[VPlan] Simplify redundant trunc (zext A) pairs to A.

[fhahn]

Summary

Add simplification for redundant trunc(zext A) pairs. Generally apply a
transform from D149903.

Depends on D159200.

[Ayal]

llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp

824

rename getTypeSizeInBits()?

Worth introducing getType(), possibly as a method of VPValue?

Callers could then get its scalar size in bits if interested, although in this case the types themselves should arguably be compared rather than their scalar sizes in bits?

829

Should this cast case be placed first?

832

(hence getType() may be more natural.)

846–852

("if Ty is equal to the type of A" also refers to types rather than their scalar size in bits.)

llvm/test/Transforms/LoopVectorize/if-pred-stores.ll

542

This "removeRedundantCasts" elimination seems to catch opportunities in the original (unoptimized) code rather than opportunities generated by LV itself, as in Mul-by-1/Add-0. But that may change once bitwidth minimization takes place as VPlan2VPlan transform (D149903)?

[fhahn]

Rebase and address comments, thanks!

ping :)

llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp

824

I think in general there we need to properly infer the types and not rely on the underlying value, see https://github.com/llvm/llvm-project/pull/69013

829

Reordered.

llvm/test/Transforms/LoopVectorize/if-pred-stores.ll

542

Yes, this will be enabled/used by D149903

[Ayal]

llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp

824

Worth a TODO to replace relying on underlying value with type analysis?

Should we return and compare types rather than their sizes in bits? For RAUW types must match not only their size in bits, although we are dealing with integer types only here. As commented below: "... if Ty is equal to the type of A".

848

nit: A in the comment does not match A above. Code suggested below setting A = ROp0->getOperand(0) tried to align them. Could be further aligned by doing something like:

// Simplify (trunc (zext A) to Ty) -> A if Ty is equal to the type of A.
VPRecipeBase *Zext = R.getOperand(0)->getDefiningRecipe();
if (!Zext || getOpcodeForRecipe(*Zext) != Instruction::ZExt)
  break;
VPValue *A = Zext->getOperand(0);
VPValue *Trunc = R.getVPSingleValue();
if (getType(Trunc) == getType(A))
  Trunc->replaceAllUsesWith(A);

[fhahn]

Address comments, thanks!

[fhahn]

llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp

824

Adjusted, thanks!

848

Adjusted, thanks!

[Ayal]

LGTM, thanks for accommodating!

[uabelho]

Hi!

The following starts crashing with this patch:

opt -passes="loop-vectorize" bbi-88061.ll -o /dev/null

It yields

PLEASE submit a bug report to https://github.com/llvm/llvm-project/issues/ and include the crash backtrace.
Stack dump:
0.	Program arguments: ../../main-github/llvm/build-all/bin/opt -passes=loop-vectorize bbi-88061.ll -o /dev/null
 #0 0x0000560ede4fecf7 llvm::sys::PrintStackTrace(llvm::raw_ostream&, int) (../../main-github/llvm/build-all/bin/opt+0x2f49cf7)
 #1 0x0000560ede4fc84e llvm::sys::RunSignalHandlers() (../../main-github/llvm/build-all/bin/opt+0x2f4784e)
 #2 0x0000560ede4ff3bf SignalHandler(int) Signals.cpp:0:0
 #3 0x00007f353ef87630 __restore_rt sigaction.c:0:0
 #4 0x0000560edea0b1d4 llvm::VPlanTransforms::optimize(llvm::VPlan&, llvm::ScalarEvolution&) (../../main-github/llvm/build-all/bin/opt+0x34561d4)
 #5 0x0000560ede8f6ec5 llvm::LoopVectorizationPlanner::buildVPlansWithVPRecipes(llvm::ElementCount, llvm::ElementCount) (../../main-github/llvm/build-all/bin/opt+0x3341ec5)
 #6 0x0000560ede8f659c llvm::LoopVectorizationPlanner::plan(llvm::ElementCount, unsigned int) (../../main-github/llvm/build-all/bin/opt+0x334159c)
 #7 0x0000560ede9098ab llvm::LoopVectorizePass::processLoop(llvm::Loop*) (../../main-github/llvm/build-all/bin/opt+0x33548ab)
 #8 0x0000560ede90f5b3 llvm::LoopVectorizePass::runImpl(llvm::Function&, llvm::ScalarEvolution&, llvm::LoopInfo&, llvm::TargetTransformInfo&, llvm::DominatorTree&, llvm::BlockFrequencyInfo*, llvm::TargetLibraryInfo*, llvm::DemandedBits&, llvm::AssumptionCache&, llvm::LoopAccessInfoManager&, llvm::OptimizationRemarkEmitter&, llvm::ProfileSummaryInfo*) (../../main-github/llvm/build-all/bin/opt+0x335a5b3)
 #9 0x0000560ede90ff18 llvm::LoopVectorizePass::run(llvm::Function&, llvm::AnalysisManager<llvm::Function>&) (../../main-github/llvm/build-all/bin/opt+0x335af18)
#10 0x0000560ede71ce4d llvm::detail::PassModel<llvm::Function, llvm::LoopVectorizePass, llvm::PreservedAnalyses, llvm::AnalysisManager<llvm::Function>>::run(llvm::Function&, llvm::AnalysisManager<llvm::Function>&) (../../main-github/llvm/build-all/bin/opt+0x3167e4d)
#11 0x0000560eddf1df64 llvm::PassManager<llvm::Function, llvm::AnalysisManager<llvm::Function>>::run(llvm::Function&, llvm::AnalysisManager<llvm::Function>&) (../../main-github/llvm/build-all/bin/opt+0x2968f64)
#12 0x0000560edc2e228d llvm::detail::PassModel<llvm::Function, llvm::PassManager<llvm::Function, llvm::AnalysisManager<llvm::Function>>, llvm::PreservedAnalyses, llvm::AnalysisManager<llvm::Function>>::run(llvm::Function&, llvm::AnalysisManager<llvm::Function>&) (../../main-github/llvm/build-all/bin/opt+0xd2d28d)
#13 0x0000560eddf2234e llvm::ModuleToFunctionPassAdaptor::run(llvm::Module&, llvm::AnalysisManager<llvm::Module>&) (../../main-github/llvm/build-all/bin/opt+0x296d34e)
#14 0x0000560edc2e202d llvm::detail::PassModel<llvm::Module, llvm::ModuleToFunctionPassAdaptor, llvm::PreservedAnalyses, llvm::AnalysisManager<llvm::Module>>::run(llvm::Module&, llvm::AnalysisManager<llvm::Module>&) (../../main-github/llvm/build-all/bin/opt+0xd2d02d)
#15 0x0000560eddf1d0f4 llvm::PassManager<llvm::Module, llvm::AnalysisManager<llvm::Module>>::run(llvm::Module&, llvm::AnalysisManager<llvm::Module>&) (../../main-github/llvm/build-all/bin/opt+0x29680f4)
#16 0x0000560edbedfc13 llvm::runPassPipeline(llvm::StringRef, llvm::Module&, llvm::TargetMachine*, llvm::TargetLibraryInfoImpl*, llvm::ToolOutputFile*, llvm::ToolOutputFile*, llvm::ToolOutputFile*, llvm::StringRef, llvm::ArrayRef<llvm::PassPlugin>, llvm::opt_tool::OutputKind, llvm::opt_tool::VerifierKind, bool, bool, bool, bool, bool, bool, bool) (../../main-github/llvm/build-all/bin/opt+0x92ac13)
#17 0x0000560edbeed25c main (../../main-github/llvm/build-all/bin/opt+0x93825c)
#18 0x00007f353c6ba555 __libc_start_main (/lib64/libc.so.6+0x22555)
#19 0x0000560edbed9db0 _start (../../main-github/llvm/build-all/bin/opt+0x924db0)
Segmentation fault (core dumped)

bbi-88061.ll521 BDownload

[uabelho]

Ping @fhahn

[fhahn]

In D159202#4655350, @uabelho wrote:

Ping @fhahn

Sorry, this dropped off my radar, looking now!

[uabelho]

In D159202#4655358, @fhahn wrote:

In D159202#4655350, @uabelho wrote:

Ping @fhahn

Sorry, this dropped off my radar, looking now!

