mirror of
https://github.com/fail0verflow/hbc.git
synced 2024-11-17 23:29:21 +01:00
354 lines
7.4 KiB
C
354 lines
7.4 KiB
C
|
#include <stdio.h>
|
||
|
#include <stdint.h>
|
||
|
#include <string.h>
|
||
|
#include <stdlib.h>
|
||
|
#include <math.h>
|
||
|
|
||
|
#if 0
|
||
|
uint16_t deftbl[16] = {
|
||
|
2048, 0,
|
||
|
4096, -2048,
|
||
|
0, 0,
|
||
|
1536, 512,
|
||
|
1920, 0,
|
||
|
2176, 0,
|
||
|
3680, -1664,
|
||
|
3136, -1856
|
||
|
};
|
||
|
#endif
|
||
|
|
||
|
#if 0
|
||
|
int16_t deftbl[16] = {
|
||
|
2048, 0,
|
||
|
0,0,
|
||
|
0,0,
|
||
|
0,0,
|
||
|
0,0,
|
||
|
0,0,
|
||
|
0,0,
|
||
|
0,0,
|
||
|
};
|
||
|
#endif
|
||
|
|
||
|
int16_t deftbl[16] = {
|
||
|
674,1040,
|
||
|
3598,-1738,
|
||
|
2270,-583,
|
||
|
3967,-1969,
|
||
|
1516,381,
|
||
|
3453, -1468,
|
||
|
2606, -617,
|
||
|
3795, -1759,
|
||
|
};
|
||
|
|
||
|
|
||
|
typedef struct {
|
||
|
|
||
|
uint32_t fourcc;
|
||
|
uint32_t unk;
|
||
|
uint32_t filesize;
|
||
|
uint16_t unk1,unk2;
|
||
|
uint32_t infooff;
|
||
|
uint32_t infosize;
|
||
|
uint32_t dataoff;
|
||
|
uint32_t datasize;
|
||
|
} bnshdr;
|
||
|
|
||
|
typedef struct {
|
||
|
uint32_t fourcc;
|
||
|
uint32_t size;
|
||
|
uint16_t looped;
|
||
|
uint16_t unk2;
|
||
|
uint16_t srate;
|
||
|
uint16_t unk3;
|
||
|
uint32_t looppoint;
|
||
|
uint32_t samples;
|
||
|
uint32_t unknown1[6];
|
||
|
uint32_t start1;
|
||
|
uint32_t start2;
|
||
|
uint32_t unknown2[2];
|
||
|
int16_t tbl1[16];
|
||
|
uint16_t unka1[8];
|
||
|
int16_t tbl2[16];
|
||
|
uint16_t unka2[8];
|
||
|
} bnsinfo;
|
||
|
|
||
|
float tables[2][16];
|
||
|
|
||
|
typedef struct {
|
||
|
uint32_t fourcc;
|
||
|
uint32_t size;
|
||
|
} bnsdatahdr;
|
||
|
|
||
|
#define SWAB16(x) ((((x)>>8)&0xFF) | (((x)&0xFF)<<8))
|
||
|
#define SWAB32(x) ((SWAB16((x)&0xFFFF)<<16)|(SWAB16(((x)>>16)&0xFFFF)))
|
||
|
|
||
|
#define ISWAB16(x) x=SWAB16(x)
|
||
|
#define ISWAB32(x) x=SWAB32(x)
|
||
|
|
||
|
typedef struct {
|
||
|
int16_t l;
|
||
|
int16_t r;
|
||
|
} sample;
|
||
|
|
||
|
int16_t lsamps[2][2] = {{0,0},{0,0}};
|
||
|
int16_t rlsamps[2][2] = {{0,0},{0,0}};
|
||
|
|
||
|
#define CLAMP(a,min,max) (((a)>(max))?(max):(((a)<(min))?(min):(a)))
|
||
|
|
||
|
void unpack_adpcm(int idx, int16_t *table, uint8_t *data, int16_t *outbuf)
|
||
|
{
|
||
|
|
||
|
int32_t index = (data[0] >> 4) & 0x7; //highest bit of byte is ignored
|
||
|
uint32_t exponent = 28 - (data[0] & 0xf);
|
||
|
int32_t factor1 = table[2*index];
