[AMDGPU] Emit 32-bit lo/hi got and pc relative variant kinds for external and global address space variables

[• tstellarAMD]

LGTM.

llvm/trunk/lib/Target/AMDGPU/AMDGPUMCInstLower.cpp

 
 
 AMDGPUMCInstLower::AMDGPUMCInstLower(MCContext &ctx, const AMDGPUSubtarget &st,
                                      const AsmPrinter &ap):
   Ctx(ctx), ST(st), AP(ap) { }
 
 static MCSymbolRefExpr::VariantKind getVariantKind(unsigned MOFlags) {
   switch (MOFlags) {
-  default: return MCSymbolRefExpr::VK_None;
-  case SIInstrInfo::MO_GOTPCREL: return MCSymbolRefExpr::VK_GOTPCREL;
+  default:
+    return MCSymbolRefExpr::VK_None;
+  case SIInstrInfo::MO_GOTPCREL:
+    return MCSymbolRefExpr::VK_GOTPCREL;
+  case SIInstrInfo::MO_GOTPCREL32_LO:
+    return MCSymbolRefExpr::VK_AMDGPU_GOTPCREL32_LO;
+  case SIInstrInfo::MO_GOTPCREL32_HI:
+    return MCSymbolRefExpr::VK_AMDGPU_GOTPCREL32_HI;
+  case SIInstrInfo::MO_REL32_LO:
+    return MCSymbolRefExpr::VK_AMDGPU_REL32_LO;
+  case SIInstrInfo::MO_REL32_HI:
+    return MCSymbolRefExpr::VK_AMDGPU_REL32_HI;
   }
 }
 
 const MCExpr *AMDGPUMCInstLower::getLongBranchBlockExpr(
   const MachineBasicBlock &SrcBB,
   const MachineOperand &MO) const {
   const MCExpr *DestBBSym
     = MCSymbolRefExpr::create(MO.getMBB()->getSymbol(), Ctx);

llvm/trunk/lib/Target/AMDGPU/SIISelLowering.cpp

SDValue SITargetLowering::lowerADDRSPACECAST(SDValue Op,
   const MachineFunction &MF = DAG.getMachineFunction();
   DiagnosticInfoUnsupported InvalidAddrSpaceCast(
     *MF.getFunction(), "invalid addrspacecast", SL.getDebugLoc());
   DAG.getContext()->diagnose(InvalidAddrSpaceCast);
 
   return DAG.getUNDEF(ASC->getValueType(0));
 }
 
+static bool shouldEmitFixup(const GlobalValue *GV,
+                            const TargetMachine &TM) {
+  // FIXME: We need to emit global variables in constant address space in a
+  // separate section, and use relocations.
+  return GV->getType()->getAddressSpace() == AMDGPUAS::CONSTANT_ADDRESS;
+}
+
 static bool shouldEmitGOTReloc(const GlobalValue *GV,
                                const TargetMachine &TM) {
   return GV->getType()->getAddressSpace() == AMDGPUAS::GLOBAL_ADDRESS &&
          !TM.shouldAssumeDSOLocal(*GV->getParent(), GV);
 }
 
+static bool shouldEmitPCReloc(const GlobalValue *GV,
+                              const TargetMachine &TM) {
+  return !shouldEmitFixup(GV, TM) && !shouldEmitGOTReloc(GV, TM);
+}
+
 bool
 SITargetLowering::isOffsetFoldingLegal(const GlobalAddressSDNode *GA) const {
   // We can fold offsets for anything that doesn't require a GOT relocation.
   return GA->getAddressSpace() == AMDGPUAS::GLOBAL_ADDRESS &&
          !shouldEmitGOTReloc(GA->getGlobal(), getTargetMachine());
 }
 
 static SDValue buildPCRelGlobalAddress(SelectionDAG &DAG, const GlobalValue *GV,
                                       SDLoc DL, unsigned Offset, EVT PtrVT,
                                       unsigned GAFlags = SIInstrInfo::MO_NONE) {
   // In order to support pc-relative addressing, the PC_ADD_REL_OFFSET SDNode is
   // lowered to the following code sequence:
-  // s_getpc_b64 s[0:1]
-  // s_add_u32 s0, s0, $symbol
-  // s_addc_u32 s1, s1, 0
+  //
+  // For constant address space:
+  //   s_getpc_b64 s[0:1]
+  //   s_add_u32 s0, s0, $symbol
+  //   s_addc_u32 s1, s1, 0
   //
-  // s_getpc_b64 returns the address of the s_add_u32 instruction and then
-  // a fixup or relocation is emitted to replace $symbol with a literal
-  // constant, which is a pc-relative offset from the encoding of the $symbol
-  // operand to the global variable.
+  //   s_getpc_b64 returns the address of the s_add_u32 instruction and then
+  //   a fixup or relocation is emitted to replace $symbol with a literal
+  //   constant, which is a pc-relative offset from the encoding of the $symbol
+  //   operand to the global variable.
   //
+  // For global address space:
+  //   s_getpc_b64 s[0:1]
+  //   s_add_u32 s0, s0, $symbol@{gotpc}rel32@lo
+  //   s_addc_u32 s1, s1, $symbol@{gotpc}rel32@hi
+  //
+  //   s_getpc_b64 returns the address of the s_add_u32 instruction and then
+  //   fixups or relocations are emitted to replace $symbol@*@lo and
+  //   $symbol@*@hi with lower 32 bits and higher 32 bits of a literal constant,
+  //   which is a 64-bit pc-relative offset from the encoding of the $symbol
+  //   operand to the global variable.
+  //
   // What we want here is an offset from the value returned by s_getpc
   // (which is the address of the s_add_u32 instruction) to the global
   // variable, but since the encoding of $symbol starts 4 bytes after the start
   // of the s_add_u32 instruction, we end up with an offset that is 4 bytes too
   // small. This requires us to add 4 to the global variable offset in order to
   // compute the correct address.
-  SDValue GA = DAG.getTargetGlobalAddress(GV, DL, MVT::i32, Offset + 4,
+  SDValue PtrLo = DAG.getTargetGlobalAddress(GV, DL, MVT::i32, Offset + 4,
                                              GAFlags);
-  return DAG.getNode(AMDGPUISD::PC_ADD_REL_OFFSET, DL, PtrVT, GA);
+  SDValue PtrHi = DAG.getTargetGlobalAddress(GV, DL, MVT::i32, Offset + 4,
+                                             GAFlags == SIInstrInfo::MO_NONE ?
+                                             GAFlags : GAFlags + 1);
+  return DAG.getNode(AMDGPUISD::PC_ADD_REL_OFFSET, DL, PtrVT, PtrLo, PtrHi);
 }
 
 SDValue SITargetLowering::LowerGlobalAddress(AMDGPUMachineFunction *MFI,
                                              SDValue Op,
                                              SelectionDAG &DAG) const {
   GlobalAddressSDNode *GSD = cast<GlobalAddressSDNode>(Op);
 
   if (GSD->getAddressSpace() != AMDGPUAS::CONSTANT_ADDRESS &&
       GSD->getAddressSpace() != AMDGPUAS::GLOBAL_ADDRESS)
     return AMDGPUTargetLowering::LowerGlobalAddress(MFI, Op, DAG);
 
   SDLoc DL(GSD);
   const GlobalValue *GV = GSD->getGlobal();
   EVT PtrVT = Op.getValueType();
 
-  if (!shouldEmitGOTReloc(GV, getTargetMachine()))
+  if (shouldEmitFixup(GV, getTargetMachine()))
     return buildPCRelGlobalAddress(DAG, GV, DL, GSD->getOffset(), PtrVT);
+  else if (shouldEmitPCReloc(GV, getTargetMachine()))
+    return buildPCRelGlobalAddress(DAG, GV, DL, GSD->getOffset(), PtrVT,
+                                   SIInstrInfo::MO_REL32);
 
   SDValue GOTAddr = buildPCRelGlobalAddress(DAG, GV, DL, 0, PtrVT,
-                                            SIInstrInfo::MO_GOTPCREL);
+                                            SIInstrInfo::MO_GOTPCREL32);
 
   Type *Ty = PtrVT.getTypeForEVT(*DAG.getContext());
   PointerType *PtrTy = PointerType::get(Ty, AMDGPUAS::CONSTANT_ADDRESS);
   const DataLayout &DataLayout = DAG.getDataLayout();
   unsigned Align = DataLayout.getABITypeAlignment(PtrTy);
   // FIXME: Use a PseudoSourceValue once those can be assigned an address space.
   MachinePointerInfo PtrInfo(UndefValue::get(PtrTy));
 

llvm/trunk/lib/Target/AMDGPU/SIInstrInfo.h

protected:
   MachineInstr *commuteInstructionImpl(MachineInstr &MI, bool NewMI,
                                        unsigned OpIdx0,
                                        unsigned OpIdx1) const override;
 
 public:
 
