SGTU	$8, R6, R7
	BNE	R7, out

	// do one byte at a time until 8-aligned
	AND	$7, R4, R8
	BEQ	R8, words
	MOVB	(R5), R7
	ADDV	$1, R5
	MOVB	R7, (R4)
	ADDV	$1, R4
	JMP	-6(PC)

words:
	// do 8 bytes at a time if there is room
	ADDV	$-7, R9, R6 // R6 is end pointer-7

	PCALIGN	$16
	SGTU	R6, R4, R8
	BEQ	R8, out
	MOVV	(R5), R7
	ADDV	$8, R5
	MOVV	R7, (R4)
	ADDV	$8, R4
	JMP	-6(PC)

out:

	SHA512ROUND0(14, 0x9bdc06a725c71235, R10, R11, R12, R13, R14, R15, R8, R9)
	SHA512ROUND0(15, 0xc19bf174cf692694, R9, R10, R11, R12, R13, R14, R15, R8)

	SHA512ROUND1(16, 0xe49b69c19ef14ad2, R8, R9, R10, R11, R12, R13, R14, R15)
	SHA512ROUND1(17, 0xefbe4786384f25e3, R15, R8, R9, R10, R11, R12, R13, R14)
	SHA512ROUND1(18, 0x0fc19dc68b8cd5b5, R14, R15, R8, R9, R10, R11, R12, R13)

	LXVD2X	(R0)(R8),V2
	VCMPEQUBCC	V2,V1,V6
	BNE	CR6,foundat0

	// Bytes 16 - 31
	LXVD2X	(R8)(R11),V2
	VCMPEQUBCC	V2,V1,V6
	BNE	CR6,foundat1		// Match found at R8+16 bytes, jump out

	BEQ	notfound		// Is length <= 32? (CR0 holds this comparison on entry to cmp32)
	CMP	R4,$48

	ADD	R4,R8,R9		// Compute &s[len(s)-16]
	ADD	$32,R8,R8
	ADD	$-16,R9,R9
	ISEL	CR0GT,R8,R9,R8		// R8 = len(s) <= 48 ? R9 : R8

		b = r4(add(p, s-4))
	case s == 8:
		a = r8(p)
		b = a
	case s <= 16:
		a = r8(p)
		b = r8(add(p, s-8))
	default:
		l := s
		if l > 48 {
			seed1 := seed
			seed2 := seed
			for ; l > 48; l -= 48 {
				seed = mix(r8(p)^hashkey[1], r8(add(p, 8))^seed)
				seed1 = mix(r8(add(p, 16))^hashkey[2], r8(add(p, 24))^seed1)
				seed2 = mix(r8(add(p, 32))^hashkey[3], r8(add(p, 40))^seed2)
				p = add(p, 48)
			}

	BL runtime·mstart(SB)

	MOVW $2, R8  // crash (not reached)
	MOVW R8, (R8)
	RET

// Exit the entire program (like C exit)
TEXT runtime·exit(SB),NOSPLIT|NOFRAME,$0
	MOVW code+0(FP), R0	// arg 1 exit status
	MOVW $SYS_exit, R7
	SWI $0
	MOVW.CS $0, R8 // crash on syscall failure
	MOVW.CS R8, (R8)
	RET

// func exitThread(wait *atomic.Uint32)

// R7: address to put result
//
// On exit:
// R4 and R8 are clobbered
TEXT countbytebody<>(SB),NOSPLIT,$0
	MOVW	$0, R8	// R8 = count of byte to search
	CMP	$0, R1
	B.EQ	done	// short path to handle 0-byte case
	ADD	R0, R1	// R1 is the end of the range
byte_loop:
	MOVBU.P	1(R0), R4
	CMP	R4, R2
	ADD.EQ	$1, R8
	CMP	R0, R1
	B.NE	byte_loop
done:
	MOVW	R8, (R7)

				{0, 49151}, // AX CX DX BX SP BP SI DI R8 R9 R10 R11 R12 R13 R15
				{1, 49151}, // AX CX DX BX SP BP SI DI R8 R9 R10 R11 R12 R13 R15
			},
		},
	},
	{
		name:   "CMPW",
		argLen: 2,
		asm:    x86.ACMPW,
		reg: regInfo{
			inputs: []inputInfo{
				{0, 49151}, // AX CX DX BX SP BP SI DI R8 R9 R10 R11 R12 R13 R15
				{1, 49151}, // AX CX DX BX SP BP SI DI R8 R9 R10 R11 R12 R13 R15
			},
		},

aligned_a:
	AND	$3, R3, R8
	CMP	$0, R8
	BNE	byte_loop	// unaligned b, use byte-wise compare
	AND	$0xfffffffc, R6, R8
	// length >= 4
chunk4_loop:
	MOVW.P	4(R2), R4
	MOVW.P	4(R3), R5
	CMP	R4, R5
	BNE	cmp
	CMP	R2, R8
	BNE	chunk4_loop
	CMP	R2, R6
	BEQ	samebytes	// all compared bytes were the same; compare lengths
byte_loop:
	MOVBU.P	1(R2), R4
	MOVBU.P	1(R3), R5
	CMP	R4, R5
	BNE	ret
	CMP	R2, R6

	MOVQ	b_len+8(FP), BX
	MOVB	c+24(FP), AL
	LEAQ	ret+32(FP), R8
	JMP	countbody<>(SB)

TEXT ·CountString(SB),NOSPLIT,$0-32
#ifndef hasPOPCNT
	CMPB	internal∕cpu·X86+const_offsetX86HasPOPCNT(SB), $1
	JEQ	2(PC)
	JMP	·countGenericString(SB)
#endif
	MOVQ	s_base+0(FP), SI
	MOVQ	s_len+8(FP), BX
	MOVB	c+16(FP), AL
	LEAQ	ret+24(FP), R8
	JMP	countbody<>(SB)

// input:
//   SI: data
//   BX: data len

	MOVD	R3, R8
	AND	$15, R8
	CMPBGT	R8, $0, notaligned

aligned:
	ADD	R6, R4, R8
	MOVD	R8, R7
	AND	$-16, R7
	// replicate c across V17
	VLVGB	$0, R5, V19
	VREPB	$0, V19, V17

vectorloop:
	CMPBGE	R3, R7, residual
	VL	0(R3), V16    // load string to be searched into V16
	ADD	$16, R3
	VFEEBS	V16, V17, V18 // search V17 in V16 and set conditional code accordingly
	BVS	vectorloop