[src/trunk]: src/lib/libc/arch/m68k/string Use dbeq loops, code submitted by ...

[is <is%NetBSD.org@localhost>]

details:   https://anonhg.NetBSD.org/src/rev/bbfa52677467
branches:  trunk
changeset: 499730:bbfa52677467
user:      is <is%NetBSD.org@localhost>
date:      Mon Nov 27 22:35:30 2000 +0000

description:
Use dbeq loops, code submitted by Hiroshi Horimoto in PR 3223, integrated
and tested by Allen Briggs.
Needs 1/3 less time on 68030, 45% less time on 68040.
XXX On 68060, 1/2 less time for the padding part, but 40% more time on the
copy part. I've stared at the 68060 User Manual, but can't understand why;
must be some wierd (branch) cache interaction.

diffstat:

 lib/libc/arch/m68k/string/strncpy.S |  23 +++++++++++++++--------
 1 files changed, 15 insertions(+), 8 deletions(-)

diffs (50 lines):

diff -r daf8ceba23be -r bbfa52677467 lib/libc/arch/m68k/string/strncpy.S
--- a/lib/libc/arch/m68k/string/strncpy.S       Mon Nov 27 22:29:26 2000 +0000
+++ b/lib/libc/arch/m68k/string/strncpy.S       Mon Nov 27 22:35:30 2000 +0000
@@ -1,4 +1,4 @@
-/*     $NetBSD: strncpy.S,v 1.10 1999/10/25 23:48:16 thorpej Exp $     */
+/*     $NetBSD: strncpy.S,v 1.11 2000/11/27 22:35:30 is Exp $  */
 
 /*-
  * Copyright (c) 1990 The Regents of the University of California.
@@ -43,7 +43,7 @@
 #if 0
        RCSID("from: @(#)strncpy.s      5.1 (Berkeley) 5/12/90")
 #else
-       RCSID("$NetBSD: strncpy.S,v 1.10 1999/10/25 23:48:16 thorpej Exp $")
+       RCSID("$NetBSD: strncpy.S,v 1.11 2000/11/27 22:35:30 is Exp $")
 #endif
 #endif /* LIBC_SCCS and not lint */
 
@@ -53,18 +53,25 @@
        jeq     Lscdone                 | nothing to do
        movl    %sp@(8),%a0             | a0 = fromaddr
        movl    %d0,%a1                 | a1 = toaddr
+       subql   #1,%d1
 Lscloop:
        movb    %a0@+,%a1@+             | copy a byte
-       jeq     Lscploop                | copied null, go pad if necessary
-       subql   #1,%d1                  | adjust count
-       jne     Lscloop                 | more room, keep going
+       dbeq    %d1,Lscloop             | loop through low word of d1
+       jeq     Lscpadding              | copied null, padding if necessary
+       clrw    %d1                     | clear low word of d1
+       subql   #1,%d1                  | adjust count for long copies
+       jcc     Lscloop                 | more room, keep going
 Lscdone:
 #ifdef __SVR4_ABI__
        moveal  %d0,%a0
 #endif
        rts
+
 Lscploop:
-       subql   #1,%d1                  | adjust count
-       jeq     Lscdone                 | no more room, all done
        clrb    %a1@+                   | clear a byte
-       jra     Lscploop                | keep going
+       dbra    %d1,Lscploop            | keep going
+       clrw    %d1
+Lscpadding:
+       subql   #1,%d1
+       jcc     Lscploop
+       jra     Lscdone