[InstCombine] refine UB-handling in shuffle-binop transform

[spatel]

Summary

As noted in rL333782, we can be both better for optimization and safer with this transform:
BinOp (shuffle V1, Mask), C --> shuffle (BinOp V1, NewC), Mask

The only potentially unsafe-to-speculate binops are integer div/rem. All other binops are always safe (although I don't see a way to assert that in code here).

For opcodes like shifts that can produce poison, it can't matter here because we know the lanes with undef are dropped by the subsequent shuffle.

[lebedev.ri]

lib/Transforms/InstCombine/InstructionCombining.cpp

1427–1428 ↗

(On Diff #149638)

No special-handling for fdiv/frem?

1431 ↗

(On Diff #149638)

We want int(1) even for floats?

test/Transforms/InstCombine/vec_shuffle.ll

855 ↗

(On Diff #149638)

This looks like a float NAN?

[spatel]

lib/Transforms/InstCombine/InstructionCombining.cpp

1427–1428 ↗

(On Diff #149638)

No - this was clarified recently following the llvm-dev discussion about FP undef:
D44216

1431 ↗

(On Diff #149638)

We're only changing integer opcodes/operands here. I can assert that if it would make the code clearer?

test/Transforms/InstCombine/vec_shuffle.ll

855 ↗

(On Diff #149638)

Yes. Interesting - I didn't notice that test was different than the rest of the FP cases!

So what happens here:

1. This transform puts in an undef element in the constant vector.
2. We canonicalize fsub with constant operand 1 to fadd: // X - C --> X + (-C)

IC: Visiting:   %1 = fsub <2 x float> %x, <float 4.200000e+01, float undef>
IC: Old =   %1 = fsub <2 x float> %x, <float 4.200000e+01, float undef>
    New =   <badref> = fadd <2 x float> %x, <float -4.200000e+01, float 0x7FF8000000000000>

...and in that constant conversion, we constant fold the undef vector element to a NaN constant.

[lebedev.ri]

This makes sense to me, but i'm not really familiar with shufflevector,
so i'm going to leave the actual LGTM to others.

lib/Transforms/InstCombine/InstructionCombining.cpp

1420 ↗

(On Diff #149638)

Then i'd suggest

// With integer div/rem instructions, it is not safe to use a vector with undef

but that will require reflowing the entire comment :S

1431 ↗

(On Diff #149638)

I think this is precisely the case of things to assert, so ideally yes please.

[spatel]

Patch updated:
Make it clearer via code comment and assert that we are only dealing with integer opcodes/constants (FP binops are always safe to speculate).

[spatel]

In D47686#1120186, @lebedev.ri wrote:

This makes sense to me, but i'm not really familiar with shufflevector,
so i'm going to leave the actual LGTM to others.

No problem - thanks for the suggestions! @efriedma noticed the potential holes in my initial fix, so I'll certainly wait for him to comment.

[efriedma]

LGTM

llvm/trunk/lib/Transforms/InstCombine/InstructionCombining.cpp

for (unsigned I = 0; I < VWidth; ++I) {
         if (!CElt || (!isa<UndefValue>(NewCElt) && NewCElt != CElt)) {
           MayChange = false;
           break;
         }
         NewVecC[ShMask[I]] = CElt;
       }
     }
     if (MayChange) {
-      // It's not safe to use a vector with undef elements because the entire
-      // instruction can be folded to undef (for example, div/rem divisors).
-      // Replace undef lanes with the first non-undef element. Vector demanded
-      // elements can change those back to undef values if that is safe.
-      Constant *SafeDummyConstant = nullptr;
-      for (unsigned i = 0; i < VWidth; ++i) {
-        if (!isa<UndefValue>(NewVecC[i])) {
-          SafeDummyConstant = NewVecC[i];
-          break;
-        }
-      }
-      assert(SafeDummyConstant && "Undef constant vector was not simplified?");
+      // With integer div/rem instructions, it is not safe to use a vector with
+      // undef elements because the entire instruction can be folded to undef.
+      // So replace undef elements with '1' because that can never induce
+      // undefined behavior. All other binop opcodes are always safe to
+      // speculate, and therefore, it is fine to include undef elements for
+      // unused lanes (and using undefs may help optimization).
+      BinaryOperator::BinaryOps Opcode = Inst.getOpcode();
+      if (Opcode == Instruction::UDiv || Opcode == Instruction::URem ||
+          Opcode == Instruction::SDiv || Opcode == Instruction::SRem) {
+        assert(C->getType()->getScalarType()->isIntegerTy() &&
+               "Not expecting FP opcodes/operands/constants here");
         for (unsigned i = 0; i < VWidth; ++i)
           if (isa<UndefValue>(NewVecC[i]))
-          NewVecC[i] = SafeDummyConstant;
+            NewVecC[i] = ConstantInt::get(NewVecC[i]->getType(), 1);
+      }
 
       // Op(shuffle(V1, Mask), C) -> shuffle(Op(V1, NewC), Mask)
       // Op(C, shuffle(V1, Mask)) -> shuffle(Op(NewC, V1), Mask)
       Constant *NewC = ConstantVector::get(NewVecC);
       Value *NewLHS = isa<Constant>(LHS) ? NewC : V1;
       Value *NewRHS = isa<Constant>(LHS) ? V1 : NewC;
       return createBinOpShuffle(NewLHS, NewRHS, Mask);
     }

llvm/trunk/test/Transforms/InstCombine/vec_shuffle.ll

;
   %r = mul <4 x i32> <i32 42, i32 42, i32 42, i32 42>, %t1
   ret <4 x i32> %r
 }
 
 ; Take 2 elements of a vector and shift each of those by a different amount
 
 define <4 x i32> @lshr_const_half_splat(<4 x i32> %v) {
 ; CHECK-LABEL: @lshr_const_half_splat(
-; CHECK-NEXT:    [[TMP1:%.*]] = lshr <4 x i32> <i32 8, i32 8, i32 9, i32 8>, [[V:%.*]]
+; CHECK-NEXT:    [[TMP1:%.*]] = lshr <4 x i32> <i32 undef, i32 8, i32 9, i32 undef>, [[V:%.*]]
 ; CHECK-NEXT:    [[R:%.*]] = shufflevector <4 x i32> [[TMP1]], <4 x i32> undef, <4 x i32> <i32 1, i32 1, i32 2, i32 2>
 ; CHECK-NEXT:    ret <4 x i32> [[R]]
 ;
   %t1 = shufflevector <4 x i32> %v, <4 x i32> undef, <4 x i32> <i32 1, i32 1, i32 2, i32 2>
   %r = lshr <4 x i32> <i32 8, i32 8, i32 9, i32 9>, %t1
   ret <4 x i32> %r
 }
 

 ;
   %splat = shufflevector <2 x i32> %x, <2 x i32> undef, <2 x i32> zeroinitializer
   %r = mul <2 x i32> %splat, <i32 42, i32 42>
   ret <2 x i32> %r
 }
 
 define <2 x i32> @shl_splat_constant0(<2 x i32> %x) {
 ; CHECK-LABEL: @shl_splat_constant0(
-; CHECK-NEXT:    [[TMP1:%.*]] = shl <2 x i32> <i32 5, i32 5>, [[X:%.*]]
+; CHECK-NEXT:    [[TMP1:%.*]] = shl <2 x i32> <i32 5, i32 undef>, [[X:%.*]]
 ; CHECK-NEXT:    [[R:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> undef, <2 x i32> zeroinitializer
 ; CHECK-NEXT:    ret <2 x i32> [[R]]
 ;
   %splat = shufflevector <2 x i32> %x, <2 x i32> undef, <2 x i32> zeroinitializer
   %r = shl <2 x i32> <i32 5, i32 5>, %splat
   ret <2 x i32> %r
 }
 
 define <2 x i32> @shl_splat_constant1(<2 x i32> %x) {
 ; CHECK-LABEL: @shl_splat_constant1(
-; CHECK-NEXT:    [[TMP1:%.*]] = shl <2 x i32> [[X:%.*]], <i32 5, i32 5>
+; CHECK-NEXT:    [[TMP1:%.*]] = shl <2 x i32> [[X:%.*]], <i32 5, i32 undef>
 ; CHECK-NEXT:    [[R:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> undef, <2 x i32> zeroinitializer
 ; CHECK-NEXT:    ret <2 x i32> [[R]]
 ;
   %splat = shufflevector <2 x i32> %x, <2 x i32> undef, <2 x i32> zeroinitializer
   %r = shl <2 x i32> %splat, <i32 5, i32 5>
   ret <2 x i32> %r
 }
 
