Math operation APIs

Defines math operations on XeTLA vector data types. More...

Collaboration diagram for Math operation APIs:

Functions
template<typename T0 , typename T1 , int SZ>
__XETLA_API xetla_vector< T0, SZ >	gpu::xetla::xetla_abs (xetla_vector< T1, SZ > src0)
	Get absolute value (vector version)
template<typename T0 , typename T1 >
std::enable_if_t<!std::is_same< remove_const_t< T0 >, remove_const_t< T1 > >::value, remove_const_t< T0 > > __XETLA_API	gpu::xetla::xetla_abs (T1 src0)
	Get absolute value (scalar version)
template<typename T1 , int SZ>
__XETLA_API xetla_vector< T1, SZ >	gpu::xetla::xetla_abs (xetla_vector< T1, SZ > src0)
	Get absolute value (vector version).
template<typename T1 >
__XETLA_API std::remove_const< T1 >::type	gpu::xetla::xetla_abs (T1 src0)
	Get absolute value (scalar version).
template<typename T , int SZ, typename Sat = xetla_saturation_off_tag>
__XETLA_API xetla_vector< T, SZ >	gpu::xetla::xetla_max (xetla_vector< T, SZ > src0, xetla_vector< T, SZ > src1, Sat sat={})
	Selects component-wise the maximum of the two vectors.
template<typename T , int SZ, typename Sat = xetla_saturation_off_tag>
__XETLA_API xetla_vector< T, SZ >	gpu::xetla::xetla_max (xetla_vector< T, SZ > src0, T src1, Sat sat={})
	Selects maximums for each element of the input vector and a scalar.
template<typename T , int SZ, typename Sat = xetla_saturation_off_tag>
__XETLA_API xetla_vector< T, SZ >	gpu::xetla::xetla_max (T src0, xetla_vector< T, SZ > src1, Sat sat={})
	Selects maximums for each element of the input scalar and a vector.
template<typename T , typename Sat = xetla_saturation_off_tag>
__XETLA_API T	gpu::xetla::xetla_max (T src0, T src1, Sat sat={})
	Selects maximum between two scalar values.
template<typename T , int SZ, typename Sat = xetla_saturation_off_tag>
__XETLA_API xetla_vector< T, SZ >	gpu::xetla::xetla_min (xetla_vector< T, SZ > src0, xetla_vector< T, SZ > src1, Sat sat={})
	Selects component-wise the minimum of the two vectors.
template<typename T , int SZ, typename Sat = xetla_saturation_off_tag>
__XETLA_API xetla_vector< T, SZ >	gpu::xetla::xetla_min (xetla_vector< T, SZ > src0, T src1, Sat sat={})
	Selects minimums for each element of the input vector and a scalar.
template<typename T , int SZ, typename Sat = xetla_saturation_off_tag>
__XETLA_API xetla_vector< T, SZ >	gpu::xetla::xetla_min (T src0, xetla_vector< T, SZ > src1, Sat sat={})
	Selects minimums for each element of the input scalar and a vector.
template<typename T , typename Sat = xetla_saturation_off_tag>
__XETLA_API T	gpu::xetla::xetla_min (T src0, T src1, Sat sat={})
	Selects minimum between two scalar values.
template<class T , int SZ, typename Sat = xetla_saturation_off_tag>
__XETLA_API xetla_vector< T, SZ >	gpu::xetla::xetla_exp (xetla_vector< T, SZ > src, Sat sat={})
	Calculate exponent value for each element of the input vector, the base is e.
template<class T , typename Sat = xetla_saturation_off_tag>
__XETLA_API T	gpu::xetla::xetla_exp (T src, Sat sat={})
	Calculate exponent value of a scalar, the base is e.
template<class T , int SZ, typename Sat = xetla_saturation_off_tag>
__XETLA_API xetla_vector< T, SZ >	gpu::xetla::xetla_exp2 (xetla_vector< T, SZ > src, Sat sat={})
	Calculate exponent value for each element of the input vector, the base is 2.
template<class T , typename Sat = xetla_saturation_off_tag>
__XETLA_API T	gpu::xetla::xetla_exp2 (T src, Sat sat={})
	Calculate exponent value of a scalar, the base is 2.
template<typename T , int SZ, typename Sat = xetla_saturation_off_tag>
__XETLA_API xetla_vector< T, SZ >	gpu::xetla::xetla_inv (xetla_vector< T, SZ > src, Sat sat={})
	Calculate the inversion, i.e.
template<typename T , typename Sat = xetla_saturation_off_tag>
__XETLA_API T	gpu::xetla::xetla_inv (T src, Sat sat={})
	Calculate the inversion, i.e.
template<typename T , int SZ, typename Sat = xetla_saturation_off_tag>
__XETLA_API xetla_vector< T, SZ >	gpu::xetla::xetla_sqrt (xetla_vector< T, SZ > src, Sat sat={})
	Calculate the square root, i.e.
template<typename T , typename Sat = xetla_saturation_off_tag>
__XETLA_API T	gpu::xetla::xetla_sqrt (T src, Sat sat={})
	Calculate the square root, i.e.
template<typename T , int SZ, typename Sat = xetla_saturation_off_tag>
__XETLA_API xetla_vector< T, SZ >	gpu::xetla::xetla_sqrt_ieee (xetla_vector< T, SZ > src, Sat sat={})
	Calculate the square root, i.e.
template<typename T , typename Sat = xetla_saturation_off_tag>
__XETLA_API T	gpu::xetla::xetla_sqrt_ieee (T src, Sat sat={})
	Calculate the square root, i.e.
template<typename T , int SZ, typename Sat = xetla_saturation_off_tag>
__XETLA_API xetla_vector< T, SZ >	gpu::xetla::xetla_rsqrt (xetla_vector< T, SZ > src, Sat sat={})
	Calculate the inversion of square root, i.e.
template<typename T , typename Sat = xetla_saturation_off_tag>
__XETLA_API T	gpu::xetla::xetla_rsqrt (T src, Sat sat={})
	Calculate the inversion of square root, i.e.
template<typename T , int SZ>
__XETLA_API xetla_vector< T, SZ >	gpu::xetla::xetla_tanh (xetla_vector< T, SZ > src)
	Calculate the tanh (vector version).
template<typename T >
__XETLA_API T	gpu::xetla::xetla_tanh (T src)
	Calculate the tanh (scalar version).
template<typename T , int SZ>
__XETLA_API xetla_vector< T, SZ >	gpu::xetla::xetla_add_c (xetla_vector< T, SZ > src0, xetla_vector< T, SZ > src1, xetla_vector_ref< T, SZ > __REF__ carry)
	Add two unsigned integer vectors, return the result and in-place update the carry.
template<typename T , int SZ>
__XETLA_API xetla_vector< T, SZ >	gpu::xetla::xetla_add_c (xetla_vector< T, SZ > src0, T src1, xetla_vector_ref< T, SZ > __REF__ carry)
	Add one unsigned integer vectors with a scalar, return the result and in-place update the carry.
template<typename T0 , typename T1 , typename T2 , int SZ>
__XETLA_API xetla_vector< T0, SZ >	gpu::xetla::xetla_imul (xetla_vector_ref< T0, SZ > __REF__ lo, xetla_vector< T1, SZ > src0, T2 src1)
	Multiply src0 with src1, return the hi part and in-place update the lo part.
template<typename T0 , typename T1 , int SZ, reduce_op BinaryOperation>
__XETLA_API T0	gpu::xetla::xetla_reduce (xetla_vector< T1, SZ > v)
	Performs reduction over elements of the input vector.
template<typename T , int SZ>
__XETLA_API xetla_vector< T, SZ >	gpu::xetla::xetla_rnde (xetla_vector< T, SZ > src0)
	Get rounded value.
template<typename T1 , typename T0 , int SZ, typename Sat = xetla_saturation_off_tag>
__XETLA_API xetla_vector< T1, SZ >	gpu::xetla::xetla_add (xetla_vector< T0, SZ > src0, xetla_vector< T0, SZ > src1, Sat sat={})
	Adds two vectors with saturation The source operands must be both of floating-point type.
template<typename T1 , typename T0 , int SZ>
__XETLA_API xetla_vector< T1, SZ >	gpu::xetla::xetla_sat (xetla_vector< T0, SZ > src)
	Saturation function.
template<argument_type src1_precision, argument_type src2_precision, int systolic_depth, int repeat_count, typename T , typename T1 , typename T2 , int N, int N1, int N2, typename Sat = xetla_saturation_off_tag>
__XETLA_API xetla_vector< T, N >	gpu::xetla::xetla_mma (xetla_vector< T, N > src0, xetla_vector< T1, N1 > src1, xetla_vector< T2, N2 > src2, Sat sat={})
	description of xetla mma perform matrix multiply add operation

Detailed Description

Defines math operations on XeTLA vector data types.

Function Documentation

xetla_abs() [1/4]

template<typename T0 , typename T1 >

std::enable_if_t<!std::is_same< remove_const_t< T0 >, remove_const_t< T1 > >::value, remove_const_t< T0 > > __XETLA_API gpu::xetla::xetla_abs

(

T1

src0

)

Get absolute value (scalar version)

Template Parameters

T0	element type of the returned value.
T1	element type of the input value.

Parameters

src0	the source operand.

Returns

absolute value.

xetla_abs() [2/4]

template<typename T1 >

__XETLA_API std::remove_const< T1 >::type gpu::xetla::xetla_abs

(

T1

src0

)

Get absolute value (scalar version).

This is a specialization of a version with two template parameters, where the types of the input and output value are the same.

Template Parameters

T1	element type of the input and output value.

Parameters

src0	the source operand.

Returns

absolute value.

xetla_abs() [3/4]

template<typename T0 , typename T1 , int SZ>

__XETLA_API xetla_vector< T0, SZ > gpu::xetla::xetla_abs

(

xetla_vector< T1, SZ >

src0

)

Get absolute value (vector version)

Template Parameters

T0	element type of the returned vector.
T1	element type of the input vector.
SZ	size of the input and returned vector.

Parameters

src0	the input vector.

Returns

vector of absolute values.

xetla_abs() [4/4]

template<typename T1 , int SZ>

__XETLA_API xetla_vector< T1, SZ > gpu::xetla::xetla_abs

(

xetla_vector< T1, SZ >

src0

)

Get absolute value (vector version).

This is a specialization of a version with three template parameters, where the element types of the input and output vector are the same.

Template Parameters

T1	element type of the input and output vectors.
SZ	size of the input and returned vectors.

Parameters

src0	the input vector.

Returns

vector of absolute values.

xetla_add()

template<typename T1 , typename T0 , int SZ, typename Sat = xetla_saturation_off_tag>

__XETLA_API xetla_vector< T1, SZ > gpu::xetla::xetla_add	(	xetla_vector< T0, SZ >	src0,
		xetla_vector< T0, SZ >	src1,
		Sat	sat = {}
	)

Adds two vectors with saturation The source operands must be both of floating-point type.

Template Parameters

T0	element type of the input vectors.
T1	element type of the return vector.
SZ	size of the input and returned vectors.

Parameters

src0	the input vector.
src1	the input vector.
sat	enables/disables the saturation (off by default). Possible values: saturation_on/saturation_off.

Returns

vector of component-wise additive elements.

for esimd, current we don't have add with saturation, this is the emulation

xetla_add_c() [1/2]

template<typename T , int SZ>

__XETLA_API xetla_vector< T, SZ > gpu::xetla::xetla_add_c	(	xetla_vector< T, SZ >	src0,
		T	src1,
		xetla_vector_ref< T, SZ > __REF__	carry
	)

Add one unsigned integer vectors with a scalar, return the result and in-place update the carry.

Template Parameters

T	element type of the src, should be uint32_t.
SZ	element num of the vector.

Parameters

src0	is the src0 of the add operation.
src1	is the src1 of the add operation.
carry	is the carry of the add operation.

Returns

result of the src0 + src1.

xetla_add_c() [2/2]

template<typename T , int SZ>

__XETLA_API xetla_vector< T, SZ > gpu::xetla::xetla_add_c	(	xetla_vector< T, SZ >	src0,
		xetla_vector< T, SZ >	src1,
		xetla_vector_ref< T, SZ > __REF__	carry
	)

Add two unsigned integer vectors, return the result and in-place update the carry.

Template Parameters

T	element type of the src, should be uint32_t.
SZ	element num of the vector.

Parameters

src0	is the src0 of the add operation.
src1	is the src1 of the add operation.
carry	is the carry of the add operation.

Returns

result of the src0 + src1.

xetla_exp() [1/2]

template<class T , typename Sat = xetla_saturation_off_tag>

__XETLA_API T gpu::xetla::xetla_exp	(	T	src,
		Sat	sat = {}
	)

Calculate exponent value of a scalar, the base is e.

Template Parameters

T	element type of the input and return a scalar.

Parameters

src	the scalar value.
sat	enables/disables the saturation (off by default). Possible values: saturation_on/saturation_off.

Returns

exponent value.

xetla_exp() [2/2]

template<class T , int SZ, typename Sat = xetla_saturation_off_tag>

__XETLA_API xetla_vector< T, SZ > gpu::xetla::xetla_exp	(	xetla_vector< T, SZ >	src,
		Sat	sat = {}
	)

Calculate exponent value for each element of the input vector, the base is e.

Template Parameters

T	element type of the input and return vectors.
SZ	size of the input and returned vectors.

Parameters

src	the input vector.
sat	enables/disables the saturation (off by default). Possible values: saturation_on/saturation_off.

Returns

vector of component-wise exponent elements.

xetla_exp2() [1/2]

template<class T , typename Sat = xetla_saturation_off_tag>

__XETLA_API T gpu::xetla::xetla_exp2	(	T	src,
		Sat	sat = {}
	)

Calculate exponent value of a scalar, the base is 2.

Template Parameters

T	element type of the input and return a scalar.

Parameters

src	the scalar value.
sat	enables/disables the saturation (off by default). Possible values: saturation_on/saturation_off.

Returns

exponent value.

xetla_exp2() [2/2]

template<class T , int SZ, typename Sat = xetla_saturation_off_tag>

__XETLA_API xetla_vector< T, SZ > gpu::xetla::xetla_exp2	(	xetla_vector< T, SZ >	src,
		Sat	sat = {}
	)

Calculate exponent value for each element of the input vector, the base is 2.

Template Parameters

T	element type of the input and return vectors.
SZ	size of the input and returned vectors.

Parameters

src	the input vector.
sat	enables/disables the saturation (off by default). Possible values: saturation_on/saturation_off.

Returns

vector of component-wise exponent elements.

xetla_imul()

template<typename T0 , typename T1 , typename T2 , int SZ>

__XETLA_API xetla_vector< T0, SZ > gpu::xetla::xetla_imul	(	xetla_vector_ref< T0, SZ > __REF__	lo,
		xetla_vector< T1, SZ >	src0,
		T2	src1
	)

Multiply src0 with src1, return the hi part and in-place update the lo part.

Template Parameters

T0	Return data type, should be 32 bit integer.
T1	Type of src0, should be 32 bit integer.
T2	Type of src1, should be 32 bit integer.
SZ	Element num of the vector.

Parameters

lo	is the low 32 bit part of the result.
src0	is the src0 of the mul operation.
src1	is the src1 of the mul operation.

Returns

return the high 32 bit part of the result.

xetla_inv() [1/2]

template<typename T , typename Sat = xetla_saturation_off_tag>

__XETLA_API T gpu::xetla::xetla_inv	(	T	src,
		Sat	sat = {}
	)

Calculate the inversion, i.e.

1/x (scalar version).

Template Parameters

T	Is the element data type
Sat	Is the saturation flag(off by default). Possible values: saturation_on/saturation_off.

Parameters

src	Is the input scalar.

Returns

Returns the result of 1/x

xetla_inv() [2/2]

template<typename T , int SZ, typename Sat = xetla_saturation_off_tag>

__XETLA_API xetla_vector< T, SZ > gpu::xetla::xetla_inv	(	xetla_vector< T, SZ >	src,
		Sat	sat = {}
	)

Calculate the inversion, i.e.

1/x (vector version).

Template Parameters

T	Is the element data type
SZ	Is the element num in the vector
Sat	Is the saturation flag(off by default). Possible values: saturation_on/saturation_off.

Parameters

src	Is the input vector.

Returns

Returns the result of 1/x

xetla_max() [1/4]

template<typename T , typename Sat = xetla_saturation_off_tag>

__XETLA_API T gpu::xetla::xetla_max	(	T	src0,
		T	src1,
		Sat	sat = {}
	)

Selects maximum between two scalar values.

The source operands must be both of integer or both of floating-point type.

Template Parameters

T	element type of the input and return vectors.

Parameters

src0	the scalar value.
src1	the scalar value.
sat	enables/disables the saturation (off by default). Possible values: saturation_on/saturation_off.

Returns

maximum value between the two inputs.

xetla_max() [2/4]

template<typename T , int SZ, typename Sat = xetla_saturation_off_tag>

__XETLA_API xetla_vector< T, SZ > gpu::xetla::xetla_max	(	T	src0,
		xetla_vector< T, SZ >	src1,
		Sat	sat = {}
	)

Selects maximums for each element of the input scalar and a vector.

The source operands must be both of integer or both of floating-point type.

Template Parameters

T	element type of the input and return vectors.
SZ	size of the input and returned vectors.

Parameters

src0	the scalar value.
src1	the input vector.
sat	enables/disables the saturation (off by default). Possible values: saturation_on/saturation_off.

Returns

vector of component-wise maximum elements.

xetla_max() [3/4]

template<typename T , int SZ, typename Sat = xetla_saturation_off_tag>

__XETLA_API xetla_vector< T, SZ > gpu::xetla::xetla_max	(	xetla_vector< T, SZ >	src0,
		T	src1,
		Sat	sat = {}
	)

Selects maximums for each element of the input vector and a scalar.

The source operands must be both of integer or both of floating-point type.

Template Parameters

T	element type of the input and return vectors.
SZ	size of the input and returned vectors.

Parameters

src0	the input vector.
src1	the scalar value.
sat	enables/disables the saturation (off by default). Possible values: saturation_on/saturation_off.

Returns

vector of component-wise maximum elements.

xetla_max() [4/4]

template<typename T , int SZ, typename Sat = xetla_saturation_off_tag>

__XETLA_API xetla_vector< T, SZ > gpu::xetla::xetla_max	(	xetla_vector< T, SZ >	src0,
		xetla_vector< T, SZ >	src1,
		Sat	sat = {}
	)

Selects component-wise the maximum of the two vectors.

The source operands must be both of integer or both of floating-point type.

Template Parameters

T	element type of the input and return vectors.
SZ	size of the input and returned vectors.

Parameters

src0	the input vector.
src1	the input vector.
sat	enables/disables the saturation (off by default). Possible values: saturation_on/saturation_off.

Returns

vector of component-wise maximum elements.

xetla_min() [1/4]

template<typename T , typename Sat = xetla_saturation_off_tag>

__XETLA_API T gpu::xetla::xetla_min	(	T	src0,
		T	src1,
		Sat	sat = {}
	)

Selects minimum between two scalar values.

The source operands must be both of integer or both of floating-point type.

Template Parameters

T	element type of the input and return vectors.

Parameters

src0	the scalar value.
src1	the scalar value.
sat	enables/disables the saturation (off by default). Possible values: saturation_on/saturation_off.

Returns

minimum value between the two inputs.

xetla_min() [2/4]

template<typename T , int SZ, typename Sat = xetla_saturation_off_tag>

__XETLA_API xetla_vector< T, SZ > gpu::xetla::xetla_min	(	T	src0,
		xetla_vector< T, SZ >	src1,
		Sat	sat = {}
	)

Selects minimums for each element of the input scalar and a vector.

The source operands must be both of integer or both of floating-point type.

Template Parameters

T	element type of the input and return vectors.
SZ	size of the input and returned vectors.

Parameters

src0	the scalar value.
src1	the input vector.
sat	enables/disables the saturation (off by default). Possible values: saturation_on/saturation_off.

Returns

vector of component-wise minimum elements.

xetla_min() [3/4]

template<typename T , int SZ, typename Sat = xetla_saturation_off_tag>

__XETLA_API xetla_vector< T, SZ > gpu::xetla::xetla_min	(	xetla_vector< T, SZ >	src0,
		T	src1,
		Sat	sat = {}
	)

Selects minimums for each element of the input vector and a scalar.

The source operands must be both of integer or both of floating-point type.

Template Parameters

T	element type of the input and return vectors.
SZ	size of the input and returned vectors.

Parameters

src0	the input vector.
src1	the scalar value.
sat	enables/disables the saturation (off by default). Possible values: saturation_on/saturation_off.

Returns

vector of component-wise minimum elements.

xetla_min() [4/4]

template<typename T , int SZ, typename Sat = xetla_saturation_off_tag>

__XETLA_API xetla_vector< T, SZ > gpu::xetla::xetla_min	(	xetla_vector< T, SZ >	src0,
		xetla_vector< T, SZ >	src1,
		Sat	sat = {}
	)

Selects component-wise the minimum of the two vectors.

The source operands must be both of integer or both of floating-point type.

Template Parameters

T	element type of the input and return vectors.
SZ	size of the input and returned vectors.

Parameters

src0	the input vector.
src1	the input vector.
sat	enables/disables the saturation (off by default). Possible values: saturation_on/saturation_off.

Returns

vector of component-wise minimum elements.

xetla_mma()

template<argument_type src1_precision, argument_type src2_precision, int systolic_depth, int repeat_count, typename T , typename T1 , typename T2 , int N, int N1, int N2, typename Sat = xetla_saturation_off_tag>

__XETLA_API xetla_vector< T, N > gpu::xetla::xetla_mma	(	xetla_vector< T, N >	src0,
		xetla_vector< T1, N1 >	src1,
		xetla_vector< T2, N2 >	src2,
		Sat	sat = {}
	)

description of xetla mma perform matrix multiply add operation

Template Parameters

src1_precision	is the data precision of src1
src2_precision	is the data precision of src2
systolic_depth	is the depth of mma (i.e k dimension size in dword)
repeat_count	is the row (m) of mma (mxkxn)
T	is the data type of src0 and dst
T1	is the data type of src1
T2	is the data type of src2
N	is the total number of elements in src0 and dst
N1	is the total number of elements in src1
N2	is the total number of elements in src2
Sat	is saturation flag

Parameters

src0	[in] is src0
src1	[in] is src1
src2	[in] is src2
sat	[in] is saturation flag

Returns

is dst, a xetla_vector of type T and element size of N todo: we need to remove argument_type and get the currect precision based on data type

xetla_reduce()

template<typename T0 , typename T1 , int SZ, reduce_op BinaryOperation>

__XETLA_API T0 gpu::xetla::xetla_reduce

(

xetla_vector< T1, SZ >

v

)

Performs reduction over elements of the input vector.

Template Parameters

T0	type of the return value.
T1	element type of the input vector.
SZ	size of the input vector.
BinaryOperation	type representing the operation. Can be an instantion of one of the following types: reduce_op::sum, performs reduce sum operation reduce_op::prod, performs reduce prod operation reduce_op::min, performs reduce min operation reduce_op::max, performs reduce max operation

Parameters

v	the vector to perform reduction on

Returns

result of the reduction

xetla_rnde()

template<typename T , int SZ>

__XETLA_API xetla_vector< T, SZ > gpu::xetla::xetla_rnde

(

xetla_vector< T, SZ >

src0

)

Get rounded value.

Template Parameters

T	element type of the input vector.
SZ	size of the input and returned vector.

Parameters

src0	the input vector.

Returns

vector of rounded values.

xetla_rsqrt() [1/2]

template<typename T , typename Sat = xetla_saturation_off_tag>

__XETLA_API T gpu::xetla::xetla_rsqrt	(	T	src,
		Sat	sat = {}
	)

Calculate the inversion of square root, i.e.

1/sqrt(x) (scalar version).

Template Parameters

T	Is the element data type
Sat	Is the saturation flag(off by default). Possible values: saturation_on/saturation_off.

