Emit callee-saved regs spills / restores

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@75943 91177308-0d34-0410-b5e6-96231b3b80d8

8 files changed, 190 insertions, 38 deletions

diff --git a/lib/Target/SystemZ/SystemZInstrInfo.cpp b/lib/Target/SystemZ/SystemZInstrInfo.cpp
index 53f8d29..335d460 100644
--- a/lib/Target/SystemZ/SystemZInstrInfo.cpp
+++ b/lib/Target/SystemZ/SystemZInstrInfo.cpp
@@ -26,7 +26,30 @@ using namespace llvm;
 
 SystemZInstrInfo::SystemZInstrInfo(SystemZTargetMachine &tm)
   : TargetInstrInfoImpl(SystemZInsts, array_lengthof(SystemZInsts)),
-    RI(tm, *this), TM(tm) {}
+    RI(tm, *this), TM(tm) {
+  // Fill the spill offsets map
+  static const unsigned SpillOffsTab[][2] = {
+    { SystemZ::R2D,  0x10 },
+    { SystemZ::R3D,  0x18 },
+    { SystemZ::R4D,  0x20 },
+    { SystemZ::R5D,  0x28 },
+    { SystemZ::R6D,  0x30 },
+    { SystemZ::R7D,  0x38 },
+    { SystemZ::R8D,  0x40 },
+    { SystemZ::R9D,  0x48 },
+    { SystemZ::R10D, 0x50 },
+    { SystemZ::R11D, 0x58 },
+    { SystemZ::R12D, 0x60 },
+    { SystemZ::R13D, 0x68 },
+    { SystemZ::R14D, 0x70 },
+    { SystemZ::R15D, 0x78 }
+  };
+
+  RegSpillOffsets.grow(SystemZ::NUM_TARGET_REGS);
+
+  for (unsigned i = 0, e = array_lengthof(SpillOffsTab); i != e; ++i)
+    RegSpillOffsets[SpillOffsTab[i][0]] = SpillOffsTab[i][1];
+}
 
 void SystemZInstrInfo::storeRegToStackSlot(MachineBasicBlock &MBB,
                                           MachineBasicBlock::iterator MI,
@@ -117,13 +140,52 @@ bool
 SystemZInstrInfo::spillCalleeSavedRegisters(MachineBasicBlock &MBB,
                                            MachineBasicBlock::iterator MI,
                                 const std::vector<CalleeSavedInfo> &CSI) const {
-  if (CSI.empty())
-    return false;
+  DebugLoc DL = DebugLoc::getUnknownLoc();
+  if (MI != MBB.end()) DL = MI->getDebugLoc();
 
   MachineFunction &MF = *MBB.getParent();
   SystemZMachineFunctionInfo *MFI = MF.getInfo<SystemZMachineFunctionInfo>();
   MFI->setCalleeSavedFrameSize(CSI.size() * 8);
 
+  // Scan the callee-saved and find the bounds of register spill area.
+  unsigned LowReg = 0, HighReg = 0, StartOffset = -1U, EndOffset = 0;
+  for (unsigned i = 0, e = CSI.size(); i != e; ++i) {
+    unsigned Reg = CSI[i].getReg();
+    unsigned Offset = RegSpillOffsets[Reg];
+    if (StartOffset > Offset) {
+      LowReg = Reg; StartOffset = Offset;
+    }
+    if (EndOffset < Offset) {
+      HighReg = Reg; EndOffset = RegSpillOffsets[Reg];
+    }
+  }
+
+  // Save information for epilogue inserter.
+  MFI->setLowReg(LowReg); MFI->setHighReg(HighReg);
+
+  // Build a store instruction. Use STORE MULTIPLE instruction if there are many
+  // registers to store, otherwise - just STORE.
+  MachineInstrBuilder MIB =
+    BuildMI(MBB, MI, DL, get((LowReg == HighReg ?
+                              SystemZ::MOV64mr : SystemZ::MOV64mrm)));
+
+  // Add store operands.
+  MIB.addReg(SystemZ::R15D).addImm(StartOffset);
+  if (LowReg == HighReg)
+    MIB.addReg(0);
+  MIB.addReg(LowReg, RegState::Kill);
+  if (LowReg != HighReg)
+    MIB.addReg(HighReg, RegState::Kill);
+
+  // Do a second scan adding regs as being killed by instruction
+  for (unsigned i = 0, e = CSI.size(); i != e; ++i) {
+    unsigned Reg = CSI[i].getReg();
+    // Add the callee-saved register as live-in. It's killed at the spill.
+    MBB.addLiveIn(Reg);
+    if (Reg != LowReg && Reg != HighReg)
+      MIB.addReg(Reg, RegState::ImplicitKill);
+  }
+
   return true;
 }
 
