rlm@15: #+title: Simulated Sense of Hearing
rlm@0: #+author: Robert McIntyre
rlm@0: #+email: rlm@mit.edu
rlm@15: #+description: Simulating multiple listeners and the sense of hearing in jMonkeyEngine3
rlm@15: #+keywords: simulated hearing, openal, clojure, jMonkeyEngine3, LWJGL, AI
rlm@15: #+SETUPFILE: ../../aurellem/org/setup.org
rlm@15: #+INCLUDE: ../../aurellem/org/level-0.org
rlm@0: #+BABEL: :exports both :noweb yes :cache no :mkdirp yes
rlm@0: 
rlm@15: * Hearing
rlm@0: 
rlm@20: I want to be able to place ears in a similar manner to how I place
rlm@0: the eyes.  I want to be able to place ears in a unique spatial
rlm@20: position, and receive as output at every tick the F.F.T. of whatever
rlm@0: signals are happening at that point.
rlm@0: 
rlm@15: Hearing is one of the more difficult senses to simulate, because there
rlm@15: is less support for obtaining the actual sound data that is processed
rlm@15: by jMonkeyEngine3.
rlm@15: 
rlm@15: jMonkeyEngine's sound system works as follows:
rlm@15: 
rlm@20:  - jMonkeyEngine uses the =AppSettings= for the particular application
rlm@15:    to determine what sort of =AudioRenderer= should be used.
rlm@15:  - although some support is provided for multiple AudioRendering
rlm@15:    backends, jMonkeyEngine at the time of this writing will either
rlm@20:    pick no AudioRenderer at all, or the =LwjglAudioRenderer=
rlm@15:  - jMonkeyEngine tries to figure out what sort of system you're
rlm@20:    running and extracts the appropriate native libraries.
rlm@18:  - the =LwjglAudioRenderer= uses the [[http://lwjgl.org/][=LWJGL=]] (LightWeight Java Game
rlm@18:    Library) bindings to interface with a C library called [[http://kcat.strangesoft.net/openal.html][=OpenAL=]]
rlm@15:  - =OpenAL= calculates the 3D sound localization and feeds a stream of
rlm@15:    sound to any of various sound output devices with which it knows
rlm@15:    how to communicate.
rlm@15: 
rlm@15: A consequence of this is that there's no way to access the actual
rlm@20: sound data produced by =OpenAL=.  Even worse, =OpenAL= only supports
rlm@15: one /listener/, which normally isn't a problem for games, but becomes
rlm@15: a problem when trying to make multiple AI creatures that can each hear
rlm@15: the world from a different perspective.
rlm@15: 
rlm@15: To make many AI creatures in jMonkeyEngine that can each hear the
rlm@15: world from their own perspective, it is necessary to go all the way
rlm@15: back to =OpenAL= and implement support for simulated hearing there.
rlm@15: 
rlm@19: * Extending =OpenAL=
rlm@15: ** =OpenAL= Devices
rlm@15:    
rlm@15: =OpenAL= goes to great lengths to support many different systems, all
rlm@20: with different sound capabilities and interfaces.  It accomplishes this
rlm@15: difficult task by providing code for many different sound backends in
rlm@15: pseudo-objects called /Devices/.  There's a device for the Linux Open
rlm@20: Sound System and the Advanced Linux Sound Architecture, there's one
rlm@15: for Direct Sound on Windows, there's even one for Solaris. =OpenAL=
rlm@15: solves the problem of platform independence by providing all these
rlm@15: Devices.
rlm@15: 
rlm@15: Wrapper libraries such as LWJGL are free to examine the system on
rlm@20: which they are running and then select an appropriate device for that
rlm@15: system. 
rlm@15: 
rlm@15: There are also a few "special" devices that don't interface with any
rlm@15: particular system.  These include the Null Device, which doesn't do
rlm@20: anything, and the Wave Device, which writes whatever sound it receives
rlm@15: to a file, if everything has been set up correctly when configuring
rlm@15: =OpenAL=.
rlm@15: 
rlm@15: Actual mixing of the sound data happens in the Devices, and they are
rlm@15: the only point in the sound rendering process where this data is
rlm@15: available. 
rlm@15: 
rlm@15: Therefore, in order to support multiple listeners, and get the sound
rlm@15: data in a form that the AIs can use, it is necessary to create a new
rlm@15: Device, which supports this features.
rlm@15: 
rlm@15: ** The Send Device
rlm@15: Adding a device to OpenAL is rather tricky -- there are five separate
rlm@15: files in the =OpenAL= source tree that must be modified to do so. I've
rlm@15: documented this process [[./add-new-device.org][here]] for anyone who is interested.
rlm@15: 
rlm@18: 
rlm@18: Onward to that actual Device!
rlm@18: 
rlm@18: again, my objectives are:
rlm@18: 
rlm@18:  - Support Multiple Listeners from jMonkeyEngine3
rlm@18:  - Get access to the rendered sound data for further processing from
rlm@18:    clojure.
