GlobalISel: Implement fewerElementsVector for basic FP ops

llvm-svn: 351866

8 files changed, 2211 insertions, 27 deletions

diff --git a/llvm/lib/CodeGen/GlobalISel/LegalizerHelper.cpp b/llvm/lib/CodeGen/GlobalISel/LegalizerHelper.cpp
index d2c507d4e18..ed56ca9e081 100644
--- a/llvm/lib/CodeGen/GlobalISel/LegalizerHelper.cpp
+++ b/llvm/lib/CodeGen/GlobalISel/LegalizerHelper.cpp
@@ -1206,7 +1206,8 @@ LegalizerHelper::fewerElementsVector(MachineInstr &MI, unsigned TypeIdx,
     return UnableToLegalize;
 
   MIRBuilder.setInstr(MI);
-  switch (MI.getOpcode()) {
+  unsigned Opc = MI.getOpcode();
+  switch (Opc) {
   default:
     return UnableToLegalize;
   case TargetOpcode::G_IMPLICIT_DEF: {
@@ -1235,9 +1236,18 @@ LegalizerHelper::fewerElementsVector(MachineInstr &MI, unsigned TypeIdx,
     MI.eraseFromParent();
     return Legalized;
   }
-  case TargetOpcode::G_ADD: {
+  case TargetOpcode::G_ADD:
+  case TargetOpcode::G_FADD:
+  case TargetOpcode::G_FMUL:
+  case TargetOpcode::G_FSUB:
+  case TargetOpcode::G_FNEG:
+  case TargetOpcode::G_FABS:
+  case TargetOpcode::G_FDIV:
+  case TargetOpcode::G_FREM:
+  case TargetOpcode::G_FMA: {
     unsigned NarrowSize = NarrowTy.getSizeInBits();
     unsigned DstReg = MI.getOperand(0).getReg();
+    unsigned Flags = MI.getFlags();
     unsigned Size = MRI.getType(DstReg).getSizeInBits();
     int NumParts = Size / NarrowSize;
     // FIXME: Don't know how to handle the situation where the small vectors
@@ -1245,17 +1255,37 @@ LegalizerHelper::fewerElementsVector(MachineInstr &MI, unsigned TypeIdx,
     if (Size % NarrowSize != 0)
       return UnableToLegalize;
 
-    SmallVector<unsigned, 2> Src1Regs, Src2Regs, DstRegs;
-    extractParts(MI.getOperand(1).getReg(), NarrowTy, NumParts, Src1Regs);
-    extractParts(MI.getOperand(2).getReg(), NarrowTy, NumParts, Src2Regs);
+    unsigned NumOps = MI.getNumOperands() - 1;
+    SmallVector<unsigned, 2> DstRegs, Src0Regs, Src1Regs, Src2Regs;
+
+    extractParts(MI.getOperand(1).getReg(), NarrowTy, NumParts, Src0Regs);
+
+    if (NumOps >= 2)
+      extractParts(MI.getOperand(2).getReg(), NarrowTy, NumParts, Src1Regs);
+
+    if (NumOps >= 3)
+      extractParts(MI.getOperand(3).getReg(), NarrowTy, NumParts, Src2Regs);
 
     for (int i = 0; i < NumParts; ++i) {
       unsigned DstReg = MRI.createGenericVirtualRegister(NarrowTy);
-      MIRBuilder.buildAdd(DstReg, Src1Regs[i], Src2Regs[i]);
+
+      if (NumOps == 1)
+        MIRBuilder.buildInstr(Opc, {DstReg}, {Src0Regs[i]}, Flags);
+      else if (NumOps == 2) {
+        MIRBuilder.buildInstr(Opc, {DstReg}, {Src0Regs[i], Src1Regs[i]}, Flags);
+      } else if (NumOps == 3) {
+        MIRBuilder.buildInstr(Opc, {DstReg},
+                              {Src0Regs[i], Src1Regs[i], Src2Regs[i]}, Flags);
+      }
+
       DstRegs.push_back(DstReg);
     }
 
-    MIRBuilder.buildConcatVectors(DstReg, DstRegs);
+    if (NarrowTy.isVector())
+      MIRBuilder.buildConcatVectors(DstReg, DstRegs);
+    else
+      MIRBuilder.buildBuildVector(DstReg, DstRegs);
+
     MI.eraseFromParent();
     return Legalized;
   }
diff --git a/llvm/lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp b/llvm/lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp
index 81fd43e73a4..a57da6493ce 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp
@@ -28,6 +28,11 @@ AMDGPULegalizerInfo::AMDGPULegalizerInfo(const GCNSubtarget &ST,
                                          const GCNTargetMachine &TM) {
   using namespace TargetOpcode;
 
+  auto scalarize = [=](const LegalityQuery &Query, unsigned TypeIdx) {
+    const LLT &Ty = Query.Types[TypeIdx];
+    return std::make_pair(TypeIdx, Ty.getElementType());
+  };
+
   auto GetAddrSpacePtr = [&TM](unsigned AS) {
     return LLT::pointer(AS, TM.getPointerSizeInBits(AS));
   };
@@ -136,10 +141,12 @@ AMDGPULegalizerInfo::AMDGPULegalizerInfo(const GCNSubtarget &ST,
 
   setAction({G_FRAME_INDEX, PrivatePtr}, Legal);
 
-  getActionDefinitionsBuilder(
-    { G_FADD, G_FMUL, G_FNEG, G_FABS, G_FMA})
-    .legalFor({S32, S64})
-    .clampScalar(0, S32, S64);
+  getActionDefinitionsBuilder({G_FADD, G_FMUL, G_FNEG, G_FABS, G_FMA})
+      .legalFor({S32, S64})
+      .fewerElementsIf(
+          [=](const LegalityQuery &Query) { return Query.Types[0].isVector(); },
+          [=](const LegalityQuery &Query) { return scalarize(Query, 0); })
+      .clampScalar(0, S32, S64);
 
   getActionDefinitionsBuilder(G_FPTRUNC)
     .legalFor({{S32, S64}, {S16, S32}});
@@ -149,11 +156,14 @@ AMDGPULegalizerInfo::AMDGPULegalizerInfo(const GCNSubtarget &ST,
     .lowerFor({{S64, S16}}); // FIXME: Implement
 
   getActionDefinitionsBuilder(G_FSUB)
-    // Use actual fsub instruction
-    .legalFor({S32})
-    // Must use fadd + fneg
-    .lowerFor({S64, S16})
-    .clampScalar(0, S32, S64);
+      // Use actual fsub instruction
+      .legalFor({S32})
+      // Must use fadd + fneg
+      .lowerFor({S64, S16, V2S16})
+      .fewerElementsIf(
+          [=](const LegalityQuery &Query) { return Query.Types[0].isVector(); },
+          [=](const LegalityQuery &Query) { return scalarize(Query, 0); })
+      .clampScalar(0, S32, S64);
 
   setAction({G_FCMP, S1}, Legal);
   setAction({G_FCMP, 1, S32}, Legal);
@@ -295,11 +305,15 @@ AMDGPULegalizerInfo::AMDGPULegalizerInfo(const GCNSubtarget &ST,
       });
 
   getActionDefinitionsBuilder(G_BUILD_VECTOR)
-    .legalForCartesianProduct(AllS32Vectors, {S32})
-    .legalForCartesianProduct(AllS64Vectors, {S64})
-    .clampNumElements(0, V16S32, V16S32)
-    .clampNumElements(0, V2S64, V8S64)
-    .minScalarSameAs(1, 0);
+      .legalForCartesianProduct(AllS32Vectors, {S32})
+      .legalForCartesianProduct(AllS64Vectors, {S64})
+      .clampNumElements(0, V16S32, V16S32)
+      .clampNumElements(0, V2S64, V8S64)
+      .minScalarSameAs(1, 0)
+      // FIXME: Sort of a hack to make progress on other legalizations.
+      .legalIf([=](const LegalityQuery &Query) {
+        return Query.Types[0].getScalarSizeInBits() < 32;
+      });
 
   // TODO: Support any combination of v2s32
   getActionDefinitionsBuilder(G_CONCAT_VECTORS)
@@ -328,12 +342,6 @@ AMDGPULegalizerInfo::AMDGPULegalizerInfo(const GCNSubtarget &ST,
       return false;
     };
 
-    auto scalarize =
-      [=](const LegalityQuery &Query, unsigned TypeIdx) {
-      const LLT &Ty = Query.Types[TypeIdx];
-      return std::make_pair(TypeIdx, Ty.getElementType());
-    };
-
     getActionDefinitionsBuilder(Op)
       // Break up vectors with weird elements into scalars
       .fewerElementsIf(
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-fabs.mir b/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-fabs.mir
index f2e397c4e62..87bd8c13859 100644
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-fabs.mir
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-fabs.mir
@@ -70,3 +70,260 @@ body: |
     %3:_(s32) = G_ANYEXT %2
     $vgpr0 = COPY %3
 ...
+
+---
+name: test_fabs_v2s32
+body: |
+  bb.0:
+    liveins: $vgpr0_vgpr1
+
+    ; SI-LABEL: name: test_fabs_v2s32
+    ; SI: [[COPY:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr0_vgpr1
+    ; SI: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](<2 x s32>)
+    ; SI: [[FABS:%[0-9]+]]:_(s32) = G_FABS [[UV]]
+    ; SI: [[FABS1:%[0-9]+]]:_(s32) = G_FABS [[UV1]]
+    ; SI: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[FABS]](s32), [[FABS1]](s32)
+    ; SI: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<2 x s32>)
+    ; VI-LABEL: name: test_fabs_v2s32
+    ; VI: [[COPY:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr0_vgpr1
+    ; VI: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](<2 x s32>)
+    ; VI: [[FABS:%[0-9]+]]:_(s32) = G_FABS [[UV]]
+    ; VI: [[FABS1:%[0-9]+]]:_(s32) = G_FABS [[UV1]]
+    ; VI: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[FABS]](s32), [[FABS1]](s32)
+    ; VI: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<2 x s32>)
+    ; GFX9-LABEL: name: test_fabs_v2s32
+    ; GFX9: [[COPY:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr0_vgpr1
+    ; GFX9: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](<2 x s32>)
+    ; GFX9: [[FABS:%[0-9]+]]:_(s32) = G_FABS [[UV]]
+    ; GFX9: [[FABS1:%[0-9]+]]:_(s32) = G_FABS [[UV1]]
+    ; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[FABS]](s32), [[FABS1]](s32)
+    ; GFX9: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<2 x s32>)
+    %0:_(<2 x s32>) = COPY $vgpr0_vgpr1
+    %1:_(<2 x s32>) = G_FABS %0
+    $vgpr0_vgpr1 = COPY %1
+...
+
+---
+name: test_fabs_v3s32
+body: |
+  bb.0:
+    liveins: $vgpr0_vgpr1_vgpr2
+
+    ; SI-LABEL: name: test_fabs_v3s32
+    ; SI: [[COPY:%[0-9]+]]:_(<3 x s32>) = COPY $vgpr0_vgpr1_vgpr2
+    ; SI: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](<3 x s32>)
+    ; SI: [[FABS:%[0-9]+]]:_(s32) = G_FABS [[UV]]
+    ; SI: [[FABS1:%[0-9]+]]:_(s32) = G_FABS [[UV1]]
+    ; SI: [[FABS2:%[0-9]+]]:_(s32) = G_FABS [[UV2]]
+    ; SI: [[BUILD_VECTOR:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[FABS]](s32), [[FABS1]](s32), [[FABS2]](s32)
+    ; SI: $vgpr0_vgpr1_vgpr2 = COPY [[BUILD_VECTOR]](<3 x s32>)
+    ; VI-LABEL: name: test_fabs_v3s32
+    ; VI: [[COPY:%[0-9]+]]:_(<3 x s32>) = COPY $vgpr0_vgpr1_vgpr2
+    ; VI: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](<3 x s32>)
+    ; VI: [[FABS:%[0-9]+]]:_(s32) = G_FABS [[UV]]
+    ; VI: [[FABS1:%[0-9]+]]:_(s32) = G_FABS [[UV1]]
+    ; VI: [[FABS2:%[0-9]+]]:_(s32) = G_FABS [[UV2]]
+    ; VI: [[BUILD_VECTOR:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[FABS]](s32), [[FABS1]](s32), [[FABS2]](s32)
+    ; VI: $vgpr0_vgpr1_vgpr2 = COPY [[BUILD_VECTOR]](<3 x s32>)
+    ; GFX9-LABEL: name: test_fabs_v3s32
+    ; GFX9: [[COPY:%[0-9]+]]:_(<3 x s32>) = COPY $vgpr0_vgpr1_vgpr2
+    ; GFX9: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](<3 x s32>)
+    ; GFX9: [[FABS:%[0-9]+]]:_(s32) = G_FABS [[UV]]
+    ; GFX9: [[FABS1:%[0-9]+]]:_(s32) = G_FABS [[UV1]]
+    ; GFX9: [[FABS2:%[0-9]+]]:_(s32) = G_FABS [[UV2]]
+    ; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[FABS]](s32), [[FABS1]](s32), [[FABS2]](s32)
+    ; GFX9: $vgpr0_vgpr1_vgpr2 = COPY [[BUILD_VECTOR]](<3 x s32>)
+    %0:_(<3 x s32>) = COPY $vgpr0_vgpr1_vgpr2
+    %1:_(<3 x  s32>) = G_FABS %0
+    $vgpr0_vgpr1_vgpr2 = COPY %1
+...
+
+---
+name: test_fabs_v2s64
+body: |
+  bb.0:
+    liveins: $vgpr0_vgpr1_vgpr2_vgpr3
+
+    ; SI-LABEL: name: test_fabs_v2s64
+    ; SI: [[COPY:%[0-9]+]]:_(<2 x s64>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3
+    ; SI: [[UV:%[0-9]+]]:_(s64), [[UV1:%[0-9]+]]:_(s64) = G_UNMERGE_VALUES [[COPY]](<2 x s64>)
+    ; SI: [[FABS:%[0-9]+]]:_(s64) = G_FABS [[UV]]
+    ; SI: [[FABS1:%[0-9]+]]:_(s64) = G_FABS [[UV1]]
+    ; SI: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s64>) = G_BUILD_VECTOR [[FABS]](s64), [[FABS1]](s64)
+    ; SI: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[BUILD_VECTOR]](<2 x s64>)
+    ; VI-LABEL: name: test_fabs_v2s64
+    ; VI: [[COPY:%[0-9]+]]:_(<2 x s64>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3
+    ; VI: [[UV:%[0-9]+]]:_(s64), [[UV1:%[0-9]+]]:_(s64) = G_UNMERGE_VALUES [[COPY]](<2 x s64>)
+    ; VI: [[FABS:%[0-9]+]]:_(s64) = G_FABS [[UV]]
+    ; VI: [[FABS1:%[0-9]+]]:_(s64) = G_FABS [[UV1]]
+    ; VI: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s64>) = G_BUILD_VECTOR [[FABS]](s64), [[FABS1]](s64)
+    ; VI: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[BUILD_VECTOR]](<2 x s64>)
+    ; GFX9-LABEL: name: test_fabs_v2s64
+    ; GFX9: [[COPY:%[0-9]+]]:_(<2 x s64>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3
+    ; GFX9: [[UV:%[0-9]+]]:_(s64), [[UV1:%[0-9]+]]:_(s64) = G_UNMERGE_VALUES [[COPY]](<2 x s64>)
+    ; GFX9: [[FABS:%[0-9]+]]:_(s64) = G_FABS [[UV]]
+    ; GFX9: [[FABS1:%[0-9]+]]:_(s64) = G_FABS [[UV1]]
+    ; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s64>) = G_BUILD_VECTOR [[FABS]](s64), [[FABS1]](s64)
+    ; GFX9: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[BUILD_VECTOR]](<2 x s64>)
+    %0:_(<2 x s64>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3
+    %1:_(<2 x s64>) = G_FABS %0
+    $vgpr0_vgpr1_vgpr2_vgpr3 = COPY %1
+...
+
+---
+name: test_fabs_v2s16
+body: |
+  bb.0:
+    liveins: $vgpr0
+
+    ; SI-LABEL: name: test_fabs_v2s16
+    ; SI: [[COPY:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0
+    ; SI: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY]](<2 x s16>)
+    ; SI: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[UV]](s16)
+    ; SI: [[FABS:%[0-9]+]]:_(s32) = G_FABS [[FPEXT]]
+    ; SI: [[FPTRUNC:%[0-9]+]]:_(s16) = G_FPTRUNC [[FABS]](s32)
+    ; SI: [[FPEXT1:%[0-9]+]]:_(s32) = G_FPEXT [[UV1]](s16)
+    ; SI: [[FABS1:%[0-9]+]]:_(s32) = G_FABS [[FPEXT1]]
+    ; SI: [[FPTRUNC1:%[0-9]+]]:_(s16) = G_FPTRUNC [[FABS1]](s32)
+    ; SI: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR [[FPTRUNC]](s16), [[FPTRUNC1]](s16)
+    ; SI: $vgpr0 = COPY [[BUILD_VECTOR]](<2 x s16>)
+    ; VI-LABEL: name: test_fabs_v2s16
+    ; VI: [[COPY:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0
+    ; VI: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY]](<2 x s16>)
+    ; VI: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[UV]](s16)
+    ; VI: [[FABS:%[0-9]+]]:_(s32) = G_FABS [[FPEXT]]
+    ; VI: [[FPTRUNC:%[0-9]+]]:_(s16) = G_FPTRUNC [[FABS]](s32)
+    ; VI: [[FPEXT1:%[0-9]+]]:_(s32) = G_FPEXT [[UV1]](s16)
+    ; VI: [[FABS1:%[0-9]+]]:_(s32) = G_FABS [[FPEXT1]]
+    ; VI: [[FPTRUNC1:%[0-9]+]]:_(s16) = G_FPTRUNC [[FABS1]](s32)
+    ; VI: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR [[FPTRUNC]](s16), [[FPTRUNC1]](s16)
+    ; VI: $vgpr0 = COPY [[BUILD_VECTOR]](<2 x s16>)
+    ; GFX9-LABEL: name: test_fabs_v2s16
+    ; GFX9: [[COPY:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0
+    ; GFX9: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY]](<2 x s16>)
+    ; GFX9: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[UV]](s16)
+    ; GFX9: [[FABS:%[0-9]+]]:_(s32) = G_FABS [[FPEXT]]
+    ; GFX9: [[FPTRUNC:%[0-9]+]]:_(s16) = G_FPTRUNC [[FABS]](s32)
+    ; GFX9: [[FPEXT1:%[0-9]+]]:_(s32) = G_FPEXT [[UV1]](s16)
+    ; GFX9: [[FABS1:%[0-9]+]]:_(s32) = G_FABS [[FPEXT1]]
+    ; GFX9: [[FPTRUNC1:%[0-9]+]]:_(s16) = G_FPTRUNC [[FABS1]](s32)
+    ; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR [[FPTRUNC]](s16), [[FPTRUNC1]](s16)
+    ; GFX9: $vgpr0 = COPY [[BUILD_VECTOR]](<2 x s16>)
+    %0:_(<2 x s16>) = COPY $vgpr0
+    %1:_(<2 x s16>) = G_FABS %0
+    $vgpr0 = COPY %1
+...
+
+---
+name: test_fabs_v3s16
+body: |
+  bb.0:
+
+    ; SI-LABEL: name: test_fabs_v3s16
+    ; SI: [[DEF:%[0-9]+]]:_(<3 x s16>) = G_IMPLICIT_DEF
+    ; SI: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[DEF]](<3 x s16>)
+    ; SI: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[UV]](s16)
+    ; SI: [[FABS:%[0-9]+]]:_(s32) = G_FABS [[FPEXT]]
+    ; SI: [[FPTRUNC:%[0-9]+]]:_(s16) = G_FPTRUNC [[FABS]](s32)
+    ; SI: [[FPEXT1:%[0-9]+]]:_(s32) = G_FPEXT [[UV1]](s16)
+    ; SI: [[FABS1:%[0-9]+]]:_(s32) = G_FABS [[FPEXT1]]
+    ; SI: [[FPTRUNC1:%[0-9]+]]:_(s16) = G_FPTRUNC [[FABS1]](s32)
+    ; SI: [[FPEXT2:%[0-9]+]]:_(s32) = G_FPEXT [[UV2]](s16)
+    ; SI: [[FABS2:%[0-9]+]]:_(s32) = G_FABS [[FPEXT2]]
+    ; SI: [[FPTRUNC2:%[0-9]+]]:_(s16) = G_FPTRUNC [[FABS2]](s32)
+    ; SI: [[BUILD_VECTOR:%[0-9]+]]:_(<3 x s16>) = G_BUILD_VECTOR [[FPTRUNC]](s16), [[FPTRUNC1]](s16), [[FPTRUNC2]](s16)
+    ; SI: S_NOP 0, implicit [[BUILD_VECTOR]](<3 x s16>)
+    ; VI-LABEL: name: test_fabs_v3s16
+    ; VI: [[DEF:%[0-9]+]]:_(<3 x s16>) = G_IMPLICIT_DEF
+    ; VI: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[DEF]](<3 x s16>)
+    ; VI: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[UV]](s16)
+    ; VI: [[FABS:%[0-9]+]]:_(s32) = G_FABS [[FPEXT]]
+    ; VI: [[FPTRUNC:%[0-9]+]]:_(s16) = G_FPTRUNC [[FABS]](s32)
+    ; VI: [[FPEXT1:%[0-9]+]]:_(s32) = G_FPEXT [[UV1]](s16)
+    ; VI: [[FABS1:%[0-9]+]]:_(s32) = G_FABS [[FPEXT1]]
+    ; VI: [[FPTRUNC1:%[0-9]+]]:_(s16) = G_FPTRUNC [[FABS1]](s32)
+    ; VI: [[FPEXT2:%[0-9]+]]:_(s32) = G_FPEXT [[UV2]](s16)
+    ; VI: [[FABS2:%[0-9]+]]:_(s32) = G_FABS [[FPEXT2]]
+    ; VI: [[FPTRUNC2:%[0-9]+]]:_(s16) = G_FPTRUNC [[FABS2]](s32)
+    ; VI: [[BUILD_VECTOR:%[0-9]+]]:_(<3 x s16>) = G_BUILD_VECTOR [[FPTRUNC]](s16), [[FPTRUNC1]](s16), [[FPTRUNC2]](s16)
+    ; VI: S_NOP 0, implicit [[BUILD_VECTOR]](<3 x s16>)
+    ; GFX9-LABEL: name: test_fabs_v3s16
+    ; GFX9: [[DEF:%[0-9]+]]:_(<3 x s16>) = G_IMPLICIT_DEF
+    ; GFX9: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[DEF]](<3 x s16>)
+    ; GFX9: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[UV]](s16)
+    ; GFX9: [[FABS:%[0-9]+]]:_(s32) = G_FABS [[FPEXT]]
+    ; GFX9: [[FPTRUNC:%[0-9]+]]:_(s16) = G_FPTRUNC [[FABS]](s32)
+    ; GFX9: [[FPEXT1:%[0-9]+]]:_(s32) = G_FPEXT [[UV1]](s16)
+    ; GFX9: [[FABS1:%[0-9]+]]:_(s32) = G_FABS [[FPEXT1]]
+    ; GFX9: [[FPTRUNC1:%[0-9]+]]:_(s16) = G_FPTRUNC [[FABS1]](s32)
+    ; GFX9: [[FPEXT2:%[0-9]+]]:_(s32) = G_FPEXT [[UV2]](s16)
+    ; GFX9: [[FABS2:%[0-9]+]]:_(s32) = G_FABS [[FPEXT2]]
+    ; GFX9: [[FPTRUNC2:%[0-9]+]]:_(s16) = G_FPTRUNC [[FABS2]](s32)
+    ; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<3 x s16>) = G_BUILD_VECTOR [[FPTRUNC]](s16), [[FPTRUNC1]](s16), [[FPTRUNC2]](s16)
+    ; GFX9: S_NOP 0, implicit [[BUILD_VECTOR]](<3 x s16>)
+    %0:_(<3 x s16>) = G_IMPLICIT_DEF
+    %1:_(<3 x s16>) = G_FABS %0
+    S_NOP 0, implicit %1
+...
+
+---
+name: test_fabs_v4s16
+body: |
+  bb.0:
+    liveins: $vgpr0_vgpr1
+
+    ; SI-LABEL: name: test_fabs_v4s16
+    ; SI: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1
+    ; SI: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16), [[UV3:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY]](<4 x s16>)
+    ; SI: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[UV]](s16)
+    ; SI: [[FABS:%[0-9]+]]:_(s32) = G_FABS [[FPEXT]]
+    ; SI: [[FPTRUNC:%[0-9]+]]:_(s16) = G_FPTRUNC [[FABS]](s32)
+    ; SI: [[FPEXT1:%[0-9]+]]:_(s32) = G_FPEXT [[UV1]](s16)
+    ; SI: [[FABS1:%[0-9]+]]:_(s32) = G_FABS [[FPEXT1]]
+    ; SI: [[FPTRUNC1:%[0-9]+]]:_(s16) = G_FPTRUNC [[FABS1]](s32)
+    ; SI: [[FPEXT2:%[0-9]+]]:_(s32) = G_FPEXT [[UV2]](s16)
+    ; SI: [[FABS2:%[0-9]+]]:_(s32) = G_FABS [[FPEXT2]]
+    ; SI: [[FPTRUNC2:%[0-9]+]]:_(s16) = G_FPTRUNC [[FABS2]](s32)
+    ; SI: [[FPEXT3:%[0-9]+]]:_(s32) = G_FPEXT [[UV3]](s16)
+    ; SI: [[FABS3:%[0-9]+]]:_(s32) = G_FABS [[FPEXT3]]
+    ; SI: [[FPTRUNC3:%[0-9]+]]:_(s16) = G_FPTRUNC [[FABS3]](s32)
+    ; SI: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s16>) = G_BUILD_VECTOR [[FPTRUNC]](s16), [[FPTRUNC1]](s16), [[FPTRUNC2]](s16), [[FPTRUNC3]](s16)
+    ; SI: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<4 x s16>)
+    ; VI-LABEL: name: test_fabs_v4s16
+    ; VI: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1
+    ; VI: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16), [[UV3:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY]](<4 x s16>)
+    ; VI: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[UV]](s16)
+    ; VI: [[FABS:%[0-9]+]]:_(s32) = G_FABS [[FPEXT]]
+    ; VI: [[FPTRUNC:%[0-9]+]]:_(s16) = G_FPTRUNC [[FABS]](s32)
+    ; VI: [[FPEXT1:%[0-9]+]]:_(s32) = G_FPEXT [[UV1]](s16)
+    ; VI: [[FABS1:%[0-9]+]]:_(s32) = G_FABS [[FPEXT1]]
+    ; VI: [[FPTRUNC1:%[0-9]+]]:_(s16) = G_FPTRUNC [[FABS1]](s32)
+    ; VI: [[FPEXT2:%[0-9]+]]:_(s32) = G_FPEXT [[UV2]](s16)
+    ; VI: [[FABS2:%[0-9]+]]:_(s32) = G_FABS [[FPEXT2]]
+    ; VI: [[FPTRUNC2:%[0-9]+]]:_(s16) = G_FPTRUNC [[FABS2]](s32)
+    ; VI: [[FPEXT3:%[0-9]+]]:_(s32) = G_FPEXT [[UV3]](s16)
+    ; VI: [[FABS3:%[0-9]+]]:_(s32) = G_FABS [[FPEXT3]]
+    ; VI: [[FPTRUNC3:%[0-9]+]]:_(s16) = G_FPTRUNC [[FABS3]](s32)
+    ; VI: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s16>) = G_BUILD_VECTOR [[FPTRUNC]](s16), [[FPTRUNC1]](s16), [[FPTRUNC2]](s16), [[FPTRUNC3]](s16)
+    ; VI: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<4 x s16>)
+    ; GFX9-LABEL: name: test_fabs_v4s16
+    ; GFX9: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1
+    ; GFX9: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16), [[UV3:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY]](<4 x s16>)
+    ; GFX9: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[UV]](s16)
+    ; GFX9: [[FABS:%[0-9]+]]:_(s32) = G_FABS [[FPEXT]]
+    ; GFX9: [[FPTRUNC:%[0-9]+]]:_(s16) = G_FPTRUNC [[FABS]](s32)
+    ; GFX9: [[FPEXT1:%[0-9]+]]:_(s32) = G_FPEXT [[UV1]](s16)
+    ; GFX9: [[FABS1:%[0-9]+]]:_(s32) = G_FABS [[FPEXT1]]
+    ; GFX9: [[FPTRUNC1:%[0-9]+]]:_(s16) = G_FPTRUNC [[FABS1]](s32)
+    ; GFX9: [[FPEXT2:%[0-9]+]]:_(s32) = G_FPEXT [[UV2]](s16)
+    ; GFX9: [[FABS2:%[0-9]+]]:_(s32) = G_FABS [[FPEXT2]]
+    ; GFX9: [[FPTRUNC2:%[0-9]+]]:_(s16) = G_FPTRUNC [[FABS2]](s32)
+    ; GFX9: [[FPEXT3:%[0-9]+]]:_(s32) = G_FPEXT [[UV3]](s16)
+    ; GFX9: [[FABS3:%[0-9]+]]:_(s32) = G_FABS [[FPEXT3]]
+    ; GFX9: [[FPTRUNC3:%[0-9]+]]:_(s16) = G_FPTRUNC [[FABS3]](s32)
+    ; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s16>) = G_BUILD_VECTOR [[FPTRUNC]](s16), [[FPTRUNC1]](s16), [[FPTRUNC2]](s16), [[FPTRUNC3]](s16)
+    ; GFX9: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<4 x s16>)
+    %0:_(<4 x s16>) = COPY $vgpr0_vgpr1
+    %1:_(<4 x s16>) = G_FABS %0
+    $vgpr0_vgpr1 = COPY %1
+...
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-fadd.mir b/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-fadd.mir
index efa450bd8f7..d4af78fc9cb 100644
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-fadd.mir
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-fadd.mir
@@ -105,3 +105,369 @@ body: |
     %5:_(s32) = G_ANYEXT %4
     $vgpr0 = COPY %5
 ...
+
+---
+name: test_fadd_v2s32
+body: |
+  bb.0.entry:
+    liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
+
+    ; SI-LABEL: name: test_fadd_v2s32
+    ; SI: [[COPY:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr0_vgpr1
+    ; SI: [[COPY1:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr2_vgpr3
+    ; SI: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](<2 x s32>)
+    ; SI: [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY1]](<2 x s32>)
+    ; SI: [[FADD:%[0-9]+]]:_(s32) = G_FADD [[UV]], [[UV2]]
+    ; SI: [[FADD1:%[0-9]+]]:_(s32) = G_FADD [[UV1]], [[UV3]]
+    ; SI: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[FADD]](s32), [[FADD1]](s32)
+    ; SI: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<2 x s32>)
+    ; VI-LABEL: name: test_fadd_v2s32
+    ; VI: [[COPY:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr0_vgpr1
+    ; VI: [[COPY1:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr2_vgpr3
+    ; VI: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](<2 x s32>)
+    ; VI: [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY1]](<2 x s32>)
+    ; VI: [[FADD:%[0-9]+]]:_(s32) = G_FADD [[UV]], [[UV2]]
+    ; VI: [[FADD1:%[0-9]+]]:_(s32) = G_FADD [[UV1]], [[UV3]]
+    ; VI: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[FADD]](s32), [[FADD1]](s32)
+    ; VI: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<2 x s32>)
+    ; GFX9-LABEL: name: test_fadd_v2s32
+    ; GFX9: [[COPY:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr0_vgpr1
+    ; GFX9: [[COPY1:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr2_vgpr3
+    ; GFX9: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](<2 x s32>)
+    ; GFX9: [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY1]](<2 x s32>)
+    ; GFX9: [[FADD:%[0-9]+]]:_(s32) = G_FADD [[UV]], [[UV2]]
+    ; GFX9: [[FADD1:%[0-9]+]]:_(s32) = G_FADD [[UV1]], [[UV3]]
+    ; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[FADD]](s32), [[FADD1]](s32)
+    ; GFX9: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<2 x s32>)
+    %0:_(<2 x s32>) = COPY $vgpr0_vgpr1
+    %1:_(<2 x s32>) = COPY $vgpr2_vgpr3
+    %2:_(<2 x s32>) = G_FADD %0, %1
+    $vgpr0_vgpr1 = COPY %2
+...
+
+---
+name: test_fadd_v2s32_flags
+body: |
+  bb.0.entry:
+    liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
+
+    ; SI-LABEL: name: test_fadd_v2s32_flags
+    ; SI: [[COPY:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr0_vgpr1
+    ; SI: [[COPY1:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr2_vgpr3
+    ; SI: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](<2 x s32>)
+    ; SI: [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY1]](<2 x s32>)
+    ; SI: %7:_(s32) = nnan G_FADD [[UV]], [[UV2]]
+    ; SI: %8:_(s32) = nnan G_FADD [[UV1]], [[UV3]]
+    ; SI: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR %7(s32), %8(s32)
+    ; SI: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<2 x s32>)
+    ; VI-LABEL: name: test_fadd_v2s32_flags
+    ; VI: [[COPY:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr0_vgpr1
+    ; VI: [[COPY1:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr2_vgpr3
+    ; VI: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](<2 x s32>)
+    ; VI: [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY1]](<2 x s32>)
+    ; VI: %7:_(s32) = nnan G_FADD [[UV]], [[UV2]]
+    ; VI: %8:_(s32) = nnan G_FADD [[UV1]], [[UV3]]
+    ; VI: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR %7(s32), %8(s32)
+    ; VI: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<2 x s32>)
+    ; GFX9-LABEL: name: test_fadd_v2s32_flags
+    ; GFX9: [[COPY:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr0_vgpr1
+    ; GFX9: [[COPY1:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr2_vgpr3
+    ; GFX9: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](<2 x s32>)
+    ; GFX9: [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY1]](<2 x s32>)
+    ; GFX9: %7:_(s32) = nnan G_FADD [[UV]], [[UV2]]
+    ; GFX9: %8:_(s32) = nnan G_FADD [[UV1]], [[UV3]]
+    ; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR %7(s32), %8(s32)
+    ; GFX9: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<2 x s32>)
+    %0:_(<2 x s32>) = COPY $vgpr0_vgpr1
+    %1:_(<2 x s32>) = COPY $vgpr2_vgpr3
+    %2:_(<2 x s32>) = nnan G_FADD %0, %1
+    $vgpr0_vgpr1 = COPY %2
+...
+
+---
+name: test_fadd_v3s32
+body: |
+  bb.0.entry:
+    liveins: $vgpr0_vgpr1_vgpr2, $vgpr3_vgpr4_vgpr5
+
+    ; SI-LABEL: name: test_fadd_v3s32
+    ; SI: [[COPY:%[0-9]+]]:_(<3 x s32>) = COPY $vgpr0_vgpr1_vgpr2
+    ; SI: [[COPY1:%[0-9]+]]:_(<3 x s32>) = COPY $vgpr3_vgpr4_vgpr5
+    ; SI: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](<3 x s32>)
+    ; SI: [[UV3:%[0-9]+]]:_(s32), [[UV4:%[0-9]+]]:_(s32), [[UV5:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY1]](<3 x s32>)
+    ; SI: [[FADD:%[0-9]+]]:_(s32) = G_FADD [[UV]], [[UV3]]
+    ; SI: [[FADD1:%[0-9]+]]:_(s32) = G_FADD [[UV1]], [[UV4]]
+    ; SI: [[FADD2:%[0-9]+]]:_(s32) = G_FADD [[UV2]], [[UV5]]
+    ; SI: [[BUILD_VECTOR:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[FADD]](s32), [[FADD1]](s32), [[FADD2]](s32)
+    ; SI: $vgpr0_vgpr1_vgpr2 = COPY [[BUILD_VECTOR]](<3 x s32>)
+    ; VI-LABEL: name: test_fadd_v3s32
+    ; VI: [[COPY:%[0-9]+]]:_(<3 x s32>) = COPY $vgpr0_vgpr1_vgpr2
+    ; VI: [[COPY1:%[0-9]+]]:_(<3 x s32>) = COPY $vgpr3_vgpr4_vgpr5
+    ; VI: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](<3 x s32>)
+    ; VI: [[UV3:%[0-9]+]]:_(s32), [[UV4:%[0-9]+]]:_(s32), [[UV5:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY1]](<3 x s32>)
+    ; VI: [[FADD:%[0-9]+]]:_(s32) = G_FADD [[UV]], [[UV3]]
+    ; VI: [[FADD1:%[0-9]+]]:_(s32) = G_FADD [[UV1]], [[UV4]]
+    ; VI: [[FADD2:%[0-9]+]]:_(s32) = G_FADD [[UV2]], [[UV5]]
+    ; VI: [[BUILD_VECTOR:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[FADD]](s32), [[FADD1]](s32), [[FADD2]](s32)
+    ; VI: $vgpr0_vgpr1_vgpr2 = COPY [[BUILD_VECTOR]](<3 x s32>)
+    ; GFX9-LABEL: name: test_fadd_v3s32
+    ; GFX9: [[COPY:%[0-9]+]]:_(<3 x s32>) = COPY $vgpr0_vgpr1_vgpr2
+    ; GFX9: [[COPY1:%[0-9]+]]:_(<3 x s32>) = COPY $vgpr3_vgpr4_vgpr5
+    ; GFX9: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](<3 x s32>)
+    ; GFX9: [[UV3:%[0-9]+]]:_(s32), [[UV4:%[0-9]+]]:_(s32), [[UV5:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY1]](<3 x s32>)
+    ; GFX9: [[FADD:%[0-9]+]]:_(s32) = G_FADD [[UV]], [[UV3]]
+    ; GFX9: [[FADD1:%[0-9]+]]:_(s32) = G_FADD [[UV1]], [[UV4]]
+    ; GFX9: [[FADD2:%[0-9]+]]:_(s32) = G_FADD [[UV2]], [[UV5]]
+    ; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[FADD]](s32), [[FADD1]](s32), [[FADD2]](s32)
+    ; GFX9: $vgpr0_vgpr1_vgpr2 = COPY [[BUILD_VECTOR]](<3 x s32>)
+    %0:_(<3 x s32>) = COPY $vgpr0_vgpr1_vgpr2
+    %1:_(<3 x s32>) = COPY $vgpr3_vgpr4_vgpr5
+    %2:_(<3 x s32>) = G_FADD %0, %1
+    $vgpr0_vgpr1_vgpr2 = COPY %2
+...
+
+---
+name: test_fadd_v2s64
+body: |
+  bb.0.entry:
+    liveins: $vgpr0_vgpr1_vgpr2_vgpr3, $vgpr4_vgpr5_vgpr6_vgpr7
+
+    ; SI-LABEL: name: test_fadd_v2s64
+    ; SI: [[COPY:%[0-9]+]]:_(<2 x s64>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3
+    ; SI: [[COPY1:%[0-9]+]]:_(<2 x s64>) = COPY $vgpr4_vgpr5_vgpr6_vgpr7
+    ; SI: [[UV:%[0-9]+]]:_(s64), [[UV1:%[0-9]+]]:_(s64) = G_UNMERGE_VALUES [[COPY]](<2 x s64>)
+    ; SI: [[UV2:%[0-9]+]]:_(s64), [[UV3:%[0-9]+]]:_(s64) = G_UNMERGE_VALUES [[COPY1]](<2 x s64>)
+    ; SI: [[FADD:%[0-9]+]]:_(s64) = G_FADD [[UV]], [[UV2]]
+    ; SI: [[FADD1:%[0-9]+]]:_(s64) = G_FADD [[UV1]], [[UV3]]
+    ; SI: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s64>) = G_BUILD_VECTOR [[FADD]](s64), [[FADD1]](s64)
+    ; SI: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[BUILD_VECTOR]](<2 x s64>)
+    ; VI-LABEL: name: test_fadd_v2s64
+    ; VI: [[COPY:%[0-9]+]]:_(<2 x s64>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3
+    ; VI: [[COPY1:%[0-9]+]]:_(<2 x s64>) = COPY $vgpr4_vgpr5_vgpr6_vgpr7
+    ; VI: [[UV:%[0-9]+]]:_(s64), [[UV1:%[0-9]+]]:_(s64) = G_UNMERGE_VALUES [[COPY]](<2 x s64>)
+    ; VI: [[UV2:%[0-9]+]]:_(s64), [[UV3:%[0-9]+]]:_(s64) = G_UNMERGE_VALUES [[COPY1]](<2 x s64>)
+    ; VI: [[FADD:%[0-9]+]]:_(s64) = G_FADD [[UV]], [[UV2]]
+    ; VI: [[FADD1:%[0-9]+]]:_(s64) = G_FADD [[UV1]], [[UV3]]
+    ; VI: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s64>) = G_BUILD_VECTOR [[FADD]](s64), [[FADD1]](s64)
+    ; VI: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[BUILD_VECTOR]](<2 x s64>)
+    ; GFX9-LABEL: name: test_fadd_v2s64
+    ; GFX9: [[COPY:%[0-9]+]]:_(<2 x s64>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3
+    ; GFX9: [[COPY1:%[0-9]+]]:_(<2 x s64>) = COPY $vgpr4_vgpr5_vgpr6_vgpr7
+    ; GFX9: [[UV:%[0-9]+]]:_(s64), [[UV1:%[0-9]+]]:_(s64) = G_UNMERGE_VALUES [[COPY]](<2 x s64>)
+    ; GFX9: [[UV2:%[0-9]+]]:_(s64), [[UV3:%[0-9]+]]:_(s64) = G_UNMERGE_VALUES [[COPY1]](<2 x s64>)
+    ; GFX9: [[FADD:%[0-9]+]]:_(s64) = G_FADD [[UV]], [[UV2]]
+    ; GFX9: [[FADD1:%[0-9]+]]:_(s64) = G_FADD [[UV1]], [[UV3]]
+    ; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s64>) = G_BUILD_VECTOR [[FADD]](s64), [[FADD1]](s64)
+    ; GFX9: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[BUILD_VECTOR]](<2 x s64>)
+    %0:_(<2 x s64>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3
+    %1:_(<2 x s64>) = COPY $vgpr4_vgpr5_vgpr6_vgpr7
+    %2:_(<2 x s64>) = G_FADD %0, %1
+    $vgpr0_vgpr1_vgpr2_vgpr3 = COPY %2
+...
+
+---
+name: test_fadd_v2s16
+body: |
+  bb.0.entry:
+    liveins: $vgpr0, $vgpr1
+
+    ; SI-LABEL: name: test_fadd_v2s16
+    ; SI: [[COPY:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0
+    ; SI: [[COPY1:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr1
+    ; SI: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY]](<2 x s16>)
+    ; SI: [[UV2:%[0-9]+]]:_(s16), [[UV3:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY1]](<2 x s16>)
+    ; SI: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[UV]](s16)
+    ; SI: [[FPEXT1:%[0-9]+]]:_(s32) = G_FPEXT [[UV2]](s16)
+    ; SI: [[FADD:%[0-9]+]]:_(s32) = G_FADD [[FPEXT]], [[FPEXT1]]
+    ; SI: [[FPTRUNC:%[0-9]+]]:_(s16) = G_FPTRUNC [[FADD]](s32)
+    ; SI: [[FPEXT2:%[0-9]+]]:_(s32) = G_FPEXT [[UV1]](s16)
+    ; SI: [[FPEXT3:%[0-9]+]]:_(s32) = G_FPEXT [[UV3]](s16)
+    ; SI: [[FADD1:%[0-9]+]]:_(s32) = G_FADD [[FPEXT2]], [[FPEXT3]]
+    ; SI: [[FPTRUNC1:%[0-9]+]]:_(s16) = G_FPTRUNC [[FADD1]](s32)
+    ; SI: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR [[FPTRUNC]](s16), [[FPTRUNC1]](s16)
+    ; SI: $vgpr0 = COPY [[BUILD_VECTOR]](<2 x s16>)
+    ; VI-LABEL: name: test_fadd_v2s16
+    ; VI: [[COPY:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0
+    ; VI: [[COPY1:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr1
+    ; VI: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY]](<2 x s16>)
+    ; VI: [[UV2:%[0-9]+]]:_(s16), [[UV3:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY1]](<2 x s16>)
+    ; VI: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[UV]](s16)
+    ; VI: [[FPEXT1:%[0-9]+]]:_(s32) = G_FPEXT [[UV2]](s16)
+    ; VI: [[FADD:%[0-9]+]]:_(s32) = G_FADD [[FPEXT]], [[FPEXT1]]
+    ; VI: [[FPTRUNC:%[0-9]+]]:_(s16) = G_FPTRUNC [[FADD]](s32)
+    ; VI: [[FPEXT2:%[0-9]+]]:_(s32) = G_FPEXT [[UV1]](s16)
+    ; VI: [[FPEXT3:%[0-9]+]]:_(s32) = G_FPEXT [[UV3]](s16)
+    ; VI: [[FADD1:%[0-9]+]]:_(s32) = G_FADD [[FPEXT2]], [[FPEXT3]]
+    ; VI: [[FPTRUNC1:%[0-9]+]]:_(s16) = G_FPTRUNC [[FADD1]](s32)
+    ; VI: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR [[FPTRUNC]](s16), [[FPTRUNC1]](s16)
+    ; VI: $vgpr0 = COPY [[BUILD_VECTOR]](<2 x s16>)
+    ; GFX9-LABEL: name: test_fadd_v2s16
+    ; GFX9: [[COPY:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0
+    ; GFX9: [[COPY1:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr1
+    ; GFX9: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY]](<2 x s16>)
+    ; GFX9: [[UV2:%[0-9]+]]:_(s16), [[UV3:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY1]](<2 x s16>)
+    ; GFX9: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[UV]](s16)
+    ; GFX9: [[FPEXT1:%[0-9]+]]:_(s32) = G_FPEXT [[UV2]](s16)
+    ; GFX9: [[FADD:%[0-9]+]]:_(s32) = G_FADD [[FPEXT]], [[FPEXT1]]
+    ; GFX9: [[FPTRUNC:%[0-9]+]]:_(s16) = G_FPTRUNC [[FADD]](s32)
+    ; GFX9: [[FPEXT2:%[0-9]+]]:_(s32) = G_FPEXT [[UV1]](s16)
+    ; GFX9: [[FPEXT3:%[0-9]+]]:_(s32) = G_FPEXT [[UV3]](s16)
+    ; GFX9: [[FADD1:%[0-9]+]]:_(s32) = G_FADD [[FPEXT2]], [[FPEXT3]]
+    ; GFX9: [[FPTRUNC1:%[0-9]+]]:_(s16) = G_FPTRUNC [[FADD1]](s32)
+    ; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR [[FPTRUNC]](s16), [[FPTRUNC1]](s16)
+    ; GFX9: $vgpr0 = COPY [[BUILD_VECTOR]](<2 x s16>)
+    %0:_(<2 x s16>) = COPY $vgpr0
+    %1:_(<2 x s16>) = COPY $vgpr1
+    %2:_(<2 x s16>) = G_FADD %0, %1
+    $vgpr0 = COPY %2
+...
+
+---
+name: test_fadd_v3s16
+body: |
+  bb.0.entry:
+    liveins: $vgpr0, $vgpr1
+
+    ; SI-LABEL: name: test_fadd_v3s16
+    ; SI: [[DEF:%[0-9]+]]:_(<3 x s16>) = G_IMPLICIT_DEF
+    ; SI: [[DEF1:%[0-9]+]]:_(<3 x s16>) = G_IMPLICIT_DEF
+    ; SI: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[DEF]](<3 x s16>)
+    ; SI: [[UV3:%[0-9]+]]:_(s16), [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[DEF1]](<3 x s16>)
+    ; SI: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[UV]](s16)
+    ; SI: [[FPEXT1:%[0-9]+]]:_(s32) = G_FPEXT [[UV3]](s16)
+    ; SI: [[FADD:%[0-9]+]]:_(s32) = G_FADD [[FPEXT]], [[FPEXT1]]
+    ; SI: [[FPTRUNC:%[0-9]+]]:_(s16) = G_FPTRUNC [[FADD]](s32)
+    ; SI: [[FPEXT2:%[0-9]+]]:_(s32) = G_FPEXT [[UV1]](s16)
+    ; SI: [[FPEXT3:%[0-9]+]]:_(s32) = G_FPEXT [[UV4]](s16)
+    ; SI: [[FADD1:%[0-9]+]]:_(s32) = G_FADD [[FPEXT2]], [[FPEXT3]]
+    ; SI: [[FPTRUNC1:%[0-9]+]]:_(s16) = G_FPTRUNC [[FADD1]](s32)
+    ; SI: [[FPEXT4:%[0-9]+]]:_(s32) = G_FPEXT [[UV2]](s16)
+    ; SI: [[FPEXT5:%[0-9]+]]:_(s32) = G_FPEXT [[UV5]](s16)
+    ; SI: [[FADD2:%[0-9]+]]:_(s32) = G_FADD [[FPEXT4]], [[FPEXT5]]
+    ; SI: [[FPTRUNC2:%[0-9]+]]:_(s16) = G_FPTRUNC [[FADD2]](s32)
+    ; SI: [[BUILD_VECTOR:%[0-9]+]]:_(<3 x s16>) = G_BUILD_VECTOR [[FPTRUNC]](s16), [[FPTRUNC1]](s16), [[FPTRUNC2]](s16)
+    ; SI: S_NOP 0, implicit [[BUILD_VECTOR]](<3 x s16>)
+    ; VI-LABEL: name: test_fadd_v3s16
+    ; VI: [[DEF:%[0-9]+]]:_(<3 x s16>) = G_IMPLICIT_DEF
+    ; VI: [[DEF1:%[0-9]+]]:_(<3 x s16>) = G_IMPLICIT_DEF
+    ; VI: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[DEF]](<3 x s16>)
+    ; VI: [[UV3:%[0-9]+]]:_(s16), [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[DEF1]](<3 x s16>)
+    ; VI: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[UV]](s16)
+    ; VI: [[FPEXT1:%[0-9]+]]:_(s32) = G_FPEXT [[UV3]](s16)
+    ; VI: [[FADD:%[0-9]+]]:_(s32) = G_FADD [[FPEXT]], [[FPEXT1]]
+    ; VI: [[FPTRUNC:%[0-9]+]]:_(s16) = G_FPTRUNC [[FADD]](s32)
+    ; VI: [[FPEXT2:%[0-9]+]]:_(s32) = G_FPEXT [[UV1]](s16)
+    ; VI: [[FPEXT3:%[0-9]+]]:_(s32) = G_FPEXT [[UV4]](s16)
+    ; VI: [[FADD1:%[0-9]+]]:_(s32) = G_FADD [[FPEXT2]], [[FPEXT3]]
+    ; VI: [[FPTRUNC1:%[0-9]+]]:_(s16) = G_FPTRUNC [[FADD1]](s32)
+    ; VI: [[FPEXT4:%[0-9]+]]:_(s32) = G_FPEXT [[UV2]](s16)
+    ; VI: [[FPEXT5:%[0-9]+]]:_(s32) = G_FPEXT [[UV5]](s16)
+    ; VI: [[FADD2:%[0-9]+]]:_(s32) = G_FADD [[FPEXT4]], [[FPEXT5]]
+    ; VI: [[FPTRUNC2:%[0-9]+]]:_(s16) = G_FPTRUNC [[FADD2]](s32)
+    ; VI: [[BUILD_VECTOR:%[0-9]+]]:_(<3 x s16>) = G_BUILD_VECTOR [[FPTRUNC]](s16), [[FPTRUNC1]](s16), [[FPTRUNC2]](s16)
+    ; VI: S_NOP 0, implicit [[BUILD_VECTOR]](<3 x s16>)
+    ; GFX9-LABEL: name: test_fadd_v3s16
+    ; GFX9: [[DEF:%[0-9]+]]:_(<3 x s16>) = G_IMPLICIT_DEF
+    ; GFX9: [[DEF1:%[0-9]+]]:_(<3 x s16>) = G_IMPLICIT_DEF
+    ; GFX9: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[DEF]](<3 x s16>)
+    ; GFX9: [[UV3:%[0-9]+]]:_(s16), [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[DEF1]](<3 x s16>)
+    ; GFX9: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[UV]](s16)
+    ; GFX9: [[FPEXT1:%[0-9]+]]:_(s32) = G_FPEXT [[UV3]](s16)
+    ; GFX9: [[FADD:%[0-9]+]]:_(s32) = G_FADD [[FPEXT]], [[FPEXT1]]
+    ; GFX9: [[FPTRUNC:%[0-9]+]]:_(s16) = G_FPTRUNC [[FADD]](s32)
+    ; GFX9: [[FPEXT2:%[0-9]+]]:_(s32) = G_FPEXT [[UV1]](s16)
+    ; GFX9: [[FPEXT3:%[0-9]+]]:_(s32) = G_FPEXT [[UV4]](s16)
+    ; GFX9: [[FADD1:%[0-9]+]]:_(s32) = G_FADD [[FPEXT2]], [[FPEXT3]]
+    ; GFX9: [[FPTRUNC1:%[0-9]+]]:_(s16) = G_FPTRUNC [[FADD1]](s32)
+    ; GFX9: [[FPEXT4:%[0-9]+]]:_(s32) = G_FPEXT [[UV2]](s16)
+    ; GFX9: [[FPEXT5:%[0-9]+]]:_(s32) = G_FPEXT [[UV5]](s16)
+    ; GFX9: [[FADD2:%[0-9]+]]:_(s32) = G_FADD [[FPEXT4]], [[FPEXT5]]
+    ; GFX9: [[FPTRUNC2:%[0-9]+]]:_(s16) = G_FPTRUNC [[FADD2]](s32)
+    ; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<3 x s16>) = G_BUILD_VECTOR [[FPTRUNC]](s16), [[FPTRUNC1]](s16), [[FPTRUNC2]](s16)
+    ; GFX9: S_NOP 0, implicit [[BUILD_VECTOR]](<3 x s16>)
+    %0:_(<3 x s16>) = G_IMPLICIT_DEF
+    %1:_(<3 x s16>) = G_IMPLICIT_DEF
+    %2:_(<3 x s16>) = G_FADD %0, %1
+    S_NOP 0, implicit %2
+...
+
+---
+name: test_fadd_v4s16
+body: |
+  bb.0.entry:
+    liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
+
+    ; SI-LABEL: name: test_fadd_v4s16
+    ; SI: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1
+    ; SI: [[COPY1:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr2_vgpr3
+    ; SI: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16), [[UV3:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY]](<4 x s16>)
+    ; SI: [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16), [[UV6:%[0-9]+]]:_(s16), [[UV7:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY1]](<4 x s16>)
+    ; SI: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[UV]](s16)
+    ; SI: [[FPEXT1:%[0-9]+]]:_(s32) = G_FPEXT [[UV4]](s16)
+    ; SI: [[FADD:%[0-9]+]]:_(s32) = G_FADD [[FPEXT]], [[FPEXT1]]
+    ; SI: [[FPTRUNC:%[0-9]+]]:_(s16) = G_FPTRUNC [[FADD]](s32)
+    ; SI: [[FPEXT2:%[0-9]+]]:_(s32) = G_FPEXT [[UV1]](s16)
+    ; SI: [[FPEXT3:%[0-9]+]]:_(s32) = G_FPEXT [[UV5]](s16)
+    ; SI: [[FADD1:%[0-9]+]]:_(s32) = G_FADD [[FPEXT2]], [[FPEXT3]]
+    ; SI: [[FPTRUNC1:%[0-9]+]]:_(s16) = G_FPTRUNC [[FADD1]](s32)
+    ; SI: [[FPEXT4:%[0-9]+]]:_(s32) = G_FPEXT [[UV2]](s16)
+    ; SI: [[FPEXT5:%[0-9]+]]:_(s32) = G_FPEXT [[UV6]](s16)
+    ; SI: [[FADD2:%[0-9]+]]:_(s32) = G_FADD [[FPEXT4]], [[FPEXT5]]
+    ; SI: [[FPTRUNC2:%[0-9]+]]:_(s16) = G_FPTRUNC [[FADD2]](s32)
+    ; SI: [[FPEXT6:%[0-9]+]]:_(s32) = G_FPEXT [[UV3]](s16)
+    ; SI: [[FPEXT7:%[0-9]+]]:_(s32) = G_FPEXT [[UV7]](s16)
+    ; SI: [[FADD3:%[0-9]+]]:_(s32) = G_FADD [[FPEXT6]], [[FPEXT7]]
+    ; SI: [[FPTRUNC3:%[0-9]+]]:_(s16) = G_FPTRUNC [[FADD3]](s32)
+    ; SI: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s16>) = G_BUILD_VECTOR [[FPTRUNC]](s16), [[FPTRUNC1]](s16), [[FPTRUNC2]](s16), [[FPTRUNC3]](s16)
+    ; SI: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<4 x s16>)
+    ; VI-LABEL: name: test_fadd_v4s16
+    ; VI: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1
+    ; VI: [[COPY1:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr2_vgpr3
+    ; VI: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16), [[UV3:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY]](<4 x s16>)
+    ; VI: [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16), [[UV6:%[0-9]+]]:_(s16), [[UV7:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY1]](<4 x s16>)
+    ; VI: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[UV]](s16)
+    ; VI: [[FPEXT1:%[0-9]+]]:_(s32) = G_FPEXT [[UV4]](s16)
+    ; VI: [[FADD:%[0-9]+]]:_(s32) = G_FADD [[FPEXT]], [[FPEXT1]]
+    ; VI: [[FPTRUNC:%[0-9]+]]:_(s16) = G_FPTRUNC [[FADD]](s32)
+    ; VI: [[FPEXT2:%[0-9]+]]:_(s32) = G_FPEXT [[UV1]](s16)
+    ; VI: [[FPEXT3:%[0-9]+]]:_(s32) = G_FPEXT [[UV5]](s16)
+    ; VI: [[FADD1:%[0-9]+]]:_(s32) = G_FADD [[FPEXT2]], [[FPEXT3]]
+    ; VI: [[FPTRUNC1:%[0-9]+]]:_(s16) = G_FPTRUNC [[FADD1]](s32)
+    ; VI: [[FPEXT4:%[0-9]+]]:_(s32) = G_FPEXT [[UV2]](s16)
+    ; VI: [[FPEXT5:%[0-9]+]]:_(s32) = G_FPEXT [[UV6]](s16)
+    ; VI: [[FADD2:%[0-9]+]]:_(s32) = G_FADD [[FPEXT4]], [[FPEXT5]]
+    ; VI: [[FPTRUNC2:%[0-9]+]]:_(s16) = G_FPTRUNC [[FADD2]](s32)
+    ; VI: [[FPEXT6:%[0-9]+]]:_(s32) = G_FPEXT [[UV3]](s16)
+    ; VI: [[FPEXT7:%[0-9]+]]:_(s32) = G_FPEXT [[UV7]](s16)
+    ; VI: [[FADD3:%[0-9]+]]:_(s32) = G_FADD [[FPEXT6]], [[FPEXT7]]
+    ; VI: [[FPTRUNC3:%[0-9]+]]:_(s16) = G_FPTRUNC [[FADD3]](s32)
+    ; VI: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s16>) = G_BUILD_VECTOR [[FPTRUNC]](s16), [[FPTRUNC1]](s16), [[FPTRUNC2]](s16), [[FPTRUNC3]](s16)
+    ; VI: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<4 x s16>)
+    ; GFX9-LABEL: name: test_fadd_v4s16
+    ; GFX9: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1
+    ; GFX9: [[COPY1:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr2_vgpr3
+    ; GFX9: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16), [[UV3:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY]](<4 x s16>)
+    ; GFX9: [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16), [[UV6:%[0-9]+]]:_(s16), [[UV7:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY1]](<4 x s16>)
+    ; GFX9: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[UV]](s16)
+    ; GFX9: [[FPEXT1:%[0-9]+]]:_(s32) = G_FPEXT [[UV4]](s16)
+    ; GFX9: [[FADD:%[0-9]+]]:_(s32) = G_FADD [[FPEXT]], [[FPEXT1]]
+    ; GFX9: [[FPTRUNC:%[0-9]+]]:_(s16) = G_FPTRUNC [[FADD]](s32)
+    ; GFX9: [[FPEXT2:%[0-9]+]]:_(s32) = G_FPEXT [[UV1]](s16)
+    ; GFX9: [[FPEXT3:%[0-9]+]]:_(s32) = G_FPEXT [[UV5]](s16)
+    ; GFX9: [[FADD1:%[0-9]+]]:_(s32) = G_FADD [[FPEXT2]], [[FPEXT3]]
+    ; GFX9: [[FPTRUNC1:%[0-9]+]]:_(s16) = G_FPTRUNC [[FADD1]](s32)
+    ; GFX9: [[FPEXT4:%[0-9]+]]:_(s32) = G_FPEXT [[UV2]](s16)
+    ; GFX9: [[FPEXT5:%[0-9]+]]:_(s32) = G_FPEXT [[UV6]](s16)
+    ; GFX9: [[FADD2:%[0-9]+]]:_(s32) = G_FADD [[FPEXT4]], [[FPEXT5]]
+    ; GFX9: [[FPTRUNC2:%[0-9]+]]:_(s16) = G_FPTRUNC [[FADD2]](s32)
+    ; GFX9: [[FPEXT6:%[0-9]+]]:_(s32) = G_FPEXT [[UV3]](s16)
+    ; GFX9: [[FPEXT7:%[0-9]+]]:_(s32) = G_FPEXT [[UV7]](s16)
+    ; GFX9: [[FADD3:%[0-9]+]]:_(s32) = G_FADD [[FPEXT6]], [[FPEXT7]]
+    ; GFX9: [[FPTRUNC3:%[0-9]+]]:_(s16) = G_FPTRUNC [[FADD3]](s32)
+    ; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s16>) = G_BUILD_VECTOR [[FPTRUNC]](s16), [[FPTRUNC1]](s16), [[FPTRUNC2]](s16), [[FPTRUNC3]](s16)
+    ; GFX9: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<4 x s16>)
+    %0:_(<4 x s16>) = COPY $vgpr0_vgpr1
+    %1:_(<4 x s16>) = COPY $vgpr2_vgpr3
+    %2:_(<4 x s16>) = G_FADD %0, %1
+    $vgpr0_vgpr1 = COPY %2
+...
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-fma.mir b/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-fma.mir
index 3928f67ea75..adc52f93b1b 100644
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-fma.mir
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-fma.mir
@@ -123,3 +123,450 @@ body: |
     %7:_(s32) = G_ANYEXT %6
     $vgpr0 = COPY %7
 ...
+
+---
+name: test_fma_v2s32
+body: |
+  bb.0:
+    liveins: $vgpr0_vgpr1, $vgpr2_vgpr3, $vgpr4_vgpr5
+
+    ; SI-LABEL: name: test_fma_v2s32
+    ; SI: [[COPY:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr0_vgpr1
+    ; SI: [[COPY1:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr2_vgpr3
+    ; SI: [[COPY2:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr4_vgpr5
+    ; SI: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](<2 x s32>)
+    ; SI: [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY1]](<2 x s32>)
+    ; SI: [[UV4:%[0-9]+]]:_(s32), [[UV5:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY2]](<2 x s32>)
+    ; SI: [[FMA:%[0-9]+]]:_(s32) = G_FMA [[UV]], [[UV2]], [[UV4]]
+    ; SI: [[FMA1:%[0-9]+]]:_(s32) = G_FMA [[UV1]], [[UV3]], [[UV5]]
+    ; SI: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[FMA]](s32), [[FMA1]](s32)
+    ; SI: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<2 x s32>)
+    ; VI-LABEL: name: test_fma_v2s32
+    ; VI: [[COPY:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr0_vgpr1
+    ; VI: [[COPY1:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr2_vgpr3
+    ; VI: [[COPY2:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr4_vgpr5
+    ; VI: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](<2 x s32>)
+    ; VI: [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY1]](<2 x s32>)
+    ; VI: [[UV4:%[0-9]+]]:_(s32), [[UV5:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY2]](<2 x s32>)
+    ; VI: [[FMA:%[0-9]+]]:_(s32) = G_FMA [[UV]], [[UV2]], [[UV4]]
+    ; VI: [[FMA1:%[0-9]+]]:_(s32) = G_FMA [[UV1]], [[UV3]], [[UV5]]
+    ; VI: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[FMA]](s32), [[FMA1]](s32)
+    ; VI: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<2 x s32>)
+    ; GFX9-LABEL: name: test_fma_v2s32
+    ; GFX9: [[COPY:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr0_vgpr1
+    ; GFX9: [[COPY1:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr2_vgpr3
+    ; GFX9: [[COPY2:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr4_vgpr5
+    ; GFX9: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](<2 x s32>)
+    ; GFX9: [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY1]](<2 x s32>)
+    ; GFX9: [[UV4:%[0-9]+]]:_(s32), [[UV5:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY2]](<2 x s32>)
+    ; GFX9: [[FMA:%[0-9]+]]:_(s32) = G_FMA [[UV]], [[UV2]], [[UV4]]
+    ; GFX9: [[FMA1:%[0-9]+]]:_(s32) = G_FMA [[UV1]], [[UV3]], [[UV5]]
+    ; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[FMA]](s32), [[FMA1]](s32)
+    ; GFX9: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<2 x s32>)
+    %0:_(<2 x s32>) = COPY $vgpr0_vgpr1
+    %1:_(<2 x s32>) = COPY $vgpr2_vgpr3
+    %2:_(<2 x s32>) = COPY $vgpr4_vgpr5
+    %3:_(<2 x s32>) = G_FMA %0, %1, %2
+    $vgpr0_vgpr1 = COPY %3
+...
+
+---
+name: test_fma_v3s32
+body: |
+  bb.0:
+    liveins: $vgpr0_vgpr1_vgpr2, $vgpr3_vgpr4_vgpr5, $vgpr6_vgpr7_vgpr8
+
+    ; SI-LABEL: name: test_fma_v3s32
+    ; SI: [[COPY:%[0-9]+]]:_(<3 x s32>) = COPY $vgpr0_vgpr1_vgpr2
+    ; SI: [[COPY1:%[0-9]+]]:_(<3 x s32>) = COPY $vgpr3_vgpr4_vgpr5
+    ; SI: [[COPY2:%[0-9]+]]:_(<3 x s32>) = COPY $vgpr6_vgpr7_vgpr8
+    ; SI: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](<3 x s32>)
+    ; SI: [[UV3:%[0-9]+]]:_(s32), [[UV4:%[0-9]+]]:_(s32), [[UV5:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY1]](<3 x s32>)
+    ; SI: [[UV6:%[0-9]+]]:_(s32), [[UV7:%[0-9]+]]:_(s32), [[UV8:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY2]](<3 x s32>)
+    ; SI: [[FMA:%[0-9]+]]:_(s32) = G_FMA [[UV]], [[UV3]], [[UV6]]
+    ; SI: [[FMA1:%[0-9]+]]:_(s32) = G_FMA [[UV1]], [[UV4]], [[UV7]]
+    ; SI: [[FMA2:%[0-9]+]]:_(s32) = G_FMA [[UV2]], [[UV5]], [[UV8]]
+    ; SI: [[BUILD_VECTOR:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[FMA]](s32), [[FMA1]](s32), [[FMA2]](s32)
+    ; SI: $vgpr0_vgpr1_vgpr2 = COPY [[BUILD_VECTOR]](<3 x s32>)
+    ; VI-LABEL: name: test_fma_v3s32
+    ; VI: [[COPY:%[0-9]+]]:_(<3 x s32>) = COPY $vgpr0_vgpr1_vgpr2
+    ; VI: [[COPY1:%[0-9]+]]:_(<3 x s32>) = COPY $vgpr3_vgpr4_vgpr5
+    ; VI: [[COPY2:%[0-9]+]]:_(<3 x s32>) = COPY $vgpr6_vgpr7_vgpr8
+    ; VI: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](<3 x s32>)
+    ; VI: [[UV3:%[0-9]+]]:_(s32), [[UV4:%[0-9]+]]:_(s32), [[UV5:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY1]](<3 x s32>)
+    ; VI: [[UV6:%[0-9]+]]:_(s32), [[UV7:%[0-9]+]]:_(s32), [[UV8:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY2]](<3 x s32>)
+    ; VI: [[FMA:%[0-9]+]]:_(s32) = G_FMA [[UV]], [[UV3]], [[UV6]]
+    ; VI: [[FMA1:%[0-9]+]]:_(s32) = G_FMA [[UV1]], [[UV4]], [[UV7]]
+    ; VI: [[FMA2:%[0-9]+]]:_(s32) = G_FMA [[UV2]], [[UV5]], [[UV8]]
+    ; VI: [[BUILD_VECTOR:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[FMA]](s32), [[FMA1]](s32), [[FMA2]](s32)
+    ; VI: $vgpr0_vgpr1_vgpr2 = COPY [[BUILD_VECTOR]](<3 x s32>)
+    ; GFX9-LABEL: name: test_fma_v3s32
+    ; GFX9: [[COPY:%[0-9]+]]:_(<3 x s32>) = COPY $vgpr0_vgpr1_vgpr2
+    ; GFX9: [[COPY1:%[0-9]+]]:_(<3 x s32>) = COPY $vgpr3_vgpr4_vgpr5
+    ; GFX9: [[COPY2:%[0-9]+]]:_(<3 x s32>) = COPY $vgpr6_vgpr7_vgpr8
+    ; GFX9: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](<3 x s32>)
+    ; GFX9: [[UV3:%[0-9]+]]:_(s32), [[UV4:%[0-9]+]]:_(s32), [[UV5:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY1]](<3 x s32>)
+    ; GFX9: [[UV6:%[0-9]+]]:_(s32), [[UV7:%[0-9]+]]:_(s32), [[UV8:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY2]](<3 x s32>)
+    ; GFX9: [[FMA:%[0-9]+]]:_(s32) = G_FMA [[UV]], [[UV3]], [[UV6]]
+    ; GFX9: [[FMA1:%[0-9]+]]:_(s32) = G_FMA [[UV1]], [[UV4]], [[UV7]]
+    ; GFX9: [[FMA2:%[0-9]+]]:_(s32) = G_FMA [[UV2]], [[UV5]], [[UV8]]
+    ; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[FMA]](s32), [[FMA1]](s32), [[FMA2]](s32)
+    ; GFX9: $vgpr0_vgpr1_vgpr2 = COPY [[BUILD_VECTOR]](<3 x s32>)
+    %0:_(<3 x s32>) = COPY $vgpr0_vgpr1_vgpr2
+    %1:_(<3 x s32>) = COPY $vgpr3_vgpr4_vgpr5
+    %2:_(<3 x s32>) = COPY $vgpr6_vgpr7_vgpr8
+    %3:_(<3 x s32>) = G_FMA %0, %1, %2
+    $vgpr0_vgpr1_vgpr2 = COPY %3
+...
+
+---
+name: test_fma_v4s32
+body: |
+  bb.0:
+    liveins: $vgpr0_vgpr1_vgpr2_vgpr3, $vgpr4_vgpr5_vgpr6_vgpr7, $vgpr8_vgpr9_vgpr10_vgpr11
+
+    ; SI-LABEL: name: test_fma_v4s32
+    ; SI: [[COPY:%[0-9]+]]:_(<4 x s32>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3
+    ; SI: [[COPY1:%[0-9]+]]:_(<4 x s32>) = COPY $vgpr4_vgpr5_vgpr6_vgpr7
+    ; SI: [[COPY2:%[0-9]+]]:_(<4 x s32>) = COPY $vgpr8_vgpr9_vgpr10_vgpr11
+    ; SI: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](<4 x s32>)
+    ; SI: [[UV4:%[0-9]+]]:_(s32), [[UV5:%[0-9]+]]:_(s32), [[UV6:%[0-9]+]]:_(s32), [[UV7:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY1]](<4 x s32>)
+    ; SI: [[UV8:%[0-9]+]]:_(s32), [[UV9:%[0-9]+]]:_(s32), [[UV10:%[0-9]+]]:_(s32), [[UV11:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY2]](<4 x s32>)
+    ; SI: [[FMA:%[0-9]+]]:_(s32) = G_FMA [[UV]], [[UV4]], [[UV8]]
+    ; SI: [[FMA1:%[0-9]+]]:_(s32) = G_FMA [[UV1]], [[UV5]], [[UV9]]
+    ; SI: [[FMA2:%[0-9]+]]:_(s32) = G_FMA [[UV2]], [[UV6]], [[UV10]]
+    ; SI: [[FMA3:%[0-9]+]]:_(s32) = G_FMA [[UV3]], [[UV7]], [[UV11]]
+    ; SI: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[FMA]](s32), [[FMA1]](s32), [[FMA2]](s32), [[FMA3]](s32)
+    ; SI: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[BUILD_VECTOR]](<4 x s32>)
+    ; VI-LABEL: name: test_fma_v4s32
+    ; VI: [[COPY:%[0-9]+]]:_(<4 x s32>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3
+    ; VI: [[COPY1:%[0-9]+]]:_(<4 x s32>) = COPY $vgpr4_vgpr5_vgpr6_vgpr7
+    ; VI: [[COPY2:%[0-9]+]]:_(<4 x s32>) = COPY $vgpr8_vgpr9_vgpr10_vgpr11
+    ; VI: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](<4 x s32>)
+    ; VI: [[UV4:%[0-9]+]]:_(s32), [[UV5:%[0-9]+]]:_(s32), [[UV6:%[0-9]+]]:_(s32), [[UV7:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY1]](<4 x s32>)
+    ; VI: [[UV8:%[0-9]+]]:_(s32), [[UV9:%[0-9]+]]:_(s32), [[UV10:%[0-9]+]]:_(s32), [[UV11:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY2]](<4 x s32>)
+    ; VI: [[FMA:%[0-9]+]]:_(s32) = G_FMA [[UV]], [[UV4]], [[UV8]]
+    ; VI: [[FMA1:%[0-9]+]]:_(s32) = G_FMA [[UV1]], [[UV5]], [[UV9]]
+    ; VI: [[FMA2:%[0-9]+]]:_(s32) = G_FMA [[UV2]], [[UV6]], [[UV10]]
+    ; VI: [[FMA3:%[0-9]+]]:_(s32) = G_FMA [[UV3]], [[UV7]], [[UV11]]
+    ; VI: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[FMA]](s32), [[FMA1]](s32), [[FMA2]](s32), [[FMA3]](s32)
+    ; VI: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[BUILD_VECTOR]](<4 x s32>)
+    ; GFX9-LABEL: name: test_fma_v4s32
+    ; GFX9: [[COPY:%[0-9]+]]:_(<4 x s32>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3
+    ; GFX9: [[COPY1:%[0-9]+]]:_(<4 x s32>) = COPY $vgpr4_vgpr5_vgpr6_vgpr7
+    ; GFX9: [[COPY2:%[0-9]+]]:_(<4 x s32>) = COPY $vgpr8_vgpr9_vgpr10_vgpr11
+    ; GFX9: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](<4 x s32>)
+    ; GFX9: [[UV4:%[0-9]+]]:_(s32), [[UV5:%[0-9]+]]:_(s32), [[UV6:%[0-9]+]]:_(s32), [[UV7:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY1]](<4 x s32>)
+    ; GFX9: [[UV8:%[0-9]+]]:_(s32), [[UV9:%[0-9]+]]:_(s32), [[UV10:%[0-9]+]]:_(s32), [[UV11:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY2]](<4 x s32>)
+    ; GFX9: [[FMA:%[0-9]+]]:_(s32) = G_FMA [[UV]], [[UV4]], [[UV8]]
+    ; GFX9: [[FMA1:%[0-9]+]]:_(s32) = G_FMA [[UV1]], [[UV5]], [[UV9]]
+    ; GFX9: [[FMA2:%[0-9]+]]:_(s32) = G_FMA [[UV2]], [[UV6]], [[UV10]]
+    ; GFX9: [[FMA3:%[0-9]+]]:_(s32) = G_FMA [[UV3]], [[UV7]], [[UV11]]
+    ; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[FMA]](s32), [[FMA1]](s32), [[FMA2]](s32), [[FMA3]](s32)
+    ; GFX9: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[BUILD_VECTOR]](<4 x s32>)
+    %0:_(<4 x s32>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3
+    %1:_(<4 x s32>) = COPY $vgpr4_vgpr5_vgpr6_vgpr7
+    %2:_(<4 x s32>) = COPY $vgpr8_vgpr9_vgpr10_vgpr11
+    %3:_(<4 x s32>) = G_FMA %0, %1, %2
+    $vgpr0_vgpr1_vgpr2_vgpr3 = COPY %3
+...
+
+---
+name: test_fma_v2s64
+body: |
+  bb.0:
+    liveins: $vgpr0_vgpr1_vgpr2_vgpr3, $vgpr4_vgpr5_vgpr6_vgpr7, $vgpr8_vgpr9_vgpr10_vgpr11
+
+    ; SI-LABEL: name: test_fma_v2s64
+    ; SI: [[COPY:%[0-9]+]]:_(<2 x s64>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3
+    ; SI: [[COPY1:%[0-9]+]]:_(<2 x s64>) = COPY $vgpr4_vgpr5_vgpr6_vgpr7
+    ; SI: [[COPY2:%[0-9]+]]:_(<2 x s64>) = COPY $vgpr8_vgpr9_vgpr10_vgpr11
+    ; SI: [[UV:%[0-9]+]]:_(s64), [[UV1:%[0-9]+]]:_(s64) = G_UNMERGE_VALUES [[COPY]](<2 x s64>)
+    ; SI: [[UV2:%[0-9]+]]:_(s64), [[UV3:%[0-9]+]]:_(s64) = G_UNMERGE_VALUES [[COPY1]](<2 x s64>)
+    ; SI: [[UV4:%[0-9]+]]:_(s64), [[UV5:%[0-9]+]]:_(s64) = G_UNMERGE_VALUES [[COPY2]](<2 x s64>)
+    ; SI: [[FMA:%[0-9]+]]:_(s64) = G_FMA [[UV]], [[UV2]], [[UV4]]
+    ; SI: [[FMA1:%[0-9]+]]:_(s64) = G_FMA [[UV1]], [[UV3]], [[UV5]]
+    ; SI: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s64>) = G_BUILD_VECTOR [[FMA]](s64), [[FMA1]](s64)
+    ; SI: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[BUILD_VECTOR]](<2 x s64>)
+    ; VI-LABEL: name: test_fma_v2s64
+    ; VI: [[COPY:%[0-9]+]]:_(<2 x s64>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3
+    ; VI: [[COPY1:%[0-9]+]]:_(<2 x s64>) = COPY $vgpr4_vgpr5_vgpr6_vgpr7
+    ; VI: [[COPY2:%[0-9]+]]:_(<2 x s64>) = COPY $vgpr8_vgpr9_vgpr10_vgpr11
+    ; VI: [[UV:%[0-9]+]]:_(s64), [[UV1:%[0-9]+]]:_(s64) = G_UNMERGE_VALUES [[COPY]](<2 x s64>)
+    ; VI: [[UV2:%[0-9]+]]:_(s64), [[UV3:%[0-9]+]]:_(s64) = G_UNMERGE_VALUES [[COPY1]](<2 x s64>)
+    ; VI: [[UV4:%[0-9]+]]:_(s64), [[UV5:%[0-9]+]]:_(s64) = G_UNMERGE_VALUES [[COPY2]](<2 x s64>)
+    ; VI: [[FMA:%[0-9]+]]:_(s64) = G_FMA [[UV]], [[UV2]], [[UV4]]
+    ; VI: [[FMA1:%[0-9]+]]:_(s64) = G_FMA [[UV1]], [[UV3]], [[UV5]]
+    ; VI: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s64>) = G_BUILD_VECTOR [[FMA]](s64), [[FMA1]](s64)
+    ; VI: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[BUILD_VECTOR]](<2 x s64>)
+    ; GFX9-LABEL: name: test_fma_v2s64
+    ; GFX9: [[COPY:%[0-9]+]]:_(<2 x s64>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3
+    ; GFX9: [[COPY1:%[0-9]+]]:_(<2 x s64>) = COPY $vgpr4_vgpr5_vgpr6_vgpr7
+    ; GFX9: [[COPY2:%[0-9]+]]:_(<2 x s64>) = COPY $vgpr8_vgpr9_vgpr10_vgpr11
+    ; GFX9: [[UV:%[0-9]+]]:_(s64), [[UV1:%[0-9]+]]:_(s64) = G_UNMERGE_VALUES [[COPY]](<2 x s64>)
+    ; GFX9: [[UV2:%[0-9]+]]:_(s64), [[UV3:%[0-9]+]]:_(s64) = G_UNMERGE_VALUES [[COPY1]](<2 x s64>)
+    ; GFX9: [[UV4:%[0-9]+]]:_(s64), [[UV5:%[0-9]+]]:_(s64) = G_UNMERGE_VALUES [[COPY2]](<2 x s64>)
+    ; GFX9: [[FMA:%[0-9]+]]:_(s64) = G_FMA [[UV]], [[UV2]], [[UV4]]
+    ; GFX9: [[FMA1:%[0-9]+]]:_(s64) = G_FMA [[UV1]], [[UV3]], [[UV5]]
+    ; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s64>) = G_BUILD_VECTOR [[FMA]](s64), [[FMA1]](s64)
+    ; GFX9: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[BUILD_VECTOR]](<2 x s64>)
+    %0:_(<2 x s64>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3
+    %1:_(<2 x s64>) = COPY $vgpr4_vgpr5_vgpr6_vgpr7
+    %2:_(<2 x s64>) = COPY $vgpr8_vgpr9_vgpr10_vgpr11
+    %3:_(<2 x s64>) = G_FMA %0, %1, %2
+    $vgpr0_vgpr1_vgpr2_vgpr3 = COPY %3
+...
+
+---
+name: test_fma_v2s16
+body: |
+  bb.0:
+    liveins: $vgpr0, $vgpr1, $vgpr2
+
+    ; SI-LABEL: name: test_fma_v2s16
+    ; SI: [[COPY:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0
+    ; SI: [[COPY1:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr1
+    ; SI: [[COPY2:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr2
+    ; SI: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY]](<2 x s16>)
+    ; SI: [[UV2:%[0-9]+]]:_(s16), [[UV3:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY1]](<2 x s16>)
+    ; SI: [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY2]](<2 x s16>)
+    ; SI: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[UV]](s16)
+    ; SI: [[FPEXT1:%[0-9]+]]:_(s32) = G_FPEXT [[UV2]](s16)
+    ; SI: [[FPEXT2:%[0-9]+]]:_(s32) = G_FPEXT [[UV4]](s16)
+    ; SI: [[FMA:%[0-9]+]]:_(s32) = G_FMA [[FPEXT]], [[FPEXT1]], [[FPEXT2]]
+    ; SI: [[FPTRUNC:%[0-9]+]]:_(s16) = G_FPTRUNC [[FMA]](s32)
+    ; SI: [[FPEXT3:%[0-9]+]]:_(s32) = G_FPEXT [[UV1]](s16)
+    ; SI: [[FPEXT4:%[0-9]+]]:_(s32) = G_FPEXT [[UV3]](s16)
+    ; SI: [[FPEXT5:%[0-9]+]]:_(s32) = G_FPEXT [[UV5]](s16)
+    ; SI: [[FMA1:%[0-9]+]]:_(s32) = G_FMA [[FPEXT3]], [[FPEXT4]], [[FPEXT5]]
+    ; SI: [[FPTRUNC1:%[0-9]+]]:_(s16) = G_FPTRUNC [[FMA1]](s32)
+    ; SI: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR [[FPTRUNC]](s16), [[FPTRUNC1]](s16)
+    ; SI: $vgpr0 = COPY [[BUILD_VECTOR]](<2 x s16>)
+    ; VI-LABEL: name: test_fma_v2s16
+    ; VI: [[COPY:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0
+    ; VI: [[COPY1:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr1
+    ; VI: [[COPY2:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr2
+    ; VI: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY]](<2 x s16>)
+    ; VI: [[UV2:%[0-9]+]]:_(s16), [[UV3:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY1]](<2 x s16>)
+    ; VI: [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY2]](<2 x s16>)
+    ; VI: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[UV]](s16)
+    ; VI: [[FPEXT1:%[0-9]+]]:_(s32) = G_FPEXT [[UV2]](s16)
+    ; VI: [[FPEXT2:%[0-9]+]]:_(s32) = G_FPEXT [[UV4]](s16)
+    ; VI: [[FMA:%[0-9]+]]:_(s32) = G_FMA [[FPEXT]], [[FPEXT1]], [[FPEXT2]]
+    ; VI: [[FPTRUNC:%[0-9]+]]:_(s16) = G_FPTRUNC [[FMA]](s32)
+    ; VI: [[FPEXT3:%[0-9]+]]:_(s32) = G_FPEXT [[UV1]](s16)
+    ; VI: [[FPEXT4:%[0-9]+]]:_(s32) = G_FPEXT [[UV3]](s16)
+    ; VI: [[FPEXT5:%[0-9]+]]:_(s32) = G_FPEXT [[UV5]](s16)
+    ; VI: [[FMA1:%[0-9]+]]:_(s32) = G_FMA [[FPEXT3]], [[FPEXT4]], [[FPEXT5]]
+    ; VI: [[FPTRUNC1:%[0-9]+]]:_(s16) = G_FPTRUNC [[FMA1]](s32)
+    ; VI: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR [[FPTRUNC]](s16), [[FPTRUNC1]](s16)
+    ; VI: $vgpr0 = COPY [[BUILD_VECTOR]](<2 x s16>)
+    ; GFX9-LABEL: name: test_fma_v2s16
+    ; GFX9: [[COPY:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0
+    ; GFX9: [[COPY1:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr1
+    ; GFX9: [[COPY2:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr2
+    ; GFX9: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY]](<2 x s16>)
+    ; GFX9: [[UV2:%[0-9]+]]:_(s16), [[UV3:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY1]](<2 x s16>)
+    ; GFX9: [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY2]](<2 x s16>)
+    ; GFX9: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[UV]](s16)
+    ; GFX9: [[FPEXT1:%[0-9]+]]:_(s32) = G_FPEXT [[UV2]](s16)
+    ; GFX9: [[FPEXT2:%[0-9]+]]:_(s32) = G_FPEXT [[UV4]](s16)
+    ; GFX9: [[FMA:%[0-9]+]]:_(s32) = G_FMA [[FPEXT]], [[FPEXT1]], [[FPEXT2]]
+    ; GFX9: [[FPTRUNC:%[0-9]+]]:_(s16) = G_FPTRUNC [[FMA]](s32)
+    ; GFX9: [[FPEXT3:%[0-9]+]]:_(s32) = G_FPEXT [[UV1]](s16)
+    ; GFX9: [[FPEXT4:%[0-9]+]]:_(s32) = G_FPEXT [[UV3]](s16)
+    ; GFX9: [[FPEXT5:%[0-9]+]]:_(s32) = G_FPEXT [[UV5]](s16)
+    ; GFX9: [[FMA1:%[0-9]+]]:_(s32) = G_FMA [[FPEXT3]], [[FPEXT4]], [[FPEXT5]]
+    ; GFX9: [[FPTRUNC1:%[0-9]+]]:_(s16) = G_FPTRUNC [[FMA1]](s32)
+    ; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR [[FPTRUNC]](s16), [[FPTRUNC1]](s16)
+    ; GFX9: $vgpr0 = COPY [[BUILD_VECTOR]](<2 x s16>)
+    %0:_(<2 x s16>) = COPY $vgpr0
+    %1:_(<2 x s16>) = COPY $vgpr1
+    %2:_(<2 x s16>) = COPY $vgpr2
+    %3:_(<2 x s16>) = G_FMA %0, %1, %2
+    $vgpr0 = COPY %3
+...
+
+---
+name: test_fma_v3s16
+body: |
+  bb.0:
+
+    ; SI-LABEL: name: test_fma_v3s16
+    ; SI: [[DEF:%[0-9]+]]:_(<3 x s16>) = G_IMPLICIT_DEF
+    ; SI: [[DEF1:%[0-9]+]]:_(<3 x s16>) = G_IMPLICIT_DEF
+    ; SI: [[DEF2:%[0-9]+]]:_(<3 x s16>) = G_IMPLICIT_DEF
+    ; SI: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[DEF]](<3 x s16>)
+    ; SI: [[UV3:%[0-9]+]]:_(s16), [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[DEF1]](<3 x s16>)
+    ; SI: [[UV6:%[0-9]+]]:_(s16), [[UV7:%[0-9]+]]:_(s16), [[UV8:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[DEF2]](<3 x s16>)
+    ; SI: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[UV]](s16)
+    ; SI: [[FPEXT1:%[0-9]+]]:_(s32) = G_FPEXT [[UV3]](s16)
+    ; SI: [[FPEXT2:%[0-9]+]]:_(s32) = G_FPEXT [[UV6]](s16)
+    ; SI: [[FMA:%[0-9]+]]:_(s32) = G_FMA [[FPEXT]], [[FPEXT1]], [[FPEXT2]]
+    ; SI: [[FPTRUNC:%[0-9]+]]:_(s16) = G_FPTRUNC [[FMA]](s32)
+    ; SI: [[FPEXT3:%[0-9]+]]:_(s32) = G_FPEXT [[UV1]](s16)
+    ; SI: [[FPEXT4:%[0-9]+]]:_(s32) = G_FPEXT [[UV4]](s16)
+    ; SI: [[FPEXT5:%[0-9]+]]:_(s32) = G_FPEXT [[UV7]](s16)
+    ; SI: [[FMA1:%[0-9]+]]:_(s32) = G_FMA [[FPEXT3]], [[FPEXT4]], [[FPEXT5]]
+    ; SI: [[FPTRUNC1:%[0-9]+]]:_(s16) = G_FPTRUNC [[FMA1]](s32)
+    ; SI: [[FPEXT6:%[0-9]+]]:_(s32) = G_FPEXT [[UV2]](s16)
+    ; SI: [[FPEXT7:%[0-9]+]]:_(s32) = G_FPEXT [[UV5]](s16)
+    ; SI: [[FPEXT8:%[0-9]+]]:_(s32) = G_FPEXT [[UV8]](s16)
+    ; SI: [[FMA2:%[0-9]+]]:_(s32) = G_FMA [[FPEXT6]], [[FPEXT7]], [[FPEXT8]]
+    ; SI: [[FPTRUNC2:%[0-9]+]]:_(s16) = G_FPTRUNC [[FMA2]](s32)
+    ; SI: [[BUILD_VECTOR:%[0-9]+]]:_(<3 x s16>) = G_BUILD_VECTOR [[FPTRUNC]](s16), [[FPTRUNC1]](s16), [[FPTRUNC2]](s16)
+    ; SI: S_NOP 0, implicit [[BUILD_VECTOR]](<3 x s16>)
+    ; VI-LABEL: name: test_fma_v3s16
+    ; VI: [[DEF:%[0-9]+]]:_(<3 x s16>) = G_IMPLICIT_DEF
+    ; VI: [[DEF1:%[0-9]+]]:_(<3 x s16>) = G_IMPLICIT_DEF
+    ; VI: [[DEF2:%[0-9]+]]:_(<3 x s16>) = G_IMPLICIT_DEF
+    ; VI: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[DEF]](<3 x s16>)
+    ; VI: [[UV3:%[0-9]+]]:_(s16), [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[DEF1]](<3 x s16>)
+    ; VI: [[UV6:%[0-9]+]]:_(s16), [[UV7:%[0-9]+]]:_(s16), [[UV8:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[DEF2]](<3 x s16>)
+    ; VI: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[UV]](s16)
+    ; VI: [[FPEXT1:%[0-9]+]]:_(s32) = G_FPEXT [[UV3]](s16)
+    ; VI: [[FPEXT2:%[0-9]+]]:_(s32) = G_FPEXT [[UV6]](s16)
+    ; VI: [[FMA:%[0-9]+]]:_(s32) = G_FMA [[FPEXT]], [[FPEXT1]], [[FPEXT2]]
+    ; VI: [[FPTRUNC:%[0-9]+]]:_(s16) = G_FPTRUNC [[FMA]](s32)
+    ; VI: [[FPEXT3:%[0-9]+]]:_(s32) = G_FPEXT [[UV1]](s16)
+    ; VI: [[FPEXT4:%[0-9]+]]:_(s32) = G_FPEXT [[UV4]](s16)
+    ; VI: [[FPEXT5:%[0-9]+]]:_(s32) = G_FPEXT [[UV7]](s16)
+    ; VI: [[FMA1:%[0-9]+]]:_(s32) = G_FMA [[FPEXT3]], [[FPEXT4]], [[FPEXT5]]
+    ; VI: [[FPTRUNC1:%[0-9]+]]:_(s16) = G_FPTRUNC [[FMA1]](s32)
+    ; VI: [[FPEXT6:%[0-9]+]]:_(s32) = G_FPEXT [[UV2]](s16)
+    ; VI: [[FPEXT7:%[0-9]+]]:_(s32) = G_FPEXT [[UV5]](s16)
+    ; VI: [[FPEXT8:%[0-9]+]]:_(s32) = G_FPEXT [[UV8]](s16)
+    ; VI: [[FMA2:%[0-9]+]]:_(s32) = G_FMA [[FPEXT6]], [[FPEXT7]], [[FPEXT8]]
+    ; VI: [[FPTRUNC2:%[0-9]+]]:_(s16) = G_FPTRUNC [[FMA2]](s32)
+    ; VI: [[BUILD_VECTOR:%[0-9]+]]:_(<3 x s16>) = G_BUILD_VECTOR [[FPTRUNC]](s16), [[FPTRUNC1]](s16), [[FPTRUNC2]](s16)
+    ; VI: S_NOP 0, implicit [[BUILD_VECTOR]](<3 x s16>)
+    ; GFX9-LABEL: name: test_fma_v3s16
+    ; GFX9: [[DEF:%[0-9]+]]:_(<3 x s16>) = G_IMPLICIT_DEF
+    ; GFX9: [[DEF1:%[0-9]+]]:_(<3 x s16>) = G_IMPLICIT_DEF
+    ; GFX9: [[DEF2:%[0-9]+]]:_(<3 x s16>) = G_IMPLICIT_DEF
+    ; GFX9: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[DEF]](<3 x s16>)
+    ; GFX9: [[UV3:%[0-9]+]]:_(s16), [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[DEF1]](<3 x s16>)
+    ; GFX9: [[UV6:%[0-9]+]]:_(s16), [[UV7:%[0-9]+]]:_(s16), [[UV8:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[DEF2]](<3 x s16>)
+    ; GFX9: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[UV]](s16)
+    ; GFX9: [[FPEXT1:%[0-9]+]]:_(s32) = G_FPEXT [[UV3]](s16)
+    ; GFX9: [[FPEXT2:%[0-9]+]]:_(s32) = G_FPEXT [[UV6]](s16)
+    ; GFX9: [[FMA:%[0-9]+]]:_(s32) = G_FMA [[FPEXT]], [[FPEXT1]], [[FPEXT2]]
+    ; GFX9: [[FPTRUNC:%[0-9]+]]:_(s16) = G_FPTRUNC [[FMA]](s32)
+    ; GFX9: [[FPEXT3:%[0-9]+]]:_(s32) = G_FPEXT [[UV1]](s16)
+    ; GFX9: [[FPEXT4:%[0-9]+]]:_(s32) = G_FPEXT [[UV4]](s16)
+    ; GFX9: [[FPEXT5:%[0-9]+]]:_(s32) = G_FPEXT [[UV7]](s16)
+    ; GFX9: [[FMA1:%[0-9]+]]:_(s32) = G_FMA [[FPEXT3]], [[FPEXT4]], [[FPEXT5]]
+    ; GFX9: [[FPTRUNC1:%[0-9]+]]:_(s16) = G_FPTRUNC [[FMA1]](s32)
+    ; GFX9: [[FPEXT6:%[0-9]+]]:_(s32) = G_FPEXT [[UV2]](s16)
+    ; GFX9: [[FPEXT7:%[0-9]+]]:_(s32) = G_FPEXT [[UV5]](s16)
+    ; GFX9: [[FPEXT8:%[0-9]+]]:_(s32) = G_FPEXT [[UV8]](s16)
+    ; GFX9: [[FMA2:%[0-9]+]]:_(s32) = G_FMA [[FPEXT6]], [[FPEXT7]], [[FPEXT8]]
+    ; GFX9: [[FPTRUNC2:%[0-9]+]]:_(s16) = G_FPTRUNC [[FMA2]](s32)
+    ; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<3 x s16>) = G_BUILD_VECTOR [[FPTRUNC]](s16), [[FPTRUNC1]](s16), [[FPTRUNC2]](s16)
+    ; GFX9: S_NOP 0, implicit [[BUILD_VECTOR]](<3 x s16>)
+    %0:_(<3 x s16>) = G_IMPLICIT_DEF
+    %1:_(<3 x s16>) = G_IMPLICIT_DEF
+    %2:_(<3 x s16>) = G_IMPLICIT_DEF
+    %3:_(<3 x s16>) = G_FMA %0, %1, %2
+    S_NOP 0, implicit %3
+...
+
+---
+name: test_fma_v4s16
+body: |
+  bb.0:
+    liveins: $vgpr0_vgpr1, $vgpr2_vgpr3, $vgpr4_vgpr5
+
+    ; SI-LABEL: name: test_fma_v4s16
+    ; SI: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1
+    ; SI: [[COPY1:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr2_vgpr3
+    ; SI: [[COPY2:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr4_vgpr5
+    ; SI: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16), [[UV3:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY]](<4 x s16>)
+    ; SI: [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16), [[UV6:%[0-9]+]]:_(s16), [[UV7:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY1]](<4 x s16>)
+    ; SI: [[UV8:%[0-9]+]]:_(s16), [[UV9:%[0-9]+]]:_(s16), [[UV10:%[0-9]+]]:_(s16), [[UV11:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY2]](<4 x s16>)
+    ; SI: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[UV]](s16)
+    ; SI: [[FPEXT1:%[0-9]+]]:_(s32) = G_FPEXT [[UV4]](s16)
+    ; SI: [[FPEXT2:%[0-9]+]]:_(s32) = G_FPEXT [[UV8]](s16)
+    ; SI: [[FMA:%[0-9]+]]:_(s32) = G_FMA [[FPEXT]], [[FPEXT1]], [[FPEXT2]]
+    ; SI: [[FPTRUNC:%[0-9]+]]:_(s16) = G_FPTRUNC [[FMA]](s32)
+    ; SI: [[FPEXT3:%[0-9]+]]:_(s32) = G_FPEXT [[UV1]](s16)
+    ; SI: [[FPEXT4:%[0-9]+]]:_(s32) = G_FPEXT [[UV5]](s16)
+    ; SI: [[FPEXT5:%[0-9]+]]:_(s32) = G_FPEXT [[UV9]](s16)
+    ; SI: [[FMA1:%[0-9]+]]:_(s32) = G_FMA [[FPEXT3]], [[FPEXT4]], [[FPEXT5]]
+    ; SI: [[FPTRUNC1:%[0-9]+]]:_(s16) = G_FPTRUNC [[FMA1]](s32)
+    ; SI: [[FPEXT6:%[0-9]+]]:_(s32) = G_FPEXT [[UV2]](s16)
+    ; SI: [[FPEXT7:%[0-9]+]]:_(s32) = G_FPEXT [[UV6]](s16)
+    ; SI: [[FPEXT8:%[0-9]+]]:_(s32) = G_FPEXT [[UV10]](s16)
+    ; SI: [[FMA2:%[0-9]+]]:_(s32) = G_FMA [[FPEXT6]], [[FPEXT7]], [[FPEXT8]]
+    ; SI: [[FPTRUNC2:%[0-9]+]]:_(s16) = G_FPTRUNC [[FMA2]](s32)
+    ; SI: [[FPEXT9:%[0-9]+]]:_(s32) = G_FPEXT [[UV3]](s16)
+    ; SI: [[FPEXT10:%[0-9]+]]:_(s32) = G_FPEXT [[UV7]](s16)
+    ; SI: [[FPEXT11:%[0-9]+]]:_(s32) = G_FPEXT [[UV11]](s16)
+    ; SI: [[FMA3:%[0-9]+]]:_(s32) = G_FMA [[FPEXT9]], [[FPEXT10]], [[FPEXT11]]
+    ; SI: [[FPTRUNC3:%[0-9]+]]:_(s16) = G_FPTRUNC [[FMA3]](s32)
+    ; SI: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s16>) = G_BUILD_VECTOR [[FPTRUNC]](s16), [[FPTRUNC1]](s16), [[FPTRUNC2]](s16), [[FPTRUNC3]](s16)
+    ; SI: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<4 x s16>)
+    ; VI-LABEL: name: test_fma_v4s16
+    ; VI: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1
+    ; VI: [[COPY1:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr2_vgpr3
+    ; VI: [[COPY2:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr4_vgpr5
+    ; VI: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16), [[UV3:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY]](<4 x s16>)
+    ; VI: [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16), [[UV6:%[0-9]+]]:_(s16), [[UV7:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY1]](<4 x s16>)
+    ; VI: [[UV8:%[0-9]+]]:_(s16), [[UV9:%[0-9]+]]:_(s16), [[UV10:%[0-9]+]]:_(s16), [[UV11:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY2]](<4 x s16>)
+    ; VI: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[UV]](s16)
+    ; VI: [[FPEXT1:%[0-9]+]]:_(s32) = G_FPEXT [[UV4]](s16)
+    ; VI: [[FPEXT2:%[0-9]+]]:_(s32) = G_FPEXT [[UV8]](s16)
+    ; VI: [[FMA:%[0-9]+]]:_(s32) = G_FMA [[FPEXT]], [[FPEXT1]], [[FPEXT2]]
+    ; VI: [[FPTRUNC:%[0-9]+]]:_(s16) = G_FPTRUNC [[FMA]](s32)
+    ; VI: [[FPEXT3:%[0-9]+]]:_(s32) = G_FPEXT [[UV1]](s16)
+    ; VI: [[FPEXT4:%[0-9]+]]:_(s32) = G_FPEXT [[UV5]](s16)
+    ; VI: [[FPEXT5:%[0-9]+]]:_(s32) = G_FPEXT [[UV9]](s16)
+    ; VI: [[FMA1:%[0-9]+]]:_(s32) = G_FMA [[FPEXT3]], [[FPEXT4]], [[FPEXT5]]
+    ; VI: [[FPTRUNC1:%[0-9]+]]:_(s16) = G_FPTRUNC [[FMA1]](s32)
+    ; VI: [[FPEXT6:%[0-9]+]]:_(s32) = G_FPEXT [[UV2]](s16)
+    ; VI: [[FPEXT7:%[0-9]+]]:_(s32) = G_FPEXT [[UV6]](s16)
+    ; VI: [[FPEXT8:%[0-9]+]]:_(s32) = G_FPEXT [[UV10]](s16)
+    ; VI: [[FMA2:%[0-9]+]]:_(s32) = G_FMA [[FPEXT6]], [[FPEXT7]], [[FPEXT8]]
+    ; VI: [[FPTRUNC2:%[0-9]+]]:_(s16) = G_FPTRUNC [[FMA2]](s32)
+    ; VI: [[FPEXT9:%[0-9]+]]:_(s32) = G_FPEXT [[UV3]](s16)
+    ; VI: [[FPEXT10:%[0-9]+]]:_(s32) = G_FPEXT [[UV7]](s16)
+    ; VI: [[FPEXT11:%[0-9]+]]:_(s32) = G_FPEXT [[UV11]](s16)
+    ; VI: [[FMA3:%[0-9]+]]:_(s32) = G_FMA [[FPEXT9]], [[FPEXT10]], [[FPEXT11]]
+    ; VI: [[FPTRUNC3:%[0-9]+]]:_(s16) = G_FPTRUNC [[FMA3]](s32)
+    ; VI: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s16>) = G_BUILD_VECTOR [[FPTRUNC]](s16), [[FPTRUNC1]](s16), [[FPTRUNC2]](s16), [[FPTRUNC3]](s16)
+    ; VI: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<4 x s16>)
+    ; GFX9-LABEL: name: test_fma_v4s16
+    ; GFX9: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1
+    ; GFX9: [[COPY1:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr2_vgpr3
+    ; GFX9: [[COPY2:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr4_vgpr5
+    ; GFX9: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16), [[UV3:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY]](<4 x s16>)
+    ; GFX9: [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16), [[UV6:%[0-9]+]]:_(s16), [[UV7:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY1]](<4 x s16>)
+    ; GFX9: [[UV8:%[0-9]+]]:_(s16), [[UV9:%[0-9]+]]:_(s16), [[UV10:%[0-9]+]]:_(s16), [[UV11:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY2]](<4 x s16>)
+    ; GFX9: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[UV]](s16)
+    ; GFX9: [[FPEXT1:%[0-9]+]]:_(s32) = G_FPEXT [[UV4]](s16)
+    ; GFX9: [[FPEXT2:%[0-9]+]]:_(s32) = G_FPEXT [[UV8]](s16)
+    ; GFX9: [[FMA:%[0-9]+]]:_(s32) = G_FMA [[FPEXT]], [[FPEXT1]], [[FPEXT2]]
+    ; GFX9: [[FPTRUNC:%[0-9]+]]:_(s16) = G_FPTRUNC [[FMA]](s32)
+    ; GFX9: [[FPEXT3:%[0-9]+]]:_(s32) = G_FPEXT [[UV1]](s16)
+    ; GFX9: [[FPEXT4:%[0-9]+]]:_(s32) = G_FPEXT [[UV5]](s16)
+    ; GFX9: [[FPEXT5:%[0-9]+]]:_(s32) = G_FPEXT [[UV9]](s16)
+    ; GFX9: [[FMA1:%[0-9]+]]:_(s32) = G_FMA [[FPEXT3]], [[FPEXT4]], [[FPEXT5]]
+    ; GFX9: [[FPTRUNC1:%[0-9]+]]:_(s16) = G_FPTRUNC [[FMA1]](s32)
+    ; GFX9: [[FPEXT6:%[0-9]+]]:_(s32) = G_FPEXT [[UV2]](s16)
+    ; GFX9: [[FPEXT7:%[0-9]+]]:_(s32) = G_FPEXT [[UV6]](s16)
+    ; GFX9: [[FPEXT8:%[0-9]+]]:_(s32) = G_FPEXT [[UV10]](s16)
+    ; GFX9: [[FMA2:%[0-9]+]]:_(s32) = G_FMA [[FPEXT6]], [[FPEXT7]], [[FPEXT8]]
+    ; GFX9: [[FPTRUNC2:%[0-9]+]]:_(s16) = G_FPTRUNC [[FMA2]](s32)
+    ; GFX9: [[FPEXT9:%[0-9]+]]:_(s32) = G_FPEXT [[UV3]](s16)
+    ; GFX9: [[FPEXT10:%[0-9]+]]:_(s32) = G_FPEXT [[UV7]](s16)
+    ; GFX9: [[FPEXT11:%[0-9]+]]:_(s32) = G_FPEXT [[UV11]](s16)
+    ; GFX9: [[FMA3:%[0-9]+]]:_(s32) = G_FMA [[FPEXT9]], [[FPEXT10]], [[FPEXT11]]
+    ; GFX9: [[FPTRUNC3:%[0-9]+]]:_(s16) = G_FPTRUNC [[FMA3]](s32)
+    ; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s16>) = G_BUILD_VECTOR [[FPTRUNC]](s16), [[FPTRUNC1]](s16), [[FPTRUNC2]](s16), [[FPTRUNC3]](s16)
+    ; GFX9: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<4 x s16>)
+    %0:_(<4 x s16>) = COPY $vgpr0_vgpr1
+    %1:_(<4 x s16>) = COPY $vgpr2_vgpr3
+    %2:_(<4 x s16>) = COPY $vgpr4_vgpr5
+    %3:_(<4 x s16>) = G_FMA %0, %1, %2
+    $vgpr0_vgpr1 = COPY %3
+...
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-fmul.mir b/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-fmul.mir
index 612110ac486..f0313e1f11d 100644
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-fmul.mir
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-fmul.mir
@@ -104,3 +104,369 @@ body: |
     %5:_(s32) = G_ANYEXT %4
     $vgpr0 = COPY %5
 ...
+
+---
+name: test_fmul_v2s32
+body: |
+  bb.0:
+    liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
+
+    ; SI-LABEL: name: test_fmul_v2s32
+    ; SI: [[COPY:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr0_vgpr1
+    ; SI: [[COPY1:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr2_vgpr3
+    ; SI: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](<2 x s32>)
+    ; SI: [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY1]](<2 x s32>)
+    ; SI: [[FMUL:%[0-9]+]]:_(s32) = G_FMUL [[UV]], [[UV2]]
+    ; SI: [[FMUL1:%[0-9]+]]:_(s32) = G_FMUL [[UV1]], [[UV3]]
+    ; SI: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[FMUL]](s32), [[FMUL1]](s32)
+    ; SI: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<2 x s32>)
+    ; VI-LABEL: name: test_fmul_v2s32
+    ; VI: [[COPY:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr0_vgpr1
+    ; VI: [[COPY1:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr2_vgpr3
+    ; VI: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](<2 x s32>)
+    ; VI: [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY1]](<2 x s32>)
+    ; VI: [[FMUL:%[0-9]+]]:_(s32) = G_FMUL [[UV]], [[UV2]]
+    ; VI: [[FMUL1:%[0-9]+]]:_(s32) = G_FMUL [[UV1]], [[UV3]]
+    ; VI: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[FMUL]](s32), [[FMUL1]](s32)
+    ; VI: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<2 x s32>)
+    ; GFX9-LABEL: name: test_fmul_v2s32
+    ; GFX9: [[COPY:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr0_vgpr1
+    ; GFX9: [[COPY1:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr2_vgpr3
+    ; GFX9: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](<2 x s32>)
+    ; GFX9: [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY1]](<2 x s32>)
+    ; GFX9: [[FMUL:%[0-9]+]]:_(s32) = G_FMUL [[UV]], [[UV2]]
+    ; GFX9: [[FMUL1:%[0-9]+]]:_(s32) = G_FMUL [[UV1]], [[UV3]]
+    ; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[FMUL]](s32), [[FMUL1]](s32)
+    ; GFX9: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<2 x s32>)
+    %0:_(<2 x s32>) = COPY $vgpr0_vgpr1
+    %1:_(<2 x s32>) = COPY $vgpr2_vgpr3
+    %2:_(<2 x s32>) = G_FMUL %0, %1
+    $vgpr0_vgpr1 = COPY %2
+...
+
+---
+name: test_fmul_v2s32_flags
+body: |
+  bb.0:
+    liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
+
+    ; SI-LABEL: name: test_fmul_v2s32_flags
+    ; SI: [[COPY:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr0_vgpr1
+    ; SI: [[COPY1:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr2_vgpr3
+    ; SI: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](<2 x s32>)
+    ; SI: [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY1]](<2 x s32>)
+    ; SI: %7:_(s32) = nnan G_FMUL [[UV]], [[UV2]]
+    ; SI: %8:_(s32) = nnan G_FMUL [[UV1]], [[UV3]]
+    ; SI: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR %7(s32), %8(s32)
+    ; SI: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<2 x s32>)
+    ; VI-LABEL: name: test_fmul_v2s32_flags
+    ; VI: [[COPY:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr0_vgpr1
+    ; VI: [[COPY1:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr2_vgpr3
+    ; VI: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](<2 x s32>)
+    ; VI: [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY1]](<2 x s32>)
+    ; VI: %7:_(s32) = nnan G_FMUL [[UV]], [[UV2]]
+    ; VI: %8:_(s32) = nnan G_FMUL [[UV1]], [[UV3]]
+    ; VI: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR %7(s32), %8(s32)
+    ; VI: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<2 x s32>)
+    ; GFX9-LABEL: name: test_fmul_v2s32_flags
+    ; GFX9: [[COPY:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr0_vgpr1
+    ; GFX9: [[COPY1:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr2_vgpr3
+    ; GFX9: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](<2 x s32>)
+    ; GFX9: [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY1]](<2 x s32>)
+    ; GFX9: %7:_(s32) = nnan G_FMUL [[UV]], [[UV2]]
+    ; GFX9: %8:_(s32) = nnan G_FMUL [[UV1]], [[UV3]]
+    ; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR %7(s32), %8(s32)
+    ; GFX9: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<2 x s32>)
+    %0:_(<2 x s32>) = COPY $vgpr0_vgpr1
+    %1:_(<2 x s32>) = COPY $vgpr2_vgpr3
+    %2:_(<2 x s32>) = nnan G_FMUL %0, %1
+    $vgpr0_vgpr1 = COPY %2
+...
+
+---
+name: test_fmul_v3s32
+body: |
+  bb.0:
+    liveins: $vgpr0_vgpr1_vgpr2, $vgpr3_vgpr4_vgpr5
+
+    ; SI-LABEL: name: test_fmul_v3s32
+    ; SI: [[COPY:%[0-9]+]]:_(<3 x s32>) = COPY $vgpr0_vgpr1_vgpr2
+    ; SI: [[COPY1:%[0-9]+]]:_(<3 x s32>) = COPY $vgpr3_vgpr4_vgpr5
+    ; SI: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](<3 x s32>)
+    ; SI: [[UV3:%[0-9]+]]:_(s32), [[UV4:%[0-9]+]]:_(s32), [[UV5:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY1]](<3 x s32>)
+    ; SI: [[FMUL:%[0-9]+]]:_(s32) = G_FMUL [[UV]], [[UV3]]
+    ; SI: [[FMUL1:%[0-9]+]]:_(s32) = G_FMUL [[UV1]], [[UV4]]
+    ; SI: [[FMUL2:%[0-9]+]]:_(s32) = G_FMUL [[UV2]], [[UV5]]
+    ; SI: [[BUILD_VECTOR:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[FMUL]](s32), [[FMUL1]](s32), [[FMUL2]](s32)
+    ; SI: $vgpr0_vgpr1_vgpr2 = COPY [[BUILD_VECTOR]](<3 x s32>)
+    ; VI-LABEL: name: test_fmul_v3s32
+    ; VI: [[COPY:%[0-9]+]]:_(<3 x s32>) = COPY $vgpr0_vgpr1_vgpr2
+    ; VI: [[COPY1:%[0-9]+]]:_(<3 x s32>) = COPY $vgpr3_vgpr4_vgpr5
+    ; VI: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](<3 x s32>)
+    ; VI: [[UV3:%[0-9]+]]:_(s32), [[UV4:%[0-9]+]]:_(s32), [[UV5:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY1]](<3 x s32>)
+    ; VI: [[FMUL:%[0-9]+]]:_(s32) = G_FMUL [[UV]], [[UV3]]
+    ; VI: [[FMUL1:%[0-9]+]]:_(s32) = G_FMUL [[UV1]], [[UV4]]
+    ; VI: [[FMUL2:%[0-9]+]]:_(s32) = G_FMUL [[UV2]], [[UV5]]
+    ; VI: [[BUILD_VECTOR:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[FMUL]](s32), [[FMUL1]](s32), [[FMUL2]](s32)
+    ; VI: $vgpr0_vgpr1_vgpr2 = COPY [[BUILD_VECTOR]](<3 x s32>)
+    ; GFX9-LABEL: name: test_fmul_v3s32
+    ; GFX9: [[COPY:%[0-9]+]]:_(<3 x s32>) = COPY $vgpr0_vgpr1_vgpr2
+    ; GFX9: [[COPY1:%[0-9]+]]:_(<3 x s32>) = COPY $vgpr3_vgpr4_vgpr5
+    ; GFX9: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](<3 x s32>)
+    ; GFX9: [[UV3:%[0-9]+]]:_(s32), [[UV4:%[0-9]+]]:_(s32), [[UV5:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY1]](<3 x s32>)
+    ; GFX9: [[FMUL:%[0-9]+]]:_(s32) = G_FMUL [[UV]], [[UV3]]
+    ; GFX9: [[FMUL1:%[0-9]+]]:_(s32) = G_FMUL [[UV1]], [[UV4]]
+    ; GFX9: [[FMUL2:%[0-9]+]]:_(s32) = G_FMUL [[UV2]], [[UV5]]
+    ; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[FMUL]](s32), [[FMUL1]](s32), [[FMUL2]](s32)
+    ; GFX9: $vgpr0_vgpr1_vgpr2 = COPY [[BUILD_VECTOR]](<3 x s32>)
+    %0:_(<3 x s32>) = COPY $vgpr0_vgpr1_vgpr2
+    %1:_(<3 x s32>) = COPY $vgpr3_vgpr4_vgpr5
+    %2:_(<3 x s32>) = G_FMUL %0, %1
+    $vgpr0_vgpr1_vgpr2 = COPY %2
+...
+
+---
+name: test_fmul_v2s64
+body: |
+  bb.0:
+    liveins: $vgpr0_vgpr1_vgpr2_vgpr3, $vgpr4_vgpr5_vgpr6_vgpr7
+
+    ; SI-LABEL: name: test_fmul_v2s64
+    ; SI: [[COPY:%[0-9]+]]:_(<2 x s64>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3
+    ; SI: [[COPY1:%[0-9]+]]:_(<2 x s64>) = COPY $vgpr4_vgpr5_vgpr6_vgpr7
+    ; SI: [[UV:%[0-9]+]]:_(s64), [[UV1:%[0-9]+]]:_(s64) = G_UNMERGE_VALUES [[COPY]](<2 x s64>)
+    ; SI: [[UV2:%[0-9]+]]:_(s64), [[UV3:%[0-9]+]]:_(s64) = G_UNMERGE_VALUES [[COPY1]](<2 x s64>)
+    ; SI: [[FMUL:%[0-9]+]]:_(s64) = G_FMUL [[UV]], [[UV2]]
+    ; SI: [[FMUL1:%[0-9]+]]:_(s64) = G_FMUL [[UV1]], [[UV3]]
+    ; SI: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s64>) = G_BUILD_VECTOR [[FMUL]](s64), [[FMUL1]](s64)
+    ; SI: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[BUILD_VECTOR]](<2 x s64>)
+    ; VI-LABEL: name: test_fmul_v2s64
+    ; VI: [[COPY:%[0-9]+]]:_(<2 x s64>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3
+    ; VI: [[COPY1:%[0-9]+]]:_(<2 x s64>) = COPY $vgpr4_vgpr5_vgpr6_vgpr7
+    ; VI: [[UV:%[0-9]+]]:_(s64), [[UV1:%[0-9]+]]:_(s64) = G_UNMERGE_VALUES [[COPY]](<2 x s64>)
