Broaden the definition of a "widenable branch"

[reames]

Summary

As a reminder, a "widenable branch" is the pattern "br i1 (and i1 X, WC()), label %taken, label %untaken" where "WC" is the widenable condition intrinsics. The semantics of such a branch (derived from the semantics of WC) is that a new condition can be added into the condition arbitrarily without violating legality.

Broaden the definition in two ways:

1. Allow swapped operands to the br (and X, WC()) form
2. Allow widenable branch w/trivial condition (i.e. true) which takes form of br i1 WC()

The former is just general robustness. The later is specifically important as partial unswitching of a widenable range check produces exactly this form above the loop.

[ebrevnov]

llvm/lib/Analysis/GuardUtils.cpp

47–59

Looks not finished...

llvm/lib/Transforms/Utils/GuardUtils.cpp

89–107

Would it be better to use parseWidenableBranch instead of repeating its logic here.

114–128

Same as above.

[reames]

llvm/lib/Analysis/GuardUtils.cpp

47–59

Why? It's the trivial case of a widenable branch?

I can simply remove the comment if it's unhelpful?

llvm/lib/Transforms/Utils/GuardUtils.cpp

89–107

Can't. I wish we could and tried to find a way, but we need to modify the existing instructions to preserve the "one use" requirement on the widenable condition to be able to determine it's a trivially widenable branch.

[apilipenko]

llvm/lib/Transforms/Utils/GuardUtils.cpp

89–107

Yeah, a naive approach here would be to replace wc with (newc && wc). It handles both trivial and non-trivial widenable branches. The only problem is it makes a pattern which is not recognized by parseWidenableBranch.

br (c && wc), ... - widenable branch
=>
br (c && (newc && wc)), ... - not (obviously) widenable anymore

Can you add a comment that we do it this way because we need to preserve the structure (c && wc) which is recognized by parseWidenableBranch?

[reames]

llvm/lib/Transforms/Utils/GuardUtils.cpp

89–107

Just to close loop, Artur and I latter released we could rephrase parseWidenableBranch in terms of Uses and simplify a lot of this code. I landed the rewrite as rG8293f7434577

[ebrevnov]

llvm/lib/Transforms/Utils/GuardUtils.cpp

89–107

Cool. That's what I assumed by asking to reuse parseWidenableBranch :-)

llvm/lib/Analysis/GuardUtils.cpp

 //===-- GuardUtils.cpp - Utils for work with guards -------------*- C++ -*-===//
 //
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
 // See https://llvm.org/LICENSE.txt for license information.
 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 //
 //===----------------------------------------------------------------------===//
 // Utils that are used to perform analyzes related to guards and their
 // conditions.
 //===----------------------------------------------------------------------===//
 
 #include "llvm/Analysis/GuardUtils.h"
 #include "llvm/IR/PatternMatch.h"
 
 using namespace llvm;
+using namespace llvm::PatternMatch;
 
 bool llvm::isGuard(const User *U) {
-  using namespace llvm::PatternMatch;
   return match(U, m_Intrinsic<Intrinsic::experimental_guard>());
 }
 
 bool llvm::isWidenableBranch(const User *U) {
   Value *Condition, *WidenableCondition;
   BasicBlock *GuardedBB, *DeoptBB;
   return parseWidenableBranch(U, Condition, WidenableCondition, GuardedBB,
                               DeoptBB);
 }
 
 bool llvm::isGuardAsWidenableBranch(const User *U) {
   Value *Condition, *WidenableCondition;
   BasicBlock *GuardedBB, *DeoptBB;
   if (!parseWidenableBranch(U, Condition, WidenableCondition, GuardedBB,
                             DeoptBB))
     return false;
-  using namespace llvm::PatternMatch;
   for (auto &Insn : *DeoptBB) {
     if (match(&Insn, m_Intrinsic<Intrinsic::experimental_deoptimize>()))
       return true;
     if (Insn.mayHaveSideEffects())
       return false;
   }
   return false;
 }
 
 bool llvm::parseWidenableBranch(const User *U, Value *&Condition,
                                 Value *&WidenableCondition,
                                 BasicBlock *&IfTrueBB, BasicBlock *&IfFalseBB) {
-  using namespace llvm::PatternMatch;
+  if (match(U, m_Br(m_Intrinsic<Intrinsic::experimental_widenable_condition>(),
+                    IfTrueBB, IfFalseBB)) &&
+      cast<BranchInst>(U)->getCondition()->hasOneUse()) {
+    WidenableCondition = cast<BranchInst>(U)->getCondition();
+    Condition = ConstantInt::getTrue(IfTrueBB->getContext());
+    return true;
+  }
+
+  // Check for two cases:
+  // 1) br (i1 (and A, WC())), label %IfTrue, label %IfFalse
+  // 2) br (i1 (and WC(), B)), label %IfTrue, label %IfFalse
+  // We do not check for more generalized and trees as we should canonicalize
+  // to the form above in instcombine. (TODO)
   if (!match(U, m_Br(m_And(m_Value(Condition), m_Value(WidenableCondition)),
                      IfTrueBB, IfFalseBB)))
     return false;
+  if (!match(WidenableCondition,
+             m_Intrinsic<Intrinsic::experimental_widenable_condition>())) {
+    if (!match(Condition,
+               m_Intrinsic<Intrinsic::experimental_widenable_condition>()))
+      return false;
+    std::swap(Condition, WidenableCondition);
+  }
+    
   // For the branch to be (easily) widenable, it must not correlate with other
   // branches.  Thus, the widenable condition must have a single use.
-  if (!WidenableCondition->hasOneUse() ||
-      !cast<BranchInst>(U)->getCondition()->hasOneUse())
-    return false;
-  // TODO: At the moment, we only recognize the branch if the WC call in this
-  // specific position.  We should generalize!
-  return match(WidenableCondition,
-               m_Intrinsic<Intrinsic::experimental_widenable_condition>());
+  return (WidenableCondition->hasOneUse() &&
+          cast<BranchInst>(U)->getCondition()->hasOneUse());
 }

