Arithmetic

AIE API provides a set of functions that implement arithmetic operations on vector types. More...

Overview

AIE API provides a set of functions that implement arithmetic operations on vector types.

Operands may be vectors, values or vector element references and the supported operand combinations are:

• Vector / Vector: the type and the size of the vectors must match. The operation is performed element-wise between the corresponding elements in each vector.
• Value / Vector: the type of the value and the type of the elements of the vector must match. The operation has the same result as if the value was broadcast to a vector and then operated with the vector argument.
• Vector element reference / Vector. Similar as Value / Vector, but using an element reference, the AIE API may optimize the operation by accessing the element directly from its original location.

The following code snippet shows an example that adds two input arrays into an output array using vectors. For simplicity count must be divisible by 8.

void add(int32 * __restrict out,
         const int32 * __restrict in1, const int32 * __restrict in2, unsigned count)
{
    for (unsigned i = 8; i < count; i += 8) {
        aie::vector<int32, 8> vec = aie::add(aie::load_v<8>(in1 + i),
                                             aie::load_v<8>(in2 + i));
 
        aie::store_v(out, vec);
    }
}

Operations that include a multiplication return an accumulator. The API defines a default accumulation for each combination of types, but users may specify a larger number of accumulation bits by explicitly passing an accumulator tag.

// Default accumulation will be used
auto acc = aie::mul(v1, v2);
 
// 64b accumulation, at least, will be used
auto acc = aie::mul<acc64>(v1, v2);
 
// For multiply-add operations, the API uses the same accumulation as the given accumulator (cannot be overriden)
auto acc2 = aie::mac(acc, v1, v2);

Functions
template<Elem E>
constexpr auto	aie::abs (const E &a) -> operand_base_type_t< E >
	Compute the absolute value of a value.
template<Vector Vec>
constexpr auto	aie::abs (const Vec &v) -> aie_dm_resource_remove_t< Vec >
	Compute the absolute value for each element in the given vector.
template<typename TR = int32, ComplexVector Vec> requires (Utils::is_one_of_v<TR, int32, int16>)
constexpr auto	aie::abs_square (const Vec &v, int shift=0)
	Compute the absolute square of each element in the given complex vector.
template<unsigned Elems> requires (arch::is(arch::AIE) || (arch::is(arch::AIE_ML) && __AIE_API_COMPLEX_FP32_EMULATION__ == 1))
constexpr vector< float, Elems >	aie::abs_square (const vector< cfloat, Elems > &v)
	Compute the absolute square of each element in the given complex vector.
template<AccumOrOp Acc, Vector Vec> requires (Acc::size() == Vec::size())
auto	aie::add (const Acc &acc, const Vec &v) -> result_type_t< Acc >
	Returns an accumulator with the element-wise addition of the input accumulator and vector.
template<AccumOrOp Acc, Elem E>
auto	aie::add (const Acc &acc, E a) -> result_type_t< Acc >
	Returns an accumulator with the addition of a value to all the elements of the input vector.
template<AccumOrOp Acc1, AccumOrOp Acc2> requires (arch::is(arch::AIE_ML) && is_same_accum_v<result_type_t<Acc1>, result_type_t<Acc2>>)
auto	aie::add (const Acc1 &acc1, const Acc2 &acc2) -> result_type_t< Acc1 >
	Returns an accumulator with the element-wise addition of the two input accumulators.
template<Vector Vec, Elem E> requires (is_valid_elem_op_v<typename Vec::value_type, E>)
auto	aie::add (const Vec &v, E a) -> aie_dm_resource_remove_t< Vec >
	Returns a vector with the addition of a value to all the elements of the input vector.
template<Vector Vec1, Vector Vec2> requires (is_same_vector_v<Vec1, Vec2>)
auto	aie::add (const Vec1 &v1, const Vec2 &v2) -> aie_dm_resource_remove_t< Vec1 >
	Returns a vector with the element-wise addition of the two input vectors.
template<Elem E, Vector Vec> requires (is_valid_elem_op_v<E, typename Vec::value_type>)
auto	aie::add (E a, const Vec &v) -> aie_dm_resource_remove_t< Vec >
	Returns a vector with the addition of a value to all the elements of the input vector.
template<ComplexElem E>
constexpr auto	aie::conj (const E &a) -> operand_base_type_t< E >
	Compute the conjugate in the given complex value.
template<ComplexVector Vec>
auto	aie::conj (const Vec &v) -> aie_dm_resource_remove_t< Vec >
	Compute the conjugate for each element in the given vector of complex elements.
template<ComplexVector Vec, ComplexElem E> requires (arch::is(arch::AIE) || (arch::is(arch::AIE_ML) && __AIE_API_COMPLEX_FP32_EMULATION__ == 1))
auto	aie::div (const Vec &a, const E &b) -> accum< detail::accum_tag_for_mul_types< typename Vec::value_type, detail::utils::get_complex_component_type_t< E > >, Vec::size()>
	Returns the quotients of the element-wise division of a complex vector 'a' by a complex scalar 'b'.
template<ComplexVector Vec> requires (arch::is(arch::AIE) || (arch::is(arch::AIE_ML) && __AIE_API_COMPLEX_FP32_EMULATION__ == 1))
auto	aie::div (const Vec &a, const Vec &b) -> accum< detail::accum_tag_for_mul_types< typename Vec::value_type, detail::utils::get_complex_component_type_t< typename Vec::value_type > >, Vec::size()>
	Returns the quotients of the element-wise division of two complex vectors 'a' and 'b'.
template<Vector Vec, RealElem E> requires ( detail::is_floating_point_v<typename Vec::value_type> && detail::is_valid_element_type_v<E> && detail::is_floating_point_v<E>)
auto	aie::div (const Vec &a, E b) -> accum< detail::accum_tag_for_mul_types< typename Vec::value_type, E >, Vec::size()>
	Returns the quotients of the element-wise division of each value of the first input vector by a scalar.
template<Vector Vec1, RealVector Vec2> requires (Vec1::size() == Vec2::size() && detail::is_floating_point_v<typename Vec2::value_type>)
auto	aie::div (const Vec1 &a, const Vec2 &b) -> accum< detail::accum_tag_for_mul_types< typename Vec1::value_type, typename Vec2::value_type >, Vec1::size()>
	Returns the quotients of the element-wise division of each value of the first input vector by the corresponding element in the second input vector.
template<AccumOrOp Acc, VectorOrOp Vec, ElemOrOp E> requires (is_valid_size_v<Acc, Vec, E> && is_valid_mul_op_v<typename Vec::value_type, E>)
constexpr auto	aie::mac (const Acc &acc, const Vec &v, E a) -> operand_base_type_t< Acc >
	Returns an accumulator with the element-wise multiply-add of input vector, value and accumulator.
template<AccumOrOp Acc, VectorOrOp Vec1, VectorOrOp Vec2> requires (is_valid_size_v<Acc, Vec1, Vec2> && is_valid_mul_op_v<typename Vec1::value_type, typename Vec2::value_type>)
constexpr auto	aie::mac (const Acc &acc, const Vec1 &v1, const Vec2 &v2) -> operand_base_type_t< Acc >
	Returns an accumulator with the element-wise multiply-add of the two input vectors and accumulator.
template<AccumOrOp Acc, ElemOrOp E, VectorOrOp Vec> requires (is_valid_size_v<Acc, E, Vec> && is_valid_mul_op_v<E, typename Vec::value_type>)
constexpr auto	aie::mac (const Acc &acc, E a, const Vec &v) -> operand_base_type_t< Acc >
	Returns an accumulator with the element-wise multiply-add of value, input vector and accumulator.
template<AccumOrOp Acc, Vector Vec> requires (Vec::size() == Acc::size())
constexpr auto	aie::mac_square (const Acc &acc, const Vec &v)
	Returns an accumulator with the addition or subtraction of the given accumulator and the element-wise square of the input vector.
template<Accum Acc, VectorOrOp Vec, ElemOrOp E> requires (is_valid_size_v<Acc, Vec, E> && is_valid_mul_op_v<typename Vec::value_type, E>)
constexpr auto	aie::msc (const Acc &acc, const Vec &v, E a) -> aie_dm_resource_remove_t< Acc >
	Returns an accumulator with the element-wise multiply-add of input vector, value and accumulator.
template<Accum Acc, VectorOrOp Vec1, VectorOrOp Vec2> requires (is_valid_size_v<Acc, Vec1, Vec2> && is_valid_mul_op_v<typename Vec1::value_type, typename Vec2::value_type>)
constexpr auto	aie::msc (const Acc &acc, const Vec1 &v1, const Vec2 &v2) -> aie_dm_resource_remove_t< Acc >
	Returns an accumulator with the element-wise multiply-add of the two input vectors and accumulator.
template<Accum Acc, ElemOrOp E, VectorOrOp Vec> requires (is_valid_size_v<Acc, E, Vec> && is_valid_mul_op_v<E, typename Vec::value_type>)
constexpr auto	aie::msc (const Acc &acc, E a, const Vec &v) -> aie_dm_resource_remove_t< Acc >
	Returns an accumulator with the element-wise multiply-add of value, input vector and accumulator.
template<Accum Acc, Vector Vec> requires (Vec::size() == Acc::size())
constexpr auto	aie::msc_square (const Acc &acc, const Vec &v)
	Returns an accumulator with the subtraction of the given accumulator and the element-wise square of the input vector.
template<VectorOrOp Vec, ElemOrOp E> requires (is_valid_size_v<E, Vec> && is_valid_mul_op_v<typename Vec::value_type, E>)
constexpr auto	aie::mul (const Vec &v, E a)
	Returns an accumulator with the element-wise multiplication of a value and all the elements of a vector.
template<AccumElemBaseType AccumTag, VectorOrOp Vec, ElemOrOp E> requires (is_valid_size_v<Vec, E> && is_valid_mul_op_v<typename Vec::value_type, E>)
constexpr auto	aie::mul (const Vec &v, E a) -> accum< AccumTag, Vec::size()>
	Returns an accumulator of the requested type with the element-wise multiplication of a value and all the elements of the input vector.
template<VectorOrOp Vec1, VectorOrOp Vec2> requires (is_valid_size_v<Vec1, Vec2> && is_valid_mul_op_v<typename Vec1::value_type, typename Vec2::value_type>)
constexpr auto	aie::mul (const Vec1 &v1, const Vec2 &v2)
	Returns an accumulator with the element-wise multiplication of the two input vectors.
template<AccumElemBaseType AccumTag, VectorOrOp Vec1, VectorOrOp Vec2> requires (is_valid_size_v<Vec1, Vec2> && is_valid_mul_op_v<typename Vec1::value_type, typename Vec2::value_type>)
constexpr auto	aie::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 vectors.
template<ElemOrOp E, VectorOrOp Vec> requires (is_valid_size_v<E, Vec> && is_valid_mul_op_v<E, typename Vec::value_type>)
constexpr auto	aie::mul (E a, const Vec &v)
	Returns an accumulator with the element-wise multiplication of a value and all the elements of a vector.
template<AccumElemBaseType AccumTag, ElemOrOp E, VectorOrOp Vec> requires (is_valid_size_v<Vec, E> && is_valid_mul_op_v<E, typename Vec::value_type>)
constexpr auto	aie::mul (E a, const Vec &v) -> accum< AccumTag, Vec::size()>
	Returns an accumulator of the requested type with the element-wise multiplication of a value and all the elements of the input vector.
template<Vector Vec>
constexpr auto	aie::mul_square (const Vec &v)
	Returns an accumulator with the element-wise square of the input vector.
template<AccumElemBaseType AccumTag, Vector Vec>
constexpr auto	aie::mul_square (const Vec &v) -> accum< AccumTag, Vec::size()>
	Returns an accumulator of the requested type with the element-wise square of the input vector.
template<Elem E>
constexpr auto	aie::neg (const E &a) -> operand_base_type_t< E >
	For values with signed types, return the input value with the sign flipped.
template<Vector Vec>
constexpr auto	aie::neg (const Vec &v) -> aie_dm_resource_remove_t< Vec >
	For vectors with signed types, return a vector whose elements are the same as in the given vector but with the sign flipped.
template<VectorOrOp Vec, ElemOrOp E> requires (is_valid_mul_op_v<typename Vec::value_type, E>)
constexpr auto	aie::negmul (const Vec &v, E a)
	Returns an accumulator with the negate of the element-wise multiplication of all the elements of the input vector and a value.
template<AccumElemBaseType AccumTag, VectorOrOp Vec, ElemOrOp E> requires (is_valid_mul_op_v<typename Vec::value_type, E>)
constexpr auto	aie::negmul (const Vec &v, E a) -> accum< AccumTag, Vec::size()>
	Returns an accumulator of the requested type with the negate of the element-wise multiplication of all the elements of the input vector and a value.
template<VectorOrOp Vec1, VectorOrOp Vec2> requires ((Vec1::size() == Vec2::size()) && is_valid_mul_op_v<typename Vec1::value_type, typename Vec2::value_type>)
constexpr auto	aie::negmul (const Vec1 &v1, const Vec2 &v2)
	Returns an accumulator with the negate of the element-wise multiplication of the two input vectors.
template<AccumElemBaseType AccumTag, VectorOrOp Vec1, VectorOrOp Vec2> requires (Vec1::size() == Vec2::size() && is_valid_mul_op_v<typename Vec1::value_type, typename Vec2::value_type>)
constexpr auto	aie::negmul (const Vec1 &v1, const Vec2 &v2) -> accum< AccumTag, Vec1::size()>
	Returns an accumulator of the requested type with the negate of the element-wise multiplication of the two input vectors.
template<ElemOrOp E, VectorOrOp Vec> requires (is_valid_mul_op_v<E, typename Vec::value_type>)
constexpr auto	aie::negmul (E a, const Vec &v)
	Returns an accumulator with the negate of the element-wise multiplication of a value and all the elements of the input vector.
template<AccumElemBaseType AccumTag, ElemOrOp E, VectorOrOp Vec> requires (is_valid_mul_op_v<E, typename Vec::value_type>)
constexpr auto	aie::negmul (E a, const Vec &v) -> accum< AccumTag, Vec::size()>
	Returns an accumulator of the requested type with the negate of the element-wise multiplication of a value and all the elements of the input vector.
template<Vector Vec, Vector... Others>
auto	aie::reduce_add_v (const Vec &v, const Others &... others) -> aie_dm_resource_remove_t< Vec >
	Returns the sums of the elements in the input vectors.
template<AccumOrOp Acc, Vector Vec> requires (Acc::size() == Vec::size())
constexpr auto	aie::sub (const Acc &acc, const Vec &v) -> result_type_t< Acc >
	Returns an accumulator with the element-wise subtraction of the input accumulator and vector.
template<AccumOrOp Acc, Elem E>
auto	aie::sub (const Acc &acc, E a) -> result_type_t< Acc >
	Returns an accumulator with the subtraction of all the elements of the input accumulator and a value.
template<AccumOrOp Acc1, AccumOrOp Acc2> requires (arch::is(arch::AIE_ML) && is_same_accum_v<result_type_t<Acc1>, result_type_t<Acc2>>)
auto	aie::sub (const Acc1 &acc1, const Acc2 &acc2) -> result_type_t< Acc1 >
	Returns an accumulator with the element-wise subtraction of the two input accumulators.
template<Vector Vec, Elem E> requires (is_valid_elem_op_v<E, typename Vec::value_type>)
auto	aie::sub (const Vec &v, E a) -> aie_dm_resource_remove_t< Vec >
	Returns a vector with the subtraction of the elements of the input vector and a value.
template<Vector Vec1, Vector Vec2> requires (is_same_vector_v<Vec1, Vec2>)
auto	aie::sub (const Vec1 &v1, const Vec2 &v2) -> aie_dm_resource_remove_t< Vec1 >
	Returns a vector with the element-wise subtraction of the two input vectors.
template<Elem E, Vector Vec> requires (is_valid_elem_op_v<E, typename Vec::value_type>)
auto	aie::sub (E a, const Vec &v) -> aie_dm_resource_remove_t< Vec >
	Returns a vector with the subtraction of a value and all the elements of the input vector.

Function Documentation

abs() [1/2]

template<Elem E>

constexpr auto aie::abs ( const E &  a ) -> operand_base_type_t<E>

constexpr

Compute the absolute value of a value.

Parameters

a	Input value

abs() [2/2]

template<Vector Vec>

constexpr auto aie::abs ( const Vec &  v ) -> aie_dm_resource_remove_t<Vec>

constexpr

Compute the absolute value for each element in the given vector.

Parameters

v	Input vector

abs_square() [1/2]

template<typename TR = int32, ComplexVector Vec>
requires (Utils::is_one_of_v<TR, int32, int16>)

constexpr auto aie::abs_square ( const Vec &  v, int  shift = 0  )

constexpr

Compute the absolute square of each element in the given complex vector.

For a vector of N complex elements, returns a vector of N integers

for (unsigned i = 0; i < Elems; ++i)
    out[i] = v[i].real² + v[i].imag²;

Parameters

v	Input complex vector
shift	Optional parameter to control the shift value used in the srs internally

Template Parameters

TR	Type of returned integer vector, default to int32 but can be set to int16

abs_square() [2/2]

template<unsigned Elems>
requires (arch::is(arch::AIE) || (arch::is(arch::AIE_ML) && __AIE_API_COMPLEX_FP32_EMULATION__ == 1))

constexpr vector< float, Elems > aie::abs_square ( const vector< cfloat, Elems > &  v )

constexpr

Compute the absolute square of each element in the given complex vector.

For a vector of N complex elements, returns a vector of N integers

for (unsigned i = 0; i < Elems; ++i)
    out[i] = v[i].real² + v[i].imag²;

Parameters

v	Input complex vector

add() [1/6]

template<AccumOrOp Acc, Vector Vec>
requires (Acc::size() == Vec::size())

auto aie::add	(	const Acc &	acc,
		const Vec &	v
	)		-> result_type_t<Acc>

Returns an accumulator with the element-wise addition of the input accumulator and vector.

for (unsigned i = 0; i < Elems; ++i)
    out[i] = acc[i] + v[i];

Parameters

acc	Input accumulator
v	Input vector

add() [2/6]

template<AccumOrOp Acc, Elem E>

auto aie::add	(	const Acc &	acc,
		E	a
	)		-> result_type_t<Acc>

Returns an accumulator with the addition of a value to all the elements of the input vector.

for (unsigned i = 0; i < Elems; ++i)
    out[i] = acc[i] + a;

Parameters

acc	Input accumulator
a	Value. The type must meet Elem.

add() [3/6]

template<AccumOrOp Acc1, AccumOrOp Acc2>
requires (arch::is(arch::AIE_ML) && is_same_accum_v<result_type_t<Acc1>, result_type_t<Acc2>>)

auto aie::add	(	const Acc1 &	acc1,
		const Acc2 &	acc2
	)		-> result_type_t<Acc1>

Returns an accumulator with the element-wise addition of the two input accumulators.

for (unsigned i = 0; i < Elems; ++i)
    out[i] = acc1[i] + acc2[i];

Parameters

acc1	First input accumulator. The type must meet AccumOrOp.
acc2	Second input accumulator. The type must meet AccumOrOp.

add() [4/6]

template<Vector Vec, Elem E>
requires (is_valid_elem_op_v<typename Vec::value_type, E>)

auto aie::add	(	const Vec &	v,
		E	a
	)		-> aie_dm_resource_remove_t<Vec>

Returns a vector with the addition of a value to all the elements of the input vector.

The type of the value and the type of the vector elements must be the same.

for (unsigned i = 0; i < Elems; ++i)
    out[i] = v[i] + a;

Parameters

v	Input vector. The type must meet Vector.
a	Value. The type must meet Elem.

add() [5/6]

template<Vector Vec1, Vector Vec2>
requires (is_same_vector_v<Vec1, Vec2>)

auto aie::add	(	const Vec1 &	v1,
		const Vec2 &	v2
	)		-> aie_dm_resource_remove_t<Vec1>

Returns a vector with the element-wise addition of the two input vectors.

The vectors must have the same type and size.

for (unsigned i = 0; i < Elems; ++i)
    out[i] = v1[i] + v2[i];

Parameters

v1	First input vector. The type must meet Vector.
v2	Second input vector. The type must meet Vector.

add() [6/6]

template<Elem E, Vector Vec>
requires (is_valid_elem_op_v<E, typename Vec::value_type>)

auto aie::add	(	E	a,
		const Vec &	v
	)		-> aie_dm_resource_remove_t<Vec>

Returns a vector with the addition of a value to all the elements of the input vector.

The type of the value and the type of the vector elements must be the same.

for (unsigned i = 0; i < Elems; ++i)
    out[i] = a + v[i];

Parameters

a	Value. The type must meet Elem.
v	Input vector. The type must meet Vector.

conj() [1/2]

template<ComplexElem E>

constexpr auto aie::conj ( const E &  a ) -> operand_base_type_t<E>

constexpr

Compute the conjugate in the given complex value.

Parameters

a	Input value

conj() [2/2]

template<ComplexVector Vec>

auto aie::conj

(

const Vec &

v

)

-> aie_dm_resource_remove_t<Vec>

Compute the conjugate for each element in the given vector of complex elements.

Parameters

v	Input vector

div() [1/4]

template<ComplexVector Vec, ComplexElem E>
requires (arch::is(arch::AIE) || (arch::is(arch::AIE_ML) && __AIE_API_COMPLEX_FP32_EMULATION__ == 1))

auto aie::div	(	const Vec &	a,
		const E &	b
	)		-> accum<detail::accum_tag_for_mul_types<typename Vec::value_type, detail::utils::get_complex_component_type_t<E>>, Vec::size()>

Returns the quotients of the element-wise division of a complex vector 'a' by a complex scalar 'b'.

for (unsigned i = 0; i < Elems; ++i)
    out[i] = a[i] / b;

Parameters

a	Vector of dividends. The type must meet ComplexVector.
b	Divisor. The type must meet ComplexElem.

div() [2/4]

template<ComplexVector Vec>
requires (arch::is(arch::AIE) || (arch::is(arch::AIE_ML) && __AIE_API_COMPLEX_FP32_EMULATION__ == 1))

auto aie::div	(	const Vec &	a,
		const Vec &	b
	)		-> accum< detail::accum_tag_for_mul_types<typename Vec::value_type, detail::utils::get_complex_component_type_t<typename Vec::value_type>>, Vec::size()>

Returns the quotients of the element-wise division of two complex vectors 'a' and 'b'.

for (unsigned i = 0; i < Elems; ++i)
    out[i] = a[i] / b[i];

Complex division is achieved by multiplying both dividend and divisor by the conjugate of the divisor:

a = A + Bi; b = C + Di
a/b = (A + Bi) / (C + Di) = ((A + Bi) * (C - Di)) / ((C + Di) * (C - Di))
    = ((A + Bi) * (C - Di)) /  (C^2 + D^2)

Parameters

a	Vector of dividends. The type must meet ComplexVector.
b	Vector of divisors. The type must meet ComplexVector.

div() [3/4]

template<Vector Vec, RealElem E>
requires ( detail::is_floating_point_v<typename Vec::value_type> && detail::is_valid_element_type_v<E> && detail::is_floating_point_v<E>)

auto aie::div	(	const Vec &	a,
		E	b
	)		-> accum<detail::accum_tag_for_mul_types<typename Vec::value_type, E>, Vec::size()>

Returns the quotients of the element-wise division of each value of the first input vector by a scalar.

for (unsigned i = 0; i < Elems; ++i)
    out[i] = a[i] / b;

Parameters

a	Vector of dividends. The type must meet Vector.
b	Divisor. The type must meet RealElem.

div() [4/4]

template<Vector Vec1, RealVector Vec2>
requires (Vec1::size() == Vec2::size() && detail::is_floating_point_v<typename Vec2::value_type>)

auto aie::div	(	const Vec1 &	a,
		const Vec2 &	b
	)		-> accum<detail::accum_tag_for_mul_types<typename Vec1::value_type, typename Vec2::value_type>, Vec1::size()>

Returns the quotients of the element-wise division of each value of the first input vector by the corresponding element in the second input vector.

for (unsigned i = 0; i < Elems; ++i)
    out[i] = a[i] / b[i];

Parameters

a	Vector of dividends. The type must meet Vector.
b	Vector of divisors. The type must meet RealVector.

mac() [1/3]

template<AccumOrOp Acc, VectorOrOp Vec, ElemOrOp E>
requires (is_valid_size_v<Acc, Vec, E> && is_valid_mul_op_v<typename Vec::value_type, E>)

constexpr auto aie::mac ( const Acc &  acc, const Vec &  v, E  a  ) -> operand_base_type_t<Acc>

constexpr

Returns an accumulator with the element-wise multiply-add of input vector, value and accumulator.

The vector and the accumulator must have the same size and the types of all operands must be compatible. Operands can also be lazy operations (see Lazy Operations). On some AIE architectures certain operations can be collapsed with the multiplication into a single instruction.

for (unsigned i = 0; i < Elems; ++i)
    out[i] = acc[i] + op1(v[i]) * op2(a);

Parameters

acc	Accumulator to which the result of the multiplication is added (or subtracted). The type must meet AccumOrOp.
v	Input vector. The type must meet VectorOrOp.
a	Input value. The type must meet ElemOrOp.

mac() [2/3]

template<AccumOrOp Acc, VectorOrOp Vec1, VectorOrOp Vec2>
requires (is_valid_size_v<Acc, Vec1, Vec2> && is_valid_mul_op_v<typename Vec1::value_type, typename Vec2::value_type>)

constexpr auto aie::mac ( const Acc &  acc, const Vec1 &  v1, const Vec2 &  v2  ) -> operand_base_type_t<Acc>

constexpr

Returns an accumulator with the element-wise multiply-add of the two input vectors and accumulator.

The vectors and the accumulator must have the same size and their types must be compatible. Operands can also be lazy operations (see Lazy Operations). On some AIE architectures certain operations can be collapsed with the multiplication into a single instruction.

for (unsigned i = 0; i < Elems; ++i)
    out[i] = acc[i] + op1(v1[i]) * op2(v2[i]);

Parameters

acc	Accumulator to which the result of the multiplication is added (or subtracted). The type must meet AccumOrOp.
v1	First input vector. The type must meet VectorOrOp.
v2	Second input vector. The type must meet VectorOrOp.

mac() [3/3]

template<AccumOrOp Acc, ElemOrOp E, VectorOrOp Vec>
requires (is_valid_size_v<Acc, E, Vec> && is_valid_mul_op_v<E, typename Vec::value_type>)

constexpr auto aie::mac ( const Acc &  acc, E  a, const Vec &  v  ) -> operand_base_type_t<Acc>

constexpr

Returns an accumulator with the element-wise multiply-add of value, input vector and accumulator.

The vector and the accumulator must have the same size and the types of all operands must be compatible. Operands can also be lazy operations (see Lazy Operations). On some AIE architectures certain operations can be collapsed with the multiplication into a single instruction.

for (unsigned i = 0; i < Elems; ++i)
    out[i] = acc[i] + op1(a) * op2(v[i]);

Parameters

acc	Accumulator to which the result of the multiplication is added (or subtracted). The type must meet AccumOrOp.
a	Input value. The type must meet ElemOrOp.
v	Input vector. The type must meet VectorOrOp.

mac_square()

template<AccumOrOp Acc, Vector Vec>
requires (Vec::size() == Acc::size())

constexpr auto aie::mac_square ( const Acc &  acc, const Vec &  v  )

constexpr

Returns an accumulator with the addition or subtraction of the given accumulator and the element-wise square of the input vector.

The vector and the accumulator must have the same size and their types must be compatible.

for (unsigned i = 0; i < Elems; ++i)
    out[i] = acc[i] + v[i] * v[i];

Parameters

acc	Input accumulator. The type must meet AccumOrOp.
v	Input vector. The type must meet Vector.

msc() [1/3]

template<Accum Acc, VectorOrOp Vec, ElemOrOp E>
requires (is_valid_size_v<Acc, Vec, E> && is_valid_mul_op_v<typename Vec::value_type, E>)

constexpr auto aie::msc ( const Acc &  acc, const Vec &  v, E  a  ) -> aie_dm_resource_remove_t<Acc>

constexpr

Returns an accumulator with the element-wise multiply-add of input vector, value and accumulator.

The vector and the accumulator must have the same size and their types must be compatible. Operands can also be lazy operations (see Lazy Operations). On some AIE architectures certain operations can be collapsed with the multiplication into a single instruction.

for (unsigned i = 0; i < Elems; ++i)
    out[i] = acc[i] - op1(v[i]) * op2(a);

Parameters

acc	Accumulator from which the result of the multiplication is subtracted. The type must meet Accum.
v	Input vector. The type must meet VectorOrOp.
a	Input value. The type must meet ElemOrOp.

msc() [2/3]

template<Accum Acc, VectorOrOp Vec1, VectorOrOp Vec2>
requires (is_valid_size_v<Acc, Vec1, Vec2> && is_valid_mul_op_v<typename Vec1::value_type, typename Vec2::value_type>)

constexpr auto aie::msc ( const Acc &  acc, const Vec1 &  v1, const Vec2 &  v2  ) -> aie_dm_resource_remove_t<Acc>

constexpr

Returns an accumulator with the element-wise multiply-add of the two input vectors and accumulator.

The vectors and the accumulator must have the same size and the types of all operands must be compatible. Operands can also be lazy operations (see Lazy Operations). On some AIE architectures certain operations can be collapsed with the multiplication into a single instruction.

for (unsigned i = 0; i < Elems; ++i)
    out[i] = acc[i] - op1(v1[i]) * op2(v2[i]);

Parameters

acc	Accumulator from which the result of the multiplication is subtracted. The type must meet Accum.
v1	First input vector. The type must meet VectorOrOp.
v2	Second input vector. The type must meet VectorOrOp.

msc() [3/3]

template<Accum Acc, ElemOrOp E, VectorOrOp Vec>
requires (is_valid_size_v<Acc, E, Vec> && is_valid_mul_op_v<E, typename Vec::value_type>)

constexpr auto aie::msc ( const Acc &  acc, E  a, const Vec &  v  ) -> aie_dm_resource_remove_t<Acc>

constexpr

Returns an accumulator with the element-wise multiply-add of value, input vector and accumulator.

The vector and the accumulator must have the same size and their types must be compatible. Operands can also be lazy operations (see Lazy Operations). On some AIE architectures certain operations can be collapsed with the multiplication into a single instruction.

for (unsigned i = 0; i < Elems; ++i)
    out[i] = acc[i] - op1(a) * op2(v[i]);

Parameters

acc	Accumulator from which the result of the multiplication is subtracted. The type must meet Accum.
a	Input value. The type must meet ElemOrOp.
v	Input vector. The type must meet VectorOrOp.

msc_square()

template<Accum Acc, Vector Vec>
requires (Vec::size() == Acc::size())

constexpr auto aie::msc_square ( const Acc &  acc, const Vec &  v  )

constexpr

Returns an accumulator with the subtraction of the given accumulator and the element-wise square of the input vector.

The vector and the accumulator must have the same size and their types must be compatible.

for (unsigned i = 0; i < Elems; ++i)
    out[i] = acc[i] - v[i] * v[i];

Parameters

acc	Input accumulator. The type must meet Accum.
v	Input vector. The type must meet Vector.

mul() [1/6]

template<VectorOrOp Vec, ElemOrOp E>
requires (is_valid_size_v<E, Vec> && is_valid_mul_op_v<typename Vec::value_type, E>)

constexpr auto aie::mul ( const Vec &  v, E  a  )

constexpr

Returns an accumulator with the element-wise multiplication of a value and all the elements of a vector.

The default accumulator accuracy is used of the input vector. The type of the value and the type of the vector elements must be compatible. Operands can also be lazy operations (see Lazy Operations). On some AIE architectures certain operations can be collapsed with the multiplication into a single instruction.

for (unsigned i = 0; i < Elems; ++i)
    out[i] = op1(v[i]) * op2(a);

Parameters

a	Value. The type must meet ElemOrOp.
v	Input vector. The type must meet VectorOrOp.

mul() [2/6]

template<AccumElemBaseType AccumTag, VectorOrOp Vec, ElemOrOp E>
requires (is_valid_size_v<Vec, E> && is_valid_mul_op_v<typename Vec::value_type, E>)

constexpr auto aie::mul ( const Vec &  v, E  a  ) -> accum<AccumTag, Vec::size()>

constexpr

Returns an accumulator of the requested type with the element-wise multiplication of a value and all the elements of the input vector.

The type of the value and the type of the vector elements must be compatible. Operands can also be lazy operations (see Lazy Operations). On some AIE architectures certain operations can be collapsed with the multiplication into a single instruction.

for (unsigned i = 0; i < Elems; ++i)
    out[i] = op1(v[i]) * op2(a);

Template Parameters

AccumTag

Accumulator tag that specifies the required accumulation bits. The class must be compatible with the vector types (real/complex).

Parameters

v	Input vector. The type must meet VectorOrOp.
a	Value. The type must meet ElemOrOp.

mul() [3/6]

template<VectorOrOp Vec1, VectorOrOp Vec2>
requires (is_valid_size_v<Vec1, Vec2> && is_valid_mul_op_v<typename Vec1::value_type, typename Vec2::value_type>)

constexpr auto aie::mul ( const Vec1 &  v1, const Vec2 &  v2  )

constexpr

Returns an accumulator with the element-wise multiplication of the two input vectors.

The default accumulator accuracy is used. The vectors must have the same size and compatible types. Operands can also be lazy operations (see Lazy Operations). On some AIE architectures certain operations can be collapsed with the multiplication into a single instruction.

for (unsigned i = 0; i < Elems; ++i)
    out[i] = op1(v1[i]) * op2(v2[i]);

Parameters

v1	First input vector. The type must meet VectorOrOp.
v2	Second input vector. The type must meet VectorOrOp.

mul() [4/6]

template<AccumElemBaseType AccumTag, VectorOrOp Vec1, VectorOrOp Vec2>
requires (is_valid_size_v<Vec1, Vec2> && is_valid_mul_op_v<typename Vec1::value_type, typename Vec2::value_type>)

constexpr auto aie::mul ( const Vec1 &  v1, const Vec2 &  v2  ) -> accum<AccumTag, Vec1::size()>

constexpr

Returns an accumulator of the requested type with the element-wise multiplication of the two input vectors.

The vectors must have the same size and compatible types. Vectors can also be lazy operations (see Lazy Operations). On some AIE architectures certain operations can be collapsed with the multiplication into a single instruction.

for (unsigned i = 0; i < Elems; ++i)
    out[i] = op1(v1[i]) * op2(v2[i]);

Template Parameters

AccumTag

Accumulator tag that specifies the required accumulation bits. The class must be compatible with the vector types (real/complex).

Parameters

v1	First input vector. The type must meet VectorOrOp.
v2	Second input vector. The type must meet VectorOrOp.

mul() [5/6]

template<ElemOrOp E, VectorOrOp Vec>
requires (is_valid_size_v<E, Vec> && is_valid_mul_op_v<E, typename Vec::value_type>)

constexpr auto aie::mul ( E  a, const Vec &  v  )

constexpr

Returns an accumulator with the element-wise multiplication of a value and all the elements of a vector.

The default accumulator accuracy is used of the input vector. The type of the value and the type of the vector elements must be compatible. Operands can also be lazy operations (see Lazy Operations). On some AIE architectures certain operations can be collapsed with the multiplication into a single instruction.

for (unsigned i = 0; i < Elems; ++i)
    out[i] = op1(a) * op2(v[i]);

Parameters

a	Value. The type must meet ElemOrOp.
v	Input vector. The type must meet VectorOrOp.

mul() [6/6]

template<AccumElemBaseType AccumTag, ElemOrOp E, VectorOrOp Vec>
requires (is_valid_size_v<Vec, E> && is_valid_mul_op_v<E, typename Vec::value_type>)

constexpr auto aie::mul ( E  a, const Vec &  v  ) -> accum<AccumTag, Vec::size()>

constexpr

Returns an accumulator of the requested type with the element-wise multiplication of a value and all the elements of the input vector.

The type of the value and the type of the vector elements must be compatible. Operands can also be lazy operations (see Lazy Operations). On some AIE architectures certain operations can be collapsed with the multiplication into a single instruction.

for (unsigned i = 0; i < Elems; ++i)
    out[i] = op1(a) * op2(v[i]);

Template Parameters

AccumTag

Accumulator tag that specifies the required accumulation bits. The class must be compatible with the vector types (real/complex).

Parameters

a	Value. The type must meet ElemOrOp.
v	Input vector. The type must meet VectorOrOp.

mul_square() [1/2]

template<Vector Vec>

constexpr auto aie::mul_square ( const Vec &  v )

constexpr

Returns an accumulator with the element-wise square of the input vector.

The default accumulator accuracy is used.

for (unsigned i = 0; i < Elems; ++i)
    out[i] = v[i] * v[i];

Parameters

v	Input vector. The type must meet Vector.

mul_square() [2/2]

template<AccumElemBaseType AccumTag, Vector Vec>

constexpr auto aie::mul_square ( const Vec &  v ) -> accum<AccumTag, Vec::size()>

constexpr

Returns an accumulator of the requested type with the element-wise square of the input vector.

for (unsigned i = 0; i < Elems; ++i)
    out[i] = v[i] * v[i];

Template Parameters

AccumTag

Accumulator tag that specifies the required accumulation bits. The class must be compatible with the vector types (real/complex).

Parameters

v	Input vector. The type must meet Vector.

neg() [1/2]

template<Elem E>

constexpr auto aie::neg ( const E &  a ) -> operand_base_type_t<E>

constexpr

For values with signed types, return the input value with the sign flipped.

If the input type is unsigned, the input value is returned.

Parameters

a	Input value

neg() [2/2]

template<Vector Vec>

constexpr auto aie::neg ( const Vec &  v ) -> aie_dm_resource_remove_t<Vec>

constexpr

For vectors with signed types, return a vector whose elements are the same as in the given vector but with the sign flipped.

If the input type is unsigned, the input vector is returned.

Parameters

v	Input vector

negmul() [1/6]

template<VectorOrOp Vec, ElemOrOp E>
requires (is_valid_mul_op_v<typename Vec::value_type, E>)

constexpr auto aie::negmul ( const Vec &  v, E  a  )

constexpr

Returns an accumulator with the negate of the element-wise multiplication of all the elements of the input vector and a value.

The default accumulator accuracy is used. The vectors must have the same size and compatible types. Operands can also be lazy operations (see Lazy Operations). On some AIE architectures certain operations can be collapsed with the multiplication into a single instruction.

for (unsigned i = 0; i < Elems; ++i)
    out[i] = -(op1(v[i]) * op2(a)));

Parameters

v	Input vector. The type must meet VectorOrOp.
a	Value. The type must meet ElemOrOp.

negmul() [2/6]

template<AccumElemBaseType AccumTag, VectorOrOp Vec, ElemOrOp E>
requires (is_valid_mul_op_v<typename Vec::value_type, E>)

constexpr auto aie::negmul ( const Vec &  v, E  a  ) -> accum<AccumTag, Vec::size()>

constexpr

Returns an accumulator of the requested type with the negate of the element-wise multiplication of all the elements of the input vector and a value.

The type of the value and the type of the vector elements must be compatible. The input arguments can also be lazy operations (see Lazy Operations). On some AIE architectures certain operations can be collapsed with the multiplication into a single instruction.

for (unsigned i = 0; i < Elems; ++i)
    out[i] = -(op1(v[i]) * op2(a));

Template Parameters

AccumTag

Accumulator tag that specifies the required accumulation bits. The class must be compatible with the vector types (real/complex).

Parameters

v	Input vector. The type must meet VectorOrOp.
a	Value. The type must meet ElemOrOp.

negmul() [3/6]

template<VectorOrOp Vec1, VectorOrOp Vec2>
requires ((Vec1::size() == Vec2::size()) && is_valid_mul_op_v<typename Vec1::value_type, typename Vec2::value_type>)

constexpr auto aie::negmul ( const Vec1 &  v1, const Vec2 &  v2  )

constexpr

Returns an accumulator with the negate of the element-wise multiplication of the two input vectors.

The default accumulator accuracy is used. The vectors must have the same size and compatible types. Operands can also be lazy operations (see Lazy Operations). On some AIE architectures certain operations can be collapsed with the multiplication into a single instruction.

for (unsigned i = 0; i < Elems; ++i)
    out[i] = -(op1(v1[i]) * op2(v2[i]));

Parameters

v1	First input vector. The type must meet VectorOrOp.
v2	Second input vector. The type must meet VectorOrOp.

negmul() [4/6]

template<AccumElemBaseType AccumTag, VectorOrOp Vec1, VectorOrOp Vec2>
requires (Vec1::size() == Vec2::size() && is_valid_mul_op_v<typename Vec1::value_type, typename Vec2::value_type>)

constexpr auto aie::negmul ( const Vec1 &  v1, const Vec2 &  v2  ) -> accum<AccumTag, Vec1::size()>

constexpr

Returns an accumulator of the requested type with the negate of the element-wise multiplication of the two input vectors.

The vectors must have the same size and compatible types. Vectors can also be lazy operations (see Lazy Operations). On some AIE architectures certain operations can be collapsed with the multiplication into a single instruction.

for (unsigned i = 0; i < Elems; ++i)
    out[i] = -(op1(v1[i]) * op1(v2[i]));

Template Parameters

AccumTag

Accumulator tag that specifies the required accumulation bits. The class must be compatible with the vector types (real/complex).

Parameters

v1	First input vector. The type must meet VectorOrOp.
v2	Second input vector. The type must meet VectorOrOp.

negmul() [5/6]

template<ElemOrOp E, VectorOrOp Vec>
requires (is_valid_mul_op_v<E, typename Vec::value_type>)

constexpr auto aie::negmul ( E  a, const Vec &  v  )

constexpr

Returns an accumulator with the negate of the element-wise multiplication of a value and all the elements of the input vector.

The default accumulator accuracy is used. The vectors must have the same size and compatible types. Operands can also be lazy operations (see Lazy Operations). On some AIE architectures certain operations can be collapsed with the multiplication into a single instruction.

for (unsigned i = 0; i < Elems; ++i)
    out[i] = -(op1(a) * op2(v[i]));

Parameters

a	Value. The type must meet ElemOrOp.
v	Input vector. The type must meet VectorOrOp.

negmul() [6/6]

template<AccumElemBaseType AccumTag, ElemOrOp E, VectorOrOp Vec>
requires (is_valid_mul_op_v<E, typename Vec::value_type>)

constexpr auto aie::negmul ( E  a, const Vec &  v  ) -> accum<AccumTag, Vec::size()>

constexpr

Returns an accumulator of the requested type with the negate of the element-wise multiplication of a value and all the elements of the input vector.

The type of the value and the type of the vector elements must be compatible. The input arguments can also be lazy operations (see Lazy Operations). On some AIE architectures certain operations can be collapsed with the multiplication into a single instruction.

for (unsigned i = 0; i < Elems; ++i)
    out[i] = -(op1(a) * op2(v[i]));

Template Parameters

AccumTag

Accumulator tag that specifies the required accumulation bits. The class must be compatible with the vector types (real/complex).

Parameters

a	Value. The type must meet ElemOrOp.
v	Input vector. The type must meet VectorOrOp.

reduce_add_v()

template<Vector Vec, Vector... Others>

auto aie::reduce_add_v	(	const Vec &	v,
		const Others &...	others
	)		-> aie_dm_resource_remove_t<Vec>

Returns the sums of the elements in the input vectors.

The sum of each input vector is stored in an element of the output vector.

Parameters

v	First input vector. The type must meet Vector.
others	Remaining input vectors. The type must meet Vector.

sub() [1/6]

template<AccumOrOp Acc, Vector Vec>
requires (Acc::size() == Vec::size())

constexpr auto aie::sub ( const Acc &  acc, const Vec &  v  ) -> result_type_t<Acc>

constexpr

Returns an accumulator with the element-wise subtraction of the input accumulator and vector.

for (unsigned i = 0; i < Elems; ++i)
    out[i] = acc[i] - v[i];

Parameters

acc	Input accumulator (minuend)
v	Input vector (subtrahend)

sub() [2/6]

template<AccumOrOp Acc, Elem E>

auto aie::sub	(	const Acc &	acc,
		E	a
	)		-> result_type_t<Acc>

Returns an accumulator with the subtraction of all the elements of the input accumulator and a value.

for (unsigned i = 0; i < Elems; ++i)
    out[i] = acc[i] - a;

Parameters

acc	Input accumulator (minuend)
a	Value (subtrahend). The type must meet Elem.

sub() [3/6]

template<AccumOrOp Acc1, AccumOrOp Acc2>
requires (arch::is(arch::AIE_ML) && is_same_accum_v<result_type_t<Acc1>, result_type_t<Acc2>>)

auto aie::sub	(	const Acc1 &	acc1,
		const Acc2 &	acc2
	)		-> result_type_t<Acc1>

Returns an accumulator with the element-wise subtraction of the two input accumulators.

for (unsigned i = 0; i < Elems; ++i)
    out[i] = acc1[i] - acc2[i];

Parameters

acc1	First input accumulator. The type must meet AccumOrOp.
acc2	Second input accumulator. The type must meet AccumOrOp.

sub() [4/6]

template<Vector Vec, Elem E>
requires (is_valid_elem_op_v<E, typename Vec::value_type>)

auto aie::sub	(	const Vec &	v,
		E	a
	)		-> aie_dm_resource_remove_t<Vec>

Returns a vector with the subtraction of the elements of the input vector and a value.

The type of the value and the type of the vector elements must be the same.

for (unsigned i = 0; i < Elems; ++i)
    out[i] = v[i] - a;

Parameters

v	Input vector (minuend)
a	Value (subtrahend)

sub() [5/6]

template<Vector Vec1, Vector Vec2>
requires (is_same_vector_v<Vec1, Vec2>)

auto aie::sub	(	const Vec1 &	v1,
		const Vec2 &	v2
	)		-> aie_dm_resource_remove_t<Vec1>

Returns a vector with the element-wise subtraction of the two input vectors.

The vectors must have the same type and size.

for (unsigned i = 0; i < Elems; ++i)
    out[i] = v1[i] - v2[i];

Parameters

v1	First input vector (minuend). The type must meet Vector.
v2	Second input vector (subtrahend). The type must meet Vector.

sub() [6/6]

template<Elem E, Vector Vec>
requires (is_valid_elem_op_v<E, typename Vec::value_type>)

auto aie::sub	(	E	a,
		const Vec &	v
	)		-> aie_dm_resource_remove_t<Vec>

Returns a vector with the subtraction of a value and all the elements of the input vector.

The type of the value and the type of the vector elements must be the same.

for (unsigned i = 0; i < Elems; ++i)
    out[i] = a - v[i];

Parameters

a	Value (minuend). The type must meet Elem.
v	Input vector (subtrahend). The type must meet Vector.