Text file src/cmd/compile/internal/ssa/_gen/S390X.rules

     1  // Copyright 2016 The Go Authors. All rights reserved.
     2  // Use of this source code is governed by a BSD-style
     3  // license that can be found in the LICENSE file.
     4  
     5  // Lowering arithmetic
     6  (Add(64|Ptr) ...) => (ADD ...)
     7  (Add(32|16|8) ...) => (ADDW ...)
     8  (Add32F x y) => (Select0 (FADDS x y))
     9  (Add64F x y) => (Select0 (FADD x y))
    10  
    11  (Sub(64|Ptr) ...) => (SUB ...)
    12  (Sub(32|16|8) ...) => (SUBW ...)
    13  (Sub32F x y) => (Select0 (FSUBS x y))
    14  (Sub64F x y) => (Select0 (FSUB x y))
    15  
    16  (Mul64 ...) => (MULLD ...)
    17  (Mul(32|16|8) ...) => (MULLW ...)
    18  (Mul32F ...) => (FMULS ...)
    19  (Mul64F ...) => (FMUL ...)
    20  (Mul64uhilo ...) => (MLGR ...)
    21  
    22  (Div32F ...) => (FDIVS ...)
    23  (Div64F ...) => (FDIV ...)
    24  
    25  (Div64 x y) => (DIVD x y)
    26  (Div64u ...) => (DIVDU ...)
    27  // DIVW/DIVWU has a 64-bit dividend and a 32-bit divisor,
    28  // so a sign/zero extension of the dividend is required.
    29  (Div32  x y) => (DIVW  (MOVWreg x) y)
    30  (Div32u x y) => (DIVWU (MOVWZreg x) y)
    31  (Div16  x y) => (DIVW  (MOVHreg x) (MOVHreg y))
    32  (Div16u x y) => (DIVWU (MOVHZreg x) (MOVHZreg y))
    33  (Div8   x y) => (DIVW  (MOVBreg x) (MOVBreg y))
    34  (Div8u  x y) => (DIVWU (MOVBZreg x) (MOVBZreg y))
    35  
    36  (Hmul(64|64u) ...) => (MULH(D|DU) ...)
    37  (Hmul32  x y) => (SRDconst [32] (MULLD (MOVWreg x) (MOVWreg y)))
    38  (Hmul32u x y) => (SRDconst [32] (MULLD (MOVWZreg x) (MOVWZreg y)))
    39  
    40  (Mod64 x y) => (MODD x y)
    41  (Mod64u ...) => (MODDU ...)
    42  // MODW/MODWU has a 64-bit dividend and a 32-bit divisor,
    43  // so a sign/zero extension of the dividend is required.
    44  (Mod32  x y) => (MODW  (MOVWreg x) y)
    45  (Mod32u x y) => (MODWU (MOVWZreg x) y)
    46  (Mod16  x y) => (MODW  (MOVHreg x) (MOVHreg y))
    47  (Mod16u x y) => (MODWU (MOVHZreg x) (MOVHZreg y))
    48  (Mod8   x y) => (MODW  (MOVBreg x) (MOVBreg y))
    49  (Mod8u  x y) => (MODWU (MOVBZreg x) (MOVBZreg y))
    50  
    51  // (x + y) / 2 with x>=y -> (x - y) / 2 + y
    52  (Avg64u <t> x y) => (ADD (SRDconst <t> (SUB <t> x y) [1]) y)
    53  
    54  (And64 ...) => (AND ...)
    55  (And(32|16|8) ...) => (ANDW ...)
    56  
    57  (Or64 ...) => (OR ...)
    58  (Or(32|16|8) ...) => (ORW ...)
    59  
    60  (Xor64 ...) => (XOR ...)
    61  (Xor(32|16|8) ...) => (XORW ...)
    62  
    63  (Neg64 ...) => (NEG ...)
    64  (Neg(32|16|8) ...) => (NEGW ...)
    65  (Neg32F ...) => (FNEGS ...)
    66  (Neg64F ...) => (FNEG ...)
    67  
    68  (Com64 ...) => (NOT ...)
    69  (Com(32|16|8) ...) => (NOTW ...)
    70  (NOT x) => (XOR (MOVDconst [-1]) x)
    71  (NOTW x) => (XORWconst [-1] x)
    72  
    73  // Lowering boolean ops
    74  (AndB ...) => (ANDW ...)
    75  (OrB ...) => (ORW ...)
    76  (Not x) => (XORWconst [1] x)
    77  
    78  // Lowering pointer arithmetic
    79  (OffPtr [off] ptr:(SP)) => (MOVDaddr [int32(off)] ptr)
    80  (OffPtr [off] ptr) && is32Bit(off) => (ADDconst [int32(off)] ptr)
    81  (OffPtr [off] ptr) => (ADD (MOVDconst [off]) ptr)
    82  
    83  (Ctz(64|32|16|8)NonZero ...) => (Ctz64 ...)
    84  
    85  // Ctz(x) = 64 - findLeftmostOne((x-1)&^x)
    86  (Ctz64 <t> x) => (SUB (MOVDconst [64]) (FLOGR (AND <t> (SUBconst <t> [1] x) (NOT <t> x))))
    87  (Ctz32 <t> x) => (SUB (MOVDconst [64]) (FLOGR (MOVWZreg (ANDW <t> (SUBWconst <t> [1] x) (NOTW <t> x)))))
    88  (Ctz16 x) => (Ctz64 (Or64 <typ.UInt64> x (MOVDconst [1<<16])))
    89  (Ctz8  x) => (Ctz64 (Or64 <typ.UInt64> x (MOVDconst [1<<8])))
    90  
    91  (BitLen64 x) => (SUB (MOVDconst [64]) (FLOGR x))
    92  (BitLen(32|16|8) x) => (BitLen64 (ZeroExt(32|16|8)to64 x))
    93  
    94  // POPCNT treats the input register as a vector of 8 bytes, producing
    95  // a population count for each individual byte. For inputs larger than
    96  // a single byte we therefore need to sum the individual bytes produced
    97  // by the POPCNT instruction. For example, the following instruction
    98  // sequence could be used to calculate the population count of a 4-byte
    99  // value:
   100  //
   101  //     MOVD   $0x12345678, R1 // R1=0x12345678 <-- input
   102  //     POPCNT R1, R2          // R2=0x02030404
   103  //     SRW    $16, R2, R3     // R3=0x00000203
   104  //     ADDW   R2, R3, R4      // R4=0x02030607
   105  //     SRW    $8, R4, R5      // R5=0x00020306
   106  //     ADDW   R4, R5, R6      // R6=0x0205090d
   107  //     MOVBZ  R6, R7          // R7=0x0000000d <-- result is 13
   108  //
   109  (PopCount8  x) => (POPCNT (MOVBZreg x))
   110  (PopCount16 x) => (MOVBZreg (SumBytes2 (POPCNT <typ.UInt16> x)))
   111  (PopCount32 x) => (MOVBZreg (SumBytes4 (POPCNT <typ.UInt32> x)))
   112  (PopCount64 x) => (MOVBZreg (SumBytes8 (POPCNT <typ.UInt64> x)))
   113  
   114  // SumBytes{2,4,8} pseudo operations sum the values of the rightmost
   115  // 2, 4 or 8 bytes respectively. The result is a single byte however
   116  // other bytes might contain junk so a zero extension is required if
   117  // the desired output type is larger than 1 byte.
   118  (SumBytes2 x) => (ADDW (SRWconst <typ.UInt8> x [8]) x)
   119  (SumBytes4 x) => (SumBytes2 (ADDW <typ.UInt16> (SRWconst <typ.UInt16> x [16]) x))
   120  (SumBytes8 x) => (SumBytes4 (ADDW <typ.UInt32> (SRDconst <typ.UInt32> x [32]) x))
   121  
   122  (Bswap64 ...) => (MOVDBR ...)
   123  (Bswap32 ...) => (MOVWBR ...)
   124  
   125  // add with carry
   126  (Select0 (Add64carry x y c))
   127    => (Select0 <typ.UInt64> (ADDE x y (Select1 <types.TypeFlags> (ADDCconst c [-1]))))
   128  (Select1 (Add64carry x y c))
   129    => (Select0 <typ.UInt64> (ADDE (MOVDconst [0]) (MOVDconst [0]) (Select1 <types.TypeFlags> (ADDE x y (Select1 <types.TypeFlags> (ADDCconst c [-1]))))))
   130  
   131  // subtract with borrow
   132  (Select0 (Sub64borrow x y c))
   133    => (Select0 <typ.UInt64> (SUBE x y (Select1 <types.TypeFlags> (SUBC (MOVDconst [0]) c))))
   134  (Select1 (Sub64borrow x y c))
   135    => (NEG (Select0 <typ.UInt64> (SUBE (MOVDconst [0]) (MOVDconst [0]) (Select1 <types.TypeFlags> (SUBE x y (Select1 <types.TypeFlags> (SUBC (MOVDconst [0]) c)))))))
   136  
   137  // math package intrinsics
   138  (Sqrt      ...) => (FSQRT ...)
   139  (Floor       x) => (FIDBR [7] x)
   140  (Ceil        x) => (FIDBR [6] x)
   141  (Trunc       x) => (FIDBR [5] x)
   142  (RoundToEven x) => (FIDBR [4] x)
   143  (Round       x) => (FIDBR [1] x)
   144  (FMA     x y z) => (FMADD z x y)
   145  
   146  (Sqrt32    ...) => (FSQRTS ...)
   147  
   148  // Atomic loads and stores.
   149  // The SYNC instruction (fast-BCR-serialization) prevents store-load
   150  // reordering. Other sequences of memory operations (load-load,
   151  // store-store and load-store) are already guaranteed not to be reordered.
   152  (AtomicLoad(8|32|Acq32|64|Ptr) ptr mem) => (MOV(BZ|WZ|WZ|D|D)atomicload ptr mem)
   153  (AtomicStore(8|32|64|PtrNoWB) ptr val mem) => (SYNC (MOV(B|W|D|D)atomicstore ptr val mem))
   154  
   155  // Store-release doesn't require store-load ordering.
   156  (AtomicStoreRel32 ptr val mem) => (MOVWatomicstore ptr val mem)
   157  
   158  // Atomic adds.
   159  (AtomicAdd32 ptr val mem) => (AddTupleFirst32 val (LAA ptr val mem))
   160  (AtomicAdd64 ptr val mem) => (AddTupleFirst64 val (LAAG ptr val mem))
   161  (Select0 <t> (AddTupleFirst32 val tuple)) => (ADDW val (Select0 <t> tuple))
   162  (Select1     (AddTupleFirst32   _ tuple)) => (Select1 tuple)
   163  (Select0 <t> (AddTupleFirst64 val tuple)) => (ADD val (Select0 <t> tuple))
   164  (Select1     (AddTupleFirst64   _ tuple)) => (Select1 tuple)
   165  
   166  // Atomic exchanges.
   167  (AtomicExchange32 ptr val mem) => (LoweredAtomicExchange32 ptr val mem)
   168  (AtomicExchange64 ptr val mem) => (LoweredAtomicExchange64 ptr val mem)
   169  
   170  // Atomic compare and swap.
   171  (AtomicCompareAndSwap32 ptr old new_ mem) => (LoweredAtomicCas32 ptr old new_ mem)
   172  (AtomicCompareAndSwap64 ptr old new_ mem) => (LoweredAtomicCas64 ptr old new_ mem)
   173  
   174  // Atomic and: *(*uint8)(ptr) &= val
   175  //
   176  // Round pointer down to nearest word boundary and pad value with ones before
   177  // applying atomic AND operation to target word.
   178  //
   179  // *(*uint32)(ptr &^ 3) &= rotateleft(uint32(val) | 0xffffff00, ((3 << 3) ^ ((ptr & 3) << 3))
   180  //
   181  (AtomicAnd8 ptr val mem)
   182    => (LANfloor
   183         ptr
   184         (RLL <typ.UInt32>
   185           (ORWconst <typ.UInt32> val [-1<<8])
   186           (RXSBG <typ.UInt32> {s390x.NewRotateParams(59, 60, 3)} (MOVDconst [3<<3]) ptr))
   187         mem)
   188  
   189  // Atomic or: *(*uint8)(ptr) |= val
   190  //
   191  // Round pointer down to nearest word boundary and pad value with zeros before
   192  // applying atomic OR operation to target word.
   193  //
   194  // *(*uint32)(ptr &^ 3) |= uint32(val) << ((3 << 3) ^ ((ptr & 3) << 3))
   195  //
   196  (AtomicOr8  ptr val mem)
   197    => (LAOfloor
   198         ptr
   199         (SLW <typ.UInt32>
   200           (MOVBZreg <typ.UInt32> val)
   201           (RXSBG <typ.UInt32> {s390x.NewRotateParams(59, 60, 3)} (MOVDconst [3<<3]) ptr))
   202         mem)
   203  
   204  (AtomicAnd32 ...) => (LAN ...)
   205  (AtomicOr32  ...) => (LAO ...)
   206  
   207  // Lowering extension
   208  // Note: we always extend to 64 bits even though some ops don't need that many result bits.
   209  (SignExt8to(16|32|64) ...) => (MOVBreg ...)
   210  (SignExt16to(32|64) ...) => (MOVHreg ...)
   211  (SignExt32to64 ...) => (MOVWreg ...)
   212  
   213  (ZeroExt8to(16|32|64) ...) => (MOVBZreg ...)
   214  (ZeroExt16to(32|64) ...) => (MOVHZreg ...)
   215  (ZeroExt32to64 ...) => (MOVWZreg ...)
   216  
   217  (Slicemask <t> x) => (SRADconst (NEG <t> x) [63])
   218  
   219  // Lowering truncation
   220  // Because we ignore high parts of registers, truncates are just copies.
   221  (Trunc(16|32|64)to8 ...) => (Copy ...)
   222  (Trunc(32|64)to16 ...) => (Copy ...)
   223  (Trunc64to32 ...) => (Copy ...)
   224  
   225  // Lowering float <-> int
   226  (Cvt32to32F ...) => (CEFBRA ...)
   227  (Cvt32to64F ...) => (CDFBRA ...)
   228  (Cvt64to32F ...) => (CEGBRA ...)
   229  (Cvt64to64F ...) => (CDGBRA ...)
   230  
   231  (Cvt32Fto32 ...) => (CFEBRA ...)
   232  (Cvt32Fto64 ...) => (CGEBRA ...)
   233  (Cvt64Fto32 ...) => (CFDBRA ...)
   234  (Cvt64Fto64 ...) => (CGDBRA ...)
   235  
   236  // Lowering float <-> uint
   237  (Cvt32Uto32F ...) => (CELFBR ...)
   238  (Cvt32Uto64F ...) => (CDLFBR ...)
   239  (Cvt64Uto32F ...) => (CELGBR ...)
   240  (Cvt64Uto64F ...) => (CDLGBR ...)
   241  
   242  (Cvt32Fto32U ...) => (CLFEBR ...)
   243  (Cvt32Fto64U ...) => (CLGEBR ...)
   244  (Cvt64Fto32U ...) => (CLFDBR ...)
   245  (Cvt64Fto64U ...) => (CLGDBR ...)
   246  
   247  // Lowering float32 <-> float64
   248  (Cvt32Fto64F ...) => (LDEBR ...)
   249  (Cvt64Fto32F ...) => (LEDBR ...)
   250  
   251  (CvtBoolToUint8 ...) => (Copy ...)
   252  
   253  (Round(32|64)F ...) => (LoweredRound(32|64)F ...)
   254  
   255  // Lowering shifts
   256  
   257  // Lower bounded shifts first. No need to check shift value.
   258  (Lsh64x(64|32|16|8)  x y) && shiftIsBounded(v) => (SLD x y)
   259  (Lsh32x(64|32|16|8)  x y) && shiftIsBounded(v) => (SLW x y)
   260  (Lsh16x(64|32|16|8)  x y) && shiftIsBounded(v) => (SLW x y)
   261  (Lsh8x(64|32|16|8)   x y) && shiftIsBounded(v) => (SLW x y)
   262  (Rsh64Ux(64|32|16|8) x y) && shiftIsBounded(v) => (SRD x y)
   263  (Rsh32Ux(64|32|16|8) x y) && shiftIsBounded(v) => (SRW x y)
   264  (Rsh16Ux(64|32|16|8) x y) && shiftIsBounded(v) => (SRW (MOVHZreg x) y)
   265  (Rsh8Ux(64|32|16|8)  x y) && shiftIsBounded(v) => (SRW (MOVBZreg x) y)
   266  (Rsh64x(64|32|16|8)  x y) && shiftIsBounded(v) => (SRAD x y)
   267  (Rsh32x(64|32|16|8)  x y) && shiftIsBounded(v) => (SRAW x y)
   268  (Rsh16x(64|32|16|8)  x y) && shiftIsBounded(v) => (SRAW (MOVHreg x) y)
   269  (Rsh8x(64|32|16|8)   x y) && shiftIsBounded(v) => (SRAW (MOVBreg x) y)
   270  
   271  // Unsigned shifts need to return 0 if shift amount is >= width of shifted value.
   272  //   result = shift >= 64 ? 0 : arg << shift
   273  (Lsh(64|32|16|8)x64 <t> x y) => (LOCGR {s390x.GreaterOrEqual} <t> (SL(D|W|W|W) <t> x y) (MOVDconst [0]) (CMPUconst y [64]))
   274  (Lsh(64|32|16|8)x32 <t> x y) => (LOCGR {s390x.GreaterOrEqual} <t> (SL(D|W|W|W) <t> x y) (MOVDconst [0]) (CMPWUconst y [64]))
   275  (Lsh(64|32|16|8)x16 <t> x y) => (LOCGR {s390x.GreaterOrEqual} <t> (SL(D|W|W|W) <t> x y) (MOVDconst [0]) (CMPWUconst (MOVHZreg y) [64]))
   276  (Lsh(64|32|16|8)x8  <t> x y) => (LOCGR {s390x.GreaterOrEqual} <t> (SL(D|W|W|W) <t> x y) (MOVDconst [0]) (CMPWUconst (MOVBZreg y) [64]))
   277  
   278  (Rsh(64|32)Ux64 <t> x y) => (LOCGR {s390x.GreaterOrEqual} <t> (SR(D|W) <t> x y) (MOVDconst [0]) (CMPUconst y [64]))
   279  (Rsh(64|32)Ux32 <t> x y) => (LOCGR {s390x.GreaterOrEqual} <t> (SR(D|W) <t> x y) (MOVDconst [0]) (CMPWUconst y [64]))
   280  (Rsh(64|32)Ux16 <t> x y) => (LOCGR {s390x.GreaterOrEqual} <t> (SR(D|W) <t> x y) (MOVDconst [0]) (CMPWUconst (MOVHZreg y) [64]))
   281  (Rsh(64|32)Ux8  <t> x y) => (LOCGR {s390x.GreaterOrEqual} <t> (SR(D|W) <t> x y) (MOVDconst [0]) (CMPWUconst (MOVBZreg y) [64]))
   282  
   283  (Rsh(16|8)Ux64 <t> x y) => (LOCGR {s390x.GreaterOrEqual} <t> (SRW <t> (MOV(H|B)Zreg x) y) (MOVDconst [0]) (CMPUconst y [64]))
   284  (Rsh(16|8)Ux32 <t> x y) => (LOCGR {s390x.GreaterOrEqual} <t> (SRW <t> (MOV(H|B)Zreg x) y) (MOVDconst [0]) (CMPWUconst y [64]))
   285  (Rsh(16|8)Ux16 <t> x y) => (LOCGR {s390x.GreaterOrEqual} <t> (SRW <t> (MOV(H|B)Zreg x) y) (MOVDconst [0]) (CMPWUconst (MOVHZreg y) [64]))
   286  (Rsh(16|8)Ux8  <t> x y) => (LOCGR {s390x.GreaterOrEqual} <t> (SRW <t> (MOV(H|B)Zreg x) y) (MOVDconst [0]) (CMPWUconst (MOVBZreg y) [64]))
   287  
   288  // Signed right shift needs to return 0/-1 if shift amount is >= width of shifted value.
   289  // We implement this by setting the shift value to 63 (all ones) if the shift value is more than 63.
   290  //   result = arg >> (shift >= 64 ? 63 : shift)
   291  (Rsh(64|32)x64 x y) => (SRA(D|W) x (LOCGR {s390x.GreaterOrEqual} <y.Type> y (MOVDconst <y.Type> [63]) (CMPUconst  y [64])))
   292  (Rsh(64|32)x32 x y) => (SRA(D|W) x (LOCGR {s390x.GreaterOrEqual} <y.Type> y (MOVDconst <y.Type> [63]) (CMPWUconst y [64])))
   293  (Rsh(64|32)x16 x y) => (SRA(D|W) x (LOCGR {s390x.GreaterOrEqual} <y.Type> y (MOVDconst <y.Type> [63]) (CMPWUconst (MOVHZreg y) [64])))
   294  (Rsh(64|32)x8  x y) => (SRA(D|W) x (LOCGR {s390x.GreaterOrEqual} <y.Type> y (MOVDconst <y.Type> [63]) (CMPWUconst (MOVBZreg y) [64])))
   295  
   296  (Rsh(16|8)x64 x y) => (SRAW (MOV(H|B)reg x) (LOCGR {s390x.GreaterOrEqual} <y.Type> y (MOVDconst <y.Type> [63]) (CMPUconst  y [64])))
   297  (Rsh(16|8)x32 x y) => (SRAW (MOV(H|B)reg x) (LOCGR {s390x.GreaterOrEqual} <y.Type> y (MOVDconst <y.Type> [63]) (CMPWUconst y [64])))
   298  (Rsh(16|8)x16 x y) => (SRAW (MOV(H|B)reg x) (LOCGR {s390x.GreaterOrEqual} <y.Type> y (MOVDconst <y.Type> [63]) (CMPWUconst (MOVHZreg y) [64])))
   299  (Rsh(16|8)x8  x y) => (SRAW (MOV(H|B)reg x) (LOCGR {s390x.GreaterOrEqual} <y.Type> y (MOVDconst <y.Type> [63]) (CMPWUconst (MOVBZreg y) [64])))
   300  
   301  // Lowering rotates
   302  (RotateLeft8 <t> x (MOVDconst [c])) => (Or8 (Lsh8x64 <t> x (MOVDconst [c&7])) (Rsh8Ux64 <t> x (MOVDconst [-c&7])))
   303  (RotateLeft16 <t> x (MOVDconst [c])) => (Or16 (Lsh16x64 <t> x (MOVDconst [c&15])) (Rsh16Ux64 <t> x (MOVDconst [-c&15])))
   304  (RotateLeft32 ...) => (RLL  ...)
   305  (RotateLeft64 ...) => (RLLG ...)
   306  
   307  // Lowering comparisons
   308  (Less64      x y) => (LOCGR {s390x.Less} (MOVDconst [0]) (MOVDconst [1]) (CMP x y))
   309  (Less32      x y) => (LOCGR {s390x.Less} (MOVDconst [0]) (MOVDconst [1]) (CMPW x y))
   310  (Less(16|8)  x y) => (LOCGR {s390x.Less} (MOVDconst [0]) (MOVDconst [1]) (CMPW (MOV(H|B)reg x) (MOV(H|B)reg y)))
   311  (Less64U     x y) => (LOCGR {s390x.Less} (MOVDconst [0]) (MOVDconst [1]) (CMPU x y))
   312  (Less32U     x y) => (LOCGR {s390x.Less} (MOVDconst [0]) (MOVDconst [1]) (CMPWU x y))
   313  (Less(16|8)U x y) => (LOCGR {s390x.Less} (MOVDconst [0]) (MOVDconst [1]) (CMPWU (MOV(H|B)Zreg x) (MOV(H|B)Zreg y)))
   314  (Less64F     x y) => (LOCGR {s390x.Less} (MOVDconst [0]) (MOVDconst [1]) (FCMP x y))
   315  (Less32F     x y) => (LOCGR {s390x.Less} (MOVDconst [0]) (MOVDconst [1]) (FCMPS x y))
   316  
   317  (Leq64      x y) => (LOCGR {s390x.LessOrEqual} (MOVDconst [0]) (MOVDconst [1]) (CMP x y))
   318  (Leq32      x y) => (LOCGR {s390x.LessOrEqual} (MOVDconst [0]) (MOVDconst [1]) (CMPW x y))
   319  (Leq(16|8)  x y) => (LOCGR {s390x.LessOrEqual} (MOVDconst [0]) (MOVDconst [1]) (CMPW (MOV(H|B)reg x) (MOV(H|B)reg y)))
   320  (Leq64U     x y) => (LOCGR {s390x.LessOrEqual} (MOVDconst [0]) (MOVDconst [1]) (CMPU x y))
   321  (Leq32U     x y) => (LOCGR {s390x.LessOrEqual} (MOVDconst [0]) (MOVDconst [1]) (CMPWU x y))
   322  (Leq(16|8)U x y) => (LOCGR {s390x.LessOrEqual} (MOVDconst [0]) (MOVDconst [1]) (CMPWU (MOV(H|B)Zreg x) (MOV(H|B)Zreg y)))
   323  (Leq64F     x y) => (LOCGR {s390x.LessOrEqual} (MOVDconst [0]) (MOVDconst [1]) (FCMP x y))
   324  (Leq32F     x y) => (LOCGR {s390x.LessOrEqual} (MOVDconst [0]) (MOVDconst [1]) (FCMPS x y))
   325  
   326  (Eq(64|Ptr) x y) => (LOCGR {s390x.Equal} (MOVDconst [0]) (MOVDconst [1]) (CMP x y))
   327  (Eq32       x y) => (LOCGR {s390x.Equal} (MOVDconst [0]) (MOVDconst [1]) (CMPW x y))
   328  (Eq(16|8|B) x y) => (LOCGR {s390x.Equal} (MOVDconst [0]) (MOVDconst [1]) (CMPW (MOV(H|B|B)reg x) (MOV(H|B|B)reg y)))
   329  (Eq64F      x y) => (LOCGR {s390x.Equal} (MOVDconst [0]) (MOVDconst [1]) (FCMP x y))
   330  (Eq32F      x y) => (LOCGR {s390x.Equal} (MOVDconst [0]) (MOVDconst [1]) (FCMPS x y))
   331  
   332  (Neq(64|Ptr) x y) => (LOCGR {s390x.NotEqual} (MOVDconst [0]) (MOVDconst [1]) (CMP x y))
   333  (Neq32       x y) => (LOCGR {s390x.NotEqual} (MOVDconst [0]) (MOVDconst [1]) (CMPW x y))
   334  (Neq(16|8|B) x y) => (LOCGR {s390x.NotEqual} (MOVDconst [0]) (MOVDconst [1]) (CMPW (MOV(H|B|B)reg x) (MOV(H|B|B)reg y)))
   335  (Neq64F      x y) => (LOCGR {s390x.NotEqual} (MOVDconst [0]) (MOVDconst [1]) (FCMP x y))
   336  (Neq32F      x y) => (LOCGR {s390x.NotEqual} (MOVDconst [0]) (MOVDconst [1]) (FCMPS x y))
   337  
   338  // Lowering loads
   339  (Load <t> ptr mem) && (is64BitInt(t) || isPtr(t)) => (MOVDload ptr mem)
   340  (Load <t> ptr mem) && is32BitInt(t) &&  t.IsSigned() => (MOVWload ptr mem)
   341  (Load <t> ptr mem) && is32BitInt(t) && !t.IsSigned() => (MOVWZload ptr mem)
   342  (Load <t> ptr mem) && is16BitInt(t) &&  t.IsSigned() => (MOVHload ptr mem)
   343  (Load <t> ptr mem) && is16BitInt(t) && !t.IsSigned() => (MOVHZload ptr mem)
   344  (Load <t> ptr mem) && is8BitInt(t)  &&  t.IsSigned() => (MOVBload ptr mem)
   345  (Load <t> ptr mem) && (t.IsBoolean() || (is8BitInt(t) && !t.IsSigned())) => (MOVBZload ptr mem)
   346  (Load <t> ptr mem) && is32BitFloat(t) => (FMOVSload ptr mem)
   347  (Load <t> ptr mem) && is64BitFloat(t) => (FMOVDload ptr mem)
   348  
   349  // Lowering stores
   350  (Store {t} ptr val mem) && t.Size() == 8 &&  t.IsFloat() => (FMOVDstore ptr val mem)
   351  (Store {t} ptr val mem) && t.Size() == 4 &&  t.IsFloat() => (FMOVSstore ptr val mem)
   352  (Store {t} ptr val mem) && t.Size() == 8 && !t.IsFloat() => (MOVDstore ptr val mem)
   353  (Store {t} ptr val mem) && t.Size() == 4 && !t.IsFloat() => (MOVWstore ptr val mem)
   354  (Store {t} ptr val mem) && t.Size() == 2 => (MOVHstore ptr val mem)
   355  (Store {t} ptr val mem) && t.Size() == 1 => (MOVBstore ptr val mem)
   356  
   357  // Lowering moves
   358  
   359  // Load and store for small copies.
   360  (Move [0] _ _ mem) => mem
   361  (Move [1] dst src mem) => (MOVBstore dst (MOVBZload src mem) mem)
   362  (Move [2] dst src mem) => (MOVHstore dst (MOVHZload src mem) mem)
   363  (Move [4] dst src mem) => (MOVWstore dst (MOVWZload src mem) mem)
   364  (Move [8] dst src mem) => (MOVDstore dst (MOVDload src mem) mem)
   365  (Move [16] dst src mem) =>
   366  	(MOVDstore [8] dst (MOVDload [8] src mem)
   367  		(MOVDstore dst (MOVDload src mem) mem))
   368  (Move [24] dst src mem) =>
   369          (MOVDstore [16] dst (MOVDload [16] src mem)
   370  	        (MOVDstore [8] dst (MOVDload [8] src mem)
   371                  (MOVDstore dst (MOVDload src mem) mem)))
   372  (Move [3] dst src mem) =>
   373  	(MOVBstore [2] dst (MOVBZload [2] src mem)
   374  		(MOVHstore dst (MOVHZload src mem) mem))
   375  (Move [5] dst src mem) =>
   376  	(MOVBstore [4] dst (MOVBZload [4] src mem)
   377  		(MOVWstore dst (MOVWZload src mem) mem))
   378  (Move [6] dst src mem) =>
   379  	(MOVHstore [4] dst (MOVHZload [4] src mem)
   380  		(MOVWstore dst (MOVWZload src mem) mem))
   381  (Move [7] dst src mem) =>
   382  	(MOVBstore [6] dst (MOVBZload [6] src mem)
   383  		(MOVHstore [4] dst (MOVHZload [4] src mem)
   384  			(MOVWstore dst (MOVWZload src mem) mem)))
   385  
   386  // MVC for other moves. Use up to 4 instructions (sizes up to 1024 bytes).
   387  (Move [s] dst src mem) && s > 0 && s <= 256 && logLargeCopy(v, s) =>
   388  	(MVC [makeValAndOff(int32(s), 0)] dst src mem)
   389  (Move [s] dst src mem) && s > 256 && s <= 512 && logLargeCopy(v, s) =>
   390  	(MVC [makeValAndOff(int32(s)-256, 256)] dst src (MVC [makeValAndOff(256, 0)] dst src mem))
   391  (Move [s] dst src mem) && s > 512 && s <= 768 && logLargeCopy(v, s) =>
   392  	(MVC [makeValAndOff(int32(s)-512, 512)] dst src (MVC [makeValAndOff(256, 256)] dst src (MVC [makeValAndOff(256, 0)] dst src mem)))
   393  (Move [s] dst src mem) && s > 768 && s <= 1024 && logLargeCopy(v, s) =>
   394  	(MVC [makeValAndOff(int32(s)-768, 768)] dst src (MVC [makeValAndOff(256, 512)] dst src (MVC [makeValAndOff(256, 256)] dst src (MVC [makeValAndOff(256, 0)] dst src mem))))
   395  
   396  // Move more than 1024 bytes using a loop.
   397  (Move [s] dst src mem) && s > 1024 && logLargeCopy(v, s) =>
   398  	(LoweredMove [s%256] dst src (ADD <src.Type> src (MOVDconst [(s/256)*256])) mem)
   399  
   400  // Lowering Zero instructions
   401  (Zero [0] _ mem) => mem
   402  (Zero [1] destptr mem) => (MOVBstoreconst [0] destptr mem)
   403  (Zero [2] destptr mem) => (MOVHstoreconst [0] destptr mem)
   404  (Zero [4] destptr mem) => (MOVWstoreconst [0] destptr mem)
   405  (Zero [8] destptr mem) => (MOVDstoreconst [0] destptr mem)
   406  (Zero [3] destptr mem) =>
   407  	(MOVBstoreconst [makeValAndOff(0,2)] destptr
   408  		(MOVHstoreconst [0] destptr mem))
   409  (Zero [5] destptr mem) =>
   410  	(MOVBstoreconst [makeValAndOff(0,4)] destptr
   411  		(MOVWstoreconst [0] destptr mem))
   412  (Zero [6] destptr mem) =>
   413  	(MOVHstoreconst [makeValAndOff(0,4)] destptr
   414  		(MOVWstoreconst [0] destptr mem))
   415  (Zero [7] destptr mem) =>
   416  	(MOVWstoreconst [makeValAndOff(0,3)] destptr
   417  		(MOVWstoreconst [0] destptr mem))
   418  
   419  (Zero [s] destptr mem) && s > 0 && s <= 1024 =>
   420  	(CLEAR [makeValAndOff(int32(s), 0)] destptr mem)
   421  
   422  // Zero more than 1024 bytes using a loop.
   423  (Zero [s] destptr mem) && s > 1024 =>
   424  	(LoweredZero [s%256] destptr (ADDconst <destptr.Type> destptr [(int32(s)/256)*256]) mem)
   425  
   426  // Lowering constants
   427  (Const(64|32|16|8) [val]) => (MOVDconst [int64(val)])
   428  (Const(32|64)F ...) => (FMOV(S|D)const ...)
   429  (ConstNil) => (MOVDconst [0])
   430  (ConstBool [t]) => (MOVDconst [b2i(t)])
   431  
   432  // Lowering calls
   433  (StaticCall ...) => (CALLstatic ...)
   434  (ClosureCall ...) => (CALLclosure ...)
   435  (InterCall ...) => (CALLinter ...)
   436  (TailCall ...) => (CALLtail ...)
   437  
   438  // Miscellaneous
   439  (IsNonNil p) => (LOCGR {s390x.NotEqual} (MOVDconst [0]) (MOVDconst [1]) (CMPconst p [0]))
   440  (IsInBounds idx len) => (LOCGR {s390x.Less} (MOVDconst [0]) (MOVDconst [1]) (CMPU idx len))
   441  (IsSliceInBounds idx len) => (LOCGR {s390x.LessOrEqual} (MOVDconst [0]) (MOVDconst [1]) (CMPU idx len))
   442  (NilCheck ...) => (LoweredNilCheck ...)
   443  (GetG ...) => (LoweredGetG ...)
   444  (GetClosurePtr ...) => (LoweredGetClosurePtr ...)
   445  (GetCallerSP ...) => (LoweredGetCallerSP ...)
   446  (GetCallerPC ...) => (LoweredGetCallerPC ...)
   447  (Addr {sym} base) => (MOVDaddr {sym} base)
   448  (LocalAddr <t> {sym} base mem) && t.Elem().HasPointers() => (MOVDaddr {sym} (SPanchored base mem))
   449  (LocalAddr <t> {sym} base _)  && !t.Elem().HasPointers() => (MOVDaddr {sym} base)
   450  (ITab (Load ptr mem)) => (MOVDload ptr mem)
   451  
   452  // block rewrites
   453  (If cond yes no) => (CLIJ {s390x.LessOrGreater} (MOVBZreg <typ.Bool> cond) [0] yes no)
   454  
   455  // Write barrier.
   456  (WB ...) => (LoweredWB ...)
   457  
   458  (PanicBounds [kind] x y mem) && boundsABI(kind) == 0 => (LoweredPanicBoundsA [kind] x y mem)
   459  (PanicBounds [kind] x y mem) && boundsABI(kind) == 1 => (LoweredPanicBoundsB [kind] x y mem)
   460  (PanicBounds [kind] x y mem) && boundsABI(kind) == 2 => (LoweredPanicBoundsC [kind] x y mem)
   461  
   462  // ***************************
   463  // Above: lowering rules
   464  // Below: optimizations
   465  // ***************************
   466  // TODO: Should the optimizations be a separate pass?
   467  
   468  // Note: when removing unnecessary sign/zero extensions.
   469  //
   470  // After a value is spilled it is restored using a sign- or zero-extension
   471  // to register-width as appropriate for its type. For example, a uint8 will
   472  // be restored using a MOVBZ (llgc) instruction which will zero extend the
   473  // 8-bit value to 64-bits.
   474  //
   475  // This is a hazard when folding sign- and zero-extensions since we need to
   476  // ensure not only that the value in the argument register is correctly
   477  // extended but also that it will still be correctly extended if it is
   478  // spilled and restored.
   479  //
   480  // In general this means we need type checks when the RHS of a rule is an
   481  // OpCopy (i.e. "(... x:(...) ...) -> x").
   482  
   483  // Merge double extensions.
   484  (MOV(H|HZ)reg e:(MOV(B|BZ)reg x)) && clobberIfDead(e) => (MOV(B|BZ)reg x)
   485  (MOV(W|WZ)reg e:(MOV(B|BZ)reg x)) && clobberIfDead(e) => (MOV(B|BZ)reg x)
   486  (MOV(W|WZ)reg e:(MOV(H|HZ)reg x)) && clobberIfDead(e) => (MOV(H|HZ)reg x)
   487  
   488  // Bypass redundant sign extensions.
   489  (MOV(B|BZ)reg e:(MOVBreg x)) && clobberIfDead(e) => (MOV(B|BZ)reg x)
   490  (MOV(B|BZ)reg e:(MOVHreg x)) && clobberIfDead(e) => (MOV(B|BZ)reg x)
   491  (MOV(B|BZ)reg e:(MOVWreg x)) && clobberIfDead(e) => (MOV(B|BZ)reg x)
   492  (MOV(H|HZ)reg e:(MOVHreg x)) && clobberIfDead(e) => (MOV(H|HZ)reg x)
   493  (MOV(H|HZ)reg e:(MOVWreg x)) && clobberIfDead(e) => (MOV(H|HZ)reg x)
   494  (MOV(W|WZ)reg e:(MOVWreg x)) && clobberIfDead(e) => (MOV(W|WZ)reg x)
   495  
   496  // Bypass redundant zero extensions.
   497  (MOV(B|BZ)reg e:(MOVBZreg x)) && clobberIfDead(e) => (MOV(B|BZ)reg x)
   498  (MOV(B|BZ)reg e:(MOVHZreg x)) && clobberIfDead(e) => (MOV(B|BZ)reg x)
   499  (MOV(B|BZ)reg e:(MOVWZreg x)) && clobberIfDead(e) => (MOV(B|BZ)reg x)
   500  (MOV(H|HZ)reg e:(MOVHZreg x)) && clobberIfDead(e) => (MOV(H|HZ)reg x)
   501  (MOV(H|HZ)reg e:(MOVWZreg x)) && clobberIfDead(e) => (MOV(H|HZ)reg x)
   502  (MOV(W|WZ)reg e:(MOVWZreg x)) && clobberIfDead(e) => (MOV(W|WZ)reg x)
   503  
   504  // Remove zero extensions after zero extending load.
   505  // Note: take care that if x is spilled it is restored correctly.
   506  (MOV(B|H|W)Zreg x:(MOVBZload    _   _)) && (!x.Type.IsSigned() || x.Type.Size() > 1) => x
   507  (MOV(H|W)Zreg   x:(MOVHZload    _   _)) && (!x.Type.IsSigned() || x.Type.Size() > 2) => x
   508  (MOVWZreg       x:(MOVWZload    _   _)) && (!x.Type.IsSigned() || x.Type.Size() > 4) => x
   509  
   510  // Remove sign extensions after sign extending load.
   511  // Note: take care that if x is spilled it is restored correctly.
   512  (MOV(B|H|W)reg x:(MOVBload    _   _)) && (x.Type.IsSigned() || x.Type.Size() == 8) => x
   513  (MOV(H|W)reg   x:(MOVHload    _   _)) && (x.Type.IsSigned() || x.Type.Size() == 8) => x
   514  (MOVWreg       x:(MOVWload    _   _)) && (x.Type.IsSigned() || x.Type.Size() == 8) => x
   515  
   516  // Remove sign extensions after zero extending load.
   517  // These type checks are probably unnecessary but do them anyway just in case.
   518  (MOV(H|W)reg x:(MOVBZload    _   _)) && (!x.Type.IsSigned() || x.Type.Size() > 1) => x
   519  (MOVWreg     x:(MOVHZload    _   _)) && (!x.Type.IsSigned() || x.Type.Size() > 2) => x
   520  
   521  // Fold sign and zero extensions into loads.
   522  //
   523  // Note: The combined instruction must end up in the same block
   524  // as the original load. If not, we end up making a value with
   525  // memory type live in two different blocks, which can lead to
   526  // multiple memory values alive simultaneously.
   527  //
   528  // Make sure we don't combine these ops if the load has another use.
   529  // This prevents a single load from being split into multiple loads
   530  // which then might return different values.  See test/atomicload.go.
   531  (MOV(B|H|W)Zreg <t> x:(MOV(B|H|W)load [o] {s} p mem))
   532    && x.Uses == 1
   533    && clobber(x)
   534    => @x.Block (MOV(B|H|W)Zload <t> [o] {s} p mem)
   535  (MOV(B|H|W)reg <t> x:(MOV(B|H|W)Zload [o] {s} p mem))
   536    && x.Uses == 1
   537    && clobber(x)
   538    => @x.Block (MOV(B|H|W)load <t> [o] {s} p mem)
   539  
   540  // Remove zero extensions after argument load.
   541  (MOVBZreg x:(Arg <t>)) && !t.IsSigned() && t.Size() == 1 => x
   542  (MOVHZreg x:(Arg <t>)) && !t.IsSigned() && t.Size() <= 2 => x
   543  (MOVWZreg x:(Arg <t>)) && !t.IsSigned() && t.Size() <= 4 => x
   544  
   545  // Remove sign extensions after argument load.
   546  (MOVBreg x:(Arg <t>)) && t.IsSigned() && t.Size() == 1 => x
   547  (MOVHreg x:(Arg <t>)) && t.IsSigned() && t.Size() <= 2 => x
   548  (MOVWreg x:(Arg <t>)) && t.IsSigned() && t.Size() <= 4 => x
   549  
   550  // Fold zero extensions into constants.
   551  (MOVBZreg (MOVDconst [c])) => (MOVDconst [int64( uint8(c))])
   552  (MOVHZreg (MOVDconst [c])) => (MOVDconst [int64(uint16(c))])
   553  (MOVWZreg (MOVDconst [c])) => (MOVDconst [int64(uint32(c))])
   554  
   555  // Fold sign extensions into constants.
   556  (MOVBreg (MOVDconst [c])) => (MOVDconst [int64( int8(c))])
   557  (MOVHreg (MOVDconst [c])) => (MOVDconst [int64(int16(c))])
   558  (MOVWreg (MOVDconst [c])) => (MOVDconst [int64(int32(c))])
   559  
   560  // Remove zero extension of conditional move.
   561  // Note: only for MOVBZreg for now since it is added as part of 'if' statement lowering.
   562  (MOVBZreg x:(LOCGR (MOVDconst [c]) (MOVDconst [d]) _))
   563    && int64(uint8(c)) == c
   564    && int64(uint8(d)) == d
   565    && (!x.Type.IsSigned() || x.Type.Size() > 1)
   566    => x
   567  
   568  // Fold boolean tests into blocks.
   569  // Note: this must match If statement lowering.
   570  (CLIJ {s390x.LessOrGreater} (LOCGR {d} (MOVDconst [0]) (MOVDconst [x]) cmp) [0] yes no)
   571    && int32(x) != 0
   572    => (BRC {d} cmp yes no)
   573  
   574  // Canonicalize BRC condition code mask by removing impossible conditions.
   575  // Integer comparisons cannot generate the unordered condition.
   576  (BRC {c} x:((CMP|CMPW|CMPU|CMPWU)    _ _) yes no) && c&s390x.Unordered != 0 => (BRC {c&^s390x.Unordered} x yes no)
   577  (BRC {c} x:((CMP|CMPW|CMPU|CMPWU)const _) yes no) && c&s390x.Unordered != 0 => (BRC {c&^s390x.Unordered} x yes no)
   578  
   579  // Compare-and-branch.
   580  // Note: bit 3 (unordered) must not be set so we mask out s390x.Unordered.
   581  (BRC {c} (CMP   x y) yes no) => (CGRJ  {c&^s390x.Unordered} x y yes no)
   582  (BRC {c} (CMPW  x y) yes no) => (CRJ   {c&^s390x.Unordered} x y yes no)
   583  (BRC {c} (CMPU  x y) yes no) => (CLGRJ {c&^s390x.Unordered} x y yes no)
   584  (BRC {c} (CMPWU x y) yes no) => (CLRJ  {c&^s390x.Unordered} x y yes no)
   585  
   586  // Compare-and-branch (immediate).
   587  // Note: bit 3 (unordered) must not be set so we mask out s390x.Unordered.
   588  (BRC {c} (CMPconst   x [y]) yes no) && y == int32( int8(y)) => (CGIJ  {c&^s390x.Unordered} x [ int8(y)] yes no)
   589  (BRC {c} (CMPWconst  x [y]) yes no) && y == int32( int8(y)) => (CIJ   {c&^s390x.Unordered} x [ int8(y)] yes no)
   590  (BRC {c} (CMPUconst  x [y]) yes no) && y == int32(uint8(y)) => (CLGIJ {c&^s390x.Unordered} x [uint8(y)] yes no)
   591  (BRC {c} (CMPWUconst x [y]) yes no) && y == int32(uint8(y)) => (CLIJ  {c&^s390x.Unordered} x [uint8(y)] yes no)
   592  
   593  // Absorb immediate into compare-and-branch.
