Why use it

A bitset often needs one set bit at a time. blsi isolates it, blsr removes it, and tzcnt gives its position. BMI1 also extracts a field without a separate shift and mask. These are useful building blocks for sparse masks, packed records and bitset iteration.

Operations

import native.x86.bmi1; provides the following operations in native. native.x86 and native re-export the module. Operands and results are std::uint32_t or std::uint64_t; tzcnt also accepts std::uint16_t.

Operation	Result
andn<Arch>(a, b)	(~a) & b
bextr<Arch>(value, control)	Extract a field; start is control bits 7:0, length is bits 15:8
bextr<Arch>(value, start, length)	Extract using the low eight bits of each unsigned control
blsi<Arch>(value)	Isolate the lowest set bit; zero stays zero
blsmsk<Arch>(value)	Set bits through the lowest set bit, inclusive; zero gives all ones
blsr<Arch>(value)	Clear the lowest set bit; zero stays zero
tzcnt<Arch>(value)	Count trailing zeros; zero gives the operand width

bextr clips the field at the operand width and clears the remaining result bits. Zero length or a start at or beyond the width gives zero. Control bits above bit 15 are ignored. All functions are noexcept.

Caveats

Use target_features<native::x86>("bmi"), a matching compiler target and classify_isa before entry. BMI1 is independent of BMI2 and AVX and needs no vector OS state. The feature check matters even for tzcnt: its encoding can execute as BSF on older CPUs, with different zero-input behavior.

Arch is an isa<x86>. The module's scalar default is the provider's NATIVE_BASELINE; caller target attributes do not change it. Feature-bearing calls are constexpr with native runtime paths. Tags without BMI1 accept constant evaluation through consteval overloads, with no runtime fallback. The API exposes values, not arithmetic flags.

See Intel's BMI1 instruction entries and TZCNT entry.