Use DomTree in MachineSink to sink over diamonds.

According to a previous FIXME comment we now not only look at MBB
successors, but also handle code sinking past them:

  x = computation
  if () {} else {}
  use x

The instruction could be sunk over the whole diamond for the
if/then/else (or loop, etc), allowing it to be sunk into other blocks
after that.

Modified test added in r204522, due to one spill less present.

Minor fixes in comments.

Patch provided by Jonas Paulsson. Reviewed by Hal Finkel.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223350 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/CodeGen/MachineSink.cpp b/lib/CodeGen/MachineSink.cpp
index ba25bca0a6e180ec46421fcbfa211af44e288c5f..8337793d960d9ab82e13b7d3875395d1f2249f9d 100644 (file)
--- a/lib/CodeGen/MachineSink.cpp
+++ b/lib/CodeGen/MachineSink.cpp
@@ -112,7 +112,7 @@ namespace {
     /// for the lifetime of an iteration.
     ///
     /// \return True if the edge is marked as toSplit, false otherwise.
-    /// False can be retruned if, for instance, this is not profitable.
+    /// False can be returned if, for instance, this is not profitable.
     bool PostponeSplitCriticalEdge(MachineInstr *MI,
                                    MachineBasicBlock *From,
                                    MachineBasicBlock *To,
@@ -504,7 +504,7 @@ bool MachineSinking::isProfitableToSinkTo(unsigned Reg, MachineInstr *MI,
   // If SuccToSinkTo post dominates then also it may be profitable if MI
   // can further profitably sinked into another block in next round.
   bool BreakPHIEdge = false;
-  // FIXME - If finding successor is compile time expensive then catch results.
+  // FIXME - If finding successor is compile time expensive then cache results.
   if (MachineBasicBlock *MBB2 = FindSuccToSinkTo(MI, SuccToSinkTo, BreakPHIEdge))
     return isProfitableToSinkTo(Reg, MI, SuccToSinkTo, MBB2);
 
@@ -553,19 +553,6 @@ MachineBasicBlock *MachineSinking::FindSuccToSinkTo(MachineInstr *MI,
       if (!TII->isSafeToMoveRegClassDefs(MRI->getRegClass(Reg)))
         return nullptr;
 
-      // FIXME: This picks a successor to sink into based on having one
-      // successor that dominates all the uses.  However, there are cases where
-      // sinking can happen but where the sink point isn't a successor.  For
-      // example:
-      //
-      //   x = computation
-      //   if () {} else {}
-      //   use x
-      //
-      // the instruction could be sunk over the whole diamond for the
-      // if/then/else (or loop, etc), allowing it to be sunk into other blocks
-      // after that.
-
       // Virtual register defs can only be sunk if all their uses are in blocks
       // dominated by one of the successors.
       if (SuccToSinkTo) {
@@ -585,6 +572,23 @@ MachineBasicBlock *MachineSinking::FindSuccToSinkTo(MachineInstr *MI,
       // higher priority, otherwise prioritize smaller loop depths.
       SmallVector<MachineBasicBlock*, 4> Succs(MBB->succ_begin(),
                                                MBB->succ_end());
+
+      // Handle cases where sinking can happen but where the sink point isn't a
+      // successor. For example:
+      //
+      //   x = computation
+      //   if () {} else {}
+      //   use x
+      //
+      const std::vector<MachineDomTreeNode *> &Children =
+        DT->getNode(MBB)->getChildren();
+      for (const auto &DTChild : Children)
+        // DomTree children of MBB that have MBB as immediate dominator are added.
+        if (DTChild->getIDom()->getBlock() == MI->getParent() &&
+            // Skip MBBs already added to the Succs vector above.
+            !MBB->isSuccessor(DTChild->getBlock()))
+          Succs.push_back(DTChild->getBlock());
+
       // Sort Successors according to their loop depth or block frequency info.
       std::stable_sort(
           Succs.begin(), Succs.end(),

diff --git a/test/CodeGen/X86/ragreedy-hoist-spill.ll b/test/CodeGen/X86/ragreedy-hoist-spill.ll
index c6b28f71af461e68c40e4087a8ec32594d6d5507..d628ae5b064d8442298add47e5d54bdd28e16686 100644 (file)
--- a/test/CodeGen/X86/ragreedy-hoist-spill.ll
+++ b/test/CodeGen/X86/ragreedy-hoist-spill.ll
@@ -202,7 +202,6 @@ lor.rhs500:
   ; CHECK: lor.rhs500
   ; Make sure that we don't hoist the spill to outer loops.
   ; CHECK: movq %r{{.*}}, {{[0-9]+}}(%rsp)
-  ; CHECK: movq %r{{.*}}, {{[0-9]+}}(%rsp)
   ; CHECK: callq {{.*}}maskrune
   %call3.i.i2792 = call i32 @__maskrune(i32 undef, i64 256)
   br i1 undef, label %land.lhs.true504, label %do.body479.backedge