[AMDGPU][GlobalISel] Transform (fadd (fmul x, y), z) -> (fma x, y, z)

[matejam]

Summary

Instead of add (sub) and mul instructions, use v_mad, v_mac or v_fma
if fma instructions are faster and are legal for the given architecture.
Combiner for a simple case that has only one add and one mul instruction
and transforms them into some of the fma instructions depending on the
architecture.

[arsenm]

llvm/lib/CodeGen/GlobalISel/CombinerHelper.cpp

3145–3146

Using bf16 here instead of regular half is wrong. We also have a utility function to get the FP type from the LLT

3159

Checking the denormal mode here isn't right. The denormal mode only impacts whether this should be used for AMDGPU and isn't a generic property

3185–3187

return compare directly

3196

The matching logic here implies this should be able to use fma also

[matejam]

Instead of add and mul instructions, use v_mad, v_mac or v_fma
if fma instructions are faster and are legal for the given architecture.
Combiner for a simple case that has only one add and one mul instruction
and transforms them into some of the fma instructions depending on the
architecture.

[arsenm]

llvm/include/llvm/CodeGen/TargetLowering.h

796

This will break for vectors

1116

This will break for vectors. Also, this should not be built on top of the DAG legalizer rules. This should be something in LegalizerInfo

2744

Broken for vectors

[foad]

llvm/include/llvm/CodeGen/GlobalISel/CombinerHelper.h

112

isLegal needs its own comment.

333

Comment should mention fmad.

334

Capitalise them like "FAdd" and "FMul" and "FMA" and "FMad". There are akready examples of this in GlobalISel.

550

"FMul"

llvm/include/llvm/CodeGen/TargetLowering.h

798

Do you need EVT() here? I thought MVT would implicitly convert to EVT.

[foad]

llvm/include/llvm/CodeGen/TargetLowering.h

795

Function needs a comment.

1114

Function needs a comment.

2693

Function needs a comment.

2734

Function needs a comment.

2739

Function needs a comment.

llvm/include/llvm/Target/GlobalISel/Combine.td

560

"combine_fadd_fmul_to_fmad_or_fma"?

llvm/lib/CodeGen/GlobalISel/CombinerHelper.cpp

3575

Function needs a comment.

[matejam]

Added support for vector types and some refactoring.

[foad]

llvm/include/llvm/Target/GlobalISel/Combine.td

608

You've lost load_or_combine.

[matejam]

Put back the accidentally deleted combiner from the list of combiners (load_or_combine).

[matejam]

Added two mir tests (prelegalize and postlegalize combiner) and a few adjustments in CombinerHelper.cpp.

[matejam]

Code formatting in CombinerHelper.cpp.

[matejam]

Adjustments in the tests.

[foad]

llvm/include/llvm/CodeGen/GlobalISel/CombinerHelper.h

110–111

Comment seems wrong. Should it be "... running after legalization ..."?

llvm/include/llvm/CodeGen/TargetLowering.h

2717

"const LLT" seems weird for a parameter type. Is there a reason for the const?

2729

"const LLT" seems weird.

But why do we need this isFMADLegal function, which calls into selectiondag legalization code? Can't you use CombinerHelper isLegal methods instead?

llvm/lib/CodeGen/GlobalISel/CombinerHelper.cpp

3906

This could be static, just like isContractable?

3910

return AllowFusionGlobally || isContractable(MI);?

3927

MI.getMF()

3939

This expression is exactly what "isLegalOrBeforeLegalization" does, so you didn't need to create isLegal.

3951

!isContractable already implies !MI.FmReassoc so you don't need to check it again here.

llvm/lib/Target/AMDGPU/AMDGPUInstructionSelector.cpp

3330 ↗

(On Diff #337401)

Can you submit this change as a separate patch? (Does it affect any existing lit tests?)

[matejam]

llvm/include/llvm/CodeGen/GlobalISel/CombinerHelper.h

110–111

That seems ok, I'll change that.

llvm/include/llvm/CodeGen/TargetLowering.h

2717

I don't think there is a good reason for that, I'll change that.

2729

I'll change the const LLT.
I tried to find a way to call methods like "hasMadF16", "hasFP64FP16Denormals", etc. from CombinerHelper, but using adjusted SelectionDAG legalization code seemed much easier and prettier to me.

llvm/lib/CodeGen/GlobalISel/CombinerHelper.cpp

3906

It probably should, I'll change that.

3910

Thanks!

3927

Thanks!

3939

"isLegalOrBeforeLegalization" returns the value of "!LI || LI->getAction(Query).Action == LegalizeActions::Legal"
and isLegal returns the value of "LI && LI->getAction(Query).Action == LegalizeActions::Legal", so I thinks it's not exactly the same.

3951

Thanks!

llvm/lib/Target/AMDGPU/AMDGPUInstructionSelector.cpp

3330 ↗

(On Diff #337401)

Yes, it should be in a separate patch, it was added here by mistake.

[foad]

llvm/lib/CodeGen/GlobalISel/CombinerHelper.cpp

3939

I mean !LegalOperations || isLegal(...) is the same as isLegalOrBeforeLegalization(...). Because LegalOperations is just another name for LI.

[foad]

llvm/include/llvm/CodeGen/TargetLowering.h

2729

I was hoping that hasFP64FP16Denormals etc would be tested in AMDGPULegalizerInfo::AMDGPULegalizerInfo, where it decides whether G_FMAD is legal, so the normal isLegal* functions would work. But now I see that hasFP64FP16Denormals depends on function attributes, so it can't be done that way.

[matejam]

llvm/include/llvm/CodeGen/TargetLowering.h

2729

So do you think I should keep it this way?

llvm/lib/CodeGen/GlobalISel/CombinerHelper.cpp

3939

You're right, I'll change that. Thanks!

[foad]

llvm/include/llvm/CodeGen/TargetLowering.h

2729

I'm not sure. Alternatively you could just check "isLegalOrBeforeLegalizer" on the FMAD instructions. Then pre-legalization we would always combine to FMAD, and then the legalizer might decide that the FMAD is not legal and turn it into either MUL+ADD or FMA. Would that work?

[foad]

I think this will look good once you've updated the patch for the outstanding minor comments.

llvm/include/llvm/CodeGen/TargetLowering.h

2729

We discussed this offline. I still think this would be an interesting experiment to try, but I also think the approach in the current patch is OK for now.

[matejam]

A few minor adjustments in CombinerHelper.cpp.

[matejam]

Minor changes in CombinerHelper.cpp and in the tests.

[matejam]

Use mi_match for comparing instructions instead of comparing the opcodes.

[foad]

LGTM but I'm adding a couple more globalisel reviewers. Please wait a couple more days in case they have comments.

llvm/include/llvm/CodeGen/TargetLowering.h

797

"benefits"

2735

"if \p MI can be combined with another instruction to form"?

[matejam]

A few typos.

[arsenm]

llvm/include/llvm/CodeGen/TargetLowering.h

2729

Right, the FP mode is not a property of the target. It's a property of the FP mode of the function, which is contextually valid or not. The legalizer rules are not allowed to differ based on context like that

[matejam]

Rebased.

[matejam]

Move isFPExtFoldable() to the patch where it is used for the first time.
Updated to reflect changes in D104247.
Update tests.

[mbrkusanin]

llvm/lib/CodeGen/GlobalISel/CombinerHelper.cpp

4627–4639

We could instead use applyBuildFn which uses a lambda as MatchInfo that details what instructions to build. I noticed this is used for some new combines. It might look more readable then this long tuple. Especially for other fma combines.

If we do this for other fma combines as well then we could merge multiple matches into one (like the ones that start with fadd D93305, D97937, D97938, D98047) and avoid running a lot of common checks again for each combine (like the ones in canCombineFMadOrFMA).

@foad, @arsenm What do you think of this idea. Would it be more preferable?
Main reason why all these matching opportunities are not part of the same match function is because we would need the large number of bools in tuple/struct to figure out what exactly to build in the apply function. A problem that can be avoided with this BuildFn lambda mechanism.

[matejam]

Use applyBuildFn instead of writing my own apply.

[mbrkusanin]

llvm/include/llvm/CodeGen/GlobalISel/CombinerHelper.h

386

Rename type to BuildFnTy since there is a:
using BuildFnTy = std::function<void(MachineIRBuilder &)>;

(same for other patches as well, might help with formatting)

[matejam]

Changes in tests.

[mbrkusanin]

LGTM

llvm/include/llvm/Target/GlobalISel/Combine.td

822–824

Might be useful to make a new combine group for this and other fma combines.

[matejam]

Added missing comments and an FMA combiner group in table gen.

llvm/include/llvm/CodeGen/GlobalISel/CombinerHelper.h

 };
 
 class CombinerHelper {
 protected:
   MachineIRBuilder &Builder;
   MachineRegisterInfo &MRI;
   GISelChangeObserver &Observer;
   GISelKnownBits *KB;
   MachineDominatorTree *MDT;
   const LegalizerInfo *LI;
   const RegisterBankInfo *RBI;
   const TargetRegisterInfo *TRI;
 
 public:
   CombinerHelper(GISelChangeObserver &Observer, MachineIRBuilder &B,
                  GISelKnownBits *KB = nullptr,
                  MachineDominatorTree *MDT = nullptr,
                  const LegalizerInfo *LI = nullptr);
 
public:
 
   /// Transform X, Y<dead> = G_UNMERGE Z -> X = G_TRUNC Z.
   bool matchCombineUnmergeWithDeadLanesToTrunc(MachineInstr &MI);
   void applyCombineUnmergeWithDeadLanesToTrunc(MachineInstr &MI);
 
   /// Transform X, Y = G_UNMERGE(G_ZEXT(Z)) -> X = G_ZEXT(Z); Y = G_CONSTANT 0
   bool matchCombineUnmergeZExtToZExt(MachineInstr &MI);
   void applyCombineUnmergeZExtToZExt(MachineInstr &MI);
 
   /// Transform fp_instr(cst) to constant result of the fp operation.
   bool matchCombineConstantFoldFpUnary(MachineInstr &MI,
                                        Optional<APFloat> &Cst);
   void applyCombineConstantFoldFpUnary(MachineInstr &MI,
                                        Optional<APFloat> &Cst);
 
   /// Transform IntToPtr(PtrToInt(x)) to x if cast is in the same address space.
   bool matchCombineI2PToP2I(MachineInstr &MI, Register &Reg);
   void applyCombineI2PToP2I(MachineInstr &MI, Register &Reg);
public:
 
   /// Transform fabs(fneg(x)) to fabs(x).
   bool matchCombineFAbsOfFNeg(MachineInstr &MI, BuildFnTy &MatchInfo);
 
   /// Transform trunc ([asz]ext x) to x or ([asz]ext x) or (trunc x).
   bool matchCombineTruncOfExt(MachineInstr &MI,
                               std::pair<Register, unsigned> &MatchInfo);
   void applyCombineTruncOfExt(MachineInstr &MI,
                               std::pair<Register, unsigned> &MatchInfo);
 
   /// Transform trunc (shl x, K) to shl (trunc x),
   /// K => K < VT.getScalarSizeInBits().
   bool matchCombineTruncOfShl(MachineInstr &MI,
                               std::pair<Register, Register> &MatchInfo);
   void applyCombineTruncOfShl(MachineInstr &MI,
                               std::pair<Register, Register> &MatchInfo);
 
public:
   bool matchLoadOrCombine(MachineInstr &MI, BuildFnTy &MatchInfo);
 
   bool matchTruncStoreMerge(MachineInstr &MI, MergeTruncStoresInfo &MatchInfo);
   void applyTruncStoreMerge(MachineInstr &MI, MergeTruncStoresInfo &MatchInfo);
 
   bool matchExtendThroughPhis(MachineInstr &MI, MachineInstr *&ExtMI);
   void applyExtendThroughPhis(MachineInstr &MI, MachineInstr *&ExtMI);
 
   bool matchExtractVecEltBuildVec(MachineInstr &MI, Register &Reg);
   void applyExtractVecEltBuildVec(MachineInstr &MI, Register &Reg);
 
   bool matchExtractAllEltsFromBuildVector(
       MachineInstr &MI,
       SmallVectorImpl<std::pair<Register, MachineInstr *>> &MatchInfo);
   void applyExtractAllEltsFromBuildVector(
       MachineInstr &MI,
       SmallVectorImpl<std::pair<Register, MachineInstr *>> &MatchInfo);
public:
   ///           (fadd fneg(x), y) -> (fsub y, x)
   ///           (fsub x, fneg(y)) -> (fadd x, y)
   ///           (fmul fneg(x), fneg(y)) -> (fmul x, y)
   ///           (fdiv fneg(x), fneg(y)) -> (fdiv x, y)
   ///           (fmad fneg(x), fneg(y), z) -> (fmad x, y, z)
   ///           (fma fneg(x), fneg(y), z) -> (fma x, y, z)
   bool matchRedundantNegOperands(MachineInstr &MI, BuildFnTy &MatchInfo);
 
+  bool canCombineFMadOrFMA(MachineInstr &MI, bool &AllowFusionGlobally,
+                           bool &HasFMAD, bool &Aggressive);
+
+  /// Transform (fadd (fmul x, y), z) -> (fma x, y, z)
+  ///           (fadd (fmul x, y), z) -> (fmad x, y, z)
+  bool matchCombineFAddFMulToFMadOrFMA(MachineInstr &MI, BuildFnTy &MatchInfo);
+
 private:
   /// Given a non-indexed load or store instruction \p MI, find an offset that
   /// can be usefully and legally folded into it as a post-indexing operation.
   ///
   /// \returns true if a candidate is found.
   bool findPostIndexCandidate(MachineInstr &MI, Register &Addr, Register &Base,
                               Register &Offset);
 

llvm/include/llvm/CodeGen/TargetLowering.h

public:
   /// Some targets may prefer one to the other.
   virtual bool preferIncOfAddToSubOfNot(EVT VT) const {
     // By default, let's assume that everyone prefers the form with two add's.
     return true;
   }
 
   /// Return true if the target wants to use the optimization that
   /// turns ext(promotableInst1(...(promotableInstN(load)))) into
   /// promotedInst1(...(promotedInstN(ext(load)))).
   bool enableExtLdPromotion() const { return EnableExtLdPromotion; }
 
   /// Return true if the target can combine store(extractelement VectorTy,
   /// Idx).
   /// \p Cost[out] gives the cost of that transformation when this is true.
   virtual bool canCombineStoreAndExtract(Type *VectorTy, Value *Idx,
                                          unsigned &Cost) const {
     return false;
   }
 
   /// Return true if inserting a scalar into a variable element of an undef
   /// vector is more efficiently handled by splatting the scalar instead.
   virtual bool shouldSplatInsEltVarIndex(EVT) const {
     return false;
   }
 
   /// Return true if target always benefits from combining into FMA for a
   /// given value type. This must typically return false on targets where FMA
   /// takes more cycles to execute than FADD.
-  virtual bool enableAggressiveFMAFusion(EVT VT) const {
-    return false;
-  }
+  virtual bool enableAggressiveFMAFusion(EVT VT) const { return false; }
+
+  /// Return true if target always benefits from combining into FMA for a
+  /// given value type. This must typically return false on targets where FMA
+  /// takes more cycles to execute than FADD.
+  virtual bool enableAggressiveFMAFusion(LLT Ty) const { return false; }
 
   /// Return the ValueType of the result of SETCC operations.
   virtual EVT getSetCCResultType(const DataLayout &DL, LLVMContext &Context,
                                  EVT VT) const;
 
   /// Return the ValueType for comparison libcalls. Comparions libcalls include
   /// floating point comparion calls, and Ordered/Unordered check calls on
   /// floating point numbers.
LegalizeAction getFixedPointOperationAction(unsigned Op, EVT VT,
     }
 
     return Supported ? Action : Expand;
   }
 
   // If Op is a strict floating-point operation, return the result
   // of getOperationAction for the equivalent non-strict operation.
   LegalizeAction getStrictFPOperationAction(unsigned Op, EVT VT) const {
     unsigned EqOpc;
     switch (Op) {
       default: llvm_unreachable("Unexpected FP pseudo-opcode");
 #define DAG_INSTRUCTION(NAME, NARG, ROUND_MODE, INTRINSIC, DAGN)               \
       case ISD::STRICT_##DAGN: EqOpc = ISD::DAGN; break;
 #define CMP_INSTRUCTION(NAME, NARG, ROUND_MODE, INTRINSIC, DAGN)               \
       case ISD::STRICT_##DAGN: EqOpc = ISD::SETCC; break;
 #include "llvm/IR/ConstrainedOps.def"
     }
 
     return getOperationAction(EqOpc, VT);
public:
   virtual bool lowerInterleavedLoad(LoadInst *LI,
                                     ArrayRef<ShuffleVectorInst *> Shuffles,
                                     ArrayRef<unsigned> Indices,
                                     unsigned Factor) const {
     return false;
   }
 
   /// Lower an interleaved store to target specific intrinsics. Return
   /// true on success.
   ///
   /// \p SI is the vector store instruction.
   /// \p SVI is the shufflevector to RE-interleave the stored vector.
   /// \p Factor is the interleave factor.
   virtual bool lowerInterleavedStore(StoreInst *SI, ShuffleVectorInst *SVI,
                                      unsigned Factor) const {
     return false;
   }
 
   /// Return true if zero-extending the specific node Val to type VT2 is free
   /// (either because it's implicitly zero-extended such as ARM ldrb / ldrh or
   /// because it's folded such as X86 zero-extending loads).
   virtual bool isZExtFree(SDValue Val, EVT VT2) const {
     return isZExtFree(Val.getValueType(), VT2);
   }
 
   /// Return true if an fpext operation is free (for instance, because
   /// single-precision floating-point numbers are implicitly extended to
   /// double-precision).
   virtual bool isFPExtFree(EVT DestVT, EVT SrcVT) const {
     assert(SrcVT.isFloatingPoint() && DestVT.isFloatingPoint() &&
            "invalid fpext types");
     return false;
   }
 
   /// Return true if an fpext operation input to an \p Opcode operation is free
   /// (for instance, because half-precision floating-point numbers are
   /// implicitly extended to float-precision) for an FMA instruction.
   virtual bool isFPExtFoldable(const SelectionDAG &DAG, unsigned Opcode,
                                EVT DestVT, EVT SrcVT) const {
     assert(DestVT.isFloatingPoint() && SrcVT.isFloatingPoint() &&
            "invalid fpext types");
     return isFPExtFree(DestVT, SrcVT);
   }
 
   /// Return true if folding a vector load into ExtVal (a sign, zero, or any
   /// extend node) is profitable.
   virtual bool isVectorLoadExtDesirable(SDValue ExtVal) const { return false; }
 
   /// Return true if an fneg operation is free to the point where it is never
   /// worthwhile to replace it with a bitwise operation.
   virtual bool isFNegFree(EVT VT) const {
     assert(VT.isFloatingPoint());
     return false;
   }
 
   /// Return true if an fabs operation is free to the point where it is never
   /// worthwhile to replace it with a bitwise operation.
   virtual bool isFAbsFree(EVT VT) const {
     assert(VT.isFloatingPoint());
     return false;
   }
 
   /// Return true if an FMA operation is faster than a pair of fmul and fadd
   /// instructions. fmuladd intrinsics will be expanded to FMAs when this method
   /// returns true, otherwise fmuladd is expanded to fmul + fadd.
   ///
   /// NOTE: This may be called before legalization on types for which FMAs are
   /// not legal, but should return true if those types will eventually legalize
   /// to types that support FMAs. After legalization, it will only be called on
   /// types that support FMAs (via Legal or Custom actions)
   virtual bool isFMAFasterThanFMulAndFAdd(const MachineFunction &MF,
                                           EVT) const {
     return false;
   }
 
+  /// Return true if an FMA operation is faster than a pair of fmul and fadd
+  /// instructions. fmuladd intrinsics will be expanded to FMAs when this method
+  /// returns true, otherwise fmuladd is expanded to fmul + fadd.
+  ///
+  /// NOTE: This may be called before legalization on types for which FMAs are
+  /// not legal, but should return true if those types will eventually legalize
+  /// to types that support FMAs. After legalization, it will only be called on
+  /// types that support FMAs (via Legal or Custom actions)
+  virtual bool isFMAFasterThanFMulAndFAdd(const MachineFunction &MF,
+                                          LLT) const {
+    return false;
+  }
+
   /// IR version
   virtual bool isFMAFasterThanFMulAndFAdd(const Function &F, Type *) const {
     return false;
   }
 
+  /// Returns true if \p MI can be combined with another instruction to
+  /// form TargetOpcode::G_FMAD. \p N may be an TargetOpcode::G_FADD,
+  /// TargetOpcode::G_FSUB, or an TargetOpcode::G_FMUL which will be
+  /// distributed into an fadd/fsub.
+  virtual bool isFMADLegal(const MachineInstr &MI, LLT Ty) const {
+    assert((MI.getOpcode() == TargetOpcode::G_FADD ||
+            MI.getOpcode() == TargetOpcode::G_FSUB ||
+            MI.getOpcode() == TargetOpcode::G_FMUL) &&
+           "unexpected node in FMAD forming combine");
+    switch (Ty.getScalarSizeInBits()) {
+    case 16:
+      return isOperationLegal(TargetOpcode::G_FMAD, MVT::f16);
+    case 32:
+      return isOperationLegal(TargetOpcode::G_FMAD, MVT::f32);
+    case 64:
+      return isOperationLegal(TargetOpcode::G_FMAD, MVT::f64);
+    default:
+      break;
+    }
+
+    return false;
+  }
+
   /// Returns true if be combined with to form an ISD::FMAD. \p N may be an
   /// ISD::FADD, ISD::FSUB, or an ISD::FMUL which will be distributed into an
   /// fadd/fsub.
   virtual bool isFMADLegal(const SelectionDAG &DAG, const SDNode *N) const {
     assert((N->getOpcode() == ISD::FADD || N->getOpcode() == ISD::FSUB ||
             N->getOpcode() == ISD::FMUL) &&
            "unexpected node in FMAD forming combine");
     return isOperationLegal(ISD::FMAD, N->getValueType(0));

llvm/include/llvm/Target/GlobalISel/Combine.td

 >;
 
 // Fold trunc ([asz]ext x) -> x or ([asz]ext x) or (trunc x).
 def trunc_ext_fold_matchinfo : GIDefMatchData<"std::pair<Register, unsigned>">;
 def trunc_ext_fold: GICombineRule <
   (defs root:$root, trunc_ext_fold_matchinfo:$matchinfo),
   (match (wip_match_opcode G_TRUNC):$root,
          [{ return Helper.matchCombineTruncOfExt(*${root}, ${matchinfo}); }]),
   (apply [{ Helper.applyCombineTruncOfExt(*${root}, ${matchinfo}); }])
 >;
 
 // Fold trunc (shl x, K) -> shl (trunc x), K => K < VT.getScalarSizeInBits().
 def trunc_shl_matchinfo : GIDefMatchData<"std::pair<Register, Register>">;
 def trunc_shl: GICombineRule <
   (defs root:$root, trunc_shl_matchinfo:$matchinfo),
   (match (wip_match_opcode G_TRUNC):$root,
          [{ return Helper.matchCombineTruncOfShl(*${root}, ${matchinfo}); }]),
def combine_insert_vec_elts_build_vector : GICombineRule<
   (match (wip_match_opcode G_INSERT_VECTOR_ELT):$root,
     [{ return Helper.matchCombineInsertVecElts(*${root}, ${info}); }]),
   (apply [{ Helper.applyCombineInsertVecElts(*${root}, ${info}); }])>;
 
 def load_or_combine : GICombineRule<
   (defs root:$root, build_fn_matchinfo:$info),
   (match (wip_match_opcode G_OR):$root,
     [{ return Helper.matchLoadOrCombine(*${root}, ${info}); }]),
   (apply [{ Helper.applyBuildFn(*${root}, ${info}); }])>;
 
 
 def truncstore_merge_matcdata : GIDefMatchData<"MergeTruncStoresInfo">;
 def truncstore_merge : GICombineRule<
   (defs root:$root, truncstore_merge_matcdata:$info),
   (match (wip_match_opcode G_STORE):$root,
    [{ return Helper.matchTruncStoreMerge(*${root}, ${info}); }]),
   (apply [{ Helper.applyTruncStoreMerge(*${root}, ${info}); }])>;

 def mulh_combines : GICombineGroup<[mulh_to_lshr]>;
 
 def redundant_neg_operands: GICombineRule<
   (defs root:$root, build_fn_matchinfo:$matchinfo),
   (match (wip_match_opcode G_FADD, G_FSUB, G_FMUL, G_FDIV, G_FMAD, G_FMA):$root,
     [{ return Helper.matchRedundantNegOperands(*${root}, ${matchinfo}); }]),
   (apply [{ Helper.applyBuildFnNoErase(*${root}, ${matchinfo}); }])>;
 
+// Transform (fadd x, (fmul y, z)) -> (fma y, z, x)
+//           (fadd x, (fmul y, z)) -> (fmad y, z, x)
+// Transform (fadd (fmul x, y), z) -> (fma x, y, z)
+//           (fadd (fmul x, y), z) -> (fmad x, y, z)
+def combine_fadd_fmul_to_fmad_or_fma: GICombineRule<
+  (defs root:$root, build_fn_matchinfo:$info),
+  (match (wip_match_opcode G_FADD):$root,
+         [{ return Helper.matchCombineFAddFMulToFMadOrFMA(*${root},
+                                                          ${info}); }]),
+  (apply [{ Helper.applyBuildFn(*${root}, ${info}); }])>;
+
+
 // FIXME: These should use the custom predicate feature once it lands.
 def undef_combines : GICombineGroup<[undef_to_fp_zero, undef_to_int_zero,
                                      undef_to_negative_one,
                                      binop_left_undef_to_zero,
                                      propagate_undef_any_op,
                                      propagate_undef_all_ops,
                                      propagate_undef_shuffle_mask,
                                      erase_undef_store]>;

 
 def phi_combines : GICombineGroup<[extend_through_phis]>;
 
 def select_combines : GICombineGroup<[select_undef_cmp, select_constant_cmp]>;
 
 def trivial_combines : GICombineGroup<[copy_prop, mul_to_shl, add_p2i_to_ptradd,
                                        mul_by_neg_one]>;
 
+def fma_combines : GICombineGroup<[combine_fadd_fmul_to_fmad_or_fma]>;
+
 def all_combines : GICombineGroup<[trivial_combines, insert_vec_elt_combines,
     extract_vec_elt_combines, combines_for_extload,
     combine_indexed_load_store, undef_combines, identity_combines, phi_combines,
     simplify_add_to_sub, hoist_logic_op_with_same_opcode_hands,
     reassocs, ptr_add_immed_chain,
     shl_ashr_to_sext_inreg, sext_inreg_of_load,
     width_reduction_combines, select_combines,
     known_bits_simplifications, ext_ext_fold,
     not_cmp_fold, opt_brcond_by_inverting_cond,
     unmerge_merge, fabs_fabs_fold, unmerge_cst, unmerge_dead_to_trunc,
     unmerge_zext_to_zext, merge_unmerge, trunc_ext_fold, trunc_shl,
     const_combines, xor_of_and_with_same_reg, ptr_add_with_zero,
     shift_immed_chain, shift_of_shifted_logic_chain, load_or_combine,
     truncstore_merge, div_rem_to_divrem, funnel_shift_combines,
     form_bitfield_extract, constant_fold, fabs_fneg_fold,
     intdiv_combines, mulh_combines, redundant_neg_operands,
-    and_or_disjoint_mask ]>;
+    and_or_disjoint_mask, fma_combines]>;
 
 // A combine group used to for prelegalizer combiners at -O0. The combines in
 // this group have been selected based on experiments to balance code size and
 // compile time performance.
 def optnone_combines : GICombineGroup<[trivial_combines,
     ptr_add_immed_chain, combines_for_extload,
     not_cmp_fold, opt_brcond_by_inverting_cond]>;

llvm/lib/CodeGen/GlobalISel/CombinerHelper.cpp

 #include "llvm/CodeGen/MachineBasicBlock.h"
 #include "llvm/CodeGen/MachineDominators.h"
 #include "llvm/CodeGen/MachineFrameInfo.h"
 #include "llvm/CodeGen/MachineInstr.h"
 #include "llvm/CodeGen/MachineMemOperand.h"
 #include "llvm/CodeGen/MachineRegisterInfo.h"
 #include "llvm/CodeGen/TargetInstrInfo.h"
 #include "llvm/CodeGen/TargetLowering.h"
+#include "llvm/Target/TargetMachine.h"
 #include "llvm/CodeGen/TargetOpcodes.h"
 #include "llvm/IR/DataLayout.h"
 #include "llvm/Support/Casting.h"
 #include "llvm/Support/DivisionByConstantInfo.h"
 #include "llvm/Support/MathExtras.h"
 #include <tuple>
 
 #define DEBUG_TYPE "gi-combiner"
CombinerHelper::findLoadOffsetsForLoadOrCombine(
 
   // Each load found for the pattern. There should be one for each RegsToVisit.
   SmallSetVector<const MachineInstr *, 8> Loads;
 
   // The lowest index used in any load. (The lowest "i" for each x[i].)
   int64_t LowestIdx = INT64_MAX;
 
   // The load which uses the lowest index.
   GZExtLoad *LowestIdxLoad = nullptr;
 
   // Keeps track of the load indices we see. We shouldn't see any indices twice.
   SmallSet<int64_t, 8> SeenIdx;
 
   // Ensure each load is in the same MBB.
   // TODO: Support multiple MachineBasicBlocks.
   MachineBasicBlock *MBB = nullptr;
   const MachineMemOperand *MMO = nullptr;
 
   // Earliest instruction-order load in the pattern.
   GZExtLoad *EarliestLoad = nullptr;
 
   // Latest instruction-order load in the pattern.
   GZExtLoad *LatestLoad = nullptr;
 
   // Base pointer which every load should share.
   Register BasePtr;
 
   // We want to find a load for each register. Each load should have some
   // appropriate bit twiddling arithmetic. During this loop, we will also keep
   // track of the load which uses the lowest index. Later, we will check if we
   // can use its pointer in the final, combined load.
for (auto Reg : RegsToVisit) {
 
     // TODO: Handle multiple MachineBasicBlocks. Currently not handled because
     // it is difficult to check for stores/calls/etc between loads.
     MachineBasicBlock *LoadMBB = Load->getParent();
     if (!MBB)
       MBB = LoadMBB;
     if (LoadMBB != MBB)
       return None;
 
     // Make sure that the MachineMemOperands of every seen load are compatible.
     auto &LoadMMO = Load->getMMO();
     if (!MMO)
       MMO = &LoadMMO;
     if (MMO->getAddrSpace() != LoadMMO.getAddrSpace())
       return None;
 
     // Find out what the base pointer and index for the load is.
     Register LoadPtr;
     int64_t Idx;
     if (!mi_match(Load->getOperand(1).getReg(), MRI,
                   m_GPtrAdd(m_Reg(LoadPtr), m_ICst(Idx)))) {
       LoadPtr = Load->getOperand(1).getReg();
       Idx = 0;
     }
 
     // Don't combine things like a[i], a[i] -> a bigger load.
     if (!SeenIdx.insert(Idx).second)
       return None;
bool CombinerHelper::matchTruncStoreMerge(MachineInstr &MI,
   const auto &DL = LastStore.getMF()->getDataLayout();
   auto &C = LastStore.getMF()->getFunction().getContext();
   // Check that a store of the wide type is both allowed and fast on the target
   bool Fast = false;
   bool Allowed = getTargetLowering().allowsMemoryAccess(
       C, DL, WideStoreTy, LowestIdxStore->getMMO(), &Fast);
   if (!Allowed || !Fast)
     return false;
 
   // Check if the pieces of the value are going to the expected places in memory
   // to merge the stores.
   unsigned NarrowBits = MemTy.getScalarSizeInBits();
   auto checkOffsets = [&](bool MatchLittleEndian) {
     if (MatchLittleEndian) {
       for (unsigned i = 0; i != NumStoresRequired; ++i)
         if (OffsetMap[i] != i * (NarrowBits / 8) + LowestIdxOffset)
           return false;
bool CombinerHelper::matchOrShiftToFunnelShift(MachineInstr &MI,
   } else {
     return false;
   }
 
   if (ShlAmt != LShrAmt)
     return false;
 
   LLT AmtTy = MRI.getType(ShlAmt);
   if (!isLegalOrBeforeLegalizer({FshOpc, {Ty, AmtTy}}))
     return false;
 
   MatchInfo = [=](MachineIRBuilder &B) {
     B.buildInstr(FshOpc, {Dst}, {ShlSrc, LShrSrc, ShlAmt});
   };
   return true;
 }
 
 /// Match an FSHL or FSHR that can be combined to a ROTR or ROTL rotate.
 bool CombinerHelper::matchFunnelShiftToRotate(MachineInstr &MI) {
   unsigned Opc = MI.getOpcode();
   assert(Opc == TargetOpcode::G_FSHL || Opc == TargetOpcode::G_FSHR);
   Register X = MI.getOperand(1).getReg();
   Register Y = MI.getOperand(2).getReg();
   if (X != Y)
     return false;
   unsigned RotateOpc =
       Opc == TargetOpcode::G_FSHL ? TargetOpcode::G_ROTL : TargetOpcode::G_ROTR;
   return isLegalOrBeforeLegalizer({RotateOpc, {MRI.getType(X), MRI.getType(Y)}});
 }
 
 void CombinerHelper::applyFunnelShiftToRotate(MachineInstr &MI) {
   unsigned Opc = MI.getOpcode();
   assert(Opc == TargetOpcode::G_FSHL || Opc == TargetOpcode::G_FSHR);
   bool IsFSHL = Opc == TargetOpcode::G_FSHL;
   Observer.changingInstr(MI);
   MI.setDesc(Builder.getTII().get(IsFSHL ? TargetOpcode::G_ROTL
                                          : TargetOpcode::G_ROTR));
   MI.RemoveOperand(2);
   Observer.changedInstr(MI);
 }
 
 // Fold (rot x, c) -> (rot x, c % BitSize)
 bool CombinerHelper::matchRotateOutOfRange(MachineInstr &MI) {
   assert(MI.getOpcode() == TargetOpcode::G_ROTL ||
          MI.getOpcode() == TargetOpcode::G_ROTR);
   unsigned Bitsize =
       MRI.getType(MI.getOperand(0).getReg()).getScalarSizeInBits();
   Register AmtReg = MI.getOperand(2).getReg();
   bool OutOfRange = false;
   auto MatchOutOfRange = [Bitsize, &OutOfRange](const Constant *C) {
     if (auto *CI = dyn_cast<ConstantInt>(C))
       OutOfRange |= CI->getValue().uge(Bitsize);
     return true;
   };
   return matchUnaryPredicate(MRI, AmtReg, MatchOutOfRange) && OutOfRange;
 }
 
 void CombinerHelper::applyRotateOutOfRange(MachineInstr &MI) {
   assert(MI.getOpcode() == TargetOpcode::G_ROTL ||
          MI.getOpcode() == TargetOpcode::G_ROTR);
   unsigned Bitsize =
       MRI.getType(MI.getOperand(0).getReg()).getScalarSizeInBits();
auto BuildUDIVPattern = [&](const Constant *C) {
     PreShifts.push_back(
         MIB.buildConstant(ScalarShiftAmtTy, PreShift).getReg(0));
     MagicFactors.push_back(MIB.buildConstant(ScalarTy, Magic).getReg(0));
     NPQFactors.push_back(
         MIB.buildConstant(ScalarTy,
                           SelNPQ ? APInt::getOneBitSet(EltBits, EltBits - 1)
                                  : APInt::getZero(EltBits))
             .getReg(0));
     PostShifts.push_back(
         MIB.buildConstant(ScalarShiftAmtTy, PostShift).getReg(0));
     UseNPQ |= SelNPQ;
     return true;
   };
 
   // Collect the shifts/magic values from each element.
   bool Matched = matchUnaryPredicate(MRI, RHS, BuildUDIVPattern);
   (void)Matched;
   assert(Matched && "Expected unary predicate match to succeed");
 
   Register PreShift, PostShift, MagicFactor, NPQFactor;
   auto *RHSDef = getOpcodeDef<GBuildVector>(RHS, MRI);
   if (RHSDef) {
     PreShift = MIB.buildBuildVector(ShiftAmtTy, PreShifts).getReg(0);
     MagicFactor = MIB.buildBuildVector(Ty, MagicFactors).getReg(0);
     NPQFactor = MIB.buildBuildVector(Ty, NPQFactors).getReg(0);
     PostShift = MIB.buildBuildVector(ShiftAmtTy, PostShifts).getReg(0);
   } else {
     assert(MRI.getType(RHS).isScalar() &&
            "Non-build_vector operation should have been a scalar");
MatchInfo = [=, &MI](MachineIRBuilder &B) {
     MI.setDesc(B.getTII().get(Opc));
     MI.getOperand(1).setReg(X);
     MI.getOperand(2).setReg(Y);
     Observer.changedInstr(MI);
   };
   return true;
 }
 
+/// Checks if \p MI is TargetOpcode::G_FMUL and contractable either
+/// due to global flags or MachineInstr flags.
+static bool isContractableFMul(MachineInstr &MI, bool AllowFusionGlobally) {
+  if (MI.getOpcode() != TargetOpcode::G_FMUL)
+    return false;
+  return AllowFusionGlobally || MI.getFlag(MachineInstr::MIFlag::FmContract);
+}
+
+static bool hasMoreUses(const MachineInstr &MI0, const MachineInstr &MI1,
+                        const MachineRegisterInfo &MRI) {
+  return std::distance(MRI.use_instr_nodbg_begin(MI0.getOperand(0).getReg()),
+                       MRI.use_instr_nodbg_end()) >
+         std::distance(MRI.use_instr_nodbg_begin(MI1.getOperand(0).getReg()),
+                       MRI.use_instr_nodbg_end());
+}
+
+bool CombinerHelper::canCombineFMadOrFMA(MachineInstr &MI,
+                                         bool &AllowFusionGlobally,
+                                         bool &HasFMAD, bool &Aggressive) {
+  auto *MF = MI.getMF();
+  const auto &TLI = *MF->getSubtarget().getTargetLowering();
+  const TargetOptions &Options = MF->getTarget().Options;
+  LLT DstType = MRI.getType(MI.getOperand(0).getReg());
+
+  // Floating-point multiply-add with intermediate rounding.
+  HasFMAD = (LI && TLI.isFMADLegal(MI, DstType));
+  // Floating-point multiply-add without intermediate rounding.
+  bool HasFMA = TLI.isFMAFasterThanFMulAndFAdd(*MF, DstType) &&
+                isLegalOrBeforeLegalizer({TargetOpcode::G_FMA, {DstType}});
+  // No valid opcode, do not combine.
+  if (!HasFMAD && !HasFMA)
+    return false;
+
+  AllowFusionGlobally = Options.AllowFPOpFusion == FPOpFusion::Fast ||
+                        Options.UnsafeFPMath || HasFMAD;
+  // If the addition is not contractable, do not combine.
+  if (!AllowFusionGlobally && !MI.getFlag(MachineInstr::MIFlag::FmContract))
+    return false;
+
+  Aggressive = TLI.enableAggressiveFMAFusion(DstType);
+  return true;
+}
+
+bool CombinerHelper::matchCombineFAddFMulToFMadOrFMA(
+    MachineInstr &MI, std::function<void(MachineIRBuilder &)> &MatchInfo) {
+  assert(MI.getOpcode() == TargetOpcode::G_FADD);
+
+  bool AllowFusionGlobally, HasFMAD, Aggressive;
+  if (!canCombineFMadOrFMA(MI, AllowFusionGlobally, HasFMAD, Aggressive))
+    return false;
+
+  MachineInstr *LHS = MRI.getVRegDef(MI.getOperand(1).getReg());
+  MachineInstr *RHS = MRI.getVRegDef(MI.getOperand(2).getReg());
+  unsigned PreferredFusedOpcode =
+      HasFMAD ? TargetOpcode::G_FMAD : TargetOpcode::G_FMA;
+
+  // If we have two choices trying to fold (fadd (fmul u, v), (fmul x, y)),
+  // prefer to fold the multiply with fewer uses.
+  if (Aggressive && isContractableFMul(*LHS, AllowFusionGlobally) &&
+      isContractableFMul(*RHS, AllowFusionGlobally)) {
+    if (hasMoreUses(*LHS, *RHS, MRI))
+      std::swap(LHS, RHS);
+  }
+
+  // fold (fadd (fmul x, y), z) -> (fma x, y, z)
+  if (isContractableFMul(*LHS, AllowFusionGlobally) &&
+      (Aggressive || MRI.hasOneNonDBGUse(LHS->getOperand(0).getReg()))) {
+    MatchInfo = [=, &MI](MachineIRBuilder &B) {
+      B.buildInstr(PreferredFusedOpcode, {MI.getOperand(0).getReg()},
+                   {LHS->getOperand(1).getReg(), LHS->getOperand(2).getReg(),
+                    RHS->getOperand(0).getReg()});
+    };
+    return true;
+  }
+
+  // fold (fadd x, (fmul y, z)) -> (fma y, z, x)
+  if (isContractableFMul(*RHS, AllowFusionGlobally) &&
+      (Aggressive || MRI.hasOneNonDBGUse(RHS->getOperand(0).getReg()))) {
+    MatchInfo = [=, &MI](MachineIRBuilder &B) {
+      B.buildInstr(PreferredFusedOpcode, {MI.getOperand(0).getReg()},
+                   {RHS->getOperand(1).getReg(), RHS->getOperand(2).getReg(),
+                    LHS->getOperand(0).getReg()});
+    };
+    return true;
+  }
+
+  return false;
+}
+
 bool CombinerHelper::tryCombine(MachineInstr &MI) {
   if (tryCombineCopy(MI))
     return true;
   if (tryCombineExtendingLoads(MI))
     return true;
   if (tryCombineIndexedLoadStore(MI))
     return true;
   return false;
 }

llvm/lib/Target/AMDGPU/SIISelLowering.h

MachineBasicBlock *emitGWSMemViolTestLoop(MachineInstr &MI,
                                             MachineBasicBlock *BB) const;
 
   MachineBasicBlock *
   EmitInstrWithCustomInserter(MachineInstr &MI,
                               MachineBasicBlock *BB) const override;
 
   bool hasBitPreservingFPLogic(EVT VT) const override;
   bool enableAggressiveFMAFusion(EVT VT) const override;
+  bool enableAggressiveFMAFusion(LLT Ty) const override;
   EVT getSetCCResultType(const DataLayout &DL, LLVMContext &Context,
                          EVT VT) const override;
   MVT getScalarShiftAmountTy(const DataLayout &, EVT) const override;
   LLT getPreferredShiftAmountTy(LLT Ty) const override;
 
   bool isFMAFasterThanFMulAndFAdd(const MachineFunction &MF,
                                   EVT VT) const override;
+  bool isFMAFasterThanFMulAndFAdd(const MachineFunction &MF,
+                                  const LLT Ty) const override;
   bool isFMADLegal(const SelectionDAG &DAG, const SDNode *N) const override;
+  bool isFMADLegal(const MachineInstr &MI, const LLT Ty) const override;
 
   SDValue splitUnaryVectorOp(SDValue Op, SelectionDAG &DAG) const;
   SDValue splitBinaryVectorOp(SDValue Op, SelectionDAG &DAG) const;
   SDValue splitTernaryVectorOp(SDValue Op, SelectionDAG &DAG) const;
   SDValue LowerOperation(SDValue Op, SelectionDAG &DAG) const override;
 
   void ReplaceNodeResults(SDNode *N, SmallVectorImpl<SDValue> &Results,
                           SelectionDAG &DAG) const override;

llvm/lib/Target/AMDGPU/SIISelLowering.cpp

bool SITargetLowering::enableAggressiveFMAFusion(EVT VT) const {
   // isFMAFasterThanFMulAndFAdd), we should perform these.
 
   // When fma is quarter rate, for f64 where add / sub are at best half rate,
   // most of these combines appear to be cycle neutral but save on instruction
   // count / code size.
   return true;
 }
 
+bool SITargetLowering::enableAggressiveFMAFusion(LLT Ty) const { return true; }
+
 EVT SITargetLowering::getSetCCResultType(const DataLayout &DL, LLVMContext &Ctx,
                                          EVT VT) const {
   if (!VT.isVector()) {
     return MVT::i1;
   }
   return EVT::getVectorVT(Ctx, MVT::i1, VT.getVectorNumElements());
 }
 
case MVT::f16:
     return Subtarget->has16BitInsts() && hasFP64FP16Denormals(MF);
   default:
     break;
   }
 
   return false;
 }
 
+bool SITargetLowering::isFMAFasterThanFMulAndFAdd(const MachineFunction &MF,
+                                                  LLT Ty) const {
+  switch (Ty.getScalarSizeInBits()) {
+  case 16:
+    return isFMAFasterThanFMulAndFAdd(MF, MVT::f16);
+  case 32:
+    return isFMAFasterThanFMulAndFAdd(MF, MVT::f32);
+  case 64:
+    return isFMAFasterThanFMulAndFAdd(MF, MVT::f64);
+  default:
+    break;
+  }
+
+  return false;
+}
+
+bool SITargetLowering::isFMADLegal(const MachineInstr &MI, LLT Ty) const {
+  if (!Ty.isScalar())
+    return false;
+
+  if (Ty.getScalarSizeInBits() == 16)
+    return Subtarget->hasMadF16() && !hasFP64FP16Denormals(*MI.getMF());
+  if (Ty.getScalarSizeInBits() == 32)
+    return Subtarget->hasMadMacF32Insts() && !hasFP32Denormals(*MI.getMF());
+
+  return false;
+}
+
 bool SITargetLowering::isFMADLegal(const SelectionDAG &DAG,
                                    const SDNode *N) const {
   // TODO: Check future ftz flag
   // v_mad_f32/v_mac_f32 do not support denormals.
   EVT VT = N->getValueType(0);
   if (VT == MVT::f32)
     return Subtarget->hasMadMacF32Insts() &&
            !hasFP32Denormals(DAG.getMachineFunction());

