// 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|64|32|16|8) ...) => (ADD ...)
(Add(64|32)F ...) => (FADD(D|S) ...)

(Sub(Ptr|64|32|16|8) ...) => (SUB ...)
(Sub(64|32)F ...) => (FSUB(D|S) ...)

(Mul64 ...) => (MUL  ...)
(Mul64uhilo ...) => (LoweredMuluhilo ...)
(Mul64uover ...) => (LoweredMuluover ...)
(Mul32 ...) => (MULW ...)
(Mul16 x y) => (MULW (SignExt16to32 x) (SignExt16to32 y))
(Mul8 x y)  => (MULW (SignExt8to32 x)  (SignExt8to32 y))
(Mul(64|32)F ...) => (FMUL(D|S) ...)

(Div(64|32)F ...) => (FDIV(D|S) ...)

(Div64 x y [false])  => (DIV x y)
(Div64u ...) => (DIVU ...)
(Div32 x y [false])  => (DIVW x y)
(Div32u ...) => (DIVUW ...)
(Div16 x y [false])  => (DIVW  (SignExt16to32 x) (SignExt16to32 y))
(Div16u x y) => (DIVUW (ZeroExt16to32 x) (ZeroExt16to32 y))
(Div8 x y)   => (DIVW  (SignExt8to32 x)  (SignExt8to32 y))
(Div8u x y)  => (DIVUW (ZeroExt8to32 x)  (ZeroExt8to32 y))

(Hmul64 ...)  => (MULH  ...)
(Hmul64u ...) => (MULHU ...)
(Hmul32 x y)  => (SRAI [32] (MUL  (SignExt32to64 x) (SignExt32to64 y)))
(Hmul32u x y) => (SRLI [32] (MUL  (ZeroExt32to64 x) (ZeroExt32to64 y)))

(Select0 (Add64carry x y c)) => (ADD (ADD <typ.UInt64> x y) c)
(Select1 (Add64carry x y c)) =>
	(OR (SLTU <typ.UInt64> s:(ADD <typ.UInt64> x y) x) (SLTU <typ.UInt64> (ADD <typ.UInt64> s c) s))

(Select0 (Sub64borrow x y c)) => (SUB (SUB <typ.UInt64> x y) c)
(Select1 (Sub64borrow x y c)) =>
	(OR (SLTU <typ.UInt64> x s:(SUB <typ.UInt64> x y)) (SLTU <typ.UInt64> s (SUB <typ.UInt64> s c)))

// (x + y) / 2 => (x / 2) + (y / 2) + (x & y & 1)
(Avg64u <t> x y) => (ADD (ADD <t> (SRLI <t> [1] x) (SRLI <t> [1] y)) (ANDI <t> [1] (AND <t> x y)))

(Mod64 x y [false])  => (REM x y)
(Mod64u ...) => (REMU  ...)
(Mod32 x y [false])  => (REMW x y)
(Mod32u ...) => (REMUW ...)
(Mod16 x y [false])  => (REMW  (SignExt16to32 x) (SignExt16to32 y))
(Mod16u x y) => (REMUW (ZeroExt16to32 x) (ZeroExt16to32 y))
(Mod8 x y)   => (REMW  (SignExt8to32  x) (SignExt8to32  y))
(Mod8u x y)  => (REMUW (ZeroExt8to32  x) (ZeroExt8to32  y))

(And(64|32|16|8) ...) => (AND ...)
(Or(64|32|16|8) ...) => (OR ...)
(Xor(64|32|16|8) ...) => (XOR ...)

(Neg(64|32|16|8) ...) => (NEG ...)
(Neg(64|32)F ...) => (FNEG(D|S) ...)

(Com(64|32|16|8) ...) => (NOT ...)


(Sqrt ...) => (FSQRTD ...)
(Sqrt32 ...) => (FSQRTS ...)

(Copysign ...) => (FSGNJD ...)

(Abs ...) => (FABSD ...)

(FMA ...) => (FMADDD ...)

// Sign and zero extension.

(SignExt8to16  ...) => (MOVBreg ...)
(SignExt8to32  ...) => (MOVBreg ...)
(SignExt8to64  ...) => (MOVBreg ...)
(SignExt16to32 ...) => (MOVHreg ...)
(SignExt16to64 ...) => (MOVHreg ...)
(SignExt32to64 ...) => (MOVWreg ...)

(ZeroExt8to16  ...) => (MOVBUreg ...)
(ZeroExt8to32  ...) => (MOVBUreg ...)
(ZeroExt8to64  ...) => (MOVBUreg ...)
(ZeroExt16to32 ...) => (MOVHUreg ...)
(ZeroExt16to64 ...) => (MOVHUreg ...)
(ZeroExt32to64 ...) => (MOVWUreg ...)

(Cvt32to32F ...) => (FCVTSW ...)
(Cvt32to64F ...) => (FCVTDW ...)
(Cvt64to32F ...) => (FCVTSL ...)
(Cvt64to64F ...) => (FCVTDL ...)

(Cvt32Fto32 ...) => (FCVTWS ...)
(Cvt32Fto64 ...) => (FCVTLS ...)
(Cvt64Fto32 ...) => (FCVTWD ...)
(Cvt64Fto64 ...) => (FCVTLD ...)

(Cvt32Fto64F ...) => (FCVTDS ...)
(Cvt64Fto32F ...) => (FCVTSD ...)

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

(Round(32|64)F ...) => (LoweredRound(32|64)F ...)

(Slicemask <t> x) => (SRAI [63] (NEG <t> x))

// Truncations
// We ignore the unused high parts of registers, so truncates are just copies.
(Trunc16to8  ...) => (Copy ...)
(Trunc32to8  ...) => (Copy ...)
(Trunc32to16 ...) => (Copy ...)
(Trunc64to8  ...) => (Copy ...)
(Trunc64to16 ...) => (Copy ...)
(Trunc64to32 ...) => (Copy ...)

// Shifts

// SLL only considers the bottom 6 bits of y. If y > 64, the result should
// always be 0.
//
// Breaking down the operation:
//
// (SLL x y) generates x << (y & 63).
//
// If y < 64, this is the value we want. Otherwise, we want zero.
//
// So, we AND with -1 * uint64(y < 64), which is 0xfffff... if y < 64 and 0 otherwise.
(Lsh8x8   <t> x y) && !shiftIsBounded(v) => (AND (SLL <t> x y) (Neg8  <t> (SLTIU <t> [64] (ZeroExt8to64  y))))
(Lsh8x16  <t> x y) && !shiftIsBounded(v) => (AND (SLL <t> x y) (Neg8  <t> (SLTIU <t> [64] (ZeroExt16to64 y))))
(Lsh8x32  <t> x y) && !shiftIsBounded(v) => (AND (SLL <t> x y) (Neg8  <t> (SLTIU <t> [64] (ZeroExt32to64 y))))
(Lsh8x64  <t> x y) && !shiftIsBounded(v) => (AND (SLL <t> x y) (Neg8  <t> (SLTIU <t> [64] y)))
(Lsh16x8  <t> x y) && !shiftIsBounded(v) => (AND (SLL <t> x y) (Neg16 <t> (SLTIU <t> [64] (ZeroExt8to64  y))))
(Lsh16x16 <t> x y) && !shiftIsBounded(v) => (AND (SLL <t> x y) (Neg16 <t> (SLTIU <t> [64] (ZeroExt16to64 y))))
(Lsh16x32 <t> x y) && !shiftIsBounded(v) => (AND (SLL <t> x y) (Neg16 <t> (SLTIU <t> [64] (ZeroExt32to64 y))))
(Lsh16x64 <t> x y) && !shiftIsBounded(v) => (AND (SLL <t> x y) (Neg16 <t> (SLTIU <t> [64] y)))
(Lsh32x8  <t> x y) && !shiftIsBounded(v) => (AND (SLL <t> x y) (Neg32 <t> (SLTIU <t> [64] (ZeroExt8to64  y))))
(Lsh32x16 <t> x y) && !shiftIsBounded(v) => (AND (SLL <t> x y) (Neg32 <t> (SLTIU <t> [64] (ZeroExt16to64 y))))
(Lsh32x32 <t> x y) && !shiftIsBounded(v) => (AND (SLL <t> x y) (Neg32 <t> (SLTIU <t> [64] (ZeroExt32to64 y))))
(Lsh32x64 <t> x y) && !shiftIsBounded(v) => (AND (SLL <t> x y) (Neg32 <t> (SLTIU <t> [64] y)))
(Lsh64x8  <t> x y) && !shiftIsBounded(v) => (AND (SLL <t> x y) (Neg64 <t> (SLTIU <t> [64] (ZeroExt8to64  y))))
(Lsh64x16 <t> x y) && !shiftIsBounded(v) => (AND (SLL <t> x y) (Neg64 <t> (SLTIU <t> [64] (ZeroExt16to64 y))))
(Lsh64x32 <t> x y) && !shiftIsBounded(v) => (AND (SLL <t> x y) (Neg64 <t> (SLTIU <t> [64] (ZeroExt32to64 y))))
(Lsh64x64 <t> x y) && !shiftIsBounded(v) => (AND (SLL <t> x y) (Neg64 <t> (SLTIU <t> [64] y)))

