github.com/Filosottile/go@v0.0.0-20170906193555-dbed9972d994/src/cmd/compile/internal/ssa/gen/AMD64.rules

// Copyright 2015 The Go Authors. All rights reserved.
// Use of this source code is governed by a BSD-style
// license that can be found in the LICENSE file.

// Lowering arithmetic
(Add64  x y) -> (ADDQ  x y)
(AddPtr x y) && config.PtrSize == 8 -> (ADDQ x y)
(AddPtr x y) && config.PtrSize == 4 -> (ADDL x y)
(Add32  x y) -> (ADDL  x y)
(Add16  x y) -> (ADDL  x y)
(Add8   x y) -> (ADDL  x y)
(Add32F x y) -> (ADDSS x y)
(Add64F x y) -> (ADDSD x y)

(Sub64  x y) -> (SUBQ  x y)
(SubPtr x y) && config.PtrSize == 8 -> (SUBQ x y)
(SubPtr x y) && config.PtrSize == 4 -> (SUBL x y)
(Sub32  x y) -> (SUBL  x y)
(Sub16  x y) -> (SUBL  x y)
(Sub8   x y) -> (SUBL  x y)
(Sub32F x y) -> (SUBSS x y)
(Sub64F x y) -> (SUBSD x y)

(Mul64  x y) -> (MULQ  x y)
(Mul32  x y) -> (MULL  x y)
(Mul16  x y) -> (MULL  x y)
(Mul8   x y) -> (MULL  x y)
(Mul32F x y) -> (MULSS x y)
(Mul64F x y) -> (MULSD x y)

(Div32F x y) -> (DIVSS x y)
(Div64F x y) -> (DIVSD x y)

(Div64  x y) -> (Select0 (DIVQ  x y))
(Div64u x y) -> (Select0 (DIVQU x y))
(Div32  x y) -> (Select0 (DIVL  x y))
(Div32u x y) -> (Select0 (DIVLU x y))
(Div16  x y) -> (Select0 (DIVW  x y))
(Div16u x y) -> (Select0 (DIVWU x y))
(Div8   x y) -> (Select0 (DIVW  (SignExt8to16 x) (SignExt8to16 y)))
(Div8u  x y) -> (Select0 (DIVWU (ZeroExt8to16 x) (ZeroExt8to16 y)))

(Hmul64  x y) -> (HMULQ  x y)
(Hmul64u x y) -> (HMULQU x y)
(Hmul32  x y) -> (HMULL  x y)
(Hmul32u x y) -> (HMULLU x y)

(Mul64uhilo x y) -> (MULQU2 x y)
(Div128u xhi xlo y) -> (DIVQU2 xhi xlo y)

(Avg64u x y) -> (AVGQU x y)

(Mod64  x y) -> (Select1 (DIVQ  x y))
(Mod64u x y) -> (Select1 (DIVQU x y))
(Mod32  x y) -> (Select1 (DIVL  x y))
(Mod32u x y) -> (Select1 (DIVLU x y))
(Mod16  x y) -> (Select1 (DIVW  x y))
(Mod16u x y) -> (Select1 (DIVWU x y))
(Mod8   x y) -> (Select1 (DIVW  (SignExt8to16 x) (SignExt8to16 y)))
(Mod8u  x y) -> (Select1 (DIVWU (ZeroExt8to16 x) (ZeroExt8to16 y)))

(And64 x y) -> (ANDQ x y)
(And32 x y) -> (ANDL x y)
(And16 x y) -> (ANDL x y)
(And8  x y) -> (ANDL x y)

(Or64 x y) -> (ORQ x y)
(Or32 x y) -> (ORL x y)
(Or16 x y) -> (ORL x y)
(Or8  x y) -> (ORL x y)

(Xor64 x y) -> (XORQ x y)
(Xor32 x y) -> (XORL x y)
(Xor16 x y) -> (XORL x y)
(Xor8  x y) -> (XORL x y)

(Neg64  x) -> (NEGQ x)
(Neg32  x) -> (NEGL x)
(Neg16  x) -> (NEGL x)
(Neg8   x) -> (NEGL x)
(Neg32F x) -> (PXOR x (MOVSSconst <typ.Float32> [f2i(math.Copysign(0, -1))]))
(Neg64F x) -> (PXOR x (MOVSDconst <typ.Float64> [f2i(math.Copysign(0, -1))]))

(Com64 x) -> (NOTQ x)
(Com32 x) -> (NOTL x)
(Com16 x) -> (NOTL x)
(Com8  x) -> (NOTL x)

// Lowering boolean ops
(AndB x y) -> (ANDL x y)
(OrB x y) -> (ORL x y)
(Not x) -> (XORLconst [1] x)

// Lowering pointer arithmetic
(OffPtr [off] ptr) && config.PtrSize == 8 && is32Bit(off) -> (ADDQconst [off] ptr)
(OffPtr [off] ptr) && config.PtrSize == 8 -> (ADDQ (MOVQconst [off]) ptr)
(OffPtr [off] ptr) && config.PtrSize == 4 -> (ADDLconst [off] ptr)

// Lowering other arithmetic
(Ctz64 <t> x) -> (CMOVQEQ (Select0 <t> (BSFQ x)) (MOVQconst <t> [64]) (Select1 <types.TypeFlags> (BSFQ x)))
(Ctz32 x) -> (Select0 (BSFQ (ORQ <typ.UInt64> (MOVQconst [1<<32]) x)))

(BitLen64 <t> x) -> (ADDQconst [1] (CMOVQEQ <t> (Select0 <t> (BSRQ x)) (MOVQconst <t> [-1]) (Select1 <types.TypeFlags> (BSRQ x))))
(BitLen32 x) -> (BitLen64 (MOVLQZX <typ.UInt64> x))

(Bswap64 x) -> (BSWAPQ x)
(Bswap32 x) -> (BSWAPL x)

(PopCount64 x) -> (POPCNTQ x)
(PopCount32 x) -> (POPCNTL x)
(PopCount16 x) -> (POPCNTL (MOVWQZX <typ.UInt32> x))
(PopCount8 x) -> (POPCNTL (MOVBQZX <typ.UInt32> x))

(Sqrt x) -> (SQRTSD x)

// Lowering extension
// Note: we always extend to 64 bits even though some ops don't need that many result bits.
(SignExt8to16  x) -> (MOVBQSX x)
(SignExt8to32  x) -> (MOVBQSX x)
(SignExt8to64  x) -> (MOVBQSX x)
(SignExt16to32 x) -> (MOVWQSX x)
(SignExt16to64 x) -> (MOVWQSX x)
(SignExt32to64 x) -> (MOVLQSX x)

(ZeroExt8to16  x) -> (MOVBQZX x)
(ZeroExt8to32  x) -> (MOVBQZX x)
(ZeroExt8to64  x) -> (MOVBQZX x)
(ZeroExt16to32 x) -> (MOVWQZX x)
(ZeroExt16to64 x) -> (MOVWQZX x)
(ZeroExt32to64 x) -> (MOVLQZX x)

(Slicemask <t> x) -> (SARQconst (NEGQ <t> x) [63])

// Lowering truncation
// Because we ignore high parts of registers, truncates are just copies.
(Trunc16to8  x) -> x
(Trunc32to8  x) -> x
(Trunc32to16 x) -> x
(Trunc64to8  x) -> x
(Trunc64to16 x) -> x
(Trunc64to32 x) -> x

// Lowering float <-> int
(Cvt32to32F x) -> (CVTSL2SS x)
(Cvt32to64F x) -> (CVTSL2SD x)
(Cvt64to32F x) -> (CVTSQ2SS x)
(Cvt64to64F x) -> (CVTSQ2SD x)

(Cvt32Fto32 x) -> (CVTTSS2SL x)
(Cvt32Fto64 x) -> (CVTTSS2SQ x)
(Cvt64Fto32 x) -> (CVTTSD2SL x)
(Cvt64Fto64 x) -> (CVTTSD2SQ x)

(Cvt32Fto64F x) -> (CVTSS2SD x)
(Cvt64Fto32F x) -> (CVTSD2SS x)

(Round32F x) -> x
(Round64F x) -> x

// Lowering shifts
// Unsigned shifts need to return 0 if shift amount is >= width of shifted value.
//   result = (arg << shift) & (shift >= argbits ? 0 : 0xffffffffffffffff)
(Lsh64x64 <t> x y) -> (ANDQ (SHLQ <t> x y) (SBBQcarrymask <t> (CMPQconst y [64])))
(Lsh64x32 <t> x y) -> (ANDQ (SHLQ <t> x y) (SBBQcarrymask <t> (CMPLconst y [64])))
(Lsh64x16 <t> x y) -> (ANDQ (SHLQ <t> x y) (SBBQcarrymask <t> (CMPWconst y [64])))
(Lsh64x8  <t> x y) -> (ANDQ (SHLQ <t> x y) (SBBQcarrymask <t> (CMPBconst y [64])))

(Lsh32x64 <t> x y) -> (ANDL (SHLL <t> x y) (SBBLcarrymask <t> (CMPQconst y [32])))
(Lsh32x32 <t> x y) -> (ANDL (SHLL <t> x y) (SBBLcarrymask <t> (CMPLconst y [32])))
(Lsh32x16 <t> x y) -> (ANDL (SHLL <t> x y) (SBBLcarrymask <t> (CMPWconst y [32])))
(Lsh32x8  <t> x y) -> (ANDL (SHLL <t> x y) (SBBLcarrymask <t> (CMPBconst y [32])))

(Lsh16x64 <t> x y) -> (ANDL (SHLL <t> x y) (SBBLcarrymask <t> (CMPQconst y [32])))
(Lsh16x32 <t> x y) -> (ANDL (SHLL <t> x y) (SBBLcarrymask <t> (CMPLconst y [32])))
(Lsh16x16 <t> x y) -> (ANDL (SHLL <t> x y) (SBBLcarrymask <t> (CMPWconst y [32])))
(Lsh16x8  <t> x y) -> (ANDL (SHLL <t> x y) (SBBLcarrymask <t> (CMPBconst y [32])))

(Lsh8x64 <t> x y)  -> (ANDL (SHLL <t> x y) (SBBLcarrymask <t> (CMPQconst y [32])))
(Lsh8x32 <t> x y)  -> (ANDL (SHLL <t> x y) (SBBLcarrymask <t> (CMPLconst y [32])))
(Lsh8x16 <t> x y)  -> (ANDL (SHLL <t> x y) (SBBLcarrymask <t> (CMPWconst y [32])))
(Lsh8x8  <t> x y)  -> (ANDL (SHLL <t> x y) (SBBLcarrymask <t> (CMPBconst y [32])))

(Rsh64Ux64 <t> x y) -> (ANDQ (SHRQ <t> x y) (SBBQcarrymask <t> (CMPQconst y [64])))
(Rsh64Ux32 <t> x y) -> (ANDQ (SHRQ <t> x y) (SBBQcarrymask <t> (CMPLconst y [64])))
(Rsh64Ux16 <t> x y) -> (ANDQ (SHRQ <t> x y) (SBBQcarrymask <t> (CMPWconst y [64])))
(Rsh64Ux8  <t> x y) -> (ANDQ (SHRQ <t> x y) (SBBQcarrymask <t> (CMPBconst y [64])))

(Rsh32Ux64 <t> x y) -> (ANDL (SHRL <t> x y) (SBBLcarrymask <t> (CMPQconst y [32])))
(Rsh32Ux32 <t> x y) -> (ANDL (SHRL <t> x y) (SBBLcarrymask <t> (CMPLconst y [32])))
(Rsh32Ux16 <t> x y) -> (ANDL (SHRL <t> x y) (SBBLcarrymask <t> (CMPWconst y [32])))
(Rsh32Ux8  <t> x y) -> (ANDL (SHRL <t> x y) (SBBLcarrymask <t> (CMPBconst y [32])))

(Rsh16Ux64 <t> x y) -> (ANDL (SHRW <t> x y) (SBBLcarrymask <t> (CMPQconst y [16])))
(Rsh16Ux32 <t> x y) -> (ANDL (SHRW <t> x y) (SBBLcarrymask <t> (CMPLconst y [16])))
(Rsh16Ux16 <t> x y) -> (ANDL (SHRW <t> x y) (SBBLcarrymask <t> (CMPWconst y [16])))
(Rsh16Ux8  <t> x y) -> (ANDL (SHRW <t> x y) (SBBLcarrymask <t> (CMPBconst y [16])))

(Rsh8Ux64 <t> x y)  -> (ANDL (SHRB <t> x y) (SBBLcarrymask <t> (CMPQconst y [8])))
(Rsh8Ux32 <t> x y)  -> (ANDL (SHRB <t> x y) (SBBLcarrymask <t> (CMPLconst y [8])))
(Rsh8Ux16 <t> x y)  -> (ANDL (SHRB <t> x y) (SBBLcarrymask <t> (CMPWconst y [8])))
(Rsh8Ux8  <t> x y)  -> (ANDL (SHRB <t> x y) (SBBLcarrymask <t> (CMPBconst y [8])))

// Signed right shift needs to return 0/-1 if shift amount is >= width of shifted value.
// We implement this by setting the shift value to -1 (all ones) if the shift value is >= width.
(Rsh64x64 <t> x y) -> (SARQ <t> x (ORQ <y.Type> y (NOTQ <y.Type> (SBBQcarrymask <y.Type> (CMPQconst y [64])))))
(Rsh64x32 <t> x y) -> (SARQ <t> x (ORL <y.Type> y (NOTL <y.Type> (SBBLcarrymask <y.Type> (CMPLconst y [64])))))
(Rsh64x16 <t> x y) -> (SARQ <t> x (ORL <y.Type> y (NOTL <y.Type> (SBBLcarrymask <y.Type> (CMPWconst y [64])))))
(Rsh64x8  <t> x y) -> (SARQ <t> x (ORL <y.Type> y (NOTL <y.Type> (SBBLcarrymask <y.Type> (CMPBconst y [64])))))

(Rsh32x64 <t> x y) -> (SARL <t> x (ORQ <y.Type> y (NOTQ <y.Type> (SBBQcarrymask <y.Type> (CMPQconst y [32])))))
(Rsh32x32 <t> x y) -> (SARL <t> x (ORL <y.Type> y (NOTL <y.Type> (SBBLcarrymask <y.Type> (CMPLconst y [32])))))
(Rsh32x16 <t> x y) -> (SARL <t> x (ORL <y.Type> y (NOTL <y.Type> (SBBLcarrymask <y.Type> (CMPWconst y [32])))))
(Rsh32x8  <t> x y) -> (SARL <t> x (ORL <y.Type> y (NOTL <y.Type> (SBBLcarrymask <y.Type> (CMPBconst y [32])))))

(Rsh16x64 <t> x y) -> (SARW <t> x (ORQ <y.Type> y (NOTQ <y.Type> (SBBQcarrymask <y.Type> (CMPQconst y [16])))))
(Rsh16x32 <t> x y) -> (SARW <t> x (ORL <y.Type> y (NOTL <y.Type> (SBBLcarrymask <y.Type> (CMPLconst y [16])))))
(Rsh16x16 <t> x y) -> (SARW <t> x (ORL <y.Type> y (NOTL <y.Type> (SBBLcarrymask <y.Type> (CMPWconst y [16])))))
(Rsh16x8  <t> x y) -> (SARW <t> x (ORL <y.Type> y (NOTL <y.Type> (SBBLcarrymask <y.Type> (CMPBconst y [16])))))

(Rsh8x64 <t> x y)  -> (SARB <t> x (ORQ <y.Type> y (NOTQ <y.Type> (SBBQcarrymask <y.Type> (CMPQconst y [8])))))
(Rsh8x32 <t> x y)  -> (SARB <t> x (ORL <y.Type> y (NOTL <y.Type> (SBBLcarrymask <y.Type> (CMPLconst y [8])))))
(Rsh8x16 <t> x y)  -> (SARB <t> x (ORL <y.Type> y (NOTL <y.Type> (SBBLcarrymask <y.Type> (CMPWconst y [8])))))
(Rsh8x8  <t> x y)  -> (SARB <t> x (ORL <y.Type> y (NOTL <y.Type> (SBBLcarrymask <y.Type> (CMPBconst y [8])))))

// Lowering comparisons
(Less64  x y) -> (SETL (CMPQ x y))
(Less32  x y) -> (SETL (CMPL x y))
(Less16  x y) -> (SETL (CMPW x y))
(Less8   x y) -> (SETL (CMPB x y))
(Less64U x y) -> (SETB (CMPQ x y))
(Less32U x y) -> (SETB (CMPL x y))
(Less16U x y) -> (SETB (CMPW x y))
(Less8U  x y) -> (SETB (CMPB x y))
// Use SETGF with reversed operands to dodge NaN case
(Less64F x y) -> (SETGF (UCOMISD y x))
(Less32F x y) -> (SETGF (UCOMISS y x))

(Leq64  x y) -> (SETLE (CMPQ x y))
(Leq32  x y) -> (SETLE (CMPL x y))
(Leq16  x y) -> (SETLE (CMPW x y))
(Leq8   x y) -> (SETLE (CMPB x y))
(Leq64U x y) -> (SETBE (CMPQ x y))
(Leq32U x y) -> (SETBE (CMPL x y))
(Leq16U x y) -> (SETBE (CMPW x y))
(Leq8U  x y) -> (SETBE (CMPB x y))
// Use SETGEF with reversed operands to dodge NaN case
(Leq64F x y) -> (SETGEF (UCOMISD y x))
(Leq32F x y) -> (SETGEF (UCOMISS y x))

(Greater64  x y) -> (SETG (CMPQ x y))
(Greater32  x y) -> (SETG (CMPL x y))
(Greater16  x y) -> (SETG (CMPW x y))
(Greater8   x y) -> (SETG (CMPB x y))
(Greater64U x y) -> (SETA (CMPQ x y))
(Greater32U x y) -> (SETA (CMPL x y))
(Greater16U x y) -> (SETA (CMPW x y))
(Greater8U  x y) -> (SETA (CMPB x y))
// Note Go assembler gets UCOMISx operand order wrong, but it is right here
// Bug is accommodated at generation of assembly language.
(Greater64F x y) -> (SETGF (UCOMISD x y))
(Greater32F x y) -> (SETGF (UCOMISS x y))

(Geq64  x y) -> (SETGE (CMPQ x y))
(Geq32  x y) -> (SETGE (CMPL x y))
(Geq16  x y) -> (SETGE (CMPW x y))
(Geq8   x y) -> (SETGE (CMPB x y))
(Geq64U x y) -> (SETAE (CMPQ x y))
(Geq32U x y) -> (SETAE (CMPL x y))
(Geq16U x y) -> (SETAE (CMPW x y))
(Geq8U  x y) -> (SETAE (CMPB x y))
// Note Go assembler gets UCOMISx operand order wrong, but it is right here
// Bug is accommodated at generation of assembly language.
(Geq64F x y) -> (SETGEF (UCOMISD x y))
(Geq32F x y) -> (SETGEF (UCOMISS x y))

(Eq64  x y) -> (SETEQ (CMPQ x y))
(Eq32  x y) -> (SETEQ (CMPL x y))
(Eq16  x y) -> (SETEQ (CMPW x y))
(Eq8   x y) -> (SETEQ (CMPB x y))
(EqB   x y) -> (SETEQ (CMPB x y))
(EqPtr x y) && config.PtrSize == 8 -> (SETEQ (CMPQ x y))
(EqPtr x y) && config.PtrSize == 4 -> (SETEQ (CMPL x y))
(Eq64F x y) -> (SETEQF (UCOMISD x y))
(Eq32F x y) -> (SETEQF (UCOMISS x y))

(Neq64  x y) -> (SETNE (CMPQ x y))
(Neq32  x y) -> (SETNE (CMPL x y))
(Neq16  x y) -> (SETNE (CMPW x y))
(Neq8   x y) -> (SETNE (CMPB x y))
(NeqB   x y) -> (SETNE (CMPB x y))
(NeqPtr x y) && config.PtrSize == 8 -> (SETNE (CMPQ x y))
(NeqPtr x y) && config.PtrSize == 4 -> (SETNE (CMPL x y))
(Neq64F x y) -> (SETNEF (UCOMISD x y))
(Neq32F x y) -> (SETNEF (UCOMISS x y))

(Int64Hi x) -> (SHRQconst [32] x) // needed for amd64p32

// Lowering loads
(Load <t> ptr mem) && (is64BitInt(t) || isPtr(t) && config.PtrSize == 8) -> (MOVQload ptr mem)
(Load <t> ptr mem) && (is32BitInt(t) || isPtr(t) && config.PtrSize == 4) -> (MOVLload ptr mem)
(Load <t> ptr mem) && is16BitInt(t) -> (MOVWload ptr mem)
(Load <t> ptr mem) && (t.IsBoolean() || is8BitInt(t)) -> (MOVBload ptr mem)
(Load <t> ptr mem) && is32BitFloat(t) -> (MOVSSload ptr mem)
(Load <t> ptr mem) && is64BitFloat(t) -> (MOVSDload ptr mem)

// Lowering stores
// These more-specific FP versions of Store pattern should come first.
(Store {t} ptr val mem) && t.(*types.Type).Size() == 8 && is64BitFloat(val.Type) -> (MOVSDstore ptr val mem)
(Store {t} ptr val mem) && t.(*types.Type).Size() == 4 && is32BitFloat(val.Type) -> (MOVSSstore ptr val mem)

(Store {t} ptr val mem) && t.(*types.Type).Size() == 8 -> (MOVQstore ptr val mem)
(Store {t} ptr val mem) && t.(*types.Type).Size() == 4 -> (MOVLstore ptr val mem)
(Store {t} ptr val mem) && t.(*types.Type).Size() == 2 -> (MOVWstore ptr val mem)
(Store {t} ptr val mem) && t.(*types.Type).Size() == 1 -> (MOVBstore ptr val mem)

// Lowering moves
(Move [0] _ _ mem) -> mem
(Move [1] dst src mem) -> (MOVBstore dst (MOVBload src mem) mem)
(Move [2] dst src mem) -> (MOVWstore dst (MOVWload src mem) mem)
(Move [4] dst src mem) -> (MOVLstore dst (MOVLload src mem) mem)
(Move [8] dst src mem) -> (MOVQstore dst (MOVQload src mem) mem)
(Move [16] dst src mem) && config.useSSE -> (MOVOstore dst (MOVOload src mem) mem)
(Move [16] dst src mem) && !config.useSSE ->
	(MOVQstore [8] dst (MOVQload [8] src mem)
		(MOVQstore dst (MOVQload src mem) mem))
(Move [3] dst src mem) ->
	(MOVBstore [2] dst (MOVBload [2] src mem)
		(MOVWstore dst (MOVWload src mem) mem))
(Move [5] dst src mem) ->
	(MOVBstore [4] dst (MOVBload [4] src mem)
		(MOVLstore dst (MOVLload src mem) mem))
(Move [6] dst src mem) ->
	(MOVWstore [4] dst (MOVWload [4] src mem)
		(MOVLstore dst (MOVLload src mem) mem))
(Move [7] dst src mem) ->
	(MOVLstore [3] dst (MOVLload [3] src mem)
		(MOVLstore dst (MOVLload src mem) mem))
(Move [s] dst src mem) && s > 8 && s < 16 ->
	(MOVQstore [s-8] dst (MOVQload [s-8] src mem)
		(MOVQstore dst (MOVQload src mem) mem))

// Adjust moves to be a multiple of 16 bytes.
(Move [s] dst src mem)
	&& s > 16 && s%16 != 0 && s%16 <= 8 ->
	(Move [s-s%16]
		(OffPtr <dst.Type> dst [s%16])
		(OffPtr <src.Type> src [s%16])
		(MOVQstore dst (MOVQload src mem) mem))
(Move [s] dst src mem)
	&& s > 16 && s%16 != 0 && s%16 > 8 && config.useSSE ->
	(Move [s-s%16]
		(OffPtr <dst.Type> dst [s%16])
		(OffPtr <src.Type> src [s%16])
		(MOVOstore dst (MOVOload src mem) mem))
(Move [s] dst src mem)
	&& s > 16 && s%16 != 0 && s%16 > 8 && !config.useSSE ->
	(Move [s-s%16]
		(OffPtr <dst.Type> dst [s%16])
		(OffPtr <src.Type> src [s%16])
		(MOVQstore [8] dst (MOVQload [8] src mem)
			(MOVQstore dst (MOVQload src mem) mem)))

// Medium copying uses a duff device.
(Move [s] dst src mem)
	&& s >= 32 && s <= 16*64 && s%16 == 0
	&& !config.noDuffDevice ->
	(DUFFCOPY [14*(64-s/16)] dst src mem)
// 14 and 64 are magic constants.  14 is the number of bytes to encode:
//	MOVUPS	(SI), X0
//	ADDQ	$16, SI
//	MOVUPS	X0, (DI)
//	ADDQ	$16, DI
// and 64 is the number of such blocks. See src/runtime/duff_amd64.s:duffcopy.

