[JumpThread] We want to fold (not thread) when all predecessor go to single BB's successor.

[trentxintong]

Summary

In case all predecessor go to a single successor of current BB. We want to fold (not thread).

I failed to update the phi nodes properly in the last patch https://reviews.llvm.org/rL300657.

Phi nodes values are per predecessor in LLVM.

[trentxintong]

@sanjoy Appreciate it if you can take a quick look :).

[sanjoy]

lgtm with minor nit

lib/Transforms/Scalar/JumpThreading.cpp

1298

The TheOnlyDest variable name threw me off a bit. How about being 100% explicit, and using a bool SeenFirstBranchToOnlyDest;?

[trentxintong]

Address comment

lib/Transforms/Scalar/JumpThreading.cpp

for (const auto &PredValue : PredValues) {
 
     PredToDestList.push_back(std::make_pair(Pred, DestBB));
   }
 
   // If all edges were unthreadable, we fail.
   if (PredToDestList.empty())
     return false;
 
+  // If all the predecessors go to a single known successor, we want to fold,
+  // not thread. By doing so, we do not need to duplicate the current block and
+  // also miss potential opportunities in case we dont/cant duplicate.
+  if (OnlyDest && OnlyDest != MultipleDestSentinel) {
+    if (PredToDestList.size() ==
+        (size_t)std::distance(pred_begin(BB), pred_end(BB))) {
+      auto *TheOnlyDest = OnlyDest;
+      for (BasicBlock *SuccBB : successors(BB)) {
+        if (SuccBB == TheOnlyDest)
+          TheOnlyDest = nullptr; // Don't modify the first branch to TheOnlyDest
+        else
+          SuccBB->removePredecessor(BB, true); // This is unreachable successor.
+      }
+
+      // Finally update the terminator.
+      TerminatorInst *Term = BB->getTerminator();
+      BranchInst::Create(OnlyDest, Term);
+      Term->eraseFromParent();
+
+      // If the condition is now dead due to the removal of the old terminator,
+      // erase it.
+      auto *CondInst = dyn_cast<Instruction>(Cond);
+      if (CondInst && CondInst->use_empty())
+        CondInst->eraseFromParent();
+      // FIXME: in case this instruction is defined in the current BB and it
+      // resolves to a single value from all predecessors, we can do RAUW.
+      return true;
+    }
+  }
+
   // Determine which is the most common successor.  If we have many inputs and
   // this block is a switch, we want to start by threading the batch that goes
   // to the most popular destination first.  If we only know about one
   // threadable destination (the common case) we can avoid this.
   BasicBlock *MostPopularDest = OnlyDest;
 
   if (MostPopularDest == MultipleDestSentinel)
     MostPopularDest = FindMostPopularDest(BB, PredToDestList);

test/Transforms/JumpThreading/fold-not-thread.ll

+; RUN: opt -jump-threading -S -verify < %s | FileCheck %s
+
+declare i32 @f1()
+declare i32 @f2()
+declare void @f3()
+declare void @f4(i32)
+
+
+; Make sure we update the phi node properly.
+;
+; CHECK-LABEL: define void @test_br_folding_not_threading_update_phi(
+; CHECK: br label %L1
+; Make sure we update the phi node properly here, i.e. we only have 2 predecessors, entry and L0
+; CHECK: %res.0 = phi i32 [ 0, %L0 ], [ 1, %entry ]
+define void @test_br_folding_not_threading_update_phi(i32 %val) nounwind {
+entry:
+  %cmp = icmp eq i32 %val, 32
+  br i1 %cmp, label %L0, label %L1
+L0:
+  call i32 @f2()
+  call i32 @f2()
+  call i32 @f2()
+  call i32 @f2()
+  call i32 @f2()
+  call i32 @f2()
+  call i32 @f2()
+  call i32 @f2()
+  call i32 @f2()
+  call i32 @f2()
+  call i32 @f2()
+  call i32 @f2()
+  call i32 @f2()
+  switch i32 %val, label %L2 [
+    i32 0, label %L1
+    i32 32, label %L1
+  ]
+
+L1:
+  %res.0 = phi i32 [ 0, %L0 ], [ 0, %L0 ], [1, %entry]
+  call void @f4(i32 %res.0)
+  ret void
+L2:
+  call void @f3()
+  ret void
+}
+
+; Make sure we can fold this branch ... We will not be able to thread it as
+; L0 is too big to duplicate. L2 is the unreachable block here.
+;
+; CHECK-LABEL: @test_br_folding_not_threading(
+; CHECK: L1:
+; CHECK: call i32 @f2()
+; CHECK: call void @f3()
+; CHECK-NEXT: ret void
+; CHECK-NOT: br
+; CHECK: L3:
+define void @test_br_folding_not_threading(i1 %cond) nounwind {
+entry:
+  br i1 %cond, label %L0, label %L3 
+L0:
+  call i32 @f2()
+  call i32 @f2()
+  call i32 @f2()
+  call i32 @f2()
+  call i32 @f2()
+  call i32 @f2()
+  call i32 @f2()
+  call i32 @f2()
+  call i32 @f2()
+  call i32 @f2()
+  call i32 @f2()
+  call i32 @f2()
+  call i32 @f2()
+  br i1 %cond, label %L1, label %L2 
+
+L1:
+  call void @f3()
+  ret void
+L2:
+  call void @f3()
+  ret void
+L3:
+  call void @f3()
+  ret void
+}
+
+
+; Make sure we can fold this branch ... We will not be able to thread it as
+; L0 is too big to duplicate. L2 is the unreachable block here.
+; With more than 1 predecessors.
+;
+; CHECK-LABEL: @test_br_folding_not_threading_multiple_preds(
+; CHECK: L1:
+; CHECK: call i32 @f2()
+; CHECK: call void @f3()
+; CHECK-NEXT: ret void
+; CHECK-NOT: br
+; CHECK: L3:
+define void @test_br_folding_not_threading_multiple_preds(i1 %condx, i1 %cond) nounwind {
+entry:
+  br i1 %condx, label %X0, label %X1
+
+X0:
+  br i1 %cond, label %L0, label %L3 
+
+X1:
+  br i1 %cond, label %L0, label %L3 
+
+L0:
+  call i32 @f2()
+  call i32 @f2()
+  call i32 @f2()
+  call i32 @f2()
+  call i32 @f2()
+  call i32 @f2()
+  call i32 @f2()
+  call i32 @f2()
+  call i32 @f2()
+  call i32 @f2()
+  call i32 @f2()
+  call i32 @f2()
+  call i32 @f2()
+  br i1 %cond, label %L1, label %L2 
+
+L1:
+  call void @f3()
+  ret void
+L2:
+  call void @f3()
+  ret void
+L3:
+  call void @f3()
+  ret void
+}
+