| OLD | NEW |
| (Empty) | |
| 1 // Copyright 2013 the V8 project authors. All rights reserved. |
| 2 // Redistribution and use in source and binary forms, with or without |
| 3 // modification, are permitted provided that the following conditions are |
| 4 // met: |
| 5 // |
| 6 // * Redistributions of source code must retain the above copyright |
| 7 // notice, this list of conditions and the following disclaimer. |
| 8 // * Redistributions in binary form must reproduce the above |
| 9 // copyright notice, this list of conditions and the following |
| 10 // disclaimer in the documentation and/or other materials provided |
| 11 // with the distribution. |
| 12 // * Neither the name of Google Inc. nor the names of its |
| 13 // contributors may be used to endorse or promote products derived |
| 14 // from this software without specific prior written permission. |
| 15 // |
| 16 // THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS |
| 17 // "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT |
| 18 // LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR |
| 19 // A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT |
| 20 // OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, |
| 21 // SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT |
| 22 // LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, |
| 23 // DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY |
| 24 // THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT |
| 25 // (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE |
| 26 // OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. |
| 27 |
| 28 #ifndef V8_A64_ASSEMBLER_A64_H_ |
| 29 #define V8_A64_ASSEMBLER_A64_H_ |
| 30 |
| 31 #include <list> |
| 32 |
| 33 #include "globals.h" |
| 34 #include "utils.h" |
| 35 #include "assembler.h" |
| 36 #include "serialize.h" |
| 37 #include "a64/instructions-a64.h" |
| 38 #include "a64/cpu-a64.h" |
| 39 |
| 40 |
| 41 namespace v8 { |
| 42 namespace internal { |
| 43 |
| 44 |
| 45 // ----------------------------------------------------------------------------- |
| 46 // Registers. |
| 47 #define REGISTER_CODE_LIST(R) \ |
| 48 R(0) R(1) R(2) R(3) R(4) R(5) R(6) R(7) \ |
| 49 R(8) R(9) R(10) R(11) R(12) R(13) R(14) R(15) \ |
| 50 R(16) R(17) R(18) R(19) R(20) R(21) R(22) R(23) \ |
| 51 R(24) R(25) R(26) R(27) R(28) R(29) R(30) R(31) |
| 52 |
| 53 |
| 54 static const int kRegListSizeInBits = sizeof(RegList) * kBitsPerByte; |
| 55 |
| 56 |
| 57 // Some CPURegister methods can return Register and FPRegister types, so we |
| 58 // need to declare them in advance. |
| 59 class Register; |
| 60 class FPRegister; |
| 61 |
| 62 |
| 63 class CPURegister { |
| 64 public: |
| 65 enum RegisterType { |
| 66 // The kInvalid value is used to detect uninitialized static instances, |
| 67 // which are always zero-initialized before any constructors are called. |
| 68 kInvalid = 0, |
| 69 kRegister, |
| 70 kFPRegister, |
| 71 kNoRegister |
| 72 }; |
| 73 |
| 74 inline CPURegister() : code_(0), size_(0), type_(kNoRegister) { |
| 75 ASSERT(!IsValid()); |
| 76 ASSERT(IsNone()); |
| 77 } |
| 78 |
| 79 inline CPURegister(unsigned code, unsigned size, RegisterType type) |
| 80 : code_(code), size_(size), type_(type) { |
| 81 ASSERT(IsValidOrNone()); |
| 82 } |
| 83 |
| 84 // This copy constructor is used by the Register and FPRegister classes for |
| 85 // converting a CPURegister into a more specialized type. It is necessary |
| 86 // because (FP)Register cannot directly access other.size_ and suchlike and |
| 87 // the accessor methods don't allow NoCPUReg. Unfortunately, it cannot be |
| 88 // explicit because otherwise the simple (FP)Register->CPURegister casts do |
| 89 // not work. |
| 90 inline CPURegister(const CPURegister& other) |
| 91 : code_(other.code_), size_(other.size_), type_(other.type_) { |
| 92 ASSERT(IsValidOrNone()); |
| 93 } |
| 94 |
| 95 unsigned code() const; |
| 96 RegisterType type() const; |
| 97 RegList Bit() const; |
| 98 unsigned SizeInBits() const; |
| 99 int SizeInBytes() const; |
| 100 bool Is32Bits() const; |
| 101 bool Is64Bits() const; |
| 102 bool IsValid() const; |
| 103 bool IsValidRegister() const; |
| 104 bool IsValidFPRegister() const; |
| 105 bool IsNone() const; |
| 106 bool Is(const CPURegister& other) const; |
| 107 bool is(const CPURegister& other) const; // V8 compatibility. |
| 108 |
| 109 bool IsZero() const; |
| 110 bool IsSP() const; |
| 111 |
| 112 bool IsRegister() const; |
| 113 bool IsFPRegister() const; |
| 114 |
| 115 const Register& X(); |
| 116 const Register& W(); |
| 117 const FPRegister& D(); |
| 118 const FPRegister& S(); |
| 119 |
| 120 bool IsSameSizeAndType(const CPURegister& other) const; |
| 121 |
| 122 protected: |
| 123 unsigned code_; |
| 124 unsigned size_; |
| 125 RegisterType type_; |
| 126 |
| 127 private: |
| 128 bool IsValidOrNone() const; |
| 129 }; |
| 130 |
| 131 |
| 132 class Register : public CPURegister { |
| 133 public: |
| 134 inline Register() : CPURegister() {} |
| 135 inline explicit Register(const CPURegister& other) : CPURegister(other) { |
| 136 ASSERT(IsValidRegister() || IsNone()); |
| 137 } |
| 138 inline Register(unsigned code, unsigned size) |
| 139 : CPURegister(code, size, kRegister) {} |
| 140 |
| 141 bool IsValid() const { |
| 142 ASSERT(IsRegister() || IsNone()); |
| 143 return IsValidRegister(); |
| 144 } |
| 145 |
| 146 static const Register& XRegFromCode(unsigned code); |
| 147 static const Register& WRegFromCode(unsigned code); |
| 148 |
| 149 // Start of V8 compatibility section --------------------- |
| 150 // These memebers are necessary for compilation. |
| 151 // A few of them may be unused for now. |
| 152 |
| 153 static const int kNumRegisters = kNumberOfRegisters; |
| 154 static int NumRegisters() { return kNumRegisters; } |
| 155 |
| 156 // We allow crankshaft to use the following registers: |
| 157 // - x0 to x15 |
| 158 // - x18 to x24 |
| 159 // |
| 160 // TODO(all): Register x25 is currently free and could be available for |
| 161 // crankshaft, but we don't use it as we might use it as a per function |
| 162 // literal pool pointer in the future. |
| 163 // |
| 164 // We split allocatable registers in two ranges called "low range" and |
| 165 // "high range" which are defined as follows. |
| 166 static const unsigned kAllocatableLowRangeBegin = 0; |
| 167 static const unsigned kAllocatableLowRangeEnd = 15; |
| 168 static const unsigned kAllocatableHighRangeBegin = 18; |
| 169 static const unsigned kAllocatableHighRangeEnd = 24; |
| 170 |
| 171 static const int kAllocatableRangeGapSize = |
| 172 (kAllocatableHighRangeBegin - kAllocatableLowRangeEnd) - 1; |
| 173 |
| 174 static const int kMaxNumAllocatableRegisters = |
| 175 (kAllocatableLowRangeEnd - kAllocatableLowRangeBegin + 1) + |
| 176 (kAllocatableHighRangeEnd - kAllocatableHighRangeBegin + 1); |
| 177 static int NumAllocatableRegisters() { return kMaxNumAllocatableRegisters; } |
| 178 |
| 179 // Return true if the register is one that crankshaft can allocate. |
| 180 bool IsAllocatable() const { |
| 181 return (code_ <= kAllocatableLowRangeEnd) || |
| 182 ((code_ >= kAllocatableHighRangeBegin) && |
| 183 (code_ <= kAllocatableHighRangeEnd)); |
| 184 } |
| 185 |
| 186 static Register FromAllocationIndex(unsigned index) { |
| 187 ASSERT(index < static_cast<unsigned>(NumAllocatableRegisters())); |
| 188 return (index <= kAllocatableLowRangeEnd) |
| 189 ? from_code(index) |
| 190 : from_code(index + kAllocatableRangeGapSize); |
| 191 } |
| 192 |
| 193 static const char* AllocationIndexToString(int index) { |
| 194 ASSERT((index >= 0) && (index < NumAllocatableRegisters())); |
| 195 ASSERT((kAllocatableLowRangeBegin == 0) && |
| 196 (kAllocatableLowRangeEnd == 15) && |
| 197 (kAllocatableHighRangeBegin == 18) && |
| 198 (kAllocatableHighRangeEnd == 24)); |
| 199 const char* const names[] = { |
| 200 "x0", "x1", "x2", "x3", "x4", |
| 201 "x5", "x6", "x7", "x8", "x9", |
| 202 "x10", "x11", "x12", "x13", "x14", |
| 203 "x15", "x18", "x19", "x20", "x21", |
| 204 "x22", "x23", "x24", |
| 205 }; |
| 206 return names[index]; |
| 207 } |
| 208 |
| 209 static int ToAllocationIndex(Register reg) { |
| 210 ASSERT(reg.IsAllocatable()); |
| 211 unsigned code = reg.code(); |
| 212 return (code <= kAllocatableLowRangeEnd) |
| 213 ? code |
| 214 : code - kAllocatableRangeGapSize; |
| 215 } |
| 216 |
| 217 static Register from_code(int code) { |
| 218 // Always return an X register. |
| 219 Register r(code, kXRegSize); |
| 220 return r; |
| 221 } |
| 222 |
| 223 // End of V8 compatibility section ----------------------- |
| 224 |
| 225 private: |
| 226 static const Register xregisters[]; |
| 227 static const Register wregisters[]; |
| 228 }; |
| 229 |
| 230 |
| 231 class FPRegister : public CPURegister { |
| 232 public: |
| 233 inline FPRegister() : CPURegister() {} |
| 234 inline explicit FPRegister(const CPURegister& other) : CPURegister(other) { |
| 235 ASSERT(IsValidFPRegister() || IsNone()); |
| 236 } |
| 237 inline FPRegister(unsigned code, unsigned size) |
| 238 : CPURegister(code, size, kFPRegister) {} |
| 239 |
| 240 bool IsValid() const { |
| 241 ASSERT(IsFPRegister() || IsNone()); |
| 242 return IsValidFPRegister(); |
| 243 } |
| 244 |
| 245 static const FPRegister& SRegFromCode(unsigned code); |
| 246 static const FPRegister& DRegFromCode(unsigned code); |
| 247 |
| 248 // Start of V8 compatibility section --------------------- |
| 249 static const int kMaxNumRegisters = kNumberOfFPRegisters; |
| 250 |
| 251 // Crankshaft can use all the FP registers except: |
| 252 // - d29 which is used in crankshaft as a double scratch register |
| 253 // - d30 which is used to keep the 0 double value |
| 254 // - d31 which is used in the MacroAssembler as a double scratch register |
| 255 static const int kNumReservedRegisters = 3; |
| 256 static const int kMaxNumAllocatableRegisters = |
| 257 kNumberOfFPRegisters - kNumReservedRegisters; |
| 258 static int NumAllocatableRegisters() { return kMaxNumAllocatableRegisters; } |