rlm@18: 
rlm@18: ** =send.c=
rlm@18: 
rlm@18: ** Header
rlm@18: #+srcname: send-header
rlm@15: #+begin_src C
rlm@15: #include "config.h"
rlm@15: #include <stdlib.h>
rlm@15: #include "alMain.h"
rlm@15: #include "AL/al.h"
rlm@15: #include "AL/alc.h"
rlm@15: #include "alSource.h"
rlm@15: #include <jni.h>
rlm@15: 
rlm@15: //////////////////// Summary
rlm@15: 
rlm@15: struct send_data;
rlm@15: struct context_data;
rlm@15: 
rlm@15: static void addContext(ALCdevice *, ALCcontext *);
rlm@15: static void syncContexts(ALCcontext *master, ALCcontext *slave);
rlm@15: static void syncSources(ALsource *master, ALsource *slave, 
rlm@15: 			ALCcontext *masterCtx, ALCcontext *slaveCtx);
rlm@15: 
rlm@15: static void syncSourcei(ALuint master, ALuint slave,
rlm@15: 			ALCcontext *masterCtx, ALCcontext *ctx2, ALenum param);
rlm@15: static void syncSourcef(ALuint master, ALuint slave,
rlm@15: 			ALCcontext *masterCtx, ALCcontext *ctx2, ALenum param);
rlm@15: static void syncSource3f(ALuint master, ALuint slave,
rlm@15: 			ALCcontext *masterCtx, ALCcontext *ctx2, ALenum param);
rlm@15: 
rlm@15: static void swapInContext(ALCdevice *, struct context_data *);
rlm@15: static void saveContext(ALCdevice *, struct context_data *);
rlm@15: static void limitContext(ALCdevice *, ALCcontext *);
rlm@15: static void unLimitContext(ALCdevice *);
rlm@15: 
rlm@15: static void init(ALCdevice *);
rlm@15: static void renderData(ALCdevice *, int samples);
rlm@15: 
rlm@15: #define UNUSED(x)  (void)(x)
rlm@18: #+end_src
rlm@15: 
rlm@20: The main idea behind the Send device is to take advantage of the fact
rlm@18: that LWJGL only manages one /context/ when using OpenAL.  A /context/
rlm@18: is like a container that holds samples and keeps track of where the
rlm@18: listener is.  In order to support multiple listeners, the Send device
rlm@18: identifies the LWJGL context as the master context, and creates any
rlm@18: number of slave contexts to represent additional listeners.  Every
rlm@18: time the device renders sound, it synchronizes every source from the
rlm@18: master LWJGL context to the slave contexts.  Then, it renders each
rlm@18: context separately, using a different listener for each one.  The
rlm@18: rendered sound is made available via JNI to jMonkeyEngine.
rlm@18: 
rlm@18: To recap, the process is: 
rlm@18:  - Set the LWJGL context as "master" in the =init()= method.
rlm@18:  - Create any number of additional contexts via =addContext()=
rlm@18:  - At every call to =renderData()= sync the master context with the
rlm@20:    slave contexts with =syncContexts()=
rlm@18:  - =syncContexts()= calls =syncSources()= to sync all the sources
rlm@18:    which are in the master context.
rlm@18:  - =limitContext()= and =unLimitContext()= make it possible to render
rlm@18:    only one context at a time.
rlm@18: 
rlm@18: ** Necessary State
rlm@21: #+srcname: send-state
rlm@18: #+begin_src C
rlm@15: ////////////////////  State
rlm@15: 
rlm@15: typedef struct context_data {
rlm@15:   ALfloat ClickRemoval[MAXCHANNELS];
rlm@15:   ALfloat PendingClicks[MAXCHANNELS];
rlm@15:   ALvoid *renderBuffer;
rlm@15:   ALCcontext *ctx;
rlm@15: } context_data;
rlm@15: 
rlm@15: typedef struct send_data {
rlm@15:   ALuint size;
rlm@15:   context_data **contexts;
rlm@15:   ALuint numContexts;
rlm@15:   ALuint maxContexts;
rlm@15: } send_data;
rlm@18: #+end_src
rlm@15: 
rlm@18: Switching between contexts is not the normal operation of a Device,
rlm@18: and one of the problems with doing so is that a Device normally keeps
rlm@18: around a few pieces of state such as the =ClickRemoval= array above
rlm@18: which will become corrupted if the contexts are not done in
rlm@18: parallel. The solution is to create a copy of this normally global
rlm@18: device state for each context, and copy it back and forth into and out
rlm@18: of the actual device state whenever a context is rendered.
rlm@15: 
rlm@18: ** Synchronization Macros
rlm@21: #+srcname: sync-macros
rlm@18: #+begin_src C
rlm@15: ////////////////////  Context Creation / Synchronization
rlm@15: 
rlm@15: #define _MAKE_SYNC(NAME, INIT_EXPR, GET_EXPR, SET_EXPR)	\
rlm@15:   void NAME (ALuint sourceID1, ALuint sourceID2,	\
rlm@15: 	     ALCcontext *ctx1, ALCcontext *ctx2,	\
rlm@15: 	     ALenum param){				\
rlm@15:     INIT_EXPR;						\
rlm@15:     ALCcontext *current = alcGetCurrentContext();	\
rlm@15:     alcMakeContextCurrent(ctx1);			\
rlm@15:     GET_EXPR;						\
rlm@15:     alcMakeContextCurrent(ctx2);			\
rlm@15:     SET_EXPR;						\
rlm@15:     alcMakeContextCurrent(current);			\
rlm@15:   }
rlm@15:   
rlm@15: #define MAKE_SYNC(NAME, TYPE, GET, SET)			\
rlm@15:   _MAKE_SYNC(NAME,					\
rlm@15: 	     TYPE value,				\
rlm@15:              GET(sourceID1, param, &value),		\
rlm@15:              SET(sourceID2, param, value))
rlm@15:   
rlm@15: #define MAKE_SYNC3(NAME, TYPE, GET, SET)				\
rlm@15:   _MAKE_SYNC(NAME,							\
rlm@15: 	     TYPE value1; TYPE value2; TYPE value3;,			\
rlm@15: 	     GET(sourceID1, param, &value1, &value2, &value3),		\
rlm@15: 	     SET(sourceID2, param,  value1,  value2,  value3))
rlm@15: 
rlm@15: MAKE_SYNC( syncSourcei,  ALint,   alGetSourcei,  alSourcei);
rlm@15: MAKE_SYNC( syncSourcef,  ALfloat, alGetSourcef,  alSourcef);
rlm@15: MAKE_SYNC3(syncSource3i, ALint,   alGetSource3i, alSource3i);
rlm@15: MAKE_SYNC3(syncSource3f, ALfloat, alGetSource3f, alSource3f);
rlm@15:   
rlm@18: #+end_src
rlm@18: 
rlm@20: Setting the state of an =OpenAL= source is done with the =alSourcei=,
rlm@18: =alSourcef=, =alSource3i=, and =alSource3f= functions.  In order to
rlm@20: completely synchronize two sources, it is necessary to use all of
rlm@18: them. These macros help to condense the otherwise repetitive
rlm@20: synchronization code involving these similar low-level =OpenAL= functions.
rlm@18: 
rlm@18: ** Source Synchronization
rlm@21: #+srcname: sync-sources
rlm@18: #+begin_src C
rlm@15: void syncSources(ALsource *masterSource, ALsource *slaveSource, 
rlm@15: 		 ALCcontext *masterCtx, ALCcontext *slaveCtx){
rlm@15:   ALuint master = masterSource->source;
rlm@15:   ALuint slave = slaveSource->source;
rlm@15:   ALCcontext *current = alcGetCurrentContext();
rlm@15: 
rlm@15:   syncSourcef(master,slave,masterCtx,slaveCtx,AL_PITCH);
rlm@15:   syncSourcef(master,slave,masterCtx,slaveCtx,AL_GAIN);
rlm@15:   syncSourcef(master,slave,masterCtx,slaveCtx,AL_MAX_DISTANCE);
rlm@15:   syncSourcef(master,slave,masterCtx,slaveCtx,AL_ROLLOFF_FACTOR);
rlm@15:   syncSourcef(master,slave,masterCtx,slaveCtx,AL_REFERENCE_DISTANCE);
rlm@15:   syncSourcef(master,slave,masterCtx,slaveCtx,AL_MIN_GAIN);
rlm@15:   syncSourcef(master,slave,masterCtx,slaveCtx,AL_MAX_GAIN);
rlm@15:   syncSourcef(master,slave,masterCtx,slaveCtx,AL_CONE_OUTER_GAIN);
rlm@15:   syncSourcef(master,slave,masterCtx,slaveCtx,AL_CONE_INNER_ANGLE);
rlm@15:   syncSourcef(master,slave,masterCtx,slaveCtx,AL_CONE_OUTER_ANGLE);
rlm@15:   syncSourcef(master,slave,masterCtx,slaveCtx,AL_SEC_OFFSET);
rlm@15:   syncSourcef(master,slave,masterCtx,slaveCtx,AL_SAMPLE_OFFSET);
rlm@15:   syncSourcef(master,slave,masterCtx,slaveCtx,AL_BYTE_OFFSET);
rlm@15:     
rlm@15:   syncSource3f(master,slave,masterCtx,slaveCtx,AL_POSITION);
rlm@15:   syncSource3f(master,slave,masterCtx,slaveCtx,AL_VELOCITY);
rlm@15:   syncSource3f(master,slave,masterCtx,slaveCtx,AL_DIRECTION);
rlm@15:   
rlm@15:   syncSourcei(master,slave,masterCtx,slaveCtx,AL_SOURCE_RELATIVE);
rlm@15:   syncSourcei(master,slave,masterCtx,slaveCtx,AL_LOOPING);
rlm@15: 
rlm@15:   alcMakeContextCurrent(masterCtx);
rlm@15:   ALint source_type;
rlm@15:   alGetSourcei(master, AL_SOURCE_TYPE, &source_type);
rlm@15: 
rlm@15:   // Only static sources are currently synchronized! 
rlm@15:   if (AL_STATIC == source_type){
rlm@15:     ALint master_buffer;
rlm@15:     ALint slave_buffer;
rlm@15:     alGetSourcei(master, AL_BUFFER, &master_buffer);
rlm@15:     alcMakeContextCurrent(slaveCtx);
rlm@15:     alGetSourcei(slave, AL_BUFFER, &slave_buffer);
rlm@15:     if (master_buffer != slave_buffer){
rlm@15:       alSourcei(slave, AL_BUFFER, master_buffer);
rlm@15:     }
rlm@15:   }
rlm@15:   
rlm@15:   // Synchronize the state of the two sources.
rlm@15:   alcMakeContextCurrent(masterCtx);
rlm@15:   ALint masterState;
rlm@15:   ALint slaveState;
rlm@15: 
rlm@15:   alGetSourcei(master, AL_SOURCE_STATE, &masterState);
rlm@15:   alcMakeContextCurrent(slaveCtx);
rlm@15:   alGetSourcei(slave, AL_SOURCE_STATE, &slaveState);
rlm@15: 
rlm@15:   if (masterState != slaveState){
rlm@15:     switch (masterState){
rlm@15:     case AL_INITIAL : alSourceRewind(slave); break;
rlm@15:     case AL_PLAYING : alSourcePlay(slave);   break;
rlm@15:     case AL_PAUSED  : alSourcePause(slave);  break;
rlm@15:     case AL_STOPPED : alSourceStop(slave);   break;
rlm@15:     }
rlm@15:   }
rlm@15:   // Restore whatever context was previously active.
rlm@15:   alcMakeContextCurrent(current);
rlm@15: }
rlm@18: #+end_src
rlm@20: This function is long because it has to exhaustively go through all the
rlm@18: possible state that a source can have and make sure that it is the
rlm@18: same between the master and slave sources.  I'd like to take this
rlm@18: moment to salute the [[http://connect.creativelabs.com/openal/Documentation/Forms/AllItems.aspx][=OpenAL= Reference Manual]], which provides a very
rlm@18: good description of =OpenAL='s internals.
