github.com/go-asm/go@v1.21.1-0.20240213172139-40c5ead50c48/cmd/compile/ssa/_gen/LOONG64.rules (about)

     1  // Copyright 2022 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) ...) => (ADDV ...)
     6  (Add(32|64)F ...) => (ADD(F|D) ...)
     7  
     8  (Sub(Ptr|64|32|16|8) ...) => (SUBV ...)
     9  (Sub(32|64)F ...) => (SUB(F|D) ...)
    10  
    11  (Mul(64|32|16|8) ...) => (MULV ...)
    12  (Mul(32|64)F ...) => (MUL(F|D) ...)
    13  (Select0 (Mul64uhilo x y)) => (MULHVU x y)
    14  (Select1 (Mul64uhilo x y)) => (MULV x y)
    15  (Select0 (Mul64uover x y)) => (MULV x y)
    16  (Select1 (Mul64uover x y)) => (SGTU <typ.Bool> (MULHVU x y) (MOVVconst <typ.UInt64> [0]))
    17  
    18  (Hmul64 ...)  => (MULHV  ...)
    19  (Hmul64u ...) => (MULHVU ...)
    20  (Hmul32 x y)  => (SRAVconst (MULV (SignExt32to64 x) (SignExt32to64 y)) [32])
    21  (Hmul32u x y) => (SRLVconst (MULV (ZeroExt32to64 x) (ZeroExt32to64 y)) [32])
    22  
    23  (Div64 x y) => (DIVV x y)
    24  (Div64u ...) => (DIVVU ...)
    25  (Div32 x y) => (DIVV (SignExt32to64 x) (SignExt32to64 y))
    26  (Div32u x y) => (DIVVU (ZeroExt32to64 x) (ZeroExt32to64 y))
    27  (Div16 x y) => (DIVV (SignExt16to64 x) (SignExt16to64 y))
    28  (Div16u x y) => (DIVVU (ZeroExt16to64 x) (ZeroExt16to64 y))
    29  (Div8 x y) => (DIVV (SignExt8to64 x) (SignExt8to64 y))
    30  (Div8u x y) => (DIVVU (ZeroExt8to64 x) (ZeroExt8to64 y))
    31  (Div(32|64)F ...) => (DIV(F|D) ...)
    32  
    33  (Mod64 x y) => (REMV x y)
    34  (Mod64u ...) => (REMVU ...)
    35  (Mod32 x y) => (REMV (SignExt32to64 x) (SignExt32to64 y))
    36  (Mod32u x y) => (REMVU (ZeroExt32to64 x) (ZeroExt32to64 y))
    37  (Mod16 x y) => (REMV (SignExt16to64 x) (SignExt16to64 y))
    38  (Mod16u x y) => (REMVU (ZeroExt16to64 x) (ZeroExt16to64 y))
    39  (Mod8 x y) => (REMV (SignExt8to64 x) (SignExt8to64 y))
    40  (Mod8u x y) => (REMVU (ZeroExt8to64 x) (ZeroExt8to64 y))
    41  
    42  (Select0 <t> (Add64carry x y c)) => (ADDV (ADDV <t> x y) c)
    43  (Select1 <t> (Add64carry x y c)) =>
    44  	(OR (SGTU <t> x s:(ADDV <t> x y)) (SGTU <t> s (ADDV <t> s c)))
    45  
    46  (Select0 <t> (Sub64borrow x y c)) => (SUBV (SUBV <t> x y) c)
    47  (Select1 <t> (Sub64borrow x y c)) =>
    48  	(OR (SGTU <t> s:(SUBV <t> x y) x) (SGTU <t> (SUBV <t> s c) s))
    49  
    50  // (x + y) / 2 with x>=y => (x - y) / 2 + y
    51  (Avg64u <t> x y) => (ADDV (SRLVconst <t> (SUBV <t> x y) [1]) y)
    52  
    53  (And(64|32|16|8) ...) => (AND ...)
    54  (Or(64|32|16|8) ...) => (OR ...)
    55  (Xor(64|32|16|8) ...) => (XOR ...)
    56  
    57  // shifts
    58  // hardware instruction uses only the low 6 bits of the shift
    59  // we compare to 64 to ensure Go semantics for large shifts
    60  (Lsh64x64 <t> x y) => (MASKEQZ (SLLV <t> x                y)  (SGTU (MOVVconst <typ.UInt64> [64])                y))
    61  (Lsh64x32 <t> x y) => (MASKEQZ (SLLV <t> x (ZeroExt32to64 y)) (SGTU (MOVVconst <typ.UInt64> [64]) (ZeroExt32to64 y)))
    62  (Lsh64x16 <t> x y) => (MASKEQZ (SLLV <t> x (ZeroExt16to64 y)) (SGTU (MOVVconst <typ.UInt64> [64]) (ZeroExt16to64 y)))
    63  (Lsh64x8  <t> x y) => (MASKEQZ (SLLV <t> x (ZeroExt8to64  y)) (SGTU (MOVVconst <typ.UInt64> [64]) (ZeroExt8to64  y)))
    64  
    65  (Lsh32x64 <t> x y) => (MASKEQZ (SLLV <t> x                y)  (SGTU (MOVVconst <typ.UInt64> [64])                y))
    66  (Lsh32x32 <t> x y) => (MASKEQZ (SLLV <t> x (ZeroExt32to64 y)) (SGTU (MOVVconst <typ.UInt64> [64]) (ZeroExt32to64 y)))
    67  (Lsh32x16 <t> x y) => (MASKEQZ (SLLV <t> x (ZeroExt16to64 y)) (SGTU (MOVVconst <typ.UInt64> [64]) (ZeroExt16to64 y)))
    68  (Lsh32x8  <t> x y) => (MASKEQZ (SLLV <t> x (ZeroExt8to64  y)) (SGTU (MOVVconst <typ.UInt64> [64]) (ZeroExt8to64  y)))
    69  
    70  (Lsh16x64 <t> x y) => (MASKEQZ (SLLV <t> x                y)  (SGTU (MOVVconst <typ.UInt64> [64])                y))
    71  (Lsh16x32 <t> x y) => (MASKEQZ (SLLV <t> x (ZeroExt32to64 y)) (SGTU (MOVVconst <typ.UInt64> [64]) (ZeroExt32to64 y)))
    72  (Lsh16x16 <t> x y) => (MASKEQZ (SLLV <t> x (ZeroExt16to64 y)) (SGTU (MOVVconst <typ.UInt64> [64]) (ZeroExt16to64 y)))
    73  (Lsh16x8  <t> x y) => (MASKEQZ (SLLV <t> x (ZeroExt8to64  y)) (SGTU (MOVVconst <typ.UInt64> [64]) (ZeroExt8to64  y)))
    74  
    75  (Lsh8x64 <t> x y) => (MASKEQZ (SLLV <t> x                y)  (SGTU (MOVVconst <typ.UInt64> [64])                y))
    76  (Lsh8x32 <t> x y) => (MASKEQZ (SLLV <t> x (ZeroExt32to64 y)) (SGTU (MOVVconst <typ.UInt64> [64]) (ZeroExt32to64 y)))
    77  (Lsh8x16 <t> x y) => (MASKEQZ (SLLV <t> x (ZeroExt16to64 y)) (SGTU (MOVVconst <typ.UInt64> [64]) (ZeroExt16to64 y)))
    78  (Lsh8x8  <t> x y) => (MASKEQZ (SLLV <t> x (ZeroExt8to64  y)) (SGTU (MOVVconst <typ.UInt64> [64]) (ZeroExt8to64  y)))
    79  