+    ; VI: [[UV2:%[0-9]+]]:_(s64), [[UV3:%[0-9]+]]:_(s64) = G_UNMERGE_VALUES [[COPY1]](<2 x s64>)
+    ; VI: [[FMUL:%[0-9]+]]:_(s64) = G_FMUL [[UV]], [[UV2]]
+    ; VI: [[FMUL1:%[0-9]+]]:_(s64) = G_FMUL [[UV1]], [[UV3]]
+    ; VI: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s64>) = G_BUILD_VECTOR [[FMUL]](s64), [[FMUL1]](s64)
+    ; VI: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[BUILD_VECTOR]](<2 x s64>)
+    ; GFX9-LABEL: name: test_fmul_v2s64
+    ; GFX9: [[COPY:%[0-9]+]]:_(<2 x s64>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3
+    ; GFX9: [[COPY1:%[0-9]+]]:_(<2 x s64>) = COPY $vgpr4_vgpr5_vgpr6_vgpr7
+    ; GFX9: [[UV:%[0-9]+]]:_(s64), [[UV1:%[0-9]+]]:_(s64) = G_UNMERGE_VALUES [[COPY]](<2 x s64>)
+    ; GFX9: [[UV2:%[0-9]+]]:_(s64), [[UV3:%[0-9]+]]:_(s64) = G_UNMERGE_VALUES [[COPY1]](<2 x s64>)
+    ; GFX9: [[FMUL:%[0-9]+]]:_(s64) = G_FMUL [[UV]], [[UV2]]
+    ; GFX9: [[FMUL1:%[0-9]+]]:_(s64) = G_FMUL [[UV1]], [[UV3]]
+    ; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s64>) = G_BUILD_VECTOR [[FMUL]](s64), [[FMUL1]](s64)
+    ; GFX9: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[BUILD_VECTOR]](<2 x s64>)
+    %0:_(<2 x s64>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3
+    %1:_(<2 x s64>) = COPY $vgpr4_vgpr5_vgpr6_vgpr7
+    %2:_(<2 x s64>) = G_FMUL %0, %1
+    $vgpr0_vgpr1_vgpr2_vgpr3 = COPY %2
+...
+
+---
+name: test_fmul_v2s16
+body: |
+  bb.0:
+    liveins: $vgpr0, $vgpr1
+
+    ; SI-LABEL: name: test_fmul_v2s16
+    ; SI: [[COPY:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0
+    ; SI: [[COPY1:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr1
+    ; SI: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY]](<2 x s16>)
+    ; SI: [[UV2:%[0-9]+]]:_(s16), [[UV3:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY1]](<2 x s16>)
+    ; SI: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[UV]](s16)
+    ; SI: [[FPEXT1:%[0-9]+]]:_(s32) = G_FPEXT [[UV2]](s16)
+    ; SI: [[FMUL:%[0-9]+]]:_(s32) = G_FMUL [[FPEXT]], [[FPEXT1]]
+    ; SI: [[FPTRUNC:%[0-9]+]]:_(s16) = G_FPTRUNC [[FMUL]](s32)
+    ; SI: [[FPEXT2:%[0-9]+]]:_(s32) = G_FPEXT [[UV1]](s16)
+    ; SI: [[FPEXT3:%[0-9]+]]:_(s32) = G_FPEXT [[UV3]](s16)
+    ; SI: [[FMUL1:%[0-9]+]]:_(s32) = G_FMUL [[FPEXT2]], [[FPEXT3]]
+    ; SI: [[FPTRUNC1:%[0-9]+]]:_(s16) = G_FPTRUNC [[FMUL1]](s32)
+    ; SI: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR [[FPTRUNC]](s16), [[FPTRUNC1]](s16)
+    ; SI: $vgpr0 = COPY [[BUILD_VECTOR]](<2 x s16>)
+    ; VI-LABEL: name: test_fmul_v2s16
+    ; VI: [[COPY:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0
+    ; VI: [[COPY1:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr1
+    ; VI: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY]](<2 x s16>)
+    ; VI: [[UV2:%[0-9]+]]:_(s16), [[UV3:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY1]](<2 x s16>)
+    ; VI: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[UV]](s16)
+    ; VI: [[FPEXT1:%[0-9]+]]:_(s32) = G_FPEXT [[UV2]](s16)
+    ; VI: [[FMUL:%[0-9]+]]:_(s32) = G_FMUL [[FPEXT]], [[FPEXT1]]
+    ; VI: [[FPTRUNC:%[0-9]+]]:_(s16) = G_FPTRUNC [[FMUL]](s32)
+    ; VI: [[FPEXT2:%[0-9]+]]:_(s32) = G_FPEXT [[UV1]](s16)
+    ; VI: [[FPEXT3:%[0-9]+]]:_(s32) = G_FPEXT [[UV3]](s16)
+    ; VI: [[FMUL1:%[0-9]+]]:_(s32) = G_FMUL [[FPEXT2]], [[FPEXT3]]
+    ; VI: [[FPTRUNC1:%[0-9]+]]:_(s16) = G_FPTRUNC [[FMUL1]](s32)
+    ; VI: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR [[FPTRUNC]](s16), [[FPTRUNC1]](s16)
+    ; VI: $vgpr0 = COPY [[BUILD_VECTOR]](<2 x s16>)
+    ; GFX9-LABEL: name: test_fmul_v2s16
+    ; GFX9: [[COPY:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0
+    ; GFX9: [[COPY1:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr1
+    ; GFX9: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY]](<2 x s16>)
+    ; GFX9: [[UV2:%[0-9]+]]:_(s16), [[UV3:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY1]](<2 x s16>)
+    ; GFX9: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[UV]](s16)
+    ; GFX9: [[FPEXT1:%[0-9]+]]:_(s32) = G_FPEXT [[UV2]](s16)
+    ; GFX9: [[FMUL:%[0-9]+]]:_(s32) = G_FMUL [[FPEXT]], [[FPEXT1]]
+    ; GFX9: [[FPTRUNC:%[0-9]+]]:_(s16) = G_FPTRUNC [[FMUL]](s32)
+    ; GFX9: [[FPEXT2:%[0-9]+]]:_(s32) = G_FPEXT [[UV1]](s16)
+    ; GFX9: [[FPEXT3:%[0-9]+]]:_(s32) = G_FPEXT [[UV3]](s16)
+    ; GFX9: [[FMUL1:%[0-9]+]]:_(s32) = G_FMUL [[FPEXT2]], [[FPEXT3]]
+    ; GFX9: [[FPTRUNC1:%[0-9]+]]:_(s16) = G_FPTRUNC [[FMUL1]](s32)
+    ; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR [[FPTRUNC]](s16), [[FPTRUNC1]](s16)
+    ; GFX9: $vgpr0 = COPY [[BUILD_VECTOR]](<2 x s16>)
+    %0:_(<2 x s16>) = COPY $vgpr0
+    %1:_(<2 x s16>) = COPY $vgpr1
+    %2:_(<2 x s16>) = G_FMUL %0, %1
+    $vgpr0 = COPY %2
+...
+
+---
+name: test_fmul_v3s16
+body: |
+  bb.0:
+    liveins: $vgpr0, $vgpr1
+
+    ; SI-LABEL: name: test_fmul_v3s16
+    ; SI: [[DEF:%[0-9]+]]:_(<3 x s16>) = G_IMPLICIT_DEF
+    ; SI: [[DEF1:%[0-9]+]]:_(<3 x s16>) = G_IMPLICIT_DEF
+    ; SI: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[DEF]](<3 x s16>)
+    ; SI: [[UV3:%[0-9]+]]:_(s16), [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[DEF1]](<3 x s16>)
+    ; SI: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[UV]](s16)
+    ; SI: [[FPEXT1:%[0-9]+]]:_(s32) = G_FPEXT [[UV3]](s16)
+    ; SI: [[FMUL:%[0-9]+]]:_(s32) = G_FMUL [[FPEXT]], [[FPEXT1]]
+    ; SI: [[FPTRUNC:%[0-9]+]]:_(s16) = G_FPTRUNC [[FMUL]](s32)
+    ; SI: [[FPEXT2:%[0-9]+]]:_(s32) = G_FPEXT [[UV1]](s16)
+    ; SI: [[FPEXT3:%[0-9]+]]:_(s32) = G_FPEXT [[UV4]](s16)
+    ; SI: [[FMUL1:%[0-9]+]]:_(s32) = G_FMUL [[FPEXT2]], [[FPEXT3]]
+    ; SI: [[FPTRUNC1:%[0-9]+]]:_(s16) = G_FPTRUNC [[FMUL1]](s32)
+    ; SI: [[FPEXT4:%[0-9]+]]:_(s32) = G_FPEXT [[UV2]](s16)
+    ; SI: [[FPEXT5:%[0-9]+]]:_(s32) = G_FPEXT [[UV5]](s16)
+    ; SI: [[FMUL2:%[0-9]+]]:_(s32) = G_FMUL [[FPEXT4]], [[FPEXT5]]
+    ; SI: [[FPTRUNC2:%[0-9]+]]:_(s16) = G_FPTRUNC [[FMUL2]](s32)
+    ; SI: [[BUILD_VECTOR:%[0-9]+]]:_(<3 x s16>) = G_BUILD_VECTOR [[FPTRUNC]](s16), [[FPTRUNC1]](s16), [[FPTRUNC2]](s16)
+    ; SI: S_NOP 0, implicit [[BUILD_VECTOR]](<3 x s16>)
+    ; VI-LABEL: name: test_fmul_v3s16
+    ; VI: [[DEF:%[0-9]+]]:_(<3 x s16>) = G_IMPLICIT_DEF
+    ; VI: [[DEF1:%[0-9]+]]:_(<3 x s16>) = G_IMPLICIT_DEF
+    ; VI: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[DEF]](<3 x s16>)
+    ; VI: [[UV3:%[0-9]+]]:_(s16), [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[DEF1]](<3 x s16>)
+    ; VI: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[UV]](s16)
+    ; VI: [[FPEXT1:%[0-9]+]]:_(s32) = G_FPEXT [[UV3]](s16)
+    ; VI: [[FMUL:%[0-9]+]]:_(s32) = G_FMUL [[FPEXT]], [[FPEXT1]]
+    ; VI: [[FPTRUNC:%[0-9]+]]:_(s16) = G_FPTRUNC [[FMUL]](s32)
+    ; VI: [[FPEXT2:%[0-9]+]]:_(s32) = G_FPEXT [[UV1]](s16)
+    ; VI: [[FPEXT3:%[0-9]+]]:_(s32) = G_FPEXT [[UV4]](s16)
+    ; VI: [[FMUL1:%[0-9]+]]:_(s32) = G_FMUL [[FPEXT2]], [[FPEXT3]]
+    ; VI: [[FPTRUNC1:%[0-9]+]]:_(s16) = G_FPTRUNC [[FMUL1]](s32)
+    ; VI: [[FPEXT4:%[0-9]+]]:_(s32) = G_FPEXT [[UV2]](s16)
+    ; VI: [[FPEXT5:%[0-9]+]]:_(s32) = G_FPEXT [[UV5]](s16)
+    ; VI: [[FMUL2:%[0-9]+]]:_(s32) = G_FMUL [[FPEXT4]], [[FPEXT5]]
+    ; VI: [[FPTRUNC2:%[0-9]+]]:_(s16) = G_FPTRUNC [[FMUL2]](s32)
+    ; VI: [[BUILD_VECTOR:%[0-9]+]]:_(<3 x s16>) = G_BUILD_VECTOR [[FPTRUNC]](s16), [[FPTRUNC1]](s16), [[FPTRUNC2]](s16)
+    ; VI: S_NOP 0, implicit [[BUILD_VECTOR]](<3 x s16>)
+    ; GFX9-LABEL: name: test_fmul_v3s16
+    ; GFX9: [[DEF:%[0-9]+]]:_(<3 x s16>) = G_IMPLICIT_DEF
+    ; GFX9: [[DEF1:%[0-9]+]]:_(<3 x s16>) = G_IMPLICIT_DEF
+    ; GFX9: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[DEF]](<3 x s16>)
+    ; GFX9: [[UV3:%[0-9]+]]:_(s16), [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[DEF1]](<3 x s16>)
+    ; GFX9: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[UV]](s16)
+    ; GFX9: [[FPEXT1:%[0-9]+]]:_(s32) = G_FPEXT [[UV3]](s16)
+    ; GFX9: [[FMUL:%[0-9]+]]:_(s32) = G_FMUL [[FPEXT]], [[FPEXT1]]
+    ; GFX9: [[FPTRUNC:%[0-9]+]]:_(s16) = G_FPTRUNC [[FMUL]](s32)
+    ; GFX9: [[FPEXT2:%[0-9]+]]:_(s32) = G_FPEXT [[UV1]](s16)
+    ; GFX9: [[FPEXT3:%[0-9]+]]:_(s32) = G_FPEXT [[UV4]](s16)
+    ; GFX9: [[FMUL1:%[0-9]+]]:_(s32) = G_FMUL [[FPEXT2]], [[FPEXT3]]
+    ; GFX9: [[FPTRUNC1:%[0-9]+]]:_(s16) = G_FPTRUNC [[FMUL1]](s32)
+    ; GFX9: [[FPEXT4:%[0-9]+]]:_(s32) = G_FPEXT [[UV2]](s16)
+    ; GFX9: [[FPEXT5:%[0-9]+]]:_(s32) = G_FPEXT [[UV5]](s16)
+    ; GFX9: [[FMUL2:%[0-9]+]]:_(s32) = G_FMUL [[FPEXT4]], [[FPEXT5]]
+    ; GFX9: [[FPTRUNC2:%[0-9]+]]:_(s16) = G_FPTRUNC [[FMUL2]](s32)
+    ; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<3 x s16>) = G_BUILD_VECTOR [[FPTRUNC]](s16), [[FPTRUNC1]](s16), [[FPTRUNC2]](s16)
+    ; GFX9: S_NOP 0, implicit [[BUILD_VECTOR]](<3 x s16>)
+    %0:_(<3 x s16>) = G_IMPLICIT_DEF
+    %1:_(<3 x s16>) = G_IMPLICIT_DEF
+    %2:_(<3 x s16>) = G_FMUL %0, %1
+    S_NOP 0, implicit %2
+...
+
+---
+name: test_fmul_v4s16
+body: |
+  bb.0:
+    liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
+
+    ; SI-LABEL: name: test_fmul_v4s16
+    ; SI: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1
+    ; SI: [[COPY1:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr2_vgpr3
+    ; SI: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16), [[UV3:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY]](<4 x s16>)
+    ; SI: [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16), [[UV6:%[0-9]+]]:_(s16), [[UV7:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY1]](<4 x s16>)
+    ; SI: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[UV]](s16)
+    ; SI: [[FPEXT1:%[0-9]+]]:_(s32) = G_FPEXT [[UV4]](s16)
+    ; SI: [[FMUL:%[0-9]+]]:_(s32) = G_FMUL [[FPEXT]], [[FPEXT1]]
+    ; SI: [[FPTRUNC:%[0-9]+]]:_(s16) = G_FPTRUNC [[FMUL]](s32)
+    ; SI: [[FPEXT2:%[0-9]+]]:_(s32) = G_FPEXT [[UV1]](s16)
+    ; SI: [[FPEXT3:%[0-9]+]]:_(s32) = G_FPEXT [[UV5]](s16)
+    ; SI: [[FMUL1:%[0-9]+]]:_(s32) = G_FMUL [[FPEXT2]], [[FPEXT3]]
+    ; SI: [[FPTRUNC1:%[0-9]+]]:_(s16) = G_FPTRUNC [[FMUL1]](s32)
+    ; SI: [[FPEXT4:%[0-9]+]]:_(s32) = G_FPEXT [[UV2]](s16)
+    ; SI: [[FPEXT5:%[0-9]+]]:_(s32) = G_FPEXT [[UV6]](s16)
+    ; SI: [[FMUL2:%[0-9]+]]:_(s32) = G_FMUL [[FPEXT4]], [[FPEXT5]]
+    ; SI: [[FPTRUNC2:%[0-9]+]]:_(s16) = G_FPTRUNC [[FMUL2]](s32)
+    ; SI: [[FPEXT6:%[0-9]+]]:_(s32) = G_FPEXT [[UV3]](s16)
+    ; SI: [[FPEXT7:%[0-9]+]]:_(s32) = G_FPEXT [[UV7]](s16)
+    ; SI: [[FMUL3:%[0-9]+]]:_(s32) = G_FMUL [[FPEXT6]], [[FPEXT7]]
+    ; SI: [[FPTRUNC3:%[0-9]+]]:_(s16) = G_FPTRUNC [[FMUL3]](s32)
+    ; SI: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s16>) = G_BUILD_VECTOR [[FPTRUNC]](s16), [[FPTRUNC1]](s16), [[FPTRUNC2]](s16), [[FPTRUNC3]](s16)
+    ; SI: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<4 x s16>)
+    ; VI-LABEL: name: test_fmul_v4s16
+    ; VI: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1
+    ; VI: [[COPY1:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr2_vgpr3
+    ; VI: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16), [[UV3:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY]](<4 x s16>)
+    ; VI: [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16), [[UV6:%[0-9]+]]:_(s16), [[UV7:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY1]](<4 x s16>)
+    ; VI: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[UV]](s16)
+    ; VI: [[FPEXT1:%[0-9]+]]:_(s32) = G_FPEXT [[UV4]](s16)
+    ; VI: [[FMUL:%[0-9]+]]:_(s32) = G_FMUL [[FPEXT]], [[FPEXT1]]
+    ; VI: [[FPTRUNC:%[0-9]+]]:_(s16) = G_FPTRUNC [[FMUL]](s32)
+    ; VI: [[FPEXT2:%[0-9]+]]:_(s32) = G_FPEXT [[UV1]](s16)
+    ; VI: [[FPEXT3:%[0-9]+]]:_(s32) = G_FPEXT [[UV5]](s16)
+    ; VI: [[FMUL1:%[0-9]+]]:_(s32) = G_FMUL [[FPEXT2]], [[FPEXT3]]
+    ; VI: [[FPTRUNC1:%[0-9]+]]:_(s16) = G_FPTRUNC [[FMUL1]](s32)
+    ; VI: [[FPEXT4:%[0-9]+]]:_(s32) = G_FPEXT [[UV2]](s16)
+    ; VI: [[FPEXT5:%[0-9]+]]:_(s32) = G_FPEXT [[UV6]](s16)
+    ; VI: [[FMUL2:%[0-9]+]]:_(s32) = G_FMUL [[FPEXT4]], [[FPEXT5]]
+    ; VI: [[FPTRUNC2:%[0-9]+]]:_(s16) = G_FPTRUNC [[FMUL2]](s32)
+    ; VI: [[FPEXT6:%[0-9]+]]:_(s32) = G_FPEXT [[UV3]](s16)
+    ; VI: [[FPEXT7:%[0-9]+]]:_(s32) = G_FPEXT [[UV7]](s16)
+    ; VI: [[FMUL3:%[0-9]+]]:_(s32) = G_FMUL [[FPEXT6]], [[FPEXT7]]
+    ; VI: [[FPTRUNC3:%[0-9]+]]:_(s16) = G_FPTRUNC [[FMUL3]](s32)
+    ; VI: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s16>) = G_BUILD_VECTOR [[FPTRUNC]](s16), [[FPTRUNC1]](s16), [[FPTRUNC2]](s16), [[FPTRUNC3]](s16)
+    ; VI: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<4 x s16>)
+    ; GFX9-LABEL: name: test_fmul_v4s16
+    ; GFX9: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1
+    ; GFX9: [[COPY1:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr2_vgpr3
+    ; GFX9: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16), [[UV3:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY]](<4 x s16>)
+    ; GFX9: [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16), [[UV6:%[0-9]+]]:_(s16), [[UV7:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY1]](<4 x s16>)
+    ; GFX9: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[UV]](s16)
+    ; GFX9: [[FPEXT1:%[0-9]+]]:_(s32) = G_FPEXT [[UV4]](s16)
+    ; GFX9: [[FMUL:%[0-9]+]]:_(s32) = G_FMUL [[FPEXT]], [[FPEXT1]]
+    ; GFX9: [[FPTRUNC:%[0-9]+]]:_(s16) = G_FPTRUNC [[FMUL]](s32)
+    ; GFX9: [[FPEXT2:%[0-9]+]]:_(s32) = G_FPEXT [[UV1]](s16)
+    ; GFX9: [[FPEXT3:%[0-9]+]]:_(s32) = G_FPEXT [[UV5]](s16)
+    ; GFX9: [[FMUL1:%[0-9]+]]:_(s32) = G_FMUL [[FPEXT2]], [[FPEXT3]]
+    ; GFX9: [[FPTRUNC1:%[0-9]+]]:_(s16) = G_FPTRUNC [[FMUL1]](s32)
+    ; GFX9: [[FPEXT4:%[0-9]+]]:_(s32) = G_FPEXT [[UV2]](s16)
+    ; GFX9: [[FPEXT5:%[0-9]+]]:_(s32) = G_FPEXT [[UV6]](s16)
+    ; GFX9: [[FMUL2:%[0-9]+]]:_(s32) = G_FMUL [[FPEXT4]], [[FPEXT5]]
+    ; GFX9: [[FPTRUNC2:%[0-9]+]]:_(s16) = G_FPTRUNC [[FMUL2]](s32)
+    ; GFX9: [[FPEXT6:%[0-9]+]]:_(s32) = G_FPEXT [[UV3]](s16)
+    ; GFX9: [[FPEXT7:%[0-9]+]]:_(s32) = G_FPEXT [[UV7]](s16)
+    ; GFX9: [[FMUL3:%[0-9]+]]:_(s32) = G_FMUL [[FPEXT6]], [[FPEXT7]]
+    ; GFX9: [[FPTRUNC3:%[0-9]+]]:_(s16) = G_FPTRUNC [[FMUL3]](s32)
+    ; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s16>) = G_BUILD_VECTOR [[FPTRUNC]](s16), [[FPTRUNC1]](s16), [[FPTRUNC2]](s16), [[FPTRUNC3]](s16)
+    ; GFX9: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<4 x s16>)
+    %0:_(<4 x s16>) = COPY $vgpr0_vgpr1
+    %1:_(<4 x s16>) = COPY $vgpr2_vgpr3
+    %2:_(<4 x s16>) = G_FMUL %0, %1
+    $vgpr0_vgpr1 = COPY %2
+...
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-fneg.mir b/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-fneg.mir
index e0f1484f628..3c5f3411419 100644
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-fneg.mir
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-fneg.mir
@@ -83,3 +83,260 @@ body: |
     %3:_(s32) = G_ANYEXT %2
     $vgpr0 = COPY %3
 ...
+
+---
+name: test_fneg_v2s32
+body: |
+  bb.0:
+    liveins: $vgpr0_vgpr1
+
+    ; SI-LABEL: name: test_fneg_v2s32
+    ; SI: [[COPY:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr0_vgpr1
+    ; SI: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](<2 x s32>)
+    ; SI: [[FNEG:%[0-9]+]]:_(s32) = G_FNEG [[UV]]
+    ; SI: [[FNEG1:%[0-9]+]]:_(s32) = G_FNEG [[UV1]]
+    ; SI: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[FNEG]](s32), [[FNEG1]](s32)
+    ; SI: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<2 x s32>)
+    ; VI-LABEL: name: test_fneg_v2s32
+    ; VI: [[COPY:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr0_vgpr1
+    ; VI: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](<2 x s32>)
+    ; VI: [[FNEG:%[0-9]+]]:_(s32) = G_FNEG [[UV]]
+    ; VI: [[FNEG1:%[0-9]+]]:_(s32) = G_FNEG [[UV1]]
+    ; VI: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[FNEG]](s32), [[FNEG1]](s32)
+    ; VI: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<2 x s32>)
+    ; GFX9-LABEL: name: test_fneg_v2s32
+    ; GFX9: [[COPY:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr0_vgpr1
+    ; GFX9: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](<2 x s32>)
+    ; GFX9: [[FNEG:%[0-9]+]]:_(s32) = G_FNEG [[UV]]
+    ; GFX9: [[FNEG1:%[0-9]+]]:_(s32) = G_FNEG [[UV1]]
+    ; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[FNEG]](s32), [[FNEG1]](s32)
+    ; GFX9: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<2 x s32>)
+    %0:_(<2 x s32>) = COPY $vgpr0_vgpr1
+    %1:_(<2 x s32>) = G_FNEG %0
+    $vgpr0_vgpr1 = COPY %1
+...
+
+---
+name: test_fneg_v3s32
+body: |
+  bb.0:
+    liveins: $vgpr0_vgpr1_vgpr2
+
+    ; SI-LABEL: name: test_fneg_v3s32
+    ; SI: [[COPY:%[0-9]+]]:_(<3 x s32>) = COPY $vgpr0_vgpr1_vgpr2
+    ; SI: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](<3 x s32>)
+    ; SI: [[FNEG:%[0-9]+]]:_(s32) = G_FNEG [[UV]]
+    ; SI: [[FNEG1:%[0-9]+]]:_(s32) = G_FNEG [[UV1]]
+    ; SI: [[FNEG2:%[0-9]+]]:_(s32) = G_FNEG [[UV2]]
+    ; SI: [[BUILD_VECTOR:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[FNEG]](s32), [[FNEG1]](s32), [[FNEG2]](s32)
+    ; SI: $vgpr0_vgpr1_vgpr2 = COPY [[BUILD_VECTOR]](<3 x s32>)
+    ; VI-LABEL: name: test_fneg_v3s32
+    ; VI: [[COPY:%[0-9]+]]:_(<3 x s32>) = COPY $vgpr0_vgpr1_vgpr2
+    ; VI: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](<3 x s32>)
+    ; VI: [[FNEG:%[0-9]+]]:_(s32) = G_FNEG [[UV]]
+    ; VI: [[FNEG1:%[0-9]+]]:_(s32) = G_FNEG [[UV1]]
+    ; VI: [[FNEG2:%[0-9]+]]:_(s32) = G_FNEG [[UV2]]
+    ; VI: [[BUILD_VECTOR:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[FNEG]](s32), [[FNEG1]](s32), [[FNEG2]](s32)
+    ; VI: $vgpr0_vgpr1_vgpr2 = COPY [[BUILD_VECTOR]](<3 x s32>)
+    ; GFX9-LABEL: name: test_fneg_v3s32
+    ; GFX9: [[COPY:%[0-9]+]]:_(<3 x s32>) = COPY $vgpr0_vgpr1_vgpr2
+    ; GFX9: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](<3 x s32>)
+    ; GFX9: [[FNEG:%[0-9]+]]:_(s32) = G_FNEG [[UV]]
+    ; GFX9: [[FNEG1:%[0-9]+]]:_(s32) = G_FNEG [[UV1]]
+    ; GFX9: [[FNEG2:%[0-9]+]]:_(s32) = G_FNEG [[UV2]]
+    ; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[FNEG]](s32), [[FNEG1]](s32), [[FNEG2]](s32)
+    ; GFX9: $vgpr0_vgpr1_vgpr2 = COPY [[BUILD_VECTOR]](<3 x s32>)
+    %0:_(<3 x s32>) = COPY $vgpr0_vgpr1_vgpr2
+    %1:_(<3 x  s32>) = G_FNEG %0
+    $vgpr0_vgpr1_vgpr2 = COPY %1
+...
+
+---
+name: test_fneg_v2s64
+body: |
+  bb.0:
+    liveins: $vgpr0_vgpr1_vgpr2_vgpr3
+
+    ; SI-LABEL: name: test_fneg_v2s64
+    ; SI: [[COPY:%[0-9]+]]:_(<2 x s64>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3
+    ; SI: [[UV:%[0-9]+]]:_(s64), [[UV1:%[0-9]+]]:_(s64) = G_UNMERGE_VALUES [[COPY]](<2 x s64>)
+    ; SI: [[FNEG:%[0-9]+]]:_(s64) = G_FNEG [[UV]]
+    ; SI: [[FNEG1:%[0-9]+]]:_(s64) = G_FNEG [[UV1]]
+    ; SI: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s64>) = G_BUILD_VECTOR [[FNEG]](s64), [[FNEG1]](s64)
+    ; SI: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[BUILD_VECTOR]](<2 x s64>)
+    ; VI-LABEL: name: test_fneg_v2s64
+    ; VI: [[COPY:%[0-9]+]]:_(<2 x s64>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3
+    ; VI: [[UV:%[0-9]+]]:_(s64), [[UV1:%[0-9]+]]:_(s64) = G_UNMERGE_VALUES [[COPY]](<2 x s64>)
+    ; VI: [[FNEG:%[0-9]+]]:_(s64) = G_FNEG [[UV]]
+    ; VI: [[FNEG1:%[0-9]+]]:_(s64) = G_FNEG [[UV1]]
+    ; VI: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s64>) = G_BUILD_VECTOR [[FNEG]](s64), [[FNEG1]](s64)
+    ; VI: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[BUILD_VECTOR]](<2 x s64>)
+    ; GFX9-LABEL: name: test_fneg_v2s64
+    ; GFX9: [[COPY:%[0-9]+]]:_(<2 x s64>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3
+    ; GFX9: [[UV:%[0-9]+]]:_(s64), [[UV1:%[0-9]+]]:_(s64) = G_UNMERGE_VALUES [[COPY]](<2 x s64>)
+    ; GFX9: [[FNEG:%[0-9]+]]:_(s64) = G_FNEG [[UV]]
+    ; GFX9: [[FNEG1:%[0-9]+]]:_(s64) = G_FNEG [[UV1]]
+    ; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s64>) = G_BUILD_VECTOR [[FNEG]](s64), [[FNEG1]](s64)
+    ; GFX9: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[BUILD_VECTOR]](<2 x s64>)
+    %0:_(<2 x s64>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3
+    %1:_(<2 x s64>) = G_FNEG %0
+    $vgpr0_vgpr1_vgpr2_vgpr3 = COPY %1
+...
+
+---
+name: test_fneg_v2s16
+body: |
+  bb.0:
+    liveins: $vgpr0
+
+    ; SI-LABEL: name: test_fneg_v2s16
+    ; SI: [[COPY:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0
+    ; SI: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY]](<2 x s16>)
+    ; SI: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[UV]](s16)
+    ; SI: [[FNEG:%[0-9]+]]:_(s32) = G_FNEG [[FPEXT]]
+    ; SI: [[FPTRUNC:%[0-9]+]]:_(s16) = G_FPTRUNC [[FNEG]](s32)
+    ; SI: [[FPEXT1:%[0-9]+]]:_(s32) = G_FPEXT [[UV1]](s16)
+    ; SI: [[FNEG1:%[0-9]+]]:_(s32) = G_FNEG [[FPEXT1]]
+    ; SI: [[FPTRUNC1:%[0-9]+]]:_(s16) = G_FPTRUNC [[FNEG1]](s32)
+    ; SI: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR [[FPTRUNC]](s16), [[FPTRUNC1]](s16)
+    ; SI: $vgpr0 = COPY [[BUILD_VECTOR]](<2 x s16>)
+    ; VI-LABEL: name: test_fneg_v2s16
+    ; VI: [[COPY:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0
+    ; VI: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY]](<2 x s16>)
+    ; VI: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[UV]](s16)
+    ; VI: [[FNEG:%[0-9]+]]:_(s32) = G_FNEG [[FPEXT]]
+    ; VI: [[FPTRUNC:%[0-9]+]]:_(s16) = G_FPTRUNC [[FNEG]](s32)
+    ; VI: [[FPEXT1:%[0-9]+]]:_(s32) = G_FPEXT [[UV1]](s16)
+    ; VI: [[FNEG1:%[0-9]+]]:_(s32) = G_FNEG [[FPEXT1]]
+    ; VI: [[FPTRUNC1:%[0-9]+]]:_(s16) = G_FPTRUNC [[FNEG1]](s32)
+    ; VI: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR [[FPTRUNC]](s16), [[FPTRUNC1]](s16)
+    ; VI: $vgpr0 = COPY [[BUILD_VECTOR]](<2 x s16>)
+    ; GFX9-LABEL: name: test_fneg_v2s16
+    ; GFX9: [[COPY:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0
+    ; GFX9: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY]](<2 x s16>)
+    ; GFX9: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[UV]](s16)
+    ; GFX9: [[FNEG:%[0-9]+]]:_(s32) = G_FNEG [[FPEXT]]
+    ; GFX9: [[FPTRUNC:%[0-9]+]]:_(s16) = G_FPTRUNC [[FNEG]](s32)
+    ; GFX9: [[FPEXT1:%[0-9]+]]:_(s32) = G_FPEXT [[UV1]](s16)
+    ; GFX9: [[FNEG1:%[0-9]+]]:_(s32) = G_FNEG [[FPEXT1]]
+    ; GFX9: [[FPTRUNC1:%[0-9]+]]:_(s16) = G_FPTRUNC [[FNEG1]](s32)
+    ; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR [[FPTRUNC]](s16), [[FPTRUNC1]](s16)
+    ; GFX9: $vgpr0 = COPY [[BUILD_VECTOR]](<2 x s16>)
+    %0:_(<2 x s16>) = COPY $vgpr0
+    %1:_(<2 x s16>) = G_FNEG %0
+    $vgpr0 = COPY %1
+...
+
+---
+name: test_fneg_v3s16
+body: |
+  bb.0:
+
+    ; SI-LABEL: name: test_fneg_v3s16
+    ; SI: [[DEF:%[0-9]+]]:_(<3 x s16>) = G_IMPLICIT_DEF
+    ; SI: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[DEF]](<3 x s16>)
+    ; SI: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[UV]](s16)
+    ; SI: [[FNEG:%[0-9]+]]:_(s32) = G_FNEG [[FPEXT]]
+    ; SI: [[FPTRUNC:%[0-9]+]]:_(s16) = G_FPTRUNC [[FNEG]](s32)
+    ; SI: [[FPEXT1:%[0-9]+]]:_(s32) = G_FPEXT [[UV1]](s16)
+    ; SI: [[FNEG1:%[0-9]+]]:_(s32) = G_FNEG [[FPEXT1]]
+    ; SI: [[FPTRUNC1:%[0-9]+]]:_(s16) = G_FPTRUNC [[FNEG1]](s32)
+    ; SI: [[FPEXT2:%[0-9]+]]:_(s32) = G_FPEXT [[UV2]](s16)
+    ; SI: [[FNEG2:%[0-9]+]]:_(s32) = G_FNEG [[FPEXT2]]
+    ; SI: [[FPTRUNC2:%[0-9]+]]:_(s16) = G_FPTRUNC [[FNEG2]](s32)
+    ; SI: [[BUILD_VECTOR:%[0-9]+]]:_(<3 x s16>) = G_BUILD_VECTOR [[FPTRUNC]](s16), [[FPTRUNC1]](s16), [[FPTRUNC2]](s16)
+    ; SI: S_NOP 0, implicit [[BUILD_VECTOR]](<3 x s16>)
+    ; VI-LABEL: name: test_fneg_v3s16
+    ; VI: [[DEF:%[0-9]+]]:_(<3 x s16>) = G_IMPLICIT_DEF
+    ; VI: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[DEF]](<3 x s16>)
+    ; VI: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[UV]](s16)
+    ; VI: [[FNEG:%[0-9]+]]:_(s32) = G_FNEG [[FPEXT]]
+    ; VI: [[FPTRUNC:%[0-9]+]]:_(s16) = G_FPTRUNC [[FNEG]](s32)
+    ; VI: [[FPEXT1:%[0-9]+]]:_(s32) = G_FPEXT [[UV1]](s16)
+    ; VI: [[FNEG1:%[0-9]+]]:_(s32) = G_FNEG [[FPEXT1]]
+    ; VI: [[FPTRUNC1:%[0-9]+]]:_(s16) = G_FPTRUNC [[FNEG1]](s32)
+    ; VI: [[FPEXT2:%[0-9]+]]:_(s32) = G_FPEXT [[UV2]](s16)
+    ; VI: [[FNEG2:%[0-9]+]]:_(s32) = G_FNEG [[FPEXT2]]
+    ; VI: [[FPTRUNC2:%[0-9]+]]:_(s16) = G_FPTRUNC [[FNEG2]](s32)
+    ; VI: [[BUILD_VECTOR:%[0-9]+]]:_(<3 x s16>) = G_BUILD_VECTOR [[FPTRUNC]](s16), [[FPTRUNC1]](s16), [[FPTRUNC2]](s16)
+    ; VI: S_NOP 0, implicit [[BUILD_VECTOR]](<3 x s16>)
+    ; GFX9-LABEL: name: test_fneg_v3s16
+    ; GFX9: [[DEF:%[0-9]+]]:_(<3 x s16>) = G_IMPLICIT_DEF
+    ; GFX9: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[DEF]](<3 x s16>)
+    ; GFX9: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[UV]](s16)
+    ; GFX9: [[FNEG:%[0-9]+]]:_(s32) = G_FNEG [[FPEXT]]
+    ; GFX9: [[FPTRUNC:%[0-9]+]]:_(s16) = G_FPTRUNC [[FNEG]](s32)
+    ; GFX9: [[FPEXT1:%[0-9]+]]:_(s32) = G_FPEXT [[UV1]](s16)
+    ; GFX9: [[FNEG1:%[0-9]+]]:_(s32) = G_FNEG [[FPEXT1]]
+    ; GFX9: [[FPTRUNC1:%[0-9]+]]:_(s16) = G_FPTRUNC [[FNEG1]](s32)
+    ; GFX9: [[FPEXT2:%[0-9]+]]:_(s32) = G_FPEXT [[UV2]](s16)
+    ; GFX9: [[FNEG2:%[0-9]+]]:_(s32) = G_FNEG [[FPEXT2]]
+    ; GFX9: [[FPTRUNC2:%[0-9]+]]:_(s16) = G_FPTRUNC [[FNEG2]](s32)
+    ; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<3 x s16>) = G_BUILD_VECTOR [[FPTRUNC]](s16), [[FPTRUNC1]](s16), [[FPTRUNC2]](s16)
+    ; GFX9: S_NOP 0, implicit [[BUILD_VECTOR]](<3 x s16>)
+    %0:_(<3 x s16>) = G_IMPLICIT_DEF
+    %1:_(<3 x s16>) = G_FNEG %0
+    S_NOP 0, implicit %1
+...
+
+---
+name: test_fneg_v4s16
+body: |
+  bb.0:
+    liveins: $vgpr0_vgpr1
+
+    ; SI-LABEL: name: test_fneg_v4s16
+    ; SI: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1
+    ; SI: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16), [[UV3:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY]](<4 x s16>)
+    ; SI: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[UV]](s16)
+    ; SI: [[FNEG:%[0-9]+]]:_(s32) = G_FNEG [[FPEXT]]
+    ; SI: [[FPTRUNC:%[0-9]+]]:_(s16) = G_FPTRUNC [[FNEG]](s32)
+    ; SI: [[FPEXT1:%[0-9]+]]:_(s32) = G_FPEXT [[UV1]](s16)
+    ; SI: [[FNEG1:%[0-9]+]]:_(s32) = G_FNEG [[FPEXT1]]
+    ; SI: [[FPTRUNC1:%[0-9]+]]:_(s16) = G_FPTRUNC [[FNEG1]](s32)
+    ; SI: [[FPEXT2:%[0-9]+]]:_(s32) = G_FPEXT [[UV2]](s16)
+    ; SI: [[FNEG2:%[0-9]+]]:_(s32) = G_FNEG [[FPEXT2]]
+    ; SI: [[FPTRUNC2:%[0-9]+]]:_(s16) = G_FPTRUNC [[FNEG2]](s32)
+    ; SI: [[FPEXT3:%[0-9]+]]:_(s32) = G_FPEXT [[UV3]](s16)
+    ; SI: [[FNEG3:%[0-9]+]]:_(s32) = G_FNEG [[FPEXT3]]
+    ; SI: [[FPTRUNC3:%[0-9]+]]:_(s16) = G_FPTRUNC [[FNEG3]](s32)
+    ; SI: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s16>) = G_BUILD_VECTOR [[FPTRUNC]](s16), [[FPTRUNC1]](s16), [[FPTRUNC2]](s16), [[FPTRUNC3]](s16)
+    ; SI: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<4 x s16>)
+    ; VI-LABEL: name: test_fneg_v4s16
+    ; VI: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1
+    ; VI: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16), [[UV3:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY]](<4 x s16>)
+    ; VI: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[UV]](s16)
+    ; VI: [[FNEG:%[0-9]+]]:_(s32) = G_FNEG [[FPEXT]]
+    ; VI: [[FPTRUNC:%[0-9]+]]:_(s16) = G_FPTRUNC [[FNEG]](s32)
+    ; VI: [[FPEXT1:%[0-9]+]]:_(s32) = G_FPEXT [[UV1]](s16)
+    ; VI: [[FNEG1:%[0-9]+]]:_(s32) = G_FNEG [[FPEXT1]]
+    ; VI: [[FPTRUNC1:%[0-9]+]]:_(s16) = G_FPTRUNC [[FNEG1]](s32)
+    ; VI: [[FPEXT2:%[0-9]+]]:_(s32) = G_FPEXT [[UV2]](s16)
+    ; VI: [[FNEG2:%[0-9]+]]:_(s32) = G_FNEG [[FPEXT2]]
+    ; VI: [[FPTRUNC2:%[0-9]+]]:_(s16) = G_FPTRUNC [[FNEG2]](s32)
+    ; VI: [[FPEXT3:%[0-9]+]]:_(s32) = G_FPEXT [[UV3]](s16)
+    ; VI: [[FNEG3:%[0-9]+]]:_(s32) = G_FNEG [[FPEXT3]]
+    ; VI: [[FPTRUNC3:%[0-9]+]]:_(s16) = G_FPTRUNC [[FNEG3]](s32)
+    ; VI: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s16>) = G_BUILD_VECTOR [[FPTRUNC]](s16), [[FPTRUNC1]](s16), [[FPTRUNC2]](s16), [[FPTRUNC3]](s16)
+    ; VI: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<4 x s16>)
+    ; GFX9-LABEL: name: test_fneg_v4s16
+    ; GFX9: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1
+    ; GFX9: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16), [[UV3:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY]](<4 x s16>)
+    ; GFX9: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[UV]](s16)
+    ; GFX9: [[FNEG:%[0-9]+]]:_(s32) = G_FNEG [[FPEXT]]
+    ; GFX9: [[FPTRUNC:%[0-9]+]]:_(s16) = G_FPTRUNC [[FNEG]](s32)
+    ; GFX9: [[FPEXT1:%[0-9]+]]:_(s32) = G_FPEXT [[UV1]](s16)
+    ; GFX9: [[FNEG1:%[0-9]+]]:_(s32) = G_FNEG [[FPEXT1]]
+    ; GFX9: [[FPTRUNC1:%[0-9]+]]:_(s16) = G_FPTRUNC [[FNEG1]](s32)
+    ; GFX9: [[FPEXT2:%[0-9]+]]:_(s32) = G_FPEXT [[UV2]](s16)
+    ; GFX9: [[FNEG2:%[0-9]+]]:_(s32) = G_FNEG [[FPEXT2]]
+    ; GFX9: [[FPTRUNC2:%[0-9]+]]:_(s16) = G_FPTRUNC [[FNEG2]](s32)
+    ; GFX9: [[FPEXT3:%[0-9]+]]:_(s32) = G_FPEXT [[UV3]](s16)
+    ; GFX9: [[FNEG3:%[0-9]+]]:_(s32) = G_FNEG [[FPEXT3]]
+    ; GFX9: [[FPTRUNC3:%[0-9]+]]:_(s16) = G_FPTRUNC [[FNEG3]](s32)
+    ; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s16>) = G_BUILD_VECTOR [[FPTRUNC]](s16), [[FPTRUNC1]](s16), [[FPTRUNC2]](s16), [[FPTRUNC3]](s16)
+    ; GFX9: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<4 x s16>)
+    %0:_(<4 x s16>) = COPY $vgpr0_vgpr1
+    %1:_(<4 x s16>) = G_FNEG %0
+    $vgpr0_vgpr1 = COPY %1
+...
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-fsub.mir b/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-fsub.mir
index 7a020aa82c2..3400c9f3581 100644
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-fsub.mir
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-fsub.mir
@@ -116,3 +116,456 @@ body: |
     %5:_(s32) = G_ANYEXT %4
     $vgpr0 = COPY %5
 ...
+
+---
+name: test_fsub_v2s32
+body: |
+  bb.0.entry:
+    liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
+
+    ; SI-LABEL: name: test_fsub_v2s32
+    ; SI: [[COPY:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr0_vgpr1
+    ; SI: [[COPY1:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr2_vgpr3
+    ; SI: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](<2 x s32>)
+    ; SI: [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY1]](<2 x s32>)
+    ; SI: [[FSUB:%[0-9]+]]:_(s32) = G_FSUB [[UV]], [[UV2]]
+    ; SI: [[FSUB1:%[0-9]+]]:_(s32) = G_FSUB [[UV1]], [[UV3]]
+    ; SI: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[FSUB]](s32), [[FSUB1]](s32)
+    ; SI: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<2 x s32>)
+    ; VI-LABEL: name: test_fsub_v2s32
+    ; VI: [[COPY:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr0_vgpr1
+    ; VI: [[COPY1:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr2_vgpr3
+    ; VI: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](<2 x s32>)
+    ; VI: [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY1]](<2 x s32>)
+    ; VI: [[FSUB:%[0-9]+]]:_(s32) = G_FSUB [[UV]], [[UV2]]
+    ; VI: [[FSUB1:%[0-9]+]]:_(s32) = G_FSUB [[UV1]], [[UV3]]
+    ; VI: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[FSUB]](s32), [[FSUB1]](s32)
+    ; VI: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<2 x s32>)
+    ; GFX9-LABEL: name: test_fsub_v2s32
+    ; GFX9: [[COPY:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr0_vgpr1
+    ; GFX9: [[COPY1:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr2_vgpr3
+    ; GFX9: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](<2 x s32>)
+    ; GFX9: [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY1]](<2 x s32>)
+    ; GFX9: [[FSUB:%[0-9]+]]:_(s32) = G_FSUB [[UV]], [[UV2]]
+    ; GFX9: [[FSUB1:%[0-9]+]]:_(s32) = G_FSUB [[UV1]], [[UV3]]
+    ; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[FSUB]](s32), [[FSUB1]](s32)
+    ; GFX9: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<2 x s32>)
+    %0:_(<2 x s32>) = COPY $vgpr0_vgpr1
+    %1:_(<2 x s32>) = COPY $vgpr2_vgpr3
+    %2:_(<2 x s32>) = G_FSUB %0, %1
+    $vgpr0_vgpr1 = COPY %2
+...
+
+---
+name: test_fsub_v2s32_flags
+body: |
+  bb.0.entry:
+    liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
+
+    ; SI-LABEL: name: test_fsub_v2s32_flags
+    ; SI: [[COPY:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr0_vgpr1
+    ; SI: [[COPY1:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr2_vgpr3
+    ; SI: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](<2 x s32>)
+    ; SI: [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY1]](<2 x s32>)
+    ; SI: %7:_(s32) = nnan G_FSUB [[UV]], [[UV2]]
+    ; SI: %8:_(s32) = nnan G_FSUB [[UV1]], [[UV3]]
+    ; SI: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR %7(s32), %8(s32)
+    ; SI: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<2 x s32>)
+    ; VI-LABEL: name: test_fsub_v2s32_flags
+    ; VI: [[COPY:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr0_vgpr1
+    ; VI: [[COPY1:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr2_vgpr3
+    ; VI: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](<2 x s32>)
+    ; VI: [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY1]](<2 x s32>)
+    ; VI: %7:_(s32) = nnan G_FSUB [[UV]], [[UV2]]
+    ; VI: %8:_(s32) = nnan G_FSUB [[UV1]], [[UV3]]
+    ; VI: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR %7(s32), %8(s32)
+    ; VI: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<2 x s32>)
+    ; GFX9-LABEL: name: test_fsub_v2s32_flags
+    ; GFX9: [[COPY:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr0_vgpr1
+    ; GFX9: [[COPY1:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr2_vgpr3
+    ; GFX9: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](<2 x s32>)
+    ; GFX9: [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY1]](<2 x s32>)
+    ; GFX9: %7:_(s32) = nnan G_FSUB [[UV]], [[UV2]]
+    ; GFX9: %8:_(s32) = nnan G_FSUB [[UV1]], [[UV3]]
+    ; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR %7(s32), %8(s32)
+    ; GFX9: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<2 x s32>)
+    %0:_(<2 x s32>) = COPY $vgpr0_vgpr1
+    %1:_(<2 x s32>) = COPY $vgpr2_vgpr3
+    %2:_(<2 x s32>) = nnan G_FSUB %0, %1
+    $vgpr0_vgpr1 = COPY %2
+...
+
+---
+name: test_fsub_v3s32
+body: |
+  bb.0.entry:
+    liveins: $vgpr0_vgpr1_vgpr2, $vgpr3_vgpr4_vgpr5
+
+    ; SI-LABEL: name: test_fsub_v3s32
+    ; SI: [[COPY:%[0-9]+]]:_(<3 x s32>) = COPY $vgpr0_vgpr1_vgpr2
+    ; SI: [[COPY1:%[0-9]+]]:_(<3 x s32>) = COPY $vgpr3_vgpr4_vgpr5
+    ; SI: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](<3 x s32>)
+    ; SI: [[UV3:%[0-9]+]]:_(s32), [[UV4:%[0-9]+]]:_(s32), [[UV5:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY1]](<3 x s32>)
+    ; SI: [[FSUB:%[0-9]+]]:_(s32) = G_FSUB [[UV]], [[UV3]]
+    ; SI: [[FSUB1:%[0-9]+]]:_(s32) = G_FSUB [[UV1]], [[UV4]]
+    ; SI: [[FSUB2:%[0-9]+]]:_(s32) = G_FSUB [[UV2]], [[UV5]]
+    ; SI: [[BUILD_VECTOR:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[FSUB]](s32), [[FSUB1]](s32), [[FSUB2]](s32)
+    ; SI: $vgpr0_vgpr1_vgpr2 = COPY [[BUILD_VECTOR]](<3 x s32>)
+    ; VI-LABEL: name: test_fsub_v3s32
+    ; VI: [[COPY:%[0-9]+]]:_(<3 x s32>) = COPY $vgpr0_vgpr1_vgpr2
+    ; VI: [[COPY1:%[0-9]+]]:_(<3 x s32>) = COPY $vgpr3_vgpr4_vgpr5
+    ; VI: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](<3 x s32>)
+    ; VI: [[UV3:%[0-9]+]]:_(s32), [[UV4:%[0-9]+]]:_(s32), [[UV5:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY1]](<3 x s32>)
+    ; VI: [[FSUB:%[0-9]+]]:_(s32) = G_FSUB [[UV]], [[UV3]]
+    ; VI: [[FSUB1:%[0-9]+]]:_(s32) = G_FSUB [[UV1]], [[UV4]]
+    ; VI: [[FSUB2:%[0-9]+]]:_(s32) = G_FSUB [[UV2]], [[UV5]]
+    ; VI: [[BUILD_VECTOR:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[FSUB]](s32), [[FSUB1]](s32), [[FSUB2]](s32)
+    ; VI: $vgpr0_vgpr1_vgpr2 = COPY [[BUILD_VECTOR]](<3 x s32>)
+    ; GFX9-LABEL: name: test_fsub_v3s32
+    ; GFX9: [[COPY:%[0-9]+]]:_(<3 x s32>) = COPY $vgpr0_vgpr1_vgpr2
+    ; GFX9: [[COPY1:%[0-9]+]]:_(<3 x s32>) = COPY $vgpr3_vgpr4_vgpr5
+    ; GFX9: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](<3 x s32>)
+    ; GFX9: [[UV3:%[0-9]+]]:_(s32), [[UV4:%[0-9]+]]:_(s32), [[UV5:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY1]](<3 x s32>)
+    ; GFX9: [[FSUB:%[0-9]+]]:_(s32) = G_FSUB [[UV]], [[UV3]]
+    ; GFX9: [[FSUB1:%[0-9]+]]:_(s32) = G_FSUB [[UV1]], [[UV4]]
+    ; GFX9: [[FSUB2:%[0-9]+]]:_(s32) = G_FSUB [[UV2]], [[UV5]]
+    ; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[FSUB]](s32), [[FSUB1]](s32), [[FSUB2]](s32)
+    ; GFX9: $vgpr0_vgpr1_vgpr2 = COPY [[BUILD_VECTOR]](<3 x s32>)
+    %0:_(<3 x s32>) = COPY $vgpr0_vgpr1_vgpr2
+    %1:_(<3 x s32>) = COPY $vgpr3_vgpr4_vgpr5
+    %2:_(<3 x s32>) = G_FSUB %0, %1
+    $vgpr0_vgpr1_vgpr2 = COPY %2
+...
+
+---
+name: test_fsub_v2s64
+body: |
+  bb.0.entry:
+    liveins: $vgpr0_vgpr1_vgpr2_vgpr3, $vgpr4_vgpr5_vgpr6_vgpr7
+
+    ; SI-LABEL: name: test_fsub_v2s64
+    ; SI: [[COPY:%[0-9]+]]:_(<2 x s64>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3
+    ; SI: [[COPY1:%[0-9]+]]:_(<2 x s64>) = COPY $vgpr4_vgpr5_vgpr6_vgpr7
+    ; SI: [[UV:%[0-9]+]]:_(s64), [[UV1:%[0-9]+]]:_(s64) = G_UNMERGE_VALUES [[COPY]](<2 x s64>)
+    ; SI: [[UV2:%[0-9]+]]:_(s64), [[UV3:%[0-9]+]]:_(s64) = G_UNMERGE_VALUES [[COPY1]](<2 x s64>)
+    ; SI: [[FNEG:%[0-9]+]]:_(s64) = G_FNEG [[UV2]]
+    ; SI: [[FADD:%[0-9]+]]:_(s64) = G_FADD [[UV]], [[FNEG]]
+    ; SI: [[FNEG1:%[0-9]+]]:_(s64) = G_FNEG [[UV3]]
+    ; SI: [[FADD1:%[0-9]+]]:_(s64) = G_FADD [[UV1]], [[FNEG1]]
+    ; SI: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s64>) = G_BUILD_VECTOR [[FADD]](s64), [[FADD1]](s64)
+    ; SI: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[BUILD_VECTOR]](<2 x s64>)
+    ; VI-LABEL: name: test_fsub_v2s64
+    ; VI: [[COPY:%[0-9]+]]:_(<2 x s64>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3
+    ; VI: [[COPY1:%[0-9]+]]:_(<2 x s64>) = COPY $vgpr4_vgpr5_vgpr6_vgpr7
+    ; VI: [[UV:%[0-9]+]]:_(s64), [[UV1:%[0-9]+]]:_(s64) = G_UNMERGE_VALUES [[COPY]](<2 x s64>)
+    ; VI: [[UV2:%[0-9]+]]:_(s64), [[UV3:%[0-9]+]]:_(s64) = G_UNMERGE_VALUES [[COPY1]](<2 x s64>)
+    ; VI: [[FNEG:%[0-9]+]]:_(s64) = G_FNEG [[UV2]]
+    ; VI: [[FADD:%[0-9]+]]:_(s64) = G_FADD [[UV]], [[FNEG]]
+    ; VI: [[FNEG1:%[0-9]+]]:_(s64) = G_FNEG [[UV3]]
+    ; VI: [[FADD1:%[0-9]+]]:_(s64) = G_FADD [[UV1]], [[FNEG1]]
+    ; VI: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s64>) = G_BUILD_VECTOR [[FADD]](s64), [[FADD1]](s64)
+    ; VI: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[BUILD_VECTOR]](<2 x s64>)
+    ; GFX9-LABEL: name: test_fsub_v2s64