|
||
|
int32_t factor2 = table[2*index + 1];
|
||
|
int i;
|
||
|
int32_t sample;
|
||
|
for(i=0;i<14;i++) {
|
||
|
sample = data[1+(i/2)];
|
||
|
if(!(i&1)) {
|
||
|
sample = (sample&0xf0)<<24;
|
||
|
} else {
|
||
|
sample = (sample)<<28;
|
||
|
}
|
||
|
sample = ((lsamps[idx][1]*factor1 + lsamps[idx][0]*factor2)>>11) + (sample>>exponent);
|
||
|
if(sample>32767) sample=32767;
|
||
|
if(sample<-32768) sample=-32768;
|
||
|
if(abs(sample)>20000) printf("dammit %d\n",sample);
|
||
|
outbuf[i] = sample;
|
||
|
lsamps[idx][0] = lsamps[idx][1];
|
||
|
lsamps[idx][1] = outbuf[i];
|
||
|
}
|
||
|
}
|
||
|
|
||
|
uint8_t findexp(float residual, uint8_t *nybble)
|
||
|
{
|
||
|
uint8_t exp = 0;
|
||
|
|
||
|
while((residual > 7.5f) || (residual < -8.5f)) {
|
||
|
exp++;
|
||
|
residual /= 2;
|
||
|
}
|
||
|
if(nybble)
|
||
|
*nybble = CLAMP((int16_t)floor(residual),-8,7);
|
||
|
return exp;
|
||
|
}
|
||
|
|
||
|
uint8_t determine_std_exponent(int idx, int16_t *table, int index, int16_t *inbuf)
|
||
|
{
|
||
|
int32_t maxres = 0;
|
||
|
int32_t factor1 = table[2*index];
|
||
|
int32_t factor2 = table[2*index + 1];
|
||
|
int32_t predictor;
|
||
|
int32_t residual;
|
||
|
int i;
|
||
|
int16_t elsamps[2];
|
||
|
memcpy(elsamps,rlsamps[idx],sizeof(int16_t)*2);
|
||
|
for(i=0;i<14;i++) {
|
||
|
predictor = (elsamps[1]*factor1 + elsamps[0]*factor2)/2048;
|
||
|
residual = inbuf[i] - predictor;
|
||
|
if(residual > maxres) maxres = residual;
|
||
|
elsamps[0] = elsamps[1];
|
||
|
elsamps[1] = inbuf[i];
|
||
|
}
|
||
|
return findexp(maxres,NULL);
|
||
|
}
|
||
|
|
||
|
float compress_adpcm(int idx, int16_t *table, uint8_t tblidx, uint8_t *data, int16_t *inbuf, int16_t *lsamps) {
|
||
|
|
||
|
int32_t factor1 = table[2*tblidx];
|
||
|
int32_t factor2 = table[2*tblidx + 1];
|
||
|
int32_t predictor;
|
||
|
int32_t residual;
|
||
|
uint8_t exp;
|
||
|
int8_t nybble;
|
||
|
int i;
|
||
|
float error = 0;
|
||
|
|
||
|
exp = determine_std_exponent(idx, table, tblidx, inbuf);
|
||
|
|
||
|
while(exp<=15) {
|
||
|
memcpy(lsamps,rlsamps[idx],sizeof(int16_t)*2);
|
||
|
data[0] = exp | tblidx<<4;
|
||
|
error = 0;
|
||
|
for(i=0;i<14;i++) {
|
||
|
predictor = (lsamps[1]*factor1 + lsamps[0]*factor2)>>11;
|
||
|
residual = inbuf[i] - predictor;
|
||
|
residual = residual>>exp;
|
||
|
if((residual > 7) || (residual < -8)) {
|
||
|
exp++;
|
||
|
break;
|
||
|
}
|
||
|
nybble = CLAMP(residual,-8,7);
|
||
|
