[LoopPeel] Reduce peel count for chains of invariant phis.

[caojoshua]

Summary

If a Phi takes a previous phi in the same block as input, the phi
becomes loop invariant in the same iteration as the input.

For example, both %a and %b become loop invariant on the second
iteration:

for (...) {
  %a = phi(0, %invariant)
  %b = phi(1, %a)
}

The peeling with size limit test is modified to test its intended use
case.

[caojoshua]

Update commit msg

[caojoshua]

rename variable

[caojoshua]

Add another test case

[nikic]

This is not how phis work. The %a in phi(1, %a) refers to the previous loop iteration.

[fhahn]

llvm/lib/Transforms/Utils/LoopPeel.cpp

233

The current patch doesn't apply cleanly on main so I cannot check, but is this covered by a test?

236

nit: redundant ()

llvm/test/Transforms/LoopUnroll/peel-loop-not-forced.ll

387

Can the new tests/this change be done separately?

[caojoshua]

In D151866#4386452, @nikic wrote:

This is not how phis work. The %a in phi(1, %a) refers to the previous loop iteration.

oops. Thanks.

llvm/lib/Transforms/Utils/LoopPeel.cpp

if (L.isLoopInvariant(&V))
     return (IterationsToInvariance[&V] = 0);
   if (const PHINode *Phi = dyn_cast<PHINode>(&V)) {
     if (Phi->getParent() != L.getHeader()) {
       // Phi is not in header block so Unknown.
       assert(IterationsToInvariance[&V] == Unknown && "unexpected value saved");
       return Unknown;
     }
     // We need to analyze the input from the back edge and add 1.
-    Value *Input = Phi->getIncomingValueForBlock(L.getLoopLatch());
+    const Value *Input = Phi->getIncomingValueForBlock(L.getLoopLatch());
     PeelCounter Iterations = calculate(*Input);
     assert(IterationsToInvariance[Input] == Iterations &&
            "unexpected value saved");
+    if (const PHINode *PhiInput = dyn_cast<PHINode>(Input))
+      if (Phi->getParent() == PhiInput->getParent() &&
+          PhiInput->comesBefore(Phi))
+        // If the input is from a previous phi in the same block, Phi becomes
+        // loop invariant in the same iteration.
+        return (IterationsToInvariance[Phi] = Iterations);
     return (IterationsToInvariance[Phi] = addOne(Iterations));
   }
   if (const Instruction *I = dyn_cast<Instruction>(&V)) {
     if (isa<CmpInst>(I) || I->isBinaryOp()) {
       // Binary instructions get the max of the operands.
       PeelCounter LHS = calculate(*I->getOperand(0));
       if (LHS == Unknown)
         return Unknown;

llvm/test/Transforms/LoopUnroll/peel-loop-not-forced.ll

loop:
 
