daiduo/TurboPFor-Integer-Compression
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

BitUtil: Delta, ZigZag, NumBits, Floating Point,...

Browse Source
This commit is contained in:
x
2017-03-19 13:20:20 +01:00
parent a95fdf21a7
commit 6a23b0813e
1 changed files with 174 additions and 229 deletions
Download Patch File Download Diff File Expand all files Collapse all files

403
bitutil.c
View File

@ -25,41 +25,24 @@
#include "conf.h" #include "conf.h"
#include "bitutil.h" #include "bitutil.h"
//------------- max. bits for bit packing ------------------- //------------- max. bits for bit packing -------------
unsigned bit8( uint8_t *in, unsigned n) { uint8_t b; BITSIZE_( in, n, b, 8); return b; } unsigned bit8( uint8_t *in, unsigned n) { uint8_t b; BITSIZE_( in, n, b, 8); return b; }
unsigned bit16(uint16_t *in, unsigned n) { uint16_t b; BITSIZE_( in, n, b, 16); return b; } unsigned bit16(uint16_t *in, unsigned n) { uint16_t b; BITSIZE_( in, n, b, 16); return b; }
unsigned bit32(uint32_t *in, unsigned n) { uint32_t b; BITSIZE32(in, n, b); return b; } unsigned bit32(uint32_t *in, unsigned n) { uint32_t b; BITSIZE32(in, n, b); return b; }
unsigned bit64(uint64_t *in, unsigned n) { uint64_t b; BITSIZE_( in, n, b, 64); return b; } unsigned bit64(uint64_t *in, unsigned n) { uint64_t b; BITSIZE_( in, n, b, 64); return b; }
//---------------- Delta ---------------------------------------- //------------ Delta ----------------------------------
#define BITD(_p_,_n_, _inc_, _start_, _act_) {\ #define DE(i) x = _p[i]-start-_md; start = _p[i]
typeof(_p_[0]) _x, *_p;\ #define BITDE(_t_, _p_, _n_, _md_, _act_) { _t_ *_p; const _md = _md_;\
for(_p = _p_; _p != _p_+(_n_&~(4-1)); _p+=4) {\ for(_p = _p_; _p != _p_+(_n_&~(4-1)); _p += 4) { DE(0);_act_; DE(1);_act_; DE(2);_act_; DE(3);_act_; }\
_x = _p[0]-_start_-_inc_; _start_ = _p[0]; _act_;\ for(;_p != _p_+_n_;_p++) { DE(0); _act_; }\
_x = _p[1]-_start_-_inc_; _start_ = _p[1]; _act_;\
_x = _p[2]-_start_-_inc_; _start_ = _p[2]; _act_;\
_x = _p[3]-_start_-_inc_; _start_ = _p[3]; _act_;\
}\
for(;_p != _p_+_n_;_p++) {\
_x = _p[0]-_start_-_inc_; _start_ = _p[0]; _act_;\
}\
} }
/*#define BITD(_p_,_n_, _inc_, _start_, _act_) {\ //---- Delta (min. Delta = 0)
typeof(_p_[0]) _x, *_p; int i;\ //-- max. bits for delta encoding
for(_p = _p_+_n_, i = n&(4-1); i > 0; i--) { --_p; _p[i] -= _p[i-1]; } unsigned bitd8( uint8_t *in, unsigned n, uint8_t start) { uint8_t b = 0,x; BITDE(uint8_t, in, n, 0, b |= x); return bsr8( b); }
while(_p != _p_) {\ unsigned bitd16(uint16_t *in, unsigned n, uint16_t start) { uint16_t b = 0,x; BITDE(uint16_t, in, n, 0, b |= x); return bsr16(b); }
_p[-1] -= _p[-2]+_inc_; _act_;\ unsigned bitd64(uint64_t *in, unsigned n, uint64_t start) { uint64_t b = 0,x; BITDE(uint64_t, in, n, 0, b |= x); return bsr64(b); }
_p[-2] -= _p[-3]+_inc_; _act_;\
_p[-3] -= _p[-4]+_inc_; _act_;\
_p[-4] -= _p[-5]+_inc_; _act_;\
}\
for(;_p != _p_+_n_;_p++) {\
_x = _p[0]-_start_-_inc_; _start_ = _p[0]; _act_;\
}\
}*/
// max. bits for delta bit packing
unsigned bitd32(uint32_t *in, unsigned n, uint32_t start) { unsigned bitd32(uint32_t *in, unsigned n, uint32_t start) {
#ifdef __SSE2__ #ifdef __SSE2__
unsigned *ip,b; __m128i bv = _mm_setzero_si128(), sv = _mm_set1_epi32(start); unsigned *ip,b; __m128i bv = _mm_setzero_si128(), sv = _mm_set1_epi32(start);
@ -77,45 +60,58 @@ unsigned bitd32(uint32_t *in, unsigned n, uint32_t start) {
b |= x; b |= x;
} }
#else #else
typeof(in[0]) b = 0; uint32_t b = 0,_x,*_p;
BITD(in,n, 0, start, b |= _x); BITDE(in,n, 0, start, b |= _x);
#endif #endif
return bsr32(b); return bsr32(b);
} }
unsigned bitd8( uint8_t *in, unsigned n, uint8_t start) { uint8_t b = 0; BITD(in, n, 0, start, b |= _x); return bsr8(b); } //----- Undelta: In-place prefix sum (min. Delta = 0) -------------------
unsigned bitd16(uint16_t *in, unsigned n, uint16_t start) { uint16_t b = 0; BITD(in, n, 0, start, b |= _x); return bsr16(b); } #define DD(i) _p[i] = (start += _p[i] + _md);
unsigned bitd64(uint64_t *in, unsigned n, uint64_t start) { uint64_t b = 0; BITD(in, n, 0, start, b |= _x); return bsr64(b); } #define BITDD(_t_, _p_, _n_, _md_) { _t_ *_p; const _md = _md_;\
for(_p = _p_; _p != _p_+(_n_&~(4-1)); _p += 4) { DD(0); DD(1); DD(2); DD(3); }\
unsigned bitdenc32(uint32_t *in, unsigned n, uint32_t *out, uint32_t start, unsigned inc) { for(;_p != _p_+_n_; _p++) DD(0);\
#ifdef __SSE2__
unsigned *ip,b,*op = out;
__m128i bv = _mm_setzero_si128(), sv = _mm_set1_epi32(start), cv = _mm_set1_epi32(inc), dv;
for(ip = in; ip != in+(n&~(4-1)); ip += 4,op += 4) {
__m128i iv = _mm_loadu_si128((__m128i *)ip);
bv = _mm_or_si128(bv, dv = _mm_sub_epi32(DELTA128x32(iv,sv),cv));
sv = iv;
_mm_storeu_si128((__m128i *)op, dv);
}
start = (unsigned)_mm_cvtsi128_si32(_mm_srli_si128(sv,12));
HOR128x32(bv, b);
while(ip != in+n) {
unsigned x = *ip-start-inc;
start = *ip++;
b |= x;
*op++ = x;
}
#else
typeof(in[0]) b = 0,*op = out;
BITD(in, n, inc, start, b |= _x;*op++ = _x);
#endif
return bsr32(b);
} }
unsigned bitdenc8( uint8_t *in, unsigned n, uint8_t *out, uint8_t start, unsigned inc) { uint8_t b = 0,*op = out; BITD(in, n, inc, start, b |= _x; *op++ = _x); return bsr8(b);} void bitddec8( uint8_t *p, unsigned n, uint8_t start) { BITDD(uint8_t, p, n, 0); }
unsigned bitdenc16(uint16_t *in, unsigned n, uint16_t *out, uint16_t start, unsigned inc) { uint16_t b = 0,*op = out; BITD(in, n, inc, start, b |= _x; *op++ = _x); return bsr16(b);} void bitddec16(uint16_t *p, unsigned n, uint16_t start) { BITDD(uint16_t, p, n, 0); }
unsigned bitdenc64(uint64_t *in, unsigned n, uint64_t *out, uint64_t start, unsigned inc) { uint64_t b = 0,*op = out; BITD(in, n, inc, start, b |= _x; *op++ = _x); return bsr64(b);} void bitddec64(uint64_t *p, unsigned n, uint64_t start) { BITDD(uint64_t, p, n, 0); }
void bitddec32(uint32_t *p, unsigned n, unsigned start) {
#if defined(__AVX2__) && defined(USE_AVX2)
__m256i sv = _mm256_set1_epi32(start),zv = _mm256_setzero_si256();
unsigned *ip;
for(ip = p; ip != p+(n&~(8-1)); ip += 8) {
__m256i v = _mm256_loadu_si256((__m256i *)ip);
SCAN256x32(v, sv);
_mm256_storeu_si256((__m256i *)ip, sv);
}
start = (unsigned)_mm256_extract_epi32(sv, 7);
while(ip != p+n) {
*ip = (start += (*ip));
ip++;
}
#elif defined(__SSE2__)
__m128i sv = _mm_set1_epi32(start);
unsigned *ip;
for(ip = p; ip != p+(n&~(4-1)); ip += 4) {
__m128i v = _mm_loadu_si128((__m128i *)ip);
SCAN128x32(v, sv);
_mm_storeu_si128((__m128i *)ip, sv);
}
start = (unsigned)_mm_cvtsi128_si32(_mm_srli_si128(sv,12));
while(ip != p+n) {
*ip = (start += (*ip));
ip++;
}
#else
BITDD(uint32_t, p, n, 0);
#endif
}
//-----Undelta: In-place prefix sum (min. Delta = 1) -------------------
unsigned bitd18( uint8_t *in, unsigned n, uint8_t start) { uint8_t b = 0,x; BITDE(uint8_t, in, n, 1, b |= x); return bsr8(b); }
unsigned bitd116(uint16_t *in, unsigned n, uint16_t start) { uint16_t b = 0,x; BITDE(uint16_t, in, n, 1, b |= x); return bsr16(b); }
unsigned bitd164(uint64_t *in, unsigned n, uint64_t start) { uint64_t b = 0,x; BITDE(uint64_t, in, n, 1, b |= x); return bsr64(b); }
unsigned bitd132(uint32_t *in, unsigned n, uint32_t start) { unsigned bitd132(uint32_t *in, unsigned n, uint32_t start) {
#ifdef __SSE2__ #ifdef __SSE2__
unsigned *ip,b; __m128i bv = _mm_setzero_si128(), sv = _mm_set1_epi32(start), cv = _mm_set1_epi32(1); unsigned *ip,b; __m128i bv = _mm_setzero_si128(), sv = _mm_set1_epi32(start), cv = _mm_set1_epi32(1);
@ -133,36 +129,30 @@ unsigned bitd132(uint32_t *in, unsigned n, uint32_t start) {
b |= x; b |= x;
} }
#else #else
uint32_t b = 0; uint32_t b = 0,x;
BITD(in, n, 1, start, b |= _x); BITDE(uint32_t, in, n, 1, b |= _x);
#endif #endif
return bsr32(b); return bsr32(b);
} }
unsigned bitd18( uint8_t *in, unsigned n, uint8_t start) { uint8_t b = 0; BITD(in, n, 1, start, b |= _x); return bsr8(b); } void bitd1dec8( uint8_t *p, unsigned n, uint8_t start) { BITDD(uint8_t, p, n, 1); }
unsigned bitd116(uint16_t *in, unsigned n, uint16_t start) { uint16_t b = 0; BITD(in, n, 1, start, b |= _x); return bsr16(b); } void bitd1dec16(uint16_t *p, unsigned n, uint16_t start) { BITDD(uint16_t, p, n, 1); }
unsigned bitd164(uint64_t *in, unsigned n, uint64_t start) { uint64_t b = 0; BITD(in, n, 1, start, b |= _x); return bsr64(b); } void bitd1dec64(uint64_t *p, unsigned n, uint64_t start) { BITDD(uint64_t, p, n, 1); }
void bitd1dec32(uint32_t *p, unsigned n, uint32_t start) {
// -- undelta: in-place prefix sum --------------- #if defined(__AVX2__) && defined(USE_AVX2)
#define BITDDEC(_p_, _n_, _start_, _inc_) {\ __m256i sv = _mm256_set1_epi32(start),zv = _mm256_setzero_si256(), cv = _mm256_set_epi32(8,7,6,5,4,3,2,1);
typeof(_p_[0]) *_p;\ unsigned *ip;
for(_p = _p_; _p != _p_+(_n_&~(4-1)); _p += 4) {\ for(ip = p; ip != p+(n&~(8-1)); ip += 8) {
_p[0] = (_start_ += _p[0] + _inc_);\ __m256i v = _mm256_loadu_si256((__m256i *)ip);
_p[1] = (_start_ += _p[1] + _inc_);\ SCANI256x32(v, sv, cv);
_p[2] = (_start_ += _p[2] + _inc_);\ _mm256_storeu_si256((__m256i *)ip, sv);
_p[3] = (_start_ += _p[3] + _inc_);\ }
}\ start = (unsigned)_mm256_extract_epi32(sv, 7);
for(;_p != _p_+_n_; _p++)\ while(ip != p+n) {
_p[0] = (_start_ += _p[0] + _inc_);\ *ip = (start += (*ip) + 1);
} ip++;
}
void bitddec8( uint8_t *p, unsigned n, uint8_t start) { BITDDEC(p, n, start, 0); } #elif defined(__SSE2__)
void bitddec16( uint16_t *p, unsigned n, uint16_t start) { BITDDEC(p, n, start, 0); }
void bitddec32( uint32_t *p, unsigned n, uint32_t start) { BITDDEC(p, n, start, 0); }
void bitddec64( uint64_t *p, unsigned n, uint64_t start) { BITDDEC(p, n, start, 0); }
void bitd1dec32(unsigned *p, unsigned n, unsigned start) {
#ifdef __SSE2__
__m128i sv = _mm_set1_epi32(start), cv = _mm_set_epi32(4,3,2,1); __m128i sv = _mm_set1_epi32(start), cv = _mm_set_epi32(4,3,2,1);
unsigned *ip; unsigned *ip;
for(ip = p; ip != p+(n&~(4-1)); ip += 4) { for(ip = p; ip != p+(n&~(4-1)); ip += 4) {
@ -176,93 +166,71 @@ void bitd1dec32(unsigned *p, unsigned n, unsigned start) {
ip++; ip++;
} }
#else #else
BITDDEC(p, n, start, 1); BITDD(uint32_t, p, n, 1);
#endif #endif
} }
void bitd1dec8( uint8_t *p, unsigned n, uint8_t start) { BITDDEC(p, n, start, 1); } //---------Delta encoding/decoding (min. Delta = mindelta) -------------------
void bitd1dec16(uint16_t *p, unsigned n, uint16_t start) { BITDDEC(p, n, start, 1); } //determine min. delta for encoding w/ bitdiencNN function
void bitd1dec64(uint64_t *p, unsigned n, uint64_t start) { BITDDEC(p, n, start, 1); } #define DI(i) x = _p[0] - start; start = _p[0]; if(x < mindelta) mindelta = x
#define BITDIE(_p_, _n_) {\
