mirror of
https://github.com/vgmstream/vgmstream.git
synced 2025-01-17 23:36:41 +01:00
commit
e58c73aab1
@ -20,7 +20,7 @@ void decode_wv6_ima(VGMSTREAMCHANNEL * stream, sample_t * outbuf, int channelspa
|
||||
void decode_alp_ima(VGMSTREAMCHANNEL * stream, sample_t * outbuf, int channelspacing, int32_t first_sample, int32_t samples_to_do);
|
||||
void decode_ffta2_ima(VGMSTREAMCHANNEL * stream, sample_t * outbuf, int channelspacing, int32_t first_sample, int32_t samples_to_do);
|
||||
void decode_blitz_ima(VGMSTREAMCHANNEL * stream, sample_t * outbuf, int channelspacing, int32_t first_sample, int32_t samples_to_do);
|
||||
void decode_mtf_ima(VGMSTREAMCHANNEL * stream, sample_t * outbuf, int channelspacing, int32_t first_sample, int32_t samples_to_do);
|
||||
void decode_mtf_ima(VGMSTREAMCHANNEL * stream, sample_t * outbuf, int channelspacing, int32_t first_sample, int32_t samples_to_do, int channel, int is_stereo);
|
||||
|
||||
void decode_ms_ima(VGMSTREAM * vgmstream,VGMSTREAMCHANNEL * stream, sample_t * outbuf, int channelspacing, int32_t first_sample, int32_t samples_to_do,int channel);
|
||||
void decode_ref_ima(VGMSTREAM * vgmstream, VGMSTREAMCHANNEL * stream, sample_t * outbuf, int channelspacing, int32_t first_sample, int32_t samples_to_do,int channel);
|
||||
|
@ -318,7 +318,7 @@ void decode_standard_ima(VGMSTREAMCHANNEL * stream, sample_t * outbuf, int chann
|
||||
stream->adpcm_step_index = step_index;
|
||||
}
|
||||
|
||||
void decode_mtf_ima(VGMSTREAMCHANNEL * stream, sample_t * outbuf, int channelspacing, int32_t first_sample, int32_t samples_to_do) {
|
||||
void decode_mtf_ima(VGMSTREAMCHANNEL * stream, sample_t * outbuf, int channelspacing, int32_t first_sample, int32_t samples_to_do, int channel, int is_stereo) {
|
||||
int i, sample_count = 0;
|
||||
int32_t hist1 = stream->adpcm_history1_32;
|
||||
int step_index = stream->adpcm_step_index;
|
||||
@ -331,8 +331,12 @@ void decode_mtf_ima(VGMSTREAMCHANNEL * stream, sample_t * outbuf, int channelspa
|
||||
|
||||
/* decode nibbles (layout: varies) */
|
||||
for (i = first_sample; i < first_sample + samples_to_do; i++, sample_count += channelspacing) {
|
||||
off_t byte_offset = stream->offset + i/2;
|
||||
int nibble_shift = ((i&1) ? 0:4);
|
||||
off_t byte_offset = is_stereo ?
|
||||
stream->offset + i : /* stereo: one nibble per channel */
|
||||
stream->offset + i/2; /* mono: consecutive nibbles */
|
||||
int nibble_shift = is_stereo ?
|
||||
((channel&1) ? 0:4) :
|
||||
((i&1) ? 0:4);
|
||||
|
||||
mtf_ima_expand_nibble(stream, byte_offset,nibble_shift, &hist1, &step_index);
|
||||
outbuf[sample_count] = clamp16(hist1 >> 4);
|
||||
|
@ -1,209 +1,201 @@
|
||||
#include "../util.h"
|
||||
#include "coding.h"
|
||||
|
||||
/* fixed point amount to scale the current step size */
|
||||
static const unsigned int scale_step_aica[16] = {
|
||||
230, 230, 230, 230, 307, 409, 512, 614,
|
||||
230, 230, 230, 230, 307, 409, 512, 614
|
||||
};
|
||||
|
||||
static const int scale_step_adpcmb[16] = {
|
||||
57, 57, 57, 57, 77, 102, 128, 153,
|
||||
57, 57, 57, 57, 77, 102, 128, 153,
|
||||
};
|
||||
|
||||
/* look-up for 'mul' IMA's sign*((code&7) * 2 + 1) for every code */
|
||||
static const int scale_delta[16] = {
|
||||
1, 3, 5, 7, 9, 11, 13, 15,
|
||||
-1, -3, -5, -7, -9,-11,-13,-15
|
||||
};
|
||||
|
||||
/* Yamaha ADPCM-B (aka DELTA-T) expand used in YM2608/YM2610/etc (cross referenced with various sources and .so) */
|
||||
static void yamaha_adpcmb_expand_nibble(VGMSTREAMCHANNEL * stream, off_t byte_offset, int nibble_shift, int32_t* hist1, int32_t* step_size, int16_t *out_sample) {