   594  (C(R|GR)J  {c} x (MOVDconst [y]) yes no) && is8Bit(y)  => (C(I|GI)J  {c} x [ int8(y)] yes no)
   595  (CL(R|GR)J {c} x (MOVDconst [y]) yes no) && isU8Bit(y) => (CL(I|GI)J {c} x [uint8(y)] yes no)
   596  (C(R|GR)J  {c} (MOVDconst [x]) y yes no) && is8Bit(x)  => (C(I|GI)J  {c.ReverseComparison()} y [ int8(x)] yes no)
   597  (CL(R|GR)J {c} (MOVDconst [x]) y yes no) && isU8Bit(x) => (CL(I|GI)J {c.ReverseComparison()} y [uint8(x)] yes no)
   598  
   599  // Prefer comparison with immediate to compare-and-branch.
   600  (CGRJ  {c} x (MOVDconst [y]) yes no) && !is8Bit(y)  && is32Bit(y)  => (BRC {c} (CMPconst   x [int32(y)]) yes no)
   601  (CRJ   {c} x (MOVDconst [y]) yes no) && !is8Bit(y)  && is32Bit(y)  => (BRC {c} (CMPWconst  x [int32(y)]) yes no)
   602  (CLGRJ {c} x (MOVDconst [y]) yes no) && !isU8Bit(y) && isU32Bit(y) => (BRC {c} (CMPUconst  x [int32(y)]) yes no)
   603  (CLRJ  {c} x (MOVDconst [y]) yes no) && !isU8Bit(y) && isU32Bit(y) => (BRC {c} (CMPWUconst x [int32(y)]) yes no)
   604  (CGRJ  {c} (MOVDconst [x]) y yes no) && !is8Bit(x)  && is32Bit(x)  => (BRC {c.ReverseComparison()} (CMPconst   y [int32(x)]) yes no)
   605  (CRJ   {c} (MOVDconst [x]) y yes no) && !is8Bit(x)  && is32Bit(x)  => (BRC {c.ReverseComparison()} (CMPWconst  y [int32(x)]) yes no)
   606  (CLGRJ {c} (MOVDconst [x]) y yes no) && !isU8Bit(x) && isU32Bit(x) => (BRC {c.ReverseComparison()} (CMPUconst  y [int32(x)]) yes no)
   607  (CLRJ  {c} (MOVDconst [x]) y yes no) && !isU8Bit(x) && isU32Bit(x) => (BRC {c.ReverseComparison()} (CMPWUconst y [int32(x)]) yes no)
   608  
   609  // Absorb sign/zero extensions into 32-bit compare-and-branch.
   610  (CIJ  {c} (MOV(W|WZ)reg x) [y] yes no) => (CIJ  {c} x [y] yes no)
   611  (CLIJ {c} (MOV(W|WZ)reg x) [y] yes no) => (CLIJ {c} x [y] yes no)
   612  
   613  // Bring out-of-range signed immediates into range by varying branch condition.
   614  (BRC {s390x.Less}           (CMPconst  x [ 128]) yes no) => (CGIJ {s390x.LessOrEqual}    x [ 127] yes no)
   615  (BRC {s390x.Less}           (CMPWconst x [ 128]) yes no) => (CIJ  {s390x.LessOrEqual}    x [ 127] yes no)
   616  (BRC {s390x.LessOrEqual}    (CMPconst  x [-129]) yes no) => (CGIJ {s390x.Less}           x [-128] yes no)
   617  (BRC {s390x.LessOrEqual}    (CMPWconst x [-129]) yes no) => (CIJ  {s390x.Less}           x [-128] yes no)
   618  (BRC {s390x.Greater}        (CMPconst  x [-129]) yes no) => (CGIJ {s390x.GreaterOrEqual} x [-128] yes no)
   619  (BRC {s390x.Greater}        (CMPWconst x [-129]) yes no) => (CIJ  {s390x.GreaterOrEqual} x [-128] yes no)
   620  (BRC {s390x.GreaterOrEqual} (CMPconst  x [ 128]) yes no) => (CGIJ {s390x.Greater}        x [ 127] yes no)
   621  (BRC {s390x.GreaterOrEqual} (CMPWconst x [ 128]) yes no) => (CIJ  {s390x.Greater}        x [ 127] yes no)
   622  
   623  // Bring out-of-range unsigned immediates into range by varying branch condition.
   624  (BRC {s390x.Less}           (CMP(WU|U)const  x [256]) yes no) => (C(L|LG)IJ {s390x.LessOrEqual} x [255] yes no)
   625  (BRC {s390x.GreaterOrEqual} (CMP(WU|U)const  x [256]) yes no) => (C(L|LG)IJ {s390x.Greater}     x [255] yes no)
   626  
   627  // Bring out-of-range immediates into range by switching signedness (only == and !=).
   628  (BRC {c} (CMPconst   x [y]) yes no) && y == int32(uint8(y)) && (c == s390x.Equal || c == s390x.LessOrGreater) => (CLGIJ {c} x [uint8(y)] yes no)
   629  (BRC {c} (CMPWconst  x [y]) yes no) && y == int32(uint8(y)) && (c == s390x.Equal || c == s390x.LessOrGreater) => (CLIJ  {c} x [uint8(y)] yes no)
   630  (BRC {c} (CMPUconst  x [y]) yes no) && y == int32( int8(y)) && (c == s390x.Equal || c == s390x.LessOrGreater) => (CGIJ  {c} x [ int8(y)] yes no)
   631  (BRC {c} (CMPWUconst x [y]) yes no) && y == int32( int8(y)) && (c == s390x.Equal || c == s390x.LessOrGreater) => (CIJ   {c} x [ int8(y)] yes no)
   632  
   633  // Fold constants into instructions.
   634  (ADD x (MOVDconst <t> [c])) && is32Bit(c) && !t.IsPtr() => (ADDconst [int32(c)] x)
   635  (ADDW x (MOVDconst [c])) => (ADDWconst [int32(c)] x)
   636  
   637  (SUB x (MOVDconst [c])) && is32Bit(c) => (SUBconst x [int32(c)])
   638  (SUB (MOVDconst [c]) x) && is32Bit(c) => (NEG (SUBconst <v.Type> x [int32(c)]))
   639  (SUBW x (MOVDconst [c])) => (SUBWconst x [int32(c)])
   640  (SUBW (MOVDconst [c]) x) => (NEGW (SUBWconst <v.Type> x [int32(c)]))
   641  
   642  (MULLD x (MOVDconst [c])) && is32Bit(c) => (MULLDconst [int32(c)] x)
   643  (MULLW x (MOVDconst [c])) => (MULLWconst [int32(c)] x)
   644  
   645  // NILF instructions leave the high 32 bits unchanged which is
   646  // equivalent to the leftmost 32 bits being set.
   647  // TODO(mundaym): modify the assembler to accept 64-bit values
   648  // and use isU32Bit(^c).
   649  (AND x (MOVDconst [c]))
   650    && s390x.NewRotateParams(0, 63, 0).OutMerge(uint64(c)) != nil
   651    => (RISBGZ x {*s390x.NewRotateParams(0, 63, 0).OutMerge(uint64(c))})
   652  (AND x (MOVDconst [c]))
   653    && is32Bit(c)
   654    && c < 0
   655    => (ANDconst [c] x)
   656  (AND x (MOVDconst [c]))
   657    && is32Bit(c)
   658    && c >= 0
   659    => (MOVWZreg (ANDWconst <typ.UInt32> [int32(c)] x))
   660  
   661  (ANDW x (MOVDconst [c])) => (ANDWconst [int32(c)] x)
   662  
   663  ((AND|ANDW)const [c] ((AND|ANDW)const [d] x)) => ((AND|ANDW)const [c&d] x)
   664  
   665  ((OR|XOR) x (MOVDconst [c])) && isU32Bit(c) => ((OR|XOR)const [c] x)
   666  ((OR|XOR)W x (MOVDconst [c])) => ((OR|XOR)Wconst [int32(c)] x)
   667  
   668  // Constant shifts.
   669  (S(LD|RD|RAD) x (MOVDconst [c])) => (S(LD|RD|RAD)const x [uint8(c&63)])
   670  (S(LW|RW|RAW) x (MOVDconst [c])) && c&32 == 0 => (S(LW|RW|RAW)const x [uint8(c&31)])
   671  (S(LW|RW)     _ (MOVDconst [c])) && c&32 != 0 => (MOVDconst [0])
   672  (SRAW         x (MOVDconst [c])) && c&32 != 0 => (SRAWconst x [31])
   673  
   674  // Shifts only use the rightmost 6 bits of the shift value.
   675  (S(LD|RD|RAD|LW|RW|RAW) x (RISBGZ y {r}))
   676    && r.Amount == 0
   677    && r.OutMask()&63 == 63
   678    => (S(LD|RD|RAD|LW|RW|RAW) x y)
   679  (S(LD|RD|RAD|LW|RW|RAW) x (AND (MOVDconst [c]) y))
   680    => (S(LD|RD|RAD|LW|RW|RAW) x (ANDWconst <typ.UInt32> [int32(c&63)] y))
   681  (S(LD|RD|RAD|LW|RW|RAW) x (ANDWconst [c] y)) && c&63 == 63
   682    => (S(LD|RD|RAD|LW|RW|RAW) x y)
   683  (SLD  x (MOV(W|H|B|WZ|HZ|BZ)reg y)) => (SLD  x y)
   684  (SRD  x (MOV(W|H|B|WZ|HZ|BZ)reg y)) => (SRD  x y)
   685  (SRAD x (MOV(W|H|B|WZ|HZ|BZ)reg y)) => (SRAD x y)
   686  (SLW  x (MOV(W|H|B|WZ|HZ|BZ)reg y)) => (SLW  x y)
   687  (SRW  x (MOV(W|H|B|WZ|HZ|BZ)reg y)) => (SRW  x y)
   688  (SRAW x (MOV(W|H|B|WZ|HZ|BZ)reg y)) => (SRAW x y)
   689  
   690  // Match rotate by constant.
   691  (RLLG x (MOVDconst [c])) => (RISBGZ x {s390x.NewRotateParams(0, 63, uint8(c&63))})
   692  (RLL  x (MOVDconst [c])) => (RLLconst x [uint8(c&31)])
   693  
   694  // Signed 64-bit comparison with immediate.
   695  (CMP x (MOVDconst [c])) && is32Bit(c) => (CMPconst x [int32(c)])
   696  (CMP (MOVDconst [c]) x) && is32Bit(c) => (InvertFlags (CMPconst x [int32(c)]))
   697  
   698  // Unsigned 64-bit comparison with immediate.
   699  (CMPU x (MOVDconst [c])) && isU32Bit(c) => (CMPUconst x [int32(c)])
   700  (CMPU (MOVDconst [c]) x) && isU32Bit(c) => (InvertFlags (CMPUconst x [int32(c)]))
   701  
   702  // Signed and unsigned 32-bit comparison with immediate.
   703  (CMP(W|WU) x (MOVDconst [c])) => (CMP(W|WU)const x [int32(c)])
   704  (CMP(W|WU) (MOVDconst [c]) x) => (InvertFlags (CMP(W|WU)const x [int32(c)]))
   705  
   706  // Match (x >> c) << d to 'rotate then insert selected bits [into zero]'.
   707  (SLDconst (SRDconst x [c]) [d]) => (RISBGZ x {s390x.NewRotateParams(uint8(max(0, int8(c-d))), 63-d, uint8(int8(d-c)&63))})
   708  
   709  // Match (x << c) >> d to 'rotate then insert selected bits [into zero]'.
   710  (SRDconst (SLDconst x [c]) [d]) => (RISBGZ x {s390x.NewRotateParams(d, uint8(min(63, int8(63-c+d))), uint8(int8(c-d)&63))})
   711  
   712  // Absorb input zero extension into 'rotate then insert selected bits [into zero]'.
   713  (RISBGZ (MOVWZreg x) {r}) && r.InMerge(0xffffffff) != nil => (RISBGZ x {*r.InMerge(0xffffffff)})
   714  (RISBGZ (MOVHZreg x) {r}) && r.InMerge(0x0000ffff) != nil => (RISBGZ x {*r.InMerge(0x0000ffff)})
   715  (RISBGZ (MOVBZreg x) {r}) && r.InMerge(0x000000ff) != nil => (RISBGZ x {*r.InMerge(0x000000ff)})
   716  
   717  // Absorb 'rotate then insert selected bits [into zero]' into zero extension.
   718  (MOVWZreg (RISBGZ x {r})) && r.OutMerge(0xffffffff) != nil => (RISBGZ x {*r.OutMerge(0xffffffff)})
   719  (MOVHZreg (RISBGZ x {r})) && r.OutMerge(0x0000ffff) != nil => (RISBGZ x {*r.OutMerge(0x0000ffff)})
   720  (MOVBZreg (RISBGZ x {r})) && r.OutMerge(0x000000ff) != nil => (RISBGZ x {*r.OutMerge(0x000000ff)})
   721  
   722  // Absorb shift into 'rotate then insert selected bits [into zero]'.
   723  //
   724  // Any unsigned shift can be represented as a rotate and mask operation:
   725  //
   726  //   x << c => RotateLeft64(x, c) & (^uint64(0) << c)
   727  //   x >> c => RotateLeft64(x, -c) & (^uint64(0) >> c)
   728  //
   729  // Therefore when a shift is used as the input to a rotate then insert
   730  // selected bits instruction we can merge the two together. We just have
   731  // to be careful that the resultant mask is representable (non-zero and
   732  // contiguous). For example, assuming that x is variable and c, y and m
   733  // are constants, a shift followed by a rotate then insert selected bits
   734  // could be represented as:
   735  //
   736  //   RotateLeft64(RotateLeft64(x, c) & (^uint64(0) << c), y) & m
   737  //
   738  // We can split the rotation by y into two, one rotate for x and one for
   739  // the mask:
   740  //
   741  //   RotateLeft64(RotateLeft64(x, c), y) & (RotateLeft64(^uint64(0) << c, y)) & m
   742  //
   743  // The rotations of x by c followed by y can then be combined:
   744  //
   745  //   RotateLeft64(x, c+y) & (RotateLeft64(^uint64(0) << c, y)) & m
   746  //   ^^^^^^^^^^^^^^^^^^^^   ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
   747  //          rotate                          mask
   748  //
   749  // To perform this optimization we therefore just need to check that it
   750  // is valid to merge the shift mask (^(uint64(0)<<c)) into the selected
   751  // bits mask (i.e. that the resultant mask is non-zero and contiguous).
   752  //
   753  (RISBGZ (SLDconst x [c]) {r}) && r.InMerge(^uint64(0)<<c) != nil => (RISBGZ x {(*r.InMerge(^uint64(0)<<c)).RotateLeft(c)})
   754  (RISBGZ (SRDconst x [c]) {r}) && r.InMerge(^uint64(0)>>c) != nil => (RISBGZ x {(*r.InMerge(^uint64(0)>>c)).RotateLeft(-c)})
   755  
   756  // Absorb 'rotate then insert selected bits [into zero]' into left shift.
   757  (SLDconst (RISBGZ x {r}) [c])
   758    && s390x.NewRotateParams(0, 63-c, c).InMerge(r.OutMask()) != nil
   759    => (RISBGZ x {(*s390x.NewRotateParams(0, 63-c, c).InMerge(r.OutMask())).RotateLeft(r.Amount)})
   760  
   761  // Absorb 'rotate then insert selected bits [into zero]' into right shift.
   762  (SRDconst (RISBGZ x {r}) [c])
   763    && s390x.NewRotateParams(c, 63, -c&63).InMerge(r.OutMask()) != nil
   764    => (RISBGZ x {(*s390x.NewRotateParams(c, 63, -c&63).InMerge(r.OutMask())).RotateLeft(r.Amount)})
   765  
   766  // Merge 'rotate then insert selected bits [into zero]' instructions together.
   767  (RISBGZ (RISBGZ x {y}) {z})
   768    && z.InMerge(y.OutMask()) != nil
   769    => (RISBGZ x {(*z.InMerge(y.OutMask())).RotateLeft(y.Amount)})
   770  
   771  // Convert RISBGZ into 64-bit shift (helps CSE).
   772  (RISBGZ x {r}) && r.End == 63 && r.Start == -r.Amount&63 => (SRDconst x [-r.Amount&63])
   773  (RISBGZ x {r}) && r.Start == 0 && r.End == 63-r.Amount => (SLDconst x [r.Amount])
   774  
   775  // Optimize single bit isolation when it is known to be equivalent to
   776  // the most significant bit due to mask produced by arithmetic shift.
   777  // Simply isolate the most significant bit itself and place it in the
   778  // correct position.
   779  //
   780  // Example: (int64(x) >> 63) & 0x8 -> RISBGZ $60, $60, $4, Rsrc, Rdst
   781  (RISBGZ (SRADconst x [c]) {r})
   782    && r.Start == r.End           // single bit selected
   783    && (r.Start+r.Amount)&63 <= c // equivalent to most significant bit of x
   784    => (RISBGZ x {s390x.NewRotateParams(r.Start, r.Start, -r.Start&63)})
   785  
   786  // Canonicalize the order of arguments to comparisons - helps with CSE.
   787  ((CMP|CMPW|CMPU|CMPWU) x y) && canonLessThan(x,y) => (InvertFlags ((CMP|CMPW|CMPU|CMPWU) y x))
   788  
   789  // Use sign/zero extend instead of RISBGZ.
   790  (RISBGZ x {r}) && r == s390x.NewRotateParams(56, 63, 0) => (MOVBZreg x)
   791  (RISBGZ x {r}) && r == s390x.NewRotateParams(48, 63, 0) => (MOVHZreg x)
   792  (RISBGZ x {r}) && r == s390x.NewRotateParams(32, 63, 0) => (MOVWZreg x)
   793  
   794  // Use sign/zero extend instead of ANDW.
   795  (ANDWconst [0x00ff] x) => (MOVBZreg x)
   796  (ANDWconst [0xffff] x) => (MOVHZreg x)
   797  
   798  // Strength reduce multiplication to the sum (or difference) of two powers of two.
   799  //
   800  // Examples:
   801  //     5x -> 4x + 1x
   802  //    10x -> 8x + 2x
   803  //   120x -> 128x - 8x
   804  //  -120x -> 8x - 128x
   805  //
   806  // We know that the rightmost bit of any positive value, once isolated, must either
   807  // be a power of 2 (because it is a single bit) or 0 (if the original value is 0).
   808  // In all of these rules we use a rightmost bit calculation to determine one operand
   809  // for the addition or subtraction. We then just need to calculate if the other
   810  // operand is a valid power of 2 before we can match the rule.
   811  //
   812  // Notes:
   813  //   - the generic rules have already matched single powers of two so we ignore them here
   814  //   - isPowerOfTwo asserts that its argument is greater than 0
   815  //   - c&(c-1) = clear rightmost bit
   816  //   - c&^(c-1) = isolate rightmost bit
   817  
   818  // c = 2Ë£ + 2ʸ => c - 2Ë£ = 2ʸ
   819  (MULL(D|W)const <t> x [c]) && isPowerOfTwo(c&(c-1))
   820    => ((ADD|ADDW) (SL(D|W)const <t> x [uint8(log32(c&(c-1)))])
   821                   (SL(D|W)const <t> x [uint8(log32(c&^(c-1)))]))
   822  
   823  // c = 2ʸ - 2Ë£ => c + 2Ë£ = 2ʸ
   824  (MULL(D|W)const <t> x [c]) && isPowerOfTwo(c+(c&^(c-1)))
   825    => ((SUB|SUBW) (SL(D|W)const <t> x [uint8(log32(c+(c&^(c-1))))])
   826                   (SL(D|W)const <t> x [uint8(log32(c&^(c-1)))]))
   827  
   828  // c = 2Ë£ - 2ʸ => -c + 2Ë£ = 2ʸ
   829  (MULL(D|W)const <t> x [c]) && isPowerOfTwo(-c+(-c&^(-c-1)))
   830    => ((SUB|SUBW) (SL(D|W)const <t> x [uint8(log32(-c&^(-c-1)))])
   831                   (SL(D|W)const <t> x [uint8(log32(-c+(-c&^(-c-1))))]))
   832  
   833  // Fold ADD into MOVDaddr. Odd offsets from SB shouldn't be folded (LARL can't handle them).
   834  (ADDconst [c] (MOVDaddr [d] {s} x:(SB))) && ((c+d)&1 == 0) && is32Bit(int64(c)+int64(d)) => (MOVDaddr [c+d] {s} x)
   835  (ADDconst [c] (MOVDaddr [d] {s} x)) && x.Op != OpSB && is20Bit(int64(c)+int64(d)) => (MOVDaddr [c+d] {s} x)
   836  (ADD idx (MOVDaddr [c] {s} ptr)) && ptr.Op != OpSB => (MOVDaddridx [c] {s} ptr idx)
   837  
   838  // fold ADDconst into MOVDaddrx
   839  (ADDconst [c] (MOVDaddridx [d] {s} x y)) && is20Bit(int64(c)+int64(d)) => (MOVDaddridx [c+d] {s} x y)
   840  (MOVDaddridx [c] {s} (ADDconst [d] x) y) && is20Bit(int64(c)+int64(d)) => (MOVDaddridx [c+d] {s} x y)
   841  (MOVDaddridx [c] {s} x (ADDconst [d] y)) && is20Bit(int64(c)+int64(d)) => (MOVDaddridx [c+d] {s} x y)
   842  
   843  // reverse ordering of compare instruction
   844  (LOCGR {c} x y (InvertFlags cmp)) => (LOCGR {c.ReverseComparison()} x y cmp)
   845  
   846  // replace load from same location as preceding store with copy
   847  (MOVDload  [off] {sym} ptr1 (MOVDstore  [off] {sym} ptr2 x _)) && isSamePtr(ptr1, ptr2) => x
   848  (MOVWload  [off] {sym} ptr1 (MOVWstore  [off] {sym} ptr2 x _)) && isSamePtr(ptr1, ptr2) => (MOVWreg x)
   849  (MOVHload  [off] {sym} ptr1 (MOVHstore  [off] {sym} ptr2 x _)) && isSamePtr(ptr1, ptr2) => (MOVHreg x)
   850  (MOVBload  [off] {sym} ptr1 (MOVBstore  [off] {sym} ptr2 x _)) && isSamePtr(ptr1, ptr2) => (MOVBreg x)
   851  (MOVWZload [off] {sym} ptr1 (MOVWstore  [off] {sym} ptr2 x _)) && isSamePtr(ptr1, ptr2) => (MOVWZreg x)
   852  (MOVHZload [off] {sym} ptr1 (MOVHstore  [off] {sym} ptr2 x _)) && isSamePtr(ptr1, ptr2) => (MOVHZreg x)
   853  (MOVBZload [off] {sym} ptr1 (MOVBstore  [off] {sym} ptr2 x _)) && isSamePtr(ptr1, ptr2) => (MOVBZreg x)
   854  (MOVDload  [off] {sym} ptr1 (FMOVDstore [off] {sym} ptr2 x _)) && isSamePtr(ptr1, ptr2) => (LGDR x)
   855  (FMOVDload [off] {sym} ptr1 (MOVDstore  [off] {sym} ptr2 x _)) && isSamePtr(ptr1, ptr2) => (LDGR x)
   856  (FMOVDload [off] {sym} ptr1 (FMOVDstore [off] {sym} ptr2 x _)) && isSamePtr(ptr1, ptr2) => x
   857  (FMOVSload [off] {sym} ptr1 (FMOVSstore [off] {sym} ptr2 x _)) && isSamePtr(ptr1, ptr2) => x
   858  
   859  // prefer FPR <-> GPR moves over combined load ops
   860  (MULLDload <t> [off] {sym} x ptr1 (FMOVDstore [off] {sym} ptr2 y _)) && isSamePtr(ptr1, ptr2) => (MULLD x (LGDR <t> y))
   861  (ADDload   <t> [off] {sym} x ptr1 (FMOVDstore [off] {sym} ptr2 y _)) && isSamePtr(ptr1, ptr2) => (ADD   x (LGDR <t> y))
   862  (SUBload   <t> [off] {sym} x ptr1 (FMOVDstore [off] {sym} ptr2 y _)) && isSamePtr(ptr1, ptr2) => (SUB   x (LGDR <t> y))
   863  (ORload    <t> [off] {sym} x ptr1 (FMOVDstore [off] {sym} ptr2 y _)) && isSamePtr(ptr1, ptr2) => (OR    x (LGDR <t> y))
   864  (ANDload   <t> [off] {sym} x ptr1 (FMOVDstore [off] {sym} ptr2 y _)) && isSamePtr(ptr1, ptr2) => (AND   x (LGDR <t> y))
   865  (XORload   <t> [off] {sym} x ptr1 (FMOVDstore [off] {sym} ptr2 y _)) && isSamePtr(ptr1, ptr2) => (XOR   x (LGDR <t> y))
   866  
   867  // detect attempts to set/clear the sign bit
   868  // may need to be reworked when NIHH/OIHH are added
   869  (RISBGZ (LGDR <t> x) {r}) && r == s390x.NewRotateParams(1, 63, 0) => (LGDR <t> (LPDFR <x.Type> x))
   870  (LDGR <t> (RISBGZ x {r})) && r == s390x.NewRotateParams(1, 63, 0) => (LPDFR (LDGR <t> x))
   871  (OR (MOVDconst [-1<<63]) (LGDR <t> x)) => (LGDR <t> (LNDFR <x.Type> x))
   872  (LDGR <t> (OR (MOVDconst [-1<<63]) x)) => (LNDFR (LDGR <t> x))
   873  
   874  // detect attempts to set the sign bit with load
   875  (LDGR <t> x:(ORload <t1> [off] {sym} (MOVDconst [-1<<63]) ptr mem)) && x.Uses == 1 && clobber(x) => @x.Block (LNDFR <t> (LDGR <t> (MOVDload <t1> [off] {sym} ptr mem)))
   876  
   877  // detect copysign
   878  (OR (RISBGZ (LGDR x) {r}) (LGDR (LPDFR <t> y)))
   879    && r == s390x.NewRotateParams(0, 0, 0)
   880    => (LGDR (CPSDR <t> y x))
   881  (OR (RISBGZ (LGDR x) {r}) (MOVDconst [c]))
   882    && c >= 0
   883    && r == s390x.NewRotateParams(0, 0, 0)
   884    => (LGDR (CPSDR <x.Type> (FMOVDconst <x.Type> [math.Float64frombits(uint64(c))]) x))
   885  (CPSDR y (FMOVDconst [c])) && !math.Signbit(c) => (LPDFR y)
   886  (CPSDR y (FMOVDconst [c])) && math.Signbit(c)  => (LNDFR y)
   887  
   888  // absorb negations into set/clear sign bit
   889  (FNEG  (LPDFR x)) => (LNDFR x)
   890  (FNEG  (LNDFR x)) => (LPDFR x)
   891  (FNEGS (LPDFR x)) => (LNDFR x)
   892  (FNEGS (LNDFR x)) => (LPDFR x)
   893  
   894  // no need to convert float32 to float64 to set/clear sign bit
   895  (LEDBR (LPDFR (LDEBR x))) => (LPDFR x)
   896  (LEDBR (LNDFR (LDEBR x))) => (LNDFR x)
   897  
   898  // remove unnecessary FPR <-> GPR moves
   899  (LDGR (LGDR x)) => x
   900  (LGDR (LDGR x)) => x
   901  
   902  // Don't extend before storing
   903  (MOVWstore [off] {sym} ptr (MOVWreg x) mem) => (MOVWstore [off] {sym} ptr x mem)
   904  (MOVHstore [off] {sym} ptr (MOVHreg x) mem) => (MOVHstore [off] {sym} ptr x mem)
   905  (MOVBstore [off] {sym} ptr (MOVBreg x) mem) => (MOVBstore [off] {sym} ptr x mem)
   906  (MOVWstore [off] {sym} ptr (MOVWZreg x) mem) => (MOVWstore [off] {sym} ptr x mem)
   907  (MOVHstore [off] {sym} ptr (MOVHZreg x) mem) => (MOVHstore [off] {sym} ptr x mem)
   908  (MOVBstore [off] {sym} ptr (MOVBZreg x) mem) => (MOVBstore [off] {sym} ptr x mem)
   909  
   910  // Fold constants into memory operations.
   911  // Note that this is not always a good idea because if not all the uses of
   912  // the ADDconst get eliminated, we still have to compute the ADDconst and we now
   913  // have potentially two live values (ptr and (ADDconst [off] ptr)) instead of one.
   914  // Nevertheless, let's do it!
