[JumpThreading] Don't select an edge that we know we can't thread

In r312664 (D36404), JumpThreading stopped threading edges into loop headers. Unfortunately, I observed a significant performance regression as a result of this change. Upon further investigation, the problematic pattern looked something like this (after many high level optimizations): while (true) { bool cond = ...; if (!cond) { <body> } if (cond) break; } Now, naturally we want jump threading to essentially eliminate the second if check and hook up the edges appropriately. However, the above mentioned change, prevented it from doing this because it would have to thread an edge into the loop header. Upon further investigation, what is happening is that since both branches are threadable, JumpThreading picks one of them at arbitrarily. In my case, because of the way that the IR ended up, it tended to pick the one to the loop header, bailing out immediately after. However, if it had picked the one to the exit block, everything would have worked out fine (because the only remaining branch would then be folded, not thraded which is acceptable). Thus, to fix this problem, we can simply eliminate loop headers from consideration as possible threading targets earlier, to make sure that if there are multiple eligible branches, we can still thread one of the ones that don't target a loop header. Patch by Keno Fischer! Differential Revision: https://reviews.llvm.org/D42260 llvm-svn: 328798
llvm · Mar 29, 2018 · c7cc879 · c7cc879
1 parent 0132baa
commit c7cc879
Show file tree

Hide file tree

Showing 2 changed files with 115 additions and 1 deletion.
diff --git a/llvm/lib/Transforms/Scalar/JumpThreading.cpp b/llvm/lib/Transforms/Scalar/JumpThreading.cpp
@@ -1499,6 +1499,9 @@ FindMostPopularDest(BasicBlock *BB,
     if (PredToDest.second)
       DestPopularity[PredToDest.second]++;
 
+  if (DestPopularity.empty())
+    return nullptr;
+
   // Find the most popular dest.
   DenseMap<BasicBlock*, unsigned>::iterator DPI = DestPopularity.begin();
   BasicBlock *MostPopularDest = DPI->first;
@@ -1679,8 +1682,20 @@ bool JumpThreadingPass::ProcessThreadableEdges(Value *Cond, BasicBlock *BB,
   // threadable destination (the common case) we can avoid this.
   BasicBlock *MostPopularDest = OnlyDest;
 
-  if (MostPopularDest == MultipleDestSentinel)
+  if (MostPopularDest == MultipleDestSentinel) {
+    // Remove any loop headers from the Dest list, ThreadEdge conservatively
+    // won't process them, but we might have other destination that are eligible
+    // and we still want to process.
+    erase_if(PredToDestList,
+             [&](const std::pair<BasicBlock *, BasicBlock *> &PredToDest) {
+               return LoopHeaders.count(PredToDest.second) != 0;
+             });
+
+    if (PredToDestList.empty())
+      return false;
+
     MostPopularDest = FindMostPopularDest(BB, PredToDestList);
+  }
 
   // Now that we know what the most popular destination is, factor all
   // predecessors that will jump to it into a single predecessor.

diff --git a/llvm/test/Transforms/JumpThreading/header-succ.ll b/llvm/test/Transforms/JumpThreading/header-succ.ll
@@ -0,0 +1,99 @@
+; RUN: opt -S -jump-threading < %s | FileCheck %s
+
+; Check that the heuristic for avoiding accidental introduction of irreducible
+; loops doesn't also prevent us from threading simple constructs where this
+; isn't a problem.
+
+declare void @opaque_body()
+
+define void @jump_threading_loopheader() {
+; CHECK-LABEL: @jump_threading_loopheader
+top:
+    br label %entry
+
+entry:
+    %ind = phi i32 [0, %top], [%nextind, %latch]
+    %nextind = add i32 %ind, 1
+    %cmp = icmp ule i32 %ind, 10
+; CHECK: br i1 %cmp, label %latch, label %exit
+    br i1 %cmp, label %body, label %latch
+
+body:
+    call void @opaque_body()
+; CHECK: br label %entry
+    br label %latch
+
+latch:
+    %cond = phi i2 [1, %entry], [2, %body]
+    switch i2 %cond, label %unreach [
+        i2 2, label %entry
+        i2 1, label %exit
+    ]
+
+unreach:
+    unreachable
+
+exit:
+    ret void
+}
+
+; We also need to check the opposite order of the branches, in the switch
+; instruction because jump-threading relies on that to decide which edge to
+; try to thread first.
+define void @jump_threading_loopheader2() {
+; CHECK-LABEL: @jump_threading_loopheader2
+top:
+    br label %entry
+
+entry:
+    %ind = phi i32 [0, %top], [%nextind, %latch]
+    %nextind = add i32 %ind, 1
+    %cmp = icmp ule i32 %ind, 10
+; CHECK: br i1 %cmp, label %exit, label %latch
+    br i1 %cmp, label %body, label %latch
+
+body:
+    call void @opaque_body()
+; CHECK: br label %entry
+    br label %latch
+
+latch:
+    %cond = phi i2 [1, %entry], [2, %body]
+    switch i2 %cond, label %unreach [
+        i2 1, label %entry
+        i2 2, label %exit
+    ]
+
+unreach:
+    unreachable
+
+exit:
+    ret void
+}
+
+; Check if we can handle undef branch condition.
+define void @jump_threading_loopheader3() {
+; CHECK-LABEL: @jump_threading_loopheader3
+top:
+    br label %entry
+
+entry:
+    %ind = phi i32 [0, %top], [%nextind, %latch]
+    %nextind = add i32 %ind, 1
+    %cmp = icmp ule i32 %ind, 10
+; CHECK: br i1 %cmp, label %latch, label %exit
+    br i1 %cmp, label %body, label %latch
+
+body:
+    call void @opaque_body()
+; CHECK: br label %entry
+    br label %latch
+
+latch:
+   %phi = phi i32 [undef, %entry], [0, %body]
+   %cmp1 = icmp eq i32 %phi, 0
+   br i1 %cmp1, label %entry, label %exit
+
+exit:
+    ret void
+}