Expand description
§SIMD Popcount Dispatch
Runtime CPU-feature dispatch for packed-bit popcount kernels. Supported backends: AVX-512, AVX2, ARM NEON, ARM SVE, RISC-V RVV.
Modules§
- avx2
- avx512
- neon
- rvv
- Scalar fallback for RISC-V Vector (RVV) targets.
Hardware RVV intrinsics are not yet implemented; all operations
use the portable scalar path from
super::mod.rs. - sve
- Scalar fallback for ARM SVE targets.
Hardware SVE intrinsics are not yet implemented; all operations
use the portable scalar path from
super::mod.rs.
Functions§
- dot_
f64_ dispatch - Dot product of two f64 slices using the best available SIMD path.
- encode_
and_ popcount_ dispatch - Fused encode+AND+popcount dispatch.
- fused_
and_ popcount_ dispatch - Fused AND+popcount dispatch using the best available SIMD path.
- fused_
xor_ popcount_ dispatch - Fused XOR+popcount dispatch using the best available SIMD path.
- hamming_
distance_ dispatch - Hamming distance between two packed bitstream slices.
- max_
f64_ dispatch - Maximum of f64 slice using the best available SIMD path.
- pack_
dispatch - Pack u8 bits into u64 words using the best available SIMD path.
- popcount_
dispatch - Count set bits in packed
u64words using the best available SIMD path. - scale_
f64_ dispatch - Scale f64 slice in-place: y[i] *= alpha, using the best available SIMD path.
- softmax_
inplace_ f64_ dispatch - In-place softmax over an f64 slice (numerically stable).
- sum_
f64_ dispatch - Sum of f64 slice using the best available SIMD path.