TargetLowering.cpp revision 7ef91d5d2ed4129318c2e297cac5601bbc8c0c75
1//===-- TargetLowering.cpp - Implement the TargetLowering class -----------===// 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 implements the TargetLowering class. 11// 12//===----------------------------------------------------------------------===// 13 14#include "llvm/Target/TargetLowering.h" 15#include "llvm/MC/MCAsmInfo.h" 16#include "llvm/MC/MCExpr.h" 17#include "llvm/Target/TargetData.h" 18#include "llvm/Target/TargetLoweringObjectFile.h" 19#include "llvm/Target/TargetMachine.h" 20#include "llvm/Target/TargetRegisterInfo.h" 21#include "llvm/Target/TargetSubtarget.h" 22#include "llvm/GlobalVariable.h" 23#include "llvm/DerivedTypes.h" 24#include "llvm/CodeGen/MachineFrameInfo.h" 25#include "llvm/CodeGen/MachineJumpTableInfo.h" 26#include "llvm/CodeGen/MachineFunction.h" 27#include "llvm/CodeGen/SelectionDAG.h" 28#include "llvm/ADT/STLExtras.h" 29#include "llvm/Support/ErrorHandling.h" 30#include "llvm/Support/MathExtras.h" 31using namespace llvm; 32 33namespace llvm { 34TLSModel::Model getTLSModel(const GlobalValue *GV, Reloc::Model reloc) { 35 bool isLocal = GV->hasLocalLinkage(); 36 bool isDeclaration = GV->isDeclaration(); 37 // FIXME: what should we do for protected and internal visibility? 38 // For variables, is internal different from hidden? 39 bool isHidden = GV->hasHiddenVisibility(); 40 41 if (reloc == Reloc::PIC_) { 42 if (isLocal || isHidden) 43 return TLSModel::LocalDynamic; 44 else 45 return TLSModel::GeneralDynamic; 46 } else { 47 if (!isDeclaration || isHidden) 48 return TLSModel::LocalExec; 49 else 50 return TLSModel::InitialExec; 51 } 52} 53} 54 55/// InitLibcallNames - Set default libcall names. 56/// 57static void InitLibcallNames(const char **Names) { 58 Names[RTLIB::SHL_I16] = "__ashlhi3"; 59 Names[RTLIB::SHL_I32] = "__ashlsi3"; 60 Names[RTLIB::SHL_I64] = "__ashldi3"; 61 Names[RTLIB::SHL_I128] = "__ashlti3"; 62 Names[RTLIB::SRL_I16] = "__lshrhi3"; 63 Names[RTLIB::SRL_I32] = "__lshrsi3"; 64 Names[RTLIB::SRL_I64] = "__lshrdi3"; 65 Names[RTLIB::SRL_I128] = "__lshrti3"; 66 Names[RTLIB::SRA_I16] = "__ashrhi3"; 67 Names[RTLIB::SRA_I32] = "__ashrsi3"; 68 Names[RTLIB::SRA_I64] = "__ashrdi3"; 69 Names[RTLIB::SRA_I128] = "__ashrti3"; 70 Names[RTLIB::MUL_I8] = "__mulqi3"; 71 Names[RTLIB::MUL_I16] = "__mulhi3"; 72 Names[RTLIB::MUL_I32] = "__mulsi3"; 73 Names[RTLIB::MUL_I64] = "__muldi3"; 74 Names[RTLIB::MUL_I128] = "__multi3"; 75 Names[RTLIB::SDIV_I8] = "__divqi3"; 76 Names[RTLIB::SDIV_I16] = "__divhi3"; 77 Names[RTLIB::SDIV_I32] = "__divsi3"; 78 Names[RTLIB::SDIV_I64] = "__divdi3"; 79 Names[RTLIB::SDIV_I128] = "__divti3"; 80 Names[RTLIB::UDIV_I8] = "__udivqi3"; 81 Names[RTLIB::UDIV_I16] = "__udivhi3"; 82 Names[RTLIB::UDIV_I32] = "__udivsi3"; 83 Names[RTLIB::UDIV_I64] = "__udivdi3"; 84 Names[RTLIB::UDIV_I128] = "__udivti3"; 85 Names[RTLIB::SREM_I8] = "__modqi3"; 86 Names[RTLIB::SREM_I16] = "__modhi3"; 87 Names[RTLIB::SREM_I32] = "__modsi3"; 88 Names[RTLIB::SREM_I64] = "__moddi3"; 89 Names[RTLIB::SREM_I128] = "__modti3"; 90 Names[RTLIB::UREM_I8] = "__umodqi3"; 91 Names[RTLIB::UREM_I16] = "__umodhi3"; 92 Names[RTLIB::UREM_I32] = "__umodsi3"; 93 Names[RTLIB::UREM_I64] = "__umoddi3"; 94 Names[RTLIB::UREM_I128] = "__umodti3"; 95 Names[RTLIB::NEG_I32] = "__negsi2"; 96 Names[RTLIB::NEG_I64] = "__negdi2"; 97 Names[RTLIB::ADD_F32] = "__addsf3"; 98 Names[RTLIB::ADD_F64] = "__adddf3"; 99 Names[RTLIB::ADD_F80] = "__addxf3"; 100 Names[RTLIB::ADD_PPCF128] = "__gcc_qadd"; 101 Names[RTLIB::SUB_F32] = "__subsf3"; 102 Names[RTLIB::SUB_F64] = "__subdf3"; 103 Names[RTLIB::SUB_F80] = "__subxf3"; 104 Names[RTLIB::SUB_PPCF128] = "__gcc_qsub"; 105 Names[RTLIB::MUL_F32] = "__mulsf3"; 106 Names[RTLIB::MUL_F64] = "__muldf3"; 107 Names[RTLIB::MUL_F80] = "__mulxf3"; 108 Names[RTLIB::MUL_PPCF128] = "__gcc_qmul"; 109 Names[RTLIB::DIV_F32] = "__divsf3"; 110 Names[RTLIB::DIV_F64] = "__divdf3"; 111 Names[RTLIB::DIV_F80] = "__divxf3"; 112 Names[RTLIB::DIV_PPCF128] = "__gcc_qdiv"; 113 Names[RTLIB::REM_F32] = "fmodf"; 114 Names[RTLIB::REM_F64] = "fmod"; 115 Names[RTLIB::REM_F80] = "fmodl"; 116 Names[RTLIB::REM_PPCF128] = "fmodl"; 117 Names[RTLIB::POWI_F32] = "__powisf2"; 118 Names[RTLIB::POWI_F64] = "__powidf2"; 119 Names[RTLIB::POWI_F80] = "__powixf2"; 120 Names[RTLIB::POWI_PPCF128] = "__powitf2"; 121 Names[RTLIB::SQRT_F32] = "sqrtf"; 122 Names[RTLIB::SQRT_F64] = "sqrt"; 123 Names[RTLIB::SQRT_F80] = "sqrtl"; 124 Names[RTLIB::SQRT_PPCF128] = "sqrtl"; 125 Names[RTLIB::LOG_F32] = "logf"; 126 Names[RTLIB::LOG_F64] = "log"; 127 Names[RTLIB::LOG_F80] = "logl"; 128 Names[RTLIB::LOG_PPCF128] = "logl"; 129 Names[RTLIB::LOG2_F32] = "log2f"; 130 Names[RTLIB::LOG2_F64] = "log2"; 131 Names[RTLIB::LOG2_F80] = "log2l"; 132 Names[RTLIB::LOG2_PPCF128] = "log2l"; 133 Names[RTLIB::LOG10_F32] = "log10f"; 134 Names[RTLIB::LOG10_F64] = "log10"; 135 Names[RTLIB::LOG10_F80] = "log10l"; 136 Names[RTLIB::LOG10_PPCF128] = "log10l"; 137 Names[RTLIB::EXP_F32] = "expf"; 138 Names[RTLIB::EXP_F64] = "exp"; 139 Names[RTLIB::EXP_F80] = "expl"; 140 Names[RTLIB::EXP_PPCF128] = "expl"; 141 Names[RTLIB::EXP2_F32] = "exp2f"; 142 Names[RTLIB::EXP2_F64] = "exp2"; 143 Names[RTLIB::EXP2_F80] = "exp2l"; 144 Names[RTLIB::EXP2_PPCF128] = "exp2l"; 145 Names[RTLIB::SIN_F32] = "sinf"; 146 Names[RTLIB::SIN_F64] = "sin"; 147 Names[RTLIB::SIN_F80] = "sinl"; 148 Names[RTLIB::SIN_PPCF128] = "sinl"; 149 Names[RTLIB::COS_F32] = "cosf"; 150 Names[RTLIB::COS_F64] = "cos"; 151 Names[RTLIB::COS_F80] = "cosl"; 152 Names[RTLIB::COS_PPCF128] = "cosl"; 153 Names[RTLIB::POW_F32] = "powf"; 154 Names[RTLIB::POW_F64] = "pow"; 155 Names[RTLIB::POW_F80] = "powl"; 156 Names[RTLIB::POW_PPCF128] = "powl"; 157 Names[RTLIB::CEIL_F32] = "ceilf"; 158 Names[RTLIB::CEIL_F64] = "ceil"; 159 Names[RTLIB::CEIL_F80] = "ceill"; 160 Names[RTLIB::CEIL_PPCF128] = "ceill"; 161 Names[RTLIB::TRUNC_F32] = "truncf"; 162 Names[RTLIB::TRUNC_F64] = "trunc"; 163 Names[RTLIB::TRUNC_F80] = "truncl"; 164 Names[RTLIB::TRUNC_PPCF128] = "truncl"; 165 Names[RTLIB::RINT_F32] = "rintf"; 166 Names[RTLIB::RINT_F64] = "rint"; 167 Names[RTLIB::RINT_F80] = "rintl"; 168 Names[RTLIB::RINT_PPCF128] = "rintl"; 169 Names[RTLIB::NEARBYINT_F32] = "nearbyintf"; 170 Names[RTLIB::NEARBYINT_F64] = "nearbyint"; 171 Names[RTLIB::NEARBYINT_F80] = "nearbyintl"; 172 Names[RTLIB::NEARBYINT_PPCF128] = "nearbyintl"; 173 Names[RTLIB::FLOOR_F32] = "floorf"; 174 Names[RTLIB::FLOOR_F64] = "floor"; 175 Names[RTLIB::FLOOR_F80] = "floorl"; 176 Names[RTLIB::FLOOR_PPCF128] = "floorl"; 177 Names[RTLIB::FPEXT_F32_F64] = "__extendsfdf2"; 178 Names[RTLIB::FPROUND_F64_F32] = "__truncdfsf2"; 179 Names[RTLIB::FPROUND_F80_F32] = "__truncxfsf2"; 180 Names[RTLIB::FPROUND_PPCF128_F32] = "__trunctfsf2"; 181 Names[RTLIB::FPROUND_F80_F64] = "__truncxfdf2"; 182 Names[RTLIB::FPROUND_PPCF128_F64] = "__trunctfdf2"; 183 Names[RTLIB::FPTOSINT_F32_I8] = "__fixsfi8"; 184 Names[RTLIB::FPTOSINT_F32_I16] = "__fixsfi16"; 185 Names[RTLIB::FPTOSINT_F32_I32] = "__fixsfsi"; 186 Names[RTLIB::FPTOSINT_F32_I64] = "__fixsfdi"; 187 Names[RTLIB::FPTOSINT_F32_I128] = "__fixsfti"; 188 Names[RTLIB::FPTOSINT_F64_I32] = "__fixdfsi"; 189 Names[RTLIB::FPTOSINT_F64_I64] = "__fixdfdi"; 190 Names[RTLIB::FPTOSINT_F64_I128] = "__fixdfti"; 191 Names[RTLIB::FPTOSINT_F80_I32] = "__fixxfsi"; 192 Names[RTLIB::FPTOSINT_F80_I64] = "__fixxfdi"; 193 Names[RTLIB::FPTOSINT_F80_I128] = "__fixxfti"; 194 Names[RTLIB::FPTOSINT_PPCF128_I32] = "__fixtfsi"; 195 Names[RTLIB::FPTOSINT_PPCF128_I64] = "__fixtfdi"; 196 Names[RTLIB::FPTOSINT_PPCF128_I128] = "__fixtfti"; 197 Names[RTLIB::FPTOUINT_F32_I8] = "__fixunssfi8"; 198 Names[RTLIB::FPTOUINT_F32_I16] = "__fixunssfi16"; 199 Names[RTLIB::FPTOUINT_F32_I32] = "__fixunssfsi"; 200 Names[RTLIB::FPTOUINT_F32_I64] = "__fixunssfdi"; 201 Names[RTLIB::FPTOUINT_F32_I128] = "__fixunssfti"; 202 Names[RTLIB::FPTOUINT_F64_I32] = "__fixunsdfsi"; 203 Names[RTLIB::FPTOUINT_F64_I64] = "__fixunsdfdi"; 204 Names[RTLIB::FPTOUINT_F64_I128] = "__fixunsdfti"; 205 Names[RTLIB::FPTOUINT_F80_I32] = "__fixunsxfsi"; 206 Names[RTLIB::FPTOUINT_F80_I64] = "__fixunsxfdi"; 207 Names[RTLIB::FPTOUINT_F80_I128] = "__fixunsxfti"; 208 Names[RTLIB::FPTOUINT_PPCF128_I32] = "__fixunstfsi"; 209 Names[RTLIB::FPTOUINT_PPCF128_I64] = "__fixunstfdi"; 210 Names[RTLIB::FPTOUINT_PPCF128_I128] = "__fixunstfti"; 211 Names[RTLIB::SINTTOFP_I32_F32] = "__floatsisf"; 212 Names[RTLIB::SINTTOFP_I32_F64] = "__floatsidf"; 213 Names[RTLIB::SINTTOFP_I32_F80] = "__floatsixf"; 214 Names[RTLIB::SINTTOFP_I32_PPCF128] = "__floatsitf"; 215 Names[RTLIB::SINTTOFP_I64_F32] = "__floatdisf"; 216 Names[RTLIB::SINTTOFP_I64_F64] = "__floatdidf"; 217 Names[RTLIB::SINTTOFP_I64_F80] = "__floatdixf"; 218 Names[RTLIB::SINTTOFP_I64_PPCF128] = "__floatditf"; 219 Names[RTLIB::SINTTOFP_I128_F32] = "__floattisf"; 220 Names[RTLIB::SINTTOFP_I128_F64] = "__floattidf"; 221 Names[RTLIB::SINTTOFP_I128_F80] = "__floattixf"; 222 Names[RTLIB::SINTTOFP_I128_PPCF128] = "__floattitf"; 223 Names[RTLIB::UINTTOFP_I32_F32] = "__floatunsisf"; 224 Names[RTLIB::UINTTOFP_I32_F64] = "__floatunsidf"; 225 Names[RTLIB::UINTTOFP_I32_F80] = "__floatunsixf"; 226 Names[RTLIB::UINTTOFP_I32_PPCF128] = "__floatunsitf"; 227 Names[RTLIB::UINTTOFP_I64_F32] = "__floatundisf"; 228 Names[RTLIB::UINTTOFP_I64_F64] = "__floatundidf"; 229 Names[RTLIB::UINTTOFP_I64_F80] = "__floatundixf"; 230 Names[RTLIB::UINTTOFP_I64_PPCF128] = "__floatunditf"; 231 Names[RTLIB::UINTTOFP_I128_F32] = "__floatuntisf"; 232 Names[RTLIB::UINTTOFP_I128_F64] = "__floatuntidf"; 233 Names[RTLIB::UINTTOFP_I128_F80] = "__floatuntixf"; 234 Names[RTLIB::UINTTOFP_I128_PPCF128] = "__floatuntitf"; 235 Names[RTLIB::OEQ_F32] = "__eqsf2"; 236 Names[RTLIB::OEQ_F64] = "__eqdf2"; 237 Names[RTLIB::UNE_F32] = "__nesf2"; 238 Names[RTLIB::UNE_F64] = "__nedf2"; 239 Names[RTLIB::OGE_F32] = "__gesf2"; 240 Names[RTLIB::OGE_F64] = "__gedf2"; 241 Names[RTLIB::OLT_F32] = "__ltsf2"; 242 Names[RTLIB::OLT_F64] = "__ltdf2"; 243 Names[RTLIB::OLE_F32] = "__lesf2"; 244 Names[RTLIB::OLE_F64] = "__ledf2"; 245 Names[RTLIB::OGT_F32] = "__gtsf2"; 246 Names[RTLIB::OGT_F64] = "__gtdf2"; 247 Names[RTLIB::UO_F32] = "__unordsf2"; 248 Names[RTLIB::UO_F64] = "__unorddf2"; 249 Names[RTLIB::O_F32] = "__unordsf2"; 250 Names[RTLIB::O_F64] = "__unorddf2"; 251 Names[RTLIB::MEMCPY] = "memcpy"; 252 Names[RTLIB::MEMMOVE] = "memmove"; 253 Names[RTLIB::MEMSET] = "memset"; 254 Names[RTLIB::UNWIND_RESUME] = "_Unwind_Resume"; 255} 256 257/// InitLibcallCallingConvs - Set default libcall CallingConvs. 258/// 259static void InitLibcallCallingConvs(CallingConv::ID *CCs) { 260 for (int i = 0; i < RTLIB::UNKNOWN_LIBCALL; ++i) { 261 CCs[i] = CallingConv::C; 262 } 263} 264 265/// getFPEXT - Return the FPEXT_*_* value for the given types, or 266/// UNKNOWN_LIBCALL if there is none. 267RTLIB::Libcall RTLIB::getFPEXT(EVT OpVT, EVT RetVT) { 268 if (OpVT == MVT::f32) { 269 if (RetVT == MVT::f64) 270 return FPEXT_F32_F64; 271 } 272 return UNKNOWN_LIBCALL; 273} 274 275/// getFPROUND - Return the FPROUND_*_* value for the given types, or 276/// UNKNOWN_LIBCALL if there is none. 277RTLIB::Libcall RTLIB::getFPROUND(EVT OpVT, EVT RetVT) { 278 if (RetVT == MVT::f32) { 279 if (OpVT == MVT::f64) 280 return FPROUND_F64_F32; 281 if (OpVT == MVT::f80) 282 return FPROUND_F80_F32; 283 if (OpVT == MVT::ppcf128) 284 return FPROUND_PPCF128_F32; 285 } else if (RetVT == MVT::f64) { 286 if (OpVT == MVT::f80) 287 return FPROUND_F80_F64; 288 if (OpVT == MVT::ppcf128) 289 return FPROUND_PPCF128_F64; 290 } 291 return UNKNOWN_LIBCALL; 292} 293 294/// getFPTOSINT - Return the FPTOSINT_*_* value for the given types, or 295/// UNKNOWN_LIBCALL if there is none. 296RTLIB::Libcall RTLIB::getFPTOSINT(EVT OpVT, EVT RetVT) { 297 if (OpVT == MVT::f32) { 298 if (RetVT == MVT::i8) 299 return FPTOSINT_F32_I8; 300 if (RetVT == MVT::i16) 301 return FPTOSINT_F32_I16; 302 if (RetVT == MVT::i32) 303 return FPTOSINT_F32_I32; 304 if (RetVT == MVT::i64) 305 return FPTOSINT_F32_I64; 306 if (RetVT == MVT::i128) 307 return FPTOSINT_F32_I128; 308 } else if (OpVT == MVT::f64) { 309 if (RetVT == MVT::i32) 310 return FPTOSINT_F64_I32; 311 if (RetVT == MVT::i64) 312 return FPTOSINT_F64_I64; 313 if (RetVT == MVT::i128) 314 return FPTOSINT_F64_I128; 315 } else if (OpVT == MVT::f80) { 316 if (RetVT == MVT::i32) 317 return FPTOSINT_F80_I32; 318 if (RetVT == MVT::i64) 319 return FPTOSINT_F80_I64; 320 if (RetVT == MVT::i128) 321 return FPTOSINT_F80_I128; 322 } else if (OpVT == MVT::ppcf128) { 323 if (RetVT == MVT::i32) 324 return FPTOSINT_PPCF128_I32; 325 if (RetVT == MVT::i64) 326 return FPTOSINT_PPCF128_I64; 327 if (RetVT == MVT::i128) 328 return FPTOSINT_PPCF128_I128; 329 } 330 return UNKNOWN_LIBCALL; 331} 332 333/// getFPTOUINT - Return the FPTOUINT_*_* value for the given types, or 334/// UNKNOWN_LIBCALL if there is none. 335RTLIB::Libcall RTLIB::getFPTOUINT(EVT OpVT, EVT RetVT) { 336 if (OpVT == MVT::f32) { 337 if (RetVT == MVT::i8) 338 return FPTOUINT_F32_I8; 339 if (RetVT == MVT::i16) 340 return FPTOUINT_F32_I16; 341 if (RetVT == MVT::i32) 342 return FPTOUINT_F32_I32; 343 if (RetVT == MVT::i64) 344 return FPTOUINT_F32_I64; 345 if (RetVT == MVT::i128) 346 return FPTOUINT_F32_I128; 347 } else if (OpVT == MVT::f64) { 348 if (RetVT == MVT::i32) 349 return FPTOUINT_F64_I32; 350 if (RetVT == MVT::i64) 351 return FPTOUINT_F64_I64; 352 if (RetVT == MVT::i128) 353 return FPTOUINT_F64_I128; 354 } else if (OpVT == MVT::f80) { 355 if (RetVT == MVT::i32) 356 return FPTOUINT_F80_I32; 357 if (RetVT == MVT::i64) 358 return FPTOUINT_F80_I64; 359 if (RetVT == MVT::i128) 360 return FPTOUINT_F80_I128; 361 } else if (OpVT == MVT::ppcf128) { 362 if (RetVT == MVT::i32) 363 return FPTOUINT_PPCF128_I32; 364 if (RetVT == MVT::i64) 365 return FPTOUINT_PPCF128_I64; 366 if (RetVT == MVT::i128) 367 return FPTOUINT_PPCF128_I128; 368 } 369 return UNKNOWN_LIBCALL; 370} 371 372/// getSINTTOFP - Return the SINTTOFP_*_* value for the given types, or 373/// UNKNOWN_LIBCALL if there is none. 374RTLIB::Libcall RTLIB::getSINTTOFP(EVT OpVT, EVT RetVT) { 375 if (OpVT == MVT::i32) { 376 if (RetVT == MVT::f32) 377 return SINTTOFP_I32_F32; 378 else if (RetVT == MVT::f64) 379 return SINTTOFP_I32_F64; 380 else if (RetVT == MVT::f80) 381 return SINTTOFP_I32_F80; 382 else if (RetVT == MVT::ppcf128) 383 return SINTTOFP_I32_PPCF128; 384 } else if (OpVT == MVT::i64) { 385 if (RetVT == MVT::f32) 386 return SINTTOFP_I64_F32; 387 else if (RetVT == MVT::f64) 388 return SINTTOFP_I64_F64; 389 else if (RetVT == MVT::f80) 390 return SINTTOFP_I64_F80; 391 else if (RetVT == MVT::ppcf128) 392 return SINTTOFP_I64_PPCF128; 393 } else if (OpVT == MVT::i128) { 394 if (RetVT == MVT::f32) 395 return SINTTOFP_I128_F32; 396 else if (RetVT == MVT::f64) 397 return SINTTOFP_I128_F64; 398 else if (RetVT == MVT::f80) 399 return SINTTOFP_I128_F80; 400 else if (RetVT == MVT::ppcf128) 401 return SINTTOFP_I128_PPCF128; 402 } 403 return UNKNOWN_LIBCALL; 404} 405 406/// getUINTTOFP - Return the UINTTOFP_*_* value for the given types, or 407/// UNKNOWN_LIBCALL if there is none. 408RTLIB::Libcall RTLIB::getUINTTOFP(EVT OpVT, EVT RetVT) { 409 if (OpVT == MVT::i32) { 410 if (RetVT == MVT::f32) 411 return UINTTOFP_I32_F32; 412 else if (RetVT == MVT::f64) 413 return UINTTOFP_I32_F64; 414 else if (RetVT == MVT::f80) 415 return UINTTOFP_I32_F80; 416 else if (RetVT == MVT::ppcf128) 417 return UINTTOFP_I32_PPCF128; 418 } else if (OpVT == MVT::i64) { 419 if (RetVT == MVT::f32) 420 return UINTTOFP_I64_F32; 421 else if (RetVT == MVT::f64) 422 return UINTTOFP_I64_F64; 423 else if (RetVT == MVT::f80) 424 return UINTTOFP_I64_F80; 425 else if (RetVT == MVT::ppcf128) 426 return UINTTOFP_I64_PPCF128; 427 } else if (OpVT == MVT::i128) { 428 if (RetVT == MVT::f32) 429 return UINTTOFP_I128_F32; 430 else if (RetVT == MVT::f64) 431 return UINTTOFP_I128_F64; 432 else if (RetVT == MVT::f80) 433 return UINTTOFP_I128_F80; 434 else if (RetVT == MVT::ppcf128) 435 return UINTTOFP_I128_PPCF128; 436 } 437 return UNKNOWN_LIBCALL; 438} 439 440/// InitCmpLibcallCCs - Set default comparison libcall CC. 441/// 442static void InitCmpLibcallCCs(ISD::CondCode *CCs) { 443 memset(CCs, ISD::SETCC_INVALID, sizeof(ISD::CondCode)*RTLIB::UNKNOWN_LIBCALL); 444 CCs[RTLIB::OEQ_F32] = ISD::SETEQ; 445 CCs[RTLIB::OEQ_F64] = ISD::SETEQ; 446 CCs[RTLIB::UNE_F32] = ISD::SETNE; 447 CCs[RTLIB::UNE_F64] = ISD::SETNE; 448 CCs[RTLIB::OGE_F32] = ISD::SETGE; 449 CCs[RTLIB::OGE_F64] = ISD::SETGE; 450 CCs[RTLIB::OLT_F32] = ISD::SETLT; 451 CCs[RTLIB::OLT_F64] = ISD::SETLT; 452 CCs[RTLIB::OLE_F32] = ISD::SETLE; 453 CCs[RTLIB::OLE_F64] = ISD::SETLE; 454 CCs[RTLIB::OGT_F32] = ISD::SETGT; 455 CCs[RTLIB::OGT_F64] = ISD::SETGT; 456 CCs[RTLIB::UO_F32] = ISD::SETNE; 457 CCs[RTLIB::UO_F64] = ISD::SETNE; 458 CCs[RTLIB::O_F32] = ISD::SETEQ; 459 CCs[RTLIB::O_F64] = ISD::SETEQ; 460} 461 462/// NOTE: The constructor takes ownership of TLOF. 463TargetLowering::TargetLowering(TargetMachine &tm,TargetLoweringObjectFile *tlof) 464 : TM(tm), TD(TM.getTargetData()), TLOF(*tlof) { 465 // All operations default to being supported. 466 memset(OpActions, 0, sizeof(OpActions)); 467 memset(LoadExtActions, 0, sizeof(LoadExtActions)); 468 memset(TruncStoreActions, 0, sizeof(TruncStoreActions)); 469 memset(IndexedModeActions, 0, sizeof(IndexedModeActions)); 470 memset(ConvertActions, 0, sizeof(ConvertActions)); 471 memset(CondCodeActions, 0, sizeof(CondCodeActions)); 472 473 // Set default actions for various operations. 474 for (unsigned VT = 0; VT != (unsigned)MVT::LAST_VALUETYPE; ++VT) { 475 // Default all indexed load / store to expand. 476 for (unsigned IM = (unsigned)ISD::PRE_INC; 477 IM != (unsigned)ISD::LAST_INDEXED_MODE; ++IM) { 478 setIndexedLoadAction(IM, (MVT::SimpleValueType)VT, Expand); 479 setIndexedStoreAction(IM, (MVT::SimpleValueType)VT, Expand); 480 } 481 482 // These operations default to expand. 483 setOperationAction(ISD::FGETSIGN, (MVT::SimpleValueType)VT, Expand); 484 setOperationAction(ISD::CONCAT_VECTORS, (MVT::SimpleValueType)VT, Expand); 485 } 486 487 // Most targets ignore the @llvm.prefetch intrinsic. 488 setOperationAction(ISD::PREFETCH, MVT::Other, Expand); 489 490 // ConstantFP nodes default to expand. Targets can either change this to 491 // Legal, in which case all fp constants are legal, or use isFPImmLegal() 492 // to optimize expansions for certain constants. 493 setOperationAction(ISD::ConstantFP, MVT::f32, Expand); 494 setOperationAction(ISD::ConstantFP, MVT::f64, Expand); 495 setOperationAction(ISD::ConstantFP, MVT::f80, Expand); 496 497 // These library functions default to expand. 498 setOperationAction(ISD::FLOG , MVT::f64, Expand); 499 setOperationAction(ISD::FLOG2, MVT::f64, Expand); 500 setOperationAction(ISD::FLOG10,MVT::f64, Expand); 501 setOperationAction(ISD::FEXP , MVT::f64, Expand); 502 setOperationAction(ISD::FEXP2, MVT::f64, Expand); 503 setOperationAction(ISD::FLOG , MVT::f32, Expand); 504 setOperationAction(ISD::FLOG2, MVT::f32, Expand); 505 setOperationAction(ISD::FLOG10,MVT::f32, Expand); 506 setOperationAction(ISD::FEXP , MVT::f32, Expand); 507 setOperationAction(ISD::FEXP2, MVT::f32, Expand); 508 509 // Default ISD::TRAP to expand (which turns it into abort). 510 setOperationAction(ISD::TRAP, MVT::Other, Expand); 511 512 IsLittleEndian = TD->isLittleEndian(); 513 ShiftAmountTy = PointerTy = MVT::getIntegerVT(8*TD->getPointerSize()); 514 memset(RegClassForVT, 0,MVT::LAST_VALUETYPE*sizeof(TargetRegisterClass*)); 515 memset(TargetDAGCombineArray, 0, array_lengthof(TargetDAGCombineArray)); 516 maxStoresPerMemset = maxStoresPerMemcpy = maxStoresPerMemmove = 8; 517 benefitFromCodePlacementOpt = false; 518 UseUnderscoreSetJmp = false; 519 UseUnderscoreLongJmp = false; 520 SelectIsExpensive = false; 521 IntDivIsCheap = false; 522 Pow2DivIsCheap = false; 523 StackPointerRegisterToSaveRestore = 0; 524 ExceptionPointerRegister = 0; 525 ExceptionSelectorRegister = 0; 526 BooleanContents = UndefinedBooleanContent; 527 SchedPreferenceInfo = SchedulingForLatency; 528 JumpBufSize = 0; 529 JumpBufAlignment = 0; 530 IfCvtBlockSizeLimit = 2; 531 IfCvtDupBlockSizeLimit = 0; 532 PrefLoopAlignment = 0; 533 534 InitLibcallNames(LibcallRoutineNames); 535 InitCmpLibcallCCs(CmpLibcallCCs); 536 InitLibcallCallingConvs(LibcallCallingConvs); 537} 538 539TargetLowering::~TargetLowering() { 540 delete &TLOF; 541} 542 543/// canOpTrap - Returns true if the operation can trap for the value type. 544/// VT must be a legal type. 545bool TargetLowering::canOpTrap(unsigned Op, EVT VT) const { 546 assert(isTypeLegal(VT)); 547 switch (Op) { 548 default: 549 return false; 550 case ISD::FDIV: 551 case ISD::FREM: 552 case ISD::SDIV: 553 case ISD::UDIV: 554 case ISD::SREM: 555 case ISD::UREM: 556 return true; 557 } 558} 559 560 561static unsigned getVectorTypeBreakdownMVT(MVT VT, MVT &IntermediateVT, 562 unsigned &NumIntermediates, 563 EVT &RegisterVT, 564 TargetLowering* TLI) { 565 // Figure out the right, legal destination reg to copy into. 566 unsigned NumElts = VT.getVectorNumElements(); 567 MVT EltTy = VT.getVectorElementType(); 568 569 unsigned NumVectorRegs = 1; 570 571 // FIXME: We don't support non-power-of-2-sized vectors for now. Ideally we 572 // could break down into LHS/RHS like LegalizeDAG does. 573 if (!isPowerOf2_32(NumElts)) { 574 NumVectorRegs = NumElts; 575 NumElts = 1; 576 } 577 578 // Divide the input until we get to a supported size. This will always 579 // end with a scalar if the target doesn't support vectors. 580 while (NumElts > 1 && !TLI->isTypeLegal(MVT::getVectorVT(EltTy, NumElts))) { 581 NumElts >>= 1; 582 NumVectorRegs <<= 1; 583 } 584 585 NumIntermediates = NumVectorRegs; 586 587 MVT NewVT = MVT::getVectorVT(EltTy, NumElts); 588 if (!TLI->isTypeLegal(NewVT)) 589 NewVT = EltTy; 590 IntermediateVT = NewVT; 591 592 EVT DestVT = TLI->getRegisterType(NewVT); 593 RegisterVT = DestVT; 594 if (EVT(DestVT).bitsLT(NewVT)) { 595 // Value is expanded, e.g. i64 -> i16. 596 return NumVectorRegs*(NewVT.getSizeInBits()/DestVT.getSizeInBits()); 597 } else { 598 // Otherwise, promotion or legal types use the same number of registers as 599 // the vector decimated to the appropriate level. 600 return NumVectorRegs; 601 } 602 603 return 1; 604} 605 606/// computeRegisterProperties - Once all of the register classes are added, 607/// this allows us to compute derived properties we expose. 608void TargetLowering::computeRegisterProperties() { 609 assert(MVT::LAST_VALUETYPE <= MVT::MAX_ALLOWED_VALUETYPE && 610 "Too many value types for ValueTypeActions to hold!"); 611 612 // Everything defaults to needing one register. 613 for (unsigned i = 0; i != MVT::LAST_VALUETYPE; ++i) { 614 NumRegistersForVT[i] = 1; 615 RegisterTypeForVT[i] = TransformToType[i] = (MVT::SimpleValueType)i; 616 } 617 // ...except isVoid, which doesn't need any registers. 618 NumRegistersForVT[MVT::isVoid] = 0; 619 620 // Find the largest integer register class. 621 unsigned LargestIntReg = MVT::LAST_INTEGER_VALUETYPE; 622 for (; RegClassForVT[LargestIntReg] == 0; --LargestIntReg) 623 assert(LargestIntReg != MVT::i1 && "No integer registers defined!"); 624 625 // Every integer value type larger than this largest register takes twice as 626 // many registers to represent as the previous ValueType. 627 for (unsigned ExpandedReg = LargestIntReg + 1; ; ++ExpandedReg) { 628 EVT ExpandedVT = (MVT::SimpleValueType)ExpandedReg; 629 if (!ExpandedVT.isInteger()) 630 break; 631 NumRegistersForVT[ExpandedReg] = 2*NumRegistersForVT[ExpandedReg-1]; 632 RegisterTypeForVT[ExpandedReg] = (MVT::SimpleValueType)LargestIntReg; 633 TransformToType[ExpandedReg] = (MVT::SimpleValueType)(ExpandedReg - 1); 634 ValueTypeActions.setTypeAction(ExpandedVT, Expand); 635 } 636 637 // Inspect all of the ValueType's smaller than the largest integer 638 // register to see which ones need promotion. 639 unsigned LegalIntReg = LargestIntReg; 640 for (unsigned IntReg = LargestIntReg - 1; 641 IntReg >= (unsigned)MVT::i1; --IntReg) { 642 EVT IVT = (MVT::SimpleValueType)IntReg; 643 if (isTypeLegal(IVT)) { 644 LegalIntReg = IntReg; 645 } else { 646 RegisterTypeForVT[IntReg] = TransformToType[IntReg] = 647 (MVT::SimpleValueType)LegalIntReg; 648 ValueTypeActions.setTypeAction(IVT, Promote); 649 } 650 } 651 652 // ppcf128 type is really two f64's. 653 if (!isTypeLegal(MVT::ppcf128)) { 654 NumRegistersForVT[MVT::ppcf128] = 2*NumRegistersForVT[MVT::f64]; 655 RegisterTypeForVT[MVT::ppcf128] = MVT::f64; 656 TransformToType[MVT::ppcf128] = MVT::f64; 657 ValueTypeActions.setTypeAction(MVT::ppcf128, Expand); 658 } 659 660 // Decide how to handle f64. If the target does not have native f64 support, 661 // expand it to i64 and we will be generating soft float library calls. 662 if (!isTypeLegal(MVT::f64)) { 663 NumRegistersForVT[MVT::f64] = NumRegistersForVT[MVT::i64]; 664 RegisterTypeForVT[MVT::f64] = RegisterTypeForVT[MVT::i64]; 665 TransformToType[MVT::f64] = MVT::i64; 666 ValueTypeActions.setTypeAction(MVT::f64, Expand); 667 } 668 669 // Decide how to handle f32. If the target does not have native support for 670 // f32, promote it to f64 if it is legal. Otherwise, expand it to i32. 671 if (!isTypeLegal(MVT::f32)) { 672 if (isTypeLegal(MVT::f64)) { 673 NumRegistersForVT[MVT::f32] = NumRegistersForVT[MVT::f64]; 674 RegisterTypeForVT[MVT::f32] = RegisterTypeForVT[MVT::f64]; 675 TransformToType[MVT::f32] = MVT::f64; 676 ValueTypeActions.setTypeAction(MVT::f32, Promote); 677 } else { 678 NumRegistersForVT[MVT::f32] = NumRegistersForVT[MVT::i32]; 679 RegisterTypeForVT[MVT::f32] = RegisterTypeForVT[MVT::i32]; 680 TransformToType[MVT::f32] = MVT::i32; 681 ValueTypeActions.setTypeAction(MVT::f32, Expand); 682 } 683 } 684 685 // Loop over all of the vector value types to see which need transformations. 