Source file src/cmd/compile/internal/ssa/_gen/ARM64Ops.go

     1  // Copyright 2016 The Go Authors. All rights reserved.
     2  // Use of this source code is governed by a BSD-style
     3  // license that can be found in the LICENSE file.
     4  
     5  package main
     6  
     7  import "strings"
     8  
     9  // Notes:
    10  //  - Integer types live in the low portion of registers. Upper portions are junk.
    11  //  - Boolean types use the low-order byte of a register. 0=false, 1=true.
    12  //    Upper bytes are junk.
    13  //  - *const instructions may use a constant larger than the instruction can encode.
    14  //    In this case the assembler expands to multiple instructions and uses tmp
    15  //    register (R27).
    16  //  - All 32-bit Ops will zero the upper 32 bits of the destination register.
    17  
    18  // Suffixes encode the bit width of various instructions.
    19  // D (double word) = 64 bit
    20  // W (word)        = 32 bit
    21  // H (half word)   = 16 bit
    22  // HU              = 16 bit unsigned
    23  // B (byte)        = 8 bit
    24  // BU              = 8 bit unsigned
    25  // S (single)      = 32 bit float
    26  // D (double)      = 64 bit float
    27  
    28  // Note: registers not used in regalloc are not included in this list,
    29  // so that regmask stays within int64
    30  // Be careful when hand coding regmasks.
    31  var regNamesARM64 = []string{
    32  	"R0",
    33  	"R1",
    34  	"R2",
    35  	"R3",
    36  	"R4",
    37  	"R5",
    38  	"R6",
    39  	"R7",
    40  	"R8",
    41  	"R9",
    42  	"R10",
    43  	"R11",
    44  	"R12",
    45  	"R13",
    46  	"R14",
    47  	"R15",
    48  	"R16",
    49  	"R17",
    50  	// R18 = platform register, not used
    51  	"R19",
    52  	"R20",
    53  	"R21",
    54  	"R22",
    55  	"R23",
    56  	"R24",
    57  	"R25",
    58  	"R26",
    59  	// R27 = REGTMP not used in regalloc
    60  	"g",    // aka R28
    61  	"R29",  // frame pointer, not used
    62  	"R30",  // aka REGLINK
    63  	"ZERO", // zero register (aka R31)
    64  	"SP",   // stack pointer (aka R31)
    65  
    66  	// Note: both ZERO and SP are register number 31!
    67  	// What r31 means in a particular instruction depends on
    68  	// the instruction.  Generally, for arguments of instructions
    69  	// which are addresses to load or store from, r31 means SP.
    70  	// In other instructions, r31 means ZERO. But there are
    71  	// exceptions.
    72  	// See https://stackoverflow.com/questions/61532867
    73  	// This does not have much of an effect here, as the
    74  	// cmd/internal/obj/arm64 interface treats them as two
    75  	// different registers and picks the right instruction
    76  	// that encodes what r31 means. But see issue 71651.
    77  
    78  	"F0",
    79  	"F1",
    80  	"F2",
    81  	"F3",
    82  	"F4",
    83  	"F5",
    84  	"F6",
    85  	"F7",
    86  	"F8",
    87  	"F9",
    88  	"F10",
    89  	"F11",
    90  	"F12",
    91  	"F13",
    92  	"F14",
    93  	"F15",
    94  	"F16",
    95  	"F17",
    96  	"F18",
    97  	"F19",
    98  	"F20",
    99  	"F21",
   100  	"F22",
   101  	"F23",
   102  	"F24",
   103  	"F25",
   104  	"F26",
   105  	"F27",
   106  	"F28",
   107  	"F29",
   108  	"F30",
   109  	"F31",
   110  
   111  	// If you add registers, update asyncPreempt in runtime.
   112  
   113  	// pseudo-registers
   114  	"SB",
   115  }
   116  
   117  func init() {
   118  	// Make map from reg names to reg integers.
   119  	if len(regNamesARM64) > 64 {
   120  		panic("too many registers")
   121  	}
   122  	num := map[string]int{}
   123  	for i, name := range regNamesARM64 {
   124  		num[name] = i
   125  	}
   126  	buildReg := func(s string) regMask {
   127  		m := regMask(0)
   128  		for _, r := range strings.Split(s, " ") {
   129  			if n, ok := num[r]; ok {
   130  				m |= regMask(1) << uint(n)
   131  				continue
   132  			}
   133  			panic("register " + r + " not found")
   134  		}
   135  		return m
   136  	}
   137  
   138  	// Common individual register masks
   139  	var (
   140  		gp         = buildReg("R0 R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R19 R20 R21 R22 R23 R24 R25 R26 R30")
   141  		gpg        = gp | buildReg("g")
   142  		gpsp       = gp | buildReg("SP")
   143  		gpspg      = gpg | buildReg("SP")
   144  		gpspsbg    = gpspg | buildReg("SB")
   145  		fp         = buildReg("F0 F1 F2 F3 F4 F5 F6 F7 F8 F9 F10 F11 F12 F13 F14 F15 F16 F17 F18 F19 F20 F21 F22 F23 F24 F25 F26 F27 F28 F29 F30 F31")
   146  		callerSave = gp | fp | buildReg("g") // runtime.setg (and anything calling it) may clobber g
   147  		r25        = buildReg("R25")
   148  		r24to25    = buildReg("R24 R25")
   149  		f16to17    = buildReg("F16 F17")
   150  		rz         = buildReg("ZERO")
   151  		first16    = buildReg("R0 R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15")
   152  	)
   153  	// Common regInfo
   154  	var (
   155  		gp01           = regInfo{inputs: nil, outputs: []regMask{gp}}
   156  		gp0flags1      = regInfo{inputs: []regMask{0}, outputs: []regMask{gp}}
   157  		gp11           = regInfo{inputs: []regMask{gpg}, outputs: []regMask{gp}}
   158  		gp11sp         = regInfo{inputs: []regMask{gpspg}, outputs: []regMask{gp}}
   159  		gp1flags       = regInfo{inputs: []regMask{gpg}}
   160  		gp1flagsflags  = regInfo{inputs: []regMask{gpg}}
   161  		gp1flags1      = regInfo{inputs: []regMask{gpg}, outputs: []regMask{gp}}
   162  		gp11flags      = regInfo{inputs: []regMask{gpg}, outputs: []regMask{gp, 0}}
   163  		gp21           = regInfo{inputs: []regMask{gpg, gpg}, outputs: []regMask{gp}}
   164  		gp21nog        = regInfo{inputs: []regMask{gp, gp}, outputs: []regMask{gp}}
   165  		gp21flags      = regInfo{inputs: []regMask{gp, gp}, outputs: []regMask{gp, 0}}
   166  		gp2flags       = regInfo{inputs: []regMask{gpg, gpg}}
   167  		gp2flagsflags  = regInfo{inputs: []regMask{gpg, gpg}}
   168  		gp2flags1      = regInfo{inputs: []regMask{gp, gp}, outputs: []regMask{gp}}
   169  		gp2flags1flags = regInfo{inputs: []regMask{gp, gp, 0}, outputs: []regMask{gp, 0}}
   170  		gp2load        = regInfo{inputs: []regMask{gpspsbg, gpg}, outputs: []regMask{gp}}
   171  		gp31           = regInfo{inputs: []regMask{gpg, gpg, gpg}, outputs: []regMask{gp}}
   172  		gpload         = regInfo{inputs: []regMask{gpspsbg}, outputs: []regMask{gp}}
   173  		gpload2        = regInfo{inputs: []regMask{gpspsbg}, outputs: []regMask{gpg, gpg}}
   174  		gpstore        = regInfo{inputs: []regMask{gpspsbg, gpg | rz}}
   175  		gpstore2       = regInfo{inputs: []regMask{gpspsbg, gpg | rz, gpg | rz}}
   176  		gpxchg         = regInfo{inputs: []regMask{gpspsbg, gpg | rz}, outputs: []regMask{gp}}
   177  		gpcas          = regInfo{inputs: []regMask{gpspsbg, gpg | rz, gpg | rz}, outputs: []regMask{gp}}
   178  		fp01           = regInfo{inputs: nil, outputs: []regMask{fp}}
   179  		fp11           = regInfo{inputs: []regMask{fp}, outputs: []regMask{fp}}
   180  		fpgp           = regInfo{inputs: []regMask{fp}, outputs: []regMask{gp}}
   181  		gpfp           = regInfo{inputs: []regMask{gp}, outputs: []regMask{fp}}
   182  		fp21           = regInfo{inputs: []regMask{fp, fp}, outputs: []regMask{fp}}
   183  		fp31           = regInfo{inputs: []regMask{fp, fp, fp}, outputs: []regMask{fp}}
   184  		fp2flags       = regInfo{inputs: []regMask{fp, fp}}
   185  		fp1flags       = regInfo{inputs: []regMask{fp}}
   186  		fpload         = regInfo{inputs: []regMask{gpspsbg}, outputs: []regMask{fp}}
   187  		fpload2        = regInfo{inputs: []regMask{gpspsbg}, outputs: []regMask{fp, fp}}
   188  		fp2load        = regInfo{inputs: []regMask{gpspsbg, gpg}, outputs: []regMask{fp}}
   189  		fpstore        = regInfo{inputs: []regMask{gpspsbg, fp}}
   190  		fpstoreidx     = regInfo{inputs: []regMask{gpspsbg, gpg, fp}}
   191  		fpstore2       = regInfo{inputs: []regMask{gpspsbg, fp, fp}}
   192  		readflags      = regInfo{inputs: nil, outputs: []regMask{gp}}
   193  		prefreg        = regInfo{inputs: []regMask{gpspsbg}}
   194  	)
   195  	ops := []opData{
   196  		// binary ops
   197  		{name: "ADCSflags", argLength: 3, reg: gp2flags1flags, typ: "(UInt64,Flags)", asm: "ADCS", commutative: true}, // arg0+arg1+carry, set flags.
   198  		{name: "ADCzerocarry", argLength: 1, reg: gp0flags1, typ: "UInt64", asm: "ADC"},                               // ZR+ZR+carry
   199  		{name: "ADD", argLength: 2, reg: gp21, asm: "ADD", commutative: true},                                         // arg0 + arg1
   200  		{name: "ADDconst", argLength: 1, reg: gp11sp, asm: "ADD", aux: "Int64"},                                       // arg0 + auxInt
   201  		{name: "ADDSconstflags", argLength: 1, reg: gp11flags, typ: "(UInt64,Flags)", asm: "ADDS", aux: "Int64"},      // arg0+auxint, set flags.
   202  		{name: "ADDSflags", argLength: 2, reg: gp21flags, typ: "(UInt64,Flags)", asm: "ADDS", commutative: true},      // arg0+arg1, set flags.
