Why use it

Byte and word population counts work directly on small bitsets, packed flags and Hamming-distance data. Bit selection turns positions within packed qwords into a predicate, avoiding a separate scalar extraction for each selected bit.

Operations

import native.x86.bitalg; supplies these operations through native::native. native.x86 and native also export them.

Operation	Input	Result	Lane counts
vpopcntb	simd<std::uint8_t,N,Arch>	Same vector type	16, 32, 64
vpopcntw	simd<std::uint16_t,N,Arch>	Same vector type	8, 16, 32
vpshufbitqmb	simd<std::uint64_t,Q,Arch>, simd<std::uint8_t,8*Q,Arch>	predicate<8*Q,Arch>	Q = 2, 4, 8

Population counts are in 0–8 or 0–16. Their ordinary forms take <Arch>(value); mask_NAME<Arch>(source, mask, value) merges inactive source lanes and maskz_NAME<Arch>(mask, value) clears them. Masks are predicate<N,Arch>.

vpshufbitqmb<Arch>(value, control) sets result bit i from bit control[i] & 63 of qword value[i / 8]. Each group of eight control bytes addresses its own qword. mask_vpshufbitqmb<Arch>(mask, value, control) clears inactive predicate bits. It has no merge-source or separate maskz_ form.

Caveats

Bit selection does not cross qword boundaries; control bits 6 and 7 are ignored. For full-vector byte permutations, use VBMI.

Runtime wrappers need AVX512F, AVX512BW and AVX512BITALG, plus AVX512VL for 128/256 bits. Use target_features<native::x86>("avx512bitalg"), adding avx512vl for shorter forms. BW is part of Clang's compiler prerequisites and mask operations. DQ and VPOPCNTDQ are independent. Scalar POPCNT does not establish BITALG support, nor does the general AVX-512 profile.

Admit the matching compiler target and OS AVX-512 state before entry. Feature-bearing overloads are constexpr with native runtime paths; weaker tags are consteval-only with complete SSE2, AVX or AVX512F storage for the chosen width. Results preserve their architecture tag.

See Intel's Software Developer's Manual and Clang's BITALG and VL BITALG headers.