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