2 files changed, 61 insertions, 1 deletions

diff --git a/lib/CodeGen/SelectionDAG/ScheduleDAGRRList.cpp b/lib/CodeGen/SelectionDAG/ScheduleDAGRRList.cpp
index 5f28cdb407..88bd4509b4 100644
--- a/lib/CodeGen/SelectionDAG/ScheduleDAGRRList.cpp
+++ b/lib/CodeGen/SelectionDAG/ScheduleDAGRRList.cpp
@@ -1732,7 +1732,17 @@ unsigned RegReductionPQBase::getNodePriority(const SUnit *SU) const {
     // If SU does not have a register def, schedule it close to its uses
     // because it does not lengthen any live ranges.
     return 0;
+#if 1
   return SethiUllmanNumbers[SU->NodeNum];
+#else
+  unsigned Priority = SethiUllmanNumbers[SU->NodeNum];
+  if (SU->isCallOp) {
+    // FIXME: This assumes all of the defs are used as call operands.
+    int NP = (int)Priority - SU->getNode()->getNumValues();
+    return (NP > 0) ? NP : 0;
+  }
+  return Priority;
+#endif
 }
 
 //===----------------------------------------------------------------------===//
@@ -2238,11 +2248,35 @@ static bool BURRSort(SUnit *left, SUnit *right, RegReductionPQBase *SPQ) {
   // Prioritize by Sethi-Ulmann number and push CopyToReg nodes down.
   unsigned LPriority = SPQ->getNodePriority(left);
   unsigned RPriority = SPQ->getNodePriority(right);
+
+  // Be really careful about hoisting call operands above previous calls.
+  // Only allows it if it would reduce register pressure.
+  if (left->isCall && right->isCallOp) {
+    unsigned RNumVals = right->getNode()->getNumValues();
+    RPriority = (RPriority > RNumVals) ? (RPriority - RNumVals) : 0;
+  }
+  if (right->isCall && left->isCallOp) {
+    unsigned LNumVals = left->getNode()->getNumValues();
+    LPriority = (LPriority > LNumVals) ? (LPriority - LNumVals) : 0;
+  }
+
   if (LPriority != RPriority) {
     DEBUG(++FactorCount[FactStatic]);
     return LPriority > RPriority;
   }
 
+  // One or both of the nodes are calls and their sethi-ullman numbers are the
+  // same, then keep source order.
+  if (left->isCall || right->isCall) {
+    unsigned LOrder = SPQ->getNodeOrdering(left);
+    unsigned ROrder = SPQ->getNodeOrdering(right);
+
+    // Prefer an ordering where the lower the non-zero order number, the higher
+    // the preference.
+    if ((LOrder || ROrder) && LOrder != ROrder)
+      return LOrder != 0 && (LOrder < ROrder || ROrder == 0);
+  }
+
   // Try schedule def + use closer when Sethi-Ullman numbers are the same.
   // e.g.
   // t1 = op t2, c1
@@ -2275,7 +2309,14 @@ static bool BURRSort(SUnit *left, SUnit *right, RegReductionPQBase *SPQ) {
     return LScratch > RScratch;
   }
 
-  if (!DisableSchedCycles) {
+  // Comparing latency against a call makes little sense unless the node
+  // is register pressure-neutral.
+  if ((left->isCall && RPriority > 0) || (right->isCall && LPriority > 0))
+    return (left->NodeQueueId > right->NodeQueueId);
+
+  // Do not compare latencies when one or both of the nodes are calls.
+  if (!DisableSchedCycles &&
+      !(left->isCall || right->isCall)) {
     int result = BUCompareLatency(left, right, false /*checkPref*/, SPQ);
     if (result != 0)
       return result > 0;
diff --git a/lib/CodeGen/SelectionDAG/ScheduleDAGSDNodes.cpp b/lib/CodeGen/SelectionDAG/ScheduleDAGSDNodes.cpp
index 94b8c2f204..9f2f0121a8 100644
--- a/lib/CodeGen/SelectionDAG/ScheduleDAGSDNodes.cpp
+++ b/lib/CodeGen/SelectionDAG/ScheduleDAGSDNodes.cpp
@@ -83,6 +83,7 @@ SUnit *ScheduleDAGSDNodes::Clone(SUnit *Old) {
   SU->Latency = Old->Latency;
   SU->isVRegCycle = Old->isVRegCycle;
   SU->isCall = Old->isCall;
+  SU->isCallOp = Old->isCallOp;
   SU->isTwoAddress = Old->isTwoAddress;
   SU->isCommutable = Old->isCommutable;
   SU->hasPhysRegDefs = Old->hasPhysRegDefs;
@@ -285,6 +286,7 @@ void ScheduleDAGSDNodes::BuildSchedUnits() {
   Worklist.push_back(DAG->getRoot().getNode());
   Visited.insert(DAG->getRoot().getNode());
 
+  SmallVector<SUnit*, 8> CallSUnits;
   while (!Worklist.empty()) {
     SDNode *NI = Worklist.pop_back_val();
 
@@ -337,6 +339,9 @@ void ScheduleDAGSDNodes::BuildSchedUnits() {
       if (!HasGlueUse) break;
     }
 
+    if (NodeSUnit->isCall)
+      CallSUnits.push_back(NodeSUnit);
+
     // Schedule zero-latency TokenFactor below any nodes that may increase the
     // schedule height. Otherwise, ancestors of the TokenFactor may appear to
     // have false stalls.
@@ -356,6 +361,20 @@ void ScheduleDAGSDNodes::BuildSchedUnits() {
     // Assign the Latency field of NodeSUnit using target-provided information.
     ComputeLatency(NodeSUnit);
   }
+
+  // Find all call operands.
+  while (!CallSUnits.empty()) {
+    SUnit *SU = CallSUnits.pop_back_val();
+    for (const SDNode *SUNode = SU->getNode(); SUNode;
+         SUNode = SUNode->getGluedNode()) {
+      if (SUNode->getOpcode() != ISD::CopyToReg)
+        continue;
+      SDNode *SrcN = SUNode->getOperand(2).getNode();
+      if (isPassiveNode(SrcN)) continue;   // Not scheduled.
+      SUnit *SrcSU = &SUnits[SrcN->getNodeId()];
+      SrcSU->isCallOp = true;
+    }
+  }
 }
 
 void ScheduleDAGSDNodes::AddSchedEdges() {