Hi Jean-Marc,
I tried to add VERY_SMALL at the input of the encoder, but that did not change
much.
Here's a list of source code locations where denormals appear for the first
time as calculation results.
This list is based on a 4 minutes recording of ambient sound that is passed to
speexenc 1.2rc1 with the command line
--narrowband --denoise --agc --abr 15000
=============================================Count File Line
Variable
=============================================6 x filters.c :367 yi
131 x ltp.c :71 sum
24 x filters.c :367 mem[0]
1057 x lpc.c :193 x[i-lag]
117 x filters.c :326 mem[0]
1295 x filters.c :326 yi
1257 x filters.c :104 mem[0]
7 x filterbank.c :165 tmp
942 x filterbank.c :166 tmp
2387 x lpc.c :193 x[i]
5 x filters.c :105 mem[1]
306618 x filters.c :370 mem[j+1]
21796 x preprocess.c :851 theta
1638 x vbr.c :139 ener2
1635 x vbr.c :136 ener1
8037 x filters.c :306 sum
17852 x lpc.c :193 d
103519 x ltp.c :67 part
77910 x ltp.c :68 part
69331 x ltp.c :69 part
13687042 x ltp.c :66 part
4209 x ltp.c :263 tmp
55351 x preprocess.c :804 st->old_ps[i]
508 x preprocess.c :807 st->prior[i]
217495 x preprocess.c :892 theta
117888 x filters.c :330 mem[j+1]
=============================================
I hope this is usefull.
Thanks
Mark
-----Original Message-----
From: speex-dev-bounces at xiph.org [mailto:speex-dev-bounces at xiph.org] On
Behalf Of Jean-Marc Valin
Sent: Thursday, September 24, 2009 2:34 AM
To: Mark Schilling
Cc: speex-dev at xiph.org
Subject: Re: [Speex-dev] High CPU usage
If you tell me where most of the denormals are happening, I may be able
to fix that. Also, can you try just adding VERY_SMALL to the encoder's
input and see what happens?
Jean-Marc
--
Neu: GMX Doppel-FLAT mit Internet-Flatrate + Telefon-Flatrate
f?r nur 19,99 Euro/mtl.!* http://portal.gmx.net/de/go/dsl02
