...

Text file src/cmd/compile/internal/ssa/_gen/S390X.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// Lowering arithmetic
     6(Add(64|Ptr) ...) => (ADD ...)
     7(Add(32|16|8) ...) => (ADDW ...)
     8(Add32F x y) => (Select0 (FADDS x y))
     9(Add64F x y) => (Select0 (FADD x y))
    10
    11(Sub(64|Ptr) ...) => (SUB ...)
    12(Sub(32|16|8) ...) => (SUBW ...)
    13(Sub32F x y) => (Select0 (FSUBS x y))
    14(Sub64F x y) => (Select0 (FSUB x y))
    15
    16(Mul64 ...) => (MULLD ...)
    17(Mul(32|16|8) ...) => (MULLW ...)
    18(Mul32F ...) => (FMULS ...)
    19(Mul64F ...) => (FMUL ...)
    20(Mul64uhilo ...) => (MLGR ...)
    21
    22(Div32F ...) => (FDIVS ...)
    23(Div64F ...) => (FDIV ...)
    24
    25(Div64 x y) => (DIVD x y)
    26(Div64u ...) => (DIVDU ...)
    27// DIVW/DIVWU has a 64-bit dividend and a 32-bit divisor,
    28// so a sign/zero extension of the dividend is required.
    29(Div32  x y) => (DIVW  (MOVWreg x) y)
    30(Div32u x y) => (DIVWU (MOVWZreg x) y)
    31(Div16  x y) => (DIVW  (MOVHreg x) (MOVHreg y))
    32(Div16u x y) => (DIVWU (MOVHZreg x) (MOVHZreg y))
    33(Div8   x y) => (DIVW  (MOVBreg x) (MOVBreg y))
    34(Div8u  x y) => (DIVWU (MOVBZreg x) (MOVBZreg y))
    35
    36(Hmul(64|64u) ...) => (MULH(D|DU) ...)
    37(Hmul32  x y) => (SRDconst [32] (MULLD (MOVWreg x) (MOVWreg y)))
    38(Hmul32u x y) => (SRDconst [32] (MULLD (MOVWZreg x) (MOVWZreg y)))
    39
    40(Mod64 x y) => (MODD x y)
    41(Mod64u ...) => (MODDU ...)
    42// MODW/MODWU has a 64-bit dividend and a 32-bit divisor,
    43// so a sign/zero extension of the dividend is required.
    44(Mod32  x y) => (MODW  (MOVWreg x) y)
    45(Mod32u x y) => (MODWU (MOVWZreg x) y)
    46(Mod16  x y) => (MODW  (MOVHreg x) (MOVHreg y))
    47(Mod16u x y) => (MODWU (MOVHZreg x) (MOVHZreg y))
    48(Mod8   x y) => (MODW  (MOVBreg x) (MOVBreg y))
    49(Mod8u  x y) => (MODWU (MOVBZreg x) (MOVBZreg y))
    50
    51// (x + y) / 2 with x>=y -> (x - y) / 2 + y
    52(Avg64u <t> x y) => (ADD (SRDconst <t> (SUB <t> x y) [1]) y)
    53
    54(And64 ...) => (AND ...)
    55(And(32|16|8) ...) => (ANDW ...)
    56
    57(Or64 ...) => (OR ...)
    58(Or(32|16|8) ...) => (ORW ...)
    59
    60(Xor64 ...) => (XOR ...)
    61(Xor(32|16|8) ...) => (XORW ...)
    62
    63(Neg64 ...) => (NEG ...)
    64(Neg(32|16|8) ...) => (NEGW ...)
    65(Neg32F ...) => (FNEGS ...)
    66(Neg64F ...) => (FNEG ...)
    67
    68(Com64 ...) => (NOT ...)
    69(Com(32|16|8) ...) => (NOTW ...)
    70(NOT x) => (XOR (MOVDconst [-1]) x)
    71(NOTW x) => (XORWconst [-1] x)
    72
    73// Lowering boolean ops
    74(AndB ...) => (ANDW ...)
    75(OrB ...) => (ORW ...)
    76(Not x) => (XORWconst [1] x)
    77
    78// Lowering pointer arithmetic
    79(OffPtr [off] ptr:(SP)) => (MOVDaddr [int32(off)] ptr)
    80(OffPtr [off] ptr) && is32Bit(off) => (ADDconst [int32(off)] ptr)
    81(OffPtr [off] ptr) => (ADD (MOVDconst [off]) ptr)
    82
    83// TODO: optimize these cases?
    84(Ctz64NonZero ...) => (Ctz64 ...)
    85(Ctz32NonZero ...) => (Ctz32 ...)
    86
    87// Ctz(x) = 64 - findLeftmostOne((x-1)&^x)
    88(Ctz64 <t> x) => (SUB (MOVDconst [64]) (FLOGR (AND <t> (SUBconst <t> [1] x) (NOT <t> x))))
    89(Ctz32 <t> x) => (SUB (MOVDconst [64]) (FLOGR (MOVWZreg (ANDW <t> (SUBWconst <t> [1] x) (NOTW <t> x)))))
    90
    91(BitLen64 x) => (SUB (MOVDconst [64]) (FLOGR x))
    92
    93// POPCNT treats the input register as a vector of 8 bytes, producing
    94// a population count for each individual byte. For inputs larger than
    95// a single byte we therefore need to sum the individual bytes produced
    96// by the POPCNT instruction. For example, the following instruction
    97// sequence could be used to calculate the population count of a 4-byte
    98// value:
    99//
   100//     MOVD   $0x12345678, R1 // R1=0x12345678 <-- input
   101//     POPCNT R1, R2          // R2=0x02030404
   102//     SRW    $16, R2, R3     // R3=0x00000203
   103//     ADDW   R2, R3, R4      // R4=0x02030607
   104//     SRW    $8, R4, R5      // R5=0x00020306
   105//     ADDW   R4, R5, R6      // R6=0x0205090d
   106//     MOVBZ  R6, R7          // R7=0x0000000d <-- result is 13
   107//
   108(PopCount8  x) => (POPCNT (MOVBZreg x))
   109(PopCount16 x) => (MOVBZreg (SumBytes2 (POPCNT <typ.UInt16> x)))
   110(PopCount32 x) => (MOVBZreg (SumBytes4 (POPCNT <typ.UInt32> x)))
   111(PopCount64 x) => (MOVBZreg (SumBytes8 (POPCNT <typ.UInt64> x)))
   112
   113// SumBytes{2,4,8} pseudo operations sum the values of the rightmost
   114// 2, 4 or 8 bytes respectively. The result is a single byte however
   115// other bytes might contain junk so a zero extension is required if
   116// the desired output type is larger than 1 byte.
   117(SumBytes2 x) => (ADDW (SRWconst <typ.UInt8> x [8]) x)
   118(SumBytes4 x) => (SumBytes2 (ADDW <typ.UInt16> (SRWconst <typ.UInt16> x [16]) x))
   119(SumBytes8 x) => (SumBytes4 (ADDW <typ.UInt32> (SRDconst <typ.UInt32> x [32]) x))
   120
   121(Bswap64 ...) => (MOVDBR ...)
   122(Bswap32 ...) => (MOVWBR ...)
   123
   124// add with carry
   125(Select0 (Add64carry x y c))
   126  => (Select0 <typ.UInt64> (ADDE x y (Select1 <types.TypeFlags> (ADDCconst c [-1]))))
   127(Select1 (Add64carry x y c))
   128  => (Select0 <typ.UInt64> (ADDE (MOVDconst [0]) (MOVDconst [0]) (Select1 <types.TypeFlags> (ADDE x y (Select1 <types.TypeFlags> (ADDCconst c [-1]))))))
   129
   130// subtract with borrow
   131(Select0 (Sub64borrow x y c))
   132  => (Select0 <typ.UInt64> (SUBE x y (Select1 <types.TypeFlags> (SUBC (MOVDconst [0]) c))))
   133(Select1 (Sub64borrow x y c))
   134  => (NEG (Select0 <typ.UInt64> (SUBE (MOVDconst [0]) (MOVDconst [0]) (Select1 <types.TypeFlags> (SUBE x y (Select1 <types.TypeFlags> (SUBC (MOVDconst [0]) c)))))))
   135
   136// math package intrinsics
   137(Sqrt      ...) => (FSQRT ...)
   138(Floor       x) => (FIDBR [7] x)
   139(Ceil        x) => (FIDBR [6] x)
   140(Trunc       x) => (FIDBR [5] x)
   141(RoundToEven x) => (FIDBR [4] x)
   142(Round       x) => (FIDBR [1] x)
   143(FMA     x y z) => (FMADD z x y)
   144
   145(Sqrt32    ...) => (FSQRTS ...)
   146
   147// Atomic loads and stores.
   148// The SYNC instruction (fast-BCR-serialization) prevents store-load
   149// reordering. Other sequences of memory operations (load-load,
   150// store-store and load-store) are already guaranteed not to be reordered.
   151(AtomicLoad(8|32|Acq32|64|Ptr) ptr mem) => (MOV(BZ|WZ|WZ|D|D)atomicload ptr mem)
   152(AtomicStore(8|32|64|PtrNoWB) ptr val mem) => (SYNC (MOV(B|W|D|D)atomicstore ptr val mem))
   153
   154// Store-release doesn't require store-load ordering.
   155(AtomicStoreRel32 ptr val mem) => (MOVWatomicstore ptr val mem)
   156
   157// Atomic adds.
   158(AtomicAdd32 ptr val mem) => (AddTupleFirst32 val (LAA ptr val mem))
   159(AtomicAdd64 ptr val mem) => (AddTupleFirst64 val (LAAG ptr val mem))
   160(Select0 <t> (AddTupleFirst32 val tuple)) => (ADDW val (Select0 <t> tuple))
   161(Select1     (AddTupleFirst32   _ tuple)) => (Select1 tuple)
   162(Select0 <t> (AddTupleFirst64 val tuple)) => (ADD val (Select0 <t> tuple))
   163(Select1     (AddTupleFirst64   _ tuple)) => (Select1 tuple)
   164
   165// Atomic exchanges.
   166(AtomicExchange32 ptr val mem) => (LoweredAtomicExchange32 ptr val mem)
   167(AtomicExchange64 ptr val mem) => (LoweredAtomicExchange64 ptr val mem)
   168
   169// Atomic compare and swap.
   170(AtomicCompareAndSwap32 ptr old new_ mem) => (LoweredAtomicCas32 ptr old new_ mem)
   171(AtomicCompareAndSwap64 ptr old new_ mem) => (LoweredAtomicCas64 ptr old new_ mem)
   172
   173// Atomic and: *(*uint8)(ptr) &= val
   174//
   175// Round pointer down to nearest word boundary and pad value with ones before
   176// applying atomic AND operation to target word.
   177//
   178// *(*uint32)(ptr &^ 3) &= rotateleft(uint32(val) | 0xffffff00, ((3 << 3) ^ ((ptr & 3) << 3))
   179//
   180(AtomicAnd8 ptr val mem)
   181  => (LANfloor
   182       ptr
   183       (RLL <typ.UInt32>
   184         (ORWconst <typ.UInt32> val [-1<<8])
   185         (RXSBG <typ.UInt32> {s390x.NewRotateParams(59, 60, 3)} (MOVDconst [3<<3]) ptr))
   186       mem)
   187
   188// Atomic or: *(*uint8)(ptr) |= val
   189//
   190// Round pointer down to nearest word boundary and pad value with zeros before
   191// applying atomic OR operation to target word.
   192//
   193// *(*uint32)(ptr &^ 3) |= uint32(val) << ((3 << 3) ^ ((ptr & 3) << 3))
   194//
   195(AtomicOr8  ptr val mem)
   196  => (LAOfloor
   197       ptr
   198       (SLW <typ.UInt32>
   199         (MOVBZreg <typ.UInt32> val)
   200         (RXSBG <typ.UInt32> {s390x.NewRotateParams(59, 60, 3)} (MOVDconst [3<<3]) ptr))
   201       mem)
   202
   203(AtomicAnd32 ...) => (LAN ...)
   204(AtomicOr32  ...) => (LAO ...)
   205
   206// Lowering extension
   207// Note: we always extend to 64 bits even though some ops don't need that many result bits.
   208(SignExt8to(16|32|64) ...) => (MOVBreg ...)
   209(SignExt16to(32|64) ...) => (MOVHreg ...)
   210(SignExt32to64 ...) => (MOVWreg ...)
   211
   212(ZeroExt8to(16|32|64) ...) => (MOVBZreg ...)
   213(ZeroExt16to(32|64) ...) => (MOVHZreg ...)
   214(ZeroExt32to64 ...) => (MOVWZreg ...)
   215
   216(Slicemask <t> x) => (SRADconst (NEG <t> x) [63])
   217
   218// Lowering truncation
   219// Because we ignore high parts of registers, truncates are just copies.
   220(Trunc(16|32|64)to8 ...) => (Copy ...)
   221(Trunc(32|64)to16 ...) => (Copy ...)
   222(Trunc64to32 ...) => (Copy ...)
   223
   224// Lowering float <-> int
   225(Cvt32to32F ...) => (CEFBRA ...)
   226(Cvt32to64F ...) => (CDFBRA ...)
   227(Cvt64to32F ...) => (CEGBRA ...)
   228(Cvt64to64F ...) => (CDGBRA ...)
   229
   230(Cvt32Fto32 ...) => (CFEBRA ...)
   231(Cvt32Fto64 ...) => (CGEBRA ...)
   232(Cvt64Fto32 ...) => (CFDBRA ...)
   233(Cvt64Fto64 ...) => (CGDBRA ...)
   234
   235// Lowering float <-> uint
   236(Cvt32Uto32F ...) => (CELFBR ...)
   237(Cvt32Uto64F ...) => (CDLFBR ...)
   238(Cvt64Uto32F ...) => (CELGBR ...)
   239(Cvt64Uto64F ...) => (CDLGBR ...)
   240
   241(Cvt32Fto32U ...) => (CLFEBR ...)
   242(Cvt32Fto64U ...) => (CLGEBR ...)
   243(Cvt64Fto32U ...) => (CLFDBR ...)
   244(Cvt64Fto64U ...) => (CLGDBR ...)
   245
   246// Lowering float32 <-> float64
   247(Cvt32Fto64F ...) => (LDEBR ...)
   248(Cvt64Fto32F ...) => (LEDBR ...)
   249
   250(CvtBoolToUint8 ...) => (Copy ...)
   251
   252(Round(32|64)F ...) => (LoweredRound(32|64)F ...)
   253
   254// Lowering shifts
   255
   256// Lower bounded shifts first. No need to check shift value.
