hi again
i made a patch, that uses several instances of an abstraction based on the partconv~ external in order to do binaural spatialization of sounds using a set of hrtf tables. the hrtf files are all loaded on loadbang into tables, so that they can be accessed quickly. it is part of the patch, that a 'set <tablename>' message to all [partconv~]s is sent quite often, which leads to dropouts on the box, where the installation is meant to run (for some reason not on my personal laptop). i also noticed a difference in performance between an old version of partconv~ (0.1) and the current svn version, whereas the latter is more likely to cause dropout at high 'set' message rates.
there are some things, that i don't know, which might affect performance. for instance, what is the recommended blocksize of the subpatch, that contains an object [partconv~]? is the optimal blocksize dependent on the size of the impulse response? is using [partconv~] generally not the best option to implement hrtf? if so, what are better ways?
thanks in advance for any help.
roman
Telefonate ohne weitere Kosten vom PC zum PC: http://messenger.yahoo.de
Hallo, Roman Haefeli hat gesagt: // Roman Haefeli wrote:
i made a patch, that uses several instances of an abstraction based on the partconv~ external in order to do binaural spatialization of sounds using a set of hrtf tables. the hrtf files are all loaded on loadbang into tables, so that they can be accessed quickly. it is part of the patch, that a 'set <tablename>' message to all [partconv~]s is sent quite often, which leads to dropouts on the box, where the installation is meant to run (for some reason not on my personal laptop). i also noticed a difference in performance between an old version of partconv~ (0.1) and the current svn version, whereas the latter is more likely to cause dropout at high 'set' message rates.
I had a quick look at the partconv~ source code and found that everytime a "set" message is sent, partconv~ does quite a bit of memory allocation to configure itself for the new convolution core. This may be a reason for the dropouts if you "set" the objects with a high rate, as memory allocation is not very realtime friendly. You could try to avoid some of the "set"s by creating more partconv~ objects that statically refer to their tables. Of course that's a bit tricky with 1400 tables ...
Ciao
Frank Barknecht _ ______footils.org__
On Mon, 2008-05-19 at 09:12 +0200, Frank Barknecht wrote:
Hallo, Roman Haefeli hat gesagt: // Roman Haefeli wrote:
i made a patch, that uses several instances of an abstraction based on the partconv~ external in order to do binaural spatialization of sounds using a set of hrtf tables. the hrtf files are all loaded on loadbang into tables, so that they can be accessed quickly. it is part of the patch, that a 'set <tablename>' message to all [partconv~]s is sent quite often, which leads to dropouts on the box, where the installation is meant to run (for some reason not on my personal laptop). i also noticed a difference in performance between an old version of partconv~ (0.1) and the current svn version, whereas the latter is more likely to cause dropout at high 'set' message rates.
I had a quick look at the partconv~ source code and found that everytime a "set" message is sent, partconv~ does quite a bit of memory allocation to configure itself for the new convolution core. This may be a reason for the dropouts if you "set" the objects with a high rate, as memory allocation is not very realtime friendly. You could try to avoid some of the "set"s by creating more partconv~ objects that statically refer to their tables. Of course that's a bit tricky with 1400 tables ...
thanks for confirming, what i guessed. so one (probably not the smartest) solution is to use a computer with fast memory access. in my case, lowering the message rate and increasing jackd buffersize a bit already helped.
roman
Telefonate ohne weitere Kosten vom PC zum PC: http://messenger.yahoo.de
-
Frank Barknecht -
Roman Haefeli