"\tPINSRW	$index, (BP)(R8*4), xmm",
			"#define LOAD(off, reg) \\",
			"\tMOVBLZX	(off*4)(R12),	reg \\",
			"\tADDB	reg,		DX",
			"KEYROUND(X0, LOAD, 8, AX, BX, 0)",
		),
		"\n.MOVBLZX.(.BP.).(.DX.*.4.).,.R8.\n.\n.MOVBLZX.(.(.8.+.1.).*.4.).(.R12.).,.BX.\n.ADDB.BX.,.DX.\n.MOVB.R8.,.(.8.*.4.).(.R12.).\n.PINSRW.$.0.,.(.BP.).(.R8.*.4.).,.X0.\n",
	},
	{
		"taken #ifdef",
		lines(
			"#define A",
			"#ifdef A",

	// amd64: `ORQ\t[$]77, 64\([A-Z]+[0-9]*\)\([A-Z]+[0-9]*\*8\)`
	x[i+8] |= 77
	// amd64: `XORQ\t[$]77, 72\([A-Z]+[0-9]*\)\([A-Z]+[0-9]*\*8\)`
	x[i+9] ^= 77
}

func idxCompare(i int) int {
	// amd64: `MOVBLZX\t1\([A-Z]+[0-9]*\)\([A-Z]+[0-9]*\*1\), [A-Z]+[0-9]*`
	if x8[i+1] < x8[0] {
		return 0
	}
	// amd64: `MOVWLZX\t2\([A-Z]+[0-9]*\)\([A-Z]+[0-9]*\*2\), [A-Z]+[0-9]*`
	if x16[i+1] < x16[0] {
		return 0
	}

// Generic atomic operation implementation.
// AX already contains target function.
TEXT	racecallatomic<>(SB), NOSPLIT|NOFRAME, $0-0
	// Trigger SIGSEGV early.
	MOVQ	16(SP), R12
	MOVBLZX	(R12), R13
	// Check that addr is within [arenastart, arenaend) or within [racedatastart, racedataend).
	CMPQ	R12, runtime·racearenastart(SB)
	JB	racecallatomic_data
	CMPQ	R12, runtime·racearenaend(SB)

		{name: "MOVBQSX", argLength: 1, reg: gp11, asm: "MOVBQSX"}, // sign extend arg0 from int8 to int64
		{name: "MOVBQZX", argLength: 1, reg: gp11, asm: "MOVBLZX"}, // zero extend arg0 from int8 to int64
		{name: "MOVWQSX", argLength: 1, reg: gp11, asm: "MOVWQSX"}, // sign extend arg0 from int16 to int64

		reg: regInfo{
			inputs: []inputInfo{
				{0, 239}, // AX CX DX BX BP SI DI
			},
			outputs: []outputInfo{
				{0, 239}, // AX CX DX BX BP SI DI
			},
		},
	},
	{
		name:   "MOVBLZX",
		argLen: 1,
		asm:    x86.AMOVBLZX,
		reg: regInfo{
			inputs: []inputInfo{
				{0, 239}, // AX CX DX BX BP SI DI
			},
			outputs: []outputInfo{
				{0, 239}, // AX CX DX BX BP SI DI