1 //=====---- ARMSubtarget.h - Define Subtarget for the ARM -----*- 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 declares the ARM specific subclass of TargetSubtargetInfo. 11 // 12 //===----------------------------------------------------------------------===// 13 14 #ifndef ARMSUBTARGET_H 15 #define ARMSUBTARGET_H 16 17 #include "MCTargetDesc/ARMMCTargetDesc.h" 18 #include "llvm/Target/TargetSubtargetInfo.h" 19 #include "llvm/MC/MCInstrItineraries.h" 20 #include "llvm/ADT/Triple.h" 21 #include <string> 22 23 #define GET_SUBTARGETINFO_HEADER 24 #include "ARMGenSubtargetInfo.inc" 25 26 namespace llvm { 27 class GlobalValue; 28 class StringRef; 29 30 class ARMSubtarget : public ARMGenSubtargetInfo { 31 protected: 32 enum ARMProcFamilyEnum { 33 Others, CortexA8, CortexA9 34 }; 35 36 /// ARMProcFamily - ARM processor family: Cortex-A8, Cortex-A9, and others. 37 ARMProcFamilyEnum ARMProcFamily; 38 39 /// HasV4TOps, HasV5TOps, HasV5TEOps, HasV6Ops, HasV6T2Ops, HasV7Ops - 40 /// Specify whether target support specific ARM ISA variants. 41 bool HasV4TOps; 42 bool HasV5TOps; 43 bool HasV5TEOps; 44 bool HasV6Ops; 45 bool HasV6T2Ops; 46 bool HasV7Ops; 47 48 /// HasVFPv2, HasVFPv3, HasNEON - Specify what floating point ISAs are 49 /// supported. 50 bool HasVFPv2; 51 bool HasVFPv3; 52 bool HasNEON; 53 54 /// UseNEONForSinglePrecisionFP - if the NEONFP attribute has been 55 /// specified. Use the method useNEONForSinglePrecisionFP() to 56 /// determine if NEON should actually be used. 57 bool UseNEONForSinglePrecisionFP; 58 59 /// SlowFPVMLx - If the VFP2 / NEON instructions are available, indicates 60 /// whether the FP VML[AS] instructions are slow (if so, don't use them). 61 bool SlowFPVMLx; 62 63 /// HasVMLxForwarding - If true, NEON has special multiplier accumulator 64 /// forwarding to allow mul + mla being issued back to back. 65 bool HasVMLxForwarding; 66 67 /// SlowFPBrcc - True if floating point compare + branch is slow. 68 bool SlowFPBrcc; 69 70 /// InThumbMode - True if compiling for Thumb, false for ARM. 71 bool InThumbMode; 72 73 /// HasThumb2 - True if Thumb2 instructions are supported. 74 bool HasThumb2; 75 76 /// IsMClass - True if the subtarget belongs to the 'M' profile of CPUs - 77 /// v6m, v7m for example. 78 bool IsMClass; 79 80 /// NoARM - True if subtarget does not support ARM mode execution. 81 bool NoARM; 82 83 /// PostRAScheduler - True if using post-register-allocation scheduler. 84 bool PostRAScheduler; 85 86 /// IsR9Reserved - True if R9 is a not available as general purpose register. 87 bool IsR9Reserved; 88 89 /// UseMovt - True if MOVT / MOVW pairs are used for materialization of 32-bit 90 /// imms (including global addresses). 91 bool UseMovt; 92 93 /// SupportsTailCall - True if the OS supports tail call. The dynamic linker 94 /// must be able to synthesize call stubs for interworking between ARM and 95 /// Thumb. 96 bool SupportsTailCall; 97 98 /// HasFP16 - True if subtarget supports half-precision FP (We support VFP+HF 99 /// only so far) 100 bool HasFP16; 101 102 /// HasD16 - True if subtarget is limited to 16 double precision 103 /// FP registers for VFPv3. 104 bool HasD16; 105 106 /// HasHardwareDivide - True if subtarget supports [su]div 107 bool HasHardwareDivide; 108 109 /// HasT2ExtractPack - True if subtarget supports thumb2 extract/pack 110 /// instructions. 111 bool HasT2ExtractPack; 112 113 /// HasDataBarrier - True if the subtarget supports DMB / DSB data barrier 114 /// instructions. 115 bool HasDataBarrier; 116 117 /// Pref32BitThumb - If true, codegen would prefer 32-bit Thumb instructions 118 /// over 16-bit ones. 119 bool Pref32BitThumb; 120 121 /// AvoidCPSRPartialUpdate - If true, codegen would avoid using instructions 122 /// that partially update CPSR and add false dependency on the previous 123 /// CPSR setting instruction. 124 bool AvoidCPSRPartialUpdate; 125 126 /// HasMPExtension - True if the subtarget supports Multiprocessing 127 /// extension (ARMv7 only). 128 bool HasMPExtension; 129 130 /// FPOnlySP - If true, the floating point unit only supports single 131 /// precision. 132 bool FPOnlySP; 133 134 /// AllowsUnalignedMem - If true, the subtarget allows unaligned memory 135 /// accesses for some types. For details, see 136 /// ARMTargetLowering::allowsUnalignedMemoryAccesses(). 137 bool AllowsUnalignedMem; 138 139 /// Thumb2DSP - If true, the subtarget supports the v7 DSP (saturating arith 140 /// and such) instructions in Thumb2 code. 141 bool Thumb2DSP; 142 143 /// stackAlignment - The minimum alignment known to hold of the stack frame on 144 /// entry to the function and which must be maintained by every function. 145 unsigned stackAlignment; 146 147 /// CPUString - String name of used CPU. 148 std::string CPUString; 149 150 /// TargetTriple - What processor and OS we're targeting. 151 Triple TargetTriple; 152 153 /// Selected instruction itineraries (one entry per itinerary class.) 154 InstrItineraryData InstrItins; 155 156 public: 157 enum { 158 isELF, isDarwin 159 } TargetType; 160 161 enum { 162 ARM_ABI_APCS, 163 ARM_ABI_AAPCS // ARM EABI 164 } TargetABI; 165 166 /// This constructor initializes the data members to match that 167 /// of the specified triple. 168 /// 169 ARMSubtarget(const std::string &TT, const std::string &CPU, 170 const std::string &FS); 171 172 /// getMaxInlineSizeThreshold - Returns the maximum memset / memcpy size 173 /// that still makes it profitable to inline the call. 174 unsigned getMaxInlineSizeThreshold() const { 175 // FIXME: For now, we don't lower memcpy's to loads / stores for Thumb1. 176 // Change this once Thumb1 ldmia / stmia support is added. 177 return isThumb1Only() ? 0 : 64; 178 } 179 /// ParseSubtargetFeatures - Parses features string setting specified 180 /// subtarget options. Definition of function is auto generated by tblgen. 181 void ParseSubtargetFeatures(StringRef CPU, StringRef FS); 182 183 void computeIssueWidth(); 184 185 bool hasV4TOps() const { return HasV4TOps; } 186 bool hasV5TOps() const { return HasV5TOps; } 187 bool hasV5TEOps() const { return HasV5TEOps; } 188 bool hasV6Ops() const { return HasV6Ops; } 189 bool hasV6T2Ops() const { return HasV6T2Ops; } 190 bool hasV7Ops() const { return HasV7Ops; } 191 192 bool isCortexA8() const { return ARMProcFamily == CortexA8; } 193 bool isCortexA9() const { return ARMProcFamily == CortexA9; } 194 195 bool hasARMOps() const { return !NoARM; } 196 197 bool hasVFP2() const { return HasVFPv2; } 198 bool hasVFP3() const { return HasVFPv3; } 199 bool hasNEON() const { return HasNEON; } 200 bool useNEONForSinglePrecisionFP() const { 201 return hasNEON() && UseNEONForSinglePrecisionFP; } 202 203 bool hasDivide() const { return HasHardwareDivide; } 204 bool hasT2ExtractPack() const { return HasT2ExtractPack; } 205 bool hasDataBarrier() const { return HasDataBarrier; } 206 bool useFPVMLx() const { return !SlowFPVMLx; } 207 bool hasVMLxForwarding() const { return HasVMLxForwarding; } 208 bool isFPBrccSlow() const { return SlowFPBrcc; } 209 bool isFPOnlySP() const { return FPOnlySP; } 210 bool prefers32BitThumb() const { return Pref32BitThumb; } 211 bool avoidCPSRPartialUpdate() const { return AvoidCPSRPartialUpdate; } 212 bool hasMPExtension() const { return HasMPExtension; } 213 bool hasThumb2DSP() const { return Thumb2DSP; } 214 215 bool hasFP16() const { return HasFP16; } 216 bool hasD16() const { return HasD16; } 217 218 const Triple &getTargetTriple() const { return TargetTriple; } 219 220 bool isTargetDarwin() const { return TargetTriple.isOSDarwin(); } 221 bool isTargetNaCl() const { 222 return TargetTriple.getOS() == Triple::NativeClient; 223 } 224 bool isTargetELF() const { return !isTargetDarwin(); } 225 226 bool isAPCS_ABI() const { return TargetABI == ARM_ABI_APCS; } 227 bool isAAPCS_ABI() const { return TargetABI == ARM_ABI_AAPCS; } 228 229 bool isThumb() const { return InThumbMode; } 230 bool isThumb1Only() const { return InThumbMode && !HasThumb2; } 231 bool isThumb2() const { return InThumbMode && HasThumb2; } 232 bool hasThumb2() const { return HasThumb2; } 233 bool isMClass() const { return IsMClass; } 234 bool isARClass() const { return !IsMClass; } 235 236 bool isR9Reserved() const { return IsR9Reserved; } 237 238 bool useMovt() const { return UseMovt && hasV6T2Ops(); } 239 bool supportsTailCall() const { return SupportsTailCall; } 240 241 bool allowsUnalignedMem() const { return AllowsUnalignedMem; } 242 243 const std::string & getCPUString() const { return CPUString; } 244 245 unsigned getMispredictionPenalty() const; 246 247 /// enablePostRAScheduler - True at 'More' optimization. 248 bool enablePostRAScheduler(CodeGenOpt::Level OptLevel, 249 TargetSubtargetInfo::AntiDepBreakMode& Mode, 250 RegClassVector& CriticalPathRCs) const; 251 252 /// getInstrItins - Return the instruction itineraies based on subtarget 253 /// selection. 254 const InstrItineraryData &getInstrItineraryData() const { return InstrItins; } 255 256 /// getStackAlignment - Returns the minimum alignment known to hold of the 257 /// stack frame on entry to the function and which must be maintained by every 258 /// function for this subtarget. 259 unsigned getStackAlignment() const { return stackAlignment; } 260 261 /// GVIsIndirectSymbol - true if the GV will be accessed via an indirect 262 /// symbol. 263 bool GVIsIndirectSymbol(const GlobalValue *GV, Reloc::Model RelocM) const; 264 }; 265 } // End llvm namespace 266 267 #endif // ARMSUBTARGET_H 268