 define <2 x i32> @ashr_splat_constant0(<2 x i32> %x) {
 ; CHECK-LABEL: @ashr_splat_constant0(
-; CHECK-NEXT:    [[TMP1:%.*]] = lshr <2 x i32> <i32 5, i32 5>, [[X:%.*]]
+; CHECK-NEXT:    [[TMP1:%.*]] = ashr <2 x i32> <i32 5, i32 undef>, [[X:%.*]]
 ; CHECK-NEXT:    [[R:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> undef, <2 x i32> zeroinitializer
 ; CHECK-NEXT:    ret <2 x i32> [[R]]
 ;
   %splat = shufflevector <2 x i32> %x, <2 x i32> undef, <2 x i32> zeroinitializer
   %r = ashr <2 x i32> <i32 5, i32 5>, %splat
   ret <2 x i32> %r
 }
 
 define <2 x i32> @ashr_splat_constant1(<2 x i32> %x) {
 ; CHECK-LABEL: @ashr_splat_constant1(
-; CHECK-NEXT:    [[TMP1:%.*]] = ashr <2 x i32> [[X:%.*]], <i32 5, i32 5>
+; CHECK-NEXT:    [[TMP1:%.*]] = ashr <2 x i32> [[X:%.*]], <i32 5, i32 undef>
 ; CHECK-NEXT:    [[R:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> undef, <2 x i32> zeroinitializer
 ; CHECK-NEXT:    ret <2 x i32> [[R]]
 ;
   %splat = shufflevector <2 x i32> %x, <2 x i32> undef, <2 x i32> zeroinitializer
   %r = ashr <2 x i32> %splat, <i32 5, i32 5>
   ret <2 x i32> %r
 }
 
 define <2 x i32> @lshr_splat_constant0(<2 x i32> %x) {
 ; CHECK-LABEL: @lshr_splat_constant0(
-; CHECK-NEXT:    [[TMP1:%.*]] = lshr <2 x i32> <i32 5, i32 5>, [[X:%.*]]
+; CHECK-NEXT:    [[TMP1:%.*]] = lshr <2 x i32> <i32 5, i32 undef>, [[X:%.*]]
 ; CHECK-NEXT:    [[R:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> undef, <2 x i32> zeroinitializer
 ; CHECK-NEXT:    ret <2 x i32> [[R]]
 ;
   %splat = shufflevector <2 x i32> %x, <2 x i32> undef, <2 x i32> zeroinitializer
   %r = lshr <2 x i32> <i32 5, i32 5>, %splat
   ret <2 x i32> %r
 }
 
 define <2 x i32> @lshr_splat_constant1(<2 x i32> %x) {
 ; CHECK-LABEL: @lshr_splat_constant1(
-; CHECK-NEXT:    [[TMP1:%.*]] = lshr <2 x i32> [[X:%.*]], <i32 5, i32 5>
+; CHECK-NEXT:    [[TMP1:%.*]] = lshr <2 x i32> [[X:%.*]], <i32 5, i32 undef>
 ; CHECK-NEXT:    [[R:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> undef, <2 x i32> zeroinitializer
 ; CHECK-NEXT:    ret <2 x i32> [[R]]
 ;
   %splat = shufflevector <2 x i32> %x, <2 x i32> undef, <2 x i32> zeroinitializer
   %r = lshr <2 x i32> %splat, <i32 5, i32 5>
   ret <2 x i32> %r
 }
 
 define <2 x i32> @urem_splat_constant0(<2 x i32> %x) {
 ; CHECK-LABEL: @urem_splat_constant0(
 ; CHECK-NEXT:    [[SPLAT:%.*]] = shufflevector <2 x i32> [[X:%.*]], <2 x i32> undef, <2 x i32> zeroinitializer
 ; CHECK-NEXT:    [[R:%.*]] = urem <2 x i32> <i32 42, i32 42>, [[SPLAT]]
 ; CHECK-NEXT:    ret <2 x i32> [[R]]
 ;
   %splat = shufflevector <2 x i32> %x, <2 x i32> undef, <2 x i32> zeroinitializer
   %r = urem <2 x i32> <i32 42, i32 42>, %splat
   ret <2 x i32> %r
 }
 
