// Copyright 2016 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
(Add(Ptr|32|16|8) ...) => (ADDL ...)
(Add(32|64)F ...) => (ADDS(S|D) ...)
(Add32carry ...) => (ADDLcarry ...)
(Add32withcarry ...) => (ADCL ...)

(Sub(Ptr|32|16|8) ...) => (SUBL ...)
(Sub(32|64)F ...) => (SUBS(S|D) ...)
(Sub32carry ...) => (SUBLcarry ...)
(Sub32withcarry ...) => (SBBL ...)

(Mul(32|16|8) ...) => (MULL ...)
(Mul(32|64)F ...) => (MULS(S|D) ...)
(Mul32uhilo ...) => (MULLQU ...)

(Select0 (Mul32uover x y)) => (Select0 <typ.UInt32> (MULLU x y))
(Select1 (Mul32uover x y)) => (SETO (Select1 <types.TypeFlags> (MULLU x y)))

(Avg32u ...) => (AVGLU ...)

(Div(32|64)F ...) => (DIVS(S|D) ...)
(Div(32|32u|16|16u) ...) => (DIV(L|LU|W|WU) ...)
(Div8   x y) => (DIVW  (SignExt8to16 x) (SignExt8to16 y))
(Div8u  x y) => (DIVWU (ZeroExt8to16 x) (ZeroExt8to16 y))

(Hmul(32|32u) ...) => (HMUL(L|LU) ...)

(Mod(32|32u|16|16u) ...) => (MOD(L|LU|W|WU) ...)
(Mod8   x y) => (MODW  (SignExt8to16 x) (SignExt8to16 y))
(Mod8u  x y) => (MODWU (ZeroExt8to16 x) (ZeroExt8to16 y))

(And(32|16|8) ...) => (ANDL ...)
(Or(32|16|8) ...) => (ORL ...)
(Xor(32|16|8) ...) => (XORL ...)

(Neg(32|16|8) ...) => (NEGL ...)
(Neg32F x) => (PXOR x (MOVSSconst <typ.Float32> [float32(math.Copysign(0, -1))]))
(Neg64F x) => (PXOR x (MOVSDconst <typ.Float64> [math.Copysign(0, -1)]))

(Com(32|16|8) ...) => (NOTL ...)

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

// Lowering pointer arithmetic
(OffPtr [off] ptr) => (ADDLconst [int32(off)] ptr)

(Bswap32 ...) => (BSWAPL ...)

(Sqrt ...) => (SQRTSD ...)
(Sqrt32 ...) => (SQRTSS ...)

(Ctz8 x) => (BSFL (ORLconst <typ.UInt32> [0x100] x))
(Ctz8NonZero ...) => (BSFL ...)
(Ctz16 x) => (BSFL (ORLconst <typ.UInt32> [0x10000] x))
(Ctz16NonZero ...) => (BSFL ...)
(Ctz32 ...) => (LoweredCtz32 ...)
(Ctz32NonZero ...) => (BSFL ...)

// Lowering extension
(SignExt8to16  ...) => (MOVBLSX ...)
(SignExt8to32  ...) => (MOVBLSX ...)
(SignExt16to32 ...) => (MOVWLSX ...)

(ZeroExt8to16  ...) => (MOVBLZX ...)
(ZeroExt8to32  ...) => (MOVBLZX ...)
(ZeroExt16to32 ...) => (MOVWLZX ...)

(Signmask x) => (SARLconst x [31])
(Zeromask <t> x) => (XORLconst [-1] (SBBLcarrymask <t> (CMPLconst x [1])))
(Slicemask <t> x) => (SARLconst (NEGL <t> x) [31])

// Lowering truncation
// Because we ignore high parts of registers, truncates are just copies.
(Trunc16to8  ...) => (Copy ...)
(Trunc32to8  ...) => (Copy ...)
(Trunc32to16 ...) => (Copy ...)

// Lowering float-int conversions
(Cvt32to32F ...) => (CVTSL2SS ...)
(Cvt32to64F ...) => (CVTSL2SD ...)

(Cvt32Fto32 ...) => (CVTTSS2SL ...)
(Cvt64Fto32 ...) => (CVTTSD2SL ...)

(Cvt32Fto64F ...) => (CVTSS2SD ...)
(Cvt64Fto32F ...) => (CVTSD2SS ...)

(Round32F ...) => (Copy ...)
(Round64F ...) => (Copy ...)

(CvtBoolToUint8 ...) => (Copy ...)

// Lowering shifts
// Unsigned shifts need to return 0 if shift amount is >= width of shifted value.
//   result = (arg << shift) & (shift >= argbits ? 0 : 0xffffffffffffffff)
(Lsh32x(32|16|8) <t> x y) && !shiftIsBounded(v) => (ANDL (SHLL <t> x y) (SBBLcarrymask <t> (CMP(L|W|B)const y [32])))
(Lsh16x(32|16|8) <t> x y) && !shiftIsBounded(v) => (ANDL (SHLL <t> x y) (SBBLcarrymask <t> (CMP(L|W|B)const y [32])))
(Lsh8x(32|16|8)  <t> x y) && !shiftIsBounded(v) => (ANDL (SHLL <t> x y) (SBBLcarrymask <t> (CMP(L|W|B)const y [32])))

(Lsh32x(32|16|8) <t> x y) && shiftIsBounded(v) => (SHLL <t> x y)
(Lsh16x(32|16|8) <t> x y) && shiftIsBounded(v) => (SHLL <t> x y)
(Lsh8x(32|16|8)  <t> x y) && shiftIsBounded(v) => (SHLL <t> x y)

(Rsh32Ux(32|16|8) <t> x y) && !shiftIsBounded(v) => (ANDL (SHRL <t> x y) (SBBLcarrymask <t> (CMP(L|W|B)const y [32])))
(Rsh16Ux(32|16|8) <t> x y) && !shiftIsBounded(v) => (ANDL (SHRW <t> x y) (SBBLcarrymask <t> (CMP(L|W|B)const y [16])))
(Rsh8Ux(32|16|8)  <t> x y) && !shiftIsBounded(v) => (ANDL (SHRB <t> x y) (SBBLcarrymask <t> (CMP(L|W|B)const y [8])))

(Rsh32Ux(32|16|8) <t> x y) && shiftIsBounded(v) => (SHRL <t> x y)
(Rsh16Ux(32|16|8) <t> x y) && shiftIsBounded(v) => (SHRW <t> x y)
(Rsh8Ux(32|16|8)  <t> x y) && shiftIsBounded(v) => (SHRB <t> x y)

// 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.

