| 1 | /* SPDX-License-Identifier: GPL-2.0-only */
|
| 2 | /*
|
| 3 | * Copyright (c) 2013-2021, Arm Limited.
|
| 4 | *
|
| 5 | * Adapted from the original at:
|
| 6 | * https://github.com/ARM-software/optimized-routines/blob/e823e3abf5f89ecb/string/aarch64/memcmp.S
|
| 7 | */
|
| 8 |
|
| 9 | |
| 10 | |
| 11 |
|
| 12 | /* Assumptions:
|
| 13 | *
|
| 14 | * ARMv8-a, AArch64, unaligned accesses.
|
| 15 | */
|
| 16 |
|
| 17 | |
| 18 |
|
| 19 | /* Parameters and result. */
|
| 20 | |
| 21 | |
| 22 | |
| 23 | |
| 24 |
|
| 25 | /* Internal variables. */
|
| 26 | |
| 27 | |
| 28 | |
| 29 | |
| 30 | |
| 31 | |
| 32 | |
| 33 | |
| 34 |
|
| 35 | SYM_FUNC_START(__pi_memcmp)
|
| 36 | subs limit, limit, 8
|
| 37 | b.lo L(less8)
|
| 38 |
|
| 39 | ldr data1, [src1], 8
|
| 40 | ldr data2, [src2], 8
|
| 41 | cmp data1, data2
|
| 42 | b.ne L(return)
|
| 43 |
|
| 44 | subs limit, limit, 8
|
| 45 | b.gt L(more16)
|
| 46 |
|
| 47 | ldr data1, [src1, limit]
|
| 48 | ldr data2, [src2, limit]
|
| 49 | b L(return)
|
| 50 |
|
| 51 | L(more16):
|
| 52 | ldr data1, [src1], 8
|
| 53 | ldr data2, [src2], 8
|
| 54 | cmp data1, data2
|
| 55 | bne L(return)
|
| 56 |
|
| 57 | /* Jump directly to comparing the last 16 bytes for 32 byte (or less)
|
| 58 | strings. */
|
| 59 | subs limit, limit, 16
|
| 60 | b.ls L(last_bytes)
|
| 61 |
|
| 62 | /* We overlap loads between 0-32 bytes at either side of SRC1 when we
|
| 63 | try to align, so limit it only to strings larger than 128 bytes. */
|
| 64 | cmp limit, 96
|
| 65 | b.ls L(loop16)
|
| 66 |
|
| 67 | /* Align src1 and adjust src2 with bytes not yet done. */
|
| 68 | and tmp1, src1, 15
|
| 69 | add limit, limit, tmp1
|
| 70 | sub src1, src1, tmp1
|
| 71 | sub src2, src2, tmp1
|
| 72 |
|
| 73 | /* Loop performing 16 bytes per iteration using aligned src1.
|
| 74 | Limit is pre-decremented by 16 and must be larger than zero.
|
| 75 | Exit if <= 16 bytes left to do or if the data is not equal. */
|
| 76 | .p2align 4
|
| 77 | L(loop16):
|
| 78 | ldp data1, data1h, [src1], 16
|
| 79 | ldp data2, data2h, [src2], 16
|
| 80 | subs limit, limit, 16
|
| 81 | ccmp data1, data2, 0, hi
|
| 82 | ccmp data1h, data2h, 0, eq
|
| 83 | b.eq L(loop16)
|
| 84 |
|
| 85 | cmp data1, data2
|
| 86 | bne L(return)
|
| 87 | mov data1, data1h
|
| 88 | mov data2, data2h
|
| 89 | cmp data1, data2
|
| 90 | bne L(return)
|
| 91 |
|
| 92 | /* Compare last 1-16 bytes using unaligned access. */
|
| 93 | L(last_bytes):
|
| 94 | add src1, src1, limit
|
| 95 | add src2, src2, limit
|
| 96 | ldp data1, data1h, [src1]
|
| 97 | ldp data2, data2h, [src2]
|
| 98 | cmp data1, data2
|
| 99 | bne L(return)
|
| 100 | mov data1, data1h
|
| 101 | mov data2, data2h
|
| 102 | cmp data1, data2
|
| 103 |
|
| 104 | /* Compare data bytes and set return value to 0, -1 or 1. */
|
| 105 | L(return):
|
| 106 | rev data1, data1
|
| 107 | rev data2, data2
|
| 108 | cmp data1, data2
|
| 109 | L(ret_eq):
|
| 110 | cset result, ne
|
| 111 | cneg result, result, lo
|
| 112 | ret
|
| 113 |
|
| 114 | .p2align 4
|
| 115 | /* Compare up to 8 bytes. Limit is [-8..-1]. */
|
| 116 | L(less8):
|
| 117 | adds limit, limit, 4
|
| 118 | b.lo L(less4)
|
| 119 | ldr data1w, [src1], 4
|
| 120 | ldr data2w, [src2], 4
|
| 121 | cmp data1w, data2w
|
| 122 | b.ne L(return)
|
| 123 | sub limit, limit, 4
|
| 124 | L(less4):
|
| 125 | adds limit, limit, 4
|
| 126 | beq L(ret_eq)
|
| 127 | L(byte_loop):
|
| 128 | ldrb data1w, [src1], 1
|
| 129 | ldrb data2w, [src2], 1
|
| 130 | subs limit, limit, 1
|
| 131 | ccmp data1w, data2w, 0, ne /* NZCV = 0b0000. */
|
| 132 | b.eq L(byte_loop)
|
| 133 | sub result, data1w, data2w
|
| 134 | ret
|
| 135 | SYM_FUNC_END(__pi_memcmp)
|
| 136 | SYM_FUNC_ALIAS_WEAK(memcmp, __pi_memcmp)
|
| 137 | EXPORT_SYMBOL_NOKASAN(memcmp)
|