github.com/gagliardetto/golang-go@v0.0.0-20201020153340-53909ea70814/cmd/compile/internal/ssa/gen/MIPS64.rules (about)

     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  (Add(Ptr|64|32|16|8) x y) -> (ADDV x y)
     6  (Add(32|64)F x y) -> (ADD(F|D) x y)
     7  
     8  (Sub(Ptr|64|32|16|8) x y) -> (SUBV x y)
     9  (Sub(32|64)F x y) -> (SUB(F|D) x y)
    10  
    11  (Mul(64|32|16|8) x y) -> (Select1 (MULVU x y))
    12  (Mul(32|64)F x y) -> (MUL(F|D) x y)
    13  (Mul64uhilo x y) -> (MULVU x y)
    14  
    15  (Hmul64 x y) -> (Select0 (MULV x y))
    16  (Hmul64u x y) -> (Select0 (MULVU x y))
    17  (Hmul32 x y) -> (SRAVconst (Select1 <typ.Int64> (MULV (SignExt32to64 x) (SignExt32to64 y))) [32])
    18  (Hmul32u x y) -> (SRLVconst (Select1 <typ.UInt64> (MULVU (ZeroExt32to64 x) (ZeroExt32to64 y))) [32])
    19  
    20  (Div64 x y) -> (Select1 (DIVV x y))
    21  (Div64u x y) -> (Select1 (DIVVU x y))
    22  (Div32 x y) -> (Select1 (DIVV (SignExt32to64 x) (SignExt32to64 y)))
    23  (Div32u x y) -> (Select1 (DIVVU (ZeroExt32to64 x) (ZeroExt32to64 y)))
    24  (Div16 x y) -> (Select1 (DIVV (SignExt16to64 x) (SignExt16to64 y)))
    25  (Div16u x y) -> (Select1 (DIVVU (ZeroExt16to64 x) (ZeroExt16to64 y)))
    26  (Div8 x y) -> (Select1 (DIVV (SignExt8to64 x) (SignExt8to64 y)))
    27  (Div8u x y) -> (Select1 (DIVVU (ZeroExt8to64 x) (ZeroExt8to64 y)))
    28  (Div(32|64)F x y) -> (DIV(F|D) x y)
    29  
    30  (Mod64 x y) -> (Select0 (DIVV x y))
    31  (Mod64u x y) -> (Select0 (DIVVU x y))
    32  (Mod32 x y) -> (Select0 (DIVV (SignExt32to64 x) (SignExt32to64 y)))
    33  (Mod32u x y) -> (Select0 (DIVVU (ZeroExt32to64 x) (ZeroExt32to64 y)))
    34  (Mod16 x y) -> (Select0 (DIVV (SignExt16to64 x) (SignExt16to64 y)))
    35  (Mod16u x y) -> (Select0 (DIVVU (ZeroExt16to64 x) (ZeroExt16to64 y)))
    36  (Mod8 x y) -> (Select0 (DIVV (SignExt8to64 x) (SignExt8to64 y)))
    37  (Mod8u x y) -> (Select0 (DIVVU (ZeroExt8to64 x) (ZeroExt8to64 y)))
    38  
    39  // (x + y) / 2 with x>=y -> (x - y) / 2 + y
    40  (Avg64u <t> x y) -> (ADDV (SRLVconst <t> (SUBV <t> x y) [1]) y)
    41  
    42  (And(64|32|16|8) x y) -> (AND x y)
    43  (Or(64|32|16|8) x y) -> (OR x y)
    44  (Xor(64|32|16|8) x y) -> (XOR x y)
    45  
    46  // shifts
    47  // hardware instruction uses only the low 6 bits of the shift
    48  // we compare to 64 to ensure Go semantics for large shifts
    49  (Lsh64x64 <t> x y) -> (AND (NEGV <t> (SGTU (MOVVconst <typ.UInt64> [64]) y)) (SLLV <t> x y))
    50  (Lsh64x32 <t> x y) -> (AND (NEGV <t> (SGTU (MOVVconst <typ.UInt64> [64]) (ZeroExt32to64 y))) (SLLV <t> x (ZeroExt32to64 y)))
    51  (Lsh64x16 <t> x y) -> (AND (NEGV <t> (SGTU (MOVVconst <typ.UInt64> [64]) (ZeroExt16to64 y))) (SLLV <t> x (ZeroExt16to64 y)))
    52  (Lsh64x8  <t> x y) -> (AND (NEGV <t> (SGTU (MOVVconst <typ.UInt64> [64]) (ZeroExt8to64  y))) (SLLV <t> x (ZeroExt8to64  y)))
    53  
    54  (Lsh32x64 <t> x y) -> (AND (NEGV <t> (SGTU (MOVVconst <typ.UInt64> [64]) y)) (SLLV <t> x y))
    55  (Lsh32x32 <t> x y) -> (AND (NEGV <t> (SGTU (MOVVconst <typ.UInt64> [64]) (ZeroExt32to64 y))) (SLLV <t> x (ZeroExt32to64 y)))
    56  (Lsh32x16 <t> x y) -> (AND (NEGV <t> (SGTU (MOVVconst <typ.UInt64> [64]) (ZeroExt16to64 y))) (SLLV <t> x (ZeroExt16to64 y)))
    57  (Lsh32x8  <t> x y) -> (AND (NEGV <t> (SGTU (MOVVconst <typ.UInt64> [64]) (ZeroExt8to64  y))) (SLLV <t> x (ZeroExt8to64  y)))
    58  
    59  (Lsh16x64 <t> x y) -> (AND (NEGV <t> (SGTU (MOVVconst <typ.UInt64> [64]) y)) (SLLV <t> x y))
    60  (Lsh16x32 <t> x y) -> (AND (NEGV <t> (SGTU (MOVVconst <typ.UInt64> [64]) (ZeroExt32to64 y))) (SLLV <t> x (ZeroExt32to64 y)))
    61  (Lsh16x16 <t> x y) -> (AND (NEGV <t> (SGTU (MOVVconst <typ.UInt64> [64]) (ZeroExt16to64 y))) (SLLV <t> x (ZeroExt16to64 y)))
    62  (Lsh16x8  <t> x y) -> (AND (NEGV <t> (SGTU (MOVVconst <typ.UInt64> [64]) (ZeroExt8to64  y))) (SLLV <t> x (ZeroExt8to64  y)))
    63  
    64  (Lsh8x64 <t> x y) -> (AND (NEGV <t> (SGTU (MOVVconst <typ.UInt64> [64]) y)) (SLLV <t> x y))
    65  (Lsh8x32 <t> x y) -> (AND (NEGV <t> (SGTU (MOVVconst <typ.UInt64> [64]) (ZeroExt32to64 y))) (SLLV <t> x (ZeroExt32to64 y)))
    66  (Lsh8x16 <t> x y) -> (AND (NEGV <t> (SGTU (MOVVconst <typ.UInt64> [64]) (ZeroExt16to64 y))) (SLLV <t> x (ZeroExt16to64 y)))
    67  (Lsh8x8  <t> x y) -> (AND (NEGV <t> (SGTU (MOVVconst <typ.UInt64> [64]) (ZeroExt8to64  y))) (SLLV <t> x (ZeroExt8to64  y)))
    68  
    69  (Rsh64Ux64 <t> x y) -> (AND (NEGV <t> (SGTU (MOVVconst <typ.UInt64> [64]) y)) (SRLV <t> x y))
    70  (Rsh64Ux32 <t> x y) -> (AND (NEGV <t> (SGTU (MOVVconst <typ.UInt64> [64]) (ZeroExt32to64 y))) (SRLV <t> x (ZeroExt32to64 y)))
    71  (Rsh64Ux16 <t> x y) -> (AND (NEGV <t> (SGTU (MOVVconst <typ.UInt64> [64]) (ZeroExt16to64 y))) (SRLV <t> x (ZeroExt16to64 y)))
    72  (Rsh64Ux8  <t> x y) -> (AND (NEGV <t> (SGTU (MOVVconst <typ.UInt64> [64]) (ZeroExt8to64  y))) (SRLV <t> x (ZeroExt8to64  y)))
    73  
    74  (Rsh32Ux64 <t> x y) -> (AND (NEGV <t> (SGTU (MOVVconst <typ.UInt64> [64]) y)) (SRLV <t> (ZeroExt32to64 x) y))
    75  (Rsh32Ux32 <t> x y) -> (AND (NEGV <t> (SGTU (MOVVconst <typ.UInt64> [64]) (ZeroExt32to64 y))) (SRLV <t> (ZeroExt32to64 x) (ZeroExt32to64 y)))
    76  (Rsh32Ux16 <t> x y) -> (AND (NEGV <t> (SGTU (MOVVconst <typ.UInt64> [64]) (ZeroExt16to64 y))) (SRLV <t> (ZeroExt32to64 x) (ZeroExt16to64 y)))
    77  (Rsh32Ux8  <t> x y) -> (AND (NEGV <t> (SGTU (MOVVconst <typ.UInt64> [64]) (ZeroExt8to64  y))) (SRLV <t> (ZeroExt32to64 x) (ZeroExt8to64  y)))
    78  
    79  (Rsh16Ux64 <t> x y) -> (AND (NEGV <t> (SGTU (MOVVconst <typ.UInt64> [64]) y)) (SRLV <t> (ZeroExt16to64 x) y))
    80  (Rsh16Ux32 <t> x y) -> (AND (NEGV <t> (SGTU (MOVVconst <typ.UInt64> [64]) (ZeroExt32to64 y))) (SRLV <t> (ZeroExt16to64 x) (ZeroExt32to64 y)))
    81  (Rsh16Ux16 <t> x y) -> (AND (NEGV <t> (SGTU (MOVVconst <typ.UInt64> [64]) (ZeroExt16to64 y))) (SRLV <t> (ZeroExt16to64 x) (ZeroExt16to64 y)))
    82  (Rsh16Ux8  <t> x y) -> (AND (NEGV <t> (SGTU (MOVVconst <typ.UInt64> [64]) (ZeroExt8to64  y))) (SRLV <t> (ZeroExt16to64 x) (ZeroExt8to64  y)))
    83  
    84  (Rsh8Ux64 <t> x y) -> (AND (NEGV <t> (SGTU (MOVVconst <typ.UInt64> [64]) y)) (SRLV <t> (ZeroExt8to64 x) y))
    85  (Rsh8Ux32 <t> x y) -> (AND (NEGV <t> (SGTU (MOVVconst <typ.UInt64> [64]) (ZeroExt32to64 y))) (SRLV <t> (ZeroExt8to64 x) (ZeroExt32to64 y)))
    86  (Rsh8Ux16 <t> x y) -> (AND (NEGV <t> (SGTU (MOVVconst <typ.UInt64> [64]) (ZeroExt16to64 y))) (SRLV <t> (ZeroExt8to64 x) (ZeroExt16to64 y)))
    87  (Rsh8Ux8  <t> x y) -> (AND (NEGV <t> (SGTU (MOVVconst <typ.UInt64> [64]) (ZeroExt8to64  y))) (SRLV <t> (ZeroExt8to64 x) (ZeroExt8to64  y)))
    88  
    89  (Rsh64x64 <t> x y) -> (SRAV x (OR <t> (NEGV <t> (SGTU y (MOVVconst <typ.UInt64> [63]))) y))
    90  (Rsh64x32 <t> x y) -> (SRAV x (OR <t> (NEGV <t> (SGTU (ZeroExt32to64 y) (MOVVconst <typ.UInt64> [63]))) (ZeroExt32to64 y)))
    91  (Rsh64x16 <t> x y) -> (SRAV x (OR <t> (NEGV <t> (SGTU (ZeroExt16to64 y) (MOVVconst <typ.UInt64> [63]))) (ZeroExt16to64 y)))
    92  (Rsh64x8  <t> x y) -> (SRAV x (OR <t> (NEGV <t> (SGTU (ZeroExt8to64  y) (MOVVconst <typ.UInt64> [63]))) (ZeroExt8to64  y)))
    93  
    94  (Rsh32x64 <t> x y) -> (SRAV (SignExt32to64 x) (OR <t> (NEGV <t> (SGTU y (MOVVconst <typ.UInt64> [63]))) y))
    95  (Rsh32x32 <t> x y) -> (SRAV (SignExt32to64 x) (OR <t> (NEGV <t> (SGTU (ZeroExt32to64 y) (MOVVconst <typ.UInt64> [63]))) (ZeroExt32to64 y)))
    96  (Rsh32x16 <t> x y) -> (SRAV (SignExt32to64 x) (OR <t> (NEGV <t> (SGTU (ZeroExt16to64 y) (MOVVconst <typ.UInt64> [63]))) (ZeroExt16to64 y)))
    97  (Rsh32x8  <t> x y) -> (SRAV (SignExt32to64 x) (OR <t> (NEGV <t> (SGTU (ZeroExt8to64  y) (MOVVconst <typ.UInt64> [63]))) (ZeroExt8to64  y)))
    98  
    99  (Rsh16x64 <t> x y) -> (SRAV (SignExt16to64 x) (OR <t> (NEGV <t> (SGTU y (MOVVconst <typ.UInt64> [63]))) y))
   100  (Rsh16x32 <t> x y) -> (SRAV (SignExt16to64 x) (OR <t> (NEGV <t> (SGTU (ZeroExt32to64 y) (MOVVconst <typ.UInt64> [63]))) (ZeroExt32to64 y)))
   101  (Rsh16x16 <t> x y) -> (SRAV (SignExt16to64 x) (OR <t> (NEGV <t> (SGTU (ZeroExt16to64 y) (MOVVconst <typ.UInt64> [63]))) (ZeroExt16to64 y)))
   102  (Rsh16x8  <t> x y) -> (SRAV (SignExt16to64 x) (OR <t> (NEGV <t> (SGTU (ZeroExt8to64  y) (MOVVconst <typ.UInt64> [63]))) (ZeroExt8to64  y)))
   103  
   104  (Rsh8x64 <t> x y) -> (SRAV (SignExt8to64 x) (OR <t> (NEGV <t> (SGTU y (MOVVconst <typ.UInt64> [63]))) y))
   105  (Rsh8x32 <t> x y) -> (SRAV (SignExt8to64 x) (OR <t> (NEGV <t> (SGTU (ZeroExt32to64 y) (MOVVconst <typ.UInt64> [63]))) (ZeroExt32to64 y)))
   106  (Rsh8x16 <t> x y) -> (SRAV (SignExt8to64 x) (OR <t> (NEGV <t> (SGTU (ZeroExt16to64 y) (MOVVconst <typ.UInt64> [63]))) (ZeroExt16to64 y)))
   107  (Rsh8x8  <t> x y) -> (SRAV (SignExt8to64 x) (OR <t> (NEGV <t> (SGTU (ZeroExt8to64  y) (MOVVconst <typ.UInt64> [63]))) (ZeroExt8to64  y)))
   108  
   109  // rotates
   110  (RotateLeft8 <t> x (MOVVconst [c])) -> (Or8 (Lsh8x64 <t> x (MOVVconst [c&7])) (Rsh8Ux64 <t> x (MOVVconst [-c&7])))
   111  (RotateLeft16 <t> x (MOVVconst [c])) -> (Or16 (Lsh16x64 <t> x (MOVVconst [c&15])) (Rsh16Ux64 <t> x (MOVVconst [-c&15])))
   112  (RotateLeft32 <t> x (MOVVconst [c])) -> (Or32 (Lsh32x64 <t> x (MOVVconst [c&31])) (Rsh32Ux64 <t> x (MOVVconst [-c&31])))
   113  (RotateLeft64 <t> x (MOVVconst [c])) -> (Or64 (Lsh64x64 <t> x (MOVVconst [c&63])) (Rsh64Ux64 <t> x (MOVVconst [-c&63])))
   114  
   115  // unary ops
   116  (Neg(64|32|16|8) x) -> (NEGV x)
   117  (Neg(32|64)F x) -> (NEG(F|D) x)
   118  
   119  (Com(64|32|16|8) x) -> (NOR (MOVVconst [0]) x)
   120  
   121  (Sqrt x) -> (SQRTD x)
   122  
   123  // boolean ops -- booleans are represented with 0=false, 1=true
   124  (AndB x y) -> (AND x y)
   125  (OrB x y) -> (OR x y)
   126  (EqB x y) -> (XOR (MOVVconst [1]) (XOR <typ.Bool> x y))
   127  (NeqB x y) -> (XOR x y)
   128  (Not x) -> (XORconst [1] x)
   129  
   130  // constants
   131  (Const(64|32|16|8) [val]) -> (MOVVconst [val])
   132  (Const(32|64)F [val]) -> (MOV(F|D)const [val])
   133  (ConstNil) -> (MOVVconst [0])
   134  (ConstBool [b]) -> (MOVVconst [b])
   135  
   136  (Slicemask <t> x) -> (SRAVconst (NEGV <t> x) [63])
   137  
   138  // truncations
   139  // Because we ignore high parts of registers, truncates are just copies.
   140  (Trunc16to8 x) -> x
   141  (Trunc32to8 x) -> x
   142  (Trunc32to16 x) -> x
   143  (Trunc64to8 x) -> x
   144  (Trunc64to16 x) -> x
   145  (Trunc64to32 x) -> x
   146  
   147  // Zero-/Sign-extensions
   148  (ZeroExt8to16 x) -> (MOVBUreg x)
   149  (ZeroExt8to32 x) -> (MOVBUreg x)
   150  (ZeroExt16to32 x) -> (MOVHUreg x)
   151  (ZeroExt8to64 x) -> (MOVBUreg x)
   152  (ZeroExt16to64 x) -> (MOVHUreg x)
   153  (ZeroExt32to64 x) -> (MOVWUreg x)
   154  
   155  (SignExt8to16 x) -> (MOVBreg x)
   156  (SignExt8to32 x) -> (MOVBreg x)
   157  (SignExt16to32 x) -> (MOVHreg x)
   158  (SignExt8to64 x) -> (MOVBreg x)
   159  (SignExt16to64 x) -> (MOVHreg x)
   160  (SignExt32to64 x) -> (MOVWreg x)
   161  
   162  // float <-> int conversion
   163  (Cvt32to32F x) -> (MOVWF x)
   164  (Cvt32to64F x) -> (MOVWD x)
   165  (Cvt64to32F x) -> (MOVVF x)
   166  (Cvt64to64F x) -> (MOVVD x)
   167  (Cvt32Fto32 x) -> (TRUNCFW x)
   168  (Cvt64Fto32 x) -> (TRUNCDW x)
   169  (Cvt32Fto64 x) -> (TRUNCFV x)
   170  (Cvt64Fto64 x) -> (TRUNCDV x)
   171  (Cvt32Fto64F x) -> (MOVFD x)
   172  (Cvt64Fto32F x) -> (MOVDF x)
   173  
   174  (Round(32|64)F x) -> x
   175  
   176  // comparisons
   177  (Eq8 x y)  -> (SGTU (MOVVconst [1]) (XOR (ZeroExt8to64 x) (ZeroExt8to64 y)))
   178  (Eq16 x y) -> (SGTU (MOVVconst [1]) (XOR (ZeroExt16to64 x) (ZeroExt16to64 y)))
   179  (Eq32 x y) -> (SGTU (MOVVconst [1]) (XOR (ZeroExt32to64 x) (ZeroExt32to64 y)))
   180  (Eq64 x y) -> (SGTU (MOVVconst [1]) (XOR x y))
   181  (EqPtr x y) -> (SGTU (MOVVconst [1]) (XOR x y))
   182  (Eq(32|64)F x y) -> (FPFlagTrue (CMPEQ(F|D) x y))
   183  
   184  (Neq8 x y)  -> (SGTU (XOR (ZeroExt8to64 x) (ZeroExt8to64 y)) (MOVVconst [0]))
   185  (Neq16 x y) -> (SGTU (XOR (ZeroExt16to32 x) (ZeroExt16to64 y)) (MOVVconst [0]))
   186  (Neq32 x y) -> (SGTU (XOR (ZeroExt32to64 x) (ZeroExt32to64 y)) (MOVVconst [0]))
   187  (Neq64 x y) -> (SGTU (XOR x y) (MOVVconst [0]))
   188  (NeqPtr x y) -> (SGTU (XOR x y) (MOVVconst [0]))
   189  (Neq(32|64)F x y) -> (FPFlagFalse (CMPEQ(F|D) x y))
   190  
   191  (Less8 x y)  -> (SGT (SignExt8to64 y) (SignExt8to64 x))
   192  (Less16 x y) -> (SGT (SignExt16to64 y) (SignExt16to64 x))
   193  (Less32 x y) -> (SGT (SignExt32to64 y) (SignExt32to64 x))
   194  (Less64 x y) -> (SGT y x)
   195  (Less(32|64)F x y) -> (FPFlagTrue (CMPGT(F|D) y x)) // reverse operands to work around NaN
   196  
   197  (Less8U x y)  -> (SGTU (ZeroExt8to64 y) (ZeroExt8to64 x))
   198  (Less16U x y) -> (SGTU (ZeroExt16to64 y) (ZeroExt16to64 x))
   199  (Less32U x y) -> (SGTU (ZeroExt32to64 y) (ZeroExt32to64 x))
   200  (Less64U x y) -> (SGTU y x)
   201  
   202  (Leq8 x y)  -> (XOR (MOVVconst [1]) (SGT (SignExt8to64 x) (SignExt8to64 y)))
   203  (Leq16 x y) -> (XOR (MOVVconst [1]) (SGT (SignExt16to64 x) (SignExt16to64 y)))
   204  (Leq32 x y) -> (XOR (MOVVconst [1]) (SGT (SignExt32to64 x) (SignExt32to64 y)))
   205  (Leq64 x y) -> (XOR (MOVVconst [1]) (SGT x y))
   206  (Leq(32|64)F x y) -> (FPFlagTrue (CMPGE(F|D) y x)) // reverse operands to work around NaN
   207  
   208  (Leq8U x y)  -> (XOR (MOVVconst [1]) (SGTU (ZeroExt8to64 x) (ZeroExt8to64 y)))
   209  (Leq16U x y) -> (XOR (MOVVconst [1]) (SGTU (ZeroExt16to64 x) (ZeroExt16to64 y)))
   210  (Leq32U x y) -> (XOR (MOVVconst [1]) (SGTU (ZeroExt32to64 x) (ZeroExt32to64 y)))
   211  (Leq64U x y) -> (XOR (MOVVconst [1]) (SGTU x y))
   212  
   213  (Greater8 x y)  -> (SGT (SignExt8to64 x) (SignExt8to64 y))
   214  (Greater16 x y) -> (SGT (SignExt16to64 x) (SignExt16to64 y))
   215  (Greater32 x y) -> (SGT (SignExt32to64 x) (SignExt32to64 y))
   216  (Greater64 x y) -> (SGT x y)
   217  (Greater(32|64)F x y) -> (FPFlagTrue (CMPGT(F|D) x y))
   218  
   219  (Greater8U x y)  -> (SGTU (ZeroExt8to64 x) (ZeroExt8to64 y))
   220  (Greater16U x y) -> (SGTU (ZeroExt16to64 x) (ZeroExt16to64 y))
   221  (Greater32U x y) -> (SGTU (ZeroExt32to64 x) (ZeroExt32to64 y))
   222  (Greater64U x y) -> (SGTU x y)
   223  
   224  (Geq8 x y)  -> (XOR (MOVVconst [1]) (SGT (SignExt8to64 y) (SignExt8to64 x)))
   225  (Geq16 x y) -> (XOR (MOVVconst [1]) (SGT (SignExt16to64 y) (SignExt16to64 x)))
   226  (Geq32 x y) -> (XOR (MOVVconst [1]) (SGT (SignExt32to64 y) (SignExt32to64 x)))
   227  (Geq64 x y) -> (XOR (MOVVconst [1]) (SGT y x))
   228  (Geq(32|64)F x y) -> (FPFlagTrue (CMPGE(F|D) x y))
   229  
   230  (Geq8U x y)  -> (XOR (MOVVconst [1]) (SGTU (ZeroExt8to64 y) (ZeroExt8to64 x)))
   231  (Geq16U x y) -> (XOR (MOVVconst [1]) (SGTU (ZeroExt16to64 y) (ZeroExt16to64 x)))
   232  (Geq32U x y) -> (XOR (MOVVconst [1]) (SGTU (ZeroExt32to64 y) (ZeroExt32to64 x)))
   233  (Geq64U x y) -> (XOR (MOVVconst [1]) (SGTU y x))
   234  
   235  (OffPtr [off] ptr:(SP)) -> (MOVVaddr [off] ptr)
   236  (OffPtr [off] ptr) -> (ADDVconst [off] ptr)
   237  
   238  (Addr {sym} base) -> (MOVVaddr {sym} base)
   239  (LocalAddr {sym} base _) -> (MOVVaddr {sym} base)
   240  
   241  // loads
   242  (Load <t> ptr mem) && t.IsBoolean() -> (MOVBUload ptr mem)
   243  (Load <t> ptr mem) && (is8BitInt(t) && isSigned(t)) -> (MOVBload ptr mem)
   244  (Load <t> ptr mem) && (is8BitInt(t) && !isSigned(t)) -> (MOVBUload ptr mem)
   245  (Load <t> ptr mem) && (is16BitInt(t) && isSigned(t)) -> (MOVHload ptr mem)
   246  (Load <t> ptr mem) && (is16BitInt(t) && !isSigned(t)) -> (MOVHUload ptr mem)
   247  (Load <t> ptr mem) && (is32BitInt(t) && isSigned(t)) -> (MOVWload ptr mem)
   248  (Load <t> ptr mem) && (is32BitInt(t) && !isSigned(t)) -> (MOVWUload ptr mem)
   249  (Load <t> ptr mem) && (is64BitInt(t) || isPtr(t)) -> (MOVVload ptr mem)
   250  (Load <t> ptr mem) && is32BitFloat(t) -> (MOVFload ptr mem)
   251  (Load <t> ptr mem) && is64BitFloat(t) -> (MOVDload ptr mem)
   252  
   253  // stores
   254  (Store {t} ptr val mem) && t.(*types.Type).Size() == 1 -> (MOVBstore ptr val mem)
   255  (Store {t} ptr val mem) && t.(*types.Type).Size() == 2 -> (MOVHstore ptr val mem)
   256  (Store {t} ptr val mem) && t.(*types.Type).Size() == 4 && !is32BitFloat(val.Type) -> (MOVWstore ptr val mem)
   257  (Store {t} ptr val mem) && t.(*types.Type).Size() == 8 && !is64BitFloat(val.Type) -> (MOVVstore ptr val mem)
   258  (Store {t} ptr val mem) && t.(*types.Type).Size() == 4 && is32BitFloat(val.Type) -> (MOVFstore ptr val mem)
   259  (Store {t} ptr val mem) && t.(*types.Type).Size() == 8 && is64BitFloat(val.Type) -> (MOVDstore ptr val mem)
   260  
   261  // zeroing
   262  (Zero [0] _ mem) -> mem
   263  (Zero [1] ptr mem) -> (MOVBstore ptr (MOVVconst [0]) mem)
   264  (Zero [2] {t} ptr mem) && t.(*types.Type).Alignment()%2 == 0 ->
   265  	(MOVHstore ptr (MOVVconst [0]) mem)
   266  (Zero [2] ptr mem) ->
   267  	(MOVBstore [1] ptr (MOVVconst [0])
   268  		(MOVBstore [0] ptr (MOVVconst [0]) mem))
   269  (Zero [4] {t} ptr mem) && t.(*types.Type).Alignment()%4 == 0 ->
   270  	(MOVWstore ptr (MOVVconst [0]) mem)
   271  (Zero [4] {t} ptr mem) && t.(*types.Type).Alignment()%2 == 0 ->
   272  	(MOVHstore [2] ptr (MOVVconst [0])
   273  		(MOVHstore [0] ptr (MOVVconst [0]) mem))
   274  (Zero [4] ptr mem) ->
   275  	(MOVBstore [3] ptr (MOVVconst [0])
   276  		(MOVBstore [2] ptr (MOVVconst [0])
   277  			(MOVBstore [1] ptr (MOVVconst [0])
   278  				(MOVBstore [0] ptr (MOVVconst [0]) mem))))
   279  (Zero [8] {t} ptr mem) && t.(*types.Type).Alignment()%8 == 0 ->
   280  	(MOVVstore ptr (MOVVconst [0]) mem)
   281  (Zero [8] {t} ptr mem) && t.(*types.Type).Alignment()%4 == 0 ->
   282  	(MOVWstore [4] ptr (MOVVconst [0])
   283  		(MOVWstore [0] ptr (MOVVconst [0]) mem))
   284  (Zero [8] {t} ptr mem) && t.(*types.Type).Alignment()%2 == 0 ->
   285  	(MOVHstore [6] ptr (MOVVconst [0])
   286  		(MOVHstore [4] ptr (MOVVconst [0])
   287  			(MOVHstore [2] ptr (MOVVconst [0])
   288  				(MOVHstore [0] ptr (MOVVconst [0]) mem))))
   289  
   290  (Zero [3] ptr mem) ->
   291  	(MOVBstore [2] ptr (MOVVconst [0])
   292  		(MOVBstore [1] ptr (MOVVconst [0])
   293  			(MOVBstore [0] ptr (MOVVconst [0]) mem)))
   294  (Zero [6] {t} ptr mem) && t.(*types.Type).Alignment()%2 == 0 ->
   295  	(MOVHstore [4] ptr (MOVVconst [0])
   296  		(MOVHstore [2] ptr (MOVVconst [0])
   297  			(MOVHstore [0] ptr (MOVVconst [0]) mem)))
   298  (Zero [12] {t} ptr mem) && t.(*types.Type).Alignment()%4 == 0 ->
   299  	(MOVWstore [8] ptr (MOVVconst [0])
   300  		(MOVWstore [4] ptr (MOVVconst [0])
   301  			(MOVWstore [0] ptr (MOVVconst [0]) mem)))
   302  (Zero [16] {t} ptr mem) && t.(*types.Type).Alignment()%8 == 0 ->
   303  	(MOVVstore [8] ptr (MOVVconst [0])
   304  		(MOVVstore [0] ptr (MOVVconst [0]) mem))
   305  (Zero [24] {t} ptr mem) && t.(*types.Type).Alignment()%8 == 0 ->
   306  	(MOVVstore [16] ptr (MOVVconst [0])
   307  		(MOVVstore [8] ptr (MOVVconst [0])
   308  			(MOVVstore [0] ptr (MOVVconst [0]) mem)))
   309  
   310  // medium zeroing uses a duff device
   311  // 8, and 128 are magic constants, see runtime/mkduff.go
   312  (Zero [s] {t} ptr mem)
   313  	&& s%8 == 0 && s > 24 && s <= 8*128
   314  	&& t.(*types.Type).Alignment()%8 == 0 && !config.noDuffDevice ->
   315  	(DUFFZERO [8 * (128 - s/8)] ptr mem)
   316  
   317  // large or unaligned zeroing uses a loop
   318  (Zero [s] {t} ptr mem)
   319  	&& (s > 8*128 || config.noDuffDevice) || t.(*types.Type).Alignment()%8 != 0 ->
   320  	(LoweredZero [t.(*types.Type).Alignment()]
   321  		ptr
   322  		(ADDVconst <ptr.Type> ptr [s-moveSize(t.(*types.Type).Alignment(), config)])
   323  		mem)
   324  
   325  // moves
   326  (Move [0] _ _ mem) -> mem
   327  (Move [1] dst src mem) -> (MOVBstore dst (MOVBload src mem) mem)
   328  (Move [2] {t} dst src mem) && t.(*types.Type).Alignment()%2 == 0 ->
   329  	(MOVHstore dst (MOVHload src mem) mem)
   330  (Move [2] dst src mem) ->
   331  	(MOVBstore [1] dst (MOVBload [1] src mem)
   332  		(MOVBstore dst (MOVBload src mem) mem))
   333  (Move [4] {t} dst src mem) && t.(*types.Type).Alignment()%4 == 0 ->
   334  	(MOVWstore dst (MOVWload src mem) mem)
   335  (Move [4] {t} dst src mem) && t.(*types.Type).Alignment()%2 == 0 ->
   336  	(MOVHstore [2] dst (MOVHload [2] src mem)
   337  		(MOVHstore dst (MOVHload src mem) mem))
   338  (Move [4] dst src mem) ->
   339  	(MOVBstore [3] dst (MOVBload [3] src mem)
   340  		(MOVBstore [2] dst (MOVBload [2] src mem)
   341  			(MOVBstore [1] dst (MOVBload [1] src mem)
   342  				(MOVBstore dst (MOVBload src mem) mem))))
   343  (Move [8] {t} dst src mem) && t.(*types.Type).Alignment()%8 == 0 ->
   344  	(MOVVstore dst (MOVVload src mem) mem)
   345  (Move [8] {t} dst src mem) && t.(*types.Type).Alignment()%4 == 0 ->
   346  	(MOVWstore [4] dst (MOVWload [4] src mem)
   347  		(MOVWstore dst (MOVWload src mem) mem))
   348  (Move [8] {t} dst src mem) && t.(*types.Type).Alignment()%2 == 0 ->
   349  	(MOVHstore [6] dst (MOVHload [6] src mem)
   350  		(MOVHstore [4] dst (MOVHload [4] src mem)
   351  			(MOVHstore [2] dst (MOVHload [2] src mem)
   352  				(MOVHstore dst (MOVHload src mem) mem))))
   353  
   354  (Move [3] dst src mem) ->
   355  	(MOVBstore [2] dst (MOVBload [2] src mem)
   356  		(MOVBstore [1] dst (MOVBload [1] src mem)
   357  			(MOVBstore dst (MOVBload src mem) mem)))
   358  (Move [6] {t} dst src mem) && t.(*types.Type).Alignment()%2 == 0 ->
   359  	(MOVHstore [4] dst (MOVHload [4] src mem)
   360  		(MOVHstore [2] dst (MOVHload [2] src mem)
   361  			(MOVHstore dst (MOVHload src mem) mem)))
   362  (Move [12] {t} dst src mem) && t.(*types.Type).Alignment()%4 == 0 ->
   363  	(MOVWstore [8] dst (MOVWload [8] src mem)
   364  		(MOVWstore [4] dst (MOVWload [4] src mem)
   365  			(MOVWstore dst (MOVWload src mem) mem)))
   366  (Move [16] {t} dst src mem) && t.(*types.Type).Alignment()%8 == 0 ->
   367  	(MOVVstore [8] dst (MOVVload [8] src mem)
   368  		(MOVVstore dst (MOVVload src mem) mem))
   369  (Move [24] {t} dst src mem) && t.(*types.Type).Alignment()%8 == 0 ->
   370  	(MOVVstore [16] dst (MOVVload [16] src mem)
   371  		(MOVVstore [8] dst (MOVVload [8] src mem)
   372  			(MOVVstore dst (MOVVload src mem) mem)))
   373  
   374  // medium move uses a duff device
   375  (Move [s] {t} dst src mem)
   376  	&& s%8 == 0 && s >= 24 && s <= 8*128 && t.(*types.Type).Alignment()%8 == 0
   377  	&& !config.noDuffDevice ->
   378  	(DUFFCOPY [16 * (128 - s/8)] dst src mem)
   379  // 16 and 128 are magic constants.  16 is the number of bytes to encode:
   380  //	MOVV	(R1), R23
   381  //	ADDV	$8, R1
   382  //	MOVV	R23, (R2)
   383  //	ADDV	$8, R2
   384  // and 128 is the number of such blocks. See runtime/duff_mips64.s:duffcopy.
   385  
   386  // large or unaligned move uses a loop
   387  (Move [s] {t} dst src mem)
   388  	&& s > 24 || t.(*types.Type).Alignment()%8 != 0 ->
   389  	(LoweredMove [t.(*types.Type).Alignment()]
   390  		dst
   391  		src
   392  		(ADDVconst <src.Type> src [s-moveSize(t.(*types.Type).Alignment(), config)])
   393  		mem)
   394  
   395  // calls
   396  (StaticCall [argwid] {target} mem) -> (CALLstatic [argwid] {target} mem)
   397  (ClosureCall [argwid] entry closure mem) -> (CALLclosure [argwid] entry closure mem)
   398  (InterCall [argwid] entry mem) -> (CALLinter [argwid] entry mem)
   399  
   400  // atomic intrinsics
   401  (AtomicLoad8   ptr mem) -> (LoweredAtomicLoad8  ptr mem)
   402  (AtomicLoad32  ptr mem) -> (LoweredAtomicLoad32 ptr mem)
   403  (AtomicLoad64  ptr mem) -> (LoweredAtomicLoad64 ptr mem)
   404  (AtomicLoadPtr ptr mem) -> (LoweredAtomicLoad64 ptr mem)
   405  
   406  (AtomicStore8       ptr val mem) -> (LoweredAtomicStore8  ptr val mem)
   407  (AtomicStore32      ptr val mem) -> (LoweredAtomicStore32 ptr val mem)
   408  (AtomicStore64      ptr val mem) -> (LoweredAtomicStore64 ptr val mem)
   409  (AtomicStorePtrNoWB ptr val mem) -> (LoweredAtomicStore64 ptr val mem)
   410  
   411  (AtomicExchange32 ptr val mem) -> (LoweredAtomicExchange32 ptr val mem)
   412  (AtomicExchange64 ptr val mem) -> (LoweredAtomicExchange64 ptr val mem)
   413  
   414  (AtomicAdd32 ptr val mem) -> (LoweredAtomicAdd32 ptr val mem)
   415  (AtomicAdd64 ptr val mem) -> (LoweredAtomicAdd64 ptr val mem)
   416  
   417  (AtomicCompareAndSwap32 ptr old new_ mem) -> (LoweredAtomicCas32 ptr old new_ mem)
   418  (AtomicCompareAndSwap64 ptr old new_ mem) -> (LoweredAtomicCas64 ptr old new_ mem)
   419  
   420  // checks
   421  (NilCheck ptr mem) -> (LoweredNilCheck ptr mem)
   422  (IsNonNil ptr) -> (SGTU ptr (MOVVconst [0]))
   423  (IsInBounds idx len) -> (SGTU len idx)
   424  (IsSliceInBounds idx len) -> (XOR (MOVVconst [1]) (SGTU idx len))
   425  
   426  // pseudo-ops
   427  (GetClosurePtr) -> (LoweredGetClosurePtr)
   428  (GetCallerSP) -> (LoweredGetCallerSP)
   429  (GetCallerPC) -> (LoweredGetCallerPC)
   430  
   431  (If cond yes no) -> (NE cond yes no)
   432  
   433  // Write barrier.
   434  (WB {fn} destptr srcptr mem) -> (LoweredWB {fn} destptr srcptr mem)
   435  
   436  (PanicBounds [kind] x y mem) && boundsABI(kind) == 0 -> (LoweredPanicBoundsA [kind] x y mem)
   437  (PanicBounds [kind] x y mem) && boundsABI(kind) == 1 -> (LoweredPanicBoundsB [kind] x y mem)
   438  (PanicBounds [kind] x y mem) && boundsABI(kind) == 2 -> (LoweredPanicBoundsC [kind] x y mem)
   439  
   440  // Optimizations
   441  
   442  // Absorb boolean tests into block
   443  (NE (FPFlagTrue cmp) yes no) -> (FPT cmp yes no)
   444  (NE (FPFlagFalse cmp) yes no) -> (FPF cmp yes no)
   445  (EQ (FPFlagTrue cmp) yes no) -> (FPF cmp yes no)
   446  (EQ (FPFlagFalse cmp) yes no) -> (FPT cmp yes no)
   447  (NE (XORconst [1] cmp:(SGT _ _)) yes no) -> (EQ cmp yes no)
   448  (NE (XORconst [1] cmp:(SGTU _ _)) yes no) -> (EQ cmp yes no)
   449  (NE (XORconst [1] cmp:(SGTconst _)) yes no) -> (EQ cmp yes no)
   450  (NE (XORconst [1] cmp:(SGTUconst _)) yes no) -> (EQ cmp yes no)
   451  (EQ (XORconst [1] cmp:(SGT _ _)) yes no) -> (NE cmp yes no)
   452  (EQ (XORconst [1] cmp:(SGTU _ _)) yes no) -> (NE cmp yes no)
   453  (EQ (XORconst [1] cmp:(SGTconst _)) yes no) -> (NE cmp yes no)
   454  (EQ (XORconst [1] cmp:(SGTUconst _)) yes no) -> (NE cmp yes no)
   455  (NE (SGTUconst [1] x) yes no) -> (EQ x yes no)
   456  (EQ (SGTUconst [1] x) yes no) -> (NE x yes no)
   457  (NE (SGTU x (MOVVconst [0])) yes no) -> (NE x yes no)
   458  (EQ (SGTU x (MOVVconst [0])) yes no) -> (EQ x yes no)
   459  (NE (SGTconst [0] x) yes no) -> (LTZ x yes no)
   460  (EQ (SGTconst [0] x) yes no) -> (GEZ x yes no)
   461  (NE (SGT x (MOVVconst [0])) yes no) -> (GTZ x yes no)
   462  (EQ (SGT x (MOVVconst [0])) yes no) -> (LEZ x yes no)
   463  
   464  // fold offset into address
   465  (ADDVconst [off1] (MOVVaddr [off2] {sym} ptr)) -> (MOVVaddr [off1+off2] {sym} ptr)
   466  
   467  // fold address into load/store
   468  (MOVBload  [off1] {sym} (ADDVconst [off2] ptr) mem) && is32Bit(off1+off2) -> (MOVBload  [off1+off2] {sym} ptr mem)
   469  (MOVBUload [off1] {sym} (ADDVconst [off2] ptr) mem) && is32Bit(off1+off2) -> (MOVBUload [off1+off2] {sym} ptr mem)
   470  (MOVHload  [off1] {sym} (ADDVconst [off2] ptr) mem) && is32Bit(off1+off2) -> (MOVHload  [off1+off2] {sym} ptr mem)
   471  (MOVHUload [off1] {sym} (ADDVconst [off2] ptr) mem) && is32Bit(off1+off2) -> (MOVHUload [off1+off2] {sym} ptr mem)
   472  (MOVWload  [off1] {sym} (ADDVconst [off2] ptr) mem) && is32Bit(off1+off2) -> (MOVWload  [off1+off2] {sym} ptr mem)
   473  (MOVWUload [off1] {sym} (ADDVconst [off2] ptr) mem) && is32Bit(off1+off2) -> (MOVWUload [off1+off2] {sym} ptr mem)
   474  (MOVVload  [off1] {sym} (ADDVconst [off2] ptr) mem) && is32Bit(off1+off2) -> (MOVVload  [off1+off2] {sym} ptr mem)
   475  (MOVFload  [off1] {sym} (ADDVconst [off2] ptr) mem) && is32Bit(off1+off2) -> (MOVFload  [off1+off2] {sym} ptr mem)
   476  (MOVDload  [off1] {sym} (ADDVconst [off2] ptr) mem) && is32Bit(off1+off2) -> (MOVDload  [off1+off2] {sym} ptr mem)
   477  
   478  (MOVBstore [off1] {sym} (ADDVconst [off2] ptr) val mem) && is32Bit(off1+off2) -> (MOVBstore [off1+off2] {sym} ptr val mem)
   479  (MOVHstore [off1] {sym} (ADDVconst [off2] ptr) val mem) && is32Bit(off1+off2) -> (MOVHstore [off1+off2] {sym} ptr val mem)
   480  (MOVWstore [off1] {sym} (ADDVconst [off2] ptr) val mem) && is32Bit(off1+off2) -> (MOVWstore [off1+off2] {sym} ptr val mem)
   481  (MOVVstore [off1] {sym} (ADDVconst [off2] ptr) val mem) && is32Bit(off1+off2) -> (MOVVstore [off1+off2] {sym} ptr val mem)
   482  (MOVFstore [off1] {sym} (ADDVconst [off2] ptr) val mem) && is32Bit(off1+off2) -> (MOVFstore [off1+off2] {sym} ptr val mem)
   483  (MOVDstore [off1] {sym} (ADDVconst [off2] ptr) val mem) && is32Bit(off1+off2) -> (MOVDstore [off1+off2] {sym} ptr val mem)
   484  (MOVBstorezero [off1] {sym} (ADDVconst [off2] ptr) mem) && is32Bit(off1+off2) -> (MOVBstorezero [off1+off2] {sym} ptr mem)
   485  (MOVHstorezero [off1] {sym} (ADDVconst [off2] ptr) mem) && is32Bit(off1+off2) -> (MOVHstorezero [off1+off2] {sym} ptr mem)
   486  (MOVWstorezero [off1] {sym} (ADDVconst [off2] ptr) mem) && is32Bit(off1+off2) -> (MOVWstorezero [off1+off2] {sym} ptr mem)
   487  (MOVVstorezero [off1] {sym} (ADDVconst [off2] ptr) mem) && is32Bit(off1+off2) -> (MOVVstorezero [off1+off2] {sym} ptr mem)
   488  
   489  (MOVBload [off1] {sym1} (MOVVaddr [off2] {sym2} ptr) mem) && canMergeSym(sym1,sym2) && is32Bit(off1+off2) ->
   490  	(MOVBload [off1+off2] {mergeSym(sym1,sym2)} ptr mem)
   491  (MOVBUload [off1] {sym1} (MOVVaddr [off2] {sym2} ptr) mem) && canMergeSym(sym1,sym2) && is32Bit(off1+off2) ->
   492  	(MOVBUload [off1+off2] {mergeSym(sym1,sym2)} ptr mem)
   493  (MOVHload [off1] {sym1} (MOVVaddr [off2] {sym2} ptr) mem) && canMergeSym(sym1,sym2) && is32Bit(off1+off2) ->
   494  	(MOVHload [off1+off2] {mergeSym(sym1,sym2)} ptr mem)
   495  (MOVHUload [off1] {sym1} (MOVVaddr [off2] {sym2} ptr) mem) && canMergeSym(sym1,sym2) && is32Bit(off1+off2) ->
   496  	(MOVHUload [off1+off2] {mergeSym(sym1,sym2)} ptr mem)
   497  (MOVWload [off1] {sym1} (MOVVaddr [off2] {sym2} ptr) mem) && canMergeSym(sym1,sym2) && is32Bit(off1+off2) ->
   498  	(MOVWload [off1+off2] {mergeSym(sym1,sym2)} ptr mem)
   499  (MOVWUload [off1] {sym1} (MOVVaddr [off2] {sym2} ptr) mem) && canMergeSym(sym1,sym2) && is32Bit(off1+off2) ->
   500  	(MOVWUload [off1+off2] {mergeSym(sym1,sym2)} ptr mem)
   501  (MOVVload [off1] {sym1} (MOVVaddr [off2] {sym2} ptr) mem) && canMergeSym(sym1,sym2) && is32Bit(off1+off2) ->
   502  	(MOVVload [off1+off2] {mergeSym(sym1,sym2)} ptr mem)
   503  (MOVFload [off1] {sym1} (MOVVaddr [off2] {sym2} ptr) mem) && canMergeSym(sym1,sym2) && is32Bit(off1+off2) ->
   504  	(MOVFload [off1+off2] {mergeSym(sym1,sym2)} ptr mem)
   505  (MOVDload [off1] {sym1} (MOVVaddr [off2] {sym2} ptr) mem) && canMergeSym(sym1,sym2) && is32Bit(off1+off2) ->
   506  	(MOVDload [off1+off2] {mergeSym(sym1,sym2)} ptr mem)
   507  
   508  (MOVBstore [off1] {sym1} (MOVVaddr [off2] {sym2} ptr) val mem) && canMergeSym(sym1,sym2) && is32Bit(off1+off2) ->
   509  	(MOVBstore [off1+off2] {mergeSym(sym1,sym2)} ptr val mem)
   510  (MOVHstore [off1] {sym1} (MOVVaddr [off2] {sym2} ptr) val mem) && canMergeSym(sym1,sym2) && is32Bit(off1+off2) ->
   511  	(MOVHstore [off1+off2] {mergeSym(sym1,sym2)} ptr val mem)
   512  (MOVWstore [off1] {sym1} (MOVVaddr [off2] {sym2} ptr) val mem) && canMergeSym(sym1,sym2) && is32Bit(off1+off2) ->
   513  	(MOVWstore [off1+off2] {mergeSym(sym1,sym2)} ptr val mem)
   514  (MOVVstore [off1] {sym1} (MOVVaddr [off2] {sym2} ptr) val mem) && canMergeSym(sym1,sym2) && is32Bit(off1+off2) ->
   515  	(MOVVstore [off1+off2] {mergeSym(sym1,sym2)} ptr val mem)
   516  (MOVFstore [off1] {sym1} (MOVVaddr [off2] {sym2} ptr) val mem) && canMergeSym(sym1,sym2) && is32Bit(off1+off2) ->
   517  	(MOVFstore [off1+off2] {mergeSym(sym1,sym2)} ptr val mem)
   518  (MOVDstore [off1] {sym1} (MOVVaddr [off2] {sym2} ptr) val mem) && canMergeSym(sym1,sym2) && is32Bit(off1+off2) ->
   519  	(MOVDstore [off1+off2] {mergeSym(sym1,sym2)} ptr val mem)
   520  (MOVBstorezero [off1] {sym1} (MOVVaddr [off2] {sym2} ptr) mem) && canMergeSym(sym1,sym2) && is32Bit(off1+off2) ->
   521  	(MOVBstorezero [off1+off2] {mergeSym(sym1,sym2)} ptr mem)
   522  (MOVHstorezero [off1] {sym1} (MOVVaddr [off2] {sym2} ptr) mem) && canMergeSym(sym1,sym2) && is32Bit(off1+off2) ->
   523  	(MOVHstorezero [off1+off2] {mergeSym(sym1,sym2)} ptr mem)
   524  (MOVWstorezero [off1] {sym1} (MOVVaddr [off2] {sym2} ptr) mem) && canMergeSym(sym1,sym2) && is32Bit(off1+off2) ->
   525  	(MOVWstorezero [off1+off2] {mergeSym(sym1,sym2)} ptr mem)
   526  (MOVVstorezero [off1] {sym1} (MOVVaddr [off2] {sym2} ptr) mem) && canMergeSym(sym1,sym2) && is32Bit(off1+off2) ->
   527  	(MOVVstorezero [off1+off2] {mergeSym(sym1,sym2)} ptr mem)
   528  
   529  // store zero
   530  (MOVBstore [off] {sym} ptr (MOVVconst [0]) mem) -> (MOVBstorezero [off] {sym} ptr mem)
   531  (MOVHstore [off] {sym} ptr (MOVVconst [0]) mem) -> (MOVHstorezero [off] {sym} ptr mem)
   532  (MOVWstore [off] {sym} ptr (MOVVconst [0]) mem) -> (MOVWstorezero [off] {sym} ptr mem)
   533  (MOVVstore [off] {sym} ptr (MOVVconst [0]) mem) -> (MOVVstorezero [off] {sym} ptr mem)
   534  
   535  // don't extend after proper load
   536  (MOVBreg x:(MOVBload _ _)) -> (MOVVreg x)
   537  (MOVBUreg x:(MOVBUload _ _)) -> (MOVVreg x)
   538  (MOVHreg x:(MOVBload _ _)) -> (MOVVreg x)
   539  (MOVHreg x:(MOVBUload _ _)) -> (MOVVreg x)
   540  (MOVHreg x:(MOVHload _ _)) -> (MOVVreg x)
   541  (MOVHUreg x:(MOVBUload _ _)) -> (MOVVreg x)
   542  (MOVHUreg x:(MOVHUload _ _)) -> (MOVVreg x)
   543  (MOVWreg x:(MOVBload _ _)) -> (MOVVreg x)
   544  (MOVWreg x:(MOVBUload _ _)) -> (MOVVreg x)
   545  (MOVWreg x:(MOVHload _ _)) -> (MOVVreg x)
   546  (MOVWreg x:(MOVHUload _ _)) -> (MOVVreg x)
   547  (MOVWreg x:(MOVWload _ _)) -> (MOVVreg x)
   548  (MOVWUreg x:(MOVBUload _ _)) -> (MOVVreg x)
   549  (MOVWUreg x:(MOVHUload _ _)) -> (MOVVreg x)
   550  (MOVWUreg x:(MOVWUload _ _)) -> (MOVVreg x)
   551  
   552  // fold double extensions
   553  (MOVBreg x:(MOVBreg _)) -> (MOVVreg x)
   554  (MOVBUreg x:(MOVBUreg _)) -> (MOVVreg x)
   555  (MOVHreg x:(MOVBreg _)) -> (MOVVreg x)
   556  (MOVHreg x:(MOVBUreg _)) -> (MOVVreg x)
   557  (MOVHreg x:(MOVHreg _)) -> (MOVVreg x)
   558  (MOVHUreg x:(MOVBUreg _)) -> (MOVVreg x)
   559  (MOVHUreg x:(MOVHUreg _)) -> (MOVVreg x)
   560  (MOVWreg x:(MOVBreg _)) -> (MOVVreg x)
   561  (MOVWreg x:(MOVBUreg _)) -> (MOVVreg x)
   562  (MOVWreg x:(MOVHreg _)) -> (MOVVreg x)
   563  (MOVWreg x:(MOVHreg _)) -> (MOVVreg x)
   564  (MOVWreg x:(MOVWreg _)) -> (MOVVreg x)
   565  (MOVWUreg x:(MOVBUreg _)) -> (MOVVreg x)
   566  (MOVWUreg x:(MOVHUreg _)) -> (MOVVreg x)
   567  (MOVWUreg x:(MOVWUreg _)) -> (MOVVreg x)
   568  
   569  // don't extend before store
   570  (MOVBstore [off] {sym} ptr (MOVBreg x) mem) -> (MOVBstore [off] {sym} ptr x mem)
   571  (MOVBstore [off] {sym} ptr (MOVBUreg x) mem) -> (MOVBstore [off] {sym} ptr x mem)
   572  (MOVBstore [off] {sym} ptr (MOVHreg x) mem) -> (MOVBstore [off] {sym} ptr x mem)
   573  (MOVBstore [off] {sym} ptr (MOVHUreg x) mem) -> (MOVBstore [off] {sym} ptr x mem)
   574  (MOVBstore [off] {sym} ptr (MOVWreg x) mem) -> (MOVBstore [off] {sym} ptr x mem)
   575  (MOVBstore [off] {sym} ptr (MOVWUreg x) mem) -> (MOVBstore [off] {sym} ptr x mem)
   576  (MOVHstore [off] {sym} ptr (MOVHreg x) mem) -> (MOVHstore [off] {sym} ptr x mem)
   577  (MOVHstore [off] {sym} ptr (MOVHUreg x) mem) -> (MOVHstore [off] {sym} ptr x mem)
   578  (MOVHstore [off] {sym} ptr (MOVWreg x) mem) -> (MOVHstore [off] {sym} ptr x mem)
   579  (MOVHstore [off] {sym} ptr (MOVWUreg x) mem) -> (MOVHstore [off] {sym} ptr x mem)
   580  (MOVWstore [off] {sym} ptr (MOVWreg x) mem) -> (MOVWstore [off] {sym} ptr x mem)
   581  (MOVWstore [off] {sym} ptr (MOVWUreg x) mem) -> (MOVWstore [off] {sym} ptr x mem)
   582  
   583  // if a register move has only 1 use, just use the same register without emitting instruction
   584  // MOVVnop doesn't emit instruction, only for ensuring the type.
   585  (MOVVreg x) && x.Uses == 1 -> (MOVVnop x)
   586  
   587  // fold constant into arithmatic ops
   588  (ADDV x (MOVVconst [c])) && is32Bit(c) -> (ADDVconst [c] x)
   589  (SUBV x (MOVVconst [c])) && is32Bit(c) -> (SUBVconst [c] x)
   590  (AND x (MOVVconst [c])) && is32Bit(c) -> (ANDconst [c] x)
   591  (OR  x (MOVVconst [c])) && is32Bit(c) -> (ORconst  [c] x)
   592  (XOR x (MOVVconst [c])) && is32Bit(c) -> (XORconst [c] x)
   593  (NOR x (MOVVconst [c])) && is32Bit(c) -> (NORconst [c] x)
   594  
   595  (SLLV _ (MOVVconst [c])) && uint64(c)>=64 -> (MOVVconst [0])
   596  (SRLV _ (MOVVconst [c])) && uint64(c)>=64 -> (MOVVconst [0])
   597  (SRAV x (MOVVconst [c])) && uint64(c)>=64 -> (SRAVconst x [63])
   598  (SLLV x (MOVVconst [c])) -> (SLLVconst x [c])
   599  (SRLV x (MOVVconst [c])) -> (SRLVconst x [c])
   600  (SRAV x (MOVVconst [c])) -> (SRAVconst x [c])
   601  
   602  (SGT  (MOVVconst [c]) x) && is32Bit(c) -> (SGTconst  [c] x)
   603  (SGTU (MOVVconst [c]) x) && is32Bit(c) -> (SGTUconst [c] x)
   604  
   605  // mul by constant
   606  (Select1 (MULVU x (MOVVconst [-1]))) -> (NEGV x)
   607  (Select1 (MULVU _ (MOVVconst [0]))) -> (MOVVconst [0])
   608  (Select1 (MULVU x (MOVVconst [1]))) -> x
   609  (Select1 (MULVU x (MOVVconst [c]))) && isPowerOfTwo(c) -> (SLLVconst [log2(c)] x)
   610  
   611  (Select1 (MULVU (MOVVconst [-1]) x)) -> (NEGV x)
   612  (Select1 (MULVU (MOVVconst [0]) _)) -> (MOVVconst [0])
   613  (Select1 (MULVU (MOVVconst [1]) x)) -> x
   614  (Select1 (MULVU (MOVVconst [c]) x)) && isPowerOfTwo(c) -> (SLLVconst [log2(c)] x)
   615  
   616  // div by constant
   617  (Select1 (DIVVU x (MOVVconst [1]))) -> x
   618  (Select1 (DIVVU x (MOVVconst [c]))) && isPowerOfTwo(c) -> (SRLVconst [log2(c)] x)
   619  (Select0 (DIVVU _ (MOVVconst [1]))) -> (MOVVconst [0])                       // mod
   620  (Select0 (DIVVU x (MOVVconst [c]))) && isPowerOfTwo(c) -> (ANDconst [c-1] x) // mod
   621  
   622  // generic simplifications
   623  (ADDV x (NEGV y)) -> (SUBV x y)
   624  (SUBV x x) -> (MOVVconst [0])
   625  (SUBV (MOVVconst [0]) x) -> (NEGV x)
   626  (AND x x) -> x
   627  (OR  x x) -> x
   628  (XOR x x) -> (MOVVconst [0])
   629  
   630  // remove redundant *const ops
   631  (ADDVconst [0]  x) -> x
   632  (SUBVconst [0]  x) -> x
   633  (ANDconst [0]  _) -> (MOVVconst [0])
   634  (ANDconst [-1] x) -> x
   635  (ORconst  [0]  x) -> x
   636  (ORconst  [-1] _) -> (MOVVconst [-1])
   637  (XORconst [0]  x) -> x
   638  (XORconst [-1] x) -> (NORconst [0] x)
   639  
   640  // generic constant folding
   641  (ADDVconst [c] (MOVVconst [d]))  -> (MOVVconst [c+d])
   642  (ADDVconst [c] (ADDVconst [d] x)) && is32Bit(c+d) -> (ADDVconst [c+d] x)
   643  (ADDVconst [c] (SUBVconst [d] x)) && is32Bit(c-d) -> (ADDVconst [c-d] x)
   644  (SUBVconst [c] (MOVVconst [d]))  -> (MOVVconst [d-c])
   645  (SUBVconst [c] (SUBVconst [d] x)) && is32Bit(-c-d) -> (ADDVconst [-c-d] x)
   646  (SUBVconst [c] (ADDVconst [d] x)) && is32Bit(-c+d) -> (ADDVconst [-c+d] x)
   647  (SLLVconst [c] (MOVVconst [d]))  -> (MOVVconst [d<<uint64(c)])
   648  (SRLVconst [c] (MOVVconst [d]))  -> (MOVVconst [int64(uint64(d)>>uint64(c))])
   649  (SRAVconst [c] (MOVVconst [d]))  -> (MOVVconst [d>>uint64(c)])
   650  (Select1 (MULVU (MOVVconst [c]) (MOVVconst [d]))) -> (MOVVconst [c*d])
   651  (Select1 (DIVV  (MOVVconst [c]) (MOVVconst [d]))) -> (MOVVconst [c/d])
   652  (Select1 (DIVVU (MOVVconst [c]) (MOVVconst [d]))) -> (MOVVconst [int64(uint64(c)/uint64(d))])
   653  (Select0 (DIVV  (MOVVconst [c]) (MOVVconst [d]))) -> (MOVVconst [c%d])   // mod
   654  (Select0 (DIVVU (MOVVconst [c]) (MOVVconst [d]))) -> (MOVVconst [int64(uint64(c)%uint64(d))]) // mod
   655  (ANDconst [c] (MOVVconst [d])) -> (MOVVconst [c&d])
   656  (ANDconst [c] (ANDconst [d] x)) -> (ANDconst [c&d] x)
   657  (ORconst [c] (MOVVconst [d])) -> (MOVVconst [c|d])
   658  (ORconst [c] (ORconst [d] x)) && is32Bit(c|d) -> (ORconst [c|d] x)
   659  (XORconst [c] (MOVVconst [d])) -> (MOVVconst [c^d])
   660  (XORconst [c] (XORconst [d] x)) && is32Bit(c^d) -> (XORconst [c^d] x)
   661  (NORconst [c] (MOVVconst [d])) -> (MOVVconst [^(c|d)])
   662  (NEGV (MOVVconst [c])) -> (MOVVconst [-c])
   663  (MOVBreg  (MOVVconst [c])) -> (MOVVconst [int64(int8(c))])
   664  (MOVBUreg (MOVVconst [c])) -> (MOVVconst [int64(uint8(c))])
   665  (MOVHreg  (MOVVconst [c])) -> (MOVVconst [int64(int16(c))])
   666  (MOVHUreg (MOVVconst [c])) -> (MOVVconst [int64(uint16(c))])
   667  (MOVWreg  (MOVVconst [c])) -> (MOVVconst [int64(int32(c))])
   668  (MOVWUreg (MOVVconst [c])) -> (MOVVconst [int64(uint32(c))])
   669  (MOVVreg  (MOVVconst [c])) -> (MOVVconst [c])
   670  (LoweredAtomicStore32 ptr (MOVVconst [0]) mem) -> (LoweredAtomicStorezero32 ptr mem)
   671  (LoweredAtomicStore64 ptr (MOVVconst [0]) mem) -> (LoweredAtomicStorezero64 ptr mem)
   672  (LoweredAtomicAdd32 ptr (MOVVconst [c]) mem) && is32Bit(c) -> (LoweredAtomicAddconst32 [c] ptr mem)
   673  (LoweredAtomicAdd64 ptr (MOVVconst [c]) mem) && is32Bit(c) -> (LoweredAtomicAddconst64 [c] ptr mem)
   674  
   675  // constant comparisons
   676  (SGTconst [c] (MOVVconst [d])) && c>d -> (MOVVconst [1])
   677  (SGTconst [c] (MOVVconst [d])) && c<=d -> (MOVVconst [0])
   678  (SGTUconst [c] (MOVVconst [d])) && uint64(c)>uint64(d) -> (MOVVconst [1])
   679  (SGTUconst [c] (MOVVconst [d])) && uint64(c)<=uint64(d) -> (MOVVconst [0])
   680  
   681  // other known comparisons
   682  (SGTconst [c] (MOVBreg _)) && 0x7f < c -> (MOVVconst [1])
   683  (SGTconst [c] (MOVBreg _)) && c <= -0x80 -> (MOVVconst [0])
   684  (SGTconst [c] (MOVBUreg _)) && 0xff < c -> (MOVVconst [1])
   685  (SGTconst [c] (MOVBUreg _)) && c < 0 -> (MOVVconst [0])
   686  (SGTUconst [c] (MOVBUreg _)) && 0xff < uint64(c) -> (MOVVconst [1])
   687  (SGTconst [c] (MOVHreg _)) && 0x7fff < c -> (MOVVconst [1])
   688  (SGTconst [c] (MOVHreg _)) && c <= -0x8000 -> (MOVVconst [0])
   689  (SGTconst [c] (MOVHUreg _)) && 0xffff < c -> (MOVVconst [1])
   690  (SGTconst [c] (MOVHUreg _)) && c < 0 -> (MOVVconst [0])
   691  (SGTUconst [c] (MOVHUreg _)) && 0xffff < uint64(c) -> (MOVVconst [1])
   692  (SGTconst [c] (MOVWUreg _)) && c < 0 -> (MOVVconst [0])
   693  (SGTconst [c] (ANDconst [m] _)) && 0 <= m && m < c -> (MOVVconst [1])
   694  (SGTUconst [c] (ANDconst [m] _)) && uint64(m) < uint64(c) -> (MOVVconst [1])
   695  (SGTconst [c] (SRLVconst _ [d])) && 0 <= c && 0 < d && d <= 63 && 0xffffffffffffffff>>uint64(d) < uint64(c) -> (MOVVconst [1])
   696  (SGTUconst [c] (SRLVconst _ [d])) && 0 < d && d <= 63 && 0xffffffffffffffff>>uint64(d) < uint64(c) -> (MOVVconst [1])
   697  
   698  // absorb constants into branches
   699  (EQ  (MOVVconst [0]) yes no) -> (First yes no)
   700  (EQ  (MOVVconst [c]) yes no) && c != 0 -> (First no yes)
   701  (NE  (MOVVconst [0]) yes no) -> (First no yes)
   702  (NE  (MOVVconst [c]) yes no) && c != 0 -> (First yes no)
   703  (LTZ (MOVVconst [c]) yes no) && c <  0 -> (First yes no)
   704  (LTZ (MOVVconst [c]) yes no) && c >= 0 -> (First no yes)
   705  (LEZ (MOVVconst [c]) yes no) && c <= 0 -> (First yes no)
   706  (LEZ (MOVVconst [c]) yes no) && c >  0 -> (First no yes)
   707  (GTZ (MOVVconst [c]) yes no) && c >  0 -> (First yes no)
   708  (GTZ (MOVVconst [c]) yes no) && c <= 0 -> (First no yes)
   709  (GEZ (MOVVconst [c]) yes no) && c >= 0 -> (First yes no)
   710  (GEZ (MOVVconst [c]) yes no) && c <  0 -> (First no yes)