[AMDGPU] gfx1010: use fmac instructions

[msearles]

LGTM

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

if (Subtarget->has16BitInsts()) {
     // F16 - VOP2 Actions.
     setOperationAction(ISD::BR_CC, MVT::f16, Expand);
     setOperationAction(ISD::SELECT_CC, MVT::f16, Expand);
 
     setOperationAction(ISD::FDIV, MVT::f16, Custom);
 
     // F16 - VOP3 Actions.
     setOperationAction(ISD::FMA, MVT::f16, Legal);
-    if (!Subtarget->hasFP16Denormals())
+    if (!Subtarget->hasFP16Denormals() && STI.hasMadF16())
       setOperationAction(ISD::FMAD, MVT::f16, Legal);
 
     for (MVT VT : {MVT::v2i16, MVT::v2f16, MVT::v4i16, MVT::v4f16}) {
       for (unsigned Op = 0; Op < ISD::BUILTIN_OP_END; ++Op) {
         switch (Op) {
         case ISD::LOAD:
         case ISD::STORE:
         case ISD::BUILD_VECTOR:

 
 unsigned SITargetLowering::getFusedOpcode(const SelectionDAG &DAG,
                                           const SDNode *N0,
                                           const SDNode *N1) const {
   EVT VT = N0->getValueType(0);
 
   // Only do this if we are not trying to support denormals. v_mad_f32 does not
   // support denormals ever.
-  if ((VT == MVT::f32 && !Subtarget->hasFP32Denormals()) ||
-      (VT == MVT::f16 && !Subtarget->hasFP16Denormals()))
+  if (((VT == MVT::f32 && !Subtarget->hasFP32Denormals()) ||
+       (VT == MVT::f16 && !Subtarget->hasFP16Denormals() &&
+        getSubtarget()->hasMadF16())) &&
+       isOperationLegal(ISD::FMAD, VT))
     return ISD::FMAD;
 
   const TargetOptions &Options = DAG.getTarget().Options;
   if ((Options.AllowFPOpFusion == FPOpFusion::Fast || Options.UnsafeFPMath ||
        (N0->getFlags().hasAllowContract() &&
         N1->getFlags().hasAllowContract())) &&
       isFMAFasterThanFMulAndFAdd(VT)) {
     return ISD::FMA;

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

if (Opc == AMDGPU::COPY) {
     unsigned NewOpc = isVGPRCopy ? AMDGPU::V_MOV_B32_e32 : AMDGPU::S_MOV_B32;
     UseMI.setDesc(get(NewOpc));
     UseMI.getOperand(1).ChangeToImmediate(ImmOp->getImm());
     UseMI.addImplicitDefUseOperands(*UseMI.getParent()->getParent());
     return true;
   }
 
   if (Opc == AMDGPU::V_MAD_F32 || Opc == AMDGPU::V_MAC_F32_e64 ||
-      Opc == AMDGPU::V_MAD_F16 || Opc == AMDGPU::V_MAC_F16_e64) {
+      Opc == AMDGPU::V_MAD_F16 || Opc == AMDGPU::V_MAC_F16_e64 ||
+      Opc == AMDGPU::V_FMA_F32 || Opc == AMDGPU::V_FMAC_F32_e64 ||
+      Opc == AMDGPU::V_FMA_F16 || Opc == AMDGPU::V_FMAC_F16_e64) {
     // Don't fold if we are using source or output modifiers. The new VOP2
     // instructions don't have them.
     if (hasAnyModifiersSet(UseMI))
       return false;
 
     // If this is a free constant, there's no reason to do this.
     // TODO: We could fold this here instead of letting SIFoldOperands do it
     // later.
     MachineOperand *Src0 = getNamedOperand(UseMI, AMDGPU::OpName::src0);
 
     // Any src operand can be used for the legality check.
     if (isInlineConstant(UseMI, *Src0, *ImmOp))
       return false;
 
-    bool IsF32 = Opc == AMDGPU::V_MAD_F32 || Opc == AMDGPU::V_MAC_F32_e64;
+    bool IsF32 = Opc == AMDGPU::V_MAD_F32 || Opc == AMDGPU::V_MAC_F32_e64 ||
+                 Opc == AMDGPU::V_FMA_F32 || Opc == AMDGPU::V_FMAC_F32_e64;
+    bool IsFMA = Opc == AMDGPU::V_FMA_F32 || Opc == AMDGPU::V_FMAC_F32_e64 ||
+                 Opc == AMDGPU::V_FMA_F16 || Opc == AMDGPU::V_FMAC_F16_e64;
     MachineOperand *Src1 = getNamedOperand(UseMI, AMDGPU::OpName::src1);
     MachineOperand *Src2 = getNamedOperand(UseMI, AMDGPU::OpName::src2);
 
     // Multiplied part is the constant: Use v_madmk_{f16, f32}.
     // We should only expect these to be on src0 due to canonicalizations.
     if (Src0->isReg() && Src0->getReg() == Reg) {
       if (!Src1->isReg() || RI.isSGPRClass(MRI->getRegClass(Src1->getReg())))
         return false;
 
       if (!Src2->isReg() || RI.isSGPRClass(MRI->getRegClass(Src2->getReg())))
         return false;
 
+      unsigned NewOpc =
+        IsFMA ? (IsF32 ? AMDGPU::V_FMAMK_F32 : AMDGPU::V_FMAMK_F16)
+              : (IsF32 ? AMDGPU::V_MADMK_F32 : AMDGPU::V_MADMK_F16);
+      if (pseudoToMCOpcode(NewOpc) == -1)
+        return false;
+
       // We need to swap operands 0 and 1 since madmk constant is at operand 1.
 
       const int64_t Imm = ImmOp->getImm();
 
       // FIXME: This would be a lot easier if we could return a new instruction
       // instead of having to modify in place.
 
       // Remove these first since they are at the end.
       UseMI.RemoveOperand(
           AMDGPU::getNamedOperandIdx(Opc, AMDGPU::OpName::omod));
       UseMI.RemoveOperand(
           AMDGPU::getNamedOperandIdx(Opc, AMDGPU::OpName::clamp));
 
       unsigned Src1Reg = Src1->getReg();
       unsigned Src1SubReg = Src1->getSubReg();
       Src0->setReg(Src1Reg);
       Src0->setSubReg(Src1SubReg);
       Src0->setIsKill(Src1->isKill());
 
       if (Opc == AMDGPU::V_MAC_F32_e64 ||
-          Opc == AMDGPU::V_MAC_F16_e64)
+          Opc == AMDGPU::V_MAC_F16_e64 ||
+          Opc == AMDGPU::V_FMAC_F32_e64 ||
+          Opc == AMDGPU::V_FMAC_F16_e64)
         UseMI.untieRegOperand(
             AMDGPU::getNamedOperandIdx(Opc, AMDGPU::OpName::src2));
 
       Src1->ChangeToImmediate(Imm);
 
       removeModOperands(UseMI);
-      UseMI.setDesc(get(IsF32 ? AMDGPU::V_MADMK_F32 : AMDGPU::V_MADMK_F16));
+      UseMI.setDesc(get(NewOpc));
 
       bool DeleteDef = MRI->hasOneNonDBGUse(Reg);
       if (DeleteDef)
         DefMI.eraseFromParent();
 
       return true;
     }
 
if (Src2->isReg() && Src2->getReg() == Reg) {
         } else if ((RI.isPhysicalRegister(Src1->getReg()) &&
             RI.isSGPRClass(RI.getPhysRegClass(Src1->getReg()))) ||
             (RI.isVirtualRegister(Src1->getReg()) &&
             RI.isSGPRClass(MRI->getRegClass(Src1->getReg()))))
           return false;
           // VGPR is okay as Src1 - fallthrough
       }
 
+      unsigned NewOpc =
+        IsFMA ? (IsF32 ? AMDGPU::V_FMAAK_F32 : AMDGPU::V_FMAAK_F16)
+              : (IsF32 ? AMDGPU::V_MADAK_F32 : AMDGPU::V_MADAK_F16);
+      if (pseudoToMCOpcode(NewOpc) == -1)
+        return false;
+
       const int64_t Imm = ImmOp->getImm();
 
       // FIXME: This would be a lot easier if we could return a new instruction
       // instead of having to modify in place.
 
       // Remove these first since they are at the end.
       UseMI.RemoveOperand(
           AMDGPU::getNamedOperandIdx(Opc, AMDGPU::OpName::omod));
       UseMI.RemoveOperand(
           AMDGPU::getNamedOperandIdx(Opc, AMDGPU::OpName::clamp));
 
       if (Opc == AMDGPU::V_MAC_F32_e64 ||
-          Opc == AMDGPU::V_MAC_F16_e64)
+          Opc == AMDGPU::V_MAC_F16_e64 ||
+          Opc == AMDGPU::V_FMAC_F32_e64 ||
+          Opc == AMDGPU::V_FMAC_F16_e64)
         UseMI.untieRegOperand(
             AMDGPU::getNamedOperandIdx(Opc, AMDGPU::OpName::src2));
 
       // ChangingToImmediate adds Src2 back to the instruction.
       Src2->ChangeToImmediate(Imm);
 
       // These come before src2.
       removeModOperands(UseMI);
-      UseMI.setDesc(get(IsF32 ? AMDGPU::V_MADAK_F32 : AMDGPU::V_MADAK_F16));
+      UseMI.setDesc(get(NewOpc));
 
       bool DeleteDef = MRI->hasOneNonDBGUse(Reg);
       if (DeleteDef)
         DefMI.eraseFromParent();
 
       return true;
     }
   }
static int64_t getFoldableImm(const MachineOperand* MO) {
   return AMDGPU::NoRegister;
 }
 
 MachineInstr *SIInstrInfo::convertToThreeAddress(MachineFunction::iterator &MBB,
                                                  MachineInstr &MI,
                                                  LiveVariables *LV) const {
   unsigned Opc = MI.getOpcode();
   bool IsF16 = false;
-  bool IsFMA = Opc == AMDGPU::V_FMAC_F32_e32 || Opc == AMDGPU::V_FMAC_F32_e64;
+  bool IsFMA = Opc == AMDGPU::V_FMAC_F32_e32 || Opc == AMDGPU::V_FMAC_F32_e64 ||
+               Opc == AMDGPU::V_FMAC_F16_e32 || Opc == AMDGPU::V_FMAC_F16_e64;
 
   switch (Opc) {
   default:
     return nullptr;
   case AMDGPU::V_MAC_F16_e64:
+  case AMDGPU::V_FMAC_F16_e64:
     IsF16 = true;
     LLVM_FALLTHROUGH;
   case AMDGPU::V_MAC_F32_e64:
   case AMDGPU::V_FMAC_F32_e64:
     break;
   case AMDGPU::V_MAC_F16_e32:
+  case AMDGPU::V_FMAC_F16_e32:
     IsF16 = true;
     LLVM_FALLTHROUGH;
   case AMDGPU::V_MAC_F32_e32:
   case AMDGPU::V_FMAC_F32_e32: {
     int Src0Idx = AMDGPU::getNamedOperandIdx(MI.getOpcode(),
                                              AMDGPU::OpName::src0);
     const MachineOperand *Src0 = &MI.getOperand(Src0Idx);
     if (!Src0->isReg() && !Src0->isImm())
const MachineOperand *Src0Mods =
     getNamedOperand(MI, AMDGPU::OpName::src0_modifiers);
   const MachineOperand *Src1 = getNamedOperand(MI, AMDGPU::OpName::src1);
   const MachineOperand *Src1Mods =
     getNamedOperand(MI, AMDGPU::OpName::src1_modifiers);
   const MachineOperand *Src2 = getNamedOperand(MI, AMDGPU::OpName::src2);
   const MachineOperand *Clamp = getNamedOperand(MI, AMDGPU::OpName::clamp);
   const MachineOperand *Omod = getNamedOperand(MI, AMDGPU::OpName::omod);
 
-  if (!IsFMA && !Src0Mods && !Src1Mods && !Clamp && !Omod &&
+  if (!Src0Mods && !Src1Mods && !Clamp && !Omod &&
       // If we have an SGPR input, we will violate the constant bus restriction.
       (ST.getConstantBusLimit(Opc) > 1 ||
        !Src0->isReg() ||
        !RI.isSGPRReg(MBB->getParent()->getRegInfo(), Src0->getReg()))) {
     if (auto Imm = getFoldableImm(Src2)) {
-      return BuildMI(*MBB, MI, MI.getDebugLoc(),
-                     get(IsF16 ? AMDGPU::V_MADAK_F16 : AMDGPU::V_MADAK_F32))
+      unsigned NewOpc =
+         IsFMA ? (IsF16 ? AMDGPU::V_FMAAK_F16 : AMDGPU::V_FMAAK_F32)
+               : (IsF16 ? AMDGPU::V_MADAK_F16 : AMDGPU::V_MADAK_F32);
+      if (pseudoToMCOpcode(NewOpc) != -1)
+        return BuildMI(*MBB, MI, MI.getDebugLoc(), get(NewOpc))
                  .add(*Dst)
                  .add(*Src0)
                  .add(*Src1)
                  .addImm(Imm);
     }
+    unsigned NewOpc =
+      IsFMA ? (IsF16 ? AMDGPU::V_FMAMK_F16 : AMDGPU::V_FMAMK_F32)
+            : (IsF16 ? AMDGPU::V_MADMK_F16 : AMDGPU::V_MADMK_F32);
     if (auto Imm = getFoldableImm(Src1)) {
-      return BuildMI(*MBB, MI, MI.getDebugLoc(),
-                     get(IsF16 ? AMDGPU::V_MADMK_F16 : AMDGPU::V_MADMK_F32))
+      if (pseudoToMCOpcode(NewOpc) != -1)
+        return BuildMI(*MBB, MI, MI.getDebugLoc(), get(NewOpc))
                  .add(*Dst)
                  .add(*Src0)
                  .addImm(Imm)
                  .add(*Src2);
     }
     if (auto Imm = getFoldableImm(Src0)) {
-      if (isOperandLegal(MI, AMDGPU::getNamedOperandIdx(AMDGPU::V_MADMK_F32,
+      if (pseudoToMCOpcode(NewOpc) != -1 &&
+          isOperandLegal(MI, AMDGPU::getNamedOperandIdx(NewOpc,
                            AMDGPU::OpName::src0), Src1))
-        return BuildMI(*MBB, MI, MI.getDebugLoc(),
-                       get(IsF16 ? AMDGPU::V_MADMK_F16 : AMDGPU::V_MADMK_F32))
+        return BuildMI(*MBB, MI, MI.getDebugLoc(), get(NewOpc))
                  .add(*Dst)
                  .add(*Src1)
                  .addImm(Imm)
                  .add(*Src2);
     }
   }
 
-  assert((!IsFMA || !IsF16) && "fmac only expected with f32");
-  unsigned NewOpc = IsFMA ? AMDGPU::V_FMA_F32 :
-    (IsF16 ? AMDGPU::V_MAD_F16 : AMDGPU::V_MAD_F32);
+  unsigned NewOpc = IsFMA ? (IsF16 ? AMDGPU::V_FMA_F16 : AMDGPU::V_FMA_F32)
+                          : (IsF16 ? AMDGPU::V_MAD_F16 : AMDGPU::V_MAD_F32);
+  if (pseudoToMCOpcode(NewOpc) == -1)
+    return nullptr;
+
   return BuildMI(*MBB, MI, MI.getDebugLoc(), get(NewOpc))
       .add(*Dst)
       .addImm(Src0Mods ? Src0Mods->getImm() : 0)
       .add(*Src0)
       .addImm(Src1Mods ? Src1Mods->getImm() : 0)
       .add(*Src1)
       .addImm(0) // Src mods
       .add(*Src2)
switch (MI.getOpcode()) {
         if (!Src1->isReg() || !RI.isVGPR(MRI, Src1->getReg()))
           return false;
         // Additional verification is needed for sdst/src2.
         return true;
       }
       case AMDGPU::V_MAC_F32_e64:
       case AMDGPU::V_MAC_F16_e64:
       case AMDGPU::V_FMAC_F32_e64:
+      case AMDGPU::V_FMAC_F16_e64:
         if (!Src2->isReg() || !RI.isVGPR(MRI, Src2->getReg()) ||
             hasModifiersSet(MI, AMDGPU::OpName::src2_modifiers))
           return false;
         break;
 
       case AMDGPU::V_CNDMASK_B32_e64:
         break;
     }
const TargetRegisterClass *SIInstrInfo::getOpRegClass(const MachineInstr &MI,
   return RI.getRegClass(RCID);
 }
 
 void SIInstrInfo::legalizeOpWithMove(MachineInstr &MI, unsigned OpIdx) const {
   MachineBasicBlock::iterator I = MI;
   MachineBasicBlock *MBB = MI.getParent();
   MachineOperand &MO = MI.getOperand(OpIdx);
   MachineRegisterInfo &MRI = MBB->getParent()->getRegInfo();
+  const SIRegisterInfo *TRI =
+      static_cast<const SIRegisterInfo*>(MRI.getTargetRegisterInfo());
   unsigned RCID = get(MI.getOpcode()).OpInfo[OpIdx].RegClass;
   const TargetRegisterClass *RC = RI.getRegClass(RCID);
-  unsigned Opcode = AMDGPU::V_MOV_B32_e32;
+  unsigned Size = TRI->getRegSizeInBits(*RC);
+  unsigned Opcode = (Size == 64) ? AMDGPU::V_MOV_B64_PSEUDO : AMDGPU::V_MOV_B32_e32;
   if (MO.isReg())
     Opcode = AMDGPU::COPY;
   else if (RI.isSGPRClass(RC))
-    Opcode = AMDGPU::S_MOV_B32;
+    Opcode = (Size == 64) ? AMDGPU::S_MOV_B64 : AMDGPU::S_MOV_B32;
 
   const TargetRegisterClass *VRC = RI.getEquivalentVGPRClass(RC);
   if (RI.getCommonSubClass(&AMDGPU::VReg_64RegClass, VRC))
     VRC = &AMDGPU::VReg_64RegClass;
   else
     VRC = &AMDGPU::VGPR_32RegClass;
 
   unsigned Reg = MRI.createVirtualRegister(VRC);
MachineOperand *SIInstrInfo::getNamedOperand(MachineInstr &MI,
   int Idx = AMDGPU::getNamedOperandIdx(MI.getOpcode(), OperandName);
   if (Idx == -1)
     return nullptr;
 
   return &MI.getOperand(Idx);
 }
 
 uint64_t SIInstrInfo::getDefaultRsrcDataFormat() const {
+  if (ST.getGeneration() >= AMDGPUSubtarget::GFX10) {
+    return (16ULL << 44) | // IMG_FORMAT_32_FLOAT
+           (1ULL << 56) | // RESOURCE_LEVEL = 1
+           (3ULL << 60); // OOB_SELECT = 3
+  }
+
   uint64_t RsrcDataFormat = AMDGPU::RSRC_DATA_FORMAT;
   if (ST.isAmdHsaOS()) {
     // Set ATC = 1. GFX9 doesn't have this bit.
     if (ST.getGeneration() <= AMDGPUSubtarget::VOLCANIC_ISLANDS)
       RsrcDataFormat |= (1ULL << 56);
 
     // Set MTYPE = 2 (MTYPE_UC = uncached). GFX9 doesn't have this.
     // BTW, it disables TC L2 and therefore decreases performance.
uint64_t Rsrc23 = getDefaultRsrcDataFormat() |
                     0xffffffff; // Size;
 
   // GFX9 doesn't have ELEMENT_SIZE.
   if (ST.getGeneration() <= AMDGPUSubtarget::VOLCANIC_ISLANDS) {
     uint64_t EltSizeValue = Log2_32(ST.getMaxPrivateElementSize()) - 1;
     Rsrc23 |= EltSizeValue << AMDGPU::RSRC_ELEMENT_SIZE_SHIFT;
   }
 
-  // IndexStride = 64.
-  Rsrc23 |= UINT64_C(3) << AMDGPU::RSRC_INDEX_STRIDE_SHIFT;
+  // IndexStride = 64 / 32.
+  uint64_t IndexStride = ST.getGeneration() <= AMDGPUSubtarget::GFX9 ? 3 : 2;
+  Rsrc23 |= IndexStride << AMDGPU::RSRC_INDEX_STRIDE_SHIFT;
 
   // If TID_ENABLE is set, DATA_FORMAT specifies stride bits [14:17].
   // Clear them unless we want a huge stride.
-  if (ST.getGeneration() >= AMDGPUSubtarget::VOLCANIC_ISLANDS)
+  if (ST.getGeneration() >= AMDGPUSubtarget::VOLCANIC_ISLANDS &&
+      ST.getGeneration() <= AMDGPUSubtarget::GFX9)
     Rsrc23 &= ~AMDGPU::RSRC_DATA_FORMAT;
 
   return Rsrc23;
 }
 
 bool SIInstrInfo::isLowLatencyInstruction(const MachineInstr &MI) const {
   unsigned Opc = MI.getOpcode();
 

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

 
 def : GCNPat<
   (fcanonicalize (f16 (fneg (VOP3Mods f16:$src, i32:$src_mods)))),
   (V_MUL_F16_e64 0, (i32 CONST.FP16_NEG_ONE), $src_mods, $src, 0, 0)
 >;
 
 def : GCNPat<
   (fcanonicalize (v2f16 (VOP3PMods v2f16:$src, i32:$src_mods))),
-  (V_PK_MUL_F16 0, (i32 CONST.V2FP16_ONE), $src_mods, $src, DSTCLAMP.NONE)
+  (V_PK_MUL_F16 0, (i32 CONST.FP16_ONE), $src_mods, $src, DSTCLAMP.NONE)
 >;
 }
 
 let OtherPredicates = [FP16Denormals] in {
 def : GCNPat<
   (fcanonicalize (f16 (VOP3Mods f16:$src, i32:$src_mods))),
   (V_MAX_F16_e64 $src_mods, $src, $src_mods, $src, 0, 0)
 >;
def : GCNPat <
   (fma (f32 (VOP3Mods0 f32:$src0, i32:$src0_modifiers, i1:$clamp, i32:$omod)),
        (f32 (VOP3Mods f32:$src1, i32:$src1_modifiers)),
        (f32 (VOP3NoMods f32:$src2))),
   (V_FMAC_F32_e64 $src0_modifiers, $src0, $src1_modifiers, $src1,
                   SRCMODS.NONE, $src2, $clamp, $omod)
 >;
 } // End OtherPredicates = [HasDLInsts]
 
+let SubtargetPredicate = isGFX10Plus in
+def : GCNPat <
+  (fma (f16 (VOP3Mods0 f32:$src0, i32:$src0_modifiers, i1:$clamp, i32:$omod)),
+       (f16 (VOP3Mods f32:$src1, i32:$src1_modifiers)),
+       (f16 (VOP3NoMods f32:$src2))),
+  (V_FMAC_F16_e64 $src0_modifiers, $src0, $src1_modifiers, $src1,
+                  SRCMODS.NONE, $src2, $clamp, $omod)
+>;
 
 // Allow integer inputs
 class ExpPattern<SDPatternOperator node, ValueType vt, Instruction Inst> : GCNPat<
   (node (i8 timm:$tgt), (i8 timm:$en), vt:$src0, vt:$src1, vt:$src2, vt:$src3, (i1 timm:$compr), (i1 timm:$vm)),
   (Inst i8:$tgt, vt:$src0, vt:$src1, vt:$src2, vt:$src3, i1:$vm, i1:$compr, i8:$en)
 >;
 
 def : ExpPattern<AMDGPUexport, i32, EXP>;

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

if (!Src ||
         } else {
           // Not legal to convert this src
           return false;
         }
       }
     }
     assert(Src && Src->isReg());
 
-    if ((MI.getOpcode() == AMDGPU::V_MAC_F16_sdwa ||
+    if ((MI.getOpcode() == AMDGPU::V_FMAC_F16_sdwa ||
+         MI.getOpcode() == AMDGPU::V_FMAC_F32_sdwa ||
+         MI.getOpcode() == AMDGPU::V_MAC_F16_sdwa ||
          MI.getOpcode() == AMDGPU::V_MAC_F32_sdwa) &&
          !isSameReg(*Src, *getReplacedOperand())) {
       // In case of v_mac_f16/32_sdwa this pass can try to apply src operand to
       // src2. This is not allowed.
       return false;
     }
 
     assert(isSameReg(*Src, *getReplacedOperand()) &&
MachineInstr *SDWADstOperand::potentialToConvert(const SIInstrInfo *TII) {
   }
 
   return PotentialMO->getParent();
 }
 
 bool SDWADstOperand::convertToSDWA(MachineInstr &MI, const SIInstrInfo *TII) {
   // Replace vdst operand in MI with target operand. Set dst_sel and dst_unused
 
-  if ((MI.getOpcode() == AMDGPU::V_MAC_F16_sdwa ||
+  if ((MI.getOpcode() == AMDGPU::V_FMAC_F16_sdwa ||
+       MI.getOpcode() == AMDGPU::V_FMAC_F32_sdwa ||
+       MI.getOpcode() == AMDGPU::V_MAC_F16_sdwa ||
        MI.getOpcode() == AMDGPU::V_MAC_F32_sdwa) &&
       getDstSel() != AMDGPU::SDWA::DWORD) {
     // v_mac_f16/32_sdwa allow dst_sel to be equal only to DWORD
     return false;
   }
 
   MachineOperand *Operand = TII->getNamedOperand(MI, AMDGPU::OpName::vdst);
   assert(Operand &&
if (!ST.hasSDWAOutModsVOPC() &&
          TII->hasModifiersSet(MI, AMDGPU::OpName::omod)))
       return false;
 
   } else if (TII->getNamedOperand(MI, AMDGPU::OpName::sdst) ||
              !TII->getNamedOperand(MI, AMDGPU::OpName::vdst)) {
     return false;
   }
 
-  if (!ST.hasSDWAMac() && (Opc == AMDGPU::V_MAC_F16_e32 ||
+  if (!ST.hasSDWAMac() && (Opc == AMDGPU::V_FMAC_F16_e32 ||
+                           Opc == AMDGPU::V_FMAC_F32_e32 ||
+                           Opc == AMDGPU::V_MAC_F16_e32 ||
                            Opc == AMDGPU::V_MAC_F32_e32))
     return false;
 
+  // Check if target supports this SDWA opcode
+  if (TII->pseudoToMCOpcode(Opc) == -1)
+    return false;
+
   // FIXME: has SDWA but require handling of implicit VCC use
   if (Opc == AMDGPU::V_CNDMASK_B32_e32)
     return false;
 
   return true;
 }
 
 bool SIPeepholeSDWA::convertToSDWA(MachineInstr &MI,
assert(
       AMDGPU::getNamedOperandIdx(SDWAOpcode, AMDGPU::OpName::src1_modifiers) != -1);
     if (auto *Mod = TII->getNamedOperand(MI, AMDGPU::OpName::src1_modifiers))
       SDWAInst.addImm(Mod->getImm());
     else
       SDWAInst.addImm(0);
     SDWAInst.add(*Src1);
   }
 
-  if (SDWAOpcode == AMDGPU::V_MAC_F16_sdwa ||
+  if (SDWAOpcode == AMDGPU::V_FMAC_F16_sdwa ||
+      SDWAOpcode == AMDGPU::V_FMAC_F32_sdwa ||
+      SDWAOpcode == AMDGPU::V_MAC_F16_sdwa ||
       SDWAOpcode == AMDGPU::V_MAC_F32_sdwa) {
     // v_mac_f16/32 has additional src2 operand tied to vdst
     MachineOperand *Src2 = TII->getNamedOperand(MI, AMDGPU::OpName::src2);
     assert(Src2);
     SDWAInst.add(*Src2);
   }
 
   // Copy clamp if present, initialize otherwise

llvm/trunk/test/CodeGen/AMDGPU/fdiv.f16.ll

 ; RUN: llc -march=amdgcn -verify-machineinstrs < %s | FileCheck -check-prefix=GCN -check-prefix=SI %s
-; RUN: llc -march=amdgcn -mcpu=fiji -mattr=-flat-for-global -mattr=+fp16-denormals -verify-machineinstrs < %s | FileCheck -check-prefix=GCN -check-prefix=GFX8_9 %s
-; RUN: llc -march=amdgcn -mcpu=fiji -mattr=-flat-for-global -mattr=-fp16-denormals -verify-machineinstrs < %s | FileCheck -check-prefix=GCN -check-prefix=GFX8_9 %s
-; RUN: llc -march=amdgcn -mcpu=gfx900 -mattr=-flat-for-global -verify-machineinstrs < %s | FileCheck -check-prefix=GCN -check-prefix=GFX8_9 %s
+; RUN: llc -march=amdgcn -mcpu=fiji -mattr=-flat-for-global -mattr=+fp16-denormals -verify-machineinstrs < %s | FileCheck -check-prefix=GCN -check-prefix=GFX8_9_10 %s
+; RUN: llc -march=amdgcn -mcpu=fiji -mattr=-flat-for-global -mattr=-fp16-denormals -verify-machineinstrs < %s | FileCheck -check-prefix=GCN -check-prefix=GFX8_9_10 %s
+; RUN: llc -march=amdgcn -mcpu=gfx900 -mattr=-flat-for-global -verify-machineinstrs < %s | FileCheck -check-prefix=GCN -check-prefix=GFX8_9_10 %s
+; RUN: llc -march=amdgcn -mcpu=gfx1010 -mattr=-flat-for-global -verify-machineinstrs < %s | FileCheck -check-prefix=GCN -check-prefix=GFX8_9_10 %s
 
 ; Make sure fdiv is promoted to f32.
 
 ; GCN-LABEL: {{^}}v_fdiv_f16
 ; SI:     v_cvt_f32_f16
 ; SI:     v_cvt_f32_f16
 ; SI:     v_div_scale_f32
 ; SI-DAG: v_div_scale_f32
 ; SI-DAG: v_rcp_f32
 ; SI:     v_fma_f32
 ; SI:     v_fma_f32
 ; SI:     v_mul_f32
 ; SI:     v_fma_f32
 ; SI:     v_fma_f32
 ; SI:     v_fma_f32
 ; SI:     v_div_fmas_f32
 ; SI:     v_div_fixup_f32
 ; SI:     v_cvt_f16_f32
 
-; GFX8_9: {{flat|global}}_load_ushort [[LHS:v[0-9]+]]
-; GFX8_9: {{flat|global}}_load_ushort [[RHS:v[0-9]+]]
+; GFX8_9_10: {{flat|global}}_load_ushort [[LHS:v[0-9]+]]
+; GFX8_9_10: {{flat|global}}_load_ushort [[RHS:v[0-9]+]]
 
-; GFX8_9-DAG: v_cvt_f32_f16_e32 [[CVT_LHS:v[0-9]+]], [[LHS]]
-; GFX8_9-DAG: v_cvt_f32_f16_e32 [[CVT_RHS:v[0-9]+]], [[RHS]]
+; GFX8_9_10-DAG: v_cvt_f32_f16_e32 [[CVT_LHS:v[0-9]+]], [[LHS]]
+; GFX8_9_10-DAG: v_cvt_f32_f16_e32 [[CVT_RHS:v[0-9]+]], [[RHS]]
 
-; GFX8_9-DAG: v_rcp_f32_e32 [[RCP_RHS:v[0-9]+]], [[CVT_RHS]]
-; GFX8_9: v_mul_f32_e32 [[MUL:v[0-9]+]], [[CVT_LHS]], [[RCP_RHS]]
-; GFX8_9: v_cvt_f16_f32_e32 [[CVT_BACK:v[0-9]+]], [[MUL]]
-; GFX8_9: v_div_fixup_f16 [[RESULT:v[0-9]+]], [[CVT_BACK]], [[RHS]], [[LHS]]
-; GFX8_9: {{flat|global}}_store_short v{{\[[0-9]+:[0-9]+\]}}, [[RESULT]]
+; GFX8_9_10-DAG: v_rcp_f32_e32 [[RCP_RHS:v[0-9]+]], [[CVT_RHS]]
+; GFX8_9_10: v_mul_f32_e32 [[MUL:v[0-9]+]], [[CVT_LHS]], [[RCP_RHS]]
+; GFX8_9_10: v_cvt_f16_f32_e32 [[CVT_BACK:v[0-9]+]], [[MUL]]
+; GFX8_9_10: v_div_fixup_f16 [[RESULT:v[0-9]+]], [[CVT_BACK]], [[RHS]], [[LHS]]
+; GFX8_9_10: {{flat|global}}_store_short v{{\[[0-9]+:[0-9]+\]}}, [[RESULT]]
 define amdgpu_kernel void @v_fdiv_f16(
     half addrspace(1)* %r,
     half addrspace(1)* %a,
     half addrspace(1)* %b) #0 {
 entry:
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %tid.ext = sext i32 %tid to i64
   %gep.a = getelementptr inbounds half, half addrspace(1)* %a, i64 %tid.ext
   %gep.b = getelementptr inbounds half, half addrspace(1)* %b, i64 %tid.ext
   %gep.r = getelementptr inbounds half, half addrspace(1)* %r, i64 %tid.ext
   %a.val = load volatile half, half addrspace(1)* %gep.a
   %b.val = load volatile half, half addrspace(1)* %gep.b
   %r.val = fdiv half %a.val, %b.val
   store half %r.val, half addrspace(1)* %gep.r
   ret void
 }
 
 ; GCN-LABEL: {{^}}v_rcp_f16:
-; GFX8_9: {{flat|global}}_load_ushort [[VAL:v[0-9]+]]
-; GFX8_9-NOT: [[VAL]]
-; GFX8_9: v_rcp_f16_e32 [[RESULT:v[0-9]+]], [[VAL]]
-; GFX8_9-NOT: [[RESULT]]
-; GFX8_9: {{flat|global}}_store_short v{{\[[0-9]+:[0-9]+\]}}, [[RESULT]]
+; GFX8_9_10: {{flat|global}}_load_ushort [[VAL:v[0-9]+]]
+; GFX8_9_10-NOT: [[VAL]]
+; GFX8_9_10: v_rcp_f16_e32 [[RESULT:v[0-9]+]], [[VAL]]
+; GFX8_9_10-NOT: [[RESULT]]
+; GFX8_9_10: {{flat|global}}_store_short v{{\[[0-9]+:[0-9]+\]}}, [[RESULT]]
 define amdgpu_kernel void @v_rcp_f16(half addrspace(1)* %r, half addrspace(1)* %b) #0 {
 entry:
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %tid.ext = sext i32 %tid to i64
   %gep.b = getelementptr inbounds half, half addrspace(1)* %b, i64 %tid.ext
   %gep.r = getelementptr inbounds half, half addrspace(1)* %r, i64 %tid.ext
   %b.val = load volatile half, half addrspace(1)* %gep.b
   %r.val = fdiv half 1.0, %b.val
   store half %r.val, half addrspace(1)* %gep.r
   ret void
 }
 
 ; GCN-LABEL: {{^}}v_rcp_f16_abs:
-; GFX8_9: {{flat|global}}_load_ushort [[VAL:v[0-9]+]]
-; GFX8_9-NOT: [[VAL]]
-; GFX8_9: v_rcp_f16_e64 [[RESULT:v[0-9]+]], |[[VAL]]|
-; GFX8_9-NOT: [RESULT]]
-; GFX8_9: {{flat|global}}_store_short v{{\[[0-9]+:[0-9]+\]}}, [[RESULT]]
+; GFX8_9_10: {{flat|global}}_load_ushort [[VAL:v[0-9]+]]
+; GFX8_9_10-NOT: [[VAL]]
+; GFX8_9_10: v_rcp_f16_e64 [[RESULT:v[0-9]+]], |[[VAL]]|
+; GFX8_9_10-NOT: [RESULT]]
+; GFX8_9_10: {{flat|global}}_store_short v{{\[[0-9]+:[0-9]+\]}}, [[RESULT]]
 define amdgpu_kernel void @v_rcp_f16_abs(half addrspace(1)* %r, half addrspace(1)* %b) #0 {
 entry:
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %tid.ext = sext i32 %tid to i64
   %gep.b = getelementptr inbounds half, half addrspace(1)* %b, i64 %tid.ext
   %gep.r = getelementptr inbounds half, half addrspace(1)* %r, i64 %tid.ext
   %b.val = load volatile half, half addrspace(1)* %gep.b
   %b.abs = call half @llvm.fabs.f16(half %b.val)
   %r.val = fdiv half 1.0, %b.abs
   store half %r.val, half addrspace(1)* %gep.r
   ret void
 }
 
 ; GCN-LABEL: {{^}}v_rcp_f16_arcp:
-; GFX8_9: {{flat|global}}_load_ushort [[VAL:v[0-9]+]]
-; GFX8_9-NOT: [[VAL]]
-; GFX8_9: v_rcp_f16_e32 [[RESULT:v[0-9]+]], [[VAL]]
-; GFX8_9-NOT: [[RESULT]]
-; GFX8_9: {{flat|global}}_store_short v{{\[[0-9]+:[0-9]+\]}}, [[RESULT]]
+; GFX8_9_10: {{flat|global}}_load_ushort [[VAL:v[0-9]+]]
+; GFX8_9_10-NOT: [[VAL]]
+; GFX8_9_10: v_rcp_f16_e32 [[RESULT:v[0-9]+]], [[VAL]]
+; GFX8_9_10-NOT: [[RESULT]]
+; GFX8_9_10: {{flat|global}}_store_short v{{\[[0-9]+:[0-9]+\]}}, [[RESULT]]
 define amdgpu_kernel void @v_rcp_f16_arcp(half addrspace(1)* %r, half addrspace(1)* %b) #0 {
 entry:
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %tid.ext = sext i32 %tid to i64
   %gep.b = getelementptr inbounds half, half addrspace(1)* %b, i64 %tid.ext
   %gep.r = getelementptr inbounds half, half addrspace(1)* %r, i64 %tid.ext
   %b.val = load volatile half, half addrspace(1)* %gep.b
   %r.val = fdiv arcp half 1.0, %b.val
   store half %r.val, half addrspace(1)* %gep.r
   ret void
 }
 
 ; GCN-LABEL: {{^}}v_rcp_f16_neg:
-; GFX8_9: {{flat|global}}_load_ushort [[VAL:v[0-9]+]]
-; GFX8_9-NOT: [[VAL]]
-; GFX8_9: v_rcp_f16_e64 [[RESULT:v[0-9]+]], -[[VAL]]
-; GFX8_9-NOT: [RESULT]]
-; GFX8_9: {{flat|global}}_store_short v{{\[[0-9]+:[0-9]+\]}}, [[RESULT]]
+; GFX8_9_10: {{flat|global}}_load_ushort [[VAL:v[0-9]+]]
+; GFX8_9_10-NOT: [[VAL]]
+; GFX8_9_10: v_rcp_f16_e64 [[RESULT:v[0-9]+]], -[[VAL]]
+; GFX8_9_10-NOT: [RESULT]]
+; GFX8_9_10: {{flat|global}}_store_short v{{\[[0-9]+:[0-9]+\]}}, [[RESULT]]
 define amdgpu_kernel void @v_rcp_f16_neg(half addrspace(1)* %r, half addrspace(1)* %b) #0 {
 entry:
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %tid.ext = sext i32 %tid to i64
   %gep.b = getelementptr inbounds half, half addrspace(1)* %b, i64 %tid.ext
   %gep.r = getelementptr inbounds half, half addrspace(1)* %r, i64 %tid.ext
   %b.val = load volatile half, half addrspace(1)* %gep.b
   %r.val = fdiv half -1.0, %b.val
   store half %r.val, half addrspace(1)* %gep.r
   ret void
 }
 
 ; GCN-LABEL: {{^}}v_rsq_f16:
-; GFX8_9: {{flat|global}}_load_ushort [[VAL:v[0-9]+]]
-; GFX8_9-NOT: [[VAL]]
-; GFX8_9: v_rsq_f16_e32 [[RESULT:v[0-9]+]], [[VAL]]
-; GFX8_9-NOT: [RESULT]]
-; GFX8_9: {{flat|global}}_store_short v{{\[[0-9]+:[0-9]+\]}}, [[RESULT]]
+; GFX8_9_10: {{flat|global}}_load_ushort [[VAL:v[0-9]+]]
+; GFX8_9_10-NOT: [[VAL]]
+; GFX8_9_10: v_rsq_f16_e32 [[RESULT:v[0-9]+]], [[VAL]]
+; GFX8_9_10-NOT: [RESULT]]
+; GFX8_9_10: {{flat|global}}_store_short v{{\[[0-9]+:[0-9]+\]}}, [[RESULT]]
 define amdgpu_kernel void @v_rsq_f16(half addrspace(1)* %r, half addrspace(1)* %b) #0 {
 entry:
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %tid.ext = sext i32 %tid to i64
   %gep.b = getelementptr inbounds half, half addrspace(1)* %b, i64 %tid.ext
   %gep.r = getelementptr inbounds half, half addrspace(1)* %r, i64 %tid.ext
   %b.val = load volatile half, half addrspace(1)* %gep.b
   %b.sqrt = call half @llvm.sqrt.f16(half %b.val)
   %r.val = fdiv half 1.0, %b.sqrt
   store half %r.val, half addrspace(1)* %gep.r
   ret void
 }
 
 ; GCN-LABEL: {{^}}v_rsq_f16_neg:
-; GFX8_9: {{flat|global}}_load_ushort [[VAL:v[0-9]+]]
-; GFX8_9-NOT: [[VAL]]
-; GFX8_9: v_sqrt_f16_e32 [[SQRT:v[0-9]+]], [[VAL]]
-; GFX8_9-NEXT: v_rcp_f16_e64 [[RESULT:v[0-9]+]], -[[SQRT]]
-; GFX8_9-NOT: [RESULT]]
-; GFX8_9: {{flat|global}}_store_short v{{\[[0-9]+:[0-9]+\]}}, [[RESULT]]
+; GFX8_9_10: {{flat|global}}_load_ushort [[VAL:v[0-9]+]]
+; GFX8_9_10-NOT: [[VAL]]
+; GFX8_9_10: v_sqrt_f16_e32 [[SQRT:v[0-9]+]], [[VAL]]
+; GFX8_9_10-NEXT: v_rcp_f16_e64 [[RESULT:v[0-9]+]], -[[SQRT]]
+; GFX8_9_10-NOT: [RESULT]]
+; GFX8_9_10: {{flat|global}}_store_short v{{\[[0-9]+:[0-9]+\]}}, [[RESULT]]
 define amdgpu_kernel void @v_rsq_f16_neg(half addrspace(1)* %r, half addrspace(1)* %b) #0 {
 entry:
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %tid.ext = sext i32 %tid to i64
   %gep.b = getelementptr inbounds half, half addrspace(1)* %b, i64 %tid.ext
   %gep.r = getelementptr inbounds half, half addrspace(1)* %r, i64 %tid.ext
   %b.val = load volatile half, half addrspace(1)* %gep.b
   %b.sqrt = call half @llvm.sqrt.f16(half %b.val)
   %r.val = fdiv half -1.0, %b.sqrt
   store half %r.val, half addrspace(1)* %gep.r
   ret void
 }
 
 ; GCN-LABEL: {{^}}v_fdiv_f16_arcp:
-; GFX8_9: {{flat|global}}_load_ushort [[LHS:v[0-9]+]]
-; GFX8_9: {{flat|global}}_load_ushort [[RHS:v[0-9]+]]
+; GFX8_9_10: {{flat|global}}_load_ushort [[LHS:v[0-9]+]]
+; GFX8_9_10: {{flat|global}}_load_ushort [[RHS:v[0-9]+]]
 
-; GFX8_9: v_rcp_f16_e32 [[RCP:v[0-9]+]], [[RHS]]
-; GFX8_9: v_mul_f16_e32 [[RESULT:v[0-9]+]], [[LHS]], [[RCP]]
+; GFX8_9_10: v_rcp_f16_e32 [[RCP:v[0-9]+]], [[RHS]]
+; GFX8_9_10: v_mul_f16_e32 [[RESULT:v[0-9]+]], [[LHS]], [[RCP]]
 
-; GFX8_9: {{flat|global}}_store_short v{{\[[0-9]+:[0-9]+\]}}, [[RESULT]]
+; GFX8_9_10: {{flat|global}}_store_short v{{\[[0-9]+:[0-9]+\]}}, [[RESULT]]
 define amdgpu_kernel void @v_fdiv_f16_arcp(half addrspace(1)* %r, half addrspace(1)* %a, half addrspace(1)* %b) #0 {
 entry:
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %tid.ext = sext i32 %tid to i64
   %gep.a = getelementptr inbounds half, half addrspace(1)* %a, i64 %tid.ext
   %gep.b = getelementptr inbounds half, half addrspace(1)* %b, i64 %tid.ext
   %gep.r = getelementptr inbounds half, half addrspace(1)* %r, i64 %tid.ext
   %a.val = load volatile half, half addrspace(1)* %gep.a
   %b.val = load volatile half, half addrspace(1)* %gep.b
   %r.val = fdiv arcp half %a.val, %b.val
   store half %r.val, half addrspace(1)* %gep.r
   ret void
 }
 
 ; GCN-LABEL: {{^}}v_fdiv_f16_unsafe:
-; GFX8_9: {{flat|global}}_load_ushort [[LHS:v[0-9]+]]
-; GFX8_9: {{flat|global}}_load_ushort [[RHS:v[0-9]+]]
+; GFX8_9_10: {{flat|global}}_load_ushort [[LHS:v[0-9]+]]
+; GFX8_9_10: {{flat|global}}_load_ushort [[RHS:v[0-9]+]]
 
-; GFX8_9: v_rcp_f16_e32 [[RCP:v[0-9]+]], [[RHS]]
-; GFX8_9: v_mul_f16_e32 [[RESULT:v[0-9]+]], [[LHS]], [[RCP]]
+; GFX8_9_10: v_rcp_f16_e32 [[RCP:v[0-9]+]], [[RHS]]
+; GFX8_9_10: v_mul_f16_e32 [[RESULT:v[0-9]+]], [[LHS]], [[RCP]]
 
-; GFX8_9: {{flat|global}}_store_short v{{\[[0-9]+:[0-9]+\]}}, [[RESULT]]
+; GFX8_9_10: {{flat|global}}_store_short v{{\[[0-9]+:[0-9]+\]}}, [[RESULT]]
 define amdgpu_kernel void @v_fdiv_f16_unsafe(half addrspace(1)* %r, half addrspace(1)* %a, half addrspace(1)* %b) #2 {
 entry:
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %tid.ext = sext i32 %tid to i64
   %gep.a = getelementptr inbounds half, half addrspace(1)* %a, i64 %tid.ext
   %gep.b = getelementptr inbounds half, half addrspace(1)* %b, i64 %tid.ext
   %gep.r = getelementptr inbounds half, half addrspace(1)* %r, i64 %tid.ext
   %a.val = load volatile half, half addrspace(1)* %gep.a
   %b.val = load volatile half, half addrspace(1)* %gep.b
   %r.val = fdiv half %a.val, %b.val
   store half %r.val, half addrspace(1)* %gep.r
   ret void
 }
 
 ; FUNC-LABEL: {{^}}div_arcp_2_x_pat_f16:
 ; SI: v_mul_f32_e32 v{{[0-9]+}}, 0.5, v{{[0-9]+}}
 
-; GFX8_9: v_mul_f16_e32 [[MUL:v[0-9]+]], 0.5, v{{[0-9]+}}
-; GFX8_9: buffer_store_short [[MUL]]
+; GFX8_9_10: v_mul_f16_e32 [[MUL:v[0-9]+]], 0.5, v{{[0-9]+}}
+; GFX8_9_10: buffer_store_short [[MUL]]
 define amdgpu_kernel void @div_arcp_2_x_pat_f16(half addrspace(1)* %out) #0 {
   %x = load half, half addrspace(1)* undef
   %rcp = fdiv arcp half %x, 2.0
   store half %rcp, half addrspace(1)* %out, align 4
   ret void
 }
 
 ; FUNC-LABEL: {{^}}div_arcp_k_x_pat_f16:
 ; SI: v_mul_f32_e32 v{{[0-9]+}}, 0x3dccc000, v{{[0-9]+}}
 
-; GFX8_9: v_mul_f16_e32 [[MUL:v[0-9]+]], 0x2e66, v{{[0-9]+}}
-; GFX8_9: buffer_store_short [[MUL]]
+; GFX8_9_10: v_mul_f16_e32 [[MUL:v[0-9]+]], 0x2e66, v{{[0-9]+}}
+; GFX8_9_10: buffer_store_short [[MUL]]
 define amdgpu_kernel void @div_arcp_k_x_pat_f16(half addrspace(1)* %out) #0 {
   %x = load half, half addrspace(1)* undef
   %rcp = fdiv arcp half %x, 10.0
   store half %rcp, half addrspace(1)* %out, align 4
   ret void
 }
 
 ; FUNC-LABEL: {{^}}div_arcp_neg_k_x_pat_f16:
 ; SI: v_mul_f32_e32 v{{[0-9]+}}, 0xbdccc000, v{{[0-9]+}}
 
-; GFX8_9: v_mul_f16_e32 [[MUL:v[0-9]+]], 0xae66, v{{[0-9]+}}
-; GFX8_9: buffer_store_short [[MUL]]
+; GFX8_9_10: v_mul_f16_e32 [[MUL:v[0-9]+]], 0xae66, v{{[0-9]+}}
+; GFX8_9_10: buffer_store_short [[MUL]]
 define amdgpu_kernel void @div_arcp_neg_k_x_pat_f16(half addrspace(1)* %out) #0 {
   %x = load half, half addrspace(1)* undef
   %rcp = fdiv arcp half %x, -10.0
   store half %rcp, half addrspace(1)* %out, align 4
   ret void
 }
 
 declare i32 @llvm.amdgcn.workitem.id.x() #1
 declare half @llvm.sqrt.f16(half) #1
 declare half @llvm.fabs.f16(half) #1
 
 attributes #0 = { nounwind }
 attributes #1 = { nounwind readnone }
 attributes #2 = { nounwind "unsafe-fp-math"="true" }

llvm/trunk/test/CodeGen/AMDGPU/fmac.sdwa.ll

+; RUN: llc -march=amdgcn -mcpu=gfx1010 -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN,GFX1010 %s
+
+; GCN-LABEL: {{^}}addMul2D:
+; GFX1010: v_fmac_f16
+; GFX1010: v_fmac_f16
+define hidden <4 x half> @addMul2D(<4 x i8>* nocapture readonly, float addrspace(4)* nocapture readonly, <2 x i32>, i32) local_unnamed_addr #0 {
+  %5 = extractelement <2 x i32> %2, i64 1
+  %6 = icmp sgt i32 %5, 0
+  br i1 %6, label %7, label %38
+
+7:                                                ; preds = %4
+  %8 = extractelement <2 x i32> %2, i64 0
+  %9 = icmp sgt i32 %8, 0
+  br label %10
+
+10:                                               ; preds = %34, %7
+  %11 = phi <4 x half> [ zeroinitializer, %7 ], [ %35, %34 ]
+  %12 = phi i32 [ 0, %7 ], [ %36, %34 ]
+  br i1 %9, label %13, label %34
+
+13:                                               ; preds = %10
+  %14 = mul nsw i32 %12, %3
+  %15 = mul nsw i32 %12, %8
+  br label %16
+
+16:                                               ; preds = %16, %13
+  %17 = phi <4 x half> [ %11, %13 ], [ %31, %16 ]
+  %18 = phi i32 [ 0, %13 ], [ %32, %16 ]
+  %19 = add nsw i32 %18, %14
+  %20 = sext i32 %19 to i64
+  %21 = getelementptr inbounds <4 x i8>, <4 x i8>* %0, i64 %20
+  %22 = load <4 x i8>, <4 x i8>* %21, align 4
+  %23 = tail call <4 x half> @_Z13convert_half4Dv4_h(<4 x i8> %22) #8
+  %24 = add nsw i32 %18, %15
+  %25 = sext i32 %24 to i64
+  %26 = getelementptr inbounds float, float addrspace(4)* %1, i64 %25
+  %27 = load float, float addrspace(4)* %26, align 4
+  %28 = fptrunc float %27 to half
+  %29 = insertelement <4 x half> undef, half %28, i32 0
+  %30 = shufflevector <4 x half> %29, <4 x half> undef, <4 x i32> zeroinitializer
+  %31 = tail call <4 x half> @llvm.fmuladd.v4f16(<4 x half> %23, <4 x half> %30, <4 x half> %17)
+  %32 = add nuw nsw i32 %18, 1
+  %33 = icmp eq i32 %32, %8
+  br i1 %33, label %34, label %16
+
+34:                                               ; preds = %16, %10
+  %35 = phi <4 x half> [ %11, %10 ], [ %31, %16 ]
+  %36 = add nuw nsw i32 %12, 1
+  %37 = icmp eq i32 %36, %5
+  br i1 %37, label %38, label %10
+
+38:                                               ; preds = %34, %4
+  %39 = phi <4 x half> [ zeroinitializer, %4 ], [ %35, %34 ]
+  ret <4 x half> %39
+}
+
+define linkonce_odr hidden <4 x half> @_Z13convert_half4Dv4_h(<4 x i8>) local_unnamed_addr #1 {
+  %2 = extractelement <4 x i8> %0, i64 0
+  %3 = uitofp i8 %2 to half
+  %4 = insertelement <4 x half> undef, half %3, i32 0
+  %5 = extractelement <4 x i8> %0, i64 1
+  %6 = uitofp i8 %5 to half
+  %7 = insertelement <4 x half> %4, half %6, i32 1
+  %8 = extractelement <4 x i8> %0, i64 2
+  %9 = uitofp i8 %8 to half
+  %10 = insertelement <4 x half> %7, half %9, i32 2
+  %11 = extractelement <4 x i8> %0, i64 3
+  %12 = uitofp i8 %11 to half
+  %13 = insertelement <4 x half> %10, half %12, i32 3
+  ret <4 x half> %13
+}
+
+declare <4 x half> @llvm.fmuladd.v4f16(<4 x half>, <4 x half>, <4 x half>)
+
+attributes #0 = { convergent nounwind readonly "correctly-rounded-divide-sqrt-fp-math"="false" "denorms-are-zero"="false" "disable-tail-calls"="false" "less-precise-fpmad"="false" "min-legal-vector-width"="64" "no-frame-pointer-elim"="true" "no-frame-pointer-elim-non-leaf" "no-infs-fp-math"="false" "no-jump-tables"="false" "no-nans-fp-math"="false" "no-signed-zeros-fp-math"="false" "no-trapping-math"="false" "stack-protector-buffer-size"="8" "target-cpu"="gfx1010" "target-features"="+16-bit-insts,+dl-insts,+dpp,+fp32-denormals,+fp64-fp16-denormals,+gfx10-insts,+gfx9-insts,+s-memrealtime,-code-object-v3,-sram-ecc,-xnack" "unsafe-fp-math"="false" "use-soft-float"="false" }
+attributes #1 = { norecurse nounwind readnone "correctly-rounded-divide-sqrt-fp-math"="false" "denorms-are-zero"="false" "disable-tail-calls"="false" "less-precise-fpmad"="false" "min-legal-vector-width"="64" "no-frame-pointer-elim"="true" "no-frame-pointer-elim-non-leaf" "no-infs-fp-math"="false" "no-jump-tables"="false" "no-nans-fp-math"="false" "no-signed-zeros-fp-math"="false" "no-trapping-math"="false" "stack-protector-buffer-size"="8" "target-features"="+fp64-fp16-denormals,-fp32-denormals" "unsafe-fp-math"="false" "use-soft-float"="false" }

llvm/trunk/test/CodeGen/AMDGPU/fmuladd.f16.ll

 ; RUN: llc -march=amdgcn -mcpu=fiji -mattr=-fp64-fp16-denormals -fp-contract=on -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefixes=GCN,GCN-STRICT,VI-FLUSH,VI %s
 ; RUN: llc -march=amdgcn -mcpu=fiji -mattr=-fp64-fp16-denormals -fp-contract=fast -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefixes=GCN,GCN-CONTRACT,VI-FLUSH,VI %s
 
-; RUN: llc -march=amdgcn -mcpu=fiji -mattr=+fp64-fp16-denormals -fp-contract=on -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefixes=GCN,GCN-STRICT,VI-DENORM-STRICT,VI-DENORM,VI %s
-; RUN: llc -march=amdgcn -mcpu=fiji -mattr=+fp64-fp16-denormals -fp-contract=fast -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefixes=GCN,GCN-CONTRACT,VI-DENORM-CONTRACT,VI-DENORM,VI %s
+; RUN: llc -march=amdgcn -mcpu=fiji -mattr=+fp64-fp16-denormals -fp-contract=on -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefixes=GCN,GCN-STRICT,GCN-DENORM,GCN-DENORM-STRICT,VI-DENORM-STRICT,VI-DENORM,VI %s
+; RUN: llc -march=amdgcn -mcpu=fiji -mattr=+fp64-fp16-denormals -fp-contract=fast -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefixes=GCN,GCN-CONTRACT,GCN-DENORM,GCN-DENORM-CONTRACT,VI-DENORM-CONTRACT,VI-DENORM,VI %s
+
+; RUN: llc -march=amdgcn -mcpu=gfx1010 -mattr=-fp64-fp16-denormals -fp-contract=on -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefixes=GCN,GCN-STRICT,GFX10-FLUSH,GFX10 %s
+; RUN: llc -march=amdgcn -mcpu=gfx1010 -mattr=-fp64-fp16-denormals -fp-contract=fast -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefixes=GCN,GCN-CONTRACT,GFX10-FLUSH,GFX10 %s
+; RUN: llc -march=amdgcn -mcpu=gfx1010 -mattr=+fp64-fp16-denormals -fp-contract=on -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefixes=GCN,GCN-STRICT,GCN-DENORM,GCN-DENORM-STRICT,GFX10-DENORM-STRICT,GFX10-DENORM,GFX10 %s
+; RUN: llc -march=amdgcn -mcpu=gfx1010 -mattr=+fp64-fp16-denormals -fp-contract=fast -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefixes=GCN,GCN-CONTRACT,GCN-DENORM,GCN-DENORM-CONTRACT,GFX10-DENORM-CONTRACT,GFX10-DENORM,GFX10 %s
 
 declare i32 @llvm.amdgcn.workitem.id.x() #1
 declare half @llvm.fmuladd.f16(half, half, half) #1
 declare half @llvm.fabs.f16(half) #1
 
 ; GCN-LABEL: {{^}}fmuladd_f16:
 ; VI-FLUSH: v_mac_f16_e32 {{v[0-9]+, v[0-9]+, v[0-9]+}}
 
 ; VI-DENORM: v_fma_f16 {{v[0-9]+, v[0-9]+, v[0-9]+}}
+
+; GFX10-FLUSH:  v_mul_f16_e32
+; GFX10-FLUSH:  v_add_f16_e32
+; GFX10-DENORM: v_fmac_f16_e32 {{v[0-9]+, v[0-9]+, v[0-9]+}}
+
 define amdgpu_kernel void @fmuladd_f16(half addrspace(1)* %out, half addrspace(1)* %in1,
                          half addrspace(1)* %in2, half addrspace(1)* %in3) #0 {
   %r0 = load half, half addrspace(1)* %in1
   %r1 = load half, half addrspace(1)* %in2
   %r2 = load half, half addrspace(1)* %in3
   %r3 = tail call half @llvm.fmuladd.f16(half %r0, half %r1, half %r2)
   store half %r3, half addrspace(1)* %out
   ret void
 }
 
 ; GCN-LABEL: {{^}}fmuladd_2.0_a_b_f16
-; GCN: {{buffer|flat}}_load_ushort [[R1:v[0-9]+]],
-; GCN: {{buffer|flat}}_load_ushort [[R2:v[0-9]+]],
+; GCN: {{buffer|flat|global}}_load_ushort [[R1:v[0-9]+]],
+; GCN: {{buffer|flat|global}}_load_ushort [[R2:v[0-9]+]],
 ; VI-FLUSH: v_mac_f16_e32 [[R2]], 2.0, [[R1]]
 ; VI-FLUSH: flat_store_short v{{\[[0-9]+:[0-9]+\]}}, [[R2]]
 
-; VI-DENORM: v_fma_f16 [[RESULT:v[0-9]+]], [[R1]], 2.0, [[R2]]
-; VI-DENORM: flat_store_short v{{\[[0-9]+:[0-9]+\]}}, [[RESULT]]
+; VI-DENORM:    v_fma_f16 [[RESULT:v[0-9]+]], [[R1]], 2.0, [[R2]]
+; GFX10-DENORM: v_fmac_f16_e32 [[R2:v[0-9]+]], 2.0, [[R1]]
+
+; GFX10-FLUSH:  v_add_f16_e32 [[MUL2:v[0-9]+]], [[R1]], [[R1]]
+; GFX10-FLUSH:  v_add_f16_e32 [[RESULT:v[0-9]+]], [[MUL2]], [[R2]]
+
+; VI-DENORM:    flat_store_short v{{\[[0-9]+:[0-9]+\]}}, [[RESULT]]
+; GFX10-DENORM: global_store_short v{{\[[0-9]+:[0-9]+\]}}, [[R2]]
+; GFX10-FLUSH:  global_store_short v{{\[[0-9]+:[0-9]+\]}}, [[RESULT]]
+
 define amdgpu_kernel void @fmuladd_2.0_a_b_f16(half addrspace(1)* %out, half addrspace(1)* %in) #0 {
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %gep.0 = getelementptr half, half addrspace(1)* %out, i32 %tid
   %gep.1 = getelementptr half, half addrspace(1)* %gep.0, i32 1
   %gep.out = getelementptr half, half addrspace(1)* %out, i32 %tid
 
   %r1 = load volatile half, half addrspace(1)* %gep.0
   %r2 = load volatile half, half addrspace(1)* %gep.1
 
   %r3 = tail call half @llvm.fmuladd.f16(half 2.0, half %r1, half %r2)
   store half %r3, half addrspace(1)* %gep.out
   ret void
 }
 
 ; GCN-LABEL: {{^}}fmuladd_a_2.0_b_f16
-; GCN: {{buffer|flat}}_load_ushort [[R1:v[0-9]+]],
-; GCN: {{buffer|flat}}_load_ushort [[R2:v[0-9]+]],
+; GCN: {{buffer|flat|global}}_load_ushort [[R1:v[0-9]+]],
+; GCN: {{buffer|flat|global}}_load_ushort [[R2:v[0-9]+]],
 ; VI-FLUSH: v_mac_f16_e32 [[R2]], 2.0, [[R1]]
 ; VI-FLUSH: flat_store_short v{{\[[0-9]+:[0-9]+\]}}, [[R2]]
 
-; VI-DENORM: v_fma_f16 [[RESULT:v[0-9]+]], [[R1]], 2.0, [[R2]]
+; VI-DENORM:    v_fma_f16 [[RESULT:v[0-9]+]], [[R1]], 2.0, [[R2]]
+; GFX10-DENORM: v_fmac_f16_e32 [[R2]], 2.0, [[R1]]
+
+; GFX10-FLUSH:  v_add_f16_e32 [[MUL2:v[0-9]+]], [[R1]], [[R1]]
+; GFX10-FLUSH:  v_add_f16_e32 [[RESULT:v[0-9]+]], [[MUL2]], [[R2]]
+
 ; VI-DENORM: flat_store_short v{{\[[0-9]+:[0-9]+\]}}, [[RESULT]]
+; GFX10-DENORM: global_store_short v{{\[[0-9]+:[0-9]+\]}}, [[R2]]
+; GFX10-FLUSH:  global_store_short v{{\[[0-9]+:[0-9]+\]}}, [[RESULT]]
+
 define amdgpu_kernel void @fmuladd_a_2.0_b_f16(half addrspace(1)* %out, half addrspace(1)* %in) #0 {
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %gep.0 = getelementptr half, half addrspace(1)* %out, i32 %tid
   %gep.1 = getelementptr half, half addrspace(1)* %gep.0, i32 1
   %gep.out = getelementptr half, half addrspace(1)* %out, i32 %tid
 
   %r1 = load volatile half, half addrspace(1)* %gep.0
   %r2 = load volatile half, half addrspace(1)* %gep.1
 
   %r3 = tail call half @llvm.fmuladd.f16(half %r1, half 2.0, half %r2)
   store half %r3, half addrspace(1)* %gep.out
   ret void
 }
 
