Universal intrinsics

Topics
	Private implementation helpers

Detailed Description

"Universal intrinsics" is a types and functions set intended to simplify vectorization of code on different platforms. Currently a few different SIMD extensions on different architectures are supported. 128 bit registers of various types support is implemented for a wide range of architectures including x86(SSE/SSE2/SSE4.2), ARM(NEON), PowerPC(VSX), MIPS(MSA). 256 bit long registers are supported on x86(AVX2) and 512 bit long registers are supported on x86(AVX512). In case when there is no SIMD extension available during compilation, fallback C++ implementation of intrinsics will be chosen and code will work as expected although it could be slower.

Types

There are several types representing packed values vector registers, each type is implemented as a structure based on a one SIMD register.

• cv::v_uint8 and cv::v_int8: 8-bit integer values (unsigned/signed) - char
• cv::v_uint16 and cv::v_int16: 16-bit integer values (unsigned/signed) - short
• cv::v_uint32 and cv::v_int32: 32-bit integer values (unsigned/signed) - int
• cv::v_uint64 and cv::v_int64: 64-bit integer values (unsigned/signed) - int64
• cv::v_float32: 32-bit floating point values (signed) - float
• cv::v_float64: 64-bit floating point values (signed) - double

Exact bit length(and value quantity) of listed types is compile time deduced and depends on architecture SIMD capabilities chosen as available during compilation of the library. All the types contains nlanes enumeration to check for exact value quantity of the type.

In case the exact bit length of the type is important it is possible to use specific fixed length register types.

There are several types representing 128-bit registers.

• cv::v_uint8x16 and cv::v_int8x16: sixteen 8-bit integer values (unsigned/signed) - char
• cv::v_uint16x8 and cv::v_int16x8: eight 16-bit integer values (unsigned/signed) - short
• cv::v_uint32x4 and cv::v_int32x4: four 32-bit integer values (unsigned/signed) - int
• cv::v_uint64x2 and cv::v_int64x2: two 64-bit integer values (unsigned/signed) - int64
• cv::v_float32x4: four 32-bit floating point values (signed) - float
• cv::v_float64x2: two 64-bit floating point values (signed) - double

There are several types representing 256-bit registers.

• cv::v_uint8x32 and cv::v_int8x32: thirty two 8-bit integer values (unsigned/signed) - char
• cv::v_uint16x16 and cv::v_int16x16: sixteen 16-bit integer values (unsigned/signed) - short
• cv::v_uint32x8 and cv::v_int32x8: eight 32-bit integer values (unsigned/signed) - int
• cv::v_uint64x4 and cv::v_int64x4: four 64-bit integer values (unsigned/signed) - int64
• cv::v_float32x8: eight 32-bit floating point values (signed) - float
• cv::v_float64x4: four 64-bit floating point values (signed) - double

Note

256 bit registers at the moment implemented for AVX2 SIMD extension only, if you want to use this type directly, don't forget to check the CV_SIMD256 preprocessor definition:

#if CV_SIMD256
//...
#endif

There are several types representing 512-bit registers.

• cv::v_uint8x64 and cv::v_int8x64: sixty four 8-bit integer values (unsigned/signed) - char
• cv::v_uint16x32 and cv::v_int16x32: thirty two 16-bit integer values (unsigned/signed) - short
• cv::v_uint32x16 and cv::v_int32x16: sixteen 32-bit integer values (unsigned/signed) - int
• cv::v_uint64x8 and cv::v_int64x8: eight 64-bit integer values (unsigned/signed) - int64
• cv::v_float32x16: sixteen 32-bit floating point values (signed) - float
• cv::v_float64x8: eight 64-bit floating point values (signed) - double

  Note

  512 bit registers at the moment implemented for AVX512 SIMD extension only, if you want to use this type directly, don't forget to check the CV_SIMD512 preprocessor definition.

  cv::v_float64x2 is not implemented in NEON variant, if you want to use this type, don't forget to check the CV_SIMD128_64F preprocessor definition.

Load and store operations

These operations allow to set contents of the register explicitly or by loading it from some memory block and to save contents of the register to memory block.

There are variable size register load operations that provide result of maximum available size depending on chosen platform capabilities.

• Constructors: from memory,
• Other create methods: vx_setall_s8, vx_setall_u8, ..., vx_setzero_u8, vx_setzero_s8, ...
• Memory load operations: vx_load, vx_load_aligned, vx_load_low, vx_load_halves,
• Memory operations with expansion of values: vx_load_expand, vx_load_expand_q

Also there are fixed size register load/store operations.

For 128 bit registers

• Constructors: from memory, from two values, ...
• Other create methods: v_setall_s8, v_setall_u8, ..., v_setzero_u8, v_setzero_s8, ...
• Memory load operations: v_load, v_load_aligned, v_load_low, v_load_halves,
• Memory operations with expansion of values: v_load_expand, v_load_expand_q

For 256 bit registers(check CV_SIMD256 preprocessor definition)

• Constructors: from memory, from four values, ...
• Other create methods: v256_setall_s8, v256_setall_u8, ..., v256_setzero_u8, v256_setzero_s8, ...
• Memory load operations: v256_load, v256_load_aligned, v256_load_low, v256_load_halves,
• Memory operations with expansion of values: v256_load_expand, v256_load_expand_q

For 512 bit registers(check CV_SIMD512 preprocessor definition)

• Constructors: from memory, from eight values, ...
• Other create methods: v512_setall_s8, v512_setall_u8, ..., v512_setzero_u8, v512_setzero_s8, ...
• Memory load operations: v512_load, v512_load_aligned, v512_load_low, v512_load_halves,
• Memory operations with expansion of values: v512_load_expand, v512_load_expand_q

Store to memory operations are similar across different platform capabilities: v_store, v_store_aligned, v_store_high, v_store_low

Value reordering

These operations allow to reorder or recombine elements in one or multiple vectors.

• Interleave, deinterleave (2, 3 and 4 channels): v_load_deinterleave, v_store_interleave
• Expand: v_expand, v_expand_low, v_expand_high
• Pack: v_pack, v_pack_u, v_pack_b, v_rshr_pack, v_rshr_pack_u, v_pack_store, v_pack_u_store, v_rshr_pack_store, v_rshr_pack_u_store
• Recombine: v_zip, v_recombine, v_combine_low, v_combine_high
• Reverse: v_reverse
• Extract: v_extract

Arithmetic, bitwise and comparison operations

Element-wise binary and unary operations.

• Arithmetics: +, -, *, /, v_mul_expand
• Non-saturating arithmetics: v_add_wrap, v_sub_wrap
• Bitwise shifts: <<, >>, v_shl, v_shr
• Bitwise logic: &, |, ^, ~
• Comparison: >, >=, <, <=, ==, !=
• min/max: v_min, v_max

Reduce and mask

Most of these operations return only one value.

• Reduce: v_reduce_min, v_reduce_max, v_reduce_sum, v_popcount
• Mask: v_signmask, v_check_all, v_check_any, v_select

Other math

• Some frequent operations: v_sqrt, v_invsqrt, v_magnitude, v_sqr_magnitude, v_exp, v_log, v_erf, v_sin, v_cos
• Absolute values: v_abs, v_absdiff, v_absdiffs

Conversions

Different type conversions and casts:

• Rounding: v_round, v_floor, v_ceil, v_trunc,
• To float: v_cvt_f32, v_cvt_f64
• Reinterpret: v_reinterpret_as_u8, v_reinterpret_as_s8, ...

Matrix operations

In these operations vectors represent matrix rows/columns: v_dotprod, v_dotprod_fast, v_dotprod_expand, v_dotprod_expand_fast, v_matmul, v_transpose4x4

Usability

Most operations are implemented only for some subset of the available types, following matrices shows the applicability of different operations to the types.

Regular integers:

Operations\Types	uint 8	int 8	uint 16	int 16	uint 32	int 32
load, store	x	x	x	x	x	x
interleave	x	x	x	x	x	x
expand	x	x	x	x	x	x
expand_low	x	x	x	x	x	x
expand_high	x	x	x	x	x	x
expand_q	x	x
add, sub	x	x	x	x	x	x
add_wrap, sub_wrap	x	x	x	x
mul_wrap	x	x	x	x
mul	x	x	x	x	x	x
mul_expand	x	x	x	x	x
compare	x	x	x	x	x	x
shift			x	x	x	x
dotprod				x		x
dotprod_fast				x		x
dotprod_expand	x	x	x	x		x
dotprod_expand_fast	x	x	x	x		x
logical	x	x	x	x	x	x
min, max	x	x	x	x	x	x
absdiff	x	x	x	x	x	x
absdiffs		x		x
reduce	x	x	x	x	x	x
mask	x	x	x	x	x	x
pack	x	x	x	x	x	x
pack_u	x		x
pack_b	x
unpack	x	x	x	x	x	x
extract	x	x	x	x	x	x
rotate (lanes)	x	x	x	x	x	x
cvt_flt32						x
cvt_flt64						x
transpose4x4					x	x
reverse	x	x	x	x	x	x
extract_n	x	x	x	x	x	x
broadcast_element					x	x

Big integers:

Operations\Types	uint 64	int 64
load, store	x	x
add, sub	x	x
shift	x	x
logical	x	x
reverse	x	x
extract	x	x
rotate (lanes)	x	x
cvt_flt64		x
extract_n	x	x

Floating point:

Operations\Types	float 32	float 64
load, store	x	x
interleave	x
add, sub	x	x
mul	x	x
div	x	x
compare	x	x
min, max	x	x
absdiff	x	x
reduce	x
mask	x	x
unpack	x	x
cvt_flt32		x
cvt_flt64	x
sqrt, abs	x	x
float math	x	x
transpose4x4	x
extract	x	x
rotate (lanes)	x	x
reverse	x	x
extract_n	x	x
broadcast_element	x
exp	x	x
log	x	x
sin, cos	x	x

Classes
struct	cv::v_reg< _Tp, n >

Macros
#define	OPENCV_HAL_HAVE_PACK_STORE_BFLOAT16   1
#define	OPENCV_HAL_MATH_HAVE_EXP   1

Typedefs
typedef v_float32x16	simd512::v_float32
	Maximum available vector register capacity 32-bit floating point values (single precision)
typedef v_reg< float, 16 >	cv::v_float32x16
	Sixteen 32-bit floating point values (single precision)
typedef v_reg< float, 4 >	cv::v_float32x4
	Four 32-bit floating point values (single precision)
typedef v_reg< float, 8 >	cv::v_float32x8
	Eight 32-bit floating point values (single precision)
typedef v_float64x8	simd512::v_float64
	Maximum available vector register capacity 64-bit floating point values (double precision)
typedef v_reg< double, 2 >	cv::v_float64x2
	Two 64-bit floating point values (double precision)
typedef v_reg< double, 4 >	cv::v_float64x4
	Four 64-bit floating point values (double precision)
typedef v_reg< double, 8 >	cv::v_float64x8
	Eight 64-bit floating point values (double precision)
typedef v_int16x32	simd512::v_int16
	Maximum available vector register capacity 16-bit signed integer values.
typedef v_reg< short, 16 >	cv::v_int16x16
	Sixteen 16-bit signed integer values.
typedef v_reg< short, 32 >	cv::v_int16x32
	Thirty two 16-bit signed integer values.
typedef v_reg< short, 8 >	cv::v_int16x8
	Eight 16-bit signed integer values.
typedef v_int32x16	simd512::v_int32
	Maximum available vector register capacity 32-bit signed integer values.
typedef v_reg< int, 16 >	cv::v_int32x16
	Sixteen 32-bit signed integer values.
typedef v_reg< int, 4 >	cv::v_int32x4
	Four 32-bit signed integer values.
typedef v_reg< int, 8 >	cv::v_int32x8
	Eight 32-bit signed integer values.
typedef v_int64x8	simd512::v_int64
	Maximum available vector register capacity 64-bit signed integer values.
typedef v_reg< int64, 2 >	cv::v_int64x2
	Two 64-bit signed integer values.
typedef v_reg< int64, 4 >	cv::v_int64x4
	Four 64-bit signed integer values.
typedef v_reg< int64, 8 >	cv::v_int64x8
	Eight 64-bit signed integer values.
typedef v_int8x64	simd512::v_int8
	Maximum available vector register capacity 8-bit signed integer values.
typedef v_reg< schar, 16 >	cv::v_int8x16
	Sixteen 8-bit signed integer values.
typedef v_reg< schar, 32 >	cv::v_int8x32
	Thirty two 8-bit signed integer values.
typedef v_reg< schar, 64 >	cv::v_int8x64
	Sixty four 8-bit signed integer values.
typedef v_uint16x32	simd512::v_uint16
	Maximum available vector register capacity 16-bit unsigned integer values.
typedef v_reg< ushort, 16 >	cv::v_uint16x16
	Sixteen 16-bit unsigned integer values.
typedef v_reg< ushort, 32 >	cv::v_uint16x32
	Thirty two 16-bit unsigned integer values.
typedef v_reg< ushort, 8 >	cv::v_uint16x8
	Eight 16-bit unsigned integer values.
typedef v_uint32x16	simd512::v_uint32
	Maximum available vector register capacity 32-bit unsigned integer values.
typedef v_reg< unsigned, 16 >	cv::v_uint32x16
	Sixteen 32-bit unsigned integer values.
typedef v_reg< unsigned, 4 >	cv::v_uint32x4
	Four 32-bit unsigned integer values.
typedef v_reg< unsigned, 8 >	cv::v_uint32x8
	Eight 32-bit unsigned integer values.
typedef v_uint64x8	simd512::v_uint64
	Maximum available vector register capacity 64-bit unsigned integer values.
typedef v_reg< uint64, 2 >	cv::v_uint64x2
	Two 64-bit unsigned integer values.
typedef v_reg< uint64, 4 >	cv::v_uint64x4
	Four 64-bit unsigned integer values.
typedef v_reg< uint64, 8 >	cv::v_uint64x8
	Eight 64-bit unsigned integer values.
typedef v_uint8x64	simd512::v_uint8
	Maximum available vector register capacity 8-bit unsigned integer values.
typedef v_reg< uchar, 16 >	cv::v_uint8x16
	Sixteen 8-bit unsigned integer values.
typedef v_reg< uchar, 32 >	cv::v_uint8x32
	Thirty two 8-bit unsigned integer values.
typedef v_reg< uchar, 64 >	cv::v_uint8x64
	Sixty four 8-bit unsigned integer values.

Enumerations
enum	{   cv::simd128_width = 16 ,   cv::simd256_width = 32 ,   cv::simd512_width = 64 ,   cv::simdmax_width = simd512_width }

Functions
void	cv::v256_cleanup ()
template<typename _Tp >
v_reg< _Tp, simd256_width/sizeof(_Tp)>	cv::v256_load (const _Tp *ptr)
	Load 256-bit length register contents from memory.
template<typename _Tp >
v_reg< _Tp, simd256_width/sizeof(_Tp)>	cv::v256_load_aligned (const _Tp *ptr)
	Load register contents from memory (aligned)
template<typename _Tp >
v_reg< typename V_TypeTraits< _Tp >::w_type, simd256_width/sizeof(typename V_TypeTraits< _Tp >::w_type)>	cv::v256_load_expand (const _Tp *ptr)
	Load register contents from memory with double expand.
v_reg< float, simd256_width/sizeof(float)>	cv::v256_load_expand (const hfloat *ptr)
template<typename _Tp >
v_reg< typename V_TypeTraits< _Tp >::q_type, simd256_width/sizeof(typename V_TypeTraits< _Tp >::q_type)>	cv::v256_load_expand_q (const _Tp *ptr)
	Load register contents from memory with quad expand.
template<typename _Tp >
v_reg< _Tp, simd256_width/sizeof(_Tp)>	cv::v256_load_halves (const _Tp *loptr, const _Tp *hiptr)
	Load register contents from two memory blocks.
template<typename _Tp >
v_reg< _Tp, simd256_width/sizeof(_Tp)>	cv::v256_load_low (const _Tp *ptr)
	Load 128-bits of data to lower part (high part is undefined).
void	cv::v512_cleanup ()
template<typename _Tp >
v_reg< _Tp, simd512_width/sizeof(_Tp)>	cv::v512_load (const _Tp *ptr)
	Load 512-bit length register contents from memory.
template<typename _Tp >
v_reg< _Tp, simd512_width/sizeof(_Tp)>	cv::v512_load_aligned (const _Tp *ptr)
	Load register contents from memory (aligned)
template<typename _Tp >
v_reg< typename V_TypeTraits< _Tp >::w_type, simd512_width/sizeof(typename V_TypeTraits< _Tp >::w_type)>	cv::v512_load_expand (const _Tp *ptr)
	Load register contents from memory with double expand.
v_reg< float, simd512_width/sizeof(float)>	cv::v512_load_expand (const hfloat *ptr)
template<typename _Tp >
v_reg< typename V_TypeTraits< _Tp >::q_type, simd512_width/sizeof(typename V_TypeTraits< _Tp >::q_type)>	cv::v512_load_expand_q (const _Tp *ptr)
	Load register contents from memory with quad expand.
template<typename _Tp >
v_reg< _Tp, simd512_width/sizeof(_Tp)>	cv::v512_load_halves (const _Tp *loptr, const _Tp *hiptr)
	Load register contents from two memory blocks.
template<typename _Tp >
v_reg< _Tp, simd512_width/sizeof(_Tp)>	cv::v512_load_low (const _Tp *ptr)
	Load 256-bits of data to lower part (high part is undefined).
template<typename _Tp , int n>
v_reg< typename V_TypeTraits< _Tp >::abs_type, n >	cv::v_abs (const v_reg< _Tp, n > &a)
	Absolute value of elements.
template<typename _Tp , int n>
v_reg< typename V_TypeTraits< _Tp >::abs_type, n >	cv::v_absdiff (const v_reg< _Tp, n > &a, const v_reg< _Tp, n > &b)
	Absolute difference.
template<int n>
v_reg< double, n >	cv::v_absdiff (const v_reg< double, n > &a, const v_reg< double, n > &b)
template<int n>
v_reg< float, n >	cv::v_absdiff (const v_reg< float, n > &a, const v_reg< float, n > &b)
template<typename _Tp , int n>
v_reg< _Tp, n >	cv::v_absdiffs (const v_reg< _Tp, n > &a, const v_reg< _Tp, n > &b)
	Saturating absolute difference.
template<typename _Tp , int n>
v_reg< _Tp, n >	cv::v_add (const v_reg< _Tp, n > &a, const v_reg< _Tp, n > &b)
	Add values.
template<typename _Tp , int n>
v_reg< _Tp, n >	cv::v_add_wrap (const v_reg< _Tp, n > &a, const v_reg< _Tp, n > &b)
	Add values without saturation.
template<typename _Tp , int n>
v_reg< _Tp, n >	cv::v_and (const v_reg< _Tp, n > &a, const v_reg< _Tp, n > &b)
	Bitwise AND.
template<int i, typename _Tp , int n>
v_reg< _Tp, n >	cv::v_broadcast_element (const v_reg< _Tp, n > &a)
	Broadcast i-th element of vector.
template<int n>
v_reg< int, n *2 >	cv::v_ceil (const v_reg< double, n > &a)
template<int n>
v_reg< int, n >	cv::v_ceil (const v_reg< float, n > &a)
	Ceil elements.
template<typename _Tp , int n>
bool	cv::v_check_all (const v_reg< _Tp, n > &a)
	Check if all packed values are less than zero.
template<typename _Tp , int n>
bool	cv::v_check_any (const v_reg< _Tp, n > &a)
	Check if any of packed values is less than zero.
void	cv::v_cleanup ()
template<typename _Tp , int n>
v_reg< _Tp, n >	cv::v_combine_high (const v_reg< _Tp, n > &a, const v_reg< _Tp, n > &b)
	Combine vector from last elements of two vectors.
template<typename _Tp , int n>
v_reg< _Tp, n >	cv::v_combine_low (const v_reg< _Tp, n > &a, const v_reg< _Tp, n > &b)
	Combine vector from first elements of two vectors.
template<typename _Tp , int n>
v_reg< _Tp, n >	cv::v_cos (const v_reg< _Tp, n > &a)
	Cosine \( cos(x) \) of elements.
template<int n>
v_reg< float, n *2 >	cv::v_cvt_f32 (const v_reg< double, n > &a)
	Convert lower half to float.
template<int n>
v_reg< float, n *2 >	cv::v_cvt_f32 (const v_reg< double, n > &a, const v_reg< double, n > &b)
	Convert to float.
template<int n>
v_reg< float, n >	cv::v_cvt_f32 (const v_reg< int, n > &a)
	Convert to float.
template<int n>
v_reg< double,(n/2)>	cv::v_cvt_f64 (const v_reg< float, n > &a)
	Convert lower half to double.
template<int n>
v_reg< double, n/2 >	cv::v_cvt_f64 (const v_reg< int, n > &a)
	Convert lower half to double.
template<int n>
v_reg< double, n >	cv::v_cvt_f64 (const v_reg< int64, n > &a)
	Convert to double.
template<int n>
v_reg< double,(n/2)>	cv::v_cvt_f64_high (const v_reg< float, n > &a)
	Convert to double high part of vector.
template<int n>
v_reg< double,(n/2)>	cv::v_cvt_f64_high (const v_reg< int, n > &a)
	Convert to double high part of vector.
template<typename _Tp , int n>
v_reg< _Tp, n >	cv::v_div (const v_reg< _Tp, n > &a, const v_reg< _Tp, n > &b)
	Divide values.
template<typename _Tp , int n>
v_reg< typename V_TypeTraits< _Tp >::w_type, n/2 >	cv::v_dotprod (const v_reg< _Tp, n > &a, const v_reg< _Tp, n > &b)
	Dot product of elements.
template<typename _Tp , int n>
v_reg< typename V_TypeTraits< _Tp >::w_type, n/2 >	cv::v_dotprod (const v_reg< _Tp, n > &a, const v_reg< _Tp, n > &b, const v_reg< typename V_TypeTraits< _Tp >::w_type, n/2 > &c)
	Dot product of elements.
template<typename _Tp , int n>
v_reg< typename V_TypeTraits< _Tp >::q_type, n/4 >	cv::v_dotprod_expand (const v_reg< _Tp, n > &a, const v_reg< _Tp, n > &b)
	Dot product of elements and expand.
template<typename _Tp , int n>
v_reg< typename V_TypeTraits< _Tp >::q_type, n/4 >	cv::v_dotprod_expand (const v_reg< _Tp, n > &a, const v_reg< _Tp, n > &b, const v_reg< typename V_TypeTraits< _Tp >::q_type, n/4 > &c)
	Dot product of elements.
template<int n>
v_reg< double, n/2 >	cv::v_dotprod_expand (const v_reg< int, n > &a, const v_reg< int, n > &b)
template<int n>
v_reg< double, n/2 >	cv::v_dotprod_expand (const v_reg< int, n > &a, const v_reg< int, n > &b, const v_reg< double, n/2 > &c)
template<typename _Tp , int n>
v_reg< typename V_TypeTraits< _Tp >::q_type, n/4 >	cv::v_dotprod_expand_fast (const v_reg< _Tp, n > &a, const v_reg< _Tp, n > &b)
	Fast Dot product of elements and expand.
template<typename _Tp , int n>
v_reg< typename V_TypeTraits< _Tp >::q_type, n/4 >	cv::v_dotprod_expand_fast (const v_reg< _Tp, n > &a, const v_reg< _Tp, n > &b, const v_reg< typename V_TypeTraits< _Tp >::q_type, n/4 > &c)
	Fast Dot product of elements.
template<int n>
v_reg< double, n/2 >	cv::v_dotprod_expand_fast (const v_reg< int, n > &a, const v_reg< int, n > &b)
template<int n>
v_reg< double, n/2 >	cv::v_dotprod_expand_fast (const v_reg< int, n > &a, const v_reg< int, n > &b, const v_reg< double, n/2 > &c)
template<typename _Tp , int n>
v_reg< typename V_TypeTraits< _Tp >::w_type, n/2 >	cv::v_dotprod_fast (const v_reg< _Tp, n > &a, const v_reg< _Tp, n > &b)
	Fast Dot product of elements.
template<typename _Tp , int n>
v_reg< typename V_TypeTraits< _Tp >::w_type, n/2 >	cv::v_dotprod_fast (const v_reg< _Tp, n > &a, const v_reg< _Tp, n > &b, const v_reg< typename V_TypeTraits< _Tp >::w_type, n/2 > &c)
	Fast Dot product of elements.
template<typename _Tp , int n>
v_reg< _Tp, n >	cv::v_eq (const v_reg< _Tp, n > &a, const v_reg< _Tp, n > &b)
	Equal comparison.
template<typename _Tp , int n>
v_reg< _Tp, n >	cv::v_erf (const v_reg< _Tp, n > &a)
	Error function.
template<typename _Tp , int n>
v_reg< _Tp, n >	cv::v_exp (const v_reg< _Tp, n > &a)
	Exponential \( e^x \) of elements.
template<typename _Tp , int n>
void	cv::v_expand (const v_reg< _Tp, n > &a, v_reg< typename V_TypeTraits< _Tp >::w_type, n/2 > &b0, v_reg< typename V_TypeTraits< _Tp >::w_type, n/2 > &b1)
	Expand values to the wider pack type.
template<typename _Tp , int n>
v_reg< typename V_TypeTraits< _Tp >::w_type, n/2 >	cv::v_expand_high (const v_reg< _Tp, n > &a)
	Expand higher values to the wider pack type.
template<typename _Tp , int n>
v_reg< typename V_TypeTraits< _Tp >::w_type, n/2 >	cv::v_expand_low (const v_reg< _Tp, n > &a)
	Expand lower values to the wider pack type.
template<int s, typename _Tp , int n>
v_reg< _Tp, n >	cv::v_extract (const v_reg< _Tp, n > &a, const v_reg< _Tp, n > &b)
	Vector extract.
template<int s, typename _Tp , int n>
_Tp	cv::v_extract_n (const v_reg< _Tp, n > &v)
	Vector extract.
template<int n>
v_reg< int, n *2 >	cv::v_floor (const v_reg< double, n > &a)
template<int n>
v_reg< int, n >	cv::v_floor (const v_reg< float, n > &a)
	Floor elements.
template<typename _Tp , int n>
v_reg< _Tp, n >	cv::v_fma (const v_reg< _Tp, n > &a, const v_reg< _Tp, n > &b, const v_reg< _Tp, n > &c)
	Multiply and add.
template<typename _Tp , int n>
v_reg< _Tp, n >	cv::v_ge (const v_reg< _Tp, n > &a, const v_reg< _Tp, n > &b)
	Greater-than or equal comparison.
template<typename _Tp , int n>
v_reg< _Tp, n >	cv::v_gt (const v_reg< _Tp, n > &a, const v_reg< _Tp, n > &b)
	Greater-than comparison.
template<typename _Tp , int n>
v_reg< _Tp, n >	cv::v_interleave_pairs (const v_reg< _Tp, n > &vec)
template<typename _Tp , int n>
v_reg< _Tp, n >	cv::v_interleave_quads (const v_reg< _Tp, n > &vec)
template<typename _Tp , int n>
v_reg< _Tp, n >	cv::v_invsqrt (const v_reg< _Tp, n > &a)
	Inversed square root.
template<typename _Tp , int n>
v_reg< _Tp, n >	cv::v_le (const v_reg< _Tp, n > &a, const v_reg< _Tp, n > &b)
	Less-than or equal comparison.
template<typename _Tp >
v_reg< _Tp, simd128_width/sizeof(_Tp)>	cv::v_load (const _Tp *ptr)
	Load register contents from memory.
template<typename _Tp >
v_reg< _Tp, simd128_width/sizeof(_Tp)>	cv::v_load_aligned (const _Tp *ptr)
	Load register contents from memory (aligned)
template<typename _Tp , int n>
void	cv::v_load_deinterleave (const _Tp *ptr, v_reg< _Tp, n > &a, v_reg< _Tp, n > &b)
	Load and deinterleave (2 channels)
template<typename _Tp , int n>
void	cv::v_load_deinterleave (const _Tp *ptr, v_reg< _Tp, n > &a, v_reg< _Tp, n > &b, v_reg< _Tp, n > &c)
	Load and deinterleave (3 channels)
template<typename _Tp , int n>
void	cv::v_load_deinterleave (const _Tp *ptr, v_reg< _Tp, n > &a, v_reg< _Tp, n > &b, v_reg< _Tp, n > &c, v_reg< _Tp, n > &d)
	Load and deinterleave (4 channels)
template<typename _Tp >
v_reg< typename V_TypeTraits< _Tp >::w_type, simd128_width/sizeof(typename V_TypeTraits< _Tp >::w_type)>	cv::v_load_expand (const _Tp *ptr)
	Load register contents from memory with double expand.
v_reg< float, simd128_width/sizeof(float)>	cv::v_load_expand (const hfloat *ptr)
template<typename _Tp >
v_reg< typename V_TypeTraits< _Tp >::q_type, simd128_width/sizeof(typename V_TypeTraits< _Tp >::q_type)>	cv::v_load_expand_q (const _Tp *ptr)
	Load register contents from memory with quad expand.
template<typename _Tp >
v_reg< _Tp, simd128_width/sizeof(_Tp)>	cv::v_load_halves (const _Tp *loptr, const _Tp *hiptr)
	Load register contents from two memory blocks.
template<typename _Tp >
v_reg< _Tp, simd128_width/sizeof(_Tp)>	cv::v_load_low (const _Tp *ptr)
	Load 64-bits of data to lower part (high part is undefined).
template<typename _Tp , int n>
v_reg< _Tp, n >	cv::v_log (const v_reg< _Tp, n > &a)
	Natural logarithm \( \log(x) \) of elements.
template<typename _Tp , int n>
v_reg< _Tp, n >	cv::v_lt (const v_reg< _Tp, n > &a, const v_reg< _Tp, n > &b)
	Less-than comparison.
template<typename _Tp >
v_reg< _Tp, simd128_width/sizeof(_Tp)>	cv::v_lut (const _Tp *tab, const int *idx)
template<int n>
v_reg< double, n/2 >	cv::v_lut (const double *tab, const v_reg< int, n > &idx)
template<int n>
v_reg< float, n >	cv::v_lut (const float *tab, const v_reg< int, n > &idx)
template<int n>
v_reg< int, n >	cv::v_lut (const int *tab, const v_reg< int, n > &idx)
template<int n>
v_reg< unsigned, n >	cv::v_lut (const unsigned *tab, const v_reg< int, n > &idx)
template<int n>
void	cv::v_lut_deinterleave (const double *tab, const v_reg< int, n *2 > &idx, v_reg< double, n > &x, v_reg< double, n > &y)
template<int n>
void	cv::v_lut_deinterleave (const float *tab, const v_reg< int, n > &idx, v_reg< float, n > &x, v_reg< float, n > &y)
template<typename _Tp >
v_reg< _Tp, simd128_width/sizeof(_Tp)>	cv::v_lut_pairs (const _Tp *tab, const int *idx)
template<typename _Tp >
v_reg< _Tp, simd128_width/sizeof(_Tp)>	cv::v_lut_quads (const _Tp *tab, const int *idx)
template<typename _Tp , int n>
v_reg< _Tp, n >	cv::v_magnitude (const v_reg< _Tp, n > &a, const v_reg< _Tp, n > &b)
	Magnitude.
template<int n>
v_reg< float, n >	cv::v_matmul (const v_reg< float, n > &v, const v_reg< float, n > &a, const v_reg< float, n > &b, const v_reg< float, n > &c, const v_reg< float, n > &d)
	Matrix multiplication.
template<int n>
v_reg< float, n >	cv::v_matmuladd (const v_reg< float, n > &v, const v_reg< float, n > &a, const v_reg< float, n > &b, const v_reg< float, n > &c, const v_reg< float, n > &d)
	Matrix multiplication and add.
template<typename _Tp , int n>
v_reg< _Tp, n >	cv::v_max (const v_reg< _Tp, n > &a, const v_reg< _Tp, n > &b)
	Choose max values for each pair.
template<typename _Tp , int n>
v_reg< _Tp, n >	cv::v_min (const v_reg< _Tp, n > &a, const v_reg< _Tp, n > &b)
	Choose min values for each pair.
template<typename _Tp , int n>
v_reg< _Tp, n >	cv::v_mul (const v_reg< _Tp, n > &a, const v_reg< _Tp, n > &b)
	Multiply values.
template<typename _Tp , int n>
void	cv::v_mul_expand (const v_reg< _Tp, n > &a, const v_reg< _Tp, n > &b, v_reg< typename V_TypeTraits< _Tp >::w_type, n/2 > &c, v_reg< typename V_TypeTraits< _Tp >::w_type, n/2 > &d)
	Multiply and expand.
template<typename _Tp , int n>
v_reg< _Tp, n >	cv::v_mul_hi (const v_reg< _Tp, n > &a, const v_reg< _Tp, n > &b)
	Multiply and extract high part.
template<typename _Tp , int n>
v_reg< _Tp, n >	cv::v_mul_wrap (const v_reg< _Tp, n > &a, const v_reg< _Tp, n > &b)
	Multiply values without saturation.
template<typename _Tp , int n>
v_reg< _Tp, n >	cv::v_muladd (const v_reg< _Tp, n > &a, const v_reg< _Tp, n > &b, const v_reg< _Tp, n > &c)
	A synonym for v_fma.
template<typename _Tp , int n>
v_reg< _Tp, n >	cv::v_ne (const v_reg< _Tp, n > &a, const v_reg< _Tp, n > &b)
	Not equal comparison.
template<typename _Tp , int n>
v_reg< _Tp, n >	cv::v_not (const v_reg< _Tp, n > &a)
	Bitwise NOT.
template<int n>
v_reg< double, n >	cv::v_not_nan (const v_reg< double, n > &a)
template<int n>
v_reg< float, n >	cv::v_not_nan (const v_reg< float, n > &a)
template<typename _Tp , int n>
v_reg< _Tp, n >	cv::v_or (const v_reg< _Tp, n > &a, const v_reg< _Tp, n > &b)
	Bitwise OR.
template<int n>
void	cv::v_pack_store (hfloat *ptr, const v_reg< float, n > &v)
template<typename _Tp , int n>
v_reg< _Tp, n >	cv::v_pack_triplets (const v_reg< _Tp, n > &vec)
template<typename _Tp , int n>
v_reg< typename V_TypeTraits< _Tp >::abs_type, n >	cv::v_popcount (const v_reg< _Tp, n > &a)
	Count the 1 bits in the vector lanes and return result as corresponding unsigned type.
template<typename _Tp , int n>
void	cv::v_recombine (const v_reg< _Tp, n > &a, const v_reg< _Tp, n > &b, v_reg< _Tp, n > &low, v_reg< _Tp, n > &high)
	Combine two vectors from lower and higher parts of two other vectors.
template<typename _Tp , int n>
_Tp	cv::v_reduce_max (const v_reg< _Tp, n > &a)
	Find one max value.
template<typename _Tp , int n>
_Tp	cv::v_reduce_min (const v_reg< _Tp, n > &a)
	Find one min value.
template<typename _Tp , int n>
V_TypeTraits< typenameV_TypeTraits< _Tp >::abs_type >::sum_type	cv::v_reduce_sad (const v_reg< _Tp, n > &a, const v_reg< _Tp, n > &b)
	Sum absolute differences of values.
template<typename _Tp , int n>
V_TypeTraits< _Tp >::sum_type	cv::v_reduce_sum (const v_reg< _Tp, n > &a)
	Sum packed values.
template<int n>
v_reg< float, n >	cv::v_reduce_sum4 (const v_reg< float, n > &a, const v_reg< float, n > &b, const v_reg< float, n > &c, const v_reg< float, n > &d)
	Sums all elements of each input vector, returns the vector of sums.
template<typename _Tp , int n>
v_reg< _Tp, n >	cv::v_reverse (const v_reg< _Tp, n > &a)
	Vector reverse order.
template<int imm, typename _Tp , int n>
v_reg< _Tp, n >	cv::v_rotate_left (const v_reg< _Tp, n > &a)
	Element shift left among vector.
template<int imm, typename _Tp , int n>
v_reg< _Tp, n >	cv::v_rotate_left (const v_reg< _Tp, n > &a, const v_reg< _Tp, n > &b)
template<int imm, typename _Tp , int n>
v_reg< _Tp, n >	cv::v_rotate_right (const v_reg< _Tp, n > &a)
	Element shift right among vector.
template<int imm, typename _Tp , int n>
v_reg< _Tp, n >	cv::v_rotate_right (const v_reg< _Tp, n > &a, const v_reg< _Tp, n > &b)
template<int n>
v_reg< int, n *2 >	cv::v_round (const v_reg< double, n > &a)
template<int n>
v_reg< int, n *2 >	cv::v_round (const v_reg< double, n > &a, const v_reg< double, n > &b)
template<int n>
v_reg< int, n >	cv::v_round (const v_reg< float, n > &a)
	Round elements.
template<typename _Tp , int n>
int	cv::v_scan_forward (const v_reg< _Tp, n > &a)
	Get first negative lane index.
template<typename _Tp , int n>
v_reg< _Tp, n >	cv::v_select (const v_reg< _Tp, n > &mask, const v_reg< _Tp, n > &a, const v_reg< _Tp, n > &b)
	Per-element select (blend operation)
template<typename _Tp , int n>
v_reg< _Tp, n >	cv::v_shl (const v_reg< _Tp, n > &a, int imm)
	Bitwise shift left.
template<typename _Tp , int n>
v_reg< _Tp, n >	cv::v_shr (const v_reg< _Tp, n > &a, int imm)
	Bitwise shift right.
template<typename _Tp , int n>
int	cv::v_signmask (const v_reg< _Tp, n > &a)
	Get negative values mask.
template<typename _Tp , int n>
v_reg< _Tp, n >	cv::v_sin (const v_reg< _Tp, n > &a)
	Sine \( sin(x) \) of elements.
template<typename _Tp , int n>
void	cv::v_sincos (const v_reg< _Tp, n > &x, v_reg< _Tp, n > &s, v_reg< _Tp, n > &c)
	Compute sine \( sin(x) \) and cosine \( cos(x) \) of elements at the same time.
template<typename _Tp , int n>
v_reg< _Tp, n >	cv::v_sqr_magnitude (const v_reg< _Tp, n > &a, const v_reg< _Tp, n > &b)
	Square of the magnitude.
template<typename _Tp , int n>
v_reg< _Tp, n >	cv::v_sqrt (const v_reg< _Tp, n > &a)
	Square root of elements.
template<typename _Tp , int n>
void	cv::v_store (_Tp *ptr, const v_reg< _Tp, n > &a)
	Store data to memory.
template<typename _Tp , int n>
void	cv::v_store (_Tp *ptr, const v_reg< _Tp, n > &a, hal::StoreMode)
template<typename _Tp , int n>
void	cv::v_store_aligned (_Tp *ptr, const v_reg< _Tp, n > &a)
	Store data to memory (aligned)
template<typename _Tp , int n>
void	cv::v_store_aligned (_Tp *ptr, const v_reg< _Tp, n > &a, hal::StoreMode)
template<typename _Tp , int n>
void	cv::v_store_aligned_nocache (_Tp *ptr, const v_reg< _Tp, n > &a)
template<typename _Tp , int n>
void	cv::v_store_high (_Tp *ptr, const v_reg< _Tp, n > &a)
	Store data to memory (higher half)
template<typename _Tp , int n>
void	cv::v_store_interleave (_Tp *ptr, const v_reg< _Tp, n > &a, const v_reg< _Tp, n > &b, const v_reg< _Tp, n > &c, const v_reg< _Tp, n > &d, hal::StoreMode=hal::STORE_UNALIGNED)
	Interleave and store (4 channels)
template<typename _Tp , int n>
void	cv::v_store_interleave (_Tp *ptr, const v_reg< _Tp, n > &a, const v_reg< _Tp, n > &b, const v_reg< _Tp, n > &c, hal::StoreMode=hal::STORE_UNALIGNED)
	Interleave and store (3 channels)
template<typename _Tp , int n>
void	cv::v_store_interleave (_Tp *ptr, const v_reg< _Tp, n > &a, const v_reg< _Tp, n > &b, hal::StoreMode=hal::STORE_UNALIGNED)
	Interleave and store (2 channels)
template<typename _Tp , int n>
void	cv::v_store_low (_Tp *ptr, const v_reg< _Tp, n > &a)
	Store data to memory (lower half)
template<typename _Tp , int n>
v_reg< _Tp, n >	cv::v_sub (const v_reg< _Tp, n > &a, const v_reg< _Tp, n > &b)
	Subtract values.
template<typename _Tp , int n>
v_reg< _Tp, n >	cv::v_sub_wrap (const v_reg< _Tp, n > &a, const v_reg< _Tp, n > &b)
	Subtract values without saturation.
template<typename _Tp , int n>
void	cv::v_transpose4x4 (v_reg< _Tp, n > &a0, const v_reg< _Tp, n > &a1, const v_reg< _Tp, n > &a2, const v_reg< _Tp, n > &a3, v_reg< _Tp, n > &b0, v_reg< _Tp, n > &b1, v_reg< _Tp, n > &b2, v_reg< _Tp, n > &b3)
	Transpose 4x4 matrix.
template<int n>
v_reg< int, n *2 >	cv::v_trunc (const v_reg< double, n > &a)
template<int n>
v_reg< int, n >	cv::v_trunc (const v_reg< float, n > &a)
	Truncate elements.
template<typename _Tp , int n>
v_reg< _Tp, n >	cv::v_xor (const v_reg< _Tp, n > &a, const v_reg< _Tp, n > &b)
	Bitwise XOR.
template<typename _Tp , int n>
void	cv::v_zip (const v_reg< _Tp, n > &a0, const v_reg< _Tp, n > &a1, v_reg< _Tp, n > &b0, v_reg< _Tp, n > &b1)
	Interleave two vectors.

Variables
static const unsigned char	cv::popCountTable []

Init with zero
Create new vector with zero elements
v_uint8x16	cv::v_setzero_u8 ()
template<>
v_uint8x16	cv::v_setzero_ ()
v_int8x16	cv::v_setzero_s8 ()
v_uint16x8	cv::v_setzero_u16 ()
v_int16x8	cv::v_setzero_s16 ()
v_uint32x4	cv::v_setzero_u32 ()
v_int32x4	cv::v_setzero_s32 ()
v_float32x4	cv::v_setzero_f32 ()
v_float64x2	cv::v_setzero_f64 ()
v_uint64x2	cv::v_setzero_u64 ()
v_int64x2	cv::v_setzero_s64 ()
v_uint8x32	cv::v256_setzero_u8 ()
v_int8x32	cv::v256_setzero_s8 ()
v_uint16x16	cv::v256_setzero_u16 ()
v_int16x16	cv::v256_setzero_s16 ()
v_uint32x8	cv::v256_setzero_u32 ()
v_int32x8	cv::v256_setzero_s32 ()
v_float32x8	cv::v256_setzero_f32 ()
v_float64x4	cv::v256_setzero_f64 ()
v_uint64x4	cv::v256_setzero_u64 ()
v_int64x4	cv::v256_setzero_s64 ()
v_uint8x64	cv::v512_setzero_u8 ()
v_int8x64	cv::v512_setzero_s8 ()
v_uint16x32	cv::v512_setzero_u16 ()
v_int16x32	cv::v512_setzero_s16 ()
v_uint32x16	cv::v512_setzero_u32 ()
v_int32x16	cv::v512_setzero_s32 ()
v_float32x16	cv::v512_setzero_f32 ()
v_float64x8	cv::v512_setzero_f64 ()
v_uint64x8	cv::v512_setzero_u64 ()
v_int64x8	cv::v512_setzero_s64 ()

Init with value
Create new vector with elements set to a specific value
v_uint8x16	cv::v_setall_u8 (uchar val)
template<>
v_uint8x16	cv::v_setall_ (uchar val)
v_int8x16	cv::v_setall_s8 (schar val)
template<>
v_int8x16	cv::v_setall_ (schar val)
v_uint16x8	cv::v_setall_u16 (ushort val)
template<>
v_uint16x8	cv::v_setall_ (ushort val)
v_int16x8	cv::v_setall_s16 (short val)
template<>
v_int16x8	cv::v_setall_ (short val)
v_uint32x4	cv::v_setall_u32 (unsigned val)
template<>
v_uint32x4	cv::v_setall_ (unsigned val)
v_int32x4	cv::v_setall_s32 (int val)
template<>
v_int32x4	cv::v_setall_ (int val)
v_float32x4	cv::v_setall_f32 (float val)
template<>
v_float32x4	cv::v_setall_ (float val)
v_float64x2	cv::v_setall_f64 (double val)
template<>
v_float64x2	cv::v_setall_ (double val)
v_uint64x2	cv::v_setall_u64 (uint64 val)
template<>
v_uint64x2	cv::v_setall_ (uint64 val)
v_int64x2	cv::v_setall_s64 (int64 val)
template<>
v_int64x2	cv::v_setall_ (int64 val)
v_uint8x32	cv::v256_setall_u8 (uchar val)
v_int8x32	cv::v256_setall_s8 (schar val)
v_uint16x16	cv::v256_setall_u16 (ushort val)
v_int16x16	cv::v256_setall_s16 (short val)
v_uint32x8	cv::v256_setall_u32 (unsigned val)
v_int32x8	cv::v256_setall_s32 (int val)
v_float32x8	cv::v256_setall_f32 (float val)
v_float64x4	cv::v256_setall_f64 (double val)
v_uint64x4	cv::v256_setall_u64 (uint64 val)
v_int64x4	cv::v256_setall_s64 (int64 val)
v_uint8x64	cv::v512_setall_u8 (uchar val)
v_int8x64	cv::v512_setall_s8 (schar val)
v_uint16x32	cv::v512_setall_u16 (ushort val)
v_int16x32	cv::v512_setall_s16 (short val)
v_uint32x16	cv::v512_setall_u32 (unsigned val)
v_int32x16	cv::v512_setall_s32 (int val)
v_float32x16	cv::v512_setall_f32 (float val)
v_float64x8	cv::v512_setall_f64 (double val)
v_uint64x8	cv::v512_setall_u64 (uint64 val)
v_int64x8	cv::v512_setall_s64 (int64 val)

Reinterpret
Convert vector to different type without modifying underlying data.
template<typename _Tp0 , int n0>
v_reg< uchar, n0 *sizeof(_Tp0)/sizeof(uchar)>	cv::v_reinterpret_as_u8 (const v_reg< _Tp0, n0 > &a)
template<typename _Tp0 , int n0>
v_reg< schar, n0 *sizeof(_Tp0)/sizeof(schar)>	cv::v_reinterpret_as_s8 (const v_reg< _Tp0, n0 > &a)
template<typename _Tp0 , int n0>
v_reg< ushort, n0 *sizeof(_Tp0)/sizeof(ushort)>	cv::v_reinterpret_as_u16 (const v_reg< _Tp0, n0 > &a)
template<typename _Tp0 , int n0>
v_reg< short, n0 *sizeof(_Tp0)/sizeof(short)>	cv::v_reinterpret_as_s16 (const v_reg< _Tp0, n0 > &a)
template<typename _Tp0 , int n0>
v_reg< unsigned, n0 *sizeof(_Tp0)/sizeof(unsigned)>	cv::v_reinterpret_as_u32 (const v_reg< _Tp0, n0 > &a)
template<typename _Tp0 , int n0>
v_reg< int, n0 *sizeof(_Tp0)/sizeof(int)>	cv::v_reinterpret_as_s32 (const v_reg< _Tp0, n0 > &a)
template<typename _Tp0 , int n0>
v_reg< float, n0 *sizeof(_Tp0)/sizeof(float)>	cv::v_reinterpret_as_f32 (const v_reg< _Tp0, n0 > &a)
template<typename _Tp0 , int n0>
v_reg< double, n0 *sizeof(_Tp0)/sizeof(double)>	cv::v_reinterpret_as_f64 (const v_reg< _Tp0, n0 > &a)
template<typename _Tp0 , int n0>
v_reg< uint64, n0 *sizeof(_Tp0)/sizeof(uint64)>	cv::v_reinterpret_as_u64 (const v_reg< _Tp0, n0 > &a)
template<typename _Tp0 , int n0>
v_reg< int64, n0 *sizeof(_Tp0)/sizeof(int64)>	cv::v_reinterpret_as_s64 (const v_reg< _Tp0, n0 > &a)

Left shift
Shift left
template<int shift, int n>
v_reg< ushort, n >	cv::v_shl (const v_reg< ushort, n > &a)
template<int shift, int n>
v_reg< short, n >	cv::v_shl (const v_reg< short, n > &a)
template<int shift, int n>
v_reg< unsigned, n >	cv::v_shl (const v_reg< unsigned, n > &a)
template<int shift, int n>
v_reg< int, n >	cv::v_shl (const v_reg< int, n > &a)
template<int shift, int n>
v_reg< uint64, n >	cv::v_shl (const v_reg< uint64, n > &a)
template<int shift, int n>
v_reg< int64, n >	cv::v_shl (const v_reg< int64, n > &a)

Right shift
Shift right
template<int shift, int n>
v_reg< ushort, n >	cv::v_shr (const v_reg< ushort, n > &a)
template<int shift, int n>
v_reg< short, n >	cv::v_shr (const v_reg< short, n > &a)
template<int shift, int n>
v_reg< unsigned, n >	cv::v_shr (const v_reg< unsigned, n > &a)
template<int shift, int n>
v_reg< int, n >	cv::v_shr (const v_reg< int, n > &a)
template<int shift, int n>
v_reg< uint64, n >	cv::v_shr (const v_reg< uint64, n > &a)
template<int shift, int n>
v_reg< int64, n >	cv::v_shr (const v_reg< int64, n > &a)

Rounding shift
Rounding shift right
template<int shift, int n>
v_reg< ushort, n >	cv::v_rshr (const v_reg< ushort, n > &a)
template<int shift, int n>
v_reg< short, n >	cv::v_rshr (const v_reg< short, n > &a)
template<int shift, int n>
v_reg< unsigned, n >	cv::v_rshr (const v_reg< unsigned, n > &a)
template<int shift, int n>
v_reg< int, n >	cv::v_rshr (const v_reg< int, n > &a)
template<int shift, int n>
v_reg< uint64, n >	cv::v_rshr (const v_reg< uint64, n > &a)
template<int shift, int n>
v_reg< int64, n >	cv::v_rshr (const v_reg< int64, n > &a)

Pack
Pack values from two vectors to one Return vector type have twice more elements than input vector types. Variant with u suffix also converts to corresponding unsigned type. pack: for 16-, 32- and 64-bit integer input types pack_u: for 16- and 32-bit signed integer input types Note All variants except 64-bit use saturation.
template<int n>
v_reg< uchar, 2 *n >	cv::v_pack (const v_reg< ushort, n > &a, const v_reg< ushort, n > &b)
template<int n>
v_reg< schar, 2 *n >	cv::v_pack (const v_reg< short, n > &a, const v_reg< short, n > &b)
template<int n>
v_reg< ushort, 2 *n >	cv::v_pack (const v_reg< unsigned, n > &a, const v_reg< unsigned, n > &b)
template<int n>
v_reg< short, 2 *n >	cv::v_pack (const v_reg< int, n > &a, const v_reg< int, n > &b)
template<int n>
v_reg< unsigned, 2 *n >	cv::v_pack (const v_reg< uint64, n > &a, const v_reg< uint64, n > &b)
template<int n>
v_reg< int, 2 *n >	cv::v_pack (const v_reg< int64, n > &a, const v_reg< int64, n > &b)
template<int n>
v_reg< uchar, 2 *n >	cv::v_pack_u (const v_reg< short, n > &a, const v_reg< short, n > &b)
template<int n>
v_reg< ushort, 2 *n >	cv::v_pack_u (const v_reg< int, n > &a, const v_reg< int, n > &b)

Pack with rounding shift
Pack values from two vectors to one with rounding shift Values from the input vectors will be shifted right by n bits with rounding, converted to narrower type and returned in the result vector. Variant with u suffix converts to unsigned type. pack: for 16-, 32- and 64-bit integer input types pack_u: for 16- and 32-bit signed integer input types Note All variants except 64-bit use saturation.
template<int shift, int n>
v_reg< uchar, 2 *n >	cv::v_rshr_pack (const v_reg< ushort, n > &a, const v_reg< ushort, n > &b)
template<int shift, int n>
v_reg< schar, 2 *n >	cv::v_rshr_pack (const v_reg< short, n > &a, const v_reg< short, n > &b)
template<int shift, int n>
v_reg< ushort, 2 *n >	cv::v_rshr_pack (const v_reg< unsigned, n > &a, const v_reg< unsigned, n > &b)
template<int shift, int n>
v_reg< short, 2 *n >	cv::v_rshr_pack (const v_reg< int, n > &a, const v_reg< int, n > &b)
template<int shift, int n>
v_reg< unsigned, 2 *n >	cv::v_rshr_pack (const v_reg< uint64, n > &a, const v_reg< uint64, n > &b)
template<int shift, int n>
v_reg< int, 2 *n >	cv::v_rshr_pack (const v_reg< int64, n > &a, const v_reg< int64, n > &b)
template<int shift, int n>
v_reg< uchar, 2 *n >	cv::v_rshr_pack_u (const v_reg< short, n > &a, const v_reg< short, n > &b)
template<int shift, int n>
v_reg< ushort, 2 *n >	cv::v_rshr_pack_u (const v_reg< int, n > &a, const v_reg< int, n > &b)

Pack and store
Store values from the input vector into memory with pack Values will be stored into memory with conversion to narrower type. Variant with u suffix converts to corresponding unsigned type. pack: for 16-, 32- and 64-bit integer input types pack_u: for 16- and 32-bit signed integer input types Note All variants except 64-bit use saturation.
template<int n>
void	cv::v_pack_store (uchar *ptr, const v_reg< ushort, n > &a)
template<int n>
void	cv::v_pack_store (schar *ptr, const v_reg< short, n > &a)
template<int n>
void	cv::v_pack_store (ushort *ptr, const v_reg< unsigned, n > &a)
template<int n>
void	cv::v_pack_store (short *ptr, const v_reg< int, n > &a)
template<int n>
void	cv::v_pack_store (unsigned *ptr, const v_reg< uint64, n > &a)
template<int n>
void	cv::v_pack_store (int *ptr, const v_reg< int64, n > &a)
template<int n>
void	cv::v_pack_u_store (uchar *ptr, const v_reg< short, n > &a)
template<int n>
void	cv::v_pack_u_store (ushort *ptr, const v_reg< int, n > &a)

Pack and store with rounding shift
Store values from the input vector into memory with pack Values will be shifted n bits right with rounding, converted to narrower type and stored into memory. Variant with u suffix converts to unsigned type. pack: for 16-, 32- and 64-bit integer input types pack_u: for 16- and 32-bit signed integer input types Note All variants except 64-bit use saturation.
template<int shift, int n>
void	cv::v_rshr_pack_store (uchar *ptr, const v_reg< ushort, n > &a)
template<int shift, int n>
void	cv::v_rshr_pack_store (schar *ptr, const v_reg< short, n > &a)
template<int shift, int n>
void	cv::v_rshr_pack_store (ushort *ptr, const v_reg< unsigned, n > &a)
template<int shift, int n>
void	cv::v_rshr_pack_store (short *ptr, const v_reg< int, n > &a)
template<int shift, int n>
void	cv::v_rshr_pack_store (unsigned *ptr, const v_reg< uint64, n > &a)
template<int shift, int n>
void	cv::v_rshr_pack_store (int *ptr, const v_reg< int64, n > &a)
template<int shift, int n>
void	cv::v_rshr_pack_u_store (uchar *ptr, const v_reg< short, n > &a)
template<int shift, int n>
void	cv::v_rshr_pack_u_store (ushort *ptr, const v_reg< int, n > &a)

Pack boolean values
Pack boolean values from multiple vectors to one unsigned 8-bit integer vector Note Must provide valid boolean values to guarantee same result for all architectures.
template<int n>
v_reg< uchar, 2 *n >	cv::v_pack_b (const v_reg< ushort, n > &a, const v_reg< ushort, n > &b)
	! For 16-bit boolean values
template<int n>
v_reg< uchar, 4 *n >	cv::v_pack_b (const v_reg< unsigned, n > &a, const v_reg< unsigned, n > &b, const v_reg< unsigned, n > &c, const v_reg< unsigned, n > &d)
template<int n>
v_reg< uchar, 8 *n >	cv::v_pack_b (const v_reg< uint64, n > &a, const v_reg< uint64, n > &b, const v_reg< uint64, n > &c, const v_reg< uint64, n > &d, const v_reg< uint64, n > &e, const v_reg< uint64, n > &f, const v_reg< uint64, n > &g, const v_reg< uint64, n > &h)

Macro Definition Documentation

OPENCV_HAL_HAVE_PACK_STORE_BFLOAT16

#define OPENCV_HAL_HAVE_PACK_STORE_BFLOAT16   1

#include <opencv2/core/hal/intrin_cpp.hpp>

OPENCV_HAL_MATH_HAVE_EXP

#define OPENCV_HAL_MATH_HAVE_EXP   1

#include <opencv2/core/hal/intrin_cpp.hpp>

Typedef Documentation

v_float32

typedef v_float32x16 simd512::v_float32

#include <opencv2/core/hal/intrin.hpp>

Maximum available vector register capacity 32-bit floating point values (single precision)

v_float32x16

typedef v_reg<float, 16> cv::v_float32x16

#include <opencv2/core/hal/intrin_cpp.hpp>

Sixteen 32-bit floating point values (single precision)

v_float32x4

typedef v_reg<float, 4> cv::v_float32x4

#include <opencv2/core/hal/intrin_cpp.hpp>

Four 32-bit floating point values (single precision)

v_float32x8

typedef v_reg<float, 8> cv::v_float32x8

#include <opencv2/core/hal/intrin_cpp.hpp>

Eight 32-bit floating point values (single precision)

v_float64

typedef v_float64x8 simd512::v_float64

#include <opencv2/core/hal/intrin.hpp>

Maximum available vector register capacity 64-bit floating point values (double precision)

v_float64x2

typedef v_reg<double, 2> cv::v_float64x2

#include <opencv2/core/hal/intrin_cpp.hpp>

Two 64-bit floating point values (double precision)

v_float64x4

typedef v_reg<double, 4> cv::v_float64x4

#include <opencv2/core/hal/intrin_cpp.hpp>

Four 64-bit floating point values (double precision)

v_float64x8

typedef v_reg<double, 8> cv::v_float64x8

#include <opencv2/core/hal/intrin_cpp.hpp>

Eight 64-bit floating point values (double precision)

v_int16

typedef v_int16x32 simd512::v_int16

#include <opencv2/core/hal/intrin.hpp>

Maximum available vector register capacity 16-bit signed integer values.

v_int16x16

typedef v_reg<short, 16> cv::v_int16x16

#include <opencv2/core/hal/intrin_cpp.hpp>

Sixteen 16-bit signed integer values.

v_int16x32

typedef v_reg<short, 32> cv::v_int16x32

#include <opencv2/core/hal/intrin_cpp.hpp>

Thirty two 16-bit signed integer values.

v_int16x8

typedef v_reg<short, 8> cv::v_int16x8

#include <opencv2/core/hal/intrin_cpp.hpp>

Eight 16-bit signed integer values.

v_int32

typedef v_int32x16 simd512::v_int32

#include <opencv2/core/hal/intrin.hpp>

Maximum available vector register capacity 32-bit signed integer values.

v_int32x16

typedef v_reg<int, 16> cv::v_int32x16

#include <opencv2/core/hal/intrin_cpp.hpp>

Sixteen 32-bit signed integer values.

v_int32x4

typedef v_reg<int, 4> cv::v_int32x4

#include <opencv2/core/hal/intrin_cpp.hpp>

Four 32-bit signed integer values.

v_int32x8

typedef v_reg<int, 8> cv::v_int32x8

#include <opencv2/core/hal/intrin_cpp.hpp>

Eight 32-bit signed integer values.

v_int64

typedef v_int64x8 simd512::v_int64

#include <opencv2/core/hal/intrin.hpp>

Maximum available vector register capacity 64-bit signed integer values.

v_int64x2

typedef v_reg<int64, 2> cv::v_int64x2

#include <opencv2/core/hal/intrin_cpp.hpp>

Two 64-bit signed integer values.

v_int64x4

typedef v_reg<int64, 4> cv::v_int64x4

#include <opencv2/core/hal/intrin_cpp.hpp>

Four 64-bit signed integer values.

v_int64x8

typedef v_reg<int64, 8> cv::v_int64x8

#include <opencv2/core/hal/intrin_cpp.hpp>

Eight 64-bit signed integer values.

v_int8

typedef v_int8x64 simd512::v_int8

#include <opencv2/core/hal/intrin.hpp>

Maximum available vector register capacity 8-bit signed integer values.

v_int8x16

typedef v_reg<schar, 16> cv::v_int8x16

#include <opencv2/core/hal/intrin_cpp.hpp>

Sixteen 8-bit signed integer values.

v_int8x32

typedef v_reg<schar, 32> cv::v_int8x32

#include <opencv2/core/hal/intrin_cpp.hpp>

Thirty two 8-bit signed integer values.

v_int8x64

typedef v_reg<schar, 64> cv::v_int8x64

#include <opencv2/core/hal/intrin_cpp.hpp>

Sixty four 8-bit signed integer values.

v_uint16

typedef v_uint16x32 simd512::v_uint16

#include <opencv2/core/hal/intrin.hpp>

Maximum available vector register capacity 16-bit unsigned integer values.

v_uint16x16

typedef v_reg<ushort, 16> cv::v_uint16x16

#include <opencv2/core/hal/intrin_cpp.hpp>

Sixteen 16-bit unsigned integer values.

v_uint16x32

typedef v_reg<ushort, 32> cv::v_uint16x32

#include <opencv2/core/hal/intrin_cpp.hpp>

Thirty two 16-bit unsigned integer values.

v_uint16x8

typedef v_reg<ushort, 8> cv::v_uint16x8

#include <opencv2/core/hal/intrin_cpp.hpp>

Eight 16-bit unsigned integer values.

v_uint32

typedef v_uint32x16 simd512::v_uint32

#include <opencv2/core/hal/intrin.hpp>

Maximum available vector register capacity 32-bit unsigned integer values.

v_uint32x16

typedef v_reg<unsigned, 16> cv::v_uint32x16

#include <opencv2/core/hal/intrin_cpp.hpp>

Sixteen 32-bit unsigned integer values.

v_uint32x4

typedef v_reg<unsigned, 4> cv::v_uint32x4

#include <opencv2/core/hal/intrin_cpp.hpp>

Four 32-bit unsigned integer values.

v_uint32x8

typedef v_reg<unsigned, 8> cv::v_uint32x8

#include <opencv2/core/hal/intrin_cpp.hpp>

Eight 32-bit unsigned integer values.

v_uint64

typedef v_uint64x8 simd512::v_uint64

#include <opencv2/core/hal/intrin.hpp>

Maximum available vector register capacity 64-bit unsigned integer values.

v_uint64x2

typedef v_reg<uint64, 2> cv::v_uint64x2

#include <opencv2/core/hal/intrin_cpp.hpp>

Two 64-bit unsigned integer values.

v_uint64x4

typedef v_reg<uint64, 4> cv::v_uint64x4

#include <opencv2/core/hal/intrin_cpp.hpp>

Four 64-bit unsigned integer values.

v_uint64x8

typedef v_reg<uint64, 8> cv::v_uint64x8

#include <opencv2/core/hal/intrin_cpp.hpp>

Eight 64-bit unsigned integer values.

v_uint8

typedef v_uint8x64 simd512::v_uint8

#include <opencv2/core/hal/intrin.hpp>

Maximum available vector register capacity 8-bit unsigned integer values.

v_uint8x16

typedef v_reg<uchar, 16> cv::v_uint8x16

#include <opencv2/core/hal/intrin_cpp.hpp>

Sixteen 8-bit unsigned integer values.

v_uint8x32

typedef v_reg<uchar, 32> cv::v_uint8x32

#include <opencv2/core/hal/intrin_cpp.hpp>

Thirty two 8-bit unsigned integer values.

v_uint8x64

typedef v_reg<uchar, 64> cv::v_uint8x64

#include <opencv2/core/hal/intrin_cpp.hpp>

Sixty four 8-bit unsigned integer values.

Enumeration Type Documentation

anonymous enum

anonymous enum

#include <opencv2/core/hal/intrin_cpp.hpp>

Enumerator
simd128_width
simd256_width
simd512_width
simdmax_width

Function Documentation

v256_cleanup()

void cv::v256_cleanup ( )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v256_load()

template<typename _Tp >

v_reg< _Tp, simd256_width/sizeof(_Tp)> cv::v256_load ( const _Tp * ptr )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

Load 256-bit length register contents from memory.

Parameters

ptr	pointer to memory block with data

Returns

register object

Note

Returned type will be detected from passed pointer type, for example uchar ==> cv::v_uint8x32, int ==> cv::v_int32x8, etc.

Check CV_SIMD256 preprocessor definition prior to use. Use vx_load version to get maximum available register length result

Alignment requirement: if CV_STRONG_ALIGNMENT=1 then passed pointer must be aligned (sizeof(lane type) should be enough). Do not cast pointer types without runtime check for pointer alignment (like uchar* => int*).

Here is the call graph for this function:

v256_load_aligned()

template<typename _Tp >

v_reg< _Tp, simd256_width/sizeof(_Tp)> cv::v256_load_aligned ( const _Tp * ptr )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

Load register contents from memory (aligned)

similar to cv::v256_load, but source memory block should be aligned (to 32-byte boundary in case of SIMD256, 64-byte - SIMD512, etc)

Note

Check CV_SIMD256 preprocessor definition prior to use. Use vx_load_aligned version to get maximum available register length result

Here is the call graph for this function:

v256_load_expand() [1/2]

template<typename _Tp >

v_reg< typename V_TypeTraits< _Tp >::w_type, simd256_width/sizeof(typename V_TypeTraits< _Tp >::w_type)> cv::v256_load_expand ( const _Tp * ptr )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

Load register contents from memory with double expand.

Same as cv::v256_load, but result pack type will be 2x wider than memory type.

short buf[8] = {1, 2, 3, 4, 5, 6, 7, 8}; // type is int16
v_int32x8 r = v256_load_expand(buf); // r = {1, 2, 3, 4, 5, 6, 7, 8} - type is int32

For 8-, 16-, 32-bit integer source types.

Note

Check CV_SIMD256 preprocessor definition prior to use. Use vx_load_expand version to get maximum available register length result

Here is the call graph for this function:

v256_load_expand() [2/2]

v_reg< float, simd256_width/sizeof(float)> cv::v256_load_expand ( const hfloat * ptr )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v256_load_expand_q()

template<typename _Tp >

v_reg< typename V_TypeTraits< _Tp >::q_type, simd256_width/sizeof(typename V_TypeTraits< _Tp >::q_type)> cv::v256_load_expand_q ( const _Tp * ptr )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

Load register contents from memory with quad expand.

Same as cv::v256_load_expand, but result type is 4 times wider than source.

char buf[8] = {1, 2, 3, 4, 5, 6, 7, 8}; // type is int8
v_int32x8 r = v256_load_expand_q(buf); // r = {1, 2, 3, 4, 5, 6, 7, 8} - type is int32

For 8-bit integer source types.

Note

Check CV_SIMD256 preprocessor definition prior to use. Use vx_load_expand_q version to get maximum available register length result

Here is the call graph for this function:

v256_load_halves()

template<typename _Tp >

v_reg< _Tp, simd256_width/sizeof(_Tp)> cv::v256_load_halves ( const _Tp * loptr, const _Tp * hiptr )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

Load register contents from two memory blocks.

Parameters

loptr	memory block containing data for first half (0..n/2)
hiptr	memory block containing data for second half (n/2..n)

int lo[4] = { 1, 2, 3, 4 }, hi[4] = { 5, 6, 7, 8 };
v_int32x8 r = v256_load_halves(lo, hi);

Note

Check CV_SIMD256 preprocessor definition prior to use. Use vx_load_halves version to get maximum available register length result

Here is the call graph for this function:

v256_load_low()

template<typename _Tp >

v_reg< _Tp, simd256_width/sizeof(_Tp)> cv::v256_load_low ( const _Tp * ptr )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

Load 128-bits of data to lower part (high part is undefined).

Parameters

ptr	memory block containing data for first half (0..n/2)

int lo[4] = { 1, 2, 3, 4 };
v_int32x8 r = v256_load_low(lo);

Note

Check CV_SIMD256 preprocessor definition prior to use. Use vx_load_low version to get maximum available register length result

Here is the call graph for this function:

v256_setall_f32()

v_float32x8 cv::v256_setall_f32 ( float val )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v256_setall_f64()

v_float64x4 cv::v256_setall_f64 ( double val )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v256_setall_s16()

v_int16x16 cv::v256_setall_s16 ( short val )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v256_setall_s32()

v_int32x8 cv::v256_setall_s32 ( int val )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v256_setall_s64()

v_int64x4 cv::v256_setall_s64 ( int64 val )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v256_setall_s8()

v_int8x32 cv::v256_setall_s8 ( schar val )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v256_setall_u16()

v_uint16x16 cv::v256_setall_u16 ( ushort val )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v256_setall_u32()

v_uint32x8 cv::v256_setall_u32 ( unsigned val )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v256_setall_u64()

v_uint64x4 cv::v256_setall_u64 ( uint64 val )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v256_setall_u8()

v_uint8x32 cv::v256_setall_u8 ( uchar val )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v256_setzero_f32()

v_float32x8 cv::v256_setzero_f32 ( )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v256_setzero_f64()

v_float64x4 cv::v256_setzero_f64 ( )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v256_setzero_s16()

v_int16x16 cv::v256_setzero_s16 ( )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v256_setzero_s32()

v_int32x8 cv::v256_setzero_s32 ( )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v256_setzero_s64()

v_int64x4 cv::v256_setzero_s64 ( )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v256_setzero_s8()

v_int8x32 cv::v256_setzero_s8 ( )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v256_setzero_u16()

v_uint16x16 cv::v256_setzero_u16 ( )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v256_setzero_u32()

v_uint32x8 cv::v256_setzero_u32 ( )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v256_setzero_u64()

v_uint64x4 cv::v256_setzero_u64 ( )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v256_setzero_u8()

v_uint8x32 cv::v256_setzero_u8 ( )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v512_cleanup()

void cv::v512_cleanup ( )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v512_load()

template<typename _Tp >

v_reg< _Tp, simd512_width/sizeof(_Tp)> cv::v512_load ( const _Tp * ptr )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

Load 512-bit length register contents from memory.

Parameters

ptr	pointer to memory block with data

Returns

register object

Note

Returned type will be detected from passed pointer type, for example uchar ==> cv::v_uint8x64, int ==> cv::v_int32x16, etc.

Check CV_SIMD512 preprocessor definition prior to use. Use vx_load version to get maximum available register length result

Alignment requirement: if CV_STRONG_ALIGNMENT=1 then passed pointer must be aligned (sizeof(lane type) should be enough). Do not cast pointer types without runtime check for pointer alignment (like uchar* => int*).

Here is the call graph for this function:

v512_load_aligned()

template<typename _Tp >

v_reg< _Tp, simd512_width/sizeof(_Tp)> cv::v512_load_aligned ( const _Tp * ptr )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

Load register contents from memory (aligned)

similar to cv::v512_load, but source memory block should be aligned (to 64-byte boundary in case of SIMD512, etc)

Note

Check CV_SIMD512 preprocessor definition prior to use. Use vx_load_aligned version to get maximum available register length result

Here is the call graph for this function:

v512_load_expand() [1/2]

template<typename _Tp >

v_reg< typename V_TypeTraits< _Tp >::w_type, simd512_width/sizeof(typename V_TypeTraits< _Tp >::w_type)> cv::v512_load_expand ( const _Tp * ptr )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

Load register contents from memory with double expand.

Same as cv::v512_load, but result pack type will be 2x wider than memory type.

short buf[8] = {1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16}; // type is int16
v_int32x16 r = v512_load_expand(buf); // r = {1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16} - type is int32

For 8-, 16-, 32-bit integer source types.

Note

Check CV_SIMD512 preprocessor definition prior to use. Use vx_load_expand version to get maximum available register length result

Here is the call graph for this function:

v512_load_expand() [2/2]

v_reg< float, simd512_width/sizeof(float)> cv::v512_load_expand ( const hfloat * ptr )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v512_load_expand_q()

template<typename _Tp >

v_reg< typename V_TypeTraits< _Tp >::q_type, simd512_width/sizeof(typename V_TypeTraits< _Tp >::q_type)> cv::v512_load_expand_q ( const _Tp * ptr )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

Load register contents from memory with quad expand.

Same as cv::v512_load_expand, but result type is 4 times wider than source.

char buf[16] = {1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16}; // type is int8
v_int32x16 r = v512_load_expand_q(buf); // r = {1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16} - type is int32

For 8-bit integer source types.

Note

Check CV_SIMD512 preprocessor definition prior to use. Use vx_load_expand_q version to get maximum available register length result

Here is the call graph for this function:

v512_load_halves()

template<typename _Tp >

v_reg< _Tp, simd512_width/sizeof(_Tp)> cv::v512_load_halves ( const _Tp * loptr, const _Tp * hiptr )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

Load register contents from two memory blocks.

Parameters

loptr	memory block containing data for first half (0..n/2)
hiptr	memory block containing data for second half (n/2..n)

int lo[4] = { 1, 2, 3, 4, 5, 6, 7, 8 }, hi[4] = { 9, 10, 11, 12, 13, 14, 15, 16 };
v_int32x16 r = v512_load_halves(lo, hi);

Note

Check CV_SIMD512 preprocessor definition prior to use. Use vx_load_halves version to get maximum available register length result

Here is the call graph for this function:

v512_load_low()

template<typename _Tp >

v_reg< _Tp, simd512_width/sizeof(_Tp)> cv::v512_load_low ( const _Tp * ptr )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

Load 256-bits of data to lower part (high part is undefined).

Parameters

ptr	memory block containing data for first half (0..n/2)

int lo[8] = { 1, 2, 3, 4, 5, 6, 7, 8 };
v_int32x16 r = v512_load_low(lo);

Note

Check CV_SIMD512 preprocessor definition prior to use. Use vx_load_low version to get maximum available register length result

Here is the call graph for this function:

v512_setall_f32()

v_float32x16 cv::v512_setall_f32 ( float val )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v512_setall_f64()

v_float64x8 cv::v512_setall_f64 ( double val )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v512_setall_s16()

v_int16x32 cv::v512_setall_s16 ( short val )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v512_setall_s32()

v_int32x16 cv::v512_setall_s32 ( int val )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v512_setall_s64()

v_int64x8 cv::v512_setall_s64 ( int64 val )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v512_setall_s8()

v_int8x64 cv::v512_setall_s8 ( schar val )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v512_setall_u16()

v_uint16x32 cv::v512_setall_u16 ( ushort val )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v512_setall_u32()

v_uint32x16 cv::v512_setall_u32 ( unsigned val )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v512_setall_u64()

v_uint64x8 cv::v512_setall_u64 ( uint64 val )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v512_setall_u8()

v_uint8x64 cv::v512_setall_u8 ( uchar val )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v512_setzero_f32()

v_float32x16 cv::v512_setzero_f32 ( )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v512_setzero_f64()

v_float64x8 cv::v512_setzero_f64 ( )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v512_setzero_s16()

v_int16x32 cv::v512_setzero_s16 ( )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v512_setzero_s32()

v_int32x16 cv::v512_setzero_s32 ( )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v512_setzero_s64()

v_int64x8 cv::v512_setzero_s64 ( )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v512_setzero_s8()

v_int8x64 cv::v512_setzero_s8 ( )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v512_setzero_u16()

v_uint16x32 cv::v512_setzero_u16 ( )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v512_setzero_u32()

v_uint32x16 cv::v512_setzero_u32 ( )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v512_setzero_u64()

v_uint64x8 cv::v512_setzero_u64 ( )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v512_setzero_u8()

v_uint8x64 cv::v512_setzero_u8 ( )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v_abs()

template<typename _Tp , int n>

v_reg< typename V_TypeTraits< _Tp >::abs_type, n > cv::v_abs ( const v_reg< _Tp, n > & a )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

Absolute value of elements.

Only for floating point types.

v_absdiff() [1/3]

template<typename _Tp , int n>

v_reg< typename V_TypeTraits< _Tp >::abs_type, n > cv::v_absdiff ( const v_reg< _Tp, n > & a, const v_reg< _Tp, n > & b )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

Absolute difference.

Returns \( |a - b| \) converted to corresponding unsigned type. Example:

v_int32x4 a, b; // {1, 2, 3, 4} and {4, 3, 2, 1}
v_uint32x4 c = v_absdiff(a, b); // result is {3, 1, 1, 3}

For 8-, 16-, 32-bit integer source types.

v_absdiff() [2/3]

template<int n>

v_reg< double, n > cv::v_absdiff ( const v_reg< double, n > & a, const v_reg< double, n > & b )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

This is an overloaded member function, provided for convenience. It differs from the above function only in what argument(s) it accepts.

For 64-bit floating point values

v_absdiff() [3/3]

template<int n>

v_reg< float, n > cv::v_absdiff ( const v_reg< float, n > & a, const v_reg< float, n > & b )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

This is an overloaded member function, provided for convenience. It differs from the above function only in what argument(s) it accepts.

For 32-bit floating point values

v_absdiffs()

template<typename _Tp , int n>

v_reg< _Tp, n > cv::v_absdiffs ( const v_reg< _Tp, n > & a, const v_reg< _Tp, n > & b )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

Saturating absolute difference.

Returns \( saturate(|a - b|) \) . For 8-, 16-bit signed integer source types.

Here is the call graph for this function:

v_add()

template<typename _Tp , int n>

v_reg< _Tp, n > cv::v_add	(	const v_reg< _Tp, n > &	a,
		const v_reg< _Tp, n > &	b )

#include <opencv2/core/hal/intrin_cpp.hpp>

Add values.

For all types.

v_add_wrap()

template<typename _Tp , int n>

v_reg< _Tp, n > cv::v_add_wrap ( const v_reg< _Tp, n > & a, const v_reg< _Tp, n > & b )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

Add values without saturation.

For 8- and 16-bit integer values.

v_and()

template<typename _Tp , int n>

v_reg< _Tp, n > cv::v_and	(	const v_reg< _Tp, n > &	a,
		const v_reg< _Tp, n > &	b )

#include <opencv2/core/hal/intrin_cpp.hpp>

Bitwise AND.

Only for integer types.

v_broadcast_element()

template<int i, typename _Tp , int n>

v_reg< _Tp, n > cv::v_broadcast_element ( const v_reg< _Tp, n > & a )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

Broadcast i-th element of vector.

Scheme:

{ v[0] v[1] v[2] ... v[SZ] } => { v[i], v[i], v[i] ... v[i] }

Restriction: 0 <= i < nlanes Supported types: 32-bit integers and floats (s32/u32/f32)

v_ceil() [1/2]

template<int n>

v_reg< int, n *2 > cv::v_ceil ( const v_reg< double, n > & a )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

This is an overloaded member function, provided for convenience. It differs from the above function only in what argument(s) it accepts.

Here is the call graph for this function:

v_ceil() [2/2]

template<int n>

v_reg< int, n > cv::v_ceil ( const v_reg< float, n > & a )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

Ceil elements.

Ceil each value. Input type is float vector ==> output type is int vector.

Note

Only for floating point types.

Here is the call graph for this function:

v_check_all()

template<typename _Tp , int n>

bool cv::v_check_all ( const v_reg< _Tp, n > & a )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

Check if all packed values are less than zero.

Unsigned values will be casted to signed: uchar 254 => char -2.

v_check_any()

template<typename _Tp , int n>

bool cv::v_check_any ( const v_reg< _Tp, n > & a )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

Check if any of packed values is less than zero.

Unsigned values will be casted to signed: uchar 254 => char -2.

v_cleanup()

void cv::v_cleanup ( )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v_combine_high()

template<typename _Tp , int n>

v_reg< _Tp, n > cv::v_combine_high ( const v_reg< _Tp, n > & a, const v_reg< _Tp, n > & b )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

Combine vector from last elements of two vectors.

Scheme:

  {A1 A2 A3 A4}
  {B1 B2 B3 B4}
---------------
  {A3 A4 B3 B4}

For all types except 64-bit.

v_combine_low()

template<typename _Tp , int n>

v_reg< _Tp, n > cv::v_combine_low ( const v_reg< _Tp, n > & a, const v_reg< _Tp, n > & b )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

Combine vector from first elements of two vectors.

Scheme:

  {A1 A2 A3 A4}
  {B1 B2 B3 B4}
---------------
  {A1 A2 B1 B2}

For all types except 64-bit.

v_cos()

template<typename _Tp , int n>

v_reg< _Tp, n > cv::v_cos ( const v_reg< _Tp, n > & a )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

Cosine \( cos(x) \) of elements.

Only for floating point types. Core implementation the same as v_sincos.

v_cvt_f32() [1/3]

template<int n>

v_reg< float, n *2 > cv::v_cvt_f32 ( const v_reg< double, n > & a )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

Convert lower half to float.

Supported input type is cv::v_float64.

v_cvt_f32() [2/3]

template<int n>

v_reg< float, n *2 > cv::v_cvt_f32 ( const v_reg< double, n > & a, const v_reg< double, n > & b )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

Convert to float.

Supported input type is cv::v_float64.

v_cvt_f32() [3/3]

template<int n>

v_reg< float, n > cv::v_cvt_f32 ( const v_reg< int, n > & a )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

Convert to float.

Supported input type is cv::v_int32.

v_cvt_f64() [1/3]

template<int n>

v_reg< double,(n/2)> cv::v_cvt_f64

(

const v_reg< float, n > &

a

)

#include <opencv2/core/hal/intrin_cpp.hpp>

Convert lower half to double.

Supported input type is cv::v_float32.

v_cvt_f64() [2/3]

template<int n>

v_reg< double, n/2 > cv::v_cvt_f64

(

const v_reg< int, n > &

a

)

#include <opencv2/core/hal/intrin_cpp.hpp>

Convert lower half to double.

Supported input type is cv::v_int32.

v_cvt_f64() [3/3]

template<int n>

v_reg< double, n > cv::v_cvt_f64

(

const v_reg< int64, n > &

a

)

#include <opencv2/core/hal/intrin_cpp.hpp>

Convert to double.

Supported input type is cv::v_int64.

v_cvt_f64_high() [1/2]

template<int n>

v_reg< double,(n/2)> cv::v_cvt_f64_high

(

const v_reg< float, n > &

a

)

#include <opencv2/core/hal/intrin_cpp.hpp>

Convert to double high part of vector.

Supported input type is cv::v_float32.

v_cvt_f64_high() [2/2]

template<int n>

v_reg< double,(n/2)> cv::v_cvt_f64_high

(

const v_reg< int, n > &

a

)

#include <opencv2/core/hal/intrin_cpp.hpp>

Convert to double high part of vector.

Supported input type is cv::v_int32.

v_div()

template<typename _Tp , int n>

v_reg< _Tp, n > cv::v_div	(	const v_reg< _Tp, n > &	a,
		const v_reg< _Tp, n > &	b )

#include <opencv2/core/hal/intrin_cpp.hpp>

Divide values.

For floating types only.

v_dotprod() [1/2]

template<typename _Tp , int n>

v_reg< typename V_TypeTraits< _Tp >::w_type, n/2 > cv::v_dotprod ( const v_reg< _Tp, n > & a, const v_reg< _Tp, n > & b )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

Dot product of elements.

Multiply values in two registers and sum adjacent result pairs.

Scheme:

  {A1 A2 ...} // 16-bit
x {B1 B2 ...} // 16-bit
-------------
{A1B1+A2B2 ...} // 32-bit

v_dotprod() [2/2]

template<typename _Tp , int n>

v_reg< typename V_TypeTraits< _Tp >::w_type, n/2 > cv::v_dotprod ( const v_reg< _Tp, n > & a, const v_reg< _Tp, n > & b, const v_reg< typename V_TypeTraits< _Tp >::w_type, n/2 > & c )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

Dot product of elements.

Same as cv::v_dotprod, but add a third element to the sum of adjacent pairs. Scheme:

  {A1 A2 ...} // 16-bit
x {B1 B2 ...} // 16-bit
-------------
  {A1B1+A2B2+C1 ...} // 32-bit

v_dotprod_expand() [1/4]

template<typename _Tp , int n>

v_reg< typename V_TypeTraits< _Tp >::q_type, n/4 > cv::v_dotprod_expand ( const v_reg< _Tp, n > & a, const v_reg< _Tp, n > & b )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

Dot product of elements and expand.

Multiply values in two registers and expand the sum of adjacent result pairs.

Scheme:

  {A1 A2 A3 A4 ...} // 8-bit
x {B1 B2 B3 B4 ...} // 8-bit
-------------
  {A1B1+A2B2+A3B3+A4B4 ...} // 32-bit

v_dotprod_expand() [2/4]

template<typename _Tp , int n>

v_reg< typename V_TypeTraits< _Tp >::q_type, n/4 > cv::v_dotprod_expand ( const v_reg< _Tp, n > & a, const v_reg< _Tp, n > & b, const v_reg< typename V_TypeTraits< _Tp >::q_type, n/4 > & c )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

Dot product of elements.

Same as cv::v_dotprod_expand, but add a third element to the sum of adjacent pairs. Scheme:

  {A1 A2 A3 A4 ...} // 8-bit
x {B1 B2 B3 B4 ...} // 8-bit
-------------
  {A1B1+A2B2+A3B3+A4B4+C1 ...} // 32-bit

v_dotprod_expand() [3/4]

template<int n>

v_reg< double, n/2 > cv::v_dotprod_expand ( const v_reg< int, n > & a, const v_reg< int, n > & b )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

Here is the call graph for this function:

v_dotprod_expand() [4/4]

template<int n>

v_reg< double, n/2 > cv::v_dotprod_expand ( const v_reg< int, n > & a, const v_reg< int, n > & b, const v_reg< double, n/2 > & c )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

Here is the call graph for this function:

v_dotprod_expand_fast() [1/4]

template<typename _Tp , int n>

v_reg< typename V_TypeTraits< _Tp >::q_type, n/4 > cv::v_dotprod_expand_fast ( const v_reg< _Tp, n > & a, const v_reg< _Tp, n > & b )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

Fast Dot product of elements and expand.

Multiply values in two registers and expand the sum of adjacent result pairs.

Same as cv::v_dotprod_expand, but it may perform unorder sum between result pairs in some platforms, this intrinsic can be used if the sum among all lanes is only matters and also it should be yielding better performance on the affected platforms.

Here is the call graph for this function:

v_dotprod_expand_fast() [2/4]

template<typename _Tp , int n>

v_reg< typename V_TypeTraits< _Tp >::q_type, n/4 > cv::v_dotprod_expand_fast ( const v_reg< _Tp, n > & a, const v_reg< _Tp, n > & b, const v_reg< typename V_TypeTraits< _Tp >::q_type, n/4 > & c )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

Fast Dot product of elements.

Same as cv::v_dotprod_expand_fast, but add a third element to the sum of adjacent pairs.

Here is the call graph for this function:

v_dotprod_expand_fast() [3/4]

template<int n>

v_reg< double, n/2 > cv::v_dotprod_expand_fast ( const v_reg< int, n > & a, const v_reg< int, n > & b )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

Here is the call graph for this function:

v_dotprod_expand_fast() [4/4]

template<int n>

v_reg< double, n/2 > cv::v_dotprod_expand_fast ( const v_reg< int, n > & a, const v_reg< int, n > & b, const v_reg< double, n/2 > & c )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

Here is the call graph for this function:

v_dotprod_fast() [1/2]

template<typename _Tp , int n>

v_reg< typename V_TypeTraits< _Tp >::w_type, n/2 > cv::v_dotprod_fast ( const v_reg< _Tp, n > & a, const v_reg< _Tp, n > & b )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

Fast Dot product of elements.

Same as cv::v_dotprod, but it may perform unorder sum between result pairs in some platforms, this intrinsic can be used if the sum among all lanes is only matters and also it should be yielding better performance on the affected platforms.

Here is the call graph for this function:

v_dotprod_fast() [2/2]

template<typename _Tp , int n>

v_reg< typename V_TypeTraits< _Tp >::w_type, n/2 > cv::v_dotprod_fast ( const v_reg< _Tp, n > & a, const v_reg< _Tp, n > & b, const v_reg< typename V_TypeTraits< _Tp >::w_type, n/2 > & c )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

Fast Dot product of elements.

Same as cv::v_dotprod_fast, but add a third element to the sum of adjacent pairs.

Here is the call graph for this function:

v_eq()

template<typename _Tp , int n>

v_reg< _Tp, n > cv::v_eq ( const v_reg< _Tp, n > & a, const v_reg< _Tp, n > & b )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

Equal comparison.

v_erf()

template<typename _Tp , int n>

v_reg< _Tp, n > cv::v_erf ( const v_reg< _Tp, n > & a )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

Error function.

Note

Support FP32 precision for now.

v_exp()

template<typename _Tp , int n>

v_reg< _Tp, n > cv::v_exp ( const v_reg< _Tp, n > & a )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

Exponential \( e^x \) of elements.

Only for floating point types. Core implementation steps:

1. Decompose Input: Convert the input to \( 2^{x \cdot \log_2e} \) and split its exponential into integer and fractional parts: \( x \cdot \log_2e = n + f \), where \( n \) is the integer part and \( f \) is the fractional part.
2. Compute \( 2^n \): Calculated by shifting the bits.
3. Adjust Fractional Part: Compute \( f \cdot \ln2 \) to convert the fractional part to base \( e \). \( C1 \) and \( C2 \) are used to adjust the fractional part.
4. Polynomial Approximation for \( e^{f \cdot \ln2} \): The closer the fractional part is to 0, the more accurate the result.
   • For float16 and float32, use a Taylor Series with 6 terms.
   • For float64, use Pade Polynomials Approximation with 4 terms.
5. Combine Results: Multiply the two parts together to get the final result: \( e^x = 2^n \cdot e^{f \cdot \ln2} \).

Note

The precision of the calculation depends on the implementation and the data type of the input vector.

v_expand()

template<typename _Tp , int n>

void cv::v_expand ( const v_reg< _Tp, n > & a, v_reg< typename V_TypeTraits< _Tp >::w_type, n/2 > & b0, v_reg< typename V_TypeTraits< _Tp >::w_type, n/2 > & b1 )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

Expand values to the wider pack type.

Copy contents of register to two registers with 2x wider pack type. Scheme:

 int32x4     int64x2 int64x2
{A B C D} ==> {A B} , {C D}

v_expand_high()

template<typename _Tp , int n>

v_reg< typename V_TypeTraits< _Tp >::w_type, n/2 > cv::v_expand_high ( const v_reg< _Tp, n > & a )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

Expand higher values to the wider pack type.

Same as cv::v_expand_low, but expand higher half of the vector instead.

Scheme:

 int32x4     int64x2
{A B C D} ==> {C D}

v_expand_low()

template<typename _Tp , int n>

v_reg< typename V_TypeTraits< _Tp >::w_type, n/2 > cv::v_expand_low ( const v_reg< _Tp, n > & a )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

Expand lower values to the wider pack type.

Same as cv::v_expand, but return lower half of the vector.

Scheme:

 int32x4     int64x2
{A B C D} ==> {A B}

v_extract()

template<int s, typename _Tp , int n>

v_reg< _Tp, n > cv::v_extract ( const v_reg< _Tp, n > & a, const v_reg< _Tp, n > & b )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

Vector extract.

Scheme:

  {A1 A2 A3 A4}
  {B1 B2 B3 B4}
========================
shift = 1  {A2 A3 A4 B1}
shift = 2  {A3 A4 B1 B2}
shift = 3  {A4 B1 B2 B3}

Restriction: 0 <= shift < nlanes

Usage:

v_int32x4 a, b, c;
c = v_extract<2>(a, b);

For all types.

v_extract_n()

template<int s, typename _Tp , int n>

_Tp cv::v_extract_n ( const v_reg< _Tp, n > & v )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

Vector extract.

Scheme: Return the s-th element of v. Restriction: 0 <= s < nlanes

Usage:

v_int32x4 a;
int r;
r = v_extract_n<2>(a);

For all types.

v_floor() [1/2]

template<int n>

v_reg< int, n *2 > cv::v_floor ( const v_reg< double, n > & a )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

This is an overloaded member function, provided for convenience. It differs from the above function only in what argument(s) it accepts.

Here is the call graph for this function:

v_floor() [2/2]

template<int n>

v_reg< int, n > cv::v_floor ( const v_reg< float, n > & a )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

Floor elements.

Floor each value. Input type is float vector ==> output type is int vector.

Note

Only for floating point types.

Here is the call graph for this function:

v_fma()

template<typename _Tp , int n>

v_reg< _Tp, n > cv::v_fma ( const v_reg< _Tp, n > & a, const v_reg< _Tp, n > & b, const v_reg< _Tp, n > & c )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

Multiply and add.

Returns \( a*b + c \) For floating point types and signed 32bit int only.

v_ge()

template<typename _Tp , int n>

v_reg< _Tp, n > cv::v_ge ( const v_reg< _Tp, n > & a, const v_reg< _Tp, n > & b )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

Greater-than or equal comparison.

For all types except 64-bit integer values.

v_gt()

template<typename _Tp , int n>

v_reg< _Tp, n > cv::v_gt ( const v_reg< _Tp, n > & a, const v_reg< _Tp, n > & b )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

Greater-than comparison.

For all types except 64-bit integer values.

v_interleave_pairs()

template<typename _Tp , int n>

v_reg< _Tp, n > cv::v_interleave_pairs ( const v_reg< _Tp, n > & vec )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v_interleave_quads()

template<typename _Tp , int n>

v_reg< _Tp, n > cv::v_interleave_quads ( const v_reg< _Tp, n > & vec )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v_invsqrt()

template<typename _Tp , int n>

v_reg< _Tp, n > cv::v_invsqrt ( const v_reg< _Tp, n > & a )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

Inversed square root.

Returns \( 1/sqrt(a) \) For floating point types only.

v_le()

template<typename _Tp , int n>

v_reg< _Tp, n > cv::v_le ( const v_reg< _Tp, n > & a, const v_reg< _Tp, n > & b )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

Less-than or equal comparison.

For all types except 64-bit integer values.

v_load()

template<typename _Tp >

v_reg< _Tp, simd128_width/sizeof(_Tp)> cv::v_load ( const _Tp * ptr )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

Load register contents from memory.

Parameters

ptr	pointer to memory block with data

Returns

register object

Note

Returned type will be detected from passed pointer type, for example uchar ==> cv::v_uint8x16, int ==> cv::v_int32x4, etc.

Use vx_load version to get maximum available register length result

Alignment requirement: if CV_STRONG_ALIGNMENT=1 then passed pointer must be aligned (sizeof(lane type) should be enough). Do not cast pointer types without runtime check for pointer alignment (like uchar* => int*).

Here is the call graph for this function:

v_load_aligned()

template<typename _Tp >

v_reg< _Tp, simd128_width/sizeof(_Tp)> cv::v_load_aligned ( const _Tp * ptr )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

Load register contents from memory (aligned)

similar to cv::v_load, but source memory block should be aligned (to 16-byte boundary in case of SIMD128, 32-byte - SIMD256, etc)

Note

Use vx_load_aligned version to get maximum available register length result

Here is the call graph for this function:

v_load_deinterleave() [1/3]

template<typename _Tp , int n>

void cv::v_load_deinterleave ( const _Tp * ptr, v_reg< _Tp, n > & a, v_reg< _Tp, n > & b )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

Load and deinterleave (2 channels)

Load data from memory deinterleave and store to 2 registers. Scheme:

{A1 B1 A2 B2 ...} ==> {A1 A2 ...}, {B1 B2 ...}

For all types except 64-bit.

Here is the call graph for this function:

v_load_deinterleave() [2/3]

template<typename _Tp , int n>

void cv::v_load_deinterleave ( const _Tp * ptr, v_reg< _Tp, n > & a, v_reg< _Tp, n > & b, v_reg< _Tp, n > & c )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

Load and deinterleave (3 channels)

Load data from memory deinterleave and store to 3 registers. Scheme:

{A1 B1 C1 A2 B2 C2 ...} ==> {A1 A2 ...}, {B1 B2 ...}, {C1 C2 ...}

For all types except 64-bit.

Here is the call graph for this function:

v_load_deinterleave() [3/3]

template<typename _Tp , int n>

void cv::v_load_deinterleave ( const _Tp * ptr, v_reg< _Tp, n > & a, v_reg< _Tp, n > & b, v_reg< _Tp, n > & c, v_reg< _Tp, n > & d )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

Load and deinterleave (4 channels)

Load data from memory deinterleave and store to 4 registers. Scheme:

{A1 B1 C1 D1 A2 B2 C2 D2 ...} ==> {A1 A2 ...}, {B1 B2 ...}, {C1 C2 ...}, {D1 D2 ...}

For all types except 64-bit.

Here is the call graph for this function:

v_load_expand() [1/2]

template<typename _Tp >

v_reg< typename V_TypeTraits< _Tp >::w_type, simd128_width/sizeof(typename V_TypeTraits< _Tp >::w_type)> cv::v_load_expand ( const _Tp * ptr )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

Load register contents from memory with double expand.

Same as cv::v_load, but result pack type will be 2x wider than memory type.

short buf[4] = {1, 2, 3, 4}; // type is int16
v_int32x4 r = v_load_expand(buf); // r = {1, 2, 3, 4} - type is int32

For 8-, 16-, 32-bit integer source types.

Note

Use vx_load_expand version to get maximum available register length result

Here is the call graph for this function:

v_load_expand() [2/2]

v_reg< float, simd128_width/sizeof(float)> cv::v_load_expand ( const hfloat * ptr )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v_load_expand_q()

template<typename _Tp >

v_reg< typename V_TypeTraits< _Tp >::q_type, simd128_width/sizeof(typename V_TypeTraits< _Tp >::q_type)> cv::v_load_expand_q ( const _Tp * ptr )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

Load register contents from memory with quad expand.

Same as cv::v_load_expand, but result type is 4 times wider than source.

char buf[4] = {1, 2, 3, 4}; // type is int8
v_int32x4 r = v_load_expand_q(buf); // r = {1, 2, 3, 4} - type is int32

For 8-bit integer source types.

Note

Use vx_load_expand_q version to get maximum available register length result

Here is the call graph for this function:

v_load_halves()

template<typename _Tp >

v_reg< _Tp, simd128_width/sizeof(_Tp)> cv::v_load_halves ( const _Tp * loptr, const _Tp * hiptr )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

Load register contents from two memory blocks.

Parameters

loptr	memory block containing data for first half (0..n/2)
hiptr	memory block containing data for second half (n/2..n)

int lo[2] = { 1, 2 }, hi[2] = { 3, 4 };
v_int32x4 r = v_load_halves(lo, hi);

Note

Use vx_load_halves version to get maximum available register length result

Here is the call graph for this function:

v_load_low()

template<typename _Tp >

v_reg< _Tp, simd128_width/sizeof(_Tp)> cv::v_load_low ( const _Tp * ptr )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

Load 64-bits of data to lower part (high part is undefined).

Parameters

ptr	memory block containing data for first half (0..n/2)

int lo[2] = { 1, 2 };
v_int32x4 r = v_load_low(lo);

Note

Use vx_load_low version to get maximum available register length result

Here is the call graph for this function:

v_log()

template<typename _Tp , int n>

v_reg< _Tp, n > cv::v_log ( const v_reg< _Tp, n > & a )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

Natural logarithm \( \log(x) \) of elements.

Only for floating point types. Core implementation steps:

1. Decompose Input: Use binary representation to decompose the input into mantissa part \( m \) and exponent part \( e \). Such that \( \log(x) = \log(m \cdot 2^e) = \log(m) + e \cdot \ln(2) \).
2. Adjust Mantissa and Exponent Parts: If the mantissa is less than \( \sqrt{0.5} \), adjust the exponent and mantissa to ensure the mantissa is in the range \( (\sqrt{0.5}, \sqrt{2}) \) for better approximation.
3. Polynomial Approximation for \( \log(m) \): The closer the \( m \) is to 1, the more accurate the result.
   • For float16 and float32, use a Taylor Series with 9 terms.
   • For float64, use Pade Polynomials Approximation with 6 terms.
4. Combine Results: Add the two parts together to get the final result.

Note

The precision of the calculation depends on the implementation and the data type of the input.

Similar to the behavior of std::log(), \( \ln(0) = -\infty \).

v_lt()

template<typename _Tp , int n>

v_reg< _Tp, n > cv::v_lt ( const v_reg< _Tp, n > & a, const v_reg< _Tp, n > & b )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

Less-than comparison.

For all types except 64-bit integer values.

v_lut() [1/5]

template<typename _Tp >

v_reg< _Tp, simd128_width/sizeof(_Tp)> cv::v_lut ( const _Tp * tab, const int * idx )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v_lut() [2/5]

template<int n>

v_reg< double, n/2 > cv::v_lut ( const double * tab, const v_reg< int, n > & idx )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v_lut() [3/5]

template<int n>

v_reg< float, n > cv::v_lut ( const float * tab, const v_reg< int, n > & idx )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v_lut() [4/5]

template<int n>

v_reg< int, n > cv::v_lut ( const int * tab, const v_reg< int, n > & idx )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v_lut() [5/5]

template<int n>

v_reg< unsigned, n > cv::v_lut ( const unsigned * tab, const v_reg< int, n > & idx )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v_lut_deinterleave() [1/2]

template<int n>

void cv::v_lut_deinterleave ( const double * tab, const v_reg< int, n *2 > & idx, v_reg< double, n > & x, v_reg< double, n > & y )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v_lut_deinterleave() [2/2]

template<int n>

void cv::v_lut_deinterleave ( const float * tab, const v_reg< int, n > & idx, v_reg< float, n > & x, v_reg< float, n > & y )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v_lut_pairs()

template<typename _Tp >

v_reg< _Tp, simd128_width/sizeof(_Tp)> cv::v_lut_pairs ( const _Tp * tab, const int * idx )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v_lut_quads()

template<typename _Tp >

v_reg< _Tp, simd128_width/sizeof(_Tp)> cv::v_lut_quads ( const _Tp * tab, const int * idx )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v_magnitude()

template<typename _Tp , int n>

v_reg< _Tp, n > cv::v_magnitude ( const v_reg< _Tp, n > & a, const v_reg< _Tp, n > & b )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

Magnitude.

Returns \( sqrt(a^2 + b^2) \) For floating point types only.

v_matmul()

template<int n>

v_reg< float, n > cv::v_matmul ( const v_reg< float, n > & v, const v_reg< float, n > & a, const v_reg< float, n > & b, const v_reg< float, n > & c, const v_reg< float, n > & d )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

Matrix multiplication.

Scheme:

{A0 A1 A2 A3}   |V0|
{B0 B1 B2 B3}   |V1|
{C0 C1 C2 C3}   |V2|
{D0 D1 D2 D3} x |V3|
====================
{R0 R1 R2 R3}, where:
R0 = A0V0 + B0V1 + C0V2 + D0V3,
R1 = A1V0 + B1V1 + C1V2 + D1V3
...

v_matmuladd()

template<int n>

v_reg< float, n > cv::v_matmuladd ( const v_reg< float, n > & v, const v_reg< float, n > & a, const v_reg< float, n > & b, const v_reg< float, n > & c, const v_reg< float, n > & d )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

Matrix multiplication and add.

Scheme:

{A0 A1 A2 A3}   |V0|   |D0|
{B0 B1 B2 B3}   |V1|   |D1|
{C0 C1 C2 C3} x |V2| + |D2|
====================   |D3|
{R0 R1 R2 R3}, where:
R0 = A0V0 + B0V1 + C0V2 + D0,
R1 = A1V0 + B1V1 + C1V2 + D1
...

v_max()

template<typename _Tp , int n>

v_reg< _Tp, n > cv::v_max ( const v_reg< _Tp, n > & a, const v_reg< _Tp, n > & b )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

Choose max values for each pair.

Scheme:

{A1 A2 ...}
{B1 B2 ...}
--------------
{max(A1,B1) max(A2,B2) ...}

For all types except 64-bit integer.

v_min()

template<typename _Tp , int n>

v_reg< _Tp, n > cv::v_min ( const v_reg< _Tp, n > & a, const v_reg< _Tp, n > & b )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

Choose min values for each pair.

Scheme:

{A1 A2 ...}
{B1 B2 ...}
--------------
{min(A1,B1) min(A2,B2) ...}

For all types except 64-bit integer.

v_mul()

template<typename _Tp , int n>

v_reg< _Tp, n > cv::v_mul	(	const v_reg< _Tp, n > &	a,
		const v_reg< _Tp, n > &	b )

#include <opencv2/core/hal/intrin_cpp.hpp>

Multiply values.

For 16- and 32-bit integer types and floating types.

v_mul_expand()

template<typename _Tp , int n>

void cv::v_mul_expand ( const v_reg< _Tp, n > & a, const v_reg< _Tp, n > & b, v_reg< typename V_TypeTraits< _Tp >::w_type, n/2 > & c, v_reg< typename V_TypeTraits< _Tp >::w_type, n/2 > & d )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

Multiply and expand.

Multiply values two registers and store results in two registers with wider pack type. Scheme:

  {A B C D} // 32-bit
x {E F G H} // 32-bit
---------------
{AE BF}         // 64-bit
        {CG DH} // 64-bit

Example:

v_uint32x4 a, b; // {1,2,3,4} and {2,2,2,2}
v_uint64x2 c, d; // results
v_mul_expand(a, b, c, d); // c, d = {2,4}, {6, 8}

Implemented only for 16- and unsigned 32-bit source types (v_int16x8, v_uint16x8, v_uint32x4).

v_mul_hi()

template<typename _Tp , int n>

v_reg< _Tp, n > cv::v_mul_hi ( const v_reg< _Tp, n > & a, const v_reg< _Tp, n > & b )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

Multiply and extract high part.

Multiply values two registers and store high part of the results. Implemented only for 16-bit source types (v_int16x8, v_uint16x8). Returns \( a*b >> 16 \)

v_mul_wrap()

template<typename _Tp , int n>

v_reg< _Tp, n > cv::v_mul_wrap ( const v_reg< _Tp, n > & a, const v_reg< _Tp, n > & b )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

Multiply values without saturation.

For 8- and 16-bit integer values.

v_muladd()

template<typename _Tp , int n>

v_reg< _Tp, n > cv::v_muladd ( const v_reg< _Tp, n > & a, const v_reg< _Tp, n > & b, const v_reg< _Tp, n > & c )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

A synonym for v_fma.

Here is the call graph for this function:

v_ne()

template<typename _Tp , int n>

v_reg< _Tp, n > cv::v_ne ( const v_reg< _Tp, n > & a, const v_reg< _Tp, n > & b )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

Not equal comparison.

v_not()

template<typename _Tp , int n>

v_reg< _Tp, n > cv::v_not

(

const v_reg< _Tp, n > &

a

)

#include <opencv2/core/hal/intrin_cpp.hpp>

Bitwise NOT.

Only for integer types.

v_not_nan() [1/2]

template<int n>

v_reg< double, n > cv::v_not_nan ( const v_reg< double, n > & a )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v_not_nan() [2/2]

template<int n>

v_reg< float, n > cv::v_not_nan ( const v_reg< float, n > & a )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v_or()

template<typename _Tp , int n>

v_reg< _Tp, n > cv::v_or	(	const v_reg< _Tp, n > &	a,
		const v_reg< _Tp, n > &	b )

#include <opencv2/core/hal/intrin_cpp.hpp>

Bitwise OR.

Only for integer types.

v_pack() [1/6]

template<int n>

v_reg< short, 2 *n > cv::v_pack ( const v_reg< int, n > & a, const v_reg< int, n > & b )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v_pack() [2/6]

template<int n>

v_reg< int, 2 *n > cv::v_pack ( const v_reg< int64, n > & a, const v_reg< int64, n > & b )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v_pack() [3/6]

template<int n>

v_reg< schar, 2 *n > cv::v_pack ( const v_reg< short, n > & a, const v_reg< short, n > & b )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v_pack() [4/6]

template<int n>

v_reg< unsigned, 2 *n > cv::v_pack ( const v_reg< uint64, n > & a, const v_reg< uint64, n > & b )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v_pack() [5/6]

template<int n>

v_reg< ushort, 2 *n > cv::v_pack ( const v_reg< unsigned, n > & a, const v_reg< unsigned, n > & b )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v_pack() [6/6]

template<int n>

v_reg< uchar, 2 *n > cv::v_pack ( const v_reg< ushort, n > & a, const v_reg< ushort, n > & b )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v_pack_b() [1/3]

template<int n>

v_reg< uchar, 8 *n > cv::v_pack_b ( const v_reg< uint64, n > & a, const v_reg< uint64, n > & b, const v_reg< uint64, n > & c, const v_reg< uint64, n > & d, const v_reg< uint64, n > & e, const v_reg< uint64, n > & f, const v_reg< uint64, n > & g, const v_reg< uint64, n > & h )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

This is an overloaded member function, provided for convenience. It differs from the above function only in what argument(s) it accepts. For 64-bit boolean values

Scheme:

a  {0xFFFF.. 0}
b  {0 0xFFFF..}
c  {0xFFFF.. 0}
d  {0 0xFFFF..}
 
e  {0xFFFF.. 0}
f  {0xFFFF.. 0}
g  {0 0xFFFF..}
h  {0 0xFFFF..}
===============
{
   0xFF 0 0 0xFF 0xFF 0 0 0xFF
   0xFF 0 0xFF 0 0 0xFF 0 0xFF
}

v_pack_b() [2/3]

template<int n>

v_reg< uchar, 4 *n > cv::v_pack_b ( const v_reg< unsigned, n > & a, const v_reg< unsigned, n > & b, const v_reg< unsigned, n > & c, const v_reg< unsigned, n > & d )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

This is an overloaded member function, provided for convenience. It differs from the above function only in what argument(s) it accepts. For 32-bit boolean values

Scheme:

a  {0xFFFF.. 0 0 0xFFFF..}
b  {0 0xFFFF.. 0xFFFF.. 0}
c  {0xFFFF.. 0 0xFFFF.. 0}
d  {0 0xFFFF.. 0 0xFFFF..}
===============
{
   0xFF 0 0 0xFF 0 0xFF 0xFF 0
   0xFF 0 0xFF 0 0 0xFF 0 0xFF
}

v_pack_b() [3/3]

template<int n>

v_reg< uchar, 2 *n > cv::v_pack_b ( const v_reg< ushort, n > & a, const v_reg< ushort, n > & b )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

! For 16-bit boolean values

Scheme:

a  {0xFFFF 0 0 0xFFFF 0 0xFFFF 0xFFFF 0}
b  {0xFFFF 0 0xFFFF 0 0 0xFFFF 0 0xFFFF}
===============
{
   0xFF 0 0 0xFF 0 0xFF 0xFF 0
   0xFF 0 0xFF 0 0 0xFF 0 0xFF
}

v_pack_store() [1/7]

template<int n>

void cv::v_pack_store ( hfloat * ptr, const v_reg< float, n > & v )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v_pack_store() [2/7]

template<int n>

void cv::v_pack_store ( int * ptr, const v_reg< int64, n > & a )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v_pack_store() [3/7]

template<int n>

void cv::v_pack_store ( schar * ptr, const v_reg< short, n > & a )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v_pack_store() [4/7]

template<int n>

void cv::v_pack_store ( short * ptr, const v_reg< int, n > & a )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v_pack_store() [5/7]

template<int n>

void cv::v_pack_store ( uchar * ptr, const v_reg< ushort, n > & a )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v_pack_store() [6/7]

template<int n>

void cv::v_pack_store ( unsigned * ptr, const v_reg< uint64, n > & a )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v_pack_store() [7/7]

template<int n>

void cv::v_pack_store ( ushort * ptr, const v_reg< unsigned, n > & a )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v_pack_triplets()

template<typename _Tp , int n>

v_reg< _Tp, n > cv::v_pack_triplets ( const v_reg< _Tp, n > & vec )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v_pack_u() [1/2]

template<int n>

v_reg< ushort, 2 *n > cv::v_pack_u ( const v_reg< int, n > & a, const v_reg< int, n > & b )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v_pack_u() [2/2]

template<int n>

v_reg< uchar, 2 *n > cv::v_pack_u ( const v_reg< short, n > & a, const v_reg< short, n > & b )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v_pack_u_store() [1/2]

template<int n>

void cv::v_pack_u_store ( uchar * ptr, const v_reg< short, n > & a )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v_pack_u_store() [2/2]

template<int n>

void cv::v_pack_u_store ( ushort * ptr, const v_reg< int, n > & a )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v_popcount()

template<typename _Tp , int n>

v_reg< typename V_TypeTraits< _Tp >::abs_type, n > cv::v_popcount ( const v_reg< _Tp, n > & a )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

Count the 1 bits in the vector lanes and return result as corresponding unsigned type.

Scheme:

{A1 A2 A3 ...} => {popcount(A1), popcount(A2), popcount(A3), ...}

For all integer types.

Here is the call graph for this function:

v_recombine()

template<typename _Tp , int n>

void cv::v_recombine ( const v_reg< _Tp, n > & a, const v_reg< _Tp, n > & b, v_reg< _Tp, n > & low, v_reg< _Tp, n > & high )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

Combine two vectors from lower and higher parts of two other vectors.

low = cv::v_combine_low(a, b);
high = cv::v_combine_high(a, b);

v_reduce_max()

template<typename _Tp , int n>

_Tp cv::v_reduce_max ( const v_reg< _Tp, n > & a )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

Find one max value.

Scheme:

{A1 A2 A3 ...} => max(A1,A2,A3,...)

For all types except 64-bit integer and 64-bit floating point types.

v_reduce_min()

template<typename _Tp , int n>

_Tp cv::v_reduce_min ( const v_reg< _Tp, n > & a )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

Find one min value.

Scheme:

{A1 A2 A3 ...} => min(A1,A2,A3,...)

For all types except 64-bit integer and 64-bit floating point types.

v_reduce_sad()

template<typename _Tp , int n>

V_TypeTraits< typenameV_TypeTraits< _Tp >::abs_type >::sum_type cv::v_reduce_sad ( const v_reg< _Tp, n > & a, const v_reg< _Tp, n > & b )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

Sum absolute differences of values.

Scheme:

{A1 A2 A3 ...} {B1 B2 B3 ...} => sum{ABS(A1-B1),abs(A2-B2),abs(A3-B3),...}

For all types except 64-bit types.

v_reduce_sum()

template<typename _Tp , int n>

V_TypeTraits< _Tp >::sum_type cv::v_reduce_sum ( const v_reg< _Tp, n > & a )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

Sum packed values.

Scheme:

{A1 A2 A3 ...} => sum{A1,A2,A3,...}

v_reduce_sum4()

template<int n>

v_reg< float, n > cv::v_reduce_sum4 ( const v_reg< float, n > & a, const v_reg< float, n > & b, const v_reg< float, n > & c, const v_reg< float, n > & d )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

Sums all elements of each input vector, returns the vector of sums.

Scheme:

result[0] = a[0] + a[1] + a[2] + a[3]
result[1] = b[0] + b[1] + b[2] + b[3]
result[2] = c[0] + c[1] + c[2] + c[3]
result[3] = d[0] + d[1] + d[2] + d[3]

v_reinterpret_as_f32()

template<typename _Tp0 , int n0>

v_reg< float, n0 *sizeof(_Tp0)/sizeof(float)> cv::v_reinterpret_as_f32 ( const v_reg< _Tp0, n0 > & a )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v_reinterpret_as_f64()

template<typename _Tp0 , int n0>

v_reg< double, n0 *sizeof(_Tp0)/sizeof(double)> cv::v_reinterpret_as_f64 ( const v_reg< _Tp0, n0 > & a )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v_reinterpret_as_s16()

template<typename _Tp0 , int n0>

v_reg< short, n0 *sizeof(_Tp0)/sizeof(short)> cv::v_reinterpret_as_s16 ( const v_reg< _Tp0, n0 > & a )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v_reinterpret_as_s32()

template<typename _Tp0 , int n0>

v_reg< int, n0 *sizeof(_Tp0)/sizeof(int)> cv::v_reinterpret_as_s32 ( const v_reg< _Tp0, n0 > & a )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v_reinterpret_as_s64()

template<typename _Tp0 , int n0>

v_reg< int64, n0 *sizeof(_Tp0)/sizeof(int64)> cv::v_reinterpret_as_s64 ( const v_reg< _Tp0, n0 > & a )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v_reinterpret_as_s8()

template<typename _Tp0 , int n0>

v_reg< schar, n0 *sizeof(_Tp0)/sizeof(schar)> cv::v_reinterpret_as_s8 ( const v_reg< _Tp0, n0 > & a )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v_reinterpret_as_u16()

template<typename _Tp0 , int n0>

v_reg< ushort, n0 *sizeof(_Tp0)/sizeof(ushort)> cv::v_reinterpret_as_u16 ( const v_reg< _Tp0, n0 > & a )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v_reinterpret_as_u32()

template<typename _Tp0 , int n0>

v_reg< unsigned, n0 *sizeof(_Tp0)/sizeof(unsigned)> cv::v_reinterpret_as_u32 ( const v_reg< _Tp0, n0 > & a )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v_reinterpret_as_u64()

template<typename _Tp0 , int n0>

v_reg< uint64, n0 *sizeof(_Tp0)/sizeof(uint64)> cv::v_reinterpret_as_u64 ( const v_reg< _Tp0, n0 > & a )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v_reinterpret_as_u8()

template<typename _Tp0 , int n0>

v_reg< uchar, n0 *sizeof(_Tp0)/sizeof(uchar)> cv::v_reinterpret_as_u8 ( const v_reg< _Tp0, n0 > & a )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v_reverse()

template<typename _Tp , int n>

v_reg< _Tp, n > cv::v_reverse ( const v_reg< _Tp, n > & a )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

Vector reverse order.

Reverse the order of the vector Scheme:

REG {A1 ... An} ==> REG {An ... A1}

For all types.

v_rotate_left() [1/2]

template<int imm, typename _Tp , int n>

v_reg< _Tp, n > cv::v_rotate_left ( const v_reg< _Tp, n > & a )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

Element shift left among vector.

For all type

v_rotate_left() [2/2]

template<int imm, typename _Tp , int n>

v_reg< _Tp, n > cv::v_rotate_left ( const v_reg< _Tp, n > & a, const v_reg< _Tp, n > & b )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v_rotate_right() [1/2]

template<int imm, typename _Tp , int n>

v_reg< _Tp, n > cv::v_rotate_right ( const v_reg< _Tp, n > & a )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

Element shift right among vector.

For all type

v_rotate_right() [2/2]

template<int imm, typename _Tp , int n>

v_reg< _Tp, n > cv::v_rotate_right ( const v_reg< _Tp, n > & a, const v_reg< _Tp, n > & b )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v_round() [1/3]

template<int n>

v_reg< int, n *2 > cv::v_round ( const v_reg< double, n > & a )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

This is an overloaded member function, provided for convenience. It differs from the above function only in what argument(s) it accepts.

Here is the call graph for this function:

v_round() [2/3]

template<int n>

v_reg< int, n *2 > cv::v_round ( const v_reg< double, n > & a, const v_reg< double, n > & b )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

This is an overloaded member function, provided for convenience. It differs from the above function only in what argument(s) it accepts.

Here is the call graph for this function:

v_round() [3/3]

template<int n>

v_reg< int, n > cv::v_round ( const v_reg< float, n > & a )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

Round elements.

Rounds each value. Input type is float vector ==> output type is int vector.

Note

Only for floating point types.

Here is the call graph for this function:

v_rshr() [1/6]

template<int shift, int n>

v_reg< int, n > cv::v_rshr ( const v_reg< int, n > & a )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v_rshr() [2/6]

template<int shift, int n>

v_reg< int64, n > cv::v_rshr ( const v_reg< int64, n > & a )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v_rshr() [3/6]

template<int shift, int n>

v_reg< short, n > cv::v_rshr ( const v_reg< short, n > & a )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v_rshr() [4/6]

template<int shift, int n>

v_reg< uint64, n > cv::v_rshr ( const v_reg< uint64, n > & a )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v_rshr() [5/6]

template<int shift, int n>

v_reg< unsigned, n > cv::v_rshr ( const v_reg< unsigned, n > & a )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v_rshr() [6/6]

template<int shift, int n>

v_reg< ushort, n > cv::v_rshr ( const v_reg< ushort, n > & a )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v_rshr_pack() [1/6]

template<int shift, int n>

v_reg< short, 2 *n > cv::v_rshr_pack ( const v_reg< int, n > & a, const v_reg< int, n > & b )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v_rshr_pack() [2/6]

template<int shift, int n>

v_reg< int, 2 *n > cv::v_rshr_pack ( const v_reg< int64, n > & a, const v_reg< int64, n > & b )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v_rshr_pack() [3/6]

template<int shift, int n>

v_reg< schar, 2 *n > cv::v_rshr_pack ( const v_reg< short, n > & a, const v_reg< short, n > & b )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v_rshr_pack() [4/6]

template<int shift, int n>

v_reg< unsigned, 2 *n > cv::v_rshr_pack ( const v_reg< uint64, n > & a, const v_reg< uint64, n > & b )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v_rshr_pack() [5/6]

template<int shift, int n>

v_reg< ushort, 2 *n > cv::v_rshr_pack ( const v_reg< unsigned, n > & a, const v_reg< unsigned, n > & b )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v_rshr_pack() [6/6]

template<int shift, int n>

v_reg< uchar, 2 *n > cv::v_rshr_pack ( const v_reg< ushort, n > & a, const v_reg< ushort, n > & b )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v_rshr_pack_store() [1/6]

template<int shift, int n>

void cv::v_rshr_pack_store ( int * ptr, const v_reg< int64, n > & a )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v_rshr_pack_store() [2/6]

template<int shift, int n>

void cv::v_rshr_pack_store ( schar * ptr, const v_reg< short, n > & a )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v_rshr_pack_store() [3/6]

template<int shift, int n>

void cv::v_rshr_pack_store ( short * ptr, const v_reg< int, n > & a )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v_rshr_pack_store() [4/6]

template<int shift, int n>

void cv::v_rshr_pack_store ( uchar * ptr, const v_reg< ushort, n > & a )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v_rshr_pack_store() [5/6]

template<int shift, int n>

void cv::v_rshr_pack_store ( unsigned * ptr, const v_reg< uint64, n > & a )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v_rshr_pack_store() [6/6]

template<int shift, int n>

void cv::v_rshr_pack_store ( ushort * ptr, const v_reg< unsigned, n > & a )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v_rshr_pack_u() [1/2]

template<int shift, int n>

v_reg< ushort, 2 *n > cv::v_rshr_pack_u ( const v_reg< int, n > & a, const v_reg< int, n > & b )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v_rshr_pack_u() [2/2]

template<int shift, int n>

v_reg< uchar, 2 *n > cv::v_rshr_pack_u ( const v_reg< short, n > & a, const v_reg< short, n > & b )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v_rshr_pack_u_store() [1/2]

template<int shift, int n>

void cv::v_rshr_pack_u_store ( uchar * ptr, const v_reg< short, n > & a )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v_rshr_pack_u_store() [2/2]

template<int shift, int n>

void cv::v_rshr_pack_u_store ( ushort * ptr, const v_reg< int, n > & a )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v_scan_forward()

template<typename _Tp , int n>

int cv::v_scan_forward ( const v_reg< _Tp, n > & a )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

Get first negative lane index.

Returned value is an index of first negative lane (undefined for input of all positive values) Example:

v_int32x4 r; // set to {0, 0, -1, -1}
int idx = v_heading_zeros(r); // idx = 2

v_select()

template<typename _Tp , int n>

v_reg< _Tp, n > cv::v_select ( const v_reg< _Tp, n > & mask, const v_reg< _Tp, n > & a, const v_reg< _Tp, n > & b )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

Per-element select (blend operation)

Return value will be built by combining values a and b using the following scheme: result[i] = mask[i] ? a[i] : b[i];

Note

: mask element values are restricted to these values:

• 0: select element from b
• 0xff/0xffff/etc: select element from a (fully compatible with bitwise-based operator)

v_setall_() [1/10]

template<>

v_float64x2 cv::v_setall_ ( double val )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v_setall_() [2/10]

template<>

v_float32x4 cv::v_setall_ ( float val )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v_setall_() [3/10]

template<>

v_int32x4 cv::v_setall_ ( int val )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v_setall_() [4/10]

template<>

v_int64x2 cv::v_setall_ ( int64 val )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v_setall_() [5/10]

template<>

v_int8x16 cv::v_setall_ ( schar val )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v_setall_() [6/10]

template<>

v_int16x8 cv::v_setall_ ( short val )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v_setall_() [7/10]

template<>

v_uint8x16 cv::v_setall_ ( uchar val )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v_setall_() [8/10]

template<>

v_uint64x2 cv::v_setall_ ( uint64 val )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v_setall_() [9/10]

template<>

v_uint32x4 cv::v_setall_ ( unsigned val )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v_setall_() [10/10]

template<>

v_uint16x8 cv::v_setall_ ( ushort val )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v_setall_f32()

v_float32x4 cv::v_setall_f32 ( float val )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v_setall_f64()

v_float64x2 cv::v_setall_f64 ( double val )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v_setall_s16()

v_int16x8 cv::v_setall_s16 ( short val )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v_setall_s32()

v_int32x4 cv::v_setall_s32 ( int val )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v_setall_s64()

v_int64x2 cv::v_setall_s64 ( int64 val )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v_setall_s8()

v_int8x16 cv::v_setall_s8 ( schar val )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v_setall_u16()

v_uint16x8 cv::v_setall_u16 ( ushort val )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v_setall_u32()

v_uint32x4 cv::v_setall_u32 ( unsigned val )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v_setall_u64()

v_uint64x2 cv::v_setall_u64 ( uint64 val )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v_setall_u8()

v_uint8x16 cv::v_setall_u8 ( uchar val )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v_setzero_()

template<>

v_uint8x16 cv::v_setzero_ ( )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v_setzero_f32()

v_float32x4 cv::v_setzero_f32 ( )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v_setzero_f64()

v_float64x2 cv::v_setzero_f64 ( )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v_setzero_s16()

v_int16x8 cv::v_setzero_s16 ( )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v_setzero_s32()

v_int32x4 cv::v_setzero_s32 ( )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v_setzero_s64()

v_int64x2 cv::v_setzero_s64 ( )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v_setzero_s8()

v_int8x16 cv::v_setzero_s8 ( )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v_setzero_u16()

v_uint16x8 cv::v_setzero_u16 ( )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v_setzero_u32()

v_uint32x4 cv::v_setzero_u32 ( )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v_setzero_u64()

v_uint64x2 cv::v_setzero_u64 ( )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v_setzero_u8()

v_uint8x16 cv::v_setzero_u8 ( )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v_shl() [1/7]

template<typename _Tp , int n>

v_reg< _Tp, n > cv::v_shl ( const v_reg< _Tp, n > & a, int imm )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

Bitwise shift left.

For 16-, 32- and 64-bit integer values.

v_shl() [2/7]

template<int shift, int n>

v_reg< int, n > cv::v_shl ( const v_reg< int, n > & a )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v_shl() [3/7]

template<int shift, int n>

v_reg< int64, n > cv::v_shl ( const v_reg< int64, n > & a )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v_shl() [4/7]

template<int shift, int n>

v_reg< short, n > cv::v_shl ( const v_reg< short, n > & a )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v_shl() [5/7]

template<int shift, int n>

v_reg< uint64, n > cv::v_shl ( const v_reg< uint64, n > & a )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v_shl() [6/7]

template<int shift, int n>

v_reg< unsigned, n > cv::v_shl ( const v_reg< unsigned, n > & a )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v_shl() [7/7]

template<int shift, int n>

v_reg< ushort, n > cv::v_shl ( const v_reg< ushort, n > & a )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v_shr() [1/7]

template<typename _Tp , int n>

v_reg< _Tp, n > cv::v_shr ( const v_reg< _Tp, n > & a, int imm )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

Bitwise shift right.

For 16-, 32- and 64-bit integer values.

v_shr() [2/7]

template<int shift, int n>

v_reg< int, n > cv::v_shr ( const v_reg< int, n > & a )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v_shr() [3/7]

template<int shift, int n>

v_reg< int64, n > cv::v_shr ( const v_reg< int64, n > & a )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v_shr() [4/7]

template<int shift, int n>

v_reg< short, n > cv::v_shr ( const v_reg< short, n > & a )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v_shr() [5/7]

template<int shift, int n>

v_reg< uint64, n > cv::v_shr ( const v_reg< uint64, n > & a )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v_shr() [6/7]

template<int shift, int n>

v_reg< unsigned, n > cv::v_shr ( const v_reg< unsigned, n > & a )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v_shr() [7/7]

template<int shift, int n>

v_reg< ushort, n > cv::v_shr ( const v_reg< ushort, n > & a )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

v_signmask()

template<typename _Tp , int n>

int cv::v_signmask ( const v_reg< _Tp, n > & a )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

Get negative values mask.

Deprecated

v_signmask depends on a lane count heavily and therefore isn't universal enough

Returned value is a bit mask with bits set to 1 on places corresponding to negative packed values indexes. Example:

v_int32x4 r; // set to {-1, -1, 1, 1}
int mask = v_signmask(r); // mask = 3 <== 00000000 00000000 00000000 00000011

v_sin()

template<typename _Tp , int n>

v_reg< _Tp, n > cv::v_sin ( const v_reg< _Tp, n > & a )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

Sine \( sin(x) \) of elements.

Only for floating point types. Core implementation the same as v_sincos.

v_sincos()

template<typename _Tp , int n>

void cv::v_sincos ( const v_reg< _Tp, n > & x, v_reg< _Tp, n > & s, v_reg< _Tp, n > & c )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

Compute sine \( sin(x) \) and cosine \( cos(x) \) of elements at the same time.

Only for floating point types. Core implementation steps:

1. Input Normalization: Scale the periodicity from 2π to 4 and reduce the angle to the range \( [0, \frac{\pi}{4}] \) using periodicity and trigonometric identities.
2. Polynomial Approximation for \( sin(x) \) and \( cos(x) \):
   • For float16 and float32, use a Taylor series with 4 terms for sine and 5 terms for cosine.
   • For float64, use a Taylor series with 7 terms for sine and 8 terms for cosine.
3. Select Results: select and convert the final sine and cosine values for the original input angle.

Note

The precision of the calculation depends on the implementation and the data type of the input vector.

v_sqr_magnitude()

template<typename _Tp , int n>

v_reg< _Tp, n > cv::v_sqr_magnitude ( const v_reg< _Tp, n > & a, const v_reg< _Tp, n > & b )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

Square of the magnitude.

Returns \( a^2 + b^2 \) For floating point types only.

v_sqrt()

template<typename _Tp , int n>

v_reg< _Tp, n > cv::v_sqrt ( const v_reg< _Tp, n > & a )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

Square root of elements.

Only for floating point types.

v_store() [1/2]

template<typename _Tp , int n>

void cv::v_store ( _Tp * ptr, const v_reg< _Tp, n > & a )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

Store data to memory.

Store register contents to memory. Scheme:

REG {A B C D} ==> MEM {A B C D}

Pointer can be unaligned.

Here is the call graph for this function:

v_store() [2/2]

template<typename _Tp , int n>

void cv::v_store ( _Tp * ptr, const v_reg< _Tp, n > & a, hal::StoreMode  )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

Here is the call graph for this function:

v_store_aligned() [1/2]

template<typename _Tp , int n>

void cv::v_store_aligned ( _Tp * ptr, const v_reg< _Tp, n > & a )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

Store data to memory (aligned)

Store register contents to memory. Scheme:

REG {A B C D} ==> MEM {A B C D}

Pointer should be aligned by 16-byte boundary.

Here is the call graph for this function:

v_store_aligned() [2/2]

template<typename _Tp , int n>

void cv::v_store_aligned ( _Tp * ptr, const v_reg< _Tp, n > & a, hal::StoreMode  )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

Here is the call graph for this function:

v_store_aligned_nocache()

template<typename _Tp , int n>

void cv::v_store_aligned_nocache ( _Tp * ptr, const v_reg< _Tp, n > & a )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

Here is the call graph for this function:

v_store_high()

template<typename _Tp , int n>

void cv::v_store_high ( _Tp * ptr, const v_reg< _Tp, n > & a )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

Store data to memory (higher half)

Store higher half of register contents to memory. Scheme:

REG {A B C D} ==> MEM {C D}

Here is the call graph for this function:

v_store_interleave() [1/3]

template<typename _Tp , int n>

void cv::v_store_interleave ( _Tp * ptr, const v_reg< _Tp, n > & a, const v_reg< _Tp, n > & b, const v_reg< _Tp, n > & c, const v_reg< _Tp, n > & d, hal::StoreMode = hal::STORE_UNALIGNED )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

Interleave and store (4 channels)

Interleave and store data from 4 registers to memory. Scheme:

{A1 A2 ...}, {B1 B2 ...}, {C1 C2 ...}, {D1 D2 ...} ==> {A1 B1 C1 D1 A2 B2 C2 D2 ...}

For all types except 64-bit.

Here is the call graph for this function:

v_store_interleave() [2/3]

template<typename _Tp , int n>

void cv::v_store_interleave ( _Tp * ptr, const v_reg< _Tp, n > & a, const v_reg< _Tp, n > & b, const v_reg< _Tp, n > & c, hal::StoreMode = hal::STORE_UNALIGNED )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

Interleave and store (3 channels)

Interleave and store data from 3 registers to memory. Scheme:

{A1 A2 ...}, {B1 B2 ...}, {C1 C2 ...} ==> {A1 B1 C1 A2 B2 C2 ...}

For all types except 64-bit.

Here is the call graph for this function:

v_store_interleave() [3/3]

template<typename _Tp , int n>

void cv::v_store_interleave ( _Tp * ptr, const v_reg< _Tp, n > & a, const v_reg< _Tp, n > & b, hal::StoreMode = hal::STORE_UNALIGNED )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

Interleave and store (2 channels)

Interleave and store data from 2 registers to memory. Scheme:

{A1 A2 ...}, {B1 B2 ...} ==> {A1 B1 A2 B2 ...}

For all types except 64-bit.

Here is the call graph for this function:

v_store_low()

template<typename _Tp , int n>

void cv::v_store_low ( _Tp * ptr, const v_reg< _Tp, n > & a )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

Store data to memory (lower half)

Store lower half of register contents to memory. Scheme:

REG {A B C D} ==> MEM {A B}

Here is the call graph for this function:

v_sub()

template<typename _Tp , int n>

v_reg< _Tp, n > cv::v_sub	(	const v_reg< _Tp, n > &	a,
		const v_reg< _Tp, n > &	b )

#include <opencv2/core/hal/intrin_cpp.hpp>

Subtract values.

For all types.

v_sub_wrap()

template<typename _Tp , int n>

v_reg< _Tp, n > cv::v_sub_wrap ( const v_reg< _Tp, n > & a, const v_reg< _Tp, n > & b )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

Subtract values without saturation.

For 8- and 16-bit integer values.

v_transpose4x4()

template<typename _Tp , int n>

void cv::v_transpose4x4 ( v_reg< _Tp, n > & a0, const v_reg< _Tp, n > & a1, const v_reg< _Tp, n > & a2, const v_reg< _Tp, n > & a3, v_reg< _Tp, n > & b0, v_reg< _Tp, n > & b1, v_reg< _Tp, n > & b2, v_reg< _Tp, n > & b3 )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

Transpose 4x4 matrix.

Scheme:

a0  {A1 A2 A3 A4}
a1  {B1 B2 B3 B4}
a2  {C1 C2 C3 C4}
a3  {D1 D2 D3 D4}
===============
b0  {A1 B1 C1 D1}
b1  {A2 B2 C2 D2}
b2  {A3 B3 C3 D3}
b3  {A4 B4 C4 D4}

v_trunc() [1/2]

template<int n>

v_reg< int, n *2 > cv::v_trunc ( const v_reg< double, n > & a )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

This is an overloaded member function, provided for convenience. It differs from the above function only in what argument(s) it accepts.

v_trunc() [2/2]

template<int n>

v_reg< int, n > cv::v_trunc ( const v_reg< float, n > & a )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

Truncate elements.

Truncate each value. Input type is float vector ==> output type is int vector.

Note

Only for floating point types.

v_xor()

template<typename _Tp , int n>

v_reg< _Tp, n > cv::v_xor	(	const v_reg< _Tp, n > &	a,
		const v_reg< _Tp, n > &	b )

#include <opencv2/core/hal/intrin_cpp.hpp>

Bitwise XOR.

Only for integer types.

v_zip()

template<typename _Tp , int n>

void cv::v_zip ( const v_reg< _Tp, n > & a0, const v_reg< _Tp, n > & a1, v_reg< _Tp, n > & b0, v_reg< _Tp, n > & b1 )

inline

#include <opencv2/core/hal/intrin_cpp.hpp>

Interleave two vectors.

Scheme:

  {A1 A2 A3 A4}
  {B1 B2 B3 B4}
---------------
  {A1 B1 A2 B2} and {A3 B3 A4 B4}

For all types except 64-bit.

Variable Documentation

popCountTable

const unsigned char cv::popCountTable[]

static

#include <opencv2/core/hal/intrin_cpp.hpp>

Initial value:

=
{
    0, 1, 1, 2, 1, 2, 2, 3, 1, 2, 2, 3, 2, 3, 3, 4,
    1, 2, 2, 3, 2, 3, 3, 4, 2, 3, 3, 4, 3, 4, 4, 5,
    1, 2, 2, 3, 2, 3, 3, 4, 2, 3, 3, 4, 3, 4, 4, 5,
    2, 3, 3, 4, 3, 4, 4, 5, 3, 4, 4, 5, 4, 5, 5, 6,
    1, 2, 2, 3, 2, 3, 3, 4, 2, 3, 3, 4, 3, 4, 4, 5,
    2, 3, 3, 4, 3, 4, 4, 5, 3, 4, 4, 5, 4, 5, 5, 6,
    2, 3, 3, 4, 3, 4, 4, 5, 3, 4, 4, 5, 4, 5, 5, 6,
    3, 4, 4, 5, 4, 5, 5, 6, 4, 5, 5, 6, 5, 6, 6, 7,
    1, 2, 2, 3, 2, 3, 3, 4, 2, 3, 3, 4, 3, 4, 4, 5,
    2, 3, 3, 4, 3, 4, 4, 5, 3, 4, 4, 5, 4, 5, 5, 6,
    2, 3, 3, 4, 3, 4, 4, 5, 3, 4, 4, 5, 4, 5, 5, 6,
    3, 4, 4, 5, 4, 5, 5, 6, 4, 5, 5, 6, 5, 6, 6, 7,
    2, 3, 3, 4, 3, 4, 4, 5, 3, 4, 4, 5, 4, 5, 5, 6,
    3, 4, 4, 5, 4, 5, 5, 6, 4, 5, 5, 6, 5, 6, 6, 7,
    3, 4, 4, 5, 4, 5, 5, 6, 4, 5, 5, 6, 5, 6, 6, 7,
    4, 5, 5, 6, 5, 6, 6, 7, 5, 6, 6, 7, 6, 7, 7, 8,
}