github.com/riscv/riscv-go@v0.0.0-20200123204226-124ebd6fcc8e/src/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  (AddPtr x y) -> (ADDV x y)
     6  (Add64 x y) -> (ADDV x y)
     7  (Add32 x y) -> (ADDV x y)
     8  (Add16 x y) -> (ADDV x y)
     9  (Add8 x y) -> (ADDV x y)
    10  (Add32F x y) -> (ADDF x y)
    11  (Add64F x y) -> (ADDD x y)
    12  
    13  (SubPtr x y) -> (SUBV x y)
    14  (Sub64 x y) -> (SUBV x y)
    15  (Sub32 x y) -> (SUBV x y)
    16  (Sub16 x y) -> (SUBV x y)
    17  (Sub8 x y) -> (SUBV x y)
    18  (Sub32F x y) -> (SUBF x y)
    19  (Sub64F x y) -> (SUBD x y)
    20  
    21  (Mul64 x y) -> (Select1 (MULVU x y))
    22  (Mul32 x y) -> (Select1 (MULVU x y))
    23  (Mul16 x y) -> (Select1 (MULVU x y))
    24  (Mul8 x y) -> (Select1 (MULVU x y))
    25  (Mul32F x y) -> (MULF x y)
    26  (Mul64F x y) -> (MULD x y)
    27  
    28  (Hmul64 x y) -> (Select0 (MULV x y))
    29  (Hmul64u x y) -> (Select0 (MULVU x y))
    30  (Hmul32 x y) -> (SRAVconst (Select1 <config.fe.TypeInt64()> (MULV (SignExt32to64 x) (SignExt32to64 y))) [32])
    31  (Hmul32u x y) -> (SRLVconst (Select1 <config.fe.TypeUInt64()> (MULVU (ZeroExt32to64 x) (ZeroExt32to64 y))) [32])
    32  (Hmul16 x y) -> (SRAVconst (Select1 <config.fe.TypeInt32()> (MULV (SignExt16to64 x) (SignExt16to64 y))) [16])
    33  (Hmul16u x y) -> (SRLVconst (Select1 <config.fe.TypeUInt32()> (MULVU (ZeroExt16to64 x) (ZeroExt16to64 y))) [16])
    34  (Hmul8 x y) -> (SRAVconst (Select1 <config.fe.TypeInt16()> (MULV (SignExt8to64 x) (SignExt8to64 y))) [8])
    35  (Hmul8u x y) -> (SRLVconst (Select1 <config.fe.TypeUInt16()> (MULVU (ZeroExt8to64 x) (ZeroExt8to64 y))) [8])
    36  
    37  (Div64 x y) -> (Select1 (DIVV x y))
    38  (Div64u x y) -> (Select1 (DIVVU x y))
    39  (Div32 x y) -> (Select1 (DIVV (SignExt32to64 x) (SignExt32to64 y)))
    40  (Div32u x y) -> (Select1 (DIVVU (ZeroExt32to64 x) (ZeroExt32to64 y)))
    41  (Div16 x y) -> (Select1 (DIVV (SignExt16to64 x) (SignExt16to64 y)))
    42  (Div16u x y) -> (Select1 (DIVVU (ZeroExt16to64 x) (ZeroExt16to64 y)))
    43  (Div8 x y) -> (Select1 (DIVV (SignExt8to64 x) (SignExt8to64 y)))
    44  (Div8u x y) -> (Select1 (DIVVU (ZeroExt8to64 x) (ZeroExt8to64 y)))
    45  (Div32F x y) -> (DIVF x y)
    46  (Div64F x y) -> (DIVD x y)
    47  
    48  (Mod64 x y) -> (Select0 (DIVV x y))
    49  (Mod64u x y) -> (Select0 (DIVVU x y))
    50  (Mod32 x y) -> (Select0 (DIVV (SignExt32to64 x) (SignExt32to64 y)))
    51  (Mod32u x y) -> (Select0 (DIVVU (ZeroExt32to64 x) (ZeroExt32to64 y)))
    52  (Mod16 x y) -> (Select0 (DIVV (SignExt16to64 x) (SignExt16to64 y)))
    53  (Mod16u x y) -> (Select0 (DIVVU (ZeroExt16to64 x) (ZeroExt16to64 y)))
    54  (Mod8 x y) -> (Select0 (DIVV (SignExt8to64 x) (SignExt8to64 y)))
    55  (Mod8u x y) -> (Select0 (DIVVU (ZeroExt8to64 x) (ZeroExt8to64 y)))
    56  
    57  (Avg64u <t> x y) -> (ADDV (ADDV <t> (SRLVconst <t> x [1]) (SRLVconst <t> y [1])) (AND <t> (AND <t> x y) (MOVVconst [1])))
    58  
    59  (And64 x y) -> (AND x y)
    60  (And32 x y) -> (AND x y)
    61  (And16 x y) -> (AND x y)
    62  (And8 x y) -> (AND x y)
    63  
    64  (Or64 x y) -> (OR x y)
    65  (Or32 x y) -> (OR x y)
    66  (Or16 x y) -> (OR x y)
    67  (Or8 x y) -> (OR x y)
    68  
    69  (Xor64 x y) -> (XOR x y)
    70  (Xor32 x y) -> (XOR x y)
    71  (Xor16 x y) -> (XOR x y)
    72  (Xor8 x y) -> (XOR x y)
    73  
    74  // shifts
    75  // hardware instruction uses only the low 6 bits of the shift
    76  // we compare to 64 to ensure Go semantics for large shifts
    77  (Lsh64x64 <t> x y) -> (AND (NEGV <t> (SGTU (Const64 <config.fe.TypeUInt64()> [64]) y)) (SLLV <t> x y))
    78  (Lsh64x32 <t> x y) -> (AND (NEGV <t> (SGTU (Const64 <config.fe.TypeUInt64()> [64]) (ZeroExt32to64 y))) (SLLV <t> x (ZeroExt32to64 y)))
    79  (Lsh64x16 <t> x y) -> (AND (NEGV <t> (SGTU (Const64 <config.fe.TypeUInt64()> [64]) (ZeroExt16to64 y))) (SLLV <t> x (ZeroExt16to64 y)))
    80  (Lsh64x8  <t> x y) -> (AND (NEGV <t> (SGTU (Const64 <config.fe.TypeUInt64()> [64]) (ZeroExt8to64  y))) (SLLV <t> x (ZeroExt8to64  y)))
    81  
    82  (Lsh32x64 <t> x y) -> (AND (NEGV <t> (SGTU (Const64 <config.fe.TypeUInt64()> [64]) y)) (SLLV <t> x y))
    83  (Lsh32x32 <t> x y) -> (AND (NEGV <t> (SGTU (Const64 <config.fe.TypeUInt64()> [64]) (ZeroExt32to64 y))) (SLLV <t> x (ZeroExt32to64 y)))
    84  (Lsh32x16 <t> x y) -> (AND (NEGV <t> (SGTU (Const64 <config.fe.TypeUInt64()> [64]) (ZeroExt16to64 y))) (SLLV <t> x (ZeroExt16to64 y)))
    85  (Lsh32x8  <t> x y) -> (AND (NEGV <t> (SGTU (Const64 <config.fe.TypeUInt64()> [64]) (ZeroExt8to64  y))) (SLLV <t> x (ZeroExt8to64  y)))
    86  
    87  (Lsh16x64 <t> x y) -> (AND (NEGV <t> (SGTU (Const64 <config.fe.TypeUInt64()> [64]) y)) (SLLV <t> x y))
    88  (Lsh16x32 <t> x y) -> (AND (NEGV <t> (SGTU (Const64 <config.fe.TypeUInt64()> [64]) (ZeroExt32to64 y))) (SLLV <t> x (ZeroExt32to64 y)))
    89  (Lsh16x16 <t> x y) -> (AND (NEGV <t> (SGTU (Const64 <config.fe.TypeUInt64()> [64]) (ZeroExt16to64 y))) (SLLV <t> x (ZeroExt16to64 y)))
    90  (Lsh16x8  <t> x y) -> (AND (NEGV <t> (SGTU (Const64 <config.fe.TypeUInt64()> [64]) (ZeroExt8to64  y))) (SLLV <t> x (ZeroExt8to64  y)))
    91  
    92  (Lsh8x64 <t> x y) -> (AND (NEGV <t> (SGTU (Const64 <config.fe.TypeUInt64()> [64]) y)) (SLLV <t> x y))
    93  (Lsh8x32 <t> x y) -> (AND (NEGV <t> (SGTU (Const64 <config.fe.TypeUInt64()> [64]) (ZeroExt32to64 y))) (SLLV <t> x (ZeroExt32to64 y)))
    94  (Lsh8x16 <t> x y) -> (AND (NEGV <t> (SGTU (Const64 <config.fe.TypeUInt64()> [64]) (ZeroExt16to64 y))) (SLLV <t> x (ZeroExt16to64 y)))
    95  (Lsh8x8  <t> x y) -> (AND (NEGV <t> (SGTU (Const64 <config.fe.TypeUInt64()> [64]) (ZeroExt8to64  y))) (SLLV <t> x (ZeroExt8to64  y)))
    96  
    97  (Rsh64Ux64 <t> x y) -> (AND (NEGV <t> (SGTU (Const64 <config.fe.TypeUInt64()> [64]) y)) (SRLV <t> x y))
    98  (Rsh64Ux32 <t> x y) -> (AND (NEGV <t> (SGTU (Const64 <config.fe.TypeUInt64()> [64]) (ZeroExt32to64 y))) (SRLV <t> x (ZeroExt32to64 y)))
    99  (Rsh64Ux16 <t> x y) -> (AND (NEGV <t> (SGTU (Const64 <config.fe.TypeUInt64()> [64]) (ZeroExt16to64 y))) (SRLV <t> x (ZeroExt16to64 y)))
   100  (Rsh64Ux8  <t> x y) -> (AND (NEGV <t> (SGTU (Const64 <config.fe.TypeUInt64()> [64]) (ZeroExt8to64  y))) (SRLV <t> x (ZeroExt8to64  y)))
   101  
   102  (Rsh32Ux64 <t> x y) -> (AND (NEGV <t> (SGTU (Const64 <config.fe.TypeUInt64()> [64]) y)) (SRLV <t> (ZeroExt32to64 x) y))
   103  (Rsh32Ux32 <t> x y) -> (AND (NEGV <t> (SGTU (Const64 <config.fe.TypeUInt64()> [64]) (ZeroExt32to64 y))) (SRLV <t> (ZeroExt32to64 x) (ZeroExt32to64 y)))
   104  (Rsh32Ux16 <t> x y) -> (AND (NEGV <t> (SGTU (Const64 <config.fe.TypeUInt64()> [64]) (ZeroExt16to64 y))) (SRLV <t> (ZeroExt32to64 x) (ZeroExt16to64 y)))
   105  (Rsh32Ux8  <t> x y) -> (AND (NEGV <t> (SGTU (Const64 <config.fe.TypeUInt64()> [64]) (ZeroExt8to64  y))) (SRLV <t> (ZeroExt32to64 x) (ZeroExt8to64  y)))
   106  
   107  (Rsh16Ux64 <t> x y) -> (AND (NEGV <t> (SGTU (Const64 <config.fe.TypeUInt64()> [64]) y)) (SRLV <t> (ZeroExt16to64 x) y))
   108  (Rsh16Ux32 <t> x y) -> (AND (NEGV <t> (SGTU (Const64 <config.fe.TypeUInt64()> [64]) (ZeroExt32to64 y))) (SRLV <t> (ZeroExt16to64 x) (ZeroExt32to64 y)))
   109  (Rsh16Ux16 <t> x y) -> (AND (NEGV <t> (SGTU (Const64 <config.fe.TypeUInt64()> [64]) (ZeroExt16to64 y))) (SRLV <t> (ZeroExt16to64 x) (ZeroExt16to64 y)))
   110  (Rsh16Ux8  <t> x y) -> (AND (NEGV <t> (SGTU (Const64 <config.fe.TypeUInt64()> [64]) (ZeroExt8to64  y))) (SRLV <t> (ZeroExt16to64 x) (ZeroExt8to64  y)))
   111  
   112  (Rsh8Ux64 <t> x y) -> (AND (NEGV <t> (SGTU (Const64 <config.fe.TypeUInt64()> [64]) y)) (SRLV <t> (ZeroExt8to64 x) y))
   113  (Rsh8Ux32 <t> x y) -> (AND (NEGV <t> (SGTU (Const64 <config.fe.TypeUInt64()> [64]) (ZeroExt32to64 y))) (SRLV <t> (ZeroExt8to64 x) (ZeroExt32to64 y)))
   114  (Rsh8Ux16 <t> x y) -> (AND (NEGV <t> (SGTU (Const64 <config.fe.TypeUInt64()> [64]) (ZeroExt16to64 y))) (SRLV <t> (ZeroExt8to64 x) (ZeroExt16to64 y)))
   115  (Rsh8Ux8  <t> x y) -> (AND (NEGV <t> (SGTU (Const64 <config.fe.TypeUInt64()> [64]) (ZeroExt8to64  y))) (SRLV <t> (ZeroExt8to64 x) (ZeroExt8to64  y)))
   116  
   117  (Rsh64x64 <t> x y) -> (SRAV x (OR <t> (NEGV <t> (SGTU y (Const64 <config.fe.TypeUInt64()> [63]))) y))
   118  (Rsh64x32 <t> x y) -> (SRAV x (OR <t> (NEGV <t> (SGTU (ZeroExt32to64 y) (Const64 <config.fe.TypeUInt64()> [63]))) (ZeroExt32to64 y)))
   119  (Rsh64x16 <t> x y) -> (SRAV x (OR <t> (NEGV <t> (SGTU (ZeroExt16to64 y) (Const64 <config.fe.TypeUInt64()> [63]))) (ZeroExt16to64 y)))
   120  (Rsh64x8  <t> x y) -> (SRAV x (OR <t> (NEGV <t> (SGTU (ZeroExt8to64  y) (Const64 <config.fe.TypeUInt64()> [63]))) (ZeroExt8to64  y)))
   121  
   122  (Rsh32x64 <t> x y) -> (SRAV (SignExt32to64 x) (OR <t> (NEGV <t> (SGTU y (Const64 <config.fe.TypeUInt64()> [63]))) y))
   123  (Rsh32x32 <t> x y) -> (SRAV (SignExt32to64 x) (OR <t> (NEGV <t> (SGTU (ZeroExt32to64 y) (Const64 <config.fe.TypeUInt64()> [63]))) (ZeroExt32to64 y)))
   124  (Rsh32x16 <t> x y) -> (SRAV (SignExt32to64 x) (OR <t> (NEGV <t> (SGTU (ZeroExt16to64 y) (Const64 <config.fe.TypeUInt64()> [63]))) (ZeroExt16to64 y)))
   125  (Rsh32x8  <t> x y) -> (SRAV (SignExt32to64 x) (OR <t> (NEGV <t> (SGTU (ZeroExt8to64  y) (Const64 <config.fe.TypeUInt64()> [63]))) (ZeroExt8to64  y)))
   126  
   127  (Rsh16x64 <t> x y) -> (SRAV (SignExt16to64 x) (OR <t> (NEGV <t> (SGTU y (Const64 <config.fe.TypeUInt64()> [63]))) y))
   128  (Rsh16x32 <t> x y) -> (SRAV (SignExt16to64 x) (OR <t> (NEGV <t> (SGTU (ZeroExt32to64 y) (Const64 <config.fe.TypeUInt64()> [63]))) (ZeroExt32to64 y)))
   129  (Rsh16x16 <t> x y) -> (SRAV (SignExt16to64 x) (OR <t> (NEGV <t> (SGTU (ZeroExt16to64 y) (Const64 <config.fe.TypeUInt64()> [63]))) (ZeroExt16to64 y)))
   130  (Rsh16x8  <t> x y) -> (SRAV (SignExt16to64 x) (OR <t> (NEGV <t> (SGTU (ZeroExt8to64  y) (Const64 <config.fe.TypeUInt64()> [63]))) (ZeroExt8to64  y)))
   131  
   132  (Rsh8x64 <t> x y) -> (SRAV (SignExt8to64 x) (OR <t> (NEGV <t> (SGTU y (Const64 <config.fe.TypeUInt64()> [63]))) y))
   133  (Rsh8x32 <t> x y) -> (SRAV (SignExt8to64 x) (OR <t> (NEGV <t> (SGTU (ZeroExt32to64 y) (Const64 <config.fe.TypeUInt64()> [63]))) (ZeroExt32to64 y)))
   134  (Rsh8x16 <t> x y) -> (SRAV (SignExt8to64 x) (OR <t> (NEGV <t> (SGTU (ZeroExt16to64 y) (Const64 <config.fe.TypeUInt64()> [63]))) (ZeroExt16to64 y)))
   135  (Rsh8x8  <t> x y) -> (SRAV (SignExt8to64 x) (OR <t> (NEGV <t> (SGTU (ZeroExt8to64  y) (Const64 <config.fe.TypeUInt64()> [63]))) (ZeroExt8to64  y)))
   136  
   137  // unary ops
   138  (Neg64 x) -> (NEGV x)
   139  (Neg32 x) -> (NEGV x)
   140  (Neg16 x) -> (NEGV x)
   141  (Neg8 x) -> (NEGV x)
   142  (Neg32F x) -> (NEGF x)
   143  (Neg64F x) -> (NEGD x)
   144  
   145  (Com64 x) -> (NOR (MOVVconst [0]) x)
   146  (Com32 x) -> (NOR (MOVVconst [0]) x)
   147  (Com16 x) -> (NOR (MOVVconst [0]) x)
   148  (Com8 x) -> (NOR (MOVVconst [0]) x)
   149  
   150  // boolean ops -- booleans are represented with 0=false, 1=true
   151  (AndB x y) -> (AND x y)
   152  (OrB x y) -> (OR x y)
   153  (EqB x y) -> (XOR (MOVVconst [1]) (XOR <config.fe.TypeBool()> x y))
   154  (NeqB x y) -> (XOR x y)
   155  (Not x) -> (XORconst [1] x)
   156  
   157  // constants
   158  (Const64 [val]) -> (MOVVconst [val])
   159  (Const32 [val]) -> (MOVVconst [val])
   160  (Const16 [val]) -> (MOVVconst [val])
   161  (Const8 [val]) -> (MOVVconst [val])
   162  (Const32F [val]) -> (MOVFconst [val])
   163  (Const64F [val]) -> (MOVDconst [val])
   164  (ConstNil) -> (MOVVconst [0])
   165  (ConstBool [b]) -> (MOVVconst [b])
   166  
   167  (Slicemask <t> x) -> (SRAVconst (NEGV <t> x) [63])
   168  
   169  // truncations
   170  // Because we ignore high parts of registers, truncates are just copies.
   171  (Trunc16to8 x) -> x
   172  (Trunc32to8 x) -> x
   173  (Trunc32to16 x) -> x
   174  (Trunc64to8 x) -> x
   175  (Trunc64to16 x) -> x
   176  (Trunc64to32 x) -> x
   177  
   178  // Zero-/Sign-extensions
   179  (ZeroExt8to16 x) -> (MOVBUreg x)
   180  (ZeroExt8to32 x) -> (MOVBUreg x)
   181  (ZeroExt16to32 x) -> (MOVHUreg x)
   182  (ZeroExt8to64 x) -> (MOVBUreg x)
   183  (ZeroExt16to64 x) -> (MOVHUreg x)
   184  (ZeroExt32to64 x) -> (MOVWUreg x)
   185  
   186  (SignExt8to16 x) -> (MOVBreg x)
   187  (SignExt8to32 x) -> (MOVBreg x)
   188  (SignExt16to32 x) -> (MOVHreg x)
   189  (SignExt8to64 x) -> (MOVBreg x)
   190  (SignExt16to64 x) -> (MOVHreg x)
   191  (SignExt32to64 x) -> (MOVWreg x)
   192  
   193  // float <-> int conversion
   194  (Cvt32to32F x) -> (MOVWF x)
   195  (Cvt32to64F x) -> (MOVWD x)
   196  (Cvt64to32F x) -> (MOVVF x)
   197  (Cvt64to64F x) -> (MOVVD x)
   198  (Cvt32Fto32 x) -> (TRUNCFW x)
   199  (Cvt64Fto32 x) -> (TRUNCDW x)
   200  (Cvt32Fto64 x) -> (TRUNCFV x)
   201  (Cvt64Fto64 x) -> (TRUNCDV x)
   202  (Cvt32Fto64F x) -> (MOVFD x)
   203  (Cvt64Fto32F x) -> (MOVDF x)
   204  
   205  // comparisons
   206  (Eq8 x y)  -> (SGTU (MOVVconst [1]) (XOR (ZeroExt8to64 x) (ZeroExt8to64 y)))
   207  (Eq16 x y) -> (SGTU (MOVVconst [1]) (XOR (ZeroExt16to64 x) (ZeroExt16to64 y)))
   208  (Eq32 x y) -> (SGTU (MOVVconst [1]) (XOR (ZeroExt32to64 x) (ZeroExt32to64 y)))
   209  (Eq64 x y) -> (SGTU (MOVVconst [1]) (XOR x y))
   210  (EqPtr x y) -> (SGTU (MOVVconst [1]) (XOR x y))
   211  (Eq32F x y) -> (FPFlagTrue (CMPEQF x y))
   212  (Eq64F x y) -> (FPFlagTrue (CMPEQD x y))
   213  
   214  (Neq8 x y)  -> (SGTU (XOR (ZeroExt8to64 x) (ZeroExt8to64 y)) (MOVVconst [0]))
   215  (Neq16 x y) -> (SGTU (XOR (ZeroExt16to32 x) (ZeroExt16to64 y)) (MOVVconst [0]))
   216  (Neq32 x y) -> (SGTU (XOR (ZeroExt32to64 x) (ZeroExt32to64 y)) (MOVVconst [0]))
   217  (Neq64 x y) -> (SGTU (XOR x y) (MOVVconst [0]))
   218  (NeqPtr x y) -> (SGTU (XOR x y) (MOVVconst [0]))
   219  (Neq32F x y) -> (FPFlagFalse (CMPEQF x y))
   220  (Neq64F x y) -> (FPFlagFalse (CMPEQD x y))
   221  
   222  (Less8 x y)  -> (SGT (SignExt8to64 y) (SignExt8to64 x))
   223  (Less16 x y) -> (SGT (SignExt16to64 y) (SignExt16to64 x))
   224  (Less32 x y) -> (SGT (SignExt32to64 y) (SignExt32to64 x))
   225  (Less64 x y) -> (SGT y x)
   226  (Less32F x y) -> (FPFlagTrue (CMPGTF y x)) // reverse operands to work around NaN
   227  (Less64F x y) -> (FPFlagTrue (CMPGTD y x)) // reverse operands to work around NaN
   228  
   229  (Less8U x y)  -> (SGTU (ZeroExt8to64 y) (ZeroExt8to64 x))
   230  (Less16U x y) -> (SGTU (ZeroExt16to64 y) (ZeroExt16to64 x))
   231  (Less32U x y) -> (SGTU (ZeroExt32to64 y) (ZeroExt32to64 x))
   232  (Less64U x y) -> (SGTU y x)
   233  
   234  (Leq8 x y)  -> (XOR (MOVVconst [1]) (SGT (SignExt8to64 x) (SignExt8to64 y)))
   235  (Leq16 x y) -> (XOR (MOVVconst [1]) (SGT (SignExt16to64 x) (SignExt16to64 y)))
   236  (Leq32 x y) -> (XOR (MOVVconst [1]) (SGT (SignExt32to64 x) (SignExt32to64 y)))
   237  (Leq64 x y) -> (XOR (MOVVconst [1]) (SGT x y))
   238  (Leq32F x y) -> (FPFlagTrue (CMPGEF y x)) // reverse operands to work around NaN
   239  (Leq64F x y) -> (FPFlagTrue (CMPGED y x)) // reverse operands to work around NaN
   240  
   241  (Leq8U x y)  -> (XOR (MOVVconst [1]) (SGTU (ZeroExt8to64 x) (ZeroExt8to64 y)))
   242  (Leq16U x y) -> (XOR (MOVVconst [1]) (SGTU (ZeroExt16to64 x) (ZeroExt16to64 y)))
   243  (Leq32U x y) -> (XOR (MOVVconst [1]) (SGTU (ZeroExt32to64 x) (ZeroExt32to64 y)))
   244  (Leq64U x y) -> (XOR (MOVVconst [1]) (SGTU x y))
   245  
   246  (Greater8 x y)  -> (SGT (SignExt8to64 x) (SignExt8to64 y))
   247  (Greater16 x y) -> (SGT (SignExt16to64 x) (SignExt16to64 y))
   248  (Greater32 x y) -> (SGT (SignExt32to64 x) (SignExt32to64 y))
   249  (Greater64 x y) -> (SGT x y)
   250  (Greater32F x y) -> (FPFlagTrue (CMPGTF x y))
   251  (Greater64F x y) -> (FPFlagTrue (CMPGTD x y))
   252  
   253  (Greater8U x y)  -> (SGTU (ZeroExt8to64 x) (ZeroExt8to64 y))
   254  (Greater16U x y) -> (SGTU (ZeroExt16to64 x) (ZeroExt16to64 y))
   255  (Greater32U x y) -> (SGTU (ZeroExt32to64 x) (ZeroExt32to64 y))
   256  (Greater64U x y) -> (SGTU x y)
   257  
   258  (Geq8 x y)  -> (XOR (MOVVconst [1]) (SGT (SignExt8to64 y) (SignExt8to64 x)))
   259  (Geq16 x y) -> (XOR (MOVVconst [1]) (SGT (SignExt16to64 y) (SignExt16to64 x)))
   260  (Geq32 x y) -> (XOR (MOVVconst [1]) (SGT (SignExt32to64 y) (SignExt32to64 x)))
   261  (Geq64 x y) -> (XOR (MOVVconst [1]) (SGT y x))
   262  (Geq32F x y) -> (FPFlagTrue (CMPGEF x y))
   263  (Geq64F x y) -> (FPFlagTrue (CMPGED x y))
   264  
   265  (Geq8U x y)  -> (XOR (MOVVconst [1]) (SGTU (ZeroExt8to64 y) (ZeroExt8to64 x)))
   266  (Geq16U x y) -> (XOR (MOVVconst [1]) (SGTU (ZeroExt16to64 y) (ZeroExt16to64 x)))
   267  (Geq32U x y) -> (XOR (MOVVconst [1]) (SGTU (ZeroExt32to64 y) (ZeroExt32to64 x)))
   268  (Geq64U x y) -> (XOR (MOVVconst [1]) (SGTU y x))
   269  
   270  (OffPtr [off] ptr:(SP)) -> (MOVVaddr [off] ptr)
   271  (OffPtr [off] ptr) -> (ADDVconst [off] ptr)
   272  
   273  (Addr {sym} base) -> (MOVVaddr {sym} base)
   274  
   275  // loads
   276  (Load <t> ptr mem) && t.IsBoolean() -> (MOVBUload ptr mem)
   277  (Load <t> ptr mem) && (is8BitInt(t) && isSigned(t)) -> (MOVBload ptr mem)
   278  (Load <t> ptr mem) && (is8BitInt(t) && !isSigned(t)) -> (MOVBUload ptr mem)