| 259 |
| 260 static FPRegister FromAllocationIndex(int index) { |
| 261 ASSERT((index >= 0) && (index < NumAllocatableRegisters())); |
| 262 return from_code(index); |
| 263 } |
| 264 |
| 265 static const char* AllocationIndexToString(int index) { |
| 266 ASSERT((index >= 0) && (index < NumAllocatableRegisters())); |
| 267 const char* const names[] = { |
| 268 "d0", "d1", "d2", "d3", "d4", "d5", "d6", "d7", |
| 269 "d8", "d9", "d10", "d11", "d12", "d13", "d14", "d15", |
| 270 "d16", "d17", "d18", "d19", "d20", "d21", "d22", "d23", |
| 271 "d24", "d25", "d26", "d27", "d28", |
| 272 }; |
| 273 return names[index]; |
| 274 } |
| 275 |
| 276 static int ToAllocationIndex(FPRegister reg) { |
| 277 int code = reg.code(); |
| 278 ASSERT(code < NumAllocatableRegisters()); |
| 279 return code; |
| 280 } |
| 281 |
| 282 static FPRegister from_code(int code) { |
| 283 // Always return a D register. |
| 284 FPRegister r(code, kDRegSize); |
| 285 return r; |
| 286 } |
| 287 // End of V8 compatibility section ----------------------- |
| 288 |
| 289 private: |
| 290 static const FPRegister sregisters[]; |
| 291 static const FPRegister dregisters[]; |
| 292 }; |
| 293 |
| 294 |
| 295 // No*Reg is used to indicate an unused argument, or an error case. Note that |
| 296 // these all compare equal (using the Is() method). The Register and FPRegister |
| 297 // variants are provided for convenience. |
| 298 const Register NoReg; |
| 299 const FPRegister NoFPReg; |
| 300 const CPURegister NoCPUReg; |
| 301 |
| 302 const Register no_reg; // v8 compatibility. |
| 303 |
| 304 |
| 305 #define DEFINE_REGISTERS(N) \ |
| 306 const Register w##N(N, kWRegSize); \ |
| 307 const Register x##N(N, kXRegSize); |
| 308 REGISTER_CODE_LIST(DEFINE_REGISTERS) |
| 309 #undef DEFINE_REGISTERS |
| 310 const Register wcsp(kSPRegInternalCode, kWRegSize); |
| 311 const Register csp(kSPRegInternalCode, kXRegSize); |
| 312 |
| 313 |
| 314 #define DEFINE_FPREGISTERS(N) \ |
| 315 const FPRegister s##N(N, kSRegSize); \ |
| 316 const FPRegister d##N(N, kDRegSize); |
| 317 REGISTER_CODE_LIST(DEFINE_FPREGISTERS) |
| 318 #undef DEFINE_FPREGISTERS |
| 319 |
| 320 |
| 321 // Registers aliases. |
| 322 const Register ip0 = x16; |
| 323 const Register ip1 = x17; |
| 324 const Register wip0 = w16; |
| 325 const Register wip1 = w17; |
| 326 // Root register. |
| 327 const Register root = x26; |
| 328 const Register rr = root; |
| 329 // Context pointer register. |
| 330 const Register cp = x27; |
| 331 // We use a register as a JS stack pointer to overcome the restriction on the |
| 332 // architectural SP alignment. |
| 333 // We chose x28 because it is contiguous with the other specific purpose |
| 334 // registers. |
| 335 STATIC_ASSERT(kJSSPCode == 28); |
| 336 const Register jssp = x28; |
| 337 const Register wjssp = w28; |
| 338 const Register fp = x29; |
| 339 const Register lr = x30; |
| 340 const Register xzr = x31; |
| 341 const Register wzr = w31; |
| 342 |
| 343 // Crankshaft double scratch register. |
| 344 const FPRegister crankshaft_fp_scratch = d29; |
| 345 // MacroAssembler double scratch register. |
| 346 const FPRegister fp_scratch = d31; |
| 347 |
| 348 // AreAliased returns true if any of the named registers overlap. Arguments set |
| 349 // to NoReg are ignored. The system stack pointer may be specified. |
| 350 bool AreAliased(const CPURegister& reg1, |
| 351 const CPURegister& reg2, |
| 352 const CPURegister& reg3 = NoReg, |
| 353 const CPURegister& reg4 = NoReg, |
| 354 const CPURegister& reg5 = NoReg, |
| 355 const CPURegister& reg6 = NoReg, |
| 356 const CPURegister& reg7 = NoReg, |
| 357 const CPURegister& reg8 = NoReg); |
| 358 |
| 359 // AreSameSizeAndType returns true if all of the specified registers have the |
| 360 // same size, and are of the same type. The system stack pointer may be |
| 361 // specified. Arguments set to NoReg are ignored, as are any subsequent |
| 362 // arguments. At least one argument (reg1) must be valid (not NoCPUReg). |
| 363 bool AreSameSizeAndType(const CPURegister& reg1, |
| 364 const CPURegister& reg2, |
| 365 const CPURegister& reg3 = NoCPUReg, |
| 366 const CPURegister& reg4 = NoCPUReg, |
| 367 const CPURegister& reg5 = NoCPUReg, |
| 368 const CPURegister& reg6 = NoCPUReg, |
| 369 const CPURegister& reg7 = NoCPUReg, |
| 370 const CPURegister& reg8 = NoCPUReg); |
| 371 |
| 372 |
| 373 typedef FPRegister DoubleRegister; |
| 374 |
| 375 |
| 376 // ----------------------------------------------------------------------------- |
| 377 // Lists of registers. |
| 378 class CPURegList { |
| 379 public: |
| 380 inline explicit CPURegList(CPURegister reg1, |
| 381 CPURegister reg2 = NoCPUReg, |
| 382 CPURegister reg3 = NoCPUReg, |
| 383 CPURegister reg4 = NoCPUReg) |
| 384 : list_(reg1.Bit() | reg2.Bit() | reg3.Bit() | reg4.Bit()), |
| 385 size_(reg1.SizeInBits()), type_(reg1.type()) { |
| 386 ASSERT(AreSameSizeAndType(reg1, reg2, reg3, reg4)); |
| 387 ASSERT(IsValid()); |
| 388 } |
| 389 |
| 390 inline CPURegList(CPURegister::RegisterType type, unsigned size, RegList list) |
| 391 : list_(list), size_(size), type_(type) { |
| 392 ASSERT(IsValid()); |
| 393 } |
| 394 |
| 395 inline CPURegList(CPURegister::RegisterType type, unsigned size, |
| 396 unsigned first_reg, unsigned last_reg) |
| 397 : size_(size), type_(type) { |
| 398 ASSERT(((type == CPURegister::kRegister) && |
| 399 (last_reg < kNumberOfRegisters)) || |
| 400 ((type == CPURegister::kFPRegister) && |
| 401 (last_reg < kNumberOfFPRegisters))); |
| 402 ASSERT(last_reg >= first_reg); |
| 403 list_ = (1UL << (last_reg + 1)) - 1; |
| 404 list_ &= ~((1UL << first_reg) - 1); |
| 405 ASSERT(IsValid()); |
| 406 } |
| 407 |
| 408 inline CPURegister::RegisterType type() const { |
| 409 ASSERT(IsValid()); |
| 410 return type_; |
| 411 } |
| 412 |
| 413 inline RegList list() const { |
| 414 ASSERT(IsValid()); |
| 415 return list_; |
| 416 } |
| 417 |
| 418 // Combine another CPURegList into this one. Registers that already exist in |
| 419 // this list are left unchanged. The type and size of the registers in the |
| 420 // 'other' list must match those in this list. |
| 421 inline void Combine(const CPURegList& other); |
| 422 |
| 423 // Remove every register in the other CPURegList from this one. Registers that |
| 424 // do not exist in this list are ignored. The type and size of the registers |
| 425 // in the 'other' list must match those in this list. |
| 426 inline void Remove(const CPURegList& other); |
| 427 |
| 428 // Variants of Combine and Remove which take a single register. |
| 429 inline void Combine(const CPURegister& other); |
| 430 inline void Remove(const CPURegister& other); |
| 431 |
| 432 // Variants of Combine and Remove which take a single register by its code; |
| 433 // the type and size of the register is inferred from this list. |
| 434 inline void Combine(int code); |
| 435 inline void Remove(int code); |
| 436 |
| 437 // Remove all callee-saved registers from the list. This can be useful when |
| 438 // preparing registers for an AAPCS64 function call, for example. |
| 439 void RemoveCalleeSaved(); |
| 440 |
| 441 CPURegister PopLowestIndex(); |
| 442 CPURegister PopHighestIndex(); |
| 443 |
| 444 // AAPCS64 callee-saved registers. |
| 445 static CPURegList GetCalleeSaved(unsigned size = kXRegSize); |
| 446 static CPURegList GetCalleeSavedFP(unsigned size = kDRegSize); |
| 447 |
| 448 // AAPCS64 caller-saved registers. Note that this includes lr. |
| 449 static CPURegList GetCallerSaved(unsigned size = kXRegSize); |
| 450 static CPURegList GetCallerSavedFP(unsigned size = kDRegSize); |
| 451 |
| 452 // Registers saved as safepoints. |
| 453 static CPURegList GetSafepointSavedRegisters(); |
| 454 |
| 455 inline bool IsEmpty() const { |
| 456 ASSERT(IsValid()); |
| 457 return list_ == 0; |
| 458 } |
| 459 |
| 460 inline bool IncludesAliasOf(const CPURegister& other) const { |
| 461 ASSERT(IsValid()); |
| 462 return (type_ == other.type()) && (other.Bit() & list_); |
| 463 } |
| 464 |
| 465 inline int Count() const { |
| 466 ASSERT(IsValid()); |
| 467 return CountSetBits(list_, kRegListSizeInBits); |
| 468 } |
| 469 |
| 470 inline unsigned RegisterSizeInBits() const { |
| 471 ASSERT(IsValid()); |
| 472 return size_; |
| 473 } |
| 474 |
| 475 inline unsigned RegisterSizeInBytes() const { |
| 476 int size_in_bits = RegisterSizeInBits(); |
| 477 ASSERT((size_in_bits % kBitsPerByte) == 0); |
| 478 return size_in_bits / kBitsPerByte; |
| 479 } |
| 480 |
| 481 private: |
| 482 RegList list_; |
| 483 unsigned size_; |
| 484 CPURegister::RegisterType type_; |
| 485 |
| 486 bool IsValid() const { |
| 487 if ((type_ == CPURegister::kRegister) || |
| 488 (type_ == CPURegister::kFPRegister)) { |
| 489 bool is_valid = true; |
| 490 // Try to create a CPURegister for each element in the list. |
| 491 for (int i = 0; i < kRegListSizeInBits; i++) { |
| 492 if (((list_ >> i) & 1) != 0) { |
| 493 is_valid &= CPURegister(i, size_, type_).IsValid(); |
| 494 } |
| 495 } |
| 496 return is_valid; |
| 497 } else if (type_ == CPURegister::kNoRegister) { |
| 498 // The kNoRegister type is valid only for empty lists. |
| 499 // We can't use IsEmpty here because that asserts IsValid(). |
| 500 return list_ == 0; |
| 501 } else { |
| 502 return false; |
| 503 } |
| 504 } |
| 505 }; |
| 506 |
| 507 |
| 508 // AAPCS64 callee-saved registers. |
| 509 extern const CPURegList kCalleeSaved; |
| 510 extern const CPURegList kCalleeSavedFP; |
| 511 |
| 512 |
| 513 // AAPCS64 caller-saved registers. Note that this includes lr. |
| 514 extern const CPURegList kCallerSaved; |
| 515 extern const CPURegList kCallerSavedFP; |
| 516 |
| 517 |
| 518 // ----------------------------------------------------------------------------- |
| 519 // Operands. |
| 520 const int kSmiShift = kSmiTagSize + kSmiShiftSize; |
| 521 const uint64_t kSmiShiftMask = (1UL << kSmiShift) - 1; |
| 522 |
| 523 // Represents an operand in a machine instruction. |
| 524 class Operand { |
| 525 // TODO(all): If necessary, study more in details which methods |