(Lsh8x(64|32|16|8)  x y) && shiftIsBounded(v) => (SLL x y)
(Lsh16x(64|32|16|8) x y) && shiftIsBounded(v) => (SLL x y)
(Lsh32x(64|32|16|8) x y) && shiftIsBounded(v) => (SLL x y)
(Lsh64x(64|32|16|8) x y) && shiftIsBounded(v) => (SLL x y)

// SRL only considers the bottom 6 bits of y, similarly SRLW only considers the
// bottom 5 bits of y. Ensure that the result is always zero if the shift exceeds
// the maximum value. See Lsh above for a detailed description.
(Rsh8Ux8   <t> x y) && !shiftIsBounded(v) => (AND (SRL  <t> (ZeroExt8to64  x) y) (Neg8  <t> (SLTIU <t> [64] (ZeroExt8to64  y))))
(Rsh8Ux16  <t> x y) && !shiftIsBounded(v) => (AND (SRL  <t> (ZeroExt8to64  x) y) (Neg8  <t> (SLTIU <t> [64] (ZeroExt16to64 y))))
(Rsh8Ux32  <t> x y) && !shiftIsBounded(v) => (AND (SRL  <t> (ZeroExt8to64  x) y) (Neg8  <t> (SLTIU <t> [64] (ZeroExt32to64 y))))
(Rsh8Ux64  <t> x y) && !shiftIsBounded(v) => (AND (SRL  <t> (ZeroExt8to64  x) y) (Neg8  <t> (SLTIU <t> [64] y)))
(Rsh16Ux8  <t> x y) && !shiftIsBounded(v) => (AND (SRL  <t> (ZeroExt16to64 x) y) (Neg16 <t> (SLTIU <t> [64] (ZeroExt8to64  y))))
(Rsh16Ux16 <t> x y) && !shiftIsBounded(v) => (AND (SRL  <t> (ZeroExt16to64 x) y) (Neg16 <t> (SLTIU <t> [64] (ZeroExt16to64 y))))
(Rsh16Ux32 <t> x y) && !shiftIsBounded(v) => (AND (SRL  <t> (ZeroExt16to64 x) y) (Neg16 <t> (SLTIU <t> [64] (ZeroExt32to64 y))))
(Rsh16Ux64 <t> x y) && !shiftIsBounded(v) => (AND (SRL  <t> (ZeroExt16to64 x) y) (Neg16 <t> (SLTIU <t> [64] y)))
(Rsh32Ux8  <t> x y) && !shiftIsBounded(v) => (AND (SRLW <t>  x                y) (Neg32 <t> (SLTIU <t> [32] (ZeroExt8to64  y))))
(Rsh32Ux16 <t> x y) && !shiftIsBounded(v) => (AND (SRLW <t>  x                y) (Neg32 <t> (SLTIU <t> [32] (ZeroExt16to64 y))))
(Rsh32Ux32 <t> x y) && !shiftIsBounded(v) => (AND (SRLW <t>  x                y) (Neg32 <t> (SLTIU <t> [32] (ZeroExt32to64 y))))
(Rsh32Ux64 <t> x y) && !shiftIsBounded(v) => (AND (SRLW <t>  x                y) (Neg32 <t> (SLTIU <t> [32] y)))
(Rsh64Ux8  <t> x y) && !shiftIsBounded(v) => (AND (SRL  <t>  x                y) (Neg64 <t> (SLTIU <t> [64] (ZeroExt8to64  y))))
(Rsh64Ux16 <t> x y) && !shiftIsBounded(v) => (AND (SRL  <t>  x                y) (Neg64 <t> (SLTIU <t> [64] (ZeroExt16to64 y))))
(Rsh64Ux32 <t> x y) && !shiftIsBounded(v) => (AND (SRL  <t>  x                y) (Neg64 <t> (SLTIU <t> [64] (ZeroExt32to64 y))))
(Rsh64Ux64 <t> x y) && !shiftIsBounded(v) => (AND (SRL  <t>  x                y) (Neg64 <t> (SLTIU <t> [64] y)))

(Rsh8Ux(64|32|16|8)  x y) && shiftIsBounded(v) => (SRL  (ZeroExt8to64  x) y)
(Rsh16Ux(64|32|16|8) x y) && shiftIsBounded(v) => (SRL  (ZeroExt16to64 x) y)
(Rsh32Ux(64|32|16|8) x y) && shiftIsBounded(v) => (SRLW x                 y)
(Rsh64Ux(64|32|16|8) x y) && shiftIsBounded(v) => (SRL  x                 y)

// SRA only considers the bottom 6 bits of y, similarly SRAW only considers the
// bottom 5 bits. If y is greater than the maximum value (either 63 or 31
// depending on the instruction),  the result of the shift should be either 0
// or -1 based on the sign bit of x.
//
// We implement this by performing the max shift (-1) if y > the maximum value.
//
// We OR (uint64(y < 64) - 1) into y before passing it to SRA. This leaves
// us with -1 (0xffff...) if y >= 64.  Similarly, we OR (uint64(y < 32) - 1) into y
// before passing it to SRAW.
//
// We don't need to sign-extend the OR result, as it will be at minimum 8 bits,
// more than the 5 or 6 bits SRAW and SRA care about.
(Rsh8x8   <t> x y) && !shiftIsBounded(v) => (SRA  <t> (SignExt8to64  x) (OR <y.Type> y (ADDI <y.Type> [-1] (SLTIU <y.Type> [64] (ZeroExt8to64  y)))))
(Rsh8x16  <t> x y) && !shiftIsBounded(v) => (SRA  <t> (SignExt8to64  x) (OR <y.Type> y (ADDI <y.Type> [-1] (SLTIU <y.Type> [64] (ZeroExt16to64 y)))))
(Rsh8x32  <t> x y) && !shiftIsBounded(v) => (SRA  <t> (SignExt8to64  x) (OR <y.Type> y (ADDI <y.Type> [-1] (SLTIU <y.Type> [64] (ZeroExt32to64 y)))))
(Rsh8x64  <t> x y) && !shiftIsBounded(v) => (SRA  <t> (SignExt8to64  x) (OR <y.Type> y (ADDI <y.Type> [-1] (SLTIU <y.Type> [64] y))))
(Rsh16x8  <t> x y) && !shiftIsBounded(v) => (SRA  <t> (SignExt16to64 x) (OR <y.Type> y (ADDI <y.Type> [-1] (SLTIU <y.Type> [64] (ZeroExt8to64  y)))))
(Rsh16x16 <t> x y) && !shiftIsBounded(v) => (SRA  <t> (SignExt16to64 x) (OR <y.Type> y (ADDI <y.Type> [-1] (SLTIU <y.Type> [64] (ZeroExt16to64 y)))))
(Rsh16x32 <t> x y) && !shiftIsBounded(v) => (SRA  <t> (SignExt16to64 x) (OR <y.Type> y (ADDI <y.Type> [-1] (SLTIU <y.Type> [64] (ZeroExt32to64 y)))))
(Rsh16x64 <t> x y) && !shiftIsBounded(v) => (SRA  <t> (SignExt16to64 x) (OR <y.Type> y (ADDI <y.Type> [-1] (SLTIU <y.Type> [64] y))))
(Rsh32x8  <t> x y) && !shiftIsBounded(v) => (SRAW <t> x                 (OR <y.Type> y (ADDI <y.Type> [-1] (SLTIU <y.Type> [32] (ZeroExt8to64  y)))))
(Rsh32x16 <t> x y) && !shiftIsBounded(v) => (SRAW <t> x                 (OR <y.Type> y (ADDI <y.Type> [-1] (SLTIU <y.Type> [32] (ZeroExt16to64 y)))))
(Rsh32x32 <t> x y) && !shiftIsBounded(v) => (SRAW <t> x                 (OR <y.Type> y (ADDI <y.Type> [-1] (SLTIU <y.Type> [32] (ZeroExt32to64 y)))))
(Rsh32x64 <t> x y) && !shiftIsBounded(v) => (SRAW <t> x                 (OR <y.Type> y (ADDI <y.Type> [-1] (SLTIU <y.Type> [32] y))))
(Rsh64x8  <t> x y) && !shiftIsBounded(v) => (SRA  <t> x                 (OR <y.Type> y (ADDI <y.Type> [-1] (SLTIU <y.Type> [64] (ZeroExt8to64  y)))))
(Rsh64x16 <t> x y) && !shiftIsBounded(v) => (SRA  <t> x                 (OR <y.Type> y (ADDI <y.Type> [-1] (SLTIU <y.Type> [64] (ZeroExt16to64 y)))))
(Rsh64x32 <t> x y) && !shiftIsBounded(v) => (SRA  <t> x                 (OR <y.Type> y (ADDI <y.Type> [-1] (SLTIU <y.Type> [64] (ZeroExt32to64 y)))))
(Rsh64x64 <t> x y) && !shiftIsBounded(v) => (SRA  <t> x                 (OR <y.Type> y (ADDI <y.Type> [-1] (SLTIU <y.Type> [64] y))))

(Rsh8x(64|32|16|8)  x y) && shiftIsBounded(v) => (SRA  (SignExt8to64  x) y)
(Rsh16x(64|32|16|8) x y) && shiftIsBounded(v) => (SRA  (SignExt16to64 x) y)
(Rsh32x(64|32|16|8) x y) && shiftIsBounded(v) => (SRAW  x                y)
(Rsh64x(64|32|16|8) x y) && shiftIsBounded(v) => (SRA   x                y)

// Rotates.
(RotateLeft8  <t> x (MOVDconst [c])) => (Or8  (Lsh8x64  <t> x (MOVDconst [c&7]))  (Rsh8Ux64  <t> x (MOVDconst [-c&7])))
(RotateLeft16 <t> x (MOVDconst [c])) => (Or16 (Lsh16x64 <t> x (MOVDconst [c&15])) (Rsh16Ux64 <t> x (MOVDconst [-c&15])))
(RotateLeft32 <t> x (MOVDconst [c])) => (Or32 (Lsh32x64 <t> x (MOVDconst [c&31])) (Rsh32Ux64 <t> x (MOVDconst [-c&31])))
(RotateLeft64 <t> x (MOVDconst [c])) => (Or64 (Lsh64x64 <t> x (MOVDconst [c&63])) (Rsh64Ux64 <t> x (MOVDconst [-c&63])))

(Less64  ...) => (SLT  ...)
(Less32  x y) => (SLT  (SignExt32to64 x) (SignExt32to64 y))
(Less16  x y) => (SLT  (SignExt16to64 x) (SignExt16to64 y))
(Less8   x y) => (SLT  (SignExt8to64  x) (SignExt8to64  y))
(Less64U ...) => (SLTU ...)
(Less32U x y) => (SLTU (ZeroExt32to64 x) (ZeroExt32to64 y))
(Less16U x y) => (SLTU (ZeroExt16to64 x) (ZeroExt16to64 y))
(Less8U  x y) => (SLTU (ZeroExt8to64  x) (ZeroExt8to64  y))
(Less(64|32)F ...) => (FLT(D|S) ...)

// Convert x <= y to !(y > x).
(Leq(64|32|16|8)  x y) => (Not (Less(64|32|16|8)  y x))
(Leq(64|32|16|8)U x y) => (Not (Less(64|32|16|8)U y x))
(Leq(64|32)F ...) => (FLE(D|S) ...)

(EqPtr x y) => (SEQZ (SUB <typ.Uintptr> x y))
(Eq64  x y) => (SEQZ (SUB <x.Type> x y))
(Eq32  x y) &&  x.Type.IsSigned() => (SEQZ (SUB <x.Type> (SignExt32to64 x) (SignExt32to64 y)))
(Eq32  x y) && !x.Type.IsSigned() => (SEQZ (SUB <x.Type> (ZeroExt32to64 x) (ZeroExt32to64 y)))
(Eq16  x y) => (SEQZ (SUB <x.Type> (ZeroExt16to64 x) (ZeroExt16to64 y)))
(Eq8   x y) => (SEQZ (SUB <x.Type> (ZeroExt8to64  x) (ZeroExt8to64  y)))
(Eq(64|32)F ...) => (FEQ(D|S) ...)

(NeqPtr x y) => (Not (EqPtr x y))
(Neq64  x y) => (Not (Eq64  x y))
(Neq32  x y) => (Not (Eq32  x y))
(Neq16  x y) => (Not (Eq16  x y))
(Neq8   x y) => (Not (Eq8   x y))
(Neq(64|32)F ...) => (FNE(D|S) ...)

// Loads
(Load <t> ptr mem) &&  t.IsBoolean()                   => (MOVBUload ptr mem)
(Load <t> ptr mem) && ( is8BitInt(t) &&  t.IsSigned()) => (MOVBload  ptr mem)
(Load <t> ptr mem) && ( is8BitInt(t) && !t.IsSigned()) => (MOVBUload ptr mem)
(Load <t> ptr mem) && (is16BitInt(t) &&  t.IsSigned()) => (MOVHload  ptr mem)
(Load <t> ptr mem) && (is16BitInt(t) && !t.IsSigned()) => (MOVHUload ptr mem)
(Load <t> ptr mem) && (is32BitInt(t) &&  t.IsSigned()) => (MOVWload  ptr mem)
(Load <t> ptr mem) && (is32BitInt(t) && !t.IsSigned()) => (MOVWUload ptr mem)
(Load <t> ptr mem) && (is64BitInt(t) || isPtr(t))      => (MOVDload  ptr mem)
(Load <t> ptr mem) &&  is32BitFloat(t)                 => (FMOVWload ptr mem)
(Load <t> ptr mem) &&  is64BitFloat(t)                 => (FMOVDload ptr mem)

// Stores
(Store {t} ptr val mem) && t.Size() == 1 => (MOVBstore ptr val mem)
(Store {t} ptr val mem) && t.Size() == 2 => (MOVHstore ptr val mem)
(Store {t} ptr val mem) && t.Size() == 4 && !t.IsFloat() => (MOVWstore ptr val mem)
(Store {t} ptr val mem) && t.Size() == 8 && !t.IsFloat() => (MOVDstore ptr val mem)
(Store {t} ptr val mem) && t.Size() == 4 &&  t.IsFloat() => (FMOVWstore ptr val mem)
(Store {t} ptr val mem) && t.Size() == 8 &&  t.IsFloat() => (FMOVDstore ptr val mem)

// We need to fold MOVaddr into the LD/MOVDstore ops so that the live variable analysis
// knows what variables are being read/written by the ops.
(MOVBUload [off1] {sym1} (MOVaddr [off2] {sym2} base) mem) && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) =>
	(MOVBUload [off1+off2] {mergeSym(sym1,sym2)} base mem)
(MOVBload  [off1] {sym1} (MOVaddr [off2] {sym2} base) mem) && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) =>
	(MOVBload  [off1+off2] {mergeSym(sym1,sym2)} base mem)
(MOVHUload [off1] {sym1} (MOVaddr [off2] {sym2} base) mem) && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) =>
	(MOVHUload [off1+off2] {mergeSym(sym1,sym2)} base mem)
(MOVHload  [off1] {sym1} (MOVaddr [off2] {sym2} base) mem) && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) =>
	(MOVHload  [off1+off2] {mergeSym(sym1,sym2)} base mem)
(MOVWUload [off1] {sym1} (MOVaddr [off2] {sym2} base) mem) && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) =>
	(MOVWUload [off1+off2] {mergeSym(sym1,sym2)} base mem)
(MOVWload  [off1] {sym1} (MOVaddr [off2] {sym2} base) mem) && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) =>
	(MOVWload  [off1+off2] {mergeSym(sym1,sym2)} base mem)
(MOVDload  [off1] {sym1} (MOVaddr [off2] {sym2} base) mem) && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) =>
	(MOVDload  [off1+off2] {mergeSym(sym1,sym2)} base mem)

(MOVBstore [off1] {sym1} (MOVaddr [off2] {sym2} base) val mem) && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) =>
	(MOVBstore [off1+off2] {mergeSym(sym1,sym2)} base val mem)
(MOVHstore [off1] {sym1} (MOVaddr [off2] {sym2} base) val mem) && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) =>
	(MOVHstore [off1+off2] {mergeSym(sym1,sym2)} base val mem)
(MOVWstore [off1] {sym1} (MOVaddr [off2] {sym2} base) val mem) && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) =>
	(MOVWstore [off1+off2] {mergeSym(sym1,sym2)} base val mem)
(MOVDstore [off1] {sym1} (MOVaddr [off2] {sym2} base) val mem) && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) =>
	(MOVDstore [off1+off2] {mergeSym(sym1,sym2)} base val mem)
(MOVBstorezero [off1] {sym1} (MOVaddr [off2] {sym2} ptr) mem) && canMergeSym(sym1,sym2) && is32Bit(int64(off1)+int64(off2)) =>
	(MOVBstorezero [off1+off2] {mergeSym(sym1,sym2)} ptr mem)
(MOVHstorezero [off1] {sym1} (MOVaddr [off2] {sym2} ptr) mem) && canMergeSym(sym1,sym2) && is32Bit(int64(off1)+int64(off2)) =>
	(MOVHstorezero [off1+off2] {mergeSym(sym1,sym2)} ptr mem)
(MOVWstorezero [off1] {sym1} (MOVaddr [off2] {sym2} ptr) mem) && canMergeSym(sym1,sym2) && is32Bit(int64(off1)+int64(off2)) =>
	(MOVWstorezero [off1+off2] {mergeSym(sym1,sym2)} ptr mem)
(MOVDstorezero [off1] {sym1} (MOVaddr [off2] {sym2} ptr) mem) && canMergeSym(sym1,sym2) && is32Bit(int64(off1)+int64(off2)) =>
	(MOVDstorezero [off1+off2] {mergeSym(sym1,sym2)} ptr mem)

(MOVBUload [off1] {sym} (ADDI [off2] base) mem) && is32Bit(int64(off1)+off2) =>
	(MOVBUload [off1+int32(off2)] {sym} base mem)
(MOVBload  [off1] {sym} (ADDI [off2] base) mem) && is32Bit(int64(off1)+off2) =>
	(MOVBload  [off1+int32(off2)] {sym} base mem)
(MOVHUload [off1] {sym} (ADDI [off2] base) mem) && is32Bit(int64(off1)+off2) =>
	(MOVHUload [off1+int32(off2)] {sym} base mem)
(MOVHload  [off1] {sym} (ADDI [off2] base) mem) && is32Bit(int64(off1)+off2) =>
	(MOVHload  [off1+int32(off2)] {sym} base mem)
(MOVWUload [off1] {sym} (ADDI [off2] base) mem) && is32Bit(int64(off1)+off2) =>
	(MOVWUload [off1+int32(off2)] {sym} base mem)
(MOVWload  [off1] {sym} (ADDI [off2] base) mem) && is32Bit(int64(off1)+off2) =>
	(MOVWload  [off1+int32(off2)] {sym} base mem)
(MOVDload  [off1] {sym} (ADDI [off2] base) mem) && is32Bit(int64(off1)+off2) =>
	(MOVDload  [off1+int32(off2)] {sym} base mem)

(MOVBstore [off1] {sym} (ADDI [off2] base) val mem) && is32Bit(int64(off1)+off2) =>
	(MOVBstore [off1+int32(off2)] {sym} base val mem)
(MOVHstore [off1] {sym} (ADDI [off2] base) val mem) && is32Bit(int64(off1)+off2) =>
	(MOVHstore [off1+int32(off2)] {sym} base val mem)
(MOVWstore [off1] {sym} (ADDI [off2] base) val mem) && is32Bit(int64(off1)+off2) =>
	(MOVWstore [off1+int32(off2)] {sym} base val mem)
(MOVDstore [off1] {sym} (ADDI [off2] base) val mem) && is32Bit(int64(off1)+off2) =>
	(MOVDstore [off1+int32(off2)] {sym} base val mem)
(MOVBstorezero [off1] {sym} (ADDI [off2] ptr) mem) && is32Bit(int64(off1)+off2) => (MOVBstorezero [off1+int32(off2)] {sym} ptr mem)
(MOVHstorezero [off1] {sym} (ADDI [off2] ptr) mem) && is32Bit(int64(off1)+off2) => (MOVHstorezero [off1+int32(off2)] {sym} ptr mem)
(MOVWstorezero [off1] {sym} (ADDI [off2] ptr) mem) && is32Bit(int64(off1)+off2) => (MOVWstorezero [off1+int32(off2)] {sym} ptr mem)
(MOVDstorezero [off1] {sym} (ADDI [off2] ptr) mem) && is32Bit(int64(off1)+off2) => (MOVDstorezero [off1+int32(off2)] {sym} ptr mem)

// Similarly, fold ADDI into MOVaddr to avoid confusing live variable analysis
// with OffPtr -> ADDI.
(ADDI [c] (MOVaddr [d] {s} x)) && is32Bit(c+int64(d)) => (MOVaddr [int32(c)+d] {s} x)