@@ -131,6 +193,41 @@ bool
 SystemZInstrInfo::restoreCalleeSavedRegisters(MachineBasicBlock &MBB,
                                              MachineBasicBlock::iterator MI,
                                 const std::vector<CalleeSavedInfo> &CSI) const {
+  if (CSI.empty())
+    return false;
+
+  DebugLoc DL = DebugLoc::getUnknownLoc();
+  if (MI != MBB.end()) DL = MI->getDebugLoc();
+
+  MachineFunction &MF = *MBB.getParent();
+  const TargetRegisterInfo *RegInfo= MF.getTarget().getRegisterInfo();
+  SystemZMachineFunctionInfo *MFI = MF.getInfo<SystemZMachineFunctionInfo>();
+
+  unsigned LowReg = MFI->getLowReg(), HighReg = MFI->getHighReg();
+  unsigned StartOffset = RegSpillOffsets[LowReg];
+
+  // Build a load instruction. Use LOAD MULTIPLE instruction if there are many
+  // registers to load, otherwise - just LOAD.
+  MachineInstrBuilder MIB =
+    BuildMI(MBB, MI, DL, get((LowReg == HighReg ?
+                              SystemZ::MOV64rm : SystemZ::MOV64rmm)));
+  // Add store operands.
+  MIB.addReg(LowReg, RegState::Define);
+  if (LowReg != HighReg)
+    MIB.addReg(HighReg, RegState::Define);
+
+  MIB.addReg((RegInfo->hasFP(MF) ? SystemZ::R11D : SystemZ::R15D));
+  MIB.addImm(StartOffset);
+  if (LowReg == HighReg)
+    MIB.addReg(0);
+
+  // Do a second scan adding regs as being defined by instruction
+  for (unsigned i = 0, e = CSI.size(); i != e; ++i) {
+    unsigned Reg = CSI[i].getReg();
+    if (Reg != LowReg && Reg != HighReg)
+      MIB.addReg(Reg, RegState::ImplicitDefine);
+  }
+
   return true;
 }
 
diff --git a/lib/Target/SystemZ/SystemZInstrInfo.h b/lib/Target/SystemZ/SystemZInstrInfo.h
index 9f46135..1e0cc82 100644
--- a/lib/Target/SystemZ/SystemZInstrInfo.h
+++ b/lib/Target/SystemZ/SystemZInstrInfo.h
@@ -14,8 +14,9 @@
 #ifndef LLVM_TARGET_SYSTEMZINSTRINFO_H
 #define LLVM_TARGET_SYSTEMZINSTRINFO_H
 
-#include "llvm/Target/TargetInstrInfo.h"
 #include "SystemZRegisterInfo.h"
+#include "llvm/ADT/IndexedMap.h"
+#include "llvm/Target/TargetInstrInfo.h"
 
 namespace llvm {
 
@@ -24,6 +25,7 @@ class SystemZTargetMachine;
 class SystemZInstrInfo : public TargetInstrInfoImpl {
   const SystemZRegisterInfo RI;
   SystemZTargetMachine &TM;
+  IndexedMap<unsigned> RegSpillOffsets;
 public:
   explicit SystemZInstrInfo(SystemZTargetMachine &TM);
 
diff --git a/lib/Target/SystemZ/SystemZInstrInfo.td b/lib/Target/SystemZ/SystemZInstrInfo.td
index decc926..007df01 100644
--- a/lib/Target/SystemZ/SystemZInstrInfo.td
+++ b/lib/Target/SystemZ/SystemZInstrInfo.td
@@ -195,19 +195,12 @@ def laaddr : Operand<i64>,
 //===----------------------------------------------------------------------===//
 // Instruction list..
 
-// ADJCALLSTACKDOWN/UP implicitly use/def SP because they may be expanded into
-// a stack adjustment and the codegen must know that they may modify the stack
-// pointer before prolog-epilog rewriting occurs.
-// Pessimistically assume ADJCALLSTACKDOWN / ADJCALLSTACKUP will become
-// sub / add which can clobber R15D.
-let Defs = [R15D], Uses = [R15D] in {
 def ADJCALLSTACKDOWN : Pseudo<(outs), (ins i64imm:$amt),
                               "#ADJCALLSTACKDOWN",
                               [(SystemZcallseq_start timm:$amt)]>;
 def ADJCALLSTACKUP   : Pseudo<(outs), (ins i64imm:$amt1, i64imm:$amt2),
                               "#ADJCALLSTACKUP",
                               [(SystemZcallseq_end timm:$amt1, timm:$amt2)]>;
-}
 
