEXT:    s_load_dwordx2 s[4:5], s[4:5], 0x0
+; GFX10-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX10-NEXT:    s_mov_b32 s33, s32
+; GFX10-NEXT:    s_add_u32 s32, s32, 0x200
+; GFX10-NEXT:    s_getpc_b64 s[6:7]
+; GFX10-NEXT:    s_add_u32 s6, s6, external_void_func_v3i16_inreg@rel32@lo+4
+; GFX10-NEXT:    s_addc_u32 s7, s7, external_void_func_v3i16_inreg@rel32@hi+12
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX10-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX10-NEXT:    s_swappc_b64 s[30:31], s[6:7]
+; GFX10-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX10-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX10-NEXT:    s_sub_u32 s32, s32, 0x200
+; GFX10-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX10-NEXT:    s_or_saveexec_b32 s6, -1
+; GFX10-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s6
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    s_setpc_b64 s[4:5]
+  %val = load <3 x i16>, <3 x i16> addrspace(4)* undef
+  call amdgpu_gfx_callable void @external_void_func_v3i16_inreg(<3 x i16> %val)
+  ret void
+}
+
+define amdgpu_gfx_callable void @test_call_external_void_func_v3f16_inreg() #0 {
+; GFX9-LABEL: test_call_external_void_func_v3f16_inreg:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    s_or_saveexec_b64 s[4:5], -1
+; GFX9-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX9-NEXT:    s_mov_b64 exec, s[4:5]
+; GFX9-NEXT:    s_load_dwordx2 s[4:5], s[4:5], 0x0
+; GFX9-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX9-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX9-NEXT:    s_mov_b32 s33, s32
+; GFX9-NEXT:    s_add_u32 s32, s32, 0x400
+; GFX9-NEXT:    s_getpc_b64 s[6:7]
+; GFX9-NEXT:    s_add_u32 s6, s6, external_void_func_v3f16_inreg@rel32@lo+4
+; GFX9-NEXT:    s_addc_u32 s7, s7, external_void_func_v3f16_inreg@rel32@hi+12
+; GFX9-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX9-NEXT:    s_swappc_b64 s[30:31], s[6:7]
+; GFX9-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX9-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX9-NEXT:    s_sub_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX9-NEXT:    s_or_saveexec_b64 s[6:7], -1
+; GFX9-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX9-NEXT:    s_mov_b64 exec, s[6:7]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    s_setpc_b64 s[4:5]
+;
+; GFX10-LABEL: test_call_external_void_func_v3f16_inreg:
+; GFX10:       ; %bb.0:
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    s_or_saveexec_b32 s4, -1
+; GFX10-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s4
+; GFX10-NEXT:    s_load_dwordx2 s[4:5], s[4:5], 0x0
+; GFX10-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX10-NEXT:    s_mov_b32 s33, s32
+; GFX10-NEXT:    s_add_u32 s32, s32, 0x200
+; GFX10-NEXT:    s_getpc_b64 s[6:7]
+; GFX10-NEXT:    s_add_u32 s6, s6, external_void_func_v3f16_inreg@rel32@lo+4
+; GFX10-NEXT:    s_addc_u32 s7, s7, external_void_func_v3f16_inreg@rel32@hi+12
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX10-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX10-NEXT:    s_swappc_b64 s[30:31], s[6:7]
+; GFX10-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX10-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX10-NEXT:    s_sub_u32 s32, s32, 0x200
+; GFX10-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX10-NEXT:    s_or_saveexec_b32 s6, -1
+; GFX10-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s6
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    s_setpc_b64 s[4:5]
+  %val = load <3 x half>, <3 x half> addrspace(4)* undef
+  call amdgpu_gfx_callable void @external_void_func_v3f16_inreg(<3 x half> %val)
+  ret void
+}
+
+define amdgpu_gfx_callable void @test_call_external_void_func_v3i16_imm_inreg() #0 {
+; GFX9-LABEL: test_call_external_void_func_v3i16_imm_inreg:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    s_or_saveexec_b64 s[4:5], -1
+; GFX9-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX9-NEXT:    s_mov_b64 exec, s[4:5]
+; GFX9-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX9-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX9-NEXT:    s_mov_b32 s33, s32
+; GFX9-NEXT:    s_add_u32 s32, s32, 0x400
+; GFX9-NEXT:    s_mov_b32 s4, 0x20001
+; GFX9-NEXT:    s_mov_b32 s5, 3
+; GFX9-NEXT:    s_getpc_b64 s[6:7]
+; GFX9-NEXT:    s_add_u32 s6, s6, external_void_func_v3i16_inreg@rel32@lo+4
+; GFX9-NEXT:    s_addc_u32 s7, s7, external_void_func_v3i16_inreg@rel32@hi+12
+; GFX9-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX9-NEXT:    s_swappc_b64 s[30:31], s[6:7]
+; GFX9-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX9-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX9-NEXT:    s_sub_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX9-NEXT:    s_or_saveexec_b64 s[6:7], -1
+; GFX9-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX9-NEXT:    s_mov_b64 exec, s[6:7]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    s_setpc_b64 s[4:5]
+;
+; GFX10-LABEL: test_call_external_void_func_v3i16_imm_inreg:
+; GFX10:       ; %bb.0:
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    s_or_saveexec_b32 s4, -1
+; GFX10-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s4
+; GFX10-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX10-NEXT:    s_mov_b32 s4, 0x20001
+; GFX10-NEXT:    s_mov_b32 s5, 3
+; GFX10-NEXT:    s_mov_b32 s33, s32
+; GFX10-NEXT:    s_add_u32 s32, s32, 0x200
+; GFX10-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX10-NEXT:    s_getpc_b64 s[6:7]
+; GFX10-NEXT:    s_add_u32 s6, s6, external_void_func_v3i16_inreg@rel32@lo+4
+; GFX10-NEXT:    s_addc_u32 s7, s7, external_void_func_v3i16_inreg@rel32@hi+12
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX10-NEXT:    s_swappc_b64 s[30:31], s[6:7]
+; GFX10-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX10-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX10-NEXT:    s_sub_u32 s32, s32, 0x200
+; GFX10-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX10-NEXT:    s_or_saveexec_b32 s6, -1
+; GFX10-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s6
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    s_setpc_b64 s[4:5]
+  call amdgpu_gfx_callable void @external_void_func_v3i16_inreg(<3 x i16> <i16 1, i16 2, i16 3>)
+  ret void
+}
+
+define amdgpu_gfx_callable void @test_call_external_void_func_v3f16_imm_inreg() #0 {
+; GFX9-LABEL: test_call_external_void_func_v3f16_imm_inreg:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    s_or_saveexec_b64 s[4:5], -1
+; GFX9-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX9-NEXT:    s_mov_b64 exec, s[4:5]
+; GFX9-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX9-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX9-NEXT:    s_mov_b32 s33, s32
+; GFX9-NEXT:    s_add_u32 s32, s32, 0x400
+; GFX9-NEXT:    s_mov_b32 s4, 0x40003c00
+; GFX9-NEXT:    s_movk_i32 s5, 0x4400
+; GFX9-NEXT:    s_getpc_b64 s[6:7]
+; GFX9-NEXT:    s_add_u32 s6, s6, external_void_func_v3f16_inreg@rel32@lo+4
+; GFX9-NEXT:    s_addc_u32 s7, s7, external_void_func_v3f16_inreg@rel32@hi+12
+; GFX9-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX9-NEXT:    s_swappc_b64 s[30:31], s[6:7]
+; GFX9-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX9-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX9-NEXT:    s_sub_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX9-NEXT:    s_or_saveexec_b64 s[6:7], -1
+; GFX9-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX9-NEXT:    s_mov_b64 exec, s[6:7]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    s_setpc_b64 s[4:5]
+;
+; GFX10-LABEL: test_call_external_void_func_v3f16_imm_inreg:
+; GFX10:       ; %bb.0:
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    s_or_saveexec_b32 s4, -1
+; GFX10-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s4
+; GFX10-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX10-NEXT:    s_mov_b32 s4, 0x40003c00
+; GFX10-NEXT:    s_movk_i32 s5, 0x4400
+; GFX10-NEXT:    s_mov_b32 s33, s32
+; GFX10-NEXT:    s_add_u32 s32, s32, 0x200
+; GFX10-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX10-NEXT:    s_getpc_b64 s[6:7]
+; GFX10-NEXT:    s_add_u32 s6, s6, external_void_func_v3f16_inreg@rel32@lo+4
+; GFX10-NEXT:    s_addc_u32 s7, s7, external_void_func_v3f16_inreg@rel32@hi+12
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX10-NEXT:    s_swappc_b64 s[30:31], s[6:7]
+; GFX10-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX10-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX10-NEXT:    s_sub_u32 s32, s32, 0x200
+; GFX10-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX10-NEXT:    s_or_saveexec_b32 s6, -1
+; GFX10-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s6
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    s_setpc_b64 s[4:5]
+  call amdgpu_gfx_callable void @external_void_func_v3f16_inreg(<3 x half> <half 1.0, half 2.0, half 4.0>)
+  ret void
+}
+
+define amdgpu_gfx_callable void @test_call_external_void_func_v4i16_inreg() #0 {
+; GFX9-LABEL: test_call_external_void_func_v4i16_inreg:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    s_or_saveexec_b64 s[4:5], -1
+; GFX9-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX9-NEXT:    s_mov_b64 exec, s[4:5]
+; GFX9-NEXT:    s_load_dwordx2 s[4:5], s[4:5], 0x0
+; GFX9-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX9-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX9-NEXT:    s_mov_b32 s33, s32
+; GFX9-NEXT:    s_add_u32 s32, s32, 0x400
+; GFX9-NEXT:    s_getpc_b64 s[6:7]
+; GFX9-NEXT:    s_add_u32 s6, s6, external_void_func_v4i16_inreg@rel32@lo+4
+; GFX9-NEXT:    s_addc_u32 s7, s7, external_void_func_v4i16_inreg@rel32@hi+12
+; GFX9-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX9-NEXT:    s_swappc_b64 s[30:31], s[6:7]
+; GFX9-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX9-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX9-NEXT:    s_sub_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX9-NEXT:    s_or_saveexec_b64 s[6:7], -1
+; GFX9-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX9-NEXT:    s_mov_b64 exec, s[6:7]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    s_setpc_b64 s[4:5]
+;
+; GFX10-LABEL: test_call_external_void_func_v4i16_inreg:
+; GFX10:       ; %bb.0:
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    s_or_saveexec_b32 s4, -1
+; GFX10-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s4
+; GFX10-NEXT:    s_load_dwordx2 s[4:5], s[4:5], 0x0
+; GFX10-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX10-NEXT:    s_mov_b32 s33, s32
+; GFX10-NEXT:    s_add_u32 s32, s32, 0x200
+; GFX10-NEXT:    s_getpc_b64 s[6:7]
+; GFX10-NEXT:    s_add_u32 s6, s6, external_void_func_v4i16_inreg@rel32@lo+4
+; GFX10-NEXT:    s_addc_u32 s7, s7, external_void_func_v4i16_inreg@rel32@hi+12
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX10-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX10-NEXT:    s_swappc_b64 s[30:31], s[6:7]
+; GFX10-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX10-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX10-NEXT:    s_sub_u32 s32, s32, 0x200
+; GFX10-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX10-NEXT:    s_or_saveexec_b32 s6, -1
+; GFX10-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s6
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    s_setpc_b64 s[4:5]
+  %val = load <4 x i16>, <4 x i16> addrspace(4)* undef
+  call amdgpu_gfx_callable void @external_void_func_v4i16_inreg(<4 x i16> %val)
+  ret void
+}
+
+define amdgpu_gfx_callable void @test_call_external_void_func_v4i16_imm_inreg() #0 {
+; GFX9-LABEL: test_call_external_void_func_v4i16_imm_inreg:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    s_or_saveexec_b64 s[4:5], -1
+; GFX9-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX9-NEXT:    s_mov_b64 exec, s[4:5]
+; GFX9-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX9-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX9-NEXT:    s_mov_b32 s33, s32
+; GFX9-NEXT:    s_add_u32 s32, s32, 0x400
+; GFX9-NEXT:    s_mov_b32 s4, 0x20001
+; GFX9-NEXT:    s_mov_b32 s5, 0x40003
+; GFX9-NEXT:    s_getpc_b64 s[6:7]
+; GFX9-NEXT:    s_add_u32 s6, s6, external_void_func_v4i16_inreg@rel32@lo+4
