DotProd

Collaboration diagram for DotProd:

Functions
template<isa< arm > Arch> requires (Arch.has(arm_feature::dotprod))
constexpr simd< std::int32_t, 2, Arch >	native::sdot (simd< std::int32_t, 2, Arch > accumulator, simd< std::int8_t, 8, Arch > lhs, simd< std::int8_t, 8, Arch > rhs) noexcept
	SDOT: four byte products per accumulator lane, modulo 2^32.
template<isa< arm > Arch, int Lane> requires (Arch.has(arm_feature::dotprod) && Lane >= 0 && Lane < 2)
constexpr simd< std::int32_t, 2, Arch >	native::sdot_lane (simd< std::int32_t, 2, Arch > accumulator, simd< std::int8_t, 8, Arch > lhs, simd< std::int8_t, 8, Arch > rhs) noexcept
	SDOT by element: broadcast rhs byte group Lane to every accumulator lane.
template<isa< arm > Arch, int Lane> requires (Arch.has(arm_feature::dotprod) && Lane >= 0 && Lane < 4)
constexpr simd< std::int32_t, 2, Arch >	native::sdot_lane (simd< std::int32_t, 2, Arch > accumulator, simd< std::int8_t, 8, Arch > lhs, simd< std::int8_t, 16, Arch > rhs) noexcept
	SDOT by element: broadcast rhs byte group Lane to every accumulator lane.
template<isa< arm > Arch> requires (Arch.has(arm_feature::dotprod))
constexpr simd< std::int32_t, 4, Arch >	native::sdot (simd< std::int32_t, 4, Arch > accumulator, simd< std::int8_t, 16, Arch > lhs, simd< std::int8_t, 16, Arch > rhs) noexcept
	SDOT: four byte products per accumulator lane, modulo 2^32.
template<isa< arm > Arch, int Lane> requires (Arch.has(arm_feature::dotprod) && Lane >= 0 && Lane < 2)
constexpr simd< std::int32_t, 4, Arch >	native::sdot_lane (simd< std::int32_t, 4, Arch > accumulator, simd< std::int8_t, 16, Arch > lhs, simd< std::int8_t, 8, Arch > rhs) noexcept
	SDOT by element: broadcast rhs byte group Lane to every accumulator lane.
template<isa< arm > Arch, int Lane> requires (Arch.has(arm_feature::dotprod) && Lane >= 0 && Lane < 4)
constexpr simd< std::int32_t, 4, Arch >	native::sdot_lane (simd< std::int32_t, 4, Arch > accumulator, simd< std::int8_t, 16, Arch > lhs, simd< std::int8_t, 16, Arch > rhs) noexcept
	SDOT by element: broadcast rhs byte group Lane to every accumulator lane.
template<isa< arm > Arch> requires (Arch.has(arm_feature::dotprod))
constexpr simd< std::uint32_t, 2, Arch >	native::udot (simd< std::uint32_t, 2, Arch > accumulator, simd< std::uint8_t, 8, Arch > lhs, simd< std::uint8_t, 8, Arch > rhs) noexcept
	UDOT: four byte products per accumulator lane, modulo 2^32.
template<isa< arm > Arch, int Lane> requires (Arch.has(arm_feature::dotprod) && Lane >= 0 && Lane < 2)
constexpr simd< std::uint32_t, 2, Arch >	native::udot_lane (simd< std::uint32_t, 2, Arch > accumulator, simd< std::uint8_t, 8, Arch > lhs, simd< std::uint8_t, 8, Arch > rhs) noexcept
	UDOT by element: broadcast rhs byte group Lane to every accumulator lane.
template<isa< arm > Arch, int Lane> requires (Arch.has(arm_feature::dotprod) && Lane >= 0 && Lane < 4)
constexpr simd< std::uint32_t, 2, Arch >	native::udot_lane (simd< std::uint32_t, 2, Arch > accumulator, simd< std::uint8_t, 8, Arch > lhs, simd< std::uint8_t, 16, Arch > rhs) noexcept
	UDOT by element: broadcast rhs byte group Lane to every accumulator lane.
template<isa< arm > Arch> requires (Arch.has(arm_feature::dotprod))
constexpr simd< std::uint32_t, 4, Arch >	native::udot (simd< std::uint32_t, 4, Arch > accumulator, simd< std::uint8_t, 16, Arch > lhs, simd< std::uint8_t, 16, Arch > rhs) noexcept
	UDOT: four byte products per accumulator lane, modulo 2^32.
template<isa< arm > Arch, int Lane> requires (Arch.has(arm_feature::dotprod) && Lane >= 0 && Lane < 2)
constexpr simd< std::uint32_t, 4, Arch >	native::udot_lane (simd< std::uint32_t, 4, Arch > accumulator, simd< std::uint8_t, 16, Arch > lhs, simd< std::uint8_t, 8, Arch > rhs) noexcept
	UDOT by element: broadcast rhs byte group Lane to every accumulator lane.
template<isa< arm > Arch, int Lane> requires (Arch.has(arm_feature::dotprod) && Lane >= 0 && Lane < 4)
constexpr simd< std::uint32_t, 4, Arch >	native::udot_lane (simd< std::uint32_t, 4, Arch > accumulator, simd< std::uint8_t, 16, Arch > lhs, simd< std::uint8_t, 16, Arch > rhs) noexcept
	UDOT by element: broadcast rhs byte group Lane to every accumulator lane.

Detailed Description

Explicit ISA-gated AArch64 instructions. Admit the feature before entering a matching target scope; runtime calls provide no software fallback. Constant operands use exact integer semantics; feature-absent overloads are immediate-only and preserve the available SIMD storage shape. SDOT and UDOT accumulate modulo 2^32, including signed overflow.