mirror of
https://github.com/ioquake/jedi-academy.git
synced 2024-11-23 04:33:12 +00:00
140 lines
2.7 KiB
C
140 lines
2.7 KiB
C
|
|
// mac_snddma.c
|
|
// all other sound mixing is portable
|
|
|
|
#include "../client/snd_local.h"
|
|
#include <sound.h>
|
|
|
|
#define MAX_MIXED_SAMPLES 0x8000
|
|
#define SUBMISSION_CHUNK 0x100
|
|
|
|
static short s_mixedSamples[MAX_MIXED_SAMPLES];
|
|
static int s_chunkCount; // number of chunks submitted
|
|
static SndChannel *s_sndChan;
|
|
static ExtSoundHeader s_sndHeader;
|
|
|
|
/*
|
|
===============
|
|
S_Callback
|
|
===============
|
|
*/
|
|
void S_Callback( SndChannel *sc, SndCommand *cmd ) {
|
|
SndCommand mySndCmd;
|
|
SndCommand mySndCmd2;
|
|
int offset;
|
|
|
|
offset = ( s_chunkCount * SUBMISSION_CHUNK ) & (MAX_MIXED_SAMPLES-1);
|
|
|
|
// queue up another sound buffer
|
|
memset( &s_sndHeader, 0, sizeof( s_sndHeader ) );
|
|
s_sndHeader.samplePtr = (void *)(s_mixedSamples + offset);
|
|
s_sndHeader.numChannels = 2;
|
|
s_sndHeader.sampleRate = rate22khz;
|
|
s_sndHeader.loopStart = 0;
|
|
s_sndHeader.loopEnd = 0;
|
|
s_sndHeader.encode = extSH;
|
|
s_sndHeader.baseFrequency = 1;
|
|
s_sndHeader.numFrames = SUBMISSION_CHUNK / 2;
|
|
s_sndHeader.markerChunk = NULL;
|
|
s_sndHeader.instrumentChunks = NULL;
|
|
s_sndHeader.AESRecording = NULL;
|
|
s_sndHeader.sampleSize = 16;
|
|
|
|
mySndCmd.cmd = bufferCmd;
|
|
mySndCmd.param1 = 0;
|
|
mySndCmd.param2 = (int)&s_sndHeader;
|
|
SndDoCommand( sc, &mySndCmd, true );
|
|
|
|
// and another callback
|
|
mySndCmd2.cmd = callBackCmd;
|
|
mySndCmd2.param1 = 0;
|
|
mySndCmd2.param2 = 0;
|
|
SndDoCommand( sc, &mySndCmd2, true );
|
|
|
|
s_chunkCount++; // this is the next buffer we will submit
|
|
}
|
|
|
|
/*
|
|
===============
|
|
S_MakeTestPattern
|
|
===============
|
|
*/
|
|
void S_MakeTestPattern( void ) {
|
|
int i;
|
|
float v;
|
|
int sample;
|
|
|
|
for ( i = 0 ; i < dma.samples / 2 ; i ++ ) {
|
|
v = sin( M_PI * 2 * i / 64 );
|
|
sample = v * 0x4000;
|
|
((short *)dma.buffer)[i*2] = sample;
|
|
((short *)dma.buffer)[i*2+1] = sample;
|
|
}
|
|
}
|
|
|
|
/*
|
|
===============
|
|
SNDDMA_Init
|
|
===============
|
|
*/
|
|
qboolean SNDDMA_Init(void) {
|
|
int err;
|
|
|
|
// create a sound channel
|
|
s_sndChan = NULL;
|
|
err = SndNewChannel( &s_sndChan, sampledSynth, initStereo, NewSndCallBackProc(S_Callback) );
|
|
if ( err ) {
|
|
return false;
|
|
}
|
|
|
|
dma.channels = 2;
|
|
dma.samples = MAX_MIXED_SAMPLES;
|
|
dma.submission_chunk = SUBMISSION_CHUNK;
|
|
dma.samplebits = 16;
|
|
dma.speed = 22050;
|
|
dma.buffer = (byte *)s_mixedSamples;
|
|
|
|
// que up the first submission-chunk sized buffer
|
|
s_chunkCount = 0;
|
|
|
|
S_Callback( s_sndChan, NULL );
|
|
|
|
return qtrue;
|
|
}
|
|
|
|
/*
|
|
===============
|
|
SNDDMA_GetDMAPos
|
|
===============
|
|
*/
|
|
int SNDDMA_GetDMAPos(void) {
|
|
return s_chunkCount * SUBMISSION_CHUNK;
|
|
}
|
|
|
|
/*
|
|
===============
|
|
SNDDMA_Shutdown
|
|
===============
|
|
*/
|
|
void SNDDMA_Shutdown(void) {
|
|
if ( s_sndChan ) {
|
|
SndDisposeChannel( s_sndChan, true );
|
|
s_sndChan = NULL;
|
|
}
|
|
}
|
|
|
|
/*
|
|
===============
|
|
SNDDMA_BeginPainting
|
|
===============
|
|
*/
|
|
void SNDDMA_BeginPainting(void) {
|
|
}
|
|
|
|
/*
|
|
===============
|
|
SNDDMA_Submit
|
|
===============
|
|
*/
|
|
void SNDDMA_Submit(void) {
|
|
}
|