[SVE] Add support for scalable vectorization of loops with selects and cmps

[david-arm]

Summary

I have removed an unnecessary assert in LoopVectorizationCostModel::getInstructionCost
that prevented a cost being calculated for select instructions when using
scalable vectors. In addition, I have changed AArch64TTIImpl::getCmpSelInstrCost
to only do special cost calculations for fixed width vectors and fall
back to the base version for scalable vectors.

I have added a simple cost model test for cmps and selects:

test/Analysis/CostModel/sve-cmpsel.ll

and some simple tests that show we vectorize loops with cmp and select:

test/Transforms/LoopVectorize/AArch64/sve-basic-vec.ll

[david-arm]

llvm/test/Analysis/CostModel/sve-cmpsel.ll

10

Here I've deliberately tried to avoid an explosion of test cases, so for the legal and illegal variants I used different element types.

[dmgreen]

llvm/test/Analysis/CostModel/sve-cmpsel.ll

10

Cost model checks don't really need valid inputs, they can just use undef. It makes adding a lot of them much simpler. See something like llvm/test/Analysis/CostModel/ARM/reduce-smax.ll or llvm/test/Analysis/CostModel/X86/arith-fma.ll. Even though they are adding many (sub-) architectures, they are still managable.

[david-arm]

llvm/test/Transforms/LoopVectorize/AArch64/sve-basic-vec.ll

13

Apologies, just realised these names are terrible! I'll fix them in another patch. I also wonder if autogenerating the CHECK lines really makes sense here.

[david-arm]

• Added more cost model test cases and optimised existing ones.
• Renamed vectorisation tests to something more useful. :) I also reduced the number of CHECK lines as it looked too messy and fragile.

[david-arm]

llvm/test/Analysis/CostModel/sve-cmpsel.ll

10

Thanks for the suggestion @dmgreen!

[dmgreen]

The code changes look simple enough, LGTM.

Up to you if you want to try and reduce the test boilerplate too.

llvm/test/Analysis/CostModel/sve-cmpsel.ll

10

I would combine all these into less functions to reduce all this boilerplate. Maybe one function for the cmp's and one for the sel's? Or maybe split out more into the legal/illegal types, like you have commented. They don't need to actually use the output of the instruction.

It makes adding lots of test much easier to check, but up to you what you think.

llvm/lib/Target/AArch64/AArch64TargetTransformInfo.cpp

int AArch64TTIImpl::getCmpSelInstrCost(unsigned Opcode, Type *ValTy,
   // TODO: Handle other cost kinds.
   if (CostKind != TTI::TCK_RecipThroughput)
     return BaseT::getCmpSelInstrCost(Opcode, ValTy, CondTy, VecPred, CostKind,
                                      I);
 
   int ISD = TLI->InstructionOpcodeToISD(Opcode);
   // We don't lower some vector selects well that are wider than the register
   // width.
-  if (ValTy->isVectorTy() && ISD == ISD::SELECT) {
+  if (isa<FixedVectorType>(ValTy) && ISD == ISD::SELECT) {
     // We would need this many instructions to hide the scalarization happening.
     const int AmortizationCost = 20;
 
     // If VecPred is not set, check if we can get a predicate from the context
     // instruction, if its type matches the requested ValTy.
     if (VecPred == CmpInst::BAD_ICMP_PREDICATE && I && I->getType() == ValTy) {
       CmpInst::Predicate CurrentPred;
       if (match(I, m_Select(m_Cmp(CurrentPred, m_Value(), m_Value()), m_Value(),
if (isa<FixedVectorType>(ValTy) && ISD == ISD::SELECT) {
     EVT SelValTy = TLI->getValueType(DL, ValTy);
     if (SelCondTy.isSimple() && SelValTy.isSimple()) {
       if (const auto *Entry = ConvertCostTableLookup(VectorSelectTbl, ISD,
                                                      SelCondTy.getSimpleVT(),
                                                      SelValTy.getSimpleVT()))
         return Entry->Cost;
     }
   }
+  // The base case handles scalable vectors fine for now, since it treats the
+  // cost as 1 * legalization cost.
   return BaseT::getCmpSelInstrCost(Opcode, ValTy, CondTy, VecPred, CostKind, I);
 }
 
 AArch64TTIImpl::TTI::MemCmpExpansionOptions
 AArch64TTIImpl::enableMemCmpExpansion(bool OptSize, bool IsZeroCmp) const {
   TTI::MemCmpExpansionOptions Options;
   if (ST->requiresStrictAlign()) {
     // TODO: Add cost modeling for strict align. Misaligned loads expand to

llvm/lib/Transforms/Vectorize/LoopVectorize.cpp

return N * TTI.getArithmeticInstrCost(
                    TargetTransformInfo::OP_None, TargetTransformInfo::OP_None,
                    I->getOperand(0), I);
   }
   case Instruction::Select: {
     SelectInst *SI = cast<SelectInst>(I);
     const SCEV *CondSCEV = SE->getSCEV(SI->getCondition());
     bool ScalarCond = (SE->isLoopInvariant(CondSCEV, TheLoop));
     Type *CondTy = SI->getCondition()->getType();
-    if (!ScalarCond) {
-      assert(!VF.isScalable() && "VF is assumed to be non scalable.");
+    if (!ScalarCond)
       CondTy = VectorType::get(CondTy, VF);
-    }
     return TTI.getCmpSelInstrCost(I->getOpcode(), VectorTy, CondTy,
                                   CmpInst::BAD_ICMP_PREDICATE, CostKind, I);
   }
   case Instruction::ICmp:
   case Instruction::FCmp: {
     Type *ValTy = I->getOperand(0)->getType();
     Instruction *Op0AsInstruction = dyn_cast<Instruction>(I->getOperand(0));
     if (canTruncateToMinimalBitwidth(Op0AsInstruction, VF))

