diff options
Diffstat (limited to 'REORG.TODO/sysdeps/i386/i586/strlen.S')
-rw-r--r-- | REORG.TODO/sysdeps/i386/i586/strlen.S | 182 |
1 files changed, 182 insertions, 0 deletions
diff --git a/REORG.TODO/sysdeps/i386/i586/strlen.S b/REORG.TODO/sysdeps/i386/i586/strlen.S new file mode 100644 index 0000000000..cfea2e020f --- /dev/null +++ b/REORG.TODO/sysdeps/i386/i586/strlen.S @@ -0,0 +1,182 @@ +/* strlen -- Compute length of NUL terminated string. + Highly optimized version for ix86, x>=5. + Copyright (C) 1995-2017 Free Software Foundation, Inc. + This file is part of the GNU C Library. + Contributed by Ulrich Drepper, <drepper@gnu.ai.mit.edu>. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <sysdep.h> +#include "asm-syntax.h" + +/* This version is especially optimized for the i586 (and following?) + processors. This is mainly done by using the two pipelines. The + version optimized for i486 is weak in this aspect because to get + as much parallelism we have to execute some *more* instructions. + + The code below is structured to reflect the pairing of the instructions + as *I think* it is. I have no processor data book to verify this. + If you find something you think is incorrect let me know. */ + + +/* The magic value which is used throughout in the whole code. */ +#define magic 0xfefefeff + +#define PARMS 4 /* no space for saved regs */ +#define STR PARMS + + .text +ENTRY (strlen) + + movl STR(%esp), %eax + movl $3, %edx /* load mask (= 3) */ + + andl %eax, %edx /* separate last two bits of address */ + + jz L(1) /* aligned => start loop */ + jp L(0) /* exactly two bits set */ + + cmpb %dh, (%eax) /* is byte NUL? */ + je L(2) /* yes => return */ + + incl %eax /* increment pointer */ + cmpb %dh, (%eax) /* is byte NUL? */ + + je L(2) /* yes => return */ + + incl %eax /* increment pointer */ + xorl $2, %edx + + jz L(1) + +L(0): cmpb %dh, (%eax) /* is byte NUL? */ + je L(2) /* yes => return */ + + incl %eax /* increment pointer */ + xorl %edx, %edx /* We need %edx == 0 for later */ + + /* We exit the loop if adding MAGIC_BITS to LONGWORD fails to + change any of the hole bits of LONGWORD. + + 1) Is this safe? Will it catch all the zero bytes? + Suppose there is a byte with all zeros. Any carry bits + propagating from its left will fall into the hole at its + least significant bit and stop. Since there will be no + carry from its most significant bit, the LSB of the + byte to the left will be unchanged, and the zero will be + detected. + + 2) Is this worthwhile? Will it ignore everything except + zero bytes? Suppose every byte of LONGWORD has a bit set + somewhere. There will be a carry into bit 8. If bit 8 + is set, this will carry into bit 16. If bit 8 is clear, + one of bits 9-15 must be set, so there will be a carry + into bit 16. Similarly, there will be a carry into bit + 24. If one of bits 24-31 is set, there will be a carry + into bit 32 (=carry flag), so all of the hole bits will + be changed. + + Note: %edx == 0 in any case here. */ + +L(1): + movl (%eax), %ecx /* get word (= 4 bytes) in question */ + addl $4, %eax /* adjust pointer for *next* word */ + + subl %ecx, %edx /* first step to negate word */ + addl $magic, %ecx /* add magic word */ + + decl %edx /* complete negation of word */ + jnc L(3) /* previous addl caused overflow? */ + + xorl %ecx, %edx /* (word+magic)^word */ + + andl $~magic, %edx /* any of the carry flags set? */ + + jne L(3) /* yes => determine byte */ + + + movl (%eax), %ecx /* get word (= 4 bytes) in question */ + addl $4, %eax /* adjust pointer for *next* word */ + + subl %ecx, %edx /* first step to negate word */ + addl $magic, %ecx /* add magic word */ + + decl %edx /* complete negation of word */ + jnc L(3) /* previous addl caused overflow? */ + + xorl %ecx, %edx /* (word+magic)^word */ + + andl $~magic, %edx /* any of the carry flags set? */ + + jne L(3) /* yes => determine byte */ + + + movl (%eax), %ecx /* get word (= 4 bytes) in question */ + addl $4, %eax /* adjust pointer for *next* word */ + + subl %ecx, %edx /* first step to negate word */ + addl $magic, %ecx /* add magic word */ + + decl %edx /* complete negation of word */ + jnc L(3) /* previous addl caused overflow? */ + + xorl %ecx, %edx /* (word+magic)^word */ + + andl $~magic, %edx /* any of the carry flags set? */ + + jne L(3) /* yes => determine byte */ + + + movl (%eax), %ecx /* get word (= 4 bytes) in question */ + addl $4, %eax /* adjust pointer for *next* word */ + + subl %ecx, %edx /* first step to negate word */ + addl $magic, %ecx /* add magic word */ + + decl %edx /* complete negation of word */ + jnc L(3) /* previous addl caused overflow? */ + + xorl %ecx, %edx /* (word+magic)^word */ + + andl $~magic, %edx /* any of the carry flags set? */ + + je L(1) /* no => start loop again */ + + +L(3): subl $4, %eax /* correct too early pointer increment */ + subl $magic, %ecx + + cmpb $0, %cl /* lowest byte NUL? */ + jz L(2) /* yes => return */ + + inc %eax /* increment pointer */ + testb %ch, %ch /* second byte NUL? */ + + jz L(2) /* yes => return */ + + shrl $16, %ecx /* make upper bytes accessible */ + incl %eax /* increment pointer */ + + cmpb $0, %cl /* is third byte NUL? */ + jz L(2) /* yes => return */ + + incl %eax /* increment pointer */ + +L(2): subl STR(%esp), %eax /* now compute the length as difference + between start and terminating NUL + character */ + ret +END (strlen) +libc_hidden_builtin_def (strlen) |