+    ; GFX9: [[COPY:%[0-9]+]]:_(<2 x s64>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3
+    ; GFX9: [[COPY1:%[0-9]+]]:_(<2 x s64>) = COPY $vgpr4_vgpr5_vgpr6_vgpr7
+    ; GFX9: [[UV:%[0-9]+]]:_(s64), [[UV1:%[0-9]+]]:_(s64) = G_UNMERGE_VALUES [[COPY]](<2 x s64>)
+    ; GFX9: [[UV2:%[0-9]+]]:_(s64), [[UV3:%[0-9]+]]:_(s64) = G_UNMERGE_VALUES [[COPY1]](<2 x s64>)
+    ; GFX9: [[FNEG:%[0-9]+]]:_(s64) = G_FNEG [[UV2]]
+    ; GFX9: [[FADD:%[0-9]+]]:_(s64) = G_FADD [[UV]], [[FNEG]]
+    ; GFX9: [[FNEG1:%[0-9]+]]:_(s64) = G_FNEG [[UV3]]
+    ; GFX9: [[FADD1:%[0-9]+]]:_(s64) = G_FADD [[UV1]], [[FNEG1]]
+    ; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s64>) = G_BUILD_VECTOR [[FADD]](s64), [[FADD1]](s64)
+    ; GFX9: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[BUILD_VECTOR]](<2 x s64>)
+    %0:_(<2 x s64>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3
+    %1:_(<2 x s64>) = COPY $vgpr4_vgpr5_vgpr6_vgpr7
+    %2:_(<2 x s64>) = G_FSUB %0, %1
+    $vgpr0_vgpr1_vgpr2_vgpr3 = COPY %2
+...
+
+---
+name: test_fsub_v2s16
+body: |
+  bb.0.entry:
+    liveins: $vgpr0, $vgpr1
+
+    ; SI-LABEL: name: test_fsub_v2s16
+    ; SI: [[COPY:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0
+    ; SI: [[COPY1:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr1
+    ; SI: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY1]](<2 x s16>)
+    ; SI: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[UV]](s16)
+    ; SI: [[FNEG:%[0-9]+]]:_(s32) = G_FNEG [[FPEXT]]
+    ; SI: [[FPTRUNC:%[0-9]+]]:_(s16) = G_FPTRUNC [[FNEG]](s32)
+    ; SI: [[FPEXT1:%[0-9]+]]:_(s32) = G_FPEXT [[UV1]](s16)
+    ; SI: [[FNEG1:%[0-9]+]]:_(s32) = G_FNEG [[FPEXT1]]
+    ; SI: [[FPTRUNC1:%[0-9]+]]:_(s16) = G_FPTRUNC [[FNEG1]](s32)
+    ; SI: [[UV2:%[0-9]+]]:_(s16), [[UV3:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY]](<2 x s16>)
+    ; SI: [[FPEXT2:%[0-9]+]]:_(s32) = G_FPEXT [[UV2]](s16)
+    ; SI: [[FPEXT3:%[0-9]+]]:_(s32) = G_FPEXT [[FPTRUNC]](s16)
+    ; SI: [[FADD:%[0-9]+]]:_(s32) = G_FADD [[FPEXT2]], [[FPEXT3]]
+    ; SI: [[FPTRUNC2:%[0-9]+]]:_(s16) = G_FPTRUNC [[FADD]](s32)
+    ; SI: [[FPEXT4:%[0-9]+]]:_(s32) = G_FPEXT [[UV3]](s16)
+    ; SI: [[FPEXT5:%[0-9]+]]:_(s32) = G_FPEXT [[FPTRUNC1]](s16)
+    ; SI: [[FADD1:%[0-9]+]]:_(s32) = G_FADD [[FPEXT4]], [[FPEXT5]]
+    ; SI: [[FPTRUNC3:%[0-9]+]]:_(s16) = G_FPTRUNC [[FADD1]](s32)
+    ; SI: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR [[FPTRUNC2]](s16), [[FPTRUNC3]](s16)
+    ; SI: $vgpr0 = COPY [[BUILD_VECTOR]](<2 x s16>)
+    ; VI-LABEL: name: test_fsub_v2s16
+    ; VI: [[COPY:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0
+    ; VI: [[COPY1:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr1
+    ; VI: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY1]](<2 x s16>)
+    ; VI: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[UV]](s16)
+    ; VI: [[FNEG:%[0-9]+]]:_(s32) = G_FNEG [[FPEXT]]
+    ; VI: [[FPTRUNC:%[0-9]+]]:_(s16) = G_FPTRUNC [[FNEG]](s32)
+    ; VI: [[FPEXT1:%[0-9]+]]:_(s32) = G_FPEXT [[UV1]](s16)
+    ; VI: [[FNEG1:%[0-9]+]]:_(s32) = G_FNEG [[FPEXT1]]
+    ; VI: [[FPTRUNC1:%[0-9]+]]:_(s16) = G_FPTRUNC [[FNEG1]](s32)
+    ; VI: [[UV2:%[0-9]+]]:_(s16), [[UV3:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY]](<2 x s16>)
+    ; VI: [[FPEXT2:%[0-9]+]]:_(s32) = G_FPEXT [[UV2]](s16)
+    ; VI: [[FPEXT3:%[0-9]+]]:_(s32) = G_FPEXT [[FPTRUNC]](s16)
+    ; VI: [[FADD:%[0-9]+]]:_(s32) = G_FADD [[FPEXT2]], [[FPEXT3]]
+    ; VI: [[FPTRUNC2:%[0-9]+]]:_(s16) = G_FPTRUNC [[FADD]](s32)
+    ; VI: [[FPEXT4:%[0-9]+]]:_(s32) = G_FPEXT [[UV3]](s16)
+    ; VI: [[FPEXT5:%[0-9]+]]:_(s32) = G_FPEXT [[FPTRUNC1]](s16)
+    ; VI: [[FADD1:%[0-9]+]]:_(s32) = G_FADD [[FPEXT4]], [[FPEXT5]]
+    ; VI: [[FPTRUNC3:%[0-9]+]]:_(s16) = G_FPTRUNC [[FADD1]](s32)
+    ; VI: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR [[FPTRUNC2]](s16), [[FPTRUNC3]](s16)
+    ; VI: $vgpr0 = COPY [[BUILD_VECTOR]](<2 x s16>)
+    ; GFX9-LABEL: name: test_fsub_v2s16
+    ; GFX9: [[COPY:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0
+    ; GFX9: [[COPY1:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr1
+    ; GFX9: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY1]](<2 x s16>)
+    ; GFX9: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[UV]](s16)
+    ; GFX9: [[FNEG:%[0-9]+]]:_(s32) = G_FNEG [[FPEXT]]
+    ; GFX9: [[FPTRUNC:%[0-9]+]]:_(s16) = G_FPTRUNC [[FNEG]](s32)
+    ; GFX9: [[FPEXT1:%[0-9]+]]:_(s32) = G_FPEXT [[UV1]](s16)
+    ; GFX9: [[FNEG1:%[0-9]+]]:_(s32) = G_FNEG [[FPEXT1]]
+    ; GFX9: [[FPTRUNC1:%[0-9]+]]:_(s16) = G_FPTRUNC [[FNEG1]](s32)
+    ; GFX9: [[UV2:%[0-9]+]]:_(s16), [[UV3:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY]](<2 x s16>)
+    ; GFX9: [[FPEXT2:%[0-9]+]]:_(s32) = G_FPEXT [[UV2]](s16)
+    ; GFX9: [[FPEXT3:%[0-9]+]]:_(s32) = G_FPEXT [[FPTRUNC]](s16)
+    ; GFX9: [[FADD:%[0-9]+]]:_(s32) = G_FADD [[FPEXT2]], [[FPEXT3]]
+    ; GFX9: [[FPTRUNC2:%[0-9]+]]:_(s16) = G_FPTRUNC [[FADD]](s32)
+    ; GFX9: [[FPEXT4:%[0-9]+]]:_(s32) = G_FPEXT [[UV3]](s16)
+    ; GFX9: [[FPEXT5:%[0-9]+]]:_(s32) = G_FPEXT [[FPTRUNC1]](s16)
+    ; GFX9: [[FADD1:%[0-9]+]]:_(s32) = G_FADD [[FPEXT4]], [[FPEXT5]]
+    ; GFX9: [[FPTRUNC3:%[0-9]+]]:_(s16) = G_FPTRUNC [[FADD1]](s32)
+    ; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR [[FPTRUNC2]](s16), [[FPTRUNC3]](s16)
+    ; GFX9: $vgpr0 = COPY [[BUILD_VECTOR]](<2 x s16>)
+    %0:_(<2 x s16>) = COPY $vgpr0
+    %1:_(<2 x s16>) = COPY $vgpr1
+    %2:_(<2 x s16>) = G_FSUB %0, %1
+    $vgpr0 = COPY %2
+...
+
+---
+name: test_fsub_v3s16
+body: |
+  bb.0.entry:
+    liveins: $vgpr0, $vgpr1
+
+    ; SI-LABEL: name: test_fsub_v3s16
+    ; SI: [[DEF:%[0-9]+]]:_(<3 x s16>) = G_IMPLICIT_DEF
+    ; SI: [[DEF1:%[0-9]+]]:_(<3 x s16>) = G_IMPLICIT_DEF
+    ; SI: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[DEF]](<3 x s16>)
+    ; SI: [[UV3:%[0-9]+]]:_(s16), [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[DEF1]](<3 x s16>)
+    ; SI: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[UV3]](s16)
+    ; SI: [[FNEG:%[0-9]+]]:_(s32) = G_FNEG [[FPEXT]]
+    ; SI: [[FPTRUNC:%[0-9]+]]:_(s16) = G_FPTRUNC [[FNEG]](s32)
+    ; SI: [[FPEXT1:%[0-9]+]]:_(s32) = G_FPEXT [[UV]](s16)
+    ; SI: [[FPEXT2:%[0-9]+]]:_(s32) = G_FPEXT [[FPTRUNC]](s16)
+    ; SI: [[FADD:%[0-9]+]]:_(s32) = G_FADD [[FPEXT1]], [[FPEXT2]]
+    ; SI: [[FPTRUNC1:%[0-9]+]]:_(s16) = G_FPTRUNC [[FADD]](s32)
+    ; SI: [[FPEXT3:%[0-9]+]]:_(s32) = G_FPEXT [[UV4]](s16)
+    ; SI: [[FNEG1:%[0-9]+]]:_(s32) = G_FNEG [[FPEXT3]]
+    ; SI: [[FPTRUNC2:%[0-9]+]]:_(s16) = G_FPTRUNC [[FNEG1]](s32)
+    ; SI: [[FPEXT4:%[0-9]+]]:_(s32) = G_FPEXT [[UV1]](s16)
+    ; SI: [[FPEXT5:%[0-9]+]]:_(s32) = G_FPEXT [[FPTRUNC2]](s16)
+    ; SI: [[FADD1:%[0-9]+]]:_(s32) = G_FADD [[FPEXT4]], [[FPEXT5]]
+    ; SI: [[FPTRUNC3:%[0-9]+]]:_(s16) = G_FPTRUNC [[FADD1]](s32)
+    ; SI: [[FPEXT6:%[0-9]+]]:_(s32) = G_FPEXT [[UV5]](s16)
+    ; SI: [[FNEG2:%[0-9]+]]:_(s32) = G_FNEG [[FPEXT6]]
+    ; SI: [[FPTRUNC4:%[0-9]+]]:_(s16) = G_FPTRUNC [[FNEG2]](s32)
+    ; SI: [[FPEXT7:%[0-9]+]]:_(s32) = G_FPEXT [[UV2]](s16)
+    ; SI: [[FPEXT8:%[0-9]+]]:_(s32) = G_FPEXT [[FPTRUNC4]](s16)
+    ; SI: [[FADD2:%[0-9]+]]:_(s32) = G_FADD [[FPEXT7]], [[FPEXT8]]
+    ; SI: [[FPTRUNC5:%[0-9]+]]:_(s16) = G_FPTRUNC [[FADD2]](s32)
+    ; SI: [[BUILD_VECTOR:%[0-9]+]]:_(<3 x s16>) = G_BUILD_VECTOR [[FPTRUNC1]](s16), [[FPTRUNC3]](s16), [[FPTRUNC5]](s16)
+    ; SI: S_NOP 0, implicit [[BUILD_VECTOR]](<3 x s16>)
+    ; VI-LABEL: name: test_fsub_v3s16
+    ; VI: [[DEF:%[0-9]+]]:_(<3 x s16>) = G_IMPLICIT_DEF
+    ; VI: [[DEF1:%[0-9]+]]:_(<3 x s16>) = G_IMPLICIT_DEF
+    ; VI: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[DEF]](<3 x s16>)
+    ; VI: [[UV3:%[0-9]+]]:_(s16), [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[DEF1]](<3 x s16>)
+    ; VI: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[UV3]](s16)
+    ; VI: [[FNEG:%[0-9]+]]:_(s32) = G_FNEG [[FPEXT]]
+    ; VI: [[FPTRUNC:%[0-9]+]]:_(s16) = G_FPTRUNC [[FNEG]](s32)
+    ; VI: [[FPEXT1:%[0-9]+]]:_(s32) = G_FPEXT [[UV]](s16)
+    ; VI: [[FPEXT2:%[0-9]+]]:_(s32) = G_FPEXT [[FPTRUNC]](s16)
+    ; VI: [[FADD:%[0-9]+]]:_(s32) = G_FADD [[FPEXT1]], [[FPEXT2]]
+    ; VI: [[FPTRUNC1:%[0-9]+]]:_(s16) = G_FPTRUNC [[FADD]](s32)
+    ; VI: [[FPEXT3:%[0-9]+]]:_(s32) = G_FPEXT [[UV4]](s16)
+    ; VI: [[FNEG1:%[0-9]+]]:_(s32) = G_FNEG [[FPEXT3]]
+    ; VI: [[FPTRUNC2:%[0-9]+]]:_(s16) = G_FPTRUNC [[FNEG1]](s32)
+    ; VI: [[FPEXT4:%[0-9]+]]:_(s32) = G_FPEXT [[UV1]](s16)
+    ; VI: [[FPEXT5:%[0-9]+]]:_(s32) = G_FPEXT [[FPTRUNC2]](s16)
+    ; VI: [[FADD1:%[0-9]+]]:_(s32) = G_FADD [[FPEXT4]], [[FPEXT5]]
+    ; VI: [[FPTRUNC3:%[0-9]+]]:_(s16) = G_FPTRUNC [[FADD1]](s32)
+    ; VI: [[FPEXT6:%[0-9]+]]:_(s32) = G_FPEXT [[UV5]](s16)
+    ; VI: [[FNEG2:%[0-9]+]]:_(s32) = G_FNEG [[FPEXT6]]
+    ; VI: [[FPTRUNC4:%[0-9]+]]:_(s16) = G_FPTRUNC [[FNEG2]](s32)
+    ; VI: [[FPEXT7:%[0-9]+]]:_(s32) = G_FPEXT [[UV2]](s16)
+    ; VI: [[FPEXT8:%[0-9]+]]:_(s32) = G_FPEXT [[FPTRUNC4]](s16)
+    ; VI: [[FADD2:%[0-9]+]]:_(s32) = G_FADD [[FPEXT7]], [[FPEXT8]]
+    ; VI: [[FPTRUNC5:%[0-9]+]]:_(s16) = G_FPTRUNC [[FADD2]](s32)
+    ; VI: [[BUILD_VECTOR:%[0-9]+]]:_(<3 x s16>) = G_BUILD_VECTOR [[FPTRUNC1]](s16), [[FPTRUNC3]](s16), [[FPTRUNC5]](s16)
+    ; VI: S_NOP 0, implicit [[BUILD_VECTOR]](<3 x s16>)
+    ; GFX9-LABEL: name: test_fsub_v3s16
+    ; GFX9: [[DEF:%[0-9]+]]:_(<3 x s16>) = G_IMPLICIT_DEF
+    ; GFX9: [[DEF1:%[0-9]+]]:_(<3 x s16>) = G_IMPLICIT_DEF
+    ; GFX9: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[DEF]](<3 x s16>)
+    ; GFX9: [[UV3:%[0-9]+]]:_(s16), [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[DEF1]](<3 x s16>)
+    ; GFX9: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[UV3]](s16)
+    ; GFX9: [[FNEG:%[0-9]+]]:_(s32) = G_FNEG [[FPEXT]]
+    ; GFX9: [[FPTRUNC:%[0-9]+]]:_(s16) = G_FPTRUNC [[FNEG]](s32)
+    ; GFX9: [[FPEXT1:%[0-9]+]]:_(s32) = G_FPEXT [[UV]](s16)
+    ; GFX9: [[FPEXT2:%[0-9]+]]:_(s32) = G_FPEXT [[FPTRUNC]](s16)
+    ; GFX9: [[FADD:%[0-9]+]]:_(s32) = G_FADD [[FPEXT1]], [[FPEXT2]]
+    ; GFX9: [[FPTRUNC1:%[0-9]+]]:_(s16) = G_FPTRUNC [[FADD]](s32)
+    ; GFX9: [[FPEXT3:%[0-9]+]]:_(s32) = G_FPEXT [[UV4]](s16)
+    ; GFX9: [[FNEG1:%[0-9]+]]:_(s32) = G_FNEG [[FPEXT3]]
+    ; GFX9: [[FPTRUNC2:%[0-9]+]]:_(s16) = G_FPTRUNC [[FNEG1]](s32)
+    ; GFX9: [[FPEXT4:%[0-9]+]]:_(s32) = G_FPEXT [[UV1]](s16)
+    ; GFX9: [[FPEXT5:%[0-9]+]]:_(s32) = G_FPEXT [[FPTRUNC2]](s16)
+    ; GFX9: [[FADD1:%[0-9]+]]:_(s32) = G_FADD [[FPEXT4]], [[FPEXT5]]
+    ; GFX9: [[FPTRUNC3:%[0-9]+]]:_(s16) = G_FPTRUNC [[FADD1]](s32)
+    ; GFX9: [[FPEXT6:%[0-9]+]]:_(s32) = G_FPEXT [[UV5]](s16)
+    ; GFX9: [[FNEG2:%[0-9]+]]:_(s32) = G_FNEG [[FPEXT6]]
+    ; GFX9: [[FPTRUNC4:%[0-9]+]]:_(s16) = G_FPTRUNC [[FNEG2]](s32)
+    ; GFX9: [[FPEXT7:%[0-9]+]]:_(s32) = G_FPEXT [[UV2]](s16)
+    ; GFX9: [[FPEXT8:%[0-9]+]]:_(s32) = G_FPEXT [[FPTRUNC4]](s16)
+    ; GFX9: [[FADD2:%[0-9]+]]:_(s32) = G_FADD [[FPEXT7]], [[FPEXT8]]
+    ; GFX9: [[FPTRUNC5:%[0-9]+]]:_(s16) = G_FPTRUNC [[FADD2]](s32)
+    ; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<3 x s16>) = G_BUILD_VECTOR [[FPTRUNC1]](s16), [[FPTRUNC3]](s16), [[FPTRUNC5]](s16)
+    ; GFX9: S_NOP 0, implicit [[BUILD_VECTOR]](<3 x s16>)
+    %0:_(<3 x s16>) = G_IMPLICIT_DEF
+    %1:_(<3 x s16>) = G_IMPLICIT_DEF
+    %2:_(<3 x s16>) = G_FSUB %0, %1
+    S_NOP 0, implicit %2
+...
+
+---
+name: test_fsub_v4s16
+body: |
+  bb.0.entry:
+    liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
+
+    ; SI-LABEL: name: test_fsub_v4s16
+    ; SI: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1
+    ; SI: [[COPY1:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr2_vgpr3
+    ; SI: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16), [[UV3:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY]](<4 x s16>)
+    ; SI: [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16), [[UV6:%[0-9]+]]:_(s16), [[UV7:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY1]](<4 x s16>)
+    ; SI: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[UV4]](s16)
+    ; SI: [[FNEG:%[0-9]+]]:_(s32) = G_FNEG [[FPEXT]]
+    ; SI: [[FPTRUNC:%[0-9]+]]:_(s16) = G_FPTRUNC [[FNEG]](s32)
+    ; SI: [[FPEXT1:%[0-9]+]]:_(s32) = G_FPEXT [[UV]](s16)
+    ; SI: [[FPEXT2:%[0-9]+]]:_(s32) = G_FPEXT [[FPTRUNC]](s16)
+    ; SI: [[FADD:%[0-9]+]]:_(s32) = G_FADD [[FPEXT1]], [[FPEXT2]]
+    ; SI: [[FPTRUNC1:%[0-9]+]]:_(s16) = G_FPTRUNC [[FADD]](s32)
+    ; SI: [[FPEXT3:%[0-9]+]]:_(s32) = G_FPEXT [[UV5]](s16)
+    ; SI: [[FNEG1:%[0-9]+]]:_(s32) = G_FNEG [[FPEXT3]]
+    ; SI: [[FPTRUNC2:%[0-9]+]]:_(s16) = G_FPTRUNC [[FNEG1]](s32)
+    ; SI: [[FPEXT4:%[0-9]+]]:_(s32) = G_FPEXT [[UV1]](s16)
+    ; SI: [[FPEXT5:%[0-9]+]]:_(s32) = G_FPEXT [[FPTRUNC2]](s16)
+    ; SI: [[FADD1:%[0-9]+]]:_(s32) = G_FADD [[FPEXT4]], [[FPEXT5]]
+    ; SI: [[FPTRUNC3:%[0-9]+]]:_(s16) = G_FPTRUNC [[FADD1]](s32)
+    ; SI: [[FPEXT6:%[0-9]+]]:_(s32) = G_FPEXT [[UV6]](s16)
+    ; SI: [[FNEG2:%[0-9]+]]:_(s32) = G_FNEG [[FPEXT6]]
+    ; SI: [[FPTRUNC4:%[0-9]+]]:_(s16) = G_FPTRUNC [[FNEG2]](s32)
+    ; SI: [[FPEXT7:%[0-9]+]]:_(s32) = G_FPEXT [[UV2]](s16)
+    ; SI: [[FPEXT8:%[0-9]+]]:_(s32) = G_FPEXT [[FPTRUNC4]](s16)
+    ; SI: [[FADD2:%[0-9]+]]:_(s32) = G_FADD [[FPEXT7]], [[FPEXT8]]
+    ; SI: [[FPTRUNC5:%[0-9]+]]:_(s16) = G_FPTRUNC [[FADD2]](s32)
+    ; SI: [[FPEXT9:%[0-9]+]]:_(s32) = G_FPEXT [[UV7]](s16)
+    ; SI: [[FNEG3:%[0-9]+]]:_(s32) = G_FNEG [[FPEXT9]]
+    ; SI: [[FPTRUNC6:%[0-9]+]]:_(s16) = G_FPTRUNC [[FNEG3]](s32)
+    ; SI: [[FPEXT10:%[0-9]+]]:_(s32) = G_FPEXT [[UV3]](s16)
+    ; SI: [[FPEXT11:%[0-9]+]]:_(s32) = G_FPEXT [[FPTRUNC6]](s16)
+    ; SI: [[FADD3:%[0-9]+]]:_(s32) = G_FADD [[FPEXT10]], [[FPEXT11]]
+    ; SI: [[FPTRUNC7:%[0-9]+]]:_(s16) = G_FPTRUNC [[FADD3]](s32)
+    ; SI: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s16>) = G_BUILD_VECTOR [[FPTRUNC1]](s16), [[FPTRUNC3]](s16), [[FPTRUNC5]](s16), [[FPTRUNC7]](s16)
+    ; SI: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<4 x s16>)
+    ; VI-LABEL: name: test_fsub_v4s16
+    ; VI: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1
+    ; VI: [[COPY1:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr2_vgpr3
+    ; VI: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16), [[UV3:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY]](<4 x s16>)
+    ; VI: [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16), [[UV6:%[0-9]+]]:_(s16), [[UV7:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY1]](<4 x s16>)
+    ; VI: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[UV4]](s16)
+    ; VI: [[FNEG:%[0-9]+]]:_(s32) = G_FNEG [[FPEXT]]
+    ; VI: [[FPTRUNC:%[0-9]+]]:_(s16) = G_FPTRUNC [[FNEG]](s32)
+    ; VI: [[FPEXT1:%[0-9]+]]:_(s32) = G_FPEXT [[UV]](s16)
+    ; VI: [[FPEXT2:%[0-9]+]]:_(s32) = G_FPEXT [[FPTRUNC]](s16)
+    ; VI: [[FADD:%[0-9]+]]:_(s32) = G_FADD [[FPEXT1]], [[FPEXT2]]
+    ; VI: [[FPTRUNC1:%[0-9]+]]:_(s16) = G_FPTRUNC [[FADD]](s32)
+    ; VI: [[FPEXT3:%[0-9]+]]:_(s32) = G_FPEXT [[UV5]](s16)
+    ; VI: [[FNEG1:%[0-9]+]]:_(s32) = G_FNEG [[FPEXT3]]
+    ; VI: [[FPTRUNC2:%[0-9]+]]:_(s16) = G_FPTRUNC [[FNEG1]](s32)
+    ; VI: [[FPEXT4:%[0-9]+]]:_(s32) = G_FPEXT [[UV1]](s16)
+    ; VI: [[FPEXT5:%[0-9]+]]:_(s32) = G_FPEXT [[FPTRUNC2]](s16)
+    ; VI: [[FADD1:%[0-9]+]]:_(s32) = G_FADD [[FPEXT4]], [[FPEXT5]]
+    ; VI: [[FPTRUNC3:%[0-9]+]]:_(s16) = G_FPTRUNC [[FADD1]](s32)
+    ; VI: [[FPEXT6:%[0-9]+]]:_(s32) = G_FPEXT [[UV6]](s16)
+    ; VI: [[FNEG2:%[0-9]+]]:_(s32) = G_FNEG [[FPEXT6]]
+    ; VI: [[FPTRUNC4:%[0-9]+]]:_(s16) = G_FPTRUNC [[FNEG2]](s32)
+    ; VI: [[FPEXT7:%[0-9]+]]:_(s32) = G_FPEXT [[UV2]](s16)
+    ; VI: [[FPEXT8:%[0-9]+]]:_(s32) = G_FPEXT [[FPTRUNC4]](s16)
+    ; VI: [[FADD2:%[0-9]+]]:_(s32) = G_FADD [[FPEXT7]], [[FPEXT8]]
+    ; VI: [[FPTRUNC5:%[0-9]+]]:_(s16) = G_FPTRUNC [[FADD2]](s32)
+    ; VI: [[FPEXT9:%[0-9]+]]:_(s32) = G_FPEXT [[UV7]](s16)
+    ; VI: [[FNEG3:%[0-9]+]]:_(s32) = G_FNEG [[FPEXT9]]
+    ; VI: [[FPTRUNC6:%[0-9]+]]:_(s16) = G_FPTRUNC [[FNEG3]](s32)
+    ; VI: [[FPEXT10:%[0-9]+]]:_(s32) = G_FPEXT [[UV3]](s16)
+    ; VI: [[FPEXT11:%[0-9]+]]:_(s32) = G_FPEXT [[FPTRUNC6]](s16)
+    ; VI: [[FADD3:%[0-9]+]]:_(s32) = G_FADD [[FPEXT10]], [[FPEXT11]]
+    ; VI: [[FPTRUNC7:%[0-9]+]]:_(s16) = G_FPTRUNC [[FADD3]](s32)
+    ; VI: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s16>) = G_BUILD_VECTOR [[FPTRUNC1]](s16), [[FPTRUNC3]](s16), [[FPTRUNC5]](s16), [[FPTRUNC7]](s16)
+    ; VI: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<4 x s16>)
+    ; GFX9-LABEL: name: test_fsub_v4s16
+    ; GFX9: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1
+    ; GFX9: [[COPY1:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr2_vgpr3
+    ; GFX9: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16), [[UV3:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY]](<4 x s16>)
+    ; GFX9: [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16), [[UV6:%[0-9]+]]:_(s16), [[UV7:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY1]](<4 x s16>)
+    ; GFX9: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[UV4]](s16)
+    ; GFX9: [[FNEG:%[0-9]+]]:_(s32) = G_FNEG [[FPEXT]]
+    ; GFX9: [[FPTRUNC:%[0-9]+]]:_(s16) = G_FPTRUNC [[FNEG]](s32)
+    ; GFX9: [[FPEXT1:%[0-9]+]]:_(s32) = G_FPEXT [[UV]](s16)
+    ; GFX9: [[FPEXT2:%[0-9]+]]:_(s32) = G_FPEXT [[FPTRUNC]](s16)
+    ; GFX9: [[FADD:%[0-9]+]]:_(s32) = G_FADD [[FPEXT1]], [[FPEXT2]]
+    ; GFX9: [[FPTRUNC1:%[0-9]+]]:_(s16) = G_FPTRUNC [[FADD]](s32)
+    ; GFX9: [[FPEXT3:%[0-9]+]]:_(s32) = G_FPEXT [[UV5]](s16)
+    ; GFX9: [[FNEG1:%[0-9]+]]:_(s32) = G_FNEG [[FPEXT3]]
+    ; GFX9: [[FPTRUNC2:%[0-9]+]]:_(s16) = G_FPTRUNC [[FNEG1]](s32)
+    ; GFX9: [[FPEXT4:%[0-9]+]]:_(s32) = G_FPEXT [[UV1]](s16)
+    ; GFX9: [[FPEXT5:%[0-9]+]]:_(s32) = G_FPEXT [[FPTRUNC2]](s16)
+    ; GFX9: [[FADD1:%[0-9]+]]:_(s32) = G_FADD [[FPEXT4]], [[FPEXT5]]
+    ; GFX9: [[FPTRUNC3:%[0-9]+]]:_(s16) = G_FPTRUNC [[FADD1]](s32)
+    ; GFX9: [[FPEXT6:%[0-9]+]]:_(s32) = G_FPEXT [[UV6]](s16)
+    ; GFX9: [[FNEG2:%[0-9]+]]:_(s32) = G_FNEG [[FPEXT6]]
+    ; GFX9: [[FPTRUNC4:%[0-9]+]]:_(s16) = G_FPTRUNC [[FNEG2]](s32)
+    ; GFX9: [[FPEXT7:%[0-9]+]]:_(s32) = G_FPEXT [[UV2]](s16)
+    ; GFX9: [[FPEXT8:%[0-9]+]]:_(s32) = G_FPEXT [[FPTRUNC4]](s16)
+    ; GFX9: [[FADD2:%[0-9]+]]:_(s32) = G_FADD [[FPEXT7]], [[FPEXT8]]
+    ; GFX9: [[FPTRUNC5:%[0-9]+]]:_(s16) = G_FPTRUNC [[FADD2]](s32)
+    ; GFX9: [[FPEXT9:%[0-9]+]]:_(s32) = G_FPEXT [[UV7]](s16)
+    ; GFX9: [[FNEG3:%[0-9]+]]:_(s32) = G_FNEG [[FPEXT9]]
+    ; GFX9: [[FPTRUNC6:%[0-9]+]]:_(s16) = G_FPTRUNC [[FNEG3]](s32)
+    ; GFX9: [[FPEXT10:%[0-9]+]]:_(s32) = G_FPEXT [[UV3]](s16)
+    ; GFX9: [[FPEXT11:%[0-9]+]]:_(s32) = G_FPEXT [[FPTRUNC6]](s16)
+    ; GFX9: [[FADD3:%[0-9]+]]:_(s32) = G_FADD [[FPEXT10]], [[FPEXT11]]
+    ; GFX9: [[FPTRUNC7:%[0-9]+]]:_(s16) = G_FPTRUNC [[FADD3]](s32)
+    ; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s16>) = G_BUILD_VECTOR [[FPTRUNC1]](s16), [[FPTRUNC3]](s16), [[FPTRUNC5]](s16), [[FPTRUNC7]](s16)
+    ; GFX9: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<4 x s16>)
+    %0:_(<4 x s16>) = COPY $vgpr0_vgpr1
+    %1:_(<4 x s16>) = COPY $vgpr2_vgpr3
+    %2:_(<4 x s16>) = G_FSUB %0, %1
+    $vgpr0_vgpr1 = COPY %2
+...