1 // Copyright (c) 1994-2006 Sun Microsystems Inc. 2 // All Rights Reserved. 3 // 4 // Redistribution and use in source and binary forms, with or without 5 // modification, are permitted provided that the following conditions are 6 // met: 7 // 8 // - Redistributions of source code must retain the above copyright notice, 9 // this list of conditions and the following disclaimer. 10 // 11 // - Redistribution in binary form must reproduce the above copyright 12 // notice, this list of conditions and the following disclaimer in the 13 // documentation and/or other materials provided with the distribution. 14 // 15 // - Neither the name of Sun Microsystems or the names of contributors may 16 // be used to endorse or promote products derived from this software without 17 // specific prior written permission. 18 // 19 // THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS 20 // IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, 21 // THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 22 // PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR 23 // CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, 24 // EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, 25 // PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR 26 // PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF 27 // LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING 28 // NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS 29 // SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 30 31 // The original source code covered by the above license above has been 32 // modified significantly by Google Inc. 33 // Copyright 2012 the V8 project authors. All rights reserved. 34 35 36 #ifndef V8_MIPS_ASSEMBLER_MIPS_H_ 37 #define V8_MIPS_ASSEMBLER_MIPS_H_ 38 39 #include <stdio.h> 40 41 #include <set> 42 43 #include "src/assembler.h" 44 #include "src/mips64/constants-mips64.h" 45 46 namespace v8 { 47 namespace internal { 48 49 // clang-format off 50 #define GENERAL_REGISTERS(V) \ 51 V(zero_reg) V(at) V(v0) V(v1) V(a0) V(a1) V(a2) V(a3) \ 52 V(a4) V(a5) V(a6) V(a7) V(t0) V(t1) V(t2) V(t3) \ 53 V(s0) V(s1) V(s2) V(s3) V(s4) V(s5) V(s6) V(s7) V(t8) V(t9) \ 54 V(k0) V(k1) V(gp) V(sp) V(fp) V(ra) 55 56 #define ALLOCATABLE_GENERAL_REGISTERS(V) \ 57 V(v0) V(v1) V(a0) V(a1) V(a2) V(a3) \ 58 V(a4) V(a5) V(a6) V(a7) V(t0) V(t1) V(t2) V(s7) 59 60 #define DOUBLE_REGISTERS(V) \ 61 V(f0) V(f1) V(f2) V(f3) V(f4) V(f5) V(f6) V(f7) \ 62 V(f8) V(f9) V(f10) V(f11) V(f12) V(f13) V(f14) V(f15) \ 63 V(f16) V(f17) V(f18) V(f19) V(f20) V(f21) V(f22) V(f23) \ 64 V(f24) V(f25) V(f26) V(f27) V(f28) V(f29) V(f30) V(f31) 65 66 #define ALLOCATABLE_DOUBLE_REGISTERS(V) \ 67 V(f0) V(f2) V(f4) V(f6) V(f8) V(f10) V(f12) V(f14) \ 68 V(f16) V(f18) V(f20) V(f22) V(f24) V(f26) 69 // clang-format on 70 71 // CPU Registers. 72 // 73 // 1) We would prefer to use an enum, but enum values are assignment- 74 // compatible with int, which has caused code-generation bugs. 75 // 76 // 2) We would prefer to use a class instead of a struct but we don't like 77 // the register initialization to depend on the particular initialization 78 // order (which appears to be different on OS X, Linux, and Windows for the 79 // installed versions of C++ we tried). Using a struct permits C-style 80 // "initialization". Also, the Register objects cannot be const as this 81 // forces initialization stubs in MSVC, making us dependent on initialization 82 // order. 83 // 84 // 3) By not using an enum, we are possibly preventing the compiler from 85 // doing certain constant folds, which may significantly reduce the 86 // code generated for some assembly instructions (because they boil down 87 // to a few constants). If this is a problem, we could change the code 88 // such that we use an enum in optimized mode, and the struct in debug 89 // mode. This way we get the compile-time error checking in debug mode 90 // and best performance in optimized code. 91 92 93 // ----------------------------------------------------------------------------- 94 // Implementation of Register and FPURegister. 95 96 struct Register { 97 static const int kCpRegister = 23; // cp (s7) is the 23rd register. 98 99 #if defined(V8_TARGET_LITTLE_ENDIAN) 100 static const int kMantissaOffset = 0; 101 static const int kExponentOffset = 4; 102 #elif defined(V8_TARGET_BIG_ENDIAN) 103 static const int kMantissaOffset = 4; 104 static const int kExponentOffset = 0; 105 #else 106 #error Unknown endianness 107 #endif 108 109 enum Code { 110 #define REGISTER_CODE(R) kCode_##R, 111 GENERAL_REGISTERS(REGISTER_CODE) 112 #undef REGISTER_CODE 113 kAfterLast, 114 kCode_no_reg = -1 115 }; 116 117 static const int kNumRegisters = Code::kAfterLast; 118 from_codeRegister119 static Register from_code(int code) { 120 DCHECK(code >= 0); 121 DCHECK(code < kNumRegisters); 122 Register r = { code }; 123 return r; 124 } 125 126 const char* ToString(); 127 bool IsAllocatable() const; is_validRegister128 bool is_valid() const { return 0 <= reg_code && reg_code < kNumRegisters; } isRegister129 bool is(Register reg) const { return reg_code == reg.reg_code; } codeRegister130 int code() const { 131 DCHECK(is_valid()); 132 return reg_code; 133 } bitRegister134 int bit() const { 135 DCHECK(is_valid()); 136 return 1 << reg_code; 137 } 138 139 // Unfortunately we can't make this private in a struct. 140 int reg_code; 141 }; 142 143 // s7: context register 144 // s3: lithium scratch 145 // s4: lithium scratch2 146 #define DECLARE_REGISTER(R) const Register R = {Register::kCode_##R}; 147 GENERAL_REGISTERS(DECLARE_REGISTER) 148 #undef DECLARE_REGISTER 149 const Register no_reg = {Register::kCode_no_reg}; 150 151 152 int ToNumber(Register reg); 153 154 Register ToRegister(int num); 155 156 // Coprocessor register. 157 struct DoubleRegister { 158 enum Code { 159 #define REGISTER_CODE(R) kCode_##R, 160 DOUBLE_REGISTERS(REGISTER_CODE) 161 #undef REGISTER_CODE 162 kAfterLast, 163 kCode_no_reg = -1 164 }; 165 166 static const int kMaxNumRegisters = Code::kAfterLast; 167 168 inline static int NumRegisters(); 169 170 // TODO(plind): Warning, inconsistent numbering here. kNumFPURegisters refers 171 // to number of 32-bit FPU regs, but kNumAllocatableRegisters refers to 172 // number of Double regs (64-bit regs, or FPU-reg-pairs). 173 174 const char* ToString(); 175 bool IsAllocatable() const; is_validDoubleRegister176 bool is_valid() const { return 0 <= reg_code && reg_code < kMaxNumRegisters; } isDoubleRegister177 bool is(DoubleRegister reg) const { return reg_code == reg.reg_code; } lowDoubleRegister178 DoubleRegister low() const { 179 // TODO(plind): Create DCHECK for FR=0 mode. This usage suspect for FR=1. 180 // Find low reg of a Double-reg pair, which is the reg itself. 181 DCHECK(reg_code % 2 == 0); // Specified Double reg must be even. 182 DoubleRegister reg; 183 reg.reg_code = reg_code; 184 DCHECK(reg.is_valid()); 185 return reg; 186 } highDoubleRegister187 DoubleRegister high() const { 188 // TODO(plind): Create DCHECK for FR=0 mode. This usage illegal in FR=1. 189 // Find high reg of a Doubel-reg pair, which is reg + 1. 190 DCHECK(reg_code % 2 == 0); // Specified Double reg must be even. 191 DoubleRegister reg; 192 reg.reg_code = reg_code + 1; 193 DCHECK(reg.is_valid()); 194 return reg; 195 } 196 codeDoubleRegister197 int code() const { 198 DCHECK(is_valid()); 199 return reg_code; 200 } bitDoubleRegister201 int bit() const { 202 DCHECK(is_valid()); 203 return 1 << reg_code; 204 } 205 from_codeDoubleRegister206 static DoubleRegister from_code(int code) { 207 DoubleRegister r = {code}; 208 return r; 209 } setcodeDoubleRegister210 void setcode(int f) { 211 reg_code = f; 212 DCHECK(is_valid()); 213 } 214 // Unfortunately we can't make this private in a struct. 