llvm/lib/Transforms/Scalar/GuardWidening.cpp

 
 // Get the condition of \p I. It can either be a guard or a conditional branch.
 static Value *getCondition(Instruction *I) {
   if (IntrinsicInst *GI = dyn_cast<IntrinsicInst>(I)) {
     assert(GI->getIntrinsicID() == Intrinsic::experimental_guard &&
            "Bad guard intrinsic?");
     return GI->getArgOperand(0);
   }
-  if (isGuardAsWidenableBranch(I)) {
-    auto *Cond = cast<BranchInst>(I)->getCondition();
-    return cast<BinaryOperator>(Cond)->getOperand(0);
-  }
+  Value *Cond, *WC;
+  BasicBlock *IfTrueBB, *IfFalseBB;
+  if (parseWidenableBranch(I, Cond, WC, IfTrueBB, IfFalseBB))
+    return Cond;
+
   return cast<BranchInst>(I)->getCondition();
 }
 
 // Set the condition for \p I to \p NewCond. \p I can either be a guard or a
-// conditional branch.
+// conditional branch.  
 static void setCondition(Instruction *I, Value *NewCond) {
   if (IntrinsicInst *GI = dyn_cast<IntrinsicInst>(I)) {
     assert(GI->getIntrinsicID() == Intrinsic::experimental_guard &&
            "Bad guard intrinsic?");
     GI->setArgOperand(0, NewCond);
     return;
   }
   cast<BranchInst>(I)->setCondition(NewCond);

llvm/lib/Transforms/Utils/GuardUtils.cpp

 //===----------------------------------------------------------------------===//
 
 #include "llvm/Transforms/Utils/GuardUtils.h"
 #include "llvm/Analysis/GuardUtils.h"
 #include "llvm/IR/Function.h"
 #include "llvm/IR/IRBuilder.h"
 #include "llvm/IR/Instructions.h"
 #include "llvm/IR/MDBuilder.h"
+#include "llvm/IR/PatternMatch.h"
 #include "llvm/Support/CommandLine.h"
 #include "llvm/Transforms/Utils/BasicBlockUtils.h"
 
 using namespace llvm;
+using namespace llvm::PatternMatch;
 
 static cl::opt<uint32_t> PredicatePassBranchWeight(
     "guards-predicate-pass-branch-weight", cl::Hidden, cl::init(1 << 20),
     cl::desc("The probability of a guard failing is assumed to be the "
              "reciprocal of this value (default = 1 << 20)"));
 
 void llvm::makeGuardControlFlowExplicit(Function *DeoptIntrinsic,
                                         CallInst *Guard, bool UseWC) {
if (UseWC) {
     assert(isWidenableBranch(CheckBI) && "sanity check");
   }
 }
 
 
 void llvm::widenWidenableBranch(BranchInst *WidenableBR, Value *NewCond) {
   assert(isWidenableBranch(WidenableBR) && "precondition");
 
+  // The tempting trivially option is to produce something like this:
+  // br (and oldcond, newcond) where oldcond is assumed to contain a widenable
+  // condition, but that doesn't match the pattern parseWidenableBranch expects
+  // so we have to be more sophisticated.
+
+  if (match(WidenableBR->getCondition(),
+            m_Intrinsic<Intrinsic::experimental_widenable_condition>())) {
+    IRBuilder<> B(WidenableBR);
+    WidenableBR->setCondition(B.CreateAnd(NewCond,
+                                          WidenableBR->getCondition()));
+  } else {
     Instruction *WCAnd = cast<Instruction>(WidenableBR->getCondition());
     // Condition is only guaranteed to dominate branch
     WCAnd->moveBefore(WidenableBR);
-  Value *OldCond = WCAnd->getOperand(0);
     IRBuilder<> B(WCAnd);
-  WCAnd->setOperand(0, B.CreateAnd(NewCond, OldCond));
-
+    const bool Op0IsWC =
+      match(WCAnd->getOperand(0),
+            m_Intrinsic<Intrinsic::experimental_widenable_condition>());
+    const unsigned CondOpIdx = Op0IsWC ? 1 : 0;
+    Value *OldCond = WCAnd->getOperand(CondOpIdx);
+    NewCond = B.CreateAnd(NewCond, OldCond);
+    WCAnd->setOperand(CondOpIdx, NewCond);
+  }
   assert(isWidenableBranch(WidenableBR) && "preserve widenabiliy");
 }
 
 void llvm::setWidenableBranchCond(BranchInst *WidenableBR, Value *NewCond) {
   assert(isWidenableBranch(WidenableBR) && "precondition");
 