void bitddecn8( uint8_t *p, unsigned n, uint8_t start, uint8_t inc) { BITDDEC(p, n, start, inc); } for(_p = _p_,mindelta = _p[0]; _p != _p_+(_n_&~(4-1)); _p+=4) { DI(0); DI(1); DI(2); DI(3); }\
void bitddecn16(uint16_t *p, unsigned n, uint16_t start, uint16_t inc) { BITDDEC(p, n, start, inc); } for(;_p != _p_+_n_;_p++) DI(0);\
void bitddecn32(uint32_t *p, unsigned n, uint32_t start, uint32_t inc) { BITDDEC(p, n, start, inc); }
void bitddecn64(uint64_t *p, unsigned n, uint64_t start, uint64_t inc) { BITDDEC(p, n, start, inc); }
#define BITDI(_p_,_n_, _start_, _inc_) {\
typeof(_p_[0]) _x, *_p; _inc_= (uint32_t)(-1);\
for(_p = _p_; _p != _p_+(_n_&~(4-1)); _p+=4) {\
_x = _p[0]-_start_; _start_ = _p[0]; if(_x<_inc_) _inc_=_x;\
_x = _p[1]-_start_; _start_ = _p[1]; if(_x<_inc_) _inc_=_x;\
_x = _p[2]-_start_; _start_ = _p[2]; if(_x<_inc_) _inc_=_x;\
_x = _p[3]-_start_; _start_ = _p[3]; if(_x<_inc_) _inc_=_x;\
}\
for(;_p != _p_+_n_;_p++) {\
_x = _p[0]-_start_; _start_ = _p[0]; if(_x<_inc_) _inc_=_x;\
}\
} }
#define BITDIE(_p_,_n_, _start_, _inc_, _act_) {\ uint8_t bitdi8( uint8_t *in, unsigned n, uint8_t start) { uint8_t mindelta,x,*_p; BITDIE(in, n); return mindelta; }
typeof(_p_[0]) _x, *_p;\ uint16_t bitdi16(uint16_t *in, unsigned n, uint16_t start) { uint16_t mindelta,x,*_p; BITDIE(in, n); return mindelta; }
for(_p = _p_; _p != _p_+(_n_&~(4-1)); _p+=4) {\ uint32_t bitdi32(uint32_t *in, unsigned n, uint32_t start) { uint32_t mindelta,x,*_p; BITDIE(in, n); return mindelta; }
_x = _p[0]-_start_-_inc_; _start_ = _p[0]; _act_; \ uint64_t bitdi64(uint64_t *in, unsigned n, uint64_t start) { uint64_t mindelta,x,*_p; BITDIE(in, n); return mindelta; }
_x = _p[1]-_start_-_inc_; _start_ = _p[1]; _act_;\
_x = _p[2]-_start_-_inc_; _start_ = _p[2]; _act_;\ unsigned bitdienc8( uint8_t *in, unsigned n, uint8_t *out, uint8_t start, uint8_t mindelta) { uint8_t b = 0,*op = out,x; BITDE(uint8_t, in, n, mindelta, b |= x;*op++ = x); return bsr8(b);}
_x = _p[3]-_start_-_inc_; _start_ = _p[3]; _act_;\ unsigned bitdienc16(uint16_t *in, unsigned n, uint16_t *out, uint16_t start, uint16_t mindelta) { uint16_t b = 0,*op = out,x; BITDE(uint16_t, in, n, mindelta, b |= x;*op++ = x); return bsr16(b);}
}\ unsigned bitdienc64(uint64_t *in, unsigned n, uint64_t *out, uint64_t start, uint64_t mindelta) { uint64_t b = 0,*op = out,x; BITDE(uint64_t, in, n, mindelta, b |= x;*op++ = x); return bsr64(b);}
for(;_p != _p_+_n_;_p++) {\ unsigned bitdienc32(uint32_t *in, unsigned n, uint32_t *out, uint32_t start, uint32_t mindelta) {
_x = _p[0]-_start_-_inc_; _start_ = _p[0]; _act_;\ #ifdef __SSE2__
}\ unsigned *ip,b,*op = out;
__m128i bv = _mm_setzero_si128(), sv = _mm_set1_epi32(start), cv = _mm_set1_epi32(mindelta), dv;
for(ip = in; ip != in+(n&~(4-1)); ip += 4,op += 4) {
__m128i iv = _mm_loadu_si128((__m128i *)ip);
bv = _mm_or_si128(bv, dv = _mm_sub_epi32(DELTA128x32(iv,sv),cv));
sv = iv;
_mm_storeu_si128((__m128i *)op, dv);
}
start = (unsigned)_mm_cvtsi128_si32(_mm_srli_si128(sv,12));
HOR128x32(bv, b);
while(ip != in+n) {
unsigned x = *ip-start-mindelta;
start = *ip++;
b |= x;
*op++ = x;
}
#else
uint32_t b = 0,*op = out, x, *_p;
BITDE(uint32_t, in, n, mindelta, b |= x; *op++ = x);
#endif
return bsr32(b);
} }
uint8_t bitdi8( uint8_t *in, unsigned n, uint8_t start) { uint8_t i; BITDI(in, n, start, i); return i; } void bitdidec8( uint8_t *p, unsigned n, uint8_t start, uint8_t mindelta) { BITDD(uint8_t, p, n, mindelta); }
uint16_t bitdi16(uint16_t *in, unsigned n, uint16_t start) { uint16_t i; BITDI(in, n, start, i); return i; } void bitdidec16( uint16_t *p, unsigned n, uint16_t start, uint16_t mindelta) { BITDD(uint16_t, p, n, mindelta); }
uint32_t bitdi32(uint32_t *in, unsigned n, uint32_t start) { uint32_t i; BITDI(in, n, start, i); return i; } void bitdidec32( uint32_t *p, unsigned n, uint32_t start, uint32_t mindelta) { BITDD(uint32_t, p, n, mindelta); }
uint64_t bitdi64(uint64_t *in, unsigned n, uint64_t start) { uint64_t i; BITDI(in, n, start, i); return i; } void bitdidec64( uint64_t *p, unsigned n, uint64_t start, uint64_t mindelta) { BITDD(uint64_t, p, n, mindelta); }
unsigned bitdienc8( uint8_t *in, unsigned n, uint8_t *out, uint8_t start, uint8_t inc) { uint8_t b = 0,*op = out; BITDIE(in, n, start, inc, b |= _x; *op++ = _x); return bsr8(b); }
unsigned bitdienc16(uint16_t *in, unsigned n, uint16_t *out, uint16_t start, uint16_t inc) { uint16_t b = 0,*op = out; BITDIE(in, n, start, inc, b |= _x; *op++ = _x); return bsr16(b);}
unsigned bitdienc32(uint32_t *in, unsigned n, uint32_t *out, uint32_t start, uint32_t inc) { uint32_t b = 0,*op = out; BITDIE(in, n, start, inc, b |= _x; *op++ = _x); return bsr32(b);}
unsigned bitdienc64(uint64_t *in, unsigned n, uint64_t *out, uint64_t start, uint64_t inc) { uint64_t b = 0,*op = out; BITDIE(in, n, start, inc, b |= _x; *op++ = _x); return bsr64(b);}
void bitdidec8( uint8_t *p, unsigned n, uint8_t start, uint8_t inc) { BITDDEC(p, n, start, inc); }
void bitdidec16( uint16_t *p, unsigned n, uint16_t start, uint16_t inc) { BITDDEC(p, n, start, inc); }
void bitdidec32( uint32_t *p, unsigned n, uint32_t start, uint32_t inc) { BITDDEC(p, n, start, inc); }
void bitdidec64( uint64_t *p, unsigned n, uint64_t start, uint64_t inc) { BITDDEC(p, n, start, inc); }
//------------------- For ------------------------------ //------------------- For ------------------------------
unsigned bitf32( uint32_t *in, unsigned n, uint32_t start) { return n?bsr32(in[n-1] - start ):0; } unsigned bitf32( uint32_t *in, unsigned n, uint32_t start) { return n?bsr32(in[n-1] - start ):0; }
unsigned bitf132( uint32_t *in, unsigned n, uint32_t start) { return n?bsr32(in[n-1] - start - n):0; } unsigned bitf132( uint32_t *in, unsigned n, uint32_t start) { return n?bsr32(in[n-1] - start - n):0; }
//----------------------------- zigzag -------------------------------------------------------- //------------------- Zigzag ---------------------------
#define BITZENC(_p_,_n_, _start_, _act_) {\ #define ZE(i) x = (int)_p[i]-(int)start; x = (x << 1) ^ (x >> (sizeof(x)*8-1)); start = _p[i]
typeof(_p_[0]) *_p;\ #define BITZENC(_t_, _p_,_n_, _act_) { _t_ *_p;\
for(_p = _p_; _p != _p_+(_n_&~(4-1)); ) {\ for(_p = _p_; _p != _p_+(_n_&~(4-1)); _p += 4) { ZE(0);_act_; ZE(1);_act_; ZE(2);_act_; ZE(3);_act_; }\
_x = ((int)(*_p)-(int)_start_); _x = (_x << 1) ^ (_x >> (sizeof(_x)*8-1)); _start_ = *_p++; _act_;\ for(;_p != _p_+_n_; _p++) { ZE(0); _act_;}\
_x = ((int)(*_p)-(int)_start_); _x = (_x << 1) ^ (_x >> (sizeof(_x)*8-1)); _start_ = *_p++; _act_;\
_x = ((int)(*_p)-(int)_start_); _x = (_x << 1) ^ (_x >> (sizeof(_x)*8-1)); _start_ = *_p++; _act_;\
_x = ((int)(*_p)-(int)_start_); _x = (_x << 1) ^ (_x >> (sizeof(_x)*8-1)); _start_ = *_p++; _act_;\
}\
while(_p != _p_+_n_) { \
_x = ((int)(*_p)-(int)_start_); _x = (_x << 1) ^ (_x >> (sizeof(_x)*8-1)); _start_ = *_p++; _act_;\
}\
}
#define BITZDEC(_p_, _n_, _start_) {\
typeof(_p_[0]) *_p, _z;\
for(_p = _p_; _p != _p_+(_n_&~(4-1)); ) {\
_z = *_p; *_p = (_start_ += (_z >> 1 ^ -(_z & 1))); _p++;\
_z = *_p; *_p = (_start_ += (_z >> 1 ^ -(_z & 1))); _p++;\
