llvm/lib/Target/X86/X86PadShortFunction.cpp

   1 //===-------- X86PadShortFunction.cpp - pad short functions -----------===//
   2 //
   3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
   4 // See https://llvm.org/LICENSE.txt for license information.
   5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
   6 //
   7 //===----------------------------------------------------------------------===//
   8 //
   9 // This file defines the pass which will pad short functions to prevent
  10 // a stall if a function returns before the return address is ready. This
  11 // is needed for some Intel Atom processors.
  12 //
  13 //===----------------------------------------------------------------------===//
  14
  15
  16 #include "X86.h"
  17 #include "X86InstrInfo.h"
  18 #include "X86Subtarget.h"
  19 #include "llvm/ADT/Statistic.h"
  20 #include "llvm/Analysis/ProfileSummaryInfo.h"
  21 #include "llvm/CodeGen/LazyMachineBlockFrequencyInfo.h"
  22 #include "llvm/CodeGen/MachineFunctionPass.h"
  23 #include "llvm/CodeGen/MachineInstrBuilder.h"
  24 #include "llvm/CodeGen/MachineSizeOpts.h"
  25 #include "llvm/CodeGen/Passes.h"
  26 #include "llvm/CodeGen/TargetSchedule.h"
  27 #include "llvm/IR/Function.h"
  28 #include "llvm/Support/Debug.h"
  29 #include "llvm/Support/raw_ostream.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 = false;
  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 = 0;
  45
  46     VisitedBBInfo() = default;
  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
  55     bool runOnMachineFunction(MachineFunction &MF) override;
  56
  57     void getAnalysisUsage(AnalysisUsage &AU) const override {
  58       AU.addRequired<ProfileSummaryInfoWrapperPass>();
  59       AU.addRequired<LazyMachineBlockFrequencyInfoPass>();
  60       AU.addPreserved<LazyMachineBlockFrequencyInfoPass>();
  61       MachineFunctionPass::getAnalysisUsage(AU);
  62     }
  63
  64     MachineFunctionProperties getRequiredProperties() const override {
  65       return MachineFunctionProperties().set(
  66           MachineFunctionProperties::Property::NoVRegs);
  67     }
  68
  69     StringRef getPassName() const override {
  70       return "X86 Atom pad short functions";
  71     }
  72
  73   private:
  74     void findReturns(MachineBasicBlock *MBB,
  75                      unsigned int Cycles = 0);
  76
  77     bool cyclesUntilReturn(MachineBasicBlock *MBB,
  78                            unsigned int &Cycles);
  79
  80     void addPadding(MachineBasicBlock *MBB,
  81                     MachineBasicBlock::iterator &MBBI,
  82                     unsigned int NOOPsToAdd);
  83
  84     const unsigned int Threshold = 4;
  85
  86     // ReturnBBs - Maps basic blocks that return to the minimum number of
  87     // cycles until the return, starting from the entry block.
  88     DenseMap<MachineBasicBlock*, unsigned int> ReturnBBs;
  89
  90     // VisitedBBs - Cache of previously visited BBs.
  91     DenseMap<MachineBasicBlock*, VisitedBBInfo> VisitedBBs;
  92
  93     TargetSchedModel TSM;
  94   };
  95
  96   char PadShortFunc::ID = 0;
  97 }
  98
  99 FunctionPass *llvm::createX86PadShortFunctions() {
 100   return new PadShortFunc();
 101 }
 102
 103 /// runOnMachineFunction - Loop over all of the basic blocks, inserting
 104 /// NOOP instructions before early exits.
 105 bool PadShortFunc::runOnMachineFunction(MachineFunction &MF) {
 106   if (skipFunction(MF.getFunction()))
 107     return false;
 108
 109   if (MF.getFunction().hasOptSize())
 110     return false;
 111
 112   if (!MF.getSubtarget<X86Subtarget>().padShortFunctions())
 113     return false;
 114
 115   TSM.init(&MF.getSubtarget());
 116
 117   auto *PSI =
 118       &getAnalysis<ProfileSummaryInfoWrapperPass>().getPSI();
 119   auto *MBFI = (PSI && PSI->hasProfileSummary()) ?
