Lazy Operations - 2023.2 English

AI Engine API User Guide (AIE) (UG1529)
Document ID
UG1529
Release Date
2023-11-16
Version
2023.2 English
AI Engine API User Guide: Lazy Operations
AI Engine API User Guide (AIE) 2023.2
Loading...
Searching...
No Matches
Functions
Lazy Operations
Basic Types

AIE architectures offer multiplication instructions that can perform additional operations on on the input arguments. More...

Overview

AIE architectures offer multiplication instructions that can perform additional operations on on the input arguments.

Instead of adding one variant for each possible combination, AIE API offers types that can wrap an existing vector, accumulator of element reference and be passed into the multiplication function. Then the API will merge the operations into a single instruction or apply the operation on the vector before the multiplication, depending on the hardware support,

The following example performs an element-wise multiplication of the absolute of vector a and the conjugate of vector b.

aie::accum<cacc48, 16> foo(aie::vector<int16, 16> a, aie::vector<cint16, 16> b)
{
aie::accum<cacc48, 16> ret;
ret = aie::mul(aie::op_abs(a), aie::op_conj(b));
return ret;
}
aie::mul
constexpr auto mul(const Vec1 &v1, const Vec2 &v2) -> accum< AccumTag, Vec1::size()>
Returns an accumulator of the requested type with the element-wise multiplication of the two input ve...
Definition aie.hpp:3860
aie::op_abs
constexpr unary_op< T, Operation::Abs > op_abs(const T &e)
Returns an absolute operation modifier for the given vector or element.
Definition aie.hpp:604
aie::op_conj
constexpr unary_op< T, Operation::Conj > op_conj(const T &e)
Returns a conjugate operation modifier for the given complex vector or element.
Definition aie.hpp:619
aie::accum
Type for vector accumulators.
Definition accum.hpp:84
aie::vector
Type for vector registers.
Definition vector.hpp:107

Functions

template<typename T >
requires (RealVector<T> || RealElem<T> || is_real_op_v<T>)
constexpr unary_op< T, Operation::Abs > aie::op_abs (const T &e)
 Returns an absolute operation modifier for the given vector or element.
 
template<Accum Acc>
constexpr unary_op< Acc, Operation::Acc_Add > aie::op_add (const Acc &acc)
 Returns an addition operation modifier for the given accumulator.
 
template<typename T >
requires (ComplexVector<T> || ComplexElem<T> || is_complex_op_v<T>)
constexpr unary_op< T, Operation::Conj > aie::op_conj (const T &e)
 Returns a conjugate operation modifier for the given complex vector or element.
 
template<VectorOrOp Vec1, VectorOrOp Vec2>
constexpr binary_op< Vec1, Vec2, Operation::Max > aie::op_max (const Vec1 &a, const Vec2 &b)
 Returns a maximum operation modifier for the given pair of vectors or elements.
 
template<VectorOrOp Vec1, VectorOrOp Vec2>
constexpr binary_op< Vec1, Vec2, Operation::Min > aie::op_min (const Vec1 &a, const Vec2 &b)
 Returns a minimum operation modifier for the given pair of vectors or elements.
 
template<typename T >
requires (Vector<T> || Elem<T> || Accum<T>)
constexpr unary_op< T, Operation::None > aie::op_none (const T &e)
 Returns an empty operation modifier for the given vector, vector element or accumulator.
 
template<typename T >
requires ((Vector<T> || Elem<T> || Accum<T> || is_op_v<T>) && (!detail::is_complex_v<typename T::value_type>) && (!detail::is_floating_point_v<typename T::value_type>))
constexpr binary_op< T, bool, Operation::Sign > aie::op_sign (const T &e, bool is_signed)
 Returns dynamically signed/unsigned operation modifier for the vector or element, based on the boolean parameter.
 
template<Accum Acc>
constexpr unary_op< Acc, Operation::Acc_Sub > aie::op_sub (const Acc &acc)
 Returns a subtraction operation modifier for the given accumulator.
 
template<Accum Acc>
constexpr binary_op< Acc, bool, Operation::Zero > aie::op_zero (const Acc &acc, bool to_zero)
 Zeroes out the given accumulator or not depending on the zero parameter.
 

Function Documentation

◆ op_abs()

template<typename T >
requires (RealVector<T> || RealElem<T> || is_real_op_v<T>)
constexpr unary_op< T, Operation::Abs > aie::op_abs ( const T &  e)
constexpr

Returns an absolute operation modifier for the given vector or element.

On some architecture versions, this operation can be collapsed with a subsequent multiplication.

Parameters
eVector or element to which the operation is performed. The type must meet RealVector or RealElem.

◆ op_add()

template<Accum Acc>
constexpr unary_op< Acc, Operation::Acc_Add > aie::op_add ( const Acc &  acc)
constexpr

Returns an addition operation modifier for the given accumulator.

Parameters
accAccumulator to which the operation is performed.

◆ op_conj()

template<typename T >
requires (ComplexVector<T> || ComplexElem<T> || is_complex_op_v<T>)
constexpr unary_op< T, Operation::Conj > aie::op_conj ( const T &  e)
constexpr

Returns a conjugate operation modifier for the given complex vector or element.

On some architecture versions, this operation can be collapsed with a subsequent multiplication.

Parameters
eVector or element to which the operation is performed. The type must meet ComplexVector or ComplexElem.

◆ op_max()

template<VectorOrOp Vec1, VectorOrOp Vec2>
constexpr binary_op< Vec1, Vec2, Operation::Max > aie::op_max ( const Vec1 &  a,
const Vec2 &  b 
)
constexpr

Returns a maximum operation modifier for the given pair of vectors or elements.

On some architecture versions, this operation can be collapsed with a subsequent multiplication.

Parameters
aFirst vector or element passed to the operation. The type must meet Vector or Elem.
bSecond vector or element passed to the operation. The type must be the same as for a.

◆ op_min()

template<VectorOrOp Vec1, VectorOrOp Vec2>
constexpr binary_op< Vec1, Vec2, Operation::Min > aie::op_min ( const Vec1 &  a,
const Vec2 &  b 
)
constexpr

Returns a minimum operation modifier for the given pair of vectors or elements.

On some architecture versions, this operation can be collapsed with a subsequent multiplication.

Parameters
aFirst vector or element passed to the operation. The type must meet Vector or Elem.
bSecond vector or element passed to the operation. The type must be the same as for a.

◆ op_none()

template<typename T >
requires (Vector<T> || Elem<T> || Accum<T>)
constexpr unary_op< T, Operation::None > aie::op_none ( const T &  e)
constexpr

Returns an empty operation modifier for the given vector, vector element or accumulator.

Parameters
eVector, vector element, or accumulator to which the operation is performed. The type must meet Vector, Elem or Accum.

◆ op_sign()

template<typename T >
requires ((Vector<T> || Elem<T> || Accum<T> || is_op_v<T>) && (!detail::is_complex_v<typename T::value_type>) && (!detail::is_floating_point_v<typename T::value_type>))
constexpr binary_op< T, bool, Operation::Sign > aie::op_sign ( const T &  e,
bool  is_signed 
)
constexpr

Returns dynamically signed/unsigned operation modifier for the vector or element, based on the boolean parameter.

On some architecture versions, this operation can be collapsed with a subsequent operation.

Parameters
eVector, element or accumulator passed to the operation. The type must meet Vector, Elem, Accum or be another operation, using an integer type.
is_signedControls if the first paramer is interpreted as signed, or not.

◆ op_sub()

template<Accum Acc>
constexpr unary_op< Acc, Operation::Acc_Sub > aie::op_sub ( const Acc &  acc)
constexpr

Returns a subtraction operation modifier for the given accumulator.

Parameters
accAccumulator to which the operation is performed.

◆ op_zero()

template<Accum Acc>
constexpr binary_op< Acc, bool, Operation::Zero > aie::op_zero ( const Acc &  acc,
bool  to_zero 
)
constexpr

Zeroes out the given accumulator or not depending on the zero parameter.

On some architectures this operation can be collapsed with a subsequent operation.

Parameters
accAccumulator to which the operation is performed.
to_zeroControls if the passed accumulator is zeroed or not.