rlm@15: 
rlm@18: ** Context Synchronization
rlm@21: #+srcname: sync-contexts
rlm@18: #+begin_src C
rlm@15: void syncContexts(ALCcontext *master, ALCcontext *slave){
rlm@15:   /* If there aren't sufficient sources in slave to mirror 
rlm@15:      the sources in master, create them. */
rlm@15:   ALCcontext *current = alcGetCurrentContext();
rlm@15: 
rlm@15:   UIntMap *masterSourceMap = &(master->SourceMap);
rlm@15:   UIntMap *slaveSourceMap = &(slave->SourceMap);
rlm@15:   ALuint numMasterSources = masterSourceMap->size;
rlm@15:   ALuint numSlaveSources = slaveSourceMap->size;
rlm@15: 
rlm@15:   alcMakeContextCurrent(slave);
rlm@15:   if (numSlaveSources < numMasterSources){
rlm@15:     ALuint numMissingSources = numMasterSources - numSlaveSources;
rlm@15:     ALuint newSources[numMissingSources];
rlm@15:     alGenSources(numMissingSources, newSources);
rlm@15:   }
rlm@15: 
rlm@20:   /* Now, slave is guaranteed to have at least as many sources
rlm@15:      as master.  Sync each source from master to the corresponding
rlm@15:      source in slave. */
rlm@15:   int i;
rlm@15:   for(i = 0; i < masterSourceMap->size; i++){
rlm@15:     syncSources((ALsource*)masterSourceMap->array[i].value,
rlm@15: 		(ALsource*)slaveSourceMap->array[i].value,
rlm@15: 		master, slave);
rlm@15:   }
rlm@15:   alcMakeContextCurrent(current);
rlm@15: }
rlm@18: #+end_src
rlm@15: 
rlm@18: Most of the hard work in Context Synchronization is done in
rlm@18: =syncSources()=. The only thing that =syncContexts()= has to worry
rlm@20: about is automatically creating new sources whenever a slave context
rlm@18: does not have the same number of sources as the master context.
rlm@18: 
rlm@19: ** Context Creation
rlm@21: #+srcname: context-creation
rlm@18: #+begin_src C
rlm@15: static void addContext(ALCdevice *Device, ALCcontext *context){
rlm@15:   send_data *data = (send_data*)Device->ExtraData;
rlm@15:   // expand array if necessary
rlm@15:   if (data->numContexts >= data->maxContexts){
rlm@15:     ALuint newMaxContexts = data->maxContexts*2 + 1;
rlm@15:     data->contexts = realloc(data->contexts, newMaxContexts*sizeof(context_data));
rlm@15:     data->maxContexts = newMaxContexts;
rlm@15:   }
rlm@15:   // create context_data and add it to the main array
rlm@15:   context_data *ctxData;
rlm@15:   ctxData = (context_data*)calloc(1, sizeof(*ctxData));
rlm@15:   ctxData->renderBuffer = 
rlm@15:     malloc(BytesFromDevFmt(Device->FmtType) * 
rlm@15: 	   Device->NumChan * Device->UpdateSize);
rlm@15:   ctxData->ctx = context;
rlm@15: 
rlm@15:   data->contexts[data->numContexts] = ctxData;
rlm@15:   data->numContexts++;
rlm@15: }
rlm@18: #+end_src
rlm@15: 
rlm@18: Here, the slave context is created, and it's data is stored in the
rlm@18: device-wide =ExtraData= structure.  The =renderBuffer= that is created
rlm@18: here is where the rendered sound samples for this slave context will
rlm@18: eventually go.
rlm@15: 
rlm@19: ** Context Switching
rlm@21: #+srcname: context-switching
rlm@18: #+begin_src C
rlm@15: ////////////////////  Context Switching 
rlm@15: 
rlm@15: /* A device brings along with it two pieces of state
rlm@15:  * which have to be swapped in and out with each context.
rlm@15:  */
rlm@15: static void swapInContext(ALCdevice *Device, context_data *ctxData){
rlm@15:   memcpy(Device->ClickRemoval, ctxData->ClickRemoval, sizeof(ALfloat)*MAXCHANNELS);
rlm@15:   memcpy(Device->PendingClicks, ctxData->PendingClicks, sizeof(ALfloat)*MAXCHANNELS);
rlm@15: }
rlm@15: 
rlm@15: static void saveContext(ALCdevice *Device, context_data *ctxData){
rlm@15:   memcpy(ctxData->ClickRemoval, Device->ClickRemoval, sizeof(ALfloat)*MAXCHANNELS);
rlm@15:   memcpy(ctxData->PendingClicks, Device->PendingClicks, sizeof(ALfloat)*MAXCHANNELS);
rlm@15: }  
rlm@15: 
rlm@15: static ALCcontext **currentContext;
rlm@15: static ALuint currentNumContext;
rlm@15: 
rlm@15: /* By default, all contexts are rendered at once for each call to aluMixData.
rlm@20:  * This function uses the internals of the ALCdevice struct to temporally 
rlm@15:  * cause aluMixData to only render the chosen context.
rlm@15:  */
rlm@15: static void limitContext(ALCdevice *Device, ALCcontext *ctx){
rlm@15:   currentContext  = Device->Contexts;
rlm@15:   currentNumContext  = Device->NumContexts;
rlm@15:   Device->Contexts = &ctx;
rlm@15:   Device->NumContexts = 1;
rlm@15: }
rlm@15: 
rlm@15: static void unLimitContext(ALCdevice *Device){
rlm@15:   Device->Contexts = currentContext;
rlm@15:   Device->NumContexts = currentNumContext;
rlm@15: }
rlm@18: #+end_src
rlm@15: 
rlm@20: =OpenAL= normally renders all Contexts in parallel, outputting the
rlm@18: whole result to the buffer.  It does this by iterating over the
rlm@18: Device->Contexts array and rendering each context to the buffer in
rlm@20: turn.  By temporally setting Device->NumContexts to 1 and adjusting
rlm@18: the Device's context list to put the desired context-to-be-rendered
rlm@18: into position 0, we can get trick =OpenAL= into rendering each slave
rlm@18: context separate from all the others.
