|
Packit |
6c4009 |
/* Copyright (C) 2012-2018 Free Software Foundation, Inc.
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
This file is part of the GNU C Library.
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
The GNU C Library is free software; you can redistribute it and/or
|
|
Packit |
6c4009 |
modify it under the terms of the GNU Lesser General Public
|
|
Packit |
6c4009 |
License as published by the Free Software Foundation; either
|
|
Packit |
6c4009 |
version 2.1 of the License, or (at your option) any later version.
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
The GNU C Library is distributed in the hope that it will be useful,
|
|
Packit |
6c4009 |
but WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
Packit |
6c4009 |
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
|
|
Packit |
6c4009 |
Lesser General Public License for more details.
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
You should have received a copy of the GNU Lesser General Public
|
|
Packit |
6c4009 |
License along with the GNU C Library. If not, see
|
|
Packit |
6c4009 |
<http://www.gnu.org/licenses/>. */
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
#include <sysdep.h>
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
/* Assumptions:
|
|
Packit |
6c4009 |
*
|
|
Packit |
6c4009 |
* ARMv8-a, AArch64, unaligned accesses.
|
|
Packit |
6c4009 |
*
|
|
Packit |
6c4009 |
*/
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
#define dstin x0
|
|
Packit |
6c4009 |
#define src x1
|
|
Packit |
6c4009 |
#define count x2
|
|
Packit |
6c4009 |
#define dst x3
|
|
Packit |
6c4009 |
#define srcend x4
|
|
Packit |
6c4009 |
#define dstend x5
|
|
Packit |
6c4009 |
#define A_l x6
|
|
Packit |
6c4009 |
#define A_lw w6
|
|
Packit |
6c4009 |
#define A_h x7
|
|
Packit |
6c4009 |
#define A_hw w7
|
|
Packit |
6c4009 |
#define B_l x8
|
|
Packit |
6c4009 |
#define B_lw w8
|
|
Packit |
6c4009 |
#define B_h x9
|
|
Packit |
6c4009 |
#define C_l x10
|
|
Packit |
6c4009 |
#define C_h x11
|
|
Packit |
6c4009 |
#define D_l x12
|
|
Packit |
6c4009 |
#define D_h x13
|
|
Packit |
6c4009 |
#define E_l src
|
|
Packit |
6c4009 |
#define E_h count
|
|
Packit |
6c4009 |
#define F_l srcend
|
|
Packit |
6c4009 |
#define F_h dst
|
|
Packit |
6c4009 |
#define G_l count
|
|
Packit |
6c4009 |
#define G_h dst
|
|
Packit |
6c4009 |
#define tmp1 x14
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
/* Copies are split into 3 main cases: small copies of up to 16 bytes,
|
|
Packit |
6c4009 |
medium copies of 17..96 bytes which are fully unrolled. Large copies
|
|
Packit |
6c4009 |
of more than 96 bytes align the destination and use an unrolled loop
|
|
Packit |
6c4009 |
processing 64 bytes per iteration.
|
|
Packit |
6c4009 |
In order to share code with memmove, small and medium copies read all
|
|
Packit |
6c4009 |
data before writing, allowing any kind of overlap. So small, medium
|
|
Packit |
6c4009 |
and large backwards memmoves are handled by falling through into memcpy.
|
|
Packit |
6c4009 |
Overlapping large forward memmoves use a loop that copies backwards.
|
|
Packit |
6c4009 |
*/
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
#ifndef MEMMOVE
|
|
Packit |
6c4009 |
# define MEMMOVE memmove
|
|
Packit |
6c4009 |
#endif
|
|
Packit |
6c4009 |
#ifndef MEMCPY
|
|
Packit |
6c4009 |
# define MEMCPY memcpy
|
|
Packit |
6c4009 |
#endif
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
ENTRY_ALIGN (MEMMOVE, 6)
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
DELOUSE (0)
|
|
Packit |
6c4009 |
DELOUSE (1)
|
|
Packit |
6c4009 |
DELOUSE (2)
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
sub tmp1, dstin, src
|
|
Packit |
6c4009 |
cmp count, 96
|
|
Packit |
6c4009 |
ccmp tmp1, count, 2, hi
|
|
Packit |
6c4009 |
b.lo L(move_long)
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
/* Common case falls through into memcpy. */
|
|
Packit |
6c4009 |
END (MEMMOVE)
|
|
Packit |
6c4009 |
libc_hidden_builtin_def (MEMMOVE)
|
|
Packit |
6c4009 |
ENTRY (MEMCPY)
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
DELOUSE (0)
|
|
Packit |
6c4009 |
DELOUSE (1)
|
|
Packit |
6c4009 |
DELOUSE (2)
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
prfm PLDL1KEEP, [src]
|
|
Packit |
6c4009 |
add srcend, src, count
|
|
Packit |
6c4009 |
add dstend, dstin, count
|
|
Packit |
6c4009 |
cmp count, 16
|
|
Packit |
6c4009 |
b.ls L(copy16)
|
|
Packit |
6c4009 |
cmp count, 96
|
|
Packit |
6c4009 |
b.hi L(copy_long)
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
/* Medium copies: 17..96 bytes. */
|
|
Packit |
6c4009 |
sub tmp1, count, 1
|
|
Packit |
6c4009 |
ldp A_l, A_h, [src]
|
|
Packit |
6c4009 |
tbnz tmp1, 6, L(copy96)
|
|
Packit |
6c4009 |
ldp D_l, D_h, [srcend, -16]
|
|
Packit |
6c4009 |
tbz tmp1, 5, 1f
|
|
Packit |
6c4009 |
ldp B_l, B_h, [src, 16]
|
|
Packit |
6c4009 |
ldp C_l, C_h, [srcend, -32]
|
|
Packit |
6c4009 |
stp B_l, B_h, [dstin, 16]
|
|
Packit |
6c4009 |
stp C_l, C_h, [dstend, -32]
|
|
Packit |
6c4009 |
1:
|
|
Packit |
6c4009 |
stp A_l, A_h, [dstin]
|
|
Packit |
6c4009 |
stp D_l, D_h, [dstend, -16]
|
|
Packit |
6c4009 |
ret
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
.p2align 4
|
|
Packit |
6c4009 |
/* Small copies: 0..16 bytes. */
|
|
Packit |
6c4009 |
L(copy16):
|
|
Packit |
6c4009 |
cmp count, 8
|
|
Packit |
6c4009 |
b.lo 1f
|
|
Packit |
6c4009 |
ldr A_l, [src]
|
|
Packit |
6c4009 |
ldr A_h, [srcend, -8]
|
|
Packit |
6c4009 |
str A_l, [dstin]
|
|
Packit |
6c4009 |
str A_h, [dstend, -8]
|
|
Packit |
6c4009 |
ret
|
|
Packit |
6c4009 |
.p2align 4
|
|
Packit |
6c4009 |
1:
|
|
Packit |
6c4009 |
tbz count, 2, 1f
|
|
Packit |
6c4009 |
ldr A_lw, [src]
|
|
Packit |
6c4009 |
ldr A_hw, [srcend, -4]
|
|
Packit |
6c4009 |
str A_lw, [dstin]
|
|
Packit |
6c4009 |
str A_hw, [dstend, -4]
|
|
Packit |
6c4009 |
ret
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
/* Copy 0..3 bytes. Use a branchless sequence that copies the same
|
|
Packit |
6c4009 |
byte 3 times if count==1, or the 2nd byte twice if count==2. */
|
|
Packit |
6c4009 |
1:
|
|
Packit |
6c4009 |
cbz count, 2f
|
|
Packit |
6c4009 |
lsr tmp1, count, 1
|
|
Packit |
6c4009 |
ldrb A_lw, [src]
|
|
Packit |
6c4009 |
ldrb A_hw, [srcend, -1]
|
|
Packit |
6c4009 |
ldrb B_lw, [src, tmp1]
|
|
Packit |
6c4009 |
strb A_lw, [dstin]
|
|
Packit |
6c4009 |
strb B_lw, [dstin, tmp1]
|
|
Packit |
6c4009 |
strb A_hw, [dstend, -1]
|
|
Packit |
6c4009 |
2: ret
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
.p2align 4
|
|
Packit |
6c4009 |
/* Copy 64..96 bytes. Copy 64 bytes from the start and
|
|
Packit |
6c4009 |
32 bytes from the end. */
|
|
Packit |
6c4009 |
L(copy96):
|
|
Packit |
6c4009 |
ldp B_l, B_h, [src, 16]
|
|
Packit |
6c4009 |
ldp C_l, C_h, [src, 32]
|
|
Packit |
6c4009 |
ldp D_l, D_h, [src, 48]
|
|
Packit |
6c4009 |
ldp E_l, E_h, [srcend, -32]
|
|
Packit |
6c4009 |
ldp F_l, F_h, [srcend, -16]
|
|
Packit |
6c4009 |
stp A_l, A_h, [dstin]
|
|
Packit |
6c4009 |
stp B_l, B_h, [dstin, 16]
|
|
Packit |
6c4009 |
stp C_l, C_h, [dstin, 32]
|
|
Packit |
6c4009 |
stp D_l, D_h, [dstin, 48]
|
|
Packit |
6c4009 |
stp E_l, E_h, [dstend, -32]
|
|
Packit |
6c4009 |
stp F_l, F_h, [dstend, -16]
|
|
Packit |
6c4009 |
ret
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
/* Align DST to 16 byte alignment so that we don't cross cache line
|
|
Packit |
6c4009 |
boundaries on both loads and stores. There are at least 96 bytes
|
|
Packit |
6c4009 |
to copy, so copy 16 bytes unaligned and then align. The loop
|
|
Packit |
6c4009 |
copies 64 bytes per iteration and prefetches one iteration ahead. */
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
.p2align 4
|
|
Packit |
6c4009 |
L(copy_long):
|
|
Packit |
6c4009 |
and tmp1, dstin, 15
|
|
Packit |
6c4009 |
bic dst, dstin, 15
|
|
Packit |
6c4009 |
ldp D_l, D_h, [src]
|
|
Packit |
6c4009 |
sub src, src, tmp1
|
|
Packit |
6c4009 |
add count, count, tmp1 /* Count is now 16 too large. */
|
|
Packit |
6c4009 |
ldp A_l, A_h, [src, 16]
|
|
Packit |
6c4009 |
stp D_l, D_h, [dstin]
|
|
Packit |
6c4009 |
ldp B_l, B_h, [src, 32]
|
|
Packit |
6c4009 |
ldp C_l, C_h, [src, 48]
|
|
Packit |
6c4009 |
ldp D_l, D_h, [src, 64]!
|
|
Packit |
6c4009 |
subs count, count, 128 + 16 /* Test and readjust count. */
|
|
Packit |
6c4009 |
b.ls L(last64)
|
|
Packit |
6c4009 |
L(loop64):
|
|
Packit |
6c4009 |
stp A_l, A_h, [dst, 16]
|
|
Packit |
6c4009 |
ldp A_l, A_h, [src, 16]
|
|
Packit |
6c4009 |
stp B_l, B_h, [dst, 32]
|
|
Packit |
6c4009 |
ldp B_l, B_h, [src, 32]
|
|
Packit |
6c4009 |
stp C_l, C_h, [dst, 48]
|
|
Packit |
6c4009 |
ldp C_l, C_h, [src, 48]
|
|
Packit |
6c4009 |
stp D_l, D_h, [dst, 64]!
|
|
Packit |
6c4009 |
ldp D_l, D_h, [src, 64]!
|
|
Packit |
6c4009 |
subs count, count, 64
|
|
Packit |
6c4009 |
b.hi L(loop64)
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
/* Write the last full set of 64 bytes. The remainder is at most 64
|
|
Packit |
6c4009 |
bytes, so it is safe to always copy 64 bytes from the end even if
|
|
Packit |
6c4009 |
there is just 1 byte left. */
|
|
Packit |
6c4009 |
L(last64):
|
|
Packit |
6c4009 |
ldp E_l, E_h, [srcend, -64]
|
|
Packit |
6c4009 |
stp A_l, A_h, [dst, 16]
|
|
Packit |
6c4009 |
ldp A_l, A_h, [srcend, -48]
|
|
Packit |
6c4009 |
stp B_l, B_h, [dst, 32]
|
|
Packit |
6c4009 |
ldp B_l, B_h, [srcend, -32]
|
|
Packit |
6c4009 |
stp C_l, C_h, [dst, 48]
|
|
Packit |
6c4009 |
ldp C_l, C_h, [srcend, -16]
|
|
Packit |
6c4009 |
stp D_l, D_h, [dst, 64]
|
|
Packit |
6c4009 |
stp E_l, E_h, [dstend, -64]
|
|
Packit |
6c4009 |
stp A_l, A_h, [dstend, -48]
|
|
Packit |
6c4009 |
stp B_l, B_h, [dstend, -32]
|
|
Packit |
6c4009 |
stp C_l, C_h, [dstend, -16]
|
|
Packit |
6c4009 |
ret
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
.p2align 4
|
|
Packit |
6c4009 |
L(move_long):
|
|
Packit |
6c4009 |
cbz tmp1, 3f
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
add srcend, src, count
|
|
Packit |
6c4009 |
add dstend, dstin, count
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
/* Align dstend to 16 byte alignment so that we don't cross cache line
|
|
Packit |
6c4009 |
boundaries on both loads and stores. There are at least 96 bytes
|
|
Packit |
6c4009 |
to copy, so copy 16 bytes unaligned and then align. The loop
|
|
Packit |
6c4009 |
copies 64 bytes per iteration and prefetches one iteration ahead. */
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
and tmp1, dstend, 15
|
|
Packit |
6c4009 |
ldp D_l, D_h, [srcend, -16]
|
|
Packit |
6c4009 |
sub srcend, srcend, tmp1
|
|
Packit |
6c4009 |
sub count, count, tmp1
|
|
Packit |
6c4009 |
ldp A_l, A_h, [srcend, -16]
|
|
Packit |
6c4009 |
stp D_l, D_h, [dstend, -16]
|
|
Packit |
6c4009 |
ldp B_l, B_h, [srcend, -32]
|
|
Packit |
6c4009 |
ldp C_l, C_h, [srcend, -48]
|
|
Packit |
6c4009 |
ldp D_l, D_h, [srcend, -64]!
|
|
Packit |
6c4009 |
sub dstend, dstend, tmp1
|
|
Packit |
6c4009 |
subs count, count, 128
|
|
Packit |
6c4009 |
b.ls 2f
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
nop
|
|
Packit |
6c4009 |
1:
|
|
Packit |
6c4009 |
stp A_l, A_h, [dstend, -16]
|
|
Packit |
6c4009 |
ldp A_l, A_h, [srcend, -16]
|
|
Packit |
6c4009 |
stp B_l, B_h, [dstend, -32]
|
|
Packit |
6c4009 |
ldp B_l, B_h, [srcend, -32]
|
|
Packit |
6c4009 |
stp C_l, C_h, [dstend, -48]
|
|
Packit |
6c4009 |
ldp C_l, C_h, [srcend, -48]
|
|
Packit |
6c4009 |
stp D_l, D_h, [dstend, -64]!
|
|
Packit |
6c4009 |
ldp D_l, D_h, [srcend, -64]!
|
|
Packit |
6c4009 |
subs count, count, 64
|
|
Packit |
6c4009 |
b.hi 1b
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
/* Write the last full set of 64 bytes. The remainder is at most 64
|
|
Packit |
6c4009 |
bytes, so it is safe to always copy 64 bytes from the start even if
|
|
Packit |
6c4009 |
there is just 1 byte left. */
|
|
Packit |
6c4009 |
2:
|
|
Packit |
6c4009 |
ldp G_l, G_h, [src, 48]
|
|
Packit |
6c4009 |
stp A_l, A_h, [dstend, -16]
|
|
Packit |
6c4009 |
ldp A_l, A_h, [src, 32]
|
|
Packit |
6c4009 |
stp B_l, B_h, [dstend, -32]
|
|
Packit |
6c4009 |
ldp B_l, B_h, [src, 16]
|
|
Packit |
6c4009 |
stp C_l, C_h, [dstend, -48]
|
|
Packit |
6c4009 |
ldp C_l, C_h, [src]
|
|
Packit |
6c4009 |
stp D_l, D_h, [dstend, -64]
|
|
Packit |
6c4009 |
stp G_l, G_h, [dstin, 48]
|
|
Packit |
6c4009 |
stp A_l, A_h, [dstin, 32]
|
|
Packit |
6c4009 |
stp B_l, B_h, [dstin, 16]
|
|
Packit |
6c4009 |
stp C_l, C_h, [dstin]
|
|
Packit |
6c4009 |
3: ret
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
END (MEMCPY)
|
|
Packit |
6c4009 |
libc_hidden_builtin_def (MEMCPY)
|