...

Text file src/cmd/compile/internal/ssa/_gen/LOONG64.rules

Documentation: cmd/compile/internal/ssa/_gen

     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
    61// left shift
    62(Lsh64x(64|32|16|8) x y) && shiftIsBounded(v) => (SLLV x y)
    63(Lsh32x(64|32|16|8) x y) && shiftIsBounded(v) => (SLL x y)
    64(Lsh16x(64|32|16|8) x y) && shiftIsBounded(v) => (SLLV x y)
    65(Lsh8x(64|32|16|8)  x y) && shiftIsBounded(v) => (SLLV x y)
    66
    67(Lsh64x64 <t> x y) && !shiftIsBounded(v) => (MASKEQZ (SLLV <t> x                y)  (SGTU (MOVVconst <typ.UInt64> [64])                y))
    68(Lsh64x32 <t> x y) && !shiftIsBounded(v) => (MASKEQZ (SLLV <t> x (ZeroExt32to64 y)) (SGTU (MOVVconst <typ.UInt64> [64]) (ZeroExt32to64 y)))
    69(Lsh64x16 <t> x y) && !shiftIsBounded(v) => (MASKEQZ (SLLV <t> x (ZeroExt16to64 y)) (SGTU (MOVVconst <typ.UInt64> [64]) (ZeroExt16to64 y)))
    70(Lsh64x8  <t> x y) && !shiftIsBounded(v) => (MASKEQZ (SLLV <t> x (ZeroExt8to64  y)) (SGTU (MOVVconst <typ.UInt64> [64]) (ZeroExt8to64  y)))
    71
    72(Lsh32x64 <t> x y) && !shiftIsBounded(v) => (MASKEQZ (SLL <t> x                y)  (SGTU (MOVVconst <typ.UInt64> [32])                y))
    73(Lsh32x32 <t> x y) && !shiftIsBounded(v) => (MASKEQZ (SLL <t> x (ZeroExt32to64 y)) (SGTU (MOVVconst <typ.UInt64> [32]) (ZeroExt32to64 y)))
    74(Lsh32x16 <t> x y) && !shiftIsBounded(v) => (MASKEQZ (SLL <t> x (ZeroExt16to64 y)) (SGTU (MOVVconst <typ.UInt64> [32]) (ZeroExt16to64 y)))
    75(Lsh32x8  <t> x y) && !shiftIsBounded(v) => (MASKEQZ (SLL <t> x (ZeroExt8to64  y)) (SGTU (MOVVconst <typ.UInt64> [32]) (ZeroExt8to64  y)))
    76
    77(Lsh16x64 <t> x y) && !shiftIsBounded(v) => (MASKEQZ (SLLV <t> x                y)  (SGTU (MOVVconst <typ.UInt64> [64])                y))
    78(Lsh16x32 <t> x y) && !shiftIsBounded(v) => (MASKEQZ (SLLV <t> x (ZeroExt32to64 y)) (SGTU (MOVVconst <typ.UInt64> [64]) (ZeroExt32to64 y)))
    79(Lsh16x16 <t> x y) && !shiftIsBounded(v) => (MASKEQZ (SLLV <t> x (ZeroExt16to64 y)) (SGTU (MOVVconst <typ.UInt64> [64]) (ZeroExt16to64 y)))
    80(Lsh16x8  <t> x y) && !shiftIsBounded(v) => (MASKEQZ (SLLV <t> x (ZeroExt8to64  y)) (SGTU (MOVVconst <typ.UInt64> [64]) (ZeroExt8to64  y)))
    81
    82(Lsh8x64 <t> x y) && !shiftIsBounded(v) => (MASKEQZ (SLLV <t> x                y)  (SGTU (MOVVconst <typ.UInt64> [64])                y))
    83(Lsh8x32 <t> x y) && !shiftIsBounded(v) => (MASKEQZ (SLLV <t> x (ZeroExt32to64 y)) (SGTU (MOVVconst <typ.UInt64> [64]) (ZeroExt32to64 y)))
    84(Lsh8x16 <t> x y) && !shiftIsBounded(v) => (MASKEQZ (SLLV <t> x (ZeroExt16to64 y)) (SGTU (MOVVconst <typ.UInt64> [64]) (ZeroExt16to64 y)))
    85(Lsh8x8  <t> x y) && !shiftIsBounded(v) => (MASKEQZ (SLLV <t> x (ZeroExt8to64  y)) (SGTU (MOVVconst <typ.UInt64> [64]) (ZeroExt8to64  y)))
    86
    87// unsigned right shift
    88(Rsh64Ux(64|32|16|8) x y) && shiftIsBounded(v) => (SRLV  x                 y)
    89(Rsh32Ux(64|32|16|8) x y) && shiftIsBounded(v) => (SRL   x                 y)
    90(Rsh16Ux(64|32|16|8) x y) && shiftIsBounded(v) => (SRLV  (ZeroExt16to64 x) y)
    91(Rsh8Ux(64|32|16|8)  x y) && shiftIsBounded(v) => (SRLV  (ZeroExt8to64  x) y)
    92
    93(Rsh64Ux64 <t> x y) && !shiftIsBounded(v) => (MASKEQZ (SRLV <t> x                y)  (SGTU (MOVVconst <typ.UInt64> [64])                y))
    94(Rsh64Ux32 <t> x y) && !shiftIsBounded(v) => (MASKEQZ (SRLV <t> x (ZeroExt32to64 y)) (SGTU (MOVVconst <typ.UInt64> [64]) (ZeroExt32to64 y)))
    95(Rsh64Ux16 <t> x y) && !shiftIsBounded(v) => (MASKEQZ (SRLV <t> x (ZeroExt16to64 y)) (SGTU (MOVVconst <typ.UInt64> [64]) (ZeroExt16to64 y)))
    96(Rsh64Ux8  <t> x y) && !shiftIsBounded(v) => (MASKEQZ (SRLV <t> x (ZeroExt8to64  y)) (SGTU (MOVVconst <typ.UInt64> [64]) (ZeroExt8to64  y)))
    97
    98(Rsh32Ux64 <t> x y) && !shiftIsBounded(v) => (MASKEQZ (SRL <t> x                y)  (SGTU (MOVVconst <typ.UInt64> [32])                y))
    99(Rsh32Ux32 <t> x y) && !shiftIsBounded(v) => (MASKEQZ (SRL <t> x (ZeroExt32to64 y)) (SGTU (MOVVconst <typ.UInt64> [32]) (ZeroExt32to64 y)))
   100(Rsh32Ux16 <t> x y) && !shiftIsBounded(v) => (MASKEQZ (SRL <t> x (ZeroExt16to64 y)) (SGTU (MOVVconst <typ.UInt64> [32]) (ZeroExt16to64 y)))
   101(Rsh32Ux8  <t> x y) && !shiftIsBounded(v) => (MASKEQZ (SRL <t> x (ZeroExt8to64  y)) (SGTU (MOVVconst <typ.UInt64> [32]) (ZeroExt8to64  y)))
   102
   103(Rsh16Ux64 <t> x y) && !shiftIsBounded(v) => (MASKEQZ (SRLV <t> (ZeroExt16to64 x)                y)  (SGTU (MOVVconst <typ.UInt64> [64])                y))
   104(Rsh16Ux32 <t> x y) && !shiftIsBounded(v) => (MASKEQZ (SRLV <t> (ZeroExt16to64 x) (ZeroExt32to64 y)) (SGTU (MOVVconst <typ.UInt64> [64]) (ZeroExt32to64 y)))
   105(Rsh16Ux16 <t> x y) && !shiftIsBounded(v) => (MASKEQZ (SRLV <t> (ZeroExt16to64 x) (ZeroExt16to64 y)) (SGTU (MOVVconst <typ.UInt64> [64]) (ZeroExt16to64 y)))
   106(Rsh16Ux8  <t> x y) && !shiftIsBounded(v) => (MASKEQZ (SRLV <t> (ZeroExt16to64 x) (ZeroExt8to64  y)) (SGTU (MOVVconst <typ.UInt64> [64]) (ZeroExt8to64  y)))
   107
   108(Rsh8Ux64 <t> x y) && !shiftIsBounded(v) => (MASKEQZ (SRLV <t> (ZeroExt8to64 x)                y)  (SGTU (MOVVconst <typ.UInt64> [64])                y))
   109(Rsh8Ux32 <t> x y) && !shiftIsBounded(v) => (MASKEQZ (SRLV <t> (ZeroExt8to64 x) (ZeroExt32to64 y)) (SGTU (MOVVconst <typ.UInt64> [64]) (ZeroExt32to64 y)))
   110(Rsh8Ux16 <t> x y) && !shiftIsBounded(v) => (MASKEQZ (SRLV <t> (ZeroExt8to64 x) (ZeroExt16to64 y)) (SGTU (MOVVconst <typ.UInt64> [64]) (ZeroExt16to64 y)))
   111(Rsh8Ux8  <t> x y) && !shiftIsBounded(v) => (MASKEQZ (SRLV <t> (ZeroExt8to64 x) (ZeroExt8to64  y)) (SGTU (MOVVconst <typ.UInt64> [64]) (ZeroExt8to64  y)))
   112
   113// signed right shift
   114(Rsh64x(64|32|16|8) x y) && shiftIsBounded(v) => (SRAV   x                y)
   115(Rsh32x(64|32|16|8) x y) && shiftIsBounded(v) => (SRA    x                y)
   116(Rsh16x(64|32|16|8) x y) && shiftIsBounded(v) => (SRAV  (SignExt16to64 x) y)
   117(Rsh8x(64|32|16|8)  x y) && shiftIsBounded(v) => (SRAV  (SignExt8to64  x) y)
   118
   119(Rsh64x64 <t> x y) && !shiftIsBounded(v) => (SRAV x (OR <t> (NEGV <t> (SGTU y (MOVVconst <typ.UInt64> [63]))) y))
   120(Rsh64x32 <t> x y) && !shiftIsBounded(v) => (SRAV x (OR <t> (NEGV <t> (SGTU (ZeroExt32to64 y) (MOVVconst <typ.UInt64> [63]))) (ZeroExt32to64 y)))
   121(Rsh64x16 <t> x y) && !shiftIsBounded(v) => (SRAV x (OR <t> (NEGV <t> (SGTU (ZeroExt16to64 y) (MOVVconst <typ.UInt64> [63]))) (ZeroExt16to64 y)))
   122(Rsh64x8  <t> x y) && !shiftIsBounded(v) => (SRAV x (OR <t> (NEGV <t> (SGTU (ZeroExt8to64  y) (MOVVconst <typ.UInt64> [63]))) (ZeroExt8to64  y)))
   123
   124(Rsh32x64 <t> x y) && !shiftIsBounded(v) => (SRA x (OR <t> (NEGV <t> (SGTU y (MOVVconst <typ.UInt64> [31]))) y))
   125(Rsh32x32 <t> x y) && !shiftIsBounded(v) => (SRA x (OR <t> (NEGV <t> (SGTU (ZeroExt32to64 y) (MOVVconst <typ.UInt64> [31]))) (ZeroExt32to64 y)))
   126(Rsh32x16 <t> x y) && !shiftIsBounded(v) => (SRA x (OR <t> (NEGV <t> (SGTU (ZeroExt16to64 y) (MOVVconst <typ.UInt64> [31]))) (ZeroExt16to64 y)))
   127(Rsh32x8  <t> x y) && !shiftIsBounded(v) => (SRA x (OR <t> (NEGV <t> (SGTU (ZeroExt8to64  y) (MOVVconst <typ.UInt64> [31]))) (ZeroExt8to64  y)))
   128
   129(Rsh16x64 <t> x y) && !shiftIsBounded(v) => (SRAV (SignExt16to64 x) (OR <t> (NEGV <t> (SGTU y (MOVVconst <typ.UInt64> [63]))) y))
   130(Rsh16x32 <t> x y) && !shiftIsBounded(v) => (SRAV (SignExt16to64 x) (OR <t> (NEGV <t> (SGTU (ZeroExt32to64 y) (MOVVconst <typ.UInt64> [63]))) (ZeroExt32to64 y)))
   131(Rsh16x16 <t> x y) && !shiftIsBounded(v) => (SRAV (SignExt16to64 x) (OR <t> (NEGV <t> (SGTU (ZeroExt16to64 y) (MOVVconst <typ.UInt64> [63]))) (ZeroExt16to64 y)))
   132(Rsh16x8  <t> x y) && !shiftIsBounded(v) => (SRAV (SignExt16to64 x) (OR <t> (NEGV <t> (SGTU (ZeroExt8to64  y) (MOVVconst <typ.UInt64> [63]))) (ZeroExt8to64  y)))
   133
   134(Rsh8x64 <t> x y) && !shiftIsBounded(v) => (SRAV (SignExt8to64 x) (OR <t> (NEGV <t> (SGTU y (MOVVconst <typ.UInt64> [63]))) y))
   135(Rsh8x32 <t> x y) && !shiftIsBounded(v) => (SRAV (SignExt8to64 x) (OR <t> (NEGV <t> (SGTU (ZeroExt32to64 y) (MOVVconst <typ.UInt64> [63]))) (ZeroExt32to64 y)))
   136(Rsh8x16 <t> x y) && !shiftIsBounded(v) => (SRAV (SignExt8to64 x) (OR <t> (NEGV <t> (SGTU (ZeroExt16to64 y) (MOVVconst <typ.UInt64> [63]))) (ZeroExt16to64 y)))
   137(Rsh8x8  <t> x y) && !shiftIsBounded(v) => (SRAV (SignExt8to64 x) (OR <t> (NEGV <t> (SGTU (ZeroExt8to64  y) (MOVVconst <typ.UInt64> [63]))) (ZeroExt8to64  y)))
   138
   139// bitfield ops
   140
   141// bstrpickv
   142// (x << lc) >> rc
   143(SRLVconst [rc] (SLLVconst [lc] x)) && lc <= rc => (BSTRPICKV [rc-lc + ((64-lc)-1)<<6] x)
   144// uint64(x) >> rc
   145(SRLVconst [rc] (MOVWUreg x)) && rc < 32 => (BSTRPICKV [rc + 31<<6] x)
   146(SRLVconst [rc] (MOVHUreg x)) && rc < 16 => (BSTRPICKV [rc + 15<<6] x)
   147(SRLVconst [rc] (MOVBUreg x)) && rc < 8 => (BSTRPICKV [rc + 7<<6] x)
   148// uint64(x >> rc)
   149(MOVWUreg (SRLVconst [rc] x)) && rc < 32 => (BSTRPICKV [rc + (31+rc)<<6] x)
   150(MOVHUreg (SRLVconst [rc] x)) && rc < 16 => (BSTRPICKV [rc + (15+rc)<<6] x)
   151(MOVBUreg (SRLVconst [rc] x)) && rc < 8 => (BSTRPICKV [rc + (7+rc)<<6] x)
   152
   153// rotates
   154(RotateLeft8 <t> x (MOVVconst [c])) => (Or8 (Lsh8x64 <t> x (MOVVconst [c&7])) (Rsh8Ux64 <t> x (MOVVconst [-c&7])))
   155(RotateLeft8 <t> x y) => (OR <t> (SLLV <t> x (ANDconst <typ.Int64> [7] y)) (SRLV <t> (ZeroExt8to64 x) (ANDconst <typ.Int64> [7] (NEGV <typ.Int64> y))))
   156(RotateLeft16 <t> x (MOVVconst [c])) => (Or16 (Lsh16x64 <t> x (MOVVconst [c&15])) (Rsh16Ux64 <t> x (MOVVconst [-c&15])))
   157(RotateLeft16 <t> x y) => (ROTR <t> (OR <typ.UInt32> (ZeroExt16to32 x) (SLLVconst <t> (ZeroExt16to32 x) [16])) (NEGV <typ.Int64> y))
   158(RotateLeft32 x y) => (ROTR  x (NEGV <y.Type> y))
   159(RotateLeft64 x y) => (ROTRV x (NEGV <y.Type> y))
   160
   161// unary ops
   162(Neg(64|32|16|8) ...) => (NEGV ...)
   163(Neg(32|64)F ...) => (NEG(F|D) ...)
   164
   165(Com(64|32|16|8) x) => (NOR (MOVVconst [0]) x)
   166
   167(BitLen64 <t> x) => (NEGV <t> (SUBVconst <t> [64] (CLZV <t> x)))
   168(BitLen32 <t> x) => (NEGV <t> (SUBVconst <t> [32] (CLZW <t> x)))
   169(BitLen(16|8) x) => (BitLen64 (ZeroExt(16|8)to64 x))
   170(Bswap(16|32|64) ...) => (REVB(2H|2W|V) ...)
   171(BitRev8 ...) => (BITREV4B ...)
   172(BitRev16 <t> x) => (REVB2H (BITREV4B <t> x))
   173(BitRev32 ...) => (BITREVW ...)
   174(BitRev64 ...) => (BITREVV ...)
   175(Ctz(64|32|16|8)NonZero ...) => (Ctz64 ...)
   176(Ctz(32|64) ...) => (CTZ(W|V) ...)
   177(Ctz16 x) => (CTZV (OR <typ.UInt64> x (MOVVconst [1<<16])))
   178(Ctz8  x) => (CTZV (OR <typ.UInt64> x (MOVVconst [1<<8])))
   179
   180(PopCount64 <t> x) => (MOVVfpgp <t> (VPCNT64 <typ.Float64> (MOVVgpfp <typ.Float64> x)))
   181(PopCount32 <t> x) => (MOVWfpgp <t> (VPCNT32 <typ.Float32> (MOVWgpfp <typ.Float32> x)))
   182(PopCount16 <t> x) => (MOVWfpgp <t> (VPCNT16 <typ.Float32> (MOVWgpfp <typ.Float32> (ZeroExt16to32 x))))
   183
   184// math package intrinsics
   185(Sqrt ...) => (SQRTD ...)
   186(Sqrt32 ...) => (SQRTF ...)
   187(Abs ...) => (ABSD ...)
   188(Copysign ...) => (FCOPYSGD ...)
   189
   190(Min(64|32)F ...) => (FMIN(D|F) ...)
   191(Max(64|32)F ...) => (FMAX(D|F) ...)
   192
   193// boolean ops -- booleans are represented with 0=false, 1=true
   194(AndB ...) => (AND ...)
   195(OrB ...) => (OR ...)
   196(EqB x y) => (XOR (MOVVconst [1]) (XOR <typ.Bool> x y))
   197(NeqB ...) => (XOR ...)
   198(Not x) => (XORconst [1] x)
   199
   200// constants
   201(Const(64|32|16|8) [val]) => (MOVVconst [int64(val)])
   202(Const(32|64)F [val]) => (MOV(F|D)const [float64(val)])
   203(ConstNil) => (MOVVconst [0])
   204(ConstBool [t]) => (MOVVconst [int64(b2i(t))])
   205
   206(Slicemask <t> x) => (SRAVconst (NEGV <t> x) [63])
   207
   208// truncations
   209// Because we ignore high parts of registers, truncates are just copies.
   210(Trunc16to8 ...) => (Copy ...)
   211(Trunc32to8 ...) => (Copy ...)
   212(Trunc32to16 ...) => (Copy ...)
   213(Trunc64to8 ...) => (Copy ...)
   214(Trunc64to16 ...) => (Copy ...)
   215(Trunc64to32 ...) => (Copy ...)
   216
   217// Zero-/Sign-extensions
   218(ZeroExt8to16 ...) => (MOVBUreg ...)