llvm/test/CodeGen/AMDGPU/GlobalISel/combine-fma-add-mul-post-legalize.mir

+# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
+# RUN: llc -global-isel -march=amdgcn -mcpu=gfx900 -run-pass=amdgpu-postlegalizer-combiner %s -o - | FileCheck -check-prefix=GFX9 %s
+# RUN: llc -global-isel -march=amdgcn -mcpu=gfx900 -run-pass=amdgpu-postlegalizer-combiner -fp-contract=fast %s -o - | FileCheck -check-prefix=GFX9-CONTRACT %s
+# RUN: llc -global-isel -march=amdgcn -mcpu=gfx900 -run-pass=amdgpu-postlegalizer-combiner --denormal-fp-math=preserve-sign %s -o - | FileCheck -check-prefix=GFX9-DENORM %s
+# RUN: llc -global-isel -march=amdgcn -mcpu=gfx900 -run-pass=amdgpu-postlegalizer-combiner -enable-unsafe-fp-math %s -o - | FileCheck -check-prefix=GFX9-UNSAFE %s
+# RUN: llc -global-isel -march=amdgcn -mcpu=gfx1010 -run-pass=amdgpu-postlegalizer-combiner %s -o - | FileCheck -check-prefix=GFX10 %s
+# RUN: llc -global-isel -march=amdgcn -mcpu=gfx1010 -run-pass=amdgpu-postlegalizer-combiner -fp-contract=fast %s -o - | FileCheck -check-prefix=GFX10-CONTRACT %s
+# RUN: llc -global-isel -march=amdgcn -mcpu=gfx1010 -run-pass=amdgpu-postlegalizer-combiner --denormal-fp-math=preserve-sign %s -o - | FileCheck -check-prefix=GFX10-DENORM %s
+# RUN: llc -global-isel -march=amdgcn -mcpu=gfx1010 -run-pass=amdgpu-postlegalizer-combiner -enable-unsafe-fp-math %s -o - | FileCheck -check-prefix=GFX10-UNSAFE %s
+
+---
+name:            test_f32_add_mul
+body:             |
+  bb.1.entry:
+    liveins: $vgpr0, $vgpr1, $vgpr2, $sgpr30_sgpr31
+
+    ; GFX9-LABEL: name: test_f32_add_mul
+    ; GFX9: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX9: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX9: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX9: [[COPY3:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX9: [[FMUL:%[0-9]+]]:_(s32) = G_FMUL [[COPY]], [[COPY1]]
+    ; GFX9: [[FADD:%[0-9]+]]:_(s32) = G_FADD [[FMUL]], [[COPY2]]
+    ; GFX9: $vgpr0 = COPY [[FADD]](s32)
+    ; GFX9: [[COPY4:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY3]]
+    ; GFX9: S_SETPC_B64_return [[COPY4]], implicit $vgpr0
+    ; GFX9-CONTRACT-LABEL: name: test_f32_add_mul
+    ; GFX9-CONTRACT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX9-CONTRACT: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX9-CONTRACT: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX9-CONTRACT: [[COPY3:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX9-CONTRACT: [[FMA:%[0-9]+]]:_(s32) = G_FMA [[COPY]], [[COPY1]], [[COPY2]]
+    ; GFX9-CONTRACT: $vgpr0 = COPY [[FMA]](s32)
+    ; GFX9-CONTRACT: [[COPY4:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY3]]
+    ; GFX9-CONTRACT: S_SETPC_B64_return [[COPY4]], implicit $vgpr0
+    ; GFX9-DENORM-LABEL: name: test_f32_add_mul
+    ; GFX9-DENORM: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX9-DENORM: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX9-DENORM: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX9-DENORM: [[COPY3:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX9-DENORM: [[FMUL:%[0-9]+]]:_(s32) = G_FMUL [[COPY]], [[COPY1]]
+    ; GFX9-DENORM: [[FADD:%[0-9]+]]:_(s32) = G_FADD [[FMUL]], [[COPY2]]
+    ; GFX9-DENORM: $vgpr0 = COPY [[FADD]](s32)
+    ; GFX9-DENORM: [[COPY4:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY3]]
+    ; GFX9-DENORM: S_SETPC_B64_return [[COPY4]], implicit $vgpr0
+    ; GFX9-UNSAFE-LABEL: name: test_f32_add_mul
+    ; GFX9-UNSAFE: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX9-UNSAFE: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX9-UNSAFE: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX9-UNSAFE: [[COPY3:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX9-UNSAFE: [[FMA:%[0-9]+]]:_(s32) = G_FMA [[COPY]], [[COPY1]], [[COPY2]]
+    ; GFX9-UNSAFE: $vgpr0 = COPY [[FMA]](s32)
+    ; GFX9-UNSAFE: [[COPY4:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY3]]
+    ; GFX9-UNSAFE: S_SETPC_B64_return [[COPY4]], implicit $vgpr0
+    ; GFX10-LABEL: name: test_f32_add_mul
+    ; GFX10: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX10: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX10: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX10: [[COPY3:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX10: [[FMUL:%[0-9]+]]:_(s32) = G_FMUL [[COPY]], [[COPY1]]
+    ; GFX10: [[FADD:%[0-9]+]]:_(s32) = G_FADD [[FMUL]], [[COPY2]]
+    ; GFX10: $vgpr0 = COPY [[FADD]](s32)
+    ; GFX10: [[COPY4:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY3]]
+    ; GFX10: S_SETPC_B64_return [[COPY4]], implicit $vgpr0
+    ; GFX10-CONTRACT-LABEL: name: test_f32_add_mul
+    ; GFX10-CONTRACT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX10-CONTRACT: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX10-CONTRACT: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX10-CONTRACT: [[COPY3:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX10-CONTRACT: [[FMA:%[0-9]+]]:_(s32) = G_FMA [[COPY]], [[COPY1]], [[COPY2]]
+    ; GFX10-CONTRACT: $vgpr0 = COPY [[FMA]](s32)
+    ; GFX10-CONTRACT: [[COPY4:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY3]]
+    ; GFX10-CONTRACT: S_SETPC_B64_return [[COPY4]], implicit $vgpr0
+    ; GFX10-DENORM-LABEL: name: test_f32_add_mul
+    ; GFX10-DENORM: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX10-DENORM: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX10-DENORM: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX10-DENORM: [[COPY3:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX10-DENORM: [[FMUL:%[0-9]+]]:_(s32) = G_FMUL [[COPY]], [[COPY1]]
+    ; GFX10-DENORM: [[FADD:%[0-9]+]]:_(s32) = G_FADD [[FMUL]], [[COPY2]]
+    ; GFX10-DENORM: $vgpr0 = COPY [[FADD]](s32)
+    ; GFX10-DENORM: [[COPY4:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY3]]
+    ; GFX10-DENORM: S_SETPC_B64_return [[COPY4]], implicit $vgpr0
+    ; GFX10-UNSAFE-LABEL: name: test_f32_add_mul
+    ; GFX10-UNSAFE: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX10-UNSAFE: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX10-UNSAFE: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX10-UNSAFE: [[COPY3:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX10-UNSAFE: [[FMA:%[0-9]+]]:_(s32) = G_FMA [[COPY]], [[COPY1]], [[COPY2]]
+    ; GFX10-UNSAFE: $vgpr0 = COPY [[FMA]](s32)
+    ; GFX10-UNSAFE: [[COPY4:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY3]]
+    ; GFX10-UNSAFE: S_SETPC_B64_return [[COPY4]], implicit $vgpr0
+    %0:_(s32) = COPY $vgpr0
+    %1:_(s32) = COPY $vgpr1
+    %2:_(s32) = COPY $vgpr2
+    %3:sgpr_64 = COPY $sgpr30_sgpr31
+    %4:_(s32) = G_FMUL %0, %1
+    %5:_(s32) = G_FADD %4, %2
+    $vgpr0 = COPY %5(s32)
+    %6:ccr_sgpr_64 = COPY %3
+    S_SETPC_B64_return %6, implicit $vgpr0
+...
+
+---
+name:            test_f32_add_mul_rhs
+body:             |
+  bb.1.entry:
+    liveins: $vgpr0, $vgpr1, $vgpr2, $sgpr30_sgpr31
+
+    ; GFX9-LABEL: name: test_f32_add_mul_rhs
+    ; GFX9: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX9: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX9: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX9: [[COPY3:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX9: [[FMUL:%[0-9]+]]:_(s32) = G_FMUL [[COPY]], [[COPY1]]
+    ; GFX9: [[FADD:%[0-9]+]]:_(s32) = G_FADD [[COPY2]], [[FMUL]]
+    ; GFX9: $vgpr0 = COPY [[FADD]](s32)
+    ; GFX9: [[COPY4:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY3]]
+    ; GFX9: S_SETPC_B64_return [[COPY4]], implicit $vgpr0
+    ; GFX9-CONTRACT-LABEL: name: test_f32_add_mul_rhs
+    ; GFX9-CONTRACT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX9-CONTRACT: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX9-CONTRACT: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX9-CONTRACT: [[COPY3:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX9-CONTRACT: [[FMA:%[0-9]+]]:_(s32) = G_FMA [[COPY]], [[COPY1]], [[COPY2]]
+    ; GFX9-CONTRACT: $vgpr0 = COPY [[FMA]](s32)
+    ; GFX9-CONTRACT: [[COPY4:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY3]]
+    ; GFX9-CONTRACT: S_SETPC_B64_return [[COPY4]], implicit $vgpr0
+    ; GFX9-DENORM-LABEL: name: test_f32_add_mul_rhs
+    ; GFX9-DENORM: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX9-DENORM: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX9-DENORM: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX9-DENORM: [[COPY3:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX9-DENORM: [[FMUL:%[0-9]+]]:_(s32) = G_FMUL [[COPY]], [[COPY1]]
+    ; GFX9-DENORM: [[FADD:%[0-9]+]]:_(s32) = G_FADD [[COPY2]], [[FMUL]]
+    ; GFX9-DENORM: $vgpr0 = COPY [[FADD]](s32)
+    ; GFX9-DENORM: [[COPY4:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY3]]
+    ; GFX9-DENORM: S_SETPC_B64_return [[COPY4]], implicit $vgpr0
+    ; GFX9-UNSAFE-LABEL: name: test_f32_add_mul_rhs
+    ; GFX9-UNSAFE: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX9-UNSAFE: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX9-UNSAFE: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX9-UNSAFE: [[COPY3:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX9-UNSAFE: [[FMA:%[0-9]+]]:_(s32) = G_FMA [[COPY]], [[COPY1]], [[COPY2]]
+    ; GFX9-UNSAFE: $vgpr0 = COPY [[FMA]](s32)
+    ; GFX9-UNSAFE: [[COPY4:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY3]]
+    ; GFX9-UNSAFE: S_SETPC_B64_return [[COPY4]], implicit $vgpr0
+    ; GFX10-LABEL: name: test_f32_add_mul_rhs
+    ; GFX10: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX10: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX10: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX10: [[COPY3:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX10: [[FMUL:%[0-9]+]]:_(s32) = G_FMUL [[COPY]], [[COPY1]]
+    ; GFX10: [[FADD:%[0-9]+]]:_(s32) = G_FADD [[COPY2]], [[FMUL]]
+    ; GFX10: $vgpr0 = COPY [[FADD]](s32)
+    ; GFX10: [[COPY4:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY3]]
+    ; GFX10: S_SETPC_B64_return [[COPY4]], implicit $vgpr0
+    ; GFX10-CONTRACT-LABEL: name: test_f32_add_mul_rhs
+    ; GFX10-CONTRACT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX10-CONTRACT: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX10-CONTRACT: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX10-CONTRACT: [[COPY3:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX10-CONTRACT: [[FMA:%[0-9]+]]:_(s32) = G_FMA [[COPY]], [[COPY1]], [[COPY2]]
+    ; GFX10-CONTRACT: $vgpr0 = COPY [[FMA]](s32)
+    ; GFX10-CONTRACT: [[COPY4:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY3]]
+    ; GFX10-CONTRACT: S_SETPC_B64_return [[COPY4]], implicit $vgpr0
+    ; GFX10-DENORM-LABEL: name: test_f32_add_mul_rhs
+    ; GFX10-DENORM: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX10-DENORM: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX10-DENORM: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX10-DENORM: [[COPY3:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX10-DENORM: [[FMUL:%[0-9]+]]:_(s32) = G_FMUL [[COPY]], [[COPY1]]
+    ; GFX10-DENORM: [[FADD:%[0-9]+]]:_(s32) = G_FADD [[COPY2]], [[FMUL]]
+    ; GFX10-DENORM: $vgpr0 = COPY [[FADD]](s32)
+    ; GFX10-DENORM: [[COPY4:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY3]]
+    ; GFX10-DENORM: S_SETPC_B64_return [[COPY4]], implicit $vgpr0
+    ; GFX10-UNSAFE-LABEL: name: test_f32_add_mul_rhs
+    ; GFX10-UNSAFE: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX10-UNSAFE: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX10-UNSAFE: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX10-UNSAFE: [[COPY3:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX10-UNSAFE: [[FMA:%[0-9]+]]:_(s32) = G_FMA [[COPY]], [[COPY1]], [[COPY2]]
+    ; GFX10-UNSAFE: $vgpr0 = COPY [[FMA]](s32)
+    ; GFX10-UNSAFE: [[COPY4:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY3]]
+    ; GFX10-UNSAFE: S_SETPC_B64_return [[COPY4]], implicit $vgpr0
+    %0:_(s32) = COPY $vgpr0
+    %1:_(s32) = COPY $vgpr1
+    %2:_(s32) = COPY $vgpr2
+    %3:sgpr_64 = COPY $sgpr30_sgpr31
+    %4:_(s32) = G_FMUL %0, %1
+    %5:_(s32) = G_FADD %2, %4
+    $vgpr0 = COPY %5(s32)
+    %6:ccr_sgpr_64 = COPY %3
+    S_SETPC_B64_return %6, implicit $vgpr0
+...
+
+---
+name:            test_half_add_mul
+body:             |
+  bb.1.entry:
+    liveins: $vgpr0, $vgpr1, $vgpr2, $sgpr30_sgpr31
+
+    ; GFX9-LABEL: name: test_half_add_mul
+    ; GFX9: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX9: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY]](s32)
+    ; GFX9: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX9: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[COPY1]](s32)
+    ; GFX9: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX9: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[COPY2]](s32)
+    ; GFX9: [[COPY3:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX9: [[FMUL:%[0-9]+]]:_(s16) = G_FMUL [[TRUNC]], [[TRUNC1]]
+    ; GFX9: [[FADD:%[0-9]+]]:_(s16) = G_FADD [[FMUL]], [[TRUNC2]]
+    ; GFX9: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[FADD]](s16)
+    ; GFX9: $vgpr0 = COPY [[ANYEXT]](s32)
+    ; GFX9: [[COPY4:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY3]]
+    ; GFX9: S_SETPC_B64_return [[COPY4]], implicit $vgpr0
+    ; GFX9-CONTRACT-LABEL: name: test_half_add_mul
+    ; GFX9-CONTRACT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX9-CONTRACT: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY]](s32)
+    ; GFX9-CONTRACT: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX9-CONTRACT: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[COPY1]](s32)
+    ; GFX9-CONTRACT: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX9-CONTRACT: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[COPY2]](s32)
+    ; GFX9-CONTRACT: [[COPY3:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX9-CONTRACT: [[FMA:%[0-9]+]]:_(s16) = G_FMA [[TRUNC]], [[TRUNC1]], [[TRUNC2]]
+    ; GFX9-CONTRACT: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[FMA]](s16)
+    ; GFX9-CONTRACT: $vgpr0 = COPY [[ANYEXT]](s32)
+    ; GFX9-CONTRACT: [[COPY4:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY3]]
+    ; GFX9-CONTRACT: S_SETPC_B64_return [[COPY4]], implicit $vgpr0
+    ; GFX9-DENORM-LABEL: name: test_half_add_mul
+    ; GFX9-DENORM: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX9-DENORM: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY]](s32)
+    ; GFX9-DENORM: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX9-DENORM: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[COPY1]](s32)
+    ; GFX9-DENORM: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX9-DENORM: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[COPY2]](s32)
+    ; GFX9-DENORM: [[COPY3:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX9-DENORM: [[FMUL:%[0-9]+]]:_(s16) = G_FMUL [[TRUNC]], [[TRUNC1]]
+    ; GFX9-DENORM: [[FADD:%[0-9]+]]:_(s16) = G_FADD [[FMUL]], [[TRUNC2]]
+    ; GFX9-DENORM: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[FADD]](s16)
+    ; GFX9-DENORM: $vgpr0 = COPY [[ANYEXT]](s32)
+    ; GFX9-DENORM: [[COPY4:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY3]]
+    ; GFX9-DENORM: S_SETPC_B64_return [[COPY4]], implicit $vgpr0
+    ; GFX9-UNSAFE-LABEL: name: test_half_add_mul
+    ; GFX9-UNSAFE: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX9-UNSAFE: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY]](s32)
+    ; GFX9-UNSAFE: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX9-UNSAFE: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[COPY1]](s32)
+    ; GFX9-UNSAFE: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX9-UNSAFE: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[COPY2]](s32)
+    ; GFX9-UNSAFE: [[COPY3:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX9-UNSAFE: [[FMA:%[0-9]+]]:_(s16) = G_FMA [[TRUNC]], [[TRUNC1]], [[TRUNC2]]
+    ; GFX9-UNSAFE: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[FMA]](s16)
+    ; GFX9-UNSAFE: $vgpr0 = COPY [[ANYEXT]](s32)
+    ; GFX9-UNSAFE: [[COPY4:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY3]]
+    ; GFX9-UNSAFE: S_SETPC_B64_return [[COPY4]], implicit $vgpr0
+    ; GFX10-LABEL: name: test_half_add_mul
+    ; GFX10: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX10: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY]](s32)
+    ; GFX10: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX10: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[COPY1]](s32)
+    ; GFX10: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX10: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[COPY2]](s32)
+    ; GFX10: [[COPY3:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX10: [[FMUL:%[0-9]+]]:_(s16) = G_FMUL [[TRUNC]], [[TRUNC1]]
+    ; GFX10: [[FADD:%[0-9]+]]:_(s16) = G_FADD [[FMUL]], [[TRUNC2]]
+    ; GFX10: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[FADD]](s16)
+    ; GFX10: $vgpr0 = COPY [[ANYEXT]](s32)
+    ; GFX10: [[COPY4:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY3]]
+    ; GFX10: S_SETPC_B64_return [[COPY4]], implicit $vgpr0
+    ; GFX10-CONTRACT-LABEL: name: test_half_add_mul
+    ; GFX10-CONTRACT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX10-CONTRACT: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY]](s32)
+    ; GFX10-CONTRACT: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX10-CONTRACT: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[COPY1]](s32)
+    ; GFX10-CONTRACT: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX10-CONTRACT: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[COPY2]](s32)
+    ; GFX10-CONTRACT: [[COPY3:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX10-CONTRACT: [[FMA:%[0-9]+]]:_(s16) = G_FMA [[TRUNC]], [[TRUNC1]], [[TRUNC2]]
+    ; GFX10-CONTRACT: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[FMA]](s16)
+    ; GFX10-CONTRACT: $vgpr0 = COPY [[ANYEXT]](s32)
+    ; GFX10-CONTRACT: [[COPY4:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY3]]
+    ; GFX10-CONTRACT: S_SETPC_B64_return [[COPY4]], implicit $vgpr0
+    ; GFX10-DENORM-LABEL: name: test_half_add_mul
+    ; GFX10-DENORM: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX10-DENORM: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY]](s32)
+    ; GFX10-DENORM: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX10-DENORM: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[COPY1]](s32)
+    ; GFX10-DENORM: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX10-DENORM: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[COPY2]](s32)
+    ; GFX10-DENORM: [[COPY3:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX10-DENORM: [[FMUL:%[0-9]+]]:_(s16) = G_FMUL [[TRUNC]], [[TRUNC1]]
+    ; GFX10-DENORM: [[FADD:%[0-9]+]]:_(s16) = G_FADD [[FMUL]], [[TRUNC2]]
+    ; GFX10-DENORM: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[FADD]](s16)
+    ; GFX10-DENORM: $vgpr0 = COPY [[ANYEXT]](s32)
+    ; GFX10-DENORM: [[COPY4:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY3]]
+    ; GFX10-DENORM: S_SETPC_B64_return [[COPY4]], implicit $vgpr0
+    ; GFX10-UNSAFE-LABEL: name: test_half_add_mul
+    ; GFX10-UNSAFE: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX10-UNSAFE: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY]](s32)
+    ; GFX10-UNSAFE: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX10-UNSAFE: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[COPY1]](s32)
+    ; GFX10-UNSAFE: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX10-UNSAFE: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[COPY2]](s32)
+    ; GFX10-UNSAFE: [[COPY3:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX10-UNSAFE: [[FMA:%[0-9]+]]:_(s16) = G_FMA [[TRUNC]], [[TRUNC1]], [[TRUNC2]]
+    ; GFX10-UNSAFE: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[FMA]](s16)
+    ; GFX10-UNSAFE: $vgpr0 = COPY [[ANYEXT]](s32)
+    ; GFX10-UNSAFE: [[COPY4:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY3]]
+    ; GFX10-UNSAFE: S_SETPC_B64_return [[COPY4]], implicit $vgpr0
+    %4:_(s32) = COPY $vgpr0
+    %0:_(s16) = G_TRUNC %4(s32)
+    %5:_(s32) = COPY $vgpr1
+    %1:_(s16) = G_TRUNC %5(s32)
+    %6:_(s32) = COPY $vgpr2
+    %2:_(s16) = G_TRUNC %6(s32)
+    %3:sgpr_64 = COPY $sgpr30_sgpr31
+    %7:_(s16) = G_FMUL %0, %1
+    %8:_(s16) = G_FADD %7, %2
+    %10:_(s32) = G_ANYEXT %8(s16)
+    $vgpr0 = COPY %10(s32)
+    %9:ccr_sgpr_64 = COPY %3
+    S_SETPC_B64_return %9, implicit $vgpr0
+...
+
+---
+name:            test_half_add_mul_rhs
+body:             |
+  bb.1.entry:
+    liveins: $vgpr0, $vgpr1, $vgpr2, $sgpr30_sgpr31
+
+    ; GFX9-LABEL: name: test_half_add_mul_rhs
+    ; GFX9: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX9: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY]](s32)
+    ; GFX9: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX9: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[COPY1]](s32)
+    ; GFX9: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX9: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[COPY2]](s32)
+    ; GFX9: [[COPY3:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX9: [[FMUL:%[0-9]+]]:_(s16) = G_FMUL [[TRUNC]], [[TRUNC1]]
+    ; GFX9: [[FADD:%[0-9]+]]:_(s16) = G_FADD [[TRUNC2]], [[FMUL]]
+    ; GFX9: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[FADD]](s16)
+    ; GFX9: $vgpr0 = COPY [[ANYEXT]](s32)
+    ; GFX9: [[COPY4:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY3]]
+    ; GFX9: S_SETPC_B64_return [[COPY4]], implicit $vgpr0
+    ; GFX9-CONTRACT-LABEL: name: test_half_add_mul_rhs
+    ; GFX9-CONTRACT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX9-CONTRACT: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY]](s32)
+    ; GFX9-CONTRACT: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX9-CONTRACT: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[COPY1]](s32)
+    ; GFX9-CONTRACT: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX9-CONTRACT: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[COPY2]](s32)
+    ; GFX9-CONTRACT: [[COPY3:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX9-CONTRACT: [[FMA:%[0-9]+]]:_(s16) = G_FMA [[TRUNC]], [[TRUNC1]], [[TRUNC2]]
+    ; GFX9-CONTRACT: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[FMA]](s16)
+    ; GFX9-CONTRACT: $vgpr0 = COPY [[ANYEXT]](s32)
+    ; GFX9-CONTRACT: [[COPY4:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY3]]
+    ; GFX9-CONTRACT: S_SETPC_B64_return [[COPY4]], implicit $vgpr0
+    ; GFX9-DENORM-LABEL: name: test_half_add_mul_rhs
+    ; GFX9-DENORM: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX9-DENORM: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY]](s32)
+    ; GFX9-DENORM: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX9-DENORM: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[COPY1]](s32)
+    ; GFX9-DENORM: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX9-DENORM: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[COPY2]](s32)
+    ; GFX9-DENORM: [[COPY3:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX9-DENORM: [[FMUL:%[0-9]+]]:_(s16) = G_FMUL [[TRUNC]], [[TRUNC1]]
+    ; GFX9-DENORM: [[FADD:%[0-9]+]]:_(s16) = G_FADD [[TRUNC2]], [[FMUL]]
+    ; GFX9-DENORM: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[FADD]](s16)
+    ; GFX9-DENORM: $vgpr0 = COPY [[ANYEXT]](s32)
+    ; GFX9-DENORM: [[COPY4:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY3]]
+    ; GFX9-DENORM: S_SETPC_B64_return [[COPY4]], implicit $vgpr0
+    ; GFX9-UNSAFE-LABEL: name: test_half_add_mul_rhs
+    ; GFX9-UNSAFE: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX9-UNSAFE: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY]](s32)
+    ; GFX9-UNSAFE: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX9-UNSAFE: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[COPY1]](s32)
+    ; GFX9-UNSAFE: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX9-UNSAFE: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[COPY2]](s32)
+    ; GFX9-UNSAFE: [[COPY3:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX9-UNSAFE: [[FMA:%[0-9]+]]:_(s16) = G_FMA [[TRUNC]], [[TRUNC1]], [[TRUNC2]]
+    ; GFX9-UNSAFE: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[FMA]](s16)
+    ; GFX9-UNSAFE: $vgpr0 = COPY [[ANYEXT]](s32)
+    ; GFX9-UNSAFE: [[COPY4:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY3]]
+    ; GFX9-UNSAFE: S_SETPC_B64_return [[COPY4]], implicit $vgpr0
+    ; GFX10-LABEL: name: test_half_add_mul_rhs
+    ; GFX10: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX10: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY]](s32)
+    ; GFX10: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX10: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[COPY1]](s32)
+    ; GFX10: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX10: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[COPY2]](s32)
+    ; GFX10: [[COPY3:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX10: [[FMUL:%[0-9]+]]:_(s16) = G_FMUL [[TRUNC]], [[TRUNC1]]
+    ; GFX10: [[FADD:%[0-9]+]]:_(s16) = G_FADD [[TRUNC2]], [[FMUL]]
+    ; GFX10: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[FADD]](s16)
+    ; GFX10: $vgpr0 = COPY [[ANYEXT]](s32)
+    ; GFX10: [[COPY4:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY3]]
+    ; GFX10: S_SETPC_B64_return [[COPY4]], implicit $vgpr0
+    ; GFX10-CONTRACT-LABEL: name: test_half_add_mul_rhs
+    ; GFX10-CONTRACT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX10-CONTRACT: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY]](s32)
+    ; GFX10-CONTRACT: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX10-CONTRACT: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[COPY1]](s32)
+    ; GFX10-CONTRACT: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX10-CONTRACT: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[COPY2]](s32)
+    ; GFX10-CONTRACT: [[COPY3:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX10-CONTRACT: [[FMA:%[0-9]+]]:_(s16) = G_FMA [[TRUNC]], [[TRUNC1]], [[TRUNC2]]
+    ; GFX10-CONTRACT: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[FMA]](s16)
+    ; GFX10-CONTRACT: $vgpr0 = COPY [[ANYEXT]](s32)
+    ; GFX10-CONTRACT: [[COPY4:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY3]]
+    ; GFX10-CONTRACT: S_SETPC_B64_return [[COPY4]], implicit $vgpr0
+    ; GFX10-DENORM-LABEL: name: test_half_add_mul_rhs
+    ; GFX10-DENORM: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX10-DENORM: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY]](s32)
+    ; GFX10-DENORM: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX10-DENORM: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[COPY1]](s32)
+    ; GFX10-DENORM: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX10-DENORM: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[COPY2]](s32)
+    ; GFX10-DENORM: [[COPY3:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX10-DENORM: [[FMUL:%[0-9]+]]:_(s16) = G_FMUL [[TRUNC]], [[TRUNC1]]
+    ; GFX10-DENORM: [[FADD:%[0-9]+]]:_(s16) = G_FADD [[TRUNC2]], [[FMUL]]
+    ; GFX10-DENORM: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[FADD]](s16)
+    ; GFX10-DENORM: $vgpr0 = COPY [[ANYEXT]](s32)
+    ; GFX10-DENORM: [[COPY4:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY3]]
+    ; GFX10-DENORM: S_SETPC_B64_return [[COPY4]], implicit $vgpr0
+    ; GFX10-UNSAFE-LABEL: name: test_half_add_mul_rhs
+    ; GFX10-UNSAFE: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX10-UNSAFE: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY]](s32)
+    ; GFX10-UNSAFE: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX10-UNSAFE: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[COPY1]](s32)
+    ; GFX10-UNSAFE: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX10-UNSAFE: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[COPY2]](s32)
+    ; GFX10-UNSAFE: [[COPY3:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX10-UNSAFE: [[FMA:%[0-9]+]]:_(s16) = G_FMA [[TRUNC]], [[TRUNC1]], [[TRUNC2]]
+    ; GFX10-UNSAFE: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[FMA]](s16)
+    ; GFX10-UNSAFE: $vgpr0 = COPY [[ANYEXT]](s32)
+    ; GFX10-UNSAFE: [[COPY4:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY3]]
+    ; GFX10-UNSAFE: S_SETPC_B64_return [[COPY4]], implicit $vgpr0
+    %4:_(s32) = COPY $vgpr0
+    %0:_(s16) = G_TRUNC %4(s32)
+    %5:_(s32) = COPY $vgpr1
+    %1:_(s16) = G_TRUNC %5(s32)
+    %6:_(s32) = COPY $vgpr2
+    %2:_(s16) = G_TRUNC %6(s32)
+    %3:sgpr_64 = COPY $sgpr30_sgpr31
+    %7:_(s16) = G_FMUL %0, %1
+    %8:_(s16) = G_FADD %2, %7
+    %10:_(s32) = G_ANYEXT %8(s16)
+    $vgpr0 = COPY %10(s32)
+    %9:ccr_sgpr_64 = COPY %3
+    S_SETPC_B64_return %9, implicit $vgpr0
+...
+
+---
+name:            test_double_add_mul
+body:             |
+  bb.1.entry:
+    liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $sgpr30_sgpr31
+
+    ; GFX9-LABEL: name: test_double_add_mul
+    ; GFX9: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX9: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX9: [[MV:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY]](s32), [[COPY1]](s32)
+    ; GFX9: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX9: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3
+    ; GFX9: [[MV1:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY2]](s32), [[COPY3]](s32)
+    ; GFX9: [[COPY4:%[0-9]+]]:_(s32) = COPY $vgpr4
+    ; GFX9: [[COPY5:%[0-9]+]]:_(s32) = COPY $vgpr5
+    ; GFX9: [[MV2:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY4]](s32), [[COPY5]](s32)
+    ; GFX9: [[COPY6:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX9: [[FMUL:%[0-9]+]]:_(s64) = G_FMUL [[MV]], [[MV1]]
+    ; GFX9: [[FADD:%[0-9]+]]:_(s64) = G_FADD [[FMUL]], [[MV2]]
+    ; GFX9: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FADD]](s64)
+    ; GFX9: $vgpr0 = COPY [[UV]](s32)
+    ; GFX9: $vgpr1 = COPY [[UV1]](s32)
+    ; GFX9: [[COPY7:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY6]]
+    ; GFX9: S_SETPC_B64_return [[COPY7]], implicit $vgpr0, implicit $vgpr1
+    ; GFX9-CONTRACT-LABEL: name: test_double_add_mul
+    ; GFX9-CONTRACT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX9-CONTRACT: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX9-CONTRACT: [[MV:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY]](s32), [[COPY1]](s32)
+    ; GFX9-CONTRACT: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX9-CONTRACT: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3
+    ; GFX9-CONTRACT: [[MV1:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY2]](s32), [[COPY3]](s32)
+    ; GFX9-CONTRACT: [[COPY4:%[0-9]+]]:_(s32) = COPY $vgpr4
+    ; GFX9-CONTRACT: [[COPY5:%[0-9]+]]:_(s32) = COPY $vgpr5
+    ; GFX9-CONTRACT: [[MV2:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY4]](s32), [[COPY5]](s32)
+    ; GFX9-CONTRACT: [[COPY6:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX9-CONTRACT: [[FMA:%[0-9]+]]:_(s64) = G_FMA [[MV]], [[MV1]], [[MV2]]
+    ; GFX9-CONTRACT: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FMA]](s64)
+    ; GFX9-CONTRACT: $vgpr0 = COPY [[UV]](s32)
+    ; GFX9-CONTRACT: $vgpr1 = COPY [[UV1]](s32)
+    ; GFX9-CONTRACT: [[COPY7:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY6]]
+    ; GFX9-CONTRACT: S_SETPC_B64_return [[COPY7]], implicit $vgpr0, implicit $vgpr1
+    ; GFX9-DENORM-LABEL: name: test_double_add_mul
+    ; GFX9-DENORM: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX9-DENORM: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX9-DENORM: [[MV:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY]](s32), [[COPY1]](s32)
+    ; GFX9-DENORM: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX9-DENORM: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3
+    ; GFX9-DENORM: [[MV1:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY2]](s32), [[COPY3]](s32)
+    ; GFX9-DENORM: [[COPY4:%[0-9]+]]:_(s32) = COPY $vgpr4
+    ; GFX9-DENORM: [[COPY5:%[0-9]+]]:_(s32) = COPY $vgpr5
+    ; GFX9-DENORM: [[MV2:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY4]](s32), [[COPY5]](s32)
+    ; GFX9-DENORM: [[COPY6:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX9-DENORM: [[FMUL:%[0-9]+]]:_(s64) = G_FMUL [[MV]], [[MV1]]
+    ; GFX9-DENORM: [[FADD:%[0-9]+]]:_(s64) = G_FADD [[FMUL]], [[MV2]]
+    ; GFX9-DENORM: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FADD]](s64)
+    ; GFX9-DENORM: $vgpr0 = COPY [[UV]](s32)
+    ; GFX9-DENORM: $vgpr1 = COPY [[UV1]](s32)
+    ; GFX9-DENORM: [[COPY7:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY6]]
+    ; GFX9-DENORM: S_SETPC_B64_return [[COPY7]], implicit $vgpr0, implicit $vgpr1
+    ; GFX9-UNSAFE-LABEL: name: test_double_add_mul
+    ; GFX9-UNSAFE: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX9-UNSAFE: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX9-UNSAFE: [[MV:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY]](s32), [[COPY1]](s32)
+    ; GFX9-UNSAFE: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX9-UNSAFE: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3
+    ; GFX9-UNSAFE: [[MV1:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY2]](s32), [[COPY3]](s32)
+    ; GFX9-UNSAFE: [[COPY4:%[0-9]+]]:_(s32) = COPY $vgpr4
+    ; GFX9-UNSAFE: [[COPY5:%[0-9]+]]:_(s32) = COPY $vgpr5
+    ; GFX9-UNSAFE: [[MV2:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY4]](s32), [[COPY5]](s32)
+    ; GFX9-UNSAFE: [[COPY6:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX9-UNSAFE: [[FMA:%[0-9]+]]:_(s64) = G_FMA [[MV]], [[MV1]], [[MV2]]
+    ; GFX9-UNSAFE: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FMA]](s64)
+    ; GFX9-UNSAFE: $vgpr0 = COPY [[UV]](s32)
+    ; GFX9-UNSAFE: $vgpr1 = COPY [[UV1]](s32)
+    ; GFX9-UNSAFE: [[COPY7:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY6]]
+    ; GFX9-UNSAFE: S_SETPC_B64_return [[COPY7]], implicit $vgpr0, implicit $vgpr1
+    ; GFX10-LABEL: name: test_double_add_mul
+    ; GFX10: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX10: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX10: [[MV:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY]](s32), [[COPY1]](s32)
+    ; GFX10: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX10: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3
+    ; GFX10: [[MV1:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY2]](s32), [[COPY3]](s32)
+    ; GFX10: [[COPY4:%[0-9]+]]:_(s32) = COPY $vgpr4
+    ; GFX10: [[COPY5:%[0-9]+]]:_(s32) = COPY $vgpr5
+    ; GFX10: [[MV2:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY4]](s32), [[COPY5]](s32)
+    ; GFX10: [[COPY6:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX10: [[FMUL:%[0-9]+]]:_(s64) = G_FMUL [[MV]], [[MV1]]
+    ; GFX10: [[FADD:%[0-9]+]]:_(s64) = G_FADD [[FMUL]], [[MV2]]
+    ; GFX10: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FADD]](s64)
+    ; GFX10: $vgpr0 = COPY [[UV]](s32)
+    ; GFX10: $vgpr1 = COPY [[UV1]](s32)
+    ; GFX10: [[COPY7:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY6]]
+    ; GFX10: S_SETPC_B64_return [[COPY7]], implicit $vgpr0, implicit $vgpr1
+    ; GFX10-CONTRACT-LABEL: name: test_double_add_mul
+    ; GFX10-CONTRACT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX10-CONTRACT: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX10-CONTRACT: [[MV:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY]](s32), [[COPY1]](s32)
+    ; GFX10-CONTRACT: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX10-CONTRACT: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3
+    ; GFX10-CONTRACT: [[MV1:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY2]](s32), [[COPY3]](s32)
+    ; GFX10-CONTRACT: [[COPY4:%[0-9]+]]:_(s32) = COPY $vgpr4
+    ; GFX10-CONTRACT: [[COPY5:%[0-9]+]]:_(s32) = COPY $vgpr5
+    ; GFX10-CONTRACT: [[MV2:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY4]](s32), [[COPY5]](s32)
+    ; GFX10-CONTRACT: [[COPY6:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX10-CONTRACT: [[FMA:%[0-9]+]]:_(s64) = G_FMA [[MV]], [[MV1]], [[MV2]]
+    ; GFX10-CONTRACT: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FMA]](s64)
+    ; GFX10-CONTRACT: $vgpr0 = COPY [[UV]](s32)
+    ; GFX10-CONTRACT: $vgpr1 = COPY [[UV1]](s32)
+    ; GFX10-CONTRACT: [[COPY7:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY6]]
+    ; GFX10-CONTRACT: S_SETPC_B64_return [[COPY7]], implicit $vgpr0, implicit $vgpr1
+    ; GFX10-DENORM-LABEL: name: test_double_add_mul
+    ; GFX10-DENORM: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX10-DENORM: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX10-DENORM: [[MV:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY]](s32), [[COPY1]](s32)
+    ; GFX10-DENORM: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX10-DENORM: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3
+    ; GFX10-DENORM: [[MV1:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY2]](s32), [[COPY3]](s32)
+    ; GFX10-DENORM: [[COPY4:%[0-9]+]]:_(s32) = COPY $vgpr4
+    ; GFX10-DENORM: [[COPY5:%[0-9]+]]:_(s32) = COPY $vgpr5
+    ; GFX10-DENORM: [[MV2:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY4]](s32), [[COPY5]](s32)
+    ; GFX10-DENORM: [[COPY6:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX10-DENORM: [[FMUL:%[0-9]+]]:_(s64) = G_FMUL [[MV]], [[MV1]]
+    ; GFX10-DENORM: [[FADD:%[0-9]+]]:_(s64) = G_FADD [[FMUL]], [[MV2]]
+    ; GFX10-DENORM: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FADD]](s64)
+    ; GFX10-DENORM: $vgpr0 = COPY [[UV]](s32)
+    ; GFX10-DENORM: $vgpr1 = COPY [[UV1]](s32)
+    ; GFX10-DENORM: [[COPY7:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY6]]
+    ; GFX10-DENORM: S_SETPC_B64_return [[COPY7]], implicit $vgpr0, implicit $vgpr1
+    ; GFX10-UNSAFE-LABEL: name: test_double_add_mul
+    ; GFX10-UNSAFE: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX10-UNSAFE: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX10-UNSAFE: [[MV:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY]](s32), [[COPY1]](s32)
+    ; GFX10-UNSAFE: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX10-UNSAFE: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3
+    ; GFX10-UNSAFE: [[MV1:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY2]](s32), [[COPY3]](s32)
+    ; GFX10-UNSAFE: [[COPY4:%[0-9]+]]:_(s32) = COPY $vgpr4
+    ; GFX10-UNSAFE: [[COPY5:%[0-9]+]]:_(s32) = COPY $vgpr5
+    ; GFX10-UNSAFE: [[MV2:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY4]](s32), [[COPY5]](s32)
+    ; GFX10-UNSAFE: [[COPY6:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX10-UNSAFE: [[FMA:%[0-9]+]]:_(s64) = G_FMA [[MV]], [[MV1]], [[MV2]]
+    ; GFX10-UNSAFE: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FMA]](s64)
+    ; GFX10-UNSAFE: $vgpr0 = COPY [[UV]](s32)
+    ; GFX10-UNSAFE: $vgpr1 = COPY [[UV1]](s32)
+    ; GFX10-UNSAFE: [[COPY7:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY6]]
+    ; GFX10-UNSAFE: S_SETPC_B64_return [[COPY7]], implicit $vgpr0, implicit $vgpr1
+    %4:_(s32) = COPY $vgpr0
+    %5:_(s32) = COPY $vgpr1
+    %0:_(s64) = G_MERGE_VALUES %4(s32), %5(s32)
+    %6:_(s32) = COPY $vgpr2
+    %7:_(s32) = COPY $vgpr3
+    %1:_(s64) = G_MERGE_VALUES %6(s32), %7(s32)
+    %8:_(s32) = COPY $vgpr4
+    %9:_(s32) = COPY $vgpr5
+    %2:_(s64) = G_MERGE_VALUES %8(s32), %9(s32)
+    %3:sgpr_64 = COPY $sgpr30_sgpr31
+    %10:_(s64) = G_FMUL %0, %1
+    %11:_(s64) = G_FADD %10, %2
+    %13:_(s32), %14:_(s32) = G_UNMERGE_VALUES %11(s64)
+    $vgpr0 = COPY %13(s32)
+    $vgpr1 = COPY %14(s32)
+    %12:ccr_sgpr_64 = COPY %3
+    S_SETPC_B64_return %12, implicit $vgpr0, implicit $vgpr1
+...
+
+---
+name:            test_double_add_mul_rhs
+body:             |
+  bb.1.entry:
+    liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $sgpr30_sgpr31
+
+    ; GFX9-LABEL: name: test_double_add_mul_rhs
+    ; GFX9: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX9: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX9: [[MV:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY]](s32), [[COPY1]](s32)
+    ; GFX9: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX9: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3
+    ; GFX9: [[MV1:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY2]](s32), [[COPY3]](s32)
+    ; GFX9: [[COPY4:%[0-9]+]]:_(s32) = COPY $vgpr4
+    ; GFX9: [[COPY5:%[0-9]+]]:_(s32) = COPY $vgpr5
+    ; GFX9: [[MV2:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY4]](s32), [[COPY5]](s32)
+    ; GFX9: [[COPY6:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX9: [[FMUL:%[0-9]+]]:_(s64) = G_FMUL [[MV]], [[MV1]]
+    ; GFX9: [[FADD:%[0-9]+]]:_(s64) = G_FADD [[MV2]], [[FMUL]]
+    ; GFX9: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FADD]](s64)
+    ; GFX9: $vgpr0 = COPY [[UV]](s32)
+    ; GFX9: $vgpr1 = COPY [[UV1]](s32)
+    ; GFX9: [[COPY7:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY6]]
+    ; GFX9: S_SETPC_B64_return [[COPY7]], implicit $vgpr0, implicit $vgpr1
+    ; GFX9-CONTRACT-LABEL: name: test_double_add_mul_rhs
+    ; GFX9-CONTRACT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX9-CONTRACT: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX9-CONTRACT: [[MV:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY]](s32), [[COPY1]](s32)
+    ; GFX9-CONTRACT: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX9-CONTRACT: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3
+    ; GFX9-CONTRACT: [[MV1:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY2]](s32), [[COPY3]](s32)
+    ; GFX9-CONTRACT: [[COPY4:%[0-9]+]]:_(s32) = COPY $vgpr4
+    ; GFX9-CONTRACT: [[COPY5:%[0-9]+]]:_(s32) = COPY $vgpr5
+    ; GFX9-CONTRACT: [[MV2:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY4]](s32), [[COPY5]](s32)
+    ; GFX9-CONTRACT: [[COPY6:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX9-CONTRACT: [[FMA:%[0-9]+]]:_(s64) = G_FMA [[MV]], [[MV1]], [[MV2]]
+    ; GFX9-CONTRACT: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FMA]](s64)
+    ; GFX9-CONTRACT: $vgpr0 = COPY [[UV]](s32)
+    ; GFX9-CONTRACT: $vgpr1 = COPY [[UV1]](s32)
+    ; GFX9-CONTRACT: [[COPY7:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY6]]
+    ; GFX9-CONTRACT: S_SETPC_B64_return [[COPY7]], implicit $vgpr0, implicit $vgpr1
+    ; GFX9-DENORM-LABEL: name: test_double_add_mul_rhs
+    ; GFX9-DENORM: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX9-DENORM: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX9-DENORM: [[MV:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY]](s32), [[COPY1]](s32)
+    ; GFX9-DENORM: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX9-DENORM: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3
+    ; GFX9-DENORM: [[MV1:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY2]](s32), [[COPY3]](s32)
+    ; GFX9-DENORM: [[COPY4:%[0-9]+]]:_(s32) = COPY $vgpr4
+    ; GFX9-DENORM: [[COPY5:%[0-9]+]]:_(s32) = COPY $vgpr5
+    ; GFX9-DENORM: [[MV2:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY4]](s32), [[COPY5]](s32)
+    ; GFX9-DENORM: [[COPY6:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX9-DENORM: [[FMUL:%[0-9]+]]:_(s64) = G_FMUL [[MV]], [[MV1]]
+    ; GFX9-DENORM: [[FADD:%[0-9]+]]:_(s64) = G_FADD [[MV2]], [[FMUL]]
+    ; GFX9-DENORM: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FADD]](s64)
+    ; GFX9-DENORM: $vgpr0 = COPY [[UV]](s32)
+    ; GFX9-DENORM: $vgpr1 = COPY [[UV1]](s32)
+    ; GFX9-DENORM: [[COPY7:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY6]]
+    ; GFX9-DENORM: S_SETPC_B64_return [[COPY7]], implicit $vgpr0, implicit $vgpr1
+    ; GFX9-UNSAFE-LABEL: name: test_double_add_mul_rhs
+    ; GFX9-UNSAFE: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX9-UNSAFE: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX9-UNSAFE: [[MV:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY]](s32), [[COPY1]](s32)
+    ; GFX9-UNSAFE: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX9-UNSAFE: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3
+    ; GFX9-UNSAFE: [[MV1:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY2]](s32), [[COPY3]](s32)
+    ; GFX9-UNSAFE: [[COPY4:%[0-9]+]]:_(s32) = COPY $vgpr4
+    ; GFX9-UNSAFE: [[COPY5:%[0-9]+]]:_(s32) = COPY $vgpr5
+    ; GFX9-UNSAFE: [[MV2:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY4]](s32), [[COPY5]](s32)
+    ; GFX9-UNSAFE: [[COPY6:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX9-UNSAFE: [[FMA:%[0-9]+]]:_(s64) = G_FMA [[MV]], [[MV1]], [[MV2]]
+    ; GFX9-UNSAFE: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FMA]](s64)
+    ; GFX9-UNSAFE: $vgpr0 = COPY [[UV]](s32)
+    ; GFX9-UNSAFE: $vgpr1 = COPY [[UV1]](s32)
+    ; GFX9-UNSAFE: [[COPY7:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY6]]
+    ; GFX9-UNSAFE: S_SETPC_B64_return [[COPY7]], implicit $vgpr0, implicit $vgpr1
+    ; GFX10-LABEL: name: test_double_add_mul_rhs
+    ; GFX10: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX10: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX10: [[MV:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY]](s32), [[COPY1]](s32)
+    ; GFX10: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX10: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3
+    ; GFX10: [[MV1:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY2]](s32), [[COPY3]](s32)
+    ; GFX10: [[COPY4:%[0-9]+]]:_(s32) = COPY $vgpr4
+    ; GFX10: [[COPY5:%[0-9]+]]:_(s32) = COPY $vgpr5
+    ; GFX10: [[MV2:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY4]](s32), [[COPY5]](s32)
+    ; GFX10: [[COPY6:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX10: [[FMUL:%[0-9]+]]:_(s64) = G_FMUL [[MV]], [[MV1]]
+    ; GFX10: [[FADD:%[0-9]+]]:_(s64) = G_FADD [[MV2]], [[FMUL]]
+    ; GFX10: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FADD]](s64)
+    ; GFX10: $vgpr0 = COPY [[UV]](s32)
+    ; GFX10: $vgpr1 = COPY [[UV1]](s32)
+    ; GFX10: [[COPY7:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY6]]
+    ; GFX10: S_SETPC_B64_return [[COPY7]], implicit $vgpr0, implicit $vgpr1
+    ; GFX10-CONTRACT-LABEL: name: test_double_add_mul_rhs
+    ; GFX10-CONTRACT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX10-CONTRACT: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX10-CONTRACT: [[MV:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY]](s32), [[COPY1]](s32)
+    ; GFX10-CONTRACT: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX10-CONTRACT: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3
+    ; GFX10-CONTRACT: [[MV1:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY2]](s32), [[COPY3]](s32)
+    ; GFX10-CONTRACT: [[COPY4:%[0-9]+]]:_(s32) = COPY $vgpr4
+    ; GFX10-CONTRACT: [[COPY5:%[0-9]+]]:_(s32) = COPY $vgpr5
+    ; GFX10-CONTRACT: [[MV2:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY4]](s32), [[COPY5]](s32)
+    ; GFX10-CONTRACT: [[COPY6:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX10-CONTRACT: [[FMA:%[0-9]+]]:_(s64) = G_FMA [[MV]], [[MV1]], [[MV2]]
+    ; GFX10-CONTRACT: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FMA]](s64)
+    ; GFX10-CONTRACT: $vgpr0 = COPY [[UV]](s32)
+    ; GFX10-CONTRACT: $vgpr1 = COPY [[UV1]](s32)
+    ; GFX10-CONTRACT: [[COPY7:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY6]]
+    ; GFX10-CONTRACT: S_SETPC_B64_return [[COPY7]], implicit $vgpr0, implicit $vgpr1
+    ; GFX10-DENORM-LABEL: name: test_double_add_mul_rhs
+    ; GFX10-DENORM: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX10-DENORM: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX10-DENORM: [[MV:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY]](s32), [[COPY1]](s32)
+    ; GFX10-DENORM: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX10-DENORM: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3
+    ; GFX10-DENORM: [[MV1:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY2]](s32), [[COPY3]](s32)
+    ; GFX10-DENORM: [[COPY4:%[0-9]+]]:_(s32) = COPY $vgpr4
+    ; GFX10-DENORM: [[COPY5:%[0-9]+]]:_(s32) = COPY $vgpr5
+    ; GFX10-DENORM: [[MV2:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY4]](s32), [[COPY5]](s32)
+    ; GFX10-DENORM: [[COPY6:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX10-DENORM: [[FMUL:%[0-9]+]]:_(s64) = G_FMUL [[MV]], [[MV1]]
+    ; GFX10-DENORM: [[FADD:%[0-9]+]]:_(s64) = G_FADD [[MV2]], [[FMUL]]
+    ; GFX10-DENORM: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FADD]](s64)
+    ; GFX10-DENORM: $vgpr0 = COPY [[UV]](s32)
+    ; GFX10-DENORM: $vgpr1 = COPY [[UV1]](s32)
+    ; GFX10-DENORM: [[COPY7:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY6]]
+    ; GFX10-DENORM: S_SETPC_B64_return [[COPY7]], implicit $vgpr0, implicit $vgpr1
+    ; GFX10-UNSAFE-LABEL: name: test_double_add_mul_rhs
+    ; GFX10-UNSAFE: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX10-UNSAFE: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX10-UNSAFE: [[MV:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY]](s32), [[COPY1]](s32)
+    ; GFX10-UNSAFE: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX10-UNSAFE: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3
+    ; GFX10-UNSAFE: [[MV1:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY2]](s32), [[COPY3]](s32)
+    ; GFX10-UNSAFE: [[COPY4:%[0-9]+]]:_(s32) = COPY $vgpr4
+    ; GFX10-UNSAFE: [[COPY5:%[0-9]+]]:_(s32) = COPY $vgpr5
+    ; GFX10-UNSAFE: [[MV2:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY4]](s32), [[COPY5]](s32)
+    ; GFX10-UNSAFE: [[COPY6:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX10-UNSAFE: [[FMA:%[0-9]+]]:_(s64) = G_FMA [[MV]], [[MV1]], [[MV2]]
+    ; GFX10-UNSAFE: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FMA]](s64)
+    ; GFX10-UNSAFE: $vgpr0 = COPY [[UV]](s32)
+    ; GFX10-UNSAFE: $vgpr1 = COPY [[UV1]](s32)
+    ; GFX10-UNSAFE: [[COPY7:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY6]]
+    ; GFX10-UNSAFE: S_SETPC_B64_return [[COPY7]], implicit $vgpr0, implicit $vgpr1
+    %4:_(s32) = COPY $vgpr0
+    %5:_(s32) = COPY $vgpr1
+    %0:_(s64) = G_MERGE_VALUES %4(s32), %5(s32)
+    %6:_(s32) = COPY $vgpr2
+    %7:_(s32) = COPY $vgpr3
+    %1:_(s64) = G_MERGE_VALUES %6(s32), %7(s32)
+    %8:_(s32) = COPY $vgpr4
+    %9:_(s32) = COPY $vgpr5
+    %2:_(s64) = G_MERGE_VALUES %8(s32), %9(s32)
+    %3:sgpr_64 = COPY $sgpr30_sgpr31
+    %10:_(s64) = G_FMUL %0, %1
+    %11:_(s64) = G_FADD %2, %10
+    %13:_(s32), %14:_(s32) = G_UNMERGE_VALUES %11(s64)
+    $vgpr0 = COPY %13(s32)
+    $vgpr1 = COPY %14(s32)
+    %12:ccr_sgpr_64 = COPY %3
+    S_SETPC_B64_return %12, implicit $vgpr0, implicit $vgpr1
+...
+
+---
+name:            test_4xfloat_add_mul
+body:             |
+  bb.1.entry:
+    liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8, $vgpr9, $vgpr10, $vgpr11, $sgpr30_sgpr31
+
+    ; GFX9-LABEL: name: test_4xfloat_add_mul
+    ; GFX9: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX9: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX9: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX9: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3
+    ; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32)
+    ; GFX9: [[COPY4:%[0-9]+]]:_(s32) = COPY $vgpr4
+    ; GFX9: [[COPY5:%[0-9]+]]:_(s32) = COPY $vgpr5
+    ; GFX9: [[COPY6:%[0-9]+]]:_(s32) = COPY $vgpr6
+    ; GFX9: [[COPY7:%[0-9]+]]:_(s32) = COPY $vgpr7
+    ; GFX9: [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+    ; GFX9: [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr8
+    ; GFX9: [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr9
+    ; GFX9: [[COPY10:%[0-9]+]]:_(s32) = COPY $vgpr10
+    ; GFX9: [[COPY11:%[0-9]+]]:_(s32) = COPY $vgpr11
+    ; GFX9: [[BUILD_VECTOR2:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+    ; GFX9: [[COPY12:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX9: [[FMUL:%[0-9]+]]:_(<4 x s32>) = G_FMUL [[BUILD_VECTOR]], [[BUILD_VECTOR1]]
+    ; GFX9: [[FADD:%[0-9]+]]:_(<4 x s32>) = G_FADD [[FMUL]], [[BUILD_VECTOR2]]
+    ; GFX9: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FADD]](<4 x s32>)
+    ; GFX9: $vgpr0 = COPY [[UV]](s32)
+    ; GFX9: $vgpr1 = COPY [[UV1]](s32)
+    ; GFX9: $vgpr2 = COPY [[UV2]](s32)
+    ; GFX9: $vgpr3 = COPY [[UV3]](s32)
+    ; GFX9: [[COPY13:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY12]]
+    ; GFX9: S_SETPC_B64_return [[COPY13]], implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+    ; GFX9-CONTRACT-LABEL: name: test_4xfloat_add_mul
+    ; GFX9-CONTRACT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX9-CONTRACT: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX9-CONTRACT: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX9-CONTRACT: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3
+    ; GFX9-CONTRACT: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32)
+    ; GFX9-CONTRACT: [[COPY4:%[0-9]+]]:_(s32) = COPY $vgpr4
+    ; GFX9-CONTRACT: [[COPY5:%[0-9]+]]:_(s32) = COPY $vgpr5
+    ; GFX9-CONTRACT: [[COPY6:%[0-9]+]]:_(s32) = COPY $vgpr6
+    ; GFX9-CONTRACT: [[COPY7:%[0-9]+]]:_(s32) = COPY $vgpr7
+    ; GFX9-CONTRACT: [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+    ; GFX9-CONTRACT: [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr8
+    ; GFX9-CONTRACT: [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr9
+    ; GFX9-CONTRACT: [[COPY10:%[0-9]+]]:_(s32) = COPY $vgpr10
+    ; GFX9-CONTRACT: [[COPY11:%[0-9]+]]:_(s32) = COPY $vgpr11
+    ; GFX9-CONTRACT: [[BUILD_VECTOR2:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+    ; GFX9-CONTRACT: [[COPY12:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX9-CONTRACT: [[FMUL:%[0-9]+]]:_(<4 x s32>) = G_FMUL [[BUILD_VECTOR]], [[BUILD_VECTOR1]]
+    ; GFX9-CONTRACT: [[FADD:%[0-9]+]]:_(<4 x s32>) = G_FADD [[FMUL]], [[BUILD_VECTOR2]]
+    ; GFX9-CONTRACT: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FADD]](<4 x s32>)
+    ; GFX9-CONTRACT: $vgpr0 = COPY [[UV]](s32)
+    ; GFX9-CONTRACT: $vgpr1 = COPY [[UV1]](s32)
+    ; GFX9-CONTRACT: $vgpr2 = COPY [[UV2]](s32)
+    ; GFX9-CONTRACT: $vgpr3 = COPY [[UV3]](s32)
+    ; GFX9-CONTRACT: [[COPY13:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY12]]
+    ; GFX9-CONTRACT: S_SETPC_B64_return [[COPY13]], implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+    ; GFX9-DENORM-LABEL: name: test_4xfloat_add_mul
+    ; GFX9-DENORM: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX9-DENORM: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX9-DENORM: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX9-DENORM: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3
+    ; GFX9-DENORM: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32)
+    ; GFX9-DENORM: [[COPY4:%[0-9]+]]:_(s32) = COPY $vgpr4
+    ; GFX9-DENORM: [[COPY5:%[0-9]+]]:_(s32) = COPY $vgpr5
+    ; GFX9-DENORM: [[COPY6:%[0-9]+]]:_(s32) = COPY $vgpr6
+    ; GFX9-DENORM: [[COPY7:%[0-9]+]]:_(s32) = COPY $vgpr7
+    ; GFX9-DENORM: [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+    ; GFX9-DENORM: [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr8
+    ; GFX9-DENORM: [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr9
+    ; GFX9-DENORM: [[COPY10:%[0-9]+]]:_(s32) = COPY $vgpr10
+    ; GFX9-DENORM: [[COPY11:%[0-9]+]]:_(s32) = COPY $vgpr11
+    ; GFX9-DENORM: [[BUILD_VECTOR2:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+    ; GFX9-DENORM: [[COPY12:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX9-DENORM: [[FMUL:%[0-9]+]]:_(<4 x s32>) = G_FMUL [[BUILD_VECTOR]], [[BUILD_VECTOR1]]
+    ; GFX9-DENORM: [[FADD:%[0-9]+]]:_(<4 x s32>) = G_FADD [[FMUL]], [[BUILD_VECTOR2]]
+    ; GFX9-DENORM: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FADD]](<4 x s32>)
+    ; GFX9-DENORM: $vgpr0 = COPY [[UV]](s32)
+    ; GFX9-DENORM: $vgpr1 = COPY [[UV1]](s32)
+    ; GFX9-DENORM: $vgpr2 = COPY [[UV2]](s32)
+    ; GFX9-DENORM: $vgpr3 = COPY [[UV3]](s32)
+    ; GFX9-DENORM: [[COPY13:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY12]]
+    ; GFX9-DENORM: S_SETPC_B64_return [[COPY13]], implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+    ; GFX9-UNSAFE-LABEL: name: test_4xfloat_add_mul
+    ; GFX9-UNSAFE: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX9-UNSAFE: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX9-UNSAFE: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX9-UNSAFE: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3
+    ; GFX9-UNSAFE: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32)
+    ; GFX9-UNSAFE: [[COPY4:%[0-9]+]]:_(s32) = COPY $vgpr4
+    ; GFX9-UNSAFE: [[COPY5:%[0-9]+]]:_(s32) = COPY $vgpr5
+    ; GFX9-UNSAFE: [[COPY6:%[0-9]+]]:_(s32) = COPY $vgpr6
+    ; GFX9-UNSAFE: [[COPY7:%[0-9]+]]:_(s32) = COPY $vgpr7
+    ; GFX9-UNSAFE: [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+    ; GFX9-UNSAFE: [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr8
+    ; GFX9-UNSAFE: [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr9
+    ; GFX9-UNSAFE: [[COPY10:%[0-9]+]]:_(s32) = COPY $vgpr10
+    ; GFX9-UNSAFE: [[COPY11:%[0-9]+]]:_(s32) = COPY $vgpr11
+    ; GFX9-UNSAFE: [[BUILD_VECTOR2:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+    ; GFX9-UNSAFE: [[COPY12:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX9-UNSAFE: [[FMUL:%[0-9]+]]:_(<4 x s32>) = G_FMUL [[BUILD_VECTOR]], [[BUILD_VECTOR1]]
+    ; GFX9-UNSAFE: [[FADD:%[0-9]+]]:_(<4 x s32>) = G_FADD [[FMUL]], [[BUILD_VECTOR2]]
+    ; GFX9-UNSAFE: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FADD]](<4 x s32>)
+    ; GFX9-UNSAFE: $vgpr0 = COPY [[UV]](s32)
+    ; GFX9-UNSAFE: $vgpr1 = COPY [[UV1]](s32)
+    ; GFX9-UNSAFE: $vgpr2 = COPY [[UV2]](s32)
+    ; GFX9-UNSAFE: $vgpr3 = COPY [[UV3]](s32)
+    ; GFX9-UNSAFE: [[COPY13:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY12]]
+    ; GFX9-UNSAFE: S_SETPC_B64_return [[COPY13]], implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+    ; GFX10-LABEL: name: test_4xfloat_add_mul
+    ; GFX10: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX10: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX10: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX10: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3
+    ; GFX10: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32)
+    ; GFX10: [[COPY4:%[0-9]+]]:_(s32) = COPY $vgpr4
+    ; GFX10: [[COPY5:%[0-9]+]]:_(s32) = COPY $vgpr5
+    ; GFX10: [[COPY6:%[0-9]+]]:_(s32) = COPY $vgpr6
+    ; GFX10: [[COPY7:%[0-9]+]]:_(s32) = COPY $vgpr7
+    ; GFX10: [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+    ; GFX10: [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr8
+    ; GFX10: [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr9
+    ; GFX10: [[COPY10:%[0-9]+]]:_(s32) = COPY $vgpr10
+    ; GFX10: [[COPY11:%[0-9]+]]:_(s32) = COPY $vgpr11
+    ; GFX10: [[BUILD_VECTOR2:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+    ; GFX10: [[COPY12:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX10: [[FMUL:%[0-9]+]]:_(<4 x s32>) = G_FMUL [[BUILD_VECTOR]], [[BUILD_VECTOR1]]
+    ; GFX10: [[FADD:%[0-9]+]]:_(<4 x s32>) = G_FADD [[FMUL]], [[BUILD_VECTOR2]]
+    ; GFX10: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FADD]](<4 x s32>)
+    ; GFX10: $vgpr0 = COPY [[UV]](s32)
+    ; GFX10: $vgpr1 = COPY [[UV1]](s32)
+    ; GFX10: $vgpr2 = COPY [[UV2]](s32)
+    ; GFX10: $vgpr3 = COPY [[UV3]](s32)
+    ; GFX10: [[COPY13:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY12]]
+    ; GFX10: S_SETPC_B64_return [[COPY13]], implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+    ; GFX10-CONTRACT-LABEL: name: test_4xfloat_add_mul
+    ; GFX10-CONTRACT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX10-CONTRACT: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX10-CONTRACT: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX10-CONTRACT: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3
+    ; GFX10-CONTRACT: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32)
+    ; GFX10-CONTRACT: [[COPY4:%[0-9]+]]:_(s32) = COPY $vgpr4
+    ; GFX10-CONTRACT: [[COPY5:%[0-9]+]]:_(s32) = COPY $vgpr5
+    ; GFX10-CONTRACT: [[COPY6:%[0-9]+]]:_(s32) = COPY $vgpr6
+    ; GFX10-CONTRACT: [[COPY7:%[0-9]+]]:_(s32) = COPY $vgpr7
+    ; GFX10-CONTRACT: [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+    ; GFX10-CONTRACT: [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr8
+    ; GFX10-CONTRACT: [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr9
+    ; GFX10-CONTRACT: [[COPY10:%[0-9]+]]:_(s32) = COPY $vgpr10
+    ; GFX10-CONTRACT: [[COPY11:%[0-9]+]]:_(s32) = COPY $vgpr11
+    ; GFX10-CONTRACT: [[BUILD_VECTOR2:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+    ; GFX10-CONTRACT: [[COPY12:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX10-CONTRACT: [[FMUL:%[0-9]+]]:_(<4 x s32>) = G_FMUL [[BUILD_VECTOR]], [[BUILD_VECTOR1]]
+    ; GFX10-CONTRACT: [[FADD:%[0-9]+]]:_(<4 x s32>) = G_FADD [[FMUL]], [[BUILD_VECTOR2]]
+    ; GFX10-CONTRACT: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FADD]](<4 x s32>)
+    ; GFX10-CONTRACT: $vgpr0 = COPY [[UV]](s32)
+    ; GFX10-CONTRACT: $vgpr1 = COPY [[UV1]](s32)
+    ; GFX10-CONTRACT: $vgpr2 = COPY [[UV2]](s32)
+    ; GFX10-CONTRACT: $vgpr3 = COPY [[UV3]](s32)