   915  (MOVDload   [off1] {sym} (ADDconst [off2] ptr) mem) && is20Bit(int64(off1)+int64(off2)) => (MOVDload  [off1+off2] {sym} ptr mem)
   916  (MOVWload   [off1] {sym} (ADDconst [off2] ptr) mem) && is20Bit(int64(off1)+int64(off2)) => (MOVWload  [off1+off2] {sym} ptr mem)
   917  (MOVHload   [off1] {sym} (ADDconst [off2] ptr) mem) && is20Bit(int64(off1)+int64(off2)) => (MOVHload  [off1+off2] {sym} ptr mem)
   918  (MOVBload   [off1] {sym} (ADDconst [off2] ptr) mem) && is20Bit(int64(off1)+int64(off2)) => (MOVBload  [off1+off2] {sym} ptr mem)
   919  (MOVWZload  [off1] {sym} (ADDconst [off2] ptr) mem) && is20Bit(int64(off1)+int64(off2)) => (MOVWZload [off1+off2] {sym} ptr mem)
   920  (MOVHZload  [off1] {sym} (ADDconst [off2] ptr) mem) && is20Bit(int64(off1)+int64(off2)) => (MOVHZload [off1+off2] {sym} ptr mem)
   921  (MOVBZload  [off1] {sym} (ADDconst [off2] ptr) mem) && is20Bit(int64(off1)+int64(off2)) => (MOVBZload [off1+off2] {sym} ptr mem)
   922  (FMOVSload  [off1] {sym} (ADDconst [off2] ptr) mem) && is20Bit(int64(off1)+int64(off2)) => (FMOVSload [off1+off2] {sym} ptr mem)
   923  (FMOVDload  [off1] {sym} (ADDconst [off2] ptr) mem) && is20Bit(int64(off1)+int64(off2)) => (FMOVDload [off1+off2] {sym} ptr mem)
   924  
   925  (MOVDstore  [off1] {sym} (ADDconst [off2] ptr) val mem) && is20Bit(int64(off1)+int64(off2)) => (MOVDstore  [off1+off2] {sym} ptr val mem)
   926  (MOVWstore  [off1] {sym} (ADDconst [off2] ptr) val mem) && is20Bit(int64(off1)+int64(off2)) => (MOVWstore  [off1+off2] {sym} ptr val mem)
   927  (MOVHstore  [off1] {sym} (ADDconst [off2] ptr) val mem) && is20Bit(int64(off1)+int64(off2)) => (MOVHstore  [off1+off2] {sym} ptr val mem)
   928  (MOVBstore  [off1] {sym} (ADDconst [off2] ptr) val mem) && is20Bit(int64(off1)+int64(off2)) => (MOVBstore  [off1+off2] {sym} ptr val mem)
   929  (FMOVSstore [off1] {sym} (ADDconst [off2] ptr) val mem) && is20Bit(int64(off1)+int64(off2)) => (FMOVSstore [off1+off2] {sym} ptr val mem)
   930  (FMOVDstore [off1] {sym} (ADDconst [off2] ptr) val mem) && is20Bit(int64(off1)+int64(off2)) => (FMOVDstore [off1+off2] {sym} ptr val mem)
   931  
   932  (ADDload   [off1] {sym} x (ADDconst [off2] ptr) mem) && ptr.Op != OpSB && is20Bit(int64(off1)+int64(off2)) => (ADDload   [off1+off2] {sym} x ptr mem)
   933  (ADDWload  [off1] {sym} x (ADDconst [off2] ptr) mem) && ptr.Op != OpSB && is20Bit(int64(off1)+int64(off2)) => (ADDWload  [off1+off2] {sym} x ptr mem)
   934  (MULLDload [off1] {sym} x (ADDconst [off2] ptr) mem) && ptr.Op != OpSB && is20Bit(int64(off1)+int64(off2)) => (MULLDload [off1+off2] {sym} x ptr mem)
   935  (MULLWload [off1] {sym} x (ADDconst [off2] ptr) mem) && ptr.Op != OpSB && is20Bit(int64(off1)+int64(off2)) => (MULLWload [off1+off2] {sym} x ptr mem)
   936  (SUBload   [off1] {sym} x (ADDconst [off2] ptr) mem) && ptr.Op != OpSB && is20Bit(int64(off1)+int64(off2)) => (SUBload   [off1+off2] {sym} x ptr mem)
   937  (SUBWload  [off1] {sym} x (ADDconst [off2] ptr) mem) && ptr.Op != OpSB && is20Bit(int64(off1)+int64(off2)) => (SUBWload  [off1+off2] {sym} x ptr mem)
   938  
   939  (ANDload   [off1] {sym} x (ADDconst [off2] ptr) mem) && ptr.Op != OpSB && is20Bit(int64(off1)+int64(off2)) => (ANDload   [off1+off2] {sym} x ptr mem)
   940  (ANDWload  [off1] {sym} x (ADDconst [off2] ptr) mem) && ptr.Op != OpSB && is20Bit(int64(off1)+int64(off2)) => (ANDWload  [off1+off2] {sym} x ptr mem)
   941  (ORload    [off1] {sym} x (ADDconst [off2] ptr) mem) && ptr.Op != OpSB && is20Bit(int64(off1)+int64(off2)) => (ORload    [off1+off2] {sym} x ptr mem)
   942  (ORWload   [off1] {sym} x (ADDconst [off2] ptr) mem) && ptr.Op != OpSB && is20Bit(int64(off1)+int64(off2)) => (ORWload   [off1+off2] {sym} x ptr mem)
   943  (XORload   [off1] {sym} x (ADDconst [off2] ptr) mem) && ptr.Op != OpSB && is20Bit(int64(off1)+int64(off2)) => (XORload   [off1+off2] {sym} x ptr mem)
   944  (XORWload  [off1] {sym} x (ADDconst [off2] ptr) mem) && ptr.Op != OpSB && is20Bit(int64(off1)+int64(off2)) => (XORWload  [off1+off2] {sym} x ptr mem)
   945  
   946  // Fold constants into stores.
   947  (MOVDstore [off] {sym} ptr (MOVDconst [c]) mem) && is16Bit(c) && isU12Bit(int64(off)) && ptr.Op != OpSB =>
   948  	(MOVDstoreconst [makeValAndOff(int32(c),off)] {sym} ptr mem)
   949  (MOVWstore [off] {sym} ptr (MOVDconst [c]) mem) && is16Bit(c) && isU12Bit(int64(off)) && ptr.Op != OpSB =>
   950  	(MOVWstoreconst [makeValAndOff(int32(c),off)] {sym} ptr mem)
   951  (MOVHstore [off] {sym} ptr (MOVDconst [c]) mem) && isU12Bit(int64(off)) && ptr.Op != OpSB =>
   952  	(MOVHstoreconst [makeValAndOff(int32(int16(c)),off)] {sym} ptr mem)
   953  (MOVBstore [off] {sym} ptr (MOVDconst [c]) mem) && is20Bit(int64(off)) && ptr.Op != OpSB =>
   954  	(MOVBstoreconst [makeValAndOff(int32(int8(c)),off)] {sym} ptr mem)
   955  
   956  // Fold address offsets into constant stores.
   957  (MOVDstoreconst [sc] {s} (ADDconst [off] ptr) mem) && isU12Bit(sc.Off64()+int64(off)) =>
   958  	(MOVDstoreconst [sc.addOffset32(off)] {s} ptr mem)
   959  (MOVWstoreconst [sc] {s} (ADDconst [off] ptr) mem) && isU12Bit(sc.Off64()+int64(off)) =>
   960  	(MOVWstoreconst [sc.addOffset32(off)] {s} ptr mem)
   961  (MOVHstoreconst [sc] {s} (ADDconst [off] ptr) mem) && isU12Bit(sc.Off64()+int64(off)) =>
   962  	(MOVHstoreconst [sc.addOffset32(off)] {s} ptr mem)
   963  (MOVBstoreconst [sc] {s} (ADDconst [off] ptr) mem) && is20Bit(sc.Off64()+int64(off)) =>
   964  	(MOVBstoreconst [sc.addOffset32(off)] {s} ptr mem)
   965  
   966  // Merge address calculations into loads and stores.
   967  // Offsets from SB must not be merged into unaligned memory accesses because
   968  // loads/stores using PC-relative addressing directly must be aligned to the
   969  // size of the target.
   970  (MOVDload   [off1] {sym1} (MOVDaddr <t> [off2] {sym2} base) mem) && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) && (base.Op != OpSB || (t.IsPtr() && t.Elem().Alignment()%8 == 0 && (off1+off2)%8 == 0)) =>
   971  	(MOVDload  [off1+off2] {mergeSym(sym1,sym2)} base mem)
   972  (MOVWZload  [off1] {sym1} (MOVDaddr <t> [off2] {sym2} base) mem) && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) && (base.Op != OpSB || (t.IsPtr() && t.Elem().Alignment()%4 == 0 && (off1+off2)%4 == 0)) =>
   973  	(MOVWZload  [off1+off2] {mergeSym(sym1,sym2)} base mem)
   974  (MOVHZload  [off1] {sym1} (MOVDaddr <t> [off2] {sym2} base) mem) && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) && (base.Op != OpSB || (t.IsPtr() && t.Elem().Alignment()%2 == 0 && (off1+off2)%2 == 0)) =>
   975  	(MOVHZload  [off1+off2] {mergeSym(sym1,sym2)} base mem)
   976  (MOVBZload  [off1] {sym1} (MOVDaddr [off2] {sym2} base) mem) && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) =>
   977  	(MOVBZload  [off1+off2] {mergeSym(sym1,sym2)} base mem)
   978  (FMOVSload [off1] {sym1} (MOVDaddr [off2] {sym2} base) mem) && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) =>
   979  	(FMOVSload [off1+off2] {mergeSym(sym1,sym2)} base mem)
   980  (FMOVDload [off1] {sym1} (MOVDaddr [off2] {sym2} base) mem) && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) =>
   981  	(FMOVDload [off1+off2] {mergeSym(sym1,sym2)} base mem)
   982  
   983  (MOVWload [off1] {sym1} (MOVDaddr <t> [off2] {sym2} base) mem) && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) && (base.Op != OpSB || (t.IsPtr() && t.Elem().Alignment()%4 == 0 && (off1+off2)%4 == 0)) =>
   984  	(MOVWload [off1+off2] {mergeSym(sym1,sym2)} base mem)
   985  (MOVHload [off1] {sym1} (MOVDaddr <t> [off2] {sym2} base) mem) && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) && (base.Op != OpSB || (t.IsPtr() && t.Elem().Alignment()%2 == 0 && (off1+off2)%2 == 0)) =>
   986  	(MOVHload [off1+off2] {mergeSym(sym1,sym2)} base mem)
   987  (MOVBload [off1] {sym1} (MOVDaddr [off2] {sym2} base) mem) && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) =>
   988  	(MOVBload [off1+off2] {mergeSym(sym1,sym2)} base mem)
   989  
   990  (MOVDstore  [off1] {sym1} (MOVDaddr <t> [off2] {sym2} base) val mem) && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) && (base.Op != OpSB || (t.IsPtr() && t.Elem().Alignment()%8 == 0 && (off1+off2)%8 == 0)) =>
   991  	(MOVDstore  [off1+off2] {mergeSym(sym1,sym2)} base val mem)
   992  (MOVWstore  [off1] {sym1} (MOVDaddr <t> [off2] {sym2} base) val mem) && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) && (base.Op != OpSB || (t.IsPtr() && t.Elem().Alignment()%4 == 0 && (off1+off2)%4 == 0)) =>
   993  	(MOVWstore  [off1+off2] {mergeSym(sym1,sym2)} base val mem)
   994  (MOVHstore  [off1] {sym1} (MOVDaddr <t> [off2] {sym2} base) val mem) && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) && (base.Op != OpSB || (t.IsPtr() && t.Elem().Alignment()%2 == 0 && (off1+off2)%2 == 0)) =>
   995  	(MOVHstore  [off1+off2] {mergeSym(sym1,sym2)} base val mem)
   996  (MOVBstore  [off1] {sym1} (MOVDaddr [off2] {sym2} base) val mem) && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) =>
   997  	(MOVBstore  [off1+off2] {mergeSym(sym1,sym2)} base val mem)
   998  (FMOVSstore [off1] {sym1} (MOVDaddr [off2] {sym2} base) val mem) && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) =>
   999  	(FMOVSstore [off1+off2] {mergeSym(sym1,sym2)} base val mem)
  1000  (FMOVDstore [off1] {sym1} (MOVDaddr [off2] {sym2} base) val mem) && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) =>
  1001  	(FMOVDstore [off1+off2] {mergeSym(sym1,sym2)} base val mem)
  1002  
  1003  (ADDload   [o1] {s1} x (MOVDaddr [o2] {s2} ptr) mem) && ptr.Op != OpSB && is20Bit(int64(o1)+int64(o2)) && canMergeSym(s1, s2) => (ADDload   [o1+o2] {mergeSym(s1, s2)} x ptr mem)
  1004  (ADDWload  [o1] {s1} x (MOVDaddr [o2] {s2} ptr) mem) && ptr.Op != OpSB && is20Bit(int64(o1)+int64(o2)) && canMergeSym(s1, s2) => (ADDWload  [o1+o2] {mergeSym(s1, s2)} x ptr mem)
  1005  (MULLDload [o1] {s1} x (MOVDaddr [o2] {s2} ptr) mem) && ptr.Op != OpSB && is20Bit(int64(o1)+int64(o2)) && canMergeSym(s1, s2) => (MULLDload [o1+o2] {mergeSym(s1, s2)} x ptr mem)
  1006  (MULLWload [o1] {s1} x (MOVDaddr [o2] {s2} ptr) mem) && ptr.Op != OpSB && is20Bit(int64(o1)+int64(o2)) && canMergeSym(s1, s2) => (MULLWload [o1+o2] {mergeSym(s1, s2)} x ptr mem)
  1007  (SUBload   [o1] {s1} x (MOVDaddr [o2] {s2} ptr) mem) && ptr.Op != OpSB && is20Bit(int64(o1)+int64(o2)) && canMergeSym(s1, s2) => (SUBload   [o1+o2] {mergeSym(s1, s2)} x ptr mem)
  1008  (SUBWload  [o1] {s1} x (MOVDaddr [o2] {s2} ptr) mem) && ptr.Op != OpSB && is20Bit(int64(o1)+int64(o2)) && canMergeSym(s1, s2) => (SUBWload  [o1+o2] {mergeSym(s1, s2)} x ptr mem)
  1009  
  1010  (ANDload   [o1] {s1} x (MOVDaddr [o2] {s2} ptr) mem) && ptr.Op != OpSB && is20Bit(int64(o1)+int64(o2)) && canMergeSym(s1, s2) => (ANDload   [o1+o2] {mergeSym(s1, s2)} x ptr mem)
  1011  (ANDWload  [o1] {s1} x (MOVDaddr [o2] {s2} ptr) mem) && ptr.Op != OpSB && is20Bit(int64(o1)+int64(o2)) && canMergeSym(s1, s2) => (ANDWload  [o1+o2] {mergeSym(s1, s2)} x ptr mem)
  1012  (ORload    [o1] {s1} x (MOVDaddr [o2] {s2} ptr) mem) && ptr.Op != OpSB && is20Bit(int64(o1)+int64(o2)) && canMergeSym(s1, s2) => (ORload    [o1+o2] {mergeSym(s1, s2)} x ptr mem)
  1013  (ORWload   [o1] {s1} x (MOVDaddr [o2] {s2} ptr) mem) && ptr.Op != OpSB && is20Bit(int64(o1)+int64(o2)) && canMergeSym(s1, s2) => (ORWload   [o1+o2] {mergeSym(s1, s2)} x ptr mem)
  1014  (XORload   [o1] {s1} x (MOVDaddr [o2] {s2} ptr) mem) && ptr.Op != OpSB && is20Bit(int64(o1)+int64(o2)) && canMergeSym(s1, s2) => (XORload   [o1+o2] {mergeSym(s1, s2)} x ptr mem)
  1015  (XORWload  [o1] {s1} x (MOVDaddr [o2] {s2} ptr) mem) && ptr.Op != OpSB && is20Bit(int64(o1)+int64(o2)) && canMergeSym(s1, s2) => (XORWload  [o1+o2] {mergeSym(s1, s2)} x ptr mem)
  1016  
  1017  // Cannot store constant to SB directly (no 'move relative long immediate' instructions).
