| Vineet Gupta | 5210d1e | 2013-01-18 15:12:18 +0530 | [diff] [blame] | 1 | /* |
| 2 | * Copyright (C) 2004, 2007-2010, 2011-2012 Synopsys, Inc. (www.synopsys.com) |
| 3 | * |
| 4 | * This program is free software; you can redistribute it and/or modify |
| 5 | * it under the terms of the GNU General Public License version 2 as |
| 6 | * published by the Free Software Foundation. |
| 7 | */ |
| 8 | |
| 9 | /* ARC700 has a relatively long pipeline and branch prediction, so we want |
| 10 | to avoid branches that are hard to predict. On the other hand, the |
| 11 | presence of the norm instruction makes it easier to operate on whole |
| 12 | words branch-free. */ |
| 13 | |
| 14 | #include <asm/linkage.h> |
| 15 | |
| 16 | ARC_ENTRY strchr |
| 17 | extb_s r1,r1 |
| 18 | asl r5,r1,8 |
| 19 | bmsk r2,r0,1 |
| 20 | or r5,r5,r1 |
| 21 | mov_s r3,0x01010101 |
| 22 | breq.d r2,r0,.Laligned |
| 23 | asl r4,r5,16 |
| 24 | sub_s r0,r0,r2 |
| 25 | asl r7,r2,3 |
| 26 | ld_s r2,[r0] |
| 27 | #ifdef __LITTLE_ENDIAN__ |
| 28 | asl r7,r3,r7 |
| 29 | #else |
| 30 | lsr r7,r3,r7 |
| 31 | #endif |
| 32 | or r5,r5,r4 |
| 33 | ror r4,r3 |
| 34 | sub r12,r2,r7 |
| 35 | bic_s r12,r12,r2 |
| 36 | and r12,r12,r4 |
| 37 | brne.d r12,0,.Lfound0_ua |
| 38 | xor r6,r2,r5 |
| 39 | ld.a r2,[r0,4] |
| 40 | sub r12,r6,r7 |
| 41 | bic r12,r12,r6 |
| 42 | and r7,r12,r4 |
| 43 | breq r7,0,.Loop ; For speed, we want this branch to be unaligned. |
| 44 | b .Lfound_char ; Likewise this one. |
| 45 | ; /* We require this code address to be unaligned for speed... */ |
| 46 | .Laligned: |
| 47 | ld_s r2,[r0] |
| 48 | or r5,r5,r4 |
| 49 | ror r4,r3 |
| 50 | ; /* ... so that this code address is aligned, for itself and ... */ |
| 51 | .Loop: |
| 52 | sub r12,r2,r3 |
| 53 | bic_s r12,r12,r2 |
| 54 | and r12,r12,r4 |
| 55 | brne.d r12,0,.Lfound0 |
| 56 | xor r6,r2,r5 |
| 57 | ld.a r2,[r0,4] |
| 58 | sub r12,r6,r3 |
| 59 | bic r12,r12,r6 |
| 60 | and r7,r12,r4 |
| 61 | breq r7,0,.Loop /* ... so that this branch is unaligned. */ |
| 62 | ; Found searched-for character. r0 has already advanced to next word. |
| 63 | #ifdef __LITTLE_ENDIAN__ |
| 64 | /* We only need the information about the first matching byte |
| 65 | (i.e. the least significant matching byte) to be exact, |
| 66 | hence there is no problem with carry effects. */ |
| 67 | .Lfound_char: |
| 68 | sub r3,r7,1 |
| 69 | bic r3,r3,r7 |
| 70 | norm r2,r3 |
| 71 | sub_s r0,r0,1 |
| 72 | asr_s r2,r2,3 |
| 73 | j.d [blink] |
| 74 | sub_s r0,r0,r2 |
| 75 | |
| 76 | .balign 4 |
| 77 | .Lfound0_ua: |
| 78 | mov r3,r7 |
| 79 | .Lfound0: |
| 80 | sub r3,r6,r3 |
| 81 | bic r3,r3,r6 |
| 82 | and r2,r3,r4 |
| 83 | or_s r12,r12,r2 |
| 84 | sub_s r3,r12,1 |
| 85 | bic_s r3,r3,r12 |
| 86 | norm r3,r3 |
| 87 | add_s r0,r0,3 |
| 88 | asr_s r12,r3,3 |
| 89 | asl.f 0,r2,r3 |
| 90 | sub_s r0,r0,r12 |
| 91 | j_s.d [blink] |
| 92 | mov.pl r0,0 |
| 93 | #else /* BIG ENDIAN */ |
| 94 | .Lfound_char: |
| 95 | lsr r7,r7,7 |
| 96 | |
| 97 | bic r2,r7,r6 |
| 98 | norm r2,r2 |
| 99 | sub_s r0,r0,4 |
| 100 | asr_s r2,r2,3 |
| 101 | j.d [blink] |
| 102 | add_s r0,r0,r2 |
| 103 | |
| 104 | .Lfound0_ua: |
| 105 | mov_s r3,r7 |
| 106 | .Lfound0: |
| 107 | asl_s r2,r2,7 |
| 108 | or r7,r6,r4 |
| 109 | bic_s r12,r12,r2 |
| 110 | sub r2,r7,r3 |
| 111 | or r2,r2,r6 |
| 112 | bic r12,r2,r12 |
| 113 | bic.f r3,r4,r12 |
| 114 | norm r3,r3 |
| 115 | |
| 116 | add.pl r3,r3,1 |
| 117 | asr_s r12,r3,3 |
| 118 | asl.f 0,r2,r3 |
| 119 | add_s r0,r0,r12 |
| 120 | j_s.d [blink] |
| 121 | mov.mi r0,0 |
| 122 | #endif /* ENDIAN */ |
| 123 | ARC_EXIT strchr |