[SimplifyCFG] Simplify switch instruction that has duplicate arms

llvm/lib/Transforms/Utils/SimplifyCFG.cpp

@@ -276,6 +276,7 @@ class SimplifyCFGOpt {
   bool simplifyCleanupReturn(CleanupReturnInst *RI);
   bool simplifyUnreachable(UnreachableInst *UI);
   bool simplifySwitch(SwitchInst *SI, IRBuilder<> &Builder);
+  bool simplifyDuplicateSwitchArms(SwitchInst *SI);
   bool simplifyIndirectBr(IndirectBrInst *IBI);
   bool simplifyBranch(BranchInst *Branch, IRBuilder<> &Builder);
   bool simplifyUncondBranch(BranchInst *BI, IRBuilder<> &Builder);
@@ -7436,6 +7437,94 @@ static bool simplifySwitchOfCmpIntrinsic(SwitchInst *SI, IRBuilderBase &Builder,
   return true;
 }
 
+bool SimplifyCFGOpt::simplifyDuplicateSwitchArms(SwitchInst *SI) {
+  // Simplify the case where multiple arms contain only a terminator, the
+  // terminators are the same, and their sucessor PHIS incoming values are the
+  // same.
+
+  // Find BBs that are candidates for simplification.
+  SmallPtrSet<BasicBlock *, 8> BBs;
+  for (auto &Case : SI->cases()) {
+    BasicBlock *BB = Case.getCaseSuccessor();
+
+    // FIXME: This case needs some extra care because the terminators other than
+    // SI need to be updated.
+    if (!BB->hasNPredecessors(1))
+      continue;
+
+    // FIXME: Relax that the terminator is a BranchInst by checking for equality
+    // on other kinds of terminators.
+    Instruction *T = BB->getTerminator();
+    if (T && BB->size() == 1 && isa<BranchInst>(T))
+      BBs.insert(BB);
+  }
+
+  auto IsBranchEq = [](BranchInst *A, BranchInst *B) {
+    if (A->isConditional() != B->isConditional())
+      return false;
+
+    if (A->isConditional() && A->getCondition() != B->getCondition())
+      return false;
+
+    if (A->getNumSuccessors() != B->getNumSuccessors())
+      return false;
+
+    for (unsigned I = 0; I < A->getNumSuccessors(); ++I)
+      if (A->getSuccessor(I) != B->getSuccessor(I))
+        return false;
+
+    // Need to check that PHIs in sucessors have matching values
+    for (auto *Succ : A->successors()) {
+      for (PHINode &Phi : Succ->phis())
+        if (Phi.getIncomingValueForBlock(A->getParent()) !=
+            Phi.getIncomingValueForBlock(B->getParent()))
+          return false;
+    }
+
+    return true;
+  };
+
+  // Construct a map from candidate basic block to an equivalent basic block
+  // to replace it with. All equivalent basic blocks should be replaced with
+  // the same basic block. To do this, if there is no equivalent BB in the map,
+  // then insert into the map BB -> BB. Otherwise, we should check only elements
+  // in the map for equivalence to ensure that all equivalent BB get replaced
+  // by the BB in the map. Replacing BB with BB has no impact, so we skip
+  // a call to setSuccessor when we do the actual replacement.
+  DenseMap<BasicBlock *, BasicBlock *> ReplaceWith;
+  for (BasicBlock *BB : BBs) {
+    bool Inserted = false;
+    for (auto KV : ReplaceWith) {
+      if (IsBranchEq(cast<BranchInst>(BB->getTerminator()),
+                     cast<BranchInst>(KV.first->getTerminator()))) {
+        ReplaceWith[BB] = KV.first;
+        Inserted = true;
+        break;
+      }
+    }
+    if (!Inserted)
+      ReplaceWith[BB] = BB;
+  }
+
+  // Do the replacement in SI.
+  bool MadeChange = false;
+  // There is no fast lookup of BasicBlock -> Cases, so we iterate over cases
+  // and check that the case was a candidate. BBs is already filtered, so
+  // hopefully calling contains on it is not too expensive.
+  for (auto &Case : SI->cases()) {
+    BasicBlock *OldSucc = Case.getCaseSuccessor();
+    if (!BBs.contains(OldSucc))
+      continue;
+    BasicBlock *NewSucc = ReplaceWith[OldSucc];
+    if (OldSucc != NewSucc) {
+      Case.setSuccessor(NewSucc);
+      MadeChange = true;
+    }
+  }
+
+  return MadeChange;
+}
+
 bool SimplifyCFGOpt::simplifySwitch(SwitchInst *SI, IRBuilder<> &Builder) {
   BasicBlock *BB = SI->getParent();
 
@@ -7496,6 +7585,9 @@ bool SimplifyCFGOpt::simplifySwitch(SwitchInst *SI, IRBuilder<> &Builder) {
       hoistCommonCodeFromSuccessors(SI, !Options.HoistCommonInsts))
     return requestResimplify();
 
+  if (simplifyDuplicateSwitchArms(SI))
+    return requestResimplify();
+
   return false;
 }
 

llvm/test/Transforms/SimplifyCFG/ForwardSwitchConditionToPHI.ll

@@ -139,16 +139,14 @@ define i32 @PR34471(i32 %x) {
 ; NO_FWD-NEXT:    switch i32 [[X:%.*]], label [[ELSE3:%.*]] [
 ; NO_FWD-NEXT:      i32 17, label [[RETURN:%.*]]
 ; NO_FWD-NEXT:      i32 19, label [[IF19:%.*]]
-; NO_FWD-NEXT:      i32 42, label [[IF42:%.*]]
+; NO_FWD-NEXT:      i32 42, label [[IF19]]
 ; NO_FWD-NEXT:    ]
 ; NO_FWD:       if19:
 ; NO_FWD-NEXT:    br label [[RETURN]]
-; NO_FWD:       if42:
-; NO_FWD-NEXT:    br label [[RETURN]]
 ; NO_FWD:       else3:
 ; NO_FWD-NEXT:    br label [[RETURN]]
 ; NO_FWD:       return:
-; NO_FWD-NEXT:    [[R:%.*]] = phi i32 [ [[X]], [[IF19]] ], [ [[X]], [[IF42]] ], [ 0, [[ELSE3]] ], [ 17, [[ENTRY:%.*]] ]
+; NO_FWD-NEXT:    [[R:%.*]] = phi i32 [ [[X]], [[IF19]] ], [ 0, [[ELSE3]] ], [ 17, [[ENTRY:%.*]] ]
 ; NO_FWD-NEXT:    ret i32 [[R]]
 ;
 ; FWD-LABEL: @PR34471(

llvm/test/Transforms/SimplifyCFG/HoistCode.ll

@@ -65,14 +65,12 @@ define float @PR39535min_switch(i64 %i, float %x) {
 ; CHECK-NEXT:  entry:
 ; CHECK-NEXT:    switch i64 [[I:%.*]], label [[END:%.*]] [
 ; CHECK-NEXT:      i64 1, label [[BB1:%.*]]
-; CHECK-NEXT:      i64 2, label [[BB2:%.*]]
+; CHECK-NEXT:      i64 2, label [[BB1]]
 ; CHECK-NEXT:    ]
 ; CHECK:       bb1:
 ; CHECK-NEXT:    br label [[END]]
-; CHECK:       bb2:
-; CHECK-NEXT:    br label [[END]]
 ; CHECK:       end:
-; CHECK-NEXT:    [[COND:%.*]] = phi fast float [ [[X:%.*]], [[BB1]] ], [ [[X]], [[BB2]] ], [ 0.000000e+00, [[ENTRY:%.*]] ]
+; CHECK-NEXT:    [[COND:%.*]] = phi fast float [ [[X:%.*]], [[BB1]] ], [ 0.000000e+00, [[ENTRY:%.*]] ]
 ; CHECK-NEXT:    ret float [[COND]]
 ;
 entry:

llvm/test/Transforms/SimplifyCFG/switch-dup-bbs.ll

@@ -0,0 +1,140 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 5
+; RUN: opt < %s -passes=simplifycfg -S | FileCheck %s -check-prefix=SIMPLIFY-CFG
+; RUN: opt < %s -O3 -S | FileCheck %s -check-prefix=O3
+
+define i32 @switch_all_duplicate_arms(i32 %0, i32 %1, i32 %2, i32 %3) {
+; SIMPLIFY-CFG-LABEL: define i32 @switch_all_duplicate_arms(
+; SIMPLIFY-CFG-SAME: i32 [[TMP0:%.*]], i32 [[TMP1:%.*]], i32 [[TMP2:%.*]], i32 [[TMP3:%.*]]) {
+; SIMPLIFY-CFG-NEXT:    switch i32 [[TMP1]], label %[[BB6:.*]] [
+; SIMPLIFY-CFG-NEXT:      i32 0, label %[[BB5:.*]]
+; SIMPLIFY-CFG-NEXT:      i32 1, label %[[BB5]]
+; SIMPLIFY-CFG-NEXT:    ]
+; SIMPLIFY-CFG:       [[BB5]]:
+; SIMPLIFY-CFG-NEXT:    br label %[[BB6]]
+; SIMPLIFY-CFG:       [[BB6]]:
+; SIMPLIFY-CFG-NEXT:    [[TMP8:%.*]] = phi i32 [ [[TMP3]], [[TMP4:%.*]] ], [ [[TMP2]], %[[BB5]] ]
+; SIMPLIFY-CFG-NEXT:    ret i32 [[TMP8]]
+;
+; O3-LABEL: define i32 @switch_all_duplicate_arms(
+; O3-SAME: i32 [[TMP0:%.*]], i32 [[TMP1:%.*]], i32 [[TMP2:%.*]], i32 [[TMP3:%.*]]) local_unnamed_addr #[[ATTR0:[0-9]+]] {
+; O3-NEXT:    [[SWITCH:%.*]] = icmp ult i32 [[TMP1]], 2
+; O3-NEXT:    [[TMP8:%.*]] = select i1 [[SWITCH]], i32 [[TMP2]], i32 [[TMP3]]
+; O3-NEXT:    ret i32 [[TMP8]]
+;
+  switch i32 %1, label %7 [
+  i32 0, label %5
+  i32 1, label %6
+  ]
+
+5:
+  br label %7
+
+6:
+  br label %7
+
+7:
+  %8 = phi i32 [ %3, %4 ], [ %2, %6 ], [ %2, %5 ]
+  ret i32 %8
+}
+
+define i32 @switch_some_duplicate_arms(i32 %0, i32 %1, i32 %2, i32 %3, i32 %4) {
+; SIMPLIFY-CFG-LABEL: define i32 @switch_some_duplicate_arms(
+; SIMPLIFY-CFG-SAME: i32 [[TMP0:%.*]], i32 [[TMP1:%.*]], i32 [[TMP2:%.*]], i32 [[TMP3:%.*]], i32 [[TMP4:%.*]]) {
+; SIMPLIFY-CFG-NEXT:    switch i32 [[TMP1]], label %[[BB8:.*]] [
+; SIMPLIFY-CFG-NEXT:      i32 0, label %[[BB6:.*]]
+; SIMPLIFY-CFG-NEXT:      i32 1, label %[[BB6]]
+; SIMPLIFY-CFG-NEXT:      i32 2, label %[[BB7:.*]]
+; SIMPLIFY-CFG-NEXT:    ]
+; SIMPLIFY-CFG:       [[BB6]]:
+; SIMPLIFY-CFG-NEXT:    br label %[[BB8]]
+; SIMPLIFY-CFG:       [[BB7]]:
+; SIMPLIFY-CFG-NEXT:    br label %[[BB8]]
+; SIMPLIFY-CFG:       [[BB8]]:
+; SIMPLIFY-CFG-NEXT:    [[TMP10:%.*]] = phi i32 [ [[TMP3]], [[TMP5:%.*]] ], [ [[TMP4]], %[[BB7]] ], [ [[TMP2]], %[[BB6]] ]
+; SIMPLIFY-CFG-NEXT:    ret i32 [[TMP10]]
+;
+; O3-LABEL: define i32 @switch_some_duplicate_arms(
+; O3-SAME: i32 [[TMP0:%.*]], i32 [[TMP1:%.*]], i32 [[TMP2:%.*]], i32 [[TMP3:%.*]], i32 [[TMP4:%.*]]) local_unnamed_addr #[[ATTR0]] {
+; O3-NEXT:    switch i32 [[TMP1]], label %[[BB8:.*]] [
+; O3-NEXT:      i32 0, label %[[BB6:.*]]
+; O3-NEXT:      i32 1, label %[[BB6]]
+; O3-NEXT:      i32 2, label %[[BB7:.*]]
+; O3-NEXT:    ]
+; O3:       [[BB6]]:
+; O3-NEXT:    br label %[[BB8]]
+; O3:       [[BB7]]:
+; O3-NEXT:    br label %[[BB8]]
+; O3:       [[BB8]]:
+; O3-NEXT:    [[TMP10:%.*]] = phi i32 [ [[TMP3]], [[TMP5:%.*]] ], [ [[TMP4]], %[[BB7]] ], [ [[TMP2]], %[[BB6]] ]
+; O3-NEXT:    ret i32 [[TMP10]]
+;
+  switch i32 %1, label %9 [
+  i32 0, label %6
+  i32 1, label %7
+  i32 2, label %8
+  ]
+
+6:
+  br label %9
+
+7:
+  br label %9
+
+8:
+  br label %9
+
+9:
+  %10 = phi i32 [ %3, %5 ], [ %4, %8 ], [ %2, %7 ], [ %2, %6 ]
+  ret i32 %10
+}
+
+define i32 @switch_duplicate_arms_multipred(i1 %0, i32 %1, i32 %2, i32 %3, i32 %4) {
+; SIMPLIFY-CFG-LABEL: define i32 @switch_duplicate_arms_multipred(
+; SIMPLIFY-CFG-SAME: i1 [[TMP0:%.*]], i32 [[TMP1:%.*]], i32 [[TMP2:%.*]], i32 [[TMP3:%.*]], i32 [[TMP4:%.*]]) {
+; SIMPLIFY-CFG-NEXT:    br i1 [[TMP0]], label %[[BB6:.*]], label %[[BB7:.*]]
+; SIMPLIFY-CFG:       [[BB6]]:
+; SIMPLIFY-CFG-NEXT:    switch i32 [[TMP2]], label %[[BB9:.*]] [
+; SIMPLIFY-CFG-NEXT:      i32 0, label %[[BB7]]
+; SIMPLIFY-CFG-NEXT:      i32 1, label %[[BB8:.*]]
+; SIMPLIFY-CFG-NEXT:    ]
+; SIMPLIFY-CFG:       [[BB7]]:
+; SIMPLIFY-CFG-NEXT:    br label %[[BB9]]
+; SIMPLIFY-CFG:       [[BB8]]:
+; SIMPLIFY-CFG-NEXT:    br label %[[BB9]]
+; SIMPLIFY-CFG:       [[BB9]]:
+; SIMPLIFY-CFG-NEXT:    [[TMP10:%.*]] = phi i32 [ [[TMP4]], %[[BB6]] ], [ [[TMP3]], %[[BB8]] ], [ [[TMP3]], %[[BB7]] ]
+; SIMPLIFY-CFG-NEXT:    ret i32 [[TMP10]]
+;
+; O3-LABEL: define i32 @switch_duplicate_arms_multipred(
+; O3-SAME: i1 [[TMP0:%.*]], i32 [[TMP1:%.*]], i32 [[TMP2:%.*]], i32 [[TMP3:%.*]], i32 [[TMP4:%.*]]) local_unnamed_addr #[[ATTR0]] {
+; O3-NEXT:    br i1 [[TMP0]], label %[[BB6:.*]], label %[[BB7:.*]]
+; O3:       [[BB6]]:
+; O3-NEXT:    switch i32 [[TMP2]], label %[[BB9:.*]] [
+; O3-NEXT:      i32 0, label %[[BB7]]
+; O3-NEXT:      i32 1, label %[[BB8:.*]]
+; O3-NEXT:    ]
+; O3:       [[BB7]]:
+; O3-NEXT:    br label %[[BB9]]
+; O3:       [[BB8]]:
+; O3-NEXT:    br label %[[BB9]]
+; O3:       [[BB9]]:
+; O3-NEXT:    [[TMP10:%.*]] = phi i32 [ [[TMP4]], %[[BB6]] ], [ [[TMP3]], %[[BB8]] ], [ [[TMP3]], %[[BB7]] ]
+; O3-NEXT:    ret i32 [[TMP10]]
+;
+  br i1 %0, label %6, label %7
+6:
+  switch i32 %2, label %9 [
+  i32 0, label %7
+  i32 1, label %8
+  ]
+
+7:
+  br label %9
+
+8:
+  br label %9
+
+9:
+  %10 = phi i32 [ %4, %6 ], [ %3, %8 ], [ %3, %7 ]
+  ret i32 %10
+}

llvm/test/Transforms/SimplifyCFG/switch-to-select-two-case.ll

@@ -272,16 +272,14 @@ define i8 @switch_to_select_two_case_results_no_default(i32 %i) {
 ; CHECK-NEXT:      i32 0, label [[END:%.*]]
 ; CHECK-NEXT:      i32 2, label [[END]]
 ; CHECK-NEXT:      i32 4, label [[CASE3:%.*]]
-; CHECK-NEXT:      i32 6, label [[CASE4:%.*]]
+; CHECK-NEXT:      i32 6, label [[CASE3]]
 ; CHECK-NEXT:    ]
 ; CHECK:       case3:
 ; CHECK-NEXT:    br label [[END]]
-; CHECK:       case4:
-; CHECK-NEXT:    br label [[END]]
 ; CHECK:       default:
 ; CHECK-NEXT:    unreachable
 ; CHECK:       end:
-; CHECK-NEXT:    [[T0:%.*]] = phi i8 [ 44, [[CASE3]] ], [ 44, [[CASE4]] ], [ 42, [[ENTRY:%.*]] ], [ 42, [[ENTRY]] ]
+; CHECK-NEXT:    [[T0:%.*]] = phi i8 [ 44, [[CASE3]] ], [ 42, [[ENTRY:%.*]] ], [ 42, [[ENTRY]] ]
 ; CHECK-NEXT:    ret i8 [[T0]]
 ;
 entry: