Processor specific optimizations for SSE4.2 instructions (internals). More...

Functions
bm::id_t	bm::sse4_bit_count (const __m128i block, const __m128i block_end) BMNOEXCEPT
bm::id_t	bm::sse42_bit_count_digest (const bm::word_t *BMRESTRICT block, bm::id64_t digest) BMNOEXCEPT
bool	bm::sse4_is_all_zero (const __m128i *BMRESTRICT block) BMNOEXCEPT
	check if block is all zero bits
BMFORCEINLINE bool	bm::sse4_is_digest_zero (const __m128i *BMRESTRICT block) BMNOEXCEPT
	check if digest stride is all zero bits
BMFORCEINLINE void	bm::sse4_block_set_digest (__m128i *dst, unsigned value) BMNOEXCEPT
	set digest stride to 0xFF.. or 0x0 value
unsigned	bm::sse4_and_block (__m128i BMRESTRICT dst, const __m128i BMRESTRICT src) BMNOEXCEPT
	AND blocks2 dst &= *src.
BMFORCEINLINE bool	bm::sse4_and_digest (__m128i BMRESTRICT dst, const __m128i BMRESTRICT src) BMNOEXCEPT
	AND block digest stride dst &= *src.
BMFORCEINLINE bool	bm::sse4_and_digest_2way (__m128i BMRESTRICT dst, const __m128i BMRESTRICT src1, const __m128i *BMRESTRICT src2) BMNOEXCEPT
	AND block digest stride dst = *src1 & src2.
bool	bm::sse4_and_or_digest_2way (__m128i BMRESTRICT dst, const __m128i BMRESTRICT src1, const __m128i *BMRESTRICT src2) BMNOEXCEPT
	AND-OR block digest stride dst \|= *src1 & src2.
bool	bm::sse4_and_digest_3way (__m128i BMRESTRICT dst, const __m128i BMRESTRICT src1, const __m128i *BMRESTRICT src2) BMNOEXCEPT
	AND block digest stride.
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) BMNOEXCEPT
	AND block digest stride.
BMFORCEINLINE bool	bm::sse4_sub_digest (__m128i BMRESTRICT dst, const __m128i BMRESTRICT src) BMNOEXCEPT
	SUB (AND NOT) block digest stride dst &= ~*src.
BMFORCEINLINE bool	bm::sse4_sub_digest_2way (__m128i BMRESTRICT dst, const __m128i BMRESTRICT src1, const __m128i *BMRESTRICT src2) BMNOEXCEPT
	2-operand SUB (AND NOT) block digest stride dst = src1 & ~*src2
bool	bm::sse4_sub_digest_5way (__m128i BMRESTRICT dst, const __m128i BMRESTRICT src1, const __m128i BMRESTRICT src2, const __m128i BMRESTRICT src3, const __m128i *BMRESTRICT src4) BMNOEXCEPT
	SUB block digest stride.
bool	bm::sse4_sub_digest_3way (__m128i BMRESTRICT dst, const __m128i BMRESTRICT src1, const __m128i *BMRESTRICT src2) BMNOEXCEPT
	SUB block digest stride.
bool	bm::sse4_is_all_one (const __m128i *BMRESTRICT block) BMNOEXCEPT
	check if block is all ONE bits
BMFORCEINLINE bool	bm::sse42_test_all_one_wave (const void *ptr) BMNOEXCEPT
	check if SSE wave is all oxFFFF...FFF
BMFORCEINLINE bool	bm::sse42_test_all_zero_wave (const void *ptr) BMNOEXCEPT
	check if wave of pointers is all NULL
BMFORCEINLINE bool	bm::sse42_test_all_zero_wave2 (const void ptr0, const void ptr1) BMNOEXCEPT
	check if 2 waves of pointers are all NULL
BMFORCEINLINE bool	bm::sse42_test_all_eq_wave2 (const void ptr0, const void ptr1) BMNOEXCEPT
	check if wave of 2 pointers are the same (null or FULL)
unsigned	bm::sse42_bit_block_calc_change (const __m128i *BMRESTRICT block, unsigned size) BMNOEXCEPT
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) BMNOEXCEPT
void	bm::sse42_bit_block_calc_change_bc (const __m128i BMRESTRICT block, unsigned gc, unsigned *bc) BMNOEXCEPT
bool	bm::sse42_bit_find_first_diff (const __m128i BMRESTRICT block1, const __m128i BMRESTRICT block2, unsigned *pos) BMNOEXCEPT
	Find first bit which is different between two bit-blocks.
bool	bm::sse42_bit_find_first (const __m128i BMRESTRICT block, unsigned off, unsigned pos) BMNOEXCEPT
	Find first non-zero bit.
unsigned	bm::sse4_gap_find (const bm::gap_word_t *BMRESTRICT pbuf, const bm::gap_word_t pos, const unsigned size) BMNOEXCEPT
unsigned	bm::sse42_gap_bfind (const unsigned short BMRESTRICT buf, unsigned pos, unsigned BMRESTRICT is_set) BMNOEXCEPT
	Hybrid binary search, starts as binary, then switches to linear scan.
unsigned	bm::sse42_gap_test (const unsigned short *BMRESTRICT buf, unsigned pos) BMNOEXCEPT
	Hybrid binary search to test GAP value, starts as binary, then switches to scan.
int	bm::sse42_cmpge_u32 (__m128i vect4, unsigned value) BMNOEXCEPT
	Experimental (test) function to do SIMD vector search (lower bound) in sorted, growing array.
bool	bm::sse42_shift_l1 (__m128i block, unsigned empty_acc, unsigned co1) BMNOEXCEPT
	block shift left by 1
bool	bm::sse42_shift_r1 (__m128i block, unsigned empty_acc, unsigned co1) BMNOEXCEPT
	block shift right by 1
bool	bm::sse42_shift_r1_and (__m128i block, bm::word_t co1, const __m128i BMRESTRICT mask_block, bm::id64_t *digest) BMNOEXCEPT
	block shift right by 1 plus AND
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) BMNOEXCEPT
	Build partial XOR product of 2 bit-blocks using digest mask.
void	bm::sse42_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.
bool	bm::sse2_sub_digest_5way (__m128i BMRESTRICT dst, const __m128i BMRESTRICT src1, const __m128i BMRESTRICT src2, const __m128i BMRESTRICT src3, const __m128i *BMRESTRICT src4) BMNOEXCEPT
	SUB block digest stride.
bool	bm::sse2_sub_digest_3way (__m128i BMRESTRICT dst, const __m128i BMRESTRICT src1, const __m128i *BMRESTRICT src2) BMNOEXCEPT
	SUB block digest stride.

Detailed Description

Processor specific optimizations for SSE4.2 instructions (internals).

Function Documentation

◆ sse2_sub_digest_3way()

bool bm::sse2_sub_digest_3way	(	__m128i *BMRESTRICT	dst,
		const __m128i *BMRESTRICT	src1,
		const __m128i *BMRESTRICT	src2 )

inline

SUB block digest stride.

Returns: true if stide is all zero

Definition at line 911 of file bmsse2.h.

References BMNOEXCEPT, and BMRESTRICT.

◆ sse2_sub_digest_5way()

bool bm::sse2_sub_digest_5way	(	__m128i *BMRESTRICT	dst,
		const __m128i *BMRESTRICT	src1,
		const __m128i *BMRESTRICT	src2,
		const __m128i *BMRESTRICT	src3,
		const __m128i *BMRESTRICT	src4 )

inline

SUB block digest stride.

Returns: true if stide is all zero

Definition at line 828 of file bmsse2.h.

References BMNOEXCEPT, and BMRESTRICT.

◆ sse42_bit_block_calc_change()

unsigned bm::sse42_bit_block_calc_change	(	const __m128i *BMRESTRICT	block,
		unsigned	size )

inline

SSE4.2 calculate number of bit changes from 0 to 1

Definition at line 948 of file bmsse4.h.

References BM_ALIGN32, BM_ALIGN32ATTR, BMNOEXCEPT, and BMRESTRICT.

◆ sse42_bit_block_calc_change_bc()

void bm::sse42_bit_block_calc_change_bc	(	const __m128i *BMRESTRICT	block,
		unsigned *	gc,
		unsigned *	bc )

inline

SSE4.2 calculate number of bit changes from 0 to 1

Definition at line 1143 of file bmsse4.h.

References BMNOEXCEPT, BMRESTRICT, and set_block_size.

◆ sse42_bit_block_calc_xor_change()

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 )

inline

SSE4.2 calculate number of bit changes from 0 to 1 of a XOR product

Definition at line 1025 of file bmsse4.h.

References BM_ALIGN32, BM_ALIGN32ATTR, BMNOEXCEPT, and BMRESTRICT.

◆ sse42_bit_block_xor()

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 )

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 1988 of file bmsse4.h.

References block_waves, BMNOEXCEPT, and set_block_digest_wave_size.

◆ sse42_bit_block_xor_2way()

void bm::sse42_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 (if 1)

Definition at line 2058 of file bmsse4.h.

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

◆ sse42_bit_count_digest()

bm::id_t bm::sse42_bit_count_digest	(	const bm::word_t *BMRESTRICT	block,
		bm::id64_t	digest )

inline

SSE4.2 optimized bitcounting, uses digest for positioning

Definition at line 127 of file bmsse4.h.

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

◆ sse42_bit_find_first()

bool bm::sse42_bit_find_first	(	const __m128i *BMRESTRICT	block,
		unsigned	off,
		unsigned *	pos )

inline

Find first non-zero bit.

Definition at line 1274 of file bmsse4.h.

References BM_ALIGN32, BM_ALIGN32ATTR, BM_ASSERT, BM_BSF32, BMNOEXCEPT, BMRESTRICT, and set_block_size.

◆ sse42_bit_find_first_diff()

bool bm::sse42_bit_find_first_diff	(	const __m128i *BMRESTRICT	block1,
		const __m128i *BMRESTRICT	block2,
		unsigned *	pos )

inline

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

Definition at line 1218 of file bmsse4.h.

References BM_ALIGN32, BM_ALIGN32ATTR, BM_ASSERT, BM_BSF32, BMNOEXCEPT, BMRESTRICT, and set_block_size.

◆ sse42_cmpge_u32()

int bm::sse42_cmpge_u32	(	__m128i	vect4,
		unsigned	value )

inline

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

Definition at line 1527 of file bmsse4.h.

References BM_BSF32, and BMNOEXCEPT.

◆ sse42_gap_bfind()

unsigned bm::sse42_gap_bfind	(	const unsigned short *BMRESTRICT	buf,
		unsigned	pos,
		unsigned *BMRESTRICT	is_set )

inline

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

Parameters

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

Returns: GAP index.

Definition at line 1390 of file bmsse4.h.

References BM_ASSERT, BMNOEXCEPT, BMRESTRICT, and sse4_gap_find().

◆ sse42_gap_test()

unsigned bm::sse42_gap_test	(	const unsigned short *BMRESTRICT	buf,
		unsigned	pos )

inline

Hybrid binary search to test GAP value, starts as binary, then switches to scan.

Returns: test result

Definition at line 1461 of file bmsse4.h.

References BM_ASSERT, BMNOEXCEPT, BMRESTRICT, and sse4_gap_find().

◆ sse42_shift_l1()

bool bm::sse42_shift_l1	(	__m128i *	block,
		unsigned *	empty_acc,
		unsigned	co1 )

inline

block shift left by 1

Definition at line 1773 of file bmsse4.h.

References BMNOEXCEPT, and set_block_size.

◆ sse42_shift_r1()

bool bm::sse42_shift_r1	(	__m128i *	block,
		unsigned *	empty_acc,
		unsigned	co1 )

inline

block shift right by 1

Definition at line 1823 of file bmsse4.h.

References BMNOEXCEPT, and set_block_size.

◆ sse42_shift_r1_and()

bool bm::sse42_shift_r1_and	(	__m128i *	block,
		bm::word_t	co1,
		const __m128i *BMRESTRICT	mask_block,
		bm::id64_t *	digest )

inline

block shift right by 1 plus AND

Returns: carry over flag

Definition at line 1873 of file bmsse4.h.

References BM_ASSERT, BMNOEXCEPT, BMRESTRICT, id_max, and set_block_digest_wave_size.

◆ sse42_test_all_eq_wave2()

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)

Definition at line 934 of file bmsse4.h.

References BMNOEXCEPT.

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

◆ sse42_test_all_one_wave()

BMFORCEINLINE bool bm::sse42_test_all_one_wave ( const void * ptr )

check if SSE wave is all oxFFFF...FFF

Definition at line 899 of file bmsse4.h.

References BMNOEXCEPT.

◆ sse42_test_all_zero_wave()

BMFORCEINLINE bool bm::sse42_test_all_zero_wave ( const void * ptr )

check if wave of pointers is all NULL

Definition at line 910 of file bmsse4.h.

References BMNOEXCEPT.

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

◆ sse42_test_all_zero_wave2()

BMFORCEINLINE bool bm::sse42_test_all_zero_wave2	(	const void *	ptr0,
		const void *	ptr1 )

check if 2 waves of pointers are all NULL

Definition at line 921 of file bmsse4.h.

References BMNOEXCEPT.

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

◆ sse4_and_block()

unsigned bm::sse4_and_block	(	__m128i *BMRESTRICT	dst,
		const __m128i *BMRESTRICT	src )

inline

AND blocks2 dst &= *src.

Returns: 0 if no bits were set

Definition at line 294 of file bmsse4.h.

References BMNOEXCEPT, BMRESTRICT, and set_block_size.

◆ sse4_and_digest()

BMFORCEINLINE bool bm::sse4_and_digest	(	__m128i *BMRESTRICT	dst,
		const __m128i *BMRESTRICT	src )

AND block digest stride dst &= *src.

Returns: true if stide is all zero

Definition at line 341 of file bmsse4.h.

References BMNOEXCEPT, and BMRESTRICT.

◆ sse4_and_digest_2way()

BMFORCEINLINE bool bm::sse4_and_digest_2way	(	__m128i *BMRESTRICT	dst,
		const __m128i *BMRESTRICT	src1,
		const __m128i *BMRESTRICT	src2 )

AND block digest stride dst = *src1 & src2.

Returns: true if stide is all zero

Definition at line 389 of file bmsse4.h.

References BMNOEXCEPT, and BMRESTRICT.

◆ sse4_and_digest_3way()

bool bm::sse4_and_digest_3way	(	__m128i *BMRESTRICT	dst,
		const __m128i *BMRESTRICT	src1,
		const __m128i *BMRESTRICT	src2 )

inline

AND block digest stride.

Returns: true if stide is all zero

Definition at line 491 of file bmsse4.h.

References BMNOEXCEPT, and BMRESTRICT.

◆ sse4_and_digest_5way()

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 )

inline

AND block digest stride.

Returns: true if stide is all zero

Definition at line 552 of file bmsse4.h.

References BMNOEXCEPT, and BMRESTRICT.

◆ sse4_and_or_digest_2way()

bool bm::sse4_and_or_digest_2way	(	__m128i *BMRESTRICT	dst,
		const __m128i *BMRESTRICT	src1,
		const __m128i *BMRESTRICT	src2 )

inline

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

Returns: true if stide is all zero

Definition at line 438 of file bmsse4.h.

References BMNOEXCEPT, and BMRESTRICT.

◆ sse4_bit_count()

bm::id_t bm::sse4_bit_count	(	const __m128i *	block,
		const __m128i *	block_end )

inline

SSE4.2 optimized bitcounting .

Definition at line 93 of file bmsse4.h.

References BMNOEXCEPT.

◆ sse4_block_set_digest()

BMFORCEINLINE void bm::sse4_block_set_digest	(	__m128i *	dst,
		unsigned	value )

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

Definition at line 276 of file bmsse4.h.

References BMNOEXCEPT.

◆ sse4_gap_find()

unsigned bm::sse4_gap_find	(	const bm::gap_word_t *BMRESTRICT	pbuf,
		const bm::gap_word_t	pos,
		const unsigned	size )

inline

SSE4.2 check for one to two (variable len) 128 bit SSE lines for gap search results (8 elements)

Definition at line 1340 of file bmsse4.h.

References BM_ASSERT, BMNOEXCEPT, and BMRESTRICT.

Referenced by sse42_gap_bfind(), and sse42_gap_test().

◆ sse4_is_all_one()

bool bm::sse4_is_all_one ( const __m128i *BMRESTRICT block )

inline

check if block is all ONE bits

Definition at line 874 of file bmsse4.h.

References BMNOEXCEPT, BMRESTRICT, and set_block_size.

◆ sse4_is_all_zero()

bool bm::sse4_is_all_zero ( const __m128i *BMRESTRICT block )

inline

check if block is all zero bits

Definition at line 232 of file bmsse4.h.

References BMNOEXCEPT, BMRESTRICT, and set_block_size.

◆ sse4_is_digest_zero()

BMFORCEINLINE bool bm::sse4_is_digest_zero ( const __m128i *BMRESTRICT block )

check if digest stride is all zero bits

Definition at line 257 of file bmsse4.h.

References BMNOEXCEPT, and BMRESTRICT.

◆ sse4_sub_digest()

BMFORCEINLINE bool bm::sse4_sub_digest	(	__m128i *BMRESTRICT	dst,
		const __m128i *BMRESTRICT	src )

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

Returns: true if stide is all zero

Definition at line 636 of file bmsse4.h.

References BMNOEXCEPT, and BMRESTRICT.

◆ sse4_sub_digest_2way()

BMFORCEINLINE 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

Returns: true if stide is all zero

Definition at line 685 of file bmsse4.h.

References BMNOEXCEPT, and BMRESTRICT.

◆ sse4_sub_digest_3way()

bool bm::sse4_sub_digest_3way	(	__m128i *BMRESTRICT	dst,
		const __m128i *BMRESTRICT	src1,
		const __m128i *BMRESTRICT	src2 )

inline

SUB block digest stride.

Returns: true if stide is all zero

Definition at line 814 of file bmsse4.h.

References BMNOEXCEPT, and BMRESTRICT.

◆ sse4_sub_digest_5way()

bool bm::sse4_sub_digest_5way	(	__m128i *BMRESTRICT	dst,
		const __m128i *BMRESTRICT	src1,
		const __m128i *BMRESTRICT	src2,
		const __m128i *BMRESTRICT	src3,
		const __m128i *BMRESTRICT	src4 )

inline

SUB block digest stride.

Returns: true if stide is all zero

Definition at line 732 of file bmsse4.h.

References BMNOEXCEPT, and BMRESTRICT.

Functions

Detailed Description

Function Documentation

◆ sse2_sub_digest_3way()

◆ sse2_sub_digest_5way()

◆ sse42_bit_block_calc_change()

◆ sse42_bit_block_calc_change_bc()

◆ sse42_bit_block_calc_xor_change()

◆ sse42_bit_block_xor()

◆ sse42_bit_block_xor_2way()

◆ sse42_bit_count_digest()

◆ sse42_bit_find_first()

◆ sse42_bit_find_first_diff()

◆ sse42_cmpge_u32()

◆ sse42_gap_bfind()

◆ sse42_gap_test()

◆ sse42_shift_l1()

◆ sse42_shift_r1()

◆ sse42_shift_r1_and()

◆ sse42_test_all_eq_wave2()

◆ sse42_test_all_one_wave()

◆ sse42_test_all_zero_wave()

◆ sse42_test_all_zero_wave2()

◆ sse4_and_block()

◆ sse4_and_digest()

◆ sse4_and_digest_2way()

◆ sse4_and_digest_3way()

◆ sse4_and_digest_5way()

◆ sse4_and_or_digest_2way()

◆ sse4_bit_count()

◆ sse4_block_set_digest()

◆ sse4_gap_find()

◆ sse4_is_all_one()

◆ sse4_is_all_zero()

◆ sse4_is_digest_zero()

◆ sse4_sub_digest()

◆ sse4_sub_digest_2way()

◆ sse4_sub_digest_3way()

◆ sse4_sub_digest_5way()