686 for (unsigned i = MVT::FIRST_VECTOR_VALUETYPE; 687 i <= (unsigned)MVT::LAST_VECTOR_VALUETYPE; ++i) { 688 MVT VT = (MVT::SimpleValueType)i; 689 if (!isTypeLegal(VT)) { 690 MVT IntermediateVT; 691 EVT RegisterVT; 692 unsigned NumIntermediates; 693 NumRegistersForVT[i] = 694 getVectorTypeBreakdownMVT(VT, IntermediateVT, NumIntermediates, 695 RegisterVT, this); 696 RegisterTypeForVT[i] = RegisterVT; 697 698 // Determine if there is a legal wider type. 699 bool IsLegalWiderType = false; 700 EVT EltVT = VT.getVectorElementType(); 701 unsigned NElts = VT.getVectorNumElements(); 702 for (unsigned nVT = i+1; nVT <= MVT::LAST_VECTOR_VALUETYPE; ++nVT) { 703 EVT SVT = (MVT::SimpleValueType)nVT; 704 if (isTypeLegal(SVT) && SVT.getVectorElementType() == EltVT && 705 SVT.getVectorNumElements() > NElts && NElts != 1) { 706 TransformToType[i] = SVT; 707 ValueTypeActions.setTypeAction(VT, Promote); 708 IsLegalWiderType = true; 709 break; 710 } 711 } 712 if (!IsLegalWiderType) { 713 EVT NVT = VT.getPow2VectorType(); 714 if (NVT == VT) { 715 // Type is already a power of 2. The default action is to split. 716 TransformToType[i] = MVT::Other; 717 ValueTypeActions.setTypeAction(VT, Expand); 718 } else { 719 TransformToType[i] = NVT; 720 ValueTypeActions.setTypeAction(VT, Promote); 721 } 722 } 723 } 724 } 725} 726 727const char *TargetLowering::getTargetNodeName(unsigned Opcode) const { 728 return NULL; 729} 730 731 732MVT::SimpleValueType TargetLowering::getSetCCResultType(EVT VT) const { 733 return PointerTy.SimpleTy; 734} 735 736MVT::SimpleValueType TargetLowering::getCmpLibcallReturnType() const { 737 return MVT::i32; // return the default value 738} 739 740/// getVectorTypeBreakdown - Vector types are broken down into some number of 741/// legal first class types. For example, MVT::v8f32 maps to 2 MVT::v4f32 742/// with Altivec or SSE1, or 8 promoted MVT::f64 values with the X86 FP stack. 743/// Similarly, MVT::v2i64 turns into 4 MVT::i32 values with both PPC and X86. 744/// 745/// This method returns the number of registers needed, and the VT for each 746/// register. It also returns the VT and quantity of the intermediate values 747/// before they are promoted/expanded. 748/// 749unsigned TargetLowering::getVectorTypeBreakdown(LLVMContext &Context, EVT VT, 750 EVT &IntermediateVT, 751 unsigned &NumIntermediates, 752 EVT &RegisterVT) const { 753 // Figure out the right, legal destination reg to copy into. 754 unsigned NumElts = VT.getVectorNumElements(); 755 EVT EltTy = VT.getVectorElementType(); 756 757 unsigned NumVectorRegs = 1; 758 759 // FIXME: We don't support non-power-of-2-sized vectors for now. Ideally we 760 // could break down into LHS/RHS like LegalizeDAG does. 761 if (!isPowerOf2_32(NumElts)) { 762 NumVectorRegs = NumElts; 763 NumElts = 1; 764 } 765 766 // Divide the input until we get to a supported size. This will always 767 // end with a scalar if the target doesn't support vectors. 768 while (NumElts > 1 && !isTypeLegal( 769 EVT::getVectorVT(Context, EltTy, NumElts))) { 770 NumElts >>= 1; 771 NumVectorRegs <<= 1; 772 } 773 774 NumIntermediates = NumVectorRegs; 775 776 EVT NewVT = EVT::getVectorVT(Context, EltTy, NumElts); 777 if (!isTypeLegal(NewVT)) 778 NewVT = EltTy; 779 IntermediateVT = NewVT; 780 781 EVT DestVT = getRegisterType(Context, NewVT); 782 RegisterVT = DestVT; 783 if (DestVT.bitsLT(NewVT)) { 784 // Value is expanded, e.g. i64 -> i16. 785 return NumVectorRegs*(NewVT.getSizeInBits()/DestVT.getSizeInBits()); 786 } else { 787 // Otherwise, promotion or legal types use the same number of registers as 788 // the vector decimated to the appropriate level. 789 return NumVectorRegs; 790 } 791 792 return 1; 793} 794 795/// getWidenVectorType: given a vector type, returns the type to widen to 796/// (e.g., v7i8 to v8i8). If the vector type is legal, it returns itself. 797/// If there is no vector type that we want to widen to, returns MVT::Other 798/// When and where to widen is target dependent based on the cost of 799/// scalarizing vs using the wider vector type. 800EVT TargetLowering::getWidenVectorType(EVT VT) const { 801 assert(VT.isVector()); 802 if (isTypeLegal(VT)) 803 return VT; 804 805 // Default is not to widen until moved to LegalizeTypes 806 return MVT::Other; 807} 808 809/// getByValTypeAlignment - Return the desired alignment for ByVal aggregate 810/// function arguments in the caller parameter area. This is the actual 811/// alignment, not its logarithm. 812unsigned TargetLowering::getByValTypeAlignment(const Type *Ty) const { 813 return TD->getCallFrameTypeAlignment(Ty); 814} 815 816/// getJumpTableEncoding - Return the entry encoding for a jump table in the 817/// current function. The returned value is a member of the 818/// MachineJumpTableInfo::JTEntryKind enum. 819unsigned TargetLowering::getJumpTableEncoding() const { 820 // In non-pic modes, just use the address of a block. 821 if (getTargetMachine().getRelocationModel() != Reloc::PIC_) 822 return MachineJumpTableInfo::EK_BlockAddress; 823 824 // In PIC mode, if the target supports a GPRel32 directive, use it. 825 if (getTargetMachine().getMCAsmInfo()->getGPRel32Directive() != 0) 826 return MachineJumpTableInfo::EK_GPRel32BlockAddress; 827 828 // Otherwise, use a label difference. 829 return MachineJumpTableInfo::EK_LabelDifference32; 830} 831 832SDValue TargetLowering::getPICJumpTableRelocBase(SDValue Table, 833 SelectionDAG &DAG) const { 834 // If our PIC model is GP relative, use the global offset table as the base. 835 if (getJumpTableEncoding() == MachineJumpTableInfo::EK_GPRel32BlockAddress) 836 return DAG.getGLOBAL_OFFSET_TABLE(getPointerTy()); 837 return Table; 838} 839 840/// getPICJumpTableRelocBaseExpr - This returns the relocation base for the 841/// given PIC jumptable, the same as getPICJumpTableRelocBase, but as an 842/// MCExpr. 843const MCExpr * 844TargetLowering::getPICJumpTableRelocBaseExpr(const MachineFunction *MF, 845 unsigned JTI,MCContext &Ctx) const{ 846 // The normal PIC reloc base is the label at the start of the jump table. 847 return MCSymbolRefExpr::Create(MF->getJTISymbol(JTI, Ctx), Ctx); 848} 849 850bool 851TargetLowering::isOffsetFoldingLegal(const GlobalAddressSDNode *GA) const { 852 // Assume that everything is safe in static mode. 853 if (getTargetMachine().getRelocationModel() == Reloc::Static) 854 return true; 855 856 // In dynamic-no-pic mode, assume that known defined values are safe. 857 if (getTargetMachine().getRelocationModel() == Reloc::DynamicNoPIC && 858 GA && 859 !GA->getGlobal()->isDeclaration() && 860 !GA->getGlobal()->isWeakForLinker()) 861 return true; 862 863 // Otherwise assume nothing is safe. 864 return false; 865} 866 867//===----------------------------------------------------------------------===// 868// Optimization Methods 869//===----------------------------------------------------------------------===// 870 871/// ShrinkDemandedConstant - Check to see if the specified operand of the 872/// specified instruction is a constant integer. If so, check to see if there 873/// are any bits set in the constant that are not demanded. If so, shrink the 874/// constant and return true. 875bool TargetLowering::TargetLoweringOpt::ShrinkDemandedConstant(SDValue Op, 876 const APInt &Demanded) { 877 DebugLoc dl = Op.getDebugLoc(); 878 879 // FIXME: ISD::SELECT, ISD::SELECT_CC 880 switch (Op.getOpcode()) { 881 default: break; 882 case ISD::XOR: 883 case ISD::AND: 884 case ISD::OR: { 885 ConstantSDNode *C = dyn_cast<ConstantSDNode>(Op.getOperand(1)); 886 if (!C) return false; 887 888 if (Op.getOpcode() == ISD::XOR && 889 (C->getAPIntValue() | (~Demanded)).isAllOnesValue()) 890 return false; 891 892 // if we can expand it to have all bits set, do it 893 if (C->getAPIntValue().intersects(~Demanded)) { 894 EVT VT = Op.getValueType(); 895 SDValue New = DAG.getNode(Op.getOpcode(), dl, VT, Op.getOperand(0), 896 DAG.getConstant(Demanded & 897 C->getAPIntValue(), 898 VT)); 899 return CombineTo(Op, New); 900 } 901 902 break; 903 } 904 } 905 906 return false; 907} 908 909/// ShrinkDemandedOp - Convert x+y to (VT)((SmallVT)x+(SmallVT)y) if the 910/// casts are free. This uses isZExtFree and ZERO_EXTEND for the widening 911/// cast, but it could be generalized for targets with other types of 912/// implicit widening casts. 913bool 914TargetLowering::TargetLoweringOpt::ShrinkDemandedOp(SDValue Op, 915 unsigned BitWidth, 916 const APInt &Demanded, 917 DebugLoc dl) { 918 assert(Op.getNumOperands() == 2 && 919 "ShrinkDemandedOp only supports binary operators!"); 920 assert(Op.getNode()->getNumValues() == 1 && 921 "ShrinkDemandedOp only supports nodes with one result!"); 922 923 // Don't do this if the node has another user, which may require the 924 // full value. 925 if (!Op.getNode()->hasOneUse()) 926 return false; 927 928 // Search for the smallest integer type with free casts to and from 929 // Op's type. For expedience, just check power-of-2 integer types. 930 const TargetLowering &TLI = DAG.getTargetLoweringInfo(); 931 unsigned SmallVTBits = BitWidth - Demanded.countLeadingZeros(); 932 if (!isPowerOf2_32(SmallVTBits)) 933 SmallVTBits = NextPowerOf2(SmallVTBits); 934 for (; SmallVTBits < BitWidth; SmallVTBits = NextPowerOf2(SmallVTBits)) { 935 EVT SmallVT = EVT::getIntegerVT(*DAG.getContext(), SmallVTBits); 936 if (TLI.isTruncateFree(Op.getValueType(), SmallVT) && 937 TLI.isZExtFree(SmallVT, Op.getValueType())) { 938 // We found a type with free casts. 939 SDValue X = DAG.getNode(Op.getOpcode(), dl, SmallVT, 940 DAG.getNode(ISD::TRUNCATE, dl, SmallVT, 941 Op.getNode()->getOperand(0)), 942 DAG.getNode(ISD::TRUNCATE, dl, SmallVT, 943 Op.getNode()->getOperand(1))); 944 SDValue Z = DAG.getNode(ISD::ZERO_EXTEND, dl, Op.getValueType(), X); 945 return CombineTo(Op, Z); 946 } 947 } 948 return false; 949} 950 951/// SimplifyDemandedBits - Look at Op. At this point, we know that only the 952/// DemandedMask bits of the result of Op are ever used downstream. If we can 953/// use this information to simplify Op, create a new simplified DAG node and 954/// return true, returning the original and new nodes in Old and New. Otherwise, 955/// analyze the expression and return a mask of KnownOne and KnownZero bits for 956/// the expression (used to simplify the caller). The KnownZero/One bits may 957/// only be accurate for those bits in the DemandedMask. 958bool TargetLowering::SimplifyDemandedBits(SDValue Op, 959 const APInt &DemandedMask, 960 APInt &KnownZero, 961 APInt &KnownOne, 962 TargetLoweringOpt &TLO, 963 unsigned Depth) const { 964 unsigned BitWidth = DemandedMask.getBitWidth(); 965 assert(Op.getValueType().getScalarType().getSizeInBits() == BitWidth && 966 "Mask size mismatches value type size!"); 967 APInt NewMask = DemandedMask; 968 DebugLoc dl = Op.getDebugLoc(); 969 970 // Don't know anything. 971 KnownZero = KnownOne = APInt(BitWidth, 0); 972 973 // Other users may use these bits. 974 if (!Op.getNode()->hasOneUse()) { 975 if (Depth != 0) { 976 // If not at the root, Just compute the KnownZero/KnownOne bits to 977 // simplify things downstream. 978 TLO.DAG.ComputeMaskedBits(Op, DemandedMask, KnownZero, KnownOne, Depth); 979 return false; 980 } 981 // If this is the root being simplified, allow it to have multiple uses, 982 // just set the NewMask to all bits. 983 NewMask = APInt::getAllOnesValue(BitWidth); 984 } else if (DemandedMask == 0) { 985 // Not demanding any bits from Op. 986 if (Op.getOpcode() != ISD::UNDEF) 987 return TLO.CombineTo(Op, TLO.DAG.getUNDEF(Op.getValueType())); 988 return false; 989 } else if (Depth == 6) { // Limit search depth. 990 return false; 991 } 992 993 APInt KnownZero2, KnownOne2, KnownZeroOut, KnownOneOut; 994 switch (Op.getOpcode()) { 995 case ISD::Constant: 996 // We know all of the bits for a constant! 997 KnownOne = cast<ConstantSDNode>(Op)->getAPIntValue() & NewMask; 998 KnownZero = ~KnownOne & NewMask; 999 return false; // Don't fall through, will infinitely loop. 1000 case ISD::AND: 1001 // If the RHS is a constant, check to see if the LHS would be zero without 1002 // using the bits from the RHS. Below, we use knowledge about the RHS to 1003 // simplify the LHS, here we're using information from the LHS to simplify 1004 // the RHS. 1005 if (ConstantSDNode *RHSC = dyn_cast<ConstantSDNode>(Op.getOperand(1))) { 1006 APInt LHSZero, LHSOne; 1007 TLO.DAG.ComputeMaskedBits(Op.getOperand(0), NewMask, 1008 LHSZero, LHSOne, Depth+1); 1009 // If the LHS already has zeros where RHSC does, this and is dead. 1010 if ((LHSZero & NewMask) == (~RHSC->getAPIntValue() & NewMask)) 1011 return TLO.CombineTo(Op, Op.getOperand(0)); 1012 // If any of the set bits in the RHS are known zero on the LHS, shrink 1013 // the constant. 1014 if (TLO.ShrinkDemandedConstant(Op, ~LHSZero & NewMask)) 1015 return true; 1016 } 1017 1018 if (SimplifyDemandedBits(Op.getOperand(1), NewMask, KnownZero, 1019 KnownOne, TLO, Depth+1)) 1020 return true; 1021 assert((KnownZero & KnownOne) == 0 && "Bits known to be one AND zero?"); 1022 if (SimplifyDemandedBits(Op.getOperand(0), ~KnownZero & NewMask, 1023 KnownZero2, KnownOne2, TLO, Depth+1)) 1024 return true; 1025 assert((KnownZero2 & KnownOne2) == 0 && "Bits known to be one AND zero?"); 1026 1027 // If all of the demanded bits are known one on one side, return the other. 1028 // These bits cannot contribute to the result of the 'and'. 1029 if ((NewMask & ~KnownZero2 & KnownOne) == (~KnownZero2 & NewMask)) 1030 return TLO.CombineTo(Op, Op.getOperand(0)); 1031 if ((NewMask & ~KnownZero & KnownOne2) == (~KnownZero & NewMask)) 1032 return TLO.CombineTo(Op, Op.getOperand(1)); 1033 // If all of the demanded bits in the inputs are known zeros, return zero. 1034 if ((NewMask & (KnownZero|KnownZero2)) == NewMask) 1035 return TLO.CombineTo(Op, TLO.DAG.getConstant(0, Op.getValueType())); 1036 // If the RHS is a constant, see if we can simplify it. 1037 if (TLO.ShrinkDemandedConstant(Op, ~KnownZero2 & NewMask)) 1038 return true; 1039 // If the operation can be done in a smaller type, do so. 1040 if (TLO.ShrinkOps && TLO.ShrinkDemandedOp(Op, BitWidth, NewMask, dl)) 1041 return true; 1042 1043 // Output known-1 bits are only known if set in both the LHS & RHS. 1044 KnownOne &= KnownOne2; 1045 // Output known-0 are known to be clear if zero in either the LHS | RHS. 1046 KnownZero |= KnownZero2; 1047 break; 1048 case ISD::OR: 1049 if (SimplifyDemandedBits(Op.getOperand(1), NewMask, KnownZero, 1050 KnownOne, TLO, Depth+1)) 1051 return true; 1052 assert((KnownZero & KnownOne) == 0 && "Bits known to be one AND zero?"); 1053 if (SimplifyDemandedBits(Op.getOperand(0), ~KnownOne & NewMask, 1054 KnownZero2, KnownOne2, TLO, Depth+1)) 1055 return true; 1056 assert((KnownZero2 & KnownOne2) == 0 && "Bits known to be one AND zero?"); 1057 1058 // If all of the demanded bits are known zero on one side, return the other. 1059 // These bits cannot contribute to the result of the 'or'. 1060 if ((NewMask & ~KnownOne2 & KnownZero) == (~KnownOne2 & NewMask)) 1061 return TLO.CombineTo(Op, Op.getOperand(0)); 1062 if ((NewMask & ~KnownOne & KnownZero2) == (~KnownOne & NewMask)) 1063 return TLO.CombineTo(Op, Op.getOperand(1)); 1064 // If all of the potentially set bits on one side are known to be set on 1065 // the other side, just use the 'other' side. 1066 if ((NewMask & ~KnownZero & KnownOne2) == (~KnownZero & NewMask)) 1067 return TLO.CombineTo(Op, Op.getOperand(0)); 1068 if ((NewMask & ~KnownZero2 & KnownOne) == (~KnownZero2 & NewMask)) 1069 return TLO.CombineTo(Op, Op.getOperand(1)); 1070 // If the RHS is a constant, see if we can simplify it. 1071 if (TLO.ShrinkDemandedConstant(Op, NewMask)) 1072 return true; 1073 // If the operation can be done in a smaller type, do so. 1074 if (TLO.ShrinkOps && TLO.ShrinkDemandedOp(Op, BitWidth, NewMask, dl)) 1075 return true; 1076 1077 // Output known-0 bits are only known if clear in both the LHS & RHS. 1078 KnownZero &= KnownZero2; 1079 // Output known-1 are known to be set if set in either the LHS | RHS. 1080 KnownOne |= KnownOne2; 1081 break; 1082 case ISD::XOR: 1083 if (SimplifyDemandedBits(Op.getOperand(1), NewMask, KnownZero, 1084 KnownOne, TLO, Depth+1)) 1085 return true; 1086 assert((KnownZero & KnownOne) == 0 && "Bits known to be one AND zero?"); 1087 if (SimplifyDemandedBits(Op.getOperand(0), NewMask, KnownZero2, 1088 KnownOne2, TLO, Depth+1)) 1089 return true; 1090 assert((KnownZero2 & KnownOne2) == 0 && "Bits known to be one AND zero?"); 1091 1092 // If all of the demanded bits are known zero on one side, return the other. 1093 // These bits cannot contribute to the result of the 'xor'. 1094 if ((KnownZero & NewMask) == NewMask) 1095 return TLO.CombineTo(Op, Op.getOperand(0)); 1096 if ((KnownZero2 & NewMask) == NewMask) 1097 return TLO.CombineTo(Op, Op.getOperand(1)); 1098 // If the operation can be done in a smaller type, do so. 1099 if (TLO.ShrinkOps && TLO.ShrinkDemandedOp(Op, BitWidth, NewMask, dl)) 1100 return true; 1101 1102 // If all of the unknown bits are known to be zero on one side or the other 1103 // (but not both) turn this into an *inclusive* or. 1104 // e.g. (A & C1)^(B & C2) -> (A & C1)|(B & C2) iff C1&C2 == 0 1105 if ((NewMask & ~KnownZero & ~KnownZero2) == 0) 1106 return TLO.CombineTo(Op, TLO.DAG.getNode(ISD::OR, dl, Op.getValueType(), 1107 Op.getOperand(0), 1108 Op.getOperand(1))); 1109 1110 // Output known-0 bits are known if clear or set in both the LHS & RHS. 1111 KnownZeroOut = (KnownZero & KnownZero2) | (KnownOne & KnownOne2); 1112 // Output known-1 are known to be set if set in only one of the LHS, RHS. 1113 KnownOneOut = (KnownZero & KnownOne2) | (KnownOne & KnownZero2); 1114 1115 // If all of the demanded bits on one side are known, and all of the set 1116 // bits on that side are also known to be set on the other side, turn this 1117 // into an AND, as we know the bits will be cleared. 1118 // e.g. (X | C1) ^ C2 --> (X | C1) & ~C2 iff (C1&C2) == C2 1119 if ((NewMask & (KnownZero|KnownOne)) == NewMask) { // all known 1120 if ((KnownOne & KnownOne2) == KnownOne) { 1121 EVT VT = Op.getValueType(); 1122 SDValue ANDC = TLO.DAG.getConstant(~KnownOne & NewMask, VT); 1123 return TLO.CombineTo(Op, TLO.DAG.getNode(ISD::AND, dl, VT, 1124 Op.getOperand(0), ANDC)); 1125 } 1126 } 1127 1128 // If the RHS is a constant, see if we can simplify it. 1129 // for XOR, we prefer to force bits to 1 if they will make a -1. 1130 // if we can't force bits, try to shrink constant 1131 if (ConstantSDNode *C = dyn_cast<ConstantSDNode>(Op.getOperand(1))) { 1132 APInt Expanded = C->getAPIntValue() | (~NewMask); 1133 // if we can expand it to have all bits set, do it 1134 if (Expanded.isAllOnesValue()) { 1135 if (Expanded != C->getAPIntValue()) { 1136 EVT VT = Op.getValueType(); 1137 SDValue New = TLO.DAG.getNode(Op.getOpcode(), dl,VT, Op.getOperand(0), 1138 TLO.DAG.getConstant(Expanded, VT)); 1139 return TLO.CombineTo(Op, New); 1140 } 1141 // if it already has all the bits set, nothing to change 1142 // but don't shrink either! 1143 } else if (TLO.ShrinkDemandedConstant(Op, NewMask)) { 1144 return true; 1145 } 1146 } 1147 1148 KnownZero = KnownZeroOut; 1149 KnownOne = KnownOneOut; 1150 break; 1151 case ISD::SELECT: 1152 if (SimplifyDemandedBits(Op.getOperand(2), NewMask, KnownZero, 1153 KnownOne, TLO, Depth+1)) 1154 return true; 1155 if (SimplifyDemandedBits(Op.getOperand(1), NewMask, KnownZero2, 1156 KnownOne2, TLO, Depth+1)) 1157 return true; 1158 assert((KnownZero & KnownOne) == 0 && "Bits known to be one AND zero?"); 1159 assert((KnownZero2 & KnownOne2) == 0 && "Bits known to be one AND zero?"); 1160 1161 // If the operands are constants, see if we can simplify them. 1162 if (TLO.ShrinkDemandedConstant(Op, NewMask)) 1163 return true; 1164 1165 // Only known if known in both the LHS and RHS. 1166 KnownOne &= KnownOne2; 1167 KnownZero &= KnownZero2; 1168 break; 1169 case ISD::SELECT_CC: 1170 if (SimplifyDemandedBits(Op.getOperand(3), NewMask, KnownZero, 1171 KnownOne, TLO, Depth+1)) 1172 return true; 1173 if (SimplifyDemandedBits(Op.getOperand(2), NewMask, KnownZero2, 1174 KnownOne2, TLO, Depth+1)) 1175 return true; 1176 assert((KnownZero & KnownOne) == 0 && "Bits known to be one AND zero?"); 1177 assert((KnownZero2 & KnownOne2) == 0 && "Bits known to be one AND zero?"); 1178 1179 // If the operands are constants, see if we can simplify them. 1180 if (TLO.ShrinkDemandedConstant(Op, NewMask)) 1181 return true; 1182 1183 // Only known if known in both the LHS and RHS. 1184 KnownOne &= KnownOne2; 1185 KnownZero &= KnownZero2; 1186 break; 1187 case ISD::SHL: 1188 if (ConstantSDNode *SA = dyn_cast<ConstantSDNode>(Op.getOperand(1))) { 1189 unsigned ShAmt = SA->getZExtValue(); 1190 SDValue InOp = Op.getOperand(0); 1191 1192 // If the shift count is an invalid immediate, don't do anything. 1193 if (ShAmt >= BitWidth) 1194 break; 1195 1196 // If this is ((X >>u C1) << ShAmt), see if we can simplify this into a 1197 // single shift. We can do this if the bottom bits (which are shifted 1198 // out) are never demanded. 1199 if (InOp.getOpcode() == ISD::SRL && 1200 isa<ConstantSDNode>(InOp.getOperand(1))) { 1201 if (ShAmt && (NewMask & APInt::getLowBitsSet(BitWidth, ShAmt)) == 0) { 1202 unsigned C1= cast<ConstantSDNode>(InOp.getOperand(1))->getZExtValue(); 1203 unsigned Opc = ISD::SHL; 1204 int Diff = ShAmt-C1; 1205 if (Diff < 0) { 1206 Diff = -Diff; 1207 Opc = ISD::SRL; 1208 } 1209 1210 SDValue NewSA = 1211 TLO.DAG.getConstant(Diff, Op.getOperand(1).getValueType()); 1212 EVT VT = Op.getValueType(); 1213 return TLO.CombineTo(Op, TLO.DAG.getNode(Opc, dl, VT, 1214 InOp.getOperand(0), NewSA)); 1215 } 1216 } 1217 1218 if (SimplifyDemandedBits(Op.getOperand(0), NewMask.lshr(ShAmt), 1219 KnownZero, KnownOne, TLO, Depth+1)) 1220 return true; 1221 KnownZero <<= SA->getZExtValue(); 1222 KnownOne <<= SA->getZExtValue(); 1223 // low bits known zero. 1224 KnownZero |= APInt::getLowBitsSet(BitWidth, SA->getZExtValue()); 1225 } 1226 break; 1227 case ISD::SRL: 1228 if (ConstantSDNode *SA = dyn_cast<ConstantSDNode>(Op.getOperand(1))) { 1229 EVT VT = Op.getValueType(); 1230 unsigned ShAmt = SA->getZExtValue(); 1231 unsigned VTSize = VT.getSizeInBits(); 1232 SDValue InOp = Op.getOperand(0); 1233 1234 // If the shift count is an invalid immediate, don't do anything. 1235 if (ShAmt >= BitWidth) 1236 break; 1237 1238 // If this is ((X << C1) >>u ShAmt), see if we can simplify this into a 1239 // single shift. We can do this if the top bits (which are shifted out) 1240 // are never demanded. 1241 if (InOp.getOpcode() == ISD::SHL && 1242 isa<ConstantSDNode>(InOp.getOperand(1))) { 1243 if (ShAmt && (NewMask & APInt::getHighBitsSet(VTSize, ShAmt)) == 0) { 1244 unsigned C1= cast<ConstantSDNode>(InOp.getOperand(1))->getZExtValue(); 1245 unsigned Opc = ISD::SRL; 1246 int Diff = ShAmt-C1; 1247 if (Diff < 0) { 1248 Diff = -Diff; 1249 Opc = ISD::SHL; 1250 } 1251 1252 SDValue NewSA = 1253 TLO.DAG.getConstant(Diff, Op.getOperand(1).getValueType()); 1254 return TLO.CombineTo(Op, TLO.DAG.getNode(Opc, dl, VT, 1255 InOp.getOperand(0), NewSA)); 1256 } 1257 } 1258 1259 // Compute the new bits that are at the top now. 1260 if (SimplifyDemandedBits(InOp, (NewMask << ShAmt), 1261 KnownZero, KnownOne, TLO, Depth+1)) 1262 return true; 1263 assert((KnownZero & KnownOne) == 0 && "Bits known to be one AND zero?"); 1264 KnownZero = KnownZero.lshr(ShAmt); 1265 KnownOne = KnownOne.lshr(ShAmt); 1266 1267 APInt HighBits = APInt::getHighBitsSet(BitWidth, ShAmt); 1268 KnownZero |= HighBits; // High bits known zero. 1269 } 1270 break; 1271 case ISD::SRA: 1272 // If this is an arithmetic shift right and only the low-bit is set, we can 1273 // always convert this into a logical shr, even if the shift amount is 1274 // variable. The low bit of the shift cannot be an input sign bit unless 1275 // the shift amount is >= the size of the datatype, which is undefined. 1276 if (DemandedMask == 1) 1277 return TLO.CombineTo(Op, TLO.DAG.getNode(ISD::SRL, dl, Op.getValueType(), 1278 Op.getOperand(0), Op.getOperand(1))); 1279 1280 if (ConstantSDNode *SA = dyn_cast<ConstantSDNode>(Op.getOperand(1))) { 1281 EVT VT = Op.getValueType(); 1282 unsigned ShAmt = SA->getZExtValue(); 1283 1284 // If the shift count is an invalid immediate, don't do anything. 1285 if (ShAmt >= BitWidth) 1286 break; 1287 1288 APInt InDemandedMask = (NewMask << ShAmt); 1289 1290 // If any of the demanded bits are produced by the sign extension, we also 1291 // demand the input sign bit. 1292 APInt HighBits = APInt::getHighBitsSet(BitWidth, ShAmt); 1293 if (HighBits.intersects(NewMask)) 1294 InDemandedMask |= APInt::getSignBit(VT.getScalarType().getSizeInBits()); 1295 1296 if (SimplifyDemandedBits(Op.getOperand(0), InDemandedMask, 1297 KnownZero, KnownOne, TLO, Depth+1)) 1298 return true; 1299 assert((KnownZero & KnownOne) == 0 && "Bits known to be one AND zero?"); 1300 KnownZero = KnownZero.lshr(ShAmt); 1301 KnownOne = KnownOne.lshr(ShAmt); 1302 1303 // Handle the sign bit, adjusted to where it is now in the mask. 1304 APInt SignBit = APInt::getSignBit(BitWidth).lshr(ShAmt); 1305 1306 // If the input sign bit is known to be zero, or if none of the top bits 1307 // are demanded, turn this into an unsigned shift right. 1308 if (KnownZero.intersects(SignBit) || (HighBits & ~NewMask) == HighBits) { 1309 return TLO.CombineTo(Op, TLO.DAG.getNode(ISD::SRL, dl, VT, 1310 Op.getOperand(0), 1311 Op.getOperand(1))); 1312 } else if (KnownOne.intersects(SignBit)) { // New bits are known one. 1313 KnownOne |= HighBits; 1314 } 1315 } 1316 break; 1317 case ISD::SIGN_EXTEND_INREG: { 1318 EVT EVT = cast<VTSDNode>(Op.getOperand(1))->getVT(); 1319 1320 // Sign extension. Compute the demanded bits in the result that are not 1321 // present in the input. 1322 APInt NewBits = 1323 APInt::getHighBitsSet(BitWidth, 1324 BitWidth - EVT.getScalarType().getSizeInBits()) & 1325 NewMask; 1326 1327 // If none of the extended bits are demanded, eliminate the sextinreg. 1328 if (NewBits == 0) 1329 return TLO.CombineTo(Op, Op.getOperand(0)); 1330 1331 APInt InSignBit = APInt::getSignBit(EVT.getScalarType().getSizeInBits()); 1332 InSignBit.zext(BitWidth); 1333 APInt InputDemandedBits = 1334 APInt::getLowBitsSet(BitWidth, 1335 EVT.getScalarType().getSizeInBits()) & 1336 NewMask; 1337 1338 // Since the sign extended bits are demanded, we know that the sign 1339 // bit is demanded. 1340 InputDemandedBits |= InSignBit; 1341 1342 if (SimplifyDemandedBits(Op.getOperand(0), InputDemandedBits, 1343 KnownZero, KnownOne, TLO, Depth+1)) 1344 return true; 1345 assert((KnownZero & KnownOne) == 0 && "Bits known to be one AND zero?"); 1346 1347 // If the sign bit of the input is known set or clear, then we know the 1348 // top bits of the result. 1349 1350 // If the input sign bit is known zero, convert this into a zero extension. 