+  if (match(WidenableBR->getCondition(),
+            m_Intrinsic<Intrinsic::experimental_widenable_condition>())) {
+    IRBuilder<> B(WidenableBR);
+    WidenableBR->setCondition(B.CreateAnd(NewCond,
+                                          WidenableBR->getCondition()));
+  } else {
     Instruction *WCAnd = cast<Instruction>(WidenableBR->getCondition());
     // Condition is only guaranteed to dominate branch
     WCAnd->moveBefore(WidenableBR);
-  WCAnd->setOperand(0, NewCond);
-
+    const bool Op0IsWC =
+      match(WCAnd->getOperand(0),
+            m_Intrinsic<Intrinsic::experimental_widenable_condition>());
+    const unsigned CondOpIdx = Op0IsWC ? 1 : 0;
+    WCAnd->setOperand(CondOpIdx, NewCond);
+  }
   assert(isWidenableBranch(WidenableBR) && "preserve widenabiliy");
 }

llvm/test/Transforms/GuardWidening/basic_widenable_condition_guards.ll

 deopt2:                                           ; preds = %guarded
   call void (...) @llvm.experimental.deoptimize.isVoid() [ "deopt"() ]
   ret void
 
 guarded1:                                         ; preds = %guarded
   ret void
 }
 
+
+define void @swapped_wb(i1 %cond_0, i1 %cond_1) {
+; CHECK-LABEL: @swapped_wb(
+; CHECK-NEXT:  entry:
+; CHECK-NEXT:    [[WIDENABLE_COND:%.*]] = call i1 @llvm.experimental.widenable.condition()
+; CHECK-NEXT:    [[WIDE_CHK:%.*]] = and i1 [[COND_0:%.*]], [[COND_1:%.*]]
+; CHECK-NEXT:    [[EXIPLICIT_GUARD_COND:%.*]] = and i1 [[WIDENABLE_COND]], [[WIDE_CHK]]
+; CHECK-NEXT:    br i1 [[EXIPLICIT_GUARD_COND]], label [[GUARDED:%.*]], label [[DEOPT:%.*]], !prof !0
+; CHECK:       deopt:
+; CHECK-NEXT:    call void (...) @llvm.experimental.deoptimize.isVoid() [ "deopt"() ]
+; CHECK-NEXT:    ret void
+; CHECK:       guarded:
+; CHECK-NEXT:    [[WIDENABLE_COND3:%.*]] = call i1 @llvm.experimental.widenable.condition()
+; CHECK-NEXT:    [[EXIPLICIT_GUARD_COND4:%.*]] = and i1 [[COND_1]], [[WIDENABLE_COND3]]
+; CHECK-NEXT:    br i1 true, label [[GUARDED1:%.*]], label [[DEOPT2:%.*]], !prof !0
+; CHECK:       deopt2:
+; CHECK-NEXT:    call void (...) @llvm.experimental.deoptimize.isVoid() [ "deopt"() ]
+; CHECK-NEXT:    ret void
+; CHECK:       guarded1:
+; CHECK-NEXT:    ret void
+;
+entry:
+  %widenable_cond = call i1 @llvm.experimental.widenable.condition()
+  %exiplicit_guard_cond = and i1 %widenable_cond, %cond_0
+  br i1 %exiplicit_guard_cond, label %guarded, label %deopt, !prof !0
+
+deopt:                                            ; preds = %entry
+  call void (...) @llvm.experimental.deoptimize.isVoid() [ "deopt"() ]
+  ret void
+
+guarded:                                          ; preds = %entry
+  %widenable_cond3 = call i1 @llvm.experimental.widenable.condition()
+  %exiplicit_guard_cond4 = and i1 %cond_1, %widenable_cond3
+  br i1 %exiplicit_guard_cond4, label %guarded1, label %deopt2, !prof !0
+
+deopt2:                                           ; preds = %guarded
+  call void (...) @llvm.experimental.deoptimize.isVoid() [ "deopt"() ]
+  ret void
+
+guarded1:                                         ; preds = %guarded
+  ret void
+}
+
+define void @trivial_wb(i1 %cond_0) {
+; CHECK-LABEL: @trivial_wb(
+; CHECK-NEXT:  entry:
+; CHECK-NEXT:    [[WIDENABLE_COND:%.*]] = call i1 @llvm.experimental.widenable.condition()
+; CHECK-NEXT:    [[WIDE_CHK:%.*]] = and i1 true, [[COND_0:%.*]]
+; CHECK-NEXT:    [[TMP0:%.*]] = and i1 [[WIDE_CHK]], [[WIDENABLE_COND]]
+; CHECK-NEXT:    br i1 [[TMP0]], label [[GUARDED:%.*]], label [[DEOPT:%.*]], !prof !0
+; CHECK:       deopt:
+; CHECK-NEXT:    call void (...) @llvm.experimental.deoptimize.isVoid() [ "deopt"() ]
+; CHECK-NEXT:    ret void
+; CHECK:       guarded:
+; CHECK-NEXT:    [[WIDENABLE_COND3:%.*]] = call i1 @llvm.experimental.widenable.condition()
+; CHECK-NEXT:    [[EXIPLICIT_GUARD_COND4:%.*]] = and i1 [[COND_0]], [[WIDENABLE_COND3]]
+; CHECK-NEXT:    br i1 true, label [[GUARDED1:%.*]], label [[DEOPT2:%.*]], !prof !0
+; CHECK:       deopt2:
+; CHECK-NEXT:    call void (...) @llvm.experimental.deoptimize.isVoid() [ "deopt"() ]
+; CHECK-NEXT:    ret void
+; CHECK:       guarded1:
+; CHECK-NEXT:    ret void
+;
+entry:
+  %widenable_cond = call i1 @llvm.experimental.widenable.condition()
+  br i1 %widenable_cond, label %guarded, label %deopt, !prof !0
+
+deopt:                                            ; preds = %entry
+  call void (...) @llvm.experimental.deoptimize.isVoid() [ "deopt"() ]
+  ret void
+
+guarded:                                          ; preds = %entry
+  %widenable_cond3 = call i1 @llvm.experimental.widenable.condition()
+  %exiplicit_guard_cond4 = and i1 %cond_0, %widenable_cond3
+  br i1 %exiplicit_guard_cond4, label %guarded1, label %deopt2, !prof !0
+
+deopt2:                                           ; preds = %guarded
+  call void (...) @llvm.experimental.deoptimize.isVoid() [ "deopt"() ]
+  ret void
+
+guarded1:                                         ; preds = %guarded
+  ret void
+}
+
+
 declare void @llvm.experimental.deoptimize.isVoid(...)
 