    80  (Rsh64Ux64 <t> x y) => (MASKEQZ (SRLV <t> x                y)  (SGTU (MOVVconst <typ.UInt64> [64])                y))
    81  (Rsh64Ux32 <t> x y) => (MASKEQZ (SRLV <t> x (ZeroExt32to64 y)) (SGTU (MOVVconst <typ.UInt64> [64]) (ZeroExt32to64 y)))
    82  (Rsh64Ux16 <t> x y) => (MASKEQZ (SRLV <t> x (ZeroExt16to64 y)) (SGTU (MOVVconst <typ.UInt64> [64]) (ZeroExt16to64 y)))
    83  (Rsh64Ux8  <t> x y) => (MASKEQZ (SRLV <t> x (ZeroExt8to64  y)) (SGTU (MOVVconst <typ.UInt64> [64]) (ZeroExt8to64  y)))
    84  
    85  (Rsh32Ux64 <t> x y) => (MASKEQZ (SRLV <t> (ZeroExt32to64 x)                y)  (SGTU (MOVVconst <typ.UInt64> [64])                y))
    86  (Rsh32Ux32 <t> x y) => (MASKEQZ (SRLV <t> (ZeroExt32to64 x) (ZeroExt32to64 y)) (SGTU (MOVVconst <typ.UInt64> [64]) (ZeroExt32to64 y)))
    87  (Rsh32Ux16 <t> x y) => (MASKEQZ (SRLV <t> (ZeroExt32to64 x) (ZeroExt16to64 y)) (SGTU (MOVVconst <typ.UInt64> [64]) (ZeroExt16to64 y)))
    88  (Rsh32Ux8  <t> x y) => (MASKEQZ (SRLV <t> (ZeroExt32to64 x) (ZeroExt8to64  y)) (SGTU (MOVVconst <typ.UInt64> [64]) (ZeroExt8to64  y)))
    89  
    90  (Rsh16Ux64 <t> x y) => (MASKEQZ (SRLV <t> (ZeroExt16to64 x)                y)  (SGTU (MOVVconst <typ.UInt64> [64])                y))
    91  (Rsh16Ux32 <t> x y) => (MASKEQZ (SRLV <t> (ZeroExt16to64 x) (ZeroExt32to64 y)) (SGTU (MOVVconst <typ.UInt64> [64]) (ZeroExt32to64 y)))
    92  (Rsh16Ux16 <t> x y) => (MASKEQZ (SRLV <t> (ZeroExt16to64 x) (ZeroExt16to64 y)) (SGTU (MOVVconst <typ.UInt64> [64]) (ZeroExt16to64 y)))
    93  (Rsh16Ux8  <t> x y) => (MASKEQZ (SRLV <t> (ZeroExt16to64 x) (ZeroExt8to64  y)) (SGTU (MOVVconst <typ.UInt64> [64]) (ZeroExt8to64  y)))
    94  
    95  (Rsh8Ux64 <t> x y) => (MASKEQZ (SRLV <t> (ZeroExt8to64 x)                y)  (SGTU (MOVVconst <typ.UInt64> [64])                y))
    96  (Rsh8Ux32 <t> x y) => (MASKEQZ (SRLV <t> (ZeroExt8to64 x) (ZeroExt32to64 y)) (SGTU (MOVVconst <typ.UInt64> [64]) (ZeroExt32to64 y)))
    97  (Rsh8Ux16 <t> x y) => (MASKEQZ (SRLV <t> (ZeroExt8to64 x) (ZeroExt16to64 y)) (SGTU (MOVVconst <typ.UInt64> [64]) (ZeroExt16to64 y)))
    98  (Rsh8Ux8  <t> x y) => (MASKEQZ (SRLV <t> (ZeroExt8to64 x) (ZeroExt8to64  y)) (SGTU (MOVVconst <typ.UInt64> [64]) (ZeroExt8to64  y)))
    99  
   100  (Rsh64x64 <t> x y) => (SRAV x (OR <t> (NEGV <t> (SGTU y (MOVVconst <typ.UInt64> [63]))) y))
   101  (Rsh64x32 <t> x y) => (SRAV x (OR <t> (NEGV <t> (SGTU (ZeroExt32to64 y) (MOVVconst <typ.UInt64> [63]))) (ZeroExt32to64 y)))
   102  (Rsh64x16 <t> x y) => (SRAV x (OR <t> (NEGV <t> (SGTU (ZeroExt16to64 y) (MOVVconst <typ.UInt64> [63]))) (ZeroExt16to64 y)))
   103  (Rsh64x8  <t> x y) => (SRAV x (OR <t> (NEGV <t> (SGTU (ZeroExt8to64  y) (MOVVconst <typ.UInt64> [63]))) (ZeroExt8to64  y)))
   104  
   105  (Rsh32x64 <t> x y) => (SRAV (SignExt32to64 x) (OR <t> (NEGV <t> (SGTU y (MOVVconst <typ.UInt64> [63]))) y))
   106  (Rsh32x32 <t> x y) => (SRAV (SignExt32to64 x) (OR <t> (NEGV <t> (SGTU (ZeroExt32to64 y) (MOVVconst <typ.UInt64> [63]))) (ZeroExt32to64 y)))
   107  (Rsh32x16 <t> x y) => (SRAV (SignExt32to64 x) (OR <t> (NEGV <t> (SGTU (ZeroExt16to64 y) (MOVVconst <typ.UInt64> [63]))) (ZeroExt16to64 y)))
   108  (Rsh32x8  <t> x y) => (SRAV (SignExt32to64 x) (OR <t> (NEGV <t> (SGTU (ZeroExt8to64  y) (MOVVconst <typ.UInt64> [63]))) (ZeroExt8to64  y)))
   109  
   110  (Rsh16x64 <t> x y) => (SRAV (SignExt16to64 x) (OR <t> (NEGV <t> (SGTU y (MOVVconst <typ.UInt64> [63]))) y))
   111  (Rsh16x32 <t> x y) => (SRAV (SignExt16to64 x) (OR <t> (NEGV <t> (SGTU (ZeroExt32to64 y) (MOVVconst <typ.UInt64> [63]))) (ZeroExt32to64 y)))
   112  (Rsh16x16 <t> x y) => (SRAV (SignExt16to64 x) (OR <t> (NEGV <t> (SGTU (ZeroExt16to64 y) (MOVVconst <typ.UInt64> [63]))) (ZeroExt16to64 y)))
   113  (Rsh16x8  <t> x y) => (SRAV (SignExt16to64 x) (OR <t> (NEGV <t> (SGTU (ZeroExt8to64  y) (MOVVconst <typ.UInt64> [63]))) (ZeroExt8to64  y)))
   114  
   115  (Rsh8x64 <t> x y) => (SRAV (SignExt8to64 x) (OR <t> (NEGV <t> (SGTU y (MOVVconst <typ.UInt64> [63]))) y))
   116  (Rsh8x32 <t> x y) => (SRAV (SignExt8to64 x) (OR <t> (NEGV <t> (SGTU (ZeroExt32to64 y) (MOVVconst <typ.UInt64> [63]))) (ZeroExt32to64 y)))
   117  (Rsh8x16 <t> x y) => (SRAV (SignExt8to64 x) (OR <t> (NEGV <t> (SGTU (ZeroExt16to64 y) (MOVVconst <typ.UInt64> [63]))) (ZeroExt16to64 y)))
   118  (Rsh8x8  <t> x y) => (SRAV (SignExt8to64 x) (OR <t> (NEGV <t> (SGTU (ZeroExt8to64  y) (MOVVconst <typ.UInt64> [63]))) (ZeroExt8to64  y)))
   119  
   120  // rotates
   121  (RotateLeft8 <t> x (MOVVconst [c])) => (Or8 (Lsh8x64 <t> x (MOVVconst [c&7])) (Rsh8Ux64 <t> x (MOVVconst [-c&7])))
   122  (RotateLeft16 <t> x (MOVVconst [c])) => (Or16 (Lsh16x64 <t> x (MOVVconst [c&15])) (Rsh16Ux64 <t> x (MOVVconst [-c&15])))
   123  (RotateLeft32 x y) => (ROTR  x (NEGV <y.Type> y))
   124  (RotateLeft64 x y) => (ROTRV x (NEGV <y.Type> y))
   125  
   126  // unary ops
   127  (Neg(64|32|16|8) ...) => (NEGV ...)
   128  (Neg(32|64)F ...) => (NEG(F|D) ...)
   129  
   130  (Com(64|32|16|8) x) => (NOR (MOVVconst [0]) x)
   131  
   132  (Sqrt ...) => (SQRTD ...)
   133  (Sqrt32 ...) => (SQRTF ...)
   134  
   135  // boolean ops -- booleans are represented with 0=false, 1=true
   136  (AndB ...) => (AND ...)
   137  (OrB ...) => (OR ...)
   138  (EqB x y) => (XOR (MOVVconst [1]) (XOR <typ.Bool> x y))
   139  (NeqB ...) => (XOR ...)
   140  (Not x) => (XORconst [1] x)
   141  
   142  // constants
   143  (Const(64|32|16|8) [val]) => (MOVVconst [int64(val)])
   144  (Const(32|64)F [val]) => (MOV(F|D)const [float64(val)])
   145  (ConstNil) => (MOVVconst [0])
   146  (ConstBool [t]) => (MOVVconst [int64(b2i(t))])
   147  
   148  (Slicemask <t> x) => (SRAVconst (NEGV <t> x) [63])
   149  
   150  // truncations
   151  // Because we ignore high parts of registers, truncates are just copies.
   152  (Trunc16to8 ...) => (Copy ...)
   153  (Trunc32to8 ...) => (Copy ...)
   154  (Trunc32to16 ...) => (Copy ...)
   155  (Trunc64to8 ...) => (Copy ...)
   156  (Trunc64to16 ...) => (Copy ...)
   157  (Trunc64to32 ...) => (Copy ...)
   158  
   159  // Zero-/Sign-extensions
   160  (ZeroExt8to16 ...) => (MOVBUreg ...)
   161  (ZeroExt8to32 ...) => (MOVBUreg ...)
   162  (ZeroExt16to32 ...) => (MOVHUreg ...)
   163  (ZeroExt8to64 ...) => (MOVBUreg ...)
   164  (ZeroExt16to64 ...) => (MOVHUreg ...)
   165  (ZeroExt32to64 ...) => (MOVWUreg ...)
   166  
   167  (SignExt8to16 ...) => (MOVBreg ...)
   168  (SignExt8to32 ...) => (MOVBreg ...)
   169  (SignExt16to32 ...) => (MOVHreg ...)
   170  (SignExt8to64 ...) => (MOVBreg ...)
   171  (SignExt16to64 ...) => (MOVHreg ...)
   172  (SignExt32to64 ...) => (MOVWreg ...)
   173  
   174  // float <=> int conversion
   175  (Cvt32to32F ...) => (MOVWF ...)
   176  (Cvt32to64F ...) => (MOVWD ...)
   177  (Cvt64to32F ...) => (MOVVF ...)
   178  (Cvt64to64F ...) => (MOVVD ...)
   179  (Cvt32Fto32 ...) => (TRUNCFW ...)
   180  (Cvt64Fto32 ...) => (TRUNCDW ...)
   181  (Cvt32Fto64 ...) => (TRUNCFV ...)
   182  (Cvt64Fto64 ...) => (TRUNCDV ...)
   183  (Cvt32Fto64F ...) => (MOVFD ...)
   184  (Cvt64Fto32F ...) => (MOVDF ...)
   185  
   186  (CvtBoolToUint8 ...) => (Copy ...)
   187  
   188  (Round(32|64)F ...) => (Copy ...)
   189  
   190  // comparisons
   191  (Eq8 x y)  => (SGTU (MOVVconst [1]) (XOR (ZeroExt8to64 x) (ZeroExt8to64 y)))
   192  (Eq16 x y) => (SGTU (MOVVconst [1]) (XOR (ZeroExt16to64 x) (ZeroExt16to64 y)))
   193  (Eq32 x y) => (SGTU (MOVVconst [1]) (XOR (ZeroExt32to64 x) (ZeroExt32to64 y)))
   194  (Eq64 x y) => (SGTU (MOVVconst [1]) (XOR x y))
   195  (EqPtr x y) => (SGTU (MOVVconst [1]) (XOR x y))
   196  (Eq(32|64)F x y) => (FPFlagTrue (CMPEQ(F|D) x y))
   197  
   198  (Neq8 x y)  => (SGTU (XOR (ZeroExt8to64 x) (ZeroExt8to64 y)) (MOVVconst [0]))
   199  (Neq16 x y) => (SGTU (XOR (ZeroExt16to32 x) (ZeroExt16to64 y)) (MOVVconst [0]))
   200  (Neq32 x y) => (SGTU (XOR (ZeroExt32to64 x) (ZeroExt32to64 y)) (MOVVconst [0]))
   201  (Neq64 x y) => (SGTU (XOR x y) (MOVVconst [0]))
   202  (NeqPtr x y) => (SGTU (XOR x y) (MOVVconst [0]))
   203  (Neq(32|64)F x y) => (FPFlagFalse (CMPEQ(F|D) x y))
   204  
   205  (Less8 x y)  => (SGT (SignExt8to64 y) (SignExt8to64 x))
   206  (Less16 x y) => (SGT (SignExt16to64 y) (SignExt16to64 x))
   207  (Less32 x y) => (SGT (SignExt32to64 y) (SignExt32to64 x))
   208  (Less64 x y) => (SGT y x)
   209  (Less(32|64)F x y) => (FPFlagTrue (CMPGT(F|D) y x)) // reverse operands to work around NaN
   210  
   211  (Less8U x y)  => (SGTU (ZeroExt8to64 y) (ZeroExt8to64 x))
   212  (Less16U x y) => (SGTU (ZeroExt16to64 y) (ZeroExt16to64 x))
   213  (Less32U x y) => (SGTU (ZeroExt32to64 y) (ZeroExt32to64 x))
   214  (Less64U x y) => (SGTU y x)
   215  
   216  (Leq8 x y)  => (XOR (MOVVconst [1]) (SGT (SignExt8to64 x) (SignExt8to64 y)))
   217  (Leq16 x y) => (XOR (MOVVconst [1]) (SGT (SignExt16to64 x) (SignExt16to64 y)))
   218  (Leq32 x y) => (XOR (MOVVconst [1]) (SGT (SignExt32to64 x) (SignExt32to64 y)))
   219  (Leq64 x y) => (XOR (MOVVconst [1]) (SGT x y))
   220  (Leq(32|64)F x y) => (FPFlagTrue (CMPGE(F|D) y x)) // reverse operands to work around NaN
   221  
   222  (Leq8U x y)  => (XOR (MOVVconst [1]) (SGTU (ZeroExt8to64 x) (ZeroExt8to64 y)))
   223  (Leq16U x y) => (XOR (MOVVconst [1]) (SGTU (ZeroExt16to64 x) (ZeroExt16to64 y)))
   224  (Leq32U x y) => (XOR (MOVVconst [1]) (SGTU (ZeroExt32to64 x) (ZeroExt32to64 y)))
   225  (Leq64U x y) => (XOR (MOVVconst [1]) (SGTU x y))
   226  
   227  (OffPtr [off] ptr:(SP)) => (MOVVaddr [int32(off)] ptr)