1351 if (KnownZero.intersects(InSignBit)) 1352 return TLO.CombineTo(Op, 1353 TLO.DAG.getZeroExtendInReg(Op.getOperand(0),dl,EVT)); 1354 1355 if (KnownOne.intersects(InSignBit)) { // Input sign bit known set 1356 KnownOne |= NewBits; 1357 KnownZero &= ~NewBits; 1358 } else { // Input sign bit unknown 1359 KnownZero &= ~NewBits; 1360 KnownOne &= ~NewBits; 1361 } 1362 break; 1363 } 1364 case ISD::ZERO_EXTEND: { 1365 unsigned OperandBitWidth = 1366 Op.getOperand(0).getValueType().getScalarType().getSizeInBits(); 1367 APInt InMask = NewMask; 1368 InMask.trunc(OperandBitWidth); 1369 1370 // If none of the top bits are demanded, convert this into an any_extend. 1371 APInt NewBits = 1372 APInt::getHighBitsSet(BitWidth, BitWidth - OperandBitWidth) & NewMask; 1373 if (!NewBits.intersects(NewMask)) 1374 return TLO.CombineTo(Op, TLO.DAG.getNode(ISD::ANY_EXTEND, dl, 1375 Op.getValueType(), 1376 Op.getOperand(0))); 1377 1378 if (SimplifyDemandedBits(Op.getOperand(0), InMask, 1379 KnownZero, KnownOne, TLO, Depth+1)) 1380 return true; 1381 assert((KnownZero & KnownOne) == 0 && "Bits known to be one AND zero?"); 1382 KnownZero.zext(BitWidth); 1383 KnownOne.zext(BitWidth); 1384 KnownZero |= NewBits; 1385 break; 1386 } 1387 case ISD::SIGN_EXTEND: { 1388 EVT InVT = Op.getOperand(0).getValueType(); 1389 unsigned InBits = InVT.getScalarType().getSizeInBits(); 1390 APInt InMask = APInt::getLowBitsSet(BitWidth, InBits); 1391 APInt InSignBit = APInt::getBitsSet(BitWidth, InBits - 1, InBits); 1392 APInt NewBits = ~InMask & NewMask; 1393 1394 // If none of the top bits are demanded, convert this into an any_extend. 1395 if (NewBits == 0) 1396 return TLO.CombineTo(Op,TLO.DAG.getNode(ISD::ANY_EXTEND, dl, 1397 Op.getValueType(), 1398 Op.getOperand(0))); 1399 1400 // Since some of the sign extended bits are demanded, we know that the sign 1401 // bit is demanded. 1402 APInt InDemandedBits = InMask & NewMask; 1403 InDemandedBits |= InSignBit; 1404 InDemandedBits.trunc(InBits); 1405 1406 if (SimplifyDemandedBits(Op.getOperand(0), InDemandedBits, KnownZero, 1407 KnownOne, TLO, Depth+1)) 1408 return true; 1409 KnownZero.zext(BitWidth); 1410 KnownOne.zext(BitWidth); 1411 1412 // If the sign bit is known zero, convert this to a zero extend. 1413 if (KnownZero.intersects(InSignBit)) 1414 return TLO.CombineTo(Op, TLO.DAG.getNode(ISD::ZERO_EXTEND, dl, 1415 Op.getValueType(), 1416 Op.getOperand(0))); 1417 1418 // If the sign bit is known one, the top bits match. 1419 if (KnownOne.intersects(InSignBit)) { 1420 KnownOne |= NewBits; 1421 KnownZero &= ~NewBits; 1422 } else { // Otherwise, top bits aren't known. 1423 KnownOne &= ~NewBits; 1424 KnownZero &= ~NewBits; 1425 } 1426 break; 1427 } 1428 case ISD::ANY_EXTEND: { 1429 unsigned OperandBitWidth = 1430 Op.getOperand(0).getValueType().getScalarType().getSizeInBits(); 1431 APInt InMask = NewMask; 1432 InMask.trunc(OperandBitWidth); 1433 if (SimplifyDemandedBits(Op.getOperand(0), InMask, 1434 KnownZero, KnownOne, TLO, Depth+1)) 1435 return true; 1436 assert((KnownZero & KnownOne) == 0 && "Bits known to be one AND zero?"); 1437 KnownZero.zext(BitWidth); 1438 KnownOne.zext(BitWidth); 1439 break; 1440 } 1441 case ISD::TRUNCATE: { 1442 // Simplify the input, using demanded bit information, and compute the known 1443 // zero/one bits live out. 1444 APInt TruncMask = NewMask; 1445 TruncMask.zext(Op.getOperand(0).getValueSizeInBits()); 1446 if (SimplifyDemandedBits(Op.getOperand(0), TruncMask, 1447 KnownZero, KnownOne, TLO, Depth+1)) 1448 return true; 1449 KnownZero.trunc(BitWidth); 1450 KnownOne.trunc(BitWidth); 1451 1452 // If the input is only used by this truncate, see if we can shrink it based 1453 // on the known demanded bits. 1454 if (Op.getOperand(0).getNode()->hasOneUse()) { 1455 SDValue In = Op.getOperand(0); 1456 unsigned InBitWidth = In.getValueSizeInBits(); 1457 switch (In.getOpcode()) { 1458 default: break; 1459 case ISD::SRL: 1460 // Shrink SRL by a constant if none of the high bits shifted in are 1461 // demanded. 1462 if (ConstantSDNode *ShAmt = dyn_cast<ConstantSDNode>(In.getOperand(1))){ 1463 APInt HighBits = APInt::getHighBitsSet(InBitWidth, 1464 InBitWidth - BitWidth); 1465 HighBits = HighBits.lshr(ShAmt->getZExtValue()); 1466 HighBits.trunc(BitWidth); 1467 1468 if (ShAmt->getZExtValue() < BitWidth && !(HighBits & NewMask)) { 1469 // None of the shifted in bits are needed. Add a truncate of the 1470 // shift input, then shift it. 1471 SDValue NewTrunc = TLO.DAG.getNode(ISD::TRUNCATE, dl, 1472 Op.getValueType(), 1473 In.getOperand(0)); 1474 return TLO.CombineTo(Op, TLO.DAG.getNode(ISD::SRL, dl, 1475 Op.getValueType(), 1476 NewTrunc, 1477 In.getOperand(1))); 1478 } 1479 } 1480 break; 1481 } 1482 } 1483 1484 assert((KnownZero & KnownOne) == 0 && "Bits known to be one AND zero?"); 1485 break; 1486 } 1487 case ISD::AssertZext: { 1488 EVT VT = cast<VTSDNode>(Op.getOperand(1))->getVT(); 1489 APInt InMask = APInt::getLowBitsSet(BitWidth, 1490 VT.getSizeInBits()); 1491 if (SimplifyDemandedBits(Op.getOperand(0), InMask & NewMask, 1492 KnownZero, KnownOne, TLO, Depth+1)) 1493 return true; 1494 assert((KnownZero & KnownOne) == 0 && "Bits known to be one AND zero?"); 1495 KnownZero |= ~InMask & NewMask; 1496 break; 1497 } 1498 case ISD::BIT_CONVERT: 1499#if 0 1500 // If this is an FP->Int bitcast and if the sign bit is the only thing that 1501 // is demanded, turn this into a FGETSIGN. 1502 if (NewMask == EVT::getIntegerVTSignBit(Op.getValueType()) && 1503 MVT::isFloatingPoint(Op.getOperand(0).getValueType()) && 1504 !MVT::isVector(Op.getOperand(0).getValueType())) { 1505 // Only do this xform if FGETSIGN is valid or if before legalize. 1506 if (!TLO.AfterLegalize || 1507 isOperationLegal(ISD::FGETSIGN, Op.getValueType())) { 1508 // Make a FGETSIGN + SHL to move the sign bit into the appropriate 1509 // place. We expect the SHL to be eliminated by other optimizations. 1510 SDValue Sign = TLO.DAG.getNode(ISD::FGETSIGN, Op.getValueType(), 1511 Op.getOperand(0)); 1512 unsigned ShVal = Op.getValueType().getSizeInBits()-1; 1513 SDValue ShAmt = TLO.DAG.getConstant(ShVal, getShiftAmountTy()); 1514 return TLO.CombineTo(Op, TLO.DAG.getNode(ISD::SHL, Op.getValueType(), 1515 Sign, ShAmt)); 1516 } 1517 } 1518#endif 1519 break; 1520 case ISD::ADD: 1521 case ISD::MUL: 1522 case ISD::SUB: { 1523 // Add, Sub, and Mul don't demand any bits in positions beyond that 1524 // of the highest bit demanded of them. 1525 APInt LoMask = APInt::getLowBitsSet(BitWidth, 1526 BitWidth - NewMask.countLeadingZeros()); 1527 if (SimplifyDemandedBits(Op.getOperand(0), LoMask, KnownZero2, 1528 KnownOne2, TLO, Depth+1)) 1529 return true; 1530 if (SimplifyDemandedBits(Op.getOperand(1), LoMask, KnownZero2, 1531 KnownOne2, TLO, Depth+1)) 1532 return true; 1533 // See if the operation should be performed at a smaller bit width. 1534 if (TLO.ShrinkOps && TLO.ShrinkDemandedOp(Op, BitWidth, NewMask, dl)) 1535 return true; 1536 } 1537 // FALL THROUGH 1538 default: 1539 // Just use ComputeMaskedBits to compute output bits. 1540 TLO.DAG.ComputeMaskedBits(Op, NewMask, KnownZero, KnownOne, Depth); 1541 break; 1542 } 1543 1544 // If we know the value of all of the demanded bits, return this as a 1545 // constant. 1546 if ((NewMask & (KnownZero|KnownOne)) == NewMask) 1547 return TLO.CombineTo(Op, TLO.DAG.getConstant(KnownOne, Op.getValueType())); 1548 1549 return false; 1550} 1551 1552/// computeMaskedBitsForTargetNode - Determine which of the bits specified 1553/// in Mask are known to be either zero or one and return them in the 1554/// KnownZero/KnownOne bitsets. 1555void TargetLowering::computeMaskedBitsForTargetNode(const SDValue Op, 1556 const APInt &Mask, 1557 APInt &KnownZero, 1558 APInt &KnownOne, 1559 const SelectionDAG &DAG, 1560 unsigned Depth) const { 1561 assert((Op.getOpcode() >= ISD::BUILTIN_OP_END || 1562 Op.getOpcode() == ISD::INTRINSIC_WO_CHAIN || 1563 Op.getOpcode() == ISD::INTRINSIC_W_CHAIN || 1564 Op.getOpcode() == ISD::INTRINSIC_VOID) && 1565 "Should use MaskedValueIsZero if you don't know whether Op" 1566 " is a target node!"); 1567 KnownZero = KnownOne = APInt(Mask.getBitWidth(), 0); 1568} 1569 1570/// ComputeNumSignBitsForTargetNode - This method can be implemented by 1571/// targets that want to expose additional information about sign bits to the 1572/// DAG Combiner. 1573unsigned TargetLowering::ComputeNumSignBitsForTargetNode(SDValue Op, 1574 unsigned Depth) const { 1575 assert((Op.getOpcode() >= ISD::BUILTIN_OP_END || 1576 Op.getOpcode() == ISD::INTRINSIC_WO_CHAIN || 1577 Op.getOpcode() == ISD::INTRINSIC_W_CHAIN || 1578 Op.getOpcode() == ISD::INTRINSIC_VOID) && 1579 "Should use ComputeNumSignBits if you don't know whether Op" 1580 " is a target node!"); 1581 return 1; 1582} 1583 1584/// ValueHasExactlyOneBitSet - Test if the given value is known to have exactly 1585/// one bit set. This differs from ComputeMaskedBits in that it doesn't need to 1586/// determine which bit is set. 1587/// 1588static bool ValueHasExactlyOneBitSet(SDValue Val, const SelectionDAG &DAG) { 1589 // A left-shift of a constant one will have exactly one bit set, because 1590 // shifting the bit off the end is undefined. 1591 if (Val.getOpcode() == ISD::SHL) 1592 if (ConstantSDNode *C = 1593 dyn_cast<ConstantSDNode>(Val.getNode()->getOperand(0))) 1594 if (C->getAPIntValue() == 1) 1595 return true; 1596 1597 // Similarly, a right-shift of a constant sign-bit will have exactly 1598 // one bit set. 1599 if (Val.getOpcode() == ISD::SRL) 1600 if (ConstantSDNode *C = 1601 dyn_cast<ConstantSDNode>(Val.getNode()->getOperand(0))) 1602 if (C->getAPIntValue().isSignBit()) 1603 return true; 1604 1605 // More could be done here, though the above checks are enough 1606 // to handle some common cases. 1607 1608 // Fall back to ComputeMaskedBits to catch other known cases. 1609 EVT OpVT = Val.getValueType(); 1610 unsigned BitWidth = OpVT.getSizeInBits(); 1611 APInt Mask = APInt::getAllOnesValue(BitWidth); 1612 APInt KnownZero, KnownOne; 1613 DAG.ComputeMaskedBits(Val, Mask, KnownZero, KnownOne); 1614 return (KnownZero.countPopulation() == BitWidth - 1) && 1615 (KnownOne.countPopulation() == 1); 1616} 1617 1618/// SimplifySetCC - Try to simplify a setcc built with the specified operands 1619/// and cc. If it is unable to simplify it, return a null SDValue. 1620SDValue 1621TargetLowering::SimplifySetCC(EVT VT, SDValue N0, SDValue N1, 1622 ISD::CondCode Cond, bool foldBooleans, 1623 DAGCombinerInfo &DCI, DebugLoc dl) const { 1624 SelectionDAG &DAG = DCI.DAG; 1625 LLVMContext &Context = *DAG.getContext(); 1626 1627 // These setcc operations always fold. 1628 switch (Cond) { 1629 default: break; 1630 case ISD::SETFALSE: 1631 case ISD::SETFALSE2: return DAG.getConstant(0, VT); 1632 case ISD::SETTRUE: 1633 case ISD::SETTRUE2: return DAG.getConstant(1, VT); 1634 } 1635 1636 if (isa<ConstantSDNode>(N0.getNode())) { 1637 // Ensure that the constant occurs on the RHS, and fold constant 1638 // comparisons. 1639 return DAG.getSetCC(dl, VT, N1, N0, ISD::getSetCCSwappedOperands(Cond)); 1640 } 1641 1642 if (ConstantSDNode *N1C = dyn_cast<ConstantSDNode>(N1.getNode())) { 1643 const APInt &C1 = N1C->getAPIntValue(); 1644 1645 // If the LHS is '(srl (ctlz x), 5)', the RHS is 0/1, and this is an 1646 // equality comparison, then we're just comparing whether X itself is 1647 // zero. 1648 if (N0.getOpcode() == ISD::SRL && (C1 == 0 || C1 == 1) && 1649 N0.getOperand(0).getOpcode() == ISD::CTLZ && 1650 N0.getOperand(1).getOpcode() == ISD::Constant) { 1651 const APInt &ShAmt 1652 = cast<ConstantSDNode>(N0.getOperand(1))->getAPIntValue(); 1653 if ((Cond == ISD::SETEQ || Cond == ISD::SETNE) && 1654 ShAmt == Log2_32(N0.getValueType().getSizeInBits())) { 1655 if ((C1 == 0) == (Cond == ISD::SETEQ)) { 1656 // (srl (ctlz x), 5) == 0 -> X != 0 1657 // (srl (ctlz x), 5) != 1 -> X != 0 1658 Cond = ISD::SETNE; 1659 } else { 1660 // (srl (ctlz x), 5) != 0 -> X == 0 1661 // (srl (ctlz x), 5) == 1 -> X == 0 1662 Cond = ISD::SETEQ; 1663 } 1664 SDValue Zero = DAG.getConstant(0, N0.getValueType()); 1665 return DAG.getSetCC(dl, VT, N0.getOperand(0).getOperand(0), 1666 Zero, Cond); 1667 } 1668 } 1669 1670 // If the LHS is '(and load, const)', the RHS is 0, 1671 // the test is for equality or unsigned, and all 1 bits of the const are 1672 // in the same partial word, see if we can shorten the load. 1673 if (DCI.isBeforeLegalize() && 1674 N0.getOpcode() == ISD::AND && C1 == 0 && 1675 N0.getNode()->hasOneUse() && 1676 isa<LoadSDNode>(N0.getOperand(0)) && 1677 N0.getOperand(0).getNode()->hasOneUse() && 1678 isa<ConstantSDNode>(N0.getOperand(1))) { 1679 LoadSDNode *Lod = cast<LoadSDNode>(N0.getOperand(0)); 1680 APInt bestMask; 1681 unsigned bestWidth = 0, bestOffset = 0; 1682 if (!Lod->isVolatile() && Lod->isUnindexed()) { 1683 unsigned origWidth = N0.getValueType().getSizeInBits(); 1684 unsigned maskWidth = origWidth; 1685 // We can narrow (e.g.) 16-bit extending loads on 32-bit target to 1686 // 8 bits, but have to be careful... 1687 if (Lod->getExtensionType() != ISD::NON_EXTLOAD) 1688 origWidth = Lod->getMemoryVT().getSizeInBits(); 1689 const APInt &Mask = 1690 cast<ConstantSDNode>(N0.getOperand(1))->getAPIntValue(); 1691 for (unsigned width = origWidth / 2; width>=8; width /= 2) { 1692 APInt newMask = APInt::getLowBitsSet(maskWidth, width); 1693 for (unsigned offset=0; offset<origWidth/width; offset++) { 1694 if ((newMask & Mask) == Mask) { 1695 if (!TD->isLittleEndian()) 1696 bestOffset = (origWidth/width - offset - 1) * (width/8); 1697 else 1698 bestOffset = (uint64_t)offset * (width/8); 1699 bestMask = Mask.lshr(offset * (width/8) * 8); 1700 bestWidth = width; 1701 break; 1702 } 1703 newMask = newMask << width; 1704 } 1705 } 1706 } 1707 if (bestWidth) { 1708 EVT newVT = EVT::getIntegerVT(Context, bestWidth); 1709 if (newVT.isRound()) { 1710 EVT PtrType = Lod->getOperand(1).getValueType(); 1711 SDValue Ptr = Lod->getBasePtr(); 1712 if (bestOffset != 0) 1713 Ptr = DAG.getNode(ISD::ADD, dl, PtrType, Lod->getBasePtr(), 1714 DAG.getConstant(bestOffset, PtrType)); 1715 unsigned NewAlign = MinAlign(Lod->getAlignment(), bestOffset); 1716 SDValue NewLoad = DAG.getLoad(newVT, dl, Lod->getChain(), Ptr, 1717 Lod->getSrcValue(), 1718 Lod->getSrcValueOffset() + bestOffset, 1719 false, false, NewAlign); 1720 return DAG.getSetCC(dl, VT, 1721 DAG.getNode(ISD::AND, dl, newVT, NewLoad, 1722 DAG.getConstant(bestMask.trunc(bestWidth), 1723 newVT)), 1724 DAG.getConstant(0LL, newVT), Cond); 1725 } 1726 } 1727 } 1728 1729 // If the LHS is a ZERO_EXTEND, perform the comparison on the input. 1730 if (N0.getOpcode() == ISD::ZERO_EXTEND) { 1731 unsigned InSize = N0.getOperand(0).getValueType().getSizeInBits(); 1732 1733 // If the comparison constant has bits in the upper part, the 1734 // zero-extended value could never match. 1735 if (C1.intersects(APInt::getHighBitsSet(C1.getBitWidth(), 1736 C1.getBitWidth() - InSize))) { 1737 switch (Cond) { 1738 case ISD::SETUGT: 1739 case ISD::SETUGE: 1740 case ISD::SETEQ: return DAG.getConstant(0, VT); 1741 case ISD::SETULT: 1742 case ISD::SETULE: 1743 case ISD::SETNE: return DAG.getConstant(1, VT); 1744 case ISD::SETGT: 1745 case ISD::SETGE: 1746 // True if the sign bit of C1 is set. 1747 return DAG.getConstant(C1.isNegative(), VT); 1748 case ISD::SETLT: 1749 case ISD::SETLE: 1750 // True if the sign bit of C1 isn't set. 1751 return DAG.getConstant(C1.isNonNegative(), VT); 1752 default: 1753 break; 1754 } 1755 } 1756 1757 // Otherwise, we can perform the comparison with the low bits. 1758 switch (Cond) { 1759 case ISD::SETEQ: 1760 case ISD::SETNE: 1761 case ISD::SETUGT: 1762 case ISD::SETUGE: 1763 case ISD::SETULT: 1764 case ISD::SETULE: { 1765 EVT newVT = N0.getOperand(0).getValueType(); 1766 if (DCI.isBeforeLegalizeOps() || 1767 (isOperationLegal(ISD::SETCC, newVT) && 1768 getCondCodeAction(Cond, newVT)==Legal)) 1769 return DAG.getSetCC(dl, VT, N0.getOperand(0), 1770 DAG.getConstant(APInt(C1).trunc(InSize), newVT), 1771 Cond); 1772 break; 1773 } 1774 default: 1775 break; // todo, be more careful with signed comparisons 1776 } 1777 } else if (N0.getOpcode() == ISD::SIGN_EXTEND_INREG && 1778 (Cond == ISD::SETEQ || Cond == ISD::SETNE)) { 1779 EVT ExtSrcTy = cast<VTSDNode>(N0.getOperand(1))->getVT(); 1780 unsigned ExtSrcTyBits = ExtSrcTy.getSizeInBits(); 1781 EVT ExtDstTy = N0.getValueType(); 1782 unsigned ExtDstTyBits = ExtDstTy.getSizeInBits(); 1783 1784 // If the extended part has any inconsistent bits, it cannot ever 1785 // compare equal. In other words, they have to be all ones or all 1786 // zeros. 1787 APInt ExtBits = 1788 APInt::getHighBitsSet(ExtDstTyBits, ExtDstTyBits - ExtSrcTyBits); 1789 if ((C1 & ExtBits) != 0 && (C1 & ExtBits) != ExtBits) 1790 return DAG.getConstant(Cond == ISD::SETNE, VT); 1791 1792 SDValue ZextOp; 1793 EVT Op0Ty = N0.getOperand(0).getValueType(); 1794 if (Op0Ty == ExtSrcTy) { 1795 ZextOp = N0.getOperand(0); 1796 } else { 1797 APInt Imm = APInt::getLowBitsSet(ExtDstTyBits, ExtSrcTyBits); 1798 ZextOp = DAG.getNode(ISD::AND, dl, Op0Ty, N0.getOperand(0), 1799 DAG.getConstant(Imm, Op0Ty)); 1800 } 1801 if (!DCI.isCalledByLegalizer()) 1802 DCI.AddToWorklist(ZextOp.getNode()); 1803 // Otherwise, make this a use of a zext. 1804 return DAG.getSetCC(dl, VT, ZextOp, 1805 DAG.getConstant(C1 & APInt::getLowBitsSet( 1806 ExtDstTyBits, 1807 ExtSrcTyBits), 1808 ExtDstTy), 1809 Cond); 1810 } else if ((N1C->isNullValue() || N1C->getAPIntValue() == 1) && 1811 (Cond == ISD::SETEQ || Cond == ISD::SETNE)) { 1812 // SETCC (SETCC), [0|1], [EQ|NE] -> SETCC 1813 if (N0.getOpcode() == ISD::SETCC) { 1814 bool TrueWhenTrue = (Cond == ISD::SETEQ) ^ (N1C->getAPIntValue() != 1); 1815 if (TrueWhenTrue) 1816 return N0; 1817 1818 // Invert the condition. 1819 ISD::CondCode CC = cast<CondCodeSDNode>(N0.getOperand(2))->get(); 1820 CC = ISD::getSetCCInverse(CC, 1821 N0.getOperand(0).getValueType().isInteger()); 1822 return DAG.getSetCC(dl, VT, N0.getOperand(0), N0.getOperand(1), CC); 1823 } 1824 1825 if ((N0.getOpcode() == ISD::XOR || 1826 (N0.getOpcode() == ISD::AND && 1827 N0.getOperand(0).getOpcode() == ISD::XOR && 1828 N0.getOperand(1) == N0.getOperand(0).getOperand(1))) && 1829 isa<ConstantSDNode>(N0.getOperand(1)) && 1830 cast<ConstantSDNode>(N0.getOperand(1))->getAPIntValue() == 1) { 1831 // If this is (X^1) == 0/1, swap the RHS and eliminate the xor. We 1832 // can only do this if the top bits are known zero. 1833 unsigned BitWidth = N0.getValueSizeInBits(); 1834 if (DAG.MaskedValueIsZero(N0, 1835 APInt::getHighBitsSet(BitWidth, 1836 BitWidth-1))) { 1837 // Okay, get the un-inverted input value. 1838 SDValue Val; 1839 if (N0.getOpcode() == ISD::XOR) 1840 Val = N0.getOperand(0); 1841 else { 1842 assert(N0.getOpcode() == ISD::AND && 1843 N0.getOperand(0).getOpcode() == ISD::XOR); 1844 // ((X^1)&1)^1 -> X & 1 1845 Val = DAG.getNode(ISD::AND, dl, N0.getValueType(), 1846 N0.getOperand(0).getOperand(0), 1847 N0.getOperand(1)); 1848 } 1849 1850 return DAG.getSetCC(dl, VT, Val, N1, 1851 Cond == ISD::SETEQ ? ISD::SETNE : ISD::SETEQ); 1852 } 1853 } else if (N1C->getAPIntValue() == 1) { 1854 // If this is (X&1) == / != 1, normalize it to (X&1) != / == 0. 1855 SDValue Op0 = N0; 1856 if (Op0.getOpcode() == ISD::TRUNCATE) 1857 Op0 = Op0.getOperand(0); 1858 1859 if ((Op0.getOpcode() == ISD::XOR || Op0.getOpcode() == ISD::AND) && 1860 Op0.getOperand(0).getOpcode() == ISD::SETCC && 1861 Op0.getOperand(1).getOpcode() == ISD::SETCC) { 1862 // (and (setcc), (setcc)) == / != 1 -> (setcc) == / != (setcc) 1863 // (xor (setcc), (setcc)) == / != 1 -> (setcc) != / == (setcc) 1864 if (Op0.getOpcode() == ISD::XOR) 1865 Cond = (Cond == ISD::SETEQ) ? ISD::SETNE : ISD::SETEQ; 1866 return DAG.getSetCC(dl, VT, Op0.getOperand(0), Op0.getOperand(1), 1867 Cond); 1868 } else if (Op0.getOpcode() == ISD::AND && 1869 isa<ConstantSDNode>(Op0.getOperand(1)) && 1870 cast<ConstantSDNode>(Op0.getOperand(1))->getAPIntValue() == 1) { 1871 if (Op0.getValueType() != VT) 1872 Op0 = DAG.getNode(ISD::AND, dl, VT, 1873 DAG.getNode(ISD::TRUNCATE, dl, VT, Op0.getOperand(0)), 1874 DAG.getConstant(1, VT)); 1875 return DAG.getSetCC(dl, VT, Op0, 1876 DAG.getConstant(0, Op0.getValueType()), 1877 Cond == ISD::SETEQ ? ISD::SETNE : ISD::SETEQ); 1878 } 1879 } 1880 } 1881 1882 APInt MinVal, MaxVal; 1883 unsigned OperandBitSize = N1C->getValueType(0).getSizeInBits(); 1884 if (ISD::isSignedIntSetCC(Cond)) { 1885 MinVal = APInt::getSignedMinValue(OperandBitSize); 1886 MaxVal = APInt::getSignedMaxValue(OperandBitSize); 1887 } else { 1888 MinVal = APInt::getMinValue(OperandBitSize); 1889 MaxVal = APInt::getMaxValue(OperandBitSize); 1890 } 1891 1892 // Canonicalize GE/LE comparisons to use GT/LT comparisons. 1893 if (Cond == ISD::SETGE || Cond == ISD::SETUGE) { 1894 if (C1 == MinVal) return DAG.getConstant(1, VT); // X >= MIN --> true 1895 // X >= C0 --> X > (C0-1) 1896 return DAG.getSetCC(dl, VT, N0, 1897 DAG.getConstant(C1-1, N1.getValueType()), 1898 (Cond == ISD::SETGE) ? ISD::SETGT : ISD::SETUGT); 1899 } 1900 1901 if (Cond == ISD::SETLE || Cond == ISD::SETULE) { 1902 if (C1 == MaxVal) return DAG.getConstant(1, VT); // X <= MAX --> true 1903 // X <= C0 --> X < (C0+1) 1904 return DAG.getSetCC(dl, VT, N0, 1905 DAG.getConstant(C1+1, N1.getValueType()), 1906 (Cond == ISD::SETLE) ? ISD::SETLT : ISD::SETULT); 1907 } 1908 1909 if ((Cond == ISD::SETLT || Cond == ISD::SETULT) && C1 == MinVal) 1910 return DAG.getConstant(0, VT); // X < MIN --> false 1911 if ((Cond == ISD::SETGE || Cond == ISD::SETUGE) && C1 == MinVal) 1912 return DAG.getConstant(1, VT); // X >= MIN --> true 1913 if ((Cond == ISD::SETGT || Cond == ISD::SETUGT) && C1 == MaxVal) 1914 return DAG.getConstant(0, VT); // X > MAX --> false 1915 if ((Cond == ISD::SETLE || Cond == ISD::SETULE) && C1 == MaxVal) 1916 return DAG.getConstant(1, VT); // X <= MAX --> true 1917 1918 // Canonicalize setgt X, Min --> setne X, Min 1919 if ((Cond == ISD::SETGT || Cond == ISD::SETUGT) && C1 == MinVal) 1920 return DAG.getSetCC(dl, VT, N0, N1, ISD::SETNE); 1921 // Canonicalize setlt X, Max --> setne X, Max 1922 if ((Cond == ISD::SETLT || Cond == ISD::SETULT) && C1 == MaxVal) 1923 return DAG.getSetCC(dl, VT, N0, N1, ISD::SETNE); 1924 1925 // If we have setult X, 1, turn it into seteq X, 0 1926 if ((Cond == ISD::SETLT || Cond == ISD::SETULT) && C1 == MinVal+1) 1927 return DAG.getSetCC(dl, VT, N0, 1928 DAG.getConstant(MinVal, N0.getValueType()), 1929 ISD::SETEQ); 1930 // If we have setugt X, Max-1, turn it into seteq X, Max 1931 else if ((Cond == ISD::SETGT || Cond == ISD::SETUGT) && C1 == MaxVal-1) 1932 return DAG.getSetCC(dl, VT, N0, 1933 DAG.getConstant(MaxVal, N0.getValueType()), 1934 ISD::SETEQ); 1935 1936 // If we have "setcc X, C0", check to see if we can shrink the immediate 1937 // by changing cc. 1938 1939 // SETUGT X, SINTMAX -> SETLT X, 0 1940 if (Cond == ISD::SETUGT && 1941 C1 == APInt::getSignedMaxValue(OperandBitSize)) 1942 return DAG.getSetCC(dl, VT, N0, 1943 DAG.getConstant(0, N1.getValueType()), 1944 ISD::SETLT); 1945 1946 // SETULT X, SINTMIN -> SETGT X, -1 1947 if (Cond == ISD::SETULT && 1948 C1 == APInt::getSignedMinValue(OperandBitSize)) { 1949 SDValue ConstMinusOne = 1950 DAG.getConstant(APInt::getAllOnesValue(OperandBitSize), 1951 N1.getValueType()); 1952 return DAG.getSetCC(dl, VT, N0, ConstMinusOne, ISD::SETGT); 1953 } 1954 1955 // Fold bit comparisons when we can. 1956 if ((Cond == ISD::SETEQ || Cond == ISD::SETNE) && 1957 (VT == N0.getValueType() || 1958 (isTypeLegal(VT) && VT.bitsLE(N0.getValueType()))) && 1959 N0.getOpcode() == ISD::AND) 1960 if (ConstantSDNode *AndRHS = 1961 dyn_cast<ConstantSDNode>(N0.getOperand(1))) { 1962 EVT ShiftTy = DCI.isBeforeLegalize() ? 1963 getPointerTy() : getShiftAmountTy(); 1964 if (Cond == ISD::SETNE && C1 == 0) {// (X & 8) != 0 --> (X & 8) >> 3 1965 // Perform the xform if the AND RHS is a single bit. 1966 if (AndRHS->getAPIntValue().isPowerOf2()) { 1967 return DAG.getNode(ISD::TRUNCATE, dl, VT, 1968 DAG.getNode(ISD::SRL, dl, N0.getValueType(), N0, 1969 DAG.getConstant(AndRHS->getAPIntValue().logBase2(), ShiftTy))); 1970 } 1971 } else if (Cond == ISD::SETEQ && C1 == AndRHS->getAPIntValue()) { 1972 // (X & 8) == 8 --> (X & 8) >> 3 1973 // Perform the xform if C1 is a single bit. 1974 if (C1.isPowerOf2()) { 1975 return DAG.getNode(ISD::TRUNCATE, dl, VT, 1976 DAG.getNode(ISD::SRL, dl, N0.getValueType(), N0, 1977 DAG.getConstant(C1.logBase2(), ShiftTy))); 1978 } 1979 } 1980 } 1981 } 1982 1983 if (isa<ConstantFPSDNode>(N0.getNode())) { 1984 // Constant fold or commute setcc. 1985 SDValue O = DAG.FoldSetCC(VT, N0, N1, Cond, dl); 1986 if (O.getNode()) return O; 1987 } else if (ConstantFPSDNode *CFP = dyn_cast<ConstantFPSDNode>(N1.getNode())) { 1988 // If the RHS of an FP comparison is a constant, simplify it away in 1989 // some cases. 1990 if (CFP->getValueAPF().isNaN()) { 1991 // If an operand is known to be a nan, we can fold it. 1992 switch (ISD::getUnorderedFlavor(Cond)) { 1993 default: llvm_unreachable("Unknown flavor!"); 1994 case 0: // Known false. 1995 return DAG.getConstant(0, VT); 1996 case 1: // Known true. 1997 return DAG.getConstant(1, VT); 1998 case 2: // Undefined. 1999 return DAG.getUNDEF(VT); 2000 } 2001 } 2002 2003 // Otherwise, we know the RHS is not a NaN. Simplify the node to drop the 2004 // constant if knowing that the operand is non-nan is enough. We prefer to 2005 // have SETO(x,x) instead of SETO(x, 0.0) because this avoids having to 2006 // materialize 0.0. 2007 if (Cond == ISD::SETO || Cond == ISD::SETUO) 2008 return DAG.getSetCC(dl, VT, N0, N0, Cond); 2009 2010 // If the condition is not legal, see if we can find an equivalent one 2011 // which is legal. 2012 if (!isCondCodeLegal(Cond, N0.getValueType())) { 2013 // If the comparison was an awkward floating-point == or != and one of 2014 // the comparison operands is infinity or negative infinity, convert the 2015 // condition to a less-awkward <= or >=. 2016 if (CFP->getValueAPF().isInfinity()) { 2017 if (CFP->getValueAPF().isNegative()) { 2018 if (Cond == ISD::SETOEQ && 2019 isCondCodeLegal(ISD::SETOLE, N0.getValueType())) 2020 return DAG.getSetCC(dl, VT, N0, N1, ISD::SETOLE); 2021 if (Cond == ISD::SETUEQ && 2022 isCondCodeLegal(ISD::SETOLE, N0.getValueType())) 2023 return DAG.getSetCC(dl, VT, N0, N1, ISD::SETULE); 2024 if (Cond == ISD::SETUNE && 2025 isCondCodeLegal(ISD::SETUGT, N0.getValueType())) 2026 return DAG.getSetCC(dl, VT, N0, N1, ISD::SETUGT); 2027 if (Cond == ISD::SETONE && 2028 isCondCodeLegal(ISD::SETUGT, N0.getValueType())) 2029 return DAG.getSetCC(dl, VT, N0, N1, ISD::SETOGT); 2030 } else { 2031 if (Cond == ISD::SETOEQ && 2032 isCondCodeLegal(ISD::SETOGE, N0.getValueType())) 2033 return DAG.getSetCC(dl, VT, N0, N1, ISD::SETOGE); 2034 if (Cond == ISD::SETUEQ && 2035 isCondCodeLegal(ISD::SETOGE, N0.getValueType())) 2036 return DAG.getSetCC(dl, VT, N0, N1, ISD::SETUGE); 2037 if (Cond == ISD::SETUNE && 2038 isCondCodeLegal(ISD::SETULT, N0.getValueType())) 2039 return DAG.getSetCC(dl, VT, N0, N1, ISD::SETULT); 2040 if (Cond == ISD::SETONE && 2041 isCondCodeLegal(ISD::SETULT, N0.getValueType())) 2042 return DAG.getSetCC(dl, VT, N0, N1, ISD::SETOLT); 2043 } 2044 } 2045 } 2046 } 2047 2048 if (N0 == N1) { 2049 // We can always fold X == X for integer setcc's. 2050 if (N0.getValueType().isInteger()) 2051 return DAG.getConstant(ISD::isTrueWhenEqual(Cond), VT); 2052 unsigned UOF = ISD::getUnorderedFlavor(Cond); 2053 if (UOF == 2) // FP operators that are undefined on NaNs. 2054 return DAG.getConstant(ISD::isTrueWhenEqual(Cond), VT); 2055 if (UOF == unsigned(ISD::isTrueWhenEqual(Cond))) 2056 return DAG.getConstant(UOF, VT); 2057 // Otherwise, we can't fold it. However, we can simplify it to SETUO/SETO 2058 // if it is not already. 2059 ISD::CondCode NewCond = UOF == 0 ? ISD::SETO : ISD::SETUO; 2060 if (NewCond != Cond) 2061 return DAG.getSetCC(dl, VT, N0, N1, NewCond); 2062 } 2063 2064 if ((Cond == ISD::SETEQ || Cond == ISD::SETNE) && 2065 N0.getValueType().isInteger()) { 2066 if (N0.getOpcode() == ISD::ADD || N0.getOpcode() == ISD::SUB || 2067 N0.getOpcode() == ISD::XOR) { 2068 // Simplify (X+Y) == (X+Z) --> Y == Z 2069 if (N0.getOpcode() == N1.getOpcode()) { 2070 if (N0.getOperand(0) == N1.getOperand(0)) 2071 return DAG.getSetCC(dl, VT, N0.getOperand(1), N1.getOperand(1), Cond); 2072 if (N0.getOperand(1) == N1.getOperand(1)) 2073 return DAG.getSetCC(dl, VT, N0.getOperand(0), N1.getOperand(0), Cond); 2074 if (DAG.isCommutativeBinOp(N0.getOpcode())) { 2075 // If X op Y == Y op X, try other combinations. 2076 if (N0.getOperand(0) == N1.getOperand(1)) 2077 return DAG.getSetCC(dl, VT, N0.getOperand(1), N1.getOperand(0), 2078 Cond); 2079 if (N0.getOperand(1) == N1.getOperand(0)) 2080 return DAG.getSetCC(dl, VT, N0.getOperand(0), N1.getOperand(1), 2081 Cond); 2082 } 2083 } 2084 2085 if (ConstantSDNode *RHSC = dyn_cast<ConstantSDNode>(N1)) { 2086 if (ConstantSDNode *LHSR = dyn_cast<ConstantSDNode>(N0.getOperand(1))) { 2087 // Turn (X+C1) == C2 --> X == C2-C1 2088 if (N0.getOpcode() == ISD::ADD && N0.getNode()->hasOneUse()) { 2089 return DAG.getSetCC(dl, VT, N0.getOperand(0), 2090 DAG.getConstant(RHSC->getAPIntValue()- 2091 LHSR->getAPIntValue(), 2092 N0.getValueType()), Cond); 2093 } 2094 2095 // Turn (X^C1) == C2 into X == C1^C2 iff X&~C1 = 0. 2096 if (N0.getOpcode() == ISD::XOR) 2097 // If we know that all of the inverted bits are zero, don't bother 2098 // performing the inversion. 2099 if (DAG.MaskedValueIsZero(N0.getOperand(0), ~LHSR->getAPIntValue())) 2100 return 2101 DAG.getSetCC(dl, VT, N0.getOperand(0), 2102 DAG.getConstant(LHSR->getAPIntValue() ^ 2103 RHSC->getAPIntValue(), 2104 N0.getValueType()), 2105 Cond); 2106 } 2107 2108 // Turn (C1-X) == C2 --> X == C1-C2 2109 if (ConstantSDNode *SUBC = dyn_cast<ConstantSDNode>(N0.getOperand(0))) { 2110 if (N0.getOpcode() == ISD::SUB && N0.getNode()->hasOneUse()) { 2111 return 2112 DAG.getSetCC(dl, VT, N0.getOperand(1), 2113 DAG.getConstant(SUBC->getAPIntValue() - 2114 RHSC->getAPIntValue(), 2115 N0.getValueType()), 2116 Cond); 2117 } 2118 } 2119 } 2120 2121 // Simplify (X+Z) == X --> Z == 0 2122 if (N0.getOperand(0) == N1) 2123 return DAG.getSetCC(dl, VT, N0.getOperand(1), 2124 DAG.getConstant(0, N0.getValueType()), Cond); 2125 if (N0.getOperand(1) == N1) { 2126 if (DAG.isCommutativeBinOp(N0.getOpcode())) 2127 return DAG.getSetCC(dl, VT, N0.getOperand(0), 2128 DAG.getConstant(0, N0.getValueType()), Cond); 2129 else if (N0.getNode()->hasOneUse()) { 2130 assert(N0.getOpcode() == ISD::SUB && "Unexpected operation!"); 2131 // (Z-X) == X --> Z == X<<1 2132 SDValue SH = DAG.getNode(ISD::SHL, dl, N1.getValueType(), 2133 N1, 2134 DAG.getConstant(1, getShiftAmountTy())); 2135 if (!DCI.isCalledByLegalizer()) 2136 DCI.AddToWorklist(SH.getNode()); 2137 return DAG.getSetCC(dl, VT, N0.getOperand(0), SH, Cond); 2138 } 2139 } 2140 } 2141 2142 if (N1.getOpcode() == ISD::ADD || N1.getOpcode() == ISD::SUB || 2143 N1.getOpcode() == ISD::XOR) { 2144 // Simplify X == (X+Z) --> Z == 0 2145 if (N1.getOperand(0) == N0) { 2146 return DAG.getSetCC(dl, VT, N1.getOperand(1), 2147 DAG.getConstant(0, N1.getValueType()), Cond); 2148 } else if (N1.getOperand(1) == N0) { 2149 if (DAG.isCommutativeBinOp(N1.getOpcode())) { 2150 return DAG.getSetCC(dl, VT, N1.getOperand(0), 2151 DAG.getConstant(0, N1.getValueType()), Cond); 2152 } else if (N1.getNode()->hasOneUse()) { 2153 assert(N1.getOpcode() == ISD::SUB && "Unexpected operation!"); 2154 // X == (Z-X) --> X<<1 == Z 2155 SDValue SH = DAG.getNode(ISD::SHL, dl, N1.getValueType(), N0, 2156 DAG.getConstant(1, getShiftAmountTy())); 2157 if (!DCI.isCalledByLegalizer()) 2158 DCI.AddToWorklist(SH.getNode()); 2159 return DAG.getSetCC(dl, VT, SH, N1.getOperand(0), Cond); 2160 } 2161 } 2162 } 2163 2164 // Simplify x&y == y to x&y != 0 if y has exactly one bit set. 2165 // Note that where y is variable and is known to have at most 2166 // one bit set (for example, if it is z&1) we cannot do this; 2167 // the expressions are not equivalent when y==0. 2168 if (N0.getOpcode() == ISD::AND) 2169 if (N0.getOperand(0) == N1 || N0.getOperand(1) == N1) { 2170 if (ValueHasExactlyOneBitSet(N1, DAG)) { 2171 Cond = ISD::getSetCCInverse(Cond, /*isInteger=*/true); 2172 SDValue Zero = DAG.getConstant(0, N1.getValueType()); 2173 return DAG.getSetCC(dl, VT, N0, Zero, Cond); 2174 } 2175 } 2176 if (N1.getOpcode() == ISD::AND) 2177 if (N1.getOperand(0) == N0 || N1.getOperand(1) == N0) { 2178 if (ValueHasExactlyOneBitSet(N0, DAG)) { 2179 Cond = ISD::getSetCCInverse(Cond, /*isInteger=*/true); 2180 SDValue Zero = DAG.getConstant(0, N0.getValueType()); 2181 return DAG.getSetCC(dl, VT, N1, Zero, Cond); 2182 } 2183 } 2184 } 2185 2186 // Fold away ALL boolean setcc's. 2187 SDValue Temp; 2188 if (N0.getValueType() == MVT::i1 && foldBooleans) { 2189 switch (Cond) { 2190 default: llvm_unreachable("Unknown integer setcc!"); 2191 case ISD::SETEQ: // X == Y -> ~(X^Y) 2192 Temp = DAG.getNode(ISD::XOR, dl, MVT::i1, N0, N1); 2193 N0 = DAG.getNOT(dl, Temp, MVT::i1); 2194 if (!DCI.isCalledByLegalizer()) 2195 DCI.AddToWorklist(Temp.getNode()); 2196 break; 2197 case ISD::SETNE: // X != Y --> (X^Y) 2198 N0 = DAG.getNode(ISD::XOR, dl, MVT::i1, N0, N1); 2199 break; 2200 case ISD::SETGT: // X >s Y --> X == 0 & Y == 1 --> ~X & Y 2201 case ISD::SETULT: // X <u Y --> X == 0 & Y == 1 --> ~X & Y 2202 Temp = DAG.getNOT(dl, N0, MVT::i1); 2203 N0 = DAG.getNode(ISD::AND, dl, MVT::i1, N1, Temp); 2204 if (!DCI.isCalledByLegalizer()) 2205 DCI.AddToWorklist(Temp.getNode()); 2206 break; 2207 case ISD::SETLT: // X <s Y --> X == 1 & Y == 0 --> ~Y & X 2208 case ISD::SETUGT: // X >u Y --> X == 1 & Y == 0 --> ~Y & X 2209 Temp = DAG.getNOT(dl, N1, MVT::i1); 2210 N0 = DAG.getNode(ISD::AND, dl, MVT::i1, N0, Temp); 2211 if (!DCI.isCalledByLegalizer()) 2212 DCI.AddToWorklist(Temp.getNode()); 2213 break; 2214 case ISD::SETULE: // X <=u Y --> X == 0 | Y == 1 --> ~X | Y 2215 case ISD::SETGE: // X >=s Y --> X == 0 | Y == 1 --> ~X | Y 2216 Temp = DAG.getNOT(dl, N0, MVT::i1); 2217 N0 = DAG.getNode(ISD::OR, dl, MVT::i1, N1, Temp); 2218 if (!DCI.isCalledByLegalizer()) 2219 DCI.AddToWorklist(Temp.getNode()); 2220 break; 2221 case ISD::SETUGE: // X >=u Y --> X == 1 | Y == 0 --> ~Y | X 2222 case ISD::SETLE: // X <=s Y --> X == 1 | Y == 0 --> ~Y | X 2223 Temp = DAG.getNOT(dl, N1, MVT::i1); 2224 N0 = DAG.getNode(ISD::OR, dl, MVT::i1, N0, Temp); 2225 break; 2226 } 2227 if (VT != MVT::i1) { 2228 if (!DCI.isCalledByLegalizer()) 2229 DCI.AddToWorklist(N0.getNode()); 2230 // FIXME: If running after legalize, we probably can't do this. 2231 N0 = DAG.getNode(ISD::ZERO_EXTEND, dl, VT, N0); 2232 } 2233 return N0; 2234 } 2235 2236 // Could not fold it. 2237 return SDValue(); 2238} 2239 2240/// isGAPlusOffset - Returns true (and the GlobalValue and the offset) if the 2241/// node is a GlobalAddress + offset. 2242bool TargetLowering::isGAPlusOffset(SDNode *N, GlobalValue* &GA, 2243 int64_t &Offset) const { 2244 if (isa<GlobalAddressSDNode>(N)) { 2245 GlobalAddressSDNode *GASD = cast<GlobalAddressSDNode>(N); 2246 GA = GASD->getGlobal(); 2247 Offset += GASD->getOffset(); 2248 return true; 2249 } 2250 2251 if (N->getOpcode() == ISD::ADD) { 2252 SDValue N1 = N->getOperand(0); 2253 SDValue N2 = N->getOperand(1); 2254 if (isGAPlusOffset(N1.getNode(), GA, Offset)) { 2255 ConstantSDNode *V = dyn_cast<ConstantSDNode>(N2); 2256 if (V) { 2257 Offset += V->getSExtValue(); 2258 return true; 2259 } 2260 } else if (isGAPlusOffset(N2.getNode(), GA, Offset)) { 2261 ConstantSDNode *V = dyn_cast<ConstantSDNode>(N1); 2262 if (V) { 2263 Offset += V->getSExtValue(); 2264 return true; 2265 } 2266 } 2267 } 2268 return false; 2269} 2270 2271 2272SDValue TargetLowering:: 2273PerformDAGCombine(SDNode *N, DAGCombinerInfo &DCI) const { 2274 // Default implementation: no optimization. 2275 return SDValue(); 2276} 2277 2278//===----------------------------------------------------------------------===// 2279// Inline Assembler Implementation Methods 2280//===----------------------------------------------------------------------===// 2281 2282 2283TargetLowering::ConstraintType 2284TargetLowering::getConstraintType(const std::string &Constraint) const { 2285 // FIXME: lots more standard ones to handle. 2286 if (Constraint.size() == 1) { 2287 switch (Constraint[0]) { 2288 default: break; 2289 case 'r': return C_RegisterClass; 2290 case 'm': // memory 2291 case 'o': // offsetable 2292 case 'V': // not offsetable 2293 return C_Memory; 2294 case 'i': // Simple Integer or Relocatable Constant 2295 case 'n': // Simple Integer 2296 case 's': // Relocatable Constant 2297 case 'X': // Allow ANY value. 2298 case 'I': // Target registers. 2299 case 'J': 2300 case 'K': 2301 case 'L': 2302 case 'M': 2303 case 'N': 2304 case 'O': 2305 case 'P': 2306 return C_Other; 2307 } 2308 } 2309 2310 if (Constraint.size() > 1 && Constraint[0] == '{' && 2311 Constraint[Constraint.size()-1] == '}') 2312 return C_Register; 2313 return C_Unknown; 2314} 2315 2316/// LowerXConstraint - try to replace an X constraint, which matches anything, 2317/// with another that has more specific requirements based on the type of the 2318/// corresponding operand. 2319const char *TargetLowering::LowerXConstraint(EVT ConstraintVT) const{ 2320 if (ConstraintVT.isInteger()) 2321 return "r"; 2322 if (ConstraintVT.isFloatingPoint()) 2323 return "f"; // works for many targets 2324 return 0; 2325} 2326 2327/// LowerAsmOperandForConstraint - Lower the specified operand into the Ops 2328/// vector. If it is invalid, don't add anything to Ops. 2329void TargetLowering::LowerAsmOperandForConstraint(SDValue Op, 2330 char ConstraintLetter, 2331 bool hasMemory, 2332 std::vector<SDValue> &Ops, 2333 SelectionDAG &DAG) const { 2334 switch (ConstraintLetter) { 2335 default: break; 2336 case 'X': // Allows any operand; labels (basic block) use this. 2337 if (Op.getOpcode() == ISD::BasicBlock) { 2338 Ops.push_back(Op); 2339 return; 2340 } 2341 // fall through 2342 case 'i': // Simple Integer or Relocatable Constant 2343 case 'n': // Simple Integer 2344 case 's': { // Relocatable Constant 2345 // These operands are interested in values of the form (GV+C), where C may 2346 // be folded in as an offset of GV, or it may be explicitly added. Also, it 2347 // is possible and fine if either GV or C are missing. 2348 ConstantSDNode *C = dyn_cast<ConstantSDNode>(Op); 2349 GlobalAddressSDNode *GA = dyn_cast<GlobalAddressSDNode>(Op); 2350 2351 // If we have "(add GV, C)", pull out GV/C 2352 if (Op.getOpcode() == ISD::ADD) { 2353 C = dyn_cast<ConstantSDNode>(Op.getOperand(1)); 2354 GA = dyn_cast<GlobalAddressSDNode>(Op.getOperand(0)); 2355 if (C == 0 || GA == 0) { 2356 C = dyn_cast<ConstantSDNode>(Op.getOperand(0)); 2357 GA = dyn_cast<GlobalAddressSDNode>(Op.getOperand(1)); 2358 } 2359 if (C == 0 || GA == 0) 2360 C = 0, GA = 0; 2361 } 2362 2363 // If we find a valid operand, map to the TargetXXX version so that the 2364 // value itself doesn't get selected. 2365 if (GA) { // Either &GV or &GV+C 2366 if (ConstraintLetter != 'n') { 2367 int64_t Offs = GA->getOffset(); 2368 if (C) Offs += C->getZExtValue(); 2369 Ops.push_back(DAG.getTargetGlobalAddress(GA->getGlobal(), 2370 Op.getValueType(), Offs)); 2371 return; 2372 } 2373 } 2374 if (C) { // just C, no GV. 2375 // Simple constants are not allowed for 's'. 2376 if (ConstraintLetter != 's') { 2377 // gcc prints these as sign extended. Sign extend value to 64 bits 2378 // now; without this it would get ZExt'd later in 2379 // ScheduleDAGSDNodes::EmitNode, which is very generic. 