   203  		{name: "SUB", argLength: 2, reg: gp21, asm: "SUB"},                                                            // arg0 - arg1
   204  		{name: "SUBconst", argLength: 1, reg: gp11, asm: "SUB", aux: "Int64"},                                         // arg0 - auxInt
   205  		{name: "SBCSflags", argLength: 3, reg: gp2flags1flags, typ: "(UInt64,Flags)", asm: "SBCS"},                    // arg0-(arg1+borrowing), set flags.
   206  		{name: "SUBSflags", argLength: 2, reg: gp21flags, typ: "(UInt64,Flags)", asm: "SUBS"},                         // arg0 - arg1, set flags.
   207  		{name: "MUL", argLength: 2, reg: gp21, asm: "MUL", commutative: true},                                         // arg0 * arg1
   208  		{name: "MULW", argLength: 2, reg: gp21, asm: "MULW", commutative: true},                                       // arg0 * arg1, 32-bit
   209  		{name: "MNEG", argLength: 2, reg: gp21, asm: "MNEG", commutative: true},                                       // -arg0 * arg1
   210  		{name: "MNEGW", argLength: 2, reg: gp21, asm: "MNEGW", commutative: true},                                     // -arg0 * arg1, 32-bit
   211  		{name: "MULH", argLength: 2, reg: gp21, asm: "SMULH", commutative: true},                                      // (arg0 * arg1) >> 64, signed
   212  		{name: "UMULH", argLength: 2, reg: gp21, asm: "UMULH", commutative: true},                                     // (arg0 * arg1) >> 64, unsigned
   213  		{name: "MULL", argLength: 2, reg: gp21, asm: "SMULL", commutative: true},                                      // arg0 * arg1, signed, 32-bit mult results in 64-bit
   214  		{name: "UMULL", argLength: 2, reg: gp21, asm: "UMULL", commutative: true},                                     // arg0 * arg1, unsigned, 32-bit mult results in 64-bit
   215  		{name: "DIV", argLength: 2, reg: gp21, asm: "SDIV"},                                                           // arg0 / arg1, signed
   216  		{name: "UDIV", argLength: 2, reg: gp21, asm: "UDIV"},                                                          // arg0 / arg1, unsigned
   217  		{name: "DIVW", argLength: 2, reg: gp21, asm: "SDIVW"},                                                         // arg0 / arg1, signed, 32 bit
   218  		{name: "UDIVW", argLength: 2, reg: gp21, asm: "UDIVW"},                                                        // arg0 / arg1, unsigned, 32 bit
   219  		{name: "MOD", argLength: 2, reg: gp21, asm: "REM"},                                                            // arg0 % arg1, signed
   220  		{name: "UMOD", argLength: 2, reg: gp21, asm: "UREM"},                                                          // arg0 % arg1, unsigned
   221  		{name: "MODW", argLength: 2, reg: gp21, asm: "REMW"},                                                          // arg0 % arg1, signed, 32 bit
   222  		{name: "UMODW", argLength: 2, reg: gp21, asm: "UREMW"},                                                        // arg0 % arg1, unsigned, 32 bit
   223  
   224  		{name: "FADDS", argLength: 2, reg: fp21, asm: "FADDS", commutative: true},   // arg0 + arg1
   225  		{name: "FADDD", argLength: 2, reg: fp21, asm: "FADDD", commutative: true},   // arg0 + arg1
   226  		{name: "FSUBS", argLength: 2, reg: fp21, asm: "FSUBS"},                      // arg0 - arg1
   227  		{name: "FSUBD", argLength: 2, reg: fp21, asm: "FSUBD"},                      // arg0 - arg1
   228  		{name: "FMULS", argLength: 2, reg: fp21, asm: "FMULS", commutative: true},   // arg0 * arg1
   229  		{name: "FMULD", argLength: 2, reg: fp21, asm: "FMULD", commutative: true},   // arg0 * arg1
   230  		{name: "FNMULS", argLength: 2, reg: fp21, asm: "FNMULS", commutative: true}, // -(arg0 * arg1)
   231  		{name: "FNMULD", argLength: 2, reg: fp21, asm: "FNMULD", commutative: true}, // -(arg0 * arg1)
   232  		{name: "FDIVS", argLength: 2, reg: fp21, asm: "FDIVS"},                      // arg0 / arg1
   233  		{name: "FDIVD", argLength: 2, reg: fp21, asm: "FDIVD"},                      // arg0 / arg1
   234  
   235  		{name: "AND", argLength: 2, reg: gp21, asm: "AND", commutative: true}, // arg0 & arg1
   236  		{name: "ANDconst", argLength: 1, reg: gp11, asm: "AND", aux: "Int64"}, // arg0 & auxInt
   237  		{name: "OR", argLength: 2, reg: gp21, asm: "ORR", commutative: true},  // arg0 | arg1
   238  		{name: "ORconst", argLength: 1, reg: gp11, asm: "ORR", aux: "Int64"},  // arg0 | auxInt
   239  		{name: "XOR", argLength: 2, reg: gp21, asm: "EOR", commutative: true}, // arg0 ^ arg1
   240  		{name: "XORconst", argLength: 1, reg: gp11, asm: "EOR", aux: "Int64"}, // arg0 ^ auxInt
   241  		{name: "BIC", argLength: 2, reg: gp21, asm: "BIC"},                    // arg0 &^ arg1
   242  		{name: "EON", argLength: 2, reg: gp21, asm: "EON"},                    // arg0 ^ ^arg1
   243  		{name: "ORN", argLength: 2, reg: gp21, asm: "ORN"},                    // arg0 | ^arg1
   244  
   245  		// unary ops
   246  		{name: "MVN", argLength: 1, reg: gp11, asm: "MVN"},                                    // ^arg0
   247  		{name: "NEG", argLength: 1, reg: gp11, asm: "NEG"},                                    // -arg0
   248  		{name: "NEGSflags", argLength: 1, reg: gp11flags, typ: "(UInt64,Flags)", asm: "NEGS"}, // -arg0, set flags.
   249  		{name: "NGCzerocarry", argLength: 1, reg: gp0flags1, typ: "UInt64", asm: "NGC"},       // -1 if borrowing, 0 otherwise.
   250  		{name: "FABSD", argLength: 1, reg: fp11, asm: "FABSD"},                                // abs(arg0), float64
   251  		{name: "FABSS", argLength: 1, reg: fp11, asm: "FABSS"},                                // abs(arg0), float32
   252  		{name: "FNEGS", argLength: 1, reg: fp11, asm: "FNEGS"},                                // -arg0, float32
   253  		{name: "FNEGD", argLength: 1, reg: fp11, asm: "FNEGD"},                                // -arg0, float64
   254  		{name: "FSQRTD", argLength: 1, reg: fp11, asm: "FSQRTD"},                              // sqrt(arg0), float64
   255  		{name: "FSQRTS", argLength: 1, reg: fp11, asm: "FSQRTS"},                              // sqrt(arg0), float32
   256  		{name: "FMIND", argLength: 2, reg: fp21, asm: "FMIND"},                                // min(arg0, arg1)
   257  		{name: "FMINS", argLength: 2, reg: fp21, asm: "FMINS"},                                // min(arg0, arg1)
   258  		{name: "FMAXD", argLength: 2, reg: fp21, asm: "FMAXD"},                                // max(arg0, arg1)
   259  		{name: "FMAXS", argLength: 2, reg: fp21, asm: "FMAXS"},                                // max(arg0, arg1)
   260  		{name: "REV", argLength: 1, reg: gp11, asm: "REV"},                                    // byte reverse, 64-bit
   261  		{name: "REVW", argLength: 1, reg: gp11, asm: "REVW"},                                  // byte reverse, 32-bit
   262  		{name: "REV16", argLength: 1, reg: gp11, asm: "REV16"},                                // byte reverse in each 16-bit halfword, 64-bit
   263  		{name: "REV16W", argLength: 1, reg: gp11, asm: "REV16W"},                              // byte reverse in each 16-bit halfword, 32-bit
   264  		{name: "RBIT", argLength: 1, reg: gp11, asm: "RBIT"},                                  // bit reverse, 64-bit
   265  		{name: "RBITW", argLength: 1, reg: gp11, asm: "RBITW"},                                // bit reverse, 32-bit
   266  		{name: "CLZ", argLength: 1, reg: gp11, asm: "CLZ"},                                    // count leading zero, 64-bit
   267  		{name: "CLZW", argLength: 1, reg: gp11, asm: "CLZW"},                                  // count leading zero, 32-bit
   268  		{name: "VCNT", argLength: 1, reg: fp11, asm: "VCNT"},                                  // count set bits for each 8-bit unit and store the result in each 8-bit unit
   269  		{name: "VUADDLV", argLength: 1, reg: fp11, asm: "VUADDLV"},                            // unsigned sum of eight bytes in a 64-bit value, zero extended to 64-bit.
   270  		{name: "LoweredRound32F", argLength: 1, reg: fp11, resultInArg0: true, zeroWidth: true},
   271  		{name: "LoweredRound64F", argLength: 1, reg: fp11, resultInArg0: true, zeroWidth: true},
   272  
   273  		// 3-operand, the addend comes first
   274  		{name: "FMADDS", argLength: 3, reg: fp31, asm: "FMADDS"},   // +arg0 + (arg1 * arg2)
   275  		{name: "FMADDD", argLength: 3, reg: fp31, asm: "FMADDD"},   // +arg0 + (arg1 * arg2)
   276  		{name: "FNMADDS", argLength: 3, reg: fp31, asm: "FNMADDS"}, // -arg0 - (arg1 * arg2)
   277  		{name: "FNMADDD", argLength: 3, reg: fp31, asm: "FNMADDD"}, // -arg0 - (arg1 * arg2)
   278  		{name: "FMSUBS", argLength: 3, reg: fp31, asm: "FMSUBS"},   // +arg0 - (arg1 * arg2)
   279  		{name: "FMSUBD", argLength: 3, reg: fp31, asm: "FMSUBD"},   // +arg0 - (arg1 * arg2)
   280  		{name: "FNMSUBS", argLength: 3, reg: fp31, asm: "FNMSUBS"}, // -arg0 + (arg1 * arg2)
   281  		{name: "FNMSUBD", argLength: 3, reg: fp31, asm: "FNMSUBD"}, // -arg0 + (arg1 * arg2)
   282  		{name: "MADD", argLength: 3, reg: gp31, asm: "MADD"},       // +arg0 + (arg1 * arg2)
   283  		{name: "MADDW", argLength: 3, reg: gp31, asm: "MADDW"},     // +arg0 + (arg1 * arg2), 32-bit
   284  		{name: "MSUB", argLength: 3, reg: gp31, asm: "MSUB"},       // +arg0 - (arg1 * arg2)
   285  		{name: "MSUBW", argLength: 3, reg: gp31, asm: "MSUBW"},     // +arg0 - (arg1 * arg2), 32-bit
   286  
   287  		// shifts
   288  		{name: "SLL", argLength: 2, reg: gp21, asm: "LSL"},                        // arg0 << arg1, shift amount is mod 64
   289  		{name: "SLLconst", argLength: 1, reg: gp11, asm: "LSL", aux: "Int64"},     // arg0 << auxInt, auxInt should be in the range 0 to 63.
