| 1 | /* |
| 2 | * __strlen_aarch64_sve - compute the length of a string |
| 3 | * |
| 4 | * Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. |
| 5 | * See https://llvm.org/LICENSE.txt for license information. |
| 6 | * SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception |
| 7 | */ |
| 8 | |
| 9 | #if __ARM_FEATURE_SVE |
| 10 | /* Assumptions: |
| 11 | * |
| 12 | * ARMv8-a, AArch64 |
| 13 | * SVE Available. |
| 14 | */ |
| 15 | |
| 16 | .arch armv8-a+sve |
| 17 | .text |
| 18 | |
| 19 | .globl __strlen_aarch64_sve |
| 20 | .type __strlen_aarch64_sve, %function |
| 21 | .p2align 4 |
| 22 | __strlen_aarch64_sve: |
| 23 | setffr /* initialize FFR */ |
| 24 | ptrue p2.b /* all ones; loop invariant */ |
| 25 | mov x1, 0 /* initialize length */ |
| 26 | nop |
| 27 | |
| 28 | /* Read a vector's worth of bytes, stopping on first fault. */ |
| 29 | 0: ldff1b z0.b, p2/z, [x0, x1] |
| 30 | nop |
| 31 | rdffrs p0.b, p2/z |
| 32 | b.nlast 2f |
| 33 | |
| 34 | /* First fault did not fail: the whole vector is valid. |
| 35 | Avoid depending on the contents of FFR beyond the branch. */ |
| 36 | incb x1, all /* speculate increment */ |
| 37 | cmpeq p1.b, p2/z, z0.b, 0 /* loop if no zeros */ |
| 38 | b.none 0b |
| 39 | decb x1, all /* undo speculate */ |
| 40 | |
| 41 | /* Zero found. Select the bytes before the first and count them. */ |
| 42 | 1: brkb p0.b, p2/z, p1.b |
| 43 | incp x1, p0.b |
| 44 | mov x0, x1 |
| 45 | ret |
| 46 | |
| 47 | /* First fault failed: only some of the vector is valid. |
| 48 | Perform the comparison only on the valid bytes. */ |
| 49 | 2: cmpeq p1.b, p0/z, z0.b, 0 |
| 50 | b.any 1b |
| 51 | |
| 52 | /* No zero found. Re-init FFR, increment, and loop. */ |
| 53 | setffr |
| 54 | incp x1, p0.b |
| 55 | b 0b |
| 56 | |
| 57 | .size __strlen_aarch64_sve, . - __strlen_aarch64_sve |
| 58 | #endif |
| 59 | |