   279  (Load <t> ptr mem) && (is16BitInt(t) && isSigned(t)) -> (MOVHload ptr mem)
   280  (Load <t> ptr mem) && (is16BitInt(t) && !isSigned(t)) -> (MOVHUload ptr mem)
   281  (Load <t> ptr mem) && (is32BitInt(t) && isSigned(t)) -> (MOVWload ptr mem)
   282  (Load <t> ptr mem) && (is32BitInt(t) && !isSigned(t)) -> (MOVWUload ptr mem)
   283  (Load <t> ptr mem) && (is64BitInt(t) || isPtr(t)) -> (MOVVload ptr mem)
   284  (Load <t> ptr mem) && is32BitFloat(t) -> (MOVFload ptr mem)
   285  (Load <t> ptr mem) && is64BitFloat(t) -> (MOVDload ptr mem)
   286  
   287  // stores
   288  (Store [1] ptr val mem) -> (MOVBstore ptr val mem)
   289  (Store [2] ptr val mem) -> (MOVHstore ptr val mem)
   290  (Store [4] ptr val mem) && !is32BitFloat(val.Type) -> (MOVWstore ptr val mem)
   291  (Store [8] ptr val mem) && !is64BitFloat(val.Type) -> (MOVVstore ptr val mem)
   292  (Store [4] ptr val mem) && is32BitFloat(val.Type) -> (MOVFstore ptr val mem)
   293  (Store [8] ptr val mem) && is64BitFloat(val.Type) -> (MOVDstore ptr val mem)
   294  
   295  // zeroing
   296  (Zero [s] _ mem) && SizeAndAlign(s).Size() == 0 -> mem
   297  (Zero [s] ptr mem) && SizeAndAlign(s).Size() == 1 -> (MOVBstore ptr (MOVVconst [0]) mem)
   298  (Zero [s] ptr mem) && SizeAndAlign(s).Size() == 2 && SizeAndAlign(s).Align()%2 == 0 ->
   299  	(MOVHstore ptr (MOVVconst [0]) mem)
   300  (Zero [s] ptr mem) && SizeAndAlign(s).Size() == 2 ->
   301  	(MOVBstore [1] ptr (MOVVconst [0])
   302  		(MOVBstore [0] ptr (MOVVconst [0]) mem))
   303  (Zero [s] ptr mem) && SizeAndAlign(s).Size() == 4 && SizeAndAlign(s).Align()%4 == 0 ->
   304  	(MOVWstore ptr (MOVVconst [0]) mem)
   305  (Zero [s] ptr mem) && SizeAndAlign(s).Size() == 4 && SizeAndAlign(s).Align()%2 == 0 ->
   306  	(MOVHstore [2] ptr (MOVVconst [0])
   307  		(MOVHstore [0] ptr (MOVVconst [0]) mem))
   308  (Zero [s] ptr mem) && SizeAndAlign(s).Size() == 4 ->
   309  	(MOVBstore [3] ptr (MOVVconst [0])
   310  		(MOVBstore [2] ptr (MOVVconst [0])
   311  			(MOVBstore [1] ptr (MOVVconst [0])
   312  				(MOVBstore [0] ptr (MOVVconst [0]) mem))))
   313  (Zero [s] ptr mem) && SizeAndAlign(s).Size() == 8 && SizeAndAlign(s).Align()%8 == 0 ->
   314  	(MOVVstore ptr (MOVVconst [0]) mem)
   315  (Zero [s] ptr mem) && SizeAndAlign(s).Size() == 8 && SizeAndAlign(s).Align()%4 == 0 ->
   316  	(MOVWstore [4] ptr (MOVVconst [0])
   317  		(MOVWstore [0] ptr (MOVVconst [0]) mem))
   318  (Zero [s] ptr mem) && SizeAndAlign(s).Size() == 4 ->
   319  	(MOVHstore [6] ptr (MOVVconst [0])
   320  		(MOVHstore [4] ptr (MOVVconst [0])
   321  			(MOVHstore [2] ptr (MOVVconst [0])
   322  				(MOVHstore [0] ptr (MOVVconst [0]) mem))))
   323  
   324  (Zero [s] ptr mem) && SizeAndAlign(s).Size() == 3 ->
   325  	(MOVBstore [2] ptr (MOVVconst [0])
   326  		(MOVBstore [1] ptr (MOVVconst [0])
   327  			(MOVBstore [0] ptr (MOVVconst [0]) mem)))
   328  (Zero [s] ptr mem) && SizeAndAlign(s).Size() == 6 && SizeAndAlign(s).Align()%2 == 0 ->
   329  	(MOVHstore [4] ptr (MOVVconst [0])
   330  		(MOVHstore [2] ptr (MOVVconst [0])
   331  			(MOVHstore [0] ptr (MOVVconst [0]) mem)))
   332  (Zero [s] ptr mem) && SizeAndAlign(s).Size() == 12 && SizeAndAlign(s).Align()%4 == 0 ->
   333  	(MOVWstore [8] ptr (MOVVconst [0])
   334  		(MOVWstore [4] ptr (MOVVconst [0])
   335  			(MOVWstore [0] ptr (MOVVconst [0]) mem)))
   336  (Zero [s] ptr mem) && SizeAndAlign(s).Size() == 16 && SizeAndAlign(s).Align()%8 == 0 ->
   337  	(MOVVstore [8] ptr (MOVVconst [0])
   338  		(MOVVstore [0] ptr (MOVVconst [0]) mem))
   339  (Zero [s] ptr mem) && SizeAndAlign(s).Size() == 24 && SizeAndAlign(s).Align()%8 == 0 ->
   340  	(MOVVstore [16] ptr (MOVVconst [0])
   341  		(MOVVstore [8] ptr (MOVVconst [0])
   342  			(MOVVstore [0] ptr (MOVVconst [0]) mem)))
   343  
   344  // medium zeroing uses a duff device
   345  // 8, and 128 are magic constants, see runtime/mkduff.go
   346  (Zero [s] ptr mem)
   347  	&& SizeAndAlign(s).Size()%8 == 0 && SizeAndAlign(s).Size() > 24 && SizeAndAlign(s).Size() <= 8*128
   348  	&& SizeAndAlign(s).Align()%8 == 0 && !config.noDuffDevice ->
   349  	(DUFFZERO [8 * (128 - int64(SizeAndAlign(s).Size()/8))] ptr mem)
   350  
   351  // large or unaligned zeroing uses a loop
   352  (Zero [s] ptr mem)
   353  	&& (SizeAndAlign(s).Size() > 8*128 || config.noDuffDevice) || SizeAndAlign(s).Align()%8 != 0 ->
   354  	(LoweredZero [SizeAndAlign(s).Align()]
   355  		ptr
   356  		(ADDVconst <ptr.Type> ptr [SizeAndAlign(s).Size()-moveSize(SizeAndAlign(s).Align(), config)])
   357  		mem)
   358  
   359  // moves
   360  (Move [s] _ _ mem) && SizeAndAlign(s).Size() == 0 -> mem
   361  (Move [s] dst src mem) && SizeAndAlign(s).Size() == 1 -> (MOVBstore dst (MOVBload src mem) mem)
   362  (Move [s] dst src mem) && SizeAndAlign(s).Size() == 2 && SizeAndAlign(s).Align()%2 == 0 ->
   363  	(MOVHstore dst (MOVHload src mem) mem)
   364  (Move [s] dst src mem) && SizeAndAlign(s).Size() == 2 ->
   365  	(MOVBstore [1] dst (MOVBload [1] src mem)
   366  		(MOVBstore dst (MOVBload src mem) mem))
   367  (Move [s] dst src mem) && SizeAndAlign(s).Size() == 4 && SizeAndAlign(s).Align()%4 == 0 ->
   368  	(MOVWstore dst (MOVWload src mem) mem)
   369  (Move [s] dst src mem) && SizeAndAlign(s).Size() == 4 && SizeAndAlign(s).Align()%2 == 0 ->
   370  	(MOVHstore [2] dst (MOVHload [2] src mem)
   371  		(MOVHstore dst (MOVHload src mem) mem))
   372  (Move [s] dst src mem) && SizeAndAlign(s).Size() == 4 ->
   373  	(MOVBstore [3] dst (MOVBload [3] src mem)
   374  		(MOVBstore [2] dst (MOVBload [2] src mem)
   375  			(MOVBstore [1] dst (MOVBload [1] src mem)
   376  				(MOVBstore dst (MOVBload src mem) mem))))
   377  (Move [s] dst src mem) && SizeAndAlign(s).Size() == 8 && SizeAndAlign(s).Align()%8 == 0 ->
   378  	(MOVVstore dst (MOVVload src mem) mem)
   379  (Move [s] dst src mem) && SizeAndAlign(s).Size() == 8 && SizeAndAlign(s).Align()%4 == 0 ->
   380  	(MOVWstore [4] dst (MOVWload [4] src mem)
   381  		(MOVWstore dst (MOVWload src mem) mem))
   382  (Move [s] dst src mem) && SizeAndAlign(s).Size() == 8 && SizeAndAlign(s).Align()%2 == 0 ->
   383  	(MOVHstore [6] dst (MOVHload [6] src mem)
   384  		(MOVHstore [4] dst (MOVHload [4] src mem)
   385  			(MOVHstore [2] dst (MOVHload [2] src mem)
   386  				(MOVHstore dst (MOVHload src mem) mem))))
   387  
   388  (Move [s] dst src mem) && SizeAndAlign(s).Size() == 3 ->
   389  	(MOVBstore [2] dst (MOVBload [2] src mem)
   390  		(MOVBstore [1] dst (MOVBload [1] src mem)
   391  			(MOVBstore dst (MOVBload src mem) mem)))
   392  (Move [s] dst src mem) && SizeAndAlign(s).Size() == 6 && SizeAndAlign(s).Align()%2 == 0 ->
   393  	(MOVHstore [4] dst (MOVHload [4] src mem)
   394  		(MOVHstore [2] dst (MOVHload [2] src mem)
   395  			(MOVHstore dst (MOVHload src mem) mem)))
   396  (Move [s] dst src mem) && SizeAndAlign(s).Size() == 12 && SizeAndAlign(s).Align()%4 == 0 ->
   397  	(MOVWstore [8] dst (MOVWload [8] src mem)
   398  		(MOVWstore [4] dst (MOVWload [4] src mem)
   399  			(MOVWstore dst (MOVWload src mem) mem)))
   400  (Move [s] dst src mem) && SizeAndAlign(s).Size() == 16 && SizeAndAlign(s).Align()%8 == 0 ->
   401  	(MOVVstore [8] dst (MOVVload [8] src mem)
   402  		(MOVVstore dst (MOVVload src mem) mem))
   403  (Move [s] dst src mem) && SizeAndAlign(s).Size() == 24 && SizeAndAlign(s).Align()%8 == 0 ->
   404  	(MOVVstore [16] dst (MOVVload [16] src mem)
   405  		(MOVVstore [8] dst (MOVVload [8] src mem)
   406  			(MOVVstore dst (MOVVload src mem) mem)))
   407  
   408  // large or unaligned move uses a loop
   409  (Move [s] dst src mem)
   410  	&& SizeAndAlign(s).Size() > 24 || SizeAndAlign(s).Align()%8 != 0 ->
   411  	(LoweredMove [SizeAndAlign(s).Align()]
   412  		dst
   413  		src
   414  		(ADDVconst <src.Type> src [SizeAndAlign(s).Size()-moveSize(SizeAndAlign(s).Align(), config)])
   415  		mem)
   416  
   417  // calls
   418  (StaticCall [argwid] {target} mem) -> (CALLstatic [argwid] {target} mem)
   419  (ClosureCall [argwid] entry closure mem) -> (CALLclosure [argwid] entry closure mem)
   420  (DeferCall [argwid] mem) -> (CALLdefer [argwid] mem)
   421  (GoCall [argwid] mem) -> (CALLgo [argwid] mem)
   422  (InterCall [argwid] entry mem) -> (CALLinter [argwid] entry mem)
   423  
   424  // checks
   425  (NilCheck ptr mem) -> (LoweredNilCheck ptr mem)
   426  (IsNonNil ptr) -> (SGTU ptr (MOVVconst [0]))
   427  (IsInBounds idx len) -> (SGTU len idx)
   428  (IsSliceInBounds idx len) -> (XOR (MOVVconst [1]) (SGTU idx len))
   429  
   430  // pseudo-ops
   431  (GetClosurePtr) -> (LoweredGetClosurePtr)
   432  (Convert x mem) -> (MOVVconvert x mem)
   433  
   434  (If cond yes no) -> (NE cond yes no)
   435  
   436  // Optimizations
   437  
   438  // Absorb boolean tests into block
   439  (NE (FPFlagTrue cmp) yes no) -> (FPT cmp yes no)
   440  (NE (FPFlagFalse cmp) yes no) -> (FPF cmp yes no)
   441  (EQ (FPFlagTrue cmp) yes no) -> (FPF cmp yes no)
   442  (EQ (FPFlagFalse cmp) yes no) -> (FPT cmp yes no)
   443  (NE (XORconst [1] cmp:(SGT _ _)) yes no) -> (EQ cmp yes no)
   444  (NE (XORconst [1] cmp:(SGTU _ _)) yes no) -> (EQ cmp yes no)
   445  (NE (XORconst [1] cmp:(SGTconst _)) yes no) -> (EQ cmp yes no)
   446  (NE (XORconst [1] cmp:(SGTUconst _)) yes no) -> (EQ cmp yes no)
   447  (EQ (XORconst [1] cmp:(SGT _ _)) yes no) -> (NE cmp yes no)
   448  (EQ (XORconst [1] cmp:(SGTU _ _)) yes no) -> (NE cmp yes no)
   449  (EQ (XORconst [1] cmp:(SGTconst _)) yes no) -> (NE cmp yes no)
   450  (EQ (XORconst [1] cmp:(SGTUconst _)) yes no) -> (NE cmp yes no)
   451  (NE (SGTUconst [1] x) yes no) -> (EQ x yes no)
   452  (EQ (SGTUconst [1] x) yes no) -> (NE x yes no)
   453  (NE (SGTU x (MOVVconst [0])) yes no) -> (NE x yes no)
   454  (EQ (SGTU x (MOVVconst [0])) yes no) -> (EQ x yes no)
   455  (NE (SGTconst [0] x) yes no) -> (LTZ x yes no)
   456  (EQ (SGTconst [0] x) yes no) -> (GEZ x yes no)
   457  (NE (SGT x (MOVVconst [0])) yes no) -> (GTZ x yes no)
   458  (EQ (SGT x (MOVVconst [0])) yes no) -> (LEZ x yes no)
   459  
   460  // fold offset into address
   461  (ADDVconst [off1] (MOVVaddr [off2] {sym} ptr)) -> (MOVVaddr [off1+off2] {sym} ptr)
   462  
   463  // fold address into load/store
   464  (MOVBload  [off1] {sym} (ADDVconst [off2] ptr) mem) && is32Bit(off1+off2) -> (MOVBload  [off1+off2] {sym} ptr mem)
   465  (MOVBUload [off1] {sym} (ADDVconst [off2] ptr) mem) && is32Bit(off1+off2) -> (MOVBUload [off1+off2] {sym} ptr mem)
   466  (MOVHload  [off1] {sym} (ADDVconst [off2] ptr) mem) && is32Bit(off1+off2) -> (MOVHload  [off1+off2] {sym} ptr mem)
   467  (MOVHUload [off1] {sym} (ADDVconst [off2] ptr) mem) && is32Bit(off1+off2) -> (MOVHUload [off1+off2] {sym} ptr mem)
   468  (MOVWload  [off1] {sym} (ADDVconst [off2] ptr) mem) && is32Bit(off1+off2) -> (MOVWload  [off1+off2] {sym} ptr mem)