   228  (OffPtr [off] ptr) => (ADDVconst [off] ptr)
   229  
   230  (Addr {sym} base) => (MOVVaddr {sym} base)
   231  (LocalAddr <t> {sym} base mem) && t.Elem().HasPointers() => (MOVVaddr {sym} (SPanchored base mem))
   232  (LocalAddr <t> {sym} base _)  && !t.Elem().HasPointers() => (MOVVaddr {sym} base)
   233  
   234  // loads
   235  (Load <t> ptr mem) && t.IsBoolean() => (MOVBUload ptr mem)
   236  (Load <t> ptr mem) && (is8BitInt(t)  &&  t.IsSigned()) => (MOVBload ptr mem)
   237  (Load <t> ptr mem) && (is8BitInt(t)  && !t.IsSigned()) => (MOVBUload ptr mem)
   238  (Load <t> ptr mem) && (is16BitInt(t) &&  t.IsSigned()) => (MOVHload ptr mem)
   239  (Load <t> ptr mem) && (is16BitInt(t) && !t.IsSigned()) => (MOVHUload ptr mem)
   240  (Load <t> ptr mem) && (is32BitInt(t) &&  t.IsSigned()) => (MOVWload ptr mem)
   241  (Load <t> ptr mem) && (is32BitInt(t) && !t.IsSigned()) => (MOVWUload ptr mem)
   242  (Load <t> ptr mem) && (is64BitInt(t) || isPtr(t)) => (MOVVload ptr mem)
   243  (Load <t> ptr mem) && is32BitFloat(t) => (MOVFload ptr mem)
   244  (Load <t> ptr mem) && is64BitFloat(t) => (MOVDload ptr mem)
   245  
   246  // stores
   247  (Store {t} ptr val mem) && t.Size() == 1 => (MOVBstore ptr val mem)
   248  (Store {t} ptr val mem) && t.Size() == 2 => (MOVHstore ptr val mem)
   249  (Store {t} ptr val mem) && t.Size() == 4 && !t.IsFloat() => (MOVWstore ptr val mem)
   250  (Store {t} ptr val mem) && t.Size() == 8 && !t.IsFloat() => (MOVVstore ptr val mem)
   251  (Store {t} ptr val mem) && t.Size() == 4 &&  t.IsFloat() => (MOVFstore ptr val mem)
   252  (Store {t} ptr val mem) && t.Size() == 8 &&  t.IsFloat() => (MOVDstore ptr val mem)
   253  
   254  // zeroing
   255  (Zero [0] _ mem) => mem
   256  (Zero [1] ptr mem) => (MOVBstore ptr (MOVVconst [0]) mem)
   257  (Zero [2] {t} ptr mem) && t.Alignment()%2 == 0 =>
   258  	(MOVHstore ptr (MOVVconst [0]) mem)
   259  (Zero [2] ptr mem) =>
   260  	(MOVBstore [1] ptr (MOVVconst [0])
   261  		(MOVBstore [0] ptr (MOVVconst [0]) mem))
   262  (Zero [4] {t} ptr mem) && t.Alignment()%4 == 0 =>
   263  	(MOVWstore ptr (MOVVconst [0]) mem)
   264  (Zero [4] {t} ptr mem) && t.Alignment()%2 == 0 =>
   265  	(MOVHstore [2] ptr (MOVVconst [0])
   266  		(MOVHstore [0] ptr (MOVVconst [0]) mem))
   267  (Zero [4] ptr mem) =>
   268  	(MOVBstore [3] ptr (MOVVconst [0])
   269  		(MOVBstore [2] ptr (MOVVconst [0])
   270  			(MOVBstore [1] ptr (MOVVconst [0])
   271  				(MOVBstore [0] ptr (MOVVconst [0]) mem))))
   272  (Zero [8] {t} ptr mem) && t.Alignment()%8 == 0 =>
   273  	(MOVVstore ptr (MOVVconst [0]) mem)
   274  (Zero [8] {t} ptr mem) && t.Alignment()%4 == 0 =>
   275  	(MOVWstore [4] ptr (MOVVconst [0])
   276  		(MOVWstore [0] ptr (MOVVconst [0]) mem))
   277  (Zero [8] {t} ptr mem) && t.Alignment()%2 == 0 =>
   278  	(MOVHstore [6] ptr (MOVVconst [0])
   279  		(MOVHstore [4] ptr (MOVVconst [0])
   280  			(MOVHstore [2] ptr (MOVVconst [0])
   281  				(MOVHstore [0] ptr (MOVVconst [0]) mem))))
   282  
   283  (Zero [3] ptr mem) =>
   284  	(MOVBstore [2] ptr (MOVVconst [0])
   285  		(MOVBstore [1] ptr (MOVVconst [0])
   286  			(MOVBstore [0] ptr (MOVVconst [0]) mem)))
   287  (Zero [6] {t} ptr mem) && t.Alignment()%2 == 0 =>
   288  	(MOVHstore [4] ptr (MOVVconst [0])
   289  		(MOVHstore [2] ptr (MOVVconst [0])
   290  			(MOVHstore [0] ptr (MOVVconst [0]) mem)))
   291  (Zero [12] {t} ptr mem) && t.Alignment()%4 == 0 =>
   292  	(MOVWstore [8] ptr (MOVVconst [0])
   293  		(MOVWstore [4] ptr (MOVVconst [0])
   294  			(MOVWstore [0] ptr (MOVVconst [0]) mem)))
   295  (Zero [16] {t} ptr mem) && t.Alignment()%8 == 0 =>
   296  	(MOVVstore [8] ptr (MOVVconst [0])
   297  		(MOVVstore [0] ptr (MOVVconst [0]) mem))
   298  (Zero [24] {t} ptr mem) && t.Alignment()%8 == 0 =>
   299  	(MOVVstore [16] ptr (MOVVconst [0])
   300  		(MOVVstore [8] ptr (MOVVconst [0])
   301  			(MOVVstore [0] ptr (MOVVconst [0]) mem)))
   302  
   303  // medium zeroing uses a duff device
   304  // 8, and 128 are magic constants, see runtime/mkduff.go
   305  (Zero [s] {t} ptr mem)
   306  	&& s%8 == 0 && s > 24 && s <= 8*128
   307  	&& t.Alignment()%8 == 0 && !config.noDuffDevice =>
   308  	(DUFFZERO [8 * (128 - s/8)] ptr mem)
   309  
   310  // large or unaligned zeroing uses a loop
   311  (Zero [s] {t} ptr mem)
   312  	&& (s > 8*128 || config.noDuffDevice) || t.Alignment()%8 != 0 =>
   313  	(LoweredZero [t.Alignment()]
   314  		ptr
   315  		(ADDVconst <ptr.Type> ptr [s-moveSize(t.Alignment(), config)])
   316  		mem)
   317  
   318  // moves
   319  (Move [0] _ _ mem) => mem
   320  (Move [1] dst src mem) => (MOVBstore dst (MOVBload src mem) mem)
   321  (Move [2] {t} dst src mem) && t.Alignment()%2 == 0 =>
   322  	(MOVHstore dst (MOVHload src mem) mem)
   323  (Move [2] dst src mem) =>
   324  	(MOVBstore [1] dst (MOVBload [1] src mem)
   325  		(MOVBstore dst (MOVBload src mem) mem))
   326  (Move [4] {t} dst src mem) && t.Alignment()%4 == 0 =>
   327  	(MOVWstore dst (MOVWload src mem) mem)
   328  (Move [4] {t} dst src mem) && t.Alignment()%2 == 0 =>
   329  	(MOVHstore [2] dst (MOVHload [2] src mem)
   330  		(MOVHstore dst (MOVHload src mem) mem))
   331  (Move [4] dst src mem) =>
   332  	(MOVBstore [3] dst (MOVBload [3] src mem)
   333  		(MOVBstore [2] dst (MOVBload [2] src mem)
   334  			(MOVBstore [1] dst (MOVBload [1] src mem)
   335  				(MOVBstore dst (MOVBload src mem) mem))))
   336  (Move [8] {t} dst src mem) && t.Alignment()%8 == 0 =>
   337  	(MOVVstore dst (MOVVload src mem) mem)
   338  (Move [8] {t} dst src mem) && t.Alignment()%4 == 0 =>
   339  	(MOVWstore [4] dst (MOVWload [4] src mem)
   340  		(MOVWstore dst (MOVWload src mem) mem))
   341  (Move [8] {t} dst src mem) && t.Alignment()%2 == 0 =>
   342  	(MOVHstore [6] dst (MOVHload [6] src mem)
   343  		(MOVHstore [4] dst (MOVHload [4] src mem)
   344  			(MOVHstore [2] dst (MOVHload [2] src mem)
   345  				(MOVHstore dst (MOVHload src mem) mem))))
   346  
   347  (Move [3] dst src mem) =>
   348  	(MOVBstore [2] dst (MOVBload [2] src mem)
   349  		(MOVBstore [1] dst (MOVBload [1] src mem)
   350  			(MOVBstore dst (MOVBload src mem) mem)))
   351  (Move [6] {t} dst src mem) && t.Alignment()%2 == 0 =>
   352  	(MOVHstore [4] dst (MOVHload [4] src mem)
   353  		(MOVHstore [2] dst (MOVHload [2] src mem)
   354  			(MOVHstore dst (MOVHload src mem) mem)))
   355  (Move [12] {t} dst src mem) && t.Alignment()%4 == 0 =>
   356  	(MOVWstore [8] dst (MOVWload [8] src mem)
   357  		(MOVWstore [4] dst (MOVWload [4] src mem)
   358  			(MOVWstore dst (MOVWload src mem) mem)))
   359  (Move [16] {t} dst src mem) && t.Alignment()%8 == 0 =>
   360  	(MOVVstore [8] dst (MOVVload [8] src mem)
   361  		(MOVVstore dst (MOVVload src mem) mem))
   362  (Move [24] {t} dst src mem) && t.Alignment()%8 == 0 =>
   363  	(MOVVstore [16] dst (MOVVload [16] src mem)
   364  		(MOVVstore [8] dst (MOVVload [8] src mem)
   365  			(MOVVstore dst (MOVVload src mem) mem)))
   366  
   367  // medium move uses a duff device
   368  (Move [s] {t} dst src mem)
   369  	&& s%8 == 0 && s >= 24 && s <= 8*128 && t.Alignment()%8 == 0
   370  	&& !config.noDuffDevice && logLargeCopy(v, s)  =>
   371  	(DUFFCOPY [16 * (128 - s/8)] dst src mem)
   372  // 16 and 128 are magic constants.  16 is the number of bytes to encode:
   373  //	MOVV	(R1), R23
   374  //	ADDV	$8, R1
   375  //	MOVV	R23, (R2)
   376  //	ADDV	$8, R2
   377  // and 128 is the number of such blocks. See runtime/duff_mips64.s:duffcopy.
   378  
   379  // large or unaligned move uses a loop
   380  (Move [s] {t} dst src mem)
   381  	&& s > 24 && logLargeCopy(v, s) || t.Alignment()%8 != 0 =>
   382  	(LoweredMove [t.Alignment()]
   383  		dst
   384  		src
   385  		(ADDVconst <src.Type> src [s-moveSize(t.Alignment(), config)])
   386  		mem)
   387  
   388  // calls
   389  (StaticCall ...) => (CALLstatic ...)
   390  (ClosureCall ...) => (CALLclosure ...)
   391  (InterCall ...) => (CALLinter ...)
   392  (TailCall ...) => (CALLtail ...)
   393  
   394  // atomic intrinsics
   395  (AtomicLoad(8|32|64)   ...) => (LoweredAtomicLoad(8|32|64)  ...)
   396  (AtomicLoadPtr ...) => (LoweredAtomicLoad64 ...)
   397  
   398  (AtomicStore(8|32|64) ...) => (LoweredAtomicStore(8|32|64)  ...)
   399  (AtomicStorePtrNoWB ...) => (LoweredAtomicStore64 ...)
   400  
   401  (AtomicExchange(32|64) ...) => (LoweredAtomicExchange(32|64) ...)
   402  
   403  (AtomicAdd(32|64) ...) => (LoweredAtomicAdd(32|64) ...)
   404  
   405  (AtomicCompareAndSwap32 ptr old new mem) => (LoweredAtomicCas32 ptr (SignExt32to64 old) new mem)
   406  (AtomicCompareAndSwap64 ...) => (LoweredAtomicCas64 ...)
   407  
   408  // checks
   409  (NilCheck ...) => (LoweredNilCheck ...)
   410  (IsNonNil ptr) => (SGTU ptr (MOVVconst [0]))
   411  (IsInBounds idx len) => (SGTU len idx)
   412  (IsSliceInBounds idx len) => (XOR (MOVVconst [1]) (SGTU idx len))
   413  
   414  // pseudo-ops
   415  (GetClosurePtr ...) => (LoweredGetClosurePtr ...)
   416  (GetCallerSP ...) => (LoweredGetCallerSP ...)
   417  (GetCallerPC ...) => (LoweredGetCallerPC ...)
   418  
   419  (If cond yes no) => (NE (MOVBUreg <typ.UInt64> cond) yes no)
   420  
   421  // Write barrier.
   422  (WB ...) => (LoweredWB ...)
   423  
   424  (PanicBounds [kind] x y mem) && boundsABI(kind) == 0 => (LoweredPanicBoundsA [kind] x y mem)
   425  (PanicBounds [kind] x y mem) && boundsABI(kind) == 1 => (LoweredPanicBoundsB [kind] x y mem)
   426  (PanicBounds [kind] x y mem) && boundsABI(kind) == 2 => (LoweredPanicBoundsC [kind] x y mem)
   427  
   428  (CondSelect <t> x y cond) => (OR (MASKEQZ <t> x cond) (MASKNEZ <t> y cond))
   429  
   430  // Optimizations
   431  
   432  // Absorb boolean tests into block
   433  (NE (FPFlagTrue cmp) yes no) => (FPT cmp yes no)
   434  (NE (FPFlagFalse cmp) yes no) => (FPF cmp yes no)
   435  (EQ (FPFlagTrue cmp) yes no) => (FPF cmp yes no)
   436  (EQ (FPFlagFalse cmp) yes no) => (FPT cmp yes no)
   437  (NE (XORconst [1] cmp:(SGT _ _)) yes no) => (EQ cmp yes no)
   438  (NE (XORconst [1] cmp:(SGTU _ _)) yes no) => (EQ cmp yes no)
   439  (NE (XORconst [1] cmp:(SGTconst _)) yes no) => (EQ cmp yes no)
   440  (NE (XORconst [1] cmp:(SGTUconst _)) yes no) => (EQ cmp yes no)
   441  (EQ (XORconst [1] cmp:(SGT _ _)) yes no) => (NE cmp yes no)
   442  (EQ (XORconst [1] cmp:(SGTU _ _)) yes no) => (NE cmp yes no)
   443  (EQ (XORconst [1] cmp:(SGTconst _)) yes no) => (NE cmp yes no)
   444  (EQ (XORconst [1] cmp:(SGTUconst _)) yes no) => (NE cmp yes no)
   445  (NE (SGTUconst [1] x) yes no) => (EQ x yes no)
   446  (EQ (SGTUconst [1] x) yes no) => (NE x yes no)
   447  (NE (SGTU x (MOVVconst [0])) yes no) => (NE x yes no)
   448  (EQ (SGTU x (MOVVconst [0])) yes no) => (EQ x yes no)
   449  (NE (SGTconst [0] x) yes no) => (LTZ x yes no)
   450  (EQ (SGTconst [0] x) yes no) => (GEZ x yes no)
   451  (NE (SGT x (MOVVconst [0])) yes no) => (GTZ x yes no)
   452  (EQ (SGT x (MOVVconst [0])) yes no) => (LEZ x yes no)
   453  (MOVBUreg x:((SGT|SGTU) _ _)) => x
   454  
   455  // fold offset into address
   456  (ADDVconst [off1] (MOVVaddr [off2] {sym} ptr)) && is32Bit(off1+int64(off2)) => (MOVVaddr [int32(off1)+int32(off2)] {sym} ptr)
   457  
   458  // fold address into load/store
   459  // Do not fold global variable access in -dynlink mode, where it will be rewritten
   460  // to use the GOT via REGTMP, which currently cannot handle large offset.
   461  (MOV(B|BU|H|HU|W|WU|V|F|D)load [off1] {sym} (ADDVconst [off2] ptr) mem) && is32Bit(int64(off1)+off2)
   462  	&& (ptr.Op != OpSB || !config.ctxt.Flag_dynlink) =>
   463  	(MOV(B|BU|H|HU|W|WU|V|F|D)load [off1+int32(off2)] {sym} ptr mem)
   464  
   465  (MOV(B|H|W|V|F|D)store [off1] {sym} (ADDVconst [off2] ptr) val mem) && is32Bit(int64(off1)+off2)
   466  	&& (ptr.Op != OpSB || !config.ctxt.Flag_dynlink) =>
   467  	(MOV(B|H|W|V|F|D)store [off1+int32(off2)] {sym} ptr val mem)
   468  
   469  (MOV(B|H|W|V)storezero [off1] {sym} (ADDVconst [off2] ptr) mem) && is32Bit(int64(off1)+off2)
   470  	&& (ptr.Op != OpSB || !config.ctxt.Flag_dynlink) =>
   471  	(MOV(B|H|W|V)storezero [off1+int32(off2)] {sym} ptr mem)
   472  
   473  (MOV(B|BU|H|HU|W|WU|V|F|D)load [off1] {sym1} (MOVVaddr [off2] {sym2} ptr) mem) && canMergeSym(sym1,sym2)
   474  	&& is32Bit(int64(off1)+int64(off2)) && (ptr.Op != OpSB || !config.ctxt.Flag_dynlink) =>
   475  	(MOV(B|BU|H|HU|W|WU|V|F|D)load [off1+int32(off2)] {mergeSym(sym1,sym2)} ptr mem)
   476  
   477  (MOV(B|H|W|V|F|D)store [off1] {sym1} (MOVVaddr [off2] {sym2} ptr) val mem) && canMergeSym(sym1,sym2)
   478  	&& is32Bit(int64(off1)+int64(off2)) && (ptr.Op != OpSB || !config.ctxt.Flag_dynlink) =>
   479  	(MOV(B|H|W|V|F|D)store [off1+int32(off2)] {mergeSym(sym1,sym2)} ptr val mem)
   480  
   481  (MOV(B|H|W|V)storezero [off1] {sym1} (MOVVaddr [off2] {sym2} ptr) mem) && canMergeSym(sym1,sym2)
   482  	&& is32Bit(int64(off1)+int64(off2)) && (ptr.Op != OpSB || !config.ctxt.Flag_dynlink) =>
   483  	(MOV(B|H|W|V)storezero [off1+int32(off2)] {mergeSym(sym1,sym2)} ptr mem)
   484  
   485  (LoweredAtomicStore(32|64) ptr (MOVVconst [0]) mem) => (LoweredAtomicStorezero(32|64) ptr mem)
   486  (LoweredAtomicAdd32 ptr (MOVVconst [c]) mem) && is32Bit(c) => (LoweredAtomicAddconst32 [int32(c)] ptr mem)
   487  (LoweredAtomicAdd64 ptr (MOVVconst [c]) mem) && is32Bit(c) => (LoweredAtomicAddconst64 [c] ptr mem)
   488  
   489  // don't extend after proper load
   490  (MOVBreg x:(MOVBload _ _)) => (MOVVreg x)
   491  (MOVBUreg x:(MOVBUload _ _)) => (MOVVreg x)
   492  (MOVHreg x:(MOVBload _ _)) => (MOVVreg x)
   493  (MOVHreg x:(MOVBUload _ _)) => (MOVVreg x)
   494  (MOVHreg x:(MOVHload _ _)) => (MOVVreg x)
   495  (MOVHUreg x:(MOVBUload _ _)) => (MOVVreg x)
   496  (MOVHUreg x:(MOVHUload _ _)) => (MOVVreg x)
   497  (MOVWreg x:(MOVBload _ _)) => (MOVVreg x)
   498  (MOVWreg x:(MOVBUload _ _)) => (MOVVreg x)
   499  (MOVWreg x:(MOVHload _ _)) => (MOVVreg x)
   500  (MOVWreg x:(MOVHUload _ _)) => (MOVVreg x)
   501  (MOVWreg x:(MOVWload _ _)) => (MOVVreg x)
   502  (MOVWUreg x:(MOVBUload _ _)) => (MOVVreg x)
   503  (MOVWUreg x:(MOVHUload _ _)) => (MOVVreg x)
   504  (MOVWUreg x:(MOVWUload _ _)) => (MOVVreg x)
   505  
   506  // fold double extensions
   507  (MOVBreg x:(MOVBreg _)) => (MOVVreg x)
   508  (MOVBUreg x:(MOVBUreg _)) => (MOVVreg x)
   509  (MOVHreg x:(MOVBreg _)) => (MOVVreg x)
   510  (MOVHreg x:(MOVBUreg _)) => (MOVVreg x)
   511  (MOVHreg x:(MOVHreg _)) => (MOVVreg x)
   512  (MOVHUreg x:(MOVBUreg _)) => (MOVVreg x)
   513  (MOVHUreg x:(MOVHUreg _)) => (MOVVreg x)
   514  (MOVWreg x:(MOVBreg _)) => (MOVVreg x)
   515  (MOVWreg x:(MOVBUreg _)) => (MOVVreg x)
   516  (MOVWreg x:(MOVHreg _)) => (MOVVreg x)
   517  (MOVWreg x:(MOVWreg _)) => (MOVVreg x)
   518  (MOVWUreg x:(MOVBUreg _)) => (MOVVreg x)
   519  (MOVWUreg x:(MOVHUreg _)) => (MOVVreg x)
   520  (MOVWUreg x:(MOVWUreg _)) => (MOVVreg x)
   521  
   522  // don't extend before store
   523  (MOVBstore [off] {sym} ptr (MOVBreg x) mem) => (MOVBstore [off] {sym} ptr x mem)
   524  (MOVBstore [off] {sym} ptr (MOVBUreg x) mem) => (MOVBstore [off] {sym} ptr x mem)
   525  (MOVBstore [off] {sym} ptr (MOVHreg x) mem) => (MOVBstore [off] {sym} ptr x mem)
   526  (MOVBstore [off] {sym} ptr (MOVHUreg x) mem) => (MOVBstore [off] {sym} ptr x mem)
   527  (MOVBstore [off] {sym} ptr (MOVWreg x) mem) => (MOVBstore [off] {sym} ptr x mem)
   528  (MOVBstore [off] {sym} ptr (MOVWUreg x) mem) => (MOVBstore [off] {sym} ptr x mem)
   529  (MOVHstore [off] {sym} ptr (MOVHreg x) mem) => (MOVHstore [off] {sym} ptr x mem)
   530  (MOVHstore [off] {sym} ptr (MOVHUreg x) mem) => (MOVHstore [off] {sym} ptr x mem)
   531  (MOVHstore [off] {sym} ptr (MOVWreg x) mem) => (MOVHstore [off] {sym} ptr x mem)
   532  (MOVHstore [off] {sym} ptr (MOVWUreg x) mem) => (MOVHstore [off] {sym} ptr x mem)
   533  (MOVWstore [off] {sym} ptr (MOVWreg x) mem) => (MOVWstore [off] {sym} ptr x mem)
   534  (MOVWstore [off] {sym} ptr (MOVWUreg x) mem) => (MOVWstore [off] {sym} ptr x mem)
   535  
   536  // if a register move has only 1 use, just use the same register without emitting instruction
   537  // MOVVnop doesn't emit instruction, only for ensuring the type.
   538  (MOVVreg x) && x.Uses == 1 => (MOVVnop x)
   539  
   540  // fold constant into arithmetic ops
   541  (ADDV x (MOVVconst <t> [c])) && is32Bit(c) && !t.IsPtr() => (ADDVconst [c] x)
   542  (SUBV x (MOVVconst [c])) && is32Bit(c) => (SUBVconst [c] x)
   543  (AND x (MOVVconst [c])) && is32Bit(c) => (ANDconst [c] x)
   544  (OR  x (MOVVconst [c])) && is32Bit(c) => (ORconst  [c] x)
   545  (XOR x (MOVVconst [c])) && is32Bit(c) => (XORconst [c] x)
   546  (NOR x (MOVVconst [c])) && is32Bit(c) => (NORconst [c] x)
   547  
   548  (SLLV _ (MOVVconst [c])) && uint64(c)>=64 => (MOVVconst [0])
   549  (SRLV _ (MOVVconst [c])) && uint64(c)>=64 => (MOVVconst [0])
   550  (SRAV x (MOVVconst [c])) && uint64(c)>=64 => (SRAVconst x [63])
   551  (SLLV x (MOVVconst [c])) => (SLLVconst x [c])
   552  (SRLV x (MOVVconst [c])) => (SRLVconst x [c])
   553  (SRAV x (MOVVconst [c])) => (SRAVconst x [c])
   554  (ROTR x (MOVVconst [c]))  => (ROTRconst x [c&31])
   555  (ROTRV x (MOVVconst [c])) => (ROTRVconst x [c&63])
   556  
   557  (SGT  (MOVVconst [c]) x) && is32Bit(c) => (SGTconst  [c] x)
   558  (SGTU (MOVVconst [c]) x) && is32Bit(c) => (SGTUconst [c] x)
   559  
   560  // mul by constant
   561  (MULV x (MOVVconst [-1])) => (NEGV x)
   562  (MULV _ (MOVVconst [0])) => (MOVVconst [0])
   563  (MULV x (MOVVconst [1])) => x
   564  (MULV x (MOVVconst [c])) && isPowerOfTwo64(c) => (SLLVconst [log64(c)] x)
   565  
   566  // div by constant
   567  (DIVVU x (MOVVconst [1])) => x
   568  (DIVVU x (MOVVconst [c])) && isPowerOfTwo64(c) => (SRLVconst [log64(c)] x)
   569  (REMVU _ (MOVVconst [1])) => (MOVVconst [0])                       // mod
   570  (REMVU x (MOVVconst [c])) && isPowerOfTwo64(c) => (ANDconst [c-1] x) // mod
   571  
   572  // generic simplifications
   573  (ADDV x (NEGV y)) => (SUBV x y)
   574  (SUBV x x) => (MOVVconst [0])
   575  (SUBV (MOVVconst [0]) x) => (NEGV x)
   576  (AND x x) => x
   577  (OR  x x) => x
   578  (XOR x x) => (MOVVconst [0])
   579  
   580  // remove redundant *const ops
   581  (ADDVconst [0]  x) => x
   582  (SUBVconst [0]  x) => x
   583  (ANDconst [0]  _) => (MOVVconst [0])
   584  (ANDconst [-1] x) => x
   585  (ORconst  [0]  x) => x
   586  (ORconst  [-1] _) => (MOVVconst [-1])
   587  (XORconst [0]  x) => x
   588  (XORconst [-1] x) => (NORconst [0] x)
   589  (MASKEQZ (MOVVconst [0]) cond) => (MOVVconst [0])
   590  (MASKNEZ (MOVVconst [0]) cond) => (MOVVconst [0])
   591  (MASKEQZ x (MOVVconst [c])) && c == 0 => (MOVVconst [0])
   592  (MASKEQZ x (MOVVconst [c])) && c != 0 => x
   593  
   594  // generic constant folding
   595  (ADDVconst [c] (MOVVconst [d]))  => (MOVVconst [c+d])
   596  (ADDVconst [c] (ADDVconst [d] x)) && is32Bit(c+d) => (ADDVconst [c+d] x)
   597  (ADDVconst [c] (SUBVconst [d] x)) && is32Bit(c-d) => (ADDVconst [c-d] x)
   598  (SUBVconst [c] (MOVVconst [d]))  => (MOVVconst [d-c])
   599  (SUBVconst [c] (SUBVconst [d] x)) && is32Bit(-c-d) => (ADDVconst [-c-d] x)
   600  (SUBVconst [c] (ADDVconst [d] x)) && is32Bit(-c+d) => (ADDVconst [-c+d] x)
   601  (SLLVconst [c] (MOVVconst [d]))  => (MOVVconst [d<<uint64(c)])
   602  (SRLVconst [c] (MOVVconst [d]))  => (MOVVconst [int64(uint64(d)>>uint64(c))])
   603  (SRAVconst [c] (MOVVconst [d]))  => (MOVVconst [d>>uint64(c)])
   604  (MULV (MOVVconst [c]) (MOVVconst [d])) => (MOVVconst [c*d])
   605  (DIVV  (MOVVconst [c]) (MOVVconst [d])) && d != 0 => (MOVVconst [c/d])
   606  (DIVVU (MOVVconst [c]) (MOVVconst [d])) && d != 0 => (MOVVconst [int64(uint64(c)/uint64(d))])
   607  (REMV  (MOVVconst [c]) (MOVVconst [d])) && d != 0 => (MOVVconst [c%d])   // mod
   608  (REMVU (MOVVconst [c]) (MOVVconst [d])) && d != 0 => (MOVVconst [int64(uint64(c)%uint64(d))]) // mod
   609  (ANDconst [c] (MOVVconst [d])) => (MOVVconst [c&d])
   610  (ANDconst [c] (ANDconst [d] x)) => (ANDconst [c&d] x)
   611  (ORconst [c] (MOVVconst [d])) => (MOVVconst [c|d])
   612  (ORconst [c] (ORconst [d] x)) && is32Bit(c|d) => (ORconst [c|d] x)
   613  (XORconst [c] (MOVVconst [d])) => (MOVVconst [c^d])
   614  (XORconst [c] (XORconst [d] x)) && is32Bit(c^d) => (XORconst [c^d] x)
   615  (NORconst [c] (MOVVconst [d])) => (MOVVconst [^(c|d)])
   616  (NEGV (MOVVconst [c])) => (MOVVconst [-c])
   617  (MOVBreg  (MOVVconst [c])) => (MOVVconst [int64(int8(c))])
   618  (MOVBUreg (MOVVconst [c])) => (MOVVconst [int64(uint8(c))])
   619  (MOVHreg  (MOVVconst [c])) => (MOVVconst [int64(int16(c))])
   620  (MOVHUreg (MOVVconst [c])) => (MOVVconst [int64(uint16(c))])
   621  (MOVWreg  (MOVVconst [c])) => (MOVVconst [int64(int32(c))])
   622  (MOVWUreg (MOVVconst [c])) => (MOVVconst [int64(uint32(c))])
   623  (MOVVreg  (MOVVconst [c])) => (MOVVconst [c])
   624  
   625  // constant comparisons
   626  (SGTconst [c] (MOVVconst [d])) && c>d => (MOVVconst [1])
   627  (SGTconst [c] (MOVVconst [d])) && c<=d => (MOVVconst [0])
   628  (SGTUconst [c] (MOVVconst [d])) && uint64(c)>uint64(d) => (MOVVconst [1])
   629  (SGTUconst [c] (MOVVconst [d])) && uint64(c)<=uint64(d) => (MOVVconst [0])
   630  
   631  // other known comparisons
   632  (SGTconst [c] (MOVBreg _)) && 0x7f < c => (MOVVconst [1])
   633  (SGTconst [c] (MOVBreg _)) && c <= -0x80 => (MOVVconst [0])
   634  (SGTconst [c] (MOVBUreg _)) && 0xff < c => (MOVVconst [1])
   635  (SGTconst [c] (MOVBUreg _)) && c < 0 => (MOVVconst [0])
   636  (SGTUconst [c] (MOVBUreg _)) && 0xff < uint64(c) => (MOVVconst [1])
   637  (SGTconst [c] (MOVHreg _)) && 0x7fff < c => (MOVVconst [1])
   638  (SGTconst [c] (MOVHreg _)) && c <= -0x8000 => (MOVVconst [0])
   639  (SGTconst [c] (MOVHUreg _)) && 0xffff < c => (MOVVconst [1])
   640  (SGTconst [c] (MOVHUreg _)) && c < 0 => (MOVVconst [0])
   641  (SGTUconst [c] (MOVHUreg _)) && 0xffff < uint64(c) => (MOVVconst [1])
   642  (SGTconst [c] (MOVWUreg _)) && c < 0 => (MOVVconst [0])
   643  (SGTconst [c] (ANDconst [m] _)) && 0 <= m && m < c => (MOVVconst [1])
   644  (SGTUconst [c] (ANDconst [m] _)) && uint64(m) < uint64(c) => (MOVVconst [1])
   645  (SGTconst [c] (SRLVconst _ [d])) && 0 <= c && 0 < d && d <= 63 && 0xffffffffffffffff>>uint64(d) < uint64(c) => (MOVVconst [1])
   646  (SGTUconst [c] (SRLVconst _ [d])) && 0 < d && d <= 63 && 0xffffffffffffffff>>uint64(d) < uint64(c) => (MOVVconst [1])
   647  
   648  // absorb constants into branches
   649  (EQ  (MOVVconst [0]) yes no) => (First yes no)
   650  (EQ  (MOVVconst [c]) yes no) && c != 0 => (First no yes)
   651  (NE  (MOVVconst [0]) yes no) => (First no yes)
   652  (NE  (MOVVconst [c]) yes no) && c != 0 => (First yes no)
   653  (LTZ (MOVVconst [c]) yes no) && c <  0 => (First yes no)
   654  (LTZ (MOVVconst [c]) yes no) && c >= 0 => (First no yes)
   655  (LEZ (MOVVconst [c]) yes no) && c <= 0 => (First yes no)
   656  (LEZ (MOVVconst [c]) yes no) && c >  0 => (First no yes)
   657  (GTZ (MOVVconst [c]) yes no) && c >  0 => (First yes no)
   658  (GTZ (MOVVconst [c]) yes no) && c <= 0 => (First no yes)
   659  (GEZ (MOVVconst [c]) yes no) && c >= 0 => (First yes no)
   660  (GEZ (MOVVconst [c]) yes no) && c <  0 => (First no yes)
   661  
   662  // SGT/SGTU with known outcomes.
   663  (SGT  x x) => (MOVVconst [0])
   664  (SGTU x x) => (MOVVconst [0])