mirror of https://git.ffmpeg.org/ffmpeg.git
Prepare to remove more duplicated code
Based on a patch by Matthieu Castet Originally committed as revision 9576 to svn://svn.ffmpeg.org/ffmpeg/trunk
This commit is contained in:
parent
a562e2e651
commit
153696a663
|
@ -57,6 +57,7 @@
|
||||||
#include "bitstream.h"
|
#include "bitstream.h"
|
||||||
|
|
||||||
#define ALAC_EXTRADATA_SIZE 36
|
#define ALAC_EXTRADATA_SIZE 36
|
||||||
|
#define MAX_CHANNELS 2
|
||||||
|
|
||||||
typedef struct {
|
typedef struct {
|
||||||
|
|
||||||
|
@ -71,11 +72,9 @@ typedef struct {
|
||||||
int bytespersample;
|
int bytespersample;
|
||||||
|
|
||||||
/* buffers */
|
/* buffers */
|
||||||
int32_t *predicterror_buffer_a;
|
int32_t *predicterror_buffer[MAX_CHANNELS];
|
||||||
int32_t *predicterror_buffer_b;
|
|
||||||
|
|
||||||
int32_t *outputsamples_buffer_a;
|
int32_t *outputsamples_buffer[MAX_CHANNELS];
|
||||||
int32_t *outputsamples_buffer_b;
|
|
||||||
|
|
||||||
/* stuff from setinfo */
|
/* stuff from setinfo */
|
||||||
uint32_t setinfo_max_samples_per_frame; /* 0x1000 = 4096 */ /* max samples per frame? */
|
uint32_t setinfo_max_samples_per_frame; /* 0x1000 = 4096 */ /* max samples per frame? */
|
||||||
|
@ -95,11 +94,14 @@ typedef struct {
|
||||||
|
|
||||||
static void allocate_buffers(ALACContext *alac)
|
static void allocate_buffers(ALACContext *alac)
|
||||||
{
|
{
|
||||||
alac->predicterror_buffer_a = av_malloc(alac->setinfo_max_samples_per_frame * 4);
|
int chan;
|
||||||
alac->predicterror_buffer_b = av_malloc(alac->setinfo_max_samples_per_frame * 4);
|
for (chan = 0; chan < MAX_CHANNELS; chan++) {
|
||||||
|
alac->predicterror_buffer[chan] =
|
||||||
|
av_malloc(alac->setinfo_max_samples_per_frame * 4);
|
||||||
|
|
||||||
alac->outputsamples_buffer_a = av_malloc(alac->setinfo_max_samples_per_frame * 4);
|
alac->outputsamples_buffer[chan] =
|
||||||
alac->outputsamples_buffer_b = av_malloc(alac->setinfo_max_samples_per_frame * 4);
|
av_malloc(alac->setinfo_max_samples_per_frame * 4);
|
||||||
|
}
|
||||||
}
|
}
|
||||||
|
|
||||||
static int alac_set_info(ALACContext *alac)
|
static int alac_set_info(ALACContext *alac)
|
||||||
|
@ -546,7 +548,7 @@ static int alac_decode_frame(AVCodecContext *avctx,
|
||||||
}
|
}
|
||||||
|
|
||||||
bastardized_rice_decompress(alac,
|
bastardized_rice_decompress(alac,
|
||||||
alac->predicterror_buffer_a,
|
alac->predicterror_buffer[0],
|
||||||
outputsamples,
|
outputsamples,
|
||||||
readsamplesize,
|
readsamplesize,
|
||||||
alac->setinfo_rice_initialhistory,
|
alac->setinfo_rice_initialhistory,
|
||||||
|
@ -556,8 +558,8 @@ static int alac_decode_frame(AVCodecContext *avctx,
|
||||||
|
|
||||||
if (prediction_type == 0) {
|
if (prediction_type == 0) {
|
||||||
/* adaptive fir */
|
/* adaptive fir */
|
||||||
predictor_decompress_fir_adapt(alac->predicterror_buffer_a,
|
predictor_decompress_fir_adapt(alac->predicterror_buffer[0],
|
||||||
alac->outputsamples_buffer_a,
|
alac->outputsamples_buffer[0],
|
||||||
outputsamples,
|
outputsamples,
|
||||||
readsamplesize,
|
readsamplesize,
|
||||||
predictor_coef_table,
|
predictor_coef_table,
|
||||||
|
@ -582,7 +584,7 @@ static int alac_decode_frame(AVCodecContext *avctx,
|
||||||
|
|
||||||
audiobits = SIGN_EXTENDED32(audiobits, readsamplesize);
|
audiobits = SIGN_EXTENDED32(audiobits, readsamplesize);
|
||||||
|
|
||||||
alac->outputsamples_buffer_a[i] = audiobits;
|
alac->outputsamples_buffer[0][i] = audiobits;
|
||||||
}
|
}
|
||||||
} else {
|
} else {
|
||||||
int i;
|
int i;
|
||||||
|
@ -597,7 +599,7 @@ static int alac_decode_frame(AVCodecContext *avctx,
|
||||||
|
|
||||||
audiobits |= get_bits(&alac->gb, readsamplesize - 16);
|
audiobits |= get_bits(&alac->gb, readsamplesize - 16);
|
||||||
|
|
||||||
alac->outputsamples_buffer_a[i] = audiobits;
|
alac->outputsamples_buffer[0][i] = audiobits;
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
/* wasted_bytes = 0; // unused */
|
/* wasted_bytes = 0; // unused */
|
||||||
|
@ -607,7 +609,7 @@ static int alac_decode_frame(AVCodecContext *avctx,
|
||||||
case 16: {
|
case 16: {
|
||||||
int i;
|
int i;
|
||||||
for (i = 0; i < outputsamples; i++) {
|
for (i = 0; i < outputsamples; i++) {
|
||||||
int16_t sample = alac->outputsamples_buffer_a[i];
|
int16_t sample = alac->outputsamples_buffer[0][i];
|
||||||
((int16_t*)outbuffer)[i * alac->numchannels] = sample;
|
((int16_t*)outbuffer)[i * alac->numchannels] = sample;
|
||||||
}
|
}
|
||||||
break;
|
break;
|
||||||
|
@ -677,7 +679,7 @@ static int alac_decode_frame(AVCodecContext *avctx,
|
||||||
|
|
||||||
/* channel 1 */
|
/* channel 1 */
|
||||||
bastardized_rice_decompress(alac,
|
bastardized_rice_decompress(alac,
|
||||||
alac->predicterror_buffer_a,
|
alac->predicterror_buffer[0],
|
||||||
outputsamples,
|
outputsamples,
|
||||||
readsamplesize,
|
readsamplesize,
|
||||||
alac->setinfo_rice_initialhistory,
|
alac->setinfo_rice_initialhistory,
|
||||||
|
@ -687,8 +689,8 @@ static int alac_decode_frame(AVCodecContext *avctx,
|
||||||
|
|
||||||
if (prediction_type_a == 0) {
|
if (prediction_type_a == 0) {
|
||||||
/* adaptive fir */
|
/* adaptive fir */
|
||||||
predictor_decompress_fir_adapt(alac->predicterror_buffer_a,
|
predictor_decompress_fir_adapt(alac->predicterror_buffer[0],
|
||||||
alac->outputsamples_buffer_a,
|
alac->outputsamples_buffer[0],
|
||||||
outputsamples,
|
outputsamples,
|
||||||
readsamplesize,
|
readsamplesize,
|
||||||
predictor_coef_table_a,
|
predictor_coef_table_a,
|
||||||
|
@ -701,7 +703,7 @@ static int alac_decode_frame(AVCodecContext *avctx,
|
||||||
|
|
||||||
/* channel 2 */
|
/* channel 2 */
|
||||||
bastardized_rice_decompress(alac,
|
bastardized_rice_decompress(alac,
|
||||||
alac->predicterror_buffer_b,
|
alac->predicterror_buffer[1],
|
||||||
outputsamples,
|
outputsamples,
|
||||||
readsamplesize,
|
readsamplesize,
|
||||||
alac->setinfo_rice_initialhistory,
|
alac->setinfo_rice_initialhistory,
|
||||||
|
@ -711,8 +713,8 @@ static int alac_decode_frame(AVCodecContext *avctx,
|
||||||
|
|
||||||
if (prediction_type_b == 0) {
|
if (prediction_type_b == 0) {
|
||||||
/* adaptive fir */
|
/* adaptive fir */
|
||||||
predictor_decompress_fir_adapt(alac->predicterror_buffer_b,
|
predictor_decompress_fir_adapt(alac->predicterror_buffer[1],
|
||||||
alac->outputsamples_buffer_b,
|
alac->outputsamples_buffer[1],
|
||||||
outputsamples,
|
outputsamples,
|
||||||
readsamplesize,
|
readsamplesize,
|
||||||
predictor_coef_table_b,
|
predictor_coef_table_b,
|
||||||
|
@ -734,8 +736,8 @@ static int alac_decode_frame(AVCodecContext *avctx,
|
||||||
audiobits_a = SIGN_EXTENDED32(audiobits_a, alac->setinfo_sample_size);
|
audiobits_a = SIGN_EXTENDED32(audiobits_a, alac->setinfo_sample_size);
|
||||||
audiobits_b = SIGN_EXTENDED32(audiobits_b, alac->setinfo_sample_size);
|
audiobits_b = SIGN_EXTENDED32(audiobits_b, alac->setinfo_sample_size);
|
||||||
|
|
||||||
alac->outputsamples_buffer_a[i] = audiobits_a;
|
alac->outputsamples_buffer[0][i] = audiobits_a;
|
||||||
alac->outputsamples_buffer_b[i] = audiobits_b;
|
alac->outputsamples_buffer[1][i] = audiobits_b;
|
||||||
}
|
}
|
||||||
} else {
|
} else {
|
||||||
int i;
|
int i;
|
||||||
|
@ -752,8 +754,8 @@ static int alac_decode_frame(AVCodecContext *avctx,
|
||||||
audiobits_b = audiobits_b >> (32 - alac->setinfo_sample_size);
|
audiobits_b = audiobits_b >> (32 - alac->setinfo_sample_size);
|
||||||
audiobits_b |= get_bits(&alac->gb, alac->setinfo_sample_size - 16);
|
audiobits_b |= get_bits(&alac->gb, alac->setinfo_sample_size - 16);
|
||||||
|
|
||||||
alac->outputsamples_buffer_a[i] = audiobits_a;
|
alac->outputsamples_buffer[0][i] = audiobits_a;
|
||||||
alac->outputsamples_buffer_b[i] = audiobits_b;
|
alac->outputsamples_buffer[1][i] = audiobits_b;
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
/* wasted_bytes = 0; */
|
/* wasted_bytes = 0; */
|
||||||
|
@ -763,8 +765,8 @@ static int alac_decode_frame(AVCodecContext *avctx,
|
||||||
|
|
||||||
switch(alac->setinfo_sample_size) {
|
switch(alac->setinfo_sample_size) {
|
||||||
case 16: {
|
case 16: {
|
||||||
deinterlace_16(alac->outputsamples_buffer_a,
|
deinterlace_16(alac->outputsamples_buffer[0],
|
||||||
alac->outputsamples_buffer_b,
|
alac->outputsamples_buffer[1],
|
||||||
(int16_t*)outbuffer,
|
(int16_t*)outbuffer,
|
||||||
alac->numchannels,
|
alac->numchannels,
|
||||||
outputsamples,
|
outputsamples,
|
||||||
|
@ -805,11 +807,11 @@ static int alac_decode_close(AVCodecContext *avctx)
|
||||||
{
|
{
|
||||||
ALACContext *alac = avctx->priv_data;
|
ALACContext *alac = avctx->priv_data;
|
||||||
|
|
||||||
av_free(alac->predicterror_buffer_a);
|
int chan;
|
||||||
av_free(alac->predicterror_buffer_b);
|
for (chan = 0; chan < MAX_CHANNELS; chan++) {
|
||||||
|
av_free(alac->predicterror_buffer[chan]);
|
||||||
av_free(alac->outputsamples_buffer_a);
|
av_free(alac->outputsamples_buffer[chan]);
|
||||||
av_free(alac->outputsamples_buffer_b);
|
}
|
||||||
|
|
||||||
return 0;
|
return 0;
|
||||||
}
|
}
|
||||||
|
|
Loading…
Reference in New Issue