 define <2 x i32> @urem_splat_constant1(<2 x i32> %x) {
 ; CHECK-LABEL: @urem_splat_constant1(
-; CHECK-NEXT:    [[TMP1:%.*]] = urem <2 x i32> [[X:%.*]], <i32 42, i32 42>
+; CHECK-NEXT:    [[TMP1:%.*]] = urem <2 x i32> [[X:%.*]], <i32 42, i32 1>
 ; CHECK-NEXT:    [[R:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> undef, <2 x i32> zeroinitializer
 ; CHECK-NEXT:    ret <2 x i32> [[R]]
 ;
   %splat = shufflevector <2 x i32> %x, <2 x i32> undef, <2 x i32> zeroinitializer
   %r = urem <2 x i32> %splat, <i32 42, i32 42>
   ret <2 x i32> %r
 }
 
 define <2 x i32> @srem_splat_constant0(<2 x i32> %x) {
 ; CHECK-LABEL: @srem_splat_constant0(
 ; CHECK-NEXT:    [[SPLAT:%.*]] = shufflevector <2 x i32> [[X:%.*]], <2 x i32> undef, <2 x i32> zeroinitializer
 ; CHECK-NEXT:    [[R:%.*]] = srem <2 x i32> <i32 42, i32 42>, [[SPLAT]]
 ; CHECK-NEXT:    ret <2 x i32> [[R]]
 ;
   %splat = shufflevector <2 x i32> %x, <2 x i32> undef, <2 x i32> zeroinitializer
   %r = srem <2 x i32> <i32 42, i32 42>, %splat
   ret <2 x i32> %r
 }
 
 define <2 x i32> @srem_splat_constant1(<2 x i32> %x) {
 ; CHECK-LABEL: @srem_splat_constant1(
-; CHECK-NEXT:    [[TMP1:%.*]] = srem <2 x i32> [[X:%.*]], <i32 42, i32 42>
+; CHECK-NEXT:    [[TMP1:%.*]] = srem <2 x i32> [[X:%.*]], <i32 42, i32 1>
 ; CHECK-NEXT:    [[R:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> undef, <2 x i32> zeroinitializer
 ; CHECK-NEXT:    ret <2 x i32> [[R]]
 ;
   %splat = shufflevector <2 x i32> %x, <2 x i32> undef, <2 x i32> zeroinitializer
   %r = srem <2 x i32> %splat, <i32 42, i32 42>
   ret <2 x i32> %r
 }
 
 define <2 x i32> @udiv_splat_constant0(<2 x i32> %x) {
 ; CHECK-LABEL: @udiv_splat_constant0(
 ; CHECK-NEXT:    [[SPLAT:%.*]] = shufflevector <2 x i32> [[X:%.*]], <2 x i32> undef, <2 x i32> zeroinitializer
 ; CHECK-NEXT:    [[R:%.*]] = udiv <2 x i32> <i32 42, i32 42>, [[SPLAT]]
 ; CHECK-NEXT:    ret <2 x i32> [[R]]
 ;
   %splat = shufflevector <2 x i32> %x, <2 x i32> undef, <2 x i32> zeroinitializer
   %r = udiv <2 x i32> <i32 42, i32 42>, %splat
   ret <2 x i32> %r
 }
 
 define <2 x i32> @udiv_splat_constant1(<2 x i32> %x) {
 ; CHECK-LABEL: @udiv_splat_constant1(
-; CHECK-NEXT:    [[TMP1:%.*]] = udiv <2 x i32> [[X:%.*]], <i32 42, i32 42>
+; CHECK-NEXT:    [[TMP1:%.*]] = udiv <2 x i32> [[X:%.*]], <i32 42, i32 1>
 ; CHECK-NEXT:    [[R:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> undef, <2 x i32> zeroinitializer
 ; CHECK-NEXT:    ret <2 x i32> [[R]]
 ;
   %splat = shufflevector <2 x i32> %x, <2 x i32> undef, <2 x i32> zeroinitializer
   %r = udiv <2 x i32> %splat, <i32 42, i32 42>
   ret <2 x i32> %r
 }
 
 define <2 x i32> @sdiv_splat_constant0(<2 x i32> %x) {
 ; CHECK-LABEL: @sdiv_splat_constant0(
 ; CHECK-NEXT:    [[SPLAT:%.*]] = shufflevector <2 x i32> [[X:%.*]], <2 x i32> undef, <2 x i32> zeroinitializer
 ; CHECK-NEXT:    [[R:%.*]] = sdiv <2 x i32> <i32 42, i32 42>, [[SPLAT]]
 ; CHECK-NEXT:    ret <2 x i32> [[R]]
 ;
   %splat = shufflevector <2 x i32> %x, <2 x i32> undef, <2 x i32> zeroinitializer
   %r = sdiv <2 x i32> <i32 42, i32 42>, %splat
   ret <2 x i32> %r
 }
 
