From: Joakim Tjernlund <joakim.tjernlund@lumentis.se>
To: torvalds@transmeta.com, alan@lxorguk.ukuu.org.uk
Cc: linux-kernel@vger.kernel.org
Subject: [PATCH] crc32 improvements for 2.5
Date: Wed, 05 Feb 2003 17:23:11 +0100 [thread overview]
Message-ID: <IGEFJKJNHJDCBKALBJLLGEPOFJAA.joakim.tjernlund@lumentis.se> (raw)
In-Reply-To: <1044365707.4067.4.camel@passion.cambridge.redhat.com>
Hi
I did the optimizations in the crc32 patch Brian Murphy submitted a while ago.
Now I have cleaned it up a little and made some more optimizations.
gcc is quite bad at loop optimizations (at least for PPC) so I have
rewritten them to make gcc to generate better code. Even recent gcc's(3.2.x) produces
better code.
Also reduced the unrolling since it did not make a noticeable difference.
Joakim Tjernlund
--- lib/crc32.c.org Thu Jan 9 00:19:02 2003
+++ lib/crc32.c Tue Feb 4 19:05:01 2003
@@ -87,55 +87,51 @@
{
# if CRC_LE_BITS == 8
const u32 *b =(u32 *)p;
- const u32 *e;
- /* load data 32 bits wide, xor data 32 bits wide. */
+ const u32 *tab = crc32table_le;
- crc = __cpu_to_le32(crc);
- /* Align it */
- for ( ; ((long)b)&3 && len ; len--){
-# ifdef __LITTLE_ENDIAN
- crc = (crc>>8) ^ crc32table_le[ (crc ^ *((u8 *)b)++) & 0xff ];
-# else
- crc = (crc<<8) ^ crc32table_le[ crc>>24 ^ *((u8 *)b)++ ];
-# endif
- }
- e = (u32 *) ( (u8 *)b + (len & ~7));
- while (b < e) {
- crc ^= *b++;
-# ifdef __LITTLE_ENDIAN
- crc = (crc>>8) ^ crc32table_le[ crc & 0xff ];
- crc = (crc>>8) ^ crc32table_le[ crc & 0xff ];
- crc = (crc>>8) ^ crc32table_le[ crc & 0xff ];
- crc = (crc>>8) ^ crc32table_le[ crc & 0xff ];
-# else
- crc = (crc<<8) ^ crc32table_le[ crc >> 24 ];
- crc = (crc<<8) ^ crc32table_le[ crc >> 24 ];
- crc = (crc<<8) ^ crc32table_le[ crc >> 24 ];
- crc = (crc<<8) ^ crc32table_le[ crc >> 24 ];
-# endif
- crc ^= *b++;
# ifdef __LITTLE_ENDIAN
- crc = (crc>>8) ^ crc32table_le[ crc & 0xff ];
- crc = (crc>>8) ^ crc32table_le[ crc & 0xff ];
- crc = (crc>>8) ^ crc32table_le[ crc & 0xff ];
- crc = (crc>>8) ^ crc32table_le[ crc & 0xff ];
+# define DO_CRC crc = (crc>>8) ^ tab[ crc & 255 ]
+# define ENDIAN_SHIFT 0
# else
- crc = (crc<<8) ^ crc32table_le[ crc >> 24 ];
- crc = (crc<<8) ^ crc32table_le[ crc >> 24 ];
- crc = (crc<<8) ^ crc32table_le[ crc >> 24 ];
- crc = (crc<<8) ^ crc32table_le[ crc >> 24 ];
+# define DO_CRC crc = (crc<<8) ^ tab[ crc >> 24 ]
+# define ENDIAN_SHIFT 24
# endif
+
+ crc = __cpu_to_le32(crc);
+ /* Align it */
+ if(unlikely(((long)b)&3 && len)){
+ do {
+ crc ^= *((u8 *)b)++ << ENDIAN_SHIFT;
+ DO_CRC;
+ } while ((--len) && ((long)b)&3 );
+ }
+ if(likely(len >= 4)){
+ /* load data 32 bits wide, xor data 32 bits wide. */
+ size_t save_len = len & 3;
+ len = len >> 2;
+ --b; /* use pre increment below(*++b) for speed */
+ do {
+ crc ^= *++b;
+ DO_CRC;
+ DO_CRC;
+ DO_CRC;
+ DO_CRC;
+ } while (--len);
+ b++; /* point to next byte(s) */
+ len = save_len;
}
/* And the last few bytes */
- e = (u32 *)((u8 *)b + (len & 7));
- while (b < e){
-# ifdef __LITTLE_ENDIAN
- crc = (crc>>8) ^ crc32table_le[ (crc ^ *((u8 *)b)++) & 0xff ];
-# else
- crc = (crc<<8) ^ crc32table_le[ crc>>24 ^ *((u8 *)b)++ ];
-# endif
+ if(len){
+ do {
+ crc ^= *((u8 *)b)++ << ENDIAN_SHIFT;
+ DO_CRC;
+ } while (--len);
}
- return __le32_to_cpu(crc) ;
+
+ return __le32_to_cpu(crc);
+#undef ENDIAN_SHIFT
+#undef DO_CRC
+
# elif CRC_LE_BITS == 4
while (len--) {
crc ^= *p++;
@@ -196,55 +192,50 @@
{
# if CRC_BE_BITS == 8
const u32 *b =(u32 *)p;
- const u32 *e;
- /* load data 32 bits wide, xor data 32 bits wide. */
+ const u32 *tab = crc32table_be;
- crc = __cpu_to_be32(crc);
- /* Align it */
- for ( ; ((long)b)&3 && len ; len--){
-# ifdef __LITTLE_ENDIAN
- crc = (crc>>8) ^ crc32table_be[ (crc ^ *((u8 *)b)++) & 0xff ];
-# else
- crc = (crc<<8) ^ crc32table_be[ crc>>24 ^ *((u8 *)b)++ ];
-# endif
- }
- e = (u32 *) ( (u8 *)b + (len & ~7));
- while (b < e) {
- crc ^= *b++;
-# ifdef __LITTLE_ENDIAN
- crc = (crc>>8) ^ crc32table_be[ crc & 0xff ];
- crc = (crc>>8) ^ crc32table_be[ crc & 0xff ];
- crc = (crc>>8) ^ crc32table_be[ crc & 0xff ];
- crc = (crc>>8) ^ crc32table_be[ crc & 0xff ];
-# else
- crc = (crc<<8) ^ crc32table_be[ crc >> 24 ];
- crc = (crc<<8) ^ crc32table_be[ crc >> 24 ];
- crc = (crc<<8) ^ crc32table_be[ crc >> 24 ];
- crc = (crc<<8) ^ crc32table_be[ crc >> 24 ];
-# endif
- crc ^= *b++;
# ifdef __LITTLE_ENDIAN
- crc = (crc>>8) ^ crc32table_be[ crc & 0xff ];
- crc = (crc>>8) ^ crc32table_be[ crc & 0xff ];
- crc = (crc>>8) ^ crc32table_be[ crc & 0xff ];
- crc = (crc>>8) ^ crc32table_be[ crc & 0xff ];
+# define DO_CRC crc = (crc>>8) ^ tab[ crc & 255 ]
+# define ENDIAN_SHIFT 24
# else
- crc = (crc<<8) ^ crc32table_be[ crc >> 24 ];
- crc = (crc<<8) ^ crc32table_be[ crc >> 24 ];
- crc = (crc<<8) ^ crc32table_be[ crc >> 24 ];
- crc = (crc<<8) ^ crc32table_be[ crc >> 24 ];
+# define DO_CRC crc = (crc<<8) ^ tab[ crc >> 24 ]
+# define ENDIAN_SHIFT 0
# endif
+
+ crc = __cpu_to_be32(crc);
+ /* Align it */
+ if(unlikely(((long)b)&3 && len)){
+ do {
+ crc ^= *((u8 *)b)++ << ENDIAN_SHIFT;
+ DO_CRC;
+ } while ((--len) && ((long)b)&3 );
+ }
+ if(likely(len >= 4)){
+ /* load data 32 bits wide, xor data 32 bits wide. */
+ size_t save_len = len & 3;
+ len = len >> 2;
+ --b; /* use pre increment below(*++b) for speed */
+ do {
+ crc ^= *++b;
+ DO_CRC;
+ DO_CRC;
+ DO_CRC;
+ DO_CRC;
+ } while (--len);
+ b++; /* point to next byte(s) */
+ len = save_len;
}
/* And the last few bytes */
- e = (u32 *)((u8 *)b + (len & 7));
- while (b < e){
-# ifdef __LITTLE_ENDIAN
- crc = (crc>>8) ^ crc32table_be[ (crc ^ *((u8 *)b)++) & 0xff ];
-# else
- crc = (crc<<8) ^ crc32table_be[ crc>>24 ^ *((u8 *)b)++ ];
-# endif
- }
- return __be32_to_cpu(crc) ;
+ if(len){
+ do {
+ crc ^= *((u8 *)b)++ << ENDIAN_SHIFT;
+ DO_CRC;
+ } while (--len);
+ }
+ return __be32_to_cpu(crc);
+#undef ENDIAN_SHIFT
+#undef DO_CRC
+
# elif CRC_BE_BITS == 4
while (len--) {
crc ^= *p++ << 24;
next parent reply other threads:[~2003-02-05 16:13 UTC|newest]
Thread overview: 5+ messages / expand[flat|nested] mbox.gz Atom feed top
[not found] <1044365707.4067.4.camel@passion.cambridge.redhat.com>
2003-02-05 16:23 ` Joakim Tjernlund [this message]
2003-02-13 11:13 ` [PATCH] crc32 improvements for 2.5 [RESEND] Joakim Tjernlund
2003-02-13 17:54 ` Andrew Morton
2003-02-13 21:45 ` Joakim Tjernlund
2003-02-18 9:28 ` [PATCH] crc32 improvements for 2.5, more optimizations Joakim Tjernlund
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=IGEFJKJNHJDCBKALBJLLGEPOFJAA.joakim.tjernlund@lumentis.se \
--to=joakim.tjernlund@lumentis.se \
--cc=alan@lxorguk.ukuu.org.uk \
--cc=linux-kernel@vger.kernel.org \
--cc=torvalds@transmeta.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).