From 3ae0a0651806b3e4c0ae91157ec657dabc8461b3 Mon Sep 17 00:00:00 2001 From: x Date: Wed, 16 Oct 2019 19:47:30 +0200 Subject: [PATCH] TurboPFor: Bit Unpacking --- bitunpack.c | 135 ++++++++++++++++++++++++++-------------------------- 1 file changed, 67 insertions(+), 68 deletions(-) diff --git a/bitunpack.c b/bitunpack.c index 77210a5..6fb6238 100644 --- a/bitunpack.c +++ b/bitunpack.c @@ -28,6 +28,7 @@ #include "bitutil.h" #include "bitpack.h" #include "vint.h" + #define PAD8(_x_) (((_x_)+7)/8) #pragma warning( disable : 4005) @@ -48,8 +49,6 @@ typedef unsigned char *(*BITUNPACK_D32)(const unsigned char *__restrict in, unsi typedef unsigned char *(*BITUNPACK_F64)(const unsigned char *__restrict in, unsigned n, uint64_t *__restrict out); typedef unsigned char *(*BITUNPACK_D64)(const unsigned char *__restrict in, unsigned n, uint64_t *__restrict out, uint64_t start); -#define PREFETCH(_ip_) __builtin_prefetch(_ip_+512,0)//#define PREFETCH(ip) - #if !defined(SSE2_ON) && !defined(AVX2_ON) #if 0 //???? @@ -91,7 +90,7 @@ typedef unsigned char *(*BITUNPACK_D64)(const unsigned char *__restrict in, unsi #define BITNUNPACK(in, n, out, _csize_, _usize_) {\ unsigned char *ip = in;\ - for(op = out,out+=n; op < out;) { unsigned oplen = out - op,b; if(oplen > _csize_) oplen = _csize_; PREFETCH(in+512);\ + for(op = out,out+=n; op < out;) { unsigned oplen = out - op,b; if(oplen > _csize_) oplen = _csize_; PREFETCH(ip+512,0);\ b = *ip++; ip = TEMPLATE2(bitunpacka, _usize_)[b](ip, oplen, op);\ op += oplen;\ } \ @@ -101,7 +100,7 @@ typedef unsigned char *(*BITUNPACK_D64)(const unsigned char *__restrict in, unsi #define BITNDUNPACK(in, n, out, _csize_, _usize_, _bitunpacka_) { if(!n) return 0;\ unsigned char *ip = in;\ TEMPLATE2(vbxget, _usize_)(ip, start);\ - for(*out++ = start,--n,op = out; op != out+(n&~(_csize_-1)); ) { PREFETCH(ip+512);\ + for(*out++ = start,--n,op = out; op != out+(n&~(_csize_-1)); ) { PREFETCH(ip+512,0);\ unsigned b = *ip++; ip = TEMPLATE2(_bitunpacka_, _usize_)[b](ip, _csize_, op, start); op += _csize_; start = op[-1];\ } if(n&=(_csize_-1)) { unsigned b = *ip++; ip = TEMPLATE2(_bitunpacka_, _usize_)[b](ip, n, op, start); }\ return ip - in;\ @@ -135,22 +134,22 @@ size_t bitnfunpack64( unsigned char *__restrict in, size_t n, uint64_t *__restri #define _BITNUNPACKV(in, n, out, _csize_, _usize_, _bitunpackv_) {\ unsigned char *ip = in;\ - for(op = out; op != out+(n&~(_csize_-1)); op += _csize_) { PREFETCH(in+512);\ + for(op = out; op != out+(n&~(_csize_-1)); op += _csize_) { PREFETCH(in+512,0);\ unsigned b = *ip++; ip = TEMPLATE2(_bitunpackv_, _usize_)(ip, _csize_, op,b);\ } if(n&=(_csize_-1)) { unsigned b = *ip++; ip = TEMPLATE2(bitunpack, _usize_)(ip, n, op,b); }\ return ip - in;\ -} +} #define _BITNDUNPACKV(in, n, out, _csize_, _usize_, _bitunpackv_, _bitunpack_) { if(!n) return 0;\ unsigned char *ip = in;\ TEMPLATE2(vbxget, _usize_)(ip, start); \ *out++ = start;\ - for(--n,op = out; op != out+(n&~(_csize_-1)); ) { PREFETCH(ip+512);\ + for(--n,op = out; op != out+(n&~(_csize_-1)); ) { PREFETCH(ip+512,0);\ unsigned b = *ip++; ip = TEMPLATE2(_bitunpackv_, _usize_)(ip, _csize_, op, start,b); op += _csize_; start = op[-1];\ } if(n&=(_csize_-1)) { unsigned b = *ip++; ip = TEMPLATE2(_bitunpack_, _usize_)(ip, n, op, start,b); }\ return ip - in;\ } - + #if (defined(__SSE2__) || defined(__ARM_NEON)) && defined(SSE2_ON) #define VO16( _op_, _i_, ov, _parm_) _mm_storeu_si128(_op_++, ov) #define VO32( _op_, _i_, ov, _parm_) _mm_storeu_si128(_op_++, ov) @@ -175,7 +174,7 @@ size_t bitnfunpack64( unsigned char *__restrict in, size_t n, uint64_t *__restri VOZ16(_op_, 13, ov, _parm_);\ VOZ16(_op_, 14, ov, _parm_);\ VOZ16(_op_, 15, ov, _parm_);\ - VOZ16(_op_, 16, ov, _parm_);\ + /*VOZ16(_op_, 16, ov, _parm_);\ VOZ16(_op_, 17, ov, _parm_);\ VOZ16(_op_, 18, ov, _parm_);\ VOZ16(_op_, 19, ov, _parm_);\ @@ -190,7 +189,7 @@ size_t bitnfunpack64( unsigned char *__restrict in, size_t n, uint64_t *__restri VOZ16(_op_, 28, ov, _parm_);\ VOZ16(_op_, 29, ov, _parm_);\ VOZ16(_op_, 30, ov, _parm_);\ - VOZ16(_op_, 31, ov, _parm_);\ + VOZ16(_op_, 31, ov, _parm_);*/\ } #define BITUNBLK128V32_0(ip, _i_, _op_, _parm_) {__m128i ov;\ @@ -260,7 +259,7 @@ unsigned char *bitunpack128v64( const unsigned char *__restrict in, unsigned n, //------------------------------SSE ----------------------------------------------- #if defined(__SSSE3__) || defined(__ARM_NEON) #define _ 0x80 -char _shuffle_32[16][16] = { +ALIGNED(char, _shuffle_32[16][16],16) = { { _,_,_,_, _,_,_,_, _,_, _, _, _, _, _,_ }, { 0,1,2,3, _,_,_,_, _,_, _, _, _, _, _,_ }, { _,_,_,_, 0,1,2,3, _,_, _, _, _, _, _,_ }, @@ -278,7 +277,7 @@ char _shuffle_32[16][16] = { { _,_,_,_, 0,1,2,3, 4,5, 6, 7, 8, 9,10,11 }, { 0,1,2,3, 4,5,6,7, 8,9,10,11, 12,13,14,15 }, }; -char _shuffle_16[256][16] = { +ALIGNED(char, _shuffle_16[256][16],16) = { { _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _ }, { 0, 1, _, _, _, _, _, _, _, _, _, _, _, _, _, _ }, { _, _, 0, 1, _, _, _, _, _, _, _, _, _, _, _, _ }, @@ -539,10 +538,10 @@ char _shuffle_16[256][16] = { #undef _ #endif -#define VO16( _op_, _i_, _ov_, _parm_) m = *bb++; _mm_storeu_si128(_op_++, _mm_add_epi16(_ov_, _mm_shuffle_epi8(_mm_slli_epi16(_mm_loadu_si128((__m128i*)pex), b), _mm_load_si128((__m128i*)_shuffle_16[m]) ) )); pex += popcnt32(m) -#define VO32( _op_, _i_, _ov_, _parm_) if((_i_) & 1) m = (*bb++) >> 4; else m = (*bb) & 0xf; _mm_storeu_si128(_op_++, _mm_add_epi32(_ov_, _mm_shuffle_epi8(_mm_slli_epi32(_mm_loadu_si128((__m128i*)pex), b), _mm_load_si128((__m128i*)_shuffle_32[m]) ) )); pex += popcnt32(m) -#define VOZ16(_op_, _i_, _ov_, _parm_) m = *bb++; _mm_storeu_si128(_op_++, _mm_shuffle_epi8( _mm_loadu_si128((__m128i*)pex), _mm_load_si128((__m128i*)_shuffle_16[m]) ) ); pex += popcnt32(m) -#define VOZ32(_op_, _i_, _ov_, _parm_) if((_i_) & 1) m = (*bb++) >> 4; else m = (*bb) & 0xf; _mm_storeu_si128(_op_++, _mm_shuffle_epi8( _mm_loadu_si128((__m128i*)pex), _mm_load_si128((__m128i*)_shuffle_32[m]) ) ); pex += popcnt32(m) +#define VO16( _op_, _i_, _ov_, _parm_) m = *bb++; _mm_storeu_si128(_op_++, _mm_add_epi16(_ov_, _mm_shuffle_epi8(_mm_slli_epi16(_mm_loadu_si128((__m128i*)pex), b), _mm_loadu_si128((__m128i*)_shuffle_16[m]) ) )); pex += popcnt32(m) +#define VO32( _op_, _i_, _ov_, _parm_) if((_i_) & 1) m = (*bb++) >> 4; else m = (*bb) & 0xf; _mm_storeu_si128(_op_++, _mm_add_epi32(_ov_, _mm_shuffle_epi8(_mm_slli_epi32(_mm_loadu_si128((__m128i*)pex), b), _mm_loadu_si128((__m128i*)_shuffle_32[m]) ) )); pex += popcnt32(m) +#define VOZ16(_op_, _i_, _ov_, _parm_) m = *bb++; _mm_storeu_si128(_op_++, _mm_shuffle_epi8( _mm_loadu_si128((__m128i*)pex), _mm_loadu_si128((__m128i*)_shuffle_16[m]) ) ); pex += popcnt32(m) +#define VOZ32(_op_, _i_, _ov_, _parm_) if((_i_) & 1) m = (*bb++) >> 4; else m = (*bb) & 0xf; _mm_storeu_si128(_op_++, _mm_shuffle_epi8( _mm_loadu_si128((__m128i*)pex), _mm_loadu_si128((__m128i*)_shuffle_32[m]) ) ); pex += popcnt32(m) #define BITUNPACK0(_parm_) //_parm_ = _mm_setzero_si128() #include "bitunpack_.h" @@ -562,8 +561,8 @@ unsigned char *_bitunpack256w32( const unsigned char *__restrict in, unsigned n, //#define STOZ64(_op_, _ov_) _mm_storeu_si128(_op_++, _ov_); _mm_storeu_si128(_op_++, _ov_) #define STO64( _op_, _ov_, _zv_) _mm_storeu_si128(_op_++, _mm_unpacklo_epi32(_ov_,_zv_));_mm_storeu_si128(_op_++, _mm_unpacklo_epi32(_mm_srli_si128(_ov_,8),_zv_)) -#define VO32( _op_, _i_, _ov_, _parm_) if((_i_) & 1) m = (*bb++) >> 4; else m = (*bb) & 0xf; { __m128i _wv = _mm_add_epi32(_ov_, _mm_shuffle_epi8(_mm_slli_epi32(_mm_loadu_si128((__m128i*)pex), b), _mm_load_si128((__m128i*)_shuffle_32[m]) ) ); STO64(_op_, _wv, zv);} pex += popcnt32(m) -#define VOZ32(_op_, _i_, _ov_, _parm_) if((_i_) & 1) m = (*bb++) >> 4; else m = (*bb) & 0xf; { __m128i _wv = _mm_shuffle_epi8( _mm_loadu_si128((__m128i*)pex), _mm_load_si128((__m128i*)_shuffle_32[m]) ) ; STO64(_op_, _wv, zv);} pex += popcnt32(m) +#define VO32( _op_, _i_, _ov_, _parm_) if((_i_) & 1) m = (*bb++) >> 4; else m = (*bb) & 0xf; { __m128i _wv = _mm_add_epi32(_ov_, _mm_shuffle_epi8(_mm_slli_epi32(_mm_loadu_si128((__m128i*)pex), b), _mm_loadu_si128((__m128i*)_shuffle_32[m]) ) ); STO64(_op_, _wv, zv);} pex += popcnt32(m) +#define VOZ32(_op_, _i_, _ov_, _parm_) if((_i_) & 1) m = (*bb++) >> 4; else m = (*bb) & 0xf; { __m128i _wv = _mm_shuffle_epi8( _mm_loadu_si128((__m128i*)pex), _mm_loadu_si128((__m128i*)_shuffle_32[m]) ) ; STO64(_op_, _wv, zv);} pex += popcnt32(m) #define BITUNPACK0(_parm_) #include "bitunpack_.h" @@ -580,8 +579,8 @@ unsigned char *_bitunpack128v64( const unsigned char *__restrict in, unsigned n, //------------------------------------------------------------------- #define VOZ16(_op_, _i_, _ov_, _parm_) _mm_storeu_si128(_op_++, _parm_) #define VOZ32(_op_, _i_, _ov_, _parm_) _mm_storeu_si128(_op_++, _parm_) -#define VO16( _op_, _i_, _ov_, _sv_) _ov_ = UNZIGZAG128x16(_ov_); SCAN128x16(_ov_,_sv_); _mm_storeu_si128(_op_++, _sv_) -#define VO32( _op_, _i_, _ov_, _sv_) _ov_ = UNZIGZAG128x32(_ov_); SCAN128x32(_ov_,_sv_); _mm_storeu_si128(_op_++, _sv_) +#define VO16( _op_, _i_, _ov_, _sv_) _ov_ = mm_zzagd_epi16(_ov_); _sv_ = mm_scan_epi16(_ov_,_sv_); _mm_storeu_si128(_op_++, _sv_) +#define VO32( _op_, _i_, _ov_, _sv_) _ov_ = mm_zzagd_epi32(_ov_); _sv_ = mm_scan_epi32(_ov_,_sv_); _mm_storeu_si128(_op_++, _sv_) #include "bitunpack_.h" #define BITUNPACK0(_parm_) unsigned char *bitzunpack128v16( const unsigned char *__restrict in, unsigned n, unsigned short *__restrict out, unsigned short start, unsigned b) { @@ -591,8 +590,8 @@ unsigned char *bitzunpack128v32( const unsigned char *__restrict in, unsigned n, const unsigned char *ip = in+PAD8(128*b); __m128i sv = _mm_set1_epi32(start); BITUNPACK128V32(in, b, out, sv); return (unsigned char *)ip; } -#define VO32(_op_, i, _ov_, _sv_) SCAN128x32(_ov_,_sv_); _mm_storeu_si128(_op_++, _sv_) -#define VO16(_op_, i, _ov_, _sv_) SCAN128x16(_ov_,_sv_); _mm_storeu_si128(_op_++, _sv_) +#define VO32(_op_, i, _ov_, _sv_) _sv_ = mm_scan_epi32(_ov_,_sv_); _mm_storeu_si128(_op_++, _sv_) +#define VO16(_op_, i, _ov_, _sv_) _sv_ = mm_scan_epi16(_ov_,_sv_); _mm_storeu_si128(_op_++, _sv_) #include "bitunpack_.h" #define BITUNPACK0(_parm_) unsigned char *bitdunpack128v16( const unsigned char *__restrict in, unsigned n, uint16_t *__restrict out, uint16_t start, unsigned b) { @@ -615,15 +614,15 @@ unsigned char *bitfunpack128v32( const unsigned char *__restrict in, unsigned n, //---------------------------- SSE ---------------------------------------------- #if defined(__SSSE3__) || defined(__ARM_NEON) -#define VX32(_i_, _ov_) if(!((_i_) & 1)) m = (*bb) & 0xf;else m = (*bb++) >> 4; _ov_ = _mm_add_epi32(_ov_, _mm_shuffle_epi8(_mm_slli_epi32(_mm_loadu_si128((__m128i*)pex), b), _mm_load_si128((__m128i*)_shuffle_32[m]))); pex += popcnt32(m) -#define VXZ32(_i_, _ov_) if(!((_i_) & 1)) m = (*bb) & 0xf;else m = (*bb++) >> 4; _ov_ = _mm_shuffle_epi8( _mm_loadu_si128((__m128i*)pex), _mm_load_si128((__m128i*)_shuffle_32[m])); pex += popcnt32(m) -#define VO32( _op_, _i_, _ov_, _sv_) VX32( _i_, _ov_); SCAN128x32(_ov_,_sv_); _mm_storeu_si128(_op_++, _sv_); -#define VOZ32(_op_, _i_, _ov_, _sv_) VXZ32( _i_, _ov_); SCAN128x32(_ov_,_sv_); _mm_storeu_si128(_op_++, _sv_); +#define VX32(_i_, _ov_) if(!((_i_) & 1)) m = (*bb) & 0xf;else m = (*bb++) >> 4; _ov_ = _mm_add_epi32(_ov_, _mm_shuffle_epi8(_mm_slli_epi32(_mm_loadu_si128((__m128i*)pex), b), _mm_loadu_si128((__m128i*)_shuffle_32[m]))); pex += popcnt32(m) +#define VXZ32(_i_, _ov_) if(!((_i_) & 1)) m = (*bb) & 0xf;else m = (*bb++) >> 4; _ov_ = _mm_shuffle_epi8( _mm_loadu_si128((__m128i*)pex), _mm_loadu_si128((__m128i*)_shuffle_32[m])); pex += popcnt32(m) +#define VO32( _op_, _i_, _ov_, _sv_) VX32( _i_, _ov_); _sv_ = mm_scan_epi32(_ov_,_sv_); _mm_storeu_si128(_op_++, _sv_); +#define VOZ32(_op_, _i_, _ov_, _sv_) VXZ32(_i_, _ov_); _sv_ = mm_scan_epi32(_ov_,_sv_); _mm_storeu_si128(_op_++, _sv_); -#define VX16(_i_, _ov_) m = *bb++; _ov_ = _mm_add_epi16(_ov_, _mm_shuffle_epi8(_mm_slli_epi16(_mm_loadu_si128((__m128i*)pex), b), _mm_load_si128((__m128i*)_shuffle_16[m]) ) ); pex += popcnt32(m) -#define VXZ16(_i_, _ov_) m = *bb++; _ov_ = _mm_shuffle_epi8( _mm_loadu_si128((__m128i*)pex), _mm_load_si128((__m128i*)_shuffle_16[m]) ); pex += popcnt32(m) -#define VO16( _op_, _i_, _ov_, _sv_) VX16( _i_, _ov_); SCAN128x16(_ov_,_sv_); _mm_storeu_si128(_op_++, _sv_); -#define VOZ16(_op_, _i_, _ov_, _sv_) VXZ16( _i_, _ov_); SCAN128x16(_ov_,_sv_); _mm_storeu_si128(_op_++, _sv_); +#define VX16(_i_, _ov_) m = *bb++; _ov_ = _mm_add_epi16(_ov_, _mm_shuffle_epi8(_mm_slli_epi16(_mm_loadu_si128((__m128i*)pex), b), _mm_loadu_si128((__m128i*)_shuffle_16[m]) ) ); pex += popcnt32(m) +#define VXZ16(_i_, _ov_) m = *bb++; _ov_ = _mm_shuffle_epi8( _mm_loadu_si128((__m128i*)pex), _mm_loadu_si128((__m128i*)_shuffle_16[m]) ); pex += popcnt32(m) +#define VO16( _op_, _i_, _ov_, _sv_) VX16( _i_, _ov_); _sv_ = mm_scan_epi16(_ov_,_sv_); _mm_storeu_si128(_op_++, _sv_); +#define VOZ16(_op_, _i_, _ov_, _sv_) VXZ16( _i_, _ov_); _sv_ = mm_scan_epi16(_ov_,_sv_); _mm_storeu_si128(_op_++, _sv_); #include "bitunpack_.h" #define BITUNPACK0(_parm_) unsigned char *_bitdunpack128v16( const unsigned char *__restrict in, unsigned n, unsigned short *__restrict out, unsigned short start, unsigned b, unsigned short *__restrict pex, unsigned char *bb) { @@ -634,21 +633,21 @@ unsigned char *_bitdunpack128v32( const unsigned char *__restrict in, unsigned n } /* -#define VO32( _op_, _i_, _ov_, _sv_) VX32( _i_, _ov_); SCAN128x32(_ov_,_sv_); STO64( _op_, _sv_) //_mm_storeu_si128(_op_++, _sv_); -#define VOZ32(_op_, _i_, _ov_, _sv_) VXZ32( _i_, _ov_); SCAN128x32(_ov_,_sv_); STOZ64( _op_, _sv_, zv) //_mm_storeu_si128(_op_++, _sv_); +#define VO32( _op_, _i_, _ov_, _sv_) VX32( _i_, _ov_); mm_scan_epi32(_ov_,_sv_); STO64( _op_, _sv_) //_mm_storeu_si128(_op_++, _sv_); +#define VOZ32(_op_, _i_, _ov_, _sv_) VXZ32( _i_, _ov_); mm_scan_epi32(_ov_,_sv_); STOZ64( _op_, _sv_, zv) //_mm_storeu_si128(_op_++, _sv_); unsigned char *_bitdunpack128v64( const unsigned char *__restrict in, unsigned n, uint64_t *__restrict out, uint64_t start, unsigned b, uint64_t *__restrict pex, unsigned char *bb) { const unsigned char *ip = in+PAD8(128*b); unsigned m; __m128i sv = _mm_set1_epi32(start),zv = _mm_setzero_si128(); BITUNPACK128V32(in, b, out, sv); return (unsigned char *)ip; }*/ -#define VX16(_i_, _ov_) m = *bb++; _ov_ = _mm_add_epi16(_ov_, _mm_shuffle_epi8(_mm_slli_epi16(_mm_loadu_si128((__m128i*)pex), b), _mm_load_si128((__m128i*)_shuffle_16[m]) ) ); pex += popcnt32(m) -#define VXZ16(_i_, _ov_) m = *bb++; _ov_ = _mm_shuffle_epi8( _mm_loadu_si128((__m128i*)pex), _mm_load_si128((__m128i*)_shuffle_16[m]) ); pex += popcnt32(m) -#define VO16( _op_, _i_, _ov_, _sv_) VX16( _i_, _ov_); _ov_ = UNZIGZAG128x16(_ov_); SCAN128x16(_ov_,_sv_); _mm_storeu_si128(_op_++, _sv_); -#define VOZ16(_op_, _i_, _ov_, _sv_) VXZ16( _i_, _ov_); _ov_ = UNZIGZAG128x16(_ov_); SCAN128x16(_ov_,_sv_); _mm_storeu_si128(_op_++, _sv_); +#define VX16(_i_, _ov_) m = *bb++; _ov_ = _mm_add_epi16(_ov_, _mm_shuffle_epi8(_mm_slli_epi16(_mm_loadu_si128((__m128i*)pex), b), _mm_loadu_si128((__m128i*)_shuffle_16[m]) ) ); pex += popcnt32(m) +#define VXZ16(_i_, _ov_) m = *bb++; _ov_ = _mm_shuffle_epi8( _mm_loadu_si128((__m128i*)pex), _mm_loadu_si128((__m128i*)_shuffle_16[m]) ); pex += popcnt32(m) +#define VO16( _op_, _i_, _ov_, _sv_) VX16( _i_, _ov_); _ov_ = mm_zzagd_epi16(_ov_); _sv_ = mm_scan_epi16(_ov_,_sv_); _mm_storeu_si128(_op_++, _sv_); +#define VOZ16(_op_, _i_, _ov_, _sv_) VXZ16( _i_, _ov_); _ov_ = mm_zzagd_epi16(_ov_); _sv_ = mm_scan_epi16(_ov_,_sv_); _mm_storeu_si128(_op_++, _sv_); -#define VX32(_i_, _ov_) if(!((_i_) & 1)) m = (*bb) & 0xf;else m = (*bb++) >> 4; _ov_ = _mm_add_epi32(_ov_, _mm_shuffle_epi8(_mm_slli_epi32(_mm_loadu_si128((__m128i*)pex), b), _mm_load_si128((__m128i*)_shuffle_32[m]) ) ); pex += popcnt32(m) -#define VXZ32(_i_, _ov_) if(!((_i_) & 1)) m = (*bb) & 0xf;else m = (*bb++) >> 4; _ov_ = _mm_shuffle_epi8( _mm_loadu_si128((__m128i*)pex), _mm_load_si128((__m128i*)_shuffle_32[m]) ); pex += popcnt32(m) -#define VO32( _op_, _i_, _ov_, _sv_) VX32( _i_, _ov_); _ov_ = UNZIGZAG128x32(_ov_); SCAN128x32(_ov_,_sv_); _mm_storeu_si128(_op_++, _sv_); -#define VOZ32(_op_, _i_, _ov_, _sv_) VXZ32(_i_, _ov_); _ov_ = UNZIGZAG128x32(_ov_); SCAN128x32(_ov_,_sv_); _mm_storeu_si128(_op_++, _sv_); +#define VX32(_i_, _ov_) if(!((_i_) & 1)) m = (*bb) & 0xf;else m = (*bb++) >> 4; _ov_ = _mm_add_epi32(_ov_, _mm_shuffle_epi8(_mm_slli_epi32(_mm_loadu_si128((__m128i*)pex), b), _mm_loadu_si128((__m128i*)_shuffle_32[m]) ) ); pex += popcnt32(m) +#define VXZ32(_i_, _ov_) if(!((_i_) & 1)) m = (*bb) & 0xf;else m = (*bb++) >> 4; _ov_ = _mm_shuffle_epi8( _mm_loadu_si128((__m128i*)pex), _mm_loadu_si128((__m128i*)_shuffle_32[m]) ); pex += popcnt32(m) +#define VO32( _op_, _i_, _ov_, _sv_) VX32( _i_, _ov_); _ov_ = mm_zzagd_epi32(_ov_); _sv_ = mm_scan_epi32(_ov_,_sv_); _mm_storeu_si128(_op_++, _sv_); +#define VOZ32(_op_, _i_, _ov_, _sv_) VXZ32(_i_, _ov_); _ov_ = mm_zzagd_epi32(_ov_); _sv_ = mm_scan_epi32(_ov_,_sv_); _mm_storeu_si128(_op_++, _sv_); #include "bitunpack_.h" #define BITUNPACK0(_parm_) @@ -660,8 +659,8 @@ unsigned char *_bitzunpack128v32( const unsigned char *__restrict in, unsigned n } #endif -#define VO16(_op_, i, _ov_, _sv_) SCANI128x16(_ov_,_sv_,cv); _mm_storeu_si128(_op_++, _sv_); -#define VO32(_op_, i, _ov_, _sv_) SCANI128x32(_ov_,_sv_,cv); _mm_storeu_si128(_op_++, _sv_); +#define VO16(_op_, i, _ov_, _sv_) _sv_ = mm_scani_epi16(_ov_,_sv_,cv); _mm_storeu_si128(_op_++, _sv_); +#define VO32(_op_, i, _ov_, _sv_) _sv_ = mm_scani_epi32(_ov_,_sv_,cv); _mm_storeu_si128(_op_++, _sv_); #define VOZ16(_op_, _i_, ov, _parm_) _mm_storeu_si128(_op_++, _parm_); _parm_ = _mm_add_epi16(_parm_, cv) #define VOZ32(_op_, _i_, ov, _parm_) _mm_storeu_si128(_op_++, _parm_); _parm_ = _mm_add_epi32(_parm_, cv) #include "bitunpack_.h" @@ -674,8 +673,8 @@ unsigned char *bitd1unpack128v32( const unsigned char *__restrict in, unsigned n const unsigned char *ip = in+PAD8(128*b); __m128i sv = _mm_set1_epi32(start), cv = _mm_set_epi32(4,3,2,1); BITUNPACK128V32(in, b, out, sv); return (unsigned char *)ip; } -#define VO16(_op_, i, _ov_, _sv_) ADDI128x16(_ov_,_sv_,cv); _mm_storeu_si128(_op_++, _sv_); -#define VO32(_op_, i, _ov_, _sv_) ADDI128x32(_ov_,_sv_,cv); _mm_storeu_si128(_op_++, _sv_); +#define VO16(_op_, i, _ov_, _sv_) ADDI16x8(_ov_,_sv_,cv); _mm_storeu_si128(_op_++, _sv_); +#define VO32(_op_, i, _ov_, _sv_) ADDI32x4(_ov_,_sv_,cv); _mm_storeu_si128(_op_++, _sv_); #define VOZ16(_op_, _i_, ov, _parm_) _mm_storeu_si128(_op_++, _parm_); _parm_ = _mm_add_epi16(_parm_, cv) #define VOZ32(_op_, _i_, ov, _parm_) _mm_storeu_si128(_op_++, _parm_); _parm_ = _mm_add_epi32(_parm_, cv) #include "bitunpack_.h" @@ -701,15 +700,15 @@ unsigned char *bitf1unpack128v32( const unsigned char *__restrict in, unsigned n } #if defined(__SSSE3__) || defined(__ARM_NEON) -#define VX16(_i_, _ov_) m = *bb++; _ov_ = _mm_add_epi16(_ov_, _mm_shuffle_epi8(_mm_slli_epi16(_mm_loadu_si128((__m128i*)pex), b), _mm_load_si128((__m128i*)_shuffle_16[m]))); pex += popcnt32(m) -#define VX32(_i_, _ov_) if(!((_i_) & 1)) m = (*bb) & 0xf;else m = (*bb++) >> 4; _ov_ = _mm_add_epi32(_ov_, _mm_shuffle_epi8(_mm_slli_epi32(_mm_loadu_si128((__m128i*)pex), b), _mm_load_si128((__m128i*)_shuffle_32[m]))); pex += popcnt32(m) -#define VXZ16(_i_, _ov_) m = *bb++; _ov_ = _mm_shuffle_epi8( _mm_loadu_si128((__m128i*)pex), _mm_load_si128((__m128i*)_shuffle_16[m])); pex += popcnt32(m) -#define VXZ32(_i_, _ov_) if(!((_i_) & 1)) m = (*bb) & 0xf;else m = (*bb++) >> 4; _ov_ = _mm_shuffle_epi8( _mm_loadu_si128((__m128i*)pex), _mm_load_si128((__m128i*)_shuffle_32[m])); pex += popcnt32(m) +#define VX16(_i_, _ov_) m = *bb++; _ov_ = _mm_add_epi16(_ov_, _mm_shuffle_epi8(_mm_slli_epi16(_mm_loadu_si128((__m128i*)pex), b), _mm_loadu_si128((__m128i*)_shuffle_16[m]))); pex += popcnt32(m) +#define VX32(_i_, _ov_) if(!((_i_) & 1)) m = (*bb) & 0xf;else m = (*bb++) >> 4; _ov_ = _mm_add_epi32(_ov_, _mm_shuffle_epi8(_mm_slli_epi32(_mm_loadu_si128((__m128i*)pex), b), _mm_loadu_si128((__m128i*)_shuffle_32[m]))); pex += popcnt32(m) +#define VXZ16(_i_, _ov_) m = *bb++; _ov_ = _mm_shuffle_epi8( _mm_loadu_si128((__m128i*)pex), _mm_loadu_si128((__m128i*)_shuffle_16[m])); pex += popcnt32(m) +#define VXZ32(_i_, _ov_) if(!((_i_) & 1)) m = (*bb) & 0xf;else m = (*bb++) >> 4; _ov_ = _mm_shuffle_epi8( _mm_loadu_si128((__m128i*)pex), _mm_loadu_si128((__m128i*)_shuffle_32[m])); pex += popcnt32(m) -#define VO16( _op_, _i_, _ov_, _sv_) VX16( _i_, _ov_); SCANI128x16(_ov_,_sv_,cv); _mm_storeu_si128(_op_++, _sv_); -#define VOZ16(_op_, _i_, _ov_, _sv_) VXZ16( _i_, _ov_); SCANI128x16(_ov_,_sv_,cv); _mm_storeu_si128(_op_++, _sv_); -#define VO32( _op_, _i_, _ov_, _sv_) VX32( _i_, _ov_); SCANI128x32(_ov_,_sv_,cv); _mm_storeu_si128(_op_++, _sv_); -#define VOZ32(_op_, _i_, _ov_, _sv_) VXZ32( _i_, _ov_); SCANI128x32(_ov_,_sv_,cv); _mm_storeu_si128(_op_++, _sv_); +#define VO16( _op_, _i_, _ov_, _sv_) VX16( _i_, _ov_); _sv_ = mm_scani_epi16(_ov_,_sv_,cv); _mm_storeu_si128(_op_++, _sv_); +#define VOZ16(_op_, _i_, _ov_, _sv_) VXZ16( _i_, _ov_); _sv_ = mm_scani_epi16(_ov_,_sv_,cv); _mm_storeu_si128(_op_++, _sv_); +#define VO32( _op_, _i_, _ov_, _sv_) VX32( _i_, _ov_); _sv_ = mm_scani_epi32(_ov_,_sv_,cv); _mm_storeu_si128(_op_++, _sv_); +#define VOZ32(_op_, _i_, _ov_, _sv_) VXZ32( _i_, _ov_); _sv_ = mm_scani_epi32(_ov_,_sv_,cv); _mm_storeu_si128(_op_++, _sv_); #include "bitunpack_.h" #define BITUNPACK0(_parm_) mv = _mm_setzero_si128() //_parm_ = _mm_setzero_si128() @@ -721,10 +720,10 @@ unsigned char *_bitd1unpack128v32( const unsigned char *__restrict in, unsigned const unsigned char *ip = in+PAD8(128*b); unsigned m; __m128i sv = _mm_set1_epi32(start), cv = _mm_set_epi32( 4,3,2,1); BITUNPACK128V32(in, b, out, sv); return (unsigned char *)ip; } -#define VO16( _op_, _i_, _ov_, _sv_) VX16( _i_, _ov_); ADDI128x16(_ov_,_sv_,cv); _mm_storeu_si128(_op_++, _sv_); -#define VOZ16(_op_, _i_, _ov_, _sv_) VXZ16( _i_, _ov_); ADDI128x16(_ov_,_sv_,cv); _mm_storeu_si128(_op_++, _sv_); -#define VO32( _op_, _i_, _ov_, _sv_) VX32( _i_, _ov_); ADDI128x32(_ov_,_sv_,cv); _mm_storeu_si128(_op_++, _sv_); -#define VOZ32(_op_, _i_, _ov_, _sv_) VXZ32( _i_, _ov_); ADDI128x32(_ov_,_sv_,cv); _mm_storeu_si128(_op_++, _sv_); +#define VO16( _op_, _i_, _ov_, _sv_) VX16( _i_, _ov_); ADDI16x8(_ov_,_sv_,cv); _mm_storeu_si128(_op_++, _sv_); +#define VOZ16(_op_, _i_, _ov_, _sv_) VXZ16( _i_, _ov_); ADDI16x8(_ov_,_sv_,cv); _mm_storeu_si128(_op_++, _sv_); +#define VO32( _op_, _i_, _ov_, _sv_) VX32( _i_, _ov_); ADDI32x4(_ov_,_sv_,cv); _mm_storeu_si128(_op_++, _sv_); +#define VOZ32(_op_, _i_, _ov_, _sv_) VXZ32( _i_, _ov_); ADDI32x4(_ov_,_sv_,cv); _mm_storeu_si128(_op_++, _sv_); #include "bitunpack_.h" #define BITUNPACK0(_parm_) mv = _mm_setzero_si128() //_parm_ = _mm_setzero_si128() @@ -1092,23 +1091,23 @@ unsigned char *_bitunpack256v32( const unsigned char *__restrict in, unsigned n, } #define VOZ32(_op_, _i_, ov, _parm_) _mm256_storeu_si256(_op_++, _parm_) -#define VO32(_op_, i, _ov_, _sv_) _ov_ = UNZIGZAG256x32(_ov_); SCAN256x32(_ov_,_sv_); _mm256_storeu_si256(_op_++, _sv_) +#define VO32(_op_, i, _ov_, _sv_) _ov_ = mm256_zzagd_epi32(_ov_); _sv_ = mm256_scan_epi32(_ov_,_sv_); _mm256_storeu_si256(_op_++, _sv_) #include "bitunpack_.h" #define BITUNPACK0(_parm_) unsigned char *bitzunpack256v32( const unsigned char *__restrict in, unsigned n, unsigned *__restrict out, unsigned start, unsigned b) { const unsigned char *ip = in+PAD8(256*b); - __m256i sv = _mm256_set1_epi32(start), zv = _mm256_setzero_si256(); + __m256i sv = _mm256_set1_epi32(start);//, zv = _mm256_setzero_si256(); BITUNPACK256V32(in, b, out, sv); return (unsigned char *)ip; } -#define VO32(_op_, i, _ov_, _sv_) SCAN256x32(_ov_,_sv_); _mm256_storeu_si256(_op_++, _sv_) +#define VO32(_op_, i, _ov_, _sv_) _sv_ = mm256_scan_epi32(_ov_,_sv_); _mm256_storeu_si256(_op_++, _sv_) #include "bitunpack_.h" #define BITUNPACK0(_parm_) unsigned char *bitdunpack256v32( const unsigned char *__restrict in, unsigned n, unsigned *__restrict out, unsigned start, unsigned b) { const unsigned char *ip = in+PAD8(256*b); - __m256i sv = _mm256_set1_epi32(start), zv = _mm256_setzero_si256(); + __m256i sv = _mm256_set1_epi32(start);// zv = _mm256_setzero_si256(); BITUNPACK256V32(in, b, out, sv); return (unsigned char *)ip; } @@ -1126,21 +1125,21 @@ unsigned char *bitfunpack256v32( const unsigned char *__restrict in, unsigned n, #define VX32(_i_, _ov_) xm = *bb++; _ov_ = _mm256_add_epi32(_ov_, _mm256_slli_epi32(mm256_maskz_loadu_epi32(xm,(__m256i*)pex), b) ); pex += popcnt32(xm) #define VXZ32(_i_, _ov_) xm = *bb++; _ov_ = mm256_maskz_loadu_epi32(xm,(__m256i*)pex); pex += popcnt32(xm) -#define VO32( _op_, _i_, _ov_, _sv_) VX32( _i_, _ov_); SCAN256x32(_ov_,_sv_); _mm256_storeu_si256(_op_++, _sv_); -#define VOZ32(_op_, _i_, _ov_, _sv_) VXZ32(_i_, _ov_); SCAN256x32(_ov_,_sv_); _mm256_storeu_si256(_op_++, _sv_); +#define VO32( _op_, _i_, _ov_, _sv_) VX32( _i_, _ov_); _sv_ = mm256_scan_epi32(_ov_,_sv_); _mm256_storeu_si256(_op_++, _sv_); +#define VOZ32(_op_, _i_, _ov_, _sv_) VXZ32(_i_, _ov_); _sv_ = mm256_scan_epi32(_ov_,_sv_); _mm256_storeu_si256(_op_++, _sv_); #include "bitunpack_.h" #define BITUNPACK0(_parm_) unsigned char *_bitdunpack256v32( const unsigned char *__restrict in, unsigned n, unsigned *__restrict out, unsigned start, unsigned b, unsigned *__restrict pex, unsigned char *bb) { const unsigned char *ip = in+PAD8(256*b); unsigned xm; - __m256i sv = _mm256_set1_epi32(start),zv = _mm256_setzero_si256(), tv = _mm256_set_epi32(0,1,2,3,4,5,6,7); + __m256i sv = _mm256_set1_epi32(start), tv = _mm256_set_epi32(0,1,2,3,4,5,6,7),zv = _mm256_setzero_si256(); BITUNPACK256V32(in, b, out, sv); return (unsigned char *)ip; } #define VX32(_i_, _ov_) xm = *bb++; _ov_ = _mm256_add_epi32(_ov_, _mm256_slli_epi32(mm256_maskz_loadu_epi32(xm,(__m256i*)pex), b) ); pex += popcnt32(xm) #define VXZ32(_i_, _ov_) xm = *bb++; _ov_ = mm256_maskz_loadu_epi32(xm,(__m256i*)pex); pex += popcnt32(xm) -#define VO32( _op_, _i_, _ov_, _sv_) VX32( _i_, _ov_); _ov_ = UNZIGZAG256x32(_ov_); SCAN256x32(_ov_,_sv_); _mm256_storeu_si256(_op_++, _sv_); -#define VOZ32(_op_, _i_, _ov_, _sv_) VXZ32(_i_, _ov_); _ov_ = UNZIGZAG256x32(_ov_); SCAN256x32(_ov_,_sv_); _mm256_storeu_si256(_op_++, _sv_); +#define VO32( _op_, _i_, _ov_, _sv_) VX32( _i_, _ov_); _ov_ = mm256_zzagd_epi32(_ov_); _sv_ = mm256_scan_epi32(_ov_,_sv_); _mm256_storeu_si256(_op_++, _sv_); +#define VOZ32(_op_, _i_, _ov_, _sv_) VXZ32(_i_, _ov_); _ov_ = mm256_zzagd_epi32(_ov_); _sv_ = mm256_scan_epi32(_ov_,_sv_); _mm256_storeu_si256(_op_++, _sv_); #include "bitunpack_.h" #define BITUNPACK0(_parm_) @@ -1148,7 +1147,7 @@ unsigned char *_bitzunpack256v32( const unsigned char *__restrict in, unsigned n const unsigned char *ip = in+PAD8(256*b); unsigned xm; __m256i sv = _mm256_set1_epi32(start),zv = _mm256_setzero_si256(), tv = _mm256_set_epi32(0,1,2,3,4,5,6,7); BITUNPACK256V32(in, b, out, sv); return (unsigned char *)ip; } -#define VO32(_op_, i, _ov_, _sv_) SCANI256x32(_ov_,_sv_,cv); _mm256_storeu_si256(_op_++, _sv_); +#define VO32(_op_, i, _ov_, _sv_) _sv_ = mm256_scani_epi32(_ov_,_sv_,cv); _mm256_storeu_si256(_op_++, _sv_); #define VOZ32(_op_, _i_, ov, _parm_) _mm256_storeu_si256(_op_++, _parm_); _parm_ = _mm256_add_epi32(_parm_, cv) #include "bitunpack_.h" #define BITUNPACK0(_parm_) _parm_ = _mm256_add_epi32(_parm_, cv); cv = _mm256_set1_epi32(8) @@ -1170,8 +1169,8 @@ unsigned char *bitf1unpack256v32( const unsigned char *__restrict in, unsigned n return (unsigned char *)ip; } -#define VO32( _op_, _i_, _ov_, _sv_) VX32( _i_, _ov_); SCANI256x32(_ov_,_sv_,cv); _mm256_storeu_si256(_op_++, _sv_); -#define VOZ32(_op_, _i_, _ov_, _sv_) VXZ32(_i_, _ov_); SCANI256x32(_ov_,_sv_,cv); _mm256_storeu_si256(_op_++, _sv_); +#define VO32( _op_, _i_, _ov_, _sv_) VX32( _i_, _ov_); _sv_ = mm256_scani_epi32(_ov_,_sv_,cv); _mm256_storeu_si256(_op_++, _sv_); +#define VOZ32(_op_, _i_, _ov_, _sv_) VXZ32(_i_, _ov_); _sv_ = mm256_scani_epi32(_ov_,_sv_,cv); _mm256_storeu_si256(_op_++, _sv_); #include "bitunpack_.h" #define BITUNPACK0(_parm_) mv = _mm256_set1_epi32(0) //_parm_ = _mm_setzero_si128() unsigned char *_bitd1unpack256v32( const unsigned char *__restrict in, unsigned n, unsigned *__restrict out, unsigned start, unsigned b, unsigned *__restrict pex, unsigned char *bb) {