/*
 * memcpy - copy memory area
 *
 * Copyright (c) 2019-2023, Arm Limited.
 * SPDX-License-Identifier: MIT OR Apache-2.0 WITH LLVM-exception
 */

/* Assumptions:
 *
 * ARMv8-a, AArch64, Advanced SIMD, unaligned accesses.
 *
 */

#include "asmdefs.h"

#define dstin	x0
#define src	x1
#define count	x2
#define dst	x3
#define srcend	x4
#define dstend	x5
#define A_l	x6
#define A_lw	w6
#define A_h	x7
#define B_l	x8
#define B_lw	w8
#define B_h	x9
#define C_lw	w10
#define tmp1	x14

#define A_q	q0
#define B_q	q1
#define C_q	q2
#define D_q	q3
#define E_q	q4
#define F_q	q5
#define G_q	q6
#define H_q	q7

/* This implementation handles overlaps and supports both memcpy and memmove
   from a single entry point.  It uses unaligned accesses and branchless
   sequences to keep the code small, simple and improve performance.

   Copies are split into 3 main cases: small copies of up to 32 bytes, medium
   copies of up to 128 bytes, and large copies.  The overhead of the overlap
   check is negligible since it is only required for large copies.

   Large copies use a software pipelined loop processing 64 bytes per iteration.
   The source pointer is 16-byte aligned to minimize unaligned accesses.
   The loop tail is handled by always copying 64 bytes from the end.
*/

ENTRY_ALIAS (__memmove_aarch64_simd)
ENTRY (__memcpy_aarch64_simd)
	PTR_ARG (0)
	PTR_ARG (1)
	SIZE_ARG (2)
	add	srcend, src, count
	cmp	count, 128
	b.hi	L(copy_long)
	add	dstend, dstin, count
	cmp	count, 32
	b.hi	L(copy32_128)
	nop

	/* Small copies: 0..32 bytes.  */
	cmp	count, 16
	b.lo	L(copy16)
	ldr	A_q, [src]
	ldr	B_q, [srcend, -16]
	str	A_q, [dstin]
	str	B_q, [dstend, -16]
	ret

	.p2align 4
	/* Medium copies: 33..128 bytes.  */
L(copy32_128):
	ldp	A_q, B_q, [src]
	ldp	C_q, D_q, [srcend, -32]
	cmp	count, 64
	b.hi	L(copy128)
	stp	A_q, B_q, [dstin]
	stp	C_q, D_q, [dstend, -32]
	ret

	.p2align 4
	/* Copy 8-15 bytes.  */
L(copy16):
	tbz	count, 3, L(copy8)
	ldr	A_l, [src]
	ldr	A_h, [srcend, -8]
	str	A_l, [dstin]
	str	A_h, [dstend, -8]
	ret

	/* Copy 4-7 bytes.  */
L(copy8):
	tbz	count, 2, L(copy4)
	ldr	A_lw, [src]
	ldr	B_lw, [srcend, -4]
	str	A_lw, [dstin]
	str	B_lw, [dstend, -4]
	ret

	/* Copy 65..128 bytes.  */
L(copy128):
	ldp	E_q, F_q, [src, 32]
	cmp	count, 96
	b.ls	L(copy96)
	ldp	G_q, H_q, [srcend, -64]
	stp	G_q, H_q, [dstend, -64]
L(copy96):
	stp	A_q, B_q, [dstin]
	stp	E_q, F_q, [dstin, 32]
	stp	C_q, D_q, [dstend, -32]
	ret

	/* Copy 0..3 bytes using a branchless sequence.  */
L(copy4):
	cbz	count, L(copy0)
	lsr	tmp1, count, 1
	ldrb	A_lw, [src]
	ldrb	C_lw, [srcend, -1]
	ldrb	B_lw, [src, tmp1]
	strb	A_lw, [dstin]
	strb	B_lw, [dstin, tmp1]
	strb	C_lw, [dstend, -1]
L(copy0):
	ret

	.p2align 3
	/* Copy more than 128 bytes.  */
L(copy_long):
	add	dstend, dstin, count

	/* Use backwards copy if there is an overlap.  */
	sub	tmp1, dstin, src
	cmp	tmp1, count
	b.lo	L(copy_long_backwards)

	/* Copy 16 bytes and then align src to 16-byte alignment.  */
	ldr	D_q, [src]
	and	tmp1, src, 15
	bic	src, src, 15
	sub	dst, dstin, tmp1
	add	count, count, tmp1	/* Count is now 16 too large.  */
	ldp	A_q, B_q, [src, 16]
	str	D_q, [dstin]
	ldp	C_q, D_q, [src, 48]
	subs	count, count, 128 + 16	/* Test and readjust count.  */
	b.ls	L(copy64_from_end)
L(loop64):
	stp	A_q, B_q, [dst, 16]
	ldp	A_q, B_q, [src, 80]
	stp	C_q, D_q, [dst, 48]
	ldp	C_q, D_q, [src, 112]
	add	src, src, 64
	add	dst, dst, 64
	subs	count, count, 64
	b.hi	L(loop64)

	/* Write the last iteration and copy 64 bytes from the end.  */
L(copy64_from_end):
	ldp	E_q, F_q, [srcend, -64]
	stp	A_q, B_q, [dst, 16]
	ldp	A_q, B_q, [srcend, -32]
	stp	C_q, D_q, [dst, 48]
	stp	E_q, F_q, [dstend, -64]
	stp	A_q, B_q, [dstend, -32]
	ret

	.p2align 4
	nop

	/* Large backwards copy for overlapping copies.
	   Copy 16 bytes and then align srcend to 16-byte alignment.  */
L(copy_long_backwards):
	cbz	tmp1, L(copy0)
	ldr	D_q, [srcend, -16]
	and	tmp1, srcend, 15
	bic	srcend, srcend, 15
	sub	count, count, tmp1
	ldp	A_q, B_q, [srcend, -32]
	str	D_q, [dstend, -16]
	ldp	C_q, D_q, [srcend, -64]
	sub	dstend, dstend, tmp1
	subs	count, count, 128
	b.ls	L(copy64_from_start)

L(loop64_backwards):
	str	B_q, [dstend, -16]
	str	A_q, [dstend, -32]
	ldp	A_q, B_q, [srcend, -96]
	str	D_q, [dstend, -48]
	str	C_q, [dstend, -64]!
	ldp	C_q, D_q, [srcend, -128]
	sub	srcend, srcend, 64
	subs	count, count, 64
	b.hi	L(loop64_backwards)

	/* Write the last iteration and copy 64 bytes from the start.  */
L(copy64_from_start):
	ldp	E_q, F_q, [src, 32]
	stp	A_q, B_q, [dstend, -32]
	ldp	A_q, B_q, [src]
	stp	C_q, D_q, [dstend, -64]
	stp	E_q, F_q, [dstin, 32]
	stp	A_q, B_q, [dstin]
	ret

END (__memcpy_aarch64_simd)