Christoph Senjak <christoph.sen...@...> writes: > > Hello. > > I dont want to use sdl-mixer if possible. I found cffi/audio.lisp, > including the SDL_OpenAudio-Command which takes a struct with a > mixer-callback. Is it possible to use this? Is there any reason not to > use it (i.e. callbacks too slow, etc.)? > > I am asking this because these functions are not mentioned in the documentation.
- sdl-mixer supports both streams and samples of many audio formats. - lisbuilder-sdl currently supports samples in WAV format only. Supporting audio in SDL posed a problem because the SDL process spawns a new thread (unknown to the Lisp process) to handle the audio callback. Lisps tend to dislike callbacks into the Lisp process from "unknown" threads. There are three mechanisms for handling audio in SDL; #1 sdl-mixer Does not need a callback into the Lisp process to fill the audio buffer. It takes case of this internally. Seems to work for all Lisps I have tried. Perhaps a little flakey under Windows if called repeatedly from the same Lisp process (e.g. during development). #2 lispbuilder-sdl without the glue library The "unknown" audio thread calls into the Lisp process to fill the audio buffer. Works fine for Lispworks but other Lisps including SBCL and CCL tend to core, especially in Win32. #3 lispbuilder-sdl with the glue library The audio callback is handled by the glue library. Lisp calls the glue library once per game loop asking (polling) if the audio buffer is empty and then filling the audio buffer if necessary. Seems to work for Lispworks/win32 and SBCL/Linux/Win32. I haven't heard of reports of the glue library crashing in OSX. Perhaps an OSX user can verify this by determining if :lispbuilder-sdl- audio is in *FEATURES*, and running (SDL-EXAMPLES:MIXER-TEST). I tend to code primarily in Lispworks which is why both #2 and #3 are included in lispbuilder-sdl (Why do something once when you can do it twice and take twice as long.) Audio processing is as follows; * lispbuilder-sdl with the glue library; Buried inside the WITH-EVENTS macro is a call to PROCESS-AUDIO that is made each game loop. PROCESS-AUDIO handles #3. If the glue library is unavailable then PROCESS-AUDIO is a no-op. PROCESS-AUDIO will call FILL-AUDIO-BUFFER if the glue library indicates that the audio buffer is empty. If you look at the source of sdl/mixer.lisp you will see that FILL-AUDIO-BUFFER is also defined as the audio callback function. But this callback is *not used* when the glue library is available. * lispbuilder-sdl without the glue library The PROCESS-AUDIO call within WITH-EVENTS is a no-op. As described in the previous section, FILL-AUDIO-BUFFER is defined as the audio callback function. The audio thread performs a callback into FILL-AUDIO-BUFFER to fill the audio buffer when necessary. lispbuilder-sdl loads WAV samples from disk. These samples are loaded in their entirety into memory. I haven't added support yet for audio streams. So right now lispbuilder-sdl is useful for creating old-school "Tracker" music. However the tracker will also have to be written by yourself. If you want to generate samples on-the-fly from Lisp then take a look at LOAD- AUDIO and LOAD-SAMPLE in sdl/mixer.lisp. I'll add 'official' support for "Lisp- generated" samples as soon as I finish up integrating sdl-gfx into lispbuilder- sdl. Basically, LOAD-SAMPLE converts the WAV file from a foreign array of bytes into a Lisp array. This is an optimization allowing the audio mixer to operate on Lisp arrays instead of having to bit-twiddle bytes via CFFI. The disadvantage is that the output from the mixer must be converted back into a foreign array when written to the audio buffer. LOAD-AUDIO calls BUILD-AUDIO-CVT to convert the output of LOAD-SAMPLE to the same audio format of the mixer. For example if the WAV sample is 8-bit unsigned and the mixer is setup for 16-bit unsigned, then following the call to BUILD- AUDIO-CVT the sample will have been converted to 16-bit unsigned. This is something to take note of if you want to create your own audio from Lisp as the mixer expects the audio samples to be in the same format at mixing time. I just realized that it is only possible to use a custom audio mixer as a replacement for FILL-AUDIO-MIXER if the glue library is not used. I'll fix this. As to your question; If you want the best performance then you have to use sdl- mixer as there is quite some overhead involved when lispbuilder-sdl performs the byte-copy over the FFI interface to fill the audio buffer. What I plan to do in the future is to use pinned vectors which should hopefully make the mixer in lispbuilder-sdl as efficient as sdl-mixer by removing the need for the the byte-copy. I hope this hasn't discouraged you. -Luke _______________________________________________ application-builder mailing list application-builder@lispniks.com http://www.lispniks.com/mailman/listinfo/application-builder