 
 //===----------------------------------------------------------------------===//
@@ -215,7 +208,7 @@ def ADJCALLSTACKUP   : Pseudo<(outs), (ins i64imm:$amt1, i64imm:$amt2),
 //
 
 // FIXME: Provide proper encoding!
-let isReturn = 1, isTerminator = 1, Uses = [R14D] in {
+let isReturn = 1, isTerminator = 1 in {
   def RET : Pseudo<(outs), (ins), "br\t%r14", [(SystemZretflag)]>;
 }
 
@@ -224,12 +217,9 @@ let isReturn = 1, isTerminator = 1, Uses = [R14D] in {
 //
 
 let isCall = 1 in
-  // All calls clobber the non-callee saved registers. R15 is marked as
-  // a use to prevent stack-pointer assignments that appear immediately
-  // before calls from potentially appearing dead. Uses for argument
-  // registers are added manually.
-  let Defs = [R0D, R1D, R3D, R4D, R5D, R14D, R15D],
-      Uses = [R15D] in {
+  // All calls clobber the non-callee saved registers (except R14 which we
+  // handle separately). Uses for argument registers are added manually.
+  let Defs = [R0D, R1D, R3D, R4D, R5D] in {
     def CALLi     : Pseudo<(outs), (ins i64imm:$dst, variable_ops),
                            "brasl\t%r14, $dst", [(SystemZcall imm:$dst)]>;
     def CALLr     : Pseudo<(outs), (ins ADDR64:$dst, variable_ops),
@@ -370,6 +360,22 @@ def MOV64m32r : Pseudo<(outs), (ins rriaddr:$dst, GR64:$src),
                        "sty\t{$src, $dst}",
                        [(truncstorei32 GR64:$src, rriaddr:$dst)]>;
 
+// multiple regs moves
+// FIXME: should we use multiple arg nodes?
+def MOV32mrm  : Pseudo<(outs), (ins riaddr:$dst, GR32:$from, GR32:$to),
+                       "stmy\t{$from, $to, $dst}",
+                       []>;
+def MOV64mrm  : Pseudo<(outs), (ins riaddr:$dst, GR64:$from, GR64:$to),
+                       "stmg\t{$from, $to, $dst}",
+                       []>;
+def MOV32rmm  : Pseudo<(outs GR32:$from, GR32:$to), (ins riaddr:$dst),
+                       "lmy\t{$from, $to, $dst}",
+                       []>;
+def MOV64rmm  : Pseudo<(outs GR64:$from, GR64:$to), (ins riaddr:$dst),
+                       "lmg\t{$from, $to, $dst}",
+                       []>;
+
+
 //===----------------------------------------------------------------------===//
 // Arithmetic Instructions
 
diff --git a/lib/Target/SystemZ/SystemZMachineFunctionInfo.h b/lib/Target/SystemZ/SystemZMachineFunctionInfo.h
index 9b45b03..e47d419 100644
--- a/lib/Target/SystemZ/SystemZMachineFunctionInfo.h
+++ b/lib/Target/SystemZ/SystemZMachineFunctionInfo.h
@@ -25,6 +25,11 @@ class SystemZMachineFunctionInfo : public MachineFunctionInfo {
   /// stack frame in bytes.
   unsigned CalleeSavedFrameSize;
 
+  /// LowReg - Low register of range of callee-saved registers to store.
+  unsigned LowReg;
+
+  /// HighReg - High register of range of callee-saved registers to store.
+  unsigned HighReg;
 public:
   SystemZMachineFunctionInfo() : CalleeSavedFrameSize(0) {}
 
@@ -32,6 +37,12 @@ public:
 
   unsigned getCalleeSavedFrameSize() const { return CalleeSavedFrameSize; }
   void setCalleeSavedFrameSize(unsigned bytes) { CalleeSavedFrameSize = bytes; }
+
+  unsigned getLowReg() const { return LowReg; }
+  void setLowReg(unsigned Reg) { LowReg = Reg; }
+
+  unsigned getHighReg() const { return HighReg; }
+  void setHighReg(unsigned Reg) { HighReg = Reg; }
 };
 
