"pmovzxbw", "pmovzxbd", "pmovzxbq", "pmovzxwd", "pmovzxwq", "pmovzxdq",
		"pmovsxbw", "pmovsxbd", "pmovsxbq", "pmovsxwd", "pmovsxwq", "pmovsxdq",
		"pblendvb",
	},
	"fma":   {"vfmadd231sd"},
	"movbe": {"movbeqq", "movbeq", "movbell", "movbel", "movbe"},
	"lzcnt": {"lzcntq", "lzcntl", "lzcnt"},
}

// Test to use POPCNT instruction, if available
func TestPopCnt(t *testing.T) {
	for _, tt := range []struct {
		x    uint64

		(MOVQstore dst (MOVQload src mem) mem))

(Move [32] dst src mem) =>
	(Move [16]
		(OffPtr <dst.Type> dst [16])
		(OffPtr <src.Type> src [16])
		(Move [16] dst src mem))

(Move [48] dst src mem) && config.useSSE =>
	(Move [32]
		(OffPtr <dst.Type> dst [16])
		(OffPtr <src.Type> src [16])
		(Move [16] dst src mem))

(Move [64] dst src mem) && config.useSSE =>
	(Move [32]
		(OffPtr <dst.Type> dst [32])

	0xF2, 2966,
	0x0, 2950,
	/*2950*/ uint16(xCondDataSize), 2954, 2960, 0,
	/*2954*/ uint16(xSetOp), uint16(MOVBE),
	/*2956*/ uint16(xReadSlashR),
	/*2957*/ uint16(xArgR16),
	/*2958*/ uint16(xArgM16),
	/*2959*/ uint16(xMatch),
	/*2960*/ uint16(xSetOp), uint16(MOVBE),
	/*2962*/ uint16(xReadSlashR),
	/*2963*/ uint16(xArgR32),
	/*2964*/ uint16(xArgM32),
	/*2965*/ uint16(xMatch),

		{name: "MOVBEQloadidx8", argLength: 3, reg: gploadidx, asm: "MOVBEQ", scale: 8, aux: "SymOff", typ: "UInt64", symEffect: "Read"},                    // load and swap 8 bytes from arg0+8*arg1+auxint+aux. arg2=mem
		// indexed MOVBE stores

	// amd64:`MOVWLZX\s\(.*\),`,-`MOVB`,-`OR`
	// ppc64le:`MOVHZ\s`,-`MOVBZ`
	// arm64:`MOVHU\s\(R[0-9]+\),`,-`MOVB`
	// s390x:`MOVHBR\s\(.*\),`
	// ppc64:`MOVHBR\s`,-`MOVBZ`
	return binary.LittleEndian.Uint16(b)
}

func load_le16_idx(b []byte, idx int) uint16 {
	// amd64:`MOVWLZX\s\(.*\),`,-`MOVB`,-`OR`
	// ppc64le:`MOVHZ\s`,-`MOVBZ`
	// ppc64:`MOVHBR\s`,-`MOVBZ`
	// arm64:`MOVHU\s\(R[0-9]+\)\(R[0-9]+\),`,-`MOVB`

	MOVB	R0>>0(R1), R2      // ERROR "bad shift"
	MOVB	R0->0(R1), R2      // ERROR "bad shift"
	MOVB	R0@>0(R1), R2      // ERROR "bad shift"
	MOVBS	R0>>0(R1), R2      // ERROR "bad shift"
	MOVBS	R0->0(R1), R2      // ERROR "bad shift"
	MOVBS	R0@>0(R1), R2      // ERROR "bad shift"
	MOVF	CPSR, F1           // ERROR "illegal combination"
	MOVD	R1, CPSR           // ERROR "illegal combination"

	//TODO MOVBW.P 77(R19), R11                // 6bd6c438
	MOVB.P 36(RSP), R27                        // fb478238
	//TODO MOVBW.W -57(R19), R13               // 6d7edc38
	MOVB.W -178(R16), R24                      // 18ee9438
	//TODO MOVBW 430(R8), R22                  // 16b9c639
	MOVB 997(R9), R23                          // 37958f39
	//TODO MOVBW (R2<<1)(R21), R15             // af7ae238

	MOVQ	$masks<>(SB), AX
	MOVQ	$shifts<>(SB), BX
	ORQ	BX, AX
	TESTQ	$15, AX
	SETEQ	ret+0(FP)
	RET

// these are arguments to pshufb. They move data down from
// the high bytes of the register to the low bytes of the register.
// index is how many bytes to move.
DATA shifts<>+0x00(SB)/8, $0x0000000000000000
DATA shifts<>+0x08(SB)/8, $0x0000000000000000
DATA shifts<>+0x10(SB)/8, $0xffffffffffffff0f

	MOVBS	0x00ffffff(R2), R1   // MOVBS 16777215(R2), R1
	MOVBS.P	0x00ffffff(R2), R1   // MOVBS.P 16777215(R2), R1
	MOVBS.W	0x00ffffff(R2), R1   // MOVBS.W 16777215(R2), R1
	MOVBS	-0x00ffffff(R2), R1  // MOVBS -16777215(R2), R1
	MOVBS.P	-0x00ffffff(R2), R1  // MOVBS.P -16777215(R2), R1
	MOVBS.W	-0x00ffffff(R2), R1  // MOVBS.W -16777215(R2), R1
	MOVBU	R1, 0x00ffffff(R2)   // MOVBU R1, 16777215(R2)

	B.NE	chunk4_loop
	CMP	R0, R1
	B.EQ	equal		// reached the end
byte_loop:
	MOVBU.P	1(R0), R4
	MOVBU.P	1(R2), R5
	CMP	R4, R5
	B.NE	notequal
	CMP	R0, R1
	B.NE	byte_loop
equal:
	MOVW	$1, R0
	MOVB	R0, (R7)
	RET
one:
	MOVBU	(R0), R4
	MOVBU	(R2), R5
	CMP	R4, R5
	B.EQ	equal
notequal:
	MOVW	$0, R0
	MOVB	R0, (R7)