+    ; GFX10-CONTRACT: [[COPY13:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY12]]
+    ; GFX10-CONTRACT: S_SETPC_B64_return [[COPY13]], implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+    ; GFX10-DENORM-LABEL: name: test_4xfloat_add_mul
+    ; GFX10-DENORM: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX10-DENORM: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX10-DENORM: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX10-DENORM: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3
+    ; GFX10-DENORM: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32)
+    ; GFX10-DENORM: [[COPY4:%[0-9]+]]:_(s32) = COPY $vgpr4
+    ; GFX10-DENORM: [[COPY5:%[0-9]+]]:_(s32) = COPY $vgpr5
+    ; GFX10-DENORM: [[COPY6:%[0-9]+]]:_(s32) = COPY $vgpr6
+    ; GFX10-DENORM: [[COPY7:%[0-9]+]]:_(s32) = COPY $vgpr7
+    ; GFX10-DENORM: [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+    ; GFX10-DENORM: [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr8
+    ; GFX10-DENORM: [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr9
+    ; GFX10-DENORM: [[COPY10:%[0-9]+]]:_(s32) = COPY $vgpr10
+    ; GFX10-DENORM: [[COPY11:%[0-9]+]]:_(s32) = COPY $vgpr11
+    ; GFX10-DENORM: [[BUILD_VECTOR2:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+    ; GFX10-DENORM: [[COPY12:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX10-DENORM: [[FMUL:%[0-9]+]]:_(<4 x s32>) = G_FMUL [[BUILD_VECTOR]], [[BUILD_VECTOR1]]
+    ; GFX10-DENORM: [[FADD:%[0-9]+]]:_(<4 x s32>) = G_FADD [[FMUL]], [[BUILD_VECTOR2]]
+    ; GFX10-DENORM: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FADD]](<4 x s32>)
+    ; GFX10-DENORM: $vgpr0 = COPY [[UV]](s32)
+    ; GFX10-DENORM: $vgpr1 = COPY [[UV1]](s32)
+    ; GFX10-DENORM: $vgpr2 = COPY [[UV2]](s32)
+    ; GFX10-DENORM: $vgpr3 = COPY [[UV3]](s32)
+    ; GFX10-DENORM: [[COPY13:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY12]]
+    ; GFX10-DENORM: S_SETPC_B64_return [[COPY13]], implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+    ; GFX10-UNSAFE-LABEL: name: test_4xfloat_add_mul
+    ; GFX10-UNSAFE: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX10-UNSAFE: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX10-UNSAFE: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX10-UNSAFE: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3
+    ; GFX10-UNSAFE: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32)
+    ; GFX10-UNSAFE: [[COPY4:%[0-9]+]]:_(s32) = COPY $vgpr4
+    ; GFX10-UNSAFE: [[COPY5:%[0-9]+]]:_(s32) = COPY $vgpr5
+    ; GFX10-UNSAFE: [[COPY6:%[0-9]+]]:_(s32) = COPY $vgpr6
+    ; GFX10-UNSAFE: [[COPY7:%[0-9]+]]:_(s32) = COPY $vgpr7
+    ; GFX10-UNSAFE: [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+    ; GFX10-UNSAFE: [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr8
+    ; GFX10-UNSAFE: [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr9
+    ; GFX10-UNSAFE: [[COPY10:%[0-9]+]]:_(s32) = COPY $vgpr10
+    ; GFX10-UNSAFE: [[COPY11:%[0-9]+]]:_(s32) = COPY $vgpr11
+    ; GFX10-UNSAFE: [[BUILD_VECTOR2:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+    ; GFX10-UNSAFE: [[COPY12:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX10-UNSAFE: [[FMUL:%[0-9]+]]:_(<4 x s32>) = G_FMUL [[BUILD_VECTOR]], [[BUILD_VECTOR1]]
+    ; GFX10-UNSAFE: [[FADD:%[0-9]+]]:_(<4 x s32>) = G_FADD [[FMUL]], [[BUILD_VECTOR2]]
+    ; GFX10-UNSAFE: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FADD]](<4 x s32>)
+    ; GFX10-UNSAFE: $vgpr0 = COPY [[UV]](s32)
+    ; GFX10-UNSAFE: $vgpr1 = COPY [[UV1]](s32)
+    ; GFX10-UNSAFE: $vgpr2 = COPY [[UV2]](s32)
+    ; GFX10-UNSAFE: $vgpr3 = COPY [[UV3]](s32)
+    ; GFX10-UNSAFE: [[COPY13:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY12]]
+    ; GFX10-UNSAFE: S_SETPC_B64_return [[COPY13]], implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+    %4:_(s32) = COPY $vgpr0
+    %5:_(s32) = COPY $vgpr1
+    %6:_(s32) = COPY $vgpr2
+    %7:_(s32) = COPY $vgpr3
+    %0:_(<4 x s32>) = G_BUILD_VECTOR %4(s32), %5(s32), %6(s32), %7(s32)
+    %8:_(s32) = COPY $vgpr4
+    %9:_(s32) = COPY $vgpr5
+    %10:_(s32) = COPY $vgpr6
+    %11:_(s32) = COPY $vgpr7
+    %1:_(<4 x s32>) = G_BUILD_VECTOR %8(s32), %9(s32), %10(s32), %11(s32)
+    %12:_(s32) = COPY $vgpr8
+    %13:_(s32) = COPY $vgpr9
+    %14:_(s32) = COPY $vgpr10
+    %15:_(s32) = COPY $vgpr11
+    %2:_(<4 x s32>) = G_BUILD_VECTOR %12(s32), %13(s32), %14(s32), %15(s32)
+    %3:sgpr_64 = COPY $sgpr30_sgpr31
+    %16:_(<4 x s32>) = G_FMUL %0, %1
+    %17:_(<4 x s32>) = G_FADD %16, %2
+    %19:_(s32), %20:_(s32), %21:_(s32), %22:_(s32) = G_UNMERGE_VALUES %17(<4 x s32>)
+    $vgpr0 = COPY %19(s32)
+    $vgpr1 = COPY %20(s32)
+    $vgpr2 = COPY %21(s32)
+    $vgpr3 = COPY %22(s32)
+    %18:ccr_sgpr_64 = COPY %3
+    S_SETPC_B64_return %18, implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+...
+
+---
+name:            test_3xfloat_add_mul_rhs
+body:             |
+  bb.1.entry:
+    liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8, $sgpr30_sgpr31
+
+    ; GFX9-LABEL: name: test_3xfloat_add_mul_rhs
+    ; GFX9: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX9: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX9: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32)
+    ; GFX9: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3
+    ; GFX9: [[COPY4:%[0-9]+]]:_(s32) = COPY $vgpr4
+    ; GFX9: [[COPY5:%[0-9]+]]:_(s32) = COPY $vgpr5
+    ; GFX9: [[BUILD_VECTOR1:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32)
+    ; GFX9: [[COPY6:%[0-9]+]]:_(s32) = COPY $vgpr6
+    ; GFX9: [[COPY7:%[0-9]+]]:_(s32) = COPY $vgpr7
+    ; GFX9: [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr8
+    ; GFX9: [[BUILD_VECTOR2:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[COPY6]](s32), [[COPY7]](s32), [[COPY8]](s32)
+    ; GFX9: [[COPY9:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX9: [[FMUL:%[0-9]+]]:_(<3 x s32>) = G_FMUL [[BUILD_VECTOR]], [[BUILD_VECTOR1]]
+    ; GFX9: [[FADD:%[0-9]+]]:_(<3 x s32>) = G_FADD [[BUILD_VECTOR2]], [[FMUL]]
+    ; GFX9: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FADD]](<3 x s32>)
+    ; GFX9: $vgpr0 = COPY [[UV]](s32)
+    ; GFX9: $vgpr1 = COPY [[UV1]](s32)
+    ; GFX9: $vgpr2 = COPY [[UV2]](s32)
+    ; GFX9: [[COPY10:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY9]]
+    ; GFX9: S_SETPC_B64_return [[COPY10]], implicit $vgpr0, implicit $vgpr1, implicit $vgpr2
+    ; GFX9-CONTRACT-LABEL: name: test_3xfloat_add_mul_rhs
+    ; GFX9-CONTRACT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX9-CONTRACT: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX9-CONTRACT: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX9-CONTRACT: [[BUILD_VECTOR:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32)
+    ; GFX9-CONTRACT: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3
+    ; GFX9-CONTRACT: [[COPY4:%[0-9]+]]:_(s32) = COPY $vgpr4
+    ; GFX9-CONTRACT: [[COPY5:%[0-9]+]]:_(s32) = COPY $vgpr5
+    ; GFX9-CONTRACT: [[BUILD_VECTOR1:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32)
+    ; GFX9-CONTRACT: [[COPY6:%[0-9]+]]:_(s32) = COPY $vgpr6
+    ; GFX9-CONTRACT: [[COPY7:%[0-9]+]]:_(s32) = COPY $vgpr7
+    ; GFX9-CONTRACT: [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr8
+    ; GFX9-CONTRACT: [[BUILD_VECTOR2:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[COPY6]](s32), [[COPY7]](s32), [[COPY8]](s32)
+    ; GFX9-CONTRACT: [[COPY9:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX9-CONTRACT: [[FMUL:%[0-9]+]]:_(<3 x s32>) = G_FMUL [[BUILD_VECTOR]], [[BUILD_VECTOR1]]
+    ; GFX9-CONTRACT: [[FADD:%[0-9]+]]:_(<3 x s32>) = G_FADD [[BUILD_VECTOR2]], [[FMUL]]
+    ; GFX9-CONTRACT: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FADD]](<3 x s32>)
+    ; GFX9-CONTRACT: $vgpr0 = COPY [[UV]](s32)
+    ; GFX9-CONTRACT: $vgpr1 = COPY [[UV1]](s32)
+    ; GFX9-CONTRACT: $vgpr2 = COPY [[UV2]](s32)
+    ; GFX9-CONTRACT: [[COPY10:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY9]]
+    ; GFX9-CONTRACT: S_SETPC_B64_return [[COPY10]], implicit $vgpr0, implicit $vgpr1, implicit $vgpr2
+    ; GFX9-DENORM-LABEL: name: test_3xfloat_add_mul_rhs
+    ; GFX9-DENORM: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX9-DENORM: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX9-DENORM: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX9-DENORM: [[BUILD_VECTOR:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32)
+    ; GFX9-DENORM: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3
+    ; GFX9-DENORM: [[COPY4:%[0-9]+]]:_(s32) = COPY $vgpr4
+    ; GFX9-DENORM: [[COPY5:%[0-9]+]]:_(s32) = COPY $vgpr5
+    ; GFX9-DENORM: [[BUILD_VECTOR1:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32)
+    ; GFX9-DENORM: [[COPY6:%[0-9]+]]:_(s32) = COPY $vgpr6
+    ; GFX9-DENORM: [[COPY7:%[0-9]+]]:_(s32) = COPY $vgpr7
+    ; GFX9-DENORM: [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr8
+    ; GFX9-DENORM: [[BUILD_VECTOR2:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[COPY6]](s32), [[COPY7]](s32), [[COPY8]](s32)
+    ; GFX9-DENORM: [[COPY9:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX9-DENORM: [[FMUL:%[0-9]+]]:_(<3 x s32>) = G_FMUL [[BUILD_VECTOR]], [[BUILD_VECTOR1]]
+    ; GFX9-DENORM: [[FADD:%[0-9]+]]:_(<3 x s32>) = G_FADD [[BUILD_VECTOR2]], [[FMUL]]
+    ; GFX9-DENORM: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FADD]](<3 x s32>)
+    ; GFX9-DENORM: $vgpr0 = COPY [[UV]](s32)
+    ; GFX9-DENORM: $vgpr1 = COPY [[UV1]](s32)
+    ; GFX9-DENORM: $vgpr2 = COPY [[UV2]](s32)
+    ; GFX9-DENORM: [[COPY10:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY9]]
+    ; GFX9-DENORM: S_SETPC_B64_return [[COPY10]], implicit $vgpr0, implicit $vgpr1, implicit $vgpr2
+    ; GFX9-UNSAFE-LABEL: name: test_3xfloat_add_mul_rhs
+    ; GFX9-UNSAFE: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX9-UNSAFE: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX9-UNSAFE: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX9-UNSAFE: [[BUILD_VECTOR:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32)
+    ; GFX9-UNSAFE: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3
+    ; GFX9-UNSAFE: [[COPY4:%[0-9]+]]:_(s32) = COPY $vgpr4
+    ; GFX9-UNSAFE: [[COPY5:%[0-9]+]]:_(s32) = COPY $vgpr5
+    ; GFX9-UNSAFE: [[BUILD_VECTOR1:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32)
+    ; GFX9-UNSAFE: [[COPY6:%[0-9]+]]:_(s32) = COPY $vgpr6
+    ; GFX9-UNSAFE: [[COPY7:%[0-9]+]]:_(s32) = COPY $vgpr7
+    ; GFX9-UNSAFE: [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr8
+    ; GFX9-UNSAFE: [[BUILD_VECTOR2:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[COPY6]](s32), [[COPY7]](s32), [[COPY8]](s32)
+    ; GFX9-UNSAFE: [[COPY9:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX9-UNSAFE: [[FMUL:%[0-9]+]]:_(<3 x s32>) = G_FMUL [[BUILD_VECTOR]], [[BUILD_VECTOR1]]
+    ; GFX9-UNSAFE: [[FADD:%[0-9]+]]:_(<3 x s32>) = G_FADD [[BUILD_VECTOR2]], [[FMUL]]
+    ; GFX9-UNSAFE: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FADD]](<3 x s32>)
+    ; GFX9-UNSAFE: $vgpr0 = COPY [[UV]](s32)
+    ; GFX9-UNSAFE: $vgpr1 = COPY [[UV1]](s32)
+    ; GFX9-UNSAFE: $vgpr2 = COPY [[UV2]](s32)
+    ; GFX9-UNSAFE: [[COPY10:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY9]]
+    ; GFX9-UNSAFE: S_SETPC_B64_return [[COPY10]], implicit $vgpr0, implicit $vgpr1, implicit $vgpr2
+    ; GFX10-LABEL: name: test_3xfloat_add_mul_rhs
+    ; GFX10: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX10: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX10: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX10: [[BUILD_VECTOR:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32)
+    ; GFX10: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3
+    ; GFX10: [[COPY4:%[0-9]+]]:_(s32) = COPY $vgpr4
+    ; GFX10: [[COPY5:%[0-9]+]]:_(s32) = COPY $vgpr5
+    ; GFX10: [[BUILD_VECTOR1:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32)
+    ; GFX10: [[COPY6:%[0-9]+]]:_(s32) = COPY $vgpr6
+    ; GFX10: [[COPY7:%[0-9]+]]:_(s32) = COPY $vgpr7
+    ; GFX10: [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr8
+    ; GFX10: [[BUILD_VECTOR2:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[COPY6]](s32), [[COPY7]](s32), [[COPY8]](s32)
+    ; GFX10: [[COPY9:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX10: [[FMUL:%[0-9]+]]:_(<3 x s32>) = G_FMUL [[BUILD_VECTOR]], [[BUILD_VECTOR1]]
+    ; GFX10: [[FADD:%[0-9]+]]:_(<3 x s32>) = G_FADD [[BUILD_VECTOR2]], [[FMUL]]
+    ; GFX10: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FADD]](<3 x s32>)
+    ; GFX10: $vgpr0 = COPY [[UV]](s32)
+    ; GFX10: $vgpr1 = COPY [[UV1]](s32)
+    ; GFX10: $vgpr2 = COPY [[UV2]](s32)
+    ; GFX10: [[COPY10:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY9]]
+    ; GFX10: S_SETPC_B64_return [[COPY10]], implicit $vgpr0, implicit $vgpr1, implicit $vgpr2
+    ; GFX10-CONTRACT-LABEL: name: test_3xfloat_add_mul_rhs
+    ; GFX10-CONTRACT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX10-CONTRACT: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX10-CONTRACT: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX10-CONTRACT: [[BUILD_VECTOR:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32)
+    ; GFX10-CONTRACT: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3
+    ; GFX10-CONTRACT: [[COPY4:%[0-9]+]]:_(s32) = COPY $vgpr4
+    ; GFX10-CONTRACT: [[COPY5:%[0-9]+]]:_(s32) = COPY $vgpr5
+    ; GFX10-CONTRACT: [[BUILD_VECTOR1:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32)
+    ; GFX10-CONTRACT: [[COPY6:%[0-9]+]]:_(s32) = COPY $vgpr6
+    ; GFX10-CONTRACT: [[COPY7:%[0-9]+]]:_(s32) = COPY $vgpr7
+    ; GFX10-CONTRACT: [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr8
+    ; GFX10-CONTRACT: [[BUILD_VECTOR2:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[COPY6]](s32), [[COPY7]](s32), [[COPY8]](s32)
+    ; GFX10-CONTRACT: [[COPY9:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX10-CONTRACT: [[FMUL:%[0-9]+]]:_(<3 x s32>) = G_FMUL [[BUILD_VECTOR]], [[BUILD_VECTOR1]]
+    ; GFX10-CONTRACT: [[FADD:%[0-9]+]]:_(<3 x s32>) = G_FADD [[BUILD_VECTOR2]], [[FMUL]]
+    ; GFX10-CONTRACT: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FADD]](<3 x s32>)
+    ; GFX10-CONTRACT: $vgpr0 = COPY [[UV]](s32)
+    ; GFX10-CONTRACT: $vgpr1 = COPY [[UV1]](s32)
+    ; GFX10-CONTRACT: $vgpr2 = COPY [[UV2]](s32)
+    ; GFX10-CONTRACT: [[COPY10:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY9]]
+    ; GFX10-CONTRACT: S_SETPC_B64_return [[COPY10]], implicit $vgpr0, implicit $vgpr1, implicit $vgpr2
+    ; GFX10-DENORM-LABEL: name: test_3xfloat_add_mul_rhs
+    ; GFX10-DENORM: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX10-DENORM: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX10-DENORM: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX10-DENORM: [[BUILD_VECTOR:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32)
+    ; GFX10-DENORM: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3
+    ; GFX10-DENORM: [[COPY4:%[0-9]+]]:_(s32) = COPY $vgpr4
+    ; GFX10-DENORM: [[COPY5:%[0-9]+]]:_(s32) = COPY $vgpr5
+    ; GFX10-DENORM: [[BUILD_VECTOR1:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32)
+    ; GFX10-DENORM: [[COPY6:%[0-9]+]]:_(s32) = COPY $vgpr6
+    ; GFX10-DENORM: [[COPY7:%[0-9]+]]:_(s32) = COPY $vgpr7
+    ; GFX10-DENORM: [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr8
+    ; GFX10-DENORM: [[BUILD_VECTOR2:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[COPY6]](s32), [[COPY7]](s32), [[COPY8]](s32)
+    ; GFX10-DENORM: [[COPY9:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX10-DENORM: [[FMUL:%[0-9]+]]:_(<3 x s32>) = G_FMUL [[BUILD_VECTOR]], [[BUILD_VECTOR1]]
+    ; GFX10-DENORM: [[FADD:%[0-9]+]]:_(<3 x s32>) = G_FADD [[BUILD_VECTOR2]], [[FMUL]]
+    ; GFX10-DENORM: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FADD]](<3 x s32>)
+    ; GFX10-DENORM: $vgpr0 = COPY [[UV]](s32)
+    ; GFX10-DENORM: $vgpr1 = COPY [[UV1]](s32)
+    ; GFX10-DENORM: $vgpr2 = COPY [[UV2]](s32)
+    ; GFX10-DENORM: [[COPY10:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY9]]
+    ; GFX10-DENORM: S_SETPC_B64_return [[COPY10]], implicit $vgpr0, implicit $vgpr1, implicit $vgpr2
+    ; GFX10-UNSAFE-LABEL: name: test_3xfloat_add_mul_rhs
+    ; GFX10-UNSAFE: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX10-UNSAFE: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX10-UNSAFE: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX10-UNSAFE: [[BUILD_VECTOR:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32)
+    ; GFX10-UNSAFE: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3
+    ; GFX10-UNSAFE: [[COPY4:%[0-9]+]]:_(s32) = COPY $vgpr4
+    ; GFX10-UNSAFE: [[COPY5:%[0-9]+]]:_(s32) = COPY $vgpr5
+    ; GFX10-UNSAFE: [[BUILD_VECTOR1:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32)
+    ; GFX10-UNSAFE: [[COPY6:%[0-9]+]]:_(s32) = COPY $vgpr6
+    ; GFX10-UNSAFE: [[COPY7:%[0-9]+]]:_(s32) = COPY $vgpr7
+    ; GFX10-UNSAFE: [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr8
+    ; GFX10-UNSAFE: [[BUILD_VECTOR2:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[COPY6]](s32), [[COPY7]](s32), [[COPY8]](s32)
+    ; GFX10-UNSAFE: [[COPY9:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX10-UNSAFE: [[FMUL:%[0-9]+]]:_(<3 x s32>) = G_FMUL [[BUILD_VECTOR]], [[BUILD_VECTOR1]]
+    ; GFX10-UNSAFE: [[FADD:%[0-9]+]]:_(<3 x s32>) = G_FADD [[BUILD_VECTOR2]], [[FMUL]]
+    ; GFX10-UNSAFE: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FADD]](<3 x s32>)
+    ; GFX10-UNSAFE: $vgpr0 = COPY [[UV]](s32)
+    ; GFX10-UNSAFE: $vgpr1 = COPY [[UV1]](s32)
+    ; GFX10-UNSAFE: $vgpr2 = COPY [[UV2]](s32)
+    ; GFX10-UNSAFE: [[COPY10:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY9]]
+    ; GFX10-UNSAFE: S_SETPC_B64_return [[COPY10]], implicit $vgpr0, implicit $vgpr1, implicit $vgpr2
+    %4:_(s32) = COPY $vgpr0
+    %5:_(s32) = COPY $vgpr1
+    %6:_(s32) = COPY $vgpr2
+    %0:_(<3 x s32>) = G_BUILD_VECTOR %4(s32), %5(s32), %6(s32)
+    %7:_(s32) = COPY $vgpr3
+    %8:_(s32) = COPY $vgpr4
+    %9:_(s32) = COPY $vgpr5
+    %1:_(<3 x s32>) = G_BUILD_VECTOR %7(s32), %8(s32), %9(s32)
+    %10:_(s32) = COPY $vgpr6
+    %11:_(s32) = COPY $vgpr7
+    %12:_(s32) = COPY $vgpr8
+    %2:_(<3 x s32>) = G_BUILD_VECTOR %10(s32), %11(s32), %12(s32)
+    %3:sgpr_64 = COPY $sgpr30_sgpr31
+    %13:_(<3 x s32>) = G_FMUL %0, %1
+    %14:_(<3 x s32>) = G_FADD %2, %13
+    %16:_(s32), %17:_(s32), %18:_(s32) = G_UNMERGE_VALUES %14(<3 x s32>)
+    $vgpr0 = COPY %16(s32)
+    $vgpr1 = COPY %17(s32)
+    $vgpr2 = COPY %18(s32)
+    %15:ccr_sgpr_64 = COPY %3
+    S_SETPC_B64_return %15, implicit $vgpr0, implicit $vgpr1, implicit $vgpr2
+...
+
+---
+name:            test_4xhalf_add_mul
+body:             |
+  bb.1.entry:
+    liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $sgpr30_sgpr31
+
+    ; GFX9-LABEL: name: test_4xhalf_add_mul
+    ; GFX9: [[COPY:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0
+    ; GFX9: [[COPY1:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr1
+    ; GFX9: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[COPY]](<2 x s16>), [[COPY1]](<2 x s16>)
+    ; GFX9: [[COPY2:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr2
+    ; GFX9: [[COPY3:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr3
+    ; GFX9: [[CONCAT_VECTORS1:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[COPY2]](<2 x s16>), [[COPY3]](<2 x s16>)
+    ; GFX9: [[COPY4:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr4
+    ; GFX9: [[COPY5:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr5
+    ; GFX9: [[CONCAT_VECTORS2:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[COPY4]](<2 x s16>), [[COPY5]](<2 x s16>)
+    ; GFX9: [[COPY6:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX9: [[FMUL:%[0-9]+]]:_(<4 x s16>) = G_FMUL [[CONCAT_VECTORS]], [[CONCAT_VECTORS1]]
+    ; GFX9: [[FADD:%[0-9]+]]:_(<4 x s16>) = G_FADD [[FMUL]], [[CONCAT_VECTORS2]]
+    ; GFX9: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[FADD]](<4 x s16>)
+    ; GFX9: $vgpr0 = COPY [[UV]](<2 x s16>)
+    ; GFX9: $vgpr1 = COPY [[UV1]](<2 x s16>)
+    ; GFX9: [[COPY7:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY6]]
+    ; GFX9: S_SETPC_B64_return [[COPY7]], implicit $vgpr0, implicit $vgpr1
+    ; GFX9-CONTRACT-LABEL: name: test_4xhalf_add_mul
+    ; GFX9-CONTRACT: [[COPY:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0
+    ; GFX9-CONTRACT: [[COPY1:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr1
+    ; GFX9-CONTRACT: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[COPY]](<2 x s16>), [[COPY1]](<2 x s16>)
+    ; GFX9-CONTRACT: [[COPY2:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr2
+    ; GFX9-CONTRACT: [[COPY3:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr3
+    ; GFX9-CONTRACT: [[CONCAT_VECTORS1:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[COPY2]](<2 x s16>), [[COPY3]](<2 x s16>)
+    ; GFX9-CONTRACT: [[COPY4:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr4
+    ; GFX9-CONTRACT: [[COPY5:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr5
+    ; GFX9-CONTRACT: [[CONCAT_VECTORS2:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[COPY4]](<2 x s16>), [[COPY5]](<2 x s16>)
+    ; GFX9-CONTRACT: [[COPY6:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX9-CONTRACT: [[FMUL:%[0-9]+]]:_(<4 x s16>) = G_FMUL [[CONCAT_VECTORS]], [[CONCAT_VECTORS1]]
+    ; GFX9-CONTRACT: [[FADD:%[0-9]+]]:_(<4 x s16>) = G_FADD [[FMUL]], [[CONCAT_VECTORS2]]
+    ; GFX9-CONTRACT: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[FADD]](<4 x s16>)
+    ; GFX9-CONTRACT: $vgpr0 = COPY [[UV]](<2 x s16>)
+    ; GFX9-CONTRACT: $vgpr1 = COPY [[UV1]](<2 x s16>)
+    ; GFX9-CONTRACT: [[COPY7:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY6]]
+    ; GFX9-CONTRACT: S_SETPC_B64_return [[COPY7]], implicit $vgpr0, implicit $vgpr1
+    ; GFX9-DENORM-LABEL: name: test_4xhalf_add_mul
+    ; GFX9-DENORM: [[COPY:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0
+    ; GFX9-DENORM: [[COPY1:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr1
+    ; GFX9-DENORM: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[COPY]](<2 x s16>), [[COPY1]](<2 x s16>)
+    ; GFX9-DENORM: [[COPY2:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr2
+    ; GFX9-DENORM: [[COPY3:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr3
+    ; GFX9-DENORM: [[CONCAT_VECTORS1:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[COPY2]](<2 x s16>), [[COPY3]](<2 x s16>)
+    ; GFX9-DENORM: [[COPY4:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr4
+    ; GFX9-DENORM: [[COPY5:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr5
+    ; GFX9-DENORM: [[CONCAT_VECTORS2:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[COPY4]](<2 x s16>), [[COPY5]](<2 x s16>)
+    ; GFX9-DENORM: [[COPY6:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX9-DENORM: [[FMUL:%[0-9]+]]:_(<4 x s16>) = G_FMUL [[CONCAT_VECTORS]], [[CONCAT_VECTORS1]]
+    ; GFX9-DENORM: [[FADD:%[0-9]+]]:_(<4 x s16>) = G_FADD [[FMUL]], [[CONCAT_VECTORS2]]
+    ; GFX9-DENORM: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[FADD]](<4 x s16>)
+    ; GFX9-DENORM: $vgpr0 = COPY [[UV]](<2 x s16>)
+    ; GFX9-DENORM: $vgpr1 = COPY [[UV1]](<2 x s16>)
+    ; GFX9-DENORM: [[COPY7:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY6]]
+    ; GFX9-DENORM: S_SETPC_B64_return [[COPY7]], implicit $vgpr0, implicit $vgpr1
+    ; GFX9-UNSAFE-LABEL: name: test_4xhalf_add_mul
+    ; GFX9-UNSAFE: [[COPY:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0
+    ; GFX9-UNSAFE: [[COPY1:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr1
+    ; GFX9-UNSAFE: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[COPY]](<2 x s16>), [[COPY1]](<2 x s16>)
+    ; GFX9-UNSAFE: [[COPY2:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr2
+    ; GFX9-UNSAFE: [[COPY3:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr3
+    ; GFX9-UNSAFE: [[CONCAT_VECTORS1:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[COPY2]](<2 x s16>), [[COPY3]](<2 x s16>)
+    ; GFX9-UNSAFE: [[COPY4:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr4
+    ; GFX9-UNSAFE: [[COPY5:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr5
+    ; GFX9-UNSAFE: [[CONCAT_VECTORS2:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[COPY4]](<2 x s16>), [[COPY5]](<2 x s16>)
+    ; GFX9-UNSAFE: [[COPY6:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX9-UNSAFE: [[FMUL:%[0-9]+]]:_(<4 x s16>) = G_FMUL [[CONCAT_VECTORS]], [[CONCAT_VECTORS1]]
+    ; GFX9-UNSAFE: [[FADD:%[0-9]+]]:_(<4 x s16>) = G_FADD [[FMUL]], [[CONCAT_VECTORS2]]
+    ; GFX9-UNSAFE: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[FADD]](<4 x s16>)
+    ; GFX9-UNSAFE: $vgpr0 = COPY [[UV]](<2 x s16>)
+    ; GFX9-UNSAFE: $vgpr1 = COPY [[UV1]](<2 x s16>)
+    ; GFX9-UNSAFE: [[COPY7:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY6]]
+    ; GFX9-UNSAFE: S_SETPC_B64_return [[COPY7]], implicit $vgpr0, implicit $vgpr1
+    ; GFX10-LABEL: name: test_4xhalf_add_mul
+    ; GFX10: [[COPY:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0
+    ; GFX10: [[COPY1:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr1
+    ; GFX10: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[COPY]](<2 x s16>), [[COPY1]](<2 x s16>)
+    ; GFX10: [[COPY2:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr2
+    ; GFX10: [[COPY3:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr3
+    ; GFX10: [[CONCAT_VECTORS1:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[COPY2]](<2 x s16>), [[COPY3]](<2 x s16>)
+    ; GFX10: [[COPY4:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr4
+    ; GFX10: [[COPY5:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr5
+    ; GFX10: [[CONCAT_VECTORS2:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[COPY4]](<2 x s16>), [[COPY5]](<2 x s16>)
+    ; GFX10: [[COPY6:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX10: [[FMUL:%[0-9]+]]:_(<4 x s16>) = G_FMUL [[CONCAT_VECTORS]], [[CONCAT_VECTORS1]]
+    ; GFX10: [[FADD:%[0-9]+]]:_(<4 x s16>) = G_FADD [[FMUL]], [[CONCAT_VECTORS2]]
+    ; GFX10: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[FADD]](<4 x s16>)
+    ; GFX10: $vgpr0 = COPY [[UV]](<2 x s16>)
+    ; GFX10: $vgpr1 = COPY [[UV1]](<2 x s16>)
+    ; GFX10: [[COPY7:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY6]]
+    ; GFX10: S_SETPC_B64_return [[COPY7]], implicit $vgpr0, implicit $vgpr1
+    ; GFX10-CONTRACT-LABEL: name: test_4xhalf_add_mul
+    ; GFX10-CONTRACT: [[COPY:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0
+    ; GFX10-CONTRACT: [[COPY1:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr1
+    ; GFX10-CONTRACT: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[COPY]](<2 x s16>), [[COPY1]](<2 x s16>)
+    ; GFX10-CONTRACT: [[COPY2:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr2
+    ; GFX10-CONTRACT: [[COPY3:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr3
+    ; GFX10-CONTRACT: [[CONCAT_VECTORS1:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[COPY2]](<2 x s16>), [[COPY3]](<2 x s16>)
+    ; GFX10-CONTRACT: [[COPY4:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr4
+    ; GFX10-CONTRACT: [[COPY5:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr5
+    ; GFX10-CONTRACT: [[CONCAT_VECTORS2:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[COPY4]](<2 x s16>), [[COPY5]](<2 x s16>)
+    ; GFX10-CONTRACT: [[COPY6:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX10-CONTRACT: [[FMUL:%[0-9]+]]:_(<4 x s16>) = G_FMUL [[CONCAT_VECTORS]], [[CONCAT_VECTORS1]]
+    ; GFX10-CONTRACT: [[FADD:%[0-9]+]]:_(<4 x s16>) = G_FADD [[FMUL]], [[CONCAT_VECTORS2]]
+    ; GFX10-CONTRACT: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[FADD]](<4 x s16>)
+    ; GFX10-CONTRACT: $vgpr0 = COPY [[UV]](<2 x s16>)
+    ; GFX10-CONTRACT: $vgpr1 = COPY [[UV1]](<2 x s16>)
+    ; GFX10-CONTRACT: [[COPY7:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY6]]
+    ; GFX10-CONTRACT: S_SETPC_B64_return [[COPY7]], implicit $vgpr0, implicit $vgpr1
+    ; GFX10-DENORM-LABEL: name: test_4xhalf_add_mul
+    ; GFX10-DENORM: [[COPY:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0
+    ; GFX10-DENORM: [[COPY1:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr1
+    ; GFX10-DENORM: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[COPY]](<2 x s16>), [[COPY1]](<2 x s16>)
+    ; GFX10-DENORM: [[COPY2:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr2
+    ; GFX10-DENORM: [[COPY3:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr3
+    ; GFX10-DENORM: [[CONCAT_VECTORS1:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[COPY2]](<2 x s16>), [[COPY3]](<2 x s16>)
+    ; GFX10-DENORM: [[COPY4:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr4
+    ; GFX10-DENORM: [[COPY5:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr5
+    ; GFX10-DENORM: [[CONCAT_VECTORS2:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[COPY4]](<2 x s16>), [[COPY5]](<2 x s16>)
+    ; GFX10-DENORM: [[COPY6:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX10-DENORM: [[FMUL:%[0-9]+]]:_(<4 x s16>) = G_FMUL [[CONCAT_VECTORS]], [[CONCAT_VECTORS1]]
+    ; GFX10-DENORM: [[FADD:%[0-9]+]]:_(<4 x s16>) = G_FADD [[FMUL]], [[CONCAT_VECTORS2]]
+    ; GFX10-DENORM: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[FADD]](<4 x s16>)
+    ; GFX10-DENORM: $vgpr0 = COPY [[UV]](<2 x s16>)
+    ; GFX10-DENORM: $vgpr1 = COPY [[UV1]](<2 x s16>)
+    ; GFX10-DENORM: [[COPY7:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY6]]
+    ; GFX10-DENORM: S_SETPC_B64_return [[COPY7]], implicit $vgpr0, implicit $vgpr1
+    ; GFX10-UNSAFE-LABEL: name: test_4xhalf_add_mul
+    ; GFX10-UNSAFE: [[COPY:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0
+    ; GFX10-UNSAFE: [[COPY1:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr1
+    ; GFX10-UNSAFE: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[COPY]](<2 x s16>), [[COPY1]](<2 x s16>)
+    ; GFX10-UNSAFE: [[COPY2:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr2
+    ; GFX10-UNSAFE: [[COPY3:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr3
+    ; GFX10-UNSAFE: [[CONCAT_VECTORS1:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[COPY2]](<2 x s16>), [[COPY3]](<2 x s16>)
+    ; GFX10-UNSAFE: [[COPY4:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr4
+    ; GFX10-UNSAFE: [[COPY5:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr5
+    ; GFX10-UNSAFE: [[CONCAT_VECTORS2:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[COPY4]](<2 x s16>), [[COPY5]](<2 x s16>)
+    ; GFX10-UNSAFE: [[COPY6:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX10-UNSAFE: [[FMUL:%[0-9]+]]:_(<4 x s16>) = G_FMUL [[CONCAT_VECTORS]], [[CONCAT_VECTORS1]]
+    ; GFX10-UNSAFE: [[FADD:%[0-9]+]]:_(<4 x s16>) = G_FADD [[FMUL]], [[CONCAT_VECTORS2]]
+    ; GFX10-UNSAFE: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[FADD]](<4 x s16>)
+    ; GFX10-UNSAFE: $vgpr0 = COPY [[UV]](<2 x s16>)
+    ; GFX10-UNSAFE: $vgpr1 = COPY [[UV1]](<2 x s16>)
+    ; GFX10-UNSAFE: [[COPY7:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY6]]
+    ; GFX10-UNSAFE: S_SETPC_B64_return [[COPY7]], implicit $vgpr0, implicit $vgpr1
+    %4:_(<2 x s16>) = COPY $vgpr0
+    %5:_(<2 x s16>) = COPY $vgpr1
+    %0:_(<4 x s16>) = G_CONCAT_VECTORS %4(<2 x s16>), %5(<2 x s16>)
+    %6:_(<2 x s16>) = COPY $vgpr2
+    %7:_(<2 x s16>) = COPY $vgpr3
+    %1:_(<4 x s16>) = G_CONCAT_VECTORS %6(<2 x s16>), %7(<2 x s16>)
+    %8:_(<2 x s16>) = COPY $vgpr4
+    %9:_(<2 x s16>) = COPY $vgpr5
+    %2:_(<4 x s16>) = G_CONCAT_VECTORS %8(<2 x s16>), %9(<2 x s16>)
+    %3:sgpr_64 = COPY $sgpr30_sgpr31
+    %10:_(<4 x s16>) = G_FMUL %0, %1
+    %11:_(<4 x s16>) = G_FADD %10, %2
+    %13:_(<2 x s16>), %14:_(<2 x s16>) = G_UNMERGE_VALUES %11(<4 x s16>)
+    $vgpr0 = COPY %13(<2 x s16>)
+    $vgpr1 = COPY %14(<2 x s16>)
+    %12:ccr_sgpr_64 = COPY %3
+    S_SETPC_B64_return %12, implicit $vgpr0, implicit $vgpr1
+...
+
+---
+name:            test_3xhalf_add_mul_rhs
+body:             |
+  bb.1.entry:
+    liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $sgpr30_sgpr31
+
+    ; GFX9-LABEL: name: test_3xhalf_add_mul_rhs
+    ; GFX9: [[COPY:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0
+    ; GFX9: [[COPY1:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr1
+    ; GFX9: [[DEF:%[0-9]+]]:_(<2 x s16>) = G_IMPLICIT_DEF
+    ; GFX9: [[CONCAT_VECTORS:%[0-9]+]]:_(<6 x s16>) = G_CONCAT_VECTORS [[COPY]](<2 x s16>), [[COPY1]](<2 x s16>), [[DEF]](<2 x s16>)
+    ; GFX9: [[BITCAST:%[0-9]+]]:_(s96) = G_BITCAST [[CONCAT_VECTORS]](<6 x s16>)
+    ; GFX9: [[TRUNC:%[0-9]+]]:_(s48) = G_TRUNC [[BITCAST]](s96)
+    ; GFX9: [[BITCAST1:%[0-9]+]]:_(<3 x s16>) = G_BITCAST [[TRUNC]](s48)
+    ; GFX9: [[COPY2:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr2
+    ; GFX9: [[COPY3:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr3
+    ; GFX9: [[CONCAT_VECTORS1:%[0-9]+]]:_(<6 x s16>) = G_CONCAT_VECTORS [[COPY2]](<2 x s16>), [[COPY3]](<2 x s16>), [[DEF]](<2 x s16>)
+    ; GFX9: [[BITCAST2:%[0-9]+]]:_(s96) = G_BITCAST [[CONCAT_VECTORS1]](<6 x s16>)
+    ; GFX9: [[TRUNC1:%[0-9]+]]:_(s48) = G_TRUNC [[BITCAST2]](s96)
+    ; GFX9: [[BITCAST3:%[0-9]+]]:_(<3 x s16>) = G_BITCAST [[TRUNC1]](s48)
+    ; GFX9: [[COPY4:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr4
+    ; GFX9: [[COPY5:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr5
+    ; GFX9: [[CONCAT_VECTORS2:%[0-9]+]]:_(<6 x s16>) = G_CONCAT_VECTORS [[COPY4]](<2 x s16>), [[COPY5]](<2 x s16>), [[DEF]](<2 x s16>)
+    ; GFX9: [[BITCAST4:%[0-9]+]]:_(s96) = G_BITCAST [[CONCAT_VECTORS2]](<6 x s16>)
+    ; GFX9: [[TRUNC2:%[0-9]+]]:_(s48) = G_TRUNC [[BITCAST4]](s96)
+    ; GFX9: [[BITCAST5:%[0-9]+]]:_(<3 x s16>) = G_BITCAST [[TRUNC2]](s48)
+    ; GFX9: [[COPY6:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX9: [[FMUL:%[0-9]+]]:_(<3 x s16>) = G_FMUL [[BITCAST1]], [[BITCAST3]]
+    ; GFX9: [[FADD:%[0-9]+]]:_(<3 x s16>) = G_FADD [[BITCAST5]], [[FMUL]]
+    ; GFX9: [[DEF1:%[0-9]+]]:_(<3 x s16>) = G_IMPLICIT_DEF
+    ; GFX9: [[CONCAT_VECTORS3:%[0-9]+]]:_(<6 x s16>) = G_CONCAT_VECTORS [[FADD]](<3 x s16>), [[DEF1]](<3 x s16>)
+    ; GFX9: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>), [[UV2:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[CONCAT_VECTORS3]](<6 x s16>)
+    ; GFX9: $vgpr0 = COPY [[UV]](<2 x s16>)
+    ; GFX9: $vgpr1 = COPY [[UV1]](<2 x s16>)
+    ; GFX9: [[COPY7:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY6]]
+    ; GFX9: S_SETPC_B64_return [[COPY7]], implicit $vgpr0, implicit $vgpr1
+    ; GFX9-CONTRACT-LABEL: name: test_3xhalf_add_mul_rhs
+    ; GFX9-CONTRACT: [[COPY:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0
+    ; GFX9-CONTRACT: [[COPY1:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr1
+    ; GFX9-CONTRACT: [[DEF:%[0-9]+]]:_(<2 x s16>) = G_IMPLICIT_DEF
+    ; GFX9-CONTRACT: [[CONCAT_VECTORS:%[0-9]+]]:_(<6 x s16>) = G_CONCAT_VECTORS [[COPY]](<2 x s16>), [[COPY1]](<2 x s16>), [[DEF]](<2 x s16>)
+    ; GFX9-CONTRACT: [[BITCAST:%[0-9]+]]:_(s96) = G_BITCAST [[CONCAT_VECTORS]](<6 x s16>)
+    ; GFX9-CONTRACT: [[TRUNC:%[0-9]+]]:_(s48) = G_TRUNC [[BITCAST]](s96)
+    ; GFX9-CONTRACT: [[BITCAST1:%[0-9]+]]:_(<3 x s16>) = G_BITCAST [[TRUNC]](s48)
+    ; GFX9-CONTRACT: [[COPY2:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr2
+    ; GFX9-CONTRACT: [[COPY3:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr3
+    ; GFX9-CONTRACT: [[CONCAT_VECTORS1:%[0-9]+]]:_(<6 x s16>) = G_CONCAT_VECTORS [[COPY2]](<2 x s16>), [[COPY3]](<2 x s16>), [[DEF]](<2 x s16>)
+    ; GFX9-CONTRACT: [[BITCAST2:%[0-9]+]]:_(s96) = G_BITCAST [[CONCAT_VECTORS1]](<6 x s16>)
+    ; GFX9-CONTRACT: [[TRUNC1:%[0-9]+]]:_(s48) = G_TRUNC [[BITCAST2]](s96)
+    ; GFX9-CONTRACT: [[BITCAST3:%[0-9]+]]:_(<3 x s16>) = G_BITCAST [[TRUNC1]](s48)
+    ; GFX9-CONTRACT: [[COPY4:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr4
+    ; GFX9-CONTRACT: [[COPY5:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr5
+    ; GFX9-CONTRACT: [[CONCAT_VECTORS2:%[0-9]+]]:_(<6 x s16>) = G_CONCAT_VECTORS [[COPY4]](<2 x s16>), [[COPY5]](<2 x s16>), [[DEF]](<2 x s16>)
+    ; GFX9-CONTRACT: [[BITCAST4:%[0-9]+]]:_(s96) = G_BITCAST [[CONCAT_VECTORS2]](<6 x s16>)
+    ; GFX9-CONTRACT: [[TRUNC2:%[0-9]+]]:_(s48) = G_TRUNC [[BITCAST4]](s96)
+    ; GFX9-CONTRACT: [[BITCAST5:%[0-9]+]]:_(<3 x s16>) = G_BITCAST [[TRUNC2]](s48)
+    ; GFX9-CONTRACT: [[COPY6:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX9-CONTRACT: [[FMUL:%[0-9]+]]:_(<3 x s16>) = G_FMUL [[BITCAST1]], [[BITCAST3]]
+    ; GFX9-CONTRACT: [[FADD:%[0-9]+]]:_(<3 x s16>) = G_FADD [[BITCAST5]], [[FMUL]]
+    ; GFX9-CONTRACT: [[DEF1:%[0-9]+]]:_(<3 x s16>) = G_IMPLICIT_DEF
+    ; GFX9-CONTRACT: [[CONCAT_VECTORS3:%[0-9]+]]:_(<6 x s16>) = G_CONCAT_VECTORS [[FADD]](<3 x s16>), [[DEF1]](<3 x s16>)
+    ; GFX9-CONTRACT: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>), [[UV2:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[CONCAT_VECTORS3]](<6 x s16>)
+    ; GFX9-CONTRACT: $vgpr0 = COPY [[UV]](<2 x s16>)
+    ; GFX9-CONTRACT: $vgpr1 = COPY [[UV1]](<2 x s16>)
+    ; GFX9-CONTRACT: [[COPY7:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY6]]
+    ; GFX9-CONTRACT: S_SETPC_B64_return [[COPY7]], implicit $vgpr0, implicit $vgpr1
+    ; GFX9-DENORM-LABEL: name: test_3xhalf_add_mul_rhs
+    ; GFX9-DENORM: [[COPY:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0
+    ; GFX9-DENORM: [[COPY1:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr1
+    ; GFX9-DENORM: [[DEF:%[0-9]+]]:_(<2 x s16>) = G_IMPLICIT_DEF
+    ; GFX9-DENORM: [[CONCAT_VECTORS:%[0-9]+]]:_(<6 x s16>) = G_CONCAT_VECTORS [[COPY]](<2 x s16>), [[COPY1]](<2 x s16>), [[DEF]](<2 x s16>)
+    ; GFX9-DENORM: [[BITCAST:%[0-9]+]]:_(s96) = G_BITCAST [[CONCAT_VECTORS]](<6 x s16>)
+    ; GFX9-DENORM: [[TRUNC:%[0-9]+]]:_(s48) = G_TRUNC [[BITCAST]](s96)
+    ; GFX9-DENORM: [[BITCAST1:%[0-9]+]]:_(<3 x s16>) = G_BITCAST [[TRUNC]](s48)
+    ; GFX9-DENORM: [[COPY2:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr2
+    ; GFX9-DENORM: [[COPY3:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr3
+    ; GFX9-DENORM: [[CONCAT_VECTORS1:%[0-9]+]]:_(<6 x s16>) = G_CONCAT_VECTORS [[COPY2]](<2 x s16>), [[COPY3]](<2 x s16>), [[DEF]](<2 x s16>)
+    ; GFX9-DENORM: [[BITCAST2:%[0-9]+]]:_(s96) = G_BITCAST [[CONCAT_VECTORS1]](<6 x s16>)
+    ; GFX9-DENORM: [[TRUNC1:%[0-9]+]]:_(s48) = G_TRUNC [[BITCAST2]](s96)
+    ; GFX9-DENORM: [[BITCAST3:%[0-9]+]]:_(<3 x s16>) = G_BITCAST [[TRUNC1]](s48)
+    ; GFX9-DENORM: [[COPY4:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr4
+    ; GFX9-DENORM: [[COPY5:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr5
+    ; GFX9-DENORM: [[CONCAT_VECTORS2:%[0-9]+]]:_(<6 x s16>) = G_CONCAT_VECTORS [[COPY4]](<2 x s16>), [[COPY5]](<2 x s16>), [[DEF]](<2 x s16>)
+    ; GFX9-DENORM: [[BITCAST4:%[0-9]+]]:_(s96) = G_BITCAST [[CONCAT_VECTORS2]](<6 x s16>)
+    ; GFX9-DENORM: [[TRUNC2:%[0-9]+]]:_(s48) = G_TRUNC [[BITCAST4]](s96)
+    ; GFX9-DENORM: [[BITCAST5:%[0-9]+]]:_(<3 x s16>) = G_BITCAST [[TRUNC2]](s48)
+    ; GFX9-DENORM: [[COPY6:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX9-DENORM: [[FMUL:%[0-9]+]]:_(<3 x s16>) = G_FMUL [[BITCAST1]], [[BITCAST3]]
+    ; GFX9-DENORM: [[FADD:%[0-9]+]]:_(<3 x s16>) = G_FADD [[BITCAST5]], [[FMUL]]
+    ; GFX9-DENORM: [[DEF1:%[0-9]+]]:_(<3 x s16>) = G_IMPLICIT_DEF
+    ; GFX9-DENORM: [[CONCAT_VECTORS3:%[0-9]+]]:_(<6 x s16>) = G_CONCAT_VECTORS [[FADD]](<3 x s16>), [[DEF1]](<3 x s16>)
+    ; GFX9-DENORM: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>), [[UV2:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[CONCAT_VECTORS3]](<6 x s16>)
+    ; GFX9-DENORM: $vgpr0 = COPY [[UV]](<2 x s16>)
+    ; GFX9-DENORM: $vgpr1 = COPY [[UV1]](<2 x s16>)
+    ; GFX9-DENORM: [[COPY7:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY6]]
+    ; GFX9-DENORM: S_SETPC_B64_return [[COPY7]], implicit $vgpr0, implicit $vgpr1
+    ; GFX9-UNSAFE-LABEL: name: test_3xhalf_add_mul_rhs
+    ; GFX9-UNSAFE: [[COPY:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0
+    ; GFX9-UNSAFE: [[COPY1:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr1
+    ; GFX9-UNSAFE: [[DEF:%[0-9]+]]:_(<2 x s16>) = G_IMPLICIT_DEF
+    ; GFX9-UNSAFE: [[CONCAT_VECTORS:%[0-9]+]]:_(<6 x s16>) = G_CONCAT_VECTORS [[COPY]](<2 x s16>), [[COPY1]](<2 x s16>), [[DEF]](<2 x s16>)
+    ; GFX9-UNSAFE: [[BITCAST:%[0-9]+]]:_(s96) = G_BITCAST [[CONCAT_VECTORS]](<6 x s16>)
+    ; GFX9-UNSAFE: [[TRUNC:%[0-9]+]]:_(s48) = G_TRUNC [[BITCAST]](s96)
+    ; GFX9-UNSAFE: [[BITCAST1:%[0-9]+]]:_(<3 x s16>) = G_BITCAST [[TRUNC]](s48)
+    ; GFX9-UNSAFE: [[COPY2:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr2
+    ; GFX9-UNSAFE: [[COPY3:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr3
+    ; GFX9-UNSAFE: [[CONCAT_VECTORS1:%[0-9]+]]:_(<6 x s16>) = G_CONCAT_VECTORS [[COPY2]](<2 x s16>), [[COPY3]](<2 x s16>), [[DEF]](<2 x s16>)
+    ; GFX9-UNSAFE: [[BITCAST2:%[0-9]+]]:_(s96) = G_BITCAST [[CONCAT_VECTORS1]](<6 x s16>)
+    ; GFX9-UNSAFE: [[TRUNC1:%[0-9]+]]:_(s48) = G_TRUNC [[BITCAST2]](s96)
+    ; GFX9-UNSAFE: [[BITCAST3:%[0-9]+]]:_(<3 x s16>) = G_BITCAST [[TRUNC1]](s48)
+    ; GFX9-UNSAFE: [[COPY4:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr4
+    ; GFX9-UNSAFE: [[COPY5:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr5
+    ; GFX9-UNSAFE: [[CONCAT_VECTORS2:%[0-9]+]]:_(<6 x s16>) = G_CONCAT_VECTORS [[COPY4]](<2 x s16>), [[COPY5]](<2 x s16>), [[DEF]](<2 x s16>)
+    ; GFX9-UNSAFE: [[BITCAST4:%[0-9]+]]:_(s96) = G_BITCAST [[CONCAT_VECTORS2]](<6 x s16>)
+    ; GFX9-UNSAFE: [[TRUNC2:%[0-9]+]]:_(s48) = G_TRUNC [[BITCAST4]](s96)
+    ; GFX9-UNSAFE: [[BITCAST5:%[0-9]+]]:_(<3 x s16>) = G_BITCAST [[TRUNC2]](s48)
+    ; GFX9-UNSAFE: [[COPY6:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX9-UNSAFE: [[FMUL:%[0-9]+]]:_(<3 x s16>) = G_FMUL [[BITCAST1]], [[BITCAST3]]
+    ; GFX9-UNSAFE: [[FADD:%[0-9]+]]:_(<3 x s16>) = G_FADD [[BITCAST5]], [[FMUL]]
+    ; GFX9-UNSAFE: [[DEF1:%[0-9]+]]:_(<3 x s16>) = G_IMPLICIT_DEF
+    ; GFX9-UNSAFE: [[CONCAT_VECTORS3:%[0-9]+]]:_(<6 x s16>) = G_CONCAT_VECTORS [[FADD]](<3 x s16>), [[DEF1]](<3 x s16>)
+    ; GFX9-UNSAFE: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>), [[UV2:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[CONCAT_VECTORS3]](<6 x s16>)
+    ; GFX9-UNSAFE: $vgpr0 = COPY [[UV]](<2 x s16>)
+    ; GFX9-UNSAFE: $vgpr1 = COPY [[UV1]](<2 x s16>)
+    ; GFX9-UNSAFE: [[COPY7:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY6]]
+    ; GFX9-UNSAFE: S_SETPC_B64_return [[COPY7]], implicit $vgpr0, implicit $vgpr1
+    ; GFX10-LABEL: name: test_3xhalf_add_mul_rhs
+    ; GFX10: [[COPY:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0
+    ; GFX10: [[COPY1:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr1
+    ; GFX10: [[DEF:%[0-9]+]]:_(<2 x s16>) = G_IMPLICIT_DEF
+    ; GFX10: [[CONCAT_VECTORS:%[0-9]+]]:_(<6 x s16>) = G_CONCAT_VECTORS [[COPY]](<2 x s16>), [[COPY1]](<2 x s16>), [[DEF]](<2 x s16>)
+    ; GFX10: [[BITCAST:%[0-9]+]]:_(s96) = G_BITCAST [[CONCAT_VECTORS]](<6 x s16>)
+    ; GFX10: [[TRUNC:%[0-9]+]]:_(s48) = G_TRUNC [[BITCAST]](s96)
+    ; GFX10: [[BITCAST1:%[0-9]+]]:_(<3 x s16>) = G_BITCAST [[TRUNC]](s48)
+    ; GFX10: [[COPY2:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr2
+    ; GFX10: [[COPY3:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr3
+    ; GFX10: [[CONCAT_VECTORS1:%[0-9]+]]:_(<6 x s16>) = G_CONCAT_VECTORS [[COPY2]](<2 x s16>), [[COPY3]](<2 x s16>), [[DEF]](<2 x s16>)
+    ; GFX10: [[BITCAST2:%[0-9]+]]:_(s96) = G_BITCAST [[CONCAT_VECTORS1]](<6 x s16>)
+    ; GFX10: [[TRUNC1:%[0-9]+]]:_(s48) = G_TRUNC [[BITCAST2]](s96)
+    ; GFX10: [[BITCAST3:%[0-9]+]]:_(<3 x s16>) = G_BITCAST [[TRUNC1]](s48)
+    ; GFX10: [[COPY4:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr4
+    ; GFX10: [[COPY5:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr5
+    ; GFX10: [[CONCAT_VECTORS2:%[0-9]+]]:_(<6 x s16>) = G_CONCAT_VECTORS [[COPY4]](<2 x s16>), [[COPY5]](<2 x s16>), [[DEF]](<2 x s16>)
+    ; GFX10: [[BITCAST4:%[0-9]+]]:_(s96) = G_BITCAST [[CONCAT_VECTORS2]](<6 x s16>)
+    ; GFX10: [[TRUNC2:%[0-9]+]]:_(s48) = G_TRUNC [[BITCAST4]](s96)
+    ; GFX10: [[BITCAST5:%[0-9]+]]:_(<3 x s16>) = G_BITCAST [[TRUNC2]](s48)
+    ; GFX10: [[COPY6:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX10: [[FMUL:%[0-9]+]]:_(<3 x s16>) = G_FMUL [[BITCAST1]], [[BITCAST3]]
+    ; GFX10: [[FADD:%[0-9]+]]:_(<3 x s16>) = G_FADD [[BITCAST5]], [[FMUL]]
+    ; GFX10: [[DEF1:%[0-9]+]]:_(<3 x s16>) = G_IMPLICIT_DEF
+    ; GFX10: [[CONCAT_VECTORS3:%[0-9]+]]:_(<6 x s16>) = G_CONCAT_VECTORS [[FADD]](<3 x s16>), [[DEF1]](<3 x s16>)
+    ; GFX10: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>), [[UV2:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[CONCAT_VECTORS3]](<6 x s16>)
+    ; GFX10: $vgpr0 = COPY [[UV]](<2 x s16>)
+    ; GFX10: $vgpr1 = COPY [[UV1]](<2 x s16>)
+    ; GFX10: [[COPY7:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY6]]
+    ; GFX10: S_SETPC_B64_return [[COPY7]], implicit $vgpr0, implicit $vgpr1
+    ; GFX10-CONTRACT-LABEL: name: test_3xhalf_add_mul_rhs
+    ; GFX10-CONTRACT: [[COPY:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0
+    ; GFX10-CONTRACT: [[COPY1:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr1
+    ; GFX10-CONTRACT: [[DEF:%[0-9]+]]:_(<2 x s16>) = G_IMPLICIT_DEF
+    ; GFX10-CONTRACT: [[CONCAT_VECTORS:%[0-9]+]]:_(<6 x s16>) = G_CONCAT_VECTORS [[COPY]](<2 x s16>), [[COPY1]](<2 x s16>), [[DEF]](<2 x s16>)
+    ; GFX10-CONTRACT: [[BITCAST:%[0-9]+]]:_(s96) = G_BITCAST [[CONCAT_VECTORS]](<6 x s16>)
+    ; GFX10-CONTRACT: [[TRUNC:%[0-9]+]]:_(s48) = G_TRUNC [[BITCAST]](s96)
+    ; GFX10-CONTRACT: [[BITCAST1:%[0-9]+]]:_(<3 x s16>) = G_BITCAST [[TRUNC]](s48)
+    ; GFX10-CONTRACT: [[COPY2:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr2
+    ; GFX10-CONTRACT: [[COPY3:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr3
+    ; GFX10-CONTRACT: [[CONCAT_VECTORS1:%[0-9]+]]:_(<6 x s16>) = G_CONCAT_VECTORS [[COPY2]](<2 x s16>), [[COPY3]](<2 x s16>), [[DEF]](<2 x s16>)
+    ; GFX10-CONTRACT: [[BITCAST2:%[0-9]+]]:_(s96) = G_BITCAST [[CONCAT_VECTORS1]](<6 x s16>)
+    ; GFX10-CONTRACT: [[TRUNC1:%[0-9]+]]:_(s48) = G_TRUNC [[BITCAST2]](s96)
+    ; GFX10-CONTRACT: [[BITCAST3:%[0-9]+]]:_(<3 x s16>) = G_BITCAST [[TRUNC1]](s48)
+    ; GFX10-CONTRACT: [[COPY4:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr4
+    ; GFX10-CONTRACT: [[COPY5:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr5
+    ; GFX10-CONTRACT: [[CONCAT_VECTORS2:%[0-9]+]]:_(<6 x s16>) = G_CONCAT_VECTORS [[COPY4]](<2 x s16>), [[COPY5]](<2 x s16>), [[DEF]](<2 x s16>)
+    ; GFX10-CONTRACT: [[BITCAST4:%[0-9]+]]:_(s96) = G_BITCAST [[CONCAT_VECTORS2]](<6 x s16>)
+    ; GFX10-CONTRACT: [[TRUNC2:%[0-9]+]]:_(s48) = G_TRUNC [[BITCAST4]](s96)
+    ; GFX10-CONTRACT: [[BITCAST5:%[0-9]+]]:_(<3 x s16>) = G_BITCAST [[TRUNC2]](s48)
+    ; GFX10-CONTRACT: [[COPY6:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX10-CONTRACT: [[FMUL:%[0-9]+]]:_(<3 x s16>) = G_FMUL [[BITCAST1]], [[BITCAST3]]
+    ; GFX10-CONTRACT: [[FADD:%[0-9]+]]:_(<3 x s16>) = G_FADD [[BITCAST5]], [[FMUL]]
+    ; GFX10-CONTRACT: [[DEF1:%[0-9]+]]:_(<3 x s16>) = G_IMPLICIT_DEF
+    ; GFX10-CONTRACT: [[CONCAT_VECTORS3:%[0-9]+]]:_(<6 x s16>) = G_CONCAT_VECTORS [[FADD]](<3 x s16>), [[DEF1]](<3 x s16>)
+    ; GFX10-CONTRACT: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>), [[UV2:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[CONCAT_VECTORS3]](<6 x s16>)
+    ; GFX10-CONTRACT: $vgpr0 = COPY [[UV]](<2 x s16>)
+    ; GFX10-CONTRACT: $vgpr1 = COPY [[UV1]](<2 x s16>)
+    ; GFX10-CONTRACT: [[COPY7:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY6]]
+    ; GFX10-CONTRACT: S_SETPC_B64_return [[COPY7]], implicit $vgpr0, implicit $vgpr1
+    ; GFX10-DENORM-LABEL: name: test_3xhalf_add_mul_rhs
+    ; GFX10-DENORM: [[COPY:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0
+    ; GFX10-DENORM: [[COPY1:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr1
+    ; GFX10-DENORM: [[DEF:%[0-9]+]]:_(<2 x s16>) = G_IMPLICIT_DEF
+    ; GFX10-DENORM: [[CONCAT_VECTORS:%[0-9]+]]:_(<6 x s16>) = G_CONCAT_VECTORS [[COPY]](<2 x s16>), [[COPY1]](<2 x s16>), [[DEF]](<2 x s16>)
+    ; GFX10-DENORM: [[BITCAST:%[0-9]+]]:_(s96) = G_BITCAST [[CONCAT_VECTORS]](<6 x s16>)
+    ; GFX10-DENORM: [[TRUNC:%[0-9]+]]:_(s48) = G_TRUNC [[BITCAST]](s96)
+    ; GFX10-DENORM: [[BITCAST1:%[0-9]+]]:_(<3 x s16>) = G_BITCAST [[TRUNC]](s48)
+    ; GFX10-DENORM: [[COPY2:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr2
+    ; GFX10-DENORM: [[COPY3:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr3
+    ; GFX10-DENORM: [[CONCAT_VECTORS1:%[0-9]+]]:_(<6 x s16>) = G_CONCAT_VECTORS [[COPY2]](<2 x s16>), [[COPY3]](<2 x s16>), [[DEF]](<2 x s16>)
+    ; GFX10-DENORM: [[BITCAST2:%[0-9]+]]:_(s96) = G_BITCAST [[CONCAT_VECTORS1]](<6 x s16>)
+    ; GFX10-DENORM: [[TRUNC1:%[0-9]+]]:_(s48) = G_TRUNC [[BITCAST2]](s96)
+    ; GFX10-DENORM: [[BITCAST3:%[0-9]+]]:_(<3 x s16>) = G_BITCAST [[TRUNC1]](s48)
+    ; GFX10-DENORM: [[COPY4:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr4
+    ; GFX10-DENORM: [[COPY5:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr5
+    ; GFX10-DENORM: [[CONCAT_VECTORS2:%[0-9]+]]:_(<6 x s16>) = G_CONCAT_VECTORS [[COPY4]](<2 x s16>), [[COPY5]](<2 x s16>), [[DEF]](<2 x s16>)
+    ; GFX10-DENORM: [[BITCAST4:%[0-9]+]]:_(s96) = G_BITCAST [[CONCAT_VECTORS2]](<6 x s16>)
+    ; GFX10-DENORM: [[TRUNC2:%[0-9]+]]:_(s48) = G_TRUNC [[BITCAST4]](s96)
+    ; GFX10-DENORM: [[BITCAST5:%[0-9]+]]:_(<3 x s16>) = G_BITCAST [[TRUNC2]](s48)
+    ; GFX10-DENORM: [[COPY6:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX10-DENORM: [[FMUL:%[0-9]+]]:_(<3 x s16>) = G_FMUL [[BITCAST1]], [[BITCAST3]]
+    ; GFX10-DENORM: [[FADD:%[0-9]+]]:_(<3 x s16>) = G_FADD [[BITCAST5]], [[FMUL]]
+    ; GFX10-DENORM: [[DEF1:%[0-9]+]]:_(<3 x s16>) = G_IMPLICIT_DEF
+    ; GFX10-DENORM: [[CONCAT_VECTORS3:%[0-9]+]]:_(<6 x s16>) = G_CONCAT_VECTORS [[FADD]](<3 x s16>), [[DEF1]](<3 x s16>)
+    ; GFX10-DENORM: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>), [[UV2:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[CONCAT_VECTORS3]](<6 x s16>)
+    ; GFX10-DENORM: $vgpr0 = COPY [[UV]](<2 x s16>)
+    ; GFX10-DENORM: $vgpr1 = COPY [[UV1]](<2 x s16>)
+    ; GFX10-DENORM: [[COPY7:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY6]]
+    ; GFX10-DENORM: S_SETPC_B64_return [[COPY7]], implicit $vgpr0, implicit $vgpr1
+    ; GFX10-UNSAFE-LABEL: name: test_3xhalf_add_mul_rhs
+    ; GFX10-UNSAFE: [[COPY:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0
+    ; GFX10-UNSAFE: [[COPY1:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr1
+    ; GFX10-UNSAFE: [[DEF:%[0-9]+]]:_(<2 x s16>) = G_IMPLICIT_DEF
+    ; GFX10-UNSAFE: [[CONCAT_VECTORS:%[0-9]+]]:_(<6 x s16>) = G_CONCAT_VECTORS [[COPY]](<2 x s16>), [[COPY1]](<2 x s16>), [[DEF]](<2 x s16>)
+    ; GFX10-UNSAFE: [[BITCAST:%[0-9]+]]:_(s96) = G_BITCAST [[CONCAT_VECTORS]](<6 x s16>)
+    ; GFX10-UNSAFE: [[TRUNC:%[0-9]+]]:_(s48) = G_TRUNC [[BITCAST]](s96)
+    ; GFX10-UNSAFE: [[BITCAST1:%[0-9]+]]:_(<3 x s16>) = G_BITCAST [[TRUNC]](s48)
+    ; GFX10-UNSAFE: [[COPY2:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr2
+    ; GFX10-UNSAFE: [[COPY3:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr3
+    ; GFX10-UNSAFE: [[CONCAT_VECTORS1:%[0-9]+]]:_(<6 x s16>) = G_CONCAT_VECTORS [[COPY2]](<2 x s16>), [[COPY3]](<2 x s16>), [[DEF]](<2 x s16>)
+    ; GFX10-UNSAFE: [[BITCAST2:%[0-9]+]]:_(s96) = G_BITCAST [[CONCAT_VECTORS1]](<6 x s16>)
+    ; GFX10-UNSAFE: [[TRUNC1:%[0-9]+]]:_(s48) = G_TRUNC [[BITCAST2]](s96)
+    ; GFX10-UNSAFE: [[BITCAST3:%[0-9]+]]:_(<3 x s16>) = G_BITCAST [[TRUNC1]](s48)
+    ; GFX10-UNSAFE: [[COPY4:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr4
+    ; GFX10-UNSAFE: [[COPY5:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr5
+    ; GFX10-UNSAFE: [[CONCAT_VECTORS2:%[0-9]+]]:_(<6 x s16>) = G_CONCAT_VECTORS [[COPY4]](<2 x s16>), [[COPY5]](<2 x s16>), [[DEF]](<2 x s16>)
+    ; GFX10-UNSAFE: [[BITCAST4:%[0-9]+]]:_(s96) = G_BITCAST [[CONCAT_VECTORS2]](<6 x s16>)
+    ; GFX10-UNSAFE: [[TRUNC2:%[0-9]+]]:_(s48) = G_TRUNC [[BITCAST4]](s96)
+    ; GFX10-UNSAFE: [[BITCAST5:%[0-9]+]]:_(<3 x s16>) = G_BITCAST [[TRUNC2]](s48)
+    ; GFX10-UNSAFE: [[COPY6:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX10-UNSAFE: [[FMUL:%[0-9]+]]:_(<3 x s16>) = G_FMUL [[BITCAST1]], [[BITCAST3]]
+    ; GFX10-UNSAFE: [[FADD:%[0-9]+]]:_(<3 x s16>) = G_FADD [[BITCAST5]], [[FMUL]]
+    ; GFX10-UNSAFE: [[DEF1:%[0-9]+]]:_(<3 x s16>) = G_IMPLICIT_DEF
+    ; GFX10-UNSAFE: [[CONCAT_VECTORS3:%[0-9]+]]:_(<6 x s16>) = G_CONCAT_VECTORS [[FADD]](<3 x s16>), [[DEF1]](<3 x s16>)
+    ; GFX10-UNSAFE: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>), [[UV2:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[CONCAT_VECTORS3]](<6 x s16>)
+    ; GFX10-UNSAFE: $vgpr0 = COPY [[UV]](<2 x s16>)
+    ; GFX10-UNSAFE: $vgpr1 = COPY [[UV1]](<2 x s16>)
+    ; GFX10-UNSAFE: [[COPY7:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY6]]
+    ; GFX10-UNSAFE: S_SETPC_B64_return [[COPY7]], implicit $vgpr0, implicit $vgpr1
+    %4:_(<2 x s16>) = COPY $vgpr0
+    %5:_(<2 x s16>) = COPY $vgpr1
+    %10:_(<2 x s16>) = G_IMPLICIT_DEF
+    %11:_(<6 x s16>) = G_CONCAT_VECTORS %4(<2 x s16>), %5(<2 x s16>), %10(<2 x s16>)
+    %0:_(<3 x s16>), %12:_(<3 x s16>) = G_UNMERGE_VALUES %11(<6 x s16>)
+    %6:_(<2 x s16>) = COPY $vgpr2
+    %7:_(<2 x s16>) = COPY $vgpr3
+    %13:_(<6 x s16>) = G_CONCAT_VECTORS %6(<2 x s16>), %7(<2 x s16>), %10(<2 x s16>)
+    %1:_(<3 x s16>), %14:_(<3 x s16>) = G_UNMERGE_VALUES %13(<6 x s16>)
+    %8:_(<2 x s16>) = COPY $vgpr4
+    %9:_(<2 x s16>) = COPY $vgpr5
+    %15:_(<6 x s16>) = G_CONCAT_VECTORS %8(<2 x s16>), %9(<2 x s16>), %10(<2 x s16>)
+    %2:_(<3 x s16>), %16:_(<3 x s16>) = G_UNMERGE_VALUES %15(<6 x s16>)
+    %3:sgpr_64 = COPY $sgpr30_sgpr31
+    %17:_(<3 x s16>) = G_FMUL %0, %1
+    %18:_(<3 x s16>) = G_FADD %2, %17
+    %22:_(<3 x s16>) = G_IMPLICIT_DEF
+    %23:_(<6 x s16>) = G_CONCAT_VECTORS %18(<3 x s16>), %22(<3 x s16>)
+    %20:_(<2 x s16>), %21:_(<2 x s16>), %24:_(<2 x s16>) = G_UNMERGE_VALUES %23(<6 x s16>)
+    $vgpr0 = COPY %20(<2 x s16>)
+    $vgpr1 = COPY %21(<2 x s16>)
+    %19:ccr_sgpr_64 = COPY %3
+    S_SETPC_B64_return %19, implicit $vgpr0, implicit $vgpr1
+...
+
+---
+name:            test_4xdouble_add_mul
+body:             |
+  bb.1.entry:
+    liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8, $vgpr9, $vgpr10, $vgpr11, $vgpr12, $vgpr13, $vgpr14, $vgpr15, $vgpr16, $vgpr17, $vgpr18, $vgpr19, $vgpr20, $vgpr21, $vgpr22, $vgpr23, $sgpr30_sgpr31
+
+    ; GFX9-LABEL: name: test_4xdouble_add_mul
+    ; GFX9: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX9: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX9: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX9: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3
+    ; GFX9: [[COPY4:%[0-9]+]]:_(s32) = COPY $vgpr4
+    ; GFX9: [[COPY5:%[0-9]+]]:_(s32) = COPY $vgpr5
+    ; GFX9: [[COPY6:%[0-9]+]]:_(s32) = COPY $vgpr6
+    ; GFX9: [[COPY7:%[0-9]+]]:_(s32) = COPY $vgpr7
+    ; GFX9: [[MV:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY]](s32), [[COPY1]](s32)
+    ; GFX9: [[MV1:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY2]](s32), [[COPY3]](s32)
+    ; GFX9: [[MV2:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY4]](s32), [[COPY5]](s32)
+    ; GFX9: [[MV3:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY6]](s32), [[COPY7]](s32)
+    ; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s64>) = G_BUILD_VECTOR [[MV]](s64), [[MV1]](s64), [[MV2]](s64), [[MV3]](s64)
+    ; GFX9: [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr8
+    ; GFX9: [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr9
+    ; GFX9: [[COPY10:%[0-9]+]]:_(s32) = COPY $vgpr10
+    ; GFX9: [[COPY11:%[0-9]+]]:_(s32) = COPY $vgpr11
+    ; GFX9: [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr12
+    ; GFX9: [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr13
+    ; GFX9: [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr14
+    ; GFX9: [[COPY15:%[0-9]+]]:_(s32) = COPY $vgpr15
+    ; GFX9: [[MV4:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY8]](s32), [[COPY9]](s32)
+    ; GFX9: [[MV5:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY10]](s32), [[COPY11]](s32)
+    ; GFX9: [[MV6:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY12]](s32), [[COPY13]](s32)
+    ; GFX9: [[MV7:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY14]](s32), [[COPY15]](s32)
+    ; GFX9: [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s64>) = G_BUILD_VECTOR [[MV4]](s64), [[MV5]](s64), [[MV6]](s64), [[MV7]](s64)
+    ; GFX9: [[COPY16:%[0-9]+]]:_(s32) = COPY $vgpr16
+    ; GFX9: [[COPY17:%[0-9]+]]:_(s32) = COPY $vgpr17
+    ; GFX9: [[COPY18:%[0-9]+]]:_(s32) = COPY $vgpr18
+    ; GFX9: [[COPY19:%[0-9]+]]:_(s32) = COPY $vgpr19
+    ; GFX9: [[COPY20:%[0-9]+]]:_(s32) = COPY $vgpr20
+    ; GFX9: [[COPY21:%[0-9]+]]:_(s32) = COPY $vgpr21
+    ; GFX9: [[COPY22:%[0-9]+]]:_(s32) = COPY $vgpr22
+    ; GFX9: [[COPY23:%[0-9]+]]:_(s32) = COPY $vgpr23
+    ; GFX9: [[MV8:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY16]](s32), [[COPY17]](s32)
+    ; GFX9: [[MV9:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY18]](s32), [[COPY19]](s32)
+    ; GFX9: [[MV10:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY20]](s32), [[COPY21]](s32)
+    ; GFX9: [[MV11:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY22]](s32), [[COPY23]](s32)
+    ; GFX9: [[BUILD_VECTOR2:%[0-9]+]]:_(<4 x s64>) = G_BUILD_VECTOR [[MV8]](s64), [[MV9]](s64), [[MV10]](s64), [[MV11]](s64)
+    ; GFX9: [[COPY24:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX9: [[FMUL:%[0-9]+]]:_(<4 x s64>) = G_FMUL [[BUILD_VECTOR]], [[BUILD_VECTOR1]]
+    ; GFX9: [[FADD:%[0-9]+]]:_(<4 x s64>) = G_FADD [[FMUL]], [[BUILD_VECTOR2]]
+    ; GFX9: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32), [[UV4:%[0-9]+]]:_(s32), [[UV5:%[0-9]+]]:_(s32), [[UV6:%[0-9]+]]:_(s32), [[UV7:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FADD]](<4 x s64>)
+    ; GFX9: $vgpr0 = COPY [[UV]](s32)
+    ; GFX9: $vgpr1 = COPY [[UV1]](s32)
+    ; GFX9: $vgpr2 = COPY [[UV2]](s32)
+    ; GFX9: $vgpr3 = COPY [[UV3]](s32)
+    ; GFX9: $vgpr4 = COPY [[UV4]](s32)
+    ; GFX9: $vgpr5 = COPY [[UV5]](s32)
+    ; GFX9: $vgpr6 = COPY [[UV6]](s32)
+    ; GFX9: $vgpr7 = COPY [[UV7]](s32)
+    ; GFX9: [[COPY25:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY24]]
+    ; GFX9: S_SETPC_B64_return [[COPY25]], implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3, implicit $vgpr4, implicit $vgpr5, implicit $vgpr6, implicit $vgpr7
+    ; GFX9-CONTRACT-LABEL: name: test_4xdouble_add_mul
+    ; GFX9-CONTRACT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX9-CONTRACT: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX9-CONTRACT: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX9-CONTRACT: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3
+    ; GFX9-CONTRACT: [[COPY4:%[0-9]+]]:_(s32) = COPY $vgpr4
+    ; GFX9-CONTRACT: [[COPY5:%[0-9]+]]:_(s32) = COPY $vgpr5
+    ; GFX9-CONTRACT: [[COPY6:%[0-9]+]]:_(s32) = COPY $vgpr6
+    ; GFX9-CONTRACT: [[COPY7:%[0-9]+]]:_(s32) = COPY $vgpr7
+    ; GFX9-CONTRACT: [[MV:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY]](s32), [[COPY1]](s32)
+    ; GFX9-CONTRACT: [[MV1:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY2]](s32), [[COPY3]](s32)
+    ; GFX9-CONTRACT: [[MV2:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY4]](s32), [[COPY5]](s32)
+    ; GFX9-CONTRACT: [[MV3:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY6]](s32), [[COPY7]](s32)
+    ; GFX9-CONTRACT: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s64>) = G_BUILD_VECTOR [[MV]](s64), [[MV1]](s64), [[MV2]](s64), [[MV3]](s64)
+    ; GFX9-CONTRACT: [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr8
+    ; GFX9-CONTRACT: [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr9
+    ; GFX9-CONTRACT: [[COPY10:%[0-9]+]]:_(s32) = COPY $vgpr10
+    ; GFX9-CONTRACT: [[COPY11:%[0-9]+]]:_(s32) = COPY $vgpr11
+    ; GFX9-CONTRACT: [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr12
+    ; GFX9-CONTRACT: [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr13
+    ; GFX9-CONTRACT: [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr14
+    ; GFX9-CONTRACT: [[COPY15:%[0-9]+]]:_(s32) = COPY $vgpr15
+    ; GFX9-CONTRACT: [[MV4:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY8]](s32), [[COPY9]](s32)
+    ; GFX9-CONTRACT: [[MV5:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY10]](s32), [[COPY11]](s32)
+    ; GFX9-CONTRACT: [[MV6:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY12]](s32), [[COPY13]](s32)
+    ; GFX9-CONTRACT: [[MV7:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY14]](s32), [[COPY15]](s32)