215 int reg_code; 216 }; 217 218 // A few double registers are reserved: one as a scratch register and one to 219 // hold 0.0. 220 // f28: 0.0 221 // f30: scratch register. 222 223 // V8 now supports the O32 ABI, and the FPU Registers are organized as 32 224 // 32-bit registers, f0 through f31. When used as 'double' they are used 225 // in pairs, starting with the even numbered register. So a double operation 226 // on f0 really uses f0 and f1. 227 // (Modern mips hardware also supports 32 64-bit registers, via setting 228 // (privileged) Status Register FR bit to 1. This is used by the N32 ABI, 229 // but it is not in common use. Someday we will want to support this in v8.) 230 231 // For O32 ABI, Floats and Doubles refer to same set of 32 32-bit registers. 232 typedef DoubleRegister FPURegister; 233 typedef DoubleRegister FloatRegister; 234 235 const DoubleRegister no_freg = {-1}; 236 237 const DoubleRegister f0 = {0}; // Return value in hard float mode. 238 const DoubleRegister f1 = {1}; 239 const DoubleRegister f2 = {2}; 240 const DoubleRegister f3 = {3}; 241 const DoubleRegister f4 = {4}; 242 const DoubleRegister f5 = {5}; 243 const DoubleRegister f6 = {6}; 244 const DoubleRegister f7 = {7}; 245 const DoubleRegister f8 = {8}; 246 const DoubleRegister f9 = {9}; 247 const DoubleRegister f10 = {10}; 248 const DoubleRegister f11 = {11}; 249 const DoubleRegister f12 = {12}; // Arg 0 in hard float mode. 250 const DoubleRegister f13 = {13}; 251 const DoubleRegister f14 = {14}; // Arg 1 in hard float mode. 252 const DoubleRegister f15 = {15}; 253 const DoubleRegister f16 = {16}; 254 const DoubleRegister f17 = {17}; 255 const DoubleRegister f18 = {18}; 256 const DoubleRegister f19 = {19}; 257 const DoubleRegister f20 = {20}; 258 const DoubleRegister f21 = {21}; 259 const DoubleRegister f22 = {22}; 260 const DoubleRegister f23 = {23}; 261 const DoubleRegister f24 = {24}; 262 const DoubleRegister f25 = {25}; 263 const DoubleRegister f26 = {26}; 264 const DoubleRegister f27 = {27}; 265 const DoubleRegister f28 = {28}; 266 const DoubleRegister f29 = {29}; 267 const DoubleRegister f30 = {30}; 268 const DoubleRegister f31 = {31}; 269 270 // Register aliases. 271 // cp is assumed to be a callee saved register. 272 // Defined using #define instead of "static const Register&" because Clang 273 // complains otherwise when a compilation unit that includes this header 274 // doesn't use the variables. 275 #define kRootRegister s6 276 #define cp s7 277 #define kLithiumScratchReg s3 278 #define kLithiumScratchReg2 s4 279 #define kLithiumScratchDouble f30 280 #define kDoubleRegZero f28 281 // Used on mips64r6 for compare operations. 282 // We use the last non-callee saved odd register for N64 ABI 283 #define kDoubleCompareReg f23 284 285 // FPU (coprocessor 1) control registers. 286 // Currently only FCSR (#31) is implemented. 287 struct FPUControlRegister { is_validFPUControlRegister288 bool is_valid() const { return reg_code == kFCSRRegister; } isFPUControlRegister289 bool is(FPUControlRegister creg) const { return reg_code == creg.reg_code; } codeFPUControlRegister290 int code() const { 291 DCHECK(is_valid()); 292 return reg_code; 293 } bitFPUControlRegister294 int bit() const { 295 DCHECK(is_valid()); 296 return 1 << reg_code; 297 } setcodeFPUControlRegister298 void setcode(int f) { 299 reg_code = f; 300 DCHECK(is_valid()); 301 } 302 // Unfortunately we can't make this private in a struct. 303 int reg_code; 304 }; 305 306 const FPUControlRegister no_fpucreg = { kInvalidFPUControlRegister }; 307 const FPUControlRegister FCSR = { kFCSRRegister }; 308 309 310 // ----------------------------------------------------------------------------- 311 // Machine instruction Operands. 312 const int kSmiShift = kSmiTagSize + kSmiShiftSize; 313 const uint64_t kSmiShiftMask = (1UL << kSmiShift) - 1; 314 // Class Operand represents a shifter operand in data processing instructions. 315 class Operand BASE_EMBEDDED { 316 public: 317 // Immediate. 318 INLINE(explicit Operand(int64_t immediate, 319 RelocInfo::Mode rmode = RelocInfo::NONE64)); 320 INLINE(explicit Operand(const ExternalReference& f)); 321 INLINE(explicit Operand(const char* s)); 322 INLINE(explicit Operand(Object** opp)); 323 INLINE(explicit Operand(Context** cpp)); 324 explicit Operand(Handle<Object> handle); 325 INLINE(explicit Operand(Smi* value)); 326 327 // Register. 328 INLINE(explicit Operand(Register rm)); 329 330 // Return true if this is a register operand. 331 INLINE(bool is_reg() const); 332 immediate()333 inline int64_t immediate() const { 334 DCHECK(!is_reg()); 335 return imm64_; 336 } 337 rm()338 Register rm() const { return rm_; } 339 340 private: 341 Register rm_; 342 int64_t imm64_; // Valid if rm_ == no_reg. 343 RelocInfo::Mode rmode_; 344 345 friend class Assembler; 346 friend class MacroAssembler; 347 }; 348 349 350 // On MIPS we have only one adressing mode with base_reg + offset. 351 // Class MemOperand represents a memory operand in load and store instructions. 352 class MemOperand : public Operand { 353 public: 354 // Immediate value attached to offset. 355 enum OffsetAddend { 356 offset_minus_one = -1, 357 offset_zero = 0 358 }; 359 360 explicit MemOperand(Register rn, int32_t offset = 0); 361 explicit MemOperand(Register rn, int32_t unit, int32_t multiplier, 362 OffsetAddend offset_addend = offset_zero); offset()363 int32_t offset() const { return offset_; } 364 OffsetIsInt16Encodable()365 bool OffsetIsInt16Encodable() const { 366 return is_int16(offset_); 367 } 368 369 private: 370 int32_t offset_; 371 372 friend class Assembler; 373 }; 374 375 376 class Assembler : public AssemblerBase { 377 public: 378 // Create an assembler. Instructions and relocation information are emitted 379 // into a buffer, with the instructions starting from the beginning and the 380 // relocation information starting from the end of the buffer. See CodeDesc 381 // for a detailed comment on the layout (globals.h). 382 // 383 // If the provided buffer is NULL, the assembler allocates and grows its own 384 // buffer, and buffer_size determines the initial buffer size. The buffer is 385 // owned by the assembler and deallocated upon destruction of the assembler. 386 // 387 // If the provided buffer is not NULL, the assembler uses the provided buffer 388 // for code generation and assumes its size to be buffer_size. If the buffer 389 // is too small, a fatal error occurs. No deallocation of the buffer is done 390 // upon destruction of the assembler. 391 Assembler(Isolate* isolate, void* buffer, int buffer_size); ~Assembler()392 virtual ~Assembler() { } 393 394 // GetCode emits any pending (non-emitted) code and fills the descriptor 395 // desc. GetCode() is idempotent; it returns the same result if no other 396 // Assembler functions are invoked in between GetCode() calls. 397 void GetCode(CodeDesc* desc); 398 399 // Label operations & relative jumps (PPUM Appendix D). 400 // 401 // Takes a branch opcode (cc) and a label (L) and generates 402 // either a backward branch or a forward branch and links it 403 // to the label fixup chain. Usage: 404 // 405 // Label L; // unbound label 406 // j(cc, &L); // forward branch to unbound label 407 // bind(&L); // bind label to the current pc 408 // j(cc, &L); // backward branch to bound label 409 // bind(&L); // illegal: a label may be bound only once 410 // 411 // Note: The same Label can be used for forward and backward branches 412 // but it may be bound only once. 413 void bind(Label* L); // Binds an unbound label L to current code position. 