+; GFX9-NEXT:    s_addc_u32 s7, s7, external_void_func_v4i16_inreg@rel32@hi+12
+; GFX9-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX9-NEXT:    s_swappc_b64 s[30:31], s[6:7]
+; GFX9-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX9-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX9-NEXT:    s_sub_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX9-NEXT:    s_or_saveexec_b64 s[6:7], -1
+; GFX9-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX9-NEXT:    s_mov_b64 exec, s[6:7]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    s_setpc_b64 s[4:5]
+;
+; GFX10-LABEL: test_call_external_void_func_v4i16_imm_inreg:
+; GFX10:       ; %bb.0:
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    s_or_saveexec_b32 s4, -1
+; GFX10-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s4
+; GFX10-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX10-NEXT:    s_mov_b32 s4, 0x20001
+; GFX10-NEXT:    s_mov_b32 s5, 0x40003
+; GFX10-NEXT:    s_mov_b32 s33, s32
+; GFX10-NEXT:    s_add_u32 s32, s32, 0x200
+; GFX10-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX10-NEXT:    s_getpc_b64 s[6:7]
+; GFX10-NEXT:    s_add_u32 s6, s6, external_void_func_v4i16_inreg@rel32@lo+4
+; GFX10-NEXT:    s_addc_u32 s7, s7, external_void_func_v4i16_inreg@rel32@hi+12
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX10-NEXT:    s_swappc_b64 s[30:31], s[6:7]
+; GFX10-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX10-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX10-NEXT:    s_sub_u32 s32, s32, 0x200
+; GFX10-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX10-NEXT:    s_or_saveexec_b32 s6, -1
+; GFX10-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s6
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    s_setpc_b64 s[4:5]
+  call amdgpu_gfx_callable void @external_void_func_v4i16_inreg(<4 x i16> <i16 1, i16 2, i16 3, i16 4>)
+  ret void
+}
+
+define amdgpu_gfx_callable void @test_call_external_void_func_v2f16_inreg() #0 {
+; GFX9-LABEL: test_call_external_void_func_v2f16_inreg:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    s_or_saveexec_b64 s[4:5], -1
+; GFX9-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX9-NEXT:    s_mov_b64 exec, s[4:5]
+; GFX9-NEXT:    s_load_dword s4, s[4:5], 0x0
+; GFX9-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX9-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX9-NEXT:    s_mov_b32 s33, s32
+; GFX9-NEXT:    s_add_u32 s32, s32, 0x400
+; GFX9-NEXT:    s_getpc_b64 s[6:7]
+; GFX9-NEXT:    s_add_u32 s6, s6, external_void_func_v2f16_inreg@rel32@lo+4
+; GFX9-NEXT:    s_addc_u32 s7, s7, external_void_func_v2f16_inreg@rel32@hi+12
+; GFX9-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX9-NEXT:    s_swappc_b64 s[30:31], s[6:7]
+; GFX9-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX9-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX9-NEXT:    s_sub_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX9-NEXT:    s_or_saveexec_b64 s[6:7], -1
+; GFX9-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX9-NEXT:    s_mov_b64 exec, s[6:7]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    s_setpc_b64 s[4:5]
+;
+; GFX10-LABEL: test_call_external_void_func_v2f16_inreg:
+; GFX10:       ; %bb.0:
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    s_or_saveexec_b32 s4, -1
+; GFX10-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s4
+; GFX10-NEXT:    s_load_dword s4, s[4:5], 0x0
+; GFX10-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX10-NEXT:    s_mov_b32 s33, s32
+; GFX10-NEXT:    s_add_u32 s32, s32, 0x200
+; GFX10-NEXT:    s_getpc_b64 s[6:7]
+; GFX10-NEXT:    s_add_u32 s6, s6, external_void_func_v2f16_inreg@rel32@lo+4
+; GFX10-NEXT:    s_addc_u32 s7, s7, external_void_func_v2f16_inreg@rel32@hi+12
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX10-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX10-NEXT:    s_swappc_b64 s[30:31], s[6:7]
+; GFX10-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX10-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX10-NEXT:    s_sub_u32 s32, s32, 0x200
+; GFX10-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX10-NEXT:    s_or_saveexec_b32 s6, -1
+; GFX10-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s6
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    s_setpc_b64 s[4:5]
+  %val = load <2 x half>, <2 x half> addrspace(4)* undef
+  call amdgpu_gfx_callable void @external_void_func_v2f16_inreg(<2 x half> %val)
+  ret void
+}
+
+define amdgpu_gfx_callable void @test_call_external_void_func_v2i32_inreg() #0 {
+; GFX9-LABEL: test_call_external_void_func_v2i32_inreg:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    s_or_saveexec_b64 s[4:5], -1
+; GFX9-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX9-NEXT:    s_mov_b64 exec, s[4:5]
+; GFX9-NEXT:    s_load_dwordx2 s[4:5], s[4:5], 0x0
+; GFX9-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX9-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX9-NEXT:    s_mov_b32 s33, s32
+; GFX9-NEXT:    s_add_u32 s32, s32, 0x400
+; GFX9-NEXT:    s_getpc_b64 s[6:7]
+; GFX9-NEXT:    s_add_u32 s6, s6, external_void_func_v2i32_inreg@rel32@lo+4
+; GFX9-NEXT:    s_addc_u32 s7, s7, external_void_func_v2i32_inreg@rel32@hi+12
+; GFX9-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX9-NEXT:    s_swappc_b64 s[30:31], s[6:7]
+; GFX9-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX9-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX9-NEXT:    s_sub_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX9-NEXT:    s_or_saveexec_b64 s[6:7], -1
+; GFX9-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX9-NEXT:    s_mov_b64 exec, s[6:7]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    s_setpc_b64 s[4:5]
+;
+; GFX10-LABEL: test_call_external_void_func_v2i32_inreg:
+; GFX10:       ; %bb.0:
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    s_or_saveexec_b32 s4, -1
+; GFX10-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s4
+; GFX10-NEXT:    s_load_dwordx2 s[4:5], s[4:5], 0x0
+; GFX10-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX10-NEXT:    s_mov_b32 s33, s32
+; GFX10-NEXT:    s_add_u32 s32, s32, 0x200
+; GFX10-NEXT:    s_getpc_b64 s[6:7]
+; GFX10-NEXT:    s_add_u32 s6, s6, external_void_func_v2i32_inreg@rel32@lo+4
+; GFX10-NEXT:    s_addc_u32 s7, s7, external_void_func_v2i32_inreg@rel32@hi+12
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX10-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX10-NEXT:    s_swappc_b64 s[30:31], s[6:7]
+; GFX10-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX10-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX10-NEXT:    s_sub_u32 s32, s32, 0x200
+; GFX10-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX10-NEXT:    s_or_saveexec_b32 s6, -1
+; GFX10-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s6
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    s_setpc_b64 s[4:5]
+  %val = load <2 x i32>, <2 x i32> addrspace(4)* undef
+  call amdgpu_gfx_callable void @external_void_func_v2i32_inreg(<2 x i32> %val)
+  ret void
+}
+
+define amdgpu_gfx_callable void @test_call_external_void_func_v2i32_imm_inreg() #0 {
+; GFX9-LABEL: test_call_external_void_func_v2i32_imm_inreg:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    s_or_saveexec_b64 s[4:5], -1
+; GFX9-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX9-NEXT:    s_mov_b64 exec, s[4:5]
+; GFX9-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX9-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX9-NEXT:    s_mov_b32 s33, s32
+; GFX9-NEXT:    s_add_u32 s32, s32, 0x400
+; GFX9-NEXT:    s_mov_b32 s4, 1
+; GFX9-NEXT:    s_mov_b32 s5, 2
+; GFX9-NEXT:    s_getpc_b64 s[6:7]
+; GFX9-NEXT:    s_add_u32 s6, s6, external_void_func_v2i32_inreg@rel32@lo+4
+; GFX9-NEXT:    s_addc_u32 s7, s7, external_void_func_v2i32_inreg@rel32@hi+12
+; GFX9-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX9-NEXT:    s_swappc_b64 s[30:31], s[6:7]
+; GFX9-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX9-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX9-NEXT:    s_sub_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX9-NEXT:    s_or_saveexec_b64 s[6:7], -1
+; GFX9-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX9-NEXT:    s_mov_b64 exec, s[6:7]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    s_setpc_b64 s[4:5]
+;
+; GFX10-LABEL: test_call_external_void_func_v2i32_imm_inreg:
+; GFX10:       ; %bb.0:
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    s_or_saveexec_b32 s4, -1
+; GFX10-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s4
+; GFX10-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX10-NEXT:    s_mov_b32 s4, 1
+; GFX10-NEXT:    s_mov_b32 s5, 2
+; GFX10-NEXT:    s_mov_b32 s33, s32
+; GFX10-NEXT:    s_add_u32 s32, s32, 0x200
+; GFX10-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX10-NEXT:    s_getpc_b64 s[6:7]
+; GFX10-NEXT:    s_add_u32 s6, s6, external_void_func_v2i32_inreg@rel32@lo+4
+; GFX10-NEXT:    s_addc_u32 s7, s7, external_void_func_v2i32_inreg@rel32@hi+12
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX10-NEXT:    s_swappc_b64 s[30:31], s[6:7]
+; GFX10-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX10-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX10-NEXT:    s_sub_u32 s32, s32, 0x200
+; GFX10-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX10-NEXT:    s_or_saveexec_b32 s6, -1
+; GFX10-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s6
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    s_setpc_b64 s[4:5]
+  call amdgpu_gfx_callable void @external_void_func_v2i32_inreg(<2 x i32> <i32 1, i32 2>)
+  ret void
+}
+
+define amdgpu_gfx_callable void @test_call_external_void_func_v3i32_imm_inreg(i32) #0 {
+; GFX9-LABEL: test_call_external_void_func_v3i32_imm_inreg:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    s_or_saveexec_b64 s[4:5], -1
+; GFX9-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX9-NEXT:    s_mov_b64 exec, s[4:5]
+; GFX9-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX9-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX9-NEXT:    s_mov_b32 s33, s32
+; GFX9-NEXT:    s_add_u32 s32, s32, 0x400
+; GFX9-NEXT:    s_mov_b32 s4, 3
+; GFX9-NEXT:    s_mov_b32 s5, 4
+; GFX9-NEXT:    s_mov_b32 s6, 5
+; GFX9-NEXT:    s_getpc_b64 s[8:9]
+; GFX9-NEXT:    s_add_u32 s8, s8, external_void_func_v3i32_inreg@rel32@lo+4
+; GFX9-NEXT:    s_addc_u32 s9, s9, external_void_func_v3i32_inreg@rel32@hi+12
+; GFX9-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX9-NEXT:    s_swappc_b64 s[30:31], s[8:9]
+; GFX9-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX9-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX9-NEXT:    s_sub_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX9-NEXT:    s_or_saveexec_b64 s[6:7], -1
+; GFX9-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX9-NEXT:    s_mov_b64 exec, s[6:7]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    s_setpc_b64 s[4:5]
+;
+; GFX10-LABEL: test_call_external_void_func_v3i32_imm_inreg:
+; GFX10:       ; %bb.0:
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    s_or_saveexec_b32 s4, -1
+; GFX10-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s4
+; GFX10-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX10-NEXT:    s_mov_b32 s4, 3
+; GFX10-NEXT:    s_mov_b32 s5, 4
+; GFX10-NEXT:    s_mov_b32 s6, 5
+; GFX10-NEXT:    s_mov_b32 s33, s32
+; GFX10-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX10-NEXT:    s_add_u32 s32, s32, 0x200
+; GFX10-NEXT:    s_getpc_b64 s[8:9]
+; GFX10-NEXT:    s_add_u32 s8, s8, external_void_func_v3i32_inreg@rel32@lo+4
+; GFX10-NEXT:    s_addc_u32 s9, s9, external_void_func_v3i32_inreg@rel32@hi+12
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX10-NEXT:    s_swappc_b64 s[30:31], s[8:9]
+; GFX10-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX10-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX10-NEXT:    s_sub_u32 s32, s32, 0x200
+; GFX10-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX10-NEXT:    s_or_saveexec_b32 s6, -1
+; GFX10-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s6
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    s_setpc_b64 s[4:5]