| 526 // TODO(all): should be inlined or not. |
| 527 public: |
| 528 // #<immediate> |
| 529 // where <immediate> is int64_t. |
| 530 // GCC complains about ambiguous aliasing if we don't explicitly declare the |
| 531 // variants. |
| 532 // The simple literal-value wrappers are allowed to be implicit constructors |
| 533 // because Operand is a wrapper class that doesn't normally perform any type |
| 534 // conversion. |
| 535 // TODO(jbramley): Check that my interpretation of this rule is correct. |
| 536 inline Operand(int64_t immediate, |
| 537 RelocInfo::Mode rmode = RelocInfo::NONE64); // NOLINT(runtime/explicit) |
| 538 inline Operand(uint64_t immediate, |
| 539 RelocInfo::Mode rmode = RelocInfo::NONE64); // NOLINT(runtime/explicit) |
| 540 inline Operand(int32_t immediate, |
| 541 RelocInfo::Mode rmode = RelocInfo::NONE32); // NOLINT(runtime/explicit) |
| 542 inline Operand(uint32_t immediate, |
| 543 RelocInfo::Mode rmode = RelocInfo::NONE32); // NOLINT(runtime/explicit) |
| 544 |
| 545 |
| 546 // rm, {<shift> #<shift_amount>} |
| 547 // where <shift> is one of {LSL, LSR, ASR, ROR}. |
| 548 // <shift_amount> is uint6_t. |
| 549 // This is allowed to be an implicit constructor because Operand is |
| 550 // a wrapper class that doesn't normally perform any type conversion. |
| 551 // TODO(jbramley): Check that my interpretation of this rule is correct. |
| 552 inline Operand(Register reg, |
| 553 Shift shift = LSL, |
| 554 unsigned shift_amount = 0); // NOLINT(runtime/explicit) |
| 555 |
| 556 // rm, {<extend> {#<shift_amount>}} |
| 557 // where <extend> is one of {UXTB, UXTH, UXTW, UXTX, SXTB, SXTH, SXTW, SXTX}. |
| 558 // <shift_amount> is uint2_t. |
| 559 inline explicit Operand(Register reg, |
| 560 Extend extend, |
| 561 unsigned shift_amount = 0); |
| 562 |
| 563 inline explicit Operand(Smi* value); |
| 564 explicit Operand(const ExternalReference& f); |
| 565 explicit Operand(Handle<Object> handle); |
| 566 |
| 567 inline bool IsImmediate() const; |
| 568 inline bool IsShiftedRegister() const; |
| 569 inline bool IsExtendedRegister() const; |
| 570 inline bool IsZero() const; |
| 571 |
| 572 // This returns an LSL shift (<= 4) operand as an equivalent extend operand, |
| 573 // which helps in the encoding of instructions that use the stack pointer. |
| 574 inline Operand ToExtendedRegister() const; |
| 575 |
| 576 inline int64_t immediate() const; |
| 577 inline Register reg() const; |
| 578 inline Shift shift() const; |
| 579 inline Extend extend() const; |
| 580 inline unsigned shift_amount() const; |
| 581 |
| 582 // Relocation information. |
| 583 inline RelocInfo::Mode rmode() const { return rmode_; } |
| 584 inline void set_rmode(RelocInfo::Mode rmode) { rmode_ = rmode; } |
| 585 bool NeedsRelocation() const; |
| 586 |
| 587 // Helpers |
| 588 inline static Operand UntagSmi(Register smi); |
| 589 inline static Operand UntagSmiAndScale(Register smi, int scale); |
| 590 |
| 591 private: |
| 592 int64_t immediate_; |
| 593 Register reg_; |
| 594 Shift shift_; |
| 595 Extend extend_; |
| 596 unsigned shift_amount_; |
| 597 RelocInfo::Mode rmode_; |
| 598 }; |
| 599 |
| 600 |
| 601 // MemOperand represents a memory operand in a load or store instruction. |
| 602 class MemOperand { |
| 603 public: |
| 604 inline explicit MemOperand(Register base, |
| 605 ptrdiff_t offset = 0, |
| 606 AddrMode addrmode = Offset); |
| 607 inline explicit MemOperand(Register base, |
| 608 Register regoffset, |
| 609 Shift shift = LSL, |
| 610 unsigned shift_amount = 0); |
| 611 inline explicit MemOperand(Register base, |
| 612 Register regoffset, |
| 613 Extend extend, |
| 614 unsigned shift_amount = 0); |
| 615 inline explicit MemOperand(Register base, |
| 616 const Operand& offset, |
| 617 AddrMode addrmode = Offset); |
| 618 |
| 619 inline const Register& base() const { return base_; } |
| 620 inline const Register& regoffset() const { return regoffset_; } |
| 621 inline ptrdiff_t offset() const { return offset_; } |
| 622 inline AddrMode addrmode() const { return addrmode_; } |
| 623 inline Shift shift() const { return shift_; } |
| 624 inline Extend extend() const { return extend_; } |
| 625 inline unsigned shift_amount() const { return shift_amount_; } |
| 626 inline bool IsImmediateOffset() const; |
| 627 inline bool IsRegisterOffset() const; |
| 628 inline bool IsPreIndex() const; |
| 629 inline bool IsPostIndex() const; |
| 630 |
| 631 // For offset modes, return the offset as an Operand. This helper cannot |
| 632 // handle indexed modes. |
| 633 inline Operand OffsetAsOperand() const; |
| 634 |
| 635 private: |
| 636 Register base_; |
| 637 Register regoffset_; |
| 638 ptrdiff_t offset_; |
| 639 AddrMode addrmode_; |
| 640 Shift shift_; |
| 641 Extend extend_; |
| 642 unsigned shift_amount_; |
| 643 }; |
| 644 |
| 645 |
| 646 // ----------------------------------------------------------------------------- |
| 647 // Assembler. |
| 648 |
| 649 class Assembler : public AssemblerBase { |
| 650 public: |
| 651 // Create an assembler. Instructions and relocation information are emitted |
| 652 // into a buffer, with the instructions starting from the beginning and the |
| 653 // relocation information starting from the end of the buffer. See CodeDesc |
| 654 // for a detailed comment on the layout (globals.h). |
| 655 // |
| 656 // If the provided buffer is NULL, the assembler allocates and grows its own |
| 657 // buffer, and buffer_size determines the initial buffer size. The buffer is |
| 658 // owned by the assembler and deallocated upon destruction of the assembler. |
| 659 // |
| 660 // If the provided buffer is not NULL, the assembler uses the provided buffer |
| 661 // for code generation and assumes its size to be buffer_size. If the buffer |
| 662 // is too small, a fatal error occurs. No deallocation of the buffer is done |
| 663 // upon destruction of the assembler. |
| 664 Assembler(Isolate* arg_isolate, void* buffer, int buffer_size); |
| 665 |
| 666 // The default destructor asserts that one of the following is true: |
| 667 // * The Assembler object has not been used. |
| 668 // * Nothing has been emitted since the last Reset() call. |
| 669 // * Nothing has been emitted since the last GetCode() call. |
| 670 virtual ~Assembler(); |
| 671 |
| 672 // System functions --------------------------------------------------------- |
| 673 // Start generating code from the beginning of the buffer, discarding any code |
| 674 // and data that has already been emitted into the buffer. |
| 675 // |
| 676 // In order to avoid any accidental transfer of state, Reset ASSERTs that the |
| 677 // constant pool is not blocked. |
| 678 void Reset(); |
| 679 |
| 680 // GetCode emits any pending (non-emitted) code and fills the descriptor |
| 681 // desc. GetCode() is idempotent; it returns the same result if no other |
| 682 // Assembler functions are invoked in between GetCode() calls. |
| 683 // |
| 684 // The descriptor (desc) can be NULL. In that case, the code is finalized as |
| 685 // usual, but the descriptor is not populated. |
| 686 void GetCode(CodeDesc* desc); |
| 687 |
| 688 // Insert the smallest number of nop instructions |
| 689 // possible to align the pc offset to a multiple |
| 690 // of m. m must be a power of 2 (>= 4). |
| 691 void Align(int m); |
| 692 |
| 693 // Label -------------------------------------------------------------------- |
| 694 // Bind a label to the current pc. Note that labels can only be bound once, |
| 695 // and if labels are linked to other instructions, they _must_ be bound |
| 696 // before they go out of scope. |
| 697 void bind(Label* label); |
| 698 |
| 699 |
| 700 // RelocInfo and constant pool ---------------------------------------------- |
| 701 |
| 702 // Record relocation information for current pc_. |
| 703 void RecordRelocInfo(RelocInfo::Mode rmode, intptr_t data = 0); |
| 704 |
| 705 // Return the address in the constant pool of the code target address used by |
| 706 // the branch/call instruction at pc. |
| 707 inline static Address target_pointer_address_at(Address pc); |
| 708 |
| 709 // Read/modify the pointer in the branch/call instruction at pc. |
| 710 inline static Address target_pointer_at(Address pc); |
| 711 inline static void set_target_pointer_at(Address pc, Address target); |
| 712 |
| 713 // Read/Modify the code target address in the branch/call instruction at pc. |
| 714 inline static Address target_address_at(Address pc); |
| 715 inline static void set_target_address_at(Address pc, Address target); |
| 716 |
| 717 // Return the code target address at a call site from the return address of |
| 718 // that call in the instruction stream. |
| 719 inline static Address target_address_from_return_address(Address pc); |
| 720 |
| 721 // Given the address of the beginning of a call, return the address in the |
| 722 // instruction stream that call will return from. |
| 723 inline static Address return_address_from_call_start(Address pc); |
| 724 |
| 725 // This sets the branch destination (which is in the constant pool on ARM). |
| 726 // This is for calls and branches within generated code. |
| 727 inline static void deserialization_set_special_target_at( |
| 728 Address constant_pool_entry, Address target); |
| 729 |
| 730 // This sets the branch destination (which is in the constant pool on ARM). |
| 731 // This is for calls and branches to runtime code. |
| 732 inline static void set_external_target_at(Address constant_pool_entry, |
| 733 Address target); |
| 734 |
| 735 // All addresses in the constant pool are the same size as pointers. |
| 736 static const int kSpecialTargetSize = kPointerSize; |
| 737 |
| 738 // The sizes of the call sequences emitted by MacroAssembler::Call. |
| 739 // Wherever possible, use MacroAssembler::CallSize instead of these constants, |
| 740 // as it will choose the correct value for a given relocation mode. |
| 741 // |
| 742 // Without relocation: |
| 743 // movz ip0, #(target & 0x000000000000ffff) |
| 744 // movk ip0, #(target & 0x00000000ffff0000) |
| 745 // movk ip0, #(target & 0x0000ffff00000000) |
| 746 // movk ip0, #(target & 0xffff000000000000) |
| 747 // blr ip0 |
| 748 // |
| 749 // With relocation: |
| 750 // ldr ip0, =target |
| 751 // blr ip0 |
| 752 static const int kCallSizeWithoutRelocation = 5 * kInstructionSize; |
| 753 static const int kCallSizeWithRelocation = 2 * kInstructionSize; |
| 754 |
| 755 // Size of the generated code in bytes |
| 756 uint64_t SizeOfGeneratedCode() const { |
| 757 ASSERT((pc_ >= buffer_) && (pc_ < (buffer_ + buffer_size_))); |
| 758 return pc_ - buffer_; |
| 759 } |
| 760 |
| 761 // Return the code size generated from label to the current position. |
| 762 uint64_t SizeOfCodeGeneratedSince(const Label* label) { |
| 763 ASSERT(label->is_bound()); |
| 764 ASSERT(pc_offset() >= label->pos()); |
| 765 ASSERT(pc_offset() < buffer_size_); |
| 766 return pc_offset() - label->pos(); |
| 767 } |
| 768 |
| 769 // Check the size of the code generated since the given label. This function |
| 770 // is used primarily to work around comparisons between signed and unsigned |
| 771 // quantities, since V8 uses both. |
| 772 // TODO(jbramley): Work out what sign to use for these things and if possible, |
| 773 // change things to be consistent. |
| 774 void AssertSizeOfCodeGeneratedSince(const Label* label, ptrdiff_t size) { |
| 775 ASSERT(size >= 0); |
| 776 ASSERT(static_cast<uint64_t>(size) == SizeOfCodeGeneratedSince(label)); |
| 777 } |
| 778 |
| 779 // Return the number of instructions generated from label to the |
| 780 // current position. |
| 781 int InstructionsGeneratedSince(const Label* label) { |
| 782 return SizeOfCodeGeneratedSince(label) / kInstructionSize; |
| 783 } |
| 784 |
| 785 // TODO(all): Initialize these constants related with code patching. |
| 786 // TODO(all): Set to -1 to hopefully crash if mistakenly used. |
| 787 |
| 788 // Number of instructions generated for the return sequence in |
| 789 // FullCodeGenerator::EmitReturnSequence. |
| 790 static const int kJSRetSequenceInstructions = 7; |
| 791 // Distance between start of patched return sequence and the emitted address |
| 792 // to jump to. |
| 793 static const int kPatchReturnSequenceAddressOffset = 0; |
| 794 static const int kPatchDebugBreakSlotAddressOffset = 0; |
| 795 |
| 796 // Number of instructions necessary to be able to later patch it to a call. |
| 797 // See Debug::GenerateSlot() and BreakLocationIterator::SetDebugBreakAtSlot(). |
| 798 static const int kDebugBreakSlotInstructions = 4; |
| 799 static const int kDebugBreakSlotLength = |
| 800 kDebugBreakSlotInstructions * kInstructionSize; |
| 801 |
| 802 static const int kPatchDebugBreakSlotReturnOffset = 2 * kInstructionSize; |
| 803 |
| 804 // Prevent contant pool emission until EndBlockConstPool is called. |
| 805 // Call to this function can be nested but must be followed by an equal |
| 806 // number of call to EndBlockConstpool. |
| 807 void StartBlockConstPool(); |
| 808 |
| 809 // Resume constant pool emission. Need to be called as many time as |
| 810 // StartBlockConstPool to have an effect. |
| 811 void EndBlockConstPool(); |
| 812 |
| 813 bool is_const_pool_blocked() const; |
| 814 static bool IsConstantPoolAt(Instruction* instr); |
| 815 static int ConstantPoolSizeAt(Instruction* instr); |
| 816 // See Assembler::CheckConstPool for more info. |
| 817 void ConstantPoolMarker(uint32_t size); |
| 818 void ConstantPoolGuard(); |
| 819 |
| 820 |
| 821 // Debugging ---------------------------------------------------------------- |
| 822 PositionsRecorder* positions_recorder() { return &positions_recorder_; } |
| 823 void RecordComment(const char* msg); |
| 824 int buffer_space() const; |
| 825 |
| 826 // Mark address of the ExitJSFrame code. |
| 827 void RecordJSReturn(); |
| 828 |
| 829 // Mark address of a debug break slot. |
| 830 void RecordDebugBreakSlot(); |
| 831 |
| 832 // Record the emission of a constant pool. |
| 833 // |
| 834 // The emission of constant pool depends on the size of the code generated and |
| 835 // the number of RelocInfo recorded. |
| 836 // The Debug mechanism needs to map code offsets between two versions of a |
| 837 // function, compiled with and without debugger support (see for example |
| 838 // Debug::PrepareForBreakPoints()). |
| 839 // Compiling functions with debugger support generates additional code |
| 840 // (Debug::GenerateSlot()). This may affect the emission of the constant |
| 841 // pools and cause the version of the code with debugger support to have |
| 842 // constant pools generated in different places. |
| 843 // Recording the position and size of emitted constant pools allows to |
| 844 // correctly compute the offset mappings between the different versions of a |
| 845 // function in all situations. |
| 846 // |
| 847 // The parameter indicates the size of the constant pool (in bytes), including |
| 848 // the marker and branch over the data. |
| 849 void RecordConstPool(int size); |
| 850 |
| 851 |
| 852 // Instruction set functions ------------------------------------------------ |
| 853 |
| 854 // Branch / Jump instructions. |
| 855 // For branches offsets are scaled, i.e. they in instrcutions not in bytes. |
| 856 // Branch to register. |
| 857 void br(const Register& xn); |
| 858 |
| 859 // Branch-link to register. |
| 860 void blr(const Register& xn); |
| 861 |
| 862 // Branch to register with return hint. |
| 863 void ret(const Register& xn = lr); |
| 864 |
| 865 // Unconditional branch to label. |
| 866 void b(Label* label); |
| 867 |
| 868 // Conditional branch to label. |
| 869 void b(Label* label, Condition cond); |
| 870 |
| 871 // Unconditional branch to PC offset. |
| 872 void b(int imm26); |
| 873 |
| 874 // Conditional branch to PC offset. |
| 875 void b(int imm19, Condition cond); |
| 876 |
| 877 // Branch-link to label / pc offset. |
| 878 void bl(Label* label); |
| 879 void bl(int imm26); |
| 880 |
| 881 // Compare and branch to label / pc offset if zero. |
| 882 void cbz(const Register& rt, Label* label); |
| 883 void cbz(const Register& rt, int imm19); |
| 884 |
| 885 // Compare and branch to label / pc offset if not zero. |
| 886 void cbnz(const Register& rt, Label* label); |
| 887 void cbnz(const Register& rt, int imm19); |
| 888 |
| 889 // Test bit and branch to label / pc offset if zero. |
| 890 void tbz(const Register& rt, unsigned bit_pos, Label* label); |
| 891 void tbz(const Register& rt, unsigned bit_pos, int imm14); |
| 892 |
| 893 // Test bit and branch to label / pc offset if not zero. |
| 894 void tbnz(const Register& rt, unsigned bit_pos, Label* label); |
| 895 void tbnz(const Register& rt, unsigned bit_pos, int imm14); |
| 896 |
| 897 // Address calculation instructions. |
| 898 // Calculate a PC-relative address. Unlike for branches the offset in adr is |
| 899 // unscaled (i.e. the result can be unaligned). |
| 900 void adr(const Register& rd, Label* label); |
| 901 void adr(const Register& rd, int imm21); |
| 902 |
| 903 // Data Processing instructions. |
| 904 // Add. |
| 905 void add(const Register& rd, |
| 906 const Register& rn, |
| 907 const Operand& operand); |
| 908 |
| 909 // Add and update status flags. |
| 910 void adds(const Register& rd, |
| 911 const Register& rn, |
| 912 const Operand& operand); |
| 913 |
| 914 // Compare negative. |
| 915 void cmn(const Register& rn, const Operand& operand); |
| 916 |
| 917 // Subtract. |
| 918 void sub(const Register& rd, |
| 919 const Register& rn, |
| 920 const Operand& operand); |
| 921 |
| 922 // Subtract and update status flags. |
| 923 void subs(const Register& rd, |
| 924 const Register& rn, |
| 925 const Operand& operand); |
| 926 |
| 927 // Compare. |
| 928 void cmp(const Register& rn, const Operand& operand); |
| 929 |
| 930 // Negate. |
| 931 void neg(const Register& rd, |
| 932 const Operand& operand); |
| 933 |
| 934 // Negate and update status flags. |
| 935 void negs(const Register& rd, |
| 936 const Operand& operand); |
| 937 |
| 938 // Add with carry bit. |
| 939 void adc(const Register& rd, |
| 940 const Register& rn, |
| 941 const Operand& operand); |
| 942 |
| 943 // Add with carry bit and update status flags. |
| 944 void adcs(const Register& rd, |
| 945 const Register& rn, |
| 946 const Operand& operand); |
| 947 |
| 948 // Subtract with carry bit. |
| 949 void sbc(const Register& rd, |
| 950 const Register& rn, |
| 951 const Operand& operand); |
| 952 |
| 953 // Subtract with carry bit and update status flags. |
| 954 void sbcs(const Register& rd, |
| 955 const Register& rn, |
| 956 const Operand& operand); |
| 957 |
| 958 // Negate with carry bit. |
| 959 void ngc(const Register& rd, |
| 960 const Operand& operand); |
| 961 |
| 962 // Negate with carry bit and update status flags. |
| 963 void ngcs(const Register& rd, |
| 964 const Operand& operand); |
| 965 |
| 966 // Logical instructions. |
| 967 // Bitwise and (A & B). |
| 968 void and_(const Register& rd, |
| 969 const Register& rn, |
| 970 const Operand& operand); |
| 971 |
| 972 // Bitwise and (A & B) and update status flags. |
| 973 void ands(const Register& rd, |
| 974 const Register& rn, |
| 975 const Operand& operand); |
| 976 |
| 977 // Bit test, and set flags. |
| 978 void tst(const Register& rn, const Operand& operand); |
| 979 |
| 980 // Bit clear (A & ~B). |
| 981 void bic(const Register& rd, |
| 982 const Register& rn, |
| 983 const Operand& operand); |
| 984 |
| 985 // Bit clear (A & ~B) and update status flags. |
| 986 void bics(const Register& rd, |
| 987 const Register& rn, |
| 988 const Operand& operand); |
| 989 |
| 990 // Bitwise or (A | B). |
| 991 void orr(const Register& rd, const Register& rn, const Operand& operand); |
| 992 |
| 993 // Bitwise nor (A | ~B). |
| 994 void orn(const Register& rd, const Register& rn, const Operand& operand); |
| 995 |
| 996 // Bitwise eor/xor (A ^ B). |
| 997 void eor(const Register& rd, const Register& rn, const Operand& operand); |
| 998 |
| 999 // Bitwise enor/xnor (A ^ ~B). |
| 1000 void eon(const Register& rd, const Register& rn, const Operand& operand); |
| 1001 |
| 1002 // Logical shift left variable. |
| 1003 void lslv(const Register& rd, const Register& rn, const Register& rm); |
| 1004 |
| 1005 // Logical shift right variable. |
| 1006 void lsrv(const Register& rd, const Register& rn, const Register& rm); |
| 1007 |
| 1008 // Arithmetic shift right variable. |
| 1009 void asrv(const Register& rd, const Register& rn, const Register& rm); |
| 1010 |
| 1011 // Rotate right variable. |
| 1012 void rorv(const Register& rd, const Register& rn, const Register& rm); |
| 1013 |
| 1014 // Bitfield instructions. |
| 1015 // Bitfield move. |
| 1016 void bfm(const Register& rd, |
| 1017 const Register& rn, |
| 1018 unsigned immr, |
| 1019 unsigned imms); |
| 1020 |
| 1021 // Signed bitfield move. |
| 1022 void sbfm(const Register& rd, |
| 1023 const Register& rn, |
| 1024 unsigned immr, |
| 1025 unsigned imms); |
| 1026 |
| 1027 // Unsigned bitfield move. |
| 1028 void ubfm(const Register& rd, |
| 1029 const Register& rn, |
| 1030 unsigned immr, |
| 1031 unsigned imms); |
| 1032 |
| 1033 // Bfm aliases. |
| 1034 // Bitfield insert. |
| 1035 inline void bfi(const Register& rd, |
| 1036 const Register& rn, |
| 1037 unsigned lsb, |
| 1038 unsigned width) { |
| 1039 ASSERT(width >= 1); |
| 1040 ASSERT(lsb + width <= rn.SizeInBits()); |
| 1041 bfm(rd, rn, (rd.SizeInBits() - lsb) & (rd.SizeInBits() - 1), width - 1); |
| 1042 } |
| 1043 |
| 1044 // Bitfield extract and insert low. |
| 1045 inline void bfxil(const Register& rd, |
| 1046 const Register& rn, |
| 1047 unsigned lsb, |
| 1048 unsigned width) { |
| 1049 ASSERT(width >= 1); |
| 1050 ASSERT(lsb + width <= rn.SizeInBits()); |
| 1051 bfm(rd, rn, lsb, lsb + width - 1); |
| 1052 } |
| 1053 |
| 1054 // Sbfm aliases. |
| 1055 // Arithmetic shift right. |
| 1056 inline void asr(const Register& rd, const Register& rn, unsigned shift) { |
| 1057 ASSERT(shift < rd.SizeInBits()); |
| 1058 sbfm(rd, rn, shift, rd.SizeInBits() - 1); |
| 1059 } |
| 1060 |
| 1061 // Signed bitfield insert in zero. |
| 1062 inline void sbfiz(const Register& rd, |
| 1063 const Register& rn, |
| 1064 unsigned lsb, |
| 1065 unsigned width) { |
| 1066 ASSERT(width >= 1); |
| 1067 ASSERT(lsb + width <= rn.SizeInBits()); |
| 1068 sbfm(rd, rn, (rd.SizeInBits() - lsb) & (rd.SizeInBits() - 1), width - 1); |
| 1069 } |
| 1070 |
| 1071 // Signed bitfield extract. |
| 1072 inline void sbfx(const Register& rd, |
| 1073 const Register& rn, |
| 1074 unsigned lsb, |
| 1075 unsigned width) { |
| 1076 ASSERT(width >= 1); |
| 1077 ASSERT(lsb + width <= rn.SizeInBits()); |
| 1078 sbfm(rd, rn, lsb, lsb + width - 1); |
| 1079 } |
| 1080 |
| 1081 // Signed extend byte. |
| 1082 inline void sxtb(const Register& rd, const Register& rn) { |
| 1083 sbfm(rd, rn, 0, 7); |
| 1084 } |
| 1085 |
| 1086 // Signed extend halfword. |
| 1087 inline void sxth(const Register& rd, const Register& rn) { |
| 1088 sbfm(rd, rn, 0, 15); |
| 1089 } |
| 1090 |
| 1091 // Signed extend word. |
| 1092 inline void sxtw(const Register& rd, const Register& rn) { |
| 1093 sbfm(rd, rn, 0, 31); |
| 1094 } |
| 1095 |
| 1096 // Ubfm aliases. |
| 1097 // Logical shift left. |
| 1098 inline void lsl(const Register& rd, const Register& rn, unsigned shift) { |
| 1099 unsigned reg_size = rd.SizeInBits(); |
| 1100 ASSERT(shift < reg_size); |
| 1101 ubfm(rd, rn, (reg_size - shift) % reg_size, reg_size - shift - 1); |
| 1102 } |
| 1103 |
| 1104 // Logical shift right. |
| 1105 inline void lsr(const Register& rd, const Register& rn, unsigned shift) { |
| 1106 ASSERT(shift < rd.SizeInBits()); |
| 1107 ubfm(rd, rn, shift, rd.SizeInBits() - 1); |
| 1108 } |
| 1109 |
| 1110 // Unsigned bitfield insert in zero. |
| 1111 inline void ubfiz(const Register& rd, |
| 1112 const Register& rn, |
| 1113 unsigned lsb, |
| 1114 unsigned width) { |
| 1115 ASSERT(width >= 1); |
| 1116 ASSERT(lsb + width <= rn.SizeInBits()); |
| 1117 ubfm(rd, rn, (rd.SizeInBits() - lsb) & (rd.SizeInBits() - 1), width - 1); |
| 1118 } |
| 1119 |
| 1120 // Unsigned bitfield extract. |
| 1121 inline void ubfx(const Register& rd, |
| 1122 const Register& rn, |
| 1123 unsigned lsb, |
| 1124 unsigned width) { |
| 1125 ASSERT(width >= 1); |
| 1126 ASSERT(lsb + width <= rn.SizeInBits()); |
| 1127 ubfm(rd, rn, lsb, lsb + width - 1); |
| 1128 } |
| 1129 |
| 1130 // Unsigned extend byte. |
| 1131 inline void uxtb(const Register& rd, const Register& rn) { |
| 1132 ubfm(rd, rn, 0, 7); |
| 1133 } |
| 1134 |
| 1135 // Unsigned extend halfword. |
| 1136 inline void uxth(const Register& rd, const Register& rn) { |
| 1137 ubfm(rd, rn, 0, 15); |
| 1138 } |
| 1139 |
| 1140 // Unsigned extend word. |
| 1141 inline void uxtw(const Register& rd, const Register& rn) { |
| 1142 ubfm(rd, rn, 0, 31); |
| 1143 } |
| 1144 |
| 1145 // Extract. |
| 1146 void extr(const Register& rd, |
| 1147 const Register& rn, |
| 1148 const Register& rm, |
| 1149 unsigned lsb); |
| 1150 |
| 1151 // Conditional select: rd = cond ? rn : rm. |
| 1152 void csel(const Register& rd, |
| 1153 const Register& rn, |
| 1154 const Register& rm, |
| 1155 Condition cond); |
| 1156 |
| 1157 // Conditional select increment: rd = cond ? rn : rm + 1. |
| 1158 void csinc(const Register& rd, |
| 1159 const Register& rn, |
| 1160 const Register& rm, |
| 1161 Condition cond); |
| 1162 |
| 1163 // Conditional select inversion: rd = cond ? rn : ~rm. |
| 1164 void csinv(const Register& rd, |
| 1165 const Register& rn, |
| 1166 const Register& rm, |
| 1167 Condition cond); |
| 1168 |
| 1169 // Conditional select negation: rd = cond ? rn : -rm. |
| 1170 void csneg(const Register& rd, |
| 1171 const Register& rn, |
| 1172 const Register& rm, |
| 1173 Condition cond); |
| 1174 |
| 1175 // Conditional set: rd = cond ? 1 : 0. |
| 1176 void cset(const Register& rd, Condition cond); |
| 1177 |
| 1178 // Conditional set minus: rd = cond ? -1 : 0. |
| 1179 void csetm(const Register& rd, Condition cond); |
| 1180 |
| 1181 // Conditional increment: rd = cond ? rn + 1 : rn. |
| 1182 void cinc(const Register& rd, const Register& rn, Condition cond); |
| 1183 |
| 1184 // Conditional invert: rd = cond ? ~rn : rn. |
| 1185 void cinv(const Register& rd, const Register& rn, Condition cond); |
| 1186 |
| 1187 // Conditional negate: rd = cond ? -rn : rn. |
| 1188 void cneg(const Register& rd, const Register& rn, Condition cond); |
| 1189 |
| 1190 // Extr aliases. |
| 1191 inline void ror(const Register& rd, const Register& rs, unsigned shift) { |
| 1192 extr(rd, rs, rs, shift); |
| 1193 } |
| 1194 |
| 1195 // Conditional comparison. |
| 1196 // Conditional compare negative. |
| 1197 void ccmn(const Register& rn, |
| 1198 const Operand& operand, |
| 1199 StatusFlags nzcv, |
| 1200 Condition cond); |
| 1201 |
| 1202 // Conditional compare. |
| 1203 void ccmp(const Register& rn, |
| 1204 const Operand& operand, |
| 1205 StatusFlags nzcv, |
| 1206 Condition cond); |
| 1207 |
| 1208 // Multiplication. |
| 1209 // 32 x 32 -> 32-bit and 64 x 64 -> 64-bit multiply. |
| 1210 void mul(const Register& rd, const Register& rn, const Register& rm); |
| 1211 |
| 1212 // 32 + 32 x 32 -> 32-bit and 64 + 64 x 64 -> 64-bit multiply accumulate. |
| 1213 void madd(const Register& rd, |
| 1214 const Register& rn, |
| 1215 const Register& rm, |
| 1216 const Register& ra); |
| 1217 |
| 1218 // -(32 x 32) -> 32-bit and -(64 x 64) -> 64-bit multiply. |
| 1219 void mneg(const Register& rd, const Register& rn, const Register& rm); |
| 1220 |
| 1221 // 32 - 32 x 32 -> 32-bit and 64 - 64 x 64 -> 64-bit multiply subtract. |
| 1222 void msub(const Register& rd, |
| 1223 const Register& rn, |
| 1224 const Register& rm, |
| 1225 const Register& ra); |
| 1226 |
| 1227 // 32 x 32 -> 64-bit multiply. |
| 1228 void smull(const Register& rd, const Register& rn, const Register& rm); |
| 1229 |
| 1230 // Xd = bits<127:64> of Xn * Xm. |
| 1231 void smulh(const Register& rd, const Register& rn, const Register& rm); |
| 1232 |
| 1233 // Signed 32 x 32 -> 64-bit multiply and accumulate. |
| 1234 void smaddl(const Register& rd, |
| 1235 const Register& rn, |
| 1236 const Register& rm, |
| 1237 const Register& ra); |
| 1238 |
| 1239 // Unsigned 32 x 32 -> 64-bit multiply and accumulate. |
| 1240 void umaddl(const Register& rd, |
| 1241 const Register& rn, |
| 1242 const Register& rm, |
| 1243 const Register& ra); |
| 1244 |
| 1245 // Signed 32 x 32 -> 64-bit multiply and subtract. |
| 1246 void smsubl(const Register& rd, |
| 1247 const Register& rn, |
| 1248 const Register& rm, |
| 1249 const Register& ra); |
| 1250 |
| 1251 // Unsigned 32 x 32 -> 64-bit multiply and subtract. |
| 1252 void umsubl(const Register& rd, |
| 1253 const Register& rn, |
| 1254 const Register& rm, |
| 1255 const Register& ra); |
| 1256 |
| 1257 // Signed integer divide. |
| 1258 void sdiv(const Register& rd, const Register& rn, const Register& rm); |
| 1259 |
| 1260 // Unsigned integer divide. |
| 1261 void udiv(const Register& rd, const Register& rn, const Register& rm); |
| 1262 |
| 1263 // Bit count, bit reverse and endian reverse. |
| 1264 void rbit(const Register& rd, const Register& rn); |
| 1265 void rev16(const Register& rd, const Register& rn); |
| 1266 void rev32(const Register& rd, const Register& rn); |
| 1267 void rev(const Register& rd, const Register& rn); |
| 1268 void clz(const Register& rd, const Register& rn); |
| 1269 void cls(const Register& rd, const Register& rn); |
| 1270 |
| 1271 // Memory instructions. |
| 1272 |
| 1273 // Load literal from pc + offset_from_pc. |
| 1274 void LoadLiteral(const CPURegister& rt, int offset_from_pc); |
| 1275 |
| 1276 // Load integer or FP register. |
| 1277 void ldr(const CPURegister& rt, const MemOperand& src); |
| 1278 |
| 1279 // Store integer or FP register. |
| 1280 void str(const CPURegister& rt, const MemOperand& dst); |
| 1281 |
| 1282 // Load word with sign extension. |
| 1283 void ldrsw(const Register& rt, const MemOperand& src); |
| 1284 |
| 1285 // Load byte. |
| 1286 void ldrb(const Register& rt, const MemOperand& src); |
| 1287 |
| 1288 // Store byte. |
| 1289 void strb(const Register& rt, const MemOperand& dst); |
| 1290 |
| 1291 // Load byte with sign extension. |
| 1292 void ldrsb(const Register& rt, const MemOperand& src); |
| 1293 |
| 1294 // Load half-word. |
| 1295 void ldrh(const Register& rt, const MemOperand& src); |
| 1296 |
| 1297 // Store half-word. |
| 1298 void strh(const Register& rt, const MemOperand& dst); |
| 1299 |
| 1300 // Load half-word with sign extension. |
| 1301 void ldrsh(const Register& rt, const MemOperand& src); |
| 1302 |
| 1303 // Load integer or FP register pair. |
| 1304 void ldp(const CPURegister& rt, const CPURegister& rt2, |
| 1305 const MemOperand& src); |
| 1306 |
| 1307 // Store integer or FP register pair. |
| 1308 void stp(const CPURegister& rt, const CPURegister& rt2, |
| 1309 const MemOperand& dst); |
| 1310 |
| 1311 // Load word pair with sign extension. |
| 1312 void ldpsw(const Register& rt, const Register& rt2, const MemOperand& src); |
| 1313 |
| 1314 // Load integer or FP register pair, non-temporal. |
| 1315 void ldnp(const CPURegister& rt, const CPURegister& rt2, |
| 1316 const MemOperand& src); |
| 1317 |
| 1318 // Store integer or FP register pair, non-temporal. |
| 1319 void stnp(const CPURegister& rt, const CPURegister& rt2, |
| 1320 const MemOperand& dst); |
| 1321 |
| 1322 // Load literal to register. |
| 1323 void ldr(const Register& rt, uint64_t imm); |
| 1324 |
| 1325 // Load literal to FP register. |
| 1326 void ldr(const FPRegister& ft, double imm); |
| 1327 |
| 1328 // Move instructions. The default shift of -1 indicates that the move |
| 1329 // instruction will calculate an appropriate 16-bit immediate and left shift |
| 1330 // that is equal to the 64-bit immediate argument. If an explicit left shift |
| 1331 // is specified (0, 16, 32 or 48), the immediate must be a 16-bit value. |
| 1332 // |
| 1333 // For movk, an explicit shift can be used to indicate which half word should |
| 1334 // be overwritten, eg. movk(x0, 0, 0) will overwrite the least-significant |
| 1335 // half word with zero, whereas movk(x0, 0, 48) will overwrite the |
| 1336 // most-significant. |
| 1337 |
| 1338 // Move and keep. |
| 1339 void movk(const Register& rd, uint64_t imm, int shift = -1) { |
| 1340 MoveWide(rd, imm, shift, MOVK); |
| 1341 } |
| 1342 |
| 1343 // Move with non-zero. |
| 1344 void movn(const Register& rd, uint64_t imm, int shift = -1) { |
| 1345 MoveWide(rd, imm, shift, MOVN); |
| 1346 } |
| 1347 |
| 1348 // Move with zero. |
| 1349 void movz(const Register& rd, uint64_t imm, int shift = -1) { |
| 1350 MoveWide(rd, imm, shift, MOVZ); |
| 1351 } |
| 1352 |
| 1353 // Misc instructions. |
| 1354 // Monitor debug-mode breakpoint. |
| 1355 void brk(int code); |
| 1356 |
| 1357 // Halting debug-mode breakpoint. |
| 1358 void hlt(int code); |
| 1359 |
| 1360 // Move register to register. |
| 1361 void mov(const Register& rd, const Register& rn); |
| 1362 |
| 1363 // Move NOT(operand) to register. |
| 1364 void mvn(const Register& rd, const Operand& operand); |
| 1365 |
| 1366 // System instructions. |
| 1367 // Move to register from system register. |
| 1368 void mrs(const Register& rt, SystemRegister sysreg); |
| 1369 |
| 1370 // Move from register to system register. |
| 1371 void msr(SystemRegister sysreg, const Register& rt); |
| 1372 |
| 1373 // System hint. |
| 1374 void hint(SystemHint code); |
| 1375 |
| 1376 // Data memory barrier |
| 1377 void dmb(BarrierDomain domain, BarrierType type); |
| 1378 |
| 1379 // Data synchronization barrier |
| 1380 void dsb(BarrierDomain domain, BarrierType type); |
| 1381 |
| 1382 // Instruction synchronization barrier |
| 1383 void isb(); |
| 1384 |
| 1385 // Alias for system instructions. |
| 1386 void nop() { hint(NOP); } |
| 1387 |
| 1388 // Different nop operations are used by the code generator to detect certain |
| 1389 // states of the generated code. |
| 1390 enum NopMarkerTypes { |
| 1391 DEBUG_BREAK_NOP, |
| 1392 FIRST_NOP_MARKER = DEBUG_BREAK_NOP, |
| 1393 LAST_NOP_MARKER = DEBUG_BREAK_NOP |
| 1394 }; |
| 1395 |
| 1396 inline void nop(NopMarkerTypes n) { |
| 1397 ASSERT((FIRST_NOP_MARKER <= n) && (n <= LAST_NOP_MARKER)); |
| 1398 mov(Register::XRegFromCode(n), Register::XRegFromCode(n)); |
| 1399 } |
| 1400 |
| 1401 // FP instructions. |
| 1402 // Move immediate to FP register. |
| 1403 void fmov(FPRegister fd, double imm); |
| 1404 |
| 1405 // Move FP register to register. |
| 1406 void fmov(Register rd, FPRegister fn); |
| 1407 |
| 1408 // Move register to FP register. |
| 1409 void fmov(FPRegister fd, Register rn); |
| 1410 |
| 1411 // Move FP register to FP register. |
| 1412 void fmov(FPRegister fd, FPRegister fn); |
| 1413 |
| 1414 // FP add. |
| 1415 void fadd(const FPRegister& fd, const FPRegister& fn, const FPRegister& fm); |
| 1416 |
| 1417 // FP subtract. |
| 1418 void fsub(const FPRegister& fd, const FPRegister& fn, const FPRegister& fm); |
| 1419 |
| 1420 // FP multiply. |
| 1421 void fmul(const FPRegister& fd, const FPRegister& fn, const FPRegister& fm); |
| 1422 |
| 1423 // FP fused multiply and add. |
| 1424 void fmadd(const FPRegister& fd, |
| 1425 const FPRegister& fn, |
| 1426 const FPRegister& fm, |
| 1427 const FPRegister& fa); |
| 1428 |
| 1429 // FP fused multiply and subtract. |
| 1430 void fmsub(const FPRegister& fd, |
| 1431 const FPRegister& fn, |
| 1432 const FPRegister& fm, |
| 1433 const FPRegister& fa); |
| 1434 |
| 1435 // FP fused multiply, add and negate. |
| 1436 void fnmadd(const FPRegister& fd, |
| 1437 const FPRegister& fn, |
| 1438 const FPRegister& fm, |
| 1439 const FPRegister& fa); |
| 1440 |
| 1441 // FP fused multiply, subtract and negate. |
| 1442 void fnmsub(const FPRegister& fd, |
| 1443 const FPRegister& fn, |
| 1444 const FPRegister& fm, |
| 1445 const FPRegister& fa); |
| 1446 |
| 1447 // FP divide. |
| 1448 void fdiv(const FPRegister& fd, const FPRegister& fn, const FPRegister& fm); |
| 1449 |
| 1450 // FP maximum. |
| 1451 void fmax(const FPRegister& fd, const FPRegister& fn, const FPRegister& fm); |
| 1452 |
| 1453 // FP minimum. |
| 1454 void fmin(const FPRegister& fd, const FPRegister& fn, const FPRegister& fm); |
| 1455 |
| 1456 // FP maximum. |
| 1457 void fmaxnm(const FPRegister& fd, const FPRegister& fn, const FPRegister& fm); |
| 1458 |
| 1459 // FP minimum. |
| 1460 void fminnm(const FPRegister& fd, const FPRegister& fn, const FPRegister& fm); |
| 1461 |
| 1462 // FP absolute. |
| 1463 void fabs(const FPRegister& fd, const FPRegister& fn); |
| 1464 |
| 1465 // FP negate. |
| 1466 void fneg(const FPRegister& fd, const FPRegister& fn); |
| 1467 |
| 1468 // FP square root. |
| 1469 void fsqrt(const FPRegister& fd, const FPRegister& fn); |
| 1470 |
| 1471 // FP round to integer (nearest with ties to away). |
| 1472 void frinta(const FPRegister& fd, const FPRegister& fn); |
| 1473 |
| 1474 // FP round to integer (nearest with ties to even). |
| 1475 void frintn(const FPRegister& fd, const FPRegister& fn); |
| 1476 |
| 1477 // FP round to integer (towards zero.) |
| 1478 void frintz(const FPRegister& fd, const FPRegister& fn); |
| 1479 |
| 1480 // FP compare registers. |
| 1481 void fcmp(const FPRegister& fn, const FPRegister& fm); |
| 1482 |
| 1483 // FP compare immediate. |
| 1484 void fcmp(const FPRegister& fn, double value); |
| 1485 |
| 1486 // FP conditional compare. |
| 1487 void fccmp(const FPRegister& fn, |
| 1488 const FPRegister& fm, |
| 1489 StatusFlags nzcv, |
| 1490 Condition cond); |
| 1491 |
| 1492 // FP conditional select. |
| 1493 void fcsel(const FPRegister& fd, |
| 1494 const FPRegister& fn, |
| 1495 const FPRegister& fm, |
| 1496 Condition cond); |
| 1497 |
| 1498 // Common FP Convert function |
| 1499 void FPConvertToInt(const Register& rd, |
| 1500 const FPRegister& fn, |
| 1501 FPIntegerConvertOp op); |
| 1502 |
| 1503 // FP convert between single and double precision. |
| 1504 void fcvt(const FPRegister& fd, const FPRegister& fn); |
| 1505 |
| 1506 // Convert FP to unsigned integer (nearest with ties to away). |
| 1507 void fcvtau(const Register& rd, const FPRegister& fn); |
| 1508 |
| 1509 // Convert FP to signed integer (nearest with ties to away). |
| 1510 void fcvtas(const Register& rd, const FPRegister& fn); |
| 1511 |
| 1512 // Convert FP to unsigned integer (round towards -infinity). |
| 1513 void fcvtmu(const Register& rd, const FPRegister& fn); |
| 1514 |
| 1515 // Convert FP to signed integer (round towards -infinity). |
| 1516 void fcvtms(const Register& rd, const FPRegister& fn); |
| 1517 |
| 1518 // Convert FP to unsigned integer (nearest with ties to even). |
| 1519 void fcvtnu(const Register& rd, const FPRegister& fn); |
| 1520 |
| 1521 // Convert FP to signed integer (nearest with ties to even). |
| 1522 void fcvtns(const Register& rd, const FPRegister& fn); |
| 1523 |
| 1524 // Convert FP to unsigned integer (round towards zero). |
| 1525 void fcvtzu(const Register& rd, const FPRegister& fn); |
| 1526 |
| 1527 // Convert FP to signed integer (rounf towards zero). |
| 1528 void fcvtzs(const Register& rd, const FPRegister& fn); |
| 1529 |
| 1530 // Convert signed integer or fixed point to FP. |
| 1531 void scvtf(const FPRegister& fd, const Register& rn, unsigned fbits = 0); |
| 1532 |
| 1533 // Convert unsigned integer or fixed point to FP. |
| 1534 void ucvtf(const FPRegister& fd, const Register& rn, unsigned fbits = 0); |
| 1535 |
| 1536 // Instruction functions used only for test, debug, and patching. |
| 1537 // Emit raw instructions in the instruction stream. |
| 1538 inline void dci(Instr raw_inst) { Emit(raw_inst); } |
| 1539 |
| 1540 // Emit 8 bits of data in the instruction stream. |
| 1541 inline void dc8(uint8_t data) { EmitData(&data, sizeof(data)); } |
| 1542 |
| 1543 // Emit 32 bits of data in the instruction stream. |
| 1544 inline void dc32(uint32_t data) { EmitData(&data, sizeof(data)); } |
| 1545 |
| 1546 // Emit 64 bits of data in the instruction stream. |
| 1547 inline void dc64(uint64_t data) { EmitData(&data, sizeof(data)); } |
| 1548 |
| 1549 // Copy a string into the instruction stream, including the terminating NULL |
| 1550 // character. The instruction pointer (pc_) is then aligned correctly for |
| 1551 // subsequent instructions. |
| 1552 void EmitStringData(const char * string) { |
| 1553 size_t len = strlen(string) + 1; |
| 1554 ASSERT(RoundUp(len, kInstructionSize) <= static_cast<size_t>(kGap)); |
| 1555 EmitData(string, len); |
| 1556 // Pad with NULL characters until pc_ is aligned. |
| 1557 const char pad[] = {'\0', '\0', '\0', '\0'}; |
| 1558 STATIC_ASSERT(sizeof(pad) == kInstructionSize); |
| 1559 byte* next_pc = AlignUp(pc_, kInstructionSize); |
| 1560 EmitData(&pad, next_pc - pc_); |
| 1561 } |
| 1562 |
| 1563 // Pseudo-instructions ------------------------------------------------------ |
| 1564 |
| 1565 // Parameters are described in a64/instructions-a64.h. |
| 1566 void debug(const char* message, uint32_t code, Instr params = BREAK); |
| 1567 |
| 1568 // Required by V8. |
| 1569 inline void dd(uint32_t data) { dc32(data); } |
| 1570 void db(uint8_t data) { dc8(data); } |
| 1571 |
| 1572 // Code generation helpers -------------------------------------------------- |
| 1573 |
| 1574 unsigned num_pending_reloc_info() const { return num_pending_reloc_info_; } |
| 1575 |
| 1576 inline Instruction* InstructionAt(int offset) const { |
| 1577 return reinterpret_cast<Instruction*>(buffer_ + offset); |
| 1578 } |
| 1579 |
| 1580 // Register encoding. |
| 1581 static inline Instr Rd(CPURegister rd) { |
| 1582 ASSERT(rd.code() != kSPRegInternalCode); |
| 1583 return rd.code() << Rd_offset; |
| 1584 } |
| 1585 |
| 1586 static inline Instr Rn(CPURegister rn) { |
| 1587 ASSERT(rn.code() != kSPRegInternalCode); |
| 1588 return rn.code() << Rn_offset; |
| 1589 } |
| 1590 |
| 1591 static inline Instr Rm(CPURegister rm) { |
| 1592 ASSERT(rm.code() != kSPRegInternalCode); |
| 1593 return rm.code() << Rm_offset; |
| 1594 } |
| 1595 |
| 1596 static inline Instr Ra(CPURegister ra) { |
| 1597 ASSERT(ra.code() != kSPRegInternalCode); |
| 1598 return ra.code() << Ra_offset; |
| 1599 } |
| 1600 |
| 1601 static inline Instr Rt(CPURegister rt) { |
| 1602 ASSERT(rt.code() != kSPRegInternalCode); |
| 1603 return rt.code() << Rt_offset; |
| 1604 } |
| 1605 |
| 1606 static inline Instr Rt2(CPURegister rt2) { |
| 1607 ASSERT(rt2.code() != kSPRegInternalCode); |
| 1608 return rt2.code() << Rt2_offset; |
| 1609 } |
| 1610 |
| 1611 // These encoding functions allow the stack pointer to be encoded, and |
| 1612 // disallow the zero register. |
| 1613 static inline Instr RdSP(Register rd) { |
| 1614 ASSERT(!rd.IsZero()); |
| 1615 return (rd.code() & kRegCodeMask) << Rd_offset; |
| 1616 } |
| 1617 |
| 1618 static inline Instr RnSP(Register rn) { |
| 1619 ASSERT(!rn.IsZero()); |
| 1620 return (rn.code() & kRegCodeMask) << Rn_offset; |
| 1621 } |
| 1622 |
| 1623 // Flags encoding. |
| 1624 inline static Instr Flags(FlagsUpdate S); |
| 1625 inline static Instr Cond(Condition cond); |
| 1626 |
| 1627 // PC-relative address encoding. |
| 1628 inline static Instr ImmPCRelAddress(int imm21); |
| 1629 |
| 1630 // Branch encoding. |
| 1631 inline static Instr ImmUncondBranch(int imm26); |
| 1632 inline static Instr ImmCondBranch(int imm19); |
| 1633 inline static Instr ImmCmpBranch(int imm19); |
| 1634 inline static Instr ImmTestBranch(int imm14); |
| 1635 inline static Instr ImmTestBranchBit(unsigned bit_pos); |
| 1636 |
| 1637 // Data Processing encoding. |
| 1638 inline static Instr SF(Register rd); |
| 1639 inline static Instr ImmAddSub(int64_t imm); |
| 1640 inline static Instr ImmS(unsigned imms, unsigned reg_size); |
| 1641 inline static Instr ImmR(unsigned immr, unsigned reg_size); |
| 1642 inline static Instr ImmSetBits(unsigned imms, unsigned reg_size); |
| 1643 inline static Instr ImmRotate(unsigned immr, unsigned reg_size); |
| 1644 inline static Instr ImmLLiteral(int imm19); |
| 1645 inline static Instr BitN(unsigned bitn, unsigned reg_size); |
| 1646 inline static Instr ShiftDP(Shift shift); |
| 1647 inline static Instr ImmDPShift(unsigned amount); |
| 1648 inline static Instr ExtendMode(Extend extend); |
| 1649 inline static Instr ImmExtendShift(unsigned left_shift); |
| 1650 inline static Instr ImmCondCmp(unsigned imm); |
| 1651 inline static Instr Nzcv(StatusFlags nzcv); |
| 1652 |
| 1653 // MemOperand offset encoding. |
| 1654 inline static Instr ImmLSUnsigned(int imm12); |
| 1655 inline static Instr ImmLS(int imm9); |
| 1656 inline static Instr ImmLSPair(int imm7, LSDataSize size); |
| 1657 inline static Instr ImmShiftLS(unsigned shift_amount); |
| 1658 inline static Instr ImmException(int imm16); |
| 1659 inline static Instr ImmSystemRegister(int imm15); |
| 1660 inline static Instr ImmHint(int imm7); |
| 1661 inline static Instr ImmBarrierDomain(int imm2); |
| 1662 inline static Instr ImmBarrierType(int imm2); |
| 1663 inline static LSDataSize CalcLSDataSize(LoadStoreOp op); |
| 1664 |
| 1665 // Move immediates encoding. |
| 1666 inline static Instr ImmMoveWide(uint64_t imm); |
| 1667 inline static Instr ShiftMoveWide(int64_t shift); |
| 1668 |
| 1669 // FP Immediates. |
| 1670 inline static Instr ImmFP32(float imm); |
| 1671 static Instr ImmFP64(double imm); |
| 1672 inline static Instr FPScale(unsigned scale); |
| 1673 |
| 1674 // FP register type. |
| 1675 inline static Instr FPType(FPRegister fd); |
| 1676 |
| 1677 // Class for scoping postponing the constant pool generation. |
| 1678 class BlockConstPoolScope { |
| 1679 public: |
| 1680 explicit BlockConstPoolScope(Assembler* assem) : assem_(assem) { |
| 1681 assem_->StartBlockConstPool(); |
| 1682 } |
| 1683 ~BlockConstPoolScope() { |
| 1684 assem_->EndBlockConstPool(); |
| 1685 } |
| 1686 |
| 1687 private: |
| 1688 Assembler* assem_; |
| 1689 |
| 1690 DISALLOW_IMPLICIT_CONSTRUCTORS(BlockConstPoolScope); |
| 1691 }; |
| 1692 |
| 1693 // Check if is time to emit a constant pool. |
| 1694 void CheckConstPool(bool force_emit, bool require_jump); |
| 1695 |
| 1696 protected: |
| 1697 inline const Register& AppropriateZeroRegFor(const CPURegister& reg) const; |
| 1698 |
| 1699 void LoadStore(const CPURegister& rt, |
| 1700 const MemOperand& addr, |
| 1701 LoadStoreOp op); |
| 1702 static bool IsImmLSUnscaled(ptrdiff_t offset); |
| 1703 static bool IsImmLSScaled(ptrdiff_t offset, LSDataSize size); |
| 1704 |
| 1705 void Logical(const Register& rd, |
| 1706 const Register& rn, |
| 1707 const Operand& operand, |
| 1708 LogicalOp op); |
| 1709 void LogicalImmediate(const Register& rd, |
| 1710 const Register& rn, |
| 1711 unsigned n, |
| 1712 unsigned imm_s, |
| 1713 unsigned imm_r, |
| 1714 LogicalOp op); |
| 1715 static bool IsImmLogical(uint64_t value, |
| 1716 unsigned width, |
| 1717 unsigned* n, |
| 1718 unsigned* imm_s, |
| 1719 unsigned* imm_r); |
| 1720 |
| 1721 void ConditionalCompare(const Register& rn, |
| 1722 const Operand& operand, |
| 1723 StatusFlags nzcv, |
| 1724 Condition cond, |
| 1725 ConditionalCompareOp op); |
| 1726 static bool IsImmConditionalCompare(int64_t immediate); |
| 1727 |
| 1728 void AddSubWithCarry(const Register& rd, |
| 1729 const Register& rn, |
| 1730 const Operand& operand, |
| 1731 FlagsUpdate S, |
| 1732 AddSubWithCarryOp op); |
| 1733 |
| 1734 // Functions for emulating operands not directly supported by the instruction |
| 1735 // set. |
| 1736 void EmitShift(const Register& rd, |
| 1737 const Register& rn, |
| 1738 Shift shift, |
| 1739 unsigned amount); |
| 1740 void EmitExtendShift(const Register& rd, |
| 1741 const Register& rn, |
| 1742 Extend extend, |
| 1743 unsigned left_shift); |
| 1744 |
| 1745 void AddSub(const Register& rd, |
| 1746 const Register& rn, |
| 1747 const Operand& operand, |
| 1748 FlagsUpdate S, |
| 1749 AddSubOp op); |
| 1750 static bool IsImmAddSub(int64_t immediate); |
| 1751 |
| 1752 inline void LoadRelocated(const CPURegister& rt, const Operand& operand); |
| 1753 |
| 1754 static bool IsImmFP32(float imm); |
| 1755 static bool IsImmFP64(double imm); |
| 1756 |
| 1757 // Find an appropriate LoadStoreOp or LoadStorePairOp for the specified |
| 1758 // registers. Only simple loads are supported; sign- and zero-extension (such |
| 1759 // as in LDPSW_x or LDRB_w) are not supported. |
| 1760 static inline LoadStoreOp LoadOpFor(const CPURegister& rt); |
| 1761 static inline LoadStorePairOp LoadPairOpFor(const CPURegister& rt, |
| 1762 const CPURegister& rt2); |
| 1763 static inline LoadStoreOp StoreOpFor(const CPURegister& rt); |
| 1764 static inline LoadStorePairOp StorePairOpFor(const CPURegister& rt, |
| 1765 const CPURegister& rt2); |
| 1766 static inline LoadStorePairNonTemporalOp LoadPairNonTemporalOpFor( |
| 1767 const CPURegister& rt, const CPURegister& rt2); |
| 1768 static inline LoadStorePairNonTemporalOp StorePairNonTemporalOpFor( |
| 1769 const CPURegister& rt, const CPURegister& rt2); |
| 1770 |
| 1771 |
| 1772 private: |
| 1773 // Instruction helpers. |
| 1774 void MoveWide(const Register& rd, |
| 1775 uint64_t imm, |
| 1776 int shift, |
| 1777 MoveWideImmediateOp mov_op); |
| 1778 void DataProcShiftedRegister(const Register& rd, |
| 1779 const Register& rn, |
| 1780 const Operand& operand, |
| 1781 FlagsUpdate S, |
| 1782 Instr op); |
| 1783 void DataProcExtendedRegister(const Register& rd, |
| 1784 const Register& rn, |
| 1785 const Operand& operand, |
| 1786 FlagsUpdate S, |
| 1787 Instr op); |
| 1788 void LoadStorePair(const CPURegister& rt, |
| 1789 const CPURegister& rt2, |
| 1790 const MemOperand& addr, |
| 1791 LoadStorePairOp op); |
| 1792 void LoadStorePairNonTemporal(const CPURegister& rt, |
| 1793 const CPURegister& rt2, |
| 1794 const MemOperand& addr, |
| 1795 LoadStorePairNonTemporalOp op); |
| 1796 // Register the relocation information for the operand and load its value |
| 1797 // into rt. |
| 1798 void LoadRelocatedValue(const CPURegister& rt, |
| 1799 const Operand& operand, |
| 1800 LoadLiteralOp op); |
| 1801 void ConditionalSelect(const Register& rd, |
| 1802 const Register& rn, |
| 1803 const Register& rm, |
