1*4882a593Smuzhiyun/* SPDX-License-Identifier: GPL-2.0-only */ 2*4882a593Smuzhiyun/* 3*4882a593Smuzhiyun * Copyright (C) 2004, 2007-2010, 2011-2012 Synopsys, Inc. (www.synopsys.com) 4*4882a593Smuzhiyun */ 5*4882a593Smuzhiyun 6*4882a593Smuzhiyun/* ARC700 has a relatively long pipeline and branch prediction, so we want 7*4882a593Smuzhiyun to avoid branches that are hard to predict. On the other hand, the 8*4882a593Smuzhiyun presence of the norm instruction makes it easier to operate on whole 9*4882a593Smuzhiyun words branch-free. */ 10*4882a593Smuzhiyun 11*4882a593Smuzhiyun#include <linux/linkage.h> 12*4882a593Smuzhiyun 13*4882a593SmuzhiyunENTRY_CFI(strchr) 14*4882a593Smuzhiyun extb_s r1,r1 15*4882a593Smuzhiyun asl r5,r1,8 16*4882a593Smuzhiyun bmsk r2,r0,1 17*4882a593Smuzhiyun or r5,r5,r1 18*4882a593Smuzhiyun mov_s r3,0x01010101 19*4882a593Smuzhiyun breq.d r2,r0,.Laligned 20*4882a593Smuzhiyun asl r4,r5,16 21*4882a593Smuzhiyun sub_s r0,r0,r2 22*4882a593Smuzhiyun asl r7,r2,3 23*4882a593Smuzhiyun ld_s r2,[r0] 24*4882a593Smuzhiyun#ifdef __LITTLE_ENDIAN__ 25*4882a593Smuzhiyun asl r7,r3,r7 26*4882a593Smuzhiyun#else 27*4882a593Smuzhiyun lsr r7,r3,r7 28*4882a593Smuzhiyun#endif 29*4882a593Smuzhiyun or r5,r5,r4 30*4882a593Smuzhiyun ror r4,r3 31*4882a593Smuzhiyun sub r12,r2,r7 32*4882a593Smuzhiyun bic_s r12,r12,r2 33*4882a593Smuzhiyun and r12,r12,r4 34*4882a593Smuzhiyun brne.d r12,0,.Lfound0_ua 35*4882a593Smuzhiyun xor r6,r2,r5 36*4882a593Smuzhiyun ld.a r2,[r0,4] 37*4882a593Smuzhiyun sub r12,r6,r7 38*4882a593Smuzhiyun bic r12,r12,r6 39*4882a593Smuzhiyun#ifdef __LITTLE_ENDIAN__ 40*4882a593Smuzhiyun and r7,r12,r4 41*4882a593Smuzhiyun breq r7,0,.Loop ; For speed, we want this branch to be unaligned. 42*4882a593Smuzhiyun b .Lfound_char ; Likewise this one. 43*4882a593Smuzhiyun#else 44*4882a593Smuzhiyun and r12,r12,r4 45*4882a593Smuzhiyun breq r12,0,.Loop ; For speed, we want this branch to be unaligned. 46*4882a593Smuzhiyun lsr_s r12,r12,7 47*4882a593Smuzhiyun bic r2,r7,r6 48*4882a593Smuzhiyun b.d .Lfound_char_b 49*4882a593Smuzhiyun and_s r2,r2,r12 50*4882a593Smuzhiyun#endif 51*4882a593Smuzhiyun; /* We require this code address to be unaligned for speed... */ 52*4882a593Smuzhiyun.Laligned: 53*4882a593Smuzhiyun ld_s r2,[r0] 54*4882a593Smuzhiyun or r5,r5,r4 55*4882a593Smuzhiyun ror r4,r3 56*4882a593Smuzhiyun; /* ... so that this code address is aligned, for itself and ... */ 57*4882a593Smuzhiyun.Loop: 58*4882a593Smuzhiyun sub r12,r2,r3 59*4882a593Smuzhiyun bic_s r12,r12,r2 60*4882a593Smuzhiyun and r12,r12,r4 61*4882a593Smuzhiyun brne.d r12,0,.Lfound0 62*4882a593Smuzhiyun xor r6,r2,r5 63*4882a593Smuzhiyun ld.a r2,[r0,4] 64*4882a593Smuzhiyun sub r12,r6,r3 65*4882a593Smuzhiyun bic r12,r12,r6 66*4882a593Smuzhiyun and r7,r12,r4 67*4882a593Smuzhiyun breq r7,0,.Loop /* ... so that this branch is unaligned. */ 68*4882a593Smuzhiyun ; Found searched-for character. r0 has already advanced to next word. 69*4882a593Smuzhiyun#ifdef __LITTLE_ENDIAN__ 70*4882a593Smuzhiyun/* We only need the information about the first matching byte 71*4882a593Smuzhiyun (i.e. the least significant matching byte) to be exact, 72*4882a593Smuzhiyun hence there is no problem with carry effects. */ 73*4882a593Smuzhiyun.Lfound_char: 74*4882a593Smuzhiyun sub r3,r7,1 75*4882a593Smuzhiyun bic r3,r3,r7 76*4882a593Smuzhiyun norm r2,r3 77*4882a593Smuzhiyun sub_s r0,r0,1 78*4882a593Smuzhiyun asr_s r2,r2,3 79*4882a593Smuzhiyun j.d [blink] 80*4882a593Smuzhiyun sub_s r0,r0,r2 81*4882a593Smuzhiyun 82*4882a593Smuzhiyun .balign 4 83*4882a593Smuzhiyun.Lfound0_ua: 84*4882a593Smuzhiyun mov r3,r7 85*4882a593Smuzhiyun.Lfound0: 86*4882a593Smuzhiyun sub r3,r6,r3 87*4882a593Smuzhiyun bic r3,r3,r6 88*4882a593Smuzhiyun and r2,r3,r4 89*4882a593Smuzhiyun or_s r12,r12,r2 90*4882a593Smuzhiyun sub_s r3,r12,1 91*4882a593Smuzhiyun bic_s r3,r3,r12 92*4882a593Smuzhiyun norm r3,r3 93*4882a593Smuzhiyun add_s r0,r0,3 94*4882a593Smuzhiyun asr_s r12,r3,3 95*4882a593Smuzhiyun asl.f 0,r2,r3 96*4882a593Smuzhiyun sub_s r0,r0,r12 97*4882a593Smuzhiyun j_s.d [blink] 98*4882a593Smuzhiyun mov.pl r0,0 99*4882a593Smuzhiyun#else /* BIG ENDIAN */ 100*4882a593Smuzhiyun.Lfound_char: 101*4882a593Smuzhiyun lsr r7,r7,7 102*4882a593Smuzhiyun 103*4882a593Smuzhiyun bic r2,r7,r6 104*4882a593Smuzhiyun.Lfound_char_b: 105*4882a593Smuzhiyun norm r2,r2 106*4882a593Smuzhiyun sub_s r0,r0,4 107*4882a593Smuzhiyun asr_s r2,r2,3 108*4882a593Smuzhiyun j.d [blink] 109*4882a593Smuzhiyun add_s r0,r0,r2 110*4882a593Smuzhiyun 111*4882a593Smuzhiyun.Lfound0_ua: 112*4882a593Smuzhiyun mov_s r3,r7 113*4882a593Smuzhiyun.Lfound0: 114*4882a593Smuzhiyun asl_s r2,r2,7 115*4882a593Smuzhiyun or r7,r6,r4 116*4882a593Smuzhiyun bic_s r12,r12,r2 117*4882a593Smuzhiyun sub r2,r7,r3 118*4882a593Smuzhiyun or r2,r2,r6 119*4882a593Smuzhiyun bic r12,r2,r12 120*4882a593Smuzhiyun bic.f r3,r4,r12 121*4882a593Smuzhiyun norm r3,r3 122*4882a593Smuzhiyun 123*4882a593Smuzhiyun add.pl r3,r3,1 124*4882a593Smuzhiyun asr_s r12,r3,3 125*4882a593Smuzhiyun asl.f 0,r2,r3 126*4882a593Smuzhiyun add_s r0,r0,r12 127*4882a593Smuzhiyun j_s.d [blink] 128*4882a593Smuzhiyun mov.mi r0,0 129*4882a593Smuzhiyun#endif /* ENDIAN */ 130*4882a593SmuzhiyunEND_CFI(strchr) 131