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:

	MOVW	b_base+12(FP), R4
	MOVW	a_len+4(FP), R1
	MOVW	b_len+16(FP), R2
	BEQ	R3, R4, samebytes
	SGTU	R1, R2, R7
	MOVW	R1, R8
	CMOVN	R7, R2, R8	// R8 is min(R1, R2)

	ADDU	R3, R8	// R3 is current byte in a, R8 is last byte in a to compare
loop:
	BEQ	R3, R8, samebytes

	MOVBU	(R3), R6
	ADDU	$1, R3
	MOVBU	(R4), R7
	ADDU	$1, R4
	BEQ	R6, R7 , loop

	SGTU	R6, R7, R8
	MOVW	$-1, R6
	CMOVZ	R8, R6, R8

	MOVV	a+0(FP), R1
	MOVV	b+8(FP), R2
	BEQ	R1, R2, eq
	MOVV	size+16(FP), R3
	ADDV	R1, R3, R4

	// chunk size is 16
	SGTU	$16, R3, R8
	BEQ	R0, R8, chunk_entry

byte_loop:
	BNE	R1, R4, byte_test
	MOVV	$1, R1
	MOVB	R1, ret+24(FP)
	RET
byte_test:
	MOVBU	(R1), R6
	ADDV	$1, R1
	MOVBU	(R2), R7
	ADDV	$1, R2
	BEQ	R6, R7, byte_loop
	JMP	not_eq

chunk_entry:

	MOVV	R18, -24(R5)
	MOVV	$1234, R16
	MOVV	R16, -32(R5)

	MOVV	$SYS_clone, R2
	SYSCALL
	BEQ	R7, 2(PC)
	SUBVU	R2, R0, R2	// caller expects negative errno

	// In parent, return.
	BEQ	R2, 3(PC)
	MOVW	R2, ret+40(FP)
	RET

	// In child, on new stack.
	MOVV	-32(R29), R16
	MOVV	$1234, R1
	BEQ	R16, R1, 2(PC)
	MOVV	R0, 0(R0)

	// Initialize m->procid to Linux tid
	MOVV	$SYS_gettid, R2

	MOVW	R12, -16(R1)
	MOVW	R13, -12(R1)
	MOVW	R14, -8(R1)
	MOVW	R15, -4(R1)
	JMP	f_large

f_words:
	BEQ	R1, R6, f_tail
	ADDU	$4, R1
	MOVW	0(R2), R8
	ADDU	$4, R2
	MOVW	R8, -4(R1)
	JMP	f_words

f_tail:
	BEQ	R1, R5, ret
	MOVWLO	-1(R4), R8
	MOVWLO	R8, -1(R5)

ret:
	RET

f_large_ua:
	BEQ	R1, R7, f_words_ua
	ADDU	$32, R1
	MOVWHI	0(R2), R8
	MOVWHI	4(R2), R9
	MOVWHI	8(R2), R10

	MOVW	R4, R1
	SC	R4, (R2)	// *R2 = R4
	BEQ	R4, try_xadd
	SYNC
	MOVW	R1, ret+8(FP)
	RET

// uint32 Xchg(ptr *uint32, new uint32)
// Atomically:
//	old := *ptr;
//	*ptr = new;
//	return old;
TEXT ·Xchg(SB),NOSPLIT,$0-12
	MOVW	ptr+0(FP), R2
	MOVW	new+4(FP), R5
	SYNC
try_xchg:
	MOVW	R5, R3
	LL	(R2), R1	// R1 = *R2
	SC	R3, (R2)	// *R2 = R3
	BEQ	R3, try_xchg
	SYNC
	MOVW	R1, ret+8(FP)
	RET

	BNE	R0, R11, byte_loop

chunk16_loop:
	BEQ	R0, R10, byte_loop
	MOVV	(R3), R6
	MOVV	(R4), R7
	BNE	R6, R7, byte_loop
	MOVV	8(R3), R13
	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:

	BEQ   zero512setup

	ANDCC $7, R3, R15
	BEQ   zero512xsetup // at least 8 byte aligned

	// zero bytes up to 8 byte alignment

	ANDCC $1, R3, R15 // check for byte alignment
	BEQ   byte2
	MOVB  R0, 0(R3)   // zero 1 byte
	ADD   $1, R3      // bump ptr by 1
	ADD   $-1, R4

byte2:
	ANDCC $2, R3, R15 // check for 2 byte alignment
	BEQ   byte4
	MOVH  R0, 0(R3)   // zero 2 bytes

	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:
	BEQ	R4, R9, done
	MOVB	(R5), R7
	ADDV	$1, R5

	// do one byte at a time until 8-aligned
	AND	$7, R1, R5
	BEQ	R5, words
	MOVB	(R2), R4
	ADDV	$1, R2
	MOVB	R4, (R1)
	ADDV	$1, R1
	JMP	-6(PC)

words:
	// do 8 bytes at a time if there is room
	ADDV	$-7, R6, R3 // R3 is end pointer-7

	SGTU	R3, R1, R5
	BEQ	R5, out
	MOVV	(R2), R4
	ADDV	$8, R2
	MOVV	R4, (R1)
	ADDV	$8, R1
	JMP	-6(PC)

out:
	BEQ	R1, R6, done
	MOVB	(R2), R4
	ADDV	$1, R2
	MOVB	R4, (R1)