[Sparc] Emit large negative adjustments to SP/FP with sethi+xor instead of sethi+or. This generates correct code for both sparc32 and sparc64.

llvm-svn: 195576

llvm/lib/Target/Sparc/Sparc.h

@@ -105,5 +105,22 @@ namespace llvm {
     }
     llvm_unreachable("Invalid cond code");
   }
+
+  inline static unsigned HI22(int64_t imm) {
+    return (unsigned)((imm >> 10) & ((1 << 22)-1));
+  }
+
+  inline static unsigned LO10(int64_t imm) {
+    return (unsigned)(imm & 0x3FF);
+  }
+
+  inline static unsigned HIX22(int64_t imm) {
+    return HI22(~imm);
+  }
+
+  inline static unsigned LOX10(int64_t imm) {
+    return ~LO10(~imm);
+  }
+
 }  // end namespace llvm
 #endif

llvm/lib/Target/Sparc/SparcFrameLowering.cpp

@@ -33,6 +33,51 @@ DisableLeafProc("disable-sparc-leaf-proc",
                 cl::Hidden);
 
 
+void SparcFrameLowering::emitSPAdjustment(MachineFunction &MF,
+                                          MachineBasicBlock &MBB,
+                                          MachineBasicBlock::iterator MBBI,
+                                          int NumBytes,
+                                          unsigned ADDrr,
+                                          unsigned ADDri) const {
+
+  DebugLoc dl = (MBBI != MBB.end()) ? MBBI->getDebugLoc() : DebugLoc();
+  const SparcInstrInfo &TII =
+    *static_cast<const SparcInstrInfo*>(MF.getTarget().getInstrInfo());
+
+  if (NumBytes >= -4096 && NumBytes < 4096) {
+    BuildMI(MBB, MBBI, dl, TII.get(ADDri), SP::O6)
+      .addReg(SP::O6).addImm(NumBytes);
+    return;
+  }
+
+  // Emit this the hard way.  This clobbers G1 which we always know is
+  // available here.
+  if (NumBytes >= 0) {
+    // Emit nonnegative numbers with sethi + or.
+    // sethi %hi(NumBytes), %g1
+    // or %g1, %lo(NumBytes), %g1
+    // add %sp, %g1, %sp
+    BuildMI(MBB, MBBI, dl, TII.get(SP::SETHIi), SP::G1)
+      .addImm(HI22(NumBytes));
+    BuildMI(MBB, MBBI, dl, TII.get(SP::ORri), SP::G1)
+      .addReg(SP::G1).addImm(LO10(NumBytes));
+    BuildMI(MBB, MBBI, dl, TII.get(ADDrr), SP::O6)
+      .addReg(SP::O6).addReg(SP::G1);
+    return ;
+  }
+
+  // Emit negative numbers with sethi + xor.
+  // sethi %hix(NumBytes), %g1
+  // xor %g1, %lox(NumBytes), %g1
+  // add %sp, %g1, %sp
+  BuildMI(MBB, MBBI, dl, TII.get(SP::SETHIi), SP::G1)
+    .addImm(HIX22(NumBytes));
+  BuildMI(MBB, MBBI, dl, TII.get(SP::XORri), SP::G1)
+    .addReg(SP::G1).addImm(LOX10(NumBytes));
+  BuildMI(MBB, MBBI, dl, TII.get(ADDrr), SP::O6)
+    .addReg(SP::O6).addReg(SP::G1);
+}
+
 void SparcFrameLowering::emitPrologue(MachineFunction &MF) const {
   SparcMachineFunctionInfo *FuncInfo = MF.getInfo<SparcMachineFunctionInfo>();
 
@@ -55,21 +100,8 @@ void SparcFrameLowering::emitPrologue(MachineFunction &MF) const {
     SAVErr = SP::ADDrr;
   }
   NumBytes = - SubTarget.getAdjustedFrameSize(NumBytes);
+  emitSPAdjustment(MF, MBB, MBBI, NumBytes, SAVErr, SAVEri);
 
-  if (NumBytes >= -4096) {
-    BuildMI(MBB, MBBI, dl, TII.get(SAVEri), SP::O6)
-      .addReg(SP::O6).addImm(NumBytes);
-  } else {
-    // Emit this the hard way.  This clobbers G1 which we always know is
-    // available here.
-    unsigned OffHi = (unsigned)NumBytes >> 10U;
-    BuildMI(MBB, MBBI, dl, TII.get(SP::SETHIi), SP::G1).addImm(OffHi);
-    // Emit G1 = G1 + I6
-    BuildMI(MBB, MBBI, dl, TII.get(SP::ORri), SP::G1)
-      .addReg(SP::G1).addImm(NumBytes & ((1 << 10)-1));
-    BuildMI(MBB, MBBI, dl, TII.get(SAVErr), SP::O6)
-      .addReg(SP::O6).addReg(SP::G1);
-  }
   MachineModuleInfo &MMI = MF.getMMI();
   const MCRegisterInfo *MRI = MMI.getContext().getRegisterInfo();
   MCSymbol *FrameLabel = MMI.getContext().CreateTempSymbol();
@@ -100,11 +132,9 @@ eliminateCallFramePseudoInstr(MachineFunction &MF, MachineBasicBlock &MBB,
     int Size = MI.getOperand(0).getImm();
     if (MI.getOpcode() == SP::ADJCALLSTACKDOWN)
       Size = -Size;
-    const SparcInstrInfo &TII =
+
-      *static_cast<const SparcInstrInfo*>(MF.getTarget().getInstrInfo());
     if (Size)
-      BuildMI(MBB, I, DL, TII.get(SP::ADDri), SP::O6).addReg(SP::O6)
+      emitSPAdjustment(MF, MBB, I, Size, SP::ADDrr, SP::ADDri);
-        .addImm(Size);
   }
   MBB.erase(I);
 }
@@ -131,21 +161,7 @@ void SparcFrameLowering::emitEpilogue(MachineFunction &MF,
     return;
 
   NumBytes = SubTarget.getAdjustedFrameSize(NumBytes);
-
+  emitSPAdjustment(MF, MBB, MBBI, NumBytes, SP::ADDrr, SP::ADDri);
-  if (NumBytes < 4096) {
-    BuildMI(MBB, MBBI, dl, TII.get(SP::ADDri), SP::O6)
-      .addReg(SP::O6).addImm(NumBytes);
-  } else {
-    // Emit this the hard way.  This clobbers G1 which we always know is
-    // available here.
-    unsigned OffHi = (unsigned)NumBytes >> 10U;
-    BuildMI(MBB, MBBI, dl, TII.get(SP::SETHIi), SP::G1).addImm(OffHi);
-    // Emit G1 = G1 + I6
-    BuildMI(MBB, MBBI, dl, TII.get(SP::ORri), SP::G1)
-      .addReg(SP::G1).addImm(NumBytes & ((1 << 10)-1));
-    BuildMI(MBB, MBBI, dl, TII.get(SP::ADDrr), SP::O6)
-      .addReg(SP::O6).addReg(SP::G1);
-  }
 }
 
 bool SparcFrameLowering::hasReservedCallFrame(const MachineFunction &MF) const {

llvm/lib/Target/Sparc/SparcFrameLowering.h

@@ -49,6 +49,14 @@ private:
 
   // Returns true if MF is a leaf procedure.
   bool isLeafProc(MachineFunction &MF) const;
+
+
+  // Emits code for adjusting SP in function prologue/epilogue.
+  void emitSPAdjustment(MachineFunction &MF,
+                        MachineBasicBlock &MBB,
+                        MachineBasicBlock::iterator MBBI,
+                        int NumBytes, unsigned ADDrr, unsigned ADDri) const;
+
 };
 