 define <2 x i32> @sdiv_splat_constant1(<2 x i32> %x) {
 ; CHECK-LABEL: @sdiv_splat_constant1(
-; CHECK-NEXT:    [[TMP1:%.*]] = sdiv <2 x i32> [[X:%.*]], <i32 42, i32 42>
+; CHECK-NEXT:    [[TMP1:%.*]] = sdiv <2 x i32> [[X:%.*]], <i32 42, i32 1>
 ; CHECK-NEXT:    [[R:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> undef, <2 x i32> zeroinitializer
 ; CHECK-NEXT:    ret <2 x i32> [[R]]
 ;
   %splat = shufflevector <2 x i32> %x, <2 x i32> undef, <2 x i32> zeroinitializer
   %r = sdiv <2 x i32> %splat, <i32 42, i32 42>
   ret <2 x i32> %r
 }
 

 ;
   %splat = shufflevector <2 x i32> %x, <2 x i32> undef, <2 x i32> zeroinitializer
   %r = xor <2 x i32> %splat, <i32 42, i32 42>
   ret <2 x i32> %r
 }
 
 define <2 x float> @fadd_splat_constant(<2 x float> %x) {
 ; CHECK-LABEL: @fadd_splat_constant(
-; CHECK-NEXT:    [[TMP1:%.*]] = fadd <2 x float> [[X:%.*]], <float 4.200000e+01, float 4.200000e+01>
+; CHECK-NEXT:    [[TMP1:%.*]] = fadd <2 x float> [[X:%.*]], <float 4.200000e+01, float undef>
 ; CHECK-NEXT:    [[R:%.*]] = shufflevector <2 x float> [[TMP1]], <2 x float> undef, <2 x i32> zeroinitializer
 ; CHECK-NEXT:    ret <2 x float> [[R]]
 ;
   %splat = shufflevector <2 x float> %x, <2 x float> undef, <2 x i32> zeroinitializer
   %r = fadd <2 x float> %splat, <float 42.0, float 42.0>
   ret <2 x float> %r
 }
 
 define <2 x float> @fsub_splat_constant0(<2 x float> %x) {
 ; CHECK-LABEL: @fsub_splat_constant0(
-; CHECK-NEXT:    [[TMP1:%.*]] = fsub <2 x float> <float 4.200000e+01, float 4.200000e+01>, [[X:%.*]]
+; CHECK-NEXT:    [[TMP1:%.*]] = fsub <2 x float> <float 4.200000e+01, float undef>, [[X:%.*]]
 ; CHECK-NEXT:    [[R:%.*]] = shufflevector <2 x float> [[TMP1]], <2 x float> undef, <2 x i32> zeroinitializer
 ; CHECK-NEXT:    ret <2 x float> [[R]]
 ;
   %splat = shufflevector <2 x float> %x, <2 x float> undef, <2 x i32> zeroinitializer
   %r = fsub <2 x float> <float 42.0, float 42.0>, %splat
   ret <2 x float> %r
 }
 
 define <2 x float> @fsub_splat_constant1(<2 x float> %x) {
 ; CHECK-LABEL: @fsub_splat_constant1(
-; CHECK-NEXT:    [[TMP1:%.*]] = fadd <2 x float> [[X:%.*]], <float -4.200000e+01, float -4.200000e+01>
+; CHECK-NEXT:    [[TMP1:%.*]] = fadd <2 x float> [[X:%.*]], <float -4.200000e+01, float 0x7FF8000000000000>
 ; CHECK-NEXT:    [[R:%.*]] = shufflevector <2 x float> [[TMP1]], <2 x float> undef, <2 x i32> zeroinitializer
 ; CHECK-NEXT:    ret <2 x float> [[R]]
 ;
   %splat = shufflevector <2 x float> %x, <2 x float> undef, <2 x i32> zeroinitializer
   %r = fsub <2 x float> %splat, <float 42.0, float 42.0>
   ret <2 x float> %r
 }
 