   219(ZeroExt8to32 ...) => (MOVBUreg ...)
   220(ZeroExt16to32 ...) => (MOVHUreg ...)
   221(ZeroExt8to64 ...) => (MOVBUreg ...)
   222(ZeroExt16to64 ...) => (MOVHUreg ...)
   223(ZeroExt32to64 ...) => (MOVWUreg ...)
   224
   225(SignExt8to16 ...) => (MOVBreg ...)
   226(SignExt8to32 ...) => (MOVBreg ...)
   227(SignExt16to32 ...) => (MOVHreg ...)
   228(SignExt8to64 ...) => (MOVBreg ...)
   229(SignExt16to64 ...) => (MOVHreg ...)
   230(SignExt32to64 ...) => (MOVWreg ...)
   231
   232// float <=> int conversion
   233(Cvt32to32F ...) => (MOVWF ...)
   234(Cvt32to64F ...) => (MOVWD ...)
   235(Cvt64to32F ...) => (MOVVF ...)
   236(Cvt64to64F ...) => (MOVVD ...)
   237(Cvt32Fto32 ...) => (TRUNCFW ...)
   238(Cvt64Fto32 ...) => (TRUNCDW ...)
   239(Cvt32Fto64 ...) => (TRUNCFV ...)
   240(Cvt64Fto64 ...) => (TRUNCDV ...)
   241(Cvt32Fto64F ...) => (MOVFD ...)
   242(Cvt64Fto32F ...) => (MOVDF ...)
   243
   244(CvtBoolToUint8 ...) => (Copy ...)
   245
   246(Round(32|64)F ...) => (LoweredRound(32|64)F ...)
   247
   248// comparisons
   249(Eq8 x y)  => (SGTU (MOVVconst [1]) (XOR (ZeroExt8to64 x) (ZeroExt8to64 y)))
   250(Eq16 x y) => (SGTU (MOVVconst [1]) (XOR (ZeroExt16to64 x) (ZeroExt16to64 y)))
   251(Eq32 x y) => (SGTU (MOVVconst [1]) (XOR (ZeroExt32to64 x) (ZeroExt32to64 y)))
   252(Eq64 x y) => (SGTU (MOVVconst [1]) (XOR x y))
   253(EqPtr x y) => (SGTU (MOVVconst [1]) (XOR x y))
   254(Eq(32|64)F x y) => (FPFlagTrue (CMPEQ(F|D) x y))
   255
   256(Neq8 x y)  => (SGTU (XOR (ZeroExt8to64 x) (ZeroExt8to64 y)) (MOVVconst [0]))
   257(Neq16 x y) => (SGTU (XOR (ZeroExt16to32 x) (ZeroExt16to64 y)) (MOVVconst [0]))
   258(Neq32 x y) => (SGTU (XOR (ZeroExt32to64 x) (ZeroExt32to64 y)) (MOVVconst [0]))
   259(Neq64 x y) => (SGTU (XOR x y) (MOVVconst [0]))
   260(NeqPtr x y) => (SGTU (XOR x y) (MOVVconst [0]))
   261(Neq(32|64)F x y) => (FPFlagFalse (CMPEQ(F|D) x y))
   262
   263(Less8 x y)  => (SGT (SignExt8to64 y) (SignExt8to64 x))
   264(Less16 x y) => (SGT (SignExt16to64 y) (SignExt16to64 x))
   265(Less32 x y) => (SGT (SignExt32to64 y) (SignExt32to64 x))
   266(Less64 x y) => (SGT y x)
   267(Less(32|64)F x y) => (FPFlagTrue (CMPGT(F|D) y x)) // reverse operands to work around NaN
   268
   269(Less8U x y)  => (SGTU (ZeroExt8to64 y) (ZeroExt8to64 x))
   270(Less16U x y) => (SGTU (ZeroExt16to64 y) (ZeroExt16to64 x))
   271(Less32U x y) => (SGTU (ZeroExt32to64 y) (ZeroExt32to64 x))
   272(Less64U x y) => (SGTU y x)
   273
   274(Leq8 x y)  => (XOR (MOVVconst [1]) (SGT (SignExt8to64 x) (SignExt8to64 y)))
   275(Leq16 x y) => (XOR (MOVVconst [1]) (SGT (SignExt16to64 x) (SignExt16to64 y)))
   276(Leq32 x y) => (XOR (MOVVconst [1]) (SGT (SignExt32to64 x) (SignExt32to64 y)))
   277(Leq64 x y) => (XOR (MOVVconst [1]) (SGT x y))
   278(Leq(32|64)F x y) => (FPFlagTrue (CMPGE(F|D) y x)) // reverse operands to work around NaN
   279
   280(Leq8U x y)  => (XOR (MOVVconst [1]) (SGTU (ZeroExt8to64 x) (ZeroExt8to64 y)))
   281(Leq16U x y) => (XOR (MOVVconst [1]) (SGTU (ZeroExt16to64 x) (ZeroExt16to64 y)))
   282(Leq32U x y) => (XOR (MOVVconst [1]) (SGTU (ZeroExt32to64 x) (ZeroExt32to64 y)))
   283(Leq64U x y) => (XOR (MOVVconst [1]) (SGTU x y))
   284
   285(OffPtr [off] ptr:(SP)) => (MOVVaddr [int32(off)] ptr)
   286(OffPtr [off] ptr) => (ADDVconst [off] ptr)
   287
   288(Addr {sym} base) => (MOVVaddr {sym} base)
   289(LocalAddr <t> {sym} base mem) && t.Elem().HasPointers() => (MOVVaddr {sym} (SPanchored base mem))
   290(LocalAddr <t> {sym} base _)  && !t.Elem().HasPointers() => (MOVVaddr {sym} base)
   291
   292// loads
   293(Load <t> ptr mem) && t.IsBoolean() => (MOVBUload ptr mem)
   294(Load <t> ptr mem) && (is8BitInt(t)  &&  t.IsSigned()) => (MOVBload ptr mem)
   295(Load <t> ptr mem) && (is8BitInt(t)  && !t.IsSigned()) => (MOVBUload ptr mem)
   296(Load <t> ptr mem) && (is16BitInt(t) &&  t.IsSigned()) => (MOVHload ptr mem)
   297(Load <t> ptr mem) && (is16BitInt(t) && !t.IsSigned()) => (MOVHUload ptr mem)
   298(Load <t> ptr mem) && (is32BitInt(t) &&  t.IsSigned()) => (MOVWload ptr mem)
   299(Load <t> ptr mem) && (is32BitInt(t) && !t.IsSigned()) => (MOVWUload ptr mem)
   300(Load <t> ptr mem) && (is64BitInt(t) || isPtr(t)) => (MOVVload ptr mem)
   301(Load <t> ptr mem) && is32BitFloat(t) => (MOVFload ptr mem)
   302(Load <t> ptr mem) && is64BitFloat(t) => (MOVDload ptr mem)
   303
   304// stores
   305(Store {t} ptr val mem) && t.Size() == 1 => (MOVBstore ptr val mem)
   306(Store {t} ptr val mem) && t.Size() == 2 => (MOVHstore ptr val mem)
   307(Store {t} ptr val mem) && t.Size() == 4 && !t.IsFloat() => (MOVWstore ptr val mem)
   308(Store {t} ptr val mem) && t.Size() == 8 && !t.IsFloat() => (MOVVstore ptr val mem)
   309(Store {t} ptr val mem) && t.Size() == 4 &&  t.IsFloat() => (MOVFstore ptr val mem)
   310(Store {t} ptr val mem) && t.Size() == 8 &&  t.IsFloat() => (MOVDstore ptr val mem)
   311
   312// zeroing
   313(Zero [0] _ mem) => mem
   314(Zero [1] ptr mem) => (MOVBstore ptr (MOVVconst [0]) mem)
   315(Zero [2] ptr mem) => (MOVHstore ptr (MOVVconst [0]) mem)
   316(Zero [3] ptr mem) =>
   317    (MOVBstore [2] ptr (MOVVconst [0])
   318        (MOVHstore ptr (MOVVconst [0]) mem))
   319(Zero [4] {t} ptr mem) => (MOVWstore ptr (MOVVconst [0]) mem)
   320(Zero [5] ptr mem) =>
   321    (MOVBstore [4] ptr (MOVVconst [0])
   322        (MOVWstore ptr (MOVVconst [0]) mem))
   323(Zero [6] ptr mem) =>
   324    (MOVHstore [4] ptr (MOVVconst [0])
   325        (MOVWstore ptr (MOVVconst [0]) mem))
   326(Zero [7] ptr mem) =>
   327    (MOVWstore [3] ptr (MOVVconst [0])
   328        (MOVWstore ptr (MOVVconst [0]) mem))
   329(Zero [8] {t} ptr mem) => (MOVVstore ptr (MOVVconst [0]) mem)
   330(Zero [9] ptr mem) =>
   331        (MOVBstore [8] ptr (MOVVconst [0])
   332                (MOVVstore ptr (MOVVconst [0]) mem))
   333(Zero [10] ptr mem) =>
   334        (MOVHstore [8] ptr (MOVVconst [0])
   335                (MOVVstore ptr (MOVVconst [0]) mem))
   336(Zero [11] ptr mem) =>
   337        (MOVWstore [7] ptr (MOVVconst [0])
   338                (MOVVstore ptr (MOVVconst [0]) mem))
   339(Zero [12] ptr mem) =>
   340        (MOVWstore [8] ptr (MOVVconst [0])
   341                (MOVVstore ptr (MOVVconst [0]) mem))
   342(Zero [13] ptr mem) =>
   343        (MOVVstore [5] ptr (MOVVconst [0])
   344                (MOVVstore ptr (MOVVconst [0]) mem))
   345(Zero [14] ptr mem) =>
   346        (MOVVstore [6] ptr (MOVVconst [0])
   347                (MOVVstore ptr (MOVVconst [0]) mem))
   348(Zero [15] ptr mem) =>
   349        (MOVVstore [7] ptr (MOVVconst [0])
   350                (MOVVstore ptr (MOVVconst [0]) mem))
   351(Zero [16] ptr mem) =>
   352        (MOVVstore [8] ptr (MOVVconst [0])
   353                (MOVVstore ptr (MOVVconst [0]) mem))
   354
   355// strip off fractional word zeroing
   356(Zero [s] ptr mem) && s%8 != 0 && s > 16 =>
   357    (Zero [s%8]
   358        (OffPtr <ptr.Type> ptr [s-s%8])
   359        (Zero [s-s%8] ptr mem))
   360
   361// medium zeroing uses a duff device
   362(Zero [s] ptr mem)
   363    && s%8 == 0 && s > 16 && s <= 8*128 =>
   364    (DUFFZERO [8 * (128 - s/8)] ptr mem)
   365
   366// large zeroing uses a loop
   367(Zero [s] ptr mem)
   368    && s%8 == 0 && s > 8*128 =>
   369    (LoweredZero
   370        ptr
   371        (ADDVconst <ptr.Type> ptr [s-8])
   372        mem)
   373
   374// moves
   375(Move [0] _ _ mem) => mem
   376(Move [1] dst src mem) => (MOVBstore dst (MOVBUload src mem) mem)
   377(Move [2] dst src mem) => (MOVHstore dst (MOVHUload src mem) mem)
   378(Move [3] dst src mem) =>
   379        (MOVBstore [2] dst (MOVBUload [2] src mem)
   380                (MOVHstore dst (MOVHUload src mem) mem))
   381(Move [4] dst src mem) => (MOVWstore dst (MOVWUload src mem) mem)
   382(Move [5] dst src mem) =>
   383        (MOVBstore [4] dst (MOVBUload [4] src mem)
   384                (MOVWstore dst (MOVWUload src mem) mem))
   385(Move [6] dst src mem) =>
   386        (MOVHstore [4] dst (MOVHUload [4] src mem)
   387                (MOVWstore dst (MOVWUload src mem) mem))
   388(Move [7] dst src mem) =>
   389        (MOVWstore [3] dst (MOVWUload [3] src mem)
   390                (MOVWstore dst (MOVWUload src mem) mem))
   391(Move [8] dst src mem) => (MOVVstore dst (MOVVload src mem) mem)
   392(Move [9] dst src mem) =>
   393        (MOVBstore [8] dst (MOVBUload [8] src mem)
   394                (MOVVstore dst (MOVVload src mem) mem))
   395(Move [10] dst src mem) =>
   396        (MOVHstore [8] dst (MOVHUload [8] src mem)
   397                (MOVVstore dst (MOVVload src mem) mem))
   398(Move [11] dst src mem) =>
   399        (MOVWstore [7] dst (MOVWload [7] src mem)
   400                (MOVVstore dst (MOVVload src mem) mem))
   401(Move [12] dst src mem) =>
   402        (MOVWstore [8] dst (MOVWUload [8] src mem)
   403                (MOVVstore dst (MOVVload src mem) mem))
   404(Move [13] dst src mem) =>
   405        (MOVVstore [5] dst (MOVVload [5] src mem)
   406                (MOVVstore dst (MOVVload src mem) mem))
   407(Move [14] dst src mem) =>
   408        (MOVVstore [6] dst (MOVVload [6] src mem)
   409                (MOVVstore dst (MOVVload src mem) mem))
   410(Move [15] dst src mem) =>
   411        (MOVVstore [7] dst (MOVVload [7] src mem)
   412                (MOVVstore dst (MOVVload src mem) mem))
   413(Move [16] dst src mem) =>
   414        (MOVVstore [8] dst (MOVVload [8] src mem)
   415                (MOVVstore dst (MOVVload src mem) mem))
   416
   417// strip off fractional word move
   418(Move [s] dst src mem) && s%8 != 0 && s > 16 =>
   419        (Move [s%8]
   420                (OffPtr <dst.Type> dst [s-s%8])
   421                (OffPtr <src.Type> src [s-s%8])
   422                (Move [s-s%8] dst src mem))
   423
   424// medium move uses a duff device
   425(Move [s] dst src mem)
   426        && s%8 == 0 && s > 16 && s <= 8*128
   427        && logLargeCopy(v, s)  =>
   428        (DUFFCOPY [16 * (128 - s/8)] dst src mem)
   429// 16 and 128 are magic constants.  16 is the number of bytes to encode:
   430//     MOVV    (R20), R30
   431//     ADDV    $8, R20
   432//     MOVV    R30, (R21)
   433//     ADDV    $8, R21
   434// and 128 is the number of such blocks. See runtime/duff_loong64.s:duffcopy.
   435
   436// large move uses a loop
   437(Move [s] dst src mem)
   438        && s%8 == 0 && s > 1024 && logLargeCopy(v, s) =>
   439        (LoweredMove
   440                dst
   441                src
   442                (ADDVconst <src.Type> src [s-8])
   443                mem)
   444
   445
   446// float <=> int register moves, with no conversion.
   447// These come up when compiling math.{Float64bits, Float64frombits, Float32bits, Float32frombits}.
   448(MOVVload  [off] {sym} ptr (MOVDstore [off] {sym} ptr val _)) => (MOVVfpgp val)
   449(MOVDload  [off] {sym} ptr (MOVVstore [off] {sym} ptr val _)) => (MOVVgpfp val)
   450(MOVWUload [off] {sym} ptr (MOVFstore [off] {sym} ptr val _)) => (ZeroExt32to64 (MOVWfpgp <typ.Float32> val))
   451(MOVFload  [off] {sym} ptr (MOVWstore [off] {sym} ptr val _)) => (MOVWgpfp val)
   452
   453// Similarly for stores, if we see a store after FPR <=> GPR move, then redirect store to use the other register set.
   454(MOVVstore [off] {sym} ptr (MOVVfpgp val) mem) => (MOVDstore [off] {sym} ptr val mem)
   455(MOVDstore [off] {sym} ptr (MOVVgpfp val) mem) => (MOVVstore [off] {sym} ptr val mem)
   456(MOVWstore [off] {sym} ptr (MOVWfpgp val) mem) => (MOVFstore [off] {sym} ptr val mem)
   457(MOVFstore [off] {sym} ptr (MOVWgpfp val) mem) => (MOVWstore [off] {sym} ptr val mem)
   458
   459// calls
   460(StaticCall ...) => (CALLstatic ...)
   461(ClosureCall ...) => (CALLclosure ...)
   462(InterCall ...) => (CALLinter ...)
   463(TailCall ...) => (CALLtail ...)
   464
   465// atomic intrinsics
   466(AtomicLoad(8|32|64)   ...) => (LoweredAtomicLoad(8|32|64)  ...)
   467(AtomicLoadPtr ...) => (LoweredAtomicLoad64 ...)
   468
   469(AtomicStore(8|32|64) ...) => (LoweredAtomicStore(8|32|64)  ...)
   470(AtomicStore(8|32|64)Variant ...) => (LoweredAtomicStore(8|32|64)Variant  ...)
   471(AtomicStorePtrNoWB ...) => (LoweredAtomicStore64 ...)
   472
   473(AtomicExchange(32|64) ...) => (LoweredAtomicExchange(32|64) ...)
   474(AtomicExchange8Variant  ...) => (LoweredAtomicExchange8Variant  ...)
   475
   476(AtomicAdd(32|64) ...) => (LoweredAtomicAdd(32|64) ...)
   477
   478// Loong64's 32-bit atomic operation instructions ll.w and amcasw are both sign-extended,
   479// so the input parameters need to be sign-extended to 64 bits, otherwise the subsequent
   480// comparison operations may not produce the expected results.
   481//
   482(AtomicCompareAndSwap32 ptr old new mem) => (LoweredAtomicCas32 ptr (SignExt32to64 old) new mem)
   483(AtomicCompareAndSwap64 ...) => (LoweredAtomicCas64 ...)
   484(AtomicCompareAndSwap32Variant ptr old new mem) => (LoweredAtomicCas32Variant ptr (SignExt32to64 old) new mem)
   485(AtomicCompareAndSwap64Variant ...) => (LoweredAtomicCas64Variant ...)
   486
   487// Atomic memory logical operations (old style).
   488//
   489// AtomicAnd8(ptr,val) => LoweredAtomicAnd32(ptr&^3, ^((uint8(val) ^ 0xff) << ((ptr & 3) * 8)))
   490// AtomicOr8(ptr,val)  => LoweredAtomicOr32(ptr&^3, uint32(val) << ((ptr & 3) * 8))
   491//
   492(AtomicAnd8 ptr val mem) =>
   493	(LoweredAtomicAnd32 (AND <typ.Uintptr> (MOVVconst [^3]) ptr)
   494		(NORconst [0] <typ.UInt32> (SLLV <typ.UInt32> (XORconst <typ.UInt32> [0xff] (ZeroExt8to32 val))
   495			(SLLVconst <typ.UInt64> [3] (ANDconst <typ.UInt64> [3] ptr)))) mem)
   496
   497(AtomicOr8 ptr val mem) =>
   498	(LoweredAtomicOr32 (AND <typ.Uintptr> (MOVVconst [^3]) ptr)
   499		(SLLV <typ.UInt32> (ZeroExt8to32 val)
   500			(SLLVconst <typ.UInt64> [3] (ANDconst <typ.UInt64> [3] ptr))) mem)
   501
   502(AtomicAnd32 ...) => (LoweredAtomicAnd32 ...)
   503(AtomicOr32  ...) => (LoweredAtomicOr32  ...)
   504
   505// Atomic memory logical operations (new style).