 ; GCN-LABEL: {{^}}fadd_a_a_b_f16:
-; GCN: {{buffer|flat}}_load_ushort [[R1:v[0-9]+]],
-; GCN: {{buffer|flat}}_load_ushort [[R2:v[0-9]+]],
+; GCN: {{buffer|flat|global}}_load_ushort [[R1:v[0-9]+]],
+; GCN: {{buffer|flat|global}}_load_ushort [[R2:v[0-9]+]],
 ; VI-FLUSH: v_mac_f16_e32 [[R2]], 2.0, [[R1]]
 ; VI-FLUSH: flat_store_short v{{\[[0-9]+:[0-9]+\]}}, [[R2]]
 
-; VI-DENORM-CONTRACT: v_fma_f16 [[RESULT:v[0-9]+]], [[R1]], 2.0, [[R2]]
+; VI-DENORM-CONTRACT:    v_fma_f16 [[RESULT:v[0-9]+]], [[R1]], 2.0, [[R2]]
+; GFX10-DENORM-CONTRACT: v_fmac_f16_e32 [[R2]], 2.0, [[R1]]
 
-; VI-DENORM-STRICT: v_add_f16_e32 [[TMP:v[0-9]+]], [[R1]], [[R1]]
-; VI-DENORM-STRICT: v_add_f16_e32 [[RESULT:v[0-9]+]], [[TMP]], [[R2]]
+; GCN-DENORM-STRICT: v_add_f16_e32 [[TMP:v[0-9]+]], [[R1]], [[R1]]
+; GCN-DENORM-STRICT: v_add_f16_e32 [[RESULT:v[0-9]+]], [[TMP]], [[R2]]
 
 ; VI-DENORM: flat_store_short v{{\[[0-9]+:[0-9]+\]}}, [[RESULT]]
+
+; GFX10-FLUSH:           v_add_f16_e32 [[MUL2:v[0-9]+]], [[R1]], [[R1]]
+; GFX10-FLUSH:           v_add_f16_e32 [[RESULT:v[0-9]+]], [[MUL2]], [[R2]]
+; GFX10-FLUSH:           global_store_short v{{\[[0-9]+:[0-9]+\]}}, [[RESULT]]
+; GFX10-DENORM-STRICT:   global_store_short v{{\[[0-9]+:[0-9]+\]}}, [[RESULT]]
+; GFX10-DENORM-CONTRACT: global_store_short v{{\[[0-9]+:[0-9]+\]}}, [[R2]]
+
 define amdgpu_kernel void @fadd_a_a_b_f16(half addrspace(1)* %out,
                             half addrspace(1)* %in1,
                             half addrspace(1)* %in2) #0 {
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %gep.0 = getelementptr half, half addrspace(1)* %out, i32 %tid
   %gep.1 = getelementptr half, half addrspace(1)* %gep.0, i32 1
   %gep.out = getelementptr half, half addrspace(1)* %out, i32 %tid
 
   %r0 = load volatile half, half addrspace(1)* %gep.0
   %r1 = load volatile half, half addrspace(1)* %gep.1
 
   %add.0 = fadd half %r0, %r0
   %add.1 = fadd half %add.0, %r1
   store half %add.1, half addrspace(1)* %gep.out
   ret void
 }
 
 ; GCN-LABEL: {{^}}fadd_b_a_a_f16:
-; GCN: {{buffer|flat}}_load_ushort [[R1:v[0-9]+]],
-; GCN: {{buffer|flat}}_load_ushort [[R2:v[0-9]+]],
+; GCN: {{buffer|flat|global}}_load_ushort [[R1:v[0-9]+]],
+; GCN: {{buffer|flat|global}}_load_ushort [[R2:v[0-9]+]],
 ; VI-FLUSH: v_mac_f16_e32 [[R2]], 2.0, [[R1]]
 ; VI-FLUSH: flat_store_short v{{\[[0-9]+:[0-9]+\]}}, [[R2]]
 
-; VI-DENORM-CONTRACT: v_fma_f16 [[RESULT:v[0-9]+]], [[R1]], 2.0, [[R2]]
+; VI-DENORM-CONTRACT:    v_fma_f16 [[RESULT:v[0-9]+]], [[R1]], 2.0, [[R2]]
+; GFX10-DENORM-CONTRACT: v_fmac_f16_e32 [[R2]], 2.0, [[R1]]
 
-; VI-DENORM-STRICT: v_add_f16_e32 [[TMP:v[0-9]+]], [[R1]], [[R1]]
-; VI-DENORM-STRICT: v_add_f16_e32 [[RESULT:v[0-9]+]],  [[R2]], [[TMP]]
+; GCN-DENORM-STRICT: v_add_f16_e32 [[TMP:v[0-9]+]], [[R1]], [[R1]]
+; GCN-DENORM-STRICT: v_add_f16_e32 [[RESULT:v[0-9]+]],  [[R2]], [[TMP]]
 
 ; VI-DENORM: flat_store_short v{{\[[0-9]+:[0-9]+\]}}, [[RESULT]]
+
+; GFX10-FLUSH: v_add_f16_e32 [[MUL2:v[0-9]+]], [[R1]], [[R1]]
+; GFX10-FLUSH: v_add_f16_e32 [[RESULT:v[0-9]+]], [[R2]], [[MUL2]]
+; GFX10-FLUSH: global_store_short v{{\[[0-9]+:[0-9]+\]}}, [[RESULT]]
+; GFX10-DENORM-STRICT:   global_store_short v{{\[[0-9]+:[0-9]+\]}}, [[RESULT]]
+; GFX10-DENORM-CONTRACT: global_store_short v{{\[[0-9]+:[0-9]+\]}}, [[R2]]
+
 define amdgpu_kernel void @fadd_b_a_a_f16(half addrspace(1)* %out,
                             half addrspace(1)* %in1,
                             half addrspace(1)* %in2) #0 {
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %gep.0 = getelementptr half, half addrspace(1)* %out, i32 %tid
   %gep.1 = getelementptr half, half addrspace(1)* %gep.0, i32 1
   %gep.out = getelementptr half, half addrspace(1)* %out, i32 %tid
 
   %r0 = load volatile half, half addrspace(1)* %gep.0
   %r1 = load volatile half, half addrspace(1)* %gep.1
 
   %add.0 = fadd half %r0, %r0
   %add.1 = fadd half %r1, %add.0
   store half %add.1, half addrspace(1)* %gep.out
   ret void
 }
 
 ; GCN-LABEL: {{^}}fmuladd_neg_2.0_a_b_f16
-; GCN: {{buffer|flat}}_load_ushort [[R1:v[0-9]+]],
-; GCN: {{buffer|flat}}_load_ushort [[R2:v[0-9]+]],
-; VI-FLUSH: v_mac_f16_e32 [[R2]], -2.0, [[R1]]
-; VI-DENORM: v_fma_f16 [[R2:v[0-9]+]], [[R1]], -2.0, [[R2]]
-; VI: flat_store_short v{{\[[0-9]+:[0-9]+\]}}, [[R2]]
+; GCN: {{buffer|flat|global}}_load_ushort [[R1:v[0-9]+]],
+; GCN: {{buffer|flat|global}}_load_ushort [[R2:v[0-9]+]],
+; VI-FLUSH:     v_mac_f16_e32 [[R2]], -2.0, [[R1]]
+; VI-DENORM:    v_fma_f16 [[RESULT:v[0-9]+]], [[R1]], -2.0, [[R2]]
+; GFX10-DENORM: v_fmac_f16_e32 [[R2]], -2.0, [[R1]]
+; VI-FLUSH:  flat_store_short v{{\[[0-9]+:[0-9]+\]}}, [[R2]]
+; VI-DENORM: flat_store_short v{{\[[0-9]+:[0-9]+\]}}, [[RESULT]]
+; GFX10-FLUSH: v_add_f16_e32 [[MUL2:v[0-9]+]], [[R1]], [[R1]]
+; GFX10-FLUSH: v_sub_f16_e32 [[RESULT:v[0-9]+]], [[R2]], [[MUL2]]
+; GFX10-FLUSH:  global_store_short v{{\[[0-9]+:[0-9]+\]}}, [[RESULT]]
+; GFX10-DENORM: global_store_short v{{\[[0-9]+:[0-9]+\]}}, [[R2]]
 define amdgpu_kernel void @fmuladd_neg_2.0_a_b_f16(half addrspace(1)* %out, half addrspace(1)* %in) #0 {
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %gep.0 = getelementptr half, half addrspace(1)* %out, i32 %tid
   %gep.1 = getelementptr half, half addrspace(1)* %gep.0, i32 1
   %gep.out = getelementptr half, half addrspace(1)* %out, i32 %tid
 
   %r1 = load volatile half, half addrspace(1)* %gep.0
   %r2 = load volatile half, half addrspace(1)* %gep.1
 
   %r3 = tail call half @llvm.fmuladd.f16(half -2.0, half %r1, half %r2)
   store half %r3, half addrspace(1)* %gep.out
   ret void
 }
 
 ; GCN-LABEL: {{^}}fmuladd_neg_2.0_neg_a_b_f16
-; GCN: {{buffer|flat}}_load_ushort [[R1:v[0-9]+]],
-; GCN: {{buffer|flat}}_load_ushort [[R2:v[0-9]+]],
+; GCN: {{buffer|flat|global}}_load_ushort [[R1:v[0-9]+]],
+; GCN: {{buffer|flat|global}}_load_ushort [[R2:v[0-9]+]],
 ; VI-FLUSH: v_mac_f16_e32 [[R2]], 2.0, [[R1]]
 ; VI-FLUSH: flat_store_short v{{\[[0-9]+:[0-9]+\]}}, [[R2]]
 
-; VI-DENORM: v_fma_f16 [[RESULT:v[0-9]+]], [[R1]], 2.0, [[R2]]
-; VI-DENORM: flat_store_short v{{\[[0-9]+:[0-9]+\]}}, [[RESULT]]
+; VI-DENORM:    v_fma_f16 [[RESULT:v[0-9]+]], [[R1]], 2.0, [[R2]]
+; VI-DENORM:  flat_store_short v{{\[[0-9]+:[0-9]+\]}}, [[RESULT]]
+
+; GFX10-FLUSH: v_add_f16_e32 [[MUL2:v[0-9]+]], [[R1]], [[R1]]
+; GFX10-FLUSH: v_add_f16_e32 [[RESULT:v[0-9]+]], [[R2]], [[MUL2]]
+; GFX10-FLUSH:  global_store_short v{{\[[0-9]+:[0-9]+\]}}, [[RESULT]]
+
+; GFX10-DENORM: v_fmac_f16_e32 [[R2]], 2.0, [[R1]]
+; GFX10-DENORM: global_store_short v{{\[[0-9]+:[0-9]+\]}}, [[R2]]
 define amdgpu_kernel void @fmuladd_neg_2.0_neg_a_b_f16(half addrspace(1)* %out, half addrspace(1)* %in) #0 {
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %gep.0 = getelementptr half, half addrspace(1)* %out, i32 %tid
   %gep.1 = getelementptr half, half addrspace(1)* %gep.0, i32 1
   %gep.out = getelementptr half, half addrspace(1)* %out, i32 %tid
 
   %r1 = load volatile half, half addrspace(1)* %gep.0
   %r2 = load volatile half, half addrspace(1)* %gep.1
 
   %r1.fneg = fsub half -0.000000e+00, %r1
 
   %r3 = tail call half @llvm.fmuladd.f16(half -2.0, half %r1.fneg, half %r2)
   store half %r3, half addrspace(1)* %gep.out
   ret void
 }
 
 ; GCN-LABEL: {{^}}fmuladd_2.0_neg_a_b_f16
-; GCN: {{buffer|flat}}_load_ushort [[R1:v[0-9]+]],
-; GCN: {{buffer|flat}}_load_ushort [[R2:v[0-9]+]],
+; GCN: {{buffer|flat|global}}_load_ushort [[R1:v[0-9]+]],
+; GCN: {{buffer|flat|global}}_load_ushort [[R2:v[0-9]+]],
 ; VI-FLUSH: v_mac_f16_e32 [[R2]], -2.0, [[R1]]
 ; VI-FLUSH: flat_store_short v{{\[[0-9]+:[0-9]+\]}}, [[R2]]
 
 ; VI-DENORM: v_fma_f16 [[RESULT:v[0-9]+]], [[R1]], -2.0, [[R2]]
-; VI-DENORM: flat_store_short v{{\[[0-9]+:[0-9]+\]}}, [[RESULT]]
+; VI-DENORM:  flat_store_short v{{\[[0-9]+:[0-9]+\]}}, [[RESULT]]
+
+; GFX10-FLUSH: v_add_f16_e32 [[MUL2:v[0-9]+]], [[R1]], [[R1]]
+; GFX10-FLUSH: v_sub_f16_e32 [[RESULT:v[0-9]+]], [[R2]], [[MUL2]]
+; GFX10-FLUSH: global_store_short v{{\[[0-9]+:[0-9]+\]}}, [[RESULT]]
+
+; GFX10-DENORM: v_fmac_f16_e32 [[R2]], -2.0, [[R1]]
+; GFX10-DENORM: global_store_short v{{\[[0-9]+:[0-9]+\]}}, [[R2]]
 define amdgpu_kernel void @fmuladd_2.0_neg_a_b_f16(half addrspace(1)* %out, half addrspace(1)* %in) #0 {
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %gep.0 = getelementptr half, half addrspace(1)* %out, i32 %tid
   %gep.1 = getelementptr half, half addrspace(1)* %gep.0, i32 1
   %gep.out = getelementptr half, half addrspace(1)* %out, i32 %tid
 
   %r1 = load volatile half, half addrspace(1)* %gep.0
   %r2 = load volatile half, half addrspace(1)* %gep.1
 
   %r1.fneg = fsub half -0.000000e+00, %r1
 
   %r3 = tail call half @llvm.fmuladd.f16(half 2.0, half %r1.fneg, half %r2)
   store half %r3, half addrspace(1)* %gep.out
   ret void
 }
 
 ; GCN-LABEL: {{^}}fmuladd_2.0_a_neg_b_f16
-; GCN: {{buffer|flat}}_load_ushort [[R1:v[0-9]+]],
-; GCN: {{buffer|flat}}_load_ushort [[R2:v[0-9]+]],
-; VI-FLUSH: v_mad_f16 [[RESULT:v[0-9]+]], [[R1]], 2.0, -[[R2]]
-; VI-DENORM: v_fma_f16 [[RESULT:v[0-9]+]], [[R1]], 2.0, -[[R2]]
+; GCN: {{buffer|flat|global}}_load_ushort [[R1:v[0-9]+]],
+; GCN: {{buffer|flat|global}}_load_ushort [[R2:v[0-9]+]],
+; VI-FLUSH:   v_mad_f16 [[RESULT:v[0-9]+]], [[R1]], 2.0, -[[R2]]
+; GCN-DENORM: v_fma_f16 [[RESULT:v[0-9]+]], [[R1]], 2.0, -[[R2]]
 ; VI: flat_store_short v{{\[[0-9]+:[0-9]+\]}}, [[RESULT]]
+; GFX10-FLUSH: v_add_f16_e32 [[MUL2:v[0-9]+]], [[R1]], [[R1]]
+; GFX10-FLUSH: v_sub_f16_e32 [[RESULT:v[0-9]+]], [[MUL2]], [[R2]]
+; GFX10:       global_store_short v{{\[[0-9]+:[0-9]+\]}}, [[RESULT]]
 define amdgpu_kernel void @fmuladd_2.0_a_neg_b_f16(half addrspace(1)* %out, half addrspace(1)* %in) #0 {
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %gep.0 = getelementptr half, half addrspace(1)* %out, i32 %tid
   %gep.1 = getelementptr half, half addrspace(1)* %gep.0, i32 1
   %gep.out = getelementptr half, half addrspace(1)* %out, i32 %tid
 
   %r1 = load volatile half, half addrspace(1)* %gep.0
   %r2 = load volatile half, half addrspace(1)* %gep.1
 
   %r2.fneg = fsub half -0.000000e+00, %r2
 
   %r3 = tail call half @llvm.fmuladd.f16(half 2.0, half %r1, half %r2.fneg)
   store half %r3, half addrspace(1)* %gep.out
   ret void
 }
 
 ; GCN-LABEL: {{^}}mad_sub_f16:
-; GCN: {{buffer|flat}}_load_ushort [[REGA:v[0-9]+]]
-; GCN: {{buffer|flat}}_load_ushort [[REGB:v[0-9]+]]
-; GCN: {{buffer|flat}}_load_ushort [[REGC:v[0-9]+]]
+; GCN: {{buffer|flat|global}}_load_ushort [[REGA:v[0-9]+]]
+; GCN: {{buffer|flat|global}}_load_ushort [[REGB:v[0-9]+]]
+; GCN: {{buffer|flat|global}}_load_ushort [[REGC:v[0-9]+]]
 
 ; VI-FLUSH: v_mad_f16 [[RESULT:v[0-9]+]], [[REGA]], [[REGB]], -[[REGC]]
 
-; VI-DENORM-CONTRACT: v_fma_f16 [[RESULT:v[0-9]+]], [[REGA]], [[REGB]], -[[REGC]]
+; GCN-DENORM-CONTRACT: v_fma_f16 [[RESULT:v[0-9]+]], [[REGA]], [[REGB]], -[[REGC]]
 
-; VI-DENORM-STRICT: v_mul_f16_e32 [[TMP:v[0-9]+]], [[REGA]], [[REGB]]
-; VI-DENORM-STRICT: v_sub_f16_e32 [[RESULT:v[0-9]+]], [[TMP]], [[REGC]]
+; GCN-DENORM-STRICT: v_mul_f16_e32 [[TMP:v[0-9]+]], [[REGA]], [[REGB]]
+; GCN-DENORM-STRICT: v_sub_f16_e32 [[RESULT:v[0-9]+]], [[TMP]], [[REGC]]
 
 ; VI: flat_store_short v{{\[[0-9]+:[0-9]+\]}}, [[RESULT]]
+
+; GFX10-FLUSH: v_mul_f16_e32 [[TMP:v[0-9]+]], [[REGA]], [[REGB]]
+; GFX10-FLUSH: v_sub_f16_e32 [[RESULT:v[0-9]+]], [[TMP]], [[REGC]]
+; GFX10:       global_store_short v{{\[[0-9]+:[0-9]+\]}}, [[RESULT]]
 define amdgpu_kernel void @mad_sub_f16(half addrspace(1)* noalias nocapture %out, half addrspace(1)* noalias nocapture readonly %ptr) #1 {
   %tid = tail call i32 @llvm.amdgcn.workitem.id.x() #0
   %tid.ext = sext i32 %tid to i64
   %gep0 = getelementptr half, half addrspace(1)* %ptr, i64 %tid.ext
   %add1 = add i64 %tid.ext, 1
   %gep1 = getelementptr half, half addrspace(1)* %ptr, i64 %add1
   %add2 = add i64 %tid.ext, 2
   %gep2 = getelementptr half, half addrspace(1)* %ptr, i64 %add2
   %outgep = getelementptr half, half addrspace(1)* %out, i64 %tid.ext
   %a = load volatile half, half addrspace(1)* %gep0, align 2
   %b = load volatile half, half addrspace(1)* %gep1, align 2
   %c = load volatile half, half addrspace(1)* %gep2, align 2
   %mul = fmul half %a, %b
   %sub = fsub half %mul, %c
   store half %sub, half addrspace(1)* %outgep, align 2
   ret void
 }
 
 ; GCN-LABEL: {{^}}mad_sub_inv_f16:
-; GCN: {{buffer|flat}}_load_ushort [[REGA:v[0-9]+]]
-; GCN: {{buffer|flat}}_load_ushort [[REGB:v[0-9]+]]
-; GCN: {{buffer|flat}}_load_ushort [[REGC:v[0-9]+]]
+; GCN: {{buffer|flat|global}}_load_ushort [[REGA:v[0-9]+]]
+; GCN: {{buffer|flat|global}}_load_ushort [[REGB:v[0-9]+]]
+; GCN: {{buffer|flat|global}}_load_ushort [[REGC:v[0-9]+]]
 ; VI-FLUSH: v_mad_f16 [[RESULT:v[0-9]+]], -[[REGA]], [[REGB]], [[REGC]]
 
-; VI-DENORM-CONTRACT: v_fma_f16 [[RESULT:v[0-9]+]], -[[REGA]], [[REGB]], [[REGC]]
+; VI-DENORM-CONTRACT:    v_fma_f16 [[RESULT:v[0-9]+]], -[[REGA]], [[REGB]], [[REGC]]
+; GFX10-DENORM-CONTRACT: v_fmac_f16_e64 [[REGC]], -[[REGA]], [[REGB]]
 
-; VI-DENORM-STRICT: v_mul_f16_e32 [[TMP:v[0-9]+]], [[REGA]], [[REGB]]
-; VI-DENORM-STRICT: v_sub_f16_e32 [[RESULT:v[0-9]+]], [[REGC]], [[TMP]]
+; GCN-DENORM-STRICT: v_mul_f16_e32 [[TMP:v[0-9]+]], [[REGA]], [[REGB]]
+; GCN-DENORM-STRICT: v_sub_f16_e32 [[RESULT:v[0-9]+]], [[REGC]], [[TMP]]
 
