From d71ff82e7019e91ae9f76af391e7f677d421e205 Mon Sep 17 00:00:00 2001 From: Mike Frysinger Date: Fri, 26 Aug 2005 00:01:54 +0000 Subject: import sparc-optimized string functions from glibc --- libc/string/sparc/sparc32/strlen.S | 101 +++++++++++++++++++++++++++++++++++++ 1 file changed, 101 insertions(+) create mode 100644 libc/string/sparc/sparc32/strlen.S (limited to 'libc/string/sparc/sparc32/strlen.S') diff --git a/libc/string/sparc/sparc32/strlen.S b/libc/string/sparc/sparc32/strlen.S new file mode 100644 index 000000000..cc455ac89 --- /dev/null +++ b/libc/string/sparc/sparc32/strlen.S @@ -0,0 +1,101 @@ +/* Determine the length of a string. + For SPARC v7. + Copyright (C) 1996, 1999, 2003 Free Software Foundation, Inc. + This file is part of the GNU C Library. + Contributed by Jakub Jelinek . + + 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, write to the Free + Software Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA + 02111-1307 USA. */ + + /* Normally, this uses ((xword - 0x01010101) & 0x80808080) test + to find out if any byte in xword could be zero. This is fast, but + also gives false alarm for any byte in range 0x81-0xff. It does + not matter for correctness, as if this test tells us there could + be some zero byte, we check it byte by byte, but if bytes with + high bits set are common in the strings, then this will give poor + performance. You can #define EIGHTBIT_NOT_RARE and the algorithm + will use one tick slower, but more precise test + ((xword - 0x01010101) & (~xword) & 0x80808080), + which does not give any false alarms (but if some bits are set, + one cannot assume from it which bytes are zero and which are not). + It is yet to be measured, what is the correct default for glibc + in these days for an average user. + */ + + .text + .align 4 +10: ldub [%o0], %o5 + cmp %o5, 0 + be 1f + add %o0, 1, %o0 + andcc %o0, 3, %g0 + be 4f + or %o4, %lo(0x80808080), %o3 + ldub [%o0], %o5 + cmp %o5, 0 + be 2f + add %o0, 1, %o0 + andcc %o0, 3, %g0 + be 5f + sethi %hi(0x01010101), %o4 + ldub [%o0], %o5 + cmp %o5, 0 + be 3f + add %o0, 1, %o0 + b 11f + or %o4, %lo(0x01010101), %o2 +1: retl + mov 0, %o0 +2: retl + mov 1, %o0 +3: retl + mov 2, %o0 + +ENTRY(strlen) + mov %o0, %o1 + andcc %o0, 3, %g0 + bne 10b + sethi %hi(0x80808080), %o4 + or %o4, %lo(0x80808080), %o3 +4: sethi %hi(0x01010101), %o4 +5: or %o4, %lo(0x01010101), %o2 +11: ld [%o0], %o5 +12: sub %o5, %o2, %o4 +#ifdef EIGHTBIT_NOT_RARE + andn %o4, %o5, %o4 +#endif + andcc %o4, %o3, %g0 + be 11b + add %o0, 4, %o0 + + srl %o5, 24, %g5 + andcc %g5, 0xff, %g0 + be 13f + add %o0, -4, %o4 + srl %o5, 16, %g5 + andcc %g5, 0xff, %g0 + be 13f + add %o4, 1, %o4 + srl %o5, 8, %g5 + andcc %g5, 0xff, %g0 + be 13f + add %o4, 1, %o4 + andcc %o5, 0xff, %g0 + bne,a 12b + ld [%o0], %o5 + add %o4, 1, %o4 +13: retl + sub %o4, %o1, %o0 +END(strlen) -- cgit v1.2.3