   257(Lsh64x(64|32|16|8)  x y) && shiftIsBounded(v) => (SLD x y)
   258(Lsh32x(64|32|16|8)  x y) && shiftIsBounded(v) => (SLW x y)
   259(Lsh16x(64|32|16|8)  x y) && shiftIsBounded(v) => (SLW x y)
   260(Lsh8x(64|32|16|8)   x y) && shiftIsBounded(v) => (SLW x y)
   261(Rsh64Ux(64|32|16|8) x y) && shiftIsBounded(v) => (SRD x y)
   262(Rsh32Ux(64|32|16|8) x y) && shiftIsBounded(v) => (SRW x y)
   263(Rsh16Ux(64|32|16|8) x y) && shiftIsBounded(v) => (SRW (MOVHZreg x) y)
   264(Rsh8Ux(64|32|16|8)  x y) && shiftIsBounded(v) => (SRW (MOVBZreg x) y)
   265(Rsh64x(64|32|16|8)  x y) && shiftIsBounded(v) => (SRAD x y)
   266(Rsh32x(64|32|16|8)  x y) && shiftIsBounded(v) => (SRAW x y)
   267(Rsh16x(64|32|16|8)  x y) && shiftIsBounded(v) => (SRAW (MOVHreg x) y)
   268(Rsh8x(64|32|16|8)   x y) && shiftIsBounded(v) => (SRAW (MOVBreg x) y)
   269
   270// Unsigned shifts need to return 0 if shift amount is >= width of shifted value.
   271//   result = shift >= 64 ? 0 : arg << shift
   272(Lsh(64|32|16|8)x64 <t> x y) => (LOCGR {s390x.GreaterOrEqual} <t> (SL(D|W|W|W) <t> x y) (MOVDconst [0]) (CMPUconst y [64]))
   273(Lsh(64|32|16|8)x32 <t> x y) => (LOCGR {s390x.GreaterOrEqual} <t> (SL(D|W|W|W) <t> x y) (MOVDconst [0]) (CMPWUconst y [64]))
   274(Lsh(64|32|16|8)x16 <t> x y) => (LOCGR {s390x.GreaterOrEqual} <t> (SL(D|W|W|W) <t> x y) (MOVDconst [0]) (CMPWUconst (MOVHZreg y) [64]))
   275(Lsh(64|32|16|8)x8  <t> x y) => (LOCGR {s390x.GreaterOrEqual} <t> (SL(D|W|W|W) <t> x y) (MOVDconst [0]) (CMPWUconst (MOVBZreg y) [64]))
   276
   277(Rsh(64|32)Ux64 <t> x y) => (LOCGR {s390x.GreaterOrEqual} <t> (SR(D|W) <t> x y) (MOVDconst [0]) (CMPUconst y [64]))
   278(Rsh(64|32)Ux32 <t> x y) => (LOCGR {s390x.GreaterOrEqual} <t> (SR(D|W) <t> x y) (MOVDconst [0]) (CMPWUconst y [64]))
   279(Rsh(64|32)Ux16 <t> x y) => (LOCGR {s390x.GreaterOrEqual} <t> (SR(D|W) <t> x y) (MOVDconst [0]) (CMPWUconst (MOVHZreg y) [64]))
   280(Rsh(64|32)Ux8  <t> x y) => (LOCGR {s390x.GreaterOrEqual} <t> (SR(D|W) <t> x y) (MOVDconst [0]) (CMPWUconst (MOVBZreg y) [64]))
   281
   282(Rsh(16|8)Ux64 <t> x y) => (LOCGR {s390x.GreaterOrEqual} <t> (SRW <t> (MOV(H|B)Zreg x) y) (MOVDconst [0]) (CMPUconst y [64]))
   283(Rsh(16|8)Ux32 <t> x y) => (LOCGR {s390x.GreaterOrEqual} <t> (SRW <t> (MOV(H|B)Zreg x) y) (MOVDconst [0]) (CMPWUconst y [64]))
   284(Rsh(16|8)Ux16 <t> x y) => (LOCGR {s390x.GreaterOrEqual} <t> (SRW <t> (MOV(H|B)Zreg x) y) (MOVDconst [0]) (CMPWUconst (MOVHZreg y) [64]))
   285(Rsh(16|8)Ux8  <t> x y) => (LOCGR {s390x.GreaterOrEqual} <t> (SRW <t> (MOV(H|B)Zreg x) y) (MOVDconst [0]) (CMPWUconst (MOVBZreg y) [64]))
   286
   287// Signed right shift needs to return 0/-1 if shift amount is >= width of shifted value.
   288// We implement this by setting the shift value to 63 (all ones) if the shift value is more than 63.
   289//   result = arg >> (shift >= 64 ? 63 : shift)
   290(Rsh(64|32)x64 x y) => (SRA(D|W) x (LOCGR {s390x.GreaterOrEqual} <y.Type> y (MOVDconst <y.Type> [63]) (CMPUconst  y [64])))
   291(Rsh(64|32)x32 x y) => (SRA(D|W) x (LOCGR {s390x.GreaterOrEqual} <y.Type> y (MOVDconst <y.Type> [63]) (CMPWUconst y [64])))
   292(Rsh(64|32)x16 x y) => (SRA(D|W) x (LOCGR {s390x.GreaterOrEqual} <y.Type> y (MOVDconst <y.Type> [63]) (CMPWUconst (MOVHZreg y) [64])))
   293(Rsh(64|32)x8  x y) => (SRA(D|W) x (LOCGR {s390x.GreaterOrEqual} <y.Type> y (MOVDconst <y.Type> [63]) (CMPWUconst (MOVBZreg y) [64])))
   294
   295(Rsh(16|8)x64 x y) => (SRAW (MOV(H|B)reg x) (LOCGR {s390x.GreaterOrEqual} <y.Type> y (MOVDconst <y.Type> [63]) (CMPUconst  y [64])))
   296(Rsh(16|8)x32 x y) => (SRAW (MOV(H|B)reg x) (LOCGR {s390x.GreaterOrEqual} <y.Type> y (MOVDconst <y.Type> [63]) (CMPWUconst y [64])))
   297(Rsh(16|8)x16 x y) => (SRAW (MOV(H|B)reg x) (LOCGR {s390x.GreaterOrEqual} <y.Type> y (MOVDconst <y.Type> [63]) (CMPWUconst (MOVHZreg y) [64])))
   298(Rsh(16|8)x8  x y) => (SRAW (MOV(H|B)reg x) (LOCGR {s390x.GreaterOrEqual} <y.Type> y (MOVDconst <y.Type> [63]) (CMPWUconst (MOVBZreg y) [64])))
   299
   300// Lowering rotates
   301(RotateLeft8 <t> x (MOVDconst [c])) => (Or8 (Lsh8x64 <t> x (MOVDconst [c&7])) (Rsh8Ux64 <t> x (MOVDconst [-c&7])))
   302(RotateLeft16 <t> x (MOVDconst [c])) => (Or16 (Lsh16x64 <t> x (MOVDconst [c&15])) (Rsh16Ux64 <t> x (MOVDconst [-c&15])))
   303(RotateLeft32 ...) => (RLL  ...)
   304(RotateLeft64 ...) => (RLLG ...)
   305
   306// Lowering comparisons
   307(Less64      x y) => (LOCGR {s390x.Less} (MOVDconst [0]) (MOVDconst [1]) (CMP x y))
   308(Less32      x y) => (LOCGR {s390x.Less} (MOVDconst [0]) (MOVDconst [1]) (CMPW x y))
   309(Less(16|8)  x y) => (LOCGR {s390x.Less} (MOVDconst [0]) (MOVDconst [1]) (CMPW (MOV(H|B)reg x) (MOV(H|B)reg y)))
   310(Less64U     x y) => (LOCGR {s390x.Less} (MOVDconst [0]) (MOVDconst [1]) (CMPU x y))
   311(Less32U     x y) => (LOCGR {s390x.Less} (MOVDconst [0]) (MOVDconst [1]) (CMPWU x y))
   312(Less(16|8)U x y) => (LOCGR {s390x.Less} (MOVDconst [0]) (MOVDconst [1]) (CMPWU (MOV(H|B)Zreg x) (MOV(H|B)Zreg y)))
   313(Less64F     x y) => (LOCGR {s390x.Less} (MOVDconst [0]) (MOVDconst [1]) (FCMP x y))
   314(Less32F     x y) => (LOCGR {s390x.Less} (MOVDconst [0]) (MOVDconst [1]) (FCMPS x y))
   315
   316(Leq64      x y) => (LOCGR {s390x.LessOrEqual} (MOVDconst [0]) (MOVDconst [1]) (CMP x y))
   317(Leq32      x y) => (LOCGR {s390x.LessOrEqual} (MOVDconst [0]) (MOVDconst [1]) (CMPW x y))
   318(Leq(16|8)  x y) => (LOCGR {s390x.LessOrEqual} (MOVDconst [0]) (MOVDconst [1]) (CMPW (MOV(H|B)reg x) (MOV(H|B)reg y)))
   319(Leq64U     x y) => (LOCGR {s390x.LessOrEqual} (MOVDconst [0]) (MOVDconst [1]) (CMPU x y))
   320(Leq32U     x y) => (LOCGR {s390x.LessOrEqual} (MOVDconst [0]) (MOVDconst [1]) (CMPWU x y))
   321(Leq(16|8)U x y) => (LOCGR {s390x.LessOrEqual} (MOVDconst [0]) (MOVDconst [1]) (CMPWU (MOV(H|B)Zreg x) (MOV(H|B)Zreg y)))
   322(Leq64F     x y) => (LOCGR {s390x.LessOrEqual} (MOVDconst [0]) (MOVDconst [1]) (FCMP x y))
   323(Leq32F     x y) => (LOCGR {s390x.LessOrEqual} (MOVDconst [0]) (MOVDconst [1]) (FCMPS x y))
   324
   325(Eq(64|Ptr) x y) => (LOCGR {s390x.Equal} (MOVDconst [0]) (MOVDconst [1]) (CMP x y))
   326(Eq32       x y) => (LOCGR {s390x.Equal} (MOVDconst [0]) (MOVDconst [1]) (CMPW x y))
   327(Eq(16|8|B) x y) => (LOCGR {s390x.Equal} (MOVDconst [0]) (MOVDconst [1]) (CMPW (MOV(H|B|B)reg x) (MOV(H|B|B)reg y)))
   328(Eq64F      x y) => (LOCGR {s390x.Equal} (MOVDconst [0]) (MOVDconst [1]) (FCMP x y))
   329(Eq32F      x y) => (LOCGR {s390x.Equal} (MOVDconst [0]) (MOVDconst [1]) (FCMPS x y))
   330
   331(Neq(64|Ptr) x y) => (LOCGR {s390x.NotEqual} (MOVDconst [0]) (MOVDconst [1]) (CMP x y))
   332(Neq32       x y) => (LOCGR {s390x.NotEqual} (MOVDconst [0]) (MOVDconst [1]) (CMPW x y))
   333(Neq(16|8|B) x y) => (LOCGR {s390x.NotEqual} (MOVDconst [0]) (MOVDconst [1]) (CMPW (MOV(H|B|B)reg x) (MOV(H|B|B)reg y)))
   334(Neq64F      x y) => (LOCGR {s390x.NotEqual} (MOVDconst [0]) (MOVDconst [1]) (FCMP x y))
   335(Neq32F      x y) => (LOCGR {s390x.NotEqual} (MOVDconst [0]) (MOVDconst [1]) (FCMPS x y))
   336
   337// Lowering loads
   338(Load <t> ptr mem) && (is64BitInt(t) || isPtr(t)) => (MOVDload ptr mem)
   339(Load <t> ptr mem) && is32BitInt(t) &&  t.IsSigned() => (MOVWload ptr mem)
   340(Load <t> ptr mem) && is32BitInt(t) && !t.IsSigned() => (MOVWZload ptr mem)
   341(Load <t> ptr mem) && is16BitInt(t) &&  t.IsSigned() => (MOVHload ptr mem)
   342(Load <t> ptr mem) && is16BitInt(t) && !t.IsSigned() => (MOVHZload ptr mem)
   343(Load <t> ptr mem) && is8BitInt(t)  &&  t.IsSigned() => (MOVBload ptr mem)
   344(Load <t> ptr mem) && (t.IsBoolean() || (is8BitInt(t) && !t.IsSigned())) => (MOVBZload ptr mem)
   345(Load <t> ptr mem) && is32BitFloat(t) => (FMOVSload ptr mem)
   346(Load <t> ptr mem) && is64BitFloat(t) => (FMOVDload ptr mem)
   347
   348// Lowering stores
   349(Store {t} ptr val mem) && t.Size() == 8 &&  t.IsFloat() => (FMOVDstore ptr val mem)
   350(Store {t} ptr val mem) && t.Size() == 4 &&  t.IsFloat() => (FMOVSstore ptr val mem)
   351(Store {t} ptr val mem) && t.Size() == 8 && !t.IsFloat() => (MOVDstore ptr val mem)
   352(Store {t} ptr val mem) && t.Size() == 4 && !t.IsFloat() => (MOVWstore ptr val mem)
   353(Store {t} ptr val mem) && t.Size() == 2 => (MOVHstore ptr val mem)
   354(Store {t} ptr val mem) && t.Size() == 1 => (MOVBstore ptr val mem)
   355
   356// Lowering moves
   357
   358// Load and store for small copies.
   359(Move [0] _ _ mem) => mem
   360(Move [1] dst src mem) => (MOVBstore dst (MOVBZload src mem) mem)
   361(Move [2] dst src mem) => (MOVHstore dst (MOVHZload src mem) mem)
   362(Move [4] dst src mem) => (MOVWstore dst (MOVWZload src mem) mem)
   363(Move [8] dst src mem) => (MOVDstore dst (MOVDload src mem) mem)
   364(Move [16] dst src mem) =>
   365	(MOVDstore [8] dst (MOVDload [8] src mem)
   366		(MOVDstore dst (MOVDload src mem) mem))
   367(Move [24] dst src mem) =>
   368        (MOVDstore [16] dst (MOVDload [16] src mem)
   369	        (MOVDstore [8] dst (MOVDload [8] src mem)
   370                (MOVDstore dst (MOVDload src mem) mem)))
   371(Move [3] dst src mem) =>
   372	(MOVBstore [2] dst (MOVBZload [2] src mem)
   373		(MOVHstore dst (MOVHZload src mem) mem))
   374(Move [5] dst src mem) =>
   375	(MOVBstore [4] dst (MOVBZload [4] src mem)
   376		(MOVWstore dst (MOVWZload src mem) mem))
   377(Move [6] dst src mem) =>
   378	(MOVHstore [4] dst (MOVHZload [4] src mem)
   379		(MOVWstore dst (MOVWZload src mem) mem))
   380(Move [7] dst src mem) =>
   381	(MOVBstore [6] dst (MOVBZload [6] src mem)
   382		(MOVHstore [4] dst (MOVHZload [4] src mem)
   383			(MOVWstore dst (MOVWZload src mem) mem)))
   384
   385// MVC for other moves. Use up to 4 instructions (sizes up to 1024 bytes).
   386(Move [s] dst src mem) && s > 0 && s <= 256 && logLargeCopy(v, s) =>
   387	(MVC [makeValAndOff(int32(s), 0)] dst src mem)
   388(Move [s] dst src mem) && s > 256 && s <= 512 && logLargeCopy(v, s) =>
   389	(MVC [makeValAndOff(int32(s)-256, 256)] dst src (MVC [makeValAndOff(256, 0)] dst src mem))
   390(Move [s] dst src mem) && s > 512 && s <= 768 && logLargeCopy(v, s) =>
   391	(MVC [makeValAndOff(int32(s)-512, 512)] dst src (MVC [makeValAndOff(256, 256)] dst src (MVC [makeValAndOff(256, 0)] dst src mem)))
   392(Move [s] dst src mem) && s > 768 && s <= 1024 && logLargeCopy(v, s) =>
   393	(MVC [makeValAndOff(int32(s)-768, 768)] dst src (MVC [makeValAndOff(256, 512)] dst src (MVC [makeValAndOff(256, 256)] dst src (MVC [makeValAndOff(256, 0)] dst src mem))))
   394
   395// Move more than 1024 bytes using a loop.
   396(Move [s] dst src mem) && s > 1024 && logLargeCopy(v, s) =>
   397	(LoweredMove [s%256] dst src (ADD <src.Type> src (MOVDconst [(s/256)*256])) mem)
   398
   399// Lowering Zero instructions
   400(Zero [0] _ mem) => mem
   401(Zero [1] destptr mem) => (MOVBstoreconst [0] destptr mem)
   402(Zero [2] destptr mem) => (MOVHstoreconst [0] destptr mem)
   403(Zero [4] destptr mem) => (MOVWstoreconst [0] destptr mem)
   404(Zero [8] destptr mem) => (MOVDstoreconst [0] destptr mem)
   405(Zero [3] destptr mem) =>
   406	(MOVBstoreconst [makeValAndOff(0,2)] destptr
   407		(MOVHstoreconst [0] destptr mem))
   408(Zero [5] destptr mem) =>
   409	(MOVBstoreconst [makeValAndOff(0,4)] destptr
   410		(MOVWstoreconst [0] destptr mem))
   411(Zero [6] destptr mem) =>
   412	(MOVHstoreconst [makeValAndOff(0,4)] destptr
   413		(MOVWstoreconst [0] destptr mem))
   414(Zero [7] destptr mem) =>
   415	(MOVWstoreconst [makeValAndOff(0,3)] destptr
   416		(MOVWstoreconst [0] destptr mem))
   417
   418(Zero [s] destptr mem) && s > 0 && s <= 1024 =>
   419	(CLEAR [makeValAndOff(int32(s), 0)] destptr mem)
   420
   421// Zero more than 1024 bytes using a loop.
   422(Zero [s] destptr mem) && s > 1024 =>
   423	(LoweredZero [s%256] destptr (ADDconst <destptr.Type> destptr [(int32(s)/256)*256]) mem)
   424
   425// Lowering constants
   426(Const(64|32|16|8) [val]) => (MOVDconst [int64(val)])
   427(Const(32|64)F ...) => (FMOV(S|D)const ...)
   428(ConstNil) => (MOVDconst [0])
   429(ConstBool [t]) => (MOVDconst [b2i(t)])
   430
   431// Lowering calls
   432(StaticCall ...) => (CALLstatic ...)
   433(ClosureCall ...) => (CALLclosure ...)
   434(InterCall ...) => (CALLinter ...)
   435(TailCall ...) => (CALLtail ...)
   436
   437// Miscellaneous
   438(IsNonNil p) => (LOCGR {s390x.NotEqual} (MOVDconst [0]) (MOVDconst [1]) (CMPconst p [0]))
   439(IsInBounds idx len) => (LOCGR {s390x.Less} (MOVDconst [0]) (MOVDconst [1]) (CMPU idx len))
   440(IsSliceInBounds idx len) => (LOCGR {s390x.LessOrEqual} (MOVDconst [0]) (MOVDconst [1]) (CMPU idx len))
   441(NilCheck ...) => (LoweredNilCheck ...)
   442(GetG ...) => (LoweredGetG ...)
   443(GetClosurePtr ...) => (LoweredGetClosurePtr ...)
   444(GetCallerSP ...) => (LoweredGetCallerSP ...)
   445(GetCallerPC ...) => (LoweredGetCallerPC ...)
   446(Addr {sym} base) => (MOVDaddr {sym} base)
   447(LocalAddr <t> {sym} base mem) && t.Elem().HasPointers() => (MOVDaddr {sym} (SPanchored base mem))
   448(LocalAddr <t> {sym} base _)  && !t.Elem().HasPointers() => (MOVDaddr {sym} base)
   449(ITab (Load ptr mem)) => (MOVDload ptr mem)
   450
   451// block rewrites
   452(If cond yes no) => (CLIJ {s390x.LessOrGreater} (MOVBZreg <typ.Bool> cond) [0] yes no)
   453
   454// Write barrier.
   455(WB ...) => (LoweredWB ...)
   456
   457(PanicBounds [kind] x y mem) && boundsABI(kind) == 0 => (LoweredPanicBoundsA [kind] x y mem)
   458(PanicBounds [kind] x y mem) && boundsABI(kind) == 1 => (LoweredPanicBoundsB [kind] x y mem)
   459(PanicBounds [kind] x y mem) && boundsABI(kind) == 2 => (LoweredPanicBoundsC [kind] x y mem)
   460
   461// ***************************
   462// Above: lowering rules
   463// Below: optimizations
   464// ***************************
   465// TODO: Should the optimizations be a separate pass?
   466
   467// Note: when removing unnecessary sign/zero extensions.
   468//
   469// After a value is spilled it is restored using a sign- or zero-extension
   470// to register-width as appropriate for its type. For example, a uint8 will
   471// be restored using a MOVBZ (llgc) instruction which will zero extend the
   472// 8-bit value to 64-bits.
   473//
   474// This is a hazard when folding sign- and zero-extensions since we need to
   475// ensure not only that the value in the argument register is correctly
   476// extended but also that it will still be correctly extended if it is
   477// spilled and restored.
   478//
   479// In general this means we need type checks when the RHS of a rule is an
   480// OpCopy (i.e. "(... x:(...) ...) -> x").
   481
   482// Merge double extensions.
   483(MOV(H|HZ)reg e:(MOV(B|BZ)reg x)) && clobberIfDead(e) => (MOV(B|BZ)reg x)
   484(MOV(W|WZ)reg e:(MOV(B|BZ)reg x)) && clobberIfDead(e) => (MOV(B|BZ)reg x)
   485(MOV(W|WZ)reg e:(MOV(H|HZ)reg x)) && clobberIfDead(e) => (MOV(H|HZ)reg x)
   486
   487// Bypass redundant sign extensions.
   488(MOV(B|BZ)reg e:(MOVBreg x)) && clobberIfDead(e) => (MOV(B|BZ)reg x)
   489(MOV(B|BZ)reg e:(MOVHreg x)) && clobberIfDead(e) => (MOV(B|BZ)reg x)
   490(MOV(B|BZ)reg e:(MOVWreg x)) && clobberIfDead(e) => (MOV(B|BZ)reg x)
   491(MOV(H|HZ)reg e:(MOVHreg x)) && clobberIfDead(e) => (MOV(H|HZ)reg x)
   492(MOV(H|HZ)reg e:(MOVWreg x)) && clobberIfDead(e) => (MOV(H|HZ)reg x)
   493(MOV(W|WZ)reg e:(MOVWreg x)) && clobberIfDead(e) => (MOV(W|WZ)reg x)
   494
   495// Bypass redundant zero extensions.
   496(MOV(B|BZ)reg e:(MOVBZreg x)) && clobberIfDead(e) => (MOV(B|BZ)reg x)
   497(MOV(B|BZ)reg e:(MOVHZreg x)) && clobberIfDead(e) => (MOV(B|BZ)reg x)
   498(MOV(B|BZ)reg e:(MOVWZreg x)) && clobberIfDead(e) => (MOV(B|BZ)reg x)
   499(MOV(H|HZ)reg e:(MOVHZreg x)) && clobberIfDead(e) => (MOV(H|HZ)reg x)
   500(MOV(H|HZ)reg e:(MOVWZreg x)) && clobberIfDead(e) => (MOV(H|HZ)reg x)
   501(MOV(W|WZ)reg e:(MOVWZreg x)) && clobberIfDead(e) => (MOV(W|WZ)reg x)
   502
   503// Remove zero extensions after zero extending load.
   504// Note: take care that if x is spilled it is restored correctly.
   505(MOV(B|H|W)Zreg x:(MOVBZload    _   _)) && (!x.Type.IsSigned() || x.Type.Size() > 1) => x
   506(MOV(H|W)Zreg   x:(MOVHZload    _   _)) && (!x.Type.IsSigned() || x.Type.Size() > 2) => x
   507(MOVWZreg       x:(MOVWZload    _   _)) && (!x.Type.IsSigned() || x.Type.Size() > 4) => x
   508
   509// Remove sign extensions after sign extending load.
   510// Note: take care that if x is spilled it is restored correctly.
   511(MOV(B|H|W)reg x:(MOVBload    _   _)) && (x.Type.IsSigned() || x.Type.Size() == 8) => x
   512(MOV(H|W)reg   x:(MOVHload    _   _)) && (x.Type.IsSigned() || x.Type.Size() == 8) => x
   513(MOVWreg       x:(MOVWload    _   _)) && (x.Type.IsSigned() || x.Type.Size() == 8) => x
   514
   515// Remove sign extensions after zero extending load.
   516// These type checks are probably unnecessary but do them anyway just in case.
   517(MOV(H|W)reg x:(MOVBZload    _   _)) && (!x.Type.IsSigned() || x.Type.Size() > 1) => x
   518(MOVWreg     x:(MOVHZload    _   _)) && (!x.Type.IsSigned() || x.Type.Size() > 2) => x
   519
   520// Fold sign and zero extensions into loads.
   521//
   522// Note: The combined instruction must end up in the same block
   523// as the original load. If not, we end up making a value with
   524// memory type live in two different blocks, which can lead to
   525// multiple memory values alive simultaneously.
   526//
   527// Make sure we don't combine these ops if the load has another use.
   528// This prevents a single load from being split into multiple loads
   529// which then might return different values.  See test/atomicload.go.
   530(MOV(B|H|W)Zreg <t> x:(MOV(B|H|W)load [o] {s} p mem))
   531  && x.Uses == 1
   532  && clobber(x)
   533  => @x.Block (MOV(B|H|W)Zload <t> [o] {s} p mem)
   534(MOV(B|H|W)reg <t> x:(MOV(B|H|W)Zload [o] {s} p mem))
   535  && x.Uses == 1
   536  && clobber(x)
   537  => @x.Block (MOV(B|H|W)load <t> [o] {s} p mem)
   538
   539// Remove zero extensions after argument load.
   540(MOVBZreg x:(Arg <t>)) && !t.IsSigned() && t.Size() == 1 => x
   541(MOVHZreg x:(Arg <t>)) && !t.IsSigned() && t.Size() <= 2 => x
   542(MOVWZreg x:(Arg <t>)) && !t.IsSigned() && t.Size() <= 4 => x
   543
   544// Remove sign extensions after argument load.
   545(MOVBreg x:(Arg <t>)) && t.IsSigned() && t.Size() == 1 => x
   546(MOVHreg x:(Arg <t>)) && t.IsSigned() && t.Size() <= 2 => x
   547(MOVWreg x:(Arg <t>)) && t.IsSigned() && t.Size() <= 4 => x
   548
   549// Fold zero extensions into constants.
   550(MOVBZreg (MOVDconst [c])) => (MOVDconst [int64( uint8(c))])
   551(MOVHZreg (MOVDconst [c])) => (MOVDconst [int64(uint16(c))])
   552(MOVWZreg (MOVDconst [c])) => (MOVDconst [int64(uint32(c))])
   553
   554// Fold sign extensions into constants.
   555(MOVBreg (MOVDconst [c])) => (MOVDconst [int64( int8(c))])
   556(MOVHreg (MOVDconst [c])) => (MOVDconst [int64(int16(c))])
   557(MOVWreg (MOVDconst [c])) => (MOVDconst [int64(int32(c))])
   558
   559// Remove zero extension of conditional move.
   560// Note: only for MOVBZreg for now since it is added as part of 'if' statement lowering.
   561(MOVBZreg x:(LOCGR (MOVDconst [c]) (MOVDconst [d]) _))
   562  && int64(uint8(c)) == c
   563  && int64(uint8(d)) == d
   564  && (!x.Type.IsSigned() || x.Type.Size() > 1)
   565  => x
   566
   567// Fold boolean tests into blocks.
   568// Note: this must match If statement lowering.
   569(CLIJ {s390x.LessOrGreater} (LOCGR {d} (MOVDconst [0]) (MOVDconst [x]) cmp) [0] yes no)
   570  && int32(x) != 0
   571  => (BRC {d} cmp yes no)
   572
   573// Canonicalize BRC condition code mask by removing impossible conditions.
   574// Integer comparisons cannot generate the unordered condition.
   575(BRC {c} x:((CMP|CMPW|CMPU|CMPWU)    _ _) yes no) && c&s390x.Unordered != 0 => (BRC {c&^s390x.Unordered} x yes no)
   576(BRC {c} x:((CMP|CMPW|CMPU|CMPWU)const _) yes no) && c&s390x.Unordered != 0 => (BRC {c&^s390x.Unordered} x yes no)
   577
   578// Compare-and-branch.
   579// Note: bit 3 (unordered) must not be set so we mask out s390x.Unordered.
   580(BRC {c} (CMP   x y) yes no) => (CGRJ  {c&^s390x.Unordered} x y yes no)
   581(BRC {c} (CMPW  x y) yes no) => (CRJ   {c&^s390x.Unordered} x y yes no)
   582(BRC {c} (CMPU  x y) yes no) => (CLGRJ {c&^s390x.Unordered} x y yes no)
   583(BRC {c} (CMPWU x y) yes no) => (CLRJ  {c&^s390x.Unordered} x y yes no)
   584
   585// Compare-and-branch (immediate).
   586// Note: bit 3 (unordered) must not be set so we mask out s390x.Unordered.
   587(BRC {c} (CMPconst   x [y]) yes no) && y == int32( int8(y)) => (CGIJ  {c&^s390x.Unordered} x [ int8(y)] yes no)
   588(BRC {c} (CMPWconst  x [y]) yes no) && y == int32( int8(y)) => (CIJ   {c&^s390x.Unordered} x [ int8(y)] yes no)
   589(BRC {c} (CMPUconst  x [y]) yes no) && y == int32(uint8(y)) => (CLGIJ {c&^s390x.Unordered} x [uint8(y)] yes no)
   590(BRC {c} (CMPWUconst x [y]) yes no) && y == int32(uint8(y)) => (CLIJ  {c&^s390x.Unordered} x [uint8(y)] yes no)
   591
   592// Absorb immediate into compare-and-branch.
   593(C(R|GR)J  {c} x (MOVDconst [y]) yes no) && is8Bit(y)  => (C(I|GI)J  {c} x [ int8(y)] yes no)
   594(CL(R|GR)J {c} x (MOVDconst [y]) yes no) && isU8Bit(y) => (CL(I|GI)J {c} x [uint8(y)] yes no)
   595(C(R|GR)J  {c} (MOVDconst [x]) y yes no) && is8Bit(x)  => (C(I|GI)J  {c.ReverseComparison()} y [ int8(x)] yes no)
   596(CL(R|GR)J {c} (MOVDconst [x]) y yes no) && isU8Bit(x) => (CL(I|GI)J {c.ReverseComparison()} y [uint8(x)] yes no)
   597
   598// Prefer comparison with immediate to compare-and-branch.
   599(CGRJ  {c} x (MOVDconst [y]) yes no) && !is8Bit(y)  && is32Bit(y)  => (BRC {c} (CMPconst   x [int32(y)]) yes no)
   600(CRJ   {c} x (MOVDconst [y]) yes no) && !is8Bit(y)  && is32Bit(y)  => (BRC {c} (CMPWconst  x [int32(y)]) yes no)
   601(CLGRJ {c} x (MOVDconst [y]) yes no) && !isU8Bit(y) && isU32Bit(y) => (BRC {c} (CMPUconst  x [int32(y)]) yes no)
   602(CLRJ  {c} x (MOVDconst [y]) yes no) && !isU8Bit(y) && isU32Bit(y) => (BRC {c} (CMPWUconst x [int32(y)]) yes no)
   603(CGRJ  {c} (MOVDconst [x]) y yes no) && !is8Bit(x)  && is32Bit(x)  => (BRC {c.ReverseComparison()} (CMPconst   y [int32(x)]) yes no)
   604(CRJ   {c} (MOVDconst [x]) y yes no) && !is8Bit(x)  && is32Bit(x)  => (BRC {c.ReverseComparison()} (CMPWconst  y [int32(x)]) yes no)
   605(CLGRJ {c} (MOVDconst [x]) y yes no) && !isU8Bit(x) && isU32Bit(x) => (BRC {c.ReverseComparison()} (CMPUconst  y [int32(x)]) yes no)
   606(CLRJ  {c} (MOVDconst [x]) y yes no) && !isU8Bit(x) && isU32Bit(x) => (BRC {c.ReverseComparison()} (CMPWUconst y [int32(x)]) yes no)
   607
   608// Absorb sign/zero extensions into 32-bit compare-and-branch.
   609(CIJ  {c} (MOV(W|WZ)reg x) [y] yes no) => (CIJ  {c} x [y] yes no)
   610(CLIJ {c} (MOV(W|WZ)reg x) [y] yes no) => (CLIJ {c} x [y] yes no)
   611
   612// Bring out-of-range signed immediates into range by varying branch condition.
   613(BRC {s390x.Less}           (CMPconst  x [ 128]) yes no) => (CGIJ {s390x.LessOrEqual}    x [ 127] yes no)
   614(BRC {s390x.Less}           (CMPWconst x [ 128]) yes no) => (CIJ  {s390x.LessOrEqual}    x [ 127] yes no)
   615(BRC {s390x.LessOrEqual}    (CMPconst  x [-129]) yes no) => (CGIJ {s390x.Less}           x [-128] yes no)
   616(BRC {s390x.LessOrEqual}    (CMPWconst x [-129]) yes no) => (CIJ  {s390x.Less}           x [-128] yes no)
   617(BRC {s390x.Greater}        (CMPconst  x [-129]) yes no) => (CGIJ {s390x.GreaterOrEqual} x [-128] yes no)
   618(BRC {s390x.Greater}        (CMPWconst x [-129]) yes no) => (CIJ  {s390x.GreaterOrEqual} x [-128] yes no)
   619(BRC {s390x.GreaterOrEqual} (CMPconst  x [ 128]) yes no) => (CGIJ {s390x.Greater}        x [ 127] yes no)
   620(BRC {s390x.GreaterOrEqual} (CMPWconst x [ 128]) yes no) => (CIJ  {s390x.Greater}        x [ 127] yes no)
   621
   622// Bring out-of-range unsigned immediates into range by varying branch condition.
   623(BRC {s390x.Less}           (CMP(WU|U)const  x [256]) yes no) => (C(L|LG)IJ {s390x.LessOrEqual} x [255] yes no)
   624(BRC {s390x.GreaterOrEqual} (CMP(WU|U)const  x [256]) yes no) => (C(L|LG)IJ {s390x.Greater}     x [255] yes no)
   625
   626// Bring out-of-range immediates into range by switching signedness (only == and !=).
   627(BRC {c} (CMPconst   x [y]) yes no) && y == int32(uint8(y)) && (c == s390x.Equal || c == s390x.LessOrGreater) => (CLGIJ {c} x [uint8(y)] yes no)
   628(BRC {c} (CMPWconst  x [y]) yes no) && y == int32(uint8(y)) && (c == s390x.Equal || c == s390x.LessOrGreater) => (CLIJ  {c} x [uint8(y)] yes no)
   629(BRC {c} (CMPUconst  x [y]) yes no) && y == int32( int8(y)) && (c == s390x.Equal || c == s390x.LessOrGreater) => (CGIJ  {c} x [ int8(y)] yes no)
   630(BRC {c} (CMPWUconst x [y]) yes no) && y == int32( int8(y)) && (c == s390x.Equal || c == s390x.LessOrGreater) => (CIJ   {c} x [ int8(y)] yes no)
   631
   632// Fold constants into instructions.
   633(ADD x (MOVDconst <t> [c])) && is32Bit(c) && !t.IsPtr() => (ADDconst [int32(c)] x)
   634(ADDW x (MOVDconst [c])) => (ADDWconst [int32(c)] x)
   635
   636(SUB x (MOVDconst [c])) && is32Bit(c) => (SUBconst x [int32(c)])
   637(SUB (MOVDconst [c]) x) && is32Bit(c) => (NEG (SUBconst <v.Type> x [int32(c)]))
   638(SUBW x (MOVDconst [c])) => (SUBWconst x [int32(c)])
   639(SUBW (MOVDconst [c]) x) => (NEGW (SUBWconst <v.Type> x [int32(c)]))
   640
   641(MULLD x (MOVDconst [c])) && is32Bit(c) => (MULLDconst [int32(c)] x)
   642(MULLW x (MOVDconst [c])) => (MULLWconst [int32(c)] x)
   643
   644// NILF instructions leave the high 32 bits unchanged which is
   645// equivalent to the leftmost 32 bits being set.
   646// TODO(mundaym): modify the assembler to accept 64-bit values
   647// and use isU32Bit(^c).
   648(AND x (MOVDconst [c]))
   649  && s390x.NewRotateParams(0, 63, 0).OutMerge(uint64(c)) != nil
   650  => (RISBGZ x {*s390x.NewRotateParams(0, 63, 0).OutMerge(uint64(c))})
   651(AND x (MOVDconst [c]))
   652  && is32Bit(c)
   653  && c < 0
   654  => (ANDconst [c] x)
   655(AND x (MOVDconst [c]))
   656  && is32Bit(c)
   657  && c >= 0
   658  => (MOVWZreg (ANDWconst <typ.UInt32> [int32(c)] x))
   659
   660(ANDW x (MOVDconst [c])) => (ANDWconst [int32(c)] x)
   661
   662((AND|ANDW)const [c] ((AND|ANDW)const [d] x)) => ((AND|ANDW)const [c&d] x)
   663
   664((OR|XOR) x (MOVDconst [c])) && isU32Bit(c) => ((OR|XOR)const [c] x)
   665((OR|XOR)W x (MOVDconst [c])) => ((OR|XOR)Wconst [int32(c)] x)
   666
   667// Constant shifts.
   668(S(LD|RD|RAD) x (MOVDconst [c])) => (S(LD|RD|RAD)const x [uint8(c&63)])
   669(S(LW|RW|RAW) x (MOVDconst [c])) && c&32 == 0 => (S(LW|RW|RAW)const x [uint8(c&31)])
   670(S(LW|RW)     _ (MOVDconst [c])) && c&32 != 0 => (MOVDconst [0])
   671(SRAW         x (MOVDconst [c])) && c&32 != 0 => (SRAWconst x [31])
   672
   673// Shifts only use the rightmost 6 bits of the shift value.
   674(S(LD|RD|RAD|LW|RW|RAW) x (RISBGZ y {r}))
   675  && r.Amount == 0
   676  && r.OutMask()&63 == 63
   677  => (S(LD|RD|RAD|LW|RW|RAW) x y)
   678(S(LD|RD|RAD|LW|RW|RAW) x (AND (MOVDconst [c]) y))
   679  => (S(LD|RD|RAD|LW|RW|RAW) x (ANDWconst <typ.UInt32> [int32(c&63)] y))
   680(S(LD|RD|RAD|LW|RW|RAW) x (ANDWconst [c] y)) && c&63 == 63
   681  => (S(LD|RD|RAD|LW|RW|RAW) x y)
   682(SLD  x (MOV(W|H|B|WZ|HZ|BZ)reg y)) => (SLD  x y)
   683(SRD  x (MOV(W|H|B|WZ|HZ|BZ)reg y)) => (SRD  x y)
   684(SRAD x (MOV(W|H|B|WZ|HZ|BZ)reg y)) => (SRAD x y)
   685(SLW  x (MOV(W|H|B|WZ|HZ|BZ)reg y)) => (SLW  x y)
   686(SRW  x (MOV(W|H|B|WZ|HZ|BZ)reg y)) => (SRW  x y)
   687(SRAW x (MOV(W|H|B|WZ|HZ|BZ)reg y)) => (SRAW x y)
   688
   689// Match rotate by constant.
   690(RLLG x (MOVDconst [c])) => (RISBGZ x {s390x.NewRotateParams(0, 63, uint8(c&63))})
   691(RLL  x (MOVDconst [c])) => (RLLconst x [uint8(c&31)])
   692
   693// Signed 64-bit comparison with immediate.
   694(CMP x (MOVDconst [c])) && is32Bit(c) => (CMPconst x [int32(c)])
   695(CMP (MOVDconst [c]) x) && is32Bit(c) => (InvertFlags (CMPconst x [int32(c)]))
   696
   697// Unsigned 64-bit comparison with immediate.
   698(CMPU x (MOVDconst [c])) && isU32Bit(c) => (CMPUconst x [int32(c)])
   699(CMPU (MOVDconst [c]) x) && isU32Bit(c) => (InvertFlags (CMPUconst x [int32(c)]))
   700
   701// Signed and unsigned 32-bit comparison with immediate.
   702(CMP(W|WU) x (MOVDconst [c])) => (CMP(W|WU)const x [int32(c)])
   703(CMP(W|WU) (MOVDconst [c]) x) => (InvertFlags (CMP(W|WU)const x [int32(c)]))
   704
   705// Match (x >> c) << d to 'rotate then insert selected bits [into zero]'.
   706(SLDconst (SRDconst x [c]) [d]) => (RISBGZ x {s390x.NewRotateParams(uint8(max8(0, int8(c-d))), 63-d, uint8(int8(d-c)&63))})
   707
   708// Match (x << c) >> d to 'rotate then insert selected bits [into zero]'.
   709(SRDconst (SLDconst x [c]) [d]) => (RISBGZ x {s390x.NewRotateParams(d, uint8(min8(63, int8(63-c+d))), uint8(int8(c-d)&63))})
   710
   711// Absorb input zero extension into 'rotate then insert selected bits [into zero]'.
   712(RISBGZ (MOVWZreg x) {r}) && r.InMerge(0xffffffff) != nil => (RISBGZ x {*r.InMerge(0xffffffff)})
   713(RISBGZ (MOVHZreg x) {r}) && r.InMerge(0x0000ffff) != nil => (RISBGZ x {*r.InMerge(0x0000ffff)})
   714(RISBGZ (MOVBZreg x) {r}) && r.InMerge(0x000000ff) != nil => (RISBGZ x {*r.InMerge(0x000000ff)})
   715
   716// Absorb 'rotate then insert selected bits [into zero]' into zero extension.
   717(MOVWZreg (RISBGZ x {r})) && r.OutMerge(0xffffffff) != nil => (RISBGZ x {*r.OutMerge(0xffffffff)})
   718(MOVHZreg (RISBGZ x {r})) && r.OutMerge(0x0000ffff) != nil => (RISBGZ x {*r.OutMerge(0x0000ffff)})
   719(MOVBZreg (RISBGZ x {r})) && r.OutMerge(0x000000ff) != nil => (RISBGZ x {*r.OutMerge(0x000000ff)})
   720
   721// Absorb shift into 'rotate then insert selected bits [into zero]'.
   722//
   723// Any unsigned shift can be represented as a rotate and mask operation:
   724//
   725//   x << c => RotateLeft64(x, c) & (^uint64(0) << c)
   726//   x >> c => RotateLeft64(x, -c) & (^uint64(0) >> c)
   727//
   728// Therefore when a shift is used as the input to a rotate then insert
   729// selected bits instruction we can merge the two together. We just have
   730// to be careful that the resultant mask is representable (non-zero and
   731// contiguous). For example, assuming that x is variable and c, y and m
   732// are constants, a shift followed by a rotate then insert selected bits
   733// could be represented as:
   734//
   735//   RotateLeft64(RotateLeft64(x, c) & (^uint64(0) << c), y) & m
   736//
   737// We can split the rotation by y into two, one rotate for x and one for
   738// the mask:
   739//
   740//   RotateLeft64(RotateLeft64(x, c), y) & (RotateLeft64(^uint64(0) << c, y)) & m
   741//
   742// The rotations of x by c followed by y can then be combined:
   743//
   744//   RotateLeft64(x, c+y) & (RotateLeft64(^uint64(0) << c, y)) & m
   745//   ^^^^^^^^^^^^^^^^^^^^   ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
   746//          rotate                          mask
   747//
   748// To perform this optimization we therefore just need to check that it
   749// is valid to merge the shift mask (^(uint64(0)<<c)) into the selected
   750// bits mask (i.e. that the resultant mask is non-zero and contiguous).
   751//
   752(RISBGZ (SLDconst x [c]) {r}) && r.InMerge(^uint64(0)<<c) != nil => (RISBGZ x {(*r.InMerge(^uint64(0)<<c)).RotateLeft(c)})
   753(RISBGZ (SRDconst x [c]) {r}) && r.InMerge(^uint64(0)>>c) != nil => (RISBGZ x {(*r.InMerge(^uint64(0)>>c)).RotateLeft(-c)})
   754
   755// Absorb 'rotate then insert selected bits [into zero]' into left shift.
   756(SLDconst (RISBGZ x {r}) [c])
   757  && s390x.NewRotateParams(0, 63-c, c).InMerge(r.OutMask()) != nil
   758  => (RISBGZ x {(*s390x.NewRotateParams(0, 63-c, c).InMerge(r.OutMask())).RotateLeft(r.Amount)})
   759
   760// Absorb 'rotate then insert selected bits [into zero]' into right shift.
   761(SRDconst (RISBGZ x {r}) [c])
   762  && s390x.NewRotateParams(c, 63, -c&63).InMerge(r.OutMask()) != nil
   763  => (RISBGZ x {(*s390x.NewRotateParams(c, 63, -c&63).InMerge(r.OutMask())).RotateLeft(r.Amount)})
   764
   765// Merge 'rotate then insert selected bits [into zero]' instructions together.
   766(RISBGZ (RISBGZ x {y}) {z})
   767  && z.InMerge(y.OutMask()) != nil
   768  => (RISBGZ x {(*z.InMerge(y.OutMask())).RotateLeft(y.Amount)})
   769
   770// Convert RISBGZ into 64-bit shift (helps CSE).
   771(RISBGZ x {r}) && r.End == 63 && r.Start == -r.Amount&63 => (SRDconst x [-r.Amount&63])
   772(RISBGZ x {r}) && r.Start == 0 && r.End == 63-r.Amount => (SLDconst x [r.Amount])
   773
   774// Optimize single bit isolation when it is known to be equivalent to
   775// the most significant bit due to mask produced by arithmetic shift.
   776// Simply isolate the most significant bit itself and place it in the
   777// correct position.
   778//
   779// Example: (int64(x) >> 63) & 0x8 -> RISBGZ $60, $60, $4, Rsrc, Rdst
   780(RISBGZ (SRADconst x [c]) {r})
   781  && r.Start == r.End           // single bit selected
   782  && (r.Start+r.Amount)&63 <= c // equivalent to most significant bit of x
   783  => (RISBGZ x {s390x.NewRotateParams(r.Start, r.Start, -r.Start&63)})
   784
   785// Canonicalize the order of arguments to comparisons - helps with CSE.
   786((CMP|CMPW|CMPU|CMPWU) x y) && canonLessThan(x,y) => (InvertFlags ((CMP|CMPW|CMPU|CMPWU) y x))
   787
   788// Use sign/zero extend instead of RISBGZ.
   789(RISBGZ x {r}) && r == s390x.NewRotateParams(56, 63, 0) => (MOVBZreg x)
   790(RISBGZ x {r}) && r == s390x.NewRotateParams(48, 63, 0) => (MOVHZreg x)
   791(RISBGZ x {r}) && r == s390x.NewRotateParams(32, 63, 0) => (MOVWZreg x)
   792
   793// Use sign/zero extend instead of ANDW.
   794(ANDWconst [0x00ff] x) => (MOVBZreg x)
   795(ANDWconst [0xffff] x) => (MOVHZreg x)
   796
   797// Strength reduce multiplication to the sum (or difference) of two powers of two.
   798//
   799// Examples:
   800//     5x -> 4x + 1x
   801//    10x -> 8x + 2x
   802//   120x -> 128x - 8x
   803//  -120x -> 8x - 128x
   804//
   805// We know that the rightmost bit of any positive value, once isolated, must either
   806// be a power of 2 (because it is a single bit) or 0 (if the original value is 0).
   807// In all of these rules we use a rightmost bit calculation to determine one operand
   808// for the addition or subtraction. We then just need to calculate if the other
   809// operand is a valid power of 2 before we can match the rule.
   810//
   811// Notes:
   812//   - the generic rules have already matched single powers of two so we ignore them here
   813//   - isPowerOfTwo32 asserts that its argument is greater than 0
   814//   - c&(c-1) = clear rightmost bit
   815//   - c&^(c-1) = isolate rightmost bit
   816
   817// c = 2ˣ + 2ʸ => c - 2ˣ = 2ʸ
   818(MULL(D|W)const <t> x [c]) && isPowerOfTwo32(c&(c-1))
   819  => ((ADD|ADDW) (SL(D|W)const <t> x [uint8(log32(c&(c-1)))])
   820                 (SL(D|W)const <t> x [uint8(log32(c&^(c-1)))]))
   821
   822// c = 2ʸ - 2ˣ => c + 2ˣ = 2ʸ
   823(MULL(D|W)const <t> x [c]) && isPowerOfTwo32(c+(c&^(c-1)))
   824  => ((SUB|SUBW) (SL(D|W)const <t> x [uint8(log32(c+(c&^(c-1))))])
   825                 (SL(D|W)const <t> x [uint8(log32(c&^(c-1)))]))
   826
   827// c = 2ˣ - 2ʸ => -c + 2ˣ = 2ʸ
   828(MULL(D|W)const <t> x [c]) && isPowerOfTwo32(-c+(-c&^(-c-1)))
   829  => ((SUB|SUBW) (SL(D|W)const <t> x [uint8(log32(-c&^(-c-1)))])
   830                 (SL(D|W)const <t> x [uint8(log32(-c+(-c&^(-c-1))))]))
   831
   832// Fold ADD into MOVDaddr. Odd offsets from SB shouldn't be folded (LARL can't handle them).
   833(ADDconst [c] (MOVDaddr [d] {s} x:(SB))) && ((c+d)&1 == 0) && is32Bit(int64(c)+int64(d)) => (MOVDaddr [c+d] {s} x)
   834(ADDconst [c] (MOVDaddr [d] {s} x)) && x.Op != OpSB && is20Bit(int64(c)+int64(d)) => (MOVDaddr [c+d] {s} x)
   835(ADD idx (MOVDaddr [c] {s} ptr)) && ptr.Op != OpSB => (MOVDaddridx [c] {s} ptr idx)
   836
   837// fold ADDconst into MOVDaddrx
   838(ADDconst [c] (MOVDaddridx [d] {s} x y)) && is20Bit(int64(c)+int64(d)) => (MOVDaddridx [c+d] {s} x y)
   839(MOVDaddridx [c] {s} (ADDconst [d] x) y) && is20Bit(int64(c)+int64(d)) => (MOVDaddridx [c+d] {s} x y)
   840(MOVDaddridx [c] {s} x (ADDconst [d] y)) && is20Bit(int64(c)+int64(d)) => (MOVDaddridx [c+d] {s} x y)
   841
   842// reverse ordering of compare instruction
   843(LOCGR {c} x y (InvertFlags cmp)) => (LOCGR {c.ReverseComparison()} x y cmp)
   844
   845// replace load from same location as preceding store with copy
   846(MOVDload  [off] {sym} ptr1 (MOVDstore  [off] {sym} ptr2 x _)) && isSamePtr(ptr1, ptr2) => x
   847(MOVWload  [off] {sym} ptr1 (MOVWstore  [off] {sym} ptr2 x _)) && isSamePtr(ptr1, ptr2) => (MOVWreg x)
   848(MOVHload  [off] {sym} ptr1 (MOVHstore  [off] {sym} ptr2 x _)) && isSamePtr(ptr1, ptr2) => (MOVHreg x)
   849(MOVBload  [off] {sym} ptr1 (MOVBstore  [off] {sym} ptr2 x _)) && isSamePtr(ptr1, ptr2) => (MOVBreg x)
   850(MOVWZload [off] {sym} ptr1 (MOVWstore  [off] {sym} ptr2 x _)) && isSamePtr(ptr1, ptr2) => (MOVWZreg x)
   851(MOVHZload [off] {sym} ptr1 (MOVHstore  [off] {sym} ptr2 x _)) && isSamePtr(ptr1, ptr2) => (MOVHZreg x)
   852(MOVBZload [off] {sym} ptr1 (MOVBstore  [off] {sym} ptr2 x _)) && isSamePtr(ptr1, ptr2) => (MOVBZreg x)
   853(MOVDload  [off] {sym} ptr1 (FMOVDstore [off] {sym} ptr2 x _)) && isSamePtr(ptr1, ptr2) => (LGDR x)
   854(FMOVDload [off] {sym} ptr1 (MOVDstore  [off] {sym} ptr2 x _)) && isSamePtr(ptr1, ptr2) => (LDGR x)
   855(FMOVDload [off] {sym} ptr1 (FMOVDstore [off] {sym} ptr2 x _)) && isSamePtr(ptr1, ptr2) => x
   856(FMOVSload [off] {sym} ptr1 (FMOVSstore [off] {sym} ptr2 x _)) && isSamePtr(ptr1, ptr2) => x
   857
   858// prefer FPR <-> GPR moves over combined load ops
   859(MULLDload <t> [off] {sym} x ptr1 (FMOVDstore [off] {sym} ptr2 y _)) && isSamePtr(ptr1, ptr2) => (MULLD x (LGDR <t> y))
   860(ADDload   <t> [off] {sym} x ptr1 (FMOVDstore [off] {sym} ptr2 y _)) && isSamePtr(ptr1, ptr2) => (ADD   x (LGDR <t> y))
   861(SUBload   <t> [off] {sym} x ptr1 (FMOVDstore [off] {sym} ptr2 y _)) && isSamePtr(ptr1, ptr2) => (SUB   x (LGDR <t> y))
   862(ORload    <t> [off] {sym} x ptr1 (FMOVDstore [off] {sym} ptr2 y _)) && isSamePtr(ptr1, ptr2) => (OR    x (LGDR <t> y))
   863(ANDload   <t> [off] {sym} x ptr1 (FMOVDstore [off] {sym} ptr2 y _)) && isSamePtr(ptr1, ptr2) => (AND   x (LGDR <t> y))
   864(XORload   <t> [off] {sym} x ptr1 (FMOVDstore [off] {sym} ptr2 y _)) && isSamePtr(ptr1, ptr2) => (XOR   x (LGDR <t> y))
   865
   866// detect attempts to set/clear the sign bit
   867// may need to be reworked when NIHH/OIHH are added
   868(RISBGZ (LGDR <t> x) {r}) && r == s390x.NewRotateParams(1, 63, 0) => (LGDR <t> (LPDFR <x.Type> x))
   869(LDGR <t> (RISBGZ x {r})) && r == s390x.NewRotateParams(1, 63, 0) => (LPDFR (LDGR <t> x))
   870(OR (MOVDconst [-1<<63]) (LGDR <t> x)) => (LGDR <t> (LNDFR <x.Type> x))
   871(LDGR <t> (OR (MOVDconst [-1<<63]) x)) => (LNDFR (LDGR <t> x))
   872
   873// detect attempts to set the sign bit with load
   874(LDGR <t> x:(ORload <t1> [off] {sym} (MOVDconst [-1<<63]) ptr mem)) && x.Uses == 1 && clobber(x) => @x.Block (LNDFR <t> (LDGR <t> (MOVDload <t1> [off] {sym} ptr mem)))
   875
   876// detect copysign
   877(OR (RISBGZ (LGDR x) {r}) (LGDR (LPDFR <t> y)))
   878  && r == s390x.NewRotateParams(0, 0, 0)
   879  => (LGDR (CPSDR <t> y x))
   880(OR (RISBGZ (LGDR x) {r}) (MOVDconst [c]))
   881  && c >= 0
   882  && r == s390x.NewRotateParams(0, 0, 0)
   883  => (LGDR (CPSDR <x.Type> (FMOVDconst <x.Type> [math.Float64frombits(uint64(c))]) x))
   884(CPSDR y (FMOVDconst [c])) && !math.Signbit(c) => (LPDFR y)
   885(CPSDR y (FMOVDconst [c])) && math.Signbit(c)  => (LNDFR y)
   886
   887// absorb negations into set/clear sign bit
   888(FNEG  (LPDFR x)) => (LNDFR x)
   889(FNEG  (LNDFR x)) => (LPDFR x)
   890(FNEGS (LPDFR x)) => (LNDFR x)
   891(FNEGS (LNDFR x)) => (LPDFR x)
   892
   893// no need to convert float32 to float64 to set/clear sign bit
   894(LEDBR (LPDFR (LDEBR x))) => (LPDFR x)
   895(LEDBR (LNDFR (LDEBR x))) => (LNDFR x)
   896
   897// remove unnecessary FPR <-> GPR moves
   898(LDGR (LGDR x)) => x
   899(LGDR (LDGR x)) => x
   900
   901// Don't extend before storing
   902(MOVWstore [off] {sym} ptr (MOVWreg x) mem) => (MOVWstore [off] {sym} ptr x mem)
   903(MOVHstore [off] {sym} ptr (MOVHreg x) mem) => (MOVHstore [off] {sym} ptr x mem)
   904(MOVBstore [off] {sym} ptr (MOVBreg x) mem) => (MOVBstore [off] {sym} ptr x mem)
   905(MOVWstore [off] {sym} ptr (MOVWZreg x) mem) => (MOVWstore [off] {sym} ptr x mem)
   906(MOVHstore [off] {sym} ptr (MOVHZreg x) mem) => (MOVHstore [off] {sym} ptr x mem)
   907(MOVBstore [off] {sym} ptr (MOVBZreg x) mem) => (MOVBstore [off] {sym} ptr x mem)
   908
   909// Fold constants into memory operations.
   910// Note that this is not always a good idea because if not all the uses of
   911// the ADDconst get eliminated, we still have to compute the ADDconst and we now
   912// have potentially two live values (ptr and (ADDconst [off] ptr)) instead of one.
   913// Nevertheless, let's do it!
   914(MOVDload   [off1] {sym} (ADDconst [off2] ptr) mem) && is20Bit(int64(off1)+int64(off2)) => (MOVDload  [off1+off2] {sym} ptr mem)
   915(MOVWload   [off1] {sym} (ADDconst [off2] ptr) mem) && is20Bit(int64(off1)+int64(off2)) => (MOVWload  [off1+off2] {sym} ptr mem)
   916(MOVHload   [off1] {sym} (ADDconst [off2] ptr) mem) && is20Bit(int64(off1)+int64(off2)) => (MOVHload  [off1+off2] {sym} ptr mem)
   917(MOVBload   [off1] {sym} (ADDconst [off2] ptr) mem) && is20Bit(int64(off1)+int64(off2)) => (MOVBload  [off1+off2] {sym} ptr mem)
   918(MOVWZload  [off1] {sym} (ADDconst [off2] ptr) mem) && is20Bit(int64(off1)+int64(off2)) => (MOVWZload [off1+off2] {sym} ptr mem)
   919(MOVHZload  [off1] {sym} (ADDconst [off2] ptr) mem) && is20Bit(int64(off1)+int64(off2)) => (MOVHZload [off1+off2] {sym} ptr mem)
   920(MOVBZload  [off1] {sym} (ADDconst [off2] ptr) mem) && is20Bit(int64(off1)+int64(off2)) => (MOVBZload [off1+off2] {sym} ptr mem)
   921(FMOVSload  [off1] {sym} (ADDconst [off2] ptr) mem) && is20Bit(int64(off1)+int64(off2)) => (FMOVSload [off1+off2] {sym} ptr mem)
   922(FMOVDload  [off1] {sym} (ADDconst [off2] ptr) mem) && is20Bit(int64(off1)+int64(off2)) => (FMOVDload [off1+off2] {sym} ptr mem)
   923
   924(MOVDstore  [off1] {sym} (ADDconst [off2] ptr) val mem) && is20Bit(int64(off1)+int64(off2)) => (MOVDstore  [off1+off2] {sym} ptr val mem)
   925(MOVWstore  [off1] {sym} (ADDconst [off2] ptr) val mem) && is20Bit(int64(off1)+int64(off2)) => (MOVWstore  [off1+off2] {sym} ptr val mem)
   926(MOVHstore  [off1] {sym} (ADDconst [off2] ptr) val mem) && is20Bit(int64(off1)+int64(off2)) => (MOVHstore  [off1+off2] {sym} ptr val mem)
   927(MOVBstore  [off1] {sym} (ADDconst [off2] ptr) val mem) && is20Bit(int64(off1)+int64(off2)) => (MOVBstore  [off1+off2] {sym} ptr val mem)
   928(FMOVSstore [off1] {sym} (ADDconst [off2] ptr) val mem) && is20Bit(int64(off1)+int64(off2)) => (FMOVSstore [off1+off2] {sym} ptr val mem)
   929(FMOVDstore [off1] {sym} (ADDconst [off2] ptr) val mem) && is20Bit(int64(off1)+int64(off2)) => (FMOVDstore [off1+off2] {sym} ptr val mem)
   930
   931(ADDload   [off1] {sym} x (ADDconst [off2] ptr) mem) && ptr.Op != OpSB && is20Bit(int64(off1)+int64(off2)) => (ADDload   [off1+off2] {sym} x ptr mem)
   932(ADDWload  [off1] {sym} x (ADDconst [off2] ptr) mem) && ptr.Op != OpSB && is20Bit(int64(off1)+int64(off2)) => (ADDWload  [off1+off2] {sym} x ptr mem)
   933(MULLDload [off1] {sym} x (ADDconst [off2] ptr) mem) && ptr.Op != OpSB && is20Bit(int64(off1)+int64(off2)) => (MULLDload [off1+off2] {sym} x ptr mem)
   934(MULLWload [off1] {sym} x (ADDconst [off2] ptr) mem) && ptr.Op != OpSB && is20Bit(int64(off1)+int64(off2)) => (MULLWload [off1+off2] {sym} x ptr mem)
   935(SUBload   [off1] {sym} x (ADDconst [off2] ptr) mem) && ptr.Op != OpSB && is20Bit(int64(off1)+int64(off2)) => (SUBload   [off1+off2] {sym} x ptr mem)
   936(SUBWload  [off1] {sym} x (ADDconst [off2] ptr) mem) && ptr.Op != OpSB && is20Bit(int64(off1)+int64(off2)) => (SUBWload  [off1+off2] {sym} x ptr mem)
   937
   938(ANDload   [off1] {sym} x (ADDconst [off2] ptr) mem) && ptr.Op != OpSB && is20Bit(int64(off1)+int64(off2)) => (ANDload   [off1+off2] {sym} x ptr mem)
   939(ANDWload  [off1] {sym} x (ADDconst [off2] ptr) mem) && ptr.Op != OpSB && is20Bit(int64(off1)+int64(off2)) => (ANDWload  [off1+off2] {sym} x ptr mem)
   940(ORload    [off1] {sym} x (ADDconst [off2] ptr) mem) && ptr.Op != OpSB && is20Bit(int64(off1)+int64(off2)) => (ORload    [off1+off2] {sym} x ptr mem)
   941(ORWload   [off1] {sym} x (ADDconst [off2] ptr) mem) && ptr.Op != OpSB && is20Bit(int64(off1)+int64(off2)) => (ORWload   [off1+off2] {sym} x ptr mem)
   942(XORload   [off1] {sym} x (ADDconst [off2] ptr) mem) && ptr.Op != OpSB && is20Bit(int64(off1)+int64(off2)) => (XORload   [off1+off2] {sym} x ptr mem)
   943(XORWload  [off1] {sym} x (ADDconst [off2] ptr) mem) && ptr.Op != OpSB && is20Bit(int64(off1)+int64(off2)) => (XORWload  [off1+off2] {sym} x ptr mem)
   944
   945// Fold constants into stores.
   946(MOVDstore [off] {sym} ptr (MOVDconst [c]) mem) && is16Bit(c) && isU12Bit(int64(off)) && ptr.Op != OpSB =>
   947	(MOVDstoreconst [makeValAndOff(int32(c),off)] {sym} ptr mem)
   948(MOVWstore [off] {sym} ptr (MOVDconst [c]) mem) && is16Bit(c) && isU12Bit(int64(off)) && ptr.Op != OpSB =>
   949	(MOVWstoreconst [makeValAndOff(int32(c),off)] {sym} ptr mem)
   950(MOVHstore [off] {sym} ptr (MOVDconst [c]) mem) && isU12Bit(int64(off)) && ptr.Op != OpSB =>
   951	(MOVHstoreconst [makeValAndOff(int32(int16(c)),off)] {sym} ptr mem)
   952(MOVBstore [off] {sym} ptr (MOVDconst [c]) mem) && is20Bit(int64(off)) && ptr.Op != OpSB =>
   953	(MOVBstoreconst [makeValAndOff(int32(int8(c)),off)] {sym} ptr mem)
   954
   955// Fold address offsets into constant stores.
   956(MOVDstoreconst [sc] {s} (ADDconst [off] ptr) mem) && isU12Bit(sc.Off64()+int64(off)) =>
   957	(MOVDstoreconst [sc.addOffset32(off)] {s} ptr mem)
   958(MOVWstoreconst [sc] {s} (ADDconst [off] ptr) mem) && isU12Bit(sc.Off64()+int64(off)) =>
   959	(MOVWstoreconst [sc.addOffset32(off)] {s} ptr mem)
   960(MOVHstoreconst [sc] {s} (ADDconst [off] ptr) mem) && isU12Bit(sc.Off64()+int64(off)) =>
   961	(MOVHstoreconst [sc.addOffset32(off)] {s} ptr mem)
   962(MOVBstoreconst [sc] {s} (ADDconst [off] ptr) mem) && is20Bit(sc.Off64()+int64(off)) =>
   963	(MOVBstoreconst [sc.addOffset32(off)] {s} ptr mem)
   964
   965// Merge address calculations into loads and stores.
   966// Offsets from SB must not be merged into unaligned memory accesses because
   967// loads/stores using PC-relative addressing directly must be aligned to the
   968// size of the target.
   969(MOVDload   [off1] {sym1} (MOVDaddr <t> [off2] {sym2} base) mem) && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) && (base.Op != OpSB || (t.IsPtr() && t.Elem().Alignment()%8 == 0 && (off1+off2)%8 == 0)) =>
   970	(MOVDload  [off1+off2] {mergeSym(sym1,sym2)} base mem)
   971(MOVWZload  [off1] {sym1} (MOVDaddr <t> [off2] {sym2} base) mem) && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) && (base.Op != OpSB || (t.IsPtr() && t.Elem().Alignment()%4 == 0 && (off1+off2)%4 == 0)) =>
   972	(MOVWZload  [off1+off2] {mergeSym(sym1,sym2)} base mem)
   973(MOVHZload  [off1] {sym1} (MOVDaddr <t> [off2] {sym2} base) mem) && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) && (base.Op != OpSB || (t.IsPtr() && t.Elem().Alignment()%2 == 0 && (off1+off2)%2 == 0)) =>
   974	(MOVHZload  [off1+off2] {mergeSym(sym1,sym2)} base mem)
   975(MOVBZload  [off1] {sym1} (MOVDaddr [off2] {sym2} base) mem) && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) =>
   976	(MOVBZload  [off1+off2] {mergeSym(sym1,sym2)} base mem)
   977(FMOVSload [off1] {sym1} (MOVDaddr [off2] {sym2} base) mem) && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) =>
   978	(FMOVSload [off1+off2] {mergeSym(sym1,sym2)} base mem)
   979(FMOVDload [off1] {sym1} (MOVDaddr [off2] {sym2} base) mem) && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) =>
   980	(FMOVDload [off1+off2] {mergeSym(sym1,sym2)} base mem)
   981
   982(MOVWload [off1] {sym1} (MOVDaddr <t> [off2] {sym2} base) mem) && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) && (base.Op != OpSB || (t.IsPtr() && t.Elem().Alignment()%4 == 0 && (off1+off2)%4 == 0)) =>
   983	(MOVWload [off1+off2] {mergeSym(sym1,sym2)} base mem)
   984(MOVHload [off1] {sym1} (MOVDaddr <t> [off2] {sym2} base) mem) && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) && (base.Op != OpSB || (t.IsPtr() && t.Elem().Alignment()%2 == 0 && (off1+off2)%2 == 0)) =>
   985	(MOVHload [off1+off2] {mergeSym(sym1,sym2)} base mem)
   986(MOVBload [off1] {sym1} (MOVDaddr [off2] {sym2} base) mem) && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) =>
   987	(MOVBload [off1+off2] {mergeSym(sym1,sym2)} base mem)
   988
   989(MOVDstore  [off1] {sym1} (MOVDaddr <t> [off2] {sym2} base) val mem) && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) && (base.Op != OpSB || (t.IsPtr() && t.Elem().Alignment()%8 == 0 && (off1+off2)%8 == 0)) =>
   990	(MOVDstore  [off1+off2] {mergeSym(sym1,sym2)} base val mem)
   991(MOVWstore  [off1] {sym1} (MOVDaddr <t> [off2] {sym2} base) val mem) && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) && (base.Op != OpSB || (t.IsPtr() && t.Elem().Alignment()%4 == 0 && (off1+off2)%4 == 0)) =>
   992	(MOVWstore  [off1+off2] {mergeSym(sym1,sym2)} base val mem)
   993(MOVHstore  [off1] {sym1} (MOVDaddr <t> [off2] {sym2} base) val mem) && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) && (base.Op != OpSB || (t.IsPtr() && t.Elem().Alignment()%2 == 0 && (off1+off2)%2 == 0)) =>
   994	(MOVHstore  [off1+off2] {mergeSym(sym1,sym2)} base val mem)
   995(MOVBstore  [off1] {sym1} (MOVDaddr [off2] {sym2} base) val mem) && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) =>
   996	(MOVBstore  [off1+off2] {mergeSym(sym1,sym2)} base val mem)
   997(FMOVSstore [off1] {sym1} (MOVDaddr [off2] {sym2} base) val mem) && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) =>
   998	(FMOVSstore [off1+off2] {mergeSym(sym1,sym2)} base val mem)
   999(FMOVDstore [off1] {sym1} (MOVDaddr [off2] {sym2} base) val mem) && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) =>
  1000	(FMOVDstore [off1+off2] {mergeSym(sym1,sym2)} base val mem)
  1001
  1002(ADDload   [o1] {s1} x (MOVDaddr [o2] {s2} ptr) mem) && ptr.Op != OpSB && is20Bit(int64(o1)+int64(o2)) && canMergeSym(s1, s2) => (ADDload   [o1+o2] {mergeSym(s1, s2)} x ptr mem)
  1003(ADDWload  [o1] {s1} x (MOVDaddr [o2] {s2} ptr) mem) && ptr.Op != OpSB && is20Bit(int64(o1)+int64(o2)) && canMergeSym(s1, s2) => (ADDWload  [o1+o2] {mergeSym(s1, s2)} x ptr mem)
  1004(MULLDload [o1] {s1} x (MOVDaddr [o2] {s2} ptr) mem) && ptr.Op != OpSB && is20Bit(int64(o1)+int64(o2)) && canMergeSym(s1, s2) => (MULLDload [o1+o2] {mergeSym(s1, s2)} x ptr mem)
  1005(MULLWload [o1] {s1} x (MOVDaddr [o2] {s2} ptr) mem) && ptr.Op != OpSB && is20Bit(int64(o1)+int64(o2)) && canMergeSym(s1, s2) => (MULLWload [o1+o2] {mergeSym(s1, s2)} x ptr mem)
  1006(SUBload   [o1] {s1} x (MOVDaddr [o2] {s2} ptr) mem) && ptr.Op != OpSB && is20Bit(int64(o1)+int64(o2)) && canMergeSym(s1, s2) => (SUBload   [o1+o2] {mergeSym(s1, s2)} x ptr mem)
  1007(SUBWload  [o1] {s1} x (MOVDaddr [o2] {s2} ptr) mem) && ptr.Op != OpSB && is20Bit(int64(o1)+int64(o2)) && canMergeSym(s1, s2) => (SUBWload  [o1+o2] {mergeSym(s1, s2)} x ptr mem)
  1008
  1009(ANDload   [o1] {s1} x (MOVDaddr [o2] {s2} ptr) mem) && ptr.Op != OpSB && is20Bit(int64(o1)+int64(o2)) && canMergeSym(s1, s2) => (ANDload   [o1+o2] {mergeSym(s1, s2)} x ptr mem)
  1010(ANDWload  [o1] {s1} x (MOVDaddr [o2] {s2} ptr) mem) && ptr.Op != OpSB && is20Bit(int64(o1)+int64(o2)) && canMergeSym(s1, s2) => (ANDWload  [o1+o2] {mergeSym(s1, s2)} x ptr mem)
  1011(ORload    [o1] {s1} x (MOVDaddr [o2] {s2} ptr) mem) && ptr.Op != OpSB && is20Bit(int64(o1)+int64(o2)) && canMergeSym(s1, s2) => (ORload    [o1+o2] {mergeSym(s1, s2)} x ptr mem)
  1012(ORWload   [o1] {s1} x (MOVDaddr [o2] {s2} ptr) mem) && ptr.Op != OpSB && is20Bit(int64(o1)+int64(o2)) && canMergeSym(s1, s2) => (ORWload   [o1+o2] {mergeSym(s1, s2)} x ptr mem)
  1013(XORload   [o1] {s1} x (MOVDaddr [o2] {s2} ptr) mem) && ptr.Op != OpSB && is20Bit(int64(o1)+int64(o2)) && canMergeSym(s1, s2) => (XORload   [o1+o2] {mergeSym(s1, s2)} x ptr mem)
  1014(XORWload  [o1] {s1} x (MOVDaddr [o2] {s2} ptr) mem) && ptr.Op != OpSB && is20Bit(int64(o1)+int64(o2)) && canMergeSym(s1, s2) => (XORWload  [o1+o2] {mergeSym(s1, s2)} x ptr mem)
  1015
  1016// Cannot store constant to SB directly (no 'move relative long immediate' instructions).
  1017(MOVDstoreconst [sc] {sym1} (MOVDaddr [off] {sym2} ptr) mem) && ptr.Op != OpSB && canMergeSym(sym1, sym2) && sc.canAdd32(off) =>
  1018	(MOVDstoreconst [sc.addOffset32(off)] {mergeSym(sym1, sym2)} ptr mem)
  1019(MOVWstoreconst [sc] {sym1} (MOVDaddr [off] {sym2} ptr) mem) && ptr.Op != OpSB && canMergeSym(sym1, sym2) && sc.canAdd32(off) =>
  1020	(MOVWstoreconst [sc.addOffset32(off)] {mergeSym(sym1, sym2)} ptr mem)
  1021(MOVHstoreconst [sc] {sym1} (MOVDaddr [off] {sym2} ptr) mem) && ptr.Op != OpSB && canMergeSym(sym1, sym2) && sc.canAdd32(off) =>
  1022	(MOVHstoreconst [sc.addOffset32(off)] {mergeSym(sym1, sym2)} ptr mem)
  1023(MOVBstoreconst [sc] {sym1} (MOVDaddr [off] {sym2} ptr) mem) && ptr.Op != OpSB && canMergeSym(sym1, sym2) && sc.canAdd32(off) =>
  1024	(MOVBstoreconst [sc.addOffset32(off)] {mergeSym(sym1, sym2)} ptr mem)
  1025
  1026// MOVDaddr into MOVDaddridx
  1027(MOVDaddridx [off1] {sym1} (MOVDaddr [off2] {sym2} x) y) && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) && x.Op != OpSB =>
  1028       (MOVDaddridx [off1+off2] {mergeSym(sym1,sym2)} x y)
  1029(MOVDaddridx [off1] {sym1} x (MOVDaddr [off2] {sym2} y)) && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) && y.Op != OpSB =>
  1030       (MOVDaddridx [off1+off2] {mergeSym(sym1,sym2)} x y)
  1031
  1032// Absorb InvertFlags into branches.
  1033(BRC {c} (InvertFlags cmp) yes no) => (BRC {c.ReverseComparison()} cmp yes no)
  1034
  1035// Constant comparisons.
  1036(CMPconst (MOVDconst [x]) [y]) && x==int64(y) => (FlagEQ)
  1037(CMPconst (MOVDconst [x]) [y]) && x<int64(y) => (FlagLT)
  1038(CMPconst (MOVDconst [x]) [y]) && x>int64(y) => (FlagGT)
  1039(CMPUconst (MOVDconst [x]) [y]) && uint64(x)==uint64(y) => (FlagEQ)
  1040(CMPUconst (MOVDconst [x]) [y]) && uint64(x)<uint64(y) => (FlagLT)
  1041(CMPUconst (MOVDconst [x]) [y]) && uint64(x)>uint64(y) => (FlagGT)
  1042
  1043(CMPWconst (MOVDconst [x]) [y]) && int32(x)==int32(y) => (FlagEQ)
  1044(CMPWconst (MOVDconst [x]) [y]) && int32(x)<int32(y) => (FlagLT)
  1045(CMPWconst (MOVDconst [x]) [y]) && int32(x)>int32(y) => (FlagGT)
  1046(CMPWUconst (MOVDconst [x]) [y]) && uint32(x)==uint32(y) => (FlagEQ)
  1047(CMPWUconst (MOVDconst [x]) [y]) && uint32(x)<uint32(y) => (FlagLT)
  1048(CMPWUconst (MOVDconst [x]) [y]) && uint32(x)>uint32(y) => (FlagGT)
  1049
  1050(CMP(W|WU)const (MOVBZreg _) [c]) &&   0xff < c => (FlagLT)
  1051(CMP(W|WU)const (MOVHZreg _) [c]) && 0xffff < c => (FlagLT)
  1052
  1053(CMPconst  (SRDconst _ [c]) [n]) && c > 0 && n < 0 => (FlagGT)
  1054(CMPWconst (SRWconst _ [c]) [n]) && c > 0 && n < 0 => (FlagGT)
  1055
  1056(CMPUconst  (SRDconst _ [c]) [n]) && c > 0 && c < 64 && (1<<uint(64-c)) <= uint64(n) => (FlagLT)
  1057(CMPWUconst (SRWconst _ [c]) [n]) && c > 0 && c < 32 && (1<<uint(32-c)) <= uint32(n) => (FlagLT)
  1058
  1059(CMPWconst  (ANDWconst _ [m]) [n]) && int32(m) >= 0 &&  int32(m) <  int32(n) => (FlagLT)
  1060(CMPWUconst (ANDWconst _ [m]) [n]) && uint32(m) < uint32(n) => (FlagLT)
  1061
  1062(CMPconst  (RISBGZ x {r}) [c]) && c > 0 && r.OutMask() < uint64(c) => (FlagLT)
  1063(CMPUconst (RISBGZ x {r}) [c]) && r.OutMask() < uint64(uint32(c)) => (FlagLT)
  1064
  1065// Constant compare-and-branch with immediate.
  1066(CGIJ  {c} (MOVDconst [x]) [y] yes no) && c&s390x.Equal   != 0 &&  int64(x) ==  int64(y) => (First yes no)
  1067(CGIJ  {c} (MOVDconst [x]) [y] yes no) && c&s390x.Less    != 0 &&  int64(x) <   int64(y) => (First yes no)
  1068(CGIJ  {c} (MOVDconst [x]) [y] yes no) && c&s390x.Greater != 0 &&  int64(x) >   int64(y) => (First yes no)
  1069(CIJ   {c} (MOVDconst [x]) [y] yes no) && c&s390x.Equal   != 0 &&  int32(x) ==  int32(y) => (First yes no)
  1070(CIJ   {c} (MOVDconst [x]) [y] yes no) && c&s390x.Less    != 0 &&  int32(x) <   int32(y) => (First yes no)
  1071(CIJ   {c} (MOVDconst [x]) [y] yes no) && c&s390x.Greater != 0 &&  int32(x) >   int32(y) => (First yes no)
  1072(CLGIJ {c} (MOVDconst [x]) [y] yes no) && c&s390x.Equal   != 0 && uint64(x) == uint64(y) => (First yes no)
  1073(CLGIJ {c} (MOVDconst [x]) [y] yes no) && c&s390x.Less    != 0 && uint64(x) <  uint64(y) => (First yes no)
  1074(CLGIJ {c} (MOVDconst [x]) [y] yes no) && c&s390x.Greater != 0 && uint64(x) >  uint64(y) => (First yes no)
  1075(CLIJ  {c} (MOVDconst [x]) [y] yes no) && c&s390x.Equal   != 0 && uint32(x) == uint32(y) => (First yes no)
  1076(CLIJ  {c} (MOVDconst [x]) [y] yes no) && c&s390x.Less    != 0 && uint32(x) <  uint32(y) => (First yes no)
  1077(CLIJ  {c} (MOVDconst [x]) [y] yes no) && c&s390x.Greater != 0 && uint32(x) >  uint32(y) => (First yes no)
  1078(CGIJ  {c} (MOVDconst [x]) [y] yes no) && c&s390x.Equal   == 0 &&  int64(x) ==  int64(y) => (First no yes)
  1079(CGIJ  {c} (MOVDconst [x]) [y] yes no) && c&s390x.Less    == 0 &&  int64(x) <   int64(y) => (First no yes)
  1080(CGIJ  {c} (MOVDconst [x]) [y] yes no) && c&s390x.Greater == 0 &&  int64(x) >   int64(y) => (First no yes)
  1081(CIJ   {c} (MOVDconst [x]) [y] yes no) && c&s390x.Equal   == 0 &&  int32(x) ==  int32(y) => (First no yes)
  1082(CIJ   {c} (MOVDconst [x]) [y] yes no) && c&s390x.Less    == 0 &&  int32(x) <   int32(y) => (First no yes)
  1083(CIJ   {c} (MOVDconst [x]) [y] yes no) && c&s390x.Greater == 0 &&  int32(x) >   int32(y) => (First no yes)
  1084(CLGIJ {c} (MOVDconst [x]) [y] yes no) && c&s390x.Equal   == 0 && uint64(x) == uint64(y) => (First no yes)
  1085(CLGIJ {c} (MOVDconst [x]) [y] yes no) && c&s390x.Less    == 0 && uint64(x) <  uint64(y) => (First no yes)
  1086(CLGIJ {c} (MOVDconst [x]) [y] yes no) && c&s390x.Greater == 0 && uint64(x) >  uint64(y) => (First no yes)
  1087(CLIJ  {c} (MOVDconst [x]) [y] yes no) && c&s390x.Equal   == 0 && uint32(x) == uint32(y) => (First no yes)
  1088(CLIJ  {c} (MOVDconst [x]) [y] yes no) && c&s390x.Less    == 0 && uint32(x) <  uint32(y) => (First no yes)
  1089(CLIJ  {c} (MOVDconst [x]) [y] yes no) && c&s390x.Greater == 0 && uint32(x) >  uint32(y) => (First no yes)
  1090
  1091// Constant compare-and-branch with immediate when unsigned comparison with zero.
  1092(C(L|LG)IJ {s390x.GreaterOrEqual} _ [0] yes no) => (First yes no)
  1093(C(L|LG)IJ {s390x.Less}           _ [0] yes no) => (First no yes)
  1094
  1095// Constant compare-and-branch when operands match.
  1096(C(GR|R|LGR|LR)J {c} x y yes no) && x == y && c&s390x.Equal != 0 => (First yes no)
  1097(C(GR|R|LGR|LR)J {c} x y yes no) && x == y && c&s390x.Equal == 0 => (First no yes)
  1098
  1099// Convert 64-bit comparisons to 32-bit comparisons and signed comparisons
  1100// to unsigned comparisons.
  1101// Helps simplify constant comparison detection.
  1102(CM(P|PU)const (MOV(W|WZ)reg x) [c]) => (CMP(W|WU)const x [c])
  1103(CM(P|P|PU|PU)const x:(MOV(H|HZ|H|HZ)reg _) [c]) => (CMP(W|W|WU|WU)const x [c])
  1104(CM(P|P|PU|PU)const x:(MOV(B|BZ|B|BZ)reg _) [c]) => (CMP(W|W|WU|WU)const x [c])
  1105(CMPconst  (MOV(WZ|W)reg x:(ANDWconst [m] _)) [c]) && int32(m) >= 0 && c >= 0 => (CMPWUconst x [c])
  1106(CMPUconst (MOV(WZ|W)reg x:(ANDWconst [m] _)) [c]) && int32(m) >= 0           => (CMPWUconst x [c])
  1107(CMPconst  x:(SRDconst _ [c]) [n]) && c > 0 && n >= 0 => (CMPUconst  x [n])
  1108(CMPWconst x:(SRWconst _ [c]) [n]) && c > 0 && n >= 0 => (CMPWUconst x [n])
  1109
  1110// Absorb sign and zero extensions into 32-bit comparisons.
  1111(CMP(W|W|WU|WU)      x (MOV(W|WZ|W|WZ)reg y))   => (CMP(W|W|WU|WU) x y)
  1112(CMP(W|W|WU|WU)      (MOV(W|WZ|W|WZ)reg x) y)   => (CMP(W|W|WU|WU) x y)
  1113(CMP(W|W|WU|WU)const (MOV(W|WZ|W|WZ)reg x) [c]) => (CMP(W|W|WU|WU)const x [c])
  1114
  1115// Absorb flag constants into branches.
  1116(BRC {c} (FlagEQ) yes no) && c&s390x.Equal     != 0 => (First yes no)
  1117(BRC {c} (FlagLT) yes no) && c&s390x.Less      != 0 => (First yes no)
  1118(BRC {c} (FlagGT) yes no) && c&s390x.Greater   != 0 => (First yes no)
  1119(BRC {c} (FlagOV) yes no) && c&s390x.Unordered != 0 => (First yes no)
  1120
  1121(BRC {c} (FlagEQ) yes no) && c&s390x.Equal     == 0 => (First no yes)
  1122(BRC {c} (FlagLT) yes no) && c&s390x.Less      == 0 => (First no yes)
  1123(BRC {c} (FlagGT) yes no) && c&s390x.Greater   == 0 => (First no yes)
  1124(BRC {c} (FlagOV) yes no) && c&s390x.Unordered == 0 => (First no yes)
  1125
  1126// Absorb flag constants into SETxx ops.
  1127(LOCGR {c} _ x (FlagEQ)) && c&s390x.Equal     != 0 => x
  1128(LOCGR {c} _ x (FlagLT)) && c&s390x.Less      != 0 => x
  1129(LOCGR {c} _ x (FlagGT)) && c&s390x.Greater   != 0 => x
  1130(LOCGR {c} _ x (FlagOV)) && c&s390x.Unordered != 0 => x
  1131
  1132(LOCGR {c} x _ (FlagEQ)) && c&s390x.Equal     == 0 => x
  1133(LOCGR {c} x _ (FlagLT)) && c&s390x.Less      == 0 => x
  1134(LOCGR {c} x _ (FlagGT)) && c&s390x.Greater   == 0 => x
  1135(LOCGR {c} x _ (FlagOV)) && c&s390x.Unordered == 0 => x
  1136
  1137// Remove redundant *const ops
  1138(ADDconst [0] x) => x
  1139(ADDWconst [c] x) && int32(c)==0 => x
  1140(SUBconst [0] x) => x
  1141(SUBWconst [c] x) && int32(c) == 0 => x
  1142(ANDconst [0] _)                 => (MOVDconst [0])
  1143(ANDWconst [c] _) && int32(c)==0  => (MOVDconst [0])
  1144(ANDconst [-1] x)                => x
  1145(ANDWconst [c] x) && int32(c)==-1 => x
  1146(ORconst [0] x)                  => x
  1147(ORWconst [c] x) && int32(c)==0   => x
  1148(ORconst [-1] _)                 => (MOVDconst [-1])
  1149(ORWconst [c] _) && int32(c)==-1  => (MOVDconst [-1])
  1150(XORconst [0] x)                  => x
  1151(XORWconst [c] x) && int32(c)==0   => x
  1152
  1153// Shifts by zero (may be inserted during multiplication strength reduction).
  1154((SLD|SLW|SRD|SRW|SRAD|SRAW)const x [0]) => x
  1155
  1156// Convert constant subtracts to constant adds.
  1157(SUBconst [c] x) && c != -(1<<31) => (ADDconst [-c] x)
  1158(SUBWconst [c] x) => (ADDWconst [-int32(c)] x)
  1159
  1160// generic constant folding
  1161// TODO: more of this
  1162(ADDconst [c] (MOVDconst [d])) => (MOVDconst [int64(c)+d])
  1163(ADDWconst [c] (MOVDconst [d])) => (MOVDconst [int64(c)+d])
  1164(ADDconst [c] (ADDconst [d] x)) && is32Bit(int64(c)+int64(d)) => (ADDconst [c+d] x)
  1165(ADDWconst [c] (ADDWconst [d] x)) => (ADDWconst [int32(c+d)] x)
  1166(SUBconst (MOVDconst [d]) [c]) => (MOVDconst [d-int64(c)])
  1167(SUBconst (SUBconst x [d]) [c]) && is32Bit(-int64(c)-int64(d)) => (ADDconst [-c-d] x)
  1168(SRADconst [c] (MOVDconst [d])) => (MOVDconst [d>>uint64(c)])
  1169(SRAWconst [c] (MOVDconst [d])) => (MOVDconst [int64(int32(d))>>uint64(c)])
  1170(NEG (MOVDconst [c])) => (MOVDconst [-c])
  1171(NEGW (MOVDconst [c])) => (MOVDconst [int64(int32(-c))])
  1172(MULLDconst [c] (MOVDconst [d])) => (MOVDconst [int64(c)*d])
  1173(MULLWconst [c] (MOVDconst [d])) => (MOVDconst [int64(c*int32(d))])
  1174(AND (MOVDconst [c]) (MOVDconst [d])) => (MOVDconst [c&d])
  1175(ANDconst [c] (MOVDconst [d])) => (MOVDconst [c&d])
  1176(ANDWconst [c] (MOVDconst [d])) => (MOVDconst [int64(c)&d])
  1177(OR (MOVDconst [c]) (MOVDconst [d])) => (MOVDconst [c|d])
  1178(ORconst [c] (MOVDconst [d])) => (MOVDconst [c|d])
  1179(ORWconst [c] (MOVDconst [d])) => (MOVDconst [int64(c)|d])
  1180(XOR (MOVDconst [c]) (MOVDconst [d])) => (MOVDconst [c^d])
  1181(XORconst [c] (MOVDconst [d])) => (MOVDconst [c^d])
  1182(XORWconst [c] (MOVDconst [d])) => (MOVDconst [int64(c)^d])
  1183(LoweredRound32F x:(FMOVSconst)) => x
  1184(LoweredRound64F x:(FMOVDconst)) => x
  1185
  1186// generic simplifications
  1187// TODO: more of this
  1188(ADD x (NEG y)) => (SUB x y)
  1189(ADDW x (NEGW y)) => (SUBW x y)
  1190(SUB x x) => (MOVDconst [0])
  1191(SUBW x x) => (MOVDconst [0])
  1192(AND x x) => x
  1193(ANDW x x) => x
  1194(OR x x) => x
  1195(ORW x x) => x
  1196(XOR x x) => (MOVDconst [0])
  1197(XORW x x) => (MOVDconst [0])
  1198(NEG (ADDconst [c] (NEG x))) && c != -(1<<31) => (ADDconst [-c] x)
  1199(MOVBZreg (ANDWconst [m] x)) => (MOVWZreg (ANDWconst <typ.UInt32> [int32( uint8(m))] x))
  1200(MOVHZreg (ANDWconst [m] x)) => (MOVWZreg (ANDWconst <typ.UInt32> [int32(uint16(m))] x))
  1201(MOVBreg  (ANDWconst [m] x)) &&  int8(m) >= 0 => (MOVWZreg (ANDWconst <typ.UInt32> [int32( uint8(m))] x))
  1202(MOVHreg  (ANDWconst [m] x)) && int16(m) >= 0 => (MOVWZreg (ANDWconst <typ.UInt32> [int32(uint16(m))] x))
  1203
  1204// carry flag generation
  1205// (only constant fold carry of zero)
  1206(Select1 (ADDCconst (MOVDconst [c]) [d]))
  1207  && uint64(c+int64(d)) >= uint64(c) && c+int64(d) == 0
  1208  => (FlagEQ)
  1209(Select1 (ADDCconst (MOVDconst [c]) [d]))
  1210  && uint64(c+int64(d)) >= uint64(c) && c+int64(d) != 0
  1211  => (FlagLT)
  1212
  1213// borrow flag generation
  1214// (only constant fold borrow of zero)
  1215(Select1 (SUBC (MOVDconst [c]) (MOVDconst [d])))
  1216  && uint64(d) <= uint64(c) && c-d == 0
  1217  => (FlagGT)
  1218(Select1 (SUBC (MOVDconst [c]) (MOVDconst [d])))
  1219  && uint64(d) <= uint64(c) && c-d != 0
  1220  => (FlagOV)
  1221
  1222// add with carry
  1223(ADDE x y (FlagEQ)) => (ADDC x y)
  1224(ADDE x y (FlagLT)) => (ADDC x y)
  1225(ADDC x (MOVDconst [c])) && is16Bit(c) => (ADDCconst x [int16(c)])
  1226(Select0 (ADDCconst (MOVDconst [c]) [d])) => (MOVDconst [c+int64(d)])
  1227
  1228// subtract with borrow
  1229(SUBE x y (FlagGT)) => (SUBC x y)
  1230(SUBE x y (FlagOV)) => (SUBC x y)
  1231(Select0 (SUBC (MOVDconst [c]) (MOVDconst [d]))) => (MOVDconst [c-d])
  1232
  1233// collapse carry chain
  1234(ADDE x y (Select1 (ADDCconst [-1] (Select0 (ADDE (MOVDconst [0]) (MOVDconst [0]) c)))))
  1235  => (ADDE x y c)
  1236
  1237// collapse borrow chain
  1238(SUBE x y (Select1 (SUBC (MOVDconst [0]) (NEG (Select0 (SUBE (MOVDconst [0]) (MOVDconst [0]) c))))))
  1239  => (SUBE x y c)
  1240
  1241// branch on carry
  1242(C(G|LG)IJ {s390x.Equal}         (Select0 (ADDE (MOVDconst [0]) (MOVDconst [0]) carry)) [0]) => (BRC {s390x.NoCarry} carry)
  1243(C(G|LG)IJ {s390x.Equal}         (Select0 (ADDE (MOVDconst [0]) (MOVDconst [0]) carry)) [1]) => (BRC {s390x.Carry}   carry)
  1244(C(G|LG)IJ {s390x.LessOrGreater} (Select0 (ADDE (MOVDconst [0]) (MOVDconst [0]) carry)) [0]) => (BRC {s390x.Carry}   carry)
  1245(C(G|LG)IJ {s390x.LessOrGreater} (Select0 (ADDE (MOVDconst [0]) (MOVDconst [0]) carry)) [1]) => (BRC {s390x.NoCarry} carry)
  1246(C(G|LG)IJ {s390x.Greater}       (Select0 (ADDE (MOVDconst [0]) (MOVDconst [0]) carry)) [0]) => (BRC {s390x.Carry}   carry)
  1247
  1248// branch on borrow
  1249(C(G|LG)IJ {s390x.Equal}         (NEG (Select0 (SUBE (MOVDconst [0]) (MOVDconst [0]) borrow))) [0]) => (BRC {s390x.NoBorrow} borrow)
  1250(C(G|LG)IJ {s390x.Equal}         (NEG (Select0 (SUBE (MOVDconst [0]) (MOVDconst [0]) borrow))) [1]) => (BRC {s390x.Borrow}   borrow)
  1251(C(G|LG)IJ {s390x.LessOrGreater} (NEG (Select0 (SUBE (MOVDconst [0]) (MOVDconst [0]) borrow))) [0]) => (BRC {s390x.Borrow}   borrow)
  1252(C(G|LG)IJ {s390x.LessOrGreater} (NEG (Select0 (SUBE (MOVDconst [0]) (MOVDconst [0]) borrow))) [1]) => (BRC {s390x.NoBorrow} borrow)
  1253(C(G|LG)IJ {s390x.Greater}       (NEG (Select0 (SUBE (MOVDconst [0]) (MOVDconst [0]) borrow))) [0]) => (BRC {s390x.Borrow}   borrow)
  1254
  1255// fused multiply-add
  1256(Select0 (F(ADD|SUB) (FMUL y z) x)) && x.Block.Func.useFMA(v) => (FM(ADD|SUB) x y z)
  1257(Select0 (F(ADDS|SUBS) (FMULS y z) x)) && x.Block.Func.useFMA(v) => (FM(ADDS|SUBS) x y z)
  1258
  1259// Convert floating point comparisons against zero into 'load and test' instructions.
  1260(F(CMP|CMPS) x (FMOV(D|S)const [0.0])) => (LT(D|E)BR x)
  1261(F(CMP|CMPS) (FMOV(D|S)const [0.0]) x) => (InvertFlags (LT(D|E)BR <v.Type> x))
  1262
  1263// FSUB, FSUBS, FADD, FADDS now produce a condition code representing the
  1264// comparison of the result with 0.0. If a compare with zero instruction
  1265// (e.g. LTDBR) is following one of those instructions, we can use the
  1266// generated flag and remove the comparison instruction.
  1267// Note: when inserting Select1 ops we need to ensure they are in the
  1268// same block as their argument. We could also use @x.Block for this
  1269// but moving the flag generating value to a different block seems to
  1270// increase the likelihood that the flags value will have to be regenerated
  1271// by flagalloc which is not what we want.
  1272(LTDBR (Select0 x:(F(ADD|SUB) _ _)))   && b == x.Block => (Select1 x)
  1273(LTEBR (Select0 x:(F(ADDS|SUBS) _ _))) && b == x.Block => (Select1 x)
  1274
  1275// Fold memory operations into operations.
  1276// Exclude global data (SB) because these instructions cannot handle relative addresses.
  1277// TODO(mundaym): indexed versions of these?
  1278((ADD|SUB|MULLD|AND|OR|XOR) <t> x g:(MOVDload [off] {sym} ptr mem))
  1279  && ptr.Op != OpSB
  1280  && is20Bit(int64(off))
  1281  && canMergeLoadClobber(v, g, x)
  1282  && clobber(g)
  1283  => ((ADD|SUB|MULLD|AND|OR|XOR)load <t> [off] {sym} x ptr mem)
  1284((ADD|SUB|MULL|AND|OR|XOR)W <t> x g:(MOVWload [off] {sym} ptr mem))
  1285  && ptr.Op != OpSB
  1286  && is20Bit(int64(off))
  1287  && canMergeLoadClobber(v, g, x)
  1288  && clobber(g)
  1289  => ((ADD|SUB|MULL|AND|OR|XOR)Wload <t> [off] {sym} x ptr mem)
  1290((ADD|SUB|MULL|AND|OR|XOR)W <t> x g:(MOVWZload [off] {sym} ptr mem))
  1291  && ptr.Op != OpSB
  1292  && is20Bit(int64(off))
  1293  && canMergeLoadClobber(v, g, x)
  1294  && clobber(g)
  1295  => ((ADD|SUB|MULL|AND|OR|XOR)Wload <t> [off] {sym} x ptr mem)
  1296
  1297// Combine stores into store multiples.
  1298// 32-bit
  1299(MOVWstore [i] {s} p w1 x:(MOVWstore [i-4] {s} p w0 mem))
  1300  && p.Op != OpSB
  1301  && x.Uses == 1
  1302  && is20Bit(int64(i)-4)
  1303  && setPos(v, x.Pos)
  1304  && clobber(x)
  1305  => (STM2 [i-4] {s} p w0 w1 mem)
  1306(MOVWstore [i] {s} p w2 x:(STM2 [i-8] {s} p w0 w1 mem))
  1307  && x.Uses == 1
  1308  && is20Bit(int64(i)-8)
  1309  && setPos(v, x.Pos)
  1310  && clobber(x)
  1311  => (STM3 [i-8] {s} p w0 w1 w2 mem)
  1312(MOVWstore [i] {s} p w3 x:(STM3 [i-12] {s} p w0 w1 w2 mem))
  1313  && x.Uses == 1
  1314  && is20Bit(int64(i)-12)
  1315  && setPos(v, x.Pos)
  1316  && clobber(x)
  1317  => (STM4 [i-12] {s} p w0 w1 w2 w3 mem)
  1318(STM2 [i] {s} p w2 w3 x:(STM2 [i-8] {s} p w0 w1 mem))
  1319  && x.Uses == 1
  1320  && is20Bit(int64(i)-8)
  1321  && setPos(v, x.Pos)
  1322  && clobber(x)
  1323  => (STM4 [i-8] {s} p w0 w1 w2 w3 mem)
  1324// 64-bit
  1325(MOVDstore [i] {s} p w1 x:(MOVDstore [i-8] {s} p w0 mem))
  1326  && p.Op != OpSB
  1327  && x.Uses == 1
  1328  && is20Bit(int64(i)-8)
  1329  && setPos(v, x.Pos)
  1330  && clobber(x)
  1331  => (STMG2 [i-8] {s} p w0 w1 mem)
  1332(MOVDstore [i] {s} p w2 x:(STMG2 [i-16] {s} p w0 w1 mem))
  1333  && x.Uses == 1
  1334  && is20Bit(int64(i)-16)
  1335  && setPos(v, x.Pos)
  1336  && clobber(x)
  1337  => (STMG3 [i-16] {s} p w0 w1 w2 mem)
  1338(MOVDstore [i] {s} p w3 x:(STMG3 [i-24] {s} p w0 w1 w2 mem))
  1339  && x.Uses == 1
  1340  && is20Bit(int64(i)-24)
  1341  && setPos(v, x.Pos)
  1342  && clobber(x)
  1343  => (STMG4 [i-24] {s} p w0 w1 w2 w3 mem)
  1344(STMG2 [i] {s} p w2 w3 x:(STMG2 [i-16] {s} p w0 w1 mem))
  1345  && x.Uses == 1
  1346  && is20Bit(int64(i)-16)
  1347  && setPos(v, x.Pos)
  1348  && clobber(x)
  1349  => (STMG4 [i-16] {s} p w0 w1 w2 w3 mem)
  1350
  1351// Convert 32-bit store multiples into 64-bit stores.
  1352(STM2 [i] {s} p (SRDconst [32] x) x mem) => (MOVDstore [i] {s} p x mem)
  1353
  1354// Fold bit reversal into loads.
  1355(MOVWBR x:(MOVWZload    [off] {sym} ptr     mem)) && x.Uses == 1 => @x.Block (MOVWZreg (MOVWBRload    [off] {sym} ptr     mem)) // need zero extension?
  1356(MOVWBR x:(MOVWZloadidx [off] {sym} ptr idx mem)) && x.Uses == 1 => @x.Block (MOVWZreg (MOVWBRloadidx [off] {sym} ptr idx mem)) // need zero extension?
  1357(MOVDBR x:(MOVDload     [off] {sym} ptr     mem)) && x.Uses == 1 => @x.Block (MOVDBRload    [off] {sym} ptr     mem)
  1358(MOVDBR x:(MOVDloadidx  [off] {sym} ptr idx mem)) && x.Uses == 1 => @x.Block (MOVDBRloadidx [off] {sym} ptr idx mem)
  1359
  1360// Fold bit reversal into stores.
  1361(MOV(D|W)store    [off] {sym} ptr     r:(MOV(D|W)BR x) mem) && r.Uses == 1 => (MOV(D|W)BRstore    [off] {sym} ptr     x mem)
  1362(MOV(D|W)storeidx [off] {sym} ptr idx r:(MOV(D|W)BR x) mem) && r.Uses == 1 => (MOV(D|W)BRstoreidx [off] {sym} ptr idx x mem)
  1363
  1364// Special bswap16 rules
  1365(Bswap16 x:(MOVHZload    [off] {sym} ptr     mem)) => @x.Block (MOVHZreg (MOVHBRload    [off] {sym} ptr     mem))
  1366(Bswap16 x:(MOVHZloadidx [off] {sym} ptr idx mem)) => @x.Block (MOVHZreg (MOVHBRloadidx [off] {sym} ptr idx mem))
  1367(MOVHstore    [off] {sym} ptr     (Bswap16 val) mem) => (MOVHBRstore    [off] {sym} ptr     val mem)
  1368(MOVHstoreidx [off] {sym} ptr idx (Bswap16 val) mem) => (MOVHBRstoreidx [off] {sym} ptr idx val mem)

View as plain text