Home | History | Annotate | Download | only in X86 
      1 //===-------- X86PadShortFunction.cpp - pad short functions -----------===//
      2 //
      3 //                     The LLVM Compiler Infrastructure
      4 //
      5 // This file is distributed under the University of Illinois Open Source
      6 // License. See LICENSE.TXT for details.
      7 //
      8 //===----------------------------------------------------------------------===//
      9 //
     10 // This file defines the pass which will pad short functions to prevent
     11 // a stall if a function returns before the return address is ready. This
     12 // is needed for some Intel Atom processors.
     13 //
     14 //===----------------------------------------------------------------------===//
     15 
     16 #include <algorithm>
     17 
     18 #include "X86.h"
     19 #include "X86InstrInfo.h"
     20 #include "X86Subtarget.h"
     21 #include "llvm/ADT/Statistic.h"
     22 #include "llvm/CodeGen/MachineFunctionPass.h"
     23 #include "llvm/CodeGen/MachineInstrBuilder.h"
     24 #include "llvm/CodeGen/MachineRegisterInfo.h"
     25 #include "llvm/CodeGen/Passes.h"
     26 #include "llvm/IR/Function.h"
     27 #include "llvm/Support/Debug.h"
     28 #include "llvm/Support/raw_ostream.h"
     29 #include "llvm/Target/TargetInstrInfo.h"
     30 
     31 using namespace llvm;
     32 
     33 #define DEBUG_TYPE "x86-pad-short-functions"
     34 
     35 STATISTIC(NumBBsPadded, "Number of basic blocks padded");
     36 
     37 namespace {
     38   struct VisitedBBInfo {
     39     // HasReturn - Whether the BB contains a return instruction
     40     bool HasReturn;
     41 
     42     // Cycles - Number of cycles until return if HasReturn is true, otherwise
     43     // number of cycles until end of the BB
     44     unsigned int Cycles;
     45 
     46     VisitedBBInfo() : HasReturn(false), Cycles(0) {}
     47     VisitedBBInfo(bool HasReturn, unsigned int Cycles)
     48       : HasReturn(HasReturn), Cycles(Cycles) {}
     49   };
     50 
     51   struct PadShortFunc : public MachineFunctionPass {
     52     static char ID;
     53     PadShortFunc() : MachineFunctionPass(ID)
     54                    , Threshold(4), STI(nullptr), TII(nullptr) {}
     55 
     56     bool runOnMachineFunction(MachineFunction &MF) override;
     57 
     58     const char *getPassName() const override {
     59       return "X86 Atom pad short functions";
     60     }
     61 
     62   private:
     63     void findReturns(MachineBasicBlock *MBB,
     64                      unsigned int Cycles = 0);
     65 
     66     bool cyclesUntilReturn(MachineBasicBlock *MBB,
     67                            unsigned int &Cycles);
     68 
     69     void addPadding(MachineBasicBlock *MBB,
     70                     MachineBasicBlock::iterator &MBBI,
     71                     unsigned int NOOPsToAdd);
     72 
     73     const unsigned int Threshold;
     74 
     75     // ReturnBBs - Maps basic blocks that return to the minimum number of
     76     // cycles until the return, starting from the entry block.
     77     DenseMap<MachineBasicBlock*, unsigned int> ReturnBBs;
     78 
     79     // VisitedBBs - Cache of previously visited BBs.
     80     DenseMap<MachineBasicBlock*, VisitedBBInfo> VisitedBBs;
     81 
     82     const X86Subtarget *STI;
     83     const TargetInstrInfo *TII;
     84   };
     85 
     86   char PadShortFunc::ID = 0;
     87 }
     88 
     89 FunctionPass *llvm::createX86PadShortFunctions() {
     90   return new PadShortFunc();
     91 }
     92 
     93 /// runOnMachineFunction - Loop over all of the basic blocks, inserting
     94 /// NOOP instructions before early exits.
     95 bool PadShortFunc::runOnMachineFunction(MachineFunction &MF) {
     96   if (MF.getFunction()->hasFnAttribute(Attribute::OptimizeForSize) ||
     97       MF.getFunction()->hasFnAttribute(Attribute::MinSize)) {
     98     return false;
     99   }
    100 
    101   STI = &MF.getSubtarget<X86Subtarget>();
    102   if (!STI->padShortFunctions())
    103     return false;
    104 
    105   TII = STI->getInstrInfo();
    106 
    107   // Search through basic blocks and mark the ones that have early returns
    108   ReturnBBs.clear();
    109   VisitedBBs.clear();
    110   findReturns(MF.begin());
    111 
    112   bool MadeChange = false;
    113 
    114   MachineBasicBlock *MBB;
    115   unsigned int Cycles = 0;
    116 
    117   // Pad the identified basic blocks with NOOPs
    118   for (DenseMap<MachineBasicBlock*, unsigned int>::iterator I = ReturnBBs.begin();
    119        I != ReturnBBs.end(); ++I) {
    120     MBB = I->first;
    121     Cycles = I->second;
    122 
    123     if (Cycles < Threshold) {
    124       // BB ends in a return. Skip over any DBG_VALUE instructions
    125       // trailing the terminator.
    126       assert(MBB->size() > 0 &&
    127              "Basic block should contain at least a RET but is empty");
    128       MachineBasicBlock::iterator ReturnLoc = --MBB->end();
    129 
    130       while (ReturnLoc->isDebugValue())
    131         --ReturnLoc;
    132       assert(ReturnLoc->isReturn() && !ReturnLoc->isCall() &&
    133              "Basic block does not end with RET");
    134 
    135       addPadding(MBB, ReturnLoc, Threshold - Cycles);
    136       NumBBsPadded++;
    137       MadeChange = true;
    138     }
    139   }
    140 
    141   return MadeChange;
    142 }
    143 
    144 /// findReturn - Starting at MBB, follow control flow and add all
    145 /// basic blocks that contain a return to ReturnBBs.
    146 void PadShortFunc::findReturns(MachineBasicBlock *MBB, unsigned int Cycles) {
    147   // If this BB has a return, note how many cycles it takes to get there.
    148   bool hasReturn = cyclesUntilReturn(MBB, Cycles);
    149   if (Cycles >= Threshold)
    150     return;
    151 
    152   if (hasReturn) {
    153     ReturnBBs[MBB] = std::max(ReturnBBs[MBB], Cycles);
    154     return;
    155   }
    156 
    157   // Follow branches in BB and look for returns
    158   for (MachineBasicBlock::succ_iterator I = MBB->succ_begin();
    159        I != MBB->succ_end(); ++I) {
    160     if (*I == MBB)
    161       continue;
    162     findReturns(*I, Cycles);
    163   }
    164 }
    165 
    166 /// cyclesUntilReturn - return true if the MBB has a return instruction,
    167 /// and return false otherwise.
    168 /// Cycles will be incremented by the number of cycles taken to reach the
    169 /// return or the end of the BB, whichever occurs first.
    170 bool PadShortFunc::cyclesUntilReturn(MachineBasicBlock *MBB,
    171                                      unsigned int &Cycles) {
    172   // Return cached result if BB was previously visited
    173   DenseMap<MachineBasicBlock*, VisitedBBInfo>::iterator it
    174     = VisitedBBs.find(MBB);
    175   if (it != VisitedBBs.end()) {
    176     VisitedBBInfo BBInfo = it->second;
    177     Cycles += BBInfo.Cycles;
    178     return BBInfo.HasReturn;
    179   }
    180 
    181   unsigned int CyclesToEnd = 0;
    182 
    183   for (MachineBasicBlock::iterator MBBI = MBB->begin();
    184         MBBI != MBB->end(); ++MBBI) {
    185     MachineInstr *MI = MBBI;
    186     // Mark basic blocks with a return instruction. Calls to other
    187     // functions do not count because the called function will be padded,
    188     // if necessary.
    189     if (MI->isReturn() && !MI->isCall()) {
    190       VisitedBBs[MBB] = VisitedBBInfo(true, CyclesToEnd);
    191       Cycles += CyclesToEnd;
    192       return true;
    193     }
    194 
    195     CyclesToEnd += TII->getInstrLatency(STI->getInstrItineraryData(), MI);
    196   }
    197 
    198   VisitedBBs[MBB] = VisitedBBInfo(false, CyclesToEnd);
    199   Cycles += CyclesToEnd;
    200   return false;
    201 }
    202 
    203 /// addPadding - Add the given number of NOOP instructions to the function
    204 /// just prior to the return at MBBI
    205 void PadShortFunc::addPadding(MachineBasicBlock *MBB,
    206                               MachineBasicBlock::iterator &MBBI,
    207                               unsigned int NOOPsToAdd) {
    208   DebugLoc DL = MBBI->getDebugLoc();
    209 
    210   while (NOOPsToAdd-- > 0) {
    211     BuildMI(*MBB, MBBI, DL, TII->get(X86::NOOP));
    212     BuildMI(*MBB, MBBI, DL, TII->get(X86::NOOP));
    213   }
    214 }
    215