ADDU	$65536, R4, R5		// 1e02001485781000
	ADDU	$4096, R4, R5		// 3e00001485781000
	ADDU	$65536, R4		// 1e02001484781000
	ADDU	$4096, R4		// 3e00001484781000
	ADDVU	$65536, R4, R5		// 1e02001485f81000
	ADDVU	$4096, R4, R5		// 3e00001485f81000
	ADDVU	$65536, R4		// 1e02001484f81000
	ADDVU	$4096, R4		// 3e00001484f81000
	OR	$65536, R4, R5		// 1e02001485781500
	OR	$4096, R4, R5		// 3e00001485781500
	OR	$65536, R4		// 1e02001484781500

	VMOVB	$0, W0

	SGTU	$128, R2, R3
	BEQ	R3, msa_large

	AND	$15, R2, R5
	XOR	R2, R5, R6
	ADDVU	R1, R6

msa_small:
	VMOVB	W0, (R1)
	ADDVU	$16, R1
	SGTU	R6, R1, R3
	BNE	R3, R0, msa_small
	BEQ	R5, R0, done
	VMOVB	W0, -16(R4)
	JMP	done

msa_large:
	AND	$127, R2, R5
	XOR	R2, R5, R6
	ADDVU	R1, R6

msa_large_loop:
	VMOVB	W0, (R1)
	VMOVB	W0, 16(R1)
	VMOVB	W0, 32(R1)
	VMOVB	W0, 48(R1)

	MOVV	8(R6), R17
	ADDV	$16, R4
	ADDV	$16, R6
	SUBVU	$1, R14
	BEQ	R16, R17, chunk16_loop
	SUBV	$8, R4
	SUBV	$8, R6

byte_loop:
	BEQ	R4, R12, samebytes
	MOVBU	(R4), R8
	ADDVU	$1, R4
	MOVBU	(R6), R9
	ADDVU	$1, R6
	BEQ	R8, R9, byte_loop

byte_cmp:
	SGTU	R8, R9, R4 // R12 = 1 if (R8 > R9)
	BNE	R0, R4, ret
	MOVV	$-1, R4
	JMP	ret

samebytes:
	SGTU	R5, R7, R8
	SGTU	R7, R5, R9

	ADDU	$74565, R4, R5 			// 5e020014de178d0385781000
	ADDU	$4097, R4 			// 3e000014de07800384781000
	ADDU	$4097, R4, R5 			// 3e000014de07800385781000
	ADDVU	$4097, R4			// 3e000014de07800384f81000
	ADDVU	$4097, R4, R5 			// 3e000014de07800385f81000
	ADDVU	$74565, R4			// 5e020014de178d0384f81000
	ADDVU	$74565, R4, R5			// 5e020014de178d0385f81000
	OR	$74565, R4			// 5e020014de178d0384781500
	OR	$74565, R4, R5			// 5e020014de178d0385781500

	MOVV	8(R4), R14
	ADDV	$16, R3
	ADDV	$16, R4
	SUBVU	$1, R10
	BEQ	R13, R14, chunk16_loop
	SUBV	$8, R3
	SUBV	$8, R4

byte_loop:
	BEQ	R3, R8, samebytes
	MOVBU	(R3), R6
	ADDVU	$1, R3
	MOVBU	(R4), R7
	ADDVU	$1, R4
	BEQ	R6, R7, byte_loop

byte_cmp:
	SGTU	R6, R7, R8 // R8 = 1 if (R6 > R7)
	BNE	R0, R8, ret
	MOVV	$-1, R8
	JMP	ret

samebytes:
	SGTU	R1, R2, R6
	SGTU	R2, R1, R7

	"TNE",
	"WORD",
	"WSBH",
	"XOR",
	"MOVV",
	"MOVVL",
	"MOVVR",
	"SLLV",
	"SRAV",
	"SRLV",
	"DIVV",
	"DIVVU",
	"REMV",
	"REMVU",
	"MULV",
	"MULVU",
	"ADDV",
	"ADDVU",
	"SUBV",
	"SUBVU",
	"DSBH",
	"DSHD",
	"TRUNCFV",
	"TRUNCDV",
	"TRUNCFW",
	"TRUNCDW",
	"MOVWU",
	"MOVFV",
	"MOVDV",
	"MOVVF",
	"MOVVD",
	"VMOVB",
	"VMOVH",
	"VMOVW",

	"MOVV",
	"MOVVL",
	"MOVVR",
	"SLLV",
	"SRAV",
	"SRLV",
	"ROTRV",
	"DIVV",
	"DIVVU",
	"REMV",
	"REMVU",
	"MULV",
	"MULVU",
	"MULHV",
	"MULHVU",
	"ADDV",
	"ADDVU",
	"SUBV",
	"SUBVU",
	"TRUNCFV",
	"TRUNCDV",
	"TRUNCFW",
	"TRUNCDW",
	"MOVWU",
	"MOVFV",
	"MOVDV",
	"MOVVF",
	"MOVVD",
	"AMSWAPB",
	"AMSWAPH",
	"AMSWAPW",
	"AMSWAPV",

	ADDU	R13, R14, R19	// 01cd9821
	ADDV	R5, R9, R10	// 0125502c
	ADDVU	R13, R14, R19	// 01cd982d

//	LADDW imm ',' sreg ',' rreg
//	{
//		outcode(int($1), &$2, int($4), &$6);
//	}
	ADD	$15176, R14, R9	// 21c93b48
	ADD	$-9, R5, R8	// 20a8fff7
	ADDU	$10, R9, R9	// 2529000a
	ADDV	$15176, R14, R9	// 61c93b48
	ADDV	$-9, R5, R8	// 60a8fff7
	ADDVU	$10, R9, R9	// 6529000a

//	LADDW rreg ',' rreg
//	{

	MOVV	R7, R6
	SC	R7, (R4)
	BEQ	R7, -4(PC)
	MOVW	R6, ret+16(FP)
	DBAR
	RET

TEXT ·Xadd64(SB), NOSPLIT, $0-24
	MOVV	ptr+0(FP), R4
	MOVV	delta+8(FP), R5
	DBAR
	LLV	(R4), R6
	ADDVU	R6, R5, R7
	MOVV	R7, R6
	SCV	R7, (R4)
	BEQ	R7, -4(PC)
	MOVV	R6, ret+16(FP)
	DBAR
	RET

TEXT ·Xchg(SB), NOSPLIT, $0-20
	MOVV	ptr+0(FP), R4
	MOVW	new+8(FP), R5

	DBAR
	MOVV	R5, R6

// uint64 Xadd64(uint64 volatile *ptr, int64 delta)
// Atomically:
//	*val += delta;
//	return *val;
TEXT ·Xadd64(SB), NOSPLIT, $0-24
	MOVV	ptr+0(FP), R2
	MOVV	delta+8(FP), R3
	SYNC
	LLV	(R2), R1
	ADDVU	R1, R3, R4
	MOVV	R4, R1
	SCV	R4, (R2)
	BEQ	R4, -4(PC)
	MOVV	R1, ret+16(FP)
	SYNC
	RET

// uint32 Xchg(ptr *uint32, new uint32)
// Atomically:
//	old := *ptr;
//	*ptr = new;