|
||||
int code, delta, sample;
|
||||
|
||||
code = (read_8bit(byte_offset,stream->streamfile) >> nibble_shift) & 0xf;
|
||||
delta = ((((code & 0x7) * 2) + 1) * (*step_size)) >> 3; /* like 'mul' IMA */
|
||||
if (code & 8)
|
||||
delta = -delta;
|
||||
sample = *hist1 + delta;
|
||||
|
||||
sample = clamp16(sample); /* this may not be needed (not done in Aska) but no byte changes */
|
||||
|
||||
*step_size = ((*step_size) * scale_step_adpcmb[code]) >> 6;
|
||||
if (*step_size < 0x7f) *step_size = 0x7f;
|
||||
else if (*step_size > 0x6000) *step_size = 0x6000;
|
||||
|
||||
*out_sample = sample;
|
||||
*hist1 = sample;
|
||||
}
|
||||
|
||||
/* Yamaha AICA expand, slightly filtered vs "ACM" Yamaha ADPCM, same as Creative ADPCM
|
||||
* (some info from https://github.com/vgmrips/vgmplay, https://wiki.multimedia.cx/index.php/Creative_ADPCM) */
|
||||
static void yamaha_aica_expand_nibble(VGMSTREAMCHANNEL * stream, off_t byte_offset, int nibble_shift, int32_t* hist1, int32_t* step_size, int16_t *out_sample) {
|
||||
int code, delta, sample;
|
||||
|
||||
*hist1 = *hist1 * 254 / 256; /* hist filter is vital to get correct waveform but not done in many emus */
|
||||
|
||||
code = ((read_8bit(byte_offset,stream->streamfile) >> nibble_shift))&0xf;
|
||||
delta = (*step_size * scale_delta[code]) / 8; /* 'mul' IMA with table (not sure if part of encoder) */
|
||||
sample = *hist1 + delta;
|
||||
|
||||
sample = clamp16(sample); /* apparently done by official encoder */
|
||||
|
||||
*step_size = ((*step_size) * scale_step_aica[code]) >> 8;
|
||||
if (*step_size < 0x7f) *step_size = 0x7f;
|
||||
else if (*step_size > 0x6000) *step_size = 0x6000;
|
||||
|
||||
*out_sample = sample;
|
||||
*hist1 = sample;
|
||||
}
|
||||
|
||||
|
||||
/* info about Yamaha ADPCM as created by official yadpcm.acm (in 'Yamaha-ADPCM-ACM-Driver-100-j')
|
||||
* - possibly RIFF codec 0x20
|
||||
* - simply called "Yamaha ADPCM Codec" (even though not quite like Yamaha ADPCM-B)
|
||||
* - block_align = (sample_rate / 0x3C + 0x04) * channels (ex. 0x2E6 for 22050 stereo, probably given in RIFF)
|
||||
* - low nibble first, stereo or mono modes (no interleave)
|
||||
* - expand (old IMA 'shift+add' style, not 'mul' style):
|
||||
* delta = step_size >> 3;
|
||||
* if (code & 1) delta += step_size >> 2;
|
||||
* if (code & 2) delta += step_size >> 1;
|
||||
* if (code & 4) delta += step_size;
|
||||
* if (code & 8) delta = -delta;
|
||||
* sample = hist + clamp16(delta);
|
||||
* though compiled more like:
|
||||
* sample = hist + (1-2*(code & 8) * (step_size/8 + step_size/2 * (code&2) + step_size/4 * (code&1) + step_size * (code&4))
|
||||
* - step_size update:
|
||||
* step_size = clamp_range(step_size * scale_step_aica[code] >> 8, 0x7F, 0x6000)
|
||||
*/
|
||||
|
||||
|
||||
/* Yamaha AICA ADPCM (also used in YMZ280B with high nibble first) */
|
||||
void decode_aica(VGMSTREAMCHANNEL * stream, sample_t * outbuf, int channelspacing, int32_t first_sample, int32_t samples_to_do, int channel, int is_stereo) {
|
||||
int i, sample_count = 0;
|
||||
int16_t out_sample;
|
||||
int32_t hist1 = stream->adpcm_history1_16;
|
||||
int step_size = stream->adpcm_step_index;
|
||||
|
||||
/* no header (external setup), pre-clamp for wrong values */
|
||||
if (step_size < 0x7f) step_size = 0x7f;
|
||||
if (step_size > 0x6000) step_size = 0x6000;
|
||||
|
||||
for (i = first_sample; i < first_sample + samples_to_do; i++) {
|
||||
off_t byte_offset = is_stereo ?
|
||||
stream->offset + i : /* stereo: one nibble per channel */
|
||||
stream->offset + i/2; /* mono: consecutive nibbles */
|
||||
int nibble_shift = is_stereo ?
|
||||
(!(channel&1) ? 0:4) : /* even = low/L, odd = high/R */
|
||||
(!(i&1) ? 0:4); /* low nibble first */
|
||||
|
||||
yamaha_aica_expand_nibble(stream, byte_offset, nibble_shift, &hist1, &step_size, &out_sample);
|
||||
outbuf[sample_count] = out_sample;
|
||||
sample_count += channelspacing;
|
||||
}
|
||||
|
||||
stream->adpcm_history1_16 = hist1;
|
||||
stream->adpcm_step_index = step_size;
|
||||
}
|
||||
|
||||
/* tri-Ace Aska ADPCM, Yamaha ADPCM-B with headered frames (reversed from Android SO's .so)
|
||||
* implements table with if-else/switchs too but that's too goofy */
|
||||
void decode_aska(VGMSTREAMCHANNEL * stream, sample_t * outbuf, int channelspacing, int32_t first_sample, int32_t samples_to_do, int channel) {
|
||||
int i, sample_count = 0, num_frame;
|
||||
int16_t out_sample;
|
||||
int32_t hist1 = stream->adpcm_history1_32;
|
||||
int step_size = stream->adpcm_step_index;
|
||||
|
||||
/* external interleave */
|
||||
int block_samples = (0x40 - 0x04*channelspacing) * 2 / channelspacing;
|
||||
num_frame = first_sample / block_samples;
|
||||
first_sample = first_sample % block_samples;
|
||||
|
||||
/* header (hist+step) */
|
||||
if (first_sample == 0) {
|
||||
off_t header_offset = stream->offset + 0x40*num_frame + 0x04*channel;
|
||||
|
||||
hist1 = read_16bitLE(header_offset+0x00,stream->streamfile);
|
||||
step_size = read_16bitLE(header_offset+0x02,stream->streamfile);
|
||||
/* in most files 1st frame has step 0 but it seems ok and accounted for */
|
||||
//if (step_size < 0x7f) step_size = 0x7f;
|
||||
//else if (step_size > 0x6000) step_size = 0x6000;
|
||||
}
|
||||
|
||||
/* decode nibbles (layout: varies) */
|
||||
for (i = first_sample; i < first_sample + samples_to_do; i++) {
|
||||
off_t byte_offset = (channelspacing == 2) ?
|
||||
(stream->offset + 0x40*num_frame + 0x04*channelspacing) + i : /* stereo: one nibble per channel */
|
||||
(stream->offset + 0x40*num_frame + 0x04*channelspacing) + i/2; /* mono: consecutive nibbles */
|
||||
int nibble_shift = (channelspacing == 2) ?
|
||||
(!(channel&1) ? 0:4) :
|
||||
(!(i&1) ? 0:4); /* even = low, odd = high */
|
||||
|
||||
yamaha_adpcmb_expand_nibble(stream, byte_offset, nibble_shift, &hist1, &step_size, &out_sample);
|
||||
outbuf[sample_count] = out_sample;
|
||||
sample_count += channelspacing;
|
||||
}
|
||||
|
||||
stream->adpcm_history1_32 = hist1;
|
||||
stream->adpcm_step_index = step_size;
|
||||
}
|
||||
|
||||
|
||||
/* Yamaha ADPCM with unknown expand variation (noisy), step size is double of normal Yamaha? */
|
||||
void decode_nxap(VGMSTREAMCHANNEL * stream, sample_t * outbuf, int channelspacing, int32_t first_sample, int32_t samples_to_do) {
|
||||
int i, sample_count = 0, num_frame;
|
||||
int32_t hist1 = stream->adpcm_history1_32;
|
||||
int step_size = stream->adpcm_step_index;
|
||||
|
||||
/* external interleave */
|
||||
int block_samples = (0x40 - 0x4) * 2;
|
||||
num_frame = first_sample / block_samples;
|
||||
first_sample = first_sample % block_samples;
|
||||
|
||||
/* header (hist+step) */
|
||||
if (first_sample == 0) {
|
||||
off_t header_offset = stream->offset + 0x40*num_frame;
|
||||
|
||||
hist1 = read_16bitLE(header_offset+0x00,stream->streamfile);
|
||||
step_size = read_16bitLE(header_offset+0x02,stream->streamfile);
|
||||
if (step_size < 0x7f) step_size = 0x7f;
|
||||
else if (step_size > 0x6000) step_size = 0x6000;
|
||||
}
|
||||
|
||||
/* decode nibbles (layout: all nibbles from one channel) */
|
||||
for (i = first_sample; i < first_sample + samples_to_do; i++) {
|
||||
int code, delta, sample;
|
||||
off_t byte_offset = (stream->offset + 0x40*num_frame + 0x04) + i/2;
