// TODO: Maybe the constants are named wrong.
	SPB: "spl",
	BPB: "bpl",
	SIB: "sil",
	DIB: "dil",

	R8L:  "r8d",
	R9L:  "r9d",
	R10L: "r10d",
	R11L: "r11d",
	R12L: "r12d",
	R13L: "r13d",
	R14L: "r14d",
	R15L: "r15d",

	ESI:  "%esi",
	EDI:  "%edi",
	R8L:  "%r8d",
	R9L:  "%r9d",
	R10L: "%r10d",
	R11L: "%r11d",
	R12L: "%r12d",
	R13L: "%r13d",
	R14L: "%r14d",
	R15L: "%r15d",
	RAX:  "%rax",
	RCX:  "%rcx",
	RDX:  "%rdx",
	RBX:  "%rbx",
	RSP:  "%rsp",
	RBP:  "%rbp",
	RSI:  "%rsi",
	RDI:  "%rdi",
	R8:   "%r8",
	R9:   "%r9",
	R10:  "%r10",
	R11:  "%r11",
	R12:  "%r12",
	R13:  "%r13",
	R14:  "%r14",

	SIB:  "SIB",
	DIB:  "DIB",
	R8B:  "R8B",
	R9B:  "R9B",
	R10B: "R10B",
	R11B: "R11B",
	R12B: "R12B",
	R13B: "R13B",
	R14B: "R14B",
	R15B: "R15B",
	AX:   "AX",
	CX:   "CX",
	BX:   "BX",
	DX:   "DX",
	SP:   "SP",
	BP:   "BP",
	SI:   "SI",
	DI:   "DI",
	R8W:  "R8W",
	R9W:  "R9W",
	R10W: "R10W",
	R11W: "R11W",
	R12W: "R12W",
	R13W: "R13W",
	R14W: "R14W",

	BPB:  "BP",
	SIB:  "SI",
	DIB:  "DI",
	R8B:  "R8",
	R9B:  "R9",
	R10B: "R10",
	R11B: "R11",
	R12B: "R12",
	R13B: "R13",
	R14B: "R14",
	R15B: "R15",
	AX:   "AX",
	CX:   "CX",
	BX:   "BX",
	DX:   "DX",
	SP:   "SP",
	BP:   "BP",
	SI:   "SI",
	DI:   "DI",
	R8W:  "R8",
	R9W:  "R9",
	R10W: "R10",
	R11W: "R11",
	R12W: "R12",
	R13W: "R13",
	R14W: "R14",
	R15W: "R15",

	"AL", // [D_AL]
	"CL",
	"DL",
	"BL",
	"SPB",
	"BPB",
	"SIB",
	"DIB",
	"R8B",
	"R9B",
	"R10B",
	"R11B",
	"R12B",
	"R13B",
	"R14B",
	"R15B",
	"AX", // [D_AX]
	"CX",
	"DX",
	"BX",
	"SP",
	"BP",
	"SI",
	"DI",
	"R8",
	"R9",
	"R10",
	"R11",
	"R12",
	"R13",
	"R14",
	"R15",
	"AH",
	"CH",
	"DH",
	"BH",
	"F0", // [D_F0]
	"F1",
	"F2",

	BLT	less_than_16
	LDP.P	16(R13), (R8, R10)
	CRC32CX	R8, R9
	CRC32CX	R10, R9
	SUB	$16, R11

	JMP	update

less_than_16:
	TBZ	$3, R11, less_than_8

	MOVD.P	8(R13), R10
	CRC32CX	R10, R9

less_than_8:
	TBZ	$2, R11, less_than_4

	MOVWU.P	4(R13), R10
	CRC32CW	R10, R9

less_than_4:
	TBZ	$1, R11, less_than_2

	MOVHU.P	2(R13), R10
	CRC32CH	R10, R9

less_than_2:
	TBZ	$0, R11, done

	MOVD 24(R8)(R10*1), R1
	MOVD R0, R11
	SUBC R4, R11            // restore CF
	MOVD 0(R9)(R10*1), R11
	SUBE R11, R5
	MOVD 8(R9)(R10*1), R11
	SUBE R11, R6
	MOVD 16(R9)(R10*1), R11
	SUBE R11, R7
	MOVD 24(R9)(R10*1), R11
	SUBE R11, R1
	MOVD R0, R4
	SUBE R4, R4             // save CF
	MOVD R5, 0(R2)(R10*1)
	MOVD R6, 8(R2)(R10*1)
	MOVD R7, 16(R2)(R10*1)
	MOVD R1, 24(R2)(R10*1)

	ADD $32, R10 // i += 4

	LXVD2X	(R0)(R6),V4
	VCMPEQUDCC	V3,V4,V1
	BGE	CR6,different

	LXVD2X	(R10)(R5),V3
	LXVD2X	(R10)(R6),V4
	VCMPEQUDCC	V3,V4,V1
	BGE	CR6,different

	BC	$12,2,LR	// beqlr
	ADD	R9,R10,R10

	LXVD2X	(R9)(R5),V3
	LXVD2X	(R9)(R6),V4
	VCMPEQUDCC	V3,V4,V1
	BGE	CR6,different

	LXVD2X	(R10)(R5),V3
	LXVD2X	(R10)(R6),V4
	VCMPEQUDCC	V3,V4,V1
	BGE	CR6,different
	RET

	PCALIGN $16

	MOVQ	R10, 96(SP)
	MOVQ	8(R11), R10
	MOVQ	R10, 104(SP)
	MOVQ	16(R11), R10
	MOVQ	R10, 112(SP)
	MOVQ	24(R11), R10
	MOVQ	R10, 120(SP)
	MOVQ	32(R11), R10
	MOVQ	R10, 128(SP)
	MOVQ	40(R11), R10
	MOVQ	R10, 136(SP)

	// save errno, it might be EINTR; stuff we do here might reset it.
	MOVQ	(m_mOS+mOS_perrno)(BP), R10
	MOVL	0(R10), R10
	MOVQ	R10, 144(SP)

	// prepare call
	MOVQ	DI, 0(SP)

	BEQ	R3, R4, samebytes // same start of a and b

	SGTU	R1, R2, R7
	BNE	R0, R7, r2_lt_r1
	MOVV	R1, R10
	JMP	entry
r2_lt_r1:
	MOVV	R2, R10	// R10 is min(R1, R2)
entry:
	ADDV	R3, R10, R8	// R3 start of a, R8 end of a
	BEQ	R3, R8, samebytes // length is 0

	SRLV	$4, R10		// R10 is number of chunks
	BEQ	R0, R10, byte_loop

	// make sure both a and b are aligned.
	OR	R3, R4, R11
	AND	$7, R11
	BNE	R0, R11, byte_loop