BitMagic-C++
Namespaces | Macros | Functions
bmsse4.h File Reference

Compute functions for SSE4.2 SIMD instruction set (internal) More...

#include <mmintrin.h>
#include <emmintrin.h>
#include <smmintrin.h>
#include <nmmintrin.h>
#include <immintrin.h>
#include "bmdef.h"
#include "bmsse_util.h"
#include "bmutil.h"
Include dependency graph for bmsse4.h:

Go to the source code of this file.

Namespaces

 bm
 

Macros

#define VECT_XOR_ARR_2_MASK(dst, src, src_end, mask)   sse2_xor_arr_2_mask((__m128i*)(dst), (__m128i*)(src), (__m128i*)(src_end), (bm::word_t)mask)
 
#define VECT_ANDNOT_ARR_2_MASK(dst, src, src_end, mask)   sse2_andnot_arr_2_mask((__m128i*)(dst), (__m128i*)(src), (__m128i*)(src_end), (bm::word_t)mask)
 
#define VECT_BITCOUNT(first, last)   sse4_bit_count((__m128i*) (first), (__m128i*) (last))
 
#define VECT_BITCOUNT_AND(first, last, mask)   sse4_bit_count_op((__m128i*) (first), (__m128i*) (last), (__m128i*) (mask), sse2_and)
 
#define VECT_BITCOUNT_OR(first, last, mask)   sse4_bit_count_op((__m128i*) (first), (__m128i*) (last), (__m128i*) (mask), sse2_or)
 
#define VECT_BITCOUNT_XOR(first, last, mask)   sse4_bit_count_op((__m128i*) (first), (__m128i*) (last), (__m128i*) (mask), sse2_xor)
 
#define VECT_BITCOUNT_SUB(first, last, mask)   sse4_bit_count_op((__m128i*) (first), (__m128i*) (last), (__m128i*) (mask), sse2_sub)
 
#define VECT_INVERT_BLOCK(first)   sse2_invert_block((__m128i*)first);
 
#define VECT_AND_BLOCK(dst, src)   sse4_and_block((__m128i*) dst, (__m128i*) (src))
 
#define VECT_AND_DIGEST(dst, src)   sse4_and_digest((__m128i*) dst, (const __m128i*) (src))
 
#define VECT_AND_DIGEST_5WAY(dst, src1, src2, src3, src4)   sse4_and_digest_5way((__m128i*) dst, (const __m128i*) (src1), (const __m128i*) (src2), (const __m128i*) (src3), (const __m128i*) (src4))
 
#define VECT_AND_DIGEST_2WAY(dst, src1, src2)   sse4_and_digest_2way((__m128i*) dst, (const __m128i*) (src1), (const __m128i*) (src2))
 
#define VECT_OR_BLOCK(dst, src)   sse2_or_block((__m128i*) dst, (__m128i*) (src))
 
#define VECT_OR_BLOCK_2WAY(dst, src1, src2)   sse2_or_block_2way((__m128i*) (dst), (const __m128i*) (src1), (const __m128i*) (src2))
 
#define VECT_OR_BLOCK_3WAY(dst, src1, src2)   sse2_or_block_3way((__m128i*) (dst), (const __m128i*) (src1), (const __m128i*) (src2))
 
#define VECT_OR_BLOCK_5WAY(dst, src1, src2, src3, src4)   sse2_or_block_5way((__m128i*) (dst), (__m128i*) (src1), (__m128i*) (src2), (__m128i*) (src3), (__m128i*) (src4))
 
#define VECT_SUB_BLOCK(dst, src)   sse2_sub_block((__m128i*) dst, (const __m128i*) (src))
 
#define VECT_SUB_DIGEST(dst, src)   sse4_sub_digest((__m128i*) dst, (const __m128i*) (src))
 
#define VECT_SUB_DIGEST_2WAY(dst, src1, src2)   sse4_sub_digest_2way((__m128i*) dst, (const __m128i*) (src1), (const __m128i*) (src2))
 
#define VECT_XOR_BLOCK(dst, src)   sse2_xor_block((__m128i*) dst, (__m128i*) (src))
 