2380 Ops.push_back(DAG.getTargetConstant(C->getAPIntValue().getSExtValue(), 2381 MVT::i64)); 2382 return; 2383 } 2384 } 2385 break; 2386 } 2387 } 2388} 2389 2390std::vector<unsigned> TargetLowering:: 2391getRegClassForInlineAsmConstraint(const std::string &Constraint, 2392 EVT VT) const { 2393 return std::vector<unsigned>(); 2394} 2395 2396 2397std::pair<unsigned, const TargetRegisterClass*> TargetLowering:: 2398getRegForInlineAsmConstraint(const std::string &Constraint, 2399 EVT VT) const { 2400 if (Constraint[0] != '{') 2401 return std::pair<unsigned, const TargetRegisterClass*>(0, 0); 2402 assert(*(Constraint.end()-1) == '}' && "Not a brace enclosed constraint?"); 2403 2404 // Remove the braces from around the name. 2405 StringRef RegName(Constraint.data()+1, Constraint.size()-2); 2406 2407 // Figure out which register class contains this reg. 2408 const TargetRegisterInfo *RI = TM.getRegisterInfo(); 2409 for (TargetRegisterInfo::regclass_iterator RCI = RI->regclass_begin(), 2410 E = RI->regclass_end(); RCI != E; ++RCI) { 2411 const TargetRegisterClass *RC = *RCI; 2412 2413 // If none of the value types for this register class are valid, we 2414 // can't use it. For example, 64-bit reg classes on 32-bit targets. 2415 bool isLegal = false; 2416 for (TargetRegisterClass::vt_iterator I = RC->vt_begin(), E = RC->vt_end(); 2417 I != E; ++I) { 2418 if (isTypeLegal(*I)) { 2419 isLegal = true; 2420 break; 2421 } 2422 } 2423 2424 if (!isLegal) continue; 2425 2426 for (TargetRegisterClass::iterator I = RC->begin(), E = RC->end(); 2427 I != E; ++I) { 2428 if (RegName.equals_lower(RI->getName(*I))) 2429 return std::make_pair(*I, RC); 2430 } 2431 } 2432 2433 return std::pair<unsigned, const TargetRegisterClass*>(0, 0); 2434} 2435 2436//===----------------------------------------------------------------------===// 2437// Constraint Selection. 2438 2439/// isMatchingInputConstraint - Return true of this is an input operand that is 2440/// a matching constraint like "4". 2441bool TargetLowering::AsmOperandInfo::isMatchingInputConstraint() const { 2442 assert(!ConstraintCode.empty() && "No known constraint!"); 2443 return isdigit(ConstraintCode[0]); 2444} 2445 2446/// getMatchedOperand - If this is an input matching constraint, this method 2447/// returns the output operand it matches. 2448unsigned TargetLowering::AsmOperandInfo::getMatchedOperand() const { 2449 assert(!ConstraintCode.empty() && "No known constraint!"); 2450 return atoi(ConstraintCode.c_str()); 2451} 2452 2453 2454/// getConstraintGenerality - Return an integer indicating how general CT 2455/// is. 2456static unsigned getConstraintGenerality(TargetLowering::ConstraintType CT) { 2457 switch (CT) { 2458 default: llvm_unreachable("Unknown constraint type!"); 2459 case TargetLowering::C_Other: 2460 case TargetLowering::C_Unknown: 2461 return 0; 2462 case TargetLowering::C_Register: 2463 return 1; 2464 case TargetLowering::C_RegisterClass: 2465 return 2; 2466 case TargetLowering::C_Memory: 2467 return 3; 2468 } 2469} 2470 2471/// ChooseConstraint - If there are multiple different constraints that we 2472/// could pick for this operand (e.g. "imr") try to pick the 'best' one. 2473/// This is somewhat tricky: constraints fall into four classes: 2474/// Other -> immediates and magic values 2475/// Register -> one specific register 2476/// RegisterClass -> a group of regs 2477/// Memory -> memory 2478/// Ideally, we would pick the most specific constraint possible: if we have 2479/// something that fits into a register, we would pick it. The problem here 2480/// is that if we have something that could either be in a register or in 2481/// memory that use of the register could cause selection of *other* 2482/// operands to fail: they might only succeed if we pick memory. Because of 2483/// this the heuristic we use is: 2484/// 2485/// 1) If there is an 'other' constraint, and if the operand is valid for 2486/// that constraint, use it. This makes us take advantage of 'i' 2487/// constraints when available. 2488/// 2) Otherwise, pick the most general constraint present. This prefers 2489/// 'm' over 'r', for example. 2490/// 2491static void ChooseConstraint(TargetLowering::AsmOperandInfo &OpInfo, 2492 bool hasMemory, const TargetLowering &TLI, 2493 SDValue Op, SelectionDAG *DAG) { 2494 assert(OpInfo.Codes.size() > 1 && "Doesn't have multiple constraint options"); 2495 unsigned BestIdx = 0; 2496 TargetLowering::ConstraintType BestType = TargetLowering::C_Unknown; 2497 int BestGenerality = -1; 2498 2499 // Loop over the options, keeping track of the most general one. 2500 for (unsigned i = 0, e = OpInfo.Codes.size(); i != e; ++i) { 2501 TargetLowering::ConstraintType CType = 2502 TLI.getConstraintType(OpInfo.Codes[i]); 2503 2504 // If this is an 'other' constraint, see if the operand is valid for it. 2505 // For example, on X86 we might have an 'rI' constraint. If the operand 2506 // is an integer in the range [0..31] we want to use I (saving a load 2507 // of a register), otherwise we must use 'r'. 2508 if (CType == TargetLowering::C_Other && Op.getNode()) { 2509 assert(OpInfo.Codes[i].size() == 1 && 2510 "Unhandled multi-letter 'other' constraint"); 2511 std::vector<SDValue> ResultOps; 2512 TLI.LowerAsmOperandForConstraint(Op, OpInfo.Codes[i][0], hasMemory, 2513 ResultOps, *DAG); 2514 if (!ResultOps.empty()) { 2515 BestType = CType; 2516 BestIdx = i; 2517 break; 2518 } 2519 } 2520 2521 // This constraint letter is more general than the previous one, use it. 2522 int Generality = getConstraintGenerality(CType); 2523 if (Generality > BestGenerality) { 2524 BestType = CType; 2525 BestIdx = i; 2526 BestGenerality = Generality; 2527 } 2528 } 2529 2530 OpInfo.ConstraintCode = OpInfo.Codes[BestIdx]; 2531 OpInfo.ConstraintType = BestType; 2532} 2533 2534/// ComputeConstraintToUse - Determines the constraint code and constraint 2535/// type to use for the specific AsmOperandInfo, setting 2536/// OpInfo.ConstraintCode and OpInfo.ConstraintType. 2537void TargetLowering::ComputeConstraintToUse(AsmOperandInfo &OpInfo, 2538 SDValue Op, 2539 bool hasMemory, 2540 SelectionDAG *DAG) const { 2541 assert(!OpInfo.Codes.empty() && "Must have at least one constraint"); 2542 2543 // Single-letter constraints ('r') are very common. 2544 if (OpInfo.Codes.size() == 1) { 2545 OpInfo.ConstraintCode = OpInfo.Codes[0]; 2546 OpInfo.ConstraintType = getConstraintType(OpInfo.ConstraintCode); 2547 } else { 2548 ChooseConstraint(OpInfo, hasMemory, *this, Op, DAG); 2549 } 2550 2551 // 'X' matches anything. 2552 if (OpInfo.ConstraintCode == "X" && OpInfo.CallOperandVal) { 2553 // Labels and constants are handled elsewhere ('X' is the only thing 2554 // that matches labels). For Functions, the type here is the type of 2555 // the result, which is not what we want to look at; leave them alone. 2556 Value *v = OpInfo.CallOperandVal; 2557 if (isa<BasicBlock>(v) || isa<ConstantInt>(v) || isa<Function>(v)) { 2558 OpInfo.CallOperandVal = v; 2559 return; 2560 } 2561 2562 // Otherwise, try to resolve it to something we know about by looking at 2563 // the actual operand type. 2564 if (const char *Repl = LowerXConstraint(OpInfo.ConstraintVT)) { 2565 OpInfo.ConstraintCode = Repl; 2566 OpInfo.ConstraintType = getConstraintType(OpInfo.ConstraintCode); 2567 } 2568 } 2569} 2570 2571//===----------------------------------------------------------------------===// 2572// Loop Strength Reduction hooks 2573//===----------------------------------------------------------------------===// 2574 2575/// isLegalAddressingMode - Return true if the addressing mode represented 2576/// by AM is legal for this target, for a load/store of the specified type. 2577bool TargetLowering::isLegalAddressingMode(const AddrMode &AM, 2578 const Type *Ty) const { 2579 // The default implementation of this implements a conservative RISCy, r+r and 2580 // r+i addr mode. 2581 2582 // Allows a sign-extended 16-bit immediate field. 2583 if (AM.BaseOffs <= -(1LL << 16) || AM.BaseOffs >= (1LL << 16)-1) 2584 return false; 2585 2586 // No global is ever allowed as a base. 2587 if (AM.BaseGV) 2588 return false; 2589 2590 // Only support r+r, 2591 switch (AM.Scale) { 2592 case 0: // "r+i" or just "i", depending on HasBaseReg. 2593 break; 2594 case 1: 2595 if (AM.HasBaseReg && AM.BaseOffs) // "r+r+i" is not allowed. 2596 return false; 2597 // Otherwise we have r+r or r+i. 2598 break; 2599 case 2: 2600 if (AM.HasBaseReg || AM.BaseOffs) // 2*r+r or 2*r+i is not allowed. 2601 return false; 2602 // Allow 2*r as r+r. 2603 break; 2604 } 2605 2606 return true; 2607} 2608 2609/// BuildSDIVSequence - Given an ISD::SDIV node expressing a divide by constant, 2610/// return a DAG expression to select that will generate the same value by 2611/// multiplying by a magic number. See: 2612/// <http://the.wall.riscom.net/books/proc/ppc/cwg/code2.html> 2613SDValue TargetLowering::BuildSDIV(SDNode *N, SelectionDAG &DAG, 2614 std::vector<SDNode*>* Created) const { 2615 EVT VT = N->getValueType(0); 2616 DebugLoc dl= N->getDebugLoc(); 2617 2618 // Check to see if we can do this. 2619 // FIXME: We should be more aggressive here. 2620 if (!isTypeLegal(VT)) 2621 return SDValue(); 2622 2623 APInt d = cast<ConstantSDNode>(N->getOperand(1))->getAPIntValue(); 2624 APInt::ms magics = d.magic(); 2625 2626 // Multiply the numerator (operand 0) by the magic value 2627 // FIXME: We should support doing a MUL in a wider type 2628 SDValue Q; 2629 if (isOperationLegalOrCustom(ISD::MULHS, VT)) 2630 Q = DAG.getNode(ISD::MULHS, dl, VT, N->getOperand(0), 2631 DAG.getConstant(magics.m, VT)); 2632 else if (isOperationLegalOrCustom(ISD::SMUL_LOHI, VT)) 2633 Q = SDValue(DAG.getNode(ISD::SMUL_LOHI, dl, DAG.getVTList(VT, VT), 2634 N->getOperand(0), 2635 DAG.getConstant(magics.m, VT)).getNode(), 1); 2636 else 2637 return SDValue(); // No mulhs or equvialent 2638 // If d > 0 and m < 0, add the numerator 2639 if (d.isStrictlyPositive() && magics.m.isNegative()) { 2640 Q = DAG.getNode(ISD::ADD, dl, VT, Q, N->getOperand(0)); 2641 if (Created) 2642 Created->push_back(Q.getNode()); 2643 } 2644 // If d < 0 and m > 0, subtract the numerator. 2645 if (d.isNegative() && magics.m.isStrictlyPositive()) { 2646 Q = DAG.getNode(ISD::SUB, dl, VT, Q, N->getOperand(0)); 2647 if (Created) 2648 Created->push_back(Q.getNode()); 2649 } 2650 // Shift right algebraic if shift value is nonzero 2651 if (magics.s > 0) { 2652 Q = DAG.getNode(ISD::SRA, dl, VT, Q, 2653 DAG.getConstant(magics.s, getShiftAmountTy())); 2654 if (Created) 2655 Created->push_back(Q.getNode()); 2656 } 2657 // Extract the sign bit and add it to the quotient 2658 SDValue T = 2659 DAG.getNode(ISD::SRL, dl, VT, Q, DAG.getConstant(VT.getSizeInBits()-1, 2660 getShiftAmountTy())); 2661 if (Created) 2662 Created->push_back(T.getNode()); 2663 return DAG.getNode(ISD::ADD, dl, VT, Q, T); 2664} 2665 2666/// BuildUDIVSequence - Given an ISD::UDIV node expressing a divide by constant, 2667/// return a DAG expression to select that will generate the same value by 2668/// multiplying by a magic number. See: 2669/// <http://the.wall.riscom.net/books/proc/ppc/cwg/code2.html> 2670SDValue TargetLowering::BuildUDIV(SDNode *N, SelectionDAG &DAG, 2671 std::vector<SDNode*>* Created) const { 2672 EVT VT = N->getValueType(0); 2673 DebugLoc dl = N->getDebugLoc(); 2674 2675 // Check to see if we can do this. 2676 // FIXME: We should be more aggressive here. 2677 if (!isTypeLegal(VT)) 2678 return SDValue(); 2679 2680 // FIXME: We should use a narrower constant when the upper 2681 // bits are known to be zero. 2682 ConstantSDNode *N1C = cast<ConstantSDNode>(N->getOperand(1)); 2683 APInt::mu magics = N1C->getAPIntValue().magicu(); 2684 2685 // Multiply the numerator (operand 0) by the magic value 2686 // FIXME: We should support doing a MUL in a wider type 2687 SDValue Q; 2688 if (isOperationLegalOrCustom(ISD::MULHU, VT)) 2689 Q = DAG.getNode(ISD::MULHU, dl, VT, N->getOperand(0), 2690 DAG.getConstant(magics.m, VT)); 2691 else if (isOperationLegalOrCustom(ISD::UMUL_LOHI, VT)) 2692 Q = SDValue(DAG.getNode(ISD::UMUL_LOHI, dl, DAG.getVTList(VT, VT), 2693 N->getOperand(0), 2694 DAG.getConstant(magics.m, VT)).getNode(), 1); 2695 else 2696 return SDValue(); // No mulhu or equvialent 2697 if (Created) 2698 Created->push_back(Q.getNode()); 2699 2700 if (magics.a == 0) { 2701 assert(magics.s < N1C->getAPIntValue().getBitWidth() && 2702 "We shouldn't generate an undefined shift!"); 2703 return DAG.getNode(ISD::SRL, dl, VT, Q, 2704 DAG.getConstant(magics.s, getShiftAmountTy())); 2705 } else { 2706 SDValue NPQ = DAG.getNode(ISD::SUB, dl, VT, N->getOperand(0), Q); 2707 if (Created) 2708 Created->push_back(NPQ.getNode()); 2709 NPQ = DAG.getNode(ISD::SRL, dl, VT, NPQ, 2710 DAG.getConstant(1, getShiftAmountTy())); 2711 if (Created) 2712 Created->push_back(NPQ.getNode()); 2713 NPQ = DAG.getNode(ISD::ADD, dl, VT, NPQ, Q); 2714 if (Created) 2715 Created->push_back(NPQ.getNode()); 2716 return DAG.getNode(ISD::SRL, dl, VT, NPQ, 2717 DAG.getConstant(magics.s-1, getShiftAmountTy())); 2718 } 2719} 2720