 120                &getAnalysis<LazyMachineBlockFrequencyInfoPass>().getBFI() :
 121                nullptr;
 122
 123   // Search through basic blocks and mark the ones that have early returns
 124   ReturnBBs.clear();
 125   VisitedBBs.clear();
 126   findReturns(&MF.front());
 127
 128   bool MadeChange = false;
 129
 130   // Pad the identified basic blocks with NOOPs
 131   for (const auto &ReturnBB : ReturnBBs) {
 132     MachineBasicBlock *MBB = ReturnBB.first;
 133     unsigned Cycles = ReturnBB.second;
 134
 135     // Function::hasOptSize is already checked above.
 136     bool OptForSize = llvm::shouldOptimizeForSize(MBB, PSI, MBFI);
 137     if (OptForSize)
 138       continue;
 139
 140     if (Cycles < Threshold) {
 141       // BB ends in a return. Skip over any DBG_VALUE instructions
 142       // trailing the terminator.
 143       assert(MBB->size() > 0 &&
 144              "Basic block should contain at least a RET but is empty");
 145       MachineBasicBlock::iterator ReturnLoc = --MBB->end();
 146
 147       while (ReturnLoc->isDebugInstr())
 148         --ReturnLoc;
 149       assert(ReturnLoc->isReturn() && !ReturnLoc->isCall() &&
 150              "Basic block does not end with RET");
 151
 152       addPadding(MBB, ReturnLoc, Threshold - Cycles);
 153       NumBBsPadded++;
 154       MadeChange = true;
 155     }
 156   }
 157
 158   return MadeChange;
 159 }
 160
 161 /// findReturn - Starting at MBB, follow control flow and add all
 162 /// basic blocks that contain a return to ReturnBBs.
 163 void PadShortFunc::findReturns(MachineBasicBlock *MBB, unsigned int Cycles) {
 164   // If this BB has a return, note how many cycles it takes to get there.
 165   bool hasReturn = cyclesUntilReturn(MBB, Cycles);
 166   if (Cycles >= Threshold)
 167     return;
 168
 169   if (hasReturn) {
 170     ReturnBBs[MBB] = std::max(ReturnBBs[MBB], Cycles);
 171     return;
 172   }
 173
 174   // Follow branches in BB and look for returns
 175   for (MachineBasicBlock *Succ : MBB->successors())
 176     if (Succ != MBB)
 177       findReturns(Succ, Cycles);
 178 }
 179
 180 /// cyclesUntilReturn - return true if the MBB has a return instruction,
 181 /// and return false otherwise.
 182 /// Cycles will be incremented by the number of cycles taken to reach the
 183 /// return or the end of the BB, whichever occurs first.
 184 bool PadShortFunc::cyclesUntilReturn(MachineBasicBlock *MBB,
 185                                      unsigned int &Cycles) {
 186   // Return cached result if BB was previously visited
 187   DenseMap<MachineBasicBlock*, VisitedBBInfo>::iterator it
 188     = VisitedBBs.find(MBB);
 189   if (it != VisitedBBs.end()) {
 190     VisitedBBInfo BBInfo = it->second;
 191     Cycles += BBInfo.Cycles;
 192     return BBInfo.HasReturn;
 193   }
 194
 195   unsigned int CyclesToEnd = 0;
 196
 197   for (MachineInstr &MI : *MBB) {
 198     // Mark basic blocks with a return instruction. Calls to other
 199     // functions do not count because the called function will be padded,
 200     // if necessary.
 201     if (MI.isReturn() && !MI.isCall()) {
 202       VisitedBBs[MBB] = VisitedBBInfo(true, CyclesToEnd);
 203       Cycles += CyclesToEnd;
 204       return true;
 205     }
 206
 207     CyclesToEnd += TSM.computeInstrLatency(&MI);
 208   }
 209
 210   VisitedBBs[MBB] = VisitedBBInfo(false, CyclesToEnd);
 211   Cycles += CyclesToEnd;
 212   return false;
 213 }
 214
 215 /// addPadding - Add the given number of NOOP instructions to the function
 216 /// just prior to the return at MBBI
 217 void PadShortFunc::addPadding(MachineBasicBlock *MBB,
 218                               MachineBasicBlock::iterator &MBBI,
 219                               unsigned int NOOPsToAdd) {
 220   const DebugLoc &DL = MBBI->getDebugLoc();
 221   unsigned IssueWidth = TSM.getIssueWidth();
 222
 223   for (unsigned i = 0, e = IssueWidth * NOOPsToAdd; i != e; ++i)
 224     BuildMI(*MBB, MBBI, DL, TSM.getInstrInfo()->get(X86::NOOP));
 225 }