lib/Target/PowerPC/PPCBranchSelector.cpp

   1 //===-- PPCBranchSelector.cpp - Emit long conditional branches-----*- C++ -*-=//
   2 //
   3 //                     The LLVM Compiler Infrastructure
   4 //
   5 // This file was developed by Nate Baegeman and is distributed under the
   6 // University of Illinois Open Source License. See LICENSE.TXT for details.
   7 //
   8 //===----------------------------------------------------------------------===//
   9 //
  10 // This file contains a pass that scans a machine function to determine which
  11 // conditional branches need more than 16 bits of displacement to reach their
  12 // target basic block.  It does this in two passes; a calculation of basic block
  13 // positions pass, and a branch psuedo op to machine branch opcode pass.  This
  14 // pass should be run last, just before the assembly printer.
  15 //
  16 //===----------------------------------------------------------------------===//
  17
  18 #include "PPC.h"
  19 #include "PPCInstrBuilder.h"
  20 #include "PPCInstrInfo.h"
  21 #include "llvm/CodeGen/MachineFunctionPass.h"
  22 #include <map>
  23 using namespace llvm;
  24
  25 namespace {
  26   struct PPCBSel : public MachineFunctionPass {
  27     // OffsetMap - Mapping between BB and byte offset from start of function
  28     std::map<MachineBasicBlock*, unsigned> OffsetMap;
  29
  30     virtual bool runOnMachineFunction(MachineFunction &Fn);
  31
  32     virtual const char *getPassName() const {
  33       return "PowerPC Branch Selection";
  34     }
  35   };
  36 }
  37
  38 /// createPPCBranchSelectionPass - returns an instance of the Branch Selection
  39 /// Pass
  40 ///
  41 FunctionPass *llvm::createPPCBranchSelectionPass() {
  42   return new PPCBSel();
  43 }
  44
  45 /// getNumBytesForInstruction - Return the number of bytes of code the specified
  46 /// instruction may be.  This returns the maximum number of bytes.
  47 ///
  48 static unsigned getNumBytesForInstruction(MachineInstr *MI) {
  49   switch (MI->getOpcode()) {
  50   case PPC::COND_BRANCH:
  51     // while this will be 4 most of the time, if we emit 8 it is just a
  52     // minor pessimization that saves us from having to worry about
  53     // keeping the offsets up to date later when we emit long branch glue.
  54     return 8;
  55   case PPC::IMPLICIT_DEF_GPR: // no asm emitted
  56   case PPC::IMPLICIT_DEF_F4: // no asm emitted
  57   case PPC::IMPLICIT_DEF_F8: // no asm emitted
  58     return 0;
  59   case PPC::INLINEASM:    // Inline Asm: Variable size.
  60     for (unsigned i = 0, e = MI->getNumOperands(); i != e; ++i)
  61       if (MI->getOperand(i).isExternalSymbol()) {
  62         const char *AsmStr = MI->getOperand(i).getSymbolName();
  63         // Count the number of newline's in the asm string.
  64         unsigned NumInstrs = 0;
  65         for (; *AsmStr; ++AsmStr)
  66           NumInstrs += *AsmStr == '\n';
  67         return NumInstrs*4;
  68       }
  69     assert(0 && "INLINEASM didn't have format string??");
  70   default:
  71     return 4; // PowerPC instructions are all 4 bytes
  72   }
  73 }
  74
  75
  76 bool PPCBSel::runOnMachineFunction(MachineFunction &Fn) {
  77   // Running total of instructions encountered since beginning of function
  78   unsigned ByteCount = 0;
  79
  80   // For each MBB, add its offset to the offset map, and count up its
  81   // instructions
  82   for (MachineFunction::iterator MFI = Fn.begin(), E = Fn.end(); MFI != E;
  83        ++MFI) {
  84     MachineBasicBlock *MBB = MFI;
  85     OffsetMap[MBB] = ByteCount;
  86
  87     for (MachineBasicBlock::iterator MBBI = MBB->begin(), EE = MBB->end();
  88          MBBI != EE; ++MBBI)
  89       ByteCount += getNumBytesForInstruction(MBBI);
  90   }
  91
  92   // We're about to run over the MBB's again, so reset the ByteCount
  93   ByteCount = 0;
  94
  95   // For each MBB, find the conditional branch pseudo instructions, and
  96   // calculate the difference between the target MBB and the current ICount
  97   // to decide whether or not to emit a short or long branch.
  98   //
  99   // short branch:
 100   // bCC .L_TARGET_MBB
 101   //
 102   // long branch:
 103   // bInverseCC $PC+8
 104   // b .L_TARGET_MBB
 105   for (MachineFunction::iterator MFI = Fn.begin(), E = Fn.end(); MFI != E;
 106        ++MFI) {
 107     MachineBasicBlock *MBB = MFI;
 108
 109     for (MachineBasicBlock::iterator MBBI = MBB->begin(), EE = MBB->end();
 110          MBBI != EE; ++MBBI) {
 111       // We may end up deleting the MachineInstr that MBBI points to, so
 112       // remember its opcode now so we can refer to it after calling erase()
 113       unsigned ByteSize = getNumBytesForInstruction(MBBI);
 114       if (MBBI->getOpcode() == PPC::COND_BRANCH) {
 115         MachineBasicBlock::iterator MBBJ = MBBI;
 116         ++MBBJ;
 117
 118         // condbranch operands:
 119         // 0. CR0 register
 120         // 1. bc opcode
 121         // 2. target MBB
 122         // 3. fallthrough MBB
 123         MachineBasicBlock *trueMBB =
 124           MBBI->getOperand(2).getMachineBasicBlock();
 125
 126         int Displacement = OffsetMap[trueMBB] - ByteCount;
 127         unsigned Opcode = MBBI->getOperand(1).getImmedValue();
 128         unsigned CRReg = MBBI->getOperand(0).getReg();
 129         unsigned Inverted = PPCInstrInfo::invertPPCBranchOpcode(Opcode);
 130
 131         if (Displacement >= -32768 && Displacement <= 32767) {
 132           BuildMI(*MBB, MBBJ, Opcode, 2).addReg(CRReg).addMBB(trueMBB);
 133         } else {
 134           BuildMI(*MBB, MBBJ, Inverted, 2).addReg(CRReg).addSImm(8);
 135           BuildMI(*MBB, MBBJ, PPC::B, 1).addMBB(trueMBB);
 136         }
 137
 138         // Erase the psuedo COND_BRANCH instruction, and then back up the
 139         // iterator so that when the for loop increments it, we end up in
 140         // the correct place rather than iterating off the end.
 141         MBB->erase(MBBI);
 142         MBBI = --MBBJ;
 143       }
 144       ByteCount += ByteSize;
 145     }
 146   }
 147
 148   OffsetMap.clear();
 149   return true;
 150 }
 151