...

Text file src/cmd/compile/internal/ssa/_gen/MIPS64.rules

Documentation: cmd/compile/internal/ssa/_gen

     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) ...) => (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) x y) => (Select1 (MULVU x y))
    12(Mul(32|64)F ...) => (MUL(F|D) ...)
    13(Mul64uhilo ...) => (MULVU ...)
    14(Select0 (Mul64uover x y)) => (Select1 <typ.UInt64> (MULVU x y))
    15(Select1 (Mul64uover x y)) => (SGTU <typ.Bool> (Select0 <typ.UInt64> (MULVU x y)) (MOVVconst <typ.UInt64> [0]))
    16
    17(Hmul64 x y) => (Select0 (MULV x y))
    18(Hmul64u x y) => (Select0 (MULVU x y))
    19(Hmul32 x y) => (SRAVconst (Select1 <typ.Int64> (MULV (SignExt32to64 x) (SignExt32to64 y))) [32])
    20(Hmul32u x y) => (SRLVconst (Select1 <typ.UInt64> (MULVU (ZeroExt32to64 x) (ZeroExt32to64 y))) [32])
    21
    22(Div64 x y) => (Select1 (DIVV x y))
    23(Div64u x y) => (Select1 (DIVVU x y))
    24(Div32 x y) => (Select1 (DIVV (SignExt32to64 x) (SignExt32to64 y)))
    25(Div32u x y) => (Select1 (DIVVU (ZeroExt32to64 x) (ZeroExt32to64 y)))
    26(Div16 x y) => (Select1 (DIVV (SignExt16to64 x) (SignExt16to64 y)))
    27(Div16u x y) => (Select1 (DIVVU (ZeroExt16to64 x) (ZeroExt16to64 y)))
    28(Div8 x y) => (Select1 (DIVV (SignExt8to64 x) (SignExt8to64 y)))
    29(Div8u x y) => (Select1 (DIVVU (ZeroExt8to64 x) (ZeroExt8to64 y)))
    30(Div(32|64)F ...) => (DIV(F|D) ...)
    31
    32(Mod64 x y) => (Select0 (DIVV x y))
    33(Mod64u x y) => (Select0 (DIVVU x y))
    34(Mod32 x y) => (Select0 (DIVV (SignExt32to64 x) (SignExt32to64 y)))
    35(Mod32u x y) => (Select0 (DIVVU (ZeroExt32to64 x) (ZeroExt32to64 y)))
    36(Mod16 x y) => (Select0 (DIVV (SignExt16to64 x) (SignExt16to64 y)))
    37(Mod16u x y) => (Select0 (DIVVU (ZeroExt16to64 x) (ZeroExt16to64 y)))
    38(Mod8 x y) => (Select0 (DIVV (SignExt8to64 x) (SignExt8to64 y)))
    39(Mod8u x y) => (Select0 (DIVVU (ZeroExt8to64 x) (ZeroExt8to64 y)))
    40
    41(Select0 <t> (Add64carry x y c)) => (ADDV (ADDV <t> x y) c)
    42(Select1 <t> (Add64carry x y c)) =>
    43	(OR (SGTU <t> x s:(ADDV <t> x y)) (SGTU <t> s (ADDV <t> s c)))
    44
    45(Select0 <t> (Sub64borrow x y c)) => (SUBV (SUBV <t> x y) c)
    46(Select1 <t> (Sub64borrow x y c)) =>
    47	(OR (SGTU <t> s:(SUBV <t> x y) x) (SGTU <t> (SUBV <t> s c) s))
    48
    49// math package intrinsics
    50(Abs ...) => (ABSD ...)
    51
    52// (x + y) / 2 with x>=y => (x - y) / 2 + y
    53(Avg64u <t> x y) => (ADDV (SRLVconst <t> (SUBV <t> x y) [1]) y)
    54
    55(And(64|32|16|8) ...) => (AND ...)
    56(Or(64|32|16|8) ...) => (OR ...)
    57(Xor(64|32|16|8) ...) => (XOR ...)
    58
    59// shifts
    60// hardware instruction uses only the low 6 bits of the shift
    61// we compare to 64 to ensure Go semantics for large shifts
    62(Lsh64x64 <t> x y) => (AND (NEGV <t> (SGTU (MOVVconst <typ.UInt64> [64]) y)) (SLLV <t> x y))
    63(Lsh64x32 <t> x y) => (AND (NEGV <t> (SGTU (MOVVconst <typ.UInt64> [64]) (ZeroExt32to64 y))) (SLLV <t> x (ZeroExt32to64 y)))
    64(Lsh64x16 <t> x y) => (AND (NEGV <t> (SGTU (MOVVconst <typ.UInt64> [64]) (ZeroExt16to64 y))) (SLLV <t> x (ZeroExt16to64 y)))
    65(Lsh64x8  <t> x y) => (AND (NEGV <t> (SGTU (MOVVconst <typ.UInt64> [64]) (ZeroExt8to64  y))) (SLLV <t> x (ZeroExt8to64  y)))
    66
    67(Lsh32x64 <t> x y) => (AND (NEGV <t> (SGTU (MOVVconst <typ.UInt64> [64]) y)) (SLLV <t> x y))
    68(Lsh32x32 <t> x y) => (AND (NEGV <t> (SGTU (MOVVconst <typ.UInt64> [64]) (ZeroExt32to64 y))) (SLLV <t> x (ZeroExt32to64 y)))
    69(Lsh32x16 <t> x y) => (AND (NEGV <t> (SGTU (MOVVconst <typ.UInt64> [64]) (ZeroExt16to64 y))) (SLLV <t> x (ZeroExt16to64 y)))
    70(Lsh32x8  <t> x y) => (AND (NEGV <t> (SGTU (MOVVconst <typ.UInt64> [64]) (ZeroExt8to64  y))) (SLLV <t> x (ZeroExt8to64  y)))
    71
    72(Lsh16x64 <t> x y) => (AND (NEGV <t> (SGTU (MOVVconst <typ.UInt64> [64]) y)) (SLLV <t> x y))
    73(Lsh16x32 <t> x y) => (AND (NEGV <t> (SGTU (MOVVconst <typ.UInt64> [64]) (ZeroExt32to64 y))) (SLLV <t> x (ZeroExt32to64 y)))
    74(Lsh16x16 <t> x y) => (AND (NEGV <t> (SGTU (MOVVconst <typ.UInt64> [64]) (ZeroExt16to64 y))) (SLLV <t> x (ZeroExt16to64 y)))
    75(Lsh16x8  <t> x y) => (AND (NEGV <t> (SGTU (MOVVconst <typ.UInt64> [64]) (ZeroExt8to64  y))) (SLLV <t> x (ZeroExt8to64  y)))
    76
    77(Lsh8x64 <t> x y) => (AND (NEGV <t> (SGTU (MOVVconst <typ.UInt64> [64]) y)) (SLLV <t> x y))
    78(Lsh8x32 <t> x y) => (AND (NEGV <t> (SGTU (MOVVconst <typ.UInt64> [64]) (ZeroExt32to64 y))) (SLLV <t> x (ZeroExt32to64 y)))
    79(Lsh8x16 <t> x y) => (AND (NEGV <t> (SGTU (MOVVconst <typ.UInt64> [64]) (ZeroExt16to64 y))) (SLLV <t> x (ZeroExt16to64 y)))
    80(Lsh8x8  <t> x y) => (AND (NEGV <t> (SGTU (MOVVconst <typ.UInt64> [64]) (ZeroExt8to64  y))) (SLLV <t> x (ZeroExt8to64  y)))
    81
    82(Rsh64Ux64 <t> x y) => (AND (NEGV <t> (SGTU (MOVVconst <typ.UInt64> [64]) y)) (SRLV <t> x y))
    83(Rsh64Ux32 <t> x y) => (AND (NEGV <t> (SGTU (MOVVconst <typ.UInt64> [64]) (ZeroExt32to64 y))) (SRLV <t> x (ZeroExt32to64 y)))
    84(Rsh64Ux16 <t> x y) => (AND (NEGV <t> (SGTU (MOVVconst <typ.UInt64> [64]) (ZeroExt16to64 y))) (SRLV <t> x (ZeroExt16to64 y)))
    85(Rsh64Ux8  <t> x y) => (AND (NEGV <t> (SGTU (MOVVconst <typ.UInt64> [64]) (ZeroExt8to64  y))) (SRLV <t> x (ZeroExt8to64  y)))
    86
    87(Rsh32Ux64 <t> x y) => (AND (NEGV <t> (SGTU (MOVVconst <typ.UInt64> [64]) y)) (SRLV <t> (ZeroExt32to64 x) y))
    88(Rsh32Ux32 <t> x y) => (AND (NEGV <t> (SGTU (MOVVconst <typ.UInt64> [64]) (ZeroExt32to64 y))) (SRLV <t> (ZeroExt32to64 x) (ZeroExt32to64 y)))
