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

TurboPFor: Bit Packing SIMD/AVX2

Browse Source
This commit is contained in:
x
2017-01-08 11:42:42 +01:00
parent ac5094618d
commit df123e5297
1 changed files with 56 additions and 44 deletions
Download Patch File Download Diff File Expand all files Collapse all files

100
bitpackv.c
View File

@ -22,26 +22,27 @@
- email : powturbo [_AT_] gmail [_DOT_] com
**/
// "Integer Compression" SIMD bit packing
#ifndef VSTI
#include <emmintrin.h>
#include "bitpack.h"
#include "bitutil.h"
#define OPPE(__op)
#define IPPE(__op)
#define PAD8(__x) (((__x)+8-1)/8)
#define VSTI(ip, i, iv, parm)
#define IPP(ip, i, iv) _mm_loadu_si128(ip++)
#include "bitpack128v.c"
#include "bitpack128v_.h"
unsigned char *bitpack128v32(unsigned *__restrict in, unsigned n, unsigned char *__restrict out, unsigned b) { unsigned char *pout = out+PAD8(128*b); BITPACK128V32(in, b, out, 0); return pout; }
//unsigned char *bitpack128v16(unsigned short *__restrict in, unsigned char *__restrict out, unsigned b) { unsigned char *pout = out+PAD8(n*b); BITPACK128V32(in, n, b, out, 0); return pout; }
#undef VSTI
#undef IPP
//------------------------------------------------------------------------------------------------------------------------------
#define VSTI(__ip, __i, __iv, __sv) v = _mm_loadu_si128(__ip++); __iv = DELTA128x32(v,__sv); __sv = v
#define IPP(ip, i, __iv) __iv
#include "bitpack128v.c"
#include "bitpack128v_.h"
unsigned char *bitdpack128v32(unsigned *__restrict in, unsigned n, unsigned char *__restrict out, unsigned start, unsigned b) { unsigned char *pout = out+PAD8(128*b);
__m128i v,sv = _mm_set1_epi32(start);
@ -67,50 +68,61 @@ unsigned char *bitzpack128v32(unsigned *__restrict in, unsigned n, unsigne
return pout;
}
#undef VSTI
#else
#undef IPP
#ifdef __AVX2__
#include <immintrin.h>
#include <strings.h>
#include <emmintrin.h>
#include <tmmintrin.h>
#define OPPE(__op)
#define IPPE(__op)
#include "bitpack128v_.h"
#define PAD8(__x) (((__x)+8-1)/8)
#define OPPE(__op)
#define IPPE(__op)
#define VSTI(ip, i, iv, parm)
#define IPP(ip, i, iv) _mm256_loadu_si256(ip++)
#include "bitpack256v_.h"
//#include "bitpack.h"
//#include "bitutil.h"
#define BITPACK128V32(__pip, __nbits, __pop, __parm) { __m128i *__ip=(__m128i *)__pip,*__op=(__m128i *)__pop;\
switch(__nbits) {\
case 0: break;\
case 1:{ BITPACK128V32_1( __ip, __op, __parm); } break;\
case 2:{ BITPACK128V32_2( __ip, __op, __parm); } break;\
case 3:{ BITPACK128V32_3( __ip, __op, __parm); } break;\
case 4:{ BITPACK128V32_4( __ip, __op, __parm); } break;\
case 5:{ BITPACK128V32_5( __ip, __op, __parm); } break;\
case 6:{ BITPACK128V32_6( __ip, __op, __parm); } break;\
case 7:{ BITPACK128V32_7( __ip, __op, __parm); } break;\
case 8:{ BITPACK128V32_8( __ip, __op, __parm); } break;\
case 9:{ BITPACK128V32_9( __ip, __op, __parm); } break;\
case 10:{ BITPACK128V32_10(__ip, __op, __parm); } break;\
case 11:{ BITPACK128V32_11(__ip, __op, __parm); } break;\
case 12:{ BITPACK128V32_12(__ip, __op, __parm); } break;\
case 13:{ BITPACK128V32_13(__ip, __op, __parm); } break;\
case 14:{ BITPACK128V32_14(__ip, __op, __parm); } break;\
case 15:{ BITPACK128V32_15(__ip, __op, __parm); } break;\
case 16:{ BITPACK128V32_16(__ip, __op, __parm); } break;\
case 17:{ BITPACK128V32_17(__ip, __op, __parm); } break;\
case 18:{ BITPACK128V32_18(__ip, __op, __parm); } break;\
case 19:{ BITPACK128V32_19(__ip, __op, __parm); } break;\
case 20:{ BITPACK128V32_20(__ip, __op, __parm); } break;\
case 21:{ BITPACK128V32_21(__ip, __op, __parm); } break;\
case 22:{ BITPACK128V32_22(__ip, __op, __parm); } break;\
case 23:{ BITPACK128V32_23(__ip, __op, __parm); } break;\
case 24:{ BITPACK128V32_24(__ip, __op, __parm); } break;\
case 25:{ BITPACK128V32_25(__ip, __op, __parm); } break;\
case 26:{ BITPACK128V32_26(__ip, __op, __parm); } break;\
case 27:{ BITPACK128V32_27(__ip, __op, __parm); } break;\
case 28:{ BITPACK128V32_28(__ip, __op, __parm); } break;\
case 29:{ BITPACK128V32_29(__ip, __op, __parm); } break;\
case 30:{ BITPACK128V32_30(__ip, __op, __parm); } break;\
case 31:{ BITPACK128V32_31(__ip, __op, __parm); } break;\
case 32:{ BITPACK128V32_32(__ip, __op, __parm); } break;\
}\
unsigned char *bitpack256v32(unsigned *__restrict in, unsigned n, unsigned char *__restrict out, unsigned b) { unsigned char *pout = out+PAD8(256*b); BITPACK256V32(in, b, out, 0); return pout; }
#undef VSTI
#undef IPP
//------------------------------------------------------------------------------------------------------------------------------
#if 0
#define VSTI(__ip, __i, __iv, __sv) v = _mm256_loadu_si256(__ip++); DELTA256x32(v,__sv, __iv) //__sv = v
#define IPP(ip, i, __iv) __iv
#include "bitpack256v_.h"
unsigned char *bitdpack256v32(unsigned *__restrict in, unsigned char *__restrict out, unsigned start, unsigned b) { unsigned char *pout = out+PAD8(256*b);
__m256i v; //,sv = _mm256_set1_epi32(start),zv = _mm256_setzero_si256();
__m128i sv = _mm_set1_epi32(start);
BITPACK256V32(in, b, out, sv);
return pout;
}
#endif
#undef VSTI
//------------------------------------------------------------------------------------------------------------------------------
#define VSTI(__ip, __i, __iv, __sv) v = _mm256_loadu_si256(__ip++); __iv = _mm256_sub_epi32(DELTA256x32(v,__sv),cv); __sv = v
unsigned char *bitd1pack256v32(unsigned *__restrict in, unsigned char *__restrict out, unsigned start, unsigned b) { unsigned char *pout = out+PAD8(256*b);
__m256i v, sv = _mm256_set1_epi32(start), cv = _mm256_set1_epi32(1);
//BITPACK256V32(in, b, out, sv); return pout;
}
#undef VSTI
//------------------------------------------------------------------------------------------------------------------------------
#define VSTI(__ip, __i, __iv, __sv) v = _mm256_loadu_si256(__ip++); __iv = DELTA256x32(v,__sv); __sv = v; __iv = ZIGZAG256x32(__iv)
unsigned char *bitzpack256v32(unsigned *__restrict in, unsigned char *__restrict out, unsigned start, unsigned b) { unsigned char *pout = out+PAD8(256*b);
__m256i v, sv = _mm256_set1_epi32(start), cv = _mm256_set1_epi32(1);
//BITPACK256V32(in, b, out, sv);
return pout;
}
#endif
#undef VSTI
#endif