1 //===-- AArch64Subtarget.cpp - AArch64 Subtarget Information ----*- C++ -*-===//
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 implements the AArch64 specific subclass of TargetSubtarget.
11 //
12 //===----------------------------------------------------------------------===//
13 
14 #include "AArch64Subtarget.h"
15 #include "AArch64InstrInfo.h"
16 #include "AArch64PBQPRegAlloc.h"
17 #include "llvm/CodeGen/MachineScheduler.h"
18 #include "llvm/IR/GlobalValue.h"
19 #include "llvm/Support/TargetRegistry.h"
20 
21 using namespace llvm;
22 
23 #define DEBUG_TYPE "aarch64-subtarget"
24 
25 #define GET_SUBTARGETINFO_CTOR
26 #define GET_SUBTARGETINFO_TARGET_DESC
27 #include "AArch64GenSubtargetInfo.inc"
28 
29 static cl::opt<bool>
30 EnableEarlyIfConvert("aarch64-early-ifcvt", cl::desc("Enable the early if "
31                      "converter pass"), cl::init(true), cl::Hidden);
32 
33 // If OS supports TBI, use this flag to enable it.
34 static cl::opt<bool>
35 UseAddressTopByteIgnored("aarch64-use-tbi", cl::desc("Assume that top byte of "
36                          "an address is ignored"), cl::init(false), cl::Hidden);
37 
38 AArch64Subtarget &
39 AArch64Subtarget::initializeSubtargetDependencies(StringRef FS,
40                                                   StringRef CPUString) {
41   // Determine default and user-specified characteristics
42 
43   if (CPUString.empty())
44     CPUString = "generic";
45 
46   ParseSubtargetFeatures(CPUString, FS);
47   initializeProperties();
48 
49   return *this;
50 }
51 
52 void AArch64Subtarget::initializeProperties() {
53   // Initialize CPU specific properties. We should add a tablegen feature for
54   // this in the future so we can specify it together with the subtarget
55   // features.
56   switch (ARMProcFamily) {
57   case Cyclone:
58     CacheLineSize = 64;
59     PrefetchDistance = 280;
60     MinPrefetchStride = 2048;
61     MaxPrefetchIterationsAhead = 3;
62     break;
63   case CortexA57:
64     MaxInterleaveFactor = 4;
65     break;
66   case ExynosM1:
67     MaxInterleaveFactor = 4;
68     PrefFunctionAlignment = 4;
69     PrefLoopAlignment = 3;
70     MaxJumpTableSize = 12;
71     break;
72   case Kryo:
73     MaxInterleaveFactor = 4;
74     VectorInsertExtractBaseCost = 2;
75     CacheLineSize = 128;
76     PrefetchDistance = 740;
77     MinPrefetchStride = 1024;
78     MaxPrefetchIterationsAhead = 11;
79     break;
80   case Vulcan:
81     MaxInterleaveFactor = 4;
82     break;
83   case CortexA35: break;
84   case CortexA53: break;
85   case CortexA72: break;
86   case CortexA73: break;
87   case Others: break;
88   }
89 }
90 
91 AArch64Subtarget::AArch64Subtarget(const Triple &TT, const std::string &CPU,
92                                    const std::string &FS,
93                                    const TargetMachine &TM, bool LittleEndian)
94     : AArch64GenSubtargetInfo(TT, CPU, FS), ReserveX18(TT.isOSDarwin()),
95       IsLittle(LittleEndian), TargetTriple(TT), FrameLowering(),
96       InstrInfo(initializeSubtargetDependencies(FS, CPU)), TSInfo(),
97       TLInfo(TM, *this), GISel() {}
98 
99 const CallLowering *AArch64Subtarget::getCallLowering() const {
100   assert(GISel && "Access to GlobalISel APIs not set");
101   return GISel->getCallLowering();
102 }
103 
104 const InstructionSelector *AArch64Subtarget::getInstructionSelector() const {
105   assert(GISel && "Access to GlobalISel APIs not set");
106   return GISel->getInstructionSelector();
107 }
108 
109 const LegalizerInfo *AArch64Subtarget::getLegalizerInfo() const {
110   assert(GISel && "Access to GlobalISel APIs not set");
111   return GISel->getLegalizerInfo();
112 }
113 
114 const RegisterBankInfo *AArch64Subtarget::getRegBankInfo() const {
115   assert(GISel && "Access to GlobalISel APIs not set");
116   return GISel->getRegBankInfo();
117 }
118 
119 /// Find the target operand flags that describe how a global value should be
120 /// referenced for the current subtarget.
121 unsigned char
122 AArch64Subtarget::ClassifyGlobalReference(const GlobalValue *GV,
123                                           const TargetMachine &TM) const {
124   // MachO large model always goes via a GOT, simply to get a single 8-byte
125   // absolute relocation on all global addresses.
126   if (TM.getCodeModel() == CodeModel::Large && isTargetMachO())
127     return AArch64II::MO_GOT;
128 
129   if (!TM.shouldAssumeDSOLocal(*GV->getParent(), GV))
130     return AArch64II::MO_GOT;
131 
132   // The small code mode's direct accesses use ADRP, which cannot necessarily
133   // produce the value 0 (if the code is above 4GB).
134   if (TM.getCodeModel() == CodeModel::Small && GV->hasExternalWeakLinkage())
135     return AArch64II::MO_GOT;
136 
137   return AArch64II::MO_NO_FLAG;
138 }
139 
140 /// This function returns the name of a function which has an interface
141 /// like the non-standard bzero function, if such a function exists on
142 /// the current subtarget and it is considered prefereable over
143 /// memset with zero passed as the second argument. Otherwise it
144 /// returns null.
145 const char *AArch64Subtarget::getBZeroEntry() const {
146   // Prefer bzero on Darwin only.
147   if(isTargetDarwin())
148     return "bzero";
149 
150   return nullptr;
151 }
152 
153 void AArch64Subtarget::overrideSchedPolicy(MachineSchedPolicy &Policy,
154                                            unsigned NumRegionInstrs) const {
155   // LNT run (at least on Cyclone) showed reasonably significant gains for
156   // bi-directional scheduling. 253.perlbmk.
157   Policy.OnlyTopDown = false;
158   Policy.OnlyBottomUp = false;
159   // Enabling or Disabling the latency heuristic is a close call: It seems to
160   // help nearly no benchmark on out-of-order architectures, on the other hand
161   // it regresses register pressure on a few benchmarking.
162   Policy.DisableLatencyHeuristic = DisableLatencySchedHeuristic;
163 }
164 
165 bool AArch64Subtarget::enableEarlyIfConversion() const {
166   return EnableEarlyIfConvert;
167 }
168 
169 bool AArch64Subtarget::supportsAddressTopByteIgnored() const {
170   if (!UseAddressTopByteIgnored)
171     return false;
172 
173   if (TargetTriple.isiOS()) {
174     unsigned Major, Minor, Micro;
175     TargetTriple.getiOSVersion(Major, Minor, Micro);
176     return Major >= 8;
177   }
178 
179   return false;
180 }
181 
182 std::unique_ptr<PBQPRAConstraint>
183 AArch64Subtarget::getCustomPBQPConstraints() const {
184   return balanceFPOps() ? llvm::make_unique<A57ChainingConstraint>() : nullptr;
185 }
186