| 1804 Condition cond, |
| 1805 ConditionalSelectOp op); |
| 1806 void DataProcessing1Source(const Register& rd, |
| 1807 const Register& rn, |
| 1808 DataProcessing1SourceOp op); |
| 1809 void DataProcessing3Source(const Register& rd, |
| 1810 const Register& rn, |
| 1811 const Register& rm, |
| 1812 const Register& ra, |
| 1813 DataProcessing3SourceOp op); |
| 1814 void FPDataProcessing1Source(const FPRegister& fd, |
| 1815 const FPRegister& fn, |
| 1816 FPDataProcessing1SourceOp op); |
| 1817 void FPDataProcessing2Source(const FPRegister& fd, |
| 1818 const FPRegister& fn, |
| 1819 const FPRegister& fm, |
| 1820 FPDataProcessing2SourceOp op); |
| 1821 void FPDataProcessing3Source(const FPRegister& fd, |
| 1822 const FPRegister& fn, |
| 1823 const FPRegister& fm, |
| 1824 const FPRegister& fa, |
| 1825 FPDataProcessing3SourceOp op); |
| 1826 |
| 1827 // Label helpers. |
| 1828 |
| 1829 // Return an offset for a label-referencing instruction, typically a branch. |
| 1830 int LinkAndGetByteOffsetTo(Label* label); |
| 1831 |
| 1832 // This is the same as LinkAndGetByteOffsetTo, but return an offset |
| 1833 // suitable for fields that take instruction offsets. |
| 1834 inline int LinkAndGetInstructionOffsetTo(Label* label); |
| 1835 |
| 1836 static const int kStartOfLabelLinkChain = 0; |
| 1837 |
| 1838 // Verify that a label's link chain is intact. |
| 1839 void CheckLabelLinkChain(Label const * label); |
| 1840 |
| 1841 void RecordLiteral(int64_t imm, unsigned size); |
| 1842 |
| 1843 // Postpone the generation of the constant pool for the specified number of |
| 1844 // instructions. |
| 1845 void BlockConstPoolFor(int instructions); |
| 1846 |
| 1847 // Emit the instruction at pc_. |
| 1848 inline void Emit(Instr instruction) { |
| 1849 STATIC_ASSERT(sizeof(*pc_) == 1); |
| 1850 STATIC_ASSERT(sizeof(instruction) == kInstructionSize); |
| 1851 ASSERT((pc_ + sizeof(instruction)) <= (buffer_ + buffer_size_)); |
| 1852 |
| 1853 #ifdef DEBUG |
| 1854 finalized_ = false; |
| 1855 #endif |
| 1856 |
| 1857 memcpy(pc_, &instruction, sizeof(instruction)); |
| 1858 pc_ += sizeof(instruction); |
| 1859 CheckBuffer(); |
| 1860 } |
| 1861 |
| 1862 // Emit data inline in the instruction stream. |
| 1863 void EmitData(void const * data, unsigned size) { |
| 1864 ASSERT(sizeof(*pc_) == 1); |
| 1865 ASSERT((pc_ + size) <= (buffer_ + buffer_size_)); |
| 1866 |
| 1867 #ifdef DEBUG |
| 1868 finalized_ = false; |
| 1869 #endif |
| 1870 |
| 1871 // TODO(all): Somehow register we have some data here. Then we can |
| 1872 // disassemble it correctly. |
| 1873 memcpy(pc_, data, size); |
| 1874 pc_ += size; |
| 1875 CheckBuffer(); |
| 1876 } |
| 1877 |
| 1878 void GrowBuffer(); |
| 1879 void CheckBuffer(); |
| 1880 |
| 1881 // Pc offset of the next buffer check. |
| 1882 int next_buffer_check_; |
| 1883 |
| 1884 // Constant pool generation |
| 1885 // Pools are emitted in the instruction stream, preferably after unconditional |
| 1886 // jumps or after returns from functions (in dead code locations). |
| 1887 // If a long code sequence does not contain unconditional jumps, it is |
| 1888 // necessary to emit the constant pool before the pool gets too far from the |
| 1889 // location it is accessed from. In this case, we emit a jump over the emitted |
| 1890 // constant pool. |
| 1891 // Constants in the pool may be addresses of functions that gets relocated; |
| 1892 // if so, a relocation info entry is associated to the constant pool entry. |
| 1893 |
| 1894 // Repeated checking whether the constant pool should be emitted is rather |
| 1895 // expensive. By default we only check again once a number of instructions |
| 1896 // has been generated. That also means that the sizing of the buffers is not |
| 1897 // an exact science, and that we rely on some slop to not overrun buffers. |
| 1898 static const int kCheckPoolIntervalInst = 128; |
| 1899 static const int kCheckPoolInterval = |
| 1900 kCheckPoolIntervalInst * kInstructionSize; |
| 1901 |
| 1902 // Constants in pools are accessed via pc relative addressing, which can |
| 1903 // reach +/-4KB thereby defining a maximum distance between the instruction |
| 1904 // and the accessed constant. |
| 1905 static const int kMaxDistToPool = 4 * KB; |
| 1906 static const int kMaxNumPendingRelocInfo = kMaxDistToPool / kInstructionSize; |
| 1907 |
| 1908 |
| 1909 // Average distance beetween a constant pool and the first instruction |
| 1910 // accessing the constant pool. Longer distance should result in less I-cache |
| 1911 // pollution. |
| 1912 // In practice the distance will be smaller since constant pool emission is |
| 1913 // forced after function return and sometimes after unconditional branches. |
| 1914 static const int kAvgDistToPool = kMaxDistToPool - kCheckPoolInterval; |
| 1915 |
| 1916 // Emission of the constant pool may be blocked in some code sequences. |
| 1917 int const_pool_blocked_nesting_; // Block emission if this is not zero. |
| 1918 int no_const_pool_before_; // Block emission before this pc offset. |
| 1919 |
| 1920 // Keep track of the first instruction requiring a constant pool entry |
| 1921 // since the previous constant pool was emitted. |
| 1922 int first_const_pool_use_; |
| 1923 |
| 1924 // Relocation info generation |
| 1925 // Each relocation is encoded as a variable size value |
| 1926 static const int kMaxRelocSize = RelocInfoWriter::kMaxSize; |
| 1927 RelocInfoWriter reloc_info_writer; |
| 1928 |
| 1929 // Relocation info records are also used during code generation as temporary |
| 1930 // containers for constants and code target addresses until they are emitted |
| 1931 // to the constant pool. These pending relocation info records are temporarily |
| 1932 // stored in a separate buffer until a constant pool is emitted. |
| 1933 // If every instruction in a long sequence is accessing the pool, we need one |
| 1934 // pending relocation entry per instruction. |
| 1935 |
| 1936 // the buffer of pending relocation info |
| 1937 RelocInfo pending_reloc_info_[kMaxNumPendingRelocInfo]; |
| 1938 // number of pending reloc info entries in the buffer |
| 1939 int num_pending_reloc_info_; |
| 1940 |
| 1941 // Relocation for a type-recording IC has the AST id added to it. This |
| 1942 // member variable is a way to pass the information from the call site to |
| 1943 // the relocation info. |
| 1944 TypeFeedbackId recorded_ast_id_; |
| 1945 |
| 1946 inline TypeFeedbackId RecordedAstId(); |
| 1947 inline void ClearRecordedAstId(); |
| 1948 |
| 1949 protected: |
| 1950 // Record the AST id of the CallIC being compiled, so that it can be placed |
| 1951 // in the relocation information. |
| 1952 void SetRecordedAstId(TypeFeedbackId ast_id) { |
| 1953 ASSERT(recorded_ast_id_.IsNone()); |
| 1954 recorded_ast_id_ = ast_id; |
| 1955 } |
| 1956 |
| 1957 // Code generation |
| 1958 // The relocation writer's position is at least kGap bytes below the end of |
| 1959 // the generated instructions. This is so that multi-instruction sequences do |
| 1960 // not have to check for overflow. The same is true for writes of large |
| 1961 // relocation info entries, and debug strings encoded in the instruction |
| 1962 // stream. |
| 1963 static const int kGap = 128; |
| 1964 |
| 1965 #ifdef DEBUG |
| 1966 bool finalized_; |
| 1967 #endif |
| 1968 |
| 1969 private: |
| 1970 // TODO(jbramley): [GOOGJSE-238] W16 uses next_literal_pool_check_ and |
| 1971 // literal_pool_monitor_ to determine when to consider emitting a literal |
| 1972 // pool. V8 doesn't use them, so they should either not be here at all, or |
| 1973 // should replace or be merged with next_buffer_check_ and |
| 1974 // const_pool_blocked_nesting_. |
| 1975 Instruction* next_literal_pool_check_; |
| 1976 unsigned literal_pool_monitor_; |
| 1977 |
| 1978 PositionsRecorder positions_recorder_; |
| 1979 friend class PositionsRecorder; |
| 1980 friend class EnsureSpace; |
| 1981 }; |
| 1982 |
| 1983 class PatchingAssembler : public Assembler { |
| 1984 public: |
| 1985 // Create an Assembler with a buffer starting at 'start'. |
| 1986 // The buffer size is |
| 1987 // size of instructions to patch + kGap |
| 1988 // Where kGap is the distance from which the Assembler tries to grow the |
| 1989 // buffer. |
| 1990 // If more or fewer instructions than expected are generated or if some |
| 1991 // relocation information takes space in the buffer, the PatchingAssembler |
| 1992 // will crash trying to grow the buffer. |
| 1993 PatchingAssembler(Instruction* start, unsigned count) |
| 1994 : Assembler(NULL, |
| 1995 reinterpret_cast<byte*>(start), |
| 1996 count * kInstructionSize + kGap) { |
| 1997 // Block constant pool emission. |
| 1998 StartBlockConstPool(); |
| 1999 } |
| 2000 |
| 2001 PatchingAssembler(byte* start, unsigned count) |
| 2002 : Assembler(NULL, start, count * kInstructionSize + kGap) { |
| 2003 // Block constant pool emission. |
| 2004 StartBlockConstPool(); |
| 2005 } |
| 2006 |
| 2007 ~PatchingAssembler() { |
| 2008 // Const pool should still be blocked. |
| 2009 ASSERT(is_const_pool_blocked()); |
| 2010 EndBlockConstPool(); |
| 2011 // Verify we have generated the number of instruction we expected. |
| 2012 ASSERT((pc_offset() + kGap) == buffer_size_); |
| 2013 // Verify no relocation information has been emitted. |
| 2014 ASSERT(num_pending_reloc_info() == 0); |
| 2015 // Flush the Instruction cache. |
| 2016 size_t length = buffer_size_ - kGap; |
| 2017 CPU::FlushICache(buffer_, length); |
| 2018 |
| 2019 #ifdef DEBUG |
| 2020 // The Patching Assembler doesn't need to be finalized, but the Assembler |
| 2021 // class's destructor will check the finalized_ flag. |
| 2022 finalized_ = true; |
| 2023 #endif |
| 2024 } |
| 2025 }; |
| 2026 |
| 2027 |
| 2028 class EnsureSpace BASE_EMBEDDED { |
| 2029 public: |
| 2030 explicit EnsureSpace(Assembler* assembler) { |
| 2031 assembler->CheckBuffer(); |
| 2032 } |
| 2033 }; |
| 2034 |
| 2035 } } // namespace v8::internal |
| 2036 |
| 2037 #endif // V8_A64_ASSEMBLER_A64_H_ |
| OLD | NEW |