No worries. Thanks!

llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp

static bool isConstantOne(VPValue *V) {
   auto *C = dyn_cast<ConstantInt>(V->getLiveInIRValue());
   return C && C->isOne();
 }
 
 /// Returns the llvm::Instruction opcode for \p R.
 static unsigned getOpcodeForRecipe(VPRecipeBase &R) {
   if (auto *WidenR = dyn_cast<VPWidenRecipe>(&R))
     return WidenR->getUnderlyingInstr()->getOpcode();
+  if (auto *WidenC = dyn_cast<VPWidenCastRecipe>(&R))
+    return WidenC->getOpcode();
   if (auto *RepR = dyn_cast<VPReplicateRecipe>(&R))
     return RepR->getUnderlyingInstr()->getOpcode();
   if (auto *VPI = dyn_cast<VPInstruction>(&R))
     return VPI->getOpcode();
   return 0;
 }
 
+/// Return the scalar size in bits for \p VPV if possible.
+static Type *getTypeForVPValue(VPValue *VPV) {
+  // TODO: Replace with VPlan type inference once ready.
+  if (auto *VPC = dyn_cast<VPWidenCastRecipe>(VPV))
+    return VPC->getResultType();
+  auto *UV = VPV->getUnderlyingValue();
+  return UV->getType();
+}
+
 /// Try to simplify recipe \p R.
 static void simplifyRecipe(VPRecipeBase &R) {
-  unsigned Opcode = getOpcodeForRecipe(R);
-  if (Opcode == Instruction::Mul) {
+  switch (getOpcodeForRecipe(R)) {
+  case Instruction::Mul: {
     VPValue *A = R.getOperand(0);
     VPValue *B = R.getOperand(1);
     if (isConstantOne(A))
       return R.getVPSingleValue()->replaceAllUsesWith(B);
     if (isConstantOne(B))
       return R.getVPSingleValue()->replaceAllUsesWith(A);
+    break;
+  }
+  case Instruction::Trunc: {
+    VPRecipeBase *Zext = R.getOperand(0)->getDefiningRecipe();
+    if (!Zext || getOpcodeForRecipe(*Zext) != Instruction::ZExt)
+      break;
+    VPValue *A = Zext->getOperand(0);
+    VPValue *Trunc = R.getVPSingleValue();
+    if (getTypeForVPValue(Trunc) == getTypeForVPValue(A))
+      Trunc->replaceAllUsesWith(A);
+    break;
+  }
+  default:
+    break;
   }
 }
 
 /// Try to simplify the recipes in \p Plan.
 static void simplifyRecipes(VPlan &Plan) {
   ReversePostOrderTraversal<VPBlockDeepTraversalWrapper<VPBlockBase *>> RPOT(
       Plan.getEntry());
   for (VPBasicBlock *VPBB : VPBlockUtils::blocksOnly<VPBasicBlock>(RPOT)) {