 define <2 x float> @fmul_splat_constant(<2 x float> %x) {
 ; CHECK-LABEL: @fmul_splat_constant(
-; CHECK-NEXT:    [[TMP1:%.*]] = fmul <2 x float> [[X:%.*]], <float 4.200000e+01, float 4.200000e+01>
+; CHECK-NEXT:    [[TMP1:%.*]] = fmul <2 x float> [[X:%.*]], <float 4.200000e+01, float undef>
 ; CHECK-NEXT:    [[R:%.*]] = shufflevector <2 x float> [[TMP1]], <2 x float> undef, <2 x i32> zeroinitializer
 ; CHECK-NEXT:    ret <2 x float> [[R]]
 ;
   %splat = shufflevector <2 x float> %x, <2 x float> undef, <2 x i32> zeroinitializer
   %r = fmul <2 x float> %splat, <float 42.0, float 42.0>
   ret <2 x float> %r
 }
 
 define <2 x float> @fdiv_splat_constant0(<2 x float> %x) {
 ; CHECK-LABEL: @fdiv_splat_constant0(
-; CHECK-NEXT:    [[TMP1:%.*]] = fdiv <2 x float> <float 4.200000e+01, float 4.200000e+01>, [[X:%.*]]
+; CHECK-NEXT:    [[TMP1:%.*]] = fdiv <2 x float> <float 4.200000e+01, float undef>, [[X:%.*]]
 ; CHECK-NEXT:    [[R:%.*]] = shufflevector <2 x float> [[TMP1]], <2 x float> undef, <2 x i32> zeroinitializer
 ; CHECK-NEXT:    ret <2 x float> [[R]]
 ;
   %splat = shufflevector <2 x float> %x, <2 x float> undef, <2 x i32> zeroinitializer
   %r = fdiv <2 x float> <float 42.0, float 42.0>, %splat
   ret <2 x float> %r
 }
 
 define <2 x float> @fdiv_splat_constant1(<2 x float> %x) {
 ; CHECK-LABEL: @fdiv_splat_constant1(
-; CHECK-NEXT:    [[TMP1:%.*]] = fdiv <2 x float> [[X:%.*]], <float 4.200000e+01, float 4.200000e+01>
+; CHECK-NEXT:    [[TMP1:%.*]] = fdiv <2 x float> [[X:%.*]], <float 4.200000e+01, float undef>
 ; CHECK-NEXT:    [[R:%.*]] = shufflevector <2 x float> [[TMP1]], <2 x float> undef, <2 x i32> zeroinitializer
 ; CHECK-NEXT:    ret <2 x float> [[R]]
 ;
   %splat = shufflevector <2 x float> %x, <2 x float> undef, <2 x i32> zeroinitializer
   %r = fdiv <2 x float> %splat, <float 42.0, float 42.0>
   ret <2 x float> %r
 }
 
 define <2 x float> @frem_splat_constant0(<2 x float> %x) {
 ; CHECK-LABEL: @frem_splat_constant0(
-; CHECK-NEXT:    [[TMP1:%.*]] = frem <2 x float> <float 4.200000e+01, float 4.200000e+01>, [[X:%.*]]
+; CHECK-NEXT:    [[TMP1:%.*]] = frem <2 x float> <float 4.200000e+01, float undef>, [[X:%.*]]
 ; CHECK-NEXT:    [[R:%.*]] = shufflevector <2 x float> [[TMP1]], <2 x float> undef, <2 x i32> zeroinitializer
 ; CHECK-NEXT:    ret <2 x float> [[R]]
 ;
   %splat = shufflevector <2 x float> %x, <2 x float> undef, <2 x i32> zeroinitializer
   %r = frem <2 x float> <float 42.0, float 42.0>, %splat
   ret <2 x float> %r
 }
 
 define <2 x float> @frem_splat_constant1(<2 x float> %x) {
 ; CHECK-LABEL: @frem_splat_constant1(
-; CHECK-NEXT:    [[TMP1:%.*]] = frem <2 x float> [[X:%.*]], <float 4.200000e+01, float 4.200000e+01>
+; CHECK-NEXT:    [[TMP1:%.*]] = frem <2 x float> [[X:%.*]], <float 4.200000e+01, float undef>
 ; CHECK-NEXT:    [[R:%.*]] = shufflevector <2 x float> [[TMP1]], <2 x float> undef, <2 x i32> zeroinitializer
 ; CHECK-NEXT:    ret <2 x float> [[R]]
 ;
   %splat = shufflevector <2 x float> %x, <2 x float> undef, <2 x i32> zeroinitializer
   %r = frem <2 x float> %splat, <float 42.0, float 42.0>
   ret <2 x float> %r
 }