+    ; GFX9-CONTRACT: [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s64>) = G_BUILD_VECTOR [[MV4]](s64), [[MV5]](s64), [[MV6]](s64), [[MV7]](s64)
+    ; GFX9-CONTRACT: [[COPY16:%[0-9]+]]:_(s32) = COPY $vgpr16
+    ; GFX9-CONTRACT: [[COPY17:%[0-9]+]]:_(s32) = COPY $vgpr17
+    ; GFX9-CONTRACT: [[COPY18:%[0-9]+]]:_(s32) = COPY $vgpr18
+    ; GFX9-CONTRACT: [[COPY19:%[0-9]+]]:_(s32) = COPY $vgpr19
+    ; GFX9-CONTRACT: [[COPY20:%[0-9]+]]:_(s32) = COPY $vgpr20
+    ; GFX9-CONTRACT: [[COPY21:%[0-9]+]]:_(s32) = COPY $vgpr21
+    ; GFX9-CONTRACT: [[COPY22:%[0-9]+]]:_(s32) = COPY $vgpr22
+    ; GFX9-CONTRACT: [[COPY23:%[0-9]+]]:_(s32) = COPY $vgpr23
+    ; GFX9-CONTRACT: [[MV8:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY16]](s32), [[COPY17]](s32)
+    ; GFX9-CONTRACT: [[MV9:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY18]](s32), [[COPY19]](s32)
+    ; GFX9-CONTRACT: [[MV10:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY20]](s32), [[COPY21]](s32)
+    ; GFX9-CONTRACT: [[MV11:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY22]](s32), [[COPY23]](s32)
+    ; GFX9-CONTRACT: [[BUILD_VECTOR2:%[0-9]+]]:_(<4 x s64>) = G_BUILD_VECTOR [[MV8]](s64), [[MV9]](s64), [[MV10]](s64), [[MV11]](s64)
+    ; GFX9-CONTRACT: [[COPY24:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX9-CONTRACT: [[FMUL:%[0-9]+]]:_(<4 x s64>) = G_FMUL [[BUILD_VECTOR]], [[BUILD_VECTOR1]]
+    ; GFX9-CONTRACT: [[FADD:%[0-9]+]]:_(<4 x s64>) = G_FADD [[FMUL]], [[BUILD_VECTOR2]]
+    ; GFX9-CONTRACT: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32), [[UV4:%[0-9]+]]:_(s32), [[UV5:%[0-9]+]]:_(s32), [[UV6:%[0-9]+]]:_(s32), [[UV7:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FADD]](<4 x s64>)
+    ; GFX9-CONTRACT: $vgpr0 = COPY [[UV]](s32)
+    ; GFX9-CONTRACT: $vgpr1 = COPY [[UV1]](s32)
+    ; GFX9-CONTRACT: $vgpr2 = COPY [[UV2]](s32)
+    ; GFX9-CONTRACT: $vgpr3 = COPY [[UV3]](s32)
+    ; GFX9-CONTRACT: $vgpr4 = COPY [[UV4]](s32)
+    ; GFX9-CONTRACT: $vgpr5 = COPY [[UV5]](s32)
+    ; GFX9-CONTRACT: $vgpr6 = COPY [[UV6]](s32)
+    ; GFX9-CONTRACT: $vgpr7 = COPY [[UV7]](s32)
+    ; GFX9-CONTRACT: [[COPY25:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY24]]
+    ; GFX9-CONTRACT: S_SETPC_B64_return [[COPY25]], implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3, implicit $vgpr4, implicit $vgpr5, implicit $vgpr6, implicit $vgpr7
+    ; GFX9-DENORM-LABEL: name: test_4xdouble_add_mul
+    ; GFX9-DENORM: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX9-DENORM: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX9-DENORM: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX9-DENORM: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3
+    ; GFX9-DENORM: [[COPY4:%[0-9]+]]:_(s32) = COPY $vgpr4
+    ; GFX9-DENORM: [[COPY5:%[0-9]+]]:_(s32) = COPY $vgpr5
+    ; GFX9-DENORM: [[COPY6:%[0-9]+]]:_(s32) = COPY $vgpr6
+    ; GFX9-DENORM: [[COPY7:%[0-9]+]]:_(s32) = COPY $vgpr7
+    ; GFX9-DENORM: [[MV:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY]](s32), [[COPY1]](s32)
+    ; GFX9-DENORM: [[MV1:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY2]](s32), [[COPY3]](s32)
+    ; GFX9-DENORM: [[MV2:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY4]](s32), [[COPY5]](s32)
+    ; GFX9-DENORM: [[MV3:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY6]](s32), [[COPY7]](s32)
+    ; GFX9-DENORM: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s64>) = G_BUILD_VECTOR [[MV]](s64), [[MV1]](s64), [[MV2]](s64), [[MV3]](s64)
+    ; GFX9-DENORM: [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr8
+    ; GFX9-DENORM: [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr9
+    ; GFX9-DENORM: [[COPY10:%[0-9]+]]:_(s32) = COPY $vgpr10
+    ; GFX9-DENORM: [[COPY11:%[0-9]+]]:_(s32) = COPY $vgpr11
+    ; GFX9-DENORM: [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr12
+    ; GFX9-DENORM: [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr13
+    ; GFX9-DENORM: [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr14
+    ; GFX9-DENORM: [[COPY15:%[0-9]+]]:_(s32) = COPY $vgpr15
+    ; GFX9-DENORM: [[MV4:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY8]](s32), [[COPY9]](s32)
+    ; GFX9-DENORM: [[MV5:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY10]](s32), [[COPY11]](s32)
+    ; GFX9-DENORM: [[MV6:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY12]](s32), [[COPY13]](s32)
+    ; GFX9-DENORM: [[MV7:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY14]](s32), [[COPY15]](s32)
+    ; GFX9-DENORM: [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s64>) = G_BUILD_VECTOR [[MV4]](s64), [[MV5]](s64), [[MV6]](s64), [[MV7]](s64)
+    ; GFX9-DENORM: [[COPY16:%[0-9]+]]:_(s32) = COPY $vgpr16
+    ; GFX9-DENORM: [[COPY17:%[0-9]+]]:_(s32) = COPY $vgpr17
+    ; GFX9-DENORM: [[COPY18:%[0-9]+]]:_(s32) = COPY $vgpr18
+    ; GFX9-DENORM: [[COPY19:%[0-9]+]]:_(s32) = COPY $vgpr19
+    ; GFX9-DENORM: [[COPY20:%[0-9]+]]:_(s32) = COPY $vgpr20
+    ; GFX9-DENORM: [[COPY21:%[0-9]+]]:_(s32) = COPY $vgpr21
+    ; GFX9-DENORM: [[COPY22:%[0-9]+]]:_(s32) = COPY $vgpr22
+    ; GFX9-DENORM: [[COPY23:%[0-9]+]]:_(s32) = COPY $vgpr23
+    ; GFX9-DENORM: [[MV8:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY16]](s32), [[COPY17]](s32)
+    ; GFX9-DENORM: [[MV9:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY18]](s32), [[COPY19]](s32)
+    ; GFX9-DENORM: [[MV10:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY20]](s32), [[COPY21]](s32)
+    ; GFX9-DENORM: [[MV11:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY22]](s32), [[COPY23]](s32)
+    ; GFX9-DENORM: [[BUILD_VECTOR2:%[0-9]+]]:_(<4 x s64>) = G_BUILD_VECTOR [[MV8]](s64), [[MV9]](s64), [[MV10]](s64), [[MV11]](s64)
+    ; GFX9-DENORM: [[COPY24:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX9-DENORM: [[FMUL:%[0-9]+]]:_(<4 x s64>) = G_FMUL [[BUILD_VECTOR]], [[BUILD_VECTOR1]]
+    ; GFX9-DENORM: [[FADD:%[0-9]+]]:_(<4 x s64>) = G_FADD [[FMUL]], [[BUILD_VECTOR2]]
+    ; GFX9-DENORM: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32), [[UV4:%[0-9]+]]:_(s32), [[UV5:%[0-9]+]]:_(s32), [[UV6:%[0-9]+]]:_(s32), [[UV7:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FADD]](<4 x s64>)
+    ; GFX9-DENORM: $vgpr0 = COPY [[UV]](s32)
+    ; GFX9-DENORM: $vgpr1 = COPY [[UV1]](s32)
+    ; GFX9-DENORM: $vgpr2 = COPY [[UV2]](s32)
+    ; GFX9-DENORM: $vgpr3 = COPY [[UV3]](s32)
+    ; GFX9-DENORM: $vgpr4 = COPY [[UV4]](s32)
+    ; GFX9-DENORM: $vgpr5 = COPY [[UV5]](s32)
+    ; GFX9-DENORM: $vgpr6 = COPY [[UV6]](s32)
+    ; GFX9-DENORM: $vgpr7 = COPY [[UV7]](s32)
+    ; GFX9-DENORM: [[COPY25:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY24]]
+    ; GFX9-DENORM: S_SETPC_B64_return [[COPY25]], implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3, implicit $vgpr4, implicit $vgpr5, implicit $vgpr6, implicit $vgpr7
+    ; GFX9-UNSAFE-LABEL: name: test_4xdouble_add_mul
+    ; GFX9-UNSAFE: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX9-UNSAFE: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX9-UNSAFE: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX9-UNSAFE: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3
+    ; GFX9-UNSAFE: [[COPY4:%[0-9]+]]:_(s32) = COPY $vgpr4
+    ; GFX9-UNSAFE: [[COPY5:%[0-9]+]]:_(s32) = COPY $vgpr5
+    ; GFX9-UNSAFE: [[COPY6:%[0-9]+]]:_(s32) = COPY $vgpr6
+    ; GFX9-UNSAFE: [[COPY7:%[0-9]+]]:_(s32) = COPY $vgpr7
+    ; GFX9-UNSAFE: [[MV:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY]](s32), [[COPY1]](s32)
+    ; GFX9-UNSAFE: [[MV1:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY2]](s32), [[COPY3]](s32)
+    ; GFX9-UNSAFE: [[MV2:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY4]](s32), [[COPY5]](s32)
+    ; GFX9-UNSAFE: [[MV3:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY6]](s32), [[COPY7]](s32)
+    ; GFX9-UNSAFE: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s64>) = G_BUILD_VECTOR [[MV]](s64), [[MV1]](s64), [[MV2]](s64), [[MV3]](s64)
+    ; GFX9-UNSAFE: [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr8
+    ; GFX9-UNSAFE: [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr9
+    ; GFX9-UNSAFE: [[COPY10:%[0-9]+]]:_(s32) = COPY $vgpr10
+    ; GFX9-UNSAFE: [[COPY11:%[0-9]+]]:_(s32) = COPY $vgpr11
+    ; GFX9-UNSAFE: [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr12
+    ; GFX9-UNSAFE: [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr13
+    ; GFX9-UNSAFE: [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr14
+    ; GFX9-UNSAFE: [[COPY15:%[0-9]+]]:_(s32) = COPY $vgpr15
+    ; GFX9-UNSAFE: [[MV4:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY8]](s32), [[COPY9]](s32)
+    ; GFX9-UNSAFE: [[MV5:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY10]](s32), [[COPY11]](s32)
+    ; GFX9-UNSAFE: [[MV6:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY12]](s32), [[COPY13]](s32)
+    ; GFX9-UNSAFE: [[MV7:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY14]](s32), [[COPY15]](s32)
+    ; GFX9-UNSAFE: [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s64>) = G_BUILD_VECTOR [[MV4]](s64), [[MV5]](s64), [[MV6]](s64), [[MV7]](s64)
+    ; GFX9-UNSAFE: [[COPY16:%[0-9]+]]:_(s32) = COPY $vgpr16
+    ; GFX9-UNSAFE: [[COPY17:%[0-9]+]]:_(s32) = COPY $vgpr17
+    ; GFX9-UNSAFE: [[COPY18:%[0-9]+]]:_(s32) = COPY $vgpr18
+    ; GFX9-UNSAFE: [[COPY19:%[0-9]+]]:_(s32) = COPY $vgpr19
+    ; GFX9-UNSAFE: [[COPY20:%[0-9]+]]:_(s32) = COPY $vgpr20
+    ; GFX9-UNSAFE: [[COPY21:%[0-9]+]]:_(s32) = COPY $vgpr21
+    ; GFX9-UNSAFE: [[COPY22:%[0-9]+]]:_(s32) = COPY $vgpr22
+    ; GFX9-UNSAFE: [[COPY23:%[0-9]+]]:_(s32) = COPY $vgpr23
+    ; GFX9-UNSAFE: [[MV8:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY16]](s32), [[COPY17]](s32)
+    ; GFX9-UNSAFE: [[MV9:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY18]](s32), [[COPY19]](s32)
+    ; GFX9-UNSAFE: [[MV10:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY20]](s32), [[COPY21]](s32)
+    ; GFX9-UNSAFE: [[MV11:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY22]](s32), [[COPY23]](s32)
+    ; GFX9-UNSAFE: [[BUILD_VECTOR2:%[0-9]+]]:_(<4 x s64>) = G_BUILD_VECTOR [[MV8]](s64), [[MV9]](s64), [[MV10]](s64), [[MV11]](s64)
+    ; GFX9-UNSAFE: [[COPY24:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX9-UNSAFE: [[FMUL:%[0-9]+]]:_(<4 x s64>) = G_FMUL [[BUILD_VECTOR]], [[BUILD_VECTOR1]]
+    ; GFX9-UNSAFE: [[FADD:%[0-9]+]]:_(<4 x s64>) = G_FADD [[FMUL]], [[BUILD_VECTOR2]]
+    ; GFX9-UNSAFE: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32), [[UV4:%[0-9]+]]:_(s32), [[UV5:%[0-9]+]]:_(s32), [[UV6:%[0-9]+]]:_(s32), [[UV7:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FADD]](<4 x s64>)
+    ; GFX9-UNSAFE: $vgpr0 = COPY [[UV]](s32)
+    ; GFX9-UNSAFE: $vgpr1 = COPY [[UV1]](s32)
+    ; GFX9-UNSAFE: $vgpr2 = COPY [[UV2]](s32)
+    ; GFX9-UNSAFE: $vgpr3 = COPY [[UV3]](s32)
+    ; GFX9-UNSAFE: $vgpr4 = COPY [[UV4]](s32)
+    ; GFX9-UNSAFE: $vgpr5 = COPY [[UV5]](s32)
+    ; GFX9-UNSAFE: $vgpr6 = COPY [[UV6]](s32)
+    ; GFX9-UNSAFE: $vgpr7 = COPY [[UV7]](s32)
+    ; GFX9-UNSAFE: [[COPY25:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY24]]
+    ; GFX9-UNSAFE: S_SETPC_B64_return [[COPY25]], implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3, implicit $vgpr4, implicit $vgpr5, implicit $vgpr6, implicit $vgpr7
+    ; GFX10-LABEL: name: test_4xdouble_add_mul
+    ; GFX10: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX10: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX10: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX10: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3
+    ; GFX10: [[COPY4:%[0-9]+]]:_(s32) = COPY $vgpr4
+    ; GFX10: [[COPY5:%[0-9]+]]:_(s32) = COPY $vgpr5
+    ; GFX10: [[COPY6:%[0-9]+]]:_(s32) = COPY $vgpr6
+    ; GFX10: [[COPY7:%[0-9]+]]:_(s32) = COPY $vgpr7
+    ; GFX10: [[MV:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY]](s32), [[COPY1]](s32)
+    ; GFX10: [[MV1:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY2]](s32), [[COPY3]](s32)
+    ; GFX10: [[MV2:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY4]](s32), [[COPY5]](s32)
+    ; GFX10: [[MV3:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY6]](s32), [[COPY7]](s32)
+    ; GFX10: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s64>) = G_BUILD_VECTOR [[MV]](s64), [[MV1]](s64), [[MV2]](s64), [[MV3]](s64)
+    ; GFX10: [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr8
+    ; GFX10: [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr9
+    ; GFX10: [[COPY10:%[0-9]+]]:_(s32) = COPY $vgpr10
+    ; GFX10: [[COPY11:%[0-9]+]]:_(s32) = COPY $vgpr11
+    ; GFX10: [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr12
+    ; GFX10: [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr13
+    ; GFX10: [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr14
+    ; GFX10: [[COPY15:%[0-9]+]]:_(s32) = COPY $vgpr15
+    ; GFX10: [[MV4:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY8]](s32), [[COPY9]](s32)
+    ; GFX10: [[MV5:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY10]](s32), [[COPY11]](s32)
+    ; GFX10: [[MV6:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY12]](s32), [[COPY13]](s32)
+    ; GFX10: [[MV7:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY14]](s32), [[COPY15]](s32)
+    ; GFX10: [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s64>) = G_BUILD_VECTOR [[MV4]](s64), [[MV5]](s64), [[MV6]](s64), [[MV7]](s64)
+    ; GFX10: [[COPY16:%[0-9]+]]:_(s32) = COPY $vgpr16
+    ; GFX10: [[COPY17:%[0-9]+]]:_(s32) = COPY $vgpr17
+    ; GFX10: [[COPY18:%[0-9]+]]:_(s32) = COPY $vgpr18
+    ; GFX10: [[COPY19:%[0-9]+]]:_(s32) = COPY $vgpr19
+    ; GFX10: [[COPY20:%[0-9]+]]:_(s32) = COPY $vgpr20
+    ; GFX10: [[COPY21:%[0-9]+]]:_(s32) = COPY $vgpr21
+    ; GFX10: [[COPY22:%[0-9]+]]:_(s32) = COPY $vgpr22
+    ; GFX10: [[COPY23:%[0-9]+]]:_(s32) = COPY $vgpr23
+    ; GFX10: [[MV8:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY16]](s32), [[COPY17]](s32)
+    ; GFX10: [[MV9:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY18]](s32), [[COPY19]](s32)
+    ; GFX10: [[MV10:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY20]](s32), [[COPY21]](s32)
+    ; GFX10: [[MV11:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY22]](s32), [[COPY23]](s32)
+    ; GFX10: [[BUILD_VECTOR2:%[0-9]+]]:_(<4 x s64>) = G_BUILD_VECTOR [[MV8]](s64), [[MV9]](s64), [[MV10]](s64), [[MV11]](s64)
+    ; GFX10: [[COPY24:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX10: [[FMUL:%[0-9]+]]:_(<4 x s64>) = G_FMUL [[BUILD_VECTOR]], [[BUILD_VECTOR1]]
+    ; GFX10: [[FADD:%[0-9]+]]:_(<4 x s64>) = G_FADD [[FMUL]], [[BUILD_VECTOR2]]
+    ; GFX10: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32), [[UV4:%[0-9]+]]:_(s32), [[UV5:%[0-9]+]]:_(s32), [[UV6:%[0-9]+]]:_(s32), [[UV7:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FADD]](<4 x s64>)
+    ; GFX10: $vgpr0 = COPY [[UV]](s32)
+    ; GFX10: $vgpr1 = COPY [[UV1]](s32)
+    ; GFX10: $vgpr2 = COPY [[UV2]](s32)
+    ; GFX10: $vgpr3 = COPY [[UV3]](s32)
+    ; GFX10: $vgpr4 = COPY [[UV4]](s32)
+    ; GFX10: $vgpr5 = COPY [[UV5]](s32)
+    ; GFX10: $vgpr6 = COPY [[UV6]](s32)
+    ; GFX10: $vgpr7 = COPY [[UV7]](s32)
+    ; GFX10: [[COPY25:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY24]]
+    ; GFX10: S_SETPC_B64_return [[COPY25]], implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3, implicit $vgpr4, implicit $vgpr5, implicit $vgpr6, implicit $vgpr7
+    ; GFX10-CONTRACT-LABEL: name: test_4xdouble_add_mul
+    ; GFX10-CONTRACT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX10-CONTRACT: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX10-CONTRACT: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX10-CONTRACT: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3
+    ; GFX10-CONTRACT: [[COPY4:%[0-9]+]]:_(s32) = COPY $vgpr4
+    ; GFX10-CONTRACT: [[COPY5:%[0-9]+]]:_(s32) = COPY $vgpr5
+    ; GFX10-CONTRACT: [[COPY6:%[0-9]+]]:_(s32) = COPY $vgpr6
+    ; GFX10-CONTRACT: [[COPY7:%[0-9]+]]:_(s32) = COPY $vgpr7
+    ; GFX10-CONTRACT: [[MV:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY]](s32), [[COPY1]](s32)
+    ; GFX10-CONTRACT: [[MV1:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY2]](s32), [[COPY3]](s32)
+    ; GFX10-CONTRACT: [[MV2:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY4]](s32), [[COPY5]](s32)
+    ; GFX10-CONTRACT: [[MV3:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY6]](s32), [[COPY7]](s32)
+    ; GFX10-CONTRACT: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s64>) = G_BUILD_VECTOR [[MV]](s64), [[MV1]](s64), [[MV2]](s64), [[MV3]](s64)
+    ; GFX10-CONTRACT: [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr8
+    ; GFX10-CONTRACT: [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr9
+    ; GFX10-CONTRACT: [[COPY10:%[0-9]+]]:_(s32) = COPY $vgpr10
+    ; GFX10-CONTRACT: [[COPY11:%[0-9]+]]:_(s32) = COPY $vgpr11
+    ; GFX10-CONTRACT: [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr12
+    ; GFX10-CONTRACT: [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr13
+    ; GFX10-CONTRACT: [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr14
+    ; GFX10-CONTRACT: [[COPY15:%[0-9]+]]:_(s32) = COPY $vgpr15
+    ; GFX10-CONTRACT: [[MV4:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY8]](s32), [[COPY9]](s32)
+    ; GFX10-CONTRACT: [[MV5:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY10]](s32), [[COPY11]](s32)
+    ; GFX10-CONTRACT: [[MV6:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY12]](s32), [[COPY13]](s32)
+    ; GFX10-CONTRACT: [[MV7:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY14]](s32), [[COPY15]](s32)
+    ; GFX10-CONTRACT: [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s64>) = G_BUILD_VECTOR [[MV4]](s64), [[MV5]](s64), [[MV6]](s64), [[MV7]](s64)
+    ; GFX10-CONTRACT: [[COPY16:%[0-9]+]]:_(s32) = COPY $vgpr16
+    ; GFX10-CONTRACT: [[COPY17:%[0-9]+]]:_(s32) = COPY $vgpr17
+    ; GFX10-CONTRACT: [[COPY18:%[0-9]+]]:_(s32) = COPY $vgpr18
+    ; GFX10-CONTRACT: [[COPY19:%[0-9]+]]:_(s32) = COPY $vgpr19
+    ; GFX10-CONTRACT: [[COPY20:%[0-9]+]]:_(s32) = COPY $vgpr20
+    ; GFX10-CONTRACT: [[COPY21:%[0-9]+]]:_(s32) = COPY $vgpr21
+    ; GFX10-CONTRACT: [[COPY22:%[0-9]+]]:_(s32) = COPY $vgpr22
+    ; GFX10-CONTRACT: [[COPY23:%[0-9]+]]:_(s32) = COPY $vgpr23
+    ; GFX10-CONTRACT: [[MV8:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY16]](s32), [[COPY17]](s32)
+    ; GFX10-CONTRACT: [[MV9:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY18]](s32), [[COPY19]](s32)
+    ; GFX10-CONTRACT: [[MV10:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY20]](s32), [[COPY21]](s32)
+    ; GFX10-CONTRACT: [[MV11:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY22]](s32), [[COPY23]](s32)
+    ; GFX10-CONTRACT: [[BUILD_VECTOR2:%[0-9]+]]:_(<4 x s64>) = G_BUILD_VECTOR [[MV8]](s64), [[MV9]](s64), [[MV10]](s64), [[MV11]](s64)
+    ; GFX10-CONTRACT: [[COPY24:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX10-CONTRACT: [[FMUL:%[0-9]+]]:_(<4 x s64>) = G_FMUL [[BUILD_VECTOR]], [[BUILD_VECTOR1]]
+    ; GFX10-CONTRACT: [[FADD:%[0-9]+]]:_(<4 x s64>) = G_FADD [[FMUL]], [[BUILD_VECTOR2]]
+    ; GFX10-CONTRACT: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32), [[UV4:%[0-9]+]]:_(s32), [[UV5:%[0-9]+]]:_(s32), [[UV6:%[0-9]+]]:_(s32), [[UV7:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FADD]](<4 x s64>)
+    ; GFX10-CONTRACT: $vgpr0 = COPY [[UV]](s32)
+    ; GFX10-CONTRACT: $vgpr1 = COPY [[UV1]](s32)
+    ; GFX10-CONTRACT: $vgpr2 = COPY [[UV2]](s32)
+    ; GFX10-CONTRACT: $vgpr3 = COPY [[UV3]](s32)
+    ; GFX10-CONTRACT: $vgpr4 = COPY [[UV4]](s32)
+    ; GFX10-CONTRACT: $vgpr5 = COPY [[UV5]](s32)
+    ; GFX10-CONTRACT: $vgpr6 = COPY [[UV6]](s32)
+    ; GFX10-CONTRACT: $vgpr7 = COPY [[UV7]](s32)
+    ; GFX10-CONTRACT: [[COPY25:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY24]]
+    ; GFX10-CONTRACT: S_SETPC_B64_return [[COPY25]], implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3, implicit $vgpr4, implicit $vgpr5, implicit $vgpr6, implicit $vgpr7
+    ; GFX10-DENORM-LABEL: name: test_4xdouble_add_mul
+    ; GFX10-DENORM: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX10-DENORM: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX10-DENORM: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX10-DENORM: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3
+    ; GFX10-DENORM: [[COPY4:%[0-9]+]]:_(s32) = COPY $vgpr4
+    ; GFX10-DENORM: [[COPY5:%[0-9]+]]:_(s32) = COPY $vgpr5
+    ; GFX10-DENORM: [[COPY6:%[0-9]+]]:_(s32) = COPY $vgpr6
+    ; GFX10-DENORM: [[COPY7:%[0-9]+]]:_(s32) = COPY $vgpr7
+    ; GFX10-DENORM: [[MV:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY]](s32), [[COPY1]](s32)
+    ; GFX10-DENORM: [[MV1:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY2]](s32), [[COPY3]](s32)
+    ; GFX10-DENORM: [[MV2:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY4]](s32), [[COPY5]](s32)
+    ; GFX10-DENORM: [[MV3:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY6]](s32), [[COPY7]](s32)
+    ; GFX10-DENORM: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s64>) = G_BUILD_VECTOR [[MV]](s64), [[MV1]](s64), [[MV2]](s64), [[MV3]](s64)
+    ; GFX10-DENORM: [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr8
+    ; GFX10-DENORM: [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr9
+    ; GFX10-DENORM: [[COPY10:%[0-9]+]]:_(s32) = COPY $vgpr10
+    ; GFX10-DENORM: [[COPY11:%[0-9]+]]:_(s32) = COPY $vgpr11
+    ; GFX10-DENORM: [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr12
+    ; GFX10-DENORM: [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr13
+    ; GFX10-DENORM: [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr14
+    ; GFX10-DENORM: [[COPY15:%[0-9]+]]:_(s32) = COPY $vgpr15
+    ; GFX10-DENORM: [[MV4:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY8]](s32), [[COPY9]](s32)
+    ; GFX10-DENORM: [[MV5:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY10]](s32), [[COPY11]](s32)
+    ; GFX10-DENORM: [[MV6:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY12]](s32), [[COPY13]](s32)
+    ; GFX10-DENORM: [[MV7:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY14]](s32), [[COPY15]](s32)
+    ; GFX10-DENORM: [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s64>) = G_BUILD_VECTOR [[MV4]](s64), [[MV5]](s64), [[MV6]](s64), [[MV7]](s64)
+    ; GFX10-DENORM: [[COPY16:%[0-9]+]]:_(s32) = COPY $vgpr16
+    ; GFX10-DENORM: [[COPY17:%[0-9]+]]:_(s32) = COPY $vgpr17
+    ; GFX10-DENORM: [[COPY18:%[0-9]+]]:_(s32) = COPY $vgpr18
+    ; GFX10-DENORM: [[COPY19:%[0-9]+]]:_(s32) = COPY $vgpr19
+    ; GFX10-DENORM: [[COPY20:%[0-9]+]]:_(s32) = COPY $vgpr20
+    ; GFX10-DENORM: [[COPY21:%[0-9]+]]:_(s32) = COPY $vgpr21
+    ; GFX10-DENORM: [[COPY22:%[0-9]+]]:_(s32) = COPY $vgpr22
+    ; GFX10-DENORM: [[COPY23:%[0-9]+]]:_(s32) = COPY $vgpr23
+    ; GFX10-DENORM: [[MV8:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY16]](s32), [[COPY17]](s32)
+    ; GFX10-DENORM: [[MV9:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY18]](s32), [[COPY19]](s32)
+    ; GFX10-DENORM: [[MV10:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY20]](s32), [[COPY21]](s32)
+    ; GFX10-DENORM: [[MV11:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY22]](s32), [[COPY23]](s32)
+    ; GFX10-DENORM: [[BUILD_VECTOR2:%[0-9]+]]:_(<4 x s64>) = G_BUILD_VECTOR [[MV8]](s64), [[MV9]](s64), [[MV10]](s64), [[MV11]](s64)
+    ; GFX10-DENORM: [[COPY24:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX10-DENORM: [[FMUL:%[0-9]+]]:_(<4 x s64>) = G_FMUL [[BUILD_VECTOR]], [[BUILD_VECTOR1]]
+    ; GFX10-DENORM: [[FADD:%[0-9]+]]:_(<4 x s64>) = G_FADD [[FMUL]], [[BUILD_VECTOR2]]
+    ; GFX10-DENORM: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32), [[UV4:%[0-9]+]]:_(s32), [[UV5:%[0-9]+]]:_(s32), [[UV6:%[0-9]+]]:_(s32), [[UV7:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FADD]](<4 x s64>)
+    ; GFX10-DENORM: $vgpr0 = COPY [[UV]](s32)
+    ; GFX10-DENORM: $vgpr1 = COPY [[UV1]](s32)
+    ; GFX10-DENORM: $vgpr2 = COPY [[UV2]](s32)
+    ; GFX10-DENORM: $vgpr3 = COPY [[UV3]](s32)
+    ; GFX10-DENORM: $vgpr4 = COPY [[UV4]](s32)
+    ; GFX10-DENORM: $vgpr5 = COPY [[UV5]](s32)
+    ; GFX10-DENORM: $vgpr6 = COPY [[UV6]](s32)
+    ; GFX10-DENORM: $vgpr7 = COPY [[UV7]](s32)
+    ; GFX10-DENORM: [[COPY25:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY24]]
+    ; GFX10-DENORM: S_SETPC_B64_return [[COPY25]], implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3, implicit $vgpr4, implicit $vgpr5, implicit $vgpr6, implicit $vgpr7
+    ; GFX10-UNSAFE-LABEL: name: test_4xdouble_add_mul
+    ; GFX10-UNSAFE: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX10-UNSAFE: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX10-UNSAFE: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX10-UNSAFE: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3
+    ; GFX10-UNSAFE: [[COPY4:%[0-9]+]]:_(s32) = COPY $vgpr4
+    ; GFX10-UNSAFE: [[COPY5:%[0-9]+]]:_(s32) = COPY $vgpr5
+    ; GFX10-UNSAFE: [[COPY6:%[0-9]+]]:_(s32) = COPY $vgpr6
+    ; GFX10-UNSAFE: [[COPY7:%[0-9]+]]:_(s32) = COPY $vgpr7
+    ; GFX10-UNSAFE: [[MV:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY]](s32), [[COPY1]](s32)
+    ; GFX10-UNSAFE: [[MV1:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY2]](s32), [[COPY3]](s32)
+    ; GFX10-UNSAFE: [[MV2:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY4]](s32), [[COPY5]](s32)
+    ; GFX10-UNSAFE: [[MV3:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY6]](s32), [[COPY7]](s32)
+    ; GFX10-UNSAFE: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s64>) = G_BUILD_VECTOR [[MV]](s64), [[MV1]](s64), [[MV2]](s64), [[MV3]](s64)
+    ; GFX10-UNSAFE: [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr8
+    ; GFX10-UNSAFE: [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr9
+    ; GFX10-UNSAFE: [[COPY10:%[0-9]+]]:_(s32) = COPY $vgpr10
+    ; GFX10-UNSAFE: [[COPY11:%[0-9]+]]:_(s32) = COPY $vgpr11
+    ; GFX10-UNSAFE: [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr12
+    ; GFX10-UNSAFE: [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr13
+    ; GFX10-UNSAFE: [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr14
+    ; GFX10-UNSAFE: [[COPY15:%[0-9]+]]:_(s32) = COPY $vgpr15
+    ; GFX10-UNSAFE: [[MV4:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY8]](s32), [[COPY9]](s32)
+    ; GFX10-UNSAFE: [[MV5:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY10]](s32), [[COPY11]](s32)
+    ; GFX10-UNSAFE: [[MV6:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY12]](s32), [[COPY13]](s32)
+    ; GFX10-UNSAFE: [[MV7:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY14]](s32), [[COPY15]](s32)
+    ; GFX10-UNSAFE: [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s64>) = G_BUILD_VECTOR [[MV4]](s64), [[MV5]](s64), [[MV6]](s64), [[MV7]](s64)
+    ; GFX10-UNSAFE: [[COPY16:%[0-9]+]]:_(s32) = COPY $vgpr16
+    ; GFX10-UNSAFE: [[COPY17:%[0-9]+]]:_(s32) = COPY $vgpr17
+    ; GFX10-UNSAFE: [[COPY18:%[0-9]+]]:_(s32) = COPY $vgpr18
+    ; GFX10-UNSAFE: [[COPY19:%[0-9]+]]:_(s32) = COPY $vgpr19
+    ; GFX10-UNSAFE: [[COPY20:%[0-9]+]]:_(s32) = COPY $vgpr20
+    ; GFX10-UNSAFE: [[COPY21:%[0-9]+]]:_(s32) = COPY $vgpr21
+    ; GFX10-UNSAFE: [[COPY22:%[0-9]+]]:_(s32) = COPY $vgpr22
+    ; GFX10-UNSAFE: [[COPY23:%[0-9]+]]:_(s32) = COPY $vgpr23
+    ; GFX10-UNSAFE: [[MV8:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY16]](s32), [[COPY17]](s32)
+    ; GFX10-UNSAFE: [[MV9:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY18]](s32), [[COPY19]](s32)
+    ; GFX10-UNSAFE: [[MV10:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY20]](s32), [[COPY21]](s32)
+    ; GFX10-UNSAFE: [[MV11:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY22]](s32), [[COPY23]](s32)
+    ; GFX10-UNSAFE: [[BUILD_VECTOR2:%[0-9]+]]:_(<4 x s64>) = G_BUILD_VECTOR [[MV8]](s64), [[MV9]](s64), [[MV10]](s64), [[MV11]](s64)
+    ; GFX10-UNSAFE: [[COPY24:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX10-UNSAFE: [[FMUL:%[0-9]+]]:_(<4 x s64>) = G_FMUL [[BUILD_VECTOR]], [[BUILD_VECTOR1]]
+    ; GFX10-UNSAFE: [[FADD:%[0-9]+]]:_(<4 x s64>) = G_FADD [[FMUL]], [[BUILD_VECTOR2]]
+    ; GFX10-UNSAFE: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32), [[UV4:%[0-9]+]]:_(s32), [[UV5:%[0-9]+]]:_(s32), [[UV6:%[0-9]+]]:_(s32), [[UV7:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FADD]](<4 x s64>)
+    ; GFX10-UNSAFE: $vgpr0 = COPY [[UV]](s32)
+    ; GFX10-UNSAFE: $vgpr1 = COPY [[UV1]](s32)
+    ; GFX10-UNSAFE: $vgpr2 = COPY [[UV2]](s32)
+    ; GFX10-UNSAFE: $vgpr3 = COPY [[UV3]](s32)
+    ; GFX10-UNSAFE: $vgpr4 = COPY [[UV4]](s32)
+    ; GFX10-UNSAFE: $vgpr5 = COPY [[UV5]](s32)
+    ; GFX10-UNSAFE: $vgpr6 = COPY [[UV6]](s32)
+    ; GFX10-UNSAFE: $vgpr7 = COPY [[UV7]](s32)
+    ; GFX10-UNSAFE: [[COPY25:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY24]]
+    ; GFX10-UNSAFE: S_SETPC_B64_return [[COPY25]], implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3, implicit $vgpr4, implicit $vgpr5, implicit $vgpr6, implicit $vgpr7
+    %4:_(s32) = COPY $vgpr0
+    %5:_(s32) = COPY $vgpr1
+    %6:_(s32) = COPY $vgpr2
+    %7:_(s32) = COPY $vgpr3
+    %8:_(s32) = COPY $vgpr4
+    %9:_(s32) = COPY $vgpr5
+    %10:_(s32) = COPY $vgpr6
+    %11:_(s32) = COPY $vgpr7
+    %28:_(s64) = G_MERGE_VALUES %4(s32), %5(s32)
+    %29:_(s64) = G_MERGE_VALUES %6(s32), %7(s32)
+    %30:_(s64) = G_MERGE_VALUES %8(s32), %9(s32)
+    %31:_(s64) = G_MERGE_VALUES %10(s32), %11(s32)
+    %0:_(<4 x s64>) = G_BUILD_VECTOR %28(s64), %29(s64), %30(s64), %31(s64)
+    %12:_(s32) = COPY $vgpr8
+    %13:_(s32) = COPY $vgpr9
+    %14:_(s32) = COPY $vgpr10
+    %15:_(s32) = COPY $vgpr11
+    %16:_(s32) = COPY $vgpr12
+    %17:_(s32) = COPY $vgpr13
+    %18:_(s32) = COPY $vgpr14
+    %19:_(s32) = COPY $vgpr15
+    %32:_(s64) = G_MERGE_VALUES %12(s32), %13(s32)
+    %33:_(s64) = G_MERGE_VALUES %14(s32), %15(s32)
+    %34:_(s64) = G_MERGE_VALUES %16(s32), %17(s32)
+    %35:_(s64) = G_MERGE_VALUES %18(s32), %19(s32)
+    %1:_(<4 x s64>) = G_BUILD_VECTOR %32(s64), %33(s64), %34(s64), %35(s64)
+    %20:_(s32) = COPY $vgpr16
+    %21:_(s32) = COPY $vgpr17
+    %22:_(s32) = COPY $vgpr18
+    %23:_(s32) = COPY $vgpr19
+    %24:_(s32) = COPY $vgpr20
+    %25:_(s32) = COPY $vgpr21
+    %26:_(s32) = COPY $vgpr22
+    %27:_(s32) = COPY $vgpr23
+    %36:_(s64) = G_MERGE_VALUES %20(s32), %21(s32)
+    %37:_(s64) = G_MERGE_VALUES %22(s32), %23(s32)
+    %38:_(s64) = G_MERGE_VALUES %24(s32), %25(s32)
+    %39:_(s64) = G_MERGE_VALUES %26(s32), %27(s32)
+    %2:_(<4 x s64>) = G_BUILD_VECTOR %36(s64), %37(s64), %38(s64), %39(s64)
+    %3:sgpr_64 = COPY $sgpr30_sgpr31
+    %40:_(<4 x s64>) = G_FMUL %0, %1
+    %41:_(<4 x s64>) = G_FADD %40, %2
+    %43:_(s32), %44:_(s32), %45:_(s32), %46:_(s32), %47:_(s32), %48:_(s32), %49:_(s32), %50:_(s32) = G_UNMERGE_VALUES %41(<4 x s64>)
+    $vgpr0 = COPY %43(s32)
+    $vgpr1 = COPY %44(s32)
+    $vgpr2 = COPY %45(s32)
+    $vgpr3 = COPY %46(s32)
+    $vgpr4 = COPY %47(s32)
+    $vgpr5 = COPY %48(s32)
+    $vgpr6 = COPY %49(s32)
+    $vgpr7 = COPY %50(s32)
+    %42:ccr_sgpr_64 = COPY %3
+    S_SETPC_B64_return %42, implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3, implicit $vgpr4, implicit $vgpr5, implicit $vgpr6, implicit $vgpr7
+...
+
+---
+name:            test_3xdouble_add_mul_rhs
+body:             |
+  bb.1.entry:
+    liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8, $vgpr9, $vgpr10, $vgpr11, $vgpr12, $vgpr13, $vgpr14, $vgpr15, $vgpr16, $vgpr17, $sgpr30_sgpr31
+
+    ; GFX9-LABEL: name: test_3xdouble_add_mul_rhs
+    ; GFX9: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX9: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX9: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX9: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3
+    ; GFX9: [[COPY4:%[0-9]+]]:_(s32) = COPY $vgpr4
+    ; GFX9: [[COPY5:%[0-9]+]]:_(s32) = COPY $vgpr5
+    ; GFX9: [[MV:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY]](s32), [[COPY1]](s32)
+    ; GFX9: [[MV1:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY2]](s32), [[COPY3]](s32)
+    ; GFX9: [[MV2:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY4]](s32), [[COPY5]](s32)
+    ; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<3 x s64>) = G_BUILD_VECTOR [[MV]](s64), [[MV1]](s64), [[MV2]](s64)
+    ; GFX9: [[COPY6:%[0-9]+]]:_(s32) = COPY $vgpr6
+    ; GFX9: [[COPY7:%[0-9]+]]:_(s32) = COPY $vgpr7
+    ; GFX9: [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr8
+    ; GFX9: [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr9
+    ; GFX9: [[COPY10:%[0-9]+]]:_(s32) = COPY $vgpr10
+    ; GFX9: [[COPY11:%[0-9]+]]:_(s32) = COPY $vgpr11
+    ; GFX9: [[MV3:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY6]](s32), [[COPY7]](s32)
+    ; GFX9: [[MV4:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY8]](s32), [[COPY9]](s32)
+    ; GFX9: [[MV5:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY10]](s32), [[COPY11]](s32)
+    ; GFX9: [[BUILD_VECTOR1:%[0-9]+]]:_(<3 x s64>) = G_BUILD_VECTOR [[MV3]](s64), [[MV4]](s64), [[MV5]](s64)
+    ; GFX9: [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr12
+    ; GFX9: [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr13
+    ; GFX9: [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr14
+    ; GFX9: [[COPY15:%[0-9]+]]:_(s32) = COPY $vgpr15
+    ; GFX9: [[COPY16:%[0-9]+]]:_(s32) = COPY $vgpr16
+    ; GFX9: [[COPY17:%[0-9]+]]:_(s32) = COPY $vgpr17
+    ; GFX9: [[MV6:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY12]](s32), [[COPY13]](s32)
+    ; GFX9: [[MV7:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY14]](s32), [[COPY15]](s32)
+    ; GFX9: [[MV8:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY16]](s32), [[COPY17]](s32)
+    ; GFX9: [[BUILD_VECTOR2:%[0-9]+]]:_(<3 x s64>) = G_BUILD_VECTOR [[MV6]](s64), [[MV7]](s64), [[MV8]](s64)
+    ; GFX9: [[COPY18:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX9: [[FMUL:%[0-9]+]]:_(<3 x s64>) = G_FMUL [[BUILD_VECTOR]], [[BUILD_VECTOR1]]
+    ; GFX9: [[FADD:%[0-9]+]]:_(<3 x s64>) = G_FADD [[BUILD_VECTOR2]], [[FMUL]]
+    ; GFX9: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32), [[UV4:%[0-9]+]]:_(s32), [[UV5:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FADD]](<3 x s64>)
+    ; GFX9: $vgpr0 = COPY [[UV]](s32)
+    ; GFX9: $vgpr1 = COPY [[UV1]](s32)
+    ; GFX9: $vgpr2 = COPY [[UV2]](s32)
+    ; GFX9: $vgpr3 = COPY [[UV3]](s32)
+    ; GFX9: $vgpr4 = COPY [[UV4]](s32)
+    ; GFX9: $vgpr5 = COPY [[UV5]](s32)
+    ; GFX9: [[COPY19:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY18]]
+    ; GFX9: S_SETPC_B64_return [[COPY19]], implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3, implicit $vgpr4, implicit $vgpr5
+    ; GFX9-CONTRACT-LABEL: name: test_3xdouble_add_mul_rhs
+    ; GFX9-CONTRACT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX9-CONTRACT: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX9-CONTRACT: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX9-CONTRACT: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3
+    ; GFX9-CONTRACT: [[COPY4:%[0-9]+]]:_(s32) = COPY $vgpr4
+    ; GFX9-CONTRACT: [[COPY5:%[0-9]+]]:_(s32) = COPY $vgpr5
+    ; GFX9-CONTRACT: [[MV:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY]](s32), [[COPY1]](s32)
+    ; GFX9-CONTRACT: [[MV1:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY2]](s32), [[COPY3]](s32)
+    ; GFX9-CONTRACT: [[MV2:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY4]](s32), [[COPY5]](s32)
+    ; GFX9-CONTRACT: [[BUILD_VECTOR:%[0-9]+]]:_(<3 x s64>) = G_BUILD_VECTOR [[MV]](s64), [[MV1]](s64), [[MV2]](s64)
+    ; GFX9-CONTRACT: [[COPY6:%[0-9]+]]:_(s32) = COPY $vgpr6
+    ; GFX9-CONTRACT: [[COPY7:%[0-9]+]]:_(s32) = COPY $vgpr7
+    ; GFX9-CONTRACT: [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr8
+    ; GFX9-CONTRACT: [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr9
+    ; GFX9-CONTRACT: [[COPY10:%[0-9]+]]:_(s32) = COPY $vgpr10
+    ; GFX9-CONTRACT: [[COPY11:%[0-9]+]]:_(s32) = COPY $vgpr11
+    ; GFX9-CONTRACT: [[MV3:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY6]](s32), [[COPY7]](s32)
+    ; GFX9-CONTRACT: [[MV4:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY8]](s32), [[COPY9]](s32)
+    ; GFX9-CONTRACT: [[MV5:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY10]](s32), [[COPY11]](s32)
+    ; GFX9-CONTRACT: [[BUILD_VECTOR1:%[0-9]+]]:_(<3 x s64>) = G_BUILD_VECTOR [[MV3]](s64), [[MV4]](s64), [[MV5]](s64)
+    ; GFX9-CONTRACT: [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr12
+    ; GFX9-CONTRACT: [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr13
+    ; GFX9-CONTRACT: [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr14
+    ; GFX9-CONTRACT: [[COPY15:%[0-9]+]]:_(s32) = COPY $vgpr15
+    ; GFX9-CONTRACT: [[COPY16:%[0-9]+]]:_(s32) = COPY $vgpr16
+    ; GFX9-CONTRACT: [[COPY17:%[0-9]+]]:_(s32) = COPY $vgpr17
+    ; GFX9-CONTRACT: [[MV6:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY12]](s32), [[COPY13]](s32)
+    ; GFX9-CONTRACT: [[MV7:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY14]](s32), [[COPY15]](s32)
+    ; GFX9-CONTRACT: [[MV8:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY16]](s32), [[COPY17]](s32)
+    ; GFX9-CONTRACT: [[BUILD_VECTOR2:%[0-9]+]]:_(<3 x s64>) = G_BUILD_VECTOR [[MV6]](s64), [[MV7]](s64), [[MV8]](s64)
+    ; GFX9-CONTRACT: [[COPY18:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX9-CONTRACT: [[FMUL:%[0-9]+]]:_(<3 x s64>) = G_FMUL [[BUILD_VECTOR]], [[BUILD_VECTOR1]]
+    ; GFX9-CONTRACT: [[FADD:%[0-9]+]]:_(<3 x s64>) = G_FADD [[BUILD_VECTOR2]], [[FMUL]]
+    ; GFX9-CONTRACT: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32), [[UV4:%[0-9]+]]:_(s32), [[UV5:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FADD]](<3 x s64>)
+    ; GFX9-CONTRACT: $vgpr0 = COPY [[UV]](s32)
+    ; GFX9-CONTRACT: $vgpr1 = COPY [[UV1]](s32)
+    ; GFX9-CONTRACT: $vgpr2 = COPY [[UV2]](s32)
+    ; GFX9-CONTRACT: $vgpr3 = COPY [[UV3]](s32)
+    ; GFX9-CONTRACT: $vgpr4 = COPY [[UV4]](s32)
+    ; GFX9-CONTRACT: $vgpr5 = COPY [[UV5]](s32)
+    ; GFX9-CONTRACT: [[COPY19:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY18]]
+    ; GFX9-CONTRACT: S_SETPC_B64_return [[COPY19]], implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3, implicit $vgpr4, implicit $vgpr5
+    ; GFX9-DENORM-LABEL: name: test_3xdouble_add_mul_rhs
+    ; GFX9-DENORM: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX9-DENORM: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX9-DENORM: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX9-DENORM: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3
+    ; GFX9-DENORM: [[COPY4:%[0-9]+]]:_(s32) = COPY $vgpr4
+    ; GFX9-DENORM: [[COPY5:%[0-9]+]]:_(s32) = COPY $vgpr5
+    ; GFX9-DENORM: [[MV:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY]](s32), [[COPY1]](s32)
+    ; GFX9-DENORM: [[MV1:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY2]](s32), [[COPY3]](s32)
+    ; GFX9-DENORM: [[MV2:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY4]](s32), [[COPY5]](s32)
+    ; GFX9-DENORM: [[BUILD_VECTOR:%[0-9]+]]:_(<3 x s64>) = G_BUILD_VECTOR [[MV]](s64), [[MV1]](s64), [[MV2]](s64)
+    ; GFX9-DENORM: [[COPY6:%[0-9]+]]:_(s32) = COPY $vgpr6
+    ; GFX9-DENORM: [[COPY7:%[0-9]+]]:_(s32) = COPY $vgpr7
+    ; GFX9-DENORM: [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr8
+    ; GFX9-DENORM: [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr9
+    ; GFX9-DENORM: [[COPY10:%[0-9]+]]:_(s32) = COPY $vgpr10
+    ; GFX9-DENORM: [[COPY11:%[0-9]+]]:_(s32) = COPY $vgpr11
+    ; GFX9-DENORM: [[MV3:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY6]](s32), [[COPY7]](s32)
+    ; GFX9-DENORM: [[MV4:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY8]](s32), [[COPY9]](s32)
+    ; GFX9-DENORM: [[MV5:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY10]](s32), [[COPY11]](s32)
+    ; GFX9-DENORM: [[BUILD_VECTOR1:%[0-9]+]]:_(<3 x s64>) = G_BUILD_VECTOR [[MV3]](s64), [[MV4]](s64), [[MV5]](s64)
+    ; GFX9-DENORM: [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr12
+    ; GFX9-DENORM: [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr13
+    ; GFX9-DENORM: [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr14
+    ; GFX9-DENORM: [[COPY15:%[0-9]+]]:_(s32) = COPY $vgpr15
+    ; GFX9-DENORM: [[COPY16:%[0-9]+]]:_(s32) = COPY $vgpr16
+    ; GFX9-DENORM: [[COPY17:%[0-9]+]]:_(s32) = COPY $vgpr17
+    ; GFX9-DENORM: [[MV6:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY12]](s32), [[COPY13]](s32)
+    ; GFX9-DENORM: [[MV7:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY14]](s32), [[COPY15]](s32)
+    ; GFX9-DENORM: [[MV8:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY16]](s32), [[COPY17]](s32)
+    ; GFX9-DENORM: [[BUILD_VECTOR2:%[0-9]+]]:_(<3 x s64>) = G_BUILD_VECTOR [[MV6]](s64), [[MV7]](s64), [[MV8]](s64)
+    ; GFX9-DENORM: [[COPY18:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX9-DENORM: [[FMUL:%[0-9]+]]:_(<3 x s64>) = G_FMUL [[BUILD_VECTOR]], [[BUILD_VECTOR1]]
+    ; GFX9-DENORM: [[FADD:%[0-9]+]]:_(<3 x s64>) = G_FADD [[BUILD_VECTOR2]], [[FMUL]]
+    ; GFX9-DENORM: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32), [[UV4:%[0-9]+]]:_(s32), [[UV5:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FADD]](<3 x s64>)
+    ; GFX9-DENORM: $vgpr0 = COPY [[UV]](s32)
+    ; GFX9-DENORM: $vgpr1 = COPY [[UV1]](s32)
+    ; GFX9-DENORM: $vgpr2 = COPY [[UV2]](s32)
+    ; GFX9-DENORM: $vgpr3 = COPY [[UV3]](s32)
+    ; GFX9-DENORM: $vgpr4 = COPY [[UV4]](s32)
+    ; GFX9-DENORM: $vgpr5 = COPY [[UV5]](s32)
+    ; GFX9-DENORM: [[COPY19:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY18]]
+    ; GFX9-DENORM: S_SETPC_B64_return [[COPY19]], implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3, implicit $vgpr4, implicit $vgpr5
+    ; GFX9-UNSAFE-LABEL: name: test_3xdouble_add_mul_rhs
+    ; GFX9-UNSAFE: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX9-UNSAFE: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX9-UNSAFE: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX9-UNSAFE: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3
+    ; GFX9-UNSAFE: [[COPY4:%[0-9]+]]:_(s32) = COPY $vgpr4
+    ; GFX9-UNSAFE: [[COPY5:%[0-9]+]]:_(s32) = COPY $vgpr5
+    ; GFX9-UNSAFE: [[MV:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY]](s32), [[COPY1]](s32)
+    ; GFX9-UNSAFE: [[MV1:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY2]](s32), [[COPY3]](s32)
+    ; GFX9-UNSAFE: [[MV2:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY4]](s32), [[COPY5]](s32)
+    ; GFX9-UNSAFE: [[BUILD_VECTOR:%[0-9]+]]:_(<3 x s64>) = G_BUILD_VECTOR [[MV]](s64), [[MV1]](s64), [[MV2]](s64)
+    ; GFX9-UNSAFE: [[COPY6:%[0-9]+]]:_(s32) = COPY $vgpr6
+    ; GFX9-UNSAFE: [[COPY7:%[0-9]+]]:_(s32) = COPY $vgpr7
+    ; GFX9-UNSAFE: [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr8
+    ; GFX9-UNSAFE: [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr9
+    ; GFX9-UNSAFE: [[COPY10:%[0-9]+]]:_(s32) = COPY $vgpr10
+    ; GFX9-UNSAFE: [[COPY11:%[0-9]+]]:_(s32) = COPY $vgpr11
+    ; GFX9-UNSAFE: [[MV3:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY6]](s32), [[COPY7]](s32)
+    ; GFX9-UNSAFE: [[MV4:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY8]](s32), [[COPY9]](s32)
+    ; GFX9-UNSAFE: [[MV5:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY10]](s32), [[COPY11]](s32)
+    ; GFX9-UNSAFE: [[BUILD_VECTOR1:%[0-9]+]]:_(<3 x s64>) = G_BUILD_VECTOR [[MV3]](s64), [[MV4]](s64), [[MV5]](s64)
+    ; GFX9-UNSAFE: [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr12
+    ; GFX9-UNSAFE: [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr13
+    ; GFX9-UNSAFE: [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr14
+    ; GFX9-UNSAFE: [[COPY15:%[0-9]+]]:_(s32) = COPY $vgpr15
+    ; GFX9-UNSAFE: [[COPY16:%[0-9]+]]:_(s32) = COPY $vgpr16
+    ; GFX9-UNSAFE: [[COPY17:%[0-9]+]]:_(s32) = COPY $vgpr17
+    ; GFX9-UNSAFE: [[MV6:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY12]](s32), [[COPY13]](s32)
+    ; GFX9-UNSAFE: [[MV7:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY14]](s32), [[COPY15]](s32)
+    ; GFX9-UNSAFE: [[MV8:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY16]](s32), [[COPY17]](s32)
+    ; GFX9-UNSAFE: [[BUILD_VECTOR2:%[0-9]+]]:_(<3 x s64>) = G_BUILD_VECTOR [[MV6]](s64), [[MV7]](s64), [[MV8]](s64)
+    ; GFX9-UNSAFE: [[COPY18:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX9-UNSAFE: [[FMUL:%[0-9]+]]:_(<3 x s64>) = G_FMUL [[BUILD_VECTOR]], [[BUILD_VECTOR1]]
+    ; GFX9-UNSAFE: [[FADD:%[0-9]+]]:_(<3 x s64>) = G_FADD [[BUILD_VECTOR2]], [[FMUL]]
+    ; GFX9-UNSAFE: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32), [[UV4:%[0-9]+]]:_(s32), [[UV5:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FADD]](<3 x s64>)
+    ; GFX9-UNSAFE: $vgpr0 = COPY [[UV]](s32)
+    ; GFX9-UNSAFE: $vgpr1 = COPY [[UV1]](s32)
+    ; GFX9-UNSAFE: $vgpr2 = COPY [[UV2]](s32)
+    ; GFX9-UNSAFE: $vgpr3 = COPY [[UV3]](s32)
+    ; GFX9-UNSAFE: $vgpr4 = COPY [[UV4]](s32)
+    ; GFX9-UNSAFE: $vgpr5 = COPY [[UV5]](s32)
+    ; GFX9-UNSAFE: [[COPY19:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY18]]
+    ; GFX9-UNSAFE: S_SETPC_B64_return [[COPY19]], implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3, implicit $vgpr4, implicit $vgpr5
+    ; GFX10-LABEL: name: test_3xdouble_add_mul_rhs
+    ; GFX10: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX10: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX10: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX10: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3
+    ; GFX10: [[COPY4:%[0-9]+]]:_(s32) = COPY $vgpr4
+    ; GFX10: [[COPY5:%[0-9]+]]:_(s32) = COPY $vgpr5
+    ; GFX10: [[MV:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY]](s32), [[COPY1]](s32)
+    ; GFX10: [[MV1:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY2]](s32), [[COPY3]](s32)
+    ; GFX10: [[MV2:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY4]](s32), [[COPY5]](s32)
+    ; GFX10: [[BUILD_VECTOR:%[0-9]+]]:_(<3 x s64>) = G_BUILD_VECTOR [[MV]](s64), [[MV1]](s64), [[MV2]](s64)
+    ; GFX10: [[COPY6:%[0-9]+]]:_(s32) = COPY $vgpr6
+    ; GFX10: [[COPY7:%[0-9]+]]:_(s32) = COPY $vgpr7
+    ; GFX10: [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr8
+    ; GFX10: [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr9
+    ; GFX10: [[COPY10:%[0-9]+]]:_(s32) = COPY $vgpr10
+    ; GFX10: [[COPY11:%[0-9]+]]:_(s32) = COPY $vgpr11
+    ; GFX10: [[MV3:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY6]](s32), [[COPY7]](s32)
+    ; GFX10: [[MV4:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY8]](s32), [[COPY9]](s32)
+    ; GFX10: [[MV5:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY10]](s32), [[COPY11]](s32)
+    ; GFX10: [[BUILD_VECTOR1:%[0-9]+]]:_(<3 x s64>) = G_BUILD_VECTOR [[MV3]](s64), [[MV4]](s64), [[MV5]](s64)
+    ; GFX10: [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr12
+    ; GFX10: [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr13
+    ; GFX10: [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr14
+    ; GFX10: [[COPY15:%[0-9]+]]:_(s32) = COPY $vgpr15
+    ; GFX10: [[COPY16:%[0-9]+]]:_(s32) = COPY $vgpr16
+    ; GFX10: [[COPY17:%[0-9]+]]:_(s32) = COPY $vgpr17
+    ; GFX10: [[MV6:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY12]](s32), [[COPY13]](s32)
+    ; GFX10: [[MV7:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY14]](s32), [[COPY15]](s32)
+    ; GFX10: [[MV8:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY16]](s32), [[COPY17]](s32)
+    ; GFX10: [[BUILD_VECTOR2:%[0-9]+]]:_(<3 x s64>) = G_BUILD_VECTOR [[MV6]](s64), [[MV7]](s64), [[MV8]](s64)
+    ; GFX10: [[COPY18:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX10: [[FMUL:%[0-9]+]]:_(<3 x s64>) = G_FMUL [[BUILD_VECTOR]], [[BUILD_VECTOR1]]
+    ; GFX10: [[FADD:%[0-9]+]]:_(<3 x s64>) = G_FADD [[BUILD_VECTOR2]], [[FMUL]]
+    ; GFX10: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32), [[UV4:%[0-9]+]]:_(s32), [[UV5:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FADD]](<3 x s64>)
+    ; GFX10: $vgpr0 = COPY [[UV]](s32)
+    ; GFX10: $vgpr1 = COPY [[UV1]](s32)
+    ; GFX10: $vgpr2 = COPY [[UV2]](s32)
+    ; GFX10: $vgpr3 = COPY [[UV3]](s32)
+    ; GFX10: $vgpr4 = COPY [[UV4]](s32)
+    ; GFX10: $vgpr5 = COPY [[UV5]](s32)
+    ; GFX10: [[COPY19:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY18]]
+    ; GFX10: S_SETPC_B64_return [[COPY19]], implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3, implicit $vgpr4, implicit $vgpr5
+    ; GFX10-CONTRACT-LABEL: name: test_3xdouble_add_mul_rhs
+    ; GFX10-CONTRACT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX10-CONTRACT: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX10-CONTRACT: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX10-CONTRACT: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3
+    ; GFX10-CONTRACT: [[COPY4:%[0-9]+]]:_(s32) = COPY $vgpr4
+    ; GFX10-CONTRACT: [[COPY5:%[0-9]+]]:_(s32) = COPY $vgpr5
+    ; GFX10-CONTRACT: [[MV:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY]](s32), [[COPY1]](s32)
+    ; GFX10-CONTRACT: [[MV1:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY2]](s32), [[COPY3]](s32)
+    ; GFX10-CONTRACT: [[MV2:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY4]](s32), [[COPY5]](s32)
+    ; GFX10-CONTRACT: [[BUILD_VECTOR:%[0-9]+]]:_(<3 x s64>) = G_BUILD_VECTOR [[MV]](s64), [[MV1]](s64), [[MV2]](s64)
+    ; GFX10-CONTRACT: [[COPY6:%[0-9]+]]:_(s32) = COPY $vgpr6
+    ; GFX10-CONTRACT: [[COPY7:%[0-9]+]]:_(s32) = COPY $vgpr7
+    ; GFX10-CONTRACT: [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr8
+    ; GFX10-CONTRACT: [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr9
+    ; GFX10-CONTRACT: [[COPY10:%[0-9]+]]:_(s32) = COPY $vgpr10
+    ; GFX10-CONTRACT: [[COPY11:%[0-9]+]]:_(s32) = COPY $vgpr11
+    ; GFX10-CONTRACT: [[MV3:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY6]](s32), [[COPY7]](s32)
+    ; GFX10-CONTRACT: [[MV4:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY8]](s32), [[COPY9]](s32)
+    ; GFX10-CONTRACT: [[MV5:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY10]](s32), [[COPY11]](s32)
+    ; GFX10-CONTRACT: [[BUILD_VECTOR1:%[0-9]+]]:_(<3 x s64>) = G_BUILD_VECTOR [[MV3]](s64), [[MV4]](s64), [[MV5]](s64)
+    ; GFX10-CONTRACT: [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr12
+    ; GFX10-CONTRACT: [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr13
+    ; GFX10-CONTRACT: [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr14
+    ; GFX10-CONTRACT: [[COPY15:%[0-9]+]]:_(s32) = COPY $vgpr15
+    ; GFX10-CONTRACT: [[COPY16:%[0-9]+]]:_(s32) = COPY $vgpr16
+    ; GFX10-CONTRACT: [[COPY17:%[0-9]+]]:_(s32) = COPY $vgpr17
+    ; GFX10-CONTRACT: [[MV6:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY12]](s32), [[COPY13]](s32)
+    ; GFX10-CONTRACT: [[MV7:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY14]](s32), [[COPY15]](s32)
+    ; GFX10-CONTRACT: [[MV8:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY16]](s32), [[COPY17]](s32)
+    ; GFX10-CONTRACT: [[BUILD_VECTOR2:%[0-9]+]]:_(<3 x s64>) = G_BUILD_VECTOR [[MV6]](s64), [[MV7]](s64), [[MV8]](s64)
+    ; GFX10-CONTRACT: [[COPY18:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX10-CONTRACT: [[FMUL:%[0-9]+]]:_(<3 x s64>) = G_FMUL [[BUILD_VECTOR]], [[BUILD_VECTOR1]]
+    ; GFX10-CONTRACT: [[FADD:%[0-9]+]]:_(<3 x s64>) = G_FADD [[BUILD_VECTOR2]], [[FMUL]]
+    ; GFX10-CONTRACT: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32), [[UV4:%[0-9]+]]:_(s32), [[UV5:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FADD]](<3 x s64>)
+    ; GFX10-CONTRACT: $vgpr0 = COPY [[UV]](s32)
+    ; GFX10-CONTRACT: $vgpr1 = COPY [[UV1]](s32)
+    ; GFX10-CONTRACT: $vgpr2 = COPY [[UV2]](s32)
+    ; GFX10-CONTRACT: $vgpr3 = COPY [[UV3]](s32)
+    ; GFX10-CONTRACT: $vgpr4 = COPY [[UV4]](s32)
+    ; GFX10-CONTRACT: $vgpr5 = COPY [[UV5]](s32)
+    ; GFX10-CONTRACT: [[COPY19:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY18]]
+    ; GFX10-CONTRACT: S_SETPC_B64_return [[COPY19]], implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3, implicit $vgpr4, implicit $vgpr5
+    ; GFX10-DENORM-LABEL: name: test_3xdouble_add_mul_rhs
+    ; GFX10-DENORM: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX10-DENORM: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX10-DENORM: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX10-DENORM: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3
+    ; GFX10-DENORM: [[COPY4:%[0-9]+]]:_(s32) = COPY $vgpr4
+    ; GFX10-DENORM: [[COPY5:%[0-9]+]]:_(s32) = COPY $vgpr5
+    ; GFX10-DENORM: [[MV:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY]](s32), [[COPY1]](s32)
+    ; GFX10-DENORM: [[MV1:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY2]](s32), [[COPY3]](s32)
+    ; GFX10-DENORM: [[MV2:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY4]](s32), [[COPY5]](s32)
+    ; GFX10-DENORM: [[BUILD_VECTOR:%[0-9]+]]:_(<3 x s64>) = G_BUILD_VECTOR [[MV]](s64), [[MV1]](s64), [[MV2]](s64)
+    ; GFX10-DENORM: [[COPY6:%[0-9]+]]:_(s32) = COPY $vgpr6
+    ; GFX10-DENORM: [[COPY7:%[0-9]+]]:_(s32) = COPY $vgpr7
+    ; GFX10-DENORM: [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr8
+    ; GFX10-DENORM: [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr9
+    ; GFX10-DENORM: [[COPY10:%[0-9]+]]:_(s32) = COPY $vgpr10
+    ; GFX10-DENORM: [[COPY11:%[0-9]+]]:_(s32) = COPY $vgpr11
+    ; GFX10-DENORM: [[MV3:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY6]](s32), [[COPY7]](s32)
+    ; GFX10-DENORM: [[MV4:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY8]](s32), [[COPY9]](s32)
+    ; GFX10-DENORM: [[MV5:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY10]](s32), [[COPY11]](s32)
+    ; GFX10-DENORM: [[BUILD_VECTOR1:%[0-9]+]]:_(<3 x s64>) = G_BUILD_VECTOR [[MV3]](s64), [[MV4]](s64), [[MV5]](s64)
+    ; GFX10-DENORM: [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr12
+    ; GFX10-DENORM: [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr13
+    ; GFX10-DENORM: [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr14
+    ; GFX10-DENORM: [[COPY15:%[0-9]+]]:_(s32) = COPY $vgpr15
+    ; GFX10-DENORM: [[COPY16:%[0-9]+]]:_(s32) = COPY $vgpr16
+    ; GFX10-DENORM: [[COPY17:%[0-9]+]]:_(s32) = COPY $vgpr17
+    ; GFX10-DENORM: [[MV6:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY12]](s32), [[COPY13]](s32)
+    ; GFX10-DENORM: [[MV7:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY14]](s32), [[COPY15]](s32)
+    ; GFX10-DENORM: [[MV8:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY16]](s32), [[COPY17]](s32)
+    ; GFX10-DENORM: [[BUILD_VECTOR2:%[0-9]+]]:_(<3 x s64>) = G_BUILD_VECTOR [[MV6]](s64), [[MV7]](s64), [[MV8]](s64)
+    ; GFX10-DENORM: [[COPY18:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX10-DENORM: [[FMUL:%[0-9]+]]:_(<3 x s64>) = G_FMUL [[BUILD_VECTOR]], [[BUILD_VECTOR1]]
+    ; GFX10-DENORM: [[FADD:%[0-9]+]]:_(<3 x s64>) = G_FADD [[BUILD_VECTOR2]], [[FMUL]]
+    ; GFX10-DENORM: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32), [[UV4:%[0-9]+]]:_(s32), [[UV5:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FADD]](<3 x s64>)
+    ; GFX10-DENORM: $vgpr0 = COPY [[UV]](s32)
+    ; GFX10-DENORM: $vgpr1 = COPY [[UV1]](s32)
+    ; GFX10-DENORM: $vgpr2 = COPY [[UV2]](s32)
+    ; GFX10-DENORM: $vgpr3 = COPY [[UV3]](s32)
+    ; GFX10-DENORM: $vgpr4 = COPY [[UV4]](s32)
+    ; GFX10-DENORM: $vgpr5 = COPY [[UV5]](s32)
+    ; GFX10-DENORM: [[COPY19:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY18]]
+    ; GFX10-DENORM: S_SETPC_B64_return [[COPY19]], implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3, implicit $vgpr4, implicit $vgpr5
+    ; GFX10-UNSAFE-LABEL: name: test_3xdouble_add_mul_rhs
+    ; GFX10-UNSAFE: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX10-UNSAFE: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX10-UNSAFE: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX10-UNSAFE: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3
+    ; GFX10-UNSAFE: [[COPY4:%[0-9]+]]:_(s32) = COPY $vgpr4
+    ; GFX10-UNSAFE: [[COPY5:%[0-9]+]]:_(s32) = COPY $vgpr5
+    ; GFX10-UNSAFE: [[MV:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY]](s32), [[COPY1]](s32)
+    ; GFX10-UNSAFE: [[MV1:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY2]](s32), [[COPY3]](s32)
+    ; GFX10-UNSAFE: [[MV2:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY4]](s32), [[COPY5]](s32)
+    ; GFX10-UNSAFE: [[BUILD_VECTOR:%[0-9]+]]:_(<3 x s64>) = G_BUILD_VECTOR [[MV]](s64), [[MV1]](s64), [[MV2]](s64)
+    ; GFX10-UNSAFE: [[COPY6:%[0-9]+]]:_(s32) = COPY $vgpr6
+    ; GFX10-UNSAFE: [[COPY7:%[0-9]+]]:_(s32) = COPY $vgpr7
+    ; GFX10-UNSAFE: [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr8
+    ; GFX10-UNSAFE: [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr9
+    ; GFX10-UNSAFE: [[COPY10:%[0-9]+]]:_(s32) = COPY $vgpr10
+    ; GFX10-UNSAFE: [[COPY11:%[0-9]+]]:_(s32) = COPY $vgpr11
+    ; GFX10-UNSAFE: [[MV3:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY6]](s32), [[COPY7]](s32)
+    ; GFX10-UNSAFE: [[MV4:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY8]](s32), [[COPY9]](s32)
+    ; GFX10-UNSAFE: [[MV5:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY10]](s32), [[COPY11]](s32)
+    ; GFX10-UNSAFE: [[BUILD_VECTOR1:%[0-9]+]]:_(<3 x s64>) = G_BUILD_VECTOR [[MV3]](s64), [[MV4]](s64), [[MV5]](s64)
+    ; GFX10-UNSAFE: [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr12
+    ; GFX10-UNSAFE: [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr13
+    ; GFX10-UNSAFE: [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr14
+    ; GFX10-UNSAFE: [[COPY15:%[0-9]+]]:_(s32) = COPY $vgpr15
+    ; GFX10-UNSAFE: [[COPY16:%[0-9]+]]:_(s32) = COPY $vgpr16
+    ; GFX10-UNSAFE: [[COPY17:%[0-9]+]]:_(s32) = COPY $vgpr17
+    ; GFX10-UNSAFE: [[MV6:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY12]](s32), [[COPY13]](s32)
+    ; GFX10-UNSAFE: [[MV7:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY14]](s32), [[COPY15]](s32)
+    ; GFX10-UNSAFE: [[MV8:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY16]](s32), [[COPY17]](s32)
+    ; GFX10-UNSAFE: [[BUILD_VECTOR2:%[0-9]+]]:_(<3 x s64>) = G_BUILD_VECTOR [[MV6]](s64), [[MV7]](s64), [[MV8]](s64)
+    ; GFX10-UNSAFE: [[COPY18:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX10-UNSAFE: [[FMUL:%[0-9]+]]:_(<3 x s64>) = G_FMUL [[BUILD_VECTOR]], [[BUILD_VECTOR1]]
+    ; GFX10-UNSAFE: [[FADD:%[0-9]+]]:_(<3 x s64>) = G_FADD [[BUILD_VECTOR2]], [[FMUL]]
+    ; GFX10-UNSAFE: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32), [[UV4:%[0-9]+]]:_(s32), [[UV5:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FADD]](<3 x s64>)
+    ; GFX10-UNSAFE: $vgpr0 = COPY [[UV]](s32)
+    ; GFX10-UNSAFE: $vgpr1 = COPY [[UV1]](s32)
+    ; GFX10-UNSAFE: $vgpr2 = COPY [[UV2]](s32)
+    ; GFX10-UNSAFE: $vgpr3 = COPY [[UV3]](s32)
+    ; GFX10-UNSAFE: $vgpr4 = COPY [[UV4]](s32)
+    ; GFX10-UNSAFE: $vgpr5 = COPY [[UV5]](s32)
+    ; GFX10-UNSAFE: [[COPY19:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY18]]
+    ; GFX10-UNSAFE: S_SETPC_B64_return [[COPY19]], implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3, implicit $vgpr4, implicit $vgpr5
+    %4:_(s32) = COPY $vgpr0
+    %5:_(s32) = COPY $vgpr1
+    %6:_(s32) = COPY $vgpr2
+    %7:_(s32) = COPY $vgpr3
+    %8:_(s32) = COPY $vgpr4
+    %9:_(s32) = COPY $vgpr5
+    %22:_(s64) = G_MERGE_VALUES %4(s32), %5(s32)
+    %23:_(s64) = G_MERGE_VALUES %6(s32), %7(s32)
+    %24:_(s64) = G_MERGE_VALUES %8(s32), %9(s32)
+    %0:_(<3 x s64>) = G_BUILD_VECTOR %22(s64), %23(s64), %24(s64)
+    %10:_(s32) = COPY $vgpr6
+    %11:_(s32) = COPY $vgpr7
+    %12:_(s32) = COPY $vgpr8
+    %13:_(s32) = COPY $vgpr9
+    %14:_(s32) = COPY $vgpr10
+    %15:_(s32) = COPY $vgpr11
+    %25:_(s64) = G_MERGE_VALUES %10(s32), %11(s32)
+    %26:_(s64) = G_MERGE_VALUES %12(s32), %13(s32)
+    %27:_(s64) = G_MERGE_VALUES %14(s32), %15(s32)
+    %1:_(<3 x s64>) = G_BUILD_VECTOR %25(s64), %26(s64), %27(s64)
+    %16:_(s32) = COPY $vgpr12
+    %17:_(s32) = COPY $vgpr13
+    %18:_(s32) = COPY $vgpr14
+    %19:_(s32) = COPY $vgpr15
+    %20:_(s32) = COPY $vgpr16
+    %21:_(s32) = COPY $vgpr17
+    %28:_(s64) = G_MERGE_VALUES %16(s32), %17(s32)
+    %29:_(s64) = G_MERGE_VALUES %18(s32), %19(s32)
+    %30:_(s64) = G_MERGE_VALUES %20(s32), %21(s32)
+    %2:_(<3 x s64>) = G_BUILD_VECTOR %28(s64), %29(s64), %30(s64)
+    %3:sgpr_64 = COPY $sgpr30_sgpr31
+    %31:_(<3 x s64>) = G_FMUL %0, %1
+    %32:_(<3 x s64>) = G_FADD %2, %31
+    %34:_(s32), %35:_(s32), %36:_(s32), %37:_(s32), %38:_(s32), %39:_(s32) = G_UNMERGE_VALUES %32(<3 x s64>)
+    $vgpr0 = COPY %34(s32)
+    $vgpr1 = COPY %35(s32)
+    $vgpr2 = COPY %36(s32)
+    $vgpr3 = COPY %37(s32)
+    $vgpr4 = COPY %38(s32)
+    $vgpr5 = COPY %39(s32)
+    %33:ccr_sgpr_64 = COPY %3
+    S_SETPC_B64_return %33, implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3, implicit $vgpr4, implicit $vgpr5
+...

