MOVV	R5, R4
	MOVV	R6, R5
	MOVV	R7, R6
	MOVV	R8, R7
	MOVV	R9, R8
	MOVV	R10, R9
	SYSCALL
	MOVV	R0, R5      // r2 is not used. Always set to 0.
	MOVW	$-4096, R12
	BGEU	R12, R4, ok
	SUBVU	R4, R0, R6  // errno
	MOVV	$-1, R4     // r1
	RET
ok:
	// r1 already in R4
	MOVV	R0, R6     // errno

	BNE	R6, check
	RET

check:
	SGTU	R4, R5, R7
	BNE	R7, backward

	ADDV	R4, R6, R9 // end pointer

	// if the two pointers are not of same alignments, do byte copying
	SUBVU	R5, R4, R7
	AND	$7, R7
	BNE	R7, out

	// if less than 8 bytes, do byte copying
	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

	"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",
	"VMOVD",
	"LAST",

	MOVV	a3+24(FP), R6
	MOVV	$0, R7
	MOVV	$0, R8
	MOVV	$0, R9
	MOVV	trap+0(FP), R11	// syscall entry
	SYSCALL
	MOVW	$-4096, R12
	BGEU	R12, R4, ok
	MOVV	$-1, R12
	MOVV	R12, r1+32(FP)	// r1
	SUBVU	R4, R0, R4
	MOVV	R4, err+40(FP)	// errno
	RET
ok:
	MOVV	R4, r1+32(FP)	// r1
	MOVV	R0, err+40(FP)	// errno
	RET

TEXT ·rawSyscallNoError(SB),NOSPLIT,$0-48
	MOVV	a1+8(FP), R4
	MOVV	a2+16(FP), R5

	"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",
	"AMCASB",
	"AMCASH",

	BNE	R0, R15, byte_loop

	PCALIGN	$16
chunk16_loop:
	BEQ	R0, R14, byte_loop
	MOVV	(R4), R8
	MOVV	(R6), R9
	BNE	R8, R9, byte_loop
	MOVV	8(R4), R16
	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:

	SUBU	R6, R26, R27	// 0346d823
	SUBV	R16, R17, R26	// 0230d02e
	SUBVU	R16, R17, R26	// 0230d02f

//	LSUBW imm ',' sreg ',' rreg
//	{
//		outcode(int($1), &$2, int($4), &$6);
//	}
	SUB	$-3126, R17, R22	// 22360c36
	SUB	$3126, R17, R22		// 2236f3ca
	SUBU	$16384, R17, R12	// 262cc000
	SUBV	$-6122, R10, R9		// 614917ea
	SUBV	$6122, R10, R9		// 6149e816
	SUBVU	$1203, R17, R12		// 662cfb4d

//	LSUBW rreg ',' rreg
//	{

	MOVV	R2, 0(R29)	// save old g on stack
	MOVV	(g_stack+stack_hi)(R2), R2
	SUBVU	R3, R2
	MOVV	R2, 8(R29)	// save depth in old g stack (can't just save SP, as stack might be copied during a callback)
	JAL	(R25)

	// Restore g, stack pointer. R2 is return value.
	MOVV	0(R29), g
	JAL	runtime·save_g(SB)
	MOVV	(g_stack+stack_hi)(g), R5
	MOVV	8(R29), R6
	SUBVU	R6, R5
	MOVV	R5, R29

	MOVW	R2, ret+16(FP)
	RET

	MOVV	R13, 0(R3)	// save old g on stack
	MOVV	(g_stack+stack_hi)(R13), R13
	SUBVU	R12, R13
	MOVV	R13, 8(R3)	// save depth in old g stack (can't just save SP, as stack might be copied during a callback)
	JAL	(R25)

	// Restore g, stack pointer. R4 is return value.
	MOVV	0(R3), g
	JAL	runtime·save_g(SB)
	MOVV	(g_stack+stack_hi)(g), R5
	MOVV	8(R3), R6
	SUBVU	R6, R5
	MOVV	R5, R3

	MOVW	R4, ret+16(FP)
	RET

		{name: "ADDVconst", argLength: 1, reg: gp11sp, asm: "ADDVU", aux: "Int64"}, // arg0 + auxInt. auxInt is 32-bit, also in other *const ops.
		{name: "SUBV", argLength: 2, reg: gp21, asm: "SUBVU"},                      // arg0 - arg1
		{name: "SUBVconst", argLength: 1, reg: gp11, asm: "SUBVU", aux: "Int64"},   // arg0 - auxInt

		{name: "MULV", argLength: 2, reg: gp21, asm: "MULV", commutative: true, typ: "Int64"},      // arg0 * arg1