   br i1 %cmp, label %loop, label %exit
 
 exit:
   ret i32 %sum
 }
 
 define i32 @invariant_backedge_2(i32 %a, i32 %b) {
-; This loop should be peeled twice because it has a Phi which becomes invariant
-; starting from 3rd iteration.
+; This loop should be peeled once because it has Phis which becomes invariant
+; starting from 2nd iteration.
 ; CHECK-LABEL: define i32 @invariant_backedge_2
 ; CHECK-SAME: (i32 [[A:%.*]], i32 [[B:%.*]]) {
 ; CHECK-NEXT:  entry:
 ; CHECK-NEXT:    br label [[LOOP_PEEL_BEGIN:%.*]]
 ; CHECK:       loop.peel.begin:
 ; CHECK-NEXT:    br label [[LOOP_PEEL:%.*]]
 ; CHECK:       loop.peel:
 ; CHECK-NEXT:    [[INCSUM_PEEL:%.*]] = add i32 0, [[A]]
 ; CHECK-NEXT:    [[INC_PEEL:%.*]] = add i32 0, 1
 ; CHECK-NEXT:    [[CMP_PEEL:%.*]] = icmp slt i32 0, 1000
 ; CHECK-NEXT:    br i1 [[CMP_PEEL]], label [[LOOP_PEEL_NEXT:%.*]], label [[EXIT:%.*]]
 ; CHECK:       loop.peel.next:
-; CHECK-NEXT:    br label [[LOOP_PEEL2:%.*]]
-; CHECK:       loop.peel2:
-; CHECK-NEXT:    [[INCSUM_PEEL3:%.*]] = add i32 [[INCSUM_PEEL]], [[A]]
-; CHECK-NEXT:    [[INC_PEEL4:%.*]] = add i32 [[INC_PEEL]], 1
-; CHECK-NEXT:    [[CMP_PEEL5:%.*]] = icmp slt i32 [[INC_PEEL]], 1000
-; CHECK-NEXT:    br i1 [[CMP_PEEL5]], label [[LOOP_PEEL_NEXT1:%.*]], label [[EXIT]]
+; CHECK-NEXT:    br label [[LOOP_PEEL_NEXT1:%.*]]
 ; CHECK:       loop.peel.next1:
-; CHECK-NEXT:    br label [[LOOP_PEEL_NEXT6:%.*]]
-; CHECK:       loop.peel.next6:
 ; CHECK-NEXT:    br label [[ENTRY_PEEL_NEWPH:%.*]]
 ; CHECK:       entry.peel.newph:
 ; CHECK-NEXT:    br label [[LOOP:%.*]]
 ; CHECK:       loop:
-; CHECK-NEXT:    [[I:%.*]] = phi i32 [ [[INC_PEEL4]], [[ENTRY_PEEL_NEWPH]] ], [ [[INC:%.*]], [[LOOP]] ]
-; CHECK-NEXT:    [[SUM:%.*]] = phi i32 [ [[INCSUM_PEEL3]], [[ENTRY_PEEL_NEWPH]] ], [ [[INCSUM:%.*]], [[LOOP]] ]
-; CHECK-NEXT:    [[INCSUM]] = add i32 [[SUM]], [[B]]
+; CHECK-NEXT:    [[I:%.*]] = phi i32 [ [[INC_PEEL]], [[ENTRY_PEEL_NEWPH]] ], [ [[INC:%.*]], [[LOOP]] ]
+; CHECK-NEXT:    [[SUM:%.*]] = phi i32 [ [[INCSUM_PEEL]], [[ENTRY_PEEL_NEWPH]] ], [ [[INCSUM:%.*]], [[LOOP]] ]
+; CHECK-NEXT:    [[PLUS:%.*]] = phi i32 [ [[A]], [[ENTRY_PEEL_NEWPH]] ], [ [[B]], [[LOOP]] ]
+; CHECK-NEXT:    [[INCSUM]] = add i32 [[SUM]], [[PLUS]]
 ; CHECK-NEXT:    [[INC]] = add nuw nsw i32 [[I]], 1
 ; CHECK-NEXT:    [[CMP:%.*]] = icmp ult i32 [[I]], 1000
 ; CHECK-NEXT:    br i1 [[CMP]], label [[LOOP]], label [[EXIT_LOOPEXIT:%.*]], !llvm.loop [[LOOP2:![0-9]+]]
 ; CHECK:       exit.loopexit:
 ; CHECK-NEXT:    [[SUM_LCSSA_PH:%.*]] = phi i32 [ [[SUM]], [[LOOP]] ]
 ; CHECK-NEXT:    br label [[EXIT]]
 ; CHECK:       exit:
-; CHECK-NEXT:    [[SUM_LCSSA:%.*]] = phi i32 [ 0, [[LOOP_PEEL]] ], [ [[INCSUM_PEEL]], [[LOOP_PEEL2]] ], [ [[SUM_LCSSA_PH]], [[EXIT_LOOPEXIT]] ]
+; CHECK-NEXT:    [[SUM_LCSSA:%.*]] = phi i32 [ 0, [[LOOP_PEEL]] ], [ [[SUM_LCSSA_PH]], [[EXIT_LOOPEXIT]] ]
 ; CHECK-NEXT:    ret i32 [[SUM_LCSSA]]
 ;
 ; DISABLE-LABEL: define i32 @invariant_backedge_2
 ; DISABLE-SAME: (i32 [[A:%.*]], i32 [[B:%.*]]) {
 ; DISABLE-NEXT:  entry:
 ; DISABLE-NEXT:    br label [[LOOP:%.*]]
 ; DISABLE:       loop:
 ; DISABLE-NEXT:    [[I:%.*]] = phi i32 [ 0, [[ENTRY:%.*]] ], [ [[INC:%.*]], [[LOOP]] ]
loop:
 