llvm/test/Analysis/CostModel/sve-cmpsel.ll

+; RUN: opt -cost-model -analyze -mtriple=aarch64--linux-gnu -mattr=+sve  < %s 2>%t | FileCheck %s
+
+; RUN: FileCheck --check-prefix=WARN --allow-empty %s <%t
+
+; If this check fails please read test/CodeGen/AArch64/README for instructions on how to resolve it.
+; WARN-NOT: warning
+
+; Check icmp for a legal integer vector.
+define <vscale x 2 x i1> @cmp_nxv2i64() {
+; CHECK-LABEL: 'cmp_nxv2i64'
+; CHECK: Cost Model: Found an estimated cost of 1 for instruction:   %res = icmp ne <vscale x 2 x i64> undef, undef
+; CHECK: Cost Model: Found an estimated cost of 0 for instruction:   ret <vscale x 2 x i1> %res
+  %res = icmp ne <vscale x 2 x i64> undef, undef
+  ret <vscale x 2 x i1> %res
+}
+
+define <vscale x 4 x i1> @cmp_nxv4i32() {
+; CHECK-LABEL: 'cmp_nxv4i32'
+; CHECK: Cost Model: Found an estimated cost of 1 for instruction:   %res = icmp ne <vscale x 4 x i32> undef, undef
+; CHECK: Cost Model: Found an estimated cost of 0 for instruction:   ret <vscale x 4 x i1> %res
+  %res = icmp ne <vscale x 4 x i32> undef, undef
+  ret <vscale x 4 x i1> %res
+}
+
+define <vscale x 8 x i1> @cmp_nxv8i16() {
+; CHECK-LABEL: 'cmp_nxv8i16'
+; CHECK: Cost Model: Found an estimated cost of 1 for instruction:   %res = icmp ne <vscale x 8 x i16> undef, undef
+; CHECK: Cost Model: Found an estimated cost of 0 for instruction:   ret <vscale x 8 x i1> %res
+  %res = icmp ne <vscale x 8 x i16> undef, undef
+  ret <vscale x 8 x i1> %res
+}
+
+define <vscale x 16 x i1> @cmp_nxv16i8() {
+; CHECK-LABEL: 'cmp_nxv16i8'
+; CHECK: Cost Model: Found an estimated cost of 1 for instruction:   %res = icmp ne <vscale x 16 x i8> undef, undef
+; CHECK: Cost Model: Found an estimated cost of 0 for instruction:   ret <vscale x 16 x i1> %res
+  %res = icmp ne <vscale x 16 x i8> undef, undef
+  ret <vscale x 16 x i1> %res
+}
+
+; Check icmp for an illegal integer vector.
+define <vscale x 4 x i1> @cmp_nxv4i64() {
+; CHECK-LABEL: 'cmp_nxv4i64'
+; CHECK: Cost Model: Found an estimated cost of 2 for instruction:   %res = icmp ne <vscale x 4 x i64> undef, undef
+; CHECK: Cost Model: Found an estimated cost of 0 for instruction:   ret <vscale x 4 x i1> %res
+  %res = icmp ne <vscale x 4 x i64> undef, undef
+  ret <vscale x 4 x i1> %res
+}
+
+; Check icmp for predicate vectors.
+define <vscale x 2 x i1> @cmp_nxv2i1() {
+; CHECK-LABEL: 'cmp_nxv2i1'
+; CHECK: Cost Model: Found an estimated cost of 1 for instruction:   %res = icmp ne <vscale x 2 x i1> undef, undef
+; CHECK: Cost Model: Found an estimated cost of 0 for instruction:   ret <vscale x 2 x i1> %res
+  %res = icmp ne <vscale x 2 x i1> undef, undef
+  ret <vscale x 2 x i1> %res
+}
+
+define <vscale x 4 x i1> @cmp_nxv4i1() {
+; CHECK-LABEL: 'cmp_nxv4i1'
+; CHECK: Cost Model: Found an estimated cost of 1 for instruction:   %res = icmp ne <vscale x 4 x i1> undef, undef
+; CHECK: Cost Model: Found an estimated cost of 0 for instruction:   ret <vscale x 4 x i1> %res
+  %res = icmp ne <vscale x 4 x i1> undef, undef
+  ret <vscale x 4 x i1> %res
+}
+
+define <vscale x 8 x i1> @cmp_nxv8i1() {
+; CHECK-LABEL: 'cmp_nxv8i1'
+; CHECK: Cost Model: Found an estimated cost of 1 for instruction:   %res = icmp ne <vscale x 8 x i1> undef, undef
+; CHECK: Cost Model: Found an estimated cost of 0 for instruction:   ret <vscale x 8 x i1> %res
+  %res = icmp ne <vscale x 8 x i1> undef, undef
+  ret <vscale x 8 x i1> %res
+}
+
+define <vscale x 16 x i1> @cmp_nxv16i1() {
+; CHECK-LABEL: 'cmp_nxv16i1'
+; CHECK: Cost Model: Found an estimated cost of 1 for instruction:   %res = icmp ne <vscale x 16 x i1> undef, undef
+; CHECK: Cost Model: Found an estimated cost of 0 for instruction:   ret <vscale x 16 x i1> %res
+  %res = icmp ne <vscale x 16 x i1> undef, undef
+  ret <vscale x 16 x i1> %res
+}
+
+; Check icmp for an illegal predicate vector.
+define <vscale x 32 x i1> @cmp_nxv32i1() {
+; CHECK-LABEL: 'cmp_nxv32i1'
+; CHECK: Cost Model: Found an estimated cost of 2 for instruction:   %res = icmp ne <vscale x 32 x i1> undef, undef
+; CHECK: Cost Model: Found an estimated cost of 0 for instruction:   ret <vscale x 32 x i1> %res
+  %res = icmp ne <vscale x 32 x i1> undef, undef
+  ret <vscale x 32 x i1> %res
+}
+
+; Check fcmp for a legal FP vector
+define <vscale x 2 x i1> @cmp_nxv2f64() {
+; CHECK-LABEL: 'cmp_nxv2f64'
+; CHECK: Cost Model: Found an estimated cost of 1 for instruction:   %res = fcmp oge <vscale x 2 x double> undef, undef
+; CHECK: Cost Model: Found an estimated cost of 0 for instruction:   ret <vscale x 2 x i1> %res
+  %res = fcmp oge <vscale x 2 x double> undef, undef
+  ret <vscale x 2 x i1> %res
+}
+
+define <vscale x 4 x i1> @cmp_nxv4f32() {
+; CHECK-LABEL: 'cmp_nxv4f32'
+; CHECK: Cost Model: Found an estimated cost of 1 for instruction:   %res = fcmp oge <vscale x 4 x float> undef, undef
+; CHECK: Cost Model: Found an estimated cost of 0 for instruction:   ret <vscale x 4 x i1> %res
+  %res = fcmp oge <vscale x 4 x float> undef, undef
+  ret <vscale x 4 x i1> %res
+}
+
+define <vscale x 8 x i1> @cmp_nxv8f16() {
+; CHECK-LABEL: 'cmp_nxv8f16'
+; CHECK: Cost Model: Found an estimated cost of 1 for instruction:   %res = fcmp oge <vscale x 8 x half> undef, undef
+; CHECK: Cost Model: Found an estimated cost of 0 for instruction:   ret <vscale x 8 x i1> %res
+  %res = fcmp oge <vscale x 8 x half> undef, undef
+  ret <vscale x 8 x i1> %res
+}
+
+define <vscale x 8 x i1> @cmp_nxv8bf16() #0 {
+; CHECK-LABEL: 'cmp_nxv8bf16'
+; CHECK: Cost Model: Found an estimated cost of 1 for instruction:   %res = fcmp oge <vscale x 8 x bfloat> undef, undef
+; CHECK: Cost Model: Found an estimated cost of 0 for instruction:   ret <vscale x 8 x i1> %res
+  %res = fcmp oge <vscale x 8 x bfloat> undef, undef
+  ret <vscale x 8 x i1> %res
+}
+
+; Check fcmp for an illegal FP vector
+define <vscale x 16 x i1> @cmp_nxv16f16() {
+; CHECK-LABEL: 'cmp_nxv16f16'
+; CHECK: Cost Model: Found an estimated cost of 2 for instruction:   %res = fcmp oge <vscale x 16 x half> undef, undef
+; CHECK: Cost Model: Found an estimated cost of 0 for instruction:   ret <vscale x 16 x i1> %res
+  %res = fcmp oge <vscale x 16 x half> undef, undef
+  ret <vscale x 16 x i1> %res
+}
+
+; Check select for a legal integer vector
+define <vscale x 2 x i64> @sel_nxv2i64() {
+; CHECK-LABEL: 'sel_nxv2i64'
+; CHECK: Cost Model: Found an estimated cost of 1 for instruction:   %res = select <vscale x 2 x i1> undef, <vscale x 2 x i64> undef, <vscale x 2 x i64> undef
+; CHECK: Cost Model: Found an estimated cost of 0 for instruction:   ret <vscale x 2 x i64> %res
+  %res = select <vscale x 2 x i1> undef, <vscale x 2 x i64> undef, <vscale x 2 x i64> undef
+  ret <vscale x 2 x i64> %res
+}
+
+define <vscale x 4 x i32> @sel_nxv4i32() {
+; CHECK-LABEL: 'sel_nxv4i32'
+; CHECK: Cost Model: Found an estimated cost of 1 for instruction:   %res = select <vscale x 4 x i1> undef, <vscale x 4 x i32> undef, <vscale x 4 x i32> undef
+; CHECK: Cost Model: Found an estimated cost of 0 for instruction:   ret <vscale x 4 x i32> %res
+  %res = select <vscale x 4 x i1> undef, <vscale x 4 x i32> undef, <vscale x 4 x i32> undef
+  ret <vscale x 4 x i32> %res
+}
+
+define <vscale x 8 x i16> @sel_nxv8i16() {
+; CHECK-LABEL: 'sel_nxv8i16'
+; CHECK: Cost Model: Found an estimated cost of 1 for instruction:   %res = select <vscale x 8 x i1> undef, <vscale x 8 x i16> undef, <vscale x 8 x i16> undef
+; CHECK: Cost Model: Found an estimated cost of 0 for instruction:   ret <vscale x 8 x i16> %res
+  %res = select <vscale x 8 x i1> undef, <vscale x 8 x i16> undef, <vscale x 8 x i16> undef
+  ret <vscale x 8 x i16> %res
+}
+
+define <vscale x 16 x i8> @sel_nxv16i8() {
+; CHECK-LABEL: 'sel_nxv16i8'
+; CHECK: Cost Model: Found an estimated cost of 1 for instruction:   %res = select <vscale x 16 x i1> undef, <vscale x 16 x i8> undef, <vscale x 16 x i8> undef
+; CHECK: Cost Model: Found an estimated cost of 0 for instruction:   ret <vscale x 16 x i8> %res
+  %res = select <vscale x 16 x i1> undef, <vscale x 16 x i8> undef, <vscale x 16 x i8> undef
+  ret <vscale x 16 x i8> %res
+}
+
+; Check select for an illegal integer vector
+define <vscale x 16 x i16> @sel_nxv16i16() {
+; CHECK-LABEL: 'sel_nxv16i16'
+; CHECK: Cost Model: Found an estimated cost of 2 for instruction:   %res = select <vscale x 16 x i1> undef, <vscale x 16 x i16> undef, <vscale x 16 x i16> undef
+; CHECK: Cost Model: Found an estimated cost of 0 for instruction:   ret <vscale x 16 x i16> %res
+  %res = select <vscale x 16 x i1> undef, <vscale x 16 x i16> undef, <vscale x 16 x i16> undef
+  ret <vscale x 16 x i16> %res
+}
+
+; Check select for a legal FP vector
+define <vscale x 2 x double> @sel_nxv2f64() {
+; CHECK-LABEL: 'sel_nxv2f64'
+; CHECK: Cost Model: Found an estimated cost of 1 for instruction:   %res = select <vscale x 2 x i1> undef, <vscale x 2 x double> undef, <vscale x 2 x double> undef
+; CHECK: Cost Model: Found an estimated cost of 0 for instruction:   ret <vscale x 2 x double> %res
+  %res = select <vscale x 2 x i1> undef, <vscale x 2 x double> undef, <vscale x 2 x double> undef
+  ret <vscale x 2 x double> %res
+}
+
+define <vscale x 4 x float> @sel_nxv4f32() {
+; CHECK-LABEL: 'sel_nxv4f32'
+; CHECK: Cost Model: Found an estimated cost of 1 for instruction:   %res = select <vscale x 4 x i1> undef, <vscale x 4 x float> undef, <vscale x 4 x float> undef
+; CHECK: Cost Model: Found an estimated cost of 0 for instruction:   ret <vscale x 4 x float> %res
+  %res = select <vscale x 4 x i1> undef, <vscale x 4 x float> undef, <vscale x 4 x float> undef
+  ret <vscale x 4 x float> %res
+}
+
+define <vscale x 8 x half> @sel_nxv8f16() {
+; CHECK-LABEL: 'sel_nxv8f16'
+; CHECK: Cost Model: Found an estimated cost of 1 for instruction:   %res = select <vscale x 8 x i1> undef, <vscale x 8 x half> undef, <vscale x 8 x half> undef
+; CHECK: Cost Model: Found an estimated cost of 0 for instruction:   ret <vscale x 8 x half> %res
+  %res = select <vscale x 8 x i1> undef, <vscale x 8 x half> undef, <vscale x 8 x half> undef
+  ret <vscale x 8 x half> %res
+}
+
+define <vscale x 8 x bfloat> @sel_nxv8bf16() #0 {
+; CHECK-LABEL: 'sel_nxv8bf16'
+; CHECK: Cost Model: Found an estimated cost of 1 for instruction:   %res = select <vscale x 8 x i1> undef, <vscale x 8 x bfloat> undef, <vscale x 8 x bfloat> undef
+; CHECK: Cost Model: Found an estimated cost of 0 for instruction:   ret <vscale x 8 x bfloat> %res
+  %res = select <vscale x 8 x i1> undef, <vscale x 8 x bfloat> undef, <vscale x 8 x bfloat> undef
+  ret <vscale x 8 x bfloat> %res
+}
+
+; Check select for an illegal FP vector
+define <vscale x 8 x float> @sel_nxv8f32() {
+; CHECK-LABEL: 'sel_nxv8f32'
+; CHECK: Cost Model: Found an estimated cost of 2 for instruction:   %res = select <vscale x 8 x i1> undef, <vscale x 8 x float> undef, <vscale x 8 x float> undef
+; CHECK: Cost Model: Found an estimated cost of 0 for instruction:   ret <vscale x 8 x float> %res
+  %res = select <vscale x 8 x i1> undef, <vscale x 8 x float> undef, <vscale x 8 x float> undef
+  ret <vscale x 8 x float> %res
+}
+
+; Check select for a legal predicate vector
+define <vscale x 2 x i1> @sel_nxv2i1() {
+; CHECK-LABEL: 'sel_nxv2i1'
+; CHECK: Cost Model: Found an estimated cost of 1 for instruction:   %res = select <vscale x 2 x i1> undef, <vscale x 2 x i1> undef, <vscale x 2 x i1> undef
+; CHECK: Cost Model: Found an estimated cost of 0 for instruction:   ret <vscale x 2 x i1> %res
+  %res = select <vscale x 2 x i1> undef, <vscale x 2 x i1> undef, <vscale x 2 x i1> undef
+  ret <vscale x 2 x i1> %res
+}
+
+define <vscale x 4 x i1> @sel_nxv4i1() {
+; CHECK-LABEL: 'sel_nxv4i1'
+; CHECK: Cost Model: Found an estimated cost of 1 for instruction:   %res = select <vscale x 4 x i1> undef, <vscale x 4 x i1> undef, <vscale x 4 x i1> undef
+; CHECK: Cost Model: Found an estimated cost of 0 for instruction:   ret <vscale x 4 x i1> %res
+  %res = select <vscale x 4 x i1> undef, <vscale x 4 x i1> undef, <vscale x 4 x i1> undef
+  ret <vscale x 4 x i1> %res
+}
+
+define <vscale x 8 x i1> @sel_nxv8i1() {
+; CHECK-LABEL: 'sel_nxv8i1'
+; CHECK: Cost Model: Found an estimated cost of 1 for instruction:   %res = select <vscale x 8 x i1> undef, <vscale x 8 x i1> undef, <vscale x 8 x i1> undef
+; CHECK: Cost Model: Found an estimated cost of 0 for instruction:   ret <vscale x 8 x i1> %res
+  %res = select <vscale x 8 x i1> undef, <vscale x 8 x i1> undef, <vscale x 8 x i1> undef
+  ret <vscale x 8 x i1> %res
+}
+
+define <vscale x 16 x i1> @sel_nxv16i1() {
+; CHECK-LABEL: 'sel_nxv16i1'
+; CHECK: Cost Model: Found an estimated cost of 1 for instruction:   %res = select <vscale x 16 x i1> undef, <vscale x 16 x i1> undef, <vscale x 16 x i1> undef
+; CHECK: Cost Model: Found an estimated cost of 0 for instruction:   ret <vscale x 16 x i1> %res
+  %res = select <vscale x 16 x i1> undef, <vscale x 16 x i1> undef, <vscale x 16 x i1> undef
+  ret <vscale x 16 x i1> %res
+}
+
+; Check select for an illegal predicate vector
+define <vscale x 32 x i1> @sel_nxv32i1() {
+; CHECK-LABEL: 'sel_nxv32i1'
+; CHECK: Cost Model: Found an estimated cost of 2 for instruction:   %res = select <vscale x 32 x i1> undef, <vscale x 32 x i1> undef, <vscale x 32 x i1> undef
+; CHECK: Cost Model: Found an estimated cost of 0 for instruction:   ret <vscale x 32 x i1> %res
+  %res = select <vscale x 32 x i1> undef, <vscale x 32 x i1> undef, <vscale x 32 x i1> undef
+  ret <vscale x 32 x i1> %res
+}
+
+attributes #0 = { "target-features"="+sve,+bf16" }

llvm/test/Transforms/LoopVectorize/AArch64/sve-basic-vec.ll

+; RUN: opt -loop-vectorize -dce -instcombine -mtriple aarch64-linux-gnu -mattr=+sve < %s -S 2>%t | FileCheck %s
+
+; RUN: FileCheck --check-prefix=WARN --allow-empty %s <%t
+
+; If this check fails please read test/CodeGen/AArch64/README for instructions on how to resolve it.
+; WARN-NOT: warning
+
+target datalayout = "e-m:e-i8:8:32-i16:16:32-i64:64-i128:128-n32:64-S128"
+target triple = "aarch64-unknown-linux-gnu"
+
+define void @cmpsel_i32(i32* noalias nocapture %a, i32* noalias nocapture readonly %b, i64 %n) {
+; CHECK-LABEL: @cmpsel_i32(
+; CHECK-NEXT:  entry:
+; CHECK:       vector.body:
+; CHECK:         [[WIDE_LOAD:%.*]] = load <vscale x 4 x i32>, <vscale x 4 x i32>* {{.*}}, align 4
+; CHECK-NEXT:    [[TMP1:%.*]] = icmp eq <vscale x 4 x i32> [[WIDE_LOAD]], shufflevector (<vscale x 4 x i32> insertelement (<vscale x 4 x i32> poison, i32 0, i32 0), <vscale x 4 x i32> poison, <vscale x 4 x i32> zeroinitializer)
+; CHECK-NEXT:    [[TMP2:%.*]] = select <vscale x 4 x i1> [[TMP1]], <vscale x 4 x i32> shufflevector (<vscale x 4 x i32> insertelement (<vscale x 4 x i32> poison, i32 2, i32 0), <vscale x 4 x i32> poison, <vscale x 4 x i32> zeroinitializer), <vscale x 4 x i32> shufflevector (<vscale x 4 x i32> insertelement (<vscale x 4 x i32> poison, i32 10, i32 0), <vscale x 4 x i32> poison, <vscale x 4 x i32> zeroinitializer)
+; CHECK:         store <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32>* {{.*}}, align 4
+;
+entry:
+  %cmp7 = icmp sgt i64 %n, 0
+  br i1 %cmp7, label %for.body, label %for.end
+
+for.body:                                         ; preds = %entry, %for.body
+  %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]
+  %arrayidx = getelementptr inbounds i32, i32* %b, i64 %indvars.iv
+  %0 = load i32, i32* %arrayidx, align 4
+  %tobool.not = icmp eq i32 %0, 0
+  %cond = select i1 %tobool.not, i32 2, i32 10
+  %arrayidx2 = getelementptr inbounds i32, i32* %a, i64 %indvars.iv
+  store i32 %cond, i32* %arrayidx2, align 4
+  %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
+  %exitcond.not = icmp eq i64 %indvars.iv.next, %n
+  br i1 %exitcond.not, label %for.end.loopexit, label %for.body, !llvm.loop !0
+
+for.end.loopexit:                                 ; preds = %for.body
+  br label %for.end
+
+for.end:                                          ; preds = %for.end.loopexit, %entry
+  ret void
+}
+
+define void @cmpsel_f32(float* noalias nocapture %a, float* noalias nocapture readonly %b, i64 %n) {
+; CHECK-LABEL: @cmpsel_f32(
+; CHECK-NEXT:  entry:
+; CHECK:       vector.body:
+; CHECK:         [[WIDE_LOAD:%.*]] = load <vscale x 4 x float>, <vscale x 4 x float>* {{.*}}, align 4
+; CHECK-NEXT:    [[TMP1:%.*]] = fcmp ogt <vscale x 4 x float> [[WIDE_LOAD]], shufflevector (<vscale x 4 x float> insertelement (<vscale x 4 x float> poison, float 3.000000e+00, i32 0), <vscale x 4 x float> poison, <vscale x 4 x i32> zeroinitializer)
+; CHECK-NEXT:    [[TMP2:%.*]] = select <vscale x 4 x i1> [[TMP1]], <vscale x 4 x float> shufflevector (<vscale x 4 x float> insertelement (<vscale x 4 x float> poison, float 1.000000e+01, i32 0), <vscale x 4 x float> poison, <vscale x 4 x i32> zeroinitializer), <vscale x 4 x float> shufflevector (<vscale x 4 x float> insertelement (<vscale x 4 x float> poison, float 2.000000e+00, i32 0), <vscale x 4 x float> poison, <vscale x 4 x i32> zeroinitializer)
+; CHECK:         store <vscale x 4 x float> [[TMP2]], <vscale x 4 x float>* {{.*}}, align 4
+
+entry:
+  %cmp8 = icmp sgt i64 %n, 0
+  br i1 %cmp8, label %for.body, label %for.end
+
+for.body:                                         ; preds = %entry, %for.body
+  %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]
+  %arrayidx = getelementptr inbounds float, float* %b, i64 %indvars.iv
+  %0 = load float, float* %arrayidx, align 4
+  %cmp1 = fcmp ogt float %0, 3.000000e+00
+  %conv = select i1 %cmp1, float 1.000000e+01, float 2.000000e+00
+  %arrayidx3 = getelementptr inbounds float, float* %a, i64 %indvars.iv
+  store float %conv, float* %arrayidx3, align 4
+  %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
+  %exitcond.not = icmp eq i64 %indvars.iv.next, %n
+  br i1 %exitcond.not, label %for.end, label %for.body, !llvm.loop !6
+
+for.end:                                          ; preds = %for.body, %entry
+  ret void
+}
+
+!0 = distinct !{!0, !1, !2, !3, !4, !5}
+!1 = !{!"llvm.loop.mustprogress"}
+!2 = !{!"llvm.loop.vectorize.width", i32 4}
+!3 = !{!"llvm.loop.vectorize.scalable.enable", i1 true}
+!4 = !{!"llvm.loop.interleave.count", i32 1}
+!5 = !{!"llvm.loop.vectorize.enable", i1 true}
+!6 = distinct !{!6, !1, !2, !3, !4, !5}