   enum TargetOperandFlags {
     MO_NONE = 0,
-    MO_GOTPCREL = 1
+    // MO_GOTPCREL -> symbol@GOTPCREL -> R_AMDGPU_GOTPCREL.
+    MO_GOTPCREL = 1,
+    // MO_GOTPCREL32_LO -> symbol@gotpcrel32@lo -> R_AMDGPU_GOTPCREL32_LO.
+    MO_GOTPCREL32 = 2,
+    MO_GOTPCREL32_LO = 2,
+    // MO_GOTPCREL32_HI -> symbol@gotpcrel32@hi -> R_AMDGPU_GOTPCREL32_HI.
+    MO_GOTPCREL32_HI = 3,
+    // MO_REL32_LO -> symbol@rel32@lo -> R_AMDGPU_REL32_LO.
+    MO_REL32 = 4,
+    MO_REL32_LO = 4,
+    // MO_REL32_HI -> symbol@rel32@hi -> R_AMDGPU_REL32_HI.
+    MO_REL32_HI = 5
   };
 
   explicit SIInstrInfo(const SISubtarget &);
 
   const SIRegisterInfo &getRegisterInfo() const {
     return RI;
   }
 

llvm/trunk/lib/Target/AMDGPU/SIInstrInfo.cpp

case AMDGPU::SI_PC_ADD_REL_OFFSET: {
     MIBundleBuilder Bundler(MBB, MI);
     Bundler.append(BuildMI(MF, DL, get(AMDGPU::S_GETPC_B64), Reg));
 
     // Add 32-bit offset from this instruction to the start of the
     // constant data.
     Bundler.append(BuildMI(MF, DL, get(AMDGPU::S_ADD_U32), RegLo)
                        .addReg(RegLo)
                        .addOperand(MI.getOperand(1)));
-    Bundler.append(BuildMI(MF, DL, get(AMDGPU::S_ADDC_U32), RegHi)
-                           .addReg(RegHi)
-                           .addImm(0));
 
+    MachineInstrBuilder MIB = BuildMI(MF, DL, get(AMDGPU::S_ADDC_U32), RegHi)
+                                  .addReg(RegHi);
+    if (MI.getOperand(2).getTargetFlags() == SIInstrInfo::MO_NONE)
+      MIB.addImm(0);
+    else
+      MIB.addOperand(MI.getOperand(2));
+
+    Bundler.append(MIB);
     llvm::finalizeBundle(MBB, Bundler.begin());
 
     MI.eraseFromParent();
     break;
   }
   }
   return true;
 }

llvm/trunk/lib/Target/AMDGPU/SIInstrInfo.td

 >;
 
 def SIsample : SDSample<"AMDGPUISD::SAMPLE">;
 def SIsampleb : SDSample<"AMDGPUISD::SAMPLEB">;
 def SIsampled : SDSample<"AMDGPUISD::SAMPLED">;
 def SIsamplel : SDSample<"AMDGPUISD::SAMPLEL">;
 
 def SIpc_add_rel_offset : SDNode<"AMDGPUISD::PC_ADD_REL_OFFSET",
-  SDTypeProfile<1, 1, [SDTCisVT<0, iPTR>, SDTCisSameAs<0,1>]>
+  SDTypeProfile<1, 2, [SDTCisVT<0, iPTR>, SDTCisSameAs<0,1>, SDTCisSameAs<0,2>]>
 >;
 
 //===----------------------------------------------------------------------===//
 // PatFrags for global memory operations
 //===----------------------------------------------------------------------===//
 
 def atomic_inc_global : global_binary_atomic_op<SIatomic_inc>;
 def atomic_dec_global : global_binary_atomic_op<SIatomic_dec>;

llvm/trunk/lib/Target/AMDGPU/SIInstructions.td

 defm SI_SPILL_V64  : SI_SPILL_VGPR <VReg_64>;
 defm SI_SPILL_V96  : SI_SPILL_VGPR <VReg_96>;
 defm SI_SPILL_V128 : SI_SPILL_VGPR <VReg_128>;
 defm SI_SPILL_V256 : SI_SPILL_VGPR <VReg_256>;
 defm SI_SPILL_V512 : SI_SPILL_VGPR <VReg_512>;
 
 def SI_PC_ADD_REL_OFFSET : SPseudoInstSI <
   (outs SReg_64:$dst),
-  (ins si_ga:$ptr),
-  [(set SReg_64:$dst, (i64 (SIpc_add_rel_offset (tglobaladdr:$ptr))))]> {
+  (ins si_ga:$ptr_lo, si_ga:$ptr_hi),
+  [(set SReg_64:$dst,
+   (i64 (SIpc_add_rel_offset (tglobaladdr:$ptr_lo), (tglobaladdr:$ptr_hi))))]> {
   let Defs = [SCC];
 }
 
 } // End SubtargetPredicate = isGCN
 
 let Predicates = [isGCN] in {
 
 def : Pat<

llvm/trunk/test/CodeGen/AMDGPU/captured-frame-index.ll

 }
 
 @g1 = external addrspace(1) global i32*
 
 ; This was leaving a dead node around resulting in failing to select
 ; on the leftover AssertZext's ValueType operand.
 
 ; GCN-LABEL: {{^}}cannot_select_assertzext_valuetype:
-; GCN: s_add_u32 s{{[0-9]+}}, s{{[0-9]+}}, g1@GOTPCREL+4
+; GCN: s_getpc_b64 s{{\[}}[[PC_LO:[0-9]+]]:[[PC_HI:[0-9]+]]{{\]}}
+; GCN: s_add_u32 s{{[0-9]+}}, s[[PC_LO]], g1@gotpcrel32@lo+4
+; GCN: s_addc_u32 s{{[0-9]+}}, s[[PC_HI]], g1@gotpcrel32@hi+4
 ; GCN: v_mov_b32_e32 [[FI:v[0-9]+]], 0{{$}}
 ; GCN: buffer_store_dword [[FI]]
 define void @cannot_select_assertzext_valuetype(i32 addrspace(1)* %out, i32 %idx) #0 {
 entry:
   %b = alloca i32, align 4
   %tmp1 = load volatile i32*, i32* addrspace(1)* @g1, align 4
   %arrayidx = getelementptr inbounds i32, i32* %tmp1, i32 %idx
   %tmp2 = load i32, i32* %arrayidx, align 4

llvm/trunk/test/CodeGen/AMDGPU/global-variable-relocs.ll

 ; RUN: llc -mtriple=amdgcn--amdhsa -mcpu=fiji < %s | FileCheck %s
 
 @private = private addrspace(1) global [256 x i32] zeroinitializer
 @internal = internal addrspace(1) global [256 x i32] zeroinitializer
 @available_externally = available_externally addrspace(1) global [256 x i32] zeroinitializer
 @linkonce = linkonce addrspace(1) global [256 x i32] zeroinitializer
 @weak= weak addrspace(1) global [256 x i32] zeroinitializer
 @common = common addrspace(1) global [256 x i32] zeroinitializer
 @extern_weak = extern_weak addrspace(1) global [256 x i32]
 @linkonce_odr = linkonce_odr addrspace(1) global [256 x i32] zeroinitializer
 @weak_odr = weak_odr addrspace(1) global [256 x i32] zeroinitializer
 @external = external addrspace(1) global [256 x i32]
 @external_w_init = addrspace(1) global [256 x i32] zeroinitializer
 
 ; CHECK-LABEL: private_test:
 ; CHECK: s_getpc_b64 s{{\[}}[[PC_LO:[0-9]+]]:[[PC_HI:[0-9]+]]{{\]}}
-; CHECK: s_add_u32 s[[ADDR_LO:[0-9]+]], s[[PC_LO]], private+8
-; CHECK: s_addc_u32 s[[ADDR_HI:[0-9]+]], s[[PC_HI]], 0
+; CHECK: s_add_u32 s[[ADDR_LO:[0-9]+]], s[[PC_LO]], private@rel32@lo+8
+; CHECK: s_addc_u32 s[[ADDR_HI:[0-9]+]], s[[PC_HI]], private@rel32@hi+8
 ; CHECK-DAG: v_mov_b32_e32 v[[V_LO:[0-9]+]], s[[ADDR_LO]]
 ; CHECK-DAG: v_mov_b32_e32 v[[V_HI:[0-9]+]], s[[ADDR_HI]]
 ; CHECK: flat_load_dword v{{[0-9]+}}, v{{\[}}[[V_LO]]:[[V_HI]]{{\]}}
 define void @private_test(i32 addrspace(1)* %out) {
   %ptr = getelementptr [256 x i32], [256 x i32] addrspace(1)* @private, i32 0, i32 1
   %val = load i32, i32 addrspace(1)* %ptr
   store i32 %val, i32 addrspace(1)* %out
   ret void
 }
 
 ; CHECK-LABEL: internal_test:
 ; CHECK: s_getpc_b64 s{{\[}}[[PC_LO:[0-9]+]]:[[PC_HI:[0-9]+]]{{\]}}
-; CHECK: s_add_u32 s[[ADDR_LO:[0-9]+]], s[[PC_LO]], internal+8
-; CHECK: s_addc_u32 s[[ADDR_HI:[0-9]+]], s[[PC_HI]], 0
+; CHECK: s_add_u32 s[[ADDR_LO:[0-9]+]], s[[PC_LO]], internal@rel32@lo+8
+; CHECK: s_addc_u32 s[[ADDR_HI:[0-9]+]], s[[PC_HI]], internal@rel32@hi+8
 ; CHECK-DAG: v_mov_b32_e32 v[[V_LO:[0-9]+]], s[[ADDR_LO]]
 ; CHECK-DAG: v_mov_b32_e32 v[[V_HI:[0-9]+]], s[[ADDR_HI]]
 ; CHECK: flat_load_dword v{{[0-9]+}}, v{{\[}}[[V_LO]]:[[V_HI]]{{\]}}
 define void @internal_test(i32 addrspace(1)* %out) {
   %ptr = getelementptr [256 x i32], [256 x i32] addrspace(1)* @internal, i32 0, i32 1
   %val = load i32, i32 addrspace(1)* %ptr
   store i32 %val, i32 addrspace(1)* %out
   ret void
 }
 