    89(Rsh32Ux16 <t> x y) => (AND (NEGV <t> (SGTU (MOVVconst <typ.UInt64> [64]) (ZeroExt16to64 y))) (SRLV <t> (ZeroExt32to64 x) (ZeroExt16to64 y)))
    90(Rsh32Ux8  <t> x y) => (AND (NEGV <t> (SGTU (MOVVconst <typ.UInt64> [64]) (ZeroExt8to64  y))) (SRLV <t> (ZeroExt32to64 x) (ZeroExt8to64  y)))
    91
    92(Rsh16Ux64 <t> x y) => (AND (NEGV <t> (SGTU (MOVVconst <typ.UInt64> [64]) y)) (SRLV <t> (ZeroExt16to64 x) y))
    93(Rsh16Ux32 <t> x y) => (AND (NEGV <t> (SGTU (MOVVconst <typ.UInt64> [64]) (ZeroExt32to64 y))) (SRLV <t> (ZeroExt16to64 x) (ZeroExt32to64 y)))
    94(Rsh16Ux16 <t> x y) => (AND (NEGV <t> (SGTU (MOVVconst <typ.UInt64> [64]) (ZeroExt16to64 y))) (SRLV <t> (ZeroExt16to64 x) (ZeroExt16to64 y)))
    95(Rsh16Ux8  <t> x y) => (AND (NEGV <t> (SGTU (MOVVconst <typ.UInt64> [64]) (ZeroExt8to64  y))) (SRLV <t> (ZeroExt16to64 x) (ZeroExt8to64  y)))
    96
    97(Rsh8Ux64 <t> x y) => (AND (NEGV <t> (SGTU (MOVVconst <typ.UInt64> [64]) y)) (SRLV <t> (ZeroExt8to64 x) y))
    98(Rsh8Ux32 <t> x y) => (AND (NEGV <t> (SGTU (MOVVconst <typ.UInt64> [64]) (ZeroExt32to64 y))) (SRLV <t> (ZeroExt8to64 x) (ZeroExt32to64 y)))
    99(Rsh8Ux16 <t> x y) => (AND (NEGV <t> (SGTU (MOVVconst <typ.UInt64> [64]) (ZeroExt16to64 y))) (SRLV <t> (ZeroExt8to64 x) (ZeroExt16to64 y)))
   100(Rsh8Ux8  <t> x y) => (AND (NEGV <t> (SGTU (MOVVconst <typ.UInt64> [64]) (ZeroExt8to64  y))) (SRLV <t> (ZeroExt8to64 x) (ZeroExt8to64  y)))
   101
   102(Rsh64x64 <t> x y) => (SRAV x (OR <t> (NEGV <t> (SGTU y (MOVVconst <typ.UInt64> [63]))) y))
   103(Rsh64x32 <t> x y) => (SRAV x (OR <t> (NEGV <t> (SGTU (ZeroExt32to64 y) (MOVVconst <typ.UInt64> [63]))) (ZeroExt32to64 y)))
   104(Rsh64x16 <t> x y) => (SRAV x (OR <t> (NEGV <t> (SGTU (ZeroExt16to64 y) (MOVVconst <typ.UInt64> [63]))) (ZeroExt16to64 y)))
   105(Rsh64x8  <t> x y) => (SRAV x (OR <t> (NEGV <t> (SGTU (ZeroExt8to64  y) (MOVVconst <typ.UInt64> [63]))) (ZeroExt8to64  y)))
   106
   107(Rsh32x64 <t> x y) => (SRAV (SignExt32to64 x) (OR <t> (NEGV <t> (SGTU y (MOVVconst <typ.UInt64> [63]))) y))
   108(Rsh32x32 <t> x y) => (SRAV (SignExt32to64 x) (OR <t> (NEGV <t> (SGTU (ZeroExt32to64 y) (MOVVconst <typ.UInt64> [63]))) (ZeroExt32to64 y)))
   109(Rsh32x16 <t> x y) => (SRAV (SignExt32to64 x) (OR <t> (NEGV <t> (SGTU (ZeroExt16to64 y) (MOVVconst <typ.UInt64> [63]))) (ZeroExt16to64 y)))
   110(Rsh32x8  <t> x y) => (SRAV (SignExt32to64 x) (OR <t> (NEGV <t> (SGTU (ZeroExt8to64  y) (MOVVconst <typ.UInt64> [63]))) (ZeroExt8to64  y)))
   111
   112(Rsh16x64 <t> x y) => (SRAV (SignExt16to64 x) (OR <t> (NEGV <t> (SGTU y (MOVVconst <typ.UInt64> [63]))) y))
   113(Rsh16x32 <t> x y) => (SRAV (SignExt16to64 x) (OR <t> (NEGV <t> (SGTU (ZeroExt32to64 y) (MOVVconst <typ.UInt64> [63]))) (ZeroExt32to64 y)))
   114(Rsh16x16 <t> x y) => (SRAV (SignExt16to64 x) (OR <t> (NEGV <t> (SGTU (ZeroExt16to64 y) (MOVVconst <typ.UInt64> [63]))) (ZeroExt16to64 y)))
   115(Rsh16x8  <t> x y) => (SRAV (SignExt16to64 x) (OR <t> (NEGV <t> (SGTU (ZeroExt8to64  y) (MOVVconst <typ.UInt64> [63]))) (ZeroExt8to64  y)))
   116
   117(Rsh8x64 <t> x y) => (SRAV (SignExt8to64 x) (OR <t> (NEGV <t> (SGTU y (MOVVconst <typ.UInt64> [63]))) y))
   118(Rsh8x32 <t> x y) => (SRAV (SignExt8to64 x) (OR <t> (NEGV <t> (SGTU (ZeroExt32to64 y) (MOVVconst <typ.UInt64> [63]))) (ZeroExt32to64 y)))
   119(Rsh8x16 <t> x y) => (SRAV (SignExt8to64 x) (OR <t> (NEGV <t> (SGTU (ZeroExt16to64 y) (MOVVconst <typ.UInt64> [63]))) (ZeroExt16to64 y)))
   120(Rsh8x8  <t> x y) => (SRAV (SignExt8to64 x) (OR <t> (NEGV <t> (SGTU (ZeroExt8to64  y) (MOVVconst <typ.UInt64> [63]))) (ZeroExt8to64  y)))
   121
   122// rotates
   123(RotateLeft8 <t> x (MOVVconst [c])) => (Or8 (Lsh8x64 <t> x (MOVVconst [c&7])) (Rsh8Ux64 <t> x (MOVVconst [-c&7])))
   124(RotateLeft16 <t> x (MOVVconst [c])) => (Or16 (Lsh16x64 <t> x (MOVVconst [c&15])) (Rsh16Ux64 <t> x (MOVVconst [-c&15])))
   125(RotateLeft32 <t> x (MOVVconst [c])) => (Or32 (Lsh32x64 <t> x (MOVVconst [c&31])) (Rsh32Ux64 <t> x (MOVVconst [-c&31])))
   126(RotateLeft64 <t> x (MOVVconst [c])) => (Or64 (Lsh64x64 <t> x (MOVVconst [c&63])) (Rsh64Ux64 <t> x (MOVVconst [-c&63])))
   127
   128// unary ops
   129(Neg(64|32|16|8) ...) => (NEGV ...)
   130(Neg(32|64)F ...) => (NEG(F|D) ...)
   131
   132(Com(64|32|16|8) x) => (NOR (MOVVconst [0]) x)
   133
   134(Sqrt ...) => (SQRTD ...)
   135(Sqrt32 ...) => (SQRTF ...)
   136
   137// boolean ops -- booleans are represented with 0=false, 1=true
   138(AndB ...) => (AND ...)
   139(OrB ...) => (OR ...)
   140(EqB x y) => (XOR (MOVVconst [1]) (XOR <typ.Bool> x y))
   141(NeqB ...) => (XOR ...)
   142(Not x) => (XORconst [1] x)
   143
   144// constants
   145(Const(64|32|16|8) [val]) => (MOVVconst [int64(val)])
   146(Const(32|64)F [val]) => (MOV(F|D)const [float64(val)])
   147(ConstNil) => (MOVVconst [0])
   148(ConstBool [t]) => (MOVVconst [int64(b2i(t))])
   149
   150(Slicemask <t> x) => (SRAVconst (NEGV <t> x) [63])
   151
   152// truncations
   153// Because we ignore high parts of registers, truncates are just copies.
   154(Trunc16to8 ...) => (Copy ...)
   155(Trunc32to8 ...) => (Copy ...)
   156(Trunc32to16 ...) => (Copy ...)
   157(Trunc64to8 ...) => (Copy ...)
   158(Trunc64to16 ...) => (Copy ...)
   159(Trunc64to32 ...) => (Copy ...)
   160
   161// Zero-/Sign-extensions
   162(ZeroExt8to16 ...) => (MOVBUreg ...)
   163(ZeroExt8to32 ...) => (MOVBUreg ...)
   164(ZeroExt16to32 ...) => (MOVHUreg ...)
   165(ZeroExt8to64 ...) => (MOVBUreg ...)
   166(ZeroExt16to64 ...) => (MOVHUreg ...)
   167(ZeroExt32to64 ...) => (MOVWUreg ...)
   168
   169(SignExt8to16 ...) => (MOVBreg ...)
   170(SignExt8to32 ...) => (MOVBreg ...)
   171(SignExt16to32 ...) => (MOVHreg ...)
   172(SignExt8to64 ...) => (MOVBreg ...)
   173(SignExt16to64 ...) => (MOVHreg ...)
   174(SignExt32to64 ...) => (MOVWreg ...)
   175
   176// float <=> int conversion
   177(Cvt32to32F ...) => (MOVWF ...)
   178(Cvt32to64F ...) => (MOVWD ...)
   179(Cvt64to32F ...) => (MOVVF ...)
   180(Cvt64to64F ...) => (MOVVD ...)
   181(Cvt32Fto32 ...) => (TRUNCFW ...)
   182(Cvt64Fto32 ...) => (TRUNCDW ...)
   183(Cvt32Fto64 ...) => (TRUNCFV ...)
   184(Cvt64Fto64 ...) => (TRUNCDV ...)
   185(Cvt32Fto64F ...) => (MOVFD ...)
   186(Cvt64Fto32F ...) => (MOVDF ...)
   187
   188(CvtBoolToUint8 ...) => (Copy ...)
   189
   190(Round(32|64)F ...) => (Copy ...)
   191
   192// comparisons
   193(Eq8 x y)  => (SGTU (MOVVconst [1]) (XOR (ZeroExt8to64 x) (ZeroExt8to64 y)))
   194(Eq16 x y) => (SGTU (MOVVconst [1]) (XOR (ZeroExt16to64 x) (ZeroExt16to64 y)))
   195(Eq32 x y) => (SGTU (MOVVconst [1]) (XOR (ZeroExt32to64 x) (ZeroExt32to64 y)))
   196(Eq64 x y) => (SGTU (MOVVconst [1]) (XOR x y))
   197(EqPtr x y) => (SGTU (MOVVconst [1]) (XOR x y))
   198(Eq(32|64)F x y) => (FPFlagTrue (CMPEQ(F|D) x y))
   199
   200(Neq8 x y)  => (SGTU (XOR (ZeroExt8to64 x) (ZeroExt8to64 y)) (MOVVconst [0]))
   201(Neq16 x y) => (SGTU (XOR (ZeroExt16to32 x) (ZeroExt16to64 y)) (MOVVconst [0]))
   202(Neq32 x y) => (SGTU (XOR (ZeroExt32to64 x) (ZeroExt32to64 y)) (MOVVconst [0]))
   203(Neq64 x y) => (SGTU (XOR x y) (MOVVconst [0]))
   204(NeqPtr x y) => (SGTU (XOR x y) (MOVVconst [0]))
   205(Neq(32|64)F x y) => (FPFlagFalse (CMPEQ(F|D) x y))
   206
   207(Less8 x y)  => (SGT (SignExt8to64 y) (SignExt8to64 x))
   208(Less16 x y) => (SGT (SignExt16to64 y) (SignExt16to64 x))
   209(Less32 x y) => (SGT (SignExt32to64 y) (SignExt32to64 x))
   210(Less64 x y) => (SGT y x)
   211(Less(32|64)F x y) => (FPFlagTrue (CMPGT(F|D) y x)) // reverse operands to work around NaN
   212
   213(Less8U x y)  => (SGTU (ZeroExt8to64 y) (ZeroExt8to64 x))
   214(Less16U x y) => (SGTU (ZeroExt16to64 y) (ZeroExt16to64 x))
   215(Less32U x y) => (SGTU (ZeroExt32to64 y) (ZeroExt32to64 x))
   216(Less64U x y) => (SGTU y x)
   217
   218(Leq8 x y)  => (XOR (MOVVconst [1]) (SGT (SignExt8to64 x) (SignExt8to64 y)))
   219(Leq16 x y) => (XOR (MOVVconst [1]) (SGT (SignExt16to64 x) (SignExt16to64 y)))
   220(Leq32 x y) => (XOR (MOVVconst [1]) (SGT (SignExt32to64 x) (SignExt32to64 y)))
   221(Leq64 x y) => (XOR (MOVVconst [1]) (SGT x y))
   222(Leq(32|64)F x y) => (FPFlagTrue (CMPGE(F|D) y x)) // reverse operands to work around NaN
   223
   224(Leq8U x y)  => (XOR (MOVVconst [1]) (SGTU (ZeroExt8to64 x) (ZeroExt8to64 y)))
   225(Leq16U x y) => (XOR (MOVVconst [1]) (SGTU (ZeroExt16to64 x) (ZeroExt16to64 y)))
   226(Leq32U x y) => (XOR (MOVVconst [1]) (SGTU (ZeroExt32to64 x) (ZeroExt32to64 y)))
   227(Leq64U x y) => (XOR (MOVVconst [1]) (SGTU x y))
   228
   229(OffPtr [off] ptr:(SP)) && is32Bit(off) => (MOVVaddr [int32(off)] ptr)
   230(OffPtr [off] ptr) => (ADDVconst [off] ptr)
   231
   232(Addr {sym} base) => (MOVVaddr {sym} base)
   233(LocalAddr <t> {sym} base mem) && t.Elem().HasPointers() => (MOVVaddr {sym} (SPanchored base mem))
   234(LocalAddr <t> {sym} base _)  && !t.Elem().HasPointers() => (MOVVaddr {sym} base)
   235
   236// loads
   237(Load <t> ptr mem) && t.IsBoolean() => (MOVBUload ptr mem)
   238(Load <t> ptr mem) && (is8BitInt(t)  &&  t.IsSigned()) => (MOVBload ptr mem)
   239(Load <t> ptr mem) && (is8BitInt(t)  && !t.IsSigned()) => (MOVBUload ptr mem)
   240(Load <t> ptr mem) && (is16BitInt(t) &&  t.IsSigned()) => (MOVHload ptr mem)
   241(Load <t> ptr mem) && (is16BitInt(t) && !t.IsSigned()) => (MOVHUload ptr mem)
   242(Load <t> ptr mem) && (is32BitInt(t) &&  t.IsSigned()) => (MOVWload ptr mem)
   243(Load <t> ptr mem) && (is32BitInt(t) && !t.IsSigned()) => (MOVWUload ptr mem)
   244(Load <t> ptr mem) && (is64BitInt(t) || isPtr(t)) => (MOVVload ptr mem)
   245(Load <t> ptr mem) && is32BitFloat(t) => (MOVFload ptr mem)
   246(Load <t> ptr mem) && is64BitFloat(t) => (MOVDload ptr mem)
   247
   248// stores
   249(Store {t} ptr val mem) && t.Size() == 1 => (MOVBstore ptr val mem)
   250(Store {t} ptr val mem) && t.Size() == 2 => (MOVHstore ptr val mem)
   251(Store {t} ptr val mem) && t.Size() == 4 && !t.IsFloat() => (MOVWstore ptr val mem)
   252(Store {t} ptr val mem) && t.Size() == 8 && !t.IsFloat() => (MOVVstore ptr val mem)
   253(Store {t} ptr val mem) && t.Size() == 4 &&  t.IsFloat() => (MOVFstore ptr val mem)
   254(Store {t} ptr val mem) && t.Size() == 8 &&  t.IsFloat() => (MOVDstore ptr val mem)
   255
   256// zeroing
   257(Zero [0] _ mem) => mem
   258(Zero [1] ptr mem) => (MOVBstore ptr (MOVVconst [0]) mem)
   259(Zero [2] {t} ptr mem) && t.Alignment()%2 == 0 =>
   260	(MOVHstore ptr (MOVVconst [0]) mem)
   261(Zero [2] ptr mem) =>
   262	(MOVBstore [1] ptr (MOVVconst [0])
   263		(MOVBstore [0] ptr (MOVVconst [0]) mem))
   264(Zero [4] {t} ptr mem) && t.Alignment()%4 == 0 =>
   265	(MOVWstore ptr (MOVVconst [0]) mem)
   266(Zero [4] {t} ptr mem) && t.Alignment()%2 == 0 =>
   267	(MOVHstore [2] ptr (MOVVconst [0])
   268		(MOVHstore [0] ptr (MOVVconst [0]) mem))
   269(Zero [4] ptr mem) =>
   270	(MOVBstore [3] ptr (MOVVconst [0])
   271		(MOVBstore [2] ptr (MOVVconst [0])
   272			(MOVBstore [1] ptr (MOVVconst [0])
   273				(MOVBstore [0] ptr (MOVVconst [0]) mem))))
   274(Zero [8] {t} ptr mem) && t.Alignment()%8 == 0 =>
   275	(MOVVstore ptr (MOVVconst [0]) mem)
   276(Zero [8] {t} ptr mem) && t.Alignment()%4 == 0 =>
   277	(MOVWstore [4] ptr (MOVVconst [0])
   278		(MOVWstore [0] ptr (MOVVconst [0]) mem))
   279(Zero [8] {t} ptr mem) && t.Alignment()%2 == 0 =>
   280	(MOVHstore [6] ptr (MOVVconst [0])
   281		(MOVHstore [4] ptr (MOVVconst [0])
   282			(MOVHstore [2] ptr (MOVVconst [0])
   283				(MOVHstore [0] ptr (MOVVconst [0]) mem))))
   284
   285(Zero [3] ptr mem) =>
   286	(MOVBstore [2] ptr (MOVVconst [0])
   287		(MOVBstore [1] ptr (MOVVconst [0])
   288			(MOVBstore [0] ptr (MOVVconst [0]) mem)))
   289(Zero [6] {t} ptr mem) && t.Alignment()%2 == 0 =>
   290	(MOVHstore [4] ptr (MOVVconst [0])
   291		(MOVHstore [2] ptr (MOVVconst [0])
   292			(MOVHstore [0] ptr (MOVVconst [0]) mem)))
   293(Zero [12] {t} ptr mem) && t.Alignment()%4 == 0 =>
   294	(MOVWstore [8] ptr (MOVVconst [0])
   295		(MOVWstore [4] ptr (MOVVconst [0])
   296			(MOVWstore [0] ptr (MOVVconst [0]) mem)))
   297(Zero [16] {t} ptr mem) && t.Alignment()%8 == 0 =>
   298	(MOVVstore [8] ptr (MOVVconst [0])
   299		(MOVVstore [0] ptr (MOVVconst [0]) mem))
   300(Zero [24] {t} ptr mem) && t.Alignment()%8 == 0 =>
   301	(MOVVstore [16] ptr (MOVVconst [0])
   302		(MOVVstore [8] ptr (MOVVconst [0])
   303			(MOVVstore [0] ptr (MOVVconst [0]) mem)))
   304
   305// medium zeroing uses a duff device
   306// 8, and 128 are magic constants, see runtime/mkduff.go
   307(Zero [s] {t} ptr mem)
   308	&& s%8 == 0 && s > 24 && s <= 8*128
   309	&& t.Alignment()%8 == 0 && !config.noDuffDevice =>
   310	(DUFFZERO [8 * (128 - s/8)] ptr mem)
   311
   312// large or unaligned zeroing uses a loop
   313(Zero [s] {t} ptr mem)
   314	&& (s > 8*128 || config.noDuffDevice) || t.Alignment()%8 != 0 =>
   315	(LoweredZero [t.Alignment()]
   316		ptr
   317		(ADDVconst <ptr.Type> ptr [s-moveSize(t.Alignment(), config)])
   318		mem)
   319
   320// moves
   321(Move [0] _ _ mem) => mem
   322(Move [1] dst src mem) => (MOVBstore dst (MOVBload src mem) mem)
   323(Move [2] {t} dst src mem) && t.Alignment()%2 == 0 =>
   324	(MOVHstore dst (MOVHload src mem) mem)
   325(Move [2] dst src mem) =>
   326	(MOVBstore [1] dst (MOVBload [1] src mem)
   327		(MOVBstore dst (MOVBload src mem) mem))
   328(Move [4] {t} dst src mem) && t.Alignment()%4 == 0 =>
   329	(MOVWstore dst (MOVWload src mem) mem)
   330(Move [4] {t} dst src mem) && t.Alignment()%2 == 0 =>
   331	(MOVHstore [2] dst (MOVHload [2] src mem)
   332		(MOVHstore dst (MOVHload src mem) mem))
   333(Move [4] dst src mem) =>
   334	(MOVBstore [3] dst (MOVBload [3] src mem)
   335		(MOVBstore [2] dst (MOVBload [2] src mem)
   336			(MOVBstore [1] dst (MOVBload [1] src mem)
   337				(MOVBstore dst (MOVBload src mem) mem))))
   338(Move [8] {t} dst src mem) && t.Alignment()%8 == 0 =>
   339	(MOVVstore dst (MOVVload src mem) mem)
   340(Move [8] {t} dst src mem) && t.Alignment()%4 == 0 =>
   341	(MOVWstore [4] dst (MOVWload [4] src mem)
   342		(MOVWstore dst (MOVWload src mem) mem))
   343(Move [8] {t} dst src mem) && t.Alignment()%2 == 0 =>
   344	(MOVHstore [6] dst (MOVHload [6] src mem)
   345		(MOVHstore [4] dst (MOVHload [4] src mem)
   346			(MOVHstore [2] dst (MOVHload [2] src mem)
   347				(MOVHstore dst (MOVHload src mem) mem))))
   348
   349(Move [3] dst src mem) =>
   350	(MOVBstore [2] dst (MOVBload [2] src mem)
   351		(MOVBstore [1] dst (MOVBload [1] src mem)
   352			(MOVBstore dst (MOVBload src mem) mem)))
   353(Move [6] {t} dst src mem) && t.Alignment()%2 == 0 =>
   354	(MOVHstore [4] dst (MOVHload [4] src mem)
   355		(MOVHstore [2] dst (MOVHload [2] src mem)
   356			(MOVHstore dst (MOVHload src mem) mem)))
   357(Move [12] {t} dst src mem) && t.Alignment()%4 == 0 =>
   358	(MOVWstore [8] dst (MOVWload [8] src mem)
   359		(MOVWstore [4] dst (MOVWload [4] src mem)
   360			(MOVWstore dst (MOVWload src mem) mem)))
   361(Move [16] {t} dst src mem) && t.Alignment()%8 == 0 =>
   362	(MOVVstore [8] dst (MOVVload [8] src mem)
   363		(MOVVstore dst (MOVVload src mem) mem))
   364(Move [24] {t} dst src mem) && t.Alignment()%8 == 0 =>
   365	(MOVVstore [16] dst (MOVVload [16] src mem)
   366		(MOVVstore [8] dst (MOVVload [8] src mem)
   367			(MOVVstore dst (MOVVload src mem) mem)))
   368
   369// float <=> int register moves, with no conversion.
   370// These come up when compiling math.{Float64bits, Float64frombits, Float32bits, Float32frombits}.
   371(MOVVload  [off] {sym} ptr (MOVDstore [off] {sym} ptr val _)) => (MOVVfpgp val)
   372(MOVDload  [off] {sym} ptr (MOVVstore [off] {sym} ptr val _)) => (MOVVgpfp val)
   373(MOVWUload [off] {sym} ptr (MOVFstore [off] {sym} ptr val _)) => (ZeroExt32to64 (MOVWfpgp <typ.Float32> val))
   374(MOVFload  [off] {sym} ptr (MOVWstore [off] {sym} ptr val _)) => (MOVWgpfp val)
   375
   376// Similarly for stores, if we see a store after FPR <=> GPR move, then redirect store to use the other register set.
   377(MOVVstore [off] {sym} ptr (MOVVfpgp val) mem) => (MOVDstore [off] {sym} ptr val mem)
   378(MOVDstore [off] {sym} ptr (MOVVgpfp val) mem) => (MOVVstore [off] {sym} ptr val mem)
   379(MOVWstore [off] {sym} ptr (MOVWfpgp val) mem) => (MOVFstore [off] {sym} ptr val mem)
   380(MOVFstore [off] {sym} ptr (MOVWgpfp val) mem) => (MOVWstore [off] {sym} ptr val mem)
   381
   382// medium move uses a duff device
   383(Move [s] {t} dst src mem)
   384	&& s%8 == 0 && s >= 24 && s <= 8*128 && t.Alignment()%8 == 0
   385	&& !config.noDuffDevice && logLargeCopy(v, s)  =>
   386	(DUFFCOPY [16 * (128 - s/8)] dst src mem)
   387// 16 and 128 are magic constants.  16 is the number of bytes to encode:
   388//	MOVV	(R1), R23
   389//	ADDV	$8, R1
   390//	MOVV	R23, (R2)
   391//	ADDV	$8, R2
   392// and 128 is the number of such blocks. See runtime/duff_mips64.s:duffcopy.
   393
   394// large or unaligned move uses a loop
   395(Move [s] {t} dst src mem)
   396	&& s > 24 && logLargeCopy(v, s) || t.Alignment()%8 != 0 =>
   397	(LoweredMove [t.Alignment()]
   398		dst
   399		src
   400		(ADDVconst <src.Type> src [s-moveSize(t.Alignment(), config)])
   401		mem)
   402
   403// calls
   404(StaticCall ...) => (CALLstatic ...)
   405(ClosureCall ...) => (CALLclosure ...)
   406(InterCall ...) => (CALLinter ...)
   407(TailCall ...) => (CALLtail ...)
   408
   409// atomic intrinsics
   410(AtomicLoad(8|32|64) ...) => (LoweredAtomicLoad(8|32|64) ...)
   411(AtomicLoadPtr ...) => (LoweredAtomicLoad64 ...)
   412
   413(AtomicStore(8|32|64) ...) => (LoweredAtomicStore(8|32|64) ...)
   414(AtomicStorePtrNoWB ...) => (LoweredAtomicStore64 ...)
   415
   416(AtomicExchange(32|64) ...) => (LoweredAtomicExchange(32|64) ...)
   417
   418(AtomicAdd(32|64) ...) => (LoweredAtomicAdd(32|64) ...)
   419
   420(AtomicCompareAndSwap32 ptr old new mem) => (LoweredAtomicCas32 ptr (SignExt32to64 old) new mem)
   421(AtomicCompareAndSwap64 ...) => (LoweredAtomicCas64 ...)
   422
   423// AtomicOr8(ptr,val)  =>   LoweredAtomicOr32(ptr&^3,uint32(val) << ((ptr & 3) * 8))
   424(AtomicOr8 ptr val mem) && !config.BigEndian =>
   425	(LoweredAtomicOr32 (AND <typ.UInt32Ptr> (MOVVconst [^3]) ptr)
   426		(SLLV <typ.UInt32> (ZeroExt8to32 val)
   427			(SLLVconst <typ.UInt64> [3]
   428				(ANDconst <typ.UInt64> [3] ptr))) mem)
   429
   430// AtomicAnd8(ptr,val)  =>  LoweredAtomicAnd32(ptr&^3,(uint32(val) << ((ptr & 3) * 8)) | ^(uint32(0xFF) << ((ptr & 3) * 8))))
   431(AtomicAnd8  ptr val mem) && !config.BigEndian =>
   432	(LoweredAtomicAnd32 (AND <typ.UInt32Ptr> (MOVVconst [^3]) ptr)
   433		(OR <typ.UInt64> (SLLV <typ.UInt32> (ZeroExt8to32 val)
   434			(SLLVconst <typ.UInt64> [3]
   435				(ANDconst  <typ.UInt64> [3] ptr)))
   436		(NORconst [0] <typ.UInt64> (SLLV <typ.UInt64>
   437			(MOVVconst [0xff]) (SLLVconst <typ.UInt64> [3]
   438				(ANDconst <typ.UInt64> [3] ptr))))) mem)
   439
   440// AtomicOr8(ptr,val)  =>  LoweredAtomicOr32(ptr&^3,uint32(val) << (((ptr^3) & 3) * 8))
   441(AtomicOr8 ptr val mem) && config.BigEndian =>
   442	(LoweredAtomicOr32 (AND <typ.UInt32Ptr> (MOVVconst [^3]) ptr)
   443		(SLLV <typ.UInt32> (ZeroExt8to32 val)
   444			(SLLVconst <typ.UInt64> [3]
   445				(ANDconst <typ.UInt64> [3]
   446					(XORconst <typ.UInt64> [3] ptr)))) mem)
   447
   448// AtomicAnd8(ptr,val)  =>  LoweredAtomicAnd32(ptr&^3,(uint32(val) << (((ptr^3) & 3) * 8)) | ^(uint32(0xFF) << (((ptr^3) & 3) * 8))))
   449(AtomicAnd8  ptr val mem) && config.BigEndian =>
   450	(LoweredAtomicAnd32 (AND <typ.UInt32Ptr> (MOVVconst [^3]) ptr)
   451		(OR <typ.UInt64> (SLLV <typ.UInt32> (ZeroExt8to32 val)
   452			(SLLVconst <typ.UInt64> [3]
   453				(ANDconst  <typ.UInt64> [3]
   454					(XORconst <typ.UInt64> [3] ptr))))
   455		(NORconst [0] <typ.UInt64> (SLLV <typ.UInt64>
   456			(MOVVconst [0xff]) (SLLVconst <typ.UInt64> [3]
   457				(ANDconst <typ.UInt64> [3]
   458					(XORconst <typ.UInt64> [3] ptr)))))) mem)
   459
   460(AtomicAnd32 ...) => (LoweredAtomicAnd32 ...)
   461(AtomicOr32  ...) => (LoweredAtomicOr32  ...)
   462
   463// checks
   464(NilCheck ...) => (LoweredNilCheck ...)
   465(IsNonNil ptr) => (SGTU ptr (MOVVconst [0]))
   466(IsInBounds idx len) => (SGTU len idx)
   467(IsSliceInBounds idx len) => (XOR (MOVVconst [1]) (SGTU idx len))
   468
   469// pseudo-ops
   470(GetClosurePtr ...) => (LoweredGetClosurePtr ...)
   471(GetCallerSP ...) => (LoweredGetCallerSP ...)
   472(GetCallerPC ...) => (LoweredGetCallerPC ...)
   473
   474(If cond yes no) => (NE cond yes no)
   475
   476// Write barrier.
   477(WB ...) => (LoweredWB ...)
   478
   479(PanicBounds [kind] x y mem) && boundsABI(kind) == 0 => (LoweredPanicBoundsA [kind] x y mem)
   480(PanicBounds [kind] x y mem) && boundsABI(kind) == 1 => (LoweredPanicBoundsB [kind] x y mem)
   481(PanicBounds [kind] x y mem) && boundsABI(kind) == 2 => (LoweredPanicBoundsC [kind] x y mem)
   482
   483// Optimizations
   484
   485// Absorb boolean tests into block
   486(NE (FPFlagTrue cmp) yes no) => (FPT cmp yes no)
   487(NE (FPFlagFalse cmp) yes no) => (FPF cmp yes no)
   488(EQ (FPFlagTrue cmp) yes no) => (FPF cmp yes no)
   489(EQ (FPFlagFalse cmp) yes no) => (FPT cmp yes no)
   490(NE (XORconst [1] cmp:(SGT _ _)) yes no) => (EQ cmp yes no)
   491(NE (XORconst [1] cmp:(SGTU _ _)) yes no) => (EQ cmp yes no)
   492(NE (XORconst [1] cmp:(SGTconst _)) yes no) => (EQ cmp yes no)
   493(NE (XORconst [1] cmp:(SGTUconst _)) yes no) => (EQ cmp yes no)
   494(EQ (XORconst [1] cmp:(SGT _ _)) yes no) => (NE cmp yes no)
   495(EQ (XORconst [1] cmp:(SGTU _ _)) yes no) => (NE cmp yes no)
   496(EQ (XORconst [1] cmp:(SGTconst _)) yes no) => (NE cmp yes no)
   497(EQ (XORconst [1] cmp:(SGTUconst _)) yes no) => (NE cmp yes no)
   498(NE (SGTUconst [1] x) yes no) => (EQ x yes no)
   499(EQ (SGTUconst [1] x) yes no) => (NE x yes no)
   500(NE (SGTU x (MOVVconst [0])) yes no) => (NE x yes no)
   501(EQ (SGTU x (MOVVconst [0])) yes no) => (EQ x yes no)
   502(NE (SGTconst [0] x) yes no) => (LTZ x yes no)
   503(EQ (SGTconst [0] x) yes no) => (GEZ x yes no)
   504(NE (SGT x (MOVVconst [0])) yes no) => (GTZ x yes no)
   505(EQ (SGT x (MOVVconst [0])) yes no) => (LEZ x yes no)
   506
   507// fold offset into address
   508(ADDVconst [off1] (MOVVaddr [off2] {sym} ptr)) && is32Bit(off1+int64(off2)) => (MOVVaddr [int32(off1)+int32(off2)] {sym} ptr)
   509
   510// fold address into load/store
   511(MOVBload  [off1] {sym} (ADDVconst [off2] ptr) mem) && is32Bit(int64(off1)+off2)
   512	 && (ptr.Op != OpSB || !config.ctxt.Flag_shared) => (MOVBload  [off1+int32(off2)] {sym} ptr mem)
   513(MOVBUload [off1] {sym} (ADDVconst [off2] ptr) mem) && is32Bit(int64(off1)+off2)
   514	 && (ptr.Op != OpSB || !config.ctxt.Flag_shared) => (MOVBUload [off1+int32(off2)] {sym} ptr mem)
   515(MOVHload  [off1] {sym} (ADDVconst [off2] ptr) mem) && is32Bit(int64(off1)+off2)
   516	 && (ptr.Op != OpSB || !config.ctxt.Flag_shared) => (MOVHload  [off1+int32(off2)] {sym} ptr mem)
   517(MOVHUload [off1] {sym} (ADDVconst [off2] ptr) mem) && is32Bit(int64(off1)+off2)
   518	 && (ptr.Op != OpSB || !config.ctxt.Flag_shared) => (MOVHUload [off1+int32(off2)] {sym} ptr mem)
   519(MOVWload  [off1] {sym} (ADDVconst [off2] ptr) mem) && is32Bit(int64(off1)+off2)
   520	 && (ptr.Op != OpSB || !config.ctxt.Flag_shared) => (MOVWload  [off1+int32(off2)] {sym} ptr mem)
   521(MOVWUload [off1] {sym} (ADDVconst [off2] ptr) mem) && is32Bit(int64(off1)+off2)
   522	 && (ptr.Op != OpSB || !config.ctxt.Flag_shared) => (MOVWUload [off1+int32(off2)] {sym} ptr mem)
   523(MOVVload  [off1] {sym} (ADDVconst [off2] ptr) mem) && is32Bit(int64(off1)+off2)
   524	 && (ptr.Op != OpSB || !config.ctxt.Flag_shared) => (MOVVload  [off1+int32(off2)] {sym} ptr mem)
   525(MOVFload  [off1] {sym} (ADDVconst [off2] ptr) mem) && is32Bit(int64(off1)+off2)
   526	 && (ptr.Op != OpSB || !config.ctxt.Flag_shared) => (MOVFload  [off1+int32(off2)] {sym} ptr mem)
   527(MOVDload  [off1] {sym} (ADDVconst [off2] ptr) mem) && is32Bit(int64(off1)+off2)
   528	 && (ptr.Op != OpSB || !config.ctxt.Flag_shared) => (MOVDload  [off1+int32(off2)] {sym} ptr mem)
   529
   530(MOVBstore [off1] {sym} (ADDVconst [off2] ptr) val mem) && is32Bit(int64(off1)+off2)
   531	 && (ptr.Op != OpSB || !config.ctxt.Flag_shared) => (MOVBstore [off1+int32(off2)] {sym} ptr val mem)
   532(MOVHstore [off1] {sym} (ADDVconst [off2] ptr) val mem) && is32Bit(int64(off1)+off2)
   533	 && (ptr.Op != OpSB || !config.ctxt.Flag_shared) => (MOVHstore [off1+int32(off2)] {sym} ptr val mem)
   534(MOVWstore [off1] {sym} (ADDVconst [off2] ptr) val mem) && is32Bit(int64(off1)+off2)
   535	 && (ptr.Op != OpSB || !config.ctxt.Flag_shared) => (MOVWstore [off1+int32(off2)] {sym} ptr val mem)
   536(MOVVstore [off1] {sym} (ADDVconst [off2] ptr) val mem) && is32Bit(int64(off1)+off2)
   537	 && (ptr.Op != OpSB || !config.ctxt.Flag_shared) => (MOVVstore [off1+int32(off2)] {sym} ptr val mem)
   538(MOVFstore [off1] {sym} (ADDVconst [off2] ptr) val mem) && is32Bit(int64(off1)+off2)
   539	 && (ptr.Op != OpSB || !config.ctxt.Flag_shared) => (MOVFstore [off1+int32(off2)] {sym} ptr val mem)
   540(MOVDstore [off1] {sym} (ADDVconst [off2] ptr) val mem) && is32Bit(int64(off1)+off2)
   541	 && (ptr.Op != OpSB || !config.ctxt.Flag_shared) => (MOVDstore [off1+int32(off2)] {sym} ptr val mem)
   542(MOVBstorezero [off1] {sym} (ADDVconst [off2] ptr) mem) && is32Bit(int64(off1)+off2)
   543	 && (ptr.Op != OpSB || !config.ctxt.Flag_shared) => (MOVBstorezero [off1+int32(off2)] {sym} ptr mem)
   544(MOVHstorezero [off1] {sym} (ADDVconst [off2] ptr) mem) && is32Bit(int64(off1)+off2)
   545	 && (ptr.Op != OpSB || !config.ctxt.Flag_shared) => (MOVHstorezero [off1+int32(off2)] {sym} ptr mem)
   546(MOVWstorezero [off1] {sym} (ADDVconst [off2] ptr) mem) && is32Bit(int64(off1)+off2)
   547	 && (ptr.Op != OpSB || !config.ctxt.Flag_shared) => (MOVWstorezero [off1+int32(off2)] {sym} ptr mem)
   548(MOVVstorezero [off1] {sym} (ADDVconst [off2] ptr) mem) && is32Bit(int64(off1)+off2)
   549	 && (ptr.Op != OpSB || !config.ctxt.Flag_shared) => (MOVVstorezero [off1+int32(off2)] {sym} ptr mem)
   550
   551(MOVBload [off1] {sym1} (MOVVaddr [off2] {sym2} ptr) mem)
   552	&& canMergeSym(sym1,sym2) && is32Bit(int64(off1)+int64(off2))
   553	&& (ptr.Op != OpSB || !config.ctxt.Flag_shared) =>
   554	(MOVBload [off1+int32(off2)] {mergeSym(sym1,sym2)} ptr mem)
   555(MOVBUload [off1] {sym1} (MOVVaddr [off2] {sym2} ptr) mem)
   556	&& canMergeSym(sym1,sym2) && is32Bit(int64(off1)+int64(off2))
   557	&& (ptr.Op != OpSB || !config.ctxt.Flag_shared) =>
   558	(MOVBUload [off1+int32(off2)] {mergeSym(sym1,sym2)} ptr mem)
   559(MOVHload [off1] {sym1} (MOVVaddr [off2] {sym2} ptr) mem)
   560	&& canMergeSym(sym1,sym2) && is32Bit(int64(off1)+int64(off2))
   561	&& (ptr.Op != OpSB || !config.ctxt.Flag_shared) =>
   562	(MOVHload [off1+int32(off2)] {mergeSym(sym1,sym2)} ptr mem)
   563(MOVHUload [off1] {sym1} (MOVVaddr [off2] {sym2} ptr) mem)
   564	&& canMergeSym(sym1,sym2) && is32Bit(int64(off1)+int64(off2))
   565	&& (ptr.Op != OpSB || !config.ctxt.Flag_shared) =>
   566	(MOVHUload [off1+int32(off2)] {mergeSym(sym1,sym2)} ptr mem)
   567(MOVWload [off1] {sym1} (MOVVaddr [off2] {sym2} ptr) mem)
   568	&& canMergeSym(sym1,sym2) && is32Bit(int64(off1)+int64(off2))
   569	&& (ptr.Op != OpSB || !config.ctxt.Flag_shared) =>
   570	(MOVWload [off1+int32(off2)] {mergeSym(sym1,sym2)} ptr mem)
   571(MOVWUload [off1] {sym1} (MOVVaddr [off2] {sym2} ptr) mem)
   572	&& canMergeSym(sym1,sym2) && is32Bit(int64(off1)+int64(off2))
   573	&& (ptr.Op != OpSB || !config.ctxt.Flag_shared) =>
   574	(MOVWUload [off1+int32(off2)] {mergeSym(sym1,sym2)} ptr mem)
   575(MOVVload [off1] {sym1} (MOVVaddr [off2] {sym2} ptr) mem)
   576	&& canMergeSym(sym1,sym2) && is32Bit(int64(off1)+int64(off2))
   577	&& (ptr.Op != OpSB || !config.ctxt.Flag_shared) =>
   578	(MOVVload [off1+int32(off2)] {mergeSym(sym1,sym2)} ptr mem)
   579(MOVFload [off1] {sym1} (MOVVaddr [off2] {sym2} ptr) mem)
   580	&& canMergeSym(sym1,sym2) && is32Bit(int64(off1)+int64(off2))
   581	&& (ptr.Op != OpSB || !config.ctxt.Flag_shared) =>
   582	(MOVFload [off1+int32(off2)] {mergeSym(sym1,sym2)} ptr mem)
   583(MOVDload [off1] {sym1} (MOVVaddr [off2] {sym2} ptr) mem)
   584	&& canMergeSym(sym1,sym2) && is32Bit(int64(off1)+int64(off2))
   585	&& (ptr.Op != OpSB || !config.ctxt.Flag_shared) =>
   586	(MOVDload [off1+int32(off2)] {mergeSym(sym1,sym2)} ptr mem)
   587
   588(MOVBstore [off1] {sym1} (MOVVaddr [off2] {sym2} ptr) val mem)
   589	&& canMergeSym(sym1,sym2) && is32Bit(int64(off1)+int64(off2))
   590	&& (ptr.Op != OpSB || !config.ctxt.Flag_shared) =>
   591	(MOVBstore [off1+int32(off2)] {mergeSym(sym1,sym2)} ptr val mem)
   592(MOVHstore [off1] {sym1} (MOVVaddr [off2] {sym2} ptr) val mem)
   593	&& canMergeSym(sym1,sym2) && is32Bit(int64(off1)+int64(off2))
   594	&& (ptr.Op != OpSB || !config.ctxt.Flag_shared) =>
   595	(MOVHstore [off1+int32(off2)] {mergeSym(sym1,sym2)} ptr val mem)
   596(MOVWstore [off1] {sym1} (MOVVaddr [off2] {sym2} ptr) val mem)
   597	&& canMergeSym(sym1,sym2) && is32Bit(int64(off1)+int64(off2))
   598	&& (ptr.Op != OpSB || !config.ctxt.Flag_shared) =>
   599	(MOVWstore [off1+int32(off2)] {mergeSym(sym1,sym2)} ptr val mem)
   600(MOVVstore [off1] {sym1} (MOVVaddr [off2] {sym2} ptr) val mem)
   601	&& canMergeSym(sym1,sym2) && is32Bit(int64(off1)+int64(off2))
   602       	&& (ptr.Op != OpSB || !config.ctxt.Flag_shared) =>
   603	(MOVVstore [off1+int32(off2)] {mergeSym(sym1,sym2)} ptr val mem)
   604(MOVFstore [off1] {sym1} (MOVVaddr [off2] {sym2} ptr) val mem)
   605	&& canMergeSym(sym1,sym2) && is32Bit(int64(off1)+int64(off2))
   606	&& (ptr.Op != OpSB || !config.ctxt.Flag_shared) =>
   607	(MOVFstore [off1+int32(off2)] {mergeSym(sym1,sym2)} ptr val mem)
   608(MOVDstore [off1] {sym1} (MOVVaddr [off2] {sym2} ptr) val mem)
   609	&& canMergeSym(sym1,sym2) && is32Bit(int64(off1)+int64(off2))
   610	&& (ptr.Op != OpSB || !config.ctxt.Flag_shared) =>
   611	(MOVDstore [off1+int32(off2)] {mergeSym(sym1,sym2)} ptr val mem)
   612(MOVBstorezero [off1] {sym1} (MOVVaddr [off2] {sym2} ptr) mem)
   613	&& canMergeSym(sym1,sym2) && is32Bit(int64(off1)+int64(off2))
   614	&& (ptr.Op != OpSB || !config.ctxt.Flag_shared) =>
   615	(MOVBstorezero [off1+int32(off2)] {mergeSym(sym1,sym2)} ptr mem)
   616(MOVHstorezero [off1] {sym1} (MOVVaddr [off2] {sym2} ptr) mem)
   617	&& canMergeSym(sym1,sym2) && is32Bit(int64(off1)+int64(off2))
   618	&& (ptr.Op != OpSB || !config.ctxt.Flag_shared) =>
   619	(MOVHstorezero [off1+int32(off2)] {mergeSym(sym1,sym2)} ptr mem)
   620(MOVWstorezero [off1] {sym1} (MOVVaddr [off2] {sym2} ptr) mem)
   621	&& canMergeSym(sym1,sym2) && is32Bit(int64(off1)+int64(off2))
   622	&& (ptr.Op != OpSB || !config.ctxt.Flag_shared) =>
   623	(MOVWstorezero [off1+int32(off2)] {mergeSym(sym1,sym2)} ptr mem)
   624(MOVVstorezero [off1] {sym1} (MOVVaddr [off2] {sym2} ptr) mem)
   625	&& canMergeSym(sym1,sym2) && is32Bit(int64(off1)+int64(off2))
   626	&& (ptr.Op != OpSB || !config.ctxt.Flag_shared) =>
   627	(MOVVstorezero [off1+int32(off2)] {mergeSym(sym1,sym2)} ptr mem)
   628
   629// store zero
   630(MOVBstore [off] {sym} ptr (MOVVconst [0]) mem) => (MOVBstorezero [off] {sym} ptr mem)
   631(MOVHstore [off] {sym} ptr (MOVVconst [0]) mem) => (MOVHstorezero [off] {sym} ptr mem)
   632(MOVWstore [off] {sym} ptr (MOVVconst [0]) mem) => (MOVWstorezero [off] {sym} ptr mem)
   633(MOVVstore [off] {sym} ptr (MOVVconst [0]) mem) => (MOVVstorezero [off] {sym} ptr mem)
   634
   635// don't extend after proper load
   636(MOVBreg x:(MOVBload _ _)) => (MOVVreg x)
   637(MOVBUreg x:(MOVBUload _ _)) => (MOVVreg x)
   638(MOVHreg x:(MOVBload _ _)) => (MOVVreg x)
   639(MOVHreg x:(MOVBUload _ _)) => (MOVVreg x)
   640(MOVHreg x:(MOVHload _ _)) => (MOVVreg x)
   641(MOVHUreg x:(MOVBUload _ _)) => (MOVVreg x)
   642(MOVHUreg x:(MOVHUload _ _)) => (MOVVreg x)
   643(MOVWreg x:(MOVBload _ _)) => (MOVVreg x)
   644(MOVWreg x:(MOVBUload _ _)) => (MOVVreg x)
   645(MOVWreg x:(MOVHload _ _)) => (MOVVreg x)
   646(MOVWreg x:(MOVHUload _ _)) => (MOVVreg x)
   647(MOVWreg x:(MOVWload _ _)) => (MOVVreg x)
   648(MOVWUreg x:(MOVBUload _ _)) => (MOVVreg x)
   649(MOVWUreg x:(MOVHUload _ _)) => (MOVVreg x)
   650(MOVWUreg x:(MOVWUload _ _)) => (MOVVreg x)
   651
   652// fold double extensions
   653(MOVBreg x:(MOVBreg _)) => (MOVVreg x)
   654(MOVBUreg x:(MOVBUreg _)) => (MOVVreg x)
   655(MOVHreg x:(MOVBreg _)) => (MOVVreg x)
   656(MOVHreg x:(MOVBUreg _)) => (MOVVreg x)
   657(MOVHreg x:(MOVHreg _)) => (MOVVreg x)
   658(MOVHUreg x:(MOVBUreg _)) => (MOVVreg x)
   659(MOVHUreg x:(MOVHUreg _)) => (MOVVreg x)
   660(MOVWreg x:(MOVBreg _)) => (MOVVreg x)
   661(MOVWreg x:(MOVBUreg _)) => (MOVVreg x)
   662(MOVWreg x:(MOVHreg _)) => (MOVVreg x)
   663(MOVWreg x:(MOVWreg _)) => (MOVVreg x)
   664(MOVWUreg x:(MOVBUreg _)) => (MOVVreg x)
   665(MOVWUreg x:(MOVHUreg _)) => (MOVVreg x)
   666(MOVWUreg x:(MOVWUreg _)) => (MOVVreg x)
   667
   668// don't extend before store
   669(MOVBstore [off] {sym} ptr (MOVBreg x) mem) => (MOVBstore [off] {sym} ptr x mem)
   670(MOVBstore [off] {sym} ptr (MOVBUreg x) mem) => (MOVBstore [off] {sym} ptr x mem)
   671(MOVBstore [off] {sym} ptr (MOVHreg x) mem) => (MOVBstore [off] {sym} ptr x mem)
   672(MOVBstore [off] {sym} ptr (MOVHUreg x) mem) => (MOVBstore [off] {sym} ptr x mem)
   673(MOVBstore [off] {sym} ptr (MOVWreg x) mem) => (MOVBstore [off] {sym} ptr x mem)
   674(MOVBstore [off] {sym} ptr (MOVWUreg x) mem) => (MOVBstore [off] {sym} ptr x mem)
   675(MOVHstore [off] {sym} ptr (MOVHreg x) mem) => (MOVHstore [off] {sym} ptr x mem)
   676(MOVHstore [off] {sym} ptr (MOVHUreg x) mem) => (MOVHstore [off] {sym} ptr x mem)
   677(MOVHstore [off] {sym} ptr (MOVWreg x) mem) => (MOVHstore [off] {sym} ptr x mem)
   678(MOVHstore [off] {sym} ptr (MOVWUreg x) mem) => (MOVHstore [off] {sym} ptr x mem)
   679(MOVWstore [off] {sym} ptr (MOVWreg x) mem) => (MOVWstore [off] {sym} ptr x mem)
   680(MOVWstore [off] {sym} ptr (MOVWUreg x) mem) => (MOVWstore [off] {sym} ptr x mem)
   681
   682// if a register move has only 1 use, just use the same register without emitting instruction
   683// MOVVnop doesn't emit instruction, only for ensuring the type.
   684(MOVVreg x) && x.Uses == 1 => (MOVVnop x)
   685
   686// TODO: we should be able to get rid of MOVVnop all together.
   687// But for now, this is enough to get rid of lots of them.
   688(MOVVnop (MOVVconst [c])) => (MOVVconst [c])
   689
   690// fold constant into arithmetic ops
   691(ADDV x (MOVVconst <t> [c])) && is32Bit(c) && !t.IsPtr() => (ADDVconst [c] x)
   692(SUBV x (MOVVconst [c])) && is32Bit(c) => (SUBVconst [c] x)
   693(AND x (MOVVconst [c])) && is32Bit(c) => (ANDconst [c] x)
   694(OR  x (MOVVconst [c])) && is32Bit(c) => (ORconst  [c] x)
   695(XOR x (MOVVconst [c])) && is32Bit(c) => (XORconst [c] x)
   696(NOR x (MOVVconst [c])) && is32Bit(c) => (NORconst [c] x)
   697
   698(SLLV _ (MOVVconst [c])) && uint64(c)>=64 => (MOVVconst [0])
   699(SRLV _ (MOVVconst [c])) && uint64(c)>=64 => (MOVVconst [0])
   700(SRAV x (MOVVconst [c])) && uint64(c)>=64 => (SRAVconst x [63])
   701(SLLV x (MOVVconst [c])) => (SLLVconst x [c])
   702(SRLV x (MOVVconst [c])) => (SRLVconst x [c])
   703(SRAV x (MOVVconst [c])) => (SRAVconst x [c])
   704
   705(SGT  (MOVVconst [c]) x) && is32Bit(c) => (SGTconst  [c] x)
   706(SGTU (MOVVconst [c]) x) && is32Bit(c) => (SGTUconst [c] x)
   707
   708// mul by constant
   709(Select1 (MULVU x (MOVVconst [-1]))) => (NEGV x)
   710(Select1 (MULVU _ (MOVVconst [0]))) => (MOVVconst [0])
   711(Select1 (MULVU x (MOVVconst [1]))) => x
   712(Select1 (MULVU x (MOVVconst [c]))) && isPowerOfTwo64(c) => (SLLVconst [log64(c)] x)
   713
   714// div by constant
   715(Select1 (DIVVU x (MOVVconst [1]))) => x
   716(Select1 (DIVVU x (MOVVconst [c]))) && isPowerOfTwo64(c) => (SRLVconst [log64(c)] x)
   717(Select0 (DIVVU _ (MOVVconst [1]))) => (MOVVconst [0])                       // mod
   718(Select0 (DIVVU x (MOVVconst [c]))) && isPowerOfTwo64(c) => (ANDconst [c-1] x) // mod
   719
   720// generic simplifications
   721(ADDV x (NEGV y)) => (SUBV x y)
   722(SUBV x x) => (MOVVconst [0])
   723(SUBV (MOVVconst [0]) x) => (NEGV x)
   724(AND x x) => x
   725(OR  x x) => x
   726(XOR x x) => (MOVVconst [0])
   727
   728// remove redundant *const ops
   729(ADDVconst [0]  x) => x
   730(SUBVconst [0]  x) => x
   731(ANDconst [0]  _) => (MOVVconst [0])
   732(ANDconst [-1] x) => x
   733(ORconst  [0]  x) => x
   734(ORconst  [-1] _) => (MOVVconst [-1])
   735(XORconst [0]  x) => x
   736(XORconst [-1] x) => (NORconst [0] x)
   737
   738// generic constant folding
   739(ADDVconst [c] (MOVVconst [d]))  => (MOVVconst [c+d])
   740(ADDVconst [c] (ADDVconst [d] x)) && is32Bit(c+d) => (ADDVconst [c+d] x)
   741(ADDVconst [c] (SUBVconst [d] x)) && is32Bit(c-d) => (ADDVconst [c-d] x)
   742(SUBVconst [c] (MOVVconst [d]))  => (MOVVconst [d-c])
   743(SUBVconst [c] (SUBVconst [d] x)) && is32Bit(-c-d) => (ADDVconst [-c-d] x)
   744(SUBVconst [c] (ADDVconst [d] x)) && is32Bit(-c+d) => (ADDVconst [-c+d] x)
   745(SLLVconst [c] (MOVVconst [d]))  => (MOVVconst [d<<uint64(c)])
   746(SRLVconst [c] (MOVVconst [d]))  => (MOVVconst [int64(uint64(d)>>uint64(c))])
   747(SRAVconst [c] (MOVVconst [d]))  => (MOVVconst [d>>uint64(c)])
   748(Select1 (MULVU (MOVVconst [c]) (MOVVconst [d]))) => (MOVVconst [c*d])
   749(Select1 (DIVV  (MOVVconst [c]) (MOVVconst [d]))) && d != 0 => (MOVVconst [c/d])
   750(Select1 (DIVVU (MOVVconst [c]) (MOVVconst [d]))) && d != 0 => (MOVVconst [int64(uint64(c)/uint64(d))])
   751(Select0 (DIVV  (MOVVconst [c]) (MOVVconst [d]))) && d != 0 => (MOVVconst [c%d])   // mod
   752(Select0 (DIVVU (MOVVconst [c]) (MOVVconst [d]))) && d != 0 => (MOVVconst [int64(uint64(c)%uint64(d))]) // mod
   753(ANDconst [c] (MOVVconst [d])) => (MOVVconst [c&d])
   754(ANDconst [c] (ANDconst [d] x)) => (ANDconst [c&d] x)
   755(ORconst [c] (MOVVconst [d])) => (MOVVconst [c|d])
   756(ORconst [c] (ORconst [d] x)) && is32Bit(c|d) => (ORconst [c|d] x)
   757(XORconst [c] (MOVVconst [d])) => (MOVVconst [c^d])
   758(XORconst [c] (XORconst [d] x)) && is32Bit(c^d) => (XORconst [c^d] x)
   759(NORconst [c] (MOVVconst [d])) => (MOVVconst [^(c|d)])
   760(NEGV (MOVVconst [c])) => (MOVVconst [-c])
   761(MOVBreg  (MOVVconst [c])) => (MOVVconst [int64(int8(c))])
   762(MOVBUreg (MOVVconst [c])) => (MOVVconst [int64(uint8(c))])
   763(MOVHreg  (MOVVconst [c])) => (MOVVconst [int64(int16(c))])
   764(MOVHUreg (MOVVconst [c])) => (MOVVconst [int64(uint16(c))])
   765(MOVWreg  (MOVVconst [c])) => (MOVVconst [int64(int32(c))])
   766(MOVWUreg (MOVVconst [c])) => (MOVVconst [int64(uint32(c))])
   767(MOVVreg  (MOVVconst [c])) => (MOVVconst [c])
   768(LoweredAtomicStore(32|64) ptr (MOVVconst [0]) mem) => (LoweredAtomicStorezero(32|64) ptr mem)
   769(LoweredAtomicAdd32 ptr (MOVVconst [c]) mem) && is32Bit(c) => (LoweredAtomicAddconst32 [int32(c)] ptr mem)
   770(LoweredAtomicAdd64 ptr (MOVVconst [c]) mem) && is32Bit(c) => (LoweredAtomicAddconst64 [c] ptr mem)
   771
   772// constant comparisons
   773(SGTconst [c] (MOVVconst [d])) && c>d => (MOVVconst [1])
   774(SGTconst [c] (MOVVconst [d])) && c<=d => (MOVVconst [0])
   775(SGTUconst [c] (MOVVconst [d])) && uint64(c)>uint64(d) => (MOVVconst [1])
   776(SGTUconst [c] (MOVVconst [d])) && uint64(c)<=uint64(d) => (MOVVconst [0])
   777
   778// other known comparisons
   779(SGTconst [c] (MOVBreg _)) && 0x7f < c => (MOVVconst [1])
   780(SGTconst [c] (MOVBreg _)) && c <= -0x80 => (MOVVconst [0])
   781(SGTconst [c] (MOVBUreg _)) && 0xff < c => (MOVVconst [1])
   782(SGTconst [c] (MOVBUreg _)) && c < 0 => (MOVVconst [0])
   783(SGTUconst [c] (MOVBUreg _)) && 0xff < uint64(c) => (MOVVconst [1])
   784(SGTconst [c] (MOVHreg _)) && 0x7fff < c => (MOVVconst [1])
   785(SGTconst [c] (MOVHreg _)) && c <= -0x8000 => (MOVVconst [0])
   786(SGTconst [c] (MOVHUreg _)) && 0xffff < c => (MOVVconst [1])
   787(SGTconst [c] (MOVHUreg _)) && c < 0 => (MOVVconst [0])
   788(SGTUconst [c] (MOVHUreg _)) && 0xffff < uint64(c) => (MOVVconst [1])
   789(SGTconst [c] (MOVWUreg _)) && c < 0 => (MOVVconst [0])
   790(SGTconst [c] (ANDconst [m] _)) && 0 <= m && m < c => (MOVVconst [1])
   791(SGTUconst [c] (ANDconst [m] _)) && uint64(m) < uint64(c) => (MOVVconst [1])
   792(SGTconst [c] (SRLVconst _ [d])) && 0 <= c && 0 < d && d <= 63 && 0xffffffffffffffff>>uint64(d) < uint64(c) => (MOVVconst [1])
   793(SGTUconst [c] (SRLVconst _ [d])) && 0 < d && d <= 63 && 0xffffffffffffffff>>uint64(d) < uint64(c) => (MOVVconst [1])
   794
   795// absorb constants into branches
   796(EQ  (MOVVconst [0]) yes no) => (First yes no)
   797(EQ  (MOVVconst [c]) yes no) && c != 0 => (First no yes)
   798(NE  (MOVVconst [0]) yes no) => (First no yes)
   799(NE  (MOVVconst [c]) yes no) && c != 0 => (First yes no)
   800(LTZ (MOVVconst [c]) yes no) && c <  0 => (First yes no)
   801(LTZ (MOVVconst [c]) yes no) && c >= 0 => (First no yes)
   802(LEZ (MOVVconst [c]) yes no) && c <= 0 => (First yes no)
   803(LEZ (MOVVconst [c]) yes no) && c >  0 => (First no yes)
   804(GTZ (MOVVconst [c]) yes no) && c >  0 => (First yes no)
   805(GTZ (MOVVconst [c]) yes no) && c <= 0 => (First no yes)
   806(GEZ (MOVVconst [c]) yes no) && c >= 0 => (First yes no)
   807(GEZ (MOVVconst [c]) yes no) && c <  0 => (First no yes)
   808
   809// SGT/SGTU with known outcomes.
   810(SGT  x x) => (MOVVconst [0])
   811(SGTU x x) => (MOVVconst [0])
   812
   813// fold readonly sym load
   814(MOVBload [off] {sym} (SB) _) && symIsRO(sym) => (MOVVconst [int64(read8(sym, int64(off)))])
   815(MOVHload [off] {sym} (SB) _) && symIsRO(sym) => (MOVVconst [int64(read16(sym, int64(off), config.ctxt.Arch.ByteOrder))])
   816(MOVWload [off] {sym} (SB) _) && symIsRO(sym) => (MOVVconst [int64(read32(sym, int64(off), config.ctxt.Arch.ByteOrder))])
   817(MOVVload [off] {sym} (SB) _) && symIsRO(sym) => (MOVVconst [int64(read64(sym, int64(off), config.ctxt.Arch.ByteOrder))])

View as plain text