1/*- 2 * Copyright (c) 2012 The NetBSD Foundation, Inc. 3 * All rights reserved. 4 * 5 * This code is derived from software contributed to The NetBSD Foundation 6 * by Matt Thomas of 3am Software Foundry. 7 * 8 * Redistribution and use in source and binary forms, with or without 9 * modification, are permitted provided that the following conditions 10 * are met: 11 * 1. Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * 2. Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in the 15 * documentation and/or other materials provided with the distribution. 16 * 17 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 18 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 19 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 20 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 21 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 22 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 23 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 24 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 25 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 26 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 27 * POSSIBILITY OF SUCH DAMAGE. 28 */ 29 30#include <machine/asm.h> 31 32RCSID("$NetBSD: strlen_neon.S,v 1.3 2012/12/28 05:15:08 matt Exp $") 33 .text 34 35ENTRY(strlen) 36 mov ip, r0 /* we use r0 for return value */ 37 ands r1, r0, #15 /* verify qword alignment */ 38 neg r0, r1 /* subtract misalignment from count */ 39 veor q2, q2, q2 /* clear mask */ 40 mov r3, #7 /* NBBY - 1 */ 41 vdup.32 q3, r3 /* dup throughout q3 */ 42 movw r3, #0x0404 /* magic since there are 4 bytes per U32 */ 43 orr r3, r3, lsl #16 /* copy to upper 16 bits */ 44 beq .Lmain_loop 45 veor q0, q0, q0 /* clear q0 */ 46 vmvn q2, q2 /* set all 16 bytes of mask to all 1s */ 47 bic ip, ip, #15 /* qword align string address */ 48 lsl r1, r1, #3 /* convert to bits */ 49 cmp r1, #64 50 rsbgt r1, r1, #128 /* > 64? BE so we are shifting LSW right */ 51 movgt r2, #0 /* > 64? leave MSW alone */ 52 rsble r2, r1, #64 /* <=64? BE so we are shifting MSW right */ 53 movle r1, #64 /* <=64? clear LSW */ 54 vmov d0, r1, r2 /* set shifts for lower and upper halves */ 55 vmovl.u32 q0, d0 /* 2 U32 -> 2 U64 */ 56 vshl.u64 q2, q2, q0 /* shift */ 57 /* 58 * Main loop. Load 16 bytes, do a clz, 59 */ 60.Lmain_loop: 61 vld1.64 {d0, d1}, [ip:128]! /* load qword */ 62#ifdef __ARMEL__ 63 vrev64.8 q0, q0 /* convert to BE for clz */ 64#endif 65 vswp d0, d1 /* swap dwords to get BE qword */ 66 vorr q0, q0, q2 /* or "in" leading byte mask */ 67 veor q2, q2, q2 /* clear leading byte mask */ 68 vceq.i8 q1, q0, #0 /* test each byte for 0 */ 69 /* Why couldn't there be a 64-bit CLZ? */ 70 vclz.u32 q1, q1 /* count leading zeroes to find the 0 byte */ 71 vadd.u32 q1, q1, q3 /* round up to byte bounary */ 72 vshr.u32 q1, q1, #3 /* convert to bytes */ 73 vmovn.u32 d0, q1 /* 4 I32 -> 4 I16 */ 74 vmovn.u16 d0, q0 /* 4 I16 -> 4 I8 */ 75 vmov r2, s0 /* get counts */ 76 eors r2, r2, r3 /* xor with 0x04040404 */ 77 addeq r0, #16 /* 0? no NULs */ 78 beq .Lmain_loop /* get next qword */ 79 clz ip, r2 /* count leading zeros */ 80 mov r2, r2, lsl ip /* discard them */ 81 mov ip, ip, lsr #3 /* divide leading zeroes by 8 */ 82 add r0, r0, ip, lsl #2 /* multiple by 4 and add to count */ 83 and r2, r2, #(3 << 29) 84 add r0, r0, r2, lsr #29 85 RET /* and return. */ 86END(strlen) 87