R600: Use bottom up scheduling algorithm

llvm-svn: 182129

19 files changed, 56 insertions, 42 deletions

diff --git a/llvm/lib/Target/R600/R600MachineScheduler.cpp b/llvm/lib/Target/R600/R600MachineScheduler.cpp
index aeb2674f4e6..4c89124f46d 100644
--- a/llvm/lib/Target/R600/R600MachineScheduler.cpp
+++ b/llvm/lib/Target/R600/R600MachineScheduler.cpp
@@ -34,7 +34,7 @@ void R600SchedStrategy::initialize(ScheduleDAGMI *dag) {
   CurEmitted = 0;
   OccupedSlotsMask = 15;
   InstKindLimit[IDAlu] = TII->getMaxAlusPerClause();
-
+  InstKindLimit[IDOther] = 32;
 
   const AMDGPUSubtarget &ST = DAG->TM.getSubtarget<AMDGPUSubtarget>();
   InstKindLimit[IDFetch] = ST.getTexVTXClauseSize();
@@ -49,12 +49,12 @@ void R600SchedStrategy::MoveUnits(std::vector<SUnit *> &QSrc,
 
 SUnit* R600SchedStrategy::pickNode(bool &IsTopNode) {
   SUnit *SU = 0;
-  IsTopNode = true;
   NextInstKind = IDOther;
 
+  IsTopNode = false;
+
   // check if we might want to switch current clause type
-  bool AllowSwitchToAlu = (CurInstKind == IDOther) ||
-      (CurEmitted >= InstKindLimit[CurInstKind]) ||
+  bool AllowSwitchToAlu = (CurEmitted >= InstKindLimit[CurInstKind]) ||
       (Available[CurInstKind].empty());
   bool AllowSwitchFromAlu = (CurEmitted >= InstKindLimit[CurInstKind]) &&
       (!Available[IDFetch].empty() || !Available[IDOther].empty());
@@ -86,10 +86,10 @@ SUnit* R600SchedStrategy::pickNode(bool &IsTopNode) {
 
   DEBUG(
       if (SU) {
-        dbgs() << "picked node: ";
+        dbgs() << " ** Pick node **\n";
         SU->dump(DAG);
       } else {
-        dbgs() << "NO NODE ";
+        dbgs() << "NO NODE \n";
         for (unsigned i = 0; i < DAG->SUnits.size(); i++) {
           const SUnit &S = DAG->SUnits[i];
           if (!S.isScheduled)
@@ -103,9 +103,6 @@ SUnit* R600SchedStrategy::pickNode(bool &IsTopNode) {
 
 void R600SchedStrategy::schedNode(SUnit *SU, bool IsTopNode) {
 
-  DEBUG(dbgs() << "scheduled: ");
-  DEBUG(SU->dump(DAG));
-
   if (NextInstKind != CurInstKind) {
     DEBUG(dbgs() << "Instruction Type Switch\n");
     if (NextInstKind != IDAlu)
@@ -141,19 +138,23 @@ void R600SchedStrategy::schedNode(SUnit *SU, bool IsTopNode) {
   if (CurInstKind != IDFetch) {
     MoveUnits(Pending[IDFetch], Available[IDFetch]);
   }
-  MoveUnits(Pending[IDOther], Available[IDOther]);
 }
 
 void R600SchedStrategy::releaseTopNode(SUnit *SU) {
-  int IK = getInstKind(SU);
-
-  DEBUG(dbgs() << IK << " <= ");
-  DEBUG(SU->dump(DAG));
+  DEBUG(dbgs() << "Top Releasing ";SU->dump(DAG););
 
-  Pending[IK].push_back(SU);
 }
 
 void R600SchedStrategy::releaseBottomNode(SUnit *SU) {
+  DEBUG(dbgs() << "Bottom Releasing ";SU->dump(DAG););
+
+  int IK = getInstKind(SU);
+  // There is no export clause, we can schedule one as soon as its ready
+  if (IK == IDOther)
+    Available[IDOther].push_back(SU);
+  else
+    Pending[IK].push_back(SU);
+
 }
 
 bool R600SchedStrategy::regBelongsToClass(unsigned Reg,
@@ -169,18 +170,15 @@ R600SchedStrategy::AluKind R600SchedStrategy::getAluKind(SUnit *SU) const {
   MachineInstr *MI = SU->getInstr();
 
     switch (MI->getOpcode()) {
+    case AMDGPU::PRED_X:
+      return AluPredX;
     case AMDGPU::INTERP_PAIR_XY:
     case AMDGPU::INTERP_PAIR_ZW:
     case AMDGPU::INTERP_VEC_LOAD:
     case AMDGPU::DOT_4:
       return AluT_XYZW;
     case AMDGPU::COPY:
-      if (TargetRegisterInfo::isPhysicalRegister(MI->getOperand(1).getReg())) {
-        // %vregX = COPY Tn_X is likely to be discarded in favor of an
-        // assignement of Tn_X to %vregX, don't considers it in scheduling
-        return AluDiscarded;
-      }
-      else if (MI->getOperand(1).isUndef()) {
+      if (MI->getOperand(1).isUndef()) {
         // MI will become a KILL, don't considers it in scheduling
         return AluDiscarded;
       }
@@ -238,6 +236,7 @@ int R600SchedStrategy::getInstKind(SUnit* SU) {
   }
 
   switch (Opcode) {
+  case AMDGPU::PRED_X:
   case AMDGPU::COPY:
   case AMDGPU::CONST_COPY:
   case AMDGPU::INTERP_PAIR_XY:
@@ -328,12 +327,18 @@ bool R600SchedStrategy::isAvailablesAluEmpty() const {
   return Pending[IDAlu].empty() && AvailableAlus[AluAny].empty() &&
       AvailableAlus[AluT_XYZW].empty() && AvailableAlus[AluT_X].empty() &&
       AvailableAlus[AluT_Y].empty() && AvailableAlus[AluT_Z].empty() &&
-      AvailableAlus[AluT_W].empty() && AvailableAlus[AluDiscarded].empty();
+      AvailableAlus[AluT_W].empty() && AvailableAlus[AluDiscarded].empty() &&
+      AvailableAlus[AluPredX].empty();
 }
 
 SUnit* R600SchedStrategy::pickAlu() {
   while (!isAvailablesAluEmpty()) {
     if (!OccupedSlotsMask) {
+      // Bottom up scheduling : predX must comes first
+      if (!AvailableAlus[AluPredX].empty()) {
+        OccupedSlotsMask = 15;
+        return PopInst(AvailableAlus[AluPredX]);
+      }
       // Flush physical reg copies (RA will discard them)
       if (!AvailableAlus[AluDiscarded].empty()) {
         OccupedSlotsMask = 15;
@@ -345,7 +350,7 @@ SUnit* R600SchedStrategy::pickAlu() {
         return PopInst(AvailableAlus[AluT_XYZW]);
       }
     }
-    for (unsigned Chan = 0; Chan < 4; ++Chan) {
+    for (int Chan = 3; Chan > -1; --Chan) {
       bool isOccupied = OccupedSlotsMask & (1 << Chan);
       if (!isOccupied) {
         SUnit *SU = AttemptFillSlot(Chan);
diff --git a/llvm/lib/Target/R600/R600MachineScheduler.h b/llvm/lib/Target/R600/R600MachineScheduler.h
index c82ee49c78b..1db877db85d 100644
--- a/llvm/lib/Target/R600/R600MachineScheduler.h
+++ b/llvm/lib/Target/R600/R600MachineScheduler.h
@@ -45,13 +45,13 @@ class R600SchedStrategy : public MachineSchedStrategy {
     AluT_Z,
     AluT_W,
     AluT_XYZW,
+    AluPredX,
     AluDiscarded, // LLVM Instructions that are going to be eliminated
     AluLast
   };
 
   std::vector<SUnit *> Available[IDLast], Pending[IDLast];
   std::vector<SUnit *> AvailableAlus[AluLast];
-  std::vector<SUnit *> FakeCopy;
 
   InstKind CurInstKind;
   int CurEmitted;
diff --git a/llvm/lib/Target/R600/R600RegisterInfo.cpp b/llvm/lib/Target/R600/R600RegisterInfo.cpp
index bbd7995d7d5..7d13420a31b 100644
--- a/llvm/lib/Target/R600/R600RegisterInfo.cpp
+++ b/llvm/lib/Target/R600/R600RegisterInfo.cpp
@@ -25,7 +25,7 @@ R600RegisterInfo::R600RegisterInfo(AMDGPUTargetMachine &tm,
 : AMDGPURegisterInfo(tm, tii),
   TM(tm),
   TII(tii)
-  { }
+  { RCW.RegWeight = 0; RCW.WeightLimit = 0;}
 
 BitVector R600RegisterInfo::getReservedRegs(const MachineFunction &MF) const {
   BitVector Reserved(getNumRegs());
@@ -97,3 +97,7 @@ unsigned R600RegisterInfo::getSubRegFromChannel(unsigned Channel) const {
   }
 }
 
+const RegClassWeight &R600RegisterInfo::getRegClassWeight(
+  const TargetRegisterClass *RC) const {
+  return RCW;
+}
diff --git a/llvm/lib/Target/R600/R600RegisterInfo.h b/llvm/lib/Target/R600/R600RegisterInfo.h
index f9ca918f246..1270a1eea60 100644
--- a/llvm/lib/Target/R600/R600RegisterInfo.h
+++ b/llvm/lib/Target/R600/R600RegisterInfo.h
@@ -26,6 +26,7 @@ class TargetInstrInfo;
 struct R600RegisterInfo : public AMDGPURegisterInfo {
   AMDGPUTargetMachine &TM;
   const TargetInstrInfo &TII;
+  RegClassWeight RCW;
 
   R600RegisterInfo(AMDGPUTargetMachine &tm, const TargetInstrInfo &tii);
 
@@ -48,6 +49,8 @@ struct R600RegisterInfo : public AMDGPURegisterInfo {
   /// (e.g. getSubRegFromChannel(0) -> AMDGPU::sel_x)
   unsigned getSubRegFromChannel(unsigned Channel) const;
 
+  virtual const RegClassWeight &getRegClassWeight(const TargetRegisterClass *RC) const;
+
 };
 
 } // End namespace llvm
diff --git a/llvm/test/CodeGen/R600/fabs.ll b/llvm/test/CodeGen/R600/fabs.ll
index 85f2882289f..17ac8951fb3 100644
--- a/llvm/test/CodeGen/R600/fabs.ll
+++ b/llvm/test/CodeGen/R600/fabs.ll
@@ -1,6 +1,6 @@
 ;RUN: llc < %s -march=r600 -mcpu=redwood | FileCheck %s
 
-;CHECK: MOV * T{{[0-9]+\.[XYZW], \|T[0-9]+\.[XYZW]\|}}
+;CHECK: MOV * T{{[0-9]+\.[XYZW], \|PV\.[xyzw]\|}}
 
 define void @test() {
    %r0 = call float @llvm.R600.load.input(i32 0)
diff --git a/llvm/test/CodeGen/R600/fadd.ll b/llvm/test/CodeGen/R600/fadd.ll
index 9a672329e75..821d329d615 100644
--- a/llvm/test/CodeGen/R600/fadd.ll
+++ b/llvm/test/CodeGen/R600/fadd.ll
@@ -1,7 +1,7 @@
 ; RUN: llc < %s -march=r600 -mcpu=redwood | FileCheck %s
 
 ; CHECK: @fadd_f32
-; CHECK: ADD * T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
+; CHECK: ADD * T{{[0-9]+\.[XYZW], PV\.[xyzw], PV\.[xyzw]}}
 
 define void @fadd_f32() {
    %r0 = call float @llvm.R600.load.input(i32 0)
diff --git a/llvm/test/CodeGen/R600/fdiv.ll b/llvm/test/CodeGen/R600/fdiv.ll
index 2e68e36be4d..003590bb53d 100644
--- a/llvm/test/CodeGen/R600/fdiv.ll
+++ b/llvm/test/CodeGen/R600/fdiv.ll
@@ -1,12 +1,12 @@
 ;RUN: llc < %s -march=r600 -mcpu=redwood | FileCheck %s
 
 ;CHECK: RECIP_IEEE * T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
+;CHECK: MUL_IEEE * T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
 ;CHECK: RECIP_IEEE * T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
 ;CHECK: RECIP_IEEE * T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
-;CHECK: RECIP_IEEE * T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
-;CHECK: MUL_IEEE T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
 ;CHECK: MUL_IEEE * T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
 ;CHECK: MUL_IEEE * T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
+;CHECK: RECIP_IEEE * T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
 ;CHECK: MUL_IEEE * T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
 
 define void @test(<4 x float> addrspace(1)* %out, <4 x float> addrspace(1)* %in) {
diff --git a/llvm/test/CodeGen/R600/floor.ll b/llvm/test/CodeGen/R600/floor.ll
index 877d69a65b4..0a807b1130a 100644
--- a/llvm/test/CodeGen/R600/floor.ll
+++ b/llvm/test/CodeGen/R600/floor.ll
@@ -1,6 +1,6 @@
 ;RUN: llc < %s -march=r600 -mcpu=redwood | FileCheck %s
 
-;CHECK: FLOOR * T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
+;CHECK: FLOOR * T{{[0-9]+\.[XYZW], PV\.[xyzw]}}
 
 define void @test() {
    %r0 = call float @llvm.R600.load.input(i32 0)
diff --git a/llvm/test/CodeGen/R600/fmad.ll b/llvm/test/CodeGen/R600/fmad.ll
index 62001edc3aa..861411548a1 100644
--- a/llvm/test/CodeGen/R600/fmad.ll
+++ b/llvm/test/CodeGen/R600/fmad.ll
@@ -1,6 +1,6 @@
 ;RUN: llc < %s -march=r600 -mcpu=redwood | FileCheck %s
 
-;CHECK: MULADD_IEEE * T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
+;CHECK: MULADD_IEEE * {{T[0-9]+\.[XYZW], PV\.[xyzw], PV.[xyzw], PV\.[xyzw]}}
 
 define void @test() {
    %r0 = call float @llvm.R600.load.input(i32 0)
diff --git a/llvm/test/CodeGen/R600/fmax.ll b/llvm/test/CodeGen/R600/fmax.ll
index 8b704e56484..ef3daad226a 100644
--- a/llvm/test/CodeGen/R600/fmax.ll
+++ b/llvm/test/CodeGen/R600/fmax.ll
@@ -1,6 +1,6 @@
 ;RUN: llc < %s -march=r600 -mcpu=redwood | FileCheck %s
 
-;CHECK: MAX * T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
+;CHECK: MAX * T{{[0-9]+\.[XYZW], PV\.[xyzw], PV\.[xyzw]}}
 
 define void @test() {
    %r0 = call float @llvm.R600.load.input(i32 0)
diff --git a/llvm/test/CodeGen/R600/fmin.ll b/llvm/test/CodeGen/R600/fmin.ll
index 5e34b7c8902..026481c2635 100644
--- a/llvm/test/CodeGen/R600/fmin.ll
+++ b/llvm/test/CodeGen/R600/fmin.ll
@@ -1,6 +1,6 @@
 ;RUN: llc < %s -march=r600 -mcpu=redwood | FileCheck %s
 
-;CHECK: MIN * T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
+;CHECK: MIN * T{{[0-9]+\.[XYZW], PV\.[xyzw], PV\.[xyzw]}}
 
 define void @test() {
    %r0 = call float @llvm.R600.load.input(i32 0)
diff --git a/llvm/test/CodeGen/R600/fmul.ll b/llvm/test/CodeGen/R600/fmul.ll
index c29294632dc..dbb6424e2cd 100644
--- a/llvm/test/CodeGen/R600/fmul.ll
+++ b/llvm/test/CodeGen/R600/fmul.ll
@@ -1,7 +1,7 @@
 ; RUN: llc < %s -march=r600 -mcpu=redwood | FileCheck %s
 
 ; CHECK: @fmul_f32
-; CHECK: MUL_IEEE * T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
+; CHECK: MUL_IEEE * {{T[0-9]+\.[XYZW], PV\.[xyzw], PV\.[xyzw]}}
 
 define void @fmul_f32() {
    %r0 = call float @llvm.R600.load.input(i32 0)
diff --git a/llvm/test/CodeGen/R600/fsub.ll b/llvm/test/CodeGen/R600/fsub.ll
index f784cde46cd..f88729ef352 100644
--- a/llvm/test/CodeGen/R600/fsub.ll
+++ b/llvm/test/CodeGen/R600/fsub.ll
@@ -1,7 +1,7 @@
 ; RUN: llc < %s -march=r600 -mcpu=redwood | FileCheck %s
 
 ; CHECK: @fsub_f32
-; CHECK: ADD * T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], -T[0-9]+\.[XYZW]}}
+; CHECK: ADD * T{{[0-9]+\.[XYZW], PV\.[xyzw], -PV\.[xyzw]}}
 
 define void @fsub_f32() {
    %r0 = call float @llvm.R600.load.input(i32 0)
diff --git a/llvm/test/CodeGen/R600/llvm.AMDGPU.mul.ll b/llvm/test/CodeGen/R600/llvm.AMDGPU.mul.ll
index cc0732b3fff..69fbe580cb5 100644
--- a/llvm/test/CodeGen/R600/llvm.AMDGPU.mul.ll
+++ b/llvm/test/CodeGen/R600/llvm.AMDGPU.mul.ll
@@ -1,6 +1,6 @@
 ;RUN: llc < %s -march=r600 -mcpu=redwood | FileCheck %s
 
-;CHECK: MUL NON-IEEE * T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
+;CHECK: MUL NON-IEEE * T{{[0-9]+\.[XYZW], PV\.[xyzw], PV\.[xyzw]}}
 
 define void @test() {
    %r0 = call float @llvm.R600.load.input(i32 0)
diff --git a/llvm/test/CodeGen/R600/llvm.pow.ll b/llvm/test/CodeGen/R600/llvm.pow.ll
index 91b77428290..3800abfc5c4 100644
--- a/llvm/test/CodeGen/R600/llvm.pow.ll
+++ b/llvm/test/CodeGen/R600/llvm.pow.ll
@@ -1,7 +1,7 @@
 ;RUN: llc < %s -march=r600 -mcpu=redwood | FileCheck %s
 
 ;CHECK: LOG_IEEE * T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
-;CHECK-NEXT: MUL NON-IEEE * T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
+;CHECK: MUL NON-IEEE * T{{[0-9]+\.[XYZW], PV\.[xyzw], T[0-9]+\.[XYZW]}}
 ;CHECK-NEXT: EXP_IEEE * T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
 
 define void @test() {
diff --git a/llvm/test/CodeGen/R600/pv.ll b/llvm/test/CodeGen/R600/pv.ll
index 37c3d9d7d6d..5481d6d0f54 100644
--- a/llvm/test/CodeGen/R600/pv.ll
+++ b/llvm/test/CodeGen/R600/pv.ll
@@ -1,7 +1,7 @@
 ; RUN: llc < %s -march=r600 | FileCheck %s
 
 ;CHECK: DOT4 * T{{[0-9]\.W}} (MASKED)
-;CHECK-NEXT: CNDGE T{{[0-9].[XYZW]}}, PV.x
+;CHECK: CNDGE T{{[0-9].[XYZW]}}, PV.x
 
 define void @main() #0 {
 main_body:
diff --git a/llvm/test/CodeGen/R600/r600-encoding.ll b/llvm/test/CodeGen/R600/r600-encoding.ll
index c8040a1b4cd..6ef3c3124b8 100644
--- a/llvm/test/CodeGen/R600/r600-encoding.ll
+++ b/llvm/test/CodeGen/R600/r600-encoding.ll
@@ -5,10 +5,10 @@
 ; the VLIW4/5 GPUs.
 
