R600/SI: Use eliminateFrameIndex() to expand SGPR spill pseudos

This will simplify the SGPR spilling and also allow us to use
MachineFrameInfo for calculating offsets, which should be more
reliable than our custom code.

This fixes a crash in some cases where a register would be spilled
in a branch such that the VGPR defined for spilling did not dominate
all the uses when restoring.

This fixes a crash in an ocl conformance test.  The test requries
register spilling and is too big to include.

llvm-svn: 216217

5 files changed, 159 insertions, 112 deletions

diff --git a/llvm/lib/Target/R600/SIInstrInfo.cpp b/llvm/lib/Target/R600/SIInstrInfo.cpp
index 3868f63b3a8..3be1b25b363 100644
--- a/llvm/lib/Target/R600/SIInstrInfo.cpp
+++ b/llvm/lib/Target/R600/SIInstrInfo.cpp
@@ -17,6 +17,7 @@
 #include "AMDGPUTargetMachine.h"
 #include "SIDefines.h"
 #include "SIMachineFunctionInfo.h"
+#include "llvm/CodeGen/MachineFrameInfo.h"
 #include "llvm/CodeGen/MachineInstrBuilder.h"
 #include "llvm/CodeGen/MachineRegisterInfo.h"
 #include "llvm/IR/Function.h"
@@ -401,38 +402,21 @@ void SIInstrInfo::storeRegToStackSlot(MachineBasicBlock &MBB,
                                       const TargetRegisterClass *RC,
                                       const TargetRegisterInfo *TRI) const {
   MachineFunction *MF = MBB.getParent();
-  SIMachineFunctionInfo *MFI = MF->getInfo<SIMachineFunctionInfo>();
-  MachineRegisterInfo &MRI = MF->getRegInfo();
+  MachineFrameInfo *FrameInfo = MF->getFrameInfo();
   DebugLoc DL = MBB.findDebugLoc(MI);
-  unsigned KillFlag = isKill ? RegState::Kill : 0;
 
   if (RI.hasVGPRs(RC)) {
     LLVMContext &Ctx = MF->getFunction()->getContext();
     Ctx.emitError("SIInstrInfo::storeRegToStackSlot - Can't spill VGPR!");
     BuildMI(MBB, MI, DL, get(AMDGPU::V_MOV_B32_e32), AMDGPU::VGPR0)
             .addReg(SrcReg);
-  } else if (TRI->getCommonSubClass(RC, &AMDGPU::SGPR_32RegClass)) {
-    unsigned Lane = MFI->SpillTracker.reserveLanes(MRI, MF);
-    unsigned TgtReg = MFI->SpillTracker.LaneVGPR;
-
-    BuildMI(MBB, MI, DL, get(AMDGPU::V_WRITELANE_B32), TgtReg)
-            .addReg(SrcReg, KillFlag)
-            .addImm(Lane);
-    MFI->SpillTracker.addSpilledReg(FrameIndex, TgtReg, Lane);
   } else if (RI.isSGPRClass(RC)) {
     // We are only allowed to create one new instruction when spilling
-    // registers, so we need to use pseudo instruction for vector
-    // registers.
-    //
-    // Reserve a spot in the spill tracker for each sub-register of
-    // the vector register.
-    unsigned NumSubRegs = RC->getSize() / 4;
-    unsigned FirstLane = MFI->SpillTracker.reserveLanes(MRI, MF, NumSubRegs);
-    MFI->SpillTracker.addSpilledReg(FrameIndex, MFI->SpillTracker.LaneVGPR,
-                                    FirstLane);
-
+    // registers, so we need to use pseudo instruction for spilling
+    // SGPRs.
     unsigned Opcode;
     switch (RC->getSize() * 8) {
+    case 32:  Opcode = AMDGPU::SI_SPILL_S32_SAVE;  break;
     case 64:  Opcode = AMDGPU::SI_SPILL_S64_SAVE;  break;
     case 128: Opcode = AMDGPU::SI_SPILL_S128_SAVE; break;
     case 256: Opcode = AMDGPU::SI_SPILL_S256_SAVE; break;
@@ -440,9 +424,10 @@ void SIInstrInfo::storeRegToStackSlot(MachineBasicBlock &MBB,
     default: llvm_unreachable("Cannot spill register class");
     }
 