_z = *_p; *_p = (_start_ += (_z >> 1 ^ -(_z & 1))); _p++;\
_z = *_p; *_p = (_start_ += (_z >> 1 ^ -(_z & 1))); _p++;\
}\
while(_p != _p_+_n_) {\
_z = *_p; *_p = (_start_ += (_z >> 1 ^ -(_z & 1))); _p++;\
}\
} }
// max. bits for zigzag encoding // max. bits for zigzag encoding
unsigned bitz8( uint8_t *in, unsigned n, uint8_t start) { uint8_t b = 0, x; BITZENC(uint8_t, in, n, b |= x); return bsr8( b); }
unsigned bitz16(uint16_t *in, unsigned n, uint16_t start) { uint16_t b = 0, x; BITZENC(uint16_t, in, n, b |= x); return bsr16(b); }
unsigned bitz64(uint64_t *in, unsigned n, uint64_t start) { uint64_t b = 0, x; BITZENC(uint64_t, in, n, b |= x); return bsr64(b); }
unsigned bitz32(unsigned *in, unsigned n, unsigned start) { unsigned bitz32(unsigned *in, unsigned n, unsigned start) {
#ifdef __SSE2__ #ifdef __SSE2__
unsigned *ip,b; unsigned *ip,b;
@ -283,18 +251,16 @@ unsigned bitz32(unsigned *in, unsigned n, unsigned start) {
b |= x; b |= x;
} }
#else #else
typeof(in[0]) b = 0; uint32_t b = 0, x;
int _x; BITZENC(uint32_t, in, n, start, b |= _x);
BITZENC(in, n, start, b |= (unsigned)_x);
#endif #endif
return bsr32(b); return bsr32(b);
} }
unsigned bitz8( uint8_t *in, unsigned n, uint8_t start) { uint8_t b = 0, _x; BITZENC(in, n, start, b |= (uint8_t )_x); return bsr8( b); } unsigned bitzenc8( uint8_t *in, unsigned n, uint8_t *out, uint8_t start, uint8_t mindelta) { uint8_t b = 0,*op = out,x; BITZENC(uint8_t, in, n, b |= x; *op++ = x); return bsr8(b); }
unsigned bitz16(uint16_t *in, unsigned n, uint16_t start) { uint16_t b = 0, _x; BITZENC(in, n, start, b |= (uint16_t)_x); return bsr16(b); } unsigned bitzenc16(uint16_t *in, unsigned n, uint16_t *out, uint16_t start, uint16_t mindelta) { uint16_t b = 0,*op = out,x; BITZENC(uint16_t, in, n, b |= x; *op++ = x); return bsr16(b); }
unsigned bitz64(uint64_t *in, unsigned n, uint64_t start) { uint64_t b = 0, _x; BITZENC(in, n, start, b |= (uint64_t)_x); return bsr64(b); } unsigned bitzenc64(uint64_t *in, unsigned n, uint64_t *out, uint64_t start, uint64_t mindelta) { uint64_t b = 0,*op = out,x; BITZENC(uint64_t, in, n, b |= x; *op++ = x); return bsr32(b); }
unsigned bitzenc32(unsigned *in, unsigned n, unsigned *out, unsigned start, uint32_t mindelta) {
unsigned bitzenc32(unsigned *in, unsigned n, unsigned *out, unsigned start, uint32_t inc) {
#ifdef __SSE2__ #ifdef __SSE2__
unsigned *ip,b,*op = out; unsigned *ip,b,*op = out;
__m128i bv = _mm_setzero_si128(), sv = _mm_set1_epi32(start), dv; __m128i bv = _mm_setzero_si128(), sv = _mm_set1_epi32(start), dv;
@ -316,15 +282,23 @@ unsigned bitzenc32(unsigned *in, unsigned n, unsigned *out, unsigned start, uint
*op++ = x; *op++ = x;
} }
#else #else
typeof(in[0]) b = 0, *op = out; uint32_t b = 0, *op = out,x;
int _x; BITZENC(uint32_t, in, n, b |= x; *op++ = x);
BITZENC(in, n, start, b |= (unsigned)_x; *op++ = _x);
#endif #endif
return bsr32(b); return bsr32(b);
} }
#define ZD(i) _z = _p[i]; _p[i] = (start += (_z >> 1 ^ -(_z & 1)))
#define BITZDEC(_t_, _p_, _n_) { _t_ *_p,_z;\
for(_p = _p_; _p != _p_+(_n_&~(4-1)); _p += 4) { ZD(0); ZD(1); ZD(2); ZD(3); }\
for(;_p != _p_+_n_;_p++) ZD(0);\
}
void bitzdec8( uint8_t *p, unsigned n, uint8_t start) { BITZDEC(uint8_t, p, n); }
void bitzdec16(uint16_t *p, unsigned n, uint16_t start) { BITZDEC(uint16_t, p, n); }
void bitzdec64(uint64_t *p, unsigned n, uint64_t start) { BITZDEC(uint64_t, p, n); }
void bitzdec32(unsigned *p, unsigned n, unsigned start) { void bitzdec32(unsigned *p, unsigned n, unsigned start) {
#ifdef __AVX2__ #if defined(__AVX2__) && defined(USE_AVX2)
__m256i sv = _mm256_set1_epi32(start), zv = _mm256_setzero_si256(); //, c1 = _mm_set1_epi32(1), cz = _mm_setzero_si128(); __m256i sv = _mm256_set1_epi32(start), zv = _mm256_setzero_si256(); //, c1 = _mm_set1_epi32(1), cz = _mm_setzero_si128();
unsigned *ip; unsigned *ip;
for(ip = p; ip != p+(n&~(8-1)); ip += 8) { for(ip = p; ip != p+(n&~(8-1)); ip += 8) {
@ -348,74 +322,45 @@ void bitzdec32(unsigned *p, unsigned n, unsigned start) {
_mm_storeu_si128((__m128i *)ip, sv); _mm_storeu_si128((__m128i *)ip, sv);
} }
start = (unsigned)_mm_cvtsi128_si32(_mm_srli_si128(sv,12)); start = (unsigned)_mm_cvtsi128_si32(_mm_srli_si128(sv,12));
while(ip != p+n) { while(ip != p+n) {
unsigned z = *ip; unsigned z = *ip;
*ip++ = (start += (z >> 1 ^ -(z & 1))); *ip++ = (start += (z >> 1 ^ -(z & 1)));
} }
#else #else
BITZDEC(p, n, start); BITZDEC(uint32_t, p, n);
#endif #endif
} }
unsigned bitzenc8( uint8_t *in, unsigned n, uint8_t *out, uint8_t start, uint8_t inc) { uint8_t b = 0,*op = out; int8_t _x; BITZENC(in, n, start, b |= (uint8_t)_x; *op++ = _x); return bsr8(b); }
unsigned bitzenc16(uint16_t *in, unsigned n, uint16_t *out, uint16_t start, uint16_t inc) { uint16_t b = 0,*op = out; int16_t _x; BITZENC(in, n, start, b |= (uint16_t)_x; *op++ = _x); return bsr16(b); }
unsigned bitzenc64(uint64_t *in, unsigned n, uint64_t *out, uint64_t start, uint64_t inc) { uint64_t b = 0,*op = out; int64_t _x; BITZENC(in, n, start, b |= (uint64_t)_x; *op++ = _x); return bsr32(b); }
void bitzdec8( uint8_t *p, unsigned n, uint8_t start) { BITZDEC(p, n, start); }
void bitzdec16(uint16_t *p, unsigned n, uint16_t start) { BITZDEC(p, n, start); }
void bitzdec64(uint64_t *p, unsigned n, uint64_t start) { BITZDEC(p, n, start); }
//----------------------- XOR --------------------------------- //----------------------- XOR ---------------------------------
#define BITXENC(_p_,_n_, _start_, _act_) {\ #define XE(i) x = _p[i] ^ start; start = _p[i]
typeof(_p_[0]) *_p;\ #define BITXENC(_t_, _p_, _n_, _act_) { _t_ *_p;\
for(_p = _p_; _p != _p_+(_n_&~(4-1)); ) {\ for(_p = _p_; _p != _p_+(_n_&~(4-1)); _p += 4) { XE(0);_act_; XE(1);_act_; XE(2);_act_; XE(3);_act_; }\
_x = (*_p)^_start_; _start_ = *_p++; _act_;\ for( ; _p != _p_+ _n_; _p++ ) { XE(0);_act_; }\
_x = (*_p)^_start_; _start_ = *_p++; _act_;\ }
_x = (*_p)^_start_; _start_ = *_p++; _act_;\ unsigned bitxenc8( uint8_t *in, unsigned n, uint8_t *out, uint8_t start) { uint8_t b = 0,*op = out,x; BITXENC(uint8_t, in, n, b |= x; *op++ = x); return bsr8(b); }
_x = (*_p)^_start_; _start_ = *_p++; _act_;\ unsigned bitxenc16(uint16_t *in, unsigned n, uint16_t *out, uint16_t start) { uint16_t b = 0,*op = out,x; BITXENC(uint16_t, in, n, b |= x; *op++ = x); return bsr16(b); }
}\ unsigned bitxenc32(uint32_t *in, unsigned n, uint32_t *out, uint32_t start) { uint32_t b = 0,*op = out,x; BITXENC(uint32_t, in, n, b |= x; *op++ = x); return bsr32(b); }
while(_p != _p_+_n_) { \ unsigned bitxenc64(uint64_t *in, unsigned n, uint64_t *out, uint64_t start) { uint64_t b = 0,*op = out,x; BITXENC(uint64_t, in, n, b |= x; *op++ = x); return bsr64(b); }
_x = (*_p)^_start_; _start_ = *_p++; _act_;\
}\ #define XD(i) _p[i] = (start ^= _p[i])
#define BITXDEC(_t_, _p_, _n_) { _t_ *_p, _x;\
for(_p = _p_;_p != _p_+(_n_&~(4-1)); _p += 4) { XD(0); XD(1); XD(2); XD(3); }\
for( ;_p != _p_+ _n_ ; _p++ ) XD(0);\
} }
#define BITXDEC(_p_, _n_, _start_) {\ void bitxdec8( uint8_t *p, unsigned n, uint8_t start) { BITXDEC(uint8_t, p, n); }
typeof(_p_[0]) *_p, _z;\ void bitxdec16(uint16_t *p, unsigned n, uint16_t start) { BITXDEC(uint16_t, p, n); }
for(_p = _p_; _p != _p_+(_n_&~(4-1)); ) {\ void bitxdec32(uint32_t *p, unsigned n, uint32_t start) { BITXDEC(uint32_t, p, n); }
_z = *_p; *_p = (_start_ ^= _z); _p++;\ void bitxdec64(uint64_t *p, unsigned n, uint64_t start) { BITXDEC(uint64_t, p, n); }
_z = *_p; *_p = (_start_ ^= _z); _p++;\
_z = *_p; *_p = (_start_ ^= _z); _p++;\ //-------------- For : calc max. bits, min,max value ------------------------
_z = *_p; *_p = (_start_ ^= _z); _p++;\ #define FM(i) mi = _p[i] < mi?_p[i]:mi; mx = _p[i] > mx?_p[i]:mx
}\ #define BITFM(_t_, _p_,_n_) { _t_ *_p; \
while(_p != _p_+_n_) {\ for(_p = _p_, mi = mx = *_p; _p != _p_+(_n_&~(4-1)); _p += 4) { FM(0); FM(1); FM(2); FM(3); }\
_z = *_p; *_p = (_start_ ^= _z); _p++;\ for(;_p != _p_+_n_; _p++) FM(0);\
}\
} }
unsigned bitxenc8( uint8_t *in, unsigned n, uint8_t *out, uint8_t start) { uint8_t b = 0,*op = out; int8_t _x; BITXENC(in, n, start, b |= (uint8_t )_x; *op++ = _x); return bsr8(b); } unsigned bitfm8( uint8_t *in, unsigned n, uint8_t *pmin) { uint8_t mi,mx; BITFM(uint8_t, in, n); *pmin = mi; return bsr8( mx - mi); }
unsigned bitxenc16(uint16_t *in, unsigned n, uint16_t *out, uint16_t start) { uint16_t b = 0,*op = out; int16_t _x; BITXENC(in, n, start, b |= (uint16_t)_x; *op++ = _x); return bsr16(b); } unsigned bitfm16(uint16_t *in, unsigned n, uint16_t *pmin) { uint16_t mi,mx; BITFM(uint16_t, in, n); *pmin = mi; return bsr16(mx - mi); }
unsigned bitxenc64(uint64_t *in, unsigned n, uint64_t *out, uint64_t start) { uint64_t b = 0,*op = out; int64_t _x; BITXENC(in, n, start, b |= (uint64_t)_x; *op++ = _x); return bsr32(b); } unsigned bitfm32(uint32_t *in, unsigned n, uint32_t *pmin) { uint32_t mi,mx; BITFM(uint32_t, in, n); *pmin = mi; return bsr32(mx - mi); }
unsigned bitxenc32(uint32_t *in, unsigned n, uint32_t *out, uint32_t start) { uint32_t b = 0,*op = out; uint32_t _x; BITXENC(in, n, start, b |= (uint32_t)_x; *op++ = _x); return bsr32(b); } unsigned bitfm64(uint64_t *in, unsigned n, uint64_t *pmin) { uint64_t mi,mx; BITFM(uint16_t, in, n); *pmin = mi; return bsr64(mx - mi); }
void bitxdec8( uint8_t *p, unsigned n, uint8_t start) { BITXDEC(p, n, start); }
void bitxdec32(uint32_t *p, unsigned n, uint32_t start) { BITXDEC(p, n, start); }
void bitxdec16(uint16_t *p, unsigned n, uint16_t start) { BITXDEC(p, n, start); }
void bitxdec64(uint64_t *p, unsigned n, uint64_t start) { BITXDEC(p, n, start); }
//-------------- For --------------------------
#define BITFM(_p_,_n_, _mi_, _mx_) {\
typeof(_p_[0]) *_p; \
for(_p = _p_, _mi_ = _mx_ = *_p; _p != _p_+(_n_&~(4-1)); ) {\
if(*_p < _mi_) _mi_ = *_p; if(*_p > _mx_) _mx_ = *_p; _p++; \
if(*_p < _mi_) _mi_ = *_p; if(*_p > _mx_) _mx_ = *_p; _p++; \
if(*_p < _mi_) _mi_ = *_p; if(*_p > _mx_) _mx_ = *_p; _p++; \
if(*_p < _mi_) _mi_ = *_p; if(*_p > _mx_) _mx_ = *_p; _p++; \
}\
while(_p != _p_+_n_) { \
if(*_p < _mi_) _mi_ = *_p; if(*_p > _mx_) _mx_ = *_p; _p++; \
}\
}
unsigned bitfm8( uint8_t *in, unsigned n, uint8_t *pmin) { uint8_t mi,mx; BITFM(in, n, mi, mx); *pmin = mi; return bsr8( mx - mi); }
unsigned bitfm16(uint16_t *in, unsigned n, uint16_t *pmin) { uint16_t mi,mx; BITFM(in, n, mi, mx); *pmin = mi; return bsr16(mx - mi); }
unsigned bitfm32(uint32_t *in, unsigned n, uint32_t *pmin) { uint32_t mi,mx; BITFM(in, n, mi, mx); *pmin = mi; return bsr32(mx - mi); }
unsigned bitfm64(uint64_t *in, unsigned n, uint64_t *pmin) { uint64_t mi,mx; BITFM(in, n, mi, mx); *pmin = mi; return bsr64(mx - mi); }