#define VECT_XOR_BLOCK_2WAY(dst, src1, src2)   sse2_xor_block_2way((__m128i*) (dst), (const __m128i*) (src1), (const __m128i*) (src2))
 
#define VECT_COPY_BLOCK(dst, src)   sse2_copy_block((__m128i*) dst, (__m128i*) (src))
 
#define VECT_STREAM_BLOCK(dst, src)   sse2_stream_block((__m128i*) dst, (__m128i*) (src))
 
#define VECT_SET_BLOCK(dst, value)   sse2_set_block((__m128i*) dst, value)
 
#define VECT_IS_ZERO_BLOCK(dst)   sse4_is_all_zero((__m128i*) dst)
 
#define VECT_IS_ONE_BLOCK(dst)   sse4_is_all_one((__m128i*) dst)
 
#define VECT_IS_DIGEST_ZERO(start)   sse4_is_digest_zero((__m128i*)start)
 
#define VECT_BLOCK_SET_DIGEST(dst, val)   sse4_block_set_digest((__m128i*)dst, val)
 
#define VECT_LOWER_BOUND_SCAN_U32(arr, target, from, to)   sse4_lower_bound_scan_u32(arr, target, from, to)
 
#define VECT_SHIFT_L1(b, acc, co)   sse42_shift_l1((__m128i*)b, acc, co)
 
#define VECT_SHIFT_R1(b, acc, co)   sse42_shift_r1((__m128i*)b, acc, co)
 
#define VECT_SHIFT_R1_AND(b, co, m, digest)   sse42_shift_r1_and((__m128i*)b, co, (__m128i*)m, digest)
 
#define VECT_ARR_BLOCK_LOOKUP(idx, size, nb, start)   sse42_idx_arr_block_lookup(idx, size, nb, start)
 
#define VECT_SET_BLOCK_BITS(block, idx, start, stop)   sse42_set_block_bits(block, idx, start, stop)
 
#define VECT_BLOCK_CHANGE(block, size)   sse42_bit_block_calc_change((__m128i*)block, size)
 
#define VECT_BLOCK_XOR_CHANGE(block, xor_block, size, gc, bc)   sse42_bit_block_calc_xor_change((__m128i*)block, (__m128i*)xor_block, size, gc, bc)
 
#define VECT_BLOCK_CHANGE_BC(block, gc, bc)   sse42_bit_block_calc_change_bc((__m128i*)block, gc, bc)
 
#define VECT_BIT_FIND_FIRST(src, pos)   sse42_bit_find_first((__m128i*) src, pos)
 
#define VECT_BIT_FIND_DIFF(src1, src2, pos)   sse42_bit_find_first_diff((__m128i*) src1, (__m128i*) (src2), pos)
 
#define VECT_BIT_BLOCK_XOR(t, src, src_xor, d)   sse42_bit_block_xor(t, src, src_xor, d)
 
#define VECT_GAP_BFIND(buf, pos, is_set)   sse42_gap_bfind(buf, pos, is_set)
 

Functions

bm::id_t bm::sse4_bit_count (const __m128i *block, const __m128i *block_end)
 
BMFORCEINLINE unsigned bm::op_xor (unsigned a, unsigned b)
 
BMFORCEINLINE unsigned bm::op_or (unsigned a, unsigned b)
 
BMFORCEINLINE unsigned bm::op_and (unsigned a, unsigned b)
 
template<class Func >
bm::id_t bm::sse4_bit_count_op (const __m128i *BMRESTRICT block, const __m128i *BMRESTRICT block_end, const __m128i *BMRESTRICT mask_block, Func sse2_func)
 
bool bm::sse4_is_all_zero (const __m128i *BMRESTRICT block)
 check if block is all zero bits More...
 
bool bm::sse4_is_digest_zero (const __m128i *BMRESTRICT block)
 check if digest stride is all zero bits More...
 
void bm::sse4_block_set_digest (__m128i *dst, unsigned value)
 set digest stride to 0xFF.. or 0x0 value More...
 
unsigned bm::sse4_and_block (__m128i *BMRESTRICT dst, const __m128i *BMRESTRICT src)
 AND blocks2 dst &= *src. More...
 
bool bm::sse4_and_digest (__m128i *BMRESTRICT dst, const __m128i *BMRESTRICT src)
 AND block digest stride dst &= *src. More...
 
bool bm::sse4_and_digest_2way (__m128i *BMRESTRICT dst, const __m128i *BMRESTRICT src1, const __m128i *BMRESTRICT src2)
 AND block digest stride dst = *src1 & src2. More...
 
bool bm::sse4_and_digest_5way (__m128i *BMRESTRICT dst, const __m128i *BMRESTRICT src1, const __m128i *BMRESTRICT src2, const __m128i *BMRESTRICT src3, const __m128i *BMRESTRICT src4)
 AND block digest stride. More...
 
bool bm::sse4_sub_digest (__m128i *BMRESTRICT dst, const __m128i *BMRESTRICT src)
 SUB (AND NOT) block digest stride dst &= ~*src. More...
 
bool bm::sse4_sub_digest_2way (__m128i *BMRESTRICT dst, const __m128i *BMRESTRICT src1, const __m128i *BMRESTRICT src2)
 2-operand SUB (AND NOT) block digest stride dst = src1 & ~*src2 More...
 
bool bm::sse4_is_all_one (const __m128i *BMRESTRICT block)
 check if block is all zero bits More...
 
BMFORCEINLINE bool bm::sse42_test_all_one_wave (const void *ptr)
 check if SSE wave is all oxFFFF...FFF More...
 
BMFORCEINLINE bool bm::sse42_test_all_zero_wave (const void *ptr)
 check if wave of pointers is all NULL More...
 
BMFORCEINLINE bool bm::sse42_test_all_zero_wave2 (const void *ptr0, const void *ptr1)
 check if 2 waves of pointers are all NULL More...
 
BMFORCEINLINE bool bm::sse42_test_all_eq_wave2 (const void *ptr0, const void *ptr1)
 check if wave of 2 pointers are the same (null or FULL) More...
 
unsigned bm::sse42_bit_block_calc_change (const __m128i *BMRESTRICT block, unsigned size)
 
void bm::sse42_bit_block_calc_xor_change (const __m128i *BMRESTRICT block, const __m128i *BMRESTRICT xor_block, unsigned size, unsigned *BMRESTRICT gc, unsigned *BMRESTRICT bc)
 
void bm::sse42_bit_block_calc_change_bc (const __m128i *BMRESTRICT block, unsigned *gc, unsigned *bc)
 
bool bm::sse42_bit_find_first_diff (const __m128i *BMRESTRICT block1, const __m128i *BMRESTRICT block2, unsigned *pos)
 Find first bit which is different between two bit-blocks. More...
 
bool bm::sse42_bit_find_first (const __m128i *BMRESTRICT block, unsigned *pos)
 Find first non-zero bit. More...
 
unsigned bm::sse4_gap_find (const bm::gap_word_t *BMRESTRICT pbuf, const bm::gap_word_t pos, const unsigned size)
 
unsigned bm::sse42_gap_bfind (const unsigned short *BMRESTRICT buf, unsigned pos, unsigned *BMRESTRICT is_set)
 Hybrid binary search, starts as binary, then switches to linear scan. More...
 
unsigned bm::sse42_gap_test (const unsigned short *BMRESTRICT buf, unsigned pos)
 Hybrid binary search, starts as binary, then switches to scan. More...
 
int bm::sse42_cmpge_u32 (__m128i vect4, unsigned value)
 Experimental (test) function to do SIMD vector search (lower bound) in sorted, growing array. More...
 
unsigned bm::sse4_lower_bound_scan_u32 (const unsigned *BMRESTRICT arr, unsigned target, unsigned from, unsigned to)
 lower bound (great or equal) linear scan in ascending order sorted array More...
 
unsigned bm::sse42_idx_arr_block_lookup (const unsigned *idx, unsigned size, unsigned nb, unsigned start)
 
void bm::sse42_set_block_bits (bm::word_t *BMRESTRICT block, const unsigned *BMRESTRICT idx, unsigned start, unsigned stop)
 
void bm::sse4_bit_block_gather_scatter (unsigned *BMRESTRICT arr, const unsigned *BMRESTRICT blk, const unsigned *BMRESTRICT idx, unsigned size, unsigned start, unsigned bit_idx)
 
bool bm::sse42_shift_l1 (__m128i *block, unsigned *empty_acc, unsigned co1)
 block shift left by 1 More...
 
bool bm::sse42_shift_r1 (__m128i *block, unsigned *empty_acc, unsigned co1)
 block shift right by 1 More...
 
bool bm::sse42_shift_r1_and (__m128i *block, bm::word_t co1, const __m128i *BMRESTRICT mask_block, bm::id64_t *digest)
 block shift right by 1 plus AND More...
 
void bm::sse42_bit_block_xor (bm::word_t *target_block, const bm::word_t *block, const bm::word_t *xor_block, bm::id64_t digest)
 Build partial XOR product of 2 bit-blocks using digest mask. More...
 

Detailed Description

Compute functions for SSE4.2 SIMD instruction set (internal)

Definition in file bmsse4.h.

Macro Definition Documentation

◆ VECT_AND_BLOCK

#define VECT_AND_BLOCK (   dst,
  src 
)    sse4_and_block((__m128i*) dst, (__m128i*) (src))

Definition at line 1769 of file bmsse4.h.

◆ VECT_AND_DIGEST

#define VECT_AND_DIGEST (   dst,
  src 
)    sse4_and_digest((__m128i*) dst, (const __m128i*) (src))

Definition at line 1772 of file bmsse4.h.

◆ VECT_AND_DIGEST_2WAY

#define VECT_AND_DIGEST_2WAY (   dst,
  src1,
  src2 
)    sse4_and_digest_2way((__m128i*) dst, (const __m128i*) (src1), (const __m128i*) (src2))

Definition at line 1778 of file bmsse4.h.

◆ VECT_AND_DIGEST_5WAY

#define VECT_AND_DIGEST_5WAY (   dst,
  src1,
  src2,
  src3,
  src4 
)    sse4_and_digest_5way((__m128i*) dst, (const __m128i*) (src1), (const __m128i*) (src2), (const __m128i*) (src3), (const __m128i*) (src4))

Definition at line 1775 of file bmsse4.h.

◆ VECT_ANDNOT_ARR_2_MASK

#define VECT_ANDNOT_ARR_2_MASK (   dst,
  src,
  src_end,
  mask 
)    sse2_andnot_arr_2_mask((__m128i*)(dst), (__m128i*)(src), (__m128i*)(src_end), (bm::word_t)mask)

Definition at line 1748 of file bmsse4.h.

◆ VECT_ARR_BLOCK_LOOKUP

#define VECT_ARR_BLOCK_LOOKUP (   idx,
  size,
  nb,
  start 
)    sse42_idx_arr_block_lookup(idx, size, nb, start)

Definition at line 1841 of file bmsse4.h.

◆ VECT_BIT_BLOCK_XOR

#define VECT_BIT_BLOCK_XOR (   t,
  src,
  src_xor,
 
)    sse42_bit_block_xor(t, src, src_xor, d)

Definition at line 1864 of file bmsse4.h.

◆ VECT_BIT_FIND_DIFF

#define VECT_BIT_FIND_DIFF (   src1,
  src2,
  pos 
)    sse42_bit_find_first_diff((__m128i*) src1, (__m128i*) (src2), pos)

Definition at line 1861 of file bmsse4.h.

◆ VECT_BIT_FIND_FIRST

#define VECT_BIT_FIND_FIRST (   src,
  pos 
)    sse42_bit_find_first((__m128i*) src, pos)

Definition at line 1858 of file bmsse4.h.

◆ VECT_BITCOUNT

#define VECT_BITCOUNT (   first,
  last 
)    sse4_bit_count((__m128i*) (first), (__m128i*) (last))

Definition at line 1751 of file bmsse4.h.

◆ VECT_BITCOUNT_AND

#define VECT_BITCOUNT_AND (   first,
  last,
  mask 
)    sse4_bit_count_op((__m128i*) (first), (__m128i*) (last), (__m128i*) (mask), sse2_and)

Definition at line 1754 of file bmsse4.h.

◆ VECT_BITCOUNT_OR

#define VECT_BITCOUNT_OR (   first,
  last,
  mask 
)    sse4_bit_count_op((__m128i*) (first), (__m128i*) (last), (__m128i*) (mask), sse2_or)

Definition at line 1757 of file bmsse4.h.

◆ VECT_BITCOUNT_SUB

#define VECT_BITCOUNT_SUB (   first,
  last,
  mask 
)    sse4_bit_count_op((__m128i*) (first), (__m128i*) (last), (__m128i*) (mask), sse2_sub)

Definition at line 1763 of file bmsse4.h.

◆ VECT_BITCOUNT_XOR

#define VECT_BITCOUNT_XOR (   first,
  last,
  mask 
)    sse4_bit_count_op((__m128i*) (first), (__m128i*) (last), (__m128i*) (mask), sse2_xor)

Definition at line 1760 of file bmsse4.h.

◆ VECT_BLOCK_CHANGE

#define VECT_BLOCK_CHANGE (   block,
  size 
)    sse42_bit_block_calc_change((__m128i*)block, size)

Definition at line 1847 of file bmsse4.h.

◆ VECT_BLOCK_CHANGE_BC

#define VECT_BLOCK_CHANGE_BC (   block,
  gc,
  bc 
)    sse42_bit_block_calc_change_bc((__m128i*)block, gc, bc)

Definition at line 1854 of file bmsse4.h.

◆ VECT_BLOCK_SET_DIGEST

#define VECT_BLOCK_SET_DIGEST (   dst,
  val 
)    sse4_block_set_digest((__m128i*)dst, val)

Definition at line 1826 of file bmsse4.h.

◆ VECT_BLOCK_XOR_CHANGE

#define VECT_BLOCK_XOR_CHANGE (   block,
  xor_block,
  size,
  gc,
  bc 
)    sse42_bit_block_calc_xor_change((__m128i*)block, (__m128i*)xor_block, size, gc, bc)

Definition at line 1850 of file bmsse4.h.

◆ VECT_COPY_BLOCK

#define VECT_COPY_BLOCK (   dst,
  src 
)    sse2_copy_block((__m128i*) dst, (__m128i*) (src))

Definition at line 1808 of file bmsse4.h.

◆ VECT_GAP_BFIND

#define VECT_GAP_BFIND (   buf,
  pos,
  is_set 
)    sse42_gap_bfind(buf, pos, is_set)

Definition at line 1867 of file bmsse4.h.

◆ VECT_INVERT_BLOCK

#define VECT_INVERT_BLOCK (   first)    sse2_invert_block((__m128i*)first);

Definition at line 1766 of file bmsse4.h.

◆ VECT_IS_DIGEST_ZERO

#define VECT_IS_DIGEST_ZERO (   start)    sse4_is_digest_zero((__m128i*)start)

Definition at line 1823 of file bmsse4.h.

◆ VECT_IS_ONE_BLOCK

#define VECT_IS_ONE_BLOCK (   dst)    sse4_is_all_one((__m128i*) dst)

Definition at line 1820 of file bmsse4.h.

◆ VECT_IS_ZERO_BLOCK

#define VECT_IS_ZERO_BLOCK (   dst)    sse4_is_all_zero((__m128i*) dst)

Definition at line 1817 of file bmsse4.h.

◆ VECT_LOWER_BOUND_SCAN_U32

#define VECT_LOWER_BOUND_SCAN_U32 (   arr,
  target,
  from,
  to 
)    sse4_lower_bound_scan_u32(arr, target, from, to)

Definition at line 1829 of file bmsse4.h.

◆ VECT_OR_BLOCK

#define VECT_OR_BLOCK (   dst,
  src 
)    sse2_or_block((__m128i*) dst, (__m128i*) (src))

Definition at line 1781 of file bmsse4.h.

◆ VECT_OR_BLOCK_2WAY

#define VECT_OR_BLOCK_2WAY (   dst,
  src1,
  src2 
)    sse2_or_block_2way((__m128i*) (dst), (const __m128i*) (src1), (const __m128i*) (src2))

Definition at line 1784 of file bmsse4.h.

◆ VECT_OR_BLOCK_3WAY

#define VECT_OR_BLOCK_3WAY (   dst,
  src1,
  src2 
)    sse2_or_block_3way((__m128i*) (dst), (const __m128i*) (src1), (const __m128i*) (src2))

Definition at line 1787 of file bmsse4.h.

◆ VECT_OR_BLOCK_5WAY

#define VECT_OR_BLOCK_5WAY (   dst,
  src1,
  src2,
  src3,
  src4 
)    sse2_or_block_5way((__m128i*) (dst), (__m128i*) (src1), (__m128i*) (src2), (__m128i*) (src3), (__m128i*) (src4))

Definition at line 1790 of file bmsse4.h.

◆ VECT_SET_BLOCK

#define VECT_SET_BLOCK (   dst,
  value 
)    sse2_set_block((__m128i*) dst, value)

Definition at line 1814 of file bmsse4.h.

◆ VECT_SET_BLOCK_BITS

#define VECT_SET_BLOCK_BITS (   block,
  idx,
  start,
  stop 
)    sse42_set_block_bits(block, idx, start, stop)

Definition at line 1844 of file bmsse4.h.

◆ VECT_SHIFT_L1

#define VECT_SHIFT_L1 (   b,
  acc,
  co 
)    sse42_shift_l1((__m128i*)b, acc, co)

Definition at line 1832 of file bmsse4.h.

◆ VECT_SHIFT_R1

#define VECT_SHIFT_R1 (   b,
  acc,
  co 
)    sse42_shift_r1((__m128i*)b, acc, co)

Definition at line 1835 of file bmsse4.h.

◆ VECT_SHIFT_R1_AND

#define VECT_SHIFT_R1_AND (   b,
  co,
  m,
  digest 
)    sse42_shift_r1_and((__m128i*)b, co, (__m128i*)m, digest)

Definition at line 1838 of file bmsse4.h.

◆ VECT_STREAM_BLOCK

#define VECT_STREAM_BLOCK (   dst,
  src 
)    sse2_stream_block((__m128i*) dst, (__m128i*) (src))

Definition at line 1811 of file bmsse4.h.

◆ VECT_SUB_BLOCK

#define VECT_SUB_BLOCK (   dst,
  src 
)    sse2_sub_block((__m128i*) dst, (const __m128i*) (src))

Definition at line 1793 of file bmsse4.h.

◆ VECT_SUB_DIGEST

#define VECT_SUB_DIGEST (   dst,
  src 
)    sse4_sub_digest((__m128i*) dst, (const __m128i*) (src))

Definition at line 1796 of file bmsse4.h.

◆ VECT_SUB_DIGEST_2WAY

#define VECT_SUB_DIGEST_2WAY (   dst,
  src1,
  src2 
)    sse4_sub_digest_2way((__m128i*) dst, (const __m128i*) (src1), (const __m128i*) (src2))

Definition at line 1799 of file bmsse4.h.

◆ VECT_XOR_ARR_2_MASK

#define VECT_XOR_ARR_2_MASK (   dst,
  src,
  src_end,
  mask 
)    sse2_xor_arr_2_mask((__m128i*)(dst), (__m128i*)(src), (__m128i*)(src_end), (bm::word_t)mask)

Definition at line 1745 of file bmsse4.h.

◆ VECT_XOR_BLOCK

#define VECT_XOR_BLOCK (   dst,
  src 
)    sse2_xor_block((__m128i*) dst, (__m128i*) (src))

Definition at line 1802 of file bmsse4.h.

◆ VECT_XOR_BLOCK_2WAY

#define VECT_XOR_BLOCK_2WAY (   dst,
  src1,
  src2 
)    sse2_xor_block_2way((__m128i*) (dst), (const __m128i*) (src1), (const __m128i*) (src2))

Definition at line 1805 of file bmsse4.h.