// Small zeroing
(Zero [0] _ mem) => mem
(Zero [1] ptr mem) => (MOVBstore ptr (MOVDconst [0]) mem)
(Zero [2] {t} ptr mem) && t.Alignment()%2 == 0 =>
	(MOVHstore ptr (MOVDconst [0]) mem)
(Zero [2] ptr mem) =>
	(MOVBstore [1] ptr (MOVDconst [0])
		(MOVBstore ptr (MOVDconst [0]) mem))
(Zero [4] {t} ptr mem) && t.Alignment()%4 == 0 =>
	(MOVWstore ptr (MOVDconst [0]) mem)
(Zero [4] {t} ptr mem) && t.Alignment()%2 == 0 =>
	(MOVHstore [2] ptr (MOVDconst [0])
		(MOVHstore ptr (MOVDconst [0]) mem))
(Zero [4] ptr mem) =>
	(MOVBstore [3] ptr (MOVDconst [0])
		(MOVBstore [2] ptr (MOVDconst [0])
			(MOVBstore [1] ptr (MOVDconst [0])
				(MOVBstore ptr (MOVDconst [0]) mem))))
(Zero [8] {t} ptr mem) && t.Alignment()%8 == 0 =>
	(MOVDstore ptr (MOVDconst [0]) mem)
(Zero [8] {t} ptr mem) && t.Alignment()%4 == 0 =>
	(MOVWstore [4] ptr (MOVDconst [0])
		(MOVWstore ptr (MOVDconst [0]) mem))
(Zero [8] {t} ptr mem) && t.Alignment()%2 == 0 =>
	(MOVHstore [6] ptr (MOVDconst [0])
		(MOVHstore [4] ptr (MOVDconst [0])
			(MOVHstore [2] ptr (MOVDconst [0])
				(MOVHstore ptr (MOVDconst [0]) mem))))

(Zero [3] ptr mem) =>
	(MOVBstore [2] ptr (MOVDconst [0])
		(MOVBstore [1] ptr (MOVDconst [0])
			(MOVBstore ptr (MOVDconst [0]) mem)))
(Zero [6] {t} ptr mem) && t.Alignment()%2 == 0 =>
	(MOVHstore [4] ptr (MOVDconst [0])
		(MOVHstore [2] ptr (MOVDconst [0])
			(MOVHstore ptr (MOVDconst [0]) mem)))
(Zero [12] {t} ptr mem) && t.Alignment()%4 == 0 =>
	(MOVWstore [8] ptr (MOVDconst [0])
		(MOVWstore [4] ptr (MOVDconst [0])
			(MOVWstore ptr (MOVDconst [0]) mem)))
(Zero [16] {t} ptr mem) && t.Alignment()%8 == 0 =>
	(MOVDstore [8] ptr (MOVDconst [0])
		(MOVDstore ptr (MOVDconst [0]) mem))
(Zero [24] {t} ptr mem) && t.Alignment()%8 == 0 =>
	(MOVDstore [16] ptr (MOVDconst [0])
		(MOVDstore [8] ptr (MOVDconst [0])
			(MOVDstore ptr (MOVDconst [0]) mem)))
(Zero [32] {t} ptr mem) && t.Alignment()%8 == 0 =>
	(MOVDstore [24] ptr (MOVDconst [0])
		(MOVDstore [16] ptr (MOVDconst [0])
			(MOVDstore [8] ptr (MOVDconst [0])
				(MOVDstore ptr (MOVDconst [0]) mem))))

// Medium 8-aligned zeroing uses a Duff's device
// 8 and 128 are magic constants, see runtime/mkduff.go
(Zero [s] {t} ptr mem)
	&& s%8 == 0 && s <= 8*128
	&& t.Alignment()%8 == 0 && !config.noDuffDevice =>
	(DUFFZERO [8 * (128 - s/8)] ptr mem)

// Generic zeroing uses a loop
(Zero [s] {t} ptr mem) =>
	(LoweredZero [t.Alignment()]
		ptr
		(ADD <ptr.Type> ptr (MOVDconst [s-moveSize(t.Alignment(), config)]))
		mem)

// Checks
(IsNonNil ...) => (SNEZ ...)
(IsInBounds ...) => (Less64U ...)
(IsSliceInBounds ...) => (Leq64U ...)

// Trivial lowering
(NilCheck ...) => (LoweredNilCheck ...)
(GetClosurePtr ...) => (LoweredGetClosurePtr ...)
(GetCallerSP ...) => (LoweredGetCallerSP ...)
(GetCallerPC ...) => (LoweredGetCallerPC ...)

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

// Publication barrier as intrinsic
(PubBarrier ...) => (LoweredPubBarrier ...)

(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)

// Small moves
(Move [0] _ _ mem) => mem
(Move [1] dst src mem) => (MOVBstore dst (MOVBload src mem) mem)
(Move [2] {t} dst src mem) && t.Alignment()%2 == 0 =>
	(MOVHstore dst (MOVHload src mem) mem)
(Move [2] dst src mem) =>
	(MOVBstore [1] dst (MOVBload [1] src mem)
		(MOVBstore dst (MOVBload src mem) mem))
(Move [4] {t} dst src mem) && t.Alignment()%4 == 0 =>
	(MOVWstore dst (MOVWload src mem) mem)
(Move [4] {t} dst src mem) && t.Alignment()%2 == 0 =>
	(MOVHstore [2] dst (MOVHload [2] src mem)
		(MOVHstore dst (MOVHload src mem) mem))
(Move [4] dst src mem) =>
	(MOVBstore [3] dst (MOVBload [3] src mem)
		(MOVBstore [2] dst (MOVBload [2] src mem)
			(MOVBstore [1] dst (MOVBload [1] src mem)
				(MOVBstore dst (MOVBload src mem) mem))))
(Move [8] {t} dst src mem) && t.Alignment()%8 == 0 =>
	(MOVDstore dst (MOVDload src mem) mem)
(Move [8] {t} dst src mem) && t.Alignment()%4 == 0 =>
	(MOVWstore [4] dst (MOVWload [4] src mem)
		(MOVWstore dst (MOVWload src mem) mem))
(Move [8] {t} dst src mem) && t.Alignment()%2 == 0 =>
	(MOVHstore [6] dst (MOVHload [6] src mem)
		(MOVHstore [4] dst (MOVHload [4] src mem)
			(MOVHstore [2] dst (MOVHload [2] src mem)
				(MOVHstore dst (MOVHload src mem) mem))))

(Move [3] dst src mem) =>
	(MOVBstore [2] dst (MOVBload [2] src mem)
		(MOVBstore [1] dst (MOVBload [1] src mem)
			(MOVBstore dst (MOVBload src mem) mem)))
(Move [6] {t} dst src mem) && t.Alignment()%2 == 0 =>
	(MOVHstore [4] dst (MOVHload [4] src mem)
		(MOVHstore [2] dst (MOVHload [2] src mem)
			(MOVHstore dst (MOVHload src mem) mem)))
(Move [12] {t} dst src mem) && t.Alignment()%4 == 0 =>
	(MOVWstore [8] dst (MOVWload [8] src mem)
		(MOVWstore [4] dst (MOVWload [4] src mem)
			(MOVWstore dst (MOVWload src mem) mem)))
(Move [16] {t} dst src mem) && t.Alignment()%8 == 0 =>
	(MOVDstore [8] dst (MOVDload [8] src mem)
		(MOVDstore dst (MOVDload src mem) mem))
(Move [24] {t} dst src mem) && t.Alignment()%8 == 0 =>
	(MOVDstore [16] dst (MOVDload [16] src mem)
		(MOVDstore [8] dst (MOVDload [8] src mem)
			(MOVDstore dst (MOVDload src mem) mem)))
(Move [32] {t} dst src mem) && t.Alignment()%8 == 0 =>
	(MOVDstore [24] dst (MOVDload [24] src mem)
		(MOVDstore [16] dst (MOVDload [16] src mem)
			(MOVDstore [8] dst (MOVDload [8] src mem)
				(MOVDstore dst (MOVDload src mem) mem))))

// Medium 8-aligned move uses a Duff's device
// 16 and 128 are magic constants, see runtime/mkduff.go
(Move [s] {t} dst src mem)
	&& s%8 == 0 && s <= 8*128 && t.Alignment()%8 == 0
	&& !config.noDuffDevice && logLargeCopy(v, s) =>
	(DUFFCOPY [16 * (128 - s/8)] dst src mem)

// Generic move uses a loop
(Move [s] {t} dst src mem) && (s <= 16 || logLargeCopy(v, s)) =>
	(LoweredMove [t.Alignment()]
		dst
		src
		(ADDI <src.Type> [s-moveSize(t.Alignment(), config)] src)
		mem)

// Boolean ops; 0=false, 1=true
(AndB ...) => (AND ...)
(OrB  ...) => (OR  ...)
(EqB  x y) => (SEQZ (SUB <typ.Bool> x y))
(NeqB x y) => (SNEZ (SUB <typ.Bool> x y))
(Not  ...) => (SEQZ ...)

// Lowering pointer arithmetic
// TODO: Special handling for SP offsets, like ARM
(OffPtr [off] ptr:(SP)) && is32Bit(off) => (MOVaddr [int32(off)] ptr)
(OffPtr [off] ptr) && is32Bit(off) => (ADDI [off] ptr)
(OffPtr [off] ptr) => (ADD (MOVDconst [off]) ptr)

(Const(64|32|16|8) [val]) => (MOVDconst [int64(val)])
(Const32F [val]) => (FMVSX (MOVDconst [int64(math.Float32bits(val))]))
(Const64F [val]) => (FMVDX (MOVDconst [int64(math.Float64bits(val))]))
(ConstNil) => (MOVDconst [0])
(ConstBool [val]) => (MOVDconst [int64(b2i(val))])

(Addr {sym} base) => (MOVaddr {sym} [0] base)
(LocalAddr <t> {sym} base mem) && t.Elem().HasPointers() => (MOVaddr {sym} (SPanchored base mem))
(LocalAddr <t> {sym} base _)  && !t.Elem().HasPointers() => (MOVaddr {sym} base)

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

// Atomic Intrinsics
(AtomicLoad(Ptr|64|32|8)  ...) => (LoweredAtomicLoad(64|64|32|8) ...)
(AtomicStore(PtrNoWB|64|32|8) ...) => (LoweredAtomicStore(64|64|32|8) ...)
(AtomicAdd(64|32) ...) => (LoweredAtomicAdd(64|32) ...)

// AtomicAnd8(ptr,val) => LoweredAtomicAnd32(ptr&^3, ^((uint8(val) ^ 0xff) << ((ptr & 3) * 8)))
(AtomicAnd8 ptr val mem) =>
	(LoweredAtomicAnd32 (ANDI <typ.Uintptr> [^3] ptr)
		(NOT <typ.UInt32> (SLL <typ.UInt32> (XORI <typ.UInt32> [0xff] (ZeroExt8to32 val))
			(SLLI <typ.UInt64> [3] (ANDI <typ.UInt64> [3] ptr)))) mem)

(AtomicAnd32 ...) => (LoweredAtomicAnd32 ...)

(AtomicCompareAndSwap32 ptr old new mem) => (LoweredAtomicCas32 ptr (SignExt32to64 old) new mem)
(AtomicCompareAndSwap64 ...) => (LoweredAtomicCas64 ...)

(AtomicExchange(64|32) ...) => (LoweredAtomicExchange(64|32) ...)

// AtomicOr8(ptr,val)  => LoweredAtomicOr32(ptr&^3, uint32(val)<<((ptr&3)*8))
(AtomicOr8 ptr val mem) =>
	(LoweredAtomicOr32 (ANDI <typ.Uintptr> [^3] ptr)
		(SLL <typ.UInt32> (ZeroExt8to32 val)
			(SLLI <typ.UInt64> [3] (ANDI <typ.UInt64> [3] ptr))) mem)

(AtomicOr32  ...) => (LoweredAtomicOr32  ...)

// Conditional branches
(If cond yes no) => (BNEZ (MOVBUreg <typ.UInt64> cond) yes no)

// Optimizations

// Absorb SEQZ/SNEZ into branch.
(BEQZ (SEQZ x) yes no) => (BNEZ x yes no)
(BEQZ (SNEZ x) yes no) => (BEQZ x yes no)
(BNEZ (SEQZ x) yes no) => (BEQZ x yes no)
(BNEZ (SNEZ x) yes no) => (BNEZ x yes no)

// Remove redundant NEG from BEQZ/BNEZ.
(BEQZ (NEG x) yes no) => (BEQZ x yes no)
(BNEZ (NEG x) yes no) => (BNEZ x yes no)

// Negate comparison with FNES/FNED.
(BEQZ (FNES <t> x y) yes no) => (BNEZ (FEQS <t> x y) yes no)
(BNEZ (FNES <t> x y) yes no) => (BEQZ (FEQS <t> x y) yes no)
(BEQZ (FNED <t> x y) yes no) => (BNEZ (FEQD <t> x y) yes no)
(BNEZ (FNED <t> x y) yes no) => (BEQZ (FEQD <t> x y) yes no)

// Convert BEQZ/BNEZ into more optimal branch conditions.
(BEQZ (SUB x y) yes no) => (BEQ x y yes no)
(BNEZ (SUB x y) yes no) => (BNE x y yes no)
(BEQZ (SLT x y) yes no) => (BGE x y yes no)
(BNEZ (SLT x y) yes no) => (BLT x y yes no)
(BEQZ (SLTU x y) yes no) => (BGEU x y yes no)
(BNEZ (SLTU x y) yes no) => (BLTU x y yes no)
(BEQZ (SLTI [x] y) yes no) => (BGE y (MOVDconst [x]) yes no)
(BNEZ (SLTI [x] y) yes no) => (BLT y (MOVDconst [x]) yes no)
(BEQZ (SLTIU [x] y) yes no) => (BGEU y (MOVDconst [x]) yes no)
(BNEZ (SLTIU [x] y) yes no) => (BLTU y (MOVDconst [x]) yes no)

// Convert branch with zero to more optimal branch zero.
(BEQ (MOVDconst [0]) cond yes no) => (BEQZ cond yes no)
(BEQ cond (MOVDconst [0]) yes no) => (BEQZ cond yes no)
(BNE (MOVDconst [0]) cond yes no) => (BNEZ cond yes no)
(BNE cond (MOVDconst [0]) yes no) => (BNEZ cond yes no)
(BLT (MOVDconst [0]) cond yes no) => (BGTZ cond yes no)
(BLT cond (MOVDconst [0]) yes no) => (BLTZ cond yes no)
(BGE (MOVDconst [0]) cond yes no) => (BLEZ cond yes no)
(BGE cond (MOVDconst [0]) yes no) => (BGEZ cond yes no)

// Remove redundant NEG from SEQZ/SNEZ.
(SEQZ (NEG x)) => (SEQZ x)
(SNEZ (NEG x)) => (SNEZ x)

// Remove redundant SEQZ/SNEZ.
(SEQZ (SEQZ x)) => (SNEZ x)
(SEQZ (SNEZ x)) => (SEQZ x)
(SNEZ (SEQZ x)) => (SEQZ x)
(SNEZ (SNEZ x)) => (SNEZ x)

// Store zero.
(MOVBstore [off] {sym} ptr (MOVDconst [0]) mem) => (MOVBstorezero [off] {sym} ptr mem)
(MOVHstore [off] {sym} ptr (MOVDconst [0]) mem) => (MOVHstorezero [off] {sym} ptr mem)
(MOVWstore [off] {sym} ptr (MOVDconst [0]) mem) => (MOVWstorezero [off] {sym} ptr mem)
(MOVDstore [off] {sym} ptr (MOVDconst [0]) mem) => (MOVDstorezero [off] {sym} ptr mem)

// Boolean ops are already extended.
(MOVBUreg x:((FLES|FLTS|FEQS|FNES) _ _)) => x
(MOVBUreg x:((FLED|FLTD|FEQD|FNED) _ _)) => x
(MOVBUreg x:((SEQZ|SNEZ) _)) => x
(MOVBUreg x:((SLT|SLTU) _ _)) => x

// Avoid extending when already sufficiently masked.
(MOVBreg  x:(ANDI [c] y)) && c >= 0 && int64(int8(c)) == c => x
(MOVHreg  x:(ANDI [c] y)) && c >= 0 && int64(int16(c)) == c => x
(MOVWreg  x:(ANDI [c] y)) && c >= 0 && int64(int32(c)) == c => x
(MOVBUreg x:(ANDI [c] y)) && c >= 0 && int64(uint8(c)) == c => x
(MOVHUreg x:(ANDI [c] y)) && c >= 0 && int64(uint16(c)) == c => x
(MOVWUreg x:(ANDI [c] y)) && c >= 0 && int64(uint32(c)) == c => x

// Combine masking and zero extension.
(MOVBUreg (ANDI [c] x)) && c < 0 => (ANDI [int64(uint8(c))] x)
(MOVHUreg (ANDI [c] x)) && c < 0 => (ANDI [int64(uint16(c))] x)
(MOVWUreg (ANDI [c] x)) && c < 0 => (AND (MOVDconst [int64(uint32(c))]) x)

// Avoid sign/zero extension for consts.
(MOVBreg  (MOVDconst [c])) => (MOVDconst [int64(int8(c))])
(MOVHreg  (MOVDconst [c])) => (MOVDconst [int64(int16(c))])
(MOVWreg  (MOVDconst [c])) => (MOVDconst [int64(int32(c))])
(MOVBUreg (MOVDconst [c])) => (MOVDconst [int64(uint8(c))])
(MOVHUreg (MOVDconst [c])) => (MOVDconst [int64(uint16(c))])
(MOVWUreg (MOVDconst [c])) => (MOVDconst [int64(uint32(c))])

// Avoid sign/zero extension after properly typed load.
(MOVBreg  x:(MOVBload  _ _)) => (MOVDreg x)
(MOVHreg  x:(MOVBload  _ _)) => (MOVDreg x)
(MOVHreg  x:(MOVBUload _ _)) => (MOVDreg x)
(MOVHreg  x:(MOVHload  _ _)) => (MOVDreg x)
(MOVWreg  x:(MOVBload  _ _)) => (MOVDreg x)
(MOVWreg  x:(MOVBUload _ _)) => (MOVDreg x)
(MOVWreg  x:(MOVHload  _ _)) => (MOVDreg x)
(MOVWreg  x:(MOVHUload _ _)) => (MOVDreg x)
(MOVWreg  x:(MOVWload  _ _)) => (MOVDreg x)
(MOVBUreg x:(MOVBUload _ _)) => (MOVDreg x)
(MOVHUreg x:(MOVBUload _ _)) => (MOVDreg x)
(MOVHUreg x:(MOVHUload _ _)) => (MOVDreg x)
(MOVWUreg x:(MOVBUload _ _)) => (MOVDreg x)
(MOVWUreg x:(MOVHUload _ _)) => (MOVDreg x)
(MOVWUreg x:(MOVWUload _ _)) => (MOVDreg x)

// Avoid zero extension after properly typed atomic operation.
(MOVBUreg x:(Select0 (LoweredAtomicLoad8 _ _))) => (MOVDreg x)
(MOVBUreg x:(Select0 (LoweredAtomicCas32 _ _ _ _))) => (MOVDreg x)
(MOVBUreg x:(Select0 (LoweredAtomicCas64 _ _ _ _))) => (MOVDreg x)

// Avoid sign extension after word arithmetic.
(MOVWreg x:(ADDIW   _)) => (MOVDreg x)
(MOVWreg x:(SUBW  _ _)) => (MOVDreg x)
(MOVWreg x:(NEGW    _)) => (MOVDreg x)
(MOVWreg x:(MULW  _ _)) => (MOVDreg x)
(MOVWreg x:(DIVW  _ _)) => (MOVDreg x)
(MOVWreg x:(DIVUW _ _)) => (MOVDreg x)
(MOVWreg x:(REMW  _ _)) => (MOVDreg x)
(MOVWreg x:(REMUW _ _)) => (MOVDreg x)

// Fold double extensions.
(MOVBreg  x:(MOVBreg  _)) => (MOVDreg x)
(MOVHreg  x:(MOVBreg  _)) => (MOVDreg x)
(MOVHreg  x:(MOVBUreg _)) => (MOVDreg x)
(MOVHreg  x:(MOVHreg  _)) => (MOVDreg x)
(MOVWreg  x:(MOVBreg  _)) => (MOVDreg x)
(MOVWreg  x:(MOVBUreg _)) => (MOVDreg x)
(MOVWreg  x:(MOVHreg  _)) => (MOVDreg x)
(MOVWreg  x:(MOVWreg  _)) => (MOVDreg x)
(MOVBUreg x:(MOVBUreg _)) => (MOVDreg x)
(MOVHUreg x:(MOVBUreg _)) => (MOVDreg x)
(MOVHUreg x:(MOVHUreg _)) => (MOVDreg x)
(MOVWUreg x:(MOVBUreg _)) => (MOVDreg x)
(MOVWUreg x:(MOVHUreg _)) => (MOVDreg x)
(MOVWUreg x:(MOVWUreg _)) => (MOVDreg x)

// Do not extend before store.
(MOVBstore [off] {sym} ptr (MOVBreg  x) mem) => (MOVBstore [off] {sym} ptr x mem)
(MOVBstore [off] {sym} ptr (MOVHreg  x) mem) => (MOVBstore [off] {sym} ptr x mem)
(MOVBstore [off] {sym} ptr (MOVWreg  x) mem) => (MOVBstore [off] {sym} ptr x mem)
(MOVBstore [off] {sym} ptr (MOVBUreg x) mem) => (MOVBstore [off] {sym} ptr x mem)
(MOVBstore [off] {sym} ptr (MOVHUreg x) mem) => (MOVBstore [off] {sym} ptr x mem)
(MOVBstore [off] {sym} ptr (MOVWUreg x) mem) => (MOVBstore [off] {sym} ptr x mem)
(MOVHstore [off] {sym} ptr (MOVHreg  x) mem) => (MOVHstore [off] {sym} ptr x mem)
(MOVHstore [off] {sym} ptr (MOVWreg  x) mem) => (MOVHstore [off] {sym} ptr x mem)
(MOVHstore [off] {sym} ptr (MOVHUreg x) mem) => (MOVHstore [off] {sym} ptr x mem)
(MOVHstore [off] {sym} ptr (MOVWUreg x) mem) => (MOVHstore [off] {sym} ptr x mem)
(MOVWstore [off] {sym} ptr (MOVWreg  x) mem) => (MOVWstore [off] {sym} ptr x mem)
(MOVWstore [off] {sym} ptr (MOVWUreg x) mem) => (MOVWstore [off] {sym} ptr x mem)

// Replace extend after load with alternate load where possible.
(MOVBreg  <t> x:(MOVBUload [off] {sym} ptr mem)) && x.Uses == 1 && clobber(x) => @x.Block (MOVBload  <t> [off] {sym} ptr mem)
(MOVHreg  <t> x:(MOVHUload [off] {sym} ptr mem)) && x.Uses == 1 && clobber(x) => @x.Block (MOVHload  <t> [off] {sym} ptr mem)
(MOVWreg  <t> x:(MOVWUload [off] {sym} ptr mem)) && x.Uses == 1 && clobber(x) => @x.Block (MOVWload  <t> [off] {sym} ptr mem)
(MOVBUreg <t> x:(MOVBload  [off] {sym} ptr mem)) && x.Uses == 1 && clobber(x) => @x.Block (MOVBUload <t> [off] {sym} ptr mem)
(MOVHUreg <t> x:(MOVHload  [off] {sym} ptr mem)) && x.Uses == 1 && clobber(x) => @x.Block (MOVHUload <t> [off] {sym} ptr mem)
(MOVWUreg <t> x:(MOVWload  [off] {sym} ptr mem)) && x.Uses == 1 && clobber(x) => @x.Block (MOVWUload <t> [off] {sym} ptr mem)

// If a register move has only 1 use, just use the same register without emitting instruction
// MOVnop does not emit an instruction, only for ensuring the type.
(MOVDreg x) && x.Uses == 1 => (MOVDnop x)

// TODO: we should be able to get rid of MOVDnop all together.
// But for now, this is enough to get rid of lots of them.
(MOVDnop (MOVDconst [c])) => (MOVDconst [c])

// Avoid unnecessary zero and sign extension when right shifting.
(SRAI <t> [x] (MOVWreg  y)) && x >= 0 && x <= 31 => (SRAIW <t> [int64(x)] y)
(SRLI <t> [x] (MOVWUreg y)) && x >= 0 && x <= 31 => (SRLIW <t> [int64(x)] y)

// Replace right shifts that exceed size of signed type.
(SRAI <t> [x] (MOVBreg y)) && x >=  8 => (SRAI  [63] (SLLI <t> [56] y))
(SRAI <t> [x] (MOVHreg y)) && x >= 16 => (SRAI  [63] (SLLI <t> [48] y))
(SRAI <t> [x] (MOVWreg y)) && x >= 32 => (SRAIW [31] y)

// Eliminate right shifts that exceed size of unsigned type.
(SRLI <t> [x] (MOVBUreg y)) && x >=  8 => (MOVDconst <t> [0])
(SRLI <t> [x] (MOVHUreg y)) && x >= 16 => (MOVDconst <t> [0])
(SRLI <t> [x] (MOVWUreg y)) && x >= 32 => (MOVDconst <t> [0])

// Fold constant into immediate instructions where possible.
(ADD (MOVDconst <t> [val]) x) && is32Bit(val) && !t.IsPtr() => (ADDI [val] x)
(AND (MOVDconst [val]) x) && is32Bit(val) => (ANDI [val] x)
(OR  (MOVDconst [val]) x) && is32Bit(val) => (ORI  [val] x)
(XOR (MOVDconst [val]) x) && is32Bit(val) => (XORI [val] x)
(SLL  x (MOVDconst [val])) => (SLLI [int64(val&63)] x)
(SRL  x (MOVDconst [val])) => (SRLI [int64(val&63)] x)
(SRLW x (MOVDconst [val])) => (SRLIW [int64(val&31)] x)
(SRA  x (MOVDconst [val])) => (SRAI [int64(val&63)] x)
(SRAW x (MOVDconst [val])) => (SRAIW [int64(val&31)] x)
(SLT  x (MOVDconst [val])) && val >= -2048 && val <= 2047 => (SLTI  [val] x)
(SLTU x (MOVDconst [val])) && val >= -2048 && val <= 2047 => (SLTIU [val] x)

// Convert const subtraction into ADDI with negative immediate, where possible.
(SUB x (MOVDconst [val])) && is32Bit(-val) => (ADDI [-val] x)
(SUB <t> (MOVDconst [val]) y) && is32Bit(-val) => (NEG (ADDI <t> [-val] y))

// Subtraction of zero.
(SUB  x (MOVDconst [0])) => x
(SUBW x (MOVDconst [0])) => (ADDIW [0] x)

// Subtraction from zero.
(SUB  (MOVDconst [0]) x) => (NEG x)
(SUBW (MOVDconst [0]) x) => (NEGW x)

// Fold negation into subtraction.
(NEG (SUB x y)) => (SUB y x)
(NEG <t> s:(ADDI [val] (SUB x y))) && s.Uses == 1 && is32Bit(-val) => (ADDI [-val] (SUB <t> y x))

// Double negation.
(NEG (NEG x)) => x

// Addition of zero or two constants.
(ADDI [0] x) => x
(ADDI [x] (MOVDconst [y])) && is32Bit(x + y) => (MOVDconst [x + y])

// ANDI with all zeros, all ones or two constants.
(ANDI [0]  x) => (MOVDconst [0])
(ANDI [-1] x) => x
(ANDI [x] (MOVDconst [y])) => (MOVDconst [x & y])

// ORI with all zeroes, all ones or two constants.
(ORI [0]  x) => x
(ORI [-1] x) => (MOVDconst [-1])
(ORI [x] (MOVDconst [y])) => (MOVDconst [x | y])

// Combine operations with immediate.
(ADDI [x] (ADDI [y] z)) && is32Bit(x + y) => (ADDI [x + y] z)
(ANDI [x] (ANDI [y] z)) => (ANDI [x & y] z)
(ORI  [x] (ORI  [y] z)) => (ORI  [x | y] z)

// Negation of a constant.
(NEG  (MOVDconst [x])) => (MOVDconst [-x])
(NEGW (MOVDconst [x])) => (MOVDconst [int64(int32(-x))])

// Shift of a constant.
(SLLI [x] (MOVDconst [y])) && is32Bit(y << uint32(x)) => (MOVDconst [y << uint32(x)])
(SRLI [x] (MOVDconst [y])) => (MOVDconst [int64(uint64(y) >> uint32(x))])
(SRAI [x] (MOVDconst [y])) => (MOVDconst [int64(y) >> uint32(x)])

// SLTI/SLTIU with constants.
(SLTI  [x] (MOVDconst [y])) => (MOVDconst [b2i(int64(y) < int64(x))])
(SLTIU [x] (MOVDconst [y])) => (MOVDconst [b2i(uint64(y) < uint64(x))])

// SLTI/SLTIU with known outcomes.
(SLTI  [x] (ANDI [y] _)) && y >= 0 && int64(y) < int64(x) => (MOVDconst [1])
(SLTIU [x] (ANDI [y] _)) && y >= 0 && uint64(y) < uint64(x) => (MOVDconst [1])
(SLTI  [x] (ORI  [y] _)) && y >= 0 && int64(y) >= int64(x) => (MOVDconst [0])
(SLTIU [x] (ORI  [y] _)) && y >= 0 && uint64(y) >= uint64(x) => (MOVDconst [0])

// SLT/SLTU with known outcomes.
(SLT  x x) => (MOVDconst [0])
(SLTU x x) => (MOVDconst [0])

// Deadcode for LoweredMuluhilo
(Select0 m:(LoweredMuluhilo x y)) && m.Uses == 1 => (MULHU x y)
(Select1 m:(LoweredMuluhilo x y)) && m.Uses == 1 => (MUL x y)

(FADD(S|D) a (FMUL(S|D) x y)) && a.Block.Func.useFMA(v) => (FMADD(S|D) x y a)
(FSUB(S|D) a (FMUL(S|D) x y)) && a.Block.Func.useFMA(v) => (FNMSUB(S|D) x y a)
(FSUB(S|D) (FMUL(S|D) x y) a) && a.Block.Func.useFMA(v) => (FMSUB(S|D) x y a)

// Merge negation into fused multiply-add and multiply-subtract.
//
// Key:
//
//   [+ -](x * y [+ -] z).
//    _ N         A S
//                D U
//                D B
//
// Note: multiplication commutativity handled by rule generator.
(F(MADD|NMADD|MSUB|NMSUB)S neg:(FNEGS x) y z) && neg.Uses == 1 => (F(NMSUB|MSUB|NMADD|MADD)S x y z)
(F(MADD|NMADD|MSUB|NMSUB)S x y neg:(FNEGS z)) && neg.Uses == 1 => (F(MSUB|NMSUB|MADD|NMADD)S x y z)
(F(MADD|NMADD|MSUB|NMSUB)D neg:(FNEGD x) y z) && neg.Uses == 1 => (F(NMSUB|MSUB|NMADD|MADD)D x y z)
(F(MADD|NMADD|MSUB|NMSUB)D x y neg:(FNEGD z)) && neg.Uses == 1 => (F(MSUB|NMSUB|MADD|NMADD)D x y z)