llvm/test/CodeGen/AMDGPU/GlobalISel/combine-fma-add-mul-pre-legalize.mir

+# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
+# RUN: llc -global-isel -march=amdgcn -mcpu=gfx900 -run-pass=amdgpu-prelegalizer-combiner %s -o - | FileCheck -check-prefix=GFX9 %s
+# RUN: llc -global-isel -march=amdgcn -mcpu=gfx900 -run-pass=amdgpu-prelegalizer-combiner -fp-contract=fast %s -o - | FileCheck -check-prefix=GFX9-CONTRACT %s
+# RUN: llc -global-isel -march=amdgcn -mcpu=gfx900 -run-pass=amdgpu-prelegalizer-combiner --denormal-fp-math=preserve-sign %s -o - | FileCheck -check-prefix=GFX9-DENORM %s
+# RUN: llc -global-isel -march=amdgcn -mcpu=gfx900 -run-pass=amdgpu-prelegalizer-combiner -enable-unsafe-fp-math %s -o - | FileCheck -check-prefix=GFX9-UNSAFE %s
+# RUN: llc -global-isel -march=amdgcn -mcpu=gfx1010 -run-pass=amdgpu-prelegalizer-combiner %s -o - | FileCheck -check-prefix=GFX10 %s
+# RUN: llc -global-isel -march=amdgcn -mcpu=gfx1010 -run-pass=amdgpu-prelegalizer-combiner -fp-contract=fast %s -o - | FileCheck -check-prefix=GFX10-CONTRACT %s
+# RUN: llc -global-isel -march=amdgcn -mcpu=gfx1010 -run-pass=amdgpu-prelegalizer-combiner --denormal-fp-math=preserve-sign %s -o - | FileCheck -check-prefix=GFX10-DENORM %s
+# RUN: llc -global-isel -march=amdgcn -mcpu=gfx1010 -run-pass=amdgpu-prelegalizer-combiner -enable-unsafe-fp-math %s -o - | FileCheck -check-prefix=GFX10-UNSAFE %s
+
+---
+name:            test_f32_add_mul
+body:             |
+  bb.1.entry:
+    liveins: $vgpr0, $vgpr1, $vgpr2, $sgpr30_sgpr31
+
+    ; GFX9-LABEL: name: test_f32_add_mul
+    ; GFX9: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX9: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX9: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX9: [[COPY3:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX9: [[FMUL:%[0-9]+]]:_(s32) = reassoc G_FMUL [[COPY]], [[COPY1]]
+    ; GFX9: [[FADD:%[0-9]+]]:_(s32) = reassoc G_FADD [[FMUL]], [[COPY2]]
+    ; GFX9: $vgpr0 = COPY [[FADD]](s32)
+    ; GFX9: [[COPY4:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY3]]
+    ; GFX9: S_SETPC_B64_return [[COPY4]], implicit $vgpr0
+    ; GFX9-CONTRACT-LABEL: name: test_f32_add_mul
+    ; GFX9-CONTRACT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX9-CONTRACT: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX9-CONTRACT: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX9-CONTRACT: [[COPY3:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX9-CONTRACT: [[FMA:%[0-9]+]]:_(s32) = G_FMA [[COPY]], [[COPY1]], [[COPY2]]
+    ; GFX9-CONTRACT: $vgpr0 = COPY [[FMA]](s32)
+    ; GFX9-CONTRACT: [[COPY4:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY3]]
+    ; GFX9-CONTRACT: S_SETPC_B64_return [[COPY4]], implicit $vgpr0
+    ; GFX9-DENORM-LABEL: name: test_f32_add_mul
+    ; GFX9-DENORM: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX9-DENORM: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX9-DENORM: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX9-DENORM: [[COPY3:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX9-DENORM: [[FMUL:%[0-9]+]]:_(s32) = reassoc G_FMUL [[COPY]], [[COPY1]]
+    ; GFX9-DENORM: [[FADD:%[0-9]+]]:_(s32) = reassoc G_FADD [[FMUL]], [[COPY2]]
+    ; GFX9-DENORM: $vgpr0 = COPY [[FADD]](s32)
+    ; GFX9-DENORM: [[COPY4:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY3]]
+    ; GFX9-DENORM: S_SETPC_B64_return [[COPY4]], implicit $vgpr0
+    ; GFX9-UNSAFE-LABEL: name: test_f32_add_mul
+    ; GFX9-UNSAFE: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX9-UNSAFE: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX9-UNSAFE: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX9-UNSAFE: [[COPY3:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX9-UNSAFE: [[FMA:%[0-9]+]]:_(s32) = G_FMA [[COPY]], [[COPY1]], [[COPY2]]
+    ; GFX9-UNSAFE: $vgpr0 = COPY [[FMA]](s32)
+    ; GFX9-UNSAFE: [[COPY4:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY3]]
+    ; GFX9-UNSAFE: S_SETPC_B64_return [[COPY4]], implicit $vgpr0
+    ; GFX10-LABEL: name: test_f32_add_mul
+    ; GFX10: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX10: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX10: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX10: [[COPY3:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX10: [[FMUL:%[0-9]+]]:_(s32) = reassoc G_FMUL [[COPY]], [[COPY1]]
+    ; GFX10: [[FADD:%[0-9]+]]:_(s32) = reassoc G_FADD [[FMUL]], [[COPY2]]
+    ; GFX10: $vgpr0 = COPY [[FADD]](s32)
+    ; GFX10: [[COPY4:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY3]]
+    ; GFX10: S_SETPC_B64_return [[COPY4]], implicit $vgpr0
+    ; GFX10-CONTRACT-LABEL: name: test_f32_add_mul
+    ; GFX10-CONTRACT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX10-CONTRACT: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX10-CONTRACT: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX10-CONTRACT: [[COPY3:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX10-CONTRACT: [[FMA:%[0-9]+]]:_(s32) = G_FMA [[COPY]], [[COPY1]], [[COPY2]]
+    ; GFX10-CONTRACT: $vgpr0 = COPY [[FMA]](s32)
+    ; GFX10-CONTRACT: [[COPY4:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY3]]
+    ; GFX10-CONTRACT: S_SETPC_B64_return [[COPY4]], implicit $vgpr0
+    ; GFX10-DENORM-LABEL: name: test_f32_add_mul
+    ; GFX10-DENORM: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX10-DENORM: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX10-DENORM: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX10-DENORM: [[COPY3:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX10-DENORM: [[FMUL:%[0-9]+]]:_(s32) = reassoc G_FMUL [[COPY]], [[COPY1]]
+    ; GFX10-DENORM: [[FADD:%[0-9]+]]:_(s32) = reassoc G_FADD [[FMUL]], [[COPY2]]
+    ; GFX10-DENORM: $vgpr0 = COPY [[FADD]](s32)
+    ; GFX10-DENORM: [[COPY4:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY3]]
+    ; GFX10-DENORM: S_SETPC_B64_return [[COPY4]], implicit $vgpr0
+    ; GFX10-UNSAFE-LABEL: name: test_f32_add_mul
+    ; GFX10-UNSAFE: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX10-UNSAFE: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX10-UNSAFE: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX10-UNSAFE: [[COPY3:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX10-UNSAFE: [[FMA:%[0-9]+]]:_(s32) = G_FMA [[COPY]], [[COPY1]], [[COPY2]]
+    ; GFX10-UNSAFE: $vgpr0 = COPY [[FMA]](s32)
+    ; GFX10-UNSAFE: [[COPY4:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY3]]
+    ; GFX10-UNSAFE: S_SETPC_B64_return [[COPY4]], implicit $vgpr0
+    %0:_(s32) = COPY $vgpr0
+    %1:_(s32) = COPY $vgpr1
+    %2:_(s32) = COPY $vgpr2
+    %3:sgpr_64 = COPY $sgpr30_sgpr31
+    %4:_(s32) = reassoc G_FMUL %0, %1
+    %5:_(s32) = reassoc G_FADD %4, %2
+    $vgpr0 = COPY %5(s32)
+    %6:ccr_sgpr_64 = COPY %3
+    S_SETPC_B64_return %6, implicit $vgpr0
+...
+
+---
+name:            test_f32_add_mul_rhs
+body:             |
+  bb.1.entry:
+    liveins: $vgpr0, $vgpr1, $vgpr2, $sgpr30_sgpr31
+
+    ; GFX9-LABEL: name: test_f32_add_mul_rhs
+    ; GFX9: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX9: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX9: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX9: [[COPY3:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX9: [[FMUL:%[0-9]+]]:_(s32) = reassoc G_FMUL [[COPY]], [[COPY1]]
+    ; GFX9: [[FADD:%[0-9]+]]:_(s32) = reassoc G_FADD [[COPY2]], [[FMUL]]
+    ; GFX9: $vgpr0 = COPY [[FADD]](s32)
+    ; GFX9: [[COPY4:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY3]]
+    ; GFX9: S_SETPC_B64_return [[COPY4]], implicit $vgpr0
+    ; GFX9-CONTRACT-LABEL: name: test_f32_add_mul_rhs
+    ; GFX9-CONTRACT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX9-CONTRACT: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX9-CONTRACT: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX9-CONTRACT: [[COPY3:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX9-CONTRACT: [[FMA:%[0-9]+]]:_(s32) = G_FMA [[COPY]], [[COPY1]], [[COPY2]]
+    ; GFX9-CONTRACT: $vgpr0 = COPY [[FMA]](s32)
+    ; GFX9-CONTRACT: [[COPY4:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY3]]
+    ; GFX9-CONTRACT: S_SETPC_B64_return [[COPY4]], implicit $vgpr0
+    ; GFX9-DENORM-LABEL: name: test_f32_add_mul_rhs
+    ; GFX9-DENORM: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX9-DENORM: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX9-DENORM: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX9-DENORM: [[COPY3:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX9-DENORM: [[FMUL:%[0-9]+]]:_(s32) = reassoc G_FMUL [[COPY]], [[COPY1]]
+    ; GFX9-DENORM: [[FADD:%[0-9]+]]:_(s32) = reassoc G_FADD [[COPY2]], [[FMUL]]
+    ; GFX9-DENORM: $vgpr0 = COPY [[FADD]](s32)
+    ; GFX9-DENORM: [[COPY4:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY3]]
+    ; GFX9-DENORM: S_SETPC_B64_return [[COPY4]], implicit $vgpr0
+    ; GFX9-UNSAFE-LABEL: name: test_f32_add_mul_rhs
+    ; GFX9-UNSAFE: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX9-UNSAFE: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX9-UNSAFE: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX9-UNSAFE: [[COPY3:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX9-UNSAFE: [[FMA:%[0-9]+]]:_(s32) = G_FMA [[COPY]], [[COPY1]], [[COPY2]]
+    ; GFX9-UNSAFE: $vgpr0 = COPY [[FMA]](s32)
+    ; GFX9-UNSAFE: [[COPY4:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY3]]
+    ; GFX9-UNSAFE: S_SETPC_B64_return [[COPY4]], implicit $vgpr0
+    ; GFX10-LABEL: name: test_f32_add_mul_rhs
+    ; GFX10: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX10: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX10: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX10: [[COPY3:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX10: [[FMUL:%[0-9]+]]:_(s32) = reassoc G_FMUL [[COPY]], [[COPY1]]
+    ; GFX10: [[FADD:%[0-9]+]]:_(s32) = reassoc G_FADD [[COPY2]], [[FMUL]]
+    ; GFX10: $vgpr0 = COPY [[FADD]](s32)
+    ; GFX10: [[COPY4:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY3]]
+    ; GFX10: S_SETPC_B64_return [[COPY4]], implicit $vgpr0
+    ; GFX10-CONTRACT-LABEL: name: test_f32_add_mul_rhs
+    ; GFX10-CONTRACT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX10-CONTRACT: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX10-CONTRACT: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX10-CONTRACT: [[COPY3:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX10-CONTRACT: [[FMA:%[0-9]+]]:_(s32) = G_FMA [[COPY]], [[COPY1]], [[COPY2]]
+    ; GFX10-CONTRACT: $vgpr0 = COPY [[FMA]](s32)
+    ; GFX10-CONTRACT: [[COPY4:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY3]]
+    ; GFX10-CONTRACT: S_SETPC_B64_return [[COPY4]], implicit $vgpr0
+    ; GFX10-DENORM-LABEL: name: test_f32_add_mul_rhs
+    ; GFX10-DENORM: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX10-DENORM: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX10-DENORM: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX10-DENORM: [[COPY3:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX10-DENORM: [[FMUL:%[0-9]+]]:_(s32) = reassoc G_FMUL [[COPY]], [[COPY1]]
+    ; GFX10-DENORM: [[FADD:%[0-9]+]]:_(s32) = reassoc G_FADD [[COPY2]], [[FMUL]]
+    ; GFX10-DENORM: $vgpr0 = COPY [[FADD]](s32)
+    ; GFX10-DENORM: [[COPY4:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY3]]
+    ; GFX10-DENORM: S_SETPC_B64_return [[COPY4]], implicit $vgpr0
+    ; GFX10-UNSAFE-LABEL: name: test_f32_add_mul_rhs
+    ; GFX10-UNSAFE: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX10-UNSAFE: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX10-UNSAFE: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX10-UNSAFE: [[COPY3:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX10-UNSAFE: [[FMA:%[0-9]+]]:_(s32) = G_FMA [[COPY]], [[COPY1]], [[COPY2]]
+    ; GFX10-UNSAFE: $vgpr0 = COPY [[FMA]](s32)
+    ; GFX10-UNSAFE: [[COPY4:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY3]]
+    ; GFX10-UNSAFE: S_SETPC_B64_return [[COPY4]], implicit $vgpr0
+    %0:_(s32) = COPY $vgpr0
+    %1:_(s32) = COPY $vgpr1
+    %2:_(s32) = COPY $vgpr2
+    %3:sgpr_64 = COPY $sgpr30_sgpr31
+    %4:_(s32) = reassoc G_FMUL %0, %1
+    %5:_(s32) = reassoc G_FADD %2, %4
+    $vgpr0 = COPY %5(s32)
+    %6:ccr_sgpr_64 = COPY %3
+    S_SETPC_B64_return %6, implicit $vgpr0
+...
+
+---
+name:            test_half_add_mul
+body:             |
+  bb.1.entry:
+    liveins: $vgpr0, $vgpr1, $vgpr2, $sgpr30_sgpr31
+
+    ; GFX9-LABEL: name: test_half_add_mul
+    ; GFX9: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX9: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY]](s32)
+    ; GFX9: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX9: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[COPY1]](s32)
+    ; GFX9: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX9: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[COPY2]](s32)
+    ; GFX9: [[COPY3:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX9: [[FMUL:%[0-9]+]]:_(s16) = reassoc G_FMUL [[TRUNC]], [[TRUNC1]]
+    ; GFX9: [[FADD:%[0-9]+]]:_(s16) = reassoc G_FADD [[FMUL]], [[TRUNC2]]
+    ; GFX9: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[FADD]](s16)
+    ; GFX9: $vgpr0 = COPY [[ANYEXT]](s32)
+    ; GFX9: [[COPY4:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY3]]
+    ; GFX9: S_SETPC_B64_return [[COPY4]], implicit $vgpr0
+    ; GFX9-CONTRACT-LABEL: name: test_half_add_mul
+    ; GFX9-CONTRACT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX9-CONTRACT: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY]](s32)
+    ; GFX9-CONTRACT: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX9-CONTRACT: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[COPY1]](s32)
+    ; GFX9-CONTRACT: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX9-CONTRACT: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[COPY2]](s32)
+    ; GFX9-CONTRACT: [[COPY3:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX9-CONTRACT: [[FMA:%[0-9]+]]:_(s16) = G_FMA [[TRUNC]], [[TRUNC1]], [[TRUNC2]]
+    ; GFX9-CONTRACT: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[FMA]](s16)
+    ; GFX9-CONTRACT: $vgpr0 = COPY [[ANYEXT]](s32)
+    ; GFX9-CONTRACT: [[COPY4:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY3]]
+    ; GFX9-CONTRACT: S_SETPC_B64_return [[COPY4]], implicit $vgpr0
+    ; GFX9-DENORM-LABEL: name: test_half_add_mul
+    ; GFX9-DENORM: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX9-DENORM: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY]](s32)
+    ; GFX9-DENORM: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX9-DENORM: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[COPY1]](s32)
+    ; GFX9-DENORM: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX9-DENORM: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[COPY2]](s32)
+    ; GFX9-DENORM: [[COPY3:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX9-DENORM: [[FMUL:%[0-9]+]]:_(s16) = reassoc G_FMUL [[TRUNC]], [[TRUNC1]]
+    ; GFX9-DENORM: [[FADD:%[0-9]+]]:_(s16) = reassoc G_FADD [[FMUL]], [[TRUNC2]]
+    ; GFX9-DENORM: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[FADD]](s16)
+    ; GFX9-DENORM: $vgpr0 = COPY [[ANYEXT]](s32)
+    ; GFX9-DENORM: [[COPY4:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY3]]
+    ; GFX9-DENORM: S_SETPC_B64_return [[COPY4]], implicit $vgpr0
+    ; GFX9-UNSAFE-LABEL: name: test_half_add_mul
+    ; GFX9-UNSAFE: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX9-UNSAFE: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY]](s32)
+    ; GFX9-UNSAFE: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX9-UNSAFE: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[COPY1]](s32)
+    ; GFX9-UNSAFE: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX9-UNSAFE: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[COPY2]](s32)
+    ; GFX9-UNSAFE: [[COPY3:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX9-UNSAFE: [[FMA:%[0-9]+]]:_(s16) = G_FMA [[TRUNC]], [[TRUNC1]], [[TRUNC2]]
+    ; GFX9-UNSAFE: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[FMA]](s16)
+    ; GFX9-UNSAFE: $vgpr0 = COPY [[ANYEXT]](s32)
+    ; GFX9-UNSAFE: [[COPY4:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY3]]
+    ; GFX9-UNSAFE: S_SETPC_B64_return [[COPY4]], implicit $vgpr0
+    ; GFX10-LABEL: name: test_half_add_mul
+    ; GFX10: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX10: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY]](s32)
+    ; GFX10: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX10: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[COPY1]](s32)
+    ; GFX10: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX10: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[COPY2]](s32)
+    ; GFX10: [[COPY3:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX10: [[FMUL:%[0-9]+]]:_(s16) = reassoc G_FMUL [[TRUNC]], [[TRUNC1]]
+    ; GFX10: [[FADD:%[0-9]+]]:_(s16) = reassoc G_FADD [[FMUL]], [[TRUNC2]]
+    ; GFX10: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[FADD]](s16)
+    ; GFX10: $vgpr0 = COPY [[ANYEXT]](s32)
+    ; GFX10: [[COPY4:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY3]]
+    ; GFX10: S_SETPC_B64_return [[COPY4]], implicit $vgpr0
+    ; GFX10-CONTRACT-LABEL: name: test_half_add_mul
+    ; GFX10-CONTRACT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX10-CONTRACT: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY]](s32)
+    ; GFX10-CONTRACT: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX10-CONTRACT: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[COPY1]](s32)
+    ; GFX10-CONTRACT: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX10-CONTRACT: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[COPY2]](s32)
+    ; GFX10-CONTRACT: [[COPY3:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX10-CONTRACT: [[FMA:%[0-9]+]]:_(s16) = G_FMA [[TRUNC]], [[TRUNC1]], [[TRUNC2]]
+    ; GFX10-CONTRACT: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[FMA]](s16)
+    ; GFX10-CONTRACT: $vgpr0 = COPY [[ANYEXT]](s32)
+    ; GFX10-CONTRACT: [[COPY4:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY3]]
+    ; GFX10-CONTRACT: S_SETPC_B64_return [[COPY4]], implicit $vgpr0
+    ; GFX10-DENORM-LABEL: name: test_half_add_mul
+    ; GFX10-DENORM: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX10-DENORM: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY]](s32)
+    ; GFX10-DENORM: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX10-DENORM: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[COPY1]](s32)
+    ; GFX10-DENORM: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX10-DENORM: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[COPY2]](s32)
+    ; GFX10-DENORM: [[COPY3:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX10-DENORM: [[FMUL:%[0-9]+]]:_(s16) = reassoc G_FMUL [[TRUNC]], [[TRUNC1]]
+    ; GFX10-DENORM: [[FADD:%[0-9]+]]:_(s16) = reassoc G_FADD [[FMUL]], [[TRUNC2]]
+    ; GFX10-DENORM: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[FADD]](s16)
+    ; GFX10-DENORM: $vgpr0 = COPY [[ANYEXT]](s32)
+    ; GFX10-DENORM: [[COPY4:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY3]]
+    ; GFX10-DENORM: S_SETPC_B64_return [[COPY4]], implicit $vgpr0
+    ; GFX10-UNSAFE-LABEL: name: test_half_add_mul
+    ; GFX10-UNSAFE: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX10-UNSAFE: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY]](s32)
+    ; GFX10-UNSAFE: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX10-UNSAFE: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[COPY1]](s32)
+    ; GFX10-UNSAFE: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX10-UNSAFE: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[COPY2]](s32)
+    ; GFX10-UNSAFE: [[COPY3:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX10-UNSAFE: [[FMA:%[0-9]+]]:_(s16) = G_FMA [[TRUNC]], [[TRUNC1]], [[TRUNC2]]
+    ; GFX10-UNSAFE: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[FMA]](s16)
+    ; GFX10-UNSAFE: $vgpr0 = COPY [[ANYEXT]](s32)
+    ; GFX10-UNSAFE: [[COPY4:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY3]]
+    ; GFX10-UNSAFE: S_SETPC_B64_return [[COPY4]], implicit $vgpr0
+    %4:_(s32) = COPY $vgpr0
+    %0:_(s16) = G_TRUNC %4(s32)
+    %5:_(s32) = COPY $vgpr1
+    %1:_(s16) = G_TRUNC %5(s32)
+    %6:_(s32) = COPY $vgpr2
+    %2:_(s16) = G_TRUNC %6(s32)
+    %3:sgpr_64 = COPY $sgpr30_sgpr31
+    %7:_(s16) = reassoc G_FMUL %0, %1
+    %8:_(s16) = reassoc G_FADD %7, %2
+    %10:_(s32) = G_ANYEXT %8(s16)
+    $vgpr0 = COPY %10(s32)
+    %9:ccr_sgpr_64 = COPY %3
+    S_SETPC_B64_return %9, implicit $vgpr0
+...
+
+---
+name:            test_half_add_mul_rhs
+body:             |
+  bb.1.entry:
+    liveins: $vgpr0, $vgpr1, $vgpr2, $sgpr30_sgpr31
+
+    ; GFX9-LABEL: name: test_half_add_mul_rhs
+    ; GFX9: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX9: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY]](s32)
+    ; GFX9: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX9: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[COPY1]](s32)
+    ; GFX9: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX9: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[COPY2]](s32)
+    ; GFX9: [[COPY3:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX9: [[FMUL:%[0-9]+]]:_(s16) = reassoc G_FMUL [[TRUNC]], [[TRUNC1]]
+    ; GFX9: [[FADD:%[0-9]+]]:_(s16) = reassoc G_FADD [[TRUNC2]], [[FMUL]]
+    ; GFX9: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[FADD]](s16)
+    ; GFX9: $vgpr0 = COPY [[ANYEXT]](s32)
+    ; GFX9: [[COPY4:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY3]]
+    ; GFX9: S_SETPC_B64_return [[COPY4]], implicit $vgpr0
+    ; GFX9-CONTRACT-LABEL: name: test_half_add_mul_rhs
+    ; GFX9-CONTRACT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX9-CONTRACT: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY]](s32)
+    ; GFX9-CONTRACT: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX9-CONTRACT: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[COPY1]](s32)
+    ; GFX9-CONTRACT: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX9-CONTRACT: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[COPY2]](s32)
+    ; GFX9-CONTRACT: [[COPY3:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX9-CONTRACT: [[FMA:%[0-9]+]]:_(s16) = G_FMA [[TRUNC]], [[TRUNC1]], [[TRUNC2]]
+    ; GFX9-CONTRACT: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[FMA]](s16)
+    ; GFX9-CONTRACT: $vgpr0 = COPY [[ANYEXT]](s32)
+    ; GFX9-CONTRACT: [[COPY4:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY3]]
+    ; GFX9-CONTRACT: S_SETPC_B64_return [[COPY4]], implicit $vgpr0
+    ; GFX9-DENORM-LABEL: name: test_half_add_mul_rhs
+    ; GFX9-DENORM: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX9-DENORM: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY]](s32)
+    ; GFX9-DENORM: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX9-DENORM: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[COPY1]](s32)
+    ; GFX9-DENORM: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX9-DENORM: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[COPY2]](s32)
+    ; GFX9-DENORM: [[COPY3:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX9-DENORM: [[FMUL:%[0-9]+]]:_(s16) = reassoc G_FMUL [[TRUNC]], [[TRUNC1]]
+    ; GFX9-DENORM: [[FADD:%[0-9]+]]:_(s16) = reassoc G_FADD [[TRUNC2]], [[FMUL]]
+    ; GFX9-DENORM: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[FADD]](s16)
+    ; GFX9-DENORM: $vgpr0 = COPY [[ANYEXT]](s32)
+    ; GFX9-DENORM: [[COPY4:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY3]]
+    ; GFX9-DENORM: S_SETPC_B64_return [[COPY4]], implicit $vgpr0
+    ; GFX9-UNSAFE-LABEL: name: test_half_add_mul_rhs
+    ; GFX9-UNSAFE: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX9-UNSAFE: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY]](s32)
+    ; GFX9-UNSAFE: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX9-UNSAFE: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[COPY1]](s32)
+    ; GFX9-UNSAFE: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX9-UNSAFE: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[COPY2]](s32)
+    ; GFX9-UNSAFE: [[COPY3:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX9-UNSAFE: [[FMA:%[0-9]+]]:_(s16) = G_FMA [[TRUNC]], [[TRUNC1]], [[TRUNC2]]
+    ; GFX9-UNSAFE: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[FMA]](s16)
+    ; GFX9-UNSAFE: $vgpr0 = COPY [[ANYEXT]](s32)
+    ; GFX9-UNSAFE: [[COPY4:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY3]]
+    ; GFX9-UNSAFE: S_SETPC_B64_return [[COPY4]], implicit $vgpr0
+    ; GFX10-LABEL: name: test_half_add_mul_rhs
+    ; GFX10: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX10: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY]](s32)
+    ; GFX10: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX10: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[COPY1]](s32)
+    ; GFX10: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX10: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[COPY2]](s32)
+    ; GFX10: [[COPY3:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX10: [[FMUL:%[0-9]+]]:_(s16) = reassoc G_FMUL [[TRUNC]], [[TRUNC1]]
+    ; GFX10: [[FADD:%[0-9]+]]:_(s16) = reassoc G_FADD [[TRUNC2]], [[FMUL]]
+    ; GFX10: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[FADD]](s16)
+    ; GFX10: $vgpr0 = COPY [[ANYEXT]](s32)
+    ; GFX10: [[COPY4:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY3]]
+    ; GFX10: S_SETPC_B64_return [[COPY4]], implicit $vgpr0
+    ; GFX10-CONTRACT-LABEL: name: test_half_add_mul_rhs
+    ; GFX10-CONTRACT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX10-CONTRACT: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY]](s32)
+    ; GFX10-CONTRACT: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX10-CONTRACT: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[COPY1]](s32)
+    ; GFX10-CONTRACT: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX10-CONTRACT: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[COPY2]](s32)
+    ; GFX10-CONTRACT: [[COPY3:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX10-CONTRACT: [[FMA:%[0-9]+]]:_(s16) = G_FMA [[TRUNC]], [[TRUNC1]], [[TRUNC2]]
+    ; GFX10-CONTRACT: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[FMA]](s16)
+    ; GFX10-CONTRACT: $vgpr0 = COPY [[ANYEXT]](s32)
+    ; GFX10-CONTRACT: [[COPY4:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY3]]
+    ; GFX10-CONTRACT: S_SETPC_B64_return [[COPY4]], implicit $vgpr0
+    ; GFX10-DENORM-LABEL: name: test_half_add_mul_rhs
+    ; GFX10-DENORM: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX10-DENORM: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY]](s32)
+    ; GFX10-DENORM: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX10-DENORM: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[COPY1]](s32)
+    ; GFX10-DENORM: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX10-DENORM: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[COPY2]](s32)
+    ; GFX10-DENORM: [[COPY3:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX10-DENORM: [[FMUL:%[0-9]+]]:_(s16) = reassoc G_FMUL [[TRUNC]], [[TRUNC1]]
+    ; GFX10-DENORM: [[FADD:%[0-9]+]]:_(s16) = reassoc G_FADD [[TRUNC2]], [[FMUL]]
+    ; GFX10-DENORM: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[FADD]](s16)
+    ; GFX10-DENORM: $vgpr0 = COPY [[ANYEXT]](s32)
+    ; GFX10-DENORM: [[COPY4:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY3]]
+    ; GFX10-DENORM: S_SETPC_B64_return [[COPY4]], implicit $vgpr0
+    ; GFX10-UNSAFE-LABEL: name: test_half_add_mul_rhs
+    ; GFX10-UNSAFE: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX10-UNSAFE: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY]](s32)
+    ; GFX10-UNSAFE: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX10-UNSAFE: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[COPY1]](s32)
+    ; GFX10-UNSAFE: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX10-UNSAFE: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[COPY2]](s32)
+    ; GFX10-UNSAFE: [[COPY3:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX10-UNSAFE: [[FMA:%[0-9]+]]:_(s16) = G_FMA [[TRUNC]], [[TRUNC1]], [[TRUNC2]]
+    ; GFX10-UNSAFE: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[FMA]](s16)
+    ; GFX10-UNSAFE: $vgpr0 = COPY [[ANYEXT]](s32)
+    ; GFX10-UNSAFE: [[COPY4:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY3]]
+    ; GFX10-UNSAFE: S_SETPC_B64_return [[COPY4]], implicit $vgpr0
+    %4:_(s32) = COPY $vgpr0
+    %0:_(s16) = G_TRUNC %4(s32)
+    %5:_(s32) = COPY $vgpr1
+    %1:_(s16) = G_TRUNC %5(s32)
+    %6:_(s32) = COPY $vgpr2
+    %2:_(s16) = G_TRUNC %6(s32)
+    %3:sgpr_64 = COPY $sgpr30_sgpr31
+    %7:_(s16) = reassoc G_FMUL %0, %1
+    %8:_(s16) = reassoc G_FADD %2, %7
+    %10:_(s32) = G_ANYEXT %8(s16)
+    $vgpr0 = COPY %10(s32)
+    %9:ccr_sgpr_64 = COPY %3
+    S_SETPC_B64_return %9, implicit $vgpr0
+...
+
+---
+name:            test_double_add_mul
+body:             |
+  bb.1.entry:
+    liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $sgpr30_sgpr31
+
+    ; GFX9-LABEL: name: test_double_add_mul
+    ; GFX9: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX9: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX9: [[MV:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY]](s32), [[COPY1]](s32)
+    ; GFX9: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX9: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3
+    ; GFX9: [[MV1:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY2]](s32), [[COPY3]](s32)
+    ; GFX9: [[COPY4:%[0-9]+]]:_(s32) = COPY $vgpr4
+    ; GFX9: [[COPY5:%[0-9]+]]:_(s32) = COPY $vgpr5
+    ; GFX9: [[MV2:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY4]](s32), [[COPY5]](s32)
+    ; GFX9: [[COPY6:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX9: [[FMUL:%[0-9]+]]:_(s64) = reassoc G_FMUL [[MV]], [[MV1]]
+    ; GFX9: [[FADD:%[0-9]+]]:_(s64) = reassoc G_FADD [[FMUL]], [[MV2]]
+    ; GFX9: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FADD]](s64)
+    ; GFX9: $vgpr0 = COPY [[UV]](s32)
+    ; GFX9: $vgpr1 = COPY [[UV1]](s32)
+    ; GFX9: [[COPY7:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY6]]
+    ; GFX9: S_SETPC_B64_return [[COPY7]], implicit $vgpr0, implicit $vgpr1
+    ; GFX9-CONTRACT-LABEL: name: test_double_add_mul
+    ; GFX9-CONTRACT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX9-CONTRACT: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX9-CONTRACT: [[MV:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY]](s32), [[COPY1]](s32)
+    ; GFX9-CONTRACT: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX9-CONTRACT: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3
+    ; GFX9-CONTRACT: [[MV1:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY2]](s32), [[COPY3]](s32)
+    ; GFX9-CONTRACT: [[COPY4:%[0-9]+]]:_(s32) = COPY $vgpr4
+    ; GFX9-CONTRACT: [[COPY5:%[0-9]+]]:_(s32) = COPY $vgpr5
+    ; GFX9-CONTRACT: [[MV2:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY4]](s32), [[COPY5]](s32)
+    ; GFX9-CONTRACT: [[COPY6:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX9-CONTRACT: [[FMA:%[0-9]+]]:_(s64) = G_FMA [[MV]], [[MV1]], [[MV2]]
+    ; GFX9-CONTRACT: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FMA]](s64)
+    ; GFX9-CONTRACT: $vgpr0 = COPY [[UV]](s32)
+    ; GFX9-CONTRACT: $vgpr1 = COPY [[UV1]](s32)
+    ; GFX9-CONTRACT: [[COPY7:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY6]]
+    ; GFX9-CONTRACT: S_SETPC_B64_return [[COPY7]], implicit $vgpr0, implicit $vgpr1
+    ; GFX9-DENORM-LABEL: name: test_double_add_mul
+    ; GFX9-DENORM: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX9-DENORM: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX9-DENORM: [[MV:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY]](s32), [[COPY1]](s32)
+    ; GFX9-DENORM: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX9-DENORM: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3
+    ; GFX9-DENORM: [[MV1:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY2]](s32), [[COPY3]](s32)
+    ; GFX9-DENORM: [[COPY4:%[0-9]+]]:_(s32) = COPY $vgpr4
+    ; GFX9-DENORM: [[COPY5:%[0-9]+]]:_(s32) = COPY $vgpr5
+    ; GFX9-DENORM: [[MV2:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY4]](s32), [[COPY5]](s32)
+    ; GFX9-DENORM: [[COPY6:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX9-DENORM: [[FMUL:%[0-9]+]]:_(s64) = reassoc G_FMUL [[MV]], [[MV1]]
+    ; GFX9-DENORM: [[FADD:%[0-9]+]]:_(s64) = reassoc G_FADD [[FMUL]], [[MV2]]
+    ; GFX9-DENORM: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FADD]](s64)
+    ; GFX9-DENORM: $vgpr0 = COPY [[UV]](s32)
+    ; GFX9-DENORM: $vgpr1 = COPY [[UV1]](s32)
+    ; GFX9-DENORM: [[COPY7:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY6]]
+    ; GFX9-DENORM: S_SETPC_B64_return [[COPY7]], implicit $vgpr0, implicit $vgpr1
+    ; GFX9-UNSAFE-LABEL: name: test_double_add_mul
+    ; GFX9-UNSAFE: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX9-UNSAFE: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX9-UNSAFE: [[MV:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY]](s32), [[COPY1]](s32)
+    ; GFX9-UNSAFE: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX9-UNSAFE: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3
+    ; GFX9-UNSAFE: [[MV1:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY2]](s32), [[COPY3]](s32)
+    ; GFX9-UNSAFE: [[COPY4:%[0-9]+]]:_(s32) = COPY $vgpr4
+    ; GFX9-UNSAFE: [[COPY5:%[0-9]+]]:_(s32) = COPY $vgpr5
+    ; GFX9-UNSAFE: [[MV2:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY4]](s32), [[COPY5]](s32)
+    ; GFX9-UNSAFE: [[COPY6:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX9-UNSAFE: [[FMA:%[0-9]+]]:_(s64) = G_FMA [[MV]], [[MV1]], [[MV2]]
+    ; GFX9-UNSAFE: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FMA]](s64)
+    ; GFX9-UNSAFE: $vgpr0 = COPY [[UV]](s32)
+    ; GFX9-UNSAFE: $vgpr1 = COPY [[UV1]](s32)
+    ; GFX9-UNSAFE: [[COPY7:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY6]]
+    ; GFX9-UNSAFE: S_SETPC_B64_return [[COPY7]], implicit $vgpr0, implicit $vgpr1
+    ; GFX10-LABEL: name: test_double_add_mul
+    ; GFX10: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX10: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX10: [[MV:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY]](s32), [[COPY1]](s32)
+    ; GFX10: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX10: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3
+    ; GFX10: [[MV1:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY2]](s32), [[COPY3]](s32)
+    ; GFX10: [[COPY4:%[0-9]+]]:_(s32) = COPY $vgpr4
+    ; GFX10: [[COPY5:%[0-9]+]]:_(s32) = COPY $vgpr5
+    ; GFX10: [[MV2:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY4]](s32), [[COPY5]](s32)
+    ; GFX10: [[COPY6:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX10: [[FMUL:%[0-9]+]]:_(s64) = reassoc G_FMUL [[MV]], [[MV1]]
+    ; GFX10: [[FADD:%[0-9]+]]:_(s64) = reassoc G_FADD [[FMUL]], [[MV2]]
+    ; GFX10: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FADD]](s64)
+    ; GFX10: $vgpr0 = COPY [[UV]](s32)
+    ; GFX10: $vgpr1 = COPY [[UV1]](s32)
+    ; GFX10: [[COPY7:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY6]]
+    ; GFX10: S_SETPC_B64_return [[COPY7]], implicit $vgpr0, implicit $vgpr1
+    ; GFX10-CONTRACT-LABEL: name: test_double_add_mul
+    ; GFX10-CONTRACT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX10-CONTRACT: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX10-CONTRACT: [[MV:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY]](s32), [[COPY1]](s32)
+    ; GFX10-CONTRACT: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX10-CONTRACT: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3
+    ; GFX10-CONTRACT: [[MV1:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY2]](s32), [[COPY3]](s32)
+    ; GFX10-CONTRACT: [[COPY4:%[0-9]+]]:_(s32) = COPY $vgpr4
+    ; GFX10-CONTRACT: [[COPY5:%[0-9]+]]:_(s32) = COPY $vgpr5
+    ; GFX10-CONTRACT: [[MV2:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY4]](s32), [[COPY5]](s32)
+    ; GFX10-CONTRACT: [[COPY6:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX10-CONTRACT: [[FMA:%[0-9]+]]:_(s64) = G_FMA [[MV]], [[MV1]], [[MV2]]
+    ; GFX10-CONTRACT: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FMA]](s64)
+    ; GFX10-CONTRACT: $vgpr0 = COPY [[UV]](s32)
+    ; GFX10-CONTRACT: $vgpr1 = COPY [[UV1]](s32)
+    ; GFX10-CONTRACT: [[COPY7:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY6]]
+    ; GFX10-CONTRACT: S_SETPC_B64_return [[COPY7]], implicit $vgpr0, implicit $vgpr1
+    ; GFX10-DENORM-LABEL: name: test_double_add_mul
+    ; GFX10-DENORM: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX10-DENORM: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX10-DENORM: [[MV:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY]](s32), [[COPY1]](s32)
+    ; GFX10-DENORM: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX10-DENORM: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3
+    ; GFX10-DENORM: [[MV1:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY2]](s32), [[COPY3]](s32)
+    ; GFX10-DENORM: [[COPY4:%[0-9]+]]:_(s32) = COPY $vgpr4
+    ; GFX10-DENORM: [[COPY5:%[0-9]+]]:_(s32) = COPY $vgpr5
+    ; GFX10-DENORM: [[MV2:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY4]](s32), [[COPY5]](s32)
+    ; GFX10-DENORM: [[COPY6:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX10-DENORM: [[FMUL:%[0-9]+]]:_(s64) = reassoc G_FMUL [[MV]], [[MV1]]
+    ; GFX10-DENORM: [[FADD:%[0-9]+]]:_(s64) = reassoc G_FADD [[FMUL]], [[MV2]]
+    ; GFX10-DENORM: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FADD]](s64)
+    ; GFX10-DENORM: $vgpr0 = COPY [[UV]](s32)
+    ; GFX10-DENORM: $vgpr1 = COPY [[UV1]](s32)
+    ; GFX10-DENORM: [[COPY7:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY6]]
+    ; GFX10-DENORM: S_SETPC_B64_return [[COPY7]], implicit $vgpr0, implicit $vgpr1
+    ; GFX10-UNSAFE-LABEL: name: test_double_add_mul
+    ; GFX10-UNSAFE: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX10-UNSAFE: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX10-UNSAFE: [[MV:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY]](s32), [[COPY1]](s32)
+    ; GFX10-UNSAFE: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX10-UNSAFE: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3
+    ; GFX10-UNSAFE: [[MV1:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY2]](s32), [[COPY3]](s32)
+    ; GFX10-UNSAFE: [[COPY4:%[0-9]+]]:_(s32) = COPY $vgpr4
+    ; GFX10-UNSAFE: [[COPY5:%[0-9]+]]:_(s32) = COPY $vgpr5
+    ; GFX10-UNSAFE: [[MV2:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY4]](s32), [[COPY5]](s32)
+    ; GFX10-UNSAFE: [[COPY6:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX10-UNSAFE: [[FMA:%[0-9]+]]:_(s64) = G_FMA [[MV]], [[MV1]], [[MV2]]
+    ; GFX10-UNSAFE: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FMA]](s64)
+    ; GFX10-UNSAFE: $vgpr0 = COPY [[UV]](s32)
+    ; GFX10-UNSAFE: $vgpr1 = COPY [[UV1]](s32)
+    ; GFX10-UNSAFE: [[COPY7:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY6]]
+    ; GFX10-UNSAFE: S_SETPC_B64_return [[COPY7]], implicit $vgpr0, implicit $vgpr1
+    %4:_(s32) = COPY $vgpr0
+    %5:_(s32) = COPY $vgpr1
+    %0:_(s64) = G_MERGE_VALUES %4(s32), %5(s32)
+    %6:_(s32) = COPY $vgpr2
+    %7:_(s32) = COPY $vgpr3
+    %1:_(s64) = G_MERGE_VALUES %6(s32), %7(s32)
+    %8:_(s32) = COPY $vgpr4
+    %9:_(s32) = COPY $vgpr5
+    %2:_(s64) = G_MERGE_VALUES %8(s32), %9(s32)
+    %3:sgpr_64 = COPY $sgpr30_sgpr31
+    %10:_(s64) = reassoc G_FMUL %0, %1
+    %11:_(s64) = reassoc G_FADD %10, %2
+    %13:_(s32), %14:_(s32) = G_UNMERGE_VALUES %11(s64)
+    $vgpr0 = COPY %13(s32)
+    $vgpr1 = COPY %14(s32)
+    %12:ccr_sgpr_64 = COPY %3
+    S_SETPC_B64_return %12, implicit $vgpr0, implicit $vgpr1
+...
+
+---
+name:            test_double_add_mul_rhs
+body:             |
+  bb.1.entry:
+    liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $sgpr30_sgpr31
+
+    ; GFX9-LABEL: name: test_double_add_mul_rhs
+    ; GFX9: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX9: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX9: [[MV:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY]](s32), [[COPY1]](s32)
+    ; GFX9: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX9: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3
+    ; GFX9: [[MV1:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY2]](s32), [[COPY3]](s32)
+    ; GFX9: [[COPY4:%[0-9]+]]:_(s32) = COPY $vgpr4
+    ; GFX9: [[COPY5:%[0-9]+]]:_(s32) = COPY $vgpr5
+    ; GFX9: [[MV2:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY4]](s32), [[COPY5]](s32)
+    ; GFX9: [[COPY6:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX9: [[FMUL:%[0-9]+]]:_(s64) = reassoc G_FMUL [[MV]], [[MV1]]
+    ; GFX9: [[FADD:%[0-9]+]]:_(s64) = reassoc G_FADD [[MV2]], [[FMUL]]
+    ; GFX9: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FADD]](s64)
+    ; GFX9: $vgpr0 = COPY [[UV]](s32)
+    ; GFX9: $vgpr1 = COPY [[UV1]](s32)
+    ; GFX9: [[COPY7:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY6]]
+    ; GFX9: S_SETPC_B64_return [[COPY7]], implicit $vgpr0, implicit $vgpr1
+    ; GFX9-CONTRACT-LABEL: name: test_double_add_mul_rhs
+    ; GFX9-CONTRACT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX9-CONTRACT: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX9-CONTRACT: [[MV:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY]](s32), [[COPY1]](s32)
+    ; GFX9-CONTRACT: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX9-CONTRACT: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3
+    ; GFX9-CONTRACT: [[MV1:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY2]](s32), [[COPY3]](s32)
+    ; GFX9-CONTRACT: [[COPY4:%[0-9]+]]:_(s32) = COPY $vgpr4
+    ; GFX9-CONTRACT: [[COPY5:%[0-9]+]]:_(s32) = COPY $vgpr5
+    ; GFX9-CONTRACT: [[MV2:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY4]](s32), [[COPY5]](s32)
+    ; GFX9-CONTRACT: [[COPY6:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX9-CONTRACT: [[FMA:%[0-9]+]]:_(s64) = G_FMA [[MV]], [[MV1]], [[MV2]]
+    ; GFX9-CONTRACT: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FMA]](s64)
+    ; GFX9-CONTRACT: $vgpr0 = COPY [[UV]](s32)
+    ; GFX9-CONTRACT: $vgpr1 = COPY [[UV1]](s32)
+    ; GFX9-CONTRACT: [[COPY7:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY6]]
+    ; GFX9-CONTRACT: S_SETPC_B64_return [[COPY7]], implicit $vgpr0, implicit $vgpr1
+    ; GFX9-DENORM-LABEL: name: test_double_add_mul_rhs
+    ; GFX9-DENORM: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX9-DENORM: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX9-DENORM: [[MV:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY]](s32), [[COPY1]](s32)
+    ; GFX9-DENORM: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX9-DENORM: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3
+    ; GFX9-DENORM: [[MV1:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY2]](s32), [[COPY3]](s32)
+    ; GFX9-DENORM: [[COPY4:%[0-9]+]]:_(s32) = COPY $vgpr4
+    ; GFX9-DENORM: [[COPY5:%[0-9]+]]:_(s32) = COPY $vgpr5
+    ; GFX9-DENORM: [[MV2:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY4]](s32), [[COPY5]](s32)
+    ; GFX9-DENORM: [[COPY6:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX9-DENORM: [[FMUL:%[0-9]+]]:_(s64) = reassoc G_FMUL [[MV]], [[MV1]]
+    ; GFX9-DENORM: [[FADD:%[0-9]+]]:_(s64) = reassoc G_FADD [[MV2]], [[FMUL]]
+    ; GFX9-DENORM: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FADD]](s64)
+    ; GFX9-DENORM: $vgpr0 = COPY [[UV]](s32)
+    ; GFX9-DENORM: $vgpr1 = COPY [[UV1]](s32)
+    ; GFX9-DENORM: [[COPY7:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY6]]
+    ; GFX9-DENORM: S_SETPC_B64_return [[COPY7]], implicit $vgpr0, implicit $vgpr1
+    ; GFX9-UNSAFE-LABEL: name: test_double_add_mul_rhs
+    ; GFX9-UNSAFE: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX9-UNSAFE: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX9-UNSAFE: [[MV:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY]](s32), [[COPY1]](s32)
+    ; GFX9-UNSAFE: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX9-UNSAFE: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3
+    ; GFX9-UNSAFE: [[MV1:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY2]](s32), [[COPY3]](s32)
+    ; GFX9-UNSAFE: [[COPY4:%[0-9]+]]:_(s32) = COPY $vgpr4
+    ; GFX9-UNSAFE: [[COPY5:%[0-9]+]]:_(s32) = COPY $vgpr5
+    ; GFX9-UNSAFE: [[MV2:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY4]](s32), [[COPY5]](s32)
+    ; GFX9-UNSAFE: [[COPY6:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX9-UNSAFE: [[FMA:%[0-9]+]]:_(s64) = G_FMA [[MV]], [[MV1]], [[MV2]]
+    ; GFX9-UNSAFE: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FMA]](s64)
+    ; GFX9-UNSAFE: $vgpr0 = COPY [[UV]](s32)
+    ; GFX9-UNSAFE: $vgpr1 = COPY [[UV1]](s32)
+    ; GFX9-UNSAFE: [[COPY7:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY6]]
+    ; GFX9-UNSAFE: S_SETPC_B64_return [[COPY7]], implicit $vgpr0, implicit $vgpr1
+    ; GFX10-LABEL: name: test_double_add_mul_rhs
+    ; GFX10: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX10: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX10: [[MV:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY]](s32), [[COPY1]](s32)
+    ; GFX10: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX10: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3
+    ; GFX10: [[MV1:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY2]](s32), [[COPY3]](s32)
+    ; GFX10: [[COPY4:%[0-9]+]]:_(s32) = COPY $vgpr4
+    ; GFX10: [[COPY5:%[0-9]+]]:_(s32) = COPY $vgpr5
+    ; GFX10: [[MV2:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY4]](s32), [[COPY5]](s32)
+    ; GFX10: [[COPY6:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX10: [[FMUL:%[0-9]+]]:_(s64) = reassoc G_FMUL [[MV]], [[MV1]]
+    ; GFX10: [[FADD:%[0-9]+]]:_(s64) = reassoc G_FADD [[MV2]], [[FMUL]]
+    ; GFX10: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FADD]](s64)
+    ; GFX10: $vgpr0 = COPY [[UV]](s32)
+    ; GFX10: $vgpr1 = COPY [[UV1]](s32)
+    ; GFX10: [[COPY7:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY6]]
+    ; GFX10: S_SETPC_B64_return [[COPY7]], implicit $vgpr0, implicit $vgpr1
+    ; GFX10-CONTRACT-LABEL: name: test_double_add_mul_rhs
+    ; GFX10-CONTRACT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX10-CONTRACT: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX10-CONTRACT: [[MV:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY]](s32), [[COPY1]](s32)
+    ; GFX10-CONTRACT: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX10-CONTRACT: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3
+    ; GFX10-CONTRACT: [[MV1:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY2]](s32), [[COPY3]](s32)
+    ; GFX10-CONTRACT: [[COPY4:%[0-9]+]]:_(s32) = COPY $vgpr4
+    ; GFX10-CONTRACT: [[COPY5:%[0-9]+]]:_(s32) = COPY $vgpr5
+    ; GFX10-CONTRACT: [[MV2:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY4]](s32), [[COPY5]](s32)
+    ; GFX10-CONTRACT: [[COPY6:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX10-CONTRACT: [[FMA:%[0-9]+]]:_(s64) = G_FMA [[MV]], [[MV1]], [[MV2]]
+    ; GFX10-CONTRACT: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FMA]](s64)
+    ; GFX10-CONTRACT: $vgpr0 = COPY [[UV]](s32)
+    ; GFX10-CONTRACT: $vgpr1 = COPY [[UV1]](s32)
+    ; GFX10-CONTRACT: [[COPY7:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY6]]
+    ; GFX10-CONTRACT: S_SETPC_B64_return [[COPY7]], implicit $vgpr0, implicit $vgpr1
+    ; GFX10-DENORM-LABEL: name: test_double_add_mul_rhs
+    ; GFX10-DENORM: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX10-DENORM: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX10-DENORM: [[MV:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY]](s32), [[COPY1]](s32)
+    ; GFX10-DENORM: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX10-DENORM: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3
+    ; GFX10-DENORM: [[MV1:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY2]](s32), [[COPY3]](s32)
+    ; GFX10-DENORM: [[COPY4:%[0-9]+]]:_(s32) = COPY $vgpr4
+    ; GFX10-DENORM: [[COPY5:%[0-9]+]]:_(s32) = COPY $vgpr5
+    ; GFX10-DENORM: [[MV2:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY4]](s32), [[COPY5]](s32)
+    ; GFX10-DENORM: [[COPY6:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX10-DENORM: [[FMUL:%[0-9]+]]:_(s64) = reassoc G_FMUL [[MV]], [[MV1]]
+    ; GFX10-DENORM: [[FADD:%[0-9]+]]:_(s64) = reassoc G_FADD [[MV2]], [[FMUL]]
+    ; GFX10-DENORM: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FADD]](s64)
+    ; GFX10-DENORM: $vgpr0 = COPY [[UV]](s32)
+    ; GFX10-DENORM: $vgpr1 = COPY [[UV1]](s32)
+    ; GFX10-DENORM: [[COPY7:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY6]]
+    ; GFX10-DENORM: S_SETPC_B64_return [[COPY7]], implicit $vgpr0, implicit $vgpr1
+    ; GFX10-UNSAFE-LABEL: name: test_double_add_mul_rhs
+    ; GFX10-UNSAFE: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX10-UNSAFE: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX10-UNSAFE: [[MV:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY]](s32), [[COPY1]](s32)
+    ; GFX10-UNSAFE: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX10-UNSAFE: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3
+    ; GFX10-UNSAFE: [[MV1:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY2]](s32), [[COPY3]](s32)
+    ; GFX10-UNSAFE: [[COPY4:%[0-9]+]]:_(s32) = COPY $vgpr4
+    ; GFX10-UNSAFE: [[COPY5:%[0-9]+]]:_(s32) = COPY $vgpr5
+    ; GFX10-UNSAFE: [[MV2:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY4]](s32), [[COPY5]](s32)
+    ; GFX10-UNSAFE: [[COPY6:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX10-UNSAFE: [[FMA:%[0-9]+]]:_(s64) = G_FMA [[MV]], [[MV1]], [[MV2]]
+    ; GFX10-UNSAFE: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FMA]](s64)
+    ; GFX10-UNSAFE: $vgpr0 = COPY [[UV]](s32)
+    ; GFX10-UNSAFE: $vgpr1 = COPY [[UV1]](s32)
+    ; GFX10-UNSAFE: [[COPY7:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY6]]
+    ; GFX10-UNSAFE: S_SETPC_B64_return [[COPY7]], implicit $vgpr0, implicit $vgpr1
+    %4:_(s32) = COPY $vgpr0
+    %5:_(s32) = COPY $vgpr1
+    %0:_(s64) = G_MERGE_VALUES %4(s32), %5(s32)
+    %6:_(s32) = COPY $vgpr2
+    %7:_(s32) = COPY $vgpr3
+    %1:_(s64) = G_MERGE_VALUES %6(s32), %7(s32)
+    %8:_(s32) = COPY $vgpr4
+    %9:_(s32) = COPY $vgpr5
+    %2:_(s64) = G_MERGE_VALUES %8(s32), %9(s32)
+    %3:sgpr_64 = COPY $sgpr30_sgpr31
+    %10:_(s64) = reassoc G_FMUL %0, %1
+    %11:_(s64) = reassoc G_FADD %2, %10
+    %13:_(s32), %14:_(s32) = G_UNMERGE_VALUES %11(s64)
+    $vgpr0 = COPY %13(s32)
+    $vgpr1 = COPY %14(s32)
+    %12:ccr_sgpr_64 = COPY %3
+    S_SETPC_B64_return %12, implicit $vgpr0, implicit $vgpr1
+...
+
+---
+name:            test_4xfloat_add_mul
+body:             |
+  bb.1.entry:
+    liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8, $vgpr9, $vgpr10, $vgpr11, $sgpr30_sgpr31
+
+    ; GFX9-LABEL: name: test_4xfloat_add_mul
+    ; GFX9: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX9: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX9: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX9: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3
+    ; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32)
+    ; GFX9: [[COPY4:%[0-9]+]]:_(s32) = COPY $vgpr4
+    ; GFX9: [[COPY5:%[0-9]+]]:_(s32) = COPY $vgpr5
+    ; GFX9: [[COPY6:%[0-9]+]]:_(s32) = COPY $vgpr6
+    ; GFX9: [[COPY7:%[0-9]+]]:_(s32) = COPY $vgpr7
+    ; GFX9: [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+    ; GFX9: [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr8
+    ; GFX9: [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr9
+    ; GFX9: [[COPY10:%[0-9]+]]:_(s32) = COPY $vgpr10
+    ; GFX9: [[COPY11:%[0-9]+]]:_(s32) = COPY $vgpr11
+    ; GFX9: [[BUILD_VECTOR2:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+    ; GFX9: [[COPY12:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX9: [[FMUL:%[0-9]+]]:_(<4 x s32>) = reassoc G_FMUL [[BUILD_VECTOR]], [[BUILD_VECTOR1]]
+    ; GFX9: [[FADD:%[0-9]+]]:_(<4 x s32>) = reassoc G_FADD [[FMUL]], [[BUILD_VECTOR2]]
+    ; GFX9: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FADD]](<4 x s32>)
+    ; GFX9: $vgpr0 = COPY [[UV]](s32)
+    ; GFX9: $vgpr1 = COPY [[UV1]](s32)
+    ; GFX9: $vgpr2 = COPY [[UV2]](s32)
+    ; GFX9: $vgpr3 = COPY [[UV3]](s32)
+    ; GFX9: [[COPY13:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY12]]
+    ; GFX9: S_SETPC_B64_return [[COPY13]], implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+    ; GFX9-CONTRACT-LABEL: name: test_4xfloat_add_mul
+    ; GFX9-CONTRACT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX9-CONTRACT: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX9-CONTRACT: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX9-CONTRACT: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3
+    ; GFX9-CONTRACT: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32)
+    ; GFX9-CONTRACT: [[COPY4:%[0-9]+]]:_(s32) = COPY $vgpr4
+    ; GFX9-CONTRACT: [[COPY5:%[0-9]+]]:_(s32) = COPY $vgpr5
+    ; GFX9-CONTRACT: [[COPY6:%[0-9]+]]:_(s32) = COPY $vgpr6
+    ; GFX9-CONTRACT: [[COPY7:%[0-9]+]]:_(s32) = COPY $vgpr7
+    ; GFX9-CONTRACT: [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+    ; GFX9-CONTRACT: [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr8
+    ; GFX9-CONTRACT: [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr9
+    ; GFX9-CONTRACT: [[COPY10:%[0-9]+]]:_(s32) = COPY $vgpr10
+    ; GFX9-CONTRACT: [[COPY11:%[0-9]+]]:_(s32) = COPY $vgpr11
+    ; GFX9-CONTRACT: [[BUILD_VECTOR2:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+    ; GFX9-CONTRACT: [[COPY12:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX9-CONTRACT: [[FMA:%[0-9]+]]:_(<4 x s32>) = G_FMA [[BUILD_VECTOR]], [[BUILD_VECTOR1]], [[BUILD_VECTOR2]]
+    ; GFX9-CONTRACT: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FMA]](<4 x s32>)
+    ; GFX9-CONTRACT: $vgpr0 = COPY [[UV]](s32)
+    ; GFX9-CONTRACT: $vgpr1 = COPY [[UV1]](s32)
+    ; GFX9-CONTRACT: $vgpr2 = COPY [[UV2]](s32)
+    ; GFX9-CONTRACT: $vgpr3 = COPY [[UV3]](s32)
+    ; GFX9-CONTRACT: [[COPY13:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY12]]
+    ; GFX9-CONTRACT: S_SETPC_B64_return [[COPY13]], implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+    ; GFX9-DENORM-LABEL: name: test_4xfloat_add_mul
+    ; GFX9-DENORM: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX9-DENORM: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX9-DENORM: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX9-DENORM: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3
+    ; GFX9-DENORM: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32)
+    ; GFX9-DENORM: [[COPY4:%[0-9]+]]:_(s32) = COPY $vgpr4
+    ; GFX9-DENORM: [[COPY5:%[0-9]+]]:_(s32) = COPY $vgpr5
+    ; GFX9-DENORM: [[COPY6:%[0-9]+]]:_(s32) = COPY $vgpr6
+    ; GFX9-DENORM: [[COPY7:%[0-9]+]]:_(s32) = COPY $vgpr7
+    ; GFX9-DENORM: [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+    ; GFX9-DENORM: [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr8
+    ; GFX9-DENORM: [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr9
+    ; GFX9-DENORM: [[COPY10:%[0-9]+]]:_(s32) = COPY $vgpr10
+    ; GFX9-DENORM: [[COPY11:%[0-9]+]]:_(s32) = COPY $vgpr11
+    ; GFX9-DENORM: [[BUILD_VECTOR2:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+    ; GFX9-DENORM: [[COPY12:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX9-DENORM: [[FMUL:%[0-9]+]]:_(<4 x s32>) = reassoc G_FMUL [[BUILD_VECTOR]], [[BUILD_VECTOR1]]
+    ; GFX9-DENORM: [[FADD:%[0-9]+]]:_(<4 x s32>) = reassoc G_FADD [[FMUL]], [[BUILD_VECTOR2]]
+    ; GFX9-DENORM: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FADD]](<4 x s32>)
+    ; GFX9-DENORM: $vgpr0 = COPY [[UV]](s32)
+    ; GFX9-DENORM: $vgpr1 = COPY [[UV1]](s32)
+    ; GFX9-DENORM: $vgpr2 = COPY [[UV2]](s32)
+    ; GFX9-DENORM: $vgpr3 = COPY [[UV3]](s32)
+    ; GFX9-DENORM: [[COPY13:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY12]]
+    ; GFX9-DENORM: S_SETPC_B64_return [[COPY13]], implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+    ; GFX9-UNSAFE-LABEL: name: test_4xfloat_add_mul
+    ; GFX9-UNSAFE: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX9-UNSAFE: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX9-UNSAFE: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX9-UNSAFE: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3
+    ; GFX9-UNSAFE: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32)
+    ; GFX9-UNSAFE: [[COPY4:%[0-9]+]]:_(s32) = COPY $vgpr4
+    ; GFX9-UNSAFE: [[COPY5:%[0-9]+]]:_(s32) = COPY $vgpr5
+    ; GFX9-UNSAFE: [[COPY6:%[0-9]+]]:_(s32) = COPY $vgpr6
+    ; GFX9-UNSAFE: [[COPY7:%[0-9]+]]:_(s32) = COPY $vgpr7
+    ; GFX9-UNSAFE: [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+    ; GFX9-UNSAFE: [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr8
+    ; GFX9-UNSAFE: [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr9
+    ; GFX9-UNSAFE: [[COPY10:%[0-9]+]]:_(s32) = COPY $vgpr10
+    ; GFX9-UNSAFE: [[COPY11:%[0-9]+]]:_(s32) = COPY $vgpr11
+    ; GFX9-UNSAFE: [[BUILD_VECTOR2:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+    ; GFX9-UNSAFE: [[COPY12:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX9-UNSAFE: [[FMA:%[0-9]+]]:_(<4 x s32>) = G_FMA [[BUILD_VECTOR]], [[BUILD_VECTOR1]], [[BUILD_VECTOR2]]
+    ; GFX9-UNSAFE: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FMA]](<4 x s32>)
+    ; GFX9-UNSAFE: $vgpr0 = COPY [[UV]](s32)
+    ; GFX9-UNSAFE: $vgpr1 = COPY [[UV1]](s32)
+    ; GFX9-UNSAFE: $vgpr2 = COPY [[UV2]](s32)
+    ; GFX9-UNSAFE: $vgpr3 = COPY [[UV3]](s32)
+    ; GFX9-UNSAFE: [[COPY13:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY12]]
+    ; GFX9-UNSAFE: S_SETPC_B64_return [[COPY13]], implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+    ; GFX10-LABEL: name: test_4xfloat_add_mul
+    ; GFX10: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX10: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX10: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX10: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3
+    ; GFX10: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32)
+    ; GFX10: [[COPY4:%[0-9]+]]:_(s32) = COPY $vgpr4
+    ; GFX10: [[COPY5:%[0-9]+]]:_(s32) = COPY $vgpr5
+    ; GFX10: [[COPY6:%[0-9]+]]:_(s32) = COPY $vgpr6
+    ; GFX10: [[COPY7:%[0-9]+]]:_(s32) = COPY $vgpr7
+    ; GFX10: [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+    ; GFX10: [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr8
+    ; GFX10: [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr9
+    ; GFX10: [[COPY10:%[0-9]+]]:_(s32) = COPY $vgpr10
+    ; GFX10: [[COPY11:%[0-9]+]]:_(s32) = COPY $vgpr11
+    ; GFX10: [[BUILD_VECTOR2:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+    ; GFX10: [[COPY12:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX10: [[FMUL:%[0-9]+]]:_(<4 x s32>) = reassoc G_FMUL [[BUILD_VECTOR]], [[BUILD_VECTOR1]]
+    ; GFX10: [[FADD:%[0-9]+]]:_(<4 x s32>) = reassoc G_FADD [[FMUL]], [[BUILD_VECTOR2]]
+    ; GFX10: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FADD]](<4 x s32>)
+    ; GFX10: $vgpr0 = COPY [[UV]](s32)
+    ; GFX10: $vgpr1 = COPY [[UV1]](s32)
+    ; GFX10: $vgpr2 = COPY [[UV2]](s32)
+    ; GFX10: $vgpr3 = COPY [[UV3]](s32)
+    ; GFX10: [[COPY13:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY12]]
+    ; GFX10: S_SETPC_B64_return [[COPY13]], implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+    ; GFX10-CONTRACT-LABEL: name: test_4xfloat_add_mul
+    ; GFX10-CONTRACT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX10-CONTRACT: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX10-CONTRACT: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX10-CONTRACT: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3
+    ; GFX10-CONTRACT: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32)
+    ; GFX10-CONTRACT: [[COPY4:%[0-9]+]]:_(s32) = COPY $vgpr4
+    ; GFX10-CONTRACT: [[COPY5:%[0-9]+]]:_(s32) = COPY $vgpr5
+    ; GFX10-CONTRACT: [[COPY6:%[0-9]+]]:_(s32) = COPY $vgpr6
+    ; GFX10-CONTRACT: [[COPY7:%[0-9]+]]:_(s32) = COPY $vgpr7
+    ; GFX10-CONTRACT: [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+    ; GFX10-CONTRACT: [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr8
+    ; GFX10-CONTRACT: [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr9
+    ; GFX10-CONTRACT: [[COPY10:%[0-9]+]]:_(s32) = COPY $vgpr10
+    ; GFX10-CONTRACT: [[COPY11:%[0-9]+]]:_(s32) = COPY $vgpr11
+    ; GFX10-CONTRACT: [[BUILD_VECTOR2:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+    ; GFX10-CONTRACT: [[COPY12:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX10-CONTRACT: [[FMA:%[0-9]+]]:_(<4 x s32>) = G_FMA [[BUILD_VECTOR]], [[BUILD_VECTOR1]], [[BUILD_VECTOR2]]
+    ; GFX10-CONTRACT: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FMA]](<4 x s32>)
+    ; GFX10-CONTRACT: $vgpr0 = COPY [[UV]](s32)
+    ; GFX10-CONTRACT: $vgpr1 = COPY [[UV1]](s32)
+    ; GFX10-CONTRACT: $vgpr2 = COPY [[UV2]](s32)
+    ; GFX10-CONTRACT: $vgpr3 = COPY [[UV3]](s32)
+    ; GFX10-CONTRACT: [[COPY13:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY12]]
+    ; GFX10-CONTRACT: S_SETPC_B64_return [[COPY13]], implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+    ; GFX10-DENORM-LABEL: name: test_4xfloat_add_mul
+    ; GFX10-DENORM: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX10-DENORM: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX10-DENORM: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX10-DENORM: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3
+    ; GFX10-DENORM: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32)
+    ; GFX10-DENORM: [[COPY4:%[0-9]+]]:_(s32) = COPY $vgpr4
+    ; GFX10-DENORM: [[COPY5:%[0-9]+]]:_(s32) = COPY $vgpr5
+    ; GFX10-DENORM: [[COPY6:%[0-9]+]]:_(s32) = COPY $vgpr6
+    ; GFX10-DENORM: [[COPY7:%[0-9]+]]:_(s32) = COPY $vgpr7
+    ; GFX10-DENORM: [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+    ; GFX10-DENORM: [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr8
+    ; GFX10-DENORM: [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr9
+    ; GFX10-DENORM: [[COPY10:%[0-9]+]]:_(s32) = COPY $vgpr10
+    ; GFX10-DENORM: [[COPY11:%[0-9]+]]:_(s32) = COPY $vgpr11
+    ; GFX10-DENORM: [[BUILD_VECTOR2:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+    ; GFX10-DENORM: [[COPY12:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX10-DENORM: [[FMUL:%[0-9]+]]:_(<4 x s32>) = reassoc G_FMUL [[BUILD_VECTOR]], [[BUILD_VECTOR1]]
+    ; GFX10-DENORM: [[FADD:%[0-9]+]]:_(<4 x s32>) = reassoc G_FADD [[FMUL]], [[BUILD_VECTOR2]]
+    ; GFX10-DENORM: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FADD]](<4 x s32>)
+    ; GFX10-DENORM: $vgpr0 = COPY [[UV]](s32)
+    ; GFX10-DENORM: $vgpr1 = COPY [[UV1]](s32)
+    ; GFX10-DENORM: $vgpr2 = COPY [[UV2]](s32)
+    ; GFX10-DENORM: $vgpr3 = COPY [[UV3]](s32)
+    ; GFX10-DENORM: [[COPY13:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY12]]
+    ; GFX10-DENORM: S_SETPC_B64_return [[COPY13]], implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+    ; GFX10-UNSAFE-LABEL: name: test_4xfloat_add_mul
+    ; GFX10-UNSAFE: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX10-UNSAFE: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX10-UNSAFE: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX10-UNSAFE: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3
+    ; GFX10-UNSAFE: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32)
+    ; GFX10-UNSAFE: [[COPY4:%[0-9]+]]:_(s32) = COPY $vgpr4
+    ; GFX10-UNSAFE: [[COPY5:%[0-9]+]]:_(s32) = COPY $vgpr5
+    ; GFX10-UNSAFE: [[COPY6:%[0-9]+]]:_(s32) = COPY $vgpr6
+    ; GFX10-UNSAFE: [[COPY7:%[0-9]+]]:_(s32) = COPY $vgpr7
+    ; GFX10-UNSAFE: [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+    ; GFX10-UNSAFE: [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr8
+    ; GFX10-UNSAFE: [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr9
+    ; GFX10-UNSAFE: [[COPY10:%[0-9]+]]:_(s32) = COPY $vgpr10
+    ; GFX10-UNSAFE: [[COPY11:%[0-9]+]]:_(s32) = COPY $vgpr11
+    ; GFX10-UNSAFE: [[BUILD_VECTOR2:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+    ; GFX10-UNSAFE: [[COPY12:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX10-UNSAFE: [[FMA:%[0-9]+]]:_(<4 x s32>) = G_FMA [[BUILD_VECTOR]], [[BUILD_VECTOR1]], [[BUILD_VECTOR2]]
+    ; GFX10-UNSAFE: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FMA]](<4 x s32>)
+    ; GFX10-UNSAFE: $vgpr0 = COPY [[UV]](s32)
+    ; GFX10-UNSAFE: $vgpr1 = COPY [[UV1]](s32)
+    ; GFX10-UNSAFE: $vgpr2 = COPY [[UV2]](s32)
+    ; GFX10-UNSAFE: $vgpr3 = COPY [[UV3]](s32)
+    ; GFX10-UNSAFE: [[COPY13:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY12]]
+    ; GFX10-UNSAFE: S_SETPC_B64_return [[COPY13]], implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+    %4:_(s32) = COPY $vgpr0
+    %5:_(s32) = COPY $vgpr1
+    %6:_(s32) = COPY $vgpr2
+    %7:_(s32) = COPY $vgpr3
+    %0:_(<4 x s32>) = G_BUILD_VECTOR %4(s32), %5(s32), %6(s32), %7(s32)
+    %8:_(s32) = COPY $vgpr4
+    %9:_(s32) = COPY $vgpr5
+    %10:_(s32) = COPY $vgpr6
+    %11:_(s32) = COPY $vgpr7
+    %1:_(<4 x s32>) = G_BUILD_VECTOR %8(s32), %9(s32), %10(s32), %11(s32)
+    %12:_(s32) = COPY $vgpr8
+    %13:_(s32) = COPY $vgpr9
+    %14:_(s32) = COPY $vgpr10
+    %15:_(s32) = COPY $vgpr11
+    %2:_(<4 x s32>) = G_BUILD_VECTOR %12(s32), %13(s32), %14(s32), %15(s32)
+    %3:sgpr_64 = COPY $sgpr30_sgpr31
+    %16:_(<4 x s32>) = reassoc G_FMUL %0, %1
+    %17:_(<4 x s32>) = reassoc G_FADD %16, %2
+    %19:_(s32), %20:_(s32), %21:_(s32), %22:_(s32) = G_UNMERGE_VALUES %17(<4 x s32>)
+    $vgpr0 = COPY %19(s32)
+    $vgpr1 = COPY %20(s32)
+    $vgpr2 = COPY %21(s32)
+    $vgpr3 = COPY %22(s32)
+    %18:ccr_sgpr_64 = COPY %3
+    S_SETPC_B64_return %18, implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+...
+
+---
+name:            test_3xfloat_add_mul_rhs
+body:             |
+  bb.1.entry:
+    liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8, $sgpr30_sgpr31
+
+    ; GFX9-LABEL: name: test_3xfloat_add_mul_rhs
+    ; GFX9: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX9: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX9: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32)
+    ; GFX9: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3
+    ; GFX9: [[COPY4:%[0-9]+]]:_(s32) = COPY $vgpr4
+    ; GFX9: [[COPY5:%[0-9]+]]:_(s32) = COPY $vgpr5
+    ; GFX9: [[BUILD_VECTOR1:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32)
+    ; GFX9: [[COPY6:%[0-9]+]]:_(s32) = COPY $vgpr6
+    ; GFX9: [[COPY7:%[0-9]+]]:_(s32) = COPY $vgpr7
+    ; GFX9: [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr8
+    ; GFX9: [[BUILD_VECTOR2:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[COPY6]](s32), [[COPY7]](s32), [[COPY8]](s32)
+    ; GFX9: [[COPY9:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX9: [[FMUL:%[0-9]+]]:_(<3 x s32>) = reassoc G_FMUL [[BUILD_VECTOR]], [[BUILD_VECTOR1]]
+    ; GFX9: [[FADD:%[0-9]+]]:_(<3 x s32>) = reassoc G_FADD [[BUILD_VECTOR2]], [[FMUL]]
+    ; GFX9: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FADD]](<3 x s32>)
+    ; GFX9: $vgpr0 = COPY [[UV]](s32)
+    ; GFX9: $vgpr1 = COPY [[UV1]](s32)
+    ; GFX9: $vgpr2 = COPY [[UV2]](s32)
+    ; GFX9: [[COPY10:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY9]]
+    ; GFX9: S_SETPC_B64_return [[COPY10]], implicit $vgpr0, implicit $vgpr1, implicit $vgpr2
+    ; GFX9-CONTRACT-LABEL: name: test_3xfloat_add_mul_rhs
+    ; GFX9-CONTRACT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX9-CONTRACT: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX9-CONTRACT: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX9-CONTRACT: [[BUILD_VECTOR:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32)
+    ; GFX9-CONTRACT: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3
+    ; GFX9-CONTRACT: [[COPY4:%[0-9]+]]:_(s32) = COPY $vgpr4
+    ; GFX9-CONTRACT: [[COPY5:%[0-9]+]]:_(s32) = COPY $vgpr5
+    ; GFX9-CONTRACT: [[BUILD_VECTOR1:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32)
+    ; GFX9-CONTRACT: [[COPY6:%[0-9]+]]:_(s32) = COPY $vgpr6
+    ; GFX9-CONTRACT: [[COPY7:%[0-9]+]]:_(s32) = COPY $vgpr7
+    ; GFX9-CONTRACT: [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr8
+    ; GFX9-CONTRACT: [[BUILD_VECTOR2:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[COPY6]](s32), [[COPY7]](s32), [[COPY8]](s32)
+    ; GFX9-CONTRACT: [[COPY9:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX9-CONTRACT: [[FMA:%[0-9]+]]:_(<3 x s32>) = G_FMA [[BUILD_VECTOR]], [[BUILD_VECTOR1]], [[BUILD_VECTOR2]]
+    ; GFX9-CONTRACT: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FMA]](<3 x s32>)
+    ; GFX9-CONTRACT: $vgpr0 = COPY [[UV]](s32)
+    ; GFX9-CONTRACT: $vgpr1 = COPY [[UV1]](s32)
+    ; GFX9-CONTRACT: $vgpr2 = COPY [[UV2]](s32)
+    ; GFX9-CONTRACT: [[COPY10:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY9]]
+    ; GFX9-CONTRACT: S_SETPC_B64_return [[COPY10]], implicit $vgpr0, implicit $vgpr1, implicit $vgpr2
+    ; GFX9-DENORM-LABEL: name: test_3xfloat_add_mul_rhs
+    ; GFX9-DENORM: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX9-DENORM: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX9-DENORM: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX9-DENORM: [[BUILD_VECTOR:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32)
+    ; GFX9-DENORM: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3
+    ; GFX9-DENORM: [[COPY4:%[0-9]+]]:_(s32) = COPY $vgpr4
+    ; GFX9-DENORM: [[COPY5:%[0-9]+]]:_(s32) = COPY $vgpr5
+    ; GFX9-DENORM: [[BUILD_VECTOR1:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32)
+    ; GFX9-DENORM: [[COPY6:%[0-9]+]]:_(s32) = COPY $vgpr6
+    ; GFX9-DENORM: [[COPY7:%[0-9]+]]:_(s32) = COPY $vgpr7
+    ; GFX9-DENORM: [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr8
+    ; GFX9-DENORM: [[BUILD_VECTOR2:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[COPY6]](s32), [[COPY7]](s32), [[COPY8]](s32)
+    ; GFX9-DENORM: [[COPY9:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX9-DENORM: [[FMUL:%[0-9]+]]:_(<3 x s32>) = reassoc G_FMUL [[BUILD_VECTOR]], [[BUILD_VECTOR1]]
+    ; GFX9-DENORM: [[FADD:%[0-9]+]]:_(<3 x s32>) = reassoc G_FADD [[BUILD_VECTOR2]], [[FMUL]]
+    ; GFX9-DENORM: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FADD]](<3 x s32>)
+    ; GFX9-DENORM: $vgpr0 = COPY [[UV]](s32)
+    ; GFX9-DENORM: $vgpr1 = COPY [[UV1]](s32)
+    ; GFX9-DENORM: $vgpr2 = COPY [[UV2]](s32)
+    ; GFX9-DENORM: [[COPY10:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY9]]
+    ; GFX9-DENORM: S_SETPC_B64_return [[COPY10]], implicit $vgpr0, implicit $vgpr1, implicit $vgpr2
+    ; GFX9-UNSAFE-LABEL: name: test_3xfloat_add_mul_rhs
+    ; GFX9-UNSAFE: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX9-UNSAFE: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX9-UNSAFE: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX9-UNSAFE: [[BUILD_VECTOR:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32)
+    ; GFX9-UNSAFE: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3
+    ; GFX9-UNSAFE: [[COPY4:%[0-9]+]]:_(s32) = COPY $vgpr4
+    ; GFX9-UNSAFE: [[COPY5:%[0-9]+]]:_(s32) = COPY $vgpr5
+    ; GFX9-UNSAFE: [[BUILD_VECTOR1:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32)
+    ; GFX9-UNSAFE: [[COPY6:%[0-9]+]]:_(s32) = COPY $vgpr6
+    ; GFX9-UNSAFE: [[COPY7:%[0-9]+]]:_(s32) = COPY $vgpr7
+    ; GFX9-UNSAFE: [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr8
+    ; GFX9-UNSAFE: [[BUILD_VECTOR2:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[COPY6]](s32), [[COPY7]](s32), [[COPY8]](s32)
+    ; GFX9-UNSAFE: [[COPY9:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX9-UNSAFE: [[FMA:%[0-9]+]]:_(<3 x s32>) = G_FMA [[BUILD_VECTOR]], [[BUILD_VECTOR1]], [[BUILD_VECTOR2]]
+    ; GFX9-UNSAFE: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FMA]](<3 x s32>)
+    ; GFX9-UNSAFE: $vgpr0 = COPY [[UV]](s32)
+    ; GFX9-UNSAFE: $vgpr1 = COPY [[UV1]](s32)
+    ; GFX9-UNSAFE: $vgpr2 = COPY [[UV2]](s32)
+    ; GFX9-UNSAFE: [[COPY10:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY9]]
+    ; GFX9-UNSAFE: S_SETPC_B64_return [[COPY10]], implicit $vgpr0, implicit $vgpr1, implicit $vgpr2
+    ; GFX10-LABEL: name: test_3xfloat_add_mul_rhs
+    ; GFX10: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX10: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX10: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX10: [[BUILD_VECTOR:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32)
+    ; GFX10: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3
+    ; GFX10: [[COPY4:%[0-9]+]]:_(s32) = COPY $vgpr4
+    ; GFX10: [[COPY5:%[0-9]+]]:_(s32) = COPY $vgpr5
+    ; GFX10: [[BUILD_VECTOR1:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32)
+    ; GFX10: [[COPY6:%[0-9]+]]:_(s32) = COPY $vgpr6
+    ; GFX10: [[COPY7:%[0-9]+]]:_(s32) = COPY $vgpr7
+    ; GFX10: [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr8
+    ; GFX10: [[BUILD_VECTOR2:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[COPY6]](s32), [[COPY7]](s32), [[COPY8]](s32)
+    ; GFX10: [[COPY9:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX10: [[FMUL:%[0-9]+]]:_(<3 x s32>) = reassoc G_FMUL [[BUILD_VECTOR]], [[BUILD_VECTOR1]]
+    ; GFX10: [[FADD:%[0-9]+]]:_(<3 x s32>) = reassoc G_FADD [[BUILD_VECTOR2]], [[FMUL]]
+    ; GFX10: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FADD]](<3 x s32>)
+    ; GFX10: $vgpr0 = COPY [[UV]](s32)
+    ; GFX10: $vgpr1 = COPY [[UV1]](s32)
+    ; GFX10: $vgpr2 = COPY [[UV2]](s32)
+    ; GFX10: [[COPY10:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY9]]
+    ; GFX10: S_SETPC_B64_return [[COPY10]], implicit $vgpr0, implicit $vgpr1, implicit $vgpr2
+    ; GFX10-CONTRACT-LABEL: name: test_3xfloat_add_mul_rhs
+    ; GFX10-CONTRACT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX10-CONTRACT: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX10-CONTRACT: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX10-CONTRACT: [[BUILD_VECTOR:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32)
+    ; GFX10-CONTRACT: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3
+    ; GFX10-CONTRACT: [[COPY4:%[0-9]+]]:_(s32) = COPY $vgpr4
+    ; GFX10-CONTRACT: [[COPY5:%[0-9]+]]:_(s32) = COPY $vgpr5
+    ; GFX10-CONTRACT: [[BUILD_VECTOR1:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32)
+    ; GFX10-CONTRACT: [[COPY6:%[0-9]+]]:_(s32) = COPY $vgpr6
+    ; GFX10-CONTRACT: [[COPY7:%[0-9]+]]:_(s32) = COPY $vgpr7
+    ; GFX10-CONTRACT: [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr8
+    ; GFX10-CONTRACT: [[BUILD_VECTOR2:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[COPY6]](s32), [[COPY7]](s32), [[COPY8]](s32)
+    ; GFX10-CONTRACT: [[COPY9:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX10-CONTRACT: [[FMA:%[0-9]+]]:_(<3 x s32>) = G_FMA [[BUILD_VECTOR]], [[BUILD_VECTOR1]], [[BUILD_VECTOR2]]
+    ; GFX10-CONTRACT: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FMA]](<3 x s32>)
+    ; GFX10-CONTRACT: $vgpr0 = COPY [[UV]](s32)
+    ; GFX10-CONTRACT: $vgpr1 = COPY [[UV1]](s32)
+    ; GFX10-CONTRACT: $vgpr2 = COPY [[UV2]](s32)
+    ; GFX10-CONTRACT: [[COPY10:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY9]]
+    ; GFX10-CONTRACT: S_SETPC_B64_return [[COPY10]], implicit $vgpr0, implicit $vgpr1, implicit $vgpr2
+    ; GFX10-DENORM-LABEL: name: test_3xfloat_add_mul_rhs
+    ; GFX10-DENORM: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX10-DENORM: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX10-DENORM: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX10-DENORM: [[BUILD_VECTOR:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32)
+    ; GFX10-DENORM: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3
+    ; GFX10-DENORM: [[COPY4:%[0-9]+]]:_(s32) = COPY $vgpr4
+    ; GFX10-DENORM: [[COPY5:%[0-9]+]]:_(s32) = COPY $vgpr5
+    ; GFX10-DENORM: [[BUILD_VECTOR1:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32)
+    ; GFX10-DENORM: [[COPY6:%[0-9]+]]:_(s32) = COPY $vgpr6
+    ; GFX10-DENORM: [[COPY7:%[0-9]+]]:_(s32) = COPY $vgpr7
+    ; GFX10-DENORM: [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr8
+    ; GFX10-DENORM: [[BUILD_VECTOR2:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[COPY6]](s32), [[COPY7]](s32), [[COPY8]](s32)
+    ; GFX10-DENORM: [[COPY9:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX10-DENORM: [[FMUL:%[0-9]+]]:_(<3 x s32>) = reassoc G_FMUL [[BUILD_VECTOR]], [[BUILD_VECTOR1]]
+    ; GFX10-DENORM: [[FADD:%[0-9]+]]:_(<3 x s32>) = reassoc G_FADD [[BUILD_VECTOR2]], [[FMUL]]
+    ; GFX10-DENORM: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FADD]](<3 x s32>)
+    ; GFX10-DENORM: $vgpr0 = COPY [[UV]](s32)
+    ; GFX10-DENORM: $vgpr1 = COPY [[UV1]](s32)
+    ; GFX10-DENORM: $vgpr2 = COPY [[UV2]](s32)
+    ; GFX10-DENORM: [[COPY10:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY9]]
+    ; GFX10-DENORM: S_SETPC_B64_return [[COPY10]], implicit $vgpr0, implicit $vgpr1, implicit $vgpr2
+    ; GFX10-UNSAFE-LABEL: name: test_3xfloat_add_mul_rhs
+    ; GFX10-UNSAFE: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX10-UNSAFE: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX10-UNSAFE: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX10-UNSAFE: [[BUILD_VECTOR:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32)
+    ; GFX10-UNSAFE: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3
+    ; GFX10-UNSAFE: [[COPY4:%[0-9]+]]:_(s32) = COPY $vgpr4
+    ; GFX10-UNSAFE: [[COPY5:%[0-9]+]]:_(s32) = COPY $vgpr5
+    ; GFX10-UNSAFE: [[BUILD_VECTOR1:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32)
+    ; GFX10-UNSAFE: [[COPY6:%[0-9]+]]:_(s32) = COPY $vgpr6
+    ; GFX10-UNSAFE: [[COPY7:%[0-9]+]]:_(s32) = COPY $vgpr7
+    ; GFX10-UNSAFE: [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr8
+    ; GFX10-UNSAFE: [[BUILD_VECTOR2:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[COPY6]](s32), [[COPY7]](s32), [[COPY8]](s32)
+    ; GFX10-UNSAFE: [[COPY9:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX10-UNSAFE: [[FMA:%[0-9]+]]:_(<3 x s32>) = G_FMA [[BUILD_VECTOR]], [[BUILD_VECTOR1]], [[BUILD_VECTOR2]]
+    ; GFX10-UNSAFE: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FMA]](<3 x s32>)
+    ; GFX10-UNSAFE: $vgpr0 = COPY [[UV]](s32)
+    ; GFX10-UNSAFE: $vgpr1 = COPY [[UV1]](s32)
+    ; GFX10-UNSAFE: $vgpr2 = COPY [[UV2]](s32)
+    ; GFX10-UNSAFE: [[COPY10:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY9]]
+    ; GFX10-UNSAFE: S_SETPC_B64_return [[COPY10]], implicit $vgpr0, implicit $vgpr1, implicit $vgpr2
+    %4:_(s32) = COPY $vgpr0
+    %5:_(s32) = COPY $vgpr1
+    %6:_(s32) = COPY $vgpr2
+    %0:_(<3 x s32>) = G_BUILD_VECTOR %4(s32), %5(s32), %6(s32)
+    %7:_(s32) = COPY $vgpr3
+    %8:_(s32) = COPY $vgpr4
+    %9:_(s32) = COPY $vgpr5
+    %1:_(<3 x s32>) = G_BUILD_VECTOR %7(s32), %8(s32), %9(s32)
+    %10:_(s32) = COPY $vgpr6
+    %11:_(s32) = COPY $vgpr7
+    %12:_(s32) = COPY $vgpr8
+    %2:_(<3 x s32>) = G_BUILD_VECTOR %10(s32), %11(s32), %12(s32)
+    %3:sgpr_64 = COPY $sgpr30_sgpr31
+    %13:_(<3 x s32>) = reassoc G_FMUL %0, %1
+    %14:_(<3 x s32>) = reassoc G_FADD %2, %13
+    %16:_(s32), %17:_(s32), %18:_(s32) = G_UNMERGE_VALUES %14(<3 x s32>)
+    $vgpr0 = COPY %16(s32)
+    $vgpr1 = COPY %17(s32)
+    $vgpr2 = COPY %18(s32)
+    %15:ccr_sgpr_64 = COPY %3
+    S_SETPC_B64_return %15, implicit $vgpr0, implicit $vgpr1, implicit $vgpr2
+...
+
+---
+name:            test_4xhalf_add_mul
+body:             |
+  bb.1.entry:
+    liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $sgpr30_sgpr31
+
+    ; GFX9-LABEL: name: test_4xhalf_add_mul
+    ; GFX9: [[COPY:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0
+    ; GFX9: [[COPY1:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr1
+    ; GFX9: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[COPY]](<2 x s16>), [[COPY1]](<2 x s16>)
+    ; GFX9: [[COPY2:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr2
+    ; GFX9: [[COPY3:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr3
+    ; GFX9: [[CONCAT_VECTORS1:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[COPY2]](<2 x s16>), [[COPY3]](<2 x s16>)
+    ; GFX9: [[COPY4:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr4
+    ; GFX9: [[COPY5:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr5
+    ; GFX9: [[CONCAT_VECTORS2:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[COPY4]](<2 x s16>), [[COPY5]](<2 x s16>)
+    ; GFX9: [[COPY6:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX9: [[FMUL:%[0-9]+]]:_(<4 x s16>) = reassoc G_FMUL [[CONCAT_VECTORS]], [[CONCAT_VECTORS1]]
+    ; GFX9: [[FADD:%[0-9]+]]:_(<4 x s16>) = reassoc G_FADD [[FMUL]], [[CONCAT_VECTORS2]]
+    ; GFX9: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[FADD]](<4 x s16>)
+    ; GFX9: $vgpr0 = COPY [[UV]](<2 x s16>)
+    ; GFX9: $vgpr1 = COPY [[UV1]](<2 x s16>)
+    ; GFX9: [[COPY7:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY6]]
+    ; GFX9: S_SETPC_B64_return [[COPY7]], implicit $vgpr0, implicit $vgpr1
+    ; GFX9-CONTRACT-LABEL: name: test_4xhalf_add_mul
+    ; GFX9-CONTRACT: [[COPY:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0
+    ; GFX9-CONTRACT: [[COPY1:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr1
+    ; GFX9-CONTRACT: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[COPY]](<2 x s16>), [[COPY1]](<2 x s16>)
+    ; GFX9-CONTRACT: [[COPY2:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr2
+    ; GFX9-CONTRACT: [[COPY3:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr3
+    ; GFX9-CONTRACT: [[CONCAT_VECTORS1:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[COPY2]](<2 x s16>), [[COPY3]](<2 x s16>)
+    ; GFX9-CONTRACT: [[COPY4:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr4
+    ; GFX9-CONTRACT: [[COPY5:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr5
+    ; GFX9-CONTRACT: [[CONCAT_VECTORS2:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[COPY4]](<2 x s16>), [[COPY5]](<2 x s16>)
+    ; GFX9-CONTRACT: [[COPY6:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX9-CONTRACT: [[FMA:%[0-9]+]]:_(<4 x s16>) = G_FMA [[CONCAT_VECTORS]], [[CONCAT_VECTORS1]], [[CONCAT_VECTORS2]]
+    ; GFX9-CONTRACT: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[FMA]](<4 x s16>)
+    ; GFX9-CONTRACT: $vgpr0 = COPY [[UV]](<2 x s16>)
+    ; GFX9-CONTRACT: $vgpr1 = COPY [[UV1]](<2 x s16>)
+    ; GFX9-CONTRACT: [[COPY7:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY6]]
+    ; GFX9-CONTRACT: S_SETPC_B64_return [[COPY7]], implicit $vgpr0, implicit $vgpr1
+    ; GFX9-DENORM-LABEL: name: test_4xhalf_add_mul
+    ; GFX9-DENORM: [[COPY:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0
+    ; GFX9-DENORM: [[COPY1:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr1
+    ; GFX9-DENORM: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[COPY]](<2 x s16>), [[COPY1]](<2 x s16>)
+    ; GFX9-DENORM: [[COPY2:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr2
+    ; GFX9-DENORM: [[COPY3:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr3
+    ; GFX9-DENORM: [[CONCAT_VECTORS1:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[COPY2]](<2 x s16>), [[COPY3]](<2 x s16>)
+    ; GFX9-DENORM: [[COPY4:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr4
+    ; GFX9-DENORM: [[COPY5:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr5
+    ; GFX9-DENORM: [[CONCAT_VECTORS2:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[COPY4]](<2 x s16>), [[COPY5]](<2 x s16>)
+    ; GFX9-DENORM: [[COPY6:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX9-DENORM: [[FMUL:%[0-9]+]]:_(<4 x s16>) = reassoc G_FMUL [[CONCAT_VECTORS]], [[CONCAT_VECTORS1]]
+    ; GFX9-DENORM: [[FADD:%[0-9]+]]:_(<4 x s16>) = reassoc G_FADD [[FMUL]], [[CONCAT_VECTORS2]]
+    ; GFX9-DENORM: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[FADD]](<4 x s16>)
+    ; GFX9-DENORM: $vgpr0 = COPY [[UV]](<2 x s16>)
+    ; GFX9-DENORM: $vgpr1 = COPY [[UV1]](<2 x s16>)
+    ; GFX9-DENORM: [[COPY7:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY6]]
+    ; GFX9-DENORM: S_SETPC_B64_return [[COPY7]], implicit $vgpr0, implicit $vgpr1
+    ; GFX9-UNSAFE-LABEL: name: test_4xhalf_add_mul
+    ; GFX9-UNSAFE: [[COPY:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0
+    ; GFX9-UNSAFE: [[COPY1:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr1
+    ; GFX9-UNSAFE: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[COPY]](<2 x s16>), [[COPY1]](<2 x s16>)
+    ; GFX9-UNSAFE: [[COPY2:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr2
+    ; GFX9-UNSAFE: [[COPY3:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr3
+    ; GFX9-UNSAFE: [[CONCAT_VECTORS1:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[COPY2]](<2 x s16>), [[COPY3]](<2 x s16>)
+    ; GFX9-UNSAFE: [[COPY4:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr4
+    ; GFX9-UNSAFE: [[COPY5:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr5
+    ; GFX9-UNSAFE: [[CONCAT_VECTORS2:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[COPY4]](<2 x s16>), [[COPY5]](<2 x s16>)
+    ; GFX9-UNSAFE: [[COPY6:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX9-UNSAFE: [[FMA:%[0-9]+]]:_(<4 x s16>) = G_FMA [[CONCAT_VECTORS]], [[CONCAT_VECTORS1]], [[CONCAT_VECTORS2]]
+    ; GFX9-UNSAFE: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[FMA]](<4 x s16>)
+    ; GFX9-UNSAFE: $vgpr0 = COPY [[UV]](<2 x s16>)
+    ; GFX9-UNSAFE: $vgpr1 = COPY [[UV1]](<2 x s16>)
+    ; GFX9-UNSAFE: [[COPY7:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY6]]
+    ; GFX9-UNSAFE: S_SETPC_B64_return [[COPY7]], implicit $vgpr0, implicit $vgpr1
+    ; GFX10-LABEL: name: test_4xhalf_add_mul
+    ; GFX10: [[COPY:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0
+    ; GFX10: [[COPY1:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr1
+    ; GFX10: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[COPY]](<2 x s16>), [[COPY1]](<2 x s16>)
+    ; GFX10: [[COPY2:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr2
+    ; GFX10: [[COPY3:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr3
+    ; GFX10: [[CONCAT_VECTORS1:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[COPY2]](<2 x s16>), [[COPY3]](<2 x s16>)
+    ; GFX10: [[COPY4:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr4
+    ; GFX10: [[COPY5:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr5
+    ; GFX10: [[CONCAT_VECTORS2:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[COPY4]](<2 x s16>), [[COPY5]](<2 x s16>)
+    ; GFX10: [[COPY6:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX10: [[FMUL:%[0-9]+]]:_(<4 x s16>) = reassoc G_FMUL [[CONCAT_VECTORS]], [[CONCAT_VECTORS1]]
+    ; GFX10: [[FADD:%[0-9]+]]:_(<4 x s16>) = reassoc G_FADD [[FMUL]], [[CONCAT_VECTORS2]]
+    ; GFX10: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[FADD]](<4 x s16>)
+    ; GFX10: $vgpr0 = COPY [[UV]](<2 x s16>)
+    ; GFX10: $vgpr1 = COPY [[UV1]](<2 x s16>)
+    ; GFX10: [[COPY7:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY6]]
+    ; GFX10: S_SETPC_B64_return [[COPY7]], implicit $vgpr0, implicit $vgpr1
+    ; GFX10-CONTRACT-LABEL: name: test_4xhalf_add_mul
+    ; GFX10-CONTRACT: [[COPY:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0
+    ; GFX10-CONTRACT: [[COPY1:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr1
+    ; GFX10-CONTRACT: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[COPY]](<2 x s16>), [[COPY1]](<2 x s16>)
+    ; GFX10-CONTRACT: [[COPY2:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr2
+    ; GFX10-CONTRACT: [[COPY3:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr3
+    ; GFX10-CONTRACT: [[CONCAT_VECTORS1:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[COPY2]](<2 x s16>), [[COPY3]](<2 x s16>)
+    ; GFX10-CONTRACT: [[COPY4:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr4
+    ; GFX10-CONTRACT: [[COPY5:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr5
+    ; GFX10-CONTRACT: [[CONCAT_VECTORS2:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[COPY4]](<2 x s16>), [[COPY5]](<2 x s16>)
+    ; GFX10-CONTRACT: [[COPY6:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX10-CONTRACT: [[FMA:%[0-9]+]]:_(<4 x s16>) = G_FMA [[CONCAT_VECTORS]], [[CONCAT_VECTORS1]], [[CONCAT_VECTORS2]]
+    ; GFX10-CONTRACT: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[FMA]](<4 x s16>)
+    ; GFX10-CONTRACT: $vgpr0 = COPY [[UV]](<2 x s16>)
+    ; GFX10-CONTRACT: $vgpr1 = COPY [[UV1]](<2 x s16>)
+    ; GFX10-CONTRACT: [[COPY7:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY6]]
+    ; GFX10-CONTRACT: S_SETPC_B64_return [[COPY7]], implicit $vgpr0, implicit $vgpr1
+    ; GFX10-DENORM-LABEL: name: test_4xhalf_add_mul
+    ; GFX10-DENORM: [[COPY:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0
+    ; GFX10-DENORM: [[COPY1:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr1
+    ; GFX10-DENORM: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[COPY]](<2 x s16>), [[COPY1]](<2 x s16>)
+    ; GFX10-DENORM: [[COPY2:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr2
+    ; GFX10-DENORM: [[COPY3:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr3
+    ; GFX10-DENORM: [[CONCAT_VECTORS1:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[COPY2]](<2 x s16>), [[COPY3]](<2 x s16>)
+    ; GFX10-DENORM: [[COPY4:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr4
+    ; GFX10-DENORM: [[COPY5:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr5
+    ; GFX10-DENORM: [[CONCAT_VECTORS2:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[COPY4]](<2 x s16>), [[COPY5]](<2 x s16>)
+    ; GFX10-DENORM: [[COPY6:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX10-DENORM: [[FMUL:%[0-9]+]]:_(<4 x s16>) = reassoc G_FMUL [[CONCAT_VECTORS]], [[CONCAT_VECTORS1]]
+    ; GFX10-DENORM: [[FADD:%[0-9]+]]:_(<4 x s16>) = reassoc G_FADD [[FMUL]], [[CONCAT_VECTORS2]]
+    ; GFX10-DENORM: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[FADD]](<4 x s16>)
+    ; GFX10-DENORM: $vgpr0 = COPY [[UV]](<2 x s16>)
+    ; GFX10-DENORM: $vgpr1 = COPY [[UV1]](<2 x s16>)
+    ; GFX10-DENORM: [[COPY7:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY6]]
+    ; GFX10-DENORM: S_SETPC_B64_return [[COPY7]], implicit $vgpr0, implicit $vgpr1
+    ; GFX10-UNSAFE-LABEL: name: test_4xhalf_add_mul
+    ; GFX10-UNSAFE: [[COPY:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0
+    ; GFX10-UNSAFE: [[COPY1:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr1
+    ; GFX10-UNSAFE: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[COPY]](<2 x s16>), [[COPY1]](<2 x s16>)
+    ; GFX10-UNSAFE: [[COPY2:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr2
+    ; GFX10-UNSAFE: [[COPY3:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr3
+    ; GFX10-UNSAFE: [[CONCAT_VECTORS1:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[COPY2]](<2 x s16>), [[COPY3]](<2 x s16>)
+    ; GFX10-UNSAFE: [[COPY4:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr4
+    ; GFX10-UNSAFE: [[COPY5:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr5
+    ; GFX10-UNSAFE: [[CONCAT_VECTORS2:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[COPY4]](<2 x s16>), [[COPY5]](<2 x s16>)
+    ; GFX10-UNSAFE: [[COPY6:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX10-UNSAFE: [[FMA:%[0-9]+]]:_(<4 x s16>) = G_FMA [[CONCAT_VECTORS]], [[CONCAT_VECTORS1]], [[CONCAT_VECTORS2]]
+    ; GFX10-UNSAFE: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[FMA]](<4 x s16>)
+    ; GFX10-UNSAFE: $vgpr0 = COPY [[UV]](<2 x s16>)
+    ; GFX10-UNSAFE: $vgpr1 = COPY [[UV1]](<2 x s16>)
+    ; GFX10-UNSAFE: [[COPY7:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY6]]
+    ; GFX10-UNSAFE: S_SETPC_B64_return [[COPY7]], implicit $vgpr0, implicit $vgpr1
+    %4:_(<2 x s16>) = COPY $vgpr0
+    %5:_(<2 x s16>) = COPY $vgpr1
+    %0:_(<4 x s16>) = G_CONCAT_VECTORS %4(<2 x s16>), %5(<2 x s16>)
+    %6:_(<2 x s16>) = COPY $vgpr2
+    %7:_(<2 x s16>) = COPY $vgpr3
+    %1:_(<4 x s16>) = G_CONCAT_VECTORS %6(<2 x s16>), %7(<2 x s16>)
+    %8:_(<2 x s16>) = COPY $vgpr4
+    %9:_(<2 x s16>) = COPY $vgpr5
+    %2:_(<4 x s16>) = G_CONCAT_VECTORS %8(<2 x s16>), %9(<2 x s16>)
+    %3:sgpr_64 = COPY $sgpr30_sgpr31
+    %10:_(<4 x s16>) = reassoc G_FMUL %0, %1
+    %11:_(<4 x s16>) = reassoc G_FADD %10, %2
+    %13:_(<2 x s16>), %14:_(<2 x s16>) = G_UNMERGE_VALUES %11(<4 x s16>)
+    $vgpr0 = COPY %13(<2 x s16>)
+    $vgpr1 = COPY %14(<2 x s16>)
+    %12:ccr_sgpr_64 = COPY %3
+    S_SETPC_B64_return %12, implicit $vgpr0, implicit $vgpr1
+...
+
+---
+name:            test_3xhalf_add_mul_rhs
+body:             |
+  bb.1.entry:
+    liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $sgpr30_sgpr31
+
+    ; GFX9-LABEL: name: test_3xhalf_add_mul_rhs
+    ; GFX9: [[COPY:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0
+    ; GFX9: [[COPY1:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr1
+    ; GFX9: [[DEF:%[0-9]+]]:_(<2 x s16>) = G_IMPLICIT_DEF
+    ; GFX9: [[CONCAT_VECTORS:%[0-9]+]]:_(<6 x s16>) = G_CONCAT_VECTORS [[COPY]](<2 x s16>), [[COPY1]](<2 x s16>), [[DEF]](<2 x s16>)
+    ; GFX9: [[BITCAST:%[0-9]+]]:_(s96) = G_BITCAST [[CONCAT_VECTORS]](<6 x s16>)
+    ; GFX9: [[TRUNC:%[0-9]+]]:_(s48) = G_TRUNC [[BITCAST]](s96)
+    ; GFX9: [[BITCAST1:%[0-9]+]]:_(<3 x s16>) = G_BITCAST [[TRUNC]](s48)
+    ; GFX9: [[COPY2:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr2
+    ; GFX9: [[COPY3:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr3
+    ; GFX9: [[CONCAT_VECTORS1:%[0-9]+]]:_(<6 x s16>) = G_CONCAT_VECTORS [[COPY2]](<2 x s16>), [[COPY3]](<2 x s16>), [[DEF]](<2 x s16>)
+    ; GFX9: [[BITCAST2:%[0-9]+]]:_(s96) = G_BITCAST [[CONCAT_VECTORS1]](<6 x s16>)
+    ; GFX9: [[TRUNC1:%[0-9]+]]:_(s48) = G_TRUNC [[BITCAST2]](s96)
+    ; GFX9: [[BITCAST3:%[0-9]+]]:_(<3 x s16>) = G_BITCAST [[TRUNC1]](s48)
+    ; GFX9: [[COPY4:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr4
+    ; GFX9: [[COPY5:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr5
+    ; GFX9: [[CONCAT_VECTORS2:%[0-9]+]]:_(<6 x s16>) = G_CONCAT_VECTORS [[COPY4]](<2 x s16>), [[COPY5]](<2 x s16>), [[DEF]](<2 x s16>)
+    ; GFX9: [[BITCAST4:%[0-9]+]]:_(s96) = G_BITCAST [[CONCAT_VECTORS2]](<6 x s16>)
+    ; GFX9: [[TRUNC2:%[0-9]+]]:_(s48) = G_TRUNC [[BITCAST4]](s96)
+    ; GFX9: [[BITCAST5:%[0-9]+]]:_(<3 x s16>) = G_BITCAST [[TRUNC2]](s48)
+    ; GFX9: [[COPY6:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX9: [[FMUL:%[0-9]+]]:_(<3 x s16>) = reassoc G_FMUL [[BITCAST1]], [[BITCAST3]]
+    ; GFX9: [[FADD:%[0-9]+]]:_(<3 x s16>) = reassoc G_FADD [[BITCAST5]], [[FMUL]]
+    ; GFX9: [[DEF1:%[0-9]+]]:_(<3 x s16>) = G_IMPLICIT_DEF
+    ; GFX9: [[CONCAT_VECTORS3:%[0-9]+]]:_(<6 x s16>) = G_CONCAT_VECTORS [[FADD]](<3 x s16>), [[DEF1]](<3 x s16>)
+    ; GFX9: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>), [[UV2:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[CONCAT_VECTORS3]](<6 x s16>)
+    ; GFX9: $vgpr0 = COPY [[UV]](<2 x s16>)
+    ; GFX9: $vgpr1 = COPY [[UV1]](<2 x s16>)
+    ; GFX9: [[COPY7:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY6]]
+    ; GFX9: S_SETPC_B64_return [[COPY7]], implicit $vgpr0, implicit $vgpr1
+    ; GFX9-CONTRACT-LABEL: name: test_3xhalf_add_mul_rhs
+    ; GFX9-CONTRACT: [[COPY:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0
+    ; GFX9-CONTRACT: [[COPY1:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr1
+    ; GFX9-CONTRACT: [[DEF:%[0-9]+]]:_(<2 x s16>) = G_IMPLICIT_DEF
+    ; GFX9-CONTRACT: [[CONCAT_VECTORS:%[0-9]+]]:_(<6 x s16>) = G_CONCAT_VECTORS [[COPY]](<2 x s16>), [[COPY1]](<2 x s16>), [[DEF]](<2 x s16>)
+    ; GFX9-CONTRACT: [[BITCAST:%[0-9]+]]:_(s96) = G_BITCAST [[CONCAT_VECTORS]](<6 x s16>)
+    ; GFX9-CONTRACT: [[TRUNC:%[0-9]+]]:_(s48) = G_TRUNC [[BITCAST]](s96)
+    ; GFX9-CONTRACT: [[BITCAST1:%[0-9]+]]:_(<3 x s16>) = G_BITCAST [[TRUNC]](s48)
+    ; GFX9-CONTRACT: [[COPY2:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr2
+    ; GFX9-CONTRACT: [[COPY3:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr3
+    ; GFX9-CONTRACT: [[CONCAT_VECTORS1:%[0-9]+]]:_(<6 x s16>) = G_CONCAT_VECTORS [[COPY2]](<2 x s16>), [[COPY3]](<2 x s16>), [[DEF]](<2 x s16>)
+    ; GFX9-CONTRACT: [[BITCAST2:%[0-9]+]]:_(s96) = G_BITCAST [[CONCAT_VECTORS1]](<6 x s16>)
+    ; GFX9-CONTRACT: [[TRUNC1:%[0-9]+]]:_(s48) = G_TRUNC [[BITCAST2]](s96)
+    ; GFX9-CONTRACT: [[BITCAST3:%[0-9]+]]:_(<3 x s16>) = G_BITCAST [[TRUNC1]](s48)
+    ; GFX9-CONTRACT: [[COPY4:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr4
+    ; GFX9-CONTRACT: [[COPY5:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr5
+    ; GFX9-CONTRACT: [[CONCAT_VECTORS2:%[0-9]+]]:_(<6 x s16>) = G_CONCAT_VECTORS [[COPY4]](<2 x s16>), [[COPY5]](<2 x s16>), [[DEF]](<2 x s16>)
+    ; GFX9-CONTRACT: [[BITCAST4:%[0-9]+]]:_(s96) = G_BITCAST [[CONCAT_VECTORS2]](<6 x s16>)
+    ; GFX9-CONTRACT: [[TRUNC2:%[0-9]+]]:_(s48) = G_TRUNC [[BITCAST4]](s96)
+    ; GFX9-CONTRACT: [[BITCAST5:%[0-9]+]]:_(<3 x s16>) = G_BITCAST [[TRUNC2]](s48)
+    ; GFX9-CONTRACT: [[COPY6:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX9-CONTRACT: [[FMA:%[0-9]+]]:_(<3 x s16>) = G_FMA [[BITCAST1]], [[BITCAST3]], [[BITCAST5]]
+    ; GFX9-CONTRACT: [[DEF1:%[0-9]+]]:_(<3 x s16>) = G_IMPLICIT_DEF
+    ; GFX9-CONTRACT: [[CONCAT_VECTORS3:%[0-9]+]]:_(<6 x s16>) = G_CONCAT_VECTORS [[FMA]](<3 x s16>), [[DEF1]](<3 x s16>)
+    ; GFX9-CONTRACT: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>), [[UV2:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[CONCAT_VECTORS3]](<6 x s16>)
+    ; GFX9-CONTRACT: $vgpr0 = COPY [[UV]](<2 x s16>)
+    ; GFX9-CONTRACT: $vgpr1 = COPY [[UV1]](<2 x s16>)
+    ; GFX9-CONTRACT: [[COPY7:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY6]]
+    ; GFX9-CONTRACT: S_SETPC_B64_return [[COPY7]], implicit $vgpr0, implicit $vgpr1
+    ; GFX9-DENORM-LABEL: name: test_3xhalf_add_mul_rhs
+    ; GFX9-DENORM: [[COPY:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0
+    ; GFX9-DENORM: [[COPY1:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr1
+    ; GFX9-DENORM: [[DEF:%[0-9]+]]:_(<2 x s16>) = G_IMPLICIT_DEF
+    ; GFX9-DENORM: [[CONCAT_VECTORS:%[0-9]+]]:_(<6 x s16>) = G_CONCAT_VECTORS [[COPY]](<2 x s16>), [[COPY1]](<2 x s16>), [[DEF]](<2 x s16>)
+    ; GFX9-DENORM: [[BITCAST:%[0-9]+]]:_(s96) = G_BITCAST [[CONCAT_VECTORS]](<6 x s16>)
+    ; GFX9-DENORM: [[TRUNC:%[0-9]+]]:_(s48) = G_TRUNC [[BITCAST]](s96)
+    ; GFX9-DENORM: [[BITCAST1:%[0-9]+]]:_(<3 x s16>) = G_BITCAST [[TRUNC]](s48)
+    ; GFX9-DENORM: [[COPY2:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr2
+    ; GFX9-DENORM: [[COPY3:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr3
+    ; GFX9-DENORM: [[CONCAT_VECTORS1:%[0-9]+]]:_(<6 x s16>) = G_CONCAT_VECTORS [[COPY2]](<2 x s16>), [[COPY3]](<2 x s16>), [[DEF]](<2 x s16>)
+    ; GFX9-DENORM: [[BITCAST2:%[0-9]+]]:_(s96) = G_BITCAST [[CONCAT_VECTORS1]](<6 x s16>)
+    ; GFX9-DENORM: [[TRUNC1:%[0-9]+]]:_(s48) = G_TRUNC [[BITCAST2]](s96)
+    ; GFX9-DENORM: [[BITCAST3:%[0-9]+]]:_(<3 x s16>) = G_BITCAST [[TRUNC1]](s48)
+    ; GFX9-DENORM: [[COPY4:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr4
+    ; GFX9-DENORM: [[COPY5:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr5
+    ; GFX9-DENORM: [[CONCAT_VECTORS2:%[0-9]+]]:_(<6 x s16>) = G_CONCAT_VECTORS [[COPY4]](<2 x s16>), [[COPY5]](<2 x s16>), [[DEF]](<2 x s16>)
+    ; GFX9-DENORM: [[BITCAST4:%[0-9]+]]:_(s96) = G_BITCAST [[CONCAT_VECTORS2]](<6 x s16>)
+    ; GFX9-DENORM: [[TRUNC2:%[0-9]+]]:_(s48) = G_TRUNC [[BITCAST4]](s96)
+    ; GFX9-DENORM: [[BITCAST5:%[0-9]+]]:_(<3 x s16>) = G_BITCAST [[TRUNC2]](s48)
+    ; GFX9-DENORM: [[COPY6:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX9-DENORM: [[FMUL:%[0-9]+]]:_(<3 x s16>) = reassoc G_FMUL [[BITCAST1]], [[BITCAST3]]
+    ; GFX9-DENORM: [[FADD:%[0-9]+]]:_(<3 x s16>) = reassoc G_FADD [[BITCAST5]], [[FMUL]]
+    ; GFX9-DENORM: [[DEF1:%[0-9]+]]:_(<3 x s16>) = G_IMPLICIT_DEF
+    ; GFX9-DENORM: [[CONCAT_VECTORS3:%[0-9]+]]:_(<6 x s16>) = G_CONCAT_VECTORS [[FADD]](<3 x s16>), [[DEF1]](<3 x s16>)
+    ; GFX9-DENORM: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>), [[UV2:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[CONCAT_VECTORS3]](<6 x s16>)
+    ; GFX9-DENORM: $vgpr0 = COPY [[UV]](<2 x s16>)
+    ; GFX9-DENORM: $vgpr1 = COPY [[UV1]](<2 x s16>)
+    ; GFX9-DENORM: [[COPY7:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY6]]
+    ; GFX9-DENORM: S_SETPC_B64_return [[COPY7]], implicit $vgpr0, implicit $vgpr1
+    ; GFX9-UNSAFE-LABEL: name: test_3xhalf_add_mul_rhs
+    ; GFX9-UNSAFE: [[COPY:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0
+    ; GFX9-UNSAFE: [[COPY1:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr1
+    ; GFX9-UNSAFE: [[DEF:%[0-9]+]]:_(<2 x s16>) = G_IMPLICIT_DEF
+    ; GFX9-UNSAFE: [[CONCAT_VECTORS:%[0-9]+]]:_(<6 x s16>) = G_CONCAT_VECTORS [[COPY]](<2 x s16>), [[COPY1]](<2 x s16>), [[DEF]](<2 x s16>)
+    ; GFX9-UNSAFE: [[BITCAST:%[0-9]+]]:_(s96) = G_BITCAST [[CONCAT_VECTORS]](<6 x s16>)
+    ; GFX9-UNSAFE: [[TRUNC:%[0-9]+]]:_(s48) = G_TRUNC [[BITCAST]](s96)
+    ; GFX9-UNSAFE: [[BITCAST1:%[0-9]+]]:_(<3 x s16>) = G_BITCAST [[TRUNC]](s48)
+    ; GFX9-UNSAFE: [[COPY2:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr2
+    ; GFX9-UNSAFE: [[COPY3:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr3
+    ; GFX9-UNSAFE: [[CONCAT_VECTORS1:%[0-9]+]]:_(<6 x s16>) = G_CONCAT_VECTORS [[COPY2]](<2 x s16>), [[COPY3]](<2 x s16>), [[DEF]](<2 x s16>)
+    ; GFX9-UNSAFE: [[BITCAST2:%[0-9]+]]:_(s96) = G_BITCAST [[CONCAT_VECTORS1]](<6 x s16>)
+    ; GFX9-UNSAFE: [[TRUNC1:%[0-9]+]]:_(s48) = G_TRUNC [[BITCAST2]](s96)
+    ; GFX9-UNSAFE: [[BITCAST3:%[0-9]+]]:_(<3 x s16>) = G_BITCAST [[TRUNC1]](s48)
+    ; GFX9-UNSAFE: [[COPY4:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr4
+    ; GFX9-UNSAFE: [[COPY5:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr5
+    ; GFX9-UNSAFE: [[CONCAT_VECTORS2:%[0-9]+]]:_(<6 x s16>) = G_CONCAT_VECTORS [[COPY4]](<2 x s16>), [[COPY5]](<2 x s16>), [[DEF]](<2 x s16>)
+    ; GFX9-UNSAFE: [[BITCAST4:%[0-9]+]]:_(s96) = G_BITCAST [[CONCAT_VECTORS2]](<6 x s16>)
+    ; GFX9-UNSAFE: [[TRUNC2:%[0-9]+]]:_(s48) = G_TRUNC [[BITCAST4]](s96)
+    ; GFX9-UNSAFE: [[BITCAST5:%[0-9]+]]:_(<3 x s16>) = G_BITCAST [[TRUNC2]](s48)
+    ; GFX9-UNSAFE: [[COPY6:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX9-UNSAFE: [[FMA:%[0-9]+]]:_(<3 x s16>) = G_FMA [[BITCAST1]], [[BITCAST3]], [[BITCAST5]]
+    ; GFX9-UNSAFE: [[DEF1:%[0-9]+]]:_(<3 x s16>) = G_IMPLICIT_DEF
+    ; GFX9-UNSAFE: [[CONCAT_VECTORS3:%[0-9]+]]:_(<6 x s16>) = G_CONCAT_VECTORS [[FMA]](<3 x s16>), [[DEF1]](<3 x s16>)
+    ; GFX9-UNSAFE: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>), [[UV2:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[CONCAT_VECTORS3]](<6 x s16>)
+    ; GFX9-UNSAFE: $vgpr0 = COPY [[UV]](<2 x s16>)
+    ; GFX9-UNSAFE: $vgpr1 = COPY [[UV1]](<2 x s16>)
+    ; GFX9-UNSAFE: [[COPY7:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY6]]
+    ; GFX9-UNSAFE: S_SETPC_B64_return [[COPY7]], implicit $vgpr0, implicit $vgpr1
+    ; GFX10-LABEL: name: test_3xhalf_add_mul_rhs
+    ; GFX10: [[COPY:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0
+    ; GFX10: [[COPY1:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr1
+    ; GFX10: [[DEF:%[0-9]+]]:_(<2 x s16>) = G_IMPLICIT_DEF
+    ; GFX10: [[CONCAT_VECTORS:%[0-9]+]]:_(<6 x s16>) = G_CONCAT_VECTORS [[COPY]](<2 x s16>), [[COPY1]](<2 x s16>), [[DEF]](<2 x s16>)
+    ; GFX10: [[BITCAST:%[0-9]+]]:_(s96) = G_BITCAST [[CONCAT_VECTORS]](<6 x s16>)
+    ; GFX10: [[TRUNC:%[0-9]+]]:_(s48) = G_TRUNC [[BITCAST]](s96)
+    ; GFX10: [[BITCAST1:%[0-9]+]]:_(<3 x s16>) = G_BITCAST [[TRUNC]](s48)
+    ; GFX10: [[COPY2:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr2
+    ; GFX10: [[COPY3:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr3
+    ; GFX10: [[CONCAT_VECTORS1:%[0-9]+]]:_(<6 x s16>) = G_CONCAT_VECTORS [[COPY2]](<2 x s16>), [[COPY3]](<2 x s16>), [[DEF]](<2 x s16>)
+    ; GFX10: [[BITCAST2:%[0-9]+]]:_(s96) = G_BITCAST [[CONCAT_VECTORS1]](<6 x s16>)
+    ; GFX10: [[TRUNC1:%[0-9]+]]:_(s48) = G_TRUNC [[BITCAST2]](s96)
+    ; GFX10: [[BITCAST3:%[0-9]+]]:_(<3 x s16>) = G_BITCAST [[TRUNC1]](s48)
+    ; GFX10: [[COPY4:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr4
+    ; GFX10: [[COPY5:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr5
+    ; GFX10: [[CONCAT_VECTORS2:%[0-9]+]]:_(<6 x s16>) = G_CONCAT_VECTORS [[COPY4]](<2 x s16>), [[COPY5]](<2 x s16>), [[DEF]](<2 x s16>)
+    ; GFX10: [[BITCAST4:%[0-9]+]]:_(s96) = G_BITCAST [[CONCAT_VECTORS2]](<6 x s16>)
+    ; GFX10: [[TRUNC2:%[0-9]+]]:_(s48) = G_TRUNC [[BITCAST4]](s96)
+    ; GFX10: [[BITCAST5:%[0-9]+]]:_(<3 x s16>) = G_BITCAST [[TRUNC2]](s48)
+    ; GFX10: [[COPY6:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX10: [[FMUL:%[0-9]+]]:_(<3 x s16>) = reassoc G_FMUL [[BITCAST1]], [[BITCAST3]]
+    ; GFX10: [[FADD:%[0-9]+]]:_(<3 x s16>) = reassoc G_FADD [[BITCAST5]], [[FMUL]]
+    ; GFX10: [[DEF1:%[0-9]+]]:_(<3 x s16>) = G_IMPLICIT_DEF
+    ; GFX10: [[CONCAT_VECTORS3:%[0-9]+]]:_(<6 x s16>) = G_CONCAT_VECTORS [[FADD]](<3 x s16>), [[DEF1]](<3 x s16>)
+    ; GFX10: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>), [[UV2:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[CONCAT_VECTORS3]](<6 x s16>)
+    ; GFX10: $vgpr0 = COPY [[UV]](<2 x s16>)
+    ; GFX10: $vgpr1 = COPY [[UV1]](<2 x s16>)
+    ; GFX10: [[COPY7:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY6]]
+    ; GFX10: S_SETPC_B64_return [[COPY7]], implicit $vgpr0, implicit $vgpr1
+    ; GFX10-CONTRACT-LABEL: name: test_3xhalf_add_mul_rhs
+    ; GFX10-CONTRACT: [[COPY:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0
+    ; GFX10-CONTRACT: [[COPY1:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr1
+    ; GFX10-CONTRACT: [[DEF:%[0-9]+]]:_(<2 x s16>) = G_IMPLICIT_DEF
+    ; GFX10-CONTRACT: [[CONCAT_VECTORS:%[0-9]+]]:_(<6 x s16>) = G_CONCAT_VECTORS [[COPY]](<2 x s16>), [[COPY1]](<2 x s16>), [[DEF]](<2 x s16>)
+    ; GFX10-CONTRACT: [[BITCAST:%[0-9]+]]:_(s96) = G_BITCAST [[CONCAT_VECTORS]](<6 x s16>)
+    ; GFX10-CONTRACT: [[TRUNC:%[0-9]+]]:_(s48) = G_TRUNC [[BITCAST]](s96)
+    ; GFX10-CONTRACT: [[BITCAST1:%[0-9]+]]:_(<3 x s16>) = G_BITCAST [[TRUNC]](s48)
+    ; GFX10-CONTRACT: [[COPY2:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr2
+    ; GFX10-CONTRACT: [[COPY3:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr3
+    ; GFX10-CONTRACT: [[CONCAT_VECTORS1:%[0-9]+]]:_(<6 x s16>) = G_CONCAT_VECTORS [[COPY2]](<2 x s16>), [[COPY3]](<2 x s16>), [[DEF]](<2 x s16>)
+    ; GFX10-CONTRACT: [[BITCAST2:%[0-9]+]]:_(s96) = G_BITCAST [[CONCAT_VECTORS1]](<6 x s16>)
+    ; GFX10-CONTRACT: [[TRUNC1:%[0-9]+]]:_(s48) = G_TRUNC [[BITCAST2]](s96)
+    ; GFX10-CONTRACT: [[BITCAST3:%[0-9]+]]:_(<3 x s16>) = G_BITCAST [[TRUNC1]](s48)
+    ; GFX10-CONTRACT: [[COPY4:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr4
+    ; GFX10-CONTRACT: [[COPY5:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr5
+    ; GFX10-CONTRACT: [[CONCAT_VECTORS2:%[0-9]+]]:_(<6 x s16>) = G_CONCAT_VECTORS [[COPY4]](<2 x s16>), [[COPY5]](<2 x s16>), [[DEF]](<2 x s16>)
+    ; GFX10-CONTRACT: [[BITCAST4:%[0-9]+]]:_(s96) = G_BITCAST [[CONCAT_VECTORS2]](<6 x s16>)
+    ; GFX10-CONTRACT: [[TRUNC2:%[0-9]+]]:_(s48) = G_TRUNC [[BITCAST4]](s96)
+    ; GFX10-CONTRACT: [[BITCAST5:%[0-9]+]]:_(<3 x s16>) = G_BITCAST [[TRUNC2]](s48)
+    ; GFX10-CONTRACT: [[COPY6:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX10-CONTRACT: [[FMA:%[0-9]+]]:_(<3 x s16>) = G_FMA [[BITCAST1]], [[BITCAST3]], [[BITCAST5]]
+    ; GFX10-CONTRACT: [[DEF1:%[0-9]+]]:_(<3 x s16>) = G_IMPLICIT_DEF
+    ; GFX10-CONTRACT: [[CONCAT_VECTORS3:%[0-9]+]]:_(<6 x s16>) = G_CONCAT_VECTORS [[FMA]](<3 x s16>), [[DEF1]](<3 x s16>)
+    ; GFX10-CONTRACT: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>), [[UV2:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[CONCAT_VECTORS3]](<6 x s16>)
+    ; GFX10-CONTRACT: $vgpr0 = COPY [[UV]](<2 x s16>)
+    ; GFX10-CONTRACT: $vgpr1 = COPY [[UV1]](<2 x s16>)
+    ; GFX10-CONTRACT: [[COPY7:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY6]]
+    ; GFX10-CONTRACT: S_SETPC_B64_return [[COPY7]], implicit $vgpr0, implicit $vgpr1
+    ; GFX10-DENORM-LABEL: name: test_3xhalf_add_mul_rhs
+    ; GFX10-DENORM: [[COPY:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0
+    ; GFX10-DENORM: [[COPY1:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr1
+    ; GFX10-DENORM: [[DEF:%[0-9]+]]:_(<2 x s16>) = G_IMPLICIT_DEF
+    ; GFX10-DENORM: [[CONCAT_VECTORS:%[0-9]+]]:_(<6 x s16>) = G_CONCAT_VECTORS [[COPY]](<2 x s16>), [[COPY1]](<2 x s16>), [[DEF]](<2 x s16>)
+    ; GFX10-DENORM: [[BITCAST:%[0-9]+]]:_(s96) = G_BITCAST [[CONCAT_VECTORS]](<6 x s16>)
+    ; GFX10-DENORM: [[TRUNC:%[0-9]+]]:_(s48) = G_TRUNC [[BITCAST]](s96)
+    ; GFX10-DENORM: [[BITCAST1:%[0-9]+]]:_(<3 x s16>) = G_BITCAST [[TRUNC]](s48)
+    ; GFX10-DENORM: [[COPY2:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr2
+    ; GFX10-DENORM: [[COPY3:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr3
+    ; GFX10-DENORM: [[CONCAT_VECTORS1:%[0-9]+]]:_(<6 x s16>) = G_CONCAT_VECTORS [[COPY2]](<2 x s16>), [[COPY3]](<2 x s16>), [[DEF]](<2 x s16>)
+    ; GFX10-DENORM: [[BITCAST2:%[0-9]+]]:_(s96) = G_BITCAST [[CONCAT_VECTORS1]](<6 x s16>)
+    ; GFX10-DENORM: [[TRUNC1:%[0-9]+]]:_(s48) = G_TRUNC [[BITCAST2]](s96)
+    ; GFX10-DENORM: [[BITCAST3:%[0-9]+]]:_(<3 x s16>) = G_BITCAST [[TRUNC1]](s48)
+    ; GFX10-DENORM: [[COPY4:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr4
+    ; GFX10-DENORM: [[COPY5:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr5
+    ; GFX10-DENORM: [[CONCAT_VECTORS2:%[0-9]+]]:_(<6 x s16>) = G_CONCAT_VECTORS [[COPY4]](<2 x s16>), [[COPY5]](<2 x s16>), [[DEF]](<2 x s16>)
+    ; GFX10-DENORM: [[BITCAST4:%[0-9]+]]:_(s96) = G_BITCAST [[CONCAT_VECTORS2]](<6 x s16>)
+    ; GFX10-DENORM: [[TRUNC2:%[0-9]+]]:_(s48) = G_TRUNC [[BITCAST4]](s96)
+    ; GFX10-DENORM: [[BITCAST5:%[0-9]+]]:_(<3 x s16>) = G_BITCAST [[TRUNC2]](s48)
+    ; GFX10-DENORM: [[COPY6:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX10-DENORM: [[FMUL:%[0-9]+]]:_(<3 x s16>) = reassoc G_FMUL [[BITCAST1]], [[BITCAST3]]
+    ; GFX10-DENORM: [[FADD:%[0-9]+]]:_(<3 x s16>) = reassoc G_FADD [[BITCAST5]], [[FMUL]]
+    ; GFX10-DENORM: [[DEF1:%[0-9]+]]:_(<3 x s16>) = G_IMPLICIT_DEF
+    ; GFX10-DENORM: [[CONCAT_VECTORS3:%[0-9]+]]:_(<6 x s16>) = G_CONCAT_VECTORS [[FADD]](<3 x s16>), [[DEF1]](<3 x s16>)
+    ; GFX10-DENORM: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>), [[UV2:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[CONCAT_VECTORS3]](<6 x s16>)
+    ; GFX10-DENORM: $vgpr0 = COPY [[UV]](<2 x s16>)
+    ; GFX10-DENORM: $vgpr1 = COPY [[UV1]](<2 x s16>)
+    ; GFX10-DENORM: [[COPY7:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY6]]
+    ; GFX10-DENORM: S_SETPC_B64_return [[COPY7]], implicit $vgpr0, implicit $vgpr1
+    ; GFX10-UNSAFE-LABEL: name: test_3xhalf_add_mul_rhs
+    ; GFX10-UNSAFE: [[COPY:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0
+    ; GFX10-UNSAFE: [[COPY1:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr1
+    ; GFX10-UNSAFE: [[DEF:%[0-9]+]]:_(<2 x s16>) = G_IMPLICIT_DEF
+    ; GFX10-UNSAFE: [[CONCAT_VECTORS:%[0-9]+]]:_(<6 x s16>) = G_CONCAT_VECTORS [[COPY]](<2 x s16>), [[COPY1]](<2 x s16>), [[DEF]](<2 x s16>)
+    ; GFX10-UNSAFE: [[BITCAST:%[0-9]+]]:_(s96) = G_BITCAST [[CONCAT_VECTORS]](<6 x s16>)
+    ; GFX10-UNSAFE: [[TRUNC:%[0-9]+]]:_(s48) = G_TRUNC [[BITCAST]](s96)
+    ; GFX10-UNSAFE: [[BITCAST1:%[0-9]+]]:_(<3 x s16>) = G_BITCAST [[TRUNC]](s48)
+    ; GFX10-UNSAFE: [[COPY2:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr2
+    ; GFX10-UNSAFE: [[COPY3:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr3
+    ; GFX10-UNSAFE: [[CONCAT_VECTORS1:%[0-9]+]]:_(<6 x s16>) = G_CONCAT_VECTORS [[COPY2]](<2 x s16>), [[COPY3]](<2 x s16>), [[DEF]](<2 x s16>)
+    ; GFX10-UNSAFE: [[BITCAST2:%[0-9]+]]:_(s96) = G_BITCAST [[CONCAT_VECTORS1]](<6 x s16>)
+    ; GFX10-UNSAFE: [[TRUNC1:%[0-9]+]]:_(s48) = G_TRUNC [[BITCAST2]](s96)
+    ; GFX10-UNSAFE: [[BITCAST3:%[0-9]+]]:_(<3 x s16>) = G_BITCAST [[TRUNC1]](s48)
+    ; GFX10-UNSAFE: [[COPY4:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr4
+    ; GFX10-UNSAFE: [[COPY5:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr5
+    ; GFX10-UNSAFE: [[CONCAT_VECTORS2:%[0-9]+]]:_(<6 x s16>) = G_CONCAT_VECTORS [[COPY4]](<2 x s16>), [[COPY5]](<2 x s16>), [[DEF]](<2 x s16>)
+    ; GFX10-UNSAFE: [[BITCAST4:%[0-9]+]]:_(s96) = G_BITCAST [[CONCAT_VECTORS2]](<6 x s16>)
+    ; GFX10-UNSAFE: [[TRUNC2:%[0-9]+]]:_(s48) = G_TRUNC [[BITCAST4]](s96)
+    ; GFX10-UNSAFE: [[BITCAST5:%[0-9]+]]:_(<3 x s16>) = G_BITCAST [[TRUNC2]](s48)
+    ; GFX10-UNSAFE: [[COPY6:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX10-UNSAFE: [[FMA:%[0-9]+]]:_(<3 x s16>) = G_FMA [[BITCAST1]], [[BITCAST3]], [[BITCAST5]]
+    ; GFX10-UNSAFE: [[DEF1:%[0-9]+]]:_(<3 x s16>) = G_IMPLICIT_DEF
+    ; GFX10-UNSAFE: [[CONCAT_VECTORS3:%[0-9]+]]:_(<6 x s16>) = G_CONCAT_VECTORS [[FMA]](<3 x s16>), [[DEF1]](<3 x s16>)
+    ; GFX10-UNSAFE: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>), [[UV2:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[CONCAT_VECTORS3]](<6 x s16>)
+    ; GFX10-UNSAFE: $vgpr0 = COPY [[UV]](<2 x s16>)
+    ; GFX10-UNSAFE: $vgpr1 = COPY [[UV1]](<2 x s16>)
+    ; GFX10-UNSAFE: [[COPY7:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY6]]
+    ; GFX10-UNSAFE: S_SETPC_B64_return [[COPY7]], implicit $vgpr0, implicit $vgpr1
+    %4:_(<2 x s16>) = COPY $vgpr0
+    %5:_(<2 x s16>) = COPY $vgpr1
+    %10:_(<2 x s16>) = G_IMPLICIT_DEF
+    %11:_(<6 x s16>) = G_CONCAT_VECTORS %4(<2 x s16>), %5(<2 x s16>), %10(<2 x s16>)
+    %0:_(<3 x s16>), %12:_(<3 x s16>) = G_UNMERGE_VALUES %11(<6 x s16>)
+    %6:_(<2 x s16>) = COPY $vgpr2
+    %7:_(<2 x s16>) = COPY $vgpr3
+    %13:_(<6 x s16>) = G_CONCAT_VECTORS %6(<2 x s16>), %7(<2 x s16>), %10(<2 x s16>)
+    %1:_(<3 x s16>), %14:_(<3 x s16>) = G_UNMERGE_VALUES %13(<6 x s16>)
+    %8:_(<2 x s16>) = COPY $vgpr4
+    %9:_(<2 x s16>) = COPY $vgpr5
+    %15:_(<6 x s16>) = G_CONCAT_VECTORS %8(<2 x s16>), %9(<2 x s16>), %10(<2 x s16>)
+    %2:_(<3 x s16>), %16:_(<3 x s16>) = G_UNMERGE_VALUES %15(<6 x s16>)
+    %3:sgpr_64 = COPY $sgpr30_sgpr31
+    %17:_(<3 x s16>) = reassoc G_FMUL %0, %1
+    %18:_(<3 x s16>) = reassoc G_FADD %2, %17
+    %22:_(<3 x s16>) = G_IMPLICIT_DEF
+    %23:_(<6 x s16>) = G_CONCAT_VECTORS %18(<3 x s16>), %22(<3 x s16>)
+    %20:_(<2 x s16>), %21:_(<2 x s16>), %24:_(<2 x s16>) = G_UNMERGE_VALUES %23(<6 x s16>)
+    $vgpr0 = COPY %20(<2 x s16>)
+    $vgpr1 = COPY %21(<2 x s16>)
+    %19:ccr_sgpr_64 = COPY %3
+    S_SETPC_B64_return %19, implicit $vgpr0, implicit $vgpr1
+...
+
+---
+name:            test_4xdouble_add_mul
+body:             |
+  bb.1.entry:
+    liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8, $vgpr9, $vgpr10, $vgpr11, $vgpr12, $vgpr13, $vgpr14, $vgpr15, $vgpr16, $vgpr17, $vgpr18, $vgpr19, $vgpr20, $vgpr21, $vgpr22, $vgpr23, $sgpr30_sgpr31
+
+    ; GFX9-LABEL: name: test_4xdouble_add_mul
+    ; GFX9: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX9: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX9: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX9: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3
+    ; GFX9: [[COPY4:%[0-9]+]]:_(s32) = COPY $vgpr4
+    ; GFX9: [[COPY5:%[0-9]+]]:_(s32) = COPY $vgpr5
+    ; GFX9: [[COPY6:%[0-9]+]]:_(s32) = COPY $vgpr6
+    ; GFX9: [[COPY7:%[0-9]+]]:_(s32) = COPY $vgpr7
+    ; GFX9: [[MV:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY]](s32), [[COPY1]](s32)
+    ; GFX9: [[MV1:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY2]](s32), [[COPY3]](s32)
+    ; GFX9: [[MV2:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY4]](s32), [[COPY5]](s32)
+    ; GFX9: [[MV3:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY6]](s32), [[COPY7]](s32)
+    ; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s64>) = G_BUILD_VECTOR [[MV]](s64), [[MV1]](s64), [[MV2]](s64), [[MV3]](s64)
+    ; GFX9: [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr8
+    ; GFX9: [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr9
+    ; GFX9: [[COPY10:%[0-9]+]]:_(s32) = COPY $vgpr10
+    ; GFX9: [[COPY11:%[0-9]+]]:_(s32) = COPY $vgpr11
+    ; GFX9: [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr12
+    ; GFX9: [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr13
+    ; GFX9: [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr14
+    ; GFX9: [[COPY15:%[0-9]+]]:_(s32) = COPY $vgpr15
+    ; GFX9: [[MV4:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY8]](s32), [[COPY9]](s32)
+    ; GFX9: [[MV5:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY10]](s32), [[COPY11]](s32)
+    ; GFX9: [[MV6:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY12]](s32), [[COPY13]](s32)
+    ; GFX9: [[MV7:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY14]](s32), [[COPY15]](s32)
+    ; GFX9: [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s64>) = G_BUILD_VECTOR [[MV4]](s64), [[MV5]](s64), [[MV6]](s64), [[MV7]](s64)
+    ; GFX9: [[COPY16:%[0-9]+]]:_(s32) = COPY $vgpr16
+    ; GFX9: [[COPY17:%[0-9]+]]:_(s32) = COPY $vgpr17
+    ; GFX9: [[COPY18:%[0-9]+]]:_(s32) = COPY $vgpr18
+    ; GFX9: [[COPY19:%[0-9]+]]:_(s32) = COPY $vgpr19
+    ; GFX9: [[COPY20:%[0-9]+]]:_(s32) = COPY $vgpr20
+    ; GFX9: [[COPY21:%[0-9]+]]:_(s32) = COPY $vgpr21
+    ; GFX9: [[COPY22:%[0-9]+]]:_(s32) = COPY $vgpr22
+    ; GFX9: [[COPY23:%[0-9]+]]:_(s32) = COPY $vgpr23
+    ; GFX9: [[MV8:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY16]](s32), [[COPY17]](s32)
+    ; GFX9: [[MV9:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY18]](s32), [[COPY19]](s32)
+    ; GFX9: [[MV10:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY20]](s32), [[COPY21]](s32)
+    ; GFX9: [[MV11:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY22]](s32), [[COPY23]](s32)
+    ; GFX9: [[BUILD_VECTOR2:%[0-9]+]]:_(<4 x s64>) = G_BUILD_VECTOR [[MV8]](s64), [[MV9]](s64), [[MV10]](s64), [[MV11]](s64)
+    ; GFX9: [[COPY24:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX9: [[FMUL:%[0-9]+]]:_(<4 x s64>) = reassoc G_FMUL [[BUILD_VECTOR]], [[BUILD_VECTOR1]]
+    ; GFX9: [[FADD:%[0-9]+]]:_(<4 x s64>) = reassoc G_FADD [[FMUL]], [[BUILD_VECTOR2]]
+    ; GFX9: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32), [[UV4:%[0-9]+]]:_(s32), [[UV5:%[0-9]+]]:_(s32), [[UV6:%[0-9]+]]:_(s32), [[UV7:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FADD]](<4 x s64>)
+    ; GFX9: $vgpr0 = COPY [[UV]](s32)
+    ; GFX9: $vgpr1 = COPY [[UV1]](s32)
+    ; GFX9: $vgpr2 = COPY [[UV2]](s32)
+    ; GFX9: $vgpr3 = COPY [[UV3]](s32)
+    ; GFX9: $vgpr4 = COPY [[UV4]](s32)
+    ; GFX9: $vgpr5 = COPY [[UV5]](s32)
+    ; GFX9: $vgpr6 = COPY [[UV6]](s32)
+    ; GFX9: $vgpr7 = COPY [[UV7]](s32)
+    ; GFX9: [[COPY25:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY24]]
+    ; GFX9: S_SETPC_B64_return [[COPY25]], implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3, implicit $vgpr4, implicit $vgpr5, implicit $vgpr6, implicit $vgpr7
+    ; GFX9-CONTRACT-LABEL: name: test_4xdouble_add_mul
+    ; GFX9-CONTRACT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX9-CONTRACT: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX9-CONTRACT: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX9-CONTRACT: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3
+    ; GFX9-CONTRACT: [[COPY4:%[0-9]+]]:_(s32) = COPY $vgpr4
+    ; GFX9-CONTRACT: [[COPY5:%[0-9]+]]:_(s32) = COPY $vgpr5
+    ; GFX9-CONTRACT: [[COPY6:%[0-9]+]]:_(s32) = COPY $vgpr6
+    ; GFX9-CONTRACT: [[COPY7:%[0-9]+]]:_(s32) = COPY $vgpr7
+    ; GFX9-CONTRACT: [[MV:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY]](s32), [[COPY1]](s32)
+    ; GFX9-CONTRACT: [[MV1:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY2]](s32), [[COPY3]](s32)
+    ; GFX9-CONTRACT: [[MV2:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY4]](s32), [[COPY5]](s32)
+    ; GFX9-CONTRACT: [[MV3:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY6]](s32), [[COPY7]](s32)
+    ; GFX9-CONTRACT: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s64>) = G_BUILD_VECTOR [[MV]](s64), [[MV1]](s64), [[MV2]](s64), [[MV3]](s64)
+    ; GFX9-CONTRACT: [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr8
+    ; GFX9-CONTRACT: [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr9
+    ; GFX9-CONTRACT: [[COPY10:%[0-9]+]]:_(s32) = COPY $vgpr10
+    ; GFX9-CONTRACT: [[COPY11:%[0-9]+]]:_(s32) = COPY $vgpr11
+    ; GFX9-CONTRACT: [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr12
+    ; GFX9-CONTRACT: [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr13
+    ; GFX9-CONTRACT: [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr14
+    ; GFX9-CONTRACT: [[COPY15:%[0-9]+]]:_(s32) = COPY $vgpr15
+    ; GFX9-CONTRACT: [[MV4:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY8]](s32), [[COPY9]](s32)
+    ; GFX9-CONTRACT: [[MV5:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY10]](s32), [[COPY11]](s32)
+    ; GFX9-CONTRACT: [[MV6:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY12]](s32), [[COPY13]](s32)
