[MemorySSA] When applying updates, clean unnecessary Phis.

Summary: After applying a set of insert updates, there may be trivial Phis left over. Clean them up.

Reviewers: george.burgess.iv

Subscribers: jlebar, Prazek, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D63033

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@363094 91177308-0d34-0410-b5e6-96231b3b80d8

lib/Analysis/MemorySSAUpdater.cpp

@@ -860,13 +860,14 @@ void MemorySSAUpdater::applyInsertUpdates(ArrayRef<CFGUpdate> Updates,
 
   SmallVector<BasicBlock *, 8> BlocksToProcess;
   SmallVector<BasicBlock *, 16> BlocksWithDefsToReplace;
+  SmallVector<WeakVH, 8> InsertedPhis;
 
   // First create MemoryPhis in all blocks that don't have one. Create in the
   // order found in Updates, not in PredMap, to get deterministic numbering.
   for (auto &Edge : Updates) {
     BasicBlock *BB = Edge.getTo();
     if (PredMap.count(BB) && !MSSA->getMemoryAccess(BB))
-      MSSA->createMemoryPhi(BB);
+      InsertedPhis.push_back(MSSA->createMemoryPhi(BB));
   }
 
   // Now we'll fill in the MemoryPhis with the right incoming values.
@@ -967,6 +968,7 @@ void MemorySSAUpdater::applyInsertUpdates(ArrayRef<CFGUpdate> Updates,
           IDFPhi->setIncomingValue(I, GetLastDef(IDFPhi->getIncomingBlock(I)));
       } else {
         IDFPhi = MSSA->createMemoryPhi(BBIDF);
+        InsertedPhis.push_back(IDFPhi);
         for (auto &Pair : children<GraphDiffInvBBPair>({GD, BBIDF})) {
           BasicBlock *Pi = Pair.second;
           IDFPhi->addIncoming(GetLastDef(Pi), Pi);
@@ -1009,6 +1011,7 @@ void MemorySSAUpdater::applyInsertUpdates(ArrayRef<CFGUpdate> Updates,
       }
     }
   }
+  tryRemoveTrivialPhis(InsertedPhis);
 }
 
 // Move What before Where in the MemorySSA IR.

test/Analysis/MemorySSA/loop_rotate_remove_trivial_phis.ll

@@ -0,0 +1,78 @@
+; RUN: opt -loop-rotate -print-memoryssa -disable-output -enable-mssa-loop-dependency -verify-memoryssa %s 2>&1 |  FileCheck %s
+; REQUIRES: asserts
+
+target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
+target triple = "x86_64-grtev4-linux-gnu"
+
+declare double @sqrt(double)
+
+; CHECK-LABEL: @f
+define internal fastcc double @f(i32* %n_, double* %dx) align 32 {
+entry:
+; CHECK: entry:
+; CHECK: MemoryUse(liveOnEntry)
+; CHECK-NOT: 7 = MemoryPhi
+; CHECK-NOT: 6 = MemoryPhi
+  %v0 = load i32, i32* %n_, align 4
+  br label %for.cond
+
+for.cond:                                         ; preds = %for.body, %entry
+  %xmax.0 = phi double [ undef, %entry ], [ %xmax.1, %for.body ]
+  %i.0 = phi i32 [ 0, %entry ], [ %inc, %for.body ]
+  %cmp = icmp slt i32 %i.0, %v0
+  br i1 %cmp, label %for.body, label %for.end
+
+for.body:                                         ; preds = %for.cond
+  %idxprom = zext i32 %i.0 to i64
+  %arrayidx = getelementptr inbounds double, double* %dx, i64 %idxprom
+  %v1 = load double, double* %arrayidx, align 8
+  %cmp1 = fcmp ueq double %v1, 0.000000e+00
+  %xmax.1 = select i1 %cmp1, double %xmax.0, double %v1
+  %inc = add nuw nsw i32 %i.0, 1
+  br label %for.cond
+
+for.end:                                          ; preds = %for.cond
+  %xmax.0.lcssa = phi double [ %xmax.0, %for.cond ]
+  %cmp2 = fcmp oeq double %xmax.0.lcssa, 0.000000e+00
+  br i1 %cmp2, label %cleanup, label %if.end4
+
+if.end4:                                          ; preds = %for.end
+  %div = fdiv double 1.000000e+00, %xmax.0.lcssa
+  %cmp61 = icmp slt i32 0, %v0
+  br i1 %cmp61, label %for.body7.lr.ph, label %for.end15
+
+for.body7.lr.ph:                                  ; preds = %if.end4
+  br label %for.body7
+
+; CHECK: for.body7:
+; CHECK: 3 = MemoryPhi({for.body7.lr.ph,liveOnEntry},{for.body7,1})
+for.body7:                                        ; preds = %for.body7.lr.ph, %for.body7
+  %i.13 = phi i32 [ 0, %for.body7.lr.ph ], [ %inc14, %for.body7 ]
+  %sum.02 = phi x86_fp80 [ undef, %for.body7.lr.ph ], [ %add, %for.body7 ]
+  %idxprom9 = zext i32 %i.13 to i64
+  %arrayidx10 = getelementptr inbounds double, double* %dx, i64 %idxprom9
+  %v3 = load double, double* %arrayidx10, align 8
+  %mul11 = fmul double %div, %v3
+  %v2 = call double @sqrt(double %v3)
+  %mul12 = fmul double %mul11, %v2
+  %conv = fpext double %mul12 to x86_fp80
+  %add = fadd x86_fp80 %sum.02, %conv
+  %inc14 = add nuw nsw i32 %i.13, 1
+  %cmp6 = icmp slt i32 %inc14, %v0
+  br i1 %cmp6, label %for.body7, label %for.cond5.for.end15_crit_edge
+
+for.cond5.for.end15_crit_edge:                    ; preds = %for.body7
+  %split = phi x86_fp80 [ %add, %for.body7 ]
+  br label %for.end15
+
+for.end15:                                        ; preds = %for.cond5.for.end15_crit_edge, %if.end4
+  %sum.0.lcssa = phi x86_fp80 [ %split, %for.cond5.for.end15_crit_edge ], [ undef, %if.end4 ]
+  %conv16 = fptrunc x86_fp80 %sum.0.lcssa to double
+  %call = call double @sqrt(double %conv16)
+  %mul17 = fmul double %call, 0.000000e+00
+  br label %cleanup
+
+cleanup:                                          ; preds = %for.end15, %for.end
+  %retval.0 = phi double [ undef, %for.end ], [ %mul17, %for.end15 ]
+  ret double %retval.0
+}