+  call amdgpu_gfx_callable void @external_void_func_v3i32_inreg(<3 x i32> <i32 3, i32 4, i32 5>)
+  ret void
+}
+
+define amdgpu_gfx_callable void @test_call_external_void_func_v3i32_i32_inreg(i32) #0 {
+; GFX9-LABEL: test_call_external_void_func_v3i32_i32_inreg:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    s_or_saveexec_b64 s[4:5], -1
+; GFX9-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX9-NEXT:    s_mov_b64 exec, s[4:5]
+; GFX9-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX9-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX9-NEXT:    s_mov_b32 s33, s32
+; GFX9-NEXT:    s_add_u32 s32, s32, 0x400
+; GFX9-NEXT:    s_mov_b32 s4, 3
+; GFX9-NEXT:    s_mov_b32 s5, 4
+; GFX9-NEXT:    s_mov_b32 s6, 5
+; GFX9-NEXT:    s_mov_b32 s7, 6
+; GFX9-NEXT:    s_getpc_b64 s[8:9]
+; GFX9-NEXT:    s_add_u32 s8, s8, external_void_func_v3i32_i32_inreg@rel32@lo+4
+; GFX9-NEXT:    s_addc_u32 s9, s9, external_void_func_v3i32_i32_inreg@rel32@hi+12
+; GFX9-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX9-NEXT:    s_swappc_b64 s[30:31], s[8:9]
+; GFX9-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX9-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX9-NEXT:    s_sub_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX9-NEXT:    s_or_saveexec_b64 s[6:7], -1
+; GFX9-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX9-NEXT:    s_mov_b64 exec, s[6:7]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    s_setpc_b64 s[4:5]
+;
+; GFX10-LABEL: test_call_external_void_func_v3i32_i32_inreg:
+; GFX10:       ; %bb.0:
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    s_or_saveexec_b32 s4, -1
+; GFX10-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s4
+; GFX10-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX10-NEXT:    s_mov_b32 s4, 3
+; GFX10-NEXT:    s_mov_b32 s5, 4
+; GFX10-NEXT:    s_mov_b32 s6, 5
+; GFX10-NEXT:    s_mov_b32 s7, 6
+; GFX10-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX10-NEXT:    s_mov_b32 s33, s32
+; GFX10-NEXT:    s_add_u32 s32, s32, 0x200
+; GFX10-NEXT:    s_getpc_b64 s[8:9]
+; GFX10-NEXT:    s_add_u32 s8, s8, external_void_func_v3i32_i32_inreg@rel32@lo+4
+; GFX10-NEXT:    s_addc_u32 s9, s9, external_void_func_v3i32_i32_inreg@rel32@hi+12
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX10-NEXT:    s_swappc_b64 s[30:31], s[8:9]
+; GFX10-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX10-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX10-NEXT:    s_sub_u32 s32, s32, 0x200
+; GFX10-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX10-NEXT:    s_or_saveexec_b32 s6, -1
+; GFX10-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s6
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    s_setpc_b64 s[4:5]
+  call amdgpu_gfx_callable void @external_void_func_v3i32_i32_inreg(<3 x i32> <i32 3, i32 4, i32 5>, i32 6)
+  ret void
+}
+
+define amdgpu_gfx_callable void @test_call_external_void_func_v4i32_inreg() #0 {
+; GFX9-LABEL: test_call_external_void_func_v4i32_inreg:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    s_or_saveexec_b64 s[4:5], -1
+; GFX9-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX9-NEXT:    s_mov_b64 exec, s[4:5]
+; GFX9-NEXT:    s_load_dwordx4 s[4:7], s[4:5], 0x0
+; GFX9-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX9-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX9-NEXT:    s_mov_b32 s33, s32
+; GFX9-NEXT:    s_add_u32 s32, s32, 0x400
+; GFX9-NEXT:    s_getpc_b64 s[8:9]
+; GFX9-NEXT:    s_add_u32 s8, s8, external_void_func_v4i32_inreg@rel32@lo+4
+; GFX9-NEXT:    s_addc_u32 s9, s9, external_void_func_v4i32_inreg@rel32@hi+12
+; GFX9-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX9-NEXT:    s_swappc_b64 s[30:31], s[8:9]
+; GFX9-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX9-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX9-NEXT:    s_sub_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX9-NEXT:    s_or_saveexec_b64 s[6:7], -1
+; GFX9-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX9-NEXT:    s_mov_b64 exec, s[6:7]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    s_setpc_b64 s[4:5]
+;
+; GFX10-LABEL: test_call_external_void_func_v4i32_inreg:
+; GFX10:       ; %bb.0:
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    s_or_saveexec_b32 s4, -1
+; GFX10-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s4
+; GFX10-NEXT:    s_load_dwordx4 s[4:7], s[4:5], 0x0
+; GFX10-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX10-NEXT:    s_mov_b32 s33, s32
+; GFX10-NEXT:    s_add_u32 s32, s32, 0x200
+; GFX10-NEXT:    s_getpc_b64 s[8:9]
+; GFX10-NEXT:    s_add_u32 s8, s8, external_void_func_v4i32_inreg@rel32@lo+4
+; GFX10-NEXT:    s_addc_u32 s9, s9, external_void_func_v4i32_inreg@rel32@hi+12
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX10-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX10-NEXT:    s_swappc_b64 s[30:31], s[8:9]
+; GFX10-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX10-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX10-NEXT:    s_sub_u32 s32, s32, 0x200
+; GFX10-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX10-NEXT:    s_or_saveexec_b32 s6, -1
+; GFX10-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s6
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    s_setpc_b64 s[4:5]
+  %val = load <4 x i32>, <4 x i32> addrspace(4)* undef
+  call amdgpu_gfx_callable void @external_void_func_v4i32_inreg(<4 x i32> %val)
+  ret void
+}
+
+define amdgpu_gfx_callable void @test_call_external_void_func_v4i32_imm_inreg() #0 {
+; GFX9-LABEL: test_call_external_void_func_v4i32_imm_inreg:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    s_or_saveexec_b64 s[4:5], -1
+; GFX9-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX9-NEXT:    s_mov_b64 exec, s[4:5]
+; GFX9-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX9-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX9-NEXT:    s_mov_b32 s33, s32
+; GFX9-NEXT:    s_add_u32 s32, s32, 0x400
+; GFX9-NEXT:    s_mov_b32 s4, 1
+; GFX9-NEXT:    s_mov_b32 s5, 2
+; GFX9-NEXT:    s_mov_b32 s6, 3
+; GFX9-NEXT:    s_mov_b32 s7, 4
+; GFX9-NEXT:    s_getpc_b64 s[8:9]
+; GFX9-NEXT:    s_add_u32 s8, s8, external_void_func_v4i32_inreg@rel32@lo+4
+; GFX9-NEXT:    s_addc_u32 s9, s9, external_void_func_v4i32_inreg@rel32@hi+12
+; GFX9-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX9-NEXT:    s_swappc_b64 s[30:31], s[8:9]
+; GFX9-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX9-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX9-NEXT:    s_sub_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX9-NEXT:    s_or_saveexec_b64 s[6:7], -1
+; GFX9-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX9-NEXT:    s_mov_b64 exec, s[6:7]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    s_setpc_b64 s[4:5]
+;
+; GFX10-LABEL: test_call_external_void_func_v4i32_imm_inreg:
+; GFX10:       ; %bb.0:
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    s_or_saveexec_b32 s4, -1
+; GFX10-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s4
+; GFX10-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX10-NEXT:    s_mov_b32 s4, 1
+; GFX10-NEXT:    s_mov_b32 s5, 2
+; GFX10-NEXT:    s_mov_b32 s6, 3
+; GFX10-NEXT:    s_mov_b32 s7, 4
+; GFX10-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX10-NEXT:    s_mov_b32 s33, s32
+; GFX10-NEXT:    s_add_u32 s32, s32, 0x200
+; GFX10-NEXT:    s_getpc_b64 s[8:9]
+; GFX10-NEXT:    s_add_u32 s8, s8, external_void_func_v4i32_inreg@rel32@lo+4
+; GFX10-NEXT:    s_addc_u32 s9, s9, external_void_func_v4i32_inreg@rel32@hi+12
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX10-NEXT:    s_swappc_b64 s[30:31], s[8:9]
+; GFX10-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX10-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX10-NEXT:    s_sub_u32 s32, s32, 0x200
+; GFX10-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX10-NEXT:    s_or_saveexec_b32 s6, -1
+; GFX10-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s6
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    s_setpc_b64 s[4:5]
+  call amdgpu_gfx_callable void @external_void_func_v4i32_inreg(<4 x i32> <i32 1, i32 2, i32 3, i32 4>)
+  ret void
+}
+
+define amdgpu_gfx_callable void @test_call_external_void_func_v5i32_imm_inreg() #0 {
+; GFX9-LABEL: test_call_external_void_func_v5i32_imm_inreg:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    s_or_saveexec_b64 s[4:5], -1
+; GFX9-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX9-NEXT:    s_mov_b64 exec, s[4:5]
+; GFX9-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX9-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX9-NEXT:    s_mov_b32 s33, s32
+; GFX9-NEXT:    s_add_u32 s32, s32, 0x400
+; GFX9-NEXT:    s_mov_b32 s4, 1
+; GFX9-NEXT:    s_mov_b32 s5, 2
+; GFX9-NEXT:    s_mov_b32 s6, 3
+; GFX9-NEXT:    s_mov_b32 s7, 4
+; GFX9-NEXT:    s_mov_b32 s8, 5
+; GFX9-NEXT:    s_getpc_b64 s[10:11]
+; GFX9-NEXT:    s_add_u32 s10, s10, external_void_func_v5i32_inreg@rel32@lo+4
+; GFX9-NEXT:    s_addc_u32 s11, s11, external_void_func_v5i32_inreg@rel32@hi+12
+; GFX9-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX9-NEXT:    s_swappc_b64 s[30:31], s[10:11]
+; GFX9-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX9-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX9-NEXT:    s_sub_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX9-NEXT:    s_or_saveexec_b64 s[6:7], -1
+; GFX9-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX9-NEXT:    s_mov_b64 exec, s[6:7]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    s_setpc_b64 s[4:5]
+;
+; GFX10-LABEL: test_call_external_void_func_v5i32_imm_inreg:
+; GFX10:       ; %bb.0:
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    s_or_saveexec_b32 s4, -1
+; GFX10-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s4
+; GFX10-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX10-NEXT:    s_mov_b32 s4, 1
+; GFX10-NEXT:    s_mov_b32 s5, 2
+; GFX10-NEXT:    s_mov_b32 s6, 3
+; GFX10-NEXT:    s_mov_b32 s7, 4
+; GFX10-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX10-NEXT:    s_mov_b32 s8, 5
+; GFX10-NEXT:    s_mov_b32 s33, s32
+; GFX10-NEXT:    s_add_u32 s32, s32, 0x200
+; GFX10-NEXT:    s_getpc_b64 s[10:11]
+; GFX10-NEXT:    s_add_u32 s10, s10, external_void_func_v5i32_inreg@rel32@lo+4
+; GFX10-NEXT:    s_addc_u32 s11, s11, external_void_func_v5i32_inreg@rel32@hi+12
+; GFX10-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    s_swappc_b64 s[30:31], s[10:11]
+; GFX10-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX10-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX10-NEXT:    s_sub_u32 s32, s32, 0x200
+; GFX10-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX10-NEXT:    s_or_saveexec_b32 s6, -1
+; GFX10-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s6
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    s_setpc_b64 s[4:5]
+  call amdgpu_gfx_callable void @external_void_func_v5i32_inreg(<5 x i32> <i32 1, i32 2, i32 3, i32 4, i32 5>)
+  ret void
+}
+
+define amdgpu_gfx_callable void @test_call_external_void_func_v8i32_inreg() #0 {
+; GFX9-LABEL: test_call_external_void_func_v8i32_inreg:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    s_or_saveexec_b64 s[4:5], -1
+; GFX9-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX9-NEXT:    s_mov_b64 exec, s[4:5]
+; GFX9-NEXT:    s_load_dwordx2 s[4:5], s[4:5], 0x0
+; GFX9-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX9-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX9-NEXT:    s_mov_b32 s33, s32
+; GFX9-NEXT:    s_add_u32 s32, s32, 0x400
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    s_load_dwordx8 s[4:11], s[4:5], 0x0
+; GFX9-NEXT:    s_getpc_b64 s[12:13]
+; GFX9-NEXT:    s_add_u32 s12, s12, external_void_func_v8i32_inreg@rel32@lo+4
+; GFX9-NEXT:    s_addc_u32 s13, s13, external_void_func_v8i32_inreg@rel32@hi+12
+; GFX9-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX9-NEXT:    s_swappc_b64 s[30:31], s[12:13]
+; GFX9-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX9-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX9-NEXT:    s_sub_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX9-NEXT:    s_or_saveexec_b64 s[6:7], -1
+; GFX9-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX9-NEXT:    s_mov_b64 exec, s[6:7]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    s_setpc_b64 s[4:5]
+;
+; GFX10-LABEL: test_call_external_void_func_v8i32_inreg:
+; GFX10:       ; %bb.0:
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    s_or_saveexec_b32 s4, -1
+; GFX10-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s4
+; GFX10-NEXT:    s_load_dwordx2 s[4:5], s[4:5], 0x0
+; GFX10-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX10-NEXT:    s_mov_b32 s33, s32
+; GFX10-NEXT:    s_add_u32 s32, s32, 0x200
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    s_getpc_b64 s[12:13]
+; GFX10-NEXT:    s_add_u32 s12, s12, external_void_func_v8i32_inreg@rel32@lo+4
+; GFX10-NEXT:    s_addc_u32 s13, s13, external_void_func_v8i32_inreg@rel32@hi+12
+; GFX10-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX10-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX10-NEXT:    s_load_dwordx8 s[4:11], s[4:5], 0x0
+; GFX10-NEXT:    s_swappc_b64 s[30:31], s[12:13]
+; GFX10-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX10-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX10-NEXT:    s_sub_u32 s32, s32, 0x200
+; GFX10-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX10-NEXT:    s_or_saveexec_b32 s6, -1
+; GFX10-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s6
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    s_setpc_b64 s[4:5]
+  %ptr = load <8 x i32> addrspace(4)*, <8 x i32> addrspace(4)* addrspace(4)* undef
+  %val = load <8 x i32>, <8 x i32> addrspace(4)* %ptr
+  call amdgpu_gfx_callable void @external_void_func_v8i32_inreg(<8 x i32> %val)
+  ret void
+}
+
+define amdgpu_gfx_callable void @test_call_external_void_func_v8i32_imm_inreg() #0 {
+; GFX9-LABEL: test_call_external_void_func_v8i32_imm_inreg:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    s_or_saveexec_b64 s[4:5], -1
+; GFX9-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX9-NEXT:    s_mov_b64 exec, s[4:5]
+; GFX9-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX9-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX9-NEXT:    s_mov_b32 s33, s32
+; GFX9-NEXT:    s_add_u32 s32, s32, 0x400
+; GFX9-NEXT:    s_mov_b32 s4, 1
+; GFX9-NEXT:    s_mov_b32 s5, 2
+; GFX9-NEXT:    s_mov_b32 s6, 3
+; GFX9-NEXT:    s_mov_b32 s7, 4
+; GFX9-NEXT:    s_mov_b32 s8, 5
+; GFX9-NEXT:    s_mov_b32 s9, 6
+; GFX9-NEXT:    s_mov_b32 s10, 7
+; GFX9-NEXT:    s_mov_b32 s11, 8
+; GFX9-NEXT:    s_getpc_b64 s[12:13]
+; GFX9-NEXT:    s_add_u32 s12, s12, external_void_func_v8i32_inreg@rel32@lo+4
+; GFX9-NEXT:    s_addc_u32 s13, s13, external_void_func_v8i32_inreg@rel32@hi+12
+; GFX9-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX9-NEXT:    s_swappc_b64 s[30:31], s[12:13]
+; GFX9-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX9-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX9-NEXT:    s_sub_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX9-NEXT:    s_or_saveexec_b64 s[6:7], -1
+; GFX9-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX9-NEXT:    s_mov_b64 exec, s[6:7]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    s_setpc_b64 s[4:5]
+;
+; GFX10-LABEL: test_call_external_void_func_v8i32_imm_inreg:
+; GFX10:       ; %bb.0:
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    s_or_saveexec_b32 s4, -1
+; GFX10-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s4
+; GFX10-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX10-NEXT:    s_mov_b32 s4, 1
+; GFX10-NEXT:    s_mov_b32 s5, 2
+; GFX10-NEXT:    s_mov_b32 s6, 3
+; GFX10-NEXT:    s_mov_b32 s7, 4
+; GFX10-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX10-NEXT:    s_mov_b32 s8, 5
+; GFX10-NEXT:    s_mov_b32 s9, 6
+; GFX10-NEXT:    s_mov_b32 s10, 7
+; GFX10-NEXT:    s_mov_b32 s11, 8
+; GFX10-NEXT:    s_mov_b32 s33, s32
+; GFX10-NEXT:    s_add_u32 s32, s32, 0x200
+; GFX10-NEXT:    s_getpc_b64 s[12:13]
+; GFX10-NEXT:    s_add_u32 s12, s12, external_void_func_v8i32_inreg@rel32@lo+4
+; GFX10-NEXT:    s_addc_u32 s13, s13, external_void_func_v8i32_inreg@rel32@hi+12
+; GFX10-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    s_swappc_b64 s[30:31], s[12:13]
+; GFX10-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX10-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX10-NEXT:    s_sub_u32 s32, s32, 0x200
+; GFX10-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX10-NEXT:    s_or_saveexec_b32 s6, -1
+; GFX10-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s6
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    s_setpc_b64 s[4:5]
+  call amdgpu_gfx_callable void @external_void_func_v8i32_inreg(<8 x i32> <i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8>)
+  ret void
+}
+
+define amdgpu_gfx_callable void @test_call_external_void_func_v16i32_inreg() #0 {
+; GFX9-LABEL: test_call_external_void_func_v16i32_inreg:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    s_or_saveexec_b64 s[4:5], -1
+; GFX9-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX9-NEXT:    s_mov_b64 exec, s[4:5]
+; GFX9-NEXT:    s_load_dwordx2 s[4:5], s[4:5], 0x0
+; GFX9-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX9-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX9-NEXT:    s_mov_b32 s33, s32
+; GFX9-NEXT:    s_add_u32 s32, s32, 0x400
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    s_load_dwordx16 s[4:19], s[4:5], 0x0
+; GFX9-NEXT:    s_getpc_b64 s[20:21]
+; GFX9-NEXT:    s_add_u32 s20, s20, external_void_func_v16i32_inreg@rel32@lo+4
+; GFX9-NEXT:    s_addc_u32 s21, s21, external_void_func_v16i32_inreg@rel32@hi+12
+; GFX9-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX9-NEXT:    s_swappc_b64 s[30:31], s[20:21]
+; GFX9-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX9-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX9-NEXT:    s_sub_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX9-NEXT:    s_or_saveexec_b64 s[6:7], -1
+; GFX9-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX9-NEXT:    s_mov_b64 exec, s[6:7]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    s_setpc_b64 s[4:5]
+;
+; GFX10-LABEL: test_call_external_void_func_v16i32_inreg:
+; GFX10:       ; %bb.0:
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    s_or_saveexec_b32 s4, -1
+; GFX10-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s4
+; GFX10-NEXT:    s_load_dwordx2 s[4:5], s[4:5], 0x0
+; GFX10-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX10-NEXT:    s_mov_b32 s33, s32
+; GFX10-NEXT:    s_add_u32 s32, s32, 0x200
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    s_getpc_b64 s[20:21]
+; GFX10-NEXT:    s_add_u32 s20, s20, external_void_func_v16i32_inreg@rel32@lo+4
+; GFX10-NEXT:    s_addc_u32 s21, s21, external_void_func_v16i32_inreg@rel32@hi+12
+; GFX10-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX10-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX10-NEXT:    s_load_dwordx16 s[4:19], s[4:5], 0x0
+; GFX10-NEXT:    s_swappc_b64 s[30:31], s[20:21]
+; GFX10-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX10-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX10-NEXT:    s_sub_u32 s32, s32, 0x200
+; GFX10-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX10-NEXT:    s_or_saveexec_b32 s6, -1
+; GFX10-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s6
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    s_setpc_b64 s[4:5]
+  %ptr = load <16 x i32> addrspace(4)*, <16 x i32> addrspace(4)* addrspace(4)* undef
+  %val = load <16 x i32>, <16 x i32> addrspace(4)* %ptr
+  call amdgpu_gfx_callable void @external_void_func_v16i32_inreg(<16 x i32> %val)
+  ret void
+}
+
+define amdgpu_gfx_callable void @test_call_external_void_func_v32i32_inreg() #0 {
+; GFX9-LABEL: test_call_external_void_func_v32i32_inreg:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    s_or_saveexec_b64 s[4:5], -1
+; GFX9-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX9-NEXT:    s_mov_b64 exec, s[4:5]
+; GFX9-NEXT:    v_writelane_b32 v40, s33, 21
+; GFX9-NEXT:    s_mov_b32 s33, s32
+; GFX9-NEXT:    v_writelane_b32 v40, s33, 0
+; GFX9-NEXT:    v_writelane_b32 v40, s34, 1
+; GFX9-NEXT:    v_writelane_b32 v40, s35, 2
+; GFX9-NEXT:    v_writelane_b32 v40, s36, 3
+; GFX9-NEXT:    v_writelane_b32 v40, s37, 4
+; GFX9-NEXT:    v_writelane_b32 v40, s38, 5
+; GFX9-NEXT:    v_writelane_b32 v40, s39, 6
+; GFX9-NEXT:    v_writelane_b32 v40, s40, 7
+; GFX9-NEXT:    v_writelane_b32 v40, s41, 8
+; GFX9-NEXT:    v_writelane_b32 v40, s42, 9
+; GFX9-NEXT:    v_writelane_b32 v40, s43, 10
+; GFX9-NEXT:    v_writelane_b32 v40, s44, 11
+; GFX9-NEXT:    v_writelane_b32 v40, s45, 12
+; GFX9-NEXT:    v_writelane_b32 v40, s46, 13
+; GFX9-NEXT:    v_writelane_b32 v40, s47, 14
+; GFX9-NEXT:    v_writelane_b32 v40, s48, 15
+; GFX9-NEXT:    s_load_dwordx2 s[20:21], s[4:5], 0x0
+; GFX9-NEXT:    v_writelane_b32 v40, s49, 16
+; GFX9-NEXT:    v_writelane_b32 v40, s50, 17
+; GFX9-NEXT:    v_writelane_b32 v40, s51, 18
+; GFX9-NEXT:    v_writelane_b32 v40, s30, 19
+; GFX9-NEXT:    v_writelane_b32 v40, s31, 20
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    s_load_dwordx16 s[4:19], s[20:21], 0x0
+; GFX9-NEXT:    s_load_dwordx16 s[36:51], s[20:21], 0x40
+; GFX9-NEXT:    s_add_u32 s32, s32, 0x400
+; GFX9-NEXT:    s_getpc_b64 s[30:31]
+; GFX9-NEXT:    s_add_u32 s30, s30, external_void_func_v32i32_inreg@rel32@lo+4
+; GFX9-NEXT:    s_addc_u32 s31, s31, external_void_func_v32i32_inreg@rel32@hi+12
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    s_mov_b32 s20, s36
+; GFX9-NEXT:    s_mov_b32 s21, s37
+; GFX9-NEXT:    s_mov_b32 s22, s38
+; GFX9-NEXT:    s_mov_b32 s33, s46
+; GFX9-NEXT:    s_mov_b32 s23, s39
+; GFX9-NEXT:    s_mov_b32 s24, s40
+; GFX9-NEXT:    s_mov_b32 s25, s41
+; GFX9-NEXT:    s_mov_b32 s26, s42
+; GFX9-NEXT:    s_mov_b32 s27, s43
+; GFX9-NEXT:    s_mov_b32 s28, s44
+; GFX9-NEXT:    s_mov_b32 s29, s45
+; GFX9-NEXT:    s_mov_b32 s34, s47
+; GFX9-NEXT:    s_mov_b32 s35, s48
+; GFX9-NEXT:    s_mov_b32 s36, s49
+; GFX9-NEXT:    s_mov_b32 s37, s50
+; GFX9-NEXT:    s_mov_b32 s38, s51
+; GFX9-NEXT:    s_swappc_b64 s[30:31], s[30:31]
+; GFX9-NEXT:    v_readlane_b32 s4, v40, 19
+; GFX9-NEXT:    v_readlane_b32 s5, v40, 20
+; GFX9-NEXT:    v_readlane_b32 s51, v40, 18
+; GFX9-NEXT:    v_readlane_b32 s50, v40, 17
+; GFX9-NEXT:    v_readlane_b32 s49, v40, 16
+; GFX9-NEXT:    v_readlane_b32 s48, v40, 15
+; GFX9-NEXT:    v_readlane_b32 s47, v40, 14
+; GFX9-NEXT:    v_readlane_b32 s46, v40, 13
+; GFX9-NEXT:    v_readlane_b32 s45, v40, 12
+; GFX9-NEXT:    v_readlane_b32 s44, v40, 11
+; GFX9-NEXT:    v_readlane_b32 s43, v40, 10
+; GFX9-NEXT:    v_readlane_b32 s42, v40, 9
+; GFX9-NEXT:    v_readlane_b32 s41, v40, 8
+; GFX9-NEXT:    v_readlane_b32 s40, v40, 7
+; GFX9-NEXT:    v_readlane_b32 s39, v40, 6
+; GFX9-NEXT:    v_readlane_b32 s38, v40, 5
+; GFX9-NEXT:    v_readlane_b32 s37, v40, 4
+; GFX9-NEXT:    v_readlane_b32 s36, v40, 3
+; GFX9-NEXT:    v_readlane_b32 s35, v40, 2
+; GFX9-NEXT:    v_readlane_b32 s34, v40, 1
+; GFX9-NEXT:    v_readlane_b32 s33, v40, 0
+; GFX9-NEXT:    s_sub_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_readlane_b32 s33, v40, 21
+; GFX9-NEXT:    s_or_saveexec_b64 s[6:7], -1
+; GFX9-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX9-NEXT:    s_mov_b64 exec, s[6:7]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    s_setpc_b64 s[4:5]
+;
+; GFX10-LABEL: test_call_external_void_func_v32i32_inreg:
+; GFX10:       ; %bb.0:
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    s_or_saveexec_b32 s4, -1
+; GFX10-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s4
+; GFX10-NEXT:    v_writelane_b32 v40, s33, 21
+; GFX10-NEXT:    s_mov_b32 s33, s32
+; GFX10-NEXT:    s_load_dwordx2 s[4:5], s[4:5], 0x0
+; GFX10-NEXT:    s_add_u32 s32, s32, 0x200
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    v_writelane_b32 v40, s33, 0
+; GFX10-NEXT:    v_writelane_b32 v40, s34, 1
+; GFX10-NEXT:    v_writelane_b32 v40, s35, 2
+; GFX10-NEXT:    v_writelane_b32 v40, s36, 3
+; GFX10-NEXT:    v_writelane_b32 v40, s37, 4
+; GFX10-NEXT:    v_writelane_b32 v40, s38, 5
+; GFX10-NEXT:    v_writelane_b32 v40, s39, 6