(Rsh32x(32|16|8) <t> x y) && !shiftIsBounded(v) => (SARL <t> x (ORL <y.Type> y (NOTL <y.Type> (SBBLcarrymask <y.Type> (CMP(L|W|B)const y [32])))))
(Rsh16x(32|16|8) <t> x y) && !shiftIsBounded(v) => (SARW <t> x (ORL <y.Type> y (NOTL <y.Type> (SBBLcarrymask <y.Type> (CMP(L|W|B)const y [16])))))
(Rsh8x(32|16|8) <t> x y)  && !shiftIsBounded(v) => (SARB <t> x (ORL <y.Type> y (NOTL <y.Type> (SBBLcarrymask <y.Type> (CMP(L|W|B)const y [8])))))

(Rsh32x(32|16|8) <t> x y) && shiftIsBounded(v) => (SARL x y)
(Rsh16x(32|16|8) <t> x y) && shiftIsBounded(v) => (SARW x y)
(Rsh8x(32|16|8) <t> x y)  && shiftIsBounded(v) => (SARB x y)

// constant shifts
// generic opt rewrites all constant shifts to shift by Const64
(Lsh32x64 x (Const64 [c])) && uint64(c) < 32 => (SHLLconst x [int32(c)])
(Rsh32x64 x (Const64 [c])) && uint64(c) < 32 => (SARLconst x [int32(c)])
(Rsh32Ux64 x (Const64 [c])) && uint64(c) < 32 => (SHRLconst x [int32(c)])
(Lsh16x64 x (Const64 [c])) && uint64(c) < 16 => (SHLLconst x [int32(c)])
(Rsh16x64 x (Const64 [c])) && uint64(c) < 16 => (SARWconst x [int16(c)])
(Rsh16Ux64 x (Const64 [c])) && uint64(c) < 16 => (SHRWconst x [int16(c)])
(Lsh8x64 x (Const64 [c])) && uint64(c) < 8 => (SHLLconst x [int32(c)])
(Rsh8x64 x (Const64 [c])) && uint64(c) < 8 => (SARBconst x [int8(c)])
(Rsh8Ux64 x (Const64 [c])) && uint64(c) < 8 => (SHRBconst x [int8(c)])

// large constant shifts
(Lsh32x64 _ (Const64 [c])) && uint64(c) >= 32 => (Const32 [0])
(Rsh32Ux64 _ (Const64 [c])) && uint64(c) >= 32 => (Const32 [0])
(Lsh16x64 _ (Const64 [c])) && uint64(c) >= 16 => (Const16 [0])
(Rsh16Ux64 _ (Const64 [c])) && uint64(c) >= 16 => (Const16 [0])
(Lsh8x64 _ (Const64 [c])) && uint64(c) >= 8 => (Const8 [0])
(Rsh8Ux64 _ (Const64 [c])) && uint64(c) >= 8 => (Const8 [0])

// large constant signed right shift, we leave the sign bit
(Rsh32x64 x (Const64 [c])) && uint64(c) >= 32 => (SARLconst x [31])
(Rsh16x64 x (Const64 [c])) && uint64(c) >= 16 => (SARWconst x [15])
(Rsh8x64 x (Const64 [c])) && uint64(c) >= 8 => (SARBconst x [7])

// rotates
(RotateLeft32 ...) => (ROLL ...)
(RotateLeft16 ...) => (ROLW ...)
(RotateLeft8  ...) => (ROLB ...)
// constant rotates
(ROLL x (MOVLconst [c])) => (ROLLconst [c&31] x)
(ROLW x (MOVLconst [c])) => (ROLWconst [int16(c&15)] x)
(ROLB x (MOVLconst [c])) => (ROLBconst [int8(c&7)] x)

// Lowering comparisons
(Less32  x y) => (SETL (CMPL x y))
(Less16  x y) => (SETL (CMPW x y))
(Less8   x y) => (SETL (CMPB 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))

(Leq32  x y) => (SETLE (CMPL x y))
(Leq16  x y) => (SETLE (CMPW x y))
(Leq8   x y) => (SETLE (CMPB 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))

(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) => (SETEQ (CMPL x y))
(Eq64F x y) => (SETEQF (UCOMISD x y))
(Eq32F x y) => (SETEQF (UCOMISS 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) => (SETNE (CMPL x y))
(Neq64F x y) => (SETNEF (UCOMISD x y))
(Neq32F x y) => (SETNEF (UCOMISS x y))

// Lowering loads
(Load <t> ptr mem) && (is32BitInt(t) || isPtr(t)) => (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
(Store {t} ptr val mem) && t.Size() == 8 &&  t.IsFloat() => (MOVSDstore ptr val mem)
(Store {t} ptr val mem) && t.Size() == 4 &&  t.IsFloat() => (MOVSSstore ptr val mem)
(Store {t} ptr val mem) && t.Size() == 4 && !t.IsFloat() => (MOVLstore ptr val mem)
(Store {t} ptr val mem) && t.Size() == 2 => (MOVWstore ptr val mem)
(Store {t} ptr val mem) && t.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 [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 [8] dst src mem) =>
	(MOVLstore [4] dst (MOVLload [4] src mem)
		(MOVLstore dst (MOVLload src mem) mem))

// Adjust moves to be a multiple of 4 bytes.
(Move [s] dst src mem)
	&& s > 8 && s%4 != 0 =>
	(Move [s-s%4]
		(ADDLconst <dst.Type> dst [int32(s%4)])
		(ADDLconst <src.Type> src [int32(s%4)])
		(MOVLstore dst (MOVLload src mem) mem))

// Medium copying uses a duff device.
(Move [s] dst src mem)
	&& s > 8 && s <= 4*128 && s%4 == 0
	&& !config.noDuffDevice && logLargeCopy(v, s) =>
	(DUFFCOPY [10*(128-s/4)] dst src mem)
// 10 and 128 are magic constants.  10 is the number of bytes to encode:
//	MOVL	(SI), CX
//	ADDL	$4, SI
//	MOVL	CX, (DI)
//	ADDL	$4, DI
// and 128 is the number of such blocks. See src/runtime/duff_386.s:duffcopy.

// Large copying uses REP MOVSL.
(Move [s] dst src mem) && (s > 4*128 || config.noDuffDevice) && s%4 == 0 && logLargeCopy(v, s) =>
	(REPMOVSL dst src (MOVLconst [int32(s/4)]) 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 [3] destptr mem) =>
	(MOVBstoreconst [makeValAndOff(0,2)] destptr
		(MOVWstoreconst [makeValAndOff(0,0)] destptr mem))
(Zero [5] destptr mem) =>
	(MOVBstoreconst [makeValAndOff(0,4)] destptr
		(MOVLstoreconst [makeValAndOff(0,0)] destptr mem))
(Zero [6] destptr mem) =>
	(MOVWstoreconst [makeValAndOff(0,4)] destptr
		(MOVLstoreconst [makeValAndOff(0,0)] destptr mem))
(Zero [7] destptr mem) =>
	(MOVLstoreconst [makeValAndOff(0,3)] destptr
		(MOVLstoreconst [makeValAndOff(0,0)] destptr mem))

// Strip off any fractional word zeroing.
(Zero [s] destptr mem) && s%4 != 0 && s > 4 =>
	(Zero [s-s%4] (ADDLconst destptr [int32(s%4)])
		(MOVLstoreconst [0] destptr mem))

// Zero small numbers of words directly.
(Zero [8] destptr mem) =>
	(MOVLstoreconst [makeValAndOff(0,4)] destptr
		(MOVLstoreconst [makeValAndOff(0,0)] destptr mem))
(Zero [12] destptr mem) =>
	(MOVLstoreconst [makeValAndOff(0,8)] destptr
		(MOVLstoreconst [makeValAndOff(0,4)] destptr
			(MOVLstoreconst [makeValAndOff(0,0)] destptr mem)))
(Zero [16] destptr mem) =>
	(MOVLstoreconst [makeValAndOff(0,12)] destptr
		(MOVLstoreconst [makeValAndOff(0,8)] destptr
			(MOVLstoreconst [makeValAndOff(0,4)] destptr
				(MOVLstoreconst [makeValAndOff(0,0)] destptr mem))))

// Medium zeroing uses a duff device.
(Zero [s] destptr mem)
  && s > 16 && s <= 4*128 && s%4 == 0
  && !config.noDuffDevice =>
	(DUFFZERO [1*(128-s/4)] destptr (MOVLconst [0]) mem)
// 1 and 128 are magic constants.  1 is the number of bytes to encode STOSL.
// 128 is the number of STOSL instructions in duffzero.
// See src/runtime/duff_386.s:duffzero.

// Large zeroing uses REP STOSQ.
(Zero [s] destptr mem)
  && (s > 4*128 || (config.noDuffDevice && s > 16))
  && s%4 == 0 =>
	(REPSTOSL destptr (MOVLconst [int32(s/4)]) (MOVLconst [0]) mem)


// Lowering constants
(Const8   [c]) => (MOVLconst [int32(c)])
(Const16  [c]) => (MOVLconst [int32(c)])
(Const32  ...) => (MOVLconst ...)
(Const(32|64)F ...) => (MOVS(S|D)const ...)
(ConstNil) => (MOVLconst [0])
(ConstBool [c]) => (MOVLconst [b2i32(c)])

// Lowering calls
(StaticCall ...) => (CALLstatic ...)
(ClosureCall ...) => (CALLclosure ...)
(InterCall ...) => (CALLinter ...)
(TailCall ...) => (CALLtail ...)

// Miscellaneous
(IsNonNil p) => (SETNE (TESTL p p))
(IsInBounds idx len) => (SETB (CMPL idx len))
(IsSliceInBounds idx len) => (SETBE (CMPL idx len))
(NilCheck ...) => (LoweredNilCheck ...)
(GetG ...) => (LoweredGetG ...)
(GetClosurePtr ...) => (LoweredGetClosurePtr ...)
(GetCallerPC ...) => (LoweredGetCallerPC ...)
(GetCallerSP ...) => (LoweredGetCallerSP ...)
(Addr {sym} base) => (LEAL {sym} base)
(LocalAddr <t> {sym} base mem) && t.Elem().HasPointers() => (LEAL {sym} (SPanchored base mem))
(LocalAddr <t> {sym} base _)  && !t.Elem().HasPointers() => (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)
(If (SETO  cmp) yes no) => (OS 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)

// Write barrier.
(WB ...) => (LoweredWB ...)

(PanicBounds [kind] x y mem) && boundsABI(kind) == 0 => (LoweredPanicBoundsA [kind] x y mem)
(PanicBounds [kind] x y mem) && boundsABI(kind) == 1 => (LoweredPanicBoundsB [kind] x y mem)
(PanicBounds [kind] x y mem) && boundsABI(kind) == 2 => (LoweredPanicBoundsC [kind] x y mem)

(PanicExtend [kind] hi lo y mem) && boundsABI(kind) == 0 => (LoweredPanicExtendA [kind] hi lo y mem)
(PanicExtend [kind] hi lo y mem) && boundsABI(kind) == 1 => (LoweredPanicExtendB [kind] hi lo y mem)
(PanicExtend [kind] hi lo y mem) && boundsABI(kind) == 2 => (LoweredPanicExtendC [kind] hi lo y 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)
(NE (TESTB (SETO cmp) (SETO cmp)) yes no) => (OS cmp yes no)

// 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)

// fold constants into instructions
(ADDL x (MOVLconst <t> [c])) && !t.IsPtr() => (ADDLconst [c] x)
(ADDLcarry x (MOVLconst [c])) => (ADDLconstcarry [c] x)
(ADCL x (MOVLconst [c]) f) => (ADCLconst [c] x f)

(SUBL x (MOVLconst [c])) => (SUBLconst x [c])
(SUBL (MOVLconst [c]) x) => (NEGL (SUBLconst <v.Type> x [c]))
(SUBLcarry x (MOVLconst [c])) => (SUBLconstcarry [c] x)
(SBBL x (MOVLconst [c]) f) => (SBBLconst [c] x f)

(MULL x (MOVLconst [c])) => (MULLconst [c] x)
(ANDL x (MOVLconst [c])) => (ANDLconst [c] x)

(ANDLconst [c] (ANDLconst [d] x)) => (ANDLconst [c & d] x)
(XORLconst [c] (XORLconst [d] x)) => (XORLconst [c ^ d] x)
(MULLconst [c] (MULLconst [d] x)) => (MULLconst [c * d] x)

(ORL x (MOVLconst [c])) => (ORLconst [c] x)
(XORL x (MOVLconst [c])) => (XORLconst [c] x)

(SHLL x (MOVLconst [c])) => (SHLLconst [c&31] x)
(SHRL x (MOVLconst [c])) => (SHRLconst [c&31] x)
(SHRW x (MOVLconst [c])) && c&31 < 16 => (SHRWconst [int16(c&31)] x)
(SHRW _ (MOVLconst [c])) && c&31 >= 16 => (MOVLconst [0])
(SHRB x (MOVLconst [c])) && c&31 < 8 => (SHRBconst [int8(c&31)] x)
(SHRB _ (MOVLconst [c])) && c&31 >= 8 => (MOVLconst [0])

(SARL x (MOVLconst [c])) => (SARLconst [c&31] x)
(SARW x (MOVLconst [c])) => (SARWconst [int16(min(int64(c&31),15))] x)
(SARB x (MOVLconst [c])) => (SARBconst [int8(min(int64(c&31),7))] x)

(SARL x (ANDLconst [31] y)) => (SARL x y)
(SHLL x (ANDLconst [31] y)) => (SHLL x y)
(SHRL x (ANDLconst [31] y)) => (SHRL x y)

// Constant shift simplifications

(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

(ROLLconst [0] x) => x
(ROLWconst [0] x) => x
(ROLBconst [0] x) => 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.

(CMPL x (MOVLconst [c])) => (CMPLconst x [c])
(CMPL (MOVLconst [c]) x) => (InvertFlags (CMPLconst x [c]))
(CMPW x (MOVLconst [c])) => (CMPWconst x [int16(c)])
(CMPW (MOVLconst [c]) x) => (InvertFlags (CMPWconst x [int16(c)]))
(CMPB x (MOVLconst [c])) => (CMPBconst x [int8(c)])
(CMPB (MOVLconst [c]) x) => (InvertFlags (CMPBconst x [int8(c)]))

// Canonicalize the order of arguments to comparisons - helps with CSE.
(CMP(L|W|B) x y) && canonLessThan(x,y) => (InvertFlags (CMP(L|W|B) y x))