   469  (MOVWUload [off1] {sym} (ADDVconst [off2] ptr) mem) && is32Bit(off1+off2) -> (MOVWUload [off1+off2] {sym} ptr mem)
   470  (MOVVload  [off1] {sym} (ADDVconst [off2] ptr) mem) && is32Bit(off1+off2) -> (MOVVload  [off1+off2] {sym} ptr mem)
   471  (MOVFload  [off1] {sym} (ADDVconst [off2] ptr) mem) && is32Bit(off1+off2) -> (MOVFload  [off1+off2] {sym} ptr mem)
   472  (MOVDload  [off1] {sym} (ADDVconst [off2] ptr) mem) && is32Bit(off1+off2) -> (MOVDload  [off1+off2] {sym} ptr mem)
   473  
   474  (MOVBstore [off1] {sym} (ADDVconst [off2] ptr) val mem) && is32Bit(off1+off2) -> (MOVBstore [off1+off2] {sym} ptr val mem)
   475  (MOVHstore [off1] {sym} (ADDVconst [off2] ptr) val mem) && is32Bit(off1+off2) -> (MOVHstore [off1+off2] {sym} ptr val mem)
   476  (MOVWstore [off1] {sym} (ADDVconst [off2] ptr) val mem) && is32Bit(off1+off2) -> (MOVWstore [off1+off2] {sym} ptr val mem)
   477  (MOVVstore [off1] {sym} (ADDVconst [off2] ptr) val mem) && is32Bit(off1+off2) -> (MOVVstore [off1+off2] {sym} ptr val mem)
   478  (MOVFstore [off1] {sym} (ADDVconst [off2] ptr) val mem) && is32Bit(off1+off2) -> (MOVFstore [off1+off2] {sym} ptr val mem)
   479  (MOVDstore [off1] {sym} (ADDVconst [off2] ptr) val mem) && is32Bit(off1+off2) -> (MOVDstore [off1+off2] {sym} ptr val mem)
   480  (MOVBstorezero [off1] {sym} (ADDVconst [off2] ptr) mem) && is32Bit(off1+off2) -> (MOVBstorezero [off1+off2] {sym} ptr mem)
   481  (MOVHstorezero [off1] {sym} (ADDVconst [off2] ptr) mem) && is32Bit(off1+off2) -> (MOVHstorezero [off1+off2] {sym} ptr mem)
   482  (MOVWstorezero [off1] {sym} (ADDVconst [off2] ptr) mem) && is32Bit(off1+off2) -> (MOVWstorezero [off1+off2] {sym} ptr mem)
   483  (MOVVstorezero [off1] {sym} (ADDVconst [off2] ptr) mem) && is32Bit(off1+off2) -> (MOVVstorezero [off1+off2] {sym} ptr mem)
   484  
   485  (MOVBload [off1] {sym1} (MOVVaddr [off2] {sym2} ptr) mem) && canMergeSym(sym1,sym2) && is32Bit(off1+off2) ->
   486  	(MOVBload [off1+off2] {mergeSym(sym1,sym2)} ptr mem)
   487  (MOVBUload [off1] {sym1} (MOVVaddr [off2] {sym2} ptr) mem) && canMergeSym(sym1,sym2) && is32Bit(off1+off2) ->
   488  	(MOVBUload [off1+off2] {mergeSym(sym1,sym2)} ptr mem)
   489  (MOVHload [off1] {sym1} (MOVVaddr [off2] {sym2} ptr) mem) && canMergeSym(sym1,sym2) && is32Bit(off1+off2) ->
   490  	(MOVHload [off1+off2] {mergeSym(sym1,sym2)} ptr mem)
   491  (MOVHUload [off1] {sym1} (MOVVaddr [off2] {sym2} ptr) mem) && canMergeSym(sym1,sym2) && is32Bit(off1+off2) ->
   492  	(MOVHUload [off1+off2] {mergeSym(sym1,sym2)} ptr mem)
   493  (MOVWload [off1] {sym1} (MOVVaddr [off2] {sym2} ptr) mem) && canMergeSym(sym1,sym2) && is32Bit(off1+off2) ->
   494  	(MOVWload [off1+off2] {mergeSym(sym1,sym2)} ptr mem)
   495  (MOVWUload [off1] {sym1} (MOVVaddr [off2] {sym2} ptr) mem) && canMergeSym(sym1,sym2) && is32Bit(off1+off2) ->
   496  	(MOVWUload [off1+off2] {mergeSym(sym1,sym2)} ptr mem)
   497  (MOVVload [off1] {sym1} (MOVVaddr [off2] {sym2} ptr) mem) && canMergeSym(sym1,sym2) && is32Bit(off1+off2) ->
   498  	(MOVVload [off1+off2] {mergeSym(sym1,sym2)} ptr mem)
   499  (MOVFload [off1] {sym1} (MOVVaddr [off2] {sym2} ptr) mem) && canMergeSym(sym1,sym2) && is32Bit(off1+off2) ->
   500  	(MOVFload [off1+off2] {mergeSym(sym1,sym2)} ptr mem)
   501  (MOVDload [off1] {sym1} (MOVVaddr [off2] {sym2} ptr) mem) && canMergeSym(sym1,sym2) && is32Bit(off1+off2) ->
   502  	(MOVDload [off1+off2] {mergeSym(sym1,sym2)} ptr mem)
   503  
   504  (MOVBstore [off1] {sym1} (MOVVaddr [off2] {sym2} ptr) val mem) && canMergeSym(sym1,sym2) && is32Bit(off1+off2) ->
   505  	(MOVBstore [off1+off2] {mergeSym(sym1,sym2)} ptr val mem)
   506  (MOVHstore [off1] {sym1} (MOVVaddr [off2] {sym2} ptr) val mem) && canMergeSym(sym1,sym2) && is32Bit(off1+off2) ->
   507  	(MOVHstore [off1+off2] {mergeSym(sym1,sym2)} ptr val mem)
   508  (MOVWstore [off1] {sym1} (MOVVaddr [off2] {sym2} ptr) val mem) && canMergeSym(sym1,sym2) && is32Bit(off1+off2) ->
   509  	(MOVWstore [off1+off2] {mergeSym(sym1,sym2)} ptr val mem)
   510  (MOVVstore [off1] {sym1} (MOVVaddr [off2] {sym2} ptr) val mem) && canMergeSym(sym1,sym2) && is32Bit(off1+off2) ->
   511  	(MOVVstore [off1+off2] {mergeSym(sym1,sym2)} ptr val mem)
   512  (MOVFstore [off1] {sym1} (MOVVaddr [off2] {sym2} ptr) val mem) && canMergeSym(sym1,sym2) && is32Bit(off1+off2) ->
   513  	(MOVFstore [off1+off2] {mergeSym(sym1,sym2)} ptr val mem)
   514  (MOVDstore [off1] {sym1} (MOVVaddr [off2] {sym2} ptr) val mem) && canMergeSym(sym1,sym2) && is32Bit(off1+off2) ->
   515  	(MOVDstore [off1+off2] {mergeSym(sym1,sym2)} ptr val mem)
   516  (MOVBstorezero [off1] {sym1} (MOVVaddr [off2] {sym2} ptr) mem) && canMergeSym(sym1,sym2) && is32Bit(off1+off2) ->
   517  	(MOVBstorezero [off1+off2] {mergeSym(sym1,sym2)} ptr mem)
   518  (MOVHstorezero [off1] {sym1} (MOVVaddr [off2] {sym2} ptr) mem) && canMergeSym(sym1,sym2) && is32Bit(off1+off2) ->
   519  	(MOVHstorezero [off1+off2] {mergeSym(sym1,sym2)} ptr mem)
   520  (MOVWstorezero [off1] {sym1} (MOVVaddr [off2] {sym2} ptr) mem) && canMergeSym(sym1,sym2) && is32Bit(off1+off2) ->
   521  	(MOVWstorezero [off1+off2] {mergeSym(sym1,sym2)} ptr mem)
   522  (MOVVstorezero [off1] {sym1} (MOVVaddr [off2] {sym2} ptr) mem) && canMergeSym(sym1,sym2) && is32Bit(off1+off2) ->
   523  	(MOVVstorezero [off1+off2] {mergeSym(sym1,sym2)} ptr mem)
   524  
   525  // store zero
   526  (MOVBstore [off] {sym} ptr (MOVVconst [0]) mem) -> (MOVBstorezero [off] {sym} ptr mem)
   527  (MOVHstore [off] {sym} ptr (MOVVconst [0]) mem) -> (MOVHstorezero [off] {sym} ptr mem)
   528  (MOVWstore [off] {sym} ptr (MOVVconst [0]) mem) -> (MOVWstorezero [off] {sym} ptr mem)
   529  (MOVVstore [off] {sym} ptr (MOVVconst [0]) mem) -> (MOVVstorezero [off] {sym} ptr mem)
   530  
   531  // don't extend after proper load
   532  (MOVBreg x:(MOVBload _ _)) -> (MOVVreg x)
   533  (MOVBUreg x:(MOVBUload _ _)) -> (MOVVreg x)
   534  (MOVHreg x:(MOVBload _ _)) -> (MOVVreg x)
   535  (MOVHreg x:(MOVBUload _ _)) -> (MOVVreg x)
   536  (MOVHreg x:(MOVHload _ _)) -> (MOVVreg x)
   537  (MOVHUreg x:(MOVBUload _ _)) -> (MOVVreg x)
   538  (MOVHUreg x:(MOVHUload _ _)) -> (MOVVreg x)
   539  (MOVWreg x:(MOVBload _ _)) -> (MOVVreg x)
   540  (MOVWreg x:(MOVBUload _ _)) -> (MOVVreg x)
   541  (MOVWreg x:(MOVHload _ _)) -> (MOVVreg x)
   542  (MOVWreg x:(MOVHUload _ _)) -> (MOVVreg x)
   543  (MOVWreg x:(MOVWload _ _)) -> (MOVVreg x)
   544  (MOVWUreg x:(MOVBUload _ _)) -> (MOVVreg x)
   545  (MOVWUreg x:(MOVHUload _ _)) -> (MOVVreg x)
   546  (MOVWUreg x:(MOVWUload _ _)) -> (MOVVreg x)
   547  
   548  // fold double extensions
   549  (MOVBreg x:(MOVBreg _)) -> (MOVVreg x)
   550  (MOVBUreg x:(MOVBUreg _)) -> (MOVVreg x)
   551  (MOVHreg x:(MOVBreg _)) -> (MOVVreg x)
   552  (MOVHreg x:(MOVBUreg _)) -> (MOVVreg x)
   553  (MOVHreg x:(MOVHreg _)) -> (MOVVreg x)
   554  (MOVHUreg x:(MOVBUreg _)) -> (MOVVreg x)
   555  (MOVHUreg x:(MOVHUreg _)) -> (MOVVreg x)
   556  (MOVWreg x:(MOVBreg _)) -> (MOVVreg x)
   557  (MOVWreg x:(MOVBUreg _)) -> (MOVVreg x)
   558  (MOVWreg x:(MOVHreg _)) -> (MOVVreg x)
   559  (MOVWreg x:(MOVHreg _)) -> (MOVVreg x)
   560  (MOVWreg x:(MOVWreg _)) -> (MOVVreg x)
   561  (MOVWUreg x:(MOVBUreg _)) -> (MOVVreg x)
   562  (MOVWUreg x:(MOVHUreg _)) -> (MOVVreg x)
   563  (MOVWUreg x:(MOVWUreg _)) -> (MOVVreg x)
   564  
   565  // don't extend before store
   566  (MOVBstore [off] {sym} ptr (MOVBreg x) mem) -> (MOVBstore [off] {sym} ptr x mem)
   567  (MOVBstore [off] {sym} ptr (MOVBUreg x) mem) -> (MOVBstore [off] {sym} ptr x mem)
   568  (MOVBstore [off] {sym} ptr (MOVHreg x) mem) -> (MOVBstore [off] {sym} ptr x mem)
   569  (MOVBstore [off] {sym} ptr (MOVHUreg x) mem) -> (MOVBstore [off] {sym} ptr x mem)
   570  (MOVBstore [off] {sym} ptr (MOVWreg x) mem) -> (MOVBstore [off] {sym} ptr x mem)
   571  (MOVBstore [off] {sym} ptr (MOVWUreg x) mem) -> (MOVBstore [off] {sym} ptr x mem)
   572  (MOVHstore [off] {sym} ptr (MOVHreg x) mem) -> (MOVHstore [off] {sym} ptr x mem)
   573  (MOVHstore [off] {sym} ptr (MOVHUreg x) mem) -> (MOVHstore [off] {sym} ptr x mem)
   574  (MOVHstore [off] {sym} ptr (MOVWreg x) mem) -> (MOVHstore [off] {sym} ptr x mem)
   575  (MOVHstore [off] {sym} ptr (MOVWUreg x) mem) -> (MOVHstore [off] {sym} ptr x mem)
   576  (MOVWstore [off] {sym} ptr (MOVWreg x) mem) -> (MOVWstore [off] {sym} ptr x mem)
   577  (MOVWstore [off] {sym} ptr (MOVWUreg x) mem) -> (MOVWstore [off] {sym} ptr x mem)
   578  
   579  // if a register move has only 1 use, just use the same register without emitting instruction
   580  // MOVVnop doesn't emit instruction, only for ensuring the type.
   581  (MOVVreg x) && x.Uses == 1 -> (MOVVnop x)
   582  
   583  // fold constant into arithmatic ops
   584  (ADDV (MOVVconst [c]) x) && is32Bit(c) -> (ADDVconst [c] x)
   585  (ADDV x (MOVVconst [c])) && is32Bit(c) -> (ADDVconst [c] x)
   586  (SUBV x (MOVVconst [c])) && is32Bit(c) -> (SUBVconst [c] x)
   587  (AND (MOVVconst [c]) x) && is32Bit(c) -> (ANDconst [c] x)
   588  (AND x (MOVVconst [c])) && is32Bit(c) -> (ANDconst [c] x)
   589  (OR  (MOVVconst [c]) x) && is32Bit(c) -> (ORconst  [c] x)
   590  (OR  x (MOVVconst [c])) && is32Bit(c) -> (ORconst  [c] x)
   591  (XOR (MOVVconst [c]) x) && is32Bit(c) -> (XORconst [c] x)
   592  (XOR x (MOVVconst [c])) && is32Bit(c) -> (XORconst [c] x)
   593  (NOR (MOVVconst [c]) x) && is32Bit(c) -> (NORconst [c] x)
   594  (NOR x (MOVVconst [c])) && is32Bit(c) -> (NORconst [c] x)
   595  
   596  (SLLV _ (MOVVconst [c])) && uint64(c)>=64 -> (MOVVconst [0])
   597  (SRLV _ (MOVVconst [c])) && uint64(c)>=64 -> (MOVVconst [0])
   598  (SRAV x (MOVVconst [c])) && uint64(c)>=64 -> (SRAVconst x [63])
   599  (SLLV x (MOVVconst [c])) -> (SLLVconst x [c])
   600  (SRLV x (MOVVconst [c])) -> (SRLVconst x [c])
   601  (SRAV x (MOVVconst [c])) -> (SRAVconst x [c])
   602  
   603  (SGT  (MOVVconst [c]) x) && is32Bit(c) -> (SGTconst  [c] x)
   604  (SGTU (MOVVconst [c]) x) && is32Bit(c) -> (SGTUconst [c] x)
   605  
   606  // mul by constant
   607  (Select1 (MULVU x (MOVVconst [-1]))) -> (NEGV x)
   608  (Select1 (MULVU _ (MOVVconst [0]))) -> (MOVVconst [0])
   609  (Select1 (MULVU x (MOVVconst [1]))) -> x
   610  (Select1 (MULVU x (MOVVconst [c]))) && isPowerOfTwo(c) -> (SLLVconst [log2(c)] x)
   611  
   612  (Select1 (MULVU (MOVVconst [-1]) x)) -> (NEGV x)
   613  (Select1 (MULVU (MOVVconst [0]) _)) -> (MOVVconst [0])
   614  (Select1 (MULVU (MOVVconst [1]) x)) -> x
   615  (Select1 (MULVU (MOVVconst [c]) x)) && isPowerOfTwo(c) -> (SLLVconst [log2(c)] x)
   616  
   617  // div by constant
   618  (Select1 (DIVVU x (MOVVconst [1]))) -> x
   619  (Select1 (DIVVU x (MOVVconst [c]))) && isPowerOfTwo(c) -> (SRLVconst [log2(c)] x)
   620  (Select0 (DIVVU _ (MOVVconst [1]))) -> (MOVVconst [0])                       // mod
   621  (Select0 (DIVVU x (MOVVconst [c]))) && isPowerOfTwo(c) -> (ANDconst [c-1] x) // mod
   622  
   623  // generic simplifications
   624  (ADDV x (NEGV y)) -> (SUBV x y)
   625  (ADDV (NEGV y) x) -> (SUBV x y)
   626  (SUBV x x) -> (MOVVconst [0])
   627  (SUBV (MOVVconst [0]) x) -> (NEGV x)
   628  (AND x x) -> x
   629  (OR  x x) -> x
   630  (XOR x x) -> (MOVVconst [0])
   631  
   632  // remove redundant *const ops
   633  (ADDVconst [0]  x) -> x
   634  (SUBVconst [0]  x) -> x
   635  (ANDconst [0]  _) -> (MOVVconst [0])
   636  (ANDconst [-1] x) -> x
   637  (ORconst  [0]  x) -> x
   638  (ORconst  [-1] _) -> (MOVVconst [-1])
   639  (XORconst [0]  x) -> x
   640  (XORconst [-1] x) -> (NORconst [0] x)
   641  
   642  // generic constant folding
   643  (ADDVconst [c] (MOVVconst [d]))  -> (MOVVconst [c+d])
   644  (ADDVconst [c] (ADDVconst [d] x)) && is32Bit(c+d) -> (ADDVconst [c+d] x)
   645  (ADDVconst [c] (SUBVconst [d] x)) && is32Bit(c-d) -> (ADDVconst [c-d] x)
   646  (SUBVconst [c] (MOVVconst [d]))  -> (MOVVconst [d-c])
   647  (SUBVconst [c] (SUBVconst [d] x)) && is32Bit(-c-d) -> (ADDVconst [-c-d] x)
   648  (SUBVconst [c] (ADDVconst [d] x)) && is32Bit(-c+d) -> (ADDVconst [-c+d] x)
   649  (SLLVconst [c] (MOVVconst [d]))  -> (MOVVconst [int64(d)<<uint64(c)])
   650  (SRLVconst [c] (MOVVconst [d]))  -> (MOVVconst [int64(uint64(d)>>uint64(c))])
   651  (SRAVconst [c] (MOVVconst [d]))  -> (MOVVconst [int64(d)>>uint64(c)])
   652  (Select1 (MULVU (MOVVconst [c]) (MOVVconst [d]))) -> (MOVVconst [c*d])
   653  (Select1 (DIVV  (MOVVconst [c]) (MOVVconst [d]))) -> (MOVVconst [int64(c)/int64(d)])
   654  (Select1 (DIVVU (MOVVconst [c]) (MOVVconst [d]))) -> (MOVVconst [int64(uint64(c)/uint64(d))])
   655  (Select0 (DIVV  (MOVVconst [c]) (MOVVconst [d]))) -> (MOVVconst [int64(c)%int64(d)])   // mod
   656  (Select0 (DIVVU (MOVVconst [c]) (MOVVconst [d]))) -> (MOVVconst [int64(uint64(c)%uint64(d))]) // mod
   657  (ANDconst [c] (MOVVconst [d])) -> (MOVVconst [c&d])
   658  (ANDconst [c] (ANDconst [d] x)) -> (ANDconst [c&d] x)
   659  (ORconst [c] (MOVVconst [d])) -> (MOVVconst [c|d])
   660  (ORconst [c] (ORconst [d] x)) && is32Bit(c|d) -> (ORconst [c|d] x)
   661  (XORconst [c] (MOVVconst [d])) -> (MOVVconst [c^d])
   662  (XORconst [c] (XORconst [d] x)) && is32Bit(c^d) -> (XORconst [c^d] x)
   663  (NORconst [c] (MOVVconst [d])) -> (MOVVconst [^(c|d)])
   664  (NEGV (MOVVconst [c])) -> (MOVVconst [-c])
   665  (MOVBreg  (MOVVconst [c])) -> (MOVVconst [int64(int8(c))])
   666  (MOVBUreg (MOVVconst [c])) -> (MOVVconst [int64(uint8(c))])
   667  (MOVHreg  (MOVVconst [c])) -> (MOVVconst [int64(int16(c))])
   668  (MOVHUreg (MOVVconst [c])) -> (MOVVconst [int64(uint16(c))])
   669  (MOVWreg  (MOVVconst [c])) -> (MOVVconst [int64(int32(c))])
   670  (MOVWUreg (MOVVconst [c])) -> (MOVVconst [int64(uint32(c))])
   671  (MOVVreg  (MOVVconst [c])) -> (MOVVconst [c])
   672  
   673  // constant comparisons
   674  (SGTconst [c] (MOVVconst [d])) && int64(c)>int64(d) -> (MOVVconst [1])
   675  (SGTconst [c] (MOVVconst [d])) && int64(c)<=int64(d) -> (MOVVconst [0])
   676  (SGTUconst [c] (MOVVconst [d])) && uint64(c)>uint64(d) -> (MOVVconst [1])
   677  (SGTUconst [c] (MOVVconst [d])) && uint64(c)<=uint64(d) -> (MOVVconst [0])
   678  
   679  // other known comparisons
   680  (SGTconst [c] (MOVBreg _)) && 0x7f < int64(c) -> (MOVVconst [1])
   681  (SGTconst [c] (MOVBreg _)) && int64(c) <= -0x80 -> (MOVVconst [0])
   682  (SGTconst [c] (MOVBUreg _)) && 0xff < int64(c) -> (MOVVconst [1])
   683  (SGTconst [c] (MOVBUreg _)) && int64(c) < 0 -> (MOVVconst [0])
   684  (SGTUconst [c] (MOVBUreg _)) && 0xff < uint64(c) -> (MOVVconst [1])
   685  (SGTconst [c] (MOVHreg _)) && 0x7fff < int64(c) -> (MOVVconst [1])
   686  (SGTconst [c] (MOVHreg _)) && int64(c) <= -0x8000 -> (MOVVconst [0])
   687  (SGTconst [c] (MOVHUreg _)) && 0xffff < int64(c) -> (MOVVconst [1])
   688  (SGTconst [c] (MOVHUreg _)) && int64(c) < 0 -> (MOVVconst [0])
   689  (SGTUconst [c] (MOVHUreg _)) && 0xffff < uint64(c) -> (MOVVconst [1])
   690  (SGTconst [c] (MOVWUreg _)) && int64(c) < 0 -> (MOVVconst [0])
   691  (SGTconst [c] (ANDconst [m] _)) && 0 <= m && m < c -> (MOVVconst [1])
   692  (SGTUconst [c] (ANDconst [m] _)) && uint64(m) < uint64(c) -> (MOVVconst [1])
   693  (SGTconst [c] (SRLVconst _ [d])) && 0 <= c && 0 < d && d <= 63 && 1<<uint64(64-d) <= c -> (MOVVconst [1])
   694  (SGTUconst [c] (SRLVconst _ [d])) && 0 < d && d <= 63 && 1<<uint64(64-d) <= uint64(c) -> (MOVVconst [1])
   695  
   696  // absorb constants into branches
   697  (EQ  (MOVVconst [0]) yes no) -> (First nil yes no)
   698  (EQ  (MOVVconst [c]) yes no) && c != 0 -> (First nil no yes)
   699  (NE  (MOVVconst [0]) yes no) -> (First nil no yes)
   700  (NE  (MOVVconst [c]) yes no) && c != 0 -> (First nil yes no)
   701  (LTZ (MOVVconst [c]) yes no) && c <  0 -> (First nil yes no)
   702  (LTZ (MOVVconst [c]) yes no) && c >= 0 -> (First nil no yes)
   703  (LEZ (MOVVconst [c]) yes no) && c <= 0 -> (First nil yes no)
   704  (LEZ (MOVVconst [c]) yes no) && c >  0 -> (First nil no yes)
   705  (GTZ (MOVVconst [c]) yes no) && c >  0 -> (First nil yes no)
   706  (GTZ (MOVVconst [c]) yes no) && c <= 0 -> (First nil no yes)
   707  (GEZ (MOVVconst [c]) yes no) && c >= 0 -> (First nil yes no)
   708  (GEZ (MOVVconst [c]) yes no) && c <  0 -> (First nil no yes)