summaryrefslogtreecommitdiff
path: root/apps/codecs
diff options
context:
space:
mode:
authorMichael Sevakis <jethead71@rockbox.org>2007-11-18 19:03:45 +0000
committerMichael Sevakis <jethead71@rockbox.org>2007-11-18 19:03:45 +0000
commit1a41c8afeefd4884a5369c79808c68b3514eec8b (patch)
tree9200eef9f09c86fd19929733ab45f1a3acee6152 /apps/codecs
parent984a6b023684de78e41dae320721b4b28f17f108 (diff)
SPC Codec: Run SPC emulation on COP and audio sample processing on CPU on dual-core PortalPlayer targets.
git-svn-id: svn://svn.rockbox.org/rockbox/trunk@15673 a1c6a512-1295-4272-9138-f99709370657
Diffstat (limited to 'apps/codecs')
-rw-r--r--apps/codecs/spc.c309
-rw-r--r--apps/codecs/spc/spc_codec.h100
-rw-r--r--apps/codecs/spc/spc_dsp.c232
3 files changed, 518 insertions, 123 deletions
diff --git a/apps/codecs/spc.c b/apps/codecs/spc.c
index f2890cd4a4..ae3026354d 100644
--- a/apps/codecs/spc.c
+++ b/apps/codecs/spc.c
@@ -185,12 +185,253 @@ static int LoadID666(unsigned char *buf) {
}
/**************** Codec ****************/
+enum {SAMPLE_RATE = 32000};
+static struct Spc_Emu spc_emu IDATA_ATTR CACHEALIGN_ATTR;
-static int32_t samples[WAV_CHUNK_SIZE*2] IBSS_ATTR;
+#if SPC_DUAL_CORE
+/** Implementations for pipelined dual-core operation **/
+static int spc_emu_thread_stack[DEFAULT_STACK_SIZE/sizeof(int)]
+ CACHEALIGN_ATTR;
-static struct Spc_Emu spc_emu IDATA_ATTR;
+static const unsigned char * const spc_emu_thread_name = "spc emu";
+static struct thread_entry *emu_thread_p;
-enum {SAMPLE_RATE = 32000};
+enum
+{
+ SPC_EMU_AUDIO = 0,
+ SPC_EMU_LOAD,
+ SPC_EMU_QUIT,
+};
+
+struct spc_load
+{
+ uint8_t *buf;
+ size_t size;
+};
+
+/* sample queue */
+#define WAV_NUM_CHUNKS 2
+#define WAV_CHUNK_MASK (WAV_NUM_CHUNKS-1)
+struct sample_queue_chunk
+{
+ long id;
+ union
+ {
+ intptr_t data;
+ int32_t audio[WAV_CHUNK_SIZE*2];
+ };
+};
+
+static struct
+{
+ int head, tail;
+ struct semaphore emu_sem_head;
+ struct semaphore emu_sem_tail;
+ struct event emu_evt_reply;
+ intptr_t retval;
+ struct sample_queue_chunk wav_chunk[WAV_NUM_CHUNKS];
+} sample_queue NOCACHEBSS_ATTR;
+
+static inline void samples_release_wrbuf(void)
+{
+ sample_queue.tail++;
+ ci->semaphore_release(&sample_queue.emu_sem_head);
+}
+
+static inline struct sample_queue_chunk * samples_get_wrbuf(void)
+{
+ ci->semaphore_wait(&sample_queue.emu_sem_tail);
+ return &sample_queue.wav_chunk[sample_queue.tail & WAV_CHUNK_MASK];
+}
+
+static inline void samples_release_rdbuf(void)
+{
+ if (sample_queue.head != sample_queue.tail) {
+ sample_queue.head++;
+ }
+
+ ci->semaphore_release(&sample_queue.emu_sem_tail);
+}
+
+static inline int32_t * samples_get_rdbuf(void)
+{
+ ci->semaphore_wait(&sample_queue.emu_sem_head);
+
+ if (ci->stop_codec || ci->new_track)
+ {
+ /* Told to stop. Buffer must be released. */
+ samples_release_rdbuf();
+ return NULL;
+ }
+
+ return sample_queue.wav_chunk[sample_queue.head & WAV_CHUNK_MASK].audio;
+}
+
+static intptr_t emu_thread_send_msg(long id, intptr_t data)
+{
+ struct sample_queue_chunk *chunk;
+ /* Grab an audio output buffer */
+ ci->semaphore_wait(&sample_queue.emu_sem_head);
+ chunk = &sample_queue.wav_chunk[sample_queue.head & WAV_CHUNK_MASK];
+ /* Place a message in it instead of audio */
+ chunk->id = id;
+ chunk->data = data;
+ /* Release it to the emu thread */
+ samples_release_rdbuf();
+ /* Wait for a response */
+ ci->event_wait(&sample_queue.emu_evt_reply, STATE_SIGNALED);
+ return sample_queue.retval;
+}
+
+/* thread function */
+static bool emu_thread_process_msg(struct sample_queue_chunk *chunk)
+{
+ long id = chunk->id;
+ bool ret = id != SPC_EMU_QUIT;
+
+ chunk->id = SPC_EMU_AUDIO; /* Reset chunk type to audio */
+ sample_queue.retval = 0;
+
+ if (id == SPC_EMU_LOAD)
+ {
+ struct spc_load *ld = (struct spc_load *)chunk->data;
+ invalidate_icache();
+ SPC_Init(&spc_emu);
+ sample_queue.retval = SPC_load_spc(&spc_emu, ld->buf, ld->size);
+ }
+
+ /* Empty the audio queue */
+ /* This is a dirty hack a timeout based wait would make unnescessary but
+ still safe because the other thread is known to be waiting for a reply
+ and is not using the objects. */
+ ci->semaphore_init(&sample_queue.emu_sem_tail, 2, 2);
+ ci->semaphore_init(&sample_queue.emu_sem_head, 2, 0);
+ sample_queue.head = sample_queue.tail = 0;
+ ci->event_set_state(&sample_queue.emu_evt_reply, STATE_SIGNALED);
+
+ return ret;
+}
+
+static void spc_emu_thread(void)
+{
+ CPU_Init(&spc_emu);
+
+ while (1) {
+ /* get a buffer for output */
+ struct sample_queue_chunk *chunk = samples_get_wrbuf();
+
+ if (chunk->id != SPC_EMU_AUDIO) {
+ /* This chunk doesn't contain audio but a command */
+ if (!emu_thread_process_msg(chunk))
+ break;
+ /* Have to re-get this pointer to keep semaphore counts correct */
+ continue;
+ }
+
+ ENTER_TIMER(render);
+ /* fill samples buffer */
+ if ( SPC_play(&spc_emu, WAV_CHUNK_SIZE*2, chunk->audio) )
+ assert( false );
+ EXIT_TIMER(render);
+
+ /* done so release it to output */
+ samples_release_wrbuf();
+ ci->yield();
+ }
+}
+
+static bool spc_emu_start(void)
+{
+ emu_thread_p = ci->create_thread(spc_emu_thread, spc_emu_thread_stack,
+ sizeof(spc_emu_thread_stack), CREATE_THREAD_FROZEN,
+ spc_emu_thread_name IF_PRIO(, PRIORITY_PLAYBACK), COP);
+
+ if (emu_thread_p == NULL)
+ return false;
+
+ /* Initialize audio queue as full to prevent emu thread from trying to run the
+ emulator before loading something */
+ ci->event_init(&sample_queue.emu_evt_reply,
+ EVENT_AUTOMATIC | STATE_NONSIGNALED);
+ ci->semaphore_init(&sample_queue.emu_sem_tail, 2, 0);
+ ci->semaphore_init(&sample_queue.emu_sem_head, 2, 2);
+ sample_queue.head = 0;
+ sample_queue.tail = 2;
+
+ /* Start it running */
+ ci->thread_thaw(emu_thread_p);
+ return true;
+}
+
+/* load a new program on the emu thread */
+static inline int load_spc_buffer(uint8_t *buf, size_t size)
+{
+ struct spc_load ld = { buf, size };
+ flush_icache();
+ return emu_thread_send_msg(SPC_EMU_LOAD, (intptr_t)&ld);
+}
+
+static inline void spc_emu_quit(void)
+{
+ emu_thread_send_msg(SPC_EMU_QUIT, 0);
+ /* Wait for emu thread to be killed */
+ ci->thread_wait(emu_thread_p);
+}
+
+static inline bool spc_play_get_samples(int32_t **samples)
+{
+ /* obtain filled samples buffer */
+ *samples = samples_get_rdbuf();
+ return *samples != NULL;
+}
+
+static inline void spc_play_send_samples(int32_t *samples)
+{
+ ci->pcmbuf_insert(samples, samples+WAV_CHUNK_SIZE, WAV_CHUNK_SIZE);
+ /* done with chunk so release it to emu thread */
+ samples_release_rdbuf();
+}
+
+#else /* !SPC_DUAL_CORE */
+/** Implementations for single-core operation **/
+int32_t wav_chunk[WAV_CHUNK_SIZE*2] IBSS_ATTR;
+
+/* load a new program into emu */
+static inline int load_spc_buffer(uint8_t *buf, size_t size)
+{
+ SPC_Init(&spc_emu);
+ return SPC_load_spc(&spc_emu, buf, size);
+}
+
+static inline bool spc_emu_start(void)
+{
+#ifdef CPU_COLDFIRE
+ /* signed integer mode with saturation */
+ coldfire_set_macsr(EMAC_SATURATE);
+#endif
+ CPU_Init(&spc_emu);
+ return true;
+}
+
+static inline void spc_play_send_samples(int32_t *samples)
+{
+ ci->pcmbuf_insert(samples, samples+WAV_CHUNK_SIZE, WAV_CHUNK_SIZE);
+}
+
+#define spc_emu_quit()
+#define samples_release_rdbuf()
+
+static inline bool spc_play_get_samples(int32_t **samples)
+{
+ ENTER_TIMER(render);
+ /* fill samples buffer */
+ if ( SPC_play(&spc_emu,WAV_CHUNK_SIZE*2,wav_chunk) )
+ assert( false );
+ EXIT_TIMER(render);
+ *samples = wav_chunk;
+ return true;
+}
+#endif /* SPC_DUAL_CORE */
/* The main decoder loop */
static int play_track( void )
@@ -206,7 +447,7 @@ static int play_track( void )
fadedec=0x7fffffffl/(fadeendsample-fadestartsample)+1;
ENTER_TIMER(total);
-
+
while ( 1 )
{
ci->yield();
@@ -224,14 +465,12 @@ static int play_track( void )
}
ci->seek_complete();
}
-
- ENTER_TIMER(render);
- /* fill samples buffer */
- if ( SPC_play(&spc_emu,WAV_CHUNK_SIZE*2,samples) )
- assert( false );
- EXIT_TIMER(render);
-
- sampleswritten+=WAV_CHUNK_SIZE;
+
+ int32_t *samples;
+ if (!spc_play_get_samples(&samples))
+ break;
+
+ sampleswritten += WAV_CHUNK_SIZE;
/* is track timed? */
if (ci->global_settings->repeat_mode!=REPEAT_ONE && ci->id3->length) {
@@ -241,11 +480,11 @@ static int play_track( void )
/* fade? */
if (curtime>ID666.length)
{
-#ifdef CPU_COLDFIRE
+ #ifdef CPU_COLDFIRE
/* Have to switch modes to do this */
long macsr = coldfire_get_macsr();
coldfire_set_macsr(EMAC_SATURATE | EMAC_FRACTIONAL | EMAC_ROUND);
-#endif
+ #endif
int i;
for (i=0;i<WAV_CHUNK_SIZE;i++) {
if (lasttimesample+i>fadestartsample) {
@@ -256,42 +495,43 @@ static int play_track( void )
fadevol-=fadedec;
}
}
-#ifdef CPU_COLDFIRE
+ #ifdef CPU_COLDFIRE
coldfire_set_macsr(macsr);
-#endif
+ #endif
}
/* end? */
if (lasttimesample>=fadeendsample)
+ {
+ samples_release_rdbuf();
break;
+ }
}
- ci->pcmbuf_insert(samples, samples+WAV_CHUNK_SIZE, WAV_CHUNK_SIZE);
+ spc_play_send_samples(samples);
if (ci->global_settings->repeat_mode!=REPEAT_ONE)
- ci->set_elapsed(sampleswritten*1000LL/SAMPLE_RATE);
+ ci->set_elapsed(sampleswritten*1000LL/SAMPLE_RATE);
else
ci->set_elapsed(0);
}
EXIT_TIMER(total);
-
return 0;
}
/* this is the codec entry point */
enum codec_status codec_main(void)
{
-#ifdef CPU_COLDFIRE
- /* signed integer mode with saturation */
- coldfire_set_macsr(EMAC_SATURATE);
-#endif
- CPU_Init(&spc_emu);
+ enum codec_status stat = CODEC_ERROR;
+
+ if (!spc_emu_start())
+ goto codec_quit;
do
{
DEBUGF("SPC: next_track\n");
if (codec_init()) {
- return CODEC_ERROR;
+ goto codec_quit;
}
DEBUGF("SPC: after init\n");
@@ -301,7 +541,7 @@ enum codec_status codec_main(void)
/* wait for track info to load */
while (!*ci->taginfo_ready && !ci->stop_codec)
- ci->sleep(1);
+ ci->yield();
codec_set_replaygain(ci->id3);
@@ -313,20 +553,19 @@ enum codec_status codec_main(void)
size_t buffersize;
uint8_t* buffer = ci->request_buffer(&buffersize, ci->filesize);
if (!buffer) {
- return CODEC_ERROR;
+ goto codec_quit;
}
DEBUGF("SPC: read size = 0x%lx\n",(unsigned long)buffersize);
do
{
- SPC_Init(&spc_emu);
- if (SPC_load_spc(&spc_emu,buffer,buffersize)) {
+ if (load_spc_buffer(buffer, buffersize)) {
DEBUGF("SPC load failure\n");
- return CODEC_ERROR;
+ goto codec_quit;
}
LoadID666(buffer+0x2e);
-
+
if (ci->global_settings->repeat_mode!=REPEAT_ONE && ID666.length==0) {
ID666.length=3*60*1000; /* 3 minutes */
ID666.fade=5*1000; /* 5 seconds */
@@ -340,12 +579,16 @@ enum codec_status codec_main(void)
reset_profile_timers();
}
-
while ( play_track() );
print_timers(ci->id3->path);
}
while ( ci->request_next_track() );
+
+ stat = CODEC_OK;
+
+codec_quit:
+ spc_emu_quit();
- return CODEC_OK;
+ return stat;
}
diff --git a/apps/codecs/spc/spc_codec.h b/apps/codecs/spc/spc_codec.h
index f2677df04a..c785acc468 100644
--- a/apps/codecs/spc/spc_codec.h
+++ b/apps/codecs/spc/spc_codec.h
@@ -32,38 +32,51 @@
/** Basic configuration options **/
-/* TGB is the only target fast enough for gaussian and realtime BRR decode */
-/* echo is almost fast enough but not quite */
-#ifndef TOSHIBA_GIGABEAT_F
- /* Cache BRR waves */
- #define SPC_BRRCACHE 1
+#define SPC_DUAL_CORE 1
- /* Disable gaussian interpolation */
- #define SPC_NOINTERP 1
-
-#ifndef CPU_COLDFIRE
- /* Disable echo processing */
- #define SPC_NOECHO 1
-#else
- /* Enable echo processing */
- #define SPC_NOECHO 0
+#if !defined(SPC_DUAL_CORE) || NUM_CORES == 1
+#undef SPC_DUAL_CORE
+#define SPC_DUAL_CORE 0
#endif
-#else
+
+/* TGB is the only target fast enough for gaussian and realtime BRR decode */
+/* echo is almost fast enough but not quite */
+#if defined(TOSHIBA_GIGABEAT_F) || defined(SIMULATOR)
/* Don't cache BRR waves */
#define SPC_BRRCACHE 0
/* Allow gaussian interpolation */
#define SPC_NOINTERP 0
+
+ /* Allow echo processing */
+ #define SPC_NOECHO 0
+#elif defined(CPU_COLDFIRE)
+ /* Cache BRR waves */
+ #define SPC_BRRCACHE 1
+ /* Disable gaussian interpolation */
+ #define SPC_NOINTERP 1
+
/* Allow echo processing */
#define SPC_NOECHO 0
-#endif
+#elif defined (CPU_PP) && SPC_DUAL_CORE
+ /* Cache BRR waves */
+ #define SPC_BRRCACHE 1
+
+ /* Disable gaussian interpolation */
+ #define SPC_NOINTERP 1
-/* Samples per channel per iteration */
-#ifdef CPU_COLDFIRE
-#define WAV_CHUNK_SIZE 1024
+ /* Allow echo processing */
+ #define SPC_NOECHO 0
#else
-#define WAV_CHUNK_SIZE 2048
+ /* Cache BRR waves */
+ #define SPC_BRRCACHE 1
+
+ /* Disable gaussian interpolation */
+ #define SPC_NOINTERP 1
+
+ /* Disable echo processing */
+ #define SPC_NOECHO 1
#endif
#ifdef CPU_ARM
@@ -72,6 +85,26 @@
#undef IDATA_ATTR
#define IDATA_ATTR
+
+ #undef ICONST_ATTR
+ #define ICONST_ATTR
+
+ #undef IBSS_ATTR
+ #define IBSS_ATTR
+
+#if SPC_DUAL_CORE
+ #undef NOCACHEBSS_ATTR
+ #define NOCACHEBSS_ATTR __attribute__ ((section(".ibss")))
+ #undef NOCACHEDATA_ATTR
+ #define NOCACHEDATA_ATTR __attribute__((section(".idata")))
+#endif
+#endif
+
+/* Samples per channel per iteration */
+#if defined(CPU_PP) && NUM_CORES == 1
+#define WAV_CHUNK_SIZE 2048
+#else
+#define WAV_CHUNK_SIZE 1024
#endif
/**************** Little-endian handling ****************/
@@ -231,16 +264,26 @@ extern int16_t BRRcache [BRR_CACHE_SIZE];
enum { FIR_BUF_HALF = 8 };
-#ifdef CPU_COLDFIRE
+#if defined(CPU_COLDFIRE)
/* global because of the large aligment requirement for hardware masking -
* L-R interleaved 16-bit samples for easy loading and mac.w use.
*/
enum
{
- FIR_BUF_SIZE = FIR_BUF_HALF * sizeof ( int32_t ),
- FIR_BUF_MASK = ~FIR_BUF_SIZE
+ FIR_BUF_CNT = FIR_BUF_HALF,
+ FIR_BUF_SIZE = FIR_BUF_CNT * sizeof ( int32_t ),
+ FIR_BUF_ALIGN = FIR_BUF_SIZE * 2,
+ FIR_BUF_MASK = ~((FIR_BUF_ALIGN / 2) | (sizeof ( int32_t ) - 1))
+};
+#elif defined (CPU_ARM)
+enum
+{
+ FIR_BUF_CNT = FIR_BUF_HALF * 2 * 2,
+ FIR_BUF_SIZE = FIR_BUF_CNT * sizeof ( int32_t ),
+ FIR_BUF_ALIGN = FIR_BUF_SIZE,
+ FIR_BUF_MASK = ~((FIR_BUF_ALIGN / 2) | (sizeof ( int32_t ) * 2 - 1))
};
-#endif /* CPU_COLDFIRE */
+#endif /* CPU_* */
struct Spc_Dsp
{
@@ -257,14 +300,19 @@ struct Spc_Dsp
int noise_count;
uint16_t noise; /* also read as int16_t */
-#ifdef CPU_COLDFIRE
+#if defined(CPU_COLDFIRE)
/* circularly hardware masked address */
int32_t *fir_ptr;
/* wrapped address just behind current position -
allows mac.w to increment and mask fir_ptr */
int32_t *last_fir_ptr;
/* copy of echo FIR constants as int16_t for use with mac.w */
- int16_t fir_coeff[VOICE_COUNT];
+ int16_t fir_coeff [VOICE_COUNT];
+#elif defined (CPU_ARM)
+ /* fir_buf [i + 8] == fir_buf [i], to avoid wrap checking in FIR code */
+ int32_t *fir_ptr;
+ /* copy of echo FIR constants as int32_t, for faster access */
+ int32_t fir_coeff [VOICE_COUNT];
#else
/* fir_buf [i + 8] == fir_buf [i], to avoid wrap checking in FIR code */
int fir_pos; /* (0 to 7) */
diff --git a/apps/codecs/spc/spc_dsp.c b/apps/codecs/spc/spc_dsp.c
index 8881788cf1..19986fd8a8 100644
--- a/apps/codecs/spc/spc_dsp.c
+++ b/apps/codecs/spc/spc_dsp.c
@@ -25,14 +25,13 @@
#include "spc_codec.h"
#include "spc_profiler.h"
-#ifdef CPU_COLDFIRE
-static int32_t fir_buf[FIR_BUF_HALF]
- __attribute__ ((aligned (FIR_BUF_SIZE*2))) IBSS_ATTR;
+#if defined(CPU_COLDFIRE) || defined (CPU_ARM)
+int32_t fir_buf[FIR_BUF_CNT]
+ __attribute__ ((aligned (FIR_BUF_ALIGN*1))) IBSS_ATTR;
#endif
-
#if SPC_BRRCACHE
/* a little extra for samples that go past end */
-int16_t BRRcache [0x20000 + 32];
+int16_t BRRcache [BRR_CACHE_SIZE];
#endif
void DSP_write( struct Spc_Dsp* this, int i, int data )
@@ -58,11 +57,12 @@ void DSP_write( struct Spc_Dsp* this, int i, int data )
/* if ( n < -32768 ) out = -32768; */
/* if ( n > 32767 ) out = 32767; */
-#define CLAMP16( n, out )\
-{\
- if ( (int16_t) n != n )\
- out = 0x7FFF ^ (n >> 31);\
-}
+#define CLAMP16( n ) \
+({ \
+ if ( (int16_t) n != n ) \
+ n = 0x7FFF ^ (n >> 31); \
+ n; \
+})
#if SPC_BRRCACHE
static void decode_brr( struct Spc_Dsp* this, unsigned start_addr,
@@ -181,7 +181,7 @@ static void decode_brr( struct Spc_Dsp* this, unsigned start_addr,
smp2 = smp1;
}
- CLAMP16( delta, delta );
+ delta = CLAMP16( delta );
smp1 = (int16_t) (delta * 2); /* sign-extend */
}
while ( (offset += 4) != 0 );
@@ -359,7 +359,7 @@ void DSP_run_( struct Spc_Dsp* this, long count, int32_t* out_buf )
#define VOICE_RATE(x) (INT16A(raw_voice->rate) & 0x3FFF)
#define IF_RBE(...)
#endif /* ROCKBOX_BIG_ENDIAN */
-
+
#if !SPC_NOINTERP
int const slow_gaussian = (this->r.g.pitch_mods >> 1) |
this->r.g.noise_enables;
@@ -431,7 +431,7 @@ void DSP_run_( struct Spc_Dsp* this, long count, int32_t* out_buf )
/* Envelope */
{
- int const env_range = 0x800;
+ int const ENV_RANGE = 0x800;
int env_mode = voice->env_mode;
int adsr0 = raw_voice->adsr [0];
int env_timer;
@@ -482,14 +482,14 @@ void DSP_run_( struct Spc_Dsp* this, long count, int32_t* out_buf )
int envx = voice->envx;
- int const step = env_range / 64;
+ int const step = ENV_RANGE / 64;
envx += step;
if ( t == 15 )
- envx += env_range / 2 - step;
+ envx += ENV_RANGE / 2 - step;
- if ( envx >= env_range )
+ if ( envx >= ENV_RANGE )
{
- envx = env_range - 1;
+ envx = ENV_RANGE - 1;
voice->env_mode = state_decay;
}
voice->envx = envx;
@@ -516,7 +516,7 @@ void DSP_run_( struct Spc_Dsp* this, long count, int32_t* out_buf )
int mode = t >> 5;
if ( mode <= 5 ) /* decay */
{
- int step = env_range / 64;
+ int step = ENV_RANGE / 64;
if ( mode == 5 ) /* exponential */
{
envx--; /* envx *= 255 / 256 */
@@ -531,14 +531,14 @@ void DSP_run_( struct Spc_Dsp* this, long count, int32_t* out_buf )
}
else /* attack */
{
- int const step = env_range / 64;
+ int const step = ENV_RANGE / 64;
envx += step;
if ( mode == 7 &&
- envx >= env_range * 3 / 4 + step )
- envx += env_range / 256 - step;
+ envx >= ENV_RANGE * 3 / 4 + step )
+ envx += ENV_RANGE / 256 - step;
- if ( envx >= env_range )
- envx = env_range - 1;
+ if ( envx >= ENV_RANGE )
+ envx = ENV_RANGE - 1;
}
voice->envx = envx;
/* TODO: should this be 8? */
@@ -550,7 +550,7 @@ void DSP_run_( struct Spc_Dsp* this, long count, int32_t* out_buf )
else /* state_release */
{
int envx = voice->envx;
- if ( (envx -= env_range / 256) > 0 )
+ if ( (envx -= ENV_RANGE / 256) > 0 )
{
voice->envx = envx;
raw_voice->envx = envx >> 8;
@@ -683,7 +683,7 @@ void DSP_run_( struct Spc_Dsp* this, long count, int32_t* out_buf )
smp2 = smp1;
}
- CLAMP16( delta, delta );
+ delta = CLAMP16( delta );
smp1 = (int16_t) (delta * 2); /* sign-extend */
}
while ( (offset += 4) != 0 );
@@ -778,7 +778,7 @@ void DSP_run_( struct Spc_Dsp* this, long count, int32_t* out_buf )
output = (output + rev [1] * interp [2]) >> 12;
output = (int16_t) (output * 2);
output += ((rev [0] * interp [3]) >> 12) * 2;
- CLAMP16( output, output );
+ output = CLAMP16( output );
}
output = (output * voice->envx) >> 11 & ~1;
@@ -788,7 +788,7 @@ void DSP_run_( struct Spc_Dsp* this, long count, int32_t* out_buf )
prev_outx = output;
raw_voice->outx = (int8_t) (output >> 8);
}
- #else
+ #else /* SPCNOINTERP */
/* two-point linear interpolation */
#ifdef CPU_COLDFIRE
int amp_0 = (int16_t)this->noise;
@@ -822,7 +822,7 @@ void DSP_run_( struct Spc_Dsp* this, long count, int32_t* out_buf )
/* output = y0 + (result >> 12) */
"asr.l %[sh], %[y1] \r\n"
"add.l %[y0], %[y1] \r\n"
- : [f]"+&d"(f), [y0]"=&a"(y0), [y1]"=&d"(amp_0)
+ : [f]"+d"(f), [y0]"=&a"(y0), [y1]"=&d"(amp_0)
: [s]"a"(voice->samples), [sh]"d"(12)
);
}
@@ -861,17 +861,49 @@ void DSP_run_( struct Spc_Dsp* this, long count, int32_t* out_buf )
"movclr.l %%acc1, %[amp_1] \r\n"
: [amp_0]"=r"(amp_0), [amp_1]"=r"(amp_1)
);
- #else
+ #elif defined (CPU_ARM)
+ int amp_0, amp_1;
+
+ if ( (this->r.g.noise_enables & vbit) != 0 ) {
+ amp_0 = *(int16_t *)&this->noise;
+ } else {
+ uint32_t f = voice->position;
+ amp_0 = (uint32_t)voice->samples;
- /* Try this one out on ARM and see - similar to above but the asm
- on coldfire removes a redundant register load worth 1 or 2%;
- switching to loading two samples at once may help too. That's
- done above and while 6 to 7% faster on cf over two 16 bit loads
- it makes it endian dependant.
-
- measured small improvement (~1.5%) - hcs
- */
+ asm volatile(
+ "mov %[y1], %[f], lsr #12 \r\n"
+ "eor %[f], %[f], %[y1], lsl #12 \r\n"
+ "add %[y1], %[y0], %[y1], lsl #1 \r\n"
+ "ldrsh %[y0], [%[y1], #2] \r\n"
+ "ldrsh %[y1], [%[y1], #4] \r\n"
+ "sub %[y1], %[y1], %[y0] \r\n"
+ "mul %[f], %[y1], %[f] \r\n"
+ "add %[y0], %[y0], %[f], asr #12 \r\n"
+ : [f]"+r"(f), [y0]"+r"(amp_0), [y1]"=&r"(amp_1)
+ );
+ }
+
+ voice->position += rate;
+
+ asm volatile(
+ "mul %[amp_1], %[amp_0], %[envx] \r\n"
+ "mov %[amp_0], %[amp_1], asr #11 \r\n"
+ "mov %[amp_1], %[amp_0], asr #8 \r\n"
+ : [amp_0]"+r"(amp_0), [amp_1]"=&r"(amp_1)
+ : [envx]"r"(voice->envx)
+ );
+
+ prev_outx = amp_0;
+ raw_voice->outx = (int8_t)amp_1;
+ asm volatile(
+ "mul %[amp_1], %[amp_0], %[vol_1] \r\n"
+ "mul %[amp_0], %[vol_0], %[amp_0] \r\n"
+ : [amp_0]"+r"(amp_0), [amp_1]"+r"(amp_1)
+ : [vol_0]"r"((int)voice->volume[0]),
+ [vol_1]"r"((int)voice->volume[1])
+ );
+ #else /* Unoptimized CPU */
int output;
if ( (this->r.g.noise_enables & vbit) == 0 )
@@ -884,19 +916,7 @@ void DSP_run_( struct Spc_Dsp* this, long count, int32_t* out_buf )
}
voice->position += rate;
-
- /* old version */
-#if 0
- int fraction = voice->position & 0xFFF;
- short const* const pos = voice->samples + (voice->position >> 12);
- voice->position += rate;
- int output =
- (pos [2] * fraction + pos [1] * (0x1000 - fraction)) >> 12;
- /* no interpolation (hardly faster, and crappy sounding) */
- /*int output = pos [0];*/
- if ( this->r.g.noise_enables & vbit )
- output = *(int16_t*) &this->noise;
-#endif
+
output = (output * voice->envx) >> 11;
/* duplicated here to give compiler more to run in parallel */
@@ -905,8 +925,8 @@ void DSP_run_( struct Spc_Dsp* this, long count, int32_t* out_buf )
prev_outx = output;
raw_voice->outx = (int8_t) (output >> 8);
- #endif /* CPU_COLDFIRE */
- #endif
+ #endif /* CPU_* */
+ #endif /* SPCNOINTERP */
#if SPC_BRRCACHE
if ( voice->position >= voice->wave_end )
@@ -1033,7 +1053,7 @@ void DSP_run_( struct Spc_Dsp* this, long count, int32_t* out_buf )
"or.l %[sh], %[e0] \r\n"
/* save final feedback into echo buffer */
"move.l %[e0], (%[echo_ptr]) \r\n"
- : [e0]"+&d"(echo_0), [e1]"+&d"(echo_1)
+ : [e0]"+d"(echo_0), [e1]"+d"(echo_1)
: [out_0]"r"(out_0), [out_1]"r"(out_1),
[ef]"r"((int)this->r.g.echo_feedback),
[echo_ptr]"a"((int32_t *)echo_ptr),
@@ -1056,7 +1076,88 @@ void DSP_run_( struct Spc_Dsp* this, long count, int32_t* out_buf )
out_buf [ 0] = out_0;
out_buf [WAV_CHUNK_SIZE] = out_1;
out_buf ++;
- #else /* !CPU_COLDFIRE */
+ #elif defined (CPU_ARM)
+ /* Read feedback from echo buffer */
+ int echo_pos = this->echo_pos;
+ uint8_t* const echo_ptr = RAM +
+ ((this->r.g.echo_page * 0x100 + echo_pos) & 0xFFFF);
+ echo_pos += 4;
+ if ( echo_pos >= (this->r.g.echo_delay & 15) * 0x800 )
+ echo_pos = 0;
+ this->echo_pos = echo_pos;
+
+ int fb_0 = GET_LE16SA( echo_ptr );
+ int fb_1 = GET_LE16SA( echo_ptr + 2 );
+
+ /* Keep last 8 samples */
+ int32_t *fir_ptr = this->fir_ptr;
+
+ /* Apply FIR */
+ asm volatile (
+ "str %[fb_0], [%[fir_p]], #4 \r\n"
+ "str %[fb_1], [%[fir_p]], #4 \r\n"
+ /* duplicate at +8 eliminates wrap checking below */
+ "str %[fb_0], [%[fir_p], #56] \r\n"
+ "str %[fb_1], [%[fir_p], #60] \r\n"
+ : [fir_p]"+r"(fir_ptr)
+ : [fb_0]"r"(fb_0), [fb_1]"r"(fb_1)
+ );
+
+ this->fir_ptr = (int32_t *)((intptr_t)fir_ptr & FIR_BUF_MASK);
+ int32_t *fir_coeff = this->fir_coeff;
+
+ asm volatile (
+ "ldmia %[fir_c]!, { r0-r1 } \r\n"
+ "ldmia %[fir_p]!, { r4-r5 } \r\n"
+ "mul %[fb_0], r0, %[fb_0] \r\n"
+ "mul %[fb_1], r0, %[fb_1] \r\n"
+ "mla %[fb_0], r4, r1, %[fb_0] \r\n"
+ "mla %[fb_1], r5, r1, %[fb_1] \r\n"
+ "ldmia %[fir_c]!, { r0-r1 } \r\n"
+ "ldmia %[fir_p]!, { r2-r5 } \r\n"
+ "mla %[fb_0], r2, r0, %[fb_0] \r\n"
+ "mla %[fb_1], r3, r0, %[fb_1] \r\n"
+ "mla %[fb_0], r4, r1, %[fb_0] \r\n"
+ "mla %[fb_1], r5, r1, %[fb_1] \r\n"
+ "ldmia %[fir_c]!, { r0-r1 } \r\n"
+ "ldmia %[fir_p]!, { r2-r5 } \r\n"
+ "mla %[fb_0], r2, r0, %[fb_0] \r\n"
+ "mla %[fb_1], r3, r0, %[fb_1] \r\n"
+ "mla %[fb_0], r4, r1, %[fb_0] \r\n"
+ "mla %[fb_1], r5, r1, %[fb_1] \r\n"
+ "ldmia %[fir_c]!, { r0-r1 } \r\n"
+ "ldmia %[fir_p]!, { r2-r5 } \r\n"
+ "mla %[fb_0], r2, r0, %[fb_0] \r\n"
+ "mla %[fb_1], r3, r0, %[fb_1] \r\n"
+ "mla %[fb_0], r4, r1, %[fb_0] \r\n"
+ "mla %[fb_1], r5, r1, %[fb_1] \r\n"
+ : [fb_0]"+r"(fb_0), [fb_1]"+r"(fb_1),
+ [fir_p]"+r"(fir_ptr), [fir_c]"+r"(fir_coeff)
+ :
+ : "r0", "r1", "r2", "r3", "r4", "r5"
+ );
+
+ /* Generate output */
+ int amp_0 = (chans_0 * global_vol_0 + fb_0 * this->r.g.echo_volume_0)
+ >> global_muting;
+ int amp_1 = (chans_1 * global_vol_1 + fb_1 * this->r.g.echo_volume_1)
+ >> global_muting;
+
+ out_buf [ 0] = amp_0;
+ out_buf [WAV_CHUNK_SIZE] = amp_1;
+ out_buf ++;
+
+ if ( !(this->r.g.flags & 0x20) )
+ {
+ /* Feedback into echo buffer */
+ int e0 = (echo_0 >> 7) + ((fb_0 * this->r.g.echo_feedback) >> 14);
+ int e1 = (echo_1 >> 7) + ((fb_1 * this->r.g.echo_feedback) >> 14);
+ e0 = CLAMP16( e0 );
+ SET_LE16A( echo_ptr , e0 );
+ e1 = CLAMP16( e1 );
+ SET_LE16A( echo_ptr + 2, e1 );
+ }
+ #else /* Unoptimized CPU */
/* Read feedback from echo buffer */
int echo_pos = this->echo_pos;
uint8_t* const echo_ptr = RAM +
@@ -1102,25 +1203,25 @@ void DSP_run_( struct Spc_Dsp* this, long count, int32_t* out_buf )
out_buf [WAV_CHUNK_SIZE] = amp_1;
out_buf ++;
- /* Feedback into echo buffer */
- int e0 = (echo_0 >> 7) + ((fb_0 * this->r.g.echo_feedback) >> 14);
- int e1 = (echo_1 >> 7) + ((fb_1 * this->r.g.echo_feedback) >> 14);
if ( !(this->r.g.flags & 0x20) )
{
- CLAMP16( e0, e0 );
+ /* Feedback into echo buffer */
+ int e0 = (echo_0 >> 7) + ((fb_0 * this->r.g.echo_feedback) >> 14);
+ int e1 = (echo_1 >> 7) + ((fb_1 * this->r.g.echo_feedback) >> 14);
+ e0 = CLAMP16( e0 );
SET_LE16A( echo_ptr , e0 );
- CLAMP16( e1, e1 );
+ e1 = CLAMP16( e1 );
SET_LE16A( echo_ptr + 2, e1 );
}
- #endif /* CPU_COLDFIRE */
- #else
+ #endif /* CPU_* */
+ #else /* SPCNOECHO == 1*/
/* Generate output */
int amp_0 = (chans_0 * global_vol_0) >> global_muting;
int amp_1 = (chans_1 * global_vol_1) >> global_muting;
out_buf [ 0] = amp_0;
out_buf [WAV_CHUNK_SIZE] = amp_1;
out_buf ++;
- #endif
+ #endif /* SPCNOECHO */
}
while ( --count );
#if 0
@@ -1155,10 +1256,13 @@ void DSP_reset( struct Spc_Dsp* this )
this->wave_entry [i].start_addr = -1;
#endif
-#ifdef CPU_COLDFIRE
- this->fir_ptr = fir_buf;
+#if defined(CPU_COLDFIRE)
+ this->fir_ptr = fir_buf;
this->last_fir_ptr = &fir_buf [7];
ci->memset( fir_buf, 0, sizeof fir_buf );
+#elif defined (CPU_ARM)
+ this->fir_ptr = fir_buf;
+ ci->memset( fir_buf, 0, sizeof fir_buf );
#else
this->fir_pos = 0;
ci->memset( this->fir_buf, 0, sizeof this->fir_buf );