llvm/test/Transforms/LoopVectorize/if-pred-stores.ll

 ; UNROLL-NEXT:    [[TMP0:%.*]] = add i64 [[INDEX]], 0
 ; UNROLL-NEXT:    [[TMP1:%.*]] = add i64 [[INDEX]], 1
 ; UNROLL-NEXT:    [[TMP2:%.*]] = getelementptr i8, ptr undef, i64 [[TMP0]]
 ; UNROLL-NEXT:    [[TMP3:%.*]] = getelementptr i8, ptr undef, i64 [[TMP1]]
 ; UNROLL-NEXT:    [[TMP4:%.*]] = load i8, ptr [[TMP2]], align 1
 ; UNROLL-NEXT:    [[TMP5:%.*]] = load i8, ptr [[TMP3]], align 1
 ; UNROLL-NEXT:    br i1 [[C:%.*]], label [[PRED_STORE_IF:%.*]], label [[PRED_STORE_CONTINUE3]]
 ; UNROLL:       pred.store.if:
-; UNROLL-NEXT:    [[TMP6:%.*]] = zext i8 [[TMP4]] to i32
-; UNROLL-NEXT:    [[TMP7:%.*]] = trunc i32 [[TMP6]] to i8
-; UNROLL-NEXT:    store i8 [[TMP7]], ptr [[TMP2]], align 1
-; UNROLL-NEXT:    [[TMP8:%.*]] = zext i8 [[TMP5]] to i32
-; UNROLL-NEXT:    [[TMP9:%.*]] = trunc i32 [[TMP8]] to i8
-; UNROLL-NEXT:    store i8 [[TMP9]], ptr [[TMP3]], align 1
+; UNROLL-NEXT:    store i8 [[TMP4]], ptr [[TMP2]], align 1
+; UNROLL-NEXT:    store i8 [[TMP5]], ptr [[TMP3]], align 1
 ; UNROLL-NEXT:    br label [[PRED_STORE_CONTINUE3]]
 ; UNROLL:       pred.store.continue3:
 ; UNROLL-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 2
-; UNROLL-NEXT:    [[TMP10:%.*]] = icmp eq i64 [[INDEX_NEXT]], 1000
-; UNROLL-NEXT:    br i1 [[TMP10]], label [[FOR_END:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]]
+; UNROLL-NEXT:    [[TMP6:%.*]] = icmp eq i64 [[INDEX_NEXT]], 1000
+; UNROLL-NEXT:    br i1 [[TMP6]], label [[FOR_END:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]]
 ; UNROLL:       for.end:
 ; UNROLL-NEXT:    ret void
 ;
 ; UNROLL-NOSIMPLIFY-LABEL: @minimal_bit_widths(
 ; UNROLL-NOSIMPLIFY-NEXT:  entry:
 ; UNROLL-NOSIMPLIFY-NEXT:    br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
 ; UNROLL-NOSIMPLIFY:       vector.ph:
 ; UNROLL-NOSIMPLIFY-NEXT:    br label [[VECTOR_BODY:%.*]]
 ; UNROLL-NOSIMPLIFY:       vector.body:
 ; UNROLL-NOSIMPLIFY-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[PRED_STORE_CONTINUE3:%.*]] ]
 ; UNROLL-NOSIMPLIFY-NEXT:    [[TMP0:%.*]] = add i64 [[INDEX]], 0
 ; UNROLL-NOSIMPLIFY-NEXT:    [[TMP1:%.*]] = add i64 [[INDEX]], 1
 ; UNROLL-NOSIMPLIFY-NEXT:    [[TMP2:%.*]] = getelementptr i8, ptr undef, i64 [[TMP0]]
 ; UNROLL-NOSIMPLIFY-NEXT:    [[TMP3:%.*]] = getelementptr i8, ptr undef, i64 [[TMP1]]
