projects / thirdparty / glibc.git / blame
| Commit | Line | Data |
|---|---|---|
| e73015f2 | 1 | /* strlen with SSE2 and BSF |
| 04277e02 | 2 | Copyright (C) 2010-2019 Free Software Foundation, Inc. |
| e73015f2 L |
3 | Contributed by Intel Corporation. |
| 4 | This file is part of the GNU C Library. | |
| 5 | ||
| 6 | The GNU C Library is free software; you can redistribute it and/or | |
| 7 | modify it under the terms of the GNU Lesser General Public | |
| 8 | License as published by the Free Software Foundation; either | |
| 9 | version 2.1 of the License, or (at your option) any later version. | |
| 10 | ||
| 11 | The GNU C Library is distributed in the hope that it will be useful, | |
| 12 | but WITHOUT ANY WARRANTY; without even the implied warranty of | |
| 13 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
| 14 | Lesser General Public License for more details. | |
| 15 | ||
| 16 | You should have received a copy of the GNU Lesser General Public | |
| 59ba27a6 PE |
17 | License along with the GNU C Library; if not, see |
| 18 | <http://www.gnu.org/licenses/>. */ | |
| e73015f2 | 19 | |
| 4f41c682 | 20 | #if defined SHARED && IS_IN (libc) |
| e73015f2 L |
21 | |
| 22 | #include <sysdep.h> | |
| e73015f2 L |
23 | |
| 24 | #define CFI_PUSH(REG) \ | |
| 25 | cfi_adjust_cfa_offset (4); \ | |
| 26 | cfi_rel_offset (REG, 0) | |
| 27 | ||
| 28 | #define CFI_POP(REG) \ | |
| 29 | cfi_adjust_cfa_offset (-4); \ | |
| 30 | cfi_restore (REG) | |
| 31 | ||
| 32 | #define PUSH(REG) pushl REG; CFI_PUSH (REG) | |
| 33 | #define POP(REG) popl REG; CFI_POP (REG) | |
| 34 | #define PARMS 4 + 8 /* Preserve ESI and EDI. */ | |
| 35 | #define STR PARMS | |
| 36 | #define ENTRANCE PUSH (%esi); PUSH (%edi); cfi_remember_state | |
| 37 | #define RETURN POP (%edi); POP (%esi); ret; \ | |
| 38 | cfi_restore_state; cfi_remember_state | |
| 39 | ||
| 48882a1a | 40 | .text |
| e73015f2 L |
41 | ENTRY ( __strlen_sse2_bsf) |
| 42 | ENTRANCE | |
| 43 | mov STR(%esp), %edi | |
| 44 | xor %eax, %eax | |
| 45 | mov %edi, %ecx | |
| 46 | and $0x3f, %ecx | |
| 47 | pxor %xmm0, %xmm0 | |
| 48 | cmp $0x30, %ecx | |
| 49 | ja L(next) | |
| 50 | movdqu (%edi), %xmm1 | |
| 51 | pcmpeqb %xmm1, %xmm0 | |
| 52 | pmovmskb %xmm0, %edx | |
| 53 | test %edx, %edx | |
| 54 | jnz L(exit_less16) | |
| 55 | mov %edi, %eax | |
| 56 | and $-16, %eax | |
| 57 | jmp L(align16_start) | |
| 58 | L(next): | |
| 59 | ||
| 60 | mov %edi, %eax | |
| 61 | and $-16, %eax | |
| 62 | pcmpeqb (%eax), %xmm0 | |
| 63 | mov $-1, %esi | |
| 64 | sub %eax, %ecx | |
| 65 | shl %cl, %esi | |
| 66 | pmovmskb %xmm0, %edx | |
| 67 | and %esi, %edx | |
| 68 | jnz L(exit) | |
| 69 | L(align16_start): | |
| 70 | pxor %xmm0, %xmm0 | |
| 71 | pxor %xmm1, %xmm1 | |
| 72 | pxor %xmm2, %xmm2 | |
| 73 | pxor %xmm3, %xmm3 | |
| 74 | .p2align 4 | |
| 75 | L(align16_loop): | |
| 76 | pcmpeqb 16(%eax), %xmm0 | |
| 77 | pmovmskb %xmm0, %edx | |
| 78 | test %edx, %edx | |
| 79 | jnz L(exit16) | |
| 80 | ||
| 81 | pcmpeqb 32(%eax), %xmm1 | |
| 82 | pmovmskb %xmm1, %edx | |
| 83 | test %edx, %edx | |
| 84 | jnz L(exit32) | |
| 85 | ||
| 86 | pcmpeqb 48(%eax), %xmm2 | |
| 87 | pmovmskb %xmm2, %edx | |
| 88 | test %edx, %edx | |
| 89 | jnz L(exit48) | |
| 90 | ||
| 91 | pcmpeqb 64(%eax), %xmm3 | |
| 92 | pmovmskb %xmm3, %edx | |
| 93 | lea 64(%eax), %eax | |
| 94 | test %edx, %edx | |
| 95 | jz L(align16_loop) | |
| 96 | L(exit): | |
| 97 | sub %edi, %eax | |
| 98 | L(exit_less16): | |
| 99 | bsf %edx, %edx | |
| 100 | add %edx, %eax | |
| 101 | RETURN | |
| 102 | L(exit16): | |
| 103 | sub %edi, %eax | |
| 104 | bsf %edx, %edx | |
| 105 | add %edx, %eax | |
| 106 | add $16, %eax | |
| 107 | RETURN | |
| 108 | L(exit32): | |
| 109 | sub %edi, %eax | |
| 110 | bsf %edx, %edx | |
| 111 | add %edx, %eax | |
| 112 | add $32, %eax | |
| 113 | RETURN | |
| 114 | L(exit48): | |
| 115 | sub %edi, %eax | |
| 116 | bsf %edx, %edx | |
| 117 | add %edx, %eax | |
| 118 | add $48, %eax | |
| 119 | POP (%edi) | |
| 120 | POP (%esi) | |
| 121 | ret | |
| 122 | ||
| 123 | END ( __strlen_sse2_bsf) | |
| 124 | ||
| 125 | #endif |