MOVD	a_base+24(FP), R2
	MOVD	b_base+48(FP), R3
	MOVD	a_len+32(FP), R4
	MOVD	b_len+56(FP), R5
	CMPBLE	R4, R5, skip
	MOVD	R5, R4
skip:
	MOVD	R4, ret+72(FP)
	MOVD	$0, R5
	CMPBLT	R4, $8, tail
loop:
	MOVD	0(R2)(R5*1), R7
	MOVD	0(R3)(R5*1), R8
	XOR	R7, R8
	MOVD	R8, 0(R1)(R5*1)
	LAY	8(R5), R5
	SUB	$8, R4
	CMPBGE	R4, $8, loop
tail:
	CMPBEQ	R4, $0, done
	MOVB	0(R2)(R5*1), R7
	MOVB	0(R3)(R5*1), R8
	XOR	R7, R8

	// Synchronous initialization.
	MOVD	$runtime·libpreinit(SB), R4
	BL	(R4)

	// Create a new thread to do the runtime initialization and return.
	MOVD	_cgo_sys_thread_create(SB), R4
	CBZ	R4, nocgo
	MOVD	$_rt0_arm64_freebsd_lib_go(SB), R0
	MOVD	$0, R1
	SUB	$16, RSP	// reserve 16 bytes for sp-8 where fp may be saved.
	BL	(R4)
	ADD	$16, RSP
	B	restore

nocgo:

	MULVU	R3, R4
	MOVV	LO, R4
	SUBVU	R4, R5
	MOVV	R5, 16(R29)		// tv_nsec

	ADDV	$8, R29, R4		// arg 1 - rqtp
	MOVV	$0, R5			// arg 2 - rmtp
	MOVV	$91, R2			// sys_nanosleep
	SYSCALL
	RET

TEXT runtime·getthrid(SB),NOSPLIT,$0-4
	MOVV	$299, R2		// sys_getthrid
	SYSCALL
	MOVW	R2, ret+0(FP)
	RET

TEXT runtime·thrkill(SB),NOSPLIT,$0-16
	MOVW	tid+0(FP), R4		// arg 1 - tid

	addMul64(r3, 1, a1, b2)
	addMul64(r3, 1, a2, b1)
	addMul64(r3, 1, a3, b0)
	addMul64(r3, 19, a4, b4)

	// r4 = a0×b4 + a1×b3 + a2×b2 + a3×b1 + a4×b0
	r4 := uint128{"r4", GP64(), GP64()}
	mul64(r4, 1, a0, b4)
	addMul64(r4, 1, a1, b3)
	addMul64(r4, 1, a2, b2)
	addMul64(r4, 1, a3, b1)
	addMul64(r4, 1, a4, b0)

	Comment("First reduction chain")
	maskLow51Bits := GP64()
	MOVQ(Imm((1<<51)-1), maskLow51Bits)

	ADDE	R1, R2                // b9880021
	SUB	R3, R4                // b9090043
	SUB	R3, R4, R5            // b9e93054
	SUB	$8192, R3             // a73be000
	SUB	$8192, R3, R4         // ec43e00000d9
	SUBC	R1, R2                // b90b0021
	SUBC	$1, R1, R2            // ec21ffff00db
	SUBC	R2, R3, R4            // b9eb2043
	SUBW	R3, R4                // 1b43
	SUBW	R3, R4, R5            // b9f93054

	MOVD	$0x8637bd06, R4
	MULLD	R3, R4, R4	// Convert usec to S.
	SRD	$51, R4, R4
	MOVD	R4, 8(R1)	// Store to tv_sec

	MOVD	$1000000, R5
	MULLW	R4, R5, R5	// Convert tv_sec back into uS
	SUB	R5, R3, R5	// Compute remainder uS.
	MULLD	$1000, R5, R5	// Convert to nsec
	MOVD	R5, 16(R1)	// Store to tv_nsec

	// nanosleep(&ts, 0)
	ADD	$8, R1, R3
	MOVW	$0, R4
	SYSCALL	$SYS_nanosleep
	RET

	VMOVB	W0, 112(R1)

	ADDVU	$128, R1
	SGTU	R6, R1, R3
	BNE	R3, R0, msa_large_loop
	BEQ	R5, R0, done
	VMOVB	W0, -128(R4)
	VMOVB	W0, -112(R4)
	VMOVB	W0, -96(R4)
	VMOVB	W0, -80(R4)
	VMOVB	W0, -64(R4)
	VMOVB	W0, -48(R4)
	VMOVB	W0, -32(R4)
	VMOVB	W0, -16(R4)
	JMP	done

no_msa:
	// if less than 8 bytes, do one byte at a time
	SGTU	$8, R2, R3
	BNE	R3, out

	// do one byte at a time until 8-aligned

	RET

TEXT runtime·usleep(SB),NOSPLIT,$28-4
	MOVW	usec+0(FP), R3
	MOVW	R3, R5
	MOVW	$1000000, R4
	DIVU	R4, R3
	MOVW	LO, R3
	MOVW	R3, 24(R29)
	MOVW	$1000, R4
	MULU	R3, R4
	MOVW	LO, R4
	SUBU	R4, R5
	MOVW	R5, 28(R29)

	// nanosleep(&ts, 0)
	ADDU	$24, R29, R4
	MOVW	$0, R5
	MOVW	$SYS_nanosleep, R2
	SYSCALL
	RET

TEXT runtime·gettid(SB),NOSPLIT,$0-4

// sep: string to search for
// R1=&s[0], R2=len(s)
// R3=&sep[0], R4=len(sep)
// R5=&ret (int)
// Caller must confirm availability of vx facility before calling.
TEXT indexbody<>(SB),NOSPLIT|NOFRAME,$0
	CMPBGT	R4, R2, notfound
	ADD	R1, R2
	SUB	R4, R2 // R2=&s[len(s)-len(sep)] (last valid index)
	CMPBEQ	R4, $0, notfound
	SUB	$1, R4 // R4=len(sep)-1 for use as VLL index
	VLL	R4, (R3), V0 // contains first 16 bytes of sep
	MOVD	R1, R7

	ADD     $(-64), R4               // LEN = LEN - 64

	CMP     R4, $64
	BGE     fold_64bytes_loop

less_than_64bytes:
	// Fold V1 to V4 into a single 128-bit value in V1
	VGFMAG  CONST_R4R3, V1, V2, V1
	VGFMAG  CONST_R4R3, V1, V3, V1
	VGFMAG  CONST_R4R3, V1, V4, V1

	// Check whether to continue with 64-bit folding
	CMP R4, $16
	BLT final_fold

fold_16bytes_loop: