github.com/mattn/go@v0.0.0-20171011075504-07f7db3ea99f/src/cmd/compile/internal/ssa/gen/S390X.rules (about)

     1  // Copyright 2016 The Go Authors. All rights reserved.
     2  // Use of this source code is governed by a BSD-style
     3  // license that can be found in the LICENSE file.
     4  
     5  // Lowering arithmetic
     6  (Add64  x y) -> (ADD  x y)
     7  (AddPtr x y) -> (ADD  x y)
     8  (Add32  x y) -> (ADDW  x y)
     9  (Add16  x y) -> (ADDW  x y)
    10  (Add8   x y) -> (ADDW  x y)
    11  (Add32F x y) -> (FADDS x y)
    12  (Add64F x y) -> (FADD x y)
    13  
    14  (Sub64  x y) -> (SUB  x y)
    15  (SubPtr x y) -> (SUB  x y)
    16  (Sub32  x y) -> (SUBW  x y)
    17  (Sub16  x y) -> (SUBW  x y)
    18  (Sub8   x y) -> (SUBW  x y)
    19  (Sub32F x y) -> (FSUBS x y)
    20  (Sub64F x y) -> (FSUB x y)
    21  
    22  (Mul64  x y) -> (MULLD  x y)
    23  (Mul32  x y) -> (MULLW  x y)
    24  (Mul16  x y) -> (MULLW  x y)
    25  (Mul8   x y) -> (MULLW  x y)
    26  (Mul32F x y) -> (FMULS x y)
    27  (Mul64F x y) -> (FMUL x y)
    28  
    29  (Div32F x y) -> (FDIVS x y)
    30  (Div64F x y) -> (FDIV x y)
    31  
    32  (Div64  x y) -> (DIVD  x y)
    33  (Div64u x y) -> (DIVDU x y)
    34  // DIVW/DIVWU has a 64-bit dividend and a 32-bit divisor,
    35  // so a sign/zero extension of the dividend is required.
    36  (Div32  x y) -> (DIVW  (MOVWreg x) y)
    37  (Div32u x y) -> (DIVWU (MOVWZreg x) y)
    38  (Div16  x y) -> (DIVW  (MOVHreg x) (MOVHreg y))
    39  (Div16u x y) -> (DIVWU (MOVHZreg x) (MOVHZreg y))
    40  (Div8   x y) -> (DIVW  (MOVBreg x) (MOVBreg y))
    41  (Div8u  x y) -> (DIVWU (MOVBZreg x) (MOVBZreg y))
    42  
    43  (Hmul64  x y) -> (MULHD  x y)
    44  (Hmul64u x y) -> (MULHDU x y)
    45  (Hmul32  x y) -> (SRDconst [32] (MULLD (MOVWreg x) (MOVWreg y)))
    46  (Hmul32u x y) -> (SRDconst [32] (MULLD (MOVWZreg x) (MOVWZreg y)))
    47  
    48  (Mod64  x y) -> (MODD  x y)
    49  (Mod64u x y) -> (MODDU x y)
    50  // MODW/MODWU has a 64-bit dividend and a 32-bit divisor,
    51  // so a sign/zero extension of the dividend is required.
    52  (Mod32  x y) -> (MODW  (MOVWreg x) y)
    53  (Mod32u x y) -> (MODWU (MOVWZreg x) y)
    54  (Mod16  x y) -> (MODW  (MOVHreg x) (MOVHreg y))
    55  (Mod16u x y) -> (MODWU (MOVHZreg x) (MOVHZreg y))
    56  (Mod8   x y) -> (MODW  (MOVBreg x) (MOVBreg y))
    57  (Mod8u  x y) -> (MODWU (MOVBZreg x) (MOVBZreg y))
    58  
    59  // (x + y) / 2 with x>=y -> (x - y) / 2 + y
    60  (Avg64u <t> x y) -> (ADD (SRDconst <t> (SUB <t> x y) [1]) y)
    61  
    62  (And64 x y) -> (AND x y)
    63  (And32 x y) -> (ANDW x y)
    64  (And16 x y) -> (ANDW x y)
    65  (And8  x y) -> (ANDW x y)
    66  
    67  (Or64 x y) -> (OR x y)
    68  (Or32 x y) -> (ORW x y)
    69  (Or16 x y) -> (ORW x y)
    70  (Or8  x y) -> (ORW x y)
    71  
    72  (Xor64 x y) -> (XOR x y)
    73  (Xor32 x y) -> (XORW x y)
    74  (Xor16 x y) -> (XORW x y)
    75  (Xor8  x y) -> (XORW x y)
    76  
    77  (Neg64  x) -> (NEG x)
    78  (Neg32  x) -> (NEGW x)
    79  (Neg16  x) -> (NEGW (MOVHreg x))
    80  (Neg8   x) -> (NEGW (MOVBreg x))
    81  (Neg32F x) -> (FNEGS x)
    82  (Neg64F x) -> (FNEG x)
    83  
    84  (Com64 x) -> (NOT x)
    85  (Com32 x) -> (NOTW x)
    86  (Com16 x) -> (NOTW x)
    87  (Com8  x) -> (NOTW x)
    88  (NOT x) && true -> (XOR (MOVDconst [-1]) x)
    89  (NOTW x) && true -> (XORWconst [-1] x)
    90  
    91  // Lowering boolean ops
    92  (AndB x y) -> (ANDW x y)
    93  (OrB x y) -> (ORW x y)
    94  (Not x) -> (XORWconst [1] x)
    95  
    96  // Lowering pointer arithmetic
    97  (OffPtr [off] ptr:(SP)) -> (MOVDaddr [off] ptr)
    98  (OffPtr [off] ptr) && is32Bit(off) -> (ADDconst [off] ptr)
    99  (OffPtr [off] ptr) -> (ADD (MOVDconst [off]) ptr)
   100  
   101  // Ctz(x) = 64 - findLeftmostOne((x-1)&^x)
   102  (Ctz64 <t> x) -> (SUB (MOVDconst [64]) (FLOGR (AND <t> (SUBconst <t> [1] x) (NOT <t> x))))
   103  (Ctz32 <t> x) -> (SUB (MOVDconst [64]) (FLOGR (MOVWZreg (ANDW <t> (SUBWconst <t> [1] x) (NOTW <t> x)))))
   104  
   105  (BitLen64 x) -> (SUB (MOVDconst [64]) (FLOGR x))
   106  
   107  (Bswap64 x) -> (MOVDBR x)
   108  (Bswap32 x) -> (MOVWBR x)
   109  
   110  // math package intrinsics
   111  (Sqrt  x) -> (FSQRT x)
   112  (Floor x) -> (FIDBR [7] x)
   113  (Ceil  x) -> (FIDBR [6] x)
   114  (Trunc x) -> (FIDBR [5] x)
   115  (Round x) -> (FIDBR [1] x)
   116  
   117  // Atomic loads.
   118  (AtomicLoad32 ptr mem) -> (MOVWZatomicload ptr mem)
   119  (AtomicLoad64 ptr mem) -> (MOVDatomicload ptr mem)
   120  (AtomicLoadPtr ptr mem) -> (MOVDatomicload ptr mem)
   121  
   122  // Atomic stores.
   123  (AtomicStore32 ptr val mem) -> (MOVWatomicstore ptr val mem)
   124  (AtomicStore64 ptr val mem) -> (MOVDatomicstore ptr val mem)
   125  (AtomicStorePtrNoWB ptr val mem) -> (MOVDatomicstore ptr val mem)
   126  
   127  // Atomic adds.
   128  (AtomicAdd32 ptr val mem) -> (AddTupleFirst32 val (LAA ptr val mem))
   129  (AtomicAdd64 ptr val mem) -> (AddTupleFirst64 val (LAAG ptr val mem))
   130  (Select0 <t> (AddTupleFirst32 val tuple)) -> (ADDW val (Select0 <t> tuple))
   131  (Select1     (AddTupleFirst32   _ tuple)) -> (Select1 tuple)
   132  (Select0 <t> (AddTupleFirst64 val tuple)) -> (ADD val (Select0 <t> tuple))
   133  (Select1     (AddTupleFirst64   _ tuple)) -> (Select1 tuple)
   134  
   135  // Atomic exchanges.
   136  (AtomicExchange32 ptr val mem) -> (LoweredAtomicExchange32 ptr val mem)
   137  (AtomicExchange64 ptr val mem) -> (LoweredAtomicExchange64 ptr val mem)
   138  
   139  // Atomic compare and swap.
   140  (AtomicCompareAndSwap32 ptr old new_ mem) -> (LoweredAtomicCas32 ptr old new_ mem)
   141  (AtomicCompareAndSwap64 ptr old new_ mem) -> (LoweredAtomicCas64 ptr old new_ mem)
   142  
   143  // Lowering extension
   144  // Note: we always extend to 64 bits even though some ops don't need that many result bits.
   145  (SignExt8to16  x) -> (MOVBreg x)
   146  (SignExt8to32  x) -> (MOVBreg x)
   147  (SignExt8to64  x) -> (MOVBreg x)
   148  (SignExt16to32 x) -> (MOVHreg x)
   149  (SignExt16to64 x) -> (MOVHreg x)
   150  (SignExt32to64 x) -> (MOVWreg x)
   151  
   152  (ZeroExt8to16  x) -> (MOVBZreg x)
   153  (ZeroExt8to32  x) -> (MOVBZreg x)
   154  (ZeroExt8to64  x) -> (MOVBZreg x)
   155  (ZeroExt16to32 x) -> (MOVHZreg x)
   156  (ZeroExt16to64 x) -> (MOVHZreg x)
   157  (ZeroExt32to64 x) -> (MOVWZreg x)
   158  
   159  (Slicemask <t> x) -> (SRADconst (NEG <t> x) [63])
   160  
   161  // Lowering truncation
   162  // Because we ignore high parts of registers, truncates are just copies.
   163  (Trunc16to8  x) -> x
   164  (Trunc32to8  x) -> x
   165  (Trunc32to16 x) -> x
   166  (Trunc64to8  x) -> x
   167  (Trunc64to16 x) -> x
   168  (Trunc64to32 x) -> x
   169  
   170  // Lowering float <-> int
   171  (Cvt32to32F x) -> (CEFBRA x)
   172  (Cvt32to64F x) -> (CDFBRA x)
   173  (Cvt64to32F x) -> (CEGBRA x)
   174  (Cvt64to64F x) -> (CDGBRA x)
   175  
   176  (Cvt32Fto32 x) -> (CFEBRA x)
   177  (Cvt32Fto64 x) -> (CGEBRA x)
   178  (Cvt64Fto32 x) -> (CFDBRA x)
   179  (Cvt64Fto64 x) -> (CGDBRA x)
   180  
   181  (Cvt32Fto64F x) -> (LDEBR x)
   182  (Cvt64Fto32F x) -> (LEDBR x)
   183  
   184  (Round32F x) -> (LoweredRound32F x)
   185  (Round64F x) -> (LoweredRound64F x)
   186  
   187  // Lowering shifts
   188  // Unsigned shifts need to return 0 if shift amount is >= width of shifted value.
   189  //   result = (arg << shift) & (shift >= argbits ? 0 : 0xffffffffffffffff)
   190  (Lsh64x64 <t> x y) -> (AND (SLD <t> x y) (SUBEcarrymask <t> (CMPUconst y [63])))
   191  (Lsh64x32 <t> x y) -> (AND (SLD <t> x y) (SUBEcarrymask <t> (CMPWUconst y [63])))
   192  (Lsh64x16 <t> x y) -> (AND (SLD <t> x y) (SUBEcarrymask <t> (CMPWUconst (MOVHZreg y) [63])))
   193  (Lsh64x8  <t> x y) -> (AND (SLD <t> x y) (SUBEcarrymask <t> (CMPWUconst (MOVBZreg y) [63])))
   194  
   195  (Lsh32x64 <t> x y) -> (ANDW (SLW <t> x y) (SUBEWcarrymask <t> (CMPUconst y [31])))
   196  (Lsh32x32 <t> x y) -> (ANDW (SLW <t> x y) (SUBEWcarrymask <t> (CMPWUconst y [31])))
   197  (Lsh32x16 <t> x y) -> (ANDW (SLW <t> x y) (SUBEWcarrymask <t> (CMPWUconst (MOVHZreg y) [31])))
   198  (Lsh32x8  <t> x y) -> (ANDW (SLW <t> x y) (SUBEWcarrymask <t> (CMPWUconst (MOVBZreg y) [31])))
   199  
   200  (Lsh16x64 <t> x y) -> (ANDW (SLW <t> x y) (SUBEWcarrymask <t> (CMPUconst y [31])))
   201  (Lsh16x32 <t> x y) -> (ANDW (SLW <t> x y) (SUBEWcarrymask <t> (CMPWUconst y [31])))
   202  (Lsh16x16 <t> x y) -> (ANDW (SLW <t> x y) (SUBEWcarrymask <t> (CMPWUconst (MOVHZreg y) [31])))
   203  (Lsh16x8  <t> x y) -> (ANDW (SLW <t> x y) (SUBEWcarrymask <t> (CMPWUconst (MOVBZreg y) [31])))
   204  
   205  (Lsh8x64 <t> x y)  -> (ANDW (SLW <t> x y) (SUBEWcarrymask <t> (CMPUconst y [31])))
   206  (Lsh8x32 <t> x y)  -> (ANDW (SLW <t> x y) (SUBEWcarrymask <t> (CMPWUconst y [31])))
   207  (Lsh8x16 <t> x y)  -> (ANDW (SLW <t> x y) (SUBEWcarrymask <t> (CMPWUconst (MOVHZreg y) [31])))
   208  (Lsh8x8  <t> x y)  -> (ANDW (SLW <t> x y) (SUBEWcarrymask <t> (CMPWUconst (MOVBZreg y) [31])))
   209  
   210  (Rsh64Ux64 <t> x y) -> (AND (SRD <t> x y) (SUBEcarrymask <t> (CMPUconst y [63])))
   211  (Rsh64Ux32 <t> x y) -> (AND (SRD <t> x y) (SUBEcarrymask <t> (CMPWUconst y [63])))
   212  (Rsh64Ux16 <t> x y) -> (AND (SRD <t> x y) (SUBEcarrymask <t> (CMPWUconst (MOVHZreg y) [63])))
   213  (Rsh64Ux8  <t> x y) -> (AND (SRD <t> x y) (SUBEcarrymask <t> (CMPWUconst (MOVBZreg y) [63])))
   214  
   215  (Rsh32Ux64 <t> x y) -> (ANDW (SRW <t> x y) (SUBEWcarrymask <t> (CMPUconst y [31])))
   216  (Rsh32Ux32 <t> x y) -> (ANDW (SRW <t> x y) (SUBEWcarrymask <t> (CMPWUconst y [31])))
   217  (Rsh32Ux16 <t> x y) -> (ANDW (SRW <t> x y) (SUBEWcarrymask <t> (CMPWUconst (MOVHZreg y) [31])))
   218  (Rsh32Ux8  <t> x y) -> (ANDW (SRW <t> x y) (SUBEWcarrymask <t> (CMPWUconst (MOVBZreg y) [31])))
   219  
   220  (Rsh16Ux64 <t> x y) -> (ANDW (SRW <t> (MOVHZreg x) y) (SUBEWcarrymask <t> (CMPUconst y [15])))
   221  (Rsh16Ux32 <t> x y) -> (ANDW (SRW <t> (MOVHZreg x) y) (SUBEWcarrymask <t> (CMPWUconst y [15])))
   222  (Rsh16Ux16 <t> x y) -> (ANDW (SRW <t> (MOVHZreg x) y) (SUBEWcarrymask <t> (CMPWUconst (MOVHZreg y) [15])))
   223  (Rsh16Ux8  <t> x y) -> (ANDW (SRW <t> (MOVHZreg x) y) (SUBEWcarrymask <t> (CMPWUconst (MOVBZreg y) [15])))
   224  
   225  (Rsh8Ux64 <t> x y)  -> (ANDW (SRW <t> (MOVBZreg x) y) (SUBEWcarrymask <t> (CMPUconst y [7])))
   226  (Rsh8Ux32 <t> x y)  -> (ANDW (SRW <t> (MOVBZreg x) y) (SUBEWcarrymask <t> (CMPWUconst y [7])))
   227  (Rsh8Ux16 <t> x y)  -> (ANDW (SRW <t> (MOVBZreg x) y) (SUBEWcarrymask <t> (CMPWUconst (MOVHZreg y) [7])))
   228  (Rsh8Ux8  <t> x y)  -> (ANDW (SRW <t> (MOVBZreg x) y) (SUBEWcarrymask <t> (CMPWUconst (MOVBZreg y) [7])))
   229  
   230  // Signed right shift needs to return 0/-1 if shift amount is >= width of shifted value.
   231  // We implement this by setting the shift value to -1 (all ones) if the shift value is >= width.
   232  (Rsh64x64 <t> x y) -> (SRAD <t> x (OR <y.Type> y (NOT <y.Type> (SUBEcarrymask <y.Type> (CMPUconst y [63])))))
   233  (Rsh64x32 <t> x y) -> (SRAD <t> x (ORW <y.Type> y (NOTW <y.Type> (SUBEWcarrymask <y.Type> (CMPWUconst y [63])))))
   234  (Rsh64x16 <t> x y) -> (SRAD <t> x (ORW <y.Type> y (NOTW <y.Type> (SUBEWcarrymask <y.Type> (CMPWUconst (MOVHZreg y) [63])))))
   235  (Rsh64x8  <t> x y) -> (SRAD <t> x (ORW <y.Type> y (NOTW <y.Type> (SUBEWcarrymask <y.Type> (CMPWUconst (MOVBZreg y) [63])))))
   236  
   237  (Rsh32x64 <t> x y) -> (SRAW <t> x (OR <y.Type> y (NOT <y.Type> (SUBEcarrymask <y.Type> (CMPUconst y [31])))))
   238  (Rsh32x32 <t> x y) -> (SRAW <t> x (ORW <y.Type> y (NOTW <y.Type> (SUBEWcarrymask <y.Type> (CMPWUconst y [31])))))
   239  (Rsh32x16 <t> x y) -> (SRAW <t> x (ORW <y.Type> y (NOTW <y.Type> (SUBEWcarrymask <y.Type> (CMPWUconst (MOVHZreg y) [31])))))
   240  (Rsh32x8  <t> x y) -> (SRAW <t> x (ORW <y.Type> y (NOTW <y.Type> (SUBEWcarrymask <y.Type> (CMPWUconst (MOVBZreg y) [31])))))
   241  
   242  (Rsh16x64 <t> x y) -> (SRAW <t> (MOVHreg x) (OR <y.Type> y (NOT <y.Type> (SUBEcarrymask <y.Type> (CMPUconst y [15])))))
   243  (Rsh16x32 <t> x y) -> (SRAW <t> (MOVHreg x) (ORW <y.Type> y (NOTW <y.Type> (SUBEWcarrymask <y.Type> (CMPWUconst y [15])))))
   244  (Rsh16x16 <t> x y) -> (SRAW <t> (MOVHreg x) (ORW <y.Type> y (NOTW <y.Type> (SUBEWcarrymask <y.Type> (CMPWUconst (MOVHZreg y) [15])))))
   245  (Rsh16x8  <t> x y) -> (SRAW <t> (MOVHreg x) (ORW <y.Type> y (NOTW <y.Type> (SUBEWcarrymask <y.Type> (CMPWUconst (MOVBZreg y) [15])))))
   246  
   247  (Rsh8x64 <t> x y)  -> (SRAW <t> (MOVBreg x) (OR <y.Type> y (NOT <y.Type> (SUBEcarrymask <y.Type> (CMPUconst y [7])))))
   248  (Rsh8x32 <t> x y)  -> (SRAW <t> (MOVBreg x) (ORW <y.Type> y (NOTW <y.Type> (SUBEWcarrymask <y.Type> (CMPWUconst y [7])))))
   249  (Rsh8x16 <t> x y)  -> (SRAW <t> (MOVBreg x) (ORW <y.Type> y (NOTW <y.Type> (SUBEWcarrymask <y.Type> (CMPWUconst (MOVHZreg y) [7])))))
   250  (Rsh8x8  <t> x y)  -> (SRAW <t> (MOVBreg x) (ORW <y.Type> y (NOTW <y.Type> (SUBEWcarrymask <y.Type> (CMPWUconst (MOVBZreg y) [7])))))
   251  
   252  // Lowering comparisons
   253  (Less64  x y) -> (MOVDLT (MOVDconst [0]) (MOVDconst [1]) (CMP x y))
   254  (Less32  x y) -> (MOVDLT (MOVDconst [0]) (MOVDconst [1]) (CMPW x y))
   255  (Less16  x y) -> (MOVDLT (MOVDconst [0]) (MOVDconst [1]) (CMP (MOVHreg x) (MOVHreg y)))
   256  (Less8   x y) -> (MOVDLT (MOVDconst [0]) (MOVDconst [1]) (CMP (MOVBreg x) (MOVBreg y)))
   257  (Less64U x y) -> (MOVDLT (MOVDconst [0]) (MOVDconst [1]) (CMPU x y))
   258  (Less32U x y) -> (MOVDLT (MOVDconst [0]) (MOVDconst [1]) (CMPWU x y))
   259  (Less16U x y) -> (MOVDLT (MOVDconst [0]) (MOVDconst [1]) (CMPU (MOVHZreg x) (MOVHZreg y)))
   260  (Less8U  x y) -> (MOVDLT (MOVDconst [0]) (MOVDconst [1]) (CMPU (MOVBZreg x) (MOVBZreg y)))
   261  // Use SETG with reversed operands to dodge NaN case.
   262  (Less64F x y) -> (MOVDGTnoinv (MOVDconst [0]) (MOVDconst [1]) (FCMP y x))
   263  (Less32F x y) -> (MOVDGTnoinv (MOVDconst [0]) (MOVDconst [1]) (FCMPS y x))
   264  
   265  (Leq64  x y) -> (MOVDLE (MOVDconst [0]) (MOVDconst [1]) (CMP x y))
   266  (Leq32  x y) -> (MOVDLE (MOVDconst [0]) (MOVDconst [1]) (CMPW x y))
   267  (Leq16  x y) -> (MOVDLE (MOVDconst [0]) (MOVDconst [1]) (CMP (MOVHreg x) (MOVHreg y)))
   268  (Leq8   x y) -> (MOVDLE (MOVDconst [0]) (MOVDconst [1]) (CMP (MOVBreg x) (MOVBreg y)))
   269  (Leq64U x y) -> (MOVDLE (MOVDconst [0]) (MOVDconst [1]) (CMPU x y))
   270  (Leq32U x y) -> (MOVDLE (MOVDconst [0]) (MOVDconst [1]) (CMPWU x y))
   271  (Leq16U x y) -> (MOVDLE (MOVDconst [0]) (MOVDconst [1]) (CMPU (MOVHZreg x) (MOVHZreg y)))
   272  (Leq8U  x y) -> (MOVDLE (MOVDconst [0]) (MOVDconst [1]) (CMPU (MOVBZreg x) (MOVBZreg y)))
   273  // Use SETGE with reversed operands to dodge NaN case.
   274  (Leq64F x y) -> (MOVDGEnoinv (MOVDconst [0]) (MOVDconst [1]) (FCMP y x))
   275  (Leq32F x y) -> (MOVDGEnoinv (MOVDconst [0]) (MOVDconst [1]) (FCMPS y x))
   276  
   277  (Greater64  x y) -> (MOVDGT (MOVDconst [0]) (MOVDconst [1]) (CMP x y))
   278  (Greater32  x y) -> (MOVDGT (MOVDconst [0]) (MOVDconst [1]) (CMPW x y))
   279  (Greater16  x y) -> (MOVDGT (MOVDconst [0]) (MOVDconst [1]) (CMP (MOVHreg x) (MOVHreg y)))
   280  (Greater8   x y) -> (MOVDGT (MOVDconst [0]) (MOVDconst [1]) (CMP (MOVBreg x) (MOVBreg y)))
   281  (Greater64U x y) -> (MOVDGT (MOVDconst [0]) (MOVDconst [1]) (CMPU x y))
   282  (Greater32U x y) -> (MOVDGT (MOVDconst [0]) (MOVDconst [1]) (CMPWU x y))
   283  (Greater16U x y) -> (MOVDGT (MOVDconst [0]) (MOVDconst [1]) (CMPU (MOVHZreg x) (MOVHZreg y)))
   284  (Greater8U  x y) -> (MOVDGT (MOVDconst [0]) (MOVDconst [1]) (CMPU (MOVBZreg x) (MOVBZreg y)))
   285  (Greater64F x y) -> (MOVDGTnoinv (MOVDconst [0]) (MOVDconst [1]) (FCMP x y))
   286  (Greater32F x y) -> (MOVDGTnoinv (MOVDconst [0]) (MOVDconst [1]) (FCMPS x y))
   287  
   288  (Geq64  x y) -> (MOVDGE (MOVDconst [0]) (MOVDconst [1]) (CMP x y))
   289  (Geq32  x y) -> (MOVDGE (MOVDconst [0]) (MOVDconst [1]) (CMPW x y))
   290  (Geq16  x y) -> (MOVDGE (MOVDconst [0]) (MOVDconst [1]) (CMP (MOVHreg x) (MOVHreg y)))
   291  (Geq8   x y) -> (MOVDGE (MOVDconst [0]) (MOVDconst [1]) (CMP (MOVBreg x) (MOVBreg y)))
   292  (Geq64U x y) -> (MOVDGE (MOVDconst [0]) (MOVDconst [1]) (CMPU x y))
   293  (Geq32U x y) -> (MOVDGE (MOVDconst [0]) (MOVDconst [1]) (CMPWU x y))
   294  (Geq16U x y) -> (MOVDGE (MOVDconst [0]) (MOVDconst [1]) (CMPU (MOVHZreg x) (MOVHZreg y)))
   295  (Geq8U  x y) -> (MOVDGE (MOVDconst [0]) (MOVDconst [1]) (CMPU (MOVBZreg x) (MOVBZreg y)))
   296  (Geq64F x y) -> (MOVDGEnoinv (MOVDconst [0]) (MOVDconst [1]) (FCMP x y))
   297  (Geq32F x y) -> (MOVDGEnoinv (MOVDconst [0]) (MOVDconst [1]) (FCMPS x y))
   298  
   299  (Eq64  x y) -> (MOVDEQ (MOVDconst [0]) (MOVDconst [1]) (CMP x y))
   300  (Eq32  x y) -> (MOVDEQ (MOVDconst [0]) (MOVDconst [1]) (CMPW x y))
   301  (Eq16  x y) -> (MOVDEQ (MOVDconst [0]) (MOVDconst [1]) (CMP (MOVHreg x) (MOVHreg y)))
   302  (Eq8   x y) -> (MOVDEQ (MOVDconst [0]) (MOVDconst [1]) (CMP (MOVBreg x) (MOVBreg y)))
   303  (EqB   x y) -> (MOVDEQ (MOVDconst [0]) (MOVDconst [1]) (CMP (MOVBreg x) (MOVBreg y)))
   304  (EqPtr x y) -> (MOVDEQ (MOVDconst [0]) (MOVDconst [1]) (CMP x y))
   305  (Eq64F x y) -> (MOVDEQ (MOVDconst [0]) (MOVDconst [1]) (FCMP x y))
   306  (Eq32F x y) -> (MOVDEQ (MOVDconst [0]) (MOVDconst [1]) (FCMPS x y))
   307  
   308  (Neq64  x y) -> (MOVDNE (MOVDconst [0]) (MOVDconst [1]) (CMP x y))
   309  (Neq32  x y) -> (MOVDNE (MOVDconst [0]) (MOVDconst [1]) (CMPW x y))
   310  (Neq16  x y) -> (MOVDNE (MOVDconst [0]) (MOVDconst [1]) (CMP (MOVHreg x) (MOVHreg y)))
   311  (Neq8   x y) -> (MOVDNE (MOVDconst [0]) (MOVDconst [1]) (CMP (MOVBreg x) (MOVBreg y)))
   312  (NeqB   x y) -> (MOVDNE (MOVDconst [0]) (MOVDconst [1]) (CMP (MOVBreg x) (MOVBreg y)))
   313  (NeqPtr x y) -> (MOVDNE (MOVDconst [0]) (MOVDconst [1]) (CMP x y))
   314  (Neq64F x y) -> (MOVDNE (MOVDconst [0]) (MOVDconst [1]) (FCMP x y))
   315  (Neq32F x y) -> (MOVDNE (MOVDconst [0]) (MOVDconst [1]) (FCMPS x y))
   316  
   317  // Lowering loads
   318  (Load <t> ptr mem) && (is64BitInt(t) || isPtr(t)) -> (MOVDload ptr mem)
   319  (Load <t> ptr mem) && is32BitInt(t) && isSigned(t) -> (MOVWload ptr mem)
   320  (Load <t> ptr mem) && is32BitInt(t) && !isSigned(t) -> (MOVWZload ptr mem)
   321  (Load <t> ptr mem) && is16BitInt(t) && isSigned(t) -> (MOVHload ptr mem)
   322  (Load <t> ptr mem) && is16BitInt(t) && !isSigned(t) -> (MOVHZload ptr mem)
   323  (Load <t> ptr mem) && is8BitInt(t) && isSigned(t) -> (MOVBload ptr mem)
   324  (Load <t> ptr mem) && (t.IsBoolean() || (is8BitInt(t) && !isSigned(t))) -> (MOVBZload ptr mem)
   325  (Load <t> ptr mem) && is32BitFloat(t) -> (FMOVSload ptr mem)
   326  (Load <t> ptr mem) && is64BitFloat(t) -> (FMOVDload ptr mem)
   327  
   328  // Lowering stores
   329  // These more-specific FP versions of Store pattern should come first.
   330  (Store {t} ptr val mem) && t.(*types.Type).Size() == 8 && is64BitFloat(val.Type) -> (FMOVDstore ptr val mem)
   331  (Store {t} ptr val mem) && t.(*types.Type).Size() == 4 && is32BitFloat(val.Type) -> (FMOVSstore ptr val mem)
   332  
   333  (Store {t} ptr val mem) && t.(*types.Type).Size() == 8 -> (MOVDstore ptr val mem)
   334  (Store {t} ptr val mem) && t.(*types.Type).Size() == 4 -> (MOVWstore ptr val mem)
   335  (Store {t} ptr val mem) && t.(*types.Type).Size() == 2 -> (MOVHstore ptr val mem)
   336  (Store {t} ptr val mem) && t.(*types.Type).Size() == 1 -> (MOVBstore ptr val mem)
   337  
   338  // Lowering moves
   339  
   340  // Load and store for small copies.
   341  (Move [0] _ _ mem) -> mem
   342  (Move [1] dst src mem) -> (MOVBstore dst (MOVBZload src mem) mem)
   343  (Move [2] dst src mem) -> (MOVHstore dst (MOVHZload src mem) mem)
   344  (Move [4] dst src mem) -> (MOVWstore dst (MOVWZload src mem) mem)
   345  (Move [8] dst src mem) -> (MOVDstore dst (MOVDload src mem) mem)
   346  (Move [16] dst src mem) ->
   347  	(MOVDstore [8] dst (MOVDload [8] src mem)
   348  		(MOVDstore dst (MOVDload src mem) mem))
   349  (Move [24] dst src mem) ->
   350          (MOVDstore [16] dst (MOVDload [16] src mem)
   351  	        (MOVDstore [8] dst (MOVDload [8] src mem)
   352                  (MOVDstore dst (MOVDload src mem) mem)))
   353  (Move [3] dst src mem) ->
   354  	(MOVBstore [2] dst (MOVBZload [2] src mem)
   355  		(MOVHstore dst (MOVHZload src mem) mem))
   356  (Move [5] dst src mem) ->
   357  	(MOVBstore [4] dst (MOVBZload [4] src mem)
   358  		(MOVWstore dst (MOVWZload src mem) mem))
   359  (Move [6] dst src mem) ->
   360  	(MOVHstore [4] dst (MOVHZload [4] src mem)
   361  		(MOVWstore dst (MOVWZload src mem) mem))
   362  (Move [7] dst src mem) ->
   363  	(MOVBstore [6] dst (MOVBZload [6] src mem)
   364  		(MOVHstore [4] dst (MOVHZload [4] src mem)
   365  			(MOVWstore dst (MOVWZload src mem) mem)))
   366  
   367  // MVC for other moves. Use up to 4 instructions (sizes up to 1024 bytes).
   368  (Move [s] dst src mem) && s > 0 && s <= 256 ->
   369  	(MVC [makeValAndOff(s, 0)] dst src mem)
   370  (Move [s] dst src mem) && s > 256 && s <= 512 ->
   371  	(MVC [makeValAndOff(s-256, 256)] dst src (MVC [makeValAndOff(256, 0)] dst src mem))
   372  (Move [s] dst src mem) && s > 512 && s <= 768 ->
   373  	(MVC [makeValAndOff(s-512, 512)] dst src (MVC [makeValAndOff(256, 256)] dst src (MVC [makeValAndOff(256, 0)] dst src mem)))
   374  (Move [s] dst src mem) && s > 768 && s <= 1024 ->
   375  	(MVC [makeValAndOff(s-768, 768)] dst src (MVC [makeValAndOff(256, 512)] dst src (MVC [makeValAndOff(256, 256)] dst src (MVC [makeValAndOff(256, 0)] dst src mem))))
   376  
   377  // Move more than 1024 bytes using a loop.
   378  (Move [s] dst src mem) && s > 1024 ->
   379  	(LoweredMove [s%256] dst src (ADDconst <src.Type> src [(s/256)*256]) mem)
   380  
   381  // Lowering Zero instructions
   382  (Zero [0] _ mem) -> mem
   383  (Zero [1] destptr mem) -> (MOVBstoreconst [0] destptr mem)
   384  (Zero [2] destptr mem) -> (MOVHstoreconst [0] destptr mem)
   385  (Zero [4] destptr mem) -> (MOVWstoreconst [0] destptr mem)
   386  (Zero [8] destptr mem) -> (MOVDstoreconst [0] destptr mem)
   387  (Zero [3] destptr mem) ->
   388  	(MOVBstoreconst [makeValAndOff(0,2)] destptr
   389  		(MOVHstoreconst [0] destptr mem))
   390  (Zero [5] destptr mem) ->
   391  	(MOVBstoreconst [makeValAndOff(0,4)] destptr
   392  		(MOVWstoreconst [0] destptr mem))
   393  (Zero [6] destptr mem) ->
   394  	(MOVHstoreconst [makeValAndOff(0,4)] destptr
   395  		(MOVWstoreconst [0] destptr mem))
   396  (Zero [7] destptr mem) ->
   397  	(MOVWstoreconst [makeValAndOff(0,3)] destptr
   398  		(MOVWstoreconst [0] destptr mem))
   399  
   400  (Zero [s] destptr mem) && s > 0 && s <= 1024 ->
   401  	(CLEAR [makeValAndOff(s, 0)] destptr mem)
   402  
   403  // Move more than 1024 bytes using a loop.
   404  (Zero [s] destptr mem) && s > 1024 ->
   405  	(LoweredZero [s%256] destptr (ADDconst <destptr.Type> destptr [(s/256)*256]) mem)
   406  
   407  // Lowering constants
   408  (Const8   [val]) -> (MOVDconst [val])
   409  (Const16  [val]) -> (MOVDconst [val])
   410  (Const32  [val]) -> (MOVDconst [val])
   411  (Const64  [val]) -> (MOVDconst [val])
   412  (Const32F [val]) -> (FMOVSconst [val])
   413  (Const64F [val]) -> (FMOVDconst [val])
   414  (ConstNil) -> (MOVDconst [0])
   415  (ConstBool [b]) -> (MOVDconst [b])
   416  
   417  // Lowering calls
   418  (StaticCall [argwid] {target} mem) -> (CALLstatic [argwid] {target} mem)
   419  (ClosureCall [argwid] entry closure mem) -> (CALLclosure [argwid] entry closure mem)
   420  (InterCall [argwid] entry mem) -> (CALLinter [argwid] entry mem)
   421  
   422  // Miscellaneous
   423  (Convert <t> x mem) -> (MOVDconvert <t> x mem)
   424  (IsNonNil p) -> (MOVDNE (MOVDconst [0]) (MOVDconst [1]) (CMPconst p [0]))
   425  (IsInBounds idx len) -> (MOVDLT (MOVDconst [0]) (MOVDconst [1]) (CMPU idx len))
   426  (IsSliceInBounds idx len) -> (MOVDLE (MOVDconst [0]) (MOVDconst [1]) (CMPU idx len))
   427  (NilCheck ptr mem) -> (LoweredNilCheck ptr mem)
   428  (GetG mem) -> (LoweredGetG mem)
   429  (GetClosurePtr) -> (LoweredGetClosurePtr)
   430  (GetCallerSP) -> (LoweredGetCallerSP)
   431  (Addr {sym} base) -> (MOVDaddr {sym} base)
   432  (ITab (Load ptr mem)) -> (MOVDload ptr mem)
   433  
   434  // block rewrites
   435  (If (MOVDLT (MOVDconst [0]) (MOVDconst [1]) cmp) yes no) -> (LT cmp yes no)
   436  (If (MOVDLE (MOVDconst [0]) (MOVDconst [1]) cmp) yes no) -> (LE cmp yes no)
   437  (If (MOVDGT (MOVDconst [0]) (MOVDconst [1]) cmp) yes no) -> (GT cmp yes no)
   438  (If (MOVDGE (MOVDconst [0]) (MOVDconst [1]) cmp) yes no) -> (GE cmp yes no)
   439  (If (MOVDEQ (MOVDconst [0]) (MOVDconst [1]) cmp) yes no) -> (EQ cmp yes no)
   440  (If (MOVDNE (MOVDconst [0]) (MOVDconst [1]) cmp) yes no) -> (NE cmp yes no)
   441  
   442  // Special case for floating point - LF/LEF not generated.
   443  (If (MOVDGTnoinv (MOVDconst [0]) (MOVDconst [1]) cmp) yes no) -> (GTF cmp yes no)
   444  (If (MOVDGEnoinv (MOVDconst [0]) (MOVDconst [1]) cmp) yes no) -> (GEF cmp yes no)
   445  
   446  (If cond yes no) -> (NE (CMPWconst [0] (MOVBZreg <typ.Bool> cond)) yes no)
   447  
   448  // ***************************
   449  // Above: lowering rules
   450  // Below: optimizations
   451  // ***************************
   452  // TODO: Should the optimizations be a separate pass?
   453  
   454  // Fold unnecessary type conversions.
   455  (MOVDreg <t> x) && t.Compare(x.Type) == types.CMPeq -> x
   456  (MOVDnop <t> x) && t.Compare(x.Type) == types.CMPeq -> x
   457  
   458  // Propagate constants through type conversions.
   459  (MOVDreg (MOVDconst [c])) -> (MOVDconst [c])
   460  (MOVDnop (MOVDconst [c])) -> (MOVDconst [c])
   461  
   462  // If a register move has only 1 use, just use the same register without emitting instruction.
   463  // MOVDnop doesn't emit instruction, only for ensuring the type.
   464  (MOVDreg x) && x.Uses == 1 -> (MOVDnop x)
   465  
   466  // Fold type changes into loads.
   467  (MOVDreg <t> x:(MOVBZload [off] {sym} ptr mem)) && x.Uses == 1 && clobber(x) -> @x.Block (MOVBZload <t> [off] {sym} ptr mem)
   468  (MOVDreg <t> x:(MOVBload  [off] {sym} ptr mem)) && x.Uses == 1 && clobber(x) -> @x.Block (MOVBload  <t> [off] {sym} ptr mem)
   469  (MOVDreg <t> x:(MOVHZload [off] {sym} ptr mem)) && x.Uses == 1 && clobber(x) -> @x.Block (MOVHZload <t> [off] {sym} ptr mem)
   470  (MOVDreg <t> x:(MOVHload  [off] {sym} ptr mem)) && x.Uses == 1 && clobber(x) -> @x.Block (MOVHload  <t> [off] {sym} ptr mem)
   471  (MOVDreg <t> x:(MOVWZload [off] {sym} ptr mem)) && x.Uses == 1 && clobber(x) -> @x.Block (MOVWZload <t> [off] {sym} ptr mem)
   472  (MOVDreg <t> x:(MOVWload  [off] {sym} ptr mem)) && x.Uses == 1 && clobber(x) -> @x.Block (MOVWload  <t> [off] {sym} ptr mem)
   473  (MOVDreg <t> x:(MOVDload  [off] {sym} ptr mem)) && x.Uses == 1 && clobber(x) -> @x.Block (MOVDload  <t> [off] {sym} ptr mem)
   474  
   475  (MOVDnop <t> x:(MOVBZload [off] {sym} ptr mem)) && x.Uses == 1 && clobber(x) -> @x.Block (MOVBZload <t> [off] {sym} ptr mem)
   476  (MOVDnop <t> x:(MOVBload  [off] {sym} ptr mem)) && x.Uses == 1 && clobber(x) -> @x.Block (MOVBload  <t> [off] {sym} ptr mem)
   477  (MOVDnop <t> x:(MOVHZload [off] {sym} ptr mem)) && x.Uses == 1 && clobber(x) -> @x.Block (MOVHZload <t> [off] {sym} ptr mem)
   478  (MOVDnop <t> x:(MOVHload  [off] {sym} ptr mem)) && x.Uses == 1 && clobber(x) -> @x.Block (MOVHload  <t> [off] {sym} ptr mem)
   479  (MOVDnop <t> x:(MOVWZload [off] {sym} ptr mem)) && x.Uses == 1 && clobber(x) -> @x.Block (MOVWZload <t> [off] {sym} ptr mem)
   480  (MOVDnop <t> x:(MOVWload  [off] {sym} ptr mem)) && x.Uses == 1 && clobber(x) -> @x.Block (MOVWload  <t> [off] {sym} ptr mem)
   481  (MOVDnop <t> x:(MOVDload  [off] {sym} ptr mem)) && x.Uses == 1 && clobber(x) -> @x.Block (MOVDload  <t> [off] {sym} ptr mem)
   482  
   483  // TODO(mundaym): uncomment rules once signed indexed loads are added.
   484  (MOVDreg <t> x:(MOVBZloadidx [off] {sym} ptr idx mem)) && x.Uses == 1 && clobber(x) -> @x.Block (MOVBZloadidx <t> [off] {sym} ptr idx mem)
   485  //(MOVDreg <t> x:(MOVBloadidx  [off] {sym} ptr idx mem)) && x.Uses == 1 && clobber(x) -> @x.Block (MOVBloadidx  <t> [off] {sym} ptr idx mem)
   486  (MOVDreg <t> x:(MOVHZloadidx [off] {sym} ptr idx mem)) && x.Uses == 1 && clobber(x) -> @x.Block (MOVHZloadidx <t> [off] {sym} ptr idx mem)
   487  //(MOVDreg <t> x:(MOVHloadidx  [off] {sym} ptr idx mem)) && x.Uses == 1 && clobber(x) -> @x.Block (MOVHloadidx  <t> [off] {sym} ptr idx mem)
   488  (MOVDreg <t> x:(MOVWZloadidx [off] {sym} ptr idx mem)) && x.Uses == 1 && clobber(x) -> @x.Block (MOVWZloadidx <t> [off] {sym} ptr idx mem)
   489  //(MOVDreg <t> x:(MOVWloadidx  [off] {sym} ptr idx mem)) && x.Uses == 1 && clobber(x) -> @x.Block (MOVWloadidx  <t> [off] {sym} ptr idx mem)
   490  (MOVDreg <t> x:(MOVDloadidx  [off] {sym} ptr idx mem)) && x.Uses == 1 && clobber(x) -> @x.Block (MOVDloadidx  <t> [off] {sym} ptr idx mem)
   491  
   492  (MOVDnop <t> x:(MOVBZloadidx [off] {sym} ptr idx mem)) && x.Uses == 1 && clobber(x) -> @x.Block (MOVBZloadidx <t> [off] {sym} ptr idx mem)
   493  //(MOVDnop <t> x:(MOVBloadidx  [off] {sym} ptr idx mem)) && x.Uses == 1 && clobber(x) -> @x.Block (MOVBloadidx  <t> [off] {sym} ptr idx mem)
   494  (MOVDnop <t> x:(MOVHZloadidx [off] {sym} ptr idx mem)) && x.Uses == 1 && clobber(x) -> @x.Block (MOVHZloadidx <t> [off] {sym} ptr idx mem)
   495  //(MOVDnop <t> x:(MOVHloadidx  [off] {sym} ptr idx mem)) && x.Uses == 1 && clobber(x) -> @x.Block (MOVHloadidx  <t> [off] {sym} ptr idx mem)
   496  (MOVDnop <t> x:(MOVWZloadidx [off] {sym} ptr idx mem)) && x.Uses == 1 && clobber(x) -> @x.Block (MOVWZloadidx <t> [off] {sym} ptr idx mem)
   497  //(MOVDnop <t> x:(MOVWloadidx  [off] {sym} ptr idx mem)) && x.Uses == 1 && clobber(x) -> @x.Block (MOVWloadidx  <t> [off] {sym} ptr idx mem)
   498  (MOVDnop <t> x:(MOVDloadidx  [off] {sym} ptr idx mem)) && x.Uses == 1 && clobber(x) -> @x.Block (MOVDloadidx  <t> [off] {sym} ptr idx mem)
   499  
   500  // Fold sign extensions into conditional moves of constants.
   501  // Designed to remove the MOVBZreg inserted by the If lowering.
   502  (MOVBZreg x:(MOVDLT (MOVDconst [c]) (MOVDconst [d]) _)) && int64(uint8(c)) == c && int64(uint8(d)) == d -> (MOVDreg x)
   503  (MOVBZreg x:(MOVDLE (MOVDconst [c]) (MOVDconst [d]) _)) && int64(uint8(c)) == c && int64(uint8(d)) == d -> (MOVDreg x)
   504  (MOVBZreg x:(MOVDGT (MOVDconst [c]) (MOVDconst [d]) _)) && int64(uint8(c)) == c && int64(uint8(d)) == d -> (MOVDreg x)
   505  (MOVBZreg x:(MOVDGE (MOVDconst [c]) (MOVDconst [d]) _)) && int64(uint8(c)) == c && int64(uint8(d)) == d -> (MOVDreg x)
   506  (MOVBZreg x:(MOVDEQ (MOVDconst [c]) (MOVDconst [d]) _)) && int64(uint8(c)) == c && int64(uint8(d)) == d -> (MOVDreg x)
   507  (MOVBZreg x:(MOVDNE (MOVDconst [c]) (MOVDconst [d]) _)) && int64(uint8(c)) == c && int64(uint8(d)) == d -> (MOVDreg x)
   508  (MOVBZreg x:(MOVDGTnoinv (MOVDconst [c]) (MOVDconst [d]) _)) && int64(uint8(c)) == c && int64(uint8(d)) == d -> (MOVDreg x)
   509  (MOVBZreg x:(MOVDGEnoinv (MOVDconst [c]) (MOVDconst [d]) _)) && int64(uint8(c)) == c && int64(uint8(d)) == d -> (MOVDreg x)
   510  
   511  // Fold boolean tests into blocks.
   512  (NE (CMPWconst [0] (MOVDLT (MOVDconst [0]) (MOVDconst [1]) cmp)) yes no) -> (LT cmp yes no)
   513  (NE (CMPWconst [0] (MOVDLE (MOVDconst [0]) (MOVDconst [1]) cmp)) yes no) -> (LE cmp yes no)
   514  (NE (CMPWconst [0] (MOVDGT (MOVDconst [0]) (MOVDconst [1]) cmp)) yes no) -> (GT cmp yes no)
   515  (NE (CMPWconst [0] (MOVDGE (MOVDconst [0]) (MOVDconst [1]) cmp)) yes no) -> (GE cmp yes no)
   516  (NE (CMPWconst [0] (MOVDEQ (MOVDconst [0]) (MOVDconst [1]) cmp)) yes no) -> (EQ cmp yes no)
   517  (NE (CMPWconst [0] (MOVDNE (MOVDconst [0]) (MOVDconst [1]) cmp)) yes no) -> (NE cmp yes no)
   518  (NE (CMPWconst [0] (MOVDGTnoinv (MOVDconst [0]) (MOVDconst [1]) cmp)) yes no) -> (GTF cmp yes no)
   519  (NE (CMPWconst [0] (MOVDGEnoinv (MOVDconst [0]) (MOVDconst [1]) cmp)) yes no) -> (GEF cmp yes no)
   520  
   521  // Fold constants into instructions.
   522  (ADD x (MOVDconst [c])) && is32Bit(c) -> (ADDconst [c] x)
   523  (ADDW x (MOVDconst [c])) -> (ADDWconst [c] x)
   524  
   525  (SUB x (MOVDconst [c])) && is32Bit(c) -> (SUBconst x [c])
   526  (SUB (MOVDconst [c]) x) && is32Bit(c) -> (NEG (SUBconst <v.Type> x [c]))
   527  (SUBW x (MOVDconst [c])) -> (SUBWconst x [c])
   528  (SUBW (MOVDconst [c]) x) -> (NEGW (SUBWconst <v.Type> x [c]))
   529  
   530  (MULLD x (MOVDconst [c])) && is32Bit(c) -> (MULLDconst [c] x)
   531  (MULLW x (MOVDconst [c])) -> (MULLWconst [c] x)
   532  
   533  // NILF instructions leave the high 32 bits unchanged which is
   534  // equivalent to the leftmost 32 bits being set.
   535  // TODO(mundaym): modify the assembler to accept 64-bit values
   536  // and use isU32Bit(^c).
   537  (AND x (MOVDconst [c])) && is32Bit(c) && c < 0 -> (ANDconst [c] x)
   538  (ANDW x (MOVDconst [c])) -> (ANDWconst [c] x)
   539  
   540  (ANDWconst [c] (ANDWconst [d] x)) -> (ANDWconst [c & d] x)
   541  (ANDconst [c] (ANDconst [d] x)) -> (ANDconst [c & d] x)
   542  
   543  (OR x (MOVDconst [c])) && isU32Bit(c) -> (ORconst [c] x)
   544  (ORW x (MOVDconst [c])) -> (ORWconst [c] x)
   545  
   546  (XOR x (MOVDconst [c])) && isU32Bit(c) -> (XORconst [c] x)
   547  (XORW x (MOVDconst [c])) -> (XORWconst [c] x)
   548  
   549  (SLD x (MOVDconst [c])) -> (SLDconst [c&63] x)
   550  (SLW x (MOVDconst [c])) -> (SLWconst [c&63] x)
   551  (SRD x (MOVDconst [c])) -> (SRDconst [c&63] x)
   552  (SRW x (MOVDconst [c])) -> (SRWconst [c&63] x)
   553  (SRAD x (MOVDconst [c])) -> (SRADconst [c&63] x)
   554  (SRAW x (MOVDconst [c])) -> (SRAWconst [c&63] x)
   555  
   556  (SRAW x (ANDWconst [63] y)) -> (SRAW x y)
   557  (SRAD x (ANDconst [63] y)) -> (SRAD x y)
   558  (SLW x (ANDWconst [63] y)) -> (SLW x y)
   559  (SLD x (ANDconst [63] y)) -> (SLD x y)
   560  (SRW x (ANDWconst [63] y)) -> (SRW x y)
   561  (SRD x (ANDconst [63] y)) -> (SRD x y)
   562  
   563  // Rotate generation
   564  (ADD (SLDconst x [c]) (SRDconst x [d])) && d == 64-c -> (RLLGconst [c] x)
   565  ( OR (SLDconst x [c]) (SRDconst x [d])) && d == 64-c -> (RLLGconst [c] x)
   566  (XOR (SLDconst x [c]) (SRDconst x [d])) && d == 64-c -> (RLLGconst [c] x)
   567  
   568  (ADDW (SLWconst x [c]) (SRWconst x [d])) && d == 32-c -> (RLLconst [c] x)
   569  ( ORW (SLWconst x [c]) (SRWconst x [d])) && d == 32-c -> (RLLconst [c] x)
   570  (XORW (SLWconst x [c]) (SRWconst x [d])) && d == 32-c -> (RLLconst [c] x)
   571  
   572  (CMP x (MOVDconst [c])) && is32Bit(c) -> (CMPconst x [c])
   573  (CMP (MOVDconst [c]) x) && is32Bit(c) -> (InvertFlags (CMPconst x [c]))
   574  (CMPW x (MOVDconst [c])) -> (CMPWconst x [c])
   575  (CMPW (MOVDconst [c]) x) -> (InvertFlags (CMPWconst x [c]))
   576  (CMPU x (MOVDconst [c])) && isU32Bit(c) -> (CMPUconst x [int64(uint32(c))])
   577  (CMPU (MOVDconst [c]) x) && isU32Bit(c) -> (InvertFlags (CMPUconst x [int64(uint32(c))]))
   578  (CMPWU x (MOVDconst [c])) -> (CMPWUconst x [int64(uint32(c))])
   579  (CMPWU (MOVDconst [c]) x) -> (InvertFlags (CMPWUconst x [int64(uint32(c))]))
   580  
   581  // Using MOV{W,H,B}Zreg instead of AND is cheaper.
   582  (AND x (MOVDconst [0xFF])) -> (MOVBZreg x)
   583  (AND x (MOVDconst [0xFFFF])) -> (MOVHZreg x)
   584  (AND x (MOVDconst [0xFFFFFFFF])) -> (MOVWZreg x)
   585  (ANDWconst [0xFF] x) -> (MOVBZreg x)
   586  (ANDWconst [0xFFFF] x) -> (MOVHZreg x)
   587  
   588  // strength reduction
   589  (MULLDconst [-1] x) -> (NEG x)
   590  (MULLDconst [0] _) -> (MOVDconst [0])
   591  (MULLDconst [1] x) -> x
   592  (MULLDconst [c] x) && isPowerOfTwo(c) -> (SLDconst [log2(c)] x)
   593  (MULLDconst [c] x) && isPowerOfTwo(c+1) && c >= 15 -> (SUB (SLDconst <v.Type> [log2(c+1)] x) x)
   594  (MULLDconst [c] x) && isPowerOfTwo(c-1) && c >= 17 -> (ADD (SLDconst <v.Type> [log2(c-1)] x) x)
   595  
   596  (MULLWconst [-1] x) -> (NEGW x)
   597  (MULLWconst [0] _) -> (MOVDconst [0])
   598  (MULLWconst [1] x) -> x
   599  (MULLWconst [c] x) && isPowerOfTwo(c) -> (SLWconst [log2(c)] x)
   600  (MULLWconst [c] x) && isPowerOfTwo(c+1) && c >= 15 -> (SUBW (SLWconst <v.Type> [log2(c+1)] x) x)
   601  (MULLWconst [c] x) && isPowerOfTwo(c-1) && c >= 17 -> (ADDW (SLWconst <v.Type> [log2(c-1)] x) x)
   602  
   603  // Fold ADD into MOVDaddr. Odd offsets from SB shouldn't be folded (LARL can't handle them).
   604  (ADDconst [c] (MOVDaddr [d] {s} x:(SB))) && ((c+d)&1 == 0) && is32Bit(c+d) -> (MOVDaddr [c+d] {s} x)
   605  (ADDconst [c] (MOVDaddr [d] {s} x)) && x.Op != OpSB && is20Bit(c+d) -> (MOVDaddr [c+d] {s} x)
   606  (ADD idx (MOVDaddr [c] {s} ptr)) && ptr.Op != OpSB && idx.Op != OpSB -> (MOVDaddridx [c] {s} ptr idx)
   607  
   608  // fold ADDconst into MOVDaddrx
   609  (ADDconst [c] (MOVDaddridx [d] {s} x y)) && is20Bit(c+d) -> (MOVDaddridx [c+d] {s} x y)
   610  (MOVDaddridx [c] {s} (ADDconst [d] x) y) && is20Bit(c+d) && x.Op != OpSB -> (MOVDaddridx [c+d] {s} x y)
   611  (MOVDaddridx [c] {s} x (ADDconst [d] y)) && is20Bit(c+d) && y.Op != OpSB -> (MOVDaddridx [c+d] {s} x y)
   612  
   613  // reverse ordering of compare instruction
   614  (MOVDLT x y (InvertFlags cmp)) -> (MOVDGT x y cmp)
   615  (MOVDGT x y (InvertFlags cmp)) -> (MOVDLT x y cmp)
   616  (MOVDLE x y (InvertFlags cmp)) -> (MOVDGE x y cmp)
   617  (MOVDGE x y (InvertFlags cmp)) -> (MOVDLE x y cmp)
   618  (MOVDEQ x y (InvertFlags cmp)) -> (MOVDEQ x y cmp)
   619  (MOVDNE x y (InvertFlags cmp)) -> (MOVDNE x y cmp)
   620  
   621  // don't extend after proper load
   622  (MOVBreg x:(MOVBload _ _)) -> (MOVDreg x)
   623  (MOVBZreg x:(MOVBZload _ _)) -> (MOVDreg x)
   624  (MOVHreg x:(MOVBload _ _)) -> (MOVDreg x)
   625  (MOVHreg x:(MOVBZload _ _)) -> (MOVDreg x)
   626  (MOVHreg x:(MOVHload _ _)) -> (MOVDreg x)
   627  (MOVHZreg x:(MOVBZload _ _)) -> (MOVDreg x)
   628  (MOVHZreg x:(MOVHZload _ _)) -> (MOVDreg x)
   629  (MOVWreg x:(MOVBload _ _)) -> (MOVDreg x)
   630  (MOVWreg x:(MOVBZload _ _)) -> (MOVDreg x)
   631  (MOVWreg x:(MOVHload _ _)) -> (MOVDreg x)
   632  (MOVWreg x:(MOVHZload _ _)) -> (MOVDreg x)
   633  (MOVWreg x:(MOVWload _ _)) -> (MOVDreg x)
   634  (MOVWZreg x:(MOVBZload _ _)) -> (MOVDreg x)
   635  (MOVWZreg x:(MOVHZload _ _)) -> (MOVDreg x)
   636  (MOVWZreg x:(MOVWZload _ _)) -> (MOVDreg x)
   637  
   638  // don't extend if argument is already extended
   639  (MOVBreg x:(Arg <t>)) && is8BitInt(t) && isSigned(t) -> (MOVDreg x)
   640  (MOVBZreg x:(Arg <t>)) && is8BitInt(t) && !isSigned(t) -> (MOVDreg x)
   641  (MOVHreg x:(Arg <t>)) && (is8BitInt(t) || is16BitInt(t)) && isSigned(t) -> (MOVDreg x)
   642  (MOVHZreg x:(Arg <t>)) && (is8BitInt(t) || is16BitInt(t)) && !isSigned(t) -> (MOVDreg x)
   643  (MOVWreg x:(Arg <t>)) && (is8BitInt(t) || is16BitInt(t) || is32BitInt(t)) && isSigned(t) -> (MOVDreg x)
   644  (MOVWZreg x:(Arg <t>)) && (is8BitInt(t) || is16BitInt(t) || is32BitInt(t)) && !isSigned(t) -> (MOVDreg x)
   645  
   646  // fold double extensions
   647  (MOVBreg x:(MOVBreg _)) -> (MOVDreg x)
   648  (MOVBZreg x:(MOVBZreg _)) -> (MOVDreg x)
   649  (MOVHreg x:(MOVBreg _)) -> (MOVDreg x)
   650  (MOVHreg x:(MOVBZreg _)) -> (MOVDreg x)
   651  (MOVHreg x:(MOVHreg _)) -> (MOVDreg x)
   652  (MOVHZreg x:(MOVBZreg _)) -> (MOVDreg x)
   653  (MOVHZreg x:(MOVHZreg _)) -> (MOVDreg x)
   654  (MOVWreg x:(MOVBreg _)) -> (MOVDreg x)
   655  (MOVWreg x:(MOVBZreg _)) -> (MOVDreg x)
   656  (MOVWreg x:(MOVHreg _)) -> (MOVDreg x)
   657  (MOVWreg x:(MOVHreg _)) -> (MOVDreg x)
   658  (MOVWreg x:(MOVWreg _)) -> (MOVDreg x)
   659  (MOVWZreg x:(MOVBZreg _)) -> (MOVDreg x)
   660  (MOVWZreg x:(MOVHZreg _)) -> (MOVDreg x)
   661  (MOVWZreg x:(MOVWZreg _)) -> (MOVDreg x)
   662  
   663  // fold extensions into constants
   664  (MOVBreg (MOVDconst [c])) -> (MOVDconst [int64(int8(c))])
   665  (MOVBZreg (MOVDconst [c])) -> (MOVDconst [int64(uint8(c))])
   666  (MOVHreg (MOVDconst [c])) -> (MOVDconst [int64(int16(c))])
   667  (MOVHZreg (MOVDconst [c])) -> (MOVDconst [int64(uint16(c))])
   668  (MOVWreg (MOVDconst [c])) -> (MOVDconst [int64(int32(c))])
   669  (MOVWZreg (MOVDconst [c])) -> (MOVDconst [int64(uint32(c))])
   670  
   671  // sign extended loads
   672  // Note: The combined instruction must end up in the same block
   673  // as the original load. If not, we end up making a value with
   674  // memory type live in two different blocks, which can lead to
   675  // multiple memory values alive simultaneously.
   676  // Make sure we don't combine these ops if the load has another use.
   677  // This prevents a single load from being split into multiple loads
   678  // which then might return different values.  See test/atomicload.go.
   679  (MOVBreg x:(MOVBZload [off] {sym} ptr mem)) && x.Uses == 1 && clobber(x) -> @x.Block (MOVBload <v.Type> [off] {sym} ptr mem)
   680  (MOVBZreg x:(MOVBZload [off] {sym} ptr mem)) && x.Uses == 1 && clobber(x) -> @x.Block (MOVBZload <v.Type> [off] {sym} ptr mem)
   681  (MOVHreg x:(MOVHZload [off] {sym} ptr mem)) && x.Uses == 1 && clobber(x) -> @x.Block (MOVHload <v.Type> [off] {sym} ptr mem)
   682  (MOVHZreg x:(MOVHZload [off] {sym} ptr mem)) && x.Uses == 1 && clobber(x) -> @x.Block (MOVHZload <v.Type> [off] {sym} ptr mem)
   683  (MOVWreg x:(MOVWZload [off] {sym} ptr mem)) && x.Uses == 1 && clobber(x) -> @x.Block (MOVWload <v.Type> [off] {sym} ptr mem)
   684  (MOVWZreg x:(MOVWZload [off] {sym} ptr mem)) && x.Uses == 1 && clobber(x) -> @x.Block (MOVWZload <v.Type> [off] {sym} ptr mem)
   685  
   686  (MOVBZreg x:(MOVBZloadidx [off] {sym} ptr idx mem)) && x.Uses == 1 && clobber(x) -> @x.Block (MOVBZloadidx <v.Type> [off] {sym} ptr idx mem)
   687  (MOVHZreg x:(MOVHZloadidx [off] {sym} ptr idx mem)) && x.Uses == 1 && clobber(x) -> @x.Block (MOVHZloadidx <v.Type> [off] {sym} ptr idx mem)
   688  (MOVWZreg x:(MOVWZloadidx [off] {sym} ptr idx mem)) && x.Uses == 1 && clobber(x) -> @x.Block (MOVWZloadidx <v.Type> [off] {sym} ptr idx mem)
   689  
   690  // replace load from same location as preceding store with copy
   691  (MOVBZload [off] {sym} ptr (MOVBstore [off2] {sym2} ptr2 x _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) -> (MOVBZreg x)
   692  (MOVHZload [off] {sym} ptr (MOVHstore [off2] {sym2} ptr2 x _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) -> (MOVHZreg x)
   693  (MOVWZload [off] {sym} ptr (MOVWstore [off2] {sym2} ptr2 x _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) -> (MOVWZreg x)
   694  (MOVDload [off] {sym} ptr (MOVDstore [off2] {sym2} ptr2 x _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) -> (MOVDreg x)
   695  
   696  // Don't extend before storing
   697  (MOVWstore [off] {sym} ptr (MOVWreg x) mem) -> (MOVWstore [off] {sym} ptr x mem)
   698  (MOVHstore [off] {sym} ptr (MOVHreg x) mem) -> (MOVHstore [off] {sym} ptr x mem)
   699  (MOVBstore [off] {sym} ptr (MOVBreg x) mem) -> (MOVBstore [off] {sym} ptr x mem)
   700  (MOVWstore [off] {sym} ptr (MOVWZreg x) mem) -> (MOVWstore [off] {sym} ptr x mem)
   701  (MOVHstore [off] {sym} ptr (MOVHZreg x) mem) -> (MOVHstore [off] {sym} ptr x mem)
   702  (MOVBstore [off] {sym} ptr (MOVBZreg x) mem) -> (MOVBstore [off] {sym} ptr x mem)
   703  
   704  // Fold constants into memory operations.
   705  // Note that this is not always a good idea because if not all the uses of
   706  // the ADDconst get eliminated, we still have to compute the ADDconst and we now
   707  // have potentially two live values (ptr and (ADDconst [off] ptr)) instead of one.
   708  // Nevertheless, let's do it!
   709  (MOVDload   [off1] {sym} (ADDconst [off2] ptr) mem) && is20Bit(off1+off2) -> (MOVDload  [off1+off2] {sym} ptr mem)
   710  (MOVWload   [off1] {sym} (ADDconst [off2] ptr) mem) && is20Bit(off1+off2) -> (MOVWload  [off1+off2] {sym} ptr mem)
   711  (MOVHload   [off1] {sym} (ADDconst [off2] ptr) mem) && is20Bit(off1+off2) -> (MOVHload  [off1+off2] {sym} ptr mem)
   712  (MOVBload   [off1] {sym} (ADDconst [off2] ptr) mem) && is20Bit(off1+off2) -> (MOVBload  [off1+off2] {sym} ptr mem)
   713  (MOVWZload  [off1] {sym} (ADDconst [off2] ptr) mem) && is20Bit(off1+off2) -> (MOVWZload [off1+off2] {sym} ptr mem)
   714  (MOVHZload  [off1] {sym} (ADDconst [off2] ptr) mem) && is20Bit(off1+off2) -> (MOVHZload [off1+off2] {sym} ptr mem)
   715  (MOVBZload  [off1] {sym} (ADDconst [off2] ptr) mem) && is20Bit(off1+off2) -> (MOVBZload [off1+off2] {sym} ptr mem)
   716  (FMOVSload  [off1] {sym} (ADDconst [off2] ptr) mem) && is20Bit(off1+off2) -> (FMOVSload [off1+off2] {sym} ptr mem)
   717  (FMOVDload  [off1] {sym} (ADDconst [off2] ptr) mem) && is20Bit(off1+off2) -> (FMOVDload [off1+off2] {sym} ptr mem)
   718  
   719  (MOVDstore  [off1] {sym} (ADDconst [off2] ptr) val mem) && is20Bit(off1+off2) -> (MOVDstore  [off1+off2] {sym} ptr val mem)
   720  (MOVWstore  [off1] {sym} (ADDconst [off2] ptr) val mem) && is20Bit(off1+off2) -> (MOVWstore  [off1+off2] {sym} ptr val mem)
   721  (MOVHstore  [off1] {sym} (ADDconst [off2] ptr) val mem) && is20Bit(off1+off2) -> (MOVHstore  [off1+off2] {sym} ptr val mem)
   722  (MOVBstore  [off1] {sym} (ADDconst [off2] ptr) val mem) && is20Bit(off1+off2) -> (MOVBstore  [off1+off2] {sym} ptr val mem)
   723  (FMOVSstore [off1] {sym} (ADDconst [off2] ptr) val mem) && is20Bit(off1+off2) -> (FMOVSstore [off1+off2] {sym} ptr val mem)
   724  (FMOVDstore [off1] {sym} (ADDconst [off2] ptr) val mem) && is20Bit(off1+off2) -> (FMOVDstore [off1+off2] {sym} ptr val mem)
   725  
   726  // Fold constants into stores.
   727  (MOVDstore [off] {sym} ptr (MOVDconst [c]) mem) && is16Bit(c) && isU12Bit(off) && ptr.Op != OpSB ->
   728  	(MOVDstoreconst [makeValAndOff(c,off)] {sym} ptr mem)
   729  (MOVWstore [off] {sym} ptr (MOVDconst [c]) mem) && is16Bit(c) && isU12Bit(off) && ptr.Op != OpSB ->
   730  	(MOVWstoreconst [makeValAndOff(int64(int32(c)),off)] {sym} ptr mem)
   731  (MOVHstore [off] {sym} ptr (MOVDconst [c]) mem) && isU12Bit(off) && ptr.Op != OpSB ->
   732  	(MOVHstoreconst [makeValAndOff(int64(int16(c)),off)] {sym} ptr mem)
   733  (MOVBstore [off] {sym} ptr (MOVDconst [c]) mem) && is20Bit(off) && ptr.Op != OpSB ->
   734  	(MOVBstoreconst [makeValAndOff(int64(int8(c)),off)] {sym} ptr mem)
   735  
   736  // Fold address offsets into constant stores.
   737  (MOVDstoreconst [sc] {s} (ADDconst [off] ptr) mem) && isU12Bit(ValAndOff(sc).Off()+off) ->
   738  	(MOVDstoreconst [ValAndOff(sc).add(off)] {s} ptr mem)
   739  (MOVWstoreconst [sc] {s} (ADDconst [off] ptr) mem) && isU12Bit(ValAndOff(sc).Off()+off) ->
   740  	(MOVWstoreconst [ValAndOff(sc).add(off)] {s} ptr mem)
   741  (MOVHstoreconst [sc] {s} (ADDconst [off] ptr) mem) && isU12Bit(ValAndOff(sc).Off()+off) ->
   742  	(MOVHstoreconst [ValAndOff(sc).add(off)] {s} ptr mem)
   743  (MOVBstoreconst [sc] {s} (ADDconst [off] ptr) mem) && is20Bit(ValAndOff(sc).Off()+off) ->
   744  	(MOVBstoreconst [ValAndOff(sc).add(off)] {s} ptr mem)
   745  
   746  // Merge address calculations into loads and stores.
   747  // Offsets from SB must not be merged into unaligned memory accesses because
   748  // loads/stores using PC-relative addressing directly must be aligned to the
   749  // size of the target.
   750  (MOVDload   [off1] {sym1} (MOVDaddr <t> [off2] {sym2} base) mem) && is32Bit(off1+off2) && canMergeSym(sym1, sym2) && (base.Op != OpSB || (t.IsPtr() && t.ElemType().Alignment()%8 == 0 && (off1+off2)%8 == 0)) ->
   751  	(MOVDload  [off1+off2] {mergeSym(sym1,sym2)} base mem)
   752  (MOVWZload  [off1] {sym1} (MOVDaddr <t> [off2] {sym2} base) mem) && is32Bit(off1+off2) && canMergeSym(sym1, sym2) && (base.Op != OpSB || (t.IsPtr() && t.ElemType().Alignment()%4 == 0 && (off1+off2)%4 == 0)) ->
   753  	(MOVWZload  [off1+off2] {mergeSym(sym1,sym2)} base mem)
   754  (MOVHZload  [off1] {sym1} (MOVDaddr <t> [off2] {sym2} base) mem) && is32Bit(off1+off2) && canMergeSym(sym1, sym2) && (base.Op != OpSB || (t.IsPtr() && t.ElemType().Alignment()%2 == 0 && (off1+off2)%2 == 0)) ->
   755  	(MOVHZload  [off1+off2] {mergeSym(sym1,sym2)} base mem)
   756  (MOVBZload  [off1] {sym1} (MOVDaddr [off2] {sym2} base) mem) && is32Bit(off1+off2) && canMergeSym(sym1, sym2) ->
   757  	(MOVBZload  [off1+off2] {mergeSym(sym1,sym2)} base mem)
   758  (FMOVSload [off1] {sym1} (MOVDaddr [off2] {sym2} base) mem) && is32Bit(off1+off2) && canMergeSym(sym1, sym2) ->
   759  	(FMOVSload [off1+off2] {mergeSym(sym1,sym2)} base mem)
   760  (FMOVDload [off1] {sym1} (MOVDaddr [off2] {sym2} base) mem) && is32Bit(off1+off2) && canMergeSym(sym1, sym2) ->
   761  	(FMOVDload [off1+off2] {mergeSym(sym1,sym2)} base mem)
   762  
   763  (MOVWload [off1] {sym1} (MOVDaddr <t> [off2] {sym2} base) mem) && is32Bit(off1+off2) && canMergeSym(sym1, sym2) && (base.Op != OpSB || (t.IsPtr() && t.ElemType().Alignment()%4 == 0 && (off1+off2)%4 == 0)) ->
   764  	(MOVWload [off1+off2] {mergeSym(sym1,sym2)} base mem)
   765  (MOVHload [off1] {sym1} (MOVDaddr <t> [off2] {sym2} base) mem) && is32Bit(off1+off2) && canMergeSym(sym1, sym2) && (base.Op != OpSB || (t.IsPtr() && t.ElemType().Alignment()%2 == 0 && (off1+off2)%2 == 0)) ->
   766  	(MOVHload [off1+off2] {mergeSym(sym1,sym2)} base mem)
   767  (MOVBload [off1] {sym1} (MOVDaddr [off2] {sym2} base) mem) && is32Bit(off1+off2) && canMergeSym(sym1, sym2) ->
   768  	(MOVBload [off1+off2] {mergeSym(sym1,sym2)} base mem)
   769  
   770  (MOVDstore  [off1] {sym1} (MOVDaddr <t> [off2] {sym2} base) val mem) && is32Bit(off1+off2) && canMergeSym(sym1, sym2) && (base.Op != OpSB || (t.IsPtr() && t.ElemType().Alignment()%8 == 0 && (off1+off2)%8 == 0)) ->
   771  	(MOVDstore  [off1+off2] {mergeSym(sym1,sym2)} base val mem)
   772  (MOVWstore  [off1] {sym1} (MOVDaddr <t> [off2] {sym2} base) val mem) && is32Bit(off1+off2) && canMergeSym(sym1, sym2) && (base.Op != OpSB || (t.IsPtr() && t.ElemType().Alignment()%4 == 0 && (off1+off2)%4 == 0)) ->
   773  	(MOVWstore  [off1+off2] {mergeSym(sym1,sym2)} base val mem)
   774  (MOVHstore  [off1] {sym1} (MOVDaddr <t> [off2] {sym2} base) val mem) && is32Bit(off1+off2) && canMergeSym(sym1, sym2) && (base.Op != OpSB || (t.IsPtr() && t.ElemType().Alignment()%2 == 0 && (off1+off2)%2 == 0)) ->
   775  	(MOVHstore  [off1+off2] {mergeSym(sym1,sym2)} base val mem)
   776  (MOVBstore  [off1] {sym1} (MOVDaddr [off2] {sym2} base) val mem) && is32Bit(off1+off2) && canMergeSym(sym1, sym2) ->
   777  	(MOVBstore  [off1+off2] {mergeSym(sym1,sym2)} base val mem)
   778  (FMOVSstore [off1] {sym1} (MOVDaddr [off2] {sym2} base) val mem) && is32Bit(off1+off2) && canMergeSym(sym1, sym2) ->
   779  	(FMOVSstore [off1+off2] {mergeSym(sym1,sym2)} base val mem)
   780  (FMOVDstore [off1] {sym1} (MOVDaddr [off2] {sym2} base) val mem) && is32Bit(off1+off2) && canMergeSym(sym1, sym2) ->
   781  	(FMOVDstore [off1+off2] {mergeSym(sym1,sym2)} base val mem)
   782  
   783  // Cannot store constant to SB directly (no 'move relative long immediate' instructions).
   784  (MOVDstoreconst [sc] {sym1} (MOVDaddr [off] {sym2} ptr) mem) && ptr.Op != OpSB && canMergeSym(sym1, sym2) && ValAndOff(sc).canAdd(off) ->
   785  	(MOVDstoreconst [ValAndOff(sc).add(off)] {mergeSym(sym1, sym2)} ptr mem)
   786  (MOVWstoreconst [sc] {sym1} (MOVDaddr [off] {sym2} ptr) mem) && ptr.Op != OpSB && canMergeSym(sym1, sym2) && ValAndOff(sc).canAdd(off) ->
   787  	(MOVWstoreconst [ValAndOff(sc).add(off)] {mergeSym(sym1, sym2)} ptr mem)
   788  (MOVHstoreconst [sc] {sym1} (MOVDaddr [off] {sym2} ptr) mem) && ptr.Op != OpSB && canMergeSym(sym1, sym2) && ValAndOff(sc).canAdd(off) ->
   789  	(MOVHstoreconst [ValAndOff(sc).add(off)] {mergeSym(sym1, sym2)} ptr mem)
   790  (MOVBstoreconst [sc] {sym1} (MOVDaddr [off] {sym2} ptr) mem) && ptr.Op != OpSB && canMergeSym(sym1, sym2) && ValAndOff(sc).canAdd(off) ->
   791  	(MOVBstoreconst [ValAndOff(sc).add(off)] {mergeSym(sym1, sym2)} ptr mem)
   792  
   793  // generating indexed loads and stores
   794  (MOVBZload [off1] {sym1} (MOVDaddridx [off2] {sym2} ptr idx) mem) && is32Bit(off1+off2) && canMergeSym(sym1, sym2) ->
   795  	(MOVBZloadidx [off1+off2] {mergeSym(sym1,sym2)} ptr idx mem)
   796  (MOVHZload [off1] {sym1} (MOVDaddridx [off2] {sym2} ptr idx) mem) && is32Bit(off1+off2) && canMergeSym(sym1, sym2) ->
   797  	(MOVHZloadidx [off1+off2] {mergeSym(sym1,sym2)} ptr idx mem)
   798  (MOVWZload [off1] {sym1} (MOVDaddridx [off2] {sym2} ptr idx) mem) && is32Bit(off1+off2) && canMergeSym(sym1, sym2) ->
   799  	(MOVWZloadidx [off1+off2] {mergeSym(sym1,sym2)} ptr idx mem)
   800  (MOVDload [off1] {sym1} (MOVDaddridx [off2] {sym2} ptr idx) mem) && is32Bit(off1+off2) && canMergeSym(sym1, sym2) ->
   801  	(MOVDloadidx [off1+off2] {mergeSym(sym1,sym2)} ptr idx mem)
   802  (FMOVSload [off1] {sym1} (MOVDaddridx [off2] {sym2} ptr idx) mem) && is32Bit(off1+off2) && canMergeSym(sym1, sym2) ->
   803  	(FMOVSloadidx [off1+off2] {mergeSym(sym1,sym2)} ptr idx mem)
   804  (FMOVDload [off1] {sym1} (MOVDaddridx [off2] {sym2} ptr idx) mem) && is32Bit(off1+off2) && canMergeSym(sym1, sym2) ->
   805  	(FMOVDloadidx [off1+off2] {mergeSym(sym1,sym2)} ptr idx mem)
   806  
   807  (MOVBstore [off1] {sym1} (MOVDaddridx [off2] {sym2} ptr idx) val mem) && is32Bit(off1+off2) && canMergeSym(sym1, sym2) ->
   808  	(MOVBstoreidx [off1+off2] {mergeSym(sym1,sym2)} ptr idx val mem)
   809  (MOVHstore [off1] {sym1} (MOVDaddridx [off2] {sym2} ptr idx) val mem) && is32Bit(off1+off2) && canMergeSym(sym1, sym2) ->
   810  	(MOVHstoreidx [off1+off2] {mergeSym(sym1,sym2)} ptr idx val mem)
   811  (MOVWstore [off1] {sym1} (MOVDaddridx [off2] {sym2} ptr idx) val mem) && is32Bit(off1+off2) && canMergeSym(sym1, sym2) ->
   812  	(MOVWstoreidx [off1+off2] {mergeSym(sym1,sym2)} ptr idx val mem)
   813  (MOVDstore [off1] {sym1} (MOVDaddridx [off2] {sym2} ptr idx) val mem) && is32Bit(off1+off2) && canMergeSym(sym1, sym2) ->
   814  	(MOVDstoreidx [off1+off2] {mergeSym(sym1,sym2)} ptr idx val mem)
   815  (FMOVSstore [off1] {sym1} (MOVDaddridx [off2] {sym2} ptr idx) val mem) && is32Bit(off1+off2) && canMergeSym(sym1, sym2) ->
   816  	(FMOVSstoreidx [off1+off2] {mergeSym(sym1,sym2)} ptr idx val mem)
   817  (FMOVDstore [off1] {sym1} (MOVDaddridx [off2] {sym2} ptr idx) val mem) && is32Bit(off1+off2) && canMergeSym(sym1, sym2) ->
   818  	(FMOVDstoreidx [off1+off2] {mergeSym(sym1,sym2)} ptr idx val mem)
   819  
   820  (MOVBZload [off] {sym} (ADD ptr idx) mem) && ptr.Op != OpSB -> (MOVBZloadidx [off] {sym} ptr idx mem)
   821  (MOVHZload [off] {sym} (ADD ptr idx) mem) && ptr.Op != OpSB -> (MOVHZloadidx [off] {sym} ptr idx mem)
   822  (MOVWZload [off] {sym} (ADD ptr idx) mem) && ptr.Op != OpSB -> (MOVWZloadidx [off] {sym} ptr idx mem)
   823  (MOVDload [off] {sym} (ADD ptr idx) mem) && ptr.Op != OpSB -> (MOVDloadidx [off] {sym} ptr idx mem)
   824  (FMOVSload [off] {sym} (ADD ptr idx) mem) && ptr.Op != OpSB -> (FMOVSloadidx [off] {sym} ptr idx mem)
   825  (FMOVDload [off] {sym} (ADD ptr idx) mem) && ptr.Op != OpSB -> (FMOVDloadidx [off] {sym} ptr idx mem)
   826  (MOVBstore [off] {sym} (ADD ptr idx) val mem) && ptr.Op != OpSB -> (MOVBstoreidx [off] {sym} ptr idx val mem)
   827  (MOVHstore [off] {sym} (ADD ptr idx) val mem) && ptr.Op != OpSB -> (MOVHstoreidx [off] {sym} ptr idx val mem)
   828  (MOVWstore [off] {sym} (ADD ptr idx) val mem) && ptr.Op != OpSB -> (MOVWstoreidx [off] {sym} ptr idx val mem)
   829  (MOVDstore [off] {sym} (ADD ptr idx) val mem) && ptr.Op != OpSB -> (MOVDstoreidx [off] {sym} ptr idx val mem)
   830  (FMOVSstore [off] {sym} (ADD ptr idx) val mem) && ptr.Op != OpSB -> (FMOVSstoreidx [off] {sym} ptr idx val mem)
   831  (FMOVDstore [off] {sym} (ADD ptr idx) val mem) && ptr.Op != OpSB -> (FMOVDstoreidx [off] {sym} ptr idx val mem)
   832  
   833  // combine ADD into indexed loads and stores
   834  (MOVBZloadidx [c] {sym} (ADDconst [d] ptr) idx mem) -> (MOVBZloadidx [c+d] {sym} ptr idx mem)
   835  (MOVHZloadidx [c] {sym} (ADDconst [d] ptr) idx mem) -> (MOVHZloadidx [c+d] {sym} ptr idx mem)
   836  (MOVWZloadidx [c] {sym} (ADDconst [d] ptr) idx mem) -> (MOVWZloadidx [c+d] {sym} ptr idx mem)
   837  (MOVDloadidx [c] {sym} (ADDconst [d] ptr) idx mem) -> (MOVDloadidx [c+d] {sym} ptr idx mem)
   838  (FMOVSloadidx [c] {sym} (ADDconst [d] ptr) idx mem) -> (FMOVSloadidx [c+d] {sym} ptr idx mem)
   839  (FMOVDloadidx [c] {sym} (ADDconst [d] ptr) idx mem) -> (FMOVDloadidx [c+d] {sym} ptr idx mem)
   840  
   841  (MOVBstoreidx [c] {sym} (ADDconst [d] ptr) idx val mem) -> (MOVBstoreidx [c+d] {sym} ptr idx val mem)
   842  (MOVHstoreidx [c] {sym} (ADDconst [d] ptr) idx val mem) -> (MOVHstoreidx [c+d] {sym} ptr idx val mem)
   843  (MOVWstoreidx [c] {sym} (ADDconst [d] ptr) idx val mem) -> (MOVWstoreidx [c+d] {sym} ptr idx val mem)
   844  (MOVDstoreidx [c] {sym} (ADDconst [d] ptr) idx val mem) -> (MOVDstoreidx [c+d] {sym} ptr idx val mem)
   845  (FMOVSstoreidx [c] {sym} (ADDconst [d] ptr) idx val mem) -> (FMOVSstoreidx [c+d] {sym} ptr idx val mem)
   846  (FMOVDstoreidx [c] {sym} (ADDconst [d] ptr) idx val mem) -> (FMOVDstoreidx [c+d] {sym} ptr idx val mem)
   847  
   848  (MOVBZloadidx [c] {sym} ptr (ADDconst [d] idx) mem) -> (MOVBZloadidx [c+d] {sym} ptr idx mem)
   849  (MOVHZloadidx [c] {sym} ptr (ADDconst [d] idx) mem) -> (MOVHZloadidx [c+d] {sym} ptr idx mem)
   850  (MOVWZloadidx [c] {sym} ptr (ADDconst [d] idx) mem) -> (MOVWZloadidx [c+d] {sym} ptr idx mem)
   851  (MOVDloadidx [c] {sym} ptr (ADDconst [d] idx) mem) -> (MOVDloadidx [c+d] {sym} ptr idx mem)
   852  (FMOVSloadidx [c] {sym} ptr (ADDconst [d] idx) mem) -> (FMOVSloadidx [c+d] {sym} ptr idx mem)
   853  (FMOVDloadidx [c] {sym} ptr (ADDconst [d] idx) mem) -> (FMOVDloadidx [c+d] {sym} ptr idx mem)
   854  
   855  (MOVBstoreidx [c] {sym} ptr (ADDconst [d] idx) val mem) -> (MOVBstoreidx [c+d] {sym} ptr idx val mem)
   856  (MOVHstoreidx [c] {sym} ptr (ADDconst [d] idx) val mem) -> (MOVHstoreidx [c+d] {sym} ptr idx val mem)
   857  (MOVWstoreidx [c] {sym} ptr (ADDconst [d] idx) val mem) -> (MOVWstoreidx [c+d] {sym} ptr idx val mem)
   858  (MOVDstoreidx [c] {sym} ptr (ADDconst [d] idx) val mem) -> (MOVDstoreidx [c+d] {sym} ptr idx val mem)
   859  (FMOVSstoreidx [c] {sym} ptr (ADDconst [d] idx) val mem) -> (FMOVSstoreidx [c+d] {sym} ptr idx val mem)
   860  (FMOVDstoreidx [c] {sym} ptr (ADDconst [d] idx) val mem) -> (FMOVDstoreidx [c+d] {sym} ptr idx val mem)
   861  
   862  // MOVDaddr into MOVDaddridx
   863  (MOVDaddridx [off1] {sym1} (MOVDaddr [off2] {sym2} x) y) && is32Bit(off1+off2) && canMergeSym(sym1, sym2) && x.Op != OpSB ->
   864         (MOVDaddridx [off1+off2] {mergeSym(sym1,sym2)} x y)
   865  (MOVDaddridx [off1] {sym1} x (MOVDaddr [off2] {sym2} y)) && is32Bit(off1+off2) && canMergeSym(sym1, sym2) && y.Op != OpSB ->
   866         (MOVDaddridx [off1+off2] {mergeSym(sym1,sym2)} x y)
   867  
   868  // Absorb InvertFlags into branches.
   869  (LT (InvertFlags cmp) yes no) -> (GT cmp yes no)
   870  (GT (InvertFlags cmp) yes no) -> (LT cmp yes no)
   871  (LE (InvertFlags cmp) yes no) -> (GE cmp yes no)
   872  (GE (InvertFlags cmp) yes no) -> (LE cmp yes no)
   873  (EQ (InvertFlags cmp) yes no) -> (EQ cmp yes no)
   874  (NE (InvertFlags cmp) yes no) -> (NE cmp yes no)
   875  
   876  // Constant comparisons.
   877  (CMPconst (MOVDconst [x]) [y]) && x==y -> (FlagEQ)
   878  (CMPconst (MOVDconst [x]) [y]) && x<y -> (FlagLT)
   879  (CMPconst (MOVDconst [x]) [y]) && x>y -> (FlagGT)
   880  (CMPUconst (MOVDconst [x]) [y]) && uint64(x)==uint64(y) -> (FlagEQ)
   881  (CMPUconst (MOVDconst [x]) [y]) && uint64(x)<uint64(y) -> (FlagLT)
   882  (CMPUconst (MOVDconst [x]) [y]) && uint64(x)>uint64(y) -> (FlagGT)
   883  
   884  (CMPWconst (MOVDconst [x]) [y]) && int32(x)==int32(y) -> (FlagEQ)
   885  (CMPWconst (MOVDconst [x]) [y]) && int32(x)<int32(y) -> (FlagLT)
   886  (CMPWconst (MOVDconst [x]) [y]) && int32(x)>int32(y) -> (FlagGT)
   887  (CMPWUconst (MOVDconst [x]) [y]) && uint32(x)==uint32(y) -> (FlagEQ)
   888  (CMPWUconst (MOVDconst [x]) [y]) && uint32(x)<uint32(y) -> (FlagLT)
   889  (CMPWUconst (MOVDconst [x]) [y]) && uint32(x)>uint32(y) -> (FlagGT)
   890  
   891  // Other known comparisons.
   892  (CMPconst (MOVBZreg _) [c]) && 0xFF < c -> (FlagLT)
   893  (CMPconst (MOVHZreg _) [c]) && 0xFFFF < c -> (FlagLT)
   894  (CMPconst (MOVWZreg _) [c]) && 0xFFFFFFFF < c -> (FlagLT)
   895  (CMPWconst (SRWconst _ [c]) [n]) && 0 <= n && 0 < c && c <= 32 && (1<<uint64(32-c)) <= uint64(n) -> (FlagLT)
   896  (CMPconst (SRDconst _ [c]) [n]) && 0 <= n && 0 < c && c <= 64 && (1<<uint64(64-c)) <= uint64(n) -> (FlagLT)
   897  (CMPconst (ANDconst _ [m]) [n]) && 0 <= m && m < n -> (FlagLT)
   898  (CMPWconst (ANDWconst _ [m]) [n]) && 0 <= int32(m) && int32(m) < int32(n) -> (FlagLT)
   899  
   900  // Absorb flag constants into SBB ops.
   901  (SUBEcarrymask (FlagEQ)) -> (MOVDconst [-1])
   902  (SUBEcarrymask (FlagLT)) -> (MOVDconst [-1])
   903  (SUBEcarrymask (FlagGT)) -> (MOVDconst [0])
   904  (SUBEWcarrymask (FlagEQ)) -> (MOVDconst [-1])
   905  (SUBEWcarrymask (FlagLT)) -> (MOVDconst [-1])
   906  (SUBEWcarrymask (FlagGT)) -> (MOVDconst [0])
   907  
   908  // Absorb flag constants into branches.
   909  (EQ (FlagEQ) yes no) -> (First nil yes no)
   910  (EQ (FlagLT) yes no) -> (First nil no yes)
   911  (EQ (FlagGT) yes no) -> (First nil no yes)
   912  
   913  (NE (FlagEQ) yes no) -> (First nil no yes)
   914  (NE (FlagLT) yes no) -> (First nil yes no)
   915  (NE (FlagGT) yes no) -> (First nil yes no)
   916  
   917  (LT (FlagEQ) yes no) -> (First nil no yes)
   918  (LT (FlagLT) yes no) -> (First nil yes no)
   919  (LT (FlagGT) yes no) -> (First nil no yes)
   920  
   921  (LE (FlagEQ) yes no) -> (First nil yes no)
   922  (LE (FlagLT) yes no) -> (First nil yes no)
   923  (LE (FlagGT) yes no) -> (First nil no yes)
   924  
   925  (GT (FlagEQ) yes no) -> (First nil no yes)
   926  (GT (FlagLT) yes no) -> (First nil no yes)
   927  (GT (FlagGT) yes no) -> (First nil yes no)
   928  
   929  (GE (FlagEQ) yes no) -> (First nil yes no)
   930  (GE (FlagLT) yes no) -> (First nil no yes)
   931  (GE (FlagGT) yes no) -> (First nil yes no)
   932  
   933  // Absorb flag constants into SETxx ops.
   934  (MOVDEQ _ x (FlagEQ)) -> x
   935  (MOVDEQ y _ (FlagLT)) -> y
   936  (MOVDEQ y _ (FlagGT)) -> y
   937  
   938  (MOVDNE y _ (FlagEQ)) -> y
   939  (MOVDNE _ x (FlagLT)) -> x
   940  (MOVDNE _ x (FlagGT)) -> x
   941  
   942  (MOVDLT y _ (FlagEQ)) -> y
   943  (MOVDLT _ x (FlagLT)) -> x
   944  (MOVDLT y _ (FlagGT)) -> y
   945  
   946  (MOVDLE _ x (FlagEQ)) -> x
   947  (MOVDLE _ x (FlagLT)) -> x
   948  (MOVDLE y _ (FlagGT)) -> y
   949  
   950  (MOVDGT y _ (FlagEQ)) -> y
   951  (MOVDGT y _ (FlagLT)) -> y
   952  (MOVDGT _ x (FlagGT)) -> x
   953  
   954  (MOVDGE _ x (FlagEQ)) -> x
   955  (MOVDGE y _ (FlagLT)) -> y
   956  (MOVDGE _ x (FlagGT)) -> x
   957  
   958  // Remove redundant *const ops
   959  (ADDconst [0] x) -> x
   960  (ADDWconst [c] x) && int32(c)==0 -> x
   961  (SUBconst [0] x) -> x
   962  (SUBWconst [c] x) && int32(c) == 0 -> x
   963  (ANDconst [0] _)                 -> (MOVDconst [0])
   964  (ANDWconst [c] _) && int32(c)==0  -> (MOVDconst [0])
   965  (ANDconst [-1] x)                -> x
   966  (ANDWconst [c] x) && int32(c)==-1 -> x
   967  (ORconst [0] x)                  -> x
   968  (ORWconst [c] x) && int32(c)==0   -> x
   969  (ORconst [-1] _)                 -> (MOVDconst [-1])
   970  (ORWconst [c] _) && int32(c)==-1  -> (MOVDconst [-1])
   971  (XORconst [0] x)                  -> x
   972  (XORWconst [c] x) && int32(c)==0   -> x
   973  
   974  // Convert constant subtracts to constant adds.
   975  (SUBconst [c] x) && c != -(1<<31) -> (ADDconst [-c] x)
   976  (SUBWconst [c] x) -> (ADDWconst [int64(int32(-c))] x)
   977  
   978  // generic constant folding
   979  // TODO: more of this
   980  (ADDconst [c] (MOVDconst [d])) -> (MOVDconst [c+d])
   981  (ADDWconst [c] (MOVDconst [d])) -> (MOVDconst [int64(int32(c+d))])
   982  (ADDconst [c] (ADDconst [d] x)) && is32Bit(c+d) -> (ADDconst [c+d] x)
   983  (ADDWconst [c] (ADDWconst [d] x)) -> (ADDWconst [int64(int32(c+d))] x)
   984  (SUBconst (MOVDconst [d]) [c]) -> (MOVDconst [d-c])
   985  (SUBconst (SUBconst x [d]) [c]) && is32Bit(-c-d) -> (ADDconst [-c-d] x)
   986  (SRADconst [c] (MOVDconst [d])) -> (MOVDconst [d>>uint64(c)])
   987  (SRAWconst [c] (MOVDconst [d])) -> (MOVDconst [d>>uint64(c)])
   988  (NEG (MOVDconst [c])) -> (MOVDconst [-c])
   989  (NEGW (MOVDconst [c])) -> (MOVDconst [int64(int32(-c))])
   990  (MULLDconst [c] (MOVDconst [d])) -> (MOVDconst [c*d])
   991  (MULLWconst [c] (MOVDconst [d])) -> (MOVDconst [int64(int32(c*d))])
   992  (AND (MOVDconst [c]) (MOVDconst [d])) -> (MOVDconst [c&d])
   993  (ANDconst [c] (MOVDconst [d])) -> (MOVDconst [c&d])
   994  (ANDWconst [c] (MOVDconst [d])) -> (MOVDconst [c&d])
   995  (OR (MOVDconst [c]) (MOVDconst [d])) -> (MOVDconst [c|d])
   996  (ORconst [c] (MOVDconst [d])) -> (MOVDconst [c|d])
   997  (ORWconst [c] (MOVDconst [d])) -> (MOVDconst [c|d])
   998  (XOR (MOVDconst [c]) (MOVDconst [d])) -> (MOVDconst [c^d])
   999  (XORconst [c] (MOVDconst [d])) -> (MOVDconst [c^d])
  1000  (XORWconst [c] (MOVDconst [d])) -> (MOVDconst [c^d])
  1001  (LoweredRound32F x:(FMOVSconst)) -> x
  1002  (LoweredRound64F x:(FMOVDconst)) -> x
  1003  
  1004  // generic simplifications
  1005  // TODO: more of this
  1006  (ADD x (NEG y)) -> (SUB x y)
  1007  (ADDW x (NEGW y)) -> (SUBW x y)
  1008  (SUB x x) -> (MOVDconst [0])
  1009  (SUBW x x) -> (MOVDconst [0])
  1010  (AND x x) -> x
  1011  (ANDW x x) -> x
  1012  (OR x x) -> x
  1013  (ORW x x) -> x
  1014  (XOR x x) -> (MOVDconst [0])
  1015  (XORW x x) -> (MOVDconst [0])
  1016  (NEG (ADDconst [c] (NEG x))) && c != -(1<<31) -> (ADDconst [-c] x)
  1017  
  1018  // fused multiply-add
  1019  (FADD (FMUL y z) x) -> (FMADD x y z)
  1020  (FADDS (FMULS y z) x) -> (FMADDS x y z)
  1021  (FSUB (FMUL y z) x) -> (FMSUB x y z)
  1022  (FSUBS (FMULS y z) x) -> (FMSUBS x y z)
  1023  
  1024  // Fold memory operations into operations.
  1025  // Exclude global data (SB) because these instructions cannot handle relative addresses.
  1026  // TODO(mundaym): use LARL in the assembler to handle SB?
  1027  // TODO(mundaym): indexed versions of these?
  1028  (ADD <t> x g:(MOVDload [off] {sym} ptr mem)) && ptr.Op != OpSB && is20Bit(off) && canMergeLoad(v, g, x) && clobber(g)
  1029  	-> (ADDload <t> [off] {sym} x ptr mem)
  1030  (ADD <t> g:(MOVDload [off] {sym} ptr mem) x) && ptr.Op != OpSB && is20Bit(off) && canMergeLoad(v, g, x) && clobber(g)
  1031  	-> (ADDload <t> [off] {sym} x ptr mem)
  1032  (ADDW <t> x g:(MOVWload [off] {sym} ptr mem)) && ptr.Op != OpSB && is20Bit(off) && canMergeLoad(v, g, x) && clobber(g)
  1033  	-> (ADDWload <t> [off] {sym} x ptr mem)
  1034  (ADDW <t> g:(MOVWload [off] {sym} ptr mem) x) && ptr.Op != OpSB && is20Bit(off) && canMergeLoad(v, g, x) && clobber(g)
  1035  	-> (ADDWload <t> [off] {sym} x ptr mem)
  1036  (ADDW <t> x g:(MOVWZload [off] {sym} ptr mem)) && ptr.Op != OpSB && is20Bit(off) && canMergeLoad(v, g, x) && clobber(g)
  1037  	-> (ADDWload <t> [off] {sym} x ptr mem)
  1038  (ADDW <t> g:(MOVWZload [off] {sym} ptr mem) x) && ptr.Op != OpSB && is20Bit(off) && canMergeLoad(v, g, x) && clobber(g)
  1039  	-> (ADDWload <t> [off] {sym} x ptr mem)
  1040  (MULLD <t> x g:(MOVDload [off] {sym} ptr mem)) && ptr.Op != OpSB && is20Bit(off) && canMergeLoad(v, g, x) && clobber(g)
  1041  	-> (MULLDload <t> [off] {sym} x ptr mem)
  1042  (MULLD <t> g:(MOVDload [off] {sym} ptr mem) x) && ptr.Op != OpSB && is20Bit(off) && canMergeLoad(v, g, x) && clobber(g)
  1043  	-> (MULLDload <t> [off] {sym} x ptr mem)
  1044  (MULLW <t> x g:(MOVWload [off] {sym} ptr mem)) && ptr.Op != OpSB && is20Bit(off) && canMergeLoad(v, g, x) && clobber(g)
  1045  	-> (MULLWload <t> [off] {sym} x ptr mem)
  1046  (MULLW <t> g:(MOVWload [off] {sym} ptr mem) x) && ptr.Op != OpSB && is20Bit(off) && canMergeLoad(v, g, x) && clobber(g)
  1047  	-> (MULLWload <t> [off] {sym} x ptr mem)
  1048  (MULLW <t> x g:(MOVWZload [off] {sym} ptr mem)) && ptr.Op != OpSB && is20Bit(off) && canMergeLoad(v, g, x) && clobber(g)
  1049  	-> (MULLWload <t> [off] {sym} x ptr mem)
  1050  (MULLW <t> g:(MOVWZload [off] {sym} ptr mem) x) && ptr.Op != OpSB && is20Bit(off) && canMergeLoad(v, g, x) && clobber(g)
  1051  	-> (MULLWload <t> [off] {sym} x ptr mem)
  1052  (SUB <t> x g:(MOVDload [off] {sym} ptr mem)) && ptr.Op != OpSB && is20Bit(off) && canMergeLoad(v, g, x) && clobber(g)
  1053  	-> (SUBload <t> [off] {sym} x ptr mem)
  1054  (SUBW <t> x g:(MOVWload [off] {sym} ptr mem)) && ptr.Op != OpSB && is20Bit(off) && canMergeLoad(v, g, x) && clobber(g)
  1055  	-> (SUBWload <t> [off] {sym} x ptr mem)
  1056  (SUBW <t> x g:(MOVWZload [off] {sym} ptr mem)) && ptr.Op != OpSB && is20Bit(off) && canMergeLoad(v, g, x) && clobber(g)
  1057  	-> (SUBWload <t> [off] {sym} x ptr mem)
  1058  (AND <t> x g:(MOVDload [off] {sym} ptr mem)) && ptr.Op != OpSB && is20Bit(off) && canMergeLoad(v, g, x) && clobber(g)
  1059  	-> (ANDload <t> [off] {sym} x ptr mem)
  1060  (AND <t> g:(MOVDload [off] {sym} ptr mem) x) && ptr.Op != OpSB && is20Bit(off) && canMergeLoad(v, g, x) && clobber(g)
  1061  	-> (ANDload <t> [off] {sym} x ptr mem)
  1062  (ANDW <t> x g:(MOVWload [off] {sym} ptr mem)) && ptr.Op != OpSB && is20Bit(off) && canMergeLoad(v, g, x) && clobber(g)
  1063  	-> (ANDWload <t> [off] {sym} x ptr mem)
  1064  (ANDW <t> g:(MOVWload [off] {sym} ptr mem) x) && ptr.Op != OpSB && is20Bit(off) && canMergeLoad(v, g, x) && clobber(g)
  1065  	-> (ANDWload <t> [off] {sym} x ptr mem)
  1066  (ANDW <t> x g:(MOVWZload [off] {sym} ptr mem)) && ptr.Op != OpSB && is20Bit(off) && canMergeLoad(v, g, x) && clobber(g)
  1067  	-> (ANDWload <t> [off] {sym} x ptr mem)
  1068  (ANDW <t> g:(MOVWZload [off] {sym} ptr mem) x) && ptr.Op != OpSB && is20Bit(off) && canMergeLoad(v, g, x) && clobber(g)
  1069  	-> (ANDWload <t> [off] {sym} x ptr mem)
  1070  (OR <t> x g:(MOVDload [off] {sym} ptr mem)) && ptr.Op != OpSB && is20Bit(off) && canMergeLoad(v, g, x) && clobber(g)
  1071  	-> (ORload <t> [off] {sym} x ptr mem)
  1072  (OR <t> g:(MOVDload [off] {sym} ptr mem) x) && ptr.Op != OpSB && is20Bit(off) && canMergeLoad(v, g, x) && clobber(g)
  1073  	-> (ORload <t> [off] {sym} x ptr mem)
  1074  (ORW <t> x g:(MOVWload [off] {sym} ptr mem)) && ptr.Op != OpSB && is20Bit(off) && canMergeLoad(v, g, x) && clobber(g)
  1075  	-> (ORWload <t> [off] {sym} x ptr mem)
  1076  (ORW <t> g:(MOVWload [off] {sym} ptr mem) x) && ptr.Op != OpSB && is20Bit(off) && canMergeLoad(v, g, x) && clobber(g)
  1077  	-> (ORWload <t> [off] {sym} x ptr mem)
  1078  (ORW <t> x g:(MOVWZload [off] {sym} ptr mem)) && ptr.Op != OpSB && is20Bit(off) && canMergeLoad(v, g, x) && clobber(g)
  1079  	-> (ORWload <t> [off] {sym} x ptr mem)
  1080  (ORW <t> g:(MOVWZload [off] {sym} ptr mem) x) && ptr.Op != OpSB && is20Bit(off) && canMergeLoad(v, g, x) && clobber(g)
  1081  	-> (ORWload <t> [off] {sym} x ptr mem)
  1082  (XOR <t> x g:(MOVDload [off] {sym} ptr mem)) && ptr.Op != OpSB && is20Bit(off) && canMergeLoad(v, g, x) && clobber(g)
  1083  	-> (XORload <t> [off] {sym} x ptr mem)
  1084  (XOR <t> g:(MOVDload [off] {sym} ptr mem) x) && ptr.Op != OpSB && is20Bit(off) && canMergeLoad(v, g, x) && clobber(g)
  1085  	-> (XORload <t> [off] {sym} x ptr mem)
  1086  (XORW <t> x g:(MOVWload [off] {sym} ptr mem)) && ptr.Op != OpSB && is20Bit(off) && canMergeLoad(v, g, x) && clobber(g)
  1087  	-> (XORWload <t> [off] {sym} x ptr mem)
  1088  (XORW <t> g:(MOVWload [off] {sym} ptr mem) x) && ptr.Op != OpSB && is20Bit(off) && canMergeLoad(v, g, x) && clobber(g)
  1089  	-> (XORWload <t> [off] {sym} x ptr mem)
  1090  (XORW <t> x g:(MOVWZload [off] {sym} ptr mem)) && ptr.Op != OpSB && is20Bit(off) && canMergeLoad(v, g, x) && clobber(g)
  1091  	-> (XORWload <t> [off] {sym} x ptr mem)
  1092  (XORW <t> g:(MOVWZload [off] {sym} ptr mem) x) && ptr.Op != OpSB && is20Bit(off) && canMergeLoad(v, g, x) && clobber(g)
  1093  	-> (XORWload <t> [off] {sym} x ptr mem)
  1094  
  1095  // Combine constant stores into larger (unaligned) stores.
  1096  // Avoid SB because constant stores to relative offsets are
  1097  // emulated by the assembler and also can't handle unaligned offsets.
  1098  (MOVBstoreconst [c] {s} p x:(MOVBstoreconst [a] {s} p mem))
  1099    && p.Op != OpSB
  1100    && x.Uses == 1
  1101    && ValAndOff(a).Off() + 1 == ValAndOff(c).Off()
  1102    && clobber(x)
  1103    -> (MOVHstoreconst [makeValAndOff(ValAndOff(c).Val()&0xff | ValAndOff(a).Val()<<8, ValAndOff(a).Off())] {s} p mem)
  1104  (MOVHstoreconst [c] {s} p x:(MOVHstoreconst [a] {s} p mem))
  1105    && p.Op != OpSB
  1106    && x.Uses == 1
  1107    && ValAndOff(a).Off() + 2 == ValAndOff(c).Off()
  1108    && clobber(x)
  1109    -> (MOVWstore [ValAndOff(a).Off()] {s} p (MOVDconst [int64(int32(ValAndOff(c).Val()&0xffff | ValAndOff(a).Val()<<16))]) mem)
  1110  (MOVWstoreconst [c] {s} p x:(MOVWstoreconst [a] {s} p mem))
  1111    && p.Op != OpSB
  1112    && x.Uses == 1
  1113    && ValAndOff(a).Off() + 4 == ValAndOff(c).Off()
  1114    && clobber(x)
  1115    -> (MOVDstore [ValAndOff(a).Off()] {s} p (MOVDconst [ValAndOff(c).Val()&0xffffffff | ValAndOff(a).Val()<<32]) mem)
  1116  
  1117  // Combine stores into larger (unaligned) stores.
  1118  // It doesn't work on global data (based on SB) because stores with relative addressing
  1119  // require that the memory operand be aligned.
  1120  (MOVBstore [i] {s} p w x:(MOVBstore [i-1] {s} p (SRDconst [8] w) mem))
  1121    && p.Op != OpSB
  1122    && x.Uses == 1
  1123    && clobber(x)
  1124    -> (MOVHstore [i-1] {s} p w mem)
  1125  (MOVBstore [i] {s} p w0:(SRDconst [j] w) x:(MOVBstore [i-1] {s} p (SRDconst [j+8] w) mem))
  1126    && p.Op != OpSB
  1127    && x.Uses == 1
  1128    && clobber(x)
  1129    -> (MOVHstore [i-1] {s} p w0 mem)
  1130  (MOVBstore [i] {s} p w x:(MOVBstore [i-1] {s} p (SRWconst [8] w) mem))
  1131    && p.Op != OpSB
  1132    && x.Uses == 1
  1133    && clobber(x)
  1134    -> (MOVHstore [i-1] {s} p w mem)
  1135  (MOVBstore [i] {s} p w0:(SRWconst [j] w) x:(MOVBstore [i-1] {s} p (SRWconst [j+8] w) mem))
  1136    && p.Op != OpSB
  1137    && x.Uses == 1
  1138    && clobber(x)
  1139    -> (MOVHstore [i-1] {s} p w0 mem)
  1140  (MOVHstore [i] {s} p w x:(MOVHstore [i-2] {s} p (SRDconst [16] w) mem))
  1141    && p.Op != OpSB
  1142    && x.Uses == 1
  1143    && clobber(x)
  1144    -> (MOVWstore [i-2] {s} p w mem)
  1145  (MOVHstore [i] {s} p w0:(SRDconst [j] w) x:(MOVHstore [i-2] {s} p (SRDconst [j+16] w) mem))
  1146    && p.Op != OpSB
  1147    && x.Uses == 1
  1148    && clobber(x)
  1149    -> (MOVWstore [i-2] {s} p w0 mem)
  1150  (MOVHstore [i] {s} p w x:(MOVHstore [i-2] {s} p (SRWconst [16] w) mem))
  1151    && p.Op != OpSB
  1152    && x.Uses == 1
  1153    && clobber(x)
  1154    -> (MOVWstore [i-2] {s} p w mem)
  1155  (MOVHstore [i] {s} p w0:(SRWconst [j] w) x:(MOVHstore [i-2] {s} p (SRWconst [j+16] w) mem))
  1156    && p.Op != OpSB
  1157    && x.Uses == 1
  1158    && clobber(x)
  1159    -> (MOVWstore [i-2] {s} p w0 mem)
  1160  (MOVWstore [i] {s} p (SRDconst [32] w) x:(MOVWstore [i-4] {s} p w mem))
  1161    && p.Op != OpSB
  1162    && x.Uses == 1
  1163    && clobber(x)
  1164    -> (MOVDstore [i-4] {s} p w mem)
  1165  (MOVWstore [i] {s} p w0:(SRDconst [j] w) x:(MOVWstore [i-4] {s} p (SRDconst [j+32] w) mem))
  1166    && p.Op != OpSB
  1167    && x.Uses == 1
  1168    && clobber(x)
  1169    -> (MOVDstore [i-4] {s} p w0 mem)
  1170  
  1171  (MOVBstoreidx [i] {s} p idx w x:(MOVBstoreidx [i-1] {s} p idx (SRDconst [8] w) mem))
  1172    && x.Uses == 1
  1173    && clobber(x)
  1174    -> (MOVHstoreidx [i-1] {s} p idx w mem)
  1175  (MOVBstoreidx [i] {s} p idx w0:(SRDconst [j] w) x:(MOVBstoreidx [i-1] {s} p idx (SRDconst [j+8] w) mem))
  1176    && x.Uses == 1
  1177    && clobber(x)
  1178    -> (MOVHstoreidx [i-1] {s} p idx w0 mem)
  1179  (MOVBstoreidx [i] {s} p idx w x:(MOVBstoreidx [i-1] {s} p idx (SRWconst [8] w) mem))
  1180    && x.Uses == 1
  1181    && clobber(x)
  1182    -> (MOVHstoreidx [i-1] {s} p idx w mem)
  1183  (MOVBstoreidx [i] {s} p idx w0:(SRWconst [j] w) x:(MOVBstoreidx [i-1] {s} p idx (SRWconst [j+8] w) mem))
  1184    && x.Uses == 1
  1185    && clobber(x)
  1186    -> (MOVHstoreidx [i-1] {s} p idx w0 mem)
  1187  (MOVHstoreidx [i] {s} p idx w x:(MOVHstoreidx [i-2] {s} p idx (SRDconst [16] w) mem))
  1188    && x.Uses == 1
  1189    && clobber(x)
  1190    -> (MOVWstoreidx [i-2] {s} p idx w mem)
  1191  (MOVHstoreidx [i] {s} p idx w0:(SRDconst [j] w) x:(MOVHstoreidx [i-2] {s} p idx (SRDconst [j+16] w) mem))
  1192    && x.Uses == 1
  1193    && clobber(x)
  1194    -> (MOVWstoreidx [i-2] {s} p idx w0 mem)
  1195  (MOVHstoreidx [i] {s} p idx w x:(MOVHstoreidx [i-2] {s} p idx (SRWconst [16] w) mem))
  1196    && x.Uses == 1
  1197    && clobber(x)
  1198    -> (MOVWstoreidx [i-2] {s} p idx w mem)
  1199  (MOVHstoreidx [i] {s} p idx w0:(SRWconst [j] w) x:(MOVHstoreidx [i-2] {s} p idx (SRWconst [j+16] w) mem))
  1200    && x.Uses == 1
  1201    && clobber(x)
  1202    -> (MOVWstoreidx [i-2] {s} p idx w0 mem)
  1203  (MOVWstoreidx [i] {s} p idx w x:(MOVWstoreidx [i-4] {s} p idx (SRDconst [32] w) mem))
  1204    && x.Uses == 1
  1205    && clobber(x)
  1206    -> (MOVDstoreidx [i-4] {s} p idx w mem)
  1207  (MOVWstoreidx [i] {s} p idx w0:(SRDconst [j] w) x:(MOVWstoreidx [i-4] {s} p idx (SRDconst [j+32] w) mem))
  1208    && x.Uses == 1
  1209    && clobber(x)
  1210    -> (MOVDstoreidx [i-4] {s} p idx w0 mem)
  1211  
  1212  // Combine stores into larger (unaligned) stores with the bytes reversed (little endian).
  1213  // Store-with-bytes-reversed instructions do not support relative memory addresses,
  1214  // so these stores can't operate on global data (SB).
  1215  (MOVBstore [i] {s} p (SRDconst [8] w) x:(MOVBstore [i-1] {s} p w mem))
  1216    && p.Op != OpSB
  1217    && x.Uses == 1
  1218    && clobber(x)
  1219    -> (MOVHBRstore [i-1] {s} p w mem)
  1220  (MOVBstore [i] {s} p (SRDconst [j] w) x:(MOVBstore [i-1] {s} p w0:(SRDconst [j-8] w) mem))
  1221    && p.Op != OpSB
  1222    && x.Uses == 1
  1223    && clobber(x)
  1224    -> (MOVHBRstore [i-1] {s} p w0 mem)
  1225  (MOVBstore [i] {s} p (SRWconst [8] w) x:(MOVBstore [i-1] {s} p w mem))
  1226    && p.Op != OpSB
  1227    && x.Uses == 1
  1228    && clobber(x)
  1229    -> (MOVHBRstore [i-1] {s} p w mem)
  1230  (MOVBstore [i] {s} p (SRWconst [j] w) x:(MOVBstore [i-1] {s} p w0:(SRWconst [j-8] w) mem))
  1231    && p.Op != OpSB
  1232    && x.Uses == 1
  1233    && clobber(x)
  1234    -> (MOVHBRstore [i-1] {s} p w0 mem)
  1235  (MOVHBRstore [i] {s} p (SRDconst [16] w) x:(MOVHBRstore [i-2] {s} p w mem))
  1236    && x.Uses == 1
  1237    && clobber(x)
  1238    -> (MOVWBRstore [i-2] {s} p w mem)
  1239  (MOVHBRstore [i] {s} p (SRDconst [j] w) x:(MOVHBRstore [i-2] {s} p w0:(SRDconst [j-16] w) mem))
  1240    && x.Uses == 1
  1241    && clobber(x)
  1242    -> (MOVWBRstore [i-2] {s} p w0 mem)
  1243  (MOVHBRstore [i] {s} p (SRWconst [16] w) x:(MOVHBRstore [i-2] {s} p w mem))
  1244    && x.Uses == 1
  1245    && clobber(x)
  1246    -> (MOVWBRstore [i-2] {s} p w mem)
  1247  (MOVHBRstore [i] {s} p (SRWconst [j] w) x:(MOVHBRstore [i-2] {s} p w0:(SRWconst [j-16] w) mem))
  1248    && x.Uses == 1
  1249    && clobber(x)
  1250    -> (MOVWBRstore [i-2] {s} p w0 mem)
  1251  (MOVWBRstore [i] {s} p (SRDconst [32] w) x:(MOVWBRstore [i-4] {s} p w mem))
  1252    && x.Uses == 1
  1253    && clobber(x)
  1254    -> (MOVDBRstore [i-4] {s} p w mem)
  1255  (MOVWBRstore [i] {s} p (SRDconst [j] w) x:(MOVWBRstore [i-4] {s} p w0:(SRDconst [j-32] w) mem))
  1256    && x.Uses == 1
  1257    && clobber(x)
  1258    -> (MOVDBRstore [i-4] {s} p w0 mem)
  1259  
  1260  (MOVBstoreidx [i] {s} p idx (SRDconst [8] w) x:(MOVBstoreidx [i-1] {s} p idx w mem))
  1261    && x.Uses == 1
  1262    && clobber(x)
  1263    -> (MOVHBRstoreidx [i-1] {s} p idx w mem)
  1264  (MOVBstoreidx [i] {s} p idx (SRDconst [j] w) x:(MOVBstoreidx [i-1] {s} p idx w0:(SRDconst [j-8] w) mem))
  1265    && x.Uses == 1
  1266    && clobber(x)
  1267    -> (MOVHBRstoreidx [i-1] {s} p idx w0 mem)
  1268  (MOVBstoreidx [i] {s} p idx (SRWconst [8] w) x:(MOVBstoreidx [i-1] {s} p idx w mem))
  1269    && x.Uses == 1
  1270    && clobber(x)
  1271    -> (MOVHBRstoreidx [i-1] {s} p idx w mem)
  1272  (MOVBstoreidx [i] {s} p idx (SRWconst [j] w) x:(MOVBstoreidx [i-1] {s} p idx w0:(SRWconst [j-8] w) mem))
  1273    && x.Uses == 1
  1274    && clobber(x)
  1275    -> (MOVHBRstoreidx [i-1] {s} p idx w0 mem)
  1276  (MOVHBRstoreidx [i] {s} p idx (SRDconst [16] w) x:(MOVHBRstoreidx [i-2] {s} p idx w mem))
  1277    && x.Uses == 1
  1278    && clobber(x)
  1279    -> (MOVWBRstoreidx [i-2] {s} p idx w mem)
  1280  (MOVHBRstoreidx [i] {s} p idx (SRDconst [j] w) x:(MOVHBRstoreidx [i-2] {s} p idx w0:(SRDconst [j-16] w) mem))
  1281    && x.Uses == 1
  1282    && clobber(x)
  1283    -> (MOVWBRstoreidx [i-2] {s} p idx w0 mem)
  1284  (MOVHBRstoreidx [i] {s} p idx (SRWconst [16] w) x:(MOVHBRstoreidx [i-2] {s} p idx w mem))
  1285    && x.Uses == 1
  1286    && clobber(x)
  1287    -> (MOVWBRstoreidx [i-2] {s} p idx w mem)
  1288  (MOVHBRstoreidx [i] {s} p idx (SRWconst [j] w) x:(MOVHBRstoreidx [i-2] {s} p idx w0:(SRWconst [j-16] w) mem))
  1289    && x.Uses == 1
  1290    && clobber(x)
  1291    -> (MOVWBRstoreidx [i-2] {s} p idx w0 mem)
  1292  (MOVWBRstoreidx [i] {s} p idx (SRDconst [32] w) x:(MOVWBRstoreidx [i-4] {s} p idx w mem))
  1293    && x.Uses == 1
  1294    && clobber(x)
  1295    -> (MOVDBRstoreidx [i-4] {s} p idx w mem)
  1296  (MOVWBRstoreidx [i] {s} p idx (SRDconst [j] w) x:(MOVWBRstoreidx [i-4] {s} p idx w0:(SRDconst [j-32] w) mem))
  1297    && x.Uses == 1
  1298    && clobber(x)
  1299    -> (MOVDBRstoreidx [i-4] {s} p idx w0 mem)
  1300  
  1301  // Combining byte loads into larger (unaligned) loads.
  1302  
  1303  // Big-endian loads
  1304  
  1305  (ORW                 x1:(MOVBZload [i1] {s} p mem)
  1306      sh:(SLWconst [8] x0:(MOVBZload [i0] {s} p mem)))
  1307    && i1 == i0+1
  1308    && p.Op != OpSB
  1309    && x0.Uses == 1
  1310    && x1.Uses == 1
  1311    && sh.Uses == 1
  1312    && mergePoint(b,x0,x1) != nil
  1313    && clobber(x0)
  1314    && clobber(x1)
  1315    && clobber(sh)
  1316    -> @mergePoint(b,x0,x1) (MOVHZload [i0] {s} p mem)
  1317  
  1318  (OR                  x1:(MOVBZload [i1] {s} p mem)
  1319      sh:(SLDconst [8] x0:(MOVBZload [i0] {s} p mem)))
  1320    && i1 == i0+1
  1321    && p.Op != OpSB
  1322    && x0.Uses == 1
  1323    && x1.Uses == 1
  1324    && sh.Uses == 1
  1325    && mergePoint(b,x0,x1) != nil
  1326    && clobber(x0)
  1327    && clobber(x1)
  1328    && clobber(sh)
  1329    -> @mergePoint(b,x0,x1) (MOVHZload [i0] {s} p mem)
  1330  
  1331  (ORW                  x1:(MOVHZload [i1] {s} p mem)
  1332      sh:(SLWconst [16] x0:(MOVHZload [i0] {s} p mem)))
  1333    && i1 == i0+2
  1334    && p.Op != OpSB
  1335    && x0.Uses == 1
  1336    && x1.Uses == 1
  1337    && sh.Uses == 1
  1338    && mergePoint(b,x0,x1) != nil
  1339    && clobber(x0)
  1340    && clobber(x1)
  1341    && clobber(sh)
  1342    -> @mergePoint(b,x0,x1) (MOVWZload [i0] {s} p mem)
  1343  
  1344  (OR                   x1:(MOVHZload [i1] {s} p mem)
  1345      sh:(SLDconst [16] x0:(MOVHZload [i0] {s} p mem)))
  1346    && i1 == i0+2
  1347    && p.Op != OpSB
  1348    && x0.Uses == 1
  1349    && x1.Uses == 1
  1350    && sh.Uses == 1
  1351    && mergePoint(b,x0,x1) != nil
  1352    && clobber(x0)
  1353    && clobber(x1)
  1354    && clobber(sh)
  1355    -> @mergePoint(b,x0,x1) (MOVWZload [i0] {s} p mem)
  1356  
  1357  (OR                   x1:(MOVWZload [i1] {s} p mem)
  1358      sh:(SLDconst [32] x0:(MOVWZload [i0] {s} p mem)))
  1359    && i1 == i0+4
  1360    && p.Op != OpSB
  1361    && x0.Uses == 1
  1362    && x1.Uses == 1
  1363    && sh.Uses == 1
  1364    && mergePoint(b,x0,x1) != nil
  1365    && clobber(x0)
  1366    && clobber(x1)
  1367    && clobber(sh)
  1368    -> @mergePoint(b,x0,x1) (MOVDload [i0] {s} p mem)
  1369  
  1370  (ORW
  1371      s0:(SLWconst [j0] x0:(MOVBZload [i0] {s} p mem))
  1372      or:(ORW
  1373          s1:(SLWconst [j1] x1:(MOVBZload [i1] {s} p mem))
  1374  	y))
  1375    && i1 == i0+1
  1376    && j1 == j0-8
  1377    && j1 % 16 == 0
  1378    && x0.Uses == 1
  1379    && x1.Uses == 1
  1380    && s0.Uses == 1
  1381    && s1.Uses == 1
  1382    && or.Uses == 1
  1383    && mergePoint(b,x0,x1) != nil
  1384    && clobber(x0)
  1385    && clobber(x1)
  1386    && clobber(s0)
  1387    && clobber(s1)
  1388    && clobber(or)
  1389    -> @mergePoint(b,x0,x1) (ORW <v.Type> (SLWconst <v.Type> [j1] (MOVHZload [i0] {s} p mem)) y)
  1390  
  1391  (OR
  1392      s0:(SLDconst [j0] x0:(MOVBZload [i0] {s} p mem))
  1393      or:(OR
  1394          s1:(SLDconst [j1] x1:(MOVBZload [i1] {s} p mem))
  1395  	y))
  1396    && i1 == i0+1
  1397    && j1 == j0-8
  1398    && j1 % 16 == 0
  1399    && x0.Uses == 1
  1400    && x1.Uses == 1
  1401    && s0.Uses == 1
  1402    && s1.Uses == 1
  1403    && or.Uses == 1
  1404    && mergePoint(b,x0,x1) != nil
  1405    && clobber(x0)
  1406    && clobber(x1)
  1407    && clobber(s0)
  1408    && clobber(s1)
  1409    && clobber(or)
  1410    -> @mergePoint(b,x0,x1) (OR <v.Type> (SLDconst <v.Type> [j1] (MOVHZload [i0] {s} p mem)) y)
  1411  
  1412  (OR
  1413      s0:(SLDconst [j0] x0:(MOVHZload [i0] {s} p mem))
  1414      or:(OR
  1415          s1:(SLDconst [j1] x1:(MOVHZload [i1] {s} p mem))
  1416  	y))
  1417    && i1 == i0+2
  1418    && j1 == j0-16
  1419    && j1 % 32 == 0
  1420    && x0.Uses == 1
  1421    && x1.Uses == 1
  1422    && s0.Uses == 1
  1423    && s1.Uses == 1
  1424    && or.Uses == 1
  1425    && mergePoint(b,x0,x1) != nil
  1426    && clobber(x0)
  1427    && clobber(x1)
  1428    && clobber(s0)
  1429    && clobber(s1)
  1430    && clobber(or)
  1431    -> @mergePoint(b,x0,x1) (OR <v.Type> (SLDconst <v.Type> [j1] (MOVWZload [i0] {s} p mem)) y)
  1432  
  1433  // Big-endian indexed loads
  1434  
  1435  (ORW                 x1:(MOVBZloadidx [i1] {s} p idx mem)
  1436      sh:(SLWconst [8] x0:(MOVBZloadidx [i0] {s} p idx mem)))
  1437    && i1 == i0+1
  1438    && p.Op != OpSB
  1439    && x0.Uses == 1
  1440    && x1.Uses == 1
  1441    && sh.Uses == 1
  1442    && mergePoint(b,x0,x1) != nil
  1443    && clobber(x0)
  1444    && clobber(x1)
  1445    && clobber(sh)
  1446    -> @mergePoint(b,x0,x1) (MOVHZloadidx [i0] {s} p idx mem)
  1447  
  1448  (OR                  x1:(MOVBZloadidx [i1] {s} p idx mem)
  1449      sh:(SLDconst [8] x0:(MOVBZloadidx [i0] {s} p idx mem)))
  1450    && i1 == i0+1
  1451    && p.Op != OpSB
  1452    && x0.Uses == 1
  1453    && x1.Uses == 1
  1454    && sh.Uses == 1
  1455    && mergePoint(b,x0,x1) != nil
  1456    && clobber(x0)
  1457    && clobber(x1)
  1458    && clobber(sh)
  1459    -> @mergePoint(b,x0,x1) (MOVHZloadidx [i0] {s} p idx mem)
  1460  
  1461  (ORW                  x1:(MOVHZloadidx [i1] {s} p idx mem)
  1462      sh:(SLWconst [16] x0:(MOVHZloadidx [i0] {s} p idx mem)))
  1463    && i1 == i0+2
  1464    && p.Op != OpSB
  1465    && x0.Uses == 1
  1466    && x1.Uses == 1
  1467    && sh.Uses == 1
  1468    && mergePoint(b,x0,x1) != nil
  1469    && clobber(x0)
  1470    && clobber(x1)
  1471    && clobber(sh)
  1472    -> @mergePoint(b,x0,x1) (MOVWZloadidx [i0] {s} p idx mem)
  1473  
  1474  (OR                   x1:(MOVHZloadidx [i1] {s} p idx mem)
  1475      sh:(SLDconst [16] x0:(MOVHZloadidx [i0] {s} p idx mem)))
  1476    && i1 == i0+2
  1477    && p.Op != OpSB
  1478    && x0.Uses == 1
  1479    && x1.Uses == 1
  1480    && sh.Uses == 1
  1481    && mergePoint(b,x0,x1) != nil
  1482    && clobber(x0)
  1483    && clobber(x1)
  1484    && clobber(sh)
  1485    -> @mergePoint(b,x0,x1) (MOVWZloadidx [i0] {s} p idx mem)
  1486  
  1487  (OR                   x1:(MOVWZloadidx [i1] {s} p idx mem)
  1488      sh:(SLDconst [32] x0:(MOVWZloadidx [i0] {s} p idx mem)))
  1489    && i1 == i0+4
  1490    && p.Op != OpSB
  1491    && x0.Uses == 1
  1492    && x1.Uses == 1
  1493    && sh.Uses == 1
  1494    && mergePoint(b,x0,x1) != nil
  1495    && clobber(x0)
  1496    && clobber(x1)
  1497    && clobber(sh)
  1498    -> @mergePoint(b,x0,x1) (MOVDloadidx [i0] {s} p idx mem)
  1499  
  1500  (ORW
  1501      s0:(SLWconst [j0] x0:(MOVBZloadidx [i0] {s} p idx mem))
  1502      or:(ORW
  1503          s1:(SLWconst [j1] x1:(MOVBZloadidx [i1] {s} p idx mem))
  1504  	y))
  1505    && i1 == i0+1
  1506    && j1 == j0-8
  1507    && j1 % 16 == 0
  1508    && x0.Uses == 1
  1509    && x1.Uses == 1
  1510    && s0.Uses == 1
  1511    && s1.Uses == 1
  1512    && or.Uses == 1
  1513    && mergePoint(b,x0,x1) != nil
  1514    && clobber(x0)
  1515    && clobber(x1)
  1516    && clobber(s0)
  1517    && clobber(s1)
  1518    && clobber(or)
  1519    -> @mergePoint(b,x0,x1) (ORW <v.Type> (SLWconst <v.Type> [j1] (MOVHZloadidx [i0] {s} p idx mem)) y)
  1520  
  1521  (OR
  1522      s0:(SLDconst [j0] x0:(MOVBZloadidx [i0] {s} p idx mem))
  1523      or:(OR
  1524          s1:(SLDconst [j1] x1:(MOVBZloadidx [i1] {s} p idx mem))
  1525  	y))
  1526    && i1 == i0+1
  1527    && j1 == j0-8
  1528    && j1 % 16 == 0
  1529    && x0.Uses == 1
  1530    && x1.Uses == 1
  1531    && s0.Uses == 1
  1532    && s1.Uses == 1
  1533    && or.Uses == 1
  1534    && mergePoint(b,x0,x1) != nil
  1535    && clobber(x0)
  1536    && clobber(x1)
  1537    && clobber(s0)
  1538    && clobber(s1)
  1539    && clobber(or)
  1540    -> @mergePoint(b,x0,x1) (OR <v.Type> (SLDconst <v.Type> [j1] (MOVHZloadidx [i0] {s} p idx mem)) y)
  1541  
  1542  (OR
  1543      s0:(SLDconst [j0] x0:(MOVHZloadidx [i0] {s} p idx mem))
  1544      or:(OR
  1545          s1:(SLDconst [j1] x1:(MOVHZloadidx [i1] {s} p idx mem))
  1546  	y))
  1547    && i1 == i0+2
  1548    && j1 == j0-16
  1549    && j1 % 32 == 0
  1550    && x0.Uses == 1
  1551    && x1.Uses == 1
  1552    && s0.Uses == 1
  1553    && s1.Uses == 1
  1554    && or.Uses == 1
  1555    && mergePoint(b,x0,x1) != nil
  1556    && clobber(x0)
  1557    && clobber(x1)
  1558    && clobber(s0)
  1559    && clobber(s1)
  1560    && clobber(or)
  1561    -> @mergePoint(b,x0,x1) (OR <v.Type> (SLDconst <v.Type> [j1] (MOVWZloadidx [i0] {s} p idx mem)) y)
  1562  
  1563  // Little-endian loads
  1564  
  1565  (ORW                 x0:(MOVBZload [i0] {s} p mem)
  1566      sh:(SLWconst [8] x1:(MOVBZload [i1] {s} p mem)))
  1567    && p.Op != OpSB
  1568    && i1 == i0+1
  1569    && x0.Uses == 1
  1570    && x1.Uses == 1
  1571    && sh.Uses == 1
  1572    && mergePoint(b,x0,x1) != nil
  1573    && clobber(x0)
  1574    && clobber(x1)
  1575    && clobber(sh)
  1576    -> @mergePoint(b,x0,x1) (MOVHZreg (MOVHBRload [i0] {s} p mem))
  1577  
  1578  (OR                  x0:(MOVBZload [i0] {s} p mem)
  1579      sh:(SLDconst [8] x1:(MOVBZload [i1] {s} p mem)))
  1580    && p.Op != OpSB
  1581    && i1 == i0+1
  1582    && x0.Uses == 1
  1583    && x1.Uses == 1
  1584    && sh.Uses == 1
  1585    && mergePoint(b,x0,x1) != nil
  1586    && clobber(x0)
  1587    && clobber(x1)
  1588    && clobber(sh)
  1589    -> @mergePoint(b,x0,x1) (MOVHZreg (MOVHBRload [i0] {s} p mem))
  1590  
  1591  (ORW                  r0:(MOVHZreg x0:(MOVHBRload [i0] {s} p mem))
  1592      sh:(SLWconst [16] r1:(MOVHZreg x1:(MOVHBRload [i1] {s} p mem))))
  1593    && i1 == i0+2
  1594    && x0.Uses == 1
  1595    && x1.Uses == 1
  1596    && r0.Uses == 1
  1597    && r1.Uses == 1
  1598    && sh.Uses == 1
  1599    && mergePoint(b,x0,x1) != nil
  1600    && clobber(x0)
  1601    && clobber(x1)
  1602    && clobber(r0)
  1603    && clobber(r1)
  1604    && clobber(sh)
  1605    -> @mergePoint(b,x0,x1) (MOVWBRload [i0] {s} p mem)
  1606  
  1607  (OR                   r0:(MOVHZreg x0:(MOVHBRload [i0] {s} p mem))
  1608      sh:(SLDconst [16] r1:(MOVHZreg x1:(MOVHBRload [i1] {s} p mem))))
  1609    && i1 == i0+2
  1610    && x0.Uses == 1
  1611    && x1.Uses == 1
  1612    && r0.Uses == 1
  1613    && r1.Uses == 1
  1614    && sh.Uses == 1
  1615    && mergePoint(b,x0,x1) != nil
  1616    && clobber(x0)
  1617    && clobber(x1)
  1618    && clobber(r0)
  1619    && clobber(r1)
  1620    && clobber(sh)
  1621    -> @mergePoint(b,x0,x1) (MOVWZreg (MOVWBRload [i0] {s} p mem))
  1622  
  1623  (OR                   r0:(MOVWZreg x0:(MOVWBRload [i0] {s} p mem))
  1624      sh:(SLDconst [32] r1:(MOVWZreg x1:(MOVWBRload [i1] {s} p mem))))
  1625    && i1 == i0+4
  1626    && x0.Uses == 1
  1627    && x1.Uses == 1
  1628    && r0.Uses == 1
  1629    && r1.Uses == 1
  1630    && sh.Uses == 1
  1631    && mergePoint(b,x0,x1) != nil
  1632    && clobber(x0)
  1633    && clobber(x1)
  1634    && clobber(r0)
  1635    && clobber(r1)
  1636    && clobber(sh)
  1637    -> @mergePoint(b,x0,x1) (MOVDBRload [i0] {s} p mem)
  1638  
  1639  (ORW
  1640      s1:(SLWconst [j1] x1:(MOVBZload [i1] {s} p mem))
  1641      or:(ORW
  1642          s0:(SLWconst [j0] x0:(MOVBZload [i0] {s} p mem))
  1643  	y))
  1644    && p.Op != OpSB
  1645    && i1 == i0+1
  1646    && j1 == j0+8
  1647    && j0 % 16 == 0
  1648    && x0.Uses == 1
  1649    && x1.Uses == 1
  1650    && s0.Uses == 1
  1651    && s1.Uses == 1
  1652    && or.Uses == 1
  1653    && mergePoint(b,x0,x1) != nil
  1654    && clobber(x0)
  1655    && clobber(x1)
  1656    && clobber(s0)
  1657    && clobber(s1)
  1658    && clobber(or)
  1659    -> @mergePoint(b,x0,x1) (ORW <v.Type> (SLWconst <v.Type> [j0] (MOVHZreg (MOVHBRload [i0] {s} p mem))) y)
  1660  
  1661  (OR
  1662      s1:(SLDconst [j1] x1:(MOVBZload [i1] {s} p mem))
  1663      or:(OR
  1664          s0:(SLDconst [j0] x0:(MOVBZload [i0] {s} p mem))
  1665  	y))
  1666    && p.Op != OpSB
  1667    && i1 == i0+1
  1668    && j1 == j0+8
  1669    && j0 % 16 == 0
  1670    && x0.Uses == 1
  1671    && x1.Uses == 1
  1672    && s0.Uses == 1
  1673    && s1.Uses == 1
  1674    && or.Uses == 1
  1675    && mergePoint(b,x0,x1) != nil
  1676    && clobber(x0)
  1677    && clobber(x1)
  1678    && clobber(s0)
  1679    && clobber(s1)
  1680    && clobber(or)
  1681    -> @mergePoint(b,x0,x1) (OR <v.Type> (SLDconst <v.Type> [j0] (MOVHZreg (MOVHBRload [i0] {s} p mem))) y)
  1682  
  1683  (OR
  1684      s1:(SLDconst [j1] r1:(MOVHZreg x1:(MOVHBRload [i1] {s} p mem)))
  1685      or:(OR
  1686          s0:(SLDconst [j0] r0:(MOVHZreg x0:(MOVHBRload [i0] {s} p mem)))
  1687  	y))
  1688    && i1 == i0+2
  1689    && j1 == j0+16
  1690    && j0 % 32 == 0
  1691    && x0.Uses == 1
  1692    && x1.Uses == 1
  1693    && r0.Uses == 1
  1694    && r1.Uses == 1
  1695    && s0.Uses == 1
  1696    && s1.Uses == 1
  1697    && or.Uses == 1
  1698    && mergePoint(b,x0,x1) != nil
  1699    && clobber(x0)
  1700    && clobber(x1)
  1701    && clobber(r0)
  1702    && clobber(r1)
  1703    && clobber(s0)
  1704    && clobber(s1)
  1705    && clobber(or)
  1706    -> @mergePoint(b,x0,x1) (OR <v.Type> (SLDconst <v.Type> [j0] (MOVWZreg (MOVWBRload [i0] {s} p mem))) y)
  1707  
  1708  // Little-endian indexed loads
  1709  
  1710  (ORW                 x0:(MOVBZloadidx [i0] {s} p idx mem)
  1711      sh:(SLWconst [8] x1:(MOVBZloadidx [i1] {s} p idx mem)))
  1712    && p.Op != OpSB
  1713    && i1 == i0+1
  1714    && x0.Uses == 1
  1715    && x1.Uses == 1
  1716    && sh.Uses == 1
  1717    && mergePoint(b,x0,x1) != nil
  1718    && clobber(x0)
  1719    && clobber(x1)
  1720    && clobber(sh)
  1721    -> @mergePoint(b,x0,x1) (MOVHZreg (MOVHBRloadidx [i0] {s} p idx mem))
  1722  
  1723  (OR                  x0:(MOVBZloadidx [i0] {s} p idx mem)
  1724      sh:(SLDconst [8] x1:(MOVBZloadidx [i1] {s} p idx mem)))
  1725    && p.Op != OpSB
  1726    && i1 == i0+1
  1727    && x0.Uses == 1
  1728    && x1.Uses == 1
  1729    && sh.Uses == 1
  1730    && mergePoint(b,x0,x1) != nil
  1731    && clobber(x0)
  1732    && clobber(x1)
  1733    && clobber(sh)
  1734    -> @mergePoint(b,x0,x1) (MOVHZreg (MOVHBRloadidx [i0] {s} p idx mem))
  1735  
  1736  (ORW                  r0:(MOVHZreg x0:(MOVHBRloadidx [i0] {s} p idx mem))
  1737      sh:(SLWconst [16] r1:(MOVHZreg x1:(MOVHBRloadidx [i1] {s} p idx mem))))
  1738    && i1 == i0+2
  1739    && x0.Uses == 1
  1740    && x1.Uses == 1
  1741    && r0.Uses == 1
  1742    && r1.Uses == 1
  1743    && sh.Uses == 1
  1744    && mergePoint(b,x0,x1) != nil
  1745    && clobber(x0)
  1746    && clobber(x1)
  1747    && clobber(r0)
  1748    && clobber(r1)
  1749    && clobber(sh)
  1750    -> @mergePoint(b,x0,x1) (MOVWBRloadidx [i0] {s} p idx mem)
  1751  
  1752  (OR                   r0:(MOVHZreg x0:(MOVHBRloadidx [i0] {s} p idx mem))
  1753      sh:(SLDconst [16] r1:(MOVHZreg x1:(MOVHBRloadidx [i1] {s} p idx mem))))
  1754    && i1 == i0+2
  1755    && x0.Uses == 1
  1756    && x1.Uses == 1
  1757    && r0.Uses == 1
  1758    && r1.Uses == 1
  1759    && sh.Uses == 1
  1760    && mergePoint(b,x0,x1) != nil
  1761    && clobber(x0)
  1762    && clobber(x1)
  1763    && clobber(r0)
  1764    && clobber(r1)
  1765    && clobber(sh)
  1766    -> @mergePoint(b,x0,x1) (MOVWZreg (MOVWBRloadidx [i0] {s} p idx mem))
  1767  
  1768  (OR                   r0:(MOVWZreg x0:(MOVWBRloadidx [i0] {s} p idx mem))
  1769      sh:(SLDconst [32] r1:(MOVWZreg x1:(MOVWBRloadidx [i1] {s} p idx mem))))
  1770    && i1 == i0+4
  1771    && x0.Uses == 1
  1772    && x1.Uses == 1
  1773    && r0.Uses == 1
  1774    && r1.Uses == 1
  1775    && sh.Uses == 1
  1776    && mergePoint(b,x0,x1) != nil
  1777    && clobber(x0)
  1778    && clobber(x1)
  1779    && clobber(r0)
  1780    && clobber(r1)
  1781    && clobber(sh)
  1782    -> @mergePoint(b,x0,x1) (MOVDBRloadidx [i0] {s} p idx mem)
  1783  
  1784  (ORW
  1785      s1:(SLWconst [j1] x1:(MOVBZloadidx [i1] {s} p idx mem))
  1786      or:(ORW
  1787          s0:(SLWconst [j0] x0:(MOVBZloadidx [i0] {s} p idx mem))
  1788  	y))
  1789    && p.Op != OpSB
  1790    && i1 == i0+1
  1791    && j1 == j0+8
  1792    && j0 % 16 == 0
  1793    && x0.Uses == 1
  1794    && x1.Uses == 1
  1795    && s0.Uses == 1
  1796    && s1.Uses == 1
  1797    && or.Uses == 1
  1798    && mergePoint(b,x0,x1) != nil
  1799    && clobber(x0)
  1800    && clobber(x1)
  1801    && clobber(s0)
  1802    && clobber(s1)
  1803    && clobber(or)
  1804    -> @mergePoint(b,x0,x1) (ORW <v.Type> (SLWconst <v.Type> [j0] (MOVHZreg (MOVHBRloadidx [i0] {s} p idx mem))) y)
  1805  
  1806  (OR
  1807      s1:(SLDconst [j1] x1:(MOVBZloadidx [i1] {s} p idx mem))
  1808      or:(OR
  1809          s0:(SLDconst [j0] x0:(MOVBZloadidx [i0] {s} p idx mem))
  1810  	y))
  1811    && p.Op != OpSB
  1812    && i1 == i0+1
  1813    && j1 == j0+8
  1814    && j0 % 16 == 0
  1815    && x0.Uses == 1
  1816    && x1.Uses == 1
  1817    && s0.Uses == 1
  1818    && s1.Uses == 1
  1819    && or.Uses == 1
  1820    && mergePoint(b,x0,x1) != nil
  1821    && clobber(x0)
  1822    && clobber(x1)
  1823    && clobber(s0)
  1824    && clobber(s1)
  1825    && clobber(or)
  1826    -> @mergePoint(b,x0,x1) (OR <v.Type> (SLDconst <v.Type> [j0] (MOVHZreg (MOVHBRloadidx [i0] {s} p idx mem))) y)
  1827  
  1828  (OR
  1829      s1:(SLDconst [j1] r1:(MOVHZreg x1:(MOVHBRloadidx [i1] {s} p idx mem)))
  1830      or:(OR
  1831          s0:(SLDconst [j0] r0:(MOVHZreg x0:(MOVHBRloadidx [i0] {s} p idx mem)))
  1832  	y))
  1833    && i1 == i0+2
  1834    && j1 == j0+16
  1835    && j0 % 32 == 0
  1836    && x0.Uses == 1
  1837    && x1.Uses == 1
  1838    && r0.Uses == 1
  1839    && r1.Uses == 1
  1840    && s0.Uses == 1
  1841    && s1.Uses == 1
  1842    && or.Uses == 1
  1843    && mergePoint(b,x0,x1) != nil
  1844    && clobber(x0)
  1845    && clobber(x1)
  1846    && clobber(r0)
  1847    && clobber(r1)
  1848    && clobber(s0)
  1849    && clobber(s1)
  1850    && clobber(or)
  1851    -> @mergePoint(b,x0,x1) (OR <v.Type> (SLDconst <v.Type> [j0] (MOVWZreg (MOVWBRloadidx [i0] {s} p idx mem))) y)
  1852  
  1853  // Combine stores into store multiples.
  1854  // 32-bit
  1855  (MOVWstore [i] {s} p w1 x:(MOVWstore [i-4] {s} p w0 mem))
  1856    && p.Op != OpSB
  1857    && x.Uses == 1
  1858    && is20Bit(i-4)
  1859    && clobber(x)
  1860    -> (STM2 [i-4] {s} p w0 w1 mem)
  1861  (MOVWstore [i] {s} p w2 x:(STM2 [i-8] {s} p w0 w1 mem))
  1862    && x.Uses == 1
  1863    && is20Bit(i-8)
  1864    && clobber(x)
  1865    -> (STM3 [i-8] {s} p w0 w1 w2 mem)
  1866  (MOVWstore [i] {s} p w3 x:(STM3 [i-12] {s} p w0 w1 w2 mem))
  1867    && x.Uses == 1
  1868    && is20Bit(i-12)
  1869    && clobber(x)
  1870    -> (STM4 [i-12] {s} p w0 w1 w2 w3 mem)
  1871  (STM2 [i] {s} p w2 w3 x:(STM2 [i-8] {s} p w0 w1 mem))
  1872    && x.Uses == 1
  1873    && is20Bit(i-8)
  1874    && clobber(x)
  1875    -> (STM4 [i-8] {s} p w0 w1 w2 w3 mem)
  1876  // 64-bit
  1877  (MOVDstore [i] {s} p w1 x:(MOVDstore [i-8] {s} p w0 mem))
  1878    && p.Op != OpSB
  1879    && x.Uses == 1
  1880    && is20Bit(i-8)
  1881    && clobber(x)
  1882    -> (STMG2 [i-8] {s} p w0 w1 mem)
  1883  (MOVDstore [i] {s} p w2 x:(STMG2 [i-16] {s} p w0 w1 mem))
  1884    && x.Uses == 1
  1885    && is20Bit(i-16)
  1886    && clobber(x)
  1887    -> (STMG3 [i-16] {s} p w0 w1 w2 mem)
  1888  (MOVDstore [i] {s} p w3 x:(STMG3 [i-24] {s} p w0 w1 w2 mem))
  1889    && x.Uses == 1
  1890    && is20Bit(i-24)
  1891    && clobber(x)
  1892    -> (STMG4 [i-24] {s} p w0 w1 w2 w3 mem)
  1893  (STMG2 [i] {s} p w2 w3 x:(STMG2 [i-16] {s} p w0 w1 mem))
  1894    && x.Uses == 1
  1895    && is20Bit(i-16)
  1896    && clobber(x)
  1897    -> (STMG4 [i-16] {s} p w0 w1 w2 w3 mem)
  1898  
  1899  // Convert 32-bit store multiples into 64-bit stores.
  1900  (STM2 [i] {s} p (SRDconst [32] x) x mem) -> (MOVDstore [i] {s} p x mem)