/* Optimized memcpy for Qualcomm Falkor processor.
   Copyright (C) 2017 Free Software Foundation, Inc.

   This file is part of the GNU C Library.

   The GNU C Library is free software; you can redistribute it and/or
   modify it under the terms of the GNU Lesser General Public
   License as published by the Free Software Foundation; either
   version 2.1 of the License, or (at your option) any later version.

   The GNU C Library is distributed in the hope that it will be useful,
   but WITHOUT ANY WARRANTY; without even the implied warranty of
   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
   Lesser General Public License for more details.

   You should have received a copy of the GNU Lesser General Public
   License along with the GNU C Library.  If not, see
   <http://www.gnu.org/licenses/>.  */

#include <sysdep.h>

/* Assumptions:

   ARMv8-a, AArch64, falkor, unaligned accesses.  */

#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 A_hw	w7
#define tmp1	x14

/* Copies are split into 3 main cases:

   1. Small copies of up to 32 bytes
   2. Medium copies of 33..128 bytes which are fully unrolled
   3. Large copies of more than 128 bytes.

   Large copies align the sourceto a quad word and use an unrolled loop
   processing 64 bytes per iteration.

   FALKOR-SPECIFIC DESIGN:

   The smallest copies (32 bytes or less) focus on optimal pipeline usage,
   which is why the redundant copies of 0-3 bytes have been replaced with
   conditionals, since the former would unnecessarily break across multiple
   issue groups.  The medium copy group has been enlarged to 128 bytes since
   bumping up the small copies up to 32 bytes allows us to do that without
   cost and also allows us to reduce the size of the prep code before loop64.

   All copies are done only via two registers r6 and r7.  This is to ensure
   that all loads hit a single hardware prefetcher which can get correctly
   trained to prefetch a single stream.

   The non-temporal stores help optimize cache utilization.  */

#if IS_IN (libc)
ENTRY_ALIGN (__memcpy_falkor, 6)

	cmp	count, 32
	add	srcend, src, count
	add	dstend, dstin, count
	b.ls	L(copy32)
	ldp	A_l, A_h, [src]
	cmp	count, 128
	stp	A_l, A_h, [dstin]
	b.hi	L(copy_long)

	/* Medium copies: 33..128 bytes.  */
	sub	tmp1, count, 1
	ldp	A_l, A_h, [src, 16]
	stp	A_l, A_h, [dstin, 16]
	tbz	tmp1, 6, 1f
	ldp	A_l, A_h, [src, 32]
	stp	A_l, A_h, [dstin, 32]
	ldp	A_l, A_h, [src, 48]
	stp	A_l, A_h, [dstin, 48]
	ldp	A_l, A_h, [srcend, -64]
	stp	A_l, A_h, [dstend, -64]
	ldp	A_l, A_h, [srcend, -48]
	stp	A_l, A_h, [dstend, -48]
1:
	ldp	A_l, A_h, [srcend, -32]
	stp	A_l, A_h, [dstend, -32]
	ldp	A_l, A_h, [srcend, -16]
	stp	A_l, A_h, [dstend, -16]
	ret

	.p2align 4
	/* Small copies: 0..32 bytes.  */
L(copy32):
	/* 16-32 */
	cmp	count, 16
	b.lo	1f
	ldp	A_l, A_h, [src]
	stp	A_l, A_h, [dstin]
	ldp	A_l, A_h, [srcend, -16]
	stp	A_l, A_h, [dstend, -16]
	ret
	.p2align 4
1:
	/* 8-15 */
	tbz	count, 3, 1f
	ldr	A_l, [src]
	str	A_l, [dstin]
	ldr	A_l, [srcend, -8]
	str	A_l, [dstend, -8]
	ret
	.p2align 4
1:
	/* 4-7 */
	tbz	count, 2, 1f
	ldr	A_lw, [src]
	str	A_lw, [dstin]
	ldr	A_lw, [srcend, -4]
	str	A_lw, [dstend, -4]
	ret
	.p2align 4
1:
	/* 2-3 */
	tbz	count, 1, 1f
	ldrh	A_lw, [src]
	strh	A_lw, [dstin]
	ldrh	A_lw, [srcend, -2]
	strh	A_lw, [dstend, -2]
	ret
	.p2align 4
1:
	/* 0-1 */
	tbz	count, 0, 1f
	ldrb	A_lw, [src]
	strb	A_lw, [dstin]
1:
	ret

	/* Align SRC to 16 bytes and copy; that way at least one of the
	   accesses is aligned throughout the copy sequence.

	   The count is off by 0 to 15 bytes, but this is OK because we trim
	   off the last 64 bytes to copy off from the end.  Due to this the
	   loop never runs out of bounds.  */
	.p2align 6
L(copy_long):
	sub	count, count, 64 + 16
	and	tmp1, src, 15
	bic	src, src, 15
	sub	dst, dstin, tmp1
	add	count, count, tmp1

L(loop64):
	ldp	A_l, A_h, [src, 16]!
	stnp	A_l, A_h, [dst, 16]
	ldp	A_l, A_h, [src, 16]!
	subs	count, count, 64
	stnp	A_l, A_h, [dst, 32]
	ldp	A_l, A_h, [src, 16]!
	stnp	A_l, A_h, [dst, 48]
	ldp	A_l, A_h, [src, 16]!
	stnp	A_l, A_h, [dst, 64]
	add	dst, dst, 64
	b.hi	L(loop64)

	/* Write the last full set of 64 bytes.  The remainder is at most 64
	   bytes, so it is safe to always copy 64 bytes from the end even if
	   there is just 1 byte left.  */
L(last64):
	ldp	A_l, A_h, [srcend, -64]
	stnp	A_l, A_h, [dstend, -64]
	ldp	A_l, A_h, [srcend, -48]
	stnp	A_l, A_h, [dstend, -48]
	ldp	A_l, A_h, [srcend, -32]
	stnp	A_l, A_h, [dstend, -32]
	ldp	A_l, A_h, [srcend, -16]
	stnp	A_l, A_h, [dstend, -16]
	ret

END (__memcpy_falkor)
libc_hidden_builtin_def (__memcpy_falkor)
#endif