   290  		{name: "SRL", argLength: 2, reg: gp21, asm: "LSR"},                        // arg0 >> arg1, unsigned, shift amount is mod 64
   291  		{name: "SRLconst", argLength: 1, reg: gp11, asm: "LSR", aux: "Int64"},     // arg0 >> auxInt, unsigned, auxInt should be in the range 0 to 63.
   292  		{name: "SRA", argLength: 2, reg: gp21, asm: "ASR"},                        // arg0 >> arg1, signed, shift amount is mod 64
   293  		{name: "SRAconst", argLength: 1, reg: gp11, asm: "ASR", aux: "Int64"},     // arg0 >> auxInt, signed, auxInt should be in the range 0 to 63.
   294  		{name: "ROR", argLength: 2, reg: gp21, asm: "ROR"},                        // arg0 right rotate by (arg1 mod 64) bits
   295  		{name: "RORW", argLength: 2, reg: gp21, asm: "RORW"},                      // arg0 right rotate by (arg1 mod 32) bits
   296  		{name: "RORconst", argLength: 1, reg: gp11, asm: "ROR", aux: "Int64"},     // arg0 right rotate by auxInt bits, auxInt should be in the range 0 to 63.
   297  		{name: "RORWconst", argLength: 1, reg: gp11, asm: "RORW", aux: "Int64"},   // uint32(arg0) right rotate by auxInt bits, auxInt should be in the range 0 to 31.
   298  		{name: "EXTRconst", argLength: 2, reg: gp21, asm: "EXTR", aux: "Int64"},   // extract 64 bits from arg0:arg1 starting at lsb auxInt, auxInt should be in the range 0 to 63.
   299  		{name: "EXTRWconst", argLength: 2, reg: gp21, asm: "EXTRW", aux: "Int64"}, // extract 32 bits from arg0[31:0]:arg1[31:0] starting at lsb auxInt and zero top 32 bits, auxInt should be in the range 0 to 31.
   300  
   301  		// comparisons
   302  		{name: "CMP", argLength: 2, reg: gp2flags, asm: "CMP", typ: "Flags"},                      // arg0 compare to arg1
   303  		{name: "CMPconst", argLength: 1, reg: gp1flags, asm: "CMP", aux: "Int64", typ: "Flags"},   // arg0 compare to auxInt
   304  		{name: "CMPW", argLength: 2, reg: gp2flags, asm: "CMPW", typ: "Flags"},                    // arg0 compare to arg1, 32 bit
   305  		{name: "CMPWconst", argLength: 1, reg: gp1flags, asm: "CMPW", aux: "Int32", typ: "Flags"}, // arg0 compare to auxInt, 32 bit
   306  		{name: "CMN", argLength: 2, reg: gp2flags, asm: "CMN", typ: "Flags", commutative: true},   // arg0 compare to -arg1, provided arg1 is not 1<<63
   307  		{name: "CMNconst", argLength: 1, reg: gp1flags, asm: "CMN", aux: "Int64", typ: "Flags"},   // arg0 compare to -auxInt
   308  		{name: "CMNW", argLength: 2, reg: gp2flags, asm: "CMNW", typ: "Flags", commutative: true}, // arg0 compare to -arg1, 32 bit, provided arg1 is not 1<<31
   309  		{name: "CMNWconst", argLength: 1, reg: gp1flags, asm: "CMNW", aux: "Int32", typ: "Flags"}, // arg0 compare to -auxInt, 32 bit
   310  		{name: "TST", argLength: 2, reg: gp2flags, asm: "TST", typ: "Flags", commutative: true},   // arg0 & arg1 compare to 0
   311  		{name: "TSTconst", argLength: 1, reg: gp1flags, asm: "TST", aux: "Int64", typ: "Flags"},   // arg0 & auxInt compare to 0
   312  		{name: "TSTW", argLength: 2, reg: gp2flags, asm: "TSTW", typ: "Flags", commutative: true}, // arg0 & arg1 compare to 0, 32 bit
   313  		{name: "TSTWconst", argLength: 1, reg: gp1flags, asm: "TSTW", aux: "Int32", typ: "Flags"}, // arg0 & auxInt compare to 0, 32 bit
   314  		{name: "FCMPS", argLength: 2, reg: fp2flags, asm: "FCMPS", typ: "Flags"},                  // arg0 compare to arg1, float32
   315  		{name: "FCMPD", argLength: 2, reg: fp2flags, asm: "FCMPD", typ: "Flags"},                  // arg0 compare to arg1, float64
   316  		{name: "FCMPS0", argLength: 1, reg: fp1flags, asm: "FCMPS", typ: "Flags"},                 // arg0 compare to 0, float32
   317  		{name: "FCMPD0", argLength: 1, reg: fp1flags, asm: "FCMPD", typ: "Flags"},                 // arg0 compare to 0, float64
   318  
   319  		// shifted ops
   320  		{name: "MVNshiftLL", argLength: 1, reg: gp11, asm: "MVN", aux: "Int64"},                   // ^(arg0<<auxInt), auxInt should be in the range 0 to 63.
   321  		{name: "MVNshiftRL", argLength: 1, reg: gp11, asm: "MVN", aux: "Int64"},                   // ^(arg0>>auxInt), unsigned shift, auxInt should be in the range 0 to 63.
   322  		{name: "MVNshiftRA", argLength: 1, reg: gp11, asm: "MVN", aux: "Int64"},                   // ^(arg0>>auxInt), signed shift, auxInt should be in the range 0 to 63.
   323  		{name: "MVNshiftRO", argLength: 1, reg: gp11, asm: "MVN", aux: "Int64"},                   // ^(arg0 ROR auxInt), signed shift, auxInt should be in the range 0 to 63.
   324  		{name: "NEGshiftLL", argLength: 1, reg: gp11, asm: "NEG", aux: "Int64"},                   // -(arg0<<auxInt), auxInt should be in the range 0 to 63.
   325  		{name: "NEGshiftRL", argLength: 1, reg: gp11, asm: "NEG", aux: "Int64"},                   // -(arg0>>auxInt), unsigned shift, auxInt should be in the range 0 to 63.
   326  		{name: "NEGshiftRA", argLength: 1, reg: gp11, asm: "NEG", aux: "Int64"},                   // -(arg0>>auxInt), signed shift, auxInt should be in the range 0 to 63.
   327  		{name: "ADDshiftLL", argLength: 2, reg: gp21, asm: "ADD", aux: "Int64"},                   // arg0 + arg1<<auxInt, auxInt should be in the range 0 to 63.
   328  		{name: "ADDshiftRL", argLength: 2, reg: gp21, asm: "ADD", aux: "Int64"},                   // arg0 + arg1>>auxInt, unsigned shift, auxInt should be in the range 0 to 63.
   329  		{name: "ADDshiftRA", argLength: 2, reg: gp21, asm: "ADD", aux: "Int64"},                   // arg0 + arg1>>auxInt, signed shift, auxInt should be in the range 0 to 63.
   330  		{name: "SUBshiftLL", argLength: 2, reg: gp21, asm: "SUB", aux: "Int64"},                   // arg0 - arg1<<auxInt, auxInt should be in the range 0 to 63.
   331  		{name: "SUBshiftRL", argLength: 2, reg: gp21, asm: "SUB", aux: "Int64"},                   // arg0 - arg1>>auxInt, unsigned shift, auxInt should be in the range 0 to 63.
   332  		{name: "SUBshiftRA", argLength: 2, reg: gp21, asm: "SUB", aux: "Int64"},                   // arg0 - arg1>>auxInt, signed shift, auxInt should be in the range 0 to 63.
   333  		{name: "ANDshiftLL", argLength: 2, reg: gp21, asm: "AND", aux: "Int64"},                   // arg0 & (arg1<<auxInt), auxInt should be in the range 0 to 63.
   334  		{name: "ANDshiftRL", argLength: 2, reg: gp21, asm: "AND", aux: "Int64"},                   // arg0 & (arg1>>auxInt), unsigned shift, auxInt should be in the range 0 to 63.
   335  		{name: "ANDshiftRA", argLength: 2, reg: gp21, asm: "AND", aux: "Int64"},                   // arg0 & (arg1>>auxInt), signed shift, auxInt should be in the range 0 to 63.
   336  		{name: "ANDshiftRO", argLength: 2, reg: gp21, asm: "AND", aux: "Int64"},                   // arg0 & (arg1 ROR auxInt), signed shift, auxInt should be in the range 0 to 63.
   337  		{name: "ORshiftLL", argLength: 2, reg: gp21, asm: "ORR", aux: "Int64"},                    // arg0 | arg1<<auxInt, auxInt should be in the range 0 to 63.
   338  		{name: "ORshiftRL", argLength: 2, reg: gp21, asm: "ORR", aux: "Int64"},                    // arg0 | arg1>>auxInt, unsigned shift, auxInt should be in the range 0 to 63.
   339  		{name: "ORshiftRA", argLength: 2, reg: gp21, asm: "ORR", aux: "Int64"},                    // arg0 | arg1>>auxInt, signed shift, auxInt should be in the range 0 to 63.
   340  		{name: "ORshiftRO", argLength: 2, reg: gp21, asm: "ORR", aux: "Int64"},                    // arg0 | arg1 ROR auxInt, signed shift, auxInt should be in the range 0 to 63.
   341  		{name: "XORshiftLL", argLength: 2, reg: gp21, asm: "EOR", aux: "Int64"},                   // arg0 ^ arg1<<auxInt, auxInt should be in the range 0 to 63.
   342  		{name: "XORshiftRL", argLength: 2, reg: gp21, asm: "EOR", aux: "Int64"},                   // arg0 ^ arg1>>auxInt, unsigned shift, auxInt should be in the range 0 to 63.
   343  		{name: "XORshiftRA", argLength: 2, reg: gp21, asm: "EOR", aux: "Int64"},                   // arg0 ^ arg1>>auxInt, signed shift, auxInt should be in the range 0 to 63.
   344  		{name: "XORshiftRO", argLength: 2, reg: gp21, asm: "EOR", aux: "Int64"},                   // arg0 ^ arg1 ROR auxInt, signed shift, auxInt should be in the range 0 to 63.
   345  		{name: "BICshiftLL", argLength: 2, reg: gp21, asm: "BIC", aux: "Int64"},                   // arg0 &^ (arg1<<auxInt), auxInt should be in the range 0 to 63.
   346  		{name: "BICshiftRL", argLength: 2, reg: gp21, asm: "BIC", aux: "Int64"},                   // arg0 &^ (arg1>>auxInt), unsigned shift, auxInt should be in the range 0 to 63.
   347  		{name: "BICshiftRA", argLength: 2, reg: gp21, asm: "BIC", aux: "Int64"},                   // arg0 &^ (arg1>>auxInt), signed shift, auxInt should be in the range 0 to 63.
   348  		{name: "BICshiftRO", argLength: 2, reg: gp21, asm: "BIC", aux: "Int64"},                   // arg0 &^ (arg1 ROR auxInt), signed shift, auxInt should be in the range 0 to 63.
   349  		{name: "EONshiftLL", argLength: 2, reg: gp21, asm: "EON", aux: "Int64"},                   // arg0 ^ ^(arg1<<auxInt), auxInt should be in the range 0 to 63.
   350  		{name: "EONshiftRL", argLength: 2, reg: gp21, asm: "EON", aux: "Int64"},                   // arg0 ^ ^(arg1>>auxInt), unsigned shift, auxInt should be in the range 0 to 63.
   351  		{name: "EONshiftRA", argLength: 2, reg: gp21, asm: "EON", aux: "Int64"},                   // arg0 ^ ^(arg1>>auxInt), signed shift, auxInt should be in the range 0 to 63.
   352  		{name: "EONshiftRO", argLength: 2, reg: gp21, asm: "EON", aux: "Int64"},                   // arg0 ^ ^(arg1 ROR auxInt), signed shift, auxInt should be in the range 0 to 63.
   353  		{name: "ORNshiftLL", argLength: 2, reg: gp21, asm: "ORN", aux: "Int64"},                   // arg0 | ^(arg1<<auxInt), auxInt should be in the range 0 to 63.
   354  		{name: "ORNshiftRL", argLength: 2, reg: gp21, asm: "ORN", aux: "Int64"},                   // arg0 | ^(arg1>>auxInt), unsigned shift, auxInt should be in the range 0 to 63.
   355  		{name: "ORNshiftRA", argLength: 2, reg: gp21, asm: "ORN", aux: "Int64"},                   // arg0 | ^(arg1>>auxInt), signed shift, auxInt should be in the range 0 to 63.
   356  		{name: "ORNshiftRO", argLength: 2, reg: gp21, asm: "ORN", aux: "Int64"},                   // arg0 | ^(arg1 ROR auxInt), signed shift, auxInt should be in the range 0 to 63.
   357  		{name: "CMPshiftLL", argLength: 2, reg: gp2flags, asm: "CMP", aux: "Int64", typ: "Flags"}, // arg0 compare to arg1<<auxInt, auxInt should be in the range 0 to 63.
   358  		{name: "CMPshiftRL", argLength: 2, reg: gp2flags, asm: "CMP", aux: "Int64", typ: "Flags"}, // arg0 compare to arg1>>auxInt, unsigned shift, auxInt should be in the range 0 to 63.
   359  		{name: "CMPshiftRA", argLength: 2, reg: gp2flags, asm: "CMP", aux: "Int64", typ: "Flags"}, // arg0 compare to arg1>>auxInt, signed shift, auxInt should be in the range 0 to 63.
   360  		{name: "CMNshiftLL", argLength: 2, reg: gp2flags, asm: "CMN", aux: "Int64", typ: "Flags"}, // (arg0 + arg1<<auxInt) compare to 0, auxInt should be in the range 0 to 63.
   361  		{name: "CMNshiftRL", argLength: 2, reg: gp2flags, asm: "CMN", aux: "Int64", typ: "Flags"}, // (arg0 + arg1>>auxInt) compare to 0, unsigned shift, auxInt should be in the range 0 to 63.
   362  		{name: "CMNshiftRA", argLength: 2, reg: gp2flags, asm: "CMN", aux: "Int64", typ: "Flags"}, // (arg0 + arg1>>auxInt) compare to 0, signed shift, auxInt should be in the range 0 to 63.
   363  		{name: "TSTshiftLL", argLength: 2, reg: gp2flags, asm: "TST", aux: "Int64", typ: "Flags"}, // (arg0 & arg1<<auxInt) compare to 0, auxInt should be in the range 0 to 63.
   364  		{name: "TSTshiftRL", argLength: 2, reg: gp2flags, asm: "TST", aux: "Int64", typ: "Flags"}, // (arg0 & arg1>>auxInt) compare to 0, unsigned shift, auxInt should be in the range 0 to 63.
   365  		{name: "TSTshiftRA", argLength: 2, reg: gp2flags, asm: "TST", aux: "Int64", typ: "Flags"}, // (arg0 & arg1>>auxInt) compare to 0, signed shift, auxInt should be in the range 0 to 63.
   366  		{name: "TSTshiftRO", argLength: 2, reg: gp2flags, asm: "TST", aux: "Int64", typ: "Flags"}, // (arg0 & arg1 ROR auxInt) compare to 0, signed shift, auxInt should be in the range 0 to 63.
   367  
   368  		// bitfield ops
   369  		// for all bitfield ops lsb is auxInt>>8, width is auxInt&0xff
   370  		// insert low width bits of arg1 into the result starting at bit lsb, copy other bits from arg0
   371  		{name: "BFI", argLength: 2, reg: gp21nog, asm: "BFI", aux: "ARM64BitField", resultInArg0: true},
   372  		// extract width bits of arg1 starting at bit lsb and insert at low end of result, copy other bits from arg0
   373  		{name: "BFXIL", argLength: 2, reg: gp21nog, asm: "BFXIL", aux: "ARM64BitField", resultInArg0: true},
   374  		// insert low width bits of arg0 into the result starting at bit lsb, bits to the left of the inserted bit field are set to the high/sign bit of the inserted bit field, bits to the right are zeroed
   375  		{name: "SBFIZ", argLength: 1, reg: gp11, asm: "SBFIZ", aux: "ARM64BitField"},
   376  		// extract width bits of arg0 starting at bit lsb and insert at low end of result, remaining high bits are set to the high/sign bit of the extracted bitfield
   377  		{name: "SBFX", argLength: 1, reg: gp11, asm: "SBFX", aux: "ARM64BitField"},
   378  		// insert low width bits of arg0 into the result starting at bit lsb, bits to the left and right of the inserted bit field are zeroed
   379  		{name: "UBFIZ", argLength: 1, reg: gp11, asm: "UBFIZ", aux: "ARM64BitField"},
   380  		// extract width bits of arg0 starting at bit lsb and insert at low end of result, remaining high bits are zeroed
   381  		{name: "UBFX", argLength: 1, reg: gp11, asm: "UBFX", aux: "ARM64BitField"},
   382  
   383  		// moves
   384  		{name: "MOVDconst", argLength: 0, reg: gp01, aux: "Int64", asm: "MOVD", typ: "UInt64", rematerializeable: true},      // 64 bits from auxint
   385  		{name: "FMOVSconst", argLength: 0, reg: fp01, aux: "Float64", asm: "FMOVS", typ: "Float32", rematerializeable: true}, // auxint as 64-bit float, convert to 32-bit float
   386  		{name: "FMOVDconst", argLength: 0, reg: fp01, aux: "Float64", asm: "FMOVD", typ: "Float64", rematerializeable: true}, // auxint as 64-bit float
   387  
   388  		{name: "MOVDaddr", argLength: 1, reg: regInfo{inputs: []regMask{buildReg("SP") | buildReg("SB")}, outputs: []regMask{gp}}, aux: "SymOff", asm: "MOVD", rematerializeable: true, symEffect: "Addr"}, // arg0 + auxInt + aux.(*gc.Sym), arg0=SP/SB
   389  
   390  		{name: "MOVBload", argLength: 2, reg: gpload, aux: "SymOff", asm: "MOVB", typ: "Int8", faultOnNilArg0: true, symEffect: "Read"},      // load from arg0 + auxInt + aux.  arg1=mem.
   391  		{name: "MOVBUload", argLength: 2, reg: gpload, aux: "SymOff", asm: "MOVBU", typ: "UInt8", faultOnNilArg0: true, symEffect: "Read"},   // load from arg0 + auxInt + aux.  arg1=mem.
   392  		{name: "MOVHload", argLength: 2, reg: gpload, aux: "SymOff", asm: "MOVH", typ: "Int16", faultOnNilArg0: true, symEffect: "Read"},     // load from arg0 + auxInt + aux.  arg1=mem.
   393  		{name: "MOVHUload", argLength: 2, reg: gpload, aux: "SymOff", asm: "MOVHU", typ: "UInt16", faultOnNilArg0: true, symEffect: "Read"},  // load from arg0 + auxInt + aux.  arg1=mem.
   394  		{name: "MOVWload", argLength: 2, reg: gpload, aux: "SymOff", asm: "MOVW", typ: "Int32", faultOnNilArg0: true, symEffect: "Read"},     // load from arg0 + auxInt + aux.  arg1=mem.
   395  		{name: "MOVWUload", argLength: 2, reg: gpload, aux: "SymOff", asm: "MOVWU", typ: "UInt32", faultOnNilArg0: true, symEffect: "Read"},  // load from arg0 + auxInt + aux.  arg1=mem.
   396  		{name: "MOVDload", argLength: 2, reg: gpload, aux: "SymOff", asm: "MOVD", typ: "UInt64", faultOnNilArg0: true, symEffect: "Read"},    // load from arg0 + auxInt + aux.  arg1=mem.
   397  		{name: "FMOVSload", argLength: 2, reg: fpload, aux: "SymOff", asm: "FMOVS", typ: "Float32", faultOnNilArg0: true, symEffect: "Read"}, // load from arg0 + auxInt + aux.  arg1=mem.
   398  		{name: "FMOVDload", argLength: 2, reg: fpload, aux: "SymOff", asm: "FMOVD", typ: "Float64", faultOnNilArg0: true, symEffect: "Read"}, // load from arg0 + auxInt + aux.  arg1=mem.
   399  		{name: "FMOVQload", argLength: 2, reg: fpload, aux: "SymOff", asm: "FMOVQ", typ: "Vec128", faultOnNilArg0: true, symEffect: "Read"},  // load from arg0 + auxInt + aux.  arg1=mem.
   400  
   401  		// LDP instructions load the contents of two adjacent locations in memory into registers.
   402  		// Address to start loading is addr = arg0 + auxInt + aux.
   403  		// x := *(*T)(addr)
   404  		// y := *(*T)(addr+sizeof(T))
   405  		// arg1=mem
   406  		// Returns the tuple <x,y>.
   407  		{name: "LDP", argLength: 2, reg: gpload2, aux: "SymOff", asm: "LDP", typ: "(UInt64,UInt64)", faultOnNilArg0: true, symEffect: "Read"},       // T=int64 (gp reg destination)
   408  		{name: "LDPW", argLength: 2, reg: gpload2, aux: "SymOff", asm: "LDPW", typ: "(UInt32,UInt32)", faultOnNilArg0: true, symEffect: "Read"},     // T=int32 (gp reg destination) unsigned extension
   409  		{name: "LDPSW", argLength: 2, reg: gpload2, aux: "SymOff", asm: "LDPSW", typ: "(Int32,Int32)", faultOnNilArg0: true, symEffect: "Read"},     // T=int32 (gp reg destination) signed extension
   410  		{name: "FLDPD", argLength: 2, reg: fpload2, aux: "SymOff", asm: "FLDPD", typ: "(Float64,Float64)", faultOnNilArg0: true, symEffect: "Read"}, // T=float64 (fp reg destination)
   411  		{name: "FLDPS", argLength: 2, reg: fpload2, aux: "SymOff", asm: "FLDPS", typ: "(Float32,Float32)", faultOnNilArg0: true, symEffect: "Read"}, // T=float32 (fp reg destination)
   412  		{name: "FLDPQ", argLength: 2, reg: fpload2, aux: "SymOff", asm: "FLDPQ", typ: "(Vec128,Vec128)", faultOnNilArg0: true, symEffect: "Read"},   // T=vec128 (fp reg destination)
   413  
   414  		// register indexed load
   415  		{name: "MOVDloadidx", argLength: 3, reg: gp2load, asm: "MOVD", typ: "UInt64"},    // load 64-bit dword from arg0 + arg1, arg2 = mem.
   416  		{name: "MOVWloadidx", argLength: 3, reg: gp2load, asm: "MOVW", typ: "Int32"},     // load 32-bit word from arg0 + arg1, sign-extended to 64-bit, arg2=mem.
   417  		{name: "MOVWUloadidx", argLength: 3, reg: gp2load, asm: "MOVWU", typ: "UInt32"},  // load 32-bit word from arg0 + arg1, zero-extended to 64-bit, arg2=mem.
   418  		{name: "MOVHloadidx", argLength: 3, reg: gp2load, asm: "MOVH", typ: "Int16"},     // load 16-bit word from arg0 + arg1, sign-extended to 64-bit, arg2=mem.
   419  		{name: "MOVHUloadidx", argLength: 3, reg: gp2load, asm: "MOVHU", typ: "UInt16"},  // load 16-bit word from arg0 + arg1, zero-extended to 64-bit, arg2=mem.
   420  		{name: "MOVBloadidx", argLength: 3, reg: gp2load, asm: "MOVB", typ: "Int8"},      // load 8-bit word from arg0 + arg1, sign-extended to 64-bit, arg2=mem.
   421  		{name: "MOVBUloadidx", argLength: 3, reg: gp2load, asm: "MOVBU", typ: "UInt8"},   // load 8-bit word from arg0 + arg1, zero-extended to 64-bit, arg2=mem.
   422  		{name: "FMOVSloadidx", argLength: 3, reg: fp2load, asm: "FMOVS", typ: "Float32"}, // load 32-bit float from arg0 + arg1, arg2=mem.
   423  		{name: "FMOVDloadidx", argLength: 3, reg: fp2load, asm: "FMOVD", typ: "Float64"}, // load 64-bit float from arg0 + arg1, arg2=mem.
   424  
   425  		// shifted register indexed load
   426  		{name: "MOVHloadidx2", argLength: 3, reg: gp2load, asm: "MOVH", typ: "Int16"},     // load 16-bit half-word from arg0 + arg1*2, sign-extended to 64-bit, arg2=mem.
   427  		{name: "MOVHUloadidx2", argLength: 3, reg: gp2load, asm: "MOVHU", typ: "UInt16"},  // load 16-bit half-word from arg0 + arg1*2, zero-extended to 64-bit, arg2=mem.
   428  		{name: "MOVWloadidx4", argLength: 3, reg: gp2load, asm: "MOVW", typ: "Int32"},     // load 32-bit word from arg0 + arg1*4, sign-extended to 64-bit, arg2=mem.
   429  		{name: "MOVWUloadidx4", argLength: 3, reg: gp2load, asm: "MOVWU", typ: "UInt32"},  // load 32-bit word from arg0 + arg1*4, zero-extended to 64-bit, arg2=mem.
   430  		{name: "MOVDloadidx8", argLength: 3, reg: gp2load, asm: "MOVD", typ: "UInt64"},    // load 64-bit double-word from arg0 + arg1*8, arg2 = mem.
   431  		{name: "FMOVSloadidx4", argLength: 3, reg: fp2load, asm: "FMOVS", typ: "Float32"}, // load 32-bit float from arg0 + arg1*4, arg2 = mem.
   432  		{name: "FMOVDloadidx8", argLength: 3, reg: fp2load, asm: "FMOVD", typ: "Float64"}, // load 64-bit float from arg0 + arg1*8, arg2 = mem.
   433  
   434  		{name: "MOVBstore", argLength: 3, reg: gpstore, aux: "SymOff", asm: "MOVB", typ: "Mem", faultOnNilArg0: true, symEffect: "Write"},   // store 1 byte of arg1 to arg0 + auxInt + aux.  arg2=mem.
   435  		{name: "MOVHstore", argLength: 3, reg: gpstore, aux: "SymOff", asm: "MOVH", typ: "Mem", faultOnNilArg0: true, symEffect: "Write"},   // store 2 bytes of arg1 to arg0 + auxInt + aux.  arg2=mem.
   436  		{name: "MOVWstore", argLength: 3, reg: gpstore, aux: "SymOff", asm: "MOVW", typ: "Mem", faultOnNilArg0: true, symEffect: "Write"},   // store 4 bytes of arg1 to arg0 + auxInt + aux.  arg2=mem.
   437  		{name: "MOVDstore", argLength: 3, reg: gpstore, aux: "SymOff", asm: "MOVD", typ: "Mem", faultOnNilArg0: true, symEffect: "Write"},   // store 8 bytes of arg1 to arg0 + auxInt + aux.  arg2=mem.
   438  		{name: "FMOVSstore", argLength: 3, reg: fpstore, aux: "SymOff", asm: "FMOVS", typ: "Mem", faultOnNilArg0: true, symEffect: "Write"}, // store 4 bytes of arg1 to arg0 + auxInt + aux.  arg2=mem.
   439  		{name: "FMOVDstore", argLength: 3, reg: fpstore, aux: "SymOff", asm: "FMOVD", typ: "Mem", faultOnNilArg0: true, symEffect: "Write"}, // store 8 bytes of arg1 to arg0 + auxInt + aux.  arg2=mem.
   440  		{name: "FMOVQstore", argLength: 3, reg: fpstore, aux: "SymOff", asm: "FMOVQ", typ: "Mem", faultOnNilArg0: true, symEffect: "Write"}, // store 16 bytes of arg1 to arg0 + auxInt + aux.  arg2=mem.
   441  
   442  		// STP instructions store the contents of two registers to adjacent locations in memory.
   443  		// Address to start storing is addr = arg0 + auxInt + aux.
   444  		// *(*T)(addr) = arg1
   445  		// *(*T)(addr+sizeof(T)) = arg2
   446  		// arg3=mem. Returns mem.
   447  		{name: "STP", argLength: 4, reg: gpstore2, aux: "SymOff", asm: "STP", typ: "Mem", faultOnNilArg0: true, symEffect: "Write"},     // T=int64 (gp reg source)
   448  		{name: "STPW", argLength: 4, reg: gpstore2, aux: "SymOff", asm: "STPW", typ: "Mem", faultOnNilArg0: true, symEffect: "Write"},   // T=int32 (gp reg source)
   449  		{name: "FSTPD", argLength: 4, reg: fpstore2, aux: "SymOff", asm: "FSTPD", typ: "Mem", faultOnNilArg0: true, symEffect: "Write"}, // T=float64 (fp reg source)
   450  		{name: "FSTPS", argLength: 4, reg: fpstore2, aux: "SymOff", asm: "FSTPS", typ: "Mem", faultOnNilArg0: true, symEffect: "Write"}, // T=float32 (fp reg source)
   451  		{name: "FSTPQ", argLength: 4, reg: fpstore2, aux: "SymOff", asm: "FSTPQ", typ: "Mem", faultOnNilArg0: true, symEffect: "Write"}, // T=vec128 (fp reg source)
   452  
   453  		// register indexed store
   454  		{name: "MOVBstoreidx", argLength: 4, reg: gpstore2, asm: "MOVB", typ: "Mem"},     // store 1 byte of arg2 to arg0 + arg1, arg3 = mem.
   455  		{name: "MOVHstoreidx", argLength: 4, reg: gpstore2, asm: "MOVH", typ: "Mem"},     // store 2 bytes of arg2 to arg0 + arg1, arg3 = mem.
   456  		{name: "MOVWstoreidx", argLength: 4, reg: gpstore2, asm: "MOVW", typ: "Mem"},     // store 4 bytes of arg2 to arg0 + arg1, arg3 = mem.
   457  		{name: "MOVDstoreidx", argLength: 4, reg: gpstore2, asm: "MOVD", typ: "Mem"},     // store 8 bytes of arg2 to arg0 + arg1, arg3 = mem.
   458  		{name: "FMOVSstoreidx", argLength: 4, reg: fpstoreidx, asm: "FMOVS", typ: "Mem"}, // store 32-bit float of arg2 to arg0 + arg1, arg3=mem.
   459  		{name: "FMOVDstoreidx", argLength: 4, reg: fpstoreidx, asm: "FMOVD", typ: "Mem"}, // store 64-bit float of arg2 to arg0 + arg1, arg3=mem.
   460  
   461  		// shifted register indexed store
   462  		{name: "MOVHstoreidx2", argLength: 4, reg: gpstore2, asm: "MOVH", typ: "Mem"},     // store 2 bytes of arg2 to arg0 + arg1*2, arg3 = mem.
   463  		{name: "MOVWstoreidx4", argLength: 4, reg: gpstore2, asm: "MOVW", typ: "Mem"},     // store 4 bytes of arg2 to arg0 + arg1*4, arg3 = mem.
   464  		{name: "MOVDstoreidx8", argLength: 4, reg: gpstore2, asm: "MOVD", typ: "Mem"},     // store 8 bytes of arg2 to arg0 + arg1*8, arg3 = mem.
   465  		{name: "FMOVSstoreidx4", argLength: 4, reg: fpstoreidx, asm: "FMOVS", typ: "Mem"}, // store 32-bit float of arg2 to arg0 + arg1*4, arg3=mem.
   466  		{name: "FMOVDstoreidx8", argLength: 4, reg: fpstoreidx, asm: "FMOVD", typ: "Mem"}, // store 64-bit float of arg2 to arg0 + arg1*8, arg3=mem.
   467  
   468  		{name: "FMOVDgpfp", argLength: 1, reg: gpfp, asm: "FMOVD"}, // move int64 to float64 (no conversion)
   469  		{name: "FMOVDfpgp", argLength: 1, reg: fpgp, asm: "FMOVD"}, // move float64 to int64 (no conversion)
   470  		{name: "FMOVSgpfp", argLength: 1, reg: gpfp, asm: "FMOVS"}, // move 32bits from int to float reg (no conversion)
   471  		{name: "FMOVSfpgp", argLength: 1, reg: fpgp, asm: "FMOVS"}, // move 32bits from float to int reg, zero extend (no conversion)
   472  
   473  		// conversions
   474  		{name: "MOVBreg", argLength: 1, reg: gp11, asm: "MOVB"},   // move from arg0, sign-extended from byte
   475  		{name: "MOVBUreg", argLength: 1, reg: gp11, asm: "MOVBU"}, // move from arg0, unsign-extended from byte
   476  		{name: "MOVHreg", argLength: 1, reg: gp11, asm: "MOVH"},   // move from arg0, sign-extended from half
   477  		{name: "MOVHUreg", argLength: 1, reg: gp11, asm: "MOVHU"}, // move from arg0, unsign-extended from half
   478  		{name: "MOVWreg", argLength: 1, reg: gp11, asm: "MOVW"},   // move from arg0, sign-extended from word
   479  		{name: "MOVWUreg", argLength: 1, reg: gp11, asm: "MOVWU"}, // move from arg0, unsign-extended from word
   480  		{name: "MOVDreg", argLength: 1, reg: gp11, asm: "MOVD"},   // move from arg0
   481  
   482  		{name: "MOVDnop", argLength: 1, reg: regInfo{inputs: []regMask{gp}, outputs: []regMask{gp}}, resultInArg0: true}, // nop, return arg0 in same register
   483  
   484  		{name: "SCVTFWS", argLength: 1, reg: gpfp, asm: "SCVTFWS"},   // int32 -> float32
   485  		{name: "SCVTFWD", argLength: 1, reg: gpfp, asm: "SCVTFWD"},   // int32 -> float64
   486  		{name: "UCVTFWS", argLength: 1, reg: gpfp, asm: "UCVTFWS"},   // uint32 -> float32
   487  		{name: "UCVTFWD", argLength: 1, reg: gpfp, asm: "UCVTFWD"},   // uint32 -> float64
   488  		{name: "SCVTFS", argLength: 1, reg: gpfp, asm: "SCVTFS"},     // int64 -> float32
   489  		{name: "SCVTFD", argLength: 1, reg: gpfp, asm: "SCVTFD"},     // int64 -> float64
   490  		{name: "UCVTFS", argLength: 1, reg: gpfp, asm: "UCVTFS"},     // uint64 -> float32
   491  		{name: "UCVTFD", argLength: 1, reg: gpfp, asm: "UCVTFD"},     // uint64 -> float64
   492  		{name: "FCVTZSSW", argLength: 1, reg: fpgp, asm: "FCVTZSSW"}, // float32 -> int32
   493  		{name: "FCVTZSDW", argLength: 1, reg: fpgp, asm: "FCVTZSDW"}, // float64 -> int32
   494  		{name: "FCVTZUSW", argLength: 1, reg: fpgp, asm: "FCVTZUSW"}, // float32 -> uint32
   495  		{name: "FCVTZUDW", argLength: 1, reg: fpgp, asm: "FCVTZUDW"}, // float64 -> uint32
   496  		{name: "FCVTZSS", argLength: 1, reg: fpgp, asm: "FCVTZSS"},   // float32 -> int64
   497  		{name: "FCVTZSD", argLength: 1, reg: fpgp, asm: "FCVTZSD"},   // float64 -> int64
   498  		{name: "FCVTZUS", argLength: 1, reg: fpgp, asm: "FCVTZUS"},   // float32 -> uint64
   499  		{name: "FCVTZUD", argLength: 1, reg: fpgp, asm: "FCVTZUD"},   // float64 -> uint64
   500  		{name: "FCVTSD", argLength: 1, reg: fp11, asm: "FCVTSD"},     // float32 -> float64
   501  		{name: "FCVTDS", argLength: 1, reg: fp11, asm: "FCVTDS"},     // float64 -> float32
   502  
   503  		// 64-bit floating-point round to integers in 64-bit FP format
   504  		{name: "FRINTAD", argLength: 1, reg: fp11, asm: "FRINTAD"}, // Round (ties Away from zero; 0.5 -> 1, -0.5 -> -1)
   505  		{name: "FRINTMD", argLength: 1, reg: fp11, asm: "FRINTMD"}, // Floor (towards Minus; 0.5 -> 0, -0.5 -> -1)
   506  		{name: "FRINTND", argLength: 1, reg: fp11, asm: "FRINTND"}, // Round (ties to even; ; 0.5 -> 0, 1.5 -> 2)
   507  		{name: "FRINTPD", argLength: 1, reg: fp11, asm: "FRINTPD"}, // Ceil (towards Positive; 0.5 -> 1, -0.5 -> 0)
   508  		{name: "FRINTZD", argLength: 1, reg: fp11, asm: "FRINTZD"}, // Trunc (towards Zero; 0.5 -> 0, -0.5 -> 0))
   509  		// 32-bit floating-point round to integers in 32-bit FP format
   510  		{name: "FRINTAS", argLength: 1, reg: fp11, asm: "FRINTAS"}, // Round (ties Away from zero; 0.5 -> 1, -0.5 -> -1)
   511  		{name: "FRINTMS", argLength: 1, reg: fp11, asm: "FRINTMS"}, // Floor (towards Minus; 0.5 -> 0, -0.5 -> -1)
   512  		{name: "FRINTNS", argLength: 1, reg: fp11, asm: "FRINTNS"}, // Round (ties to even; ; 0.5 -> 0, 1.5 -> 2)
   513  		{name: "FRINTPS", argLength: 1, reg: fp11, asm: "FRINTPS"}, // Ceil (towards Positive; 0.5 -> 1, -0.5 -> 0)
   514  		{name: "FRINTZS", argLength: 1, reg: fp11, asm: "FRINTZS"}, // Trunc (towards Zero; 0.5 -> 0, -0.5 -> 0))
   515  
   516  		// conditional instructions; auxint is
   517  		// one of the arm64 comparison pseudo-ops (LessThan, LessThanU, etc.)
   518  		{name: "CSEL", argLength: 3, reg: gp2flags1, asm: "CSEL", aux: "CCop"},   // auxint(flags) ? arg0 : arg1
   519  		{name: "CSEL0", argLength: 2, reg: gp1flags1, asm: "CSEL", aux: "CCop"},  // auxint(flags) ? arg0 : 0
   520  		{name: "CSINC", argLength: 3, reg: gp2flags1, asm: "CSINC", aux: "CCop"}, // auxint(flags) ? arg0 : arg1 + 1
   521  		{name: "CSINV", argLength: 3, reg: gp2flags1, asm: "CSINV", aux: "CCop"}, // auxint(flags) ? arg0 : ^arg1
   522  		{name: "CSNEG", argLength: 3, reg: gp2flags1, asm: "CSNEG", aux: "CCop"}, // auxint(flags) ? arg0 : -arg1
   523  		{name: "CSETM", argLength: 1, reg: readflags, asm: "CSETM", aux: "CCop"}, // auxint(flags) ? -1 : 0
   524  
   525  		// conditional comparison instructions; auxint is
   526  		// combination of Cond, Nzcv and optional ConstValue
   527  		// Behavior:
   528  		//   If the condition 'Cond' evaluates to true against current flags,
   529  		//   flags are set to the result of the comparison operation.
   530  		//   Otherwise, flags are set to the fallback value 'Nzcv'.
   531  		{name: "CCMP", argLength: 3, reg: gp2flagsflags, asm: "CCMP", aux: "ARM64ConditionalParams", typ: "Flags"},      // If Cond then flags = CMP arg0 arg1 else flags = Nzcv
   532  		{name: "CCMN", argLength: 3, reg: gp2flagsflags, asm: "CCMN", aux: "ARM64ConditionalParams", typ: "Flags"},      // If Cond then flags = CMN arg0 arg1 else flags = Nzcv
   533  		{name: "CCMPconst", argLength: 2, reg: gp1flagsflags, asm: "CCMP", aux: "ARM64ConditionalParams", typ: "Flags"}, // If Cond then flags = CMPconst [ConstValue] arg0 else flags = Nzcv
   534  		{name: "CCMNconst", argLength: 2, reg: gp1flagsflags, asm: "CCMN", aux: "ARM64ConditionalParams", typ: "Flags"}, // If Cond then flags = CMNconst [ConstValue] arg0 else flags = Nzcv
   535  
   536  		{name: "CCMPW", argLength: 3, reg: gp2flagsflags, asm: "CCMPW", aux: "ARM64ConditionalParams", typ: "Flags"},      // If Cond then flags = CMPW arg0 arg1 else flags = Nzcv
   537  		{name: "CCMNW", argLength: 3, reg: gp2flagsflags, asm: "CCMNW", aux: "ARM64ConditionalParams", typ: "Flags"},      // If Cond then flags = CMNW arg0 arg1 else flags = Nzcv
   538  		{name: "CCMPWconst", argLength: 2, reg: gp1flagsflags, asm: "CCMPW", aux: "ARM64ConditionalParams", typ: "Flags"}, // If Cond then flags = CCMPWconst [ConstValue] arg0 else flags = Nzcv
   539  		{name: "CCMNWconst", argLength: 2, reg: gp1flagsflags, asm: "CCMNW", aux: "ARM64ConditionalParams", typ: "Flags"}, // If Cond then flags = CCMNWconst [ConstValue] arg0 else flags = Nzcv
   540  
   541  		// function calls
   542  		{name: "CALLstatic", argLength: -1, reg: regInfo{clobbers: callerSave}, aux: "CallOff", clobberFlags: true, call: true},                                               // call static function aux.(*obj.LSym).  last arg=mem, auxint=argsize, returns mem
   543  		{name: "CALLtail", argLength: -1, reg: regInfo{clobbers: callerSave}, aux: "CallOff", clobberFlags: true, call: true, tailCall: true},                                 // tail call static function aux.(*obj.LSym).  last arg=mem, auxint=argsize, returns mem
   544  		{name: "CALLtailinter", argLength: -1, reg: regInfo{inputs: []regMask{gp}, clobbers: callerSave}, aux: "CallOff", clobberFlags: true, call: true, tailCall: true},     // tail call fn by pointer. arg0=codeptr, last arg=mem, auxint=argsize, returns mem
   545  		{name: "CALLclosure", argLength: -1, reg: regInfo{inputs: []regMask{gpsp, buildReg("R26"), 0}, clobbers: callerSave}, aux: "CallOff", clobberFlags: true, call: true}, // call function via closure.  arg0=codeptr, arg1=closure, last arg=mem, auxint=argsize, returns mem
   546  		{name: "CALLinter", argLength: -1, reg: regInfo{inputs: []regMask{gp}, clobbers: callerSave}, aux: "CallOff", clobberFlags: true, call: true},                         // call fn by pointer.  arg0=codeptr, last arg=mem, auxint=argsize, returns mem
   547  
   548  		// pseudo-ops
   549  		{name: "LoweredNilCheck", argLength: 2, reg: regInfo{inputs: []regMask{gpg}}, nilCheck: true, faultOnNilArg0: true},                                                                                                                                                      // panic if arg0 is nil.  arg1=mem.
   550  		{name: "LoweredMemEq", argLength: 4, reg: regInfo{inputs: []regMask{buildReg("R0"), buildReg("R1"), buildReg("R2")}, outputs: []regMask{buildReg("R0")}, clobbers: callerSave}, typ: "Bool", faultOnNilArg0: true, faultOnNilArg1: true, clobberFlags: true, call: true}, // arg0, arg1 - pointers to memory, arg2=size, arg3=mem.
   551  
   552  		{name: "Equal", argLength: 1, reg: readflags},            // bool, true flags encode x==y false otherwise.
   553  		{name: "NotEqual", argLength: 1, reg: readflags},         // bool, true flags encode x!=y false otherwise.
   554  		{name: "LessThan", argLength: 1, reg: readflags},         // bool, true flags encode signed x<y false otherwise.
   555  		{name: "LessEqual", argLength: 1, reg: readflags},        // bool, true flags encode signed x<=y false otherwise.
   556  		{name: "GreaterThan", argLength: 1, reg: readflags},      // bool, true flags encode signed x>y false otherwise.
   557  		{name: "GreaterEqual", argLength: 1, reg: readflags},     // bool, true flags encode signed x>=y false otherwise.
   558  		{name: "LessThanU", argLength: 1, reg: readflags},        // bool, true flags encode unsigned x<y false otherwise.
   559  		{name: "LessEqualU", argLength: 1, reg: readflags},       // bool, true flags encode unsigned x<=y false otherwise.
   560  		{name: "GreaterThanU", argLength: 1, reg: readflags},     // bool, true flags encode unsigned x>y false otherwise.
   561  		{name: "GreaterEqualU", argLength: 1, reg: readflags},    // bool, true flags encode unsigned x>=y false otherwise.
   562  		{name: "LessThanF", argLength: 1, reg: readflags},        // bool, true flags encode floating-point x<y false otherwise.
   563  		{name: "LessEqualF", argLength: 1, reg: readflags},       // bool, true flags encode floating-point x<=y false otherwise.
   564  		{name: "GreaterThanF", argLength: 1, reg: readflags},     // bool, true flags encode floating-point x>y false otherwise.
   565  		{name: "GreaterEqualF", argLength: 1, reg: readflags},    // bool, true flags encode floating-point x>=y false otherwise.
   566  		{name: "NotLessThanF", argLength: 1, reg: readflags},     // bool, true flags encode floating-point x>=y || x is unordered with y, false otherwise.
   567  		{name: "NotLessEqualF", argLength: 1, reg: readflags},    // bool, true flags encode floating-point x>y || x is unordered with y, false otherwise.
   568  		{name: "NotGreaterThanF", argLength: 1, reg: readflags},  // bool, true flags encode floating-point x<=y || x is unordered with y, false otherwise.
   569  		{name: "NotGreaterEqualF", argLength: 1, reg: readflags}, // bool, true flags encode floating-point x<y || x is unordered with y, false otherwise.
   570  		{name: "LessThanNoov", argLength: 1, reg: readflags},     // bool, true flags encode signed x<y but without honoring overflow, false otherwise.
   571  		{name: "GreaterEqualNoov", argLength: 1, reg: readflags}, // bool, true flags encode signed x>=y but without honoring overflow, false otherwise.
   572  
   573  		// medium zeroing
   574  		// arg0 = address of memory to zero
   575  		// arg1 = mem
   576  		// auxint = # of bytes to zero
   577  		// returns mem
   578  		{
   579  			name:      "LoweredZero",
   580  			aux:       "Int64",
   581  			argLength: 2,
   582  			reg: regInfo{
   583  				inputs: []regMask{gp},
   584  			},
   585  			faultOnNilArg0: true,
   586  		},
   587  
   588  		// large zeroing
   589  		// arg0 = address of memory to zero
   590  		// arg1 = mem
   591  		// auxint = # of bytes to zero
   592  		// returns mem
   593  		{
   594  			name:      "LoweredZeroLoop",
   595  			aux:       "Int64",
   596  			argLength: 2,
   597  			reg: regInfo{
   598  				inputs:       []regMask{gp},
   599  				clobbersArg0: true,
   600  			},
   601  			faultOnNilArg0: true,
   602  			needIntTemp:    true,
   603  		},
   604  
   605  		// medium copying
   606  		// arg0 = address of dst memory
   607  		// arg1 = address of src memory
   608  		// arg2 = mem
   609  		// auxint = # of bytes to copy
   610  		// returns mem
   611  		{
   612  			name:      "LoweredMove",
   613  			aux:       "Int64",
   614  			argLength: 3,
   615  			reg: regInfo{
   616  				inputs:   []regMask{gp &^ r25, gp &^ r25},
   617  				clobbers: r25 | f16to17, // TODO: figure out needIntTemp + x2 for floats
   618  			},
   619  			faultOnNilArg0: true,
   620  			faultOnNilArg1: true,
   621  		},
   622  
   623  		// large copying
   624  		// arg0 = address of dst memory
   625  		// arg1 = address of src memory
   626  		// arg2 = mem
   627  		// auxint = # of bytes to copy
   628  		// returns mem
   629  		{
   630  			name:      "LoweredMoveLoop",
   631  			aux:       "Int64",
   632  			argLength: 3,
   633  			reg: regInfo{
   634  				inputs:       []regMask{gp &^ r24to25, gp &^ r24to25},
   635  				clobbers:     r24to25 | f16to17, // TODO: figure out needIntTemp x2 + x2 for floats
   636  				clobbersArg0: true,
   637  				clobbersArg1: true,
   638  			},
   639  			faultOnNilArg0: true,
   640  			faultOnNilArg1: true,
   641  		},
   642  
   643  		// Scheduler ensures LoweredGetClosurePtr occurs only in entry block,
   644  		// and sorts it to the very beginning of the block to prevent other
   645  		// use of R26 (arm64.REGCTXT, the closure pointer)
   646  		{name: "LoweredGetClosurePtr", reg: regInfo{outputs: []regMask{buildReg("R26")}}, zeroWidth: true},
   647  
   648  		// LoweredGetCallerSP returns the SP of the caller of the current function. arg0=mem
   649  		{name: "LoweredGetCallerSP", argLength: 1, reg: gp01, rematerializeable: true},
   650  
   651  		// LoweredGetCallerPC evaluates to the PC to which its "caller" will return.
   652  		// I.e., if f calls g "calls" sys.GetCallerPC,
   653  		// the result should be the PC within f that g will return to.
   654  		// See runtime/stubs.go for a more detailed discussion.
   655  		{name: "LoweredGetCallerPC", reg: gp01, rematerializeable: true},
   656  
   657  		// Constant flag value.
   658  		// Note: there's an "unordered" outcome for floating-point
   659  		// comparisons, but we don't use such a beast yet.
   660  		// This op is for temporary use by rewrite rules. It
   661  		// cannot appear in the generated assembly.
   662  		{name: "FlagConstant", aux: "FlagConstant"},
   663  
   664  		// (InvertFlags (CMP a b)) == (CMP b a)
   665  		// InvertFlags is a pseudo-op which can't appear in assembly output.
   666  		{name: "InvertFlags", argLength: 1}, // reverse direction of arg0
   667  
   668  		// atomic loads.
   669  		// load from arg0. arg1=mem. auxint must be zero.
   670  		// returns <value,memory> so they can be properly ordered with other loads.
   671  		{name: "LDAR", argLength: 2, reg: gpload, asm: "LDAR", faultOnNilArg0: true},
   672  		{name: "LDARB", argLength: 2, reg: gpload, asm: "LDARB", faultOnNilArg0: true},
   673  		{name: "LDARW", argLength: 2, reg: gpload, asm: "LDARW", faultOnNilArg0: true},
   674  
   675  		// atomic stores.
   676  		// store arg1 to arg0. arg2=mem. returns memory. auxint must be zero.
   677  		{name: "STLRB", argLength: 3, reg: gpstore, asm: "STLRB", faultOnNilArg0: true, hasSideEffects: true},
   678  		{name: "STLR", argLength: 3, reg: gpstore, asm: "STLR", faultOnNilArg0: true, hasSideEffects: true},
   679  		{name: "STLRW", argLength: 3, reg: gpstore, asm: "STLRW", faultOnNilArg0: true, hasSideEffects: true},
   680  
   681  		// atomic exchange.
   682  		// store arg1 to arg0. arg2=mem. returns <old content of *arg0, memory>. auxint must be zero.
   683  		// LDAXR	(Rarg0), Rout
   684  		// STLXR	Rarg1, (Rarg0), Rtmp
   685  		// CBNZ		Rtmp, -2(PC)
   686  		{name: "LoweredAtomicExchange64", argLength: 3, reg: gpxchg, resultNotInArgs: true, faultOnNilArg0: true, hasSideEffects: true, unsafePoint: true},
   687  		{name: "LoweredAtomicExchange32", argLength: 3, reg: gpxchg, resultNotInArgs: true, faultOnNilArg0: true, hasSideEffects: true, unsafePoint: true},
   688  		{name: "LoweredAtomicExchange8", argLength: 3, reg: gpxchg, resultNotInArgs: true, faultOnNilArg0: true, hasSideEffects: true, unsafePoint: true},
   689  
   690  		// atomic exchange variant.
   691  		// store arg1 to arg0. arg2=mem. returns <old content of *arg0, memory>. auxint must be zero.
   692  		// SWPALD	Rarg1, (Rarg0), Rout
   693  		{name: "LoweredAtomicExchange64Variant", argLength: 3, reg: gpxchg, resultNotInArgs: true, faultOnNilArg0: true, hasSideEffects: true},
   694  		{name: "LoweredAtomicExchange32Variant", argLength: 3, reg: gpxchg, resultNotInArgs: true, faultOnNilArg0: true, hasSideEffects: true},
   695  		{name: "LoweredAtomicExchange8Variant", argLength: 3, reg: gpxchg, resultNotInArgs: true, faultOnNilArg0: true, hasSideEffects: true, unsafePoint: true},
   696  
   697  		// atomic add.
   698  		// *arg0 += arg1. arg2=mem. returns <new content of *arg0, memory>. auxint must be zero.
   699  		// LDAXR	(Rarg0), Rout
   700  		// ADD		Rarg1, Rout
   701  		// STLXR	Rout, (Rarg0), Rtmp
   702  		// CBNZ		Rtmp, -3(PC)
   703  		{name: "LoweredAtomicAdd64", argLength: 3, reg: gpxchg, resultNotInArgs: true, faultOnNilArg0: true, hasSideEffects: true, unsafePoint: true},
   704  		{name: "LoweredAtomicAdd32", argLength: 3, reg: gpxchg, resultNotInArgs: true, faultOnNilArg0: true, hasSideEffects: true, unsafePoint: true},
   705  
   706  		// atomic add variant.
   707  		// *arg0 += arg1. arg2=mem. returns <new content of *arg0, memory>. auxint must be zero.
   708  		// LDADDAL	(Rarg0), Rarg1, Rout
   709  		// ADD		Rarg1, Rout
   710  		{name: "LoweredAtomicAdd64Variant", argLength: 3, reg: gpxchg, resultNotInArgs: true, faultOnNilArg0: true, hasSideEffects: true},
   711  		{name: "LoweredAtomicAdd32Variant", argLength: 3, reg: gpxchg, resultNotInArgs: true, faultOnNilArg0: true, hasSideEffects: true},
   712  
   713  		// atomic compare and swap.
   714  		// arg0 = pointer, arg1 = old value, arg2 = new value, arg3 = memory. auxint must be zero.
   715  		// if *arg0 == arg1 {
   716  		//   *arg0 = arg2
   717  		//   return (true, memory)
   718  		// } else {
   719  		//   return (false, memory)
   720  		// }
   721  		// LDAXR	(Rarg0), Rtmp
   722  		// CMP		Rarg1, Rtmp
   723  		// BNE		3(PC)
   724  		// STLXR	Rarg2, (Rarg0), Rtmp
   725  		// CBNZ		Rtmp, -4(PC)
   726  		// CSET		EQ, Rout
   727  		{name: "LoweredAtomicCas64", argLength: 4, reg: gpcas, resultNotInArgs: true, clobberFlags: true, faultOnNilArg0: true, hasSideEffects: true, unsafePoint: true},
   728  		{name: "LoweredAtomicCas32", argLength: 4, reg: gpcas, resultNotInArgs: true, clobberFlags: true, faultOnNilArg0: true, hasSideEffects: true, unsafePoint: true},
   729  
   730  		// atomic compare and swap variant.
   731  		// arg0 = pointer, arg1 = old value, arg2 = new value, arg3 = memory. auxint must be zero.
   732  		// if *arg0 == arg1 {
   733  		//   *arg0 = arg2
   734  		//   return (true, memory)
   735  		// } else {
   736  		//   return (false, memory)
   737  		// }
   738  		// MOV  	Rarg1, Rtmp
   739  		// CASAL	Rtmp, (Rarg0), Rarg2
   740  		// CMP  	Rarg1, Rtmp
   741  		// CSET 	EQ, Rout
   742  		{name: "LoweredAtomicCas64Variant", argLength: 4, reg: gpcas, resultNotInArgs: true, clobberFlags: true, faultOnNilArg0: true, hasSideEffects: true, unsafePoint: true},
   743  		{name: "LoweredAtomicCas32Variant", argLength: 4, reg: gpcas, resultNotInArgs: true, clobberFlags: true, faultOnNilArg0: true, hasSideEffects: true, unsafePoint: true},
   744  
   745  		// atomic and/or.
   746  		// *arg0 &= (|=) arg1. arg2=mem. returns <old content of *arg0, memory>. auxint must be zero.
   747  		// LDAXR	(Rarg0), Rout
   748  		// AND/OR	Rarg1, Rout, tempReg
   749  		// STLXR	tempReg, (Rarg0), Rtmp
   750  		// CBNZ		Rtmp, -3(PC)
   751  		{name: "LoweredAtomicAnd8", argLength: 3, reg: gpxchg, resultNotInArgs: true, asm: "AND", faultOnNilArg0: true, hasSideEffects: true, unsafePoint: true, needIntTemp: true},
   752  		{name: "LoweredAtomicOr8", argLength: 3, reg: gpxchg, resultNotInArgs: true, asm: "ORR", faultOnNilArg0: true, hasSideEffects: true, unsafePoint: true, needIntTemp: true},
   753  		{name: "LoweredAtomicAnd64", argLength: 3, reg: gpxchg, resultNotInArgs: true, asm: "AND", faultOnNilArg0: true, hasSideEffects: true, unsafePoint: true, needIntTemp: true},
   754  		{name: "LoweredAtomicOr64", argLength: 3, reg: gpxchg, resultNotInArgs: true, asm: "ORR", faultOnNilArg0: true, hasSideEffects: true, unsafePoint: true, needIntTemp: true},
   755  		{name: "LoweredAtomicAnd32", argLength: 3, reg: gpxchg, resultNotInArgs: true, asm: "AND", faultOnNilArg0: true, hasSideEffects: true, unsafePoint: true, needIntTemp: true},
   756  		{name: "LoweredAtomicOr32", argLength: 3, reg: gpxchg, resultNotInArgs: true, asm: "ORR", faultOnNilArg0: true, hasSideEffects: true, unsafePoint: true, needIntTemp: true},
   757  
   758  		// atomic and/or variant.
   759  		// *arg0 &= (|=) arg1. arg2=mem. returns <old content of *arg0, memory>. auxint must be zero.
   760  		//   AND:
   761  		// MNV       Rarg1, Rtemp
   762  		// LDANDALB  Rtemp, (Rarg0), Rout
   763  		//   OR:
   764  		// LDORALB  Rarg1, (Rarg0), Rout
   765  		{name: "LoweredAtomicAnd8Variant", argLength: 3, reg: gpxchg, resultNotInArgs: true, faultOnNilArg0: true, hasSideEffects: true, unsafePoint: true},
   766  		{name: "LoweredAtomicOr8Variant", argLength: 3, reg: gpxchg, resultNotInArgs: true, faultOnNilArg0: true, hasSideEffects: true},
   767  		{name: "LoweredAtomicAnd64Variant", argLength: 3, reg: gpxchg, resultNotInArgs: true, faultOnNilArg0: true, hasSideEffects: true, unsafePoint: true},
   768  		{name: "LoweredAtomicOr64Variant", argLength: 3, reg: gpxchg, resultNotInArgs: true, faultOnNilArg0: true, hasSideEffects: true},
   769  		{name: "LoweredAtomicAnd32Variant", argLength: 3, reg: gpxchg, resultNotInArgs: true, faultOnNilArg0: true, hasSideEffects: true, unsafePoint: true},
   770  		{name: "LoweredAtomicOr32Variant", argLength: 3, reg: gpxchg, resultNotInArgs: true, faultOnNilArg0: true, hasSideEffects: true},
   771  
   772  		// LoweredWB invokes runtime.gcWriteBarrier. arg0=mem, auxint=# of buffer entries needed
   773  		// It saves all GP registers if necessary,
   774  		// but clobbers R30 (LR) because it's a call.
   775  		// R16 and R17 may be clobbered by linker trampoline.
   776  		// Returns a pointer to a write barrier buffer in R25.
   777  		{name: "LoweredWB", argLength: 1, reg: regInfo{clobbers: (callerSave &^ gpg) | buildReg("R16 R17 R30"), outputs: []regMask{buildReg("R25")}}, clobberFlags: true, aux: "Int64"},
   778  
   779  		// LoweredPanicBoundsRR takes x and y, two values that caused a bounds check to fail.
   780  		// the RC and CR versions are used when one of the arguments is a constant. CC is used
   781  		// when both are constant (normally both 0, as prove derives the fact that a [0] bounds
   782  		// failure means the length must have also been 0).
   783  		// AuxInt contains a report code (see PanicBounds in genericOps.go).
   784  		{name: "LoweredPanicBoundsRR", argLength: 3, aux: "Int64", reg: regInfo{inputs: []regMask{first16, first16}}, typ: "Mem", call: true}, // arg0=x, arg1=y, arg2=mem, returns memory.
   785  		{name: "LoweredPanicBoundsRC", argLength: 2, aux: "PanicBoundsC", reg: regInfo{inputs: []regMask{first16}}, typ: "Mem", call: true},   // arg0=x, arg1=mem, returns memory.
   786  		{name: "LoweredPanicBoundsCR", argLength: 2, aux: "PanicBoundsC", reg: regInfo{inputs: []regMask{first16}}, typ: "Mem", call: true},   // arg0=y, arg1=mem, returns memory.
   787  		{name: "LoweredPanicBoundsCC", argLength: 1, aux: "PanicBoundsCC", reg: regInfo{}, typ: "Mem", call: true},                            // arg0=mem, returns memory.
   788  
   789  		// Prefetch instruction
   790  		// Do prefetch arg0 address with option aux. arg0=addr, arg1=memory, aux=option.
   791  		{name: "PRFM", argLength: 2, aux: "Int64", reg: prefreg, asm: "PRFM", hasSideEffects: true},
   792  
   793  		// Publication barrier
   794  		{name: "DMB", argLength: 1, aux: "Int64", asm: "DMB", hasSideEffects: true}, // Do data barrier. arg0=memory, aux=option.
   795  		{name: "ZERO", zeroWidth: true, fixedReg: true},                             // reads-as-zero register
   796  	}
   797  
   798  	blocks := []blockData{
   799  		{name: "EQ", controls: 1},
   800  		{name: "NE", controls: 1},
   801  		{name: "LT", controls: 1},
   802  		{name: "LE", controls: 1},
   803  		{name: "GT", controls: 1},
   804  		{name: "GE", controls: 1},
   805  		{name: "ULT", controls: 1},
   806  		{name: "ULE", controls: 1},
   807  		{name: "UGT", controls: 1},
   808  		{name: "UGE", controls: 1},
   809  		{name: "Z", controls: 1},                  // Control == 0 (take a register instead of flags)
   810  		{name: "NZ", controls: 1},                 // Control != 0
   811  		{name: "ZW", controls: 1},                 // Control == 0, 32-bit
   812  		{name: "NZW", controls: 1},                // Control != 0, 32-bit
   813  		{name: "TBZ", controls: 1, aux: "Int64"},  // Control & (1 << AuxInt) == 0
   814  		{name: "TBNZ", controls: 1, aux: "Int64"}, // Control & (1 << AuxInt) != 0
   815  		{name: "FLT", controls: 1},
   816  		{name: "FLE", controls: 1},
   817  		{name: "FGT", controls: 1},
   818  		{name: "FGE", controls: 1},
   819  		{name: "LTnoov", controls: 1}, // 'LT' but without honoring overflow
   820  		{name: "LEnoov", controls: 1}, // 'LE' but without honoring overflow
   821  		{name: "GTnoov", controls: 1}, // 'GT' but without honoring overflow
   822  		{name: "GEnoov", controls: 1}, // 'GE' but without honoring overflow
   823  
   824  		// JUMPTABLE implements jump tables.
   825  		// Aux is the symbol (an *obj.LSym) for the jump table.
   826  		// control[0] is the index into the jump table.
   827  		// control[1] is the address of the jump table (the address of the symbol stored in Aux).
   828  		{name: "JUMPTABLE", controls: 2, aux: "Sym"},
   829  	}
   830  
   831  	archs = append(archs, arch{
   832  		name:               "ARM64",
   833  		pkg:                "cmd/internal/obj/arm64",
   834  		genfile:            "../../arm64/ssa.go",
   835  		ops:                ops,
   836  		blocks:             blocks,
   837  		regnames:           regNamesARM64,
   838  		ParamIntRegNames:   "R0 R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15",
   839  		ParamFloatRegNames: "F0 F1 F2 F3 F4 F5 F6 F7 F8 F9 F10 F11 F12 F13 F14 F15",
   840  		gpregmask:          gp,
   841  		fpregmask:          fp,
   842  		framepointerreg:    -1, // not used
   843  		linkreg:            int8(num["R30"]),
   844  	})
   845  }
   846  

View as plain text