+; GFX10-NEXT:    v_writelane_b32 v40, s40, 7
+; GFX10-NEXT:    v_writelane_b32 v40, s41, 8
+; GFX10-NEXT:    v_writelane_b32 v40, s42, 9
+; GFX10-NEXT:    v_writelane_b32 v40, s43, 10
+; GFX10-NEXT:    v_writelane_b32 v40, s44, 11
+; GFX10-NEXT:    v_writelane_b32 v40, s45, 12
+; GFX10-NEXT:    v_writelane_b32 v40, s46, 13
+; GFX10-NEXT:    v_writelane_b32 v40, s47, 14
+; GFX10-NEXT:    v_writelane_b32 v40, s48, 15
+; GFX10-NEXT:    v_writelane_b32 v40, s49, 16
+; GFX10-NEXT:    v_writelane_b32 v40, s50, 17
+; GFX10-NEXT:    v_writelane_b32 v40, s51, 18
+; GFX10-NEXT:    v_writelane_b32 v40, s30, 19
+; GFX10-NEXT:    v_writelane_b32 v40, s31, 20
+; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX10-NEXT:    s_clause 0x1
+; GFX10-NEXT:    s_load_dwordx16 s[36:51], s[4:5], 0x40
+; GFX10-NEXT:    s_load_dwordx16 s[4:19], s[4:5], 0x0
+; GFX10-NEXT:    s_getpc_b64 s[30:31]
+; GFX10-NEXT:    s_add_u32 s30, s30, external_void_func_v32i32_inreg@rel32@lo+4
+; GFX10-NEXT:    s_addc_u32 s31, s31, external_void_func_v32i32_inreg@rel32@hi+12
+; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX10-NEXT:    s_mov_b32 s20, s36
+; GFX10-NEXT:    s_mov_b32 s21, s37
+; GFX10-NEXT:    s_mov_b32 s22, s38
+; GFX10-NEXT:    s_mov_b32 s33, s46
+; GFX10-NEXT:    s_mov_b32 s23, s39
+; GFX10-NEXT:    s_mov_b32 s24, s40
+; GFX10-NEXT:    s_mov_b32 s25, s41
+; GFX10-NEXT:    s_mov_b32 s26, s42
+; GFX10-NEXT:    s_mov_b32 s27, s43
+; GFX10-NEXT:    s_mov_b32 s28, s44
+; GFX10-NEXT:    s_mov_b32 s29, s45
+; GFX10-NEXT:    s_mov_b32 s34, s47
+; GFX10-NEXT:    s_mov_b32 s35, s48
+; GFX10-NEXT:    s_mov_b32 s36, s49
+; GFX10-NEXT:    s_mov_b32 s37, s50
+; GFX10-NEXT:    s_mov_b32 s38, s51
+; GFX10-NEXT:    s_swappc_b64 s[30:31], s[30:31]
+; GFX10-NEXT:    v_readlane_b32 s4, v40, 19
+; GFX10-NEXT:    v_readlane_b32 s5, v40, 20
+; GFX10-NEXT:    v_readlane_b32 s51, v40, 18
+; GFX10-NEXT:    v_readlane_b32 s50, v40, 17
+; GFX10-NEXT:    v_readlane_b32 s49, v40, 16
+; GFX10-NEXT:    v_readlane_b32 s48, v40, 15
+; GFX10-NEXT:    v_readlane_b32 s47, v40, 14
+; GFX10-NEXT:    v_readlane_b32 s46, v40, 13
+; GFX10-NEXT:    v_readlane_b32 s45, v40, 12
+; GFX10-NEXT:    v_readlane_b32 s44, v40, 11
+; GFX10-NEXT:    v_readlane_b32 s43, v40, 10
+; GFX10-NEXT:    v_readlane_b32 s42, v40, 9
+; GFX10-NEXT:    v_readlane_b32 s41, v40, 8
+; GFX10-NEXT:    v_readlane_b32 s40, v40, 7
+; GFX10-NEXT:    v_readlane_b32 s39, v40, 6
+; GFX10-NEXT:    v_readlane_b32 s38, v40, 5
+; GFX10-NEXT:    v_readlane_b32 s37, v40, 4
+; GFX10-NEXT:    v_readlane_b32 s36, v40, 3
+; GFX10-NEXT:    v_readlane_b32 s35, v40, 2
+; GFX10-NEXT:    v_readlane_b32 s34, v40, 1
+; GFX10-NEXT:    v_readlane_b32 s33, v40, 0
+; GFX10-NEXT:    s_sub_u32 s32, s32, 0x200
+; GFX10-NEXT:    v_readlane_b32 s33, v40, 21
+; GFX10-NEXT:    s_or_saveexec_b32 s6, -1
+; GFX10-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s6
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    s_setpc_b64 s[4:5]
+  %ptr = load <32 x i32> addrspace(4)*, <32 x i32> addrspace(4)* addrspace(4)* undef
+  %val = load <32 x i32>, <32 x i32> addrspace(4)* %ptr
+  call amdgpu_gfx_callable void @external_void_func_v32i32_inreg(<32 x i32> %val)
+  ret void
+}
+
+define amdgpu_gfx_callable void @test_call_external_void_func_v32i32_i32_inreg(i32) #0 {
+; GFX9-LABEL: test_call_external_void_func_v32i32_i32_inreg:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    s_or_saveexec_b64 s[4:5], -1
+; GFX9-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX9-NEXT:    s_mov_b64 exec, s[4:5]
+; GFX9-NEXT:    v_writelane_b32 v40, s33, 25
+; GFX9-NEXT:    s_mov_b32 s33, s32
+; GFX9-NEXT:    v_writelane_b32 v40, s33, 0
+; GFX9-NEXT:    v_writelane_b32 v40, s34, 1
+; GFX9-NEXT:    v_writelane_b32 v40, s35, 2
+; GFX9-NEXT:    v_writelane_b32 v40, s36, 3
+; GFX9-NEXT:    v_writelane_b32 v40, s37, 4
+; GFX9-NEXT:    v_writelane_b32 v40, s38, 5
+; GFX9-NEXT:    v_writelane_b32 v40, s39, 6
+; GFX9-NEXT:    v_writelane_b32 v40, s40, 7
+; GFX9-NEXT:    v_writelane_b32 v40, s41, 8
+; GFX9-NEXT:    v_writelane_b32 v40, s42, 9
+; GFX9-NEXT:    v_writelane_b32 v40, s43, 10
+; GFX9-NEXT:    v_writelane_b32 v40, s44, 11
+; GFX9-NEXT:    v_writelane_b32 v40, s45, 12
+; GFX9-NEXT:    v_writelane_b32 v40, s46, 13
+; GFX9-NEXT:    v_writelane_b32 v40, s47, 14
+; GFX9-NEXT:    v_writelane_b32 v40, s48, 15
+; GFX9-NEXT:    v_writelane_b32 v40, s49, 16
+; GFX9-NEXT:    v_writelane_b32 v40, s50, 17
+; GFX9-NEXT:    v_writelane_b32 v40, s51, 18
+; GFX9-NEXT:    v_writelane_b32 v40, s52, 19
+; GFX9-NEXT:    v_writelane_b32 v40, s53, 20
+; GFX9-NEXT:    v_writelane_b32 v40, s54, 21
+; GFX9-NEXT:    v_writelane_b32 v40, s55, 22
+; GFX9-NEXT:    v_writelane_b32 v40, s30, 23
+; GFX9-NEXT:    v_writelane_b32 v40, s31, 24
+; GFX9-NEXT:    s_load_dwordx2 s[20:21], s[4:5], 0x0
+; GFX9-NEXT:    s_load_dword s39, s[4:5], 0x0
+; GFX9-NEXT:    s_add_u32 s32, s32, 0x400
+; GFX9-NEXT:    s_getpc_b64 s[30:31]
+; GFX9-NEXT:    s_add_u32 s30, s30, external_void_func_v32i32_i32_inreg@rel32@lo+4
+; GFX9-NEXT:    s_addc_u32 s31, s31, external_void_func_v32i32_i32_inreg@rel32@hi+12
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    s_load_dwordx16 s[4:19], s[20:21], 0x0
+; GFX9-NEXT:    s_load_dwordx16 s[40:55], s[20:21], 0x40
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    s_mov_b32 s33, s50
+; GFX9-NEXT:    s_mov_b32 s20, s40
+; GFX9-NEXT:    s_mov_b32 s21, s41
+; GFX9-NEXT:    s_mov_b32 s22, s42
+; GFX9-NEXT:    s_mov_b32 s23, s43
+; GFX9-NEXT:    s_mov_b32 s24, s44
+; GFX9-NEXT:    s_mov_b32 s25, s45
+; GFX9-NEXT:    s_mov_b32 s26, s46
+; GFX9-NEXT:    s_mov_b32 s27, s47
+; GFX9-NEXT:    s_mov_b32 s28, s48
+; GFX9-NEXT:    s_mov_b32 s29, s49
+; GFX9-NEXT:    s_mov_b32 s34, s51
+; GFX9-NEXT:    s_mov_b32 s35, s52
+; GFX9-NEXT:    s_mov_b32 s36, s53
+; GFX9-NEXT:    s_mov_b32 s37, s54
+; GFX9-NEXT:    s_mov_b32 s38, s55
+; GFX9-NEXT:    s_swappc_b64 s[30:31], s[30:31]
+; GFX9-NEXT:    v_readlane_b32 s4, v40, 23
+; GFX9-NEXT:    v_readlane_b32 s5, v40, 24
+; GFX9-NEXT:    v_readlane_b32 s55, v40, 22
+; GFX9-NEXT:    v_readlane_b32 s54, v40, 21
+; GFX9-NEXT:    v_readlane_b32 s53, v40, 20
+; GFX9-NEXT:    v_readlane_b32 s52, v40, 19
+; GFX9-NEXT:    v_readlane_b32 s51, v40, 18
+; GFX9-NEXT:    v_readlane_b32 s50, v40, 17
+; GFX9-NEXT:    v_readlane_b32 s49, v40, 16
+; GFX9-NEXT:    v_readlane_b32 s48, v40, 15
+; GFX9-NEXT:    v_readlane_b32 s47, v40, 14
+; GFX9-NEXT:    v_readlane_b32 s46, v40, 13
+; GFX9-NEXT:    v_readlane_b32 s45, v40, 12
+; GFX9-NEXT:    v_readlane_b32 s44, v40, 11
+; GFX9-NEXT:    v_readlane_b32 s43, v40, 10
+; GFX9-NEXT:    v_readlane_b32 s42, v40, 9
+; GFX9-NEXT:    v_readlane_b32 s41, v40, 8
+; GFX9-NEXT:    v_readlane_b32 s40, v40, 7
+; GFX9-NEXT:    v_readlane_b32 s39, v40, 6
+; GFX9-NEXT:    v_readlane_b32 s38, v40, 5
+; GFX9-NEXT:    v_readlane_b32 s37, v40, 4
+; GFX9-NEXT:    v_readlane_b32 s36, v40, 3
+; GFX9-NEXT:    v_readlane_b32 s35, v40, 2
+; GFX9-NEXT:    v_readlane_b32 s34, v40, 1
+; GFX9-NEXT:    v_readlane_b32 s33, v40, 0
+; GFX9-NEXT:    s_sub_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_readlane_b32 s33, v40, 25
+; GFX9-NEXT:    s_or_saveexec_b64 s[6:7], -1
+; GFX9-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX9-NEXT:    s_mov_b64 exec, s[6:7]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    s_setpc_b64 s[4:5]
+;
+; GFX10-LABEL: test_call_external_void_func_v32i32_i32_inreg:
+; GFX10:       ; %bb.0:
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    s_or_saveexec_b32 s4, -1
+; GFX10-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s4
+; GFX10-NEXT:    v_writelane_b32 v40, s33, 25
+; GFX10-NEXT:    s_mov_b32 s33, s32
+; GFX10-NEXT:    s_load_dwordx2 s[4:5], s[4:5], 0x0
+; GFX10-NEXT:    s_add_u32 s32, s32, 0x200
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    v_writelane_b32 v40, s33, 0
+; GFX10-NEXT:    v_writelane_b32 v40, s34, 1
+; GFX10-NEXT:    v_writelane_b32 v40, s35, 2
+; GFX10-NEXT:    v_writelane_b32 v40, s36, 3
+; GFX10-NEXT:    v_writelane_b32 v40, s37, 4
+; GFX10-NEXT:    v_writelane_b32 v40, s38, 5
+; GFX10-NEXT:    v_writelane_b32 v40, s39, 6
+; GFX10-NEXT:    v_writelane_b32 v40, s40, 7
+; GFX10-NEXT:    v_writelane_b32 v40, s41, 8
+; GFX10-NEXT:    v_writelane_b32 v40, s42, 9
+; GFX10-NEXT:    v_writelane_b32 v40, s43, 10
+; GFX10-NEXT:    v_writelane_b32 v40, s44, 11
+; GFX10-NEXT:    v_writelane_b32 v40, s45, 12
+; GFX10-NEXT:    v_writelane_b32 v40, s46, 13
+; GFX10-NEXT:    v_writelane_b32 v40, s47, 14
+; GFX10-NEXT:    v_writelane_b32 v40, s48, 15
+; GFX10-NEXT:    v_writelane_b32 v40, s49, 16
+; GFX10-NEXT:    v_writelane_b32 v40, s50, 17
+; GFX10-NEXT:    v_writelane_b32 v40, s51, 18
+; GFX10-NEXT:    v_writelane_b32 v40, s52, 19
+; GFX10-NEXT:    v_writelane_b32 v40, s53, 20
+; GFX10-NEXT:    v_writelane_b32 v40, s54, 21
+; GFX10-NEXT:    v_writelane_b32 v40, s55, 22
+; GFX10-NEXT:    v_writelane_b32 v40, s30, 23
+; GFX10-NEXT:    v_writelane_b32 v40, s31, 24
+; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX10-NEXT:    s_clause 0x1
+; GFX10-NEXT:    s_load_dwordx16 s[40:55], s[4:5], 0x40
+; GFX10-NEXT:    s_load_dwordx16 s[4:19], s[4:5], 0x0
+; GFX10-NEXT:    s_getpc_b64 s[30:31]
+; GFX10-NEXT:    s_add_u32 s30, s30, external_void_func_v32i32_i32_inreg@rel32@lo+4
+; GFX10-NEXT:    s_addc_u32 s31, s31, external_void_func_v32i32_i32_inreg@rel32@hi+12
+; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX10-NEXT:    s_mov_b32 s33, s50
+; GFX10-NEXT:    s_load_dword s39, s[4:5], 0x0
+; GFX10-NEXT:    s_mov_b32 s20, s40
+; GFX10-NEXT:    s_mov_b32 s21, s41
+; GFX10-NEXT:    s_mov_b32 s22, s42
+; GFX10-NEXT:    s_mov_b32 s23, s43
+; GFX10-NEXT:    s_mov_b32 s24, s44
+; GFX10-NEXT:    s_mov_b32 s25, s45
+; GFX10-NEXT:    s_mov_b32 s26, s46
+; GFX10-NEXT:    s_mov_b32 s27, s47
+; GFX10-NEXT:    s_mov_b32 s28, s48
+; GFX10-NEXT:    s_mov_b32 s29, s49
+; GFX10-NEXT:    s_mov_b32 s34, s51
+; GFX10-NEXT:    s_mov_b32 s35, s52
+; GFX10-NEXT:    s_mov_b32 s36, s53
+; GFX10-NEXT:    s_mov_b32 s37, s54
+; GFX10-NEXT:    s_mov_b32 s38, s55
+; GFX10-NEXT:    s_swappc_b64 s[30:31], s[30:31]
+; GFX10-NEXT:    v_readlane_b32 s4, v40, 23
+; GFX10-NEXT:    v_readlane_b32 s5, v40, 24
+; GFX10-NEXT:    v_readlane_b32 s55, v40, 22
+; GFX10-NEXT:    v_readlane_b32 s54, v40, 21
+; GFX10-NEXT:    v_readlane_b32 s53, v40, 20
+; GFX10-NEXT:    v_readlane_b32 s52, v40, 19
+; GFX10-NEXT:    v_readlane_b32 s51, v40, 18
+; GFX10-NEXT:    v_readlane_b32 s50, v40, 17
+; GFX10-NEXT:    v_readlane_b32 s49, v40, 16
+; GFX10-NEXT:    v_readlane_b32 s48, v40, 15
+; GFX10-NEXT:    v_readlane_b32 s47, v40, 14
+; GFX10-NEXT:    v_readlane_b32 s46, v40, 13
+; GFX10-NEXT:    v_readlane_b32 s45, v40, 12
+; GFX10-NEXT:    v_readlane_b32 s44, v40, 11
+; GFX10-NEXT:    v_readlane_b32 s43, v40, 10
+; GFX10-NEXT:    v_readlane_b32 s42, v40, 9
+; GFX10-NEXT:    v_readlane_b32 s41, v40, 8
+; GFX10-NEXT:    v_readlane_b32 s40, v40, 7
+; GFX10-NEXT:    v_readlane_b32 s39, v40, 6
+; GFX10-NEXT:    v_readlane_b32 s38, v40, 5
+; GFX10-NEXT:    v_readlane_b32 s37, v40, 4
+; GFX10-NEXT:    v_readlane_b32 s36, v40, 3
+; GFX10-NEXT:    v_readlane_b32 s35, v40, 2
+; GFX10-NEXT:    v_readlane_b32 s34, v40, 1
+; GFX10-NEXT:    v_readlane_b32 s33, v40, 0
+; GFX10-NEXT:    s_sub_u32 s32, s32, 0x200
+; GFX10-NEXT:    v_readlane_b32 s33, v40, 25
+; GFX10-NEXT:    s_or_saveexec_b32 s6, -1
+; GFX10-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s6
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    s_setpc_b64 s[4:5]
+  %ptr0 = load <32 x i32> addrspace(4)*, <32 x i32> addrspace(4)* addrspace(4)* undef
+  %val0 = load <32 x i32>, <32 x i32> addrspace(4)* %ptr0
+  %val1 = load i32, i32 addrspace(4)* undef
+  call amdgpu_gfx_callable void @external_void_func_v32i32_i32_inreg(<32 x i32> %val0, i32 %val1)
+  ret void
+}
+
+
+declare hidden amdgpu_gfx_callable void @byval_align16_f64_arg(<32 x i32>, double addrspace(5)* byval align 16) #0
+declare hidden amdgpu_gfx_callable void @external_void_func_12xv3i32(<3 x i32>, <3 x i32>, <3 x i32>, <3 x i32>,
+    <3 x i32>, <3 x i32>, <3 x i32>, <3 x i32>, <3 x i32>, <3 x i32>, <3 x i32>, <3 x i32>) #0
+declare hidden amdgpu_gfx_callable void @external_void_func_28xv5i32(<5 x i32>, <5 x i32>, <5 x i32>, <5 x i32>,
+    <5 x i32>, <5 x i32>, <5 x i32>, <5 x i32>, <5 x i32>, <5 x i32>, <5 x i32>, <5 x i32>, <5 x i32>, <5 x i32>,
+    <5 x i32>, <5 x i32>, <5 x i32>, <5 x i32>, <5 x i32>, <5 x i32>, <5 x i32>, <5 x i32>, <5 x i32>, <5 x i32>,
+    <5 x i32>, <5 x i32>, <5 x i32>, <5 x i32>) #0
+declare hidden amdgpu_gfx_callable void @external_void_func_12xv3f32(<3 x float>, <3 x float>, <3 x float>, <3 x float>,
+    <3 x float>, <3 x float>, <3 x float>, <3 x float>, <3 x float>, <3 x float>, <3 x float>, <3 x float>) #0
+declare hidden amdgpu_gfx_callable void @external_void_func_8xv5f32(<5 x float>, <5 x float>, <5 x float>, <5 x float>,
+    <5 x float>, <5 x float>, <5 x float>, <5 x float>) #0
+attributes #0 = { nounwind }
+attributes #1 = { nounwind noinline }

llvm/test/CodeGen/AMDGPU/gfx-callable-preserved-registers.ll

+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc -mtriple=amdgcn--amdpal -mcpu=gfx900 -enable-ipra=0 -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN,GFX9 %s
+; RUN: llc -mtriple=amdgcn--amdpal -mcpu=gfx1010 -enable-ipra=0 -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN,GFX10 %s
+
+declare hidden amdgpu_gfx_callable void @external_void_func_void() #0
+
+define amdgpu_gfx_callable void @test_call_external_void_func_void_clobber_s30_s31_call_external_void_func_void() #0 {
+; GFX9-LABEL: test_call_external_void_func_void_clobber_s30_s31_call_external_void_func_void:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    s_or_saveexec_b64 s[4:5], -1
+; GFX9-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX9-NEXT:    s_mov_b64 exec, s[4:5]
+; GFX9-NEXT:    v_writelane_b32 v40, s33, 4
+; GFX9-NEXT:    v_writelane_b32 v40, s34, 0
+; GFX9-NEXT:    v_writelane_b32 v40, s35, 1
+; GFX9-NEXT:    v_writelane_b32 v40, s30, 2
+; GFX9-NEXT:    s_mov_b32 s33, s32
+; GFX9-NEXT:    s_add_u32 s32, s32, 0x400
+; GFX9-NEXT:    s_getpc_b64 s[34:35]
+; GFX9-NEXT:    s_add_u32 s34, s34, external_void_func_void@rel32@lo+4
+; GFX9-NEXT:    s_addc_u32 s35, s35, external_void_func_void@rel32@hi+12
+; GFX9-NEXT:    v_writelane_b32 v40, s31, 3
+; GFX9-NEXT:    s_swappc_b64 s[30:31], s[34:35]
+; GFX9-NEXT:    ;;#ASMSTART
+; GFX9-NEXT:    ;;#ASMEND
+; GFX9-NEXT:    s_swappc_b64 s[30:31], s[34:35]
+; GFX9-NEXT:    v_readlane_b32 s4, v40, 2
+; GFX9-NEXT:    v_readlane_b32 s5, v40, 3
+; GFX9-NEXT:    v_readlane_b32 s35, v40, 1
+; GFX9-NEXT:    v_readlane_b32 s34, v40, 0
+; GFX9-NEXT:    s_sub_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_readlane_b32 s33, v40, 4
+; GFX9-NEXT:    s_or_saveexec_b64 s[6:7], -1
+; GFX9-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX9-NEXT:    s_mov_b64 exec, s[6:7]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    s_setpc_b64 s[4:5]
+;
+; GFX10-LABEL: test_call_external_void_func_void_clobber_s30_s31_call_external_void_func_void:
+; GFX10:       ; %bb.0:
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    s_or_saveexec_b32 s4, -1
+; GFX10-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s4
+; GFX10-NEXT:    v_writelane_b32 v40, s33, 4
+; GFX10-NEXT:    s_mov_b32 s33, s32
+; GFX10-NEXT:    s_add_u32 s32, s32, 0x200
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    v_writelane_b32 v40, s34, 0
+; GFX10-NEXT:    v_writelane_b32 v40, s35, 1
+; GFX10-NEXT:    s_getpc_b64 s[34:35]
+; GFX10-NEXT:    s_add_u32 s34, s34, external_void_func_void@rel32@lo+4
+; GFX10-NEXT:    s_addc_u32 s35, s35, external_void_func_void@rel32@hi+12
+; GFX10-NEXT:    v_writelane_b32 v40, s30, 2
+; GFX10-NEXT:    v_writelane_b32 v40, s31, 3
+; GFX10-NEXT:    s_swappc_b64 s[30:31], s[34:35]
+; GFX10-NEXT:    ;;#ASMSTART
+; GFX10-NEXT:    ;;#ASMEND
+; GFX10-NEXT:    s_swappc_b64 s[30:31], s[34:35]
+; GFX10-NEXT:    v_readlane_b32 s4, v40, 2
+; GFX10-NEXT:    v_readlane_b32 s5, v40, 3
+; GFX10-NEXT:    v_readlane_b32 s35, v40, 1
+; GFX10-NEXT:    v_readlane_b32 s34, v40, 0
+; GFX10-NEXT:    s_sub_u32 s32, s32, 0x200
+; GFX10-NEXT:    v_readlane_b32 s33, v40, 4
+; GFX10-NEXT:    s_or_saveexec_b32 s6, -1
+; GFX10-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s6
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    s_setpc_b64 s[4:5]
+  call amdgpu_gfx_callable void @external_void_func_void()
+  call void asm sideeffect "", ""() #0
+  call amdgpu_gfx_callable void @external_void_func_void()
+  ret void
+}
+
+define amdgpu_gfx_callable void @void_func_void_clobber_s30_s31() #1 {
+; GFX9-LABEL: void_func_void_clobber_s30_s31:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    s_mov_b64 s[4:5], s[30:31]
+; GFX9-NEXT:    ;;#ASMSTART
+; GFX9-NEXT:    ; clobber
+; GFX9-NEXT:    ;;#ASMEND
+; GFX9-NEXT:    s_setpc_b64 s[4:5]
+;
+; GFX10-LABEL: void_func_void_clobber_s30_s31:
+; GFX10:       ; %bb.0:
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    s_mov_b64 s[4:5], s[30:31]
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    ;;#ASMSTART
+; GFX10-NEXT:    ; clobber
+; GFX10-NEXT:    ;;#ASMEND
+; GFX10-NEXT:    s_setpc_b64 s[4:5]
+  call void asm sideeffect "; clobber", "~{s[30:31]}"() #0
+  ret void
+}
+
+define amdgpu_gfx_callable void @test_call_void_func_void_mayclobber_s31(i32 addrspace(1)* %out) #0 {
+; GFX9-LABEL: test_call_void_func_void_mayclobber_s31:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    s_or_saveexec_b64 s[4:5], -1
+; GFX9-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX9-NEXT:    s_mov_b64 exec, s[4:5]
+; GFX9-NEXT:    v_writelane_b32 v40, s33, 3
+; GFX9-NEXT:    v_writelane_b32 v40, s34, 0
+; GFX9-NEXT:    v_writelane_b32 v40, s30, 1
+; GFX9-NEXT:    s_mov_b32 s33, s32
+; GFX9-NEXT:    s_add_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_writelane_b32 v40, s31, 2
+; GFX9-NEXT:    ;;#ASMSTART
+; GFX9-NEXT:    ; def s31
+; GFX9-NEXT:    ;;#ASMEND
+; GFX9-NEXT:    s_getpc_b64 s[4:5]
+; GFX9-NEXT:    s_add_u32 s4, s4, external_void_func_void@rel32@lo+4
+; GFX9-NEXT:    s_addc_u32 s5, s5, external_void_func_void@rel32@hi+12
+; GFX9-NEXT:    s_mov_b32 s34, s31
+; GFX9-NEXT:    s_swappc_b64 s[30:31], s[4:5]
+; GFX9-NEXT:    s_mov_b32 s31, s34
+; GFX9-NEXT:    ;;#ASMSTART
+; GFX9-NEXT:    ; use s31
+; GFX9-NEXT:    ;;#ASMEND
+; GFX9-NEXT:    v_readlane_b32 s4, v40, 1
+; GFX9-NEXT:    v_readlane_b32 s5, v40, 2
+; GFX9-NEXT:    v_readlane_b32 s34, v40, 0
+; GFX9-NEXT:    s_sub_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_readlane_b32 s33, v40, 3
+; GFX9-NEXT:    s_or_saveexec_b64 s[6:7], -1
+; GFX9-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX9-NEXT:    s_mov_b64 exec, s[6:7]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    s_setpc_b64 s[4:5]
+;
+; GFX10-LABEL: test_call_void_func_void_mayclobber_s31:
+; GFX10:       ; %bb.0:
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    s_or_saveexec_b32 s4, -1
+; GFX10-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s4
+; GFX10-NEXT:    v_writelane_b32 v40, s33, 3
+; GFX10-NEXT:    s_mov_b32 s33, s32
+; GFX10-NEXT:    s_add_u32 s32, s32, 0x200
+; GFX10-NEXT:    s_getpc_b64 s[4:5]
+; GFX10-NEXT:    s_add_u32 s4, s4, external_void_func_void@rel32@lo+4
+; GFX10-NEXT:    s_addc_u32 s5, s5, external_void_func_void@rel32@hi+12
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    v_writelane_b32 v40, s34, 0
+; GFX10-NEXT:    v_writelane_b32 v40, s30, 1
+; GFX10-NEXT:    v_writelane_b32 v40, s31, 2
+; GFX10-NEXT:    ;;#ASMSTART
+; GFX10-NEXT:    ; def s31
+; GFX10-NEXT:    ;;#ASMEND
+; GFX10-NEXT:    s_mov_b32 s34, s31
+; GFX10-NEXT:    s_swappc_b64 s[30:31], s[4:5]
+; GFX10-NEXT:    s_mov_b32 s31, s34
+; GFX10-NEXT:    ;;#ASMSTART
+; GFX10-NEXT:    ; use s31
+; GFX10-NEXT:    ;;#ASMEND
+; GFX10-NEXT:    v_readlane_b32 s4, v40, 1
+; GFX10-NEXT:    v_readlane_b32 s5, v40, 2
+; GFX10-NEXT:    v_readlane_b32 s34, v40, 0
+; GFX10-NEXT:    s_sub_u32 s32, s32, 0x200
+; GFX10-NEXT:    v_readlane_b32 s33, v40, 3
+; GFX10-NEXT:    s_or_saveexec_b32 s6, -1
+; GFX10-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s6
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    s_setpc_b64 s[4:5]
+  %s31 = call i32 asm sideeffect "; def $0", "={s31}"()
+  call amdgpu_gfx_callable void @external_void_func_void()
+  call void asm sideeffect "; use $0", "{s31}"(i32 %s31)
+  ret void
+}
+
+define amdgpu_gfx_callable void @test_call_void_func_void_mayclobber_v31(i32 addrspace(1)* %out) #0 {
+; GFX9-LABEL: test_call_void_func_void_mayclobber_v31:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    s_or_saveexec_b64 s[4:5], -1
+; GFX9-NEXT:    buffer_store_dword v41, off, s[0:3], s32 offset:4 ; 4-byte Folded Spill
+; GFX9-NEXT:    s_mov_b64 exec, s[4:5]
+; GFX9-NEXT:    v_writelane_b32 v41, s33, 2
+; GFX9-NEXT:    s_mov_b32 s33, s32
+; GFX9-NEXT:    s_add_u32 s32, s32, 0x400
+; GFX9-NEXT:    s_getpc_b64 s[4:5]
+; GFX9-NEXT:    s_add_u32 s4, s4, external_void_func_void@rel32@lo+4
+; GFX9-NEXT:    s_addc_u32 s5, s5, external_void_func_void@rel32@hi+12
+; GFX9-NEXT:    buffer_store_dword v40, off, s[0:3], s33 ; 4-byte Folded Spill
+; GFX9-NEXT:    v_writelane_b32 v41, s30, 0
+; GFX9-NEXT:    v_writelane_b32 v41, s31, 1
+; GFX9-NEXT:    ;;#ASMSTART
+; GFX9-NEXT:    ; def v31
+; GFX9-NEXT:    ;;#ASMEND
+; GFX9-NEXT:    v_mov_b32_e32 v40, v31
+; GFX9-NEXT:    s_swappc_b64 s[30:31], s[4:5]
+; GFX9-NEXT:    v_mov_b32_e32 v31, v40
+; GFX9-NEXT:    ;;#ASMSTART
+; GFX9-NEXT:    ; use v31
+; GFX9-NEXT:    ;;#ASMEND
+; GFX9-NEXT:    v_readlane_b32 s4, v41, 0
+; GFX9-NEXT:    v_readlane_b32 s5, v41, 1
+; GFX9-NEXT:    buffer_load_dword v40, off, s[0:3], s33 ; 4-byte Folded Reload
+; GFX9-NEXT:    s_sub_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_readlane_b32 s33, v41, 2
+; GFX9-NEXT:    s_or_saveexec_b64 s[6:7], -1
+; GFX9-NEXT:    buffer_load_dword v41, off, s[0:3], s32 offset:4 ; 4-byte Folded Reload
+; GFX9-NEXT:    s_mov_b64 exec, s[6:7]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    s_setpc_b64 s[4:5]
+;
+; GFX10-LABEL: test_call_void_func_void_mayclobber_v31:
+; GFX10:       ; %bb.0:
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    s_or_saveexec_b32 s4, -1
+; GFX10-NEXT:    buffer_store_dword v41, off, s[0:3], s32 offset:4 ; 4-byte Folded Spill
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s4
+; GFX10-NEXT:    v_writelane_b32 v41, s33, 2
+; GFX10-NEXT:    s_mov_b32 s33, s32
+; GFX10-NEXT:    s_add_u32 s32, s32, 0x200
+; GFX10-NEXT:    s_getpc_b64 s[4:5]
+; GFX10-NEXT:    s_add_u32 s4, s4, external_void_func_void@rel32@lo+4
+; GFX10-NEXT:    s_addc_u32 s5, s5, external_void_func_void@rel32@hi+12
+; GFX10-NEXT:    buffer_store_dword v40, off, s[0:3], s33 ; 4-byte Folded Spill
+; GFX10-NEXT:    v_writelane_b32 v41, s30, 0
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    v_writelane_b32 v41, s31, 1
+; GFX10-NEXT:    ;;#ASMSTART
+; GFX10-NEXT:    ; def v31
+; GFX10-NEXT:    ;;#ASMEND
+; GFX10-NEXT:    v_mov_b32_e32 v40, v31
+; GFX10-NEXT:    s_swappc_b64 s[30:31], s[4:5]
+; GFX10-NEXT:    v_mov_b32_e32 v31, v40
+; GFX10-NEXT:    ;;#ASMSTART
+; GFX10-NEXT:    ; use v31
+; GFX10-NEXT:    ;;#ASMEND
+; GFX10-NEXT:    v_readlane_b32 s4, v41, 0
+; GFX10-NEXT:    v_readlane_b32 s5, v41, 1
+; GFX10-NEXT:    buffer_load_dword v40, off, s[0:3], s33 ; 4-byte Folded Reload
+; GFX10-NEXT:    s_sub_u32 s32, s32, 0x200
+; GFX10-NEXT:    v_readlane_b32 s33, v41, 2
+; GFX10-NEXT:    s_or_saveexec_b32 s6, -1
+; GFX10-NEXT:    buffer_load_dword v41, off, s[0:3], s32 offset:4 ; 4-byte Folded Reload
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s6
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    s_setpc_b64 s[4:5]
+  %v31 = call i32 asm sideeffect "; def $0", "={v31}"()
+  call amdgpu_gfx_callable void @external_void_func_void()
+  call void asm sideeffect "; use $0", "{v31}"(i32 %v31)
+  ret void
+}
+
+
+define amdgpu_gfx_callable void @test_call_void_func_void_preserves_s33(i32 addrspace(1)* %out) #0 {
+; GFX9-LABEL: test_call_void_func_void_preserves_s33:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    s_or_saveexec_b64 s[4:5], -1
+; GFX9-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX9-NEXT:    s_mov_b64 exec, s[4:5]
+; GFX9-NEXT:    v_writelane_b32 v40, s33, 3
+; GFX9-NEXT:    s_mov_b32 s33, s32
+; GFX9-NEXT:    v_writelane_b32 v40, s33, 0
+; GFX9-NEXT:    v_writelane_b32 v40, s30, 1
+; GFX9-NEXT:    s_add_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_writelane_b32 v40, s31, 2
+; GFX9-NEXT:    s_getpc_b64 s[4:5]
+; GFX9-NEXT:    s_add_u32 s4, s4, external_void_func_void@rel32@lo+4
+; GFX9-NEXT:    s_addc_u32 s5, s5, external_void_func_void@rel32@hi+12
+; GFX9-NEXT:    ;;#ASMSTART
+; GFX9-NEXT:    ; def s33
+; GFX9-NEXT:    ;;#ASMEND
+; GFX9-NEXT:    s_swappc_b64 s[30:31], s[4:5]
+; GFX9-NEXT:    ;;#ASMSTART
+; GFX9-NEXT:    ; use s33
+; GFX9-NEXT:    ;;#ASMEND
+; GFX9-NEXT:    v_readlane_b32 s4, v40, 1
+; GFX9-NEXT:    v_readlane_b32 s5, v40, 2
+; GFX9-NEXT:    v_readlane_b32 s33, v40, 0
+; GFX9-NEXT:    s_sub_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_readlane_b32 s33, v40, 3
+; GFX9-NEXT:    s_or_saveexec_b64 s[6:7], -1
+; GFX9-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX9-NEXT:    s_mov_b64 exec, s[6:7]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    s_setpc_b64 s[4:5]
+;
+; GFX10-LABEL: test_call_void_func_void_preserves_s33:
+; GFX10:       ; %bb.0:
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    s_or_saveexec_b32 s4, -1
+; GFX10-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s4
+; GFX10-NEXT:    v_writelane_b32 v40, s33, 3
+; GFX10-NEXT:    s_mov_b32 s33, s32
+; GFX10-NEXT:    s_add_u32 s32, s32, 0x200
+; GFX10-NEXT:    s_getpc_b64 s[4:5]
+; GFX10-NEXT:    s_add_u32 s4, s4, external_void_func_void@rel32@lo+4
+; GFX10-NEXT:    s_addc_u32 s5, s5, external_void_func_void@rel32@hi+12
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    v_writelane_b32 v40, s33, 0
+; GFX10-NEXT:    v_writelane_b32 v40, s30, 1
+; GFX10-NEXT:    v_writelane_b32 v40, s31, 2
+; GFX10-NEXT:    ;;#ASMSTART
+; GFX10-NEXT:    ; def s33
+; GFX10-NEXT:    ;;#ASMEND
+; GFX10-NEXT:    s_swappc_b64 s[30:31], s[4:5]
+; GFX10-NEXT:    ;;#ASMSTART
+; GFX10-NEXT:    ; use s33
+; GFX10-NEXT:    ;;#ASMEND
+; GFX10-NEXT:    v_readlane_b32 s4, v40, 1
+; GFX10-NEXT:    v_readlane_b32 s5, v40, 2
+; GFX10-NEXT:    v_readlane_b32 s33, v40, 0
+; GFX10-NEXT:    s_sub_u32 s32, s32, 0x200
+; GFX10-NEXT:    v_readlane_b32 s33, v40, 3
+; GFX10-NEXT:    s_or_saveexec_b32 s6, -1
+; GFX10-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s6
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    s_setpc_b64 s[4:5]
+  %s33 = call i32 asm sideeffect "; def $0", "={s33}"()
+  call amdgpu_gfx_callable void @external_void_func_void()
+  call void asm sideeffect "; use $0", "{s33}"(i32 %s33)
+  ret void
+}
+
+define amdgpu_gfx_callable void @test_call_void_func_void_preserves_s34(i32 addrspace(1)* %out) #0 {
+; GFX9-LABEL: test_call_void_func_void_preserves_s34:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    s_or_saveexec_b64 s[4:5], -1
+; GFX9-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX9-NEXT:    s_mov_b64 exec, s[4:5]
+; GFX9-NEXT:    v_writelane_b32 v40, s33, 3
+; GFX9-NEXT:    v_writelane_b32 v40, s34, 0
+; GFX9-NEXT:    v_writelane_b32 v40, s30, 1
+; GFX9-NEXT:    s_mov_b32 s33, s32
+; GFX9-NEXT:    s_add_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_writelane_b32 v40, s31, 2
+; GFX9-NEXT:    s_getpc_b64 s[4:5]
+; GFX9-NEXT:    s_add_u32 s4, s4, external_void_func_void@rel32@lo+4
+; GFX9-NEXT:    s_addc_u32 s5, s5, external_void_func_void@rel32@hi+12
+; GFX9-NEXT:    ;;#ASMSTART
+; GFX9-NEXT:    ; def s34
+; GFX9-NEXT:    ;;#ASMEND
+; GFX9-NEXT:    s_swappc_b64 s[30:31], s[4:5]
+; GFX9-NEXT:    ;;#ASMSTART
+; GFX9-NEXT:    ; use s34
+; GFX9-NEXT:    ;;#ASMEND
+; GFX9-NEXT:    v_readlane_b32 s4, v40, 1
+; GFX9-NEXT:    v_readlane_b32 s5, v40, 2
+; GFX9-NEXT:    v_readlane_b32 s34, v40, 0
+; GFX9-NEXT:    s_sub_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_readlane_b32 s33, v40, 3
+; GFX9-NEXT:    s_or_saveexec_b64 s[6:7], -1
+; GFX9-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX9-NEXT:    s_mov_b64 exec, s[6:7]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    s_setpc_b64 s[4:5]
+;
+; GFX10-LABEL: test_call_void_func_void_preserves_s34:
+; GFX10:       ; %bb.0:
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    s_or_saveexec_b32 s4, -1
+; GFX10-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s4
+; GFX10-NEXT:    v_writelane_b32 v40, s33, 3
+; GFX10-NEXT:    s_mov_b32 s33, s32
+; GFX10-NEXT:    s_add_u32 s32, s32, 0x200
+; GFX10-NEXT:    s_getpc_b64 s[4:5]
+; GFX10-NEXT:    s_add_u32 s4, s4, external_void_func_void@rel32@lo+4
+; GFX10-NEXT:    s_addc_u32 s5, s5, external_void_func_void@rel32@hi+12
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    v_writelane_b32 v40, s34, 0
+; GFX10-NEXT:    v_writelane_b32 v40, s30, 1
+; GFX10-NEXT:    v_writelane_b32 v40, s31, 2
+; GFX10-NEXT:    ;;#ASMSTART
+; GFX10-NEXT:    ; def s34
+; GFX10-NEXT:    ;;#ASMEND
+; GFX10-NEXT:    s_swappc_b64 s[30:31], s[4:5]
+; GFX10-NEXT:    ;;#ASMSTART
+; GFX10-NEXT:    ; use s34
+; GFX10-NEXT:    ;;#ASMEND
+; GFX10-NEXT:    v_readlane_b32 s4, v40, 1
+; GFX10-NEXT:    v_readlane_b32 s5, v40, 2
+; GFX10-NEXT:    v_readlane_b32 s34, v40, 0
+; GFX10-NEXT:    s_sub_u32 s32, s32, 0x200
+; GFX10-NEXT:    v_readlane_b32 s33, v40, 3
+; GFX10-NEXT:    s_or_saveexec_b32 s6, -1
+; GFX10-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s6
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    s_setpc_b64 s[4:5]
+  %s34 = call i32 asm sideeffect "; def $0", "={s34}"()
+  call amdgpu_gfx_callable void @external_void_func_void()
+  call void asm sideeffect "; use $0", "{s34}"(i32 %s34)
+  ret void
+}
+
+define amdgpu_gfx_callable void @test_call_void_func_void_preserves_v40(i32 addrspace(1)* %out) #0 {
+; GFX9-LABEL: test_call_void_func_void_preserves_v40:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    s_or_saveexec_b64 s[4:5], -1
+; GFX9-NEXT:    buffer_store_dword v41, off, s[0:3], s32 offset:4 ; 4-byte Folded Spill
+; GFX9-NEXT:    s_mov_b64 exec, s[4:5]
+; GFX9-NEXT:    v_writelane_b32 v41, s33, 2
+; GFX9-NEXT:    s_mov_b32 s33, s32
+; GFX9-NEXT:    s_add_u32 s32, s32, 0x400
+; GFX9-NEXT:    s_getpc_b64 s[4:5]
+; GFX9-NEXT:    s_add_u32 s4, s4, external_void_func_void@rel32@lo+4
+; GFX9-NEXT:    s_addc_u32 s5, s5, external_void_func_void@rel32@hi+12
+; GFX9-NEXT:    buffer_store_dword v40, off, s[0:3], s33 ; 4-byte Folded Spill
+; GFX9-NEXT:    v_writelane_b32 v41, s30, 0
+; GFX9-NEXT:    v_writelane_b32 v41, s31, 1
+; GFX9-NEXT:    ;;#ASMSTART
+; GFX9-NEXT:    ; def v40
+; GFX9-NEXT:    ;;#ASMEND
+; GFX9-NEXT:    s_swappc_b64 s[30:31], s[4:5]
+; GFX9-NEXT:    ;;#ASMSTART
+; GFX9-NEXT:    ; use v40
+; GFX9-NEXT:    ;;#ASMEND
+; GFX9-NEXT:    v_readlane_b32 s4, v41, 0
+; GFX9-NEXT:    v_readlane_b32 s5, v41, 1
+; GFX9-NEXT:    buffer_load_dword v40, off, s[0:3], s33 ; 4-byte Folded Reload
+; GFX9-NEXT:    s_sub_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_readlane_b32 s33, v41, 2
+; GFX9-NEXT:    s_or_saveexec_b64 s[6:7], -1
+; GFX9-NEXT:    buffer_load_dword v41, off, s[0:3], s32 offset:4 ; 4-byte Folded Reload
+; GFX9-NEXT:    s_mov_b64 exec, s[6:7]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    s_setpc_b64 s[4:5]
+;
+; GFX10-LABEL: test_call_void_func_void_preserves_v40:
+; GFX10:       ; %bb.0:
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    s_or_saveexec_b32 s4, -1
+; GFX10-NEXT:    buffer_store_dword v41, off, s[0:3], s32 offset:4 ; 4-byte Folded Spill
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s4
+; GFX10-NEXT:    v_writelane_b32 v41, s33, 2
+; GFX10-NEXT:    s_mov_b32 s33, s32
+; GFX10-NEXT:    s_add_u32 s32, s32, 0x200
+; GFX10-NEXT:    s_getpc_b64 s[4:5]
+; GFX10-NEXT:    s_add_u32 s4, s4, external_void_func_void@rel32@lo+4
+; GFX10-NEXT:    s_addc_u32 s5, s5, external_void_func_void@rel32@hi+12
+; GFX10-NEXT:    buffer_store_dword v40, off, s[0:3], s33 ; 4-byte Folded Spill
+; GFX10-NEXT:    v_writelane_b32 v41, s30, 0
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    v_writelane_b32 v41, s31, 1
+; GFX10-NEXT:    ;;#ASMSTART
+; GFX10-NEXT:    ; def v40
+; GFX10-NEXT:    ;;#ASMEND
+; GFX10-NEXT:    s_swappc_b64 s[30:31], s[4:5]
+; GFX10-NEXT:    ;;#ASMSTART
+; GFX10-NEXT:    ; use v40
+; GFX10-NEXT:    ;;#ASMEND
+; GFX10-NEXT:    v_readlane_b32 s4, v41, 0
+; GFX10-NEXT:    v_readlane_b32 s5, v41, 1
+; GFX10-NEXT:    buffer_load_dword v40, off, s[0:3], s33 ; 4-byte Folded Reload
+; GFX10-NEXT:    s_sub_u32 s32, s32, 0x200
+; GFX10-NEXT:    v_readlane_b32 s33, v41, 2
+; GFX10-NEXT:    s_or_saveexec_b32 s6, -1
+; GFX10-NEXT:    buffer_load_dword v41, off, s[0:3], s32 offset:4 ; 4-byte Folded Reload
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s6
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    s_setpc_b64 s[4:5]
+  %v40 = call i32 asm sideeffect "; def $0", "={v40}"()
+  call amdgpu_gfx_callable void @external_void_func_void()
+  call void asm sideeffect "; use $0", "{v40}"(i32 %v40)
+  ret void
+}
+
+define hidden void @void_func_void_clobber_s33() #1 {
+; GFX9-LABEL: void_func_void_clobber_s33:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    v_writelane_b32 v0, s33, 0
+; GFX9-NEXT:    ;;#ASMSTART
+; GFX9-NEXT:    ; clobber
+; GFX9-NEXT:    ;;#ASMEND
+; GFX9-NEXT:    v_readlane_b32 s33, v0, 0
+; GFX9-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX10-LABEL: void_func_void_clobber_s33:
+; GFX10:       ; %bb.0:
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    v_writelane_b32 v0, s33, 0
+; GFX10-NEXT:    ;;#ASMSTART
+; GFX10-NEXT:    ; clobber
+; GFX10-NEXT:    ;;#ASMEND
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    v_readlane_b32 s33, v0, 0
+; GFX10-NEXT:    s_setpc_b64 s[30:31]
+  call void asm sideeffect "; clobber", "~{s33}"() #0
+  ret void
+}
+
+define hidden void @void_func_void_clobber_s34() #1 {
+; GFX9-LABEL: void_func_void_clobber_s34:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    v_writelane_b32 v0, s34, 0
+; GFX9-NEXT:    ;;#ASMSTART
+; GFX9-NEXT:    ; clobber
+; GFX9-NEXT:    ;;#ASMEND
+; GFX9-NEXT:    v_readlane_b32 s34, v0, 0
+; GFX9-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX10-LABEL: void_func_void_clobber_s34:
+; GFX10:       ; %bb.0:
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    v_writelane_b32 v0, s34, 0
+; GFX10-NEXT:    ;;#ASMSTART
+; GFX10-NEXT:    ; clobber
+; GFX10-NEXT:    ;;#ASMEND
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    v_readlane_b32 s34, v0, 0
+; GFX10-NEXT:    s_setpc_b64 s[30:31]
+  call void asm sideeffect "; clobber", "~{s34}"() #0
+  ret void
+}
+
+define amdgpu_gfx_callable void @test_call_void_func_void_clobber_s33() #0 {
+; GFX9-LABEL: test_call_void_func_void_clobber_s33:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    s_or_saveexec_b64 s[4:5], -1
+; GFX9-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX9-NEXT:    s_mov_b64 exec, s[4:5]
+; GFX9-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX9-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX9-NEXT:    s_mov_b32 s33, s32
+; GFX9-NEXT:    s_add_u32 s32, s32, 0x400
+; GFX9-NEXT:    s_getpc_b64 s[4:5]
+; GFX9-NEXT:    s_add_u32 s4, s4, void_func_void_clobber_s33@rel32@lo+4
+; GFX9-NEXT:    s_addc_u32 s5, s5, void_func_void_clobber_s33@rel32@hi+12
+; GFX9-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX9-NEXT:    s_swappc_b64 s[30:31], s[4:5]
+; GFX9-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX9-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX9-NEXT:    s_sub_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX9-NEXT:    s_or_saveexec_b64 s[6:7], -1
+; GFX9-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX9-NEXT:    s_mov_b64 exec, s[6:7]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    s_setpc_b64 s[4:5]
+;
+; GFX10-LABEL: test_call_void_func_void_clobber_s33:
+; GFX10:       ; %bb.0:
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    s_or_saveexec_b32 s4, -1
+; GFX10-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s4
+; GFX10-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX10-NEXT:    s_mov_b32 s33, s32
+; GFX10-NEXT:    s_add_u32 s32, s32, 0x200
+; GFX10-NEXT:    s_getpc_b64 s[4:5]
+; GFX10-NEXT:    s_add_u32 s4, s4, void_func_void_clobber_s33@rel32@lo+4
+; GFX10-NEXT:    s_addc_u32 s5, s5, void_func_void_clobber_s33@rel32@hi+12
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX10-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX10-NEXT:    s_swappc_b64 s[30:31], s[4:5]
+; GFX10-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX10-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX10-NEXT:    s_sub_u32 s32, s32, 0x200
+; GFX10-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX10-NEXT:    s_or_saveexec_b32 s6, -1
+; GFX10-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s6
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    s_setpc_b64 s[4:5]
+  call amdgpu_gfx_callable void @void_func_void_clobber_s33()
+  ret void
+}
+
+define amdgpu_gfx_callable void @test_call_void_func_void_clobber_s34() #0 {
+; GFX9-LABEL: test_call_void_func_void_clobber_s34:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    s_or_saveexec_b64 s[4:5], -1
+; GFX9-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX9-NEXT:    s_mov_b64 exec, s[4:5]
+; GFX9-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX9-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX9-NEXT:    s_mov_b32 s33, s32
+; GFX9-NEXT:    s_add_u32 s32, s32, 0x400
+; GFX9-NEXT:    s_getpc_b64 s[4:5]
+; GFX9-NEXT:    s_add_u32 s4, s4, void_func_void_clobber_s34@rel32@lo+4
+; GFX9-NEXT:    s_addc_u32 s5, s5, void_func_void_clobber_s34@rel32@hi+12
+; GFX9-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX9-NEXT:    s_swappc_b64 s[30:31], s[4:5]
+; GFX9-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX9-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX9-NEXT:    s_sub_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX9-NEXT:    s_or_saveexec_b64 s[6:7], -1
+; GFX9-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX9-NEXT:    s_mov_b64 exec, s[6:7]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    s_setpc_b64 s[4:5]
+;
+; GFX10-LABEL: test_call_void_func_void_clobber_s34:
+; GFX10:       ; %bb.0:
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    s_or_saveexec_b32 s4, -1
+; GFX10-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s4
+; GFX10-NEXT:    v_writelane_b32 v40, s33, 2
+; GFX10-NEXT:    s_mov_b32 s33, s32
+; GFX10-NEXT:    s_add_u32 s32, s32, 0x200
+; GFX10-NEXT:    s_getpc_b64 s[4:5]
+; GFX10-NEXT:    s_add_u32 s4, s4, void_func_void_clobber_s34@rel32@lo+4
+; GFX10-NEXT:    s_addc_u32 s5, s5, void_func_void_clobber_s34@rel32@hi+12
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    v_writelane_b32 v40, s30, 0
+; GFX10-NEXT:    v_writelane_b32 v40, s31, 1
+; GFX10-NEXT:    s_swappc_b64 s[30:31], s[4:5]
+; GFX10-NEXT:    v_readlane_b32 s4, v40, 0
+; GFX10-NEXT:    v_readlane_b32 s5, v40, 1
+; GFX10-NEXT:    s_sub_u32 s32, s32, 0x200
+; GFX10-NEXT:    v_readlane_b32 s33, v40, 2
+; GFX10-NEXT:    s_or_saveexec_b32 s6, -1
+; GFX10-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s6
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    s_setpc_b64 s[4:5]
+  call amdgpu_gfx_callable void @void_func_void_clobber_s34()
+  ret void
+}
+
+define amdgpu_gfx_callable void @callee_saved_sgpr_kernel() #1 {
+; GFX9-LABEL: callee_saved_sgpr_kernel:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    s_or_saveexec_b64 s[4:5], -1
+; GFX9-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX9-NEXT:    s_mov_b64 exec, s[4:5]
+; GFX9-NEXT:    v_writelane_b32 v40, s33, 3
+; GFX9-NEXT:    v_writelane_b32 v40, s40, 0
+; GFX9-NEXT:    v_writelane_b32 v40, s30, 1
+; GFX9-NEXT:    s_mov_b32 s33, s32
+; GFX9-NEXT:    s_add_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_writelane_b32 v40, s31, 2
+; GFX9-NEXT:    s_getpc_b64 s[4:5]
+; GFX9-NEXT:    s_add_u32 s4, s4, external_void_func_void@rel32@lo+4
+; GFX9-NEXT:    s_addc_u32 s5, s5, external_void_func_void@rel32@hi+12
+; GFX9-NEXT:    ;;#ASMSTART
+; GFX9-NEXT:    ; def s40
+; GFX9-NEXT:    ;;#ASMEND
+; GFX9-NEXT:    s_swappc_b64 s[30:31], s[4:5]
+; GFX9-NEXT:    ;;#ASMSTART
+; GFX9-NEXT:    ; use s40
+; GFX9-NEXT:    ;;#ASMEND
+; GFX9-NEXT:    v_readlane_b32 s4, v40, 1
+; GFX9-NEXT:    v_readlane_b32 s5, v40, 2
+; GFX9-NEXT:    v_readlane_b32 s40, v40, 0
+; GFX9-NEXT:    s_sub_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_readlane_b32 s33, v40, 3
+; GFX9-NEXT:    s_or_saveexec_b64 s[6:7], -1
+; GFX9-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX9-NEXT:    s_mov_b64 exec, s[6:7]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    s_setpc_b64 s[4:5]
+;
+; GFX10-LABEL: callee_saved_sgpr_kernel:
+; GFX10:       ; %bb.0:
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    s_or_saveexec_b32 s4, -1
+; GFX10-NEXT:    buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s4
+; GFX10-NEXT:    v_writelane_b32 v40, s33, 3
+; GFX10-NEXT:    s_mov_b32 s33, s32
+; GFX10-NEXT:    s_add_u32 s32, s32, 0x200
+; GFX10-NEXT:    s_getpc_b64 s[4:5]
+; GFX10-NEXT:    s_add_u32 s4, s4, external_void_func_void@rel32@lo+4
+; GFX10-NEXT:    s_addc_u32 s5, s5, external_void_func_void@rel32@hi+12
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    v_writelane_b32 v40, s40, 0
+; GFX10-NEXT:    v_writelane_b32 v40, s30, 1
+; GFX10-NEXT:    v_writelane_b32 v40, s31, 2
+; GFX10-NEXT:    ;;#ASMSTART
+; GFX10-NEXT:    ; def s40
+; GFX10-NEXT:    ;;#ASMEND
+; GFX10-NEXT:    s_swappc_b64 s[30:31], s[4:5]
+; GFX10-NEXT:    ;;#ASMSTART
+; GFX10-NEXT:    ; use s40
+; GFX10-NEXT:    ;;#ASMEND
+; GFX10-NEXT:    v_readlane_b32 s4, v40, 1
+; GFX10-NEXT:    v_readlane_b32 s5, v40, 2
+; GFX10-NEXT:    v_readlane_b32 s40, v40, 0
+; GFX10-NEXT:    s_sub_u32 s32, s32, 0x200
+; GFX10-NEXT:    v_readlane_b32 s33, v40, 3
+; GFX10-NEXT:    s_or_saveexec_b32 s6, -1
+; GFX10-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s6
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    s_setpc_b64 s[4:5]
+  %s40 = call i32 asm sideeffect "; def s40", "={s40}"() #0
+  call amdgpu_gfx_callable void @external_void_func_void()
+  call void asm sideeffect "; use $0", "s"(i32 %s40) #0
+  ret void
+}
+
+define amdgpu_gfx_callable void @callee_saved_sgpr_vgpr_kernel() #1 {
+; GFX9-LABEL: callee_saved_sgpr_vgpr_kernel:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    s_or_saveexec_b64 s[4:5], -1
+; GFX9-NEXT:    buffer_store_dword v41, off, s[0:3], s32 offset:4 ; 4-byte Folded Spill
+; GFX9-NEXT:    s_mov_b64 exec, s[4:5]
+; GFX9-NEXT:    v_writelane_b32 v41, s33, 3
+; GFX9-NEXT:    s_mov_b32 s33, s32
+; GFX9-NEXT:    s_add_u32 s32, s32, 0x400
+; GFX9-NEXT:    s_getpc_b64 s[4:5]
+; GFX9-NEXT:    s_add_u32 s4, s4, external_void_func_void@rel32@lo+4
+; GFX9-NEXT:    s_addc_u32 s5, s5, external_void_func_void@rel32@hi+12
+; GFX9-NEXT:    buffer_store_dword v40, off, s[0:3], s33 ; 4-byte Folded Spill
+; GFX9-NEXT:    v_writelane_b32 v41, s40, 0
+; GFX9-NEXT:    v_writelane_b32 v41, s30, 1
+; GFX9-NEXT:    v_writelane_b32 v41, s31, 2
+; GFX9-NEXT:    ;;#ASMSTART
+; GFX9-NEXT:    ; def s40
+; GFX9-NEXT:    ;;#ASMEND
+; GFX9-NEXT:    ;;#ASMSTART
+; GFX9-NEXT:    ; def v32
+; GFX9-NEXT:    ;;#ASMEND
+; GFX9-NEXT:    v_mov_b32_e32 v40, v32
+; GFX9-NEXT:    s_swappc_b64 s[30:31], s[4:5]
+; GFX9-NEXT:    ;;#ASMSTART
+; GFX9-NEXT:    ; use s40
+; GFX9-NEXT:    ;;#ASMEND
+; GFX9-NEXT:    ;;#ASMSTART
+; GFX9-NEXT:    ; use v40
+; GFX9-NEXT:    ;;#ASMEND
+; GFX9-NEXT:    v_readlane_b32 s4, v41, 1
+; GFX9-NEXT:    v_readlane_b32 s5, v41, 2
+; GFX9-NEXT:    v_readlane_b32 s40, v41, 0
+; GFX9-NEXT:    buffer_load_dword v40, off, s[0:3], s33 ; 4-byte Folded Reload
+; GFX9-NEXT:    s_sub_u32 s32, s32, 0x400
+; GFX9-NEXT:    v_readlane_b32 s33, v41, 3
+; GFX9-NEXT:    s_or_saveexec_b64 s[6:7], -1
+; GFX9-NEXT:    buffer_load_dword v41, off, s[0:3], s32 offset:4 ; 4-byte Folded Reload
+; GFX9-NEXT:    s_mov_b64 exec, s[6:7]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    s_setpc_b64 s[4:5]
+;
+; GFX10-LABEL: callee_saved_sgpr_vgpr_kernel:
+; GFX10:       ; %bb.0:
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    s_or_saveexec_b32 s4, -1
+; GFX10-NEXT:    buffer_store_dword v41, off, s[0:3], s32 offset:4 ; 4-byte Folded Spill
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s4
+; GFX10-NEXT:    v_writelane_b32 v41, s33, 3
+; GFX10-NEXT:    s_mov_b32 s33, s32
+; GFX10-NEXT:    s_add_u32 s32, s32, 0x200
+; GFX10-NEXT:    s_getpc_b64 s[4:5]
+; GFX10-NEXT:    s_add_u32 s4, s4, external_void_func_void@rel32@lo+4
+; GFX10-NEXT:    s_addc_u32 s5, s5, external_void_func_void@rel32@hi+12
+; GFX10-NEXT:    buffer_store_dword v40, off, s[0:3], s33 ; 4-byte Folded Spill
+; GFX10-NEXT:    v_writelane_b32 v41, s40, 0
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    v_writelane_b32 v41, s30, 1
+; GFX10-NEXT:    v_writelane_b32 v41, s31, 2
+; GFX10-NEXT:    ;;#ASMSTART
+; GFX10-NEXT:    ; def s40
+; GFX10-NEXT:    ;;#ASMEND
+; GFX10-NEXT:    ;;#ASMSTART
+; GFX10-NEXT:    ; def v32
+; GFX10-NEXT:    ;;#ASMEND
+; GFX10-NEXT:    v_mov_b32_e32 v40, v32
+; GFX10-NEXT:    s_swappc_b64 s[30:31], s[4:5]
+; GFX10-NEXT:    ;;#ASMSTART
+; GFX10-NEXT:    ; use s40
+; GFX10-NEXT:    ;;#ASMEND
+; GFX10-NEXT:    ;;#ASMSTART
+; GFX10-NEXT:    ; use v40
+; GFX10-NEXT:    ;;#ASMEND
+; GFX10-NEXT:    v_readlane_b32 s4, v41, 1
+; GFX10-NEXT:    v_readlane_b32 s5, v41, 2
+; GFX10-NEXT:    v_readlane_b32 s40, v41, 0
+; GFX10-NEXT:    buffer_load_dword v40, off, s[0:3], s33 ; 4-byte Folded Reload
+; GFX10-NEXT:    s_sub_u32 s32, s32, 0x200
+; GFX10-NEXT:    v_readlane_b32 s33, v41, 3
+; GFX10-NEXT:    s_or_saveexec_b32 s6, -1
+; GFX10-NEXT:    buffer_load_dword v41, off, s[0:3], s32 offset:4 ; 4-byte Folded Reload
+; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX10-NEXT:    s_mov_b32 exec_lo, s6
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    s_setpc_b64 s[4:5]
+  %s40 = call i32 asm sideeffect "; def s40", "={s40}"() #0
+  %v32 = call i32 asm sideeffect "; def v32", "={v32}"() #0
+  call amdgpu_gfx_callable void @external_void_func_void()
+  call void asm sideeffect "; use $0", "s"(i32 %s40) #0
+  call void asm sideeffect "; use $0", "v"(i32 %v32) #0
+  ret void
+}
+
+attributes #0 = { nounwind }
+attributes #1 = { nounwind noinline }

llvm/test/CodeGen/AMDGPU/unsupported-calls.ll

-; RUN: not llc -march=amdgcn -mtriple=amdgcn-- -tailcallopt < %s 2>&1 | FileCheck -check-prefix=GCN %s
+; RUN: not llc -march=amdgcn -mtriple=amdgcn-mesa-mesa3d -tailcallopt < %s 2>&1 | FileCheck -check-prefixes=GCN,MESA %s
+; RUN: not llc -march=amdgcn -mtriple=amdgcn--amdpal -tailcallopt < %s 2>&1 | FileCheck -check-prefixes=GCN,PAL %s
 ; RUN: not llc -march=r600 -mtriple=r600-- -mcpu=cypress -tailcallopt < %s 2>&1 | FileCheck -check-prefix=R600 %s
 
 declare i32 @external_function(i32) nounwind
 
 ; GCN-NOT: error
 ; R600: in function test_call_external{{.*}}: unsupported call to function external_function
 define amdgpu_kernel void @test_call_external(i32 addrspace(1)* %out, i32 addrspace(1)* %in) {
   %b_ptr = getelementptr i32, i32 addrspace(1)* %in, i32 1

 
 ; GCN: :0:0: in function test_indirect_call void (void ()*): unsupported indirect call to function <unknown>
 ; R600: in function test_indirect_call{{.*}}: unsupported call to function <unknown>
 define void @test_indirect_call(void()* %fptr) {
   call void %fptr()
   ret void
 }
 
-; GCN: :0:0: in function test_call_from_shader i32 (): unsupported call from graphics shader of function defined_function
+; GCN-NOT: :0:0: in function test_call_from_shader i32 (): unsupported call from graphics shader of function defined_function
 ; R600: in function test_call{{.*}}: unsupported call to function defined_function
 define amdgpu_ps i32 @test_call_from_shader() {
   %call = call i32 @defined_function(i32 0)
   ret i32 %call
 }
 
 ; FIXME: Bad error message
 ; GCN: error: <unknown>:0:0: in function test_call_absolute void (): unsupported indirect call to function <unknown>
 ; R600: error: <unknown>:0:0: in function test_call_absolute void (): unsupported call to function <unknown>
 define amdgpu_kernel void @test_call_absolute() #0 {
   %val = call i32 inttoptr (i64 1234 to i32(i32)*) (i32 1)
   %op = add i32 %val, 1
   store volatile i32 %op, i32 addrspace(1)* undef
   ret void
 }