// strength reduction
// Assumes that the following costs from https://gmplib.org/~tege/x86-timing.pdf:
//    1 - addl, shll, leal, negl, subl
//    3 - imull
// This limits the rewrites to two instructions.
// Note that negl always operates in-place,
// which can require a register-register move
// to preserve the original value,
// so it must be used with care.
(MULLconst [-9] x) => (NEGL (LEAL8 <v.Type> x x))
(MULLconst [-5] x) => (NEGL (LEAL4 <v.Type> x x))
(MULLconst [-3] x) => (NEGL (LEAL2 <v.Type> x x))
(MULLconst [-1] x) => (NEGL x)
(MULLconst [0] _) => (MOVLconst [0])
(MULLconst [1] x) => x
(MULLconst [3] x) => (LEAL2 x x)
(MULLconst [5] x) => (LEAL4 x x)
(MULLconst [7] x) => (LEAL2 x (LEAL2 <v.Type> x x))
(MULLconst [9] x) => (LEAL8 x x)
(MULLconst [11] x) => (LEAL2 x (LEAL4 <v.Type> x x))
(MULLconst [13] x) => (LEAL4 x (LEAL2 <v.Type> x x))
(MULLconst [19] x) => (LEAL2 x (LEAL8 <v.Type> x x))
(MULLconst [21] x) => (LEAL4 x (LEAL4 <v.Type> x x))
(MULLconst [25] x) => (LEAL8 x (LEAL2 <v.Type> x x))
(MULLconst [27] x) => (LEAL8 (LEAL2 <v.Type> x x) (LEAL2 <v.Type> x x))
(MULLconst [37] x) => (LEAL4 x (LEAL8 <v.Type> x x))
(MULLconst [41] x) => (LEAL8 x (LEAL4 <v.Type> x x))
(MULLconst [45] x) => (LEAL8 (LEAL4 <v.Type> x x) (LEAL4 <v.Type> x x))
(MULLconst [73] x) => (LEAL8 x (LEAL8 <v.Type> x x))
(MULLconst [81] x) => (LEAL8 (LEAL8 <v.Type> x x) (LEAL8 <v.Type> x x))

(MULLconst [c] x) && isPowerOfTwo32(c+1) && c >= 15 => (SUBL (SHLLconst <v.Type> [int32(log32(c+1))] x) x)
(MULLconst [c] x) && isPowerOfTwo32(c-1) && c >= 17 => (LEAL1 (SHLLconst <v.Type> [int32(log32(c-1))] x) x)
(MULLconst [c] x) && isPowerOfTwo32(c-2) && c >= 34 => (LEAL2 (SHLLconst <v.Type> [int32(log32(c-2))] x) x)
(MULLconst [c] x) && isPowerOfTwo32(c-4) && c >= 68 => (LEAL4 (SHLLconst <v.Type> [int32(log32(c-4))] x) x)
(MULLconst [c] x) && isPowerOfTwo32(c-8) && c >= 136 => (LEAL8 (SHLLconst <v.Type> [int32(log32(c-8))] x) x)
(MULLconst [c] x) && c%3 == 0 && isPowerOfTwo32(c/3) => (SHLLconst [int32(log32(c/3))] (LEAL2 <v.Type> x x))
(MULLconst [c] x) && c%5 == 0 && isPowerOfTwo32(c/5) => (SHLLconst [int32(log32(c/5))] (LEAL4 <v.Type> x x))
(MULLconst [c] x) && c%9 == 0 && isPowerOfTwo32(c/9) => (SHLLconst [int32(log32(c/9))] (LEAL8 <v.Type> x x))

// combine add/shift into LEAL
(ADDL x (SHLLconst [3] y)) => (LEAL8 x y)
(ADDL x (SHLLconst [2] y)) => (LEAL4 x y)
(ADDL x (SHLLconst [1] y)) => (LEAL2 x y)
(ADDL x (ADDL y y)) => (LEAL2 x y)
(ADDL x (ADDL x y)) => (LEAL2 y x)

// combine ADDL/ADDLconst into LEAL1
(ADDLconst [c] (ADDL x y)) => (LEAL1 [c] x y)
(ADDL (ADDLconst [c] x) y) => (LEAL1 [c] x y)

// fold ADDL into LEAL
(ADDLconst [c] (LEAL [d] {s} x)) && is32Bit(int64(c)+int64(d)) => (LEAL [c+d] {s} x)
(LEAL [c] {s} (ADDLconst [d] x)) && is32Bit(int64(c)+int64(d)) => (LEAL [c+d] {s} x)
(ADDLconst [c] x:(SP)) => (LEAL [c] x) // so it is rematerializeable
(LEAL [c] {s} (ADDL x y)) && x.Op != OpSB && y.Op != OpSB => (LEAL1 [c] {s} x y)
(ADDL x (LEAL [c] {s} y)) && x.Op != OpSB && y.Op != OpSB => (LEAL1 [c] {s} x y)

// fold ADDLconst into LEALx
(ADDLconst [c] (LEAL1 [d] {s} x y)) && is32Bit(int64(c)+int64(d)) => (LEAL1 [c+d] {s} x y)
(ADDLconst [c] (LEAL2 [d] {s} x y)) && is32Bit(int64(c)+int64(d)) => (LEAL2 [c+d] {s} x y)
(ADDLconst [c] (LEAL4 [d] {s} x y)) && is32Bit(int64(c)+int64(d)) => (LEAL4 [c+d] {s} x y)
(ADDLconst [c] (LEAL8 [d] {s} x y)) && is32Bit(int64(c)+int64(d)) => (LEAL8 [c+d] {s} x y)
(LEAL1 [c] {s} (ADDLconst [d] x) y) && is32Bit(int64(c)+int64(d))   && x.Op != OpSB => (LEAL1 [c+d] {s} x y)
(LEAL2 [c] {s} (ADDLconst [d] x) y) && is32Bit(int64(c)+int64(d))   && x.Op != OpSB => (LEAL2 [c+d] {s} x y)
(LEAL2 [c] {s} x (ADDLconst [d] y)) && is32Bit(int64(c)+2*int64(d)) && y.Op != OpSB => (LEAL2 [c+2*d] {s} x y)
(LEAL4 [c] {s} (ADDLconst [d] x) y) && is32Bit(int64(c)+int64(d))   && x.Op != OpSB => (LEAL4 [c+d] {s} x y)
(LEAL4 [c] {s} x (ADDLconst [d] y)) && is32Bit(int64(c)+4*int64(d)) && y.Op != OpSB => (LEAL4 [c+4*d] {s} x y)
(LEAL8 [c] {s} (ADDLconst [d] x) y) && is32Bit(int64(c)+int64(d))   && x.Op != OpSB => (LEAL8 [c+d] {s} x y)
(LEAL8 [c] {s} x (ADDLconst [d] y)) && is32Bit(int64(c)+8*int64(d)) && y.Op != OpSB => (LEAL8 [c+8*d] {s} x y)