|
||||
int nibble_shift = (i&1?4:0); /* low nibble first? */
|
||||
|
||||
code = (read_8bit(byte_offset,stream->streamfile) >> nibble_shift)&0xf;
|
||||
delta = (step_size * scale_delta[code]) / 8; //todo wrong
|
||||
sample = hist1 + delta;
|
||||
|
||||
outbuf[sample_count] = clamp16(sample);
|
||||
hist1 = outbuf[sample_count];
|
||||
|
||||
step_size = (step_size * scale_step_aica[code]) / 260.0; //todo wrong
|
||||
if (step_size < 0x7f) step_size = 0x7f;
|
||||
else if (step_size > 0x6000) step_size = 0x6000;
|
||||
|
||||
sample_count += channelspacing;
|
||||
}
|
||||
|
||||
stream->adpcm_history1_32 = hist1;
|
||||
stream->adpcm_step_index = step_size;
|
||||
}
|
||||
|
||||
size_t yamaha_bytes_to_samples(size_t bytes, int channels) {
|
||||
if (channels <= 0) return 0;
|
||||
/* 2 samples per byte (2 nibbles) in stereo or mono config */
|
||||
return bytes * 2 / channels;
|
||||
}
|
||||
|
||||
size_t aska_bytes_to_samples(size_t bytes, int channels) {
|
||||
int block_align = 0x40;
|
||||
if (channels <= 0) return 0;
|
||||
return (bytes / block_align) * (block_align - 0x04*channels) * 2 / channels
|
||||
+ ((bytes % block_align) ? ((bytes % block_align) - 0x04*channels) * 2 / channels : 0);
|
||||
}
|
||||
#include "../util.h"
|
||||
#include "coding.h"
|
||||
|
||||
/* fixed point amount to scale the current step size */
|
||||
static const unsigned int scale_step_aica[16] = {
|
||||
230, 230, 230, 230, 307, 409, 512, 614,
|
||||
230, 230, 230, 230, 307, 409, 512, 614
|
||||
};
|
||||
|
||||
static const int scale_step_adpcmb[16] = {
|
||||
57, 57, 57, 57, 77, 102, 128, 153,
|
||||
57, 57, 57, 57, 77, 102, 128, 153,
|
||||
};
|
||||
|
||||
/* look-up for 'mul' IMA's sign*((code&7) * 2 + 1) for every code */
|
||||
static const int scale_delta[16] = {
|
||||
1, 3, 5, 7, 9, 11, 13, 15,
|
||||
-1, -3, -5, -7, -9,-11,-13,-15
|
||||
};
|
||||
|
||||
/* Yamaha ADPCM-B (aka DELTA-T) expand used in YM2608/YM2610/etc (cross referenced with various sources and .so) */
|
||||
static void yamaha_adpcmb_expand_nibble(VGMSTREAMCHANNEL * stream, off_t byte_offset, int nibble_shift, int32_t* hist1, int32_t* step_size, int16_t *out_sample) {
|
||||
int code, delta, sample;
|
||||
|
||||
code = (read_8bit(byte_offset,stream->streamfile) >> nibble_shift) & 0xf;
|
||||
delta = ((((code & 0x7) * 2) + 1) * (*step_size)) >> 3; /* like 'mul' IMA */
|
||||
if (code & 8)
|
||||
delta = -delta;
|
||||
sample = *hist1 + delta;
|
||||
|
||||
sample = clamp16(sample); /* not needed in Aska but seems others do */
|
||||
|
||||
*step_size = ((*step_size) * scale_step_adpcmb[code]) >> 6;
|
||||
if (*step_size < 0x7f) *step_size = 0x7f;
|
||||
else if (*step_size > 0x6000) *step_size = 0x6000;
|
||||
|
||||
*out_sample = sample;
|
||||
*hist1 = sample;
|
||||
}
|
||||
|
||||
/* Yamaha AICA expand, slightly filtered vs "ACM" Yamaha ADPCM, same as Creative ADPCM
|
||||
* (some info from https://github.com/vgmrips/vgmplay, https://wiki.multimedia.cx/index.php/Creative_ADPCM) */
|
||||
static void yamaha_aica_expand_nibble(VGMSTREAMCHANNEL * stream, off_t byte_offset, int nibble_shift, int32_t* hist1, int32_t* step_size, int16_t *out_sample) {
|
||||
int code, delta, sample;
|
||||
|
||||
*hist1 = *hist1 * 254 / 256; /* hist filter is vital to get correct waveform but not done in many emus */
|
||||
|
||||
code = ((read_8bit(byte_offset,stream->streamfile) >> nibble_shift))&0xf;
|
||||
delta = (*step_size * scale_delta[code]) / 8; /* 'mul' IMA with table (not sure if part of encoder) */
|
||||
sample = *hist1 + delta;
|
||||
|
||||
sample = clamp16(sample); /* apparently done by official encoder */
|
||||
|
||||
*step_size = ((*step_size) * scale_step_aica[code]) >> 8;
|
||||
if (*step_size < 0x7f) *step_size = 0x7f;
|
||||