   506(AtomicAnd(64|32)value ...) => (LoweredAtomicAnd(64|32)value ...)
   507(AtomicOr(64|32)value  ...) => (LoweredAtomicOr(64|32)value  ...)
   508
   509// checks
   510(NilCheck ...) => (LoweredNilCheck ...)
   511(IsNonNil ptr) => (SGTU ptr (MOVVconst [0]))
   512(IsInBounds idx len) => (SGTU len idx)
   513(IsSliceInBounds idx len) => (XOR (MOVVconst [1]) (SGTU idx len))
   514
   515// pseudo-ops
   516(GetClosurePtr ...) => (LoweredGetClosurePtr ...)
   517(GetCallerSP ...) => (LoweredGetCallerSP ...)
   518(GetCallerPC ...) => (LoweredGetCallerPC ...)
   519
   520(If cond yes no) => (NE (MOVBUreg <typ.UInt64> cond) yes no)
   521(MOVBUreg x:((SGT|SGTU) _ _)) => x
   522(MOVBUreg x:(XOR (MOVVconst [1]) ((SGT|SGTU) _ _))) => x
   523
   524// Write barrier.
   525(WB ...) => (LoweredWB ...)
   526
   527// Publication barrier as intrinsic
   528(PubBarrier ...) => (LoweredPubBarrier ...)
   529
   530(PanicBounds [kind] x y mem) && boundsABI(kind) == 0 => (LoweredPanicBoundsA [kind] x y mem)
   531(PanicBounds [kind] x y mem) && boundsABI(kind) == 1 => (LoweredPanicBoundsB [kind] x y mem)
   532(PanicBounds [kind] x y mem) && boundsABI(kind) == 2 => (LoweredPanicBoundsC [kind] x y mem)
   533
   534(CondSelect <t> x y cond) => (OR (MASKEQZ <t> x cond) (MASKNEZ <t> y cond))
   535
   536// c > d-x => x > d-c
   537(SGT (MOVVconst [c]) (NEGV (SUBVconst [d] x))) && is32Bit(d-c) => (SGT x (MOVVconst [d-c]))
   538
   539(SGT  (MOVVconst [c]) x) && is32Bit(c) => (SGTconst  [c] x)
   540(SGTU (MOVVconst [c]) x) && is32Bit(c) => (SGTUconst [c] x)
   541
   542// fold offset into address
   543(ADDVconst [off1] (MOVVaddr [off2] {sym} ptr)) && is32Bit(off1+int64(off2)) => (MOVVaddr [int32(off1)+int32(off2)] {sym} ptr)
   544
   545// fold address into load/store
   546// Do not fold global variable access in -dynlink mode, where it will be rewritten
   547// to use the GOT via REGTMP, which currently cannot handle large offset.
   548(MOV(B|BU|H|HU|W|WU|V|F|D)load [off1] {sym} (ADDVconst [off2] ptr) mem) && is32Bit(int64(off1)+off2)
   549	&& (ptr.Op != OpSB || !config.ctxt.Flag_dynlink) =>
   550	(MOV(B|BU|H|HU|W|WU|V|F|D)load [off1+int32(off2)] {sym} ptr mem)
   551
   552(MOV(B|H|W|V|F|D)store [off1] {sym} (ADDVconst [off2] ptr) val mem) && is32Bit(int64(off1)+off2)
   553	&& (ptr.Op != OpSB || !config.ctxt.Flag_dynlink) =>
   554	(MOV(B|H|W|V|F|D)store [off1+int32(off2)] {sym} ptr val mem)
   555
   556(MOV(B|H|W|V)storezero [off1] {sym} (ADDVconst [off2] ptr) mem) && is32Bit(int64(off1)+off2)
   557	&& (ptr.Op != OpSB || !config.ctxt.Flag_dynlink) =>
   558	(MOV(B|H|W|V)storezero [off1+int32(off2)] {sym} ptr mem)
   559
   560(MOV(B|BU|H|HU|W|WU|V|F|D)load [off1] {sym1} (MOVVaddr [off2] {sym2} ptr) mem) && canMergeSym(sym1,sym2)
   561	&& is32Bit(int64(off1)+int64(off2)) && (ptr.Op != OpSB || !config.ctxt.Flag_dynlink) =>
   562	(MOV(B|BU|H|HU|W|WU|V|F|D)load [off1+int32(off2)] {mergeSym(sym1,sym2)} ptr mem)
   563
   564(MOV(B|H|W|V|F|D)store [off1] {sym1} (MOVVaddr [off2] {sym2} ptr) val mem) && canMergeSym(sym1,sym2)
   565	&& is32Bit(int64(off1)+int64(off2)) && (ptr.Op != OpSB || !config.ctxt.Flag_dynlink) =>
   566	(MOV(B|H|W|V|F|D)store [off1+int32(off2)] {mergeSym(sym1,sym2)} ptr val mem)
   567
   568(MOV(B|H|W|V)storezero [off1] {sym1} (MOVVaddr [off2] {sym2} ptr) mem) && canMergeSym(sym1,sym2)
   569	&& is32Bit(int64(off1)+int64(off2)) && (ptr.Op != OpSB || !config.ctxt.Flag_dynlink) =>
   570	(MOV(B|H|W|V)storezero [off1+int32(off2)] {mergeSym(sym1,sym2)} ptr mem)
   571
   572// don't extend after proper load
   573(MOVBreg x:(MOVBload _ _)) => (MOVVreg x)
   574(MOVBUreg x:(MOVBUload _ _)) => (MOVVreg x)
   575(MOVHreg x:(MOVBload _ _)) => (MOVVreg x)
   576(MOVHreg x:(MOVBUload _ _)) => (MOVVreg x)
   577(MOVHreg x:(MOVHload _ _)) => (MOVVreg x)
   578(MOVHUreg x:(MOVBUload _ _)) => (MOVVreg x)
   579(MOVHUreg x:(MOVHUload _ _)) => (MOVVreg x)
   580(MOVWreg x:(MOVBload _ _)) => (MOVVreg x)
   581(MOVWreg x:(MOVBUload _ _)) => (MOVVreg x)
   582(MOVWreg x:(MOVHload _ _)) => (MOVVreg x)
   583(MOVWreg x:(MOVHUload _ _)) => (MOVVreg x)
   584(MOVWreg x:(MOVWload _ _)) => (MOVVreg x)
   585(MOVWUreg x:(MOVBUload _ _)) => (MOVVreg x)
   586(MOVWUreg x:(MOVHUload _ _)) => (MOVVreg x)
   587(MOVWUreg x:(MOVWUload _ _)) => (MOVVreg x)
   588
   589// fold double extensions
   590(MOVBreg x:(MOVBreg _)) => (MOVVreg x)
   591(MOVBUreg x:(MOVBUreg _)) => (MOVVreg x)
   592(MOVHreg x:(MOVBreg _)) => (MOVVreg x)
   593(MOVHreg x:(MOVBUreg _)) => (MOVVreg x)
   594(MOVHreg x:(MOVHreg _)) => (MOVVreg x)
   595(MOVHUreg x:(MOVBUreg _)) => (MOVVreg x)
   596(MOVHUreg x:(MOVHUreg _)) => (MOVVreg x)
   597(MOVWreg x:(MOVBreg _)) => (MOVVreg x)
   598(MOVWreg x:(MOVBUreg _)) => (MOVVreg x)
   599(MOVWreg x:(MOVHreg _)) => (MOVVreg x)
   600(MOVWreg x:(MOVWreg _)) => (MOVVreg x)
   601(MOVWUreg x:(MOVBUreg _)) => (MOVVreg x)
   602(MOVWUreg x:(MOVHUreg _)) => (MOVVreg x)
   603(MOVWUreg x:(MOVWUreg _)) => (MOVVreg x)
   604
   605// don't extend before store
   606(MOVBstore [off] {sym} ptr (MOVBreg x) mem) => (MOVBstore [off] {sym} ptr x mem)
   607(MOVBstore [off] {sym} ptr (MOVBUreg x) mem) => (MOVBstore [off] {sym} ptr x mem)
   608(MOVBstore [off] {sym} ptr (MOVHreg x) mem) => (MOVBstore [off] {sym} ptr x mem)
   609(MOVBstore [off] {sym} ptr (MOVHUreg x) mem) => (MOVBstore [off] {sym} ptr x mem)
   610(MOVBstore [off] {sym} ptr (MOVWreg x) mem) => (MOVBstore [off] {sym} ptr x mem)
   611(MOVBstore [off] {sym} ptr (MOVWUreg x) mem) => (MOVBstore [off] {sym} ptr x mem)
   612(MOVHstore [off] {sym} ptr (MOVHreg x) mem) => (MOVHstore [off] {sym} ptr x mem)
   613(MOVHstore [off] {sym} ptr (MOVHUreg x) mem) => (MOVHstore [off] {sym} ptr x mem)
   614(MOVHstore [off] {sym} ptr (MOVWreg x) mem) => (MOVHstore [off] {sym} ptr x mem)
   615(MOVHstore [off] {sym} ptr (MOVWUreg x) mem) => (MOVHstore [off] {sym} ptr x mem)
   616(MOVWstore [off] {sym} ptr (MOVWreg x) mem) => (MOVWstore [off] {sym} ptr x mem)
   617(MOVWstore [off] {sym} ptr (MOVWUreg x) mem) => (MOVWstore [off] {sym} ptr x mem)
   618
   619(MOVBstore [off] {sym} ptr (MOVVconst [0]) mem) => (MOVBstorezero [off] {sym} ptr mem)
   620(MOVHstore [off] {sym} ptr (MOVVconst [0]) mem) => (MOVHstorezero [off] {sym} ptr mem)
   621(MOVWstore [off] {sym} ptr (MOVVconst [0]) mem) => (MOVWstorezero [off] {sym} ptr mem)
   622(MOVVstore [off] {sym} ptr (MOVVconst [0]) mem) => (MOVVstorezero [off] {sym} ptr mem)
   623
   624// register indexed load
   625(MOVVload [off] {sym} (ADDV ptr idx) mem) && off == 0 && sym == nil => (MOVVloadidx ptr idx mem)
   626(MOVWUload [off] {sym} (ADDV ptr idx) mem) && off == 0 && sym == nil => (MOVWUloadidx ptr idx mem)
   627(MOVWload [off] {sym} (ADDV ptr idx) mem) && off == 0 && sym == nil => (MOVWloadidx ptr idx mem)
   628(MOVHUload [off] {sym} (ADDV ptr idx) mem) && off == 0 && sym == nil => (MOVHUloadidx ptr idx mem)
   629(MOVHload [off] {sym} (ADDV ptr idx) mem) && off == 0 && sym == nil => (MOVHloadidx ptr idx mem)
   630(MOVBUload [off] {sym} (ADDV ptr idx) mem) && off == 0 && sym == nil => (MOVBUloadidx ptr idx mem)
   631(MOVBload [off] {sym} (ADDV ptr idx) mem) && off == 0 && sym == nil => (MOVBloadidx ptr idx mem)
   632(MOVFload [off] {sym} (ADDV ptr idx) mem) && off == 0 && sym == nil => (MOVFloadidx ptr idx mem)
   633(MOVDload [off] {sym} (ADDV ptr idx) mem) && off == 0 && sym == nil => (MOVDloadidx ptr idx mem)
   634(MOVVloadidx ptr (MOVVconst [c]) mem) && is32Bit(c) => (MOVVload [int32(c)] ptr mem)
   635(MOVVloadidx (MOVVconst [c]) ptr mem) && is32Bit(c) => (MOVVload [int32(c)] ptr mem)
   636(MOVWUloadidx ptr (MOVVconst [c]) mem) && is32Bit(c) => (MOVWUload [int32(c)] ptr mem)
   637(MOVWUloadidx (MOVVconst [c]) ptr mem) && is32Bit(c) => (MOVWUload [int32(c)] ptr mem)
   638(MOVWloadidx ptr (MOVVconst [c]) mem) && is32Bit(c) => (MOVWload [int32(c)] ptr mem)
   639(MOVWloadidx (MOVVconst [c]) ptr mem) && is32Bit(c) => (MOVWload [int32(c)] ptr mem)
   640(MOVHUloadidx ptr (MOVVconst [c]) mem) && is32Bit(c) => (MOVHUload [int32(c)] ptr mem)
   641(MOVHUloadidx (MOVVconst [c]) ptr mem) && is32Bit(c) => (MOVHUload [int32(c)] ptr mem)
   642(MOVHloadidx ptr (MOVVconst [c]) mem) && is32Bit(c) => (MOVHload [int32(c)] ptr mem)
   643(MOVHloadidx (MOVVconst [c]) ptr mem) && is32Bit(c) => (MOVHload [int32(c)] ptr mem)
   644(MOVBUloadidx ptr (MOVVconst [c]) mem) && is32Bit(c) => (MOVBUload [int32(c)] ptr mem)
   645(MOVBUloadidx (MOVVconst [c]) ptr mem) && is32Bit(c) => (MOVBUload [int32(c)] ptr mem)
   646(MOVBloadidx ptr (MOVVconst [c]) mem) && is32Bit(c) => (MOVBload [int32(c)] ptr mem)
   647(MOVBloadidx (MOVVconst [c]) ptr mem) && is32Bit(c) => (MOVBload [int32(c)] ptr mem)
   648(MOVFloadidx ptr (MOVVconst [c]) mem) && is32Bit(c) => (MOVFload [int32(c)] ptr mem)
   649(MOVFloadidx (MOVVconst [c]) ptr mem) && is32Bit(c) => (MOVFload [int32(c)] ptr mem)
   650(MOVDloadidx ptr (MOVVconst [c]) mem) && is32Bit(c) => (MOVDload [int32(c)] ptr mem)
   651(MOVDloadidx (MOVVconst [c]) ptr mem) && is32Bit(c) => (MOVDload [int32(c)] ptr mem)
   652
   653// register indexed store
   654(MOVVstore [off] {sym} (ADDV ptr idx) val mem) && off == 0 && sym == nil => (MOVVstoreidx ptr idx val mem)
   655(MOVWstore [off] {sym} (ADDV ptr idx) val mem) && off == 0 && sym == nil => (MOVWstoreidx ptr idx val mem)
   656(MOVHstore [off] {sym} (ADDV ptr idx) val mem) && off == 0 && sym == nil => (MOVHstoreidx ptr idx val mem)
   657(MOVBstore [off] {sym} (ADDV ptr idx) val mem) && off == 0 && sym == nil => (MOVBstoreidx ptr idx val mem)
   658(MOVFstore [off] {sym} (ADDV ptr idx) val mem) && off == 0 && sym == nil => (MOVFstoreidx ptr idx val mem)
   659(MOVDstore [off] {sym} (ADDV ptr idx) val mem) && off == 0 && sym == nil => (MOVDstoreidx ptr idx val mem)
   660(MOVVstoreidx ptr (MOVVconst [c]) val mem) && is32Bit(c) => (MOVVstore [int32(c)] ptr val mem)
   661(MOVVstoreidx (MOVVconst [c]) idx val mem) && is32Bit(c) => (MOVVstore [int32(c)] idx val mem)
   662(MOVWstoreidx ptr (MOVVconst [c]) val mem) && is32Bit(c) => (MOVWstore [int32(c)] ptr val mem)
   663(MOVWstoreidx (MOVVconst [c]) idx val mem) && is32Bit(c) => (MOVWstore [int32(c)] idx val mem)
   664(MOVHstoreidx ptr (MOVVconst [c]) val mem) && is32Bit(c) => (MOVHstore [int32(c)] ptr val mem)
   665(MOVHstoreidx (MOVVconst [c]) idx val mem) && is32Bit(c) => (MOVHstore [int32(c)] idx val mem)
   666(MOVBstoreidx ptr (MOVVconst [c]) val mem) && is32Bit(c) => (MOVBstore [int32(c)] ptr val mem)
   667(MOVBstoreidx (MOVVconst [c]) idx val mem) && is32Bit(c) => (MOVBstore [int32(c)] idx val mem)
   668(MOVFstoreidx ptr (MOVVconst [c]) val mem) && is32Bit(c) => (MOVFstore [int32(c)] ptr val mem)
   669(MOVFstoreidx (MOVVconst [c]) idx val mem) && is32Bit(c) => (MOVFstore [int32(c)] idx val mem)
   670(MOVDstoreidx ptr (MOVVconst [c]) val mem) && is32Bit(c) => (MOVDstore [int32(c)] ptr val mem)
   671(MOVDstoreidx (MOVVconst [c]) idx val mem) && is32Bit(c) => (MOVDstore [int32(c)] idx val mem)
   672
   673// register indexed store zero
   674(MOVVstorezero [off] {sym} (ADDV ptr idx) mem) && off == 0 && sym == nil => (MOVVstorezeroidx ptr idx mem)
   675(MOVWstorezero [off] {sym} (ADDV ptr idx) mem) && off == 0 && sym == nil => (MOVWstorezeroidx ptr idx mem)
   676(MOVHstorezero [off] {sym} (ADDV ptr idx) mem) && off == 0 && sym == nil => (MOVHstorezeroidx ptr idx mem)
   677(MOVBstorezero [off] {sym} (ADDV ptr idx) mem) && off == 0 && sym == nil => (MOVBstorezeroidx ptr idx mem)
   678(MOVVstoreidx ptr idx (MOVVconst [0]) mem) => (MOVVstorezeroidx ptr idx mem)
   679(MOVWstoreidx ptr idx (MOVVconst [0]) mem) => (MOVWstorezeroidx ptr idx mem)
   680(MOVHstoreidx ptr idx (MOVVconst [0]) mem) => (MOVHstorezeroidx ptr idx mem)
   681(MOVBstoreidx ptr idx (MOVVconst [0]) mem) => (MOVBstorezeroidx ptr idx mem)
   682(MOVVstorezeroidx ptr (MOVVconst [c]) mem) && is32Bit(c) => (MOVVstorezero [int32(c)] ptr mem)
   683(MOVVstorezeroidx (MOVVconst [c]) idx mem) && is32Bit(c) => (MOVVstorezero [int32(c)] idx mem)
   684(MOVWstorezeroidx ptr (MOVVconst [c]) mem) && is32Bit(c) => (MOVWstorezero [int32(c)] ptr mem)
   685(MOVWstorezeroidx (MOVVconst [c]) idx mem) && is32Bit(c) => (MOVWstorezero [int32(c)] idx mem)
   686(MOVHstorezeroidx ptr (MOVVconst [c]) mem) && is32Bit(c) => (MOVHstorezero [int32(c)] ptr mem)
   687(MOVHstorezeroidx (MOVVconst [c]) idx mem) && is32Bit(c) => (MOVHstorezero [int32(c)] idx mem)
   688(MOVBstorezeroidx ptr (MOVVconst [c]) mem) && is32Bit(c) => (MOVBstorezero [int32(c)] ptr mem)
   689(MOVBstorezeroidx (MOVVconst [c]) idx mem) && is32Bit(c) => (MOVBstorezero [int32(c)] idx mem)
   690
   691// if a register move has only 1 use, just use the same register without emitting instruction
   692// MOVVnop doesn't emit instruction, only for ensuring the type.
   693(MOVVreg x) && x.Uses == 1 => (MOVVnop x)
   694
   695// TODO: we should be able to get rid of MOVVnop all together.
   696// But for now, this is enough to get rid of lots of them.
   697(MOVVnop (MOVVconst [c])) => (MOVVconst [c])
   698
   699// fold constant into arithmetic ops
   700(ADDV x (MOVVconst <t> [c])) && is32Bit(c) && !t.IsPtr() => (ADDVconst [c] x)
   701(SUBV x (MOVVconst [c])) && is32Bit(c) => (SUBVconst [c] x)
   702(AND x (MOVVconst [c])) && is32Bit(c) => (ANDconst [c] x)
   703(OR  x (MOVVconst [c])) && is32Bit(c) => (ORconst  [c] x)
   704(XOR x (MOVVconst [c])) && is32Bit(c) => (XORconst [c] x)
   705(NOR x (MOVVconst [c])) && is32Bit(c) => (NORconst [c] x)
   706
   707(SLL _ (MOVVconst [c])) && uint64(c)>=32 => (MOVVconst [0])
   708(SLLV _ (MOVVconst [c])) && uint64(c)>=64 => (MOVVconst [0])
   709(SRL _ (MOVVconst [c])) && uint64(c)>=32 => (MOVVconst [0])
   710(SRLV _ (MOVVconst [c])) && uint64(c)>=64 => (MOVVconst [0])
   711(SRA x (MOVVconst [c])) && uint64(c)>=32 => (SRAconst x [31])
   712(SRAV x (MOVVconst [c])) && uint64(c)>=64 => (SRAVconst x [63])
   713(SLL x (MOVVconst [c])) && uint64(c) >=0 && uint64(c) <=31 => (SLLconst x [c])
   714(SLLV x (MOVVconst [c])) => (SLLVconst x [c])
   715(SRL x (MOVVconst [c])) && uint64(c) >=0 && uint64(c) <=31 => (SRLconst x [c])
   716(SRLV x (MOVVconst [c])) => (SRLVconst x [c])
   717(SRA x (MOVVconst [c])) && uint64(c) >=0 && uint64(c) <=31 => (SRAconst x [c])
   718(SRAV x (MOVVconst [c])) => (SRAVconst x [c])
   719(ROTR x (MOVVconst [c]))  => (ROTRconst x [c&31])
   720(ROTRV x (MOVVconst [c])) => (ROTRVconst x [c&63])
   721
   722// SLLV/SRLV/SRAV only considers the bottom 6 bits of y, similarly SLL/SRL/SRA only considers the
   723// bottom 5 bits of y.
   724(SLL x (ANDconst [31] y)) => (SLL x y)
   725(SRL x (ANDconst [31] y)) => (SRL x y)
   726(SRA x (ANDconst [31] y)) => (SRA x y)
   727(SLLV x (ANDconst [63] y)) => (SLLV x y)
   728(SRLV x (ANDconst [63] y)) => (SRLV x y)
   729(SRAV x (ANDconst [63] y)) => (SRAV x y)
   730
   731// Avoid unnecessary zero and sign extension when right shifting.
   732(SRLVconst [rc] (MOVWUreg y)) && rc >= 0 && rc <= 31 => (SRLconst [int64(rc)] y)
   733(SRAVconst [rc] (MOVWreg y)) && rc >= 0 && rc <= 31 => (SRAconst [int64(rc)] y)
   734
   735// Replace right shifts that exceed size of signed type.
   736(SRAVconst <t> [rc] (MOVBreg y)) && rc >=  8 => (SRAVconst [63] (SLLVconst <t> [56] y))
   737(SRAVconst <t> [rc] (MOVHreg y)) && rc >= 16 => (SRAVconst  [63] (SLLVconst <t> [48] y))
   738(SRAVconst <t> [rc] (MOVWreg y)) && rc >= 32 => (SRAconst [31] y)
   739
   740// If the shift amount is larger than the datasize(32, 16, 8), we can optimize to constant 0.
   741(MOVWUreg (SLLVconst [lc] x)) && lc >= 32 => (MOVVconst [0])
   742(MOVHUreg (SLLVconst [lc] x)) && lc >= 16 => (MOVVconst [0])
   743(MOVBUreg (SLLVconst [lc] x)) && lc >= 8 => (MOVVconst [0])
   744
   745// After zero extension, the upper (64-datasize(32|16|8)) bits are zero, we can optimize to constant 0.
   746(SRLVconst [rc] (MOVWUreg x)) && rc >= 32 => (MOVVconst [0])
   747(SRLVconst [rc] (MOVHUreg x)) && rc >= 16 => (MOVVconst [0])
   748(SRLVconst [rc] (MOVBUreg x)) && rc >= 8 => (MOVVconst [0])
   749
   750// mul by constant
   751(MULV x (MOVVconst [-1])) => (NEGV x)
   752(MULV _ (MOVVconst [0])) => (MOVVconst [0])
   753(MULV x (MOVVconst [1])) => x
   754(MULV x (MOVVconst [c])) && isPowerOfTwo(c) => (SLLVconst [log64(c)] x)
   755
   756// div by constant
   757(DIVVU x (MOVVconst [1])) => x
   758(DIVVU x (MOVVconst [c])) && isPowerOfTwo(c) => (SRLVconst [log64(c)] x)
   759(REMVU _ (MOVVconst [1])) => (MOVVconst [0])                       // mod
   760(REMVU x (MOVVconst [c])) && isPowerOfTwo(c) => (ANDconst [c-1] x) // mod
   761
   762// FMA
   763(FMA ...) => (FMADDD ...)
   764((ADD|SUB)F (MULF x y) z) && z.Block.Func.useFMA(v) => (FM(ADD|SUB)F x y z)
   765((ADD|SUB)D (MULD x y) z) && z.Block.Func.useFMA(v) => (FM(ADD|SUB)D x y z)
   766// z - xy -> -(xy - z)
   767(SUBF z (MULF x y)) && z.Block.Func.useFMA(v) => (FNMSUBF x y z)
   768(SUBD z (MULD x y)) && z.Block.Func.useFMA(v) => (FNMSUBD x y z)
   769// z + (-xy) -> -(xy - z)
   770// z - (-xy) -> xy + z
   771((ADD|SUB)F z (NEGF (MULF x y))) && z.Block.Func.useFMA(v) => (F(NMSUB|MADD)F x y z)
   772((ADD|SUB)D z (NEGD (MULD x y))) && z.Block.Func.useFMA(v) => (F(NMSUB|MADD)D x y z)
   773// -xy - z -> -(xy + z)
   774(SUBF (NEGF (MULF x y)) z) && z.Block.Func.useFMA(v) => (FNMADDF x y z)
   775(SUBD (NEGD (MULD x y)) z) && z.Block.Func.useFMA(v) => (FNMADDD x y z)
   776
   777// generic simplifications
   778(ADDV x (NEGV y)) => (SUBV x y)
   779(SUBV x (NEGV y)) => (ADDV x y)
   780(SUBV x x) => (MOVVconst [0])
   781(SUBV (MOVVconst [0]) x) => (NEGV x)
   782(AND x x) => x
   783(OR  x x) => x
   784(XOR x x) => (MOVVconst [0])
   785(ORN x (MOVVconst [-1])) => x
   786(AND x (NORconst [0] y)) => (ANDN x y)
   787(OR x (NORconst [0] y)) => (ORN x y)
   788
   789// Fold negation into subtraction.
   790(NEGV (SUBV x y)) => (SUBV y x)
   791(NEGV <t> s:(ADDVconst [c] (SUBV x y))) && s.Uses == 1 && is12Bit(-c) => (ADDVconst [-c] (SUBV <t> y x))
   792
   793// Double negation.
   794(NEGV (NEGV x)) => x
   795// Fold NEGV into ADDVconst. Take care to keep c in 12 bit range.
   796(NEGV <t> s:(ADDVconst [c] (NEGV x))) && s.Uses == 1 && is12Bit(-c) => (ADDVconst [-c] x)
   797
   798// remove redundant *const ops
   799(ADDVconst [0]  x) => x
   800(SUBVconst [0]  x) => x
   801(ANDconst [0]  _) => (MOVVconst [0])
   802(ANDconst [-1] x) => x
   803(ORconst  [0]  x) => x
   804(ORconst  [-1] _) => (MOVVconst [-1])
   805(XORconst [0]  x) => x
   806(XORconst [-1] x) => (NORconst [0] x)
   807(MASKEQZ (MOVVconst [0]) cond) => (MOVVconst [0])
   808(MASKNEZ (MOVVconst [0]) cond) => (MOVVconst [0])
   809(MASKEQZ x (MOVVconst [c])) && c == 0 => (MOVVconst [0])
   810(MASKEQZ x (MOVVconst [c])) && c != 0 => x
   811
   812// generic constant folding
   813(ADDVconst [c] (MOVVconst [d]))  => (MOVVconst [c+d])
   814(ADDVconst [c] (ADDVconst [d] x)) && is32Bit(c+d) => (ADDVconst [c+d] x)
   815(ADDVconst [c] (SUBVconst [d] x)) && is32Bit(c-d) => (ADDVconst [c-d] x)
   816(SUBVconst [c] (MOVVconst [d]))  => (MOVVconst [d-c])
   817(SUBVconst [c] (SUBVconst [d] x)) && is32Bit(-c-d) => (ADDVconst [-c-d] x)
   818(SUBVconst [c] (ADDVconst [d] x)) && is32Bit(-c+d) => (ADDVconst [-c+d] x)
   819(SUBV (MOVVconst [c]) (NEGV (SUBVconst [d] x))) => (ADDVconst [c-d] x)
   820(SLLVconst [c] (MOVVconst [d]))  => (MOVVconst [d<<uint64(c)])
   821(SRLVconst [c] (MOVVconst [d]))  => (MOVVconst [int64(uint64(d)>>uint64(c))])
   822(SRAVconst [c] (MOVVconst [d]))  => (MOVVconst [d>>uint64(c)])
   823(MULV (MOVVconst [c]) (MOVVconst [d])) => (MOVVconst [c*d])
   824(DIVV  (MOVVconst [c]) (MOVVconst [d])) && d != 0 => (MOVVconst [c/d])
   825(DIVVU (MOVVconst [c]) (MOVVconst [d])) && d != 0 => (MOVVconst [int64(uint64(c)/uint64(d))])
   826(REMV  (MOVVconst [c]) (MOVVconst [d])) && d != 0 => (MOVVconst [c%d])   // mod
   827(REMVU (MOVVconst [c]) (MOVVconst [d])) && d != 0 => (MOVVconst [int64(uint64(c)%uint64(d))]) // mod
   828(ANDconst [c] (MOVVconst [d])) => (MOVVconst [c&d])
   829(ANDconst [c] (ANDconst [d] x)) => (ANDconst [c&d] x)
   830(ORconst [c] (MOVVconst [d])) => (MOVVconst [c|d])
   831(ORconst [c] (ORconst [d] x)) && is32Bit(c|d) => (ORconst [c|d] x)
   832(XORconst [c] (MOVVconst [d])) => (MOVVconst [c^d])
   833(XORconst [c] (XORconst [d] x)) && is32Bit(c^d) => (XORconst [c^d] x)
   834(NORconst [c] (MOVVconst [d])) => (MOVVconst [^(c|d)])
   835(NEGV (MOVVconst [c])) => (MOVVconst [-c])
   836(MOVBreg  (MOVVconst [c])) => (MOVVconst [int64(int8(c))])
   837(MOVBUreg (MOVVconst [c])) => (MOVVconst [int64(uint8(c))])
   838(MOVHreg  (MOVVconst [c])) => (MOVVconst [int64(int16(c))])
   839(MOVHUreg (MOVVconst [c])) => (MOVVconst [int64(uint16(c))])
   840(MOVWreg  (MOVVconst [c])) => (MOVVconst [int64(int32(c))])
   841(MOVWUreg (MOVVconst [c])) => (MOVVconst [int64(uint32(c))])
   842(MOVVreg  (MOVVconst [c])) => (MOVVconst [c])
   843
   844(MOVBUreg (ANDconst [c] x)) => (ANDconst [c&0xff] x)
   845
   846// Prefetch instructions (hint specified using aux field)
   847// For PRELD{,X} A value of hint indicates:
   848//    hint=0 is defined as load prefetch to L1-cache
   849//    hint=2 is defined as load prefetch to L3-cache
   850// The PrefetchCacheStreamed implementation prefetches 512 bytes of data
   851// into L3. The aux field are defined as follows:
   852//    bit[4:0]:
   853//       $hint parameter of PRELDX instruction
   854//    bit[41:5]:
   855//       $n parameter of PRELDX instruction, bit[0] of $n is the address
   856//       sequence, bits[11:1] is the block size, bits[20:12] is the block
   857//       num, bits[36:21] is the stride, for more details about $n, refer
   858//       to src/cmd/internal/obj/loong64/doc.go
   859(PrefetchCache addr mem)         => (PRELD  addr mem [0])
   860(PrefetchCacheStreamed addr mem) => (PRELDX addr mem [(((512 << 1) + (1 << 12)) << 5) + 2])
   861
   862// constant comparisons
   863(SGTconst [c] (MOVVconst [d])) && c>d => (MOVVconst [1])
   864(SGTconst [c] (MOVVconst [d])) && c<=d => (MOVVconst [0])
   865(SGTUconst [c] (MOVVconst [d])) && uint64(c)>uint64(d) => (MOVVconst [1])
   866(SGTUconst [c] (MOVVconst [d])) && uint64(c)<=uint64(d) => (MOVVconst [0])
   867
   868// other known comparisons
   869(SGTconst [c] (MOVBreg _)) && 0x7f < c => (MOVVconst [1])
   870(SGTconst [c] (MOVBreg _)) && c <= -0x80 => (MOVVconst [0])
   871(SGTconst [c] (MOVBUreg _)) && 0xff < c => (MOVVconst [1])
   872(SGTconst [c] (MOVBUreg _)) && c < 0 => (MOVVconst [0])
   873(SGTUconst [c] (MOVBUreg _)) && 0xff < uint64(c) => (MOVVconst [1])
   874(SGTconst [c] (MOVHreg _)) && 0x7fff < c => (MOVVconst [1])
   875(SGTconst [c] (MOVHreg _)) && c <= -0x8000 => (MOVVconst [0])
   876(SGTconst [c] (MOVHUreg _)) && 0xffff < c => (MOVVconst [1])
   877(SGTconst [c] (MOVHUreg _)) && c < 0 => (MOVVconst [0])
   878(SGTUconst [c] (MOVHUreg _)) && 0xffff < uint64(c) => (MOVVconst [1])
   879(SGTconst [c] (MOVWUreg _)) && c < 0 => (MOVVconst [0])
   880(SGTconst [c] (ANDconst [m] _)) && 0 <= m && m < c => (MOVVconst [1])
   881(SGTUconst [c] (ANDconst [m] _)) && uint64(m) < uint64(c) => (MOVVconst [1])
   882(SGTconst [c] (SRLVconst _ [d])) && 0 <= c && 0 < d && d <= 63 && 0xffffffffffffffff>>uint64(d) < uint64(c) => (MOVVconst [1])
   883(SGTUconst [c] (SRLVconst _ [d])) && 0 < d && d <= 63 && 0xffffffffffffffff>>uint64(d) < uint64(c) => (MOVVconst [1])
   884
   885// SGT/SGTU with known outcomes.
   886(SGT  x x) => (MOVVconst [0])
   887(SGTU x x) => (MOVVconst [0])
   888
   889// Optimizations
   890
   891// Absorb boolean tests into block
   892(NE (FPFlagTrue cmp) yes no) => (FPT cmp yes no)
   893(NE (FPFlagFalse cmp) yes no) => (FPF cmp yes no)
   894(EQ (FPFlagTrue cmp) yes no) => (FPF cmp yes no)
   895(EQ (FPFlagFalse cmp) yes no) => (FPT cmp yes no)
   896(NE (XORconst [1] cmp:(SGT _ _)) yes no) => (EQ cmp yes no)
   897(NE (XORconst [1] cmp:(SGTU _ _)) yes no) => (EQ cmp yes no)
   898(NE (XORconst [1] cmp:(SGTconst _)) yes no) => (EQ cmp yes no)
   899(NE (XORconst [1] cmp:(SGTUconst _)) yes no) => (EQ cmp yes no)
   900(EQ (XORconst [1] cmp:(SGT _ _)) yes no) => (NE cmp yes no)
   901(EQ (XORconst [1] cmp:(SGTU _ _)) yes no) => (NE cmp yes no)
   902(EQ (XORconst [1] cmp:(SGTconst _)) yes no) => (NE cmp yes no)
   903(EQ (XORconst [1] cmp:(SGTUconst _)) yes no) => (NE cmp yes no)
   904(NE (SGTUconst [1] x) yes no) => (EQ x yes no)
   905(EQ (SGTUconst [1] x) yes no) => (NE x yes no)
   906(NE (SGTU x (MOVVconst [0])) yes no) => (NE x yes no)
   907(EQ (SGTU x (MOVVconst [0])) yes no) => (EQ x yes no)
   908(NE (SGTconst [0] x) yes no) => (LTZ x yes no)
   909(EQ (SGTconst [0] x) yes no) => (GEZ x yes no)
   910(NE (SGT x (MOVVconst [0])) yes no) => (GTZ x yes no)
   911(EQ (SGT x (MOVVconst [0])) yes no) => (LEZ x yes no)
   912
   913(EQ (SGTU (MOVVconst [c]) y) yes no) && c >= -2048 && c <= 2047 => (EQ (SGTUconst [c] y) yes no)
   914(NE (SGTU (MOVVconst [c]) y) yes no) && c >= -2048 && c <= 2047 => (NE (SGTUconst [c] y) yes no)
   915(EQ (SUBV x y) yes no) => (BEQ x y yes no)
   916(NE (SUBV x y) yes no) => (BNE x y yes no)
   917(EQ (SGT x y) yes no) => (BGE y x yes no)
   918(NE (SGT x y) yes no) => (BLT y x yes no)
   919(EQ (SGTU x y) yes no) => (BGEU y x yes no)
   920(NE (SGTU x y) yes no) => (BLTU y x yes no)
   921
   922// absorb constants into branches
   923(EQ  (MOVVconst [0]) yes no) => (First yes no)
   924(EQ  (MOVVconst [c]) yes no) && c != 0 => (First no yes)
   925(NE  (MOVVconst [0]) yes no) => (First no yes)
   926(NE  (MOVVconst [c]) yes no) && c != 0 => (First yes no)
   927(LTZ (MOVVconst [c]) yes no) && c <  0 => (First yes no)
   928(LTZ (MOVVconst [c]) yes no) && c >= 0 => (First no yes)
   929(LEZ (MOVVconst [c]) yes no) && c <= 0 => (First yes no)
   930(LEZ (MOVVconst [c]) yes no) && c >  0 => (First no yes)
   931(GTZ (MOVVconst [c]) yes no) && c >  0 => (First yes no)
   932(GTZ (MOVVconst [c]) yes no) && c <= 0 => (First no yes)
   933(GEZ (MOVVconst [c]) yes no) && c >= 0 => (First yes no)
   934(GEZ (MOVVconst [c]) yes no) && c <  0 => (First no yes)
   935
   936// Arch-specific inlining for small or disjoint runtime.memmove
   937// Match post-lowering calls, register version.
   938(SelectN [0] call:(CALLstatic {sym} dst src (MOVVconst [sz]) mem))
   939	&& sz >= 0
   940	&& isSameCall(sym, "runtime.memmove")
   941	&& call.Uses == 1
   942	&& isInlinableMemmove(dst, src, sz, config)
   943	&& clobber(call)
   944	=> (Move [sz] dst src mem)

View as plain text