   br i1 %cmp, label %loop, label %exit
 
 exit:
   ret i32 %sum
 }
 
 define i32 @invariant_backedge_3(i32 %a, i32 %b) {
-; This loop should be peeled thrice because it has a Phi which becomes invariant
-; starting from 4th iteration.
+; This loop should be peeled once because it has Phis which becomes invariant
+; starting from 2nd iteration.
 ; CHECK-LABEL: define i32 @invariant_backedge_3
 ; CHECK-SAME: (i32 [[A:%.*]], i32 [[B:%.*]]) {
 ; CHECK-NEXT:  entry:
 ; CHECK-NEXT:    br label [[LOOP_PEEL_BEGIN:%.*]]
 ; CHECK:       loop.peel.begin:
 ; CHECK-NEXT:    br label [[LOOP_PEEL:%.*]]
 ; CHECK:       loop.peel:
 ; CHECK-NEXT:    [[INCSUM_PEEL:%.*]] = add i32 0, [[A]]
 ; CHECK-NEXT:    [[INC_PEEL:%.*]] = add i32 0, 1
 ; CHECK-NEXT:    [[CMP_PEEL:%.*]] = icmp slt i32 0, 1000
 ; CHECK-NEXT:    br i1 [[CMP_PEEL]], label [[LOOP_PEEL_NEXT:%.*]], label [[EXIT:%.*]]
 ; CHECK:       loop.peel.next:
+; CHECK-NEXT:    br label [[LOOP_PEEL_NEXT1:%.*]]
+; CHECK:       loop.peel.next1:
+; CHECK-NEXT:    br label [[ENTRY_PEEL_NEWPH:%.*]]
+; CHECK:       entry.peel.newph:
+; CHECK-NEXT:    br label [[LOOP:%.*]]
+; CHECK:       loop:
+; CHECK-NEXT:    [[I:%.*]] = phi i32 [ [[INC_PEEL]], [[ENTRY_PEEL_NEWPH]] ], [ [[INC:%.*]], [[LOOP]] ]
+; CHECK-NEXT:    [[SUM:%.*]] = phi i32 [ [[INCSUM_PEEL]], [[ENTRY_PEEL_NEWPH]] ], [ [[INCSUM:%.*]], [[LOOP]] ]
+; CHECK-NEXT:    [[HALF_INV_2:%.*]] = phi i32 [ [[A]], [[ENTRY_PEEL_NEWPH]] ], [ [[B]], [[LOOP]] ]
+; CHECK-NEXT:    [[PLUS:%.*]] = phi i32 [ [[A]], [[ENTRY_PEEL_NEWPH]] ], [ [[HALF_INV_2]], [[LOOP]] ]
+; CHECK-NEXT:    [[INCSUM]] = add i32 [[SUM]], [[PLUS]]
+; CHECK-NEXT:    [[INC]] = add nuw nsw i32 [[I]], 1
+; CHECK-NEXT:    [[CMP:%.*]] = icmp ult i32 [[I]], 1000
+; CHECK-NEXT:    br i1 [[CMP]], label [[LOOP]], label [[EXIT_LOOPEXIT:%.*]], !llvm.loop [[LOOP3:![0-9]+]]
+; CHECK:       exit.loopexit:
+; CHECK-NEXT:    [[SUM_LCSSA_PH:%.*]] = phi i32 [ [[SUM]], [[LOOP]] ]
+; CHECK-NEXT:    br label [[EXIT]]
+; CHECK:       exit:
+; CHECK-NEXT:    [[SUM_LCSSA:%.*]] = phi i32 [ 0, [[LOOP_PEEL]] ], [ [[SUM_LCSSA_PH]], [[EXIT_LOOPEXIT]] ]
+; CHECK-NEXT:    ret i32 [[SUM_LCSSA]]
+;
+; DISABLE-LABEL: define i32 @invariant_backedge_3
+; DISABLE-SAME: (i32 [[A:%.*]], i32 [[B:%.*]]) {
+; DISABLE-NEXT:  entry:
+; DISABLE-NEXT:    br label [[LOOP:%.*]]
+; DISABLE:       loop:
+; DISABLE-NEXT:    [[I:%.*]] = phi i32 [ 0, [[ENTRY:%.*]] ], [ [[INC:%.*]], [[LOOP]] ]
+; DISABLE-NEXT:    [[SUM:%.*]] = phi i32 [ 0, [[ENTRY]] ], [ [[INCSUM:%.*]], [[LOOP]] ]
+; DISABLE-NEXT:    [[HALF_INV:%.*]] = phi i32 [ [[A]], [[ENTRY]] ], [ [[B]], [[LOOP]] ]
+; DISABLE-NEXT:    [[HALF_INV_2:%.*]] = phi i32 [ [[A]], [[ENTRY]] ], [ [[HALF_INV]], [[LOOP]] ]
+; DISABLE-NEXT:    [[PLUS:%.*]] = phi i32 [ [[A]], [[ENTRY]] ], [ [[HALF_INV_2]], [[LOOP]] ]
+; DISABLE-NEXT:    [[INCSUM]] = add i32 [[SUM]], [[PLUS]]
+; DISABLE-NEXT:    [[INC]] = add i32 [[I]], 1
+; DISABLE-NEXT:    [[CMP:%.*]] = icmp slt i32 [[I]], 1000
+; DISABLE-NEXT:    br i1 [[CMP]], label [[LOOP]], label [[EXIT:%.*]]
+; DISABLE:       exit:
+; DISABLE-NEXT:    [[SUM_LCSSA:%.*]] = phi i32 [ [[SUM]], [[LOOP]] ]
+; DISABLE-NEXT:    ret i32 [[SUM_LCSSA]]
+;
+entry:
+  br label %loop
+
+loop:
+  %i = phi i32 [ 0, %entry ], [ %inc, %loop ]
+  %sum = phi i32 [ 0, %entry ], [ %incsum, %loop ]
+  %half.inv = phi i32 [ %a, %entry ], [ %b, %loop ]
+  %half.inv.2 = phi i32 [ %a, %entry ], [ %half.inv, %loop ]
+  %plus = phi i32 [ %a, %entry ], [ %half.inv.2, %loop ]
+
+  %incsum = add i32 %sum, %plus
+  %inc = add i32 %i, 1
+  %cmp = icmp slt i32 %i, 1000
+
+  br i1 %cmp, label %loop, label %exit
+
+exit:
+  ret i32 %sum
+}
+
+define i32 @invariant_backedge_phi_input_after(i32 %a, i32 %b) {
+; This loop should be peeled twice because %plus does becomes loop invariant
+; on the third iteration, one iteration after %half.inv.
+; CHECK-LABEL: define i32 @invariant_backedge_phi_input_after
+; CHECK-SAME: (i32 [[A:%.*]], i32 [[B:%.*]]) {
+; CHECK-NEXT:  entry:
+; CHECK-NEXT:    br label [[LOOP_PEEL_BEGIN:%.*]]
+; CHECK:       loop.peel.begin:
+; CHECK-NEXT:    br label [[LOOP_PEEL:%.*]]
+; CHECK:       loop.peel:
+; CHECK-NEXT:    [[INCSUM_PEEL:%.*]] = add i32 0, [[A]]
+; CHECK-NEXT:    [[INC_PEEL:%.*]] = add i32 0, 1
+; CHECK-NEXT:    [[CMP_PEEL:%.*]] = icmp slt i32 0, 1000
+; CHECK-NEXT:    br i1 [[CMP_PEEL]], label [[LOOP_PEEL_NEXT:%.*]], label [[EXIT:%.*]]
+; CHECK:       loop.peel.next:
 ; CHECK-NEXT:    br label [[LOOP_PEEL2:%.*]]
 ; CHECK:       loop.peel2:
 ; CHECK-NEXT:    [[INCSUM_PEEL3:%.*]] = add i32 [[INCSUM_PEEL]], [[A]]
 ; CHECK-NEXT:    [[INC_PEEL4:%.*]] = add i32 [[INC_PEEL]], 1
 ; CHECK-NEXT:    [[CMP_PEEL5:%.*]] = icmp slt i32 [[INC_PEEL]], 1000
 ; CHECK-NEXT:    br i1 [[CMP_PEEL5]], label [[LOOP_PEEL_NEXT1:%.*]], label [[EXIT]]
 ; CHECK:       loop.peel.next1:
-; CHECK-NEXT:    br label [[LOOP_PEEL7:%.*]]
-; CHECK:       loop.peel7:
-; CHECK-NEXT:    [[INCSUM_PEEL8:%.*]] = add i32 [[INCSUM_PEEL3]], [[A]]
-; CHECK-NEXT:    [[INC_PEEL9:%.*]] = add i32 [[INC_PEEL4]], 1
-; CHECK-NEXT:    [[CMP_PEEL10:%.*]] = icmp slt i32 [[INC_PEEL4]], 1000
-; CHECK-NEXT:    br i1 [[CMP_PEEL10]], label [[LOOP_PEEL_NEXT6:%.*]], label [[EXIT]]
+; CHECK-NEXT:    br label [[LOOP_PEEL_NEXT6:%.*]]
 ; CHECK:       loop.peel.next6:
-; CHECK-NEXT:    br label [[LOOP_PEEL_NEXT11:%.*]]
-; CHECK:       loop.peel.next11:
 ; CHECK-NEXT:    br label [[ENTRY_PEEL_NEWPH:%.*]]
 ; CHECK:       entry.peel.newph:
 ; CHECK-NEXT:    br label [[LOOP:%.*]]
 ; CHECK:       loop:
-; CHECK-NEXT:    [[I:%.*]] = phi i32 [ [[INC_PEEL9]], [[ENTRY_PEEL_NEWPH]] ], [ [[INC:%.*]], [[LOOP]] ]
-; CHECK-NEXT:    [[SUM:%.*]] = phi i32 [ [[INCSUM_PEEL8]], [[ENTRY_PEEL_NEWPH]] ], [ [[INCSUM:%.*]], [[LOOP]] ]
+; CHECK-NEXT:    [[I:%.*]] = phi i32 [ [[INC_PEEL4]], [[ENTRY_PEEL_NEWPH]] ], [ [[INC:%.*]], [[LOOP]] ]
+; CHECK-NEXT:    [[SUM:%.*]] = phi i32 [ [[INCSUM_PEEL3]], [[ENTRY_PEEL_NEWPH]] ], [ [[INCSUM:%.*]], [[LOOP]] ]
 ; CHECK-NEXT:    [[INCSUM]] = add i32 [[SUM]], [[B]]
 ; CHECK-NEXT:    [[INC]] = add nuw nsw i32 [[I]], 1
 ; CHECK-NEXT:    [[CMP:%.*]] = icmp ult i32 [[I]], 1000
 ; CHECK-NEXT:    br i1 [[CMP]], label [[LOOP]], label [[EXIT_LOOPEXIT:%.*]], !llvm.loop [[LOOP4:![0-9]+]]
 ; CHECK:       exit.loopexit:
 ; CHECK-NEXT:    [[SUM_LCSSA_PH:%.*]] = phi i32 [ [[SUM]], [[LOOP]] ]
 ; CHECK-NEXT:    br label [[EXIT]]
 ; CHECK:       exit:
-; CHECK-NEXT:    [[SUM_LCSSA:%.*]] = phi i32 [ 0, [[LOOP_PEEL]] ], [ [[INCSUM_PEEL]], [[LOOP_PEEL2]] ], [ [[INCSUM_PEEL3]], [[LOOP_PEEL7]] ], [ [[SUM_LCSSA_PH]], [[EXIT_LOOPEXIT]] ]
+; CHECK-NEXT:    [[SUM_LCSSA:%.*]] = phi i32 [ 0, [[LOOP_PEEL]] ], [ [[INCSUM_PEEL]], [[LOOP_PEEL2]] ], [ [[SUM_LCSSA_PH]], [[EXIT_LOOPEXIT]] ]
 ; CHECK-NEXT:    ret i32 [[SUM_LCSSA]]
 ;
-; DISABLE-LABEL: define i32 @invariant_backedge_3
+; DISABLE-LABEL: define i32 @invariant_backedge_phi_input_after
 ; DISABLE-SAME: (i32 [[A:%.*]], i32 [[B:%.*]]) {
 ; DISABLE-NEXT:  entry:
 ; DISABLE-NEXT:    br label [[LOOP:%.*]]
 ; DISABLE:       loop:
 ; DISABLE-NEXT:    [[I:%.*]] = phi i32 [ 0, [[ENTRY:%.*]] ], [ [[INC:%.*]], [[LOOP]] ]
 ; DISABLE-NEXT:    [[SUM:%.*]] = phi i32 [ 0, [[ENTRY]] ], [ [[INCSUM:%.*]], [[LOOP]] ]
-; DISABLE-NEXT:    [[HALF_INV:%.*]] = phi i32 [ [[A]], [[ENTRY]] ], [ [[B]], [[LOOP]] ]
-; DISABLE-NEXT:    [[HALF_INV_2:%.*]] = phi i32 [ [[A]], [[ENTRY]] ], [ [[HALF_INV]], [[LOOP]] ]
-; DISABLE-NEXT:    [[PLUS:%.*]] = phi i32 [ [[A]], [[ENTRY]] ], [ [[HALF_INV_2]], [[LOOP]] ]
+; DISABLE-NEXT:    [[PLUS:%.*]] = phi i32 [ [[A]], [[ENTRY]] ], [ [[HALF_INV:%.*]], [[LOOP]] ]
+; DISABLE-NEXT:    [[HALF_INV]] = phi i32 [ [[A]], [[ENTRY]] ], [ [[B]], [[LOOP]] ]
 ; DISABLE-NEXT:    [[INCSUM]] = add i32 [[SUM]], [[PLUS]]
 ; DISABLE-NEXT:    [[INC]] = add i32 [[I]], 1
 ; DISABLE-NEXT:    [[CMP:%.*]] = icmp slt i32 [[I]], 1000
 ; DISABLE-NEXT:    br i1 [[CMP]], label [[LOOP]], label [[EXIT:%.*]]
 ; DISABLE:       exit:
 ; DISABLE-NEXT:    [[SUM_LCSSA:%.*]] = phi i32 [ [[SUM]], [[LOOP]] ]
 ; DISABLE-NEXT:    ret i32 [[SUM_LCSSA]]
 ;
 entry:
   br label %loop
 
 loop:
   %i = phi i32 [ 0, %entry ], [ %inc, %loop ]
   %sum = phi i32 [ 0, %entry ], [ %incsum, %loop ]
+  %plus = phi i32 [ %a, %entry ], [ %half.inv, %loop ]
   %half.inv = phi i32 [ %a, %entry ], [ %b, %loop ]
-  %half.inv.2 = phi i32 [ %a, %entry ], [ %half.inv, %loop ]
-  %plus = phi i32 [ %a, %entry ], [ %half.inv.2, %loop ]
 
   %incsum = add i32 %sum, %plus
   %inc = add i32 %i, 1
   %cmp = icmp slt i32 %i, 1000
 
   br i1 %cmp, label %loop, label %exit
 
 exit:

 ; CHECK-NEXT:  entry:
 ; CHECK-NEXT:    br label [[LOOP_PEEL_BEGIN:%.*]]
 ; CHECK:       loop.peel.begin:
 ; CHECK-NEXT:    br label [[LOOP_PEEL:%.*]]
 ; CHECK:       loop.peel:
 ; CHECK-NEXT:    [[INCSUM_PEEL:%.*]] = add i32 0, [[A]]
 ; CHECK-NEXT:    [[INC_PEEL:%.*]] = add i32 0, 1
 ; CHECK-NEXT:    [[CMP_PEEL:%.*]] = icmp slt i32 0, 1000
+; CHECK-NEXT:    [[PHIBACKEDGE1_PEEL:%.*]] = add i32 [[A]], 1
+; CHECK-NEXT:    [[PHIBACKEDGE2_PEEL:%.*]] = add i32 [[A]], 1
 ; CHECK-NEXT:    [[INCSUM2_PEEL:%.*]] = add i32 [[INCSUM_PEEL]], [[A]]
 ; CHECK-NEXT:    [[INCSUM3_PEEL:%.*]] = add i32 [[INCSUM_PEEL]], [[A]]
 ; CHECK-NEXT:    [[INCSUM4_PEEL:%.*]] = add i32 [[INCSUM_PEEL]], [[A]]
 ; CHECK-NEXT:    [[INCSUM5_PEEL:%.*]] = add i32 [[INCSUM_PEEL]], [[A]]
-; CHECK-NEXT:    [[INCSUM6_PEEL:%.*]] = add i32 [[INCSUM_PEEL]], [[A]]
-; CHECK-NEXT:    [[INCSUM7_PEEL:%.*]] = add i32 [[INCSUM_PEEL]], [[A]]
 ; CHECK-NEXT:    br i1 [[CMP_PEEL]], label [[LOOP_PEEL_NEXT:%.*]], label [[EXIT:%.*]]
 ; CHECK:       loop.peel.next:
 ; CHECK-NEXT:    br label [[LOOP_PEEL2:%.*]]
 ; CHECK:       loop.peel2:
-; CHECK-NEXT:    [[INCSUM_PEEL3:%.*]] = add i32 [[INCSUM_PEEL]], [[A]]
+; CHECK-NEXT:    [[INCSUM_PEEL3:%.*]] = add i32 [[INCSUM_PEEL]], [[PHIBACKEDGE2_PEEL]]
 ; CHECK-NEXT:    [[INC_PEEL4:%.*]] = add i32 [[INC_PEEL]], 1
 ; CHECK-NEXT:    [[CMP_PEEL5:%.*]] = icmp slt i32 [[INC_PEEL]], 1000
-; CHECK-NEXT:    [[INCSUM2_PEEL6:%.*]] = add i32 [[INCSUM_PEEL3]], [[A]]
-; CHECK-NEXT:    [[INCSUM3_PEEL7:%.*]] = add i32 [[INCSUM_PEEL3]], [[A]]
-; CHECK-NEXT:    [[INCSUM4_PEEL8:%.*]] = add i32 [[INCSUM_PEEL3]], [[A]]
-; CHECK-NEXT:    [[INCSUM5_PEEL9:%.*]] = add i32 [[INCSUM_PEEL3]], [[A]]
-; CHECK-NEXT:    [[INCSUM6_PEEL10:%.*]] = add i32 [[INCSUM_PEEL3]], [[A]]
-; CHECK-NEXT:    [[INCSUM7_PEEL11:%.*]] = add i32 [[INCSUM_PEEL3]], [[A]]
+; CHECK-NEXT:    [[PHIBACKEDGE2_PEEL7:%.*]] = add i32 [[B]], 1
+; CHECK-NEXT:    [[INCSUM2_PEEL8:%.*]] = add i32 [[INCSUM_PEEL3]], [[PHIBACKEDGE2_PEEL]]
+; CHECK-NEXT:    [[INCSUM3_PEEL9:%.*]] = add i32 [[INCSUM_PEEL3]], [[PHIBACKEDGE2_PEEL]]
+; CHECK-NEXT:    [[INCSUM4_PEEL10:%.*]] = add i32 [[INCSUM_PEEL3]], [[PHIBACKEDGE2_PEEL]]
+; CHECK-NEXT:    [[INCSUM5_PEEL11:%.*]] = add i32 [[INCSUM_PEEL3]], [[PHIBACKEDGE2_PEEL]]
 ; CHECK-NEXT:    br i1 [[CMP_PEEL5]], label [[LOOP_PEEL_NEXT1:%.*]], label [[EXIT]]
 ; CHECK:       loop.peel.next1:
 ; CHECK-NEXT:    br label [[LOOP_PEEL_NEXT12:%.*]]
 ; CHECK:       loop.peel.next12:
 ; CHECK-NEXT:    br label [[ENTRY_PEEL_NEWPH:%.*]]
 ; CHECK:       entry.peel.newph:
 ; CHECK-NEXT:    br label [[LOOP:%.*]]
 ; CHECK:       loop:
 ; CHECK-NEXT:    [[I:%.*]] = phi i32 [ [[INC_PEEL4]], [[ENTRY_PEEL_NEWPH]] ], [ [[INC:%.*]], [[LOOP]] ]
 ; CHECK-NEXT:    [[SUM:%.*]] = phi i32 [ [[INCSUM_PEEL3]], [[ENTRY_PEEL_NEWPH]] ], [ [[INCSUM:%.*]], [[LOOP]] ]
-; CHECK-NEXT:    [[PLUS:%.*]] = phi i32 [ [[A]], [[ENTRY_PEEL_NEWPH]] ], [ [[B]], [[LOOP]] ]
+; CHECK-NEXT:    [[PLUS:%.*]] = phi i32 [ [[PHIBACKEDGE2_PEEL7]], [[ENTRY_PEEL_NEWPH]] ], [ [[PHIBACKEDGE2:%.*]], [[LOOP]] ]
 ; CHECK-NEXT:    [[INCSUM]] = add i32 [[SUM]], [[PLUS]]
 ; CHECK-NEXT:    [[INC]] = add nuw nsw i32 [[I]], 1
 ; CHECK-NEXT:    [[CMP:%.*]] = icmp ult i32 [[I]], 1000
+; CHECK-NEXT:    [[PHIBACKEDGE2]] = add i32 [[B]], 1
 ; CHECK-NEXT:    br i1 [[CMP]], label [[LOOP]], label [[EXIT_LOOPEXIT:%.*]], !llvm.loop [[LOOP6:![0-9]+]]
 ; CHECK:       exit.loopexit:
 ; CHECK-NEXT:    [[SUM_LCSSA_PH:%.*]] = phi i32 [ [[SUM]], [[LOOP]] ]
 ; CHECK-NEXT:    br label [[EXIT]]
 ; CHECK:       exit:
 ; CHECK-NEXT:    [[SUM_LCSSA:%.*]] = phi i32 [ 0, [[LOOP_PEEL]] ], [ [[INCSUM_PEEL]], [[LOOP_PEEL2]] ], [ [[SUM_LCSSA_PH]], [[EXIT_LOOPEXIT]] ]
 ; CHECK-NEXT:    ret i32 [[SUM_LCSSA]]
 ;
 ; DISABLE-LABEL: define i32 @invariant_backedge_limited_by_size
 ; DISABLE-SAME: (i32 [[A:%.*]], i32 [[B:%.*]]) {
 ; DISABLE-NEXT:  entry:
 ; DISABLE-NEXT:    br label [[LOOP:%.*]]
 ; DISABLE:       loop:
 ; DISABLE-NEXT:    [[I:%.*]] = phi i32 [ 0, [[ENTRY:%.*]] ], [ [[INC:%.*]], [[LOOP]] ]
 ; DISABLE-NEXT:    [[SUM:%.*]] = phi i32 [ 0, [[ENTRY]] ], [ [[INCSUM:%.*]], [[LOOP]] ]
 ; DISABLE-NEXT:    [[HALF_INV:%.*]] = phi i32 [ [[A]], [[ENTRY]] ], [ [[B]], [[LOOP]] ]
-; DISABLE-NEXT:    [[HALF_INV_2:%.*]] = phi i32 [ [[A]], [[ENTRY]] ], [ [[HALF_INV]], [[LOOP]] ]
-; DISABLE-NEXT:    [[PLUS:%.*]] = phi i32 [ [[A]], [[ENTRY]] ], [ [[HALF_INV_2]], [[LOOP]] ]
+; DISABLE-NEXT:    [[HALF_INV_2:%.*]] = phi i32 [ [[A]], [[ENTRY]] ], [ [[PHIBACKEDGE1:%.*]], [[LOOP]] ]
+; DISABLE-NEXT:    [[PLUS:%.*]] = phi i32 [ [[A]], [[ENTRY]] ], [ [[PHIBACKEDGE2:%.*]], [[LOOP]] ]
 ; DISABLE-NEXT:    [[INCSUM]] = add i32 [[SUM]], [[PLUS]]
 ; DISABLE-NEXT:    [[INC]] = add i32 [[I]], 1
 ; DISABLE-NEXT:    [[CMP:%.*]] = icmp slt i32 [[I]], 1000
+; DISABLE-NEXT:    [[PHIBACKEDGE1]] = add i32 [[HALF_INV]], 1
+; DISABLE-NEXT:    [[PHIBACKEDGE2]] = add i32 [[HALF_INV]], 1
 ; DISABLE-NEXT:    [[INCSUM2:%.*]] = add i32 [[INCSUM]], [[PLUS]]
 ; DISABLE-NEXT:    [[INCSUM3:%.*]] = add i32 [[INCSUM]], [[PLUS]]
 ; DISABLE-NEXT:    [[INCSUM4:%.*]] = add i32 [[INCSUM]], [[PLUS]]
 ; DISABLE-NEXT:    [[INCSUM5:%.*]] = add i32 [[INCSUM]], [[PLUS]]
-; DISABLE-NEXT:    [[INCSUM6:%.*]] = add i32 [[INCSUM]], [[PLUS]]
-; DISABLE-NEXT:    [[INCSUM7:%.*]] = add i32 [[INCSUM]], [[PLUS]]
 ; DISABLE-NEXT:    br i1 [[CMP]], label [[LOOP]], label [[EXIT:%.*]]
 ; DISABLE:       exit:
 ; DISABLE-NEXT:    [[SUM_LCSSA:%.*]] = phi i32 [ [[SUM]], [[LOOP]] ]
 ; DISABLE-NEXT:    ret i32 [[SUM_LCSSA]]
 ;
 entry:
   br label %loop
 
 loop:
   %i = phi i32 [ 0, %entry ], [ %inc, %loop ]
   %sum = phi i32 [ 0, %entry ], [ %incsum, %loop ]
   %half.inv = phi i32 [ %a, %entry ], [ %b, %loop ]
-  %half.inv.2 = phi i32 [ %a, %entry ], [ %half.inv, %loop ]
-  %plus = phi i32 [ %a, %entry ], [ %half.inv.2, %loop ]
+  %half.inv.2 = phi i32 [ %a, %entry ], [ %phibackedge1, %loop ]
+  %plus = phi i32 [ %a, %entry ], [ %phibackedge2, %loop ]
 
   %incsum = add i32 %sum, %plus
   %inc = add i32 %i, 1
   %cmp = icmp slt i32 %i, 1000
 
+  %phibackedge1 = add i32 %half.inv, 1
+  %phibackedge2 = add i32 %half.inv, 1
+
   %incsum2 = add i32 %incsum, %plus
   %incsum3 = add i32 %incsum, %plus
   %incsum4 = add i32 %incsum, %plus
   %incsum5 = add i32 %incsum, %plus
-  %incsum6 = add i32 %incsum, %plus
-  %incsum7 = add i32 %incsum, %plus
 
   br i1 %cmp, label %loop, label %exit
 
 exit:
   ret i32 %sum
 }
 
 ; Peeling should fail due to method size.

llvm/test/Transforms/LoopUnroll/peel-loop-phi-analysis.ll

 ; CHECK:       for.body.peel2:
 ; CHECK-NEXT:    tail call void @_Z1gi(i32 noundef signext [[CONV_PEEL]])
 ; CHECK-NEXT:    [[CONV_PEEL3:%.*]] = fptosi float 0.000000e+00 to i32
 ; CHECK-NEXT:    tail call void @_Z1ff(float noundef 0.000000e+00)
 ; CHECK-NEXT:    [[INC_PEEL4:%.*]] = add nuw nsw i32 [[INC_PEEL]], 1
 ; CHECK-NEXT:    [[EXITCOND_PEEL5:%.*]] = icmp ne i32 [[INC_PEEL4]], 100000
 ; CHECK-NEXT:    br i1 [[EXITCOND_PEEL5]], label [[FOR_BODY_PEEL_NEXT1:%.*]], label [[FOR_COND_CLEANUP]]
 ; CHECK:       for.body.peel.next1:
-; CHECK-NEXT:    br label [[FOR_BODY_PEEL7:%.*]]
-; CHECK:       for.body.peel7:
-; CHECK-NEXT:    tail call void @_Z1gi(i32 noundef signext [[CONV_PEEL3]])
-; CHECK-NEXT:    [[CONV_PEEL8:%.*]] = fptosi float 5.000000e+00 to i32
-; CHECK-NEXT:    tail call void @_Z1ff(float noundef 5.000000e+00)
-; CHECK-NEXT:    [[INC_PEEL9:%.*]] = add nuw nsw i32 [[INC_PEEL4]], 1
-; CHECK-NEXT:    [[EXITCOND_PEEL10:%.*]] = icmp ne i32 [[INC_PEEL9]], 100000
-; CHECK-NEXT:    br i1 [[EXITCOND_PEEL10]], label [[FOR_BODY_PEEL_NEXT6:%.*]], label [[FOR_COND_CLEANUP]]
+; CHECK-NEXT:    br label [[FOR_BODY_PEEL_NEXT6:%.*]]
 ; CHECK:       for.body.peel.next6:
-; CHECK-NEXT:    br label [[FOR_BODY_PEEL_NEXT11:%.*]]
-; CHECK:       for.body.peel.next11:
 ; CHECK-NEXT:    br label [[ENTRY_PEEL_NEWPH:%.*]]
 ; CHECK:       entry.peel.newph:
 ; CHECK-NEXT:    br label [[FOR_BODY:%.*]]
 ; CHECK:       for.cond.cleanup.loopexit:
 ; CHECK-NEXT:    br label [[FOR_COND_CLEANUP]]
 ; CHECK:       for.cond.cleanup:
 ; CHECK-NEXT:    ret void
 ; CHECK:       for.body:
-; CHECK-NEXT:    [[I:%.*]] = phi i32 [ [[INC_PEEL9]], [[ENTRY_PEEL_NEWPH]] ], [ [[INC:%.*]], [[FOR_BODY]] ]
-; CHECK-NEXT:    [[X:%.*]] = phi i32 [ [[CONV_PEEL8]], [[ENTRY_PEEL_NEWPH]] ], [ 5, [[FOR_BODY]] ]
+; CHECK-NEXT:    [[I:%.*]] = phi i32 [ [[INC_PEEL4]], [[ENTRY_PEEL_NEWPH]] ], [ [[INC:%.*]], [[FOR_BODY]] ]
+; CHECK-NEXT:    [[X:%.*]] = phi i32 [ [[CONV_PEEL3]], [[ENTRY_PEEL_NEWPH]] ], [ 5, [[FOR_BODY]] ]
 ; CHECK-NEXT:    tail call void @_Z1gi(i32 noundef signext [[X]])
 ; CHECK-NEXT:    tail call void @_Z1ff(float noundef 5.000000e+00)
 ; CHECK-NEXT:    [[INC]] = add nuw nsw i32 [[I]], 1
 ; CHECK-NEXT:    [[EXITCOND:%.*]] = icmp ne i32 [[INC]], 100000
 ; CHECK-NEXT:    br i1 [[EXITCOND]], label [[FOR_BODY]], label [[FOR_COND_CLEANUP_LOOPEXIT:%.*]], !llvm.loop [[LOOP0:![0-9]+]]
 ;
 entry:
   br label %for.body

 ; CHECK-NEXT:    [[INC_PEEL:%.*]] = add nuw nsw i32 0, 1
 ; CHECK-NEXT:    [[EXITCOND_PEEL:%.*]] = icmp eq i32 [[INC_PEEL]], 100000
 ; CHECK-NEXT:    br i1 [[EXITCOND_PEEL]], label [[FOR_COND_CLEANUP:%.*]], label [[FOR_BODY_PEEL_NEXT:%.*]]
 ; CHECK:       for.body.peel.next:
 ; CHECK-NEXT:    br label [[FOR_BODY_PEEL2:%.*]]
 ; CHECK:       for.body.peel2:
 ; CHECK-NEXT:    tail call void @_Z1gi(i32 signext 0)
 ; CHECK-NEXT:    tail call void @_Z1gi(i32 signext 5)
+; CHECK-NEXT:    [[ADD_PEEL3:%.*]] = add nuw nsw i32 5, 1
 ; CHECK-NEXT:    [[INC_PEEL4:%.*]] = add nuw nsw i32 [[INC_PEEL]], 1
 ; CHECK-NEXT:    [[EXITCOND_PEEL5:%.*]] = icmp eq i32 [[INC_PEEL4]], 100000
 ; CHECK-NEXT:    br i1 [[EXITCOND_PEEL5]], label [[FOR_COND_CLEANUP]], label [[FOR_BODY_PEEL_NEXT1:%.*]]
 ; CHECK:       for.body.peel.next1:
-; CHECK-NEXT:    br label [[FOR_BODY_PEEL7:%.*]]
-; CHECK:       for.body.peel7:
-; CHECK-NEXT:    tail call void @_Z1gi(i32 signext [[ADD_PEEL]])
-; CHECK-NEXT:    tail call void @_Z1gi(i32 signext 5)
-; CHECK-NEXT:    [[INC_PEEL9:%.*]] = add nuw nsw i32 [[INC_PEEL4]], 1
-; CHECK-NEXT:    [[EXITCOND_PEEL10:%.*]] = icmp eq i32 [[INC_PEEL9]], 100000
-; CHECK-NEXT:    br i1 [[EXITCOND_PEEL10]], label [[FOR_COND_CLEANUP]], label [[FOR_BODY_PEEL_NEXT6:%.*]]
+; CHECK-NEXT:    br label [[FOR_BODY_PEEL_NEXT6:%.*]]
 ; CHECK:       for.body.peel.next6:
-; CHECK-NEXT:    br label [[FOR_BODY_PEEL_NEXT11:%.*]]
-; CHECK:       for.body.peel.next11:
 ; CHECK-NEXT:    br label [[ENTRY_PEEL_NEWPH:%.*]]
 ; CHECK:       entry.peel.newph:
 ; CHECK-NEXT:    br label [[FOR_BODY:%.*]]
 ; CHECK:       for.cond.cleanup.loopexit:
 ; CHECK-NEXT:    br label [[FOR_COND_CLEANUP]]
 ; CHECK:       for.cond.cleanup:
 ; CHECK-NEXT:    ret void
 ; CHECK:       for.body:
-; CHECK-NEXT:    [[I:%.*]] = phi i32 [ [[INC_PEEL9]], [[ENTRY_PEEL_NEWPH]] ], [ [[INC:%.*]], [[FOR_BODY]] ]
-; CHECK-NEXT:    tail call void @_Z1gi(i32 signext 6)
+; CHECK-NEXT:    [[I:%.*]] = phi i32 [ [[INC_PEEL4]], [[ENTRY_PEEL_NEWPH]] ], [ [[INC:%.*]], [[FOR_BODY]] ]
+; CHECK-NEXT:    [[Y:%.*]] = phi i32 [ [[ADD_PEEL3]], [[ENTRY_PEEL_NEWPH]] ], [ 6, [[FOR_BODY]] ]
+; CHECK-NEXT:    [[X:%.*]] = phi i32 [ [[ADD_PEEL]], [[ENTRY_PEEL_NEWPH]] ], [ [[Y]], [[FOR_BODY]] ]
+; CHECK-NEXT:    tail call void @_Z1gi(i32 signext [[X]])
 ; CHECK-NEXT:    tail call void @_Z1gi(i32 signext 5)
 ; CHECK-NEXT:    [[INC]] = add nuw nsw i32 [[I]], 1
 ; CHECK-NEXT:    [[EXITCOND:%.*]] = icmp eq i32 [[INC]], 100000
 ; CHECK-NEXT:    br i1 [[EXITCOND]], label [[FOR_COND_CLEANUP_LOOPEXIT:%.*]], label [[FOR_BODY]], !llvm.loop [[LOOP2:![0-9]+]]
 ;
 entry:
   br label %for.body