Index: lib/CodeGen/GlobalISel/LegalizerHelper.cpp
===================================================================
--- lib/CodeGen/GlobalISel/LegalizerHelper.cpp
+++ lib/CodeGen/GlobalISel/LegalizerHelper.cpp
@@ -469,6 +469,38 @@
     MI.eraseFromParent();
     return Legalized;
   }
+  case TargetOpcode::G_CONSTANT: {
+    LLT Ty = MRI.getType(MI.getOperand(0).getReg());
+    const APInt &Val = MI.getOperand(1).getCImm()->getValue();
+    unsigned TotalSize = Ty.getSizeInBits();
+    unsigned NarrowSize = NarrowTy.getSizeInBits();
+    int NumParts = TotalSize / NarrowSize;
+
+    SmallVector<unsigned, 4> PartRegs;
+    for (int I = 0; I != NumParts; ++I) {
+      unsigned Offset = I * NarrowSize;
+      auto K = MIRBuilder.buildConstant(NarrowTy,
+                                        Val.lshr(Offset).trunc(NarrowSize));
+      PartRegs.push_back(K.getReg(0));
+    }
+
+    LLT LeftoverTy;
+    unsigned LeftoverBits = TotalSize - NumParts * NarrowSize;
+    SmallVector<unsigned, 1> LeftoverRegs;
+    if (LeftoverBits != 0) {
+      LeftoverTy = LLT::scalar(LeftoverBits);
+      auto K = MIRBuilder.buildConstant(
+        LeftoverTy,
+        Val.lshr(NumParts * NarrowSize).trunc(LeftoverBits));
+      LeftoverRegs.push_back(K.getReg(0));
+    }
+
+    insertParts(MI.getOperand(0).getReg(),
+                Ty, NarrowTy, PartRegs, LeftoverTy, LeftoverRegs);
+
+    MI.eraseFromParent();
+    return Legalized;
+  }
   case TargetOpcode::G_ADD: {
     // FIXME: add support for when SizeOp0 isn't an exact multiple of
     // NarrowSize.
@@ -611,31 +643,6 @@
 
     return reduceLoadStoreWidth(MI, 0, NarrowTy);
   }
-  case TargetOpcode::G_CONSTANT: {
-    // FIXME: add support for when SizeOp0 isn't an exact multiple of
-    // NarrowSize.
-    if (SizeOp0 % NarrowSize != 0)
-      return UnableToLegalize;
-    int NumParts = SizeOp0 / NarrowSize;
-    const APInt &Cst = MI.getOperand(1).getCImm()->getValue();
-    LLVMContext &Ctx = MIRBuilder.getMF().getFunction().getContext();
-
-    SmallVector<unsigned, 2> DstRegs;
-    for (int i = 0; i < NumParts; ++i) {
-      unsigned DstReg = MRI.createGenericVirtualRegister(NarrowTy);
-      ConstantInt *CI =
-          ConstantInt::get(Ctx, Cst.lshr(NarrowSize * i).trunc(NarrowSize));
-      MIRBuilder.buildConstant(DstReg, *CI);
-      DstRegs.push_back(DstReg);
-    }
-    unsigned DstReg = MI.getOperand(0).getReg();
-    if(MRI.getType(DstReg).isVector())
-      MIRBuilder.buildBuildVector(DstReg, DstRegs);
-    else
-      MIRBuilder.buildMerge(DstReg, DstRegs);
-    MI.eraseFromParent();
-    return Legalized;
-  }
   case TargetOpcode::G_SELECT:
     return narrowScalarSelect(MI, TypeIdx, NarrowTy);
   case TargetOpcode::G_AND:
Index: test/CodeGen/AArch64/GlobalISel/arm64-fallback.ll
===================================================================
--- test/CodeGen/AArch64/GlobalISel/arm64-fallback.ll
+++ test/CodeGen/AArch64/GlobalISel/arm64-fallback.ll
@@ -215,14 +215,6 @@
   ret void
 }
 
-; FALLBACK-WITH-REPORT-ERR: remark: <unknown>:0:0: unable to legalize instruction: %0:_(s96) = G_CONSTANT i96 0 (in function: nonpow2_constant_narrowing)
-; FALLBACK-WITH-REPORT-ERR: warning: Instruction selection used fallback path for nonpow2_constant_narrowing
-; FALLBACK-WITH-REPORT-OUT-LABEL: nonpow2_constant_narrowing:
-define void @nonpow2_constant_narrowing() {
-  store i96 0, i96* undef
-  ret void
-}
-
 ; Currently can't handle vector lengths that aren't an exact multiple of
 ; natively supported vector lengths. Test that the fall-back works for those.
 ; FALLBACK-WITH-REPORT-ERR-G_IMPLICIT_DEF-LEGALIZABLE: (FIXME: this is what is expected once we can legalize non-pow-of-2 G_IMPLICIT_DEF) remark: <unknown>:0:0: unable to legalize instruction: %1:_(<7 x s64>) = G_ADD %0, %0 (in function: nonpow2_vector_add_fewerelements
Index: test/CodeGen/AMDGPU/GlobalISel/legalize-constant.mir
===================================================================
--- test/CodeGen/AMDGPU/GlobalISel/legalize-constant.mir
+++ test/CodeGen/AMDGPU/GlobalISel/legalize-constant.mir
@@ -25,6 +25,23 @@
 
 ...
 
+---
+name:            test_constant_s96
+body: |
+  bb.0:
+
+    ; CHECK-LABEL: name: test_constant_s96
+    ; CHECK: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 -4780896129847249538
+    ; CHECK: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 -547834910
+    ; CHECK: [[DEF:%[0-9]+]]:_(s96) = G_IMPLICIT_DEF
+    ; CHECK: [[INSERT:%[0-9]+]]:_(s96) = G_INSERT [[DEF]], [[C]](s64), 0
+    ; CHECK: [[INSERT1:%[0-9]+]]:_(s96) = G_INSERT [[INSERT]], [[C1]](s32), 64
+    ; CHECK: $vgpr0_vgpr1_vgpr2 = COPY [[INSERT1]](s96)
+    %0:_(s96) = G_CONSTANT i96  -10105770365747857631829412482
+    $vgpr0_vgpr1_vgpr2 = COPY %0
+
+...
+
 ---
 name:            test_constant_s1
 body: |