projects / apple / libc.git / blame
| Commit | Line | Data |
|---|---|---|
| eb1cde05 A |
1 | /* |
| 2 | * Copyright (c) 2005 Apple Computer, Inc. All rights reserved. | |
| 3 | * | |
| 4 | * @APPLE_LICENSE_HEADER_START@ | |
| 5 | * | |
| 6 | * The contents of this file constitute Original Code as defined in and | |
| 7 | * are subject to the Apple Public Source License Version 1.1 (the | |
| 8 | * "License"). You may not use this file except in compliance with the | |
| 9 | * License. Please obtain a copy of the License at | |
| 10 | * http://www.apple.com/publicsource and read it before using this file. | |
| 11 | * | |
| 12 | * This Original Code and all software distributed under the License are | |
| 13 | * distributed on an "AS IS" basis, WITHOUT WARRANTY OF ANY KIND, EITHER | |
| 14 | * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES, | |
| 15 | * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY, | |
| 16 | * FITNESS FOR A PARTICULAR PURPOSE OR NON-INFRINGEMENT. Please see the | |
| 17 | * License for the specific language governing rights and limitations | |
| 18 | * under the License. | |
| 19 | * | |
| 20 | * @APPLE_LICENSE_HEADER_END@ | |
| 21 | */ | |
| 22 | ||
| 23 | /* | |
| 24 | * Strlen, for processors with SSE3. | |
| 25 | * | |
| 26 | * Note that all memory references must be aligned, in order to avoid spurious | |
| 27 | * page faults. Thus we have to load the aligned 16-byte chunk containing the | |
| 28 | * first byte of the operand, then mask out false 0s that may occur before the | |
| 29 | * first byte. | |
| 30 | * | |
| 31 | * We favor the fall-through (ie, short operand) path. | |
| 32 | */ | |
| 33 | ||
| 34 | .text | |
| 35 | .globl _strlen | |
| 36 | .align 4, 0x90 | |
| 37 | _strlen: // size_t strlen(char *b); | |
| 38 | movl 4(%esp),%edx // get ptr to string | |
| 39 | pxor %xmm0,%xmm0 // zero %xmm0 | |
| 40 | movl %edx,%ecx // copy ptr | |
| 41 | andl $(-16),%edx // 16-byte align ptr | |
| 42 | orl $(-1),%eax | |
| 43 | pcmpeqb (%edx),%xmm0 // check whole qw for 0s | |
| 44 | andl $15,%ecx // get #bytes in aligned dq before operand | |
| 45 | shl %cl,%eax // create mask for the bytes of aligned dq in operand | |
| 46 | pmovmskb %xmm0,%ecx // collect mask of 0-bytes | |
| 47 | andl %eax,%ecx // mask out any 0s that occur before 1st byte | |
| 48 | jz LEnterLoop // no 0-bytes (ie, 1-bits), so enter by-16 loop | |
| 49 | ||
| 50 | // We've found a 0-byte. | |
| 51 | // %edx = aligned address of 16-byte block containing the terminating 0-byte | |
| 52 | // %ecx = compare bit vector | |
| 53 | ||
| 54 | LFoundIt: | |
| 55 | bsf %ecx,%eax // find first 1-bit (ie, first 0-byte) | |
| 56 | movl 4(%esp),%ecx // recover ptr to 1st byte in string | |
| 57 | addl %edx,%eax // get address of the 0-byte | |
| 58 | subl %ecx,%eax // subtract address of 1st byte to get string length | |
| 59 | ret | |
| 60 | ||
| 61 | // Loop over aligned 16-byte blocks: | |
| 62 | // %edx = address of previous block | |
| 63 | ||
| 64 | LEnterLoop: | |
| 65 | pxor %xmm0,%xmm0 // get some 0-bytes | |
| 66 | addl $16,%edx // advance ptr | |
| 67 | LLoop: | |
| 68 | movdqa (%edx),%xmm1 // get next chunk | |
| 69 | addl $16,%edx | |
| 70 | pcmpeqb %xmm0,%xmm1 // check for 0s | |
| 71 | pmovmskb %xmm1,%ecx // collect mask of 0-bytes | |
| 72 | test %ecx,%ecx // any 0-bytes? | |
| 73 | jz LLoop // no 0-bytes, so get next dq | |
| 74 | ||
| 75 | subl $16,%edx // back up ptr | |
| 76 | jmp LFoundIt | |
| 77 | ||
| 78 |