 ; Function Attrs: inaccessiblememonly nounwind
 declare i1 @llvm.experimental.widenable.condition() #0
 
 attributes #0 = { inaccessiblememonly nounwind }
 
 !0 = !{!"branch_weights", i32 1048576, i32 1}

llvm/test/Transforms/LoopPredication/basic_widenable_branch_guards.ll

guarded:                                          ; preds = %loop
   %i.next = add nuw i32 %i, 1
   %continue = icmp ult i32 %i.next, %n
   br i1 %continue, label %loop, label %exit, !prof !2
 
 exit:                                             ; preds = %guarded, %entry
   ret i32 0
 }
 
+define i32 @swapped_wb(i32* %array, i32 %length, i32 %n) {
+; CHECK-LABEL: @swapped_wb(
+; CHECK-NEXT:  entry:
+; CHECK-NEXT:    [[TMP5:%.*]] = icmp eq i32 [[N:%.*]], 0
+; CHECK-NEXT:    br i1 [[TMP5]], label [[EXIT:%.*]], label [[LOOP_PREHEADER:%.*]]
+; CHECK:       loop.preheader:
+; CHECK-NEXT:    [[WIDENABLE_COND:%.*]] = call i1 @llvm.experimental.widenable.condition()
+; CHECK-NEXT:    [[TMP0:%.*]] = icmp ule i32 [[N]], [[LENGTH:%.*]]
+; CHECK-NEXT:    [[TMP1:%.*]] = icmp ult i32 0, [[LENGTH]]
+; CHECK-NEXT:    [[TMP2:%.*]] = and i1 [[TMP1]], [[TMP0]]
+; CHECK-NEXT:    [[TMP3:%.*]] = and i1 [[TMP2]], [[WIDENABLE_COND]]
+; CHECK-NEXT:    br label [[LOOP:%.*]]
+; CHECK:       loop:
+; CHECK-NEXT:    [[LOOP_ACC:%.*]] = phi i32 [ [[LOOP_ACC_NEXT:%.*]], [[GUARDED:%.*]] ], [ 0, [[LOOP_PREHEADER]] ]
+; CHECK-NEXT:    [[I:%.*]] = phi i32 [ [[I_NEXT:%.*]], [[GUARDED]] ], [ 0, [[LOOP_PREHEADER]] ]
+; CHECK-NEXT:    br i1 [[TMP3]], label [[GUARDED]], label [[DEOPT:%.*]], !prof !0
+; CHECK:       deopt:
+; CHECK-NEXT:    [[DEOPTCALL:%.*]] = call i32 (...) @llvm.experimental.deoptimize.i32(i32 9) [ "deopt"() ]
+; CHECK-NEXT:    ret i32 [[DEOPTCALL]]
+; CHECK:       guarded:
+; CHECK-NEXT:    [[I_I64:%.*]] = zext i32 [[I]] to i64
+; CHECK-NEXT:    [[ARRAY_I_PTR:%.*]] = getelementptr inbounds i32, i32* [[ARRAY:%.*]], i64 [[I_I64]]
+; CHECK-NEXT:    [[ARRAY_I:%.*]] = load i32, i32* [[ARRAY_I_PTR]], align 4
+; CHECK-NEXT:    [[LOOP_ACC_NEXT]] = add i32 [[LOOP_ACC]], [[ARRAY_I]]
+; CHECK-NEXT:    [[I_NEXT]] = add nuw i32 [[I]], 1
+; CHECK-NEXT:    [[CONTINUE:%.*]] = icmp ult i32 [[I_NEXT]], [[N]]
+; CHECK-NEXT:    br i1 [[CONTINUE]], label [[LOOP]], label [[EXIT_LOOPEXIT:%.*]], !prof !1
+; CHECK:       exit.loopexit:
+; CHECK-NEXT:    [[LOOP_ACC_NEXT_LCSSA:%.*]] = phi i32 [ [[LOOP_ACC_NEXT]], [[GUARDED]] ]
+; CHECK-NEXT:    br label [[EXIT]]
+; CHECK:       exit:
+; CHECK-NEXT:    [[RESULT:%.*]] = phi i32 [ 0, [[ENTRY:%.*]] ], [ [[LOOP_ACC_NEXT_LCSSA]], [[EXIT_LOOPEXIT]] ]
+; CHECK-NEXT:    ret i32 [[RESULT]]
+;
+entry:
+  %tmp5 = icmp eq i32 %n, 0
+  br i1 %tmp5, label %exit, label %loop.preheader
+
+loop.preheader:                                   ; preds = %entry
+  %widenable_cond = call i1 @llvm.experimental.widenable.condition()
+  br label %loop
+
+loop:                                             ; preds = %guarded, %loop.preheader
+  %loop.acc = phi i32 [ %loop.acc.next, %guarded ], [ 0, %loop.preheader ]
+  %i = phi i32 [ %i.next, %guarded ], [ 0, %loop.preheader ]
+  %within.bounds = icmp ult i32 %i, %length
+  %exiplicit_guard_cond = and i1 %widenable_cond, %within.bounds
+  br i1 %exiplicit_guard_cond, label %guarded, label %deopt, !prof !0
+
+deopt:                                            ; preds = %loop
+  %deoptcall = call i32 (...) @llvm.experimental.deoptimize.i32(i32 9) [ "deopt"() ]
+  ret i32 %deoptcall
+
+guarded:                                          ; preds = %loop
+  %i.i64 = zext i32 %i to i64
+  %array.i.ptr = getelementptr inbounds i32, i32* %array, i64 %i.i64
+  %array.i = load i32, i32* %array.i.ptr, align 4
+  %loop.acc.next = add i32 %loop.acc, %array.i
+  %i.next = add nuw i32 %i, 1
+  %continue = icmp ult i32 %i.next, %n
+  br i1 %continue, label %loop, label %exit, !prof !2
+
+exit:                                             ; preds = %guarded, %entry
+  %result = phi i32 [ 0, %entry ], [ %loop.acc.next, %guarded ]
+  ret i32 %result
+}
 
 declare i32 @llvm.experimental.deoptimize.i32(...)
 
 ; Function Attrs: inaccessiblememonly nounwind
 declare i1 @llvm.experimental.widenable.condition() #0
 
 attributes #0 = { inaccessiblememonly nounwind }
 
 !0 = !{!"branch_weights", i32 1048576, i32 1}
 !1 = !{i32 1, i32 -2147483648}
 !2 = !{!"branch_weights", i32 1024, i32 1}

llvm/test/Transforms/LoopPredication/predicate-exits.ll

guarded:
   %continue = icmp ult i32 %i.next, %n
   br i1 %continue, label %loop, label %exit
 
 exit:
   %result = phi i32 [ %loop.acc.next, %guarded ]
   ret i32 %result
 }
 
+define i32 @swapped_wb(i32* %array, i32 %length, i32 %n, i1 %cond_0) {
+; CHECK-LABEL: @swapped_wb(
+; CHECK-NEXT:  entry:
+; CHECK-NEXT:    [[WIDENABLE_COND:%.*]] = call i1 @llvm.experimental.widenable.condition()
+; CHECK-NEXT:    [[TMP0:%.*]] = icmp ugt i32 [[N:%.*]], 1
+; CHECK-NEXT:    [[UMAX:%.*]] = select i1 [[TMP0]], i32 [[N]], i32 1
+; CHECK-NEXT:    [[TMP1:%.*]] = add i32 [[UMAX]], -1
+; CHECK-NEXT:    [[TMP2:%.*]] = icmp ult i32 [[LENGTH:%.*]], [[TMP1]]
+; CHECK-NEXT:    [[UMIN:%.*]] = select i1 [[TMP2]], i32 [[LENGTH]], i32 [[TMP1]]
+; CHECK-NEXT:    [[TMP3:%.*]] = icmp ugt i32 [[LENGTH]], [[UMIN]]
+; CHECK-NEXT:    [[TMP4:%.*]] = freeze i1 [[TMP3]]
+; CHECK-NEXT:    [[TMP5:%.*]] = and i1 [[TMP4]], [[COND_0:%.*]]
+; CHECK-NEXT:    [[EXIPLICIT_GUARD_COND:%.*]] = and i1 [[WIDENABLE_COND]], [[TMP5]]
+; CHECK-NEXT:    br i1 [[EXIPLICIT_GUARD_COND]], label [[LOOP_PREHEADER:%.*]], label [[DEOPT:%.*]], !prof !0
+; CHECK:       deopt:
+; CHECK-NEXT:    [[DEOPTRET:%.*]] = call i32 (...) @llvm.experimental.deoptimize.i32() [ "deopt"() ]
+; CHECK-NEXT:    ret i32 [[DEOPTRET]]
+; CHECK:       loop.preheader:
+; CHECK-NEXT:    br label [[LOOP:%.*]]
+; CHECK:       loop:
+; CHECK-NEXT:    [[LOOP_ACC:%.*]] = phi i32 [ [[LOOP_ACC_NEXT:%.*]], [[GUARDED:%.*]] ], [ 0, [[LOOP_PREHEADER]] ]
+; CHECK-NEXT:    [[I:%.*]] = phi i32 [ [[I_NEXT:%.*]], [[GUARDED]] ], [ 0, [[LOOP_PREHEADER]] ]
+; CHECK-NEXT:    call void @unknown()
+; CHECK-NEXT:    [[WITHIN_BOUNDS:%.*]] = icmp ult i32 [[I]], [[LENGTH]]
+; CHECK-NEXT:    br i1 true, label [[GUARDED]], label [[DEOPT2:%.*]], !prof !0
+; CHECK:       deopt2:
+; CHECK-NEXT:    call void @unknown()
+; CHECK-NEXT:    [[DEOPTRET2:%.*]] = call i32 (...) @llvm.experimental.deoptimize.i32() [ "deopt"() ]
+; CHECK-NEXT:    ret i32 [[DEOPTRET2]]
+; CHECK:       guarded:
+; CHECK-NEXT:    [[I_I64:%.*]] = zext i32 [[I]] to i64
+; CHECK-NEXT:    [[ARRAY_I_PTR:%.*]] = getelementptr inbounds i32, i32* [[ARRAY:%.*]], i64 [[I_I64]]
+; CHECK-NEXT:    [[ARRAY_I:%.*]] = load i32, i32* [[ARRAY_I_PTR]], align 4
+; CHECK-NEXT:    store i32 0, i32* [[ARRAY_I_PTR]]
+; CHECK-NEXT:    [[LOOP_ACC_NEXT]] = add i32 [[LOOP_ACC]], [[ARRAY_I]]
+; CHECK-NEXT:    [[I_NEXT]] = add nuw i32 [[I]], 1
+; CHECK-NEXT:    [[CONTINUE:%.*]] = icmp ult i32 [[I_NEXT]], [[N]]
+; CHECK-NEXT:    br i1 [[CONTINUE]], label [[LOOP]], label [[EXIT:%.*]]
+; CHECK:       exit:
+; CHECK-NEXT:    [[RESULT:%.*]] = phi i32 [ [[LOOP_ACC_NEXT]], [[GUARDED]] ]
+; CHECK-NEXT:    ret i32 [[RESULT]]
+;
+entry:
+  %widenable_cond = call i1 @llvm.experimental.widenable.condition()
+  %exiplicit_guard_cond = and i1 %widenable_cond, %cond_0
+  br i1 %exiplicit_guard_cond, label %loop.preheader, label %deopt, !prof !0
+
+deopt:
+  %deoptret = call i32 (...) @llvm.experimental.deoptimize.i32() [ "deopt"() ]
+  ret i32 %deoptret
+
+loop.preheader:
+  br label %loop
+
+loop:
+  %loop.acc = phi i32 [ %loop.acc.next, %guarded ], [ 0, %loop.preheader ]
+  %i = phi i32 [ %i.next, %guarded ], [ 0, %loop.preheader ]
+  call void @unknown()
+  %within.bounds = icmp ult i32 %i, %length
+  br i1 %within.bounds, label %guarded, label %deopt2, !prof !0
+
+deopt2:
+  call void @unknown()
+  %deoptret2 = call i32 (...) @llvm.experimental.deoptimize.i32() [ "deopt"() ]
+  ret i32 %deoptret2
+
+guarded:
+  %i.i64 = zext i32 %i to i64
+  %array.i.ptr = getelementptr inbounds i32, i32* %array, i64 %i.i64
+  %array.i = load i32, i32* %array.i.ptr, align 4
+  store i32 0, i32* %array.i.ptr
+  %loop.acc.next = add i32 %loop.acc, %array.i
+  %i.next = add nuw i32 %i, 1
+  %continue = icmp ult i32 %i.next, %n
+  br i1 %continue, label %loop, label %exit
+
+exit:
+  %result = phi i32 [ %loop.acc.next, %guarded ]
+  ret i32 %result
+}
+
+define i32 @trivial_wb(i32* %array, i32 %length, i32 %n) {
+; CHECK-LABEL: @trivial_wb(
+; CHECK-NEXT:  entry:
+; CHECK-NEXT:    [[TMP0:%.*]] = icmp ugt i32 [[N:%.*]], 1
+; CHECK-NEXT:    [[UMAX:%.*]] = select i1 [[TMP0]], i32 [[N]], i32 1
+; CHECK-NEXT:    [[TMP1:%.*]] = add i32 [[UMAX]], -1
+; CHECK-NEXT:    [[TMP2:%.*]] = icmp ult i32 [[LENGTH:%.*]], [[TMP1]]
+; CHECK-NEXT:    [[UMIN:%.*]] = select i1 [[TMP2]], i32 [[LENGTH]], i32 [[TMP1]]
+; CHECK-NEXT:    [[TMP3:%.*]] = icmp ugt i32 [[LENGTH]], [[UMIN]]
+; CHECK-NEXT:    [[TMP4:%.*]] = freeze i1 [[TMP3]]
+; CHECK-NEXT:    [[WIDENABLE_COND:%.*]] = call i1 @llvm.experimental.widenable.condition()
+; CHECK-NEXT:    [[TMP5:%.*]] = and i1 [[TMP4]], [[WIDENABLE_COND]]
+; CHECK-NEXT:    br i1 [[TMP5]], label [[LOOP_PREHEADER:%.*]], label [[DEOPT:%.*]], !prof !0
+; CHECK:       deopt:
+; CHECK-NEXT:    [[DEOPTRET:%.*]] = call i32 (...) @llvm.experimental.deoptimize.i32() [ "deopt"() ]
+; CHECK-NEXT:    ret i32 [[DEOPTRET]]
+; CHECK:       loop.preheader:
+; CHECK-NEXT:    br label [[LOOP:%.*]]
+; CHECK:       loop:
+; CHECK-NEXT:    [[LOOP_ACC:%.*]] = phi i32 [ [[LOOP_ACC_NEXT:%.*]], [[GUARDED:%.*]] ], [ 0, [[LOOP_PREHEADER]] ]
+; CHECK-NEXT:    [[I:%.*]] = phi i32 [ [[I_NEXT:%.*]], [[GUARDED]] ], [ 0, [[LOOP_PREHEADER]] ]
+; CHECK-NEXT:    call void @unknown()
+; CHECK-NEXT:    [[WITHIN_BOUNDS:%.*]] = icmp ult i32 [[I]], [[LENGTH]]
+; CHECK-NEXT:    br i1 true, label [[GUARDED]], label [[DEOPT2:%.*]], !prof !0
+; CHECK:       deopt2:
+; CHECK-NEXT:    call void @unknown()
+; CHECK-NEXT:    [[DEOPTRET2:%.*]] = call i32 (...) @llvm.experimental.deoptimize.i32() [ "deopt"() ]
+; CHECK-NEXT:    ret i32 [[DEOPTRET2]]
+; CHECK:       guarded:
+; CHECK-NEXT:    [[I_I64:%.*]] = zext i32 [[I]] to i64
+; CHECK-NEXT:    [[ARRAY_I_PTR:%.*]] = getelementptr inbounds i32, i32* [[ARRAY:%.*]], i64 [[I_I64]]
+; CHECK-NEXT:    [[ARRAY_I:%.*]] = load i32, i32* [[ARRAY_I_PTR]], align 4
+; CHECK-NEXT:    store i32 0, i32* [[ARRAY_I_PTR]]
+; CHECK-NEXT:    [[LOOP_ACC_NEXT]] = add i32 [[LOOP_ACC]], [[ARRAY_I]]
+; CHECK-NEXT:    [[I_NEXT]] = add nuw i32 [[I]], 1
+; CHECK-NEXT:    [[CONTINUE:%.*]] = icmp ult i32 [[I_NEXT]], [[N]]
+; CHECK-NEXT:    br i1 [[CONTINUE]], label [[LOOP]], label [[EXIT:%.*]]
+; CHECK:       exit:
+; CHECK-NEXT:    [[RESULT:%.*]] = phi i32 [ [[LOOP_ACC_NEXT]], [[GUARDED]] ]
+; CHECK-NEXT:    ret i32 [[RESULT]]
+;
+entry:
+  %widenable_cond = call i1 @llvm.experimental.widenable.condition()
+  br i1 %widenable_cond, label %loop.preheader, label %deopt, !prof !0
+
+deopt:
+  %deoptret = call i32 (...) @llvm.experimental.deoptimize.i32() [ "deopt"() ]
+  ret i32 %deoptret
+
+loop.preheader:
+  br label %loop
+
+loop:
+  %loop.acc = phi i32 [ %loop.acc.next, %guarded ], [ 0, %loop.preheader ]
+  %i = phi i32 [ %i.next, %guarded ], [ 0, %loop.preheader ]
+  call void @unknown()
+  %within.bounds = icmp ult i32 %i, %length
+  br i1 %within.bounds, label %guarded, label %deopt2, !prof !0
+
+deopt2:
+  call void @unknown()
+  %deoptret2 = call i32 (...) @llvm.experimental.deoptimize.i32() [ "deopt"() ]
+  ret i32 %deoptret2
+
+guarded:
+  %i.i64 = zext i32 %i to i64
+  %array.i.ptr = getelementptr inbounds i32, i32* %array, i64 %i.i64
+  %array.i = load i32, i32* %array.i.ptr, align 4
+  store i32 0, i32* %array.i.ptr
+  %loop.acc.next = add i32 %loop.acc, %array.i
+  %i.next = add nuw i32 %i, 1
+  %continue = icmp ult i32 %i.next, %n
+  br i1 %continue, label %loop, label %exit
+
+exit:
+  %result = phi i32 [ %loop.acc.next, %guarded ]
+  ret i32 %result
+}
+
 ; TODO: Non-latch exits can still be predicated
 ; This is currently prevented by an overly restrictive profitability check.
 define i32 @todo_unconditional_latch(i32* %array, i32 %length, i1 %cond_0) {
 ; CHECK-LABEL: @todo_unconditional_latch(
 ; CHECK-NEXT:  entry:
 ; CHECK-NEXT:    [[WIDENABLE_COND:%.*]] = call i1 @llvm.experimental.widenable.condition()
 ; CHECK-NEXT:    [[EXIPLICIT_GUARD_COND:%.*]] = and i1 [[COND_0:%.*]], [[WIDENABLE_COND]]
 ; CHECK-NEXT:    br i1 [[EXIPLICIT_GUARD_COND]], label [[LOOP_PREHEADER:%.*]], label [[DEOPT:%.*]], !prof !0

