[hts-users:01936] Re: Fast & memory efficient HHEd clustering patch
- Subject: [hts-users:01936] Re: Fast & memory efficient HHEd clustering patch
- From: 康世胤 <kangshiyin@xxxxxxxxx>
- Date: Tue, 7 Apr 2009 15:22:45 +0800
- Delivered-to: hts-users@xxxxxxxxxxxxxxx
- Dkim-signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=gmail.com; s=gamma; h=domainkey-signature:mime-version:received:in-reply-to:references :date:message-id:subject:from:to:content-type; bh=DONOmvrMQvRahTEI9CldApvM9Pqgzt/XvrMwtTvEIs8=; b=gPYEIcsBpiqrHu491gUE8+XAuDA1LBjwb0OsRwT0zeLU7sy/gf1mSpecIOn3Nps1dK AdxcE3sKC2AhXtoGd1OSwAu30nR4it3mNquAGsCE+96kSsgzCxmluB6nc+n6PuBgAhCP E8xGL9PZjltzfs8aJzarVwK+TSz0miNlVDc1A=
- Domainkey-signature: a=rsa-sha1; c=nofws; d=gmail.com; s=gamma; h=mime-version:in-reply-to:references:date:message-id:subject:from:to :content-type; b=McqtCeF1fxnr4LSLVO68j0Vqm0g8I5QC8mLalMv0rtuqT57q6WRXSQL9EGaybm6W6y pwxS6F2RGQBRtZyRTr/JbP1AKeNPNPLnI/8zbLLlvuXaAbPpmuUwKMUYcr2a3u/MQBZt gnJuNBfZKl/7OatNahYStseEGg+/NhMtufSx4=
thank you all.
i forgot the option '-O2'. i didn't realized it was overwritten when CFLAGS is specified by hand.
the clustering process is about 5% faster than before.
2009/4/6 Junichi Yamagishi <jyamagis@xxxxxxxxxxxx>
In my environment, this works perfectly fine.
Although I have checked actual speed, it would be much faster.
I used icc and OMP_NUM_THREADS is set to 4.
FYI, please create Makefiles like this and recompile all tools
ICC: ./configure CC=icc F77=ifort CFLAGS="-O2 -static-intel -openmp" FFLAGS="-O2 -static-intel -openmp"
GCC: ./configure CFLAGS="-O2 -fopenmp" FFLAGS="-O2 -fopenmp"
Please note that you should include '-O2' as welll as '-fopenmp',
otherwise your binary does not have any optimization process.
On 6 Apr 2009, at 00:48, Heiga ZEN (Byung Ha CHUN) wrote:
kangshiyin@xxxxxxxxx wrote (2009/04/03 10:08):
i have tried these 2 patches. the fastHHEd.patch works well,
but the HHEd.c.patch seems not so good.
it is even slower after this patch is used on my computer.
and i find that all other programs such as HInit & HRest become slower if the -fopenmp flag is applied when compiling.
my compiler is gcc4/Fedora9
is this gcc or openmp's problem? should i use icc or something else?
This sounds strange. HInit & HRest should not be affected by this modification because they have no OpenMP directives. So I believe that this is the problem of gcc4 or running environment rather than my HHEd.c.patch.
Note that you need to set the Linux environmental variable OMP_NUM_THREADS (the number of threads to be generated) to a proper value. For example, if your PC has 4 cores, OMP_NUM_THREADS should be less than 4. Otherwise, it sometimes becomes slower because of overhead while generating threads.
Heiga ZEN (Byung Ha CHUN)
Heiga ZEN (Byung Ha CHUN)
Speech Technology Group
Cambridge Research Lab
Toshiba Research Europe
phone: +44 1223 436975
This email has been scanned by the MessageLabs Email Security System.
For more information please visit http://www.messagelabs.com/email ______________________________________________________________________
The University of Edinburgh is a charitable body, registered in
Scotland, with registration number SC005336.
Lab of Human-Computer Speech Interaction,
Department of Computer Science & Technology,
Tsinghua University,Beijing, P.R.China, 100084
- [hts-users:01875] Re: Fast & memory efficient HHEd clustering patch, Heiga Zen (Byung Ha CHUN)
- [hts-users:01933] Re: Fast & memory efficient HHEd clustering patch, 康世胤
- [hts-users:01934] Re: Fast & memory efficient HHEd clustering patch, Heiga ZEN (Byung Ha CHUN)
- [hts-users:01935] Re: Fast & memory efficient HHEd clustering patch, Junichi Yamagishi