+    ; GFX9-CONTRACT: [[MV7:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY14]](s32), [[COPY15]](s32)
+    ; GFX9-CONTRACT: [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s64>) = G_BUILD_VECTOR [[MV4]](s64), [[MV5]](s64), [[MV6]](s64), [[MV7]](s64)
+    ; GFX9-CONTRACT: [[COPY16:%[0-9]+]]:_(s32) = COPY $vgpr16
+    ; GFX9-CONTRACT: [[COPY17:%[0-9]+]]:_(s32) = COPY $vgpr17
+    ; GFX9-CONTRACT: [[COPY18:%[0-9]+]]:_(s32) = COPY $vgpr18
+    ; GFX9-CONTRACT: [[COPY19:%[0-9]+]]:_(s32) = COPY $vgpr19
+    ; GFX9-CONTRACT: [[COPY20:%[0-9]+]]:_(s32) = COPY $vgpr20
+    ; GFX9-CONTRACT: [[COPY21:%[0-9]+]]:_(s32) = COPY $vgpr21
+    ; GFX9-CONTRACT: [[COPY22:%[0-9]+]]:_(s32) = COPY $vgpr22
+    ; GFX9-CONTRACT: [[COPY23:%[0-9]+]]:_(s32) = COPY $vgpr23
+    ; GFX9-CONTRACT: [[MV8:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY16]](s32), [[COPY17]](s32)
+    ; GFX9-CONTRACT: [[MV9:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY18]](s32), [[COPY19]](s32)
+    ; GFX9-CONTRACT: [[MV10:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY20]](s32), [[COPY21]](s32)
+    ; GFX9-CONTRACT: [[MV11:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY22]](s32), [[COPY23]](s32)
+    ; GFX9-CONTRACT: [[BUILD_VECTOR2:%[0-9]+]]:_(<4 x s64>) = G_BUILD_VECTOR [[MV8]](s64), [[MV9]](s64), [[MV10]](s64), [[MV11]](s64)
+    ; GFX9-CONTRACT: [[COPY24:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX9-CONTRACT: [[FMA:%[0-9]+]]:_(<4 x s64>) = G_FMA [[BUILD_VECTOR]], [[BUILD_VECTOR1]], [[BUILD_VECTOR2]]
+    ; GFX9-CONTRACT: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32), [[UV4:%[0-9]+]]:_(s32), [[UV5:%[0-9]+]]:_(s32), [[UV6:%[0-9]+]]:_(s32), [[UV7:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FMA]](<4 x s64>)
+    ; GFX9-CONTRACT: $vgpr0 = COPY [[UV]](s32)
+    ; GFX9-CONTRACT: $vgpr1 = COPY [[UV1]](s32)
+    ; GFX9-CONTRACT: $vgpr2 = COPY [[UV2]](s32)
+    ; GFX9-CONTRACT: $vgpr3 = COPY [[UV3]](s32)
+    ; GFX9-CONTRACT: $vgpr4 = COPY [[UV4]](s32)
+    ; GFX9-CONTRACT: $vgpr5 = COPY [[UV5]](s32)
+    ; GFX9-CONTRACT: $vgpr6 = COPY [[UV6]](s32)
+    ; GFX9-CONTRACT: $vgpr7 = COPY [[UV7]](s32)
+    ; GFX9-CONTRACT: [[COPY25:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY24]]
+    ; GFX9-CONTRACT: S_SETPC_B64_return [[COPY25]], implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3, implicit $vgpr4, implicit $vgpr5, implicit $vgpr6, implicit $vgpr7
+    ; GFX9-DENORM-LABEL: name: test_4xdouble_add_mul
+    ; GFX9-DENORM: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX9-DENORM: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX9-DENORM: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX9-DENORM: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3
+    ; GFX9-DENORM: [[COPY4:%[0-9]+]]:_(s32) = COPY $vgpr4
+    ; GFX9-DENORM: [[COPY5:%[0-9]+]]:_(s32) = COPY $vgpr5
+    ; GFX9-DENORM: [[COPY6:%[0-9]+]]:_(s32) = COPY $vgpr6
+    ; GFX9-DENORM: [[COPY7:%[0-9]+]]:_(s32) = COPY $vgpr7
+    ; GFX9-DENORM: [[MV:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY]](s32), [[COPY1]](s32)
+    ; GFX9-DENORM: [[MV1:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY2]](s32), [[COPY3]](s32)
+    ; GFX9-DENORM: [[MV2:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY4]](s32), [[COPY5]](s32)
+    ; GFX9-DENORM: [[MV3:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY6]](s32), [[COPY7]](s32)
+    ; GFX9-DENORM: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s64>) = G_BUILD_VECTOR [[MV]](s64), [[MV1]](s64), [[MV2]](s64), [[MV3]](s64)
+    ; GFX9-DENORM: [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr8
+    ; GFX9-DENORM: [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr9
+    ; GFX9-DENORM: [[COPY10:%[0-9]+]]:_(s32) = COPY $vgpr10
+    ; GFX9-DENORM: [[COPY11:%[0-9]+]]:_(s32) = COPY $vgpr11
+    ; GFX9-DENORM: [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr12
+    ; GFX9-DENORM: [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr13
+    ; GFX9-DENORM: [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr14
+    ; GFX9-DENORM: [[COPY15:%[0-9]+]]:_(s32) = COPY $vgpr15
+    ; GFX9-DENORM: [[MV4:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY8]](s32), [[COPY9]](s32)
+    ; GFX9-DENORM: [[MV5:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY10]](s32), [[COPY11]](s32)
+    ; GFX9-DENORM: [[MV6:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY12]](s32), [[COPY13]](s32)
+    ; GFX9-DENORM: [[MV7:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY14]](s32), [[COPY15]](s32)
+    ; GFX9-DENORM: [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s64>) = G_BUILD_VECTOR [[MV4]](s64), [[MV5]](s64), [[MV6]](s64), [[MV7]](s64)
+    ; GFX9-DENORM: [[COPY16:%[0-9]+]]:_(s32) = COPY $vgpr16
+    ; GFX9-DENORM: [[COPY17:%[0-9]+]]:_(s32) = COPY $vgpr17
+    ; GFX9-DENORM: [[COPY18:%[0-9]+]]:_(s32) = COPY $vgpr18
+    ; GFX9-DENORM: [[COPY19:%[0-9]+]]:_(s32) = COPY $vgpr19
+    ; GFX9-DENORM: [[COPY20:%[0-9]+]]:_(s32) = COPY $vgpr20
+    ; GFX9-DENORM: [[COPY21:%[0-9]+]]:_(s32) = COPY $vgpr21
+    ; GFX9-DENORM: [[COPY22:%[0-9]+]]:_(s32) = COPY $vgpr22
+    ; GFX9-DENORM: [[COPY23:%[0-9]+]]:_(s32) = COPY $vgpr23
+    ; GFX9-DENORM: [[MV8:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY16]](s32), [[COPY17]](s32)
+    ; GFX9-DENORM: [[MV9:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY18]](s32), [[COPY19]](s32)
+    ; GFX9-DENORM: [[MV10:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY20]](s32), [[COPY21]](s32)
+    ; GFX9-DENORM: [[MV11:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY22]](s32), [[COPY23]](s32)
+    ; GFX9-DENORM: [[BUILD_VECTOR2:%[0-9]+]]:_(<4 x s64>) = G_BUILD_VECTOR [[MV8]](s64), [[MV9]](s64), [[MV10]](s64), [[MV11]](s64)
+    ; GFX9-DENORM: [[COPY24:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX9-DENORM: [[FMUL:%[0-9]+]]:_(<4 x s64>) = reassoc G_FMUL [[BUILD_VECTOR]], [[BUILD_VECTOR1]]
+    ; GFX9-DENORM: [[FADD:%[0-9]+]]:_(<4 x s64>) = reassoc G_FADD [[FMUL]], [[BUILD_VECTOR2]]
+    ; GFX9-DENORM: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32), [[UV4:%[0-9]+]]:_(s32), [[UV5:%[0-9]+]]:_(s32), [[UV6:%[0-9]+]]:_(s32), [[UV7:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FADD]](<4 x s64>)
+    ; GFX9-DENORM: $vgpr0 = COPY [[UV]](s32)
+    ; GFX9-DENORM: $vgpr1 = COPY [[UV1]](s32)
+    ; GFX9-DENORM: $vgpr2 = COPY [[UV2]](s32)
+    ; GFX9-DENORM: $vgpr3 = COPY [[UV3]](s32)
+    ; GFX9-DENORM: $vgpr4 = COPY [[UV4]](s32)
+    ; GFX9-DENORM: $vgpr5 = COPY [[UV5]](s32)
+    ; GFX9-DENORM: $vgpr6 = COPY [[UV6]](s32)
+    ; GFX9-DENORM: $vgpr7 = COPY [[UV7]](s32)
+    ; GFX9-DENORM: [[COPY25:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY24]]
+    ; GFX9-DENORM: S_SETPC_B64_return [[COPY25]], implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3, implicit $vgpr4, implicit $vgpr5, implicit $vgpr6, implicit $vgpr7
+    ; GFX9-UNSAFE-LABEL: name: test_4xdouble_add_mul
+    ; GFX9-UNSAFE: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX9-UNSAFE: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX9-UNSAFE: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX9-UNSAFE: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3
+    ; GFX9-UNSAFE: [[COPY4:%[0-9]+]]:_(s32) = COPY $vgpr4
+    ; GFX9-UNSAFE: [[COPY5:%[0-9]+]]:_(s32) = COPY $vgpr5
+    ; GFX9-UNSAFE: [[COPY6:%[0-9]+]]:_(s32) = COPY $vgpr6
+    ; GFX9-UNSAFE: [[COPY7:%[0-9]+]]:_(s32) = COPY $vgpr7
+    ; GFX9-UNSAFE: [[MV:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY]](s32), [[COPY1]](s32)
+    ; GFX9-UNSAFE: [[MV1:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY2]](s32), [[COPY3]](s32)
+    ; GFX9-UNSAFE: [[MV2:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY4]](s32), [[COPY5]](s32)
+    ; GFX9-UNSAFE: [[MV3:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY6]](s32), [[COPY7]](s32)
+    ; GFX9-UNSAFE: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s64>) = G_BUILD_VECTOR [[MV]](s64), [[MV1]](s64), [[MV2]](s64), [[MV3]](s64)
+    ; GFX9-UNSAFE: [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr8
+    ; GFX9-UNSAFE: [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr9
+    ; GFX9-UNSAFE: [[COPY10:%[0-9]+]]:_(s32) = COPY $vgpr10
+    ; GFX9-UNSAFE: [[COPY11:%[0-9]+]]:_(s32) = COPY $vgpr11
+    ; GFX9-UNSAFE: [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr12
+    ; GFX9-UNSAFE: [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr13
+    ; GFX9-UNSAFE: [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr14
+    ; GFX9-UNSAFE: [[COPY15:%[0-9]+]]:_(s32) = COPY $vgpr15
+    ; GFX9-UNSAFE: [[MV4:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY8]](s32), [[COPY9]](s32)
+    ; GFX9-UNSAFE: [[MV5:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY10]](s32), [[COPY11]](s32)
+    ; GFX9-UNSAFE: [[MV6:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY12]](s32), [[COPY13]](s32)
+    ; GFX9-UNSAFE: [[MV7:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY14]](s32), [[COPY15]](s32)
+    ; GFX9-UNSAFE: [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s64>) = G_BUILD_VECTOR [[MV4]](s64), [[MV5]](s64), [[MV6]](s64), [[MV7]](s64)
+    ; GFX9-UNSAFE: [[COPY16:%[0-9]+]]:_(s32) = COPY $vgpr16
+    ; GFX9-UNSAFE: [[COPY17:%[0-9]+]]:_(s32) = COPY $vgpr17
+    ; GFX9-UNSAFE: [[COPY18:%[0-9]+]]:_(s32) = COPY $vgpr18
+    ; GFX9-UNSAFE: [[COPY19:%[0-9]+]]:_(s32) = COPY $vgpr19
+    ; GFX9-UNSAFE: [[COPY20:%[0-9]+]]:_(s32) = COPY $vgpr20
+    ; GFX9-UNSAFE: [[COPY21:%[0-9]+]]:_(s32) = COPY $vgpr21
+    ; GFX9-UNSAFE: [[COPY22:%[0-9]+]]:_(s32) = COPY $vgpr22
+    ; GFX9-UNSAFE: [[COPY23:%[0-9]+]]:_(s32) = COPY $vgpr23
+    ; GFX9-UNSAFE: [[MV8:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY16]](s32), [[COPY17]](s32)
+    ; GFX9-UNSAFE: [[MV9:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY18]](s32), [[COPY19]](s32)
+    ; GFX9-UNSAFE: [[MV10:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY20]](s32), [[COPY21]](s32)
+    ; GFX9-UNSAFE: [[MV11:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY22]](s32), [[COPY23]](s32)
+    ; GFX9-UNSAFE: [[BUILD_VECTOR2:%[0-9]+]]:_(<4 x s64>) = G_BUILD_VECTOR [[MV8]](s64), [[MV9]](s64), [[MV10]](s64), [[MV11]](s64)
+    ; GFX9-UNSAFE: [[COPY24:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX9-UNSAFE: [[FMA:%[0-9]+]]:_(<4 x s64>) = G_FMA [[BUILD_VECTOR]], [[BUILD_VECTOR1]], [[BUILD_VECTOR2]]
+    ; GFX9-UNSAFE: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32), [[UV4:%[0-9]+]]:_(s32), [[UV5:%[0-9]+]]:_(s32), [[UV6:%[0-9]+]]:_(s32), [[UV7:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FMA]](<4 x s64>)
+    ; GFX9-UNSAFE: $vgpr0 = COPY [[UV]](s32)
+    ; GFX9-UNSAFE: $vgpr1 = COPY [[UV1]](s32)
+    ; GFX9-UNSAFE: $vgpr2 = COPY [[UV2]](s32)
+    ; GFX9-UNSAFE: $vgpr3 = COPY [[UV3]](s32)
+    ; GFX9-UNSAFE: $vgpr4 = COPY [[UV4]](s32)
+    ; GFX9-UNSAFE: $vgpr5 = COPY [[UV5]](s32)
+    ; GFX9-UNSAFE: $vgpr6 = COPY [[UV6]](s32)
+    ; GFX9-UNSAFE: $vgpr7 = COPY [[UV7]](s32)
+    ; GFX9-UNSAFE: [[COPY25:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY24]]
+    ; GFX9-UNSAFE: S_SETPC_B64_return [[COPY25]], implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3, implicit $vgpr4, implicit $vgpr5, implicit $vgpr6, implicit $vgpr7
+    ; GFX10-LABEL: name: test_4xdouble_add_mul
+    ; GFX10: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX10: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX10: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX10: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3
+    ; GFX10: [[COPY4:%[0-9]+]]:_(s32) = COPY $vgpr4
+    ; GFX10: [[COPY5:%[0-9]+]]:_(s32) = COPY $vgpr5
+    ; GFX10: [[COPY6:%[0-9]+]]:_(s32) = COPY $vgpr6
+    ; GFX10: [[COPY7:%[0-9]+]]:_(s32) = COPY $vgpr7
+    ; GFX10: [[MV:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY]](s32), [[COPY1]](s32)
+    ; GFX10: [[MV1:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY2]](s32), [[COPY3]](s32)
+    ; GFX10: [[MV2:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY4]](s32), [[COPY5]](s32)
+    ; GFX10: [[MV3:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY6]](s32), [[COPY7]](s32)
+    ; GFX10: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s64>) = G_BUILD_VECTOR [[MV]](s64), [[MV1]](s64), [[MV2]](s64), [[MV3]](s64)
+    ; GFX10: [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr8
+    ; GFX10: [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr9
+    ; GFX10: [[COPY10:%[0-9]+]]:_(s32) = COPY $vgpr10
+    ; GFX10: [[COPY11:%[0-9]+]]:_(s32) = COPY $vgpr11
+    ; GFX10: [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr12
+    ; GFX10: [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr13
+    ; GFX10: [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr14
+    ; GFX10: [[COPY15:%[0-9]+]]:_(s32) = COPY $vgpr15
+    ; GFX10: [[MV4:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY8]](s32), [[COPY9]](s32)
+    ; GFX10: [[MV5:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY10]](s32), [[COPY11]](s32)
+    ; GFX10: [[MV6:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY12]](s32), [[COPY13]](s32)
+    ; GFX10: [[MV7:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY14]](s32), [[COPY15]](s32)
+    ; GFX10: [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s64>) = G_BUILD_VECTOR [[MV4]](s64), [[MV5]](s64), [[MV6]](s64), [[MV7]](s64)
+    ; GFX10: [[COPY16:%[0-9]+]]:_(s32) = COPY $vgpr16
+    ; GFX10: [[COPY17:%[0-9]+]]:_(s32) = COPY $vgpr17
+    ; GFX10: [[COPY18:%[0-9]+]]:_(s32) = COPY $vgpr18
+    ; GFX10: [[COPY19:%[0-9]+]]:_(s32) = COPY $vgpr19
+    ; GFX10: [[COPY20:%[0-9]+]]:_(s32) = COPY $vgpr20
+    ; GFX10: [[COPY21:%[0-9]+]]:_(s32) = COPY $vgpr21
+    ; GFX10: [[COPY22:%[0-9]+]]:_(s32) = COPY $vgpr22
+    ; GFX10: [[COPY23:%[0-9]+]]:_(s32) = COPY $vgpr23
+    ; GFX10: [[MV8:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY16]](s32), [[COPY17]](s32)
+    ; GFX10: [[MV9:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY18]](s32), [[COPY19]](s32)
+    ; GFX10: [[MV10:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY20]](s32), [[COPY21]](s32)
+    ; GFX10: [[MV11:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY22]](s32), [[COPY23]](s32)
+    ; GFX10: [[BUILD_VECTOR2:%[0-9]+]]:_(<4 x s64>) = G_BUILD_VECTOR [[MV8]](s64), [[MV9]](s64), [[MV10]](s64), [[MV11]](s64)
+    ; GFX10: [[COPY24:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX10: [[FMUL:%[0-9]+]]:_(<4 x s64>) = reassoc G_FMUL [[BUILD_VECTOR]], [[BUILD_VECTOR1]]
+    ; GFX10: [[FADD:%[0-9]+]]:_(<4 x s64>) = reassoc G_FADD [[FMUL]], [[BUILD_VECTOR2]]
+    ; GFX10: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32), [[UV4:%[0-9]+]]:_(s32), [[UV5:%[0-9]+]]:_(s32), [[UV6:%[0-9]+]]:_(s32), [[UV7:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FADD]](<4 x s64>)
+    ; GFX10: $vgpr0 = COPY [[UV]](s32)
+    ; GFX10: $vgpr1 = COPY [[UV1]](s32)
+    ; GFX10: $vgpr2 = COPY [[UV2]](s32)
+    ; GFX10: $vgpr3 = COPY [[UV3]](s32)
+    ; GFX10: $vgpr4 = COPY [[UV4]](s32)
+    ; GFX10: $vgpr5 = COPY [[UV5]](s32)
+    ; GFX10: $vgpr6 = COPY [[UV6]](s32)
+    ; GFX10: $vgpr7 = COPY [[UV7]](s32)
+    ; GFX10: [[COPY25:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY24]]
+    ; GFX10: S_SETPC_B64_return [[COPY25]], implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3, implicit $vgpr4, implicit $vgpr5, implicit $vgpr6, implicit $vgpr7
+    ; GFX10-CONTRACT-LABEL: name: test_4xdouble_add_mul
+    ; GFX10-CONTRACT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX10-CONTRACT: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX10-CONTRACT: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX10-CONTRACT: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3
+    ; GFX10-CONTRACT: [[COPY4:%[0-9]+]]:_(s32) = COPY $vgpr4
+    ; GFX10-CONTRACT: [[COPY5:%[0-9]+]]:_(s32) = COPY $vgpr5
+    ; GFX10-CONTRACT: [[COPY6:%[0-9]+]]:_(s32) = COPY $vgpr6
+    ; GFX10-CONTRACT: [[COPY7:%[0-9]+]]:_(s32) = COPY $vgpr7
+    ; GFX10-CONTRACT: [[MV:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY]](s32), [[COPY1]](s32)
+    ; GFX10-CONTRACT: [[MV1:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY2]](s32), [[COPY3]](s32)
+    ; GFX10-CONTRACT: [[MV2:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY4]](s32), [[COPY5]](s32)
+    ; GFX10-CONTRACT: [[MV3:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY6]](s32), [[COPY7]](s32)
+    ; GFX10-CONTRACT: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s64>) = G_BUILD_VECTOR [[MV]](s64), [[MV1]](s64), [[MV2]](s64), [[MV3]](s64)
+    ; GFX10-CONTRACT: [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr8
+    ; GFX10-CONTRACT: [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr9
+    ; GFX10-CONTRACT: [[COPY10:%[0-9]+]]:_(s32) = COPY $vgpr10
+    ; GFX10-CONTRACT: [[COPY11:%[0-9]+]]:_(s32) = COPY $vgpr11
+    ; GFX10-CONTRACT: [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr12
+    ; GFX10-CONTRACT: [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr13
+    ; GFX10-CONTRACT: [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr14
+    ; GFX10-CONTRACT: [[COPY15:%[0-9]+]]:_(s32) = COPY $vgpr15
+    ; GFX10-CONTRACT: [[MV4:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY8]](s32), [[COPY9]](s32)
+    ; GFX10-CONTRACT: [[MV5:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY10]](s32), [[COPY11]](s32)
+    ; GFX10-CONTRACT: [[MV6:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY12]](s32), [[COPY13]](s32)
+    ; GFX10-CONTRACT: [[MV7:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY14]](s32), [[COPY15]](s32)
+    ; GFX10-CONTRACT: [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s64>) = G_BUILD_VECTOR [[MV4]](s64), [[MV5]](s64), [[MV6]](s64), [[MV7]](s64)
+    ; GFX10-CONTRACT: [[COPY16:%[0-9]+]]:_(s32) = COPY $vgpr16
+    ; GFX10-CONTRACT: [[COPY17:%[0-9]+]]:_(s32) = COPY $vgpr17
+    ; GFX10-CONTRACT: [[COPY18:%[0-9]+]]:_(s32) = COPY $vgpr18
+    ; GFX10-CONTRACT: [[COPY19:%[0-9]+]]:_(s32) = COPY $vgpr19
+    ; GFX10-CONTRACT: [[COPY20:%[0-9]+]]:_(s32) = COPY $vgpr20
+    ; GFX10-CONTRACT: [[COPY21:%[0-9]+]]:_(s32) = COPY $vgpr21
+    ; GFX10-CONTRACT: [[COPY22:%[0-9]+]]:_(s32) = COPY $vgpr22
+    ; GFX10-CONTRACT: [[COPY23:%[0-9]+]]:_(s32) = COPY $vgpr23
+    ; GFX10-CONTRACT: [[MV8:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY16]](s32), [[COPY17]](s32)
+    ; GFX10-CONTRACT: [[MV9:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY18]](s32), [[COPY19]](s32)
+    ; GFX10-CONTRACT: [[MV10:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY20]](s32), [[COPY21]](s32)
+    ; GFX10-CONTRACT: [[MV11:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY22]](s32), [[COPY23]](s32)
+    ; GFX10-CONTRACT: [[BUILD_VECTOR2:%[0-9]+]]:_(<4 x s64>) = G_BUILD_VECTOR [[MV8]](s64), [[MV9]](s64), [[MV10]](s64), [[MV11]](s64)
+    ; GFX10-CONTRACT: [[COPY24:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX10-CONTRACT: [[FMA:%[0-9]+]]:_(<4 x s64>) = G_FMA [[BUILD_VECTOR]], [[BUILD_VECTOR1]], [[BUILD_VECTOR2]]
+    ; GFX10-CONTRACT: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32), [[UV4:%[0-9]+]]:_(s32), [[UV5:%[0-9]+]]:_(s32), [[UV6:%[0-9]+]]:_(s32), [[UV7:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FMA]](<4 x s64>)
+    ; GFX10-CONTRACT: $vgpr0 = COPY [[UV]](s32)
+    ; GFX10-CONTRACT: $vgpr1 = COPY [[UV1]](s32)
+    ; GFX10-CONTRACT: $vgpr2 = COPY [[UV2]](s32)
+    ; GFX10-CONTRACT: $vgpr3 = COPY [[UV3]](s32)
+    ; GFX10-CONTRACT: $vgpr4 = COPY [[UV4]](s32)
+    ; GFX10-CONTRACT: $vgpr5 = COPY [[UV5]](s32)
+    ; GFX10-CONTRACT: $vgpr6 = COPY [[UV6]](s32)
+    ; GFX10-CONTRACT: $vgpr7 = COPY [[UV7]](s32)
+    ; GFX10-CONTRACT: [[COPY25:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY24]]
+    ; GFX10-CONTRACT: S_SETPC_B64_return [[COPY25]], implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3, implicit $vgpr4, implicit $vgpr5, implicit $vgpr6, implicit $vgpr7
+    ; GFX10-DENORM-LABEL: name: test_4xdouble_add_mul
+    ; GFX10-DENORM: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX10-DENORM: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX10-DENORM: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX10-DENORM: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3
+    ; GFX10-DENORM: [[COPY4:%[0-9]+]]:_(s32) = COPY $vgpr4
+    ; GFX10-DENORM: [[COPY5:%[0-9]+]]:_(s32) = COPY $vgpr5
+    ; GFX10-DENORM: [[COPY6:%[0-9]+]]:_(s32) = COPY $vgpr6
+    ; GFX10-DENORM: [[COPY7:%[0-9]+]]:_(s32) = COPY $vgpr7
+    ; GFX10-DENORM: [[MV:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY]](s32), [[COPY1]](s32)
+    ; GFX10-DENORM: [[MV1:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY2]](s32), [[COPY3]](s32)
+    ; GFX10-DENORM: [[MV2:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY4]](s32), [[COPY5]](s32)
+    ; GFX10-DENORM: [[MV3:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY6]](s32), [[COPY7]](s32)
+    ; GFX10-DENORM: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s64>) = G_BUILD_VECTOR [[MV]](s64), [[MV1]](s64), [[MV2]](s64), [[MV3]](s64)
+    ; GFX10-DENORM: [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr8
+    ; GFX10-DENORM: [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr9
+    ; GFX10-DENORM: [[COPY10:%[0-9]+]]:_(s32) = COPY $vgpr10
+    ; GFX10-DENORM: [[COPY11:%[0-9]+]]:_(s32) = COPY $vgpr11
+    ; GFX10-DENORM: [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr12
+    ; GFX10-DENORM: [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr13
+    ; GFX10-DENORM: [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr14
+    ; GFX10-DENORM: [[COPY15:%[0-9]+]]:_(s32) = COPY $vgpr15
+    ; GFX10-DENORM: [[MV4:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY8]](s32), [[COPY9]](s32)
+    ; GFX10-DENORM: [[MV5:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY10]](s32), [[COPY11]](s32)
+    ; GFX10-DENORM: [[MV6:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY12]](s32), [[COPY13]](s32)
+    ; GFX10-DENORM: [[MV7:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY14]](s32), [[COPY15]](s32)
+    ; GFX10-DENORM: [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s64>) = G_BUILD_VECTOR [[MV4]](s64), [[MV5]](s64), [[MV6]](s64), [[MV7]](s64)
+    ; GFX10-DENORM: [[COPY16:%[0-9]+]]:_(s32) = COPY $vgpr16
+    ; GFX10-DENORM: [[COPY17:%[0-9]+]]:_(s32) = COPY $vgpr17
+    ; GFX10-DENORM: [[COPY18:%[0-9]+]]:_(s32) = COPY $vgpr18
+    ; GFX10-DENORM: [[COPY19:%[0-9]+]]:_(s32) = COPY $vgpr19
+    ; GFX10-DENORM: [[COPY20:%[0-9]+]]:_(s32) = COPY $vgpr20
+    ; GFX10-DENORM: [[COPY21:%[0-9]+]]:_(s32) = COPY $vgpr21
+    ; GFX10-DENORM: [[COPY22:%[0-9]+]]:_(s32) = COPY $vgpr22
+    ; GFX10-DENORM: [[COPY23:%[0-9]+]]:_(s32) = COPY $vgpr23
+    ; GFX10-DENORM: [[MV8:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY16]](s32), [[COPY17]](s32)
+    ; GFX10-DENORM: [[MV9:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY18]](s32), [[COPY19]](s32)
+    ; GFX10-DENORM: [[MV10:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY20]](s32), [[COPY21]](s32)
+    ; GFX10-DENORM: [[MV11:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY22]](s32), [[COPY23]](s32)
+    ; GFX10-DENORM: [[BUILD_VECTOR2:%[0-9]+]]:_(<4 x s64>) = G_BUILD_VECTOR [[MV8]](s64), [[MV9]](s64), [[MV10]](s64), [[MV11]](s64)
+    ; GFX10-DENORM: [[COPY24:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX10-DENORM: [[FMUL:%[0-9]+]]:_(<4 x s64>) = reassoc G_FMUL [[BUILD_VECTOR]], [[BUILD_VECTOR1]]
+    ; GFX10-DENORM: [[FADD:%[0-9]+]]:_(<4 x s64>) = reassoc G_FADD [[FMUL]], [[BUILD_VECTOR2]]
+    ; GFX10-DENORM: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32), [[UV4:%[0-9]+]]:_(s32), [[UV5:%[0-9]+]]:_(s32), [[UV6:%[0-9]+]]:_(s32), [[UV7:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FADD]](<4 x s64>)
+    ; GFX10-DENORM: $vgpr0 = COPY [[UV]](s32)
+    ; GFX10-DENORM: $vgpr1 = COPY [[UV1]](s32)
+    ; GFX10-DENORM: $vgpr2 = COPY [[UV2]](s32)
+    ; GFX10-DENORM: $vgpr3 = COPY [[UV3]](s32)
+    ; GFX10-DENORM: $vgpr4 = COPY [[UV4]](s32)
+    ; GFX10-DENORM: $vgpr5 = COPY [[UV5]](s32)
+    ; GFX10-DENORM: $vgpr6 = COPY [[UV6]](s32)
+    ; GFX10-DENORM: $vgpr7 = COPY [[UV7]](s32)
+    ; GFX10-DENORM: [[COPY25:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY24]]
+    ; GFX10-DENORM: S_SETPC_B64_return [[COPY25]], implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3, implicit $vgpr4, implicit $vgpr5, implicit $vgpr6, implicit $vgpr7
+    ; GFX10-UNSAFE-LABEL: name: test_4xdouble_add_mul
+    ; GFX10-UNSAFE: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX10-UNSAFE: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX10-UNSAFE: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX10-UNSAFE: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3
+    ; GFX10-UNSAFE: [[COPY4:%[0-9]+]]:_(s32) = COPY $vgpr4
+    ; GFX10-UNSAFE: [[COPY5:%[0-9]+]]:_(s32) = COPY $vgpr5
+    ; GFX10-UNSAFE: [[COPY6:%[0-9]+]]:_(s32) = COPY $vgpr6
+    ; GFX10-UNSAFE: [[COPY7:%[0-9]+]]:_(s32) = COPY $vgpr7
+    ; GFX10-UNSAFE: [[MV:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY]](s32), [[COPY1]](s32)
+    ; GFX10-UNSAFE: [[MV1:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY2]](s32), [[COPY3]](s32)
+    ; GFX10-UNSAFE: [[MV2:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY4]](s32), [[COPY5]](s32)
+    ; GFX10-UNSAFE: [[MV3:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY6]](s32), [[COPY7]](s32)
+    ; GFX10-UNSAFE: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s64>) = G_BUILD_VECTOR [[MV]](s64), [[MV1]](s64), [[MV2]](s64), [[MV3]](s64)
+    ; GFX10-UNSAFE: [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr8
+    ; GFX10-UNSAFE: [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr9
+    ; GFX10-UNSAFE: [[COPY10:%[0-9]+]]:_(s32) = COPY $vgpr10
+    ; GFX10-UNSAFE: [[COPY11:%[0-9]+]]:_(s32) = COPY $vgpr11
+    ; GFX10-UNSAFE: [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr12
+    ; GFX10-UNSAFE: [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr13
+    ; GFX10-UNSAFE: [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr14
+    ; GFX10-UNSAFE: [[COPY15:%[0-9]+]]:_(s32) = COPY $vgpr15
+    ; GFX10-UNSAFE: [[MV4:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY8]](s32), [[COPY9]](s32)
+    ; GFX10-UNSAFE: [[MV5:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY10]](s32), [[COPY11]](s32)
+    ; GFX10-UNSAFE: [[MV6:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY12]](s32), [[COPY13]](s32)
+    ; GFX10-UNSAFE: [[MV7:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY14]](s32), [[COPY15]](s32)
+    ; GFX10-UNSAFE: [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s64>) = G_BUILD_VECTOR [[MV4]](s64), [[MV5]](s64), [[MV6]](s64), [[MV7]](s64)
+    ; GFX10-UNSAFE: [[COPY16:%[0-9]+]]:_(s32) = COPY $vgpr16
+    ; GFX10-UNSAFE: [[COPY17:%[0-9]+]]:_(s32) = COPY $vgpr17
+    ; GFX10-UNSAFE: [[COPY18:%[0-9]+]]:_(s32) = COPY $vgpr18
+    ; GFX10-UNSAFE: [[COPY19:%[0-9]+]]:_(s32) = COPY $vgpr19
+    ; GFX10-UNSAFE: [[COPY20:%[0-9]+]]:_(s32) = COPY $vgpr20
+    ; GFX10-UNSAFE: [[COPY21:%[0-9]+]]:_(s32) = COPY $vgpr21
+    ; GFX10-UNSAFE: [[COPY22:%[0-9]+]]:_(s32) = COPY $vgpr22
+    ; GFX10-UNSAFE: [[COPY23:%[0-9]+]]:_(s32) = COPY $vgpr23
+    ; GFX10-UNSAFE: [[MV8:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY16]](s32), [[COPY17]](s32)
+    ; GFX10-UNSAFE: [[MV9:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY18]](s32), [[COPY19]](s32)
+    ; GFX10-UNSAFE: [[MV10:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY20]](s32), [[COPY21]](s32)
+    ; GFX10-UNSAFE: [[MV11:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY22]](s32), [[COPY23]](s32)
+    ; GFX10-UNSAFE: [[BUILD_VECTOR2:%[0-9]+]]:_(<4 x s64>) = G_BUILD_VECTOR [[MV8]](s64), [[MV9]](s64), [[MV10]](s64), [[MV11]](s64)
+    ; GFX10-UNSAFE: [[COPY24:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX10-UNSAFE: [[FMA:%[0-9]+]]:_(<4 x s64>) = G_FMA [[BUILD_VECTOR]], [[BUILD_VECTOR1]], [[BUILD_VECTOR2]]
+    ; GFX10-UNSAFE: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32), [[UV4:%[0-9]+]]:_(s32), [[UV5:%[0-9]+]]:_(s32), [[UV6:%[0-9]+]]:_(s32), [[UV7:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FMA]](<4 x s64>)
+    ; GFX10-UNSAFE: $vgpr0 = COPY [[UV]](s32)
+    ; GFX10-UNSAFE: $vgpr1 = COPY [[UV1]](s32)
+    ; GFX10-UNSAFE: $vgpr2 = COPY [[UV2]](s32)
+    ; GFX10-UNSAFE: $vgpr3 = COPY [[UV3]](s32)
+    ; GFX10-UNSAFE: $vgpr4 = COPY [[UV4]](s32)
+    ; GFX10-UNSAFE: $vgpr5 = COPY [[UV5]](s32)
+    ; GFX10-UNSAFE: $vgpr6 = COPY [[UV6]](s32)
+    ; GFX10-UNSAFE: $vgpr7 = COPY [[UV7]](s32)
+    ; GFX10-UNSAFE: [[COPY25:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY24]]
+    ; GFX10-UNSAFE: S_SETPC_B64_return [[COPY25]], implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3, implicit $vgpr4, implicit $vgpr5, implicit $vgpr6, implicit $vgpr7
+    %4:_(s32) = COPY $vgpr0
+    %5:_(s32) = COPY $vgpr1
+    %6:_(s32) = COPY $vgpr2
+    %7:_(s32) = COPY $vgpr3
+    %8:_(s32) = COPY $vgpr4
+    %9:_(s32) = COPY $vgpr5
+    %10:_(s32) = COPY $vgpr6
+    %11:_(s32) = COPY $vgpr7
+    %28:_(s64) = G_MERGE_VALUES %4(s32), %5(s32)
+    %29:_(s64) = G_MERGE_VALUES %6(s32), %7(s32)
+    %30:_(s64) = G_MERGE_VALUES %8(s32), %9(s32)
+    %31:_(s64) = G_MERGE_VALUES %10(s32), %11(s32)
+    %0:_(<4 x s64>) = G_BUILD_VECTOR %28(s64), %29(s64), %30(s64), %31(s64)
+    %12:_(s32) = COPY $vgpr8
+    %13:_(s32) = COPY $vgpr9
+    %14:_(s32) = COPY $vgpr10
+    %15:_(s32) = COPY $vgpr11
+    %16:_(s32) = COPY $vgpr12
+    %17:_(s32) = COPY $vgpr13
+    %18:_(s32) = COPY $vgpr14
+    %19:_(s32) = COPY $vgpr15
+    %32:_(s64) = G_MERGE_VALUES %12(s32), %13(s32)
+    %33:_(s64) = G_MERGE_VALUES %14(s32), %15(s32)
+    %34:_(s64) = G_MERGE_VALUES %16(s32), %17(s32)
+    %35:_(s64) = G_MERGE_VALUES %18(s32), %19(s32)
+    %1:_(<4 x s64>) = G_BUILD_VECTOR %32(s64), %33(s64), %34(s64), %35(s64)
+    %20:_(s32) = COPY $vgpr16
+    %21:_(s32) = COPY $vgpr17
+    %22:_(s32) = COPY $vgpr18
+    %23:_(s32) = COPY $vgpr19
+    %24:_(s32) = COPY $vgpr20
+    %25:_(s32) = COPY $vgpr21
+    %26:_(s32) = COPY $vgpr22
+    %27:_(s32) = COPY $vgpr23
+    %36:_(s64) = G_MERGE_VALUES %20(s32), %21(s32)
+    %37:_(s64) = G_MERGE_VALUES %22(s32), %23(s32)
+    %38:_(s64) = G_MERGE_VALUES %24(s32), %25(s32)
+    %39:_(s64) = G_MERGE_VALUES %26(s32), %27(s32)
+    %2:_(<4 x s64>) = G_BUILD_VECTOR %36(s64), %37(s64), %38(s64), %39(s64)
+    %3:sgpr_64 = COPY $sgpr30_sgpr31
+    %40:_(<4 x s64>) = reassoc G_FMUL %0, %1
+    %41:_(<4 x s64>) = reassoc G_FADD %40, %2
+    %43:_(s32), %44:_(s32), %45:_(s32), %46:_(s32), %47:_(s32), %48:_(s32), %49:_(s32), %50:_(s32) = G_UNMERGE_VALUES %41(<4 x s64>)
+    $vgpr0 = COPY %43(s32)
+    $vgpr1 = COPY %44(s32)
+    $vgpr2 = COPY %45(s32)
+    $vgpr3 = COPY %46(s32)
+    $vgpr4 = COPY %47(s32)
+    $vgpr5 = COPY %48(s32)
+    $vgpr6 = COPY %49(s32)
+    $vgpr7 = COPY %50(s32)
+    %42:ccr_sgpr_64 = COPY %3
+    S_SETPC_B64_return %42, implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3, implicit $vgpr4, implicit $vgpr5, implicit $vgpr6, implicit $vgpr7
+...
+
+---
+name:            test_3xdouble_add_mul_rhs
+body:             |
+  bb.1.entry:
+    liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8, $vgpr9, $vgpr10, $vgpr11, $vgpr12, $vgpr13, $vgpr14, $vgpr15, $vgpr16, $vgpr17, $sgpr30_sgpr31
+
+    ; GFX9-LABEL: name: test_3xdouble_add_mul_rhs
+    ; GFX9: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX9: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX9: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX9: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3
+    ; GFX9: [[COPY4:%[0-9]+]]:_(s32) = COPY $vgpr4
+    ; GFX9: [[COPY5:%[0-9]+]]:_(s32) = COPY $vgpr5
+    ; GFX9: [[MV:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY]](s32), [[COPY1]](s32)
+    ; GFX9: [[MV1:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY2]](s32), [[COPY3]](s32)
+    ; GFX9: [[MV2:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY4]](s32), [[COPY5]](s32)
+    ; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<3 x s64>) = G_BUILD_VECTOR [[MV]](s64), [[MV1]](s64), [[MV2]](s64)
+    ; GFX9: [[COPY6:%[0-9]+]]:_(s32) = COPY $vgpr6
+    ; GFX9: [[COPY7:%[0-9]+]]:_(s32) = COPY $vgpr7
+    ; GFX9: [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr8
+    ; GFX9: [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr9
+    ; GFX9: [[COPY10:%[0-9]+]]:_(s32) = COPY $vgpr10
+    ; GFX9: [[COPY11:%[0-9]+]]:_(s32) = COPY $vgpr11
+    ; GFX9: [[MV3:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY6]](s32), [[COPY7]](s32)
+    ; GFX9: [[MV4:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY8]](s32), [[COPY9]](s32)
+    ; GFX9: [[MV5:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY10]](s32), [[COPY11]](s32)
+    ; GFX9: [[BUILD_VECTOR1:%[0-9]+]]:_(<3 x s64>) = G_BUILD_VECTOR [[MV3]](s64), [[MV4]](s64), [[MV5]](s64)
+    ; GFX9: [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr12
+    ; GFX9: [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr13
+    ; GFX9: [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr14
+    ; GFX9: [[COPY15:%[0-9]+]]:_(s32) = COPY $vgpr15
+    ; GFX9: [[COPY16:%[0-9]+]]:_(s32) = COPY $vgpr16
+    ; GFX9: [[COPY17:%[0-9]+]]:_(s32) = COPY $vgpr17
+    ; GFX9: [[MV6:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY12]](s32), [[COPY13]](s32)
+    ; GFX9: [[MV7:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY14]](s32), [[COPY15]](s32)
+    ; GFX9: [[MV8:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY16]](s32), [[COPY17]](s32)
+    ; GFX9: [[BUILD_VECTOR2:%[0-9]+]]:_(<3 x s64>) = G_BUILD_VECTOR [[MV6]](s64), [[MV7]](s64), [[MV8]](s64)
+    ; GFX9: [[COPY18:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX9: [[FMUL:%[0-9]+]]:_(<3 x s64>) = reassoc G_FMUL [[BUILD_VECTOR]], [[BUILD_VECTOR1]]
+    ; GFX9: [[FADD:%[0-9]+]]:_(<3 x s64>) = reassoc G_FADD [[BUILD_VECTOR2]], [[FMUL]]
+    ; GFX9: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32), [[UV4:%[0-9]+]]:_(s32), [[UV5:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FADD]](<3 x s64>)
+    ; GFX9: $vgpr0 = COPY [[UV]](s32)
+    ; GFX9: $vgpr1 = COPY [[UV1]](s32)
+    ; GFX9: $vgpr2 = COPY [[UV2]](s32)
+    ; GFX9: $vgpr3 = COPY [[UV3]](s32)
+    ; GFX9: $vgpr4 = COPY [[UV4]](s32)
+    ; GFX9: $vgpr5 = COPY [[UV5]](s32)
+    ; GFX9: [[COPY19:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY18]]
+    ; GFX9: S_SETPC_B64_return [[COPY19]], implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3, implicit $vgpr4, implicit $vgpr5
+    ; GFX9-CONTRACT-LABEL: name: test_3xdouble_add_mul_rhs
+    ; GFX9-CONTRACT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX9-CONTRACT: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX9-CONTRACT: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX9-CONTRACT: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3
+    ; GFX9-CONTRACT: [[COPY4:%[0-9]+]]:_(s32) = COPY $vgpr4
+    ; GFX9-CONTRACT: [[COPY5:%[0-9]+]]:_(s32) = COPY $vgpr5
+    ; GFX9-CONTRACT: [[MV:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY]](s32), [[COPY1]](s32)
+    ; GFX9-CONTRACT: [[MV1:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY2]](s32), [[COPY3]](s32)
+    ; GFX9-CONTRACT: [[MV2:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY4]](s32), [[COPY5]](s32)
+    ; GFX9-CONTRACT: [[BUILD_VECTOR:%[0-9]+]]:_(<3 x s64>) = G_BUILD_VECTOR [[MV]](s64), [[MV1]](s64), [[MV2]](s64)
+    ; GFX9-CONTRACT: [[COPY6:%[0-9]+]]:_(s32) = COPY $vgpr6
+    ; GFX9-CONTRACT: [[COPY7:%[0-9]+]]:_(s32) = COPY $vgpr7
+    ; GFX9-CONTRACT: [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr8
+    ; GFX9-CONTRACT: [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr9
+    ; GFX9-CONTRACT: [[COPY10:%[0-9]+]]:_(s32) = COPY $vgpr10
+    ; GFX9-CONTRACT: [[COPY11:%[0-9]+]]:_(s32) = COPY $vgpr11
+    ; GFX9-CONTRACT: [[MV3:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY6]](s32), [[COPY7]](s32)
+    ; GFX9-CONTRACT: [[MV4:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY8]](s32), [[COPY9]](s32)
+    ; GFX9-CONTRACT: [[MV5:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY10]](s32), [[COPY11]](s32)
+    ; GFX9-CONTRACT: [[BUILD_VECTOR1:%[0-9]+]]:_(<3 x s64>) = G_BUILD_VECTOR [[MV3]](s64), [[MV4]](s64), [[MV5]](s64)
+    ; GFX9-CONTRACT: [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr12
+    ; GFX9-CONTRACT: [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr13
+    ; GFX9-CONTRACT: [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr14
+    ; GFX9-CONTRACT: [[COPY15:%[0-9]+]]:_(s32) = COPY $vgpr15
+    ; GFX9-CONTRACT: [[COPY16:%[0-9]+]]:_(s32) = COPY $vgpr16
+    ; GFX9-CONTRACT: [[COPY17:%[0-9]+]]:_(s32) = COPY $vgpr17
+    ; GFX9-CONTRACT: [[MV6:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY12]](s32), [[COPY13]](s32)
+    ; GFX9-CONTRACT: [[MV7:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY14]](s32), [[COPY15]](s32)
+    ; GFX9-CONTRACT: [[MV8:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY16]](s32), [[COPY17]](s32)
+    ; GFX9-CONTRACT: [[BUILD_VECTOR2:%[0-9]+]]:_(<3 x s64>) = G_BUILD_VECTOR [[MV6]](s64), [[MV7]](s64), [[MV8]](s64)
+    ; GFX9-CONTRACT: [[COPY18:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX9-CONTRACT: [[FMA:%[0-9]+]]:_(<3 x s64>) = G_FMA [[BUILD_VECTOR]], [[BUILD_VECTOR1]], [[BUILD_VECTOR2]]
+    ; GFX9-CONTRACT: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32), [[UV4:%[0-9]+]]:_(s32), [[UV5:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FMA]](<3 x s64>)
+    ; GFX9-CONTRACT: $vgpr0 = COPY [[UV]](s32)
+    ; GFX9-CONTRACT: $vgpr1 = COPY [[UV1]](s32)
+    ; GFX9-CONTRACT: $vgpr2 = COPY [[UV2]](s32)
+    ; GFX9-CONTRACT: $vgpr3 = COPY [[UV3]](s32)
+    ; GFX9-CONTRACT: $vgpr4 = COPY [[UV4]](s32)
+    ; GFX9-CONTRACT: $vgpr5 = COPY [[UV5]](s32)
+    ; GFX9-CONTRACT: [[COPY19:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY18]]
+    ; GFX9-CONTRACT: S_SETPC_B64_return [[COPY19]], implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3, implicit $vgpr4, implicit $vgpr5
+    ; GFX9-DENORM-LABEL: name: test_3xdouble_add_mul_rhs
+    ; GFX9-DENORM: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX9-DENORM: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX9-DENORM: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX9-DENORM: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3
+    ; GFX9-DENORM: [[COPY4:%[0-9]+]]:_(s32) = COPY $vgpr4
+    ; GFX9-DENORM: [[COPY5:%[0-9]+]]:_(s32) = COPY $vgpr5
+    ; GFX9-DENORM: [[MV:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY]](s32), [[COPY1]](s32)
+    ; GFX9-DENORM: [[MV1:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY2]](s32), [[COPY3]](s32)
+    ; GFX9-DENORM: [[MV2:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY4]](s32), [[COPY5]](s32)
+    ; GFX9-DENORM: [[BUILD_VECTOR:%[0-9]+]]:_(<3 x s64>) = G_BUILD_VECTOR [[MV]](s64), [[MV1]](s64), [[MV2]](s64)
+    ; GFX9-DENORM: [[COPY6:%[0-9]+]]:_(s32) = COPY $vgpr6
+    ; GFX9-DENORM: [[COPY7:%[0-9]+]]:_(s32) = COPY $vgpr7
+    ; GFX9-DENORM: [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr8
+    ; GFX9-DENORM: [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr9
+    ; GFX9-DENORM: [[COPY10:%[0-9]+]]:_(s32) = COPY $vgpr10
+    ; GFX9-DENORM: [[COPY11:%[0-9]+]]:_(s32) = COPY $vgpr11
+    ; GFX9-DENORM: [[MV3:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY6]](s32), [[COPY7]](s32)
+    ; GFX9-DENORM: [[MV4:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY8]](s32), [[COPY9]](s32)
+    ; GFX9-DENORM: [[MV5:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY10]](s32), [[COPY11]](s32)
+    ; GFX9-DENORM: [[BUILD_VECTOR1:%[0-9]+]]:_(<3 x s64>) = G_BUILD_VECTOR [[MV3]](s64), [[MV4]](s64), [[MV5]](s64)
+    ; GFX9-DENORM: [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr12
+    ; GFX9-DENORM: [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr13
+    ; GFX9-DENORM: [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr14
+    ; GFX9-DENORM: [[COPY15:%[0-9]+]]:_(s32) = COPY $vgpr15
+    ; GFX9-DENORM: [[COPY16:%[0-9]+]]:_(s32) = COPY $vgpr16
+    ; GFX9-DENORM: [[COPY17:%[0-9]+]]:_(s32) = COPY $vgpr17
+    ; GFX9-DENORM: [[MV6:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY12]](s32), [[COPY13]](s32)
+    ; GFX9-DENORM: [[MV7:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY14]](s32), [[COPY15]](s32)
+    ; GFX9-DENORM: [[MV8:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY16]](s32), [[COPY17]](s32)
+    ; GFX9-DENORM: [[BUILD_VECTOR2:%[0-9]+]]:_(<3 x s64>) = G_BUILD_VECTOR [[MV6]](s64), [[MV7]](s64), [[MV8]](s64)
+    ; GFX9-DENORM: [[COPY18:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX9-DENORM: [[FMUL:%[0-9]+]]:_(<3 x s64>) = reassoc G_FMUL [[BUILD_VECTOR]], [[BUILD_VECTOR1]]
+    ; GFX9-DENORM: [[FADD:%[0-9]+]]:_(<3 x s64>) = reassoc G_FADD [[BUILD_VECTOR2]], [[FMUL]]
+    ; GFX9-DENORM: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32), [[UV4:%[0-9]+]]:_(s32), [[UV5:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FADD]](<3 x s64>)
+    ; GFX9-DENORM: $vgpr0 = COPY [[UV]](s32)
+    ; GFX9-DENORM: $vgpr1 = COPY [[UV1]](s32)
+    ; GFX9-DENORM: $vgpr2 = COPY [[UV2]](s32)
+    ; GFX9-DENORM: $vgpr3 = COPY [[UV3]](s32)
+    ; GFX9-DENORM: $vgpr4 = COPY [[UV4]](s32)
+    ; GFX9-DENORM: $vgpr5 = COPY [[UV5]](s32)
+    ; GFX9-DENORM: [[COPY19:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY18]]
+    ; GFX9-DENORM: S_SETPC_B64_return [[COPY19]], implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3, implicit $vgpr4, implicit $vgpr5
+    ; GFX9-UNSAFE-LABEL: name: test_3xdouble_add_mul_rhs
+    ; GFX9-UNSAFE: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX9-UNSAFE: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX9-UNSAFE: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX9-UNSAFE: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3
+    ; GFX9-UNSAFE: [[COPY4:%[0-9]+]]:_(s32) = COPY $vgpr4
+    ; GFX9-UNSAFE: [[COPY5:%[0-9]+]]:_(s32) = COPY $vgpr5
+    ; GFX9-UNSAFE: [[MV:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY]](s32), [[COPY1]](s32)
+    ; GFX9-UNSAFE: [[MV1:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY2]](s32), [[COPY3]](s32)
+    ; GFX9-UNSAFE: [[MV2:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY4]](s32), [[COPY5]](s32)
+    ; GFX9-UNSAFE: [[BUILD_VECTOR:%[0-9]+]]:_(<3 x s64>) = G_BUILD_VECTOR [[MV]](s64), [[MV1]](s64), [[MV2]](s64)
+    ; GFX9-UNSAFE: [[COPY6:%[0-9]+]]:_(s32) = COPY $vgpr6
+    ; GFX9-UNSAFE: [[COPY7:%[0-9]+]]:_(s32) = COPY $vgpr7
+    ; GFX9-UNSAFE: [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr8
+    ; GFX9-UNSAFE: [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr9
+    ; GFX9-UNSAFE: [[COPY10:%[0-9]+]]:_(s32) = COPY $vgpr10
+    ; GFX9-UNSAFE: [[COPY11:%[0-9]+]]:_(s32) = COPY $vgpr11
+    ; GFX9-UNSAFE: [[MV3:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY6]](s32), [[COPY7]](s32)
+    ; GFX9-UNSAFE: [[MV4:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY8]](s32), [[COPY9]](s32)
+    ; GFX9-UNSAFE: [[MV5:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY10]](s32), [[COPY11]](s32)
+    ; GFX9-UNSAFE: [[BUILD_VECTOR1:%[0-9]+]]:_(<3 x s64>) = G_BUILD_VECTOR [[MV3]](s64), [[MV4]](s64), [[MV5]](s64)
+    ; GFX9-UNSAFE: [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr12
+    ; GFX9-UNSAFE: [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr13
+    ; GFX9-UNSAFE: [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr14
+    ; GFX9-UNSAFE: [[COPY15:%[0-9]+]]:_(s32) = COPY $vgpr15
+    ; GFX9-UNSAFE: [[COPY16:%[0-9]+]]:_(s32) = COPY $vgpr16
+    ; GFX9-UNSAFE: [[COPY17:%[0-9]+]]:_(s32) = COPY $vgpr17
+    ; GFX9-UNSAFE: [[MV6:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY12]](s32), [[COPY13]](s32)
+    ; GFX9-UNSAFE: [[MV7:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY14]](s32), [[COPY15]](s32)
+    ; GFX9-UNSAFE: [[MV8:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY16]](s32), [[COPY17]](s32)
+    ; GFX9-UNSAFE: [[BUILD_VECTOR2:%[0-9]+]]:_(<3 x s64>) = G_BUILD_VECTOR [[MV6]](s64), [[MV7]](s64), [[MV8]](s64)
+    ; GFX9-UNSAFE: [[COPY18:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX9-UNSAFE: [[FMA:%[0-9]+]]:_(<3 x s64>) = G_FMA [[BUILD_VECTOR]], [[BUILD_VECTOR1]], [[BUILD_VECTOR2]]
+    ; GFX9-UNSAFE: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32), [[UV4:%[0-9]+]]:_(s32), [[UV5:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FMA]](<3 x s64>)
+    ; GFX9-UNSAFE: $vgpr0 = COPY [[UV]](s32)
+    ; GFX9-UNSAFE: $vgpr1 = COPY [[UV1]](s32)
+    ; GFX9-UNSAFE: $vgpr2 = COPY [[UV2]](s32)
+    ; GFX9-UNSAFE: $vgpr3 = COPY [[UV3]](s32)
+    ; GFX9-UNSAFE: $vgpr4 = COPY [[UV4]](s32)
+    ; GFX9-UNSAFE: $vgpr5 = COPY [[UV5]](s32)
+    ; GFX9-UNSAFE: [[COPY19:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY18]]
+    ; GFX9-UNSAFE: S_SETPC_B64_return [[COPY19]], implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3, implicit $vgpr4, implicit $vgpr5
+    ; GFX10-LABEL: name: test_3xdouble_add_mul_rhs
+    ; GFX10: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX10: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX10: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX10: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3
+    ; GFX10: [[COPY4:%[0-9]+]]:_(s32) = COPY $vgpr4
+    ; GFX10: [[COPY5:%[0-9]+]]:_(s32) = COPY $vgpr5
+    ; GFX10: [[MV:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY]](s32), [[COPY1]](s32)
+    ; GFX10: [[MV1:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY2]](s32), [[COPY3]](s32)
+    ; GFX10: [[MV2:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY4]](s32), [[COPY5]](s32)
+    ; GFX10: [[BUILD_VECTOR:%[0-9]+]]:_(<3 x s64>) = G_BUILD_VECTOR [[MV]](s64), [[MV1]](s64), [[MV2]](s64)
+    ; GFX10: [[COPY6:%[0-9]+]]:_(s32) = COPY $vgpr6
+    ; GFX10: [[COPY7:%[0-9]+]]:_(s32) = COPY $vgpr7
+    ; GFX10: [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr8
+    ; GFX10: [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr9
+    ; GFX10: [[COPY10:%[0-9]+]]:_(s32) = COPY $vgpr10
+    ; GFX10: [[COPY11:%[0-9]+]]:_(s32) = COPY $vgpr11
+    ; GFX10: [[MV3:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY6]](s32), [[COPY7]](s32)
+    ; GFX10: [[MV4:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY8]](s32), [[COPY9]](s32)
+    ; GFX10: [[MV5:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY10]](s32), [[COPY11]](s32)
+    ; GFX10: [[BUILD_VECTOR1:%[0-9]+]]:_(<3 x s64>) = G_BUILD_VECTOR [[MV3]](s64), [[MV4]](s64), [[MV5]](s64)
+    ; GFX10: [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr12
+    ; GFX10: [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr13
+    ; GFX10: [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr14
+    ; GFX10: [[COPY15:%[0-9]+]]:_(s32) = COPY $vgpr15
+    ; GFX10: [[COPY16:%[0-9]+]]:_(s32) = COPY $vgpr16
+    ; GFX10: [[COPY17:%[0-9]+]]:_(s32) = COPY $vgpr17
+    ; GFX10: [[MV6:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY12]](s32), [[COPY13]](s32)
+    ; GFX10: [[MV7:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY14]](s32), [[COPY15]](s32)
+    ; GFX10: [[MV8:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY16]](s32), [[COPY17]](s32)
+    ; GFX10: [[BUILD_VECTOR2:%[0-9]+]]:_(<3 x s64>) = G_BUILD_VECTOR [[MV6]](s64), [[MV7]](s64), [[MV8]](s64)
+    ; GFX10: [[COPY18:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX10: [[FMUL:%[0-9]+]]:_(<3 x s64>) = reassoc G_FMUL [[BUILD_VECTOR]], [[BUILD_VECTOR1]]
+    ; GFX10: [[FADD:%[0-9]+]]:_(<3 x s64>) = reassoc G_FADD [[BUILD_VECTOR2]], [[FMUL]]
+    ; GFX10: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32), [[UV4:%[0-9]+]]:_(s32), [[UV5:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FADD]](<3 x s64>)
+    ; GFX10: $vgpr0 = COPY [[UV]](s32)
+    ; GFX10: $vgpr1 = COPY [[UV1]](s32)
+    ; GFX10: $vgpr2 = COPY [[UV2]](s32)
+    ; GFX10: $vgpr3 = COPY [[UV3]](s32)
+    ; GFX10: $vgpr4 = COPY [[UV4]](s32)
+    ; GFX10: $vgpr5 = COPY [[UV5]](s32)
+    ; GFX10: [[COPY19:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY18]]
+    ; GFX10: S_SETPC_B64_return [[COPY19]], implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3, implicit $vgpr4, implicit $vgpr5
+    ; GFX10-CONTRACT-LABEL: name: test_3xdouble_add_mul_rhs
+    ; GFX10-CONTRACT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX10-CONTRACT: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX10-CONTRACT: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX10-CONTRACT: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3
+    ; GFX10-CONTRACT: [[COPY4:%[0-9]+]]:_(s32) = COPY $vgpr4
+    ; GFX10-CONTRACT: [[COPY5:%[0-9]+]]:_(s32) = COPY $vgpr5
+    ; GFX10-CONTRACT: [[MV:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY]](s32), [[COPY1]](s32)
+    ; GFX10-CONTRACT: [[MV1:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY2]](s32), [[COPY3]](s32)
+    ; GFX10-CONTRACT: [[MV2:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY4]](s32), [[COPY5]](s32)
+    ; GFX10-CONTRACT: [[BUILD_VECTOR:%[0-9]+]]:_(<3 x s64>) = G_BUILD_VECTOR [[MV]](s64), [[MV1]](s64), [[MV2]](s64)
+    ; GFX10-CONTRACT: [[COPY6:%[0-9]+]]:_(s32) = COPY $vgpr6
+    ; GFX10-CONTRACT: [[COPY7:%[0-9]+]]:_(s32) = COPY $vgpr7
+    ; GFX10-CONTRACT: [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr8
+    ; GFX10-CONTRACT: [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr9
+    ; GFX10-CONTRACT: [[COPY10:%[0-9]+]]:_(s32) = COPY $vgpr10
+    ; GFX10-CONTRACT: [[COPY11:%[0-9]+]]:_(s32) = COPY $vgpr11
+    ; GFX10-CONTRACT: [[MV3:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY6]](s32), [[COPY7]](s32)
+    ; GFX10-CONTRACT: [[MV4:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY8]](s32), [[COPY9]](s32)
+    ; GFX10-CONTRACT: [[MV5:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY10]](s32), [[COPY11]](s32)
+    ; GFX10-CONTRACT: [[BUILD_VECTOR1:%[0-9]+]]:_(<3 x s64>) = G_BUILD_VECTOR [[MV3]](s64), [[MV4]](s64), [[MV5]](s64)
+    ; GFX10-CONTRACT: [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr12
+    ; GFX10-CONTRACT: [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr13
+    ; GFX10-CONTRACT: [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr14
+    ; GFX10-CONTRACT: [[COPY15:%[0-9]+]]:_(s32) = COPY $vgpr15
+    ; GFX10-CONTRACT: [[COPY16:%[0-9]+]]:_(s32) = COPY $vgpr16
+    ; GFX10-CONTRACT: [[COPY17:%[0-9]+]]:_(s32) = COPY $vgpr17
+    ; GFX10-CONTRACT: [[MV6:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY12]](s32), [[COPY13]](s32)
+    ; GFX10-CONTRACT: [[MV7:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY14]](s32), [[COPY15]](s32)
+    ; GFX10-CONTRACT: [[MV8:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY16]](s32), [[COPY17]](s32)
+    ; GFX10-CONTRACT: [[BUILD_VECTOR2:%[0-9]+]]:_(<3 x s64>) = G_BUILD_VECTOR [[MV6]](s64), [[MV7]](s64), [[MV8]](s64)
+    ; GFX10-CONTRACT: [[COPY18:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX10-CONTRACT: [[FMA:%[0-9]+]]:_(<3 x s64>) = G_FMA [[BUILD_VECTOR]], [[BUILD_VECTOR1]], [[BUILD_VECTOR2]]
+    ; GFX10-CONTRACT: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32), [[UV4:%[0-9]+]]:_(s32), [[UV5:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FMA]](<3 x s64>)
+    ; GFX10-CONTRACT: $vgpr0 = COPY [[UV]](s32)
+    ; GFX10-CONTRACT: $vgpr1 = COPY [[UV1]](s32)
+    ; GFX10-CONTRACT: $vgpr2 = COPY [[UV2]](s32)
+    ; GFX10-CONTRACT: $vgpr3 = COPY [[UV3]](s32)
+    ; GFX10-CONTRACT: $vgpr4 = COPY [[UV4]](s32)
+    ; GFX10-CONTRACT: $vgpr5 = COPY [[UV5]](s32)
+    ; GFX10-CONTRACT: [[COPY19:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY18]]
+    ; GFX10-CONTRACT: S_SETPC_B64_return [[COPY19]], implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3, implicit $vgpr4, implicit $vgpr5
+    ; GFX10-DENORM-LABEL: name: test_3xdouble_add_mul_rhs
+    ; GFX10-DENORM: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX10-DENORM: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX10-DENORM: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX10-DENORM: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3
+    ; GFX10-DENORM: [[COPY4:%[0-9]+]]:_(s32) = COPY $vgpr4
+    ; GFX10-DENORM: [[COPY5:%[0-9]+]]:_(s32) = COPY $vgpr5
+    ; GFX10-DENORM: [[MV:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY]](s32), [[COPY1]](s32)
+    ; GFX10-DENORM: [[MV1:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY2]](s32), [[COPY3]](s32)
+    ; GFX10-DENORM: [[MV2:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY4]](s32), [[COPY5]](s32)
+    ; GFX10-DENORM: [[BUILD_VECTOR:%[0-9]+]]:_(<3 x s64>) = G_BUILD_VECTOR [[MV]](s64), [[MV1]](s64), [[MV2]](s64)
+    ; GFX10-DENORM: [[COPY6:%[0-9]+]]:_(s32) = COPY $vgpr6
+    ; GFX10-DENORM: [[COPY7:%[0-9]+]]:_(s32) = COPY $vgpr7
+    ; GFX10-DENORM: [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr8
+    ; GFX10-DENORM: [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr9
+    ; GFX10-DENORM: [[COPY10:%[0-9]+]]:_(s32) = COPY $vgpr10
+    ; GFX10-DENORM: [[COPY11:%[0-9]+]]:_(s32) = COPY $vgpr11
+    ; GFX10-DENORM: [[MV3:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY6]](s32), [[COPY7]](s32)
+    ; GFX10-DENORM: [[MV4:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY8]](s32), [[COPY9]](s32)
+    ; GFX10-DENORM: [[MV5:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY10]](s32), [[COPY11]](s32)
+    ; GFX10-DENORM: [[BUILD_VECTOR1:%[0-9]+]]:_(<3 x s64>) = G_BUILD_VECTOR [[MV3]](s64), [[MV4]](s64), [[MV5]](s64)
+    ; GFX10-DENORM: [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr12
+    ; GFX10-DENORM: [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr13
+    ; GFX10-DENORM: [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr14
+    ; GFX10-DENORM: [[COPY15:%[0-9]+]]:_(s32) = COPY $vgpr15
+    ; GFX10-DENORM: [[COPY16:%[0-9]+]]:_(s32) = COPY $vgpr16
+    ; GFX10-DENORM: [[COPY17:%[0-9]+]]:_(s32) = COPY $vgpr17
+    ; GFX10-DENORM: [[MV6:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY12]](s32), [[COPY13]](s32)
+    ; GFX10-DENORM: [[MV7:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY14]](s32), [[COPY15]](s32)
+    ; GFX10-DENORM: [[MV8:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY16]](s32), [[COPY17]](s32)
+    ; GFX10-DENORM: [[BUILD_VECTOR2:%[0-9]+]]:_(<3 x s64>) = G_BUILD_VECTOR [[MV6]](s64), [[MV7]](s64), [[MV8]](s64)
+    ; GFX10-DENORM: [[COPY18:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX10-DENORM: [[FMUL:%[0-9]+]]:_(<3 x s64>) = reassoc G_FMUL [[BUILD_VECTOR]], [[BUILD_VECTOR1]]
+    ; GFX10-DENORM: [[FADD:%[0-9]+]]:_(<3 x s64>) = reassoc G_FADD [[BUILD_VECTOR2]], [[FMUL]]
+    ; GFX10-DENORM: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32), [[UV4:%[0-9]+]]:_(s32), [[UV5:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FADD]](<3 x s64>)
+    ; GFX10-DENORM: $vgpr0 = COPY [[UV]](s32)
+    ; GFX10-DENORM: $vgpr1 = COPY [[UV1]](s32)
+    ; GFX10-DENORM: $vgpr2 = COPY [[UV2]](s32)
+    ; GFX10-DENORM: $vgpr3 = COPY [[UV3]](s32)
+    ; GFX10-DENORM: $vgpr4 = COPY [[UV4]](s32)
+    ; GFX10-DENORM: $vgpr5 = COPY [[UV5]](s32)
+    ; GFX10-DENORM: [[COPY19:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY18]]
+    ; GFX10-DENORM: S_SETPC_B64_return [[COPY19]], implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3, implicit $vgpr4, implicit $vgpr5
+    ; GFX10-UNSAFE-LABEL: name: test_3xdouble_add_mul_rhs
+    ; GFX10-UNSAFE: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX10-UNSAFE: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX10-UNSAFE: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX10-UNSAFE: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3
+    ; GFX10-UNSAFE: [[COPY4:%[0-9]+]]:_(s32) = COPY $vgpr4
+    ; GFX10-UNSAFE: [[COPY5:%[0-9]+]]:_(s32) = COPY $vgpr5
+    ; GFX10-UNSAFE: [[MV:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY]](s32), [[COPY1]](s32)
+    ; GFX10-UNSAFE: [[MV1:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY2]](s32), [[COPY3]](s32)
+    ; GFX10-UNSAFE: [[MV2:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY4]](s32), [[COPY5]](s32)
+    ; GFX10-UNSAFE: [[BUILD_VECTOR:%[0-9]+]]:_(<3 x s64>) = G_BUILD_VECTOR [[MV]](s64), [[MV1]](s64), [[MV2]](s64)
+    ; GFX10-UNSAFE: [[COPY6:%[0-9]+]]:_(s32) = COPY $vgpr6
+    ; GFX10-UNSAFE: [[COPY7:%[0-9]+]]:_(s32) = COPY $vgpr7
+    ; GFX10-UNSAFE: [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr8
+    ; GFX10-UNSAFE: [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr9
+    ; GFX10-UNSAFE: [[COPY10:%[0-9]+]]:_(s32) = COPY $vgpr10
+    ; GFX10-UNSAFE: [[COPY11:%[0-9]+]]:_(s32) = COPY $vgpr11
+    ; GFX10-UNSAFE: [[MV3:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY6]](s32), [[COPY7]](s32)
+    ; GFX10-UNSAFE: [[MV4:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY8]](s32), [[COPY9]](s32)
+    ; GFX10-UNSAFE: [[MV5:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY10]](s32), [[COPY11]](s32)
+    ; GFX10-UNSAFE: [[BUILD_VECTOR1:%[0-9]+]]:_(<3 x s64>) = G_BUILD_VECTOR [[MV3]](s64), [[MV4]](s64), [[MV5]](s64)
+    ; GFX10-UNSAFE: [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr12
+    ; GFX10-UNSAFE: [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr13
+    ; GFX10-UNSAFE: [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr14
+    ; GFX10-UNSAFE: [[COPY15:%[0-9]+]]:_(s32) = COPY $vgpr15
+    ; GFX10-UNSAFE: [[COPY16:%[0-9]+]]:_(s32) = COPY $vgpr16
+    ; GFX10-UNSAFE: [[COPY17:%[0-9]+]]:_(s32) = COPY $vgpr17
+    ; GFX10-UNSAFE: [[MV6:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY12]](s32), [[COPY13]](s32)
+    ; GFX10-UNSAFE: [[MV7:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY14]](s32), [[COPY15]](s32)
+    ; GFX10-UNSAFE: [[MV8:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY16]](s32), [[COPY17]](s32)
+    ; GFX10-UNSAFE: [[BUILD_VECTOR2:%[0-9]+]]:_(<3 x s64>) = G_BUILD_VECTOR [[MV6]](s64), [[MV7]](s64), [[MV8]](s64)
+    ; GFX10-UNSAFE: [[COPY18:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
+    ; GFX10-UNSAFE: [[FMA:%[0-9]+]]:_(<3 x s64>) = G_FMA [[BUILD_VECTOR]], [[BUILD_VECTOR1]], [[BUILD_VECTOR2]]
+    ; GFX10-UNSAFE: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32), [[UV4:%[0-9]+]]:_(s32), [[UV5:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FMA]](<3 x s64>)
+    ; GFX10-UNSAFE: $vgpr0 = COPY [[UV]](s32)
+    ; GFX10-UNSAFE: $vgpr1 = COPY [[UV1]](s32)
+    ; GFX10-UNSAFE: $vgpr2 = COPY [[UV2]](s32)
+    ; GFX10-UNSAFE: $vgpr3 = COPY [[UV3]](s32)
+    ; GFX10-UNSAFE: $vgpr4 = COPY [[UV4]](s32)
+    ; GFX10-UNSAFE: $vgpr5 = COPY [[UV5]](s32)
+    ; GFX10-UNSAFE: [[COPY19:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY18]]
+    ; GFX10-UNSAFE: S_SETPC_B64_return [[COPY19]], implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3, implicit $vgpr4, implicit $vgpr5
+    %4:_(s32) = COPY $vgpr0
+    %5:_(s32) = COPY $vgpr1
+    %6:_(s32) = COPY $vgpr2
+    %7:_(s32) = COPY $vgpr3
+    %8:_(s32) = COPY $vgpr4
+    %9:_(s32) = COPY $vgpr5
+    %22:_(s64) = G_MERGE_VALUES %4(s32), %5(s32)
+    %23:_(s64) = G_MERGE_VALUES %6(s32), %7(s32)
+    %24:_(s64) = G_MERGE_VALUES %8(s32), %9(s32)
+    %0:_(<3 x s64>) = G_BUILD_VECTOR %22(s64), %23(s64), %24(s64)
+    %10:_(s32) = COPY $vgpr6
+    %11:_(s32) = COPY $vgpr7
+    %12:_(s32) = COPY $vgpr8
+    %13:_(s32) = COPY $vgpr9
+    %14:_(s32) = COPY $vgpr10
+    %15:_(s32) = COPY $vgpr11
+    %25:_(s64) = G_MERGE_VALUES %10(s32), %11(s32)
+    %26:_(s64) = G_MERGE_VALUES %12(s32), %13(s32)
+    %27:_(s64) = G_MERGE_VALUES %14(s32), %15(s32)
+    %1:_(<3 x s64>) = G_BUILD_VECTOR %25(s64), %26(s64), %27(s64)
+    %16:_(s32) = COPY $vgpr12
+    %17:_(s32) = COPY $vgpr13
+    %18:_(s32) = COPY $vgpr14
+    %19:_(s32) = COPY $vgpr15
+    %20:_(s32) = COPY $vgpr16
+    %21:_(s32) = COPY $vgpr17
+    %28:_(s64) = G_MERGE_VALUES %16(s32), %17(s32)
+    %29:_(s64) = G_MERGE_VALUES %18(s32), %19(s32)
+    %30:_(s64) = G_MERGE_VALUES %20(s32), %21(s32)
+    %2:_(<3 x s64>) = G_BUILD_VECTOR %28(s64), %29(s64), %30(s64)
+    %3:sgpr_64 = COPY $sgpr30_sgpr31
+    %31:_(<3 x s64>) = reassoc G_FMUL %0, %1
+    %32:_(<3 x s64>) = reassoc G_FADD %2, %31
+    %34:_(s32), %35:_(s32), %36:_(s32), %37:_(s32), %38:_(s32), %39:_(s32) = G_UNMERGE_VALUES %32(<3 x s64>)
+    $vgpr0 = COPY %34(s32)
+    $vgpr1 = COPY %35(s32)
+    $vgpr2 = COPY %36(s32)
+    $vgpr3 = COPY %37(s32)
+    $vgpr4 = COPY %38(s32)
+    $vgpr5 = COPY %39(s32)
+    %33:ccr_sgpr_64 = COPY %3
+    S_SETPC_B64_return %33, implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3, implicit $vgpr4, implicit $vgpr5
+...

llvm/test/CodeGen/AMDGPU/GlobalISel/combine-fma-add-mul.ll

+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc -global-isel -march=amdgcn -mcpu=gfx900 < %s | FileCheck -check-prefix=GFX9 %s
+; RUN: llc -global-isel -march=amdgcn -mcpu=gfx900 -fp-contract=fast < %s | FileCheck -check-prefix=GFX9-CONTRACT %s
+; RUN: llc -global-isel -march=amdgcn -mcpu=gfx900 --denormal-fp-math=preserve-sign < %s | FileCheck -check-prefix=GFX9-DENORM %s
+; RUN: llc -global-isel -march=amdgcn -mcpu=gfx900 -enable-unsafe-fp-math < %s | FileCheck -check-prefix=GFX9-UNSAFE %s
+; RUN: llc -global-isel -march=amdgcn -mcpu=gfx1010 < %s | FileCheck -check-prefix=GFX10 %s
+; RUN: llc -global-isel -march=amdgcn -mcpu=gfx1010 -fp-contract=fast < %s | FileCheck -check-prefix=GFX10-CONTRACT %s
+; RUN: llc -global-isel -march=amdgcn -mcpu=gfx1010 --denormal-fp-math=preserve-sign < %s | FileCheck -check-prefix=GFX10-DENORM %s
+; RUN: llc -global-isel -march=amdgcn -mcpu=gfx1010 -enable-unsafe-fp-math < %s | FileCheck -check-prefix=GFX10-UNSAFE %s
+
+define float @test_f32_add_mul(float %x, float %y, float %z) {
+; GFX9-LABEL: test_f32_add_mul:
+; GFX9:       ; %bb.0: ; %.entry
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    v_mul_f32_e32 v0, v0, v1
+; GFX9-NEXT:    v_add_f32_e32 v0, v0, v2
+; GFX9-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX9-CONTRACT-LABEL: test_f32_add_mul:
+; GFX9-CONTRACT:       ; %bb.0: ; %.entry
+; GFX9-CONTRACT-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-CONTRACT-NEXT:    v_fma_f32 v0, v0, v1, v2
+; GFX9-CONTRACT-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX9-DENORM-LABEL: test_f32_add_mul:
+; GFX9-DENORM:       ; %bb.0: ; %.entry
+; GFX9-DENORM-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-DENORM-NEXT:    v_mad_f32 v0, v0, v1, v2
+; GFX9-DENORM-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX9-UNSAFE-LABEL: test_f32_add_mul:
+; GFX9-UNSAFE:       ; %bb.0: ; %.entry
+; GFX9-UNSAFE-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-UNSAFE-NEXT:    v_fma_f32 v0, v0, v1, v2
+; GFX9-UNSAFE-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX10-LABEL: test_f32_add_mul:
+; GFX10:       ; %bb.0: ; %.entry
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    v_mul_f32_e32 v0, v0, v1
+; GFX10-NEXT:    v_add_f32_e32 v0, v0, v2
+; GFX10-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX10-CONTRACT-LABEL: test_f32_add_mul:
+; GFX10-CONTRACT:       ; %bb.0: ; %.entry
+; GFX10-CONTRACT-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-CONTRACT-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-CONTRACT-NEXT:    v_fma_f32 v0, v0, v1, v2
+; GFX10-CONTRACT-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX10-DENORM-LABEL: test_f32_add_mul:
+; GFX10-DENORM:       ; %bb.0: ; %.entry
+; GFX10-DENORM-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-DENORM-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-DENORM-NEXT:    v_mad_f32 v0, v0, v1, v2
+; GFX10-DENORM-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX10-UNSAFE-LABEL: test_f32_add_mul:
+; GFX10-UNSAFE:       ; %bb.0: ; %.entry
+; GFX10-UNSAFE-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-UNSAFE-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-UNSAFE-NEXT:    v_fma_f32 v0, v0, v1, v2
+; GFX10-UNSAFE-NEXT:    s_setpc_b64 s[30:31]
+.entry:
+  %a = fmul float %x, %y
+  %b = fadd float %a, %z
+  ret float %b
+}
+
+define float @test_f32_add_mul_rhs(float %x, float %y, float %z) {
+; GFX9-LABEL: test_f32_add_mul_rhs:
+; GFX9:       ; %bb.0: ; %.entry
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    v_mul_f32_e32 v0, v0, v1
+; GFX9-NEXT:    v_add_f32_e32 v0, v2, v0
+; GFX9-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX9-CONTRACT-LABEL: test_f32_add_mul_rhs:
+; GFX9-CONTRACT:       ; %bb.0: ; %.entry
+; GFX9-CONTRACT-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-CONTRACT-NEXT:    v_fma_f32 v0, v0, v1, v2
+; GFX9-CONTRACT-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX9-DENORM-LABEL: test_f32_add_mul_rhs:
+; GFX9-DENORM:       ; %bb.0: ; %.entry
+; GFX9-DENORM-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-DENORM-NEXT:    v_mad_f32 v0, v0, v1, v2
+; GFX9-DENORM-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX9-UNSAFE-LABEL: test_f32_add_mul_rhs:
+; GFX9-UNSAFE:       ; %bb.0: ; %.entry
+; GFX9-UNSAFE-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-UNSAFE-NEXT:    v_fma_f32 v0, v0, v1, v2
+; GFX9-UNSAFE-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX10-LABEL: test_f32_add_mul_rhs:
+; GFX10:       ; %bb.0: ; %.entry
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    v_mul_f32_e32 v0, v0, v1
+; GFX10-NEXT:    v_add_f32_e32 v0, v2, v0
+; GFX10-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX10-CONTRACT-LABEL: test_f32_add_mul_rhs:
+; GFX10-CONTRACT:       ; %bb.0: ; %.entry
+; GFX10-CONTRACT-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-CONTRACT-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-CONTRACT-NEXT:    v_fma_f32 v0, v0, v1, v2
+; GFX10-CONTRACT-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX10-DENORM-LABEL: test_f32_add_mul_rhs:
+; GFX10-DENORM:       ; %bb.0: ; %.entry
+; GFX10-DENORM-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-DENORM-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-DENORM-NEXT:    v_mad_f32 v0, v0, v1, v2
+; GFX10-DENORM-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX10-UNSAFE-LABEL: test_f32_add_mul_rhs:
+; GFX10-UNSAFE:       ; %bb.0: ; %.entry
+; GFX10-UNSAFE-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-UNSAFE-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-UNSAFE-NEXT:    v_fma_f32 v0, v0, v1, v2
+; GFX10-UNSAFE-NEXT:    s_setpc_b64 s[30:31]
+.entry:
+  %a = fmul float %x, %y
+  %b = fadd float %z, %a
+  ret float %b
+}
+
+define half @test_half_add_mul(half %x, half %y, half %z) {
+; GFX9-LABEL: test_half_add_mul:
+; GFX9:       ; %bb.0: ; %.entry
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    v_mul_f16_e32 v0, v0, v1
+; GFX9-NEXT:    v_add_f16_e32 v0, v0, v2
+; GFX9-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX9-CONTRACT-LABEL: test_half_add_mul:
+; GFX9-CONTRACT:       ; %bb.0: ; %.entry
+; GFX9-CONTRACT-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-CONTRACT-NEXT:    v_fma_f16 v0, v0, v1, v2
+; GFX9-CONTRACT-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX9-DENORM-LABEL: test_half_add_mul:
+; GFX9-DENORM:       ; %bb.0: ; %.entry
+; GFX9-DENORM-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-DENORM-NEXT:    v_mad_legacy_f16 v0, v0, v1, v2
+; GFX9-DENORM-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX9-UNSAFE-LABEL: test_half_add_mul:
+; GFX9-UNSAFE:       ; %bb.0: ; %.entry
+; GFX9-UNSAFE-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-UNSAFE-NEXT:    v_fma_f16 v0, v0, v1, v2
+; GFX9-UNSAFE-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX10-LABEL: test_half_add_mul:
+; GFX10:       ; %bb.0: ; %.entry
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    v_mul_f16_e32 v0, v0, v1
+; GFX10-NEXT:    v_add_f16_e32 v0, v0, v2
+; GFX10-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX10-CONTRACT-LABEL: test_half_add_mul:
+; GFX10-CONTRACT:       ; %bb.0: ; %.entry
+; GFX10-CONTRACT-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-CONTRACT-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-CONTRACT-NEXT:    v_fmac_f16_e32 v2, v0, v1
+; GFX10-CONTRACT-NEXT:    v_mov_b32_e32 v0, v2
+; GFX10-CONTRACT-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX10-DENORM-LABEL: test_half_add_mul:
+; GFX10-DENORM:       ; %bb.0: ; %.entry
+; GFX10-DENORM-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-DENORM-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-DENORM-NEXT:    v_mul_f16_e32 v0, v0, v1
+; GFX10-DENORM-NEXT:    v_add_f16_e32 v0, v0, v2
+; GFX10-DENORM-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX10-UNSAFE-LABEL: test_half_add_mul:
+; GFX10-UNSAFE:       ; %bb.0: ; %.entry
+; GFX10-UNSAFE-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-UNSAFE-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-UNSAFE-NEXT:    v_fmac_f16_e32 v2, v0, v1
+; GFX10-UNSAFE-NEXT:    v_mov_b32_e32 v0, v2
+; GFX10-UNSAFE-NEXT:    s_setpc_b64 s[30:31]
+.entry:
+  %a = fmul half %x, %y
+  %b = fadd half %a, %z
+  ret half %b
+}
+
+define half @test_half_add_mul_rhs(half %x, half %y, half %z) {
+; GFX9-LABEL: test_half_add_mul_rhs:
+; GFX9:       ; %bb.0: ; %.entry
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    v_mul_f16_e32 v0, v0, v1
+; GFX9-NEXT:    v_add_f16_e32 v0, v2, v0
+; GFX9-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX9-CONTRACT-LABEL: test_half_add_mul_rhs:
+; GFX9-CONTRACT:       ; %bb.0: ; %.entry
+; GFX9-CONTRACT-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-CONTRACT-NEXT:    v_fma_f16 v0, v0, v1, v2
+; GFX9-CONTRACT-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX9-DENORM-LABEL: test_half_add_mul_rhs:
+; GFX9-DENORM:       ; %bb.0: ; %.entry
+; GFX9-DENORM-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-DENORM-NEXT:    v_mad_legacy_f16 v0, v0, v1, v2
+; GFX9-DENORM-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX9-UNSAFE-LABEL: test_half_add_mul_rhs:
+; GFX9-UNSAFE:       ; %bb.0: ; %.entry
+; GFX9-UNSAFE-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-UNSAFE-NEXT:    v_fma_f16 v0, v0, v1, v2
+; GFX9-UNSAFE-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX10-LABEL: test_half_add_mul_rhs:
+; GFX10:       ; %bb.0: ; %.entry
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    v_mul_f16_e32 v0, v0, v1
+; GFX10-NEXT:    v_add_f16_e32 v0, v2, v0
+; GFX10-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX10-CONTRACT-LABEL: test_half_add_mul_rhs:
+; GFX10-CONTRACT:       ; %bb.0: ; %.entry
+; GFX10-CONTRACT-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-CONTRACT-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-CONTRACT-NEXT:    v_fmac_f16_e32 v2, v0, v1
+; GFX10-CONTRACT-NEXT:    v_mov_b32_e32 v0, v2
+; GFX10-CONTRACT-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX10-DENORM-LABEL: test_half_add_mul_rhs:
+; GFX10-DENORM:       ; %bb.0: ; %.entry
+; GFX10-DENORM-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-DENORM-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-DENORM-NEXT:    v_mul_f16_e32 v0, v0, v1
+; GFX10-DENORM-NEXT:    v_add_f16_e32 v0, v2, v0
+; GFX10-DENORM-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX10-UNSAFE-LABEL: test_half_add_mul_rhs:
+; GFX10-UNSAFE:       ; %bb.0: ; %.entry
+; GFX10-UNSAFE-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-UNSAFE-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-UNSAFE-NEXT:    v_fmac_f16_e32 v2, v0, v1
+; GFX10-UNSAFE-NEXT:    v_mov_b32_e32 v0, v2
+; GFX10-UNSAFE-NEXT:    s_setpc_b64 s[30:31]
+.entry:
+  %a = fmul half %x, %y
+  %b = fadd half %z, %a
+  ret half %b
+}
+
+define double @test_double_add_mul(double %x, double %y, double %z) {
+; GFX9-LABEL: test_double_add_mul:
+; GFX9:       ; %bb.0: ; %.entry
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    v_mul_f64 v[0:1], v[0:1], v[2:3]
+; GFX9-NEXT:    v_add_f64 v[0:1], v[0:1], v[4:5]
+; GFX9-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX9-CONTRACT-LABEL: test_double_add_mul:
+; GFX9-CONTRACT:       ; %bb.0: ; %.entry
+; GFX9-CONTRACT-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-CONTRACT-NEXT:    v_fma_f64 v[0:1], v[0:1], v[2:3], v[4:5]
+; GFX9-CONTRACT-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX9-DENORM-LABEL: test_double_add_mul:
+; GFX9-DENORM:       ; %bb.0: ; %.entry
+; GFX9-DENORM-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-DENORM-NEXT:    v_mul_f64 v[0:1], v[0:1], v[2:3]
+; GFX9-DENORM-NEXT:    v_add_f64 v[0:1], v[0:1], v[4:5]
+; GFX9-DENORM-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX9-UNSAFE-LABEL: test_double_add_mul:
+; GFX9-UNSAFE:       ; %bb.0: ; %.entry
+; GFX9-UNSAFE-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-UNSAFE-NEXT:    v_fma_f64 v[0:1], v[0:1], v[2:3], v[4:5]
+; GFX9-UNSAFE-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX10-LABEL: test_double_add_mul:
+; GFX10:       ; %bb.0: ; %.entry
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    v_mul_f64 v[0:1], v[0:1], v[2:3]
+; GFX10-NEXT:    v_add_f64 v[0:1], v[0:1], v[4:5]
+; GFX10-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX10-CONTRACT-LABEL: test_double_add_mul:
+; GFX10-CONTRACT:       ; %bb.0: ; %.entry
+; GFX10-CONTRACT-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-CONTRACT-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-CONTRACT-NEXT:    v_fma_f64 v[0:1], v[0:1], v[2:3], v[4:5]
+; GFX10-CONTRACT-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX10-DENORM-LABEL: test_double_add_mul:
+; GFX10-DENORM:       ; %bb.0: ; %.entry
+; GFX10-DENORM-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-DENORM-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-DENORM-NEXT:    v_mul_f64 v[0:1], v[0:1], v[2:3]
+; GFX10-DENORM-NEXT:    v_add_f64 v[0:1], v[0:1], v[4:5]
+; GFX10-DENORM-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX10-UNSAFE-LABEL: test_double_add_mul:
+; GFX10-UNSAFE:       ; %bb.0: ; %.entry
+; GFX10-UNSAFE-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-UNSAFE-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-UNSAFE-NEXT:    v_fma_f64 v[0:1], v[0:1], v[2:3], v[4:5]
+; GFX10-UNSAFE-NEXT:    s_setpc_b64 s[30:31]
+.entry:
+  %a = fmul double %x, %y
+  %b = fadd double %a, %z
+  ret double %b
+}
+
+define double @test_double_add_mul_rhs(double %x, double %y, double %z) {
+; GFX9-LABEL: test_double_add_mul_rhs:
+; GFX9:       ; %bb.0: ; %.entry
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    v_mul_f64 v[0:1], v[0:1], v[2:3]
+; GFX9-NEXT:    v_add_f64 v[0:1], v[4:5], v[0:1]
+; GFX9-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX9-CONTRACT-LABEL: test_double_add_mul_rhs:
+; GFX9-CONTRACT:       ; %bb.0: ; %.entry
+; GFX9-CONTRACT-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-CONTRACT-NEXT:    v_fma_f64 v[0:1], v[0:1], v[2:3], v[4:5]
+; GFX9-CONTRACT-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX9-DENORM-LABEL: test_double_add_mul_rhs:
+; GFX9-DENORM:       ; %bb.0: ; %.entry
+; GFX9-DENORM-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-DENORM-NEXT:    v_mul_f64 v[0:1], v[0:1], v[2:3]
+; GFX9-DENORM-NEXT:    v_add_f64 v[0:1], v[4:5], v[0:1]
+; GFX9-DENORM-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX9-UNSAFE-LABEL: test_double_add_mul_rhs:
+; GFX9-UNSAFE:       ; %bb.0: ; %.entry
+; GFX9-UNSAFE-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-UNSAFE-NEXT:    v_fma_f64 v[0:1], v[0:1], v[2:3], v[4:5]
+; GFX9-UNSAFE-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX10-LABEL: test_double_add_mul_rhs:
+; GFX10:       ; %bb.0: ; %.entry
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    v_mul_f64 v[0:1], v[0:1], v[2:3]
+; GFX10-NEXT:    v_add_f64 v[0:1], v[4:5], v[0:1]
+; GFX10-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX10-CONTRACT-LABEL: test_double_add_mul_rhs:
+; GFX10-CONTRACT:       ; %bb.0: ; %.entry
+; GFX10-CONTRACT-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-CONTRACT-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-CONTRACT-NEXT:    v_fma_f64 v[0:1], v[0:1], v[2:3], v[4:5]
+; GFX10-CONTRACT-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX10-DENORM-LABEL: test_double_add_mul_rhs:
+; GFX10-DENORM:       ; %bb.0: ; %.entry
+; GFX10-DENORM-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-DENORM-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-DENORM-NEXT:    v_mul_f64 v[0:1], v[0:1], v[2:3]
+; GFX10-DENORM-NEXT:    v_add_f64 v[0:1], v[4:5], v[0:1]
+; GFX10-DENORM-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX10-UNSAFE-LABEL: test_double_add_mul_rhs:
+; GFX10-UNSAFE:       ; %bb.0: ; %.entry
+; GFX10-UNSAFE-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-UNSAFE-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-UNSAFE-NEXT:    v_fma_f64 v[0:1], v[0:1], v[2:3], v[4:5]
+; GFX10-UNSAFE-NEXT:    s_setpc_b64 s[30:31]
+.entry:
+  %a = fmul double %x, %y
+  %b = fadd double %z, %a
+  ret double %b
+}
+
+define <4 x float> @test_4xfloat_add_mul(<4 x float> %x, <4 x float> %y, <4 x float> %z) {
+; GFX9-LABEL: test_4xfloat_add_mul:
+; GFX9:       ; %bb.0: ; %.entry
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    v_mul_f32_e32 v0, v0, v4
+; GFX9-NEXT:    v_mul_f32_e32 v1, v1, v5
+; GFX9-NEXT:    v_mul_f32_e32 v2, v2, v6
+; GFX9-NEXT:    v_mul_f32_e32 v3, v3, v7
+; GFX9-NEXT:    v_add_f32_e32 v0, v0, v8
+; GFX9-NEXT:    v_add_f32_e32 v1, v1, v9
+; GFX9-NEXT:    v_add_f32_e32 v2, v2, v10
+; GFX9-NEXT:    v_add_f32_e32 v3, v3, v11
+; GFX9-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX9-CONTRACT-LABEL: test_4xfloat_add_mul:
+; GFX9-CONTRACT:       ; %bb.0: ; %.entry
+; GFX9-CONTRACT-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-CONTRACT-NEXT:    v_fma_f32 v0, v0, v4, v8
+; GFX9-CONTRACT-NEXT:    v_fma_f32 v1, v1, v5, v9
+; GFX9-CONTRACT-NEXT:    v_fma_f32 v2, v2, v6, v10
+; GFX9-CONTRACT-NEXT:    v_fma_f32 v3, v3, v7, v11
+; GFX9-CONTRACT-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX9-DENORM-LABEL: test_4xfloat_add_mul:
+; GFX9-DENORM:       ; %bb.0: ; %.entry
+; GFX9-DENORM-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-DENORM-NEXT:    v_mad_f32 v0, v0, v4, v8
+; GFX9-DENORM-NEXT:    v_mad_f32 v1, v1, v5, v9
+; GFX9-DENORM-NEXT:    v_mad_f32 v2, v2, v6, v10
+; GFX9-DENORM-NEXT:    v_mad_f32 v3, v3, v7, v11
+; GFX9-DENORM-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX9-UNSAFE-LABEL: test_4xfloat_add_mul:
+; GFX9-UNSAFE:       ; %bb.0: ; %.entry
+; GFX9-UNSAFE-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-UNSAFE-NEXT:    v_fma_f32 v0, v0, v4, v8
+; GFX9-UNSAFE-NEXT:    v_fma_f32 v1, v1, v5, v9
+; GFX9-UNSAFE-NEXT:    v_fma_f32 v2, v2, v6, v10
+; GFX9-UNSAFE-NEXT:    v_fma_f32 v3, v3, v7, v11
+; GFX9-UNSAFE-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX10-LABEL: test_4xfloat_add_mul:
+; GFX10:       ; %bb.0: ; %.entry
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    v_mul_f32_e32 v0, v0, v4
+; GFX10-NEXT:    v_mul_f32_e32 v1, v1, v5
+; GFX10-NEXT:    v_mul_f32_e32 v2, v2, v6
+; GFX10-NEXT:    v_mul_f32_e32 v3, v3, v7
+; GFX10-NEXT:    v_add_f32_e32 v0, v0, v8
+; GFX10-NEXT:    v_add_f32_e32 v1, v1, v9
+; GFX10-NEXT:    v_add_f32_e32 v2, v2, v10
+; GFX10-NEXT:    v_add_f32_e32 v3, v3, v11
+; GFX10-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX10-CONTRACT-LABEL: test_4xfloat_add_mul:
+; GFX10-CONTRACT:       ; %bb.0: ; %.entry
+; GFX10-CONTRACT-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-CONTRACT-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-CONTRACT-NEXT:    v_fma_f32 v0, v0, v4, v8
+; GFX10-CONTRACT-NEXT:    v_fma_f32 v1, v1, v5, v9
+; GFX10-CONTRACT-NEXT:    v_fma_f32 v2, v2, v6, v10
+; GFX10-CONTRACT-NEXT:    v_fma_f32 v3, v3, v7, v11
+; GFX10-CONTRACT-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX10-DENORM-LABEL: test_4xfloat_add_mul:
+; GFX10-DENORM:       ; %bb.0: ; %.entry
+; GFX10-DENORM-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-DENORM-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-DENORM-NEXT:    v_mad_f32 v0, v0, v4, v8
+; GFX10-DENORM-NEXT:    v_mad_f32 v1, v1, v5, v9
+; GFX10-DENORM-NEXT:    v_mad_f32 v2, v2, v6, v10
+; GFX10-DENORM-NEXT:    v_mad_f32 v3, v3, v7, v11
+; GFX10-DENORM-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX10-UNSAFE-LABEL: test_4xfloat_add_mul:
+; GFX10-UNSAFE:       ; %bb.0: ; %.entry
+; GFX10-UNSAFE-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-UNSAFE-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-UNSAFE-NEXT:    v_fma_f32 v0, v0, v4, v8
+; GFX10-UNSAFE-NEXT:    v_fma_f32 v1, v1, v5, v9
+; GFX10-UNSAFE-NEXT:    v_fma_f32 v2, v2, v6, v10
+; GFX10-UNSAFE-NEXT:    v_fma_f32 v3, v3, v7, v11
+; GFX10-UNSAFE-NEXT:    s_setpc_b64 s[30:31]
+.entry:
+  %a = fmul <4 x float> %x, %y
+  %b = fadd <4 x float> %a, %z
+  ret <4 x float> %b
+}
+
+define <3 x float> @test_3xfloat_add_mul_rhs(<3 x float> %x, <3 x float> %y, <3 x float> %z) {
+; GFX9-LABEL: test_3xfloat_add_mul_rhs:
+; GFX9:       ; %bb.0: ; %.entry
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    v_mul_f32_e32 v0, v0, v3
+; GFX9-NEXT:    v_mul_f32_e32 v1, v1, v4
+; GFX9-NEXT:    v_mul_f32_e32 v2, v2, v5
+; GFX9-NEXT:    v_add_f32_e32 v0, v6, v0
+; GFX9-NEXT:    v_add_f32_e32 v1, v7, v1
+; GFX9-NEXT:    v_add_f32_e32 v2, v8, v2
+; GFX9-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX9-CONTRACT-LABEL: test_3xfloat_add_mul_rhs:
+; GFX9-CONTRACT:       ; %bb.0: ; %.entry
+; GFX9-CONTRACT-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-CONTRACT-NEXT:    v_fma_f32 v0, v0, v3, v6
+; GFX9-CONTRACT-NEXT:    v_fma_f32 v1, v1, v4, v7
+; GFX9-CONTRACT-NEXT:    v_fma_f32 v2, v2, v5, v8
+; GFX9-CONTRACT-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX9-DENORM-LABEL: test_3xfloat_add_mul_rhs:
+; GFX9-DENORM:       ; %bb.0: ; %.entry
+; GFX9-DENORM-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-DENORM-NEXT:    v_mad_f32 v0, v0, v3, v6
+; GFX9-DENORM-NEXT:    v_mad_f32 v1, v1, v4, v7
+; GFX9-DENORM-NEXT:    v_mad_f32 v2, v2, v5, v8
+; GFX9-DENORM-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX9-UNSAFE-LABEL: test_3xfloat_add_mul_rhs:
+; GFX9-UNSAFE:       ; %bb.0: ; %.entry
+; GFX9-UNSAFE-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-UNSAFE-NEXT:    v_fma_f32 v0, v0, v3, v6
+; GFX9-UNSAFE-NEXT:    v_fma_f32 v1, v1, v4, v7
+; GFX9-UNSAFE-NEXT:    v_fma_f32 v2, v2, v5, v8
+; GFX9-UNSAFE-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX10-LABEL: test_3xfloat_add_mul_rhs:
+; GFX10:       ; %bb.0: ; %.entry
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    v_mul_f32_e32 v0, v0, v3
+; GFX10-NEXT:    v_mul_f32_e32 v1, v1, v4
+; GFX10-NEXT:    v_mul_f32_e32 v2, v2, v5
+; GFX10-NEXT:    v_add_f32_e32 v0, v6, v0
+; GFX10-NEXT:    v_add_f32_e32 v1, v7, v1
+; GFX10-NEXT:    v_add_f32_e32 v2, v8, v2
+; GFX10-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX10-CONTRACT-LABEL: test_3xfloat_add_mul_rhs:
+; GFX10-CONTRACT:       ; %bb.0: ; %.entry
+; GFX10-CONTRACT-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-CONTRACT-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-CONTRACT-NEXT:    v_fma_f32 v0, v0, v3, v6
+; GFX10-CONTRACT-NEXT:    v_fma_f32 v1, v1, v4, v7
+; GFX10-CONTRACT-NEXT:    v_fma_f32 v2, v2, v5, v8
+; GFX10-CONTRACT-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX10-DENORM-LABEL: test_3xfloat_add_mul_rhs:
+; GFX10-DENORM:       ; %bb.0: ; %.entry
+; GFX10-DENORM-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-DENORM-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-DENORM-NEXT:    v_mad_f32 v0, v0, v3, v6
+; GFX10-DENORM-NEXT:    v_mad_f32 v1, v1, v4, v7
+; GFX10-DENORM-NEXT:    v_mad_f32 v2, v2, v5, v8
+; GFX10-DENORM-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX10-UNSAFE-LABEL: test_3xfloat_add_mul_rhs:
+; GFX10-UNSAFE:       ; %bb.0: ; %.entry
+; GFX10-UNSAFE-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-UNSAFE-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-UNSAFE-NEXT:    v_fma_f32 v0, v0, v3, v6
+; GFX10-UNSAFE-NEXT:    v_fma_f32 v1, v1, v4, v7
+; GFX10-UNSAFE-NEXT:    v_fma_f32 v2, v2, v5, v8
+; GFX10-UNSAFE-NEXT:    s_setpc_b64 s[30:31]
+.entry:
+  %a = fmul <3 x float> %x, %y
+  %b = fadd <3 x float> %z, %a
+  ret <3 x float> %b
+}
+
+define <4 x half> @test_4xhalf_add_mul(<4 x half> %x, <4 x half> %y, <4 x half> %z) {
+; GFX9-LABEL: test_4xhalf_add_mul:
+; GFX9:       ; %bb.0: ; %.entry
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    v_pk_mul_f16 v0, v0, v2
+; GFX9-NEXT:    v_pk_mul_f16 v1, v1, v3
+; GFX9-NEXT:    v_pk_add_f16 v0, v0, v4
+; GFX9-NEXT:    v_pk_add_f16 v1, v1, v5
+; GFX9-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX9-CONTRACT-LABEL: test_4xhalf_add_mul:
+; GFX9-CONTRACT:       ; %bb.0: ; %.entry
+; GFX9-CONTRACT-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-CONTRACT-NEXT:    v_pk_fma_f16 v0, v0, v2, v4
+; GFX9-CONTRACT-NEXT:    v_pk_fma_f16 v1, v1, v3, v5
+; GFX9-CONTRACT-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX9-DENORM-LABEL: test_4xhalf_add_mul:
+; GFX9-DENORM:       ; %bb.0: ; %.entry
+; GFX9-DENORM-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-DENORM-NEXT:    v_pk_mul_f16 v0, v0, v2
+; GFX9-DENORM-NEXT:    v_pk_mul_f16 v1, v1, v3
+; GFX9-DENORM-NEXT:    v_pk_add_f16 v0, v0, v4
+; GFX9-DENORM-NEXT:    v_pk_add_f16 v1, v1, v5
+; GFX9-DENORM-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX9-UNSAFE-LABEL: test_4xhalf_add_mul:
+; GFX9-UNSAFE:       ; %bb.0: ; %.entry
+; GFX9-UNSAFE-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-UNSAFE-NEXT:    v_pk_fma_f16 v0, v0, v2, v4
+; GFX9-UNSAFE-NEXT:    v_pk_fma_f16 v1, v1, v3, v5
+; GFX9-UNSAFE-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX10-LABEL: test_4xhalf_add_mul:
+; GFX10:       ; %bb.0: ; %.entry
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    v_pk_mul_f16 v0, v0, v2
+; GFX10-NEXT:    v_pk_mul_f16 v1, v1, v3
+; GFX10-NEXT:    v_pk_add_f16 v0, v0, v4
+; GFX10-NEXT:    v_pk_add_f16 v1, v1, v5
+; GFX10-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX10-CONTRACT-LABEL: test_4xhalf_add_mul:
+; GFX10-CONTRACT:       ; %bb.0: ; %.entry
+; GFX10-CONTRACT-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-CONTRACT-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-CONTRACT-NEXT:    v_pk_fma_f16 v0, v0, v2, v4
+; GFX10-CONTRACT-NEXT:    v_pk_fma_f16 v1, v1, v3, v5
+; GFX10-CONTRACT-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX10-DENORM-LABEL: test_4xhalf_add_mul:
+; GFX10-DENORM:       ; %bb.0: ; %.entry
+; GFX10-DENORM-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-DENORM-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-DENORM-NEXT:    v_pk_mul_f16 v0, v0, v2
+; GFX10-DENORM-NEXT:    v_pk_mul_f16 v1, v1, v3
+; GFX10-DENORM-NEXT:    v_pk_add_f16 v0, v0, v4
+; GFX10-DENORM-NEXT:    v_pk_add_f16 v1, v1, v5
+; GFX10-DENORM-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX10-UNSAFE-LABEL: test_4xhalf_add_mul:
+; GFX10-UNSAFE:       ; %bb.0: ; %.entry
+; GFX10-UNSAFE-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-UNSAFE-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-UNSAFE-NEXT:    v_pk_fma_f16 v0, v0, v2, v4
+; GFX10-UNSAFE-NEXT:    v_pk_fma_f16 v1, v1, v3, v5
+; GFX10-UNSAFE-NEXT:    s_setpc_b64 s[30:31]
+.entry:
+  %a = fmul <4 x half> %x, %y
+  %b = fadd <4 x half> %a, %z
+  ret <4 x half> %b
+}
+
+define <3 x half> @test_3xhalf_add_mul_rhs(<3 x half> %x, <3 x half> %y, <3 x half> %z) {
+; GFX9-LABEL: test_3xhalf_add_mul_rhs:
+; GFX9:       ; %bb.0: ; %.entry
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    v_lshrrev_b32_e32 v6, 16, v0
+; GFX9-NEXT:    v_lshrrev_b32_e32 v7, 16, v2
+; GFX9-NEXT:    v_mov_b32_e32 v9, 0xffff
+; GFX9-NEXT:    v_lshlrev_b32_e32 v6, 16, v6
+; GFX9-NEXT:    v_and_or_b32 v0, v0, v9, v6
+; GFX9-NEXT:    v_lshlrev_b32_e32 v6, 16, v7
+; GFX9-NEXT:    v_lshrrev_b32_e32 v8, 16, v4
+; GFX9-NEXT:    s_lshl_b32 s4, s4, 16
+; GFX9-NEXT:    v_and_or_b32 v2, v2, v9, v6
+; GFX9-NEXT:    v_and_or_b32 v1, v1, v9, s4
+; GFX9-NEXT:    v_and_or_b32 v3, v3, v9, s4
+; GFX9-NEXT:    v_pk_mul_f16 v0, v0, v2
+; GFX9-NEXT:    v_lshlrev_b32_e32 v2, 16, v8
+; GFX9-NEXT:    v_pk_mul_f16 v1, v1, v3
+; GFX9-NEXT:    v_and_or_b32 v2, v4, v9, v2
+; GFX9-NEXT:    v_lshrrev_b32_e32 v4, 16, v0
+; GFX9-NEXT:    v_and_or_b32 v3, v5, v9, s4
+; GFX9-NEXT:    v_lshlrev_b32_e32 v4, 16, v4
+; GFX9-NEXT:    v_and_or_b32 v1, v1, v9, s4
+; GFX9-NEXT:    v_and_or_b32 v0, v0, v9, v4
+; GFX9-NEXT:    v_pk_add_f16 v1, v3, v1
+; GFX9-NEXT:    s_lshl_b32 s4, s4, 16
+; GFX9-NEXT:    v_pk_add_f16 v0, v2, v0
+; GFX9-NEXT:    v_and_or_b32 v1, v1, v9, s4
+; GFX9-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX9-CONTRACT-LABEL: test_3xhalf_add_mul_rhs:
+; GFX9-CONTRACT:       ; %bb.0: ; %.entry
+; GFX9-CONTRACT-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-CONTRACT-NEXT:    v_lshrrev_b32_e32 v6, 16, v0
+; GFX9-CONTRACT-NEXT:    v_lshrrev_b32_e32 v7, 16, v2
+; GFX9-CONTRACT-NEXT:    v_mov_b32_e32 v9, 0xffff
+; GFX9-CONTRACT-NEXT:    v_lshlrev_b32_e32 v6, 16, v6
+; GFX9-CONTRACT-NEXT:    v_lshrrev_b32_e32 v8, 16, v4
+; GFX9-CONTRACT-NEXT:    v_and_or_b32 v0, v0, v9, v6
+; GFX9-CONTRACT-NEXT:    s_lshl_b32 s4, s4, 16
+; GFX9-CONTRACT-NEXT:    v_lshlrev_b32_e32 v6, 16, v7
+; GFX9-CONTRACT-NEXT:    v_and_or_b32 v1, v1, v9, s4
+; GFX9-CONTRACT-NEXT:    v_and_or_b32 v2, v2, v9, v6
+; GFX9-CONTRACT-NEXT:    v_and_or_b32 v3, v3, v9, s4
+; GFX9-CONTRACT-NEXT:    v_lshlrev_b32_e32 v6, 16, v8
+; GFX9-CONTRACT-NEXT:    v_and_or_b32 v5, v5, v9, s4
+; GFX9-CONTRACT-NEXT:    v_and_or_b32 v4, v4, v9, v6
+; GFX9-CONTRACT-NEXT:    v_pk_fma_f16 v1, v1, v3, v5
+; GFX9-CONTRACT-NEXT:    s_lshl_b32 s4, s4, 16
+; GFX9-CONTRACT-NEXT:    v_pk_fma_f16 v0, v0, v2, v4
+; GFX9-CONTRACT-NEXT:    v_and_or_b32 v1, v1, v9, s4
+; GFX9-CONTRACT-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX9-DENORM-LABEL: test_3xhalf_add_mul_rhs:
+; GFX9-DENORM:       ; %bb.0: ; %.entry
+; GFX9-DENORM-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-DENORM-NEXT:    v_lshrrev_b32_e32 v6, 16, v0
+; GFX9-DENORM-NEXT:    v_lshrrev_b32_e32 v7, 16, v2
+; GFX9-DENORM-NEXT:    v_mov_b32_e32 v9, 0xffff
+; GFX9-DENORM-NEXT:    v_lshlrev_b32_e32 v6, 16, v6
+; GFX9-DENORM-NEXT:    v_and_or_b32 v0, v0, v9, v6
+; GFX9-DENORM-NEXT:    v_lshlrev_b32_e32 v6, 16, v7
+; GFX9-DENORM-NEXT:    v_lshrrev_b32_e32 v8, 16, v4
+; GFX9-DENORM-NEXT:    s_lshl_b32 s4, s4, 16
+; GFX9-DENORM-NEXT:    v_and_or_b32 v2, v2, v9, v6
+; GFX9-DENORM-NEXT:    v_and_or_b32 v1, v1, v9, s4
+; GFX9-DENORM-NEXT:    v_and_or_b32 v3, v3, v9, s4
+; GFX9-DENORM-NEXT:    v_pk_mul_f16 v0, v0, v2
+; GFX9-DENORM-NEXT:    v_lshlrev_b32_e32 v2, 16, v8
+; GFX9-DENORM-NEXT:    v_pk_mul_f16 v1, v1, v3
+; GFX9-DENORM-NEXT:    v_and_or_b32 v2, v4, v9, v2
+; GFX9-DENORM-NEXT:    v_lshrrev_b32_e32 v4, 16, v0
+; GFX9-DENORM-NEXT:    v_and_or_b32 v3, v5, v9, s4
+; GFX9-DENORM-NEXT:    v_lshlrev_b32_e32 v4, 16, v4
+; GFX9-DENORM-NEXT:    v_and_or_b32 v1, v1, v9, s4
+; GFX9-DENORM-NEXT:    v_and_or_b32 v0, v0, v9, v4
+; GFX9-DENORM-NEXT:    v_pk_add_f16 v1, v3, v1
+; GFX9-DENORM-NEXT:    s_lshl_b32 s4, s4, 16
+; GFX9-DENORM-NEXT:    v_pk_add_f16 v0, v2, v0
+; GFX9-DENORM-NEXT:    v_and_or_b32 v1, v1, v9, s4
+; GFX9-DENORM-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX9-UNSAFE-LABEL: test_3xhalf_add_mul_rhs:
+; GFX9-UNSAFE:       ; %bb.0: ; %.entry
+; GFX9-UNSAFE-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-UNSAFE-NEXT:    v_lshrrev_b32_e32 v6, 16, v0
+; GFX9-UNSAFE-NEXT:    v_lshrrev_b32_e32 v7, 16, v2
+; GFX9-UNSAFE-NEXT:    v_mov_b32_e32 v9, 0xffff
+; GFX9-UNSAFE-NEXT:    v_lshlrev_b32_e32 v6, 16, v6
+; GFX9-UNSAFE-NEXT:    v_lshrrev_b32_e32 v8, 16, v4
+; GFX9-UNSAFE-NEXT:    v_and_or_b32 v0, v0, v9, v6
+; GFX9-UNSAFE-NEXT:    s_lshl_b32 s4, s4, 16
+; GFX9-UNSAFE-NEXT:    v_lshlrev_b32_e32 v6, 16, v7
+; GFX9-UNSAFE-NEXT:    v_and_or_b32 v1, v1, v9, s4
+; GFX9-UNSAFE-NEXT:    v_and_or_b32 v2, v2, v9, v6
+; GFX9-UNSAFE-NEXT:    v_and_or_b32 v3, v3, v9, s4
+; GFX9-UNSAFE-NEXT:    v_lshlrev_b32_e32 v6, 16, v8
+; GFX9-UNSAFE-NEXT:    v_and_or_b32 v5, v5, v9, s4
+; GFX9-UNSAFE-NEXT:    v_and_or_b32 v4, v4, v9, v6
+; GFX9-UNSAFE-NEXT:    v_pk_fma_f16 v1, v1, v3, v5
+; GFX9-UNSAFE-NEXT:    s_lshl_b32 s4, s4, 16
+; GFX9-UNSAFE-NEXT:    v_pk_fma_f16 v0, v0, v2, v4
+; GFX9-UNSAFE-NEXT:    v_and_or_b32 v1, v1, v9, s4
+; GFX9-UNSAFE-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX10-LABEL: test_3xhalf_add_mul_rhs:
+; GFX10:       ; %bb.0: ; %.entry
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    v_lshrrev_b32_e32 v6, 16, v0
+; GFX10-NEXT:    v_lshrrev_b32_e32 v7, 16, v2
+; GFX10-NEXT:    v_mov_b32_e32 v8, 0xffff
+; GFX10-NEXT:    s_lshl_b32 s4, s4, 16
+; GFX10-NEXT:    v_lshlrev_b32_e32 v6, 16, v6
+; GFX10-NEXT:    v_lshlrev_b32_e32 v7, 16, v7
+; GFX10-NEXT:    v_and_or_b32 v1, v1, v8, s4
+; GFX10-NEXT:    v_and_or_b32 v0, v0, v8, v6
+; GFX10-NEXT:    v_and_or_b32 v2, v2, v8, v7
+; GFX10-NEXT:    v_pk_mul_f16 v0, v0, v2
+; GFX10-NEXT:    v_and_or_b32 v2, v3, v8, s4
+; GFX10-NEXT:    v_lshrrev_b32_e32 v3, 16, v4
+; GFX10-NEXT:    v_lshrrev_b32_e32 v6, 16, v0
+; GFX10-NEXT:    v_pk_mul_f16 v1, v1, v2
+; GFX10-NEXT:    v_and_or_b32 v2, v5, v8, s4
+; GFX10-NEXT:    v_lshlrev_b32_e32 v3, 16, v3
+; GFX10-NEXT:    v_lshlrev_b32_e32 v5, 16, v6
+; GFX10-NEXT:    v_and_or_b32 v1, v1, v8, s4
+; GFX10-NEXT:    s_lshl_b32 s4, s4, 16
+; GFX10-NEXT:    v_and_or_b32 v3, v4, v8, v3
+; GFX10-NEXT:    v_and_or_b32 v0, v0, v8, v5
+; GFX10-NEXT:    v_pk_add_f16 v1, v2, v1
+; GFX10-NEXT:    v_pk_add_f16 v0, v3, v0
+; GFX10-NEXT:    v_and_or_b32 v1, v1, v8, s4
+; GFX10-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX10-CONTRACT-LABEL: test_3xhalf_add_mul_rhs:
+; GFX10-CONTRACT:       ; %bb.0: ; %.entry
+; GFX10-CONTRACT-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-CONTRACT-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-CONTRACT-NEXT:    v_lshrrev_b32_e32 v6, 16, v0
+; GFX10-CONTRACT-NEXT:    v_mov_b32_e32 v7, 0xffff
+; GFX10-CONTRACT-NEXT:    v_lshrrev_b32_e32 v8, 16, v2
+; GFX10-CONTRACT-NEXT:    v_lshrrev_b32_e32 v9, 16, v4
+; GFX10-CONTRACT-NEXT:    s_lshl_b32 s4, s4, 16
+; GFX10-CONTRACT-NEXT:    v_lshlrev_b32_e32 v6, 16, v6
+; GFX10-CONTRACT-NEXT:    v_and_or_b32 v1, v1, v7, s4
+; GFX10-CONTRACT-NEXT:    v_lshlrev_b32_e32 v8, 16, v8
+; GFX10-CONTRACT-NEXT:    v_and_or_b32 v3, v3, v7, s4
+; GFX10-CONTRACT-NEXT:    v_lshlrev_b32_e32 v9, 16, v9
+; GFX10-CONTRACT-NEXT:    v_and_or_b32 v5, v5, v7, s4
+; GFX10-CONTRACT-NEXT:    v_and_or_b32 v0, v0, v7, v6
+; GFX10-CONTRACT-NEXT:    v_and_or_b32 v2, v2, v7, v8
+; GFX10-CONTRACT-NEXT:    s_lshl_b32 s4, s4, 16
+; GFX10-CONTRACT-NEXT:    v_and_or_b32 v4, v4, v7, v9
+; GFX10-CONTRACT-NEXT:    v_pk_fma_f16 v1, v1, v3, v5
+; GFX10-CONTRACT-NEXT:    v_pk_fma_f16 v0, v0, v2, v4
+; GFX10-CONTRACT-NEXT:    v_and_or_b32 v1, v1, v7, s4
+; GFX10-CONTRACT-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX10-DENORM-LABEL: test_3xhalf_add_mul_rhs:
+; GFX10-DENORM:       ; %bb.0: ; %.entry
+; GFX10-DENORM-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-DENORM-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-DENORM-NEXT:    v_lshrrev_b32_e32 v6, 16, v0
+; GFX10-DENORM-NEXT:    v_lshrrev_b32_e32 v7, 16, v2
+; GFX10-DENORM-NEXT:    v_mov_b32_e32 v8, 0xffff
+; GFX10-DENORM-NEXT:    s_lshl_b32 s4, s4, 16
+; GFX10-DENORM-NEXT:    v_lshlrev_b32_e32 v6, 16, v6
+; GFX10-DENORM-NEXT:    v_lshlrev_b32_e32 v7, 16, v7
+; GFX10-DENORM-NEXT:    v_and_or_b32 v1, v1, v8, s4
+; GFX10-DENORM-NEXT:    v_and_or_b32 v0, v0, v8, v6
+; GFX10-DENORM-NEXT:    v_and_or_b32 v2, v2, v8, v7
+; GFX10-DENORM-NEXT:    v_pk_mul_f16 v0, v0, v2
+; GFX10-DENORM-NEXT:    v_and_or_b32 v2, v3, v8, s4
+; GFX10-DENORM-NEXT:    v_lshrrev_b32_e32 v3, 16, v4
+; GFX10-DENORM-NEXT:    v_lshrrev_b32_e32 v6, 16, v0
+; GFX10-DENORM-NEXT:    v_pk_mul_f16 v1, v1, v2
+; GFX10-DENORM-NEXT:    v_and_or_b32 v2, v5, v8, s4
+; GFX10-DENORM-NEXT:    v_lshlrev_b32_e32 v3, 16, v3
+; GFX10-DENORM-NEXT:    v_lshlrev_b32_e32 v5, 16, v6
+; GFX10-DENORM-NEXT:    v_and_or_b32 v1, v1, v8, s4
+; GFX10-DENORM-NEXT:    s_lshl_b32 s4, s4, 16
+; GFX10-DENORM-NEXT:    v_and_or_b32 v3, v4, v8, v3
+; GFX10-DENORM-NEXT:    v_and_or_b32 v0, v0, v8, v5
+; GFX10-DENORM-NEXT:    v_pk_add_f16 v1, v2, v1
+; GFX10-DENORM-NEXT:    v_pk_add_f16 v0, v3, v0
+; GFX10-DENORM-NEXT:    v_and_or_b32 v1, v1, v8, s4
+; GFX10-DENORM-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX10-UNSAFE-LABEL: test_3xhalf_add_mul_rhs:
+; GFX10-UNSAFE:       ; %bb.0: ; %.entry
+; GFX10-UNSAFE-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-UNSAFE-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-UNSAFE-NEXT:    v_lshrrev_b32_e32 v6, 16, v0
+; GFX10-UNSAFE-NEXT:    v_mov_b32_e32 v7, 0xffff
+; GFX10-UNSAFE-NEXT:    v_lshrrev_b32_e32 v8, 16, v2
+; GFX10-UNSAFE-NEXT:    v_lshrrev_b32_e32 v9, 16, v4
+; GFX10-UNSAFE-NEXT:    s_lshl_b32 s4, s4, 16
+; GFX10-UNSAFE-NEXT:    v_lshlrev_b32_e32 v6, 16, v6
+; GFX10-UNSAFE-NEXT:    v_and_or_b32 v1, v1, v7, s4
+; GFX10-UNSAFE-NEXT:    v_lshlrev_b32_e32 v8, 16, v8
+; GFX10-UNSAFE-NEXT:    v_and_or_b32 v3, v3, v7, s4
+; GFX10-UNSAFE-NEXT:    v_lshlrev_b32_e32 v9, 16, v9
+; GFX10-UNSAFE-NEXT:    v_and_or_b32 v5, v5, v7, s4
+; GFX10-UNSAFE-NEXT:    v_and_or_b32 v0, v0, v7, v6
+; GFX10-UNSAFE-NEXT:    v_and_or_b32 v2, v2, v7, v8
+; GFX10-UNSAFE-NEXT:    s_lshl_b32 s4, s4, 16
+; GFX10-UNSAFE-NEXT:    v_and_or_b32 v4, v4, v7, v9
+; GFX10-UNSAFE-NEXT:    v_pk_fma_f16 v1, v1, v3, v5
+; GFX10-UNSAFE-NEXT:    v_pk_fma_f16 v0, v0, v2, v4
+; GFX10-UNSAFE-NEXT:    v_and_or_b32 v1, v1, v7, s4
+; GFX10-UNSAFE-NEXT:    s_setpc_b64 s[30:31]
+.entry:
+  %a = fmul <3 x half> %x, %y
+  %b = fadd <3 x half> %z, %a
+  ret <3 x half> %b
+}
+
+define <4 x double> @test_4xdouble_add_mul(<4 x double> %x, <4 x double> %y, <4 x double> %z) {
+; GFX9-LABEL: test_4xdouble_add_mul:
+; GFX9:       ; %bb.0: ; %.entry
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    v_mul_f64 v[0:1], v[0:1], v[8:9]
+; GFX9-NEXT:    v_mul_f64 v[2:3], v[2:3], v[10:11]
+; GFX9-NEXT:    v_mul_f64 v[4:5], v[4:5], v[12:13]
+; GFX9-NEXT:    v_mul_f64 v[6:7], v[6:7], v[14:15]
+; GFX9-NEXT:    v_add_f64 v[0:1], v[0:1], v[16:17]
+; GFX9-NEXT:    v_add_f64 v[2:3], v[2:3], v[18:19]
+; GFX9-NEXT:    v_add_f64 v[4:5], v[4:5], v[20:21]
+; GFX9-NEXT:    v_add_f64 v[6:7], v[6:7], v[22:23]
+; GFX9-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX9-CONTRACT-LABEL: test_4xdouble_add_mul:
+; GFX9-CONTRACT:       ; %bb.0: ; %.entry
+; GFX9-CONTRACT-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-CONTRACT-NEXT:    v_fma_f64 v[0:1], v[0:1], v[8:9], v[16:17]
+; GFX9-CONTRACT-NEXT:    v_fma_f64 v[2:3], v[2:3], v[10:11], v[18:19]
+; GFX9-CONTRACT-NEXT:    v_fma_f64 v[4:5], v[4:5], v[12:13], v[20:21]
+; GFX9-CONTRACT-NEXT:    v_fma_f64 v[6:7], v[6:7], v[14:15], v[22:23]
+; GFX9-CONTRACT-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX9-DENORM-LABEL: test_4xdouble_add_mul:
+; GFX9-DENORM:       ; %bb.0: ; %.entry
+; GFX9-DENORM-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-DENORM-NEXT:    v_mul_f64 v[0:1], v[0:1], v[8:9]
+; GFX9-DENORM-NEXT:    v_mul_f64 v[2:3], v[2:3], v[10:11]
+; GFX9-DENORM-NEXT:    v_mul_f64 v[4:5], v[4:5], v[12:13]
+; GFX9-DENORM-NEXT:    v_mul_f64 v[6:7], v[6:7], v[14:15]
+; GFX9-DENORM-NEXT:    v_add_f64 v[0:1], v[0:1], v[16:17]
+; GFX9-DENORM-NEXT:    v_add_f64 v[2:3], v[2:3], v[18:19]
+; GFX9-DENORM-NEXT:    v_add_f64 v[4:5], v[4:5], v[20:21]
+; GFX9-DENORM-NEXT:    v_add_f64 v[6:7], v[6:7], v[22:23]
+; GFX9-DENORM-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX9-UNSAFE-LABEL: test_4xdouble_add_mul:
+; GFX9-UNSAFE:       ; %bb.0: ; %.entry
+; GFX9-UNSAFE-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-UNSAFE-NEXT:    v_fma_f64 v[0:1], v[0:1], v[8:9], v[16:17]
+; GFX9-UNSAFE-NEXT:    v_fma_f64 v[2:3], v[2:3], v[10:11], v[18:19]
+; GFX9-UNSAFE-NEXT:    v_fma_f64 v[4:5], v[4:5], v[12:13], v[20:21]
+; GFX9-UNSAFE-NEXT:    v_fma_f64 v[6:7], v[6:7], v[14:15], v[22:23]
+; GFX9-UNSAFE-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX10-LABEL: test_4xdouble_add_mul:
+; GFX10:       ; %bb.0: ; %.entry
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    v_mul_f64 v[0:1], v[0:1], v[8:9]
+; GFX10-NEXT:    v_mul_f64 v[2:3], v[2:3], v[10:11]
+; GFX10-NEXT:    v_mul_f64 v[4:5], v[4:5], v[12:13]
+; GFX10-NEXT:    v_mul_f64 v[6:7], v[6:7], v[14:15]
+; GFX10-NEXT:    v_add_f64 v[0:1], v[0:1], v[16:17]
+; GFX10-NEXT:    v_add_f64 v[2:3], v[2:3], v[18:19]
+; GFX10-NEXT:    v_add_f64 v[4:5], v[4:5], v[20:21]
+; GFX10-NEXT:    v_add_f64 v[6:7], v[6:7], v[22:23]
+; GFX10-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX10-CONTRACT-LABEL: test_4xdouble_add_mul:
+; GFX10-CONTRACT:       ; %bb.0: ; %.entry
+; GFX10-CONTRACT-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-CONTRACT-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-CONTRACT-NEXT:    v_fma_f64 v[0:1], v[0:1], v[8:9], v[16:17]
+; GFX10-CONTRACT-NEXT:    v_fma_f64 v[2:3], v[2:3], v[10:11], v[18:19]
+; GFX10-CONTRACT-NEXT:    v_fma_f64 v[4:5], v[4:5], v[12:13], v[20:21]
+; GFX10-CONTRACT-NEXT:    v_fma_f64 v[6:7], v[6:7], v[14:15], v[22:23]
+; GFX10-CONTRACT-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX10-DENORM-LABEL: test_4xdouble_add_mul:
+; GFX10-DENORM:       ; %bb.0: ; %.entry
+; GFX10-DENORM-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-DENORM-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-DENORM-NEXT:    v_mul_f64 v[0:1], v[0:1], v[8:9]
+; GFX10-DENORM-NEXT:    v_mul_f64 v[2:3], v[2:3], v[10:11]
+; GFX10-DENORM-NEXT:    v_mul_f64 v[4:5], v[4:5], v[12:13]
+; GFX10-DENORM-NEXT:    v_mul_f64 v[6:7], v[6:7], v[14:15]
+; GFX10-DENORM-NEXT:    v_add_f64 v[0:1], v[0:1], v[16:17]
+; GFX10-DENORM-NEXT:    v_add_f64 v[2:3], v[2:3], v[18:19]
+; GFX10-DENORM-NEXT:    v_add_f64 v[4:5], v[4:5], v[20:21]
+; GFX10-DENORM-NEXT:    v_add_f64 v[6:7], v[6:7], v[22:23]
+; GFX10-DENORM-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX10-UNSAFE-LABEL: test_4xdouble_add_mul:
+; GFX10-UNSAFE:       ; %bb.0: ; %.entry
+; GFX10-UNSAFE-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-UNSAFE-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-UNSAFE-NEXT:    v_fma_f64 v[0:1], v[0:1], v[8:9], v[16:17]
+; GFX10-UNSAFE-NEXT:    v_fma_f64 v[2:3], v[2:3], v[10:11], v[18:19]
+; GFX10-UNSAFE-NEXT:    v_fma_f64 v[4:5], v[4:5], v[12:13], v[20:21]
+; GFX10-UNSAFE-NEXT:    v_fma_f64 v[6:7], v[6:7], v[14:15], v[22:23]
+; GFX10-UNSAFE-NEXT:    s_setpc_b64 s[30:31]
+.entry:
+  %a = fmul <4 x double> %x, %y
+  %b = fadd <4 x double> %a, %z
+  ret <4 x double> %b
+}
+
+define <3 x double> @test_3xdouble_add_mul_rhs(<3 x double> %x, <3 x double> %y, <3 x double> %z) {
+; GFX9-LABEL: test_3xdouble_add_mul_rhs:
+; GFX9:       ; %bb.0: ; %.entry
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    v_mul_f64 v[0:1], v[0:1], v[6:7]
+; GFX9-NEXT:    v_mul_f64 v[2:3], v[2:3], v[8:9]
+; GFX9-NEXT:    v_mul_f64 v[4:5], v[4:5], v[10:11]
+; GFX9-NEXT:    v_add_f64 v[0:1], v[12:13], v[0:1]
+; GFX9-NEXT:    v_add_f64 v[2:3], v[14:15], v[2:3]
+; GFX9-NEXT:    v_add_f64 v[4:5], v[16:17], v[4:5]
+; GFX9-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX9-CONTRACT-LABEL: test_3xdouble_add_mul_rhs:
+; GFX9-CONTRACT:       ; %bb.0: ; %.entry
+; GFX9-CONTRACT-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-CONTRACT-NEXT:    v_fma_f64 v[0:1], v[0:1], v[6:7], v[12:13]
+; GFX9-CONTRACT-NEXT:    v_fma_f64 v[2:3], v[2:3], v[8:9], v[14:15]
+; GFX9-CONTRACT-NEXT:    v_fma_f64 v[4:5], v[4:5], v[10:11], v[16:17]
+; GFX9-CONTRACT-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX9-DENORM-LABEL: test_3xdouble_add_mul_rhs:
+; GFX9-DENORM:       ; %bb.0: ; %.entry
+; GFX9-DENORM-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-DENORM-NEXT:    v_mul_f64 v[0:1], v[0:1], v[6:7]
+; GFX9-DENORM-NEXT:    v_mul_f64 v[2:3], v[2:3], v[8:9]
+; GFX9-DENORM-NEXT:    v_mul_f64 v[4:5], v[4:5], v[10:11]
+; GFX9-DENORM-NEXT:    v_add_f64 v[0:1], v[12:13], v[0:1]
+; GFX9-DENORM-NEXT:    v_add_f64 v[2:3], v[14:15], v[2:3]
+; GFX9-DENORM-NEXT:    v_add_f64 v[4:5], v[16:17], v[4:5]
+; GFX9-DENORM-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX9-UNSAFE-LABEL: test_3xdouble_add_mul_rhs:
+; GFX9-UNSAFE:       ; %bb.0: ; %.entry
+; GFX9-UNSAFE-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-UNSAFE-NEXT:    v_fma_f64 v[0:1], v[0:1], v[6:7], v[12:13]
+; GFX9-UNSAFE-NEXT:    v_fma_f64 v[2:3], v[2:3], v[8:9], v[14:15]
+; GFX9-UNSAFE-NEXT:    v_fma_f64 v[4:5], v[4:5], v[10:11], v[16:17]
+; GFX9-UNSAFE-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX10-LABEL: test_3xdouble_add_mul_rhs:
+; GFX10:       ; %bb.0: ; %.entry
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    v_mul_f64 v[0:1], v[0:1], v[6:7]
+; GFX10-NEXT:    v_mul_f64 v[2:3], v[2:3], v[8:9]
+; GFX10-NEXT:    v_mul_f64 v[4:5], v[4:5], v[10:11]
+; GFX10-NEXT:    v_add_f64 v[0:1], v[12:13], v[0:1]
+; GFX10-NEXT:    v_add_f64 v[2:3], v[14:15], v[2:3]
+; GFX10-NEXT:    v_add_f64 v[4:5], v[16:17], v[4:5]
+; GFX10-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX10-CONTRACT-LABEL: test_3xdouble_add_mul_rhs:
+; GFX10-CONTRACT:       ; %bb.0: ; %.entry
+; GFX10-CONTRACT-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-CONTRACT-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-CONTRACT-NEXT:    v_fma_f64 v[0:1], v[0:1], v[6:7], v[12:13]
+; GFX10-CONTRACT-NEXT:    v_fma_f64 v[2:3], v[2:3], v[8:9], v[14:15]
+; GFX10-CONTRACT-NEXT:    v_fma_f64 v[4:5], v[4:5], v[10:11], v[16:17]
+; GFX10-CONTRACT-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX10-DENORM-LABEL: test_3xdouble_add_mul_rhs:
+; GFX10-DENORM:       ; %bb.0: ; %.entry
+; GFX10-DENORM-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-DENORM-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-DENORM-NEXT:    v_mul_f64 v[0:1], v[0:1], v[6:7]
+; GFX10-DENORM-NEXT:    v_mul_f64 v[2:3], v[2:3], v[8:9]
+; GFX10-DENORM-NEXT:    v_mul_f64 v[4:5], v[4:5], v[10:11]
+; GFX10-DENORM-NEXT:    v_add_f64 v[0:1], v[12:13], v[0:1]
+; GFX10-DENORM-NEXT:    v_add_f64 v[2:3], v[14:15], v[2:3]
+; GFX10-DENORM-NEXT:    v_add_f64 v[4:5], v[16:17], v[4:5]
+; GFX10-DENORM-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX10-UNSAFE-LABEL: test_3xdouble_add_mul_rhs:
+; GFX10-UNSAFE:       ; %bb.0: ; %.entry
+; GFX10-UNSAFE-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-UNSAFE-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-UNSAFE-NEXT:    v_fma_f64 v[0:1], v[0:1], v[6:7], v[12:13]
+; GFX10-UNSAFE-NEXT:    v_fma_f64 v[2:3], v[2:3], v[8:9], v[14:15]
+; GFX10-UNSAFE-NEXT:    v_fma_f64 v[4:5], v[4:5], v[10:11], v[16:17]
+; GFX10-UNSAFE-NEXT:    s_setpc_b64 s[30:31]
+.entry:
+  %a = fmul <3 x double> %x, %y
+  %b = fadd <3 x double> %z, %a
+  ret <3 x double> %b
+}