 } // End llvm namespace

llvm/lib/Target/Sparc/SparcRegisterInfo.cpp

@@ -105,19 +105,46 @@ static void replaceFI(MachineFunction &MF,
     // encode it.
     MI.getOperand(FIOperandNum).ChangeToRegister(FramePtr, false);
     MI.getOperand(FIOperandNum + 1).ChangeToImmediate(Offset);
-  } else {
+    return;
-    // Otherwise, emit a G1 = SETHI %hi(offset).  FIXME: it would be better to
+  }
-    // scavenge a register here instead of reserving G1 all of the time.
+
-    const TargetInstrInfo &TII = *MF.getTarget().getInstrInfo();
+  const TargetInstrInfo &TII = *MF.getTarget().getInstrInfo();
-    unsigned OffHi = (unsigned)Offset >> 10U;
+
-    BuildMI(*MI.getParent(), II, dl, TII.get(SP::SETHIi), SP::G1).addImm(OffHi);
+  // FIXME: it would be better to scavenge a register here instead of
+  // reserving G1 all of the time.
+  if (Offset >= 0) {
+    // Emit nonnegaive immediates with sethi + or.
+    // sethi %hi(Offset), %g1
+    // add %g1, %fp, %g1
+    // Insert G1+%lo(offset) into the user.
+    BuildMI(*MI.getParent(), II, dl, TII.get(SP::SETHIi), SP::G1)
+      .addImm(HI22(Offset));
+
+
     // Emit G1 = G1 + I6
     BuildMI(*MI.getParent(), II, dl, TII.get(SP::ADDrr), SP::G1).addReg(SP::G1)
       .addReg(FramePtr);
     // Insert: G1+%lo(offset) into the user.
     MI.getOperand(FIOperandNum).ChangeToRegister(SP::G1, false);
-    MI.getOperand(FIOperandNum + 1).ChangeToImmediate(Offset & ((1 << 10)-1));
+    MI.getOperand(FIOperandNum + 1).ChangeToImmediate(LO10(Offset));
+    return;
   }
+
+  // Emit Negative numbers with sethi + xor
+  // sethi %hix(Offset), %g1
+  // xor  %g1, %lox(offset), %g1
+  // add %g1, %fp, %g1
+  // Insert: G1 + 0 into the user.
+  BuildMI(*MI.getParent(), II, dl, TII.get(SP::SETHIi), SP::G1)
+    .addImm(HIX22(Offset));
+  BuildMI(*MI.getParent(), II, dl, TII.get(SP::XORri), SP::G1)
+    .addReg(SP::G1).addImm(LOX10(Offset));
+
+  BuildMI(*MI.getParent(), II, dl, TII.get(SP::ADDrr), SP::G1).addReg(SP::G1)
+    .addReg(FramePtr);
+  // Insert: G1+%lo(offset) into the user.
+  MI.getOperand(FIOperandNum).ChangeToRegister(SP::G1, false);
+  MI.getOperand(FIOperandNum + 1).ChangeToImmediate(0);
 }
 
 

llvm/test/CodeGen/SPARC/64abi.ll

@@ -390,3 +390,24 @@ entry:
   %2 = add nsw i32 %0, %1
   ret i32 %2
 }
+
+; CHECK-LABEL: test_large_stack
+
+; CHECK:       sethi 16, %g1
+; CHECK:       xor %g1, -176, %g1
+; CHECK:       save %sp, %g1, %sp
+
+; CHECK:       sethi 14, %g1
+; CHECK:       xor %g1, -1, %g1
+; CHECK:       add %g1, %fp, %g1
+; CHECK:       call use_buf
+
+define i32 @test_large_stack() {
+entry:
+  %buffer1 = alloca [16384 x i8], align 8
+  %buffer1.sub = getelementptr inbounds [16384 x i8]* %buffer1, i32 0, i32 0
+  %0 = call i32 @use_buf(i32 16384, i8* %buffer1.sub)
+  ret i32 %0
+}
+
+declare i32 @use_buf(i32, i8*)