(Ctz(64|32|16|8)NonZero ...) => (I64Ctz ...)

(BitLen64 x) => (I64Sub (I64Const [64]) (I64Clz x))

(PopCount64 ...) => (I64Popcnt ...)
(PopCount32 x) => (I64Popcnt (ZeroExt32to64 x))
(PopCount16 x) => (I64Popcnt (ZeroExt16to64 x))
(PopCount8  x) => (I64Popcnt (ZeroExt8to64  x))

(CondSelect ...) => (Select ...)

// --- Optimizations ---
(I64Add (I64Const [x]) (I64Const [y])) => (I64Const [x + y])

	{name: "BitRev32", argLength: 1}, // Reverse the bits in arg[0]
	{name: "BitRev64", argLength: 1}, // Reverse the bits in arg[0]

	{name: "PopCount8", argLength: 1},  // Count bits in arg[0]
	{name: "PopCount16", argLength: 1}, // Count bits in arg[0]
	{name: "PopCount32", argLength: 1}, // Count bits in arg[0]
	{name: "PopCount64", argLength: 1}, // Count bits in arg[0]

//     SRW    $8, R4, R5      // R5=0x00020306
//     ADDW   R4, R5, R6      // R6=0x0205090d
//     MOVBZ  R6, R7          // R7=0x0000000d <-- result is 13
//
(PopCount8  x) => (POPCNT (MOVBZreg x))
(PopCount16 x) => (MOVBZreg (SumBytes2 (POPCNT <typ.UInt16> x)))
(PopCount32 x) => (MOVBZreg (SumBytes4 (POPCNT <typ.UInt32> x)))
(PopCount64 x) => (MOVBZreg (SumBytes8 (POPCNT <typ.UInt64> x)))

(BitLen8 <t> x) && buildcfg.GOAMD64 >= 3 => (NEGQ (ADDQconst <t> [-32] (LZCNTL (MOVBQZX <x.Type> x))))

(Bswap(64|32) ...) => (BSWAP(Q|L) ...)
(Bswap16 x) => (ROLWconst [8] x)

(PopCount(64|32) ...) => (POPCNT(Q|L) ...)
(PopCount16 x) => (POPCNTL (MOVWQZX <typ.UInt32> x))
(PopCount8 x) => (POPCNTL (MOVBQZX <typ.UInt32> x))

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

(RoundToEven x) => (ROUNDSD [0] x)

(PopCount64 <t> x) => (FMOVDfpgp <t> (VUADDLV <typ.Float64> (VCNT <typ.Float64> (FMOVDgpfp <typ.Float64> x))))
(PopCount32 <t> x) => (FMOVDfpgp <t> (VUADDLV <typ.Float64> (VCNT <typ.Float64> (FMOVDgpfp <typ.Float64> (ZeroExt32to64 x)))))
(PopCount16 <t> x) => (FMOVDfpgp <t> (VUADDLV <typ.Float64> (VCNT <typ.Float64> (FMOVDgpfp <typ.Float64> (ZeroExt16to64 x)))))

// Load args directly into the register class where it will be used.

		v1.AddArg(y)
		v.AddArg2(v0, v1)
		return true
	}
}
func rewriteValueWasm_OpPopCount16(v *Value) bool {
	v_0 := v.Args[0]
	b := v.Block
	typ := &b.Func.Config.Types
	// match: (PopCount16 x)
	// result: (I64Popcnt (ZeroExt16to64 x))
	for {
		x := v_0
		v.reset(OpWasmI64Popcnt)
		v0 := b.NewValue0(v.Pos, OpZeroExt16to64, typ.UInt64)
		v0.AddArg(x)
		v.AddArg(v0)
		return true

		v.AddArg3(x, y, mem)
		return true
	}
	return false
}
func rewriteValuePPC64_OpPopCount16(v *Value) bool {
	v_0 := v.Args[0]
	b := v.Block
	typ := &b.Func.Config.Types
	// match: (PopCount16 x)
	// result: (POPCNTW (MOVHZreg x))
	for {
		x := v_0
		v.reset(OpPPC64POPCNTW)
		v0 := b.NewValue0(v.Pos, OpPPC64MOVHZreg, typ.Int64)
		v0.AddArg(x)
		v.AddArg(v0)
		return true

		v.AddArg3(x, y, mem)
		return true
	}
	return false
}
func rewriteValueS390X_OpPopCount16(v *Value) bool {
	v_0 := v.Args[0]
	b := v.Block
	typ := &b.Func.Config.Types
	// match: (PopCount16 x)
	// result: (MOVBZreg (SumBytes2 (POPCNT <typ.UInt16> x)))
	for {
		x := v_0
		v.reset(OpS390XMOVBZreg)
		v0 := b.NewValue0(v.Pos, OpS390XSumBytes2, typ.UInt8)

		v.AddArg3(x, y, mem)
		return true
	}
	return false
}
func rewriteValueARM64_OpPopCount16(v *Value) bool {
	v_0 := v.Args[0]
	b := v.Block
	typ := &b.Func.Config.Types
	// match: (PopCount16 <t> x)
	// result: (FMOVDfpgp <t> (VUADDLV <typ.Float64> (VCNT <typ.Float64> (FMOVDgpfp <typ.Float64> (ZeroExt16to64 x)))))
	for {
		t := v.Type
		x := v_0
		v.reset(OpARM64FMOVDfpgp)
		v.Type = t

		v.AddArg3(x, y, mem)
		return true
	}
	return false
}
func rewriteValueAMD64_OpPopCount16(v *Value) bool {
	v_0 := v.Args[0]
	b := v.Block
	typ := &b.Func.Config.Types
	// match: (PopCount16 x)
	// result: (POPCNTL (MOVWQZX <typ.UInt32> x))
	for {
		x := v_0
		v.reset(OpAMD64POPCNTL)
		v0 := b.NewValue0(v.Pos, OpAMD64MOVWQZX, typ.UInt32)
		v0.AddArg(x)
		v.AddArg(v0)