Reapply 155668: Fix the SD scheduler to avoid gluing the same node twice.

This time, also fix the caller of AddGlue to properly handle
incomplete chains. AddGlue had failure modes, but shamefully hid them
from its caller. It's luck ran out.

Fixes rdar://11314175: BuildSchedUnits assert.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@155749 91177308-0d34-0410-b5e6-96231b3b80d8

2 files changed, 104 insertions, 26 deletions

diff --git a/lib/CodeGen/SelectionDAG/ScheduleDAGSDNodes.cpp b/lib/CodeGen/SelectionDAG/ScheduleDAGSDNodes.cpp
index 52c47767ca..748668cdf6 100644
--- a/lib/CodeGen/SelectionDAG/ScheduleDAGSDNodes.cpp
+++ b/lib/CodeGen/SelectionDAG/ScheduleDAGSDNodes.cpp
@@ -131,28 +131,16 @@ static void CheckForPhysRegDependency(SDNode *Def, SDNode *User, unsigned Op,
   }
 }
 
-static void AddGlue(SDNode *N, SDValue Glue, bool AddGlue, SelectionDAG *DAG) {
-  SmallVector<EVT, 4> VTs;
-  SDNode *GlueDestNode = Glue.getNode();
-
-  // Don't add glue from a node to itself.
-  if (GlueDestNode == N) return;
-
-  // Don't add glue to something that already has it, either as a use or value.
-  if (N->getValueType(N->getNumValues() - 1) == MVT::Glue) return;
-
-  for (unsigned I = 0, E = N->getNumValues(); I != E; ++I)
-    VTs.push_back(N->getValueType(I));
-
-  if (AddGlue)
-    VTs.push_back(MVT::Glue);
-
+// Helper for AddGlue to clone node operands.
+static void CloneNodeWithValues(SDNode *N, SelectionDAG *DAG,
+                                SmallVectorImpl<EVT> &VTs,
+                                SDValue ExtraOper = SDValue()) {
   SmallVector<SDValue, 4> Ops;
   for (unsigned I = 0, E = N->getNumOperands(); I != E; ++I)
     Ops.push_back(N->getOperand(I));
 
-  if (GlueDestNode)
-    Ops.push_back(Glue);
+  if (ExtraOper.getNode())
+    Ops.push_back(ExtraOper);
 
   SDVTList VTList = DAG->getVTList(&VTs[0], VTs.size());
   MachineSDNode::mmo_iterator Begin = 0, End = 0;
@@ -171,6 +159,46 @@ static void AddGlue(SDNode *N, SDValue Glue, bool AddGlue, SelectionDAG *DAG) {
     MN->setMemRefs(Begin, End);
 }
 
+static bool AddGlue(SDNode *N, SDValue Glue, bool AddGlue, SelectionDAG *DAG) {
+  SmallVector<EVT, 4> VTs;
+  SDNode *GlueDestNode = Glue.getNode();
+
+  // Don't add glue from a node to itself.
+  if (GlueDestNode == N) return false;
+
+  // Don't add a glue operand to something that already uses glue.
+  if (GlueDestNode &&
+      N->getOperand(N->getNumOperands()-1).getValueType() == MVT::Glue) {
+    return false;
+  }
+  // Don't add glue to something that already has a glue value.
+  if (N->getValueType(N->getNumValues() - 1) == MVT::Glue) return false;
+
+  for (unsigned I = 0, E = N->getNumValues(); I != E; ++I)
+    VTs.push_back(N->getValueType(I));
+
+  if (AddGlue)
+    VTs.push_back(MVT::Glue);
+
+  CloneNodeWithValues(N, DAG, VTs, Glue);
+
+  return true;
+}
+
+// Cleanup after unsuccessful AddGlue. Use the standard method of morphing the
+// node even though simply shrinking the value list is sufficient.
+static void RemoveUnusedGlue(SDNode *N, SelectionDAG *DAG) {
+  assert((N->getValueType(N->getNumValues() - 1) == MVT::Glue &&
+          !N->hasAnyUseOfValue(N->getNumValues() - 1)) &&
+         "expected an unused glue value");
+
+  SmallVector<EVT, 4> VTs;
+  for (unsigned I = 0, E = N->getNumValues()-1; I != E; ++I)
+    VTs.push_back(N->getValueType(I));
+
+  CloneNodeWithValues(N, DAG, VTs);
+}
+
 /// ClusterNeighboringLoads - Force nearby loads together by "gluing" them.
 /// This function finds loads of the same base and different offsets. If the
 /// offsets are not far apart (target specific), it add MVT::Glue inputs and
@@ -238,19 +266,23 @@ void ScheduleDAGSDNodes::ClusterNeighboringLoads(SDNode *Node) {
   // Cluster loads by adding MVT::Glue outputs and inputs. This also
   // ensure they are scheduled in order of increasing addresses.
   SDNode *Lead = Loads[0];
-  AddGlue(Lead, SDValue(0, 0), true, DAG);
-
-  SDValue InGlue = SDValue(Lead, Lead->getNumValues() - 1);
+  SDValue InGlue = SDValue(0, 0);
+  if (AddGlue(Lead, InGlue, true, DAG))
+    InGlue = SDValue(Lead, Lead->getNumValues() - 1);
   for (unsigned I = 1, E = Loads.size(); I != E; ++I) {
     bool OutGlue = I < E - 1;
     SDNode *Load = Loads[I];
 
-    AddGlue(Load, InGlue, OutGlue, DAG);
+    // If AddGlue fails, we could leave an unsused glue value. This should not
+    // cause any
+    if (AddGlue(Load, InGlue, OutGlue, DAG)) {
+      if (OutGlue)
+        InGlue = SDValue(Load, Load->getNumValues() - 1);
 
-    if (OutGlue)
-      InGlue = SDValue(Load, Load->getNumValues() - 1);
-
-    ++LoadsClustered;
+      ++LoadsClustered;
+    }
+    else if (!OutGlue && InGlue.getNode())
+      RemoveUnusedGlue(InGlue.getNode(), DAG);
   }
 }
 
diff --git a/test/CodeGen/X86/2012-04-26-sdglue.ll b/test/CodeGen/X86/2012-04-26-sdglue.ll
new file mode 100644
index 0000000000..9543587747
--- /dev/null
+++ b/test/CodeGen/X86/2012-04-26-sdglue.ll
@@ -0,0 +1,46 @@
+; RUN: llc < %s -mtriple=x86_64-apple-darwin -mcpu=core-avx2 -mattr=+avx | FileCheck %s
+; rdar://11314175: SD Scheduler, BuildSchedUnits assert:
+;                  N->getNodeId() == -1 && "Node already inserted!
+
+; It's hard to test for the ISEL condition because CodeGen optimizes
+; away the bugpointed code. Just ensure the basics are still there.
+;CHECK: func:
+;CHECK: vmovups
+;CHECK: vpshufd
+;CHECK: vpshufd
+;CHECK: vmulps
+;CHECK: vmulps
+;CHECK: ret
+
+define void @func() nounwind ssp {
+  %tmp = load <4 x float>* null, align 1
+  %tmp14 = getelementptr <4 x float>* null, i32 2
+  %tmp15 = load <4 x float>* %tmp14, align 1
+  %tmp16 = shufflevector <4 x float> %tmp, <4 x float> <float 0.000000e+00, float undef, float undef, float undef>, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 4, i32 4, i32 4>
+  %tmp17 = call <8 x float> @llvm.x86.avx.vinsertf128.ps.256(<8 x float> %tmp16, <4 x float> undef, i8 1)
+  %tmp18 = bitcast <4 x float> %tmp to <16 x i8>
+  %tmp19 = shufflevector <16 x i8> %tmp18, <16 x i8> undef, <16 x i32> <i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19>
+  %tmp20 = bitcast <16 x i8> %tmp19 to <4 x float>
+  %tmp21 = bitcast <4 x float> %tmp15 to <16 x i8>
+  %tmp22 = shufflevector <16 x i8> undef, <16 x i8> %tmp21, <16 x i32> <i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19>
+  %tmp23 = bitcast <16 x i8> %tmp22 to <4 x float>
+  %tmp24 = shufflevector <4 x float> %tmp20, <4 x float> <float 0.000000e+00, float undef, float undef, float undef>, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 4, i32 4, i32 4>
+  %tmp25 = call <8 x float> @llvm.x86.avx.vinsertf128.ps.256(<8 x float> %tmp24, <4 x float> %tmp23, i8 1)
+  %tmp26 = fmul <8 x float> %tmp17, undef
+  %tmp27 = fmul <8 x float> %tmp25, undef
+  %tmp28 = fadd <8 x float> %tmp26, %tmp27
+  %tmp29 = fadd <8 x float> %tmp28, undef
+  %tmp30 = shufflevector <8 x float> %tmp29, <8 x float> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %tmp31 = fmul <4 x float> undef, %tmp30
+  %tmp32 = call <8 x float> @llvm.x86.avx.vinsertf128.ps.256(<8 x float> zeroinitializer, <4 x float> %tmp31, i8 1)
+  %tmp33 = fadd <8 x float> undef, %tmp32
+  %tmp34 = call <8 x float> @llvm.x86.avx.hadd.ps.256(<8 x float> %tmp33, <8 x float> undef) nounwind
+  %tmp35 = fsub <8 x float> %tmp34, undef
+  %tmp36 = call <8 x float> @llvm.x86.avx.hadd.ps.256(<8 x float> zeroinitializer, <8 x float> %tmp35) nounwind
+  store <8 x float> %tmp36, <8 x float>* undef, align 32
+  ret void
+}
+
+declare <8 x float> @llvm.x86.avx.vinsertf128.ps.256(<8 x float>, <4 x float>, i8) nounwind readnone
+
+declare <8 x float> @llvm.x86.avx.hadd.ps.256(<8 x float>, <8 x float>) nounwind readnone