diff options
Diffstat (limited to 'libc/string/bfin/strcmp.S')
| -rw-r--r-- | libc/string/bfin/strcmp.S | 38 |
1 files changed, 18 insertions, 20 deletions
diff --git a/libc/string/bfin/strcmp.S b/libc/string/bfin/strcmp.S index 07629c03b..c0a706d2b 100644 --- a/libc/string/bfin/strcmp.S +++ b/libc/string/bfin/strcmp.S @@ -1,5 +1,5 @@ /* strcmp.S - * Copyright (C) 2003, 2005, 2006 Analog Devices Inc., All Rights Reserved. + * Copyright (C) 2003-2007 Analog Devices Inc., All Rights Reserved. * * This file is subject to the terms and conditions of the GNU Library General * Public License. See the file "COPYING.LIB" in the main directory of this @@ -9,6 +9,8 @@ * http://www.analog.com/processors/resources/crosscore/visualDspDevSoftware.html */ +#include <sysdep.h> + /* Fast strcmp() for Blackfin. * When both strings are aligned, this processes four characters at * a time. Uses a hw loop with "very big" count to loop "forever", @@ -17,15 +19,11 @@ * loop to check more carefully (same as the unaligned case). */ -#include <features.h> - .text .align 2 -.global _strcmp -.type _strcmp, STT_FUNC -_strcmp: +ENTRY(_strcmp) [--sp] = (R7:4); p1 = r0; p2 = r1; @@ -36,13 +34,13 @@ _strcmp: r0 = r0 | r1; // check both pointers at same time r0 <<= 30; // dump all but last 2 bits cc = az; // are they zero? - if !cc jump unaligned; // no; use unaligned code. + if !cc jump .Lunaligned; // no; use unaligned code. // fall-thru for aligned case.. // note that r0 is zero from the previous... // p0 set to -1 - lsetup (beginloop, endloop) lc0=p0; + LSETUP (.Lbeginloop, .Lendloop) lc0=p0; // pick up first words r1 = [p1++]; r2 = [p2++]; @@ -51,8 +49,8 @@ _strcmp: r7.h = 0xFF; // loop : 9 cycles to check 4 characters cc = r1 == r2; -beginloop: - if !cc jump notequal4; // compare failure, exit loop +.Lbeginloop: + if !cc jump .Lnotequal4; // compare failure, exit loop // starting with 44332211 // see if char 3 or char 1 is 0 @@ -65,18 +63,18 @@ beginloop: // add to zero, and (r1 is free, reload) r6 = r3 +|+ r0 || r1 = [p1++] || nop; cc |= az; // true if either is zero - if cc jump zero4; // leave if a zero somewhere -endloop: + if cc jump .Lzero4; // leave if a zero somewhere +.Lendloop: cc = r1 == r2; // loop exits -notequal4: // compare failure on 4-char compare +.Lnotequal4: // compare failure on 4-char compare // address pointers are one word ahead; // faster to use zero4 exit code p1 += 4; p2 += 4; -zero4: // one of the bytes in word 1 is zero +.Lzero4: // one of the bytes in word 1 is zero // but we've already fetched the next word; so // backup two to look at failing word again p1 += -8; @@ -87,27 +85,27 @@ zero4: // one of the bytes in word 1 is zero // here when pointers are unaligned: checks one // character at a time. Also use at the end of // the word-check algorithm to figure out what happened -unaligned: +.Lunaligned: // R0 is non-zero from before. // p0 set to -1 r0 = 0 (Z); r1 = B[p1++] (Z); r2 = B[p2++] (Z); - lsetup (beginloop1, endloop1) lc0=p0; + LSETUP (.Lbeginloop1, .Lendloop1) lc0=p0; -beginloop1: +.Lbeginloop1: cc = r1; // first char must be non-zero // chars must be the same r3 = r2 - r1 (NS) || r1 = B[p1++] (Z) || nop; cc &= az; r3 = r0 - r2; // second char must be non-zero cc &= an; - if !cc jump exitloop1; -endloop1: + if !cc jump .Lexitloop1; +.Lendloop1: r2 = B[p2++] (Z); -exitloop1: // here means we found a zero or a difference. +.Lexitloop1: // here means we found a zero or a difference. // we have r2(N), p2(N), r1(N+1), p1(N+2) r1=B[p1+ -2] (Z); r0 = r1 - r2; |