Parameters

src	Is the input vector.

Returns

Returns the result of 1/sqrt(x)

xetla_rsqrt() [2/2]

template<typename T , int SZ, typename Sat = xetla_saturation_off_tag>

__XETLA_API xetla_vector< T, SZ > gpu::xetla::xetla_rsqrt	(	xetla_vector< T, SZ >	src,
		Sat	sat = {}
	)

Calculate the inversion of square root, i.e.

1/sqrt(x) (vector version).

Template Parameters

T	Is the element data type
SZ	Is the element num in the vector
Sat	Is the saturation flag(off by default). Possible values: saturation_on/saturation_off.

Parameters

src	Is the input vector.

Returns

Returns the result of 1/sqrt(x)

xetla_sat()

template<typename T1 , typename T0 , int SZ>

__XETLA_API xetla_vector< T1, SZ > gpu::xetla::xetla_sat

(

xetla_vector< T0, SZ >

src

)

Saturation function.

Template Parameters

T0	element type of the input vectors.
T1	element type of the return vector.
SZ	size of the input and returned vectors.

xetla_sqrt() [1/2]

template<typename T , typename Sat = xetla_saturation_off_tag>

__XETLA_API T gpu::xetla::xetla_sqrt	(	T	src,
		Sat	sat = {}
	)

Calculate the square root, i.e.

x^(1/2), this is not IEEE754-compatible (scalar version).

Template Parameters

T	Is the element data type
Sat	Is the saturation flag(off by default). Possible values: saturation_on/saturation_off.

Parameters

src	Is the input vector.

Returns

Returns the result of x^(1/2)

xetla_sqrt() [2/2]

template<typename T , int SZ, typename Sat = xetla_saturation_off_tag>

__XETLA_API xetla_vector< T, SZ > gpu::xetla::xetla_sqrt	(	xetla_vector< T, SZ >	src,
		Sat	sat = {}
	)

Calculate the square root, i.e.

x^(1/2), this is not IEEE754-compatible (vector version).

Template Parameters

T	Is the element data type
SZ	Is the element num in the vector
Sat	Is the saturation flag(off by default). Possible values: saturation_on/saturation_off.

Parameters

src	Is the input vector.

Returns

Returns the result of x^(1/2)

xetla_sqrt_ieee() [1/2]

template<typename T , typename Sat = xetla_saturation_off_tag>

__XETLA_API T gpu::xetla::xetla_sqrt_ieee	(	T	src,
		Sat	sat = {}
	)

Calculate the square root, i.e.

x^(1/2), IEEE754-compatible (scalar version).

Template Parameters

T	Is the element data type
Sat	Is the saturation flag(off by default). Possible values: saturation_on/saturation_off.

Parameters

src	Is the input vector.

Returns

Returns the result of x^(1/2)

xetla_sqrt_ieee() [2/2]

template<typename T , int SZ, typename Sat = xetla_saturation_off_tag>

__XETLA_API xetla_vector< T, SZ > gpu::xetla::xetla_sqrt_ieee	(	xetla_vector< T, SZ >	src,
		Sat	sat = {}
	)

Calculate the square root, i.e.

x^(1/2), IEEE754-compatible (vector version).

Template Parameters

T	Is the element data type
SZ	Is the element num in the vector
Sat	Is the saturation flag(off by default). Possible values: saturation_on/saturation_off.

Parameters

src	Is the input vector.

Returns

Returns the result of x^(1/2)

xetla_tanh() [1/2]

template<typename T >

__XETLA_API T gpu::xetla::xetla_tanh

(

T

src

)

Calculate the tanh (scalar version).

Template Parameters

T	Is the element data type
SZ	Is the element num in the vector

Parameters

src	Is the input vector.

Returns

Returns the result of tanh(x)

xetla_tanh() [2/2]

template<typename T , int SZ>

__XETLA_API xetla_vector< T, SZ > gpu::xetla::xetla_tanh

(

xetla_vector< T, SZ >

src

)

Calculate the tanh (vector version).

Template Parameters

T	Is the element data type
SZ	Is the element num in the vector

Parameters

src	Is the input vector.

Returns

Returns the result of tanh(x)