if(i&1) {
|
||
|
data[i/2+1] |= nybble&0xf;
|
||
|
} else {
|
||
|
data[i/2+1] = nybble<<4;
|
||
|
}
|
||
|
predictor += nybble<<exp;
|
||
|
lsamps[0] = lsamps[1];
|
||
|
lsamps[1] = CLAMP(predictor,-32768,32767);
|
||
|
error += powf(lsamps[1] - inbuf[i],2.0f);
|
||
|
}
|
||
|
if(i == 14) break;
|
||
|
}
|
||
|
return error;
|
||
|
}
|
||
|
|
||
|
|
||
|
void repack_adpcm(int idx, int16_t *table, uint8_t *data, int16_t *inbuf)
|
||
|
{
|
||
|
uint8_t tblidx;
|
||
|
uint8_t testdata[8];
|
||
|
int16_t tlsamps[2];
|
||
|
int16_t blsamps[2];
|
||
|
float error;
|
||
|
float besterror = 99999999.0f;
|
||
|
for(tblidx = 0; tblidx < 8; tblidx++) {
|
||
|
error = compress_adpcm(idx, table, tblidx, testdata, inbuf, tlsamps);
|
||
|
if(error < besterror) {
|
||
|
besterror = error;
|
||
|
memcpy(data, testdata, 8);
|
||
|
memcpy(blsamps, tlsamps, sizeof(int16_t)*2);
|
||
|
}
|
||
|
}
|
||
|
memcpy(rlsamps[idx], blsamps, sizeof(int16_t)*2);
|
||
|
|
||
|
}
|
||
|
|
||
|
int main(int argc, char **argv)
|
||
|
{
|
||
|
|
||
|
FILE *f;
|
||
|
FILE *f2 = NULL;
|
||
|
FILE *fo;
|
||
|
int i,j;
|
||
|
int16_t sampbuf[14];
|
||
|
|
||
|
bnshdr hdr;
|
||
|
bnsinfo info;
|
||
|
bnsdatahdr datahdr;
|
||
|
uint8_t *databuf;
|
||
|
uint8_t *data1;
|
||
|
uint8_t *data2;
|
||
|
sample *datain;
|
||
|
|
||
|
int samples;
|
||
|
int loop_pt = 0;
|
||
|
int blocks;
|
||
|
int separated_loop = 0;
|
||
|
|
||
|
if(argc > 4 && (atoi(argv[3]) == 1))
|
||
|
separated_loop = 1;
|
||
|
|
||
|
f = fopen(argv[1],"r");
|
||
|
fo = fopen(argv[2],"w");
|
||
|
|
||
|
fseek(f,0,SEEK_END);
|
||
|
|
||
|
samples = ftell(f)/(sizeof(uint16_t)*2);
|
||
|
|
||
|
if(separated_loop) {
|
||
|
f2 = fopen(argv[4],"r");
|
||
|
fseek(f2,0,SEEK_END);
|
||
|
loop_pt = samples;
|
||
|
samples += ftell(f2)/(sizeof(uint16_t)*2);
|
||
|
}
|
||
|
|
||
|
blocks = (samples+13)/14;
|
||
|
|
||
|
memset(&hdr,0,sizeof(hdr));
|
||
|
memset(&info,0,sizeof(info));
|
||
|
memset(&datahdr,0,sizeof(datahdr));
|
||
|
|
||
|
hdr.fourcc = 0x20534e42;
|
||
|
hdr.unk = SWAB32(0xfeff0100);
|
||
|
hdr.filesize = SWAB32(blocks * 16 + sizeof(hdr) + sizeof(info) + sizeof(datahdr));
|
||
|
hdr.unk1 = SWAB16(32);
|
||
|
hdr.unk2 = SWAB16(2);
|
||
|
hdr.infooff = SWAB32(sizeof(hdr));
|
||
|
hdr.infosize = SWAB32(sizeof(info));
|
||
|
hdr.dataoff = SWAB32(sizeof(hdr) + sizeof(info));
|
||
|
hdr.datasize = SWAB32(sizeof(datahdr) + blocks * 16);
|
||
|
|
||
|
info.fourcc = 0x4f464e49;