llvm/test/Transforms/SimplifyCFG/wc-widen-block.ll

 deopt3:
   %deoptret3 = call i32 (...) @llvm.experimental.deoptimize.i32() [ "deopt"() ]
   ret i32 %deoptret3
 
 return:
   ret i32 0
 }
 
+define i32 @trivial_wb(i1 %cond_0, i32* %p) {
+; CHECK-LABEL: @trivial_wb(
+; CHECK-NEXT:  entry:
+; CHECK-NEXT:    [[WIDENABLE_COND:%.*]] = call i1 @llvm.experimental.widenable.condition()
+; CHECK-NEXT:    br i1 [[WIDENABLE_COND]], label [[GUARDED:%.*]], label [[DEOPT:%.*]], !prof !0
+; CHECK:       deopt:
+; CHECK-NEXT:    [[DEOPTRET:%.*]] = call i32 (...) @llvm.experimental.deoptimize.i32() [ "deopt"() ]
+; CHECK-NEXT:    ret i32 [[DEOPTRET]]
+; CHECK:       guarded:
+; CHECK-NEXT:    [[V:%.*]] = load i32, i32* [[P:%.*]]
+; CHECK-NEXT:    [[COND_1:%.*]] = icmp eq i32 [[V]], 0
+; CHECK-NEXT:    br i1 [[COND_1]], label [[RETURN:%.*]], label [[DEOPT]], !prof !0
+; CHECK:       return:
+; CHECK-NEXT:    ret i32 0
+;
+entry:
+  %widenable_cond = call i1 @llvm.experimental.widenable.condition()
+  br i1 %widenable_cond, label %guarded, label %deopt, !prof !0
+
+deopt:
+  %deoptret = call i32 (...) @llvm.experimental.deoptimize.i32() [ "deopt"() ]
+  ret i32 %deoptret
+
+guarded:
+  %v = load i32, i32* %p
+  %cond_1 = icmp eq i32 %v, 0
+  br i1 %cond_1, label %return, label %deopt2, !prof !0
+
+deopt2:
+  %deoptret2 = call i32 (...) @llvm.experimental.deoptimize.i32() [ "deopt"() ]
+  ret i32 %deoptret2
+
+return:
+  ret i32 0
+}
+
+
+define i32 @swapped_wb(i1 %cond_0, i32* %p) {
+; CHECK-LABEL: @swapped_wb(
+; CHECK-NEXT:  entry:
+; CHECK-NEXT:    [[WIDENABLE_COND:%.*]] = call i1 @llvm.experimental.widenable.condition()
+; CHECK-NEXT:    [[EXIPLICIT_GUARD_COND:%.*]] = and i1 [[WIDENABLE_COND]], [[COND_0:%.*]]
+; CHECK-NEXT:    br i1 [[EXIPLICIT_GUARD_COND]], label [[GUARDED:%.*]], label [[DEOPT:%.*]], !prof !0
+; CHECK:       deopt:
+; CHECK-NEXT:    [[DEOPTRET:%.*]] = call i32 (...) @llvm.experimental.deoptimize.i32() [ "deopt"() ]
+; CHECK-NEXT:    ret i32 [[DEOPTRET]]
+; CHECK:       guarded:
+; CHECK-NEXT:    [[V:%.*]] = load i32, i32* [[P:%.*]]
+; CHECK-NEXT:    [[COND_1:%.*]] = icmp eq i32 [[V]], 0
+; CHECK-NEXT:    br i1 [[COND_1]], label [[RETURN:%.*]], label [[DEOPT]], !prof !0
+; CHECK:       return:
+; CHECK-NEXT:    ret i32 0
+;
+entry:
+  %widenable_cond = call i1 @llvm.experimental.widenable.condition()
+  %exiplicit_guard_cond = and i1 %widenable_cond, %cond_0
+  br i1 %exiplicit_guard_cond, label %guarded, label %deopt, !prof !0
+
+deopt:
+  %deoptret = call i32 (...) @llvm.experimental.deoptimize.i32() [ "deopt"() ]
+  ret i32 %deoptret
+
+guarded:
+  %v = load i32, i32* %p
+  %cond_1 = icmp eq i32 %v, 0
+  br i1 %cond_1, label %return, label %deopt2, !prof !0
+
+deopt2:
+  %deoptret2 = call i32 (...) @llvm.experimental.deoptimize.i32() [ "deopt"() ]
+  ret i32 %deoptret2
+
+return:
+  ret i32 0
+}
+
 
 
 
 declare void @unknown()
 declare i32 @unknown_i32()
 
 declare i1 @llvm.experimental.widenable.condition()
 declare i32 @llvm.experimental.deoptimize.i32(...)
 
 !0 = !{!"branch_weights", i32 1048576, i32 1}
 !1 = !{i32 1, i32 -2147483648}