* [PATCH] zlib: Optimize inffast even more
@ 2009-11-09 10:22 Joakim Tjernlund
2009-11-11 23:24 ` Andrew Morton
2009-11-11 23:46 ` roel kluin
0 siblings, 2 replies; 5+ messages in thread
From: Joakim Tjernlund @ 2009-11-09 10:22 UTC (permalink / raw)
To: akpm, Richard Purdie, linux-kernel; +Cc: Joakim Tjernlund
This improves zlib: Optimize inffast when copying direct from output
and gives another 3-4% improvement for my MPC8321 target.
Does not need CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS,
uses get_unaligned() but only in one place.
The copy loop just above this one can also use this
optimization, but I havn't done so as I have not tested if it
is a win there too.
Signed-off-by: Joakim Tjernlund <Joakim.Tjernlund@transmode.se>
---
Someone with a little endian target should test too.
arch/powerpc/boot/Makefile | 1 +
lib/zlib_inflate/inffast.c | 54 ++++++++++++++++++++++----------------------
2 files changed, 28 insertions(+), 27 deletions(-)
diff --git a/arch/powerpc/boot/Makefile b/arch/powerpc/boot/Makefile
index 9ae7b7e..33335be 100644
--- a/arch/powerpc/boot/Makefile
+++ b/arch/powerpc/boot/Makefile
@@ -34,6 +34,7 @@ BOOTCFLAGS += -fno-stack-protector
endif
BOOTCFLAGS += -I$(obj) -I$(srctree)/$(obj)
+BOOTCFLAGS += -Iarch/powerpc/include
DTS_FLAGS ?= -p 1024
diff --git a/lib/zlib_inflate/inffast.c b/lib/zlib_inflate/inffast.c
index 0588fbf..4f78567 100644
--- a/lib/zlib_inflate/inffast.c
+++ b/lib/zlib_inflate/inffast.c
@@ -4,6 +4,7 @@
*/
#include <linux/zutil.h>
+#include <asm/unaligned.h>
#include "inftrees.h"
#include "inflate.h"
#include "inffast.h"
@@ -24,9 +25,11 @@
#ifdef POSTINC
# define OFF 0
# define PUP(a) *(a)++
+# define UP_UNALIGNED(a) get_unaligned((a)++)
#else
# define OFF 1
# define PUP(a) *++(a)
+# define UP_UNALIGNED(a) get_unaligned(++(a))
#endif
/*
@@ -240,52 +243,49 @@ void inflate_fast(z_streamp strm, unsigned start)
}
else {
from = out - dist; /* copy direct from output */
-#ifdef CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS
/* minimum length is three */
if (dist > 2 ) {
- unsigned short *sout = (unsigned short *)(out - OFF);
- unsigned short *sfrom = (unsigned short *)(from - OFF);
- unsigned long loops = len >> 1;
+ unsigned short *sout;
+ unsigned short *sfrom;
+ unsigned long loops;
+ /* Align out addr, only sfrom might be unaligned */
+ if (!((long)(out - 1 + OFF)) & 1) {
+ PUP(out) = PUP(from);
+ len--;
+ }
+ sout = (unsigned short *)(out - OFF);
+ sfrom = (unsigned short *)(from - OFF);
+ loops = len >> 1;
do
- PUP(sout) = PUP(sfrom);
+ PUP(sout) = UP_UNALIGNED(sfrom);
while (--loops);
out = (unsigned char *)sout + OFF;
from = (unsigned char *)sfrom + OFF;
if (len & 1)
PUP(out) = PUP(from);
- } else if (dist == 2) {
- unsigned short *sout = (unsigned short *)(out - OFF);
+ } else { /* dist == 1 or dist == 2 */
+ unsigned short *sout;
unsigned short pat16;
- unsigned long loops = len >> 1;
+ unsigned long loops;
+ /* Align out addr */
+ if (!((long)(out - 1 + OFF)) & 1) {
+ PUP(out) = PUP(from);
+ len--;
+ }
+ sout = (unsigned short *)(out - OFF);
pat16 = *(sout-2+2*OFF);
+ if (dist == 1)
+ pat16 = (pat16 & 0xff) | ((pat16 & 0xff ) << 8);
+ loops = len >> 1;
do
PUP(sout) = pat16;
while (--loops);
out = (unsigned char *)sout + OFF;
if (len & 1)
PUP(out) = PUP(from);
- } else {
- unsigned char pat8 = *(out - 1 + OFF);
-
- do {
- PUP(out) = pat8;
- } while (--len);
}
-#else
- do { /* minimum length is three */
- PUP(out) = PUP(from);
- PUP(out) = PUP(from);
- PUP(out) = PUP(from);
- len -= 3;
- } while (len > 2);
- if (len) {
- PUP(out) = PUP(from);
- if (len > 1)
- PUP(out) = PUP(from);
- }
-#endif
}
}
else if ((op & 64) == 0) { /* 2nd level distance code */
--
1.6.4.4
^ permalink raw reply related [flat|nested] 5+ messages in thread* Re: [PATCH] zlib: Optimize inffast even more
2009-11-09 10:22 [PATCH] zlib: Optimize inffast even more Joakim Tjernlund
@ 2009-11-11 23:24 ` Andrew Morton
2009-11-12 8:51 ` Joakim Tjernlund
2009-11-11 23:46 ` roel kluin
1 sibling, 1 reply; 5+ messages in thread
From: Andrew Morton @ 2009-11-11 23:24 UTC (permalink / raw)
To: Joakim Tjernlund; +Cc: Richard Purdie, linux-kernel
On Mon, 9 Nov 2009 11:22:16 +0100
Joakim Tjernlund <Joakim.Tjernlund@transmode.se> wrote:
> This improves zlib: Optimize inffast when copying direct from output
> and gives another 3-4% improvement for my MPC8321 target.
> Does not need CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS,
> uses get_unaligned() but only in one place.
> The copy loop just above this one can also use this
> optimization, but I havn't done so as I have not tested if it
> is a win there too.
>
> Signed-off-by: Joakim Tjernlund <Joakim.Tjernlund@transmode.se>
> ---
>
> Someone with a little endian target should test too.
eh? That requires a PC. Can't you test it??
^ permalink raw reply [flat|nested] 5+ messages in thread
* Re: [PATCH] zlib: Optimize inffast even more
2009-11-11 23:24 ` Andrew Morton
@ 2009-11-12 8:51 ` Joakim Tjernlund
0 siblings, 0 replies; 5+ messages in thread
From: Joakim Tjernlund @ 2009-11-12 8:51 UTC (permalink / raw)
To: Andrew Morton; +Cc: linux-kernel, Richard Purdie
Andrew Morton <akpm@linux-foundation.org> wrote on 12/11/2009 00:24:45:
>
> On Mon, 9 Nov 2009 11:22:16 +0100
> Joakim Tjernlund <Joakim.Tjernlund@transmode.se> wrote:
>
> > This improves zlib: Optimize inffast when copying direct from output
> > and gives another 3-4% improvement for my MPC8321 target.
> > Does not need CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS,
> > uses get_unaligned() but only in one place.
> > The copy loop just above this one can also use this
> > optimization, but I havn't done so as I have not tested if it
> > is a win there too.
> >
> > Signed-off-by: Joakim Tjernlund <Joakim.Tjernlund@transmode.se>
> > ---
> >
> > Someone with a little endian target should test too.
>
> eh? That requires a PC. Can't you test it??
It needs something that actually uses the inflate too on that PC and
I don't have that handy.
Anyhow, I downloaded zip and made the changes and tested a little, found
a problem. Will resend a new patch with everything in it soon.
^ permalink raw reply [flat|nested] 5+ messages in thread
* Re: [PATCH] zlib: Optimize inffast even more
2009-11-09 10:22 [PATCH] zlib: Optimize inffast even more Joakim Tjernlund
2009-11-11 23:24 ` Andrew Morton
@ 2009-11-11 23:46 ` roel kluin
2009-11-12 7:23 ` Joakim Tjernlund
1 sibling, 1 reply; 5+ messages in thread
From: roel kluin @ 2009-11-11 23:46 UTC (permalink / raw)
To: Joakim Tjernlund; +Cc: akpm, Richard Purdie, linux-kernel
On Mon, Nov 9, 2009 at 11:22 AM, Joakim Tjernlund<Joakim.Tjernlund@transmode.se> wrote:> This improves zlib: Optimize inffast when copying direct from output> and gives another 3-4% improvement for my MPC8321 target.> Does not need CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS,> uses get_unaligned() but only in one place.> The copy loop just above this one can also use this> optimization, but I havn't done so as I have not tested if it> is a win there too.>> Signed-off-by: Joakim Tjernlund <Joakim.Tjernlund@transmode.se>> ---
> @@ -240,52 +243,49 @@ void inflate_fast(z_streamp strm, unsigned start)> }> else {> from = out - dist; /* copy direct from output */> -#ifdef CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS> /* minimum length is three */> if (dist > 2 ) {> - unsigned short *sout = (unsigned short *)(out - OFF);> - unsigned short *sfrom = (unsigned short *)(from - OFF);> - unsigned long loops = len >> 1;> + unsigned short *sout;> + unsigned short *sfrom;> + unsigned long loops;>> + /* Align out addr, only sfrom might be unaligned */> + if (!((long)(out - 1 + OFF)) & 1) {
I think this is wrong
did you mean
if (!((long)(out - 1 + OFF) & 1))
> + PUP(out) = PUP(from);> + len--;> + }> + sout = (unsigned short *)(out - OFF);> + sfrom = (unsigned short *)(from - OFF);> + loops = len >> 1;> do> - PUP(sout) = PUP(sfrom);> + PUP(sout) = UP_UNALIGNED(sfrom);> while (--loops);> out = (unsigned char *)sout + OFF;> from = (unsigned char *)sfrom + OFF;> if (len & 1)> PUP(out) = PUP(from);> - } else if (dist == 2) {> - unsigned short *sout = (unsigned short *)(out - OFF);> + } else { /* dist == 1 or dist == 2 */> + unsigned short *sout;> unsigned short pat16;> - unsigned long loops = len >> 1;> + unsigned long loops;>> + /* Align out addr */> + if (!((long)(out - 1 + OFF)) & 1) {
and
if (!((long)(out - 1 + OFF) & 1))
> + PUP(out) = PUP(from);> + len--;> + }> + sout = (unsigned short *)(out - OFF);> pat16 = *(sout-2+2*OFF);> + if (dist == 1)> + pat16 = (pat16 & 0xff) | ((pat16 & 0xff ) << 8);> + loops = len >> 1;> do> PUP(sout) = pat16;> while (--loops);> out = (unsigned char *)sout + OFF;> if (len & 1)> PUP(out) = PUP(from);> - } else {> - unsigned char pat8 = *(out - 1 + OFF);> -> - do {> - PUP(out) = pat8;> - } while (--len);> }> -#else> - do { /* minimum length is three */> - PUP(out) = PUP(from);> - PUP(out) = PUP(from);> - PUP(out) = PUP(from);> - len -= 3;> - } while (len > 2);> - if (len) {> - PUP(out) = PUP(from);> - if (len > 1)> - PUP(out) = PUP(from);> - }> -#endif> }> }> else if ((op & 64) == 0) { /* 2nd level distance code */> --> 1.6.4.4>> --> To unsubscribe from this list: send the line "unsubscribe linux-kernel" in> the body of a message to majordomo@vger.kernel.org> More majordomo info at http://vger.kernel.org/majordomo-info.html> Please read the FAQ at http://www.tux.org/lkml/>ÿôèº{.nÇ+‰·Ÿ®‰†+%ŠËÿ±éݶ\x17¥Šwÿº{.nÇ+‰·¥Š{±þG«éÿŠ{ayº\x1dʇڙë,j\a¢f£¢·hšïêÿ‘êçz_è®\x03(階ŠÝ¢j"ú\x1a¶^[m§ÿÿ¾\a«þG«éÿ¢¸?™¨èÚ&£ø§~á¶iO•æ¬z·švØ^\x14\x04\x1a¶^[m§ÿÿÃ\fÿ¶ìÿ¢¸?–I¥
^ permalink raw reply [flat|nested] 5+ messages in thread* Re: [PATCH] zlib: Optimize inffast even more
2009-11-11 23:46 ` roel kluin
@ 2009-11-12 7:23 ` Joakim Tjernlund
0 siblings, 0 replies; 5+ messages in thread
From: Joakim Tjernlund @ 2009-11-12 7:23 UTC (permalink / raw)
To: roel kluin; +Cc: akpm, linux-kernel, Richard Purdie
roel kluin <roel.kluin@gmail.com> wrote on 12/11/2009 00:46:41:
>
> On Mon, Nov 9, 2009 at 11:22 AM, Joakim Tjernlund
> <Joakim.Tjernlund@transmode.se> wrote:
> > This improves zlib: Optimize inffast when copying direct from output
> > and gives another 3-4% improvement for my MPC8321 target.
> > Does not need CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS,
> > uses get_unaligned() but only in one place.
> > The copy loop just above this one can also use this
> > optimization, but I havn't done so as I have not tested if it
> > is a win there too.
> >
> > Signed-off-by: Joakim Tjernlund <Joakim.Tjernlund@transmode.se>
> > ---
>
>
> > @@ -240,52 +243,49 @@ void inflate_fast(z_streamp strm, unsigned start)
> > }
> > else {
> > from = out - dist; /* copy direct from output */
> > -#ifdef CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS
> > /* minimum length is three */
> > if (dist > 2 ) {
> > - unsigned short *sout = (unsigned short *)(out - OFF);
> > - unsigned short *sfrom = (unsigned short *)(from - OFF);
> > - unsigned long loops = len >> 1;
> > + unsigned short *sout;
> > + unsigned short *sfrom;
> > + unsigned long loops;
> >
> > + /* Align out addr, only sfrom might be unaligned */
> > + if (!((long)(out - 1 + OFF)) & 1) {
>
> I think this is wrong
>
> did you mean
>
> if (!((long)(out - 1 + OFF) & 1))
Yes, will fix and send out a new patch with
cleanups and fixes for CPUs that cannot do unaligned
accesses. Thanks
Jocke
^ permalink raw reply [flat|nested] 5+ messages in thread
end of thread, other threads:[~2009-11-12 8:52 UTC | newest]
Thread overview: 5+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2009-11-09 10:22 [PATCH] zlib: Optimize inffast even more Joakim Tjernlund
2009-11-11 23:24 ` Andrew Morton
2009-11-12 8:51 ` Joakim Tjernlund
2009-11-11 23:46 ` roel kluin
2009-11-12 7:23 ` Joakim Tjernlund
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.