414 415 enum OffsetSize : int { kOffset26 = 26, kOffset21 = 21, kOffset16 = 16 }; 416 417 // Determines if Label is bound and near enough so that branch instruction 418 // can be used to reach it, instead of jump instruction. 419 bool is_near(Label* L); 420 bool is_near(Label* L, OffsetSize bits); 421 bool is_near_branch(Label* L); is_near_pre_r6(Label * L)422 inline bool is_near_pre_r6(Label* L) { 423 DCHECK(!(kArchVariant == kMips64r6)); 424 return pc_offset() - L->pos() < kMaxBranchOffset - 4 * kInstrSize; 425 } is_near_r6(Label * L)426 inline bool is_near_r6(Label* L) { 427 DCHECK(kArchVariant == kMips64r6); 428 return pc_offset() - L->pos() < kMaxCompactBranchOffset - 4 * kInstrSize; 429 } 430 431 int BranchOffset(Instr instr); 432 433 // Returns the branch offset to the given label from the current code 434 // position. Links the label to the current position if it is still unbound. 435 // Manages the jump elimination optimization if the second parameter is true. 436 int32_t branch_offset_helper(Label* L, OffsetSize bits); branch_offset(Label * L)437 inline int32_t branch_offset(Label* L) { 438 return branch_offset_helper(L, OffsetSize::kOffset16); 439 } branch_offset21(Label * L)440 inline int32_t branch_offset21(Label* L) { 441 return branch_offset_helper(L, OffsetSize::kOffset21); 442 } branch_offset26(Label * L)443 inline int32_t branch_offset26(Label* L) { 444 return branch_offset_helper(L, OffsetSize::kOffset26); 445 } shifted_branch_offset(Label * L)446 inline int32_t shifted_branch_offset(Label* L) { 447 return branch_offset(L) >> 2; 448 } shifted_branch_offset21(Label * L)449 inline int32_t shifted_branch_offset21(Label* L) { 450 return branch_offset21(L) >> 2; 451 } shifted_branch_offset26(Label * L)452 inline int32_t shifted_branch_offset26(Label* L) { 453 return branch_offset26(L) >> 2; 454 } 455 uint64_t jump_address(Label* L); 456 uint64_t jump_offset(Label* L); 457 458 // Puts a labels target address at the given position. 459 // The high 8 bits are set to zero. 460 void label_at_put(Label* L, int at_offset); 461 462 // Read/Modify the code target address in the branch/call instruction at pc. 463 static Address target_address_at(Address pc); 464 static void set_target_address_at( 465 Isolate* isolate, Address pc, Address target, 466 ICacheFlushMode icache_flush_mode = FLUSH_ICACHE_IF_NEEDED); 467 // On MIPS there is no Constant Pool so we skip that parameter. INLINE(static Address target_address_at (Address pc,Address constant_pool))468 INLINE(static Address target_address_at(Address pc, Address constant_pool)) { 469 return target_address_at(pc); 470 } INLINE(static void set_target_address_at (Isolate * isolate,Address pc,Address constant_pool,Address target,ICacheFlushMode icache_flush_mode=FLUSH_ICACHE_IF_NEEDED))471 INLINE(static void set_target_address_at( 472 Isolate* isolate, Address pc, Address constant_pool, Address target, 473 ICacheFlushMode icache_flush_mode = FLUSH_ICACHE_IF_NEEDED)) { 474 set_target_address_at(isolate, pc, target, icache_flush_mode); 475 } INLINE(static Address target_address_at (Address pc,Code * code))476 INLINE(static Address target_address_at(Address pc, Code* code)) { 477 Address constant_pool = code ? code->constant_pool() : NULL; 478 return target_address_at(pc, constant_pool); 479 } INLINE(static void set_target_address_at (Isolate * isolate,Address pc,Code * code,Address target,ICacheFlushMode icache_flush_mode=FLUSH_ICACHE_IF_NEEDED))480 INLINE(static void set_target_address_at( 481 Isolate* isolate, Address pc, Code* code, Address target, 482 ICacheFlushMode icache_flush_mode = FLUSH_ICACHE_IF_NEEDED)) { 483 Address constant_pool = code ? code->constant_pool() : NULL; 484 set_target_address_at(isolate, pc, constant_pool, target, 485 icache_flush_mode); 486 } 487 488 // Return the code target address at a call site from the return address 489 // of that call in the instruction stream. 490 inline static Address target_address_from_return_address(Address pc); 491 492 static void JumpLabelToJumpRegister(Address pc); 493 494 static void QuietNaN(HeapObject* nan); 495 496 // This sets the branch destination (which gets loaded at the call address). 497 // This is for calls and branches within generated code. The serializer 498 // has already deserialized the lui/ori instructions etc. deserialization_set_special_target_at(Isolate * isolate,Address instruction_payload,Code * code,Address target)499 inline static void deserialization_set_special_target_at( 500 Isolate* isolate, Address instruction_payload, Code* code, 501 Address target) { 502 set_target_address_at( 503 isolate, 504 instruction_payload - kInstructionsFor64BitConstant * kInstrSize, code, 505 target); 506 } 507 508 // This sets the internal reference at the pc. 509 inline static void deserialization_set_target_internal_reference_at( 510 Isolate* isolate, Address pc, Address target, 511 RelocInfo::Mode mode = RelocInfo::INTERNAL_REFERENCE); 512 513 // Size of an instruction. 514 static const int kInstrSize = sizeof(Instr); 515 516 // Difference between address of current opcode and target address offset. 517 static const int kBranchPCOffset = 4; 518 519 // Here we are patching the address in the LUI/ORI instruction pair. 520 // These values are used in the serialization process and must be zero for 521 // MIPS platform, as Code, Embedded Object or External-reference pointers 522 // are split across two consecutive instructions and don't exist separately 523 // in the code, so the serializer should not step forwards in memory after 524 // a target is resolved and written. 525 static const int kSpecialTargetSize = 0; 526 527 // Number of consecutive instructions used to store 32bit/64bit constant. 528 // This constant was used in RelocInfo::target_address_address() function 529 // to tell serializer address of the instruction that follows 530 // LUI/ORI instruction pair. 531 static const int kInstructionsFor32BitConstant = 2; 532 static const int kInstructionsFor64BitConstant = 4; 533 534 // Distance between the instruction referring to the address of the call 535 // target and the return address. 536 static const int kCallTargetAddressOffset = 6 * kInstrSize; 537 538 // Distance between start of patched debug break slot and the emitted address 539 // to jump to. 540 static const int kPatchDebugBreakSlotAddressOffset = 6 * kInstrSize; 541 542 // Difference between address of current opcode and value read from pc 543 // register. 544 static const int kPcLoadDelta = 4; 545 546 static const int kDebugBreakSlotInstructions = 6; 547 static const int kDebugBreakSlotLength = 548 kDebugBreakSlotInstructions * kInstrSize; 549 550 551 // --------------------------------------------------------------------------- 552 // Code generation. 553 554 // Insert the smallest number of nop instructions 555 // possible to align the pc offset to a multiple 556 // of m. m must be a power of 2 (>= 4). 557 void Align(int m); 558 // Insert the smallest number of zero bytes possible to align the pc offset 559 // to a mulitple of m. m must be a power of 2 (>= 2). 560 void DataAlign(int m); 561 // Aligns code to something that's optimal for a jump target for the platform. 562 void CodeTargetAlign(); 563 564 // Different nop operations are used by the code generator to detect certain 565 // states of the generated code. 566 enum NopMarkerTypes { 567 NON_MARKING_NOP = 0, 568 DEBUG_BREAK_NOP, 569 // IC markers. 570 PROPERTY_ACCESS_INLINED, 571 PROPERTY_ACCESS_INLINED_CONTEXT, 572 PROPERTY_ACCESS_INLINED_CONTEXT_DONT_DELETE, 573 // Helper values. 574 LAST_CODE_MARKER, 575 FIRST_IC_MARKER = PROPERTY_ACCESS_INLINED, 576 // Code aging 577 CODE_AGE_MARKER_NOP = 6, 578 CODE_AGE_SEQUENCE_NOP 579 }; 580 581 // Type == 0 is the default non-marking nop. For mips this is a 582 // sll(zero_reg, zero_reg, 0). We use rt_reg == at for non-zero 583 // marking, to avoid conflict with ssnop and ehb instructions. 584 void nop(unsigned int type = 0) { 585 DCHECK(type < 32); 586 Register nop_rt_reg = (type == 0) ? zero_reg : at; 587 sll(zero_reg, nop_rt_reg, type, true); 588 } 589 590 591 // --------Branch-and-jump-instructions---------- 592 // We don't use likely variant of instructions. 593 void b(int16_t offset); b(Label * L)594 inline void b(Label* L) { b(shifted_branch_offset(L)); } 595 void bal(int16_t offset); bal(Label * L)596 inline void bal(Label* L) { bal(shifted_branch_offset(L)); } 597 void bc(int32_t offset); bc(Label * L)598 inline void bc(Label* L) { bc(shifted_branch_offset26(L)); } 599 void balc(int32_t offset); balc(Label * L)600 inline void balc(Label* L) { balc(shifted_branch_offset26(L)); } 601 602 void beq(Register rs, Register rt, int16_t offset); beq(Register rs,Register rt,Label * L)603 inline void beq(Register rs, Register rt, Label* L) { 604 beq(rs, rt, shifted_branch_offset(L)); 605 } 606 void bgez(Register rs, int16_t offset); 607 void bgezc(Register rt, int16_t offset); bgezc(Register rt,Label * L)608 inline void bgezc(Register rt, Label* L) { 609 bgezc(rt, shifted_branch_offset(L)); 610 } 611 void bgeuc(Register rs, Register rt, int16_t offset); bgeuc(Register rs,Register rt,Label * L)612 inline void bgeuc(Register rs, Register rt, Label* L) { 613 bgeuc(rs, rt, shifted_branch_offset(L)); 614 } 615 void bgec(Register rs, Register rt, int16_t offset); bgec(Register rs,Register rt,Label * L)616 inline void bgec(Register rs, Register rt, Label* L) { 617 bgec(rs, rt, shifted_branch_offset(L)); 618 } 619 void bgezal(Register rs, int16_t offset); 620 void bgezalc(Register rt, int16_t offset); bgezalc(Register rt,Label * L)621 inline void bgezalc(Register rt, Label* L) { 622 bgezalc(rt, shifted_branch_offset(L)); 623 } 624 void bgezall(Register rs, int16_t offset); bgezall(Register rs,Label * L)625 inline void bgezall(Register rs, Label* L) { 626 bgezall(rs, branch_offset(L) >> 2); 627 } 628 void bgtz(Register rs, int16_t offset); 629 void bgtzc(Register rt, int16_t offset); bgtzc(Register rt,Label * L)630 inline void bgtzc(Register rt, Label* L) { 631 bgtzc(rt, shifted_branch_offset(L)); 632 } 633 void blez(Register rs, int16_t offset); 634 void blezc(Register rt, int16_t offset); blezc(Register rt,Label * L)635 inline void blezc(Register rt, Label* L) { 636 blezc(rt, shifted_branch_offset(L)); 637 } 638 void bltz(Register rs, int16_t offset); 639 void bltzc(Register rt, int16_t offset); bltzc(Register rt,Label * L)640 inline void bltzc(Register rt, Label* L) { 641 bltzc(rt, shifted_branch_offset(L)); 642 } 643 void bltuc(Register rs, Register rt, int16_t offset); bltuc(Register rs,Register rt,Label * L)644 inline void bltuc(Register rs, Register rt, Label* L) { 645 bltuc(rs, rt, shifted_branch_offset(L)); 646 } 647 void bltc(Register rs, Register rt, int16_t offset); bltc(Register rs,Register rt,Label * L)648 inline void bltc(Register rs, Register rt, Label* L) { 649 bltc(rs, rt, shifted_branch_offset(L)); 650 } 651 void bltzal(Register rs, int16_t offset); 652 void blezalc(Register rt, int16_t offset); blezalc(Register rt,Label * L)653 inline void blezalc(Register rt, Label* L) { 654 blezalc(rt, shifted_branch_offset(L)); 655 } 656 void bltzalc(Register rt, int16_t offset); bltzalc(Register rt,Label * L)657 inline void bltzalc(Register rt, Label* L) { 658 bltzalc(rt, shifted_branch_offset(L)); 659 } 660 void bgtzalc(Register rt, int16_t offset); bgtzalc(Register rt,Label * L)661 inline void bgtzalc(Register rt, Label* L) { 662 bgtzalc(rt, shifted_branch_offset(L)); 663 } 664 void beqzalc(Register rt, int16_t offset); beqzalc(Register rt,Label * L)665 inline void beqzalc(Register rt, Label* L) { 666 beqzalc(rt, shifted_branch_offset(L)); 667 } 668 void beqc(Register rs, Register rt, int16_t offset); beqc(Register rs,Register rt,Label * L)669 inline void beqc(Register rs, Register rt, Label* L) { 670 beqc(rs, rt, shifted_branch_offset(L)); 671 } 672 void beqzc(Register rs, int32_t offset); beqzc(Register rs,Label * L)673 inline void beqzc(Register rs, Label* L) { 674 beqzc(rs, shifted_branch_offset21(L)); 675 } 676 void bnezalc(Register rt, int16_t offset); bnezalc(Register rt,Label * L)677 inline void bnezalc(Register rt, Label* L) { 678 bnezalc(rt, shifted_branch_offset(L)); 679 } 680 void bnec(Register rs, Register rt, int16_t offset); bnec(Register rs,Register rt,Label * L)681 inline void bnec(Register rs, Register rt, Label* L) { 682 bnec(rs, rt, shifted_branch_offset(L)); 683 } 684 void bnezc(Register rt, int32_t offset); bnezc(Register rt,Label * L)685 inline void bnezc(Register rt, Label* L) { 686 bnezc(rt, shifted_branch_offset21(L)); 687 } 688 void bne(Register rs, Register rt, int16_t offset); bne(Register rs,Register rt,Label * L)689 inline void bne(Register rs, Register rt, Label* L) { 690 bne(rs, rt, shifted_branch_offset(L)); 691 } 692 void bovc(Register rs, Register rt, int16_t offset); bovc(Register rs,Register rt,Label * L)693 inline void bovc(Register rs, Register rt, Label* L) { 694 bovc(rs, rt, shifted_branch_offset(L)); 695 } 696 void bnvc(Register rs, Register rt, int16_t offset); bnvc(Register rs,Register rt,Label * L)697 inline void bnvc(Register rs, Register rt, Label* L) { 698 bnvc(rs, rt, shifted_branch_offset(L)); 699 } 700 701 // Never use the int16_t b(l)cond version with a branch offset 702 // instead of using the Label* version. 703 704 // Jump targets must be in the current 256 MB-aligned region. i.e. 28 bits. 705 void j(int64_t target); 706 void jal(int64_t target); 707 void j(Label* target); 708 void jal(Label* target); 709 void jalr(Register rs, Register rd = ra); 710 void jr(Register target); 711 void jic(Register rt, int16_t offset); 712 void jialc(Register rt, int16_t offset); 713 714 715 // -------Data-processing-instructions--------- 716 717 // Arithmetic. 718 void addu(Register rd, Register rs, Register rt); 719 void subu(Register rd, Register rs, Register rt); 720 721 void div(Register rs, Register rt); 722 void divu(Register rs, Register rt); 723 void ddiv(Register rs, Register rt); 724 void ddivu(Register rs, Register rt); 725 void div(Register rd, Register rs, Register rt); 726 void divu(Register rd, Register rs, Register rt); 727 void ddiv(Register rd, Register rs, Register rt); 728 void ddivu(Register rd, Register rs, Register rt); 729 void mod(Register rd, Register rs, Register rt); 730 void modu(Register rd, Register rs, Register rt); 731 void dmod(Register rd, Register rs, Register rt); 732 void dmodu(Register rd, Register rs, Register rt); 733 734 void mul(Register rd, Register rs, Register rt); 735 void muh(Register rd, Register rs, Register rt); 736 void mulu(Register rd, Register rs, Register rt); 737 void muhu(Register rd, Register rs, Register rt); 738 void mult(Register rs, Register rt); 739 void multu(Register rs, Register rt); 740 void dmul(Register rd, Register rs, Register rt); 741 void dmuh(Register rd, Register rs, Register rt); 742 void dmulu(Register rd, Register rs, Register rt); 743 void dmuhu(Register rd, Register rs, Register rt); 744 void daddu(Register rd, Register rs, Register rt); 745 void dsubu(Register rd, Register rs, Register rt); 746 void dmult(Register rs, Register rt); 747 void dmultu(Register rs, Register rt); 748 749 void addiu(Register rd, Register rs, int32_t j); 750 void daddiu(Register rd, Register rs, int32_t j); 751 752 // Logical. 753 void and_(Register rd, Register rs, Register rt); 754 void or_(Register rd, Register rs, Register rt); 755 void xor_(Register rd, Register rs, Register rt); 756 void nor(Register rd, Register rs, Register rt); 757 758 void andi(Register rd, Register rs, int32_t j); 759 void ori(Register rd, Register rs, int32_t j); 760 void xori(Register rd, Register rs, int32_t j); 761 void lui(Register rd, int32_t j); 762 void aui(Register rt, Register rs, int32_t j); 763 void daui(Register rt, Register rs, int32_t j); 764 void dahi(Register rs, int32_t j); 765 void dati(Register rs, int32_t j); 766 767 // Shifts. 768 // Please note: sll(zero_reg, zero_reg, x) instructions are reserved as nop 769 // and may cause problems in normal code. coming_from_nop makes sure this 770 // doesn't happen. 771 void sll(Register rd, Register rt, uint16_t sa, bool coming_from_nop = false); 772 void sllv(Register rd, Register rt, Register rs); 773 void srl(Register rd, Register rt, uint16_t sa); 774 void srlv(Register rd, Register rt, Register rs); 775 void sra(Register rt, Register rd, uint16_t sa); 776 void srav(Register rt, Register rd, Register rs); 777 void rotr(Register rd, Register rt, uint16_t sa); 778 void rotrv(Register rd, Register rt, Register rs); 779 void dsll(Register rd, Register rt, uint16_t sa); 780 void dsllv(Register rd, Register rt, Register rs); 781 void dsrl(Register rd, Register rt, uint16_t sa); 782 void dsrlv(Register rd, Register rt, Register rs); 783 void drotr(Register rd, Register rt, uint16_t sa); 784 void drotrv(Register rd, Register rt, Register rs); 785 void dsra(Register rt, Register rd, uint16_t sa); 786 void dsrav(Register rd, Register rt, Register rs); 787 void dsll32(Register rt, Register rd, uint16_t sa); 788 void dsrl32(Register rt, Register rd, uint16_t sa); 789 void dsra32(Register rt, Register rd, uint16_t sa); 790 791 // Address computing instructions with shift. 792 void lsa(Register rd, Register rt, Register rs, uint8_t sa); 793 void dlsa(Register rd, Register rt, Register rs, uint8_t sa); 794 795 // ------------Memory-instructions------------- 796 797 void lb(Register rd, const MemOperand& rs); 798 void lbu(Register rd, const MemOperand& rs); 799 void lh(Register rd, const MemOperand& rs); 800 void lhu(Register rd, const MemOperand& rs); 801 void lw(Register rd, const MemOperand& rs); 802 void lwu(Register rd, const MemOperand& rs); 803 void lwl(Register rd, const MemOperand& rs); 804 void lwr(Register rd, const MemOperand& rs); 805 void sb(Register rd, const MemOperand& rs); 806 void sh(Register rd, const MemOperand& rs); 807 void sw(Register rd, const MemOperand& rs); 808 void swl(Register rd, const MemOperand& rs); 809 void swr(Register rd, const MemOperand& rs); 810 void ldl(Register rd, const MemOperand& rs); 811 void ldr(Register rd, const MemOperand& rs); 812 void sdl(Register rd, const MemOperand& rs); 813 void sdr(Register rd, const MemOperand& rs); 814 void ld(Register rd, const MemOperand& rs); 815 void sd(Register rd, const MemOperand& rs); 816 817 818 // ---------PC-Relative-instructions----------- 819 820 void addiupc(Register rs, int32_t imm19); 821 void lwpc(Register rs, int32_t offset19); 822 void lwupc(Register rs, int32_t offset19); 823 void ldpc(Register rs, int32_t offset18); 824 void auipc(Register rs, int16_t imm16); 825 void aluipc(Register rs, int16_t imm16); 826 827 828 // ----------------Prefetch-------------------- 829 830 void pref(int32_t hint, const MemOperand& rs); 831 832 833 // -------------Misc-instructions-------------- 834 835 // Break / Trap instructions. 836 void break_(uint32_t code, bool break_as_stop = false); 837 void stop(const char* msg, uint32_t code = kMaxStopCode); 838 void tge(Register rs, Register rt, uint16_t code); 839 void tgeu(Register rs, Register rt, uint16_t code); 840 void tlt(Register rs, Register rt, uint16_t code); 841 void tltu(Register rs, Register rt, uint16_t code); 842 void teq(Register rs, Register rt, uint16_t code); 843 void tne(Register rs, Register rt, uint16_t code); 844 845 // Move from HI/LO register. 846 void mfhi(Register rd); 847 void mflo(Register rd); 848 849 // Set on less than. 850 void slt(Register rd, Register rs, Register rt); 851 void sltu(Register rd, Register rs, Register rt); 852 void slti(Register rd, Register rs, int32_t j); 853 void sltiu(Register rd, Register rs, int32_t j); 854 855 // Conditional move. 856 void movz(Register rd, Register rs, Register rt); 857 void movn(Register rd, Register rs, Register rt); 858 void movt(Register rd, Register rs, uint16_t cc = 0); 859 void movf(Register rd, Register rs, uint16_t cc = 0); 860 861 void sel(SecondaryField fmt, FPURegister fd, FPURegister fs, FPURegister ft); 862 void sel_s(FPURegister fd, FPURegister fs, FPURegister ft); 863 void sel_d(FPURegister fd, FPURegister fs, FPURegister ft); 864 void seleqz(Register rd, Register rs, Register rt); 865 void seleqz(SecondaryField fmt, FPURegister fd, FPURegister fs, 866 FPURegister ft); 867 void selnez(Register rs, Register rt, Register rd); 868 void selnez(SecondaryField fmt, FPURegister fd, FPURegister fs, 869 FPURegister ft); 870 void seleqz_d(FPURegister fd, FPURegister fs, FPURegister ft); 871 void seleqz_s(FPURegister fd, FPURegister fs, FPURegister ft); 872 void selnez_d(FPURegister fd, FPURegister fs, FPURegister ft); 873 void selnez_s(FPURegister fd, FPURegister fs, FPURegister ft); 874 875 void movz_s(FPURegister fd, FPURegister fs, Register rt); 876 void movz_d(FPURegister fd, FPURegister fs, Register rt); 877 void movt_s(FPURegister fd, FPURegister fs, uint16_t cc = 0); 878 void movt_d(FPURegister fd, FPURegister fs, uint16_t cc = 0); 879 void movf_s(FPURegister fd, FPURegister fs, uint16_t cc = 0); 880 void movf_d(FPURegister fd, FPURegister fs, uint16_t cc = 0); 881 void movn_s(FPURegister fd, FPURegister fs, Register rt); 882 void movn_d(FPURegister fd, FPURegister fs, Register rt); 883 // Bit twiddling. 884 void clz(Register rd, Register rs); 885 void dclz(Register rd, Register rs); 886 void ins_(Register rt, Register rs, uint16_t pos, uint16_t size); 887 void ext_(Register rt, Register rs, uint16_t pos, uint16_t size); 888 void dext_(Register rt, Register rs, uint16_t pos, uint16_t size); 889 void dextm(Register rt, Register rs, uint16_t pos, uint16_t size); 890 void dextu(Register rt, Register rs, uint16_t pos, uint16_t size); 891 void dins_(Register rt, Register rs, uint16_t pos, uint16_t size); 892 void bitswap(Register rd, Register rt); 893 void dbitswap(Register rd, Register rt); 894 void align(Register rd, Register rs, Register rt, uint8_t bp); 895 void dalign(Register rd, Register rs, Register rt, uint8_t bp); 896 897 // --------Coprocessor-instructions---------------- 898 899 // Load, store, and move. 900 void lwc1(FPURegister fd, const MemOperand& src); 901 void ldc1(FPURegister fd, const MemOperand& src); 902 903 void swc1(FPURegister fs, const MemOperand& dst); 904 void sdc1(FPURegister fs, const MemOperand& dst); 905 906 void mtc1(Register rt, FPURegister fs); 907 void mthc1(Register rt, FPURegister fs); 908 void dmtc1(Register rt, FPURegister fs); 909 910 void mfc1(Register rt, FPURegister fs); 911 void mfhc1(Register rt, FPURegister fs); 912 void dmfc1(Register rt, FPURegister fs); 913 914 void ctc1(Register rt, FPUControlRegister fs); 915 void cfc1(Register rt, FPUControlRegister fs); 916 917 // Arithmetic. 918 void add_s(FPURegister fd, FPURegister fs, FPURegister ft); 919 void add_d(FPURegister fd, FPURegister fs, FPURegister ft); 920 void sub_s(FPURegister fd, FPURegister fs, FPURegister ft); 921 void sub_d(FPURegister fd, FPURegister fs, FPURegister ft); 922 void mul_s(FPURegister fd, FPURegister fs, FPURegister ft); 923 void mul_d(FPURegister fd, FPURegister fs, FPURegister ft); 924 void madd_d(FPURegister fd, FPURegister fr, FPURegister fs, FPURegister ft); 925 void div_s(FPURegister fd, FPURegister fs, FPURegister ft); 926 void div_d(FPURegister fd, FPURegister fs, FPURegister ft); 927 void abs_s(FPURegister fd, FPURegister fs); 928 void abs_d(FPURegister fd, FPURegister fs); 929 void mov_d(FPURegister fd, FPURegister fs); 930 void mov_s(FPURegister fd, FPURegister fs); 931 void neg_s(FPURegister fd, FPURegister fs); 932 void neg_d(FPURegister fd, FPURegister fs); 933 void sqrt_s(FPURegister fd, FPURegister fs); 934 void sqrt_d(FPURegister fd, FPURegister fs); 935 void rsqrt_s(FPURegister fd, FPURegister fs); 936 void rsqrt_d(FPURegister fd, FPURegister fs); 937 void recip_d(FPURegister fd, FPURegister fs); 938 void recip_s(FPURegister fd, FPURegister fs); 939 940 // Conversion. 941 void cvt_w_s(FPURegister fd, FPURegister fs); 942 void cvt_w_d(FPURegister fd, FPURegister fs); 943 void trunc_w_s(FPURegister fd, FPURegister fs); 944 void trunc_w_d(FPURegister fd, FPURegister fs); 945 void round_w_s(FPURegister fd, FPURegister fs); 946 void round_w_d(FPURegister fd, FPURegister fs); 947 void floor_w_s(FPURegister fd, FPURegister fs); 948 void floor_w_d(FPURegister fd, FPURegister fs); 949 void ceil_w_s(FPURegister fd, FPURegister fs); 950 void ceil_w_d(FPURegister fd, FPURegister fs); 951 void rint_s(FPURegister fd, FPURegister fs); 952 void rint_d(FPURegister fd, FPURegister fs); 953 void rint(SecondaryField fmt, FPURegister fd, FPURegister fs); 954 955 956 void cvt_l_s(FPURegister fd, FPURegister fs); 957 void cvt_l_d(FPURegister fd, FPURegister fs); 958 void trunc_l_s(FPURegister fd, FPURegister fs); 959 void trunc_l_d(FPURegister fd, FPURegister fs); 960 void round_l_s(FPURegister fd, FPURegister fs); 961 void round_l_d(FPURegister fd, FPURegister fs); 962 void floor_l_s(FPURegister fd, FPURegister fs); 963 void floor_l_d(FPURegister fd, FPURegister fs); 964 void ceil_l_s(FPURegister fd, FPURegister fs); 965 void ceil_l_d(FPURegister fd, FPURegister fs); 966 967 void class_s(FPURegister fd, FPURegister fs); 968 void class_d(FPURegister fd, FPURegister fs); 969 970 void min(SecondaryField fmt, FPURegister fd, FPURegister fs, FPURegister ft); 971 void mina(SecondaryField fmt, FPURegister fd, FPURegister fs, FPURegister ft); 972 void max(SecondaryField fmt, FPURegister fd, FPURegister fs, FPURegister ft); 973 void maxa(SecondaryField fmt, FPURegister fd, FPURegister fs, FPURegister ft); 974 void min_s(FPURegister fd, FPURegister fs, FPURegister ft); 975 void min_d(FPURegister fd, FPURegister fs, FPURegister ft); 976 void max_s(FPURegister fd, FPURegister fs, FPURegister ft); 977 void max_d(FPURegister fd, FPURegister fs, FPURegister ft); 978 void mina_s(FPURegister fd, FPURegister fs, FPURegister ft); 979 void mina_d(FPURegister fd, FPURegister fs, FPURegister ft); 980 void maxa_s(FPURegister fd, FPURegister fs, FPURegister ft); 981 void maxa_d(FPURegister fd, FPURegister fs, FPURegister ft); 982 983 void cvt_s_w(FPURegister fd, FPURegister fs); 984 void cvt_s_l(FPURegister fd, FPURegister fs); 985 void cvt_s_d(FPURegister fd, FPURegister fs); 986 987 void cvt_d_w(FPURegister fd, FPURegister fs); 988 void cvt_d_l(FPURegister fd, FPURegister fs); 989 void cvt_d_s(FPURegister fd, FPURegister fs); 990 991 // Conditions and branches for MIPSr6. 992 void cmp(FPUCondition cond, SecondaryField fmt, 993 FPURegister fd, FPURegister ft, FPURegister fs); 994 void cmp_s(FPUCondition cond, FPURegister fd, FPURegister fs, FPURegister ft); 995 void cmp_d(FPUCondition cond, FPURegister fd, FPURegister fs, FPURegister ft); 996 997 void bc1eqz(int16_t offset, FPURegister ft); bc1eqz(Label * L,FPURegister ft)998 inline void bc1eqz(Label* L, FPURegister ft) { 999 bc1eqz(shifted_branch_offset(L), ft); 1000 } 1001 void bc1nez(int16_t offset, FPURegister ft); bc1nez(Label * L,FPURegister ft)1002 inline void bc1nez(Label* L, FPURegister ft) { 1003 bc1nez(shifted_branch_offset(L), ft); 1004 } 1005 1006 // Conditions and branches for non MIPSr6. 1007 void c(FPUCondition cond, SecondaryField fmt, 1008 FPURegister ft, FPURegister fs, uint16_t cc = 0); 1009 void c_s(FPUCondition cond, FPURegister ft, FPURegister fs, uint16_t cc = 0); 1010 void c_d(FPUCondition cond, FPURegister ft, FPURegister fs, uint16_t cc = 0); 1011 1012 void bc1f(int16_t offset, uint16_t cc = 0); 1013 inline void bc1f(Label* L, uint16_t cc = 0) { 1014 bc1f(shifted_branch_offset(L), cc); 1015 } 1016 void bc1t(int16_t offset, uint16_t cc = 0); 1017 inline void bc1t(Label* L, uint16_t cc = 0) { 1018 bc1t(shifted_branch_offset(L), cc); 1019 } 1020 void fcmp(FPURegister src1, const double src2, FPUCondition cond); 1021 1022 // Check the code size generated from label to here. SizeOfCodeGeneratedSince(Label * label)1023 int SizeOfCodeGeneratedSince(Label* label) { 1024 return pc_offset() - label->pos(); 1025 } 1026 1027 // Check the number of instructions generated from label to here. InstructionsGeneratedSince(Label * label)1028 int InstructionsGeneratedSince(Label* label) { 1029 return SizeOfCodeGeneratedSince(label) / kInstrSize; 1030 } 1031 1032 // Class for scoping postponing the trampoline pool generation. 1033 class BlockTrampolinePoolScope { 1034 public: BlockTrampolinePoolScope(Assembler * assem)1035 explicit BlockTrampolinePoolScope(Assembler* assem) : assem_(assem) { 1036 assem_->StartBlockTrampolinePool(); 1037 } ~BlockTrampolinePoolScope()1038 ~BlockTrampolinePoolScope() { 1039 assem_->EndBlockTrampolinePool(); 1040 } 1041 1042 private: 1043 Assembler* assem_; 1044 1045 DISALLOW_IMPLICIT_CONSTRUCTORS(BlockTrampolinePoolScope); 1046 }; 1047 1048 // Class for postponing the assembly buffer growth. Typically used for 1049 // sequences of instructions that must be emitted as a unit, before 1050 // buffer growth (and relocation) can occur. 1051 // This blocking scope is not nestable. 1052 class BlockGrowBufferScope { 1053 public: BlockGrowBufferScope(Assembler * assem)1054 explicit BlockGrowBufferScope(Assembler* assem) : assem_(assem) { 1055 assem_->StartBlockGrowBuffer(); 1056 } ~BlockGrowBufferScope()1057 ~BlockGrowBufferScope() { 1058 assem_->EndBlockGrowBuffer(); 1059 } 1060 1061 private: 1062 Assembler* assem_; 1063 1064 DISALLOW_IMPLICIT_CONSTRUCTORS(BlockGrowBufferScope); 1065 }; 1066 1067 // Debugging. 1068 1069 // Mark generator continuation. 1070 void RecordGeneratorContinuation(); 1071 1072 // Mark address of a debug break slot. 1073 void RecordDebugBreakSlot(RelocInfo::Mode mode); 1074 1075 // Record the AST id of the CallIC being compiled, so that it can be placed 1076 // in the relocation information. SetRecordedAstId(TypeFeedbackId ast_id)1077 void SetRecordedAstId(TypeFeedbackId ast_id) { 1078 DCHECK(recorded_ast_id_.IsNone()); 1079 recorded_ast_id_ = ast_id; 1080 } 1081 RecordedAstId()1082 TypeFeedbackId RecordedAstId() { 1083 DCHECK(!recorded_ast_id_.IsNone()); 1084 return recorded_ast_id_; 1085 } 1086 ClearRecordedAstId()1087 void ClearRecordedAstId() { recorded_ast_id_ = TypeFeedbackId::None(); } 1088 1089 // Record a comment relocation entry that can be used by a disassembler. 1090 // Use --code-comments to enable. 1091 void RecordComment(const char* msg); 1092 1093 // Record a deoptimization reason that can be used by a log or cpu profiler. 