 ; EG-CHECK: @test
-; EG-CHECK: MUL_IEEE {{[ *TXYZW.,0-9]+}} ; encoding: [{{0x[0-9a-f]+,0x[0-9a-f]+,0x[0-9a-f]+,0x[0-9a-f]+,0x10,0x01,0x[0-9a-f]+,0x[0-9a-f]+}}]
+; EG-CHECK: MUL_IEEE {{[ *TXYZWPVxyzw.,0-9]+}} ; encoding: [{{0x[0-9a-f]+,0x[0-9a-f]+,0x[0-9a-f]+,0x[0-9a-f]+,0x10,0x01,0x[0-9a-f]+,0x[0-9a-f]+}}]
 
 ; R600-CHECK: @test
-; R600-CHECK: MUL_IEEE {{[ *TXYZW.,0-9]+}} ; encoding: [{{0x[0-9a-f]+,0x[0-9a-f]+,0x[0-9a-f]+,0x[0-9a-f]+,0x10,0x02,0x[0-9a-f]+,0x[0-9a-f]+}}]
+; R600-CHECK: MUL_IEEE {{[ *TXYZWPVxyzw.,0-9]+}} ; encoding: [{{0x[0-9a-f]+,0x[0-9a-f]+,0x[0-9a-f]+,0x[0-9a-f]+,0x10,0x02,0x[0-9a-f]+,0x[0-9a-f]+}}]
 
 define void @test() {
 entry:
diff --git a/llvm/test/CodeGen/R600/selectcc-opt.ll b/llvm/test/CodeGen/R600/selectcc-opt.ll
index 02d93539042..7f568fc0e67 100644
--- a/llvm/test/CodeGen/R600/selectcc-opt.ll
+++ b/llvm/test/CodeGen/R600/selectcc-opt.ll
@@ -29,8 +29,10 @@ ENDIF:
 ; for the icmp instruction
 
 ; CHECK: @test_b
+; CHECK: VTX_READ
 ; CHECK: SET{{[GTEQN]+}}_DX10
 ; CHECK-NEXT: PRED_
+; CHECK-NEXT: ALU clause starting
 define void @test_b(i32 addrspace(1)* %out, float %in) {
 entry:
   %0 = fcmp ult float %in, 0.0
diff --git a/llvm/test/CodeGen/R600/vselect.ll b/llvm/test/CodeGen/R600/vselect.ll
index 6e459df847e..a74a114a331 100644
--- a/llvm/test/CodeGen/R600/vselect.ll
+++ b/llvm/test/CodeGen/R600/vselect.ll
@@ -3,8 +3,8 @@
 ; CHECK: @test_select_v4i32
 ; CHECK: CNDE_INT T{{[0-9]+\.[XYZW], PV\.[xyzw], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
 ; CHECK: CNDE_INT * T{{[0-9]+\.[XYZW], PV\.[xyzw], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
-; CHECK: CNDE_INT * T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
-; CHECK: CNDE_INT * T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
+; CHECK: CNDE_INT T{{[0-9]+\.[XYZW], PV\.[xyzw], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
+; CHECK: CNDE_INT * T{{[0-9]+\.[XYZW], PV\.[xyzw], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
 
 define void @test_select_v4i32(<4 x i32> addrspace(1)* %out, <4 x i32> addrspace(1)* %in0, <4 x i32> addrspace(1)* %in1) {
 entry: