Processor specific optimizations for AVX2 instructions (internals). More...

Collaboration diagram for AVX2 functions:

Functions
bm::id_t	bm::avx2_bit_count (const __m256i BMRESTRICT block, const __m256i BMRESTRICT block_end)
	AVX2 Harley-Seal popcount The algorithm is based on the paper "Faster Population Counts using AVX2 Instructions" by Daniel Lemire, Nathan Kurz and Wojciech Mula (23 Nov 2016).
bm::id_t	bm::avx2_bit_block_count (const bm::word_t *const block, bm::id64_t digest)
	Calculate population count based on digest.
bm::id_t	bm::avx2_bit_count_and (const __m256i BMRESTRICT block, const __m256i BMRESTRICT block_end, const __m256i *BMRESTRICT mask_block)
	AND bit count for two aligned bit-blocks.
bm::id_t	bm::avx2_bit_count_xor (const __m256i BMRESTRICT block, const __m256i BMRESTRICT block_end, const __m256i *BMRESTRICT mask_block)
	XOR bit count for two aligned bit-blocks.
bm::id_t	bm::avx2_bit_count_sub (const __m256i BMRESTRICT block, const __m256i BMRESTRICT block_end, const __m256i *BMRESTRICT mask_block)
	AND NOT bit count for two aligned bit-blocks.
void	bm::avx2_xor_arr_2_mask (__m256i BMRESTRICT dst, const __m256i BMRESTRICT src, const __m256i *BMRESTRICT src_end, bm::word_t mask)
	XOR array elements to specified mask dst = *src ^ mask.
void	bm::avx2_andnot_arr_2_mask (__m256i BMRESTRICT dst, const __m256i BMRESTRICT src, const __m256i *BMRESTRICT src_end, bm::word_t mask)
	Inverts array elements and NOT them to specified mask dst = ~*src & mask.
unsigned	bm::avx2_and_block (__m256i BMRESTRICT dst, const __m256i BMRESTRICT src)
	AND array elements against another array dst &= *src.
bool	bm::avx2_and_digest (__m256i BMRESTRICT dst, const __m256i BMRESTRICT src)
	AND block digest stride dst &= *src.
bool	bm::avx2_and_digest_2way (__m256i BMRESTRICT dst, const __m256i BMRESTRICT src1, const __m256i *BMRESTRICT src2)
	AND block digest stride 2 way dst = src1 & src2.
bool	bm::avx2_and_or_digest_2way (__m256i BMRESTRICT dst, const __m256i BMRESTRICT src1, const __m256i *BMRESTRICT src2)
	AND-OR block digest stride 2 way dst \|= src1 & src2.
bool	bm::avx2_and_digest_5way (__m256i BMRESTRICT dst, const __m256i BMRESTRICT src1, const __m256i BMRESTRICT src2, const __m256i BMRESTRICT src3, const __m256i *BMRESTRICT src4)
	AND block digest stride.
bool	bm::avx2_and_digest_3way (__m256i BMRESTRICT dst, const __m256i BMRESTRICT src1, const __m256i *BMRESTRICT src2)
	AND block digest stride.
unsigned	bm::avx2_and_arr_unal (__m256i BMRESTRICT dst, const __m256i BMRESTRICT src, const __m256i *BMRESTRICT src_end)
	AND array elements against another array (unaligned) dst &= *src.
bool	bm::avx2_or_block (__m256i BMRESTRICT dst, const __m256i BMRESTRICT src)
	OR array elements against another array dst \|= *src.
bool	bm::avx2_or_arr_unal (__m256i BMRESTRICT dst, const __m256i BMRESTRICT src, const __m256i *BMRESTRICT src_end)
	OR array elements against another unaligned array dst \|= *src.
bool	bm::avx2_or_block_2way (__m256i BMRESTRICT dst, const __m256i BMRESTRICT src1, const __m256i *BMRESTRICT src2)
	OR 2 arrays and copy to the destination dst = *src1 \| src2.
bool	bm::avx2_or_block_3way (__m256i BMRESTRICT dst, const __m256i BMRESTRICT src1, const __m256i *BMRESTRICT src2)
	OR array elements against another 2 arrays dst \|= *src1 \| src2.
bool	bm::avx2_or_block_5way (__m256i BMRESTRICT dst, const __m256i BMRESTRICT src1, const __m256i BMRESTRICT src2, const __m256i BMRESTRICT src3, const __m256i *BMRESTRICT src4)
	OR array elements against another 4 arrays dst \|= *src1 \| src2.
unsigned	bm::avx2_xor_block (__m256i BMRESTRICT dst, const __m256i BMRESTRICT src)
	XOR block against another dst ^= *src.
unsigned	bm::avx2_xor_block_2way (__m256i BMRESTRICT dst, const __m256i BMRESTRICT src1, const __m256i *BMRESTRICT src2)
	3 operand XOR dst = *src1 ^ src2
unsigned	bm::avx2_sub_block (__m256i BMRESTRICT dst, const __m256i BMRESTRICT src)
	AND-NOT (SUB) array elements against another array dst &= ~*src.
bool	bm::avx2_sub_digest (__m256i BMRESTRICT dst, const __m256i BMRESTRICT src)
	SUB (AND NOT) block digest stride dst &= ~*src.
bool	bm::avx2_sub_digest_2way (__m256i BMRESTRICT dst, const __m256i BMRESTRICT src1, const __m256i *BMRESTRICT src2)
	2-operand SUB (AND NOT) block digest stride dst = src1 & ~src2
bool	bm::avx2_sub_digest_5way (__m256i BMRESTRICT dst, const __m256i BMRESTRICT src1, const __m256i BMRESTRICT src2, const __m256i BMRESTRICT src3, const __m256i *BMRESTRICT src4)
	SUB block digest stride.
bool	bm::avx2_sub_digest_3way (__m256i BMRESTRICT dst, const __m256i BMRESTRICT src1, const __m256i *BMRESTRICT src2)
	SUB block digest stride.
BMFORCEINLINE void	bm::avx2_set_block (__m256i *BMRESTRICT dst, bm::word_t value)
	AVX2 block memset dst = value.
void	bm::avx2_copy_block (__m256i BMRESTRICT dst, const __m256i BMRESTRICT src)
	AVX2 block copy dst = *src.
void	bm::avx2_copy_block_unalign (__m256i BMRESTRICT dst, const __m256i BMRESTRICT src)
	AVX2 block copy (unaligned SRC) dst = *src.
void	bm::avx2_stream_block (__m256i BMRESTRICT dst, const __m256i BMRESTRICT src)
	AVX2 block copy dst = *src.
void	bm::avx2_stream_block_unalign (__m256i BMRESTRICT dst, const __m256i BMRESTRICT src)
	AVX2 block copy (unaligned SRC) dst = *src.
void	bm::avx2_invert_block (__m256i *BMRESTRICT dst)
	Invert bit-block dst = ~dst or dst ^= dst.
bool	bm::avx2_is_all_zero (const __m256i *BMRESTRICT block)
	check if block is all zero bits
bool	bm::avx2_is_digest_zero (const __m256i *BMRESTRICT block)
	check if digest stride is all zero bits
void	bm::avx2_block_set_digest (__m256i *dst, unsigned value)
	set digest stride to 0xFF.. or 0x0 value
bool	bm::avx2_is_all_one (const __m256i *BMRESTRICT block)
	check if block is all one bits
BMFORCEINLINE bool	bm::avx2_test_all_one_wave (const void *ptr)
	check if wave of pointers is all 0xFFF
BMFORCEINLINE bool	bm::avx2_test_all_zero_wave (const void *ptr)
	check if wave of pointers is all NULL
BMFORCEINLINE bool	bm::avx2_test_all_zero_wave2 (const void ptr0, const void ptr1)
	check if 2 wave of pointers are all NULL
BMFORCEINLINE bool	bm::avx2_test_all_eq_wave2 (const void ptr0, const void ptr1)
	check if 2 wave of pointers are all the same (NULL or FULL)
bool	bm::avx2_shift_l1 (__m256i block, bm::word_t empty_acc, unsigned co1)
	block shift left by 1
bool	bm::avx2_shift_r1 (__m256i block, bm::word_t empty_acc, unsigned co1)
	block shift right by 1
bool	bm::avx2_shift_r1_and (__m256i BMRESTRICT block, bm::word_t co1, const __m256i BMRESTRICT mask_block, bm::id64_t *BMRESTRICT digest)
	fused block shift right by 1 plus AND
unsigned	bm::avx2_bit_block_calc_change (const __m256i *BMRESTRICT block, unsigned size)
void	bm::avx2_bit_block_calc_xor_change (const __m256i BMRESTRICT block, const __m256i BMRESTRICT xor_block, unsigned size, unsigned BMRESTRICT gcount, unsigned BMRESTRICT bcount)
void	bm::avx2_bit_block_calc_change_bc (const __m256i BMRESTRICT block, unsigned gcount, unsigned *bcount)
bool	bm::avx2_bit_find_first_diff (const __m256i BMRESTRICT block1, const __m256i BMRESTRICT block2, unsigned *pos)
	Find first bit which is different between two bit-blocks.
bool	bm::avx2_bit_find_first (const __m256i BMRESTRICT block, unsigned off, unsigned pos)
	Find first bit set.
int	bm::avx2_cmpge_u32 (__m256i vect8, unsigned value)
	Experimental (test) function to do SIMD vector search (lower bound) in sorted, growing array.
int	bm::avx2_cmpge_u16 (__m256i vect16, unsigned short value)
	Experimental (test) function to do SIMD vector search in sorted, growing array.
template<bool RET_TEST = false>
unsigned	bm::avx2_gap_bfind (const unsigned short BMRESTRICT buf, unsigned pos, unsigned BMRESTRICT is_set)
	Hybrid binary search, starts as binary, then switches to scan.
unsigned	bm::avx2_gap_test (const unsigned short *BMRESTRICT buf, unsigned pos)
	Hybrid binary search, starts as binary, then switches to scan.
unsigned	bm::avx2_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
unsigned	bm::avx2_bit_to_gap (gap_word_t BMRESTRICT dest, const unsigned BMRESTRICT block, unsigned dest_len)
	Convert bit block to GAP block.
void	bm::avx2_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.
void	bm::avx2_bit_block_xor_2way (bm::word_t target_block, const bm::word_t xor_block, bm::id64_t digest) BMNOEXCEPT
	Build partial XOR product of 2 bit-blocks using digest mask.

Detailed Description

Processor specific optimizations for AVX2 instructions (internals).

Function Documentation

◆ avx2_and_arr_unal()

unsigned bm::avx2_and_arr_unal	(	__m256i *BMRESTRICT	dst,
		const __m256i *BMRESTRICT	src,
		const __m256i *BMRESTRICT	src_end )

inline

AND array elements against another array (unaligned) dst &= *src.

Returns: 0 if destination does not have any bits

Definition at line 777 of file bmavx2.h.

References BMRESTRICT.

Referenced by bm::decoder::get_32_AND().

◆ avx2_and_block()

unsigned bm::avx2_and_block	(	__m256i *BMRESTRICT	dst,
		const __m256i *BMRESTRICT	src )

inline

AND array elements against another array dst &= *src.

Returns: 0 if destination does not have any bits

Definition at line 496 of file bmavx2.h.

References BMRESTRICT, and set_block_size.

◆ avx2_and_digest()

bool bm::avx2_and_digest	(	__m256i *BMRESTRICT	dst,
		const __m256i *BMRESTRICT	src )

inline

AND block digest stride dst &= *src.

Returns: true if stide is all zero

Definition at line 543 of file bmavx2.h.

References BMRESTRICT.

◆ avx2_and_digest_2way()

bool bm::avx2_and_digest_2way	(	__m256i *BMRESTRICT	dst,
		const __m256i *BMRESTRICT	src1,
		const __m256i *BMRESTRICT	src2 )

inline

AND block digest stride 2 way dst = *src1 & *src2.

Returns: true if stide is all zero

Definition at line 573 of file bmavx2.h.

References BMRESTRICT.

◆ avx2_and_digest_3way()

bool bm::avx2_and_digest_3way	(	__m256i *BMRESTRICT	dst,
		const __m256i *BMRESTRICT	src1,
		const __m256i *BMRESTRICT	src2 )

inline

AND block digest stride.

Definition at line 727 of file bmavx2.h.

References BMRESTRICT.

◆ avx2_and_digest_5way()

bool bm::avx2_and_digest_5way	(	__m256i *BMRESTRICT	dst,
		const __m256i *BMRESTRICT	src1,
		const __m256i *BMRESTRICT	src2,
		const __m256i *BMRESTRICT	src3,
		const __m256i *BMRESTRICT	src4 )

inline

AND block digest stride.

Definition at line 659 of file bmavx2.h.

References BMRESTRICT.

◆ avx2_and_or_digest_2way()

bool bm::avx2_and_or_digest_2way	(	__m256i *BMRESTRICT	dst,
		const __m256i *BMRESTRICT	src1,
		const __m256i *BMRESTRICT	src2 )

inline

AND-OR block digest stride 2 way dst |= *src1 & *src2.

Returns: true if stide is all zero

Definition at line 604 of file bmavx2.h.

References BMRESTRICT.

◆ avx2_andnot_arr_2_mask()

void bm::avx2_andnot_arr_2_mask	(	__m256i *BMRESTRICT	dst,
		const __m256i *BMRESTRICT	src,
		const __m256i *BMRESTRICT	src_end,
		bm::word_t	mask )

inline

Inverts array elements and NOT them to specified mask dst = ~*src & mask.

Definition at line 472 of file bmavx2.h.

References BMRESTRICT.

◆ avx2_bit_block_calc_change()

unsigned bm::avx2_bit_block_calc_change	(	const __m256i *BMRESTRICT	block,
		unsigned	size )

inline

AVX2 calculate number of bit changes from 0 to 1

Definition at line 2083 of file bmavx2.h.

References BM_ALIGN32, BM_ALIGN32ATTR, BM_AVX2_BIT_COUNT, BM_AVX2_POPCNT_PROLOG, and BMRESTRICT.

◆ avx2_bit_block_calc_change_bc()

void bm::avx2_bit_block_calc_change_bc	(	const __m256i *BMRESTRICT	block,
		unsigned *	gcount,
		unsigned *	bcount )

inline

AVX2 calculate number of bit changes from 0 to 1 and bitcount

Definition at line 2251 of file bmavx2.h.

References BM_ALIGN32, BM_ALIGN32ATTR, BM_AVX2_BIT_COUNT, BM_AVX2_POPCNT_PROLOG, BMRESTRICT, and set_block_size.

◆ avx2_bit_block_calc_xor_change()

void bm::avx2_bit_block_calc_xor_change	(	const __m256i *BMRESTRICT	block,
		const __m256i *BMRESTRICT	xor_block,
		unsigned	size,
		unsigned *BMRESTRICT	gcount,
		unsigned *BMRESTRICT	bcount )

inline

AVX2 calculate number of bit changes from 0 to 1 from a XOR product

Definition at line 2154 of file bmavx2.h.

References BM_ALIGN32, BM_ALIGN32ATTR, BM_AVX2_BIT_COUNT, BM_AVX2_POPCNT_PROLOG, and BMRESTRICT.

◆ avx2_bit_block_count()

bm::id_t bm::avx2_bit_block_count	(	const bm::word_t *const	block,
		bm::id64_t	digest )

inline

Calculate population count based on digest.

Returns: popcnt

Definition at line 232 of file bmavx2.h.

References BM_AVX2_BIT_COUNT, BM_AVX2_POPCNT_PROLOG, bmi_blsi_u64(), bmi_bslr_u64(), BMRESTRICT, and set_block_digest_wave_size.

◆ avx2_bit_block_xor()

void bm::avx2_bit_block_xor	(	bm::word_t *	target_block,
		const bm::word_t *	block,
		const bm::word_t *	xor_block,
		bm::id64_t	digest )

inline

Build partial XOR product of 2 bit-blocks using digest mask.

Parameters

target_block	- target := block ^ xor_block
block	- arg1
xor_block	- arg2
digest	- mask for each block wave to XOR (1) or just copy (0)

Definition at line 3341 of file bmavx2.h.

References block_waves, and set_block_digest_wave_size.

◆ avx2_bit_block_xor_2way()

void bm::avx2_bit_block_xor_2way	(	bm::word_t *	target_block,
		const bm::word_t *	xor_block,
		bm::id64_t	digest )

inline

Build partial XOR product of 2 bit-blocks using digest mask.

Parameters

target_block	- target ^= xor_block
xor_block	- arg1
digest	- mask for each block wave to XOR (1)

Definition at line 3392 of file bmavx2.h.

References bmi_blsi_u64(), bmi_bslr_u64(), BMNOEXCEPT, and set_block_digest_wave_size.

◆ avx2_bit_count()

bm::id_t bm::avx2_bit_count	(	const __m256i *BMRESTRICT	block,
		const __m256i *BMRESTRICT	block_end )

inline

AVX2 Harley-Seal popcount The algorithm is based on the paper "Faster Population Counts using AVX2 Instructions" by Daniel Lemire, Nathan Kurz and Wojciech Mula (23 Nov 2016).

See also: https://arxiv.org/abs/1611.07612

Definition at line 156 of file bmavx2.h.

References BM_AVX2_BIT_COUNT, BM_AVX2_POPCNT_PROLOG, BM_CSA256, and BMRESTRICT.

◆ avx2_bit_count_and()

bm::id_t bm::avx2_bit_count_and	(	const __m256i *BMRESTRICT	block,
		const __m256i *BMRESTRICT	block_end,
		const __m256i *BMRESTRICT	mask_block )

inline

AND bit count for two aligned bit-blocks.

Definition at line 290 of file bmavx2.h.

References BM_AVX2_BIT_COUNT, BM_AVX2_POPCNT_PROLOG, and BMRESTRICT.

◆ avx2_bit_count_sub()

bm::id_t bm::avx2_bit_count_sub	(	const __m256i *BMRESTRICT	block,
		const __m256i *BMRESTRICT	block_end,
		const __m256i *BMRESTRICT	mask_block )

inline

AND NOT bit count for two aligned bit-blocks.

Definition at line 413 of file bmavx2.h.

References BM_AVX2_BIT_COUNT, BM_AVX2_POPCNT_PROLOG, and BMRESTRICT.

◆ avx2_bit_count_xor()

bm::id_t bm::avx2_bit_count_xor	(	const __m256i *BMRESTRICT	block,
		const __m256i *BMRESTRICT	block_end,
		const __m256i *BMRESTRICT	mask_block )

inline

XOR bit count for two aligned bit-blocks.

Definition at line 368 of file bmavx2.h.

References BM_AVX2_BIT_COUNT, BM_AVX2_POPCNT_PROLOG, and BMRESTRICT.

◆ avx2_bit_find_first()

bool bm::avx2_bit_find_first	(	const __m256i *BMRESTRICT	block,
		unsigned	off,
		unsigned *	pos )

inline

Find first bit set.

Definition at line 2394 of file bmavx2.h.

References BM_ALIGN32, BM_ALIGN32ATTR, BM_ASSERT, BMRESTRICT, and set_block_size.

◆ avx2_bit_find_first_diff()

bool bm::avx2_bit_find_first_diff	(	const __m256i *BMRESTRICT	block1,
		const __m256i *BMRESTRICT	block2,
		unsigned *	pos )

inline

Find first bit which is different between two bit-blocks.

Definition at line 2336 of file bmavx2.h.

References BM_ALIGN32, BM_ALIGN32ATTR, BM_ASSERT, BMRESTRICT, and set_block_size.

◆ avx2_bit_to_gap()

unsigned bm::avx2_bit_to_gap	(	gap_word_t *BMRESTRICT	dest,
		const unsigned *BMRESTRICT	block,
		unsigned	dest_len )

inline

Convert bit block to GAP block.

Definition at line 3227 of file bmavx2.h.

References BM_ASSERT, BMRESTRICT, and set_block_size.

◆ avx2_block_set_digest()

void bm::avx2_block_set_digest	(	__m256i *	dst,
		unsigned	value )

inline

set digest stride to 0xFF.. or 0x0 value

Definition at line 1752 of file bmavx2.h.

◆ avx2_cmpge_u16()

int bm::avx2_cmpge_u16	(	__m256i	vect16,
		unsigned short	value )

inline

Experimental (test) function to do SIMD vector search in sorted, growing array.

Definition at line 2907 of file bmavx2.h.

◆ avx2_cmpge_u32()

int bm::avx2_cmpge_u32	(	__m256i	vect8,
		unsigned	value )

inline

Experimental (test) function to do SIMD vector search (lower bound) in sorted, growing array.

Definition at line 2875 of file bmavx2.h.

◆ avx2_copy_block()

void bm::avx2_copy_block	(	__m256i *BMRESTRICT	dst,
		const __m256i *BMRESTRICT	src )

inline

AVX2 block copy dst = *src.

Definition at line 1503 of file bmavx2.h.

References BMRESTRICT, and set_block_size.

◆ avx2_copy_block_unalign()

void bm::avx2_copy_block_unalign	(	__m256i *BMRESTRICT	dst,
		const __m256i *BMRESTRICT	src )

inline

AVX2 block copy (unaligned SRC) dst = *src.

Definition at line 1545 of file bmavx2.h.

References BMRESTRICT, and set_block_size.

◆ avx2_gap_bfind()

template<bool RET_TEST = false>

unsigned bm::avx2_gap_bfind	(	const unsigned short *BMRESTRICT	buf,
		unsigned	pos,
		unsigned *BMRESTRICT	is_set )

Hybrid binary search, starts as binary, then switches to scan.

NOTE: AVX code uses _mm256_subs_epu16 - saturated substraction which gives 0 if A-B=0 if A < B (not negative a value).

Parameters

buf	- GAP buffer pointer.
pos	- index of the element.
is_set	- output. GAP value (0 or 1).

Returns: GAP index OR bit-test

Definition at line 2939 of file bmavx2.h.

References BM_ASSERT, BMRESTRICT, and gap_max_bits.

Referenced by avx2_gap_test().

◆ avx2_gap_test()

unsigned bm::avx2_gap_test	(	const unsigned short *BMRESTRICT	buf,
		unsigned	pos )

inline

Hybrid binary search, starts as binary, then switches to scan.

Definition at line 3057 of file bmavx2.h.

References avx2_gap_bfind(), and BMRESTRICT.

◆ avx2_invert_block()

void bm::avx2_invert_block ( __m256i *BMRESTRICT dst )

inline

Invert bit-block dst = ~*dst or dst ^= *dst.

Definition at line 1677 of file bmavx2.h.

References BMRESTRICT, and set_block_size.

◆ avx2_is_all_one()

bool bm::avx2_is_all_one ( const __m256i *BMRESTRICT block )

inline

check if block is all one bits

Returns: true if all bits are 1

Definition at line 1767 of file bmavx2.h.

References BMRESTRICT, and set_block_size.

◆ avx2_is_all_zero()

bool bm::avx2_is_all_zero ( const __m256i *BMRESTRICT block )

inline

check if block is all zero bits

Definition at line 1708 of file bmavx2.h.

References BMRESTRICT, and set_block_size.

◆ avx2_is_digest_zero()

bool bm::avx2_is_digest_zero ( const __m256i *BMRESTRICT block )

inline

check if digest stride is all zero bits

Definition at line 1738 of file bmavx2.h.

References BMRESTRICT.

◆ avx2_lower_bound_scan_u32()

unsigned bm::avx2_lower_bound_scan_u32	(	const unsigned *BMRESTRICT	arr,
		unsigned	target,
		unsigned	from,
		unsigned	to )

inline

lower bound (great or equal) linear scan in ascending order sorted array

Definition at line 3068 of file bmavx2.h.

References BMRESTRICT.

◆ avx2_or_arr_unal()

bool bm::avx2_or_arr_unal	(	__m256i *BMRESTRICT	dst,
		const __m256i *BMRESTRICT	src,
		const __m256i *BMRESTRICT	src_end )

inline

OR array elements against another unaligned array dst |= *src.

Returns: true if all bits are 1

Definition at line 888 of file bmavx2.h.

References BMRESTRICT.

Referenced by bm::decoder::get_32_OR().

◆ avx2_or_block()

bool bm::avx2_or_block	(	__m256i *BMRESTRICT	dst,
		const __m256i *BMRESTRICT	src )

inline

OR array elements against another array dst |= *src.

Returns: true if all bits are 1

Definition at line 835 of file bmavx2.h.

References BMRESTRICT, and set_block_size.

◆ avx2_or_block_2way()

bool bm::avx2_or_block_2way	(	__m256i *BMRESTRICT	dst,
		const __m256i *BMRESTRICT	src1,
		const __m256i *BMRESTRICT	src2 )

inline

OR 2 arrays and copy to the destination dst = *src1 | src2.

Returns: true if all bits are 1

Definition at line 941 of file bmavx2.h.

References BMRESTRICT, and set_block_size.

◆ avx2_or_block_3way()

bool bm::avx2_or_block_3way	(	__m256i *BMRESTRICT	dst,
		const __m256i *BMRESTRICT	src1,
		const __m256i *BMRESTRICT	src2 )

inline

OR array elements against another 2 arrays dst |= *src1 | src2.

Returns: true if all bits are 1

Definition at line 987 of file bmavx2.h.

References BMRESTRICT, and set_block_size.

◆ avx2_or_block_5way()

bool bm::avx2_or_block_5way	(	__m256i *BMRESTRICT	dst,
		const __m256i *BMRESTRICT	src1,
		const __m256i *BMRESTRICT	src2,
		const __m256i *BMRESTRICT	src3,
		const __m256i *BMRESTRICT	src4 )

inline

OR array elements against another 4 arrays dst |= *src1 | src2.

Returns: true if all bits are 1

Definition at line 1039 of file bmavx2.h.

References BMRESTRICT, and set_block_size.

◆ avx2_set_block()

BMFORCEINLINE void bm::avx2_set_block	(	__m256i *BMRESTRICT	dst,
		bm::word_t	value )

AVX2 block memset dst = value.

Definition at line 1477 of file bmavx2.h.

References BMRESTRICT, and set_block_size.

◆ avx2_shift_l1()

bool bm::avx2_shift_l1	(	__m256i *	block,
		bm::word_t *	empty_acc,
		unsigned	co1 )

inline

block shift left by 1

Definition at line 1842 of file bmavx2.h.

References set_block_size.

◆ avx2_shift_r1()

bool bm::avx2_shift_r1	(	__m256i *	block,
		bm::word_t *	empty_acc,
		unsigned	co1 )

inline

block shift right by 1

Definition at line 1903 of file bmavx2.h.

References set_block_size.

◆ avx2_shift_r1_and()

bool bm::avx2_shift_r1_and	(	__m256i *BMRESTRICT	block,
		bm::word_t	co1,
		const __m256i *BMRESTRICT	mask_block,
		bm::id64_t *BMRESTRICT	digest )

inline

fused block shift right by 1 plus AND

Definition at line 1959 of file bmavx2.h.

References BM_ASSERT, BMRESTRICT, and set_block_digest_wave_size.

◆ avx2_stream_block()

void bm::avx2_stream_block	(	__m256i *BMRESTRICT	dst,
		const __m256i *BMRESTRICT	src )

inline

AVX2 block copy dst = *src.

Definition at line 1589 of file bmavx2.h.

References BMRESTRICT, and set_block_size.

◆ avx2_stream_block_unalign()

void bm::avx2_stream_block_unalign	(	__m256i *BMRESTRICT	dst,
		const __m256i *BMRESTRICT	src )

inline

AVX2 block copy (unaligned SRC) dst = *src.

Definition at line 1631 of file bmavx2.h.

References BMRESTRICT, and set_block_size.

◆ avx2_sub_block()

unsigned bm::avx2_sub_block	(	__m256i *BMRESTRICT	dst,
		const __m256i *BMRESTRICT	src )

inline

AND-NOT (SUB) array elements against another array dst &= ~*src.

Returns: 0 if destination does not have any bits

Definition at line 1204 of file bmavx2.h.

References BMRESTRICT, and set_block_size.

◆ avx2_sub_digest()

bool bm::avx2_sub_digest	(	__m256i *BMRESTRICT	dst,
		const __m256i *BMRESTRICT	src )

inline

SUB (AND NOT) block digest stride dst &= ~*src.

Returns: true if stide is all zero

Definition at line 1250 of file bmavx2.h.

References BMRESTRICT.

◆ avx2_sub_digest_2way()

bool bm::avx2_sub_digest_2way	(	__m256i *BMRESTRICT	dst,
		const __m256i *BMRESTRICT	src1,
		const __m256i *BMRESTRICT	src2 )

inline

2-operand SUB (AND NOT) block digest stride dst = *src1 & ~*src2

Returns: true if stide is all zero

Definition at line 1280 of file bmavx2.h.

References BMRESTRICT.

◆ avx2_sub_digest_3way()

bool bm::avx2_sub_digest_3way	(	__m256i *BMRESTRICT	dst,
		const __m256i *BMRESTRICT	src1,
		const __m256i *BMRESTRICT	src2 )

inline

SUB block digest stride.

Definition at line 1392 of file bmavx2.h.

References BMRESTRICT.

◆ avx2_sub_digest_5way()

bool bm::avx2_sub_digest_5way	(	__m256i *BMRESTRICT	dst,
		const __m256i *BMRESTRICT	src1,
		const __m256i *BMRESTRICT	src2,
		const __m256i *BMRESTRICT	src3,
		const __m256i *BMRESTRICT	src4 )

inline

SUB block digest stride.

Definition at line 1310 of file bmavx2.h.

References BMRESTRICT.

◆ avx2_test_all_eq_wave2()

BMFORCEINLINE bool bm::avx2_test_all_eq_wave2	(	const void *	ptr0,
		const void *	ptr1 )

check if 2 wave of pointers are all the same (NULL or FULL)

Definition at line 1829 of file bmavx2.h.

Referenced by bm::bvector< Alloc >::combine_operation_or().

◆ avx2_test_all_one_wave()

BMFORCEINLINE bool bm::avx2_test_all_one_wave ( const void * ptr )

check if wave of pointers is all 0xFFF

Definition at line 1791 of file bmavx2.h.

◆ avx2_test_all_zero_wave()

BMFORCEINLINE bool bm::avx2_test_all_zero_wave ( const void * ptr )

check if wave of pointers is all NULL

Definition at line 1805 of file bmavx2.h.

Referenced by bm::bvector< Alloc >::combine_operation_and(), bm::bvector< Alloc >::combine_operation_sub(), find_not_null_ptr(), for_each_bit(), and for_each_nzblock().

◆ avx2_test_all_zero_wave2()

BMFORCEINLINE bool bm::avx2_test_all_zero_wave2	(	const void *	ptr0,
		const void *	ptr1 )

check if 2 wave of pointers are all NULL

Definition at line 1816 of file bmavx2.h.

Referenced by bm::bvector< Alloc >::combine_operation_xor().

◆ avx2_xor_arr_2_mask()

void bm::avx2_xor_arr_2_mask	(	__m256i *BMRESTRICT	dst,
		const __m256i *BMRESTRICT	src,
		const __m256i *BMRESTRICT	src_end,
		bm::word_t	mask )

inline

XOR array elements to specified mask dst = *src ^ mask.

Definition at line 447 of file bmavx2.h.

References BMRESTRICT.

◆ avx2_xor_block()

unsigned bm::avx2_xor_block	(	__m256i *BMRESTRICT	dst,
		const __m256i *BMRESTRICT	src )

inline

XOR block against another dst ^= *src.

Returns: 0 if destination does not have any bits

Definition at line 1108 of file bmavx2.h.

References BMRESTRICT, and set_block_size.

◆ avx2_xor_block_2way()

unsigned bm::avx2_xor_block_2way	(	__m256i *BMRESTRICT	dst,
		const __m256i *BMRESTRICT	src1,
		const __m256i *BMRESTRICT	src2 )

inline

3 operand XOR dst = *src1 ^ src2

Returns: 0 if destination does not have any bits

Definition at line 1154 of file bmavx2.h.

References BMRESTRICT, and set_block_size.

Functions

Detailed Description

Function Documentation

◆ avx2_and_arr_unal()

◆ avx2_and_block()

◆ avx2_and_digest()

◆ avx2_and_digest_2way()

◆ avx2_and_digest_3way()

◆ avx2_and_digest_5way()

◆ avx2_and_or_digest_2way()

◆ avx2_andnot_arr_2_mask()

◆ avx2_bit_block_calc_change()

◆ avx2_bit_block_calc_change_bc()

◆ avx2_bit_block_calc_xor_change()

◆ avx2_bit_block_count()

◆ avx2_bit_block_xor()

◆ avx2_bit_block_xor_2way()

◆ avx2_bit_count()

◆ avx2_bit_count_and()

◆ avx2_bit_count_sub()

◆ avx2_bit_count_xor()

◆ avx2_bit_find_first()

◆ avx2_bit_find_first_diff()

◆ avx2_bit_to_gap()

◆ avx2_block_set_digest()

◆ avx2_cmpge_u16()

◆ avx2_cmpge_u32()

◆ avx2_copy_block()

◆ avx2_copy_block_unalign()

◆ avx2_gap_bfind()

◆ avx2_gap_test()

◆ avx2_invert_block()

◆ avx2_is_all_one()

◆ avx2_is_all_zero()

◆ avx2_is_digest_zero()

◆ avx2_lower_bound_scan_u32()

◆ avx2_or_arr_unal()

◆ avx2_or_block()

◆ avx2_or_block_2way()

◆ avx2_or_block_3way()

◆ avx2_or_block_5way()

◆ avx2_set_block()

◆ avx2_shift_l1()

◆ avx2_shift_r1()

◆ avx2_shift_r1_and()

◆ avx2_stream_block()

◆ avx2_stream_block_unalign()

◆ avx2_sub_block()

◆ avx2_sub_digest()

◆ avx2_sub_digest_2way()

◆ avx2_sub_digest_3way()

◆ avx2_sub_digest_5way()

◆ avx2_test_all_eq_wave2()

◆ avx2_test_all_one_wave()

◆ avx2_test_all_zero_wave()

◆ avx2_test_all_zero_wave2()

◆ avx2_xor_arr_2_mask()

◆ avx2_xor_block()

◆ avx2_xor_block_2way()