| Summary: | FluidSynth 1.0.5 makes a horrible noise after some playing | ||
|---|---|---|---|
| Product: | Gentoo Linux | Reporter: | Kimmo Sundqvist <kimmo.sundqvist> |
| Component: | Current packages | Assignee: | Gentoo Sound Team <sound> |
| Status: | RESOLVED UPSTREAM | ||
| Severity: | major | ||
| Priority: | High | ||
| Version: | unspecified | ||
| Hardware: | All | ||
| OS: | Linux | ||
| URL: | http://lists.gnu.org/archive/html/fluid-dev/2005-01/msg00002.html | ||
| Whiteboard: | Can't reproduce => upstream | ||
| Package list: | Runtime testing required: | --- | |
|
Description
Kimmo Sundqvist
2005-03-18 11:10:34 UTC
I just committed fluidsynth-1.0.6 to portage cvs. According to Josh Green's release note[1] it fixed quite some bugs. Could you give it a spin and let me know? thanks. [1] http://lists.gnu.org/archive/html/fluid-dev/2005-06/msg00008.html Tested 1.0.6 and it is as if nothing has been fixed. I loaded the Merlin soundfont and tried playing a midi file. pmidi blader.mid -p 128:0 It went ok for two seconds, then came the noise. After about 10 seconds of noise, it could produce no output, i.e it was totally silent. I quitted it, reloaded it and the Merlin soundfont. Tried playing sabre_dance.mid. It was ok for 10 seconds, then I noticed a clicking that got faster and faster, and suddenly the noise (screech, scream, positive feedback loop, whatever) was there again. It went away after some seconds, and then came back again. I stopped pmidi, and fluidsynth segfaulted. It is as if fluidsynth was doing some kind of echo or chorus, and had a positive feedback loop somewhere, thus resulting in an overload. Disabling chorus and reverb on the cmdline did nothing to help with the feedback, though. My emerge info is: Portage 2.0.51.22-r1 (default-linux/x86/2005.0, gcc-3.3.5-20050130, glibc-2.3.4.20041102-r1, 2.6.11-ck9 i686) ================================================================= System uname: 2.6.11-ck9 i686 Pentium III (Coppermine) Gentoo Base System version 1.6.12 distcc 2.18.3 i686-pc-linux-gnu (protocols 1 and 2) (default port 3632) [disabled] dev-lang/python: 2.3.5 sys-apps/sandbox: 1.2.10 sys-devel/autoconf: 2.13, 2.59-r6 sys-devel/automake: 1.4_p6, 1.5, 1.6.3, 1.7.9, 1.8.5-r3, 1.9.5 sys-devel/binutils: 2.15.92.0.2-r10 sys-devel/libtool: 1.5.18-r1 virtual/os-headers: 2.6.11-r2 ACCEPT_KEYWORDS="x86" AUTOCLEAN="yes" CBUILD="i686-pc-linux-gnu" CFLAGS="-Os -mcpu=pentium3 -pipe" CHOST="i686-pc-linux-gnu" CONFIG_PROTECT="/etc /usr/kde/2/share/config /usr/kde/3.4/env /usr/kde/3.4/share/config /usr/kde/3.4/shutdown /usr/kde/3/share/config /usr/lib/X11/xkb /usr/share/config /var/qmail/control" CONFIG_PROTECT_MASK="/etc/gconf /etc/terminfo /etc/env.d" CXXFLAGS="-Os -mcpu=pentium3 -pipe" DISTDIR="/usr/portage/distfiles" FEATURES="autoconfig distlocks sandbox sfperms strict" GENTOO_MIRRORS="http://trumpetti.atm.tut.fi/gentoo/ ftp://trumpetti.atm.tut.fi/gentoo/" LANG="C" MAKEOPTS="-j2" PKGDIR="/usr/portage/packages" PORTAGE_TMPDIR="/var/tmp" PORTDIR="/usr/portage" PORTDIR_OVERLAY="/usr/local/portage" SYNC="rsync://rsync.europe.gentoo.org/gentoo-portage" USE="X acl alsa apm arts avi berkdb bitmap-fonts bonobo cdr crypt cups curl dga eds emboss encode f77 fam fbcon flac foomaticdb fortran gd gdbm gif glut gnome gphoto2 gpm gstreamer gtk gtk2 gtkhtml guile imagemagick imlib ipv6 jack java jikes jpeg jpg junit kde ladcca libg++ libwww mad mikmod mmx motif mp3 mpeg mpi nas ncurses nls nptl objc ogg oggvorbis opengl pam pdflib perl png ppds python qt quicktime readline rtc samba sdl slang speex sqlite sse ssl svga tcltk tcpd tiff truetype truetype-fonts type1-fonts unicode vorbis wmf x86 xine xml xml2 xmms xv xvid zlib userland_GNU kernel_linux elibc_glibc" Unset: ASFLAGS, CTARGET, LC_ALL, LDFLAGS, LINGUAS Tried with -O2 instead of -Os, didn't help. In sabre_dance.mid, the noise only seems to happen when a certain instrument is playing. There is something in the midi file that either changes the program for a channel, enables or disables chorus or reverb, changes volume or something, that starts and stops the screech. The screech is caused by different instrument or channel in the midi than the clicking sound. The clicking sounds like a vinyl turntable skipping a track. After some playing, it just segfaults. And in other midi files, the noise, once it starts, doesn't stop until I quit fluidsynth, fluidsynth segfaults, or the noise dies suddenly. After the sudden silence there is no sound output from fluidsynth at all until I quit or something triggers a segfault. I know that typing 'settings' in the fluidsynth console lists all settings. I see that I can disable both from the cmdline, by --chorus 0 --reverb 0. But typing "chorus 0" or "reverb 0" (or on, off or 1) on the fluidsynth's console doesn't change the settings displayed by typing "settings." Typing "set synth.chorus.active whoa" sets it to whoa. I also tried "toilet" and "toothfairy" and it didn't complain. Wonder if it was enabled or disabled really. Either the synth really doesn't disable chorus and reverb, or it does, but I have no way to figure out if it indeed has, and the problem is not related to reverb and chorus after all. Fluidsynth also randomly, maybe 50% of the time, crashes on startup with fluidsynth: fluid_ladspa.c:719: fluid_LADSPA_run: Assertion `fluid_settings_getint(FxUnit->synth->settings, "synth.effects-channels", &temp)' failed. Aborted I'll try disabling ladcca and recompiling. Disabling ladcca in the USE variable didn't help. Fluidsynth keeps randomly segfaulting, and after the Sabre dance starts playing, Fluidsynth's CPU usage slowly (in about 10 seconds, from Gnome system monitor) climbs to 99% and stays there. Before anyone says anything about my hardware, I think I should mention that I can run memtest86 or The Great Internet Mersenne Prime Search's test for 20 hours with no problems. I'm leaving this alone now. If anyone has suggestions or 1.0.7 comes out, I'll maybe have a look and try again. Hi Kimmo, looking at your updates you've tried already I tried fluidsynth with different soundfounds and different instrument in these soundfonts but was not able to reproduce your problem. Machines used in these tests are pentium4 and amd64. To get this issued nailed down (and hopefully fixed in the 1.0.7 release) you might want to explain your problem to the fluidsynth developers. (http://savannah.nongnu.org/mail/?group=fluid). Please update any progress you make to this bug. Tried installing fluidsynth-1.0.3 and it doesn't have this screech problem. Some instruments still sound like they were missing, especially drums can be heard only in some songs, and CPU usage climbs to 99% with no good reason. Are the reverb and chorus commands supposed to take action immediately or when? Will do more tests with 1.0.3 now that I need not be afraid of the noise. Maybe the instruments are not missing, but their relative volumes are seriously screwed up. I am comparing fluidsynth to Timidity++ and I have noticed that some songs, which come out perfectly with Timidity, have their melody missing or barely audible with fluidsynth, and some background instrument buries everything else. Drums and percussion is usually among the first victims. Still using the Merlin_22 soundfont that has caused me no problems with Timidity. All of this is sad, because I can see (hear) that fluidsynth is really much better at some things than Timidity. I found this inside the ebuild for 1.0.6:
if use sse; then
myconf="--enable-SSE ${myconf}"
# If your CFLAGS include optimizations for sse, ie:
# -march=pentium4 -mfpmath=sse -msse2
# AND your USE flags include sse, ie: USE=sse,
# the sounds with fluidsynth will be distorted.
if [ `is-flag "-march=pentium4"` ]; then
filter-flags "-msse2"
filter-flags "-mfpmath=sse"
fi
fi
I have "-Os -mcpu=pentium3 -pipe"
I have noticed some possibly related issues. These apply to all the builds in portage, as well the cvs version. First, I am unable to use the SSE flag. It causes bad aliasing noise on any notes that are being resampled from their original frequency. Second, any -march setting pentium3 or pentium4 causes certain notes in several soundfonts to be played an octave higher than normal when -O is 2 or 3. It does not happen for me with -Os or -O1 and -march=pentium3. I remerged fluidsynth with CFLAGS="$CFLAGS -march=pentium2" emerge fluidsynth or alternatively CFLAGS="$CFLAGS -O1" emerge fluidsynth and it works fine now (In reply to comment #10) > I have noticed some possibly related issues. These apply to all the builds in > portage, as well the cvs version. > > First, I am unable to use the SSE flag. It causes bad aliasing noise on any > notes that are being resampled from their original frequency. > > Second, any -march setting pentium3 or pentium4 causes certain notes in several > soundfonts to be played an octave higher than normal when -O is 2 or 3. It does > not happen for me with -Os or -O1 and -march=pentium3. > > I remerged fluidsynth with > CFLAGS="$CFLAGS -march=pentium2" emerge fluidsynth > or alternatively > CFLAGS="$CFLAGS -O1" emerge fluidsynth > and it works fine now > (In reply to comment #9) > I found this inside the ebuild for 1.0.6: > > if use sse; then > myconf="--enable-SSE ${myconf}" > # If your CFLAGS include optimizations for sse, ie: > # -march=pentium4 -mfpmath=sse -msse2 > # AND your USE flags include sse, ie: USE=sse, > # the sounds with fluidsynth will be distorted. > if [ `is-flag "-march=pentium4"` ]; then > filter-flags "-msse2" > filter-flags "-mfpmath=sse" > fi > fi > > I have "-Os -mcpu=pentium3 -pipe" Hi, I had a problem with too aggressive cflags: CFLAGS="-O2 -mfpmath=sse,387 -march=pentium4 -pipe -fmove-all-movables -freduce-all-givs -frename-registers -fweb -ffast-math -funswitch-loops -funit-at-a-time" CXXFLAGS="${CFLAGS}" gcc-3.4.5 note resampling dies... recompiled fluidsynth with CFLAGS="" CXXFLAGS="" emerge fluidsynth and qsynth/fluidsynth worked out of the box! Seems like the useflag-filtering is buggy, too. Not reproducable, send this upstream if it's still an issue. |