-    BuildMI(MBB, MI, DL, get(Opcode), MFI->SpillTracker.LaneVGPR)
+    FrameInfo->setObjectAlignment(FrameIndex, 4);
+    BuildMI(MBB, MI, DL, get(Opcode))
             .addReg(SrcReg)
-            .addImm(FrameIndex);
+            .addFrameIndex(FrameIndex);
   } else {
     llvm_unreachable("VGPR spilling not supported");
   }
@@ -454,7 +439,7 @@ void SIInstrInfo::loadRegFromStackSlot(MachineBasicBlock &MBB,
                                        const TargetRegisterClass *RC,
                                        const TargetRegisterInfo *TRI) const {
   MachineFunction *MF = MBB.getParent();
-  SIMachineFunctionInfo *MFI = MF->getInfo<SIMachineFunctionInfo>();
+  MachineFrameInfo *FrameInfo = MF->getFrameInfo();
   DebugLoc DL = MBB.findDebugLoc(MI);
 
   if (RI.hasVGPRs(RC)) {
@@ -473,38 +458,14 @@ void SIInstrInfo::loadRegFromStackSlot(MachineBasicBlock &MBB,
     default: llvm_unreachable("Cannot spill register class");
     }
 
-    SIMachineFunctionInfo::SpilledReg Spill =
-        MFI->SpillTracker.getSpilledReg(FrameIndex);
-
+    FrameInfo->setObjectAlignment(FrameIndex, 4);
     BuildMI(MBB, MI, DL, get(Opcode), DestReg)
-            .addReg(Spill.VGPR)
-            .addImm(FrameIndex);
+            .addFrameIndex(FrameIndex);
   } else {
     llvm_unreachable("VGPR spilling not supported");
   }
 }
 
-static unsigned getNumSubRegsForSpillOp(unsigned Op) {
-
-  switch (Op) {
-  case AMDGPU::SI_SPILL_S512_SAVE:
-  case AMDGPU::SI_SPILL_S512_RESTORE:
-    return 16;
-  case AMDGPU::SI_SPILL_S256_SAVE:
-  case AMDGPU::SI_SPILL_S256_RESTORE:
-    return 8;
-  case AMDGPU::SI_SPILL_S128_SAVE:
-  case AMDGPU::SI_SPILL_S128_RESTORE:
-    return 4;
-  case AMDGPU::SI_SPILL_S64_SAVE:
-  case AMDGPU::SI_SPILL_S64_RESTORE:
-    return 2;
-  case AMDGPU::SI_SPILL_S32_RESTORE:
-    return 1;
-  default: llvm_unreachable("Invalid spill opcode");
-  }
-}
-
 void SIInstrInfo::insertNOPs(MachineBasicBlock::iterator MI,
                              int Count) const {
   while (Count > 0) {
@@ -520,59 +481,11 @@ void SIInstrInfo::insertNOPs(MachineBasicBlock::iterator MI,
 }
 
 bool SIInstrInfo::expandPostRAPseudo(MachineBasicBlock::iterator MI) const {
-  SIMachineFunctionInfo *MFI =
-      MI->getParent()->getParent()->getInfo<SIMachineFunctionInfo>();
   MachineBasicBlock &MBB = *MI->getParent();
   DebugLoc DL = MBB.findDebugLoc(MI);
   switch (MI->getOpcode()) {
   default: return AMDGPUInstrInfo::expandPostRAPseudo(MI);
 
-  // SGPR register spill
-  case AMDGPU::SI_SPILL_S512_SAVE:
-  case AMDGPU::SI_SPILL_S256_SAVE:
-  case AMDGPU::SI_SPILL_S128_SAVE:
-  case AMDGPU::SI_SPILL_S64_SAVE: {
-    unsigned NumSubRegs = getNumSubRegsForSpillOp(MI->getOpcode());
-    unsigned FrameIndex = MI->getOperand(2).getImm();
-
-    for (unsigned i = 0, e = NumSubRegs; i < e; ++i) {
-      SIMachineFunctionInfo::SpilledReg Spill;
-      unsigned SubReg = RI.getPhysRegSubReg(MI->getOperand(1).getReg(),
-                                            &AMDGPU::SGPR_32RegClass, i);
-      Spill = MFI->SpillTracker.getSpilledReg(FrameIndex);
-
-      BuildMI(MBB, MI, DL, get(AMDGPU::V_WRITELANE_B32),
-              MI->getOperand(0).getReg())
-              .addReg(SubReg)
-              .addImm(Spill.Lane + i);
-    }
-    MI->eraseFromParent();
-    break;
-  }
-
-  // SGPR register restore
-  case AMDGPU::SI_SPILL_S512_RESTORE:
-  case AMDGPU::SI_SPILL_S256_RESTORE:
-  case AMDGPU::SI_SPILL_S128_RESTORE:
-  case AMDGPU::SI_SPILL_S64_RESTORE:
-  case AMDGPU::SI_SPILL_S32_RESTORE: {
-    unsigned NumSubRegs = getNumSubRegsForSpillOp(MI->getOpcode());
-
-    for (unsigned i = 0, e = NumSubRegs; i < e; ++i) {
-      SIMachineFunctionInfo::SpilledReg Spill;
-      unsigned FrameIndex = MI->getOperand(2).getImm();
-      unsigned SubReg = RI.getPhysRegSubReg(MI->getOperand(0).getReg(),
-                                   &AMDGPU::SGPR_32RegClass, i);
-      Spill = MFI->SpillTracker.getSpilledReg(FrameIndex);
-
-      BuildMI(MBB, MI, DL, get(AMDGPU::V_READLANE_B32), SubReg)
-              .addReg(MI->getOperand(1).getReg())
-              .addImm(Spill.Lane + i);
-    }
-    insertNOPs(MI, 3);
-    MI->eraseFromParent();
-    break;
-  }
   case AMDGPU::SI_CONSTDATA_PTR: {
     unsigned Reg = MI->getOperand(0).getReg();
     unsigned RegLo = RI.getSubReg(Reg, AMDGPU::sub0);
diff --git a/llvm/lib/Target/R600/SIInstructions.td b/llvm/lib/Target/R600/SIInstructions.td
index fd011a16067..01db459e3f2 100644
--- a/llvm/lib/Target/R600/SIInstructions.td
+++ b/llvm/lib/Target/R600/SIInstructions.td
@@ -1729,14 +1729,14 @@ def V_SUB_F64 : InstSI <
 multiclass SI_SPILL_SGPR <RegisterClass sgpr_class> {
 
   def _SAVE : InstSI <
-    (outs VReg_32:$dst),
+    (outs),
     (ins sgpr_class:$src, i32imm:$frame_idx),
     "", []
   >;
 
   def _RESTORE : InstSI <
     (outs sgpr_class:$dst),
-    (ins VReg_32:$src, i32imm:$frame_idx),
+    (ins i32imm:$frame_idx),
     "", []
   >;
 
diff --git a/llvm/lib/Target/R600/SIMachineFunctionInfo.cpp b/llvm/lib/Target/R600/SIMachineFunctionInfo.cpp
index c53a7e10d54..086f0908703 100644
--- a/llvm/lib/Target/R600/SIMachineFunctionInfo.cpp
+++ b/llvm/lib/Target/R600/SIMachineFunctionInfo.cpp
@@ -12,6 +12,7 @@
 #include "SIMachineFunctionInfo.h"
 #include "SIInstrInfo.h"
 #include "SIRegisterInfo.h"
+#include "llvm/CodeGen/MachineFrameInfo.h"
 #include "llvm/CodeGen/MachineRegisterInfo.h"
 #include "llvm/IR/Function.h"
 #include "llvm/IR/LLVMContext.h"
@@ -90,8 +91,49 @@ void SIMachineFunctionInfo::RegSpillTracker::addSpilledReg(unsigned FrameIndex,
                                                            int Lane) {
   SpilledRegisters[FrameIndex] = SpilledReg(Reg, Lane);
 }
+/// \brief Returns a register that is not used at any point in the function.
+///        If all registers are used, then this function will return
+//         AMDGPU::NoRegister.
+static unsigned findUnusedVGPR(const MachineRegisterInfo &MRI) {
 
-const SIMachineFunctionInfo::SpilledReg&
-SIMachineFunctionInfo::RegSpillTracker::getSpilledReg(unsigned FrameIndex) {
-  return SpilledRegisters[FrameIndex];
+  const TargetRegisterClass *RC = &AMDGPU::VGPR_32RegClass;
+
+  for (TargetRegisterClass::iterator I = RC->begin(), E = RC->end();
+       I != E; ++I) {
+    if (!MRI.isPhysRegUsed(*I))
+      return *I;
+  }
+  return AMDGPU::NoRegister;
+}
+
+SIMachineFunctionInfo::SpilledReg SIMachineFunctionInfo::getSpilledReg(
+                                                       MachineFunction *MF,
+                                                       unsigned FrameIndex,
+                                                       unsigned SubIdx) {
+  const MachineFrameInfo *FrameInfo = MF->getFrameInfo();
+  MachineRegisterInfo &MRI = MF->getRegInfo();
+  int64_t Offset = FrameInfo->getObjectOffset(FrameIndex);
+  Offset += SubIdx * 4;
+
+  unsigned LaneVGPRIdx = Offset / (64 * 4);
+  unsigned Lane = (Offset / 4) % 64;
+
+  struct SpilledReg Spill;
+
+  if (!LaneVGPRs.count(LaneVGPRIdx)) {
+    unsigned LaneVGPR = findUnusedVGPR(MRI);
+    LaneVGPRs[LaneVGPRIdx] = LaneVGPR;
+    MRI.setPhysRegUsed(LaneVGPR);
+
+    // Add this register as live-in to all blocks to avoid machine verifer
+    // complaining about use of an undefined physical register.
+    for (MachineFunction::iterator BI = MF->begin(), BE = MF->end();
+         BI != BE; ++BI) {
+      BI->addLiveIn(LaneVGPR);
+    }
+  }
+
+  Spill.VGPR = LaneVGPRs[LaneVGPRIdx];
+  Spill.Lane = Lane;
+  return Spill;
 }
diff --git a/llvm/lib/Target/R600/SIMachineFunctionInfo.h b/llvm/lib/Target/R600/SIMachineFunctionInfo.h
index 707c41aba71..fcaa23a619b 100644
--- a/llvm/lib/Target/R600/SIMachineFunctionInfo.h
+++ b/llvm/lib/Target/R600/SIMachineFunctionInfo.h
@@ -57,9 +57,12 @@ public:
   // SIMachineFunctionInfo definition
 
   SIMachineFunctionInfo(const MachineFunction &MF);
+  SpilledReg getSpilledReg(MachineFunction *MF, unsigned FrameIndex,
+                           unsigned SubIdx);
   unsigned PSInputAddr;
   struct RegSpillTracker SpillTracker;
   unsigned NumUserSGPRs;
+  std::map<unsigned, unsigned> LaneVGPRs;
 };
 
 } // End namespace llvm
diff --git a/llvm/lib/Target/R600/SIRegisterInfo.cpp b/llvm/lib/Target/R600/SIRegisterInfo.cpp
index 531c5047f50..8663df88922 100644
--- a/llvm/lib/Target/R600/SIRegisterInfo.cpp
+++ b/llvm/lib/Target/R600/SIRegisterInfo.cpp
@@ -20,6 +20,8 @@
 #include "llvm/CodeGen/MachineFrameInfo.h"
 #include "llvm/CodeGen/MachineInstrBuilder.h"
 #include "llvm/CodeGen/RegisterScavenging.h"
+#include "llvm/IR/Function.h"
+#include "llvm/IR/LLVMContext.h"
 
 using namespace llvm;
 
@@ -43,23 +45,110 @@ bool SIRegisterInfo::requiresRegisterScavenging(const MachineFunction &Fn) const
   return Fn.getFrameInfo()->hasStackObjects();
 }
 
+static unsigned getNumSubRegsForSpillOp(unsigned Op) {
+
+  switch (Op) {
+  case AMDGPU::SI_SPILL_S512_SAVE:
+  case AMDGPU::SI_SPILL_S512_RESTORE:
+    return 16;
+  case AMDGPU::SI_SPILL_S256_SAVE:
+  case AMDGPU::SI_SPILL_S256_RESTORE:
+    return 8;
+  case AMDGPU::SI_SPILL_S128_SAVE:
+  case AMDGPU::SI_SPILL_S128_RESTORE:
+    return 4;
+  case AMDGPU::SI_SPILL_S64_SAVE:
+  case AMDGPU::SI_SPILL_S64_RESTORE:
+    return 2;
+  case AMDGPU::SI_SPILL_S32_SAVE:
+  case AMDGPU::SI_SPILL_S32_RESTORE:
+    return 1;
+  default: llvm_unreachable("Invalid spill opcode");
+  }
+}
+
 void SIRegisterInfo::eliminateFrameIndex(MachineBasicBlock::iterator MI,
                                         int SPAdj, unsigned FIOperandNum,
                                         RegScavenger *RS) const {
   MachineFunction *MF = MI->getParent()->getParent();
+  MachineBasicBlock *MBB = MI->getParent();
+  SIMachineFunctionInfo *MFI = MF->getInfo<SIMachineFunctionInfo>();
   MachineFrameInfo *FrameInfo = MF->getFrameInfo();
   const SIInstrInfo *TII = static_cast<const SIInstrInfo*>(ST.getInstrInfo());
+  DebugLoc DL = MI->getDebugLoc();
+
   MachineOperand &FIOp = MI->getOperand(FIOperandNum);
   int Index = MI->getOperand(FIOperandNum).getIndex();
-  int64_t Offset = FrameInfo->getObjectOffset(Index);
-
-  FIOp.ChangeToImmediate(Offset);
-  if (!TII->isImmOperandLegal(MI, FIOperandNum, FIOp)) {
-    unsigned TmpReg = RS->scavengeRegister(&AMDGPU::VReg_32RegClass, MI, SPAdj);
-    BuildMI(*MI->getParent(), MI, MI->getDebugLoc(),
-            TII->get(AMDGPU::V_MOV_B32_e32), TmpReg)
-            .addImm(Offset);
-    FIOp.ChangeToRegister(TmpReg, false);
+
+  switch (MI->getOpcode()) {
+    // SGPR register spill
+    case AMDGPU::SI_SPILL_S512_SAVE:
+    case AMDGPU::SI_SPILL_S256_SAVE:
+    case AMDGPU::SI_SPILL_S128_SAVE:
+    case AMDGPU::SI_SPILL_S64_SAVE:
+    case AMDGPU::SI_SPILL_S32_SAVE: {
+      unsigned NumSubRegs = getNumSubRegsForSpillOp(MI->getOpcode());
+
+      for (unsigned i = 0, e = NumSubRegs; i < e; ++i) {
+        unsigned SubReg = getPhysRegSubReg(MI->getOperand(0).getReg(),
+                                           &AMDGPU::SGPR_32RegClass, i);
+        struct SIMachineFunctionInfo::SpilledReg Spill =
+            MFI->getSpilledReg(MF, Index, i);
+
+        if (Spill.VGPR == AMDGPU::NoRegister) {
+           LLVMContext &Ctx = MF->getFunction()->getContext();
+           Ctx.emitError("Ran out of VGPRs for spilling SGPR");
+        }
+
+        BuildMI(*MBB, MI, DL, TII->get(AMDGPU::V_WRITELANE_B32), Spill.VGPR)
+                .addReg(SubReg)
+                .addImm(Spill.Lane);
+
+      }
+      MI->eraseFromParent();
+      break;
+    }
+
+    // SGPR register restore
+    case AMDGPU::SI_SPILL_S512_RESTORE:
+    case AMDGPU::SI_SPILL_S256_RESTORE:
+    case AMDGPU::SI_SPILL_S128_RESTORE:
+    case AMDGPU::SI_SPILL_S64_RESTORE:
+    case AMDGPU::SI_SPILL_S32_RESTORE: {
+      unsigned NumSubRegs = getNumSubRegsForSpillOp(MI->getOpcode());
+
+      for (unsigned i = 0, e = NumSubRegs; i < e; ++i) {
+        unsigned SubReg = getPhysRegSubReg(MI->getOperand(0).getReg(),
+                                           &AMDGPU::SGPR_32RegClass, i);
+        struct SIMachineFunctionInfo::SpilledReg Spill =
+            MFI->getSpilledReg(MF, Index, i);
+
+        if (Spill.VGPR == AMDGPU::NoRegister) {
+           LLVMContext &Ctx = MF->getFunction()->getContext();
+           Ctx.emitError("Ran out of VGPRs for spilling SGPR");
+        }
+
+        BuildMI(*MBB, MI, DL, TII->get(AMDGPU::V_READLANE_B32), SubReg)
+                .addReg(Spill.VGPR)
+                .addImm(Spill.Lane);
+
+      }
+      TII->insertNOPs(MI, 3);
+      MI->eraseFromParent();
+      break;
+    }
+
+    default: {
+      int64_t Offset = FrameInfo->getObjectOffset(Index);
+      FIOp.ChangeToImmediate(Offset);
+      if (!TII->isImmOperandLegal(MI, FIOperandNum, FIOp)) {
+        unsigned TmpReg = RS->scavengeRegister(&AMDGPU::VReg_32RegClass, MI, SPAdj);
+        BuildMI(*MBB, MI, MI->getDebugLoc(),
+                TII->get(AMDGPU::V_MOV_B32_e32), TmpReg)
+                .addImm(Offset);
+        FIOp.ChangeToRegister(TmpReg, false);
+      }
+    }
   }
 }