 ; VI: flat_store_short v{{\[[0-9]+:[0-9]+\]}}, [[RESULT]]
+
+; GFX10-FLUSH: v_mul_f16_e32 [[TMP:v[0-9]+]], [[REGA]], [[REGB]]
+; GFX10-FLUSH: v_sub_f16_e32 [[RESULT:v[0-9]+]], [[REGC]], [[TMP]]
+; GFX10-FLUSH:  global_store_short v{{\[[0-9]+:[0-9]+\]}}, [[RESULT]]
+; GFX10-DENORM: global_store_short v{{\[[0-9]+:[0-9]+\]}}, [[REGC]]
 define amdgpu_kernel void @mad_sub_inv_f16(half addrspace(1)* noalias nocapture %out, half addrspace(1)* noalias nocapture readonly %ptr) #1 {
   %tid = tail call i32 @llvm.amdgcn.workitem.id.x() #0
   %tid.ext = sext i32 %tid to i64
   %gep0 = getelementptr half, half addrspace(1)* %ptr, i64 %tid.ext
   %add1 = add i64 %tid.ext, 1
   %gep1 = getelementptr half, half addrspace(1)* %ptr, i64 %add1
   %add2 = add i64 %tid.ext, 2
   %gep2 = getelementptr half, half addrspace(1)* %ptr, i64 %add2
   %outgep = getelementptr half, half addrspace(1)* %out, i64 %tid.ext
   %a = load volatile half, half addrspace(1)* %gep0, align 2
   %b = load volatile half, half addrspace(1)* %gep1, align 2
   %c = load volatile half, half addrspace(1)* %gep2, align 2
   %mul = fmul half %a, %b
   %sub = fsub half %c, %mul
   store half %sub, half addrspace(1)* %outgep, align 2
   ret void
 }
 
 ; GCN-LABEL: {{^}}mad_sub_fabs_f16:
-; GCN: {{buffer|flat}}_load_ushort [[REGA:v[0-9]+]]
-; GCN: {{buffer|flat}}_load_ushort [[REGB:v[0-9]+]]
-; GCN: {{buffer|flat}}_load_ushort [[REGC:v[0-9]+]]
+; GCN: {{buffer|flat|global}}_load_ushort [[REGA:v[0-9]+]]
+; GCN: {{buffer|flat|global}}_load_ushort [[REGB:v[0-9]+]]
+; GCN: {{buffer|flat|global}}_load_ushort [[REGC:v[0-9]+]]
 ; VI-FLUSH: v_mad_f16 [[RESULT:v[0-9]+]], [[REGA]], [[REGB]], -|[[REGC]]|
 
-; VI-DENORM-CONTRACT: v_fma_f16 [[RESULT:v[0-9]+]], [[REGA]], [[REGB]], -|[[REGC]]|
+; GCN-DENORM-CONTRACT: v_fma_f16 [[RESULT:v[0-9]+]], [[REGA]], [[REGB]], -|[[REGC]]|
 
-; VI-DENORM-STRICT: v_mul_f16_e32 [[TMP:v[0-9]+]], [[REGA]], [[REGB]]
-; VI-DENORM-STRICT: v_sub_f16_e64 [[RESULT:v[0-9]+]], [[TMP]], |[[REGC]]|
+; GCN-DENORM-STRICT: v_mul_f16_e32 [[TMP:v[0-9]+]], [[REGA]], [[REGB]]
+; GCN-DENORM-STRICT: v_sub_f16_e64 [[RESULT:v[0-9]+]], [[TMP]], |[[REGC]]|
 
 ; VI: flat_store_short v{{\[[0-9]+:[0-9]+\]}}, [[RESULT]]
+
+; GFX10-FLUSH: v_mul_f16_e32 [[TMP:v[0-9]+]], [[REGA]], [[REGB]]
+; GFX10-FLUSH: v_sub_f16_e64 [[RESULT:v[0-9]+]], [[TMP]], |[[REGC]]|
+; GFX10:       global_store_short v{{\[[0-9]+:[0-9]+\]}}, [[RESULT]]
 define amdgpu_kernel void @mad_sub_fabs_f16(half addrspace(1)* noalias nocapture %out, half addrspace(1)* noalias nocapture readonly %ptr) #1 {
   %tid = tail call i32 @llvm.amdgcn.workitem.id.x() #0
   %tid.ext = sext i32 %tid to i64
   %gep0 = getelementptr half, half addrspace(1)* %ptr, i64 %tid.ext
   %add1 = add i64 %tid.ext, 1
   %gep1 = getelementptr half, half addrspace(1)* %ptr, i64 %add1
   %add2 = add i64 %tid.ext, 2
   %gep2 = getelementptr half, half addrspace(1)* %ptr, i64 %add2
   %outgep = getelementptr half, half addrspace(1)* %out, i64 %tid.ext
   %a = load volatile half, half addrspace(1)* %gep0, align 2
   %b = load volatile half, half addrspace(1)* %gep1, align 2
   %c = load volatile half, half addrspace(1)* %gep2, align 2
   %c.abs = call half @llvm.fabs.f16(half %c) #0
   %mul = fmul half %a, %b
   %sub = fsub half %mul, %c.abs
   store half %sub, half addrspace(1)* %outgep, align 2
   ret void
 }
 
 ; GCN-LABEL: {{^}}mad_sub_fabs_inv_f16:
-; GCN: {{buffer|flat}}_load_ushort [[REGA:v[0-9]+]]
-; GCN: {{buffer|flat}}_load_ushort [[REGB:v[0-9]+]]
-; GCN: {{buffer|flat}}_load_ushort [[REGC:v[0-9]+]]
+; GCN: {{buffer|flat|global}}_load_ushort [[REGA:v[0-9]+]]
+; GCN: {{buffer|flat|global}}_load_ushort [[REGB:v[0-9]+]]
+; GCN: {{buffer|flat|global}}_load_ushort [[REGC:v[0-9]+]]
 
 ; VI-FLUSH: v_mad_f16 [[RESULT:v[0-9]+]], -[[REGA]], [[REGB]], |[[REGC]]|
 
-; VI-DENORM-CONTRACT: v_fma_f16 [[RESULT:v[0-9]+]], -[[REGA]], [[REGB]], |[[REGC]]|
+; GCN-DENORM-CONTRACT: v_fma_f16 [[RESULT:v[0-9]+]], -[[REGA]], [[REGB]], |[[REGC]]|
 
-; VI-DENORM-STRICT: v_mul_f16_e32 [[TMP:v[0-9]+]], [[REGA]], [[REGB]]
-; VI-DENORM-STRICT: v_sub_f16_e64 [[RESULT:v[0-9]+]], |[[REGC]]|, [[TMP]]
+; GCN-DENORM-STRICT: v_mul_f16_e32 [[TMP:v[0-9]+]], [[REGA]], [[REGB]]
+; GCN-DENORM-STRICT: v_sub_f16_e64 [[RESULT:v[0-9]+]], |[[REGC]]|, [[TMP]]
 
 ; VI: flat_store_short v{{\[[0-9]+:[0-9]+\]}}, [[RESULT]]
+
+; GFX10-FLUSH: v_mul_f16_e32 [[TMP:v[0-9]+]], [[REGA]], [[REGB]]
+; GFX10-FLUSH: v_sub_f16_e64 [[RESULT:v[0-9]+]], |[[REGC]]|, [[TMP]]
+; GFX10:       global_store_short v{{\[[0-9]+:[0-9]+\]}}, [[RESULT]]
 define amdgpu_kernel void @mad_sub_fabs_inv_f16(half addrspace(1)* noalias nocapture %out, half addrspace(1)* noalias nocapture readonly %ptr) #1 {
   %tid = tail call i32 @llvm.amdgcn.workitem.id.x() #0
   %tid.ext = sext i32 %tid to i64
   %gep0 = getelementptr half, half addrspace(1)* %ptr, i64 %tid.ext
   %add1 = add i64 %tid.ext, 1
   %gep1 = getelementptr half, half addrspace(1)* %ptr, i64 %add1
   %add2 = add i64 %tid.ext, 2
   %gep2 = getelementptr half, half addrspace(1)* %ptr, i64 %add2
   %outgep = getelementptr half, half addrspace(1)* %out, i64 %tid.ext
   %a = load volatile half, half addrspace(1)* %gep0, align 2
   %b = load volatile half, half addrspace(1)* %gep1, align 2
   %c = load volatile half, half addrspace(1)* %gep2, align 2
   %c.abs = call half @llvm.fabs.f16(half %c) #0
   %mul = fmul half %a, %b
   %sub = fsub half %c.abs, %mul
   store half %sub, half addrspace(1)* %outgep, align 2
   ret void
 }
 
 ; GCN-LABEL: {{^}}neg_neg_mad_f16:
-; GCN: {{buffer|flat}}_load_ushort [[REGA:v[0-9]+]]
-; GCN: {{buffer|flat}}_load_ushort [[REGB:v[0-9]+]]
-; GCN: {{buffer|flat}}_load_ushort [[REGC:v[0-9]+]]
+; GCN: {{buffer|flat|global}}_load_ushort [[REGA:v[0-9]+]]
+; GCN: {{buffer|flat|global}}_load_ushort [[REGB:v[0-9]+]]
+; GCN: {{buffer|flat|global}}_load_ushort [[REGC:v[0-9]+]]
 
 ; VI-FLUSH: v_mac_f16_e32 [[REGC]], [[REGA]], [[REGB]]
 ; VI-FLUSH: flat_store_short v{{\[[0-9]+:[0-9]+\]}}, [[REGC]]
 
-; VI-DENORM-CONTRACT: v_fma_f16 [[RESULT:v[0-9]+]], [[REGA]], [[REGB]], [[REGC]]
+; VI-DENORM-CONTRACT:    v_fma_f16 [[RESULT:v[0-9]+]], [[REGA]], [[REGB]], [[REGC]]
+; GFX10-DENORM-CONTRACT: v_fmac_f16_e32 [[REGC]], [[REGA]], [[REGB]]
 
-; VI-DENORM-STRICT: v_mul_f16_e32 [[TMP:v[0-9]+]], [[REGA]], [[REGB]]
-; VI-DENORM-STRICT: v_add_f16_e32 [[RESULT:v[0-9]+]], [[REGC]], [[TMP]]
+; GCN-DENORM-STRICT: v_mul_f16_e32 [[TMP:v[0-9]+]], [[REGA]], [[REGB]]
+; GCN-DENORM-STRICT: v_add_f16_e32 [[RESULT:v[0-9]+]], [[REGC]], [[TMP]]
 ; VI-DENORM: flat_store_short v{{\[[0-9]+:[0-9]+\]}}, [[RESULT]]
+
+; GFX10-FLUSH: v_mul_f16_e32 [[TMP:v[0-9]+]], [[REGA]], [[REGB]]
+; GFX10-FLUSH: v_add_f16_e32 [[RESULT:v[0-9]+]], [[REGC]], [[TMP]]
+; GFX10-FLUSH:  global_store_short v{{\[[0-9]+:[0-9]+\]}}, [[RESULT]]
+; GFX10-DENORM: global_store_short v{{\[[0-9]+:[0-9]+\]}}, [[REGC]]
 define amdgpu_kernel void @neg_neg_mad_f16(half addrspace(1)* noalias nocapture %out, half addrspace(1)* noalias nocapture readonly %ptr) #1 {
   %tid = tail call i32 @llvm.amdgcn.workitem.id.x() #0
   %tid.ext = sext i32 %tid to i64
   %gep0 = getelementptr half, half addrspace(1)* %ptr, i64 %tid.ext
   %add1 = add i64 %tid.ext, 1
   %gep1 = getelementptr half, half addrspace(1)* %ptr, i64 %add1
   %add2 = add i64 %tid.ext, 2
   %gep2 = getelementptr half, half addrspace(1)* %ptr, i64 %add2
   %outgep = getelementptr half, half addrspace(1)* %out, i64 %tid.ext
   %a = load volatile half, half addrspace(1)* %gep0, align 2
   %b = load volatile half, half addrspace(1)* %gep1, align 2
   %c = load volatile half, half addrspace(1)* %gep2, align 2
   %nega = fsub half -0.000000e+00, %a
   %negb = fsub half -0.000000e+00, %b
   %mul = fmul half %nega, %negb
   %sub = fadd half %mul, %c
   store half %sub, half addrspace(1)* %outgep, align 2
   ret void
 }
 
 ; GCN-LABEL: {{^}}mad_fabs_sub_f16:
-; GCN: {{buffer|flat}}_load_ushort [[REGA:v[0-9]+]]
-; GCN: {{buffer|flat}}_load_ushort [[REGB:v[0-9]+]]
-; GCN: {{buffer|flat}}_load_ushort [[REGC:v[0-9]+]]
+; GCN: {{buffer|flat|global}}_load_ushort [[REGA:v[0-9]+]]
+; GCN: {{buffer|flat|global}}_load_ushort [[REGB:v[0-9]+]]
+; GCN: {{buffer|flat|global}}_load_ushort [[REGC:v[0-9]+]]
 
 ; VI-FLUSH: v_mad_f16 [[RESULT:v[0-9]+]], [[REGA]], |[[REGB]]|, -[[REGC]]
 
-; VI-DENORM-CONTRACT: v_fma_f16 [[RESULT:v[0-9]+]], [[REGA]], |[[REGB]]|, -[[REGC]]
+; GCN-DENORM-CONTRACT: v_fma_f16 [[RESULT:v[0-9]+]], [[REGA]], |[[REGB]]|, -[[REGC]]
 
-; VI-DENORM-STRICT: v_mul_f16_e64 [[TMP:v[0-9]+]], [[REGA]], |[[REGB]]|
-; VI-DENORM-STRICT: v_sub_f16_e32 [[RESULT:v[0-9]+]], [[TMP]], [[REGC]]
+; GCN-DENORM-STRICT: v_mul_f16_e64 [[TMP:v[0-9]+]], [[REGA]], |[[REGB]]|
+; GCN-DENORM-STRICT: v_sub_f16_e32 [[RESULT:v[0-9]+]], [[TMP]], [[REGC]]
 
 ; VI: flat_store_short v{{\[[0-9]+:[0-9]+\]}}, [[RESULT]]
+
+; GFX10-FLUSH: v_mul_f16_e64 [[TMP:v[0-9]+]], [[REGA]], |[[REGB]]|
+; GFX10-FLUSH: v_sub_f16_e32 [[RESULT:v[0-9]+]], [[TMP]], [[REGC]]
+; GFX10:       global_store_short v{{\[[0-9]+:[0-9]+\]}}, [[RESULT]]
 define amdgpu_kernel void @mad_fabs_sub_f16(half addrspace(1)* noalias nocapture %out, half addrspace(1)* noalias nocapture readonly %ptr) #1 {
   %tid = tail call i32 @llvm.amdgcn.workitem.id.x() #0
   %tid.ext = sext i32 %tid to i64
   %gep0 = getelementptr half, half addrspace(1)* %ptr, i64 %tid.ext
   %add1 = add i64 %tid.ext, 1
   %gep1 = getelementptr half, half addrspace(1)* %ptr, i64 %add1
   %add2 = add i64 %tid.ext, 2
   %gep2 = getelementptr half, half addrspace(1)* %ptr, i64 %add2
   %outgep = getelementptr half, half addrspace(1)* %out, i64 %tid.ext
   %a = load volatile half, half addrspace(1)* %gep0, align 2
   %b = load volatile half, half addrspace(1)* %gep1, align 2
   %c = load volatile half, half addrspace(1)* %gep2, align 2
   %b.abs = call half @llvm.fabs.f16(half %b) #0
   %mul = fmul half %a, %b.abs
   %sub = fsub half %mul, %c
   store half %sub, half addrspace(1)* %outgep, align 2
   ret void
 }
 
 ; GCN-LABEL: {{^}}fsub_c_fadd_a_a_f16:
-; GCN: {{buffer|flat}}_load_ushort [[R1:v[0-9]+]],
-; GCN: {{buffer|flat}}_load_ushort [[R2:v[0-9]+]],
+; GCN: {{buffer|flat|global}}_load_ushort [[R1:v[0-9]+]],
+; GCN: {{buffer|flat|global}}_load_ushort [[R2:v[0-9]+]],
 ; VI-FLUSH: v_mac_f16_e32 [[R2]], -2.0, [[R1]]
 ; VI-FLUSH: flat_store_short v{{\[[0-9]+:[0-9]+\]}}, [[R2]]
 
-; VI-DENORM-CONTRACT: v_fma_f16 [[RESULT:v[0-9]+]], [[R1]], -2.0, [[R2]]
+; VI-DENORM-CONTRACT:    v_fma_f16 [[RESULT:v[0-9]+]], [[R1]], -2.0, [[R2]]
+; GFX10-DENORM-CONTRACT: v_fmac_f16_e32 [[R2]], -2.0, [[R1]]
 
-; VI-DENORM-STRICT: v_add_f16_e32 [[TMP:v[0-9]+]], [[R1]], [[R1]]
-; VI-DENORM-STRICT: v_sub_f16_e32 [[RESULT:v[0-9]+]], [[R2]], [[TMP]]
+; GCN-DENORM-STRICT: v_add_f16_e32 [[TMP:v[0-9]+]], [[R1]], [[R1]]
+; GCN-DENORM-STRICT: v_sub_f16_e32 [[RESULT:v[0-9]+]], [[R2]], [[TMP]]
 
 ; VI-DENORM: flat_store_short v{{\[[0-9]+:[0-9]+\]}}, [[RESULT]]
+
+; GFX10-FLUSH: v_add_f16_e32 [[TMP:v[0-9]+]], [[R1]], [[R1]]
+; GFX10-FLUSH: v_sub_f16_e32 [[RESULT:v[0-9]+]], [[R2]], [[TMP]]
+; GFX10-FLUSH:  global_store_short v{{\[[0-9]+:[0-9]+\]}}, [[RESULT]]
+; GFX10-DENORM-STRICT:   global_store_short v{{\[[0-9]+:[0-9]+\]}}, [[RESULT]]
+; GFX10-DENORM-CONTRACT: global_store_short v{{\[[0-9]+:[0-9]+\]}}, [[R2]]
 define amdgpu_kernel void @fsub_c_fadd_a_a_f16(half addrspace(1)* %out, half addrspace(1)* %in) {
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %gep.0 = getelementptr half, half addrspace(1)* %out, i32 %tid
   %gep.1 = getelementptr half, half addrspace(1)* %gep.0, i32 1
   %gep.out = getelementptr half, half addrspace(1)* %out, i32 %tid
 
   %r1 = load volatile half, half addrspace(1)* %gep.0
   %r2 = load volatile half, half addrspace(1)* %gep.1
 
   %add = fadd half %r1, %r1
   %r3 = fsub half %r2, %add
 
   store half %r3, half addrspace(1)* %gep.out
   ret void
 }
 
 ; GCN-LABEL: {{^}}fsub_fadd_a_a_c_f16:
-; GCN: {{buffer|flat}}_load_ushort [[R1:v[0-9]+]],
-; GCN: {{buffer|flat}}_load_ushort [[R2:v[0-9]+]],
+; GCN: {{buffer|flat|global}}_load_ushort [[R1:v[0-9]+]],
+; GCN: {{buffer|flat|global}}_load_ushort [[R2:v[0-9]+]],
 
 ; VI-FLUSH: v_mad_f16 [[RESULT:v[0-9]+]], [[R1]], 2.0, -[[R2]]
 
-; VI-DENORM-CONTRACT: v_fma_f16 [[RESULT:v[0-9]+]], [[R1]], 2.0, -[[R2]]
+; GCN-DENORM-CONTRACT: v_fma_f16 [[RESULT:v[0-9]+]], [[R1]], 2.0, -[[R2]]
 
-; VI-DENORM-STRICT: v_add_f16_e32 [[TMP:v[0-9]+]], [[R1]], [[R1]]
-; VI-DENORM-STRICT: v_sub_f16_e32 [[RESULT:v[0-9]+]], [[TMP]], [[R2]]
+; GCN-DENORM-STRICT: v_add_f16_e32 [[TMP:v[0-9]+]], [[R1]], [[R1]]
+; GCN-DENORM-STRICT: v_sub_f16_e32 [[RESULT:v[0-9]+]], [[TMP]], [[R2]]
 
 ; VI: flat_store_short v{{\[[0-9]+:[0-9]+\]}}, [[RESULT]]
+
+; GFX10-FLUSH: v_add_f16_e32 [[TMP:v[0-9]+]], [[R1]], [[R1]]
+; GFX10-FLUSH: v_sub_f16_e32 [[RESULT:v[0-9]+]], [[TMP]], [[R2]]
+; GFX10:       global_store_short v{{\[[0-9]+:[0-9]+\]}}, [[RESULT]]
 define amdgpu_kernel void @fsub_fadd_a_a_c_f16(half addrspace(1)* %out, half addrspace(1)* %in) {
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %gep.0 = getelementptr half, half addrspace(1)* %out, i32 %tid
   %gep.1 = getelementptr half, half addrspace(1)* %gep.0, i32 1
   %gep.out = getelementptr half, half addrspace(1)* %out, i32 %tid
 
   %r1 = load volatile half, half addrspace(1)* %gep.0
   %r2 = load volatile half, half addrspace(1)* %gep.1

llvm/trunk/test/CodeGen/AMDGPU/llvm.fmuladd.f16.ll

 ; RUN: llc -amdgpu-scalarize-global-loads=false -march=amdgcn -mcpu=tahiti -mattr=-fp64-fp16-denormals -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefix=GCN -check-prefix=SI -check-prefix=SI-FLUSH %s
 ; RUN: llc -amdgpu-scalarize-global-loads=false -march=amdgcn -mcpu=fiji -mattr=-fp64-fp16-denormals,-flat-for-global -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefix=GCN -check-prefix=VI -check-prefix=VI-FLUSH %s
 ; RUN: llc -amdgpu-scalarize-global-loads=false -march=amdgcn -mcpu=tahiti -mattr=+fp64-fp16-denormals -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefix=GCN -check-prefix=SI -check-prefix=SI-DENORM %s
 ; RUN: llc -amdgpu-scalarize-global-loads=false -march=amdgcn -mcpu=fiji -mattr=+fp64-fp16-denormals,-flat-for-global -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefix=GCN -check-prefix=VI -check-prefix=VI-DENORM %s
+; RUN: llc -amdgpu-scalarize-global-loads=false -march=amdgcn -mcpu=gfx1010 -mattr=-fp64-fp16-denormals,-flat-for-global -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefix=GCN -check-prefix=GFX10 -check-prefix=GFX10-FLUSH %s
+; RUN: llc -amdgpu-scalarize-global-loads=false -march=amdgcn -mcpu=gfx1010 -mattr=+fp64-fp16-denormals,-flat-for-global -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefix=GCN -check-prefix=GFX10 -check-prefix=GFX10-DENORM %s
 
 declare half @llvm.fmuladd.f16(half %a, half %b, half %c)
 declare <2 x half> @llvm.fmuladd.v2f16(<2 x half> %a, <2 x half> %b, <2 x half> %c)
 
 ; GCN-LABEL: {{^}}fmuladd_f16
 ; GCN: buffer_load_ushort v[[A_F16:[0-9]+]]
 ; GCN: buffer_load_ushort v[[B_F16:[0-9]+]]
 ; GCN: buffer_load_ushort v[[C_F16:[0-9]+]]
 ; SI:  v_cvt_f32_f16_e32 v[[A_F32:[0-9]+]], v[[A_F16]]
 ; SI:  v_cvt_f32_f16_e32 v[[B_F32:[0-9]+]], v[[B_F16]]
 ; SI:  v_cvt_f32_f16_e32 v[[C_F32:[0-9]+]], v[[C_F16]]
 ; SI:  v_mac_f32_e32 v[[C_F32]], v[[A_F32]], v[[B_F32]]
 ; SI:  v_cvt_f16_f32_e32 v[[R_F16:[0-9]+]], v[[C_F32]]
 ; SI:  buffer_store_short v[[R_F16]]
 
 ; VI-FLUSH: v_mac_f16_e32 v[[C_F16]], v[[A_F16]], v[[B_F16]]
 ; VI-FLUSH: buffer_store_short v[[C_F16]]
 
 ; VI-DENORM: v_fma_f16 [[RESULT:v[0-9]+]], v[[A_F16]], v[[B_F16]], v[[C_F16]]
 ; VI-DENORM: buffer_store_short [[RESULT]]
 
+; GFX10-FLUSH: v_mul_f16_e32 [[MUL:v[0-9]+]], v[[A_F16]], v[[B_F16]]
+; GFX10-FLUSH: v_add_f16_e32 [[ADD:v[0-9]+]], [[MUL]], v[[C_F16]]
+; GFX10-FLUSH: buffer_store_short [[ADD]]
+
+; GFX10-DENORM: v_fmac_f16_e32 v[[C_F16]], v[[A_F16]], v[[B_F16]]
+; GFX10-DENORM: buffer_store_short v[[C_F16]],
+
 ; GCN: s_endpgm
 define amdgpu_kernel void @fmuladd_f16(
     half addrspace(1)* %r,
     half addrspace(1)* %a,
     half addrspace(1)* %b,
     half addrspace(1)* %c) {
   %a.val = load half, half addrspace(1)* %a
   %b.val = load half, half addrspace(1)* %b

 
 ; VI-FLUSH: v_mac_f16_e32 v[[C_F16]], 0x4200, v[[B_F16]]
 ; VI-FLUSH: buffer_store_short v[[C_F16]]
 
 ; VI-DENORM: s_movk_i32 [[KA:s[0-9]+]], 0x4200
 ; VI-DENORM: v_fma_f16 [[RESULT:v[0-9]+]], v[[B_F16]], [[KA]], v[[C_F16]]
 ; VI-DENORM: buffer_store_short [[RESULT]]
 
+; GFX10-FLUSH: v_mul_f16_e32 [[MUL:v[0-9]+]], 0x4200, v[[B_F16]]
+; GFX10-FLUSH: v_add_f16_e32 [[ADD:v[0-9]+]], [[MUL]], v[[C_F16]]
+; GFX10-FLUSH: buffer_store_short [[ADD]]
+
+; GFX10-DENORM: v_fmac_f16_e32 v[[C_F16]], 0x4200, v[[B_F16]]
+; GFX10-DENORM: buffer_store_short v[[C_F16]],
+
 ; GCN: s_endpgm
 define amdgpu_kernel void @fmuladd_f16_imm_a(
     half addrspace(1)* %r,
     half addrspace(1)* %b,
     half addrspace(1)* %c) {
   %b.val = load volatile half, half addrspace(1)* %b
   %c.val = load volatile half, half addrspace(1)* %c
   %r.val = call half @llvm.fmuladd.f16(half 3.0, half %b.val, half %c.val)

 
 ; VI-FLUSH: v_mac_f16_e32 v[[C_F16]], 0x4200, v[[A_F16]]
 ; VI-FLUSH: buffer_store_short v[[C_F16]]
 
 ; VI-DENORM: s_movk_i32 [[KA:s[0-9]+]], 0x4200
 ; VI-DENORM: v_fma_f16 [[RESULT:v[0-9]+]], v[[A_F16]], [[KA]], v[[C_F16]]
 ; VI-DENORM buffer_store_short [[RESULT]]
 
+; GFX10-FLUSH: v_mul_f16_e32 [[MUL:v[0-9]+]], 0x4200, v[[A_F16]]
+; GFX10-FLUSH: v_add_f16_e32 [[ADD:v[0-9]+]], [[MUL]], v[[C_F16]]
+; GFX10-FLUSH: buffer_store_short [[ADD]]
+
+; GFX10-DENORM: v_fmac_f16_e32 v[[C_F16]], 0x4200, v[[A_F16]]
+; GFX10-DENORM buffer_store_short v[[C_F16]],
 
 ; GCN: s_endpgm
 define amdgpu_kernel void @fmuladd_f16_imm_b(
     half addrspace(1)* %r,
     half addrspace(1)* %a,
     half addrspace(1)* %c) {
   %a.val = load volatile half, half addrspace(1)* %a
   %c.val = load volatile half, half addrspace(1)* %c

 ; VI-FLUSH: buffer_load_dword v[[A_V2_F16:[0-9]+]]
 ; VI-FLUSH: buffer_load_dword v[[C_V2_F16:[0-9]+]]
 ; VI-FLUSH: buffer_load_dword v[[B_V2_F16:[0-9]+]]
 
 ; VI-DENORM: buffer_load_dword v[[A_V2_F16:[0-9]+]]
 ; VI-DENORM: buffer_load_dword v[[B_V2_F16:[0-9]+]]
 ; VI-DENORM: buffer_load_dword v[[C_V2_F16:[0-9]+]]
 
+; GFX10: buffer_load_dword v[[A_V2_F16:[0-9]+]]
+; GFX10: buffer_load_dword v[[B_V2_F16:[0-9]+]]
+; GFX10: buffer_load_dword v[[C_V2_F16:[0-9]+]]
 
 ; SI: v_cvt_f32_f16_e32 v[[A_F32_0:[0-9]+]], v[[A_V2_F16]]
 ; SI: v_lshrrev_b32_e32 v[[B_F16_1:[0-9]+]], 16, v[[B_V2_F16]]
 ; SI: v_lshrrev_b32_e32 v[[C_F16_1:[0-9]+]], 16, v[[C_V2_F16]]
 ; SI: v_lshrrev_b32_e32 v[[A_F16_1:[0-9]+]], 16, v[[A_V2_F16]]
 
 ; SI-DAG: v_cvt_f32_f16_e32 v[[B_F32_0:[0-9]+]], v[[B_V2_F16]]
 ; SI-DAG: v_cvt_f32_f16_e32 v[[C_F32_0:[0-9]+]], v[[C_V2_F16]]
 
-
 ; SI-DAG:  v_cvt_f32_f16_e32 v[[A_F32_1:[0-9]+]], v[[A_F16_1]]
 ; SI-DAG:  v_cvt_f32_f16_e32 v[[B_F32_1:[0-9]+]], v[[B_F16_1]]
 ; SI-DAG:  v_cvt_f32_f16_e32 v[[C_F32_1:[0-9]+]], v[[C_F16_1]]
 ; SI-DAG:  v_mac_f32_e32 v[[C_F32_0]], v[[A_F32_0]], v[[B_F32_0]]
 ; SI-DAG:  v_mac_f32_e32 v[[C_F32_1]], v[[A_F32_1]], v[[B_F32_1]]
 ; SI-DAG:  v_cvt_f16_f32_e32 v[[R_F16_1:[0-9]+]], v[[C_F32_1]]
 ; SI-DAG:  v_cvt_f16_f32_e32 v[[R_F16_LO:[0-9]+]], v[[C_F32_0]]
 ; SI-DAG:  v_lshlrev_b32_e32 v[[R_F16_HI:[0-9]+]], 16, v[[R_F16_1]]
 ; SI:  v_or_b32_e32 v[[R_V2_F16:[0-9]+]], v[[R_F16_LO]], v[[R_F16_HI]]
 
-
 ; VI-FLUSH:     v_lshrrev_b32_e32 v[[C_F16_1:[0-9]+]], 16, v[[C_V2_F16]]
 ; VI-FLUSH-DAG: v_mac_f16_sdwa v[[C_F16_1]], v[[A_V2_F16]], v[[B_V2_F16]] dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
 ; VI-FLUSH-DAG: v_mac_f16_e32 v[[C_V2_F16]], v[[A_V2_F16]], v[[B_V2_F16]]
 ; VI-FLUSH-DAG: v_lshlrev_b32_e32 v[[R_F16_HI:[0-9]+]], 16, v[[C_F16_1]]
 ; VI-FLUSH-NOT: v_and_b32
 ; VI-FLUSH:     v_or_b32_e32 v[[R_V2_F16:[0-9]+]], v[[C_V2_F16]], v[[R_F16_HI]]
 
 ; VI-DENORM-DAG: v_lshrrev_b32_e32 v[[A_F16_1:[0-9]+]], 16, v[[A_V2_F16]]
 ; VI-DENORM-DAG: v_lshrrev_b32_e32 v[[B_F16_1:[0-9]+]], 16, v[[B_V2_F16]]
 ; VI-DENORM-DAG: v_lshrrev_b32_e32 v[[C_F16_1:[0-9]+]], 16, v[[C_V2_F16]]
 ; VI-DENORM-DAG: v_fma_f16 v[[RES0:[0-9]+]], v[[A_V2_F16]], v[[B_V2_F16]], v[[C_V2_F16]]
 ; VI-DENORM-DAG: v_fma_f16 v[[RES1:[0-9]+]], v[[A_F16_1]], v[[B_F16_1]], v[[C_F16_1]]
 ; VI-DENORM-DAG: v_lshlrev_b32_e32 v[[R_F16_HI:[0-9]+]], 16, v[[RES1]]
 ; VI-DENORM-NOT: v_and_b32
 ; VI-DENORM: v_or_b32_e32 v[[R_V2_F16:[0-9]+]], v[[RES0]], v[[R_F16_HI]]
 
+; GFX10-FLUSH: v_pk_mul_f16 [[MUL:v[0-9]+]], v[[A_V2_F16]], v[[B_V2_F16]]
+; GFX10-FLUSH: v_pk_add_f16 v[[R_V2_F16:[0-9]+]], [[MUL]], v[[C_V2_F16]]
+
+; GFX10-DENORM: v_pk_fma_f16 v[[R_V2_F16:[0-9]+]], v[[A_V2_F16]], v[[B_V2_F16]], v[[C_V2_F16]]
+
 ; GCN: buffer_store_dword v[[R_V2_F16]]
 define amdgpu_kernel void @fmuladd_v2f16(
     <2 x half> addrspace(1)* %r,
     <2 x half> addrspace(1)* %a,
     <2 x half> addrspace(1)* %b,
     <2 x half> addrspace(1)* %c) {
   %a.val = load <2 x half>, <2 x half> addrspace(1)* %a
   %b.val = load <2 x half>, <2 x half> addrspace(1)* %b
   %c.val = load <2 x half>, <2 x half> addrspace(1)* %c
   %r.val = call <2 x half> @llvm.fmuladd.v2f16(<2 x half> %a.val, <2 x half> %b.val, <2 x half> %c.val)
   store <2 x half> %r.val, <2 x half> addrspace(1)* %r
   ret void
 }

llvm/trunk/test/CodeGen/AMDGPU/madak.ll

-; RUN: llc -march=amdgcn -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN,GFX6 %s
-; RUN: llc -march=amdgcn -mcpu=tonga -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN,GFX8,GFX8_9 %s
-; RUN: llc -march=amdgcn -mcpu=gfx900 -verify-machineinstrs -amdgpu-enable-global-sgpr-addr < %s | FileCheck -check-prefixes=GCN,GFX9,GFX8_9 %s
+; RUN: llc -march=amdgcn -mcpu=tahiti -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN,GFX6,GFX6_8_9,MAD %s
+; RUN: llc -march=amdgcn -mcpu=tonga -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN,GFX8,GFX6_8_9,GFX8_9,GFX8_9_10,MAD %s
+; RUN: llc -march=amdgcn -mcpu=gfx900 -verify-machineinstrs  -amdgpu-enable-global-sgpr-addr < %s | FileCheck -check-prefixes=GCN,GFX9,GFX6_8_9,GFX8_9,GFX8_9_10,MAD %s
+; RUN: llc -march=amdgcn -mcpu=gfx1010 -verify-machineinstrs -amdgpu-enable-global-sgpr-addr < %s | FileCheck -check-prefixes=GCN,GFX10,GFX8_9_10,MAD,GFX10-MAD %s
+; RUN: llc -march=amdgcn -mcpu=gfx1010 -verify-machineinstrs -fp-contract=fast -amdgpu-enable-global-sgpr-addr < %s | FileCheck -check-prefixes=GCN,GFX10,GFX8_9_10,FMA %s
 
 declare i32 @llvm.amdgcn.workitem.id.x() nounwind readnone
 declare float @llvm.fabs.f32(float) nounwind readnone
 
 ; GCN-LABEL: {{^}}madak_f32:
 ; GFX6:   buffer_load_dword [[VA:v[0-9]+]]
 ; GFX6:   buffer_load_dword [[VB:v[0-9]+]]
 ; GFX8: {{flat|global}}_load_dword [[VB:v[0-9]+]]
 ; GFX8: {{flat|global}}_load_dword [[VA:v[0-9]+]]
 ; GFX9: {{flat|global}}_load_dword [[VA:v[0-9]+]]
 ; GFX9: {{flat|global}}_load_dword [[VB:v[0-9]+]]
-; GCN:    v_madak_f32 {{v[0-9]+}}, [[VA]], [[VB]], 0x41200000
+; GFX10: {{flat|global}}_load_dword [[VA:v[0-9]+]]
+; GFX10: {{flat|global}}_load_dword [[VB:v[0-9]+]]
+; MAD:   v_madak_f32 {{v[0-9]+}}, [[VA]], [[VB]], 0x41200000
+; FMA:   v_fmaak_f32 {{v[0-9]+}}, [[VA]], [[VB]], 0x41200000
 define amdgpu_kernel void @madak_f32(float addrspace(1)* noalias %out, float addrspace(1)* noalias %in.a, float addrspace(1)* noalias %in.b) nounwind {
   %tid = tail call i32 @llvm.amdgcn.workitem.id.x() nounwind readnone
   %in.a.gep = getelementptr float, float addrspace(1)* %in.a, i32 %tid
   %in.b.gep = getelementptr float, float addrspace(1)* %in.b, i32 %tid
   %out.gep = getelementptr float, float addrspace(1)* %out, i32 %tid
 
   %a = load float, float addrspace(1)* %in.a.gep, align 4
   %b = load float, float addrspace(1)* %in.b.gep, align 4
 
   %mul = fmul float %a, %b
   %madak = fadd float %mul, 10.0
   store float %madak, float addrspace(1)* %out.gep, align 4
   ret void
 }
 
 ; Make sure this is only folded with one use. This is a code size
 ; optimization and if we fold the immediate multiple times, we'll undo
 ; it.
 
 ; GCN-LABEL: {{^}}madak_2_use_f32:
-; GFX8_9:   v_mov_b32_e32 [[VK:v[0-9]+]], 0x41200000
-; GFX6-DAG: buffer_load_dword [[VA:v[0-9]+]], {{v\[[0-9]+:[0-9]+\]}}, {{s\[[0-9]+:[0-9]+\]}}, 0 addr64{{$}}
-; GFX6-DAG: buffer_load_dword [[VB:v[0-9]+]], {{v\[[0-9]+:[0-9]+\]}}, {{s\[[0-9]+:[0-9]+\]}}, 0 addr64 offset:4
-; GFX6-DAG: buffer_load_dword [[VC:v[0-9]+]], {{v\[[0-9]+:[0-9]+\]}}, {{s\[[0-9]+:[0-9]+\]}}, 0 addr64 offset:8
-; GFX8_9:   {{flat|global}}_load_dword [[VA:v[0-9]+]], {{v\[[0-9]+:[0-9]+\]}}
-; GFX8_9:   {{flat|global}}_load_dword [[VB:v[0-9]+]], {{v\[[0-9]+:[0-9]+\]}}
-; GFX8_9:   {{flat|global}}_load_dword [[VC:v[0-9]+]], {{v\[[0-9]+:[0-9]+\]}}
-; GFX6-DAG: v_mov_b32_e32 [[VK:v[0-9]+]], 0x41200000
-; GCN-DAG:  v_madak_f32 {{v[0-9]+}}, [[VA]], [[VB]], 0x41200000
-; GCN-DAG:  v_mac_f32_e32 [[VK]], [[VA]], [[VC]]
-; GCN: s_endpgm
+; GFX8_9_10:    v_mov_b32_e32 [[VK:v[0-9]+]], 0x41200000
+; GFX6-DAG:     buffer_load_dword [[VA:v[0-9]+]], {{v\[[0-9]+:[0-9]+\]}}, {{s\[[0-9]+:[0-9]+\]}}, 0 addr64{{$}}
+; GFX6-DAG:     buffer_load_dword [[VB:v[0-9]+]], {{v\[[0-9]+:[0-9]+\]}}, {{s\[[0-9]+:[0-9]+\]}}, 0 addr64 offset:4
+; GFX6-DAG:     buffer_load_dword [[VC:v[0-9]+]], {{v\[[0-9]+:[0-9]+\]}}, {{s\[[0-9]+:[0-9]+\]}}, 0 addr64 offset:8
+; GFX8_9_10:    {{flat|global}}_load_dword [[VA:v[0-9]+]], {{v\[[0-9]+:[0-9]+\]}}
+; GFX8_9_10:    {{flat|global}}_load_dword [[VB:v[0-9]+]], {{v\[[0-9]+:[0-9]+\]}}
+; GFX8_9_10:    {{flat|global}}_load_dword [[VC:v[0-9]+]], {{v\[[0-9]+:[0-9]+\]}}
+; GFX6-DAG:     v_mov_b32_e32 [[VK:v[0-9]+]], 0x41200000
+; GFX6_8_9-DAG: v_madak_f32 {{v[0-9]+}}, [[VA]], [[VB]], 0x41200000
+; GFX10-MAD-DAG:v_madak_f32 {{v[0-9]+}}, [[VA]], [[VB]], 0x41200000
+; FMA-DAG:      v_fmaak_f32 {{v[0-9]+}}, [[VA]], [[VB]], 0x41200000
+; MAD-DAG:      v_mac_f32_e32 [[VK]], [[VA]], [[VC]]
+; FMA-DAG:      v_fmac_f32_e32 [[VK]], [[VA]], [[VC]]
+; GCN:          s_endpgm
 define amdgpu_kernel void @madak_2_use_f32(float addrspace(1)* noalias %out, float addrspace(1)* noalias %in) nounwind {
   %tid = tail call i32 @llvm.amdgcn.workitem.id.x() nounwind readnone
 
   %in.gep.0 = getelementptr float, float addrspace(1)* %in, i32 %tid
   %in.gep.1 = getelementptr float, float addrspace(1)* %in.gep.0, i32 1
   %in.gep.2 = getelementptr float, float addrspace(1)* %in.gep.0, i32 2
 
   %out.gep.0 = getelementptr float, float addrspace(1)* %out, i32 %tid
define amdgpu_kernel void @madak_2_use_f32(float addrspace(1)* noalias %out, float addrspace(1)* noalias %in) nounwind {
 
   store volatile float %madak0, float addrspace(1)* %out.gep.0, align 4
   store volatile float %madak1, float addrspace(1)* %out.gep.1, align 4
   ret void
 }
 
 ; GCN-LABEL: {{^}}madak_m_inline_imm_f32:
 ; GCN: {{buffer|flat|global}}_load_dword [[VA:v[0-9]+]]
-; GCN: v_madak_f32 {{v[0-9]+}}, 4.0, [[VA]], 0x41200000
+; MAD: v_madak_f32 {{v[0-9]+}}, 4.0, [[VA]], 0x41200000
+; FMA: v_fmaak_f32 {{v[0-9]+}}, 4.0, [[VA]], 0x41200000
 define amdgpu_kernel void @madak_m_inline_imm_f32(float addrspace(1)* noalias %out, float addrspace(1)* noalias %in.a) nounwind {
   %tid = tail call i32 @llvm.amdgcn.workitem.id.x() nounwind readnone
   %in.a.gep = getelementptr float, float addrspace(1)* %in.a, i32 %tid
   %out.gep = getelementptr float, float addrspace(1)* %out, i32 %tid
 
   %a = load float, float addrspace(1)* %in.a.gep, align 4
 
   %mul = fmul float 4.0, %a
   %madak = fadd float %mul, 10.0
   store float %madak, float addrspace(1)* %out.gep, align 4
   ret void
 }
 
 ; Make sure nothing weird happens with a value that is also allowed as
 ; an inline immediate.
 
 ; GCN-LABEL: {{^}}madak_inline_imm_f32:
 ; GFX6:   buffer_load_dword [[VA:v[0-9]+]]
 ; GFX6:   buffer_load_dword [[VB:v[0-9]+]]
 ; GFX8: {{flat|global}}_load_dword [[VB:v[0-9]+]]
 ; GFX8: {{flat|global}}_load_dword [[VA:v[0-9]+]]
 ; GFX9: {{flat|global}}_load_dword [[VA:v[0-9]+]]
 ; GFX9: {{flat|global}}_load_dword [[VB:v[0-9]+]]
-; GCN:    v_mad_f32 {{v[0-9]+}}, [[VA]], [[VB]], 4.0
+; GFX10: {{flat|global}}_load_dword [[VA:v[0-9]+]]
+; GFX10: {{flat|global}}_load_dword [[VB:v[0-9]+]]
+; MAD:   v_mad_f32 {{v[0-9]+}}, [[VA]], [[VB]], 4.0
+; FMA:   v_fma_f32 {{v[0-9]+}}, [[VA]], [[VB]], 4.0
 define amdgpu_kernel void @madak_inline_imm_f32(float addrspace(1)* noalias %out, float addrspace(1)* noalias %in.a, float addrspace(1)* noalias %in.b) nounwind {
   %tid = tail call i32 @llvm.amdgcn.workitem.id.x() nounwind readnone
   %in.a.gep = getelementptr float, float addrspace(1)* %in.a, i32 %tid
   %in.b.gep = getelementptr float, float addrspace(1)* %in.b, i32 %tid
   %out.gep = getelementptr float, float addrspace(1)* %out, i32 %tid
 
   %a = load float, float addrspace(1)* %in.a.gep, align 4
   %b = load float, float addrspace(1)* %in.b.gep, align 4
 
   %mul = fmul float %a, %b
   %madak = fadd float %mul, 4.0
   store float %madak, float addrspace(1)* %out.gep, align 4
   ret void
 }
 
 ; We can't use an SGPR when forming madak
 ; GCN-LABEL: {{^}}s_v_madak_f32:
-; GCN-DAG: s_load_dword [[SB:s[0-9]+]]
-; GCN-DAG: v_mov_b32_e32 [[VK:v[0-9]+]], 0x41200000
-; GCN-DAG: {{buffer|flat|global}}_load_dword [[VA:v[0-9]+]]
-; GCN-NOT: v_madak_f32
-; GCN: v_mac_f32_e32 [[VK]], [[SB]], [[VA]]
+; GCN-DAG:      s_load_dword [[SB:s[0-9]+]]
+; GFX6_8_9-DAG: v_mov_b32_e32 [[VK:v[0-9]+]], 0x41200000
+; GCN-DAG:      {{buffer|flat|global}}_load_dword{{(_addtid)?}} [[VA:v[0-9]+]]
+; GCN-NOT:      v_madak_f32
+; GFX6_8_9:     v_mac_f32_e32 [[VK]], [[SB]], [[VA]]
+; GFX10-MAD:    v_mad_f32 v{{[0-9]+}}, [[VA]], [[SB]], 0x41200000
+; FMA:          v_fma_f32 v{{[0-9]+}}, [[VA]], [[SB]], 0x41200000
 define amdgpu_kernel void @s_v_madak_f32(float addrspace(1)* noalias %out, float addrspace(1)* noalias %in.a, float %b) nounwind {
   %tid = tail call i32 @llvm.amdgcn.workitem.id.x() nounwind readnone
   %in.a.gep = getelementptr float, float addrspace(1)* %in.a, i32 %tid
   %out.gep = getelementptr float, float addrspace(1)* %out, i32 %tid
 
   %a = load float, float addrspace(1)* %in.a.gep, align 4
 
   %mul = fmul float %a, %b
   %madak = fadd float %mul, 10.0
   store float %madak, float addrspace(1)* %out.gep, align 4
   ret void
 }
 
 ; GCN-LABEL: @v_s_madak_f32
-; GCN-DAG: s_load_dword [[SB:s[0-9]+]]
-; GCN-DAG: v_mov_b32_e32 [[VK:v[0-9]+]], 0x41200000
-; GCN-DAG: {{buffer|flat|global}}_load_dword [[VA:v[0-9]+]]
-; GCN-NOT: v_madak_f32
-; GCN: v_mac_f32_e32 [[VK]], [[SB]], [[VA]]
+; GCN-DAG:       s_load_dword [[SB:s[0-9]+]]
+; GFX6_8_9-DAG:  v_mov_b32_e32 [[VK:v[0-9]+]], 0x41200000
+; GCN-DAG:       {{buffer|flat|global}}_load_dword{{(_addtid)?}} [[VA:v[0-9]+]]
+; GFX6_8_9-NOT:  v_madak_f32
+; GFX6_8_9:      v_mac_f32_e32 [[VK]], [[SB]], [[VA]]
+; GFX10-MAD:     v_madak_f32 v{{[0-9]+}}, [[SB]], [[VA]], 0x41200000
+; FMA:           v_fmaak_f32 v{{[0-9]+}}, [[SB]], [[VA]], 0x41200000
 define amdgpu_kernel void @v_s_madak_f32(float addrspace(1)* noalias %out, float %a, float addrspace(1)* noalias %in.b) nounwind {
   %tid = tail call i32 @llvm.amdgcn.workitem.id.x() nounwind readnone
   %in.b.gep = getelementptr float, float addrspace(1)* %in.b, i32 %tid
   %out.gep = getelementptr float, float addrspace(1)* %out, i32 %tid
 
   %b = load float, float addrspace(1)* %in.b.gep, align 4
 
   %mul = fmul float %a, %b
   %madak = fadd float %mul, 10.0
   store float %madak, float addrspace(1)* %out.gep, align 4
   ret void
 }
 
 ; GCN-LABEL: {{^}}s_s_madak_f32:
 ; GCN-NOT: v_madak_f32
-; GCN: v_mac_f32_e32 {{v[0-9]+}}, {{s[0-9]+}}, {{v[0-9]+}}
+; GFX8_9:  v_mac_f32_e32 {{v[0-9]+}}, {{s[0-9]+}}, {{v[0-9]+}}
+; GFX10-MAD: v_mac_f32_e64 {{v[0-9]+}}, {{s[0-9]+}}, {{s[0-9]+}}
+; FMA:       v_fmac_f32_e64 {{v[0-9]+}}, {{s[0-9]+}}, {{s[0-9]+}}
 define amdgpu_kernel void @s_s_madak_f32(float addrspace(1)* %out, float %a, float %b) nounwind {
   %mul = fmul float %a, %b
   %madak = fadd float %mul, 10.0
   store float %madak, float addrspace(1)* %out, align 4
   ret void
 }
 
 ; GCN-LABEL: {{^}}no_madak_src0_modifier_f32:
-; GFX6:   buffer_load_dword [[VA:v[0-9]+]]
-; GFX6:   buffer_load_dword [[VB:v[0-9]+]]
-; GFX8_9: {{flat|global}}_load_dword [[VB:v[0-9]+]]
-; GFX8_9: {{flat|global}}_load_dword [[VA:v[0-9]+]]
-; GCN:    v_mad_f32 {{v[0-9]+}}, |{{v[0-9]+}}|, {{v[0-9]+}}, {{[sv][0-9]+}}
-; GCN:    s_endpgm
+; GFX6:      buffer_load_dword [[VA:v[0-9]+]]
+; GFX6:      buffer_load_dword [[VB:v[0-9]+]]
+; GFX8_9_10: {{flat|global}}_load_dword [[VB:v[0-9]+]]
+; GFX8_9_10: {{flat|global}}_load_dword [[VA:v[0-9]+]]
+; GFX6_8_9:  v_mad_f32 {{v[0-9]+}}, |{{v[0-9]+}}|, {{v[0-9]+}}, {{[sv][0-9]+}}
+; GFX10-MAD: v_mad_f32 {{v[0-9]+}}, |{{v[0-9]+}}|, {{v[0-9]+}}, 0x41200000
+; FMA:       v_fma_f32 {{v[0-9]+}}, |{{v[0-9]+}}|, {{v[0-9]+}}, 0x41200000
+; GCN:       s_endpgm
 define amdgpu_kernel void @no_madak_src0_modifier_f32(float addrspace(1)* noalias %out, float addrspace(1)* noalias %in.a, float addrspace(1)* noalias %in.b) nounwind {
   %tid = tail call i32 @llvm.amdgcn.workitem.id.x() nounwind readnone
   %in.a.gep = getelementptr float, float addrspace(1)* %in.a, i32 %tid
   %in.b.gep = getelementptr float, float addrspace(1)* %in.b, i32 %tid
   %out.gep = getelementptr float, float addrspace(1)* %out, i32 %tid
 
   %a = load float, float addrspace(1)* %in.a.gep, align 4
   %b = load float, float addrspace(1)* %in.b.gep, align 4
 
   %a.fabs = call float @llvm.fabs.f32(float %a) nounwind readnone
 
   %mul = fmul float %a.fabs, %b
   %madak = fadd float %mul, 10.0
   store float %madak, float addrspace(1)* %out.gep, align 4
   ret void
 }
 
 ; GCN-LABEL: {{^}}no_madak_src1_modifier_f32:
-; GFX6:   buffer_load_dword [[VA:v[0-9]+]]
-; GFX6:   buffer_load_dword [[VB:v[0-9]+]]
-; GFX8_9: {{flat|global}}_load_dword [[VB:v[0-9]+]]
-; GFX8_9: {{flat|global}}_load_dword [[VA:v[0-9]+]]
-; GCN:    v_mad_f32 {{v[0-9]+}}, {{v[0-9]+}}, |{{v[0-9]+}}|, {{[sv][0-9]+}}
-; GCN:    s_endpgm
+; GFX6:      buffer_load_dword [[VA:v[0-9]+]]
+; GFX6:      buffer_load_dword [[VB:v[0-9]+]]
+; GFX8_9_10: {{flat|global}}_load_dword [[VB:v[0-9]+]]
+; GFX8_9_10: {{flat|global}}_load_dword [[VA:v[0-9]+]]
+; GFX6_8_9:  v_mad_f32 {{v[0-9]+}}, {{v[0-9]+}}, |{{v[0-9]+}}|, {{[sv][0-9]+}}
+; GFX10-MAD: v_mad_f32 {{v[0-9]+}}, {{v[0-9]+}}, |{{v[0-9]+}}|, 0x41200000
+; FMA:       v_fma_f32 {{v[0-9]+}}, {{v[0-9]+}}, |{{v[0-9]+}}|, 0x41200000
+; GCN:       s_endpgm
 define amdgpu_kernel void @no_madak_src1_modifier_f32(float addrspace(1)* noalias %out, float addrspace(1)* noalias %in.a, float addrspace(1)* noalias %in.b) nounwind {
   %tid = tail call i32 @llvm.amdgcn.workitem.id.x() nounwind readnone
   %in.a.gep = getelementptr float, float addrspace(1)* %in.a, i32 %tid
   %in.b.gep = getelementptr float, float addrspace(1)* %in.b, i32 %tid
   %out.gep = getelementptr float, float addrspace(1)* %out, i32 %tid
 
   %a = load float, float addrspace(1)* %in.a.gep, align 4
   %b = load float, float addrspace(1)* %in.b.gep, align 4
 
   %b.fabs = call float @llvm.fabs.f32(float %b) nounwind readnone
 
   %mul = fmul float %a, %b.fabs
   %madak = fadd float %mul, 10.0
   store float %madak, float addrspace(1)* %out.gep, align 4
   ret void
 }
 
-; SIFoldOperands should not fold the SGPR copy into the instruction
+; SIFoldOperands should not fold the SGPR copy into the instruction before GFX10
 ; because the implicit immediate already uses the constant bus.
+; On GFX10+ we can use two scalar operands.
 ; GCN-LABEL: {{^}}madak_constant_bus_violation:
-; GCN:    s_load_dword [[SGPR0:s[0-9]+]], s{{\[[0-9]+:[0-9]+\]}}, {{0x12|0x48}}
-; GCN:    v_mov_b32_e32 [[SGPR0_VCOPY:v[0-9]+]], [[SGPR0]]
-; GCN:    {{buffer|flat|global}}_load_dword [[VGPR:v[0-9]+]]
-; GCN:    v_madak_f32 [[MADAK:v[0-9]+]], 0.5, [[SGPR0_VCOPY]], 0x42280000
-; GCN:    v_mul_f32_e32 [[MUL:v[0-9]+]], [[MADAK]], [[VGPR]]
-; GFX6:   buffer_store_dword [[MUL]]
-; GFX8_9: {{flat|global}}_store_dword v[{{[0-9:]+}}], [[MUL]]
+; GCN:       s_load_dword [[SGPR0:s[0-9]+]], s{{\[[0-9]+:[0-9]+\]}}, {{0x12|0x48}}
+; GCN:       v_mov_b32_e32 [[SGPR0_VCOPY:v[0-9]+]], [[SGPR0]]
+; GCN:       {{buffer|flat|global}}_load_dword [[VGPR:v[0-9]+]]
+; MAD:       v_madak_f32 [[MADAK:v[0-9]+]], 0.5, [[SGPR0_VCOPY]], 0x42280000
+; FMA:       v_fmaak_f32 [[MADAK:v[0-9]+]], 0.5, [[SGPR0_VCOPY]], 0x42280000
+; GCN:       v_mul_f32_e32 [[MUL:v[0-9]+]], [[MADAK]], [[VGPR]]
+; GFX6:      buffer_store_dword [[MUL]]
+; GFX8_9_10: {{flat|global}}_store_dword v[{{[0-9:]+}}], [[MUL]]
 define amdgpu_kernel void @madak_constant_bus_violation(i32 %arg1, [8 x i32], float %sgpr0, float %sgpr1) #0 {
 bb:
   %tmp = icmp eq i32 %arg1, 0
   br i1 %tmp, label %bb3, label %bb4
 
 bb3:
   store volatile float 0.0, float addrspace(1)* undef
   br label %bb4

llvm/trunk/test/CodeGen/AMDGPU/sdwa-peephole-instr-gfx10.mir

+# RUN: llc -march=amdgcn -mcpu=gfx1010 -run-pass=si-peephole-sdwa -verify-machineinstrs -o - %s | FileCheck -check-prefix=GFX1010 -check-prefix=GCN %s
+
+# GCN-LABEL: {{^}}name: vop1_instructions
+
+# GFX1010: %{{[0-9]+}}:vgpr_32 = V_MOV_B32_sdwa 0, %{{[0-9]+}}, 0, 5, 0, 5, implicit $exec
+# GFX1010: %{{[0-9]+}}:vgpr_32 = V_FRACT_F32_sdwa 0, %{{[0-9]+}}, 0, 0, 5, 0, 6, implicit $exec
+# GFX1010: %{{[0-9]+}}:vgpr_32 = V_SIN_F32_sdwa 0, %{{[0-9]+}}, 0, 0, 5, 0, 5, implicit $exec
+# GFX1010: %{{[0-9]+}}:vgpr_32 = V_CVT_U32_F32_sdwa 0, %{{[0-9]+}}, 0, 5, 0, 5, implicit $exec
+# GFX1010: %{{[0-9]+}}:vgpr_32 = V_CVT_F32_I32_sdwa 0, %{{[0-9]+}}, 0, 0, 5, 0, 6, implicit $exec
+
+# GFX1010: %{{[0-9]+}}:vgpr_32 = V_MOV_B32_sdwa 0, %{{[0-9]+}}, 0, 6, 0, 5, implicit $exec
+# GFX1010: %{{[0-9]+}}:vgpr_32 = V_FRACT_F32_sdwa 0, %{{[0-9]+}}, 0, 0, 5, 0, 6, implicit $exec
+# GFX1010: %{{[0-9]+}}:vgpr_32 = V_SIN_F32_sdwa 0, %{{[0-9]+}}, 0, 0, 5, 0, 5, implicit $exec
+# GFX1010: %{{[0-9]+}}:vgpr_32 = V_CVT_U32_F32_sdwa 0, %{{[0-9]+}}, 0, 5, 0, 5, implicit $exec
+# GFX1010: %{{[0-9]+}}:vgpr_32 = V_CVT_F32_I32_sdwa 0, %{{[0-9]+}}, 0, 0, 5, 0, 6, implicit $exec
+
+# GFX1010: %{{[0-9]+}}:vgpr_32 = V_FRACT_F32_sdwa 1, %{{[0-9]+}}, 0, 0, 5, 0, 5, implicit $exec
+# GFX1010: %{{[0-9]+}}:vgpr_32 = V_SIN_F32_sdwa 0, %{{[0-9]+}}, 1, 0, 5, 0, 5, implicit $exec
+# GFX1010: %{{[0-9]+}}:vgpr_32 = V_CVT_U32_F32_sdwa 1, %{{[0-9]+}}, 0, 5, 0, 5, implicit $exec
+# GFX1010: %{{[0-9]+}}:vgpr_32 = V_CVT_F32_I32_sdwa 0, %{{[0-9]+}}, 0, 1, 5, 0, 5, implicit $exec
+
+---
+name:            vop1_instructions
+tracksRegLiveness: true
+registers:
+  - { id: 0, class: vreg_64 }
+  - { id: 1, class: vreg_64 }
+  - { id: 2, class: sreg_64 }
+  - { id: 3, class: vgpr_32 }
+  - { id: 4, class: sreg_32_xm0 }
+  - { id: 5, class: sreg_32_xm0 }
+  - { id: 6, class: sreg_32_xm0 }
+  - { id: 7, class: sreg_32_xm0 }
+  - { id: 8, class: sreg_32 }
+  - { id: 9, class: vgpr_32 }
+  - { id: 10, class: vgpr_32 }
+  - { id: 11, class: vgpr_32 }
+  - { id: 12, class: vgpr_32 }
+  - { id: 13, class: vgpr_32 }
+  - { id: 14, class: vgpr_32 }
+  - { id: 15, class: vgpr_32 }
+  - { id: 16, class: vgpr_32 }
+  - { id: 17, class: vgpr_32 }
+  - { id: 18, class: vgpr_32 }
+  - { id: 19, class: vgpr_32 }
+  - { id: 20, class: vgpr_32 }
+  - { id: 21, class: vgpr_32 }
+  - { id: 22, class: vgpr_32 }
+  - { id: 23, class: vgpr_32 }
+  - { id: 24, class: vgpr_32 }
+  - { id: 25, class: vgpr_32 }
+  - { id: 26, class: vgpr_32 }
+  - { id: 27, class: vgpr_32 }
+  - { id: 28, class: vgpr_32 }
+  - { id: 29, class: vgpr_32 }
+  - { id: 30, class: vgpr_32 }
+  - { id: 31, class: vgpr_32 }
+  - { id: 32, class: vgpr_32 }
+  - { id: 33, class: vgpr_32 }
+  - { id: 34, class: vgpr_32 }
+  - { id: 35, class: vgpr_32 }
+  - { id: 36, class: vgpr_32 }
+  - { id: 37, class: vgpr_32 }
+  - { id: 38, class: vgpr_32 }
+  - { id: 39, class: vgpr_32 }
+  - { id: 40, class: vgpr_32 }
+  - { id: 41, class: vgpr_32 }
+  - { id: 42, class: vgpr_32 }
+  - { id: 43, class: vgpr_32 }
+  - { id: 44, class: vgpr_32 }
+  - { id: 45, class: vgpr_32 }
+  - { id: 46, class: vgpr_32 }
+  - { id: 47, class: vgpr_32 }
+  - { id: 48, class: vgpr_32 }
+  - { id: 100, class: vgpr_32 }
+body:             |
+  bb.0:
+    liveins: $vgpr0_vgpr1, $vgpr2_vgpr3, $sgpr30_sgpr31
+
+    %2 = COPY $sgpr30_sgpr31
+    %1 = COPY $vgpr2_vgpr3
+    %0 = COPY $vgpr0_vgpr1
+    %3 = FLAT_LOAD_DWORD %1, 0, 0, 0, 0, implicit $exec, implicit $flat_scr :: (load 4)
+
+    %5 = S_MOV_B32 65535
+    %6 = S_MOV_B32 65535
+
+    %10 = V_LSHRREV_B32_e64 16, %3, implicit $exec
+    %11 = V_MOV_B32_e32 %10, implicit $exec
+    %12 = V_LSHLREV_B32_e64 16, %11, implicit $exec
+    %14 = V_FRACT_F32_e32 123, implicit $exec
+    %15 = V_LSHLREV_B32_e64 16, %14, implicit $exec
+    %16 = V_LSHRREV_B32_e64 16, %15, implicit $exec
+    %17 = V_SIN_F32_e32 %16, implicit $exec
+    %18 = V_LSHLREV_B32_e64 16, %17, implicit $exec
+    %19 = V_LSHRREV_B32_e64 16, %18, implicit $exec
+    %20 = V_CVT_U32_F32_e32 %19, implicit $exec
+    %21 = V_LSHLREV_B32_e64 16, %20, implicit $exec
+    %23 = V_CVT_F32_I32_e32 123, implicit $exec
+    %24 = V_LSHLREV_B32_e64 16, %23, implicit $exec
+
+    %25 = V_LSHRREV_B32_e64 16, %3, implicit $exec
+    %26 = V_MOV_B32_e64 %25, implicit $exec
+    %26 = V_LSHLREV_B32_e64 16, %26, implicit $exec
+    %27 = V_FRACT_F32_e64 0, %6, 0, 0, implicit $exec
+    %28 = V_LSHLREV_B32_e64 16, %27, implicit $exec
+    %29 = V_LSHRREV_B32_e64 16, %28, implicit $exec
+    %30 = V_SIN_F32_e64 0, %29, 0, 0, implicit $exec
+    %31 = V_LSHLREV_B32_e64 16, %30, implicit $exec
+    %32 = V_LSHRREV_B32_e64 16, %31, implicit $exec
+    %33 = V_CVT_U32_F32_e64 0, %32, 0, 0, implicit $exec
+    %34 = V_LSHLREV_B32_e64 16, %33, implicit $exec
+    %35 = V_CVT_F32_I32_e64 %6, 0, 0, implicit $exec
+    %36 = V_LSHLREV_B32_e64 16, %35, implicit $exec
+
+
+    %37 = V_LSHRREV_B32_e64 16, %36, implicit $exec
+    %38 = V_FRACT_F32_e64 1, %37, 0, 0, implicit $exec
+    %39 = V_LSHLREV_B32_e64 16, %38, implicit $exec
+    %40 = V_LSHRREV_B32_e64 16, %39, implicit $exec
+    %41 = V_SIN_F32_e64 0, %40, 1, 0, implicit $exec
+    %42 = V_LSHLREV_B32_e64 16, %41, implicit $exec
+    %43 = V_LSHRREV_B32_e64 16, %42, implicit $exec
+    %44 = V_CVT_U32_F32_e64 1, %43, 0, 0, implicit $exec
+    %45 = V_LSHLREV_B32_e64 16, %44, implicit $exec
+    %46 = V_LSHRREV_B32_e64 16, %45, implicit $exec
+    %47 = V_CVT_F32_I32_e64 %46, 0, 1, implicit $exec
+    %48 = V_LSHLREV_B32_e64 16, %47, implicit $exec
+
+
+    %100 = V_MOV_B32_e32 %48, implicit $exec
+
+    FLAT_STORE_DWORD %0, %100, 0, 0, 0, 0, implicit $exec, implicit $flat_scr :: (store 4)
+    $sgpr30_sgpr31 = COPY %2
+    S_SETPC_B64_return $sgpr30_sgpr31
+
+...
+---
+# GCN-LABEL: {{^}}name: vop2_instructions
+
+# GFX1010: %{{[0-9]+}}:vgpr_32 = V_AND_B32_sdwa 0, %{{[0-9]+}}, 0, %{{[0-9]+}}, 0, 5, 0, 6, 5, implicit $exec
+# GFX1010: %{{[0-9]+}}:vgpr_32 = V_ADD_F32_sdwa 0, %{{[0-9]+}}, 0, %{{[0-9]+}}, 0, 0, 5, 0, 5, 1, implicit $exec
+# GFX1010: %{{[0-9]+}}:vgpr_32 = V_SUB_F16_sdwa 0, %{{[0-9]+}}, 0, %{{[0-9]+}}, 0, 0, 6, 0, 5, 1, implicit $exec
+# GFX1010: %{{[0-9]+}}:vgpr_32 = V_FMAC_F32_e32 %{{[0-9]+}}, %{{[0-9]+}}, %{{[0-9]+}}, implicit $exec
+# GFX1010: %{{[0-9]+}}:vgpr_32 = V_FMAC_F16_e32 %{{[0-9]+}}, %{{[0-9]+}}, %{{[0-9]+}}, implicit $exec
+
+# GFX1010: %{{[0-9]+}}:vgpr_32 = V_AND_B32_sdwa 0, %{{[0-9]+}}, 0, %{{[0-9]+}}, 0, 5, 0, 6, 5, implicit $exec
+# GFX1010: %{{[0-9]+}}:vgpr_32 = V_ADD_F32_sdwa 0, %{{[0-9]+}}, 0, %{{[0-9]+}}, 0, 0, 5, 0, 5, 1, implicit $exec
+# GFX1010: %{{[0-9]+}}:vgpr_32 = V_SUB_F16_sdwa 0, %{{[0-9]+}}, 0, %{{[0-9]+}}, 0, 0, 5, 0, 6, 1, implicit $exec
+# GFX1010: %{{[0-9]+}}:vgpr_32 = V_FMAC_F32_e64 0, 23, 0, %{{[0-9]+}}, 0, %{{[0-9]+}}, 0, 0, implicit $exec
+# GFX1010: %{{[0-9]+}}:vgpr_32 = V_FMAC_F16_e64 0, %{{[0-9]+}}, 0, %{{[0-9]+}}, 0, %{{[0-9]+}}, 0, 0, implicit $exec
+
+# GFX1010: %{{[0-9]+}}:vgpr_32 = V_ADD_F32_sdwa 0, %{{[0-9]+}}, 1, %{{[0-9]+}}, 0, 0, 5, 0, 5, 1, implicit $exec
+# GFX1010: %{{[0-9]+}}:vgpr_32 = V_SUB_F16_sdwa 1, %{{[0-9]+}}, 1, %{{[0-9]+}}, 0, 0, 5, 0, 6, 1, implicit $exec
+# GFX1010: %{{[0-9]+}}:vgpr_32 = V_FMAC_F32_e64 1, 23, 1, %{{[0-9]+}}, 1, %{{[0-9]+}}, 1, 0, implicit $exec
+# GFX1010: %{{[0-9]+}}:vgpr_32 = V_FMAC_F16_e64 1, %{{[0-9]+}}, 1, %{{[0-9]+}}, 1, %{{[0-9]+}}, 0, 2, implicit $exec
+
+name:            vop2_instructions
+tracksRegLiveness: true
+registers:
+  - { id: 0, class: vreg_64 }
+  - { id: 1, class: vreg_64 }
+  - { id: 2, class: sreg_64 }
+  - { id: 3, class: vgpr_32 }
+  - { id: 4, class: sreg_32_xm0 }
+  - { id: 5, class: sreg_32_xm0 }
+  - { id: 6, class: sreg_32_xm0 }
+  - { id: 7, class: sreg_32_xm0 }
+  - { id: 8, class: sreg_32 }
+  - { id: 9, class: vgpr_32 }
+  - { id: 10, class: vgpr_32 }
+  - { id: 11, class: vgpr_32 }
+  - { id: 12, class: vgpr_32 }
+  - { id: 13, class: vgpr_32 }
+  - { id: 14, class: vgpr_32 }
+  - { id: 15, class: vgpr_32 }
+  - { id: 16, class: vgpr_32 }
+  - { id: 17, class: vgpr_32 }
+  - { id: 18, class: vgpr_32 }
+  - { id: 19, class: vgpr_32 }
+  - { id: 20, class: vgpr_32 }
+  - { id: 21, class: vgpr_32 }
+  - { id: 22, class: vgpr_32 }
+  - { id: 23, class: vgpr_32 }
+  - { id: 24, class: vgpr_32 }
+  - { id: 25, class: vgpr_32 }
+  - { id: 26, class: vgpr_32 }
+  - { id: 27, class: vgpr_32 }
+  - { id: 28, class: vgpr_32 }
+  - { id: 29, class: vgpr_32 }
+  - { id: 30, class: vgpr_32 }
+  - { id: 31, class: vgpr_32 }
+  - { id: 32, class: vgpr_32 }
+  - { id: 33, class: vgpr_32 }
+  - { id: 34, class: vgpr_32 }
+  - { id: 35, class: vgpr_32 }
+  - { id: 36, class: vgpr_32 }
+  - { id: 37, class: vgpr_32 }
+  - { id: 38, class: vgpr_32 }
+  - { id: 39, class: vgpr_32 }
+  - { id: 40, class: vgpr_32 }
+  - { id: 41, class: vgpr_32 }
+  - { id: 42, class: vgpr_32 }
+  - { id: 43, class: vgpr_32 }
+  - { id: 44, class: vgpr_32 }
+  - { id: 45, class: vgpr_32 }
+  - { id: 46, class: vgpr_32 }
+  - { id: 47, class: vgpr_32 }
+  - { id: 48, class: vgpr_32 }
+  - { id: 49, class: vgpr_32 }
+  - { id: 50, class: vgpr_32 }
+  - { id: 51, class: vgpr_32 }
+  - { id: 52, class: vgpr_32 }
+  - { id: 53, class: vgpr_32 }
+  - { id: 54, class: vgpr_32 }
+  - { id: 55, class: vgpr_32 }
+  - { id: 56, class: vgpr_32 }
+  - { id: 57, class: vgpr_32 }
+  - { id: 58, class: vgpr_32 }
+  - { id: 59, class: vgpr_32 }
+  - { id: 60, class: vgpr_32 }
+  - { id: 100, class: vgpr_32 }
+body:             |
+  bb.0:
+    liveins: $vgpr0_vgpr1, $vgpr2_vgpr3, $sgpr30_sgpr31
+
+    %2 = COPY $sgpr30_sgpr31
+    %1 = COPY $vgpr2_vgpr3
+    %0 = COPY $vgpr0_vgpr1
+    %3 = FLAT_LOAD_DWORD %1, 0, 0, 0, 0, implicit $exec, implicit $flat_scr :: (load 4)
+
+    %5 = S_MOV_B32 65535
+    %6 = S_MOV_B32 65535
+
+    %11 = V_LSHRREV_B32_e64 16, %3, implicit $exec
+    %12 = V_AND_B32_e32 %6, %11, implicit $exec
+    %13 = V_LSHLREV_B32_e64 16, %12, implicit $exec
+    %14 = V_LSHRREV_B32_e64 16, %13, implicit $exec
+    %15 = V_BFE_U32 %13, 8, 8, implicit $exec
+    %16 = V_ADD_F32_e32 %14, %15, implicit $exec
+    %17 = V_LSHLREV_B32_e64 16, %16, implicit $exec
+    %18 = V_LSHRREV_B32_e64 16, %17, implicit $exec
+    %19 = V_BFE_U32 %17, 8, 8, implicit $exec
+    %20 = V_SUB_F16_e32 %18, %19, implicit $exec
+    %21 = V_LSHLREV_B32_e64 16, %20, implicit $exec
+    %22 = V_BFE_U32 %20, 8, 8, implicit $exec
+    %23 = V_FMAC_F32_e32 %21, %22, %22, implicit $exec
+    %24 = V_LSHLREV_B32_e64 16, %23, implicit $exec
+    %25 = V_LSHRREV_B32_e64 16, %24, implicit $exec
+    %26 = V_BFE_U32 %24, 8, 8, implicit $exec
+    %27 = V_FMAC_F16_e32 %25, %26, %26, implicit $exec
+    %28 = V_LSHLREV_B32_e64 16, %27, implicit $exec
+
+    %29 = V_LSHRREV_B32_e64 16, %28, implicit $exec
+    %30 = V_AND_B32_e64 23, %29, implicit $exec
+    %31 = V_LSHLREV_B32_e64 16, %30, implicit $exec
+    %32 = V_LSHRREV_B32_e64 16, %31, implicit $exec
+    %33 = V_BFE_U32 %31, 8, 8, implicit $exec
+    %34 = V_ADD_F32_e64 0, %32, 0, %33, 0, 0, implicit $exec
+    %35 = V_LSHLREV_B32_e64 16, %34, implicit $exec
+    %37 = V_BFE_U32 %35, 8, 8, implicit $exec
+    %38 = V_SUB_F16_e64 0, 23, 0, %37, 0, 0, implicit $exec
+    %39 = V_LSHLREV_B32_e64 16, %38, implicit $exec
+    %40 = V_BFE_U32 %39, 8, 8, implicit $exec
+    %41 = V_FMAC_F32_e64 0, 23, 0, %40, 0, %40, 0, 0, implicit $exec
+    %42 = V_LSHLREV_B32_e64 16, %41, implicit $exec
+    %43 = V_LSHRREV_B32_e64 16, %42, implicit $exec
+    %44 = V_BFE_U32 %42, 8, 8, implicit $exec
+    %45 = V_FMAC_F16_e64 0, %43, 0, %44, 0, %44, 0, 0, implicit $exec
+    %46 = V_LSHLREV_B32_e64 16, %45, implicit $exec
+
+    %47 = V_LSHRREV_B32_e64 16, %46, implicit $exec
+    %48 = V_BFE_U32 %46, 8, 8, implicit $exec
+    %49 = V_ADD_F32_e64 0, %47, 1, %48, 0, 0, implicit $exec
+    %50 = V_LSHLREV_B32_e64 16, %49, implicit $exec
+    %51 = V_BFE_U32 %50, 8, 8, implicit $exec
+    %52 = V_SUB_F16_e64 1, 23, 1, %51, 0, 0, implicit $exec
+    %53 = V_LSHLREV_B32_e64 16, %52, implicit $exec
+    %54 = V_BFE_U32 %53, 8, 8, implicit $exec
+    %55 = V_FMAC_F32_e64 1, 23, 1, %54, 1, %54, 1, 0, implicit $exec
+    %56 = V_LSHLREV_B32_e64 16, %55, implicit $exec
+    %57 = V_LSHRREV_B32_e64 16, %56, implicit $exec
+    %58 = V_BFE_U32 %56, 8, 8, implicit $exec
+    %59 = V_FMAC_F16_e64 1, %57, 1, %58, 1, %58, 0, 2, implicit $exec
+    %60 = V_LSHLREV_B32_e64 16, %59, implicit $exec
+
+    %100 = V_MOV_B32_e32 %60, implicit $exec
+
+    FLAT_STORE_DWORD %0, %100, 0, 0, 0, 0, implicit $exec, implicit $flat_scr :: (store 4)
+    $sgpr30_sgpr31 = COPY %2
+    S_SETPC_B64_return $sgpr30_sgpr31
+
+...

llvm/trunk/test/CodeGen/AMDGPU/twoaddr-fma.mir

+# RUN: llc -march=amdgcn -mcpu=gfx1010 %s -run-pass twoaddressinstruction -verify-machineinstrs -o - | FileCheck -check-prefix=GCN %s
+
+# GCN-LABEL: name: test_fmamk_reg_imm_f32
+# GCN: V_FMAMK_F32 killed %0.sub0, 1078523331, killed %1, implicit $exec
+---
+name:            test_fmamk_reg_imm_f32
+registers:
+  - { id: 0, class: vreg_64 }
+  - { id: 1, class: vgpr_32 }
+  - { id: 2, class: vgpr_32 }
+  - { id: 3, class: vgpr_32 }
+body:             |
+  bb.0:
+
+    %0 = IMPLICIT_DEF
+    %1 = COPY %0.sub1
+    %2 = V_MOV_B32_e32 1078523331, implicit $exec
+    %3 = V_FMAC_F32_e32 killed %0.sub0, %2, killed %1, implicit $exec
+
+...
+
+# GCN-LABEL: name: test_fmamk_imm_reg_f32
+# GCN: V_FMAMK_F32 killed %0.sub0, 1078523331, killed %1, implicit $exec
+---
+name:            test_fmamk_imm_reg_f32
+registers:
+  - { id: 0, class: vreg_64 }
+  - { id: 1, class: vgpr_32 }
+  - { id: 2, class: vgpr_32 }
+  - { id: 3, class: vgpr_32 }
+body:             |
+  bb.0:
+
+    %0 = IMPLICIT_DEF
+    %1 = COPY %0.sub1
+    %2 = V_MOV_B32_e32 1078523331, implicit $exec
+    %3 = V_FMAC_F32_e32 %2, killed %0.sub0, killed %1, implicit $exec
+
+...
+
+# GCN-LABEL: name: test_fmaak_f32
+# GCN: V_FMAAK_F32 killed %0.sub0, %0.sub1, 1078523331, implicit $exec
+---
+name:            test_fmaak_f32
+registers:
+  - { id: 0, class: vreg_64 }
+  - { id: 1, class: vgpr_32 }
+  - { id: 2, class: vgpr_32 }
+body:             |
+  bb.0:
+
+    %0 = IMPLICIT_DEF
+    %1 = V_MOV_B32_e32 1078523331, implicit $exec
+    %2 = V_FMAC_F32_e32 killed %0.sub0, %0.sub1, %1, implicit $exec
+
+...
+
+# GCN-LABEL: name: test_fmamk_reg_imm_f16
+# GCN: V_FMAMK_F16 killed %0.sub0, 1078523331, killed %1, implicit $exec
+---
+name:            test_fmamk_reg_imm_f16
+registers:
+  - { id: 0, class: vreg_64 }
+  - { id: 1, class: vgpr_32 }
+  - { id: 2, class: vgpr_32 }
+  - { id: 3, class: vgpr_32 }
+body:             |
+  bb.0:
+
+    %0 = IMPLICIT_DEF
+    %1 = COPY %0.sub1
+    %2 = V_MOV_B32_e32 1078523331, implicit $exec
+    %3 = V_FMAC_F16_e32 killed %0.sub0, %2, killed %1, implicit $exec
+
+...
+
+# GCN-LABEL: name: test_fmamk_imm_reg_f16
+# GCN: V_FMAMK_F16 killed %0.sub0, 1078523331, killed %1, implicit $exec
+---
+name:            test_fmamk_imm_reg_f16
+registers:
+  - { id: 0, class: vreg_64 }
+  - { id: 1, class: vgpr_32 }
+  - { id: 2, class: vgpr_32 }
+  - { id: 3, class: vgpr_32 }
+body:             |
+  bb.0:
+
+    %0 = IMPLICIT_DEF
+    %1 = COPY %0.sub1
+    %2 = V_MOV_B32_e32 1078523331, implicit $exec
+    %3 = V_FMAC_F16_e32 %2, killed %0.sub0, killed %1, implicit $exec
+
+...
+
+# GCN-LABEL: name: test_fmaak_f16
+# GCN: V_FMAAK_F16 killed %0.sub0, %0.sub1, 1078523331, implicit $exec
+---
+name:            test_fmaak_f16
+registers:
+  - { id: 0, class: vreg_64 }
+  - { id: 1, class: vgpr_32 }
+  - { id: 2, class: vgpr_32 }
+body:             |
+  bb.0:
+
+    %0 = IMPLICIT_DEF
+    %1 = V_MOV_B32_e32 1078523331, implicit $exec
+    %2 = V_FMAC_F16_e32 killed %0.sub0, %0.sub1, %1, implicit $exec
+...
+
+# GCN-LABEL: name: test_fmaak_sgpr_src0_f32
+# GCN: %2:vgpr_32 = V_FMAMK_F32 killed %0, 1078523331, %3:vgpr_32, implicit $exec
+
+---
+name:            test_fmaak_sgpr_src0_f32
+registers:
+  - { id: 0, class: sreg_32_xm0 }
+  - { id: 1, class: vgpr_32}
+  - { id: 2, class: vgpr_32 }
+  - { id: 3, class: vgpr_32 }
+body:             |
+  bb.0:
+
+    %0 = IMPLICIT_DEF
+    %1 = V_MOV_B32_e32 1078523331, implicit $exec
+    %2 = V_FMAC_F32_e32 killed %0, %1, %3, implicit $exec
+
+...
+
+# GCN-LABEL: name: test_fmaak_inlineimm_src0_f32
+# GCN: %1:vgpr_32 = V_FMAMK_F32 1073741824, 1078523331, %2:vgpr_32, implicit $exec
+
+---
+name:            test_fmaak_inlineimm_src0_f32
+registers:
+  - { id: 0, class: vgpr_32}
+  - { id: 1, class: vgpr_32 }
+  - { id: 2, class: vgpr_32 }
+body:             |
+  bb.0:
+
+    %0 = V_MOV_B32_e32 1078523331, implicit $exec
+    %1 = V_FMAC_F32_e32 1073741824, %0, %2, implicit $exec
+
+...
+
+# GCN-LABEL: name: test_fmaak_otherimm_src0_f32
+# GCN: %1:vgpr_32 = V_FMAC_F32_e32 1120403456, %0, %1, implicit $exec
+
+---
+name:            test_fmaak_otherimm_src0_f32
+registers:
+  - { id: 0, class: vgpr_32}
+  - { id: 1, class: vgpr_32 }
+  - { id: 2, class: vgpr_32 }
+body:             |
+  bb.0:
+
+    %0 = V_MOV_B32_e32 1078523331, implicit $exec
+    %1 = V_FMAC_F32_e32 1120403456, %0, %2, implicit $exec
+
+...
+
+# GCN-LABEL: name: test_fmaak_other_constantlike_src0_f32
+# GCN: %1:vgpr_32 = V_FMAC_F32_e32 %stack.0, %0, %1, implicit $exec
+---
+name:            test_fmaak_other_constantlike_src0_f32
+registers:
+  - { id: 0, class: vgpr_32}
+  - { id: 1, class: vgpr_32 }
+  - { id: 2, class: vgpr_32 }
+stack:
+  - { id: 0, name: "", type: default, offset: 0, size: 128, alignment: 8,
+      callee-saved-register: '', local-offset: 0, debug-info-variable: '',
+      debug-info-expression: '', debug-info-location: '' }
+body:             |
+  bb.0:
+
+    %0 = V_MOV_B32_e32 1078523331, implicit $exec
+    %1 = V_FMAC_F32_e32 %stack.0, %0, %2, implicit $exec
+
+...