1094 // Use --trace-deopt to enable. 1095 void RecordDeoptReason(const int reason, const SourcePosition position); 1096 1097 static int RelocateInternalReference(RelocInfo::Mode rmode, byte* pc, 1098 intptr_t pc_delta); 1099 1100 // Writes a single byte or word of data in the code stream. Used for 1101 // inline tables, e.g., jump-tables. 1102 void db(uint8_t data); 1103 void dd(uint32_t data); 1104 void dq(uint64_t data); dp(uintptr_t data)1105 void dp(uintptr_t data) { dq(data); } 1106 void dd(Label* label); 1107 positions_recorder()1108 PositionsRecorder* positions_recorder() { return &positions_recorder_; } 1109 1110 // Postpone the generation of the trampoline pool for the specified number of 1111 // instructions. 1112 void BlockTrampolinePoolFor(int instructions); 1113 1114 // Check if there is less than kGap bytes available in the buffer. 1115 // If this is the case, we need to grow the buffer before emitting 1116 // an instruction or relocation information. overflow()1117 inline bool overflow() const { return pc_ >= reloc_info_writer.pos() - kGap; } 1118 1119 // Get the number of bytes available in the buffer. available_space()1120 inline intptr_t available_space() const { 1121 return reloc_info_writer.pos() - pc_; 1122 } 1123 1124 // Read/patch instructions. instr_at(byte * pc)1125 static Instr instr_at(byte* pc) { return *reinterpret_cast<Instr*>(pc); } instr_at_put(byte * pc,Instr instr)1126 static void instr_at_put(byte* pc, Instr instr) { 1127 *reinterpret_cast<Instr*>(pc) = instr; 1128 } instr_at(int pos)1129 Instr instr_at(int pos) { return *reinterpret_cast<Instr*>(buffer_ + pos); } instr_at_put(int pos,Instr instr)1130 void instr_at_put(int pos, Instr instr) { 1131 *reinterpret_cast<Instr*>(buffer_ + pos) = instr; 1132 } 1133 1134 // Check if an instruction is a branch of some kind. 1135 static bool IsBranch(Instr instr); 1136 static bool IsBc(Instr instr); 1137 static bool IsBzc(Instr instr); 1138 1139 static bool IsBeq(Instr instr); 1140 static bool IsBne(Instr instr); 1141 static bool IsBeqzc(Instr instr); 1142 static bool IsBnezc(Instr instr); 1143 static bool IsBeqc(Instr instr); 1144 static bool IsBnec(Instr instr); 1145 1146 1147 static bool IsJump(Instr instr); 1148 static bool IsJ(Instr instr); 1149 static bool IsLui(Instr instr); 1150 static bool IsOri(Instr instr); 1151 1152 static bool IsJal(Instr instr); 1153 static bool IsJr(Instr instr); 1154 static bool IsJalr(Instr instr); 1155 1156 static bool IsNop(Instr instr, unsigned int type); 1157 static bool IsPop(Instr instr); 1158 static bool IsPush(Instr instr); 1159 static bool IsLwRegFpOffset(Instr instr); 1160 static bool IsSwRegFpOffset(Instr instr); 1161 static bool IsLwRegFpNegOffset(Instr instr); 1162 static bool IsSwRegFpNegOffset(Instr instr); 1163 1164 static Register GetRtReg(Instr instr); 1165 static Register GetRsReg(Instr instr); 1166 static Register GetRdReg(Instr instr); 1167 1168 static uint32_t GetRt(Instr instr); 1169 static uint32_t GetRtField(Instr instr); 1170 static uint32_t GetRs(Instr instr); 1171 static uint32_t GetRsField(Instr instr); 1172 static uint32_t GetRd(Instr instr); 1173 static uint32_t GetRdField(Instr instr); 1174 static uint32_t GetSa(Instr instr); 1175 static uint32_t GetSaField(Instr instr); 1176 static uint32_t GetOpcodeField(Instr instr); 1177 static uint32_t GetFunction(Instr instr); 1178 static uint32_t GetFunctionField(Instr instr); 1179 static uint32_t GetImmediate16(Instr instr); 1180 static uint32_t GetLabelConst(Instr instr); 1181 1182 static int32_t GetBranchOffset(Instr instr); 1183 static bool IsLw(Instr instr); 1184 static int16_t GetLwOffset(Instr instr); 1185 static Instr SetLwOffset(Instr instr, int16_t offset); 1186 1187 static bool IsSw(Instr instr); 1188 static Instr SetSwOffset(Instr instr, int16_t offset); 1189 static bool IsAddImmediate(Instr instr); 1190 static Instr SetAddImmediateOffset(Instr instr, int16_t offset); 1191 1192 static bool IsAndImmediate(Instr instr); 1193 static bool IsEmittedConstant(Instr instr); 1194 1195 void CheckTrampolinePool(); 1196 PatchConstantPoolAccessInstruction(int pc_offset,int offset,ConstantPoolEntry::Access access,ConstantPoolEntry::Type type)1197 void PatchConstantPoolAccessInstruction(int pc_offset, int offset, 1198 ConstantPoolEntry::Access access, 1199 ConstantPoolEntry::Type type) { 1200 // No embedded constant pool support. 1201 UNREACHABLE(); 1202 } 1203 IsPrevInstrCompactBranch()1204 bool IsPrevInstrCompactBranch() { return prev_instr_compact_branch_; } 1205 1206 protected: 1207 // Relocation for a type-recording IC has the AST id added to it. This 1208 // member variable is a way to pass the information from the call site to 1209 // the relocation info. 1210 TypeFeedbackId recorded_ast_id_; 1211 1212 inline static void set_target_internal_reference_encoded_at(Address pc, 1213 Address target); 1214 buffer_space()1215 int64_t buffer_space() const { return reloc_info_writer.pos() - pc_; } 1216 1217 // Decode branch instruction at pos and return branch target pos. 1218 int target_at(int pos, bool is_internal); 1219 1220 // Patch branch instruction at pos to branch to given branch target pos. 1221 void target_at_put(int pos, int target_pos, bool is_internal); 1222 1223 // Say if we need to relocate with this mode. 1224 bool MustUseReg(RelocInfo::Mode rmode); 1225 1226 // Record reloc info for current pc_. 1227 void RecordRelocInfo(RelocInfo::Mode rmode, intptr_t data = 0); 1228 1229 // Block the emission of the trampoline pool before pc_offset. BlockTrampolinePoolBefore(int pc_offset)1230 void BlockTrampolinePoolBefore(int pc_offset) { 1231 if (no_trampoline_pool_before_ < pc_offset) 1232 no_trampoline_pool_before_ = pc_offset; 1233 } 1234 StartBlockTrampolinePool()1235 void StartBlockTrampolinePool() { 1236 trampoline_pool_blocked_nesting_++; 1237 } 1238 EndBlockTrampolinePool()1239 void EndBlockTrampolinePool() { 1240 trampoline_pool_blocked_nesting_--; 1241 } 1242 is_trampoline_pool_blocked()1243 bool is_trampoline_pool_blocked() const { 1244 return trampoline_pool_blocked_nesting_ > 0; 1245 } 1246 has_exception()1247 bool has_exception() const { 1248 return internal_trampoline_exception_; 1249 } 1250 1251 void DoubleAsTwoUInt32(double d, uint32_t* lo, uint32_t* hi); 1252 is_trampoline_emitted()1253 bool is_trampoline_emitted() const { 1254 return trampoline_emitted_; 1255 } 1256 1257 // Temporarily block automatic assembly buffer growth. StartBlockGrowBuffer()1258 void StartBlockGrowBuffer() { 1259 DCHECK(!block_buffer_growth_); 1260 block_buffer_growth_ = true; 1261 } 1262 EndBlockGrowBuffer()1263 void EndBlockGrowBuffer() { 1264 DCHECK(block_buffer_growth_); 1265 block_buffer_growth_ = false; 1266 } 1267 is_buffer_growth_blocked()1268 bool is_buffer_growth_blocked() const { 1269 return block_buffer_growth_; 1270 } 1271 EmitForbiddenSlotInstruction()1272 void EmitForbiddenSlotInstruction() { 1273 if (IsPrevInstrCompactBranch()) { 1274 nop(); 1275 ClearCompactBranchState(); 1276 } 1277 } 1278 1279 inline void CheckTrampolinePoolQuick(int extra_instructions = 0); 1280 1281 private: 1282 // Buffer size and constant pool distance are checked together at regular 1283 // intervals of kBufferCheckInterval emitted bytes. 1284 static const int kBufferCheckInterval = 1*KB/2; 1285 1286 // Code generation. 1287 // The relocation writer's position is at least kGap bytes below the end of 1288 // the generated instructions. This is so that multi-instruction sequences do 1289 // not have to check for overflow. The same is true for writes of large 1290 // relocation info entries. 1291 static const int kGap = 32; 1292 1293 1294 // Repeated checking whether the trampoline pool should be emitted is rather 1295 // expensive. By default we only check again once a number of instructions 1296 // has been generated. 1297 static const int kCheckConstIntervalInst = 32; 1298 static const int kCheckConstInterval = kCheckConstIntervalInst * kInstrSize; 1299 1300 int next_buffer_check_; // pc offset of next buffer check. 1301 1302 // Emission of the trampoline pool may be blocked in some code sequences. 1303 int trampoline_pool_blocked_nesting_; // Block emission if this is not zero. 1304 int no_trampoline_pool_before_; // Block emission before this pc offset. 1305 1306 // Keep track of the last emitted pool to guarantee a maximal distance. 1307 int last_trampoline_pool_end_; // pc offset of the end of the last pool. 1308 1309 // Automatic growth of the assembly buffer may be blocked for some sequences. 1310 bool block_buffer_growth_; // Block growth when true. 1311 1312 // Relocation information generation. 1313 // Each relocation is encoded as a variable size value. 1314 static const int kMaxRelocSize = RelocInfoWriter::kMaxSize; 1315 RelocInfoWriter reloc_info_writer; 1316 1317 // The bound position, before this we cannot do instruction elimination. 1318 int last_bound_pos_; 1319 1320 // Readable constants for compact branch handling in emit() 1321 enum class CompactBranchType : bool { NO = false, COMPACT_BRANCH = true }; 1322 1323 // Code emission. 1324 inline void CheckBuffer(); 1325 void GrowBuffer(); 1326 inline void emit(Instr x, 1327 CompactBranchType is_compact_branch = CompactBranchType::NO); 1328 inline void emit(uint64_t x); 1329 inline void CheckForEmitInForbiddenSlot(); 1330 template <typename T> 1331 inline void EmitHelper(T x); 1332 inline void EmitHelper(Instr x, CompactBranchType is_compact_branch); 1333 1334 // Instruction generation. 1335 // We have 3 different kind of encoding layout on MIPS. 1336 // However due to many different types of objects encoded in the same fields 1337 // we have quite a few aliases for each mode. 1338 // Using the same structure to refer to Register and FPURegister would spare a 1339 // few aliases, but mixing both does not look clean to me. 1340 // Anyway we could surely implement this differently. 1341 1342 void GenInstrRegister(Opcode opcode, 1343 Register rs, 1344 Register rt, 1345 Register rd, 1346 uint16_t sa = 0, 1347 SecondaryField func = NULLSF); 1348 1349 void GenInstrRegister(Opcode opcode, 1350 Register rs, 1351 Register rt, 1352 uint16_t msb, 1353 uint16_t lsb, 1354 SecondaryField func); 1355 1356 void GenInstrRegister(Opcode opcode, 1357 SecondaryField fmt, 1358 FPURegister ft, 1359 FPURegister fs, 1360 FPURegister fd, 1361 SecondaryField func = NULLSF); 1362 1363 void GenInstrRegister(Opcode opcode, 1364 FPURegister fr, 1365 FPURegister ft, 1366 FPURegister fs, 1367 FPURegister fd, 1368 SecondaryField func = NULLSF); 1369 1370 void GenInstrRegister(Opcode opcode, 1371 SecondaryField fmt, 1372 Register rt, 1373 FPURegister fs, 1374 FPURegister fd, 1375 SecondaryField func = NULLSF); 1376 1377 void GenInstrRegister(Opcode opcode, 1378 SecondaryField fmt, 1379 Register rt, 1380 FPUControlRegister fs, 1381 SecondaryField func = NULLSF); 1382 1383 1384 void GenInstrImmediate( 1385 Opcode opcode, Register rs, Register rt, int32_t j, 1386 CompactBranchType is_compact_branch = CompactBranchType::NO); 1387 void GenInstrImmediate( 1388 Opcode opcode, Register rs, SecondaryField SF, int32_t j, 1389 CompactBranchType is_compact_branch = CompactBranchType::NO); 1390 void GenInstrImmediate( 1391 Opcode opcode, Register r1, FPURegister r2, int32_t j, 1392 CompactBranchType is_compact_branch = CompactBranchType::NO); 1393 void GenInstrImmediate( 1394 Opcode opcode, Register rs, int32_t offset21, 1395 CompactBranchType is_compact_branch = CompactBranchType::NO); 1396 void GenInstrImmediate(Opcode opcode, Register rs, uint32_t offset21); 1397 void GenInstrImmediate( 1398 Opcode opcode, int32_t offset26, 1399 CompactBranchType is_compact_branch = CompactBranchType::NO); 1400 1401 void GenInstrJump(Opcode opcode, 1402 uint32_t address); 1403 1404 // Helpers. 1405 void LoadRegPlusOffsetToAt(const MemOperand& src); 1406 1407 // Labels. 1408 void print(Label* L); 1409 void bind_to(Label* L, int pos); 1410 void next(Label* L, bool is_internal); 1411 1412 // One trampoline consists of: 1413 // - space for trampoline slots, 1414 // - space for labels. 1415 // 1416 // Space for trampoline slots is equal to slot_count * 2 * kInstrSize. 1417 // Space for trampoline slots preceeds space for labels. Each label is of one 1418 // instruction size, so total amount for labels is equal to 1419 // label_count * kInstrSize. 1420 class Trampoline { 1421 public: Trampoline()1422 Trampoline() { 1423 start_ = 0; 1424 next_slot_ = 0; 1425 free_slot_count_ = 0; 1426 end_ = 0; 1427 } Trampoline(int start,int slot_count)1428 Trampoline(int start, int slot_count) { 1429 start_ = start; 1430 next_slot_ = start; 1431 free_slot_count_ = slot_count; 1432 end_ = start + slot_count * kTrampolineSlotsSize; 1433 } start()1434 int start() { 1435 return start_; 1436 } end()1437 int end() { 1438 return end_; 1439 } take_slot()1440 int take_slot() { 1441 int trampoline_slot = kInvalidSlotPos; 1442 if (free_slot_count_ <= 0) { 1443 // We have run out of space on trampolines. 1444 // Make sure we fail in debug mode, so we become aware of each case 1445 // when this happens. 1446 DCHECK(0); 1447 // Internal exception will be caught. 1448 } else { 1449 trampoline_slot = next_slot_; 1450 free_slot_count_--; 1451 next_slot_ += kTrampolineSlotsSize; 1452 } 1453 return trampoline_slot; 1454 } 1455 1456 private: 1457 int start_; 1458 int end_; 1459 int next_slot_; 1460 int free_slot_count_; 1461 }; 1462 1463 int32_t get_trampoline_entry(int32_t pos); 1464 int unbound_labels_count_; 1465 // After trampoline is emitted, long branches are used in generated code for 1466 // the forward branches whose target offsets could be beyond reach of branch 1467 // instruction. We use this information to trigger different mode of 1468 // branch instruction generation, where we use jump instructions rather 1469 // than regular branch instructions. 1470 bool trampoline_emitted_; 1471 static const int kTrampolineSlotsSize = 2 * kInstrSize; 1472 static const int kMaxBranchOffset = (1 << (18 - 1)) - 1; 1473 static const int kMaxCompactBranchOffset = (1 << (28 - 1)) - 1; 1474 static const int kInvalidSlotPos = -1; 1475 1476 // Internal reference positions, required for unbounded internal reference 1477 // labels. 1478 std::set<int64_t> internal_reference_positions_; 1479 EmittedCompactBranchInstruction()1480 void EmittedCompactBranchInstruction() { prev_instr_compact_branch_ = true; } ClearCompactBranchState()1481 void ClearCompactBranchState() { prev_instr_compact_branch_ = false; } 1482 bool prev_instr_compact_branch_ = false; 1483 1484 Trampoline trampoline_; 1485 bool internal_trampoline_exception_; 1486 1487 friend class RegExpMacroAssemblerMIPS; 1488 friend class RelocInfo; 1489 friend class CodePatcher; 1490 friend class BlockTrampolinePoolScope; 1491 1492 PositionsRecorder positions_recorder_; 1493 friend class PositionsRecorder; 1494 friend class EnsureSpace; 1495 }; 1496 1497 1498 class EnsureSpace BASE_EMBEDDED { 1499 public: EnsureSpace(Assembler * assembler)1500 explicit EnsureSpace(Assembler* assembler) { 1501 assembler->CheckBuffer(); 1502 } 1503 }; 1504 1505 } // namespace internal 1506 } // namespace v8 1507 1508 #endif // V8_ARM_ASSEMBLER_MIPS_H_ 1509