// Large copying uses REP MOVSQ.
(Move [s] dst src mem) && (s > 16*64 || config.noDuffDevice) && s%8 == 0 ->
	(REPMOVSQ dst src (MOVQconst [s/8]) mem)

// Lowering Zero instructions
(Zero [0] _ mem) -> mem
(Zero [1] destptr mem) -> (MOVBstoreconst [0] destptr mem)
(Zero [2] destptr mem) -> (MOVWstoreconst [0] destptr mem)
(Zero [4] destptr mem) -> (MOVLstoreconst [0] destptr mem)
(Zero [8] destptr mem) -> (MOVQstoreconst [0] destptr mem)

(Zero [3] destptr mem) ->
	(MOVBstoreconst [makeValAndOff(0,2)] destptr
		(MOVWstoreconst [0] destptr mem))
(Zero [5] destptr mem) ->
	(MOVBstoreconst [makeValAndOff(0,4)] destptr
		(MOVLstoreconst [0] destptr mem))
(Zero [6] destptr mem) ->
	(MOVWstoreconst [makeValAndOff(0,4)] destptr
		(MOVLstoreconst [0] destptr mem))
(Zero [7] destptr mem) ->
	(MOVLstoreconst [makeValAndOff(0,3)] destptr
		(MOVLstoreconst [0] destptr mem))

// Strip off any fractional word zeroing.
(Zero [s] destptr mem) && s%8 != 0 && s > 8 && !config.useSSE ->
	(Zero [s-s%8] (OffPtr <destptr.Type> destptr [s%8])
		(MOVQstoreconst [0] destptr mem))

// Zero small numbers of words directly.
(Zero [16] destptr mem) && !config.useSSE ->
	(MOVQstoreconst [makeValAndOff(0,8)] destptr
		(MOVQstoreconst [0] destptr mem))
(Zero [24] destptr mem) && !config.useSSE ->
	(MOVQstoreconst [makeValAndOff(0,16)] destptr
		(MOVQstoreconst [makeValAndOff(0,8)] destptr
			(MOVQstoreconst [0] destptr mem)))
(Zero [32] destptr mem) && !config.useSSE ->
	(MOVQstoreconst [makeValAndOff(0,24)] destptr
		(MOVQstoreconst [makeValAndOff(0,16)] destptr
			(MOVQstoreconst [makeValAndOff(0,8)] destptr
				(MOVQstoreconst [0] destptr mem))))

(Zero [s] destptr mem) && s > 8 && s < 16 && config.useSSE ->
	(MOVQstoreconst [makeValAndOff(0,s-8)] destptr
		(MOVQstoreconst [0] destptr mem))

// Adjust zeros to be a multiple of 16 bytes.
(Zero [s] destptr mem) && s%16 != 0 && s > 16 && s%16 > 8 && config.useSSE ->
	(Zero [s-s%16] (OffPtr <destptr.Type> destptr [s%16])
		(MOVOstore destptr (MOVOconst [0]) mem))

(Zero [s] destptr mem) && s%16 != 0 && s > 16 && s%16 <= 8 && config.useSSE ->
	(Zero [s-s%16] (OffPtr <destptr.Type> destptr [s%16])
		(MOVQstoreconst [0] destptr mem))

(Zero [16] destptr mem) && config.useSSE ->
	(MOVOstore destptr (MOVOconst [0]) mem)
(Zero [32] destptr mem) && config.useSSE ->
	(MOVOstore (OffPtr <destptr.Type> destptr [16]) (MOVOconst [0])
		(MOVOstore destptr (MOVOconst [0]) mem))
(Zero [48] destptr mem) && config.useSSE ->
	(MOVOstore (OffPtr <destptr.Type> destptr [32]) (MOVOconst [0])
		(MOVOstore (OffPtr <destptr.Type> destptr [16]) (MOVOconst [0])
			(MOVOstore destptr (MOVOconst [0]) mem)))
(Zero [64] destptr mem) && config.useSSE ->
	(MOVOstore (OffPtr <destptr.Type> destptr [48]) (MOVOconst [0])
		(MOVOstore (OffPtr <destptr.Type> destptr [32]) (MOVOconst [0])
			(MOVOstore (OffPtr <destptr.Type> destptr [16]) (MOVOconst [0])
				(MOVOstore destptr (MOVOconst [0]) mem))))

// Medium zeroing uses a duff device.
(Zero [s] destptr mem)
	&& s > 64 && s <= 1024 && s%16 == 0 && !config.noDuffDevice ->
	(DUFFZERO [s] destptr (MOVOconst [0]) mem)

// Large zeroing uses REP STOSQ.
(Zero [s] destptr mem)
	&& (s > 1024 || (config.noDuffDevice && s > 64 || !config.useSSE && s > 32))
	&& s%8 == 0 ->
	(REPSTOSQ destptr (MOVQconst [s/8]) (MOVQconst [0]) mem)

// Lowering constants
(Const8   [val]) -> (MOVLconst [val])
(Const16  [val]) -> (MOVLconst [val])
(Const32  [val]) -> (MOVLconst [val])
(Const64  [val]) -> (MOVQconst [val])
(Const32F [val]) -> (MOVSSconst [val])
(Const64F [val]) -> (MOVSDconst [val])
(ConstNil) && config.PtrSize == 8 -> (MOVQconst [0])
(ConstNil) && config.PtrSize == 4 -> (MOVLconst [0])
(ConstBool [b]) -> (MOVLconst [b])

// Lowering calls
(StaticCall [argwid] {target} mem) -> (CALLstatic [argwid] {target} mem)
(ClosureCall [argwid] entry closure mem) -> (CALLclosure [argwid] entry closure mem)
(InterCall [argwid] entry mem) -> (CALLinter [argwid] entry mem)

// Miscellaneous
(Convert <t> x mem) && config.PtrSize == 8 -> (MOVQconvert <t> x mem)
(Convert <t> x mem) && config.PtrSize == 4 -> (MOVLconvert <t> x mem)
(IsNonNil p) && config.PtrSize == 8 -> (SETNE (TESTQ p p))
(IsNonNil p) && config.PtrSize == 4 -> (SETNE (TESTL p p))
(IsInBounds idx len) && config.PtrSize == 8 -> (SETB (CMPQ idx len))
(IsInBounds idx len) && config.PtrSize == 4 -> (SETB (CMPL idx len))
(IsSliceInBounds idx len) && config.PtrSize == 8 -> (SETBE (CMPQ idx len))
(IsSliceInBounds idx len) && config.PtrSize == 4 -> (SETBE (CMPL idx len))
(NilCheck ptr mem) -> (LoweredNilCheck ptr mem)
(GetG mem) -> (LoweredGetG mem)
(GetClosurePtr) -> (LoweredGetClosurePtr)
(Addr {sym} base) && config.PtrSize == 8 -> (LEAQ {sym} base)
(Addr {sym} base) && config.PtrSize == 4 -> (LEAL {sym} base)

// block rewrites
(If (SETL  cmp) yes no) -> (LT  cmp yes no)
(If (SETLE cmp) yes no) -> (LE  cmp yes no)
(If (SETG  cmp) yes no) -> (GT  cmp yes no)
(If (SETGE cmp) yes no) -> (GE  cmp yes no)
(If (SETEQ cmp) yes no) -> (EQ  cmp yes no)
(If (SETNE cmp) yes no) -> (NE  cmp yes no)
(If (SETB  cmp) yes no) -> (ULT cmp yes no)
(If (SETBE cmp) yes no) -> (ULE cmp yes no)
(If (SETA  cmp) yes no) -> (UGT cmp yes no)
(If (SETAE cmp) yes no) -> (UGE cmp yes no)

// Special case for floating point - LF/LEF not generated
(If (SETGF  cmp) yes no) -> (UGT  cmp yes no)
(If (SETGEF cmp) yes no) -> (UGE  cmp yes no)
(If (SETEQF cmp) yes no) -> (EQF  cmp yes no)
(If (SETNEF cmp) yes no) -> (NEF  cmp yes no)

(If cond yes no) -> (NE (TESTB cond cond) yes no)

// Atomic loads.  Other than preserving their ordering with respect to other loads, nothing special here.
(AtomicLoad32 ptr mem) -> (MOVLatomicload ptr mem)
(AtomicLoad64 ptr mem) -> (MOVQatomicload ptr mem)
(AtomicLoadPtr ptr mem) && config.PtrSize == 8 -> (MOVQatomicload ptr mem)
(AtomicLoadPtr ptr mem) && config.PtrSize == 4 -> (MOVLatomicload ptr mem)

// Atomic stores.  We use XCHG to prevent the hardware reordering a subsequent load.
// TODO: most runtime uses of atomic stores don't need that property.  Use normal stores for those?
(AtomicStore32 ptr val mem) -> (Select1 (XCHGL <types.NewTuple(typ.UInt32,types.TypeMem)> val ptr mem))
(AtomicStore64 ptr val mem) -> (Select1 (XCHGQ <types.NewTuple(typ.UInt64,types.TypeMem)> val ptr mem))
(AtomicStorePtrNoWB ptr val mem) && config.PtrSize == 8 -> (Select1 (XCHGQ <types.NewTuple(typ.BytePtr,types.TypeMem)> val ptr mem))
(AtomicStorePtrNoWB ptr val mem) && config.PtrSize == 4 -> (Select1 (XCHGL <types.NewTuple(typ.BytePtr,types.TypeMem)> val ptr mem))

// Atomic exchanges.
(AtomicExchange32 ptr val mem) -> (XCHGL val ptr mem)
(AtomicExchange64 ptr val mem) -> (XCHGQ val ptr mem)

// Atomic adds.
(AtomicAdd32 ptr val mem) -> (AddTupleFirst32 val (XADDLlock val ptr mem))
(AtomicAdd64 ptr val mem) -> (AddTupleFirst64 val (XADDQlock val ptr mem))
(Select0 <t> (AddTupleFirst32 val tuple)) -> (ADDL val (Select0 <t> tuple))
(Select1     (AddTupleFirst32   _ tuple)) -> (Select1 tuple)
(Select0 <t> (AddTupleFirst64 val tuple)) -> (ADDQ val (Select0 <t> tuple))
(Select1     (AddTupleFirst64   _ tuple)) -> (Select1 tuple)

// Atomic compare and swap.
(AtomicCompareAndSwap32 ptr old new_ mem) -> (CMPXCHGLlock ptr old new_ mem)
(AtomicCompareAndSwap64 ptr old new_ mem) -> (CMPXCHGQlock ptr old new_ mem)

// Atomic memory updates.
(AtomicAnd8 ptr val mem) -> (ANDBlock ptr val mem)
(AtomicOr8 ptr val mem) -> (ORBlock ptr val mem)

// ***************************
// Above: lowering rules
// Below: optimizations
// ***************************
// TODO: Should the optimizations be a separate pass?

// Fold boolean tests into blocks
(NE (TESTB (SETL  cmp) (SETL  cmp)) yes no) -> (LT  cmp yes no)
(NE (TESTB (SETLE cmp) (SETLE cmp)) yes no) -> (LE  cmp yes no)
(NE (TESTB (SETG  cmp) (SETG  cmp)) yes no) -> (GT  cmp yes no)
(NE (TESTB (SETGE cmp) (SETGE cmp)) yes no) -> (GE  cmp yes no)
(NE (TESTB (SETEQ cmp) (SETEQ cmp)) yes no) -> (EQ  cmp yes no)
(NE (TESTB (SETNE cmp) (SETNE cmp)) yes no) -> (NE  cmp yes no)
(NE (TESTB (SETB  cmp) (SETB  cmp)) yes no) -> (ULT cmp yes no)
(NE (TESTB (SETBE cmp) (SETBE cmp)) yes no) -> (ULE cmp yes no)
(NE (TESTB (SETA  cmp) (SETA  cmp)) yes no) -> (UGT cmp yes no)
(NE (TESTB (SETAE cmp) (SETAE cmp)) yes no) -> (UGE cmp yes no)

// Recognize bit tests: a&(1<<b) != 0 for b suitably bounded
// Note that ULT and SETB check the carry flag; they are identical to CS and SETCS.
// Same, mutatis mutandis, for UGE and SETAE, and CC and SETCC.
(NE (TESTL (SHLL (MOVLconst [1]) x) y)) && !config.nacl -> (ULT (BTL x y))
(EQ (TESTL (SHLL (MOVLconst [1]) x) y)) && !config.nacl -> (UGE (BTL x y))
(NE (TESTQ (SHLQ (MOVQconst [1]) x) y)) && !config.nacl -> (ULT (BTQ x y))
(EQ (TESTQ (SHLQ (MOVQconst [1]) x) y)) && !config.nacl -> (UGE (BTQ x y))
(NE (TESTLconst [c] x)) && isPowerOfTwo(c) && log2(c) < 32 && !config.nacl -> (ULT (BTLconst [log2(c)] x))
(EQ (TESTLconst [c] x)) && isPowerOfTwo(c) && log2(c) < 32 && !config.nacl -> (UGE (BTLconst [log2(c)] x))
(NE (TESTQconst [c] x)) && isPowerOfTwo(c) && log2(c) < 64 && !config.nacl -> (ULT (BTQconst [log2(c)] x))
(EQ (TESTQconst [c] x)) && isPowerOfTwo(c) && log2(c) < 64 && !config.nacl -> (UGE (BTQconst [log2(c)] x))
(NE (TESTQ (MOVQconst [c]) x)) && isPowerOfTwo(c) && log2(c) < 64 && !config.nacl -> (ULT (BTQconst [log2(c)] x))
(EQ (TESTQ (MOVQconst [c]) x)) && isPowerOfTwo(c) && log2(c) < 64 && !config.nacl -> (UGE (BTQconst [log2(c)] x))
(SETNE (TESTL (SHLL (MOVLconst [1]) x) y)) && !config.nacl -> (SETB  (BTL x y))
(SETEQ (TESTL (SHLL (MOVLconst [1]) x) y)) && !config.nacl -> (SETAE (BTL x y))
(SETNE (TESTQ (SHLQ (MOVQconst [1]) x) y)) && !config.nacl -> (SETB  (BTQ x y))
(SETEQ (TESTQ (SHLQ (MOVQconst [1]) x) y)) && !config.nacl -> (SETAE (BTQ x y))
(SETNE (TESTLconst [c] x)) && isPowerOfTwo(c) && log2(c) < 32 && !config.nacl -> (SETB  (BTLconst [log2(c)] x))
(SETEQ (TESTLconst [c] x)) && isPowerOfTwo(c) && log2(c) < 32 && !config.nacl -> (SETAE (BTLconst [log2(c)] x))
(SETNE (TESTQconst [c] x)) && isPowerOfTwo(c) && log2(c) < 64 && !config.nacl -> (SETB  (BTQconst [log2(c)] x))
(SETEQ (TESTQconst [c] x)) && isPowerOfTwo(c) && log2(c) < 64 && !config.nacl -> (SETAE (BTQconst [log2(c)] x))
(SETNE (TESTQ (MOVQconst [c]) x)) && isPowerOfTwo(c) && log2(c) < 64 && !config.nacl -> (SETB  (BTQconst [log2(c)] x))
(SETEQ (TESTQ (MOVQconst [c]) x)) && isPowerOfTwo(c) && log2(c) < 64 && !config.nacl -> (SETAE (BTQconst [log2(c)] x))

// Fold boolean negation into SETcc.
(XORLconst [1] (SETNE x)) -> (SETEQ x)
(XORLconst [1] (SETEQ x)) -> (SETNE x)
(XORLconst [1] (SETL  x)) -> (SETGE x)
(XORLconst [1] (SETGE x)) -> (SETL  x)
(XORLconst [1] (SETLE x)) -> (SETG  x)
(XORLconst [1] (SETG  x)) -> (SETLE x)
(XORLconst [1] (SETB  x)) -> (SETAE x)
(XORLconst [1] (SETAE x)) -> (SETB  x)
(XORLconst [1] (SETBE x)) -> (SETA  x)
(XORLconst [1] (SETA  x)) -> (SETBE x)

// Convert BTQconst to BTLconst if possible. It has a shorter encoding.
(BTQconst [c] x) && c < 32 -> (BTLconst [c] x)

// Special case for floating point - LF/LEF not generated
(NE (TESTB (SETGF  cmp) (SETGF  cmp)) yes no) -> (UGT  cmp yes no)
(NE (TESTB (SETGEF cmp) (SETGEF cmp)) yes no) -> (UGE  cmp yes no)
(NE (TESTB (SETEQF cmp) (SETEQF cmp)) yes no) -> (EQF  cmp yes no)
(NE (TESTB (SETNEF cmp) (SETNEF cmp)) yes no) -> (NEF  cmp yes no)

// Disabled because it interferes with the pattern match above and makes worse code.
// (SETNEF x) -> (ORQ (SETNE <typ.Int8> x) (SETNAN <typ.Int8> x))
// (SETEQF x) -> (ANDQ (SETEQ <typ.Int8> x) (SETORD <typ.Int8> x))

// fold constants into instructions
(ADDQ x (MOVQconst [c])) && is32Bit(c) -> (ADDQconst [c] x)
(ADDL x (MOVLconst [c])) -> (ADDLconst [c] x)

(SUBQ x (MOVQconst [c])) && is32Bit(c) -> (SUBQconst x [c])
(SUBQ (MOVQconst [c]) x) && is32Bit(c) -> (NEGQ (SUBQconst <v.Type> x [c]))
(SUBL x (MOVLconst [c])) -> (SUBLconst x [c])
(SUBL (MOVLconst [c]) x) -> (NEGL (SUBLconst <v.Type> x [c]))

(MULQ x (MOVQconst [c])) && is32Bit(c) -> (MULQconst [c] x)
(MULL x (MOVLconst [c])) -> (MULLconst [c] x)

(ANDQ x (MOVQconst [c])) && is32Bit(c) -> (ANDQconst [c] x)
(ANDL x (MOVLconst [c])) -> (ANDLconst [c] x)

(ANDLconst [c] (ANDLconst [d] x)) -> (ANDLconst [c & d] x)
(ANDQconst [c] (ANDQconst [d] x)) -> (ANDQconst [c & d] x)

(XORLconst [c] (XORLconst [d] x)) -> (XORLconst [c ^ d] x)
(XORQconst [c] (XORQconst [d] x)) -> (XORQconst [c ^ d] x)

(MULLconst [c] (MULLconst [d] x)) -> (MULLconst [int64(int32(c * d))] x)
(MULQconst [c] (MULQconst [d] x)) && is32Bit(c*d) -> (MULQconst [c * d] x)

(ORQ x (MOVQconst [c])) && is32Bit(c) -> (ORQconst [c] x)
(ORL x (MOVLconst [c])) -> (ORLconst [c] x)

(XORQ x (MOVQconst [c])) && is32Bit(c) -> (XORQconst [c] x)
(XORL x (MOVLconst [c])) -> (XORLconst [c] x)

(SHLQ x (MOVQconst [c])) -> (SHLQconst [c&63] x)
(SHLQ x (MOVLconst [c])) -> (SHLQconst [c&63] x)

(SHLL x (MOVQconst [c])) -> (SHLLconst [c&31] x)
(SHLL x (MOVLconst [c])) -> (SHLLconst [c&31] x)

(SHRQ x (MOVQconst [c])) -> (SHRQconst [c&63] x)
(SHRQ x (MOVLconst [c])) -> (SHRQconst [c&63] x)

(SHRL x (MOVQconst [c])) -> (SHRLconst [c&31] x)
(SHRL x (MOVLconst [c])) -> (SHRLconst [c&31] x)

(SHRW x (MOVQconst [c])) && c&31 < 16 -> (SHRWconst [c&31] x)
(SHRW x (MOVLconst [c])) && c&31 < 16 -> (SHRWconst [c&31] x)
(SHRW _ (MOVQconst [c])) && c&31 >= 16 -> (MOVLconst [0])
(SHRW _ (MOVLconst [c])) && c&31 >= 16 -> (MOVLconst [0])

(SHRB x (MOVQconst [c])) && c&31 < 8 -> (SHRBconst [c&31] x)
(SHRB x (MOVLconst [c])) && c&31 < 8 -> (SHRBconst [c&31] x)
(SHRB _ (MOVQconst [c])) && c&31 >= 8 -> (MOVLconst [0])
(SHRB _ (MOVLconst [c])) && c&31 >= 8 -> (MOVLconst [0])

(SARQ x (MOVQconst [c])) -> (SARQconst [c&63] x)
(SARQ x (MOVLconst [c])) -> (SARQconst [c&63] x)

(SARL x (MOVQconst [c])) -> (SARLconst [c&31] x)
(SARL x (MOVLconst [c])) -> (SARLconst [c&31] x)

(SARW x (MOVQconst [c])) -> (SARWconst [min(c&31,15)] x)
(SARW x (MOVLconst [c])) -> (SARWconst [min(c&31,15)] x)

(SARB x (MOVQconst [c])) -> (SARBconst [min(c&31,7)] x)
(SARB x (MOVLconst [c])) -> (SARBconst [min(c&31,7)] x)

// Operations which don't affect the low 6/5 bits of the shift amount are NOPs.
(SHLQ x (ADDQconst [c] y)) && c & 63 == 0 -> (SHLQ x y)
(SHRQ x (ADDQconst [c] y)) && c & 63 == 0 -> (SHRQ x y)
(SARQ x (ADDQconst [c] y)) && c & 63 == 0 -> (SARQ x y)
(SHLQ x (NEGQ <t> (ADDQconst [c] y))) && c & 63 == 0 -> (SHLQ x (NEGQ <t> y))
(SHRQ x (NEGQ <t> (ADDQconst [c] y))) && c & 63 == 0 -> (SHRQ x (NEGQ <t> y))
(SARQ x (NEGQ <t> (ADDQconst [c] y))) && c & 63 == 0 -> (SARQ x (NEGQ <t> y))
(SHLQ x (ANDQconst [c] y)) && c & 63 == 63 -> (SHLQ x y)
(SHRQ x (ANDQconst [c] y)) && c & 63 == 63 -> (SHRQ x y)
(SARQ x (ANDQconst [c] y)) && c & 63 == 63 -> (SARQ x y)
(SHLQ x (NEGQ <t> (ANDQconst [c] y))) && c & 63 == 63 -> (SHLQ x (NEGQ <t> y))
(SHRQ x (NEGQ <t> (ANDQconst [c] y))) && c & 63 == 63 -> (SHRQ x (NEGQ <t> y))
(SARQ x (NEGQ <t> (ANDQconst [c] y))) && c & 63 == 63 -> (SARQ x (NEGQ <t> y))

(SHLL x (ADDQconst [c] y)) && c & 31 == 0 -> (SHLL x y)
(SHRL x (ADDQconst [c] y)) && c & 31 == 0 -> (SHRL x y)
(SARL x (ADDQconst [c] y)) && c & 31 == 0 -> (SARL x y)
(SHLL x (NEGQ <t> (ADDQconst [c] y))) && c & 31 == 0 -> (SHLL x (NEGQ <t> y))
(SHRL x (NEGQ <t> (ADDQconst [c] y))) && c & 31 == 0 -> (SHRL x (NEGQ <t> y))
(SARL x (NEGQ <t> (ADDQconst [c] y))) && c & 31 == 0 -> (SARL x (NEGQ <t> y))
(SHLL x (ANDQconst [c] y)) && c & 31 == 31 -> (SHLL x y)
(SHRL x (ANDQconst [c] y)) && c & 31 == 31 -> (SHRL x y)
(SARL x (ANDQconst [c] y)) && c & 31 == 31 -> (SARL x y)
(SHLL x (NEGQ <t> (ANDQconst [c] y))) && c & 31 == 31 -> (SHLL x (NEGQ <t> y))
(SHRL x (NEGQ <t> (ANDQconst [c] y))) && c & 31 == 31 -> (SHRL x (NEGQ <t> y))
(SARL x (NEGQ <t> (ANDQconst [c] y))) && c & 31 == 31 -> (SARL x (NEGQ <t> y))

(SHLQ x (ADDLconst [c] y)) && c & 63 == 0 -> (SHLQ x y)
(SHRQ x (ADDLconst [c] y)) && c & 63 == 0 -> (SHRQ x y)
(SARQ x (ADDLconst [c] y)) && c & 63 == 0 -> (SARQ x y)
(SHLQ x (NEGL <t> (ADDLconst [c] y))) && c & 63 == 0 -> (SHLQ x (NEGL <t> y))
(SHRQ x (NEGL <t> (ADDLconst [c] y))) && c & 63 == 0 -> (SHRQ x (NEGL <t> y))
(SARQ x (NEGL <t> (ADDLconst [c] y))) && c & 63 == 0 -> (SARQ x (NEGL <t> y))
(SHLQ x (ANDLconst [c] y)) && c & 63 == 63 -> (SHLQ x y)
(SHRQ x (ANDLconst [c] y)) && c & 63 == 63 -> (SHRQ x y)
(SARQ x (ANDLconst [c] y)) && c & 63 == 63 -> (SARQ x y)
(SHLQ x (NEGL <t> (ANDLconst [c] y))) && c & 63 == 63 -> (SHLQ x (NEGL <t> y))
(SHRQ x (NEGL <t> (ANDLconst [c] y))) && c & 63 == 63 -> (SHRQ x (NEGL <t> y))
(SARQ x (NEGL <t> (ANDLconst [c] y))) && c & 63 == 63 -> (SARQ x (NEGL <t> y))

(SHLL x (ADDLconst [c] y)) && c & 31 == 0 -> (SHLL x y)
(SHRL x (ADDLconst [c] y)) && c & 31 == 0 -> (SHRL x y)
(SARL x (ADDLconst [c] y)) && c & 31 == 0 -> (SARL x y)
(SHLL x (NEGL <t> (ADDLconst [c] y))) && c & 31 == 0 -> (SHLL x (NEGL <t> y))
(SHRL x (NEGL <t> (ADDLconst [c] y))) && c & 31 == 0 -> (SHRL x (NEGL <t> y))
(SARL x (NEGL <t> (ADDLconst [c] y))) && c & 31 == 0 -> (SARL x (NEGL <t> y))
(SHLL x (ANDLconst [c] y)) && c & 31 == 31 -> (SHLL x y)
(SHRL x (ANDLconst [c] y)) && c & 31 == 31 -> (SHRL x y)
(SARL x (ANDLconst [c] y)) && c & 31 == 31 -> (SARL x y)
(SHLL x (NEGL <t> (ANDLconst [c] y))) && c & 31 == 31 -> (SHLL x (NEGL <t> y))
(SHRL x (NEGL <t> (ANDLconst [c] y))) && c & 31 == 31 -> (SHRL x (NEGL <t> y))
(SARL x (NEGL <t> (ANDLconst [c] y))) && c & 31 == 31 -> (SARL x (NEGL <t> y))

// Constant rotate instructions
(ADDQ (SHLQconst x [c]) (SHRQconst x [d])) && d==64-c -> (ROLQconst x [c])
( ORQ (SHLQconst x [c]) (SHRQconst x [d])) && d==64-c -> (ROLQconst x [c])
(XORQ (SHLQconst x [c]) (SHRQconst x [d])) && d==64-c -> (ROLQconst x [c])

(ADDL (SHLLconst x [c]) (SHRLconst x [d])) && d==32-c -> (ROLLconst x [c])
( ORL (SHLLconst x [c]) (SHRLconst x [d])) && d==32-c -> (ROLLconst x [c])
(XORL (SHLLconst x [c]) (SHRLconst x [d])) && d==32-c -> (ROLLconst x [c])

(ADDL <t> (SHLLconst x [c]) (SHRWconst x [d])) && d==16-c && c < 16 && t.Size() == 2 -> (ROLWconst x [c])
( ORL <t> (SHLLconst x [c]) (SHRWconst x [d])) && d==16-c && c < 16 && t.Size() == 2 -> (ROLWconst x [c])
(XORL <t> (SHLLconst x [c]) (SHRWconst x [d])) && d==16-c && c < 16 && t.Size() == 2 -> (ROLWconst x [c])

(ADDL <t> (SHLLconst x [c]) (SHRBconst x [d])) && d==8-c  && c < 8 && t.Size() == 1 -> (ROLBconst x [c])
( ORL <t> (SHLLconst x [c]) (SHRBconst x [d])) && d==8-c  && c < 8 && t.Size() == 1 -> (ROLBconst x [c])
(XORL <t> (SHLLconst x [c]) (SHRBconst x [d])) && d==8-c  && c < 8 && t.Size() == 1 -> (ROLBconst x [c])

(ROLQconst [c] (ROLQconst [d] x)) -> (ROLQconst [(c+d)&63] x)
(ROLLconst [c] (ROLLconst [d] x)) -> (ROLLconst [(c+d)&31] x)
(ROLWconst [c] (ROLWconst [d] x)) -> (ROLWconst [(c+d)&15] x)
(ROLBconst [c] (ROLBconst [d] x)) -> (ROLBconst [(c+d)& 7] x)

// Non-constant rotates.
// We want to issue a rotate when the Go source contains code like
//     y &= 63
//     x << y | x >> (64-y)
// The shift rules above convert << to SHLx and >> to SHRx.
// SHRx converts its shift argument from 64-y to -y.
// A tricky situation occurs when y==0. Then the original code would be:
//     x << 0 | x >> 64
// But x >> 64 is 0, not x. So there's an additional mask that is ANDed in
// to force the second term to 0. We don't need that mask, but we must match
// it in order to strip it out.
(ORQ (SHLQ x y) (ANDQ (SHRQ x (NEGQ y)) (SBBQcarrymask (CMPQconst (NEGQ (ADDQconst (ANDQconst y [63]) [-64])) [64])))) -> (ROLQ x y)
(ORQ (SHLQ x y) (ANDQ (SHRQ x (NEGL y)) (SBBQcarrymask (CMPLconst (NEGL (ADDLconst (ANDLconst y [63]) [-64])) [64])))) -> (ROLQ x y)
(ORQ (SHRQ x y) (ANDQ (SHLQ x (NEGQ y)) (SBBQcarrymask (CMPQconst (NEGQ (ADDQconst (ANDQconst y [63]) [-64])) [64])))) -> (RORQ x y)
(ORQ (SHRQ x y) (ANDQ (SHLQ x (NEGL y)) (SBBQcarrymask (CMPLconst (NEGL (ADDLconst (ANDLconst y [63]) [-64])) [64])))) -> (RORQ x y)

(ORL (SHLL x y) (ANDL (SHRL x (NEGQ y)) (SBBLcarrymask (CMPQconst (NEGQ (ADDQconst (ANDQconst y [31]) [-32])) [32])))) -> (ROLL x y)
(ORL (SHLL x y) (ANDL (SHRL x (NEGL y)) (SBBLcarrymask (CMPLconst (NEGL (ADDLconst (ANDLconst y [31]) [-32])) [32])))) -> (ROLL x y)
(ORL (SHRL x y) (ANDL (SHLL x (NEGQ y)) (SBBLcarrymask (CMPQconst (NEGQ (ADDQconst (ANDQconst y [31]) [-32])) [32])))) -> (RORL x y)
(ORL (SHRL x y) (ANDL (SHLL x (NEGL y)) (SBBLcarrymask (CMPLconst (NEGL (ADDLconst (ANDLconst y [31]) [-32])) [32])))) -> (RORL x y)

// Help with rotate detection
(CMPQconst (NEGQ (ADDQconst [-16] (ANDQconst [15] _))) [32]) -> (FlagLT_ULT)
(CMPQconst (NEGQ (ADDQconst [ -8] (ANDQconst  [7] _))) [32]) -> (FlagLT_ULT)

(ORL (SHLL x (ANDQconst y [15]))
     (ANDL (SHRW x (NEGQ (ADDQconst (ANDQconst y [15]) [-16])))
           (SBBLcarrymask (CMPQconst (NEGQ (ADDQconst (ANDQconst y [15]) [-16])) [16]))))
  && v.Type.Size() == 2
  -> (ROLW x y)
(ORL (SHLL x (ANDLconst y [15]))
     (ANDL (SHRW x (NEGL (ADDLconst (ANDLconst y [15]) [-16])))
           (SBBLcarrymask (CMPLconst (NEGL (ADDLconst (ANDLconst y [15]) [-16])) [16]))))
  && v.Type.Size() == 2
  -> (ROLW x y)
(ORL (SHRW x (ANDQconst y [15]))
     (SHLL x (NEGQ (ADDQconst (ANDQconst y [15]) [-16]))))
  && v.Type.Size() == 2
  -> (RORW x y)
(ORL (SHRW x (ANDLconst y [15]))
     (SHLL x (NEGL (ADDLconst (ANDLconst y [15]) [-16]))))
  && v.Type.Size() == 2
  -> (RORW x y)

(ORL (SHLL x (ANDQconst y [ 7]))
     (ANDL (SHRB x (NEGQ (ADDQconst (ANDQconst y [ 7]) [ -8])))
           (SBBLcarrymask (CMPQconst (NEGQ (ADDQconst (ANDQconst y [ 7]) [ -8])) [ 8]))))
  && v.Type.Size() == 1
  -> (ROLB x y)
(ORL (SHLL x (ANDLconst y [ 7]))
     (ANDL (SHRB x (NEGL (ADDLconst (ANDLconst y [ 7]) [ -8])))
           (SBBLcarrymask (CMPLconst (NEGL (ADDLconst (ANDLconst y [ 7]) [ -8])) [ 8]))))
  && v.Type.Size() == 1
  -> (ROLB x y)
(ORL (SHRB x (ANDQconst y [ 7]))
     (SHLL x (NEGQ (ADDQconst (ANDQconst y [ 7]) [ -8]))))
  && v.Type.Size() == 1
  -> (RORB x y)
(ORL (SHRB x (ANDLconst y [ 7]))
     (SHLL x (NEGL (ADDLconst (ANDLconst y [ 7]) [ -8]))))
  && v.Type.Size() == 1
  -> (RORB x y)

// rotate left negative = rotate right
(ROLQ x (NEGQ y)) -> (RORQ x y)
(ROLQ x (NEGL y)) -> (RORQ x y)
(ROLL x (NEGQ y)) -> (RORL x y)
(ROLL x (NEGL y)) -> (RORL x y)
(ROLW x (NEGQ y)) -> (RORW x y)
(ROLW x (NEGL y)) -> (RORW x y)
(ROLB x (NEGQ y)) -> (RORB x y)
(ROLB x (NEGL y)) -> (RORB x y)

// rotate right negative = rotate left
(RORQ x (NEGQ y)) -> (ROLQ x y)
(RORQ x (NEGL y)) -> (ROLQ x y)
(RORL x (NEGQ y)) -> (ROLL x y)
(RORL x (NEGL y)) -> (ROLL x y)
(RORW x (NEGQ y)) -> (ROLW x y)
(RORW x (NEGL y)) -> (ROLW x y)
(RORB x (NEGQ y)) -> (ROLB x y)
(RORB x (NEGL y)) -> (ROLB x y)

// rotate by constants
(ROLQ x (MOVQconst [c])) -> (ROLQconst [c&63] x)
(ROLQ x (MOVLconst [c])) -> (ROLQconst [c&63] x)
(ROLL x (MOVQconst [c])) -> (ROLLconst [c&31] x)
(ROLL x (MOVLconst [c])) -> (ROLLconst [c&31] x)
(ROLW x (MOVQconst [c])) -> (ROLWconst [c&15] x)
(ROLW x (MOVLconst [c])) -> (ROLWconst [c&15] x)
(ROLB x (MOVQconst [c])) -> (ROLBconst [c&7 ] x)
(ROLB x (MOVLconst [c])) -> (ROLBconst [c&7 ] x)

(RORQ x (MOVQconst [c])) -> (ROLQconst [(-c)&63] x)
(RORQ x (MOVLconst [c])) -> (ROLQconst [(-c)&63] x)
(RORL x (MOVQconst [c])) -> (ROLLconst [(-c)&31] x)
(RORL x (MOVLconst [c])) -> (ROLLconst [(-c)&31] x)
(RORW x (MOVQconst [c])) -> (ROLWconst [(-c)&15] x)
(RORW x (MOVLconst [c])) -> (ROLWconst [(-c)&15] x)
(RORB x (MOVQconst [c])) -> (ROLBconst [(-c)&7 ] x)
(RORB x (MOVLconst [c])) -> (ROLBconst [(-c)&7 ] x)

// Constant shift simplifications
(SHLQconst x [0]) -> x
(SHRQconst x [0]) -> x
(SARQconst x [0]) -> x

(SHLLconst x [0]) -> x
(SHRLconst x [0]) -> x
(SARLconst x [0]) -> x

(SHRWconst x [0]) -> x
(SARWconst x [0]) -> x

(SHRBconst x [0]) -> x
(SARBconst x [0]) -> x

(ROLQconst x [0]) -> x
(ROLLconst x [0]) -> x
(ROLWconst x [0]) -> x
(ROLBconst x [0]) -> x

// Note: the word and byte shifts keep the low 5 bits (not the low 4 or 3 bits)
// because the x86 instructions are defined to use all 5 bits of the shift even
// for the small shifts. I don't think we'll ever generate a weird shift (e.g.
// (SHRW x (MOVLconst [24])), but just in case.

(CMPQ x (MOVQconst [c])) && is32Bit(c) -> (CMPQconst x [c])
(CMPQ (MOVQconst [c]) x) && is32Bit(c) -> (InvertFlags (CMPQconst x [c]))
(CMPL x (MOVLconst [c])) -> (CMPLconst x [c])
(CMPL (MOVLconst [c]) x) -> (InvertFlags (CMPLconst x [c]))
(CMPW x (MOVLconst [c])) -> (CMPWconst x [int64(int16(c))])
(CMPW (MOVLconst [c]) x) -> (InvertFlags (CMPWconst x [int64(int16(c))]))
(CMPB x (MOVLconst [c])) -> (CMPBconst x [int64(int8(c))])
(CMPB (MOVLconst [c]) x) -> (InvertFlags (CMPBconst x [int64(int8(c))]))

// Using MOVZX instead of AND is cheaper.
(ANDLconst [0xFF] x) -> (MOVBQZX x)
(ANDLconst [0xFFFF] x) -> (MOVWQZX x)
(ANDQconst [0xFF] x) -> (MOVBQZX x)
(ANDQconst [0xFFFF] x) -> (MOVWQZX x)
(ANDQconst [0xFFFFFFFF] x) -> (MOVLQZX x)

// strength reduction
// Assumes that the following costs from https://gmplib.org/~tege/x86-timing.pdf:
//    1 - addq, shlq, leaq, negq
//    3 - imulq
// This limits the rewrites to two instructions.
// TODO: 27, 81
(MULQconst [-1] x) -> (NEGQ x)
(MULQconst [0] _) -> (MOVQconst [0])
(MULQconst [1] x) -> x
(MULQconst [3] x) -> (LEAQ2 x x)
(MULQconst [5] x) -> (LEAQ4 x x)
(MULQconst [7] x) -> (LEAQ8 (NEGQ <v.Type> x) x)
(MULQconst [9] x) -> (LEAQ8 x x)
(MULQconst [11] x) -> (LEAQ2 x (LEAQ4 <v.Type> x x))
(MULQconst [13] x) -> (LEAQ4 x (LEAQ2 <v.Type> x x))
(MULQconst [21] x) -> (LEAQ4 x (LEAQ4 <v.Type> x x))
(MULQconst [25] x) -> (LEAQ8 x (LEAQ2 <v.Type> x x))
(MULQconst [37] x) -> (LEAQ4 x (LEAQ8 <v.Type> x x))
(MULQconst [41] x) -> (LEAQ8 x (LEAQ4 <v.Type> x x))
(MULQconst [73] x) -> (LEAQ8 x (LEAQ8 <v.Type> x x))

(MULQconst [c] x) && isPowerOfTwo(c) -> (SHLQconst [log2(c)] x)
(MULQconst [c] x) && isPowerOfTwo(c+1) && c >= 15 -> (SUBQ (SHLQconst <v.Type> [log2(c+1)] x) x)
(MULQconst [c] x) && isPowerOfTwo(c-1) && c >= 17 -> (LEAQ1 (SHLQconst <v.Type> [log2(c-1)] x) x)
(MULQconst [c] x) && isPowerOfTwo(c-2) && c >= 34 -> (LEAQ2 (SHLQconst <v.Type> [log2(c-2)] x) x)
(MULQconst [c] x) && isPowerOfTwo(c-4) && c >= 68 -> (LEAQ4 (SHLQconst <v.Type> [log2(c-4)] x) x)
(MULQconst [c] x) && isPowerOfTwo(c-8) && c >= 136 -> (LEAQ8 (SHLQconst <v.Type> [log2(c-8)] x) x)
(MULQconst [c] x) && c%3 == 0 && isPowerOfTwo(c/3) -> (SHLQconst [log2(c/3)] (LEAQ2 <v.Type> x x))
(MULQconst [c] x) && c%5 == 0 && isPowerOfTwo(c/5) -> (SHLQconst [log2(c/5)] (LEAQ4 <v.Type> x x))
(MULQconst [c] x) && c%9 == 0 && isPowerOfTwo(c/9) -> (SHLQconst [log2(c/9)] (LEAQ8 <v.Type> x x))

// combine add/shift into LEAQ
(ADDQ x (SHLQconst [3] y)) -> (LEAQ8 x y)
(ADDQ x (SHLQconst [2] y)) -> (LEAQ4 x y)
(ADDQ x (SHLQconst [1] y)) -> (LEAQ2 x y)
(ADDQ x (ADDQ y y)) -> (LEAQ2 x y)
(ADDQ x (ADDQ x y)) -> (LEAQ2 y x)

// combine ADDQ/ADDQconst into LEAQ1
(ADDQconst [c] (ADDQ x y)) -> (LEAQ1 [c] x y)
(ADDQ (ADDQconst [c] x) y) -> (LEAQ1 [c] x y)

// fold ADDQ into LEAQ
(ADDQconst [c] (LEAQ [d] {s} x)) && is32Bit(c+d) -> (LEAQ [c+d] {s} x)
(LEAQ [c] {s} (ADDQconst [d] x)) && is32Bit(c+d) -> (LEAQ [c+d] {s} x)
(LEAQ [c] {s} (ADDQ x y)) && x.Op != OpSB && y.Op != OpSB -> (LEAQ1 [c] {s} x y)
(ADDQ x (LEAQ [c] {s} y)) && x.Op != OpSB && y.Op != OpSB -> (LEAQ1 [c] {s} x y)

// fold ADDQconst into LEAQx
(ADDQconst [c] (LEAQ1 [d] {s} x y)) && is32Bit(c+d) -> (LEAQ1 [c+d] {s} x y)
(ADDQconst [c] (LEAQ2 [d] {s} x y)) && is32Bit(c+d) -> (LEAQ2 [c+d] {s} x y)
(ADDQconst [c] (LEAQ4 [d] {s} x y)) && is32Bit(c+d) -> (LEAQ4 [c+d] {s} x y)
(ADDQconst [c] (LEAQ8 [d] {s} x y)) && is32Bit(c+d) -> (LEAQ8 [c+d] {s} x y)
(LEAQ1 [c] {s} (ADDQconst [d] x) y) && is32Bit(c+d)   && x.Op != OpSB -> (LEAQ1 [c+d] {s} x y)
(LEAQ2 [c] {s} (ADDQconst [d] x) y) && is32Bit(c+d)   && x.Op != OpSB -> (LEAQ2 [c+d] {s} x y)
(LEAQ2 [c] {s} x (ADDQconst [d] y)) && is32Bit(c+2*d) && y.Op != OpSB -> (LEAQ2 [c+2*d] {s} x y)
(LEAQ4 [c] {s} (ADDQconst [d] x) y) && is32Bit(c+d)   && x.Op != OpSB -> (LEAQ4 [c+d] {s} x y)
(LEAQ4 [c] {s} x (ADDQconst [d] y)) && is32Bit(c+4*d) && y.Op != OpSB -> (LEAQ4 [c+4*d] {s} x y)
(LEAQ8 [c] {s} (ADDQconst [d] x) y) && is32Bit(c+d)   && x.Op != OpSB -> (LEAQ8 [c+d] {s} x y)
(LEAQ8 [c] {s} x (ADDQconst [d] y)) && is32Bit(c+8*d) && y.Op != OpSB -> (LEAQ8 [c+8*d] {s} x y)

// fold shifts into LEAQx
(LEAQ1 [c] {s} x (SHLQconst [1] y)) -> (LEAQ2 [c] {s} x y)
(LEAQ1 [c] {s} x (SHLQconst [2] y)) -> (LEAQ4 [c] {s} x y)
(LEAQ1 [c] {s} x (SHLQconst [3] y)) -> (LEAQ8 [c] {s} x y)
(LEAQ2 [c] {s} x (SHLQconst [1] y)) -> (LEAQ4 [c] {s} x y)
(LEAQ2 [c] {s} x (SHLQconst [2] y)) -> (LEAQ8 [c] {s} x y)
(LEAQ4 [c] {s} x (SHLQconst [1] y)) -> (LEAQ8 [c] {s} x y)

// reverse ordering of compare instruction
(SETL (InvertFlags x)) -> (SETG x)
(SETG (InvertFlags x)) -> (SETL x)
(SETB (InvertFlags x)) -> (SETA x)
(SETA (InvertFlags x)) -> (SETB x)
(SETLE (InvertFlags x)) -> (SETGE x)
(SETGE (InvertFlags x)) -> (SETLE x)
(SETBE (InvertFlags x)) -> (SETAE x)
(SETAE (InvertFlags x)) -> (SETBE x)
(SETEQ (InvertFlags x)) -> (SETEQ x)
(SETNE (InvertFlags x)) -> (SETNE x)

// sign extended loads
// Note: The combined instruction must end up in the same block
// as the original load. If not, we end up making a value with
// memory type live in two different blocks, which can lead to
// multiple memory values alive simultaneously.
// Make sure we don't combine these ops if the load has another use.
// This prevents a single load from being split into multiple loads
// which then might return different values.  See test/atomicload.go.
(MOVBQSX x:(MOVBload [off] {sym} ptr mem)) && x.Uses == 1 && clobber(x) -> @x.Block (MOVBQSXload <v.Type> [off] {sym} ptr mem)
(MOVBQSX x:(MOVWload [off] {sym} ptr mem)) && x.Uses == 1 && clobber(x) -> @x.Block (MOVBQSXload <v.Type> [off] {sym} ptr mem)
(MOVBQSX x:(MOVLload [off] {sym} ptr mem)) && x.Uses == 1 && clobber(x) -> @x.Block (MOVBQSXload <v.Type> [off] {sym} ptr mem)
(MOVBQSX x:(MOVQload [off] {sym} ptr mem)) && x.Uses == 1 && clobber(x) -> @x.Block (MOVBQSXload <v.Type> [off] {sym} ptr mem)
(MOVBQZX x:(MOVBload [off] {sym} ptr mem)) && x.Uses == 1 && clobber(x) -> @x.Block (MOVBload <v.Type> [off] {sym} ptr mem)
(MOVBQZX x:(MOVWload [off] {sym} ptr mem)) && x.Uses == 1 && clobber(x) -> @x.Block (MOVBload <v.Type> [off] {sym} ptr mem)
(MOVBQZX x:(MOVLload [off] {sym} ptr mem)) && x.Uses == 1 && clobber(x) -> @x.Block (MOVBload <v.Type> [off] {sym} ptr mem)
(MOVBQZX x:(MOVQload [off] {sym} ptr mem)) && x.Uses == 1 && clobber(x) -> @x.Block (MOVBload <v.Type> [off] {sym} ptr mem)
(MOVWQSX x:(MOVWload [off] {sym} ptr mem)) && x.Uses == 1 && clobber(x) -> @x.Block (MOVWQSXload <v.Type> [off] {sym} ptr mem)
(MOVWQSX x:(MOVLload [off] {sym} ptr mem)) && x.Uses == 1 && clobber(x) -> @x.Block (MOVWQSXload <v.Type> [off] {sym} ptr mem)
(MOVWQSX x:(MOVQload [off] {sym} ptr mem)) && x.Uses == 1 && clobber(x) -> @x.Block (MOVWQSXload <v.Type> [off] {sym} ptr mem)
(MOVWQZX x:(MOVWload [off] {sym} ptr mem)) && x.Uses == 1 && clobber(x) -> @x.Block (MOVWload <v.Type> [off] {sym} ptr mem)
(MOVWQZX x:(MOVLload [off] {sym} ptr mem)) && x.Uses == 1 && clobber(x) -> @x.Block (MOVWload <v.Type> [off] {sym} ptr mem)
(MOVWQZX x:(MOVQload [off] {sym} ptr mem)) && x.Uses == 1 && clobber(x) -> @x.Block (MOVWload <v.Type> [off] {sym} ptr mem)
(MOVLQSX x:(MOVLload [off] {sym} ptr mem)) && x.Uses == 1 && clobber(x) -> @x.Block (MOVLQSXload <v.Type> [off] {sym} ptr mem)
(MOVLQSX x:(MOVQload [off] {sym} ptr mem)) && x.Uses == 1 && clobber(x) -> @x.Block (MOVLQSXload <v.Type> [off] {sym} ptr mem)
(MOVLQZX x:(MOVLload [off] {sym} ptr mem)) && x.Uses == 1 && clobber(x) -> @x.Block (MOVLload <v.Type> [off] {sym} ptr mem)
(MOVLQZX x:(MOVQload [off] {sym} ptr mem)) && x.Uses == 1 && clobber(x) -> @x.Block (MOVLload <v.Type> [off] {sym} ptr mem)

(MOVLQZX x) && zeroUpper32Bits(x,3) -> x

(MOVBQZX x:(MOVBloadidx1 [off] {sym} ptr idx mem)) && x.Uses == 1 && clobber(x) -> @x.Block (MOVBloadidx1 <v.Type> [off] {sym} ptr idx mem)
(MOVWQZX x:(MOVWloadidx1 [off] {sym} ptr idx mem)) && x.Uses == 1 && clobber(x) -> @x.Block (MOVWloadidx1 <v.Type> [off] {sym} ptr idx mem)
(MOVWQZX x:(MOVWloadidx2 [off] {sym} ptr idx mem)) && x.Uses == 1 && clobber(x) -> @x.Block (MOVWloadidx2 <v.Type> [off] {sym} ptr idx mem)
(MOVLQZX x:(MOVLloadidx1 [off] {sym} ptr idx mem)) && x.Uses == 1 && clobber(x) -> @x.Block (MOVLloadidx1 <v.Type> [off] {sym} ptr idx mem)
(MOVLQZX x:(MOVLloadidx4 [off] {sym} ptr idx mem)) && x.Uses == 1 && clobber(x) -> @x.Block (MOVLloadidx4 <v.Type> [off] {sym} ptr idx mem)

// replace load from same location as preceding store with zero/sign extension (or copy in case of full width)
(MOVBload [off] {sym} ptr (MOVBstore [off2] {sym2} ptr2 x _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) -> (MOVBQZX x)
(MOVWload [off] {sym} ptr (MOVWstore [off2] {sym2} ptr2 x _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) -> (MOVWQZX x)
(MOVLload [off] {sym} ptr (MOVLstore [off2] {sym2} ptr2 x _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) -> (MOVLQZX x)
(MOVQload [off] {sym} ptr (MOVQstore [off2] {sym2} ptr2 x _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) -> x
(MOVBQSXload [off] {sym} ptr (MOVBstore [off2] {sym2} ptr2 x _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) -> (MOVBQSX x)
(MOVWQSXload [off] {sym} ptr (MOVWstore [off2] {sym2} ptr2 x _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) -> (MOVWQSX x)
(MOVLQSXload [off] {sym} ptr (MOVLstore [off2] {sym2} ptr2 x _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) -> (MOVLQSX x)

// Fold extensions and ANDs together.
(MOVBQZX (ANDLconst [c] x)) -> (ANDLconst [c & 0xff] x)
(MOVWQZX (ANDLconst [c] x)) -> (ANDLconst [c & 0xffff] x)
(MOVLQZX (ANDLconst [c] x)) -> (ANDLconst [c] x)
(MOVBQSX (ANDLconst [c] x)) && c & 0x80 == 0 -> (ANDLconst [c & 0x7f] x)
(MOVWQSX (ANDLconst [c] x)) && c & 0x8000 == 0 -> (ANDLconst [c & 0x7fff] x)
(MOVLQSX (ANDLconst [c] x)) && c & 0x80000000 == 0 -> (ANDLconst [c & 0x7fffffff] x)

// Don't extend before storing
(MOVLstore [off] {sym} ptr (MOVLQSX x) mem) -> (MOVLstore [off] {sym} ptr x mem)
(MOVWstore [off] {sym} ptr (MOVWQSX x) mem) -> (MOVWstore [off] {sym} ptr x mem)
(MOVBstore [off] {sym} ptr (MOVBQSX x) mem) -> (MOVBstore [off] {sym} ptr x mem)
(MOVLstore [off] {sym} ptr (MOVLQZX x) mem) -> (MOVLstore [off] {sym} ptr x mem)
(MOVWstore [off] {sym} ptr (MOVWQZX x) mem) -> (MOVWstore [off] {sym} ptr x mem)
(MOVBstore [off] {sym} ptr (MOVBQZX x) mem) -> (MOVBstore [off] {sym} ptr x mem)

// fold constants into memory operations
// Note that this is not always a good idea because if not all the uses of
// the ADDQconst get eliminated, we still have to compute the ADDQconst and we now
// have potentially two live values (ptr and (ADDQconst [off] ptr)) instead of one.
// Nevertheless, let's do it!
(MOVQload  [off1] {sym} (ADDQconst [off2] ptr) mem) && is32Bit(off1+off2) -> (MOVQload  [off1+off2] {sym} ptr mem)
(MOVLload  [off1] {sym} (ADDQconst [off2] ptr) mem) && is32Bit(off1+off2) -> (MOVLload  [off1+off2] {sym} ptr mem)
(MOVWload  [off1] {sym} (ADDQconst [off2] ptr) mem) && is32Bit(off1+off2) -> (MOVWload  [off1+off2] {sym} ptr mem)
(MOVBload  [off1] {sym} (ADDQconst [off2] ptr) mem) && is32Bit(off1+off2) -> (MOVBload  [off1+off2] {sym} ptr mem)
(MOVSSload [off1] {sym} (ADDQconst [off2] ptr) mem) && is32Bit(off1+off2) -> (MOVSSload [off1+off2] {sym} ptr mem)
(MOVSDload [off1] {sym} (ADDQconst [off2] ptr) mem) && is32Bit(off1+off2) -> (MOVSDload [off1+off2] {sym} ptr mem)
(MOVOload  [off1] {sym} (ADDQconst [off2] ptr) mem) && is32Bit(off1+off2) -> (MOVOload  [off1+off2] {sym} ptr mem)

(MOVQstore  [off1] {sym} (ADDQconst [off2] ptr) val mem) && is32Bit(off1+off2) -> (MOVQstore  [off1+off2] {sym} ptr val mem)
(MOVLstore  [off1] {sym} (ADDQconst [off2] ptr) val mem) && is32Bit(off1+off2) -> (MOVLstore  [off1+off2] {sym} ptr val mem)
(MOVWstore  [off1] {sym} (ADDQconst [off2] ptr) val mem) && is32Bit(off1+off2) -> (MOVWstore  [off1+off2] {sym} ptr val mem)
(MOVBstore  [off1] {sym} (ADDQconst [off2] ptr) val mem) && is32Bit(off1+off2) -> (MOVBstore  [off1+off2] {sym} ptr val mem)
(MOVSSstore [off1] {sym} (ADDQconst [off2] ptr) val mem) && is32Bit(off1+off2) -> (MOVSSstore [off1+off2] {sym} ptr val mem)
(MOVSDstore [off1] {sym} (ADDQconst [off2] ptr) val mem) && is32Bit(off1+off2) -> (MOVSDstore [off1+off2] {sym} ptr val mem)
(MOVOstore  [off1] {sym} (ADDQconst [off2] ptr) val mem) && is32Bit(off1+off2) -> (MOVOstore  [off1+off2] {sym} ptr val mem)

// Fold constants into stores.
(MOVQstore [off] {sym} ptr (MOVQconst [c]) mem) && validValAndOff(c,off) ->
	(MOVQstoreconst [makeValAndOff(c,off)] {sym} ptr mem)
(MOVLstore [off] {sym} ptr (MOVLconst [c]) mem) && validOff(off) ->
	(MOVLstoreconst [makeValAndOff(int64(int32(c)),off)] {sym} ptr mem)
(MOVWstore [off] {sym} ptr (MOVLconst [c]) mem) && validOff(off) ->
	(MOVWstoreconst [makeValAndOff(int64(int16(c)),off)] {sym} ptr mem)
(MOVBstore [off] {sym} ptr (MOVLconst [c]) mem) && validOff(off) ->
	(MOVBstoreconst [makeValAndOff(int64(int8(c)),off)] {sym} ptr mem)

// Fold address offsets into constant stores.
(MOVQstoreconst [sc] {s} (ADDQconst [off] ptr) mem) && ValAndOff(sc).canAdd(off) ->
	(MOVQstoreconst [ValAndOff(sc).add(off)] {s} ptr mem)
(MOVLstoreconst [sc] {s} (ADDQconst [off] ptr) mem) && ValAndOff(sc).canAdd(off) ->
	(MOVLstoreconst [ValAndOff(sc).add(off)] {s} ptr mem)
(MOVWstoreconst [sc] {s} (ADDQconst [off] ptr) mem) && ValAndOff(sc).canAdd(off) ->
	(MOVWstoreconst [ValAndOff(sc).add(off)] {s} ptr mem)
(MOVBstoreconst [sc] {s} (ADDQconst [off] ptr) mem) && ValAndOff(sc).canAdd(off) ->
	(MOVBstoreconst [ValAndOff(sc).add(off)] {s} ptr mem)

// We need to fold LEAQ into the MOVx ops so that the live variable analysis knows
// what variables are being read/written by the ops.
(MOVQload  [off1] {sym1} (LEAQ [off2] {sym2} base) mem) && is32Bit(off1+off2) && canMergeSym(sym1, sym2) ->
	(MOVQload  [off1+off2] {mergeSym(sym1,sym2)} base mem)
(MOVLload  [off1] {sym1} (LEAQ [off2] {sym2} base) mem) && is32Bit(off1+off2) && canMergeSym(sym1, sym2) ->
	(MOVLload  [off1+off2] {mergeSym(sym1,sym2)} base mem)
(MOVWload  [off1] {sym1} (LEAQ [off2] {sym2} base) mem) && is32Bit(off1+off2) && canMergeSym(sym1, sym2) ->
	(MOVWload  [off1+off2] {mergeSym(sym1,sym2)} base mem)
(MOVBload  [off1] {sym1} (LEAQ [off2] {sym2} base) mem) && is32Bit(off1+off2) && canMergeSym(sym1, sym2) ->
	(MOVBload  [off1+off2] {mergeSym(sym1,sym2)} base mem)
(MOVSSload [off1] {sym1} (LEAQ [off2] {sym2} base) mem) && is32Bit(off1+off2) && canMergeSym(sym1, sym2) ->
	(MOVSSload [off1+off2] {mergeSym(sym1,sym2)} base mem)
(MOVSDload [off1] {sym1} (LEAQ [off2] {sym2} base) mem) && is32Bit(off1+off2) && canMergeSym(sym1, sym2) ->
	(MOVSDload [off1+off2] {mergeSym(sym1,sym2)} base mem)
(MOVOload [off1] {sym1} (LEAQ [off2] {sym2} base) mem) && is32Bit(off1+off2) && canMergeSym(sym1, sym2) ->
	(MOVOload [off1+off2] {mergeSym(sym1,sym2)} base mem)

(MOVBQSXload [off1] {sym1} (LEAQ [off2] {sym2} base) mem) && is32Bit(off1+off2) && canMergeSym(sym1, sym2) ->
	(MOVBQSXload [off1+off2] {mergeSym(sym1,sym2)} base mem)
(MOVWQSXload [off1] {sym1} (LEAQ [off2] {sym2} base) mem) && is32Bit(off1+off2) && canMergeSym(sym1, sym2) ->
	(MOVWQSXload [off1+off2] {mergeSym(sym1,sym2)} base mem)
(MOVLQSXload [off1] {sym1} (LEAQ [off2] {sym2} base) mem) && is32Bit(off1+off2) && canMergeSym(sym1, sym2) ->
	(MOVLQSXload [off1+off2] {mergeSym(sym1,sym2)} base mem)

(MOVQstore  [off1] {sym1} (LEAQ [off2] {sym2} base) val mem) && is32Bit(off1+off2) && canMergeSym(sym1, sym2) ->
	(MOVQstore  [off1+off2] {mergeSym(sym1,sym2)} base val mem)
(MOVLstore  [off1] {sym1} (LEAQ [off2] {sym2} base) val mem) && is32Bit(off1+off2) && canMergeSym(sym1, sym2) ->
	(MOVLstore  [off1+off2] {mergeSym(sym1,sym2)} base val mem)
(MOVWstore  [off1] {sym1} (LEAQ [off2] {sym2} base) val mem) && is32Bit(off1+off2) && canMergeSym(sym1, sym2) ->
	(MOVWstore  [off1+off2] {mergeSym(sym1,sym2)} base val mem)
(MOVBstore  [off1] {sym1} (LEAQ [off2] {sym2} base) val mem) && is32Bit(off1+off2) && canMergeSym(sym1, sym2) ->
	(MOVBstore  [off1+off2] {mergeSym(sym1,sym2)} base val mem)
(MOVSSstore [off1] {sym1} (LEAQ [off2] {sym2} base) val mem) && is32Bit(off1+off2) && canMergeSym(sym1, sym2) ->
	(MOVSSstore [off1+off2] {mergeSym(sym1,sym2)} base val mem)
(MOVSDstore [off1] {sym1} (LEAQ [off2] {sym2} base) val mem) && is32Bit(off1+off2) && canMergeSym(sym1, sym2) ->
	(MOVSDstore [off1+off2] {mergeSym(sym1,sym2)} base val mem)
(MOVOstore [off1] {sym1} (LEAQ [off2] {sym2} base) val mem) && is32Bit(off1+off2) && canMergeSym(sym1, sym2) ->
	(MOVOstore [off1+off2] {mergeSym(sym1,sym2)} base val mem)

(MOVQstoreconst [sc] {sym1} (LEAQ [off] {sym2} ptr) mem) && canMergeSym(sym1, sym2) && ValAndOff(sc).canAdd(off) ->
	(MOVQstoreconst [ValAndOff(sc).add(off)] {mergeSym(sym1, sym2)} ptr mem)
(MOVLstoreconst [sc] {sym1} (LEAQ [off] {sym2} ptr) mem) && canMergeSym(sym1, sym2) && ValAndOff(sc).canAdd(off) ->
	(MOVLstoreconst [ValAndOff(sc).add(off)] {mergeSym(sym1, sym2)} ptr mem)
(MOVWstoreconst [sc] {sym1} (LEAQ [off] {sym2} ptr) mem) && canMergeSym(sym1, sym2) && ValAndOff(sc).canAdd(off) ->
	(MOVWstoreconst [ValAndOff(sc).add(off)] {mergeSym(sym1, sym2)} ptr mem)
(MOVBstoreconst [sc] {sym1} (LEAQ [off] {sym2} ptr) mem) && canMergeSym(sym1, sym2) && ValAndOff(sc).canAdd(off) ->
	(MOVBstoreconst [ValAndOff(sc).add(off)] {mergeSym(sym1, sym2)} ptr mem)

// generating indexed loads and stores
(MOVBload [off1] {sym1} (LEAQ1 [off2] {sym2} ptr idx) mem) && is32Bit(off1+off2) && canMergeSym(sym1, sym2) ->
	(MOVBloadidx1 [off1+off2] {mergeSym(sym1,sym2)} ptr idx mem)
(MOVWload [off1] {sym1} (LEAQ1 [off2] {sym2} ptr idx) mem) && is32Bit(off1+off2) && canMergeSym(sym1, sym2) ->
	(MOVWloadidx1 [off1+off2] {mergeSym(sym1,sym2)} ptr idx mem)
(MOVWload [off1] {sym1} (LEAQ2 [off2] {sym2} ptr idx) mem) && is32Bit(off1+off2) && canMergeSym(sym1, sym2) ->
	(MOVWloadidx2 [off1+off2] {mergeSym(sym1,sym2)} ptr idx mem)
(MOVLload [off1] {sym1} (LEAQ1 [off2] {sym2} ptr idx) mem) && is32Bit(off1+off2) && canMergeSym(sym1, sym2) ->
	(MOVLloadidx1 [off1+off2] {mergeSym(sym1,sym2)} ptr idx mem)
(MOVLload [off1] {sym1} (LEAQ4 [off2] {sym2} ptr idx) mem) && is32Bit(off1+off2) && canMergeSym(sym1, sym2) ->
	(MOVLloadidx4 [off1+off2] {mergeSym(sym1,sym2)} ptr idx mem)
(MOVLload [off1] {sym1} (LEAQ8 [off2] {sym2} ptr idx) mem) && is32Bit(off1+off2) && canMergeSym(sym1, sym2) ->
	(MOVLloadidx8 [off1+off2] {mergeSym(sym1,sym2)} ptr idx mem)
(MOVQload [off1] {sym1} (LEAQ1 [off2] {sym2} ptr idx) mem) && is32Bit(off1+off2) && canMergeSym(sym1, sym2) ->
	(MOVQloadidx1 [off1+off2] {mergeSym(sym1,sym2)} ptr idx mem)
(MOVQload [off1] {sym1} (LEAQ8 [off2] {sym2} ptr idx) mem) && is32Bit(off1+off2) && canMergeSym(sym1, sym2) ->
	(MOVQloadidx8 [off1+off2] {mergeSym(sym1,sym2)} ptr idx mem)
(MOVSSload [off1] {sym1} (LEAQ1 [off2] {sym2} ptr idx) mem) && is32Bit(off1+off2) && canMergeSym(sym1, sym2) ->
	(MOVSSloadidx1 [off1+off2] {mergeSym(sym1,sym2)} ptr idx mem)
(MOVSSload [off1] {sym1} (LEAQ4 [off2] {sym2} ptr idx) mem) && is32Bit(off1+off2) && canMergeSym(sym1, sym2) ->
	(MOVSSloadidx4 [off1+off2] {mergeSym(sym1,sym2)} ptr idx mem)
(MOVSDload [off1] {sym1} (LEAQ1 [off2] {sym2} ptr idx) mem) && is32Bit(off1+off2) && canMergeSym(sym1, sym2) ->
	(MOVSDloadidx1 [off1+off2] {mergeSym(sym1,sym2)} ptr idx mem)
(MOVSDload [off1] {sym1} (LEAQ8 [off2] {sym2} ptr idx) mem) && is32Bit(off1+off2) && canMergeSym(sym1, sym2) ->
	(MOVSDloadidx8 [off1+off2] {mergeSym(sym1,sym2)} ptr idx mem)

(MOVBstore [off1] {sym1} (LEAQ1 [off2] {sym2} ptr idx) val mem) && is32Bit(off1+off2) && canMergeSym(sym1, sym2) ->
	(MOVBstoreidx1 [off1+off2] {mergeSym(sym1,sym2)} ptr idx val mem)
(MOVWstore [off1] {sym1} (LEAQ1 [off2] {sym2} ptr idx) val mem) && is32Bit(off1+off2) && canMergeSym(sym1, sym2) ->
	(MOVWstoreidx1 [off1+off2] {mergeSym(sym1,sym2)} ptr idx val mem)
(MOVWstore [off1] {sym1} (LEAQ2 [off2] {sym2} ptr idx) val mem) && is32Bit(off1+off2) && canMergeSym(sym1, sym2) ->
	(MOVWstoreidx2 [off1+off2] {mergeSym(sym1,sym2)} ptr idx val mem)
(MOVLstore [off1] {sym1} (LEAQ1 [off2] {sym2} ptr idx) val mem) && is32Bit(off1+off2) && canMergeSym(sym1, sym2) ->
	(MOVLstoreidx1 [off1+off2] {mergeSym(sym1,sym2)} ptr idx val mem)
(MOVLstore [off1] {sym1} (LEAQ4 [off2] {sym2} ptr idx) val mem) && is32Bit(off1+off2) && canMergeSym(sym1, sym2) ->
	(MOVLstoreidx4 [off1+off2] {mergeSym(sym1,sym2)} ptr idx val mem)
(MOVLstore [off1] {sym1} (LEAQ8 [off2] {sym2} ptr idx) val mem) && is32Bit(off1+off2) && canMergeSym(sym1, sym2) ->
	(MOVLstoreidx8 [off1+off2] {mergeSym(sym1,sym2)} ptr idx val mem)
(MOVQstore [off1] {sym1} (LEAQ1 [off2] {sym2} ptr idx) val mem) && is32Bit(off1+off2) && canMergeSym(sym1, sym2) ->
	(MOVQstoreidx1 [off1+off2] {mergeSym(sym1,sym2)} ptr idx val mem)
(MOVQstore [off1] {sym1} (LEAQ8 [off2] {sym2} ptr idx) val mem) && is32Bit(off1+off2) && canMergeSym(sym1, sym2) ->
	(MOVQstoreidx8 [off1+off2] {mergeSym(sym1,sym2)} ptr idx val mem)
(MOVSSstore [off1] {sym1} (LEAQ1 [off2] {sym2} ptr idx) val mem) && is32Bit(off1+off2) && canMergeSym(sym1, sym2) ->
	(MOVSSstoreidx1 [off1+off2] {mergeSym(sym1,sym2)} ptr idx val mem)
(MOVSSstore [off1] {sym1} (LEAQ4 [off2] {sym2} ptr idx) val mem) && is32Bit(off1+off2) && canMergeSym(sym1, sym2) ->
	(MOVSSstoreidx4 [off1+off2] {mergeSym(sym1,sym2)} ptr idx val mem)
(MOVSDstore [off1] {sym1} (LEAQ1 [off2] {sym2} ptr idx) val mem) && is32Bit(off1+off2) && canMergeSym(sym1, sym2) ->
	(MOVSDstoreidx1 [off1+off2] {mergeSym(sym1,sym2)} ptr idx val mem)
(MOVSDstore [off1] {sym1} (LEAQ8 [off2] {sym2} ptr idx) val mem) && is32Bit(off1+off2) && canMergeSym(sym1, sym2) ->
	(MOVSDstoreidx8 [off1+off2] {mergeSym(sym1,sym2)} ptr idx val mem)

(MOVBload [off] {sym} (ADDQ ptr idx) mem) && ptr.Op != OpSB -> (MOVBloadidx1 [off] {sym} ptr idx mem)
(MOVWload [off] {sym} (ADDQ ptr idx) mem) && ptr.Op != OpSB -> (MOVWloadidx1 [off] {sym} ptr idx mem)
(MOVLload [off] {sym} (ADDQ ptr idx) mem) && ptr.Op != OpSB -> (MOVLloadidx1 [off] {sym} ptr idx mem)
(MOVQload [off] {sym} (ADDQ ptr idx) mem) && ptr.Op != OpSB -> (MOVQloadidx1 [off] {sym} ptr idx mem)
(MOVSSload [off] {sym} (ADDQ ptr idx) mem) && ptr.Op != OpSB -> (MOVSSloadidx1 [off] {sym} ptr idx mem)
(MOVSDload [off] {sym} (ADDQ ptr idx) mem) && ptr.Op != OpSB -> (MOVSDloadidx1 [off] {sym} ptr idx mem)
(MOVBstore [off] {sym} (ADDQ ptr idx) val mem) && ptr.Op != OpSB -> (MOVBstoreidx1 [off] {sym} ptr idx val mem)
(MOVWstore [off] {sym} (ADDQ ptr idx) val mem) && ptr.Op != OpSB -> (MOVWstoreidx1 [off] {sym} ptr idx val mem)
(MOVLstore [off] {sym} (ADDQ ptr idx) val mem) && ptr.Op != OpSB -> (MOVLstoreidx1 [off] {sym} ptr idx val mem)
(MOVQstore [off] {sym} (ADDQ ptr idx) val mem) && ptr.Op != OpSB -> (MOVQstoreidx1 [off] {sym} ptr idx val mem)
(MOVSSstore [off] {sym} (ADDQ ptr idx) val mem) && ptr.Op != OpSB -> (MOVSSstoreidx1 [off] {sym} ptr idx val mem)
(MOVSDstore [off] {sym} (ADDQ ptr idx) val mem) && ptr.Op != OpSB -> (MOVSDstoreidx1 [off] {sym} ptr idx val mem)

(MOVBstoreconst [x] {sym1} (LEAQ1 [off] {sym2} ptr idx) mem) && canMergeSym(sym1, sym2) ->
	(MOVBstoreconstidx1 [ValAndOff(x).add(off)] {mergeSym(sym1,sym2)} ptr idx mem)
(MOVWstoreconst [x] {sym1} (LEAQ1 [off] {sym2} ptr idx) mem) && canMergeSym(sym1, sym2) ->
	(MOVWstoreconstidx1 [ValAndOff(x).add(off)] {mergeSym(sym1,sym2)} ptr idx mem)
(MOVWstoreconst [x] {sym1} (LEAQ2 [off] {sym2} ptr idx) mem) && canMergeSym(sym1, sym2) ->
	(MOVWstoreconstidx2 [ValAndOff(x).add(off)] {mergeSym(sym1,sym2)} ptr idx mem)
(MOVLstoreconst [x] {sym1} (LEAQ1 [off] {sym2} ptr idx) mem) && canMergeSym(sym1, sym2) ->
	(MOVLstoreconstidx1 [ValAndOff(x).add(off)] {mergeSym(sym1,sym2)} ptr idx mem)
(MOVLstoreconst [x] {sym1} (LEAQ4 [off] {sym2} ptr idx) mem) && canMergeSym(sym1, sym2) ->
	(MOVLstoreconstidx4 [ValAndOff(x).add(off)] {mergeSym(sym1,sym2)} ptr idx mem)
(MOVQstoreconst [x] {sym1} (LEAQ1 [off] {sym2} ptr idx) mem) && canMergeSym(sym1, sym2) ->
	(MOVQstoreconstidx1 [ValAndOff(x).add(off)] {mergeSym(sym1,sym2)} ptr idx mem)
(MOVQstoreconst [x] {sym1} (LEAQ8 [off] {sym2} ptr idx) mem) && canMergeSym(sym1, sym2) ->
	(MOVQstoreconstidx8 [ValAndOff(x).add(off)] {mergeSym(sym1,sym2)} ptr idx mem)

(MOVBstoreconst [x] {sym} (ADDQ ptr idx) mem) -> (MOVBstoreconstidx1 [x] {sym} ptr idx mem)
(MOVWstoreconst [x] {sym} (ADDQ ptr idx) mem) -> (MOVWstoreconstidx1 [x] {sym} ptr idx mem)
(MOVLstoreconst [x] {sym} (ADDQ ptr idx) mem) -> (MOVLstoreconstidx1 [x] {sym} ptr idx mem)
(MOVQstoreconst [x] {sym} (ADDQ ptr idx) mem) -> (MOVQstoreconstidx1 [x] {sym} ptr idx mem)

// combine SHLQ into indexed loads and stores
(MOVWloadidx1 [c] {sym} ptr (SHLQconst [1] idx) mem) -> (MOVWloadidx2 [c] {sym} ptr idx mem)
(MOVLloadidx1 [c] {sym} ptr (SHLQconst [2] idx) mem) -> (MOVLloadidx4 [c] {sym} ptr idx mem)
(MOVLloadidx1 [c] {sym} ptr (SHLQconst [3] idx) mem) -> (MOVLloadidx8 [c] {sym} ptr idx mem)
(MOVQloadidx1 [c] {sym} ptr (SHLQconst [3] idx) mem) -> (MOVQloadidx8 [c] {sym} ptr idx mem)
(MOVSSloadidx1 [c] {sym} ptr (SHLQconst [2] idx) mem) -> (MOVSSloadidx4 [c] {sym} ptr idx mem)
(MOVSDloadidx1 [c] {sym} ptr (SHLQconst [3] idx) mem) -> (MOVSDloadidx8 [c] {sym} ptr idx mem)
(MOVWstoreidx1 [c] {sym} ptr (SHLQconst [1] idx) val mem) -> (MOVWstoreidx2 [c] {sym} ptr idx val mem)
(MOVLstoreidx1 [c] {sym} ptr (SHLQconst [2] idx) val mem) -> (MOVLstoreidx4 [c] {sym} ptr idx val mem)
(MOVLstoreidx1 [c] {sym} ptr (SHLQconst [3] idx) val mem) -> (MOVLstoreidx8 [c] {sym} ptr idx val mem)
(MOVQstoreidx1 [c] {sym} ptr (SHLQconst [3] idx) val mem) -> (MOVQstoreidx8 [c] {sym} ptr idx val mem)
(MOVSSstoreidx1 [c] {sym} ptr (SHLQconst [2] idx) val mem) -> (MOVSSstoreidx4 [c] {sym} ptr idx val mem)
(MOVSDstoreidx1 [c] {sym} ptr (SHLQconst [3] idx) val mem) -> (MOVSDstoreidx8 [c] {sym} ptr idx val mem)
(MOVWstoreconstidx1 [c] {sym} ptr (SHLQconst [1] idx) mem) -> (MOVWstoreconstidx2 [c] {sym} ptr idx mem)
(MOVLstoreconstidx1 [c] {sym} ptr (SHLQconst [2] idx) mem) -> (MOVLstoreconstidx4 [c] {sym} ptr idx mem)
(MOVQstoreconstidx1 [c] {sym} ptr (SHLQconst [3] idx) mem) -> (MOVQstoreconstidx8 [c] {sym} ptr idx mem)

// combine ADDQ into indexed loads and stores
(MOVBloadidx1 [c] {sym} (ADDQconst [d] ptr) idx mem) && is32Bit(c+d) -> (MOVBloadidx1 [c+d] {sym} ptr idx mem)
(MOVWloadidx1 [c] {sym} (ADDQconst [d] ptr) idx mem) && is32Bit(c+d) -> (MOVWloadidx1 [c+d] {sym} ptr idx mem)
(MOVWloadidx2 [c] {sym} (ADDQconst [d] ptr) idx mem) && is32Bit(c+d) -> (MOVWloadidx2 [c+d] {sym} ptr idx mem)
(MOVLloadidx1 [c] {sym} (ADDQconst [d] ptr) idx mem) && is32Bit(c+d) -> (MOVLloadidx1 [c+d] {sym} ptr idx mem)
(MOVLloadidx4 [c] {sym} (ADDQconst [d] ptr) idx mem) && is32Bit(c+d) -> (MOVLloadidx4 [c+d] {sym} ptr idx mem)
(MOVLloadidx8 [c] {sym} (ADDQconst [d] ptr) idx mem) && is32Bit(c+d) -> (MOVLloadidx8 [c+d] {sym} ptr idx mem)
(MOVQloadidx1 [c] {sym} (ADDQconst [d] ptr) idx mem) && is32Bit(c+d) -> (MOVQloadidx1 [c+d] {sym} ptr idx mem)
(MOVQloadidx8 [c] {sym} (ADDQconst [d] ptr) idx mem) && is32Bit(c+d) -> (MOVQloadidx8 [c+d] {sym} ptr idx mem)
(MOVSSloadidx1 [c] {sym} (ADDQconst [d] ptr) idx mem) && is32Bit(c+d) -> (MOVSSloadidx1 [c+d] {sym} ptr idx mem)
(MOVSSloadidx4 [c] {sym} (ADDQconst [d] ptr) idx mem) && is32Bit(c+d) -> (MOVSSloadidx4 [c+d] {sym} ptr idx mem)
(MOVSDloadidx1 [c] {sym} (ADDQconst [d] ptr) idx mem) && is32Bit(c+d) -> (MOVSDloadidx1 [c+d] {sym} ptr idx mem)
(MOVSDloadidx8 [c] {sym} (ADDQconst [d] ptr) idx mem) && is32Bit(c+d) -> (MOVSDloadidx8 [c+d] {sym} ptr idx mem)

(MOVBstoreidx1 [c] {sym} (ADDQconst [d] ptr) idx val mem) && is32Bit(c+d) -> (MOVBstoreidx1 [c+d] {sym} ptr idx val mem)
(MOVWstoreidx1 [c] {sym} (ADDQconst [d] ptr) idx val mem) && is32Bit(c+d) -> (MOVWstoreidx1 [c+d] {sym} ptr idx val mem)
(MOVWstoreidx2 [c] {sym} (ADDQconst [d] ptr) idx val mem) && is32Bit(c+d) -> (MOVWstoreidx2 [c+d] {sym} ptr idx val mem)
(MOVLstoreidx1 [c] {sym} (ADDQconst [d] ptr) idx val mem) && is32Bit(c+d) -> (MOVLstoreidx1 [c+d] {sym} ptr idx val mem)
(MOVLstoreidx4 [c] {sym} (ADDQconst [d] ptr) idx val mem) && is32Bit(c+d) -> (MOVLstoreidx4 [c+d] {sym} ptr idx val mem)
(MOVLstoreidx8 [c] {sym} (ADDQconst [d] ptr) idx val mem) && is32Bit(c+d) -> (MOVLstoreidx8 [c+d] {sym} ptr idx val mem)
(MOVQstoreidx1 [c] {sym} (ADDQconst [d] ptr) idx val mem) && is32Bit(c+d) -> (MOVQstoreidx1 [c+d] {sym} ptr idx val mem)
(MOVQstoreidx8 [c] {sym} (ADDQconst [d] ptr) idx val mem) && is32Bit(c+d) -> (MOVQstoreidx8 [c+d] {sym} ptr idx val mem)
(MOVSSstoreidx1 [c] {sym} (ADDQconst [d] ptr) idx val mem) && is32Bit(c+d) -> (MOVSSstoreidx1 [c+d] {sym} ptr idx val mem)
(MOVSSstoreidx4 [c] {sym} (ADDQconst [d] ptr) idx val mem) && is32Bit(c+d) -> (MOVSSstoreidx4 [c+d] {sym} ptr idx val mem)
(MOVSDstoreidx1 [c] {sym} (ADDQconst [d] ptr) idx val mem) && is32Bit(c+d) -> (MOVSDstoreidx1 [c+d] {sym} ptr idx val mem)
(MOVSDstoreidx8 [c] {sym} (ADDQconst [d] ptr) idx val mem) && is32Bit(c+d) -> (MOVSDstoreidx8 [c+d] {sym} ptr idx val mem)

(MOVBloadidx1 [c] {sym} ptr (ADDQconst [d] idx) mem)  && is32Bit(c+d)   -> (MOVBloadidx1 [c+d] {sym} ptr idx mem)
(MOVWloadidx1 [c] {sym} ptr (ADDQconst [d] idx) mem)  && is32Bit(c+d)   -> (MOVWloadidx1 [c+d] {sym} ptr idx mem)
(MOVWloadidx2 [c] {sym} ptr (ADDQconst [d] idx) mem)  && is32Bit(c+2*d) -> (MOVWloadidx2 [c+2*d] {sym} ptr idx mem)
(MOVLloadidx1 [c] {sym} ptr (ADDQconst [d] idx) mem)  && is32Bit(c+d)   -> (MOVLloadidx1 [c+d] {sym} ptr idx mem)
(MOVLloadidx4 [c] {sym} ptr (ADDQconst [d] idx) mem)  && is32Bit(c+4*d) -> (MOVLloadidx4 [c+4*d] {sym} ptr idx mem)
(MOVLloadidx8 [c] {sym} ptr (ADDQconst [d] idx) mem)  && is32Bit(c+8*d) -> (MOVLloadidx8 [c+8*d] {sym} ptr idx mem)
(MOVQloadidx1 [c] {sym} ptr (ADDQconst [d] idx) mem)  && is32Bit(c+d)   -> (MOVQloadidx1 [c+d] {sym} ptr idx mem)
(MOVQloadidx8 [c] {sym} ptr (ADDQconst [d] idx) mem)  && is32Bit(c+8*d) -> (MOVQloadidx8 [c+8*d] {sym} ptr idx mem)
(MOVSSloadidx1 [c] {sym} ptr (ADDQconst [d] idx) mem) && is32Bit(c+d)   -> (MOVSSloadidx1 [c+d] {sym} ptr idx mem)
(MOVSSloadidx4 [c] {sym} ptr (ADDQconst [d] idx) mem) && is32Bit(c+4*d) -> (MOVSSloadidx4 [c+4*d] {sym} ptr idx mem)
(MOVSDloadidx1 [c] {sym} ptr (ADDQconst [d] idx) mem) && is32Bit(c+d)   -> (MOVSDloadidx1 [c+d] {sym} ptr idx mem)
(MOVSDloadidx8 [c] {sym} ptr (ADDQconst [d] idx) mem) && is32Bit(c+8*d) -> (MOVSDloadidx8 [c+8*d] {sym} ptr idx mem)

(MOVBstoreidx1 [c] {sym} ptr (ADDQconst [d] idx) val mem)  && is32Bit(c+d)   -> (MOVBstoreidx1 [c+d] {sym} ptr idx val mem)
(MOVWstoreidx1 [c] {sym} ptr (ADDQconst [d] idx) val mem)  && is32Bit(c+d)   -> (MOVWstoreidx1 [c+d] {sym} ptr idx val mem)
(MOVWstoreidx2 [c] {sym} ptr (ADDQconst [d] idx) val mem)  && is32Bit(c+2*d) -> (MOVWstoreidx2 [c+2*d] {sym} ptr idx val mem)
(MOVLstoreidx1 [c] {sym} ptr (ADDQconst [d] idx) val mem)  && is32Bit(c+d)   -> (MOVLstoreidx1 [c+d] {sym} ptr idx val mem)
(MOVLstoreidx4 [c] {sym} ptr (ADDQconst [d] idx) val mem)  && is32Bit(c+4*d) -> (MOVLstoreidx4 [c+4*d] {sym} ptr idx val mem)
(MOVLstoreidx8 [c] {sym} ptr (ADDQconst [d] idx) val mem)  && is32Bit(c+8*d) -> (MOVLstoreidx8 [c+8*d] {sym} ptr idx val mem)
(MOVQstoreidx1 [c] {sym} ptr (ADDQconst [d] idx) val mem)  && is32Bit(c+d)   -> (MOVQstoreidx1 [c+d] {sym} ptr idx val mem)
(MOVQstoreidx8 [c] {sym} ptr (ADDQconst [d] idx) val mem)  && is32Bit(c+8*d) -> (MOVQstoreidx8 [c+8*d] {sym} ptr idx val mem)
(MOVSSstoreidx1 [c] {sym} ptr (ADDQconst [d] idx) val mem) && is32Bit(c+d)   -> (MOVSSstoreidx1 [c+d] {sym} ptr idx val mem)
(MOVSSstoreidx4 [c] {sym} ptr (ADDQconst [d] idx) val mem) && is32Bit(c+4*d) -> (MOVSSstoreidx4 [c+4*d] {sym} ptr idx val mem)
(MOVSDstoreidx1 [c] {sym} ptr (ADDQconst [d] idx) val mem) && is32Bit(c+d)   -> (MOVSDstoreidx1 [c+d] {sym} ptr idx val mem)
(MOVSDstoreidx8 [c] {sym} ptr (ADDQconst [d] idx) val mem) && is32Bit(c+8*d) -> (MOVSDstoreidx8 [c+8*d] {sym} ptr idx val mem)

(MOVBstoreconstidx1 [x] {sym} (ADDQconst [c] ptr) idx mem) && ValAndOff(x).canAdd(c) ->
	(MOVBstoreconstidx1 [ValAndOff(x).add(c)] {sym} ptr idx mem)
(MOVWstoreconstidx1 [x] {sym} (ADDQconst [c] ptr) idx mem) && ValAndOff(x).canAdd(c) ->
	(MOVWstoreconstidx1 [ValAndOff(x).add(c)] {sym} ptr idx mem)
(MOVWstoreconstidx2 [x] {sym} (ADDQconst [c] ptr) idx mem) && ValAndOff(x).canAdd(c) ->
	(MOVWstoreconstidx2 [ValAndOff(x).add(c)] {sym} ptr idx mem)
(MOVLstoreconstidx1 [x] {sym} (ADDQconst [c] ptr) idx mem) && ValAndOff(x).canAdd(c) ->
	(MOVLstoreconstidx1 [ValAndOff(x).add(c)] {sym} ptr idx mem)
(MOVLstoreconstidx4 [x] {sym} (ADDQconst [c] ptr) idx mem) && ValAndOff(x).canAdd(c) ->
	(MOVLstoreconstidx4 [ValAndOff(x).add(c)] {sym} ptr idx mem)
(MOVQstoreconstidx1 [x] {sym} (ADDQconst [c] ptr) idx mem) && ValAndOff(x).canAdd(c) ->
	(MOVQstoreconstidx1 [ValAndOff(x).add(c)] {sym} ptr idx mem)
(MOVQstoreconstidx8 [x] {sym} (ADDQconst [c] ptr) idx mem) && ValAndOff(x).canAdd(c) ->
	(MOVQstoreconstidx8 [ValAndOff(x).add(c)] {sym} ptr idx mem)

(MOVBstoreconstidx1 [x] {sym} ptr (ADDQconst [c] idx) mem) && ValAndOff(x).canAdd(c) ->
	(MOVBstoreconstidx1 [ValAndOff(x).add(c)] {sym} ptr idx mem)
(MOVWstoreconstidx1 [x] {sym} ptr (ADDQconst [c] idx) mem) && ValAndOff(x).canAdd(c) ->
	(MOVWstoreconstidx1 [ValAndOff(x).add(c)] {sym} ptr idx mem)
(MOVWstoreconstidx2 [x] {sym} ptr (ADDQconst [c] idx) mem) && ValAndOff(x).canAdd(2*c) ->
	(MOVWstoreconstidx2 [ValAndOff(x).add(2*c)] {sym} ptr idx mem)
(MOVLstoreconstidx1 [x] {sym} ptr (ADDQconst [c] idx) mem) && ValAndOff(x).canAdd(c) ->
	(MOVLstoreconstidx1 [ValAndOff(x).add(c)] {sym} ptr idx mem)
(MOVLstoreconstidx4 [x] {sym} ptr (ADDQconst [c] idx) mem) && ValAndOff(x).canAdd(4*c) ->
	(MOVLstoreconstidx4 [ValAndOff(x).add(4*c)] {sym} ptr idx mem)
(MOVQstoreconstidx1 [x] {sym} ptr (ADDQconst [c] idx) mem) && ValAndOff(x).canAdd(c) ->
	(MOVQstoreconstidx1 [ValAndOff(x).add(c)] {sym} ptr idx mem)
(MOVQstoreconstidx8 [x] {sym} ptr (ADDQconst [c] idx) mem) && ValAndOff(x).canAdd(8*c) ->
	(MOVQstoreconstidx8 [ValAndOff(x).add(8*c)] {sym} ptr idx mem)

// fold LEAQs together
(LEAQ [off1] {sym1} (LEAQ [off2] {sym2} x)) && is32Bit(off1+off2) && canMergeSym(sym1, sym2) ->
      (LEAQ [off1+off2] {mergeSym(sym1,sym2)} x)

// LEAQ into LEAQ1
(LEAQ1 [off1] {sym1} (LEAQ [off2] {sym2} x) y) && is32Bit(off1+off2) && canMergeSym(sym1, sym2) && x.Op != OpSB ->
       (LEAQ1 [off1+off2] {mergeSym(sym1,sym2)} x y)

// LEAQ1 into LEAQ
(LEAQ [off1] {sym1} (LEAQ1 [off2] {sym2} x y)) && is32Bit(off1+off2) && canMergeSym(sym1, sym2) ->
       (LEAQ1 [off1+off2] {mergeSym(sym1,sym2)} x y)

// LEAQ into LEAQ[248]
(LEAQ2 [off1] {sym1} (LEAQ [off2] {sym2} x) y) && is32Bit(off1+off2) && canMergeSym(sym1, sym2) && x.Op != OpSB ->
       (LEAQ2 [off1+off2] {mergeSym(sym1,sym2)} x y)
(LEAQ4 [off1] {sym1} (LEAQ [off2] {sym2} x) y) && is32Bit(off1+off2) && canMergeSym(sym1, sym2) && x.Op != OpSB ->
       (LEAQ4 [off1+off2] {mergeSym(sym1,sym2)} x y)
(LEAQ8 [off1] {sym1} (LEAQ [off2] {sym2} x) y) && is32Bit(off1+off2) && canMergeSym(sym1, sym2) && x.Op != OpSB ->
       (LEAQ8 [off1+off2] {mergeSym(sym1,sym2)} x y)

// LEAQ[248] into LEAQ
(LEAQ [off1] {sym1} (LEAQ2 [off2] {sym2} x y)) && is32Bit(off1+off2) && canMergeSym(sym1, sym2) ->
      (LEAQ2 [off1+off2] {mergeSym(sym1,sym2)} x y)
(LEAQ [off1] {sym1} (LEAQ4 [off2] {sym2} x y)) && is32Bit(off1+off2) && canMergeSym(sym1, sym2) ->
      (LEAQ4 [off1+off2] {mergeSym(sym1,sym2)} x y)
(LEAQ [off1] {sym1} (LEAQ8 [off2] {sym2} x y)) && is32Bit(off1+off2) && canMergeSym(sym1, sym2) ->
      (LEAQ8 [off1+off2] {mergeSym(sym1,sym2)} x y)

// Absorb InvertFlags into branches.
(LT (InvertFlags cmp) yes no) -> (GT cmp yes no)
(GT (InvertFlags cmp) yes no) -> (LT cmp yes no)
(LE (InvertFlags cmp) yes no) -> (GE cmp yes no)
(GE (InvertFlags cmp) yes no) -> (LE cmp yes no)
(ULT (InvertFlags cmp) yes no) -> (UGT cmp yes no)
(UGT (InvertFlags cmp) yes no) -> (ULT cmp yes no)
(ULE (InvertFlags cmp) yes no) -> (UGE cmp yes no)
(UGE (InvertFlags cmp) yes no) -> (ULE cmp yes no)
(EQ (InvertFlags cmp) yes no) -> (EQ cmp yes no)
(NE (InvertFlags cmp) yes no) -> (NE cmp yes no)

// Constant comparisons.
(CMPQconst (MOVQconst [x]) [y]) && x==y -> (FlagEQ)
(CMPQconst (MOVQconst [x]) [y]) && x<y && uint64(x)<uint64(y) -> (FlagLT_ULT)
(CMPQconst (MOVQconst [x]) [y]) && x<y && uint64(x)>uint64(y) -> (FlagLT_UGT)
(CMPQconst (MOVQconst [x]) [y]) && x>y && uint64(x)<uint64(y) -> (FlagGT_ULT)
(CMPQconst (MOVQconst [x]) [y]) && x>y && uint64(x)>uint64(y) -> (FlagGT_UGT)
(CMPLconst (MOVLconst [x]) [y]) && int32(x)==int32(y) -> (FlagEQ)
(CMPLconst (MOVLconst [x]) [y]) && int32(x)<int32(y) && uint32(x)<uint32(y) -> (FlagLT_ULT)
(CMPLconst (MOVLconst [x]) [y]) && int32(x)<int32(y) && uint32(x)>uint32(y) -> (FlagLT_UGT)
(CMPLconst (MOVLconst [x]) [y]) && int32(x)>int32(y) && uint32(x)<uint32(y) -> (FlagGT_ULT)
(CMPLconst (MOVLconst [x]) [y]) && int32(x)>int32(y) && uint32(x)>uint32(y) -> (FlagGT_UGT)
(CMPWconst (MOVLconst [x]) [y]) && int16(x)==int16(y) -> (FlagEQ)
(CMPWconst (MOVLconst [x]) [y]) && int16(x)<int16(y) && uint16(x)<uint16(y) -> (FlagLT_ULT)
(CMPWconst (MOVLconst [x]) [y]) && int16(x)<int16(y) && uint16(x)>uint16(y) -> (FlagLT_UGT)
(CMPWconst (MOVLconst [x]) [y]) && int16(x)>int16(y) && uint16(x)<uint16(y) -> (FlagGT_ULT)
(CMPWconst (MOVLconst [x]) [y]) && int16(x)>int16(y) && uint16(x)>uint16(y) -> (FlagGT_UGT)
(CMPBconst (MOVLconst [x]) [y]) && int8(x)==int8(y) -> (FlagEQ)
(CMPBconst (MOVLconst [x]) [y]) && int8(x)<int8(y) && uint8(x)<uint8(y) -> (FlagLT_ULT)
(CMPBconst (MOVLconst [x]) [y]) && int8(x)<int8(y) && uint8(x)>uint8(y) -> (FlagLT_UGT)
(CMPBconst (MOVLconst [x]) [y]) && int8(x)>int8(y) && uint8(x)<uint8(y) -> (FlagGT_ULT)
(CMPBconst (MOVLconst [x]) [y]) && int8(x)>int8(y) && uint8(x)>uint8(y) -> (FlagGT_UGT)

// Other known comparisons.
(CMPQconst (MOVBQZX _) [c]) && 0xFF < c -> (FlagLT_ULT)
(CMPQconst (MOVWQZX _) [c]) && 0xFFFF < c -> (FlagLT_ULT)
(CMPQconst (MOVLQZX _) [c]) && 0xFFFFFFFF < c -> (FlagLT_ULT)
(CMPLconst (SHRLconst _ [c]) [n]) && 0 <= n && 0 < c && c <= 32 && (1<<uint64(32-c)) <= uint64(n) -> (FlagLT_ULT)
(CMPQconst (SHRQconst _ [c]) [n]) && 0 <= n && 0 < c && c <= 64 && (1<<uint64(64-c)) <= uint64(n) -> (FlagLT_ULT)
(CMPQconst (ANDQconst _ [m]) [n]) && 0 <= m && m < n -> (FlagLT_ULT)
(CMPQconst (ANDLconst _ [m]) [n]) && 0 <= m && m < n -> (FlagLT_ULT)
(CMPLconst (ANDLconst _ [m]) [n]) && 0 <= int32(m) && int32(m) < int32(n) -> (FlagLT_ULT)
(CMPWconst (ANDLconst _ [m]) [n]) && 0 <= int16(m) && int16(m) < int16(n) -> (FlagLT_ULT)
(CMPBconst (ANDLconst _ [m]) [n]) && 0 <= int8(m) && int8(m) < int8(n) -> (FlagLT_ULT)

// TODO: DIVxU also.

// Absorb flag constants into SBB ops.
(SBBQcarrymask (FlagEQ)) -> (MOVQconst [0])
(SBBQcarrymask (FlagLT_ULT)) -> (MOVQconst [-1])
(SBBQcarrymask (FlagLT_UGT)) -> (MOVQconst [0])
(SBBQcarrymask (FlagGT_ULT)) -> (MOVQconst [-1])
(SBBQcarrymask (FlagGT_UGT)) -> (MOVQconst [0])
(SBBLcarrymask (FlagEQ)) -> (MOVLconst [0])
(SBBLcarrymask (FlagLT_ULT)) -> (MOVLconst [-1])
(SBBLcarrymask (FlagLT_UGT)) -> (MOVLconst [0])
(SBBLcarrymask (FlagGT_ULT)) -> (MOVLconst [-1])
(SBBLcarrymask (FlagGT_UGT)) -> (MOVLconst [0])

// Absorb flag constants into branches.
(EQ (FlagEQ) yes no) -> (First nil yes no)
(EQ (FlagLT_ULT) yes no) -> (First nil no yes)
(EQ (FlagLT_UGT) yes no) -> (First nil no yes)
(EQ (FlagGT_ULT) yes no) -> (First nil no yes)
(EQ (FlagGT_UGT) yes no) -> (First nil no yes)

(NE (FlagEQ) yes no) -> (First nil no yes)
(NE (FlagLT_ULT) yes no) -> (First nil yes no)
(NE (FlagLT_UGT) yes no) -> (First nil yes no)
(NE (FlagGT_ULT) yes no) -> (First nil yes no)
(NE (FlagGT_UGT) yes no) -> (First nil yes no)

(LT (FlagEQ) yes no) -> (First nil no yes)
(LT (FlagLT_ULT) yes no) -> (First nil yes no)
(LT (FlagLT_UGT) yes no) -> (First nil yes no)
(LT (FlagGT_ULT) yes no) -> (First nil no yes)
(LT (FlagGT_UGT) yes no) -> (First nil no yes)

(LE (FlagEQ) yes no) -> (First nil yes no)
(LE (FlagLT_ULT) yes no) -> (First nil yes no)
(LE (FlagLT_UGT) yes no) -> (First nil yes no)
(LE (FlagGT_ULT) yes no) -> (First nil no yes)
(LE (FlagGT_UGT) yes no) -> (First nil no yes)

(GT (FlagEQ) yes no) -> (First nil no yes)
(GT (FlagLT_ULT) yes no) -> (First nil no yes)
(GT (FlagLT_UGT) yes no) -> (First nil no yes)
(GT (FlagGT_ULT) yes no) -> (First nil yes no)
(GT (FlagGT_UGT) yes no) -> (First nil yes no)

(GE (FlagEQ) yes no) -> (First nil yes no)
(GE (FlagLT_ULT) yes no) -> (First nil no yes)
(GE (FlagLT_UGT) yes no) -> (First nil no yes)
(GE (FlagGT_ULT) yes no) -> (First nil yes no)
(GE (FlagGT_UGT) yes no) -> (First nil yes no)

(ULT (FlagEQ) yes no) -> (First nil no yes)
(ULT (FlagLT_ULT) yes no) -> (First nil yes no)
(ULT (FlagLT_UGT) yes no) -> (First nil no yes)
(ULT (FlagGT_ULT) yes no) -> (First nil yes no)
(ULT (FlagGT_UGT) yes no) -> (First nil no yes)

(ULE (FlagEQ) yes no) -> (First nil yes no)
(ULE (FlagLT_ULT) yes no) -> (First nil yes no)
(ULE (FlagLT_UGT) yes no) -> (First nil no yes)
(ULE (FlagGT_ULT) yes no) -> (First nil yes no)
(ULE (FlagGT_UGT) yes no) -> (First nil no yes)

(UGT (FlagEQ) yes no) -> (First nil no yes)
(UGT (FlagLT_ULT) yes no) -> (First nil no yes)
(UGT (FlagLT_UGT) yes no) -> (First nil yes no)
(UGT (FlagGT_ULT) yes no) -> (First nil no yes)
(UGT (FlagGT_UGT) yes no) -> (First nil yes no)

(UGE (FlagEQ) yes no) -> (First nil yes no)
(UGE (FlagLT_ULT) yes no) -> (First nil no yes)
(UGE (FlagLT_UGT) yes no) -> (First nil yes no)
(UGE (FlagGT_ULT) yes no) -> (First nil no yes)
(UGE (FlagGT_UGT) yes no) -> (First nil yes no)

// Absorb flag constants into SETxx ops.
(SETEQ (FlagEQ)) -> (MOVLconst [1])
(SETEQ (FlagLT_ULT)) -> (MOVLconst [0])
(SETEQ (FlagLT_UGT)) -> (MOVLconst [0])
(SETEQ (FlagGT_ULT)) -> (MOVLconst [0])
(SETEQ (FlagGT_UGT)) -> (MOVLconst [0])

(SETNE (FlagEQ)) -> (MOVLconst [0])
(SETNE (FlagLT_ULT)) -> (MOVLconst [1])
(SETNE (FlagLT_UGT)) -> (MOVLconst [1])
(SETNE (FlagGT_ULT)) -> (MOVLconst [1])
(SETNE (FlagGT_UGT)) -> (MOVLconst [1])

(SETL (FlagEQ)) -> (MOVLconst [0])
(SETL (FlagLT_ULT)) -> (MOVLconst [1])
(SETL (FlagLT_UGT)) -> (MOVLconst [1])
(SETL (FlagGT_ULT)) -> (MOVLconst [0])
(SETL (FlagGT_UGT)) -> (MOVLconst [0])

(SETLE (FlagEQ)) -> (MOVLconst [1])
(SETLE (FlagLT_ULT)) -> (MOVLconst [1])
(SETLE (FlagLT_UGT)) -> (MOVLconst [1])
(SETLE (FlagGT_ULT)) -> (MOVLconst [0])
(SETLE (FlagGT_UGT)) -> (MOVLconst [0])

(SETG (FlagEQ)) -> (MOVLconst [0])
(SETG (FlagLT_ULT)) -> (MOVLconst [0])
(SETG (FlagLT_UGT)) -> (MOVLconst [0])
(SETG (FlagGT_ULT)) -> (MOVLconst [1])
(SETG (FlagGT_UGT)) -> (MOVLconst [1])

(SETGE (FlagEQ)) -> (MOVLconst [1])
(SETGE (FlagLT_ULT)) -> (MOVLconst [0])
(SETGE (FlagLT_UGT)) -> (MOVLconst [0])
(SETGE (FlagGT_ULT)) -> (MOVLconst [1])
(SETGE (FlagGT_UGT)) -> (MOVLconst [1])

(SETB (FlagEQ)) -> (MOVLconst [0])
(SETB (FlagLT_ULT)) -> (MOVLconst [1])
(SETB (FlagLT_UGT)) -> (MOVLconst [0])
(SETB (FlagGT_ULT)) -> (MOVLconst [1])
(SETB (FlagGT_UGT)) -> (MOVLconst [0])

(SETBE (FlagEQ)) -> (MOVLconst [1])
(SETBE (FlagLT_ULT)) -> (MOVLconst [1])
(SETBE (FlagLT_UGT)) -> (MOVLconst [0])
(SETBE (FlagGT_ULT)) -> (MOVLconst [1])
(SETBE (FlagGT_UGT)) -> (MOVLconst [0])

(SETA (FlagEQ)) -> (MOVLconst [0])
(SETA (FlagLT_ULT)) -> (MOVLconst [0])
(SETA (FlagLT_UGT)) -> (MOVLconst [1])
(SETA (FlagGT_ULT)) -> (MOVLconst [0])
(SETA (FlagGT_UGT)) -> (MOVLconst [1])

(SETAE (FlagEQ)) -> (MOVLconst [1])
(SETAE (FlagLT_ULT)) -> (MOVLconst [0])
(SETAE (FlagLT_UGT)) -> (MOVLconst [1])
(SETAE (FlagGT_ULT)) -> (MOVLconst [0])
(SETAE (FlagGT_UGT)) -> (MOVLconst [1])

// Remove redundant *const ops
(ADDQconst [0] x) -> x
(ADDLconst [c] x) && int32(c)==0 -> x
(SUBQconst [0] x) -> x
(SUBLconst [c] x) && int32(c) == 0 -> x
(ANDQconst [0] _)                 -> (MOVQconst [0])
(ANDLconst [c] _) && int32(c)==0  -> (MOVLconst [0])
(ANDQconst [-1] x)                -> x
(ANDLconst [c] x) && int32(c)==-1 -> x
(ORQconst [0] x)                  -> x
(ORLconst [c] x) && int32(c)==0   -> x
(ORQconst [-1] _)                 -> (MOVQconst [-1])
(ORLconst [c] _) && int32(c)==-1  -> (MOVLconst [-1])
(XORQconst [0] x)                  -> x
(XORLconst [c] x) && int32(c)==0   -> x
// TODO: since we got rid of the W/B versions, we might miss
// things like (ANDLconst [0x100] x) which were formerly
// (ANDBconst [0] x).  Probably doesn't happen very often.
// If we cared, we might do:
//  (ANDLconst <t> [c] x) && t.Size()==1 && int8(x)==0 -> (MOVLconst [0])

// Convert constant subtracts to constant adds
(SUBQconst [c] x) && c != -(1<<31) -> (ADDQconst [-c] x)
(SUBLconst [c] x) -> (ADDLconst [int64(int32(-c))] x)

// generic constant folding
// TODO: more of this
(ADDQconst [c] (MOVQconst [d])) -> (MOVQconst [c+d])
(ADDLconst [c] (MOVLconst [d])) -> (MOVLconst [int64(int32(c+d))])
(ADDQconst [c] (ADDQconst [d] x)) && is32Bit(c+d) -> (ADDQconst [c+d] x)
(ADDLconst [c] (ADDLconst [d] x)) -> (ADDLconst [int64(int32(c+d))] x)
(SUBQconst (MOVQconst [d]) [c]) -> (MOVQconst [d-c])
(SUBQconst (SUBQconst x [d]) [c]) && is32Bit(-c-d) -> (ADDQconst [-c-d] x)
(SARQconst [c] (MOVQconst [d])) -> (MOVQconst [d>>uint64(c)])
(SARLconst [c] (MOVQconst [d])) -> (MOVQconst [d>>uint64(c)])
(SARWconst [c] (MOVQconst [d])) -> (MOVQconst [d>>uint64(c)])
(SARBconst [c] (MOVQconst [d])) -> (MOVQconst [d>>uint64(c)])
(NEGQ (MOVQconst [c])) -> (MOVQconst [-c])
(NEGL (MOVLconst [c])) -> (MOVLconst [int64(int32(-c))])
(MULQconst [c] (MOVQconst [d])) -> (MOVQconst [c*d])
(MULLconst [c] (MOVLconst [d])) -> (MOVLconst [int64(int32(c*d))])
(ANDQconst [c] (MOVQconst [d])) -> (MOVQconst [c&d])
(ANDLconst [c] (MOVLconst [d])) -> (MOVLconst [c&d])
(ORQconst [c] (MOVQconst [d])) -> (MOVQconst [c|d])
(ORLconst [c] (MOVLconst [d])) -> (MOVLconst [c|d])
(XORQconst [c] (MOVQconst [d])) -> (MOVQconst [c^d])
(XORLconst [c] (MOVLconst [d])) -> (MOVLconst [c^d])
(NOTQ (MOVQconst [c])) -> (MOVQconst [^c])
(NOTL (MOVLconst [c])) -> (MOVLconst [^c])

// generic simplifications
// TODO: more of this
(ADDQ x (NEGQ y)) -> (SUBQ x y)
(ADDL x (NEGL y)) -> (SUBL x y)
(SUBQ x x) -> (MOVQconst [0])
(SUBL x x) -> (MOVLconst [0])
(ANDQ x x) -> x
(ANDL x x) -> x
(ORQ x x) -> x
(ORL x x) -> x
(XORQ x x) -> (MOVQconst [0])
(XORL x x) -> (MOVLconst [0])
(NEGQ (ADDQconst [c] (NEGQ x))) && c != -(1<<31) -> (ADDQconst [-c] x)

// checking AND against 0.
(CMPQconst (ANDQ x y) [0]) -> (TESTQ x y)
(CMPLconst (ANDL x y) [0]) -> (TESTL x y)
(CMPWconst (ANDL x y) [0]) -> (TESTW x y)
(CMPBconst (ANDL x y) [0]) -> (TESTB x y)
(CMPQconst (ANDQconst [c] x) [0]) -> (TESTQconst [c] x)
(CMPLconst (ANDLconst [c] x) [0]) -> (TESTLconst [c] x)
(CMPWconst (ANDLconst [c] x) [0]) -> (TESTWconst [int64(int16(c))] x)
(CMPBconst (ANDLconst [c] x) [0]) -> (TESTBconst [int64(int8(c))] x)

// Convert TESTx to TESTxconst if possible.
(TESTQ (MOVQconst [c]) x) && is32Bit(c) -> (TESTQconst [c] x)
(TESTL (MOVLconst [c]) x) -> (TESTLconst [c] x)
(TESTW (MOVLconst [c]) x) -> (TESTWconst [c] x)
(TESTB (MOVLconst [c]) x) -> (TESTBconst [c] x)

// TEST %reg,%reg is shorter than CMP
(CMPQconst x [0]) -> (TESTQ x x)
(CMPLconst x [0]) -> (TESTL x x)
(CMPWconst x [0]) -> (TESTW x x)
(CMPBconst x [0]) -> (TESTB x x)

// Combining byte loads into larger (unaligned) loads.
// There are many ways these combinations could occur.  This is
// designed to match the way encoding/binary.LittleEndian does it.

// Little-endian loads

(ORL                  x0:(MOVBload [i0] {s} p mem)
    sh:(SHLLconst [8] x1:(MOVBload [i1] {s} p mem)))
  && i1 == i0+1
  && x0.Uses == 1
  && x1.Uses == 1
  && sh.Uses == 1
  && mergePoint(b,x0,x1) != nil
  && clobber(x0)
  && clobber(x1)
  && clobber(sh)
  -> @mergePoint(b,x0,x1) (MOVWload [i0] {s} p mem)

(ORQ                  x0:(MOVBload [i0] {s} p mem)
    sh:(SHLQconst [8] x1:(MOVBload [i1] {s} p mem)))
  && i1 == i0+1
  && x0.Uses == 1
  && x1.Uses == 1
  && sh.Uses == 1
  && mergePoint(b,x0,x1) != nil
  && clobber(x0)
  && clobber(x1)
  && clobber(sh)
  -> @mergePoint(b,x0,x1) (MOVWload [i0] {s} p mem)

(ORL                   x0:(MOVWload [i0] {s} p mem)
    sh:(SHLLconst [16] x1:(MOVWload [i1] {s} p mem)))
  && i1 == i0+2
  && x0.Uses == 1
  && x1.Uses == 1
  && sh.Uses == 1
  && mergePoint(b,x0,x1) != nil
  && clobber(x0)
  && clobber(x1)
  && clobber(sh)
  -> @mergePoint(b,x0,x1) (MOVLload [i0] {s} p mem)

(ORQ                   x0:(MOVWload [i0] {s} p mem)
    sh:(SHLQconst [16] x1:(MOVWload [i1] {s} p mem)))
  && i1 == i0+2
  && x0.Uses == 1
  && x1.Uses == 1
  && sh.Uses == 1
  && mergePoint(b,x0,x1) != nil
  && clobber(x0)
  && clobber(x1)
  && clobber(sh)
  -> @mergePoint(b,x0,x1) (MOVLload [i0] {s} p mem)

(ORQ                   x0:(MOVLload [i0] {s} p mem)
    sh:(SHLQconst [32] x1:(MOVLload [i1] {s} p mem)))
  && i1 == i0+4
  && x0.Uses == 1
  && x1.Uses == 1
  && sh.Uses == 1
  && mergePoint(b,x0,x1) != nil
  && clobber(x0)
  && clobber(x1)
  && clobber(sh)
  -> @mergePoint(b,x0,x1) (MOVQload [i0] {s} p mem)

(ORL
    s1:(SHLLconst [j1] x1:(MOVBload [i1] {s} p mem))
    or:(ORL
        s0:(SHLLconst [j0] x0:(MOVBload [i0] {s} p mem))
	y))
  && i1 == i0+1
  && j1 == j0+8
  && j0 % 16 == 0
  && x0.Uses == 1
  && x1.Uses == 1
  && s0.Uses == 1
  && s1.Uses == 1
  && or.Uses == 1
  && mergePoint(b,x0,x1) != nil
  && clobber(x0)
  && clobber(x1)
  && clobber(s0)
  && clobber(s1)
  && clobber(or)
  -> @mergePoint(b,x0,x1) (ORL <v.Type> (SHLLconst <v.Type> [j0] (MOVWload [i0] {s} p mem)) y)

(ORQ
    s1:(SHLQconst [j1] x1:(MOVBload [i1] {s} p mem))
    or:(ORQ
        s0:(SHLQconst [j0] x0:(MOVBload [i0] {s} p mem))
	y))
  && i1 == i0+1
  && j1 == j0+8
  && j0 % 16 == 0
  && x0.Uses == 1
  && x1.Uses == 1
  && s0.Uses == 1
  && s1.Uses == 1
  && or.Uses == 1
  && mergePoint(b,x0,x1) != nil
  && clobber(x0)
  && clobber(x1)
  && clobber(s0)
  && clobber(s1)
  && clobber(or)
  -> @mergePoint(b,x0,x1) (ORQ <v.Type> (SHLQconst <v.Type> [j0] (MOVWload [i0] {s} p mem)) y)

(ORQ
    s1:(SHLQconst [j1] x1:(MOVWload [i1] {s} p mem))
    or:(ORQ
        s0:(SHLQconst [j0] x0:(MOVWload [i0] {s} p mem))
	y))
  && i1 == i0+2
  && j1 == j0+16
  && j0 % 32 == 0
  && x0.Uses == 1
  && x1.Uses == 1
  && s0.Uses == 1
  && s1.Uses == 1
  && or.Uses == 1
  && mergePoint(b,x0,x1) != nil
  && clobber(x0)
  && clobber(x1)
  && clobber(s0)
  && clobber(s1)
  && clobber(or)
  -> @mergePoint(b,x0,x1) (ORQ <v.Type> (SHLQconst <v.Type> [j0] (MOVLload [i0] {s} p mem)) y)

// Little-endian indexed loads

(ORL                  x0:(MOVBloadidx1 [i0] {s} p idx mem)
    sh:(SHLLconst [8] x1:(MOVBloadidx1 [i1] {s} p idx mem)))
  && i1 == i0+1
  && x0.Uses == 1
  && x1.Uses == 1
  && sh.Uses == 1
  && mergePoint(b,x0,x1) != nil
  && clobber(x0)
  && clobber(x1)
  && clobber(sh)
  -> @mergePoint(b,x0,x1) (MOVWloadidx1 <v.Type> [i0] {s} p idx mem)

(ORQ                  x0:(MOVBloadidx1 [i0] {s} p idx mem)
    sh:(SHLQconst [8] x1:(MOVBloadidx1 [i1] {s} p idx mem)))
  && i1 == i0+1
  && x0.Uses == 1
  && x1.Uses == 1
  && sh.Uses == 1
  && mergePoint(b,x0,x1) != nil
  && clobber(x0)
  && clobber(x1)
  && clobber(sh)
  -> @mergePoint(b,x0,x1) (MOVWloadidx1 <v.Type> [i0] {s} p idx mem)

(ORL                   x0:(MOVWloadidx1 [i0] {s} p idx mem)
    sh:(SHLLconst [16] x1:(MOVWloadidx1 [i1] {s} p idx mem)))
  && i1 == i0+2
  && x0.Uses == 1
  && x1.Uses == 1
  && sh.Uses == 1
  && mergePoint(b,x0,x1) != nil
  && clobber(x0)
  && clobber(x1)
  && clobber(sh)
  -> @mergePoint(b,x0,x1) (MOVLloadidx1 [i0] {s} p idx mem)

(ORQ                   x0:(MOVWloadidx1 [i0] {s} p idx mem)
    sh:(SHLQconst [16] x1:(MOVWloadidx1 [i1] {s} p idx mem)))
  && i1 == i0+2
  && x0.Uses == 1
  && x1.Uses == 1
  && sh.Uses == 1
  && mergePoint(b,x0,x1) != nil
  && clobber(x0)
  && clobber(x1)
  && clobber(sh)
  -> @mergePoint(b,x0,x1) (MOVLloadidx1 [i0] {s} p idx mem)

(ORQ                   x0:(MOVLloadidx1 [i0] {s} p idx mem)
    sh:(SHLQconst [32] x1:(MOVLloadidx1 [i1] {s} p idx mem)))
  && i1 == i0+4
  && x0.Uses == 1
  && x1.Uses == 1
  && sh.Uses == 1
  && mergePoint(b,x0,x1) != nil
  && clobber(x0)
  && clobber(x1)
  && clobber(sh)
  -> @mergePoint(b,x0,x1) (MOVQloadidx1 [i0] {s} p idx mem)

(ORL
    s1:(SHLLconst [j1] x1:(MOVBloadidx1 [i1] {s} p idx mem))
    or:(ORL
        s0:(SHLLconst [j0] x0:(MOVBloadidx1 [i0] {s} p idx mem))
	y))
  && i1 == i0+1
  && j1 == j0+8
  && j0 % 16 == 0
  && x0.Uses == 1
  && x1.Uses == 1
  && s0.Uses == 1
  && s1.Uses == 1
  && or.Uses == 1
  && mergePoint(b,x0,x1) != nil
  && clobber(x0)
  && clobber(x1)
  && clobber(s0)
  && clobber(s1)
  && clobber(or)
  -> @mergePoint(b,x0,x1) (ORL <v.Type> (SHLLconst <v.Type> [j0] (MOVWloadidx1 [i0] {s} p idx mem)) y)

(ORQ
    s1:(SHLQconst [j1] x1:(MOVBloadidx1 [i1] {s} p idx mem))
    or:(ORQ
        s0:(SHLQconst [j0] x0:(MOVBloadidx1 [i0] {s} p idx mem))
	y))
  && i1 == i0+1
  && j1 == j0+8
  && j0 % 16 == 0
  && x0.Uses == 1
  && x1.Uses == 1
  && s0.Uses == 1
  && s1.Uses == 1
  && or.Uses == 1
  && mergePoint(b,x0,x1) != nil
  && clobber(x0)
  && clobber(x1)
  && clobber(s0)
  && clobber(s1)
  && clobber(or)
  -> @mergePoint(b,x0,x1) (ORQ <v.Type> (SHLQconst <v.Type> [j0] (MOVWloadidx1 [i0] {s} p idx mem)) y)

(ORQ
    s1:(SHLQconst [j1] x1:(MOVWloadidx1 [i1] {s} p idx mem))
    or:(ORQ
        s0:(SHLQconst [j0] x0:(MOVWloadidx1 [i0] {s} p idx mem))
	y))
  && i1 == i0+2
  && j1 == j0+16
  && j0 % 32 == 0
  && x0.Uses == 1
  && x1.Uses == 1
  && s0.Uses == 1
  && s1.Uses == 1
  && or.Uses == 1
  && mergePoint(b,x0,x1) != nil
  && clobber(x0)
  && clobber(x1)
  && clobber(s0)
  && clobber(s1)
  && clobber(or)
  -> @mergePoint(b,x0,x1) (ORQ <v.Type> (SHLQconst <v.Type> [j0] (MOVLloadidx1 [i0] {s} p idx mem)) y)

// Big-endian loads

(ORL
                       x1:(MOVBload [i1] {s} p mem)
    sh:(SHLLconst [8]  x0:(MOVBload [i0] {s} p mem)))
  && i1 == i0+1
  && x0.Uses == 1
  && x1.Uses == 1
  && sh.Uses == 1
  && mergePoint(b,x0,x1) != nil
  && clobber(x0)
  && clobber(x1)
  && clobber(sh)
  -> @mergePoint(b,x0,x1) (ROLWconst <v.Type> [8] (MOVWload [i0] {s} p mem))

(ORQ
                       x1:(MOVBload [i1] {s} p mem)
    sh:(SHLQconst [8]  x0:(MOVBload [i0] {s} p mem)))
  && i1 == i0+1
  && x0.Uses == 1
  && x1.Uses == 1
  && sh.Uses == 1
  && mergePoint(b,x0,x1) != nil
  && clobber(x0)
  && clobber(x1)
  && clobber(sh)
  -> @mergePoint(b,x0,x1) (ROLWconst <v.Type> [8] (MOVWload [i0] {s} p mem))

(ORL
                        r1:(ROLWconst [8] x1:(MOVWload [i1] {s} p mem))
    sh:(SHLLconst [16]  r0:(ROLWconst [8] x0:(MOVWload [i0] {s} p mem))))
  && i1 == i0+2
  && x0.Uses == 1
  && x1.Uses == 1
  && r0.Uses == 1
  && r1.Uses == 1
  && sh.Uses == 1
  && mergePoint(b,x0,x1) != nil
  && clobber(x0)
  && clobber(x1)
  && clobber(r0)
  && clobber(r1)
  && clobber(sh)
  -> @mergePoint(b,x0,x1) (BSWAPL <v.Type> (MOVLload [i0] {s} p mem))

(ORQ
                        r1:(ROLWconst [8] x1:(MOVWload [i1] {s} p mem))
    sh:(SHLQconst [16]  r0:(ROLWconst [8] x0:(MOVWload [i0] {s} p mem))))
  && i1 == i0+2
  && x0.Uses == 1
  && x1.Uses == 1
  && r0.Uses == 1
  && r1.Uses == 1
  && sh.Uses == 1
  && mergePoint(b,x0,x1) != nil
  && clobber(x0)
  && clobber(x1)
  && clobber(r0)
  && clobber(r1)
  && clobber(sh)
  -> @mergePoint(b,x0,x1) (BSWAPL <v.Type> (MOVLload [i0] {s} p mem))

(ORQ
                        r1:(BSWAPL x1:(MOVLload [i1] {s} p mem))
    sh:(SHLQconst [32]  r0:(BSWAPL x0:(MOVLload [i0] {s} p mem))))
  && i1 == i0+4
  && x0.Uses == 1
  && x1.Uses == 1
  && r0.Uses == 1
  && r1.Uses == 1
  && sh.Uses == 1
  && mergePoint(b,x0,x1) != nil
  && clobber(x0)
  && clobber(x1)
  && clobber(r0)
  && clobber(r1)
  && clobber(sh)
  -> @mergePoint(b,x0,x1) (BSWAPQ <v.Type> (MOVQload [i0] {s} p mem))

(ORL
    s0:(SHLLconst [j0] x0:(MOVBload [i0] {s} p mem))
    or:(ORL
        s1:(SHLLconst [j1] x1:(MOVBload [i1] {s} p mem))
	y))
  && i1 == i0+1
  && j1 == j0-8
  && j1 % 16 == 0
  && x0.Uses == 1
  && x1.Uses == 1
  && s0.Uses == 1
  && s1.Uses == 1
  && or.Uses == 1
  && mergePoint(b,x0,x1) != nil
  && clobber(x0)
  && clobber(x1)
  && clobber(s0)
  && clobber(s1)
  && clobber(or)
  -> @mergePoint(b,x0,x1) (ORL <v.Type> (SHLLconst <v.Type> [j1] (ROLWconst <typ.UInt16> [8] (MOVWload [i0] {s} p mem))) y)

(ORQ
    s0:(SHLQconst [j0] x0:(MOVBload [i0] {s} p mem))
    or:(ORQ
        s1:(SHLQconst [j1] x1:(MOVBload [i1] {s} p mem))
	y))
  && i1 == i0+1
  && j1 == j0-8
  && j1 % 16 == 0
  && x0.Uses == 1
  && x1.Uses == 1
  && s0.Uses == 1
  && s1.Uses == 1
  && or.Uses == 1
  && mergePoint(b,x0,x1) != nil
  && clobber(x0)
  && clobber(x1)
  && clobber(s0)
  && clobber(s1)
  && clobber(or)
  -> @mergePoint(b,x0,x1) (ORQ <v.Type> (SHLQconst <v.Type> [j1] (ROLWconst <typ.UInt16> [8] (MOVWload [i0] {s} p mem))) y)

(ORQ
    s0:(SHLQconst [j0] r0:(ROLWconst [8] x0:(MOVWload [i0] {s} p mem)))
    or:(ORQ
        s1:(SHLQconst [j1] r1:(ROLWconst [8] x1:(MOVWload [i1] {s} p mem)))
	y))
  && i1 == i0+2
  && j1 == j0-16
  && j1 % 32 == 0
  && x0.Uses == 1
  && x1.Uses == 1
  && r0.Uses == 1
  && r1.Uses == 1
  && s0.Uses == 1
  && s1.Uses == 1
  && or.Uses == 1
  && mergePoint(b,x0,x1) != nil
  && clobber(x0)
  && clobber(x1)
  && clobber(r0)
  && clobber(r1)
  && clobber(s0)
  && clobber(s1)
  && clobber(or)
  -> @mergePoint(b,x0,x1) (ORQ <v.Type> (SHLQconst <v.Type> [j1] (BSWAPL <typ.UInt32> (MOVLload [i0] {s} p mem))) y)

// Big-endian indexed loads

(ORL
                       x1:(MOVBloadidx1 [i1] {s} p idx mem)
    sh:(SHLLconst [8]  x0:(MOVBloadidx1 [i0] {s} p idx mem)))
  && i1 == i0+1
  && x0.Uses == 1
  && x1.Uses == 1
  && sh.Uses == 1
  && mergePoint(b,x0,x1) != nil
  && clobber(x0)
  && clobber(x1)
  && clobber(sh)
  -> @mergePoint(b,x0,x1) (ROLWconst <v.Type> [8] (MOVWloadidx1 [i0] {s} p idx mem))

(ORQ
                       x1:(MOVBloadidx1 [i1] {s} p idx mem)
    sh:(SHLQconst [8]  x0:(MOVBloadidx1 [i0] {s} p idx mem)))
  && i1 == i0+1
  && x0.Uses == 1
  && x1.Uses == 1
  && sh.Uses == 1
  && mergePoint(b,x0,x1) != nil
  && clobber(x0)
  && clobber(x1)
  && clobber(sh)
  -> @mergePoint(b,x0,x1) (ROLWconst <v.Type> [8] (MOVWloadidx1 [i0] {s} p idx mem))

(ORL
                        r1:(ROLWconst [8] x1:(MOVWloadidx1 [i1] {s} p idx mem))
    sh:(SHLLconst [16]  r0:(ROLWconst [8] x0:(MOVWloadidx1 [i0] {s} p idx mem))))
  && i1 == i0+2
  && x0.Uses == 1
  && x1.Uses == 1
  && r0.Uses == 1
  && r1.Uses == 1
  && sh.Uses == 1
  && mergePoint(b,x0,x1) != nil
  && clobber(x0)
  && clobber(x1)
  && clobber(r0)
  && clobber(r1)
  && clobber(sh)
  -> @mergePoint(b,x0,x1) (BSWAPL <v.Type> (MOVLloadidx1 [i0] {s} p idx mem))

(ORQ
                        r1:(ROLWconst [8] x1:(MOVWloadidx1 [i1] {s} p idx mem))
    sh:(SHLQconst [16]  r0:(ROLWconst [8] x0:(MOVWloadidx1 [i0] {s} p idx mem))))
  && i1 == i0+2
  && x0.Uses == 1
  && x1.Uses == 1
  && r0.Uses == 1
  && r1.Uses == 1
  && sh.Uses == 1
  && mergePoint(b,x0,x1) != nil
  && clobber(x0)
  && clobber(x1)
  && clobber(r0)
  && clobber(r1)
  && clobber(sh)
  -> @mergePoint(b,x0,x1) (BSWAPL <v.Type> (MOVLloadidx1 [i0] {s} p idx mem))

(ORQ
                        r1:(BSWAPL x1:(MOVLloadidx1 [i1] {s} p idx mem))
    sh:(SHLQconst [32]  r0:(BSWAPL x0:(MOVLloadidx1 [i0] {s} p idx mem))))
  && i1 == i0+4
  && x0.Uses == 1
  && x1.Uses == 1
  && r0.Uses == 1
  && r1.Uses == 1
  && sh.Uses == 1
  && mergePoint(b,x0,x1) != nil
  && clobber(x0)
  && clobber(x1)
  && clobber(r0)
  && clobber(r1)
  && clobber(sh)
  -> @mergePoint(b,x0,x1) (BSWAPQ <v.Type> (MOVQloadidx1 [i0] {s} p idx mem))

(ORL
    s0:(SHLLconst [j0] x0:(MOVBloadidx1 [i0] {s} p idx mem))
    or:(ORL
        s1:(SHLLconst [j1] x1:(MOVBloadidx1 [i1] {s} p idx mem))
	y))
  && i1 == i0+1
  && j1 == j0-8
  && j1 % 16 == 0
  && x0.Uses == 1
  && x1.Uses == 1
  && s0.Uses == 1
  && s1.Uses == 1
  && or.Uses == 1
  && mergePoint(b,x0,x1) != nil
  && clobber(x0)
  && clobber(x1)
  && clobber(s0)
  && clobber(s1)
  && clobber(or)
  -> @mergePoint(b,x0,x1) (ORL <v.Type> (SHLLconst <v.Type> [j1] (ROLWconst <typ.UInt16> [8] (MOVWloadidx1 [i0] {s} p idx mem))) y)

(ORQ
    s0:(SHLQconst [j0] x0:(MOVBloadidx1 [i0] {s} p idx mem))
    or:(ORQ
        s1:(SHLQconst [j1] x1:(MOVBloadidx1 [i1] {s} p idx mem))
	y))
  && i1 == i0+1
  && j1 == j0-8
  && j1 % 16 == 0
  && x0.Uses == 1
  && x1.Uses == 1
  && s0.Uses == 1
  && s1.Uses == 1
  && or.Uses == 1
  && mergePoint(b,x0,x1) != nil
  && clobber(x0)
  && clobber(x1)
  && clobber(s0)
  && clobber(s1)
  && clobber(or)
  -> @mergePoint(b,x0,x1) (ORQ <v.Type> (SHLQconst <v.Type> [j1] (ROLWconst <typ.UInt16> [8] (MOVWloadidx1 [i0] {s} p idx mem))) y)

(ORQ
    s0:(SHLQconst [j0] r0:(ROLWconst [8] x0:(MOVWloadidx1 [i0] {s} p idx mem)))
    or:(ORQ
        s1:(SHLQconst [j1] r1:(ROLWconst [8] x1:(MOVWloadidx1 [i1] {s} p idx mem)))
	y))
  && i1 == i0+2
  && j1 == j0-16
  && j1 % 32 == 0
  && x0.Uses == 1
  && x1.Uses == 1
  && r0.Uses == 1
  && r1.Uses == 1
  && s0.Uses == 1
  && s1.Uses == 1
  && or.Uses == 1
  && mergePoint(b,x0,x1) != nil
  && clobber(x0)
  && clobber(x1)
  && clobber(r0)
  && clobber(r1)
  && clobber(s0)
  && clobber(s1)
  && clobber(or)
  -> @mergePoint(b,x0,x1) (ORQ <v.Type> (SHLQconst <v.Type> [j1] (BSWAPL <typ.UInt32> (MOVLloadidx1 [i0] {s} p idx mem))) y)

// Combine 2 byte stores + shift into rolw 8 + word store
(MOVBstore [i] {s} p w
  x0:(MOVBstore [i-1] {s} p (SHRWconst [8] w) mem))
  && x0.Uses == 1
  && clobber(x0)
  -> (MOVWstore [i-1] {s} p (ROLWconst <w.Type> [8] w) mem)

(MOVBstoreidx1 [i] {s} p idx w
  x0:(MOVBstoreidx1 [i-1] {s} p idx (SHRWconst [8] w) mem))
  && x0.Uses == 1
  && clobber(x0)
  -> (MOVWstoreidx1 [i-1] {s} p idx (ROLWconst <w.Type> [8] w) mem)

// Combine stores + shifts into bswap and larger (unaligned) stores
(MOVBstore [i] {s} p w
  x2:(MOVBstore [i-1] {s} p (SHRLconst [8] w)
  x1:(MOVBstore [i-2] {s} p (SHRLconst [16] w)
  x0:(MOVBstore [i-3] {s} p (SHRLconst [24] w) mem))))
  && x0.Uses == 1
  && x1.Uses == 1
  && x2.Uses == 1
  && clobber(x0)
  && clobber(x1)
  && clobber(x2)
  -> (MOVLstore [i-3] {s} p (BSWAPL <w.Type> w) mem)

(MOVBstoreidx1 [i] {s} p idx w
  x2:(MOVBstoreidx1 [i-1] {s} p idx (SHRLconst [8] w)
  x1:(MOVBstoreidx1 [i-2] {s} p idx (SHRLconst [16] w)
  x0:(MOVBstoreidx1 [i-3] {s} p idx (SHRLconst [24] w) mem))))
  && x0.Uses == 1
  && x1.Uses == 1
  && x2.Uses == 1
  && clobber(x0)
  && clobber(x1)
  && clobber(x2)
  -> (MOVLstoreidx1 [i-3] {s} p idx (BSWAPL <w.Type> w) mem)

(MOVBstore [i] {s} p w
  x6:(MOVBstore [i-1] {s} p (SHRQconst [8] w)
  x5:(MOVBstore [i-2] {s} p (SHRQconst [16] w)
  x4:(MOVBstore [i-3] {s} p (SHRQconst [24] w)
  x3:(MOVBstore [i-4] {s} p (SHRQconst [32] w)
  x2:(MOVBstore [i-5] {s} p (SHRQconst [40] w)
  x1:(MOVBstore [i-6] {s} p (SHRQconst [48] w)
  x0:(MOVBstore [i-7] {s} p (SHRQconst [56] w) mem))))))))
  && x0.Uses == 1
  && x1.Uses == 1
  && x2.Uses == 1
  && x3.Uses == 1
  && x4.Uses == 1
  && x5.Uses == 1
  && x6.Uses == 1
  && clobber(x0)
  && clobber(x1)
  && clobber(x2)
  && clobber(x3)
  && clobber(x4)
  && clobber(x5)
  && clobber(x6)
  -> (MOVQstore [i-7] {s} p (BSWAPQ <w.Type> w) mem)

(MOVBstoreidx1 [i] {s} p idx w
  x6:(MOVBstoreidx1 [i-1] {s} p idx (SHRQconst [8] w)
  x5:(MOVBstoreidx1 [i-2] {s} p idx (SHRQconst [16] w)
  x4:(MOVBstoreidx1 [i-3] {s} p idx (SHRQconst [24] w)
  x3:(MOVBstoreidx1 [i-4] {s} p idx (SHRQconst [32] w)
  x2:(MOVBstoreidx1 [i-5] {s} p idx (SHRQconst [40] w)
  x1:(MOVBstoreidx1 [i-6] {s} p idx (SHRQconst [48] w)
  x0:(MOVBstoreidx1 [i-7] {s} p idx (SHRQconst [56] w) mem))))))))
  && x0.Uses == 1
  && x1.Uses == 1
  && x2.Uses == 1
  && x3.Uses == 1
  && x4.Uses == 1
  && x5.Uses == 1
  && x6.Uses == 1
  && clobber(x0)
  && clobber(x1)
  && clobber(x2)
  && clobber(x3)
  && clobber(x4)
  && clobber(x5)
  && clobber(x6)
  -> (MOVQstoreidx1 [i-7] {s} p idx (BSWAPQ <w.Type> w) mem)

// Combine constant stores into larger (unaligned) stores.
(MOVBstoreconst [c] {s} p x:(MOVBstoreconst [a] {s} p mem))
  && x.Uses == 1
  && ValAndOff(a).Off() + 1 == ValAndOff(c).Off()
  && clobber(x)
  -> (MOVWstoreconst [makeValAndOff(ValAndOff(a).Val()&0xff | ValAndOff(c).Val()<<8, ValAndOff(a).Off())] {s} p mem)
(MOVWstoreconst [c] {s} p x:(MOVWstoreconst [a] {s} p mem))
  && x.Uses == 1
  && ValAndOff(a).Off() + 2 == ValAndOff(c).Off()
  && clobber(x)
  -> (MOVLstoreconst [makeValAndOff(ValAndOff(a).Val()&0xffff | ValAndOff(c).Val()<<16, ValAndOff(a).Off())] {s} p mem)
(MOVLstoreconst [c] {s} p x:(MOVLstoreconst [a] {s} p mem))
  && x.Uses == 1
  && ValAndOff(a).Off() + 4 == ValAndOff(c).Off()
  && clobber(x)
  -> (MOVQstore [ValAndOff(a).Off()] {s} p (MOVQconst [ValAndOff(a).Val()&0xffffffff | ValAndOff(c).Val()<<32]) mem)
(MOVQstoreconst [c] {s} p x:(MOVQstoreconst [c2] {s} p mem))
  && config.useSSE
  && x.Uses == 1
  && ValAndOff(c2).Off() + 8 == ValAndOff(c).Off()
  && ValAndOff(c).Val() == 0
  && ValAndOff(c2).Val() == 0
  && clobber(x)
  -> (MOVOstore [ValAndOff(c2).Off()] {s} p (MOVOconst [0]) mem)

(MOVBstoreconstidx1 [c] {s} p i x:(MOVBstoreconstidx1 [a] {s} p i mem))
  && x.Uses == 1
  && ValAndOff(a).Off() + 1 == ValAndOff(c).Off()
  && clobber(x)
  -> (MOVWstoreconstidx1 [makeValAndOff(ValAndOff(a).Val()&0xff | ValAndOff(c).Val()<<8, ValAndOff(a).Off())] {s} p i mem)
(MOVWstoreconstidx1 [c] {s} p i x:(MOVWstoreconstidx1 [a] {s} p i mem))
  && x.Uses == 1
  && ValAndOff(a).Off() + 2 == ValAndOff(c).Off()
  && clobber(x)
  -> (MOVLstoreconstidx1 [makeValAndOff(ValAndOff(a).Val()&0xffff | ValAndOff(c).Val()<<16, ValAndOff(a).Off())] {s} p i mem)
(MOVLstoreconstidx1 [c] {s} p i x:(MOVLstoreconstidx1 [a] {s} p i mem))
  && x.Uses == 1
  && ValAndOff(a).Off() + 4 == ValAndOff(c).Off()
  && clobber(x)
  -> (MOVQstoreidx1 [ValAndOff(a).Off()] {s} p i (MOVQconst [ValAndOff(a).Val()&0xffffffff | ValAndOff(c).Val()<<32]) mem)

(MOVWstoreconstidx2 [c] {s} p i x:(MOVWstoreconstidx2 [a] {s} p i mem))
  && x.Uses == 1
  && ValAndOff(a).Off() + 2 == ValAndOff(c).Off()
  && clobber(x)
  -> (MOVLstoreconstidx1 [makeValAndOff(ValAndOff(a).Val()&0xffff | ValAndOff(c).Val()<<16, ValAndOff(a).Off())] {s} p (SHLQconst <i.Type> [1] i) mem)
(MOVLstoreconstidx4 [c] {s} p i x:(MOVLstoreconstidx4 [a] {s} p i mem))
  && x.Uses == 1
  && ValAndOff(a).Off() + 4 == ValAndOff(c).Off()
  && clobber(x)
  -> (MOVQstoreidx1 [ValAndOff(a).Off()] {s} p (SHLQconst <i.Type> [2] i) (MOVQconst [ValAndOff(a).Val()&0xffffffff | ValAndOff(c).Val()<<32]) mem)

// Combine stores into larger (unaligned) stores.
(MOVBstore [i] {s} p (SHRQconst [8] w) x:(MOVBstore [i-1] {s} p w mem))
  && x.Uses == 1
  && clobber(x)
  -> (MOVWstore [i-1] {s} p w mem)
(MOVBstore [i] {s} p (SHRQconst [j] w) x:(MOVBstore [i-1] {s} p w0:(SHRQconst [j-8] w) mem))
  && x.Uses == 1
  && clobber(x)
  -> (MOVWstore [i-1] {s} p w0 mem)
(MOVWstore [i] {s} p (SHRQconst [16] w) x:(MOVWstore [i-2] {s} p w mem))
  && x.Uses == 1
  && clobber(x)
  -> (MOVLstore [i-2] {s} p w mem)
(MOVWstore [i] {s} p (SHRQconst [j] w) x:(MOVWstore [i-2] {s} p w0:(SHRQconst [j-16] w) mem))
  && x.Uses == 1
  && clobber(x)
  -> (MOVLstore [i-2] {s} p w0 mem)
(MOVLstore [i] {s} p (SHRQconst [32] w) x:(MOVLstore [i-4] {s} p w mem))
  && x.Uses == 1
  && clobber(x)
  -> (MOVQstore [i-4] {s} p w mem)
(MOVLstore [i] {s} p (SHRQconst [j] w) x:(MOVLstore [i-4] {s} p w0:(SHRQconst [j-32] w) mem))
  && x.Uses == 1
  && clobber(x)
  -> (MOVQstore [i-4] {s} p w0 mem)

(MOVBstoreidx1 [i] {s} p idx (SHRQconst [8] w) x:(MOVBstoreidx1 [i-1] {s} p idx w mem))
  && x.Uses == 1
  && clobber(x)
  -> (MOVWstoreidx1 [i-1] {s} p idx w mem)
(MOVBstoreidx1 [i] {s} p idx (SHRQconst [j] w) x:(MOVBstoreidx1 [i-1] {s} p idx w0:(SHRQconst [j-8] w) mem))
  && x.Uses == 1
  && clobber(x)
  -> (MOVWstoreidx1 [i-1] {s} p idx w0 mem)
(MOVWstoreidx1 [i] {s} p idx (SHRQconst [16] w) x:(MOVWstoreidx1 [i-2] {s} p idx w mem))
  && x.Uses == 1
  && clobber(x)
  -> (MOVLstoreidx1 [i-2] {s} p idx w mem)
(MOVWstoreidx1 [i] {s} p idx (SHRQconst [j] w) x:(MOVWstoreidx1 [i-2] {s} p idx w0:(SHRQconst [j-16] w) mem))
  && x.Uses == 1
  && clobber(x)
  -> (MOVLstoreidx1 [i-2] {s} p idx w0 mem)
(MOVLstoreidx1 [i] {s} p idx (SHRQconst [32] w) x:(MOVLstoreidx1 [i-4] {s} p idx w mem))
  && x.Uses == 1
  && clobber(x)
  -> (MOVQstoreidx1 [i-4] {s} p idx w mem)
(MOVLstoreidx1 [i] {s} p idx (SHRQconst [j] w) x:(MOVLstoreidx1 [i-4] {s} p idx w0:(SHRQconst [j-32] w) mem))
  && x.Uses == 1
  && clobber(x)
  -> (MOVQstoreidx1 [i-4] {s} p idx w0 mem)

(MOVWstoreidx2 [i] {s} p idx (SHRQconst [16] w) x:(MOVWstoreidx2 [i-2] {s} p idx w mem))
  && x.Uses == 1
  && clobber(x)
  -> (MOVLstoreidx1 [i-2] {s} p (SHLQconst <idx.Type> [1] idx) w mem)
(MOVWstoreidx2 [i] {s} p idx (SHRQconst [j] w) x:(MOVWstoreidx2 [i-2] {s} p idx w0:(SHRQconst [j-16] w) mem))
  && x.Uses == 1
  && clobber(x)
  -> (MOVLstoreidx1 [i-2] {s} p (SHLQconst <idx.Type> [1] idx) w0 mem)
(MOVLstoreidx4 [i] {s} p idx (SHRQconst [32] w) x:(MOVLstoreidx4 [i-4] {s} p idx w mem))
  && x.Uses == 1
  && clobber(x)
  -> (MOVQstoreidx1 [i-4] {s} p (SHLQconst <idx.Type> [2] idx) w mem)
(MOVLstoreidx4 [i] {s} p idx (SHRQconst [j] w) x:(MOVLstoreidx4 [i-4] {s} p idx w0:(SHRQconst [j-32] w) mem))
  && x.Uses == 1
  && clobber(x)
  -> (MOVQstoreidx1 [i-4] {s} p (SHLQconst <idx.Type> [2] idx) w0 mem)

(MOVBstore [i] {s} p
  x1:(MOVBload [j] {s2} p2 mem)
    mem2:(MOVBstore [i-1] {s} p
      x2:(MOVBload [j-1] {s2} p2 mem) mem))
  && x1.Uses == 1
  && x2.Uses == 1
  && mem2.Uses == 1
  && clobber(x1)
  && clobber(x2)
  && clobber(mem2)
  -> (MOVWstore [i-1] {s} p (MOVWload [j-1] {s2} p2 mem) mem)

(MOVWstore [i] {s} p
  x1:(MOVWload [j] {s2} p2 mem)
    mem2:(MOVWstore [i-2] {s} p
      x2:(MOVWload [j-2] {s2} p2 mem) mem))
  && x1.Uses == 1
  && x2.Uses == 1
  && mem2.Uses == 1
  && clobber(x1)
  && clobber(x2)
  && clobber(mem2)
  -> (MOVLstore [i-2] {s} p (MOVLload [j-2] {s2} p2 mem) mem)

(MOVLstore [i] {s} p
  x1:(MOVLload [j] {s2} p2 mem)
    mem2:(MOVLstore [i-4] {s} p
      x2:(MOVLload [j-4] {s2} p2 mem) mem))
  && x1.Uses == 1
  && x2.Uses == 1
  && mem2.Uses == 1
  && clobber(x1)
  && clobber(x2)
  && clobber(mem2)
  -> (MOVQstore [i-4] {s} p (MOVQload [j-4] {s2} p2 mem) mem)

// This is somewhat tricky. There may be pointers in SSE registers due to rule below.
// However those register shouldn't live across GC safepoint.
(MOVQstore [i] {s} p
  x1:(MOVQload [j] {s2} p2 mem)
    mem2:(MOVQstore [i-8] {s} p
      x2:(MOVQload [j-8] {s2} p2 mem) mem))
  && x1.Uses == 1
  && x2.Uses == 1
  && mem2.Uses == 1
  && config.useSSE
  && clobber(x1)
  && clobber(x2)
  && clobber(mem2)
  -> (MOVOstore [i-8] {s} p (MOVOload [j-8] {s2} p2 mem) mem)


// amd64p32 rules
// same as the rules above, but with 32 instead of 64 bit pointer arithmetic.
// LEAQ,ADDQ -> LEAL,ADDL
(ADDLconst [c] (LEAL [d] {s} x)) && is32Bit(c+d) -> (LEAL [c+d] {s} x)
(LEAL [c] {s} (ADDLconst [d] x)) && is32Bit(c+d) -> (LEAL [c+d] {s} x)

(MOVQload  [off1] {sym1} (LEAL [off2] {sym2} base) mem) && canMergeSym(sym1, sym2) && is32Bit(off1+off2) ->
	(MOVQload  [off1+off2] {mergeSym(sym1,sym2)} base mem)
(MOVLload  [off1] {sym1} (LEAL [off2] {sym2} base) mem) && canMergeSym(sym1, sym2) && is32Bit(off1+off2) ->
	(MOVLload  [off1+off2] {mergeSym(sym1,sym2)} base mem)
(MOVWload  [off1] {sym1} (LEAL [off2] {sym2} base) mem) && canMergeSym(sym1, sym2) && is32Bit(off1+off2) ->
	(MOVWload  [off1+off2] {mergeSym(sym1,sym2)} base mem)
(MOVBload  [off1] {sym1} (LEAL [off2] {sym2} base) mem) && canMergeSym(sym1, sym2) && is32Bit(off1+off2) ->
	(MOVBload  [off1+off2] {mergeSym(sym1,sym2)} base mem)

(MOVQstore  [off1] {sym1} (LEAL [off2] {sym2} base) val mem) && canMergeSym(sym1, sym2) && is32Bit(off1+off2) ->
	(MOVQstore  [off1+off2] {mergeSym(sym1,sym2)} base val mem)
(MOVLstore  [off1] {sym1} (LEAL [off2] {sym2} base) val mem) && canMergeSym(sym1, sym2) && is32Bit(off1+off2) ->
	(MOVLstore  [off1+off2] {mergeSym(sym1,sym2)} base val mem)
(MOVWstore  [off1] {sym1} (LEAL [off2] {sym2} base) val mem) && canMergeSym(sym1, sym2) && is32Bit(off1+off2) ->
	(MOVWstore  [off1+off2] {mergeSym(sym1,sym2)} base val mem)
(MOVBstore  [off1] {sym1} (LEAL [off2] {sym2} base) val mem) && canMergeSym(sym1, sym2) && is32Bit(off1+off2) ->
	(MOVBstore  [off1+off2] {mergeSym(sym1,sym2)} base val mem)

(MOVQstoreconst [sc] {sym1} (LEAL [off] {sym2} ptr) mem) && canMergeSym(sym1, sym2) && ValAndOff(sc).canAdd(off) ->
	(MOVQstoreconst [ValAndOff(sc).add(off)] {mergeSym(sym1, sym2)} ptr mem)
(MOVLstoreconst [sc] {sym1} (LEAL [off] {sym2} ptr) mem) && canMergeSym(sym1, sym2) && ValAndOff(sc).canAdd(off) ->
	(MOVLstoreconst [ValAndOff(sc).add(off)] {mergeSym(sym1, sym2)} ptr mem)
(MOVWstoreconst [sc] {sym1} (LEAL [off] {sym2} ptr) mem) && canMergeSym(sym1, sym2) && ValAndOff(sc).canAdd(off) ->
	(MOVWstoreconst [ValAndOff(sc).add(off)] {mergeSym(sym1, sym2)} ptr mem)
(MOVBstoreconst [sc] {sym1} (LEAL [off] {sym2} ptr) mem) && canMergeSym(sym1, sym2) && ValAndOff(sc).canAdd(off) ->
	(MOVBstoreconst [ValAndOff(sc).add(off)] {mergeSym(sym1, sym2)} ptr mem)

(MOVQload  [off1] {sym} (ADDLconst [off2] ptr) mem) && is32Bit(off1+off2) -> (MOVQload  [off1+off2] {sym} ptr mem)
(MOVLload  [off1] {sym} (ADDLconst [off2] ptr) mem) && is32Bit(off1+off2) -> (MOVLload  [off1+off2] {sym} ptr mem)
(MOVWload  [off1] {sym} (ADDLconst [off2] ptr) mem) && is32Bit(off1+off2) -> (MOVWload  [off1+off2] {sym} ptr mem)
(MOVBload  [off1] {sym} (ADDLconst [off2] ptr) mem) && is32Bit(off1+off2) -> (MOVBload  [off1+off2] {sym} ptr mem)
(MOVQstore  [off1] {sym} (ADDLconst [off2] ptr) val mem) && is32Bit(off1+off2) -> (MOVQstore  [off1+off2] {sym} ptr val mem)
(MOVLstore  [off1] {sym} (ADDLconst [off2] ptr) val mem) && is32Bit(off1+off2) -> (MOVLstore  [off1+off2] {sym} ptr val mem)
(MOVWstore  [off1] {sym} (ADDLconst [off2] ptr) val mem) && is32Bit(off1+off2) -> (MOVWstore  [off1+off2] {sym} ptr val mem)
(MOVBstore  [off1] {sym} (ADDLconst [off2] ptr) val mem) && is32Bit(off1+off2) -> (MOVBstore  [off1+off2] {sym} ptr val mem)
(MOVQstoreconst [sc] {s} (ADDLconst [off] ptr) mem) && ValAndOff(sc).canAdd(off) ->
	(MOVQstoreconst [ValAndOff(sc).add(off)] {s} ptr mem)
(MOVLstoreconst [sc] {s} (ADDLconst [off] ptr) mem) && ValAndOff(sc).canAdd(off) ->
	(MOVLstoreconst [ValAndOff(sc).add(off)] {s} ptr mem)
(MOVWstoreconst [sc] {s} (ADDLconst [off] ptr) mem) && ValAndOff(sc).canAdd(off) ->
	(MOVWstoreconst [ValAndOff(sc).add(off)] {s} ptr mem)
(MOVBstoreconst [sc] {s} (ADDLconst [off] ptr) mem) && ValAndOff(sc).canAdd(off) ->
	(MOVBstoreconst [ValAndOff(sc).add(off)] {s} ptr mem)

// Merge load and op
// TODO: add indexed variants?
(ADDQ x l:(MOVQload [off] {sym} ptr mem)) && canMergeLoad(v, l, x) && clobber(l) -> (ADDQmem x [off] {sym} ptr mem)
(ADDL x l:(MOVLload [off] {sym} ptr mem)) && canMergeLoad(v, l, x) && clobber(l) -> (ADDLmem x [off] {sym} ptr mem)
(SUBQ x l:(MOVQload [off] {sym} ptr mem)) && canMergeLoad(v, l, x) && clobber(l) -> (SUBQmem x [off] {sym} ptr mem)
(SUBL x l:(MOVLload [off] {sym} ptr mem)) && canMergeLoad(v, l, x) && clobber(l) -> (SUBLmem x [off] {sym} ptr mem)
(ANDQ x l:(MOVQload [off] {sym} ptr mem)) && canMergeLoad(v, l, x) && clobber(l) -> (ANDQmem x [off] {sym} ptr mem)
(ANDL x l:(MOVLload [off] {sym} ptr mem)) && canMergeLoad(v, l, x) && clobber(l) -> (ANDLmem x [off] {sym} ptr mem)
(ORQ  x l:(MOVQload [off] {sym} ptr mem)) && canMergeLoad(v, l, x) && clobber(l) -> (ORQmem x [off] {sym} ptr mem)
(ORL  x l:(MOVLload [off] {sym} ptr mem)) && canMergeLoad(v, l, x) && clobber(l) -> (ORLmem x [off] {sym} ptr mem)
(XORQ x l:(MOVQload [off] {sym} ptr mem)) && canMergeLoad(v, l, x) && clobber(l) -> (XORQmem x [off] {sym} ptr mem)
(XORL x l:(MOVLload [off] {sym} ptr mem)) && canMergeLoad(v, l, x) && clobber(l) -> (XORLmem x [off] {sym} ptr mem)
(ADDSD x l:(MOVSDload [off] {sym} ptr mem)) && canMergeLoad(v, l, x) && clobber(l) -> (ADDSDmem x [off] {sym} ptr mem)
(ADDSS x l:(MOVSSload [off] {sym} ptr mem)) && canMergeLoad(v, l, x) && clobber(l) -> (ADDSSmem x [off] {sym} ptr mem)
(SUBSD x l:(MOVSDload [off] {sym} ptr mem)) && canMergeLoad(v, l, x) && clobber(l) -> (SUBSDmem x [off] {sym} ptr mem)
(SUBSS x l:(MOVSSload [off] {sym} ptr mem)) && canMergeLoad(v, l, x) && clobber(l) -> (SUBSSmem x [off] {sym} ptr mem)
(MULSD x l:(MOVSDload [off] {sym} ptr mem)) && canMergeLoad(v, l, x) && clobber(l) -> (MULSDmem x [off] {sym} ptr mem)
(MULSS x l:(MOVSSload [off] {sym} ptr mem)) && canMergeLoad(v, l, x) && clobber(l) -> (MULSSmem x [off] {sym} ptr mem)

// Merge ADDQconst and LEAQ into atomic loads.
(MOVQatomicload [off1] {sym} (ADDQconst [off2] ptr) mem) && is32Bit(off1+off2) ->
	(MOVQatomicload [off1+off2] {sym} ptr mem)
(MOVLatomicload [off1] {sym} (ADDQconst [off2] ptr) mem) && is32Bit(off1+off2) ->
	(MOVLatomicload [off1+off2] {sym} ptr mem)
(MOVQatomicload [off1] {sym1} (LEAQ [off2] {sym2} ptr) mem) && is32Bit(off1+off2) && canMergeSym(sym1, sym2) ->
	(MOVQatomicload [off1+off2] {mergeSym(sym1,sym2)} ptr mem)
(MOVLatomicload [off1] {sym1} (LEAQ [off2] {sym2} ptr) mem) && is32Bit(off1+off2) && canMergeSym(sym1, sym2) ->
	(MOVLatomicload [off1+off2] {mergeSym(sym1,sym2)} ptr mem)

// Merge ADDQconst and LEAQ into atomic stores.
(XCHGQ [off1] {sym} val (ADDQconst [off2] ptr) mem) && is32Bit(off1+off2) ->
	(XCHGQ [off1+off2] {sym} val ptr mem)
(XCHGQ [off1] {sym1} val (LEAQ [off2] {sym2} ptr) mem) && is32Bit(off1+off2) && canMergeSym(sym1, sym2) && ptr.Op != OpSB ->
	(XCHGQ [off1+off2] {mergeSym(sym1,sym2)} val ptr mem)
(XCHGL [off1] {sym} val (ADDQconst [off2] ptr) mem) && is32Bit(off1+off2) ->
	(XCHGL [off1+off2] {sym} val ptr mem)
(XCHGL [off1] {sym1} val (LEAQ [off2] {sym2} ptr) mem) && is32Bit(off1+off2) && canMergeSym(sym1, sym2) && ptr.Op != OpSB ->
	(XCHGL [off1+off2] {mergeSym(sym1,sym2)} val ptr mem)

// Merge ADDQconst into atomic adds.
// TODO: merging LEAQ doesn't work, assembler doesn't like the resulting instructions.
(XADDQlock [off1] {sym} val (ADDQconst [off2] ptr) mem) && is32Bit(off1+off2) ->
	(XADDQlock [off1+off2] {sym} val ptr mem)
(XADDLlock [off1] {sym} val (ADDQconst [off2] ptr) mem) && is32Bit(off1+off2) ->
	(XADDLlock [off1+off2] {sym} val ptr mem)

// Merge ADDQconst into atomic compare and swaps.
// TODO: merging LEAQ doesn't work, assembler doesn't like the resulting instructions.
(CMPXCHGQlock [off1] {sym} (ADDQconst [off2] ptr) old new_ mem) && is32Bit(off1+off2) ->
	(CMPXCHGQlock [off1+off2] {sym} ptr old new_ mem)
(CMPXCHGLlock [off1] {sym} (ADDQconst [off2] ptr) old new_ mem) && is32Bit(off1+off2) ->
	(CMPXCHGLlock [off1+off2] {sym} ptr old new_ mem)

// We don't need the conditional move if we know the arg of BSF is not zero.
(CMOVQEQ x _ (Select1 (BSFQ (ORQconst [c] _)))) && c != 0 -> x
// Extension is unnecessary for trailing zeros.
(BSFQ (ORQconst <t> [1<<8] (MOVBQZX x))) -> (BSFQ (ORQconst <t> [1<<8] x))
(BSFQ (ORQconst <t> [1<<16] (MOVWQZX x))) -> (BSFQ (ORQconst <t> [1<<16] x))

// Redundant sign/zero extensions
(MOVLQSX x:(MOVLQSX _)) -> x
(MOVLQSX x:(MOVWQSX _)) -> x
(MOVLQSX x:(MOVBQSX _)) -> x
(MOVWQSX x:(MOVWQSX _)) -> x
(MOVWQSX x:(MOVBQSX _)) -> x
(MOVBQSX x:(MOVBQSX _)) -> x
(MOVLQZX x:(MOVLQZX _)) -> x
(MOVLQZX x:(MOVWQZX _)) -> x
(MOVLQZX x:(MOVBQZX _)) -> x
(MOVWQZX x:(MOVWQZX _)) -> x
(MOVWQZX x:(MOVBQZX _)) -> x
(MOVBQZX x:(MOVBQZX _)) -> x

(MOVQstore [off] {sym} ptr a:(ADDQconst [c] l:(MOVQload [off] {sym} ptr2 mem)) mem)
	&& isSamePtr(ptr, ptr2) && a.Uses == 1 && l.Uses == 1 && validValAndOff(c,off) ->
	(ADDQconstmem {sym} [makeValAndOff(c,off)] ptr mem)
(MOVLstore [off] {sym} ptr a:(ADDLconst [c] l:(MOVLload [off] {sym} ptr2 mem)) mem)
	&& isSamePtr(ptr, ptr2) && a.Uses == 1 && l.Uses == 1 && validValAndOff(c,off) ->
	(ADDLconstmem {sym} [makeValAndOff(c,off)] ptr mem)

// float <-> int register moves, with no conversion.
// These come up when compiling math.{Float{32,64}bits,Float{32,64}frombits}.
(MOVQload  [off] {sym} ptr (MOVSDstore [off] {sym} ptr val _)) -> (MOVQf2i val)
(MOVLload  [off] {sym} ptr (MOVSSstore [off] {sym} ptr val _)) -> (MOVLf2i val)
(MOVSDload [off] {sym} ptr (MOVQstore  [off] {sym} ptr val _)) -> (MOVQi2f val)
(MOVSSload [off] {sym} ptr (MOVLstore  [off] {sym} ptr val _)) -> (MOVLi2f val)

// Other load-like ops.
(ADDQmem x [off] {sym} ptr (MOVSDstore [off] {sym} ptr y _)) -> (ADDQ x (MOVQf2i y))
(ADDLmem x [off] {sym} ptr (MOVSSstore [off] {sym} ptr y _)) -> (ADDL x (MOVLf2i y))
(SUBQmem x [off] {sym} ptr (MOVSDstore [off] {sym} ptr y _)) -> (SUBQ x (MOVQf2i y))
(SUBLmem x [off] {sym} ptr (MOVSSstore [off] {sym} ptr y _)) -> (SUBL x (MOVLf2i y))
(ANDQmem x [off] {sym} ptr (MOVSDstore [off] {sym} ptr y _)) -> (ANDQ x (MOVQf2i y))
(ANDLmem x [off] {sym} ptr (MOVSSstore [off] {sym} ptr y _)) -> (ANDL x (MOVLf2i y))
( ORQmem x [off] {sym} ptr (MOVSDstore [off] {sym} ptr y _)) -> ( ORQ x (MOVQf2i y))
( ORLmem x [off] {sym} ptr (MOVSSstore [off] {sym} ptr y _)) -> ( ORL x (MOVLf2i y))
(XORQmem x [off] {sym} ptr (MOVSDstore [off] {sym} ptr y _)) -> (XORQ x (MOVQf2i y))
(XORLmem x [off] {sym} ptr (MOVSSstore [off] {sym} ptr y _)) -> (XORL x (MOVLf2i y))

(ADDQconstmem [valOff] {sym} ptr (MOVSDstore [ValAndOff(valOff).Off()] {sym} ptr x _)) ->
  (ADDQconst [ValAndOff(valOff).Val()] (MOVQf2i x))
(ADDLconstmem [valOff] {sym} ptr (MOVSSstore [ValAndOff(valOff).Off()] {sym} ptr x _)) ->
  (ADDLconst [ValAndOff(valOff).Val()] (MOVLf2i x))

(ADDSDmem x [off] {sym} ptr (MOVQstore [off] {sym} ptr y _)) -> (ADDSD x (MOVQi2f y))
(ADDSSmem x [off] {sym} ptr (MOVLstore [off] {sym} ptr y _)) -> (ADDSS x (MOVLi2f y))
(SUBSDmem x [off] {sym} ptr (MOVQstore [off] {sym} ptr y _)) -> (SUBSD x (MOVQi2f y))
(SUBSSmem x [off] {sym} ptr (MOVLstore [off] {sym} ptr y _)) -> (SUBSS x (MOVLi2f y))
(MULSDmem x [off] {sym} ptr (MOVQstore [off] {sym} ptr y _)) -> (MULSD x (MOVQi2f y))
(MULSSmem x [off] {sym} ptr (MOVLstore [off] {sym} ptr y _)) -> (MULSS x (MOVLi2f y))

// Redirect stores to use the other register set.
(MOVQstore  [off] {sym} ptr (MOVQf2i val) mem) -> (MOVSDstore [off] {sym} ptr val mem)
(MOVLstore  [off] {sym} ptr (MOVLf2i val) mem) -> (MOVSSstore [off] {sym} ptr val mem)
(MOVSDstore [off] {sym} ptr (MOVQi2f val) mem) -> (MOVQstore  [off] {sym} ptr val mem)
(MOVSSstore [off] {sym} ptr (MOVLi2f val) mem) -> (MOVLstore  [off] {sym} ptr val mem)

// Load args directly into the register class where it will be used.
// We do this by just modifying the type of the Arg.
(MOVQf2i <t> (Arg [off] {sym})) -> @b.Func.Entry (Arg <t> [off] {sym})
(MOVLf2i <t> (Arg [off] {sym})) -> @b.Func.Entry (Arg <t> [off] {sym})
(MOVQi2f <t> (Arg [off] {sym})) -> @b.Func.Entry (Arg <t> [off] {sym})
(MOVLi2f <t> (Arg [off] {sym})) -> @b.Func.Entry (Arg <t> [off] {sym})