rlm@15: 
rlm@18: ** Main Device Loop
rlm@21: #+srcname: main-loop
rlm@18: #+begin_src C
rlm@15: ////////////////////   Main Device Loop
rlm@15: 
rlm@18: /* Establish the LWJGL context as the master context, which will
rlm@15:  * be synchronized to all the slave contexts
rlm@15:  */
rlm@15: static void init(ALCdevice *Device){
rlm@15:   ALCcontext *masterContext = alcGetCurrentContext();
rlm@15:   addContext(Device, masterContext);
rlm@15: }
rlm@15: 
rlm@15: 
rlm@15: static void renderData(ALCdevice *Device, int samples){
rlm@15:   if(!Device->Connected){return;}
rlm@15:   send_data *data = (send_data*)Device->ExtraData;
rlm@15:   ALCcontext *current = alcGetCurrentContext();
rlm@15: 
rlm@15:   ALuint i;
rlm@15:   for (i = 1; i < data->numContexts; i++){
rlm@15:     syncContexts(data->contexts[0]->ctx , data->contexts[i]->ctx);
rlm@15:   }
rlm@15:   
rlm@15:   if ((uint) samples > Device->UpdateSize){
rlm@15:     printf("exceeding internal buffer size; dropping samples\n");
rlm@15:     printf("requested %d; available %d\n", samples, Device->UpdateSize);
rlm@15:     samples = (int) Device->UpdateSize;
rlm@15:   }
rlm@15: 
rlm@15:   for (i = 0; i < data->numContexts; i++){
rlm@15:     context_data *ctxData = data->contexts[i];
rlm@15:     ALCcontext *ctx = ctxData->ctx;
rlm@15:     alcMakeContextCurrent(ctx);
rlm@15:     limitContext(Device, ctx);
rlm@15:     swapInContext(Device, ctxData);
rlm@15:     aluMixData(Device, ctxData->renderBuffer, samples);
rlm@15:     saveContext(Device, ctxData);
rlm@15:     unLimitContext(Device);
rlm@15:   }
rlm@15:   alcMakeContextCurrent(current);
rlm@15: }
rlm@18: #+end_src
rlm@15: 
rlm@18: The main loop synchronizes the master LWJGL context with all the slave
rlm@18: contexts, then walks each context, rendering just that context to it's
rlm@18: audio-sample storage buffer.
rlm@15: 
rlm@19: ** JNI Methods
rlm@19: 
rlm@19: At this point, we have the ability to create multiple listeners by
rlm@19: using the master/slave context trick, and the rendered audio data is
rlm@19: waiting patiently in internal buffers, one for each listener.  We need
rlm@19: a way to transport this information to Java, and also a way to drive
rlm@19: this device from Java.  The following JNI interface code is inspired
rlm@19: by the way LWJGL interfaces with =OpenAL=.
rlm@19: 
rlm@19: *** step
rlm@21: #+srcname: jni-step
rlm@18: #+begin_src C
rlm@15: ////////////////////   JNI Methods
rlm@15: 
rlm@15: #include "com_aurellem_send_AudioSend.h"
rlm@15: 
rlm@15: /*
rlm@15:  * Class:     com_aurellem_send_AudioSend
rlm@15:  * Method:    nstep
rlm@15:  * Signature: (JI)V
rlm@15:  */
rlm@15: JNIEXPORT void JNICALL Java_com_aurellem_send_AudioSend_nstep
rlm@15: (JNIEnv *env, jclass clazz, jlong device, jint samples){
rlm@15:   UNUSED(env);UNUSED(clazz);UNUSED(device);
rlm@15:   renderData((ALCdevice*)((intptr_t)device), samples);
rlm@15: }
rlm@19: #+end_src
rlm@19: This device, unlike most of the other devices in =OpenAL=, does not
rlm@19: render sound unless asked. This enables the system to slow down or
rlm@19: speed up depending on the needs of the AIs who are using it to
rlm@19: listen.  If the device tried to render samples in real-time, a
rlm@19: complicated AI whose mind takes 100 seconds of computer time to
rlm@19: simulate 1 second of AI-time would miss almost all of the sound in
rlm@19: its environment.
rlm@15: 
rlm@19: 
rlm@19: *** getSamples
rlm@21: #+srcname: jni-get-samples
rlm@19: #+begin_src C
rlm@15: /*
rlm@15:  * Class:     com_aurellem_send_AudioSend
rlm@15:  * Method:    ngetSamples
rlm@15:  * Signature: (JLjava/nio/ByteBuffer;III)V
rlm@15:  */
rlm@15: JNIEXPORT void JNICALL Java_com_aurellem_send_AudioSend_ngetSamples
rlm@15: (JNIEnv *env, jclass clazz, jlong device, jobject buffer, jint position, 
rlm@15:  jint samples, jint n){
rlm@15:   UNUSED(clazz);  
rlm@15: 
rlm@15:   ALvoid *buffer_address = 
rlm@15:     ((ALbyte *)(((char*)(*env)->GetDirectBufferAddress(env, buffer)) + position));
rlm@15:   ALCdevice *recorder = (ALCdevice*) ((intptr_t)device);
rlm@15:   send_data *data = (send_data*)recorder->ExtraData;
rlm@15:   if ((ALuint)n > data->numContexts){return;}
rlm@15:   memcpy(buffer_address, data->contexts[n]->renderBuffer, 
rlm@15: 	 BytesFromDevFmt(recorder->FmtType) * recorder->NumChan * samples);
rlm@15: }
rlm@19: #+end_src
rlm@15: 
rlm@19: This is the transport layer between C and Java that will eventually
rlm@19: allow us to access rendered sound data from clojure.
rlm@19: 
rlm@19: *** Listener Management
rlm@19: 
rlm@19: =addListener=, =setNthListenerf=, and =setNthListener3f= are
rlm@19: necessary to change the properties of any listener other than the
rlm@19: master one, since only the listener of the current active context is
rlm@19: affected by the normal =OpenAL= listener calls.
rlm@21: #+srcname: listener-manage
rlm@19: #+begin_src C
rlm@15: /*
rlm@15:  * Class:     com_aurellem_send_AudioSend
rlm@15:  * Method:    naddListener
rlm@15:  * Signature: (J)V
rlm@15:  */
rlm@15: JNIEXPORT void JNICALL Java_com_aurellem_send_AudioSend_naddListener
rlm@15: (JNIEnv *env, jclass clazz, jlong device){
rlm@15:   UNUSED(env); UNUSED(clazz);
rlm@15:   //printf("creating new context via naddListener\n");
rlm@15:   ALCdevice *Device = (ALCdevice*) ((intptr_t)device);
rlm@15:   ALCcontext *new = alcCreateContext(Device, NULL);
rlm@15:   addContext(Device, new);
rlm@15: }
rlm@15: 
rlm@15: /*
rlm@15:  * Class:     com_aurellem_send_AudioSend
rlm@15:  * Method:    nsetNthListener3f
rlm@15:  * Signature: (IFFFJI)V
rlm@15:  */
rlm@15: JNIEXPORT void JNICALL Java_com_aurellem_send_AudioSend_nsetNthListener3f
rlm@15:   (JNIEnv *env, jclass clazz, jint param, 
rlm@15:    jfloat v1, jfloat v2, jfloat v3, jlong device, jint contextNum){
rlm@15:   UNUSED(env);UNUSED(clazz);
rlm@15: 
rlm@15:   ALCdevice *Device = (ALCdevice*) ((intptr_t)device);
rlm@15:   send_data *data = (send_data*)Device->ExtraData;
rlm@15:   
rlm@15:   ALCcontext *current = alcGetCurrentContext();
rlm@15:   if ((ALuint)contextNum > data->numContexts){return;}
rlm@15:   alcMakeContextCurrent(data->contexts[contextNum]->ctx);
rlm@15:   alListener3f(param, v1, v2, v3);
rlm@15:   alcMakeContextCurrent(current);
rlm@15: }
rlm@15: 
rlm@15: /*
rlm@15:  * Class:     com_aurellem_send_AudioSend
rlm@15:  * Method:    nsetNthListenerf
rlm@15:  * Signature: (IFJI)V
rlm@15:  */
rlm@15: JNIEXPORT void JNICALL Java_com_aurellem_send_AudioSend_nsetNthListenerf
rlm@15: (JNIEnv *env, jclass clazz, jint param, jfloat v1, jlong device, 
rlm@15:  jint contextNum){
rlm@15: 
rlm@15:   UNUSED(env);UNUSED(clazz);
rlm@15:   
rlm@15:   ALCdevice *Device = (ALCdevice*) ((intptr_t)device);
rlm@15:   send_data *data = (send_data*)Device->ExtraData;
rlm@15:   
rlm@15:   ALCcontext *current = alcGetCurrentContext();
rlm@15:   if ((ALuint)contextNum > data->numContexts){return;}
rlm@15:   alcMakeContextCurrent(data->contexts[contextNum]->ctx);
rlm@15:   alListenerf(param, v1);
rlm@15:   alcMakeContextCurrent(current);
rlm@15: }
rlm@19: #+end_src
rlm@15: 
rlm@20: *** Initialization
rlm@19: =initDevice= is called from the Java side after LWJGL has created its
rlm@19: context, and before any calls to =addListener=. It establishes the
rlm@19: LWJGL context as the master context.
rlm@19: 
rlm@20: =getAudioFormat= is a convenience function that uses JNI to build up a
rlm@19: =javax.sound.sampled.AudioFormat= object from data in the Device. This
rlm@19: way, there is no ambiguity about what the bits created by =step= and
rlm@19: returned by =getSamples= mean.
rlm@21: #+srcname: jni-init
rlm@19: #+begin_src C
rlm@15: /*
rlm@15:  * Class:     com_aurellem_send_AudioSend
rlm@15:  * Method:    ninitDevice
rlm@15:  * Signature: (J)V
rlm@15:  */                     
rlm@15: JNIEXPORT void JNICALL Java_com_aurellem_send_AudioSend_ninitDevice
rlm@15: (JNIEnv *env, jclass clazz, jlong device){
rlm@15:   UNUSED(env);UNUSED(clazz);
rlm@15:   ALCdevice *Device = (ALCdevice*) ((intptr_t)device);
rlm@15:   init(Device);
rlm@15: }
rlm@15: 
rlm@15: /*
rlm@15:  * Class:     com_aurellem_send_AudioSend
rlm@15:  * Method:    ngetAudioFormat
rlm@15:  * Signature: (J)Ljavax/sound/sampled/AudioFormat;
rlm@15:  */
rlm@15: JNIEXPORT jobject JNICALL Java_com_aurellem_send_AudioSend_ngetAudioFormat
rlm@15: (JNIEnv *env, jclass clazz, jlong device){
rlm@15:   UNUSED(clazz);
rlm@15:   jclass AudioFormatClass = 
rlm@15:     (*env)->FindClass(env, "javax/sound/sampled/AudioFormat");
rlm@15:   jmethodID AudioFormatConstructor = 
rlm@15:     (*env)->GetMethodID(env, AudioFormatClass, "<init>", "(FIIZZ)V");
rlm@15:   
rlm@15:   ALCdevice *Device = (ALCdevice*) ((intptr_t)device);
rlm@15:   int isSigned;
rlm@15:   switch (Device->FmtType)
rlm@15:     {
rlm@15:     case DevFmtUByte: 
rlm@15:     case DevFmtUShort: isSigned = 0; break;  
rlm@15:     default : isSigned = 1;   
rlm@15:     }
rlm@15:   float frequency = Device->Frequency;
rlm@15:   int bitsPerFrame = (8 * BytesFromDevFmt(Device->FmtType));
rlm@15:   int channels = Device->NumChan;
rlm@15:   jobject format = (*env)->
rlm@15:     NewObject(
rlm@15: 	      env,AudioFormatClass,AudioFormatConstructor,
rlm@15: 	      frequency,
rlm@15: 	      bitsPerFrame,
rlm@15: 	      channels,
rlm@15: 	      isSigned,
rlm@15: 	      0);
rlm@15:   return format;
rlm@15: }
rlm@19: #+end_src
rlm@15: 
rlm@22: ** Boring Device management stuff
rlm@19: This code is more-or-less copied verbatim from the other =OpenAL=
rlm@19: backends. It's the basis for =OpenAL='s primitive object system.
rlm@21: #+srcname: device-init
rlm@19: #+begin_src C
rlm@20: ////////////////////   Device Initialization / Management
rlm@15: 
rlm@15: static const ALCchar sendDevice[] = "Multiple Audio Send";
rlm@15: 
rlm@15: static ALCboolean send_open_playback(ALCdevice *device, 
rlm@15:  const ALCchar *deviceName)
rlm@15: {
rlm@15:   send_data *data;
rlm@15:   // stop any buffering for stdout, so that I can 
rlm@20:   // see the printf statements in my terminal immediately
rlm@15:   setbuf(stdout, NULL);
rlm@15: 
rlm@15:   if(!deviceName)
rlm@15:     deviceName = sendDevice;
rlm@15:   else if(strcmp(deviceName, sendDevice) != 0)
rlm@15:     return ALC_FALSE;
rlm@15:   data = (send_data*)calloc(1, sizeof(*data));
rlm@15:   device->szDeviceName = strdup(deviceName);
rlm@15:   device->ExtraData = data;
rlm@15:   return ALC_TRUE;
rlm@15: }
rlm@15: 
rlm@15: static void send_close_playback(ALCdevice *device)
rlm@15: {
rlm@15:   send_data *data = (send_data*)device->ExtraData;
rlm@15:   alcMakeContextCurrent(NULL);
rlm@15:   ALuint i;
rlm@15:   // Destroy all slave contexts. LWJGL will take care of 
rlm@15:   // its own context.
rlm@15:   for (i = 1; i < data->numContexts; i++){
rlm@15:     context_data *ctxData = data->contexts[i];
rlm@15:     alcDestroyContext(ctxData->ctx);
rlm@15:     free(ctxData->renderBuffer);
rlm@15:     free(ctxData);
rlm@15:   }
rlm@15:   free(data);
rlm@15:   device->ExtraData = NULL;
rlm@15: }
rlm@15: 
rlm@15: static ALCboolean send_reset_playback(ALCdevice *device)
rlm@15: {
rlm@15:   SetDefaultWFXChannelOrder(device);
rlm@15:   return ALC_TRUE;
rlm@15: }
rlm@15: 
rlm@15: static void send_stop_playback(ALCdevice *Device){
rlm@15:   UNUSED(Device);
rlm@15: }
rlm@15: 
rlm@15: static const BackendFuncs send_funcs = {
rlm@15:   send_open_playback,
rlm@15:   send_close_playback,
rlm@15:   send_reset_playback,
rlm@15:   send_stop_playback,
rlm@15:   NULL,
rlm@15:   NULL,  /* These would be filled with functions to    */
rlm@15:   NULL,  /* handle capturing audio if we we into that  */
rlm@15:   NULL,  /* sort of thing...                           */
rlm@15:   NULL,
rlm@15:   NULL
rlm@15: };
rlm@15: 
rlm@15: ALCboolean alc_send_init(BackendFuncs *func_list){
rlm@15:   *func_list = send_funcs;
rlm@15:   return ALC_TRUE;
rlm@15: }
rlm@15: 
rlm@15: void alc_send_deinit(void){}
rlm@15: 
rlm@15: void alc_send_probe(enum DevProbe type)
rlm@15: {
rlm@15:   switch(type)
rlm@15:     {
rlm@15:     case DEVICE_PROBE:
rlm@15:       AppendDeviceList(sendDevice);
rlm@15:       break;
rlm@15:     case ALL_DEVICE_PROBE:
rlm@15:       AppendAllDeviceList(sendDevice);
rlm@15:       break;
rlm@15:     case CAPTURE_DEVICE_PROBE:
rlm@15:       break;
rlm@15:     }
rlm@15: }
rlm@15: #+end_src
rlm@15: 
rlm@19: * The Java interface, =AudioSend=
rlm@15: 
rlm@19: The Java interface to the Send Device follows naturally from the JNI
rlm@19: definitions. It is included here for completeness. The only thing here
rlm@19: of note is the =deviceID=. This is available from LWJGL, but to only
rlm@20: way to get it is reflection. Unfortunately, there is no other way to
rlm@19: control the Send device than to obtain a pointer to it.
rlm@15: 
rlm@19: #+include: "../java/src/com/aurellem/send/AudioSend.java" src java :exports code
rlm@15: 
rlm@19: * Finally, Ears in clojure! 
rlm@15: 
rlm@20: Now that the infrastructure is complete (modulo a few patches to
rlm@19: jMonkeyEngine3 to support accessing this modified version of =OpenAL=
rlm@19: that are not worth discussing), the clojure ear abstraction is rather
rlm@19: simple.  Just as there were =SceneProcessors= for vision, there are
rlm@19: now =SoundProcessors= for hearing.
rlm@15: 
rlm@19: #+include "../../jmeCapture/src/com/aurellem/capture/audio/SoundProcessor.java" src java 
rlm@15: 
rlm@15: #+srcname: ears
rlm@0: #+begin_src clojure
rlm@19: (ns cortex.hearing
rlm@19:   "Simulate the sense of hearing in jMonkeyEngine3. Enables multiple
rlm@19:   listeners at different positions in the same world. Passes vectors
rlm@20:   of floats in the range [-1.0 -- 1.0] in PCM format to any arbitrary
rlm@19:   function."
rlm@19:   {:author "Robert McIntyre"}
rlm@19:   (:use (cortex world util))
rlm@19:   (:import java.nio.ByteBuffer)
rlm@19:   (:import org.tritonus.share.sampled.FloatSampleTools)
rlm@19:   (:import com.aurellem.capture.audio.SoundProcessor)
rlm@19:   (:import javax.sound.sampled.AudioFormat))
rlm@19:   
rlm@0: (defn sound-processor
rlm@19:   "Deals with converting ByteBuffers into Vectors of floats so that
rlm@19:   the continuation functions can be defined in terms of immutable
rlm@19:   stuff."
rlm@0:   [continuation]
rlm@0:   (proxy [SoundProcessor] []
rlm@0:     (cleanup [])
rlm@0:     (process
rlm@19:       [#^ByteBuffer audioSamples numSamples #^AudioFormat audioFormat]
rlm@19:       (let [bytes  (byte-array numSamples)
rlm@19:             floats (float-array numSamples)]
rlm@19:         (.get audioSamples bytes 0 numSamples)
rlm@19:         (FloatSampleTools/byte2floatInterleaved
rlm@19:          bytes 0 floats 0
rlm@19:          (/ numSamples (.getFrameSize audioFormat)) audioFormat)
rlm@19:         (continuation
rlm@19:          (vec floats))))))
rlm@0: 
rlm@0: (defn add-ear 
rlm@19:   "Add an ear to the world.  The continuation function will be called
rlm@0:   on the FFT or the sounds which the ear hears in the given
rlm@0:   timeframe. Sound is 3D."
rlm@0:   [world listener continuation]
rlm@0:   (let [renderer (.getAudioRenderer world)]
rlm@0:     (.addListener renderer listener)
rlm@0:     (.registerSoundProcessor renderer listener
rlm@0: 			     (sound-processor continuation))
rlm@0:     listener))
rlm@0: #+end_src
rlm@0: 
rlm@19: * Example
rlm@0: 
rlm@0: #+srcname: test-hearing
rlm@0: #+begin_src clojure :results silent
rlm@19: (ns test.hearing
rlm@19:   (:use (cortex world util hearing))
rlm@19:   (:import (com.jme3.audio AudioNode Listener))
rlm@19:   (:import com.jme3.scene.Node))
rlm@0: 
rlm@0: (defn setup-fn [world]
rlm@0:   (let [listener (Listener.)]
rlm@19:     (add-ear world listener #(println-repl (nth % 0)))))
rlm@0:   
rlm@0: (defn play-sound [node world value]
rlm@0:   (if (not value)
rlm@0:     (do
rlm@0:       (.playSource (.getAudioRenderer world) node))))
rlm@0: 
rlm@19: (defn test-basic-hearing []
rlm@19:   (.start
rlm@19:    (let [node1 (AudioNode. (asset-manager) "Sounds/pure.wav" false false)]
rlm@19:      (world
rlm@19:       (Node.)
rlm@19:       {"key-space" (partial play-sound node1)}
rlm@19:       setup-fn
rlm@19:       no-op))))  
rlm@0: #+end_src
rlm@0: 
rlm@19: This extremely basic program prints out the first sample it encounters
rlm@19: at every time stamp. You can see the rendered sound begin printed at
rlm@19: the REPL.
rlm@15: 
rlm@22:  - As a bonus, this method of capturing audio for AI can also be used
rlm@22:    to capture perfect audio from a jMonkeyEngine application, for use
rlm@22:    in demos and the like.
rlm@22: 
rlm@22: 
rlm@0: * COMMENT Code Generation
rlm@0: 
rlm@15: #+begin_src clojure :tangle ../../cortex/src/cortex/hearing.clj
rlm@15: <<ears>>
rlm@0: #+end_src
rlm@0: 
rlm@15: #+begin_src clojure :tangle ../../cortex/src/test/hearing.clj
rlm@0: <<test-hearing>>
rlm@0: #+end_src
rlm@0: 
rlm@0: 
rlm@15: #+begin_src C :tangle ../Alc/backends/send.c
rlm@21: <<send-header>>
rlm@21: <<send-state>>
rlm@21: <<sync-macros>>
rlm@21: <<sync-sources>>
rlm@21: <<sync-contexts>>
rlm@21: <<context-creation>>
rlm@21: <<context-switching>>
rlm@21: <<main-loop>>
rlm@21: <<jni-step>>
rlm@21: <<jni-get-samples>>
rlm@21: <<listener-manage>>
rlm@21: <<jni-init>>
rlm@21: <<device-init>>
rlm@15: #+end_src
rlm@19: 
rlm@19: