Commit d910274f authored by Timothy B. Terriberry's avatar Timothy B. Terriberry Committed by Jean-Marc Valin
Browse files

Change CWRS indexing to use Pyramid VQ's magnitude ordering.

This lets us encode and decode directly from the pulse vector without an
 intermediate transformation.
This makes old streams undecodable.
Additionally, ncwrs_u32() has been sped up for large N by using the sliding
 recurrence from Mohorko et al.
ncwrs_u64 could be sped up in a similar manner, but would require a larger
 table of multiplicative inverses (or several 32x32->64 bit multiplies).
Note that U(N,M) is now everywhere 1/2 the value it used to be.
parent 7b0cb4ba
......@@ -139,6 +139,75 @@ int fits_in64(int _n, int _m)
}
}
#define MASK32 (0xFFFFFFFF)
/*INV_TABLE[i] holds the multiplicative inverse of (2*i-1) mod 2**32.*/
static const unsigned INV_TABLE[128]={
0x00000001,0xAAAAAAAB,0xCCCCCCCD,0xB6DB6DB7,
0x38E38E39,0xBA2E8BA3,0xC4EC4EC5,0xEEEEEEEF,
0xF0F0F0F1,0x286BCA1B,0x3CF3CF3D,0xE9BD37A7,
0xC28F5C29,0x684BDA13,0x4F72C235,0xBDEF7BDF,
0x3E0F83E1,0x8AF8AF8B,0x914C1BAD,0x96F96F97,
0xC18F9C19,0x2FA0BE83,0xA4FA4FA5,0x677D46CF,
0x1A1F58D1,0xFAFAFAFB,0x8C13521D,0x586FB587,
0xB823EE09,0xA08AD8F3,0xC10C9715,0xBEFBEFBF,
0xC0FC0FC1,0x07A44C6B,0xA33F128D,0xE327A977,
0xC7E3F1F9,0x962FC963,0x3F2B3885,0x613716AF,
0x781948B1,0x2B2E43DB,0xFCFCFCFD,0x6FD0EB67,
0xFA3F47E9,0xD2FD2FD3,0x3F4FD3F5,0xD4E25B9F,
0x5F02A3A1,0xBF5A814B,0x7C32B16D,0xD3431B57,
0xD8FD8FD9,0x8D28AC43,0xDA6C0965,0xDB195E8F,
0x0FDBC091,0x61F2A4BB,0xDCFDCFDD,0x46FDD947,
0x56BE69C9,0xEB2FDEB3,0x26E978D5,0xEFDFBF7F,
0x0FE03F81,0xC9484E2B,0xE133F84D,0xE1A8C537,
0x077975B9,0x70586723,0xCD29C245,0xFAA11E6F,
0x0FE3C071,0x08B51D9B,0x8CE2CABD,0xBF937F27,
0xA8FE53A9,0x592FE593,0x2C0685B5,0x2EB11B5F,
0xFCD1E361,0x451AB30B,0x72CFE72D,0xDB35A717,
0xFB74A399,0xE80BFA03,0x0D516325,0x1BCB564F,
0xE02E4851,0xD962AE7B,0x10F8ED9D,0x95AEDD07,
0xE9DC0589,0xA18A4473,0xEA53FA95,0xEE936F3F,
0x90948F41,0xEAFEAFEB,0x3D137E0D,0xEF46C0F7,
0x028C1979,0x791064E3,0xC04FEC05,0xE115062F,
0x32385831,0x6E68575B,0xA10D387D,0x6FECF2E7,
0x3FB47F69,0xED4BFB53,0x74FED775,0xDB43BB1F,
0x87654321,0x9BA144CB,0x478BBCED,0xBFB912D7,
0x1FDCD759,0x14B2A7C3,0xCB125CE5,0x437B2E0F,
0x10FEF011,0xD2B3183B,0x386CAB5D,0xEF6AC0C7,
0x0E64C149,0x9A020A33,0xE6B41C55,0xFEFEFEFF
};
/*Computes (_a*_b-_c)/(2*_d-1) when the quotient is known to be exact.
_a, _b, _c, and _d may be arbitrary so long as the arbitrary precision result
fits in 32 bits, but currently the table for multiplicative inverses is only
valid for _d<128.*/
static inline celt_uint32_t imusdiv32odd(celt_uint32_t _a,celt_uint32_t _b,
celt_uint32_t _c,celt_uint32_t _d){
return (_a*_b-_c)*INV_TABLE[_d]&MASK32;
}
/*Computes (_a*_b-_c)/_d when the quotient is known to be exact.
_d does not actually have to be even, but imusdiv32odd will be faster when
it's odd, so you should use that instead.
_a and _d are assumed to be small (e.g., _a*_d fits in 32 bits; currently the
table for multiplicative inverses is only valid for _d<256).
_b and _c may be arbitrary so long as the arbitrary precision reuslt fits in
32 bits.*/
static inline celt_uint32_t imusdiv32even(celt_uint32_t _a,celt_uint32_t _b,
celt_uint32_t _c,celt_uint32_t _d){
unsigned inv;
int mask;
int shift;
int one;
shift=EC_ILOG(_d^_d-1);
inv=INV_TABLE[_d-1>>shift];
shift--;
one=1<<shift;
mask=one-1;
return (_a*(_b>>shift)-(_c>>shift)+
(_a*(_b&mask)+one-(_c&mask)>>shift)-1)*inv&MASK32;
}
/*Computes the next row/column of any recurrence that obeys the relation
u[i][j]=u[i-1][j]+u[i][j-1]+u[i-1][j-1].
_ui0 is the base case for the new row/column.*/
......@@ -219,7 +288,7 @@ celt_uint64_t ncwrs_unext64(int _n,celt_uint64_t *_ui){
celt_uint64_t ui0;
celt_uint64_t ui1;
int j;
ret=ui0=2;
ret=ui0=1;
celt_assert(_n>=2);
j=1; do {
ui1=_ui[j]+_ui[j-1]+ui0;
......@@ -228,19 +297,50 @@ celt_uint64_t ncwrs_unext64(int _n,celt_uint64_t *_ui){
ret+=ui0;
} while (++j<_n);
_ui[j-1]=ui0;
return ret;
return ret<<=1;
}
/*Returns the number of ways of choosing _m elements from a set of size _n with
replacement when a sign bit is needed for each unique element.
On exit, _u will be initialized to column _m of U(n,m).*/
_u: On exit, _u[i] contains U(i+1,_m).*/
celt_uint32_t ncwrs_u32(int _n,int _m,celt_uint32_t *_u){
int k;
CELT_MEMSET(_u,0,_n);
if(_m<=0)return 1;
if(_n<=0)return 0;
for(k=1;k<_m;k++)unext32(_u,_n,2);
return ncwrs_unext32(_n,_u);
celt_uint32_t ret;
celt_uint32_t um2;
int k;
/*If _m==0, _u[] should be set to zero and the return should be 1.*/
celt_assert(_m>0);
/*We'll overflow our buffer unless _n>=2.*/
celt_assert(_n>=2);
um2=_u[0]=1;
if(_m<=6){
if(_m<2){
k=1;
do _u[k]=1;
while(++k<_n);
}
else{
k=1;
do _u[k]=(k<<1)+1;
while(++k<_n);
for(k=2;k<_m;k++)unext32(_u,_n,1);
}
}
else{
celt_uint32_t um1;
celt_uint32_t n2m1;
_u[1]=n2m1=um1=(_m<<1)-1;
for(k=2;k<_n;k++){
/*U(n,m) = ((2*n-1)*U(n,m-1)-U(n,m-2))/(m-1) + U(n,m-2)*/
_u[k]=um2=imusdiv32even(n2m1,um1,um2,k)+um2;
if(++k>=_n)break;
_u[k]=um1=imusdiv32odd(n2m1,um2,um1,k>>1)+um1;
}
}
ret=1;
k=1;
do ret+=_u[k];
while(++k<_n);
return ret<<1;
}
celt_uint64_t ncwrs_u64(int _n,int _m,celt_uint64_t *_u){
......@@ -248,171 +348,173 @@ celt_uint64_t ncwrs_u64(int _n,int _m,celt_uint64_t *_u){
CELT_MEMSET(_u,0,_n);
if(_m<=0)return 1;
if(_n<=0)return 0;
for(k=1;k<_m;k++)unext64(_u,_n,2);
for(k=1;k<_m;k++)unext64(_u,_n,1);
return ncwrs_unext64(_n,_u);
}
/*Returns the _i'th combination of _m elements chosen from a set of size _n
with associated sign bits.
_x: Returns the combination with elements sorted in ascending order.
_s: Returns the associated sign bits.
_u: Temporary storage already initialized to column _m of U(n,m).
Its contents will be overwritten.*/
void cwrsi32(int _n,int _m,celt_uint32_t _i,int *_x,int *_s,celt_uint32_t *_u){
int j;
int k;
for(k=j=0;k<_m;k++){
celt_uint32_t p;
celt_uint32_t t;
p=_u[_n-j-1];
if(k>0){
t=p>>1;
if(t<=_i||_s[k-1])_i+=t;
}
while(p<=_i){
_i-=p;
j++;
p=_u[_n-j-1];
_y: Returns the vector of pulses.
_u: Must contain entries [1..._n] of column _m of U() on input.
Its contents will be destructively modified.*/
void cwrsi32(int _n,int _m,celt_uint32_t _i,celt_uint32_t _nc,int *_y,
celt_uint32_t *_u){
celt_uint32_t p;
celt_uint32_t q;
int j;
int k;
celt_assert(_n>0);
p=_nc;
q=0;
j=0;
k=_m;
do{
int s;
int yj;
p-=q;
q=_u[_n-j-1];
p-=q;
s=_i>=p;
if(s)_i-=p;
yj=k;
while(q>_i){
uprev32(_u,_n-j,--k>0);
p=q;
q=_u[_n-j-1];
}
t=p>>1;
_s[k]=_i>=t;
_x[k]=j;
if(_s[k])_i-=t;
uprev32(_u,_n-j,2);
_i-=q;
yj-=k;
_y[j]=yj-(yj<<1&-s);
}
while(++j<_n);
}
void cwrsi64(int _n,int _m,celt_uint64_t _i,int *_x,int *_s,celt_uint64_t *_u){
int j;
int k;
for(k=j=0;k<_m;k++){
celt_uint64_t p;
celt_uint64_t t;
p=_u[_n-j-1];
if(k>0){
t=p>>1;
if(t<=_i||_s[k-1])_i+=t;
}
while(p<=_i){
_i-=p;
j++;
p=_u[_n-j-1];
/*Returns the _i'th combination of _m elements chosen from a set of size _n
with associated sign bits.
_y: Returns the vector of pulses.
_u: Must contain entries [1..._n] of column _m of U() on input.
Its contents will be destructively modified.*/
void cwrsi64(int _n,int _m,celt_uint64_t _i,celt_uint64_t _nc,int *_y,
celt_uint64_t *_u){
celt_uint64_t p;
celt_uint64_t q;
int j;
int k;
celt_assert(_n>0);
p=_nc;
q=0;
j=0;
k=_m;
do{
int s;
int yj;
p-=q;
q=_u[_n-j-1];
p-=q;
s=_i>=p;
if(s)_i-=p;
yj=k;
while(q>_i){
uprev64(_u,_n-j,--k>0);
p=q;
q=_u[_n-j-1];
}
t=p>>1;
_s[k]=_i>=t;
_x[k]=j;
if(_s[k])_i-=t;
uprev64(_u,_n-j,2);
_i-=q;
yj-=k;
_y[j]=yj-(yj<<1&-s);
}
while(++j<_n);
}
/*Returns the index of the given combination of _m elements chosen from a set
of size _n with associated sign bits.
_x: The combination with elements sorted in ascending order.
_s: The associated sign bits.
_u: Temporary storage already initialized to column _m of U(n,m).
Its contents will be overwritten.*/
celt_uint32_t icwrs32(int _n,int _m,const int *_x,const int *_s,
_y: The vector of pulses, whose sum of absolute values must be _m.
_nc: Returns V(_n,_m).*/
celt_uint32_t icwrs32(int _n,int _m,celt_uint32_t *_nc,const int *_y,
celt_uint32_t *_u){
celt_uint32_t nc;
celt_uint32_t i;
int j;
int k;
i=0;
for(k=j=0;k<_m;k++){
celt_uint32_t p;
p=_u[_n-j-1];
if(k>0)p>>=1;
while(j<_x[k]){
i+=p;
j++;
p=_u[_n-j-1];
}
if((k==0||_x[k]!=_x[k-1])&&_s[k])i+=p>>1;
uprev32(_u,_n-j,2);
/*We can't unroll the first two iterations of the loop unless _n>=2.*/
celt_assert(_n>=2);
nc=1;
i=_y[_n-1]<0;
_u[0]=0;
for(k=1;k<=_m+1;k++)_u[k]=(k<<1)-1;
k=abs(_y[_n-1]);
j=_n-2;
nc+=_u[_m];
i+=_u[k];
k+=abs(_y[j]);
if(_y[j]<0)i+=_u[k+1];
while(j-->0){
unext32(_u,_m+2,0);
nc+=_u[_m];
i+=_u[k];
k+=abs(_y[j]);
if(_y[j]<0)i+=_u[k+1];
}
/*If _m==0, nc should not be doubled.*/
celt_assert(_m>0);
*_nc=nc<<1;
return i;
}
celt_uint64_t icwrs64(int _n,int _m,const int *_x,const int *_s,
/*Returns the index of the given combination of _m elements chosen from a set
of size _n with associated sign bits.
_y: The vector of pulses, whose sum of absolute values must be _m.
_nc: Returns V(_n,_m).*/
celt_uint64_t icwrs64(int _n,int _m,celt_uint64_t *_nc,const int *_y,
celt_uint64_t *_u){
celt_uint64_t nc;
celt_uint64_t i;
int j;
int k;
i=0;
for(k=j=0;k<_m;k++){
celt_uint64_t p;
p=_u[_n-j-1];
if(k>0)p>>=1;
while(j<_x[k]){
i+=p;
j++;
p=_u[_n-j-1];
}
if((k==0||_x[k]!=_x[k-1])&&_s[k])i+=p>>1;
uprev64(_u,_n-j,2);
/*We can't unroll the first two iterations of the loop unless _n>=2.*/
celt_assert(_n>=2);
nc=1;
i=_y[_n-1]<0;
_u[0]=0;
for(k=1;k<=_m+1;k++)_u[k]=(k<<1)-1;
k=abs(_y[_n-1]);
j=_n-2;
nc+=_u[_m];
i+=_u[k];
k+=abs(_y[j]);
if(_y[j]<0)i+=_u[k+1];
while(j-->0){
unext64(_u,_m+2,0);
nc+=_u[_m];
i+=_u[k];
k+=abs(_y[j]);
if(_y[j]<0)i+=_u[k+1];
}
/*If _m==0, nc should not be doubled.*/
celt_assert(_m>0);
*_nc=nc<<1;
return i;
}
/*Converts a combination _x of _m unit pulses with associated sign bits _s into
a pulse vector _y of length _n.
_y: Returns the vector of pulses.
_x: The combination with elements sorted in ascending order. _x[_m] = -1
_s: The associated sign bits.*/
void comb2pulse(int _n,int _m,int * restrict _y,const int *_x,const int *_s){
int k;
const int signs[2]={1,-1};
CELT_MEMSET(_y, 0, _n);
k=0; do {
_y[_x[k]]+=signs[_s[k]];
} while (++k<_m);
}
/*Converts a pulse vector vector _y of length _n into a combination of _m unit
pulses with associated sign bits _s.
_x: Returns the combination with elements sorted in ascending order.
_s: Returns the associated sign bits.
_y: The vector of pulses, whose sum of absolute values must be _m.*/
void pulse2comb(int _n,int _m,int *_x,int *_s,const int *_y){
int j;
int k;
for(k=j=0;j<_n;j++){
if(_y[j]){
int n;
int s;
n=abs(_y[j]);
s=_y[j]<0;
do {
_x[k]=j;
_s[k]=s;
k++;
} while (--n>0);
}
}
}
static inline void encode_comb32(int _n,int _m,const int *_x,const int *_s,
ec_enc *_enc){
static inline void encode_pulse32(int _n,int _m,const int *_y,ec_enc *_enc){
VARDECL(celt_uint32_t,u);
celt_uint32_t nc;
celt_uint32_t i;
SAVE_STACK;
ALLOC(u,_n,celt_uint32_t);
nc=ncwrs_u32(_n,_m,u);
i=icwrs32(_n,_m,_x,_s,u);
ALLOC(u,_m+2,celt_uint32_t);
i=icwrs32(_n,_m,&nc,_y,u);
ec_enc_uint(_enc,i,nc);
RESTORE_STACK;
}
static inline void encode_comb64(int _n,int _m,const int *_x,const int *_s,
ec_enc *_enc){
static inline void encode_pulse64(int _n,int _m,const int *_y,ec_enc *_enc){
VARDECL(celt_uint64_t,u);
celt_uint64_t nc;
celt_uint64_t i;
SAVE_STACK;
ALLOC(u,_n,celt_uint64_t);
nc=ncwrs_u64(_n,_m,u);
i=icwrs64(_n,_m,_x,_s,u);
ALLOC(u,_m+2,celt_uint64_t);
i=icwrs64(_n,_m,&nc,_y,u);
ec_enc_uint64(_enc,i,nc);
RESTORE_STACK;
}
......@@ -438,23 +540,15 @@ int get_required_bits(int N, int K, int frac)
void encode_pulses(int *_y, int N, int K, ec_enc *enc)
{
VARDECL(int, comb);
VARDECL(int, signs);
SAVE_STACK;
ALLOC(comb, K, int);
ALLOC(signs, K, int);
pulse2comb(N, K, comb, signs, _y);
if (K==0) {
} else if (N==1)
{
ec_enc_bits(enc, _y[0]<0, 1);
} else if(fits_in32(N,K))
{
encode_comb32(N, K, comb, signs, enc);
encode_pulse32(N, K, _y, enc);
} else if(fits_in64(N,K)) {
encode_comb64(N, K, comb, signs, enc);
encode_pulse64(N, K, _y, enc);
} else {
int i;
int count=0;
......@@ -466,33 +560,30 @@ void encode_pulses(int *_y, int N, int K, ec_enc *enc)
encode_pulses(_y, split, count, enc);
encode_pulses(_y+split, N-split, K-count, enc);
}
RESTORE_STACK;
}
static inline void decode_comb32(int _n,int _m,int *_x,int *_s,ec_dec *_dec){
static inline void decode_pulse32(int _n,int _m,int *_y,ec_dec *_dec){
VARDECL(celt_uint32_t,u);
celt_uint32_t nc;
SAVE_STACK;
ALLOC(u,_n,celt_uint32_t);
cwrsi32(_n,_m,ec_dec_uint(_dec,ncwrs_u32(_n,_m,u)),_x,_s,u);
nc=ncwrs_u32(_n,_m,u);
cwrsi32(_n,_m,ec_dec_uint(_dec,nc),nc,_y,u);
RESTORE_STACK;
}
static inline void decode_comb64(int _n,int _m,int *_x,int *_s,ec_dec *_dec){
static inline void decode_pulse64(int _n,int _m,int *_y,ec_dec *_dec){
VARDECL(celt_uint64_t,u);
celt_uint64_t nc;
SAVE_STACK;
ALLOC(u,_n,celt_uint64_t);
cwrsi64(_n,_m,ec_dec_uint64(_dec,ncwrs_u64(_n,_m,u)),_x,_s,u);
nc=ncwrs_u64(_n,_m,u);
cwrsi64(_n,_m,ec_dec_uint64(_dec,nc),nc,_y,u);
RESTORE_STACK;
}
void decode_pulses(int *_y, int N, int K, ec_dec *dec)
{
VARDECL(int, comb);
VARDECL(int, signs);
SAVE_STACK;
ALLOC(comb, K, int);
ALLOC(signs, K, int);
if (K==0) {
int i;
for (i=0;i<N;i++)
......@@ -506,11 +597,9 @@ void decode_pulses(int *_y, int N, int K, ec_dec *dec)
_y[0] = -K;
} else if(fits_in32(N,K))
{
decode_comb32(N, K, comb, signs, dec);
comb2pulse(N, K, _y, comb, signs);
decode_pulse32(N, K, _y, dec);
} else if(fits_in64(N,K)) {
decode_comb64(N, K, comb, signs, dec);
comb2pulse(N, K, _y, comb, signs);
decode_pulse64(N, K, _y, dec);
} else {
int split;
int count = ec_dec_uint(dec,K+1);
......@@ -518,5 +607,4 @@ void decode_pulses(int *_y, int N, int K, ec_dec *dec)
decode_pulses(_y, split, count, dec);
decode_pulses(_y+split, N-split, K-count, dec);
}
RESTORE_STACK;
}
......@@ -46,28 +46,22 @@ int fits_in64(int _n, int _m);
/* 32-bit versions */
celt_uint32_t ncwrs_u32(int _n,int _m,celt_uint32_t *_u);
void cwrsi32(int _n,int _m,celt_uint32_t _i,int *_x,int *_s,
void cwrsi32(int _n,int _m,celt_uint32_t _i,celt_uint32_t _nc,int *_y,
celt_uint32_t *_u);
celt_uint32_t icwrs32(int _n,int _m,const int *_x,const int *_s,
celt_uint32_t icwrs32(int _n,int _m,celt_uint32_t *_nc,const int *_y,
celt_uint32_t *_u);
/* 64-bit versions */
celt_uint64_t ncwrs_u64(int _n,int _m,celt_uint64_t *_u);
celt_uint64_t ncwrs_unext64(int _n,celt_uint64_t *_u);
void cwrsi64(int _n,int _m,celt_uint64_t _i,int *_x,int *_s,
void cwrsi64(int _n,int _m,celt_uint64_t _i,celt_uint64_t _nc,int *_y,
celt_uint64_t *_u);
celt_uint64_t icwrs64(int _n,int _m,const int *_x,const int *_s,
celt_uint64_t icwrs64(int _n,int _m,celt_uint64_t *_nc,const int *_y,
celt_uint64_t *_u);
void comb2pulse(int _n,int _m,int * restrict _y,const int *_x,const int *_s);
void pulse2comb(int _n,int _m,int *_x,int *_s,const int *_y);
int get_required_bits(int N, int K, int frac);
void encode_pulses(int *_y, int N, int K, ec_enc *enc);
......
......@@ -21,33 +21,24 @@ int main(int _argc,char **_argv){
inc=nc/10000;
if(inc<1)inc=1;
for(i=0;i<nc;i+=inc){
celt_uint32_t u[NMAX];
int x[MMAX];
int s[MMAX];
int x2[MMAX];
int s2[MMAX];
celt_uint32_t u[NMAX>MMAX+2?NMAX:MMAX+2];
int y[NMAX];
celt_uint32_t v;
int k;
memcpy(u,uu,n*sizeof(*u));
cwrsi32(n,m,i,x,s,u);
/*printf("%6u of %u:",i,nc);*/
/*for(k=0;k<m;k++){
printf(" %c%i",k>0&&x[k]==x[k-1]?' ':s[k]?'-':'+',x[k]);
}
cwrsi32(n,m,i,nc,y,u);
/*printf("%6u of %u:",i,nc);
for(k=0;k<n;k++)printf(" %+3i",y[k]);
printf(" ->");*/
memcpy(u,uu,n*sizeof(*u));
if(icwrs32(n,m,x,s,u)!=i){
if(icwrs32(n,m,&v,y,u)!=i){
fprintf(stderr,"Combination-index mismatch.\n");
return 1;
}
comb2pulse(n,m,y,x,s);
/*for(j=0;j<n;j++)printf(" %c%i",y[j]?y[j]<0?'-':'+':' ',abs(y[j]));
printf("\n");*/
pulse2comb(n,m,x2,s2,y);
for(k=0;k<m;k++)if(x[k]!=x2[k]||s[k]!=s2[k]){
fprintf(stderr,"Pulse-combination mismatch.\n");
return 1;
if(v!=nc){
fprintf(stderr,"Combination count mismatch.\n");
return 2;
}
/*printf(" %6u\n",i);*/
}
/*printf("\n");*/
}
......
......@@ -24,33 +24,24 @@ int main(int _argc,char **_argv){
if(inc<1)inc=1;
/*printf("%d/%d: %llu",n,m, nc);*/