 ; UNROLL-NOSIMPLIFY-NEXT:    [[TMP4:%.*]] = load i8, ptr [[TMP2]], align 1
 ; UNROLL-NOSIMPLIFY-NEXT:    [[TMP5:%.*]] = load i8, ptr [[TMP3]], align 1
 ; UNROLL-NOSIMPLIFY-NEXT:    br i1 [[C:%.*]], label [[PRED_STORE_IF:%.*]], label [[PRED_STORE_CONTINUE:%.*]]
 ; UNROLL-NOSIMPLIFY:       pred.store.if:
-; UNROLL-NOSIMPLIFY-NEXT:    [[TMP6:%.*]] = zext i8 [[TMP4]] to i32
-; UNROLL-NOSIMPLIFY-NEXT:    [[TMP7:%.*]] = trunc i32 [[TMP6]] to i8
-; UNROLL-NOSIMPLIFY-NEXT:    store i8 [[TMP7]], ptr [[TMP2]], align 1
+; UNROLL-NOSIMPLIFY-NEXT:    store i8 [[TMP4]], ptr [[TMP2]], align 1
 ; UNROLL-NOSIMPLIFY-NEXT:    br label [[PRED_STORE_CONTINUE]]
 ; UNROLL-NOSIMPLIFY:       pred.store.continue:
 ; UNROLL-NOSIMPLIFY-NEXT:    br i1 [[C]], label [[PRED_STORE_IF2:%.*]], label [[PRED_STORE_CONTINUE3]]
 ; UNROLL-NOSIMPLIFY:       pred.store.if2:
-; UNROLL-NOSIMPLIFY-NEXT:    [[TMP8:%.*]] = zext i8 [[TMP5]] to i32
-; UNROLL-NOSIMPLIFY-NEXT:    [[TMP9:%.*]] = trunc i32 [[TMP8]] to i8
-; UNROLL-NOSIMPLIFY-NEXT:    store i8 [[TMP9]], ptr [[TMP3]], align 1
+; UNROLL-NOSIMPLIFY-NEXT:    store i8 [[TMP5]], ptr [[TMP3]], align 1
 ; UNROLL-NOSIMPLIFY-NEXT:    br label [[PRED_STORE_CONTINUE3]]
 ; UNROLL-NOSIMPLIFY:       pred.store.continue3:
 ; UNROLL-NOSIMPLIFY-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 2
-; UNROLL-NOSIMPLIFY-NEXT:    [[TMP10:%.*]] = icmp eq i64 [[INDEX_NEXT]], 1000
-; UNROLL-NOSIMPLIFY-NEXT:    br i1 [[TMP10]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP6:![0-9]+]]
+; UNROLL-NOSIMPLIFY-NEXT:    [[TMP6:%.*]] = icmp eq i64 [[INDEX_NEXT]], 1000
+; UNROLL-NOSIMPLIFY-NEXT:    br i1 [[TMP6]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP6:![0-9]+]]
 ; UNROLL-NOSIMPLIFY:       middle.block:
 ; UNROLL-NOSIMPLIFY-NEXT:    br i1 true, label [[FOR_END:%.*]], label [[SCALAR_PH]]
 ; UNROLL-NOSIMPLIFY:       scalar.ph:
 ; UNROLL-NOSIMPLIFY-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i64 [ 1000, [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ]
 ; UNROLL-NOSIMPLIFY-NEXT:    [[BC_RESUME_VAL1:%.*]] = phi i64 [ 0, [[MIDDLE_BLOCK]] ], [ 1000, [[ENTRY]] ]
 ; UNROLL-NOSIMPLIFY-NEXT:    br label [[FOR_BODY:%.*]]
 ; UNROLL-NOSIMPLIFY:       for.body:
 ; UNROLL-NOSIMPLIFY-NEXT:    [[TMP0:%.*]] = phi i64 [ [[TMP6:%.*]], [[FOR_INC:%.*]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ]

 ; VEC-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[ENTRY:%.*]] ], [ [[INDEX_NEXT:%.*]], [[PRED_STORE_CONTINUE3:%.*]] ]
 ; VEC-NEXT:    [[TMP0:%.*]] = add i64 [[INDEX]], 0
 ; VEC-NEXT:    [[TMP1:%.*]] = getelementptr i8, ptr undef, i64 [[TMP0]]
 ; VEC-NEXT:    [[TMP2:%.*]] = getelementptr i8, ptr [[TMP1]], i32 0
 ; VEC-NEXT:    [[WIDE_LOAD:%.*]] = load <2 x i8>, ptr [[TMP2]], align 1
 ; VEC-NEXT:    [[TMP3:%.*]] = extractelement <2 x i1> [[BROADCAST_SPLAT]], i32 0
 ; VEC-NEXT:    br i1 [[TMP3]], label [[PRED_STORE_IF:%.*]], label [[PRED_STORE_CONTINUE:%.*]]
 ; VEC:       pred.store.if:
-; VEC-NEXT:    [[TMP4:%.*]] = extractelement <2 x i8> [[WIDE_LOAD]], i32 0
-; VEC-NEXT:    [[TMP5:%.*]] = zext i8 [[TMP4]] to i32
-; VEC-NEXT:    [[TMP6:%.*]] = getelementptr i8, ptr undef, i64 [[TMP0]]
-; VEC-NEXT:    [[TMP7:%.*]] = trunc i32 [[TMP5]] to i8
-; VEC-NEXT:    store i8 [[TMP7]], ptr [[TMP6]], align 1
+; VEC-NEXT:    [[TMP4:%.*]] = getelementptr i8, ptr undef, i64 [[TMP0]]
+; VEC-NEXT:    [[TMP5:%.*]] = extractelement <2 x i8> [[WIDE_LOAD]], i32 0
+; VEC-NEXT:    store i8 [[TMP5]], ptr [[TMP4]], align 1
 ; VEC-NEXT:    br label [[PRED_STORE_CONTINUE]]
 ; VEC:       pred.store.continue:
-; VEC-NEXT:    [[TMP8:%.*]] = extractelement <2 x i1> [[BROADCAST_SPLAT]], i32 1
-; VEC-NEXT:    br i1 [[TMP8]], label [[PRED_STORE_IF2:%.*]], label [[PRED_STORE_CONTINUE3]]
+; VEC-NEXT:    [[TMP6:%.*]] = extractelement <2 x i1> [[BROADCAST_SPLAT]], i32 1
+; VEC-NEXT:    br i1 [[TMP6]], label [[PRED_STORE_IF2:%.*]], label [[PRED_STORE_CONTINUE3]]
 ; VEC:       pred.store.if2:
-; VEC-NEXT:    [[TMP9:%.*]] = add i64 [[INDEX]], 1
-; VEC-NEXT:    [[TMP10:%.*]] = extractelement <2 x i8> [[WIDE_LOAD]], i32 1
-; VEC-NEXT:    [[TMP11:%.*]] = zext i8 [[TMP10]] to i32
-; VEC-NEXT:    [[TMP12:%.*]] = getelementptr i8, ptr undef, i64 [[TMP9]]
-; VEC-NEXT:    [[TMP13:%.*]] = trunc i32 [[TMP11]] to i8
-; VEC-NEXT:    store i8 [[TMP13]], ptr [[TMP12]], align 1
+; VEC-NEXT:    [[TMP7:%.*]] = add i64 [[INDEX]], 1
+; VEC-NEXT:    [[TMP8:%.*]] = getelementptr i8, ptr undef, i64 [[TMP7]]
+; VEC-NEXT:    [[TMP9:%.*]] = extractelement <2 x i8> [[WIDE_LOAD]], i32 1
+; VEC-NEXT:    store i8 [[TMP9]], ptr [[TMP8]], align 1
 ; VEC-NEXT:    br label [[PRED_STORE_CONTINUE3]]
 ; VEC:       pred.store.continue3:
 ; VEC-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 2
-; VEC-NEXT:    [[TMP14:%.*]] = icmp eq i64 [[INDEX_NEXT]], 1000
-; VEC-NEXT:    br i1 [[TMP14]], label [[FOR_END:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]]
+; VEC-NEXT:    [[TMP10:%.*]] = icmp eq i64 [[INDEX_NEXT]], 1000
+; VEC-NEXT:    br i1 [[TMP10]], label [[FOR_END:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]]
 ; VEC:       for.end:
 ; VEC-NEXT:    ret void
 ;
 entry:
   br label %for.body
 