  1018  (MOVDstoreconst [sc] {sym1} (MOVDaddr [off] {sym2} ptr) mem) && ptr.Op != OpSB && canMergeSym(sym1, sym2) && sc.canAdd32(off) =>
  1019  	(MOVDstoreconst [sc.addOffset32(off)] {mergeSym(sym1, sym2)} ptr mem)
  1020  (MOVWstoreconst [sc] {sym1} (MOVDaddr [off] {sym2} ptr) mem) && ptr.Op != OpSB && canMergeSym(sym1, sym2) && sc.canAdd32(off) =>
  1021  	(MOVWstoreconst [sc.addOffset32(off)] {mergeSym(sym1, sym2)} ptr mem)
  1022  (MOVHstoreconst [sc] {sym1} (MOVDaddr [off] {sym2} ptr) mem) && ptr.Op != OpSB && canMergeSym(sym1, sym2) && sc.canAdd32(off) =>
  1023  	(MOVHstoreconst [sc.addOffset32(off)] {mergeSym(sym1, sym2)} ptr mem)
  1024  (MOVBstoreconst [sc] {sym1} (MOVDaddr [off] {sym2} ptr) mem) && ptr.Op != OpSB && canMergeSym(sym1, sym2) && sc.canAdd32(off) =>
  1025  	(MOVBstoreconst [sc.addOffset32(off)] {mergeSym(sym1, sym2)} ptr mem)
  1026  
  1027  // MOVDaddr into MOVDaddridx
  1028  (MOVDaddridx [off1] {sym1} (MOVDaddr [off2] {sym2} x) y) && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) && x.Op != OpSB =>
  1029         (MOVDaddridx [off1+off2] {mergeSym(sym1,sym2)} x y)
  1030  (MOVDaddridx [off1] {sym1} x (MOVDaddr [off2] {sym2} y)) && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) && y.Op != OpSB =>
  1031         (MOVDaddridx [off1+off2] {mergeSym(sym1,sym2)} x y)
  1032  
  1033  // Absorb InvertFlags into branches.
  1034  (BRC {c} (InvertFlags cmp) yes no) => (BRC {c.ReverseComparison()} cmp yes no)
  1035  
  1036  // Constant comparisons.
  1037  (CMPconst (MOVDconst [x]) [y]) && x==int64(y) => (FlagEQ)
  1038  (CMPconst (MOVDconst [x]) [y]) && x<int64(y) => (FlagLT)
  1039  (CMPconst (MOVDconst [x]) [y]) && x>int64(y) => (FlagGT)
  1040  (CMPUconst (MOVDconst [x]) [y]) && uint64(x)==uint64(y) => (FlagEQ)
  1041  (CMPUconst (MOVDconst [x]) [y]) && uint64(x)<uint64(y) => (FlagLT)
  1042  (CMPUconst (MOVDconst [x]) [y]) && uint64(x)>uint64(y) => (FlagGT)
  1043  
  1044  (CMPWconst (MOVDconst [x]) [y]) && int32(x)==int32(y) => (FlagEQ)
  1045  (CMPWconst (MOVDconst [x]) [y]) && int32(x)<int32(y) => (FlagLT)
  1046  (CMPWconst (MOVDconst [x]) [y]) && int32(x)>int32(y) => (FlagGT)
  1047  (CMPWUconst (MOVDconst [x]) [y]) && uint32(x)==uint32(y) => (FlagEQ)
  1048  (CMPWUconst (MOVDconst [x]) [y]) && uint32(x)<uint32(y) => (FlagLT)
  1049  (CMPWUconst (MOVDconst [x]) [y]) && uint32(x)>uint32(y) => (FlagGT)
  1050  
  1051  (CMP(W|WU)const (MOVBZreg _) [c]) &&   0xff < c => (FlagLT)
  1052  (CMP(W|WU)const (MOVHZreg _) [c]) && 0xffff < c => (FlagLT)
  1053  
  1054  (CMPconst  (SRDconst _ [c]) [n]) && c > 0 && n < 0 => (FlagGT)
  1055  (CMPWconst (SRWconst _ [c]) [n]) && c > 0 && n < 0 => (FlagGT)
  1056  
  1057  (CMPUconst  (SRDconst _ [c]) [n]) && c > 0 && c < 64 && (1<<uint(64-c)) <= uint64(n) => (FlagLT)
  1058  (CMPWUconst (SRWconst _ [c]) [n]) && c > 0 && c < 32 && (1<<uint(32-c)) <= uint32(n) => (FlagLT)
  1059  
  1060  (CMPWconst  (ANDWconst _ [m]) [n]) && int32(m) >= 0 &&  int32(m) <  int32(n) => (FlagLT)
  1061  (CMPWUconst (ANDWconst _ [m]) [n]) && uint32(m) < uint32(n) => (FlagLT)
  1062  
  1063  (CMPconst  (RISBGZ x {r}) [c]) && c > 0 && r.OutMask() < uint64(c) => (FlagLT)
  1064  (CMPUconst (RISBGZ x {r}) [c]) && r.OutMask() < uint64(uint32(c)) => (FlagLT)
  1065  
  1066  // Constant compare-and-branch with immediate.
  1067  (CGIJ  {c} (MOVDconst [x]) [y] yes no) && c&s390x.Equal   != 0 &&  int64(x) ==  int64(y) => (First yes no)
  1068  (CGIJ  {c} (MOVDconst [x]) [y] yes no) && c&s390x.Less    != 0 &&  int64(x) <   int64(y) => (First yes no)
  1069  (CGIJ  {c} (MOVDconst [x]) [y] yes no) && c&s390x.Greater != 0 &&  int64(x) >   int64(y) => (First yes no)
  1070  (CIJ   {c} (MOVDconst [x]) [y] yes no) && c&s390x.Equal   != 0 &&  int32(x) ==  int32(y) => (First yes no)
  1071  (CIJ   {c} (MOVDconst [x]) [y] yes no) && c&s390x.Less    != 0 &&  int32(x) <   int32(y) => (First yes no)
  1072  (CIJ   {c} (MOVDconst [x]) [y] yes no) && c&s390x.Greater != 0 &&  int32(x) >   int32(y) => (First yes no)
  1073  (CLGIJ {c} (MOVDconst [x]) [y] yes no) && c&s390x.Equal   != 0 && uint64(x) == uint64(y) => (First yes no)
  1074  (CLGIJ {c} (MOVDconst [x]) [y] yes no) && c&s390x.Less    != 0 && uint64(x) <  uint64(y) => (First yes no)
  1075  (CLGIJ {c} (MOVDconst [x]) [y] yes no) && c&s390x.Greater != 0 && uint64(x) >  uint64(y) => (First yes no)
  1076  (CLIJ  {c} (MOVDconst [x]) [y] yes no) && c&s390x.Equal   != 0 && uint32(x) == uint32(y) => (First yes no)
  1077  (CLIJ  {c} (MOVDconst [x]) [y] yes no) && c&s390x.Less    != 0 && uint32(x) <  uint32(y) => (First yes no)
  1078  (CLIJ  {c} (MOVDconst [x]) [y] yes no) && c&s390x.Greater != 0 && uint32(x) >  uint32(y) => (First yes no)
  1079  (CGIJ  {c} (MOVDconst [x]) [y] yes no) && c&s390x.Equal   == 0 &&  int64(x) ==  int64(y) => (First no yes)
  1080  (CGIJ  {c} (MOVDconst [x]) [y] yes no) && c&s390x.Less    == 0 &&  int64(x) <   int64(y) => (First no yes)
  1081  (CGIJ  {c} (MOVDconst [x]) [y] yes no) && c&s390x.Greater == 0 &&  int64(x) >   int64(y) => (First no yes)
  1082  (CIJ   {c} (MOVDconst [x]) [y] yes no) && c&s390x.Equal   == 0 &&  int32(x) ==  int32(y) => (First no yes)
  1083  (CIJ   {c} (MOVDconst [x]) [y] yes no) && c&s390x.Less    == 0 &&  int32(x) <   int32(y) => (First no yes)
  1084  (CIJ   {c} (MOVDconst [x]) [y] yes no) && c&s390x.Greater == 0 &&  int32(x) >   int32(y) => (First no yes)
  1085  (CLGIJ {c} (MOVDconst [x]) [y] yes no) && c&s390x.Equal   == 0 && uint64(x) == uint64(y) => (First no yes)
  1086  (CLGIJ {c} (MOVDconst [x]) [y] yes no) && c&s390x.Less    == 0 && uint64(x) <  uint64(y) => (First no yes)
  1087  (CLGIJ {c} (MOVDconst [x]) [y] yes no) && c&s390x.Greater == 0 && uint64(x) >  uint64(y) => (First no yes)
  1088  (CLIJ  {c} (MOVDconst [x]) [y] yes no) && c&s390x.Equal   == 0 && uint32(x) == uint32(y) => (First no yes)
  1089  (CLIJ  {c} (MOVDconst [x]) [y] yes no) && c&s390x.Less    == 0 && uint32(x) <  uint32(y) => (First no yes)
  1090  (CLIJ  {c} (MOVDconst [x]) [y] yes no) && c&s390x.Greater == 0 && uint32(x) >  uint32(y) => (First no yes)
  1091  
  1092  // Constant compare-and-branch with immediate when unsigned comparison with zero.
  1093  (C(L|LG)IJ {s390x.GreaterOrEqual} _ [0] yes no) => (First yes no)
  1094  (C(L|LG)IJ {s390x.Less}           _ [0] yes no) => (First no yes)
  1095  
  1096  // Constant compare-and-branch when operands match.
  1097  (C(GR|R|LGR|LR)J {c} x y yes no) && x == y && c&s390x.Equal != 0 => (First yes no)
  1098  (C(GR|R|LGR|LR)J {c} x y yes no) && x == y && c&s390x.Equal == 0 => (First no yes)
  1099  
  1100  // Convert 64-bit comparisons to 32-bit comparisons and signed comparisons
  1101  // to unsigned comparisons.
  1102  // Helps simplify constant comparison detection.
  1103  (CM(P|PU)const (MOV(W|WZ)reg x) [c]) => (CMP(W|WU)const x [c])
  1104  (CM(P|P|PU|PU)const x:(MOV(H|HZ|H|HZ)reg _) [c]) => (CMP(W|W|WU|WU)const x [c])
  1105  (CM(P|P|PU|PU)const x:(MOV(B|BZ|B|BZ)reg _) [c]) => (CMP(W|W|WU|WU)const x [c])
  1106  (CMPconst  (MOV(WZ|W)reg x:(ANDWconst [m] _)) [c]) && int32(m) >= 0 && c >= 0 => (CMPWUconst x [c])
  1107  (CMPUconst (MOV(WZ|W)reg x:(ANDWconst [m] _)) [c]) && int32(m) >= 0           => (CMPWUconst x [c])
  1108  (CMPconst  x:(SRDconst _ [c]) [n]) && c > 0 && n >= 0 => (CMPUconst  x [n])
  1109  (CMPWconst x:(SRWconst _ [c]) [n]) && c > 0 && n >= 0 => (CMPWUconst x [n])
  1110  
  1111  // Absorb sign and zero extensions into 32-bit comparisons.
  1112  (CMP(W|W|WU|WU)      x (MOV(W|WZ|W|WZ)reg y))   => (CMP(W|W|WU|WU) x y)
  1113  (CMP(W|W|WU|WU)      (MOV(W|WZ|W|WZ)reg x) y)   => (CMP(W|W|WU|WU) x y)
  1114  (CMP(W|W|WU|WU)const (MOV(W|WZ|W|WZ)reg x) [c]) => (CMP(W|W|WU|WU)const x [c])
  1115  
  1116  // Absorb flag constants into branches.
  1117  (BRC {c} (FlagEQ) yes no) && c&s390x.Equal     != 0 => (First yes no)
  1118  (BRC {c} (FlagLT) yes no) && c&s390x.Less      != 0 => (First yes no)
  1119  (BRC {c} (FlagGT) yes no) && c&s390x.Greater   != 0 => (First yes no)
  1120  (BRC {c} (FlagOV) yes no) && c&s390x.Unordered != 0 => (First yes no)
  1121  
  1122  (BRC {c} (FlagEQ) yes no) && c&s390x.Equal     == 0 => (First no yes)
  1123  (BRC {c} (FlagLT) yes no) && c&s390x.Less      == 0 => (First no yes)
  1124  (BRC {c} (FlagGT) yes no) && c&s390x.Greater   == 0 => (First no yes)
  1125  (BRC {c} (FlagOV) yes no) && c&s390x.Unordered == 0 => (First no yes)
  1126  
  1127  // Absorb flag constants into SETxx ops.
  1128  (LOCGR {c} _ x (FlagEQ)) && c&s390x.Equal     != 0 => x
  1129  (LOCGR {c} _ x (FlagLT)) && c&s390x.Less      != 0 => x
  1130  (LOCGR {c} _ x (FlagGT)) && c&s390x.Greater   != 0 => x
  1131  (LOCGR {c} _ x (FlagOV)) && c&s390x.Unordered != 0 => x
  1132  
  1133  (LOCGR {c} x _ (FlagEQ)) && c&s390x.Equal     == 0 => x
  1134  (LOCGR {c} x _ (FlagLT)) && c&s390x.Less      == 0 => x
  1135  (LOCGR {c} x _ (FlagGT)) && c&s390x.Greater   == 0 => x
  1136  (LOCGR {c} x _ (FlagOV)) && c&s390x.Unordered == 0 => x
  1137  
  1138  // Remove redundant *const ops
  1139  (ADDconst [0] x) => x
  1140  (ADDWconst [c] x) && int32(c)==0 => x
  1141  (SUBconst [0] x) => x
  1142  (SUBWconst [c] x) && int32(c) == 0 => x
  1143  (ANDconst [0] _)                 => (MOVDconst [0])
  1144  (ANDWconst [c] _) && int32(c)==0  => (MOVDconst [0])
  1145  (ANDconst [-1] x)                => x
  1146  (ANDWconst [c] x) && int32(c)==-1 => x
  1147  (ORconst [0] x)                  => x
  1148  (ORWconst [c] x) && int32(c)==0   => x
  1149  (ORconst [-1] _)                 => (MOVDconst [-1])
  1150  (ORWconst [c] _) && int32(c)==-1  => (MOVDconst [-1])
  1151  (XORconst [0] x)                  => x
  1152  (XORWconst [c] x) && int32(c)==0   => x
  1153  
  1154  // Shifts by zero (may be inserted during multiplication strength reduction).
  1155  ((SLD|SLW|SRD|SRW|SRAD|SRAW)const x [0]) => x
  1156  
  1157  // Convert constant subtracts to constant adds.
