|
83
|
1 //===-- ARMTargetTransformInfo.h - ARM specific TTI -------------*- 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 /// \file
|
|
|
10 /// This file a TargetTransformInfo::Concept conforming object specific to the
|
|
|
11 /// ARM target machine. It uses the target's detailed information to
|
|
|
12 /// provide more precise answers to certain TTI queries, while letting the
|
|
|
13 /// target independent and default TTI implementations handle the rest.
|
|
|
14 ///
|
|
|
15 //===----------------------------------------------------------------------===//
|
|
|
16
|
|
|
17 #ifndef LLVM_LIB_TARGET_ARM_ARMTARGETTRANSFORMINFO_H
|
|
|
18 #define LLVM_LIB_TARGET_ARM_ARMTARGETTRANSFORMINFO_H
|
|
|
19
|
|
|
20 #include "ARM.h"
|
|
|
21 #include "ARMTargetMachine.h"
|
|
|
22 #include "llvm/Analysis/TargetTransformInfo.h"
|
|
|
23 #include "llvm/CodeGen/BasicTTIImpl.h"
|
|
|
24 #include "llvm/Target/TargetLowering.h"
|
|
|
25
|
|
|
26 namespace llvm {
|
|
|
27
|
|
|
28 class ARMTTIImpl : public BasicTTIImplBase<ARMTTIImpl> {
|
|
|
29 typedef BasicTTIImplBase<ARMTTIImpl> BaseT;
|
|
|
30 typedef TargetTransformInfo TTI;
|
|
|
31 friend BaseT;
|
|
|
32
|
|
|
33 const ARMSubtarget *ST;
|
|
|
34 const ARMTargetLowering *TLI;
|
|
|
35
|
|
|
36 /// Estimate the overhead of scalarizing an instruction. Insert and Extract
|
|
|
37 /// are set if the result needs to be inserted and/or extracted from vectors.
|
|
|
38 unsigned getScalarizationOverhead(Type *Ty, bool Insert, bool Extract);
|
|
|
39
|
|
|
40 const ARMSubtarget *getST() const { return ST; }
|
|
|
41 const ARMTargetLowering *getTLI() const { return TLI; }
|
|
|
42
|
|
|
43 public:
|
|
95
|
44 explicit ARMTTIImpl(const ARMBaseTargetMachine *TM, const Function &F)
|
|
|
45 : BaseT(TM, F.getParent()->getDataLayout()), ST(TM->getSubtargetImpl(F)),
|
|
|
46 TLI(ST->getTargetLowering()) {}
|
|
83
|
47
|
|
120
|
48 bool enableInterleavedAccessVectorization() { return true; }
|
|
95
|
49
|
|
120
|
50 /// Floating-point computation using ARMv8 AArch32 Advanced
|
|
|
51 /// SIMD instructions remains unchanged from ARMv7. Only AArch64 SIMD
|
|
|
52 /// is IEEE-754 compliant, but it's not covered in this target.
|
|
|
53 bool isFPVectorizationPotentiallyUnsafe() {
|
|
|
54 return !ST->isTargetDarwin();
|
|
|
55 }
|
|
83
|
56
|
|
|
57 /// \name Scalar TTI Implementations
|
|
|
58 /// @{
|
|
|
59
|
|
120
|
60 int getIntImmCodeSizeCost(unsigned Opcode, unsigned Idx, const APInt &Imm,
|
|
|
61 Type *Ty);
|
|
|
62
|
|
83
|
63 using BaseT::getIntImmCost;
|
|
95
|
64 int getIntImmCost(const APInt &Imm, Type *Ty);
|
|
83
|
65
|
|
120
|
66 int getIntImmCost(unsigned Opcode, unsigned Idx, const APInt &Imm, Type *Ty);
|
|
|
67
|
|
83
|
68 /// @}
|
|
|
69
|
|
|
70 /// \name Vector TTI Implementations
|
|
|
71 /// @{
|
|
|
72
|
|
|
73 unsigned getNumberOfRegisters(bool Vector) {
|
|
|
74 if (Vector) {
|
|
|
75 if (ST->hasNEON())
|
|
|
76 return 16;
|
|
|
77 return 0;
|
|
|
78 }
|
|
|
79
|
|
|
80 if (ST->isThumb1Only())
|
|
|
81 return 8;
|
|
|
82 return 13;
|
|
|
83 }
|
|
|
84
|
|
|
85 unsigned getRegisterBitWidth(bool Vector) {
|
|
|
86 if (Vector) {
|
|
|
87 if (ST->hasNEON())
|
|
|
88 return 128;
|
|
|
89 return 0;
|
|
|
90 }
|
|
|
91
|
|
|
92 return 32;
|
|
|
93 }
|
|
|
94
|
|
95
|
95 unsigned getMaxInterleaveFactor(unsigned VF) {
|
|
120
|
96 return ST->getMaxInterleaveFactor();
|
|
83
|
97 }
|
|
|
98
|
|
95
|
99 int getShuffleCost(TTI::ShuffleKind Kind, Type *Tp, int Index, Type *SubTp);
|
|
83
|
100
|
|
95
|
101 int getCastInstrCost(unsigned Opcode, Type *Dst, Type *Src);
|
|
83
|
102
|
|
95
|
103 int getCmpSelInstrCost(unsigned Opcode, Type *ValTy, Type *CondTy);
|
|
83
|
104
|
|
95
|
105 int getVectorInstrCost(unsigned Opcode, Type *Val, unsigned Index);
|
|
83
|
106
|
|
95
|
107 int getAddressComputationCost(Type *Val, bool IsComplex);
|
|
83
|
108
|
|
95
|
109 int getFPOpCost(Type *Ty);
|
|
83
|
110
|
|
95
|
111 int getArithmeticInstrCost(
|
|
83
|
112 unsigned Opcode, Type *Ty,
|
|
|
113 TTI::OperandValueKind Op1Info = TTI::OK_AnyValue,
|
|
|
114 TTI::OperandValueKind Op2Info = TTI::OK_AnyValue,
|
|
|
115 TTI::OperandValueProperties Opd1PropInfo = TTI::OP_None,
|
|
|
116 TTI::OperandValueProperties Opd2PropInfo = TTI::OP_None);
|
|
|
117
|
|
95
|
118 int getMemoryOpCost(unsigned Opcode, Type *Src, unsigned Alignment,
|
|
|
119 unsigned AddressSpace);
|
|
83
|
120
|
|
95
|
121 int getInterleavedMemoryOpCost(unsigned Opcode, Type *VecTy, unsigned Factor,
|
|
|
122 ArrayRef<unsigned> Indices, unsigned Alignment,
|
|
|
123 unsigned AddressSpace);
|
|
120
|
124
|
|
|
125 bool shouldBuildLookupTablesForConstant(Constant *C) const {
|
|
|
126 // In the ROPI and RWPI relocation models we can't have pointers to global
|
|
|
127 // variables or functions in constant data, so don't convert switches to
|
|
|
128 // lookup tables if any of the values would need relocation.
|
|
|
129 if (ST->isROPI() || ST->isRWPI())
|
|
|
130 return !C->needsRelocation();
|
|
|
131
|
|
|
132 return true;
|
|
|
133 }
|
|
83
|
134 /// @}
|
|
|
135 };
|
|
|
136
|
|
|
137 } // end namespace llvm
|
|
|
138
|
|
|
139 #endif
|