 for.body:
   %tmp0 = phi i64 [ %tmp6, %for.inc ], [ 0, %entry ]
   %tmp1 = phi i64 [ %tmp7, %for.inc ], [ 1000, %entry ]
   %tmp2 = getelementptr i8, ptr undef, i64 %tmp0
   %tmp3 = load i8, ptr %tmp2, align 1
   br i1 %c, label %if.then, label %for.inc
 
 if.then:
   %tmp4 = zext i8 %tmp3 to i32
   %tmp5 = trunc i32 %tmp4 to i8
   store i8 %tmp5, ptr %tmp2, align 1
   br label %for.inc
 
 for.inc:
   %tmp6 = add nuw nsw i64 %tmp0, 1
   %tmp7 = add i64 %tmp1, -1
   %tmp8 = icmp eq i64 %tmp7, 0
   br i1 %tmp8, label %for.end, label %for.body

 ; UNROLL-NOSIMPLIFY-NEXT:    [[TMP2:%.*]] = getelementptr i8, ptr [[PTR:%.*]], i64 [[TMP0]]
 ; UNROLL-NOSIMPLIFY-NEXT:    [[TMP3:%.*]] = getelementptr i8, ptr [[PTR]], i64 [[TMP1]]
 ; UNROLL-NOSIMPLIFY-NEXT:    [[TMP4:%.*]] = load i8, ptr [[TMP2]], align 1
 ; UNROLL-NOSIMPLIFY-NEXT:    [[TMP5:%.*]] = load i8, ptr [[TMP3]], align 1
 ; UNROLL-NOSIMPLIFY-NEXT:    store i8 0, ptr [[TMP2]], align 1
 ; UNROLL-NOSIMPLIFY-NEXT:    store i8 0, ptr [[TMP3]], align 1
 ; UNROLL-NOSIMPLIFY-NEXT:    br i1 [[C:%.*]], label [[PRED_STORE_IF:%.*]], label [[PRED_STORE_CONTINUE:%.*]]
 ; UNROLL-NOSIMPLIFY:       pred.store.if:
-; UNROLL-NOSIMPLIFY-NEXT:    [[TMP6:%.*]] = zext i8 [[TMP4]] to i32
-; UNROLL-NOSIMPLIFY-NEXT:    [[TMP7:%.*]] = trunc i32 [[TMP6]] to i8
-; UNROLL-NOSIMPLIFY-NEXT:    store i8 [[TMP7]], ptr [[TMP2]], align 1
+; UNROLL-NOSIMPLIFY-NEXT:    store i8 [[TMP4]], ptr [[TMP2]], align 1
 ; UNROLL-NOSIMPLIFY-NEXT:    br label [[PRED_STORE_CONTINUE]]
 ; UNROLL-NOSIMPLIFY:       pred.store.continue:
 ; UNROLL-NOSIMPLIFY-NEXT:    br i1 [[C]], label [[PRED_STORE_IF2:%.*]], label [[PRED_STORE_CONTINUE3]]
 ; UNROLL-NOSIMPLIFY:       pred.store.if2:
-; UNROLL-NOSIMPLIFY-NEXT:    [[TMP8:%.*]] = zext i8 [[TMP5]] to i32
-; UNROLL-NOSIMPLIFY-NEXT:    [[TMP9:%.*]] = trunc i32 [[TMP8]] to i8
-; UNROLL-NOSIMPLIFY-NEXT:    store i8 [[TMP9]], ptr [[TMP3]], align 1
+; UNROLL-NOSIMPLIFY-NEXT:    store i8 [[TMP5]], ptr [[TMP3]], align 1
 ; UNROLL-NOSIMPLIFY-NEXT:    br label [[PRED_STORE_CONTINUE3]]
 ; UNROLL-NOSIMPLIFY:       pred.store.continue3:
 ; UNROLL-NOSIMPLIFY-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 2
-; UNROLL-NOSIMPLIFY-NEXT:    [[TMP10:%.*]] = icmp eq i64 [[INDEX_NEXT]], 0
-; UNROLL-NOSIMPLIFY-NEXT:    br i1 [[TMP10]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP8:![0-9]+]]
+; UNROLL-NOSIMPLIFY-NEXT:    [[TMP6:%.*]] = icmp eq i64 [[INDEX_NEXT]], 0
+; UNROLL-NOSIMPLIFY-NEXT:    br i1 [[TMP6]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP8:![0-9]+]]
 ; UNROLL-NOSIMPLIFY:       middle.block:
 ; UNROLL-NOSIMPLIFY-NEXT:    br i1 true, label [[FOR_END:%.*]], label [[SCALAR_PH]]
 ; UNROLL-NOSIMPLIFY:       scalar.ph:
 ; UNROLL-NOSIMPLIFY-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i64 [ 0, [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ]
 ; UNROLL-NOSIMPLIFY-NEXT:    [[BC_RESUME_VAL1:%.*]] = phi i64 [ 0, [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY]] ]
 ; UNROLL-NOSIMPLIFY-NEXT:    br label [[FOR_BODY:%.*]]
 ; UNROLL-NOSIMPLIFY:       for.body:
 ; UNROLL-NOSIMPLIFY-NEXT:    [[TMP0:%.*]] = phi i64 [ [[TMP6:%.*]], [[FOR_INC:%.*]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ]

llvm/test/Transforms/LoopVectorize/vplan-sink-scalars-and-merge-vf1.ll

 ; CHECK-NEXT: Successor(s): pred.store
 
 ; CHECK:      <xVFxUF> pred.store: {
 ; CHECK-NEXT:   pred.store.entry:
 ; CHECK-NEXT:     BRANCH-ON-MASK ir<%c>
 ; CHECK-NEXT:   Successor(s): pred.store.if, pred.store.continue
 
 ; CHECK:      pred.store.if:
-; CHECK-NEXT:   CLONE ir<%tmp4> = zext ir<%tmp3>
-; CHECK-NEXT:   CLONE ir<%tmp5> = trunc ir<%tmp4>
-; CHECK-NEXT:   CLONE store ir<%tmp5>, ir<%tmp2>
+; CHECK-NEXT:   CLONE store ir<%tmp3>, ir<%tmp2>
 ; CHECK-NEXT:   Successor(s): pred.store.continue
 
 ; CHECK:      pred.store.continue:
 ; CHECK-NEXT:   No successors
 ; CHECK-NEXT: }
 
 ; CHECK:      if.then.0:
 ; CHECK-NEXT:  EMIT vp<[[CAN_IV_NEXT:%.+]]> = VF * UF + nuw vp<[[CAN_IV]]>