 } // End llvm namespace
diff --git a/lib/Target/SystemZ/SystemZRegisterInfo.cpp b/lib/Target/SystemZ/SystemZRegisterInfo.cpp
index 0411569..087da0b 100644
--- a/lib/Target/SystemZ/SystemZRegisterInfo.cpp
+++ b/lib/Target/SystemZ/SystemZRegisterInfo.cpp
@@ -18,6 +18,7 @@
 #include "llvm/CodeGen/MachineInstrBuilder.h"
 #include "llvm/CodeGen/MachineFrameInfo.h"
 #include "llvm/CodeGen/MachineFunction.h"
+#include "llvm/CodeGen/MachineRegisterInfo.h"
 #include "llvm/Target/TargetFrameInfo.h"
 #include "llvm/Target/TargetInstrInfo.h"
 #include "llvm/Target/TargetMachine.h"
@@ -35,7 +36,8 @@ const unsigned*
 SystemZRegisterInfo::getCalleeSavedRegs(const MachineFunction *MF) const {
   static const unsigned CalleeSavedRegs[] = {
     SystemZ::R6D,  SystemZ::R7D,  SystemZ::R8D,  SystemZ::R9D,
-    SystemZ::R10D, SystemZ::R11D, SystemZ::R12D, SystemZ::R13D, SystemZ::R14D,
+    SystemZ::R10D, SystemZ::R11D, SystemZ::R12D, SystemZ::R13D,
+    SystemZ::R14D, SystemZ::R15D,
     SystemZ::F1,  SystemZ::F3,  SystemZ::F5,  SystemZ::F7,
     0
   };
@@ -50,7 +52,7 @@ SystemZRegisterInfo::getCalleeSavedRegClasses(const MachineFunction *MF) const {
     &SystemZ::GR64RegClass, &SystemZ::GR64RegClass,
     &SystemZ::GR64RegClass, &SystemZ::GR64RegClass,
     &SystemZ::GR64RegClass, &SystemZ::GR64RegClass,
-    &SystemZ::GR64RegClass,
+    &SystemZ::GR64RegClass, &SystemZ::GR64RegClass,
     &SystemZ::FP64RegClass, &SystemZ::FP64RegClass,
     &SystemZ::FP64RegClass, &SystemZ::FP64RegClass, 0
   };
@@ -66,16 +68,16 @@ BitVector SystemZRegisterInfo::getReservedRegs(const MachineFunction &MF) const
   return Reserved;
 }
 
-// needsFP - Return true if the specified function should have a dedicated frame
-// pointer register.  This is true if the function has variable sized allocas or
-// if frame pointer elimination is disabled.
-//
+/// needsFP - Return true if the specified function should have a dedicated
+/// frame pointer register.  This is true if the function has variable sized
+/// allocas or if frame pointer elimination is disabled.
 bool SystemZRegisterInfo::hasFP(const MachineFunction &MF) const {
   const MachineFrameInfo *MFI = MF.getFrameInfo();
   return NoFramePointerElim || MFI->hasVarSizedObjects();
 }
 
 bool SystemZRegisterInfo::hasReservedCallFrame(MachineFunction &MF) const {
+  // FIXME: Should we always have reserved call frame?
   return !MF.getFrameInfo()->hasVarSizedObjects();
 }
 
@@ -137,6 +139,25 @@ void SystemZRegisterInfo::eliminateFrameIndex(MachineBasicBlock::iterator II,
   MI.getOperand(i+1).ChangeToImmediate(Offset);
 }
 
+void
+SystemZRegisterInfo::processFunctionBeforeCalleeSavedScan(MachineFunction &MF,
+                                                       RegScavenger *RS) const {
+  // Determine whether R15/R14 will ever be clobbered inside the function. And
+  // if yes - mark it as 'callee' saved.
+  MachineFrameInfo *FFI = MF.getFrameInfo();
+
+  if (FFI->hasCalls()
+      /* FIXME: function is varargs */
+      /* FIXME: function grabs RA */
+      /* FIXME: function calls eh_return */)
+    MF.getRegInfo().setPhysRegUsed(SystemZ::R14D);
+
+  if (FFI->getObjectIndexEnd() != 0 || // Contains automatic variables
+      FFI->hasVarSizedObjects() // Function calls dynamic alloca's
+      /* FIXME: function is varargs */)
+    MF.getRegInfo().setPhysRegUsed(SystemZ::R15D);
+}
+
 /// emitSPUpdate - Emit a series of instructions to increment / decrement the
 /// stack pointer by a constant value.
 static
@@ -177,7 +198,11 @@ void SystemZRegisterInfo::emitPrologue(MachineFunction &MF) const {
   uint64_t StackSize =
     MFI->getStackSize() - SystemZMFI->getCalleeSavedFrameSize();
 
-  // FIXME: Skip the callee-saved push instructions.
+  // Skip the callee-saved push instructions.
+  while (MBBI != MBB.end() &&
+         (MBBI->getOpcode() == SystemZ::MOV64mr ||
+          MBBI->getOpcode() == SystemZ::MOV64mrm))
+    ++MBBI;
 
   if (MBBI != MBB.end())
     DL = MBBI->getDebugLoc();
@@ -223,23 +248,30 @@ void SystemZRegisterInfo::emitEpilogue(MachineFunction &MF,
     MFI->getStackSize() - SystemZMFI->getCalleeSavedFrameSize();
   uint64_t NumBytes = StackSize - TFI.getOffsetOfLocalArea();
 
-  // Skip the callee-saved regs load instructions.
-  MachineBasicBlock::iterator LastCSPop = MBBI;
+  // Skip the final terminator instruction.
   while (MBBI != MBB.begin()) {
     MachineBasicBlock::iterator PI = prior(MBBI);
+    --MBBI;
     if (!PI->getDesc().isTerminator())
       break;
-    --MBBI;
   }
 
-  DL = MBBI->getDebugLoc();
-
-  if (MFI->hasVarSizedObjects()) {
-    assert(0 && "Not implemented yet!");
-  } else {
-    // adjust stack pointer back: R15 += numbytes
-    if (StackSize)
-      emitSPUpdate(MBB, MBBI, NumBytes, TII);
+  // During callee-saved restores emission stack frame was not yet finialized
+  // (and thus - the stack size was unknown). Tune the offset having full stack
+  // size in hands.
+  if (SystemZMFI->getCalleeSavedFrameSize()) {
+    assert((MBBI->getOpcode() == SystemZ::MOV64rmm ||
+            MBBI->getOpcode() == SystemZ::MOV64rm) &&
+           "Expected to see callee-save register restore code");
+
+    unsigned i = 0;
+    MachineInstr &MI = *MBBI;
+    while (!MI.getOperand(i).isImm()) {
+      ++i;
+      assert(i < MI.getNumOperands() && "Unexpected restore code!");
+    }
+
+    MI.getOperand(i).ChangeToImmediate(NumBytes + MI.getOperand(i).getImm());
   }
 }
 
diff --git a/lib/Target/SystemZ/SystemZRegisterInfo.h b/lib/Target/SystemZ/SystemZRegisterInfo.h
index 74db3d5..d800f29 100644
--- a/lib/Target/SystemZ/SystemZRegisterInfo.h
+++ b/lib/Target/SystemZ/SystemZRegisterInfo.h
@@ -49,6 +49,10 @@ struct SystemZRegisterInfo : public SystemZGenRegisterInfo {
   void eliminateFrameIndex(MachineBasicBlock::iterator II,
                            int SPAdj, RegScavenger *RS = NULL) const;
 
+
+  void processFunctionBeforeCalleeSavedScan(MachineFunction &MF,
+                                            RegScavenger *RS) const;
+
   void emitPrologue(MachineFunction &MF) const;
   void emitEpilogue(MachineFunction &MF, MachineBasicBlock &MBB) const;
 
diff --git a/test/CodeGen/SystemZ/06-CallViaStack.ll b/test/CodeGen/SystemZ/06-CallViaStack.ll
index 7b222d9..03cb164 100644
--- a/test/CodeGen/SystemZ/06-CallViaStack.ll
+++ b/test/CodeGen/SystemZ/06-CallViaStack.ll
@@ -1,4 +1,4 @@
-; RUN: llvm-as < %s | llc | grep 168 | count 2
+; RUN: llvm-as < %s | llc | grep 168 | count 1
 ; RUN: llvm-as < %s | llc | grep 160 | count 3
 ; RUN: llvm-as < %s | llc | grep 328 | count 1
 
diff --git a/test/CodeGen/SystemZ/06-LocalFrame.ll b/test/CodeGen/SystemZ/06-LocalFrame.ll
index 588c46d..4f7535d 100644
--- a/test/CodeGen/SystemZ/06-LocalFrame.ll
+++ b/test/CodeGen/SystemZ/06-LocalFrame.ll
@@ -1,6 +1,6 @@
 ; RUN: llvm-as < %s | llc | grep 160 | count 1
 ; RUN: llvm-as < %s | llc | grep 328 | count 1
-; RUN: llvm-as < %s | llc | grep 168 | count 2
+; RUN: llvm-as < %s | llc | grep 168 | count 1
 
 target datalayout = "E-p:64:64:64-i1:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-f128:128:128"
 target triple = "s390x-unknown-linux-gnu"