  1158  (SUBconst [c] x) && c != -(1<<31) => (ADDconst [-c] x)
  1159  (SUBWconst [c] x) => (ADDWconst [-int32(c)] x)
  1160  
  1161  // generic constant folding
  1162  // TODO: more of this
  1163  (ADDconst [c] (MOVDconst [d])) => (MOVDconst [int64(c)+d])
  1164  (ADDWconst [c] (MOVDconst [d])) => (MOVDconst [int64(c)+d])
  1165  (ADDconst [c] (ADDconst [d] x)) && is32Bit(int64(c)+int64(d)) => (ADDconst [c+d] x)
  1166  (ADDWconst [c] (ADDWconst [d] x)) => (ADDWconst [int32(c+d)] x)
  1167  (SUBconst (MOVDconst [d]) [c]) => (MOVDconst [d-int64(c)])
  1168  (SUBconst (SUBconst x [d]) [c]) && is32Bit(-int64(c)-int64(d)) => (ADDconst [-c-d] x)
  1169  (SRADconst [c] (MOVDconst [d])) => (MOVDconst [d>>uint64(c)])
  1170  (SRAWconst [c] (MOVDconst [d])) => (MOVDconst [int64(int32(d))>>uint64(c)])
  1171  (NEG (MOVDconst [c])) => (MOVDconst [-c])
  1172  (NEGW (MOVDconst [c])) => (MOVDconst [int64(int32(-c))])
  1173  (MULLDconst [c] (MOVDconst [d])) => (MOVDconst [int64(c)*d])
  1174  (MULLWconst [c] (MOVDconst [d])) => (MOVDconst [int64(c*int32(d))])
  1175  (AND (MOVDconst [c]) (MOVDconst [d])) => (MOVDconst [c&d])
  1176  (ANDconst [c] (MOVDconst [d])) => (MOVDconst [c&d])
  1177  (ANDWconst [c] (MOVDconst [d])) => (MOVDconst [int64(c)&d])
  1178  (OR (MOVDconst [c]) (MOVDconst [d])) => (MOVDconst [c|d])
  1179  (ORconst [c] (MOVDconst [d])) => (MOVDconst [c|d])
  1180  (ORWconst [c] (MOVDconst [d])) => (MOVDconst [int64(c)|d])
  1181  (XOR (MOVDconst [c]) (MOVDconst [d])) => (MOVDconst [c^d])
  1182  (XORconst [c] (MOVDconst [d])) => (MOVDconst [c^d])
  1183  (XORWconst [c] (MOVDconst [d])) => (MOVDconst [int64(c)^d])
  1184  (LoweredRound32F x:(FMOVSconst)) => x
  1185  (LoweredRound64F x:(FMOVDconst)) => x
  1186  
  1187  // generic simplifications
  1188  // TODO: more of this
  1189  (ADD x (NEG y)) => (SUB x y)
  1190  (ADDW x (NEGW y)) => (SUBW x y)
  1191  (SUB x x) => (MOVDconst [0])
  1192  (SUBW x x) => (MOVDconst [0])
  1193  (AND x x) => x
  1194  (ANDW x x) => x
  1195  (OR x x) => x
  1196  (ORW x x) => x
  1197  (XOR x x) => (MOVDconst [0])
  1198  (XORW x x) => (MOVDconst [0])
  1199  (NEG (ADDconst [c] (NEG x))) && c != -(1<<31) => (ADDconst [-c] x)
  1200  (MOVBZreg (ANDWconst [m] x)) => (MOVWZreg (ANDWconst <typ.UInt32> [int32( uint8(m))] x))
  1201  (MOVHZreg (ANDWconst [m] x)) => (MOVWZreg (ANDWconst <typ.UInt32> [int32(uint16(m))] x))
  1202  (MOVBreg  (ANDWconst [m] x)) &&  int8(m) >= 0 => (MOVWZreg (ANDWconst <typ.UInt32> [int32( uint8(m))] x))
  1203  (MOVHreg  (ANDWconst [m] x)) && int16(m) >= 0 => (MOVWZreg (ANDWconst <typ.UInt32> [int32(uint16(m))] x))
  1204  
  1205  // carry flag generation
  1206  // (only constant fold carry of zero)
  1207  (Select1 (ADDCconst (MOVDconst [c]) [d]))
  1208    && uint64(c+int64(d)) >= uint64(c) && c+int64(d) == 0
  1209    => (FlagEQ)
  1210  (Select1 (ADDCconst (MOVDconst [c]) [d]))
  1211    && uint64(c+int64(d)) >= uint64(c) && c+int64(d) != 0
  1212    => (FlagLT)
  1213  
  1214  // borrow flag generation
  1215  // (only constant fold borrow of zero)
  1216  (Select1 (SUBC (MOVDconst [c]) (MOVDconst [d])))
  1217    && uint64(d) <= uint64(c) && c-d == 0
  1218    => (FlagGT)
  1219  (Select1 (SUBC (MOVDconst [c]) (MOVDconst [d])))
  1220    && uint64(d) <= uint64(c) && c-d != 0
  1221    => (FlagOV)
  1222  
  1223  // add with carry
  1224  (ADDE x y (FlagEQ)) => (ADDC x y)
  1225  (ADDE x y (FlagLT)) => (ADDC x y)
  1226  (ADDC x (MOVDconst [c])) && is16Bit(c) => (ADDCconst x [int16(c)])
  1227  (Select0 (ADDCconst (MOVDconst [c]) [d])) => (MOVDconst [c+int64(d)])
  1228  
  1229  // subtract with borrow
  1230  (SUBE x y (FlagGT)) => (SUBC x y)
  1231  (SUBE x y (FlagOV)) => (SUBC x y)
  1232  (Select0 (SUBC (MOVDconst [c]) (MOVDconst [d]))) => (MOVDconst [c-d])
  1233  
  1234  // collapse carry chain
  1235  (ADDE x y (Select1 (ADDCconst [-1] (Select0 (ADDE (MOVDconst [0]) (MOVDconst [0]) c)))))
  1236    => (ADDE x y c)
  1237  
  1238  // collapse borrow chain
  1239  (SUBE x y (Select1 (SUBC (MOVDconst [0]) (NEG (Select0 (SUBE (MOVDconst [0]) (MOVDconst [0]) c))))))
  1240    => (SUBE x y c)
  1241  
  1242  // branch on carry
  1243  (C(G|LG)IJ {s390x.Equal}         (Select0 (ADDE (MOVDconst [0]) (MOVDconst [0]) carry)) [0]) => (BRC {s390x.NoCarry} carry)
  1244  (C(G|LG)IJ {s390x.Equal}         (Select0 (ADDE (MOVDconst [0]) (MOVDconst [0]) carry)) [1]) => (BRC {s390x.Carry}   carry)
  1245  (C(G|LG)IJ {s390x.LessOrGreater} (Select0 (ADDE (MOVDconst [0]) (MOVDconst [0]) carry)) [0]) => (BRC {s390x.Carry}   carry)
  1246  (C(G|LG)IJ {s390x.LessOrGreater} (Select0 (ADDE (MOVDconst [0]) (MOVDconst [0]) carry)) [1]) => (BRC {s390x.NoCarry} carry)
  1247  (C(G|LG)IJ {s390x.Greater}       (Select0 (ADDE (MOVDconst [0]) (MOVDconst [0]) carry)) [0]) => (BRC {s390x.Carry}   carry)
  1248  
  1249  // branch on borrow
  1250  (C(G|LG)IJ {s390x.Equal}         (NEG (Select0 (SUBE (MOVDconst [0]) (MOVDconst [0]) borrow))) [0]) => (BRC {s390x.NoBorrow} borrow)
  1251  (C(G|LG)IJ {s390x.Equal}         (NEG (Select0 (SUBE (MOVDconst [0]) (MOVDconst [0]) borrow))) [1]) => (BRC {s390x.Borrow}   borrow)
  1252  (C(G|LG)IJ {s390x.LessOrGreater} (NEG (Select0 (SUBE (MOVDconst [0]) (MOVDconst [0]) borrow))) [0]) => (BRC {s390x.Borrow}   borrow)
  1253  (C(G|LG)IJ {s390x.LessOrGreater} (NEG (Select0 (SUBE (MOVDconst [0]) (MOVDconst [0]) borrow))) [1]) => (BRC {s390x.NoBorrow} borrow)
  1254  (C(G|LG)IJ {s390x.Greater}       (NEG (Select0 (SUBE (MOVDconst [0]) (MOVDconst [0]) borrow))) [0]) => (BRC {s390x.Borrow}   borrow)
  1255  
  1256  // fused multiply-add
  1257  (Select0 (F(ADD|SUB) (FMUL y z) x)) && x.Block.Func.useFMA(v) => (FM(ADD|SUB) x y z)
  1258  (Select0 (F(ADDS|SUBS) (FMULS y z) x)) && x.Block.Func.useFMA(v) => (FM(ADDS|SUBS) x y z)
  1259  
  1260  // Convert floating point comparisons against zero into 'load and test' instructions.
  1261  (F(CMP|CMPS) x (FMOV(D|S)const [0.0])) => (LT(D|E)BR x)
  1262  (F(CMP|CMPS) (FMOV(D|S)const [0.0]) x) => (InvertFlags (LT(D|E)BR <v.Type> x))
  1263  
  1264  // FSUB, FSUBS, FADD, FADDS now produce a condition code representing the
  1265  // comparison of the result with 0.0. If a compare with zero instruction
  1266  // (e.g. LTDBR) is following one of those instructions, we can use the
  1267  // generated flag and remove the comparison instruction.
  1268  // Note: when inserting Select1 ops we need to ensure they are in the
  1269  // same block as their argument. We could also use @x.Block for this
  1270  // but moving the flag generating value to a different block seems to
  1271  // increase the likelihood that the flags value will have to be regenerated
  1272  // by flagalloc which is not what we want.
  1273  (LTDBR (Select0 x:(F(ADD|SUB) _ _)))   && b == x.Block => (Select1 x)
  1274  (LTEBR (Select0 x:(F(ADDS|SUBS) _ _))) && b == x.Block => (Select1 x)
  1275  
  1276  // Fold memory operations into operations.
  1277  // Exclude global data (SB) because these instructions cannot handle relative addresses.
  1278  // TODO(mundaym): indexed versions of these?
  1279  ((ADD|SUB|MULLD|AND|OR|XOR) <t> x g:(MOVDload [off] {sym} ptr mem))
  1280    && ptr.Op != OpSB
  1281    && is20Bit(int64(off))
  1282    && canMergeLoadClobber(v, g, x)
  1283    && clobber(g)
  1284    => ((ADD|SUB|MULLD|AND|OR|XOR)load <t> [off] {sym} x ptr mem)
  1285  ((ADD|SUB|MULL|AND|OR|XOR)W <t> x g:(MOVWload [off] {sym} ptr mem))
  1286    && ptr.Op != OpSB
  1287    && is20Bit(int64(off))
  1288    && canMergeLoadClobber(v, g, x)
  1289    && clobber(g)
  1290    => ((ADD|SUB|MULL|AND|OR|XOR)Wload <t> [off] {sym} x ptr mem)
  1291  ((ADD|SUB|MULL|AND|OR|XOR)W <t> x g:(MOVWZload [off] {sym} ptr mem))
  1292    && ptr.Op != OpSB
  1293    && is20Bit(int64(off))
  1294    && canMergeLoadClobber(v, g, x)
  1295    && clobber(g)
  1296    => ((ADD|SUB|MULL|AND|OR|XOR)Wload <t> [off] {sym} x ptr mem)
  1297  
  1298  // Combine stores into store multiples.
  1299  // 32-bit
  1300  (MOVWstore [i] {s} p w1 x:(MOVWstore [i-4] {s} p w0 mem))
  1301    && p.Op != OpSB
  1302    && x.Uses == 1
  1303    && is20Bit(int64(i)-4)
  1304    && setPos(v, x.Pos)
  1305    && clobber(x)
  1306    => (STM2 [i-4] {s} p w0 w1 mem)
  1307  (MOVWstore [i] {s} p w2 x:(STM2 [i-8] {s} p w0 w1 mem))
  1308    && x.Uses == 1
  1309    && is20Bit(int64(i)-8)
  1310    && setPos(v, x.Pos)
  1311    && clobber(x)
  1312    => (STM3 [i-8] {s} p w0 w1 w2 mem)
  1313  (MOVWstore [i] {s} p w3 x:(STM3 [i-12] {s} p w0 w1 w2 mem))
  1314    && x.Uses == 1
  1315    && is20Bit(int64(i)-12)
  1316    && setPos(v, x.Pos)
  1317    && clobber(x)
  1318    => (STM4 [i-12] {s} p w0 w1 w2 w3 mem)
  1319  (STM2 [i] {s} p w2 w3 x:(STM2 [i-8] {s} p w0 w1 mem))
  1320    && x.Uses == 1
  1321    && is20Bit(int64(i)-8)
  1322    && setPos(v, x.Pos)
  1323    && clobber(x)
  1324    => (STM4 [i-8] {s} p w0 w1 w2 w3 mem)
  1325  // 64-bit
  1326  (MOVDstore [i] {s} p w1 x:(MOVDstore [i-8] {s} p w0 mem))
  1327    && p.Op != OpSB
  1328    && x.Uses == 1
  1329    && is20Bit(int64(i)-8)
  1330    && setPos(v, x.Pos)
  1331    && clobber(x)
  1332    => (STMG2 [i-8] {s} p w0 w1 mem)
  1333  (MOVDstore [i] {s} p w2 x:(STMG2 [i-16] {s} p w0 w1 mem))
  1334    && x.Uses == 1
  1335    && is20Bit(int64(i)-16)
  1336    && setPos(v, x.Pos)
  1337    && clobber(x)
  1338    => (STMG3 [i-16] {s} p w0 w1 w2 mem)
  1339  (MOVDstore [i] {s} p w3 x:(STMG3 [i-24] {s} p w0 w1 w2 mem))
  1340    && x.Uses == 1
  1341    && is20Bit(int64(i)-24)
  1342    && setPos(v, x.Pos)
  1343    && clobber(x)
  1344    => (STMG4 [i-24] {s} p w0 w1 w2 w3 mem)
  1345  (STMG2 [i] {s} p w2 w3 x:(STMG2 [i-16] {s} p w0 w1 mem))
  1346    && x.Uses == 1
  1347    && is20Bit(int64(i)-16)
  1348    && setPos(v, x.Pos)
  1349    && clobber(x)
  1350    => (STMG4 [i-16] {s} p w0 w1 w2 w3 mem)
  1351  
  1352  // Convert 32-bit store multiples into 64-bit stores.
  1353  (STM2 [i] {s} p (SRDconst [32] x) x mem) => (MOVDstore [i] {s} p x mem)
  1354  
  1355  // Fold bit reversal into loads.
  1356  (MOVWBR x:(MOVWZload    [off] {sym} ptr     mem)) && x.Uses == 1 => @x.Block (MOVWZreg (MOVWBRload    [off] {sym} ptr     mem)) // need zero extension?
  1357  (MOVWBR x:(MOVWZloadidx [off] {sym} ptr idx mem)) && x.Uses == 1 => @x.Block (MOVWZreg (MOVWBRloadidx [off] {sym} ptr idx mem)) // need zero extension?
  1358  (MOVDBR x:(MOVDload     [off] {sym} ptr     mem)) && x.Uses == 1 => @x.Block (MOVDBRload    [off] {sym} ptr     mem)
  1359  (MOVDBR x:(MOVDloadidx  [off] {sym} ptr idx mem)) && x.Uses == 1 => @x.Block (MOVDBRloadidx [off] {sym} ptr idx mem)
  1360  
  1361  // Fold bit reversal into stores.
  1362  (MOV(D|W)store    [off] {sym} ptr     r:(MOV(D|W)BR x) mem) && r.Uses == 1 => (MOV(D|W)BRstore    [off] {sym} ptr     x mem)
  1363  (MOV(D|W)storeidx [off] {sym} ptr idx r:(MOV(D|W)BR x) mem) && r.Uses == 1 => (MOV(D|W)BRstoreidx [off] {sym} ptr idx x mem)
  1364  
  1365  // Special bswap16 rules
  1366  (Bswap16 x:(MOVHZload    [off] {sym} ptr     mem)) => @x.Block (MOVHZreg (MOVHBRload    [off] {sym} ptr     mem))
  1367  (Bswap16 x:(MOVHZloadidx [off] {sym} ptr idx mem)) => @x.Block (MOVHZreg (MOVHBRloadidx [off] {sym} ptr idx mem))
  1368  (MOVHstore    [off] {sym} ptr     (Bswap16 val) mem) => (MOVHBRstore    [off] {sym} ptr     val mem)
  1369  (MOVHstoreidx [off] {sym} ptr idx (Bswap16 val) mem) => (MOVHBRstoreidx [off] {sym} ptr idx val mem)
  1370  

View as plain text