Invoke SimplifyIndVar when we partially unroll a loop. Fixes PR10534.

llvm-svn: 137203

llvm/lib/Transforms/Scalar/LoopUnrollPass.cpp

@@ -79,6 +79,7 @@ namespace {
       AU.addPreservedID(LoopSimplifyID);
       AU.addRequiredID(LCSSAID);
       AU.addPreservedID(LCSSAID);
+      AU.addRequired<ScalarEvolution>();
       AU.addPreserved<ScalarEvolution>();
       // FIXME: Loop unroll requires LCSSA. And LCSSA requires dom info.
       // If loop unroll does not preserve dom info then LCSSA pass on next
@@ -187,12 +188,8 @@ bool LoopUnroll::runOnLoop(Loop *L, LPPassManager &LPM) {
   }
 
   // Unroll the loop.
-  Function *F = L->getHeader()->getParent();
   if (!UnrollLoop(L, Count, TripCount, TripMultiple, LI, &LPM))
     return false;
 
-  // FIXME: Reconstruct dom info, because it is not preserved properly.
-  if (DominatorTree *DT = getAnalysisIfAvailable<DominatorTree>())
-    DT->runOnFunction(*F);
   return true;
 }

llvm/lib/Transforms/Utils/LoopUnroll.cpp

@@ -29,6 +29,7 @@
 #include "llvm/Transforms/Utils/BasicBlockUtils.h"
 #include "llvm/Transforms/Utils/Cloning.h"
 #include "llvm/Transforms/Utils/Local.h"
+#include "llvm/Transforms/Utils/SimplifyIndVar.h"
 using namespace llvm;
 
 // TODO: Should these be here or in LoopUnroll?
@@ -130,6 +131,9 @@ static BasicBlock *FoldBlockIntoPredecessor(BasicBlock *BB, LoopInfo* LI,
 ///
 /// If a LoopPassManager is passed in, and the loop is fully removed, it will be
 /// removed from the LoopPassManager as well. LPM can also be NULL.
+///
+/// This utility preserves LoopInfo. If DominatorTree or ScalarEvolution are
+/// available it must also preseve those analyses.
 bool llvm::UnrollLoop(Loop *L, unsigned Count, unsigned TripCount,
                       unsigned TripMultiple, LoopInfo *LI, LPPassManager *LPM) {
   BasicBlock *Preheader = L->getLoopPreheader();
@@ -163,7 +167,8 @@ bool llvm::UnrollLoop(Loop *L, unsigned Count, unsigned TripCount,
 
   // Notify ScalarEvolution that the loop will be substantially changed,
   // if not outright eliminated.
-  if (ScalarEvolution *SE = LPM->getAnalysisIfAvailable<ScalarEvolution>())
+  ScalarEvolution *SE = LPM->getAnalysisIfAvailable<ScalarEvolution>();
+  if (SE)
     SE->forgetLoop(L);
 
   if (TripCount != 0)
@@ -374,6 +379,24 @@ bool llvm::UnrollLoop(Loop *L, unsigned Count, unsigned TripCount,
     }
   }
 
+  // FIXME: Reconstruct dom info, because it is not preserved properly.
+  // Incrementally updating domtree after loop unrolling woud be easy.
+  if (DominatorTree *DT = LPM->getAnalysisIfAvailable<DominatorTree>())
+    DT->runOnFunction(*L->getHeader()->getParent());
+
+  // Simplify any new induction variables in the partially unrolled loop.
+  if (SE && !CompletelyUnroll) {
+    SmallVector<WeakVH, 16> DeadInsts;
+    simplifyLoopIVs(L, SE, LPM, DeadInsts);
+
+    // Aggressively clean up dead instructions that simplifyLoopIVs already
+    // identified. Any remaining should be cleaned up below.
+    while (!DeadInsts.empty())
+      if (Instruction *Inst =
+          dyn_cast_or_null<Instruction>(&*DeadInsts.pop_back_val()))
+        RecursivelyDeleteTriviallyDeadInstructions(Inst);
+  }
+
   // At this point, the code is well formed.  We now do a quick sweep over the
   // inserted code, doing constant propagation and dead code elimination as we
   // go.

llvm/test/Transforms/LoopUnroll/2011-08-09-IVSimplify.ll

@@ -0,0 +1,39 @@
+; RUN: opt -S < %s -loop-unroll -unroll-count=4 -disable-iv-rewrite | FileCheck %s
+;
+; Test induction variable simplify after loop unrolling. It should
+; expose nice opportunities for GVN.
+
+target datalayout = "e-p:32:32:32-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-f80:128:128-v64:64:64-v128:128:128-a0:0:64-f80:32:32-n8:16:32"
+
+; PR10534: LoopUnroll not keeping canonical induction variable...
+; CHECK: while.body.1:
+; CHECK: %shr.1 = lshr i32 %bit_addr.addr.01, 5
+; CHECK: %arrayidx.1 = getelementptr inbounds i32* %bitmap, i32 %shr.1
+; CHECK: while.body.2:
+; CHECK: %shr.2 = lshr i32 %bit_addr.addr.01, 5
+; CHECK: %arrayidx.2 = getelementptr inbounds i32* %bitmap, i32 %shr.2
+; CHECK: while.body.3:
+; CHECK: %shr.3 = lshr i32 %bit_addr.addr.01, 5
+; CHECK: %arrayidx.3 = getelementptr inbounds i32* %bitmap, i32 %shr.3
+define void @FlipBit(i32* nocapture %bitmap, i32 %bit_addr, i32 %nbits) nounwind {
+entry:
+  br label %while.body
+
+while.body:
+  %nbits.addr.02 = phi i32 [ 128, %entry ], [ %dec, %while.body ]
+  %bit_addr.addr.01 = phi i32 [ 0, %entry ], [ %inc, %while.body ]
+  %dec = add i32 %nbits.addr.02, -1
+  %shr = lshr i32 %bit_addr.addr.01, 5
+  %rem = and i32 %bit_addr.addr.01, 31
+  %shl = shl i32 1, %rem
+  %arrayidx = getelementptr inbounds i32* %bitmap, i32 %shr
+  %tmp6 = load i32* %arrayidx, align 4
+  %xor = xor i32 %tmp6, %shl
+  store i32 %xor, i32* %arrayidx, align 4
+  %inc = add i32 %bit_addr.addr.01, 1
+  %tobool = icmp eq i32 %dec, 0
+  br i1 %tobool, label %while.end, label %while.body
+
+while.end:
+  ret void
+}