Btw, Good work. 5ms is a huge different, espically in optimizing terms. I've added a few flags and shaved off another ms

here's my flags: (only for p4/xeon)

-fPIC -O3 -march=pentium4 -funroll-loops -fomit-frame-pointer -msse -msse2 -mfpmath=sse

keep up the good work,

Adam

brian k. west wrote:

Yes I realized my error in my wording but it was early :P  It doesn't
improve alot but does give you some ways to get a better idea of translation
times if your box is loaded up with calls.

bkw
PS this patch was added to CVS-HEAD

----- Original Message ----- From: "Adam Hart" <[EMAIL PROTECTED]>
To: <[EMAIL PROTECTED]>
Sent: Monday, May 17, 2004 5:49 PM
Subject: Re: [Asterisk-Users] speex




Actually it encodes a second of data, which with a 20ms codec would be
50 frames. The timing shows better than expected results due to caching.

-Adam

brian wrote:


http://asterisk.bkw.org/diff/translate.patch.txt

If you try that patch out it adds a nice feature...

show translation recalc [xx]

You can throw more than 1 sample thru it and recalculate your

translation

matrix.  It also allows you to see TRUE translation under a load or just
when ever you feel like seeing them updated.  When * loads the codec it
shoot one frame thru and times it.  Now under real world scenarios you

will

be shooting more than one frame thru so LETS have the option to update

the

matrix with these types of tests.  200 is the max.

bkw




-----Original Message-----
From: [EMAIL PROTECTED] [mailto:asterisk-users-
[EMAIL PROTECTED] On Behalf Of James H. Cloos Jr.
Sent: Monday, May 17, 2004 11:24 AM
To: [EMAIL PROTECTED]
Subject: [Asterisk-Users] speex

Just a suggestion to anyone using speex:

Try running the 1.1.5 or svn code rather than 1.0.3.

As a quick example, here are the show translation outputs from * on a
2.8 GHz P4 with speex 1.0.3 (from debian sid's .deb) and from 1.1.5
(compiled with CFLAGS=-march=pentium4 and --enable-sse).

Note how encoding from slin went from 25 to 15 ms.  That is from the
re-write of the sse optimized routines in libspeex.  The % change is
similar to what I saw on my p3 notebook, where both 1.0.3 and 1.1.5
were compiled with --enable-sse and -march=pentium3.

(As a side note, these were captured before Brian's ilbc Makefile
patch made it to the anon cvs tree; that optimization shaved 5ms
off the time to encode to iLBC on that box.)

spx103*CLI> show translation
       Translation times between formats (in milliseconds)
        Source Format (Rows) Destination Format(Columns)

G723 GSM ULAW ALAW G726 ADPCM SLINR LPC10 G729A SPEEX

ILBC

23     -     -     -     -     -     -     -     -     -     -     -

GSM - - 2 2 2 2 1 6 - 26

19

ULAW - 3 - 1 2 2 1 6 - 26

19

ALAW - 3 1 - 2 2 1 6 - 26

19

G726 - 3 2 2 - 2 1 6 - 26

19

ADPCM - 3 2 2 2 - 1 6 - 26

19

SLINR - 2 1 1 1 1 - 5 - 25

18

LPC10 - 4 3 3 3 3 2 - - 27

20

9A     -     -     -     -     -     -     -     -     -     -     -

SPEEX - 3 2 2 2 2 1 6 - -

19

ILBC - 5 4 4 4 4 3 8 -

8 -

spx115*CLI> show translation
       Translation times between formats (in milliseconds)
        Source Format (Rows) Destination Format(Columns)

G723 GSM ULAW ALAW G726 ADPCM SLINR LPC10 G729A SPEEX

ILBC

23     -     -     -     -     -     -     -     -     -     -     -

GSM - - 2 2 2 2 1 6 - 16

19

ULAW - 3 - 1 2 2 1 6 - 16

19

ALAW - 3 1 - 2 2 1 6 - 16

19

G726 - 3 2 2 - 2 1 6 - 16

19

ADPCM - 3 2 2 2 - 1 6 - 16

19

SLINR - 2 1 1 1 1 - 5 - 15

18

LPC10 - 4 3 3 3 3 2 - - 17

20

9A     -     -     -     -     -     -     -     -     -     -     -

SPEEX - 3 2 2 2 2 1 6 - -

19

ILBC - 5 4 4 4 4 3 8 -

8 -


-JimC -- James H. Cloos, Jr. <[EMAIL PROTECTED]> <http://jhcloos.com/voip> _______________________________________________ Asterisk-Users mailing list [EMAIL PROTECTED] http://lists.digium.com/mailman/listinfo/asterisk-users To UNSUBSCRIBE or update options visit: http://lists.digium.com/mailman/listinfo/asterisk-users



_______________________________________________ Asterisk-Users mailing list [EMAIL PROTECTED] http://lists.digium.com/mailman/listinfo/asterisk-users To UNSUBSCRIBE or update options visit: http://lists.digium.com/mailman/listinfo/asterisk-users


_______________________________________________ Asterisk-Users mailing list [EMAIL PROTECTED] http://lists.digium.com/mailman/listinfo/asterisk-users To UNSUBSCRIBE or update options visit: http://lists.digium.com/mailman/listinfo/asterisk-users




_______________________________________________ Asterisk-Users mailing list [EMAIL PROTECTED] http://lists.digium.com/mailman/listinfo/asterisk-users To UNSUBSCRIBE or update options visit: http://lists.digium.com/mailman/listinfo/asterisk-users


_______________________________________________ Asterisk-Users mailing list [EMAIL PROTECTED] http://lists.digium.com/mailman/listinfo/asterisk-users To UNSUBSCRIBE or update options visit: http://lists.digium.com/mailman/listinfo/asterisk-users

Reply via email to