|
||
|
info.size = SWAB32(sizeof(info));
|
||
|
info.srate = SWAB16(32000);
|
||
|
if(argc > 3 && (atoi(argv[3]) == 1))
|
||
|
info.looped = SWAB16(1);
|
||
|
info.unk2 = SWAB16(0x200);
|
||
|
info.looppoint = SWAB32(loop_pt);
|
||
|
info.samples = SWAB32(samples);
|
||
|
info.unknown1[0] = SWAB32(0x18);
|
||
|
info.unknown1[1] = SWAB32(0x00);
|
||
|
info.unknown1[2] = SWAB32(0x20);
|
||
|
info.unknown1[3] = SWAB32(0x2c);
|
||
|
info.unknown1[4] = SWAB32(0x00);
|
||
|
info.unknown1[5] = SWAB32(0x38);
|
||
|
info.unknown2[0] = SWAB32(0x68);
|
||
|
info.unknown2[1] = SWAB32(0x00);
|
||
|
info.start1 = SWAB32(0);
|
||
|
info.start2 = SWAB32(blocks * 8);
|
||
|
for(i=0;i<16;i++) {
|
||
|
info.tbl1[i] = SWAB16((int16_t)(deftbl[i]));
|
||
|
info.tbl2[i] = SWAB16((int16_t)(deftbl[i]));
|
||
|
}
|
||
|
|
||
|
datahdr.fourcc = 0x41544144;
|
||
|
datahdr.size = SWAB32(blocks * 16);
|
||
|
|
||
|
fwrite(&hdr,sizeof(hdr),1,fo);
|
||
|
fwrite(&info,sizeof(info),1,fo);
|
||
|
fwrite(&datahdr,sizeof(datahdr),1,fo);
|
||
|
|
||
|
|
||
|
datain = malloc(sizeof(uint16_t)*2*blocks*14);
|
||
|
memset(datain,0,sizeof(uint16_t)*2*blocks*14);
|
||
|
|
||
|
databuf = malloc(blocks * 16);
|
||
|
data1 = databuf;
|
||
|
data2 = databuf + blocks * 8;
|
||
|
|
||
|
if(separated_loop) {
|
||
|
fseek(f,0,SEEK_SET);
|
||
|
fread(datain,sizeof(uint16_t)*2,loop_pt,f);
|
||
|
fseek(f2,0,SEEK_SET);
|
||
|
fread(&datain[loop_pt],sizeof(uint16_t)*2,samples-loop_pt,f2);
|
||
|
fclose(f);
|
||
|
fclose(f2);
|
||
|
} else {
|
||
|
fseek(f,0,SEEK_SET);
|
||
|
fread(datain,sizeof(uint16_t)*2,samples,f);
|
||
|
fclose(f);
|
||
|
}
|
||
|
|
||
|
printf("Samples: 0x%x\n",samples);
|
||
|
printf("Blocks: 0x%x Size ADPCM: 0x%x Size PCM: 0x%x\n",blocks,blocks*8,blocks*14);
|
||
|
if(separated_loop)
|
||
|
printf("Loop point: 0x%x samples\n",loop_pt);
|
||
|
|
||
|
for(i=0;i<blocks;i++) {
|
||
|
//printf("Block %d\n",i);
|
||
|
for(j=0;j<14;j++) {
|
||
|
sampbuf[j] = datain[i*14+j].l;
|
||
|
}
|
||
|
repack_adpcm(0,deftbl,data1,sampbuf);
|
||
|
//unpack_adpcm(0,deftbl,data1,sampbuf);
|
||
|
for(j=0;j<14;j++) {
|
||
|
sampbuf[j] = datain[i*14+j].r;
|
||
|
}
|
||
|
repack_adpcm(1,deftbl,data2,sampbuf);
|
||
|
//unpack_adpcm(1,deftbl,data2,sampbuf);
|
||
|
data1 += 8;
|
||
|
data2 += 8;
|
||
|
}
|
||
|
|
||
|
fwrite(databuf, blocks*16, 1, fo);
|
||
|
|
||
|
fclose(fo);
|
||
|
|
||
|
free(datain); free(databuf);
|
||
|
return 0;
|
||
|
}
|