else if (*step_size > 0x6000) *step_size = 0x6000;
|
||||
|
||||
*out_sample = sample;
|
||||
*hist1 = sample;
|
||||
}
|
||||
|
||||
|
||||
/* info about Yamaha ADPCM as created by official yadpcm.acm (in 'Yamaha-ADPCM-ACM-Driver-100-j')
|
||||
* - possibly RIFF codec 0x20
|
||||
* - simply called "Yamaha ADPCM Codec" (even though not quite like Yamaha ADPCM-B)
|
||||
* - block_align = (sample_rate / 0x3C + 0x04) * channels (ex. 0x2E6 for 22050 stereo, probably given in RIFF)
|
||||
* - low nibble first, stereo or mono modes (no interleave)
|
||||
* - expand (old IMA 'shift+add' style, not 'mul' style):
|
||||
* delta = step_size >> 3;
|
||||
* if (code & 1) delta += step_size >> 2;
|
||||
* if (code & 2) delta += step_size >> 1;
|
||||
* if (code & 4) delta += step_size;
|
||||
* if (code & 8) delta = -delta;
|
||||
* sample = hist + clamp16(delta);
|
||||
* though compiled more like:
|
||||
* sample = hist + (1-2*(code & 8) * (step_size/8 + step_size/2 * (code&2) + step_size/4 * (code&1) + step_size * (code&4))
|
||||
* - step_size update:
|
||||
* step_size = clamp_range(step_size * scale_step_aica[code] >> 8, 0x7F, 0x6000)
|
||||
*/
|
||||
|
||||
|
||||
/* Yamaha AICA ADPCM (also used in YMZ280B with high nibble first) */
|
||||
void decode_aica(VGMSTREAMCHANNEL * stream, sample_t * outbuf, int channelspacing, int32_t first_sample, int32_t samples_to_do, int channel, int is_stereo) {
|
||||
int i, sample_count = 0;
|
||||
int16_t out_sample;
|
||||
int32_t hist1 = stream->adpcm_history1_16;
|
||||
int step_size = stream->adpcm_step_index;
|
||||
|
||||
/* no header (external setup), pre-clamp for wrong values */
|
||||
if (step_size < 0x7f) step_size = 0x7f;
|
||||
if (step_size > 0x6000) step_size = 0x6000;
|
||||
|
||||
for (i = first_sample; i < first_sample + samples_to_do; i++) {
|
||||
off_t byte_offset = is_stereo ?
|
||||
stream->offset + i : /* stereo: one nibble per channel */
|
||||
stream->offset + i/2; /* mono: consecutive nibbles */
|
||||
int nibble_shift = is_stereo ?
|
||||
(!(channel&1) ? 0:4) : /* even = low/L, odd = high/R */
|
||||
(!(i&1) ? 0:4); /* low nibble first */
|
||||
|
||||
yamaha_aica_expand_nibble(stream, byte_offset, nibble_shift, &hist1, &step_size, &out_sample);
|
||||
outbuf[sample_count] = out_sample;
|
||||
sample_count += channelspacing;
|
||||
}
|
||||
|
||||
stream->adpcm_history1_16 = hist1;
|
||||
stream->adpcm_step_index = step_size;
|
||||
}
|
||||
|
||||
/* tri-Ace Aska ADPCM, Yamaha ADPCM-B with headered frames (reversed from Android SO's .so)
|
||||
* implements table with if-else/switchs too but that's too goofy */
|
||||
void decode_aska(VGMSTREAMCHANNEL * stream, sample_t * outbuf, int channelspacing, int32_t first_sample, int32_t samples_to_do, int channel) {
|
||||
int i, sample_count = 0, num_frame;
|
||||
int16_t out_sample;
|
||||
int32_t hist1 = stream->adpcm_history1_32;
|
||||
int step_size = stream->adpcm_step_index;
|
||||
|
||||
/* external interleave */
|
||||
int block_samples = (0x40 - 0x04*channelspacing) * 2 / channelspacing;
|
||||
num_frame = first_sample / block_samples;
|
||||
first_sample = first_sample % block_samples;
|
||||
|
||||
/* header (hist+step) */
|
||||
if (first_sample == 0) {
|
||||
off_t header_offset = stream->offset + 0x40*num_frame + 0x04*channel;
|
||||
|
||||
hist1 = read_16bitLE(header_offset+0x00,stream->streamfile);
|
||||
step_size = read_16bitLE(header_offset+0x02,stream->streamfile);
|
||||
/* in most files 1st frame has step 0 but it seems ok and accounted for */
|
||||
//if (step_size < 0x7f) step_size = 0x7f;
|
||||
//else if (step_size > 0x6000) step_size = 0x6000;
|
||||
}
|
||||
|
||||
/* decode nibbles (layout: varies) */
|
||||
for (i = first_sample; i < first_sample + samples_to_do; i++) {
|
||||
off_t byte_offset = (channelspacing == 2) ?
|
||||
(stream->offset + 0x40*num_frame + 0x04*channelspacing) + i : /* stereo: one nibble per channel */
|
||||
(stream->offset + 0x40*num_frame + 0x04*channelspacing) + i/2; /* mono: consecutive nibbles */
|
||||
int nibble_shift = (channelspacing == 2) ?
|
||||
(!(channel&1) ? 0:4) :
|
||||
(!(i&1) ? 0:4); /* even = low, odd = high */
|
||||
|
||||
yamaha_adpcmb_expand_nibble(stream, byte_offset, nibble_shift, &hist1, &step_size, &out_sample);
|
||||
outbuf[sample_count] = out_sample;
|
||||
sample_count += channelspacing;
|
||||
}
|
||||
|
||||
stream->adpcm_history1_32 = hist1;
|
||||
stream->adpcm_step_index = step_size;
|
||||
}
|
||||
|
||||
|
||||
/* NXAP ADPCM, Yamaha ADPCM-B with weird headered frames */
|
||||
void decode_nxap(VGMSTREAMCHANNEL * stream, sample_t * outbuf, int channelspacing, int32_t first_sample, int32_t samples_to_do) {
|
||||
int i, sample_count = 0, num_frame;
|
||||
int32_t hist1 = stream->adpcm_history1_32;
|
||||
int step_size = stream->adpcm_step_index;
|
||||
int16_t out_sample;
|
||||
|
||||
/* external interleave, mono */
|
||||
int block_samples = (0x40 - 0x4) * 2;
|
||||
num_frame = first_sample / block_samples;
|
||||
first_sample = first_sample % block_samples;
|
||||
|
||||
/* header (hist+step) */
|
||||
if (first_sample == 0) {
|
||||
off_t header_offset = stream->offset + 0x40*num_frame;
|
||||
|
||||
hist1 = read_s16le(header_offset+0x00,stream->streamfile);
|
||||
step_size = read_u16le(header_offset+0x02,stream->streamfile) >> 1; /* remove lower bit, also note unsignedness */
|
||||
if (step_size < 0x7f) step_size = 0x7f;
|
||||
else if (step_size > 0x6000) step_size = 0x6000;
|
||||
/* step's lower bit is hist1 sign (useless), and code doesn't seem to do anything useful with it? */
|
||||
}
|
||||
|
||||
/* decode nibbles (layout: all nibbles from one channel) */
|
||||
for (i = first_sample; i < first_sample + samples_to_do; i++) {
|
||||
off_t byte_offset = (stream->offset + 0x40*num_frame + 0x04) + i/2;
|
||||
int nibble_shift = (i&1?0:4);
|
||||
|
||||
yamaha_adpcmb_expand_nibble(stream, byte_offset, nibble_shift, &hist1, &step_size, &out_sample);
|
||||
outbuf[sample_count] = out_sample;
|
||||
sample_count += channelspacing;
|
||||
}
|
||||
|
||||
stream->adpcm_history1_32 = hist1;
|
||||
stream->adpcm_step_index = step_size;
|
||||
}
|
||||
|
||||
size_t yamaha_bytes_to_samples(size_t bytes, int channels) {
|
||||
if (channels <= 0) return 0;
|
||||
/* 2 samples per byte (2 nibbles) in stereo or mono config */
|
||||
return bytes * 2 / channels;
|
||||
}
|
||||
|
||||
size_t aska_bytes_to_samples(size_t bytes, int channels) {
|
||||
int block_align = 0x40;
|
||||
if (channels <= 0) return 0;
|
||||
return (bytes / block_align) * (block_align - 0x04*channels) * 2 / channels
|
||||
+ ((bytes % block_align) ? ((bytes % block_align) - 0x04*channels) * 2 / channels : 0);
|
||||
}
|
||||
|
@ -4,40 +4,51 @@
|
||||
/* FWSE - Capcom's MT Framework V1.x sound file */
|
||||
VGMSTREAM *init_vgmstream_fwse(STREAMFILE *streamFile) {
|
||||
VGMSTREAM *vgmstream = NULL;
|
||||
uint32_t version, /*file_size,*/ buffer_offset,
|
||||
channel_count, sample_count, sample_rate;
|
||||
uint32_t version, /*data_size,*/ buffer_offset;
|
||||
int channel_count, loop_flag, sample_count, sample_rate, loop_start, loop_end;
|
||||
|
||||
|
||||
/* checks*/
|
||||
/* .fwse: header id, no apparent extension in bigfiles */
|
||||
if (!check_extensions(streamFile,"fwse"))
|
||||
goto fail;
|
||||
|
||||
if ((read_32bitLE(0x00,streamFile)) != 0x45535746)
|
||||
if ((read_32bitBE(0x00,streamFile)) != 0x46575345) /* "FWSE" */
|
||||
goto fail;
|
||||
|
||||
version = read_32bitLE(0x04,streamFile);
|
||||
|
||||
if (version != 2)
|
||||
/* v2: Resident Evil 5 (PC)
|
||||
* v3: Ace Attourney: Dual Destinies (Android) */
|
||||
if (version != 2 && version != 3)
|
||||
goto fail;
|
||||
|
||||
//file_size = read_32bitLE(0x08,streamFile);
|
||||
//data_size = read_32bitLE(0x08,streamFile);
|
||||
buffer_offset = read_32bitLE(0x0C,streamFile);
|
||||
channel_count = read_32bitLE(0x10,streamFile);
|
||||
|
||||
if (channel_count > 1)
|
||||
if (channel_count > 2)
|
||||
goto fail;
|
||||
|
||||
sample_count = read_32bitLE(0x14,streamFile);
|
||||
sample_rate = read_32bitLE(0x18,streamFile);
|
||||
loop_start = read_32bitLE(0x20,streamFile);
|
||||
loop_end = read_32bitLE(0x24,streamFile);
|
||||
loop_flag = (loop_start != -1);
|
||||
/* 0x28: some kind of setup? */
|
||||
/* 0x40: some kind of seek table with ADPCM hist/steps? */
|
||||
|
||||
|
||||
/* build the VGMSTREAM */
|
||||
vgmstream = allocate_vgmstream(channel_count, 0);
|
||||
vgmstream = allocate_vgmstream(channel_count, loop_flag);
|
||||
if (!vgmstream) goto fail;
|
||||
|
||||
vgmstream->meta_type = meta_FWSE;
|
||||
vgmstream->sample_rate = sample_rate;
|
||||
vgmstream->num_samples = sample_count;
|
||||
vgmstream->loop_start_sample = loop_start;
|
||||
vgmstream->loop_end_sample = loop_end;
|
||||
vgmstream->coding_type = coding_MTF_IMA;
|
||||
vgmstream->layout_type = channel_count == 1 ? layout_none : layout_interleave;
|
||||
vgmstream->interleave_block_size = 1;
|
||||
vgmstream->layout_type = layout_none;
|
||||
|
||||
|
||||
if (!vgmstream_open_stream(vgmstream,streamFile,buffer_offset))
|
||||
|
@ -1,49 +1,48 @@
|
||||
#include "meta.h"
|
||||
#include "../coding/coding.h"
|
||||
|
||||
/* NXAP - Nex Entertainment header [Time Crisis 4 (PS3), Time Crisis Razing Storm (PS3)] */
|
||||
VGMSTREAM * init_vgmstream_nxap(STREAMFILE *streamFile) {
|
||||
VGMSTREAM * vgmstream = NULL;
|
||||
off_t start_offset;
|
||||
int loop_flag, channel_count;
|
||||
|
||||
|
||||
/* checks */
|
||||
if (!check_extensions(streamFile, "adp"))
|
||||
goto fail;
|
||||
if (read_32bitBE(0x00,streamFile) != 0x4E584150) /* "NXAP" */
|
||||
goto fail;
|
||||
if (read_32bitLE(0x14,streamFile) != 0x40 || /* expected frame size? */
|
||||
read_32bitLE(0x18,streamFile) != 0x40) /* expected interleave? */
|
||||
goto fail;
|
||||
|
||||
start_offset = read_32bitLE(0x04,streamFile);
|
||||
channel_count = read_32bitLE(0x0c,streamFile);
|
||||
loop_flag = 0; //(read_32bitLE(0x24,streamFile) > 0); //todo
|
||||
|
||||
|
||||
/* build the VGMSTREAM */
|
||||
vgmstream = allocate_vgmstream(channel_count, loop_flag);
|
||||
if (!vgmstream) goto fail;
|
||||
|
||||
vgmstream->sample_rate = read_32bitLE(0x10, streamFile);
|
||||
vgmstream->num_samples = read_32bitLE(0x1c,streamFile) * (0x40-0x04)*2 / channel_count; /* number of frames */
|
||||
|
||||
/* unknown loop format, also 0x28/2c values seem related */
|
||||
//vgmstream->loop_start_sample = read_32bitLE(0x20,streamFile) * (0x40-0x04)*2 / channel_count;
|
||||
//vgmstream->loop_end_sample = read_32bitLE(0x24,streamFile) * (0x40-0x04)*2 / channel_count;
|
||||
//vgmstream->loop_end_sample = vgmstream->loop_start_sample + vgmstream->loop_end_sample;
|
||||
|
||||
vgmstream->meta_type = meta_NXAP;
|
||||
vgmstream->coding_type = coding_NXAP;
|
||||
vgmstream->layout_type = layout_interleave;
|
||||
vgmstream->interleave_block_size = 0x40;
|
||||
|
||||
if (!vgmstream_open_stream(vgmstream,streamFile,start_offset))
|
||||
goto fail;
|
||||
return vgmstream;
|
||||
|
||||
fail:
|
||||
close_vgmstream(vgmstream);
|
||||
return NULL;
|
||||
}
|
||||
#include "meta.h"
|
||||
#include "../coding/coding.h"
|
||||
|
||||
/* NXAP - Nex Entertainment header [Time Crisis 4 (PS3), Time Crisis Razing Storm (PS3)] */
|
||||
VGMSTREAM * init_vgmstream_nxap(STREAMFILE *streamFile) {
|
||||
VGMSTREAM * vgmstream = NULL;
|
||||
off_t start_offset;
|
||||
int loop_flag, channel_count;
|
||||
|
||||
|
||||
/* checks */
|
||||
if (!check_extensions(streamFile, "adp"))
|
||||
goto fail;
|
||||
if (read_32bitBE(0x00,streamFile) != 0x4E584150) /* "NXAP" */
|
||||
goto fail;
|
||||
if (read_32bitLE(0x14,streamFile) != 0x40 || /* expected frame size? */
|
||||
read_32bitLE(0x18,streamFile) != 0x40) /* expected interleave? */
|
||||
goto fail;
|
||||
|
||||
start_offset = read_32bitLE(0x04,streamFile);
|
||||
channel_count = read_32bitLE(0x0c,streamFile);
|
||||
loop_flag = (read_32bitLE(0x24,streamFile) > 0);
|
||||
|
||||
|
||||
/* build the VGMSTREAM */
|
||||
vgmstream = allocate_vgmstream(channel_count, loop_flag);
|
||||
if (!vgmstream) goto fail;
|
||||
|
||||
vgmstream->sample_rate = read_32bitLE(0x10, streamFile);
|
||||
vgmstream->num_samples = read_32bitLE(0x1c,streamFile) * (0x40-0x04)*2 / channel_count; /* number of frames */
|
||||
|
||||
/* in channel blocks, also 0x28/2c values seem related */
|
||||
vgmstream->loop_start_sample = read_32bitLE(0x20,streamFile) * (0x40-0x04)*2;
|
||||
vgmstream->loop_end_sample = read_32bitLE(0x24,streamFile) * (0x40-0x04)*2;
|
||||
|
||||
vgmstream->meta_type = meta_NXAP;
|
||||
vgmstream->coding_type = coding_NXAP;
|
||||
vgmstream->layout_type = layout_interleave;
|
||||
vgmstream->interleave_block_size = 0x40;
|
||||
|
||||
if (!vgmstream_open_stream(vgmstream,streamFile,start_offset))
|
||||
goto fail;
|
||||
return vgmstream;
|
||||
|
||||
fail:
|
||||
close_vgmstream(vgmstream);
|
||||
return NULL;
|
||||
}
|
||||
|
@ -1846,8 +1846,10 @@ void decode_vgmstream(VGMSTREAM * vgmstream, int samples_written, int samples_to
|
||||
break;
|
||||
case coding_MTF_IMA:
|
||||
for (ch = 0; ch < vgmstream->channels; ch++) {
|
||||
int is_stereo = (vgmstream->channels > 1);
|
||||
decode_mtf_ima(&vgmstream->ch[ch],buffer+samples_written*vgmstream->channels+ch,
|
||||
vgmstream->channels,vgmstream->samples_into_block,samples_to_do);
|
||||
vgmstream->channels,vgmstream->samples_into_block,samples_to_do, ch,
|
||||
is_stereo);
|
||||
}
|
||||
break;
|
||||
case coding_3DS_IMA:
|
||||
|
Loading…
x
Reference in New Issue
Block a user