[GuardWidening] Choose right point for generating wide condition for branches. PR60234

[mkazantsev]

Summary

When guards are represented as widenable branches, there is a tricky
situation when the branch stays in loop but widenable condition doesn't.
It means that the widenable condition is loop-invariant, and some other
optimizations could have done changes using this fact.

If widening is allowed to create widened condition inside this loop,
and join the loop-invariant wc with some non-invariant facts, it can
cause miscompile. See example of this at https://github.com/llvm/llvm-project/issues/60234.

The solution is to adjust the point of generationg the wide condition,
and therefore of hoisting all related parts there. It should not be before
the branch, but before the widenable_condition call. The fact that wc()
and the wide condition are in the same block guarantees that they will
not violate the invariance property for any loop.

llvm/lib/Transforms/Scalar/GuardWidening.cpp

 // Eliminates the guard instruction properly.
 static void eliminateGuard(Instruction *GuardInst, MemorySSAUpdater *MSSAU) {
   GuardInst->eraseFromParent();
   if (MSSAU)
     MSSAU->removeMemoryAccess(GuardInst);
   ++GuardsEliminated;
 }
 
+/// Find a point at which the widened condition of \p Guard should be inserted.
+/// When it is represented as intrinsic call, we can do it right before the call
+/// instruction. However, when we are dealing with widenable branch, we must
+/// account for the following situation: widening should not turn a
+/// loop-invariant condition into a loop-variant. It means that if
+/// widenable.condition() call is invariant (w.r.t. any loop), the new wide
+/// condition should stay invariant. Otherwise there can be a miscompile, like
+/// the one described at https://github.com/llvm/llvm-project/issues/60234. The
+/// safest way to do it is to expand the new condition at WC's block.
+static Instruction *findInsertionPointForWideCondition(Instruction *Guard) {
+  Value *Condition, *WC;
+  BasicBlock *IfTrue, *IfFalse;
+  if (parseWidenableBranch(Guard, Condition, WC, IfTrue, IfFalse))
+    return cast<Instruction>(WC);
+  return Guard;
+}
+
 class GuardWideningImpl {
   DominatorTree &DT;
   PostDominatorTree *PDT;
   LoopInfo &LI;
   AssumptionCache &AC;
   MemorySSAUpdater *MSSAU;
 
   /// Together, these describe the region of interest.  This might be all of
class GuardWideningImpl {
   }
 
   /// If \p InvertCondition is false, Widen \p ToWiden to fail if
   /// \p NewCondition is false, otherwise make it fail if \p NewCondition is
   /// true (in addition to whatever it is already checking).
   void widenGuard(Instruction *ToWiden, Value *NewCondition,
                   bool InvertCondition) {
     Value *Result;
-
-    widenCondCommon(getCondition(ToWiden), NewCondition, ToWiden, Result,
+    Instruction *InsertPt = findInsertionPointForWideCondition(ToWiden);
+    widenCondCommon(getCondition(ToWiden), NewCondition, InsertPt, Result,
                     InvertCondition);
     if (isGuardAsWidenableBranch(ToWiden)) {
       setWidenableBranchCond(cast<BranchInst>(ToWiden), Result);
       return;
     }
     setCondition(ToWiden, Result);
   }
 
if (DominatingGuardLoop != DominatedInstrLoop) {
     // sibling is colder, we should consider allowing this.
     if (DominatingGuardLoop &&
         !DominatingGuardLoop->contains(DominatedInstrLoop))
       return WS_IllegalOrNegative;
 
     HoistingOutOfLoop = true;
   }
 
-  if (!isAvailableAt(getCondition(DominatedInstr), DominatingGuard))
+  auto *WideningPoint = findInsertionPointForWideCondition(DominatingGuard);
+  if (!isAvailableAt(getCondition(DominatedInstr), WideningPoint))
     return WS_IllegalOrNegative;
 
   // If the guard was conditional executed, it may never be reached
   // dynamically.  There are two potential downsides to hoisting it out of the
   // conditionally executed region: 1) we may spuriously deopt without need and
   // 2) we have the extra cost of computing the guard condition in the common
   // case.  At the moment, we really only consider the second in our heuristic
   // here.  TODO: evaluate cost model for spurious deopt

llvm/test/Transforms/GuardWidening/basic_widenable_condition_guards.ll

 ; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
 ; RUN: opt -S -guard-widening-widen-branch-guards=true -passes=guard-widening < %s        | FileCheck %s
 ; RUN: opt -S -guard-widening-widen-branch-guards=true -passes=guard-widening < %s | FileCheck %s
 