// fold shifts into LEALx
(LEAL1 [c] {s} x (SHLLconst [1] y)) => (LEAL2 [c] {s} x y)
(LEAL1 [c] {s} x (SHLLconst [2] y)) => (LEAL4 [c] {s} x y)
(LEAL1 [c] {s} x (SHLLconst [3] y)) => (LEAL8 [c] {s} x y)
(LEAL2 [c] {s} x (SHLLconst [1] y)) => (LEAL4 [c] {s} x y)
(LEAL2 [c] {s} x (SHLLconst [2] y)) => (LEAL8 [c] {s} x y)
(LEAL4 [c] {s} x (SHLLconst [1] y)) => (LEAL8 [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.
(MOVBLSX x:(MOVBload [off] {sym} ptr mem)) && x.Uses == 1 && clobber(x) => @x.Block (MOVBLSXload <v.Type> [off] {sym} ptr mem)
(MOVBLZX x:(MOVBload [off] {sym} ptr mem)) && x.Uses == 1 && clobber(x) => @x.Block (MOVBload <v.Type> [off] {sym} ptr mem)
(MOVWLSX x:(MOVWload [off] {sym} ptr mem)) && x.Uses == 1 && clobber(x) => @x.Block (MOVWLSXload <v.Type> [off] {sym} ptr mem)
(MOVWLZX x:(MOVWload [off] {sym} ptr mem)) && x.Uses == 1 && clobber(x) => @x.Block (MOVWload <v.Type> [off] {sym} ptr 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) => (MOVBLZX x)
(MOVWload [off] {sym} ptr (MOVWstore [off2] {sym2} ptr2 x _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) => (MOVWLZX x)
(MOVLload [off] {sym} ptr (MOVLstore [off2] {sym2} ptr2 x _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) => x
(MOVBLSXload [off] {sym} ptr (MOVBstore [off2] {sym2} ptr2 x _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) => (MOVBLSX x)
(MOVWLSXload [off] {sym} ptr (MOVWstore [off2] {sym2} ptr2 x _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) => (MOVWLSX x)

// Fold extensions and ANDs together.
(MOVBLZX (ANDLconst [c] x)) => (ANDLconst [c & 0xff] x)
(MOVWLZX (ANDLconst [c] x)) => (ANDLconst [c & 0xffff] x)
(MOVBLSX (ANDLconst [c] x)) && c & 0x80 == 0 => (ANDLconst [c & 0x7f] x)
(MOVWLSX (ANDLconst [c] x)) && c & 0x8000 == 0 => (ANDLconst [c & 0x7fff] x)

// Don't extend before storing
(MOVWstore [off] {sym} ptr (MOVWL(S|Z)X x) mem) => (MOVWstore [off] {sym} ptr x mem)
(MOVBstore [off] {sym} ptr (MOVBL(S|Z)X 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 ADDLconst get eliminated, we still have to compute the ADDLconst and we now
// have potentially two live values (ptr and (ADDLconst [off] ptr)) instead of one.
// Nevertheless, let's do it!
(MOV(L|W|B|SS|SD)load  [off1] {sym} (ADDLconst [off2] ptr) mem) && is32Bit(int64(off1)+int64(off2)) =>
    (MOV(L|W|B|SS|SD)load  [off1+off2] {sym} ptr mem)
(MOV(L|W|B|SS|SD)store  [off1] {sym} (ADDLconst [off2] ptr) val mem) && is32Bit(int64(off1)+int64(off2)) =>
    (MOV(L|W|B|SS|SD)store  [off1+off2] {sym} ptr val mem)

((ADD|SUB|MUL|AND|OR|XOR)Lload [off1] {sym} val (ADDLconst [off2] base) mem) && is32Bit(int64(off1)+int64(off2)) =>
	((ADD|SUB|MUL|AND|OR|XOR)Lload [off1+off2] {sym} val base mem)
((ADD|SUB|MUL|DIV)SSload [off1] {sym} val (ADDLconst [off2] base) mem) && is32Bit(int64(off1)+int64(off2)) =>
	((ADD|SUB|MUL|DIV)SSload [off1+off2] {sym} val base mem)
((ADD|SUB|MUL|DIV)SDload [off1] {sym} val (ADDLconst [off2] base) mem) && is32Bit(int64(off1)+int64(off2)) =>
	((ADD|SUB|MUL|DIV)SDload [off1+off2] {sym} val base mem)
((ADD|SUB|AND|OR|XOR)Lmodify [off1] {sym} (ADDLconst [off2] base) val mem) && is32Bit(int64(off1)+int64(off2)) =>
	((ADD|SUB|AND|OR|XOR)Lmodify [off1+off2] {sym} base val mem)
((ADD|AND|OR|XOR)Lconstmodify [valoff1] {sym} (ADDLconst [off2] base) mem) && valoff1.canAdd32(off2) =>
	((ADD|AND|OR|XOR)Lconstmodify [valoff1.addOffset32(off2)] {sym} base mem)

// Fold constants into stores.
(MOVLstore [off] {sym} ptr (MOVLconst [c]) mem) =>
	(MOVLstoreconst [makeValAndOff(c,off)] {sym} ptr mem)
(MOVWstore [off] {sym} ptr (MOVLconst [c]) mem) =>
	(MOVWstoreconst [makeValAndOff(c,off)] {sym} ptr mem)
(MOVBstore [off] {sym} ptr (MOVLconst [c]) mem) =>
	(MOVBstoreconst [makeValAndOff(c,off)] {sym} ptr mem)

// Fold address offsets into constant stores.
(MOV(L|W|B)storeconst [sc] {s} (ADDLconst [off] ptr) mem) && sc.canAdd32(off) =>
	(MOV(L|W|B)storeconst [sc.addOffset32(off)] {s} ptr mem)

// We need to fold LEAL into the MOVx ops so that the live variable analysis knows
// what variables are being read/written by the ops.
// Note: we turn off this merging for operations on globals when building
// position-independent code (when Flag_shared is set).
// PIC needs a spare register to load the PC into.  Having the LEAL be
// a separate instruction gives us that register.  Having the LEAL be
// a separate instruction also allows it to be CSEd (which is good because
// it compiles to a thunk call).
(MOV(L|W|B|SS|SD|BLSX|WLSX)load  [off1] {sym1} (LEAL [off2] {sym2} base) mem) && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2)
  && (base.Op != OpSB || !config.ctxt.Flag_shared) =>
        (MOV(L|W|B|SS|SD|BLSX|WLSX)load  [off1+off2] {mergeSym(sym1,sym2)} base mem)

(MOV(L|W|B|SS|SD)store  [off1] {sym1} (LEAL [off2] {sym2} base) val mem) && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2)
  && (base.Op != OpSB || !config.ctxt.Flag_shared) =>
	(MOV(L|W|B|SS|SD)store  [off1+off2] {mergeSym(sym1,sym2)} base val mem)

(MOV(L|W|B)storeconst [sc] {sym1} (LEAL [off] {sym2} ptr) mem) && canMergeSym(sym1, sym2) && sc.canAdd32(off)
  && (ptr.Op != OpSB || !config.ctxt.Flag_shared) =>
	(MOV(L|W|B)storeconst [sc.addOffset32(off)] {mergeSym(sym1, sym2)} ptr mem)

((ADD|SUB|MUL|AND|OR|XOR)Lload [off1] {sym1} val (LEAL [off2] {sym2} base) mem)
	&& is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) && (base.Op != OpSB || !config.ctxt.Flag_shared) =>
	((ADD|SUB|MUL|AND|OR|XOR)Lload [off1+off2] {mergeSym(sym1,sym2)} val base mem)
((ADD|SUB|MUL|DIV)SSload [off1] {sym1} val (LEAL [off2] {sym2} base) mem)
	&& is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) && (base.Op != OpSB || !config.ctxt.Flag_shared) =>
	((ADD|SUB|MUL|DIV)SSload [off1+off2] {mergeSym(sym1,sym2)} val base mem)
((ADD|SUB|MUL|DIV)SDload [off1] {sym1} val (LEAL [off2] {sym2} base) mem)
	&& is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) && (base.Op != OpSB || !config.ctxt.Flag_shared) =>
	((ADD|SUB|MUL|DIV)SDload [off1+off2] {mergeSym(sym1,sym2)} val base mem)
((ADD|SUB|AND|OR|XOR)Lmodify [off1] {sym1} (LEAL [off2] {sym2} base) val mem)
	&& is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) && (base.Op != OpSB || !config.ctxt.Flag_shared) =>
	((ADD|SUB|AND|OR|XOR)Lmodify [off1+off2] {mergeSym(sym1,sym2)} base val mem)
((ADD|AND|OR|XOR)Lconstmodify [valoff1] {sym1} (LEAL [off2] {sym2} base) mem)
	&& valoff1.canAdd32(off2) && canMergeSym(sym1, sym2) && (base.Op != OpSB || !config.ctxt.Flag_shared) =>
	((ADD|AND|OR|XOR)Lconstmodify [valoff1.addOffset32(off2)] {mergeSym(sym1,sym2)} base mem)

// Merge load/store to op
((ADD|AND|OR|XOR|SUB|MUL)L x l:(MOVLload [off] {sym} ptr mem)) && canMergeLoadClobber(v, l, x) && clobber(l) => ((ADD|AND|OR|XOR|SUB|MUL)Lload x [off] {sym} ptr mem)
((ADD|SUB|MUL|DIV)SD x l:(MOVSDload [off] {sym} ptr mem)) && canMergeLoadClobber(v, l, x) && clobber(l) => ((ADD|SUB|MUL|DIV)SDload x [off] {sym} ptr mem)
((ADD|SUB|MUL|DIV)SS x l:(MOVSSload [off] {sym} ptr mem)) && canMergeLoadClobber(v, l, x) && clobber(l) => ((ADD|SUB|MUL|DIV)SSload x [off] {sym} ptr mem)
(MOVLstore {sym} [off] ptr y:((ADD|AND|OR|XOR)Lload x [off] {sym} ptr mem) mem) && y.Uses==1 && clobber(y) => ((ADD|AND|OR|XOR)Lmodify [off] {sym} ptr x mem)
(MOVLstore {sym} [off] ptr y:((ADD|SUB|AND|OR|XOR)L l:(MOVLload [off] {sym} ptr mem) x) mem) && y.Uses==1 && l.Uses==1 && clobber(y, l) =>
	((ADD|SUB|AND|OR|XOR)Lmodify [off] {sym} ptr x mem)
(MOVLstore {sym} [off] ptr y:((ADD|AND|OR|XOR)Lconst [c] l:(MOVLload [off] {sym} ptr mem)) mem)
	&& y.Uses==1 && l.Uses==1 && clobber(y, l) =>
	((ADD|AND|OR|XOR)Lconstmodify [makeValAndOff(c,off)] {sym} ptr mem)

// fold LEALs together
(LEAL [off1] {sym1} (LEAL [off2] {sym2} x)) && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) =>
      (LEAL [off1+off2] {mergeSym(sym1,sym2)} x)

// LEAL into LEAL1
(LEAL1 [off1] {sym1} (LEAL [off2] {sym2} x) y) && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) && x.Op != OpSB =>
       (LEAL1 [off1+off2] {mergeSym(sym1,sym2)} x y)

// LEAL1 into LEAL
(LEAL [off1] {sym1} (LEAL1 [off2] {sym2} x y)) && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) =>
       (LEAL1 [off1+off2] {mergeSym(sym1,sym2)} x y)

// LEAL into LEAL[248]
(LEAL2 [off1] {sym1} (LEAL [off2] {sym2} x) y) && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) && x.Op != OpSB =>
       (LEAL2 [off1+off2] {mergeSym(sym1,sym2)} x y)
(LEAL4 [off1] {sym1} (LEAL [off2] {sym2} x) y) && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) && x.Op != OpSB =>
       (LEAL4 [off1+off2] {mergeSym(sym1,sym2)} x y)
(LEAL8 [off1] {sym1} (LEAL [off2] {sym2} x) y) && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) && x.Op != OpSB =>
       (LEAL8 [off1+off2] {mergeSym(sym1,sym2)} x y)

// LEAL[248] into LEAL
(LEAL [off1] {sym1} (LEAL2 [off2] {sym2} x y)) && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) =>
      (LEAL2 [off1+off2] {mergeSym(sym1,sym2)} x y)
(LEAL [off1] {sym1} (LEAL4 [off2] {sym2} x y)) && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) =>
      (LEAL4 [off1+off2] {mergeSym(sym1,sym2)} x y)
(LEAL [off1] {sym1} (LEAL8 [off2] {sym2} x y)) && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) =>
      (LEAL8 [off1+off2] {mergeSym(sym1,sym2)} x y)

// LEAL[1248] into LEAL[1248]. Only some such merges are possible.
(LEAL1 [off1] {sym1} x (LEAL1 [off2] {sym2} y y)) && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) =>
      (LEAL2 [off1+off2] {mergeSym(sym1, sym2)} x y)
(LEAL1 [off1] {sym1} x (LEAL1 [off2] {sym2} x y)) && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) =>
      (LEAL2 [off1+off2] {mergeSym(sym1, sym2)} y x)
(LEAL2 [off1] {sym} x (LEAL1 [off2] {nil} y y)) && is32Bit(int64(off1)+2*int64(off2)) =>
      (LEAL4 [off1+2*off2] {sym} x y)
(LEAL4 [off1] {sym} x (LEAL1 [off2] {nil} y y)) && is32Bit(int64(off1)+4*int64(off2)) =>
      (LEAL8 [off1+4*off2] {sym} 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.
(CMPLconst (MOVLconst [x]) [y]) && x==y                       => (FlagEQ)
(CMPLconst (MOVLconst [x]) [y]) && x<y && uint32(x)<uint32(y) => (FlagLT_ULT)
(CMPLconst (MOVLconst [x]) [y]) && x<y && uint32(x)>uint32(y) => (FlagLT_UGT)
(CMPLconst (MOVLconst [x]) [y]) && x>y && uint32(x)<uint32(y) => (FlagGT_ULT)
(CMPLconst (MOVLconst [x]) [y]) && x>y && uint32(x)>uint32(y) => (FlagGT_UGT)

(CMPWconst (MOVLconst [x]) [y]) && int16(x)==y                       => (FlagEQ)
(CMPWconst (MOVLconst [x]) [y]) && int16(x)<y && uint16(x)<uint16(y) => (FlagLT_ULT)
(CMPWconst (MOVLconst [x]) [y]) && int16(x)<y && uint16(x)>uint16(y) => (FlagLT_UGT)
(CMPWconst (MOVLconst [x]) [y]) && int16(x)>y && uint16(x)<uint16(y) => (FlagGT_ULT)
(CMPWconst (MOVLconst [x]) [y]) && int16(x)>y && uint16(x)>uint16(y) => (FlagGT_UGT)

(CMPBconst (MOVLconst [x]) [y]) && int8(x)==y                      => (FlagEQ)
(CMPBconst (MOVLconst [x]) [y]) && int8(x)<y && uint8(x)<uint8(y) => (FlagLT_ULT)
(CMPBconst (MOVLconst [x]) [y]) && int8(x)<y && uint8(x)>uint8(y) => (FlagLT_UGT)
(CMPBconst (MOVLconst [x]) [y]) && int8(x)>y && uint8(x)<uint8(y) => (FlagGT_ULT)
(CMPBconst (MOVLconst [x]) [y]) && int8(x)>y && uint8(x)>uint8(y) => (FlagGT_UGT)

// Other known comparisons.
(CMPLconst (SHRLconst _ [c]) [n]) && 0 <= n && 0 < c && c <= 32 && (1<<uint64(32-c)) <= uint64(n) => (FlagLT_ULT)
(CMPLconst (ANDLconst _ [m]) [n]) && 0 <= m && m < n => (FlagLT_ULT)
(CMPWconst (ANDLconst _ [m]) [n]) && 0 <= int16(m) && int16(m) < n => (FlagLT_ULT)
(CMPBconst (ANDLconst _ [m]) [n]) && 0 <= int8(m) && int8(m) < n => (FlagLT_ULT)
// TODO: DIVxU also.

// Absorb flag constants into SBB ops.
(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 yes no)
(EQ (FlagLT_ULT) yes no) => (First no yes)
(EQ (FlagLT_UGT) yes no) => (First no yes)
(EQ (FlagGT_ULT) yes no) => (First no yes)
(EQ (FlagGT_UGT) yes no) => (First no yes)

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

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

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

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

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

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

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

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

(UGE (FlagEQ) yes no) => (First yes no)
(UGE (FlagLT_ULT) yes no) => (First no yes)
(UGE (FlagLT_UGT) yes no) => (First yes no)
(UGE (FlagGT_ULT) yes no) => (First no yes)
(UGE (FlagGT_UGT) yes no) => (First 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
(ADDLconst [c] x) && c==0  => x
(SUBLconst [c] x) && c==0  => x
(ANDLconst [c] _) && c==0  => (MOVLconst [0])
(ANDLconst [c] x) && c==-1 => x
(ORLconst [c] x)  && c==0  => x
(ORLconst [c] _)  && c==-1 => (MOVLconst [-1])
(XORLconst [c] x) && 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
(SUBLconst [c] x) => (ADDLconst [-c] x)

// generic constant folding
// TODO: more of this
(ADDLconst [c] (MOVLconst [d])) => (MOVLconst [c+d])
(ADDLconst [c] (ADDLconst [d] x)) => (ADDLconst [c+d] x)
(SARLconst [c] (MOVLconst [d])) => (MOVLconst [d>>uint64(c)])
(SARWconst [c] (MOVLconst [d])) => (MOVLconst [d>>uint64(c)])
(SARBconst [c] (MOVLconst [d])) => (MOVLconst [d>>uint64(c)])
(NEGL (MOVLconst [c])) => (MOVLconst [-c])
(MULLconst [c] (MOVLconst [d])) => (MOVLconst [c*d])
(ANDLconst [c] (MOVLconst [d])) => (MOVLconst [c&d])
(ORLconst [c] (MOVLconst [d])) => (MOVLconst [c|d])
(XORLconst [c] (MOVLconst [d])) => (MOVLconst [c^d])
(NOTL (MOVLconst [c])) => (MOVLconst [^c])

// generic simplifications
// TODO: more of this
(ADDL x (NEGL y)) => (SUBL x y)
(SUBL x x) => (MOVLconst [0])
(ANDL x x) => x
(ORL x x) => x
(XORL x x) => (MOVLconst [0])

// checking AND against 0.
(CMP(L|W|B)const l:(ANDL x y) [0]) && l.Uses==1 => (TEST(L|W|B) x y)
(CMPLconst l:(ANDLconst [c] x) [0]) && l.Uses==1 => (TESTLconst [c] x)
(CMPWconst l:(ANDLconst [c] x) [0]) && l.Uses==1 => (TESTWconst [int16(c)] x)
(CMPBconst l:(ANDLconst [c] x) [0]) && l.Uses==1 => (TESTBconst [int8(c)] x)

// TEST %reg,%reg is shorter than CMP
(CMP(L|W|B)const x [0]) => (TEST(L|W|B) x x)

// Convert LEAL1 back to ADDL if we can
(LEAL1 [0] {nil} x y) => (ADDL x y)

// 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.
(ORL                  x0:(MOVBload [i0] {s} p mem)
    s0:(SHLLconst [8] x1:(MOVBload [i1] {s} p mem)))
  && i1 == i0+1
  && x0.Uses == 1
  && x1.Uses == 1
  && s0.Uses == 1
  && mergePoint(b,x0,x1) != nil
  && clobber(x0, x1, s0)
  => @mergePoint(b,x0,x1) (MOVWload [i0] {s} p mem)

(ORL                  x0:(MOVBload [i] {s} p0 mem)
    s0:(SHLLconst [8] x1:(MOVBload [i] {s} p1 mem)))
  && x0.Uses == 1
  && x1.Uses == 1
  && s0.Uses == 1
  && sequentialAddresses(p0, p1, 1)
  && mergePoint(b,x0,x1) != nil
  && clobber(x0, x1, s0)
  => @mergePoint(b,x0,x1) (MOVWload [i] {s} p0 mem)

(ORL o0:(ORL
                       x0:(MOVWload [i0] {s} p mem)
    s0:(SHLLconst [16] x1:(MOVBload [i2] {s} p mem)))
    s1:(SHLLconst [24] x2:(MOVBload [i3] {s} p mem)))
  && i2 == i0+2
  && i3 == i0+3
  && x0.Uses == 1
  && x1.Uses == 1
  && x2.Uses == 1
  && s0.Uses == 1
  && s1.Uses == 1
  && o0.Uses == 1
  && mergePoint(b,x0,x1,x2) != nil
  && clobber(x0, x1, x2, s0, s1, o0)
  => @mergePoint(b,x0,x1,x2) (MOVLload [i0] {s} p mem)

(ORL o0:(ORL
                       x0:(MOVWload [i] {s} p0 mem)
    s0:(SHLLconst [16] x1:(MOVBload [i] {s} p1 mem)))
    s1:(SHLLconst [24] x2:(MOVBload [i] {s} p2 mem)))
  && x0.Uses == 1
  && x1.Uses == 1
  && x2.Uses == 1
  && s0.Uses == 1
  && s1.Uses == 1
  && o0.Uses == 1
  && sequentialAddresses(p0, p1, 2)
  && sequentialAddresses(p1, p2, 1)
  && mergePoint(b,x0,x1,x2) != nil
  && clobber(x0, x1, x2, s0, s1, o0)
  => @mergePoint(b,x0,x1,x2) (MOVLload [i] {s} p0 mem)

// Combine constant stores into larger (unaligned) stores.
(MOVBstoreconst [c] {s} p x:(MOVBstoreconst [a] {s} p mem))
  && x.Uses == 1
  && a.Off() + 1 == c.Off()
  && clobber(x)
  => (MOVWstoreconst [makeValAndOff(a.Val()&0xff | c.Val()<<8, a.Off())] {s} p mem)
(MOVBstoreconst [a] {s} p x:(MOVBstoreconst [c] {s} p mem))
  && x.Uses == 1
  && a.Off() + 1 == c.Off()
  && clobber(x)
  => (MOVWstoreconst [makeValAndOff(a.Val()&0xff | c.Val()<<8, a.Off())] {s} p mem)

(MOVBstoreconst [c] {s} p1 x:(MOVBstoreconst [a] {s} p0 mem))
  && x.Uses == 1
  && a.Off() == c.Off()
  && sequentialAddresses(p0, p1, 1)
  && clobber(x)
  => (MOVWstoreconst [makeValAndOff(a.Val()&0xff | c.Val()<<8, a.Off())] {s} p0 mem)
(MOVBstoreconst [a] {s} p0 x:(MOVBstoreconst [c] {s} p1 mem))
  && x.Uses == 1
  && a.Off() == c.Off()
  && sequentialAddresses(p0, p1, 1)
  && clobber(x)
  => (MOVWstoreconst [makeValAndOff(a.Val()&0xff | c.Val()<<8, a.Off())] {s} p0 mem)

(MOVWstoreconst [c] {s} p x:(MOVWstoreconst [a] {s} p mem))
  && x.Uses == 1
  && a.Off() + 2 == c.Off()
  && clobber(x)
  => (MOVLstoreconst [makeValAndOff(a.Val()&0xffff | c.Val()<<16, a.Off())] {s} p mem)
(MOVWstoreconst [a] {s} p x:(MOVWstoreconst [c] {s} p mem))
  && x.Uses == 1
  && ValAndOff(a).Off() + 2 == ValAndOff(c).Off()
  && clobber(x)
  => (MOVLstoreconst [makeValAndOff(a.Val()&0xffff | c.Val()<<16, a.Off())] {s} p mem)

(MOVWstoreconst [c] {s} p1 x:(MOVWstoreconst [a] {s} p0 mem))
  && x.Uses == 1
  && a.Off() == c.Off()
  && sequentialAddresses(p0, p1, 2)
  && clobber(x)
  => (MOVLstoreconst [makeValAndOff(a.Val()&0xffff | c.Val()<<16, a.Off())] {s} p0 mem)
(MOVWstoreconst [a] {s} p0 x:(MOVWstoreconst [c] {s} p1 mem))
  && x.Uses == 1
  && a.Off() == c.Off()
  && sequentialAddresses(p0, p1, 2)
  && clobber(x)
  => (MOVLstoreconst [makeValAndOff(a.Val()&0xffff | c.Val()<<16, a.Off())] {s} p0 mem)

// Combine stores into larger (unaligned) stores.
(MOVBstore [i] {s} p (SHR(W|L)const [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 w x:(MOVBstore {s} [i+1] p (SHR(W|L)const [8] w) mem))
  && x.Uses == 1
  && clobber(x)
  => (MOVWstore [i] {s} p w mem)
(MOVBstore [i] {s} p (SHRLconst [j] w) x:(MOVBstore [i-1] {s} p w0:(SHRLconst [j-8] w) mem))
  && x.Uses == 1
  && clobber(x)
  => (MOVWstore [i-1] {s} p w0 mem)

(MOVBstore [i] {s} p1 (SHR(W|L)const [8] w) x:(MOVBstore [i] {s} p0 w mem))
  && x.Uses == 1
  && sequentialAddresses(p0, p1, 1)
  && clobber(x)
  => (MOVWstore [i] {s} p0 w mem)
(MOVBstore [i] {s} p0 w x:(MOVBstore {s} [i] p1 (SHR(W|L)const [8] w) mem))
  && x.Uses == 1
  && sequentialAddresses(p0, p1, 1)
  && clobber(x)
  => (MOVWstore [i] {s} p0 w mem)
(MOVBstore [i] {s} p1 (SHRLconst [j] w) x:(MOVBstore [i] {s} p0 w0:(SHRLconst [j-8] w) mem))
  && x.Uses == 1
  && sequentialAddresses(p0, p1, 1)
  && clobber(x)
  => (MOVWstore [i] {s} p0 w0 mem)

(MOVWstore [i] {s} p (SHRLconst [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 (SHRLconst [j] w) x:(MOVWstore [i-2] {s} p w0:(SHRLconst [j-16] w) mem))
  && x.Uses == 1
  && clobber(x)
  => (MOVLstore [i-2] {s} p w0 mem)

(MOVWstore [i] {s} p1 (SHRLconst [16] w) x:(MOVWstore [i] {s} p0 w mem))
  && x.Uses == 1
  && sequentialAddresses(p0, p1, 2)
  && clobber(x)
  => (MOVLstore [i] {s} p0 w mem)
(MOVWstore [i] {s} p1 (SHRLconst [j] w) x:(MOVWstore [i] {s} p0 w0:(SHRLconst [j-16] w) mem))
  && x.Uses == 1
  && sequentialAddresses(p0, p1, 2)
  && clobber(x)
  => (MOVLstore [i] {s} p0 w0 mem)

// For PIC, break floating-point constant loading into two instructions so we have
// a register to use for holding the address of the constant pool entry.
(MOVSSconst [c]) && config.ctxt.Flag_shared => (MOVSSconst2 (MOVSSconst1 [c]))
(MOVSDconst [c]) && config.ctxt.Flag_shared => (MOVSDconst2 (MOVSDconst1 [c]))

(CMP(L|W|B) l:(MOV(L|W|B)load {sym} [off] ptr mem) x) && canMergeLoad(v, l) && clobber(l) => (CMP(L|W|B)load {sym} [off] ptr x mem)
(CMP(L|W|B) x l:(MOV(L|W|B)load {sym} [off] ptr mem)) && canMergeLoad(v, l) && clobber(l) => (InvertFlags (CMP(L|W|B)load {sym} [off] ptr x mem))

(CMP(L|W|B)const l:(MOV(L|W|B)load {sym} [off] ptr mem) [c])
	&& l.Uses == 1
	&& clobber(l) =>
  @l.Block (CMP(L|W|B)constload {sym} [makeValAndOff(int32(c),off)] ptr mem)

(CMPLload {sym} [off] ptr (MOVLconst [c]) mem) => (CMPLconstload {sym} [makeValAndOff(c,off)] ptr mem)
(CMPWload {sym} [off] ptr (MOVLconst [c]) mem) => (CMPWconstload {sym} [makeValAndOff(int32(int16(c)),off)] ptr mem)
(CMPBload {sym} [off] ptr (MOVLconst [c]) mem) => (CMPBconstload {sym} [makeValAndOff(int32(int8(c)),off)] ptr mem)

(MOVBload [off] {sym} (SB) _) && symIsRO(sym) => (MOVLconst [int32(read8(sym, int64(off)))])
(MOVWload [off] {sym} (SB) _) && symIsRO(sym) => (MOVLconst [int32(read16(sym, int64(off), config.ctxt.Arch.ByteOrder))])
(MOVLload [off] {sym} (SB) _) && symIsRO(sym) => (MOVLconst [int32(read32(sym, int64(off), config.ctxt.Arch.ByteOrder))])