 ; CHECK-LABEL: available_externally_test:
 ; CHECK: s_getpc_b64 s{{\[}}[[PC_LO:[0-9]+]]:[[PC_HI:[0-9]+]]{{\]}}
-; CHECK: s_add_u32 s[[GOTADDR_LO:[0-9]+]], s[[PC_LO]], available_externally@GOTPCREL+4
-; CHECK: s_addc_u32 s[[GOTADDR_HI:[0-9]+]], s[[PC_HI]], 0
+; CHECK: s_add_u32 s[[GOTADDR_LO:[0-9]+]], s[[PC_LO]], available_externally@gotpcrel32@lo+4
+; CHECK: s_addc_u32 s[[GOTADDR_HI:[0-9]+]], s[[PC_HI]], available_externally@gotpcrel32@hi+4
 ; CHECK: s_load_dwordx2 s{{\[}}[[ADDR_LO:[0-9]+]]:[[ADDR_HI:[0-9]+]]{{\]}}, s{{\[}}[[GOTADDR_LO]]:[[GOTADDR_HI]]{{\]}}, 0x0
 ; CHECK: s_add_u32 s[[GEP_LO:[0-9]+]], s[[ADDR_LO]], 4
 ; CHECK: s_addc_u32 s[[GEP_HI:[0-9]+]], s[[ADDR_HI]], 0
 ; CHECK-DAG: v_mov_b32_e32 v[[V_LO:[0-9]+]], s[[GEP_LO]]
 ; CHECK-DAG: v_mov_b32_e32 v[[V_HI:[0-9]+]], s[[GEP_HI]]
 ; CHECK: flat_load_dword v{{[0-9]+}}, v{{\[}}[[V_LO]]:[[V_HI]]{{\]}}
 define void @available_externally_test(i32 addrspace(1)* %out) {
   %ptr = getelementptr [256 x i32], [256 x i32] addrspace(1)* @available_externally, i32 0, i32 1
   %val = load i32, i32 addrspace(1)* %ptr
   store i32 %val, i32 addrspace(1)* %out
   ret void
 }
 
 ; CHECK-LABEL: linkonce_test:
 ; CHECK: s_getpc_b64 s{{\[}}[[PC_LO:[0-9]+]]:[[PC_HI:[0-9]+]]{{\]}}
-; CHECK: s_add_u32 s[[GOTADDR_LO:[0-9]+]], s[[PC_LO]], linkonce@GOTPCREL+4
-; CHECK: s_addc_u32 s[[GOTADDR_HI:[0-9]+]], s[[PC_HI]], 0
+; CHECK: s_add_u32 s[[GOTADDR_LO:[0-9]+]], s[[PC_LO]], linkonce@gotpcrel32@lo+4
+; CHECK: s_addc_u32 s[[GOTADDR_HI:[0-9]+]], s[[PC_HI]], linkonce@gotpcrel32@hi+4
 ; CHECK: s_load_dwordx2 s{{\[}}[[ADDR_LO:[0-9]+]]:[[ADDR_HI:[0-9]+]]{{\]}}, s{{\[}}[[GOTADDR_LO]]:[[GOTADDR_HI]]{{\]}}, 0x0
 ; CHECK: s_add_u32 s[[GEP_LO:[0-9]+]], s[[ADDR_LO]], 4
 ; CHECK: s_addc_u32 s[[GEP_HI:[0-9]+]], s[[ADDR_HI]], 0
 ; CHECK-DAG: v_mov_b32_e32 v[[V_LO:[0-9]+]], s[[GEP_LO]]
 ; CHECK-DAG: v_mov_b32_e32 v[[V_HI:[0-9]+]], s[[GEP_HI]]
 ; CHECK: flat_load_dword v{{[0-9]+}}, v{{\[}}[[V_LO]]:[[V_HI]]{{\]}}
 define void @linkonce_test(i32 addrspace(1)* %out) {
   %ptr = getelementptr [256 x i32], [256 x i32] addrspace(1)* @linkonce, i32 0, i32 1
   %val = load i32, i32 addrspace(1)* %ptr
   store i32 %val, i32 addrspace(1)* %out
   ret void
 }
 
 ; CHECK-LABEL: weak_test:
 ; CHECK: s_getpc_b64 s{{\[}}[[PC_LO:[0-9]+]]:[[PC_HI:[0-9]+]]{{\]}}
-; CHECK: s_add_u32 s[[GOTADDR_LO:[0-9]+]], s[[PC_LO]], weak@GOTPCREL+4
-; CHECK: s_addc_u32 s[[GOTADDR_HI:[0-9]+]], s[[PC_HI]], 0
+; CHECK: s_add_u32 s[[GOTADDR_LO:[0-9]+]], s[[PC_LO]], weak@gotpcrel32@lo+4
+; CHECK: s_addc_u32 s[[GOTADDR_HI:[0-9]+]], s[[PC_HI]], weak@gotpcrel32@hi+4
 ; CHECK: s_load_dwordx2 s{{\[}}[[ADDR_LO:[0-9]+]]:[[ADDR_HI:[0-9]+]]{{\]}}, s{{\[}}[[GOTADDR_LO]]:[[GOTADDR_HI]]{{\]}}, 0x0
 ; CHECK: s_add_u32 s[[GEP_LO:[0-9]+]], s[[ADDR_LO]], 4
 ; CHECK: s_addc_u32 s[[GEP_HI:[0-9]+]], s[[ADDR_HI]], 0
 ; CHECK-DAG: v_mov_b32_e32 v[[V_LO:[0-9]+]], s[[GEP_LO]]
 ; CHECK-DAG: v_mov_b32_e32 v[[V_HI:[0-9]+]], s[[GEP_HI]]
 ; CHECK: flat_load_dword v{{[0-9]+}}, v{{\[}}[[V_LO]]:[[V_HI]]{{\]}}
 define void @weak_test(i32 addrspace(1)* %out) {
   %ptr = getelementptr [256 x i32], [256 x i32] addrspace(1)* @weak, i32 0, i32 1
   %val = load i32, i32 addrspace(1)* %ptr
   store i32 %val, i32 addrspace(1)* %out
   ret void
 }
 
 ; CHECK-LABEL: common_test:
 ; CHECK: s_getpc_b64 s{{\[}}[[PC_LO:[0-9]+]]:[[PC_HI:[0-9]+]]{{\]}}
-; CHECK: s_add_u32 s[[GOTADDR_LO:[0-9]+]], s[[PC_LO]], common@GOTPCREL+4
-; CHECK: s_addc_u32 s[[GOTADDR_HI:[0-9]+]], s[[PC_HI]], 0
+; CHECK: s_add_u32 s[[GOTADDR_LO:[0-9]+]], s[[PC_LO]], common@gotpcrel32@lo+4
+; CHECK: s_addc_u32 s[[GOTADDR_HI:[0-9]+]], s[[PC_HI]], common@gotpcrel32@hi+4
 ; CHECK: s_load_dwordx2 s{{\[}}[[ADDR_LO:[0-9]+]]:[[ADDR_HI:[0-9]+]]{{\]}}, s{{\[}}[[GOTADDR_LO]]:[[GOTADDR_HI]]{{\]}}, 0x0
 ; CHECK: s_add_u32 s[[GEP_LO:[0-9]+]], s[[ADDR_LO]], 4
 ; CHECK: s_addc_u32 s[[GEP_HI:[0-9]+]], s[[ADDR_HI]], 0
 ; CHECK-DAG: v_mov_b32_e32 v[[V_LO:[0-9]+]], s[[GEP_LO]]
 ; CHECK-DAG: v_mov_b32_e32 v[[V_HI:[0-9]+]], s[[GEP_HI]]
 ; CHECK: flat_load_dword v{{[0-9]+}}, v{{\[}}[[V_LO]]:[[V_HI]]{{\]}}
 define void @common_test(i32 addrspace(1)* %out) {
   %ptr = getelementptr [256 x i32], [256 x i32] addrspace(1)* @common, i32 0, i32 1
   %val = load i32, i32 addrspace(1)* %ptr
   store i32 %val, i32 addrspace(1)* %out
   ret void
 }
 
 ; CHECK-LABEL: extern_weak_test:
 ; CHECK: s_getpc_b64 s{{\[}}[[PC_LO:[0-9]+]]:[[PC_HI:[0-9]+]]{{\]}}
-; CHECK: s_add_u32 s[[GOTADDR_LO:[0-9]+]], s[[PC_LO]], extern_weak@GOTPCREL+4
-; CHECK: s_addc_u32 s[[GOTADDR_HI:[0-9]+]], s[[PC_HI]], 0
+; CHECK: s_add_u32 s[[GOTADDR_LO:[0-9]+]], s[[PC_LO]], extern_weak@gotpcrel32@lo+4
+; CHECK: s_addc_u32 s[[GOTADDR_HI:[0-9]+]], s[[PC_HI]], extern_weak@gotpcrel32@hi+4
 ; CHECK: s_load_dwordx2 s{{\[}}[[ADDR_LO:[0-9]+]]:[[ADDR_HI:[0-9]+]]{{\]}}, s{{\[}}[[GOTADDR_LO]]:[[GOTADDR_HI]]{{\]}}, 0x0
 ; CHECK: s_add_u32 s[[GEP_LO:[0-9]+]], s[[ADDR_LO]], 4
 ; CHECK: s_addc_u32 s[[GEP_HI:[0-9]+]], s[[ADDR_HI]], 0
 ; CHECK-DAG: v_mov_b32_e32 v[[V_LO:[0-9]+]], s[[GEP_LO]]
 ; CHECK-DAG: v_mov_b32_e32 v[[V_HI:[0-9]+]], s[[GEP_HI]]
 ; CHECK: flat_load_dword v{{[0-9]+}}, v{{\[}}[[V_LO]]:[[V_HI]]{{\]}}
 define void @extern_weak_test(i32 addrspace(1)* %out) {
   %ptr = getelementptr [256 x i32], [256 x i32] addrspace(1)* @extern_weak, i32 0, i32 1
   %val = load i32, i32 addrspace(1)* %ptr
   store i32 %val, i32 addrspace(1)* %out
   ret void
 }
 
 ; CHECK-LABEL: linkonce_odr_test:
 ; CHECK: s_getpc_b64 s{{\[}}[[PC_LO:[0-9]+]]:[[PC_HI:[0-9]+]]{{\]}}
-; CHECK: s_add_u32 s[[GOTADDR_LO:[0-9]+]], s[[PC_LO]], linkonce_odr@GOTPCREL+4
-; CHECK: s_addc_u32 s[[GOTADDR_HI:[0-9]+]], s[[PC_HI]], 0
+; CHECK: s_add_u32 s[[GOTADDR_LO:[0-9]+]], s[[PC_LO]], linkonce_odr@gotpcrel32@lo+4
+; CHECK: s_addc_u32 s[[GOTADDR_HI:[0-9]+]], s[[PC_HI]], linkonce_odr@gotpcrel32@hi+4
 ; CHECK: s_load_dwordx2 s{{\[}}[[ADDR_LO:[0-9]+]]:[[ADDR_HI:[0-9]+]]{{\]}}, s{{\[}}[[GOTADDR_LO]]:[[GOTADDR_HI]]{{\]}}, 0x0
 ; CHECK: s_add_u32 s[[GEP_LO:[0-9]+]], s[[ADDR_LO]], 4
 ; CHECK: s_addc_u32 s[[GEP_HI:[0-9]+]], s[[ADDR_HI]], 0
 ; CHECK-DAG: v_mov_b32_e32 v[[V_LO:[0-9]+]], s[[GEP_LO]]
 ; CHECK-DAG: v_mov_b32_e32 v[[V_HI:[0-9]+]], s[[GEP_HI]]
 ; CHECK: flat_load_dword v{{[0-9]+}}, v{{\[}}[[V_LO]]:[[V_HI]]{{\]}}
 define void @linkonce_odr_test(i32 addrspace(1)* %out) {
   %ptr = getelementptr [256 x i32], [256 x i32] addrspace(1)* @linkonce_odr, i32 0, i32 1
   %val = load i32, i32 addrspace(1)* %ptr
   store i32 %val, i32 addrspace(1)* %out
   ret void
 }
 
 ; CHECK-LABEL: weak_odr_test:
 ; CHECK: s_getpc_b64 s{{\[}}[[PC_LO:[0-9]+]]:[[PC_HI:[0-9]+]]{{\]}}
-; CHECK: s_add_u32 s[[GOTADDR_LO:[0-9]+]], s[[PC_LO]], weak_odr@GOTPCREL+4
-; CHECK: s_addc_u32 s[[GOTADDR_HI:[0-9]+]], s[[PC_HI]], 0
+; CHECK: s_add_u32 s[[GOTADDR_LO:[0-9]+]], s[[PC_LO]], weak_odr@gotpcrel32@lo+4
+; CHECK: s_addc_u32 s[[GOTADDR_HI:[0-9]+]], s[[PC_HI]], weak_odr@gotpcrel32@hi+4
 ; CHECK: s_load_dwordx2 s{{\[}}[[ADDR_LO:[0-9]+]]:[[ADDR_HI:[0-9]+]]{{\]}}, s{{\[}}[[GOTADDR_LO]]:[[GOTADDR_HI]]{{\]}}, 0x0
 ; CHECK: s_add_u32 s[[GEP_LO:[0-9]+]], s[[ADDR_LO]], 4
 ; CHECK: s_addc_u32 s[[GEP_HI:[0-9]+]], s[[ADDR_HI]], 0
 ; CHECK-DAG: v_mov_b32_e32 v[[V_LO:[0-9]+]], s[[GEP_LO]]
 ; CHECK-DAG: v_mov_b32_e32 v[[V_HI:[0-9]+]], s[[GEP_HI]]
 ; CHECK: flat_load_dword v{{[0-9]+}}, v{{\[}}[[V_LO]]:[[V_HI]]{{\]}}
 define void @weak_odr_test(i32 addrspace(1)* %out) {
   %ptr = getelementptr [256 x i32], [256 x i32] addrspace(1)* @weak_odr, i32 0, i32 1
   %val = load i32, i32 addrspace(1)* %ptr
   store i32 %val, i32 addrspace(1)* %out
   ret void
 }
 
 ; CHECK-LABEL: external_test:
 ; CHECK: s_getpc_b64 s{{\[}}[[PC_LO:[0-9]+]]:[[PC_HI:[0-9]+]]{{\]}}
-; CHECK: s_add_u32 s[[GOTADDR_LO:[0-9]+]], s[[PC_LO]], external@GOTPCREL+4
-; CHECK: s_addc_u32 s[[GOTADDR_HI:[0-9]+]], s[[PC_HI]], 0
+; CHECK: s_add_u32 s[[GOTADDR_LO:[0-9]+]], s[[PC_LO]], external@gotpcrel32@lo+4
+; CHECK: s_addc_u32 s[[GOTADDR_HI:[0-9]+]], s[[PC_HI]], external@gotpcrel32@hi+4
 ; CHECK: s_load_dwordx2 s{{\[}}[[ADDR_LO:[0-9]+]]:[[ADDR_HI:[0-9]+]]{{\]}}, s{{\[}}[[GOTADDR_LO]]:[[GOTADDR_HI]]{{\]}}, 0x0
 ; CHECK: s_add_u32 s[[GEP_LO:[0-9]+]], s[[ADDR_LO]], 4
 ; CHECK: s_addc_u32 s[[GEP_HI:[0-9]+]], s[[ADDR_HI]], 0
 ; CHECK-DAG: v_mov_b32_e32 v[[V_LO:[0-9]+]], s[[GEP_LO]]
 ; CHECK-DAG: v_mov_b32_e32 v[[V_HI:[0-9]+]], s[[GEP_HI]]
 ; CHECK: flat_load_dword v{{[0-9]+}}, v{{\[}}[[V_LO]]:[[V_HI]]{{\]}}
 define void @external_test(i32 addrspace(1)* %out) {
   %ptr = getelementptr [256 x i32], [256 x i32] addrspace(1)* @external, i32 0, i32 1
   %val = load i32, i32 addrspace(1)* %ptr
   store i32 %val, i32 addrspace(1)* %out
   ret void
 }
 
 ; CHECK-LABEL: external_w_init_test:
 ; CHECK: s_getpc_b64 s{{\[}}[[PC_LO:[0-9]+]]:[[PC_HI:[0-9]+]]{{\]}}
-; CHECK: s_add_u32 s[[GOTADDR_LO:[0-9]+]], s[[PC_LO]], external_w_init@GOTPCREL+4
-; CHECK: s_addc_u32 s[[GOTADDR_HI:[0-9]+]], s[[PC_HI]], 0
+; CHECK: s_add_u32 s[[GOTADDR_LO:[0-9]+]], s[[PC_LO]], external_w_init@gotpcrel32@lo+4
+; CHECK: s_addc_u32 s[[GOTADDR_HI:[0-9]+]], s[[PC_HI]], external_w_init@gotpcrel32@hi+4
 ; CHECK: s_load_dwordx2 s{{\[}}[[ADDR_LO:[0-9]+]]:[[ADDR_HI:[0-9]+]]{{\]}}, s{{\[}}[[GOTADDR_LO]]:[[GOTADDR_HI]]{{\]}}, 0x0
 ; CHECK: s_add_u32 s[[GEP_LO:[0-9]+]], s[[ADDR_LO]], 4
 ; CHECK: s_addc_u32 s[[GEP_HI:[0-9]+]], s[[ADDR_HI]], 0
 ; CHECK-DAG: v_mov_b32_e32 v[[V_LO:[0-9]+]], s[[GEP_LO]]
 ; CHECK-DAG: v_mov_b32_e32 v[[V_HI:[0-9]+]], s[[GEP_HI]]
 ; CHECK: flat_load_dword v{{[0-9]+}}, v{{\[}}[[V_LO]]:[[V_HI]]{{\]}}
 define void @external_w_init_test(i32 addrspace(1)* %out) {
   %ptr = getelementptr [256 x i32], [256 x i32] addrspace(1)* @external_w_init, i32 0, i32 1