 ; Basic test case: we wide the first check to check both the
 ; conditions.
 define void @f_0(i1 %cond_0, i1 %cond_1) {
 ; CHECK-LABEL: @f_0(
 ; 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:    [[WIDENABLE_COND:%.*]] = call i1 @llvm.experimental.widenable.condition()
 ; CHECK-NEXT:    [[EXIPLICIT_GUARD_COND:%.*]] = and i1 [[WIDE_CHK]], [[WIDENABLE_COND]]
 ; CHECK-NEXT:    br i1 [[EXIPLICIT_GUARD_COND]], label [[GUARDED:%.*]], label [[DEOPT:%.*]], !prof [[PROF0:![0-9]+]]
 ; 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]]

 guarded1:                                         ; preds = %guarded
   ret void
 }
 
 ; Same as @f_0, but with using a more general notion of postdominance.
 define void @f_1(i1 %cond_0, i1 %cond_1) {
 ; CHECK-LABEL: @f_1(
 ; 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:    [[WIDENABLE_COND:%.*]] = call i1 @llvm.experimental.widenable.condition()
 ; CHECK-NEXT:    [[EXIPLICIT_GUARD_COND:%.*]] = and i1 [[WIDE_CHK]], [[WIDENABLE_COND]]
 ; CHECK-NEXT:    br i1 [[EXIPLICIT_GUARD_COND]], label [[GUARDED:%.*]], label [[DEOPT:%.*]], !prof [[PROF0]]
 ; CHECK:       deopt:
 ; CHECK-NEXT:    call void (...) @llvm.experimental.deoptimize.isVoid() [ "deopt"() ]
 ; CHECK-NEXT:    ret void
 ; CHECK:       guarded:
 ; CHECK-NEXT:    br i1 undef, label [[LEFT:%.*]], label [[RIGHT:%.*]]
 ; CHECK:       left:

 }
 
 ; Like @f_1, but we have some code we need to hoist before we can
 ; widen a dominanting check.
 define void @f_2(i32 %a, i32 %b) {
 ; CHECK-LABEL: @f_2(
 ; CHECK-NEXT:  entry:
 ; CHECK-NEXT:    [[COND_0:%.*]] = icmp ult i32 [[A:%.*]], 10
-; CHECK-NEXT:    [[WIDENABLE_COND:%.*]] = call i1 @llvm.experimental.widenable.condition()
 ; CHECK-NEXT:    [[COND_1:%.*]] = icmp ult i32 [[B:%.*]], 10
 ; CHECK-NEXT:    [[WIDE_CHK:%.*]] = and i1 [[COND_0]], [[COND_1]]
+; CHECK-NEXT:    [[WIDENABLE_COND:%.*]] = call i1 @llvm.experimental.widenable.condition()
 ; CHECK-NEXT:    [[EXIPLICIT_GUARD_COND:%.*]] = and i1 [[WIDE_CHK]], [[WIDENABLE_COND]]
 ; CHECK-NEXT:    br i1 [[EXIPLICIT_GUARD_COND]], label [[GUARDED:%.*]], label [[DEOPT:%.*]], !prof [[PROF0]]
 ; CHECK:       deopt:
 ; CHECK-NEXT:    call void (...) @llvm.experimental.deoptimize.isVoid() [ "deopt"() ]
 ; CHECK-NEXT:    ret void
 ; CHECK:       guarded:
 ; CHECK-NEXT:    br i1 undef, label [[LEFT:%.*]], label [[RIGHT:%.*]]
 ; CHECK:       left:

 
 ; But hoisting out of control flow is fine if it makes a loop computed
 ; condition loop invariant.  This behavior may require some tuning in
 ; the future.
 define void @f_4(i32 %a, i32 %b) {
 ; CHECK-LABEL: @f_4(
 ; CHECK-NEXT:  entry:
 ; CHECK-NEXT:    [[COND_0:%.*]] = icmp ult i32 [[A:%.*]], 10
-; CHECK-NEXT:    [[WIDENABLE_COND:%.*]] = call i1 @llvm.experimental.widenable.condition()
 ; CHECK-NEXT:    [[COND_1:%.*]] = icmp ult i32 [[B:%.*]], 10
 ; CHECK-NEXT:    [[WIDE_CHK:%.*]] = and i1 [[COND_0]], [[COND_1]]
+; CHECK-NEXT:    [[WIDENABLE_COND:%.*]] = call i1 @llvm.experimental.widenable.condition()
 ; CHECK-NEXT:    [[EXIPLICIT_GUARD_COND:%.*]] = and i1 [[WIDE_CHK]], [[WIDENABLE_COND]]
 ; CHECK-NEXT:    br i1 [[EXIPLICIT_GUARD_COND]], label [[GUARDED:%.*]], label [[DEOPT:%.*]], !prof [[PROF0]]
 ; CHECK:       deopt:
 ; CHECK-NEXT:    call void (...) @llvm.experimental.deoptimize.isVoid() [ "deopt"() ]
 ; CHECK-NEXT:    ret void
 ; CHECK:       guarded:
 ; CHECK-NEXT:    br i1 undef, label [[LOOP:%.*]], label [[LEAVE:%.*]]
 ; CHECK:       loop:

 }
 
 ; Hoisting out of control flow is also fine if we can widen the
 ; dominating check without doing any extra work.
 define void @f_5(i32 %a) {
 ; CHECK-LABEL: @f_5(
 ; CHECK-NEXT:  entry:
 ; CHECK-NEXT:    [[COND_0:%.*]] = icmp ugt i32 [[A:%.*]], 7
-; CHECK-NEXT:    [[WIDENABLE_COND:%.*]] = call i1 @llvm.experimental.widenable.condition()
 ; CHECK-NEXT:    [[WIDE_CHK:%.*]] = icmp uge i32 [[A]], 11
+; CHECK-NEXT:    [[WIDENABLE_COND:%.*]] = call i1 @llvm.experimental.widenable.condition()
 ; CHECK-NEXT:    [[EXIPLICIT_GUARD_COND:%.*]] = and i1 [[WIDE_CHK]], [[WIDENABLE_COND]]
 ; CHECK-NEXT:    br i1 [[EXIPLICIT_GUARD_COND]], label [[GUARDED:%.*]], label [[DEOPT:%.*]], !prof [[PROF0]]
 ; CHECK:       deopt:
 ; CHECK-NEXT:    call void (...) @llvm.experimental.deoptimize.isVoid() [ "deopt"() ]
 ; CHECK-NEXT:    ret void
 ; CHECK:       guarded:
 ; CHECK-NEXT:    br i1 undef, label [[LEFT:%.*]], label [[RIGHT:%.*]]
 ; CHECK:       left:

 }
 
 ; All else equal, we try to widen the earliest guard we can.  This
 ; heuristic can use some tuning.
 define void @f_7(i32 %a, ptr %cond_buf) {
 ; CHECK-LABEL: @f_7(
 ; CHECK-NEXT:  entry:
 ; CHECK-NEXT:    [[COND_1:%.*]] = load volatile i1, ptr [[COND_BUF:%.*]], align 1
-; CHECK-NEXT:    [[WIDENABLE_COND:%.*]] = call i1 @llvm.experimental.widenable.condition()
 ; CHECK-NEXT:    [[COND_3:%.*]] = icmp ult i32 [[A:%.*]], 7
 ; CHECK-NEXT:    [[WIDE_CHK:%.*]] = and i1 [[COND_1]], [[COND_3]]
+; CHECK-NEXT:    [[WIDENABLE_COND:%.*]] = call i1 @llvm.experimental.widenable.condition()
 ; CHECK-NEXT:    [[EXIPLICIT_GUARD_COND:%.*]] = and i1 [[WIDE_CHK]], [[WIDENABLE_COND]]
 ; CHECK-NEXT:    br i1 [[EXIPLICIT_GUARD_COND]], label [[GUARDED:%.*]], label [[DEOPT:%.*]], !prof [[PROF0]]
 ; CHECK:       deopt:
 ; CHECK-NEXT:    call void (...) @llvm.experimental.deoptimize.isVoid() [ "deopt"() ]
 ; CHECK-NEXT:    ret void
 ; CHECK:       guarded:
 ; CHECK-NEXT:    [[COND_2:%.*]] = load volatile i1, ptr [[COND_BUF]], align 1
 ; CHECK-NEXT:    [[WIDENABLE_COND3:%.*]] = call i1 @llvm.experimental.widenable.condition()

 ; CHECK-NEXT:    [[EXIPLICIT_GUARD_COND:%.*]] = and i1 [[COND_1:%.*]], [[WIDENABLE_COND]]
 ; CHECK-NEXT:    br i1 [[EXIPLICIT_GUARD_COND]], label [[GUARDED:%.*]], label [[DEOPT:%.*]], !prof [[PROF0]]
 ; CHECK:       deopt:
 ; CHECK-NEXT:    call void (...) @llvm.experimental.deoptimize.isVoid() [ "deopt"() ]
 ; CHECK-NEXT:    ret void
 ; CHECK:       guarded:
 ; CHECK-NEXT:    br i1 undef, label [[LOOP]], label [[LEAVE:%.*]]
 ; CHECK:       leave:
-; CHECK-NEXT:    [[WIDENABLE_COND3:%.*]] = call i1 @llvm.experimental.widenable.condition()
 ; CHECK-NEXT:    [[COND_3:%.*]] = icmp ult i32 [[A:%.*]], 7
 ; CHECK-NEXT:    [[WIDE_CHK:%.*]] = and i1 [[COND_2:%.*]], [[COND_3]]
+; CHECK-NEXT:    [[WIDENABLE_COND3:%.*]] = call i1 @llvm.experimental.widenable.condition()
 ; CHECK-NEXT:    [[EXIPLICIT_GUARD_COND4:%.*]] = and i1 [[WIDE_CHK]], [[WIDENABLE_COND3]]
 ; CHECK-NEXT:    br i1 [[EXIPLICIT_GUARD_COND4]], label [[GUARDED1:%.*]], label [[DEOPT2:%.*]], !prof [[PROF0]]
 ; CHECK:       deopt2:
 ; CHECK-NEXT:    call void (...) @llvm.experimental.deoptimize.isVoid() [ "deopt"() ]
 ; CHECK-NEXT:    ret void
 ; CHECK:       guarded1:
 ; CHECK-NEXT:    br i1 undef, label [[LOOP2:%.*]], label [[LEAVE2:%.*]]
 ; CHECK:       loop2:

 
 ; With guards in loops, we're okay hoisting out the guard into the
 ; containing loop.
 define void @f_11(i32 %a, i1 %cond_0, i1 %cond_1) {
 ; CHECK-LABEL: @f_11(
 ; CHECK-NEXT:  entry:
 ; CHECK-NEXT:    br label [[OUTER_HEADER:%.*]]
 ; CHECK:       outer_header:
-; CHECK-NEXT:    [[WIDENABLE_COND:%.*]] = call i1 @llvm.experimental.widenable.condition()
 ; CHECK-NEXT:    [[WIDE_CHK:%.*]] = and i1 [[COND_0:%.*]], [[COND_1:%.*]]
+; CHECK-NEXT:    [[WIDENABLE_COND:%.*]] = call i1 @llvm.experimental.widenable.condition()
 ; CHECK-NEXT:    [[EXIPLICIT_GUARD_COND:%.*]] = and i1 [[WIDE_CHK]], [[WIDENABLE_COND]]
 ; CHECK-NEXT:    br i1 [[EXIPLICIT_GUARD_COND]], label [[GUARDED:%.*]], label [[DEOPT:%.*]], !prof [[PROF0]]
 ; CHECK:       deopt:
 ; CHECK-NEXT:    call void (...) @llvm.experimental.deoptimize.isVoid() [ "deopt"() ]
 ; CHECK-NEXT:    ret void
 ; CHECK:       guarded:
 ; CHECK-NEXT:    br label [[INNER:%.*]]
 ; CHECK:       inner:
exit:                                             ; preds = %outer_latch
   ret void
 }
 
 ; Checks that we are adequately guarded against exponential-time
 ; behavior when hoisting code.
 define void @f_12(i32 %a0) {
 ; CHECK-LABEL: @f_12(
 ; CHECK-NEXT:  entry:
-; CHECK-NEXT:    [[WIDENABLE_COND:%.*]] = call i1 @llvm.experimental.widenable.condition()
 ; CHECK-NEXT:    [[A1:%.*]] = mul i32 [[A0:%.*]], [[A0]]
 ; CHECK-NEXT:    [[A2:%.*]] = mul i32 [[A1]], [[A1]]
 ; CHECK-NEXT:    [[A3:%.*]] = mul i32 [[A2]], [[A2]]
 ; CHECK-NEXT:    [[A4:%.*]] = mul i32 [[A3]], [[A3]]
 ; CHECK-NEXT:    [[A5:%.*]] = mul i32 [[A4]], [[A4]]
 ; CHECK-NEXT:    [[A6:%.*]] = mul i32 [[A5]], [[A5]]
 ; CHECK-NEXT:    [[A7:%.*]] = mul i32 [[A6]], [[A6]]
 ; CHECK-NEXT:    [[A8:%.*]] = mul i32 [[A7]], [[A7]]

 ; CHECK-NEXT:    [[A25:%.*]] = mul i32 [[A24]], [[A24]]
 ; CHECK-NEXT:    [[A26:%.*]] = mul i32 [[A25]], [[A25]]
 ; CHECK-NEXT:    [[A27:%.*]] = mul i32 [[A26]], [[A26]]
 ; CHECK-NEXT:    [[A28:%.*]] = mul i32 [[A27]], [[A27]]
 ; CHECK-NEXT:    [[A29:%.*]] = mul i32 [[A28]], [[A28]]
 ; CHECK-NEXT:    [[A30:%.*]] = mul i32 [[A29]], [[A29]]
 ; CHECK-NEXT:    [[COND:%.*]] = trunc i32 [[A30]] to i1
 ; CHECK-NEXT:    [[WIDE_CHK:%.*]] = and i1 true, [[COND]]
+; CHECK-NEXT:    [[WIDENABLE_COND:%.*]] = call i1 @llvm.experimental.widenable.condition()
 ; CHECK-NEXT:    [[EXIPLICIT_GUARD_COND:%.*]] = and i1 [[WIDE_CHK]], [[WIDENABLE_COND]]
 ; CHECK-NEXT:    br i1 [[EXIPLICIT_GUARD_COND]], label [[GUARDED:%.*]], label [[DEOPT:%.*]], !prof [[PROF0]]
 ; 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]], [[WIDENABLE_COND3]]

 guarded1:                                         ; preds = %guarded
   ret void
 }
 
 define void @f_13(i32 %a) {
 ; CHECK-LABEL: @f_13(
 ; CHECK-NEXT:  entry:
 ; CHECK-NEXT:    [[COND_0:%.*]] = icmp ult i32 [[A:%.*]], 14
-; CHECK-NEXT:    [[WIDENABLE_COND:%.*]] = call i1 @llvm.experimental.widenable.condition()
 ; CHECK-NEXT:    [[WIDE_CHK:%.*]] = icmp ult i32 [[A]], 10
+; CHECK-NEXT:    [[WIDENABLE_COND:%.*]] = call i1 @llvm.experimental.widenable.condition()
 ; CHECK-NEXT:    [[EXIPLICIT_GUARD_COND:%.*]] = and i1 [[WIDE_CHK]], [[WIDENABLE_COND]]
 ; CHECK-NEXT:    br i1 [[EXIPLICIT_GUARD_COND]], label [[GUARDED:%.*]], label [[DEOPT:%.*]], !prof [[PROF0]]
 ; CHECK:       deopt:
 ; CHECK-NEXT:    call void (...) @llvm.experimental.deoptimize.isVoid() [ "deopt"() ]
 ; CHECK-NEXT:    ret void
 ; CHECK:       guarded:
 ; CHECK-NEXT:    br i1 undef, label [[LEFT:%.*]], label [[RIGHT:%.*]]
 ; CHECK:       left:

 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:    [[WIDENABLE_COND:%.*]] = call i1 @llvm.experimental.widenable.condition()
 ; CHECK-NEXT:    [[EXIPLICIT_GUARD_COND:%.*]] = and i1 [[WIDENABLE_COND]], [[WIDE_CHK]]
 ; CHECK-NEXT:    br i1 [[EXIPLICIT_GUARD_COND]], label [[GUARDED:%.*]], label [[DEOPT:%.*]], !prof [[PROF0]]
 ; 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]]

 
 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:    [[WIDENABLE_COND:%.*]] = call i1 @llvm.experimental.widenable.condition()
 ; CHECK-NEXT:    [[TMP0:%.*]] = and i1 [[WIDE_CHK]], [[WIDENABLE_COND]]
 ; CHECK-NEXT:    br i1 [[TMP0]], label [[GUARDED:%.*]], label [[DEOPT:%.*]], !prof [[PROF0]]
 ; 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]]

llvm/test/Transforms/GuardWidening/loop_invariant_widenable_condition.ll

 ; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
 ; RUN: opt -passes=guard-widening -S < %s | FileCheck %s
 
 declare i32 @llvm.experimental.deoptimize.i32(...)
 
-; FIXME: Make sure the two loop-invariant conditions can be widened together,
-;        and the widening point is outside the loop.
+; Make sure the two loop-invariant conditions can be widened together,
+; and the widening point is outside the loop.
 define i32 @test_01(i32 %start, i32 %x) {
 ; CHECK-LABEL: @test_01(
 ; CHECK-NEXT:  entry:
 ; CHECK-NEXT:    [[COND:%.*]] = icmp eq i32 [[START:%.*]], [[X:%.*]]
+; CHECK-NEXT:    [[WIDE_CHK:%.*]] = and i1 true, [[COND]]
 ; CHECK-NEXT:    [[WC1:%.*]] = call i1 @llvm.experimental.widenable.condition()
 ; CHECK-NEXT:    br label [[LOOP:%.*]]
 ; CHECK:       loop:
 ; CHECK-NEXT:    [[IV:%.*]] = phi i32 [ [[START]], [[ENTRY:%.*]] ], [ [[IV_NEXT:%.*]], [[BACKEDGE:%.*]] ]
-; CHECK-NEXT:    [[WIDE_CHK:%.*]] = and i1 true, [[COND]]
 ; CHECK-NEXT:    [[TMP0:%.*]] = and i1 [[WIDE_CHK]], [[WC1]]
 ; CHECK-NEXT:    br i1 [[TMP0]], label [[GUARD_BLOCK:%.*]], label [[EXIT_BY_WC:%.*]]
 ; CHECK:       exit_by_wc:
 ; CHECK-NEXT:    [[RVAL1:%.*]] = call i32 (...) @llvm.experimental.deoptimize.i32() [ "deopt"(i32 [[IV]]) ]
 ; CHECK-NEXT:    ret i32 [[RVAL1]]
 ; CHECK:       guard_block:
 ; CHECK-NEXT:    [[WC2:%.*]] = call i1 @llvm.experimental.widenable.condition()
 ; CHECK-NEXT:    [[GUARD:%.*]] = and i1 [[COND]], [[WC2]]
exit:
   ret i32 -1
 
 failure:
   %rval2 = call i32(...) @llvm.experimental.deoptimize.i32() [ "deopt"(i32 %iv) ]
   ret i32 %rval2
 }
 
 
-; FIXME: Make sure the loop-variant condition is not widened into loop-invariant.
+; Make sure the loop-variant condition is not widened into loop-invariant.
 define i32 @test_02(i32 %start, i32 %x) {
 ; CHECK-LABEL: @test_02(
 ; CHECK-NEXT:  entry:
 ; CHECK-NEXT:    [[WC1:%.*]] = call i1 @llvm.experimental.widenable.condition()
 ; CHECK-NEXT:    br label [[LOOP:%.*]]
 ; CHECK:       loop:
 ; CHECK-NEXT:    [[IV:%.*]] = phi i32 [ [[START:%.*]], [[ENTRY:%.*]] ], [ [[IV_NEXT:%.*]], [[BACKEDGE:%.*]] ]
-; CHECK-NEXT:    [[COND:%.*]] = icmp eq i32 [[IV]], [[X:%.*]]
-; CHECK-NEXT:    [[WIDE_CHK:%.*]] = and i1 true, [[COND]]
-; CHECK-NEXT:    [[TMP0:%.*]] = and i1 [[WIDE_CHK]], [[WC1]]
-; CHECK-NEXT:    br i1 [[TMP0]], label [[GUARD_BLOCK:%.*]], label [[EXIT_BY_WC:%.*]]
+; CHECK-NEXT:    br i1 [[WC1]], label [[GUARD_BLOCK:%.*]], label [[EXIT_BY_WC:%.*]]
 ; CHECK:       exit_by_wc:
 ; CHECK-NEXT:    [[RVAL1:%.*]] = call i32 (...) @llvm.experimental.deoptimize.i32() [ "deopt"(i32 [[IV]]) ]
 ; CHECK-NEXT:    ret i32 [[RVAL1]]
 ; CHECK:       guard_block:
+; CHECK-NEXT:    [[COND:%.*]] = icmp eq i32 [[IV]], [[X:%.*]]
 ; CHECK-NEXT:    [[WC2:%.*]] = call i1 @llvm.experimental.widenable.condition()
 ; CHECK-NEXT:    [[GUARD:%.*]] = and i1 [[COND]], [[WC2]]
-; CHECK-NEXT:    br i1 true, label [[BACKEDGE]], label [[FAILURE:%.*]]
+; CHECK-NEXT:    br i1 [[GUARD]], label [[BACKEDGE]], label [[FAILURE:%.*]]
 ; CHECK:       backedge:
 ; CHECK-NEXT:    call void @side_effect()
 ; CHECK-NEXT:    [[IV_NEXT]] = add i32 [[IV]], 1
 ; CHECK-NEXT:    br label [[LOOP]]
 ; CHECK:       exit:
 ; CHECK-NEXT:    ret i32 -1
 ; CHECK:       failure:
 ; CHECK-NEXT:    [[RVAL2:%.*]] = call i32 (...) @llvm.experimental.deoptimize.i32() [ "deopt"(i32 [[IV]]) ]

 exit:
   ret i32 -1
 
 failure:
   %rval2 = call i32(...) @llvm.experimental.deoptimize.i32() [ "deopt"(i32 %iv) ]
   ret i32 %rval2
 }
 
-; FIXME: Same as test_01, but the initial condition is not immediately WC.
+; Same as test_01, but the initial condition is not immediately WC.
 define i32 @test_03(i32 %start, i32 %x, i1 %c) {
 ; CHECK-LABEL: @test_03(
 ; CHECK-NEXT:  entry:
 ; CHECK-NEXT:    [[COND:%.*]] = icmp eq i32 [[START:%.*]], [[X:%.*]]
+; CHECK-NEXT:    [[WIDE_CHK:%.*]] = and i1 [[C:%.*]], [[COND]]
 ; CHECK-NEXT:    [[WC1:%.*]] = call i1 @llvm.experimental.widenable.condition()
 ; CHECK-NEXT:    br label [[LOOP:%.*]]
 ; CHECK:       loop:
 ; CHECK-NEXT:    [[IV:%.*]] = phi i32 [ [[START]], [[ENTRY:%.*]] ], [ [[IV_NEXT:%.*]], [[BACKEDGE:%.*]] ]
-; CHECK-NEXT:    [[WIDE_CHK:%.*]] = and i1 [[C:%.*]], [[COND]]
 ; CHECK-NEXT:    [[INVARIANT:%.*]] = and i1 [[WIDE_CHK]], [[WC1]]
 ; CHECK-NEXT:    br i1 [[INVARIANT]], label [[GUARD_BLOCK:%.*]], label [[EXIT_BY_WC:%.*]]
 ; CHECK:       exit_by_wc:
 ; CHECK-NEXT:    [[RVAL1:%.*]] = call i32 (...) @llvm.experimental.deoptimize.i32() [ "deopt"(i32 [[IV]]) ]
 ; CHECK-NEXT:    ret i32 [[RVAL1]]
 ; CHECK:       guard_block:
 ; CHECK-NEXT:    [[WC2:%.*]] = call i1 @llvm.experimental.widenable.condition()
 ; CHECK-NEXT:    [[GUARD:%.*]] = and i1 [[COND]], [[WC2]]

llvm/test/Transforms/GuardWidening/mixed_guards.ll

 guarded1:                                         ; preds = %guarded
   ret void
 }
 
 ; Widen condition of widenable condition guard with a condition from intrinsic.
 define void @test_02(i1 %cond_0, i1 %cond_1) {
 ; CHECK-LABEL: @test_02(
 ; 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:    [[WIDENABLE_COND:%.*]] = call i1 @llvm.experimental.widenable.condition()
 ; CHECK-NEXT:    [[EXIPLICIT_GUARD_COND:%.*]] = and i1 [[WIDE_CHK]], [[WIDENABLE_COND]]
 ; CHECK-NEXT:    br i1 [[EXIPLICIT_GUARD_COND]], label [[GUARDED:%.*]], label [[DEOPT:%.*]], !prof [[PROF0]]
 ; CHECK:       deopt:
 ; CHECK-NEXT:    call void (...) @llvm.experimental.deoptimize.isVoid() [ "deopt"() ]
 ; CHECK-NEXT:    ret void
 ; CHECK:       guarded:
 ; CHECK-NEXT:    ret void
 ;

llvm/test/Transforms/GuardWidening/pr60234.ll

 ; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
 ; RUN: opt -passes=indvars,guard-widening -S < %s | FileCheck %s
 
 declare i32 @llvm.experimental.deoptimize.i32(...)
 
-; FIXME: Make sure that guard widening does not turn loop-invariant condition
-;        (that then gets optimized basing on this fact) into non-invariant.
-;        https://github.com/llvm/llvm-project/issues/60234 explains how it causes
-;        a miscompile.
+; Make sure that guard widening does not turn loop-invariant condition
+; (that then gets optimized basing on this fact) into non-invariant.
+; https://github.com/llvm/llvm-project/issues/60234 explains how it causes
+; a miscompile.
 define i32 @test(i32 %start) {
 ; CHECK-LABEL: @test(
 ; CHECK-NEXT:  entry:
 ; CHECK-NEXT:    [[WC1:%.*]] = call i1 @llvm.experimental.widenable.condition()
 ; CHECK-NEXT:    br label [[LOOP:%.*]]
 ; CHECK:       loop:
 ; CHECK-NEXT:    [[IV:%.*]] = phi i32 [ [[START:%.*]], [[ENTRY:%.*]] ], [ [[IV_NEXT:%.*]], [[BACKEDGE:%.*]] ]
-; CHECK-NEXT:    [[START_PLUS_1:%.*]] = add i32 [[START]], 1
-; CHECK-NEXT:    [[COND:%.*]] = icmp ne i32 [[START_PLUS_1]], [[IV]]
-; CHECK-NEXT:    [[WIDE_CHK:%.*]] = and i1 true, [[COND]]
-; CHECK-NEXT:    [[TMP0:%.*]] = and i1 [[WIDE_CHK]], [[WC1]]
-; CHECK-NEXT:    br i1 [[TMP0]], label [[GUARD_BLOCK:%.*]], label [[EXIT_BY_WC:%.*]]
+; CHECK-NEXT:    br i1 [[WC1]], label [[GUARD_BLOCK:%.*]], label [[EXIT_BY_WC:%.*]]
 ; CHECK:       exit_by_wc:
 ; CHECK-NEXT:    [[IV_LCSSA:%.*]] = phi i32 [ [[START]], [[LOOP]] ]
 ; CHECK-NEXT:    [[RVAL1:%.*]] = call i32 (...) @llvm.experimental.deoptimize.i32() [ "deopt"(i32 [[IV_LCSSA]]) ]
 ; CHECK-NEXT:    ret i32 [[RVAL1]]
 ; CHECK:       guard_block:
+; CHECK-NEXT:    [[START_PLUS_1:%.*]] = add i32 [[START]], 1
+; CHECK-NEXT:    [[COND:%.*]] = icmp ne i32 [[START_PLUS_1]], [[IV]]
 ; CHECK-NEXT:    [[WC2:%.*]] = call i1 @llvm.experimental.widenable.condition()
 ; CHECK-NEXT:    [[GUARD:%.*]] = and i1 [[COND]], [[WC2]]
-; CHECK-NEXT:    br i1 true, label [[BACKEDGE]], label [[FAILURE:%.*]]
+; CHECK-NEXT:    br i1 [[GUARD]], label [[BACKEDGE]], label [[FAILURE:%.*]]
 ; CHECK:       backedge:
 ; CHECK-NEXT:    call void @side_effect()
 ; CHECK-NEXT:    [[IV_NEXT]] = add i32 [[IV]], 1
 ; CHECK-NEXT:    br label [[LOOP]]
 ; CHECK:       exit:
 ; CHECK-NEXT:    ret i32 -1
 ; CHECK:       failure:
 ; CHECK-NEXT:    [[IV_LCSSA1:%.*]] = phi i32 [ [[IV]], [[GUARD_BLOCK]] ]